Added Ruby to conformance tests.

This involved fixing a few important bugs in the
Ruby implementation -- mostly cases of mixing
upb field types and descriptor types (upb field
types do not distinguish between int/sint/fixed/sfixed
like descriptor types do).

Also added protobuf-specific exceptions so parse
errors can be caught specifically.

Change-Id: Ib49d3db976900b2c6f3455c8b88af52cfb86e036

6 files changed, 486 insertions, 540 deletions

diff --git a/ruby/ext/google/protobuf_c/defs.c b/ruby/ext/google/protobuf_c/defs.c
index 8f9f33e2..0b2f977f 100644
--- a/ruby/ext/google/protobuf_c/defs.c
+++ b/ruby/ext/google/protobuf_c/defs.c
@@ -548,7 +548,7 @@ upb_fieldtype_t ruby_to_fieldtype(VALUE type) {
 
 #define CONVERT(upb, ruby)                                           \
   if (SYM2ID(type) == rb_intern( # ruby )) {                         \
-    return UPB_TYPE_ ## upb;                                          \
+    return UPB_TYPE_ ## upb;                                         \
   }
 
   CONVERT(FLOAT, float);
@@ -589,6 +589,68 @@ VALUE fieldtype_to_ruby(upb_fieldtype_t type) {
   return Qnil;
 }
 
+upb_descriptortype_t ruby_to_descriptortype(VALUE type) {
+  if (TYPE(type) != T_SYMBOL) {
+    rb_raise(rb_eArgError, "Expected symbol for field type.");
+  }
+
+#define CONVERT(upb, ruby)                                           \
+  if (SYM2ID(type) == rb_intern( # ruby )) {                         \
+    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
+
+  rb_raise(rb_eArgError, "Unknown field type.");
+  return 0;
+}
+
+VALUE descriptortype_to_ruby(upb_descriptortype_t type) {
+  switch (type) {
+#define CONVERT(upb, ruby)                                           \
+    case UPB_DESCRIPTOR_TYPE_ ## upb : return ID2SYM(rb_intern( # ruby ));
+    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
+  }
+  return Qnil;
+}
+
 /*
  * call-seq:
  *     FieldDescriptor.type => type
@@ -604,7 +666,7 @@ VALUE FieldDescriptor_type(VALUE _self) {
   if (!upb_fielddef_typeisset(self->fielddef)) {
     return Qnil;
   }
-  return fieldtype_to_ruby(upb_fielddef_type(self->fielddef));
+  return descriptortype_to_ruby(upb_fielddef_descriptortype(self->fielddef));
 }
 
 /*
@@ -617,7 +679,7 @@ VALUE FieldDescriptor_type(VALUE _self) {
 VALUE FieldDescriptor_type_set(VALUE _self, VALUE type) {
   DEFINE_SELF(FieldDescriptor, self, _self);
   upb_fielddef* mut_def = check_field_notfrozen(self->fielddef);
-  upb_fielddef_settype(mut_def, ruby_to_fieldtype(type));
+  upb_fielddef_setdescriptortype(mut_def, ruby_to_descriptortype(type));
   return Qnil;
 }
 
diff --git a/ruby/ext/google/protobuf_c/encode_decode.c b/ruby/ext/google/protobuf_c/encode_decode.c
index f789f6d4..e488f05b 100644
--- a/ruby/ext/google/protobuf_c/encode_decode.c
+++ b/ruby/ext/google/protobuf_c/encode_decode.c
@@ -656,8 +656,10 @@ static bool env_error_func(void* ud, const upb_status* status) {
   // 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);
-  rb_raise(rb_eRuntimeError, se->ruby_error_template,
-           upb_status_errmsg(status));
+
+  // TODO(haberman): have a way to verify that this is actually a parse error,
+  // instead of just throwing "parse error" unconditionally.
+  rb_raise(cParseError, se->ruby_error_template, upb_status_errmsg(status));
   // Never reached: rb_raise() always longjmp()s up the stack, past all of our
   // code, back to Ruby.
   return false;
diff --git a/ruby/ext/google/protobuf_c/protobuf.c b/ruby/ext/google/protobuf_c/protobuf.c
index d0625a10..ca0f7b7c 100644
--- a/ruby/ext/google/protobuf_c/protobuf.c
+++ b/ruby/ext/google/protobuf_c/protobuf.c
@@ -39,6 +39,9 @@
 // Ruby integers) to MessageDef/EnumDef instances (as Ruby values).
 VALUE upb_def_to_ruby_obj_map;
 
+VALUE cError;
+VALUE cParseError;
+
 void add_def_obj(const void* def, VALUE value) {
   rb_hash_aset(upb_def_to_ruby_obj_map, ULL2NUM((intptr_t)def), value);
 }
@@ -96,6 +99,9 @@ void Init_protobuf_c() {
   RepeatedField_register(protobuf);
   Map_register(protobuf);
 
+  cError = rb_const_get(protobuf, rb_intern("Error"));
+  cParseError = rb_const_get(protobuf, rb_intern("ParseError"));
+
   rb_define_singleton_method(protobuf, "deep_copy",
                              Google_Protobuf_deep_copy, 1);
 
diff --git a/ruby/ext/google/protobuf_c/protobuf.h b/ruby/ext/google/protobuf_c/protobuf.h
index f8667486..8750c93d 100644
--- a/ruby/ext/google/protobuf_c/protobuf.h
+++ b/ruby/ext/google/protobuf_c/protobuf.h
@@ -161,6 +161,9 @@ extern VALUE cOneofBuilderContext;
 extern VALUE cEnumBuilderContext;
 extern VALUE cBuilder;
 
+extern VALUE cError;
+extern VALUE cParseError;
+
 // We forward-declare all of the Ruby method implementations here because we
 // sometimes call the methods directly across .c files, rather than going
 // through Ruby's method dispatching (e.g. during message parse). It's cleaner
diff --git a/ruby/ext/google/protobuf_c/upb.c b/ruby/ext/google/protobuf_c/upb.c
index f99c7a70..8ef8e31d 100644
--- a/ruby/ext/google/protobuf_c/upb.c
+++ b/ruby/ext/google/protobuf_c/upb.c
@@ -1,11 +1,5 @@
 // Amalgamated source file
 #include "upb.h"
-/*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2008-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
 
 
 #include <stdlib.h>
@@ -1701,12 +1695,6 @@ upb_fielddef *upb_oneof_iter_field(const upb_oneof_iter *iter) {
 void upb_oneof_iter_setdone(upb_oneof_iter *iter) {
   upb_inttable_iter_setdone(iter);
 }
-/*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
 
 
 #include <stdlib.h>
@@ -1980,14 +1968,9 @@ upb_alloc_func *upb_seededalloc_getallocfunc(upb_seededalloc *a) {
   return seeded_alloc;
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2011-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * TODO(haberman): it's unclear whether a lot of the consistency checks should
- * assert() or return false.
- */
+** TODO(haberman): it's unclear whether a lot of the consistency checks should
+** assert() or return false.
+*/
 
 
 #include <stdlib.h>
@@ -2668,24 +2651,21 @@ bool upb_byteshandler_setendstr(upb_byteshandler *h,
   return true;
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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>
@@ -3514,12 +3494,6 @@ bool upb_refcounted_freeze(upb_refcounted *const*roots, int n, upb_status *s,
   }
   return freeze(roots, n, s, maxdepth);
 }
-/*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2013 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
 
 
 #include <stdlib.h>
@@ -3605,12 +3579,6 @@ const upb_shim_data *upb_shim_getdata(const upb_handlers *h, upb_selector_t s,
 
   return (const upb_shim_data*)upb_handlers_gethandlerdata(h, s);
 }
-/*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2008-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
 
 
 #include <stdlib.h>
@@ -4041,13 +4009,10 @@ const upb_def *upb_symtab_iter_def(const upb_symtab_iter *iter) {
   return upb_value_getptr(upb_strtable_iter_value(&iter->iter));
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * Implementation is heavily inspired by Lua's ltable.c.
- */
+** upb_table Implementation
+**
+** Implementation is heavily inspired by Lua's ltable.c.
+*/
 
 
 #include <stdlib.h>
@@ -4931,12 +4896,6 @@ uint32_t MurmurHash2(const void * key, size_t len, uint32_t seed) {
 #undef MIX
 
 #endif /* UPB_UNALIGNED_READS_OK */
-/*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
 
 #include <errno.h>
 #include <stdarg.h>
@@ -5860,17 +5819,12 @@ static upb_inttable reftables[212] = {
 #endif
 
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2008-2009 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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>
@@ -6518,21 +6472,18 @@ const upb_handlers *upb_descreader_newhandlers(const void *owner) {
   return h;
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2013 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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>
 
@@ -7502,24 +7453,19 @@ void upb_pbdecodermethodopts_setlazy(upb_pbdecodermethodopts *opts, bool lazy) {
   opts->lazy = lazy;
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2008-2013 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * This file implements a VM for the interpreted (bytecode) decoder.
- *
- * 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.
- */
+** 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>
@@ -8529,63 +8475,60 @@ bool upb_pbdecoder_setmaxnesting(upb_pbdecoder *d, size_t max) {
   return true;
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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>
@@ -9095,12 +9038,6 @@ upb_pb_encoder *upb_pb_encoder_create(upb_env *env, const upb_handlers *h,
 }
 
 upb_sink *upb_pb_encoder_input(upb_pb_encoder *e) { return &e->input_; }
-/*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2010-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
 
 
 #include <stdio.h>
@@ -9189,10 +9126,7 @@ bool upb_load_descriptor_file_into_symtab(upb_symtab *symtab, const char *fname,
   return success;
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
+ * 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.
@@ -9529,12 +9463,6 @@ 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;
 }
-/*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2011 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
 
 
 /* Index is descriptor type. */
@@ -9662,28 +9590,25 @@ upb_decoderet upb_vdecode_max8_wright(upb_decoderet r) {
 
 #line 1 "upb/json/parser.rl"
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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>
@@ -9731,7 +9656,7 @@ struct upb_json_parser {
   upb_jsonparser_frame *top;
   upb_jsonparser_frame *limit;
 
-  upb_status *status;
+  upb_status status;
 
   /* Ragel's internal parsing stack for the parsing state machine. */
   int current_state;
@@ -9778,7 +9703,8 @@ static upb_selector_t parser_getsel(upb_json_parser *p) {
 
 static bool check_stack(upb_json_parser *p) {
   if ((p->top + 1) == p->limit) {
-    upb_status_seterrmsg(p->status, "Nesting too deep");
+    upb_status_seterrmsg(&p->status, "Nesting too deep");
+    upb_env_reporterror(p->env, &p->status);
     return false;
   }
 
@@ -9860,9 +9786,10 @@ static bool base64_push(upb_json_parser *p, upb_selector_t sel, const char *ptr,
     char output[3];
 
     if (limit - ptr < 4) {
-      upb_status_seterrf(p->status,
+      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;
     }
 
@@ -9886,9 +9813,10 @@ static bool base64_push(upb_json_parser *p, upb_selector_t sel, const char *ptr,
 otherchar:
   if (nonbase64(ptr[0]) || nonbase64(ptr[1]) || nonbase64(ptr[2]) ||
       nonbase64(ptr[3]) ) {
-    upb_status_seterrf(p->status,
+    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;
@@ -9926,10 +9854,11 @@ otherchar:
   }
 
 badpadding:
-  upb_status_seterrf(p->status,
+  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;
 }
 
@@ -9976,7 +9905,8 @@ static bool accumulate_realloc(upb_json_parser *p, size_t need) {
 
   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_status_seterrmsg(&p->status, "Out of memory allocating buffer.");
+    upb_env_reporterror(p->env, &p->status);
     return false;
   }
 
@@ -9999,7 +9929,8 @@ static bool accumulate_append(upb_json_parser *p, const char *buf, size_t len,
   }
 
   if (!checked_add(p->accumulated_len, len, &need)) {
-    upb_status_seterrmsg(p->status, "Integer overflow.");
+    upb_status_seterrmsg(&p->status, "Integer overflow.");
+    upb_env_reporterror(p->env, &p->status);
     return false;
   }
 
@@ -10077,7 +10008,8 @@ static bool multipart_text(upb_json_parser *p, const char *buf, size_t len,
   switch (p->multipart_state) {
     case MULTIPART_INACTIVE:
       upb_status_seterrmsg(
-          p->status, "Internal error: unexpected state MULTIPART_INACTIVE");
+          &p->status, "Internal error: unexpected state MULTIPART_INACTIVE");
+      upb_env_reporterror(p->env, &p->status);
       return false;
 
     case MULTIPART_ACCUMULATE:
@@ -10336,7 +10268,8 @@ static bool parse_number(upb_json_parser *p) {
   return true;
 
 err:
-  upb_status_seterrf(p->status, "error parsing number: %s", buf);
+  upb_status_seterrf(&p->status, "error parsing number: %s", buf);
+  upb_env_reporterror(p->env, &p->status);
   multipart_end(p);
   return false;
 }
@@ -10345,9 +10278,10 @@ 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,
+    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;
   }
 
@@ -10398,9 +10332,10 @@ static bool start_stringval(upb_json_parser *p) {
     multipart_startaccum(p);
     return true;
   } else {
-    upb_status_seterrf(p->status,
+    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;
   }
 }
@@ -10438,7 +10373,8 @@ static bool end_stringval(upb_json_parser *p) {
         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_status_seterrf(&p->status, "Enum value unknown: '%.*s'", len, buf);
+        upb_env_reporterror(p->env, &p->status);
       }
 
       break;
@@ -10446,7 +10382,8 @@ static bool end_stringval(upb_json_parser *p) {
 
     default:
       assert(false);
-      upb_status_seterrmsg(p->status, "Internal error in JSON decoder");
+      upb_status_seterrmsg(&p->status, "Internal error in JSON decoder");
+      upb_env_reporterror(p->env, &p->status);
       ok = false;
       break;
   }
@@ -10476,7 +10413,8 @@ static bool parse_mapentry_key(upb_json_parser *p) {
 
   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_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)) {
@@ -10499,8 +10437,9 @@ static bool parse_mapentry_key(upb_json_parser *p) {
           return false;
         }
       } else {
-        upb_status_seterrmsg(p->status,
+        upb_status_seterrmsg(&p->status,
                              "Map bool key not 'true' or 'false'");
+        upb_env_reporterror(p->env, &p->status);
         return false;
       }
       multipart_end(p);
@@ -10518,7 +10457,8 @@ static bool parse_mapentry_key(upb_json_parser *p) {
       break;
     }
     default:
-      upb_status_seterrmsg(p->status, "Invalid field type for map key");
+      upb_status_seterrmsg(&p->status, "Invalid field type for map key");
+      upb_env_reporterror(p->env, &p->status);
       return false;
   }
 
@@ -10573,7 +10513,8 @@ static bool handle_mapentry(upb_json_parser *p) {
   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_status_seterrmsg(&p->status, "mapentry message has no value");
+    upb_env_reporterror(p->env, &p->status);
     return false;
   }
 
@@ -10593,7 +10534,8 @@ static bool end_membername(upb_json_parser *p) {
     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_status_seterrf(&p->status, "No such field: %.*s\n", (int)len, buf);
+      upb_env_reporterror(p->env, &p->status);
       return false;
     }
 
@@ -10669,9 +10611,10 @@ static bool start_subobject(upb_json_parser *p) {
 
     return true;
   } else {
-    upb_status_seterrf(p->status,
+    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;
   }
 }
@@ -10697,9 +10640,10 @@ static bool start_array(upb_json_parser *p) {
   assert(p->top->f);
 
   if (!upb_fielddef_isseq(p->top->f)) {
-    upb_status_seterrf(p->status,
+    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;
   }
 
@@ -10736,7 +10680,11 @@ static void start_object(upb_json_parser *p) {
 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);
+    }
   }
 }
 
@@ -10762,11 +10710,11 @@ static void end_object(upb_json_parser *p) {
  * final state once, when the closing '"' is seen. */
 
 
-#line 1198 "upb/json/parser.rl"
+#line 1218 "upb/json/parser.rl"
 
 
 
-#line 1110 "upb/json/parser.c"
+#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, 
@@ -10915,7 +10863,7 @@ static const int json_en_value_machine = 27;
 static const int json_en_main = 1;
 
 
-#line 1201 "upb/json/parser.rl"
+#line 1221 "upb/json/parser.rl"
 
 size_t parse(void *closure, const void *hd, const char *buf, size_t size,
              const upb_bufhandle *handle) {
@@ -10937,7 +10885,7 @@ size_t parse(void *closure, const void *hd, const char *buf, size_t size,
   capture_resume(parser, buf);
 
   
-#line 1281 "upb/json/parser.c"
+#line 1301 "upb/json/parser.c"
 	{
 	int _klen;
 	unsigned int _trans;
@@ -11012,118 +10960,118 @@ _match:
 		switch ( *_acts++ )
 		{
 	case 0:
-#line 1113 "upb/json/parser.rl"
+#line 1133 "upb/json/parser.rl"
 	{ p--; {cs = stack[--top]; goto _again;} }
 	break;
 	case 1:
-#line 1114 "upb/json/parser.rl"
+#line 1134 "upb/json/parser.rl"
 	{ p--; {stack[top++] = cs; cs = 10; goto _again;} }
 	break;
 	case 2:
-#line 1118 "upb/json/parser.rl"
+#line 1138 "upb/json/parser.rl"
 	{ start_text(parser, p); }
 	break;
 	case 3:
-#line 1119 "upb/json/parser.rl"
+#line 1139 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(end_text(parser, p)); }
 	break;
 	case 4:
-#line 1125 "upb/json/parser.rl"
+#line 1145 "upb/json/parser.rl"
 	{ start_hex(parser); }
 	break;
 	case 5:
-#line 1126 "upb/json/parser.rl"
+#line 1146 "upb/json/parser.rl"
 	{ hexdigit(parser, p); }
 	break;
 	case 6:
-#line 1127 "upb/json/parser.rl"
+#line 1147 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(end_hex(parser)); }
 	break;
 	case 7:
-#line 1133 "upb/json/parser.rl"
+#line 1153 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(escape(parser, p)); }
 	break;
 	case 8:
-#line 1139 "upb/json/parser.rl"
+#line 1159 "upb/json/parser.rl"
 	{ p--; {cs = stack[--top]; goto _again;} }
 	break;
 	case 9:
-#line 1142 "upb/json/parser.rl"
+#line 1162 "upb/json/parser.rl"
 	{ {stack[top++] = cs; cs = 19; goto _again;} }
 	break;
 	case 10:
-#line 1144 "upb/json/parser.rl"
+#line 1164 "upb/json/parser.rl"
 	{ p--; {stack[top++] = cs; cs = 27; goto _again;} }
 	break;
 	case 11:
-#line 1149 "upb/json/parser.rl"
+#line 1169 "upb/json/parser.rl"
 	{ start_member(parser); }
 	break;
 	case 12:
-#line 1150 "upb/json/parser.rl"
+#line 1170 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(end_membername(parser)); }
 	break;
 	case 13:
-#line 1153 "upb/json/parser.rl"
+#line 1173 "upb/json/parser.rl"
 	{ end_member(parser); }
 	break;
 	case 14:
-#line 1159 "upb/json/parser.rl"
+#line 1179 "upb/json/parser.rl"
 	{ start_object(parser); }
 	break;
 	case 15:
-#line 1162 "upb/json/parser.rl"
+#line 1182 "upb/json/parser.rl"
 	{ end_object(parser); }
 	break;
 	case 16:
-#line 1168 "upb/json/parser.rl"
+#line 1188 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(start_array(parser)); }
 	break;
 	case 17:
-#line 1172 "upb/json/parser.rl"
+#line 1192 "upb/json/parser.rl"
 	{ end_array(parser); }
 	break;
 	case 18:
-#line 1177 "upb/json/parser.rl"
+#line 1197 "upb/json/parser.rl"
 	{ start_number(parser, p); }
 	break;
 	case 19:
-#line 1178 "upb/json/parser.rl"
+#line 1198 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(end_number(parser, p)); }
 	break;
 	case 20:
-#line 1180 "upb/json/parser.rl"
+#line 1200 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(start_stringval(parser)); }
 	break;
 	case 21:
-#line 1181 "upb/json/parser.rl"
+#line 1201 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(end_stringval(parser)); }
 	break;
 	case 22:
-#line 1183 "upb/json/parser.rl"
+#line 1203 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(parser_putbool(parser, true)); }
 	break;
 	case 23:
-#line 1185 "upb/json/parser.rl"
+#line 1205 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(parser_putbool(parser, false)); }
 	break;
 	case 24:
-#line 1187 "upb/json/parser.rl"
+#line 1207 "upb/json/parser.rl"
 	{ /* null value */ }
 	break;
 	case 25:
-#line 1189 "upb/json/parser.rl"
+#line 1209 "upb/json/parser.rl"
 	{ CHECK_RETURN_TOP(start_subobject(parser)); }
 	break;
 	case 26:
-#line 1190 "upb/json/parser.rl"
+#line 1210 "upb/json/parser.rl"
 	{ end_subobject(parser); }
 	break;
 	case 27:
-#line 1195 "upb/json/parser.rl"
+#line 1215 "upb/json/parser.rl"
 	{ p--; {cs = stack[--top]; goto _again;} }
 	break;
-#line 1467 "upb/json/parser.c"
+#line 1487 "upb/json/parser.c"
 		}
 	}
 
@@ -11136,10 +11084,11 @@ _again:
 	_out: {}
 	}
 
-#line 1222 "upb/json/parser.rl"
+#line 1242 "upb/json/parser.rl"
 
   if (p != pe) {
-    upb_status_seterrf(parser->status, "Parse error at %s\n", p);
+    upb_status_seterrf(&parser->status, "Parse error at %s\n", p);
+    upb_env_reporterror(parser->env, &parser->status);
   } else {
     capture_suspend(parser, &p);
   }
@@ -11176,19 +11125,20 @@ static void json_parser_reset(upb_json_parser *p) {
 
   /* Emit Ragel initialization of the parser. */
   
-#line 1520 "upb/json/parser.c"
+#line 1541 "upb/json/parser.c"
 	{
 	cs = json_start;
 	top = 0;
 	}
 
-#line 1261 "upb/json/parser.rl"
+#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);
 }
 
 
@@ -11214,8 +11164,8 @@ upb_json_parser *upb_json_parser_create(upb_env *env, upb_sink *output) {
   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); */
+  /* 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;
 }
@@ -11224,14 +11174,9 @@ upb_bytessink *upb_json_parser_input(upb_json_parser *p) {
   return &p->input_;
 }
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * This currently uses snprintf() to format primitives, and could be optimized
- * further.
- */
+** This currently uses snprintf() to format primitives, and could be optimized
+** further.
+*/
 
 
 #include <stdlib.h>
diff --git a/ruby/ext/google/protobuf_c/upb.h b/ruby/ext/google/protobuf_c/upb.h
index b4dcd558..b31f5c08 100644
--- a/ruby/ext/google/protobuf_c/upb.h
+++ b/ruby/ext/google/protobuf_c/upb.h
@@ -1,70 +1,62 @@
 // Amalgamated source file
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * Defs are upb's internal representation of the constructs that can appear
- * in a .proto file:
- *
- * - upb_msgdef: describes a "message" construct.
- * - upb_fielddef: describes a message field.
- * - upb_enumdef: describes an enum.
- * (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.
- */
+** 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 - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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 - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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_
@@ -73,16 +65,11 @@
 #include <stdint.h>
 #include <string.h>
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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_
@@ -3006,25 +2993,20 @@ inline bool OneofDef::const_iterator::operator!=(
 
 #endif /* UPB_DEF_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2015 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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_
@@ -3181,25 +3163,22 @@ struct upb_symtab {
 
 #endif  /* UPB_STATICINIT_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2010-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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
@@ -3980,14 +3959,9 @@ uint32_t upb_handlers_selectorcount(const upb_fielddef *f);
 UPB_END_EXTERN_C
 
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2011-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * Inline definitions for handlers.h, which are particularly long and a bit
- * tricky.
- */
+** Inline definitions for handlers.h, which are particularly long and a bit
+** tricky.
+*/
 
 #ifndef UPB_HANDLERS_INL_H_
 #define UPB_HANDLERS_INL_H_
@@ -5128,21 +5102,18 @@ inline BytesHandler::~BytesHandler() {}
 
 #endif  /* UPB_HANDLERS_H */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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_
@@ -5392,23 +5363,21 @@ inline upb_alloc_func *SeededAllocator::GetAllocationFunction() {
 
 #endif  /* UPB_ENV_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2010-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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).
- */
+** 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
@@ -5921,21 +5890,16 @@ inline bool BufferSource::PutBuffer(const char *buf, size_t len,
 
 #endif
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2013 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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
@@ -5994,19 +5958,16 @@ inline const Shim::Data* Shim::GetData(const Handlers* h, Handlers::Selector s,
 
 #endif  /* UPB_SHIM_H */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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_
@@ -6182,14 +6143,10 @@ inline bool SymbolTable::Add(
 
 #endif  /* UPB_SYMTAB_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2011 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * upb::descriptor::Reader provides a way of building upb::Defs from
- * data in descriptor.proto format.
- */
+** 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
@@ -7067,34 +7024,26 @@ inline upb::reffed_ptr<const upb::FieldDef> name_part() { RETURN_REFFED(upb::Fie
 
 #endif  /* GOOGLE_PROTOBUF_DESCRIPTOR_UPB_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009-2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * Internal-only definitions for the decoder.
- */
+** Internal-only definitions for the decoder.
+*/
 
 #ifndef UPB_DECODER_INT_H_
 #define UPB_DECODER_INT_H_
 
 #include <stdlib.h>
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009-2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * upb::pb::Decoder implements 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.
- */
+** 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_
@@ -7702,14 +7651,9 @@ UPB_INLINE void upb_pbdecoder_unpackdispatch(uint64_t dispatch, uint64_t *ofs,
 
 #endif  /* UPB_DECODER_INT_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2011 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * A number of routines for varint manipulation (we keep them all around to
- * have multiple approaches available for benchmarking).
- */
+** 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_
@@ -7873,18 +7817,15 @@ UPB_INLINE uint64_t upb_vencode32(uint32_t val) {
 
 #endif  /* UPB_VARINT_DECODER_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009-2010 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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.
- */
+** 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_
@@ -7966,29 +7907,24 @@ inline reffed_ptr<const Handlers> Encoder::NewHandlers(
 
 #endif  /* UPB_ENCODER_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2011-2012 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * 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).
- */
+** 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
@@ -8047,11 +7983,10 @@ bool LoadDescriptorIntoSymtab(SymbolTable* s, const T& desc, Status* status) {
 
 #endif  /* UPB_GLUE_H */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2009 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- */
+** upb::pb::TextPrinter (upb_textprinter)
+**
+** Handlers for writing to protobuf text format.
+*/
 
 #ifndef UPB_TEXT_H_
 #define UPB_TEXT_H_
@@ -8127,14 +8062,11 @@ inline reffed_ptr<const Handlers> TextPrinter::NewHandlers(
 
 #endif  /* UPB_TEXT_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * upb::json::Parser can parse JSON according to a specific schema.
- * Support for parsing arbitrary JSON (schema-less) will be added later.
- */
+** 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_
@@ -8156,7 +8088,7 @@ UPB_DECLARE_TYPE(upb::json::Parser, upb_json_parser)
  * 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 3568
+#define UPB_JSON_PARSER_SIZE 3704
 
 #ifdef __cplusplus
 
@@ -8199,14 +8131,10 @@ inline BytesSink* Parser::input() {
 
 #endif  /* UPB_JSON_PARSER_H_ */
 /*
- * upb - a minimalist implementation of protocol buffers.
- *
- * Copyright (c) 2014 Google Inc.  See LICENSE for details.
- * Author: Josh Haberman <jhaberman@gmail.com>
- *
- * upb::json::Printer allows you to create handlers that emit JSON
- * according to a specific protobuf schema.
- */
+** 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_