Integrate google internal changes.

9 files changed, 21795 insertions, 0 deletions

diff --git a/php/ext/google/protobuf/config.m4 b/php/ext/google/protobuf/config.m4
new file mode 100644
index 00000000..b5946f74
--- /dev/null
+++ b/php/ext/google/protobuf/config.m4
@@ -0,0 +1,10 @@
+dnl lines starting with "dnl" are comments
+
+PHP_ARG_ENABLE(protobuf, whether to enable Protobuf extension, [  --enable-protobuf   Enable Protobuf extension])
+
+if test "$PHP_PROTOBUF" != "no"; then
+
+  dnl this defines the extension
+  PHP_NEW_EXTENSION(protobuf, upb.c protobuf.c def.c message.c storage.c, $ext_shared)
+
+fi
diff --git a/php/ext/google/protobuf/def.c b/php/ext/google/protobuf/def.c
new file mode 100644
index 00000000..fc188069
--- /dev/null
+++ b/php/ext/google/protobuf/def.c
@@ -0,0 +1,381 @@
+#include "protobuf.h"
+
+// -----------------------------------------------------------------------------
+// Common Utilities
+// -----------------------------------------------------------------------------
+
+void check_upb_status(const upb_status* status, const char* msg) {
+  if (!upb_ok(status)) {
+    zend_error("%s: %s\n", msg, upb_status_errmsg(status));
+  }
+}
+
+
+static upb_def *check_notfrozen(const upb_def *def) {
+  if (upb_def_isfrozen(def)) {
+    zend_error(E_ERROR,
+               "Attempt to modify a frozen descriptor. Once descriptors are "
+               "added to the descriptor pool, they may not be modified.");
+  }
+  return (upb_def *)def;
+}
+
+static upb_msgdef *check_msgdef_notfrozen(const upb_msgdef *def) {
+  return upb_downcast_msgdef_mutable(check_notfrozen((const upb_def *)def));
+}
+
+static upb_fielddef *check_fielddef_notfrozen(const upb_fielddef *def) {
+  return upb_downcast_fielddef_mutable(check_notfrozen((const upb_def *)def));
+}
+
+#define PROTOBUF_WRAP_INTERN(wrapper, intern, intern_dtor)            \
+  Z_TYPE_P(wrapper) = IS_OBJECT;                                      \
+  Z_OBJVAL_P(wrapper)                                                 \
+      .handle = zend_objects_store_put(                               \
+      intern, (zend_objects_store_dtor_t)zend_objects_destroy_object, \
+      intern_dtor, NULL TSRMLS_CC);                                   \
+  Z_OBJVAL_P(wrapper).handlers = zend_get_std_object_handlers();
+
+#define PROTOBUF_SETUP_ZEND_WRAPPER(class_name, class_name_lower, wrapper,    \
+                                    intern)                                   \
+  Z_TYPE_P(wrapper) = IS_OBJECT;                                              \
+  class_name *intern = ALLOC(class_name);                                     \
+  memset(intern, 0, sizeof(class_name));                                      \
+  class_name_lower##_init_c_instance(intern TSRMLS_CC);                       \
+  Z_OBJVAL_P(wrapper)                                                         \
+      .handle = zend_objects_store_put(intern, NULL, class_name_lower##_free, \
+                                       NULL TSRMLS_CC);                       \
+  Z_OBJVAL_P(wrapper).handlers = zend_get_std_object_handlers();
+
+#define PROTOBUF_CREATE_ZEND_WRAPPER(class_name, class_name_lower, wrapper, \
+                                     intern)                                \
+  MAKE_STD_ZVAL(wrapper);                                                   \
+  PROTOBUF_SETUP_ZEND_WRAPPER(class_name, class_name_lower, wrapper, intern);
+
+#define DEFINE_CLASS(name, name_lower, string_name)                          \
+  zend_class_entry *name_lower##_type;                                       \
+  void name_lower##_init(TSRMLS_D) {                                         \
+    zend_class_entry class_type;                                             \
+    INIT_CLASS_ENTRY(class_type, string_name, name_lower##_methods);         \
+    name_lower##_type = zend_register_internal_class(&class_type TSRMLS_CC); \
+    name_lower##_type->create_object = name_lower##_create;                  \
+  }                                                                          \
+  name *php_to_##name_lower(zval *val TSRMLS_DC) {                           \
+    return (name *)zend_object_store_get_object(val TSRMLS_CC);              \
+  }                                                                          \
+  void name_lower##_free(void *object TSRMLS_DC) {                           \
+    name *intern = (name *)object;                                           \
+    name_lower##_free_c(intern TSRMLS_CC);                                   \
+    efree(object);                                                           \
+  }                                                                          \
+  zend_object_value name_lower##_create(zend_class_entry *ce TSRMLS_DC) {    \
+    zend_object_value return_value;                                          \
+    name *intern = (name *)emalloc(sizeof(name));                            \
+    memset(intern, 0, sizeof(name));                                         \
+    name_lower##_init_c_instance(intern TSRMLS_CC);                          \
+    return_value.handle = zend_objects_store_put(                            \
+        intern, (zend_objects_store_dtor_t)zend_objects_destroy_object,      \
+        name_lower##_free, NULL TSRMLS_CC);                                  \
+    return_value.handlers = zend_get_std_object_handlers();                  \
+    return return_value;                                                     \
+  }
+
+// -----------------------------------------------------------------------------
+// DescriptorPool
+// -----------------------------------------------------------------------------
+
+static zend_function_entry descriptor_pool_methods[] = {
+  PHP_ME(DescriptorPool, addMessage, NULL, ZEND_ACC_PUBLIC)
+  PHP_ME(DescriptorPool, finalize, NULL, ZEND_ACC_PUBLIC)
+  ZEND_FE_END
+};
+
+DEFINE_CLASS(DescriptorPool, descriptor_pool,
+             "Google\\Protobuf\\DescriptorPool");
+
+DescriptorPool *generated_pool;  // The actual generated pool
+
+ZEND_FUNCTION(get_generated_pool) {
+  if (PROTOBUF_G(generated_pool) == NULL) {
+    MAKE_STD_ZVAL(PROTOBUF_G(generated_pool));
+    Z_TYPE_P(PROTOBUF_G(generated_pool)) = IS_OBJECT;
+    generated_pool = ALLOC(DescriptorPool);
+    descriptor_pool_init_c_instance(generated_pool TSRMLS_CC);
+    Z_OBJ_HANDLE_P(PROTOBUF_G(generated_pool)) = zend_objects_store_put(
+        generated_pool, NULL,
+        (zend_objects_free_object_storage_t)descriptor_pool_free, NULL TSRMLS_CC);
+    Z_OBJ_HT_P(PROTOBUF_G(generated_pool)) = zend_get_std_object_handlers();
+  }
+  RETURN_ZVAL(PROTOBUF_G(generated_pool), 1, 0);
+}
+
+void descriptor_pool_init_c_instance(DescriptorPool* pool TSRMLS_DC) {
+  zend_object_std_init(&pool->std, descriptor_pool_type TSRMLS_CC);
+  pool->symtab = upb_symtab_new(&pool->symtab);
+
+  ALLOC_HASHTABLE(pool->pending_list);
+  zend_hash_init(pool->pending_list, 1, NULL, ZVAL_PTR_DTOR, 0);
+}
+
+void descriptor_pool_free_c(DescriptorPool *pool TSRMLS_DC) {
+  upb_symtab_unref(pool->symtab, &pool->symtab);
+  zend_hash_destroy(pool->pending_list);
+  FREE_HASHTABLE(pool->pending_list);
+}
+
+PHP_METHOD(DescriptorPool, addMessage) {
+  char *name = NULL;
+  int str_len;
+  if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "s", &name, &str_len) ==
+      FAILURE) {
+    return;
+  }
+
+  zval* retval = NULL;
+  PROTOBUF_CREATE_ZEND_WRAPPER(MessageBuilderContext, message_builder_context,
+                               retval, context);
+
+  MAKE_STD_ZVAL(context->pool);
+  ZVAL_ZVAL(context->pool, getThis(), 1, 0);
+
+  Descriptor *desc = php_to_descriptor(context->descriptor TSRMLS_CC);
+  Descriptor_name_set(desc, name);
+
+  RETURN_ZVAL(retval, 0, 1);
+}
+
+static void validate_msgdef(const upb_msgdef* msgdef) {
+  // Verify that no required fields exist. proto3 does not support these.
+  upb_msg_field_iter it;
+  for (upb_msg_field_begin(&it, msgdef);
+       !upb_msg_field_done(&it);
+       upb_msg_field_next(&it)) {
+    const upb_fielddef* field = upb_msg_iter_field(&it);
+    if (upb_fielddef_label(field) == UPB_LABEL_REQUIRED) {
+      zend_error(E_ERROR, "Required fields are unsupported in proto3.");
+    }
+  }
+}
+
+PHP_METHOD(DescriptorPool, finalize) {
+  DescriptorPool *self = php_to_descriptor_pool(getThis() TSRMLS_CC);
+  Bucket *temp;
+  int i, num;
+
+  num = zend_hash_num_elements(self->pending_list);
+  upb_def **defs = emalloc(sizeof(upb_def *) * num);
+
+  for (i = 0, temp = self->pending_list->pListHead; temp != NULL;
+       temp = temp->pListNext) {
+    zval *def_php = *(zval **)temp->pData;
+    Descriptor* desc = php_to_descriptor(def_php TSRMLS_CC);
+    defs[i] = (upb_def *)desc->msgdef;
+    validate_msgdef((const upb_msgdef *)defs[i++]);
+  }
+
+  CHECK_UPB(upb_symtab_add(self->symtab, (upb_def **)defs, num, NULL, &status),
+            "Unable to add defs to DescriptorPool");
+
+  for (temp = self->pending_list->pListHead; temp != NULL;
+       temp = temp->pListNext) {
+    // zval *def_php = *(zval **)temp->pData;
+    // Descriptor* desc = php_to_descriptor(def_php TSRMLS_CC);
+    build_class_from_descriptor((zval *)temp->pDataPtr TSRMLS_CC);
+  }
+
+  FREE(defs);
+  zend_hash_destroy(self->pending_list);
+  zend_hash_init(self->pending_list, 1, NULL, ZVAL_PTR_DTOR, 0);
+}
+
+// -----------------------------------------------------------------------------
+// Descriptor
+// -----------------------------------------------------------------------------
+
+static zend_function_entry descriptor_methods[] = {
+  ZEND_FE_END
+};
+
+DEFINE_CLASS(Descriptor, descriptor, "Google\\Protobuf\\Descriptor");
+
+void descriptor_free_c(Descriptor *self TSRMLS_DC) {
+  upb_msg_field_iter iter;
+  upb_msg_field_begin(&iter, self->msgdef);
+  while (!upb_msg_field_done(&iter)) {
+    upb_fielddef *fielddef = upb_msg_iter_field(&iter);
+    upb_fielddef_unref(fielddef, &fielddef);
+    upb_msg_field_next(&iter);
+  }
+  upb_msgdef_unref(self->msgdef, &self->msgdef);
+  if (self->layout) {
+    free_layout(self->layout);
+  }
+}
+
+static void descriptor_add_field(Descriptor *desc,
+                                 const upb_fielddef *fielddef) {
+  upb_msgdef *mut_def = check_msgdef_notfrozen(desc->msgdef);
+  upb_fielddef *mut_field_def = check_fielddef_notfrozen(fielddef);
+  CHECK_UPB(upb_msgdef_addfield(mut_def, mut_field_def, NULL, &status),
+            "Adding field to Descriptor failed");
+  // add_def_obj(fielddef, obj);
+}
+
+void descriptor_init_c_instance(Descriptor* desc TSRMLS_DC) {
+  zend_object_std_init(&desc->std, descriptor_type TSRMLS_CC);
+  desc->msgdef = upb_msgdef_new(&desc->msgdef);
+  desc->layout = NULL;
+  // MAKE_STD_ZVAL(intern->klass);
+  // ZVAL_NULL(intern->klass);
+  desc->pb_serialize_handlers = NULL;
+}
+
+void Descriptor_name_set(Descriptor *desc, const char *name) {
+  upb_msgdef *mut_def = check_msgdef_notfrozen(desc->msgdef);
+  CHECK_UPB(upb_msgdef_setfullname(mut_def, name, &status),
+            "Error setting Descriptor name");
+}
+
+// -----------------------------------------------------------------------------
+// FieldDescriptor
+// -----------------------------------------------------------------------------
+
+static void field_descriptor_name_set(const upb_fielddef* fielddef,
+                                      const char *name) {
+  upb_fielddef *mut_def = check_fielddef_notfrozen(fielddef);
+  CHECK_UPB(upb_fielddef_setname(mut_def, name, &status),
+            "Error setting FieldDescriptor name");
+}
+
+static void field_descriptor_label_set(const upb_fielddef* fielddef,
+                                       upb_label_t upb_label) {
+  upb_fielddef *mut_def = check_fielddef_notfrozen(fielddef);
+  upb_fielddef_setlabel(mut_def, upb_label);
+}
+
+upb_fieldtype_t string_to_descriptortype(const char *type) {
+#define CONVERT(upb, str)   \
+  if (!strcmp(type, str)) { \
+    return UPB_DESCRIPTOR_TYPE_##upb;  \
+  }
+
+  CONVERT(FLOAT, "float");
+  CONVERT(DOUBLE, "double");
+  CONVERT(BOOL, "bool");
+  CONVERT(STRING, "string");
+  CONVERT(BYTES, "bytes");
+  CONVERT(MESSAGE, "message");
+  CONVERT(GROUP, "group");
+  CONVERT(ENUM, "enum");
+  CONVERT(INT32, "int32");
+  CONVERT(INT64, "int64");
+  CONVERT(UINT32, "uint32");
+  CONVERT(UINT64, "uint64");
+  CONVERT(SINT32, "sint32");
+  CONVERT(SINT64, "sint64");
+  CONVERT(FIXED32, "fixed32");
+  CONVERT(FIXED64, "fixed64");
+  CONVERT(SFIXED32, "sfixed32");
+  CONVERT(SFIXED64, "sfixed64");
+
+#undef CONVERT
+
+  zend_error(E_ERROR, "Unknown field type.");
+  return 0;
+}
+
+static void field_descriptor_type_set(const upb_fielddef* fielddef,
+                                      const char *type) {
+  upb_fielddef *mut_def = check_fielddef_notfrozen(fielddef);
+  upb_fielddef_setdescriptortype(mut_def, string_to_descriptortype(type));
+}
+
+static void field_descriptor_number_set(const upb_fielddef* fielddef,
+                                        int number) {
+  upb_fielddef *mut_def = check_fielddef_notfrozen(fielddef);
+  CHECK_UPB(upb_fielddef_setnumber(mut_def, number, &status),
+            "Error setting field number");
+}
+
+// -----------------------------------------------------------------------------
+// MessageBuilderContext
+// -----------------------------------------------------------------------------
+
+static zend_function_entry message_builder_context_methods[] = {
+    PHP_ME(MessageBuilderContext, finalizeToPool, NULL, ZEND_ACC_PUBLIC)
+    PHP_ME(MessageBuilderContext, optional, NULL, ZEND_ACC_PUBLIC)
+    {NULL, NULL, NULL}
+};
+
+DEFINE_CLASS(MessageBuilderContext, message_builder_context,
+             "Google\\Protobuf\\Internal\\MessageBuilderContext");
+
+void message_builder_context_free_c(MessageBuilderContext *context TSRMLS_DC) {
+  zval_ptr_dtor(&context->descriptor);
+  zval_ptr_dtor(&context->pool);
+}
+
+void message_builder_context_init_c_instance(
+    MessageBuilderContext *context TSRMLS_DC) {
+  zend_object_std_init(&context->std, message_builder_context_type TSRMLS_CC);
+  PROTOBUF_CREATE_ZEND_WRAPPER(Descriptor, descriptor, context->descriptor,
+                               desc);
+}
+
+static void msgdef_add_field(Descriptor *desc, upb_label_t upb_label,
+                             const char *name, const char *type, int number,
+                             const char *type_class) {
+  upb_fielddef *fielddef = upb_fielddef_new(&fielddef);
+  upb_fielddef_setpacked(fielddef, false);
+
+  field_descriptor_label_set(fielddef, upb_label);
+  field_descriptor_name_set(fielddef, name);
+  field_descriptor_type_set(fielddef, type);
+  field_descriptor_number_set(fielddef, number);
+
+// //   if (type_class != Qnil) {
+// //     if (TYPE(type_class) != T_STRING) {
+// //       rb_raise(rb_eArgError, "Expected string for type class");
+// //     }
+// //     // Make it an absolute type name by prepending a dot.
+// //     type_class = rb_str_append(rb_str_new2("."), type_class);
+// //     rb_funcall(fielddef, rb_intern("submsg_name="), 1, type_class);
+// //   }
+  descriptor_add_field(desc, fielddef);
+}
+
+PHP_METHOD(MessageBuilderContext, optional) {
+  MessageBuilderContext *self = php_to_message_builder_context(getThis() TSRMLS_CC);
+  Descriptor *desc = php_to_descriptor(self->descriptor TSRMLS_CC);
+  // VALUE name, type, number, type_class;
+  const char *name, *type, *type_class;
+  int number, name_str_len, type_str_len, type_class_str_len;
+  if (ZEND_NUM_ARGS() == 3) {
+    if (zend_parse_parameters(3 TSRMLS_CC, "ssl", &name,
+                              &name_str_len, &type, &type_str_len, &number) == FAILURE) {
+      return;
+    }
+  } else {
+    if (zend_parse_parameters(4 TSRMLS_CC, "ssls", &name,
+                              &name_str_len, &type, &type_str_len, &number, &type_class,
+                              &type_class_str_len) == FAILURE) {
+      return;
+    }
+  }
+
+  msgdef_add_field(desc, UPB_LABEL_OPTIONAL, name, type, number, type_class);
+
+  zval_copy_ctor(getThis());
+  RETURN_ZVAL(getThis(), 1, 0);
+}
+
+PHP_METHOD(MessageBuilderContext, finalizeToPool) {
+  MessageBuilderContext *self = php_to_message_builder_context(getThis() TSRMLS_CC);
+  DescriptorPool *pool = php_to_descriptor_pool(self->pool TSRMLS_CC);
+  Descriptor* desc = php_to_descriptor(self->descriptor TSRMLS_CC);
+
+  Z_ADDREF_P(self->descriptor);
+  zend_hash_next_index_insert(pool->pending_list, &self->descriptor,
+                              sizeof(zval *), NULL);
+  RETURN_ZVAL(self->pool, 1, 0);
+}
diff --git a/php/ext/google/protobuf/message.c b/php/ext/google/protobuf/message.c
new file mode 100644
index 00000000..c062d665
--- /dev/null
+++ b/php/ext/google/protobuf/message.c
@@ -0,0 +1,273 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2014 Google Inc.  All rights reserved.
+// https://developers.google.com/protocol-buffers/
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include <php.h>
+
+#include "protobuf.h"
+
+// -----------------------------------------------------------------------------
+// Class/module creation from msgdefs and enumdefs, respectively.
+// -----------------------------------------------------------------------------
+
+void* message_data(void* msg) {
+  return ((uint8_t *)msg) + sizeof(MessageHeader);
+}
+
+void message_set_property(zval* object, zval* field_name, zval* value,
+                          const zend_literal* key TSRMLS_DC) {
+  const upb_fielddef* field;
+
+  MessageHeader* self = zend_object_store_get_object(object TSRMLS_CC);
+
+  CHECK_TYPE(field_name, IS_STRING);
+  field = upb_msgdef_ntofz(self->descriptor->msgdef, Z_STRVAL_P(field_name));
+  if (field == NULL) {
+    zend_error(E_ERROR, "Unknown field: %s", Z_STRVAL_P(field_name));
+  }
+  layout_set(self->descriptor->layout, message_data(self), field, value);
+}
+
+zval* message_get_property(zval* object, zval* member, int type,
+                             const zend_literal* key TSRMLS_DC) {
+  MessageHeader* self =
+      (MessageHeader*)zend_object_store_get_object(object TSRMLS_CC);
+  CHECK_TYPE(member, IS_STRING);
+
+  const upb_fielddef* field;
+  field = upb_msgdef_ntofz(self->descriptor->msgdef, Z_STRVAL_P(member));
+  if (field == NULL) {
+    return EG(uninitialized_zval_ptr);
+  }
+  zval* retval = layout_get(self->descriptor->layout, message_data(self), field TSRMLS_CC);
+  return retval;
+}
+
+static  zend_function_entry message_methods[] = {
+  PHP_ME(Message, encode, NULL, ZEND_ACC_PUBLIC)
+  {NULL, NULL, NULL}
+};
+
+/* stringsink *****************************************************************/
+
+// This should probably be factored into a common upb component.
+
+typedef struct {
+  upb_byteshandler handler;
+  upb_bytessink sink;
+  char *ptr;
+  size_t len, size;
+} stringsink;
+
+static void *stringsink_start(void *_sink, const void *hd, size_t size_hint) {
+  stringsink *sink = _sink;
+  sink->len = 0;
+  return sink;
+}
+
+static size_t stringsink_string(void *_sink, const void *hd, const char *ptr,
+                                size_t len, const upb_bufhandle *handle) {
+  stringsink *sink = _sink;
+  size_t new_size = sink->size;
+
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+
+  while (sink->len + len > new_size) {
+    new_size *= 2;
+  }
+
+  if (new_size != sink->size) {
+    sink->ptr = realloc(sink->ptr, new_size);
+    sink->size = new_size;
+  }
+
+  memcpy(sink->ptr + sink->len, ptr, len);
+  sink->len += len;
+
+  return len;
+}
+
+void stringsink_init(stringsink *sink) {
+  upb_byteshandler_init(&sink->handler);
+  upb_byteshandler_setstartstr(&sink->handler, stringsink_start, NULL);
+  upb_byteshandler_setstring(&sink->handler, stringsink_string, NULL);
+
+  upb_bytessink_reset(&sink->sink, &sink->handler, sink);
+
+  sink->size = 32;
+  sink->ptr = malloc(sink->size);
+  sink->len = 0;
+}
+
+void stringsink_uninit(stringsink *sink) { free(sink->ptr); }
+
+// Stack-allocated context during an encode/decode operation. Contains the upb
+// environment and its stack-based allocator, an initial buffer for allocations
+// to avoid malloc() when possible, and a template for PHP exception messages
+// if any error occurs.
+#define STACK_ENV_STACKBYTES 4096
+typedef struct {
+  upb_env env;
+  upb_seededalloc alloc;
+  const char *php_error_template;
+  char allocbuf[STACK_ENV_STACKBYTES];
+} stackenv;
+
+static void stackenv_init(stackenv* se, const char* errmsg);
+static void stackenv_uninit(stackenv* se);
+
+// Callback invoked by upb if any error occurs during parsing or serialization.
+static bool env_error_func(void* ud, const upb_status* status) {
+    stackenv* se = ud;
+    // Free the env -- rb_raise will longjmp up the stack past the encode/decode
+    // function so it would not otherwise have been freed.
+    stackenv_uninit(se);
+
+    // TODO(teboring): have a way to verify that this is actually a parse error,
+    // instead of just throwing "parse error" unconditionally.
+    zend_error(E_ERROR, se->php_error_template);
+    // Never reached.
+    return false;
+}
+
+static void stackenv_init(stackenv* se, const char* errmsg) {
+  se->php_error_template = errmsg;
+  upb_env_init(&se->env);
+  upb_seededalloc_init(&se->alloc, &se->allocbuf, STACK_ENV_STACKBYTES);
+  upb_env_setallocfunc(&se->env, upb_seededalloc_getallocfunc(&se->alloc),
+                       &se->alloc);
+  upb_env_seterrorfunc(&se->env, env_error_func, se);
+}
+
+static void stackenv_uninit(stackenv* se) {
+  upb_env_uninit(&se->env);
+  upb_seededalloc_uninit(&se->alloc);
+}
+
+// -----------------------------------------------------------------------------
+// Message
+// -----------------------------------------------------------------------------
+
+static const upb_handlers* msgdef_pb_serialize_handlers(Descriptor* desc) {
+  if (desc->pb_serialize_handlers == NULL) {
+    desc->pb_serialize_handlers =
+        upb_pb_encoder_newhandlers(desc->msgdef, &desc->pb_serialize_handlers);
+  }
+  return desc->pb_serialize_handlers;
+}
+
+PHP_METHOD(Message, encode) {
+  Descriptor* desc = (Descriptor*)zend_object_store_get_object(
+      CE_STATIC_MEMBERS(Z_OBJCE_P(getThis()))[0] TSRMLS_CC);
+
+  stringsink sink;
+  stringsink_init(&sink);
+
+  {
+    const upb_handlers* serialize_handlers = msgdef_pb_serialize_handlers(desc);
+
+    stackenv se;
+    upb_pb_encoder* encoder;
+
+    stackenv_init(&se, "Error occurred during encoding: %s");
+    encoder = upb_pb_encoder_create(&se.env, serialize_handlers, &sink.sink);
+
+    putmsg(getThis(), desc, upb_pb_encoder_input(encoder), 0);
+
+    RETVAL_STRINGL(sink.ptr, sink.len, 1);
+
+    stackenv_uninit(&se);
+    stringsink_uninit(&sink);
+  }
+}
+
+void message_free(void * object TSRMLS_DC) {
+  FREE(object);
+}
+
+zend_object_value message_create(zend_class_entry* ce TSRMLS_DC) {
+  zend_object_value return_value;
+
+  zval* php_descriptor = get_def_obj(ce);
+
+  Descriptor* desc = zend_object_store_get_object(php_descriptor TSRMLS_CC);
+  MessageHeader* msg = (MessageHeader*)ALLOC_N(
+      uint8_t, sizeof(MessageHeader) + desc->layout->size);
+  memset(message_data(msg), 0, desc->layout->size);
+
+  // We wrap first so that everything in the message object is GC-rooted in case
+  // a collection happens during object creation in layout_init().
+  msg->descriptor = desc;
+
+  layout_init(desc->layout, message_data(msg));
+  zend_object_std_init(&msg->std, ce TSRMLS_CC);
+
+  return_value.handle = zend_objects_store_put(
+      msg, (zend_objects_store_dtor_t)zend_objects_destroy_object,
+      message_free, NULL TSRMLS_CC);
+
+  return_value.handlers = PROTOBUF_G(message_handlers);
+  return return_value;
+}
+
+const zend_class_entry* build_class_from_descriptor(
+    zval* php_descriptor TSRMLS_DC) {
+  Descriptor* desc = php_to_descriptor(php_descriptor);
+  if (desc->layout == NULL) {
+    MessageLayout* layout = create_layout(desc->msgdef);
+    desc->layout = layout;
+  }
+  // TODO(teboring): Add it back.
+  // if (desc->fill_method == NULL) {
+  //   desc->fill_method = new_fillmsg_decodermethod(desc, &desc->fill_method);
+  // }
+
+  const char* name = upb_msgdef_fullname(desc->msgdef);
+  if (name == NULL) {
+    zend_error(E_ERROR, "Descriptor does not have assigned name.");
+  }
+
+  zend_class_entry class_entry;
+  INIT_CLASS_ENTRY_EX(class_entry, name, strlen(name), message_methods);
+  zend_class_entry* registered_ce =
+      zend_register_internal_class(&class_entry TSRMLS_CC);
+
+  add_def_obj(registered_ce, php_descriptor);
+
+  if (PROTOBUF_G(message_handlers) == NULL) {
+    PROTOBUF_G(message_handlers) = ALLOC(zend_object_handlers);
+    memcpy(PROTOBUF_G(message_handlers), zend_get_std_object_handlers(),
+           sizeof(zend_object_handlers));
+    PROTOBUF_G(message_handlers)->write_property = message_set_property;
+    PROTOBUF_G(message_handlers)->read_property = message_get_property;
+  }
+
+  registered_ce->create_object = message_create;
+}
diff --git a/php/ext/google/protobuf/protobuf.c b/php/ext/google/protobuf/protobuf.c
new file mode 100644
index 00000000..b1ace8b0
--- /dev/null
+++ b/php/ext/google/protobuf/protobuf.c
@@ -0,0 +1,89 @@
+#include "protobuf.h"
+
+#include <zend_hash.h>
+
+ZEND_DECLARE_MODULE_GLOBALS(protobuf)
+static PHP_GINIT_FUNCTION(protobuf);
+static PHP_GSHUTDOWN_FUNCTION(protobuf);
+
+// -----------------------------------------------------------------------------
+// Global map from upb {msg,enum}defs to wrapper Descriptor/EnumDescriptor
+// instances.
+// -----------------------------------------------------------------------------
+
+void add_def_obj(const void* def, zval* value) {
+  uint nIndex = (ulong)def & PROTOBUF_G(upb_def_to_php_obj_map).nTableMask;
+
+  zval* pDest = NULL;
+  Z_ADDREF_P(value);
+  zend_hash_index_update(&PROTOBUF_G(upb_def_to_php_obj_map), (zend_ulong)def,
+                         &value, sizeof(zval*), &pDest);
+}
+
+zval* get_def_obj(const void* def) {
+  zval** value;
+  if (zend_hash_index_find(&PROTOBUF_G(upb_def_to_php_obj_map), (zend_ulong)def,
+                           &value) == FAILURE) {
+    zend_error(E_ERROR, "PHP object not found for given definition.\n");
+    return NULL;
+  }
+  return *value;
+}
+
+// -----------------------------------------------------------------------------
+// Utilities.
+// -----------------------------------------------------------------------------
+
+// define the function(s) we want to add
+zend_function_entry protobuf_functions[] = {
+  ZEND_FE(get_generated_pool, NULL)
+  ZEND_FE_END
+};
+
+// "protobuf_functions" refers to the struct defined above
+// we'll be filling in more of this later: you can use this to specify
+// globals, php.ini info, startup and teardown functions, etc.
+zend_module_entry protobuf_module_entry = {
+  STANDARD_MODULE_HEADER,
+  PHP_PROTOBUF_EXTNAME, // extension name
+  protobuf_functions,   // function list
+  PHP_MINIT(protobuf),  // process startup
+  NULL,  // process shutdown
+  NULL,  // request startup
+  NULL,  // request shutdown
+  NULL,  // extension info
+  PHP_PROTOBUF_VERSION, // extension version
+  PHP_MODULE_GLOBALS(protobuf),  // globals descriptor
+  PHP_GINIT(protobuf), // globals ctor
+  PHP_GSHUTDOWN(protobuf),  // globals dtor
+  NULL,  // post deactivate
+  STANDARD_MODULE_PROPERTIES_EX
+};
+
+// install module
+ZEND_GET_MODULE(protobuf)
+
+// global variables
+static PHP_GINIT_FUNCTION(protobuf) {
+  protobuf_globals->generated_pool = NULL;
+  generated_pool = NULL;
+  protobuf_globals->message_handlers = NULL;
+  zend_hash_init(&protobuf_globals->upb_def_to_php_obj_map, 16, NULL,
+                 ZVAL_PTR_DTOR, 0);
+}
+
+static PHP_GSHUTDOWN_FUNCTION(protobuf) {
+  if (protobuf_globals->generated_pool != NULL) {
+    FREE_ZVAL(protobuf_globals->generated_pool);
+  }
+  if (protobuf_globals->message_handlers != NULL) {
+    FREE(protobuf_globals->message_handlers);
+  }
+  zend_hash_destroy(&protobuf_globals->upb_def_to_php_obj_map);
+}
+
+PHP_MINIT_FUNCTION(protobuf) {
+  descriptor_pool_init(TSRMLS_C);
+  descriptor_init(TSRMLS_C);
+  message_builder_context_init(TSRMLS_C);
+}
diff --git a/php/ext/google/protobuf/protobuf.h b/php/ext/google/protobuf/protobuf.h
new file mode 100644
index 00000000..f9038550
--- /dev/null
+++ b/php/ext/google/protobuf/protobuf.h
@@ -0,0 +1,281 @@
+#ifndef __GOOGLE_PROTOBUF_PHP_PROTOBUF_H__
+#define __GOOGLE_PROTOBUF_PHP_PROTOBUF_H__
+
+#include <php.h>
+
+#include "upb.h"
+
+#define PHP_PROTOBUF_EXTNAME "protobuf"
+#define PHP_PROTOBUF_VERSION "0.01"
+
+// Forward decls.
+struct DescriptorPool;
+struct Descriptor;
+struct FieldDescriptor;
+struct EnumDescriptor;
+struct MessageLayout;
+struct MessageField;
+struct MessageHeader;
+struct MessageBuilderContext;
+struct EnumBuilderContext;
+
+typedef struct DescriptorPool DescriptorPool;
+typedef struct Descriptor Descriptor;
+typedef struct FieldDescriptor FieldDescriptor;
+typedef struct OneofDescriptor OneofDescriptor;
+typedef struct EnumDescriptor EnumDescriptor;
+typedef struct MessageLayout MessageLayout;
+typedef struct MessageField MessageField;
+typedef struct MessageHeader MessageHeader;
+typedef struct MessageBuilderContext MessageBuilderContext;
+typedef struct OneofBuilderContext OneofBuilderContext;
+typedef struct EnumBuilderContext EnumBuilderContext;
+
+extern zend_class_entry* builder_type;
+extern zend_class_entry* descriptor_type;
+extern zend_class_entry* message_builder_context_type;
+
+extern DescriptorPool* generated_pool;  // The actual generated pool
+
+ZEND_BEGIN_MODULE_GLOBALS(protobuf)
+  zval* generated_pool;
+  zend_object_handlers* message_handlers;
+  HashTable upb_def_to_php_obj_map;
+ZEND_END_MODULE_GLOBALS(protobuf)
+
+ZEND_DECLARE_MODULE_GLOBALS(protobuf)
+
+#ifdef ZTS
+#define PROTOBUF_G(v) TSRMG(protobuf_globals_id, zend_protobuf_globals*, v)
+#else
+#define PROTOBUF_G(v) (protobuf_globals.v)
+#endif
+
+// -----------------------------------------------------------------------------
+// PHP functions and global variables.
+// -----------------------------------------------------------------------------
+
+PHP_MINIT_FUNCTION(protobuf);
+
+// -----------------------------------------------------------------------------
+// PHP class structure.
+// -----------------------------------------------------------------------------
+
+struct DescriptorPool {
+  zend_object std;
+  upb_symtab* symtab;
+  HashTable* pending_list;
+};
+
+struct Descriptor {
+  zend_object std;
+  const upb_msgdef* msgdef;
+  MessageLayout* layout;
+  // zval* klass;  // begins as NULL
+  // const upb_handlers* fill_handlers;
+  // const upb_pbdecodermethod* fill_method;
+  const upb_handlers* pb_serialize_handlers;
+  // const upb_handlers* json_serialize_handlers;
+  // Handlers hold type class references for sub-message fields directly in some
+  // cases. We need to keep these rooted because they might otherwise be
+  // collected.
+  // zval_array typeclass_references;
+};
+
+struct FieldDescriptor {
+  zend_object std;
+  const upb_fielddef* fielddef;
+};
+
+struct OneofDescriptor {
+  zend_object std;
+  const upb_oneofdef* oneofdef;
+};
+
+struct EnumDescriptor {
+  zend_object std;
+  const upb_enumdef* enumdef;
+  // zval* module;  // begins as NULL
+};
+
+// -----------------------------------------------------------------------------
+// Native slot storage abstraction.
+// -----------------------------------------------------------------------------
+
+#define NATIVE_SLOT_MAX_SIZE sizeof(uint64_t)
+
+size_t native_slot_size(upb_fieldtype_t type);
+
+#define MAP_KEY_FIELD 1
+#define MAP_VALUE_FIELD 2
+
+// Oneof case slot value to indicate that no oneof case is set. The value `0` is
+// safe because field numbers are used as case identifiers, and no field can
+// have a number of 0.
+#define ONEOF_CASE_NONE 0
+
+// These operate on a map field (i.e., a repeated field of submessages whose
+// submessage type is a map-entry msgdef).
+bool is_map_field(const upb_fielddef* field);
+const upb_fielddef* map_field_key(const upb_fielddef* field);
+const upb_fielddef* map_field_value(const upb_fielddef* field);
+
+// These operate on a map-entry msgdef.
+const upb_fielddef* map_entry_key(const upb_msgdef* msgdef);
+const upb_fielddef* map_entry_value(const upb_msgdef* msgdef);
+
+// -----------------------------------------------------------------------------
+// Message layout / storage.
+// -----------------------------------------------------------------------------
+
+#define MESSAGE_FIELD_NO_CASE ((size_t)-1)
+
+struct MessageField {
+  size_t offset;
+  size_t case_offset;  // for oneofs, a uint32. Else, MESSAGE_FIELD_NO_CASE.
+};
+
+struct MessageLayout {
+  const upb_msgdef* msgdef;
+  MessageField* fields;
+  size_t size;
+};
+
+void layout_init(MessageLayout* layout, void* storage);
+zval* layout_get(MessageLayout* layout, const void* storage,
+                 const upb_fielddef* field TSRMLS_DC);
+MessageLayout* create_layout(const upb_msgdef* msgdef);
+void free_layout(MessageLayout* layout);
+zval* native_slot_get(upb_fieldtype_t type, /*VALUE type_class,*/
+                      const void* memory TSRMLS_DC);
+
+// -----------------------------------------------------------------------------
+// Message class creation.
+// -----------------------------------------------------------------------------
+
+struct MessageHeader {
+  zend_object std;
+  Descriptor* descriptor;  // kept alive by self.class.descriptor reference.
+                           // Data comes after this.
+};
+
+struct MessageBuilderContext {
+  zend_object std;
+  zval* descriptor;
+  zval* pool;
+};
+
+struct OneofBuilderContext {
+  zend_object std;
+  // VALUE descriptor;
+  // VALUE builder;
+};
+
+struct EnumBuilderContext {
+  zend_object std;
+  // VALUE enumdesc;
+};
+
+// Forward-declare all of the PHP method implementations.
+
+DescriptorPool* php_to_descriptor_pool(zval* value TSRMLS_DC);
+zend_object_value descriptor_pool_create(zend_class_entry *ce TSRMLS_DC);
+void descriptor_pool_free_c(DescriptorPool* object TSRMLS_DC);
+void descriptor_pool_free(void* object TSRMLS_DC);
+void descriptor_pool_init_c_instance(DescriptorPool* pool TSRMLS_DC);
+PHP_METHOD(DescriptorPool, addMessage);
+PHP_METHOD(DescriptorPool, finalize);
+
+Descriptor* php_to_descriptor(zval* value TSRMLS_DC);
+zend_object_value descriptor_create(zend_class_entry *ce TSRMLS_DC);
+void descriptor_init_c_instance(Descriptor* intern TSRMLS_DC);
+void descriptor_free_c(Descriptor* object TSRMLS_DC);
+void descriptor_free(void* object TSRMLS_DC);
+void descriptor_name_set(Descriptor *desc, const char *name);
+
+MessageBuilderContext* php_to_message_builder_context(zval* value TSRMLS_DC);
+zend_object_value message_builder_context_create(
+    zend_class_entry* ce TSRMLS_DC);
+void message_builder_context_init_c_instance(
+    MessageBuilderContext* intern TSRMLS_DC);
+void message_builder_context_free_c(MessageBuilderContext* object TSRMLS_DC);
+void message_builder_context_free(void* object TSRMLS_DC);
+PHP_METHOD(MessageBuilderContext, optional);
+PHP_METHOD(MessageBuilderContext, finalizeToPool);
+
+PHP_METHOD(Message, encode);
+const zend_class_entry* build_class_from_descriptor(
+    zval* php_descriptor TSRMLS_DC);
+
+PHP_FUNCTION(get_generated_pool);
+
+// -----------------------------------------------------------------------------
+// Global map from upb {msg,enum}defs to wrapper Descriptor/EnumDescriptor
+// instances.
+// ----------------------------------------------------------------------------
+
+void add_def_obj(const void* def, zval* value);
+zval* get_def_obj(const void* def);
+
+// -----------------------------------------------------------------------------
+// Utilities.
+// -----------------------------------------------------------------------------
+
+// PHP Array utils.
+#define Z_ARRVAL_SIZE_P(zval_p) zend_hash_num_elements(Z_ARRVAL_P(zval_p))
+#define Z_ARRVAL_BEGIN_P(zval_p) Z_ARRVAL_P(zval_p)->pListHead
+#define Z_BUCKET_NEXT_PP(bucket_pp) *bucket_pp = (*bucket_pp)->pListNext
+
+#define DEFINE_PHP_OBJECT(class_name, class_name_lower, name) \
+  do {                                                        \
+    zval* name;                                               \
+    MAKE_STD_ZVAL(name);                                      \
+    object_init_ex(name, class_name_lower##_type);            \
+  } while (0)
+
+#define DEFINE_PHP_WRAPPER(class_name, class_name_lower, name, intern) \
+  zval* name;                                                          \
+  MAKE_STD_ZVAL(name);                                                 \
+  object_init_ex(name, class_name_lower##_type);                       \
+  Z_OBJVAL_P(name)                                                     \
+      .handle = zend_objects_store_put(                                \
+      intern, (zend_objects_store_dtor_t)zend_objects_destroy_object,  \
+      class_name_lower##_free, NULL TSRMLS_CC);
+
+#define DEFINE_PHP_ZVAL(name) \
+  do {                        \
+    zval* name;               \
+    MAKE_STD_ZVAL(name);      \
+  } while (0)
+
+#define DEFINE_PHP_STRING(name, value) \
+  do {                                 \
+    zval* name;                        \
+    MAKE_STD_ZVAL(name);               \
+    ZVAL_STRING(name, value, 1);       \
+  } while (0)
+
+// Upb Utilities
+
+void check_upb_status(const upb_status* status, const char* msg);
+
+#define CHECK_UPB(code, msg)             \
+  do {                                   \
+    upb_status status = UPB_STATUS_INIT; \
+    code;                                \
+    check_upb_status(&status, msg);      \
+  } while (0)
+
+// Memory management
+
+#define ALLOC(class_name) (class_name*) emalloc(sizeof(class_name))
+#define ALLOC_N(class_name, n) (class_name*) emalloc(sizeof(class_name) * n)
+#define FREE(object) efree(object)
+
+// Type Checking
+#define CHECK_TYPE(field, type)             \
+  if (Z_TYPE_P(field) != type) {            \
+    zend_error(E_ERROR, "Unexpected type"); \
+  }
+
+#endif  // __GOOGLE_PROTOBUF_PHP_PROTOBUF_H__
diff --git a/php/ext/google/protobuf/storage.c b/php/ext/google/protobuf/storage.c
new file mode 100644
index 00000000..e5a09c17
--- /dev/null
+++ b/php/ext/google/protobuf/storage.c
@@ -0,0 +1,539 @@
+#include <stdint.h>
+#include <protobuf.h>
+
+// -----------------------------------------------------------------------------
+// PHP <-> native slot management.
+// -----------------------------------------------------------------------------
+
+static zval* int32_to_zval(int32_t value) {
+  zval* tmp;
+  MAKE_STD_ZVAL(tmp);
+  ZVAL_LONG(tmp, value);
+  php_printf("int32 to zval\n");
+  // ZVAL_LONG(tmp, 1);
+  return tmp;
+}
+
+#define DEREF(memory, type) *(type*)(memory)
+
+size_t native_slot_size(upb_fieldtype_t type) {
+  switch (type) {
+    case UPB_TYPE_FLOAT: return 4;
+    case UPB_TYPE_DOUBLE: return 8;
+    case UPB_TYPE_BOOL: return 1;
+    case UPB_TYPE_STRING: return sizeof(zval*);
+    case UPB_TYPE_BYTES: return sizeof(zval*);
+    case UPB_TYPE_MESSAGE: return sizeof(zval*);
+    case UPB_TYPE_ENUM: return 4;
+    case UPB_TYPE_INT32: return 4;
+    case UPB_TYPE_INT64: return 8;
+    case UPB_TYPE_UINT32: return 4;
+    case UPB_TYPE_UINT64: return 8;
+    default: return 0;
+  }
+}
+
+static bool is_php_num(zval* value) {
+  // Is numerial string also valid?
+  return (Z_TYPE_P(value) == IS_LONG ||
+          Z_TYPE_P(value) == IS_DOUBLE);
+}
+
+void native_slot_check_int_range_precision(upb_fieldtype_t type, zval* val) {
+  // TODO(teboring): Add it back.
+  // if (!is_php_num(val)) {
+  //   zend_error(E_ERROR, "Expected number type for integral field.");
+  // }
+
+  // if (Z_TYPE_P(val) == IS_DOUBLE) {
+  //   double dbl_val = NUM2DBL(val);
+  //   if (floor(dbl_val) != dbl_val) {
+  //     zend_error(E_ERROR,
+  //              "Non-integral floating point value assigned to integer field.");
+  //   }
+  // }
+  // if (type == UPB_TYPE_UINT32 || type == UPB_TYPE_UINT64) {
+  //   if (NUM2DBL(val) < 0) {
+  //     zend_error(E_ERROR,
+  //              "Assigning negative value to unsigned integer field.");
+  //   }
+  // }
+}
+
+zval* native_slot_get(upb_fieldtype_t type, /*VALUE type_class,*/
+                      const void* memory TSRMLS_DC) {
+  zval* retval = NULL;
+  switch (type) {
+    // TODO(teboring): Add it back.
+    // case UPB_TYPE_FLOAT:
+    //   return DBL2NUM(DEREF(memory, float));
+    // case UPB_TYPE_DOUBLE:
+    //   return DBL2NUM(DEREF(memory, double));
+    // case UPB_TYPE_BOOL:
+    //   return DEREF(memory, int8_t) ? Qtrue : Qfalse;
+    // case UPB_TYPE_STRING:
+    // case UPB_TYPE_BYTES:
+    // case UPB_TYPE_MESSAGE:
+    //   return DEREF(memory, VALUE);
+    // case UPB_TYPE_ENUM: {
+    //   int32_t val = DEREF(memory, int32_t);
+    //   VALUE symbol = enum_lookup(type_class, INT2NUM(val));
+    //   if (symbol == Qnil) {
+    //     return INT2NUM(val);
+    //   } else {
+    //     return symbol;
+    //   }
+    // }
+    case UPB_TYPE_INT32:
+      return int32_to_zval(DEREF(memory, int32_t));
+    // TODO(teboring): Add it back.
+    // case UPB_TYPE_INT64:
+    //   return LL2NUM(DEREF(memory, int64_t));
+    // case UPB_TYPE_UINT32:
+    //   return UINT2NUM(DEREF(memory, uint32_t));
+    // case UPB_TYPE_UINT64:
+    //   return ULL2NUM(DEREF(memory, uint64_t));
+    default:
+      return EG(uninitialized_zval_ptr);
+  }
+}
+
+void native_slot_init(upb_fieldtype_t type, void* memory) {
+  switch (type) {
+    case UPB_TYPE_FLOAT:
+      DEREF(memory, float) = 0.0;
+      break;
+    case UPB_TYPE_DOUBLE:
+      DEREF(memory, double) = 0.0;
+      break;
+    case UPB_TYPE_BOOL:
+      DEREF(memory, int8_t) = 0;
+      break;
+    // TODO(teboring): Add it back.
+    // case UPB_TYPE_STRING:
+    // case UPB_TYPE_BYTES:
+    //   DEREF(memory, VALUE) = php_str_new2("");
+    //   php_enc_associate(DEREF(memory, VALUE), (type == UPB_TYPE_BYTES)
+    //                                              ? kRubyString8bitEncoding
+    //                                              : kRubyStringUtf8Encoding);
+    //   break;
+    // case UPB_TYPE_MESSAGE:
+    //   DEREF(memory, VALUE) = Qnil;
+    //   break;
+    case UPB_TYPE_ENUM:
+    case UPB_TYPE_INT32:
+      DEREF(memory, int32_t) = 0;
+      break;
+    case UPB_TYPE_INT64:
+      DEREF(memory, int64_t) = 0;
+      break;
+    case UPB_TYPE_UINT32:
+      DEREF(memory, uint32_t) = 0;
+      break;
+    case UPB_TYPE_UINT64:
+      DEREF(memory, uint64_t) = 0;
+      break;
+    default:
+      break;
+  }
+}
+
+void native_slot_set(upb_fieldtype_t type, /*VALUE type_class,*/ void* memory,
+                     zval* value) {
+  native_slot_set_value_and_case(type, /*type_class,*/ memory, value, NULL, 0);
+}
+
+void native_slot_set_value_and_case(upb_fieldtype_t type, /*VALUE type_class,*/
+                                    void* memory, zval* value,
+                                    uint32_t* case_memory,
+                                    uint32_t case_number) {
+  switch (type) {
+    case UPB_TYPE_FLOAT:
+      if (!Z_TYPE_P(value) == IS_LONG) {
+        zend_error(E_ERROR, "Expected number type for float field.");
+      }
+      DEREF(memory, float) = Z_DVAL_P(value);
+      break;
+    case UPB_TYPE_DOUBLE:
+      // TODO(teboring): Add it back.
+      // if (!is_php_num(value)) {
+      //   zend_error(E_ERROR, "Expected number type for double field.");
+      // }
+      // DEREF(memory, double) = Z_DVAL_P(value);
+      break;
+    case UPB_TYPE_BOOL: {
+      int8_t val = -1;
+      if (zval_is_true(value)) {
+        val = 1;
+      } else {
+        val = 0;
+      }
+      // TODO(teboring): Add it back.
+      // else if (value == Qfalse) {
+      //   val = 0;
+      // }
+      // else {
+      //   php_raise(php_eTypeError, "Invalid argument for boolean field.");
+      // }
+      DEREF(memory, int8_t) = val;
+      break;
+    }
+    case UPB_TYPE_STRING:
+    case UPB_TYPE_BYTES: {
+      // TODO(teboring): Add it back.
+      // if (Z_TYPE_P(value) != IS_STRING) {
+      //   zend_error(E_ERROR, "Invalid argument for string field.");
+      // }
+      // native_slot_validate_string_encoding(type, value);
+      // DEREF(memory, zval*) = value;
+      break;
+    }
+    case UPB_TYPE_MESSAGE: {
+      // TODO(teboring): Add it back.
+      // if (CLASS_OF(value) == CLASS_OF(Qnil)) {
+      //   value = Qnil;
+      // } else if (CLASS_OF(value) != type_class) {
+      //   php_raise(php_eTypeError,
+      //            "Invalid type %s to assign to submessage field.",
+      //            php_class2name(CLASS_OF(value)));
+      // }
+      // DEREF(memory, VALUE) = value;
+      break;
+    }
+    case UPB_TYPE_ENUM: {
+      // TODO(teboring): Add it back.
+      // int32_t int_val = 0;
+      // if (!is_php_num(value) && TYPE(value) != T_SYMBOL) {
+      //   php_raise(php_eTypeError,
+      //            "Expected number or symbol type for enum field.");
+      // }
+      // if (TYPE(value) == T_SYMBOL) {
+      //   // Ensure that the given symbol exists in the enum module.
+      //   VALUE lookup = php_funcall(type_class, php_intern("resolve"), 1, value);
+      //   if (lookup == Qnil) {
+      //     php_raise(php_eRangeError, "Unknown symbol value for enum field.");
+      //   } else {
+      //     int_val = NUM2INT(lookup);
+      //   }
+      // } else {
+      //   native_slot_check_int_range_precision(UPB_TYPE_INT32, value);
+      //   int_val = NUM2INT(value);
+      // }
+      // DEREF(memory, int32_t) = int_val;
+      // break;
+    }
+    case UPB_TYPE_INT32:
+    case UPB_TYPE_INT64:
+    case UPB_TYPE_UINT32:
+    case UPB_TYPE_UINT64:
+      native_slot_check_int_range_precision(type, value);
+      switch (type) {
+        case UPB_TYPE_INT32:
+          php_printf("Setting INT32 field\n");
+          DEREF(memory, int32_t) = Z_LVAL_P(value);
+          break;
+        case UPB_TYPE_INT64:
+          // TODO(teboring): Add it back.
+          // DEREF(memory, int64_t) = NUM2LL(value);
+          break;
+        case UPB_TYPE_UINT32:
+          // TODO(teboring): Add it back.
+          // DEREF(memory, uint32_t) = NUM2UINT(value);
+          break;
+        case UPB_TYPE_UINT64:
+          // TODO(teboring): Add it back.
+          // DEREF(memory, uint64_t) = NUM2ULL(value);
+          break;
+        default:
+          break;
+      }
+      break;
+    default:
+      break;
+  }
+
+  if (case_memory != NULL) {
+    *case_memory = case_number;
+  }
+}
+
+// -----------------------------------------------------------------------------
+// Map field utilities.
+// ----------------------------------------------------------------------------
+
+const upb_msgdef* tryget_map_entry_msgdef(const upb_fielddef* field) {
+  const upb_msgdef* subdef;
+  if (upb_fielddef_label(field) != UPB_LABEL_REPEATED ||
+      upb_fielddef_type(field) != UPB_TYPE_MESSAGE) {
+    return NULL;
+  }
+  subdef = upb_fielddef_msgsubdef(field);
+  return upb_msgdef_mapentry(subdef) ? subdef : NULL;
+}
+
+const upb_msgdef* map_entry_msgdef(const upb_fielddef* field) {
+  const upb_msgdef* subdef = tryget_map_entry_msgdef(field);
+  assert(subdef);
+  return subdef;
+}
+
+bool is_map_field(const upb_fielddef* field) {
+  return tryget_map_entry_msgdef(field) != NULL;
+}
+
+// -----------------------------------------------------------------------------
+// Memory layout management.
+// -----------------------------------------------------------------------------
+
+static size_t align_up_to(size_t offset, size_t granularity) {
+  // Granularity must be a power of two.
+  return (offset + granularity - 1) & ~(granularity - 1);
+}
+
+MessageLayout* create_layout(const upb_msgdef* msgdef) {
+  MessageLayout* layout = ALLOC(MessageLayout);
+  int nfields = upb_msgdef_numfields(msgdef);
+  upb_msg_field_iter it;
+  upb_msg_oneof_iter oit;
+  size_t off = 0;
+
+  layout->fields = ALLOC_N(MessageField, nfields);
+
+  for (upb_msg_field_begin(&it, msgdef); !upb_msg_field_done(&it);
+       upb_msg_field_next(&it)) {
+    const upb_fielddef* field = upb_msg_iter_field(&it);
+    size_t field_size;
+
+    if (upb_fielddef_containingoneof(field)) {
+      // Oneofs are handled separately below.
+      continue;
+    }
+
+    // Allocate |field_size| bytes for this field in the layout.
+    field_size = 0;
+    if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
+      field_size = sizeof(zval*);
+    } else {
+      field_size = native_slot_size(upb_fielddef_type(field));
+    }
+
+    // Align current offset up to | size | granularity.
+    off = align_up_to(off, field_size);
+    layout->fields[upb_fielddef_index(field)].offset = off;
+    layout->fields[upb_fielddef_index(field)].case_offset =
+        MESSAGE_FIELD_NO_CASE;
+    off += field_size;
+  }
+
+  // Handle oneofs now -- we iterate over oneofs specifically and allocate only
+  // one slot per oneof.
+  //
+  // We assign all value slots first, then pack the 'case' fields at the end,
+  // since in the common case (modern 64-bit platform) these are 8 bytes and 4
+  // bytes respectively and we want to avoid alignment overhead.
+  //
+  // Note that we reserve 4 bytes (a uint32) per 'case' slot because the value
+  // space for oneof cases is conceptually as wide as field tag numbers.  In
+  // practice, it's unlikely that a oneof would have more than e.g.  256 or 64K
+  // members (8 or 16 bits respectively), so conceivably we could assign
+  // consecutive case numbers and then pick a smaller oneof case slot size, but
+  // the complexity to implement this indirection is probably not worthwhile.
+  for (upb_msg_oneof_begin(&oit, msgdef); !upb_msg_oneof_done(&oit);
+       upb_msg_oneof_next(&oit)) {
+    const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit);
+    upb_oneof_iter fit;
+
+    // Always allocate NATIVE_SLOT_MAX_SIZE bytes, but share the slot between
+    // all fields.
+    size_t field_size = NATIVE_SLOT_MAX_SIZE;
+    // Align the offset .
+    off = align_up_to( off, field_size);
+    // Assign all fields in the oneof this same offset.
+    for (upb_oneof_begin(&fit, oneof); !upb_oneof_done(&fit);
+         upb_oneof_next(&fit)) {
+      const upb_fielddef* field = upb_oneof_iter_field(&fit);
+      layout->fields[upb_fielddef_index(field)].offset = off;
+    }
+    off += field_size;
+  }
+
+  // Now the case fields.
+  for (upb_msg_oneof_begin(&oit, msgdef); !upb_msg_oneof_done(&oit);
+       upb_msg_oneof_next(&oit)) {
+    const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit);
+    upb_oneof_iter fit;
+
+    size_t field_size = sizeof(uint32_t);
+    // Align the offset .
+    off = (off + field_size - 1) & ~(field_size - 1);
+    // Assign all fields in the oneof this same offset.
+    for (upb_oneof_begin(&fit, oneof); !upb_oneof_done(&fit);
+         upb_oneof_next(&fit)) {
+      const upb_fielddef* field = upb_oneof_iter_field(&fit);
+      layout->fields[upb_fielddef_index(field)].case_offset = off;
+    }
+    off += field_size;
+  }
+
+  layout->size = off;
+
+  layout->msgdef = msgdef;
+  upb_msgdef_ref(layout->msgdef, &layout->msgdef);
+
+  return layout;
+}
+
+void free_layout(MessageLayout* layout) {
+  FREE(layout->fields);
+  upb_msgdef_unref(layout->msgdef, &layout->msgdef);
+  FREE(layout);
+}
+
+// TODO(teboring): Add it back.
+// VALUE field_type_class(const upb_fielddef* field) {
+//   VALUE type_class = Qnil;
+//   if (upb_fielddef_type(field) == UPB_TYPE_MESSAGE) {
+//     VALUE submsgdesc = get_def_obj(upb_fielddef_subdef(field));
+//     type_class = Descriptor_msgclass(submsgdesc);
+//   } else if (upb_fielddef_type(field) == UPB_TYPE_ENUM) {
+//     VALUE subenumdesc = get_def_obj(upb_fielddef_subdef(field));
+//     type_class = EnumDescriptor_enummodule(subenumdesc);
+//   }
+//   return type_class;
+// }
+
+static void* slot_memory(MessageLayout* layout, const void* storage,
+                         const upb_fielddef* field) {
+  return ((uint8_t*)storage) + layout->fields[upb_fielddef_index(field)].offset;
+}
+
+static uint32_t* slot_oneof_case(MessageLayout* layout, const void* storage,
+                                 const upb_fielddef* field) {
+  return (uint32_t*)(((uint8_t*)storage) +
+                     layout->fields[upb_fielddef_index(field)].case_offset);
+}
+
+void layout_set(MessageLayout* layout, void* storage, const upb_fielddef* field,
+                zval* val) {
+  void* memory = slot_memory(layout, storage, field);
+  uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
+
+  if (upb_fielddef_containingoneof(field)) {
+    if (Z_TYPE_P(val) == IS_NULL) {
+      // Assigning nil to a oneof field clears the oneof completely.
+      *oneof_case = ONEOF_CASE_NONE;
+      memset(memory, 0, NATIVE_SLOT_MAX_SIZE);
+    } else {
+      // The transition between field types for a single oneof (union) slot is
+      // somewhat complex because we need to ensure that a GC triggered at any
+      // point by a call into the Ruby VM sees a valid state for this field and
+      // does not either go off into the weeds (following what it thinks is a
+      // VALUE but is actually a different field type) or miss an object (seeing
+      // what it thinks is a primitive field but is actually a VALUE for the new
+      // field type).
+      //
+      // In order for the transition to be safe, the oneof case slot must be in
+      // sync with the value slot whenever the Ruby VM has been called. Thus, we
+      // use native_slot_set_value_and_case(), which ensures that both the value
+      // and case number are altered atomically (w.r.t. the Ruby VM).
+      native_slot_set_value_and_case(upb_fielddef_type(field),
+                                     /*field_type_class(field),*/ memory, val,
+                                     oneof_case, upb_fielddef_number(field));
+    }
+  } else if (is_map_field(field)) {
+    // TODO(teboring): Add it back.
+    // check_map_field_type(val, field);
+    // DEREF(memory, zval*) = val;
+  } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
+    // TODO(teboring): Add it back.
+    // check_repeated_field_type(val, field);
+    // DEREF(memory, zval*) = val;
+  } else {
+    native_slot_set(upb_fielddef_type(field), /*field_type_class(field),*/ memory,
+                    val);
+  }
+}
+
+void layout_init(MessageLayout* layout, void* storage) {
+  upb_msg_field_iter it;
+  for (upb_msg_field_begin(&it, layout->msgdef); !upb_msg_field_done(&it);
+       upb_msg_field_next(&it)) {
+    const upb_fielddef* field = upb_msg_iter_field(&it);
+    void* memory = slot_memory(layout, storage, field);
+    uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
+
+    if (upb_fielddef_containingoneof(field)) {
+      // TODO(teboring): Add it back.
+      // memset(memory, 0, NATIVE_SLOT_MAX_SIZE);
+      // *oneof_case = ONEOF_CASE_NONE;
+    } else if (is_map_field(field)) {
+      // TODO(teboring): Add it back.
+      // VALUE map = Qnil;
+
+      // const upb_fielddef* key_field = map_field_key(field);
+      // const upb_fielddef* value_field = map_field_value(field);
+      // VALUE type_class = field_type_class(value_field);
+
+      // if (type_class != Qnil) {
+      //   VALUE args[3] = {
+      //       fieldtype_to_php(upb_fielddef_type(key_field)),
+      //       fieldtype_to_php(upb_fielddef_type(value_field)), type_class,
+      //   };
+      //   map = php_class_new_instance(3, args, cMap);
+      // } else {
+      //   VALUE args[2] = {
+      //       fieldtype_to_php(upb_fielddef_type(key_field)),
+      //       fieldtype_to_php(upb_fielddef_type(value_field)),
+      //   };
+      //   map = php_class_new_instance(2, args, cMap);
+      // }
+
+      // DEREF(memory, VALUE) = map;
+    } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
+      // TODO(teboring): Add it back.
+      // VALUE ary = Qnil;
+
+      // VALUE type_class = field_type_class(field);
+
+      // if (type_class != Qnil) {
+      //   VALUE args[2] = {
+      //       fieldtype_to_php(upb_fielddef_type(field)), type_class,
+      //   };
+      //   ary = php_class_new_instance(2, args, cRepeatedField);
+      // } else {
+      //   VALUE args[1] = {fieldtype_to_php(upb_fielddef_type(field))};
+      //   ary = php_class_new_instance(1, args, cRepeatedField);
+      // }
+
+      // DEREF(memory, VALUE) = ary;
+    } else {
+      native_slot_init(upb_fielddef_type(field), memory);
+    }
+  }
+}
+
+zval* layout_get(MessageLayout* layout, const void* storage,
+                 const upb_fielddef* field TSRMLS_DC) {
+  void* memory = slot_memory(layout, storage, field);
+  uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
+
+  if (upb_fielddef_containingoneof(field)) {
+    if (*oneof_case != upb_fielddef_number(field)) {
+      return NULL;
+      // TODO(teboring): Add it back.
+      // return Qnil;
+    }
+      return NULL;
+    // TODO(teboring): Add it back.
+    // return native_slot_get(upb_fielddef_type(field), field_type_class(field),
+    //                        memory);
+  } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
+      return NULL;
+    // TODO(teboring): Add it back.
+    // return *((VALUE*)memory);
+  } else {
+    return native_slot_get(
+        upb_fielddef_type(field), /*field_type_class(field), */
+        memory TSRMLS_CC);
+  }
+}
diff --git a/php/ext/google/protobuf/test.php b/php/ext/google/protobuf/test.php
new file mode 100644
index 00000000..6bbadd48
--- /dev/null
+++ b/php/ext/google/protobuf/test.php
@@ -0,0 +1,15 @@
+<?php
+
+
+namespace Google\Protobuf;
+
+$pool = get_generated_pool();
+$pool->addMessage("TestMessage")
+    ->optional("optional_int32_a", "int32", 1)
+    ->optional("optional_int32_b", "int32", 2)
+    ->finalizeToPool()
+    ->finalize();
+
+$test_message = new \TestMessage();
+
+?>
diff --git a/php/ext/google/protobuf/upb.c b/php/ext/google/protobuf/upb.c
new file mode 100644
index 00000000..048a163a
--- /dev/null
+++ b/php/ext/google/protobuf/upb.c
@@ -0,0 +1,11990 @@
+// Amalgamated source file
+#include "upb.h"
+
+
+#include <stdlib.h>
+#include <string.h>
+
+typedef struct {
+  size_t len;
+  char str[1];  /* Null-terminated string data follows. */
+} str_t;
+
+static str_t *newstr(const char *data, size_t len) {
+  str_t *ret = malloc(sizeof(*ret) + len);
+  if (!ret) return NULL;
+  ret->len = len;
+  memcpy(ret->str, data, len);
+  ret->str[len] = '\0';
+  return ret;
+}
+
+static void freestr(str_t *s) { free(s); }
+
+/* isalpha() etc. from <ctype.h> are locale-dependent, which we don't want. */
+static bool upb_isbetween(char c, char low, char high) {
+  return c >= low && c <= high;
+}
+
+static bool upb_isletter(char c) {
+  return upb_isbetween(c, 'A', 'Z') || upb_isbetween(c, 'a', 'z') || c == '_';
+}
+
+static bool upb_isalphanum(char c) {
+  return upb_isletter(c) || upb_isbetween(c, '0', '9');
+}
+
+static bool upb_isident(const char *str, size_t len, bool full, upb_status *s) {
+  bool start = true;
+  size_t i;
+  for (i = 0; i < len; i++) {
+    char c = str[i];
+    if (c == '.') {
+      if (start || !full) {
+        upb_status_seterrf(s, "invalid name: unexpected '.' (%s)", str);
+        return false;
+      }
+      start = true;
+    } else if (start) {
+      if (!upb_isletter(c)) {
+        upb_status_seterrf(
+            s, "invalid name: path components must start with a letter (%s)",
+            str);
+        return false;
+      }
+      start = false;
+    } else {
+      if (!upb_isalphanum(c)) {
+        upb_status_seterrf(s, "invalid name: non-alphanumeric character (%s)",
+                           str);
+        return false;
+      }
+    }
+  }
+  return !start;
+}
+
+
+/* upb_def ********************************************************************/
+
+upb_deftype_t upb_def_type(const upb_def *d) { return d->type; }
+
+const char *upb_def_fullname(const upb_def *d) { return d->fullname; }
+
+bool upb_def_setfullname(upb_def *def, const char *fullname, upb_status *s) {
+  assert(!upb_def_isfrozen(def));
+  if (!upb_isident(fullname, strlen(fullname), true, s)) return false;
+  free((void*)def->fullname);
+  def->fullname = upb_strdup(fullname);
+  return true;
+}
+
+upb_def *upb_def_dup(const upb_def *def, const void *o) {
+  switch (def->type) {
+    case UPB_DEF_MSG:
+      return upb_msgdef_upcast_mutable(
+          upb_msgdef_dup(upb_downcast_msgdef(def), o));
+    case UPB_DEF_FIELD:
+      return upb_fielddef_upcast_mutable(
+          upb_fielddef_dup(upb_downcast_fielddef(def), o));
+    case UPB_DEF_ENUM:
+      return upb_enumdef_upcast_mutable(
+          upb_enumdef_dup(upb_downcast_enumdef(def), o));
+    default: assert(false); return NULL;
+  }
+}
+
+static bool upb_def_init(upb_def *def, upb_deftype_t type,
+                         const struct upb_refcounted_vtbl *vtbl,
+                         const void *owner) {
+  if (!upb_refcounted_init(upb_def_upcast_mutable(def), vtbl, owner)) return false;
+  def->type = type;
+  def->fullname = NULL;
+  def->came_from_user = false;
+  return true;
+}
+
+static void upb_def_uninit(upb_def *def) {
+  free((void*)def->fullname);
+}
+
+static const char *msgdef_name(const upb_msgdef *m) {
+  const char *name = upb_def_fullname(upb_msgdef_upcast(m));
+  return name ? name : "(anonymous)";
+}
+
+static bool upb_validate_field(upb_fielddef *f, upb_status *s) {
+  if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) {
+    upb_status_seterrmsg(s, "fielddef must have name and number set");
+    return false;
+  }
+
+  if (!f->type_is_set_) {
+    upb_status_seterrmsg(s, "fielddef type was not initialized");
+    return false;
+  }
+
+  if (upb_fielddef_lazy(f) &&
+      upb_fielddef_descriptortype(f) != UPB_DESCRIPTOR_TYPE_MESSAGE) {
+    upb_status_seterrmsg(s,
+                         "only length-delimited submessage fields may be lazy");
+    return false;
+  }
+
+  if (upb_fielddef_hassubdef(f)) {
+    const upb_def *subdef;
+
+    if (f->subdef_is_symbolic) {
+      upb_status_seterrf(s, "field '%s.%s' has not been resolved",
+                         msgdef_name(f->msg.def), upb_fielddef_name(f));
+      return false;
+    }
+
+    subdef = upb_fielddef_subdef(f);
+    if (subdef == NULL) {
+      upb_status_seterrf(s, "field %s.%s is missing required subdef",
+                         msgdef_name(f->msg.def), upb_fielddef_name(f));
+      return false;
+    }
+
+    if (!upb_def_isfrozen(subdef) && !subdef->came_from_user) {
+      upb_status_seterrf(s,
+                         "subdef of field %s.%s is not frozen or being frozen",
+                         msgdef_name(f->msg.def), upb_fielddef_name(f));
+      return false;
+    }
+  }
+
+  if (upb_fielddef_type(f) == UPB_TYPE_ENUM) {
+    bool has_default_name = upb_fielddef_enumhasdefaultstr(f);
+    bool has_default_number = upb_fielddef_enumhasdefaultint32(f);
+
+    /* Previously verified by upb_validate_enumdef(). */
+    assert(upb_enumdef_numvals(upb_fielddef_enumsubdef(f)) > 0);
+
+    /* We've already validated that we have an associated enumdef and that it
+     * has at least one member, so at least one of these should be true.
+     * Because if the user didn't set anything, we'll pick up the enum's
+     * default, but if the user *did* set something we should at least pick up
+     * the one they set (int32 or string). */
+    assert(has_default_name || has_default_number);
+
+    if (!has_default_name) {
+      upb_status_seterrf(s,
+                         "enum default for field %s.%s (%d) is not in the enum",
+                         msgdef_name(f->msg.def), upb_fielddef_name(f),
+                         upb_fielddef_defaultint32(f));
+      return false;
+    }
+
+    if (!has_default_number) {
+      upb_status_seterrf(s,
+                         "enum default for field %s.%s (%s) is not in the enum",
+                         msgdef_name(f->msg.def), upb_fielddef_name(f),
+                         upb_fielddef_defaultstr(f, NULL));
+      return false;
+    }
+
+    /* Lift the effective numeric default into the field's default slot, in case
+     * we were only getting it "by reference" from the enumdef. */
+    upb_fielddef_setdefaultint32(f, upb_fielddef_defaultint32(f));
+  }
+
+  /* Ensure that MapEntry submessages only appear as repeated fields, not
+   * optional/required (singular) fields. */
+  if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE &&
+      upb_fielddef_msgsubdef(f) != NULL) {
+    const upb_msgdef *subdef = upb_fielddef_msgsubdef(f);
+    if (upb_msgdef_mapentry(subdef) && !upb_fielddef_isseq(f)) {
+      upb_status_seterrf(s,
+                         "Field %s refers to mapentry message but is not "
+                         "a repeated field",
+                         upb_fielddef_name(f) ? upb_fielddef_name(f) :
+                         "(unnamed)");
+      return false;
+    }
+  }
+
+  return true;
+}
+
+static bool upb_validate_enumdef(const upb_enumdef *e, upb_status *s) {
+  if (upb_enumdef_numvals(e) == 0) {
+    upb_status_seterrf(s, "enum %s has no members (must have at least one)",
+                       upb_enumdef_fullname(e));
+    return false;
+  }
+
+  return true;
+}
+
+/* All submessage fields are lower than all other fields.
+ * Secondly, fields are increasing in order. */
+uint32_t field_rank(const upb_fielddef *f) {
+  uint32_t ret = upb_fielddef_number(f);
+  const uint32_t high_bit = 1 << 30;
+  assert(ret < high_bit);
+  if (!upb_fielddef_issubmsg(f))
+    ret |= high_bit;
+  return ret;
+}
+
+int cmp_fields(const void *p1, const void *p2) {
+  const upb_fielddef *f1 = *(upb_fielddef*const*)p1;
+  const upb_fielddef *f2 = *(upb_fielddef*const*)p2;
+  return field_rank(f1) - field_rank(f2);
+}
+
+static bool assign_msg_indices(upb_msgdef *m, upb_status *s) {
+  /* Sort fields.  upb internally relies on UPB_TYPE_MESSAGE fields having the
+   * lowest indexes, but we do not publicly guarantee this. */
+  upb_msg_field_iter j;
+  int i;
+  uint32_t selector;
+  int n = upb_msgdef_numfields(m);
+  upb_fielddef **fields = malloc(n * sizeof(*fields));
+  if (!fields) return false;
+
+  m->submsg_field_count = 0;
+  for(i = 0, upb_msg_field_begin(&j, m);
+      !upb_msg_field_done(&j);
+      upb_msg_field_next(&j), i++) {
+    upb_fielddef *f = upb_msg_iter_field(&j);
+    assert(f->msg.def == m);
+    if (!upb_validate_field(f, s)) {
+      free(fields);
+      return false;
+    }
+    if (upb_fielddef_issubmsg(f)) {
+      m->submsg_field_count++;
+    }
+    fields[i] = f;
+  }
+
+  qsort(fields, n, sizeof(*fields), cmp_fields);
+
+  selector = UPB_STATIC_SELECTOR_COUNT + m->submsg_field_count;
+  for (i = 0; i < n; i++) {
+    upb_fielddef *f = fields[i];
+    f->index_ = i;
+    f->selector_base = selector + upb_handlers_selectorbaseoffset(f);
+    selector += upb_handlers_selectorcount(f);
+  }
+  m->selector_count = selector;
+
+#ifndef NDEBUG
+  {
+    /* Verify that all selectors for the message are distinct. */
+#define TRY(type) \
+    if (upb_handlers_getselector(f, type, &sel)) upb_inttable_insert(&t, sel, v);
+
+    upb_inttable t;
+    upb_value v;
+    upb_selector_t sel;
+
+    upb_inttable_init(&t, UPB_CTYPE_BOOL);
+    v = upb_value_bool(true);
+    upb_inttable_insert(&t, UPB_STARTMSG_SELECTOR, v);
+    upb_inttable_insert(&t, UPB_ENDMSG_SELECTOR, v);
+    for(upb_msg_field_begin(&j, m);
+        !upb_msg_field_done(&j);
+        upb_msg_field_next(&j)) {
+      upb_fielddef *f = upb_msg_iter_field(&j);
+      /* These calls will assert-fail in upb_table if the value already
+       * exists. */
+      TRY(UPB_HANDLER_INT32);
+      TRY(UPB_HANDLER_INT64)
+      TRY(UPB_HANDLER_UINT32)
+      TRY(UPB_HANDLER_UINT64)
+      TRY(UPB_HANDLER_FLOAT)
+      TRY(UPB_HANDLER_DOUBLE)
+      TRY(UPB_HANDLER_BOOL)
+      TRY(UPB_HANDLER_STARTSTR)
+      TRY(UPB_HANDLER_STRING)
+      TRY(UPB_HANDLER_ENDSTR)
+      TRY(UPB_HANDLER_STARTSUBMSG)
+      TRY(UPB_HANDLER_ENDSUBMSG)
+      TRY(UPB_HANDLER_STARTSEQ)
+      TRY(UPB_HANDLER_ENDSEQ)
+    }
+    upb_inttable_uninit(&t);
+  }
+#undef TRY
+#endif
+
+  free(fields);
+  return true;
+}
+
+bool upb_def_freeze(upb_def *const* defs, int n, upb_status *s) {
+  int i;
+  int maxdepth;
+  bool ret;
+  upb_status_clear(s);
+
+  /* First perform validation, in two passes so we can check that we have a
+   * transitive closure without needing to search. */
+  for (i = 0; i < n; i++) {
+    upb_def *def = defs[i];
+    if (upb_def_isfrozen(def)) {
+      /* Could relax this requirement if it's annoying. */
+      upb_status_seterrmsg(s, "def is already frozen");
+      goto err;
+    } else if (def->type == UPB_DEF_FIELD) {
+      upb_status_seterrmsg(s, "standalone fielddefs can not be frozen");
+      goto err;
+    } else if (def->type == UPB_DEF_ENUM) {
+      if (!upb_validate_enumdef(upb_dyncast_enumdef(def), s)) {
+        goto err;
+      }
+    } else {
+      /* Set now to detect transitive closure in the second pass. */
+      def->came_from_user = true;
+    }
+  }
+
+  /* Second pass of validation.  Also assign selector bases and indexes, and
+   * compact tables. */
+  for (i = 0; i < n; i++) {
+    upb_msgdef *m = upb_dyncast_msgdef_mutable(defs[i]);
+    upb_enumdef *e = upb_dyncast_enumdef_mutable(defs[i]);
+    if (m) {
+      upb_inttable_compact(&m->itof);
+      if (!assign_msg_indices(m, s)) {
+        goto err;
+      }
+    } else if (e) {
+      upb_inttable_compact(&e->iton);
+    }
+  }
+
+  /* Def graph contains FieldDefs between each MessageDef, so double the
+   * limit. */
+  maxdepth = UPB_MAX_MESSAGE_DEPTH * 2;
+
+  /* Validation all passed; freeze the defs. */
+  ret = upb_refcounted_freeze((upb_refcounted * const *)defs, n, s, maxdepth);
+  assert(!(s && ret != upb_ok(s)));
+  return ret;
+
+err:
+  for (i = 0; i < n; i++) {
+    defs[i]->came_from_user = false;
+  }
+  assert(!(s && upb_ok(s)));
+  return false;
+}
+
+
+/* upb_enumdef ****************************************************************/
+
+static void upb_enumdef_free(upb_refcounted *r) {
+  upb_enumdef *e = (upb_enumdef*)r;
+  upb_inttable_iter i;
+  upb_inttable_begin(&i, &e->iton);
+  for( ; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    /* To clean up the upb_strdup() from upb_enumdef_addval(). */
+    free(upb_value_getcstr(upb_inttable_iter_value(&i)));
+  }
+  upb_strtable_uninit(&e->ntoi);
+  upb_inttable_uninit(&e->iton);
+  upb_def_uninit(upb_enumdef_upcast_mutable(e));
+  free(e);
+}
+
+upb_enumdef *upb_enumdef_new(const void *owner) {
+  static const struct upb_refcounted_vtbl vtbl = {NULL, &upb_enumdef_free};
+  upb_enumdef *e = malloc(sizeof(*e));
+  if (!e) return NULL;
+  if (!upb_def_init(upb_enumdef_upcast_mutable(e), UPB_DEF_ENUM, &vtbl, owner))
+    goto err2;
+  if (!upb_strtable_init(&e->ntoi, UPB_CTYPE_INT32)) goto err2;
+  if (!upb_inttable_init(&e->iton, UPB_CTYPE_CSTR)) goto err1;
+  return e;
+
+err1:
+  upb_strtable_uninit(&e->ntoi);
+err2:
+  free(e);
+  return NULL;
+}
+
+upb_enumdef *upb_enumdef_dup(const upb_enumdef *e, const void *owner) {
+  upb_enum_iter i;
+  upb_enumdef *new_e = upb_enumdef_new(owner);
+  if (!new_e) return NULL;
+  for(upb_enum_begin(&i, e); !upb_enum_done(&i); upb_enum_next(&i)) {
+    bool success = upb_enumdef_addval(
+        new_e, upb_enum_iter_name(&i),upb_enum_iter_number(&i), NULL);
+    if (!success) {
+      upb_enumdef_unref(new_e, owner);
+      return NULL;
+    }
+  }
+  return new_e;
+}
+
+bool upb_enumdef_freeze(upb_enumdef *e, upb_status *status) {
+  upb_def *d = upb_enumdef_upcast_mutable(e);
+  return upb_def_freeze(&d, 1, status);
+}
+
+const char *upb_enumdef_fullname(const upb_enumdef *e) {
+  return upb_def_fullname(upb_enumdef_upcast(e));
+}
+
+bool upb_enumdef_setfullname(upb_enumdef *e, const char *fullname,
+                             upb_status *s) {
+  return upb_def_setfullname(upb_enumdef_upcast_mutable(e), fullname, s);
+}
+
+bool upb_enumdef_addval(upb_enumdef *e, const char *name, int32_t num,
+                        upb_status *status) {
+  if (!upb_isident(name, strlen(name), false, status)) {
+    return false;
+  }
+  if (upb_enumdef_ntoiz(e, name, NULL)) {
+    upb_status_seterrf(status, "name '%s' is already defined", name);
+    return false;
+  }
+  if (!upb_strtable_insert(&e->ntoi, name, upb_value_int32(num))) {
+    upb_status_seterrmsg(status, "out of memory");
+    return false;
+  }
+  if (!upb_inttable_lookup(&e->iton, num, NULL) &&
+      !upb_inttable_insert(&e->iton, num, upb_value_cstr(upb_strdup(name)))) {
+    upb_status_seterrmsg(status, "out of memory");
+    upb_strtable_remove(&e->ntoi, name, NULL);
+    return false;
+  }
+  if (upb_enumdef_numvals(e) == 1) {
+    bool ok = upb_enumdef_setdefault(e, num, NULL);
+    UPB_ASSERT_VAR(ok, ok);
+  }
+  return true;
+}
+
+int32_t upb_enumdef_default(const upb_enumdef *e) {
+  assert(upb_enumdef_iton(e, e->defaultval));
+  return e->defaultval;
+}
+
+bool upb_enumdef_setdefault(upb_enumdef *e, int32_t val, upb_status *s) {
+  assert(!upb_enumdef_isfrozen(e));
+  if (!upb_enumdef_iton(e, val)) {
+    upb_status_seterrf(s, "number '%d' is not in the enum.", val);
+    return false;
+  }
+  e->defaultval = val;
+  return true;
+}
+
+int upb_enumdef_numvals(const upb_enumdef *e) {
+  return upb_strtable_count(&e->ntoi);
+}
+
+void upb_enum_begin(upb_enum_iter *i, const upb_enumdef *e) {
+  /* We iterate over the ntoi table, to account for duplicate numbers. */
+  upb_strtable_begin(i, &e->ntoi);
+}
+
+void upb_enum_next(upb_enum_iter *iter) { upb_strtable_next(iter); }
+bool upb_enum_done(upb_enum_iter *iter) { return upb_strtable_done(iter); }
+
+bool upb_enumdef_ntoi(const upb_enumdef *def, const char *name,
+                      size_t len, int32_t *num) {
+  upb_value v;
+  if (!upb_strtable_lookup2(&def->ntoi, name, len, &v)) {
+    return false;
+  }
+  if (num) *num = upb_value_getint32(v);
+  return true;
+}
+
+const char *upb_enumdef_iton(const upb_enumdef *def, int32_t num) {
+  upb_value v;
+  return upb_inttable_lookup32(&def->iton, num, &v) ?
+      upb_value_getcstr(v) : NULL;
+}
+
+const char *upb_enum_iter_name(upb_enum_iter *iter) {
+  return upb_strtable_iter_key(iter);
+}
+
+int32_t upb_enum_iter_number(upb_enum_iter *iter) {
+  return upb_value_getint32(upb_strtable_iter_value(iter));
+}
+
+
+/* upb_fielddef ***************************************************************/
+
+static void upb_fielddef_init_default(upb_fielddef *f);
+
+static void upb_fielddef_uninit_default(upb_fielddef *f) {
+  if (f->type_is_set_ && f->default_is_string && f->defaultval.bytes)
+    freestr(f->defaultval.bytes);
+}
+
+static void visitfield(const upb_refcounted *r, upb_refcounted_visit *visit,
+                       void *closure) {
+  const upb_fielddef *f = (const upb_fielddef*)r;
+  if (upb_fielddef_containingtype(f)) {
+    visit(r, upb_msgdef_upcast2(upb_fielddef_containingtype(f)), closure);
+  }
+  if (upb_fielddef_containingoneof(f)) {
+    visit(r, upb_oneofdef_upcast2(upb_fielddef_containingoneof(f)), closure);
+  }
+  if (upb_fielddef_subdef(f)) {
+    visit(r, upb_def_upcast(upb_fielddef_subdef(f)), closure);
+  }
+}
+
+static void freefield(upb_refcounted *r) {
+  upb_fielddef *f = (upb_fielddef*)r;
+  upb_fielddef_uninit_default(f);
+  if (f->subdef_is_symbolic)
+    free(f->sub.name);
+  upb_def_uninit(upb_fielddef_upcast_mutable(f));
+  free(f);
+}
+
+static const char *enumdefaultstr(const upb_fielddef *f) {
+  const upb_enumdef *e;
+  assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM);
+  e = upb_fielddef_enumsubdef(f);
+  if (f->default_is_string && f->defaultval.bytes) {
+    /* Default was explicitly set as a string. */
+    str_t *s = f->defaultval.bytes;
+    return s->str;
+  } else if (e) {
+    if (!f->default_is_string) {
+      /* Default was explicitly set as an integer; look it up in enumdef. */
+      const char *name = upb_enumdef_iton(e, f->defaultval.sint);
+      if (name) {
+        return name;
+      }
+    } else {
+      /* Default is completely unset; pull enumdef default. */
+      if (upb_enumdef_numvals(e) > 0) {
+        const char *name = upb_enumdef_iton(e, upb_enumdef_default(e));
+        assert(name);
+        return name;
+      }
+    }
+  }
+  return NULL;
+}
+
+static bool enumdefaultint32(const upb_fielddef *f, int32_t *val) {
+  const upb_enumdef *e;
+  assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM);
+  e = upb_fielddef_enumsubdef(f);
+  if (!f->default_is_string) {
+    /* Default was explicitly set as an integer. */
+    *val = f->defaultval.sint;
+    return true;
+  } else if (e) {
+    if (f->defaultval.bytes) {
+      /* Default was explicitly set as a str; try to lookup corresponding int. */
+      str_t *s = f->defaultval.bytes;
+      if (upb_enumdef_ntoiz(e, s->str, val)) {
+        return true;
+      }
+    } else {
+      /* Default is unset; try to pull in enumdef default. */
+      if (upb_enumdef_numvals(e) > 0) {
+        *val = upb_enumdef_default(e);
+        return true;
+      }
+    }
+  }
+  return false;
+}
+
+upb_fielddef *upb_fielddef_new(const void *o) {
+  static const struct upb_refcounted_vtbl vtbl = {visitfield, freefield};
+  upb_fielddef *f = malloc(sizeof(*f));
+  if (!f) return NULL;
+  if (!upb_def_init(upb_fielddef_upcast_mutable(f), UPB_DEF_FIELD, &vtbl, o)) {
+    free(f);
+    return NULL;
+  }
+  f->msg.def = NULL;
+  f->sub.def = NULL;
+  f->oneof = NULL;
+  f->subdef_is_symbolic = false;
+  f->msg_is_symbolic = false;
+  f->label_ = UPB_LABEL_OPTIONAL;
+  f->type_ = UPB_TYPE_INT32;
+  f->number_ = 0;
+  f->type_is_set_ = false;
+  f->tagdelim = false;
+  f->is_extension_ = false;
+  f->lazy_ = false;
+  f->packed_ = true;
+
+  /* For the moment we default this to UPB_INTFMT_VARIABLE, since it will work
+   * with all integer types and is in some since more "default" since the most
+   * normal-looking proto2 types int32/int64/uint32/uint64 use variable.
+   *
+   * Other options to consider:
+   * - there is no default; users must set this manually (like type).
+   * - default signed integers to UPB_INTFMT_ZIGZAG, since it's more likely to
+   *   be an optimal default for signed integers. */
+  f->intfmt = UPB_INTFMT_VARIABLE;
+  return f;
+}
+
+upb_fielddef *upb_fielddef_dup(const upb_fielddef *f, const void *owner) {
+  const char *srcname;
+  upb_fielddef *newf = upb_fielddef_new(owner);
+  if (!newf) return NULL;
+  upb_fielddef_settype(newf, upb_fielddef_type(f));
+  upb_fielddef_setlabel(newf, upb_fielddef_label(f));
+  upb_fielddef_setnumber(newf, upb_fielddef_number(f), NULL);
+  upb_fielddef_setname(newf, upb_fielddef_name(f), NULL);
+  if (f->default_is_string && f->defaultval.bytes) {
+    str_t *s = f->defaultval.bytes;
+    upb_fielddef_setdefaultstr(newf, s->str, s->len, NULL);
+  } else {
+    newf->default_is_string = f->default_is_string;
+    newf->defaultval = f->defaultval;
+  }
+
+  if (f->subdef_is_symbolic) {
+    srcname = f->sub.name;  /* Might be NULL. */
+  } else {
+    srcname = f->sub.def ? upb_def_fullname(f->sub.def) : NULL;
+  }
+  if (srcname) {
+    char *newname = malloc(strlen(f->sub.def->fullname) + 2);
+    if (!newname) {
+      upb_fielddef_unref(newf, owner);
+      return NULL;
+    }
+    strcpy(newname, ".");
+    strcat(newname, f->sub.def->fullname);
+    upb_fielddef_setsubdefname(newf, newname, NULL);
+    free(newname);
+  }
+
+  return newf;
+}
+
+bool upb_fielddef_typeisset(const upb_fielddef *f) {
+  return f->type_is_set_;
+}
+
+upb_fieldtype_t upb_fielddef_type(const upb_fielddef *f) {
+  assert(f->type_is_set_);
+  return f->type_;
+}
+
+uint32_t upb_fielddef_index(const upb_fielddef *f) {
+  return f->index_;
+}
+
+upb_label_t upb_fielddef_label(const upb_fielddef *f) {
+  return f->label_;
+}
+
+upb_intfmt_t upb_fielddef_intfmt(const upb_fielddef *f) {
+  return f->intfmt;
+}
+
+bool upb_fielddef_istagdelim(const upb_fielddef *f) {
+  return f->tagdelim;
+}
+
+uint32_t upb_fielddef_number(const upb_fielddef *f) {
+  return f->number_;
+}
+
+bool upb_fielddef_isextension(const upb_fielddef *f) {
+  return f->is_extension_;
+}
+
+bool upb_fielddef_lazy(const upb_fielddef *f) {
+  return f->lazy_;
+}
+
+bool upb_fielddef_packed(const upb_fielddef *f) {
+  return f->packed_;
+}
+
+const char *upb_fielddef_name(const upb_fielddef *f) {
+  return upb_def_fullname(upb_fielddef_upcast(f));
+}
+
+const upb_msgdef *upb_fielddef_containingtype(const upb_fielddef *f) {
+  return f->msg_is_symbolic ? NULL : f->msg.def;
+}
+
+const upb_oneofdef *upb_fielddef_containingoneof(const upb_fielddef *f) {
+  return f->oneof;
+}
+
+upb_msgdef *upb_fielddef_containingtype_mutable(upb_fielddef *f) {
+  return (upb_msgdef*)upb_fielddef_containingtype(f);
+}
+
+const char *upb_fielddef_containingtypename(upb_fielddef *f) {
+  return f->msg_is_symbolic ? f->msg.name : NULL;
+}
+
+static void release_containingtype(upb_fielddef *f) {
+  if (f->msg_is_symbolic) free(f->msg.name);
+}
+
+bool upb_fielddef_setcontainingtypename(upb_fielddef *f, const char *name,
+                                        upb_status *s) {
+  assert(!upb_fielddef_isfrozen(f));
+  if (upb_fielddef_containingtype(f)) {
+    upb_status_seterrmsg(s, "field has already been added to a message.");
+    return false;
+  }
+  /* TODO: validate name (upb_isident() doesn't quite work atm because this name
+   * may have a leading "."). */
+  release_containingtype(f);
+  f->msg.name = upb_strdup(name);
+  f->msg_is_symbolic = true;
+  return true;
+}
+
+bool upb_fielddef_setname(upb_fielddef *f, const char *name, upb_status *s) {
+  if (upb_fielddef_containingtype(f) || upb_fielddef_containingoneof(f)) {
+    upb_status_seterrmsg(s, "Already added to message or oneof");
+    return false;
+  }
+  return upb_def_setfullname(upb_fielddef_upcast_mutable(f), name, s);
+}
+
+static void chkdefaulttype(const upb_fielddef *f, upb_fieldtype_t type) {
+  UPB_UNUSED(f);
+  UPB_UNUSED(type);
+  assert(f->type_is_set_ && upb_fielddef_type(f) == type);
+}
+
+int64_t upb_fielddef_defaultint64(const upb_fielddef *f) {
+  chkdefaulttype(f, UPB_TYPE_INT64);
+  return f->defaultval.sint;
+}
+
+int32_t upb_fielddef_defaultint32(const upb_fielddef *f) {
+  if (f->type_is_set_ && upb_fielddef_type(f) == UPB_TYPE_ENUM) {
+    int32_t val;
+    bool ok = enumdefaultint32(f, &val);
+    UPB_ASSERT_VAR(ok, ok);
+    return val;
+  } else {
+    chkdefaulttype(f, UPB_TYPE_INT32);
+    return f->defaultval.sint;
+  }
+}
+
+uint64_t upb_fielddef_defaultuint64(const upb_fielddef *f) {
+  chkdefaulttype(f, UPB_TYPE_UINT64);
+  return f->defaultval.uint;
+}
+
+uint32_t upb_fielddef_defaultuint32(const upb_fielddef *f) {
+  chkdefaulttype(f, UPB_TYPE_UINT32);
+  return f->defaultval.uint;
+}
+
+bool upb_fielddef_defaultbool(const upb_fielddef *f) {
+  chkdefaulttype(f, UPB_TYPE_BOOL);
+  return f->defaultval.uint;
+}
+
+float upb_fielddef_defaultfloat(const upb_fielddef *f) {
+  chkdefaulttype(f, UPB_TYPE_FLOAT);
+  return f->defaultval.flt;
+}
+
+double upb_fielddef_defaultdouble(const upb_fielddef *f) {
+  chkdefaulttype(f, UPB_TYPE_DOUBLE);
+  return f->defaultval.dbl;
+}
+
+const char *upb_fielddef_defaultstr(const upb_fielddef *f, size_t *len) {
+  assert(f->type_is_set_);
+  assert(upb_fielddef_type(f) == UPB_TYPE_STRING ||
+         upb_fielddef_type(f) == UPB_TYPE_BYTES ||
+         upb_fielddef_type(f) == UPB_TYPE_ENUM);
+
+  if (upb_fielddef_type(f) == UPB_TYPE_ENUM) {
+    const char *ret = enumdefaultstr(f);
+    assert(ret);
+    /* Enum defaults can't have embedded NULLs. */
+    if (len) *len = strlen(ret);
+    return ret;
+  }
+
+  if (f->default_is_string) {
+    str_t *str = f->defaultval.bytes;
+    if (len) *len = str->len;
+    return str->str;
+  }
+
+  return NULL;
+}
+
+static void upb_fielddef_init_default(upb_fielddef *f) {
+  f->default_is_string = false;
+  switch (upb_fielddef_type(f)) {
+    case UPB_TYPE_DOUBLE: f->defaultval.dbl = 0; break;
+    case UPB_TYPE_FLOAT: f->defaultval.flt = 0; break;
+    case UPB_TYPE_INT32:
+    case UPB_TYPE_INT64: f->defaultval.sint = 0; break;
+    case UPB_TYPE_UINT64:
+    case UPB_TYPE_UINT32:
+    case UPB_TYPE_BOOL: f->defaultval.uint = 0; break;
+    case UPB_TYPE_STRING:
+    case UPB_TYPE_BYTES:
+      f->defaultval.bytes = newstr("", 0);
+      f->default_is_string = true;
+      break;
+    case UPB_TYPE_MESSAGE: break;
+    case UPB_TYPE_ENUM:
+      /* This is our special sentinel that indicates "not set" for an enum. */
+      f->default_is_string = true;
+      f->defaultval.bytes = NULL;
+      break;
+  }
+}
+
+const upb_def *upb_fielddef_subdef(const upb_fielddef *f) {
+  return f->subdef_is_symbolic ? NULL : f->sub.def;
+}
+
+const upb_msgdef *upb_fielddef_msgsubdef(const upb_fielddef *f) {
+  const upb_def *def = upb_fielddef_subdef(f);
+  return def ? upb_dyncast_msgdef(def) : NULL;
+}
+
+const upb_enumdef *upb_fielddef_enumsubdef(const upb_fielddef *f) {
+  const upb_def *def = upb_fielddef_subdef(f);
+  return def ? upb_dyncast_enumdef(def) : NULL;
+}
+
+upb_def *upb_fielddef_subdef_mutable(upb_fielddef *f) {
+  return (upb_def*)upb_fielddef_subdef(f);
+}
+
+const char *upb_fielddef_subdefname(const upb_fielddef *f) {
+  if (f->subdef_is_symbolic) {
+    return f->sub.name;
+  } else if (f->sub.def) {
+    return upb_def_fullname(f->sub.def);
+  } else {
+    return NULL;
+  }
+}
+
+bool upb_fielddef_setnumber(upb_fielddef *f, uint32_t number, upb_status *s) {
+  if (upb_fielddef_containingtype(f)) {
+    upb_status_seterrmsg(
+        s, "cannot change field number after adding to a message");
+    return false;
+  }
+  if (number == 0 || number > UPB_MAX_FIELDNUMBER) {
+    upb_status_seterrf(s, "invalid field number (%u)", number);
+    return false;
+  }
+  f->number_ = number;
+  return true;
+}
+
+void upb_fielddef_settype(upb_fielddef *f, upb_fieldtype_t type) {
+  assert(!upb_fielddef_isfrozen(f));
+  assert(upb_fielddef_checktype(type));
+  upb_fielddef_uninit_default(f);
+  f->type_ = type;
+  f->type_is_set_ = true;
+  upb_fielddef_init_default(f);
+}
+
+void upb_fielddef_setdescriptortype(upb_fielddef *f, int type) {
+  assert(!upb_fielddef_isfrozen(f));
+  switch (type) {
+    case UPB_DESCRIPTOR_TYPE_DOUBLE:
+      upb_fielddef_settype(f, UPB_TYPE_DOUBLE);
+      break;
+    case UPB_DESCRIPTOR_TYPE_FLOAT:
+      upb_fielddef_settype(f, UPB_TYPE_FLOAT);
+      break;
+    case UPB_DESCRIPTOR_TYPE_INT64:
+    case UPB_DESCRIPTOR_TYPE_SFIXED64:
+    case UPB_DESCRIPTOR_TYPE_SINT64:
+      upb_fielddef_settype(f, UPB_TYPE_INT64);
+      break;
+    case UPB_DESCRIPTOR_TYPE_UINT64:
+    case UPB_DESCRIPTOR_TYPE_FIXED64:
+      upb_fielddef_settype(f, UPB_TYPE_UINT64);
+      break;
+    case UPB_DESCRIPTOR_TYPE_INT32:
+    case UPB_DESCRIPTOR_TYPE_SFIXED32:
+    case UPB_DESCRIPTOR_TYPE_SINT32:
+      upb_fielddef_settype(f, UPB_TYPE_INT32);
+      break;
+    case UPB_DESCRIPTOR_TYPE_UINT32:
+    case UPB_DESCRIPTOR_TYPE_FIXED32:
+      upb_fielddef_settype(f, UPB_TYPE_UINT32);
+      break;
+    case UPB_DESCRIPTOR_TYPE_BOOL:
+      upb_fielddef_settype(f, UPB_TYPE_BOOL);
+      break;
+    case UPB_DESCRIPTOR_TYPE_STRING:
+      upb_fielddef_settype(f, UPB_TYPE_STRING);
+      break;
+    case UPB_DESCRIPTOR_TYPE_BYTES:
+      upb_fielddef_settype(f, UPB_TYPE_BYTES);
+      break;
+    case UPB_DESCRIPTOR_TYPE_GROUP:
+    case UPB_DESCRIPTOR_TYPE_MESSAGE:
+      upb_fielddef_settype(f, UPB_TYPE_MESSAGE);
+      break;
+    case UPB_DESCRIPTOR_TYPE_ENUM:
+      upb_fielddef_settype(f, UPB_TYPE_ENUM);
+      break;
+    default: assert(false);
+  }
+
+  if (type == UPB_DESCRIPTOR_TYPE_FIXED64 ||
+      type == UPB_DESCRIPTOR_TYPE_FIXED32 ||
+      type == UPB_DESCRIPTOR_TYPE_SFIXED64 ||
+      type == UPB_DESCRIPTOR_TYPE_SFIXED32) {
+    upb_fielddef_setintfmt(f, UPB_INTFMT_FIXED);
+  } else if (type == UPB_DESCRIPTOR_TYPE_SINT64 ||
+             type == UPB_DESCRIPTOR_TYPE_SINT32) {
+    upb_fielddef_setintfmt(f, UPB_INTFMT_ZIGZAG);
+  } else {
+    upb_fielddef_setintfmt(f, UPB_INTFMT_VARIABLE);
+  }
+
+  upb_fielddef_settagdelim(f, type == UPB_DESCRIPTOR_TYPE_GROUP);
+}
+
+upb_descriptortype_t upb_fielddef_descriptortype(const upb_fielddef *f) {
+  switch (upb_fielddef_type(f)) {
+    case UPB_TYPE_FLOAT:  return UPB_DESCRIPTOR_TYPE_FLOAT;
+    case UPB_TYPE_DOUBLE: return UPB_DESCRIPTOR_TYPE_DOUBLE;
+    case UPB_TYPE_BOOL:   return UPB_DESCRIPTOR_TYPE_BOOL;
+    case UPB_TYPE_STRING: return UPB_DESCRIPTOR_TYPE_STRING;
+    case UPB_TYPE_BYTES:  return UPB_DESCRIPTOR_TYPE_BYTES;
+    case UPB_TYPE_ENUM:   return UPB_DESCRIPTOR_TYPE_ENUM;
+    case UPB_TYPE_INT32:
+      switch (upb_fielddef_intfmt(f)) {
+        case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_INT32;
+        case UPB_INTFMT_FIXED:    return UPB_DESCRIPTOR_TYPE_SFIXED32;
+        case UPB_INTFMT_ZIGZAG:   return UPB_DESCRIPTOR_TYPE_SINT32;
+      }
+    case UPB_TYPE_INT64:
+      switch (upb_fielddef_intfmt(f)) {
+        case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_INT64;
+        case UPB_INTFMT_FIXED:    return UPB_DESCRIPTOR_TYPE_SFIXED64;
+        case UPB_INTFMT_ZIGZAG:   return UPB_DESCRIPTOR_TYPE_SINT64;
+      }
+    case UPB_TYPE_UINT32:
+      switch (upb_fielddef_intfmt(f)) {
+        case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_UINT32;
+        case UPB_INTFMT_FIXED:    return UPB_DESCRIPTOR_TYPE_FIXED32;
+        case UPB_INTFMT_ZIGZAG:   return -1;
+      }
+    case UPB_TYPE_UINT64:
+      switch (upb_fielddef_intfmt(f)) {
+        case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_UINT64;
+        case UPB_INTFMT_FIXED:    return UPB_DESCRIPTOR_TYPE_FIXED64;
+        case UPB_INTFMT_ZIGZAG:   return -1;
+      }
+    case UPB_TYPE_MESSAGE:
+      return upb_fielddef_istagdelim(f) ?
+          UPB_DESCRIPTOR_TYPE_GROUP : UPB_DESCRIPTOR_TYPE_MESSAGE;
+  }
+  return 0;
+}
+
+void upb_fielddef_setisextension(upb_fielddef *f, bool is_extension) {
+  assert(!upb_fielddef_isfrozen(f));
+  f->is_extension_ = is_extension;
+}
+
+void upb_fielddef_setlazy(upb_fielddef *f, bool lazy) {
+  assert(!upb_fielddef_isfrozen(f));
+  f->lazy_ = lazy;
+}
+
+void upb_fielddef_setpacked(upb_fielddef *f, bool packed) {
+  assert(!upb_fielddef_isfrozen(f));
+  f->packed_ = packed;
+}
+
+void upb_fielddef_setlabel(upb_fielddef *f, upb_label_t label) {
+  assert(!upb_fielddef_isfrozen(f));
+  assert(upb_fielddef_checklabel(label));
+  f->label_ = label;
+}
+
+void upb_fielddef_setintfmt(upb_fielddef *f, upb_intfmt_t fmt) {
+  assert(!upb_fielddef_isfrozen(f));
+  assert(upb_fielddef_checkintfmt(fmt));
+  f->intfmt = fmt;
+}
+
+void upb_fielddef_settagdelim(upb_fielddef *f, bool tag_delim) {
+  assert(!upb_fielddef_isfrozen(f));
+  f->tagdelim = tag_delim;
+  f->tagdelim = tag_delim;
+}
+
+static bool checksetdefault(upb_fielddef *f, upb_fieldtype_t type) {
+  if (!f->type_is_set_ || upb_fielddef_isfrozen(f) ||
+      upb_fielddef_type(f) != type) {
+    assert(false);
+    return false;
+  }
+  if (f->default_is_string) {
+    str_t *s = f->defaultval.bytes;
+    assert(s || type == UPB_TYPE_ENUM);
+    if (s) freestr(s);
+  }
+  f->default_is_string = false;
+  return true;
+}
+
+void upb_fielddef_setdefaultint64(upb_fielddef *f, int64_t value) {
+  if (checksetdefault(f, UPB_TYPE_INT64))
+    f->defaultval.sint = value;
+}
+
+void upb_fielddef_setdefaultint32(upb_fielddef *f, int32_t value) {
+  if ((upb_fielddef_type(f) == UPB_TYPE_ENUM &&
+       checksetdefault(f, UPB_TYPE_ENUM)) ||
+      checksetdefault(f, UPB_TYPE_INT32)) {
+    f->defaultval.sint = value;
+  }
+}
+
+void upb_fielddef_setdefaultuint64(upb_fielddef *f, uint64_t value) {
+  if (checksetdefault(f, UPB_TYPE_UINT64))
+    f->defaultval.uint = value;
+}
+
+void upb_fielddef_setdefaultuint32(upb_fielddef *f, uint32_t value) {
+  if (checksetdefault(f, UPB_TYPE_UINT32))
+    f->defaultval.uint = value;
+}
+
+void upb_fielddef_setdefaultbool(upb_fielddef *f, bool value) {
+  if (checksetdefault(f, UPB_TYPE_BOOL))
+    f->defaultval.uint = value;
+}
+
+void upb_fielddef_setdefaultfloat(upb_fielddef *f, float value) {
+  if (checksetdefault(f, UPB_TYPE_FLOAT))
+    f->defaultval.flt = value;
+}
+
+void upb_fielddef_setdefaultdouble(upb_fielddef *f, double value) {
+  if (checksetdefault(f, UPB_TYPE_DOUBLE))
+    f->defaultval.dbl = value;
+}
+
+bool upb_fielddef_setdefaultstr(upb_fielddef *f, const void *str, size_t len,
+                                upb_status *s) {
+  str_t *str2;
+  assert(upb_fielddef_isstring(f) || f->type_ == UPB_TYPE_ENUM);
+  if (f->type_ == UPB_TYPE_ENUM && !upb_isident(str, len, false, s))
+    return false;
+
+  if (f->default_is_string) {
+    str_t *s = f->defaultval.bytes;
+    assert(s || f->type_ == UPB_TYPE_ENUM);
+    if (s) freestr(s);
+  } else {
+    assert(f->type_ == UPB_TYPE_ENUM);
+  }
+
+  str2 = newstr(str, len);
+  f->defaultval.bytes = str2;
+  f->default_is_string = true;
+  return true;
+}
+
+void upb_fielddef_setdefaultcstr(upb_fielddef *f, const char *str,
+                                 upb_status *s) {
+  assert(f->type_is_set_);
+  upb_fielddef_setdefaultstr(f, str, str ? strlen(str) : 0, s);
+}
+
+bool upb_fielddef_enumhasdefaultint32(const upb_fielddef *f) {
+  int32_t val;
+  assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM);
+  return enumdefaultint32(f, &val);
+}
+
+bool upb_fielddef_enumhasdefaultstr(const upb_fielddef *f) {
+  assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM);
+  return enumdefaultstr(f) != NULL;
+}
+
+static bool upb_subdef_typecheck(upb_fielddef *f, const upb_def *subdef,
+                                 upb_status *s) {
+  if (f->type_ == UPB_TYPE_MESSAGE) {
+    if (upb_dyncast_msgdef(subdef)) return true;
+    upb_status_seterrmsg(s, "invalid subdef type for this submessage field");
+    return false;
+  } else if (f->type_ == UPB_TYPE_ENUM) {
+    if (upb_dyncast_enumdef(subdef)) return true;
+    upb_status_seterrmsg(s, "invalid subdef type for this enum field");
+    return false;
+  } else {
+    upb_status_seterrmsg(s, "only message and enum fields can have a subdef");
+    return false;
+  }
+}
+
+static void release_subdef(upb_fielddef *f) {
+  if (f->subdef_is_symbolic) {
+    free(f->sub.name);
+  } else if (f->sub.def) {
+    upb_unref2(f->sub.def, f);
+  }
+}
+
+bool upb_fielddef_setsubdef(upb_fielddef *f, const upb_def *subdef,
+                            upb_status *s) {
+  assert(!upb_fielddef_isfrozen(f));
+  assert(upb_fielddef_hassubdef(f));
+  if (subdef && !upb_subdef_typecheck(f, subdef, s)) return false;
+  release_subdef(f);
+  f->sub.def = subdef;
+  f->subdef_is_symbolic = false;
+  if (f->sub.def) upb_ref2(f->sub.def, f);
+  return true;
+}
+
+bool upb_fielddef_setmsgsubdef(upb_fielddef *f, const upb_msgdef *subdef,
+                               upb_status *s) {
+  return upb_fielddef_setsubdef(f, upb_msgdef_upcast(subdef), s);
+}
+
+bool upb_fielddef_setenumsubdef(upb_fielddef *f, const upb_enumdef *subdef,
+                                upb_status *s) {
+  return upb_fielddef_setsubdef(f, upb_enumdef_upcast(subdef), s);
+}
+
+bool upb_fielddef_setsubdefname(upb_fielddef *f, const char *name,
+                                upb_status *s) {
+  assert(!upb_fielddef_isfrozen(f));
+  if (!upb_fielddef_hassubdef(f)) {
+    upb_status_seterrmsg(s, "field type does not accept a subdef");
+    return false;
+  }
+  /* TODO: validate name (upb_isident() doesn't quite work atm because this name
+   * may have a leading "."). */
+  release_subdef(f);
+  f->sub.name = upb_strdup(name);
+  f->subdef_is_symbolic = true;
+  return true;
+}
+
+bool upb_fielddef_issubmsg(const upb_fielddef *f) {
+  return upb_fielddef_type(f) == UPB_TYPE_MESSAGE;
+}
+
+bool upb_fielddef_isstring(const upb_fielddef *f) {
+  return upb_fielddef_type(f) == UPB_TYPE_STRING ||
+         upb_fielddef_type(f) == UPB_TYPE_BYTES;
+}
+
+bool upb_fielddef_isseq(const upb_fielddef *f) {
+  return upb_fielddef_label(f) == UPB_LABEL_REPEATED;
+}
+
+bool upb_fielddef_isprimitive(const upb_fielddef *f) {
+  return !upb_fielddef_isstring(f) && !upb_fielddef_issubmsg(f);
+}
+
+bool upb_fielddef_ismap(const upb_fielddef *f) {
+  return upb_fielddef_isseq(f) && upb_fielddef_issubmsg(f) &&
+         upb_msgdef_mapentry(upb_fielddef_msgsubdef(f));
+}
+
+bool upb_fielddef_hassubdef(const upb_fielddef *f) {
+  return upb_fielddef_issubmsg(f) || upb_fielddef_type(f) == UPB_TYPE_ENUM;
+}
+
+static bool between(int32_t x, int32_t low, int32_t high) {
+  return x >= low && x <= high;
+}
+
+bool upb_fielddef_checklabel(int32_t label) { return between(label, 1, 3); }
+bool upb_fielddef_checktype(int32_t type) { return between(type, 1, 11); }
+bool upb_fielddef_checkintfmt(int32_t fmt) { return between(fmt, 1, 3); }
+
+bool upb_fielddef_checkdescriptortype(int32_t type) {
+  return between(type, 1, 18);
+}
+
+/* upb_msgdef *****************************************************************/
+
+static void visitmsg(const upb_refcounted *r, upb_refcounted_visit *visit,
+                     void *closure) {
+  upb_msg_oneof_iter o;
+  const upb_msgdef *m = (const upb_msgdef*)r;
+  upb_msg_field_iter i;
+  for(upb_msg_field_begin(&i, m);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    upb_fielddef *f = upb_msg_iter_field(&i);
+    visit(r, upb_fielddef_upcast2(f), closure);
+  }
+  for(upb_msg_oneof_begin(&o, m);
+      !upb_msg_oneof_done(&o);
+      upb_msg_oneof_next(&o)) {
+    upb_oneofdef *f = upb_msg_iter_oneof(&o);
+    visit(r, upb_oneofdef_upcast2(f), closure);
+  }
+}
+
+static void freemsg(upb_refcounted *r) {
+  upb_msgdef *m = (upb_msgdef*)r;
+  upb_strtable_uninit(&m->ntoo);
+  upb_strtable_uninit(&m->ntof);
+  upb_inttable_uninit(&m->itof);
+  upb_def_uninit(upb_msgdef_upcast_mutable(m));
+  free(m);
+}
+
+upb_msgdef *upb_msgdef_new(const void *owner) {
+  static const struct upb_refcounted_vtbl vtbl = {visitmsg, freemsg};
+  upb_msgdef *m = malloc(sizeof(*m));
+  if (!m) return NULL;
+  if (!upb_def_init(upb_msgdef_upcast_mutable(m), UPB_DEF_MSG, &vtbl, owner))
+    goto err2;
+  if (!upb_inttable_init(&m->itof, UPB_CTYPE_PTR)) goto err3;
+  if (!upb_strtable_init(&m->ntof, UPB_CTYPE_PTR)) goto err2;
+  if (!upb_strtable_init(&m->ntoo, UPB_CTYPE_PTR)) goto err1;
+  m->map_entry = false;
+  return m;
+
+err1:
+  upb_strtable_uninit(&m->ntof);
+err2:
+  upb_inttable_uninit(&m->itof);
+err3:
+  free(m);
+  return NULL;
+}
+
+upb_msgdef *upb_msgdef_dup(const upb_msgdef *m, const void *owner) {
+  bool ok;
+  upb_msg_field_iter i;
+  upb_msg_oneof_iter o;
+
+  upb_msgdef *newm = upb_msgdef_new(owner);
+  if (!newm) return NULL;
+  ok = upb_def_setfullname(upb_msgdef_upcast_mutable(newm),
+                           upb_def_fullname(upb_msgdef_upcast(m)),
+                           NULL);
+  newm->map_entry = m->map_entry;
+  UPB_ASSERT_VAR(ok, ok);
+  for(upb_msg_field_begin(&i, m);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    upb_fielddef *f = upb_fielddef_dup(upb_msg_iter_field(&i), &f);
+    /* Fields in oneofs are dup'd below. */
+    if (upb_fielddef_containingoneof(f)) continue;
+    if (!f || !upb_msgdef_addfield(newm, f, &f, NULL)) {
+      upb_msgdef_unref(newm, owner);
+      return NULL;
+    }
+  }
+  for(upb_msg_oneof_begin(&o, m);
+      !upb_msg_oneof_done(&o);
+      upb_msg_oneof_next(&o)) {
+    upb_oneofdef *f = upb_oneofdef_dup(upb_msg_iter_oneof(&o), &f);
+    if (!f || !upb_msgdef_addoneof(newm, f, &f, NULL)) {
+      upb_msgdef_unref(newm, owner);
+      return NULL;
+    }
+  }
+  return newm;
+}
+
+bool upb_msgdef_freeze(upb_msgdef *m, upb_status *status) {
+  upb_def *d = upb_msgdef_upcast_mutable(m);
+  return upb_def_freeze(&d, 1, status);
+}
+
+const char *upb_msgdef_fullname(const upb_msgdef *m) {
+  return upb_def_fullname(upb_msgdef_upcast(m));
+}
+
+bool upb_msgdef_setfullname(upb_msgdef *m, const char *fullname,
+                            upb_status *s) {
+  return upb_def_setfullname(upb_msgdef_upcast_mutable(m), fullname, s);
+}
+
+/* Helper: check that the field |f| is safe to add to msgdef |m|. Set an error
+ * on status |s| and return false if not. */
+static bool check_field_add(const upb_msgdef *m, const upb_fielddef *f,
+                            upb_status *s) {
+  if (upb_fielddef_containingtype(f) != NULL) {
+    upb_status_seterrmsg(s, "fielddef already belongs to a message");
+    return false;
+  } else if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) {
+    upb_status_seterrmsg(s, "field name or number were not set");
+    return false;
+  } else if (upb_msgdef_ntofz(m, upb_fielddef_name(f)) ||
+             upb_msgdef_itof(m, upb_fielddef_number(f))) {
+    upb_status_seterrmsg(s, "duplicate field name or number for field");
+    return false;
+  }
+  return true;
+}
+
+static void add_field(upb_msgdef *m, upb_fielddef *f, const void *ref_donor) {
+  release_containingtype(f);
+  f->msg.def = m;
+  f->msg_is_symbolic = false;
+  upb_inttable_insert(&m->itof, upb_fielddef_number(f), upb_value_ptr(f));
+  upb_strtable_insert(&m->ntof, upb_fielddef_name(f), upb_value_ptr(f));
+  upb_ref2(f, m);
+  upb_ref2(m, f);
+  if (ref_donor) upb_fielddef_unref(f, ref_donor);
+}
+
+bool upb_msgdef_addfield(upb_msgdef *m, upb_fielddef *f, const void *ref_donor,
+                         upb_status *s) {
+  /* TODO: extensions need to have a separate namespace, because proto2 allows a
+   * top-level extension (ie. one not in any package) to have the same name as a
+   * field from the message.
+   *
+   * This also implies that there needs to be a separate lookup-by-name method
+   * for extensions.  It seems desirable for iteration to return both extensions
+   * and non-extensions though.
+   *
+   * We also need to validate that the field number is in an extension range iff
+   * it is an extension.
+   *
+   * This method is idempotent. Check if |f| is already part of this msgdef and
+   * return immediately if so. */
+  if (upb_fielddef_containingtype(f) == m) {
+    return true;
+  }
+
+  /* Check constraints for all fields before performing any action. */
+  if (!check_field_add(m, f, s)) {
+    return false;
+  } else if (upb_fielddef_containingoneof(f) != NULL) {
+    /* Fields in a oneof can only be added by adding the oneof to the msgdef. */
+    upb_status_seterrmsg(s, "fielddef is part of a oneof");
+    return false;
+  }
+
+  /* Constraint checks ok, perform the action. */
+  add_field(m, f, ref_donor);
+  return true;
+}
+
+bool upb_msgdef_addoneof(upb_msgdef *m, upb_oneofdef *o, const void *ref_donor,
+                         upb_status *s) {
+  upb_oneof_iter it;
+
+  /* Check various conditions that would prevent this oneof from being added. */
+  if (upb_oneofdef_containingtype(o)) {
+    upb_status_seterrmsg(s, "oneofdef already belongs to a message");
+    return false;
+  } else if (upb_oneofdef_name(o) == NULL) {
+    upb_status_seterrmsg(s, "oneofdef name was not set");
+    return false;
+  } else if (upb_msgdef_ntooz(m, upb_oneofdef_name(o))) {
+    upb_status_seterrmsg(s, "duplicate oneof name");
+    return false;
+  }
+
+  /* Check that all of the oneof's fields do not conflict with names or numbers
+   * of fields already in the message. */
+  for (upb_oneof_begin(&it, o); !upb_oneof_done(&it); upb_oneof_next(&it)) {
+    const upb_fielddef *f = upb_oneof_iter_field(&it);
+    if (!check_field_add(m, f, s)) {
+      return false;
+    }
+  }
+
+  /* Everything checks out -- commit now. */
+
+  /* Add oneof itself first. */
+  o->parent = m;
+  upb_strtable_insert(&m->ntoo, upb_oneofdef_name(o), upb_value_ptr(o));
+  upb_ref2(o, m);
+  upb_ref2(m, o);
+
+  /* Add each field of the oneof directly to the msgdef. */
+  for (upb_oneof_begin(&it, o); !upb_oneof_done(&it); upb_oneof_next(&it)) {
+    upb_fielddef *f = upb_oneof_iter_field(&it);
+    add_field(m, f, NULL);
+  }
+
+  if (ref_donor) upb_oneofdef_unref(o, ref_donor);
+
+  return true;
+}
+
+const upb_fielddef *upb_msgdef_itof(const upb_msgdef *m, uint32_t i) {
+  upb_value val;
+  return upb_inttable_lookup32(&m->itof, i, &val) ?
+      upb_value_getptr(val) : NULL;
+}
+
+const upb_fielddef *upb_msgdef_ntof(const upb_msgdef *m, const char *name,
+                                    size_t len) {
+  upb_value val;
+  return upb_strtable_lookup2(&m->ntof, name, len, &val) ?
+      upb_value_getptr(val) : NULL;
+}
+
+const upb_oneofdef *upb_msgdef_ntoo(const upb_msgdef *m, const char *name,
+                                    size_t len) {
+  upb_value val;
+  return upb_strtable_lookup2(&m->ntoo, name, len, &val) ?
+      upb_value_getptr(val) : NULL;
+}
+
+int upb_msgdef_numfields(const upb_msgdef *m) {
+  return upb_strtable_count(&m->ntof);
+}
+
+int upb_msgdef_numoneofs(const upb_msgdef *m) {
+  return upb_strtable_count(&m->ntoo);
+}
+
+void upb_msgdef_setmapentry(upb_msgdef *m, bool map_entry) {
+  assert(!upb_msgdef_isfrozen(m));
+  m->map_entry = map_entry;
+}
+
+bool upb_msgdef_mapentry(const upb_msgdef *m) {
+  return m->map_entry;
+}
+
+void upb_msg_field_begin(upb_msg_field_iter *iter, const upb_msgdef *m) {
+  upb_inttable_begin(iter, &m->itof);
+}
+
+void upb_msg_field_next(upb_msg_field_iter *iter) { upb_inttable_next(iter); }
+
+bool upb_msg_field_done(const upb_msg_field_iter *iter) {
+  return upb_inttable_done(iter);
+}
+
+upb_fielddef *upb_msg_iter_field(const upb_msg_field_iter *iter) {
+  return (upb_fielddef*)upb_value_getptr(upb_inttable_iter_value(iter));
+}
+
+void upb_msg_field_iter_setdone(upb_msg_field_iter *iter) {
+  upb_inttable_iter_setdone(iter);
+}
+
+void upb_msg_oneof_begin(upb_msg_oneof_iter *iter, const upb_msgdef *m) {
+  upb_strtable_begin(iter, &m->ntoo);
+}
+
+void upb_msg_oneof_next(upb_msg_oneof_iter *iter) { upb_strtable_next(iter); }
+
+bool upb_msg_oneof_done(const upb_msg_oneof_iter *iter) {
+  return upb_strtable_done(iter);
+}
+
+upb_oneofdef *upb_msg_iter_oneof(const upb_msg_oneof_iter *iter) {
+  return (upb_oneofdef*)upb_value_getptr(upb_strtable_iter_value(iter));
+}
+
+void upb_msg_oneof_iter_setdone(upb_msg_oneof_iter *iter) {
+  upb_strtable_iter_setdone(iter);
+}
+
+/* upb_oneofdef ***************************************************************/
+
+static void visitoneof(const upb_refcounted *r, upb_refcounted_visit *visit,
+                       void *closure) {
+  const upb_oneofdef *o = (const upb_oneofdef*)r;
+  upb_oneof_iter i;
+  for (upb_oneof_begin(&i, o); !upb_oneof_done(&i); upb_oneof_next(&i)) {
+    const upb_fielddef *f = upb_oneof_iter_field(&i);
+    visit(r, upb_fielddef_upcast2(f), closure);
+  }
+  if (o->parent) {
+    visit(r, upb_msgdef_upcast2(o->parent), closure);
+  }
+}
+
+static void freeoneof(upb_refcounted *r) {
+  upb_oneofdef *o = (upb_oneofdef*)r;
+  upb_strtable_uninit(&o->ntof);
+  upb_inttable_uninit(&o->itof);
+  upb_def_uninit(upb_oneofdef_upcast_mutable(o));
+  free(o);
+}
+
+upb_oneofdef *upb_oneofdef_new(const void *owner) {
+  static const struct upb_refcounted_vtbl vtbl = {visitoneof, freeoneof};
+  upb_oneofdef *o = malloc(sizeof(*o));
+  o->parent = NULL;
+  if (!o) return NULL;
+  if (!upb_def_init(upb_oneofdef_upcast_mutable(o), UPB_DEF_ONEOF, &vtbl,
+                    owner))
+    goto err2;
+  if (!upb_inttable_init(&o->itof, UPB_CTYPE_PTR)) goto err2;
+  if (!upb_strtable_init(&o->ntof, UPB_CTYPE_PTR)) goto err1;
+  return o;
+
+err1:
+  upb_inttable_uninit(&o->itof);
+err2:
+  free(o);
+  return NULL;
+}
+
+upb_oneofdef *upb_oneofdef_dup(const upb_oneofdef *o, const void *owner) {
+  bool ok;
+  upb_oneof_iter i;
+  upb_oneofdef *newo = upb_oneofdef_new(owner);
+  if (!newo) return NULL;
+  ok = upb_def_setfullname(upb_oneofdef_upcast_mutable(newo),
+                           upb_def_fullname(upb_oneofdef_upcast(o)), NULL);
+  UPB_ASSERT_VAR(ok, ok);
+  for (upb_oneof_begin(&i, o); !upb_oneof_done(&i); upb_oneof_next(&i)) {
+    upb_fielddef *f = upb_fielddef_dup(upb_oneof_iter_field(&i), &f);
+    if (!f || !upb_oneofdef_addfield(newo, f, &f, NULL)) {
+      upb_oneofdef_unref(newo, owner);
+      return NULL;
+    }
+  }
+  return newo;
+}
+
+const char *upb_oneofdef_name(const upb_oneofdef *o) {
+  return upb_def_fullname(upb_oneofdef_upcast(o));
+}
+
+bool upb_oneofdef_setname(upb_oneofdef *o, const char *fullname,
+                             upb_status *s) {
+  if (upb_oneofdef_containingtype(o)) {
+    upb_status_seterrmsg(s, "oneof already added to a message");
+    return false;
+  }
+  return upb_def_setfullname(upb_oneofdef_upcast_mutable(o), fullname, s);
+}
+
+const upb_msgdef *upb_oneofdef_containingtype(const upb_oneofdef *o) {
+  return o->parent;
+}
+
+int upb_oneofdef_numfields(const upb_oneofdef *o) {
+  return upb_strtable_count(&o->ntof);
+}
+
+bool upb_oneofdef_addfield(upb_oneofdef *o, upb_fielddef *f,
+                           const void *ref_donor,
+                           upb_status *s) {
+  assert(!upb_oneofdef_isfrozen(o));
+  assert(!o->parent || !upb_msgdef_isfrozen(o->parent));
+
+  /* This method is idempotent. Check if |f| is already part of this oneofdef
+   * and return immediately if so. */
+  if (upb_fielddef_containingoneof(f) == o) {
+    return true;
+  }
+
+  /* The field must have an OPTIONAL label. */
+  if (upb_fielddef_label(f) != UPB_LABEL_OPTIONAL) {
+    upb_status_seterrmsg(s, "fields in oneof must have OPTIONAL label");
+    return false;
+  }
+
+  /* Check that no field with this name or number exists already in the oneof.
+   * Also check that the field is not already part of a oneof. */
+  if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) {
+    upb_status_seterrmsg(s, "field name or number were not set");
+    return false;
+  } else if (upb_oneofdef_itof(o, upb_fielddef_number(f)) ||
+             upb_oneofdef_ntofz(o, upb_fielddef_name(f))) {
+    upb_status_seterrmsg(s, "duplicate field name or number");
+    return false;
+  } else if (upb_fielddef_containingoneof(f) != NULL) {
+    upb_status_seterrmsg(s, "fielddef already belongs to a oneof");
+    return false;
+  }
+
+  /* We allow adding a field to the oneof either if the field is not part of a
+   * msgdef, or if it is and we are also part of the same msgdef. */
+  if (o->parent == NULL) {
+    /* If we're not in a msgdef, the field cannot be either. Otherwise we would
+     * need to magically add this oneof to a msgdef to remain consistent, which
+     * is surprising behavior. */
+    if (upb_fielddef_containingtype(f) != NULL) {
+      upb_status_seterrmsg(s, "fielddef already belongs to a message, but "
+                              "oneof does not");
+      return false;
+    }
+  } else {
+    /* If we're in a msgdef, the user can add fields that either aren't in any
+     * msgdef (in which case they're added to our msgdef) or already a part of
+     * our msgdef. */
+    if (upb_fielddef_containingtype(f) != NULL &&
+        upb_fielddef_containingtype(f) != o->parent) {
+      upb_status_seterrmsg(s, "fielddef belongs to a different message "
+                              "than oneof");
+      return false;
+    }
+  }
+
+  /* Commit phase. First add the field to our parent msgdef, if any, because
+   * that may fail; then add the field to our own tables. */
+
+  if (o->parent != NULL && upb_fielddef_containingtype(f) == NULL) {
+    if (!upb_msgdef_addfield((upb_msgdef*)o->parent, f, NULL, s)) {
+      return false;
+    }
+  }
+
+  release_containingtype(f);
+  f->oneof = o;
+  upb_inttable_insert(&o->itof, upb_fielddef_number(f), upb_value_ptr(f));
+  upb_strtable_insert(&o->ntof, upb_fielddef_name(f), upb_value_ptr(f));
+  upb_ref2(f, o);
+  upb_ref2(o, f);
+  if (ref_donor) upb_fielddef_unref(f, ref_donor);
+
+  return true;
+}
+
+const upb_fielddef *upb_oneofdef_ntof(const upb_oneofdef *o,
+                                      const char *name, size_t length) {
+  upb_value val;
+  return upb_strtable_lookup2(&o->ntof, name, length, &val) ?
+      upb_value_getptr(val) : NULL;
+}
+
+const upb_fielddef *upb_oneofdef_itof(const upb_oneofdef *o, uint32_t num) {
+  upb_value val;
+  return upb_inttable_lookup32(&o->itof, num, &val) ?
+      upb_value_getptr(val) : NULL;
+}
+
+void upb_oneof_begin(upb_oneof_iter *iter, const upb_oneofdef *o) {
+  upb_inttable_begin(iter, &o->itof);
+}
+
+void upb_oneof_next(upb_oneof_iter *iter) {
+  upb_inttable_next(iter);
+}
+
+bool upb_oneof_done(upb_oneof_iter *iter) {
+  return upb_inttable_done(iter);
+}
+
+upb_fielddef *upb_oneof_iter_field(const upb_oneof_iter *iter) {
+  return (upb_fielddef*)upb_value_getptr(upb_inttable_iter_value(iter));
+}
+
+void upb_oneof_iter_setdone(upb_oneof_iter *iter) {
+  upb_inttable_iter_setdone(iter);
+}
+
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+
+typedef struct cleanup_ent {
+  upb_cleanup_func *cleanup;
+  void *ud;
+  struct cleanup_ent *next;
+} cleanup_ent;
+
+static void *seeded_alloc(void *ud, void *ptr, size_t oldsize, size_t size);
+
+/* Default allocator **********************************************************/
+
+/* Just use realloc, keeping all allocated blocks in a linked list to destroy at
+ * the end. */
+
+typedef struct mem_block {
+  /* List is doubly-linked, because in cases where realloc() moves an existing
+   * block, we need to be able to remove the old pointer from the list
+   * efficiently. */
+  struct mem_block *prev, *next;
+#ifndef NDEBUG
+  size_t size;  /* Doesn't include mem_block structure. */
+#endif
+} mem_block;
+
+typedef struct {
+  mem_block *head;
+} default_alloc_ud;
+
+static void *default_alloc(void *_ud, void *ptr, size_t oldsize, size_t size) {
+  default_alloc_ud *ud = _ud;
+  mem_block *from, *block;
+  void *ret;
+  UPB_UNUSED(oldsize);
+
+  from = ptr ? (void*)((char*)ptr - sizeof(mem_block)) : NULL;
+
+#ifndef NDEBUG
+  if (from) {
+    assert(oldsize <= from->size);
+  }
+#endif
+
+  /* TODO(haberman): we probably need to provide even better alignment here,
+   * like 16-byte alignment of the returned data pointer. */
+  block = realloc(from, size + sizeof(mem_block));
+  if (!block) return NULL;
+  ret = (char*)block + sizeof(*block);
+
+#ifndef NDEBUG
+  block->size = size;
+#endif
+
+  if (from) {
+    if (block != from) {
+      /* The block was moved, so pointers in next and prev blocks must be
+       * updated to its new location. */
+      if (block->next) block->next->prev = block;
+      if (block->prev) block->prev->next = block;
+      if (ud->head == from) ud->head = block;
+    }
+  } else {
+    /* Insert at head of linked list. */
+    block->prev = NULL;
+    block->next = ud->head;
+    if (block->next) block->next->prev = block;
+    ud->head = block;
+  }
+
+  return ret;
+}
+
+static void default_alloc_cleanup(void *_ud) {
+  default_alloc_ud *ud = _ud;
+  mem_block *block = ud->head;
+
+  while (block) {
+    void *to_free = block;
+    block = block->next;
+    free(to_free);
+  }
+}
+
+
+/* Standard error functions ***************************************************/
+
+static bool default_err(void *ud, const upb_status *status) {
+  UPB_UNUSED(ud);
+  UPB_UNUSED(status);
+  return false;
+}
+
+static bool write_err_to(void *ud, const upb_status *status) {
+  upb_status *copy_to = ud;
+  upb_status_copy(copy_to, status);
+  return false;
+}
+
+
+/* upb_env ********************************************************************/
+
+void upb_env_init(upb_env *e) {
+  default_alloc_ud *ud = (default_alloc_ud*)&e->default_alloc_ud;
+  e->ok_ = true;
+  e->bytes_allocated = 0;
+  e->cleanup_head = NULL;
+
+  ud->head = NULL;
+
+  /* Set default functions. */
+  upb_env_setallocfunc(e, default_alloc, ud);
+  upb_env_seterrorfunc(e, default_err, NULL);
+}
+
+void upb_env_uninit(upb_env *e) {
+  cleanup_ent *ent = e->cleanup_head;
+
+  while (ent) {
+    ent->cleanup(ent->ud);
+    ent = ent->next;
+  }
+
+  /* Must do this after running cleanup functions, because this will delete
+     the memory we store our cleanup entries in! */
+  if (e->alloc == default_alloc) {
+    default_alloc_cleanup(e->alloc_ud);
+  }
+}
+
+UPB_FORCEINLINE void upb_env_setallocfunc(upb_env *e, upb_alloc_func *alloc,
+                                          void *ud) {
+  e->alloc = alloc;
+  e->alloc_ud = ud;
+}
+
+UPB_FORCEINLINE void upb_env_seterrorfunc(upb_env *e, upb_error_func *func,
+                                          void *ud) {
+  e->err = func;
+  e->err_ud = ud;
+}
+
+void upb_env_reporterrorsto(upb_env *e, upb_status *status) {
+  e->err = write_err_to;
+  e->err_ud = status;
+}
+
+bool upb_env_ok(const upb_env *e) {
+  return e->ok_;
+}
+
+bool upb_env_reporterror(upb_env *e, const upb_status *status) {
+  e->ok_ = false;
+  return e->err(e->err_ud, status);
+}
+
+bool upb_env_addcleanup(upb_env *e, upb_cleanup_func *func, void *ud) {
+  cleanup_ent *ent = upb_env_malloc(e, sizeof(cleanup_ent));
+  if (!ent) return false;
+
+  ent->cleanup = func;
+  ent->ud = ud;
+  ent->next = e->cleanup_head;
+  e->cleanup_head = ent;
+
+  return true;
+}
+
+void *upb_env_malloc(upb_env *e, size_t size) {
+  e->bytes_allocated += size;
+  if (e->alloc == seeded_alloc) {
+    /* This is equivalent to the next branch, but allows inlining for a
+     * measurable perf benefit. */
+    return seeded_alloc(e->alloc_ud, NULL, 0, size);
+  } else {
+    return e->alloc(e->alloc_ud, NULL, 0, size);
+  }
+}
+
+void *upb_env_realloc(upb_env *e, void *ptr, size_t oldsize, size_t size) {
+  char *ret;
+  assert(oldsize <= size);
+  ret = e->alloc(e->alloc_ud, ptr, oldsize, size);
+
+#ifndef NDEBUG
+  /* Overwrite non-preserved memory to ensure callers are passing the oldsize
+   * that they truly require. */
+  memset(ret + oldsize, 0xff, size - oldsize);
+#endif
+
+  return ret;
+}
+
+size_t upb_env_bytesallocated(const upb_env *e) {
+  return e->bytes_allocated;
+}
+
+
+/* upb_seededalloc ************************************************************/
+
+/* Be conservative and choose 16 in case anyone is using SSE. */
+static const size_t maxalign = 16;
+
+static size_t align_up(size_t size) {
+  return ((size + maxalign - 1) / maxalign) * maxalign;
+}
+
+UPB_FORCEINLINE static void *seeded_alloc(void *ud, void *ptr, size_t oldsize,
+                                          size_t size) {
+  upb_seededalloc *a = ud;
+
+  size = align_up(size);
+
+  assert(a->mem_limit >= a->mem_ptr);
+
+  if (oldsize == 0 && size <= (size_t)(a->mem_limit - a->mem_ptr)) {
+    /* Fast path: we can satisfy from the initial allocation. */
+    void *ret = a->mem_ptr;
+    a->mem_ptr += size;
+    return ret;
+  } else {
+    char *chptr = ptr;
+    /* Slow path: fallback to other allocator. */
+    a->need_cleanup = true;
+    /* Is `ptr` part of the user-provided initial block? Don't pass it to the
+     * default allocator if so; otherwise, it may try to realloc() the block. */
+    if (chptr >= a->mem_base && chptr < a->mem_limit) {
+      void *ret;
+      assert(chptr + oldsize <= a->mem_limit);
+      ret = a->alloc(a->alloc_ud, NULL, 0, size);
+      if (ret) memcpy(ret, ptr, oldsize);
+      return ret;
+    } else {
+      return a->alloc(a->alloc_ud, ptr, oldsize, size);
+    }
+  }
+}
+
+void upb_seededalloc_init(upb_seededalloc *a, void *mem, size_t len) {
+  default_alloc_ud *ud = (default_alloc_ud*)&a->default_alloc_ud;
+  a->mem_base = mem;
+  a->mem_ptr = mem;
+  a->mem_limit = (char*)mem + len;
+  a->need_cleanup = false;
+  a->returned_allocfunc = false;
+
+  ud->head = NULL;
+
+  upb_seededalloc_setfallbackalloc(a, default_alloc, ud);
+}
+
+void upb_seededalloc_uninit(upb_seededalloc *a) {
+  if (a->alloc == default_alloc && a->need_cleanup) {
+    default_alloc_cleanup(a->alloc_ud);
+  }
+}
+
+UPB_FORCEINLINE void upb_seededalloc_setfallbackalloc(upb_seededalloc *a,
+                                                      upb_alloc_func *alloc,
+                                                      void *ud) {
+  assert(!a->returned_allocfunc);
+  a->alloc = alloc;
+  a->alloc_ud = ud;
+}
+
+upb_alloc_func *upb_seededalloc_getallocfunc(upb_seededalloc *a) {
+  a->returned_allocfunc = true;
+  return seeded_alloc;
+}
+/*
+** TODO(haberman): it's unclear whether a lot of the consistency checks should
+** assert() or return false.
+*/
+
+
+#include <stdlib.h>
+#include <string.h>
+
+
+
+/* Defined for the sole purpose of having a unique pointer value for
+ * UPB_NO_CLOSURE. */
+char _upb_noclosure;
+
+static void freehandlers(upb_refcounted *r) {
+  upb_handlers *h = (upb_handlers*)r;
+
+  upb_inttable_iter i;
+  upb_inttable_begin(&i, &h->cleanup_);
+  for(; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    void *val = (void*)upb_inttable_iter_key(&i);
+    upb_value func_val = upb_inttable_iter_value(&i);
+    upb_handlerfree *func = upb_value_getfptr(func_val);
+    func(val);
+  }
+
+  upb_inttable_uninit(&h->cleanup_);
+  upb_msgdef_unref(h->msg, h);
+  free(h->sub);
+  free(h);
+}
+
+static void visithandlers(const upb_refcounted *r, upb_refcounted_visit *visit,
+                          void *closure) {
+  const upb_handlers *h = (const upb_handlers*)r;
+  upb_msg_field_iter i;
+  for(upb_msg_field_begin(&i, h->msg);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    upb_fielddef *f = upb_msg_iter_field(&i);
+    const upb_handlers *sub;
+    if (!upb_fielddef_issubmsg(f)) continue;
+    sub = upb_handlers_getsubhandlers(h, f);
+    if (sub) visit(r, upb_handlers_upcast(sub), closure);
+  }
+}
+
+static const struct upb_refcounted_vtbl vtbl = {visithandlers, freehandlers};
+
+typedef struct {
+  upb_inttable tab;  /* maps upb_msgdef* -> upb_handlers*. */
+  upb_handlers_callback *callback;
+  const void *closure;
+} dfs_state;
+
+/* TODO(haberman): discard upb_handlers* objects that do not actually have any
+ * handlers set and cannot reach any upb_handlers* object that does.  This is
+ * slightly tricky to do correctly. */
+static upb_handlers *newformsg(const upb_msgdef *m, const void *owner,
+                               dfs_state *s) {
+  upb_msg_field_iter i;
+  upb_handlers *h = upb_handlers_new(m, owner);
+  if (!h) return NULL;
+  if (!upb_inttable_insertptr(&s->tab, m, upb_value_ptr(h))) goto oom;
+
+  s->callback(s->closure, h);
+
+  /* For each submessage field, get or create a handlers object and set it as
+   * the subhandlers. */
+  for(upb_msg_field_begin(&i, m);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    upb_fielddef *f = upb_msg_iter_field(&i);
+    const upb_msgdef *subdef;
+    upb_value subm_ent;
+
+    if (!upb_fielddef_issubmsg(f)) continue;
+
+    subdef = upb_downcast_msgdef(upb_fielddef_subdef(f));
+    if (upb_inttable_lookupptr(&s->tab, subdef, &subm_ent)) {
+      upb_handlers_setsubhandlers(h, f, upb_value_getptr(subm_ent));
+    } else {
+      upb_handlers *sub_mh = newformsg(subdef, &sub_mh, s);
+      if (!sub_mh) goto oom;
+      upb_handlers_setsubhandlers(h, f, sub_mh);
+      upb_handlers_unref(sub_mh, &sub_mh);
+    }
+  }
+  return h;
+
+oom:
+  upb_handlers_unref(h, owner);
+  return NULL;
+}
+
+/* Given a selector for a STARTSUBMSG handler, resolves to a pointer to the
+ * subhandlers for this submessage field. */
+#define SUBH(h, selector) (h->sub[selector])
+
+/* The selector for a submessage field is the field index. */
+#define SUBH_F(h, f) SUBH(h, f->index_)
+
+static int32_t trygetsel(upb_handlers *h, const upb_fielddef *f,
+                         upb_handlertype_t type) {
+  upb_selector_t sel;
+  assert(!upb_handlers_isfrozen(h));
+  if (upb_handlers_msgdef(h) != upb_fielddef_containingtype(f)) {
+    upb_status_seterrf(
+        &h->status_, "type mismatch: field %s does not belong to message %s",
+        upb_fielddef_name(f), upb_msgdef_fullname(upb_handlers_msgdef(h)));
+    return -1;
+  }
+  if (!upb_handlers_getselector(f, type, &sel)) {
+    upb_status_seterrf(
+        &h->status_,
+        "type mismatch: cannot register handler type %d for field %s",
+        type, upb_fielddef_name(f));
+    return -1;
+  }
+  return sel;
+}
+
+static upb_selector_t handlers_getsel(upb_handlers *h, const upb_fielddef *f,
+                             upb_handlertype_t type) {
+  int32_t sel = trygetsel(h, f, type);
+  assert(sel >= 0);
+  return sel;
+}
+
+static const void **returntype(upb_handlers *h, const upb_fielddef *f,
+                               upb_handlertype_t type) {
+  return &h->table[handlers_getsel(h, f, type)].attr.return_closure_type_;
+}
+
+static bool doset(upb_handlers *h, int32_t sel, const upb_fielddef *f,
+                  upb_handlertype_t type, upb_func *func,
+                  upb_handlerattr *attr) {
+  upb_handlerattr set_attr = UPB_HANDLERATTR_INITIALIZER;
+  const void *closure_type;
+  const void **context_closure_type;
+
+  assert(!upb_handlers_isfrozen(h));
+
+  if (sel < 0) {
+    upb_status_seterrmsg(&h->status_,
+                         "incorrect handler type for this field.");
+    return false;
+  }
+
+  if (h->table[sel].func) {
+    upb_status_seterrmsg(&h->status_,
+                         "cannot change handler once it has been set.");
+    return false;
+  }
+
+  if (attr) {
+    set_attr = *attr;
+  }
+
+  /* Check that the given closure type matches the closure type that has been
+   * established for this context (if any). */
+  closure_type = upb_handlerattr_closuretype(&set_attr);
+
+  if (type == UPB_HANDLER_STRING) {
+    context_closure_type = returntype(h, f, UPB_HANDLER_STARTSTR);
+  } else if (f && upb_fielddef_isseq(f) &&
+             type != UPB_HANDLER_STARTSEQ &&
+             type != UPB_HANDLER_ENDSEQ) {
+    context_closure_type = returntype(h, f, UPB_HANDLER_STARTSEQ);
+  } else {
+    context_closure_type = &h->top_closure_type;
+  }
+
+  if (closure_type && *context_closure_type &&
+      closure_type != *context_closure_type) {
+    /* TODO(haberman): better message for debugging. */
+    if (f) {
+      upb_status_seterrf(&h->status_,
+                         "closure type does not match for field %s",
+                         upb_fielddef_name(f));
+    } else {
+      upb_status_seterrmsg(
+          &h->status_, "closure type does not match for message-level handler");
+    }
+    return false;
+  }
+
+  if (closure_type)
+    *context_closure_type = closure_type;
+
+  /* If this is a STARTSEQ or STARTSTR handler, check that the returned pointer
+   * matches any pre-existing expectations about what type is expected. */
+  if (type == UPB_HANDLER_STARTSEQ || type == UPB_HANDLER_STARTSTR) {
+    const void *return_type = upb_handlerattr_returnclosuretype(&set_attr);
+    const void *table_return_type =
+        upb_handlerattr_returnclosuretype(&h->table[sel].attr);
+    if (return_type && table_return_type && return_type != table_return_type) {
+      upb_status_seterrmsg(&h->status_, "closure return type does not match");
+      return false;
+    }
+
+    if (table_return_type && !return_type)
+      upb_handlerattr_setreturnclosuretype(&set_attr, table_return_type);
+  }
+
+  h->table[sel].func = (upb_func*)func;
+  h->table[sel].attr = set_attr;
+  return true;
+}
+
+/* Returns the effective closure type for this handler (which will propagate
+ * from outer frames if this frame has no START* handler).  Not implemented for
+ * UPB_HANDLER_STRING at the moment since this is not needed.  Returns NULL is
+ * the effective closure type is unspecified (either no handler was registered
+ * to specify it or the handler that was registered did not specify the closure
+ * type). */
+const void *effective_closure_type(upb_handlers *h, const upb_fielddef *f,
+                                   upb_handlertype_t type) {
+  const void *ret;
+  upb_selector_t sel;
+
+  assert(type != UPB_HANDLER_STRING);
+  ret = h->top_closure_type;
+
+  if (upb_fielddef_isseq(f) &&
+      type != UPB_HANDLER_STARTSEQ &&
+      type != UPB_HANDLER_ENDSEQ &&
+      h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSEQ)].func) {
+    ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr);
+  }
+
+  if (type == UPB_HANDLER_STRING &&
+      h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSTR)].func) {
+    ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr);
+  }
+
+  /* The effective type of the submessage; not used yet.
+   * if (type == SUBMESSAGE &&
+   *     h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSUBMSG)].func) {
+   *   ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr);
+   * } */
+
+  return ret;
+}
+
+/* Checks whether the START* handler specified by f & type is missing even
+ * though it is required to convert the established type of an outer frame
+ * ("closure_type") into the established type of an inner frame (represented in
+ * the return closure type of this handler's attr. */
+bool checkstart(upb_handlers *h, const upb_fielddef *f, upb_handlertype_t type,
+                upb_status *status) {
+  const void *closure_type;
+  const upb_handlerattr *attr;
+  const void *return_closure_type;
+
+  upb_selector_t sel = handlers_getsel(h, f, type);
+  if (h->table[sel].func) return true;
+  closure_type = effective_closure_type(h, f, type);
+  attr = &h->table[sel].attr;
+  return_closure_type = upb_handlerattr_returnclosuretype(attr);
+  if (closure_type && return_closure_type &&
+      closure_type != return_closure_type) {
+    upb_status_seterrf(status,
+                       "expected start handler to return sub type for field %f",
+                       upb_fielddef_name(f));
+    return false;
+  }
+  return true;
+}
+
+/* Public interface ***********************************************************/
+
+upb_handlers *upb_handlers_new(const upb_msgdef *md, const void *owner) {
+  int extra;
+  upb_handlers *h;
+
+  assert(upb_msgdef_isfrozen(md));
+
+  extra = sizeof(upb_handlers_tabent) * (md->selector_count - 1);
+  h = calloc(sizeof(*h) + extra, 1);
+  if (!h) return NULL;
+
+  h->msg = md;
+  upb_msgdef_ref(h->msg, h);
+  upb_status_clear(&h->status_);
+  h->sub = calloc(md->submsg_field_count, sizeof(*h->sub));
+  if (!h->sub) goto oom;
+  if (!upb_refcounted_init(upb_handlers_upcast_mutable(h), &vtbl, owner))
+    goto oom;
+  if (!upb_inttable_init(&h->cleanup_, UPB_CTYPE_FPTR)) goto oom;
+
+  /* calloc() above initialized all handlers to NULL. */
+  return h;
+
+oom:
+  freehandlers(upb_handlers_upcast_mutable(h));
+  return NULL;
+}
+
+const upb_handlers *upb_handlers_newfrozen(const upb_msgdef *m,
+                                           const void *owner,
+                                           upb_handlers_callback *callback,
+                                           const void *closure) {
+  dfs_state state;
+  upb_handlers *ret;
+  bool ok;
+  upb_refcounted *r;
+
+  state.callback = callback;
+  state.closure = closure;
+  if (!upb_inttable_init(&state.tab, UPB_CTYPE_PTR)) return NULL;
+
+  ret = newformsg(m, owner, &state);
+
+  upb_inttable_uninit(&state.tab);
+  if (!ret) return NULL;
+
+  r = upb_handlers_upcast_mutable(ret);
+  ok = upb_refcounted_freeze(&r, 1, NULL, UPB_MAX_HANDLER_DEPTH);
+  UPB_ASSERT_VAR(ok, ok);
+
+  return ret;
+}
+
+const upb_status *upb_handlers_status(upb_handlers *h) {
+  assert(!upb_handlers_isfrozen(h));
+  return &h->status_;
+}
+
+void upb_handlers_clearerr(upb_handlers *h) {
+  assert(!upb_handlers_isfrozen(h));
+  upb_status_clear(&h->status_);
+}
+
+#define SETTER(name, handlerctype, handlertype) \
+  bool upb_handlers_set ## name(upb_handlers *h, const upb_fielddef *f, \
+                                handlerctype func, upb_handlerattr *attr) { \
+    int32_t sel = trygetsel(h, f, handlertype); \
+    return doset(h, sel, f, handlertype, (upb_func*)func, attr); \
+  }
+
+SETTER(int32,       upb_int32_handlerfunc*,       UPB_HANDLER_INT32)
+SETTER(int64,       upb_int64_handlerfunc*,       UPB_HANDLER_INT64)
+SETTER(uint32,      upb_uint32_handlerfunc*,      UPB_HANDLER_UINT32)
+SETTER(uint64,      upb_uint64_handlerfunc*,      UPB_HANDLER_UINT64)
+SETTER(float,       upb_float_handlerfunc*,       UPB_HANDLER_FLOAT)
+SETTER(double,      upb_double_handlerfunc*,      UPB_HANDLER_DOUBLE)
+SETTER(bool,        upb_bool_handlerfunc*,        UPB_HANDLER_BOOL)
+SETTER(startstr,    upb_startstr_handlerfunc*,    UPB_HANDLER_STARTSTR)
+SETTER(string,      upb_string_handlerfunc*,      UPB_HANDLER_STRING)
+SETTER(endstr,      upb_endfield_handlerfunc*,    UPB_HANDLER_ENDSTR)
+SETTER(startseq,    upb_startfield_handlerfunc*,  UPB_HANDLER_STARTSEQ)
+SETTER(startsubmsg, upb_startfield_handlerfunc*,  UPB_HANDLER_STARTSUBMSG)
+SETTER(endsubmsg,   upb_endfield_handlerfunc*,    UPB_HANDLER_ENDSUBMSG)
+SETTER(endseq,      upb_endfield_handlerfunc*,    UPB_HANDLER_ENDSEQ)
+
+#undef SETTER
+
+bool upb_handlers_setstartmsg(upb_handlers *h, upb_startmsg_handlerfunc *func,
+                              upb_handlerattr *attr) {
+  return doset(h, UPB_STARTMSG_SELECTOR, NULL, UPB_HANDLER_INT32,
+               (upb_func *)func, attr);
+}
+
+bool upb_handlers_setendmsg(upb_handlers *h, upb_endmsg_handlerfunc *func,
+                            upb_handlerattr *attr) {
+  assert(!upb_handlers_isfrozen(h));
+  return doset(h, UPB_ENDMSG_SELECTOR, NULL, UPB_HANDLER_INT32,
+               (upb_func *)func, attr);
+}
+
+bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f,
+                                 const upb_handlers *sub) {
+  assert(sub);
+  assert(!upb_handlers_isfrozen(h));
+  assert(upb_fielddef_issubmsg(f));
+  if (SUBH_F(h, f)) return false;  /* Can't reset. */
+  if (upb_msgdef_upcast(upb_handlers_msgdef(sub)) != upb_fielddef_subdef(f)) {
+    return false;
+  }
+  SUBH_F(h, f) = sub;
+  upb_ref2(sub, h);
+  return true;
+}
+
+const upb_handlers *upb_handlers_getsubhandlers(const upb_handlers *h,
+                                                const upb_fielddef *f) {
+  assert(upb_fielddef_issubmsg(f));
+  return SUBH_F(h, f);
+}
+
+bool upb_handlers_getattr(const upb_handlers *h, upb_selector_t sel,
+                          upb_handlerattr *attr) {
+  if (!upb_handlers_gethandler(h, sel))
+    return false;
+  *attr = h->table[sel].attr;
+  return true;
+}
+
+const upb_handlers *upb_handlers_getsubhandlers_sel(const upb_handlers *h,
+                                                    upb_selector_t sel) {
+  /* STARTSUBMSG selector in sel is the field's selector base. */
+  return SUBH(h, sel - UPB_STATIC_SELECTOR_COUNT);
+}
+
+const upb_msgdef *upb_handlers_msgdef(const upb_handlers *h) { return h->msg; }
+
+bool upb_handlers_addcleanup(upb_handlers *h, void *p, upb_handlerfree *func) {
+  bool ok;
+  if (upb_inttable_lookupptr(&h->cleanup_, p, NULL)) {
+    return false;
+  }
+  ok = upb_inttable_insertptr(&h->cleanup_, p, upb_value_fptr(func));
+  UPB_ASSERT_VAR(ok, ok);
+  return true;
+}
+
+
+/* "Static" methods ***********************************************************/
+
+bool upb_handlers_freeze(upb_handlers *const*handlers, int n, upb_status *s) {
+  /* TODO: verify we have a transitive closure. */
+  int i;
+  for (i = 0; i < n; i++) {
+    upb_msg_field_iter j;
+    upb_handlers *h = handlers[i];
+
+    if (!upb_ok(&h->status_)) {
+      upb_status_seterrf(s, "handlers for message %s had error status: %s",
+                         upb_msgdef_fullname(upb_handlers_msgdef(h)),
+                         upb_status_errmsg(&h->status_));
+      return false;
+    }
+
+    /* Check that there are no closure mismatches due to missing Start* handlers
+     * or subhandlers with different type-level types. */
+    for(upb_msg_field_begin(&j, h->msg);
+        !upb_msg_field_done(&j);
+        upb_msg_field_next(&j)) {
+
+      const upb_fielddef *f = upb_msg_iter_field(&j);
+      if (upb_fielddef_isseq(f)) {
+        if (!checkstart(h, f, UPB_HANDLER_STARTSEQ, s))
+          return false;
+      }
+
+      if (upb_fielddef_isstring(f)) {
+        if (!checkstart(h, f, UPB_HANDLER_STARTSTR, s))
+          return false;
+      }
+
+      if (upb_fielddef_issubmsg(f)) {
+        bool hashandler = false;
+        if (upb_handlers_gethandler(
+                h, handlers_getsel(h, f, UPB_HANDLER_STARTSUBMSG)) ||
+            upb_handlers_gethandler(
+                h, handlers_getsel(h, f, UPB_HANDLER_ENDSUBMSG))) {
+          hashandler = true;
+        }
+
+        if (upb_fielddef_isseq(f) &&
+            (upb_handlers_gethandler(
+                 h, handlers_getsel(h, f, UPB_HANDLER_STARTSEQ)) ||
+             upb_handlers_gethandler(
+                 h, handlers_getsel(h, f, UPB_HANDLER_ENDSEQ)))) {
+          hashandler = true;
+        }
+
+        if (hashandler && !upb_handlers_getsubhandlers(h, f)) {
+          /* For now we add an empty subhandlers in this case.  It makes the
+           * decoder code generator simpler, because it only has to handle two
+           * cases (submessage has handlers or not) as opposed to three
+           * (submessage has handlers in enclosing message but no subhandlers).
+           *
+           * This makes parsing less efficient in the case that we want to
+           * notice a submessage but skip its contents (like if we're testing
+           * for submessage presence or counting the number of repeated
+           * submessages).  In this case we will end up parsing the submessage
+           * field by field and throwing away the results for each, instead of
+           * skipping the whole delimited thing at once.  If this is an issue we
+           * can revisit it, but do remember that this only arises when you have
+           * handlers (startseq/startsubmsg/endsubmsg/endseq) set for the
+           * submessage but no subhandlers.  The uses cases for this are
+           * limited. */
+          upb_handlers *sub = upb_handlers_new(upb_fielddef_msgsubdef(f), &sub);
+          upb_handlers_setsubhandlers(h, f, sub);
+          upb_handlers_unref(sub, &sub);
+        }
+
+        /* TODO(haberman): check type of submessage.
+         * This is slightly tricky; also consider whether we should check that
+         * they match at setsubhandlers time. */
+      }
+    }
+  }
+
+  if (!upb_refcounted_freeze((upb_refcounted*const*)handlers, n, s,
+                             UPB_MAX_HANDLER_DEPTH)) {
+    return false;
+  }
+
+  return true;
+}
+
+upb_handlertype_t upb_handlers_getprimitivehandlertype(const upb_fielddef *f) {
+  switch (upb_fielddef_type(f)) {
+    case UPB_TYPE_INT32:
+    case UPB_TYPE_ENUM: return UPB_HANDLER_INT32;
+    case UPB_TYPE_INT64: return UPB_HANDLER_INT64;
+    case UPB_TYPE_UINT32: return UPB_HANDLER_UINT32;
+    case UPB_TYPE_UINT64: return UPB_HANDLER_UINT64;
+    case UPB_TYPE_FLOAT: return UPB_HANDLER_FLOAT;
+    case UPB_TYPE_DOUBLE: return UPB_HANDLER_DOUBLE;
+    case UPB_TYPE_BOOL: return UPB_HANDLER_BOOL;
+    default: assert(false); return -1;  /* Invalid input. */
+  }
+}
+
+bool upb_handlers_getselector(const upb_fielddef *f, upb_handlertype_t type,
+                              upb_selector_t *s) {
+  switch (type) {
+    case UPB_HANDLER_INT32:
+    case UPB_HANDLER_INT64:
+    case UPB_HANDLER_UINT32:
+    case UPB_HANDLER_UINT64:
+    case UPB_HANDLER_FLOAT:
+    case UPB_HANDLER_DOUBLE:
+    case UPB_HANDLER_BOOL:
+      if (!upb_fielddef_isprimitive(f) ||
+          upb_handlers_getprimitivehandlertype(f) != type)
+        return false;
+      *s = f->selector_base;
+      break;
+    case UPB_HANDLER_STRING:
+      if (upb_fielddef_isstring(f)) {
+        *s = f->selector_base;
+      } else if (upb_fielddef_lazy(f)) {
+        *s = f->selector_base + 3;
+      } else {
+        return false;
+      }
+      break;
+    case UPB_HANDLER_STARTSTR:
+      if (upb_fielddef_isstring(f) || upb_fielddef_lazy(f)) {
+        *s = f->selector_base + 1;
+      } else {
+        return false;
+      }
+      break;
+    case UPB_HANDLER_ENDSTR:
+      if (upb_fielddef_isstring(f) || upb_fielddef_lazy(f)) {
+        *s = f->selector_base + 2;
+      } else {
+        return false;
+      }
+      break;
+    case UPB_HANDLER_STARTSEQ:
+      if (!upb_fielddef_isseq(f)) return false;
+      *s = f->selector_base - 2;
+      break;
+    case UPB_HANDLER_ENDSEQ:
+      if (!upb_fielddef_isseq(f)) return false;
+      *s = f->selector_base - 1;
+      break;
+    case UPB_HANDLER_STARTSUBMSG:
+      if (!upb_fielddef_issubmsg(f)) return false;
+      /* Selectors for STARTSUBMSG are at the beginning of the table so that the
+       * selector can also be used as an index into the "sub" array of
+       * subhandlers.  The indexes for the two into these two tables are the
+       * same, except that in the handler table the static selectors come first. */
+      *s = f->index_ + UPB_STATIC_SELECTOR_COUNT;
+      break;
+    case UPB_HANDLER_ENDSUBMSG:
+      if (!upb_fielddef_issubmsg(f)) return false;
+      *s = f->selector_base;
+      break;
+  }
+  assert((size_t)*s < upb_fielddef_containingtype(f)->selector_count);
+  return true;
+}
+
+uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f) {
+  return upb_fielddef_isseq(f) ? 2 : 0;
+}
+
+uint32_t upb_handlers_selectorcount(const upb_fielddef *f) {
+  uint32_t ret = 1;
+  if (upb_fielddef_isseq(f)) ret += 2;    /* STARTSEQ/ENDSEQ */
+  if (upb_fielddef_isstring(f)) ret += 2; /* [STRING]/STARTSTR/ENDSTR */
+  if (upb_fielddef_issubmsg(f)) {
+    /* ENDSUBMSG (STARTSUBMSG is at table beginning) */
+    ret += 0;
+    if (upb_fielddef_lazy(f)) {
+      /* STARTSTR/ENDSTR/STRING (for lazy) */
+      ret += 3;
+    }
+  }
+  return ret;
+}
+
+
+/* upb_handlerattr ************************************************************/
+
+void upb_handlerattr_init(upb_handlerattr *attr) {
+  upb_handlerattr from = UPB_HANDLERATTR_INITIALIZER;
+  memcpy(attr, &from, sizeof(*attr));
+}
+
+void upb_handlerattr_uninit(upb_handlerattr *attr) {
+  UPB_UNUSED(attr);
+}
+
+bool upb_handlerattr_sethandlerdata(upb_handlerattr *attr, const void *hd) {
+  attr->handler_data_ = hd;
+  return true;
+}
+
+bool upb_handlerattr_setclosuretype(upb_handlerattr *attr, const void *type) {
+  attr->closure_type_ = type;
+  return true;
+}
+
+const void *upb_handlerattr_closuretype(const upb_handlerattr *attr) {
+  return attr->closure_type_;
+}
+
+bool upb_handlerattr_setreturnclosuretype(upb_handlerattr *attr,
+                                          const void *type) {
+  attr->return_closure_type_ = type;
+  return true;
+}
+
+const void *upb_handlerattr_returnclosuretype(const upb_handlerattr *attr) {
+  return attr->return_closure_type_;
+}
+
+bool upb_handlerattr_setalwaysok(upb_handlerattr *attr, bool alwaysok) {
+  attr->alwaysok_ = alwaysok;
+  return true;
+}
+
+bool upb_handlerattr_alwaysok(const upb_handlerattr *attr) {
+  return attr->alwaysok_;
+}
+
+/* upb_bufhandle **************************************************************/
+
+size_t upb_bufhandle_objofs(const upb_bufhandle *h) {
+  return h->objofs_;
+}
+
+/* upb_byteshandler ***********************************************************/
+
+void upb_byteshandler_init(upb_byteshandler* h) {
+  memset(h, 0, sizeof(*h));
+}
+
+/* For when we support handlerfree callbacks. */
+void upb_byteshandler_uninit(upb_byteshandler* h) {
+  UPB_UNUSED(h);
+}
+
+bool upb_byteshandler_setstartstr(upb_byteshandler *h,
+                                  upb_startstr_handlerfunc *func, void *d) {
+  h->table[UPB_STARTSTR_SELECTOR].func = (upb_func*)func;
+  h->table[UPB_STARTSTR_SELECTOR].attr.handler_data_ = d;
+  return true;
+}
+
+bool upb_byteshandler_setstring(upb_byteshandler *h,
+                                upb_string_handlerfunc *func, void *d) {
+  h->table[UPB_STRING_SELECTOR].func = (upb_func*)func;
+  h->table[UPB_STRING_SELECTOR].attr.handler_data_ = d;
+  return true;
+}
+
+bool upb_byteshandler_setendstr(upb_byteshandler *h,
+                                upb_endfield_handlerfunc *func, void *d) {
+  h->table[UPB_ENDSTR_SELECTOR].func = (upb_func*)func;
+  h->table[UPB_ENDSTR_SELECTOR].attr.handler_data_ = d;
+  return true;
+}
+/*
+** upb::RefCounted Implementation
+**
+** Our key invariants are:
+** 1. reference cycles never span groups
+** 2. for ref2(to, from), we increment to's count iff group(from) != group(to)
+**
+** The previous two are how we avoid leaking cycles.  Other important
+** invariants are:
+** 3. for mutable objects "from" and "to", if there exists a ref2(to, from)
+**    this implies group(from) == group(to).  (In practice, what we implement
+**    is even stronger; "from" and "to" will share a group if there has *ever*
+**    been a ref2(to, from), but all that is necessary for correctness is the
+**    weaker one).
+** 4. mutable and immutable objects are never in the same group.
+*/
+
+
+#include <setjmp.h>
+#include <stdlib.h>
+
+static void freeobj(upb_refcounted *o);
+
+const char untracked_val;
+const void *UPB_UNTRACKED_REF = &untracked_val;
+
+/* arch-specific atomic primitives  *******************************************/
+
+#ifdef UPB_THREAD_UNSAFE /*---------------------------------------------------*/
+
+static void atomic_inc(uint32_t *a) { (*a)++; }
+static bool atomic_dec(uint32_t *a) { return --(*a) == 0; }
+
+#elif defined(__GNUC__) || defined(__clang__) /*------------------------------*/
+
+static void atomic_inc(uint32_t *a) { __sync_fetch_and_add(a, 1); }
+static bool atomic_dec(uint32_t *a) { return __sync_sub_and_fetch(a, 1) == 0; }
+
+#elif defined(WIN32) /*-------------------------------------------------------*/
+
+#include <Windows.h>
+
+static void atomic_inc(upb_atomic_t *a) { InterlockedIncrement(&a->val); }
+static bool atomic_dec(upb_atomic_t *a) {
+  return InterlockedDecrement(&a->val) == 0;
+}
+
+#else
+#error Atomic primitives not defined for your platform/CPU.  \
+       Implement them or compile with UPB_THREAD_UNSAFE.
+#endif
+
+/* All static objects point to this refcount.
+ * It is special-cased in ref/unref below.  */
+uint32_t static_refcount = -1;
+
+/* We can avoid atomic ops for statically-declared objects.
+ * This is a minor optimization but nice since we can avoid degrading under
+ * contention in this case. */
+
+static void refgroup(uint32_t *group) {
+  if (group != &static_refcount)
+    atomic_inc(group);
+}
+
+static bool unrefgroup(uint32_t *group) {
+  if (group == &static_refcount) {
+    return false;
+  } else {
+    return atomic_dec(group);
+  }
+}
+
+
+/* Reference tracking (debug only) ********************************************/
+
+#ifdef UPB_DEBUG_REFS
+
+#ifdef UPB_THREAD_UNSAFE
+
+static void upb_lock() {}
+static void upb_unlock() {}
+
+#else
+
+/* User must define functions that lock/unlock a global mutex and link this
+ * file against them. */
+void upb_lock();
+void upb_unlock();
+
+#endif
+
+/* UPB_DEBUG_REFS mode counts on being able to malloc() memory in some
+ * code-paths that can normally never fail, like upb_refcounted_ref().  Since
+ * we have no way to propagage out-of-memory errors back to the user, and since
+ * these errors can only occur in UPB_DEBUG_REFS mode, we immediately fail. */
+#define CHECK_OOM(predicate) if (!(predicate)) { assert(predicate); exit(1); }
+
+typedef struct {
+  int count;  /* How many refs there are (duplicates only allowed for ref2). */
+  bool is_ref2;
+} trackedref;
+
+static trackedref *trackedref_new(bool is_ref2) {
+  trackedref *ret = malloc(sizeof(*ret));
+  CHECK_OOM(ret);
+  ret->count = 1;
+  ret->is_ref2 = is_ref2;
+  return ret;
+}
+
+static void track(const upb_refcounted *r, const void *owner, bool ref2) {
+  upb_value v;
+
+  assert(owner);
+  if (owner == UPB_UNTRACKED_REF) return;
+
+  upb_lock();
+  if (upb_inttable_lookupptr(r->refs, owner, &v)) {
+    trackedref *ref = upb_value_getptr(v);
+    /* Since we allow multiple ref2's for the same to/from pair without
+     * allocating separate memory for each one, we lose the fine-grained
+     * tracking behavior we get with regular refs.  Since ref2s only happen
+     * inside upb, we'll accept this limitation until/unless there is a really
+     * difficult upb-internal bug that can't be figured out without it. */
+    assert(ref2);
+    assert(ref->is_ref2);
+    ref->count++;
+  } else {
+    trackedref *ref = trackedref_new(ref2);
+    bool ok = upb_inttable_insertptr(r->refs, owner, upb_value_ptr(ref));
+    CHECK_OOM(ok);
+    if (ref2) {
+      /* We know this cast is safe when it is a ref2, because it's coming from
+       * another refcounted object. */
+      const upb_refcounted *from = owner;
+      assert(!upb_inttable_lookupptr(from->ref2s, r, NULL));
+      ok = upb_inttable_insertptr(from->ref2s, r, upb_value_ptr(NULL));
+      CHECK_OOM(ok);
+    }
+  }
+  upb_unlock();
+}
+
+static void untrack(const upb_refcounted *r, const void *owner, bool ref2) {
+  upb_value v;
+  bool found;
+  trackedref *ref;
+
+  assert(owner);
+  if (owner == UPB_UNTRACKED_REF) return;
+
+  upb_lock();
+  found = upb_inttable_lookupptr(r->refs, owner, &v);
+  /* This assert will fail if an owner attempts to release a ref it didn't have. */
+  UPB_ASSERT_VAR(found, found);
+  ref = upb_value_getptr(v);
+  assert(ref->is_ref2 == ref2);
+  if (--ref->count == 0) {
+    free(ref);
+    upb_inttable_removeptr(r->refs, owner, NULL);
+    if (ref2) {
+      /* We know this cast is safe when it is a ref2, because it's coming from
+       * another refcounted object. */
+      const upb_refcounted *from = owner;
+      bool removed = upb_inttable_removeptr(from->ref2s, r, NULL);
+      assert(removed);
+    }
+  }
+  upb_unlock();
+}
+
+static void checkref(const upb_refcounted *r, const void *owner, bool ref2) {
+  upb_value v;
+  bool found;
+  trackedref *ref;
+
+  upb_lock();
+  found = upb_inttable_lookupptr(r->refs, owner, &v);
+  UPB_ASSERT_VAR(found, found);
+  ref = upb_value_getptr(v);
+  assert(ref->is_ref2 == ref2);
+  upb_unlock();
+}
+
+/* Populates the given UPB_CTYPE_INT32 inttable with counts of ref2's that
+ * originate from the given owner. */
+static void getref2s(const upb_refcounted *owner, upb_inttable *tab) {
+  upb_inttable_iter i;
+
+  upb_lock();
+  upb_inttable_begin(&i, owner->ref2s);
+  for(; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    upb_value v;
+    upb_value count;
+    trackedref *ref;
+    bool ok;
+    bool found;
+
+    upb_refcounted *to = (upb_refcounted*)upb_inttable_iter_key(&i);
+
+    /* To get the count we need to look in the target's table. */
+    found = upb_inttable_lookupptr(to->refs, owner, &v);
+    assert(found);
+    ref = upb_value_getptr(v);
+    count = upb_value_int32(ref->count);
+
+    ok = upb_inttable_insertptr(tab, to, count);
+    CHECK_OOM(ok);
+  }
+  upb_unlock();
+}
+
+typedef struct {
+  upb_inttable ref2;
+  const upb_refcounted *obj;
+} check_state;
+
+static void visit_check(const upb_refcounted *obj, const upb_refcounted *subobj,
+                        void *closure) {
+  check_state *s = closure;
+  upb_inttable *ref2 = &s->ref2;
+  upb_value v;
+  bool removed;
+  int32_t newcount;
+
+  assert(obj == s->obj);
+  assert(subobj);
+  removed = upb_inttable_removeptr(ref2, subobj, &v);
+  /* The following assertion will fail if the visit() function visits a subobj
+   * that it did not have a ref2 on, or visits the same subobj too many times. */
+  assert(removed);
+  newcount = upb_value_getint32(v) - 1;
+  if (newcount > 0) {
+    upb_inttable_insert(ref2, (uintptr_t)subobj, upb_value_int32(newcount));
+  }
+}
+
+static void visit(const upb_refcounted *r, upb_refcounted_visit *v,
+                  void *closure) {
+  bool ok;
+
+  /* In DEBUG_REFS mode we know what existing ref2 refs there are, so we know
+   * exactly the set of nodes that visit() should visit.  So we verify visit()'s
+   * correctness here. */
+  check_state state;
+  state.obj = r;
+  ok = upb_inttable_init(&state.ref2, UPB_CTYPE_INT32);
+  CHECK_OOM(ok);
+  getref2s(r, &state.ref2);
+
+  /* This should visit any children in the ref2 table. */
+  if (r->vtbl->visit) r->vtbl->visit(r, visit_check, &state);
+
+  /* This assertion will fail if the visit() function missed any children. */
+  assert(upb_inttable_count(&state.ref2) == 0);
+  upb_inttable_uninit(&state.ref2);
+  if (r->vtbl->visit) r->vtbl->visit(r, v, closure);
+}
+
+static bool trackinit(upb_refcounted *r) {
+  r->refs = malloc(sizeof(*r->refs));
+  r->ref2s = malloc(sizeof(*r->ref2s));
+  if (!r->refs || !r->ref2s) goto err1;
+
+  if (!upb_inttable_init(r->refs, UPB_CTYPE_PTR)) goto err1;
+  if (!upb_inttable_init(r->ref2s, UPB_CTYPE_PTR)) goto err2;
+  return true;
+
+err2:
+  upb_inttable_uninit(r->refs);
+err1:
+  free(r->refs);
+  free(r->ref2s);
+  return false;
+}
+
+static void trackfree(const upb_refcounted *r) {
+  upb_inttable_uninit(r->refs);
+  upb_inttable_uninit(r->ref2s);
+  free(r->refs);
+  free(r->ref2s);
+}
+
+#else
+
+static void track(const upb_refcounted *r, const void *owner, bool ref2) {
+  UPB_UNUSED(r);
+  UPB_UNUSED(owner);
+  UPB_UNUSED(ref2);
+}
+
+static void untrack(const upb_refcounted *r, const void *owner, bool ref2) {
+  UPB_UNUSED(r);
+  UPB_UNUSED(owner);
+  UPB_UNUSED(ref2);
+}
+
+static void checkref(const upb_refcounted *r, const void *owner, bool ref2) {
+  UPB_UNUSED(r);
+  UPB_UNUSED(owner);
+  UPB_UNUSED(ref2);
+}
+
+static bool trackinit(upb_refcounted *r) {
+  UPB_UNUSED(r);
+  return true;
+}
+
+static void trackfree(const upb_refcounted *r) {
+  UPB_UNUSED(r);
+}
+
+static void visit(const upb_refcounted *r, upb_refcounted_visit *v,
+                  void *closure) {
+  if (r->vtbl->visit) r->vtbl->visit(r, v, closure);
+}
+
+#endif  /* UPB_DEBUG_REFS */
+
+
+/* freeze() *******************************************************************/
+
+/* The freeze() operation is by far the most complicated part of this scheme.
+ * We compute strongly-connected components and then mutate the graph such that
+ * we preserve the invariants documented at the top of this file.  And we must
+ * handle out-of-memory errors gracefully (without leaving the graph
+ * inconsistent), which adds to the fun. */
+
+/* The state used by the freeze operation (shared across many functions). */
+typedef struct {
+  int depth;
+  int maxdepth;
+  uint64_t index;
+  /* Maps upb_refcounted* -> attributes (color, etc).  attr layout varies by
+   * color. */
+  upb_inttable objattr;
+  upb_inttable stack;   /* stack of upb_refcounted* for Tarjan's algorithm. */
+  upb_inttable groups;  /* array of uint32_t*, malloc'd refcounts for new groups */
+  upb_status *status;
+  jmp_buf err;
+} tarjan;
+
+static void release_ref2(const upb_refcounted *obj,
+                         const upb_refcounted *subobj,
+                         void *closure);
+
+/* Node attributes -----------------------------------------------------------*/
+
+/* After our analysis phase all nodes will be either GRAY or WHITE. */
+
+typedef enum {
+  BLACK = 0,  /* Object has not been seen. */
+  GRAY,   /* Object has been found via a refgroup but may not be reachable. */
+  GREEN,  /* Object is reachable and is currently on the Tarjan stack. */
+  WHITE   /* Object is reachable and has been assigned a group (SCC). */
+} color_t;
+
+UPB_NORETURN static void err(tarjan *t) { longjmp(t->err, 1); }
+UPB_NORETURN static void oom(tarjan *t) {
+  upb_status_seterrmsg(t->status, "out of memory");
+  err(t);
+}
+
+static uint64_t trygetattr(const tarjan *t, const upb_refcounted *r) {
+  upb_value v;
+  return upb_inttable_lookupptr(&t->objattr, r, &v) ?
+      upb_value_getuint64(v) : 0;
+}
+
+static uint64_t getattr(const tarjan *t, const upb_refcounted *r) {
+  upb_value v;
+  bool found = upb_inttable_lookupptr(&t->objattr, r, &v);
+  UPB_ASSERT_VAR(found, found);
+  return upb_value_getuint64(v);
+}
+
+static void setattr(tarjan *t, const upb_refcounted *r, uint64_t attr) {
+  upb_inttable_removeptr(&t->objattr, r, NULL);
+  upb_inttable_insertptr(&t->objattr, r, upb_value_uint64(attr));
+}
+
+static color_t color(tarjan *t, const upb_refcounted *r) {
+  return trygetattr(t, r) & 0x3;  /* Color is always stored in the low 2 bits. */
+}
+
+static void set_gray(tarjan *t, const upb_refcounted *r) {
+  assert(color(t, r) == BLACK);
+  setattr(t, r, GRAY);
+}
+
+/* Pushes an obj onto the Tarjan stack and sets it to GREEN. */
+static void push(tarjan *t, const upb_refcounted *r) {
+  assert(color(t, r) == BLACK || color(t, r) == GRAY);
+  /* This defines the attr layout for the GREEN state.  "index" and "lowlink"
+   * get 31 bits, which is plenty (limit of 2B objects frozen at a time). */
+  setattr(t, r, GREEN | (t->index << 2) | (t->index << 33));
+  if (++t->index == 0x80000000) {
+    upb_status_seterrmsg(t->status, "too many objects to freeze");
+    err(t);
+  }
+  upb_inttable_push(&t->stack, upb_value_ptr((void*)r));
+}
+
+/* Pops an obj from the Tarjan stack and sets it to WHITE, with a ptr to its
+ * SCC group. */
+static upb_refcounted *pop(tarjan *t) {
+  upb_refcounted *r = upb_value_getptr(upb_inttable_pop(&t->stack));
+  assert(color(t, r) == GREEN);
+  /* This defines the attr layout for nodes in the WHITE state.
+   * Top of group stack is [group, NULL]; we point at group. */
+  setattr(t, r, WHITE | (upb_inttable_count(&t->groups) - 2) << 8);
+  return r;
+}
+
+static void tarjan_newgroup(tarjan *t) {
+  uint32_t *group = malloc(sizeof(*group));
+  if (!group) oom(t);
+  /* Push group and empty group leader (we'll fill in leader later). */
+  if (!upb_inttable_push(&t->groups, upb_value_ptr(group)) ||
+      !upb_inttable_push(&t->groups, upb_value_ptr(NULL))) {
+    free(group);
+    oom(t);
+  }
+  *group = 0;
+}
+
+static uint32_t idx(tarjan *t, const upb_refcounted *r) {
+  assert(color(t, r) == GREEN);
+  return (getattr(t, r) >> 2) & 0x7FFFFFFF;
+}
+
+static uint32_t lowlink(tarjan *t, const upb_refcounted *r) {
+  if (color(t, r) == GREEN) {
+    return getattr(t, r) >> 33;
+  } else {
+    return UINT32_MAX;
+  }
+}
+
+static void set_lowlink(tarjan *t, const upb_refcounted *r, uint32_t lowlink) {
+  assert(color(t, r) == GREEN);
+  setattr(t, r, ((uint64_t)lowlink << 33) | (getattr(t, r) & 0x1FFFFFFFF));
+}
+
+static uint32_t *group(tarjan *t, upb_refcounted *r) {
+  uint64_t groupnum;
+  upb_value v;
+  bool found;
+
+  assert(color(t, r) == WHITE);
+  groupnum = getattr(t, r) >> 8;
+  found = upb_inttable_lookup(&t->groups, groupnum, &v);
+  UPB_ASSERT_VAR(found, found);
+  return upb_value_getptr(v);
+}
+
+/* If the group leader for this object's group has not previously been set,
+ * the given object is assigned to be its leader. */
+static upb_refcounted *groupleader(tarjan *t, upb_refcounted *r) {
+  uint64_t leader_slot;
+  upb_value v;
+  bool found;
+
+  assert(color(t, r) == WHITE);
+  leader_slot = (getattr(t, r) >> 8) + 1;
+  found = upb_inttable_lookup(&t->groups, leader_slot, &v);
+  UPB_ASSERT_VAR(found, found);
+  if (upb_value_getptr(v)) {
+    return upb_value_getptr(v);
+  } else {
+    upb_inttable_remove(&t->groups, leader_slot, NULL);
+    upb_inttable_insert(&t->groups, leader_slot, upb_value_ptr(r));
+    return r;
+  }
+}
+
+
+/* Tarjan's algorithm --------------------------------------------------------*/
+
+/* See:
+ *   http://en.wikipedia.org/wiki/Tarjan%27s_strongly_connected_components_algorithm */
+static void do_tarjan(const upb_refcounted *obj, tarjan *t);
+
+static void tarjan_visit(const upb_refcounted *obj,
+                         const upb_refcounted *subobj,
+                         void *closure) {
+  tarjan *t = closure;
+  if (++t->depth > t->maxdepth) {
+    upb_status_seterrf(t->status, "graph too deep to freeze (%d)", t->maxdepth);
+    err(t);
+  } else if (subobj->is_frozen || color(t, subobj) == WHITE) {
+    /* Do nothing: we don't want to visit or color already-frozen nodes,
+     * and WHITE nodes have already been assigned a SCC. */
+  } else if (color(t, subobj) < GREEN) {
+    /* Subdef has not yet been visited; recurse on it. */
+    do_tarjan(subobj, t);
+    set_lowlink(t, obj, UPB_MIN(lowlink(t, obj), lowlink(t, subobj)));
+  } else if (color(t, subobj) == GREEN) {
+    /* Subdef is in the stack and hence in the current SCC. */
+    set_lowlink(t, obj, UPB_MIN(lowlink(t, obj), idx(t, subobj)));
+  }
+  --t->depth;
+}
+
+static void do_tarjan(const upb_refcounted *obj, tarjan *t) {
+  if (color(t, obj) == BLACK) {
+    /* We haven't seen this object's group; mark the whole group GRAY. */
+    const upb_refcounted *o = obj;
+    do { set_gray(t, o); } while ((o = o->next) != obj);
+  }
+
+  push(t, obj);
+  visit(obj, tarjan_visit, t);
+  if (lowlink(t, obj) == idx(t, obj)) {
+    tarjan_newgroup(t);
+    while (pop(t) != obj)
+      ;
+  }
+}
+
+
+/* freeze() ------------------------------------------------------------------*/
+
+static void crossref(const upb_refcounted *r, const upb_refcounted *subobj,
+                     void *_t) {
+  tarjan *t = _t;
+  assert(color(t, r) > BLACK);
+  if (color(t, subobj) > BLACK && r->group != subobj->group) {
+    /* Previously this ref was not reflected in subobj->group because they
+     * were in the same group; now that they are split a ref must be taken. */
+    refgroup(subobj->group);
+  }
+}
+
+static bool freeze(upb_refcounted *const*roots, int n, upb_status *s,
+                   int maxdepth) {
+  volatile bool ret = false;
+  int i;
+  upb_inttable_iter iter;
+
+  /* We run in two passes so that we can allocate all memory before performing
+   * any mutation of the input -- this allows us to leave the input unchanged
+   * in the case of memory allocation failure. */
+  tarjan t;
+  t.index = 0;
+  t.depth = 0;
+  t.maxdepth = maxdepth;
+  t.status = s;
+  if (!upb_inttable_init(&t.objattr, UPB_CTYPE_UINT64)) goto err1;
+  if (!upb_inttable_init(&t.stack, UPB_CTYPE_PTR)) goto err2;
+  if (!upb_inttable_init(&t.groups, UPB_CTYPE_PTR)) goto err3;
+  if (setjmp(t.err) != 0) goto err4;
+
+
+  for (i = 0; i < n; i++) {
+    if (color(&t, roots[i]) < GREEN) {
+      do_tarjan(roots[i], &t);
+    }
+  }
+
+  /* If we've made it this far, no further errors are possible so it's safe to
+   * mutate the objects without risk of leaving them in an inconsistent state. */
+  ret = true;
+
+  /* The transformation that follows requires care.  The preconditions are:
+   * - all objects in attr map are WHITE or GRAY, and are in mutable groups
+   *   (groups of all mutable objs)
+   * - no ref2(to, from) refs have incremented count(to) if both "to" and
+   *   "from" are in our attr map (this follows from invariants (2) and (3)) */
+
+  /* Pass 1: we remove WHITE objects from their mutable groups, and add them to
+   * new groups  according to the SCC's we computed.  These new groups will
+   * consist of only frozen objects.  None will be immediately collectible,
+   * because WHITE objects are by definition reachable from one of "roots",
+   * which the caller must own refs on. */
+  upb_inttable_begin(&iter, &t.objattr);
+  for(; !upb_inttable_done(&iter); upb_inttable_next(&iter)) {
+    upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&iter);
+    /* Since removal from a singly-linked list requires access to the object's
+     * predecessor, we consider obj->next instead of obj for moving.  With the
+     * while() loop we guarantee that we will visit every node's predecessor.
+     * Proof:
+     *  1. every node's predecessor is in our attr map.
+     *  2. though the loop body may change a node's predecessor, it will only
+     *     change it to be the node we are currently operating on, so with a
+     *     while() loop we guarantee ourselves the chance to remove each node. */
+    while (color(&t, obj->next) == WHITE &&
+           group(&t, obj->next) != obj->next->group) {
+      upb_refcounted *leader;
+
+      /* Remove from old group. */
+      upb_refcounted *move = obj->next;
+      if (obj == move) {
+        /* Removing the last object from a group. */
+        assert(*obj->group == obj->individual_count);
+        free(obj->group);
+      } else {
+        obj->next = move->next;
+        /* This may decrease to zero; we'll collect GRAY objects (if any) that
+         * remain in the group in the third pass. */
+        assert(*move->group >= move->individual_count);
+        *move->group -= move->individual_count;
+      }
+
+      /* Add to new group. */
+      leader = groupleader(&t, move);
+      if (move == leader) {
+        /* First object added to new group is its leader. */
+        move->group = group(&t, move);
+        move->next = move;
+        *move->group = move->individual_count;
+      } else {
+        /* Group already has at least one object in it. */
+        assert(leader->group == group(&t, move));
+        move->group = group(&t, move);
+        move->next = leader->next;
+        leader->next = move;
+        *move->group += move->individual_count;
+      }
+
+      move->is_frozen = true;
+    }
+  }
+
+  /* Pass 2: GRAY and WHITE objects "obj" with ref2(to, obj) references must
+   * increment count(to) if group(obj) != group(to) (which could now be the
+   * case if "to" was just frozen). */
+  upb_inttable_begin(&iter, &t.objattr);
+  for(; !upb_inttable_done(&iter); upb_inttable_next(&iter)) {
+    upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&iter);
+    visit(obj, crossref, &t);
+  }
+
+  /* Pass 3: GRAY objects are collected if their group's refcount dropped to
+   * zero when we removed its white nodes.  This can happen if they had only
+   * been kept alive by virtue of sharing a group with an object that was just
+   * frozen.
+   *
+   * It is important that we do this last, since the GRAY object's free()
+   * function could call unref2() on just-frozen objects, which will decrement
+   * refs that were added in pass 2. */
+  upb_inttable_begin(&iter, &t.objattr);
+  for(; !upb_inttable_done(&iter); upb_inttable_next(&iter)) {
+    upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&iter);
+    if (obj->group == NULL || *obj->group == 0) {
+      if (obj->group) {
+        upb_refcounted *o;
+
+        /* We eagerly free() the group's count (since we can't easily determine
+         * the group's remaining size it's the easiest way to ensure it gets
+         * done). */
+        free(obj->group);
+
+        /* Visit to release ref2's (done in a separate pass since release_ref2
+         * depends on o->group being unmodified so it can test merged()). */
+        o = obj;
+        do { visit(o, release_ref2, NULL); } while ((o = o->next) != obj);
+
+        /* Mark "group" fields as NULL so we know to free the objects later in
+         * this loop, but also don't try to delete the group twice. */
+        o = obj;
+        do { o->group = NULL; } while ((o = o->next) != obj);
+      }
+      freeobj(obj);
+    }
+  }
+
+err4:
+  if (!ret) {
+    upb_inttable_begin(&iter, &t.groups);
+    for(; !upb_inttable_done(&iter); upb_inttable_next(&iter))
+      free(upb_value_getptr(upb_inttable_iter_value(&iter)));
+  }
+  upb_inttable_uninit(&t.groups);
+err3:
+  upb_inttable_uninit(&t.stack);
+err2:
+  upb_inttable_uninit(&t.objattr);
+err1:
+  return ret;
+}
+
+
+/* Misc internal functions  ***************************************************/
+
+static bool merged(const upb_refcounted *r, const upb_refcounted *r2) {
+  return r->group == r2->group;
+}
+
+static void merge(upb_refcounted *r, upb_refcounted *from) {
+  upb_refcounted *base;
+  upb_refcounted *tmp;
+
+  if (merged(r, from)) return;
+  *r->group += *from->group;
+  free(from->group);
+  base = from;
+
+  /* Set all refcount pointers in the "from" chain to the merged refcount.
+   *
+   * TODO(haberman): this linear algorithm can result in an overall O(n^2) bound
+   * if the user continuously extends a group by one object.  Prevent this by
+   * using one of the techniques in this paper:
+   *     ftp://www.ncedc.org/outgoing/geomorph/dino/orals/p245-tarjan.pdf */
+  do { from->group = r->group; } while ((from = from->next) != base);
+
+  /* Merge the two circularly linked lists by swapping their next pointers. */
+  tmp = r->next;
+  r->next = base->next;
+  base->next = tmp;
+}
+
+static void unref(const upb_refcounted *r);
+
+static void release_ref2(const upb_refcounted *obj,
+                         const upb_refcounted *subobj,
+                         void *closure) {
+  UPB_UNUSED(closure);
+  untrack(subobj, obj, true);
+  if (!merged(obj, subobj)) {
+    assert(subobj->is_frozen);
+    unref(subobj);
+  }
+}
+
+static void unref(const upb_refcounted *r) {
+  if (unrefgroup(r->group)) {
+    const upb_refcounted *o;
+
+    free(r->group);
+
+    /* In two passes, since release_ref2 needs a guarantee that any subobjs
+     * are alive. */
+    o = r;
+    do { visit(o, release_ref2, NULL); } while((o = o->next) != r);
+
+    o = r;
+    do {
+      const upb_refcounted *next = o->next;
+      assert(o->is_frozen || o->individual_count == 0);
+      freeobj((upb_refcounted*)o);
+      o = next;
+    } while(o != r);
+  }
+}
+
+static void freeobj(upb_refcounted *o) {
+  trackfree(o);
+  o->vtbl->free((upb_refcounted*)o);
+}
+
+
+/* Public interface ***********************************************************/
+
+bool upb_refcounted_init(upb_refcounted *r,
+                         const struct upb_refcounted_vtbl *vtbl,
+                         const void *owner) {
+#ifndef NDEBUG
+  /* Endianness check.  This is unrelated to upb_refcounted, it's just a
+   * convenient place to put the check that we can be assured will run for
+   * basically every program using upb. */
+  const int x = 1;
+#ifdef UPB_BIG_ENDIAN
+  assert(*(char*)&x != 1);
+#else
+  assert(*(char*)&x == 1);
+#endif
+#endif
+
+  r->next = r;
+  r->vtbl = vtbl;
+  r->individual_count = 0;
+  r->is_frozen = false;
+  r->group = malloc(sizeof(*r->group));
+  if (!r->group) return false;
+  *r->group = 0;
+  if (!trackinit(r)) {
+    free(r->group);
+    return false;
+  }
+  upb_refcounted_ref(r, owner);
+  return true;
+}
+
+bool upb_refcounted_isfrozen(const upb_refcounted *r) {
+  return r->is_frozen;
+}
+
+void upb_refcounted_ref(const upb_refcounted *r, const void *owner) {
+  track(r, owner, false);
+  if (!r->is_frozen)
+    ((upb_refcounted*)r)->individual_count++;
+  refgroup(r->group);
+}
+
+void upb_refcounted_unref(const upb_refcounted *r, const void *owner) {
+  untrack(r, owner, false);
+  if (!r->is_frozen)
+    ((upb_refcounted*)r)->individual_count--;
+  unref(r);
+}
+
+void upb_refcounted_ref2(const upb_refcounted *r, upb_refcounted *from) {
+  assert(!from->is_frozen);  /* Non-const pointer implies this. */
+  track(r, from, true);
+  if (r->is_frozen) {
+    refgroup(r->group);
+  } else {
+    merge((upb_refcounted*)r, from);
+  }
+}
+
+void upb_refcounted_unref2(const upb_refcounted *r, upb_refcounted *from) {
+  assert(!from->is_frozen);  /* Non-const pointer implies this. */
+  untrack(r, from, true);
+  if (r->is_frozen) {
+    unref(r);
+  } else {
+    assert(merged(r, from));
+  }
+}
+
+void upb_refcounted_donateref(
+    const upb_refcounted *r, const void *from, const void *to) {
+  assert(from != to);
+  if (to != NULL)
+    upb_refcounted_ref(r, to);
+  if (from != NULL)
+    upb_refcounted_unref(r, from);
+}
+
+void upb_refcounted_checkref(const upb_refcounted *r, const void *owner) {
+  checkref(r, owner, false);
+}
+
+bool upb_refcounted_freeze(upb_refcounted *const*roots, int n, upb_status *s,
+                           int maxdepth) {
+  int i;
+  for (i = 0; i < n; i++) {
+    assert(!roots[i]->is_frozen);
+  }
+  return freeze(roots, n, s, maxdepth);
+}
+
+
+#include <stdlib.h>
+
+/* Fallback implementation if the shim is not specialized by the JIT. */
+#define SHIM_WRITER(type, ctype)                                              \
+  bool upb_shim_set ## type (void *c, const void *hd, ctype val) {            \
+    uint8_t *m = c;                                                           \
+    const upb_shim_data *d = hd;                                              \
+    if (d->hasbit > 0)                                                        \
+      *(uint8_t*)&m[d->hasbit / 8] |= 1 << (d->hasbit % 8);                   \
+    *(ctype*)&m[d->offset] = val;                                             \
+    return true;                                                              \
+  }                                                                           \
+
+SHIM_WRITER(double, double)
+SHIM_WRITER(float,  float)
+SHIM_WRITER(int32,  int32_t)
+SHIM_WRITER(int64,  int64_t)
+SHIM_WRITER(uint32, uint32_t)
+SHIM_WRITER(uint64, uint64_t)
+SHIM_WRITER(bool,   bool)
+#undef SHIM_WRITER
+
+bool upb_shim_set(upb_handlers *h, const upb_fielddef *f, size_t offset,
+                  int32_t hasbit) {
+  upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER;
+  bool ok;
+
+  upb_shim_data *d = malloc(sizeof(*d));
+  if (!d) return false;
+  d->offset = offset;
+  d->hasbit = hasbit;
+
+  upb_handlerattr_sethandlerdata(&attr, d);
+  upb_handlerattr_setalwaysok(&attr, true);
+  upb_handlers_addcleanup(h, d, free);
+
+#define TYPE(u, l) \
+  case UPB_TYPE_##u: \
+    ok = upb_handlers_set##l(h, f, upb_shim_set##l, &attr); break;
+
+  ok = false;
+
+  switch (upb_fielddef_type(f)) {
+    TYPE(INT64,  int64);
+    TYPE(INT32,  int32);
+    TYPE(ENUM,   int32);
+    TYPE(UINT64, uint64);
+    TYPE(UINT32, uint32);
+    TYPE(DOUBLE, double);
+    TYPE(FLOAT,  float);
+    TYPE(BOOL,   bool);
+    default: assert(false); break;
+  }
+#undef TYPE
+
+  upb_handlerattr_uninit(&attr);
+  return ok;
+}
+
+const upb_shim_data *upb_shim_getdata(const upb_handlers *h, upb_selector_t s,
+                                      upb_fieldtype_t *type) {
+  upb_func *f = upb_handlers_gethandler(h, s);
+
+  if ((upb_int64_handlerfunc*)f == upb_shim_setint64) {
+    *type = UPB_TYPE_INT64;
+  } else if ((upb_int32_handlerfunc*)f == upb_shim_setint32) {
+    *type = UPB_TYPE_INT32;
+  } else if ((upb_uint64_handlerfunc*)f == upb_shim_setuint64) {
+    *type = UPB_TYPE_UINT64;
+  } else if ((upb_uint32_handlerfunc*)f == upb_shim_setuint32) {
+    *type = UPB_TYPE_UINT32;
+  } else if ((upb_double_handlerfunc*)f == upb_shim_setdouble) {
+    *type = UPB_TYPE_DOUBLE;
+  } else if ((upb_float_handlerfunc*)f == upb_shim_setfloat) {
+    *type = UPB_TYPE_FLOAT;
+  } else if ((upb_bool_handlerfunc*)f == upb_shim_setbool) {
+    *type = UPB_TYPE_BOOL;
+  } else {
+    return NULL;
+  }
+
+  return (const upb_shim_data*)upb_handlers_gethandlerdata(h, s);
+}
+
+
+#include <stdlib.h>
+#include <string.h>
+
+static void upb_symtab_free(upb_refcounted *r) {
+  upb_symtab *s = (upb_symtab*)r;
+  upb_strtable_iter i;
+  upb_strtable_begin(&i, &s->symtab);
+  for (; !upb_strtable_done(&i); upb_strtable_next(&i)) {
+    const upb_def *def = upb_value_getptr(upb_strtable_iter_value(&i));
+    upb_def_unref(def, s);
+  }
+  upb_strtable_uninit(&s->symtab);
+  free(s);
+}
+
+
+upb_symtab *upb_symtab_new(const void *owner) {
+  static const struct upb_refcounted_vtbl vtbl = {NULL, &upb_symtab_free};
+  upb_symtab *s = malloc(sizeof(*s));
+  upb_refcounted_init(upb_symtab_upcast_mutable(s), &vtbl, owner);
+  upb_strtable_init(&s->symtab, UPB_CTYPE_PTR);
+  return s;
+}
+
+void upb_symtab_freeze(upb_symtab *s) {
+  upb_refcounted *r;
+  bool ok;
+
+  assert(!upb_symtab_isfrozen(s));
+  r = upb_symtab_upcast_mutable(s);
+  /* The symtab does not take ref2's (see refcounted.h) on the defs, because
+   * defs cannot refer back to the table and therefore cannot create cycles.  So
+   * 0 will suffice for maxdepth here. */
+  ok = upb_refcounted_freeze(&r, 1, NULL, 0);
+  UPB_ASSERT_VAR(ok, ok);
+}
+
+const upb_def *upb_symtab_lookup(const upb_symtab *s, const char *sym) {
+  upb_value v;
+  upb_def *ret = upb_strtable_lookup(&s->symtab, sym, &v) ?
+      upb_value_getptr(v) : NULL;
+  return ret;
+}
+
+const upb_msgdef *upb_symtab_lookupmsg(const upb_symtab *s, const char *sym) {
+  upb_value v;
+  upb_def *def = upb_strtable_lookup(&s->symtab, sym, &v) ?
+      upb_value_getptr(v) : NULL;
+  return def ? upb_dyncast_msgdef(def) : NULL;
+}
+
+const upb_enumdef *upb_symtab_lookupenum(const upb_symtab *s, const char *sym) {
+  upb_value v;
+  upb_def *def = upb_strtable_lookup(&s->symtab, sym, &v) ?
+      upb_value_getptr(v) : NULL;
+  return def ? upb_dyncast_enumdef(def) : NULL;
+}
+
+/* Given a symbol and the base symbol inside which it is defined, find the
+ * symbol's definition in t. */
+static upb_def *upb_resolvename(const upb_strtable *t,
+                                const char *base, const char *sym) {
+  if(strlen(sym) == 0) return NULL;
+  if(sym[0] == '.') {
+    /* Symbols starting with '.' are absolute, so we do a single lookup.
+     * Slice to omit the leading '.' */
+    upb_value v;
+    return upb_strtable_lookup(t, sym + 1, &v) ? upb_value_getptr(v) : NULL;
+  } else {
+    /* Remove components from base until we find an entry or run out.
+     * TODO: This branch is totally broken, but currently not used. */
+    (void)base;
+    assert(false);
+    return NULL;
+  }
+}
+
+const upb_def *upb_symtab_resolve(const upb_symtab *s, const char *base,
+                                  const char *sym) {
+  upb_def *ret = upb_resolvename(&s->symtab, base, sym);
+  return ret;
+}
+
+/* Starts a depth-first traversal at "def", recursing into any subdefs
+ * (ie. submessage types).  Adds duplicates of existing defs to addtab
+ * wherever necessary, so that the resulting symtab will be consistent once
+ * addtab is added.
+ *
+ * More specifically, if any def D is found in the DFS that:
+ *
+ *   1. can reach a def that is being replaced by something in addtab, AND
+ *
+ *   2. is not itself being replaced already (ie. this name doesn't already
+ *      exist in addtab)
+ *
+ * ...then a duplicate (new copy) of D will be added to addtab.
+ *
+ * Returns true if this happened for any def reachable from "def."
+ *
+ * It is slightly tricky to do this correctly in the presence of cycles.  If we
+ * detect that our DFS has hit a cycle, we might not yet know if any SCCs on
+ * our stack can reach a def in addtab or not.  Once we figure this out, that
+ * answer needs to apply to *all* defs in these SCCs, even if we visited them
+ * already.  So a straight up one-pass cycle-detecting DFS won't work.
+ *
+ * To work around this problem, we traverse each SCC (which we already
+ * computed, since these defs are frozen) as a single node.  We first compute
+ * whether the SCC as a whole can reach any def in addtab, then we dup (or not)
+ * the entire SCC.  This requires breaking the encapsulation of upb_refcounted,
+ * since that is where we get the data about what SCC we are in. */
+static bool upb_resolve_dfs(const upb_def *def, upb_strtable *addtab,
+                            const void *new_owner, upb_inttable *seen,
+                            upb_status *s) {
+  upb_value v;
+  bool need_dup;
+  const upb_def *base;
+  const void* memoize_key;
+
+  /* Memoize results of this function for efficiency (since we're traversing a
+   * DAG this is not needed to limit the depth of the search).
+   *
+   * We memoize by SCC instead of by individual def. */
+  memoize_key = def->base.group;
+
+  if (upb_inttable_lookupptr(seen, memoize_key, &v))
+    return upb_value_getbool(v);
+
+  /* Visit submessages for all messages in the SCC. */
+  need_dup = false;
+  base = def;
+  do {
+    upb_value v;
+    const upb_msgdef *m;
+
+    assert(upb_def_isfrozen(def));
+    if (def->type == UPB_DEF_FIELD) continue;
+    if (upb_strtable_lookup(addtab, upb_def_fullname(def), &v)) {
+      need_dup = true;
+    }
+
+    /* For messages, continue the recursion by visiting all subdefs, but only
+     * ones in different SCCs. */
+    m = upb_dyncast_msgdef(def);
+    if (m) {
+      upb_msg_field_iter i;
+      for(upb_msg_field_begin(&i, m);
+          !upb_msg_field_done(&i);
+          upb_msg_field_next(&i)) {
+        upb_fielddef *f = upb_msg_iter_field(&i);
+        const upb_def *subdef;
+
+        if (!upb_fielddef_hassubdef(f)) continue;
+        subdef = upb_fielddef_subdef(f);
+
+        /* Skip subdefs in this SCC. */
+        if (def->base.group == subdef->base.group) continue;
+
+        /* |= to avoid short-circuit; we need its side-effects. */
+        need_dup |= upb_resolve_dfs(subdef, addtab, new_owner, seen, s);
+        if (!upb_ok(s)) return false;
+      }
+    }
+  } while ((def = (upb_def*)def->base.next) != base);
+
+  if (need_dup) {
+    /* Dup all defs in this SCC that don't already have entries in addtab. */
+    def = base;
+    do {
+      const char *name;
+
+      if (def->type == UPB_DEF_FIELD) continue;
+      name = upb_def_fullname(def);
+      if (!upb_strtable_lookup(addtab, name, NULL)) {
+        upb_def *newdef = upb_def_dup(def, new_owner);
+        if (!newdef) goto oom;
+        newdef->came_from_user = false;
+        if (!upb_strtable_insert(addtab, name, upb_value_ptr(newdef)))
+          goto oom;
+      }
+    } while ((def = (upb_def*)def->base.next) != base);
+  }
+
+  upb_inttable_insertptr(seen, memoize_key, upb_value_bool(need_dup));
+  return need_dup;
+
+oom:
+  upb_status_seterrmsg(s, "out of memory");
+  return false;
+}
+
+/* TODO(haberman): we need a lot more testing of error conditions.
+ * The came_from_user stuff in particular is not tested. */
+bool upb_symtab_add(upb_symtab *s, upb_def *const*defs, int n, void *ref_donor,
+                    upb_status *status) {
+  int i;
+  upb_strtable_iter iter;
+  upb_def **add_defs = NULL;
+  upb_strtable addtab;
+  upb_inttable seen;
+
+  assert(!upb_symtab_isfrozen(s));
+  if (!upb_strtable_init(&addtab, UPB_CTYPE_PTR)) {
+    upb_status_seterrmsg(status, "out of memory");
+    return false;
+  }
+
+  /* Add new defs to our "add" set. */
+  for (i = 0; i < n; i++) {
+    upb_def *def = defs[i];
+    const char *fullname;
+    upb_fielddef *f;
+
+    if (upb_def_isfrozen(def)) {
+      upb_status_seterrmsg(status, "added defs must be mutable");
+      goto err;
+    }
+    assert(!upb_def_isfrozen(def));
+    fullname = upb_def_fullname(def);
+    if (!fullname) {
+      upb_status_seterrmsg(
+          status, "Anonymous defs cannot be added to a symtab");
+      goto err;
+    }
+
+    f = upb_dyncast_fielddef_mutable(def);
+
+    if (f) {
+      if (!upb_fielddef_containingtypename(f)) {
+        upb_status_seterrmsg(status,
+                             "Standalone fielddefs must have a containing type "
+                             "(extendee) name set");
+        goto err;
+      }
+    } else {
+      if (upb_strtable_lookup(&addtab, fullname, NULL)) {
+        upb_status_seterrf(status, "Conflicting defs named '%s'", fullname);
+        goto err;
+      }
+      /* We need this to back out properly, because if there is a failure we
+       * need to donate the ref back to the caller. */
+      def->came_from_user = true;
+      upb_def_donateref(def, ref_donor, s);
+      if (!upb_strtable_insert(&addtab, fullname, upb_value_ptr(def)))
+        goto oom_err;
+    }
+  }
+
+  /* Add standalone fielddefs (ie. extensions) to the appropriate messages.
+   * If the appropriate message only exists in the existing symtab, duplicate
+   * it so we have a mutable copy we can add the fields to. */
+  for (i = 0; i < n; i++) {
+    upb_def *def = defs[i];
+    upb_fielddef *f = upb_dyncast_fielddef_mutable(def);
+    const char *msgname;
+    upb_value v;
+    upb_msgdef *m;
+
+    if (!f) continue;
+    msgname = upb_fielddef_containingtypename(f);
+    /* We validated this earlier in this function. */
+    assert(msgname);
+
+    /* If the extendee name is absolutely qualified, move past the initial ".".
+     * TODO(haberman): it is not obvious what it would mean if this was not
+     * absolutely qualified. */
+    if (msgname[0] == '.') {
+      msgname++;
+    }
+
+    if (upb_strtable_lookup(&addtab, msgname, &v)) {
+      /* Extendee is in the set of defs the user asked us to add. */
+      m = upb_value_getptr(v);
+    } else {
+      /* Need to find and dup the extendee from the existing symtab. */
+      const upb_msgdef *frozen_m = upb_symtab_lookupmsg(s, msgname);
+      if (!frozen_m) {
+        upb_status_seterrf(status,
+                           "Tried to extend message %s that does not exist "
+                           "in this SymbolTable.",
+                           msgname);
+        goto err;
+      }
+      m = upb_msgdef_dup(frozen_m, s);
+      if (!m) goto oom_err;
+      if (!upb_strtable_insert(&addtab, msgname, upb_value_ptr(m))) {
+        upb_msgdef_unref(m, s);
+        goto oom_err;
+      }
+    }
+
+    if (!upb_msgdef_addfield(m, f, ref_donor, status)) {
+      goto err;
+    }
+  }
+
+  /* Add dups of any existing def that can reach a def with the same name as
+   * anything in our "add" set. */
+  if (!upb_inttable_init(&seen, UPB_CTYPE_BOOL)) goto oom_err;
+  upb_strtable_begin(&iter, &s->symtab);
+  for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
+    upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter));
+    upb_resolve_dfs(def, &addtab, s, &seen, status);
+    if (!upb_ok(status)) goto err;
+  }
+  upb_inttable_uninit(&seen);
+
+  /* Now using the table, resolve symbolic references for subdefs. */
+  upb_strtable_begin(&iter, &addtab);
+  for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
+    const char *base;
+    upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter));
+    upb_msgdef *m = upb_dyncast_msgdef_mutable(def);
+    upb_msg_field_iter j;
+
+    if (!m) continue;
+    /* Type names are resolved relative to the message in which they appear. */
+    base = upb_msgdef_fullname(m);
+
+    for(upb_msg_field_begin(&j, m);
+        !upb_msg_field_done(&j);
+        upb_msg_field_next(&j)) {
+      upb_fielddef *f = upb_msg_iter_field(&j);
+      const char *name = upb_fielddef_subdefname(f);
+      if (name && !upb_fielddef_subdef(f)) {
+        /* Try the lookup in the current set of to-be-added defs first. If not
+         * there, try existing defs. */
+        upb_def *subdef = upb_resolvename(&addtab, base, name);
+        if (subdef == NULL) {
+          subdef = upb_resolvename(&s->symtab, base, name);
+        }
+        if (subdef == NULL) {
+          upb_status_seterrf(
+              status, "couldn't resolve name '%s' in message '%s'", name, base);
+          goto err;
+        } else if (!upb_fielddef_setsubdef(f, subdef, status)) {
+          goto err;
+        }
+      }
+    }
+  }
+
+  /* We need an array of the defs in addtab, for passing to upb_def_freeze. */
+  add_defs = malloc(sizeof(void*) * upb_strtable_count(&addtab));
+  if (add_defs == NULL) goto oom_err;
+  upb_strtable_begin(&iter, &addtab);
+  for (n = 0; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
+    add_defs[n++] = upb_value_getptr(upb_strtable_iter_value(&iter));
+  }
+
+  if (!upb_def_freeze(add_defs, n, status)) goto err;
+
+  /* This must be delayed until all errors have been detected, since error
+   * recovery code uses this table to cleanup defs. */
+  upb_strtable_uninit(&addtab);
+
+  /* TODO(haberman) we don't properly handle errors after this point (like
+   * OOM in upb_strtable_insert() below). */
+  for (i = 0; i < n; i++) {
+    upb_def *def = add_defs[i];
+    const char *name = upb_def_fullname(def);
+    upb_value v;
+    bool success;
+
+    if (upb_strtable_remove(&s->symtab, name, &v)) {
+      const upb_def *def = upb_value_getptr(v);
+      upb_def_unref(def, s);
+    }
+    success = upb_strtable_insert(&s->symtab, name, upb_value_ptr(def));
+    UPB_ASSERT_VAR(success, success == true);
+  }
+  free(add_defs);
+  return true;
+
+oom_err:
+  upb_status_seterrmsg(status, "out of memory");
+err: {
+    /* For defs the user passed in, we need to donate the refs back.  For defs
+     * we dup'd, we need to just unref them. */
+    upb_strtable_begin(&iter, &addtab);
+    for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
+      upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter));
+      bool came_from_user = def->came_from_user;
+      def->came_from_user = false;
+      if (came_from_user) {
+        upb_def_donateref(def, s, ref_donor);
+      } else {
+        upb_def_unref(def, s);
+      }
+    }
+  }
+  upb_strtable_uninit(&addtab);
+  free(add_defs);
+  assert(!upb_ok(status));
+  return false;
+}
+
+/* Iteration. */
+
+static void advance_to_matching(upb_symtab_iter *iter) {
+  if (iter->type == UPB_DEF_ANY)
+    return;
+
+  while (!upb_strtable_done(&iter->iter) &&
+         iter->type != upb_symtab_iter_def(iter)->type) {
+    upb_strtable_next(&iter->iter);
+  }
+}
+
+void upb_symtab_begin(upb_symtab_iter *iter, const upb_symtab *s,
+                      upb_deftype_t type) {
+  upb_strtable_begin(&iter->iter, &s->symtab);
+  iter->type = type;
+  advance_to_matching(iter);
+}
+
+void upb_symtab_next(upb_symtab_iter *iter) {
+  upb_strtable_next(&iter->iter);
+  advance_to_matching(iter);
+}
+
+bool upb_symtab_done(const upb_symtab_iter *iter) {
+  return upb_strtable_done(&iter->iter);
+}
+
+const upb_def *upb_symtab_iter_def(const upb_symtab_iter *iter) {
+  return upb_value_getptr(upb_strtable_iter_value(&iter->iter));
+}
+/*
+** upb_table Implementation
+**
+** Implementation is heavily inspired by Lua's ltable.c.
+*/
+
+
+#include <stdlib.h>
+#include <string.h>
+
+#define UPB_MAXARRSIZE 16  /* 64k. */
+
+/* From Chromium. */
+#define ARRAY_SIZE(x) \
+    ((sizeof(x)/sizeof(0[x])) / ((size_t)(!(sizeof(x) % sizeof(0[x])))))
+
+static const double MAX_LOAD = 0.85;
+
+/* The minimum utilization of the array part of a mixed hash/array table.  This
+ * is a speed/memory-usage tradeoff (though it's not straightforward because of
+ * cache effects).  The lower this is, the more memory we'll use. */
+static const double MIN_DENSITY = 0.1;
+
+bool is_pow2(uint64_t v) { return v == 0 || (v & (v - 1)) == 0; }
+
+int log2ceil(uint64_t v) {
+  int ret = 0;
+  bool pow2 = is_pow2(v);
+  while (v >>= 1) ret++;
+  ret = pow2 ? ret : ret + 1;  /* Ceiling. */
+  return UPB_MIN(UPB_MAXARRSIZE, ret);
+}
+
+char *upb_strdup(const char *s) {
+  return upb_strdup2(s, strlen(s));
+}
+
+char *upb_strdup2(const char *s, size_t len) {
+  size_t n;
+  char *p;
+
+  /* Prevent overflow errors. */
+  if (len == SIZE_MAX) return NULL;
+  /* Always null-terminate, even if binary data; but don't rely on the input to
+   * have a null-terminating byte since it may be a raw binary buffer. */
+  n = len + 1;
+  p = malloc(n);
+  if (p) {
+    memcpy(p, s, len);
+    p[len] = 0;
+  }
+  return p;
+}
+
+/* A type to represent the lookup key of either a strtable or an inttable. */
+typedef union {
+  uintptr_t num;
+  struct {
+    const char *str;
+    size_t len;
+  } str;
+} lookupkey_t;
+
+static lookupkey_t strkey2(const char *str, size_t len) {
+  lookupkey_t k;
+  k.str.str = str;
+  k.str.len = len;
+  return k;
+}
+
+static lookupkey_t intkey(uintptr_t key) {
+  lookupkey_t k;
+  k.num = key;
+  return k;
+}
+
+typedef uint32_t hashfunc_t(upb_tabkey key);
+typedef bool eqlfunc_t(upb_tabkey k1, lookupkey_t k2);
+
+/* Base table (shared code) ***************************************************/
+
+/* For when we need to cast away const. */
+static upb_tabent *mutable_entries(upb_table *t) {
+  return (upb_tabent*)t->entries;
+}
+
+static bool isfull(upb_table *t) {
+  return (double)(t->count + 1) / upb_table_size(t) > MAX_LOAD;
+}
+
+static bool init(upb_table *t, upb_ctype_t ctype, uint8_t size_lg2) {
+  size_t bytes;
+
+  t->count = 0;
+  t->ctype = ctype;
+  t->size_lg2 = size_lg2;
+  t->mask = upb_table_size(t) ? upb_table_size(t) - 1 : 0;
+  bytes = upb_table_size(t) * sizeof(upb_tabent);
+  if (bytes > 0) {
+    t->entries = malloc(bytes);
+    if (!t->entries) return false;
+    memset(mutable_entries(t), 0, bytes);
+  } else {
+    t->entries = NULL;
+  }
+  return true;
+}
+
+static void uninit(upb_table *t) { free(mutable_entries(t)); }
+
+static upb_tabent *emptyent(upb_table *t) {
+  upb_tabent *e = mutable_entries(t) + upb_table_size(t);
+  while (1) { if (upb_tabent_isempty(--e)) return e; assert(e > t->entries); }
+}
+
+static upb_tabent *getentry_mutable(upb_table *t, uint32_t hash) {
+  return (upb_tabent*)upb_getentry(t, hash);
+}
+
+static const upb_tabent *findentry(const upb_table *t, lookupkey_t key,
+                                   uint32_t hash, eqlfunc_t *eql) {
+  const upb_tabent *e;
+
+  if (t->size_lg2 == 0) return NULL;
+  e = upb_getentry(t, hash);
+  if (upb_tabent_isempty(e)) return NULL;
+  while (1) {
+    if (eql(e->key, key)) return e;
+    if ((e = e->next) == NULL) return NULL;
+  }
+}
+
+static upb_tabent *findentry_mutable(upb_table *t, lookupkey_t key,
+                                     uint32_t hash, eqlfunc_t *eql) {
+  return (upb_tabent*)findentry(t, key, hash, eql);
+}
+
+static bool lookup(const upb_table *t, lookupkey_t key, upb_value *v,
+                   uint32_t hash, eqlfunc_t *eql) {
+  const upb_tabent *e = findentry(t, key, hash, eql);
+  if (e) {
+    if (v) {
+      _upb_value_setval(v, e->val.val, t->ctype);
+    }
+    return true;
+  } else {
+    return false;
+  }
+}
+
+/* The given key must not already exist in the table. */
+static void insert(upb_table *t, lookupkey_t key, upb_tabkey tabkey,
+                   upb_value val, uint32_t hash,
+                   hashfunc_t *hashfunc, eqlfunc_t *eql) {
+  upb_tabent *mainpos_e;
+  upb_tabent *our_e;
+
+  UPB_UNUSED(eql);
+  UPB_UNUSED(key);
+  assert(findentry(t, key, hash, eql) == NULL);
+  assert(val.ctype == t->ctype);
+
+  t->count++;
+  mainpos_e = getentry_mutable(t, hash);
+  our_e = mainpos_e;
+
+  if (upb_tabent_isempty(mainpos_e)) {
+    /* Our main position is empty; use it. */
+    our_e->next = NULL;
+  } else {
+    /* Collision. */
+    upb_tabent *new_e = emptyent(t);
+    /* Head of collider's chain. */
+    upb_tabent *chain = getentry_mutable(t, hashfunc(mainpos_e->key));
+    if (chain == mainpos_e) {
+      /* Existing ent is in its main posisiton (it has the same hash as us, and
+       * is the head of our chain).  Insert to new ent and append to this chain. */
+      new_e->next = mainpos_e->next;
+      mainpos_e->next = new_e;
+      our_e = new_e;
+    } else {
+      /* Existing ent is not in its main position (it is a node in some other
+       * chain).  This implies that no existing ent in the table has our hash.
+       * Evict it (updating its chain) and use its ent for head of our chain. */
+      *new_e = *mainpos_e;  /* copies next. */
+      while (chain->next != mainpos_e) {
+        chain = (upb_tabent*)chain->next;
+        assert(chain);
+      }
+      chain->next = new_e;
+      our_e = mainpos_e;
+      our_e->next = NULL;
+    }
+  }
+  our_e->key = tabkey;
+  our_e->val.val = val.val;
+  assert(findentry(t, key, hash, eql) == our_e);
+}
+
+static bool rm(upb_table *t, lookupkey_t key, upb_value *val,
+               upb_tabkey *removed, uint32_t hash, eqlfunc_t *eql) {
+  upb_tabent *chain = getentry_mutable(t, hash);
+  if (upb_tabent_isempty(chain)) return false;
+  if (eql(chain->key, key)) {
+    /* Element to remove is at the head of its chain. */
+    t->count--;
+    if (val) {
+      _upb_value_setval(val, chain->val.val, t->ctype);
+    }
+    if (chain->next) {
+      upb_tabent *move = (upb_tabent*)chain->next;
+      *chain = *move;
+      if (removed) *removed = move->key;
+      move->key = 0;  /* Make the slot empty. */
+    } else {
+      if (removed) *removed = chain->key;
+      chain->key = 0;  /* Make the slot empty. */
+    }
+    return true;
+  } else {
+    /* Element to remove is either in a non-head position or not in the
+     * table. */
+    while (chain->next && !eql(chain->next->key, key))
+      chain = (upb_tabent*)chain->next;
+    if (chain->next) {
+      /* Found element to remove. */
+      upb_tabent *rm;
+
+      if (val) {
+        _upb_value_setval(val, chain->next->val.val, t->ctype);
+      }
+      rm = (upb_tabent*)chain->next;
+      if (removed) *removed = rm->key;
+      rm->key = 0;
+      chain->next = rm->next;
+      t->count--;
+      return true;
+    } else {
+      return false;
+    }
+  }
+}
+
+static size_t next(const upb_table *t, size_t i) {
+  do {
+    if (++i >= upb_table_size(t))
+      return SIZE_MAX;
+  } while(upb_tabent_isempty(&t->entries[i]));
+
+  return i;
+}
+
+static size_t begin(const upb_table *t) {
+  return next(t, -1);
+}
+
+
+/* upb_strtable ***************************************************************/
+
+/* A simple "subclass" of upb_table that only adds a hash function for strings. */
+
+static upb_tabkey strcopy(lookupkey_t k2) {
+  char *str = malloc(k2.str.len + sizeof(uint32_t) + 1);
+  if (str == NULL) return 0;
+  memcpy(str, &k2.str.len, sizeof(uint32_t));
+  memcpy(str + sizeof(uint32_t), k2.str.str, k2.str.len + 1);
+  return (uintptr_t)str;
+}
+
+static uint32_t strhash(upb_tabkey key) {
+  uint32_t len;
+  char *str = upb_tabstr(key, &len);
+  return MurmurHash2(str, len, 0);
+}
+
+static bool streql(upb_tabkey k1, lookupkey_t k2) {
+  uint32_t len;
+  char *str = upb_tabstr(k1, &len);
+  return len == k2.str.len && memcmp(str, k2.str.str, len) == 0;
+}
+
+bool upb_strtable_init(upb_strtable *t, upb_ctype_t ctype) {
+  return init(&t->t, ctype, 2);
+}
+
+void upb_strtable_uninit(upb_strtable *t) {
+  size_t i;
+  for (i = 0; i < upb_table_size(&t->t); i++)
+    free((void*)t->t.entries[i].key);
+  uninit(&t->t);
+}
+
+bool upb_strtable_resize(upb_strtable *t, size_t size_lg2) {
+  upb_strtable new_table;
+  upb_strtable_iter i;
+
+  if (!init(&new_table.t, t->t.ctype, size_lg2))
+    return false;
+  upb_strtable_begin(&i, t);
+  for ( ; !upb_strtable_done(&i); upb_strtable_next(&i)) {
+    upb_strtable_insert2(
+        &new_table,
+        upb_strtable_iter_key(&i),
+        upb_strtable_iter_keylength(&i),
+        upb_strtable_iter_value(&i));
+  }
+  upb_strtable_uninit(t);
+  *t = new_table;
+  return true;
+}
+
+bool upb_strtable_insert2(upb_strtable *t, const char *k, size_t len,
+                          upb_value v) {
+  lookupkey_t key;
+  upb_tabkey tabkey;
+  uint32_t hash;
+
+  if (isfull(&t->t)) {
+    /* Need to resize.  New table of double the size, add old elements to it. */
+    if (!upb_strtable_resize(t, t->t.size_lg2 + 1)) {
+      return false;
+    }
+  }
+
+  key = strkey2(k, len);
+  tabkey = strcopy(key);
+  if (tabkey == 0) return false;
+
+  hash = MurmurHash2(key.str.str, key.str.len, 0);
+  insert(&t->t, key, tabkey, v, hash, &strhash, &streql);
+  return true;
+}
+
+bool upb_strtable_lookup2(const upb_strtable *t, const char *key, size_t len,
+                          upb_value *v) {
+  uint32_t hash = MurmurHash2(key, len, 0);
+  return lookup(&t->t, strkey2(key, len), v, hash, &streql);
+}
+
+bool upb_strtable_remove2(upb_strtable *t, const char *key, size_t len,
+                         upb_value *val) {
+  uint32_t hash = MurmurHash2(key, strlen(key), 0);
+  upb_tabkey tabkey;
+  if (rm(&t->t, strkey2(key, len), val, &tabkey, hash, &streql)) {
+    free((void*)tabkey);
+    return true;
+  } else {
+    return false;
+  }
+}
+
+/* Iteration */
+
+static const upb_tabent *str_tabent(const upb_strtable_iter *i) {
+  return &i->t->t.entries[i->index];
+}
+
+void upb_strtable_begin(upb_strtable_iter *i, const upb_strtable *t) {
+  i->t = t;
+  i->index = begin(&t->t);
+}
+
+void upb_strtable_next(upb_strtable_iter *i) {
+  i->index = next(&i->t->t, i->index);
+}
+
+bool upb_strtable_done(const upb_strtable_iter *i) {
+  return i->index >= upb_table_size(&i->t->t) ||
+         upb_tabent_isempty(str_tabent(i));
+}
+
+const char *upb_strtable_iter_key(upb_strtable_iter *i) {
+  assert(!upb_strtable_done(i));
+  return upb_tabstr(str_tabent(i)->key, NULL);
+}
+
+size_t upb_strtable_iter_keylength(upb_strtable_iter *i) {
+  uint32_t len;
+  assert(!upb_strtable_done(i));
+  upb_tabstr(str_tabent(i)->key, &len);
+  return len;
+}
+
+upb_value upb_strtable_iter_value(const upb_strtable_iter *i) {
+  assert(!upb_strtable_done(i));
+  return _upb_value_val(str_tabent(i)->val.val, i->t->t.ctype);
+}
+
+void upb_strtable_iter_setdone(upb_strtable_iter *i) {
+  i->index = SIZE_MAX;
+}
+
+bool upb_strtable_iter_isequal(const upb_strtable_iter *i1,
+                               const upb_strtable_iter *i2) {
+  if (upb_strtable_done(i1) && upb_strtable_done(i2))
+    return true;
+  return i1->t == i2->t && i1->index == i2->index;
+}
+
+
+/* upb_inttable ***************************************************************/
+
+/* For inttables we use a hybrid structure where small keys are kept in an
+ * array and large keys are put in the hash table. */
+
+static uint32_t inthash(upb_tabkey key) { return upb_inthash(key); }
+
+static bool inteql(upb_tabkey k1, lookupkey_t k2) {
+  return k1 == k2.num;
+}
+
+static upb_tabval *mutable_array(upb_inttable *t) {
+  return (upb_tabval*)t->array;
+}
+
+static upb_tabval *inttable_val(upb_inttable *t, uintptr_t key) {
+  if (key < t->array_size) {
+    return upb_arrhas(t->array[key]) ? &(mutable_array(t)[key]) : NULL;
+  } else {
+    upb_tabent *e =
+        findentry_mutable(&t->t, intkey(key), upb_inthash(key), &inteql);
+    return e ? &e->val : NULL;
+  }
+}
+
+static const upb_tabval *inttable_val_const(const upb_inttable *t,
+                                            uintptr_t key) {
+  return inttable_val((upb_inttable*)t, key);
+}
+
+size_t upb_inttable_count(const upb_inttable *t) {
+  return t->t.count + t->array_count;
+}
+
+static void check(upb_inttable *t) {
+  UPB_UNUSED(t);
+#if defined(UPB_DEBUG_TABLE) && !defined(NDEBUG)
+  {
+    /* This check is very expensive (makes inserts/deletes O(N)). */
+    size_t count = 0;
+    upb_inttable_iter i;
+    upb_inttable_begin(&i, t);
+    for(; !upb_inttable_done(&i); upb_inttable_next(&i), count++) {
+      assert(upb_inttable_lookup(t, upb_inttable_iter_key(&i), NULL));
+    }
+    assert(count == upb_inttable_count(t));
+  }
+#endif
+}
+
+bool upb_inttable_sizedinit(upb_inttable *t, upb_ctype_t ctype,
+                            size_t asize, int hsize_lg2) {
+  size_t array_bytes;
+
+  if (!init(&t->t, ctype, hsize_lg2)) return false;
+  /* Always make the array part at least 1 long, so that we know key 0
+   * won't be in the hash part, which simplifies things. */
+  t->array_size = UPB_MAX(1, asize);
+  t->array_count = 0;
+  array_bytes = t->array_size * sizeof(upb_value);
+  t->array = malloc(array_bytes);
+  if (!t->array) {
+    uninit(&t->t);
+    return false;
+  }
+  memset(mutable_array(t), 0xff, array_bytes);
+  check(t);
+  return true;
+}
+
+bool upb_inttable_init(upb_inttable *t, upb_ctype_t ctype) {
+  return upb_inttable_sizedinit(t, ctype, 0, 4);
+}
+
+void upb_inttable_uninit(upb_inttable *t) {
+  uninit(&t->t);
+  free(mutable_array(t));
+}
+
+bool upb_inttable_insert(upb_inttable *t, uintptr_t key, upb_value val) {
+  /* XXX: Table can't store value (uint64_t)-1.  Need to somehow statically
+   * guarantee that this is not necessary, or fix the limitation. */
+  upb_tabval tabval;
+  tabval.val = val.val;
+  UPB_UNUSED(tabval);
+  assert(upb_arrhas(tabval));
+
+  if (key < t->array_size) {
+    assert(!upb_arrhas(t->array[key]));
+    t->array_count++;
+    mutable_array(t)[key].val = val.val;
+  } else {
+    if (isfull(&t->t)) {
+      /* Need to resize the hash part, but we re-use the array part. */
+      size_t i;
+      upb_table new_table;
+      if (!init(&new_table, t->t.ctype, t->t.size_lg2 + 1))
+        return false;
+      for (i = begin(&t->t); i < upb_table_size(&t->t); i = next(&t->t, i)) {
+        const upb_tabent *e = &t->t.entries[i];
+        uint32_t hash;
+        upb_value v;
+
+        _upb_value_setval(&v, e->val.val, t->t.ctype);
+        hash = upb_inthash(e->key);
+        insert(&new_table, intkey(e->key), e->key, v, hash, &inthash, &inteql);
+      }
+
+      assert(t->t.count == new_table.count);
+
+      uninit(&t->t);
+      t->t = new_table;
+    }
+    insert(&t->t, intkey(key), key, val, upb_inthash(key), &inthash, &inteql);
+  }
+  check(t);
+  return true;
+}
+
+bool upb_inttable_lookup(const upb_inttable *t, uintptr_t key, upb_value *v) {
+  const upb_tabval *table_v = inttable_val_const(t, key);
+  if (!table_v) return false;
+  if (v) _upb_value_setval(v, table_v->val, t->t.ctype);
+  return true;
+}
+
+bool upb_inttable_replace(upb_inttable *t, uintptr_t key, upb_value val) {
+  upb_tabval *table_v = inttable_val(t, key);
+  if (!table_v) return false;
+  table_v->val = val.val;
+  return true;
+}
+
+bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val) {
+  bool success;
+  if (key < t->array_size) {
+    if (upb_arrhas(t->array[key])) {
+      upb_tabval empty = UPB_TABVALUE_EMPTY_INIT;
+      t->array_count--;
+      if (val) {
+        _upb_value_setval(val, t->array[key].val, t->t.ctype);
+      }
+      mutable_array(t)[key] = empty;
+      success = true;
+    } else {
+      success = false;
+    }
+  } else {
+    upb_tabkey removed;
+    uint32_t hash = upb_inthash(key);
+    success = rm(&t->t, intkey(key), val, &removed, hash, &inteql);
+  }
+  check(t);
+  return success;
+}
+
+bool upb_inttable_push(upb_inttable *t, upb_value val) {
+  return upb_inttable_insert(t, upb_inttable_count(t), val);
+}
+
+upb_value upb_inttable_pop(upb_inttable *t) {
+  upb_value val;
+  bool ok = upb_inttable_remove(t, upb_inttable_count(t) - 1, &val);
+  UPB_ASSERT_VAR(ok, ok);
+  return val;
+}
+
+bool upb_inttable_insertptr(upb_inttable *t, const void *key, upb_value val) {
+  return upb_inttable_insert(t, (uintptr_t)key, val);
+}
+
+bool upb_inttable_lookupptr(const upb_inttable *t, const void *key,
+                            upb_value *v) {
+  return upb_inttable_lookup(t, (uintptr_t)key, v);
+}
+
+bool upb_inttable_removeptr(upb_inttable *t, const void *key, upb_value *val) {
+  return upb_inttable_remove(t, (uintptr_t)key, val);
+}
+
+void upb_inttable_compact(upb_inttable *t) {
+  /* Create a power-of-two histogram of the table keys. */
+  int counts[UPB_MAXARRSIZE + 1] = {0};
+  uintptr_t max_key = 0;
+  upb_inttable_iter i;
+  size_t arr_size;
+  int arr_count;
+  upb_inttable new_t;
+
+  upb_inttable_begin(&i, t);
+  for (; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    uintptr_t key = upb_inttable_iter_key(&i);
+    if (key > max_key) {
+      max_key = key;
+    }
+    counts[log2ceil(key)]++;
+  }
+
+  arr_size = 1;
+  arr_count = upb_inttable_count(t);
+
+  if (upb_inttable_count(t) >= max_key * MIN_DENSITY) {
+    /* We can put 100% of the entries in the array part. */
+    arr_size = max_key + 1;
+  } else {
+    /* Find the largest power of two that satisfies the MIN_DENSITY
+     * definition. */
+    int size_lg2;
+    for (size_lg2 = ARRAY_SIZE(counts) - 1; size_lg2 > 1; size_lg2--) {
+      arr_size = 1 << size_lg2;
+      arr_count -= counts[size_lg2];
+      if (arr_count >= arr_size * MIN_DENSITY) {
+        break;
+      }
+    }
+  }
+
+  /* Array part must always be at least 1 entry large to catch lookups of key
+   * 0.  Key 0 must always be in the array part because "0" in the hash part
+   * denotes an empty entry. */
+  arr_size = UPB_MAX(arr_size, 1);
+
+  {
+    /* Insert all elements into new, perfectly-sized table. */
+    int hash_count = upb_inttable_count(t) - arr_count;
+    int hash_size = hash_count ? (hash_count / MAX_LOAD) + 1 : 0;
+    int hashsize_lg2 = log2ceil(hash_size);
+
+    assert(hash_count >= 0);
+    upb_inttable_sizedinit(&new_t, t->t.ctype, arr_size, hashsize_lg2);
+    upb_inttable_begin(&i, t);
+    for (; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+      uintptr_t k = upb_inttable_iter_key(&i);
+      upb_inttable_insert(&new_t, k, upb_inttable_iter_value(&i));
+    }
+    assert(new_t.array_size == arr_size);
+    assert(new_t.t.size_lg2 == hashsize_lg2);
+  }
+  upb_inttable_uninit(t);
+  *t = new_t;
+}
+
+/* Iteration. */
+
+static const upb_tabent *int_tabent(const upb_inttable_iter *i) {
+  assert(!i->array_part);
+  return &i->t->t.entries[i->index];
+}
+
+static upb_tabval int_arrent(const upb_inttable_iter *i) {
+  assert(i->array_part);
+  return i->t->array[i->index];
+}
+
+void upb_inttable_begin(upb_inttable_iter *i, const upb_inttable *t) {
+  i->t = t;
+  i->index = -1;
+  i->array_part = true;
+  upb_inttable_next(i);
+}
+
+void upb_inttable_next(upb_inttable_iter *iter) {
+  const upb_inttable *t = iter->t;
+  if (iter->array_part) {
+    while (++iter->index < t->array_size) {
+      if (upb_arrhas(int_arrent(iter))) {
+        return;
+      }
+    }
+    iter->array_part = false;
+    iter->index = begin(&t->t);
+  } else {
+    iter->index = next(&t->t, iter->index);
+  }
+}
+
+bool upb_inttable_done(const upb_inttable_iter *i) {
+  if (i->array_part) {
+    return i->index >= i->t->array_size ||
+           !upb_arrhas(int_arrent(i));
+  } else {
+    return i->index >= upb_table_size(&i->t->t) ||
+           upb_tabent_isempty(int_tabent(i));
+  }
+}
+
+uintptr_t upb_inttable_iter_key(const upb_inttable_iter *i) {
+  assert(!upb_inttable_done(i));
+  return i->array_part ? i->index : int_tabent(i)->key;
+}
+
+upb_value upb_inttable_iter_value(const upb_inttable_iter *i) {
+  assert(!upb_inttable_done(i));
+  return _upb_value_val(
+      i->array_part ? i->t->array[i->index].val : int_tabent(i)->val.val,
+      i->t->t.ctype);
+}
+
+void upb_inttable_iter_setdone(upb_inttable_iter *i) {
+  i->index = SIZE_MAX;
+  i->array_part = false;
+}
+
+bool upb_inttable_iter_isequal(const upb_inttable_iter *i1,
+                                          const upb_inttable_iter *i2) {
+  if (upb_inttable_done(i1) && upb_inttable_done(i2))
+    return true;
+  return i1->t == i2->t && i1->index == i2->index &&
+         i1->array_part == i2->array_part;
+}
+
+#ifdef UPB_UNALIGNED_READS_OK
+/* -----------------------------------------------------------------------------
+ * MurmurHash2, by Austin Appleby (released as public domain).
+ * Reformatted and C99-ified by Joshua Haberman.
+ * Note - This code makes a few assumptions about how your machine behaves -
+ *   1. We can read a 4-byte value from any address without crashing
+ *   2. sizeof(int) == 4 (in upb this limitation is removed by using uint32_t
+ * And it has a few limitations -
+ *   1. It will not work incrementally.
+ *   2. It will not produce the same results on little-endian and big-endian
+ *      machines. */
+uint32_t MurmurHash2(const void *key, size_t len, uint32_t seed) {
+  /* 'm' and 'r' are mixing constants generated offline.
+   * They're not really 'magic', they just happen to work well. */
+  const uint32_t m = 0x5bd1e995;
+  const int32_t r = 24;
+
+  /* Initialize the hash to a 'random' value */
+  uint32_t h = seed ^ len;
+
+  /* Mix 4 bytes at a time into the hash */
+  const uint8_t * data = (const uint8_t *)key;
+  while(len >= 4) {
+    uint32_t k = *(uint32_t *)data;
+
+    k *= m;
+    k ^= k >> r;
+    k *= m;
+
+    h *= m;
+    h ^= k;
+
+    data += 4;
+    len -= 4;
+  }
+
+  /* Handle the last few bytes of the input array */
+  switch(len) {
+    case 3: h ^= data[2] << 16;
+    case 2: h ^= data[1] << 8;
+    case 1: h ^= data[0]; h *= m;
+  };
+
+  /* Do a few final mixes of the hash to ensure the last few
+   * bytes are well-incorporated. */
+  h ^= h >> 13;
+  h *= m;
+  h ^= h >> 15;
+
+  return h;
+}
+
+#else /* !UPB_UNALIGNED_READS_OK */
+
+/* -----------------------------------------------------------------------------
+ * MurmurHashAligned2, by Austin Appleby
+ * Same algorithm as MurmurHash2, but only does aligned reads - should be safer
+ * on certain platforms.
+ * Performance will be lower than MurmurHash2 */
+
+#define MIX(h,k,m) { k *= m; k ^= k >> r; k *= m; h *= m; h ^= k; }
+
+uint32_t MurmurHash2(const void * key, size_t len, uint32_t seed) {
+  const uint32_t m = 0x5bd1e995;
+  const int32_t r = 24;
+  const uint8_t * data = (const uint8_t *)key;
+  uint32_t h = seed ^ len;
+  uint8_t align = (uintptr_t)data & 3;
+
+  if(align && (len >= 4)) {
+    /* Pre-load the temp registers */
+    uint32_t t = 0, d = 0;
+    int32_t sl;
+    int32_t sr;
+
+    switch(align) {
+      case 1: t |= data[2] << 16;
+      case 2: t |= data[1] << 8;
+      case 3: t |= data[0];
+    }
+
+    t <<= (8 * align);
+
+    data += 4-align;
+    len -= 4-align;
+
+    sl = 8 * (4-align);
+    sr = 8 * align;
+
+    /* Mix */
+
+    while(len >= 4) {
+      uint32_t k;
+
+      d = *(uint32_t *)data;
+      t = (t >> sr) | (d << sl);
+
+      k = t;
+
+      MIX(h,k,m);
+
+      t = d;
+
+      data += 4;
+      len -= 4;
+    }
+
+    /* Handle leftover data in temp registers */
+
+    d = 0;
+
+    if(len >= align) {
+      uint32_t k;
+
+      switch(align) {
+        case 3: d |= data[2] << 16;
+        case 2: d |= data[1] << 8;
+        case 1: d |= data[0];
+      }
+
+      k = (t >> sr) | (d << sl);
+      MIX(h,k,m);
+
+      data += align;
+      len -= align;
+
+      /* ----------
+       * Handle tail bytes */
+
+      switch(len) {
+        case 3: h ^= data[2] << 16;
+        case 2: h ^= data[1] << 8;
+        case 1: h ^= data[0]; h *= m;
+      };
+    } else {
+      switch(len) {
+        case 3: d |= data[2] << 16;
+        case 2: d |= data[1] << 8;
+        case 1: d |= data[0];
+        case 0: h ^= (t >> sr) | (d << sl); h *= m;
+      }
+    }
+
+    h ^= h >> 13;
+    h *= m;
+    h ^= h >> 15;
+
+    return h;
+  } else {
+    while(len >= 4) {
+      uint32_t k = *(uint32_t *)data;
+
+      MIX(h,k,m);
+
+      data += 4;
+      len -= 4;
+    }
+
+    /* ----------
+     * Handle tail bytes */
+
+    switch(len) {
+      case 3: h ^= data[2] << 16;
+      case 2: h ^= data[1] << 8;
+      case 1: h ^= data[0]; h *= m;
+    };
+
+    h ^= h >> 13;
+    h *= m;
+    h ^= h >> 15;
+
+    return h;
+  }
+}
+#undef MIX
+
+#endif /* UPB_UNALIGNED_READS_OK */
+
+#include <errno.h>
+#include <stdarg.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+bool upb_dumptostderr(void *closure, const upb_status* status) {
+  UPB_UNUSED(closure);
+  fprintf(stderr, "%s\n", upb_status_errmsg(status));
+  return false;
+}
+
+/* Guarantee null-termination and provide ellipsis truncation.
+ * It may be tempting to "optimize" this by initializing these final
+ * four bytes up-front and then being careful never to overwrite them,
+ * this is safer and simpler. */
+static void nullz(upb_status *status) {
+  const char *ellipsis = "...";
+  size_t len = strlen(ellipsis);
+  assert(sizeof(status->msg) > len);
+  memcpy(status->msg + sizeof(status->msg) - len, ellipsis, len);
+}
+
+void upb_status_clear(upb_status *status) {
+  if (!status) return;
+  status->ok_ = true;
+  status->code_ = 0;
+  status->msg[0] = '\0';
+}
+
+bool upb_ok(const upb_status *status) { return status->ok_; }
+
+upb_errorspace *upb_status_errspace(const upb_status *status) {
+  return status->error_space_;
+}
+
+int upb_status_errcode(const upb_status *status) { return status->code_; }
+
+const char *upb_status_errmsg(const upb_status *status) { return status->msg; }
+
+void upb_status_seterrmsg(upb_status *status, const char *msg) {
+  if (!status) return;
+  status->ok_ = false;
+  strncpy(status->msg, msg, sizeof(status->msg));
+  nullz(status);
+}
+
+void upb_status_seterrf(upb_status *status, const char *fmt, ...) {
+  va_list args;
+  va_start(args, fmt);
+  upb_status_vseterrf(status, fmt, args);
+  va_end(args);
+}
+
+void upb_status_vseterrf(upb_status *status, const char *fmt, va_list args) {
+  if (!status) return;
+  status->ok_ = false;
+  _upb_vsnprintf(status->msg, sizeof(status->msg), fmt, args);
+  nullz(status);
+}
+
+void upb_status_seterrcode(upb_status *status, upb_errorspace *space,
+                           int code) {
+  if (!status) return;
+  status->ok_ = false;
+  status->error_space_ = space;
+  status->code_ = code;
+  space->set_message(status, code);
+}
+
+void upb_status_copy(upb_status *to, const upb_status *from) {
+  if (!to) return;
+  *to = *from;
+}
+/* This file was generated by upbc (the upb compiler).
+ * Do not edit -- your changes will be discarded when the file is
+ * regenerated. */
+
+
+static const upb_msgdef msgs[20];
+static const upb_fielddef fields[81];
+static const upb_enumdef enums[4];
+static const upb_tabent strentries[236];
+static const upb_tabent intentries[14];
+static const upb_tabval arrays[232];
+
+#ifdef UPB_DEBUG_REFS
+static upb_inttable reftables[212];
+#endif
+
+static const upb_msgdef msgs[20] = {
+  UPB_MSGDEF_INIT("google.protobuf.DescriptorProto", 27, 6, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[0], 8, 7), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[0]),&reftables[0], &reftables[1]),
+  UPB_MSGDEF_INIT("google.protobuf.DescriptorProto.ExtensionRange", 4, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[8], 3, 2), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[16]),&reftables[2], &reftables[3]),
+  UPB_MSGDEF_INIT("google.protobuf.EnumDescriptorProto", 11, 2, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[11], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[20]),&reftables[4], &reftables[5]),
+  UPB_MSGDEF_INIT("google.protobuf.EnumOptions", 7, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[0], &arrays[15], 8, 1), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[24]),&reftables[6], &reftables[7]),
+  UPB_MSGDEF_INIT("google.protobuf.EnumValueDescriptorProto", 8, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[23], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[28]),&reftables[8], &reftables[9]),
+  UPB_MSGDEF_INIT("google.protobuf.EnumValueOptions", 6, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[2], &arrays[27], 4, 0), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[32]),&reftables[10], &reftables[11]),
+  UPB_MSGDEF_INIT("google.protobuf.FieldDescriptorProto", 19, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[31], 9, 8), UPB_STRTABLE_INIT(8, 15, UPB_CTYPE_PTR, 4, &strentries[36]),&reftables[12], &reftables[13]),
+  UPB_MSGDEF_INIT("google.protobuf.FieldOptions", 14, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[4], &arrays[40], 32, 6), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[52]),&reftables[14], &reftables[15]),
+  UPB_MSGDEF_INIT("google.protobuf.FileDescriptorProto", 39, 6, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[72], 12, 11), UPB_STRTABLE_INIT(11, 15, UPB_CTYPE_PTR, 4, &strentries[68]),&reftables[16], &reftables[17]),
+  UPB_MSGDEF_INIT("google.protobuf.FileDescriptorSet", 6, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[84], 2, 1), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[84]),&reftables[18], &reftables[19]),
+  UPB_MSGDEF_INIT("google.protobuf.FileOptions", 21, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[6], &arrays[86], 64, 9), UPB_STRTABLE_INIT(10, 15, UPB_CTYPE_PTR, 4, &strentries[88]),&reftables[20], &reftables[21]),
+  UPB_MSGDEF_INIT("google.protobuf.MessageOptions", 8, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[8], &arrays[150], 16, 2), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[104]),&reftables[22], &reftables[23]),
+  UPB_MSGDEF_INIT("google.protobuf.MethodDescriptorProto", 13, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[166], 5, 4), UPB_STRTABLE_INIT(4, 7, UPB_CTYPE_PTR, 3, &strentries[108]),&reftables[24], &reftables[25]),
+  UPB_MSGDEF_INIT("google.protobuf.MethodOptions", 6, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[10], &arrays[171], 4, 0), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[116]),&reftables[26], &reftables[27]),
+  UPB_MSGDEF_INIT("google.protobuf.ServiceDescriptorProto", 11, 2, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[175], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[120]),&reftables[28], &reftables[29]),
+  UPB_MSGDEF_INIT("google.protobuf.ServiceOptions", 6, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[12], &arrays[179], 4, 0), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[124]),&reftables[30], &reftables[31]),
+  UPB_MSGDEF_INIT("google.protobuf.SourceCodeInfo", 6, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[183], 2, 1), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[128]),&reftables[32], &reftables[33]),
+  UPB_MSGDEF_INIT("google.protobuf.SourceCodeInfo.Location", 14, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[185], 5, 4), UPB_STRTABLE_INIT(4, 7, UPB_CTYPE_PTR, 3, &strentries[132]),&reftables[34], &reftables[35]),
+  UPB_MSGDEF_INIT("google.protobuf.UninterpretedOption", 18, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[190], 9, 7), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[140]),&reftables[36], &reftables[37]),
+  UPB_MSGDEF_INIT("google.protobuf.UninterpretedOption.NamePart", 6, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[199], 3, 2), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[156]),&reftables[38], &reftables[39]),
+};
+
+static const upb_fielddef fields[81] = {
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "aggregate_value", 8, &msgs[18], NULL, 15, 6, {0},&reftables[40], &reftables[41]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "allow_alias", 2, &msgs[3], NULL, 6, 1, {0},&reftables[42], &reftables[43]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "cc_generic_services", 16, &msgs[10], NULL, 17, 6, {0},&reftables[44], &reftables[45]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "ctype", 1, &msgs[7], (const upb_def*)(&enums[2]), 6, 1, {0},&reftables[46], &reftables[47]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "default_value", 7, &msgs[6], NULL, 16, 7, {0},&reftables[48], &reftables[49]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_STRING, 0, false, false, false, false, "dependency", 3, &msgs[8], NULL, 30, 8, {0},&reftables[50], &reftables[51]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 3, &msgs[7], NULL, 8, 3, {0},&reftables[52], &reftables[53]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_DOUBLE, 0, false, false, false, false, "double_value", 6, &msgs[18], NULL, 11, 4, {0},&reftables[54], &reftables[55]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "end", 2, &msgs[1], NULL, 3, 1, {0},&reftables[56], &reftables[57]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "enum_type", 4, &msgs[0], (const upb_def*)(&msgs[2]), 16, 2, {0},&reftables[58], &reftables[59]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "enum_type", 5, &msgs[8], (const upb_def*)(&msgs[2]), 13, 1, {0},&reftables[60], &reftables[61]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "experimental_map_key", 9, &msgs[7], NULL, 10, 5, {0},&reftables[62], &reftables[63]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "extendee", 2, &msgs[6], NULL, 7, 2, {0},&reftables[64], &reftables[65]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension", 7, &msgs[8], (const upb_def*)(&msgs[6]), 19, 3, {0},&reftables[66], &reftables[67]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension", 6, &msgs[0], (const upb_def*)(&msgs[6]), 22, 4, {0},&reftables[68], &reftables[69]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension_range", 5, &msgs[0], (const upb_def*)(&msgs[1]), 19, 3, {0},&reftables[70], &reftables[71]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "field", 2, &msgs[0], (const upb_def*)(&msgs[6]), 10, 0, {0},&reftables[72], &reftables[73]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "file", 1, &msgs[9], (const upb_def*)(&msgs[8]), 5, 0, {0},&reftables[74], &reftables[75]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "go_package", 11, &msgs[10], NULL, 14, 5, {0},&reftables[76], &reftables[77]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "identifier_value", 3, &msgs[18], NULL, 6, 1, {0},&reftables[78], &reftables[79]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "input_type", 2, &msgs[12], NULL, 7, 2, {0},&reftables[80], &reftables[81]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REQUIRED, UPB_TYPE_BOOL, 0, false, false, false, false, "is_extension", 2, &msgs[19], NULL, 5, 1, {0},&reftables[82], &reftables[83]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_generate_equals_and_hash", 20, &msgs[10], NULL, 20, 9, {0},&reftables[84], &reftables[85]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_generic_services", 17, &msgs[10], NULL, 18, 7, {0},&reftables[86], &reftables[87]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_multiple_files", 10, &msgs[10], NULL, 13, 4, {0},&reftables[88], &reftables[89]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "java_outer_classname", 8, &msgs[10], NULL, 9, 2, {0},&reftables[90], &reftables[91]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "java_package", 1, &msgs[10], NULL, 6, 1, {0},&reftables[92], &reftables[93]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "label", 4, &msgs[6], (const upb_def*)(&enums[0]), 11, 4, {0},&reftables[94], &reftables[95]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "lazy", 5, &msgs[7], NULL, 9, 4, {0},&reftables[96], &reftables[97]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "leading_comments", 3, &msgs[17], NULL, 8, 2, {0},&reftables[98], &reftables[99]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "location", 1, &msgs[16], (const upb_def*)(&msgs[17]), 5, 0, {0},&reftables[100], &reftables[101]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "message_set_wire_format", 1, &msgs[11], NULL, 6, 1, {0},&reftables[102], &reftables[103]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "message_type", 4, &msgs[8], (const upb_def*)(&msgs[0]), 10, 0, {0},&reftables[104], &reftables[105]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "method", 2, &msgs[14], (const upb_def*)(&msgs[12]), 6, 0, {0},&reftables[106], &reftables[107]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[8], NULL, 22, 6, {0},&reftables[108], &reftables[109]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[14], NULL, 8, 2, {0},&reftables[110], &reftables[111]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "name", 2, &msgs[18], (const upb_def*)(&msgs[19]), 5, 0, {0},&reftables[112], &reftables[113]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[4], NULL, 4, 1, {0},&reftables[114], &reftables[115]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[0], NULL, 24, 6, {0},&reftables[116], &reftables[117]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[12], NULL, 4, 1, {0},&reftables[118], &reftables[119]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[2], NULL, 8, 2, {0},&reftables[120], &reftables[121]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[6], NULL, 4, 1, {0},&reftables[122], &reftables[123]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REQUIRED, UPB_TYPE_STRING, 0, false, false, false, false, "name_part", 1, &msgs[19], NULL, 2, 0, {0},&reftables[124], &reftables[125]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT64, UPB_INTFMT_VARIABLE, false, false, false, false, "negative_int_value", 5, &msgs[18], NULL, 10, 3, {0},&reftables[126], &reftables[127]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "nested_type", 3, &msgs[0], (const upb_def*)(&msgs[0]), 13, 1, {0},&reftables[128], &reftables[129]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "no_standard_descriptor_accessor", 2, &msgs[11], NULL, 7, 2, {0},&reftables[130], &reftables[131]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "number", 3, &msgs[6], NULL, 10, 3, {0},&reftables[132], &reftables[133]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "number", 2, &msgs[4], NULL, 7, 2, {0},&reftables[134], &reftables[135]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "optimize_for", 9, &msgs[10], (const upb_def*)(&enums[3]), 12, 3, {0},&reftables[136], &reftables[137]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 7, &msgs[0], (const upb_def*)(&msgs[11]), 23, 5, {0},&reftables[138], &reftables[139]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[2], (const upb_def*)(&msgs[3]), 7, 1, {0},&reftables[140], &reftables[141]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 8, &msgs[6], (const upb_def*)(&msgs[7]), 3, 0, {0},&reftables[142], &reftables[143]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[4], (const upb_def*)(&msgs[5]), 3, 0, {0},&reftables[144], &reftables[145]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 8, &msgs[8], (const upb_def*)(&msgs[10]), 20, 4, {0},&reftables[146], &reftables[147]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[14], (const upb_def*)(&msgs[15]), 7, 1, {0},&reftables[148], &reftables[149]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 4, &msgs[12], (const upb_def*)(&msgs[13]), 3, 0, {0},&reftables[150], &reftables[151]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "output_type", 3, &msgs[12], NULL, 10, 3, {0},&reftables[152], &reftables[153]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "package", 2, &msgs[8], NULL, 25, 7, {0},&reftables[154], &reftables[155]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "packed", 2, &msgs[7], NULL, 7, 2, {0},&reftables[156], &reftables[157]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, true, "path", 1, &msgs[17], NULL, 4, 0, {0},&reftables[158], &reftables[159]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_UINT64, UPB_INTFMT_VARIABLE, false, false, false, false, "positive_int_value", 4, &msgs[18], NULL, 9, 2, {0},&reftables[160], &reftables[161]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "public_dependency", 10, &msgs[8], NULL, 35, 9, {0},&reftables[162], &reftables[163]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "py_generic_services", 18, &msgs[10], NULL, 19, 8, {0},&reftables[164], &reftables[165]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "service", 6, &msgs[8], (const upb_def*)(&msgs[14]), 16, 2, {0},&reftables[166], &reftables[167]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "source_code_info", 9, &msgs[8], (const upb_def*)(&msgs[16]), 21, 5, {0},&reftables[168], &reftables[169]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, true, "span", 2, &msgs[17], NULL, 7, 1, {0},&reftables[170], &reftables[171]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "start", 1, &msgs[1], NULL, 2, 0, {0},&reftables[172], &reftables[173]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BYTES, 0, false, false, false, false, "string_value", 7, &msgs[18], NULL, 12, 5, {0},&reftables[174], &reftables[175]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "trailing_comments", 4, &msgs[17], NULL, 11, 3, {0},&reftables[176], &reftables[177]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "type", 5, &msgs[6], (const upb_def*)(&enums[1]), 12, 5, {0},&reftables[178], &reftables[179]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "type_name", 6, &msgs[6], NULL, 13, 6, {0},&reftables[180], &reftables[181]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[5], (const upb_def*)(&msgs[18]), 5, 0, {0},&reftables[182], &reftables[183]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[15], (const upb_def*)(&msgs[18]), 5, 0, {0},&reftables[184], &reftables[185]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[3], (const upb_def*)(&msgs[18]), 5, 0, {0},&reftables[186], &reftables[187]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[13], (const upb_def*)(&msgs[18]), 5, 0, {0},&reftables[188], &reftables[189]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[10], (const upb_def*)(&msgs[18]), 5, 0, {0},&reftables[190], &reftables[191]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[11], (const upb_def*)(&msgs[18]), 5, 0, {0},&reftables[192], &reftables[193]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[7], (const upb_def*)(&msgs[18]), 5, 0, {0},&reftables[194], &reftables[195]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "value", 2, &msgs[2], (const upb_def*)(&msgs[4]), 6, 0, {0},&reftables[196], &reftables[197]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "weak", 10, &msgs[7], NULL, 13, 6, {0},&reftables[198], &reftables[199]),
+  UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "weak_dependency", 11, &msgs[8], NULL, 38, 10, {0},&reftables[200], &reftables[201]),
+};
+
+static const upb_enumdef enums[4] = {
+  UPB_ENUMDEF_INIT("google.protobuf.FieldDescriptorProto.Label", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[160]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[202], 4, 3), 0, &reftables[202], &reftables[203]),
+  UPB_ENUMDEF_INIT("google.protobuf.FieldDescriptorProto.Type", UPB_STRTABLE_INIT(18, 31, UPB_CTYPE_INT32, 5, &strentries[164]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[206], 19, 18), 0, &reftables[204], &reftables[205]),
+  UPB_ENUMDEF_INIT("google.protobuf.FieldOptions.CType", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[196]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[225], 3, 3), 0, &reftables[206], &reftables[207]),
+  UPB_ENUMDEF_INIT("google.protobuf.FileOptions.OptimizeMode", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[200]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[228], 4, 3), 0, &reftables[208], &reftables[209]),
+};
+
+static const upb_tabent strentries[236] = {
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "extension"), UPB_TABVALUE_PTR_INIT(&fields[14]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[38]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "field"), UPB_TABVALUE_PTR_INIT(&fields[16]), NULL},
+  {UPB_TABKEY_STR("\017", "\000", "\000", "\000", "extension_range"), UPB_TABVALUE_PTR_INIT(&fields[15]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "nested_type"), UPB_TABVALUE_PTR_INIT(&fields[44]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[49]), NULL},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "enum_type"), UPB_TABVALUE_PTR_INIT(&fields[9]), &strentries[14]},
+  {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "start"), UPB_TABVALUE_PTR_INIT(&fields[66]), NULL},
+  {UPB_TABKEY_STR("\003", "\000", "\000", "\000", "end"), UPB_TABVALUE_PTR_INIT(&fields[8]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "value"), UPB_TABVALUE_PTR_INIT(&fields[78]), NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[50]), NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[40]), &strentries[22]},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[73]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "allow_alias"), UPB_TABVALUE_PTR_INIT(&fields[1]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "number"), UPB_TABVALUE_PTR_INIT(&fields[47]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[52]), NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[37]), &strentries[30]},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[71]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "label"), UPB_TABVALUE_PTR_INIT(&fields[27]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[41]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "number"), UPB_TABVALUE_PTR_INIT(&fields[46]), &strentries[49]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "type_name"), UPB_TABVALUE_PTR_INIT(&fields[70]), NULL},
+  {UPB_TABKEY_STR("\010", "\000", "\000", "\000", "extendee"), UPB_TABVALUE_PTR_INIT(&fields[12]), NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "type"), UPB_TABVALUE_PTR_INIT(&fields[69]), &strentries[48]},
+  {UPB_TABKEY_STR("\015", "\000", "\000", "\000", "default_value"), UPB_TABVALUE_PTR_INIT(&fields[4]), NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[51]), NULL},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "experimental_map_key"), UPB_TABVALUE_PTR_INIT(&fields[11]), &strentries[67]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "weak"), UPB_TABVALUE_PTR_INIT(&fields[79]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "packed"), UPB_TABVALUE_PTR_INIT(&fields[58]), NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "lazy"), UPB_TABVALUE_PTR_INIT(&fields[28]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "ctype"), UPB_TABVALUE_PTR_INIT(&fields[3]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[6]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[77]), NULL},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "extension"), UPB_TABVALUE_PTR_INIT(&fields[13]), NULL},
+  {UPB_TABKEY_STR("\017", "\000", "\000", "\000", "weak_dependency"), UPB_TABVALUE_PTR_INIT(&fields[80]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[34]), NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "service"), UPB_TABVALUE_PTR_INIT(&fields[63]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "source_code_info"), UPB_TABVALUE_PTR_INIT(&fields[64]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "dependency"), UPB_TABVALUE_PTR_INIT(&fields[5]), NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "message_type"), UPB_TABVALUE_PTR_INIT(&fields[32]), NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "package"), UPB_TABVALUE_PTR_INIT(&fields[57]), NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[53]), &strentries[82]},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "enum_type"), UPB_TABVALUE_PTR_INIT(&fields[10]), NULL},
+  {UPB_TABKEY_STR("\021", "\000", "\000", "\000", "public_dependency"), UPB_TABVALUE_PTR_INIT(&fields[61]), &strentries[81]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "file"), UPB_TABVALUE_PTR_INIT(&fields[17]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[75]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\023", "\000", "\000", "\000", "cc_generic_services"), UPB_TABVALUE_PTR_INIT(&fields[2]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\023", "\000", "\000", "\000", "java_multiple_files"), UPB_TABVALUE_PTR_INIT(&fields[24]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\025", "\000", "\000", "\000", "java_generic_services"), UPB_TABVALUE_PTR_INIT(&fields[23]), &strentries[102]},
+  {UPB_TABKEY_STR("\035", "\000", "\000", "\000", "java_generate_equals_and_hash"), UPB_TABVALUE_PTR_INIT(&fields[22]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "go_package"), UPB_TABVALUE_PTR_INIT(&fields[18]), NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "java_package"), UPB_TABVALUE_PTR_INIT(&fields[26]), NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "optimize_for"), UPB_TABVALUE_PTR_INIT(&fields[48]), NULL},
+  {UPB_TABKEY_STR("\023", "\000", "\000", "\000", "py_generic_services"), UPB_TABVALUE_PTR_INIT(&fields[62]), NULL},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "java_outer_classname"), UPB_TABVALUE_PTR_INIT(&fields[25]), NULL},
+  {UPB_TABKEY_STR("\027", "\000", "\000", "\000", "message_set_wire_format"), UPB_TABVALUE_PTR_INIT(&fields[31]), &strentries[106]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[76]), NULL},
+  {UPB_TABKEY_STR("\037", "\000", "\000", "\000", "no_standard_descriptor_accessor"), UPB_TABVALUE_PTR_INIT(&fields[45]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[39]), NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "input_type"), UPB_TABVALUE_PTR_INIT(&fields[20]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "output_type"), UPB_TABVALUE_PTR_INIT(&fields[56]), NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[55]), NULL},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[74]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[54]), &strentries[122]},
+  {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "method"), UPB_TABVALUE_PTR_INIT(&fields[33]), NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[35]), &strentries[121]},
+  {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[72]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\010", "\000", "\000", "\000", "location"), UPB_TABVALUE_PTR_INIT(&fields[30]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "span"), UPB_TABVALUE_PTR_INIT(&fields[65]), &strentries[139]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\021", "\000", "\000", "\000", "trailing_comments"), UPB_TABVALUE_PTR_INIT(&fields[68]), NULL},
+  {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "leading_comments"), UPB_TABVALUE_PTR_INIT(&fields[29]), &strentries[137]},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "path"), UPB_TABVALUE_PTR_INIT(&fields[59]), NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "double_value"), UPB_TABVALUE_PTR_INIT(&fields[7]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[36]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\022", "\000", "\000", "\000", "negative_int_value"), UPB_TABVALUE_PTR_INIT(&fields[43]), NULL},
+  {UPB_TABKEY_STR("\017", "\000", "\000", "\000", "aggregate_value"), UPB_TABVALUE_PTR_INIT(&fields[0]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\022", "\000", "\000", "\000", "positive_int_value"), UPB_TABVALUE_PTR_INIT(&fields[60]), NULL},
+  {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "identifier_value"), UPB_TABVALUE_PTR_INIT(&fields[19]), NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "string_value"), UPB_TABVALUE_PTR_INIT(&fields[67]), &strentries[154]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "is_extension"), UPB_TABVALUE_PTR_INIT(&fields[21]), NULL},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "name_part"), UPB_TABVALUE_PTR_INIT(&fields[42]), NULL},
+  {UPB_TABKEY_STR("\016", "\000", "\000", "\000", "LABEL_REQUIRED"), UPB_TABVALUE_INT_INIT(2), &strentries[162]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\016", "\000", "\000", "\000", "LABEL_REPEATED"), UPB_TABVALUE_INT_INIT(3), NULL},
+  {UPB_TABKEY_STR("\016", "\000", "\000", "\000", "LABEL_OPTIONAL"), UPB_TABVALUE_INT_INIT(1), NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "TYPE_FIXED64"), UPB_TABVALUE_INT_INIT(6), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_STRING"), UPB_TABVALUE_INT_INIT(9), NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_FLOAT"), UPB_TABVALUE_INT_INIT(2), &strentries[193]},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_DOUBLE"), UPB_TABVALUE_INT_INIT(1), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_INT32"), UPB_TABVALUE_INT_INIT(5), NULL},
+  {UPB_TABKEY_STR("\015", "\000", "\000", "\000", "TYPE_SFIXED32"), UPB_TABVALUE_INT_INIT(15), NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "TYPE_FIXED32"), UPB_TABVALUE_INT_INIT(7), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "TYPE_MESSAGE"), UPB_TABVALUE_INT_INIT(11), &strentries[194]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_INT64"), UPB_TABVALUE_INT_INIT(3), &strentries[191]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "TYPE_ENUM"), UPB_TABVALUE_INT_INIT(14), NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_UINT32"), UPB_TABVALUE_INT_INIT(13), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_UINT64"), UPB_TABVALUE_INT_INIT(4), &strentries[190]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\015", "\000", "\000", "\000", "TYPE_SFIXED64"), UPB_TABVALUE_INT_INIT(16), NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_BYTES"), UPB_TABVALUE_INT_INIT(12), NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_SINT64"), UPB_TABVALUE_INT_INIT(18), NULL},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "TYPE_BOOL"), UPB_TABVALUE_INT_INIT(8), NULL},
+  {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_GROUP"), UPB_TABVALUE_INT_INIT(10), NULL},
+  {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_SINT32"), UPB_TABVALUE_INT_INIT(17), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "CORD"), UPB_TABVALUE_INT_INIT(1), NULL},
+  {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "STRING"), UPB_TABVALUE_INT_INIT(0), &strentries[197]},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "STRING_PIECE"), UPB_TABVALUE_INT_INIT(2), NULL},
+  {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "CODE_SIZE"), UPB_TABVALUE_INT_INIT(2), NULL},
+  {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "SPEED"), UPB_TABVALUE_INT_INIT(1), &strentries[203]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "LITE_RUNTIME"), UPB_TABVALUE_INT_INIT(3), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\047", "\000", "\000", "\000", "google.protobuf.SourceCodeInfo.Location"), UPB_TABVALUE_PTR_INIT(&msgs[17]), NULL},
+  {UPB_TABKEY_STR("\043", "\000", "\000", "\000", "google.protobuf.UninterpretedOption"), UPB_TABVALUE_PTR_INIT(&msgs[18]), NULL},
+  {UPB_TABKEY_STR("\043", "\000", "\000", "\000", "google.protobuf.FileDescriptorProto"), UPB_TABVALUE_PTR_INIT(&msgs[8]), NULL},
+  {UPB_TABKEY_STR("\045", "\000", "\000", "\000", "google.protobuf.MethodDescriptorProto"), UPB_TABVALUE_PTR_INIT(&msgs[12]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\040", "\000", "\000", "\000", "google.protobuf.EnumValueOptions"), UPB_TABVALUE_PTR_INIT(&msgs[5]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\037", "\000", "\000", "\000", "google.protobuf.DescriptorProto"), UPB_TABVALUE_PTR_INIT(&msgs[0]), &strentries[228]},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\036", "\000", "\000", "\000", "google.protobuf.SourceCodeInfo"), UPB_TABVALUE_PTR_INIT(&msgs[16]), NULL},
+  {UPB_TABKEY_STR("\051", "\000", "\000", "\000", "google.protobuf.FieldDescriptorProto.Type"), UPB_TABVALUE_PTR_INIT(&enums[1]), NULL},
+  {UPB_TABKEY_STR("\056", "\000", "\000", "\000", "google.protobuf.DescriptorProto.ExtensionRange"), UPB_TABVALUE_PTR_INIT(&msgs[1]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_STR("\050", "\000", "\000", "\000", "google.protobuf.EnumValueDescriptorProto"), UPB_TABVALUE_PTR_INIT(&msgs[4]), NULL},
+  {UPB_TABKEY_STR("\034", "\000", "\000", "\000", "google.protobuf.FieldOptions"), UPB_TABVALUE_PTR_INIT(&msgs[7]), NULL},
+  {UPB_TABKEY_STR("\033", "\000", "\000", "\000", "google.protobuf.FileOptions"), UPB_TABVALUE_PTR_INIT(&msgs[10]), NULL},
+  {UPB_TABKEY_STR("\043", "\000", "\000", "\000", "google.protobuf.EnumDescriptorProto"), UPB_TABVALUE_PTR_INIT(&msgs[2]), &strentries[233]},
+  {UPB_TABKEY_STR("\052", "\000", "\000", "\000", "google.protobuf.FieldDescriptorProto.Label"), UPB_TABVALUE_PTR_INIT(&enums[0]), NULL},
+  {UPB_TABKEY_STR("\046", "\000", "\000", "\000", "google.protobuf.ServiceDescriptorProto"), UPB_TABVALUE_PTR_INIT(&msgs[14]), NULL},
+  {UPB_TABKEY_STR("\042", "\000", "\000", "\000", "google.protobuf.FieldOptions.CType"), UPB_TABVALUE_PTR_INIT(&enums[2]), &strentries[229]},
+  {UPB_TABKEY_STR("\041", "\000", "\000", "\000", "google.protobuf.FileDescriptorSet"), UPB_TABVALUE_PTR_INIT(&msgs[9]), &strentries[235]},
+  {UPB_TABKEY_STR("\033", "\000", "\000", "\000", "google.protobuf.EnumOptions"), UPB_TABVALUE_PTR_INIT(&msgs[3]), NULL},
+  {UPB_TABKEY_STR("\044", "\000", "\000", "\000", "google.protobuf.FieldDescriptorProto"), UPB_TABVALUE_PTR_INIT(&msgs[6]), NULL},
+  {UPB_TABKEY_STR("\050", "\000", "\000", "\000", "google.protobuf.FileOptions.OptimizeMode"), UPB_TABVALUE_PTR_INIT(&enums[3]), &strentries[221]},
+  {UPB_TABKEY_STR("\036", "\000", "\000", "\000", "google.protobuf.ServiceOptions"), UPB_TABVALUE_PTR_INIT(&msgs[15]), NULL},
+  {UPB_TABKEY_STR("\036", "\000", "\000", "\000", "google.protobuf.MessageOptions"), UPB_TABVALUE_PTR_INIT(&msgs[11]), NULL},
+  {UPB_TABKEY_STR("\035", "\000", "\000", "\000", "google.protobuf.MethodOptions"), UPB_TABVALUE_PTR_INIT(&msgs[13]), &strentries[226]},
+  {UPB_TABKEY_STR("\054", "\000", "\000", "\000", "google.protobuf.UninterpretedOption.NamePart"), UPB_TABVALUE_PTR_INIT(&msgs[19]), NULL},
+};
+
+static const upb_tabent intentries[14] = {
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[73]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[71]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[77]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[75]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[76]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[74]), NULL},
+  {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL},
+  {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[72]), NULL},
+};
+
+static const upb_tabval arrays[232] = {
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[38]),
+  UPB_TABVALUE_PTR_INIT(&fields[16]),
+  UPB_TABVALUE_PTR_INIT(&fields[44]),
+  UPB_TABVALUE_PTR_INIT(&fields[9]),
+  UPB_TABVALUE_PTR_INIT(&fields[15]),
+  UPB_TABVALUE_PTR_INIT(&fields[14]),
+  UPB_TABVALUE_PTR_INIT(&fields[49]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[66]),
+  UPB_TABVALUE_PTR_INIT(&fields[8]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[40]),
+  UPB_TABVALUE_PTR_INIT(&fields[78]),
+  UPB_TABVALUE_PTR_INIT(&fields[50]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[1]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[37]),
+  UPB_TABVALUE_PTR_INIT(&fields[47]),
+  UPB_TABVALUE_PTR_INIT(&fields[52]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[41]),
+  UPB_TABVALUE_PTR_INIT(&fields[12]),
+  UPB_TABVALUE_PTR_INIT(&fields[46]),
+  UPB_TABVALUE_PTR_INIT(&fields[27]),
+  UPB_TABVALUE_PTR_INIT(&fields[69]),
+  UPB_TABVALUE_PTR_INIT(&fields[70]),
+  UPB_TABVALUE_PTR_INIT(&fields[4]),
+  UPB_TABVALUE_PTR_INIT(&fields[51]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[3]),
+  UPB_TABVALUE_PTR_INIT(&fields[58]),
+  UPB_TABVALUE_PTR_INIT(&fields[6]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[28]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[11]),
+  UPB_TABVALUE_PTR_INIT(&fields[79]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[34]),
+  UPB_TABVALUE_PTR_INIT(&fields[57]),
+  UPB_TABVALUE_PTR_INIT(&fields[5]),
+  UPB_TABVALUE_PTR_INIT(&fields[32]),
+  UPB_TABVALUE_PTR_INIT(&fields[10]),
+  UPB_TABVALUE_PTR_INIT(&fields[63]),
+  UPB_TABVALUE_PTR_INIT(&fields[13]),
+  UPB_TABVALUE_PTR_INIT(&fields[53]),
+  UPB_TABVALUE_PTR_INIT(&fields[64]),
+  UPB_TABVALUE_PTR_INIT(&fields[61]),
+  UPB_TABVALUE_PTR_INIT(&fields[80]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[17]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[26]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[25]),
+  UPB_TABVALUE_PTR_INIT(&fields[48]),
+  UPB_TABVALUE_PTR_INIT(&fields[24]),
+  UPB_TABVALUE_PTR_INIT(&fields[18]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[2]),
+  UPB_TABVALUE_PTR_INIT(&fields[23]),
+  UPB_TABVALUE_PTR_INIT(&fields[62]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[22]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[31]),
+  UPB_TABVALUE_PTR_INIT(&fields[45]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[39]),
+  UPB_TABVALUE_PTR_INIT(&fields[20]),
+  UPB_TABVALUE_PTR_INIT(&fields[56]),
+  UPB_TABVALUE_PTR_INIT(&fields[55]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[35]),
+  UPB_TABVALUE_PTR_INIT(&fields[33]),
+  UPB_TABVALUE_PTR_INIT(&fields[54]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[30]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[59]),
+  UPB_TABVALUE_PTR_INIT(&fields[65]),
+  UPB_TABVALUE_PTR_INIT(&fields[29]),
+  UPB_TABVALUE_PTR_INIT(&fields[68]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[36]),
+  UPB_TABVALUE_PTR_INIT(&fields[19]),
+  UPB_TABVALUE_PTR_INIT(&fields[60]),
+  UPB_TABVALUE_PTR_INIT(&fields[43]),
+  UPB_TABVALUE_PTR_INIT(&fields[7]),
+  UPB_TABVALUE_PTR_INIT(&fields[67]),
+  UPB_TABVALUE_PTR_INIT(&fields[0]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT(&fields[42]),
+  UPB_TABVALUE_PTR_INIT(&fields[21]),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT("LABEL_OPTIONAL"),
+  UPB_TABVALUE_PTR_INIT("LABEL_REQUIRED"),
+  UPB_TABVALUE_PTR_INIT("LABEL_REPEATED"),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT("TYPE_DOUBLE"),
+  UPB_TABVALUE_PTR_INIT("TYPE_FLOAT"),
+  UPB_TABVALUE_PTR_INIT("TYPE_INT64"),
+  UPB_TABVALUE_PTR_INIT("TYPE_UINT64"),
+  UPB_TABVALUE_PTR_INIT("TYPE_INT32"),
+  UPB_TABVALUE_PTR_INIT("TYPE_FIXED64"),
+  UPB_TABVALUE_PTR_INIT("TYPE_FIXED32"),
+  UPB_TABVALUE_PTR_INIT("TYPE_BOOL"),
+  UPB_TABVALUE_PTR_INIT("TYPE_STRING"),
+  UPB_TABVALUE_PTR_INIT("TYPE_GROUP"),
+  UPB_TABVALUE_PTR_INIT("TYPE_MESSAGE"),
+  UPB_TABVALUE_PTR_INIT("TYPE_BYTES"),
+  UPB_TABVALUE_PTR_INIT("TYPE_UINT32"),
+  UPB_TABVALUE_PTR_INIT("TYPE_ENUM"),
+  UPB_TABVALUE_PTR_INIT("TYPE_SFIXED32"),
+  UPB_TABVALUE_PTR_INIT("TYPE_SFIXED64"),
+  UPB_TABVALUE_PTR_INIT("TYPE_SINT32"),
+  UPB_TABVALUE_PTR_INIT("TYPE_SINT64"),
+  UPB_TABVALUE_PTR_INIT("STRING"),
+  UPB_TABVALUE_PTR_INIT("CORD"),
+  UPB_TABVALUE_PTR_INIT("STRING_PIECE"),
+  UPB_TABVALUE_EMPTY_INIT,
+  UPB_TABVALUE_PTR_INIT("SPEED"),
+  UPB_TABVALUE_PTR_INIT("CODE_SIZE"),
+  UPB_TABVALUE_PTR_INIT("LITE_RUNTIME"),
+};
+
+static const upb_symtab symtab = UPB_SYMTAB_INIT(UPB_STRTABLE_INIT(24, 31, UPB_CTYPE_PTR, 5, &strentries[204]), &reftables[210], &reftables[211]);
+
+const upb_symtab *upbdefs_google_protobuf_descriptor(const void *owner) {
+  upb_symtab_ref(&symtab, owner);
+  return &symtab;
+}
+
+#ifdef UPB_DEBUG_REFS
+static upb_inttable reftables[212] = {
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+  UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR),
+};
+#endif
+
+/*
+** XXX: The routines in this file that consume a string do not currently
+** support having the string span buffers.  In the future, as upb_sink and
+** its buffering/sharing functionality evolve there should be an easy and
+** idiomatic way of correctly handling this case.  For now, we accept this
+** limitation since we currently only parse descriptors from single strings.
+*/
+
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+/* upb_deflist is an internal-only dynamic array for storing a growing list of
+ * upb_defs. */
+typedef struct {
+  upb_def **defs;
+  size_t len;
+  size_t size;
+  bool owned;
+} upb_deflist;
+
+/* We keep a stack of all the messages scopes we are currently in, as well as
+ * the top-level file scope.  This is necessary to correctly qualify the
+ * definitions that are contained inside.  "name" tracks the name of the
+ * message or package (a bare name -- not qualified by any enclosing scopes). */
+typedef struct {
+  char *name;
+  /* Index of the first def that is under this scope.  For msgdefs, the
+   * msgdef itself is at start-1. */
+  int start;
+} upb_descreader_frame;
+
+/* The maximum number of nested declarations that are allowed, ie.
+ * message Foo {
+ *   message Bar {
+ *     message Baz {
+ *     }
+ *   }
+ * }
+ *
+ * This is a resource limit that affects how big our runtime stack can grow.
+ * TODO: make this a runtime-settable property of the Reader instance. */
+#define UPB_MAX_MESSAGE_NESTING 64
+
+struct upb_descreader {
+  upb_sink sink;
+  upb_deflist defs;
+  upb_descreader_frame stack[UPB_MAX_MESSAGE_NESTING];
+  int stack_len;
+
+  uint32_t number;
+  char *name;
+  bool saw_number;
+  bool saw_name;
+
+  char *default_string;
+
+  upb_fielddef *f;
+};
+
+static char *upb_strndup(const char *buf, size_t n) {
+  char *ret = malloc(n + 1);
+  if (!ret) return NULL;
+  memcpy(ret, buf, n);
+  ret[n] = '\0';
+  return ret;
+}
+
+/* Returns a newly allocated string that joins input strings together, for
+ * example:
+ *   join("Foo.Bar", "Baz") -> "Foo.Bar.Baz"
+ *   join("", "Baz") -> "Baz"
+ * Caller owns a ref on the returned string. */
+static char *upb_join(const char *base, const char *name) {
+  if (!base || strlen(base) == 0) {
+    return upb_strdup(name);
+  } else {
+    char *ret = malloc(strlen(base) + strlen(name) + 2);
+    ret[0] = '\0';
+    strcat(ret, base);
+    strcat(ret, ".");
+    strcat(ret, name);
+    return ret;
+  }
+}
+
+
+/* upb_deflist ****************************************************************/
+
+void upb_deflist_init(upb_deflist *l) {
+  l->size = 0;
+  l->defs = NULL;
+  l->len = 0;
+  l->owned = true;
+}
+
+void upb_deflist_uninit(upb_deflist *l) {
+  size_t i;
+  if (l->owned)
+    for(i = 0; i < l->len; i++)
+      upb_def_unref(l->defs[i], l);
+  free(l->defs);
+}
+
+bool upb_deflist_push(upb_deflist *l, upb_def *d) {
+  if(++l->len >= l->size) {
+    size_t new_size = UPB_MAX(l->size, 4);
+    new_size *= 2;
+    l->defs = realloc(l->defs, new_size * sizeof(void *));
+    if (!l->defs) return false;
+    l->size = new_size;
+  }
+  l->defs[l->len - 1] = d;
+  return true;
+}
+
+void upb_deflist_donaterefs(upb_deflist *l, void *owner) {
+  size_t i;
+  assert(l->owned);
+  for (i = 0; i < l->len; i++)
+    upb_def_donateref(l->defs[i], l, owner);
+  l->owned = false;
+}
+
+static upb_def *upb_deflist_last(upb_deflist *l) {
+  return l->defs[l->len-1];
+}
+
+/* Qualify the defname for all defs starting with offset "start" with "str". */
+static void upb_deflist_qualify(upb_deflist *l, char *str, int32_t start) {
+  uint32_t i;
+  for (i = start; i < l->len; i++) {
+    upb_def *def = l->defs[i];
+    char *name = upb_join(str, upb_def_fullname(def));
+    upb_def_setfullname(def, name, NULL);
+    free(name);
+  }
+}
+
+
+/* upb_descreader  ************************************************************/
+
+static upb_msgdef *upb_descreader_top(upb_descreader *r) {
+  int index;
+  assert(r->stack_len > 1);
+  index = r->stack[r->stack_len-1].start - 1;
+  assert(index >= 0);
+  return upb_downcast_msgdef_mutable(r->defs.defs[index]);
+}
+
+static upb_def *upb_descreader_last(upb_descreader *r) {
+  return upb_deflist_last(&r->defs);
+}
+
+/* Start/end handlers for FileDescriptorProto and DescriptorProto (the two
+ * entities that have names and can contain sub-definitions. */
+void upb_descreader_startcontainer(upb_descreader *r) {
+  upb_descreader_frame *f = &r->stack[r->stack_len++];
+  f->start = r->defs.len;
+  f->name = NULL;
+}
+
+void upb_descreader_endcontainer(upb_descreader *r) {
+  upb_descreader_frame *f = &r->stack[--r->stack_len];
+  upb_deflist_qualify(&r->defs, f->name, f->start);
+  free(f->name);
+  f->name = NULL;
+}
+
+void upb_descreader_setscopename(upb_descreader *r, char *str) {
+  upb_descreader_frame *f = &r->stack[r->stack_len-1];
+  free(f->name);
+  f->name = str;
+}
+
+/* Handlers for google.protobuf.FileDescriptorProto. */
+static bool file_startmsg(void *r, const void *hd) {
+  UPB_UNUSED(hd);
+  upb_descreader_startcontainer(r);
+  return true;
+}
+
+static bool file_endmsg(void *closure, const void *hd, upb_status *status) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  UPB_UNUSED(status);
+  upb_descreader_endcontainer(r);
+  return true;
+}
+
+static size_t file_onpackage(void *closure, const void *hd, const char *buf,
+                             size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+  /* XXX: see comment at the top of the file. */
+  upb_descreader_setscopename(r, upb_strndup(buf, n));
+  return n;
+}
+
+/* Handlers for google.protobuf.EnumValueDescriptorProto. */
+static bool enumval_startmsg(void *closure, const void *hd) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  r->saw_number = false;
+  r->saw_name = false;
+  return true;
+}
+
+static size_t enumval_onname(void *closure, const void *hd, const char *buf,
+                             size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+  /* XXX: see comment at the top of the file. */
+  free(r->name);
+  r->name = upb_strndup(buf, n);
+  r->saw_name = true;
+  return n;
+}
+
+static bool enumval_onnumber(void *closure, const void *hd, int32_t val) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  r->number = val;
+  r->saw_number = true;
+  return true;
+}
+
+static bool enumval_endmsg(void *closure, const void *hd, upb_status *status) {
+  upb_descreader *r = closure;
+  upb_enumdef *e;
+  UPB_UNUSED(hd);
+
+  if(!r->saw_number || !r->saw_name) {
+    upb_status_seterrmsg(status, "Enum value missing name or number.");
+    return false;
+  }
+  e = upb_downcast_enumdef_mutable(upb_descreader_last(r));
+  upb_enumdef_addval(e, r->name, r->number, status);
+  free(r->name);
+  r->name = NULL;
+  return true;
+}
+
+
+/* Handlers for google.protobuf.EnumDescriptorProto. */
+static bool enum_startmsg(void *closure, const void *hd) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  upb_deflist_push(&r->defs,
+                   upb_enumdef_upcast_mutable(upb_enumdef_new(&r->defs)));
+  return true;
+}
+
+static bool enum_endmsg(void *closure, const void *hd, upb_status *status) {
+  upb_descreader *r = closure;
+  upb_enumdef *e;
+  UPB_UNUSED(hd);
+
+  e = upb_downcast_enumdef_mutable(upb_descreader_last(r));
+  if (upb_def_fullname(upb_descreader_last(r)) == NULL) {
+    upb_status_seterrmsg(status, "Enum had no name.");
+    return false;
+  }
+  if (upb_enumdef_numvals(e) == 0) {
+    upb_status_seterrmsg(status, "Enum had no values.");
+    return false;
+  }
+  return true;
+}
+
+static size_t enum_onname(void *closure, const void *hd, const char *buf,
+                          size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  char *fullname = upb_strndup(buf, n);
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+  /* XXX: see comment at the top of the file. */
+  upb_def_setfullname(upb_descreader_last(r), fullname, NULL);
+  free(fullname);
+  return n;
+}
+
+/* Handlers for google.protobuf.FieldDescriptorProto */
+static bool field_startmsg(void *closure, const void *hd) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  r->f = upb_fielddef_new(&r->defs);
+  free(r->default_string);
+  r->default_string = NULL;
+
+  /* fielddefs default to packed, but descriptors default to non-packed. */
+  upb_fielddef_setpacked(r->f, false);
+  return true;
+}
+
+/* Converts the default value in string "str" into "d".  Passes a ref on str.
+ * Returns true on success. */
+static bool parse_default(char *str, upb_fielddef *f) {
+  bool success = true;
+  char *end;
+  switch (upb_fielddef_type(f)) {
+    case UPB_TYPE_INT32: {
+      long val = strtol(str, &end, 0);
+      if (val > INT32_MAX || val < INT32_MIN || errno == ERANGE || *end)
+        success = false;
+      else
+        upb_fielddef_setdefaultint32(f, val);
+      break;
+    }
+    case UPB_TYPE_INT64: {
+      /* XXX: Need to write our own strtoll, since it's not available in c89. */
+      long long val = strtol(str, &end, 0);
+      if (val > INT64_MAX || val < INT64_MIN || errno == ERANGE || *end)
+        success = false;
+      else
+        upb_fielddef_setdefaultint64(f, val);
+      break;
+    }
+    case UPB_TYPE_UINT32: {
+      unsigned long val = strtoul(str, &end, 0);
+      if (val > UINT32_MAX || errno == ERANGE || *end)
+        success = false;
+      else
+        upb_fielddef_setdefaultuint32(f, val);
+      break;
+    }
+    case UPB_TYPE_UINT64: {
+      /* XXX: Need to write our own strtoull, since it's not available in c89. */
+      unsigned long long val = strtoul(str, &end, 0);
+      if (val > UINT64_MAX || errno == ERANGE || *end)
+        success = false;
+      else
+        upb_fielddef_setdefaultuint64(f, val);
+      break;
+    }
+    case UPB_TYPE_DOUBLE: {
+      double val = strtod(str, &end);
+      if (errno == ERANGE || *end)
+        success = false;
+      else
+        upb_fielddef_setdefaultdouble(f, val);
+      break;
+    }
+    case UPB_TYPE_FLOAT: {
+      /* XXX: Need to write our own strtof, since it's not available in c89. */
+      float val = strtod(str, &end);
+      if (errno == ERANGE || *end)
+        success = false;
+      else
+        upb_fielddef_setdefaultfloat(f, val);
+      break;
+    }
+    case UPB_TYPE_BOOL: {
+      if (strcmp(str, "false") == 0)
+        upb_fielddef_setdefaultbool(f, false);
+      else if (strcmp(str, "true") == 0)
+        upb_fielddef_setdefaultbool(f, true);
+      else
+        success = false;
+      break;
+    }
+    default: abort();
+  }
+  return success;
+}
+
+static bool field_endmsg(void *closure, const void *hd, upb_status *status) {
+  upb_descreader *r = closure;
+  upb_fielddef *f = r->f;
+  UPB_UNUSED(hd);
+
+  /* TODO: verify that all required fields were present. */
+  assert(upb_fielddef_number(f) != 0);
+  assert(upb_fielddef_name(f) != NULL);
+  assert((upb_fielddef_subdefname(f) != NULL) == upb_fielddef_hassubdef(f));
+
+  if (r->default_string) {
+    if (upb_fielddef_issubmsg(f)) {
+      upb_status_seterrmsg(status, "Submessages cannot have defaults.");
+      return false;
+    }
+    if (upb_fielddef_isstring(f) || upb_fielddef_type(f) == UPB_TYPE_ENUM) {
+      upb_fielddef_setdefaultcstr(f, r->default_string, NULL);
+    } else {
+      if (r->default_string && !parse_default(r->default_string, f)) {
+        /* We don't worry too much about giving a great error message since the
+         * compiler should have ensured this was correct. */
+        upb_status_seterrmsg(status, "Error converting default value.");
+        return false;
+      }
+    }
+  }
+  return true;
+}
+
+static bool field_onlazy(void *closure, const void *hd, bool val) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+
+  upb_fielddef_setlazy(r->f, val);
+  return true;
+}
+
+static bool field_onpacked(void *closure, const void *hd, bool val) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+
+  upb_fielddef_setpacked(r->f, val);
+  return true;
+}
+
+static bool field_ontype(void *closure, const void *hd, int32_t val) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+
+  upb_fielddef_setdescriptortype(r->f, val);
+  return true;
+}
+
+static bool field_onlabel(void *closure, const void *hd, int32_t val) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+
+  upb_fielddef_setlabel(r->f, val);
+  return true;
+}
+
+static bool field_onnumber(void *closure, const void *hd, int32_t val) {
+  upb_descreader *r = closure;
+  bool ok = upb_fielddef_setnumber(r->f, val, NULL);
+  UPB_UNUSED(hd);
+
+  UPB_ASSERT_VAR(ok, ok);
+  return true;
+}
+
+static size_t field_onname(void *closure, const void *hd, const char *buf,
+                           size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  char *name = upb_strndup(buf, n);
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+
+  /* XXX: see comment at the top of the file. */
+  upb_fielddef_setname(r->f, name, NULL);
+  free(name);
+  return n;
+}
+
+static size_t field_ontypename(void *closure, const void *hd, const char *buf,
+                               size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  char *name = upb_strndup(buf, n);
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+
+  /* XXX: see comment at the top of the file. */
+  upb_fielddef_setsubdefname(r->f, name, NULL);
+  free(name);
+  return n;
+}
+
+static size_t field_onextendee(void *closure, const void *hd, const char *buf,
+                               size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  char *name = upb_strndup(buf, n);
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+
+  /* XXX: see comment at the top of the file. */
+  upb_fielddef_setcontainingtypename(r->f, name, NULL);
+  free(name);
+  return n;
+}
+
+static size_t field_ondefaultval(void *closure, const void *hd, const char *buf,
+                                 size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+
+  /* Have to convert from string to the correct type, but we might not know the
+   * type yet, so we save it as a string until the end of the field.
+   * XXX: see comment at the top of the file. */
+  free(r->default_string);
+  r->default_string = upb_strndup(buf, n);
+  return n;
+}
+
+/* Handlers for google.protobuf.DescriptorProto (representing a message). */
+static bool msg_startmsg(void *closure, const void *hd) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+
+  upb_deflist_push(&r->defs,
+                   upb_msgdef_upcast_mutable(upb_msgdef_new(&r->defs)));
+  upb_descreader_startcontainer(r);
+  return true;
+}
+
+static bool msg_endmsg(void *closure, const void *hd, upb_status *status) {
+  upb_descreader *r = closure;
+  upb_msgdef *m = upb_descreader_top(r);
+  UPB_UNUSED(hd);
+
+  if(!upb_def_fullname(upb_msgdef_upcast_mutable(m))) {
+    upb_status_seterrmsg(status, "Encountered message with no name.");
+    return false;
+  }
+  upb_descreader_endcontainer(r);
+  return true;
+}
+
+static size_t msg_onname(void *closure, const void *hd, const char *buf,
+                         size_t n, const upb_bufhandle *handle) {
+  upb_descreader *r = closure;
+  upb_msgdef *m = upb_descreader_top(r);
+  /* XXX: see comment at the top of the file. */
+  char *name = upb_strndup(buf, n);
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+
+  upb_def_setfullname(upb_msgdef_upcast_mutable(m), name, NULL);
+  upb_descreader_setscopename(r, name);  /* Passes ownership of name. */
+  return n;
+}
+
+static bool msg_onendfield(void *closure, const void *hd) {
+  upb_descreader *r = closure;
+  upb_msgdef *m = upb_descreader_top(r);
+  UPB_UNUSED(hd);
+
+  upb_msgdef_addfield(m, r->f, &r->defs, NULL);
+  r->f = NULL;
+  return true;
+}
+
+static bool pushextension(void *closure, const void *hd) {
+  upb_descreader *r = closure;
+  UPB_UNUSED(hd);
+
+  assert(upb_fielddef_containingtypename(r->f));
+  upb_fielddef_setisextension(r->f, true);
+  upb_deflist_push(&r->defs, upb_fielddef_upcast_mutable(r->f));
+  r->f = NULL;
+  return true;
+}
+
+#define D(name) upbdefs_google_protobuf_ ## name(s)
+
+static void reghandlers(const void *closure, upb_handlers *h) {
+  const upb_symtab *s = closure;
+  const upb_msgdef *m = upb_handlers_msgdef(h);
+
+  if (m == D(DescriptorProto)) {
+    upb_handlers_setstartmsg(h, &msg_startmsg, NULL);
+    upb_handlers_setendmsg(h, &msg_endmsg, NULL);
+    upb_handlers_setstring(h, D(DescriptorProto_name), &msg_onname, NULL);
+    upb_handlers_setendsubmsg(h, D(DescriptorProto_field), &msg_onendfield,
+                              NULL);
+    upb_handlers_setendsubmsg(h, D(DescriptorProto_extension), &pushextension,
+                              NULL);
+  } else if (m == D(FileDescriptorProto)) {
+    upb_handlers_setstartmsg(h, &file_startmsg, NULL);
+    upb_handlers_setendmsg(h, &file_endmsg, NULL);
+    upb_handlers_setstring(h, D(FileDescriptorProto_package), &file_onpackage,
+                           NULL);
+    upb_handlers_setendsubmsg(h, D(FileDescriptorProto_extension), &pushextension,
+                              NULL);
+  } else if (m == D(EnumValueDescriptorProto)) {
+    upb_handlers_setstartmsg(h, &enumval_startmsg, NULL);
+    upb_handlers_setendmsg(h, &enumval_endmsg, NULL);
+    upb_handlers_setstring(h, D(EnumValueDescriptorProto_name), &enumval_onname, NULL);
+    upb_handlers_setint32(h, D(EnumValueDescriptorProto_number), &enumval_onnumber,
+                          NULL);
+  } else if (m == D(EnumDescriptorProto)) {
+    upb_handlers_setstartmsg(h, &enum_startmsg, NULL);
+    upb_handlers_setendmsg(h, &enum_endmsg, NULL);
+    upb_handlers_setstring(h, D(EnumDescriptorProto_name), &enum_onname, NULL);
+  } else if (m == D(FieldDescriptorProto)) {
+    upb_handlers_setstartmsg(h, &field_startmsg, NULL);
+    upb_handlers_setendmsg(h, &field_endmsg, NULL);
+    upb_handlers_setint32(h, D(FieldDescriptorProto_type), &field_ontype,
+                          NULL);
+    upb_handlers_setint32(h, D(FieldDescriptorProto_label), &field_onlabel,
+                          NULL);
+    upb_handlers_setint32(h, D(FieldDescriptorProto_number), &field_onnumber,
+                          NULL);
+    upb_handlers_setstring(h, D(FieldDescriptorProto_name), &field_onname,
+                           NULL);
+    upb_handlers_setstring(h, D(FieldDescriptorProto_type_name),
+                           &field_ontypename, NULL);
+    upb_handlers_setstring(h, D(FieldDescriptorProto_extendee),
+                           &field_onextendee, NULL);
+    upb_handlers_setstring(h, D(FieldDescriptorProto_default_value),
+                           &field_ondefaultval, NULL);
+  } else if (m == D(FieldOptions)) {
+    upb_handlers_setbool(h, D(FieldOptions_lazy), &field_onlazy, NULL);
+    upb_handlers_setbool(h, D(FieldOptions_packed), &field_onpacked, NULL);
+  }
+}
+
+#undef D
+
+void descreader_cleanup(void *_r) {
+  upb_descreader *r = _r;
+  free(r->name);
+  upb_deflist_uninit(&r->defs);
+  free(r->default_string);
+  while (r->stack_len > 0) {
+    upb_descreader_frame *f = &r->stack[--r->stack_len];
+    free(f->name);
+  }
+}
+
+
+/* Public API  ****************************************************************/
+
+upb_descreader *upb_descreader_create(upb_env *e, const upb_handlers *h) {
+  upb_descreader *r = upb_env_malloc(e, sizeof(upb_descreader));
+  if (!r || !upb_env_addcleanup(e, descreader_cleanup, r)) {
+    return NULL;
+  }
+
+  upb_deflist_init(&r->defs);
+  upb_sink_reset(upb_descreader_input(r), h, r);
+  r->stack_len = 0;
+  r->name = NULL;
+  r->default_string = NULL;
+
+  return r;
+}
+
+upb_def **upb_descreader_getdefs(upb_descreader *r, void *owner, int *n) {
+  *n = r->defs.len;
+  upb_deflist_donaterefs(&r->defs, owner);
+  return r->defs.defs;
+}
+
+upb_sink *upb_descreader_input(upb_descreader *r) {
+  return &r->sink;
+}
+
+const upb_handlers *upb_descreader_newhandlers(const void *owner) {
+  const upb_symtab *s = upbdefs_google_protobuf_descriptor(&s);
+  const upb_handlers *h = upb_handlers_newfrozen(
+      upbdefs_google_protobuf_FileDescriptorSet(s), owner, reghandlers, s);
+  upb_symtab_unref(s, &s);
+  return h;
+}
+/*
+** protobuf decoder bytecode compiler
+**
+** Code to compile a upb::Handlers into bytecode for decoding a protobuf
+** according to that specific schema and destination handlers.
+**
+** Compiling to bytecode is always the first step.  If we are using the
+** interpreted decoder we leave it as bytecode and interpret that.  If we are
+** using a JIT decoder we use a code generator to turn the bytecode into native
+** code, LLVM IR, etc.
+**
+** Bytecode definition is in decoder.int.h.
+*/
+
+#include <stdarg.h>
+
+#ifdef UPB_DUMP_BYTECODE
+#include <stdio.h>
+#endif
+
+#define MAXLABEL 5
+#define EMPTYLABEL -1
+
+/* mgroup *********************************************************************/
+
+static void freegroup(upb_refcounted *r) {
+  mgroup *g = (mgroup*)r;
+  upb_inttable_uninit(&g->methods);
+#ifdef UPB_USE_JIT_X64
+  upb_pbdecoder_freejit(g);
+#endif
+  free(g->bytecode);
+  free(g);
+}
+
+static void visitgroup(const upb_refcounted *r, upb_refcounted_visit *visit,
+                       void *closure) {
+  const mgroup *g = (const mgroup*)r;
+  upb_inttable_iter i;
+  upb_inttable_begin(&i, &g->methods);
+  for(; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    upb_pbdecodermethod *method = upb_value_getptr(upb_inttable_iter_value(&i));
+    visit(r, upb_pbdecodermethod_upcast(method), closure);
+  }
+}
+
+mgroup *newgroup(const void *owner) {
+  mgroup *g = malloc(sizeof(*g));
+  static const struct upb_refcounted_vtbl vtbl = {visitgroup, freegroup};
+  upb_refcounted_init(mgroup_upcast_mutable(g), &vtbl, owner);
+  upb_inttable_init(&g->methods, UPB_CTYPE_PTR);
+  g->bytecode = NULL;
+  g->bytecode_end = NULL;
+  return g;
+}
+
+
+/* upb_pbdecodermethod ********************************************************/
+
+static void freemethod(upb_refcounted *r) {
+  upb_pbdecodermethod *method = (upb_pbdecodermethod*)r;
+
+  if (method->dest_handlers_) {
+    upb_handlers_unref(method->dest_handlers_, method);
+  }
+
+  upb_inttable_uninit(&method->dispatch);
+  free(method);
+}
+
+static void visitmethod(const upb_refcounted *r, upb_refcounted_visit *visit,
+                        void *closure) {
+  const upb_pbdecodermethod *m = (const upb_pbdecodermethod*)r;
+  visit(r, m->group, closure);
+}
+
+static upb_pbdecodermethod *newmethod(const upb_handlers *dest_handlers,
+                                      mgroup *group) {
+  static const struct upb_refcounted_vtbl vtbl = {visitmethod, freemethod};
+  upb_pbdecodermethod *ret = malloc(sizeof(*ret));
+  upb_refcounted_init(upb_pbdecodermethod_upcast_mutable(ret), &vtbl, &ret);
+  upb_byteshandler_init(&ret->input_handler_);
+
+  /* The method references the group and vice-versa, in a circular reference. */
+  upb_ref2(ret, group);
+  upb_ref2(group, ret);
+  upb_inttable_insertptr(&group->methods, dest_handlers, upb_value_ptr(ret));
+  upb_pbdecodermethod_unref(ret, &ret);
+
+  ret->group = mgroup_upcast_mutable(group);
+  ret->dest_handlers_ = dest_handlers;
+  ret->is_native_ = false;  /* If we JIT, it will update this later. */
+  upb_inttable_init(&ret->dispatch, UPB_CTYPE_UINT64);
+
+  if (ret->dest_handlers_) {
+    upb_handlers_ref(ret->dest_handlers_, ret);
+  }
+  return ret;
+}
+
+const upb_handlers *upb_pbdecodermethod_desthandlers(
+    const upb_pbdecodermethod *m) {
+  return m->dest_handlers_;
+}
+
+const upb_byteshandler *upb_pbdecodermethod_inputhandler(
+    const upb_pbdecodermethod *m) {
+  return &m->input_handler_;
+}
+
+bool upb_pbdecodermethod_isnative(const upb_pbdecodermethod *m) {
+  return m->is_native_;
+}
+
+const upb_pbdecodermethod *upb_pbdecodermethod_new(
+    const upb_pbdecodermethodopts *opts, const void *owner) {
+  const upb_pbdecodermethod *ret;
+  upb_pbcodecache cache;
+
+  upb_pbcodecache_init(&cache);
+  ret = upb_pbcodecache_getdecodermethod(&cache, opts);
+  upb_pbdecodermethod_ref(ret, owner);
+  upb_pbcodecache_uninit(&cache);
+  return ret;
+}
+
+
+/* bytecode compiler **********************************************************/
+
+/* Data used only at compilation time. */
+typedef struct {
+  mgroup *group;
+
+  uint32_t *pc;
+  int fwd_labels[MAXLABEL];
+  int back_labels[MAXLABEL];
+
+  /* For fields marked "lazy", parse them lazily or eagerly? */
+  bool lazy;
+} compiler;
+
+static compiler *newcompiler(mgroup *group, bool lazy) {
+  compiler *ret = malloc(sizeof(*ret));
+  int i;
+
+  ret->group = group;
+  ret->lazy = lazy;
+  for (i = 0; i < MAXLABEL; i++) {
+    ret->fwd_labels[i] = EMPTYLABEL;
+    ret->back_labels[i] = EMPTYLABEL;
+  }
+  return ret;
+}
+
+static void freecompiler(compiler *c) {
+  free(c);
+}
+
+const size_t ptr_words = sizeof(void*) / sizeof(uint32_t);
+
+/* How many words an instruction is. */
+static int instruction_len(uint32_t instr) {
+  switch (getop(instr)) {
+    case OP_SETDISPATCH: return 1 + ptr_words;
+    case OP_TAGN: return 3;
+    case OP_SETBIGGROUPNUM: return 2;
+    default: return 1;
+  }
+}
+
+bool op_has_longofs(int32_t instruction) {
+  switch (getop(instruction)) {
+    case OP_CALL:
+    case OP_BRANCH:
+    case OP_CHECKDELIM:
+      return true;
+    /* The "tag" instructions only have 8 bytes available for the jump target,
+     * but that is ok because these opcodes only require short jumps. */
+    case OP_TAG1:
+    case OP_TAG2:
+    case OP_TAGN:
+      return false;
+    default:
+      assert(false);
+      return false;
+  }
+}
+
+static int32_t getofs(uint32_t instruction) {
+  if (op_has_longofs(instruction)) {
+    return (int32_t)instruction >> 8;
+  } else {
+    return (int8_t)(instruction >> 8);
+  }
+}
+
+static void setofs(uint32_t *instruction, int32_t ofs) {
+  if (op_has_longofs(*instruction)) {
+    *instruction = getop(*instruction) | ofs << 8;
+  } else {
+    *instruction = (*instruction & ~0xff00) | ((ofs & 0xff) << 8);
+  }
+  assert(getofs(*instruction) == ofs);  /* Would fail in cases of overflow. */
+}
+
+static uint32_t pcofs(compiler *c) { return c->pc - c->group->bytecode; }
+
+/* Defines a local label at the current PC location.  All previous forward
+ * references are updated to point to this location.  The location is noted
+ * for any future backward references. */
+static void label(compiler *c, unsigned int label) {
+  int val;
+  uint32_t *codep;
+
+  assert(label < MAXLABEL);
+  val = c->fwd_labels[label];
+  codep = (val == EMPTYLABEL) ? NULL : c->group->bytecode + val;
+  while (codep) {
+    int ofs = getofs(*codep);
+    setofs(codep, c->pc - codep - instruction_len(*codep));
+    codep = ofs ? codep + ofs : NULL;
+  }
+  c->fwd_labels[label] = EMPTYLABEL;
+  c->back_labels[label] = pcofs(c);
+}
+
+/* Creates a reference to a numbered label; either a forward reference
+ * (positive arg) or backward reference (negative arg).  For forward references
+ * the value returned now is actually a "next" pointer into a linked list of all
+ * instructions that use this label and will be patched later when the label is
+ * defined with label().
+ *
+ * The returned value is the offset that should be written into the instruction.
+ */
+static int32_t labelref(compiler *c, int label) {
+  assert(label < MAXLABEL);
+  if (label == LABEL_DISPATCH) {
+    /* No resolving required. */
+    return 0;
+  } else if (label < 0) {
+    /* Backward local label.  Relative to the next instruction. */
+    uint32_t from = (c->pc + 1) - c->group->bytecode;
+    return c->back_labels[-label] - from;
+  } else {
+    /* Forward local label: prepend to (possibly-empty) linked list. */
+    int *lptr = &c->fwd_labels[label];
+    int32_t ret = (*lptr == EMPTYLABEL) ? 0 : *lptr - pcofs(c);
+    *lptr = pcofs(c);
+    return ret;
+  }
+}
+
+static void put32(compiler *c, uint32_t v) {
+  mgroup *g = c->group;
+  if (c->pc == g->bytecode_end) {
+    int ofs = pcofs(c);
+    size_t oldsize = g->bytecode_end - g->bytecode;
+    size_t newsize = UPB_MAX(oldsize * 2, 64);
+    /* TODO(haberman): handle OOM. */
+    g->bytecode = realloc(g->bytecode, newsize * sizeof(uint32_t));
+    g->bytecode_end = g->bytecode + newsize;
+    c->pc = g->bytecode + ofs;
+  }
+  *c->pc++ = v;
+}
+
+static void putop(compiler *c, opcode op, ...) {
+  va_list ap;
+  va_start(ap, op);
+
+  switch (op) {
+    case OP_SETDISPATCH: {
+      uintptr_t ptr = (uintptr_t)va_arg(ap, void*);
+      put32(c, OP_SETDISPATCH);
+      put32(c, ptr);
+      if (sizeof(uintptr_t) > sizeof(uint32_t))
+        put32(c, (uint64_t)ptr >> 32);
+      break;
+    }
+    case OP_STARTMSG:
+    case OP_ENDMSG:
+    case OP_PUSHLENDELIM:
+    case OP_POP:
+    case OP_SETDELIM:
+    case OP_HALT:
+    case OP_RET:
+    case OP_DISPATCH:
+      put32(c, op);
+      break;
+    case OP_PARSE_DOUBLE:
+    case OP_PARSE_FLOAT:
+    case OP_PARSE_INT64:
+    case OP_PARSE_UINT64:
+    case OP_PARSE_INT32:
+    case OP_PARSE_FIXED64:
+    case OP_PARSE_FIXED32:
+    case OP_PARSE_BOOL:
+    case OP_PARSE_UINT32:
+    case OP_PARSE_SFIXED32:
+    case OP_PARSE_SFIXED64:
+    case OP_PARSE_SINT32:
+    case OP_PARSE_SINT64:
+    case OP_STARTSEQ:
+    case OP_ENDSEQ:
+    case OP_STARTSUBMSG:
+    case OP_ENDSUBMSG:
+    case OP_STARTSTR:
+    case OP_STRING:
+    case OP_ENDSTR:
+    case OP_PUSHTAGDELIM:
+      put32(c, op | va_arg(ap, upb_selector_t) << 8);
+      break;
+    case OP_SETBIGGROUPNUM:
+      put32(c, op);
+      put32(c, va_arg(ap, int));
+      break;
+    case OP_CALL: {
+      const upb_pbdecodermethod *method = va_arg(ap, upb_pbdecodermethod *);
+      put32(c, op | (method->code_base.ofs - (pcofs(c) + 1)) << 8);
+      break;
+    }
+    case OP_CHECKDELIM:
+    case OP_BRANCH: {
+      uint32_t instruction = op;
+      int label = va_arg(ap, int);
+      setofs(&instruction, labelref(c, label));
+      put32(c, instruction);
+      break;
+    }
+    case OP_TAG1:
+    case OP_TAG2: {
+      int label = va_arg(ap, int);
+      uint64_t tag = va_arg(ap, uint64_t);
+      uint32_t instruction = op | (tag << 16);
+      assert(tag <= 0xffff);
+      setofs(&instruction, labelref(c, label));
+      put32(c, instruction);
+      break;
+    }
+    case OP_TAGN: {
+      int label = va_arg(ap, int);
+      uint64_t tag = va_arg(ap, uint64_t);
+      uint32_t instruction = op | (upb_value_size(tag) << 16);
+      setofs(&instruction, labelref(c, label));
+      put32(c, instruction);
+      put32(c, tag);
+      put32(c, tag >> 32);
+      break;
+    }
+  }
+
+  va_end(ap);
+}
+
+#if defined(UPB_USE_JIT_X64) || defined(UPB_DUMP_BYTECODE)
+
+const char *upb_pbdecoder_getopname(unsigned int op) {
+#define QUOTE(x) #x
+#define EXPAND_AND_QUOTE(x) QUOTE(x)
+#define OPNAME(x) OP_##x
+#define OP(x) case OPNAME(x): return EXPAND_AND_QUOTE(OPNAME(x));
+#define T(x) OP(PARSE_##x)
+  /* Keep in sync with list in decoder.int.h. */
+  switch ((opcode)op) {
+    T(DOUBLE) T(FLOAT) T(INT64) T(UINT64) T(INT32) T(FIXED64) T(FIXED32)
+    T(BOOL) T(UINT32) T(SFIXED32) T(SFIXED64) T(SINT32) T(SINT64)
+    OP(STARTMSG) OP(ENDMSG) OP(STARTSEQ) OP(ENDSEQ) OP(STARTSUBMSG)
+    OP(ENDSUBMSG) OP(STARTSTR) OP(STRING) OP(ENDSTR) OP(CALL) OP(RET)
+    OP(PUSHLENDELIM) OP(PUSHTAGDELIM) OP(SETDELIM) OP(CHECKDELIM)
+    OP(BRANCH) OP(TAG1) OP(TAG2) OP(TAGN) OP(SETDISPATCH) OP(POP)
+    OP(SETBIGGROUPNUM) OP(DISPATCH) OP(HALT)
+  }
+  return "<unknown op>";
+#undef OP
+#undef T
+}
+
+#endif
+
+#ifdef UPB_DUMP_BYTECODE
+
+static void dumpbc(uint32_t *p, uint32_t *end, FILE *f) {
+
+  uint32_t *begin = p;
+
+  while (p < end) {
+    fprintf(f, "%p  %8tx", p, p - begin);
+    uint32_t instr = *p++;
+    uint8_t op = getop(instr);
+    fprintf(f, " %s", upb_pbdecoder_getopname(op));
+    switch ((opcode)op) {
+      case OP_SETDISPATCH: {
+        const upb_inttable *dispatch;
+        memcpy(&dispatch, p, sizeof(void*));
+        p += ptr_words;
+        const upb_pbdecodermethod *method =
+            (void *)((char *)dispatch -
+                     offsetof(upb_pbdecodermethod, dispatch));
+        fprintf(f, " %s", upb_msgdef_fullname(
+                              upb_handlers_msgdef(method->dest_handlers_)));
+        break;
+      }
+      case OP_DISPATCH:
+      case OP_STARTMSG:
+      case OP_ENDMSG:
+      case OP_PUSHLENDELIM:
+      case OP_POP:
+      case OP_SETDELIM:
+      case OP_HALT:
+      case OP_RET:
+        break;
+      case OP_PARSE_DOUBLE:
+      case OP_PARSE_FLOAT:
+      case OP_PARSE_INT64:
+      case OP_PARSE_UINT64:
+      case OP_PARSE_INT32:
+      case OP_PARSE_FIXED64:
+      case OP_PARSE_FIXED32:
+      case OP_PARSE_BOOL:
+      case OP_PARSE_UINT32:
+      case OP_PARSE_SFIXED32:
+      case OP_PARSE_SFIXED64:
+      case OP_PARSE_SINT32:
+      case OP_PARSE_SINT64:
+      case OP_STARTSEQ:
+      case OP_ENDSEQ:
+      case OP_STARTSUBMSG:
+      case OP_ENDSUBMSG:
+      case OP_STARTSTR:
+      case OP_STRING:
+      case OP_ENDSTR:
+      case OP_PUSHTAGDELIM:
+        fprintf(f, " %d", instr >> 8);
+        break;
+      case OP_SETBIGGROUPNUM:
+        fprintf(f, " %d", *p++);
+        break;
+      case OP_CHECKDELIM:
+      case OP_CALL:
+      case OP_BRANCH:
+        fprintf(f, " =>0x%tx", p + getofs(instr) - begin);
+        break;
+      case OP_TAG1:
+      case OP_TAG2: {
+        fprintf(f, " tag:0x%x", instr >> 16);
+        if (getofs(instr)) {
+          fprintf(f, " =>0x%tx", p + getofs(instr) - begin);
+        }
+        break;
+      }
+      case OP_TAGN: {
+        uint64_t tag = *p++;
+        tag |= (uint64_t)*p++ << 32;
+        fprintf(f, " tag:0x%llx", (long long)tag);
+        fprintf(f, " n:%d", instr >> 16);
+        if (getofs(instr)) {
+          fprintf(f, " =>0x%tx", p + getofs(instr) - begin);
+        }
+        break;
+      }
+    }
+    fputs("\n", f);
+  }
+}
+
+#endif
+
+static uint64_t get_encoded_tag(const upb_fielddef *f, int wire_type) {
+  uint32_t tag = (upb_fielddef_number(f) << 3) | wire_type;
+  uint64_t encoded_tag = upb_vencode32(tag);
+  /* No tag should be greater than 5 bytes. */
+  assert(encoded_tag <= 0xffffffffff);
+  return encoded_tag;
+}
+
+static void putchecktag(compiler *c, const upb_fielddef *f,
+                        int wire_type, int dest) {
+  uint64_t tag = get_encoded_tag(f, wire_type);
+  switch (upb_value_size(tag)) {
+    case 1:
+      putop(c, OP_TAG1, dest, tag);
+      break;
+    case 2:
+      putop(c, OP_TAG2, dest, tag);
+      break;
+    default:
+      putop(c, OP_TAGN, dest, tag);
+      break;
+  }
+}
+
+static upb_selector_t getsel(const upb_fielddef *f, upb_handlertype_t type) {
+  upb_selector_t selector;
+  bool ok = upb_handlers_getselector(f, type, &selector);
+  UPB_ASSERT_VAR(ok, ok);
+  return selector;
+}
+
+/* Takes an existing, primary dispatch table entry and repacks it with a
+ * different alternate wire type.  Called when we are inserting a secondary
+ * dispatch table entry for an alternate wire type. */
+static uint64_t repack(uint64_t dispatch, int new_wt2) {
+  uint64_t ofs;
+  uint8_t wt1;
+  uint8_t old_wt2;
+  upb_pbdecoder_unpackdispatch(dispatch, &ofs, &wt1, &old_wt2);
+  assert(old_wt2 == NO_WIRE_TYPE);  /* wt2 should not be set yet. */
+  return upb_pbdecoder_packdispatch(ofs, wt1, new_wt2);
+}
+
+/* Marks the current bytecode position as the dispatch target for this message,
+ * field, and wire type. */
+static void dispatchtarget(compiler *c, upb_pbdecodermethod *method,
+                           const upb_fielddef *f, int wire_type) {
+  /* Offset is relative to msg base. */
+  uint64_t ofs = pcofs(c) - method->code_base.ofs;
+  uint32_t fn = upb_fielddef_number(f);
+  upb_inttable *d = &method->dispatch;
+  upb_value v;
+  if (upb_inttable_remove(d, fn, &v)) {
+    /* TODO: prioritize based on packed setting in .proto file. */
+    uint64_t repacked = repack(upb_value_getuint64(v), wire_type);
+    upb_inttable_insert(d, fn, upb_value_uint64(repacked));
+    upb_inttable_insert(d, fn + UPB_MAX_FIELDNUMBER, upb_value_uint64(ofs));
+  } else {
+    uint64_t val = upb_pbdecoder_packdispatch(ofs, wire_type, NO_WIRE_TYPE);
+    upb_inttable_insert(d, fn, upb_value_uint64(val));
+  }
+}
+
+static void putpush(compiler *c, const upb_fielddef *f) {
+  if (upb_fielddef_descriptortype(f) == UPB_DESCRIPTOR_TYPE_MESSAGE) {
+    putop(c, OP_PUSHLENDELIM);
+  } else {
+    uint32_t fn = upb_fielddef_number(f);
+    if (fn >= 1 << 24) {
+      putop(c, OP_PUSHTAGDELIM, 0);
+      putop(c, OP_SETBIGGROUPNUM, fn);
+    } else {
+      putop(c, OP_PUSHTAGDELIM, fn);
+    }
+  }
+}
+
+static upb_pbdecodermethod *find_submethod(const compiler *c,
+                                           const upb_pbdecodermethod *method,
+                                           const upb_fielddef *f) {
+  const upb_handlers *sub =
+      upb_handlers_getsubhandlers(method->dest_handlers_, f);
+  upb_value v;
+  return upb_inttable_lookupptr(&c->group->methods, sub, &v)
+             ? upb_value_getptr(v)
+             : NULL;
+}
+
+static void putsel(compiler *c, opcode op, upb_selector_t sel,
+                   const upb_handlers *h) {
+  if (upb_handlers_gethandler(h, sel)) {
+    putop(c, op, sel);
+  }
+}
+
+/* Puts an opcode to call a callback, but only if a callback actually exists for
+ * this field and handler type. */
+static void maybeput(compiler *c, opcode op, const upb_handlers *h,
+                     const upb_fielddef *f, upb_handlertype_t type) {
+  putsel(c, op, getsel(f, type), h);
+}
+
+static bool haslazyhandlers(const upb_handlers *h, const upb_fielddef *f) {
+  if (!upb_fielddef_lazy(f))
+    return false;
+
+  return upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_STARTSTR)) ||
+         upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_STRING)) ||
+         upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_ENDSTR));
+}
+
+
+/* bytecode compiler code generation ******************************************/
+
+/* Symbolic names for our local labels. */
+#define LABEL_LOOPSTART 1  /* Top of a repeated field loop. */
+#define LABEL_LOOPBREAK 2  /* To jump out of a repeated loop */
+#define LABEL_FIELD     3  /* Jump backward to find the most recent field. */
+#define LABEL_ENDMSG    4  /* To reach the OP_ENDMSG instr for this msg. */
+
+/* Generates bytecode to parse a single non-lazy message field. */
+static void generate_msgfield(compiler *c, const upb_fielddef *f,
+                              upb_pbdecodermethod *method) {
+  const upb_handlers *h = upb_pbdecodermethod_desthandlers(method);
+  const upb_pbdecodermethod *sub_m = find_submethod(c, method, f);
+  int wire_type;
+
+  if (!sub_m) {
+    /* Don't emit any code for this field at all; it will be parsed as an
+     * unknown field. */
+    return;
+  }
+
+  label(c, LABEL_FIELD);
+
+  wire_type =
+      (upb_fielddef_descriptortype(f) == UPB_DESCRIPTOR_TYPE_MESSAGE)
+          ? UPB_WIRE_TYPE_DELIMITED
+          : UPB_WIRE_TYPE_START_GROUP;
+
+  if (upb_fielddef_isseq(f)) {
+    putop(c, OP_CHECKDELIM, LABEL_ENDMSG);
+    putchecktag(c, f, wire_type, LABEL_DISPATCH);
+   dispatchtarget(c, method, f, wire_type);
+    putop(c, OP_PUSHTAGDELIM, 0);
+    putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ));
+   label(c, LABEL_LOOPSTART);
+    putpush(c, f);
+    putop(c, OP_STARTSUBMSG, getsel(f, UPB_HANDLER_STARTSUBMSG));
+    putop(c, OP_CALL, sub_m);
+    putop(c, OP_POP);
+    maybeput(c, OP_ENDSUBMSG, h, f, UPB_HANDLER_ENDSUBMSG);
+    if (wire_type == UPB_WIRE_TYPE_DELIMITED) {
+      putop(c, OP_SETDELIM);
+    }
+    putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK);
+    putchecktag(c, f, wire_type, LABEL_LOOPBREAK);
+    putop(c, OP_BRANCH, -LABEL_LOOPSTART);
+   label(c, LABEL_LOOPBREAK);
+    putop(c, OP_POP);
+    maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ);
+  } else {
+    putop(c, OP_CHECKDELIM, LABEL_ENDMSG);
+    putchecktag(c, f, wire_type, LABEL_DISPATCH);
+   dispatchtarget(c, method, f, wire_type);
+    putpush(c, f);
+    putop(c, OP_STARTSUBMSG, getsel(f, UPB_HANDLER_STARTSUBMSG));
+    putop(c, OP_CALL, sub_m);
+    putop(c, OP_POP);
+    maybeput(c, OP_ENDSUBMSG, h, f, UPB_HANDLER_ENDSUBMSG);
+    if (wire_type == UPB_WIRE_TYPE_DELIMITED) {
+      putop(c, OP_SETDELIM);
+    }
+  }
+}
+
+/* Generates bytecode to parse a single string or lazy submessage field. */
+static void generate_delimfield(compiler *c, const upb_fielddef *f,
+                                upb_pbdecodermethod *method) {
+  const upb_handlers *h = upb_pbdecodermethod_desthandlers(method);
+
+  label(c, LABEL_FIELD);
+  if (upb_fielddef_isseq(f)) {
+    putop(c, OP_CHECKDELIM, LABEL_ENDMSG);
+    putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH);
+   dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED);
+    putop(c, OP_PUSHTAGDELIM, 0);
+    putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ));
+   label(c, LABEL_LOOPSTART);
+    putop(c, OP_PUSHLENDELIM);
+    putop(c, OP_STARTSTR, getsel(f, UPB_HANDLER_STARTSTR));
+    /* Need to emit even if no handler to skip past the string. */
+    putop(c, OP_STRING, getsel(f, UPB_HANDLER_STRING));
+    putop(c, OP_POP);
+    maybeput(c, OP_ENDSTR, h, f, UPB_HANDLER_ENDSTR);
+    putop(c, OP_SETDELIM);
+    putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK);
+    putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_LOOPBREAK);
+    putop(c, OP_BRANCH, -LABEL_LOOPSTART);
+   label(c, LABEL_LOOPBREAK);
+    putop(c, OP_POP);
+    maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ);
+  } else {
+    putop(c, OP_CHECKDELIM, LABEL_ENDMSG);
+    putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH);
+   dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED);
+    putop(c, OP_PUSHLENDELIM);
+    putop(c, OP_STARTSTR, getsel(f, UPB_HANDLER_STARTSTR));
+    putop(c, OP_STRING, getsel(f, UPB_HANDLER_STRING));
+    putop(c, OP_POP);
+    maybeput(c, OP_ENDSTR, h, f, UPB_HANDLER_ENDSTR);
+    putop(c, OP_SETDELIM);
+  }
+}
+
+/* Generates bytecode to parse a single primitive field. */
+static void generate_primitivefield(compiler *c, const upb_fielddef *f,
+                                    upb_pbdecodermethod *method) {
+  const upb_handlers *h = upb_pbdecodermethod_desthandlers(method);
+  upb_descriptortype_t descriptor_type = upb_fielddef_descriptortype(f);
+  opcode parse_type;
+  upb_selector_t sel;
+  int wire_type;
+
+  label(c, LABEL_FIELD);
+
+  /* From a decoding perspective, ENUM is the same as INT32. */
+  if (descriptor_type == UPB_DESCRIPTOR_TYPE_ENUM)
+    descriptor_type = UPB_DESCRIPTOR_TYPE_INT32;
+
+  parse_type = (opcode)descriptor_type;
+
+  /* TODO(haberman): generate packed or non-packed first depending on "packed"
+   * setting in the fielddef.  This will favor (in speed) whichever was
+   * specified. */
+
+  assert((int)parse_type >= 0 && parse_type <= OP_MAX);
+  sel = getsel(f, upb_handlers_getprimitivehandlertype(f));
+  wire_type = upb_pb_native_wire_types[upb_fielddef_descriptortype(f)];
+  if (upb_fielddef_isseq(f)) {
+    putop(c, OP_CHECKDELIM, LABEL_ENDMSG);
+    putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH);
+   dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED);
+    putop(c, OP_PUSHLENDELIM);
+    putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ));  /* Packed */
+   label(c, LABEL_LOOPSTART);
+    putop(c, parse_type, sel);
+    putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK);
+    putop(c, OP_BRANCH, -LABEL_LOOPSTART);
+   dispatchtarget(c, method, f, wire_type);
+    putop(c, OP_PUSHTAGDELIM, 0);
+    putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ));  /* Non-packed */
+   label(c, LABEL_LOOPSTART);
+    putop(c, parse_type, sel);
+    putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK);
+    putchecktag(c, f, wire_type, LABEL_LOOPBREAK);
+    putop(c, OP_BRANCH, -LABEL_LOOPSTART);
+   label(c, LABEL_LOOPBREAK);
+    putop(c, OP_POP);  /* Packed and non-packed join. */
+    maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ);
+    putop(c, OP_SETDELIM);  /* Could remove for non-packed by dup ENDSEQ. */
+  } else {
+    putop(c, OP_CHECKDELIM, LABEL_ENDMSG);
+    putchecktag(c, f, wire_type, LABEL_DISPATCH);
+   dispatchtarget(c, method, f, wire_type);
+    putop(c, parse_type, sel);
+  }
+}
+
+/* Adds bytecode for parsing the given message to the given decoderplan,
+ * while adding all dispatch targets to this message's dispatch table. */
+static void compile_method(compiler *c, upb_pbdecodermethod *method) {
+  const upb_handlers *h;
+  const upb_msgdef *md;
+  uint32_t* start_pc;
+  upb_msg_field_iter i;
+  upb_value val;
+
+  assert(method);
+
+  /* Clear all entries in the dispatch table. */
+  upb_inttable_uninit(&method->dispatch);
+  upb_inttable_init(&method->dispatch, UPB_CTYPE_UINT64);
+
+  h = upb_pbdecodermethod_desthandlers(method);
+  md = upb_handlers_msgdef(h);
+
+ method->code_base.ofs = pcofs(c);
+  putop(c, OP_SETDISPATCH, &method->dispatch);
+  putsel(c, OP_STARTMSG, UPB_STARTMSG_SELECTOR, h);
+ label(c, LABEL_FIELD);
+  start_pc = c->pc;
+  for(upb_msg_field_begin(&i, md);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    const upb_fielddef *f = upb_msg_iter_field(&i);
+    upb_fieldtype_t type = upb_fielddef_type(f);
+
+    if (type == UPB_TYPE_MESSAGE && !(haslazyhandlers(h, f) && c->lazy)) {
+      generate_msgfield(c, f, method);
+    } else if (type == UPB_TYPE_STRING || type == UPB_TYPE_BYTES ||
+               type == UPB_TYPE_MESSAGE) {
+      generate_delimfield(c, f, method);
+    } else {
+      generate_primitivefield(c, f, method);
+    }
+  }
+
+  /* If there were no fields, or if no handlers were defined, we need to
+   * generate a non-empty loop body so that we can at least dispatch for unknown
+   * fields and check for the end of the message. */
+  if (c->pc == start_pc) {
+    /* Check for end-of-message. */
+    putop(c, OP_CHECKDELIM, LABEL_ENDMSG);
+    /* Unconditionally dispatch. */
+    putop(c, OP_DISPATCH, 0);
+  }
+
+  /* For now we just loop back to the last field of the message (or if none,
+   * the DISPATCH opcode for the message). */
+  putop(c, OP_BRANCH, -LABEL_FIELD);
+
+  /* Insert both a label and a dispatch table entry for this end-of-msg. */
+ label(c, LABEL_ENDMSG);
+  val = upb_value_uint64(pcofs(c) - method->code_base.ofs);
+  upb_inttable_insert(&method->dispatch, DISPATCH_ENDMSG, val);
+
+  putsel(c, OP_ENDMSG, UPB_ENDMSG_SELECTOR, h);
+  putop(c, OP_RET);
+
+  upb_inttable_compact(&method->dispatch);
+}
+
+/* Populate "methods" with new upb_pbdecodermethod objects reachable from "h".
+ * Returns the method for these handlers.
+ *
+ * Generates a new method for every destination handlers reachable from "h". */
+static void find_methods(compiler *c, const upb_handlers *h) {
+  upb_value v;
+  upb_msg_field_iter i;
+  const upb_msgdef *md;
+
+  if (upb_inttable_lookupptr(&c->group->methods, h, &v))
+    return;
+  newmethod(h, c->group);
+
+  /* Find submethods. */
+  md = upb_handlers_msgdef(h);
+  for(upb_msg_field_begin(&i, md);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    const upb_fielddef *f = upb_msg_iter_field(&i);
+    const upb_handlers *sub_h;
+    if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE &&
+        (sub_h = upb_handlers_getsubhandlers(h, f)) != NULL) {
+      /* We only generate a decoder method for submessages with handlers.
+       * Others will be parsed as unknown fields. */
+      find_methods(c, sub_h);
+    }
+  }
+}
+
+/* (Re-)compile bytecode for all messages in "msgs."
+ * Overwrites any existing bytecode in "c". */
+static void compile_methods(compiler *c) {
+  upb_inttable_iter i;
+
+  /* Start over at the beginning of the bytecode. */
+  c->pc = c->group->bytecode;
+
+  upb_inttable_begin(&i, &c->group->methods);
+  for(; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    upb_pbdecodermethod *method = upb_value_getptr(upb_inttable_iter_value(&i));
+    compile_method(c, method);
+  }
+}
+
+static void set_bytecode_handlers(mgroup *g) {
+  upb_inttable_iter i;
+  upb_inttable_begin(&i, &g->methods);
+  for(; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    upb_pbdecodermethod *m = upb_value_getptr(upb_inttable_iter_value(&i));
+    upb_byteshandler *h = &m->input_handler_;
+
+    m->code_base.ptr = g->bytecode + m->code_base.ofs;
+
+    upb_byteshandler_setstartstr(h, upb_pbdecoder_startbc, m->code_base.ptr);
+    upb_byteshandler_setstring(h, upb_pbdecoder_decode, g);
+    upb_byteshandler_setendstr(h, upb_pbdecoder_end, m);
+  }
+}
+
+
+/* JIT setup. *****************************************************************/
+
+#ifdef UPB_USE_JIT_X64
+
+static void sethandlers(mgroup *g, bool allowjit) {
+  g->jit_code = NULL;
+  if (allowjit) {
+    /* Compile byte-code into machine code, create handlers. */
+    upb_pbdecoder_jit(g);
+  } else {
+    set_bytecode_handlers(g);
+  }
+}
+
+#else  /* UPB_USE_JIT_X64 */
+
+static void sethandlers(mgroup *g, bool allowjit) {
+  /* No JIT compiled in; use bytecode handlers unconditionally. */
+  UPB_UNUSED(allowjit);
+  set_bytecode_handlers(g);
+}
+
+#endif  /* UPB_USE_JIT_X64 */
+
+
+/* TODO(haberman): allow this to be constructed for an arbitrary set of dest
+ * handlers and other mgroups (but verify we have a transitive closure). */
+const mgroup *mgroup_new(const upb_handlers *dest, bool allowjit, bool lazy,
+                         const void *owner) {
+  mgroup *g;
+  compiler *c;
+
+  UPB_UNUSED(allowjit);
+  assert(upb_handlers_isfrozen(dest));
+
+  g = newgroup(owner);
+  c = newcompiler(g, lazy);
+  find_methods(c, dest);
+
+  /* We compile in two passes:
+   * 1. all messages are assigned relative offsets from the beginning of the
+   *    bytecode (saved in method->code_base).
+   * 2. forwards OP_CALL instructions can be correctly linked since message
+   *    offsets have been previously assigned.
+   *
+   * Could avoid the second pass by linking OP_CALL instructions somehow. */
+  compile_methods(c);
+  compile_methods(c);
+  g->bytecode_end = c->pc;
+  freecompiler(c);
+
+#ifdef UPB_DUMP_BYTECODE
+  {
+    FILE *f = fopen("/tmp/upb-bytecode", "wb");
+    assert(f);
+    dumpbc(g->bytecode, g->bytecode_end, stderr);
+    dumpbc(g->bytecode, g->bytecode_end, f);
+    fclose(f);
+  }
+#endif
+
+  sethandlers(g, allowjit);
+  return g;
+}
+
+
+/* upb_pbcodecache ************************************************************/
+
+void upb_pbcodecache_init(upb_pbcodecache *c) {
+  upb_inttable_init(&c->groups, UPB_CTYPE_CONSTPTR);
+  c->allow_jit_ = true;
+}
+
+void upb_pbcodecache_uninit(upb_pbcodecache *c) {
+  upb_inttable_iter i;
+  upb_inttable_begin(&i, &c->groups);
+  for(; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+    const mgroup *group = upb_value_getconstptr(upb_inttable_iter_value(&i));
+    mgroup_unref(group, c);
+  }
+  upb_inttable_uninit(&c->groups);
+}
+
+bool upb_pbcodecache_allowjit(const upb_pbcodecache *c) {
+  return c->allow_jit_;
+}
+
+bool upb_pbcodecache_setallowjit(upb_pbcodecache *c, bool allow) {
+  if (upb_inttable_count(&c->groups) > 0)
+    return false;
+  c->allow_jit_ = allow;
+  return true;
+}
+
+const upb_pbdecodermethod *upb_pbcodecache_getdecodermethod(
+    upb_pbcodecache *c, const upb_pbdecodermethodopts *opts) {
+  upb_value v;
+  bool ok;
+
+  /* Right now we build a new DecoderMethod every time.
+   * TODO(haberman): properly cache methods by their true key. */
+  const mgroup *g = mgroup_new(opts->handlers, c->allow_jit_, opts->lazy, c);
+  upb_inttable_push(&c->groups, upb_value_constptr(g));
+
+  ok = upb_inttable_lookupptr(&g->methods, opts->handlers, &v);
+  UPB_ASSERT_VAR(ok, ok);
+  return upb_value_getptr(v);
+}
+
+
+/* upb_pbdecodermethodopts ****************************************************/
+
+void upb_pbdecodermethodopts_init(upb_pbdecodermethodopts *opts,
+                                  const upb_handlers *h) {
+  opts->handlers = h;
+  opts->lazy = false;
+}
+
+void upb_pbdecodermethodopts_setlazy(upb_pbdecodermethodopts *opts, bool lazy) {
+  opts->lazy = lazy;
+}
+/*
+** upb::Decoder (Bytecode Decoder VM)
+**
+** Bytecode must previously have been generated using the bytecode compiler in
+** compile_decoder.c.  This decoder then walks through the bytecode op-by-op to
+** parse the input.
+**
+** Decoding is fully resumable; we just keep a pointer to the current bytecode
+** instruction and resume from there.  A fair amount of the logic here is to
+** handle the fact that values can span buffer seams and we have to be able to
+** be capable of suspending/resuming from any byte in the stream.  This
+** sometimes requires keeping a few trailing bytes from the last buffer around
+** in the "residual" buffer.
+*/
+
+#include <inttypes.h>
+#include <stddef.h>
+
+#ifdef UPB_DUMP_BYTECODE
+#include <stdio.h>
+#endif
+
+#define CHECK_SUSPEND(x) if (!(x)) return upb_pbdecoder_suspend(d);
+
+/* Error messages that are shared between the bytecode and JIT decoders. */
+const char *kPbDecoderStackOverflow = "Nesting too deep.";
+const char *kPbDecoderSubmessageTooLong =
+    "Submessage end extends past enclosing submessage.";
+
+/* Error messages shared within this file. */
+static const char *kUnterminatedVarint = "Unterminated varint.";
+
+/* upb_pbdecoder **************************************************************/
+
+static opcode halt = OP_HALT;
+
+/* Whether an op consumes any of the input buffer. */
+static bool consumes_input(opcode op) {
+  switch (op) {
+    case OP_SETDISPATCH:
+    case OP_STARTMSG:
+    case OP_ENDMSG:
+    case OP_STARTSEQ:
+    case OP_ENDSEQ:
+    case OP_STARTSUBMSG:
+    case OP_ENDSUBMSG:
+    case OP_STARTSTR:
+    case OP_ENDSTR:
+    case OP_PUSHTAGDELIM:
+    case OP_POP:
+    case OP_SETDELIM:
+    case OP_SETBIGGROUPNUM:
+    case OP_CHECKDELIM:
+    case OP_CALL:
+    case OP_RET:
+    case OP_BRANCH:
+      return false;
+    default:
+      return true;
+  }
+}
+
+static size_t stacksize(upb_pbdecoder *d, size_t entries) {
+  UPB_UNUSED(d);
+  return entries * sizeof(upb_pbdecoder_frame);
+}
+
+static size_t callstacksize(upb_pbdecoder *d, size_t entries) {
+  UPB_UNUSED(d);
+
+#ifdef UPB_USE_JIT_X64
+  if (d->method_->is_native_) {
+    /* Each native stack frame needs two pointers, plus we need a few frames for
+     * the enter/exit trampolines. */
+    size_t ret = entries * sizeof(void*) * 2;
+    ret += sizeof(void*) * 10;
+    return ret;
+  }
+#endif
+
+  return entries * sizeof(uint32_t*);
+}
+
+
+static bool in_residual_buf(const upb_pbdecoder *d, const char *p);
+
+/* It's unfortunate that we have to micro-manage the compiler with
+ * UPB_FORCEINLINE and UPB_NOINLINE, especially since this tuning is necessarily
+ * specific to one hardware configuration.  But empirically on a Core i7,
+ * performance increases 30-50% with these annotations.  Every instance where
+ * these appear, gcc 4.2.1 made the wrong decision and degraded performance in
+ * benchmarks. */
+
+static void seterr(upb_pbdecoder *d, const char *msg) {
+  upb_status status = UPB_STATUS_INIT;
+  upb_status_seterrmsg(&status, msg);
+  upb_env_reporterror(d->env, &status);
+}
+
+void upb_pbdecoder_seterr(upb_pbdecoder *d, const char *msg) {
+  seterr(d, msg);
+}
+
+
+/* Buffering ******************************************************************/
+
+/* We operate on one buffer at a time, which is either the user's buffer passed
+ * to our "decode" callback or some residual bytes from the previous buffer. */
+
+/* How many bytes can be safely read from d->ptr without reading past end-of-buf
+ * or past the current delimited end. */
+static size_t curbufleft(const upb_pbdecoder *d) {
+  assert(d->data_end >= d->ptr);
+  return d->data_end - d->ptr;
+}
+
+/* How many bytes are available before end-of-buffer. */
+static size_t bufleft(const upb_pbdecoder *d) {
+  return d->end - d->ptr;
+}
+
+/* Overall stream offset of d->ptr. */
+uint64_t offset(const upb_pbdecoder *d) {
+  return d->bufstart_ofs + (d->ptr - d->buf);
+}
+
+/* How many bytes are available before the end of this delimited region. */
+size_t delim_remaining(const upb_pbdecoder *d) {
+  return d->top->end_ofs - offset(d);
+}
+
+/* Advances d->ptr. */
+static void advance(upb_pbdecoder *d, size_t len) {
+  assert(curbufleft(d) >= len);
+  d->ptr += len;
+}
+
+static bool in_buf(const char *p, const char *buf, const char *end) {
+  return p >= buf && p <= end;
+}
+
+static bool in_residual_buf(const upb_pbdecoder *d, const char *p) {
+  return in_buf(p, d->residual, d->residual_end);
+}
+
+/* Calculates the delim_end value, which is affected by both the current buffer
+ * and the parsing stack, so must be called whenever either is updated. */
+static void set_delim_end(upb_pbdecoder *d) {
+  size_t delim_ofs = d->top->end_ofs - d->bufstart_ofs;
+  if (delim_ofs <= (size_t)(d->end - d->buf)) {
+    d->delim_end = d->buf + delim_ofs;
+    d->data_end = d->delim_end;
+  } else {
+    d->data_end = d->end;
+    d->delim_end = NULL;
+  }
+}
+
+static void switchtobuf(upb_pbdecoder *d, const char *buf, const char *end) {
+  d->ptr = buf;
+  d->buf = buf;
+  d->end = end;
+  set_delim_end(d);
+}
+
+static void advancetobuf(upb_pbdecoder *d, const char *buf, size_t len) {
+  assert(curbufleft(d) == 0);
+  d->bufstart_ofs += (d->end - d->buf);
+  switchtobuf(d, buf, buf + len);
+}
+
+static void checkpoint(upb_pbdecoder *d) {
+  /* The assertion here is in the interests of efficiency, not correctness.
+   * We are trying to ensure that we don't checkpoint() more often than
+   * necessary. */
+  assert(d->checkpoint != d->ptr);
+  d->checkpoint = d->ptr;
+}
+
+/* Skips "bytes" bytes in the stream, which may be more than available.  If we
+ * skip more bytes than are available, we return a long read count to the caller
+ * indicating how many bytes can be skipped over before passing actual data
+ * again.  Skipped bytes can pass a NULL buffer and the decoder guarantees they
+ * won't actually be read.
+ */
+static int32_t skip(upb_pbdecoder *d, size_t bytes) {
+  assert(!in_residual_buf(d, d->ptr) || d->size_param == 0);
+  assert(d->skip == 0);
+  if (bytes > delim_remaining(d)) {
+    seterr(d, "Skipped value extended beyond enclosing submessage.");
+    return upb_pbdecoder_suspend(d);
+  } else if (bufleft(d) > bytes) {
+    /* Skipped data is all in current buffer, and more is still available. */
+    advance(d, bytes);
+    d->skip = 0;
+    return DECODE_OK;
+  } else {
+    /* Skipped data extends beyond currently available buffers. */
+    d->pc = d->last;
+    d->skip = bytes - curbufleft(d);
+    d->bufstart_ofs += (d->end - d->buf);
+    d->residual_end = d->residual;
+    switchtobuf(d, d->residual, d->residual_end);
+    return d->size_param + d->skip;
+  }
+}
+
+
+/* Resumes the decoder from an initial state or from a previous suspend. */
+int32_t upb_pbdecoder_resume(upb_pbdecoder *d, void *p, const char *buf,
+                             size_t size, const upb_bufhandle *handle) {
+  UPB_UNUSED(p);  /* Useless; just for the benefit of the JIT. */
+
+  d->buf_param = buf;
+  d->size_param = size;
+  d->handle = handle;
+
+  if (d->residual_end > d->residual) {
+    /* We have residual bytes from the last buffer. */
+    assert(d->ptr == d->residual);
+  } else {
+    switchtobuf(d, buf, buf + size);
+  }
+
+  d->checkpoint = d->ptr;
+
+  if (d->skip) {
+    size_t skip_bytes = d->skip;
+    d->skip = 0;
+    CHECK_RETURN(skip(d, skip_bytes));
+    d->checkpoint = d->ptr;
+  }
+
+  if (!buf) {
+    /* NULL buf is ok if its entire span is covered by the "skip" above, but
+     * by this point we know that "skip" doesn't cover the buffer. */
+    seterr(d, "Passed NULL buffer over non-skippable region.");
+    return upb_pbdecoder_suspend(d);
+  }
+
+  if (d->top->groupnum < 0) {
+    CHECK_RETURN(upb_pbdecoder_skipunknown(d, -1, 0));
+    d->checkpoint = d->ptr;
+  }
+
+  return DECODE_OK;
+}
+
+/* Suspends the decoder at the last checkpoint, without saving any residual
+ * bytes.  If there are any unconsumed bytes, returns a short byte count. */
+size_t upb_pbdecoder_suspend(upb_pbdecoder *d) {
+  d->pc = d->last;
+  if (d->checkpoint == d->residual) {
+    /* Checkpoint was in residual buf; no user bytes were consumed. */
+    d->ptr = d->residual;
+    return 0;
+  } else {
+    size_t consumed;
+    assert(!in_residual_buf(d, d->checkpoint));
+    assert(d->buf == d->buf_param);
+
+    consumed = d->checkpoint - d->buf;
+    d->bufstart_ofs += consumed;
+    d->residual_end = d->residual;
+    switchtobuf(d, d->residual, d->residual_end);
+    return consumed;
+  }
+}
+
+/* Suspends the decoder at the last checkpoint, and saves any unconsumed
+ * bytes in our residual buffer.  This is necessary if we need more user
+ * bytes to form a complete value, which might not be contiguous in the
+ * user's buffers.  Always consumes all user bytes. */
+static size_t suspend_save(upb_pbdecoder *d) {
+  /* We hit end-of-buffer before we could parse a full value.
+   * Save any unconsumed bytes (if any) to the residual buffer. */
+  d->pc = d->last;
+
+  if (d->checkpoint == d->residual) {
+    /* Checkpoint was in residual buf; append user byte(s) to residual buf. */
+    assert((d->residual_end - d->residual) + d->size_param <=
+           sizeof(d->residual));
+    if (!in_residual_buf(d, d->ptr)) {
+      d->bufstart_ofs -= (d->residual_end - d->residual);
+    }
+    memcpy(d->residual_end, d->buf_param, d->size_param);
+    d->residual_end += d->size_param;
+  } else {
+    /* Checkpoint was in user buf; old residual bytes not needed. */
+    size_t save;
+    assert(!in_residual_buf(d, d->checkpoint));
+
+    d->ptr = d->checkpoint;
+    save = curbufleft(d);
+    assert(save <= sizeof(d->residual));
+    memcpy(d->residual, d->ptr, save);
+    d->residual_end = d->residual + save;
+    d->bufstart_ofs = offset(d);
+  }
+
+  switchtobuf(d, d->residual, d->residual_end);
+  return d->size_param;
+}
+
+/* Copies the next "bytes" bytes into "buf" and advances the stream.
+ * Requires that this many bytes are available in the current buffer. */
+UPB_FORCEINLINE static void consumebytes(upb_pbdecoder *d, void *buf,
+                                         size_t bytes) {
+  assert(bytes <= curbufleft(d));
+  memcpy(buf, d->ptr, bytes);
+  advance(d, bytes);
+}
+
+/* Slow path for getting the next "bytes" bytes, regardless of whether they are
+ * available in the current buffer or not.  Returns a status code as described
+ * in decoder.int.h. */
+UPB_NOINLINE static int32_t getbytes_slow(upb_pbdecoder *d, void *buf,
+                                          size_t bytes) {
+  const size_t avail = curbufleft(d);
+  consumebytes(d, buf, avail);
+  bytes -= avail;
+  assert(bytes > 0);
+  if (in_residual_buf(d, d->ptr)) {
+    advancetobuf(d, d->buf_param, d->size_param);
+  }
+  if (curbufleft(d) >= bytes) {
+    consumebytes(d, (char *)buf + avail, bytes);
+    return DECODE_OK;
+  } else if (d->data_end == d->delim_end) {
+    seterr(d, "Submessage ended in the middle of a value or group");
+    return upb_pbdecoder_suspend(d);
+  } else {
+    return suspend_save(d);
+  }
+}
+
+/* Gets the next "bytes" bytes, regardless of whether they are available in the
+ * current buffer or not.  Returns a status code as described in decoder.int.h.
+ */
+UPB_FORCEINLINE static int32_t getbytes(upb_pbdecoder *d, void *buf,
+                                        size_t bytes) {
+  if (curbufleft(d) >= bytes) {
+    /* Buffer has enough data to satisfy. */
+    consumebytes(d, buf, bytes);
+    return DECODE_OK;
+  } else {
+    return getbytes_slow(d, buf, bytes);
+  }
+}
+
+UPB_NOINLINE static size_t peekbytes_slow(upb_pbdecoder *d, void *buf,
+                                          size_t bytes) {
+  size_t ret = curbufleft(d);
+  memcpy(buf, d->ptr, ret);
+  if (in_residual_buf(d, d->ptr)) {
+    size_t copy = UPB_MIN(bytes - ret, d->size_param);
+    memcpy((char *)buf + ret, d->buf_param, copy);
+    ret += copy;
+  }
+  return ret;
+}
+
+UPB_FORCEINLINE static size_t peekbytes(upb_pbdecoder *d, void *buf,
+                                        size_t bytes) {
+  if (curbufleft(d) >= bytes) {
+    memcpy(buf, d->ptr, bytes);
+    return bytes;
+  } else {
+    return peekbytes_slow(d, buf, bytes);
+  }
+}
+
+
+/* Decoding of wire types *****************************************************/
+
+/* Slow path for decoding a varint from the current buffer position.
+ * Returns a status code as described in decoder.int.h. */
+UPB_NOINLINE int32_t upb_pbdecoder_decode_varint_slow(upb_pbdecoder *d,
+                                                      uint64_t *u64) {
+  uint8_t byte = 0x80;
+  int bitpos;
+  *u64 = 0;
+  for(bitpos = 0; bitpos < 70 && (byte & 0x80); bitpos += 7) {
+    int32_t ret = getbytes(d, &byte, 1);
+    if (ret >= 0) return ret;
+    *u64 |= (uint64_t)(byte & 0x7F) << bitpos;
+  }
+  if(bitpos == 70 && (byte & 0x80)) {
+    seterr(d, kUnterminatedVarint);
+    return upb_pbdecoder_suspend(d);
+  }
+  return DECODE_OK;
+}
+
+/* Decodes a varint from the current buffer position.
+ * Returns a status code as described in decoder.int.h. */
+UPB_FORCEINLINE static int32_t decode_varint(upb_pbdecoder *d, uint64_t *u64) {
+  if (curbufleft(d) > 0 && !(*d->ptr & 0x80)) {
+    *u64 = *d->ptr;
+    advance(d, 1);
+    return DECODE_OK;
+  } else if (curbufleft(d) >= 10) {
+    /* Fast case. */
+    upb_decoderet r = upb_vdecode_fast(d->ptr);
+    if (r.p == NULL) {
+      seterr(d, kUnterminatedVarint);
+      return upb_pbdecoder_suspend(d);
+    }
+    advance(d, r.p - d->ptr);
+    *u64 = r.val;
+    return DECODE_OK;
+  } else {
+    /* Slow case -- varint spans buffer seam. */
+    return upb_pbdecoder_decode_varint_slow(d, u64);
+  }
+}
+
+/* Decodes a 32-bit varint from the current buffer position.
+ * Returns a status code as described in decoder.int.h. */
+UPB_FORCEINLINE static int32_t decode_v32(upb_pbdecoder *d, uint32_t *u32) {
+  uint64_t u64;
+  int32_t ret = decode_varint(d, &u64);
+  if (ret >= 0) return ret;
+  if (u64 > UINT32_MAX) {
+    seterr(d, "Unterminated 32-bit varint");
+    /* TODO(haberman) guarantee that this function return is >= 0 somehow,
+     * so we know this path will always be treated as error by our caller.
+     * Right now the size_t -> int32_t can overflow and produce negative values.
+     */
+    *u32 = 0;
+    return upb_pbdecoder_suspend(d);
+  }
+  *u32 = u64;
+  return DECODE_OK;
+}
+
+/* Decodes a fixed32 from the current buffer position.
+ * Returns a status code as described in decoder.int.h.
+ * TODO: proper byte swapping for big-endian machines. */
+UPB_FORCEINLINE static int32_t decode_fixed32(upb_pbdecoder *d, uint32_t *u32) {
+  return getbytes(d, u32, 4);
+}
+
+/* Decodes a fixed64 from the current buffer position.
+ * Returns a status code as described in decoder.int.h.
+ * TODO: proper byte swapping for big-endian machines. */
+UPB_FORCEINLINE static int32_t decode_fixed64(upb_pbdecoder *d, uint64_t *u64) {
+  return getbytes(d, u64, 8);
+}
+
+/* Non-static versions of the above functions.
+ * These are called by the JIT for fallback paths. */
+int32_t upb_pbdecoder_decode_f32(upb_pbdecoder *d, uint32_t *u32) {
+  return decode_fixed32(d, u32);
+}
+
+int32_t upb_pbdecoder_decode_f64(upb_pbdecoder *d, uint64_t *u64) {
+  return decode_fixed64(d, u64);
+}
+
+static double as_double(uint64_t n) { double d; memcpy(&d, &n, 8); return d; }
+static float  as_float(uint32_t n)  { float  f; memcpy(&f, &n, 4); return f; }
+
+/* Pushes a frame onto the decoder stack. */
+static bool decoder_push(upb_pbdecoder *d, uint64_t end) {
+  upb_pbdecoder_frame *fr = d->top;
+
+  if (end > fr->end_ofs) {
+    seterr(d, kPbDecoderSubmessageTooLong);
+    return false;
+  } else if (fr == d->limit) {
+    seterr(d, kPbDecoderStackOverflow);
+    return false;
+  }
+
+  fr++;
+  fr->end_ofs = end;
+  fr->dispatch = NULL;
+  fr->groupnum = 0;
+  d->top = fr;
+  return true;
+}
+
+static bool pushtagdelim(upb_pbdecoder *d, uint32_t arg) {
+  /* While we expect to see an "end" tag (either ENDGROUP or a non-sequence
+   * field number) prior to hitting any enclosing submessage end, pushing our
+   * existing delim end prevents us from continuing to parse values from a
+   * corrupt proto that doesn't give us an END tag in time. */
+  if (!decoder_push(d, d->top->end_ofs))
+    return false;
+  d->top->groupnum = arg;
+  return true;
+}
+
+/* Pops a frame from the decoder stack. */
+static void decoder_pop(upb_pbdecoder *d) { d->top--; }
+
+UPB_NOINLINE int32_t upb_pbdecoder_checktag_slow(upb_pbdecoder *d,
+                                                 uint64_t expected) {
+  uint64_t data = 0;
+  size_t bytes = upb_value_size(expected);
+  size_t read = peekbytes(d, &data, bytes);
+  if (read == bytes && data == expected) {
+    /* Advance past matched bytes. */
+    int32_t ok = getbytes(d, &data, read);
+    UPB_ASSERT_VAR(ok, ok < 0);
+    return DECODE_OK;
+  } else if (read < bytes && memcmp(&data, &expected, read) == 0) {
+    return suspend_save(d);
+  } else {
+    return DECODE_MISMATCH;
+  }
+}
+
+int32_t upb_pbdecoder_skipunknown(upb_pbdecoder *d, int32_t fieldnum,
+                                  uint8_t wire_type) {
+  if (fieldnum >= 0)
+    goto have_tag;
+
+  while (true) {
+    uint32_t tag;
+    CHECK_RETURN(decode_v32(d, &tag));
+    wire_type = tag & 0x7;
+    fieldnum = tag >> 3;
+
+have_tag:
+    if (fieldnum == 0) {
+      seterr(d, "Saw invalid field number (0)");
+      return upb_pbdecoder_suspend(d);
+    }
+
+    /* TODO: deliver to unknown field callback. */
+    switch (wire_type) {
+      case UPB_WIRE_TYPE_32BIT:
+        CHECK_RETURN(skip(d, 4));
+        break;
+      case UPB_WIRE_TYPE_64BIT:
+        CHECK_RETURN(skip(d, 8));
+        break;
+      case UPB_WIRE_TYPE_VARINT: {
+        uint64_t u64;
+        CHECK_RETURN(decode_varint(d, &u64));
+        break;
+      }
+      case UPB_WIRE_TYPE_DELIMITED: {
+        uint32_t len;
+        CHECK_RETURN(decode_v32(d, &len));
+        CHECK_RETURN(skip(d, len));
+        break;
+      }
+      case UPB_WIRE_TYPE_START_GROUP:
+        CHECK_SUSPEND(pushtagdelim(d, -fieldnum));
+        break;
+      case UPB_WIRE_TYPE_END_GROUP:
+        if (fieldnum == -d->top->groupnum) {
+          decoder_pop(d);
+        } else if (fieldnum == d->top->groupnum) {
+          return DECODE_ENDGROUP;
+        } else {
+          seterr(d, "Unmatched ENDGROUP tag.");
+          return upb_pbdecoder_suspend(d);
+        }
+        break;
+      default:
+        seterr(d, "Invalid wire type");
+        return upb_pbdecoder_suspend(d);
+    }
+
+    if (d->top->groupnum >= 0) {
+      return DECODE_OK;
+    }
+
+    /* Unknown group -- continue looping over unknown fields. */
+    checkpoint(d);
+  }
+}
+
+static void goto_endmsg(upb_pbdecoder *d) {
+  upb_value v;
+  bool found = upb_inttable_lookup32(d->top->dispatch, DISPATCH_ENDMSG, &v);
+  UPB_ASSERT_VAR(found, found);
+  d->pc = d->top->base + upb_value_getuint64(v);
+}
+
+/* Parses a tag and jumps to the corresponding bytecode instruction for this
+ * field.
+ *
+ * If the tag is unknown (or the wire type doesn't match), parses the field as
+ * unknown.  If the tag is a valid ENDGROUP tag, jumps to the bytecode
+ * instruction for the end of message. */
+static int32_t dispatch(upb_pbdecoder *d) {
+  upb_inttable *dispatch = d->top->dispatch;
+  uint32_t tag;
+  uint8_t wire_type;
+  uint32_t fieldnum;
+  upb_value val;
+  int32_t retval;
+
+  /* Decode tag. */
+  CHECK_RETURN(decode_v32(d, &tag));
+  wire_type = tag & 0x7;
+  fieldnum = tag >> 3;
+
+  /* Lookup tag.  Because of packed/non-packed compatibility, we have to
+   * check the wire type against two possibilities. */
+  if (fieldnum != DISPATCH_ENDMSG &&
+      upb_inttable_lookup32(dispatch, fieldnum, &val)) {
+    uint64_t v = upb_value_getuint64(val);
+    if (wire_type == (v & 0xff)) {
+      d->pc = d->top->base + (v >> 16);
+      return DECODE_OK;
+    } else if (wire_type == ((v >> 8) & 0xff)) {
+      bool found =
+          upb_inttable_lookup(dispatch, fieldnum + UPB_MAX_FIELDNUMBER, &val);
+      UPB_ASSERT_VAR(found, found);
+      d->pc = d->top->base + upb_value_getuint64(val);
+      return DECODE_OK;
+    }
+  }
+
+  /* We have some unknown fields (or ENDGROUP) to parse.  The DISPATCH or TAG
+   * bytecode that triggered this is preceded by a CHECKDELIM bytecode which
+   * we need to back up to, so that when we're done skipping unknown data we
+   * can re-check the delimited end. */
+  d->last--;  /* Necessary if we get suspended */
+  d->pc = d->last;
+  assert(getop(*d->last) == OP_CHECKDELIM);
+
+  /* Unknown field or ENDGROUP. */
+  retval = upb_pbdecoder_skipunknown(d, fieldnum, wire_type);
+
+  CHECK_RETURN(retval);
+
+  if (retval == DECODE_ENDGROUP) {
+    goto_endmsg(d);
+    return DECODE_OK;
+  }
+
+  return DECODE_OK;
+}
+
+/* Callers know that the stack is more than one deep because the opcodes that
+ * call this only occur after PUSH operations. */
+upb_pbdecoder_frame *outer_frame(upb_pbdecoder *d) {
+  assert(d->top != d->stack);
+  return d->top - 1;
+}
+
+
+/* The main decoding loop *****************************************************/
+
+/* The main decoder VM function.  Uses traditional bytecode dispatch loop with a
+ * switch() statement. */
+size_t run_decoder_vm(upb_pbdecoder *d, const mgroup *group,
+                      const upb_bufhandle* handle) {
+
+#define VMCASE(op, code) \
+  case op: { code; if (consumes_input(op)) checkpoint(d); break; }
+#define PRIMITIVE_OP(type, wt, name, convfunc, ctype) \
+  VMCASE(OP_PARSE_ ## type, { \
+    ctype val; \
+    CHECK_RETURN(decode_ ## wt(d, &val)); \
+    upb_sink_put ## name(&d->top->sink, arg, (convfunc)(val)); \
+  })
+
+  while(1) {
+    int32_t instruction;
+    opcode op;
+    uint32_t arg;
+    int32_t longofs;
+
+    d->last = d->pc;
+    instruction = *d->pc++;
+    op = getop(instruction);
+    arg = instruction >> 8;
+    longofs = arg;
+    assert(d->ptr != d->residual_end);
+    UPB_UNUSED(group);
+#ifdef UPB_DUMP_BYTECODE
+    fprintf(stderr, "s_ofs=%d buf_ofs=%d data_rem=%d buf_rem=%d delim_rem=%d "
+                    "%x %s (%d)\n",
+            (int)offset(d),
+            (int)(d->ptr - d->buf),
+            (int)(d->data_end - d->ptr),
+            (int)(d->end - d->ptr),
+            (int)((d->top->end_ofs - d->bufstart_ofs) - (d->ptr - d->buf)),
+            (int)(d->pc - 1 - group->bytecode),
+            upb_pbdecoder_getopname(op),
+            arg);
+#endif
+    switch (op) {
+      /* Technically, we are losing data if we see a 32-bit varint that is not
+       * properly sign-extended.  We could detect this and error about the data
+       * loss, but proto2 does not do this, so we pass. */
+      PRIMITIVE_OP(INT32,    varint,  int32,  int32_t,      uint64_t)
+      PRIMITIVE_OP(INT64,    varint,  int64,  int64_t,      uint64_t)
+      PRIMITIVE_OP(UINT32,   varint,  uint32, uint32_t,     uint64_t)
+      PRIMITIVE_OP(UINT64,   varint,  uint64, uint64_t,     uint64_t)
+      PRIMITIVE_OP(FIXED32,  fixed32, uint32, uint32_t,     uint32_t)
+      PRIMITIVE_OP(FIXED64,  fixed64, uint64, uint64_t,     uint64_t)
+      PRIMITIVE_OP(SFIXED32, fixed32, int32,  int32_t,      uint32_t)
+      PRIMITIVE_OP(SFIXED64, fixed64, int64,  int64_t,      uint64_t)
+      PRIMITIVE_OP(BOOL,     varint,  bool,   bool,         uint64_t)
+      PRIMITIVE_OP(DOUBLE,   fixed64, double, as_double,    uint64_t)
+      PRIMITIVE_OP(FLOAT,    fixed32, float,  as_float,     uint32_t)
+      PRIMITIVE_OP(SINT32,   varint,  int32,  upb_zzdec_32, uint64_t)
+      PRIMITIVE_OP(SINT64,   varint,  int64,  upb_zzdec_64, uint64_t)
+
+      VMCASE(OP_SETDISPATCH,
+        d->top->base = d->pc - 1;
+        memcpy(&d->top->dispatch, d->pc, sizeof(void*));
+        d->pc += sizeof(void*) / sizeof(uint32_t);
+      )
+      VMCASE(OP_STARTMSG,
+        CHECK_SUSPEND(upb_sink_startmsg(&d->top->sink));
+      )
+      VMCASE(OP_ENDMSG,
+        CHECK_SUSPEND(upb_sink_endmsg(&d->top->sink, d->status));
+      )
+      VMCASE(OP_STARTSEQ,
+        upb_pbdecoder_frame *outer = outer_frame(d);
+        CHECK_SUSPEND(upb_sink_startseq(&outer->sink, arg, &d->top->sink));
+      )
+      VMCASE(OP_ENDSEQ,
+        CHECK_SUSPEND(upb_sink_endseq(&d->top->sink, arg));
+      )
+      VMCASE(OP_STARTSUBMSG,
+        upb_pbdecoder_frame *outer = outer_frame(d);
+        CHECK_SUSPEND(upb_sink_startsubmsg(&outer->sink, arg, &d->top->sink));
+      )
+      VMCASE(OP_ENDSUBMSG,
+        CHECK_SUSPEND(upb_sink_endsubmsg(&d->top->sink, arg));
+      )
+      VMCASE(OP_STARTSTR,
+        uint32_t len = delim_remaining(d);
+        upb_pbdecoder_frame *outer = outer_frame(d);
+        CHECK_SUSPEND(upb_sink_startstr(&outer->sink, arg, len, &d->top->sink));
+        if (len == 0) {
+          d->pc++;  /* Skip OP_STRING. */
+        }
+      )
+      VMCASE(OP_STRING,
+        uint32_t len = curbufleft(d);
+        size_t n = upb_sink_putstring(&d->top->sink, arg, d->ptr, len, handle);
+        if (n > len) {
+          if (n > delim_remaining(d)) {
+            seterr(d, "Tried to skip past end of string.");
+            return upb_pbdecoder_suspend(d);
+          } else {
+            int32_t ret = skip(d, n);
+            /* This shouldn't return DECODE_OK, because n > len. */
+            assert(ret >= 0);
+            return ret;
+          }
+        }
+        advance(d, n);
+        if (n < len || d->delim_end == NULL) {
+          /* We aren't finished with this string yet. */
+          d->pc--;  /* Repeat OP_STRING. */
+          if (n > 0) checkpoint(d);
+          return upb_pbdecoder_suspend(d);
+        }
+      )
+      VMCASE(OP_ENDSTR,
+        CHECK_SUSPEND(upb_sink_endstr(&d->top->sink, arg));
+      )
+      VMCASE(OP_PUSHTAGDELIM,
+        CHECK_SUSPEND(pushtagdelim(d, arg));
+      )
+      VMCASE(OP_SETBIGGROUPNUM,
+        d->top->groupnum = *d->pc++;
+      )
+      VMCASE(OP_POP,
+        assert(d->top > d->stack);
+        decoder_pop(d);
+      )
+      VMCASE(OP_PUSHLENDELIM,
+        uint32_t len;
+        CHECK_RETURN(decode_v32(d, &len));
+        CHECK_SUSPEND(decoder_push(d, offset(d) + len));
+        set_delim_end(d);
+      )
+      VMCASE(OP_SETDELIM,
+        set_delim_end(d);
+      )
+      VMCASE(OP_CHECKDELIM,
+        /* We are guaranteed of this assert because we never allow ourselves to
+         * consume bytes beyond data_end, which covers delim_end when non-NULL.
+         */
+        assert(!(d->delim_end && d->ptr > d->delim_end));
+        if (d->ptr == d->delim_end)
+          d->pc += longofs;
+      )
+      VMCASE(OP_CALL,
+        d->callstack[d->call_len++] = d->pc;
+        d->pc += longofs;
+      )
+      VMCASE(OP_RET,
+        assert(d->call_len > 0);
+        d->pc = d->callstack[--d->call_len];
+      )
+      VMCASE(OP_BRANCH,
+        d->pc += longofs;
+      )
+      VMCASE(OP_TAG1,
+        uint8_t expected;
+        CHECK_SUSPEND(curbufleft(d) > 0);
+        expected = (arg >> 8) & 0xff;
+        if (*d->ptr == expected) {
+          advance(d, 1);
+        } else {
+          int8_t shortofs;
+         badtag:
+          shortofs = arg;
+          if (shortofs == LABEL_DISPATCH) {
+            CHECK_RETURN(dispatch(d));
+          } else {
+            d->pc += shortofs;
+            break; /* Avoid checkpoint(). */
+          }
+        }
+      )
+      VMCASE(OP_TAG2,
+        uint16_t expected;
+        CHECK_SUSPEND(curbufleft(d) > 0);
+        expected = (arg >> 8) & 0xffff;
+        if (curbufleft(d) >= 2) {
+          uint16_t actual;
+          memcpy(&actual, d->ptr, 2);
+          if (expected == actual) {
+            advance(d, 2);
+          } else {
+            goto badtag;
+          }
+        } else {
+          int32_t result = upb_pbdecoder_checktag_slow(d, expected);
+          if (result == DECODE_MISMATCH) goto badtag;
+          if (result >= 0) return result;
+        }
+      )
+      VMCASE(OP_TAGN, {
+        uint64_t expected;
+        int32_t result;
+        memcpy(&expected, d->pc, 8);
+        d->pc += 2;
+        result = upb_pbdecoder_checktag_slow(d, expected);
+        if (result == DECODE_MISMATCH) goto badtag;
+        if (result >= 0) return result;
+      })
+      VMCASE(OP_DISPATCH, {
+        CHECK_RETURN(dispatch(d));
+      })
+      VMCASE(OP_HALT, {
+        return d->size_param;
+      })
+    }
+  }
+}
+
+
+/* BytesHandler handlers ******************************************************/
+
+void *upb_pbdecoder_startbc(void *closure, const void *pc, size_t size_hint) {
+  upb_pbdecoder *d = closure;
+  UPB_UNUSED(size_hint);
+  d->top->end_ofs = UINT64_MAX;
+  d->bufstart_ofs = 0;
+  d->call_len = 1;
+  d->callstack[0] = &halt;
+  d->pc = pc;
+  d->skip = 0;
+  return d;
+}
+
+void *upb_pbdecoder_startjit(void *closure, const void *hd, size_t size_hint) {
+  upb_pbdecoder *d = closure;
+  UPB_UNUSED(hd);
+  UPB_UNUSED(size_hint);
+  d->top->end_ofs = UINT64_MAX;
+  d->bufstart_ofs = 0;
+  d->call_len = 0;
+  d->skip = 0;
+  return d;
+}
+
+bool upb_pbdecoder_end(void *closure, const void *handler_data) {
+  upb_pbdecoder *d = closure;
+  const upb_pbdecodermethod *method = handler_data;
+  uint64_t end;
+  char dummy;
+
+  if (d->residual_end > d->residual) {
+    seterr(d, "Unexpected EOF: decoder still has buffered unparsed data");
+    return false;
+  }
+
+  if (d->skip) {
+    seterr(d, "Unexpected EOF inside skipped data");
+    return false;
+  }
+
+  if (d->top->end_ofs != UINT64_MAX) {
+    seterr(d, "Unexpected EOF inside delimited string");
+    return false;
+  }
+
+  /* The user's end() call indicates that the message ends here. */
+  end = offset(d);
+  d->top->end_ofs = end;
+
+#ifdef UPB_USE_JIT_X64
+  if (method->is_native_) {
+    const mgroup *group = (const mgroup*)method->group;
+    if (d->top != d->stack)
+      d->stack->end_ofs = 0;
+    group->jit_code(closure, method->code_base.ptr, &dummy, 0, NULL);
+  } else
+#endif
+  {
+    const uint32_t *p = d->pc;
+    d->stack->end_ofs = end;
+    /* Check the previous bytecode, but guard against beginning. */
+    if (p != method->code_base.ptr) p--;
+    if (getop(*p) == OP_CHECKDELIM) {
+      /* Rewind from OP_TAG* to OP_CHECKDELIM. */
+      assert(getop(*d->pc) == OP_TAG1 ||
+             getop(*d->pc) == OP_TAG2 ||
+             getop(*d->pc) == OP_TAGN ||
+             getop(*d->pc) == OP_DISPATCH);
+      d->pc = p;
+    }
+    upb_pbdecoder_decode(closure, handler_data, &dummy, 0, NULL);
+  }
+
+  if (d->call_len != 0) {
+    seterr(d, "Unexpected EOF inside submessage or group");
+    return false;
+  }
+
+  return true;
+}
+
+size_t upb_pbdecoder_decode(void *decoder, const void *group, const char *buf,
+                            size_t size, const upb_bufhandle *handle) {
+  int32_t result = upb_pbdecoder_resume(decoder, NULL, buf, size, handle);
+
+  if (result == DECODE_ENDGROUP) goto_endmsg(decoder);
+  CHECK_RETURN(result);
+
+  return run_decoder_vm(decoder, group, handle);
+}
+
+
+/* Public API *****************************************************************/
+
+void upb_pbdecoder_reset(upb_pbdecoder *d) {
+  d->top = d->stack;
+  d->top->groupnum = 0;
+  d->ptr = d->residual;
+  d->buf = d->residual;
+  d->end = d->residual;
+  d->residual_end = d->residual;
+}
+
+upb_pbdecoder *upb_pbdecoder_create(upb_env *e, const upb_pbdecodermethod *m,
+                                    upb_sink *sink) {
+  const size_t default_max_nesting = 64;
+#ifndef NDEBUG
+  size_t size_before = upb_env_bytesallocated(e);
+#endif
+
+  upb_pbdecoder *d = upb_env_malloc(e, sizeof(upb_pbdecoder));
+  if (!d) return NULL;
+
+  d->method_ = m;
+  d->callstack = upb_env_malloc(e, callstacksize(d, default_max_nesting));
+  d->stack = upb_env_malloc(e, stacksize(d, default_max_nesting));
+  if (!d->stack || !d->callstack) {
+    return NULL;
+  }
+
+  d->env = e;
+  d->limit = d->stack + default_max_nesting - 1;
+  d->stack_size = default_max_nesting;
+
+  upb_pbdecoder_reset(d);
+  upb_bytessink_reset(&d->input_, &m->input_handler_, d);
+
+  assert(sink);
+  if (d->method_->dest_handlers_) {
+    if (sink->handlers != d->method_->dest_handlers_)
+      return NULL;
+  }
+  upb_sink_reset(&d->top->sink, sink->handlers, sink->closure);
+
+  /* If this fails, increase the value in decoder.h. */
+  assert(upb_env_bytesallocated(e) - size_before <= UPB_PB_DECODER_SIZE);
+  return d;
+}
+
+uint64_t upb_pbdecoder_bytesparsed(const upb_pbdecoder *d) {
+  return offset(d);
+}
+
+const upb_pbdecodermethod *upb_pbdecoder_method(const upb_pbdecoder *d) {
+  return d->method_;
+}
+
+upb_bytessink *upb_pbdecoder_input(upb_pbdecoder *d) {
+  return &d->input_;
+}
+
+size_t upb_pbdecoder_maxnesting(const upb_pbdecoder *d) {
+  return d->stack_size;
+}
+
+bool upb_pbdecoder_setmaxnesting(upb_pbdecoder *d, size_t max) {
+  assert(d->top >= d->stack);
+
+  if (max < (size_t)(d->top - d->stack)) {
+    /* Can't set a limit smaller than what we are currently at. */
+    return false;
+  }
+
+  if (max > d->stack_size) {
+    /* Need to reallocate stack and callstack to accommodate. */
+    size_t old_size = stacksize(d, d->stack_size);
+    size_t new_size = stacksize(d, max);
+    void *p = upb_env_realloc(d->env, d->stack, old_size, new_size);
+    if (!p) {
+      return false;
+    }
+    d->stack = p;
+
+    old_size = callstacksize(d, d->stack_size);
+    new_size = callstacksize(d, max);
+    p = upb_env_realloc(d->env, d->callstack, old_size, new_size);
+    if (!p) {
+      return false;
+    }
+    d->callstack = p;
+
+    d->stack_size = max;
+  }
+
+  d->limit = d->stack + max - 1;
+  return true;
+}
+/*
+** upb::Encoder
+**
+** Since we are implementing pure handlers (ie. without any out-of-band access
+** to pre-computed lengths), we have to buffer all submessages before we can
+** emit even their first byte.
+**
+** Not knowing the size of submessages also means we can't write a perfect
+** zero-copy implementation, even with buffering.  Lengths are stored as
+** varints, which means that we don't know how many bytes to reserve for the
+** length until we know what the length is.
+**
+** This leaves us with three main choices:
+**
+** 1. buffer all submessage data in a temporary buffer, then copy it exactly
+**    once into the output buffer.
+**
+** 2. attempt to buffer data directly into the output buffer, estimating how
+**    many bytes each length will take.  When our guesses are wrong, use
+**    memmove() to grow or shrink the allotted space.
+**
+** 3. buffer directly into the output buffer, allocating a max length
+**    ahead-of-time for each submessage length.  If we overallocated, we waste
+**    space, but no memcpy() or memmove() is required.  This approach requires
+**    defining a maximum size for submessages and rejecting submessages that
+**    exceed that size.
+**
+** (2) and (3) have the potential to have better performance, but they are more
+** complicated and subtle to implement:
+**
+**   (3) requires making an arbitrary choice of the maximum message size; it
+**       wastes space when submessages are shorter than this and fails
+**       completely when they are longer.  This makes it more finicky and
+**       requires configuration based on the input.  It also makes it impossible
+**       to perfectly match the output of reference encoders that always use the
+**       optimal amount of space for each length.
+**
+**   (2) requires guessing the the size upfront, and if multiple lengths are
+**       guessed wrong the minimum required number of memmove() operations may
+**       be complicated to compute correctly.  Implemented properly, it may have
+**       a useful amortized or average cost, but more investigation is required
+**       to determine this and what the optimal algorithm is to achieve it.
+**
+**   (1) makes you always pay for exactly one copy, but its implementation is
+**       the simplest and its performance is predictable.
+**
+** So for now, we implement (1) only.  If we wish to optimize later, we should
+** be able to do it without affecting users.
+**
+** The strategy is to buffer the segments of data that do *not* depend on
+** unknown lengths in one buffer, and keep a separate buffer of segment pointers
+** and lengths.  When the top-level submessage ends, we can go beginning to end,
+** alternating the writing of lengths with memcpy() of the rest of the data.
+** At the top level though, no buffering is required.
+*/
+
+
+#include <stdlib.h>
+
+/* The output buffer is divided into segments; a segment is a string of data
+ * that is "ready to go" -- it does not need any varint lengths inserted into
+ * the middle.  The seams between segments are where varints will be inserted
+ * once they are known.
+ *
+ * We also use the concept of a "run", which is a range of encoded bytes that
+ * occur at a single submessage level.  Every segment contains one or more runs.
+ *
+ * A segment can span messages.  Consider:
+ *
+ *                  .--Submessage lengths---------.
+ *                  |       |                     |
+ *                  |       V                     V
+ *                  V      | |---------------    | |-----------------
+ * Submessages:    | |-----------------------------------------------
+ * Top-level msg: ------------------------------------------------------------
+ *
+ * Segments:          -----   -------------------   -----------------
+ * Runs:              *----   *--------------*---   *----------------
+ * (* marks the start)
+ *
+ * Note that the top-level menssage is not in any segment because it does not
+ * have any length preceding it.
+ *
+ * A segment is only interrupted when another length needs to be inserted.  So
+ * observe how the second segment spans both the inner submessage and part of
+ * the next enclosing message. */
+typedef struct {
+  uint32_t msglen;  /* The length to varint-encode before this segment. */
+  uint32_t seglen;  /* Length of the segment. */
+} upb_pb_encoder_segment;
+
+struct upb_pb_encoder {
+  upb_env *env;
+
+  /* Our input and output. */
+  upb_sink input_;
+  upb_bytessink *output_;
+
+  /* The "subclosure" -- used as the inner closure as part of the bytessink
+   * protocol. */
+  void *subc;
+
+  /* The output buffer and limit, and our current write position.  "buf"
+   * initially points to "initbuf", but is dynamically allocated if we need to
+   * grow beyond the initial size. */
+  char *buf, *ptr, *limit;
+
+  /* The beginning of the current run, or undefined if we are at the top
+   * level. */
+  char *runbegin;
+
+  /* The list of segments we are accumulating. */
+  upb_pb_encoder_segment *segbuf, *segptr, *seglimit;
+
+  /* The stack of enclosing submessages.  Each entry in the stack points to the
+   * segment where this submessage's length is being accumulated. */
+  int *stack, *top, *stacklimit;
+
+  /* Depth of startmsg/endmsg calls. */
+  int depth;
+};
+
+/* low-level buffering ********************************************************/
+
+/* Low-level functions for interacting with the output buffer. */
+
+/* TODO(haberman): handle pushback */
+static void putbuf(upb_pb_encoder *e, const char *buf, size_t len) {
+  size_t n = upb_bytessink_putbuf(e->output_, e->subc, buf, len, NULL);
+  UPB_ASSERT_VAR(n, n == len);
+}
+
+static upb_pb_encoder_segment *top(upb_pb_encoder *e) {
+  return &e->segbuf[*e->top];
+}
+
+/* Call to ensure that at least "bytes" bytes are available for writing at
+ * e->ptr.  Returns false if the bytes could not be allocated. */
+static bool reserve(upb_pb_encoder *e, size_t bytes) {
+  if ((size_t)(e->limit - e->ptr) < bytes) {
+    /* Grow buffer. */
+    char *new_buf;
+    size_t needed = bytes + (e->ptr - e->buf);
+    size_t old_size = e->limit - e->buf;
+
+    size_t new_size = old_size;
+
+    while (new_size < needed) {
+      new_size *= 2;
+    }
+
+    new_buf = upb_env_realloc(e->env, e->buf, old_size, new_size);
+
+    if (new_buf == NULL) {
+      return false;
+    }
+
+    e->ptr = new_buf + (e->ptr - e->buf);
+    e->runbegin = new_buf + (e->runbegin - e->buf);
+    e->limit = new_buf + new_size;
+    e->buf = new_buf;
+  }
+
+  return true;
+}
+
+/* Call when "bytes" bytes have been writte at e->ptr.  The caller *must* have
+ * previously called reserve() with at least this many bytes. */
+static void encoder_advance(upb_pb_encoder *e, size_t bytes) {
+  assert((size_t)(e->limit - e->ptr) >= bytes);
+  e->ptr += bytes;
+}
+
+/* Call when all of the bytes for a handler have been written.  Flushes the
+ * bytes if possible and necessary, returning false if this failed. */
+static bool commit(upb_pb_encoder *e) {
+  if (!e->top) {
+    /* We aren't inside a delimited region.  Flush our accumulated bytes to
+     * the output.
+     *
+     * TODO(haberman): in the future we may want to delay flushing for
+     * efficiency reasons. */
+    putbuf(e, e->buf, e->ptr - e->buf);
+    e->ptr = e->buf;
+  }
+
+  return true;
+}
+
+/* Writes the given bytes to the buffer, handling reserve/advance. */
+static bool encode_bytes(upb_pb_encoder *e, const void *data, size_t len) {
+  if (!reserve(e, len)) {
+    return false;
+  }
+
+  memcpy(e->ptr, data, len);
+  encoder_advance(e, len);
+  return true;
+}
+
+/* Finish the current run by adding the run totals to the segment and message
+ * length. */
+static void accumulate(upb_pb_encoder *e) {
+  size_t run_len;
+  assert(e->ptr >= e->runbegin);
+  run_len = e->ptr - e->runbegin;
+  e->segptr->seglen += run_len;
+  top(e)->msglen += run_len;
+  e->runbegin = e->ptr;
+}
+
+/* Call to indicate the start of delimited region for which the full length is
+ * not yet known.  All data will be buffered until the length is known.
+ * Delimited regions may be nested; their lengths will all be tracked properly. */
+static bool start_delim(upb_pb_encoder *e) {
+  if (e->top) {
+    /* We are already buffering, advance to the next segment and push it on the
+     * stack. */
+    accumulate(e);
+
+    if (++e->top == e->stacklimit) {
+      /* TODO(haberman): grow stack? */
+      return false;
+    }
+
+    if (++e->segptr == e->seglimit) {
+      /* Grow segment buffer. */
+      size_t old_size =
+          (e->seglimit - e->segbuf) * sizeof(upb_pb_encoder_segment);
+      size_t new_size = old_size * 2;
+      upb_pb_encoder_segment *new_buf =
+          upb_env_realloc(e->env, e->segbuf, old_size, new_size);
+
+      if (new_buf == NULL) {
+        return false;
+      }
+
+      e->segptr = new_buf + (e->segptr - e->segbuf);
+      e->seglimit = new_buf + (new_size / sizeof(upb_pb_encoder_segment));
+      e->segbuf = new_buf;
+    }
+  } else {
+    /* We were previously at the top level, start buffering. */
+    e->segptr = e->segbuf;
+    e->top = e->stack;
+    e->runbegin = e->ptr;
+  }
+
+  *e->top = e->segptr - e->segbuf;
+  e->segptr->seglen = 0;
+  e->segptr->msglen = 0;
+
+  return true;
+}
+
+/* Call to indicate the end of a delimited region.  We now know the length of
+ * the delimited region.  If we are not nested inside any other delimited
+ * regions, we can now emit all of the buffered data we accumulated. */
+static bool end_delim(upb_pb_encoder *e) {
+  size_t msglen;
+  accumulate(e);
+  msglen = top(e)->msglen;
+
+  if (e->top == e->stack) {
+    /* All lengths are now available, emit all buffered data. */
+    char buf[UPB_PB_VARINT_MAX_LEN];
+    upb_pb_encoder_segment *s;
+    const char *ptr = e->buf;
+    for (s = e->segbuf; s <= e->segptr; s++) {
+      size_t lenbytes = upb_vencode64(s->msglen, buf);
+      putbuf(e, buf, lenbytes);
+      putbuf(e, ptr, s->seglen);
+      ptr += s->seglen;
+    }
+
+    e->ptr = e->buf;
+    e->top = NULL;
+  } else {
+    /* Need to keep buffering; propagate length info into enclosing
+     * submessages. */
+    --e->top;
+    top(e)->msglen += msglen + upb_varint_size(msglen);
+  }
+
+  return true;
+}
+
+
+/* tag_t **********************************************************************/
+
+/* A precomputed (pre-encoded) tag and length. */
+
+typedef struct {
+  uint8_t bytes;
+  char tag[7];
+} tag_t;
+
+/* Allocates a new tag for this field, and sets it in these handlerattr. */
+static void new_tag(upb_handlers *h, const upb_fielddef *f, upb_wiretype_t wt,
+                    upb_handlerattr *attr) {
+  uint32_t n = upb_fielddef_number(f);
+
+  tag_t *tag = malloc(sizeof(tag_t));
+  tag->bytes = upb_vencode64((n << 3) | wt, tag->tag);
+
+  upb_handlerattr_init(attr);
+  upb_handlerattr_sethandlerdata(attr, tag);
+  upb_handlers_addcleanup(h, tag, free);
+}
+
+static bool encode_tag(upb_pb_encoder *e, const tag_t *tag) {
+  return encode_bytes(e, tag->tag, tag->bytes);
+}
+
+
+/* encoding of wire types *****************************************************/
+
+static bool encode_fixed64(upb_pb_encoder *e, uint64_t val) {
+  /* TODO(haberman): byte-swap for big endian. */
+  return encode_bytes(e, &val, sizeof(uint64_t));
+}
+
+static bool encode_fixed32(upb_pb_encoder *e, uint32_t val) {
+  /* TODO(haberman): byte-swap for big endian. */
+  return encode_bytes(e, &val, sizeof(uint32_t));
+}
+
+static bool encode_varint(upb_pb_encoder *e, uint64_t val) {
+  if (!reserve(e, UPB_PB_VARINT_MAX_LEN)) {
+    return false;
+  }
+
+  encoder_advance(e, upb_vencode64(val, e->ptr));
+  return true;
+}
+
+static uint64_t dbl2uint64(double d) {
+  uint64_t ret;
+  memcpy(&ret, &d, sizeof(uint64_t));
+  return ret;
+}
+
+static uint32_t flt2uint32(float d) {
+  uint32_t ret;
+  memcpy(&ret, &d, sizeof(uint32_t));
+  return ret;
+}
+
+
+/* encoding of proto types ****************************************************/
+
+static bool startmsg(void *c, const void *hd) {
+  upb_pb_encoder *e = c;
+  UPB_UNUSED(hd);
+  if (e->depth++ == 0) {
+    upb_bytessink_start(e->output_, 0, &e->subc);
+  }
+  return true;
+}
+
+static bool endmsg(void *c, const void *hd, upb_status *status) {
+  upb_pb_encoder *e = c;
+  UPB_UNUSED(hd);
+  UPB_UNUSED(status);
+  if (--e->depth == 0) {
+    upb_bytessink_end(e->output_);
+  }
+  return true;
+}
+
+static void *encode_startdelimfield(void *c, const void *hd) {
+  bool ok = encode_tag(c, hd) && commit(c) && start_delim(c);
+  return ok ? c : UPB_BREAK;
+}
+
+static bool encode_enddelimfield(void *c, const void *hd) {
+  UPB_UNUSED(hd);
+  return end_delim(c);
+}
+
+static void *encode_startgroup(void *c, const void *hd) {
+  return (encode_tag(c, hd) && commit(c)) ? c : UPB_BREAK;
+}
+
+static bool encode_endgroup(void *c, const void *hd) {
+  return encode_tag(c, hd) && commit(c);
+}
+
+static void *encode_startstr(void *c, const void *hd, size_t size_hint) {
+  UPB_UNUSED(size_hint);
+  return encode_startdelimfield(c, hd);
+}
+
+static size_t encode_strbuf(void *c, const void *hd, const char *buf,
+                            size_t len, const upb_bufhandle *h) {
+  UPB_UNUSED(hd);
+  UPB_UNUSED(h);
+  return encode_bytes(c, buf, len) ? len : 0;
+}
+
+#define T(type, ctype, convert, encode)                                  \
+  static bool encode_scalar_##type(void *e, const void *hd, ctype val) { \
+    return encode_tag(e, hd) && encode(e, (convert)(val)) && commit(e);  \
+  }                                                                      \
+  static bool encode_packed_##type(void *e, const void *hd, ctype val) { \
+    UPB_UNUSED(hd);                                                      \
+    return encode(e, (convert)(val));                                    \
+  }
+
+T(double,   double,   dbl2uint64,   encode_fixed64)
+T(float,    float,    flt2uint32,   encode_fixed32)
+T(int64,    int64_t,  uint64_t,     encode_varint)
+T(int32,    int32_t,  uint32_t,     encode_varint)
+T(fixed64,  uint64_t, uint64_t,     encode_fixed64)
+T(fixed32,  uint32_t, uint32_t,     encode_fixed32)
+T(bool,     bool,     bool,         encode_varint)
+T(uint32,   uint32_t, uint32_t,     encode_varint)
+T(uint64,   uint64_t, uint64_t,     encode_varint)
+T(enum,     int32_t,  uint32_t,     encode_varint)
+T(sfixed32, int32_t,  uint32_t,     encode_fixed32)
+T(sfixed64, int64_t,  uint64_t,     encode_fixed64)
+T(sint32,   int32_t,  upb_zzenc_32, encode_varint)
+T(sint64,   int64_t,  upb_zzenc_64, encode_varint)
+
+#undef T
+
+
+/* code to build the handlers *************************************************/
+
+static void newhandlers_callback(const void *closure, upb_handlers *h) {
+  const upb_msgdef *m;
+  upb_msg_field_iter i;
+
+  UPB_UNUSED(closure);
+
+  upb_handlers_setstartmsg(h, startmsg, NULL);
+  upb_handlers_setendmsg(h, endmsg, NULL);
+
+  m = upb_handlers_msgdef(h);
+  for(upb_msg_field_begin(&i, m);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    const upb_fielddef *f = upb_msg_iter_field(&i);
+    bool packed = upb_fielddef_isseq(f) && upb_fielddef_isprimitive(f) &&
+                  upb_fielddef_packed(f);
+    upb_handlerattr attr;
+    upb_wiretype_t wt =
+        packed ? UPB_WIRE_TYPE_DELIMITED
+               : upb_pb_native_wire_types[upb_fielddef_descriptortype(f)];
+
+    /* Pre-encode the tag for this field. */
+    new_tag(h, f, wt, &attr);
+
+    if (packed) {
+      upb_handlers_setstartseq(h, f, encode_startdelimfield, &attr);
+      upb_handlers_setendseq(h, f, encode_enddelimfield, &attr);
+    }
+
+#define T(upper, lower, upbtype)                                     \
+  case UPB_DESCRIPTOR_TYPE_##upper:                                  \
+    if (packed) {                                                    \
+      upb_handlers_set##upbtype(h, f, encode_packed_##lower, &attr); \
+    } else {                                                         \
+      upb_handlers_set##upbtype(h, f, encode_scalar_##lower, &attr); \
+    }                                                                \
+    break;
+
+    switch (upb_fielddef_descriptortype(f)) {
+      T(DOUBLE,   double,   double);
+      T(FLOAT,    float,    float);
+      T(INT64,    int64,    int64);
+      T(INT32,    int32,    int32);
+      T(FIXED64,  fixed64,  uint64);
+      T(FIXED32,  fixed32,  uint32);
+      T(BOOL,     bool,     bool);
+      T(UINT32,   uint32,   uint32);
+      T(UINT64,   uint64,   uint64);
+      T(ENUM,     enum,     int32);
+      T(SFIXED32, sfixed32, int32);
+      T(SFIXED64, sfixed64, int64);
+      T(SINT32,   sint32,   int32);
+      T(SINT64,   sint64,   int64);
+      case UPB_DESCRIPTOR_TYPE_STRING:
+      case UPB_DESCRIPTOR_TYPE_BYTES:
+        upb_handlers_setstartstr(h, f, encode_startstr, &attr);
+        upb_handlers_setendstr(h, f, encode_enddelimfield, &attr);
+        upb_handlers_setstring(h, f, encode_strbuf, &attr);
+        break;
+      case UPB_DESCRIPTOR_TYPE_MESSAGE:
+        upb_handlers_setstartsubmsg(h, f, encode_startdelimfield, &attr);
+        upb_handlers_setendsubmsg(h, f, encode_enddelimfield, &attr);
+        break;
+      case UPB_DESCRIPTOR_TYPE_GROUP: {
+        /* Endgroup takes a different tag (wire_type = END_GROUP). */
+        upb_handlerattr attr2;
+        new_tag(h, f, UPB_WIRE_TYPE_END_GROUP, &attr2);
+
+        upb_handlers_setstartsubmsg(h, f, encode_startgroup, &attr);
+        upb_handlers_setendsubmsg(h, f, encode_endgroup, &attr2);
+
+        upb_handlerattr_uninit(&attr2);
+        break;
+      }
+    }
+
+#undef T
+
+    upb_handlerattr_uninit(&attr);
+  }
+}
+
+void upb_pb_encoder_reset(upb_pb_encoder *e) {
+  e->segptr = NULL;
+  e->top = NULL;
+  e->depth = 0;
+}
+
+
+/* public API *****************************************************************/
+
+const upb_handlers *upb_pb_encoder_newhandlers(const upb_msgdef *m,
+                                               const void *owner) {
+  return upb_handlers_newfrozen(m, owner, newhandlers_callback, NULL);
+}
+
+upb_pb_encoder *upb_pb_encoder_create(upb_env *env, const upb_handlers *h,
+                                      upb_bytessink *output) {
+  const size_t initial_bufsize = 256;
+  const size_t initial_segbufsize = 16;
+  /* TODO(haberman): make this configurable. */
+  const size_t stack_size = 64;
+#ifndef NDEBUG
+  const size_t size_before = upb_env_bytesallocated(env);
+#endif
+
+  upb_pb_encoder *e = upb_env_malloc(env, sizeof(upb_pb_encoder));
+  if (!e) return NULL;
+
+  e->buf = upb_env_malloc(env, initial_bufsize);
+  e->segbuf = upb_env_malloc(env, initial_segbufsize * sizeof(*e->segbuf));
+  e->stack = upb_env_malloc(env, stack_size * sizeof(*e->stack));
+
+  if (!e->buf || !e->segbuf || !e->stack) {
+    return NULL;
+  }
+
+  e->limit = e->buf + initial_bufsize;
+  e->seglimit = e->segbuf + initial_segbufsize;
+  e->stacklimit = e->stack + stack_size;
+
+  upb_pb_encoder_reset(e);
+  upb_sink_reset(&e->input_, h, e);
+
+  e->env = env;
+  e->output_ = output;
+  e->subc = output->closure;
+  e->ptr = e->buf;
+
+  /* If this fails, increase the value in encoder.h. */
+  assert(upb_env_bytesallocated(env) - size_before <= UPB_PB_ENCODER_SIZE);
+  return e;
+}
+
+upb_sink *upb_pb_encoder_input(upb_pb_encoder *e) { return &e->input_; }
+
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+upb_def **upb_load_defs_from_descriptor(const char *str, size_t len, int *n,
+                                        void *owner, upb_status *status) {
+  /* Create handlers. */
+  const upb_pbdecodermethod *decoder_m;
+  const upb_handlers *reader_h = upb_descreader_newhandlers(&reader_h);
+  upb_env env;
+  upb_pbdecodermethodopts opts;
+  upb_pbdecoder *decoder;
+  upb_descreader *reader;
+  bool ok;
+  upb_def **ret = NULL;
+  upb_def **defs;
+
+  upb_pbdecodermethodopts_init(&opts, reader_h);
+  decoder_m = upb_pbdecodermethod_new(&opts, &decoder_m);
+
+  upb_env_init(&env);
+  upb_env_reporterrorsto(&env, status);
+
+  reader = upb_descreader_create(&env, reader_h);
+  decoder = upb_pbdecoder_create(&env, decoder_m, upb_descreader_input(reader));
+
+  /* Push input data. */
+  ok = upb_bufsrc_putbuf(str, len, upb_pbdecoder_input(decoder));
+
+  if (!ok) goto cleanup;
+  defs = upb_descreader_getdefs(reader, owner, n);
+  ret = malloc(sizeof(upb_def*) * (*n));
+  memcpy(ret, defs, sizeof(upb_def*) * (*n));
+
+cleanup:
+  upb_env_uninit(&env);
+  upb_handlers_unref(reader_h, &reader_h);
+  upb_pbdecodermethod_unref(decoder_m, &decoder_m);
+  return ret;
+}
+
+bool upb_load_descriptor_into_symtab(upb_symtab *s, const char *str, size_t len,
+                                     upb_status *status) {
+  int n;
+  bool success;
+  upb_def **defs = upb_load_defs_from_descriptor(str, len, &n, &defs, status);
+  if (!defs) return false;
+  success = upb_symtab_add(s, defs, n, &defs, status);
+  free(defs);
+  return success;
+}
+
+char *upb_readfile(const char *filename, size_t *len) {
+  long size;
+  char *buf;
+  FILE *f = fopen(filename, "rb");
+  if(!f) return NULL;
+  if(fseek(f, 0, SEEK_END) != 0) goto error;
+  size = ftell(f);
+  if(size < 0) goto error;
+  if(fseek(f, 0, SEEK_SET) != 0) goto error;
+  buf = malloc(size + 1);
+  if(size && fread(buf, size, 1, f) != 1) goto error;
+  fclose(f);
+  if (len) *len = size;
+  return buf;
+
+error:
+  fclose(f);
+  return NULL;
+}
+
+bool upb_load_descriptor_file_into_symtab(upb_symtab *symtab, const char *fname,
+                                          upb_status *status) {
+  size_t len;
+  bool success;
+  char *data = upb_readfile(fname, &len);
+  if (!data) {
+    if (status) upb_status_seterrf(status, "Couldn't read file: %s", fname);
+    return false;
+  }
+  success = upb_load_descriptor_into_symtab(symtab, data, len, status);
+  free(data);
+  return success;
+}
+/*
+ * upb::pb::TextPrinter
+ *
+ * OPT: This is not optimized at all.  It uses printf() which parses the format
+ * string every time, and it allocates memory for every put.
+ */
+
+
+#include <ctype.h>
+#include <float.h>
+#include <inttypes.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+
+struct upb_textprinter {
+  upb_sink input_;
+  upb_bytessink *output_;
+  int indent_depth_;
+  bool single_line_;
+  void *subc;
+};
+
+#define CHECK(x) if ((x) < 0) goto err;
+
+static const char *shortname(const char *longname) {
+  const char *last = strrchr(longname, '.');
+  return last ? last + 1 : longname;
+}
+
+static int indent(upb_textprinter *p) {
+  int i;
+  if (!p->single_line_)
+    for (i = 0; i < p->indent_depth_; i++)
+      upb_bytessink_putbuf(p->output_, p->subc, "  ", 2, NULL);
+  return 0;
+}
+
+static int endfield(upb_textprinter *p) {
+  const char ch = (p->single_line_ ? ' ' : '\n');
+  upb_bytessink_putbuf(p->output_, p->subc, &ch, 1, NULL);
+  return 0;
+}
+
+static int putescaped(upb_textprinter *p, const char *buf, size_t len,
+                      bool preserve_utf8) {
+  /* Based on CEscapeInternal() from Google's protobuf release. */
+  char dstbuf[4096], *dst = dstbuf, *dstend = dstbuf + sizeof(dstbuf);
+  const char *end = buf + len;
+
+  /* I think hex is prettier and more useful, but proto2 uses octal; should
+   * investigate whether it can parse hex also. */
+  const bool use_hex = false;
+  bool last_hex_escape = false; /* true if last output char was \xNN */
+
+  for (; buf < end; buf++) {
+    bool is_hex_escape;
+
+    if (dstend - dst < 4) {
+      upb_bytessink_putbuf(p->output_, p->subc, dstbuf, dst - dstbuf, NULL);
+      dst = dstbuf;
+    }
+
+    is_hex_escape = false;
+    switch (*buf) {
+      case '\n': *(dst++) = '\\'; *(dst++) = 'n';  break;
+      case '\r': *(dst++) = '\\'; *(dst++) = 'r';  break;
+      case '\t': *(dst++) = '\\'; *(dst++) = 't';  break;
+      case '\"': *(dst++) = '\\'; *(dst++) = '\"'; break;
+      case '\'': *(dst++) = '\\'; *(dst++) = '\''; break;
+      case '\\': *(dst++) = '\\'; *(dst++) = '\\'; break;
+      default:
+        /* Note that if we emit \xNN and the buf character after that is a hex
+         * digit then that digit must be escaped too to prevent it being
+         * interpreted as part of the character code by C. */
+        if ((!preserve_utf8 || (uint8_t)*buf < 0x80) &&
+            (!isprint(*buf) || (last_hex_escape && isxdigit(*buf)))) {
+          sprintf(dst, (use_hex ? "\\x%02x" : "\\%03o"), (uint8_t)*buf);
+          is_hex_escape = use_hex;
+          dst += 4;
+        } else {
+          *(dst++) = *buf; break;
+        }
+    }
+    last_hex_escape = is_hex_escape;
+  }
+  /* Flush remaining data. */
+  upb_bytessink_putbuf(p->output_, p->subc, dstbuf, dst - dstbuf, NULL);
+  return 0;
+}
+
+bool putf(upb_textprinter *p, const char *fmt, ...) {
+  va_list args;
+  va_list args_copy;
+  char *str;
+  int written;
+  int len;
+  bool ok;
+
+  va_start(args, fmt);
+
+  /* Run once to get the length of the string. */
+  _upb_va_copy(args_copy, args);
+  len = _upb_vsnprintf(NULL, 0, fmt, args_copy);
+  va_end(args_copy);
+
+  /* + 1 for NULL terminator (vsprintf() requires it even if we don't). */
+  str = malloc(len + 1);
+  if (!str) return false;
+  written = vsprintf(str, fmt, args);
+  va_end(args);
+  UPB_ASSERT_VAR(written, written == len);
+
+  ok = upb_bytessink_putbuf(p->output_, p->subc, str, len, NULL);
+  free(str);
+  return ok;
+}
+
+
+/* handlers *******************************************************************/
+
+static bool textprinter_startmsg(void *c, const void *hd) {
+  upb_textprinter *p = c;
+  UPB_UNUSED(hd);
+  if (p->indent_depth_ == 0) {
+    upb_bytessink_start(p->output_, 0, &p->subc);
+  }
+  return true;
+}
+
+static bool textprinter_endmsg(void *c, const void *hd, upb_status *s) {
+  upb_textprinter *p = c;
+  UPB_UNUSED(hd);
+  UPB_UNUSED(s);
+  if (p->indent_depth_ == 0) {
+    upb_bytessink_end(p->output_);
+  }
+  return true;
+}
+
+#define TYPE(name, ctype, fmt) \
+  static bool textprinter_put ## name(void *closure, const void *handler_data, \
+                                      ctype val) {                             \
+    upb_textprinter *p = closure;                                              \
+    const upb_fielddef *f = handler_data;                                      \
+    CHECK(indent(p));                                                          \
+    putf(p, "%s: " fmt, upb_fielddef_name(f), val);                            \
+    CHECK(endfield(p));                                                        \
+    return true;                                                               \
+  err:                                                                         \
+    return false;                                                              \
+}
+
+static bool textprinter_putbool(void *closure, const void *handler_data,
+                                bool val) {
+  upb_textprinter *p = closure;
+  const upb_fielddef *f = handler_data;
+  CHECK(indent(p));
+  putf(p, "%s: %s", upb_fielddef_name(f), val ? "true" : "false");
+  CHECK(endfield(p));
+  return true;
+err:
+  return false;
+}
+
+#define STRINGIFY_HELPER(x) #x
+#define STRINGIFY_MACROVAL(x) STRINGIFY_HELPER(x)
+
+TYPE(int32,  int32_t,  "%" PRId32)
+TYPE(int64,  int64_t,  "%" PRId64)
+TYPE(uint32, uint32_t, "%" PRIu32)
+TYPE(uint64, uint64_t, "%" PRIu64)
+TYPE(float,  float,    "%." STRINGIFY_MACROVAL(FLT_DIG) "g")
+TYPE(double, double,   "%." STRINGIFY_MACROVAL(DBL_DIG) "g")
+
+#undef TYPE
+
+/* Output a symbolic value from the enum if found, else just print as int32. */
+static bool textprinter_putenum(void *closure, const void *handler_data,
+                                int32_t val) {
+  upb_textprinter *p = closure;
+  const upb_fielddef *f = handler_data;
+  const upb_enumdef *enum_def = upb_downcast_enumdef(upb_fielddef_subdef(f));
+  const char *label = upb_enumdef_iton(enum_def, val);
+  if (label) {
+    indent(p);
+    putf(p, "%s: %s", upb_fielddef_name(f), label);
+    endfield(p);
+  } else {
+    if (!textprinter_putint32(closure, handler_data, val))
+      return false;
+  }
+  return true;
+}
+
+static void *textprinter_startstr(void *closure, const void *handler_data,
+                      size_t size_hint) {
+  upb_textprinter *p = closure;
+  const upb_fielddef *f = handler_data;
+  UPB_UNUSED(size_hint);
+  indent(p);
+  putf(p, "%s: \"", upb_fielddef_name(f));
+  return p;
+}
+
+static bool textprinter_endstr(void *closure, const void *handler_data) {
+  upb_textprinter *p = closure;
+  UPB_UNUSED(handler_data);
+  putf(p, "\"");
+  endfield(p);
+  return true;
+}
+
+static size_t textprinter_putstr(void *closure, const void *hd, const char *buf,
+                                 size_t len, const upb_bufhandle *handle) {
+  upb_textprinter *p = closure;
+  const upb_fielddef *f = hd;
+  UPB_UNUSED(handle);
+  CHECK(putescaped(p, buf, len, upb_fielddef_type(f) == UPB_TYPE_STRING));
+  return len;
+err:
+  return 0;
+}
+
+static void *textprinter_startsubmsg(void *closure, const void *handler_data) {
+  upb_textprinter *p = closure;
+  const char *name = handler_data;
+  CHECK(indent(p));
+  putf(p, "%s {%c", name, p->single_line_ ? ' ' : '\n');
+  p->indent_depth_++;
+  return p;
+err:
+  return UPB_BREAK;
+}
+
+static bool textprinter_endsubmsg(void *closure, const void *handler_data) {
+  upb_textprinter *p = closure;
+  UPB_UNUSED(handler_data);
+  p->indent_depth_--;
+  CHECK(indent(p));
+  upb_bytessink_putbuf(p->output_, p->subc, "}", 1, NULL);
+  CHECK(endfield(p));
+  return true;
+err:
+  return false;
+}
+
+static void onmreg(const void *c, upb_handlers *h) {
+  const upb_msgdef *m = upb_handlers_msgdef(h);
+  upb_msg_field_iter i;
+  UPB_UNUSED(c);
+
+  upb_handlers_setstartmsg(h, textprinter_startmsg, NULL);
+  upb_handlers_setendmsg(h, textprinter_endmsg, NULL);
+
+  for(upb_msg_field_begin(&i, m);
+      !upb_msg_field_done(&i);
+      upb_msg_field_next(&i)) {
+    upb_fielddef *f = upb_msg_iter_field(&i);
+    upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER;
+    upb_handlerattr_sethandlerdata(&attr, f);
+    switch (upb_fielddef_type(f)) {
+      case UPB_TYPE_INT32:
+        upb_handlers_setint32(h, f, textprinter_putint32, &attr);
+        break;
+      case UPB_TYPE_INT64:
+        upb_handlers_setint64(h, f, textprinter_putint64, &attr);
+        break;
+      case UPB_TYPE_UINT32:
+        upb_handlers_setuint32(h, f, textprinter_putuint32, &attr);
+        break;
+      case UPB_TYPE_UINT64:
+        upb_handlers_setuint64(h, f, textprinter_putuint64, &attr);
+        break;
+      case UPB_TYPE_FLOAT:
+        upb_handlers_setfloat(h, f, textprinter_putfloat, &attr);
+        break;
+      case UPB_TYPE_DOUBLE:
+        upb_handlers_setdouble(h, f, textprinter_putdouble, &attr);
+        break;
+      case UPB_TYPE_BOOL:
+        upb_handlers_setbool(h, f, textprinter_putbool, &attr);
+        break;
+      case UPB_TYPE_STRING:
+      case UPB_TYPE_BYTES:
+        upb_handlers_setstartstr(h, f, textprinter_startstr, &attr);
+        upb_handlers_setstring(h, f, textprinter_putstr, &attr);
+        upb_handlers_setendstr(h, f, textprinter_endstr, &attr);
+        break;
+      case UPB_TYPE_MESSAGE: {
+        const char *name =
+            upb_fielddef_istagdelim(f)
+                ? shortname(upb_msgdef_fullname(upb_fielddef_msgsubdef(f)))
+                : upb_fielddef_name(f);
+        upb_handlerattr_sethandlerdata(&attr, name);
+        upb_handlers_setstartsubmsg(h, f, textprinter_startsubmsg, &attr);
+        upb_handlers_setendsubmsg(h, f, textprinter_endsubmsg, &attr);
+        break;
+      }
+      case UPB_TYPE_ENUM:
+        upb_handlers_setint32(h, f, textprinter_putenum, &attr);
+        break;
+    }
+  }
+}
+
+static void textprinter_reset(upb_textprinter *p, bool single_line) {
+  p->single_line_ = single_line;
+  p->indent_depth_ = 0;
+}
+
+
+/* Public API *****************************************************************/
+
+upb_textprinter *upb_textprinter_create(upb_env *env, const upb_handlers *h,
+                                        upb_bytessink *output) {
+  upb_textprinter *p = upb_env_malloc(env, sizeof(upb_textprinter));
+  if (!p) return NULL;
+
+  p->output_ = output;
+  upb_sink_reset(&p->input_, h, p);
+  textprinter_reset(p, false);
+
+  return p;
+}
+
+const upb_handlers *upb_textprinter_newhandlers(const upb_msgdef *m,
+                                                const void *owner) {
+  return upb_handlers_newfrozen(m, owner, &onmreg, NULL);
+}
+
+upb_sink *upb_textprinter_input(upb_textprinter *p) { return &p->input_; }
+
+void upb_textprinter_setsingleline(upb_textprinter *p, bool single_line) {
+  p->single_line_ = single_line;
+}
+
+
+/* Index is descriptor type. */
+const uint8_t upb_pb_native_wire_types[] = {
+  UPB_WIRE_TYPE_END_GROUP,     /* ENDGROUP */
+  UPB_WIRE_TYPE_64BIT,         /* DOUBLE */
+  UPB_WIRE_TYPE_32BIT,         /* FLOAT */
+  UPB_WIRE_TYPE_VARINT,        /* INT64 */
+  UPB_WIRE_TYPE_VARINT,        /* UINT64 */
+  UPB_WIRE_TYPE_VARINT,        /* INT32 */
+  UPB_WIRE_TYPE_64BIT,         /* FIXED64 */
+  UPB_WIRE_TYPE_32BIT,         /* FIXED32 */
+  UPB_WIRE_TYPE_VARINT,        /* BOOL */
+  UPB_WIRE_TYPE_DELIMITED,     /* STRING */
+  UPB_WIRE_TYPE_START_GROUP,   /* GROUP */
+  UPB_WIRE_TYPE_DELIMITED,     /* MESSAGE */
+  UPB_WIRE_TYPE_DELIMITED,     /* BYTES */
+  UPB_WIRE_TYPE_VARINT,        /* UINT32 */
+  UPB_WIRE_TYPE_VARINT,        /* ENUM */
+  UPB_WIRE_TYPE_32BIT,         /* SFIXED32 */
+  UPB_WIRE_TYPE_64BIT,         /* SFIXED64 */
+  UPB_WIRE_TYPE_VARINT,        /* SINT32 */
+  UPB_WIRE_TYPE_VARINT,        /* SINT64 */
+};
+
+/* A basic branch-based decoder, uses 32-bit values to get good performance
+ * on 32-bit architectures (but performs well on 64-bits also).
+ * This scheme comes from the original Google Protobuf implementation
+ * (proto2). */
+upb_decoderet upb_vdecode_max8_branch32(upb_decoderet r) {
+  upb_decoderet err = {NULL, 0};
+  const char *p = r.p;
+  uint32_t low = (uint32_t)r.val;
+  uint32_t high = 0;
+  uint32_t b;
+  b = *(p++); low  |= (b & 0x7fU) << 14; if (!(b & 0x80)) goto done;
+  b = *(p++); low  |= (b & 0x7fU) << 21; if (!(b & 0x80)) goto done;
+  b = *(p++); low  |= (b & 0x7fU) << 28;
+              high  = (b & 0x7fU) >>  4; if (!(b & 0x80)) goto done;
+  b = *(p++); high |= (b & 0x7fU) <<  3; if (!(b & 0x80)) goto done;
+  b = *(p++); high |= (b & 0x7fU) << 10; if (!(b & 0x80)) goto done;
+  b = *(p++); high |= (b & 0x7fU) << 17; if (!(b & 0x80)) goto done;
+  b = *(p++); high |= (b & 0x7fU) << 24; if (!(b & 0x80)) goto done;
+  b = *(p++); high |= (b & 0x7fU) << 31; if (!(b & 0x80)) goto done;
+  return err;
+
+done:
+  r.val = ((uint64_t)high << 32) | low;
+  r.p = p;
+  return r;
+}
+
+/* Like the previous, but uses 64-bit values. */
+upb_decoderet upb_vdecode_max8_branch64(upb_decoderet r) {
+  const char *p = r.p;
+  uint64_t val = r.val;
+  uint64_t b;
+  upb_decoderet err = {NULL, 0};
+  b = *(p++); val |= (b & 0x7fU) << 14; if (!(b & 0x80)) goto done;
+  b = *(p++); val |= (b & 0x7fU) << 21; if (!(b & 0x80)) goto done;
+  b = *(p++); val |= (b & 0x7fU) << 28; if (!(b & 0x80)) goto done;
+  b = *(p++); val |= (b & 0x7fU) << 35; if (!(b & 0x80)) goto done;
+  b = *(p++); val |= (b & 0x7fU) << 42; if (!(b & 0x80)) goto done;
+  b = *(p++); val |= (b & 0x7fU) << 49; if (!(b & 0x80)) goto done;
+  b = *(p++); val |= (b & 0x7fU) << 56; if (!(b & 0x80)) goto done;
+  b = *(p++); val |= (b & 0x7fU) << 63; if (!(b & 0x80)) goto done;
+  return err;
+
+done:
+  r.val = val;
+  r.p = p;
+  return r;
+}
+
+/* Given an encoded varint v, returns an integer with a single bit set that
+ * indicates the end of the varint.  Subtracting one from this value will
+ * yield a mask that leaves only bits that are part of the varint.  Returns
+ * 0 if the varint is unterminated. */
+static uint64_t upb_get_vstopbit(uint64_t v) {
+  uint64_t cbits = v | 0x7f7f7f7f7f7f7f7fULL;
+  return ~cbits & (cbits+1);
+}
+
+/* A branchless decoder.  Credit to Pascal Massimino for the bit-twiddling. */
+upb_decoderet upb_vdecode_max8_massimino(upb_decoderet r) {
+  uint64_t b;
+  uint64_t stop_bit;
+  upb_decoderet my_r;
+  memcpy(&b, r.p, sizeof(b));
+  stop_bit = upb_get_vstopbit(b);
+  b =  (b & 0x7f7f7f7f7f7f7f7fULL) & (stop_bit - 1);
+  b +=       b & 0x007f007f007f007fULL;
+  b +=  3 * (b & 0x0000ffff0000ffffULL);
+  b += 15 * (b & 0x00000000ffffffffULL);
+  if (stop_bit == 0) {
+    /* Error: unterminated varint. */
+    upb_decoderet err_r = {(void*)0, 0};
+    return err_r;
+  }
+  my_r = upb_decoderet_make(r.p + ((__builtin_ctzll(stop_bit) + 1) / 8),
+                            r.val | (b << 7));
+  return my_r;
+}
+
+/* A branchless decoder.  Credit to Daniel Wright for the bit-twiddling. */
+upb_decoderet upb_vdecode_max8_wright(upb_decoderet r) {
+  uint64_t b;
+  uint64_t stop_bit;
+  upb_decoderet my_r;
+  memcpy(&b, r.p, sizeof(b));
+  stop_bit = upb_get_vstopbit(b);
+  b &= (stop_bit - 1);
+  b = ((b & 0x7f007f007f007f00ULL) >> 1) | (b & 0x007f007f007f007fULL);
+  b = ((b & 0xffff0000ffff0000ULL) >> 2) | (b & 0x0000ffff0000ffffULL);
+  b = ((b & 0xffffffff00000000ULL) >> 4) | (b & 0x00000000ffffffffULL);
+  if (stop_bit == 0) {
+    /* Error: unterminated varint. */
+    upb_decoderet err_r = {(void*)0, 0};
+    return err_r;
+  }
+  my_r = upb_decoderet_make(r.p + ((__builtin_ctzll(stop_bit) + 1) / 8),
+                            r.val | (b << 14));
+  return my_r;
+}
+
+#line 1 "upb/json/parser.rl"
+/*
+** upb::json::Parser (upb_json_parser)
+**
+** A parser that uses the Ragel State Machine Compiler to generate
+** the finite automata.
+**
+** Ragel only natively handles regular languages, but we can manually
+** program it a bit to handle context-free languages like JSON, by using
+** the "fcall" and "fret" constructs.
+**
+** This parser can handle the basics, but needs several things to be fleshed
+** out:
+**
+** - handling of unicode escape sequences (including high surrogate pairs).
+** - properly check and report errors for unknown fields, stack overflow,
+**   improper array nesting (or lack of nesting).
+** - handling of base64 sequences with padding characters.
+** - handling of push-back (non-success returns from sink functions).
+** - handling of keys/escape-sequences/etc that span input buffers.
+*/
+
+#include <stdio.h>
+#include <stdint.h>
+#include <assert.h>
+#include <string.h>
+#include <stdlib.h>
+#include <errno.h>
+
+
+#define UPB_JSON_MAX_DEPTH 64
+
+typedef struct {
+  upb_sink sink;
+
+  /* The current message in which we're parsing, and the field whose value we're
+   * expecting next. */
+  const upb_msgdef *m;
+  const upb_fielddef *f;
+
+  /* We are in a repeated-field context, ready to emit mapentries as
+   * submessages. This flag alters the start-of-object (open-brace) behavior to
+   * begin a sequence of mapentry messages rather than a single submessage. */
+  bool is_map;
+
+  /* We are in a map-entry message context. This flag is set when parsing the
+   * value field of a single map entry and indicates to all value-field parsers
+   * (subobjects, strings, numbers, and bools) that the map-entry submessage
+   * should end as soon as the value is parsed. */
+  bool is_mapentry;
+
+  /* If |is_map| or |is_mapentry| is true, |mapfield| refers to the parent
+   * message's map field that we're currently parsing. This differs from |f|
+   * because |f| is the field in the *current* message (i.e., the map-entry
+   * message itself), not the parent's field that leads to this map. */
+  const upb_fielddef *mapfield;
+} upb_jsonparser_frame;
+
+struct upb_json_parser {
+  upb_env *env;
+  upb_byteshandler input_handler_;
+  upb_bytessink input_;
+
+  /* Stack to track the JSON scopes we are in. */
+  upb_jsonparser_frame stack[UPB_JSON_MAX_DEPTH];
+  upb_jsonparser_frame *top;
+  upb_jsonparser_frame *limit;
+
+  upb_status status;
+
+  /* Ragel's internal parsing stack for the parsing state machine. */
+  int current_state;
+  int parser_stack[UPB_JSON_MAX_DEPTH];
+  int parser_top;
+
+  /* The handle for the current buffer. */
+  const upb_bufhandle *handle;
+
+  /* Accumulate buffer.  See details in parser.rl. */
+  const char *accumulated;
+  size_t accumulated_len;
+  char *accumulate_buf;
+  size_t accumulate_buf_size;
+
+  /* Multi-part text data.  See details in parser.rl. */
+  int multipart_state;
+  upb_selector_t string_selector;
+
+  /* Input capture.  See details in parser.rl. */
+  const char *capture;
+
+  /* Intermediate result of parsing a unicode escape sequence. */
+  uint32_t digit;
+};
+
+#define PARSER_CHECK_RETURN(x) if (!(x)) return false
+
+/* Used to signal that a capture has been suspended. */
+static char suspend_capture;
+
+static upb_selector_t getsel_for_handlertype(upb_json_parser *p,
+                                             upb_handlertype_t type) {
+  upb_selector_t sel;
+  bool ok = upb_handlers_getselector(p->top->f, type, &sel);
+  UPB_ASSERT_VAR(ok, ok);
+  return sel;
+}
+
+static upb_selector_t parser_getsel(upb_json_parser *p) {
+  return getsel_for_handlertype(
+      p, upb_handlers_getprimitivehandlertype(p->top->f));
+}
+
+static bool check_stack(upb_json_parser *p) {
+  if ((p->top + 1) == p->limit) {
+    upb_status_seterrmsg(&p->status, "Nesting too deep");
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+
+  return true;
+}
+
+/* There are GCC/Clang built-ins for overflow checking which we could start
+ * using if there was any performance benefit to it. */
+
+static bool checked_add(size_t a, size_t b, size_t *c) {
+  if (SIZE_MAX - a < b) return false;
+  *c = a + b;
+  return true;
+}
+
+static size_t saturating_multiply(size_t a, size_t b) {
+  /* size_t is unsigned, so this is defined behavior even on overflow. */
+  size_t ret = a * b;
+  if (b != 0 && ret / b != a) {
+    ret = SIZE_MAX;
+  }
+  return ret;
+}
+
+
+/* Base64 decoding ************************************************************/
+
+/* TODO(haberman): make this streaming. */
+
+static const signed char b64table[] = {
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      62/*+*/, -1,      -1,      -1,      63/*/ */,
+  52/*0*/, 53/*1*/, 54/*2*/, 55/*3*/, 56/*4*/, 57/*5*/, 58/*6*/, 59/*7*/,
+  60/*8*/, 61/*9*/, -1,      -1,      -1,      -1,      -1,      -1,
+  -1,       0/*A*/,  1/*B*/,  2/*C*/,  3/*D*/,  4/*E*/,  5/*F*/,  6/*G*/,
+  07/*H*/,  8/*I*/,  9/*J*/, 10/*K*/, 11/*L*/, 12/*M*/, 13/*N*/, 14/*O*/,
+  15/*P*/, 16/*Q*/, 17/*R*/, 18/*S*/, 19/*T*/, 20/*U*/, 21/*V*/, 22/*W*/,
+  23/*X*/, 24/*Y*/, 25/*Z*/, -1,      -1,      -1,      -1,      -1,
+  -1,      26/*a*/, 27/*b*/, 28/*c*/, 29/*d*/, 30/*e*/, 31/*f*/, 32/*g*/,
+  33/*h*/, 34/*i*/, 35/*j*/, 36/*k*/, 37/*l*/, 38/*m*/, 39/*n*/, 40/*o*/,
+  41/*p*/, 42/*q*/, 43/*r*/, 44/*s*/, 45/*t*/, 46/*u*/, 47/*v*/, 48/*w*/,
+  49/*x*/, 50/*y*/, 51/*z*/, -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1,
+  -1,      -1,      -1,      -1,      -1,      -1,      -1,      -1
+};
+
+/* Returns the table value sign-extended to 32 bits.  Knowing that the upper
+ * bits will be 1 for unrecognized characters makes it easier to check for
+ * this error condition later (see below). */
+int32_t b64lookup(unsigned char ch) { return b64table[ch]; }
+
+/* Returns true if the given character is not a valid base64 character or
+ * padding. */
+bool nonbase64(unsigned char ch) { return b64lookup(ch) == -1 && ch != '='; }
+
+static bool base64_push(upb_json_parser *p, upb_selector_t sel, const char *ptr,
+                        size_t len) {
+  const char *limit = ptr + len;
+  for (; ptr < limit; ptr += 4) {
+    uint32_t val;
+    char output[3];
+
+    if (limit - ptr < 4) {
+      upb_status_seterrf(&p->status,
+                         "Base64 input for bytes field not a multiple of 4: %s",
+                         upb_fielddef_name(p->top->f));
+      upb_env_reporterror(p->env, &p->status);
+      return false;
+    }
+
+    val = b64lookup(ptr[0]) << 18 |
+          b64lookup(ptr[1]) << 12 |
+          b64lookup(ptr[2]) << 6  |
+          b64lookup(ptr[3]);
+
+    /* Test the upper bit; returns true if any of the characters returned -1. */
+    if (val & 0x80000000) {
+      goto otherchar;
+    }
+
+    output[0] = val >> 16;
+    output[1] = (val >> 8) & 0xff;
+    output[2] = val & 0xff;
+    upb_sink_putstring(&p->top->sink, sel, output, 3, NULL);
+  }
+  return true;
+
+otherchar:
+  if (nonbase64(ptr[0]) || nonbase64(ptr[1]) || nonbase64(ptr[2]) ||
+      nonbase64(ptr[3]) ) {
+    upb_status_seterrf(&p->status,
+                       "Non-base64 characters in bytes field: %s",
+                       upb_fielddef_name(p->top->f));
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  } if (ptr[2] == '=') {
+    uint32_t val;
+    char output;
+
+    /* Last group contains only two input bytes, one output byte. */
+    if (ptr[0] == '=' || ptr[1] == '=' || ptr[3] != '=') {
+      goto badpadding;
+    }
+
+    val = b64lookup(ptr[0]) << 18 |
+          b64lookup(ptr[1]) << 12;
+
+    assert(!(val & 0x80000000));
+    output = val >> 16;
+    upb_sink_putstring(&p->top->sink, sel, &output, 1, NULL);
+    return true;
+  } else {
+    uint32_t val;
+    char output[2];
+
+    /* Last group contains only three input bytes, two output bytes. */
+    if (ptr[0] == '=' || ptr[1] == '=' || ptr[2] == '=') {
+      goto badpadding;
+    }
+
+    val = b64lookup(ptr[0]) << 18 |
+          b64lookup(ptr[1]) << 12 |
+          b64lookup(ptr[2]) << 6;
+
+    output[0] = val >> 16;
+    output[1] = (val >> 8) & 0xff;
+    upb_sink_putstring(&p->top->sink, sel, output, 2, NULL);
+    return true;
+  }
+
+badpadding:
+  upb_status_seterrf(&p->status,
+                     "Incorrect base64 padding for field: %s (%.*s)",
+                     upb_fielddef_name(p->top->f),
+                     4, ptr);
+  upb_env_reporterror(p->env, &p->status);
+  return false;
+}
+
+
+/* Accumulate buffer **********************************************************/
+
+/* Functionality for accumulating a buffer.
+ *
+ * Some parts of the parser need an entire value as a contiguous string.  For
+ * example, to look up a member name in a hash table, or to turn a string into
+ * a number, the relevant library routines need the input string to be in
+ * contiguous memory, even if the value spanned two or more buffers in the
+ * input.  These routines handle that.
+ *
+ * In the common case we can just point to the input buffer to get this
+ * contiguous string and avoid any actual copy.  So we optimistically begin
+ * this way.  But there are a few cases where we must instead copy into a
+ * separate buffer:
+ *
+ *   1. The string was not contiguous in the input (it spanned buffers).
+ *
+ *   2. The string included escape sequences that need to be interpreted to get
+ *      the true value in a contiguous buffer. */
+
+static void assert_accumulate_empty(upb_json_parser *p) {
+  UPB_UNUSED(p);
+  assert(p->accumulated == NULL);
+  assert(p->accumulated_len == 0);
+}
+
+static void accumulate_clear(upb_json_parser *p) {
+  p->accumulated = NULL;
+  p->accumulated_len = 0;
+}
+
+/* Used internally by accumulate_append(). */
+static bool accumulate_realloc(upb_json_parser *p, size_t need) {
+  void *mem;
+  size_t old_size = p->accumulate_buf_size;
+  size_t new_size = UPB_MAX(old_size, 128);
+  while (new_size < need) {
+    new_size = saturating_multiply(new_size, 2);
+  }
+
+  mem = upb_env_realloc(p->env, p->accumulate_buf, old_size, new_size);
+  if (!mem) {
+    upb_status_seterrmsg(&p->status, "Out of memory allocating buffer.");
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+
+  p->accumulate_buf = mem;
+  p->accumulate_buf_size = new_size;
+  return true;
+}
+
+/* Logically appends the given data to the append buffer.
+ * If "can_alias" is true, we will try to avoid actually copying, but the buffer
+ * must be valid until the next accumulate_append() call (if any). */
+static bool accumulate_append(upb_json_parser *p, const char *buf, size_t len,
+                              bool can_alias) {
+  size_t need;
+
+  if (!p->accumulated && can_alias) {
+    p->accumulated = buf;
+    p->accumulated_len = len;
+    return true;
+  }
+
+  if (!checked_add(p->accumulated_len, len, &need)) {
+    upb_status_seterrmsg(&p->status, "Integer overflow.");
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+
+  if (need > p->accumulate_buf_size && !accumulate_realloc(p, need)) {
+    return false;
+  }
+
+  if (p->accumulated != p->accumulate_buf) {
+    memcpy(p->accumulate_buf, p->accumulated, p->accumulated_len);
+    p->accumulated = p->accumulate_buf;
+  }
+
+  memcpy(p->accumulate_buf + p->accumulated_len, buf, len);
+  p->accumulated_len += len;
+  return true;
+}
+
+/* Returns a pointer to the data accumulated since the last accumulate_clear()
+ * call, and writes the length to *len.  This with point either to the input
+ * buffer or a temporary accumulate buffer. */
+static const char *accumulate_getptr(upb_json_parser *p, size_t *len) {
+  assert(p->accumulated);
+  *len = p->accumulated_len;
+  return p->accumulated;
+}
+
+
+/* Mult-part text data ********************************************************/
+
+/* When we have text data in the input, it can often come in multiple segments.
+ * For example, there may be some raw string data followed by an escape
+ * sequence.  The two segments are processed with different logic.  Also buffer
+ * seams in the input can cause multiple segments.
+ *
+ * As we see segments, there are two main cases for how we want to process them:
+ *
+ *  1. we want to push the captured input directly to string handlers.
+ *
+ *  2. we need to accumulate all the parts into a contiguous buffer for further
+ *     processing (field name lookup, string->number conversion, etc). */
+
+/* This is the set of states for p->multipart_state. */
+enum {
+  /* We are not currently processing multipart data. */
+  MULTIPART_INACTIVE = 0,
+
+  /* We are processing multipart data by accumulating it into a contiguous
+   * buffer. */
+  MULTIPART_ACCUMULATE = 1,
+
+  /* We are processing multipart data by pushing each part directly to the
+   * current string handlers. */
+  MULTIPART_PUSHEAGERLY = 2
+};
+
+/* Start a multi-part text value where we accumulate the data for processing at
+ * the end. */
+static void multipart_startaccum(upb_json_parser *p) {
+  assert_accumulate_empty(p);
+  assert(p->multipart_state == MULTIPART_INACTIVE);
+  p->multipart_state = MULTIPART_ACCUMULATE;
+}
+
+/* Start a multi-part text value where we immediately push text data to a string
+ * value with the given selector. */
+static void multipart_start(upb_json_parser *p, upb_selector_t sel) {
+  assert_accumulate_empty(p);
+  assert(p->multipart_state == MULTIPART_INACTIVE);
+  p->multipart_state = MULTIPART_PUSHEAGERLY;
+  p->string_selector = sel;
+}
+
+static bool multipart_text(upb_json_parser *p, const char *buf, size_t len,
+                           bool can_alias) {
+  switch (p->multipart_state) {
+    case MULTIPART_INACTIVE:
+      upb_status_seterrmsg(
+          &p->status, "Internal error: unexpected state MULTIPART_INACTIVE");
+      upb_env_reporterror(p->env, &p->status);
+      return false;
+
+    case MULTIPART_ACCUMULATE:
+      if (!accumulate_append(p, buf, len, can_alias)) {
+        return false;
+      }
+      break;
+
+    case MULTIPART_PUSHEAGERLY: {
+      const upb_bufhandle *handle = can_alias ? p->handle : NULL;
+      upb_sink_putstring(&p->top->sink, p->string_selector, buf, len, handle);
+      break;
+    }
+  }
+
+  return true;
+}
+
+/* Note: this invalidates the accumulate buffer!  Call only after reading its
+ * contents. */
+static void multipart_end(upb_json_parser *p) {
+  assert(p->multipart_state != MULTIPART_INACTIVE);
+  p->multipart_state = MULTIPART_INACTIVE;
+  accumulate_clear(p);
+}
+
+
+/* Input capture **************************************************************/
+
+/* Functionality for capturing a region of the input as text.  Gracefully
+ * handles the case where a buffer seam occurs in the middle of the captured
+ * region. */
+
+static void capture_begin(upb_json_parser *p, const char *ptr) {
+  assert(p->multipart_state != MULTIPART_INACTIVE);
+  assert(p->capture == NULL);
+  p->capture = ptr;
+}
+
+static bool capture_end(upb_json_parser *p, const char *ptr) {
+  assert(p->capture);
+  if (multipart_text(p, p->capture, ptr - p->capture, true)) {
+    p->capture = NULL;
+    return true;
+  } else {
+    return false;
+  }
+}
+
+/* This is called at the end of each input buffer (ie. when we have hit a
+ * buffer seam).  If we are in the middle of capturing the input, this
+ * processes the unprocessed capture region. */
+static void capture_suspend(upb_json_parser *p, const char **ptr) {
+  if (!p->capture) return;
+
+  if (multipart_text(p, p->capture, *ptr - p->capture, false)) {
+    /* We use this as a signal that we were in the middle of capturing, and
+     * that capturing should resume at the beginning of the next buffer.
+     * 
+     * We can't use *ptr here, because we have no guarantee that this pointer
+     * will be valid when we resume (if the underlying memory is freed, then
+     * using the pointer at all, even to compare to NULL, is likely undefined
+     * behavior). */
+    p->capture = &suspend_capture;
+  } else {
+    /* Need to back up the pointer to the beginning of the capture, since
+     * we were not able to actually preserve it. */
+    *ptr = p->capture;
+  }
+}
+
+static void capture_resume(upb_json_parser *p, const char *ptr) {
+  if (p->capture) {
+    assert(p->capture == &suspend_capture);
+    p->capture = ptr;
+  }
+}
+
+
+/* Callbacks from the parser **************************************************/
+
+/* These are the functions called directly from the parser itself.
+ * We define these in the same order as their declarations in the parser. */
+
+static char escape_char(char in) {
+  switch (in) {
+    case 'r': return '\r';
+    case 't': return '\t';
+    case 'n': return '\n';
+    case 'f': return '\f';
+    case 'b': return '\b';
+    case '/': return '/';
+    case '"': return '"';
+    case '\\': return '\\';
+    default:
+      assert(0);
+      return 'x';
+  }
+}
+
+static bool escape(upb_json_parser *p, const char *ptr) {
+  char ch = escape_char(*ptr);
+  return multipart_text(p, &ch, 1, false);
+}
+
+static void start_hex(upb_json_parser *p) {
+  p->digit = 0;
+}
+
+static void hexdigit(upb_json_parser *p, const char *ptr) {
+  char ch = *ptr;
+
+  p->digit <<= 4;
+
+  if (ch >= '0' && ch <= '9') {
+    p->digit += (ch - '0');
+  } else if (ch >= 'a' && ch <= 'f') {
+    p->digit += ((ch - 'a') + 10);
+  } else {
+    assert(ch >= 'A' && ch <= 'F');
+    p->digit += ((ch - 'A') + 10);
+  }
+}
+
+static bool end_hex(upb_json_parser *p) {
+  uint32_t codepoint = p->digit;
+
+  /* emit the codepoint as UTF-8. */
+  char utf8[3]; /* support \u0000 -- \uFFFF -- need only three bytes. */
+  int length = 0;
+  if (codepoint <= 0x7F) {
+    utf8[0] = codepoint;
+    length = 1;
+  } else if (codepoint <= 0x07FF) {
+    utf8[1] = (codepoint & 0x3F) | 0x80;
+    codepoint >>= 6;
+    utf8[0] = (codepoint & 0x1F) | 0xC0;
+    length = 2;
+  } else /* codepoint <= 0xFFFF */ {
+    utf8[2] = (codepoint & 0x3F) | 0x80;
+    codepoint >>= 6;
+    utf8[1] = (codepoint & 0x3F) | 0x80;
+    codepoint >>= 6;
+    utf8[0] = (codepoint & 0x0F) | 0xE0;
+    length = 3;
+  }
+  /* TODO(haberman): Handle high surrogates: if codepoint is a high surrogate
+   * we have to wait for the next escape to get the full code point). */
+
+  return multipart_text(p, utf8, length, false);
+}
+
+static void start_text(upb_json_parser *p, const char *ptr) {
+  capture_begin(p, ptr);
+}
+
+static bool end_text(upb_json_parser *p, const char *ptr) {
+  return capture_end(p, ptr);
+}
+
+static void start_number(upb_json_parser *p, const char *ptr) {
+  multipart_startaccum(p);
+  capture_begin(p, ptr);
+}
+
+static bool parse_number(upb_json_parser *p);
+
+static bool end_number(upb_json_parser *p, const char *ptr) {
+  if (!capture_end(p, ptr)) {
+    return false;
+  }
+
+  return parse_number(p);
+}
+
+static bool parse_number(upb_json_parser *p) {
+  size_t len;
+  const char *buf;
+  const char *myend;
+  char *end;
+
+  /* strtol() and friends unfortunately do not support specifying the length of
+   * the input string, so we need to force a copy into a NULL-terminated buffer. */
+  if (!multipart_text(p, "\0", 1, false)) {
+    return false;
+  }
+
+  buf = accumulate_getptr(p, &len);
+  myend = buf + len - 1;  /* One for NULL. */
+
+  /* XXX: We are using strtol to parse integers, but this is wrong as even
+   * integers can be represented as 1e6 (for example), which strtol can't
+   * handle correctly.
+   *
+   * XXX: Also, we can't handle large integers properly because strto[u]ll
+   * isn't in C89.
+   *
+   * XXX: Also, we don't properly check floats for overflow, since strtof
+   * isn't in C89. */
+  switch (upb_fielddef_type(p->top->f)) {
+    case UPB_TYPE_ENUM:
+    case UPB_TYPE_INT32: {
+      long val = strtol(p->accumulated, &end, 0);
+      if (val > INT32_MAX || val < INT32_MIN || errno == ERANGE || end != myend)
+        goto err;
+      else
+        upb_sink_putint32(&p->top->sink, parser_getsel(p), val);
+      break;
+    }
+    case UPB_TYPE_INT64: {
+      long long val = strtol(p->accumulated, &end, 0);
+      if (val > INT64_MAX || val < INT64_MIN || errno == ERANGE || end != myend)
+        goto err;
+      else
+        upb_sink_putint64(&p->top->sink, parser_getsel(p), val);
+      break;
+    }
+    case UPB_TYPE_UINT32: {
+      unsigned long val = strtoul(p->accumulated, &end, 0);
+      if (val > UINT32_MAX || errno == ERANGE || end != myend)
+        goto err;
+      else
+        upb_sink_putuint32(&p->top->sink, parser_getsel(p), val);
+      break;
+    }
+    case UPB_TYPE_UINT64: {
+      unsigned long long val = strtoul(p->accumulated, &end, 0);
+      if (val > UINT64_MAX || errno == ERANGE || end != myend)
+        goto err;
+      else
+        upb_sink_putuint64(&p->top->sink, parser_getsel(p), val);
+      break;
+    }
+    case UPB_TYPE_DOUBLE: {
+      double val = strtod(p->accumulated, &end);
+      if (errno == ERANGE || end != myend)
+        goto err;
+      else
+        upb_sink_putdouble(&p->top->sink, parser_getsel(p), val);
+      break;
+    }
+    case UPB_TYPE_FLOAT: {
+      float val = strtod(p->accumulated, &end);
+      if (errno == ERANGE || end != myend)
+        goto err;
+      else
+        upb_sink_putfloat(&p->top->sink, parser_getsel(p), val);
+      break;
+    }
+    default:
+      assert(false);
+  }
+
+  multipart_end(p);
+
+  return true;
+
+err:
+  upb_status_seterrf(&p->status, "error parsing number: %s", buf);
+  upb_env_reporterror(p->env, &p->status);
+  multipart_end(p);
+  return false;
+}
+
+static bool parser_putbool(upb_json_parser *p, bool val) {
+  bool ok;
+
+  if (upb_fielddef_type(p->top->f) != UPB_TYPE_BOOL) {
+    upb_status_seterrf(&p->status,
+                       "Boolean value specified for non-bool field: %s",
+                       upb_fielddef_name(p->top->f));
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+
+  ok = upb_sink_putbool(&p->top->sink, parser_getsel(p), val);
+  UPB_ASSERT_VAR(ok, ok);
+
+  return true;
+}
+
+static bool start_stringval(upb_json_parser *p) {
+  assert(p->top->f);
+
+  if (upb_fielddef_isstring(p->top->f)) {
+    upb_jsonparser_frame *inner;
+    upb_selector_t sel;
+
+    if (!check_stack(p)) return false;
+
+    /* Start a new parser frame: parser frames correspond one-to-one with
+     * handler frames, and string events occur in a sub-frame. */
+    inner = p->top + 1;
+    sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSTR);
+    upb_sink_startstr(&p->top->sink, sel, 0, &inner->sink);
+    inner->m = p->top->m;
+    inner->f = p->top->f;
+    inner->is_map = false;
+    inner->is_mapentry = false;
+    p->top = inner;
+
+    if (upb_fielddef_type(p->top->f) == UPB_TYPE_STRING) {
+      /* For STRING fields we push data directly to the handlers as it is
+       * parsed.  We don't do this yet for BYTES fields, because our base64
+       * decoder is not streaming.
+       *
+       * TODO(haberman): make base64 decoding streaming also. */
+      multipart_start(p, getsel_for_handlertype(p, UPB_HANDLER_STRING));
+      return true;
+    } else {
+      multipart_startaccum(p);
+      return true;
+    }
+  } else if (upb_fielddef_type(p->top->f) == UPB_TYPE_ENUM) {
+    /* No need to push a frame -- symbolic enum names in quotes remain in the
+     * current parser frame.
+     *
+     * Enum string values must accumulate so we can look up the value in a table
+     * once it is complete. */
+    multipart_startaccum(p);
+    return true;
+  } else {
+    upb_status_seterrf(&p->status,
+                       "String specified for non-string/non-enum field: %s",
+                       upb_fielddef_name(p->top->f));
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+}
+
+static bool end_stringval(upb_json_parser *p) {
+  bool ok = true;
+
+  switch (upb_fielddef_type(p->top->f)) {
+    case UPB_TYPE_BYTES:
+      if (!base64_push(p, getsel_for_handlertype(p, UPB_HANDLER_STRING),
+                       p->accumulated, p->accumulated_len)) {
+        return false;
+      }
+      /* Fall through. */
+
+    case UPB_TYPE_STRING: {
+      upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSTR);
+      upb_sink_endstr(&p->top->sink, sel);
+      p->top--;
+      break;
+    }
+
+    case UPB_TYPE_ENUM: {
+      /* Resolve enum symbolic name to integer value. */
+      const upb_enumdef *enumdef =
+          (const upb_enumdef*)upb_fielddef_subdef(p->top->f);
+
+      size_t len;
+      const char *buf = accumulate_getptr(p, &len);
+
+      int32_t int_val = 0;
+      ok = upb_enumdef_ntoi(enumdef, buf, len, &int_val);
+
+      if (ok) {
+        upb_selector_t sel = parser_getsel(p);
+        upb_sink_putint32(&p->top->sink, sel, int_val);
+      } else {
+        upb_status_seterrf(&p->status, "Enum value unknown: '%.*s'", len, buf);
+        upb_env_reporterror(p->env, &p->status);
+      }
+
+      break;
+    }
+
+    default:
+      assert(false);
+      upb_status_seterrmsg(&p->status, "Internal error in JSON decoder");
+      upb_env_reporterror(p->env, &p->status);
+      ok = false;
+      break;
+  }
+
+  multipart_end(p);
+
+  return ok;
+}
+
+static void start_member(upb_json_parser *p) {
+  assert(!p->top->f);
+  multipart_startaccum(p);
+}
+
+/* Helper: invoked during parse_mapentry() to emit the mapentry message's key
+ * field based on the current contents of the accumulate buffer. */
+static bool parse_mapentry_key(upb_json_parser *p) {
+
+  size_t len;
+  const char *buf = accumulate_getptr(p, &len);
+
+  /* Emit the key field. We do a bit of ad-hoc parsing here because the
+   * parser state machine has already decided that this is a string field
+   * name, and we are reinterpreting it as some arbitrary key type. In
+   * particular, integer and bool keys are quoted, so we need to parse the
+   * quoted string contents here. */
+
+  p->top->f = upb_msgdef_itof(p->top->m, UPB_MAPENTRY_KEY);
+  if (p->top->f == NULL) {
+    upb_status_seterrmsg(&p->status, "mapentry message has no key");
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+  switch (upb_fielddef_type(p->top->f)) {
+    case UPB_TYPE_INT32:
+    case UPB_TYPE_INT64:
+    case UPB_TYPE_UINT32:
+    case UPB_TYPE_UINT64:
+      /* Invoke end_number. The accum buffer has the number's text already. */
+      if (!parse_number(p)) {
+        return false;
+      }
+      break;
+    case UPB_TYPE_BOOL:
+      if (len == 4 && !strncmp(buf, "true", 4)) {
+        if (!parser_putbool(p, true)) {
+          return false;
+        }
+      } else if (len == 5 && !strncmp(buf, "false", 5)) {
+        if (!parser_putbool(p, false)) {
+          return false;
+        }
+      } else {
+        upb_status_seterrmsg(&p->status,
+                             "Map bool key not 'true' or 'false'");
+        upb_env_reporterror(p->env, &p->status);
+        return false;
+      }
+      multipart_end(p);
+      break;
+    case UPB_TYPE_STRING:
+    case UPB_TYPE_BYTES: {
+      upb_sink subsink;
+      upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSTR);
+      upb_sink_startstr(&p->top->sink, sel, len, &subsink);
+      sel = getsel_for_handlertype(p, UPB_HANDLER_STRING);
+      upb_sink_putstring(&subsink, sel, buf, len, NULL);
+      sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSTR);
+      upb_sink_endstr(&subsink, sel);
+      multipart_end(p);
+      break;
+    }
+    default:
+      upb_status_seterrmsg(&p->status, "Invalid field type for map key");
+      upb_env_reporterror(p->env, &p->status);
+      return false;
+  }
+
+  return true;
+}
+
+/* Helper: emit one map entry (as a submessage in the map field sequence). This
+ * is invoked from end_membername(), at the end of the map entry's key string,
+ * with the map key in the accumulate buffer. It parses the key from that
+ * buffer, emits the handler calls to start the mapentry submessage (setting up
+ * its subframe in the process), and sets up state in the subframe so that the
+ * value parser (invoked next) will emit the mapentry's value field and then
+ * end the mapentry message. */
+
+static bool handle_mapentry(upb_json_parser *p) {
+  const upb_fielddef *mapfield;
+  const upb_msgdef *mapentrymsg;
+  upb_jsonparser_frame *inner;
+  upb_selector_t sel;
+
+  /* Map entry: p->top->sink is the seq frame, so we need to start a frame
+   * for the mapentry itself, and then set |f| in that frame so that the map
+   * value field is parsed, and also set a flag to end the frame after the
+   * map-entry value is parsed. */
+  if (!check_stack(p)) return false;
+
+  mapfield = p->top->mapfield;
+  mapentrymsg = upb_fielddef_msgsubdef(mapfield);
+
+  inner = p->top + 1;
+  p->top->f = mapfield;
+  sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSUBMSG);
+  upb_sink_startsubmsg(&p->top->sink, sel, &inner->sink);
+  inner->m = mapentrymsg;
+  inner->mapfield = mapfield;
+  inner->is_map = false;
+
+  /* Don't set this to true *yet* -- we reuse parsing handlers below to push
+   * the key field value to the sink, and these handlers will pop the frame
+   * if they see is_mapentry (when invoked by the parser state machine, they
+   * would have just seen the map-entry value, not key). */
+  inner->is_mapentry = false;
+  p->top = inner;
+
+  /* send STARTMSG in submsg frame. */
+  upb_sink_startmsg(&p->top->sink);
+
+  parse_mapentry_key(p);
+
+  /* Set up the value field to receive the map-entry value. */
+  p->top->f = upb_msgdef_itof(p->top->m, UPB_MAPENTRY_VALUE);
+  p->top->is_mapentry = true;  /* set up to pop frame after value is parsed. */
+  p->top->mapfield = mapfield;
+  if (p->top->f == NULL) {
+    upb_status_seterrmsg(&p->status, "mapentry message has no value");
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+
+  return true;
+}
+
+static bool end_membername(upb_json_parser *p) {
+  assert(!p->top->f);
+
+  if (p->top->is_map) {
+    return handle_mapentry(p);
+  } else {
+    size_t len;
+    const char *buf = accumulate_getptr(p, &len);
+    const upb_fielddef *f = upb_msgdef_ntof(p->top->m, buf, len);
+
+    if (!f) {
+      /* TODO(haberman): Ignore unknown fields if requested/configured to do
+       * so. */
+      upb_status_seterrf(&p->status, "No such field: %.*s\n", (int)len, buf);
+      upb_env_reporterror(p->env, &p->status);
+      return false;
+    }
+
+    p->top->f = f;
+    multipart_end(p);
+
+    return true;
+  }
+}
+
+static void end_member(upb_json_parser *p) {
+  /* If we just parsed a map-entry value, end that frame too. */
+  if (p->top->is_mapentry) {
+    upb_status s = UPB_STATUS_INIT;
+    upb_selector_t sel;
+    bool ok;
+    const upb_fielddef *mapfield;
+
+    assert(p->top > p->stack);
+    /* send ENDMSG on submsg. */
+    upb_sink_endmsg(&p->top->sink, &s);
+    mapfield = p->top->mapfield;
+
+    /* send ENDSUBMSG in repeated-field-of-mapentries frame. */
+    p->top--;
+    ok = upb_handlers_getselector(mapfield, UPB_HANDLER_ENDSUBMSG, &sel);
+    UPB_ASSERT_VAR(ok, ok);
+    upb_sink_endsubmsg(&p->top->sink, sel);
+  }
+
+  p->top->f = NULL;
+}
+
+static bool start_subobject(upb_json_parser *p) {
+  assert(p->top->f);
+
+  if (upb_fielddef_ismap(p->top->f)) {
+    upb_jsonparser_frame *inner;
+    upb_selector_t sel;
+
+    /* Beginning of a map. Start a new parser frame in a repeated-field
+     * context. */
+    if (!check_stack(p)) return false;
+
+    inner = p->top + 1;
+    sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSEQ);
+    upb_sink_startseq(&p->top->sink, sel, &inner->sink);
+    inner->m = upb_fielddef_msgsubdef(p->top->f);
+    inner->mapfield = p->top->f;
+    inner->f = NULL;
+    inner->is_map = true;
+    inner->is_mapentry = false;
+    p->top = inner;
+
+    return true;
+  } else if (upb_fielddef_issubmsg(p->top->f)) {
+    upb_jsonparser_frame *inner;
+    upb_selector_t sel;
+
+    /* Beginning of a subobject. Start a new parser frame in the submsg
+     * context. */
+    if (!check_stack(p)) return false;
+
+    inner = p->top + 1;
+
+    sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSUBMSG);
+    upb_sink_startsubmsg(&p->top->sink, sel, &inner->sink);
+    inner->m = upb_fielddef_msgsubdef(p->top->f);
+    inner->f = NULL;
+    inner->is_map = false;
+    inner->is_mapentry = false;
+    p->top = inner;
+
+    return true;
+  } else {
+    upb_status_seterrf(&p->status,
+                       "Object specified for non-message/group field: %s",
+                       upb_fielddef_name(p->top->f));
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+}
+
+static void end_subobject(upb_json_parser *p) {
+  if (p->top->is_map) {
+    upb_selector_t sel;
+    p->top--;
+    sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSEQ);
+    upb_sink_endseq(&p->top->sink, sel);
+  } else {
+    upb_selector_t sel;
+    p->top--;
+    sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSUBMSG);
+    upb_sink_endsubmsg(&p->top->sink, sel);
+  }
+}
+
+static bool start_array(upb_json_parser *p) {
+  upb_jsonparser_frame *inner;
+  upb_selector_t sel;
+
+  assert(p->top->f);
+
+  if (!upb_fielddef_isseq(p->top->f)) {
+    upb_status_seterrf(&p->status,
+                       "Array specified for non-repeated field: %s",
+                       upb_fielddef_name(p->top->f));
+    upb_env_reporterror(p->env, &p->status);
+    return false;
+  }
+
+  if (!check_stack(p)) return false;
+
+  inner = p->top + 1;
+  sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSEQ);
+  upb_sink_startseq(&p->top->sink, sel, &inner->sink);
+  inner->m = p->top->m;
+  inner->f = p->top->f;
+  inner->is_map = false;
+  inner->is_mapentry = false;
+  p->top = inner;
+
+  return true;
+}
+
+static void end_array(upb_json_parser *p) {
+  upb_selector_t sel;
+
+  assert(p->top > p->stack);
+
+  p->top--;
+  sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSEQ);
+  upb_sink_endseq(&p->top->sink, sel);
+}
+
+static void start_object(upb_json_parser *p) {
+  if (!p->top->is_map) {
+    upb_sink_startmsg(&p->top->sink);
+  }
+}
+
+static void end_object(upb_json_parser *p) {
+  if (!p->top->is_map) {
+    upb_status status;
+    upb_status_clear(&status);
+    upb_sink_endmsg(&p->top->sink, &status);
+    if (!upb_ok(&status)) {
+      upb_env_reporterror(p->env, &status);
+    }
+  }
+}
+
+
+#define CHECK_RETURN_TOP(x) if (!(x)) goto error
+
+
+/* The actual parser **********************************************************/
+
+/* What follows is the Ragel parser itself.  The language is specified in Ragel
+ * and the actions call our C functions above.
+ *
+ * Ragel has an extensive set of functionality, and we use only a small part of
+ * it.  There are many action types but we only use a few:
+ *
+ *   ">" -- transition into a machine
+ *   "%" -- transition out of a machine
+ *   "@" -- transition into a final state of a machine.
+ *
+ * "@" transitions are tricky because a machine can transition into a final
+ * state repeatedly.  But in some cases we know this can't happen, for example
+ * a string which is delimited by a final '"' can only transition into its
+ * final state once, when the closing '"' is seen. */
+
+
+#line 1218 "upb/json/parser.rl"
+
+
+
+#line 1130 "upb/json/parser.c"
+static const char _json_actions[] = {
+	0, 1, 0, 1, 2, 1, 3, 1, 
+	5, 1, 6, 1, 7, 1, 8, 1, 
+	10, 1, 12, 1, 13, 1, 14, 1, 
+	15, 1, 16, 1, 17, 1, 21, 1, 
+	25, 1, 27, 2, 3, 8, 2, 4, 
+	5, 2, 6, 2, 2, 6, 8, 2, 
+	11, 9, 2, 13, 15, 2, 14, 15, 
+	2, 18, 1, 2, 19, 27, 2, 20, 
+	9, 2, 22, 27, 2, 23, 27, 2, 
+	24, 27, 2, 26, 27, 3, 14, 11, 
+	9
+};
+
+static const unsigned char _json_key_offsets[] = {
+	0, 0, 4, 9, 14, 15, 19, 24, 
+	29, 34, 38, 42, 45, 48, 50, 54, 
+	58, 60, 62, 67, 69, 71, 80, 86, 
+	92, 98, 104, 106, 115, 116, 116, 116, 
+	121, 126, 131, 132, 133, 134, 135, 135, 
+	136, 137, 138, 138, 139, 140, 141, 141, 
+	146, 151, 152, 156, 161, 166, 171, 175, 
+	175, 178, 178, 178
+};
+
+static const char _json_trans_keys[] = {
+	32, 123, 9, 13, 32, 34, 125, 9, 
+	13, 32, 34, 125, 9, 13, 34, 32, 
+	58, 9, 13, 32, 93, 125, 9, 13, 
+	32, 44, 125, 9, 13, 32, 44, 125, 
+	9, 13, 32, 34, 9, 13, 45, 48, 
+	49, 57, 48, 49, 57, 46, 69, 101, 
+	48, 57, 69, 101, 48, 57, 43, 45, 
+	48, 57, 48, 57, 48, 57, 46, 69, 
+	101, 48, 57, 34, 92, 34, 92, 34, 
+	47, 92, 98, 102, 110, 114, 116, 117, 
+	48, 57, 65, 70, 97, 102, 48, 57, 
+	65, 70, 97, 102, 48, 57, 65, 70, 
+	97, 102, 48, 57, 65, 70, 97, 102, 
+	34, 92, 34, 45, 91, 102, 110, 116, 
+	123, 48, 57, 34, 32, 93, 125, 9, 
+	13, 32, 44, 93, 9, 13, 32, 93, 
+	125, 9, 13, 97, 108, 115, 101, 117, 
+	108, 108, 114, 117, 101, 32, 34, 125, 
+	9, 13, 32, 34, 125, 9, 13, 34, 
+	32, 58, 9, 13, 32, 93, 125, 9, 
+	13, 32, 44, 125, 9, 13, 32, 44, 
+	125, 9, 13, 32, 34, 9, 13, 32, 
+	9, 13, 0
+};
+
+static const char _json_single_lengths[] = {
+	0, 2, 3, 3, 1, 2, 3, 3, 
+	3, 2, 2, 1, 3, 0, 2, 2, 
+	0, 0, 3, 2, 2, 9, 0, 0, 
+	0, 0, 2, 7, 1, 0, 0, 3, 
+	3, 3, 1, 1, 1, 1, 0, 1, 
+	1, 1, 0, 1, 1, 1, 0, 3, 
+	3, 1, 2, 3, 3, 3, 2, 0, 
+	1, 0, 0, 0
+};
+
+static const char _json_range_lengths[] = {
+	0, 1, 1, 1, 0, 1, 1, 1, 
+	1, 1, 1, 1, 0, 1, 1, 1, 
+	1, 1, 1, 0, 0, 0, 3, 3, 
+	3, 3, 0, 1, 0, 0, 0, 1, 
+	1, 1, 0, 0, 0, 0, 0, 0, 
+	0, 0, 0, 0, 0, 0, 0, 1, 
+	1, 0, 1, 1, 1, 1, 1, 0, 
+	1, 0, 0, 0
+};
+
+static const short _json_index_offsets[] = {
+	0, 0, 4, 9, 14, 16, 20, 25, 
+	30, 35, 39, 43, 46, 50, 52, 56, 
+	60, 62, 64, 69, 72, 75, 85, 89, 
+	93, 97, 101, 104, 113, 115, 116, 117, 
+	122, 127, 132, 134, 136, 138, 140, 141, 
+	143, 145, 147, 148, 150, 152, 154, 155, 
+	160, 165, 167, 171, 176, 181, 186, 190, 
+	191, 194, 195, 196
+};
+
+static const char _json_indicies[] = {
+	0, 2, 0, 1, 3, 4, 5, 3, 
+	1, 6, 7, 8, 6, 1, 9, 1, 
+	10, 11, 10, 1, 11, 1, 1, 11, 
+	12, 13, 14, 15, 13, 1, 16, 17, 
+	8, 16, 1, 17, 7, 17, 1, 18, 
+	19, 20, 1, 19, 20, 1, 22, 23, 
+	23, 21, 24, 1, 23, 23, 24, 21, 
+	25, 25, 26, 1, 26, 1, 26, 21, 
+	22, 23, 23, 20, 21, 28, 29, 27, 
+	31, 32, 30, 33, 33, 33, 33, 33, 
+	33, 33, 33, 34, 1, 35, 35, 35, 
+	1, 36, 36, 36, 1, 37, 37, 37, 
+	1, 38, 38, 38, 1, 40, 41, 39, 
+	42, 43, 44, 45, 46, 47, 48, 43, 
+	1, 49, 1, 50, 51, 53, 54, 1, 
+	53, 52, 55, 56, 54, 55, 1, 56, 
+	1, 1, 56, 52, 57, 1, 58, 1, 
+	59, 1, 60, 1, 61, 62, 1, 63, 
+	1, 64, 1, 65, 66, 1, 67, 1, 
+	68, 1, 69, 70, 71, 72, 70, 1, 
+	73, 74, 75, 73, 1, 76, 1, 77, 
+	78, 77, 1, 78, 1, 1, 78, 79, 
+	80, 81, 82, 80, 1, 83, 84, 75, 
+	83, 1, 84, 74, 84, 1, 85, 86, 
+	86, 1, 1, 1, 1, 0
+};
+
+static const char _json_trans_targs[] = {
+	1, 0, 2, 3, 4, 56, 3, 4, 
+	56, 5, 5, 6, 7, 8, 9, 56, 
+	8, 9, 11, 12, 18, 57, 13, 15, 
+	14, 16, 17, 20, 58, 21, 20, 58, 
+	21, 19, 22, 23, 24, 25, 26, 20, 
+	58, 21, 28, 30, 31, 34, 39, 43, 
+	47, 29, 59, 59, 32, 31, 29, 32, 
+	33, 35, 36, 37, 38, 59, 40, 41, 
+	42, 59, 44, 45, 46, 59, 48, 49, 
+	55, 48, 49, 55, 50, 50, 51, 52, 
+	53, 54, 55, 53, 54, 59, 56
+};
+
+static const char _json_trans_actions[] = {
+	0, 0, 0, 21, 77, 53, 0, 47, 
+	23, 17, 0, 0, 15, 19, 19, 50, 
+	0, 0, 0, 0, 0, 1, 0, 0, 
+	0, 0, 0, 3, 13, 0, 0, 35, 
+	5, 11, 0, 38, 7, 7, 7, 41, 
+	44, 9, 62, 56, 25, 0, 0, 0, 
+	31, 29, 33, 59, 15, 0, 27, 0, 
+	0, 0, 0, 0, 0, 68, 0, 0, 
+	0, 71, 0, 0, 0, 65, 21, 77, 
+	53, 0, 47, 23, 17, 0, 0, 15, 
+	19, 19, 50, 0, 0, 74, 0
+};
+
+static const int json_start = 1;
+
+static const int json_en_number_machine = 10;
+static const int json_en_string_machine = 19;
+static const int json_en_value_machine = 27;
+static const int json_en_main = 1;
+
+
+#line 1221 "upb/json/parser.rl"
+
+size_t parse(void *closure, const void *hd, const char *buf, size_t size,
+             const upb_bufhandle *handle) {
+  upb_json_parser *parser = closure;
+
+  /* Variables used by Ragel's generated code. */
+  int cs = parser->current_state;
+  int *stack = parser->parser_stack;
+  int top = parser->parser_top;
+
+  const char *p = buf;
+  const char *pe = buf + size;
+
+  parser->handle = handle;
+
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+
+  capture_resume(parser, buf);
+
+  
+#line 1301 "upb/json/parser.c"
+	{
+	int _klen;
+	unsigned int _trans;
+	const char *_acts;
+	unsigned int _nacts;
+	const char *_keys;
+
+	if ( p == pe )
+		goto _test_eof;
+	if ( cs == 0 )
+		goto _out;
+_resume:
+	_keys = _json_trans_keys + _json_key_offsets[cs];
+	_trans = _json_index_offsets[cs];
+
+	_klen = _json_single_lengths[cs];
+	if ( _klen > 0 ) {
+		const char *_lower = _keys;
+		const char *_mid;
+		const char *_upper = _keys + _klen - 1;
+		while (1) {
+			if ( _upper < _lower )
+				break;
+
+			_mid = _lower + ((_upper-_lower) >> 1);
+			if ( (*p) < *_mid )
+				_upper = _mid - 1;
+			else if ( (*p) > *_mid )
+				_lower = _mid + 1;
+			else {
+				_trans += (unsigned int)(_mid - _keys);
+				goto _match;
+			}
+		}
+		_keys += _klen;
+		_trans += _klen;
+	}
+
+	_klen = _json_range_lengths[cs];
+	if ( _klen > 0 ) {
+		const char *_lower = _keys;
+		const char *_mid;
+		const char *_upper = _keys + (_klen<<1) - 2;
+		while (1) {
+			if ( _upper < _lower )
+				break;
+
+			_mid = _lower + (((_upper-_lower) >> 1) & ~1);
+			if ( (*p) < _mid[0] )
+				_upper = _mid - 2;
+			else if ( (*p) > _mid[1] )
+				_lower = _mid + 2;
+			else {
+				_trans += (unsigned int)((_mid - _keys)>>1);
+				goto _match;
+			}
+		}
+		_trans += _klen;
+	}
+
+_match:
+	_trans = _json_indicies[_trans];
+	cs = _json_trans_targs[_trans];
+
+	if ( _json_trans_actions[_trans] == 0 )
+		goto _again;
+
+	_acts = _json_actions + _json_trans_actions[_trans];
+	_nacts = (unsigned int) *_acts++;
+	while ( _nacts-- > 0 )
+	{
+		switch ( *_acts++ )
+		{
+	case 0:
+#line 1133 "upb/json/parser.rl"
+	{ p--; {cs = stack[--top]; goto _again;} }
+	break;
+	case 1:
+#line 1134 "upb/json/parser.rl"
+	{ p--; {stack[top++] = cs; cs = 10; goto _again;} }
+	break;
+	case 2:
+#line 1138 "upb/json/parser.rl"
+	{ start_text(parser, p); }
+	break;
+	case 3:
+#line 1139 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(end_text(parser, p)); }
+	break;
+	case 4:
+#line 1145 "upb/json/parser.rl"
+	{ start_hex(parser); }
+	break;
+	case 5:
+#line 1146 "upb/json/parser.rl"
+	{ hexdigit(parser, p); }
+	break;
+	case 6:
+#line 1147 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(end_hex(parser)); }
+	break;
+	case 7:
+#line 1153 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(escape(parser, p)); }
+	break;
+	case 8:
+#line 1159 "upb/json/parser.rl"
+	{ p--; {cs = stack[--top]; goto _again;} }
+	break;
+	case 9:
+#line 1162 "upb/json/parser.rl"
+	{ {stack[top++] = cs; cs = 19; goto _again;} }
+	break;
+	case 10:
+#line 1164 "upb/json/parser.rl"
+	{ p--; {stack[top++] = cs; cs = 27; goto _again;} }
+	break;
+	case 11:
+#line 1169 "upb/json/parser.rl"
+	{ start_member(parser); }
+	break;
+	case 12:
+#line 1170 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(end_membername(parser)); }
+	break;
+	case 13:
+#line 1173 "upb/json/parser.rl"
+	{ end_member(parser); }
+	break;
+	case 14:
+#line 1179 "upb/json/parser.rl"
+	{ start_object(parser); }
+	break;
+	case 15:
+#line 1182 "upb/json/parser.rl"
+	{ end_object(parser); }
+	break;
+	case 16:
+#line 1188 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(start_array(parser)); }
+	break;
+	case 17:
+#line 1192 "upb/json/parser.rl"
+	{ end_array(parser); }
+	break;
+	case 18:
+#line 1197 "upb/json/parser.rl"
+	{ start_number(parser, p); }
+	break;
+	case 19:
+#line 1198 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(end_number(parser, p)); }
+	break;
+	case 20:
+#line 1200 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(start_stringval(parser)); }
+	break;
+	case 21:
+#line 1201 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(end_stringval(parser)); }
+	break;
+	case 22:
+#line 1203 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(parser_putbool(parser, true)); }
+	break;
+	case 23:
+#line 1205 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(parser_putbool(parser, false)); }
+	break;
+	case 24:
+#line 1207 "upb/json/parser.rl"
+	{ /* null value */ }
+	break;
+	case 25:
+#line 1209 "upb/json/parser.rl"
+	{ CHECK_RETURN_TOP(start_subobject(parser)); }
+	break;
+	case 26:
+#line 1210 "upb/json/parser.rl"
+	{ end_subobject(parser); }
+	break;
+	case 27:
+#line 1215 "upb/json/parser.rl"
+	{ p--; {cs = stack[--top]; goto _again;} }
+	break;
+#line 1487 "upb/json/parser.c"
+		}
+	}
+
+_again:
+	if ( cs == 0 )
+		goto _out;
+	if ( ++p != pe )
+		goto _resume;
+	_test_eof: {}
+	_out: {}
+	}
+
+#line 1242 "upb/json/parser.rl"
+
+  if (p != pe) {
+    upb_status_seterrf(&parser->status, "Parse error at %s\n", p);
+    upb_env_reporterror(parser->env, &parser->status);
+  } else {
+    capture_suspend(parser, &p);
+  }
+
+error:
+  /* Save parsing state back to parser. */
+  parser->current_state = cs;
+  parser->parser_top = top;
+
+  return p - buf;
+}
+
+bool end(void *closure, const void *hd) {
+  UPB_UNUSED(closure);
+  UPB_UNUSED(hd);
+
+  /* Prevent compile warning on unused static constants. */
+  UPB_UNUSED(json_start);
+  UPB_UNUSED(json_en_number_machine);
+  UPB_UNUSED(json_en_string_machine);
+  UPB_UNUSED(json_en_value_machine);
+  UPB_UNUSED(json_en_main);
+  return true;
+}
+
+static void json_parser_reset(upb_json_parser *p) {
+  int cs;
+  int top;
+
+  p->top = p->stack;
+  p->top->f = NULL;
+  p->top->is_map = false;
+  p->top->is_mapentry = false;
+
+  /* Emit Ragel initialization of the parser. */
+  
+#line 1541 "upb/json/parser.c"
+	{
+	cs = json_start;
+	top = 0;
+	}
+
+#line 1282 "upb/json/parser.rl"
+  p->current_state = cs;
+  p->parser_top = top;
+  accumulate_clear(p);
+  p->multipart_state = MULTIPART_INACTIVE;
+  p->capture = NULL;
+  p->accumulated = NULL;
+  upb_status_clear(&p->status);
+}
+
+
+/* Public API *****************************************************************/
+
+upb_json_parser *upb_json_parser_create(upb_env *env, upb_sink *output) {
+#ifndef NDEBUG
+  const size_t size_before = upb_env_bytesallocated(env);
+#endif
+  upb_json_parser *p = upb_env_malloc(env, sizeof(upb_json_parser));
+  if (!p) return false;
+
+  p->env = env;
+  p->limit = p->stack + UPB_JSON_MAX_DEPTH;
+  p->accumulate_buf = NULL;
+  p->accumulate_buf_size = 0;
+  upb_byteshandler_init(&p->input_handler_);
+  upb_byteshandler_setstring(&p->input_handler_, parse, NULL);
+  upb_byteshandler_setendstr(&p->input_handler_, end, NULL);
+  upb_bytessink_reset(&p->input_, &p->input_handler_, p);
+
+  json_parser_reset(p);
+  upb_sink_reset(&p->top->sink, output->handlers, output->closure);
+  p->top->m = upb_handlers_msgdef(output->handlers);
+
+  /* If this fails, uncomment and increase the value in parser.h. */
+  /* fprintf(stderr, "%zd\n", upb_env_bytesallocated(env) - size_before); */
+  assert(upb_env_bytesallocated(env) - size_before <= UPB_JSON_PARSER_SIZE);
+  return p;
+}
+
+upb_bytessink *upb_json_parser_input(upb_json_parser *p) {
+  return &p->input_;
+}
+/*
+** This currently uses snprintf() to format primitives, and could be optimized
+** further.
+*/
+
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdint.h>
+
+struct upb_json_printer {
+  upb_sink input_;
+  /* BytesSink closure. */
+  void *subc_;
+  upb_bytessink *output_;
+
+  /* We track the depth so that we know when to emit startstr/endstr on the
+   * output. */
+  int depth_;
+
+  /* Have we emitted the first element? This state is necessary to emit commas
+   * without leaving a trailing comma in arrays/maps. We keep this state per
+   * frame depth.
+   *
+   * Why max_depth * 2? UPB_MAX_HANDLER_DEPTH counts depth as nested messages.
+   * We count frames (contexts in which we separate elements by commas) as both
+   * repeated fields and messages (maps), and the worst case is a
+   * message->repeated field->submessage->repeated field->... nesting. */
+  bool first_elem_[UPB_MAX_HANDLER_DEPTH * 2];
+};
+
+/* StringPiece; a pointer plus a length. */
+typedef struct {
+  const char *ptr;
+  size_t len;
+} strpc;
+
+strpc *newstrpc(upb_handlers *h, const upb_fielddef *f) {
+  strpc *ret = malloc(sizeof(*ret));
+  ret->ptr = upb_fielddef_name(f);
+  ret->len = strlen(ret->ptr);
+  upb_handlers_addcleanup(h, ret, free);
+  return ret;
+}
+
+/* ------------ JSON string printing: values, maps, arrays ------------------ */
+
+static void print_data(
+    upb_json_printer *p, const char *buf, unsigned int len) {
+  /* TODO: Will need to change if we support pushback from the sink. */
+  size_t n = upb_bytessink_putbuf(p->output_, p->subc_, buf, len, NULL);
+  UPB_ASSERT_VAR(n, n == len);
+}
+
+static void print_comma(upb_json_printer *p) {
+  if (!p->first_elem_[p->depth_]) {
+    print_data(p, ",", 1);
+  }
+  p->first_elem_[p->depth_] = false;
+}
+
+/* Helpers that print properly formatted elements to the JSON output stream. */
+
+/* Used for escaping control chars in strings. */
+static const char kControlCharLimit = 0x20;
+
+UPB_INLINE bool is_json_escaped(char c) {
+  /* See RFC 4627. */
+  unsigned char uc = (unsigned char)c;
+  return uc < kControlCharLimit || uc == '"' || uc == '\\';
+}
+
+UPB_INLINE char* json_nice_escape(char c) {
+  switch (c) {
+    case '"':  return "\\\"";
+    case '\\': return "\\\\";
+    case '\b': return "\\b";
+    case '\f': return "\\f";
+    case '\n': return "\\n";
+    case '\r': return "\\r";
+    case '\t': return "\\t";
+    default:   return NULL;
+  }
+}
+
+/* Write a properly escaped string chunk. The surrounding quotes are *not*
+ * printed; this is so that the caller has the option of emitting the string
+ * content in chunks. */
+static void putstring(upb_json_printer *p, const char *buf, unsigned int len) {
+  const char* unescaped_run = NULL;
+  unsigned int i;
+  for (i = 0; i < len; i++) {
+    char c = buf[i];
+    /* Handle escaping. */
+    if (is_json_escaped(c)) {
+      /* Use a "nice" escape, like \n, if one exists for this character. */
+      const char* escape = json_nice_escape(c);
+      /* If we don't have a specific 'nice' escape code, use a \uXXXX-style
+       * escape. */
+      char escape_buf[8];
+      if (!escape) {
+        unsigned char byte = (unsigned char)c;
+        _upb_snprintf(escape_buf, sizeof(escape_buf), "\\u%04x", (int)byte);
+        escape = escape_buf;
+      }
+
+      /* N.B. that we assume that the input encoding is equal to the output
+       * encoding (both UTF-8 for  now), so for chars >= 0x20 and != \, ", we
+       * can simply pass the bytes through. */
+
+      /* If there's a current run of unescaped chars, print that run first. */
+      if (unescaped_run) {
+        print_data(p, unescaped_run, &buf[i] - unescaped_run);
+        unescaped_run = NULL;
+      }
+      /* Then print the escape code. */
+      print_data(p, escape, strlen(escape));
+    } else {
+      /* Add to the current unescaped run of characters. */
+      if (unescaped_run == NULL) {
+        unescaped_run = &buf[i];
+      }
+    }
+  }
+
+  /* If the string ended in a run of unescaped characters, print that last run. */
+  if (unescaped_run) {
+    print_data(p, unescaped_run, &buf[len] - unescaped_run);
+  }
+}
+
+#define CHKLENGTH(x) if (!(x)) return -1;
+
+/* Helpers that format floating point values according to our custom formats.
+ * Right now we use %.8g and %.17g for float/double, respectively, to match
+ * proto2::util::JsonFormat's defaults.  May want to change this later. */
+
+static size_t fmt_double(double val, char* buf, size_t length) {
+  size_t n = _upb_snprintf(buf, length, "%.17g", val);
+  CHKLENGTH(n > 0 && n < length);
+  return n;
+}
+
+static size_t fmt_float(float val, char* buf, size_t length) {
+  size_t n = _upb_snprintf(buf, length, "%.8g", val);
+  CHKLENGTH(n > 0 && n < length);
+  return n;
+}
+
+static size_t fmt_bool(bool val, char* buf, size_t length) {
+  size_t n = _upb_snprintf(buf, length, "%s", (val ? "true" : "false"));
+  CHKLENGTH(n > 0 && n < length);
+  return n;
+}
+
+static size_t fmt_int64(long val, char* buf, size_t length) {
+  size_t n = _upb_snprintf(buf, length, "%ld", val);
+  CHKLENGTH(n > 0 && n < length);
+  return n;
+}
+
+static size_t fmt_uint64(unsigned long long val, char* buf, size_t length) {
+  size_t n = _upb_snprintf(buf, length, "%llu", val);
+  CHKLENGTH(n > 0 && n < length);
+  return n;
+}
+
+/* Print a map key given a field name. Called by scalar field handlers and by
+ * startseq for repeated fields. */
+static bool putkey(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  const strpc *key = handler_data;
+  print_comma(p);
+  print_data(p, "\"", 1);
+  putstring(p, key->ptr, key->len);
+  print_data(p, "\":", 2);
+  return true;
+}
+
+#define CHKFMT(val) if ((val) == (size_t)-1) return false;
+#define CHK(val)    if (!(val)) return false;
+
+#define TYPE_HANDLERS(type, fmt_func)                                        \
+  static bool put##type(void *closure, const void *handler_data, type val) { \
+    upb_json_printer *p = closure;                                           \
+    char data[64];                                                           \
+    size_t length = fmt_func(val, data, sizeof(data));                       \
+    UPB_UNUSED(handler_data);                                                \
+    CHKFMT(length);                                                          \
+    print_data(p, data, length);                                             \
+    return true;                                                             \
+  }                                                                          \
+  static bool scalar_##type(void *closure, const void *handler_data,         \
+                            type val) {                                      \
+    CHK(putkey(closure, handler_data));                                      \
+    CHK(put##type(closure, handler_data, val));                              \
+    return true;                                                             \
+  }                                                                          \
+  static bool repeated_##type(void *closure, const void *handler_data,       \
+                              type val) {                                    \
+    upb_json_printer *p = closure;                                           \
+    print_comma(p);                                                          \
+    CHK(put##type(closure, handler_data, val));                              \
+    return true;                                                             \
+  }
+
+#define TYPE_HANDLERS_MAPKEY(type, fmt_func)                                 \
+  static bool putmapkey_##type(void *closure, const void *handler_data,      \
+                            type val) {                                      \
+    upb_json_printer *p = closure;                                           \
+    print_data(p, "\"", 1);                                                  \
+    CHK(put##type(closure, handler_data, val));                              \
+    print_data(p, "\":", 2);                                                 \
+    return true;                                                             \
+  }
+
+TYPE_HANDLERS(double,   fmt_double)
+TYPE_HANDLERS(float,    fmt_float)
+TYPE_HANDLERS(bool,     fmt_bool)
+TYPE_HANDLERS(int32_t,  fmt_int64)
+TYPE_HANDLERS(uint32_t, fmt_int64)
+TYPE_HANDLERS(int64_t,  fmt_int64)
+TYPE_HANDLERS(uint64_t, fmt_uint64)
+
+/* double and float are not allowed to be map keys. */
+TYPE_HANDLERS_MAPKEY(bool,     fmt_bool)
+TYPE_HANDLERS_MAPKEY(int32_t,  fmt_int64)
+TYPE_HANDLERS_MAPKEY(uint32_t, fmt_int64)
+TYPE_HANDLERS_MAPKEY(int64_t,  fmt_int64)
+TYPE_HANDLERS_MAPKEY(uint64_t, fmt_uint64)
+
+#undef TYPE_HANDLERS
+#undef TYPE_HANDLERS_MAPKEY
+
+typedef struct {
+  void *keyname;
+  const upb_enumdef *enumdef;
+} EnumHandlerData;
+
+static bool scalar_enum(void *closure, const void *handler_data,
+                        int32_t val) {
+  const EnumHandlerData *hd = handler_data;
+  upb_json_printer *p = closure;
+  const char *symbolic_name;
+
+  CHK(putkey(closure, hd->keyname));
+
+  symbolic_name = upb_enumdef_iton(hd->enumdef, val);
+  if (symbolic_name) {
+    print_data(p, "\"", 1);
+    putstring(p, symbolic_name, strlen(symbolic_name));
+    print_data(p, "\"", 1);
+  } else {
+    putint32_t(closure, NULL, val);
+  }
+
+  return true;
+}
+
+static void print_enum_symbolic_name(upb_json_printer *p,
+                                     const upb_enumdef *def,
+                                     int32_t val) {
+  const char *symbolic_name = upb_enumdef_iton(def, val);
+  if (symbolic_name) {
+    print_data(p, "\"", 1);
+    putstring(p, symbolic_name, strlen(symbolic_name));
+    print_data(p, "\"", 1);
+  } else {
+    putint32_t(p, NULL, val);
+  }
+}
+
+static bool repeated_enum(void *closure, const void *handler_data,
+                          int32_t val) {
+  const EnumHandlerData *hd = handler_data;
+  upb_json_printer *p = closure;
+  print_comma(p);
+
+  print_enum_symbolic_name(p, hd->enumdef, val);
+
+  return true;
+}
+
+static bool mapvalue_enum(void *closure, const void *handler_data,
+                          int32_t val) {
+  const EnumHandlerData *hd = handler_data;
+  upb_json_printer *p = closure;
+
+  print_enum_symbolic_name(p, hd->enumdef, val);
+
+  return true;
+}
+
+static void *scalar_startsubmsg(void *closure, const void *handler_data) {
+  return putkey(closure, handler_data) ? closure : UPB_BREAK;
+}
+
+static void *repeated_startsubmsg(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  print_comma(p);
+  return closure;
+}
+
+static void start_frame(upb_json_printer *p) {
+  p->depth_++;
+  p->first_elem_[p->depth_] = true;
+  print_data(p, "{", 1);
+}
+
+static void end_frame(upb_json_printer *p) {
+  print_data(p, "}", 1);
+  p->depth_--;
+}
+
+static bool printer_startmsg(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  if (p->depth_ == 0) {
+    upb_bytessink_start(p->output_, 0, &p->subc_);
+  }
+  start_frame(p);
+  return true;
+}
+
+static bool printer_endmsg(void *closure, const void *handler_data, upb_status *s) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  UPB_UNUSED(s);
+  end_frame(p);
+  if (p->depth_ == 0) {
+    upb_bytessink_end(p->output_);
+  }
+  return true;
+}
+
+static void *startseq(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  CHK(putkey(closure, handler_data));
+  p->depth_++;
+  p->first_elem_[p->depth_] = true;
+  print_data(p, "[", 1);
+  return closure;
+}
+
+static bool endseq(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  print_data(p, "]", 1);
+  p->depth_--;
+  return true;
+}
+
+static void *startmap(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  CHK(putkey(closure, handler_data));
+  p->depth_++;
+  p->first_elem_[p->depth_] = true;
+  print_data(p, "{", 1);
+  return closure;
+}
+
+static bool endmap(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  print_data(p, "}", 1);
+  p->depth_--;
+  return true;
+}
+
+static size_t putstr(void *closure, const void *handler_data, const char *str,
+                     size_t len, const upb_bufhandle *handle) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  UPB_UNUSED(handle);
+  putstring(p, str, len);
+  return len;
+}
+
+/* This has to Base64 encode the bytes, because JSON has no "bytes" type. */
+static size_t putbytes(void *closure, const void *handler_data, const char *str,
+                       size_t len, const upb_bufhandle *handle) {
+  upb_json_printer *p = closure;
+
+  /* This is the regular base64, not the "web-safe" version. */
+  static const char base64[] =
+      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+
+  /* Base64-encode. */
+  char data[16000];
+  const char *limit = data + sizeof(data);
+  const unsigned char *from = (const unsigned char*)str;
+  char *to = data;
+  size_t remaining = len;
+  size_t bytes;
+
+  UPB_UNUSED(handler_data);
+  UPB_UNUSED(handle);
+
+  while (remaining > 2) {
+    /* TODO(haberman): handle encoded lengths > sizeof(data) */
+    UPB_ASSERT_VAR(limit, (limit - to) >= 4);
+
+    to[0] = base64[from[0] >> 2];
+    to[1] = base64[((from[0] & 0x3) << 4) | (from[1] >> 4)];
+    to[2] = base64[((from[1] & 0xf) << 2) | (from[2] >> 6)];
+    to[3] = base64[from[2] & 0x3f];
+
+    remaining -= 3;
+    to += 4;
+    from += 3;
+  }
+
+  switch (remaining) {
+    case 2:
+      to[0] = base64[from[0] >> 2];
+      to[1] = base64[((from[0] & 0x3) << 4) | (from[1] >> 4)];
+      to[2] = base64[(from[1] & 0xf) << 2];
+      to[3] = '=';
+      to += 4;
+      from += 2;
+      break;
+    case 1:
+      to[0] = base64[from[0] >> 2];
+      to[1] = base64[((from[0] & 0x3) << 4)];
+      to[2] = '=';
+      to[3] = '=';
+      to += 4;
+      from += 1;
+      break;
+  }
+
+  bytes = to - data;
+  print_data(p, "\"", 1);
+  putstring(p, data, bytes);
+  print_data(p, "\"", 1);
+  return len;
+}
+
+static void *scalar_startstr(void *closure, const void *handler_data,
+                             size_t size_hint) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  UPB_UNUSED(size_hint);
+  CHK(putkey(closure, handler_data));
+  print_data(p, "\"", 1);
+  return p;
+}
+
+static size_t scalar_str(void *closure, const void *handler_data,
+                         const char *str, size_t len,
+                         const upb_bufhandle *handle) {
+  CHK(putstr(closure, handler_data, str, len, handle));
+  return len;
+}
+
+static bool scalar_endstr(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  print_data(p, "\"", 1);
+  return true;
+}
+
+static void *repeated_startstr(void *closure, const void *handler_data,
+                               size_t size_hint) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  UPB_UNUSED(size_hint);
+  print_comma(p);
+  print_data(p, "\"", 1);
+  return p;
+}
+
+static size_t repeated_str(void *closure, const void *handler_data,
+                           const char *str, size_t len,
+                           const upb_bufhandle *handle) {
+  CHK(putstr(closure, handler_data, str, len, handle));
+  return len;
+}
+
+static bool repeated_endstr(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  print_data(p, "\"", 1);
+  return true;
+}
+
+static void *mapkeyval_startstr(void *closure, const void *handler_data,
+                                size_t size_hint) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  UPB_UNUSED(size_hint);
+  print_data(p, "\"", 1);
+  return p;
+}
+
+static size_t mapkey_str(void *closure, const void *handler_data,
+                         const char *str, size_t len,
+                         const upb_bufhandle *handle) {
+  CHK(putstr(closure, handler_data, str, len, handle));
+  return len;
+}
+
+static bool mapkey_endstr(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  print_data(p, "\":", 2);
+  return true;
+}
+
+static bool mapvalue_endstr(void *closure, const void *handler_data) {
+  upb_json_printer *p = closure;
+  UPB_UNUSED(handler_data);
+  print_data(p, "\"", 1);
+  return true;
+}
+
+static size_t scalar_bytes(void *closure, const void *handler_data,
+                           const char *str, size_t len,
+                           const upb_bufhandle *handle) {
+  CHK(putkey(closure, handler_data));
+  CHK(putbytes(closure, handler_data, str, len, handle));
+  return len;
+}
+
+static size_t repeated_bytes(void *closure, const void *handler_data,
+                             const char *str, size_t len,
+                             const upb_bufhandle *handle) {
+  upb_json_printer *p = closure;
+  print_comma(p);
+  CHK(putbytes(closure, handler_data, str, len, handle));
+  return len;
+}
+
+static size_t mapkey_bytes(void *closure, const void *handler_data,
+                           const char *str, size_t len,
+                           const upb_bufhandle *handle) {
+  upb_json_printer *p = closure;
+  CHK(putbytes(closure, handler_data, str, len, handle));
+  print_data(p, ":", 1);
+  return len;
+}
+
+static void set_enum_hd(upb_handlers *h,
+                        const upb_fielddef *f,
+                        upb_handlerattr *attr) {
+  EnumHandlerData *hd = malloc(sizeof(EnumHandlerData));
+  hd->enumdef = (const upb_enumdef *)upb_fielddef_subdef(f);
+  hd->keyname = newstrpc(h, f);
+  upb_handlers_addcleanup(h, hd, free);
+  upb_handlerattr_sethandlerdata(attr, hd);
+}
+
+/* Set up handlers for a mapentry submessage (i.e., an individual key/value pair
+ * in a map).
+ *
+ * TODO: Handle missing key, missing value, out-of-order key/value, or repeated
+ * key or value cases properly. The right way to do this is to allocate a
+ * temporary structure at the start of a mapentry submessage, store key and
+ * value data in it as key and value handlers are called, and then print the
+ * key/value pair once at the end of the submessage. If we don't do this, we
+ * should at least detect the case and throw an error. However, so far all of
+ * our sources that emit mapentry messages do so canonically (with one key
+ * field, and then one value field), so this is not a pressing concern at the
+ * moment. */
+void printer_sethandlers_mapentry(const void *closure, upb_handlers *h) {
+  const upb_msgdef *md = upb_handlers_msgdef(h);
+
+  /* A mapentry message is printed simply as '"key": value'. Rather than
+   * special-case key and value for every type below, we just handle both
+   * fields explicitly here. */
+  const upb_fielddef* key_field = upb_msgdef_itof(md, UPB_MAPENTRY_KEY);
+  const upb_fielddef* value_field = upb_msgdef_itof(md, UPB_MAPENTRY_VALUE);
+
+  upb_handlerattr empty_attr = UPB_HANDLERATTR_INITIALIZER;
+
+  UPB_UNUSED(closure);
+
+  switch (upb_fielddef_type(key_field)) {
+    case UPB_TYPE_INT32:
+      upb_handlers_setint32(h, key_field, putmapkey_int32_t, &empty_attr);
+      break;
+    case UPB_TYPE_INT64:
+      upb_handlers_setint64(h, key_field, putmapkey_int64_t, &empty_attr);
+      break;
+    case UPB_TYPE_UINT32:
+      upb_handlers_setuint32(h, key_field, putmapkey_uint32_t, &empty_attr);
+      break;
+    case UPB_TYPE_UINT64:
+      upb_handlers_setuint64(h, key_field, putmapkey_uint64_t, &empty_attr);
+      break;
+    case UPB_TYPE_BOOL:
+      upb_handlers_setbool(h, key_field, putmapkey_bool, &empty_attr);
+      break;
+    case UPB_TYPE_STRING:
+      upb_handlers_setstartstr(h, key_field, mapkeyval_startstr, &empty_attr);
+      upb_handlers_setstring(h, key_field, mapkey_str, &empty_attr);
+      upb_handlers_setendstr(h, key_field, mapkey_endstr, &empty_attr);
+      break;
+    case UPB_TYPE_BYTES:
+      upb_handlers_setstring(h, key_field, mapkey_bytes, &empty_attr);
+      break;
+    default:
+      assert(false);
+      break;
+  }
+
+  switch (upb_fielddef_type(value_field)) {
+    case UPB_TYPE_INT32:
+      upb_handlers_setint32(h, value_field, putint32_t, &empty_attr);
+      break;
+    case UPB_TYPE_INT64:
+      upb_handlers_setint64(h, value_field, putint64_t, &empty_attr);
+      break;
+    case UPB_TYPE_UINT32:
+      upb_handlers_setuint32(h, value_field, putuint32_t, &empty_attr);
+      break;
+    case UPB_TYPE_UINT64:
+      upb_handlers_setuint64(h, value_field, putuint64_t, &empty_attr);
+      break;
+    case UPB_TYPE_BOOL:
+      upb_handlers_setbool(h, value_field, putbool, &empty_attr);
+      break;
+    case UPB_TYPE_FLOAT:
+      upb_handlers_setfloat(h, value_field, putfloat, &empty_attr);
+      break;
+    case UPB_TYPE_DOUBLE:
+      upb_handlers_setdouble(h, value_field, putdouble, &empty_attr);
+      break;
+    case UPB_TYPE_STRING:
+      upb_handlers_setstartstr(h, value_field, mapkeyval_startstr, &empty_attr);
+      upb_handlers_setstring(h, value_field, putstr, &empty_attr);
+      upb_handlers_setendstr(h, value_field, mapvalue_endstr, &empty_attr);
+      break;
+    case UPB_TYPE_BYTES:
+      upb_handlers_setstring(h, value_field, putbytes, &empty_attr);
+      break;
+    case UPB_TYPE_ENUM: {
+      upb_handlerattr enum_attr = UPB_HANDLERATTR_INITIALIZER;
+      set_enum_hd(h, value_field, &enum_attr);
+      upb_handlers_setint32(h, value_field, mapvalue_enum, &enum_attr);
+      upb_handlerattr_uninit(&enum_attr);
+      break;
+    }
+    case UPB_TYPE_MESSAGE:
+      /* No handler necessary -- the submsg handlers will print the message
+       * as appropriate. */
+      break;
+  }
+
+  upb_handlerattr_uninit(&empty_attr);
+}
+
+void printer_sethandlers(const void *closure, upb_handlers *h) {
+  const upb_msgdef *md = upb_handlers_msgdef(h);
+  bool is_mapentry = upb_msgdef_mapentry(md);
+  upb_handlerattr empty_attr = UPB_HANDLERATTR_INITIALIZER;
+  upb_msg_field_iter i;
+
+  UPB_UNUSED(closure);
+
+  if (is_mapentry) {
+    /* mapentry messages are sufficiently different that we handle them
+     * separately. */
+    printer_sethandlers_mapentry(closure, h);
+    return;
+  }
+
+  upb_handlers_setstartmsg(h, printer_startmsg, &empty_attr);
+  upb_handlers_setendmsg(h, printer_endmsg, &empty_attr);
+
+#define TYPE(type, name, ctype)                                               \
+  case type:                                                                  \
+    if (upb_fielddef_isseq(f)) {                                              \
+      upb_handlers_set##name(h, f, repeated_##ctype, &empty_attr);            \
+    } else {                                                                  \
+      upb_handlers_set##name(h, f, scalar_##ctype, &name_attr);               \
+    }                                                                         \
+    break;
+
+  upb_msg_field_begin(&i, md);
+  for(; !upb_msg_field_done(&i); upb_msg_field_next(&i)) {
+    const upb_fielddef *f = upb_msg_iter_field(&i);
+
+    upb_handlerattr name_attr = UPB_HANDLERATTR_INITIALIZER;
+    upb_handlerattr_sethandlerdata(&name_attr, newstrpc(h, f));
+
+    if (upb_fielddef_ismap(f)) {
+      upb_handlers_setstartseq(h, f, startmap, &name_attr);
+      upb_handlers_setendseq(h, f, endmap, &name_attr);
+    } else if (upb_fielddef_isseq(f)) {
+      upb_handlers_setstartseq(h, f, startseq, &name_attr);
+      upb_handlers_setendseq(h, f, endseq, &empty_attr);
+    }
+
+    switch (upb_fielddef_type(f)) {
+      TYPE(UPB_TYPE_FLOAT,  float,  float);
+      TYPE(UPB_TYPE_DOUBLE, double, double);
+      TYPE(UPB_TYPE_BOOL,   bool,   bool);
+      TYPE(UPB_TYPE_INT32,  int32,  int32_t);
+      TYPE(UPB_TYPE_UINT32, uint32, uint32_t);
+      TYPE(UPB_TYPE_INT64,  int64,  int64_t);
+      TYPE(UPB_TYPE_UINT64, uint64, uint64_t);
+      case UPB_TYPE_ENUM: {
+        /* For now, we always emit symbolic names for enums. We may want an
+         * option later to control this behavior, but we will wait for a real
+         * need first. */
+        upb_handlerattr enum_attr = UPB_HANDLERATTR_INITIALIZER;
+        set_enum_hd(h, f, &enum_attr);
+
+        if (upb_fielddef_isseq(f)) {
+          upb_handlers_setint32(h, f, repeated_enum, &enum_attr);
+        } else {
+          upb_handlers_setint32(h, f, scalar_enum, &enum_attr);
+        }
+
+        upb_handlerattr_uninit(&enum_attr);
+        break;
+      }
+      case UPB_TYPE_STRING:
+        if (upb_fielddef_isseq(f)) {
+          upb_handlers_setstartstr(h, f, repeated_startstr, &empty_attr);
+          upb_handlers_setstring(h, f, repeated_str, &empty_attr);
+          upb_handlers_setendstr(h, f, repeated_endstr, &empty_attr);
+        } else {
+          upb_handlers_setstartstr(h, f, scalar_startstr, &name_attr);
+          upb_handlers_setstring(h, f, scalar_str, &empty_attr);
+          upb_handlers_setendstr(h, f, scalar_endstr, &empty_attr);
+        }
+        break;
+      case UPB_TYPE_BYTES:
+        /* XXX: this doesn't support strings that span buffers yet. The base64
+         * encoder will need to be made resumable for this to work properly. */
+        if (upb_fielddef_isseq(f)) {
+          upb_handlers_setstring(h, f, repeated_bytes, &empty_attr);
+        } else {
+          upb_handlers_setstring(h, f, scalar_bytes, &name_attr);
+        }
+        break;
+      case UPB_TYPE_MESSAGE:
+        if (upb_fielddef_isseq(f)) {
+          upb_handlers_setstartsubmsg(h, f, repeated_startsubmsg, &name_attr);
+        } else {
+          upb_handlers_setstartsubmsg(h, f, scalar_startsubmsg, &name_attr);
+        }
+        break;
+    }
+
+    upb_handlerattr_uninit(&name_attr);
+  }
+
+  upb_handlerattr_uninit(&empty_attr);
+#undef TYPE
+}
+
+static void json_printer_reset(upb_json_printer *p) {
+  p->depth_ = 0;
+}
+
+
+/* Public API *****************************************************************/
+
+upb_json_printer *upb_json_printer_create(upb_env *e, const upb_handlers *h,
+                                          upb_bytessink *output) {
+#ifndef NDEBUG
+  size_t size_before = upb_env_bytesallocated(e);
+#endif
+
+  upb_json_printer *p = upb_env_malloc(e, sizeof(upb_json_printer));
+  if (!p) return NULL;
+
+  p->output_ = output;
+  json_printer_reset(p);
+  upb_sink_reset(&p->input_, h, p);
+
+  /* If this fails, increase the value in printer.h. */
+  assert(upb_env_bytesallocated(e) - size_before <= UPB_JSON_PRINTER_SIZE);
+  return p;
+}
+
+upb_sink *upb_json_printer_input(upb_json_printer *p) {
+  return &p->input_;
+}
+
+const upb_handlers *upb_json_printer_newhandlers(const upb_msgdef *md,
+                                                 const void *owner) {
+  return upb_handlers_newfrozen(md, owner, printer_sethandlers, NULL);
+}
diff --git a/php/ext/google/protobuf/upb.h b/php/ext/google/protobuf/upb.h
new file mode 100644
index 00000000..078e2a28
--- /dev/null
+++ b/php/ext/google/protobuf/upb.h
@@ -0,0 +1,8217 @@
+// Amalgamated source file
+/*
+** Defs are upb's internal representation of the constructs that can appear
+** in a .proto file:
+**
+** - upb::MessageDef (upb_msgdef): describes a "message" construct.
+** - upb::FieldDef (upb_fielddef): describes a message field.
+** - upb::EnumDef (upb_enumdef): describes an enum.
+** - upb::OneofDef (upb_oneofdef): describes a oneof.
+** - upb::Def (upb_def): base class of all the others.
+**
+** TODO: definitions of services.
+**
+** Like upb_refcounted objects, defs are mutable only until frozen, and are
+** only thread-safe once frozen.
+**
+** This is a mixed C/C++ interface that offers a full API to both languages.
+** See the top-level README for more information.
+*/
+
+#ifndef UPB_DEF_H_
+#define UPB_DEF_H_
+
+/*
+** upb::RefCounted (upb_refcounted)
+**
+** A refcounting scheme that supports circular refs.  It accomplishes this by
+** partitioning the set of objects into groups such that no cycle spans groups;
+** we can then reference-count the group as a whole and ignore refs within the
+** group.  When objects are mutable, these groups are computed very
+** conservatively; we group any objects that have ever had a link between them.
+** When objects are frozen, we compute strongly-connected components which
+** allows us to be precise and only group objects that are actually cyclic.
+**
+** This is a mixed C/C++ interface that offers a full API to both languages.
+** See the top-level README for more information.
+*/
+
+#ifndef UPB_REFCOUNTED_H_
+#define UPB_REFCOUNTED_H_
+
+/*
+** upb_table
+**
+** This header is INTERNAL-ONLY!  Its interfaces are not public or stable!
+** This file defines very fast int->upb_value (inttable) and string->upb_value
+** (strtable) hash tables.
+**
+** The table uses chained scatter with Brent's variation (inspired by the Lua
+** implementation of hash tables).  The hash function for strings is Austin
+** Appleby's "MurmurHash."
+**
+** The inttable uses uintptr_t as its key, which guarantees it can be used to
+** store pointers or integers of at least 32 bits (upb isn't really useful on
+** systems where sizeof(void*) < 4).
+**
+** The table must be homogenous (all values of the same type).  In debug
+** mode, we check this on insert and lookup.
+*/
+
+#ifndef UPB_TABLE_H_
+#define UPB_TABLE_H_
+
+#include <assert.h>
+#include <stdint.h>
+#include <string.h>
+/*
+** This file contains shared definitions that are widely used across upb.
+**
+** This is a mixed C/C++ interface that offers a full API to both languages.
+** See the top-level README for more information.
+*/
+
+#ifndef UPB_H_
+#define UPB_H_
+
+#include <assert.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stddef.h>
+
+/* UPB_INLINE: inline if possible, emit standalone code if required. */
+#ifdef __cplusplus
+#define UPB_INLINE inline
+#elif defined (__GNUC__)
+#define UPB_INLINE static __inline__
+#else
+#define UPB_INLINE static
+#endif
+
+/* Define UPB_BIG_ENDIAN manually if you're on big endian and your compiler
+ * doesn't provide these preprocessor symbols. */
+#if defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#define UPB_BIG_ENDIAN
+#endif
+
+/* Macros for function attributes on compilers that support them. */
+#ifdef __GNUC__
+#define UPB_FORCEINLINE __inline__ __attribute__((always_inline))
+#define UPB_NOINLINE __attribute__((noinline))
+#define UPB_NORETURN __attribute__((__noreturn__))
+#else  /* !defined(__GNUC__) */
+#define UPB_FORCEINLINE
+#define UPB_NOINLINE
+#define UPB_NORETURN
+#endif
+
+/* A few hacky workarounds for functions not in C89.
+ * For internal use only!
+ * TODO(haberman): fix these by including our own implementations, or finding
+ * another workaround.
+ */
+#ifdef __GNUC__
+#define _upb_snprintf __builtin_snprintf
+#define _upb_vsnprintf __builtin_vsnprintf
+#define _upb_va_copy(a, b) __va_copy(a, b)
+#elif __STDC_VERSION__ >= 199901L
+/* C99 versions. */
+#define _upb_snprintf snprintf
+#define _upb_vsnprintf vsnprintf
+#define _upb_va_copy(a, b) va_copy(a, b)
+#else
+#error Need implementations of [v]snprintf and va_copy
+#endif
+
+
+#if ((defined(__cplusplus) && __cplusplus >= 201103L) || \
+      defined(__GXX_EXPERIMENTAL_CXX0X__)) && !defined(UPB_NO_CXX11)
+#define UPB_CXX11
+#endif
+
+/* UPB_DISALLOW_COPY_AND_ASSIGN()
+ * UPB_DISALLOW_POD_OPS()
+ *
+ * Declare these in the "private" section of a C++ class to forbid copy/assign
+ * or all POD ops (construct, destruct, copy, assign) on that class. */
+#ifdef UPB_CXX11
+#include <type_traits>
+#define UPB_DISALLOW_COPY_AND_ASSIGN(class_name) \
+  class_name(const class_name&) = delete; \
+  void operator=(const class_name&) = delete;
+#define UPB_DISALLOW_POD_OPS(class_name, full_class_name) \
+  class_name() = delete; \
+  ~class_name() = delete; \
+  UPB_DISALLOW_COPY_AND_ASSIGN(class_name)
+#define UPB_ASSERT_STDLAYOUT(type) \
+  static_assert(std::is_standard_layout<type>::value, \
+                #type " must be standard layout");
+#else  /* !defined(UPB_CXX11) */
+#define UPB_DISALLOW_COPY_AND_ASSIGN(class_name) \
+  class_name(const class_name&); \
+  void operator=(const class_name&);
+#define UPB_DISALLOW_POD_OPS(class_name, full_class_name) \
+  class_name(); \
+  ~class_name(); \
+  UPB_DISALLOW_COPY_AND_ASSIGN(class_name)
+#define UPB_ASSERT_STDLAYOUT(type)
+#endif
+
+/* UPB_DECLARE_TYPE()
+ * UPB_DECLARE_DERIVED_TYPE()
+ * UPB_DECLARE_DERIVED_TYPE2()
+ *
+ * Macros for declaring C and C++ types both, including inheritance.
+ * The inheritance doesn't use real C++ inheritance, to stay compatible with C.
+ *
+ * These macros also provide upcasts:
+ *  - in C: types-specific functions (ie. upb_foo_upcast(foo))
+ *  - in C++: upb::upcast(foo) along with implicit conversions
+ *
+ * Downcasts are not provided, but upb/def.h defines downcasts for upb::Def. */
+
+#define UPB_C_UPCASTS(ty, base)                                      \
+  UPB_INLINE base *ty ## _upcast_mutable(ty *p) { return (base*)p; } \
+  UPB_INLINE const base *ty ## _upcast(const ty *p) { return (const base*)p; }
+
+#define UPB_C_UPCASTS2(ty, base, base2)                                 \
+  UPB_C_UPCASTS(ty, base)                                               \
+  UPB_INLINE base2 *ty ## _upcast2_mutable(ty *p) { return (base2*)p; } \
+  UPB_INLINE const base2 *ty ## _upcast2(const ty *p) { return (const base2*)p; }
+
+#ifdef __cplusplus
+
+#define UPB_BEGIN_EXTERN_C extern "C" {
+#define UPB_END_EXTERN_C }
+#define UPB_PRIVATE_FOR_CPP private:
+#define UPB_DECLARE_TYPE(cppname, cname) typedef cppname cname;
+
+#define UPB_DECLARE_DERIVED_TYPE(cppname, cppbase, cname, cbase)  \
+  UPB_DECLARE_TYPE(cppname, cname)                                \
+  UPB_C_UPCASTS(cname, cbase)                                     \
+  namespace upb {                                                 \
+  template <>                                                     \
+  class Pointer<cppname> : public PointerBase<cppname, cppbase> { \
+   public:                                                        \
+    explicit Pointer(cppname* ptr) : PointerBase(ptr) {}          \
+  };                                                              \
+  template <>                                                     \
+  class Pointer<const cppname>                                    \
+      : public PointerBase<const cppname, const cppbase> {        \
+   public:                                                        \
+    explicit Pointer(const cppname* ptr) : PointerBase(ptr) {}    \
+  };                                                              \
+  }
+
+#define UPB_DECLARE_DERIVED_TYPE2(cppname, cppbase, cppbase2, cname, cbase,  \
+                                  cbase2)                                    \
+  UPB_DECLARE_TYPE(cppname, cname)                                           \
+  UPB_C_UPCASTS2(cname, cbase, cbase2)                                       \
+  namespace upb {                                                            \
+  template <>                                                                \
+  class Pointer<cppname> : public PointerBase2<cppname, cppbase, cppbase2> { \
+   public:                                                                   \
+    explicit Pointer(cppname* ptr) : PointerBase2(ptr) {}                    \
+  };                                                                         \
+  template <>                                                                \
+  class Pointer<const cppname>                                               \
+      : public PointerBase2<const cppname, const cppbase, const cppbase2> {  \
+   public:                                                                   \
+    explicit Pointer(const cppname* ptr) : PointerBase2(ptr) {}              \
+  };                                                                         \
+  }
+
+#else  /* !defined(__cplusplus) */
+
+#define UPB_BEGIN_EXTERN_C
+#define UPB_END_EXTERN_C
+#define UPB_PRIVATE_FOR_CPP
+#define UPB_DECLARE_TYPE(cppname, cname) \
+  struct cname;                          \
+  typedef struct cname cname;
+#define UPB_DECLARE_DERIVED_TYPE(cppname, cppbase, cname, cbase) \
+  UPB_DECLARE_TYPE(cppname, cname)                               \
+  UPB_C_UPCASTS(cname, cbase)
+#define UPB_DECLARE_DERIVED_TYPE2(cppname, cppbase, cppbase2,    \
+                                  cname, cbase, cbase2)          \
+  UPB_DECLARE_TYPE(cppname, cname)                               \
+  UPB_C_UPCASTS2(cname, cbase, cbase2)
+
+#endif  /* defined(__cplusplus) */
+
+#define UPB_MAX(x, y) ((x) > (y) ? (x) : (y))
+#define UPB_MIN(x, y) ((x) < (y) ? (x) : (y))
+
+#define UPB_UNUSED(var) (void)var
+
+/* For asserting something about a variable when the variable is not used for
+ * anything else.  This prevents "unused variable" warnings when compiling in
+ * debug mode. */
+#define UPB_ASSERT_VAR(var, predicate) UPB_UNUSED(var); assert(predicate)
+
+/* Generic function type. */
+typedef void upb_func();
+
+/* C++ Casts ******************************************************************/
+
+#ifdef __cplusplus
+
+namespace upb {
+
+template <class T> class Pointer;
+
+/* Casts to a subclass.  The caller must know that cast is correct; an
+ * incorrect cast will throw an assertion failure in debug mode.
+ *
+ * Example:
+ *   upb::Def* def = GetDef();
+ *   // Assert-fails if this was not actually a MessageDef.
+ *   upb::MessgeDef* md = upb::down_cast<upb::MessageDef>(def);
+ *
+ * Note that downcasts are only defined for some types (at the moment you can
+ * only downcast from a upb::Def to a specific Def type). */
+template<class To, class From> To down_cast(From* f);
+
+/* Casts to a subclass.  If the class does not actually match the given To type,
+ * returns NULL.
+ *
+ * Example:
+ *   upb::Def* def = GetDef();
+ *   // md will be NULL if this was not actually a MessageDef.
+ *   upb::MessgeDef* md = upb::down_cast<upb::MessageDef>(def);
+ *
+ * Note that dynamic casts are only defined for some types (at the moment you
+ * can only downcast from a upb::Def to a specific Def type).. */
+template<class To, class From> To dyn_cast(From* f);
+
+/* Casts to any base class, or the type itself (ie. can be a no-op).
+ *
+ * Example:
+ *   upb::MessageDef* md = GetDef();
+ *   // This will fail to compile if this wasn't actually a base class.
+ *   upb::Def* def = upb::upcast(md);
+ */
+template <class T> inline Pointer<T> upcast(T *f) { return Pointer<T>(f); }
+
+/* Attempt upcast to specific base class.
+ *
+ * Example:
+ *   upb::MessageDef* md = GetDef();
+ *   upb::upcast_to<upb::Def>(md)->MethodOnDef();
+ */
+template <class T, class F> inline T* upcast_to(F *f) {
+  return static_cast<T*>(upcast(f));
+}
+
+/* PointerBase<T>: implementation detail of upb::upcast().
+ * It is implicitly convertable to pointers to the Base class(es).
+ */
+template <class T, class Base>
+class PointerBase {
+ public:
+  explicit PointerBase(T* ptr) : ptr_(ptr) {}
+  operator T*() { return ptr_; }
+  operator Base*() { return (Base*)ptr_; }
+
+ private:
+  T* ptr_;
+};
+
+template <class T, class Base, class Base2>
+class PointerBase2 : public PointerBase<T, Base> {
+ public:
+  explicit PointerBase2(T* ptr) : PointerBase<T, Base>(ptr) {}
+  operator Base2*() { return Pointer<Base>(*this); }
+};
+
+}
+
+#endif
+
+
+/* upb::reffed_ptr ************************************************************/
+
+#ifdef __cplusplus
+
+#include <algorithm>  /* For std::swap(). */
+
+namespace upb {
+
+/* Provides RAII semantics for upb refcounted objects.  Each reffed_ptr owns a
+ * ref on whatever object it points to (if any). */
+template <class T> class reffed_ptr {
+ public:
+  reffed_ptr() : ptr_(NULL) {}
+
+  /* If ref_donor is NULL, takes a new ref, otherwise adopts from ref_donor. */
+  template <class U>
+  reffed_ptr(U* val, const void* ref_donor = NULL)
+      : ptr_(upb::upcast(val)) {
+    if (ref_donor) {
+      assert(ptr_);
+      ptr_->DonateRef(ref_donor, this);
+    } else if (ptr_) {
+      ptr_->Ref(this);
+    }
+  }
+
+  template <class U>
+  reffed_ptr(const reffed_ptr<U>& other)
+      : ptr_(upb::upcast(other.get())) {
+    if (ptr_) ptr_->Ref(this);
+  }
+
+  ~reffed_ptr() { if (ptr_) ptr_->Unref(this); }
+
+  template <class U>
+  reffed_ptr& operator=(const reffed_ptr<U>& other) {
+    reset(other.get());
+    return *this;
+  }
+
+  reffed_ptr& operator=(const reffed_ptr& other) {
+    reset(other.get());
+    return *this;
+  }
+
+  /* TODO(haberman): add C++11 move construction/assignment for greater
+   * efficiency. */
+
+  void swap(reffed_ptr& other) {
+    if (ptr_ == other.ptr_) {
+      return;
+    }
+
+    if (ptr_) ptr_->DonateRef(this, &other);
+    if (other.ptr_) other.ptr_->DonateRef(&other, this);
+    std::swap(ptr_, other.ptr_);
+  }
+
+  T& operator*() const {
+    assert(ptr_);
+    return *ptr_;
+  }
+
+  T* operator->() const {
+    assert(ptr_);
+    return ptr_;
+  }
+
+  T* get() const { return ptr_; }
+
+  /* If ref_donor is NULL, takes a new ref, otherwise adopts from ref_donor. */
+  template <class U>
+  void reset(U* ptr = NULL, const void* ref_donor = NULL) {
+    reffed_ptr(ptr, ref_donor).swap(*this);
+  }
+
+  template <class U>
+  reffed_ptr<U> down_cast() {
+    return reffed_ptr<U>(upb::down_cast<U*>(get()));
+  }
+
+  template <class U>
+  reffed_ptr<U> dyn_cast() {
+    return reffed_ptr<U>(upb::dyn_cast<U*>(get()));
+  }
+
+  /* Plain release() is unsafe; if we were the only owner, it would leak the
+   * object.  Instead we provide this: */
+  T* ReleaseTo(const void* new_owner) {
+    T* ret = NULL;
+    ptr_->DonateRef(this, new_owner);
+    std::swap(ret, ptr_);
+    return ret;
+  }
+
+ private:
+  T* ptr_;
+};
+
+}  /* namespace upb */
+
+#endif  /* __cplusplus */
+
+
+/* upb::Status ****************************************************************/
+
+#ifdef __cplusplus
+namespace upb {
+class ErrorSpace;
+class Status;
+}
+#endif
+
+UPB_DECLARE_TYPE(upb::ErrorSpace, upb_errorspace)
+UPB_DECLARE_TYPE(upb::Status, upb_status)
+
+/* The maximum length of an error message before it will get truncated. */
+#define UPB_STATUS_MAX_MESSAGE 128
+
+/* An error callback function is used to report errors from some component.
+ * The function can return "true" to indicate that the component should try
+ * to recover and proceed, but this is not always possible. */
+typedef bool upb_errcb_t(void *closure, const upb_status* status);
+
+#ifdef __cplusplus
+class upb::ErrorSpace {
+#else
+struct upb_errorspace {
+#endif
+  const char *name;
+  /* Should the error message in the status object according to this code. */
+  void (*set_message)(upb_status* status, int code);
+};
+
+#ifdef __cplusplus
+
+/* Object representing a success or failure status.
+ * It owns no resources and allocates no memory, so it should work
+ * even in OOM situations. */
+
+class upb::Status {
+ public:
+  Status();
+
+  /* Returns true if there is no error. */
+  bool ok() const;
+
+  /* Optional error space and code, useful if the caller wants to
+   * programmatically check the specific kind of error. */
+  ErrorSpace* error_space();
+  int code() const;
+
+  const char *error_message() const;
+
+  /* The error message will be truncated if it is longer than
+   * UPB_STATUS_MAX_MESSAGE-4. */
+  void SetErrorMessage(const char* msg);
+  void SetFormattedErrorMessage(const char* fmt, ...);
+
+  /* If there is no error message already, this will use the ErrorSpace to
+   * populate the error message for this code.  The caller can still call
+   * SetErrorMessage() to give a more specific message. */
+  void SetErrorCode(ErrorSpace* space, int code);
+
+  /* Resets the status to a successful state with no message. */
+  void Clear();
+
+  void CopyFrom(const Status& other);
+
+ private:
+  UPB_DISALLOW_COPY_AND_ASSIGN(Status)
+#else
+struct upb_status {
+#endif
+  bool ok_;
+
+  /* Specific status code defined by some error space (optional). */
+  int code_;
+  upb_errorspace *error_space_;
+
+  /* Error message; NULL-terminated. */
+  char msg[UPB_STATUS_MAX_MESSAGE];
+};
+
+#define UPB_STATUS_INIT {true, 0, NULL, {0}}
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* The returned string is invalidated by any other call into the status. */
+const char *upb_status_errmsg(const upb_status *status);
+bool upb_ok(const upb_status *status);
+upb_errorspace *upb_status_errspace(const upb_status *status);
+int upb_status_errcode(const upb_status *status);
+
+/* Any of the functions that write to a status object allow status to be NULL,
+ * to support use cases where the function's caller does not care about the
+ * status message. */
+void upb_status_clear(upb_status *status);
+void upb_status_seterrmsg(upb_status *status, const char *msg);
+void upb_status_seterrf(upb_status *status, const char *fmt, ...);
+void upb_status_vseterrf(upb_status *status, const char *fmt, va_list args);
+void upb_status_seterrcode(upb_status *status, upb_errorspace *space, int code);
+void upb_status_copy(upb_status *to, const upb_status *from);
+
+#ifdef __cplusplus
+}  /* extern "C" */
+
+namespace upb {
+
+/* C++ Wrappers */
+inline Status::Status() { Clear(); }
+inline bool Status::ok() const { return upb_ok(this); }
+inline const char* Status::error_message() const {
+  return upb_status_errmsg(this);
+}
+inline void Status::SetErrorMessage(const char* msg) {
+  upb_status_seterrmsg(this, msg);
+}
+inline void Status::SetFormattedErrorMessage(const char* fmt, ...) {
+  va_list args;
+  va_start(args, fmt);
+  upb_status_vseterrf(this, fmt, args);
+  va_end(args);
+}
+inline void Status::SetErrorCode(ErrorSpace* space, int code) {
+  upb_status_seterrcode(this, space, code);
+}
+inline void Status::Clear() { upb_status_clear(this); }
+inline void Status::CopyFrom(const Status& other) {
+  upb_status_copy(this, &other);
+}
+
+}  /* namespace upb */
+
+#endif
+
+#endif  /* UPB_H_ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/* upb_value ******************************************************************/
+
+/* A tagged union (stored untagged inside the table) so that we can check that
+ * clients calling table accessors are correctly typed without having to have
+ * an explosion of accessors. */
+typedef enum {
+  UPB_CTYPE_INT32    = 1,
+  UPB_CTYPE_INT64    = 2,
+  UPB_CTYPE_UINT32   = 3,
+  UPB_CTYPE_UINT64   = 4,
+  UPB_CTYPE_BOOL     = 5,
+  UPB_CTYPE_CSTR     = 6,
+  UPB_CTYPE_PTR      = 7,
+  UPB_CTYPE_CONSTPTR = 8,
+  UPB_CTYPE_FPTR     = 9
+} upb_ctype_t;
+
+typedef struct {
+  uint64_t val;
+#ifndef NDEBUG
+  /* In debug mode we carry the value type around also so we can check accesses
+   * to be sure the right member is being read. */
+  upb_ctype_t ctype;
+#endif
+} upb_value;
+
+#ifdef NDEBUG
+#define SET_TYPE(dest, val)      UPB_UNUSED(val)
+#else
+#define SET_TYPE(dest, val) dest = val
+#endif
+
+/* Like strdup(), which isn't always available since it's not ANSI C. */
+char *upb_strdup(const char *s);
+/* Variant that works with a length-delimited rather than NULL-delimited string,
+ * as supported by strtable. */
+char *upb_strdup2(const char *s, size_t len);
+
+UPB_INLINE void _upb_value_setval(upb_value *v, uint64_t val,
+                                  upb_ctype_t ctype) {
+  v->val = val;
+  SET_TYPE(v->ctype, ctype);
+}
+
+UPB_INLINE upb_value _upb_value_val(uint64_t val, upb_ctype_t ctype) {
+  upb_value ret;
+  _upb_value_setval(&ret, val, ctype);
+  return ret;
+}
+
+/* For each value ctype, define the following set of functions:
+ *
+ * // Get/set an int32 from a upb_value.
+ * int32_t upb_value_getint32(upb_value val);
+ * void upb_value_setint32(upb_value *val, int32_t cval);
+ *
+ * // Construct a new upb_value from an int32.
+ * upb_value upb_value_int32(int32_t val); */
+#define FUNCS(name, membername, type_t, converter, proto_type) \
+  UPB_INLINE void upb_value_set ## name(upb_value *val, type_t cval) { \
+    val->val = (converter)cval; \
+    SET_TYPE(val->ctype, proto_type); \
+  } \
+  UPB_INLINE upb_value upb_value_ ## name(type_t val) { \
+    upb_value ret; \
+    upb_value_set ## name(&ret, val); \
+    return ret; \
+  } \
+  UPB_INLINE type_t upb_value_get ## name(upb_value val) { \
+    assert(val.ctype == proto_type); \
+    return (type_t)(converter)val.val; \
+  }
+
+FUNCS(int32,    int32,        int32_t,      int32_t,    UPB_CTYPE_INT32)
+FUNCS(int64,    int64,        int64_t,      int64_t,    UPB_CTYPE_INT64)
+FUNCS(uint32,   uint32,       uint32_t,     uint32_t,   UPB_CTYPE_UINT32)
+FUNCS(uint64,   uint64,       uint64_t,     uint64_t,   UPB_CTYPE_UINT64)
+FUNCS(bool,     _bool,        bool,         bool,       UPB_CTYPE_BOOL)
+FUNCS(cstr,     cstr,         char*,        uintptr_t,  UPB_CTYPE_CSTR)
+FUNCS(ptr,      ptr,          void*,        uintptr_t,  UPB_CTYPE_PTR)
+FUNCS(constptr, constptr,     const void*,  uintptr_t,  UPB_CTYPE_CONSTPTR)
+FUNCS(fptr,     fptr,         upb_func*,    uintptr_t,  UPB_CTYPE_FPTR)
+
+#undef FUNCS
+#undef SET_TYPE
+
+
+/* upb_tabkey *****************************************************************/
+
+/* Either:
+ *   1. an actual integer key, or
+ *   2. a pointer to a string prefixed by its uint32_t length, owned by us.
+ *
+ * ...depending on whether this is a string table or an int table.  We would
+ * make this a union of those two types, but C89 doesn't support statically
+ * initializing a non-first union member. */
+typedef uintptr_t upb_tabkey;
+
+#define UPB_TABKEY_NUM(n) n
+#define UPB_TABKEY_NONE 0
+/* The preprocessor isn't quite powerful enough to turn the compile-time string
+ * length into a byte-wise string representation, so code generation needs to
+ * help it along.
+ *
+ * "len1" is the low byte and len4 is the high byte. */
+#ifdef UPB_BIG_ENDIAN
+#define UPB_TABKEY_STR(len1, len2, len3, len4, strval) \
+    (uintptr_t)(len4 len3 len2 len1 strval)
+#else
+#define UPB_TABKEY_STR(len1, len2, len3, len4, strval) \
+    (uintptr_t)(len1 len2 len3 len4 strval)
+#endif
+
+UPB_INLINE char *upb_tabstr(upb_tabkey key, uint32_t *len) {
+  char* mem = (char*)key;
+  if (len) memcpy(len, mem, sizeof(*len));
+  return mem + sizeof(*len);
+}
+
+
+/* upb_tabval *****************************************************************/
+
+#ifdef __cplusplus
+
+/* Status initialization not supported.
+ *
+ * This separate definition is necessary because in C++, UINTPTR_MAX isn't
+ * reliably available. */
+typedef struct {
+  uint64_t val;
+} upb_tabval;
+
+#else
+
+/* C -- supports static initialization, but to support static initialization of
+ * both integers and points for both 32 and 64 bit targets, it takes a little
+ * bit of doing. */
+
+#if UINTPTR_MAX == 0xffffffffffffffffULL
+#define UPB_PTR_IS_64BITS
+#elif UINTPTR_MAX != 0xffffffff
+#error Could not determine how many bits pointers are.
+#endif
+
+typedef union {
+  /* For static initialization.
+   *
+   * Unfortunately this ugliness is necessary -- it is the only way that we can,
+   * with -std=c89 -pedantic, statically initialize this to either a pointer or
+   * an integer on 32-bit platforms. */
+  struct {
+#ifdef UPB_PTR_IS_64BITS
+    uintptr_t val;
+#else
+    uintptr_t val1;
+    uintptr_t val2;
+#endif
+  } staticinit;
+
+  /* The normal accessor that we use for everything at runtime. */
+  uint64_t val;
+} upb_tabval;
+
+#ifdef UPB_PTR_IS_64BITS
+#define UPB_TABVALUE_INT_INIT(v) {{v}}
+#define UPB_TABVALUE_EMPTY_INIT  {{-1}}
+#else
+
+/* 32-bit pointers */
+
+#ifdef UPB_BIG_ENDIAN
+#define UPB_TABVALUE_INT_INIT(v) {{0, v}}
+#define UPB_TABVALUE_EMPTY_INIT  {{-1, -1}}
+#else
+#define UPB_TABVALUE_INT_INIT(v) {{v, 0}}
+#define UPB_TABVALUE_EMPTY_INIT  {{-1, -1}}
+#endif
+
+#endif
+
+#define UPB_TABVALUE_PTR_INIT(v) UPB_TABVALUE_INT_INIT((uintptr_t)v)
+
+#undef UPB_PTR_IS_64BITS
+
+#endif  /* __cplusplus */
+
+
+/* upb_table ******************************************************************/
+
+typedef struct _upb_tabent {
+  upb_tabkey key;
+  upb_tabval val;
+
+  /* Internal chaining.  This is const so we can create static initializers for
+   * tables.  We cast away const sometimes, but *only* when the containing
+   * upb_table is known to be non-const.  This requires a bit of care, but
+   * the subtlety is confined to table.c. */
+  const struct _upb_tabent *next;
+} upb_tabent;
+
+typedef struct {
+  size_t count;          /* Number of entries in the hash part. */
+  size_t mask;           /* Mask to turn hash value -> bucket. */
+  upb_ctype_t ctype;     /* Type of all values. */
+  uint8_t size_lg2;      /* Size of the hashtable part is 2^size_lg2 entries. */
+
+  /* Hash table entries.
+   * Making this const isn't entirely accurate; what we really want is for it to
+   * have the same const-ness as the table it's inside.  But there's no way to
+   * declare that in C.  So we have to make it const so that we can statically
+   * initialize const hash tables.  Then we cast away const when we have to.
+   */
+  const upb_tabent *entries;
+} upb_table;
+
+typedef struct {
+  upb_table t;
+} upb_strtable;
+
+#define UPB_STRTABLE_INIT(count, mask, ctype, size_lg2, entries) \
+  {{count, mask, ctype, size_lg2, entries}}
+
+#define UPB_EMPTY_STRTABLE_INIT(ctype)                           \
+  UPB_STRTABLE_INIT(0, 0, ctype, 0, NULL)
+
+typedef struct {
+  upb_table t;              /* For entries that don't fit in the array part. */
+  const upb_tabval *array;  /* Array part of the table. See const note above. */
+  size_t array_size;        /* Array part size. */
+  size_t array_count;       /* Array part number of elements. */
+} upb_inttable;
+
+#define UPB_INTTABLE_INIT(count, mask, ctype, size_lg2, ent, a, asize, acount) \
+  {{count, mask, ctype, size_lg2, ent}, a, asize, acount}
+
+#define UPB_EMPTY_INTTABLE_INIT(ctype) \
+  UPB_INTTABLE_INIT(0, 0, ctype, 0, NULL, NULL, 0, 0)
+
+#define UPB_ARRAY_EMPTYENT -1
+
+UPB_INLINE size_t upb_table_size(const upb_table *t) {
+  if (t->size_lg2 == 0)
+    return 0;
+  else
+    return 1 << t->size_lg2;
+}
+
+/* Internal-only functions, in .h file only out of necessity. */
+UPB_INLINE bool upb_tabent_isempty(const upb_tabent *e) {
+  return e->key == 0;
+}
+
+/* Used by some of the unit tests for generic hashing functionality. */
+uint32_t MurmurHash2(const void * key, size_t len, uint32_t seed);
+
+UPB_INLINE uintptr_t upb_intkey(uintptr_t key) {
+  return key;
+}
+
+UPB_INLINE uint32_t upb_inthash(uintptr_t key) {
+  return (uint32_t)key;
+}
+
+static const upb_tabent *upb_getentry(const upb_table *t, uint32_t hash) {
+  return t->entries + (hash & t->mask);
+}
+
+UPB_INLINE bool upb_arrhas(upb_tabval key) {
+  return key.val != (uint64_t)-1;
+}
+
+/* Initialize and uninitialize a table, respectively.  If memory allocation
+ * failed, false is returned that the table is uninitialized. */
+bool upb_inttable_init(upb_inttable *table, upb_ctype_t ctype);
+bool upb_strtable_init(upb_strtable *table, upb_ctype_t ctype);
+void upb_inttable_uninit(upb_inttable *table);
+void upb_strtable_uninit(upb_strtable *table);
+
+/* Returns the number of values in the table. */
+size_t upb_inttable_count(const upb_inttable *t);
+UPB_INLINE size_t upb_strtable_count(const upb_strtable *t) {
+  return t->t.count;
+}
+
+/* Inserts the given key into the hashtable with the given value.  The key must
+ * not already exist in the hash table.  For string tables, the key must be
+ * NULL-terminated, and the table will make an internal copy of the key.
+ * Inttables must not insert a value of UINTPTR_MAX.
+ *
+ * If a table resize was required but memory allocation failed, false is
+ * returned and the table is unchanged. */
+bool upb_inttable_insert(upb_inttable *t, uintptr_t key, upb_value val);
+bool upb_strtable_insert2(upb_strtable *t, const char *key, size_t len,
+                          upb_value val);
+
+/* For NULL-terminated strings. */
+UPB_INLINE bool upb_strtable_insert(upb_strtable *t, const char *key,
+                                    upb_value val) {
+  return upb_strtable_insert2(t, key, strlen(key), val);
+}
+
+/* Looks up key in this table, returning "true" if the key was found.
+ * If v is non-NULL, copies the value for this key into *v. */
+bool upb_inttable_lookup(const upb_inttable *t, uintptr_t key, upb_value *v);
+bool upb_strtable_lookup2(const upb_strtable *t, const char *key, size_t len,
+                          upb_value *v);
+
+/* For NULL-terminated strings. */
+UPB_INLINE bool upb_strtable_lookup(const upb_strtable *t, const char *key,
+                                    upb_value *v) {
+  return upb_strtable_lookup2(t, key, strlen(key), v);
+}
+
+/* Removes an item from the table.  Returns true if the remove was successful,
+ * and stores the removed item in *val if non-NULL. */
+bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val);
+bool upb_strtable_remove2(upb_strtable *t, const char *key, size_t len,
+                          upb_value *val);
+
+/* For NULL-terminated strings. */
+UPB_INLINE bool upb_strtable_remove(upb_strtable *t, const char *key,
+                                    upb_value *v) {
+  return upb_strtable_remove2(t, key, strlen(key), v);
+}
+
+/* Updates an existing entry in an inttable.  If the entry does not exist,
+ * returns false and does nothing.  Unlike insert/remove, this does not
+ * invalidate iterators. */
+bool upb_inttable_replace(upb_inttable *t, uintptr_t key, upb_value val);
+
+/* Handy routines for treating an inttable like a stack.  May not be mixed with
+ * other insert/remove calls. */
+bool upb_inttable_push(upb_inttable *t, upb_value val);
+upb_value upb_inttable_pop(upb_inttable *t);
+
+/* Convenience routines for inttables with pointer keys. */
+bool upb_inttable_insertptr(upb_inttable *t, const void *key, upb_value val);
+bool upb_inttable_removeptr(upb_inttable *t, const void *key, upb_value *val);
+bool upb_inttable_lookupptr(
+    const upb_inttable *t, const void *key, upb_value *val);
+
+/* Optimizes the table for the current set of entries, for both memory use and
+ * lookup time.  Client should call this after all entries have been inserted;
+ * inserting more entries is legal, but will likely require a table resize. */
+void upb_inttable_compact(upb_inttable *t);
+
+/* A special-case inlinable version of the lookup routine for 32-bit
+ * integers. */
+UPB_INLINE bool upb_inttable_lookup32(const upb_inttable *t, uint32_t key,
+                                      upb_value *v) {
+  *v = upb_value_int32(0);  /* Silence compiler warnings. */
+  if (key < t->array_size) {
+    upb_tabval arrval = t->array[key];
+    if (upb_arrhas(arrval)) {
+      _upb_value_setval(v, arrval.val, t->t.ctype);
+      return true;
+    } else {
+      return false;
+    }
+  } else {
+    const upb_tabent *e;
+    if (t->t.entries == NULL) return false;
+    for (e = upb_getentry(&t->t, upb_inthash(key)); true; e = e->next) {
+      if ((uint32_t)e->key == key) {
+        _upb_value_setval(v, e->val.val, t->t.ctype);
+        return true;
+      }
+      if (e->next == NULL) return false;
+    }
+  }
+}
+
+/* Exposed for testing only. */
+bool upb_strtable_resize(upb_strtable *t, size_t size_lg2);
+
+/* Iterators ******************************************************************/
+
+/* Iterators for int and string tables.  We are subject to some kind of unusual
+ * design constraints:
+ *
+ * For high-level languages:
+ *  - we must be able to guarantee that we don't crash or corrupt memory even if
+ *    the program accesses an invalidated iterator.
+ *
+ * For C++11 range-based for:
+ *  - iterators must be copyable
+ *  - iterators must be comparable
+ *  - it must be possible to construct an "end" value.
+ *
+ * Iteration order is undefined.
+ *
+ * Modifying the table invalidates iterators.  upb_{str,int}table_done() is
+ * guaranteed to work even on an invalidated iterator, as long as the table it
+ * is iterating over has not been freed.  Calling next() or accessing data from
+ * an invalidated iterator yields unspecified elements from the table, but it is
+ * guaranteed not to crash and to return real table elements (except when done()
+ * is true). */
+
+
+/* upb_strtable_iter **********************************************************/
+
+/*   upb_strtable_iter i;
+ *   upb_strtable_begin(&i, t);
+ *   for(; !upb_strtable_done(&i); upb_strtable_next(&i)) {
+ *     const char *key = upb_strtable_iter_key(&i);
+ *     const upb_value val = upb_strtable_iter_value(&i);
+ *     // ...
+ *   }
+ */
+
+typedef struct {
+  const upb_strtable *t;
+  size_t index;
+} upb_strtable_iter;
+
+void upb_strtable_begin(upb_strtable_iter *i, const upb_strtable *t);
+void upb_strtable_next(upb_strtable_iter *i);
+bool upb_strtable_done(const upb_strtable_iter *i);
+const char *upb_strtable_iter_key(upb_strtable_iter *i);
+size_t upb_strtable_iter_keylength(upb_strtable_iter *i);
+upb_value upb_strtable_iter_value(const upb_strtable_iter *i);
+void upb_strtable_iter_setdone(upb_strtable_iter *i);
+bool upb_strtable_iter_isequal(const upb_strtable_iter *i1,
+                               const upb_strtable_iter *i2);
+
+
+/* upb_inttable_iter **********************************************************/
+
+/*   upb_inttable_iter i;
+ *   upb_inttable_begin(&i, t);
+ *   for(; !upb_inttable_done(&i); upb_inttable_next(&i)) {
+ *     uintptr_t key = upb_inttable_iter_key(&i);
+ *     upb_value val = upb_inttable_iter_value(&i);
+ *     // ...
+ *   }
+ */
+
+typedef struct {
+  const upb_inttable *t;
+  size_t index;
+  bool array_part;
+} upb_inttable_iter;
+
+void upb_inttable_begin(upb_inttable_iter *i, const upb_inttable *t);
+void upb_inttable_next(upb_inttable_iter *i);
+bool upb_inttable_done(const upb_inttable_iter *i);
+uintptr_t upb_inttable_iter_key(const upb_inttable_iter *i);
+upb_value upb_inttable_iter_value(const upb_inttable_iter *i);
+void upb_inttable_iter_setdone(upb_inttable_iter *i);
+bool upb_inttable_iter_isequal(const upb_inttable_iter *i1,
+                               const upb_inttable_iter *i2);
+
+
+#ifdef __cplusplus
+}  /* extern "C" */
+#endif
+
+#endif  /* UPB_TABLE_H_ */
+
+/* Reference tracking will check ref()/unref() operations to make sure the
+ * ref ownership is correct.  Where possible it will also make tools like
+ * Valgrind attribute ref leaks to the code that took the leaked ref, not
+ * the code that originally created the object.
+ *
+ * Enabling this requires the application to define upb_lock()/upb_unlock()
+ * functions that acquire/release a global mutex (or #define UPB_THREAD_UNSAFE).
+ * For this reason we don't enable it by default, even in debug builds.
+ */
+
+/* #define UPB_DEBUG_REFS */
+
+#ifdef __cplusplus
+namespace upb { class RefCounted; }
+#endif
+
+UPB_DECLARE_TYPE(upb::RefCounted, upb_refcounted)
+
+struct upb_refcounted_vtbl;
+
+#ifdef __cplusplus
+
+class upb::RefCounted {
+ public:
+  /* Returns true if the given object is frozen. */
+  bool IsFrozen() const;
+
+  /* Increases the ref count, the new ref is owned by "owner" which must not
+   * already own a ref (and should not itself be a refcounted object if the ref
+   * could possibly be circular; see below).
+   * Thread-safe iff "this" is frozen. */
+  void Ref(const void *owner) const;
+
+  /* Release a ref that was acquired from upb_refcounted_ref() and collects any
+   * objects it can. */
+  void Unref(const void *owner) const;
+
+  /* Moves an existing ref from "from" to "to", without changing the overall
+   * ref count.  DonateRef(foo, NULL, owner) is the same as Ref(foo, owner),
+   * but "to" may not be NULL. */
+  void DonateRef(const void *from, const void *to) const;
+
+  /* Verifies that a ref to the given object is currently held by the given
+   * owner.  Only effective in UPB_DEBUG_REFS builds. */
+  void CheckRef(const void *owner) const;
+
+ private:
+  UPB_DISALLOW_POD_OPS(RefCounted, upb::RefCounted)
+#else
+struct upb_refcounted {
+#endif
+  /* TODO(haberman): move the actual structure definition to structdefs.int.h.
+   * The only reason they are here is because inline functions need to see the
+   * definition of upb_handlers, which needs to see this definition.  But we
+   * can change the upb_handlers inline functions to deal in raw offsets
+   * instead.
+   */
+
+  /* A single reference count shared by all objects in the group. */
+  uint32_t *group;
+
+  /* A singly-linked list of all objects in the group. */
+  upb_refcounted *next;
+
+  /* Table of function pointers for this type. */
+  const struct upb_refcounted_vtbl *vtbl;
+
+  /* Maintained only when mutable, this tracks the number of refs (but not
+   * ref2's) to this object.  *group should be the sum of all individual_count
+   * in the group. */
+  uint32_t individual_count;
+
+  bool is_frozen;
+
+#ifdef UPB_DEBUG_REFS
+  upb_inttable *refs;  /* Maps owner -> trackedref for incoming refs. */
+  upb_inttable *ref2s; /* Set of targets for outgoing ref2s. */
+#endif
+};
+
+#ifdef UPB_DEBUG_REFS
+#define UPB_REFCOUNT_INIT(refs, ref2s) \
+    {&static_refcount, NULL, NULL, 0, true, refs, ref2s}
+#else
+#define UPB_REFCOUNT_INIT(refs, ref2s) {&static_refcount, NULL, NULL, 0, true}
+#endif
+
+UPB_BEGIN_EXTERN_C
+
+/* It is better to use tracked refs when possible, for the extra debugging
+ * capability.  But if this is not possible (because you don't have easy access
+ * to a stable pointer value that is associated with the ref), you can pass
+ * UPB_UNTRACKED_REF instead.  */
+extern const void *UPB_UNTRACKED_REF;
+
+/* Native C API. */
+bool upb_refcounted_isfrozen(const upb_refcounted *r);
+void upb_refcounted_ref(const upb_refcounted *r, const void *owner);
+void upb_refcounted_unref(const upb_refcounted *r, const void *owner);
+void upb_refcounted_donateref(
+    const upb_refcounted *r, const void *from, const void *to);
+void upb_refcounted_checkref(const upb_refcounted *r, const void *owner);
+
+#define UPB_REFCOUNTED_CMETHODS(type, upcastfunc) \
+  UPB_INLINE bool type ## _isfrozen(const type *v) { \
+    return upb_refcounted_isfrozen(upcastfunc(v)); \
+  } \
+  UPB_INLINE void type ## _ref(const type *v, const void *owner) { \
+    upb_refcounted_ref(upcastfunc(v), owner); \
+  } \
+  UPB_INLINE void type ## _unref(const type *v, const void *owner) { \
+    upb_refcounted_unref(upcastfunc(v), owner); \
+  } \
+  UPB_INLINE void type ## _donateref(const type *v, const void *from, const void *to) { \
+    upb_refcounted_donateref(upcastfunc(v), from, to); \
+  } \
+  UPB_INLINE void type ## _checkref(const type *v, const void *owner) { \
+    upb_refcounted_checkref(upcastfunc(v), owner); \
+  }
+
+#define UPB_REFCOUNTED_CPPMETHODS \
+  bool IsFrozen() const { \
+    return upb::upcast_to<const upb::RefCounted>(this)->IsFrozen(); \
+  } \
+  void Ref(const void *owner) const { \
+    return upb::upcast_to<const upb::RefCounted>(this)->Ref(owner); \
+  } \
+  void Unref(const void *owner) const { \
+    return upb::upcast_to<const upb::RefCounted>(this)->Unref(owner); \
+  } \
+  void DonateRef(const void *from, const void *to) const { \
+    return upb::upcast_to<const upb::RefCounted>(this)->DonateRef(from, to); \
+  } \
+  void CheckRef(const void *owner) const { \
+    return upb::upcast_to<const upb::RefCounted>(this)->CheckRef(owner); \
+  }
+
+/* Internal-to-upb Interface **************************************************/
+
+typedef void upb_refcounted_visit(const upb_refcounted *r,
+                                  const upb_refcounted *subobj,
+                                  void *closure);
+
+struct upb_refcounted_vtbl {
+  /* Must visit all subobjects that are currently ref'd via upb_refcounted_ref2.
+   * Must be longjmp()-safe. */
+  void (*visit)(const upb_refcounted *r, upb_refcounted_visit *visit, void *c);
+
+  /* Must free the object and release all references to other objects. */
+  void (*free)(upb_refcounted *r);
+};
+
+/* Initializes the refcounted with a single ref for the given owner.  Returns
+ * false if memory could not be allocated. */
+bool upb_refcounted_init(upb_refcounted *r,
+                         const struct upb_refcounted_vtbl *vtbl,
+                         const void *owner);
+
+/* Adds a ref from one refcounted object to another ("from" must not already
+ * own a ref).  These refs may be circular; cycles will be collected correctly
+ * (if conservatively).  These refs do not need to be freed in from's free()
+ * function. */
+void upb_refcounted_ref2(const upb_refcounted *r, upb_refcounted *from);
+
+/* Removes a ref that was acquired from upb_refcounted_ref2(), and collects any
+ * object it can.  This is only necessary when "from" no longer points to "r",
+ * and not from from's "free" function. */
+void upb_refcounted_unref2(const upb_refcounted *r, upb_refcounted *from);
+
+#define upb_ref2(r, from) \
+    upb_refcounted_ref2((const upb_refcounted*)r, (upb_refcounted*)from)
+#define upb_unref2(r, from) \
+    upb_refcounted_unref2((const upb_refcounted*)r, (upb_refcounted*)from)
+
+/* Freezes all mutable object reachable by ref2() refs from the given roots.
+ * This will split refcounting groups into precise SCC groups, so that
+ * refcounting of frozen objects can be more aggressive.  If memory allocation
+ * fails, or if more than 2**31 mutable objects are reachable from "roots", or
+ * if the maximum depth of the graph exceeds "maxdepth", false is returned and
+ * the objects are unchanged.
+ *
+ * After this operation succeeds, the objects are frozen/const, and may not be
+ * used through non-const pointers.  In particular, they may not be passed as
+ * the second parameter of upb_refcounted_{ref,unref}2().  On the upside, all
+ * operations on frozen refcounteds are threadsafe, and objects will be freed
+ * at the precise moment that they become unreachable.
+ *
+ * Caller must own refs on each object in the "roots" list. */
+bool upb_refcounted_freeze(upb_refcounted *const*roots, int n, upb_status *s,
+                           int maxdepth);
+
+/* Shared by all compiled-in refcounted objects. */
+extern uint32_t static_refcount;
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+/* C++ Wrappers. */
+namespace upb {
+inline bool RefCounted::IsFrozen() const {
+  return upb_refcounted_isfrozen(this);
+}
+inline void RefCounted::Ref(const void *owner) const {
+  upb_refcounted_ref(this, owner);
+}
+inline void RefCounted::Unref(const void *owner) const {
+  upb_refcounted_unref(this, owner);
+}
+inline void RefCounted::DonateRef(const void *from, const void *to) const {
+  upb_refcounted_donateref(this, from, to);
+}
+inline void RefCounted::CheckRef(const void *owner) const {
+  upb_refcounted_checkref(this, owner);
+}
+}  /* namespace upb */
+#endif
+
+#endif  /* UPB_REFCOUNT_H_ */
+
+#ifdef __cplusplus
+#include <cstring>
+#include <string>
+#include <vector>
+
+namespace upb {
+class Def;
+class EnumDef;
+class FieldDef;
+class MessageDef;
+class OneofDef;
+}
+#endif
+
+UPB_DECLARE_DERIVED_TYPE(upb::Def, upb::RefCounted, upb_def, upb_refcounted)
+
+/* The maximum message depth that the type graph can have.  This is a resource
+ * limit for the C stack since we sometimes need to recursively traverse the
+ * graph.  Cycles are ok; the traversal will stop when it detects a cycle, but
+ * we must hit the cycle before the maximum depth is reached.
+ *
+ * If having a single static limit is too inflexible, we can add another variant
+ * of Def::Freeze that allows specifying this as a parameter. */
+#define UPB_MAX_MESSAGE_DEPTH 64
+
+
+/* upb::Def: base class for defs  *********************************************/
+
+/* All the different kind of defs we support.  These correspond 1:1 with
+ * declarations in a .proto file. */
+typedef enum {
+  UPB_DEF_MSG,
+  UPB_DEF_FIELD,
+  UPB_DEF_ENUM,
+  UPB_DEF_ONEOF,
+  UPB_DEF_SERVICE,   /* Not yet implemented. */
+  UPB_DEF_ANY = -1   /* Wildcard for upb_symtab_get*() */
+} upb_deftype_t;
+
+#ifdef __cplusplus
+
+/* The base class of all defs.  Its base is upb::RefCounted (use upb::upcast()
+ * to convert). */
+class upb::Def {
+ public:
+  typedef upb_deftype_t Type;
+
+  Def* Dup(const void *owner) const;
+
+  /* upb::RefCounted methods like Ref()/Unref(). */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  Type def_type() const;
+
+  /* "fullname" is the def's fully-qualified name (eg. foo.bar.Message). */
+  const char *full_name() const;
+
+  /* The def must be mutable.  Caller retains ownership of fullname.  Defs are
+   * not required to have a name; if a def has no name when it is frozen, it
+   * will remain an anonymous def.  On failure, returns false and details in "s"
+   * if non-NULL. */
+  bool set_full_name(const char* fullname, upb::Status* s);
+  bool set_full_name(const std::string &fullname, upb::Status* s);
+
+  /* Freezes the given defs; this validates all constraints and marks the defs
+   * as frozen (read-only).  "defs" may not contain any fielddefs, but fields
+   * of any msgdefs will be frozen.
+   *
+   * Symbolic references to sub-types and enum defaults must have already been
+   * resolved.  Any mutable defs reachable from any of "defs" must also be in
+   * the list; more formally, "defs" must be a transitive closure of mutable
+   * defs.
+   *
+   * After this operation succeeds, the finalized defs must only be accessed
+   * through a const pointer! */
+  static bool Freeze(Def* const* defs, int n, Status* status);
+  static bool Freeze(const std::vector<Def*>& defs, Status* status);
+
+ private:
+  UPB_DISALLOW_POD_OPS(Def, upb::Def)
+};
+
+#endif  /* __cplusplus */
+
+UPB_BEGIN_EXTERN_C
+
+/* Native C API. */
+upb_def *upb_def_dup(const upb_def *def, const void *owner);
+
+/* Include upb_refcounted methods like upb_def_ref()/upb_def_unref(). */
+UPB_REFCOUNTED_CMETHODS(upb_def, upb_def_upcast)
+
+upb_deftype_t upb_def_type(const upb_def *d);
+const char *upb_def_fullname(const upb_def *d);
+bool upb_def_setfullname(upb_def *def, const char *fullname, upb_status *s);
+bool upb_def_freeze(upb_def *const *defs, int n, upb_status *s);
+
+UPB_END_EXTERN_C
+
+
+/* upb::Def casts *************************************************************/
+
+#ifdef __cplusplus
+#define UPB_CPP_CASTS(cname, cpptype)                                          \
+  namespace upb {                                                              \
+  template <>                                                                  \
+  inline cpptype *down_cast<cpptype *, Def>(Def * def) {                       \
+    return upb_downcast_##cname##_mutable(def);                                \
+  }                                                                            \
+  template <>                                                                  \
+  inline cpptype *dyn_cast<cpptype *, Def>(Def * def) {                        \
+    return upb_dyncast_##cname##_mutable(def);                                 \
+  }                                                                            \
+  template <>                                                                  \
+  inline const cpptype *down_cast<const cpptype *, const Def>(                 \
+      const Def *def) {                                                        \
+    return upb_downcast_##cname(def);                                          \
+  }                                                                            \
+  template <>                                                                  \
+  inline const cpptype *dyn_cast<const cpptype *, const Def>(const Def *def) { \
+    return upb_dyncast_##cname(def);                                           \
+  }                                                                            \
+  template <>                                                                  \
+  inline const cpptype *down_cast<const cpptype *, Def>(Def * def) {           \
+    return upb_downcast_##cname(def);                                          \
+  }                                                                            \
+  template <>                                                                  \
+  inline const cpptype *dyn_cast<const cpptype *, Def>(Def * def) {            \
+    return upb_dyncast_##cname(def);                                           \
+  }                                                                            \
+  }  /* namespace upb */
+#else
+#define UPB_CPP_CASTS(cname, cpptype)
+#endif  /* __cplusplus */
+
+/* Dynamic casts, for determining if a def is of a particular type at runtime.
+ * Downcasts, for when some wants to assert that a def is of a particular type.
+ * These are only checked if we are building debug. */
+#define UPB_DEF_CASTS(lower, upper, cpptype)                               \
+  UPB_INLINE const upb_##lower *upb_dyncast_##lower(const upb_def *def) {  \
+    if (upb_def_type(def) != UPB_DEF_##upper) return NULL;                 \
+    return (upb_##lower *)def;                                             \
+  }                                                                        \
+  UPB_INLINE const upb_##lower *upb_downcast_##lower(const upb_def *def) { \
+    assert(upb_def_type(def) == UPB_DEF_##upper);                          \
+    return (const upb_##lower *)def;                                       \
+  }                                                                        \
+  UPB_INLINE upb_##lower *upb_dyncast_##lower##_mutable(upb_def *def) {    \
+    return (upb_##lower *)upb_dyncast_##lower(def);                        \
+  }                                                                        \
+  UPB_INLINE upb_##lower *upb_downcast_##lower##_mutable(upb_def *def) {   \
+    return (upb_##lower *)upb_downcast_##lower(def);                       \
+  }                                                                        \
+  UPB_CPP_CASTS(lower, cpptype)
+
+#define UPB_DEFINE_DEF(cppname, lower, upper, cppmethods, members)             \
+  UPB_DEFINE_CLASS2(cppname, upb::Def, upb::RefCounted, cppmethods,            \
+                   members)                                                    \
+  UPB_DEF_CASTS(lower, upper, cppname)
+
+#define UPB_DECLARE_DEF_TYPE(cppname, lower, upper) \
+  UPB_DECLARE_DERIVED_TYPE2(cppname, upb::Def, upb::RefCounted, \
+                            upb_ ## lower, upb_def, upb_refcounted) \
+  UPB_DEF_CASTS(lower, upper, cppname)
+
+UPB_DECLARE_DEF_TYPE(upb::FieldDef, fielddef, FIELD)
+UPB_DECLARE_DEF_TYPE(upb::MessageDef, msgdef, MSG)
+UPB_DECLARE_DEF_TYPE(upb::EnumDef, enumdef, ENUM)
+UPB_DECLARE_DEF_TYPE(upb::OneofDef, oneofdef, ONEOF)
+
+#undef UPB_DECLARE_DEF_TYPE
+#undef UPB_DEF_CASTS
+#undef UPB_CPP_CASTS
+
+
+/* upb::FieldDef **************************************************************/
+
+/* The types a field can have.  Note that this list is not identical to the
+ * types defined in descriptor.proto, which gives INT32 and SINT32 separate
+ * types (we distinguish the two with the "integer encoding" enum below). */
+typedef enum {
+  UPB_TYPE_FLOAT    = 1,
+  UPB_TYPE_DOUBLE   = 2,
+  UPB_TYPE_BOOL     = 3,
+  UPB_TYPE_STRING   = 4,
+  UPB_TYPE_BYTES    = 5,
+  UPB_TYPE_MESSAGE  = 6,
+  UPB_TYPE_ENUM     = 7,  /* Enum values are int32. */
+  UPB_TYPE_INT32    = 8,
+  UPB_TYPE_UINT32   = 9,
+  UPB_TYPE_INT64    = 10,
+  UPB_TYPE_UINT64   = 11
+} upb_fieldtype_t;
+
+/* The repeated-ness of each field; this matches descriptor.proto. */
+typedef enum {
+  UPB_LABEL_OPTIONAL = 1,
+  UPB_LABEL_REQUIRED = 2,
+  UPB_LABEL_REPEATED = 3
+} upb_label_t;
+
+/* How integers should be encoded in serializations that offer multiple
+ * integer encoding methods. */
+typedef enum {
+  UPB_INTFMT_VARIABLE = 1,
+  UPB_INTFMT_FIXED = 2,
+  UPB_INTFMT_ZIGZAG = 3   /* Only for signed types (INT32/INT64). */
+} upb_intfmt_t;
+
+/* Descriptor types, as defined in descriptor.proto. */
+typedef enum {
+  UPB_DESCRIPTOR_TYPE_DOUBLE   = 1,
+  UPB_DESCRIPTOR_TYPE_FLOAT    = 2,
+  UPB_DESCRIPTOR_TYPE_INT64    = 3,
+  UPB_DESCRIPTOR_TYPE_UINT64   = 4,
+  UPB_DESCRIPTOR_TYPE_INT32    = 5,
+  UPB_DESCRIPTOR_TYPE_FIXED64  = 6,
+  UPB_DESCRIPTOR_TYPE_FIXED32  = 7,
+  UPB_DESCRIPTOR_TYPE_BOOL     = 8,
+  UPB_DESCRIPTOR_TYPE_STRING   = 9,
+  UPB_DESCRIPTOR_TYPE_GROUP    = 10,
+  UPB_DESCRIPTOR_TYPE_MESSAGE  = 11,
+  UPB_DESCRIPTOR_TYPE_BYTES    = 12,
+  UPB_DESCRIPTOR_TYPE_UINT32   = 13,
+  UPB_DESCRIPTOR_TYPE_ENUM     = 14,
+  UPB_DESCRIPTOR_TYPE_SFIXED32 = 15,
+  UPB_DESCRIPTOR_TYPE_SFIXED64 = 16,
+  UPB_DESCRIPTOR_TYPE_SINT32   = 17,
+  UPB_DESCRIPTOR_TYPE_SINT64   = 18
+} upb_descriptortype_t;
+
+/* Maximum field number allowed for FieldDefs.  This is an inherent limit of the
+ * protobuf wire format. */
+#define UPB_MAX_FIELDNUMBER ((1 << 29) - 1)
+
+#ifdef __cplusplus
+
+/* A upb_fielddef describes a single field in a message.  It is most often
+ * found as a part of a upb_msgdef, but can also stand alone to represent
+ * an extension.
+ *
+ * Its base class is upb::Def (use upb::upcast() to convert). */
+class upb::FieldDef {
+ public:
+  typedef upb_fieldtype_t Type;
+  typedef upb_label_t Label;
+  typedef upb_intfmt_t IntegerFormat;
+  typedef upb_descriptortype_t DescriptorType;
+
+  /* These return true if the given value is a valid member of the enumeration. */
+  static bool CheckType(int32_t val);
+  static bool CheckLabel(int32_t val);
+  static bool CheckDescriptorType(int32_t val);
+  static bool CheckIntegerFormat(int32_t val);
+
+  /* These convert to the given enumeration; they require that the value is
+   * valid. */
+  static Type ConvertType(int32_t val);
+  static Label ConvertLabel(int32_t val);
+  static DescriptorType ConvertDescriptorType(int32_t val);
+  static IntegerFormat ConvertIntegerFormat(int32_t val);
+
+  /* Returns NULL if memory allocation failed. */
+  static reffed_ptr<FieldDef> New();
+
+  /* Duplicates the given field, returning NULL if memory allocation failed.
+   * When a fielddef is duplicated, the subdef (if any) is made symbolic if it
+   * wasn't already.  If the subdef is set but has no name (which is possible
+   * since msgdefs are not required to have a name) the new fielddef's subdef
+   * will be unset. */
+  FieldDef* Dup(const void* owner) const;
+
+  /* upb::RefCounted methods like Ref()/Unref(). */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  /* Functionality from upb::Def. */
+  const char* full_name() const;
+
+  bool type_is_set() const;  /* set_[descriptor_]type() has been called? */
+  Type type() const;         /* Requires that type_is_set() == true. */
+  Label label() const;       /* Defaults to UPB_LABEL_OPTIONAL. */
+  const char* name() const;  /* NULL if uninitialized. */
+  uint32_t number() const;   /* Returns 0 if uninitialized. */
+  bool is_extension() const;
+
+  /* For UPB_TYPE_MESSAGE fields only where is_tag_delimited() == false,
+   * indicates whether this field should have lazy parsing handlers that yield
+   * the unparsed string for the submessage.
+   *
+   * TODO(haberman): I think we want to move this into a FieldOptions container
+   * when we add support for custom options (the FieldOptions struct will
+   * contain both regular FieldOptions like "lazy" *and* custom options). */
+  bool lazy() const;
+
+  /* For non-string, non-submessage fields, this indicates whether binary
+   * protobufs are encoded in packed or non-packed format.
+   *
+   * TODO(haberman): see note above about putting options like this into a
+   * FieldOptions container. */
+  bool packed() const;
+
+  /* An integer that can be used as an index into an array of fields for
+   * whatever message this field belongs to.  Guaranteed to be less than
+   * f->containing_type()->field_count().  May only be accessed once the def has
+   * been finalized. */
+  int index() const;
+
+  /* The MessageDef to which this field belongs.
+   *
+   * If this field has been added to a MessageDef, that message can be retrieved
+   * directly (this is always the case for frozen FieldDefs).
+   *
+   * If the field has not yet been added to a MessageDef, you can set the name
+   * of the containing type symbolically instead.  This is mostly useful for
+   * extensions, where the extension is declared separately from the message. */
+  const MessageDef* containing_type() const;
+  const char* containing_type_name();
+
+  /* The OneofDef to which this field belongs, or NULL if this field is not part
+   * of a oneof. */
+  const OneofDef* containing_oneof() const;
+
+  /* The field's type according to the enum in descriptor.proto.  This is not
+   * the same as UPB_TYPE_*, because it distinguishes between (for example)
+   * INT32 and SINT32, whereas our "type" enum does not.  This return of
+   * descriptor_type() is a function of type(), integer_format(), and
+   * is_tag_delimited().  Likewise set_descriptor_type() sets all three
+   * appropriately. */
+  DescriptorType descriptor_type() const;
+
+  /* Convenient field type tests. */
+  bool IsSubMessage() const;
+  bool IsString() const;
+  bool IsSequence() const;
+  bool IsPrimitive() const;
+  bool IsMap() const;
+
+  /* How integers are encoded.  Only meaningful for integer types.
+   * Defaults to UPB_INTFMT_VARIABLE, and is reset when "type" changes. */
+  IntegerFormat integer_format() const;
+
+  /* Whether a submessage field is tag-delimited or not (if false, then
+   * length-delimited).  May only be set when type() == UPB_TYPE_MESSAGE. */
+  bool is_tag_delimited() const;
+
+  /* Returns the non-string default value for this fielddef, which may either
+   * be something the client set explicitly or the "default default" (0 for
+   * numbers, empty for strings).  The field's type indicates the type of the
+   * returned value, except for enum fields that are still mutable.
+   *
+   * Requires that the given function matches the field's current type. */
+  int64_t default_int64() const;
+  int32_t default_int32() const;
+  uint64_t default_uint64() const;
+  uint32_t default_uint32() const;
+  bool default_bool() const;
+  float default_float() const;
+  double default_double() const;
+
+  /* The resulting string is always NULL-terminated.  If non-NULL, the length
+   * will be stored in *len. */
+  const char *default_string(size_t* len) const;
+
+  /* For frozen UPB_TYPE_ENUM fields, enum defaults can always be read as either
+   * string or int32, and both of these methods will always return true.
+   *
+   * For mutable UPB_TYPE_ENUM fields, the story is a bit more complicated.
+   * Enum defaults are unusual. They can be specified either as string or int32,
+   * but to be valid the enum must have that value as a member.  And if no
+   * default is specified, the "default default" comes from the EnumDef.
+   *
+   * We allow reading the default as either an int32 or a string, but only if
+   * we have a meaningful value to report.  We have a meaningful value if it was
+   * set explicitly, or if we could get the "default default" from the EnumDef.
+   * Also if you explicitly set the name and we find the number in the EnumDef */
+  bool EnumHasStringDefault() const;
+  bool EnumHasInt32Default() const;
+
+  /* Submessage and enum fields must reference a "subdef", which is the
+   * upb::MessageDef or upb::EnumDef that defines their type.  Note that when
+   * the FieldDef is mutable it may not have a subdef *yet*, but this function
+   * still returns true to indicate that the field's type requires a subdef. */
+  bool HasSubDef() const;
+
+  /* Returns the enum or submessage def for this field, if any.  The field's
+   * type must match (ie. you may only call enum_subdef() for fields where
+   * type() == UPB_TYPE_ENUM).  Returns NULL if the subdef has not been set or
+   * is currently set symbolically. */
+  const EnumDef* enum_subdef() const;
+  const MessageDef* message_subdef() const;
+
+  /* Returns the generic subdef for this field.  Requires that HasSubDef() (ie.
+   * only works for UPB_TYPE_ENUM and UPB_TYPE_MESSAGE fields). */
+  const Def* subdef() const;
+
+  /* Returns the symbolic name of the subdef.  If the subdef is currently set
+   * unresolved (ie. set symbolically) returns the symbolic name.  If it has
+   * been resolved to a specific subdef, returns the name from that subdef. */
+  const char* subdef_name() const;
+
+  /* Setters (non-const methods), only valid for mutable FieldDefs! ***********/
+
+  bool set_full_name(const char* fullname, upb::Status* s);
+  bool set_full_name(const std::string& fullname, upb::Status* s);
+
+  /* This may only be called if containing_type() == NULL (ie. the field has not
+   * been added to a message yet). */
+  bool set_containing_type_name(const char *name, Status* status);
+  bool set_containing_type_name(const std::string& name, Status* status);
+
+  /* Defaults to false.  When we freeze, we ensure that this can only be true
+   * for length-delimited message fields.  Prior to freezing this can be true or
+   * false with no restrictions. */
+  void set_lazy(bool lazy);
+
+  /* Defaults to true.  Sets whether this field is encoded in packed format. */
+  void set_packed(bool packed);
+
+  /* "type" or "descriptor_type" MUST be set explicitly before the fielddef is
+   * finalized.  These setters require that the enum value is valid; if the
+   * value did not come directly from an enum constant, the caller should
+   * validate it first with the functions above (CheckFieldType(), etc). */
+  void set_type(Type type);
+  void set_label(Label label);
+  void set_descriptor_type(DescriptorType type);
+  void set_is_extension(bool is_extension);
+
+  /* "number" and "name" must be set before the FieldDef is added to a
+   * MessageDef, and may not be set after that.
+   *
+   * "name" is the same as full_name()/set_full_name(), but since fielddefs
+   * most often use simple, non-qualified names, we provide this accessor
+   * also.  Generally only extensions will want to think of this name as
+   * fully-qualified. */
+  bool set_number(uint32_t number, upb::Status* s);
+  bool set_name(const char* name, upb::Status* s);
+  bool set_name(const std::string& name, upb::Status* s);
+
+  void set_integer_format(IntegerFormat format);
+  bool set_tag_delimited(bool tag_delimited, upb::Status* s);
+
+  /* Sets default value for the field.  The call must exactly match the type
+   * of the field.  Enum fields may use either setint32 or setstring to set
+   * the default numerically or symbolically, respectively, but symbolic
+   * defaults must be resolved before finalizing (see ResolveEnumDefault()).
+   *
+   * Changing the type of a field will reset its default. */
+  void set_default_int64(int64_t val);
+  void set_default_int32(int32_t val);
+  void set_default_uint64(uint64_t val);
+  void set_default_uint32(uint32_t val);
+  void set_default_bool(bool val);
+  void set_default_float(float val);
+  void set_default_double(double val);
+  bool set_default_string(const void *str, size_t len, Status *s);
+  bool set_default_string(const std::string &str, Status *s);
+  void set_default_cstr(const char *str, Status *s);
+
+  /* Before a fielddef is frozen, its subdef may be set either directly (with a
+   * upb::Def*) or symbolically.  Symbolic refs must be resolved before the
+   * containing msgdef can be frozen (see upb_resolve() above).  upb always
+   * guarantees that any def reachable from a live def will also be kept alive.
+   *
+   * Both methods require that upb_hassubdef(f) (so the type must be set prior
+   * to calling these methods).  Returns false if this is not the case, or if
+   * the given subdef is not of the correct type.  The subdef is reset if the
+   * field's type is changed.  The subdef can be set to NULL to clear it. */
+  bool set_subdef(const Def* subdef, Status* s);
+  bool set_enum_subdef(const EnumDef* subdef, Status* s);
+  bool set_message_subdef(const MessageDef* subdef, Status* s);
+  bool set_subdef_name(const char* name, Status* s);
+  bool set_subdef_name(const std::string &name, Status* s);
+
+ private:
+  UPB_DISALLOW_POD_OPS(FieldDef, upb::FieldDef)
+};
+
+# endif  /* defined(__cplusplus) */
+
+UPB_BEGIN_EXTERN_C
+
+/* Native C API. */
+upb_fielddef *upb_fielddef_new(const void *owner);
+upb_fielddef *upb_fielddef_dup(const upb_fielddef *f, const void *owner);
+
+/* Include upb_refcounted methods like upb_fielddef_ref(). */
+UPB_REFCOUNTED_CMETHODS(upb_fielddef, upb_fielddef_upcast2)
+
+/* Methods from upb_def. */
+const char *upb_fielddef_fullname(const upb_fielddef *f);
+bool upb_fielddef_setfullname(upb_fielddef *f, const char *fullname,
+                              upb_status *s);
+
+bool upb_fielddef_typeisset(const upb_fielddef *f);
+upb_fieldtype_t upb_fielddef_type(const upb_fielddef *f);
+upb_descriptortype_t upb_fielddef_descriptortype(const upb_fielddef *f);
+upb_label_t upb_fielddef_label(const upb_fielddef *f);
+uint32_t upb_fielddef_number(const upb_fielddef *f);
+const char *upb_fielddef_name(const upb_fielddef *f);
+bool upb_fielddef_isextension(const upb_fielddef *f);
+bool upb_fielddef_lazy(const upb_fielddef *f);
+bool upb_fielddef_packed(const upb_fielddef *f);
+const upb_msgdef *upb_fielddef_containingtype(const upb_fielddef *f);
+const upb_oneofdef *upb_fielddef_containingoneof(const upb_fielddef *f);
+upb_msgdef *upb_fielddef_containingtype_mutable(upb_fielddef *f);
+const char *upb_fielddef_containingtypename(upb_fielddef *f);
+upb_intfmt_t upb_fielddef_intfmt(const upb_fielddef *f);
+uint32_t upb_fielddef_index(const upb_fielddef *f);
+bool upb_fielddef_istagdelim(const upb_fielddef *f);
+bool upb_fielddef_issubmsg(const upb_fielddef *f);
+bool upb_fielddef_isstring(const upb_fielddef *f);
+bool upb_fielddef_isseq(const upb_fielddef *f);
+bool upb_fielddef_isprimitive(const upb_fielddef *f);
+bool upb_fielddef_ismap(const upb_fielddef *f);
+int64_t upb_fielddef_defaultint64(const upb_fielddef *f);
+int32_t upb_fielddef_defaultint32(const upb_fielddef *f);
+uint64_t upb_fielddef_defaultuint64(const upb_fielddef *f);
+uint32_t upb_fielddef_defaultuint32(const upb_fielddef *f);
+bool upb_fielddef_defaultbool(const upb_fielddef *f);
+float upb_fielddef_defaultfloat(const upb_fielddef *f);
+double upb_fielddef_defaultdouble(const upb_fielddef *f);
+const char *upb_fielddef_defaultstr(const upb_fielddef *f, size_t *len);
+bool upb_fielddef_enumhasdefaultint32(const upb_fielddef *f);
+bool upb_fielddef_enumhasdefaultstr(const upb_fielddef *f);
+bool upb_fielddef_hassubdef(const upb_fielddef *f);
+const upb_def *upb_fielddef_subdef(const upb_fielddef *f);
+const upb_msgdef *upb_fielddef_msgsubdef(const upb_fielddef *f);
+const upb_enumdef *upb_fielddef_enumsubdef(const upb_fielddef *f);
+const char *upb_fielddef_subdefname(const upb_fielddef *f);
+
+void upb_fielddef_settype(upb_fielddef *f, upb_fieldtype_t type);
+void upb_fielddef_setdescriptortype(upb_fielddef *f, int type);
+void upb_fielddef_setlabel(upb_fielddef *f, upb_label_t label);
+bool upb_fielddef_setnumber(upb_fielddef *f, uint32_t number, upb_status *s);
+bool upb_fielddef_setname(upb_fielddef *f, const char *name, upb_status *s);
+bool upb_fielddef_setcontainingtypename(upb_fielddef *f, const char *name,
+                                        upb_status *s);
+void upb_fielddef_setisextension(upb_fielddef *f, bool is_extension);
+void upb_fielddef_setlazy(upb_fielddef *f, bool lazy);
+void upb_fielddef_setpacked(upb_fielddef *f, bool packed);
+void upb_fielddef_setintfmt(upb_fielddef *f, upb_intfmt_t fmt);
+void upb_fielddef_settagdelim(upb_fielddef *f, bool tag_delim);
+void upb_fielddef_setdefaultint64(upb_fielddef *f, int64_t val);
+void upb_fielddef_setdefaultint32(upb_fielddef *f, int32_t val);
+void upb_fielddef_setdefaultuint64(upb_fielddef *f, uint64_t val);
+void upb_fielddef_setdefaultuint32(upb_fielddef *f, uint32_t val);
+void upb_fielddef_setdefaultbool(upb_fielddef *f, bool val);
+void upb_fielddef_setdefaultfloat(upb_fielddef *f, float val);
+void upb_fielddef_setdefaultdouble(upb_fielddef *f, double val);
+bool upb_fielddef_setdefaultstr(upb_fielddef *f, const void *str, size_t len,
+                                upb_status *s);
+void upb_fielddef_setdefaultcstr(upb_fielddef *f, const char *str,
+                                 upb_status *s);
+bool upb_fielddef_setsubdef(upb_fielddef *f, const upb_def *subdef,
+                            upb_status *s);
+bool upb_fielddef_setmsgsubdef(upb_fielddef *f, const upb_msgdef *subdef,
+                               upb_status *s);
+bool upb_fielddef_setenumsubdef(upb_fielddef *f, const upb_enumdef *subdef,
+                                upb_status *s);
+bool upb_fielddef_setsubdefname(upb_fielddef *f, const char *name,
+                                upb_status *s);
+
+bool upb_fielddef_checklabel(int32_t label);
+bool upb_fielddef_checktype(int32_t type);
+bool upb_fielddef_checkdescriptortype(int32_t type);
+bool upb_fielddef_checkintfmt(int32_t fmt);
+
+UPB_END_EXTERN_C
+
+
+/* upb::MessageDef ************************************************************/
+
+typedef upb_inttable_iter upb_msg_field_iter;
+typedef upb_strtable_iter upb_msg_oneof_iter;
+
+#ifdef __cplusplus
+
+/* Structure that describes a single .proto message type.
+ *
+ * Its base class is upb::Def (use upb::upcast() to convert). */
+class upb::MessageDef {
+ public:
+  /* Returns NULL if memory allocation failed. */
+  static reffed_ptr<MessageDef> New();
+
+  /* upb::RefCounted methods like Ref()/Unref(). */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  /* Functionality from upb::Def. */
+  const char* full_name() const;
+  bool set_full_name(const char* fullname, Status* s);
+  bool set_full_name(const std::string& fullname, Status* s);
+
+  /* Call to freeze this MessageDef.
+   * WARNING: this will fail if this message has any unfrozen submessages!
+   * Messages with cycles must be frozen as a batch using upb::Def::Freeze(). */
+  bool Freeze(Status* s);
+
+  /* The number of fields that belong to the MessageDef. */
+  int field_count() const;
+
+  /* The number of oneofs that belong to the MessageDef. */
+  int oneof_count() const;
+
+  /* Adds a field (upb_fielddef object) to a msgdef.  Requires that the msgdef
+   * and the fielddefs are mutable.  The fielddef's name and number must be
+   * set, and the message may not already contain any field with this name or
+   * number, and this fielddef may not be part of another message.  In error
+   * cases false is returned and the msgdef is unchanged.
+   *
+   * If the given field is part of a oneof, this call succeeds if and only if
+   * that oneof is already part of this msgdef. (Note that adding a oneof to a
+   * msgdef automatically adds all of its fields to the msgdef at the time that
+   * the oneof is added, so it is usually more idiomatic to add the oneof's
+   * fields first then add the oneof to the msgdef. This case is supported for
+   * convenience.)
+   *
+   * If |f| is already part of this MessageDef, this method performs no action
+   * and returns true (success). Thus, this method is idempotent. */
+  bool AddField(FieldDef* f, Status* s);
+  bool AddField(const reffed_ptr<FieldDef>& f, Status* s);
+
+  /* Adds a oneof (upb_oneofdef object) to a msgdef. Requires that the msgdef,
+   * oneof, and any fielddefs are mutable, that the fielddefs contained in the
+   * oneof do not have any name or number conflicts with existing fields in the
+   * msgdef, and that the oneof's name is unique among all oneofs in the msgdef.
+   * If the oneof is added successfully, all of its fields will be added
+   * directly to the msgdef as well. In error cases, false is returned and the
+   * msgdef is unchanged. */
+  bool AddOneof(OneofDef* o, Status* s);
+  bool AddOneof(const reffed_ptr<OneofDef>& o, Status* s);
+
+  /* These return NULL if the field is not found. */
+  FieldDef* FindFieldByNumber(uint32_t number);
+  FieldDef* FindFieldByName(const char *name, size_t len);
+  const FieldDef* FindFieldByNumber(uint32_t number) const;
+  const FieldDef* FindFieldByName(const char* name, size_t len) const;
+
+
+  FieldDef* FindFieldByName(const char *name) {
+    return FindFieldByName(name, strlen(name));
+  }
+  const FieldDef* FindFieldByName(const char *name) const {
+    return FindFieldByName(name, strlen(name));
+  }
+
+  template <class T>
+  FieldDef* FindFieldByName(const T& str) {
+    return FindFieldByName(str.c_str(), str.size());
+  }
+  template <class T>
+  const FieldDef* FindFieldByName(const T& str) const {
+    return FindFieldByName(str.c_str(), str.size());
+  }
+
+  OneofDef* FindOneofByName(const char* name, size_t len);
+  const OneofDef* FindOneofByName(const char* name, size_t len) const;
+
+  OneofDef* FindOneofByName(const char* name) {
+    return FindOneofByName(name, strlen(name));
+  }
+  const OneofDef* FindOneofByName(const char* name) const {
+    return FindOneofByName(name, strlen(name));
+  }
+
+  template<class T>
+  OneofDef* FindOneofByName(const T& str) {
+    return FindOneofByName(str.c_str(), str.size());
+  }
+  template<class T>
+  const OneofDef* FindOneofByName(const T& str) const {
+    return FindOneofByName(str.c_str(), str.size());
+  }
+
+  /* Returns a new msgdef that is a copy of the given msgdef (and a copy of all
+   * the fields) but with any references to submessages broken and replaced
+   * with just the name of the submessage.  Returns NULL if memory allocation
+   * failed.
+   *
+   * TODO(haberman): which is more useful, keeping fields resolved or
+   * unresolving them?  If there's no obvious answer, Should this functionality
+   * just be moved into symtab.c? */
+  MessageDef* Dup(const void* owner) const;
+
+  /* Is this message a map entry? */
+  void setmapentry(bool map_entry);
+  bool mapentry() const;
+
+  /* Iteration over fields.  The order is undefined. */
+  class field_iterator
+      : public std::iterator<std::forward_iterator_tag, FieldDef*> {
+   public:
+    explicit field_iterator(MessageDef* md);
+    static field_iterator end(MessageDef* md);
+
+    void operator++();
+    FieldDef* operator*() const;
+    bool operator!=(const field_iterator& other) const;
+    bool operator==(const field_iterator& other) const;
+
+   private:
+    upb_msg_field_iter iter_;
+  };
+
+  class const_field_iterator
+      : public std::iterator<std::forward_iterator_tag, const FieldDef*> {
+   public:
+    explicit const_field_iterator(const MessageDef* md);
+    static const_field_iterator end(const MessageDef* md);
+
+    void operator++();
+    const FieldDef* operator*() const;
+    bool operator!=(const const_field_iterator& other) const;
+    bool operator==(const const_field_iterator& other) const;
+
+   private:
+    upb_msg_field_iter iter_;
+  };
+
+  /* Iteration over oneofs. The order is undefined. */
+  class oneof_iterator
+      : public std::iterator<std::forward_iterator_tag, FieldDef*> {
+   public:
+    explicit oneof_iterator(MessageDef* md);
+    static oneof_iterator end(MessageDef* md);
+
+    void operator++();
+    OneofDef* operator*() const;
+    bool operator!=(const oneof_iterator& other) const;
+    bool operator==(const oneof_iterator& other) const;
+
+   private:
+    upb_msg_oneof_iter iter_;
+  };
+
+  class const_oneof_iterator
+      : public std::iterator<std::forward_iterator_tag, const FieldDef*> {
+   public:
+    explicit const_oneof_iterator(const MessageDef* md);
+    static const_oneof_iterator end(const MessageDef* md);
+
+    void operator++();
+    const OneofDef* operator*() const;
+    bool operator!=(const const_oneof_iterator& other) const;
+    bool operator==(const const_oneof_iterator& other) const;
+
+   private:
+    upb_msg_oneof_iter iter_;
+  };
+
+  class FieldAccessor {
+   public:
+    explicit FieldAccessor(MessageDef* msg) : msg_(msg) {}
+    field_iterator begin() { return msg_->field_begin(); }
+    field_iterator end() { return msg_->field_end(); }
+   private:
+    MessageDef* msg_;
+  };
+
+  class ConstFieldAccessor {
+   public:
+    explicit ConstFieldAccessor(const MessageDef* msg) : msg_(msg) {}
+    const_field_iterator begin() { return msg_->field_begin(); }
+    const_field_iterator end() { return msg_->field_end(); }
+   private:
+    const MessageDef* msg_;
+  };
+
+  class OneofAccessor {
+   public:
+    explicit OneofAccessor(MessageDef* msg) : msg_(msg) {}
+    oneof_iterator begin() { return msg_->oneof_begin(); }
+    oneof_iterator end() { return msg_->oneof_end(); }
+   private:
+    MessageDef* msg_;
+  };
+
+  class ConstOneofAccessor {
+   public:
+    explicit ConstOneofAccessor(const MessageDef* msg) : msg_(msg) {}
+    const_oneof_iterator begin() { return msg_->oneof_begin(); }
+    const_oneof_iterator end() { return msg_->oneof_end(); }
+   private:
+    const MessageDef* msg_;
+  };
+
+  field_iterator field_begin();
+  field_iterator field_end();
+  const_field_iterator field_begin() const;
+  const_field_iterator field_end() const;
+
+  oneof_iterator oneof_begin();
+  oneof_iterator oneof_end();
+  const_oneof_iterator oneof_begin() const;
+  const_oneof_iterator oneof_end() const;
+
+  FieldAccessor fields() { return FieldAccessor(this); }
+  ConstFieldAccessor fields() const { return ConstFieldAccessor(this); }
+  OneofAccessor oneofs() { return OneofAccessor(this); }
+  ConstOneofAccessor oneofs() const { return ConstOneofAccessor(this); }
+
+ private:
+  UPB_DISALLOW_POD_OPS(MessageDef, upb::MessageDef)
+};
+
+#endif  /* __cplusplus */
+
+UPB_BEGIN_EXTERN_C
+
+/* Returns NULL if memory allocation failed. */
+upb_msgdef *upb_msgdef_new(const void *owner);
+
+/* Include upb_refcounted methods like upb_msgdef_ref(). */
+UPB_REFCOUNTED_CMETHODS(upb_msgdef, upb_msgdef_upcast2)
+
+bool upb_msgdef_freeze(upb_msgdef *m, upb_status *status);
+
+const char *upb_msgdef_fullname(const upb_msgdef *m);
+bool upb_msgdef_setfullname(upb_msgdef *m, const char *fullname, upb_status *s);
+
+upb_msgdef *upb_msgdef_dup(const upb_msgdef *m, const void *owner);
+bool upb_msgdef_addfield(upb_msgdef *m, upb_fielddef *f, const void *ref_donor,
+                         upb_status *s);
+bool upb_msgdef_addoneof(upb_msgdef *m, upb_oneofdef *o, const void *ref_donor,
+                         upb_status *s);
+
+/* Field lookup in a couple of different variations:
+ *   - itof = int to field
+ *   - ntof = name to field
+ *   - ntofz = name to field, null-terminated string. */
+const upb_fielddef *upb_msgdef_itof(const upb_msgdef *m, uint32_t i);
+const upb_fielddef *upb_msgdef_ntof(const upb_msgdef *m, const char *name,
+                                    size_t len);
+int upb_msgdef_numfields(const upb_msgdef *m);
+
+UPB_INLINE const upb_fielddef *upb_msgdef_ntofz(const upb_msgdef *m,
+                                                const char *name) {
+  return upb_msgdef_ntof(m, name, strlen(name));
+}
+
+UPB_INLINE upb_fielddef *upb_msgdef_itof_mutable(upb_msgdef *m, uint32_t i) {
+  return (upb_fielddef*)upb_msgdef_itof(m, i);
+}
+
+UPB_INLINE upb_fielddef *upb_msgdef_ntof_mutable(upb_msgdef *m,
+                                                 const char *name, size_t len) {
+  return (upb_fielddef *)upb_msgdef_ntof(m, name, len);
+}
+
+/* Oneof lookup:
+ *   - ntoo = name to oneof
+ *   - ntooz = name to oneof, null-terminated string. */
+const upb_oneofdef *upb_msgdef_ntoo(const upb_msgdef *m, const char *name,
+                                    size_t len);
+int upb_msgdef_numoneofs(const upb_msgdef *m);
+
+UPB_INLINE const upb_oneofdef *upb_msgdef_ntooz(const upb_msgdef *m,
+                                               const char *name) {
+  return upb_msgdef_ntoo(m, name, strlen(name));
+}
+
+UPB_INLINE upb_oneofdef *upb_msgdef_ntoo_mutable(upb_msgdef *m,
+                                                 const char *name, size_t len) {
+  return (upb_oneofdef *)upb_msgdef_ntoo(m, name, len);
+}
+
+void upb_msgdef_setmapentry(upb_msgdef *m, bool map_entry);
+bool upb_msgdef_mapentry(const upb_msgdef *m);
+
+/* Well-known field tag numbers for map-entry messages. */
+#define UPB_MAPENTRY_KEY   1
+#define UPB_MAPENTRY_VALUE 2
+
+const upb_oneofdef *upb_msgdef_findoneof(const upb_msgdef *m,
+                                          const char *name);
+int upb_msgdef_numoneofs(const upb_msgdef *m);
+
+/* upb_msg_field_iter i;
+ * for(upb_msg_field_begin(&i, m);
+ *     !upb_msg_field_done(&i);
+ *     upb_msg_field_next(&i)) {
+ *   upb_fielddef *f = upb_msg_iter_field(&i);
+ *   // ...
+ * }
+ *
+ * For C we don't have separate iterators for const and non-const.
+ * It is the caller's responsibility to cast the upb_fielddef* to
+ * const if the upb_msgdef* is const. */
+void upb_msg_field_begin(upb_msg_field_iter *iter, const upb_msgdef *m);
+void upb_msg_field_next(upb_msg_field_iter *iter);
+bool upb_msg_field_done(const upb_msg_field_iter *iter);
+upb_fielddef *upb_msg_iter_field(const upb_msg_field_iter *iter);
+void upb_msg_field_iter_setdone(upb_msg_field_iter *iter);
+
+/* Similar to above, we also support iterating through the oneofs in a
+ * msgdef. */
+void upb_msg_oneof_begin(upb_msg_oneof_iter *iter, const upb_msgdef *m);
+void upb_msg_oneof_next(upb_msg_oneof_iter *iter);
+bool upb_msg_oneof_done(const upb_msg_oneof_iter *iter);
+upb_oneofdef *upb_msg_iter_oneof(const upb_msg_oneof_iter *iter);
+void upb_msg_oneof_iter_setdone(upb_msg_oneof_iter *iter);
+
+UPB_END_EXTERN_C
+
+
+/* upb::EnumDef ***************************************************************/
+
+typedef upb_strtable_iter upb_enum_iter;
+
+#ifdef __cplusplus
+
+/* Class that represents an enum.  Its base class is upb::Def (convert with
+ * upb::upcast()). */
+class upb::EnumDef {
+ public:
+  /* Returns NULL if memory allocation failed. */
+  static reffed_ptr<EnumDef> New();
+
+  /* upb::RefCounted methods like Ref()/Unref(). */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  /* Functionality from upb::Def. */
+  const char* full_name() const;
+  bool set_full_name(const char* fullname, Status* s);
+  bool set_full_name(const std::string& fullname, Status* s);
+
+  /* Call to freeze this EnumDef. */
+  bool Freeze(Status* s);
+
+  /* The value that is used as the default when no field default is specified.
+   * If not set explicitly, the first value that was added will be used.
+   * The default value must be a member of the enum.
+   * Requires that value_count() > 0. */
+  int32_t default_value() const;
+
+  /* Sets the default value.  If this value is not valid, returns false and an
+   * error message in status. */
+  bool set_default_value(int32_t val, Status* status);
+
+  /* Returns the number of values currently defined in the enum.  Note that
+   * multiple names can refer to the same number, so this may be greater than
+   * the total number of unique numbers. */
+  int value_count() const;
+
+  /* Adds a single name/number pair to the enum.  Fails if this name has
+   * already been used by another value. */
+  bool AddValue(const char* name, int32_t num, Status* status);
+  bool AddValue(const std::string& name, int32_t num, Status* status);
+
+  /* Lookups from name to integer, returning true if found. */
+  bool FindValueByName(const char* name, int32_t* num) const;
+
+  /* Finds the name corresponding to the given number, or NULL if none was
+   * found.  If more than one name corresponds to this number, returns the
+   * first one that was added. */
+  const char* FindValueByNumber(int32_t num) const;
+
+  /* Returns a new EnumDef with all the same values.  The new EnumDef will be
+   * owned by the given owner. */
+  EnumDef* Dup(const void* owner) const;
+
+  /* Iteration over name/value pairs.  The order is undefined.
+   * Adding an enum val invalidates any iterators.
+   *
+   * TODO: make compatible with range-for, with elements as pairs? */
+  class Iterator {
+   public:
+    explicit Iterator(const EnumDef*);
+
+    int32_t number();
+    const char *name();
+    bool Done();
+    void Next();
+
+   private:
+    upb_enum_iter iter_;
+  };
+
+ private:
+  UPB_DISALLOW_POD_OPS(EnumDef, upb::EnumDef)
+};
+
+#endif  /* __cplusplus */
+
+UPB_BEGIN_EXTERN_C
+
+/* Native C API. */
+upb_enumdef *upb_enumdef_new(const void *owner);
+upb_enumdef *upb_enumdef_dup(const upb_enumdef *e, const void *owner);
+
+/* Include upb_refcounted methods like upb_enumdef_ref(). */
+UPB_REFCOUNTED_CMETHODS(upb_enumdef, upb_enumdef_upcast2)
+
+bool upb_enumdef_freeze(upb_enumdef *e, upb_status *status);
+
+/* From upb_def. */
+const char *upb_enumdef_fullname(const upb_enumdef *e);
+bool upb_enumdef_setfullname(upb_enumdef *e, const char *fullname,
+                             upb_status *s);
+
+int32_t upb_enumdef_default(const upb_enumdef *e);
+bool upb_enumdef_setdefault(upb_enumdef *e, int32_t val, upb_status *s);
+int upb_enumdef_numvals(const upb_enumdef *e);
+bool upb_enumdef_addval(upb_enumdef *e, const char *name, int32_t num,
+                        upb_status *status);
+
+/* Enum lookups:
+ * - ntoi:  look up a name with specified length.
+ * - ntoiz: look up a name provided as a null-terminated string.
+ * - iton:  look up an integer, returning the name as a null-terminated
+ *          string. */
+bool upb_enumdef_ntoi(const upb_enumdef *e, const char *name, size_t len,
+                      int32_t *num);
+UPB_INLINE bool upb_enumdef_ntoiz(const upb_enumdef *e,
+                                  const char *name, int32_t *num) {
+  return upb_enumdef_ntoi(e, name, strlen(name), num);
+}
+const char *upb_enumdef_iton(const upb_enumdef *e, int32_t num);
+
+/*  upb_enum_iter i;
+ *  for(upb_enum_begin(&i, e); !upb_enum_done(&i); upb_enum_next(&i)) {
+ *    // ...
+ *  }
+ */
+void upb_enum_begin(upb_enum_iter *iter, const upb_enumdef *e);
+void upb_enum_next(upb_enum_iter *iter);
+bool upb_enum_done(upb_enum_iter *iter);
+const char *upb_enum_iter_name(upb_enum_iter *iter);
+int32_t upb_enum_iter_number(upb_enum_iter *iter);
+
+UPB_END_EXTERN_C
+
+/* upb::OneofDef **************************************************************/
+
+typedef upb_inttable_iter upb_oneof_iter;
+
+#ifdef __cplusplus
+
+/* Class that represents a oneof.  Its base class is upb::Def (convert with
+ * upb::upcast()). */
+class upb::OneofDef {
+ public:
+  /* Returns NULL if memory allocation failed. */
+  static reffed_ptr<OneofDef> New();
+
+  /* upb::RefCounted methods like Ref()/Unref(). */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  /* Functionality from upb::Def. */
+  const char* full_name() const;
+
+  /* Returns the MessageDef that owns this OneofDef. */
+  const MessageDef* containing_type() const;
+
+  /* Returns the name of this oneof. This is the name used to look up the oneof
+   * by name once added to a message def. */
+  const char* name() const;
+  bool set_name(const char* name, Status* s);
+
+  /* Returns the number of fields currently defined in the oneof. */
+  int field_count() const;
+
+  /* Adds a field to the oneof. The field must not have been added to any other
+   * oneof or msgdef. If the oneof is not yet part of a msgdef, then when the
+   * oneof is eventually added to a msgdef, all fields added to the oneof will
+   * also be added to the msgdef at that time. If the oneof is already part of a
+   * msgdef, the field must either be a part of that msgdef already, or must not
+   * be a part of any msgdef; in the latter case, the field is added to the
+   * msgdef as a part of this operation.
+   *
+   * The field may only have an OPTIONAL label, never REQUIRED or REPEATED.
+   *
+   * If |f| is already part of this MessageDef, this method performs no action
+   * and returns true (success). Thus, this method is idempotent. */
+  bool AddField(FieldDef* field, Status* s);
+  bool AddField(const reffed_ptr<FieldDef>& field, Status* s);
+
+  /* Looks up by name. */
+  const FieldDef* FindFieldByName(const char* name, size_t len) const;
+  FieldDef* FindFieldByName(const char* name, size_t len);
+  const FieldDef* FindFieldByName(const char* name) const {
+    return FindFieldByName(name, strlen(name));
+  }
+  FieldDef* FindFieldByName(const char* name) {
+    return FindFieldByName(name, strlen(name));
+  }
+
+  template <class T>
+  FieldDef* FindFieldByName(const T& str) {
+    return FindFieldByName(str.c_str(), str.size());
+  }
+  template <class T>
+  const FieldDef* FindFieldByName(const T& str) const {
+    return FindFieldByName(str.c_str(), str.size());
+  }
+
+  /* Looks up by tag number. */
+  const FieldDef* FindFieldByNumber(uint32_t num) const;
+
+  /* Returns a new OneofDef with all the same fields. The OneofDef will be owned
+   * by the given owner. */
+  OneofDef* Dup(const void* owner) const;
+
+  /* Iteration over fields.  The order is undefined. */
+  class iterator : public std::iterator<std::forward_iterator_tag, FieldDef*> {
+   public:
+    explicit iterator(OneofDef* md);
+    static iterator end(OneofDef* md);
+
+    void operator++();
+    FieldDef* operator*() const;
+    bool operator!=(const iterator& other) const;
+    bool operator==(const iterator& other) const;
+
+   private:
+    upb_oneof_iter iter_;
+  };
+
+  class const_iterator
+      : public std::iterator<std::forward_iterator_tag, const FieldDef*> {
+   public:
+    explicit const_iterator(const OneofDef* md);
+    static const_iterator end(const OneofDef* md);
+
+    void operator++();
+    const FieldDef* operator*() const;
+    bool operator!=(const const_iterator& other) const;
+    bool operator==(const const_iterator& other) const;
+
+   private:
+    upb_oneof_iter iter_;
+  };
+
+  iterator begin();
+  iterator end();
+  const_iterator begin() const;
+  const_iterator end() const;
+
+ private:
+  UPB_DISALLOW_POD_OPS(OneofDef, upb::OneofDef)
+};
+
+#endif  /* __cplusplus */
+
+UPB_BEGIN_EXTERN_C
+
+/* Native C API. */
+upb_oneofdef *upb_oneofdef_new(const void *owner);
+upb_oneofdef *upb_oneofdef_dup(const upb_oneofdef *o, const void *owner);
+
+/* Include upb_refcounted methods like upb_oneofdef_ref(). */
+UPB_REFCOUNTED_CMETHODS(upb_oneofdef, upb_oneofdef_upcast2)
+
+const char *upb_oneofdef_name(const upb_oneofdef *o);
+bool upb_oneofdef_setname(upb_oneofdef *o, const char *name, upb_status *s);
+
+const upb_msgdef *upb_oneofdef_containingtype(const upb_oneofdef *o);
+int upb_oneofdef_numfields(const upb_oneofdef *o);
+bool upb_oneofdef_addfield(upb_oneofdef *o, upb_fielddef *f,
+                           const void *ref_donor,
+                           upb_status *s);
+
+/* Oneof lookups:
+ * - ntof:  look up a field by name.
+ * - ntofz: look up a field by name (as a null-terminated string).
+ * - itof:  look up a field by number. */
+const upb_fielddef *upb_oneofdef_ntof(const upb_oneofdef *o,
+                                      const char *name, size_t length);
+UPB_INLINE const upb_fielddef *upb_oneofdef_ntofz(const upb_oneofdef *o,
+                                                  const char *name) {
+  return upb_oneofdef_ntof(o, name, strlen(name));
+}
+const upb_fielddef *upb_oneofdef_itof(const upb_oneofdef *o, uint32_t num);
+
+/*  upb_oneof_iter i;
+ *  for(upb_oneof_begin(&i, e); !upb_oneof_done(&i); upb_oneof_next(&i)) {
+ *    // ...
+ *  }
+ */
+void upb_oneof_begin(upb_oneof_iter *iter, const upb_oneofdef *o);
+void upb_oneof_next(upb_oneof_iter *iter);
+bool upb_oneof_done(upb_oneof_iter *iter);
+upb_fielddef *upb_oneof_iter_field(const upb_oneof_iter *iter);
+void upb_oneof_iter_setdone(upb_oneof_iter *iter);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+UPB_INLINE const char* upb_safecstr(const std::string& str) {
+  assert(str.size() == std::strlen(str.c_str()));
+  return str.c_str();
+}
+
+/* Inline C++ wrappers. */
+namespace upb {
+
+inline Def* Def::Dup(const void* owner) const {
+  return upb_def_dup(this, owner);
+}
+inline Def::Type Def::def_type() const { return upb_def_type(this); }
+inline const char* Def::full_name() const { return upb_def_fullname(this); }
+inline bool Def::set_full_name(const char* fullname, Status* s) {
+  return upb_def_setfullname(this, fullname, s);
+}
+inline bool Def::set_full_name(const std::string& fullname, Status* s) {
+  return upb_def_setfullname(this, upb_safecstr(fullname), s);
+}
+inline bool Def::Freeze(Def* const* defs, int n, Status* status) {
+  return upb_def_freeze(defs, n, status);
+}
+inline bool Def::Freeze(const std::vector<Def*>& defs, Status* status) {
+  return upb_def_freeze((Def* const*)&defs[0], defs.size(), status);
+}
+
+inline bool FieldDef::CheckType(int32_t val) {
+  return upb_fielddef_checktype(val);
+}
+inline bool FieldDef::CheckLabel(int32_t val) {
+  return upb_fielddef_checklabel(val);
+}
+inline bool FieldDef::CheckDescriptorType(int32_t val) {
+  return upb_fielddef_checkdescriptortype(val);
+}
+inline bool FieldDef::CheckIntegerFormat(int32_t val) {
+  return upb_fielddef_checkintfmt(val);
+}
+inline FieldDef::Type FieldDef::ConvertType(int32_t val) {
+  assert(CheckType(val));
+  return static_cast<FieldDef::Type>(val);
+}
+inline FieldDef::Label FieldDef::ConvertLabel(int32_t val) {
+  assert(CheckLabel(val));
+  return static_cast<FieldDef::Label>(val);
+}
+inline FieldDef::DescriptorType FieldDef::ConvertDescriptorType(int32_t val) {
+  assert(CheckDescriptorType(val));
+  return static_cast<FieldDef::DescriptorType>(val);
+}
+inline FieldDef::IntegerFormat FieldDef::ConvertIntegerFormat(int32_t val) {
+  assert(CheckIntegerFormat(val));
+  return static_cast<FieldDef::IntegerFormat>(val);
+}
+
+inline reffed_ptr<FieldDef> FieldDef::New() {
+  upb_fielddef *f = upb_fielddef_new(&f);
+  return reffed_ptr<FieldDef>(f, &f);
+}
+inline FieldDef* FieldDef::Dup(const void* owner) const {
+  return upb_fielddef_dup(this, owner);
+}
+inline const char* FieldDef::full_name() const {
+  return upb_fielddef_fullname(this);
+}
+inline bool FieldDef::set_full_name(const char* fullname, Status* s) {
+  return upb_fielddef_setfullname(this, fullname, s);
+}
+inline bool FieldDef::set_full_name(const std::string& fullname, Status* s) {
+  return upb_fielddef_setfullname(this, upb_safecstr(fullname), s);
+}
+inline bool FieldDef::type_is_set() const {
+  return upb_fielddef_typeisset(this);
+}
+inline FieldDef::Type FieldDef::type() const { return upb_fielddef_type(this); }
+inline FieldDef::DescriptorType FieldDef::descriptor_type() const {
+  return upb_fielddef_descriptortype(this);
+}
+inline FieldDef::Label FieldDef::label() const {
+  return upb_fielddef_label(this);
+}
+inline uint32_t FieldDef::number() const { return upb_fielddef_number(this); }
+inline const char* FieldDef::name() const { return upb_fielddef_name(this); }
+inline bool FieldDef::is_extension() const {
+  return upb_fielddef_isextension(this);
+}
+inline bool FieldDef::lazy() const {
+  return upb_fielddef_lazy(this);
+}
+inline void FieldDef::set_lazy(bool lazy) {
+  upb_fielddef_setlazy(this, lazy);
+}
+inline bool FieldDef::packed() const {
+  return upb_fielddef_packed(this);
+}
+inline void FieldDef::set_packed(bool packed) {
+  upb_fielddef_setpacked(this, packed);
+}
+inline const MessageDef* FieldDef::containing_type() const {
+  return upb_fielddef_containingtype(this);
+}
+inline const OneofDef* FieldDef::containing_oneof() const {
+  return upb_fielddef_containingoneof(this);
+}
+inline const char* FieldDef::containing_type_name() {
+  return upb_fielddef_containingtypename(this);
+}
+inline bool FieldDef::set_number(uint32_t number, Status* s) {
+  return upb_fielddef_setnumber(this, number, s);
+}
+inline bool FieldDef::set_name(const char *name, Status* s) {
+  return upb_fielddef_setname(this, name, s);
+}
+inline bool FieldDef::set_name(const std::string& name, Status* s) {
+  return upb_fielddef_setname(this, upb_safecstr(name), s);
+}
+inline bool FieldDef::set_containing_type_name(const char *name, Status* s) {
+  return upb_fielddef_setcontainingtypename(this, name, s);
+}
+inline bool FieldDef::set_containing_type_name(const std::string &name,
+                                               Status *s) {
+  return upb_fielddef_setcontainingtypename(this, upb_safecstr(name), s);
+}
+inline void FieldDef::set_type(upb_fieldtype_t type) {
+  upb_fielddef_settype(this, type);
+}
+inline void FieldDef::set_is_extension(bool is_extension) {
+  upb_fielddef_setisextension(this, is_extension);
+}
+inline void FieldDef::set_descriptor_type(FieldDef::DescriptorType type) {
+  upb_fielddef_setdescriptortype(this, type);
+}
+inline void FieldDef::set_label(upb_label_t label) {
+  upb_fielddef_setlabel(this, label);
+}
+inline bool FieldDef::IsSubMessage() const {
+  return upb_fielddef_issubmsg(this);
+}
+inline bool FieldDef::IsString() const { return upb_fielddef_isstring(this); }
+inline bool FieldDef::IsSequence() const { return upb_fielddef_isseq(this); }
+inline bool FieldDef::IsMap() const { return upb_fielddef_ismap(this); }
+inline int64_t FieldDef::default_int64() const {
+  return upb_fielddef_defaultint64(this);
+}
+inline int32_t FieldDef::default_int32() const {
+  return upb_fielddef_defaultint32(this);
+}
+inline uint64_t FieldDef::default_uint64() const {
+  return upb_fielddef_defaultuint64(this);
+}
+inline uint32_t FieldDef::default_uint32() const {
+  return upb_fielddef_defaultuint32(this);
+}
+inline bool FieldDef::default_bool() const {
+  return upb_fielddef_defaultbool(this);
+}
+inline float FieldDef::default_float() const {
+  return upb_fielddef_defaultfloat(this);
+}
+inline double FieldDef::default_double() const {
+  return upb_fielddef_defaultdouble(this);
+}
+inline const char* FieldDef::default_string(size_t* len) const {
+  return upb_fielddef_defaultstr(this, len);
+}
+inline void FieldDef::set_default_int64(int64_t value) {
+  upb_fielddef_setdefaultint64(this, value);
+}
+inline void FieldDef::set_default_int32(int32_t value) {
+  upb_fielddef_setdefaultint32(this, value);
+}
+inline void FieldDef::set_default_uint64(uint64_t value) {
+  upb_fielddef_setdefaultuint64(this, value);
+}
+inline void FieldDef::set_default_uint32(uint32_t value) {
+  upb_fielddef_setdefaultuint32(this, value);
+}
+inline void FieldDef::set_default_bool(bool value) {
+  upb_fielddef_setdefaultbool(this, value);
+}
+inline void FieldDef::set_default_float(float value) {
+  upb_fielddef_setdefaultfloat(this, value);
+}
+inline void FieldDef::set_default_double(double value) {
+  upb_fielddef_setdefaultdouble(this, value);
+}
+inline bool FieldDef::set_default_string(const void *str, size_t len,
+                                         Status *s) {
+  return upb_fielddef_setdefaultstr(this, str, len, s);
+}
+inline bool FieldDef::set_default_string(const std::string& str, Status* s) {
+  return upb_fielddef_setdefaultstr(this, str.c_str(), str.size(), s);
+}
+inline void FieldDef::set_default_cstr(const char* str, Status* s) {
+  return upb_fielddef_setdefaultcstr(this, str, s);
+}
+inline bool FieldDef::HasSubDef() const { return upb_fielddef_hassubdef(this); }
+inline const Def* FieldDef::subdef() const { return upb_fielddef_subdef(this); }
+inline const MessageDef *FieldDef::message_subdef() const {
+  return upb_fielddef_msgsubdef(this);
+}
+inline const EnumDef *FieldDef::enum_subdef() const {
+  return upb_fielddef_enumsubdef(this);
+}
+inline const char* FieldDef::subdef_name() const {
+  return upb_fielddef_subdefname(this);
+}
+inline bool FieldDef::set_subdef(const Def* subdef, Status* s) {
+  return upb_fielddef_setsubdef(this, subdef, s);
+}
+inline bool FieldDef::set_enum_subdef(const EnumDef* subdef, Status* s) {
+  return upb_fielddef_setenumsubdef(this, subdef, s);
+}
+inline bool FieldDef::set_message_subdef(const MessageDef* subdef, Status* s) {
+  return upb_fielddef_setmsgsubdef(this, subdef, s);
+}
+inline bool FieldDef::set_subdef_name(const char* name, Status* s) {
+  return upb_fielddef_setsubdefname(this, name, s);
+}
+inline bool FieldDef::set_subdef_name(const std::string& name, Status* s) {
+  return upb_fielddef_setsubdefname(this, upb_safecstr(name), s);
+}
+
+inline reffed_ptr<MessageDef> MessageDef::New() {
+  upb_msgdef *m = upb_msgdef_new(&m);
+  return reffed_ptr<MessageDef>(m, &m);
+}
+inline const char *MessageDef::full_name() const {
+  return upb_msgdef_fullname(this);
+}
+inline bool MessageDef::set_full_name(const char* fullname, Status* s) {
+  return upb_msgdef_setfullname(this, fullname, s);
+}
+inline bool MessageDef::set_full_name(const std::string& fullname, Status* s) {
+  return upb_msgdef_setfullname(this, upb_safecstr(fullname), s);
+}
+inline bool MessageDef::Freeze(Status* status) {
+  return upb_msgdef_freeze(this, status);
+}
+inline int MessageDef::field_count() const {
+  return upb_msgdef_numfields(this);
+}
+inline int MessageDef::oneof_count() const {
+  return upb_msgdef_numoneofs(this);
+}
+inline bool MessageDef::AddField(upb_fielddef* f, Status* s) {
+  return upb_msgdef_addfield(this, f, NULL, s);
+}
+inline bool MessageDef::AddField(const reffed_ptr<FieldDef>& f, Status* s) {
+  return upb_msgdef_addfield(this, f.get(), NULL, s);
+}
+inline bool MessageDef::AddOneof(upb_oneofdef* o, Status* s) {
+  return upb_msgdef_addoneof(this, o, NULL, s);
+}
+inline bool MessageDef::AddOneof(const reffed_ptr<OneofDef>& o, Status* s) {
+  return upb_msgdef_addoneof(this, o.get(), NULL, s);
+}
+inline FieldDef* MessageDef::FindFieldByNumber(uint32_t number) {
+  return upb_msgdef_itof_mutable(this, number);
+}
+inline FieldDef* MessageDef::FindFieldByName(const char* name, size_t len) {
+  return upb_msgdef_ntof_mutable(this, name, len);
+}
+inline const FieldDef* MessageDef::FindFieldByNumber(uint32_t number) const {
+  return upb_msgdef_itof(this, number);
+}
+inline const FieldDef *MessageDef::FindFieldByName(const char *name,
+                                                   size_t len) const {
+  return upb_msgdef_ntof(this, name, len);
+}
+inline OneofDef* MessageDef::FindOneofByName(const char* name, size_t len) {
+  return upb_msgdef_ntoo_mutable(this, name, len);
+}
+inline const OneofDef* MessageDef::FindOneofByName(const char* name,
+                                                   size_t len) const {
+  return upb_msgdef_ntoo(this, name, len);
+}
+inline MessageDef* MessageDef::Dup(const void *owner) const {
+  return upb_msgdef_dup(this, owner);
+}
+inline void MessageDef::setmapentry(bool map_entry) {
+  upb_msgdef_setmapentry(this, map_entry);
+}
+inline bool MessageDef::mapentry() const {
+  return upb_msgdef_mapentry(this);
+}
+inline MessageDef::field_iterator MessageDef::field_begin() {
+  return field_iterator(this);
+}
+inline MessageDef::field_iterator MessageDef::field_end() {
+  return field_iterator::end(this);
+}
+inline MessageDef::const_field_iterator MessageDef::field_begin() const {
+  return const_field_iterator(this);
+}
+inline MessageDef::const_field_iterator MessageDef::field_end() const {
+  return const_field_iterator::end(this);
+}
+
+inline MessageDef::oneof_iterator MessageDef::oneof_begin() {
+  return oneof_iterator(this);
+}
+inline MessageDef::oneof_iterator MessageDef::oneof_end() {
+  return oneof_iterator::end(this);
+}
+inline MessageDef::const_oneof_iterator MessageDef::oneof_begin() const {
+  return const_oneof_iterator(this);
+}
+inline MessageDef::const_oneof_iterator MessageDef::oneof_end() const {
+  return const_oneof_iterator::end(this);
+}
+
+inline MessageDef::field_iterator::field_iterator(MessageDef* md) {
+  upb_msg_field_begin(&iter_, md);
+}
+inline MessageDef::field_iterator MessageDef::field_iterator::end(
+    MessageDef* md) {
+  MessageDef::field_iterator iter(md);
+  upb_msg_field_iter_setdone(&iter.iter_);
+  return iter;
+}
+inline FieldDef* MessageDef::field_iterator::operator*() const {
+  return upb_msg_iter_field(&iter_);
+}
+inline void MessageDef::field_iterator::operator++() {
+  return upb_msg_field_next(&iter_);
+}
+inline bool MessageDef::field_iterator::operator==(
+    const field_iterator &other) const {
+  return upb_inttable_iter_isequal(&iter_, &other.iter_);
+}
+inline bool MessageDef::field_iterator::operator!=(
+    const field_iterator &other) const {
+  return !(*this == other);
+}
+
+inline MessageDef::const_field_iterator::const_field_iterator(
+    const MessageDef* md) {
+  upb_msg_field_begin(&iter_, md);
+}
+inline MessageDef::const_field_iterator MessageDef::const_field_iterator::end(
+    const MessageDef *md) {
+  MessageDef::const_field_iterator iter(md);
+  upb_msg_field_iter_setdone(&iter.iter_);
+  return iter;
+}
+inline const FieldDef* MessageDef::const_field_iterator::operator*() const {
+  return upb_msg_iter_field(&iter_);
+}
+inline void MessageDef::const_field_iterator::operator++() {
+  return upb_msg_field_next(&iter_);
+}
+inline bool MessageDef::const_field_iterator::operator==(
+    const const_field_iterator &other) const {
+  return upb_inttable_iter_isequal(&iter_, &other.iter_);
+}
+inline bool MessageDef::const_field_iterator::operator!=(
+    const const_field_iterator &other) const {
+  return !(*this == other);
+}
+
+inline MessageDef::oneof_iterator::oneof_iterator(MessageDef* md) {
+  upb_msg_oneof_begin(&iter_, md);
+}
+inline MessageDef::oneof_iterator MessageDef::oneof_iterator::end(
+    MessageDef* md) {
+  MessageDef::oneof_iterator iter(md);
+  upb_msg_oneof_iter_setdone(&iter.iter_);
+  return iter;
+}
+inline OneofDef* MessageDef::oneof_iterator::operator*() const {
+  return upb_msg_iter_oneof(&iter_);
+}
+inline void MessageDef::oneof_iterator::operator++() {
+  return upb_msg_oneof_next(&iter_);
+}
+inline bool MessageDef::oneof_iterator::operator==(
+    const oneof_iterator &other) const {
+  return upb_strtable_iter_isequal(&iter_, &other.iter_);
+}
+inline bool MessageDef::oneof_iterator::operator!=(
+    const oneof_iterator &other) const {
+  return !(*this == other);
+}
+
+inline MessageDef::const_oneof_iterator::const_oneof_iterator(
+    const MessageDef* md) {
+  upb_msg_oneof_begin(&iter_, md);
+}
+inline MessageDef::const_oneof_iterator MessageDef::const_oneof_iterator::end(
+    const MessageDef *md) {
+  MessageDef::const_oneof_iterator iter(md);
+  upb_msg_oneof_iter_setdone(&iter.iter_);
+  return iter;
+}
+inline const OneofDef* MessageDef::const_oneof_iterator::operator*() const {
+  return upb_msg_iter_oneof(&iter_);
+}
+inline void MessageDef::const_oneof_iterator::operator++() {
+  return upb_msg_oneof_next(&iter_);
+}
+inline bool MessageDef::const_oneof_iterator::operator==(
+    const const_oneof_iterator &other) const {
+  return upb_strtable_iter_isequal(&iter_, &other.iter_);
+}
+inline bool MessageDef::const_oneof_iterator::operator!=(
+    const const_oneof_iterator &other) const {
+  return !(*this == other);
+}
+
+inline reffed_ptr<EnumDef> EnumDef::New() {
+  upb_enumdef *e = upb_enumdef_new(&e);
+  return reffed_ptr<EnumDef>(e, &e);
+}
+inline const char* EnumDef::full_name() const {
+  return upb_enumdef_fullname(this);
+}
+inline bool EnumDef::set_full_name(const char* fullname, Status* s) {
+  return upb_enumdef_setfullname(this, fullname, s);
+}
+inline bool EnumDef::set_full_name(const std::string& fullname, Status* s) {
+  return upb_enumdef_setfullname(this, upb_safecstr(fullname), s);
+}
+inline bool EnumDef::Freeze(Status* status) {
+  return upb_enumdef_freeze(this, status);
+}
+inline int32_t EnumDef::default_value() const {
+  return upb_enumdef_default(this);
+}
+inline bool EnumDef::set_default_value(int32_t val, Status* status) {
+  return upb_enumdef_setdefault(this, val, status);
+}
+inline int EnumDef::value_count() const { return upb_enumdef_numvals(this); }
+inline bool EnumDef::AddValue(const char* name, int32_t num, Status* status) {
+  return upb_enumdef_addval(this, name, num, status);
+}
+inline bool EnumDef::AddValue(const std::string& name, int32_t num,
+                              Status* status) {
+  return upb_enumdef_addval(this, upb_safecstr(name), num, status);
+}
+inline bool EnumDef::FindValueByName(const char* name, int32_t *num) const {
+  return upb_enumdef_ntoiz(this, name, num);
+}
+inline const char* EnumDef::FindValueByNumber(int32_t num) const {
+  return upb_enumdef_iton(this, num);
+}
+inline EnumDef* EnumDef::Dup(const void* owner) const {
+  return upb_enumdef_dup(this, owner);
+}
+
+inline EnumDef::Iterator::Iterator(const EnumDef* e) {
+  upb_enum_begin(&iter_, e);
+}
+inline int32_t EnumDef::Iterator::number() {
+  return upb_enum_iter_number(&iter_);
+}
+inline const char* EnumDef::Iterator::name() {
+  return upb_enum_iter_name(&iter_);
+}
+inline bool EnumDef::Iterator::Done() { return upb_enum_done(&iter_); }
+inline void EnumDef::Iterator::Next() { return upb_enum_next(&iter_); }
+
+inline reffed_ptr<OneofDef> OneofDef::New() {
+  upb_oneofdef *o = upb_oneofdef_new(&o);
+  return reffed_ptr<OneofDef>(o, &o);
+}
+inline const char* OneofDef::full_name() const {
+  return upb_oneofdef_name(this);
+}
+
+inline const MessageDef* OneofDef::containing_type() const {
+  return upb_oneofdef_containingtype(this);
+}
+inline const char* OneofDef::name() const {
+  return upb_oneofdef_name(this);
+}
+inline bool OneofDef::set_name(const char* name, Status* s) {
+  return upb_oneofdef_setname(this, name, s);
+}
+inline int OneofDef::field_count() const {
+  return upb_oneofdef_numfields(this);
+}
+inline bool OneofDef::AddField(FieldDef* field, Status* s) {
+  return upb_oneofdef_addfield(this, field, NULL, s);
+}
+inline bool OneofDef::AddField(const reffed_ptr<FieldDef>& field, Status* s) {
+  return upb_oneofdef_addfield(this, field.get(), NULL, s);
+}
+inline const FieldDef* OneofDef::FindFieldByName(const char* name,
+                                                 size_t len) const {
+  return upb_oneofdef_ntof(this, name, len);
+}
+inline const FieldDef* OneofDef::FindFieldByNumber(uint32_t num) const {
+  return upb_oneofdef_itof(this, num);
+}
+inline OneofDef::iterator OneofDef::begin() { return iterator(this); }
+inline OneofDef::iterator OneofDef::end() { return iterator::end(this); }
+inline OneofDef::const_iterator OneofDef::begin() const {
+  return const_iterator(this);
+}
+inline OneofDef::const_iterator OneofDef::end() const {
+  return const_iterator::end(this);
+}
+
+inline OneofDef::iterator::iterator(OneofDef* o) {
+  upb_oneof_begin(&iter_, o);
+}
+inline OneofDef::iterator OneofDef::iterator::end(OneofDef* o) {
+  OneofDef::iterator iter(o);
+  upb_oneof_iter_setdone(&iter.iter_);
+  return iter;
+}
+inline FieldDef* OneofDef::iterator::operator*() const {
+  return upb_oneof_iter_field(&iter_);
+}
+inline void OneofDef::iterator::operator++() { return upb_oneof_next(&iter_); }
+inline bool OneofDef::iterator::operator==(const iterator &other) const {
+  return upb_inttable_iter_isequal(&iter_, &other.iter_);
+}
+inline bool OneofDef::iterator::operator!=(const iterator &other) const {
+  return !(*this == other);
+}
+
+inline OneofDef::const_iterator::const_iterator(const OneofDef* md) {
+  upb_oneof_begin(&iter_, md);
+}
+inline OneofDef::const_iterator OneofDef::const_iterator::end(
+    const OneofDef *md) {
+  OneofDef::const_iterator iter(md);
+  upb_oneof_iter_setdone(&iter.iter_);
+  return iter;
+}
+inline const FieldDef* OneofDef::const_iterator::operator*() const {
+  return upb_msg_iter_field(&iter_);
+}
+inline void OneofDef::const_iterator::operator++() {
+  return upb_oneof_next(&iter_);
+}
+inline bool OneofDef::const_iterator::operator==(
+    const const_iterator &other) const {
+  return upb_inttable_iter_isequal(&iter_, &other.iter_);
+}
+inline bool OneofDef::const_iterator::operator!=(
+    const const_iterator &other) const {
+  return !(*this == other);
+}
+
+}  /* namespace upb */
+#endif
+
+#endif /* UPB_DEF_H_ */
+/*
+** This file contains definitions of structs that should be considered private
+** and NOT stable across versions of upb.
+**
+** The only reason they are declared here and not in .c files is to allow upb
+** and the application (if desired) to embed statically-initialized instances
+** of structures like defs.
+**
+** If you include this file, all guarantees of ABI compatibility go out the
+** window!  Any code that includes this file needs to recompile against the
+** exact same version of upb that they are linking against.
+**
+** You also need to recompile if you change the value of the UPB_DEBUG_REFS
+** flag.
+*/
+
+
+#ifndef UPB_STATICINIT_H_
+#define UPB_STATICINIT_H_
+
+#ifdef __cplusplus
+/* Because of how we do our typedefs, this header can't be included from C++. */
+#error This file cannot be included from C++
+#endif
+
+/* upb_refcounted *************************************************************/
+
+
+/* upb_def ********************************************************************/
+
+struct upb_def {
+  upb_refcounted base;
+
+  const char *fullname;
+  char type;  /* A upb_deftype_t (char to save space) */
+
+  /* Used as a flag during the def's mutable stage.  Must be false unless
+   * it is currently being used by a function on the stack.  This allows
+   * us to easily determine which defs were passed into the function's
+   * current invocation. */
+  bool came_from_user;
+};
+
+#define UPB_DEF_INIT(name, type, refs, ref2s) \
+    { UPB_REFCOUNT_INIT(refs, ref2s), name, type, false }
+
+
+/* upb_fielddef ***************************************************************/
+
+struct upb_fielddef {
+  upb_def base;
+
+  union {
+    int64_t sint;
+    uint64_t uint;
+    double dbl;
+    float flt;
+    void *bytes;
+  } defaultval;
+  union {
+    const upb_msgdef *def;  /* If !msg_is_symbolic. */
+    char *name;             /* If msg_is_symbolic. */
+  } msg;
+  union {
+    const upb_def *def;  /* If !subdef_is_symbolic. */
+    char *name;          /* If subdef_is_symbolic. */
+  } sub;  /* The msgdef or enumdef for this field, if upb_hassubdef(f). */
+  bool subdef_is_symbolic;
+  bool msg_is_symbolic;
+  const upb_oneofdef *oneof;
+  bool default_is_string;
+  bool type_is_set_;     /* False until type is explicitly set. */
+  bool is_extension_;
+  bool lazy_;
+  bool packed_;
+  upb_intfmt_t intfmt;
+  bool tagdelim;
+  upb_fieldtype_t type_;
+  upb_label_t label_;
+  uint32_t number_;
+  uint32_t selector_base;  /* Used to index into a upb::Handlers table. */
+  uint32_t index_;
+};
+
+#define UPB_FIELDDEF_INIT(label, type, intfmt, tagdelim, is_extension, lazy,   \
+                          packed, name, num, msgdef, subdef, selector_base,    \
+                          index, defaultval, refs, ref2s)                      \
+  {                                                                            \
+    UPB_DEF_INIT(name, UPB_DEF_FIELD, refs, ref2s), defaultval, {msgdef},      \
+        {subdef}, NULL, false, false,                                          \
+        type == UPB_TYPE_STRING || type == UPB_TYPE_BYTES, true, is_extension, \
+        lazy, packed, intfmt, tagdelim, type, label, num, selector_base, index \
+  }
+
+
+/* upb_msgdef *****************************************************************/
+
+struct upb_msgdef {
+  upb_def base;
+
+  size_t selector_count;
+  uint32_t submsg_field_count;
+
+  /* Tables for looking up fields by number and name. */
+  upb_inttable itof;  /* int to field */
+  upb_strtable ntof;  /* name to field */
+
+  /* Tables for looking up oneofs by name. */
+  upb_strtable ntoo;  /* name to oneof */
+
+  /* Is this a map-entry message?
+   * TODO: set this flag properly for static descriptors; regenerate
+   * descriptor.upb.c. */
+  bool map_entry;
+
+  /* TODO(haberman): proper extension ranges (there can be multiple). */
+};
+
+/* TODO: also support static initialization of the oneofs table. This will be
+ * needed if we compile in descriptors that contain oneofs. */
+#define UPB_MSGDEF_INIT(name, selector_count, submsg_field_count, itof, ntof, \
+                        refs, ref2s)                                          \
+  {                                                                           \
+    UPB_DEF_INIT(name, UPB_DEF_MSG, refs, ref2s), selector_count,             \
+        submsg_field_count, itof, ntof,                                       \
+        UPB_EMPTY_STRTABLE_INIT(UPB_CTYPE_PTR), false                         \
+  }
+
+
+/* upb_enumdef ****************************************************************/
+
+struct upb_enumdef {
+  upb_def base;
+
+  upb_strtable ntoi;
+  upb_inttable iton;
+  int32_t defaultval;
+};
+
+#define UPB_ENUMDEF_INIT(name, ntoi, iton, defaultval, refs, ref2s) \
+  { UPB_DEF_INIT(name, UPB_DEF_ENUM, refs, ref2s), ntoi, iton, defaultval }
+
+
+/* upb_oneofdef ***************************************************************/
+
+struct upb_oneofdef {
+  upb_def base;
+
+  upb_strtable ntof;
+  upb_inttable itof;
+  const upb_msgdef *parent;
+};
+
+#define UPB_ONEOFDEF_INIT(name, ntof, itof, refs, ref2s) \
+  { UPB_DEF_INIT(name, UPB_DEF_ENUM, refs, ref2s), ntof, itof }
+
+
+/* upb_symtab *****************************************************************/
+
+struct upb_symtab {
+  upb_refcounted base;
+
+  upb_strtable symtab;
+};
+
+#define UPB_SYMTAB_INIT(symtab, refs, ref2s) \
+  { UPB_REFCOUNT_INIT(refs, ref2s), symtab }
+
+
+#endif  /* UPB_STATICINIT_H_ */
+/*
+** upb::Handlers (upb_handlers)
+**
+** A upb_handlers is like a virtual table for a upb_msgdef.  Each field of the
+** message can have associated functions that will be called when we are
+** parsing or visiting a stream of data.  This is similar to how handlers work
+** in SAX (the Simple API for XML).
+**
+** The handlers have no idea where the data is coming from, so a single set of
+** handlers could be used with two completely different data sources (for
+** example, a parser and a visitor over in-memory objects).  This decoupling is
+** the most important feature of upb, because it allows parsers and serializers
+** to be highly reusable.
+**
+** This is a mixed C/C++ interface that offers a full API to both languages.
+** See the top-level README for more information.
+*/
+
+#ifndef UPB_HANDLERS_H
+#define UPB_HANDLERS_H
+
+
+#ifdef __cplusplus
+namespace upb {
+class BufferHandle;
+class BytesHandler;
+class HandlerAttributes;
+class Handlers;
+template <class T> class Handler;
+template <class T> struct CanonicalType;
+}  /* namespace upb */
+#endif
+
+UPB_DECLARE_TYPE(upb::BufferHandle, upb_bufhandle)
+UPB_DECLARE_TYPE(upb::BytesHandler, upb_byteshandler)
+UPB_DECLARE_TYPE(upb::HandlerAttributes, upb_handlerattr)
+UPB_DECLARE_DERIVED_TYPE(upb::Handlers, upb::RefCounted,
+                         upb_handlers, upb_refcounted)
+
+/* The maximum depth that the handler graph can have.  This is a resource limit
+ * for the C stack since we sometimes need to recursively traverse the graph.
+ * Cycles are ok; the traversal will stop when it detects a cycle, but we must
+ * hit the cycle before the maximum depth is reached.
+ *
+ * If having a single static limit is too inflexible, we can add another variant
+ * of Handlers::Freeze that allows specifying this as a parameter. */
+#define UPB_MAX_HANDLER_DEPTH 64
+
+/* All the different types of handlers that can be registered.
+ * Only needed for the advanced functions in upb::Handlers. */
+typedef enum {
+  UPB_HANDLER_INT32,
+  UPB_HANDLER_INT64,
+  UPB_HANDLER_UINT32,
+  UPB_HANDLER_UINT64,
+  UPB_HANDLER_FLOAT,
+  UPB_HANDLER_DOUBLE,
+  UPB_HANDLER_BOOL,
+  UPB_HANDLER_STARTSTR,
+  UPB_HANDLER_STRING,
+  UPB_HANDLER_ENDSTR,
+  UPB_HANDLER_STARTSUBMSG,
+  UPB_HANDLER_ENDSUBMSG,
+  UPB_HANDLER_STARTSEQ,
+  UPB_HANDLER_ENDSEQ
+} upb_handlertype_t;
+
+#define UPB_HANDLER_MAX (UPB_HANDLER_ENDSEQ+1)
+
+#define UPB_BREAK NULL
+
+/* A convenient definition for when no closure is needed. */
+extern char _upb_noclosure;
+#define UPB_NO_CLOSURE &_upb_noclosure
+
+/* A selector refers to a specific field handler in the Handlers object
+ * (for example: the STARTSUBMSG handler for field "field15"). */
+typedef int32_t upb_selector_t;
+
+UPB_BEGIN_EXTERN_C
+
+/* Forward-declares for C inline accessors.  We need to declare these here
+ * so we can "friend" them in the class declarations in C++. */
+UPB_INLINE upb_func *upb_handlers_gethandler(const upb_handlers *h,
+                                             upb_selector_t s);
+UPB_INLINE const void *upb_handlerattr_handlerdata(const upb_handlerattr *attr);
+UPB_INLINE const void *upb_handlers_gethandlerdata(const upb_handlers *h,
+                                                   upb_selector_t s);
+
+UPB_INLINE void upb_bufhandle_init(upb_bufhandle *h);
+UPB_INLINE void upb_bufhandle_setobj(upb_bufhandle *h, const void *obj,
+                                     const void *type);
+UPB_INLINE void upb_bufhandle_setbuf(upb_bufhandle *h, const char *buf,
+                                     size_t ofs);
+UPB_INLINE const void *upb_bufhandle_obj(const upb_bufhandle *h);
+UPB_INLINE const void *upb_bufhandle_objtype(const upb_bufhandle *h);
+UPB_INLINE const char *upb_bufhandle_buf(const upb_bufhandle *h);
+
+UPB_END_EXTERN_C
+
+
+/* Static selectors for upb::Handlers. */
+#define UPB_STARTMSG_SELECTOR 0
+#define UPB_ENDMSG_SELECTOR 1
+#define UPB_STATIC_SELECTOR_COUNT 2
+
+/* Static selectors for upb::BytesHandler. */
+#define UPB_STARTSTR_SELECTOR 0
+#define UPB_STRING_SELECTOR 1
+#define UPB_ENDSTR_SELECTOR 2
+
+typedef void upb_handlerfree(void *d);
+
+#ifdef __cplusplus
+
+/* A set of attributes that accompanies a handler's function pointer. */
+class upb::HandlerAttributes {
+ public:
+  HandlerAttributes();
+  ~HandlerAttributes();
+
+  /* Sets the handler data that will be passed as the second parameter of the
+   * handler.  To free this pointer when the handlers are freed, call
+   * Handlers::AddCleanup(). */
+  bool SetHandlerData(const void *handler_data);
+  const void* handler_data() const;
+
+  /* Use this to specify the type of the closure.  This will be checked against
+   * all other closure types for handler that use the same closure.
+   * Registration will fail if this does not match all other non-NULL closure
+   * types. */
+  bool SetClosureType(const void *closure_type);
+  const void* closure_type() const;
+
+  /* Use this to specify the type of the returned closure.  Only used for
+   * Start*{String,SubMessage,Sequence} handlers.  This must match the closure
+   * type of any handlers that use it (for example, the StringBuf handler must
+   * match the closure returned from StartString). */
+  bool SetReturnClosureType(const void *return_closure_type);
+  const void* return_closure_type() const;
+
+  /* Set to indicate that the handler always returns "ok" (either "true" or a
+   * non-NULL closure).  This is a hint that can allow code generators to
+   * generate more efficient code. */
+  bool SetAlwaysOk(bool always_ok);
+  bool always_ok() const;
+
+ private:
+  friend UPB_INLINE const void * ::upb_handlerattr_handlerdata(
+      const upb_handlerattr *attr);
+#else
+struct upb_handlerattr {
+#endif
+  const void *handler_data_;
+  const void *closure_type_;
+  const void *return_closure_type_;
+  bool alwaysok_;
+};
+
+#define UPB_HANDLERATTR_INITIALIZER {NULL, NULL, NULL, false}
+
+typedef struct {
+  upb_func *func;
+
+  /* It is wasteful to include the entire attributes here:
+   *
+   * * Some of the information is redundant (like storing the closure type
+   *   separately for each handler that must match).
+   * * Some of the info is only needed prior to freeze() (like closure types).
+   * * alignment padding wastes a lot of space for alwaysok_.
+   *
+   * If/when the size and locality of handlers is an issue, we can optimize this
+   * not to store the entire attr like this.  We do not expose the table's
+   * layout to allow this optimization in the future. */
+  upb_handlerattr attr;
+} upb_handlers_tabent;
+
+#ifdef __cplusplus
+
+/* Extra information about a buffer that is passed to a StringBuf handler.
+ * TODO(haberman): allow the handle to be pinned so that it will outlive
+ * the handler invocation. */
+class upb::BufferHandle {
+ public:
+  BufferHandle();
+  ~BufferHandle();
+
+  /* The beginning of the buffer.  This may be different than the pointer
+   * passed to a StringBuf handler because the handler may receive data
+   * that is from the middle or end of a larger buffer. */
+  const char* buffer() const;
+
+  /* The offset within the attached object where this buffer begins.  Only
+   * meaningful if there is an attached object. */
+  size_t object_offset() const;
+
+  /* Note that object_offset is the offset of "buf" within the attached
+   * object. */
+  void SetBuffer(const char* buf, size_t object_offset);
+
+  /* The BufferHandle can have an "attached object", which can be used to
+   * tunnel through a pointer to the buffer's underlying representation. */
+  template <class T>
+  void SetAttachedObject(const T* obj);
+
+  /* Returns NULL if the attached object is not of this type. */
+  template <class T>
+  const T* GetAttachedObject() const;
+
+ private:
+  friend UPB_INLINE void ::upb_bufhandle_init(upb_bufhandle *h);
+  friend UPB_INLINE void ::upb_bufhandle_setobj(upb_bufhandle *h,
+                                                const void *obj,
+                                                const void *type);
+  friend UPB_INLINE void ::upb_bufhandle_setbuf(upb_bufhandle *h,
+                                                const char *buf, size_t ofs);
+  friend UPB_INLINE const void* ::upb_bufhandle_obj(const upb_bufhandle *h);
+  friend UPB_INLINE const void* ::upb_bufhandle_objtype(
+      const upb_bufhandle *h);
+  friend UPB_INLINE const char* ::upb_bufhandle_buf(const upb_bufhandle *h);
+#else
+struct upb_bufhandle {
+#endif
+  const char *buf_;
+  const void *obj_;
+  const void *objtype_;
+  size_t objofs_;
+};
+
+#ifdef __cplusplus
+
+/* A upb::Handlers object represents the set of handlers associated with a
+ * message in the graph of messages.  You can think of it as a big virtual
+ * table with functions corresponding to all the events that can fire while
+ * parsing or visiting a message of a specific type.
+ *
+ * Any handlers that are not set behave as if they had successfully consumed
+ * the value.  Any unset Start* handlers will propagate their closure to the
+ * inner frame.
+ *
+ * The easiest way to create the *Handler objects needed by the Set* methods is
+ * with the UpbBind() and UpbMakeHandler() macros; see below. */
+class upb::Handlers {
+ public:
+  typedef upb_selector_t Selector;
+  typedef upb_handlertype_t Type;
+
+  typedef Handler<void *(*)(void *, const void *)> StartFieldHandler;
+  typedef Handler<bool (*)(void *, const void *)> EndFieldHandler;
+  typedef Handler<bool (*)(void *, const void *)> StartMessageHandler;
+  typedef Handler<bool (*)(void *, const void *, Status*)> EndMessageHandler;
+  typedef Handler<void *(*)(void *, const void *, size_t)> StartStringHandler;
+  typedef Handler<size_t (*)(void *, const void *, const char *, size_t,
+                             const BufferHandle *)> StringHandler;
+
+  template <class T> struct ValueHandler {
+    typedef Handler<bool(*)(void *, const void *, T)> H;
+  };
+
+  typedef ValueHandler<int32_t>::H     Int32Handler;
+  typedef ValueHandler<int64_t>::H     Int64Handler;
+  typedef ValueHandler<uint32_t>::H    UInt32Handler;
+  typedef ValueHandler<uint64_t>::H    UInt64Handler;
+  typedef ValueHandler<float>::H       FloatHandler;
+  typedef ValueHandler<double>::H      DoubleHandler;
+  typedef ValueHandler<bool>::H        BoolHandler;
+
+  /* Any function pointer can be converted to this and converted back to its
+   * correct type. */
+  typedef void GenericFunction();
+
+  typedef void HandlersCallback(const void *closure, upb_handlers *h);
+
+  /* Returns a new handlers object for the given frozen msgdef.
+   * Returns NULL if memory allocation failed. */
+  static reffed_ptr<Handlers> New(const MessageDef *m);
+
+  /* Convenience function for registering a graph of handlers that mirrors the
+   * graph of msgdefs for some message.  For "m" and all its children a new set
+   * of handlers will be created and the given callback will be invoked,
+   * allowing the client to register handlers for this message.  Note that any
+   * subhandlers set by the callback will be overwritten. */
+  static reffed_ptr<const Handlers> NewFrozen(const MessageDef *m,
+                                              HandlersCallback *callback,
+                                              const void *closure);
+
+  /* Functionality from upb::RefCounted. */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  /* All handler registration functions return bool to indicate success or
+   * failure; details about failures are stored in this status object.  If a
+   * failure does occur, it must be cleared before the Handlers are frozen,
+   * otherwise the freeze() operation will fail.  The functions may *only* be
+   * used while the Handlers are mutable. */
+  const Status* status();
+  void ClearError();
+
+  /* Call to freeze these Handlers.  Requires that any SubHandlers are already
+   * frozen.  For cycles, you must use the static version below and freeze the
+   * whole graph at once. */
+  bool Freeze(Status* s);
+
+  /* Freezes the given set of handlers.  You may not freeze a handler without
+   * also freezing any handlers they point to. */
+  static bool Freeze(Handlers*const* handlers, int n, Status* s);
+  static bool Freeze(const std::vector<Handlers*>& handlers, Status* s);
+
+  /* Returns the msgdef associated with this handlers object. */
+  const MessageDef* message_def() const;
+
+  /* Adds the given pointer and function to the list of cleanup functions that
+   * will be run when these handlers are freed.  If this pointer has previously
+   * been registered, the function returns false and does nothing. */
+  bool AddCleanup(void *ptr, upb_handlerfree *cleanup);
+
+  /* Sets the startmsg handler for the message, which is defined as follows:
+   *
+   *   bool startmsg(MyType* closure) {
+   *     // Called when the message begins.  Returns true if processing should
+   *     // continue.
+   *     return true;
+   *   }
+   */
+  bool SetStartMessageHandler(const StartMessageHandler& handler);
+
+  /* Sets the endmsg handler for the message, which is defined as follows:
+   *
+   *   bool endmsg(MyType* closure, upb_status *status) {
+   *     // Called when processing of this message ends, whether in success or
+   *     // failure.  "status" indicates the final status of processing, and
+   *     // can also be modified in-place to update the final status.
+   *   }
+   */
+  bool SetEndMessageHandler(const EndMessageHandler& handler);
+
+  /* Sets the value handler for the given field, which is defined as follows
+   * (this is for an int32 field; other field types will pass their native
+   * C/C++ type for "val"):
+   *
+   *   bool OnValue(MyClosure* c, const MyHandlerData* d, int32_t val) {
+   *     // Called when the field's value is encountered.  "d" contains
+   *     // whatever data was bound to this field when it was registered.
+   *     // Returns true if processing should continue.
+   *     return true;
+   *   }
+   *
+   *   handers->SetInt32Handler(f, UpbBind(OnValue, new MyHandlerData(...)));
+   *
+   * The value type must exactly match f->type().
+   * For example, a handler that takes an int32_t parameter may only be used for
+   * fields of type UPB_TYPE_INT32 and UPB_TYPE_ENUM.
+   *
+   * Returns false if the handler failed to register; in this case the cleanup
+   * handler (if any) will be called immediately.
+   */
+  bool SetInt32Handler (const FieldDef* f,  const Int32Handler& h);
+  bool SetInt64Handler (const FieldDef* f,  const Int64Handler& h);
+  bool SetUInt32Handler(const FieldDef* f, const UInt32Handler& h);
+  bool SetUInt64Handler(const FieldDef* f, const UInt64Handler& h);
+  bool SetFloatHandler (const FieldDef* f,  const FloatHandler& h);
+  bool SetDoubleHandler(const FieldDef* f, const DoubleHandler& h);
+  bool SetBoolHandler  (const FieldDef* f,   const BoolHandler& h);
+
+  /* Like the previous, but templated on the type on the value (ie. int32).
+   * This is mostly useful to call from other templates.  To call this you must
+   * specify the template parameter explicitly, ie:
+   *   h->SetValueHandler<T>(f, UpbBind(MyHandler<T>, MyData)); */
+  template <class T>
+  bool SetValueHandler(
+      const FieldDef *f,
+      const typename ValueHandler<typename CanonicalType<T>::Type>::H& handler);
+
+  /* Sets handlers for a string field, which are defined as follows:
+   *
+   *   MySubClosure* startstr(MyClosure* c, const MyHandlerData* d,
+   *                          size_t size_hint) {
+   *     // Called when a string value begins.  The return value indicates the
+   *     // closure for the string.  "size_hint" indicates the size of the
+   *     // string if it is known, however if the string is length-delimited
+   *     // and the end-of-string is not available size_hint will be zero.
+   *     // This case is indistinguishable from the case where the size is
+   *     // known to be zero.
+   *     //
+   *     // TODO(haberman): is it important to distinguish these cases?
+   *     // If we had ssize_t as a type we could make -1 "unknown", but
+   *     // ssize_t is POSIX (not ANSI) and therefore less portable.
+   *     // In practice I suspect it won't be important to distinguish.
+   *     return closure;
+   *   }
+   *
+   *   size_t str(MyClosure* closure, const MyHandlerData* d,
+   *              const char *str, size_t len) {
+   *     // Called for each buffer of string data; the multiple physical buffers
+   *     // are all part of the same logical string.  The return value indicates
+   *     // how many bytes were consumed.  If this number is less than "len",
+   *     // this will also indicate that processing should be halted for now,
+   *     // like returning false or UPB_BREAK from any other callback.  If
+   *     // number is greater than "len", the excess bytes will be skipped over
+   *     // and not passed to the callback.
+   *     return len;
+   *   }
+   *
+   *   bool endstr(MyClosure* c, const MyHandlerData* d) {
+   *     // Called when a string value ends.  Return value indicates whether
+   *     // processing should continue.
+   *     return true;
+   *   }
+   */
+  bool SetStartStringHandler(const FieldDef* f, const StartStringHandler& h);
+  bool SetStringHandler(const FieldDef* f, const StringHandler& h);
+  bool SetEndStringHandler(const FieldDef* f, const EndFieldHandler& h);
+
+  /* Sets the startseq handler, which is defined as follows:
+   *
+   *   MySubClosure *startseq(MyClosure* c, const MyHandlerData* d) {
+   *     // Called when a sequence (repeated field) begins.  The returned
+   *     // pointer indicates the closure for the sequence (or UPB_BREAK
+   *     // to interrupt processing).
+   *     return closure;
+   *   }
+   *
+   *   h->SetStartSequenceHandler(f, UpbBind(startseq, new MyHandlerData(...)));
+   *
+   * Returns "false" if "f" does not belong to this message or is not a
+   * repeated field.
+   */
+  bool SetStartSequenceHandler(const FieldDef* f, const StartFieldHandler& h);
+
+  /* Sets the startsubmsg handler for the given field, which is defined as
+   * follows:
+   *
+   *   MySubClosure* startsubmsg(MyClosure* c, const MyHandlerData* d) {
+   *     // Called when a submessage begins.  The returned pointer indicates the
+   *     // closure for the sequence (or UPB_BREAK to interrupt processing).
+   *     return closure;
+   *   }
+   *
+   *   h->SetStartSubMessageHandler(f, UpbBind(startsubmsg,
+   *                                           new MyHandlerData(...)));
+   *
+   * Returns "false" if "f" does not belong to this message or is not a
+   * submessage/group field.
+   */
+  bool SetStartSubMessageHandler(const FieldDef* f, const StartFieldHandler& h);
+
+  /* Sets the endsubmsg handler for the given field, which is defined as
+   * follows:
+   *
+   *   bool endsubmsg(MyClosure* c, const MyHandlerData* d) {
+   *     // Called when a submessage ends.  Returns true to continue processing.
+   *     return true;
+   *   }
+   *
+   * Returns "false" if "f" does not belong to this message or is not a
+   * submessage/group field.
+   */
+  bool SetEndSubMessageHandler(const FieldDef *f, const EndFieldHandler &h);
+
+  /* Starts the endsubseq handler for the given field, which is defined as
+   * follows:
+   *
+   *   bool endseq(MyClosure* c, const MyHandlerData* d) {
+   *     // Called when a sequence ends.  Returns true continue processing.
+   *     return true;
+   *   }
+   *
+   * Returns "false" if "f" does not belong to this message or is not a
+   * repeated field.
+   */
+  bool SetEndSequenceHandler(const FieldDef* f, const EndFieldHandler& h);
+
+  /* Sets or gets the object that specifies handlers for the given field, which
+   * must be a submessage or group.  Returns NULL if no handlers are set. */
+  bool SetSubHandlers(const FieldDef* f, const Handlers* sub);
+  const Handlers* GetSubHandlers(const FieldDef* f) const;
+
+  /* Equivalent to GetSubHandlers, but takes the STARTSUBMSG selector for the
+   * field. */
+  const Handlers* GetSubHandlers(Selector startsubmsg) const;
+
+  /* A selector refers to a specific field handler in the Handlers object
+   * (for example: the STARTSUBMSG handler for field "field15").
+   * On success, returns true and stores the selector in "s".
+   * If the FieldDef or Type are invalid, returns false.
+   * The returned selector is ONLY valid for Handlers whose MessageDef
+   * contains this FieldDef. */
+  static bool GetSelector(const FieldDef* f, Type type, Selector* s);
+
+  /* Given a START selector of any kind, returns the corresponding END selector. */
+  static Selector GetEndSelector(Selector start_selector);
+
+  /* Returns the function pointer for this handler.  It is the client's
+   * responsibility to cast to the correct function type before calling it. */
+  GenericFunction* GetHandler(Selector selector);
+
+  /* Sets the given attributes to the attributes for this selector. */
+  bool GetAttributes(Selector selector, HandlerAttributes* attr);
+
+  /* Returns the handler data that was registered with this handler. */
+  const void* GetHandlerData(Selector selector);
+
+  /* Could add any of the following functions as-needed, with some minor
+   * implementation changes:
+   *
+   * const FieldDef* GetFieldDef(Selector selector);
+   * static bool IsSequence(Selector selector); */
+
+ private:
+  UPB_DISALLOW_POD_OPS(Handlers, upb::Handlers)
+
+  friend UPB_INLINE GenericFunction *::upb_handlers_gethandler(
+      const upb_handlers *h, upb_selector_t s);
+  friend UPB_INLINE const void *::upb_handlers_gethandlerdata(
+      const upb_handlers *h, upb_selector_t s);
+#else
+struct upb_handlers {
+#endif
+  upb_refcounted base;
+
+  const upb_msgdef *msg;
+  const upb_handlers **sub;
+  const void *top_closure_type;
+  upb_inttable cleanup_;
+  upb_status status_;  /* Used only when mutable. */
+  upb_handlers_tabent table[1];  /* Dynamically-sized field handler array. */
+};
+
+#ifdef __cplusplus
+
+namespace upb {
+
+/* Convenience macros for creating a Handler object that is wrapped with a
+ * type-safe wrapper function that converts the "void*" parameters/returns
+ * of the underlying C API into nice C++ function.
+ *
+ * Sample usage:
+ *   void OnValue1(MyClosure* c, const MyHandlerData* d, int32_t val) {
+ *     // do stuff ...
+ *   }
+ *
+ *   // Handler that doesn't need any data bound to it.
+ *   void OnValue2(MyClosure* c, int32_t val) {
+ *     // do stuff ...
+ *   }
+ *
+ *   // Handler that returns bool so it can return failure if necessary.
+ *   bool OnValue3(MyClosure* c, int32_t val) {
+ *     // do stuff ...
+ *     return ok;
+ *   }
+ *
+ *   // Member function handler.
+ *   class MyClosure {
+ *    public:
+ *     void OnValue(int32_t val) {
+ *       // do stuff ...
+ *     }
+ *   };
+ *
+ *   // Takes ownership of the MyHandlerData.
+ *   handlers->SetInt32Handler(f1, UpbBind(OnValue1, new MyHandlerData(...)));
+ *   handlers->SetInt32Handler(f2, UpbMakeHandler(OnValue2));
+ *   handlers->SetInt32Handler(f1, UpbMakeHandler(OnValue3));
+ *   handlers->SetInt32Handler(f2, UpbMakeHandler(&MyClosure::OnValue));
+ */
+
+#ifdef UPB_CXX11
+
+/* In C++11, the "template" disambiguator can appear even outside templates,
+ * so all calls can safely use this pair of macros. */
+
+#define UpbMakeHandler(f) upb::MatchFunc(f).template GetFunc<f>()
+
+/* We have to be careful to only evaluate "d" once. */
+#define UpbBind(f, d) upb::MatchFunc(f).template GetFunc<f>((d))
+
+#else
+
+/* Prior to C++11, the "template" disambiguator may only appear inside a
+ * template, so the regular macro must not use "template" */
+
+#define UpbMakeHandler(f) upb::MatchFunc(f).GetFunc<f>()
+
+#define UpbBind(f, d) upb::MatchFunc(f).GetFunc<f>((d))
+
+#endif  /* UPB_CXX11 */
+
+/* This macro must be used in C++98 for calls from inside a template.  But we
+ * define this variant in all cases; code that wants to be compatible with both
+ * C++98 and C++11 should always use this macro when calling from a template. */
+#define UpbMakeHandlerT(f) upb::MatchFunc(f).template GetFunc<f>()
+
+/* We have to be careful to only evaluate "d" once. */
+#define UpbBindT(f, d) upb::MatchFunc(f).template GetFunc<f>((d))
+
+/* Handler: a struct that contains the (handler, data, deleter) tuple that is
+ * used to register all handlers.  Users can Make() these directly but it's
+ * more convenient to use the UpbMakeHandler/UpbBind macros above. */
+template <class T> class Handler {
+ public:
+  /* The underlying, handler function signature that upb uses internally. */
+  typedef T FuncPtr;
+
+  /* Intentionally implicit. */
+  template <class F> Handler(F func);
+  ~Handler();
+
+ private:
+  void AddCleanup(Handlers* h) const {
+    if (cleanup_func_) {
+      bool ok = h->AddCleanup(cleanup_data_, cleanup_func_);
+      UPB_ASSERT_VAR(ok, ok);
+    }
+  }
+
+  UPB_DISALLOW_COPY_AND_ASSIGN(Handler)
+  friend class Handlers;
+  FuncPtr handler_;
+  mutable HandlerAttributes attr_;
+  mutable bool registered_;
+  void *cleanup_data_;
+  upb_handlerfree *cleanup_func_;
+};
+
+}  /* namespace upb */
+
+#endif  /* __cplusplus */
+
+UPB_BEGIN_EXTERN_C
+
+/* Native C API. */
+
+/* Handler function typedefs. */
+typedef bool upb_startmsg_handlerfunc(void *c, const void*);
+typedef bool upb_endmsg_handlerfunc(void *c, const void *, upb_status *status);
+typedef void* upb_startfield_handlerfunc(void *c, const void *hd);
+typedef bool upb_endfield_handlerfunc(void *c, const void *hd);
+typedef bool upb_int32_handlerfunc(void *c, const void *hd, int32_t val);
+typedef bool upb_int64_handlerfunc(void *c, const void *hd, int64_t val);
+typedef bool upb_uint32_handlerfunc(void *c, const void *hd, uint32_t val);
+typedef bool upb_uint64_handlerfunc(void *c, const void *hd, uint64_t val);
+typedef bool upb_float_handlerfunc(void *c, const void *hd, float val);
+typedef bool upb_double_handlerfunc(void *c, const void *hd, double val);
+typedef bool upb_bool_handlerfunc(void *c, const void *hd, bool val);
+typedef void *upb_startstr_handlerfunc(void *c, const void *hd,
+                                       size_t size_hint);
+typedef size_t upb_string_handlerfunc(void *c, const void *hd, const char *buf,
+                                      size_t n, const upb_bufhandle* handle);
+
+/* upb_bufhandle */
+size_t upb_bufhandle_objofs(const upb_bufhandle *h);
+
+/* upb_handlerattr */
+void upb_handlerattr_init(upb_handlerattr *attr);
+void upb_handlerattr_uninit(upb_handlerattr *attr);
+
+bool upb_handlerattr_sethandlerdata(upb_handlerattr *attr, const void *hd);
+bool upb_handlerattr_setclosuretype(upb_handlerattr *attr, const void *type);
+const void *upb_handlerattr_closuretype(const upb_handlerattr *attr);
+bool upb_handlerattr_setreturnclosuretype(upb_handlerattr *attr,
+                                          const void *type);
+const void *upb_handlerattr_returnclosuretype(const upb_handlerattr *attr);
+bool upb_handlerattr_setalwaysok(upb_handlerattr *attr, bool alwaysok);
+bool upb_handlerattr_alwaysok(const upb_handlerattr *attr);
+
+UPB_INLINE const void *upb_handlerattr_handlerdata(
+    const upb_handlerattr *attr) {
+  return attr->handler_data_;
+}
+
+/* upb_handlers */
+typedef void upb_handlers_callback(const void *closure, upb_handlers *h);
+upb_handlers *upb_handlers_new(const upb_msgdef *m,
+                               const void *owner);
+const upb_handlers *upb_handlers_newfrozen(const upb_msgdef *m,
+                                           const void *owner,
+                                           upb_handlers_callback *callback,
+                                           const void *closure);
+
+/* Include refcounted methods like upb_handlers_ref(). */
+UPB_REFCOUNTED_CMETHODS(upb_handlers, upb_handlers_upcast)
+
+const upb_status *upb_handlers_status(upb_handlers *h);
+void upb_handlers_clearerr(upb_handlers *h);
+const upb_msgdef *upb_handlers_msgdef(const upb_handlers *h);
+bool upb_handlers_addcleanup(upb_handlers *h, void *p, upb_handlerfree *hfree);
+
+bool upb_handlers_setstartmsg(upb_handlers *h, upb_startmsg_handlerfunc *func,
+                              upb_handlerattr *attr);
+bool upb_handlers_setendmsg(upb_handlers *h, upb_endmsg_handlerfunc *func,
+                            upb_handlerattr *attr);
+bool upb_handlers_setint32(upb_handlers *h, const upb_fielddef *f,
+                           upb_int32_handlerfunc *func, upb_handlerattr *attr);
+bool upb_handlers_setint64(upb_handlers *h, const upb_fielddef *f,
+                           upb_int64_handlerfunc *func, upb_handlerattr *attr);
+bool upb_handlers_setuint32(upb_handlers *h, const upb_fielddef *f,
+                            upb_uint32_handlerfunc *func,
+                            upb_handlerattr *attr);
+bool upb_handlers_setuint64(upb_handlers *h, const upb_fielddef *f,
+                            upb_uint64_handlerfunc *func,
+                            upb_handlerattr *attr);
+bool upb_handlers_setfloat(upb_handlers *h, const upb_fielddef *f,
+                           upb_float_handlerfunc *func, upb_handlerattr *attr);
+bool upb_handlers_setdouble(upb_handlers *h, const upb_fielddef *f,
+                            upb_double_handlerfunc *func,
+                            upb_handlerattr *attr);
+bool upb_handlers_setbool(upb_handlers *h, const upb_fielddef *f,
+                          upb_bool_handlerfunc *func,
+                          upb_handlerattr *attr);
+bool upb_handlers_setstartstr(upb_handlers *h, const upb_fielddef *f,
+                              upb_startstr_handlerfunc *func,
+                              upb_handlerattr *attr);
+bool upb_handlers_setstring(upb_handlers *h, const upb_fielddef *f,
+                            upb_string_handlerfunc *func,
+                            upb_handlerattr *attr);
+bool upb_handlers_setendstr(upb_handlers *h, const upb_fielddef *f,
+                            upb_endfield_handlerfunc *func,
+                            upb_handlerattr *attr);
+bool upb_handlers_setstartseq(upb_handlers *h, const upb_fielddef *f,
+                              upb_startfield_handlerfunc *func,
+                              upb_handlerattr *attr);
+bool upb_handlers_setstartsubmsg(upb_handlers *h, const upb_fielddef *f,
+                                 upb_startfield_handlerfunc *func,
+                                 upb_handlerattr *attr);
+bool upb_handlers_setendsubmsg(upb_handlers *h, const upb_fielddef *f,
+                               upb_endfield_handlerfunc *func,
+                               upb_handlerattr *attr);
+bool upb_handlers_setendseq(upb_handlers *h, const upb_fielddef *f,
+                            upb_endfield_handlerfunc *func,
+                            upb_handlerattr *attr);
+
+bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f,
+                                 const upb_handlers *sub);
+const upb_handlers *upb_handlers_getsubhandlers(const upb_handlers *h,
+                                                const upb_fielddef *f);
+const upb_handlers *upb_handlers_getsubhandlers_sel(const upb_handlers *h,
+                                                    upb_selector_t sel);
+
+UPB_INLINE upb_func *upb_handlers_gethandler(const upb_handlers *h,
+                                             upb_selector_t s) {
+  return (upb_func *)h->table[s].func;
+}
+
+bool upb_handlers_getattr(const upb_handlers *h, upb_selector_t s,
+                          upb_handlerattr *attr);
+
+UPB_INLINE const void *upb_handlers_gethandlerdata(const upb_handlers *h,
+                                                   upb_selector_t s) {
+  return upb_handlerattr_handlerdata(&h->table[s].attr);
+}
+
+#ifdef __cplusplus
+
+/* Handler types for single fields.
+ * Right now we only have one for TYPE_BYTES but ones for other types
+ * should follow.
+ *
+ * These follow the same handlers protocol for fields of a message. */
+class upb::BytesHandler {
+ public:
+  BytesHandler();
+  ~BytesHandler();
+#else
+struct upb_byteshandler {
+#endif
+  upb_handlers_tabent table[3];
+};
+
+void upb_byteshandler_init(upb_byteshandler *h);
+
+/* Caller must ensure that "d" outlives the handlers.
+ * TODO(haberman): should this have a "freeze" operation?  It's not necessary
+ * for memory management, but could be useful to force immutability and provide
+ * a convenient moment to verify that all registration succeeded. */
+bool upb_byteshandler_setstartstr(upb_byteshandler *h,
+                                  upb_startstr_handlerfunc *func, void *d);
+bool upb_byteshandler_setstring(upb_byteshandler *h,
+                                upb_string_handlerfunc *func, void *d);
+bool upb_byteshandler_setendstr(upb_byteshandler *h,
+                                upb_endfield_handlerfunc *func, void *d);
+
+/* "Static" methods */
+bool upb_handlers_freeze(upb_handlers *const *handlers, int n, upb_status *s);
+upb_handlertype_t upb_handlers_getprimitivehandlertype(const upb_fielddef *f);
+bool upb_handlers_getselector(const upb_fielddef *f, upb_handlertype_t type,
+                              upb_selector_t *s);
+UPB_INLINE upb_selector_t upb_handlers_getendselector(upb_selector_t start) {
+  return start + 1;
+}
+
+/* Internal-only. */
+uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f);
+uint32_t upb_handlers_selectorcount(const upb_fielddef *f);
+
+UPB_END_EXTERN_C
+
+/*
+** Inline definitions for handlers.h, which are particularly long and a bit
+** tricky.
+*/
+
+#ifndef UPB_HANDLERS_INL_H_
+#define UPB_HANDLERS_INL_H_
+
+#include <limits.h>
+
+/* C inline methods. */
+
+/* upb_bufhandle */
+UPB_INLINE void upb_bufhandle_init(upb_bufhandle *h) {
+  h->obj_ = NULL;
+  h->objtype_ = NULL;
+  h->buf_ = NULL;
+  h->objofs_ = 0;
+}
+UPB_INLINE void upb_bufhandle_uninit(upb_bufhandle *h) {
+  UPB_UNUSED(h);
+}
+UPB_INLINE void upb_bufhandle_setobj(upb_bufhandle *h, const void *obj,
+                                     const void *type) {
+  h->obj_ = obj;
+  h->objtype_ = type;
+}
+UPB_INLINE void upb_bufhandle_setbuf(upb_bufhandle *h, const char *buf,
+                                     size_t ofs) {
+  h->buf_ = buf;
+  h->objofs_ = ofs;
+}
+UPB_INLINE const void *upb_bufhandle_obj(const upb_bufhandle *h) {
+  return h->obj_;
+}
+UPB_INLINE const void *upb_bufhandle_objtype(const upb_bufhandle *h) {
+  return h->objtype_;
+}
+UPB_INLINE const char *upb_bufhandle_buf(const upb_bufhandle *h) {
+  return h->buf_;
+}
+
+
+#ifdef __cplusplus
+
+/* Type detection and typedefs for integer types.
+ * For platforms where there are multiple 32-bit or 64-bit types, we need to be
+ * able to enumerate them so we can properly create overloads for all variants.
+ *
+ * If any platform existed where there were three integer types with the same
+ * size, this would have to become more complicated.  For example, short, int,
+ * and long could all be 32-bits.  Even more diabolically, short, int, long,
+ * and long long could all be 64 bits and still be standard-compliant.
+ * However, few platforms are this strange, and it's unlikely that upb will be
+ * used on the strangest ones. */
+
+/* Can't count on stdint.h limits like INT32_MAX, because in C++ these are
+ * only defined when __STDC_LIMIT_MACROS are defined before the *first* include
+ * of stdint.h.  We can't guarantee that someone else didn't include these first
+ * without defining __STDC_LIMIT_MACROS. */
+#define UPB_INT32_MAX 0x7fffffffLL
+#define UPB_INT32_MIN (-UPB_INT32_MAX - 1)
+#define UPB_INT64_MAX 0x7fffffffffffffffLL
+#define UPB_INT64_MIN (-UPB_INT64_MAX - 1)
+
+#if INT_MAX == UPB_INT32_MAX && INT_MIN == UPB_INT32_MIN
+#define UPB_INT_IS_32BITS 1
+#endif
+
+#if LONG_MAX == UPB_INT32_MAX && LONG_MIN == UPB_INT32_MIN
+#define UPB_LONG_IS_32BITS 1
+#endif
+
+#if LONG_MAX == UPB_INT64_MAX && LONG_MIN == UPB_INT64_MIN
+#define UPB_LONG_IS_64BITS 1
+#endif
+
+#if LLONG_MAX == UPB_INT64_MAX && LLONG_MIN == UPB_INT64_MIN
+#define UPB_LLONG_IS_64BITS 1
+#endif
+
+/* We use macros instead of typedefs so we can undefine them later and avoid
+ * leaking them outside this header file. */
+#if UPB_INT_IS_32BITS
+#define UPB_INT32_T int
+#define UPB_UINT32_T unsigned int
+
+#if UPB_LONG_IS_32BITS
+#define UPB_TWO_32BIT_TYPES 1
+#define UPB_INT32ALT_T long
+#define UPB_UINT32ALT_T unsigned long
+#endif  /* UPB_LONG_IS_32BITS */
+
+#elif UPB_LONG_IS_32BITS  /* && !UPB_INT_IS_32BITS */
+#define UPB_INT32_T long
+#define UPB_UINT32_T unsigned long
+#endif  /* UPB_INT_IS_32BITS */
+
+
+#if UPB_LONG_IS_64BITS
+#define UPB_INT64_T long
+#define UPB_UINT64_T unsigned long
+
+#if UPB_LLONG_IS_64BITS
+#define UPB_TWO_64BIT_TYPES 1
+#define UPB_INT64ALT_T long long
+#define UPB_UINT64ALT_T unsigned long long
+#endif  /* UPB_LLONG_IS_64BITS */
+
+#elif UPB_LLONG_IS_64BITS  /* && !UPB_LONG_IS_64BITS */
+#define UPB_INT64_T long long
+#define UPB_UINT64_T unsigned long long
+#endif  /* UPB_LONG_IS_64BITS */
+
+#undef UPB_INT32_MAX
+#undef UPB_INT32_MIN
+#undef UPB_INT64_MAX
+#undef UPB_INT64_MIN
+#undef UPB_INT_IS_32BITS
+#undef UPB_LONG_IS_32BITS
+#undef UPB_LONG_IS_64BITS
+#undef UPB_LLONG_IS_64BITS
+
+
+namespace upb {
+
+typedef void CleanupFunc(void *ptr);
+
+/* Template to remove "const" from "const T*" and just return "T*".
+ *
+ * We define a nonsense default because otherwise it will fail to instantiate as
+ * a function parameter type even in cases where we don't expect any caller to
+ * actually match the overload. */
+class CouldntRemoveConst {};
+template <class T> struct remove_constptr { typedef CouldntRemoveConst type; };
+template <class T> struct remove_constptr<const T *> { typedef T *type; };
+
+/* Template that we use below to remove a template specialization from
+ * consideration if it matches a specific type. */
+template <class T, class U> struct disable_if_same { typedef void Type; };
+template <class T> struct disable_if_same<T, T> {};
+
+template <class T> void DeletePointer(void *p) { delete static_cast<T>(p); }
+
+template <class T1, class T2>
+struct FirstUnlessVoidOrBool {
+  typedef T1 value;
+};
+
+template <class T2>
+struct FirstUnlessVoidOrBool<void, T2> {
+  typedef T2 value;
+};
+
+template <class T2>
+struct FirstUnlessVoidOrBool<bool, T2> {
+  typedef T2 value;
+};
+
+template<class T, class U>
+struct is_same {
+  static bool value;
+};
+
+template<class T>
+struct is_same<T, T> {
+  static bool value;
+};
+
+template<class T, class U>
+bool is_same<T, U>::value = false;
+
+template<class T>
+bool is_same<T, T>::value = true;
+
+/* FuncInfo *******************************************************************/
+
+/* Info about the user's original, pre-wrapped function. */
+template <class C, class R = void>
+struct FuncInfo {
+  /* The type of the closure that the function takes (its first param). */
+  typedef C Closure;
+
+  /* The return type. */
+  typedef R Return;
+};
+
+/* Func ***********************************************************************/
+
+/* Func1, Func2, Func3: Template classes representing a function and its
+ * signature.
+ *
+ * Since the function is a template parameter, calling the function can be
+ * inlined at compile-time and does not require a function pointer at runtime.
+ * These functions are not bound to a handler data so have no data or cleanup
+ * handler. */
+struct UnboundFunc {
+  CleanupFunc *GetCleanup() { return NULL; }
+  void *GetData() { return NULL; }
+};
+
+template <class R, class P1, R F(P1), class I>
+struct Func1 : public UnboundFunc {
+  typedef R Return;
+  typedef I FuncInfo;
+  static R Call(P1 p1) { return F(p1); }
+};
+
+template <class R, class P1, class P2, R F(P1, P2), class I>
+struct Func2 : public UnboundFunc {
+  typedef R Return;
+  typedef I FuncInfo;
+  static R Call(P1 p1, P2 p2) { return F(p1, p2); }
+};
+
+template <class R, class P1, class P2, class P3, R F(P1, P2, P3), class I>
+struct Func3 : public UnboundFunc {
+  typedef R Return;
+  typedef I FuncInfo;
+  static R Call(P1 p1, P2 p2, P3 p3) { return F(p1, p2, p3); }
+};
+
+template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4),
+          class I>
+struct Func4 : public UnboundFunc {
+  typedef R Return;
+  typedef I FuncInfo;
+  static R Call(P1 p1, P2 p2, P3 p3, P4 p4) { return F(p1, p2, p3, p4); }
+};
+
+template <class R, class P1, class P2, class P3, class P4, class P5,
+          R F(P1, P2, P3, P4, P5), class I>
+struct Func5 : public UnboundFunc {
+  typedef R Return;
+  typedef I FuncInfo;
+  static R Call(P1 p1, P2 p2, P3 p3, P4 p4, P5 p5) {
+    return F(p1, p2, p3, p4, p5);
+  }
+};
+
+/* BoundFunc ******************************************************************/
+
+/* BoundFunc2, BoundFunc3: Like Func2/Func3 except also contains a value that
+ * shall be bound to the function's second parameter.
+ * 
+ * Note that the second parameter is a const pointer, but our stored bound value
+ * is non-const so we can free it when the handlers are destroyed. */
+template <class T>
+struct BoundFunc {
+  typedef typename remove_constptr<T>::type MutableP2;
+  explicit BoundFunc(MutableP2 data_) : data(data_) {}
+  CleanupFunc *GetCleanup() { return &DeletePointer<MutableP2>; }
+  MutableP2 GetData() { return data; }
+  MutableP2 data;
+};
+
+template <class R, class P1, class P2, R F(P1, P2), class I>
+struct BoundFunc2 : public BoundFunc<P2> {
+  typedef BoundFunc<P2> Base;
+  typedef I FuncInfo;
+  explicit BoundFunc2(typename Base::MutableP2 arg) : Base(arg) {}
+};
+
+template <class R, class P1, class P2, class P3, R F(P1, P2, P3), class I>
+struct BoundFunc3 : public BoundFunc<P2> {
+  typedef BoundFunc<P2> Base;
+  typedef I FuncInfo;
+  explicit BoundFunc3(typename Base::MutableP2 arg) : Base(arg) {}
+};
+
+template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4),
+          class I>
+struct BoundFunc4 : public BoundFunc<P2> {
+  typedef BoundFunc<P2> Base;
+  typedef I FuncInfo;
+  explicit BoundFunc4(typename Base::MutableP2 arg) : Base(arg) {}
+};
+
+template <class R, class P1, class P2, class P3, class P4, class P5,
+          R F(P1, P2, P3, P4, P5), class I>
+struct BoundFunc5 : public BoundFunc<P2> {
+  typedef BoundFunc<P2> Base;
+  typedef I FuncInfo;
+  explicit BoundFunc5(typename Base::MutableP2 arg) : Base(arg) {}
+};
+
+/* FuncSig ********************************************************************/
+
+/* FuncSig1, FuncSig2, FuncSig3: template classes reflecting a function
+ * *signature*, but without a specific function attached.
+ *
+ * These classes contain member functions that can be invoked with a
+ * specific function to return a Func/BoundFunc class. */
+template <class R, class P1>
+struct FuncSig1 {
+  template <R F(P1)>
+  Func1<R, P1, F, FuncInfo<P1, R> > GetFunc() {
+    return Func1<R, P1, F, FuncInfo<P1, R> >();
+  }
+};
+
+template <class R, class P1, class P2>
+struct FuncSig2 {
+  template <R F(P1, P2)>
+  Func2<R, P1, P2, F, FuncInfo<P1, R> > GetFunc() {
+    return Func2<R, P1, P2, F, FuncInfo<P1, R> >();
+  }
+
+  template <R F(P1, P2)>
+  BoundFunc2<R, P1, P2, F, FuncInfo<P1, R> > GetFunc(
+      typename remove_constptr<P2>::type param2) {
+    return BoundFunc2<R, P1, P2, F, FuncInfo<P1, R> >(param2);
+  }
+};
+
+template <class R, class P1, class P2, class P3>
+struct FuncSig3 {
+  template <R F(P1, P2, P3)>
+  Func3<R, P1, P2, P3, F, FuncInfo<P1, R> > GetFunc() {
+    return Func3<R, P1, P2, P3, F, FuncInfo<P1, R> >();
+  }
+
+  template <R F(P1, P2, P3)>
+  BoundFunc3<R, P1, P2, P3, F, FuncInfo<P1, R> > GetFunc(
+      typename remove_constptr<P2>::type param2) {
+    return BoundFunc3<R, P1, P2, P3, F, FuncInfo<P1, R> >(param2);
+  }
+};
+
+template <class R, class P1, class P2, class P3, class P4>
+struct FuncSig4 {
+  template <R F(P1, P2, P3, P4)>
+  Func4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> > GetFunc() {
+    return Func4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> >();
+  }
+
+  template <R F(P1, P2, P3, P4)>
+  BoundFunc4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> > GetFunc(
+      typename remove_constptr<P2>::type param2) {
+    return BoundFunc4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> >(param2);
+  }
+};
+
+template <class R, class P1, class P2, class P3, class P4, class P5>
+struct FuncSig5 {
+  template <R F(P1, P2, P3, P4, P5)>
+  Func5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> > GetFunc() {
+    return Func5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> >();
+  }
+
+  template <R F(P1, P2, P3, P4, P5)>
+  BoundFunc5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> > GetFunc(
+      typename remove_constptr<P2>::type param2) {
+    return BoundFunc5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> >(param2);
+  }
+};
+
+/* Overloaded template function that can construct the appropriate FuncSig*
+ * class given a function pointer by deducing the template parameters. */
+template <class R, class P1>
+inline FuncSig1<R, P1> MatchFunc(R (*f)(P1)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return FuncSig1<R, P1>();
+}
+
+template <class R, class P1, class P2>
+inline FuncSig2<R, P1, P2> MatchFunc(R (*f)(P1, P2)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return FuncSig2<R, P1, P2>();
+}
+
+template <class R, class P1, class P2, class P3>
+inline FuncSig3<R, P1, P2, P3> MatchFunc(R (*f)(P1, P2, P3)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return FuncSig3<R, P1, P2, P3>();
+}
+
+template <class R, class P1, class P2, class P3, class P4>
+inline FuncSig4<R, P1, P2, P3, P4> MatchFunc(R (*f)(P1, P2, P3, P4)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return FuncSig4<R, P1, P2, P3, P4>();
+}
+
+template <class R, class P1, class P2, class P3, class P4, class P5>
+inline FuncSig5<R, P1, P2, P3, P4, P5> MatchFunc(R (*f)(P1, P2, P3, P4, P5)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return FuncSig5<R, P1, P2, P3, P4, P5>();
+}
+
+/* MethodSig ******************************************************************/
+
+/* CallMethod*: a function template that calls a given method. */
+template <class R, class C, R (C::*F)()>
+R CallMethod0(C *obj) {
+  return ((*obj).*F)();
+}
+
+template <class R, class C, class P1, R (C::*F)(P1)>
+R CallMethod1(C *obj, P1 arg1) {
+  return ((*obj).*F)(arg1);
+}
+
+template <class R, class C, class P1, class P2, R (C::*F)(P1, P2)>
+R CallMethod2(C *obj, P1 arg1, P2 arg2) {
+  return ((*obj).*F)(arg1, arg2);
+}
+
+template <class R, class C, class P1, class P2, class P3, R (C::*F)(P1, P2, P3)>
+R CallMethod3(C *obj, P1 arg1, P2 arg2, P3 arg3) {
+  return ((*obj).*F)(arg1, arg2, arg3);
+}
+
+template <class R, class C, class P1, class P2, class P3, class P4,
+          R (C::*F)(P1, P2, P3, P4)>
+R CallMethod4(C *obj, P1 arg1, P2 arg2, P3 arg3, P4 arg4) {
+  return ((*obj).*F)(arg1, arg2, arg3, arg4);
+}
+
+/* MethodSig: like FuncSig, but for member functions.
+ *
+ * GetFunc() returns a normal FuncN object, so after calling GetFunc() no
+ * more logic is required to special-case methods. */
+template <class R, class C>
+struct MethodSig0 {
+  template <R (C::*F)()>
+  Func1<R, C *, CallMethod0<R, C, F>, FuncInfo<C *, R> > GetFunc() {
+    return Func1<R, C *, CallMethod0<R, C, F>, FuncInfo<C *, R> >();
+  }
+};
+
+template <class R, class C, class P1>
+struct MethodSig1 {
+  template <R (C::*F)(P1)>
+  Func2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> > GetFunc() {
+    return Func2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> >();
+  }
+
+  template <R (C::*F)(P1)>
+  BoundFunc2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> > GetFunc(
+      typename remove_constptr<P1>::type param1) {
+    return BoundFunc2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> >(
+        param1);
+  }
+};
+
+template <class R, class C, class P1, class P2>
+struct MethodSig2 {
+  template <R (C::*F)(P1, P2)>
+  Func3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>, FuncInfo<C *, R> >
+  GetFunc() {
+    return Func3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>,
+                 FuncInfo<C *, R> >();
+  }
+
+  template <R (C::*F)(P1, P2)>
+  BoundFunc3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>, FuncInfo<C *, R> >
+  GetFunc(typename remove_constptr<P1>::type param1) {
+    return BoundFunc3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>,
+                      FuncInfo<C *, R> >(param1);
+  }
+};
+
+template <class R, class C, class P1, class P2, class P3>
+struct MethodSig3 {
+  template <R (C::*F)(P1, P2, P3)>
+  Func4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>, FuncInfo<C *, R> >
+  GetFunc() {
+    return Func4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>,
+                 FuncInfo<C *, R> >();
+  }
+
+  template <R (C::*F)(P1, P2, P3)>
+  BoundFunc4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>,
+             FuncInfo<C *, R> >
+  GetFunc(typename remove_constptr<P1>::type param1) {
+    return BoundFunc4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>,
+                      FuncInfo<C *, R> >(param1);
+  }
+};
+
+template <class R, class C, class P1, class P2, class P3, class P4>
+struct MethodSig4 {
+  template <R (C::*F)(P1, P2, P3, P4)>
+  Func5<R, C *, P1, P2, P3, P4, CallMethod4<R, C, P1, P2, P3, P4, F>,
+        FuncInfo<C *, R> >
+  GetFunc() {
+    return Func5<R, C *, P1, P2, P3, P4, CallMethod4<R, C, P1, P2, P3, P4, F>,
+                 FuncInfo<C *, R> >();
+  }
+
+  template <R (C::*F)(P1, P2, P3, P4)>
+  BoundFunc5<R, C *, P1, P2, P3, P4, CallMethod4<R, C, P1, P2, P3, P4, F>,
+             FuncInfo<C *, R> >
+  GetFunc(typename remove_constptr<P1>::type param1) {
+    return BoundFunc5<R, C *, P1, P2, P3, P4,
+                      CallMethod4<R, C, P1, P2, P3, P4, F>, FuncInfo<C *, R> >(
+        param1);
+  }
+};
+
+template <class R, class C>
+inline MethodSig0<R, C> MatchFunc(R (C::*f)()) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return MethodSig0<R, C>();
+}
+
+template <class R, class C, class P1>
+inline MethodSig1<R, C, P1> MatchFunc(R (C::*f)(P1)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return MethodSig1<R, C, P1>();
+}
+
+template <class R, class C, class P1, class P2>
+inline MethodSig2<R, C, P1, P2> MatchFunc(R (C::*f)(P1, P2)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return MethodSig2<R, C, P1, P2>();
+}
+
+template <class R, class C, class P1, class P2, class P3>
+inline MethodSig3<R, C, P1, P2, P3> MatchFunc(R (C::*f)(P1, P2, P3)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return MethodSig3<R, C, P1, P2, P3>();
+}
+
+template <class R, class C, class P1, class P2, class P3, class P4>
+inline MethodSig4<R, C, P1, P2, P3, P4> MatchFunc(R (C::*f)(P1, P2, P3, P4)) {
+  UPB_UNUSED(f);  /* Only used for template parameter deduction. */
+  return MethodSig4<R, C, P1, P2, P3, P4>();
+}
+
+/* MaybeWrapReturn ************************************************************/
+
+/* Template class that attempts to wrap the return value of the function so it
+ * matches the expected type.  There are two main adjustments it may make:
+ *
+ *   1. If the function returns void, make it return the expected type and with
+ *      a value that always indicates success.
+ *   2. If the function returns bool, make it return the expected type with a
+ *      value that indicates success or failure.
+ *
+ * The "expected type" for return is:
+ *   1. void* for start handlers.  If the closure parameter has a different type
+ *      we will cast it to void* for the return in the success case.
+ *   2. size_t for string buffer handlers.
+ *   3. bool for everything else. */
+
+/* Template parameters are FuncN type and desired return type. */
+template <class F, class R, class Enable = void>
+struct MaybeWrapReturn;
+
+/* If the return type matches, return the given function unwrapped. */
+template <class F>
+struct MaybeWrapReturn<F, typename F::Return> {
+  typedef F Func;
+};
+
+/* Function wrapper that munges the return value from void to (bool)true. */
+template <class P1, class P2, void F(P1, P2)>
+bool ReturnTrue2(P1 p1, P2 p2) {
+  F(p1, p2);
+  return true;
+}
+
+template <class P1, class P2, class P3, void F(P1, P2, P3)>
+bool ReturnTrue3(P1 p1, P2 p2, P3 p3) {
+  F(p1, p2, p3);
+  return true;
+}
+
+/* Function wrapper that munges the return value from void to (void*)arg1  */
+template <class P1, class P2, void F(P1, P2)>
+void *ReturnClosure2(P1 p1, P2 p2) {
+  F(p1, p2);
+  return p1;
+}
+
+template <class P1, class P2, class P3, void F(P1, P2, P3)>
+void *ReturnClosure3(P1 p1, P2 p2, P3 p3) {
+  F(p1, p2, p3);
+  return p1;
+}
+
+/* Function wrapper that munges the return value from R to void*. */
+template <class R, class P1, class P2, R F(P1, P2)>
+void *CastReturnToVoidPtr2(P1 p1, P2 p2) {
+  return F(p1, p2);
+}
+
+template <class R, class P1, class P2, class P3, R F(P1, P2, P3)>
+void *CastReturnToVoidPtr3(P1 p1, P2 p2, P3 p3) {
+  return F(p1, p2, p3);
+}
+
+/* Function wrapper that munges the return value from bool to void*. */
+template <class P1, class P2, bool F(P1, P2)>
+void *ReturnClosureOrBreak2(P1 p1, P2 p2) {
+  return F(p1, p2) ? p1 : UPB_BREAK;
+}
+
+template <class P1, class P2, class P3, bool F(P1, P2, P3)>
+void *ReturnClosureOrBreak3(P1 p1, P2 p2, P3 p3) {
+  return F(p1, p2, p3) ? p1 : UPB_BREAK;
+}
+
+/* For the string callback, which takes five params, returns the size param. */
+template <class P1, class P2,
+          void F(P1, P2, const char *, size_t, const BufferHandle *)>
+size_t ReturnStringLen(P1 p1, P2 p2, const char *p3, size_t p4,
+                       const BufferHandle *p5) {
+  F(p1, p2, p3, p4, p5);
+  return p4;
+}
+
+/* For the string callback, which takes five params, returns the size param or
+ * zero. */
+template <class P1, class P2,
+          bool F(P1, P2, const char *, size_t, const BufferHandle *)>
+size_t ReturnNOr0(P1 p1, P2 p2, const char *p3, size_t p4,
+                  const BufferHandle *p5) {
+  return F(p1, p2, p3, p4, p5) ? p4 : 0;
+}
+
+/* If we have a function returning void but want a function returning bool, wrap
+ * it in a function that returns true. */
+template <class P1, class P2, void F(P1, P2), class I>
+struct MaybeWrapReturn<Func2<void, P1, P2, F, I>, bool> {
+  typedef Func2<bool, P1, P2, ReturnTrue2<P1, P2, F>, I> Func;
+};
+
+template <class P1, class P2, class P3, void F(P1, P2, P3), class I>
+struct MaybeWrapReturn<Func3<void, P1, P2, P3, F, I>, bool> {
+  typedef Func3<bool, P1, P2, P3, ReturnTrue3<P1, P2, P3, F>, I> Func;
+};
+
+/* If our function returns void but we want one returning void*, wrap it in a
+ * function that returns the first argument. */
+template <class P1, class P2, void F(P1, P2), class I>
+struct MaybeWrapReturn<Func2<void, P1, P2, F, I>, void *> {
+  typedef Func2<void *, P1, P2, ReturnClosure2<P1, P2, F>, I> Func;
+};
+
+template <class P1, class P2, class P3, void F(P1, P2, P3), class I>
+struct MaybeWrapReturn<Func3<void, P1, P2, P3, F, I>, void *> {
+  typedef Func3<void *, P1, P2, P3, ReturnClosure3<P1, P2, P3, F>, I> Func;
+};
+
+/* If our function returns R* but we want one returning void*, wrap it in a
+ * function that casts to void*. */
+template <class R, class P1, class P2, R *F(P1, P2), class I>
+struct MaybeWrapReturn<Func2<R *, P1, P2, F, I>, void *,
+                       typename disable_if_same<R *, void *>::Type> {
+  typedef Func2<void *, P1, P2, CastReturnToVoidPtr2<R *, P1, P2, F>, I> Func;
+};
+
+template <class R, class P1, class P2, class P3, R *F(P1, P2, P3), class I>
+struct MaybeWrapReturn<Func3<R *, P1, P2, P3, F, I>, void *,
+                       typename disable_if_same<R *, void *>::Type> {
+  typedef Func3<void *, P1, P2, P3, CastReturnToVoidPtr3<R *, P1, P2, P3, F>, I>
+      Func;
+};
+
+/* If our function returns bool but we want one returning void*, wrap it in a
+ * function that returns either the first param or UPB_BREAK. */
+template <class P1, class P2, bool F(P1, P2), class I>
+struct MaybeWrapReturn<Func2<bool, P1, P2, F, I>, void *> {
+  typedef Func2<void *, P1, P2, ReturnClosureOrBreak2<P1, P2, F>, I> Func;
+};
+
+template <class P1, class P2, class P3, bool F(P1, P2, P3), class I>
+struct MaybeWrapReturn<Func3<bool, P1, P2, P3, F, I>, void *> {
+  typedef Func3<void *, P1, P2, P3, ReturnClosureOrBreak3<P1, P2, P3, F>, I>
+      Func;
+};
+
+/* If our function returns void but we want one returning size_t, wrap it in a
+ * function that returns the size argument. */
+template <class P1, class P2,
+          void F(P1, P2, const char *, size_t, const BufferHandle *), class I>
+struct MaybeWrapReturn<
+    Func5<void, P1, P2, const char *, size_t, const BufferHandle *, F, I>,
+          size_t> {
+  typedef Func5<size_t, P1, P2, const char *, size_t, const BufferHandle *,
+                ReturnStringLen<P1, P2, F>, I> Func;
+};
+
+/* If our function returns bool but we want one returning size_t, wrap it in a
+ * function that returns either 0 or the buf size. */
+template <class P1, class P2,
+          bool F(P1, P2, const char *, size_t, const BufferHandle *), class I>
+struct MaybeWrapReturn<
+    Func5<bool, P1, P2, const char *, size_t, const BufferHandle *, F, I>,
+    size_t> {
+  typedef Func5<size_t, P1, P2, const char *, size_t, const BufferHandle *,
+                ReturnNOr0<P1, P2, F>, I> Func;
+};
+
+/* ConvertParams **************************************************************/
+
+/* Template class that converts the function parameters if necessary, and
+ * ignores the HandlerData parameter if appropriate.
+ *
+ * Template parameter is the are FuncN function type. */
+template <class F, class T>
+struct ConvertParams;
+
+/* Function that discards the handler data parameter. */
+template <class R, class P1, R F(P1)>
+R IgnoreHandlerData2(void *p1, const void *hd) {
+  UPB_UNUSED(hd);
+  return F(static_cast<P1>(p1));
+}
+
+template <class R, class P1, class P2Wrapper, class P2Wrapped,
+          R F(P1, P2Wrapped)>
+R IgnoreHandlerData3(void *p1, const void *hd, P2Wrapper p2) {
+  UPB_UNUSED(hd);
+  return F(static_cast<P1>(p1), p2);
+}
+
+template <class R, class P1, class P2, class P3, R F(P1, P2, P3)>
+R IgnoreHandlerData4(void *p1, const void *hd, P2 p2, P3 p3) {
+  UPB_UNUSED(hd);
+  return F(static_cast<P1>(p1), p2, p3);
+}
+
+template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4)>
+R IgnoreHandlerData5(void *p1, const void *hd, P2 p2, P3 p3, P4 p4) {
+  UPB_UNUSED(hd);
+  return F(static_cast<P1>(p1), p2, p3, p4);
+}
+
+template <class R, class P1, R F(P1, const char*, size_t)>
+R IgnoreHandlerDataIgnoreHandle(void *p1, const void *hd, const char *p2,
+                                size_t p3, const BufferHandle *handle) {
+  UPB_UNUSED(hd);
+  UPB_UNUSED(handle);
+  return F(static_cast<P1>(p1), p2, p3);
+}
+
+/* Function that casts the handler data parameter. */
+template <class R, class P1, class P2, R F(P1, P2)>
+R CastHandlerData2(void *c, const void *hd) {
+  return F(static_cast<P1>(c), static_cast<P2>(hd));
+}
+
+template <class R, class P1, class P2, class P3Wrapper, class P3Wrapped,
+          R F(P1, P2, P3Wrapped)>
+R CastHandlerData3(void *c, const void *hd, P3Wrapper p3) {
+  return F(static_cast<P1>(c), static_cast<P2>(hd), p3);
+}
+
+template <class R, class P1, class P2, class P3, class P4, class P5,
+          R F(P1, P2, P3, P4, P5)>
+R CastHandlerData5(void *c, const void *hd, P3 p3, P4 p4, P5 p5) {
+  return F(static_cast<P1>(c), static_cast<P2>(hd), p3, p4, p5);
+}
+
+template <class R, class P1, class P2, R F(P1, P2, const char *, size_t)>
+R CastHandlerDataIgnoreHandle(void *c, const void *hd, const char *p3,
+                              size_t p4, const BufferHandle *handle) {
+  UPB_UNUSED(handle);
+  return F(static_cast<P1>(c), static_cast<P2>(hd), p3, p4);
+}
+
+/* For unbound functions, ignore the handler data. */
+template <class R, class P1, R F(P1), class I, class T>
+struct ConvertParams<Func1<R, P1, F, I>, T> {
+  typedef Func2<R, void *, const void *, IgnoreHandlerData2<R, P1, F>, I> Func;
+};
+
+template <class R, class P1, class P2, R F(P1, P2), class I,
+          class R2, class P1_2, class P2_2, class P3_2>
+struct ConvertParams<Func2<R, P1, P2, F, I>,
+                     R2 (*)(P1_2, P2_2, P3_2)> {
+  typedef Func3<R, void *, const void *, P3_2,
+                IgnoreHandlerData3<R, P1, P3_2, P2, F>, I> Func;
+};
+
+/* For StringBuffer only; this ignores both the handler data and the
+ * BufferHandle. */
+template <class R, class P1, R F(P1, const char *, size_t), class I, class T>
+struct ConvertParams<Func3<R, P1, const char *, size_t, F, I>, T> {
+  typedef Func5<R, void *, const void *, const char *, size_t,
+                const BufferHandle *, IgnoreHandlerDataIgnoreHandle<R, P1, F>,
+                I> Func;
+};
+
+template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4),
+          class I, class T>
+struct ConvertParams<Func4<R, P1, P2, P3, P4, F, I>, T> {
+  typedef Func5<R, void *, const void *, P2, P3, P4,
+                IgnoreHandlerData5<R, P1, P2, P3, P4, F>, I> Func;
+};
+
+/* For bound functions, cast the handler data. */
+template <class R, class P1, class P2, R F(P1, P2), class I, class T>
+struct ConvertParams<BoundFunc2<R, P1, P2, F, I>, T> {
+  typedef Func2<R, void *, const void *, CastHandlerData2<R, P1, P2, F>, I>
+      Func;
+};
+
+template <class R, class P1, class P2, class P3, R F(P1, P2, P3), class I,
+          class R2, class P1_2, class P2_2, class P3_2>
+struct ConvertParams<BoundFunc3<R, P1, P2, P3, F, I>,
+                     R2 (*)(P1_2, P2_2, P3_2)> {
+  typedef Func3<R, void *, const void *, P3_2,
+                CastHandlerData3<R, P1, P2, P3_2, P3, F>, I> Func;
+};
+
+/* For StringBuffer only; this ignores the BufferHandle. */
+template <class R, class P1, class P2, R F(P1, P2, const char *, size_t),
+          class I, class T>
+struct ConvertParams<BoundFunc4<R, P1, P2, const char *, size_t, F, I>, T> {
+  typedef Func5<R, void *, const void *, const char *, size_t,
+                const BufferHandle *, CastHandlerDataIgnoreHandle<R, P1, P2, F>,
+                I> Func;
+};
+
+template <class R, class P1, class P2, class P3, class P4, class P5,
+          R F(P1, P2, P3, P4, P5), class I, class T>
+struct ConvertParams<BoundFunc5<R, P1, P2, P3, P4, P5, F, I>, T> {
+  typedef Func5<R, void *, const void *, P3, P4, P5,
+                CastHandlerData5<R, P1, P2, P3, P4, P5, F>, I> Func;
+};
+
+/* utype/ltype are upper/lower-case, ctype is canonical C type, vtype is
+ * variant C type. */
+#define TYPE_METHODS(utype, ltype, ctype, vtype)                               \
+  template <> struct CanonicalType<vtype> {                                    \
+    typedef ctype Type;                                                        \
+  };                                                                           \
+  template <>                                                                  \
+  inline bool Handlers::SetValueHandler<vtype>(                                \
+      const FieldDef *f,                                                       \
+      const Handlers::utype ## Handler& handler) {                             \
+    assert(!handler.registered_);                                              \
+    handler.AddCleanup(this);                                                  \
+    handler.registered_ = true;                                                \
+    return upb_handlers_set##ltype(this, f, handler.handler_, &handler.attr_); \
+  }                                                                            \
+
+TYPE_METHODS(Double, double, double,   double)
+TYPE_METHODS(Float,  float,  float,    float)
+TYPE_METHODS(UInt64, uint64, uint64_t, UPB_UINT64_T)
+TYPE_METHODS(UInt32, uint32, uint32_t, UPB_UINT32_T)
+TYPE_METHODS(Int64,  int64,  int64_t,  UPB_INT64_T)
+TYPE_METHODS(Int32,  int32,  int32_t,  UPB_INT32_T)
+TYPE_METHODS(Bool,   bool,   bool,     bool)
+
+#ifdef UPB_TWO_32BIT_TYPES
+TYPE_METHODS(Int32,  int32,  int32_t,  UPB_INT32ALT_T)
+TYPE_METHODS(UInt32, uint32, uint32_t, UPB_UINT32ALT_T)
+#endif
+
+#ifdef UPB_TWO_64BIT_TYPES
+TYPE_METHODS(Int64,  int64,  int64_t,  UPB_INT64ALT_T)
+TYPE_METHODS(UInt64, uint64, uint64_t, UPB_UINT64ALT_T)
+#endif
+#undef TYPE_METHODS
+
+template <> struct CanonicalType<Status*> {
+  typedef Status* Type;
+};
+
+/* Type methods that are only one-per-canonical-type and not
+ * one-per-cvariant. */
+
+#define TYPE_METHODS(utype, ctype) \
+    inline bool Handlers::Set##utype##Handler(const FieldDef *f, \
+                                              const utype##Handler &h) { \
+      return SetValueHandler<ctype>(f, h); \
+    } \
+
+TYPE_METHODS(Double, double)
+TYPE_METHODS(Float,  float)
+TYPE_METHODS(UInt64, uint64_t)
+TYPE_METHODS(UInt32, uint32_t)
+TYPE_METHODS(Int64,  int64_t)
+TYPE_METHODS(Int32,  int32_t)
+TYPE_METHODS(Bool,   bool)
+#undef TYPE_METHODS
+
+template <class F> struct ReturnOf;
+
+template <class R, class P1, class P2>
+struct ReturnOf<R (*)(P1, P2)> {
+  typedef R Return;
+};
+
+template <class R, class P1, class P2, class P3>
+struct ReturnOf<R (*)(P1, P2, P3)> {
+  typedef R Return;
+};
+
+template <class R, class P1, class P2, class P3, class P4>
+struct ReturnOf<R (*)(P1, P2, P3, P4)> {
+  typedef R Return;
+};
+
+template <class R, class P1, class P2, class P3, class P4, class P5>
+struct ReturnOf<R (*)(P1, P2, P3, P4, P5)> {
+  typedef R Return;
+};
+
+template<class T> const void *UniquePtrForType() {
+  static const char ch = 0;
+  return &ch;
+}
+
+template <class T>
+template <class F>
+inline Handler<T>::Handler(F func)
+    : registered_(false),
+      cleanup_data_(func.GetData()),
+      cleanup_func_(func.GetCleanup()) {
+  upb_handlerattr_sethandlerdata(&attr_, func.GetData());
+  typedef typename ReturnOf<T>::Return Return;
+  typedef typename ConvertParams<F, T>::Func ConvertedParamsFunc;
+  typedef typename MaybeWrapReturn<ConvertedParamsFunc, Return>::Func
+      ReturnWrappedFunc;
+  handler_ = ReturnWrappedFunc().Call;
+
+  /* Set attributes based on what templates can statically tell us about the
+   * user's function. */
+
+  /* If the original function returns void, then we know that we wrapped it to
+   * always return ok. */
+  bool always_ok = is_same<typename F::FuncInfo::Return, void>::value;
+  attr_.SetAlwaysOk(always_ok);
+
+  /* Closure parameter and return type. */
+  attr_.SetClosureType(UniquePtrForType<typename F::FuncInfo::Closure>());
+
+  /* We use the closure type (from the first parameter) if the return type is
+   * void or bool, since these are the two cases we wrap to return the closure's
+   * type anyway.
+   *
+   * This is all nonsense for non START* handlers, but it doesn't matter because
+   * in that case the value will be ignored. */
+  typedef typename FirstUnlessVoidOrBool<typename F::FuncInfo::Return,
+                                         typename F::FuncInfo::Closure>::value
+      EffectiveReturn;
+  attr_.SetReturnClosureType(UniquePtrForType<EffectiveReturn>());
+}
+
+template <class T>
+inline Handler<T>::~Handler() {
+  assert(registered_);
+}
+
+inline HandlerAttributes::HandlerAttributes() { upb_handlerattr_init(this); }
+inline HandlerAttributes::~HandlerAttributes() { upb_handlerattr_uninit(this); }
+inline bool HandlerAttributes::SetHandlerData(const void *hd) {
+  return upb_handlerattr_sethandlerdata(this, hd);
+}
+inline const void* HandlerAttributes::handler_data() const {
+  return upb_handlerattr_handlerdata(this);
+}
+inline bool HandlerAttributes::SetClosureType(const void *type) {
+  return upb_handlerattr_setclosuretype(this, type);
+}
+inline const void* HandlerAttributes::closure_type() const {
+  return upb_handlerattr_closuretype(this);
+}
+inline bool HandlerAttributes::SetReturnClosureType(const void *type) {
+  return upb_handlerattr_setreturnclosuretype(this, type);
+}
+inline const void* HandlerAttributes::return_closure_type() const {
+  return upb_handlerattr_returnclosuretype(this);
+}
+inline bool HandlerAttributes::SetAlwaysOk(bool always_ok) {
+  return upb_handlerattr_setalwaysok(this, always_ok);
+}
+inline bool HandlerAttributes::always_ok() const {
+  return upb_handlerattr_alwaysok(this);
+}
+
+inline BufferHandle::BufferHandle() { upb_bufhandle_init(this); }
+inline BufferHandle::~BufferHandle() { upb_bufhandle_uninit(this); }
+inline const char* BufferHandle::buffer() const {
+  return upb_bufhandle_buf(this);
+}
+inline size_t BufferHandle::object_offset() const {
+  return upb_bufhandle_objofs(this);
+}
+inline void BufferHandle::SetBuffer(const char* buf, size_t ofs) {
+  upb_bufhandle_setbuf(this, buf, ofs);
+}
+template <class T>
+void BufferHandle::SetAttachedObject(const T* obj) {
+  upb_bufhandle_setobj(this, obj, UniquePtrForType<T>());
+}
+template <class T>
+const T* BufferHandle::GetAttachedObject() const {
+  return upb_bufhandle_objtype(this) == UniquePtrForType<T>()
+      ? static_cast<const T *>(upb_bufhandle_obj(this))
+                               : NULL;
+}
+
+inline reffed_ptr<Handlers> Handlers::New(const MessageDef *m) {
+  upb_handlers *h = upb_handlers_new(m, &h);
+  return reffed_ptr<Handlers>(h, &h);
+}
+inline reffed_ptr<const Handlers> Handlers::NewFrozen(
+    const MessageDef *m, upb_handlers_callback *callback,
+    const void *closure) {
+  const upb_handlers *h = upb_handlers_newfrozen(m, &h, callback, closure);
+  return reffed_ptr<const Handlers>(h, &h);
+}
+inline const Status* Handlers::status() {
+  return upb_handlers_status(this);
+}
+inline void Handlers::ClearError() {
+  return upb_handlers_clearerr(this);
+}
+inline bool Handlers::Freeze(Status *s) {
+  upb::Handlers* h = this;
+  return upb_handlers_freeze(&h, 1, s);
+}
+inline bool Handlers::Freeze(Handlers *const *handlers, int n, Status *s) {
+  return upb_handlers_freeze(handlers, n, s);
+}
+inline bool Handlers::Freeze(const std::vector<Handlers*>& h, Status* status) {
+  return upb_handlers_freeze((Handlers* const*)&h[0], h.size(), status);
+}
+inline const MessageDef *Handlers::message_def() const {
+  return upb_handlers_msgdef(this);
+}
+inline bool Handlers::AddCleanup(void *p, upb_handlerfree *func) {
+  return upb_handlers_addcleanup(this, p, func);
+}
+inline bool Handlers::SetStartMessageHandler(
+    const Handlers::StartMessageHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setstartmsg(this, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetEndMessageHandler(
+    const Handlers::EndMessageHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setendmsg(this, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetStartStringHandler(const FieldDef *f,
+                                            const StartStringHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setstartstr(this, f, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetEndStringHandler(const FieldDef *f,
+                                          const EndFieldHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setendstr(this, f, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetStringHandler(const FieldDef *f,
+                                       const StringHandler& handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setstring(this, f, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetStartSequenceHandler(
+    const FieldDef *f, const StartFieldHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setstartseq(this, f, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetStartSubMessageHandler(
+    const FieldDef *f, const StartFieldHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setstartsubmsg(this, f, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetEndSubMessageHandler(const FieldDef *f,
+                                              const EndFieldHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setendsubmsg(this, f, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetEndSequenceHandler(const FieldDef *f,
+                                            const EndFieldHandler &handler) {
+  assert(!handler.registered_);
+  handler.registered_ = true;
+  handler.AddCleanup(this);
+  return upb_handlers_setendseq(this, f, handler.handler_, &handler.attr_);
+}
+inline bool Handlers::SetSubHandlers(const FieldDef *f, const Handlers *sub) {
+  return upb_handlers_setsubhandlers(this, f, sub);
+}
+inline const Handlers *Handlers::GetSubHandlers(const FieldDef *f) const {
+  return upb_handlers_getsubhandlers(this, f);
+}
+inline const Handlers *Handlers::GetSubHandlers(Handlers::Selector sel) const {
+  return upb_handlers_getsubhandlers_sel(this, sel);
+}
+inline bool Handlers::GetSelector(const FieldDef *f, Handlers::Type type,
+                                  Handlers::Selector *s) {
+  return upb_handlers_getselector(f, type, s);
+}
+inline Handlers::Selector Handlers::GetEndSelector(Handlers::Selector start) {
+  return upb_handlers_getendselector(start);
+}
+inline Handlers::GenericFunction *Handlers::GetHandler(
+    Handlers::Selector selector) {
+  return upb_handlers_gethandler(this, selector);
+}
+inline const void *Handlers::GetHandlerData(Handlers::Selector selector) {
+  return upb_handlers_gethandlerdata(this, selector);
+}
+
+inline BytesHandler::BytesHandler() {
+  upb_byteshandler_init(this);
+}
+
+inline BytesHandler::~BytesHandler() {}
+
+}  /* namespace upb */
+
+#endif  /* __cplusplus */
+
+
+#undef UPB_TWO_32BIT_TYPES
+#undef UPB_TWO_64BIT_TYPES
+#undef UPB_INT32_T
+#undef UPB_UINT32_T
+#undef UPB_INT32ALT_T
+#undef UPB_UINT32ALT_T
+#undef UPB_INT64_T
+#undef UPB_UINT64_T
+#undef UPB_INT64ALT_T
+#undef UPB_UINT64ALT_T
+
+#endif  /* UPB_HANDLERS_INL_H_ */
+
+#endif  /* UPB_HANDLERS_H */
+/*
+** upb::Environment (upb_env)
+**
+** A upb::Environment provides a means for injecting malloc and an
+** error-reporting callback into encoders/decoders.  This allows them to be
+** independent of nearly all assumptions about their actual environment.
+**
+** It is also a container for allocating the encoders/decoders themselves that
+** insulates clients from knowing their actual size.  This provides ABI
+** compatibility even if the size of the objects change.  And this allows the
+** structure definitions to be in the .c files instead of the .h files, making
+** the .h files smaller and more readable.
+*/
+
+
+#ifndef UPB_ENV_H_
+#define UPB_ENV_H_
+
+#ifdef __cplusplus
+namespace upb {
+class Environment;
+class SeededAllocator;
+}
+#endif
+
+UPB_DECLARE_TYPE(upb::Environment, upb_env)
+UPB_DECLARE_TYPE(upb::SeededAllocator, upb_seededalloc)
+
+typedef void *upb_alloc_func(void *ud, void *ptr, size_t oldsize, size_t size);
+typedef void upb_cleanup_func(void *ud);
+typedef bool upb_error_func(void *ud, const upb_status *status);
+
+#ifdef __cplusplus
+
+/* An environment is *not* thread-safe. */
+class upb::Environment {
+ public:
+  Environment();
+  ~Environment();
+
+  /* Set a custom memory allocation function for the environment.  May ONLY
+   * be called before any calls to Malloc()/Realloc()/AddCleanup() below.
+   * If this is not called, the system realloc() function will be used.
+   * The given user pointer "ud" will be passed to the allocation function.
+   *
+   * The allocation function will not receive corresponding "free" calls.  it
+   * must ensure that the memory is valid for the lifetime of the Environment,
+   * but it may be reclaimed any time thereafter.  The likely usage is that
+   * "ud" points to a stateful allocator, and that the allocator frees all
+   * memory, arena-style, when it is destroyed.  In this case the allocator must
+   * outlive the Environment.  Another possibility is that the allocation
+   * function returns GC-able memory that is guaranteed to be GC-rooted for the
+   * life of the Environment. */
+  void SetAllocationFunction(upb_alloc_func* alloc, void* ud);
+
+  template<class T>
+  void SetAllocator(T* allocator) {
+    SetAllocationFunction(allocator->GetAllocationFunction(), allocator);
+  }
+
+  /* Set a custom error reporting function. */
+  void SetErrorFunction(upb_error_func* func, void* ud);
+
+  /* Set the error reporting function to simply copy the status to the given
+   * status and abort. */
+  void ReportErrorsTo(Status* status);
+
+  /* Returns true if all allocations and AddCleanup() calls have succeeded,
+   * and no errors were reported with ReportError() (except ones that recovered
+   * successfully). */
+  bool ok() const;
+
+  /* Functions for use by encoders/decoders. **********************************/
+
+  /* Reports an error to this environment's callback, returning true if
+   * the caller should try to recover. */
+  bool ReportError(const Status* status);
+
+  /* Allocate memory.  Uses the environment's allocation function.
+   *
+   * There is no need to free(). All memory will be freed automatically, but is
+   * guaranteed to outlive the Environment. */
+  void* Malloc(size_t size);
+
+  /* Reallocate memory.  Preserves "oldsize" bytes from the existing buffer
+   * Requires: oldsize <= existing_size.
+   *
+   * TODO(haberman): should we also enforce that oldsize <= size? */
+  void* Realloc(void* ptr, size_t oldsize, size_t size);
+
+  /* Add a cleanup function to run when the environment is destroyed.
+   * Returns false on out-of-memory.
+   *
+   * The first call to AddCleanup() after SetAllocationFunction() is guaranteed
+   * to return true -- this makes it possible to robustly set a cleanup handler
+   * for a custom allocation function. */
+  bool AddCleanup(upb_cleanup_func* func, void* ud);
+
+  /* Total number of bytes that have been allocated.  It is undefined what
+   * Realloc() does to this counter. */
+  size_t BytesAllocated() const;
+
+ private:
+  UPB_DISALLOW_COPY_AND_ASSIGN(Environment)
+
+#else
+struct upb_env {
+#endif  /* __cplusplus */
+
+  bool ok_;
+  size_t bytes_allocated;
+
+  /* Alloc function. */
+  upb_alloc_func *alloc;
+  void *alloc_ud;
+
+  /* Error-reporting function. */
+  upb_error_func *err;
+  void *err_ud;
+
+  /* Userdata for default alloc func. */
+  void *default_alloc_ud;
+
+  /* Cleanup entries.  Pointer to a cleanup_ent, defined in env.c */
+  void *cleanup_head;
+
+  /* For future expansion, since the size of this struct is exposed to users. */
+  void *future1;
+  void *future2;
+};
+
+UPB_BEGIN_EXTERN_C
+
+void upb_env_init(upb_env *e);
+void upb_env_uninit(upb_env *e);
+void upb_env_setallocfunc(upb_env *e, upb_alloc_func *func, void *ud);
+void upb_env_seterrorfunc(upb_env *e, upb_error_func *func, void *ud);
+void upb_env_reporterrorsto(upb_env *e, upb_status *status);
+bool upb_env_ok(const upb_env *e);
+bool upb_env_reporterror(upb_env *e, const upb_status *status);
+void *upb_env_malloc(upb_env *e, size_t size);
+void *upb_env_realloc(upb_env *e, void *ptr, size_t oldsize, size_t size);
+bool upb_env_addcleanup(upb_env *e, upb_cleanup_func *func, void *ud);
+size_t upb_env_bytesallocated(const upb_env *e);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+/* An allocator that allocates from an initial memory region (likely the stack)
+ * before falling back to another allocator. */
+class upb::SeededAllocator {
+ public:
+  SeededAllocator(void *mem, size_t len);
+  ~SeededAllocator();
+
+  /* Set a custom fallback memory allocation function for the allocator, to use
+   * once the initial region runs out.
+   *
+   * May ONLY be called before GetAllocationFunction().  If this is not
+   * called, the system realloc() will be the fallback allocator. */
+  void SetFallbackAllocator(upb_alloc_func *alloc, void *ud);
+
+  /* Gets the allocation function for this allocator. */
+  upb_alloc_func* GetAllocationFunction();
+
+ private:
+  UPB_DISALLOW_COPY_AND_ASSIGN(SeededAllocator)
+
+#else
+struct upb_seededalloc {
+#endif  /* __cplusplus */
+
+  /* Fallback alloc function.  */
+  upb_alloc_func *alloc;
+  upb_cleanup_func *alloc_cleanup;
+  void *alloc_ud;
+  bool need_cleanup;
+  bool returned_allocfunc;
+
+  /* Userdata for default alloc func. */
+  void *default_alloc_ud;
+
+  /* Pointers for the initial memory region. */
+  char *mem_base;
+  char *mem_ptr;
+  char *mem_limit;
+
+  /* For future expansion, since the size of this struct is exposed to users. */
+  void *future1;
+  void *future2;
+};
+
+UPB_BEGIN_EXTERN_C
+
+void upb_seededalloc_init(upb_seededalloc *a, void *mem, size_t len);
+void upb_seededalloc_uninit(upb_seededalloc *a);
+void upb_seededalloc_setfallbackalloc(upb_seededalloc *a, upb_alloc_func *func,
+                                      void *ud);
+upb_alloc_func *upb_seededalloc_getallocfunc(upb_seededalloc *a);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upb {
+
+inline Environment::Environment() {
+  upb_env_init(this);
+}
+inline Environment::~Environment() {
+  upb_env_uninit(this);
+}
+inline void Environment::SetAllocationFunction(upb_alloc_func *alloc,
+                                               void *ud) {
+  upb_env_setallocfunc(this, alloc, ud);
+}
+inline void Environment::SetErrorFunction(upb_error_func *func, void *ud) {
+  upb_env_seterrorfunc(this, func, ud);
+}
+inline void Environment::ReportErrorsTo(Status* status) {
+  upb_env_reporterrorsto(this, status);
+}
+inline bool Environment::ok() const {
+  return upb_env_ok(this);
+}
+inline bool Environment::ReportError(const Status* status) {
+  return upb_env_reporterror(this, status);
+}
+inline void *Environment::Malloc(size_t size) {
+  return upb_env_malloc(this, size);
+}
+inline void *Environment::Realloc(void *ptr, size_t oldsize, size_t size) {
+  return upb_env_realloc(this, ptr, oldsize, size);
+}
+inline bool Environment::AddCleanup(upb_cleanup_func *func, void *ud) {
+  return upb_env_addcleanup(this, func, ud);
+}
+inline size_t Environment::BytesAllocated() const {
+  return upb_env_bytesallocated(this);
+}
+
+inline SeededAllocator::SeededAllocator(void *mem, size_t len) {
+  upb_seededalloc_init(this, mem, len);
+}
+inline SeededAllocator::~SeededAllocator() {
+  upb_seededalloc_uninit(this);
+}
+inline void SeededAllocator::SetFallbackAllocator(upb_alloc_func *alloc,
+                                                  void *ud) {
+  upb_seededalloc_setfallbackalloc(this, alloc, ud);
+}
+inline upb_alloc_func *SeededAllocator::GetAllocationFunction() {
+  return upb_seededalloc_getallocfunc(this);
+}
+
+}  /* namespace upb */
+
+#endif  /* __cplusplus */
+
+#endif  /* UPB_ENV_H_ */
+/*
+** upb::Sink (upb_sink)
+** upb::BytesSink (upb_bytessink)
+**
+** A upb_sink is an object that binds a upb_handlers object to some runtime
+** state.  It is the object that can actually receive data via the upb_handlers
+** interface.
+**
+** Unlike upb_def and upb_handlers, upb_sink is never frozen, immutable, or
+** thread-safe.  You can create as many of them as you want, but each one may
+** only be used in a single thread at a time.
+**
+** If we compare with class-based OOP, a you can think of a upb_def as an
+** abstract base class, a upb_handlers as a concrete derived class, and a
+** upb_sink as an object (class instance).
+*/
+
+#ifndef UPB_SINK_H
+#define UPB_SINK_H
+
+
+#ifdef __cplusplus
+namespace upb {
+class BufferSource;
+class BytesSink;
+class Sink;
+}
+#endif
+
+UPB_DECLARE_TYPE(upb::BufferSource, upb_bufsrc)
+UPB_DECLARE_TYPE(upb::BytesSink, upb_bytessink)
+UPB_DECLARE_TYPE(upb::Sink, upb_sink)
+
+#ifdef __cplusplus
+
+/* A upb::Sink is an object that binds a upb::Handlers object to some runtime
+ * state.  It represents an endpoint to which data can be sent.
+ *
+ * TODO(haberman): right now all of these functions take selectors.  Should they
+ * take selectorbase instead?
+ *
+ * ie. instead of calling:
+ *   sink->StartString(FOO_FIELD_START_STRING, ...)
+ * a selector base would let you say:
+ *   sink->StartString(FOO_FIELD, ...)
+ *
+ * This would make call sites a little nicer and require emitting fewer selector
+ * definitions in .h files.
+ *
+ * But the current scheme has the benefit that you can retrieve a function
+ * pointer for any handler with handlers->GetHandler(selector), without having
+ * to have a separate GetHandler() function for each handler type.  The JIT
+ * compiler uses this.  To accommodate we'd have to expose a separate
+ * GetHandler() for every handler type.
+ *
+ * Also to ponder: selectors right now are independent of a specific Handlers
+ * instance.  In other words, they allocate a number to every possible handler
+ * that *could* be registered, without knowing anything about what handlers
+ * *are* registered.  That means that using selectors as table offsets prohibits
+ * us from compacting the handler table at Freeze() time.  If the table is very
+ * sparse, this could be wasteful.
+ *
+ * Having another selector-like thing that is specific to a Handlers instance
+ * would allow this compacting, but then it would be impossible to write code
+ * ahead-of-time that can be bound to any Handlers instance at runtime.  For
+ * example, a .proto file parser written as straight C will not know what
+ * Handlers it will be bound to, so when it calls sink->StartString() what
+ * selector will it pass?  It needs a selector like we have today, that is
+ * independent of any particular upb::Handlers.
+ *
+ * Is there a way then to allow Handlers table compaction? */
+class upb::Sink {
+ public:
+  /* Constructor with no initialization; must be Reset() before use. */
+  Sink() {}
+
+  /* Constructs a new sink for the given frozen handlers and closure.
+   *
+   * TODO: once the Handlers know the expected closure type, verify that T
+   * matches it. */
+  template <class T> Sink(const Handlers* handlers, T* closure);
+
+  /* Resets the value of the sink. */
+  template <class T> void Reset(const Handlers* handlers, T* closure);
+
+  /* Returns the top-level object that is bound to this sink.
+   *
+   * TODO: once the Handlers know the expected closure type, verify that T
+   * matches it. */
+  template <class T> T* GetObject() const;
+
+  /* Functions for pushing data into the sink.
+   *
+   * These return false if processing should stop (either due to error or just
+   * to suspend).
+   *
+   * These may not be called from within one of the same sink's handlers (in
+   * other words, handlers are not re-entrant). */
+
+  /* Should be called at the start and end of every message; both the top-level
+   * message and submessages.  This means that submessages should use the
+   * following sequence:
+   *   sink->StartSubMessage(startsubmsg_selector);
+   *   sink->StartMessage();
+   *   // ...
+   *   sink->EndMessage(&status);
+   *   sink->EndSubMessage(endsubmsg_selector); */
+  bool StartMessage();
+  bool EndMessage(Status* status);
+
+  /* Putting of individual values.  These work for both repeated and
+   * non-repeated fields, but for repeated fields you must wrap them in
+   * calls to StartSequence()/EndSequence(). */
+  bool PutInt32(Handlers::Selector s, int32_t val);
+  bool PutInt64(Handlers::Selector s, int64_t val);
+  bool PutUInt32(Handlers::Selector s, uint32_t val);
+  bool PutUInt64(Handlers::Selector s, uint64_t val);
+  bool PutFloat(Handlers::Selector s, float val);
+  bool PutDouble(Handlers::Selector s, double val);
+  bool PutBool(Handlers::Selector s, bool val);
+
+  /* Putting of string/bytes values.  Each string can consist of zero or more
+   * non-contiguous buffers of data.
+   *
+   * For StartString(), the function will write a sink for the string to "sub."
+   * The sub-sink must be used for any/all PutStringBuffer() calls. */
+  bool StartString(Handlers::Selector s, size_t size_hint, Sink* sub);
+  size_t PutStringBuffer(Handlers::Selector s, const char *buf, size_t len,
+                         const BufferHandle *handle);
+  bool EndString(Handlers::Selector s);
+
+  /* For submessage fields.
+   *
+   * For StartSubMessage(), the function will write a sink for the string to
+   * "sub." The sub-sink must be used for any/all handlers called within the
+   * submessage. */
+  bool StartSubMessage(Handlers::Selector s, Sink* sub);
+  bool EndSubMessage(Handlers::Selector s);
+
+  /* For repeated fields of any type, the sequence of values must be wrapped in
+   * these calls.
+   *
+   * For StartSequence(), the function will write a sink for the string to
+   * "sub." The sub-sink must be used for any/all handlers called within the
+   * sequence. */
+  bool StartSequence(Handlers::Selector s, Sink* sub);
+  bool EndSequence(Handlers::Selector s);
+
+  /* Copy and assign specifically allowed.
+   * We don't even bother making these members private because so many
+   * functions need them and this is mainly just a dumb data container anyway.
+   */
+#else
+struct upb_sink {
+#endif
+  const upb_handlers *handlers;
+  void *closure;
+};
+
+#ifdef __cplusplus
+class upb::BytesSink {
+ public:
+  BytesSink() {}
+
+  /* Constructs a new sink for the given frozen handlers and closure.
+   *
+   * TODO(haberman): once the Handlers know the expected closure type, verify
+   * that T matches it. */
+  template <class T> BytesSink(const BytesHandler* handler, T* closure);
+
+  /* Resets the value of the sink. */
+  template <class T> void Reset(const BytesHandler* handler, T* closure);
+
+  bool Start(size_t size_hint, void **subc);
+  size_t PutBuffer(void *subc, const char *buf, size_t len,
+                   const BufferHandle *handle);
+  bool End();
+#else
+struct upb_bytessink {
+#endif
+  const upb_byteshandler *handler;
+  void *closure;
+};
+
+#ifdef __cplusplus
+
+/* A class for pushing a flat buffer of data to a BytesSink.
+ * You can construct an instance of this to get a resumable source,
+ * or just call the static PutBuffer() to do a non-resumable push all in one
+ * go. */
+class upb::BufferSource {
+ public:
+  BufferSource();
+  BufferSource(const char* buf, size_t len, BytesSink* sink);
+
+  /* Returns true if the entire buffer was pushed successfully.  Otherwise the
+   * next call to PutNext() will resume where the previous one left off.
+   * TODO(haberman): implement this. */
+  bool PutNext();
+
+  /* A static version; with this version is it not possible to resume in the
+   * case of failure or a partially-consumed buffer. */
+  static bool PutBuffer(const char* buf, size_t len, BytesSink* sink);
+
+  template <class T> static bool PutBuffer(const T& str, BytesSink* sink) {
+    return PutBuffer(str.c_str(), str.size(), sink);
+  }
+#else
+struct upb_bufsrc {
+  char dummy;
+#endif
+};
+
+UPB_BEGIN_EXTERN_C
+
+/* Inline definitions. */
+
+UPB_INLINE void upb_bytessink_reset(upb_bytessink *s, const upb_byteshandler *h,
+                                    void *closure) {
+  s->handler = h;
+  s->closure = closure;
+}
+
+UPB_INLINE bool upb_bytessink_start(upb_bytessink *s, size_t size_hint,
+                                    void **subc) {
+  typedef upb_startstr_handlerfunc func;
+  func *start;
+  *subc = s->closure;
+  if (!s->handler) return true;
+  start = (func *)s->handler->table[UPB_STARTSTR_SELECTOR].func;
+
+  if (!start) return true;
+  *subc = start(s->closure, upb_handlerattr_handlerdata(
+                                &s->handler->table[UPB_STARTSTR_SELECTOR].attr),
+                size_hint);
+  return *subc != NULL;
+}
+
+UPB_INLINE size_t upb_bytessink_putbuf(upb_bytessink *s, void *subc,
+                                       const char *buf, size_t size,
+                                       const upb_bufhandle* handle) {
+  typedef upb_string_handlerfunc func;
+  func *putbuf;
+  if (!s->handler) return true;
+  putbuf = (func *)s->handler->table[UPB_STRING_SELECTOR].func;
+
+  if (!putbuf) return true;
+  return putbuf(subc, upb_handlerattr_handlerdata(
+                          &s->handler->table[UPB_STRING_SELECTOR].attr),
+                buf, size, handle);
+}
+
+UPB_INLINE bool upb_bytessink_end(upb_bytessink *s) {
+  typedef upb_endfield_handlerfunc func;
+  func *end;
+  if (!s->handler) return true;
+  end = (func *)s->handler->table[UPB_ENDSTR_SELECTOR].func;
+
+  if (!end) return true;
+  return end(s->closure,
+             upb_handlerattr_handlerdata(
+                 &s->handler->table[UPB_ENDSTR_SELECTOR].attr));
+}
+
+UPB_INLINE bool upb_bufsrc_putbuf(const char *buf, size_t len,
+                                  upb_bytessink *sink) {
+  void *subc;
+  bool ret;
+  upb_bufhandle handle;
+  upb_bufhandle_init(&handle);
+  upb_bufhandle_setbuf(&handle, buf, 0);
+  ret = upb_bytessink_start(sink, len, &subc);
+  if (ret && len != 0) {
+    ret = (upb_bytessink_putbuf(sink, subc, buf, len, &handle) >= len);
+  }
+  if (ret) {
+    ret = upb_bytessink_end(sink);
+  }
+  upb_bufhandle_uninit(&handle);
+  return ret;
+}
+
+#define PUTVAL(type, ctype)                                                    \
+  UPB_INLINE bool upb_sink_put##type(upb_sink *s, upb_selector_t sel,          \
+                                     ctype val) {                              \
+    typedef upb_##type##_handlerfunc functype;                                 \
+    functype *func;                                                            \
+    const void *hd;                                                            \
+    if (!s->handlers) return true;                                             \
+    func = (functype *)upb_handlers_gethandler(s->handlers, sel);              \
+    if (!func) return true;                                                    \
+    hd = upb_handlers_gethandlerdata(s->handlers, sel);                        \
+    return func(s->closure, hd, val);                                          \
+  }
+
+PUTVAL(int32,  int32_t)
+PUTVAL(int64,  int64_t)
+PUTVAL(uint32, uint32_t)
+PUTVAL(uint64, uint64_t)
+PUTVAL(float,  float)
+PUTVAL(double, double)
+PUTVAL(bool,   bool)
+#undef PUTVAL
+
+UPB_INLINE void upb_sink_reset(upb_sink *s, const upb_handlers *h, void *c) {
+  s->handlers = h;
+  s->closure = c;
+}
+
+UPB_INLINE size_t upb_sink_putstring(upb_sink *s, upb_selector_t sel,
+                                     const char *buf, size_t n,
+                                     const upb_bufhandle *handle) {
+  typedef upb_string_handlerfunc func;
+  func *handler;
+  const void *hd;
+  if (!s->handlers) return n;
+  handler = (func *)upb_handlers_gethandler(s->handlers, sel);
+
+  if (!handler) return n;
+  hd = upb_handlers_gethandlerdata(s->handlers, sel);
+  return handler(s->closure, hd, buf, n, handle);
+}
+
+UPB_INLINE bool upb_sink_startmsg(upb_sink *s) {
+  typedef upb_startmsg_handlerfunc func;
+  func *startmsg;
+  const void *hd;
+  if (!s->handlers) return true;
+  startmsg = (func*)upb_handlers_gethandler(s->handlers, UPB_STARTMSG_SELECTOR);
+
+  if (!startmsg) return true;
+  hd = upb_handlers_gethandlerdata(s->handlers, UPB_STARTMSG_SELECTOR);
+  return startmsg(s->closure, hd);
+}
+
+UPB_INLINE bool upb_sink_endmsg(upb_sink *s, upb_status *status) {
+  typedef upb_endmsg_handlerfunc func;
+  func *endmsg;
+  const void *hd;
+  if (!s->handlers) return true;
+  endmsg = (func *)upb_handlers_gethandler(s->handlers, UPB_ENDMSG_SELECTOR);
+
+  if (!endmsg) return true;
+  hd = upb_handlers_gethandlerdata(s->handlers, UPB_ENDMSG_SELECTOR);
+  return endmsg(s->closure, hd, status);
+}
+
+UPB_INLINE bool upb_sink_startseq(upb_sink *s, upb_selector_t sel,
+                                  upb_sink *sub) {
+  typedef upb_startfield_handlerfunc func;
+  func *startseq;
+  const void *hd;
+  sub->closure = s->closure;
+  sub->handlers = s->handlers;
+  if (!s->handlers) return true;
+  startseq = (func*)upb_handlers_gethandler(s->handlers, sel);
+
+  if (!startseq) return true;
+  hd = upb_handlers_gethandlerdata(s->handlers, sel);
+  sub->closure = startseq(s->closure, hd);
+  return sub->closure ? true : false;
+}
+
+UPB_INLINE bool upb_sink_endseq(upb_sink *s, upb_selector_t sel) {
+  typedef upb_endfield_handlerfunc func;
+  func *endseq;
+  const void *hd;
+  if (!s->handlers) return true;
+  endseq = (func*)upb_handlers_gethandler(s->handlers, sel);
+
+  if (!endseq) return true;
+  hd = upb_handlers_gethandlerdata(s->handlers, sel);
+  return endseq(s->closure, hd);
+}
+
+UPB_INLINE bool upb_sink_startstr(upb_sink *s, upb_selector_t sel,
+                                  size_t size_hint, upb_sink *sub) {
+  typedef upb_startstr_handlerfunc func;
+  func *startstr;
+  const void *hd;
+  sub->closure = s->closure;
+  sub->handlers = s->handlers;
+  if (!s->handlers) return true;
+  startstr = (func*)upb_handlers_gethandler(s->handlers, sel);
+
+  if (!startstr) return true;
+  hd = upb_handlers_gethandlerdata(s->handlers, sel);
+  sub->closure = startstr(s->closure, hd, size_hint);
+  return sub->closure ? true : false;
+}
+
+UPB_INLINE bool upb_sink_endstr(upb_sink *s, upb_selector_t sel) {
+  typedef upb_endfield_handlerfunc func;
+  func *endstr;
+  const void *hd;
+  if (!s->handlers) return true;
+  endstr = (func*)upb_handlers_gethandler(s->handlers, sel);
+
+  if (!endstr) return true;
+  hd = upb_handlers_gethandlerdata(s->handlers, sel);
+  return endstr(s->closure, hd);
+}
+
+UPB_INLINE bool upb_sink_startsubmsg(upb_sink *s, upb_selector_t sel,
+                                     upb_sink *sub) {
+  typedef upb_startfield_handlerfunc func;
+  func *startsubmsg;
+  const void *hd;
+  sub->closure = s->closure;
+  if (!s->handlers) {
+    sub->handlers = NULL;
+    return true;
+  }
+  sub->handlers = upb_handlers_getsubhandlers_sel(s->handlers, sel);
+  startsubmsg = (func*)upb_handlers_gethandler(s->handlers, sel);
+
+  if (!startsubmsg) return true;
+  hd = upb_handlers_gethandlerdata(s->handlers, sel);
+  sub->closure = startsubmsg(s->closure, hd);
+  return sub->closure ? true : false;
+}
+
+UPB_INLINE bool upb_sink_endsubmsg(upb_sink *s, upb_selector_t sel) {
+  typedef upb_endfield_handlerfunc func;
+  func *endsubmsg;
+  const void *hd;
+  if (!s->handlers) return true;
+  endsubmsg = (func*)upb_handlers_gethandler(s->handlers, sel);
+
+  if (!endsubmsg) return s->closure;
+  hd = upb_handlers_gethandlerdata(s->handlers, sel);
+  return endsubmsg(s->closure, hd);
+}
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upb {
+
+template <class T> Sink::Sink(const Handlers* handlers, T* closure) {
+  upb_sink_reset(this, handlers, closure);
+}
+template <class T>
+inline void Sink::Reset(const Handlers* handlers, T* closure) {
+  upb_sink_reset(this, handlers, closure);
+}
+inline bool Sink::StartMessage() {
+  return upb_sink_startmsg(this);
+}
+inline bool Sink::EndMessage(Status* status) {
+  return upb_sink_endmsg(this, status);
+}
+inline bool Sink::PutInt32(Handlers::Selector sel, int32_t val) {
+  return upb_sink_putint32(this, sel, val);
+}
+inline bool Sink::PutInt64(Handlers::Selector sel, int64_t val) {
+  return upb_sink_putint64(this, sel, val);
+}
+inline bool Sink::PutUInt32(Handlers::Selector sel, uint32_t val) {
+  return upb_sink_putuint32(this, sel, val);
+}
+inline bool Sink::PutUInt64(Handlers::Selector sel, uint64_t val) {
+  return upb_sink_putuint64(this, sel, val);
+}
+inline bool Sink::PutFloat(Handlers::Selector sel, float val) {
+  return upb_sink_putfloat(this, sel, val);
+}
+inline bool Sink::PutDouble(Handlers::Selector sel, double val) {
+  return upb_sink_putdouble(this, sel, val);
+}
+inline bool Sink::PutBool(Handlers::Selector sel, bool val) {
+  return upb_sink_putbool(this, sel, val);
+}
+inline bool Sink::StartString(Handlers::Selector sel, size_t size_hint,
+                              Sink *sub) {
+  return upb_sink_startstr(this, sel, size_hint, sub);
+}
+inline size_t Sink::PutStringBuffer(Handlers::Selector sel, const char *buf,
+                                    size_t len, const BufferHandle* handle) {
+  return upb_sink_putstring(this, sel, buf, len, handle);
+}
+inline bool Sink::EndString(Handlers::Selector sel) {
+  return upb_sink_endstr(this, sel);
+}
+inline bool Sink::StartSubMessage(Handlers::Selector sel, Sink* sub) {
+  return upb_sink_startsubmsg(this, sel, sub);
+}
+inline bool Sink::EndSubMessage(Handlers::Selector sel) {
+  return upb_sink_endsubmsg(this, sel);
+}
+inline bool Sink::StartSequence(Handlers::Selector sel, Sink* sub) {
+  return upb_sink_startseq(this, sel, sub);
+}
+inline bool Sink::EndSequence(Handlers::Selector sel) {
+  return upb_sink_endseq(this, sel);
+}
+
+template <class T>
+BytesSink::BytesSink(const BytesHandler* handler, T* closure) {
+  Reset(handler, closure);
+}
+
+template <class T>
+void BytesSink::Reset(const BytesHandler *handler, T *closure) {
+  upb_bytessink_reset(this, handler, closure);
+}
+inline bool BytesSink::Start(size_t size_hint, void **subc) {
+  return upb_bytessink_start(this, size_hint, subc);
+}
+inline size_t BytesSink::PutBuffer(void *subc, const char *buf, size_t len,
+                                   const BufferHandle *handle) {
+  return upb_bytessink_putbuf(this, subc, buf, len, handle);
+}
+inline bool BytesSink::End() {
+  return upb_bytessink_end(this);
+}
+
+inline bool BufferSource::PutBuffer(const char *buf, size_t len,
+                                    BytesSink *sink) {
+  return upb_bufsrc_putbuf(buf, len, sink);
+}
+
+}  /* namespace upb */
+#endif
+
+#endif
+/*
+** For handlers that do very tiny, very simple operations, the function call
+** overhead of calling a handler can be significant.  This file allows the
+** user to define handlers that do something very simple like store the value
+** to memory and/or set a hasbit.  JIT compilers can then special-case these
+** handlers and emit specialized code for them instead of actually calling the
+** handler.
+**
+** The functionality is very simple/limited right now but may expand to be able
+** to call another function.
+*/
+
+#ifndef UPB_SHIM_H
+#define UPB_SHIM_H
+
+
+typedef struct {
+  size_t offset;
+  int32_t hasbit;
+} upb_shim_data;
+
+#ifdef __cplusplus
+
+namespace upb {
+
+struct Shim {
+  typedef upb_shim_data Data;
+
+  /* Sets a handler for the given field that writes the value to the given
+   * offset and, if hasbit >= 0, sets a bit at the given bit offset.  Returns
+   * true if the handler was set successfully. */
+  static bool Set(Handlers *h, const FieldDef *f, size_t ofs, int32_t hasbit);
+
+  /* If this handler is a shim, returns the corresponding upb::Shim::Data and
+   * stores the type in "type".  Otherwise returns NULL. */
+  static const Data* GetData(const Handlers* h, Handlers::Selector s,
+                             FieldDef::Type* type);
+};
+
+}  /* namespace upb */
+
+#endif
+
+UPB_BEGIN_EXTERN_C
+
+/* C API. */
+bool upb_shim_set(upb_handlers *h, const upb_fielddef *f, size_t offset,
+                  int32_t hasbit);
+const upb_shim_data *upb_shim_getdata(const upb_handlers *h, upb_selector_t s,
+                                      upb_fieldtype_t *type);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+/* C++ Wrappers. */
+namespace upb {
+inline bool Shim::Set(Handlers* h, const FieldDef* f, size_t ofs,
+                      int32_t hasbit) {
+  return upb_shim_set(h, f, ofs, hasbit);
+}
+inline const Shim::Data* Shim::GetData(const Handlers* h, Handlers::Selector s,
+                                       FieldDef::Type* type) {
+  return upb_shim_getdata(h, s, type);
+}
+}  /* namespace upb */
+#endif
+
+#endif  /* UPB_SHIM_H */
+/*
+** upb::SymbolTable (upb_symtab)
+**
+** A symtab (symbol table) stores a name->def map of upb_defs.  Clients could
+** always create such tables themselves, but upb_symtab has logic for resolving
+** symbolic references, and in particular, for keeping a whole set of consistent
+** defs when replacing some subset of those defs.  This logic is nontrivial.
+**
+** This is a mixed C/C++ interface that offers a full API to both languages.
+** See the top-level README for more information.
+*/
+
+#ifndef UPB_SYMTAB_H_
+#define UPB_SYMTAB_H_
+
+
+#ifdef __cplusplus
+#include <vector>
+namespace upb { class SymbolTable; }
+#endif
+
+UPB_DECLARE_DERIVED_TYPE(upb::SymbolTable, upb::RefCounted,
+                         upb_symtab, upb_refcounted)
+
+typedef struct {
+ UPB_PRIVATE_FOR_CPP
+  upb_strtable_iter iter;
+  upb_deftype_t type;
+} upb_symtab_iter;
+
+#ifdef __cplusplus
+
+/* Non-const methods in upb::SymbolTable are NOT thread-safe. */
+class upb::SymbolTable {
+ public:
+  /* Returns a new symbol table with a single ref owned by "owner."
+   * Returns NULL if memory allocation failed. */
+  static reffed_ptr<SymbolTable> New();
+
+  /* Include RefCounted base methods. */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  /* For all lookup functions, the returned pointer is not owned by the
+   * caller; it may be invalidated by any non-const call or unref of the
+   * SymbolTable!  To protect against this, take a ref if desired. */
+
+  /* Freezes the symbol table: prevents further modification of it.
+   * After the Freeze() operation is successful, the SymbolTable must only be
+   * accessed via a const pointer.
+   *
+   * Unlike with upb::MessageDef/upb::EnumDef/etc, freezing a SymbolTable is not
+   * a necessary step in using a SymbolTable.  If you have no need for it to be
+   * immutable, there is no need to freeze it ever.  However sometimes it is
+   * useful, and SymbolTables that are statically compiled into the binary are
+   * always frozen by nature. */
+  void Freeze();
+
+  /* Resolves the given symbol using the rules described in descriptor.proto,
+   * namely:
+   *
+   *    If the name starts with a '.', it is fully-qualified.  Otherwise,
+   *    C++-like scoping rules are used to find the type (i.e. first the nested
+   *    types within this message are searched, then within the parent, on up
+   *    to the root namespace).
+   *
+   * If not found, returns NULL. */
+  const Def* Resolve(const char* base, const char* sym) const;
+
+  /* Finds an entry in the symbol table with this exact name.  If not found,
+   * returns NULL. */
+  const Def* Lookup(const char *sym) const;
+  const MessageDef* LookupMessage(const char *sym) const;
+  const EnumDef* LookupEnum(const char *sym) const;
+
+  /* TODO: introduce a C++ iterator, but make it nice and templated so that if
+   * you ask for an iterator of MessageDef the iterated elements are strongly
+   * typed as MessageDef*. */
+
+  /* Adds the given mutable defs to the symtab, resolving all symbols
+   * (including enum default values) and finalizing the defs.  Only one def per
+   * name may be in the list, but defs can replace existing defs in the symtab.
+   * All defs must have a name -- anonymous defs are not allowed.  Anonymous
+   * defs can still be frozen by calling upb_def_freeze() directly.
+   *
+   * Any existing defs that can reach defs that are being replaced will
+   * themselves be replaced also, so that the resulting set of defs is fully
+   * consistent.
+   *
+   * This logic implemented in this method is a convenience; ultimately it
+   * calls some combination of upb_fielddef_setsubdef(), upb_def_dup(), and
+   * upb_freeze(), any of which the client could call themself.  However, since
+   * the logic for doing so is nontrivial, we provide it here.
+   *
+   * The entire operation either succeeds or fails.  If the operation fails,
+   * the symtab is unchanged, false is returned, and status indicates the
+   * error.  The caller passes a ref on all defs to the symtab (even if the
+   * operation fails).
+   *
+   * TODO(haberman): currently failure will leave the symtab unchanged, but may
+   * leave the defs themselves partially resolved.  Does this matter?  If so we
+   * could do a prepass that ensures that all symbols are resolvable and bail
+   * if not, so we don't mutate anything until we know the operation will
+   * succeed.
+   *
+   * TODO(haberman): since the defs must be mutable, refining a frozen def
+   * requires making mutable copies of the entire tree.  This is wasteful if
+   * only a few messages are changing.  We may want to add a way of adding a
+   * tree of frozen defs to the symtab (perhaps an alternate constructor where
+   * you pass the root of the tree?) */
+  bool Add(Def*const* defs, int n, void* ref_donor, upb_status* status);
+
+  bool Add(const std::vector<Def*>& defs, void *owner, Status* status) {
+    return Add((Def*const*)&defs[0], defs.size(), owner, status);
+  }
+
+ private:
+  UPB_DISALLOW_POD_OPS(SymbolTable, upb::SymbolTable)
+};
+
+#endif  /* __cplusplus */
+
+UPB_BEGIN_EXTERN_C
+
+/* Native C API. */
+
+/* Include refcounted methods like upb_symtab_ref(). */
+UPB_REFCOUNTED_CMETHODS(upb_symtab, upb_symtab_upcast)
+
+upb_symtab *upb_symtab_new(const void *owner);
+void upb_symtab_freeze(upb_symtab *s);
+const upb_def *upb_symtab_resolve(const upb_symtab *s, const char *base,
+                                  const char *sym);
+const upb_def *upb_symtab_lookup(const upb_symtab *s, const char *sym);
+const upb_msgdef *upb_symtab_lookupmsg(const upb_symtab *s, const char *sym);
+const upb_enumdef *upb_symtab_lookupenum(const upb_symtab *s, const char *sym);
+bool upb_symtab_add(upb_symtab *s, upb_def *const*defs, int n, void *ref_donor,
+                    upb_status *status);
+
+/* upb_symtab_iter i;
+ * for(upb_symtab_begin(&i, s, type); !upb_symtab_done(&i);
+ *     upb_symtab_next(&i)) {
+ *   const upb_def *def = upb_symtab_iter_def(&i);
+ *    // ...
+ * }
+ *
+ * For C we don't have separate iterators for const and non-const.
+ * It is the caller's responsibility to cast the upb_fielddef* to
+ * const if the upb_msgdef* is const. */
+void upb_symtab_begin(upb_symtab_iter *iter, const upb_symtab *s,
+                      upb_deftype_t type);
+void upb_symtab_next(upb_symtab_iter *iter);
+bool upb_symtab_done(const upb_symtab_iter *iter);
+const upb_def *upb_symtab_iter_def(const upb_symtab_iter *iter);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+/* C++ inline wrappers. */
+namespace upb {
+inline reffed_ptr<SymbolTable> SymbolTable::New() {
+  upb_symtab *s = upb_symtab_new(&s);
+  return reffed_ptr<SymbolTable>(s, &s);
+}
+
+inline void SymbolTable::Freeze() {
+  return upb_symtab_freeze(this);
+}
+inline const Def *SymbolTable::Resolve(const char *base,
+                                       const char *sym) const {
+  return upb_symtab_resolve(this, base, sym);
+}
+inline const Def* SymbolTable::Lookup(const char *sym) const {
+  return upb_symtab_lookup(this, sym);
+}
+inline const MessageDef *SymbolTable::LookupMessage(const char *sym) const {
+  return upb_symtab_lookupmsg(this, sym);
+}
+inline bool SymbolTable::Add(
+    Def*const* defs, int n, void* ref_donor, upb_status* status) {
+  return upb_symtab_add(this, (upb_def*const*)defs, n, ref_donor, status);
+}
+}  /* namespace upb */
+#endif
+
+#endif  /* UPB_SYMTAB_H_ */
+/*
+** upb::descriptor::Reader (upb_descreader)
+**
+** Provides a way of building upb::Defs from data in descriptor.proto format.
+*/
+
+#ifndef UPB_DESCRIPTOR_H
+#define UPB_DESCRIPTOR_H
+
+
+#ifdef __cplusplus
+namespace upb {
+namespace descriptor {
+class Reader;
+}  /* namespace descriptor */
+}  /* namespace upb */
+#endif
+
+UPB_DECLARE_TYPE(upb::descriptor::Reader, upb_descreader)
+
+#ifdef __cplusplus
+
+/* Class that receives descriptor data according to the descriptor.proto schema
+ * and use it to build upb::Defs corresponding to that schema. */
+class upb::descriptor::Reader {
+ public:
+  /* These handlers must have come from NewHandlers() and must outlive the
+   * Reader.
+   *
+   * TODO: generate the handlers statically (like we do with the
+   * descriptor.proto defs) so that there is no need to pass this parameter (or
+   * to build/memory-manage the handlers at runtime at all).  Unfortunately this
+   * is a bit tricky to implement for Handlers, but necessary to simplify this
+   * interface. */
+  static Reader* Create(Environment* env, const Handlers* handlers);
+
+  /* The reader's input; this is where descriptor.proto data should be sent. */
+  Sink* input();
+
+  /* Returns an array of all defs that have been parsed, and transfers ownership
+   * of them to "owner".  The number of defs is stored in *n.  Ownership of the
+   * returned array is retained and is invalidated by any other call into
+   * Reader.
+   *
+   * These defs are not frozen or resolved; they are ready to be added to a
+   * symtab. */
+  upb::Def** GetDefs(void* owner, int* n);
+
+  /* Builds and returns handlers for the reader, owned by "owner." */
+  static Handlers* NewHandlers(const void* owner);
+
+ private:
+  UPB_DISALLOW_POD_OPS(Reader, upb::descriptor::Reader)
+};
+
+#endif
+
+UPB_BEGIN_EXTERN_C
+
+/* C API. */
+upb_descreader *upb_descreader_create(upb_env *e, const upb_handlers *h);
+upb_sink *upb_descreader_input(upb_descreader *r);
+upb_def **upb_descreader_getdefs(upb_descreader *r, void *owner, int *n);
+const upb_handlers *upb_descreader_newhandlers(const void *owner);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+/* C++ implementation details. ************************************************/
+namespace upb {
+namespace descriptor {
+inline Reader* Reader::Create(Environment* e, const Handlers *h) {
+  return upb_descreader_create(e, h);
+}
+inline Sink* Reader::input() { return upb_descreader_input(this); }
+inline upb::Def** Reader::GetDefs(void* owner, int* n) {
+  return upb_descreader_getdefs(this, owner, n);
+}
+}  /* namespace descriptor */
+}  /* namespace upb */
+#endif
+
+#endif  /* UPB_DESCRIPTOR_H */
+/* This file contains accessors for a set of compiled-in defs.
+ * Note that unlike Google's protobuf, it does *not* define
+ * generated classes or any other kind of data structure for
+ * actually storing protobufs.  It only contains *defs* which
+ * let you reflect over a protobuf *schema*.
+ */
+/* This file was generated by upbc (the upb compiler).
+ * Do not edit -- your changes will be discarded when the file is
+ * regenerated. */
+
+#ifndef GOOGLE_PROTOBUF_DESCRIPTOR_UPB_H_
+#define GOOGLE_PROTOBUF_DESCRIPTOR_UPB_H_
+
+
+#ifdef __cplusplus
+UPB_BEGIN_EXTERN_C
+#endif
+
+/* Enums */
+
+typedef enum {
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_OPTIONAL = 1,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_REQUIRED = 2,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_REPEATED = 3
+} google_protobuf_FieldDescriptorProto_Label;
+
+typedef enum {
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_DOUBLE = 1,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_FLOAT = 2,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_INT64 = 3,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_UINT64 = 4,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_INT32 = 5,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_FIXED64 = 6,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_FIXED32 = 7,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_BOOL = 8,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_STRING = 9,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_GROUP = 10,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_MESSAGE = 11,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_BYTES = 12,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_UINT32 = 13,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_ENUM = 14,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SFIXED32 = 15,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SFIXED64 = 16,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SINT32 = 17,
+  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SINT64 = 18
+} google_protobuf_FieldDescriptorProto_Type;
+
+typedef enum {
+  GOOGLE_PROTOBUF_FIELDOPTIONS_STRING = 0,
+  GOOGLE_PROTOBUF_FIELDOPTIONS_CORD = 1,
+  GOOGLE_PROTOBUF_FIELDOPTIONS_STRING_PIECE = 2
+} google_protobuf_FieldOptions_CType;
+
+typedef enum {
+  GOOGLE_PROTOBUF_FILEOPTIONS_SPEED = 1,
+  GOOGLE_PROTOBUF_FILEOPTIONS_CODE_SIZE = 2,
+  GOOGLE_PROTOBUF_FILEOPTIONS_LITE_RUNTIME = 3
+} google_protobuf_FileOptions_OptimizeMode;
+
+/* Selectors */
+
+/* google.protobuf.DescriptorProto */
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_STARTSUBMSG 3
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_STARTSUBMSG 4
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_STARTSUBMSG 5
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_STARTSUBMSG 6
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_OPTIONS_STARTSUBMSG 7
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_STARTSEQ 8
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_ENDSEQ 9
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_ENDSUBMSG 10
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_STARTSEQ 11
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_ENDSEQ 12
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_ENDSUBMSG 13
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_STARTSEQ 14
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_ENDSEQ 15
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_ENDSUBMSG 16
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_STARTSEQ 17
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_ENDSEQ 18
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_ENDSUBMSG 19
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_STARTSEQ 20
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_ENDSEQ 21
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_ENDSUBMSG 22
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_OPTIONS_ENDSUBMSG 23
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NAME_STRING 24
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NAME_STARTSTR 25
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NAME_ENDSTR 26
+
+/* google.protobuf.DescriptorProto.ExtensionRange */
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSIONRANGE_START_INT32 2
+#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSIONRANGE_END_INT32 3
+
+/* google.protobuf.EnumDescriptorProto */
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 3
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_STARTSEQ 4
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_ENDSEQ 5
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_ENDSUBMSG 6
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 7
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_NAME_STRING 8
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_NAME_STARTSTR 9
+#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_NAME_ENDSTR 10
+
+/* google.protobuf.EnumOptions */
+#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5
+#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_ALLOW_ALIAS_BOOL 6
+
+/* google.protobuf.EnumValueDescriptorProto */
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 3
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NAME_STRING 4
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NAME_STARTSTR 5
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NAME_ENDSTR 6
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NUMBER_INT32 7
+
+/* google.protobuf.EnumValueOptions */
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5
+
+/* google.protobuf.FieldDescriptorProto */
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 3
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NAME_STRING 4
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NAME_STARTSTR 5
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NAME_ENDSTR 6
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_EXTENDEE_STRING 7
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_EXTENDEE_STARTSTR 8
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_EXTENDEE_ENDSTR 9
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NUMBER_INT32 10
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_INT32 11
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_INT32 12
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_NAME_STRING 13
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_NAME_STARTSTR 14
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_NAME_ENDSTR 15
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_DEFAULT_VALUE_STRING 16
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_DEFAULT_VALUE_STARTSTR 17
+#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_DEFAULT_VALUE_ENDSTR 18
+
+/* google.protobuf.FieldOptions */
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_CTYPE_INT32 6
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_PACKED_BOOL 7
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_DEPRECATED_BOOL 8
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_LAZY_BOOL 9
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_EXPERIMENTAL_MAP_KEY_STRING 10
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_EXPERIMENTAL_MAP_KEY_STARTSTR 11
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_EXPERIMENTAL_MAP_KEY_ENDSTR 12
+#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_WEAK_BOOL 13
+
+/* google.protobuf.FileDescriptorProto */
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_STARTSUBMSG 3
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_STARTSUBMSG 4
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_STARTSUBMSG 5
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 6
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SOURCE_CODE_INFO_STARTSUBMSG 7
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_STARTSEQ 8
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_ENDSEQ 9
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_ENDSUBMSG 10
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_STARTSEQ 11
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_ENDSEQ 12
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_ENDSUBMSG 13
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_STARTSEQ 14
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_ENDSEQ 15
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_ENDSUBMSG 16
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_STARTSEQ 17
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_ENDSEQ 18
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_ENDSUBMSG 19
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 20
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SOURCE_CODE_INFO_ENDSUBMSG 21
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_NAME_STRING 22
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_NAME_STARTSTR 23
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_NAME_ENDSTR 24
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PACKAGE_STRING 25
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PACKAGE_STARTSTR 26
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PACKAGE_ENDSTR 27
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_STARTSEQ 28
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_ENDSEQ 29
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_STRING 30
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_STARTSTR 31
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_ENDSTR 32
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PUBLIC_DEPENDENCY_STARTSEQ 33
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PUBLIC_DEPENDENCY_ENDSEQ 34
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PUBLIC_DEPENDENCY_INT32 35
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_WEAK_DEPENDENCY_STARTSEQ 36
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_WEAK_DEPENDENCY_ENDSEQ 37
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_WEAK_DEPENDENCY_INT32 38
+
+/* google.protobuf.FileDescriptorSet */
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_ENDSUBMSG 5
+
+/* google.protobuf.FileOptions */
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_PACKAGE_STRING 6
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_PACKAGE_STARTSTR 7
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_PACKAGE_ENDSTR 8
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_OUTER_CLASSNAME_STRING 9
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_OUTER_CLASSNAME_STARTSTR 10
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_OUTER_CLASSNAME_ENDSTR 11
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_OPTIMIZE_FOR_INT32 12
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_MULTIPLE_FILES_BOOL 13
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_GO_PACKAGE_STRING 14
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_GO_PACKAGE_STARTSTR 15
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_GO_PACKAGE_ENDSTR 16
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_CC_GENERIC_SERVICES_BOOL 17
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_GENERIC_SERVICES_BOOL 18
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_PY_GENERIC_SERVICES_BOOL 19
+#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_GENERATE_EQUALS_AND_HASH_BOOL 20
+
+/* google.protobuf.MessageOptions */
+#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5
+#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_MESSAGE_SET_WIRE_FORMAT_BOOL 6
+#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_NO_STANDARD_DESCRIPTOR_ACCESSOR_BOOL 7
+
+/* google.protobuf.MethodDescriptorProto */
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 3
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_NAME_STRING 4
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_NAME_STARTSTR 5
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_NAME_ENDSTR 6
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_INPUT_TYPE_STRING 7
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_INPUT_TYPE_STARTSTR 8
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_INPUT_TYPE_ENDSTR 9
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OUTPUT_TYPE_STRING 10
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OUTPUT_TYPE_STARTSTR 11
+#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OUTPUT_TYPE_ENDSTR 12
+
+/* google.protobuf.MethodOptions */
+#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5
+
+/* google.protobuf.ServiceDescriptorProto */
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 3
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_STARTSEQ 4
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_ENDSEQ 5
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_ENDSUBMSG 6
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 7
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_NAME_STRING 8
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_NAME_STARTSTR 9
+#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_NAME_ENDSTR 10
+
+/* google.protobuf.ServiceOptions */
+#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5
+
+/* google.protobuf.SourceCodeInfo */
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_ENDSUBMSG 5
+
+/* google.protobuf.SourceCodeInfo.Location */
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_PATH_STARTSEQ 2
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_PATH_ENDSEQ 3
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_PATH_INT32 4
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_SPAN_STARTSEQ 5
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_SPAN_ENDSEQ 6
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_SPAN_INT32 7
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_LEADING_COMMENTS_STRING 8
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_LEADING_COMMENTS_STARTSTR 9
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_LEADING_COMMENTS_ENDSTR 10
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_TRAILING_COMMENTS_STRING 11
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_TRAILING_COMMENTS_STARTSTR 12
+#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_TRAILING_COMMENTS_ENDSTR 13
+
+/* google.protobuf.UninterpretedOption */
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_STARTSUBMSG 2
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_STARTSEQ 3
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_ENDSEQ 4
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_ENDSUBMSG 5
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_IDENTIFIER_VALUE_STRING 6
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_IDENTIFIER_VALUE_STARTSTR 7
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_IDENTIFIER_VALUE_ENDSTR 8
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_POSITIVE_INT_VALUE_UINT64 9
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NEGATIVE_INT_VALUE_INT64 10
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_DOUBLE_VALUE_DOUBLE 11
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_STRING_VALUE_STRING 12
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_STRING_VALUE_STARTSTR 13
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_STRING_VALUE_ENDSTR 14
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_AGGREGATE_VALUE_STRING 15
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_AGGREGATE_VALUE_STARTSTR 16
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_AGGREGATE_VALUE_ENDSTR 17
+
+/* google.protobuf.UninterpretedOption.NamePart */
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_NAME_PART_STRING 2
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_NAME_PART_STARTSTR 3
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_NAME_PART_ENDSTR 4
+#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_IS_EXTENSION_BOOL 5
+
+const upb_symtab *upbdefs_google_protobuf_descriptor(const void *owner);
+
+/* MessageDefs */
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_DescriptorProto(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.DescriptorProto");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.DescriptorProto.ExtensionRange");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumDescriptorProto(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumDescriptorProto");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumOptions(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumOptions");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumValueDescriptorProto(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumValueDescriptorProto");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumValueOptions(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumValueOptions");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FieldDescriptorProto(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FieldDescriptorProto");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FieldOptions(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FieldOptions");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FileDescriptorProto(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FileDescriptorProto");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FileDescriptorSet(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FileDescriptorSet");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FileOptions(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FileOptions");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_MessageOptions(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.MessageOptions");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_MethodDescriptorProto(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.MethodDescriptorProto");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_MethodOptions(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.MethodOptions");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_ServiceDescriptorProto(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.ServiceDescriptorProto");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_ServiceOptions(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.ServiceOptions");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.SourceCodeInfo");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo_Location(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.SourceCodeInfo.Location");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.UninterpretedOption");
+  assert(m);
+  return m;
+}
+UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption_NamePart(const upb_symtab *s) {
+  const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.UninterpretedOption.NamePart");
+  assert(m);
+  return m;
+}
+
+
+/* EnumDefs */
+UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Label(const upb_symtab *s) {
+  const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FieldDescriptorProto.Label");
+  assert(e);
+  return e;
+}
+UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Type(const upb_symtab *s) {
+  const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FieldDescriptorProto.Type");
+  assert(e);
+  return e;
+}
+UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FieldOptions_CType(const upb_symtab *s) {
+  const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FieldOptions.CType");
+  assert(e);
+  return e;
+}
+UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FileOptions_OptimizeMode(const upb_symtab *s) {
+  const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FileOptions.OptimizeMode");
+  assert(e);
+  return e;
+}
+
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_end(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto_ExtensionRange(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_start(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto_ExtensionRange(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_enum_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 4); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_extension(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 6); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_extension_range(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 5); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_field(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_nested_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 7); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumDescriptorProto(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumDescriptorProto(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumDescriptorProto(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumOptions_allow_alias(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumOptions(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumOptions(s), 999); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueDescriptorProto(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_number(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueDescriptorProto(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueDescriptorProto(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueOptions(s), 999); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_default_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 7); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_extendee(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_label(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 4); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_number(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 8); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 5); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_type_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 6); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_ctype(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_deprecated(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_experimental_map_key(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 9); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_lazy(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 5); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_packed(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 999); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_weak(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 10); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_dependency(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_enum_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 5); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_extension(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 7); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_message_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 4); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 8); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_package(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_public_dependency(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 10); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_service(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 6); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_source_code_info(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 9); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_weak_dependency(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 11); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorSet_file(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorSet(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_cc_generic_services(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 16); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_go_package(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 11); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_generate_equals_and_hash(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 20); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_generic_services(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 17); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_multiple_files(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 10); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_outer_classname(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 8); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_package(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_optimize_for(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 9); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_py_generic_services(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 18); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 999); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_message_set_wire_format(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MessageOptions(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_no_standard_descriptor_accessor(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MessageOptions(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MessageOptions(s), 999); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_input_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 4); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_output_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodOptions(s), 999); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_method(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceDescriptorProto(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceDescriptorProto(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceDescriptorProto(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceOptions(s), 999); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_leading_comments(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_path(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_span(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_trailing_comments(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 4); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_location(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_NamePart_is_extension(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption_NamePart(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_NamePart_name_part(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption_NamePart(s), 1); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_aggregate_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 8); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_double_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 6); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_identifier_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 3); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 2); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_negative_int_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 5); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_positive_int_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 4); }
+UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_string_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 7); }
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upbdefs {
+namespace google {
+namespace protobuf {
+namespace descriptor {
+inline upb::reffed_ptr<const upb::SymbolTable> SymbolTable() {
+  const upb::SymbolTable* s = upbdefs_google_protobuf_descriptor(&s);
+  return upb::reffed_ptr<const upb::SymbolTable>(s, &s);
+}
+}  /* namespace descriptor */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+#define RETURN_REFFED(type, func) \
+    const type* obj = func(upbdefs::google::protobuf::descriptor::SymbolTable().get()); \
+    return upb::reffed_ptr<const type>(obj);
+
+namespace google {
+namespace protobuf {
+namespace DescriptorProto {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_DescriptorProto) }
+inline upb::reffed_ptr<const upb::FieldDef> enum_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_enum_type) }
+inline upb::reffed_ptr<const upb::FieldDef> extension() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_extension) }
+inline upb::reffed_ptr<const upb::FieldDef> extension_range() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_extension_range) }
+inline upb::reffed_ptr<const upb::FieldDef> field() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_field) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_name) }
+inline upb::reffed_ptr<const upb::FieldDef> nested_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_nested_type) }
+inline upb::reffed_ptr<const upb::FieldDef> options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_options) }
+}  /* namespace DescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace DescriptorProto {
+namespace ExtensionRange {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_DescriptorProto_ExtensionRange) }
+inline upb::reffed_ptr<const upb::FieldDef> end() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_ExtensionRange_end) }
+inline upb::reffed_ptr<const upb::FieldDef> start() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_ExtensionRange_start) }
+}  /* namespace ExtensionRange */
+}  /* namespace DescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace EnumDescriptorProto {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumDescriptorProto) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumDescriptorProto_name) }
+inline upb::reffed_ptr<const upb::FieldDef> options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumDescriptorProto_options) }
+inline upb::reffed_ptr<const upb::FieldDef> value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumDescriptorProto_value) }
+}  /* namespace EnumDescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace EnumOptions {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumOptions) }
+inline upb::reffed_ptr<const upb::FieldDef> allow_alias() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumOptions_allow_alias) }
+inline upb::reffed_ptr<const upb::FieldDef> uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumOptions_uninterpreted_option) }
+}  /* namespace EnumOptions */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace EnumValueDescriptorProto {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumValueDescriptorProto) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueDescriptorProto_name) }
+inline upb::reffed_ptr<const upb::FieldDef> number() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueDescriptorProto_number) }
+inline upb::reffed_ptr<const upb::FieldDef> options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueDescriptorProto_options) }
+}  /* namespace EnumValueDescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace EnumValueOptions {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumValueOptions) }
+inline upb::reffed_ptr<const upb::FieldDef> uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueOptions_uninterpreted_option) }
+}  /* namespace EnumValueOptions */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace FieldDescriptorProto {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FieldDescriptorProto) }
+inline upb::reffed_ptr<const upb::FieldDef> default_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_default_value) }
+inline upb::reffed_ptr<const upb::FieldDef> extendee() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_extendee) }
+inline upb::reffed_ptr<const upb::FieldDef> label() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_label) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_name) }
+inline upb::reffed_ptr<const upb::FieldDef> number() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_number) }
+inline upb::reffed_ptr<const upb::FieldDef> options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_options) }
+inline upb::reffed_ptr<const upb::FieldDef> type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_type) }
+inline upb::reffed_ptr<const upb::FieldDef> type_name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_type_name) }
+inline upb::reffed_ptr<const upb::EnumDef> Label() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FieldDescriptorProto_Label) }
+inline upb::reffed_ptr<const upb::EnumDef> Type() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FieldDescriptorProto_Type) }
+}  /* namespace FieldDescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace FieldOptions {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FieldOptions) }
+inline upb::reffed_ptr<const upb::FieldDef> ctype() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_ctype) }
+inline upb::reffed_ptr<const upb::FieldDef> deprecated() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_deprecated) }
+inline upb::reffed_ptr<const upb::FieldDef> experimental_map_key() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_experimental_map_key) }
+inline upb::reffed_ptr<const upb::FieldDef> lazy() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_lazy) }
+inline upb::reffed_ptr<const upb::FieldDef> packed() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_packed) }
+inline upb::reffed_ptr<const upb::FieldDef> uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_uninterpreted_option) }
+inline upb::reffed_ptr<const upb::FieldDef> weak() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_weak) }
+inline upb::reffed_ptr<const upb::EnumDef> CType() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FieldOptions_CType) }
+}  /* namespace FieldOptions */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace FileDescriptorProto {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FileDescriptorProto) }
+inline upb::reffed_ptr<const upb::FieldDef> dependency() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_dependency) }
+inline upb::reffed_ptr<const upb::FieldDef> enum_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_enum_type) }
+inline upb::reffed_ptr<const upb::FieldDef> extension() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_extension) }
+inline upb::reffed_ptr<const upb::FieldDef> message_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_message_type) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_name) }
+inline upb::reffed_ptr<const upb::FieldDef> options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_options) }
+inline upb::reffed_ptr<const upb::FieldDef> package() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_package) }
+inline upb::reffed_ptr<const upb::FieldDef> public_dependency() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_public_dependency) }
+inline upb::reffed_ptr<const upb::FieldDef> service() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_service) }
+inline upb::reffed_ptr<const upb::FieldDef> source_code_info() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_source_code_info) }
+inline upb::reffed_ptr<const upb::FieldDef> weak_dependency() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_weak_dependency) }
+}  /* namespace FileDescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace FileDescriptorSet {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FileDescriptorSet) }
+inline upb::reffed_ptr<const upb::FieldDef> file() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorSet_file) }
+}  /* namespace FileDescriptorSet */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace FileOptions {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FileOptions) }
+inline upb::reffed_ptr<const upb::FieldDef> cc_generic_services() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_cc_generic_services) }
+inline upb::reffed_ptr<const upb::FieldDef> go_package() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_go_package) }
+inline upb::reffed_ptr<const upb::FieldDef> java_generate_equals_and_hash() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_generate_equals_and_hash) }
+inline upb::reffed_ptr<const upb::FieldDef> java_generic_services() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_generic_services) }
+inline upb::reffed_ptr<const upb::FieldDef> java_multiple_files() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_multiple_files) }
+inline upb::reffed_ptr<const upb::FieldDef> java_outer_classname() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_outer_classname) }
+inline upb::reffed_ptr<const upb::FieldDef> java_package() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_package) }
+inline upb::reffed_ptr<const upb::FieldDef> optimize_for() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_optimize_for) }
+inline upb::reffed_ptr<const upb::FieldDef> py_generic_services() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_py_generic_services) }
+inline upb::reffed_ptr<const upb::FieldDef> uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_uninterpreted_option) }
+inline upb::reffed_ptr<const upb::EnumDef> OptimizeMode() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FileOptions_OptimizeMode) }
+}  /* namespace FileOptions */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace MessageOptions {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_MessageOptions) }
+inline upb::reffed_ptr<const upb::FieldDef> message_set_wire_format() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MessageOptions_message_set_wire_format) }
+inline upb::reffed_ptr<const upb::FieldDef> no_standard_descriptor_accessor() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MessageOptions_no_standard_descriptor_accessor) }
+inline upb::reffed_ptr<const upb::FieldDef> uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MessageOptions_uninterpreted_option) }
+}  /* namespace MessageOptions */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace MethodDescriptorProto {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_MethodDescriptorProto) }
+inline upb::reffed_ptr<const upb::FieldDef> input_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_input_type) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_name) }
+inline upb::reffed_ptr<const upb::FieldDef> options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_options) }
+inline upb::reffed_ptr<const upb::FieldDef> output_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_output_type) }
+}  /* namespace MethodDescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace MethodOptions {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_MethodOptions) }
+inline upb::reffed_ptr<const upb::FieldDef> uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodOptions_uninterpreted_option) }
+}  /* namespace MethodOptions */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace ServiceDescriptorProto {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_ServiceDescriptorProto) }
+inline upb::reffed_ptr<const upb::FieldDef> method() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceDescriptorProto_method) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceDescriptorProto_name) }
+inline upb::reffed_ptr<const upb::FieldDef> options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceDescriptorProto_options) }
+}  /* namespace ServiceDescriptorProto */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace ServiceOptions {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_ServiceOptions) }
+inline upb::reffed_ptr<const upb::FieldDef> uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceOptions_uninterpreted_option) }
+}  /* namespace ServiceOptions */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace SourceCodeInfo {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_SourceCodeInfo) }
+inline upb::reffed_ptr<const upb::FieldDef> location() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_location) }
+}  /* namespace SourceCodeInfo */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace SourceCodeInfo {
+namespace Location {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_SourceCodeInfo_Location) }
+inline upb::reffed_ptr<const upb::FieldDef> leading_comments() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_leading_comments) }
+inline upb::reffed_ptr<const upb::FieldDef> path() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_path) }
+inline upb::reffed_ptr<const upb::FieldDef> span() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_span) }
+inline upb::reffed_ptr<const upb::FieldDef> trailing_comments() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_trailing_comments) }
+}  /* namespace Location */
+}  /* namespace SourceCodeInfo */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace UninterpretedOption {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_UninterpretedOption) }
+inline upb::reffed_ptr<const upb::FieldDef> aggregate_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_aggregate_value) }
+inline upb::reffed_ptr<const upb::FieldDef> double_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_double_value) }
+inline upb::reffed_ptr<const upb::FieldDef> identifier_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_identifier_value) }
+inline upb::reffed_ptr<const upb::FieldDef> name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_name) }
+inline upb::reffed_ptr<const upb::FieldDef> negative_int_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_negative_int_value) }
+inline upb::reffed_ptr<const upb::FieldDef> positive_int_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_positive_int_value) }
+inline upb::reffed_ptr<const upb::FieldDef> string_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_string_value) }
+}  /* namespace UninterpretedOption */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+namespace google {
+namespace protobuf {
+namespace UninterpretedOption {
+namespace NamePart {
+inline upb::reffed_ptr<const upb::MessageDef> MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_UninterpretedOption_NamePart) }
+inline upb::reffed_ptr<const upb::FieldDef> is_extension() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_NamePart_is_extension) }
+inline upb::reffed_ptr<const upb::FieldDef> name_part() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_NamePart_name_part) }
+}  /* namespace NamePart */
+}  /* namespace UninterpretedOption */
+}  /* namespace protobuf */
+}  /* namespace google */
+
+}  /* namespace upbdefs */
+
+
+#undef RETURN_REFFED
+#endif /* __cplusplus */
+
+#endif  /* GOOGLE_PROTOBUF_DESCRIPTOR_UPB_H_ */
+/*
+** Internal-only definitions for the decoder.
+*/
+
+#ifndef UPB_DECODER_INT_H_
+#define UPB_DECODER_INT_H_
+
+#include <stdlib.h>
+/*
+** upb::pb::Decoder
+**
+** A high performance, streaming, resumable decoder for the binary protobuf
+** format.
+**
+** This interface works the same regardless of what decoder backend is being
+** used.  A client of this class does not need to know whether decoding is using
+** a JITted decoder (DynASM, LLVM, etc) or an interpreted decoder.  By default,
+** it will always use the fastest available decoder.  However, you can call
+** set_allow_jit(false) to disable any JIT decoder that might be available.
+** This is primarily useful for testing purposes.
+*/
+
+#ifndef UPB_DECODER_H_
+#define UPB_DECODER_H_
+
+
+#ifdef __cplusplus
+namespace upb {
+namespace pb {
+class CodeCache;
+class Decoder;
+class DecoderMethod;
+class DecoderMethodOptions;
+}  /* namespace pb */
+}  /* namespace upb */
+#endif
+
+UPB_DECLARE_TYPE(upb::pb::CodeCache, upb_pbcodecache)
+UPB_DECLARE_TYPE(upb::pb::Decoder, upb_pbdecoder)
+UPB_DECLARE_TYPE(upb::pb::DecoderMethodOptions, upb_pbdecodermethodopts)
+
+UPB_DECLARE_DERIVED_TYPE(upb::pb::DecoderMethod, upb::RefCounted,
+                         upb_pbdecodermethod, upb_refcounted)
+
+#ifdef __cplusplus
+
+/* The parameters one uses to construct a DecoderMethod.
+ * TODO(haberman): move allowjit here?  Seems more convenient for users.
+ * TODO(haberman): move this to be heap allocated for ABI stability. */
+class upb::pb::DecoderMethodOptions {
+ public:
+  /* Parameter represents the destination handlers that this method will push
+   * to. */
+  explicit DecoderMethodOptions(const Handlers* dest_handlers);
+
+  /* Should the decoder push submessages to lazy handlers for fields that have
+   * them?  The caller should set this iff the lazy handlers expect data that is
+   * in protobuf binary format and the caller wishes to lazy parse it. */
+  void set_lazy(bool lazy);
+#else
+struct upb_pbdecodermethodopts {
+#endif
+  const upb_handlers *handlers;
+  bool lazy;
+};
+
+#ifdef __cplusplus
+
+/* Represents the code to parse a protobuf according to a destination
+ * Handlers. */
+class upb::pb::DecoderMethod {
+ public:
+  /* Include base methods from upb::ReferenceCounted. */
+  UPB_REFCOUNTED_CPPMETHODS
+
+  /* The destination handlers that are statically bound to this method.
+   * This method is only capable of outputting to a sink that uses these
+   * handlers. */
+  const Handlers* dest_handlers() const;
+
+  /* The input handlers for this decoder method. */
+  const BytesHandler* input_handler() const;
+
+  /* Whether this method is native. */
+  bool is_native() const;
+
+  /* Convenience method for generating a DecoderMethod without explicitly
+   * creating a CodeCache. */
+  static reffed_ptr<const DecoderMethod> New(const DecoderMethodOptions& opts);
+
+ private:
+  UPB_DISALLOW_POD_OPS(DecoderMethod, upb::pb::DecoderMethod)
+};
+
+#endif
+
+/* Preallocation hint: decoder won't allocate more bytes than this when first
+ * constructed.  This hint may be an overestimate for some build configurations.
+ * But if the decoder library is upgraded without recompiling the application,
+ * it may be an underestimate. */
+#define UPB_PB_DECODER_SIZE 4408
+
+#ifdef __cplusplus
+
+/* A Decoder receives binary protobuf data on its input sink and pushes the
+ * decoded data to its output sink. */
+class upb::pb::Decoder {
+ public:
+  /* Constructs a decoder instance for the given method, which must outlive this
+   * decoder.  Any errors during parsing will be set on the given status, which
+   * must also outlive this decoder.
+   *
+   * The sink must match the given method. */
+  static Decoder* Create(Environment* env, const DecoderMethod* method,
+                         Sink* output);
+
+  /* Returns the DecoderMethod this decoder is parsing from. */
+  const DecoderMethod* method() const;
+
+  /* The sink on which this decoder receives input. */
+  BytesSink* input();
+
+  /* Returns number of bytes successfully parsed.
+   *
+   * This can be useful for determining the stream position where an error
+   * occurred.
+   *
+   * This value may not be up-to-date when called from inside a parsing
+   * callback. */
+  uint64_t BytesParsed() const;
+
+  /* Gets/sets the parsing nexting limit.  If the total number of nested
+   * submessages and repeated fields hits this limit, parsing will fail.  This
+   * is a resource limit that controls the amount of memory used by the parsing
+   * stack.
+   *
+   * Setting the limit will fail if the parser is currently suspended at a depth
+   * greater than this, or if memory allocation of the stack fails. */
+  size_t max_nesting() const;
+  bool set_max_nesting(size_t max);
+
+  void Reset();
+
+  static const size_t kSize = UPB_PB_DECODER_SIZE;
+
+ private:
+  UPB_DISALLOW_POD_OPS(Decoder, upb::pb::Decoder)
+};
+
+#endif  /* __cplusplus */
+
+#ifdef __cplusplus
+
+/* A class for caching protobuf processing code, whether bytecode for the
+ * interpreted decoder or machine code for the JIT.
+ *
+ * This class is not thread-safe.
+ *
+ * TODO(haberman): move this to be heap allocated for ABI stability. */
+class upb::pb::CodeCache {
+ public:
+  CodeCache();
+  ~CodeCache();
+
+  /* Whether the cache is allowed to generate machine code.  Defaults to true.
+   * There is no real reason to turn it off except for testing or if you are
+   * having a specific problem with the JIT.
+   *
+   * Note that allow_jit = true does not *guarantee* that the code will be JIT
+   * compiled.  If this platform is not supported or the JIT was not compiled
+   * in, the code may still be interpreted. */
+  bool allow_jit() const;
+
+  /* This may only be called when the object is first constructed, and prior to
+   * any code generation, otherwise returns false and does nothing. */
+  bool set_allow_jit(bool allow);
+
+  /* Returns a DecoderMethod that can push data to the given handlers.
+   * If a suitable method already exists, it will be returned from the cache.
+   *
+   * Specifying the destination handlers here allows the DecoderMethod to be
+   * statically bound to the destination handlers if possible, which can allow
+   * more efficient decoding.  However the returned method may or may not
+   * actually be statically bound.  But in all cases, the returned method can
+   * push data to the given handlers. */
+  const DecoderMethod *GetDecoderMethod(const DecoderMethodOptions& opts);
+
+  /* If/when someone needs to explicitly create a dynamically-bound
+   * DecoderMethod*, we can add a method to get it here. */
+
+ private:
+  UPB_DISALLOW_COPY_AND_ASSIGN(CodeCache)
+#else
+struct upb_pbcodecache {
+#endif
+  bool allow_jit_;
+
+  /* Array of mgroups. */
+  upb_inttable groups;
+};
+
+UPB_BEGIN_EXTERN_C
+
+upb_pbdecoder *upb_pbdecoder_create(upb_env *e,
+                                    const upb_pbdecodermethod *method,
+                                    upb_sink *output);
+const upb_pbdecodermethod *upb_pbdecoder_method(const upb_pbdecoder *d);
+upb_bytessink *upb_pbdecoder_input(upb_pbdecoder *d);
+uint64_t upb_pbdecoder_bytesparsed(const upb_pbdecoder *d);
+size_t upb_pbdecoder_maxnesting(const upb_pbdecoder *d);
+bool upb_pbdecoder_setmaxnesting(upb_pbdecoder *d, size_t max);
+void upb_pbdecoder_reset(upb_pbdecoder *d);
+
+void upb_pbdecodermethodopts_init(upb_pbdecodermethodopts *opts,
+                                  const upb_handlers *h);
+void upb_pbdecodermethodopts_setlazy(upb_pbdecodermethodopts *opts, bool lazy);
+
+
+/* Include refcounted methods like upb_pbdecodermethod_ref(). */
+UPB_REFCOUNTED_CMETHODS(upb_pbdecodermethod, upb_pbdecodermethod_upcast)
+
+const upb_handlers *upb_pbdecodermethod_desthandlers(
+    const upb_pbdecodermethod *m);
+const upb_byteshandler *upb_pbdecodermethod_inputhandler(
+    const upb_pbdecodermethod *m);
+bool upb_pbdecodermethod_isnative(const upb_pbdecodermethod *m);
+const upb_pbdecodermethod *upb_pbdecodermethod_new(
+    const upb_pbdecodermethodopts *opts, const void *owner);
+
+void upb_pbcodecache_init(upb_pbcodecache *c);
+void upb_pbcodecache_uninit(upb_pbcodecache *c);
+bool upb_pbcodecache_allowjit(const upb_pbcodecache *c);
+bool upb_pbcodecache_setallowjit(upb_pbcodecache *c, bool allow);
+const upb_pbdecodermethod *upb_pbcodecache_getdecodermethod(
+    upb_pbcodecache *c, const upb_pbdecodermethodopts *opts);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upb {
+
+namespace pb {
+
+/* static */
+inline Decoder* Decoder::Create(Environment* env, const DecoderMethod* m,
+                                Sink* sink) {
+  return upb_pbdecoder_create(env, m, sink);
+}
+inline const DecoderMethod* Decoder::method() const {
+  return upb_pbdecoder_method(this);
+}
+inline BytesSink* Decoder::input() {
+  return upb_pbdecoder_input(this);
+}
+inline uint64_t Decoder::BytesParsed() const {
+  return upb_pbdecoder_bytesparsed(this);
+}
+inline size_t Decoder::max_nesting() const {
+  return upb_pbdecoder_maxnesting(this);
+}
+inline bool Decoder::set_max_nesting(size_t max) {
+  return upb_pbdecoder_setmaxnesting(this, max);
+}
+inline void Decoder::Reset() { upb_pbdecoder_reset(this); }
+
+inline DecoderMethodOptions::DecoderMethodOptions(const Handlers* h) {
+  upb_pbdecodermethodopts_init(this, h);
+}
+inline void DecoderMethodOptions::set_lazy(bool lazy) {
+  upb_pbdecodermethodopts_setlazy(this, lazy);
+}
+
+inline const Handlers* DecoderMethod::dest_handlers() const {
+  return upb_pbdecodermethod_desthandlers(this);
+}
+inline const BytesHandler* DecoderMethod::input_handler() const {
+  return upb_pbdecodermethod_inputhandler(this);
+}
+inline bool DecoderMethod::is_native() const {
+  return upb_pbdecodermethod_isnative(this);
+}
+/* static */
+inline reffed_ptr<const DecoderMethod> DecoderMethod::New(
+    const DecoderMethodOptions &opts) {
+  const upb_pbdecodermethod *m = upb_pbdecodermethod_new(&opts, &m);
+  return reffed_ptr<const DecoderMethod>(m, &m);
+}
+
+inline CodeCache::CodeCache() {
+  upb_pbcodecache_init(this);
+}
+inline CodeCache::~CodeCache() {
+  upb_pbcodecache_uninit(this);
+}
+inline bool CodeCache::allow_jit() const {
+  return upb_pbcodecache_allowjit(this);
+}
+inline bool CodeCache::set_allow_jit(bool allow) {
+  return upb_pbcodecache_setallowjit(this, allow);
+}
+inline const DecoderMethod *CodeCache::GetDecoderMethod(
+    const DecoderMethodOptions& opts) {
+  return upb_pbcodecache_getdecodermethod(this, &opts);
+}
+
+}  /* namespace pb */
+}  /* namespace upb */
+
+#endif  /* __cplusplus */
+
+#endif  /* UPB_DECODER_H_ */
+
+/* C++ names are not actually used since this type isn't exposed to users. */
+#ifdef __cplusplus
+namespace upb {
+namespace pb {
+class MessageGroup;
+}  /* namespace pb */
+}  /* namespace upb */
+#endif
+UPB_DECLARE_DERIVED_TYPE(upb::pb::MessageGroup, upb::RefCounted,
+                         mgroup, upb_refcounted)
+
+/* Opcode definitions.  The canonical meaning of each opcode is its
+ * implementation in the interpreter (the JIT is written to match this).
+ *
+ * All instructions have the opcode in the low byte.
+ * Instruction format for most instructions is:
+ *
+ * +-------------------+--------+
+ * |     arg (24)      | op (8) |
+ * +-------------------+--------+
+ *
+ * Exceptions are indicated below.  A few opcodes are multi-word. */
+typedef enum {
+  /* Opcodes 1-8, 13, 15-18 parse their respective descriptor types.
+   * Arg for all of these is the upb selector for this field. */
+#define T(type) OP_PARSE_ ## type = UPB_DESCRIPTOR_TYPE_ ## type
+  T(DOUBLE), T(FLOAT), T(INT64), T(UINT64), T(INT32), T(FIXED64), T(FIXED32),
+  T(BOOL), T(UINT32), T(SFIXED32), T(SFIXED64), T(SINT32), T(SINT64),
+#undef T
+  OP_STARTMSG       = 9,   /* No arg. */
+  OP_ENDMSG         = 10,  /* No arg. */
+  OP_STARTSEQ       = 11,
+  OP_ENDSEQ         = 12,
+  OP_STARTSUBMSG    = 14,
+  OP_ENDSUBMSG      = 19,
+  OP_STARTSTR       = 20,
+  OP_STRING         = 21,
+  OP_ENDSTR         = 22,
+
+  OP_PUSHTAGDELIM   = 23,  /* No arg. */
+  OP_PUSHLENDELIM   = 24,  /* No arg. */
+  OP_POP            = 25,  /* No arg. */
+  OP_SETDELIM       = 26,  /* No arg. */
+  OP_SETBIGGROUPNUM = 27,  /* two words:
+                            *   | unused (24)     | opc (8) |
+                            *   |        groupnum (32)      | */
+  OP_CHECKDELIM     = 28,
+  OP_CALL           = 29,
+  OP_RET            = 30,
+  OP_BRANCH         = 31,
+
+  /* Different opcodes depending on how many bytes expected. */
+  OP_TAG1           = 32,  /* | match tag (16) | jump target (8) | opc (8) | */
+  OP_TAG2           = 33,  /* | match tag (16) | jump target (8) | opc (8) | */
+  OP_TAGN           = 34,  /* three words: */
+                           /*   | unused (16) | jump target(8) | opc (8) | */
+                           /*   |           match tag 1 (32)             | */
+                           /*   |           match tag 2 (32)             | */
+
+  OP_SETDISPATCH    = 35,  /* N words: */
+                           /*   | unused (24)         | opc | */
+                           /*   | upb_inttable* (32 or 64)  | */
+
+  OP_DISPATCH       = 36,  /* No arg. */
+
+  OP_HALT           = 37   /* No arg. */
+} opcode;
+
+#define OP_MAX OP_HALT
+
+UPB_INLINE opcode getop(uint32_t instr) { return instr & 0xff; }
+
+/* Method group; represents a set of decoder methods that had their code
+ * emitted together, and must therefore be freed together.  Immutable once
+ * created.  It is possible we may want to expose this to users at some point.
+ *
+ * Overall ownership of Decoder objects looks like this:
+ *
+ *                +----------+
+ *                |          | <---> DecoderMethod
+ *                | method   |
+ * CodeCache ---> |  group   | <---> DecoderMethod
+ *                |          |
+ *                | (mgroup) | <---> DecoderMethod
+ *                +----------+
+ */
+struct mgroup {
+  upb_refcounted base;
+
+  /* Maps upb_msgdef/upb_handlers -> upb_pbdecodermethod.  We own refs on the
+   * methods. */
+  upb_inttable methods;
+
+  /* When we add the ability to link to previously existing mgroups, we'll
+   * need an array of mgroups we reference here, and own refs on them. */
+
+  /* The bytecode for our methods, if any exists.  Owned by us. */
+  uint32_t *bytecode;
+  uint32_t *bytecode_end;
+
+#ifdef UPB_USE_JIT_X64
+  /* JIT-generated machine code, if any. */
+  upb_string_handlerfunc *jit_code;
+  /* The size of the jit_code (required to munmap()). */
+  size_t jit_size;
+  char *debug_info;
+  void *dl;
+#endif
+};
+
+/* The maximum that any submessages can be nested.  Matches proto2's limit.
+ * This specifies the size of the decoder's statically-sized array and therefore
+ * setting it high will cause the upb::pb::Decoder object to be larger.
+ *
+ * If necessary we can add a runtime-settable property to Decoder that allow
+ * this to be larger than the compile-time setting, but this would add
+ * complexity, particularly since we would have to decide how/if to give users
+ * the ability to set a custom memory allocation function. */
+#define UPB_DECODER_MAX_NESTING 64
+
+/* Internal-only struct used by the decoder. */
+typedef struct {
+  /* Space optimization note: we store two pointers here that the JIT
+   * doesn't need at all; the upb_handlers* inside the sink and
+   * the dispatch table pointer.  We can optimze so that the JIT uses
+   * smaller stack frames than the interpreter.  The only thing we need
+   * to guarantee is that the fallback routines can find end_ofs. */
+  upb_sink sink;
+
+  /* The absolute stream offset of the end-of-frame delimiter.
+   * Non-delimited frames (groups and non-packed repeated fields) reuse the
+   * delimiter of their parent, even though the frame may not end there.
+   *
+   * NOTE: the JIT stores a slightly different value here for non-top frames.
+   * It stores the value relative to the end of the enclosed message.  But the
+   * top frame is still stored the same way, which is important for ensuring
+   * that calls from the JIT into C work correctly. */
+  uint64_t end_ofs;
+  const uint32_t *base;
+
+  /* 0 indicates a length-delimited field.
+   * A positive number indicates a known group.
+   * A negative number indicates an unknown group. */
+  int32_t groupnum;
+  upb_inttable *dispatch;  /* Not used by the JIT. */
+} upb_pbdecoder_frame;
+
+struct upb_pbdecodermethod {
+  upb_refcounted base;
+
+  /* While compiling, the base is relative in "ofs", after compiling it is
+   * absolute in "ptr". */
+  union {
+    uint32_t ofs;     /* PC offset of method. */
+    void *ptr;        /* Pointer to bytecode or machine code for this method. */
+  } code_base;
+
+  /* The decoder method group to which this method belongs.  We own a ref.
+   * Owning a ref on the entire group is more coarse-grained than is strictly
+   * necessary; all we truly require is that methods we directly reference
+   * outlive us, while the group could contain many other messages we don't
+   * require.  But the group represents the messages that were
+   * allocated+compiled together, so it makes the most sense to free them
+   * together also. */
+  const upb_refcounted *group;
+
+  /* Whether this method is native code or bytecode. */
+  bool is_native_;
+
+  /* The handler one calls to invoke this method. */
+  upb_byteshandler input_handler_;
+
+  /* The destination handlers this method is bound to.  We own a ref. */
+  const upb_handlers *dest_handlers_;
+
+  /* Dispatch table -- used by both bytecode decoder and JIT when encountering a
+   * field number that wasn't the one we were expecting to see.  See
+   * decoder.int.h for the layout of this table. */
+  upb_inttable dispatch;
+};
+
+struct upb_pbdecoder {
+  upb_env *env;
+
+  /* Our input sink. */
+  upb_bytessink input_;
+
+  /* The decoder method we are parsing with (owned). */
+  const upb_pbdecodermethod *method_;
+
+  size_t call_len;
+  const uint32_t *pc, *last;
+
+  /* Current input buffer and its stream offset. */
+  const char *buf, *ptr, *end, *checkpoint;
+
+  /* End of the delimited region, relative to ptr, NULL if not in this buf. */
+  const char *delim_end;
+
+  /* End of the delimited region, relative to ptr, end if not in this buf. */
+  const char *data_end;
+
+  /* Overall stream offset of "buf." */
+  uint64_t bufstart_ofs;
+
+  /* Buffer for residual bytes not parsed from the previous buffer.
+   * The maximum number of residual bytes we require is 12; a five-byte
+   * unknown tag plus an eight-byte value, less one because the value
+   * is only a partial value. */
+  char residual[12];
+  char *residual_end;
+
+  /* Bytes of data that should be discarded from the input beore we start
+   * parsing again.  We set this when we internally determine that we can
+   * safely skip the next N bytes, but this region extends past the current
+   * user buffer. */
+  size_t skip;
+
+  /* Stores the user buffer passed to our decode function. */
+  const char *buf_param;
+  size_t size_param;
+  const upb_bufhandle *handle;
+
+  /* Our internal stack. */
+  upb_pbdecoder_frame *stack, *top, *limit;
+  const uint32_t **callstack;
+  size_t stack_size;
+
+  upb_status *status;
+
+#ifdef UPB_USE_JIT_X64
+  /* Used momentarily by the generated code to store a value while a user
+   * function is called. */
+  uint32_t tmp_len;
+
+  const void *saved_rsp;
+#endif
+};
+
+/* Decoder entry points; used as handlers. */
+void *upb_pbdecoder_startbc(void *closure, const void *pc, size_t size_hint);
+void *upb_pbdecoder_startjit(void *closure, const void *hd, size_t size_hint);
+size_t upb_pbdecoder_decode(void *closure, const void *hd, const char *buf,
+                            size_t size, const upb_bufhandle *handle);
+bool upb_pbdecoder_end(void *closure, const void *handler_data);
+
+/* Decoder-internal functions that the JIT calls to handle fallback paths. */
+int32_t upb_pbdecoder_resume(upb_pbdecoder *d, void *p, const char *buf,
+                             size_t size, const upb_bufhandle *handle);
+size_t upb_pbdecoder_suspend(upb_pbdecoder *d);
+int32_t upb_pbdecoder_skipunknown(upb_pbdecoder *d, int32_t fieldnum,
+                                  uint8_t wire_type);
+int32_t upb_pbdecoder_checktag_slow(upb_pbdecoder *d, uint64_t expected);
+int32_t upb_pbdecoder_decode_varint_slow(upb_pbdecoder *d, uint64_t *u64);
+int32_t upb_pbdecoder_decode_f32(upb_pbdecoder *d, uint32_t *u32);
+int32_t upb_pbdecoder_decode_f64(upb_pbdecoder *d, uint64_t *u64);
+void upb_pbdecoder_seterr(upb_pbdecoder *d, const char *msg);
+
+/* Error messages that are shared between the bytecode and JIT decoders. */
+extern const char *kPbDecoderStackOverflow;
+extern const char *kPbDecoderSubmessageTooLong;
+
+/* Access to decoderplan members needed by the decoder. */
+const char *upb_pbdecoder_getopname(unsigned int op);
+
+/* JIT codegen entry point. */
+void upb_pbdecoder_jit(mgroup *group);
+void upb_pbdecoder_freejit(mgroup *group);
+UPB_REFCOUNTED_CMETHODS(mgroup, mgroup_upcast)
+
+/* A special label that means "do field dispatch for this message and branch to
+ * wherever that takes you." */
+#define LABEL_DISPATCH 0
+
+/* A special slot in the dispatch table that stores the epilogue (ENDMSG and/or
+ * RET) for branching to when we find an appropriate ENDGROUP tag. */
+#define DISPATCH_ENDMSG 0
+
+/* It's important to use this invalid wire type instead of 0 (which is a valid
+ * wire type). */
+#define NO_WIRE_TYPE 0xff
+
+/* The dispatch table layout is:
+ *   [field number] -> [ 48-bit offset ][ 8-bit wt2 ][ 8-bit wt1 ]
+ *
+ * If wt1 matches, jump to the 48-bit offset.  If wt2 matches, lookup
+ * (UPB_MAX_FIELDNUMBER + fieldnum) and jump there.
+ *
+ * We need two wire types because of packed/non-packed compatibility.  A
+ * primitive repeated field can use either wire type and be valid.  While we
+ * could key the table on fieldnum+wiretype, the table would be 8x sparser.
+ *
+ * Storing two wire types in the primary value allows us to quickly rule out
+ * the second wire type without needing to do a separate lookup (this case is
+ * less common than an unknown field). */
+UPB_INLINE uint64_t upb_pbdecoder_packdispatch(uint64_t ofs, uint8_t wt1,
+                                               uint8_t wt2) {
+  return (ofs << 16) | (wt2 << 8) | wt1;
+}
+
+UPB_INLINE void upb_pbdecoder_unpackdispatch(uint64_t dispatch, uint64_t *ofs,
+                                             uint8_t *wt1, uint8_t *wt2) {
+  *wt1 = (uint8_t)dispatch;
+  *wt2 = (uint8_t)(dispatch >> 8);
+  *ofs = dispatch >> 16;
+}
+
+/* All of the functions in decoder.c that return int32_t return values according
+ * to the following scheme:
+ *   1. negative values indicate a return code from the following list.
+ *   2. positive values indicate that error or end of buffer was hit, and
+ *      that the decode function should immediately return the given value
+ *      (the decoder state has already been suspended and is ready to be
+ *      resumed). */
+#define DECODE_OK -1
+#define DECODE_MISMATCH -2  /* Used only from checktag_slow(). */
+#define DECODE_ENDGROUP -3  /* Used only from checkunknown(). */
+
+#define CHECK_RETURN(x) { int32_t ret = x; if (ret >= 0) return ret; }
+
+#endif  /* UPB_DECODER_INT_H_ */
+/*
+** A number of routines for varint manipulation (we keep them all around to
+** have multiple approaches available for benchmarking).
+*/
+
+#ifndef UPB_VARINT_DECODER_H_
+#define UPB_VARINT_DECODER_H_
+
+#include <assert.h>
+#include <stdint.h>
+#include <string.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* A list of types as they are encoded on-the-wire. */
+typedef enum {
+  UPB_WIRE_TYPE_VARINT      = 0,
+  UPB_WIRE_TYPE_64BIT       = 1,
+  UPB_WIRE_TYPE_DELIMITED   = 2,
+  UPB_WIRE_TYPE_START_GROUP = 3,
+  UPB_WIRE_TYPE_END_GROUP   = 4,
+  UPB_WIRE_TYPE_32BIT       = 5
+} upb_wiretype_t;
+
+#define UPB_MAX_WIRE_TYPE 5
+
+/* The maximum number of bytes that it takes to encode a 64-bit varint.
+ * Note that with a better encoding this could be 9 (TODO: write up a
+ * wiki document about this). */
+#define UPB_PB_VARINT_MAX_LEN 10
+
+/* Array of the "native" (ie. non-packed-repeated) wire type for the given a
+ * descriptor type (upb_descriptortype_t). */
+extern const uint8_t upb_pb_native_wire_types[];
+
+/* Zig-zag encoding/decoding **************************************************/
+
+UPB_INLINE int32_t upb_zzdec_32(uint32_t n) {
+  return (n >> 1) ^ -(int32_t)(n & 1);
+}
+UPB_INLINE int64_t upb_zzdec_64(uint64_t n) {
+  return (n >> 1) ^ -(int64_t)(n & 1);
+}
+UPB_INLINE uint32_t upb_zzenc_32(int32_t n) { return (n << 1) ^ (n >> 31); }
+UPB_INLINE uint64_t upb_zzenc_64(int64_t n) { return (n << 1) ^ (n >> 63); }
+
+/* Decoding *******************************************************************/
+
+/* All decoding functions return this struct by value. */
+typedef struct {
+  const char *p;  /* NULL if the varint was unterminated. */
+  uint64_t val;
+} upb_decoderet;
+
+UPB_INLINE upb_decoderet upb_decoderet_make(const char *p, uint64_t val) {
+  upb_decoderet ret;
+  ret.p = p;
+  ret.val = val;
+  return ret;
+}
+
+/* Four functions for decoding a varint of at most eight bytes.  They are all
+ * functionally identical, but are implemented in different ways and likely have
+ * different performance profiles.  We keep them around for performance testing.
+ *
+ * Note that these functions may not read byte-by-byte, so they must not be used
+ * unless there are at least eight bytes left in the buffer! */
+upb_decoderet upb_vdecode_max8_branch32(upb_decoderet r);
+upb_decoderet upb_vdecode_max8_branch64(upb_decoderet r);
+upb_decoderet upb_vdecode_max8_wright(upb_decoderet r);
+upb_decoderet upb_vdecode_max8_massimino(upb_decoderet r);
+
+/* Template for a function that checks the first two bytes with branching
+ * and dispatches 2-10 bytes with a separate function.  Note that this may read
+ * up to 10 bytes, so it must not be used unless there are at least ten bytes
+ * left in the buffer! */
+#define UPB_VARINT_DECODER_CHECK2(name, decode_max8_function)                  \
+UPB_INLINE upb_decoderet upb_vdecode_check2_ ## name(const char *_p) {         \
+  uint8_t *p = (uint8_t*)_p;                                                   \
+  upb_decoderet r;                                                             \
+  if ((*p & 0x80) == 0) {                                                      \
+  /* Common case: one-byte varint. */                                          \
+    return upb_decoderet_make(_p + 1, *p & 0x7fU);                             \
+  }                                                                            \
+  r = upb_decoderet_make(_p + 2, (*p & 0x7fU) | ((*(p + 1) & 0x7fU) << 7));    \
+  if ((*(p + 1) & 0x80) == 0) {                                                \
+    /* Two-byte varint. */                                                     \
+    return r;                                                                  \
+  }                                                                            \
+  /* Longer varint, fallback to out-of-line function. */                       \
+  return decode_max8_function(r);                                              \
+}
+
+UPB_VARINT_DECODER_CHECK2(branch32, upb_vdecode_max8_branch32)
+UPB_VARINT_DECODER_CHECK2(branch64, upb_vdecode_max8_branch64)
+UPB_VARINT_DECODER_CHECK2(wright, upb_vdecode_max8_wright)
+UPB_VARINT_DECODER_CHECK2(massimino, upb_vdecode_max8_massimino)
+#undef UPB_VARINT_DECODER_CHECK2
+
+/* Our canonical functions for decoding varints, based on the currently
+ * favored best-performing implementations. */
+UPB_INLINE upb_decoderet upb_vdecode_fast(const char *p) {
+  if (sizeof(long) == 8)
+    return upb_vdecode_check2_branch64(p);
+  else
+    return upb_vdecode_check2_branch32(p);
+}
+
+UPB_INLINE upb_decoderet upb_vdecode_max8_fast(upb_decoderet r) {
+  return upb_vdecode_max8_massimino(r);
+}
+
+
+/* Encoding *******************************************************************/
+
+UPB_INLINE int upb_value_size(uint64_t val) {
+#ifdef __GNUC__
+  int high_bit = 63 - __builtin_clzll(val);  /* 0-based, undef if val == 0. */
+#else
+  int high_bit = 0;
+  uint64_t tmp = val;
+  while(tmp >>= 1) high_bit++;
+#endif
+  return val == 0 ? 1 : high_bit / 8 + 1;
+}
+
+/* Encodes a 64-bit varint into buf (which must be >=UPB_PB_VARINT_MAX_LEN
+ * bytes long), returning how many bytes were used.
+ *
+ * TODO: benchmark and optimize if necessary. */
+UPB_INLINE size_t upb_vencode64(uint64_t val, char *buf) {
+  size_t i;
+  if (val == 0) { buf[0] = 0; return 1; }
+  i = 0;
+  while (val) {
+    uint8_t byte = val & 0x7fU;
+    val >>= 7;
+    if (val) byte |= 0x80U;
+    buf[i++] = byte;
+  }
+  return i;
+}
+
+UPB_INLINE size_t upb_varint_size(uint64_t val) {
+  char buf[UPB_PB_VARINT_MAX_LEN];
+  return upb_vencode64(val, buf);
+}
+
+/* Encodes a 32-bit varint, *not* sign-extended. */
+UPB_INLINE uint64_t upb_vencode32(uint32_t val) {
+  char buf[UPB_PB_VARINT_MAX_LEN];
+  size_t bytes = upb_vencode64(val, buf);
+  uint64_t ret = 0;
+  assert(bytes <= 5);
+  memcpy(&ret, buf, bytes);
+  assert(ret <= 0xffffffffffU);
+  return ret;
+}
+
+#ifdef __cplusplus
+}  /* extern "C" */
+#endif
+
+#endif  /* UPB_VARINT_DECODER_H_ */
+/*
+** upb::pb::Encoder (upb_pb_encoder)
+**
+** Implements a set of upb_handlers that write protobuf data to the binary wire
+** format.
+**
+** This encoder implementation does not have any access to any out-of-band or
+** precomputed lengths for submessages, so it must buffer submessages internally
+** before it can emit the first byte.
+*/
+
+#ifndef UPB_ENCODER_H_
+#define UPB_ENCODER_H_
+
+
+#ifdef __cplusplus
+namespace upb {
+namespace pb {
+class Encoder;
+}  /* namespace pb */
+}  /* namespace upb */
+#endif
+
+UPB_DECLARE_TYPE(upb::pb::Encoder, upb_pb_encoder)
+
+#define UPB_PBENCODER_MAX_NESTING 100
+
+/* upb::pb::Encoder ***********************************************************/
+
+/* Preallocation hint: decoder won't allocate more bytes than this when first
+ * constructed.  This hint may be an overestimate for some build configurations.
+ * But if the decoder library is upgraded without recompiling the application,
+ * it may be an underestimate. */
+#define UPB_PB_ENCODER_SIZE 768
+
+#ifdef __cplusplus
+
+class upb::pb::Encoder {
+ public:
+  /* Creates a new encoder in the given environment.  The Handlers must have
+   * come from NewHandlers() below. */
+  static Encoder* Create(Environment* env, const Handlers* handlers,
+                         BytesSink* output);
+
+  /* The input to the encoder. */
+  Sink* input();
+
+  /* Creates a new set of handlers for this MessageDef. */
+  static reffed_ptr<const Handlers> NewHandlers(const MessageDef* msg);
+
+  static const size_t kSize = UPB_PB_ENCODER_SIZE;
+
+ private:
+  UPB_DISALLOW_POD_OPS(Encoder, upb::pb::Encoder)
+};
+
+#endif
+
+UPB_BEGIN_EXTERN_C
+
+const upb_handlers *upb_pb_encoder_newhandlers(const upb_msgdef *m,
+                                               const void *owner);
+upb_sink *upb_pb_encoder_input(upb_pb_encoder *p);
+upb_pb_encoder* upb_pb_encoder_create(upb_env* e, const upb_handlers* h,
+                                      upb_bytessink* output);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upb {
+namespace pb {
+inline Encoder* Encoder::Create(Environment* env, const Handlers* handlers,
+                                BytesSink* output) {
+  return upb_pb_encoder_create(env, handlers, output);
+}
+inline Sink* Encoder::input() {
+  return upb_pb_encoder_input(this);
+}
+inline reffed_ptr<const Handlers> Encoder::NewHandlers(
+    const upb::MessageDef *md) {
+  const Handlers* h = upb_pb_encoder_newhandlers(md, &h);
+  return reffed_ptr<const Handlers>(h, &h);
+}
+}  /* namespace pb */
+}  /* namespace upb */
+
+#endif
+
+#endif  /* UPB_ENCODER_H_ */
+/*
+** upb's core components like upb_decoder and upb_msg are carefully designed to
+** avoid depending on each other for maximum orthogonality.  In other words,
+** you can use a upb_decoder to decode into *any* kind of structure; upb_msg is
+** just one such structure.  A upb_msg can be serialized/deserialized into any
+** format, protobuf binary format is just one such format.
+**
+** However, for convenience we provide functions here for doing common
+** operations like deserializing protobuf binary format into a upb_msg.  The
+** compromise is that this file drags in almost all of upb as a dependency,
+** which could be undesirable if you're trying to use a trimmed-down build of
+** upb.
+**
+** While these routines are convenient, they do not reuse any encoding/decoding
+** state.  For example, if a decoder is JIT-based, it will be re-JITted every
+** time these functions are called.  For this reason, if you are parsing lots
+** of data and efficiency is an issue, these may not be the best functions to
+** use (though they are useful for prototyping, before optimizing).
+*/
+
+#ifndef UPB_GLUE_H
+#define UPB_GLUE_H
+
+#include <stdbool.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Loads all defs from the given protobuf binary descriptor, setting default
+ * accessors and a default layout on all messages.  The caller owns the
+ * returned array of defs, which will be of length *n.  On error NULL is
+ * returned and status is set (if non-NULL). */
+upb_def **upb_load_defs_from_descriptor(const char *str, size_t len, int *n,
+                                        void *owner, upb_status *status);
+
+/* Like the previous but also adds the loaded defs to the given symtab. */
+bool upb_load_descriptor_into_symtab(upb_symtab *symtab, const char *str,
+                                     size_t len, upb_status *status);
+
+/* Like the previous but also reads the descriptor from the given filename. */
+bool upb_load_descriptor_file_into_symtab(upb_symtab *symtab, const char *fname,
+                                          upb_status *status);
+
+/* Reads the given filename into a character string, returning NULL if there
+ * was an error. */
+char *upb_readfile(const char *filename, size_t *len);
+
+#ifdef __cplusplus
+}  /* extern "C" */
+
+namespace upb {
+
+/* All routines that load descriptors expect the descriptor to be a
+ * FileDescriptorSet. */
+inline bool LoadDescriptorFileIntoSymtab(SymbolTable* s, const char *fname,
+                                         Status* status) {
+  return upb_load_descriptor_file_into_symtab(s, fname, status);
+}
+
+inline bool LoadDescriptorIntoSymtab(SymbolTable* s, const char* str,
+                                     size_t len, Status* status) {
+  return upb_load_descriptor_into_symtab(s, str, len, status);
+}
+
+/* Templated so it can accept both string and std::string. */
+template <typename T>
+bool LoadDescriptorIntoSymtab(SymbolTable* s, const T& desc, Status* status) {
+  return upb_load_descriptor_into_symtab(s, desc.c_str(), desc.size(), status);
+}
+
+}  /* namespace upb */
+
+#endif
+
+#endif  /* UPB_GLUE_H */
+/*
+** upb::pb::TextPrinter (upb_textprinter)
+**
+** Handlers for writing to protobuf text format.
+*/
+
+#ifndef UPB_TEXT_H_
+#define UPB_TEXT_H_
+
+
+#ifdef __cplusplus
+namespace upb {
+namespace pb {
+class TextPrinter;
+}  /* namespace pb */
+}  /* namespace upb */
+#endif
+
+UPB_DECLARE_TYPE(upb::pb::TextPrinter, upb_textprinter)
+
+#ifdef __cplusplus
+
+class upb::pb::TextPrinter {
+ public:
+  /* The given handlers must have come from NewHandlers().  It must outlive the
+   * TextPrinter. */
+  static TextPrinter *Create(Environment *env, const upb::Handlers *handlers,
+                             BytesSink *output);
+
+  void SetSingleLineMode(bool single_line);
+
+  Sink* input();
+
+  /* If handler caching becomes a requirement we can add a code cache as in
+   * decoder.h */
+  static reffed_ptr<const Handlers> NewHandlers(const MessageDef* md);
+};
+
+#endif
+
+UPB_BEGIN_EXTERN_C
+
+/* C API. */
+upb_textprinter *upb_textprinter_create(upb_env *env, const upb_handlers *h,
+                                        upb_bytessink *output);
+void upb_textprinter_setsingleline(upb_textprinter *p, bool single_line);
+upb_sink *upb_textprinter_input(upb_textprinter *p);
+
+const upb_handlers *upb_textprinter_newhandlers(const upb_msgdef *m,
+                                                const void *owner);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upb {
+namespace pb {
+inline TextPrinter *TextPrinter::Create(Environment *env,
+                                        const upb::Handlers *handlers,
+                                        BytesSink *output) {
+  return upb_textprinter_create(env, handlers, output);
+}
+inline void TextPrinter::SetSingleLineMode(bool single_line) {
+  upb_textprinter_setsingleline(this, single_line);
+}
+inline Sink* TextPrinter::input() {
+  return upb_textprinter_input(this);
+}
+inline reffed_ptr<const Handlers> TextPrinter::NewHandlers(
+    const MessageDef *md) {
+  const Handlers* h = upb_textprinter_newhandlers(md, &h);
+  return reffed_ptr<const Handlers>(h, &h);
+}
+}  /* namespace pb */
+}  /* namespace upb */
+
+#endif
+
+#endif  /* UPB_TEXT_H_ */
+/*
+** upb::json::Parser (upb_json_parser)
+**
+** Parses JSON according to a specific schema.
+** Support for parsing arbitrary JSON (schema-less) will be added later.
+*/
+
+#ifndef UPB_JSON_PARSER_H_
+#define UPB_JSON_PARSER_H_
+
+
+#ifdef __cplusplus
+namespace upb {
+namespace json {
+class Parser;
+}  /* namespace json */
+}  /* namespace upb */
+#endif
+
+UPB_DECLARE_TYPE(upb::json::Parser, upb_json_parser)
+
+/* upb::json::Parser **********************************************************/
+
+/* Preallocation hint: parser won't allocate more bytes than this when first
+ * constructed.  This hint may be an overestimate for some build configurations.
+ * But if the parser library is upgraded without recompiling the application,
+ * it may be an underestimate. */
+#define UPB_JSON_PARSER_SIZE 3704
+
+#ifdef __cplusplus
+
+/* Parses an incoming BytesStream, pushing the results to the destination
+ * sink. */
+class upb::json::Parser {
+ public:
+  static Parser* Create(Environment* env, Sink* output);
+
+  BytesSink* input();
+
+ private:
+  UPB_DISALLOW_POD_OPS(Parser, upb::json::Parser)
+};
+
+#endif
+
+UPB_BEGIN_EXTERN_C
+
+upb_json_parser *upb_json_parser_create(upb_env *e, upb_sink *output);
+upb_bytessink *upb_json_parser_input(upb_json_parser *p);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upb {
+namespace json {
+inline Parser* Parser::Create(Environment* env, Sink* output) {
+  return upb_json_parser_create(env, output);
+}
+inline BytesSink* Parser::input() {
+  return upb_json_parser_input(this);
+}
+}  /* namespace json */
+}  /* namespace upb */
+
+#endif
+
+
+#endif  /* UPB_JSON_PARSER_H_ */
+/*
+** upb::json::Printer
+**
+** Handlers that emit JSON according to a specific protobuf schema.
+*/
+
+#ifndef UPB_JSON_TYPED_PRINTER_H_
+#define UPB_JSON_TYPED_PRINTER_H_
+
+
+#ifdef __cplusplus
+namespace upb {
+namespace json {
+class Printer;
+}  /* namespace json */
+}  /* namespace upb */
+#endif
+
+UPB_DECLARE_TYPE(upb::json::Printer, upb_json_printer)
+
+
+/* upb::json::Printer *********************************************************/
+
+#define UPB_JSON_PRINTER_SIZE 168
+
+#ifdef __cplusplus
+
+/* Prints an incoming stream of data to a BytesSink in JSON format. */
+class upb::json::Printer {
+ public:
+  static Printer* Create(Environment* env, const upb::Handlers* handlers,
+                         BytesSink* output);
+
+  /* The input to the printer. */
+  Sink* input();
+
+  /* Returns handlers for printing according to the specified schema. */
+  static reffed_ptr<const Handlers> NewHandlers(const upb::MessageDef* md);
+
+  static const size_t kSize = UPB_JSON_PRINTER_SIZE;
+
+ private:
+  UPB_DISALLOW_POD_OPS(Printer, upb::json::Printer)
+};
+
+#endif
+
+UPB_BEGIN_EXTERN_C
+
+/* Native C API. */
+upb_json_printer *upb_json_printer_create(upb_env *e, const upb_handlers *h,
+                                          upb_bytessink *output);
+upb_sink *upb_json_printer_input(upb_json_printer *p);
+const upb_handlers *upb_json_printer_newhandlers(const upb_msgdef *md,
+                                                 const void *owner);
+
+UPB_END_EXTERN_C
+
+#ifdef __cplusplus
+
+namespace upb {
+namespace json {
+inline Printer* Printer::Create(Environment* env, const upb::Handlers* handlers,
+                                BytesSink* output) {
+  return upb_json_printer_create(env, handlers, output);
+}
+inline Sink* Printer::input() { return upb_json_printer_input(this); }
+inline reffed_ptr<const Handlers> Printer::NewHandlers(
+    const upb::MessageDef *md) {
+  const Handlers* h = upb_json_printer_newhandlers(md, &h);
+  return reffed_ptr<const Handlers>(h, &h);
+}
+}  /* namespace json */
+}  /* namespace upb */
+
+#endif
+
+#endif  /* UPB_JSON_TYPED_PRINTER_H_ */