Initial checkin.

64 files changed, 16449 insertions, 0 deletions

diff --git a/src/google/protobuf/compiler/code_generator.cc b/src/google/protobuf/compiler/code_generator.cc
new file mode 100644
index 00000000..d3a051d0
--- /dev/null
+++ b/src/google/protobuf/compiler/code_generator.cc
@@ -0,0 +1,32 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/code_generator.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+
+CodeGenerator::~CodeGenerator() {}
+OutputDirectory::~OutputDirectory() {}
+
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/code_generator.h b/src/google/protobuf/compiler/code_generator.h
new file mode 100644
index 00000000..8f9938e3
--- /dev/null
+++ b/src/google/protobuf/compiler/code_generator.h
@@ -0,0 +1,98 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// Defines the abstract interface implemented by each of the language-specific
+// code generators.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CODE_GENERATOR_H__
+#define GOOGLE_PROTOBUF_COMPILER_CODE_GENERATOR_H__
+
+#include <google/protobuf/stubs/common.h>
+#include <string>
+
+namespace google {
+namespace protobuf {
+
+namespace io { class ZeroCopyOutputStream; }
+class FileDescriptor;
+
+namespace compiler {
+
+// Defined in this file.
+class CodeGenerator;
+class OutputDirectory;
+
+// The abstract interface to a class which generates code implementing a
+// particular proto file in a particular language.  A number of these may
+// be registered with CommandLineInterface to support various languages.
+class LIBPROTOC_EXPORT CodeGenerator {
+ public:
+  inline CodeGenerator() {}
+  virtual ~CodeGenerator();
+
+  // Generates code for the given proto file, generating one or more files in
+  // the given output directory.
+  //
+  // A parameter to be passed to the generator can be specified on the
+  // command line.  This is intended to be used by Java and similar languages
+  // to specify which specific class from the proto file is to be generated,
+  // though it could have other uses as well.  It is empty if no parameter was
+  // given.
+  //
+  // Returns true if successful.  Otherwise, sets *error to a description of
+  // the problem (e.g. "invalid parameter") and returns false.
+  virtual bool Generate(const FileDescriptor* file,
+                        const string& parameter,
+                        OutputDirectory* output_directory,
+                        string* error) const = 0;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(CodeGenerator);
+};
+
+// CodeGenerators generate one or more files in a given directory.  This
+// abstract interface represents the directory to which the CodeGenerator is
+// to write.
+class LIBPROTOC_EXPORT OutputDirectory {
+ public:
+  inline OutputDirectory() {}
+  virtual ~OutputDirectory();
+
+  // Opens the given file, truncating it if it exists, and returns a
+  // ZeroCopyOutputStream that writes to the file.  The caller takes ownership
+  // of the returned object.  This method never fails (a dummy stream will be
+  // returned instead).
+  //
+  // The filename given should be relative to the root of the source tree.
+  // E.g. the C++ generator, when generating code for "foo/bar.proto", will
+  // generate the files "foo/bar.pb2.h" and "foo/bar.pb2.cc"; note that
+  // "foo/" is included in these filenames.  The filename is not allowed to
+  // contain "." or ".." components.
+  virtual io::ZeroCopyOutputStream* Open(const string& filename) = 0;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(OutputDirectory);
+};
+
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CODE_GENERATOR_H__
diff --git a/src/google/protobuf/compiler/command_line_interface.cc b/src/google/protobuf/compiler/command_line_interface.cc
new file mode 100644
index 00000000..68e88a8e
--- /dev/null
+++ b/src/google/protobuf/compiler/command_line_interface.cc
@@ -0,0 +1,579 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#ifdef _MSC_VER
+#include <io.h>
+#include <direct.h>
+#else
+#include <unistd.h>
+#endif
+#include <errno.h>
+#include <iostream>
+
+#include <google/protobuf/compiler/command_line_interface.h>
+#include <google/protobuf/compiler/importer.h>
+#include <google/protobuf/compiler/code_generator.h>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/io/zero_copy_stream_impl.h>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/stubs/strutil.h>
+
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+
+#if defined(_WIN32)
+#define mkdir(name, mode) mkdir(name)
+#ifndef W_OK
+#define W_OK 02  // not defined by MSVC for whatever reason
+#endif
+#ifndef F_OK
+#define F_OK 00  // not defined by MSVC for whatever reason
+#endif
+#endif
+
+#ifndef O_BINARY
+#ifdef _O_BINARY
+#define O_BINARY _O_BINARY
+#else
+#define O_BINARY 0     // If this isn't defined, the platform doesn't need it.
+#endif
+#endif
+
+namespace {
+#if defined(_WIN32) && !defined(__CYGWIN__)
+static const char* kPathSeparator = ";";
+#else
+static const char* kPathSeparator = ":";
+#endif
+}  // namespace
+
+// A MultiFileErrorCollector that prints errors to stderr.
+class CommandLineInterface::ErrorPrinter : public MultiFileErrorCollector {
+ public:
+  ErrorPrinter() {}
+  ~ErrorPrinter() {}
+
+  // implements MultiFileErrorCollector ------------------------------
+  void AddError(const string& filename, int line, int column,
+                const string& message) {
+    // Users typically expect 1-based line/column numbers, so we add 1
+    // to each here.
+    cerr << filename;
+    if (line != -1) {
+      cerr << ":" << (line + 1) << ":" << (column + 1);
+    }
+    cerr << ": " << message << endl;
+  }
+};
+
+// -------------------------------------------------------------------
+
+// An OutputDirectory implementation that writes to disk.
+class CommandLineInterface::DiskOutputDirectory : public OutputDirectory {
+ public:
+  DiskOutputDirectory(const string& root);
+  ~DiskOutputDirectory();
+
+  bool VerifyExistence();
+
+  inline bool had_error() { return had_error_; }
+  inline void set_had_error(bool value) { had_error_ = value; }
+
+  // implements OutputDirectory --------------------------------------
+  io::ZeroCopyOutputStream* Open(const string& filename);
+
+ private:
+  string root_;
+  bool had_error_;
+};
+
+// A FileOutputStream that checks for errors in the destructor and reports
+// them.  We extend FileOutputStream via wrapping rather than inheritance
+// for two reasons:
+// 1) Implementation inheritance is evil.
+// 2) We need to close the file descriptor *after* the FileOutputStream's
+//    destructor is run to make sure it flushes the file contents.
+class CommandLineInterface::ErrorReportingFileOutput
+    : public io::ZeroCopyOutputStream {
+ public:
+  ErrorReportingFileOutput(int file_descriptor,
+                           const string& filename,
+                           DiskOutputDirectory* directory);
+  ~ErrorReportingFileOutput();
+
+  // implements ZeroCopyOutputStream ---------------------------------
+  bool Next(void** data, int* size) { return file_stream_->Next(data, size); }
+  void BackUp(int count)            {        file_stream_->BackUp(count);    }
+  int64 ByteCount() const           { return file_stream_->ByteCount();      }
+
+ private:
+  scoped_ptr<io::FileOutputStream> file_stream_;
+  int file_descriptor_;
+  string filename_;
+  DiskOutputDirectory* directory_;
+};
+
+// -------------------------------------------------------------------
+
+CommandLineInterface::DiskOutputDirectory::DiskOutputDirectory(
+    const string& root)
+  : root_(root), had_error_(false) {
+  // Add a '/' to the end if it doesn't already have one.  But don't add a
+  // '/' to an empty string since this probably means the current directory.
+  if (!root_.empty() && root[root_.size() - 1] != '/') {
+    root_ += '/';
+  }
+}
+
+CommandLineInterface::DiskOutputDirectory::~DiskOutputDirectory() {
+}
+
+bool CommandLineInterface::DiskOutputDirectory::VerifyExistence() {
+  if (!root_.empty()) {
+    // Make sure the directory exists.  If it isn't a directory, this will fail
+    // because we added a '/' to the end of the name in the constructor.
+    if (access(root_.c_str(), W_OK) == -1) {
+      cerr << root_ << ": " << strerror(errno) << endl;
+      return false;
+    }
+  }
+
+  return true;
+}
+
+io::ZeroCopyOutputStream* CommandLineInterface::DiskOutputDirectory::Open(
+    const string& filename) {
+  // Recursively create parent directories to the output file.
+  vector<string> parts;
+  SplitStringUsing(filename, "/", &parts);
+  string path_so_far = root_;
+  for (int i = 0; i < parts.size() - 1; i++) {
+    path_so_far += parts[i];
+    if (mkdir(path_so_far.c_str(), 0777) != 0) {
+      if (errno != EEXIST) {
+        cerr << filename << ": while trying to create directory "
+             << path_so_far << ": " << strerror(errno) << endl;
+        had_error_ = true;
+        // Return a dummy stream.
+        return new io::ArrayOutputStream(NULL, 0);
+      }
+    }
+    path_so_far += '/';
+  }
+
+  // Create the output file.
+  int file_descriptor;
+  do {
+    file_descriptor =
+      open((root_ + filename).c_str(),
+           O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0777);
+  } while (file_descriptor < 0 && errno == EINTR);
+
+  if (file_descriptor < 0) {
+    // Failed to open.
+    cerr << filename << ": " << strerror(errno) << endl;
+    had_error_ = true;
+    // Return a dummy stream.
+    return new io::ArrayOutputStream(NULL, 0);
+  }
+
+  return new ErrorReportingFileOutput(file_descriptor, filename, this);
+}
+
+CommandLineInterface::ErrorReportingFileOutput::ErrorReportingFileOutput(
+    int file_descriptor,
+    const string& filename,
+    DiskOutputDirectory* directory)
+  : file_stream_(new io::FileOutputStream(file_descriptor)),
+    file_descriptor_(file_descriptor),
+    filename_(filename),
+    directory_(directory) {}
+
+CommandLineInterface::ErrorReportingFileOutput::~ErrorReportingFileOutput() {
+  // Check if we had any errors while writing.
+  if (file_stream_->GetErrno() != 0) {
+    cerr << filename_ << ": " << strerror(file_stream_->GetErrno()) << endl;
+    directory_->set_had_error(true);
+  }
+
+  // Close the file stream.
+  if (!file_stream_->Close()) {
+    cerr << filename_ << ": " << strerror(file_stream_->GetErrno()) << endl;
+    directory_->set_had_error(true);
+  }
+}
+
+// ===================================================================
+
+CommandLineInterface::CommandLineInterface()
+  : disallow_services_(false),
+    inputs_are_proto_path_relative_(false) {}
+CommandLineInterface::~CommandLineInterface() {}
+
+void CommandLineInterface::RegisterGenerator(const string& flag_name,
+                                             CodeGenerator* generator,
+                                             const string& help_text) {
+  GeneratorInfo info;
+  info.generator = generator;
+  info.help_text = help_text;
+  generators_[flag_name] = info;
+}
+
+int CommandLineInterface::Run(int argc, const char* const argv[]) {
+  Clear();
+  if (!ParseArguments(argc, argv)) return -1;
+
+  // Set up the source tree.
+  DiskSourceTree source_tree;
+  for (int i = 0; i < proto_path_.size(); i++) {
+    source_tree.MapPath(proto_path_[i].first, proto_path_[i].second);
+  }
+
+  // Map input files to virtual paths if necessary.
+  if (!inputs_are_proto_path_relative_) {
+    if (!MakeInputsBeProtoPathRelative(&source_tree)) {
+      return -1;
+    }
+  }
+
+  // Allocate the Importer.
+  ErrorPrinter error_collector;
+  DescriptorPool pool;
+  Importer importer(&source_tree, &error_collector);
+
+  // Parse each file and generate output.
+  for (int i = 0; i < input_files_.size(); i++) {
+    // Import the file.
+    const FileDescriptor* parsed_file = importer.Import(input_files_[i]);
+    if (parsed_file == NULL) return -1;
+
+    // Enforce --disallow_services.
+    if (disallow_services_ && parsed_file->service_count() > 0) {
+      cerr << parsed_file->name() << ": This file contains services, but "
+              "--disallow_services was used." << endl;
+      return -1;
+    }
+
+    // Generate output files.
+    for (int i = 0; i < output_directives_.size(); i++) {
+      if (!GenerateOutput(parsed_file, output_directives_[i])) {
+        return -1;
+      }
+    }
+  }
+
+  return 0;
+}
+
+void CommandLineInterface::Clear() {
+  proto_path_.clear();
+  input_files_.clear();
+  output_directives_.clear();
+}
+
+bool CommandLineInterface::MakeInputsBeProtoPathRelative(
+    DiskSourceTree* source_tree) {
+  for (int i = 0; i < input_files_.size(); i++) {
+    string virtual_file, shadowing_disk_file;
+    switch (source_tree->DiskFileToVirtualFile(
+        input_files_[i], &virtual_file, &shadowing_disk_file)) {
+      case DiskSourceTree::SUCCESS:
+        input_files_[i] = virtual_file;
+        break;
+      case DiskSourceTree::SHADOWED:
+        cerr << input_files_[i] << ": Input is shadowed in the --proto_path "
+                "by \"" << shadowing_disk_file << "\".  Either use the latter "
+                "file as your input or reorder the --proto_path so that the "
+                "former file's location comes first." << endl;
+        return false;
+      case DiskSourceTree::CANNOT_OPEN:
+        cerr << input_files_[i] << ": " << strerror(errno) << endl;
+        return false;
+      case DiskSourceTree::NO_MAPPING:
+        // First check if the file exists at all.
+        if (access(input_files_[i].c_str(), F_OK) < 0) {
+          // File does not even exist.
+          cerr << input_files_[i] << ": " << strerror(ENOENT) << endl;
+        } else {
+          cerr << input_files_[i] << ": File does not reside within any path "
+                  "specified using --proto_path (or -I).  You must specify a "
+                  "--proto_path which encompasses this file." << endl;
+        }
+        return false;
+    }
+  }
+
+  return true;
+}
+
+bool CommandLineInterface::ParseArguments(int argc, const char* const argv[]) {
+  executable_name_ = argv[0];
+
+  // Iterate through all arguments and parse them.
+  for (int i = 1; i < argc; i++) {
+    string name, value;
+
+    if (ParseArgument(argv[i], &name, &value)) {
+      // Retured true => Use the next argument as the flag value.
+      if (i + 1 == argc || argv[i+1][0] == '-') {
+        cerr << "Missing value for flag: " << name << endl;
+        return false;
+      } else {
+        ++i;
+        value = argv[i];
+      }
+    }
+
+    if (!InterpretArgument(name, value)) return false;
+  }
+
+  // If no --proto_path was given, use the current working directory.
+  if (proto_path_.empty()) {
+    proto_path_.push_back(make_pair("", "."));
+  }
+
+  // Check some errror cases.
+  if (input_files_.empty()) {
+    cerr << "Missing input file." << endl;
+    return false;
+  }
+  if (output_directives_.empty()) {
+    cerr << "Missing output directives." << endl;
+    return false;
+  }
+
+  return true;
+}
+
+bool CommandLineInterface::ParseArgument(const char* arg,
+                                         string* name, string* value) {
+  bool parsed_value = false;
+
+  if (arg[0] != '-') {
+    // Not a flag.
+    name->clear();
+    parsed_value = true;
+    *value = arg;
+  } else if (arg[1] == '-') {
+    // Two dashes:  Multi-character name, with '=' separating name and
+    //   value.
+    const char* equals_pos = strchr(arg, '=');
+    if (equals_pos != NULL) {
+      *name = string(arg, equals_pos - arg);
+      *value = equals_pos + 1;
+      parsed_value = true;
+    } else {
+      *name = arg;
+    }
+  } else {
+    // One dash:  One-character name, all subsequent characters are the
+    //   value.
+    if (arg[1] == '\0') {
+      // arg is just "-".  We treat this as an input file, except that at
+      // present this will just lead to a "file not found" error.
+      name->clear();
+      *value = arg;
+      parsed_value = true;
+    } else {
+      *name = string(arg, 2);
+      *value = arg + 2;
+      parsed_value = !value->empty();
+    }
+  }
+
+  // Need to return true iff the next arg should be used as the value for this
+  // one, false otherwise.
+
+  if (parsed_value) {
+    // We already parsed a value for this flag.
+    return false;
+  }
+
+  if (*name == "-h" || *name == "--help" ||
+      *name == "--disallow_services" ||
+      *name == "--version") {
+    // HACK:  These are the only flags that don't take a value.
+    //   They probably should not be hard-coded like this but for now it's
+    //   not worth doing better.
+    return false;
+  }
+
+  // Next argument is the flag value.
+  return true;
+}
+
+bool CommandLineInterface::InterpretArgument(const string& name,
+                                             const string& value) {
+  if (name.empty()) {
+    // Not a flag.  Just a filename.
+    if (value.empty()) {
+      cerr << "You seem to have passed an empty string as one of the "
+              "arguments to " << executable_name_ << ".  This is actually "
+              "sort of hard to do.  Congrats.  Unfortunately it is not valid "
+              "input so the program is going to die now." << endl;
+      return false;
+    }
+
+    input_files_.push_back(value);
+
+  } else if (name == "-I" || name == "--proto_path") {
+    // Java's -classpath (and some other languages) delimits path components
+    // with colons.  Let's accept that syntax too just to make things more
+    // intuitive.
+    vector<string> parts;
+    SplitStringUsing(value, kPathSeparator, &parts);
+
+    for (int i = 0; i < parts.size(); i++) {
+      string virtual_path;
+      string disk_path;
+
+      int equals_pos = parts[i].find_first_of('=');
+      if (equals_pos == string::npos) {
+        virtual_path = "";
+        disk_path = parts[i];
+      } else {
+        virtual_path = parts[i].substr(0, equals_pos);
+        disk_path = parts[i].substr(equals_pos + 1);
+      }
+
+      if (disk_path.empty()) {
+        cerr << "--proto_path passed empty directory name.  (Use \".\" for "
+                "current directory.)" << endl;
+        return false;
+      }
+
+      // Make sure disk path exists, warn otherwise.
+      if (access(disk_path.c_str(), F_OK) < 0) {
+        cerr << disk_path << ": warning: directory does not exist." << endl;
+      }
+
+      proto_path_.push_back(make_pair(virtual_path, disk_path));
+    }
+
+  } else if (name == "-h" || name == "--help") {
+    PrintHelpText();
+    return false;  // Exit without running compiler.
+
+  } else if (name == "--version") {
+    if (!version_info_.empty()) {
+      cout << version_info_ << endl;
+    }
+    cout << "libprotoc "
+         << protobuf::internal::VersionString(GOOGLE_PROTOBUF_VERSION)
+         << endl;
+    return false;  // Exit without running compiler.
+
+  } else if (name == "--disallow_services") {
+    disallow_services_ = true;
+
+  } else {
+    // Some other flag.  Look it up in the generators list.
+    GeneratorMap::const_iterator iter = generators_.find(name);
+    if (iter == generators_.end()) {
+      cerr << "Unknown flag: " << name << endl;
+      return false;
+    }
+
+    // It's an output flag.  Add it to the output directives.
+    OutputDirective directive;
+    directive.name = name;
+    directive.generator = iter->second.generator;
+
+    // Split value at ':' to separate the generator parameter from the
+    // filename.
+    vector<string> parts;
+    SplitStringUsing(value, ":", &parts);
+
+    if (parts.size() == 1) {
+      directive.output_location = parts[0];
+    } else if (parts.size() == 2) {
+      directive.parameter = parts[0];
+      directive.output_location = parts[1];
+    } else {
+      cerr << "Invalid value for flag " << name << "." << endl;
+      return false;
+    }
+
+    output_directives_.push_back(directive);
+  }
+
+  return true;
+}
+
+void CommandLineInterface::PrintHelpText() {
+  // Sorry for indentation here; line wrapping would be uglier.
+  cerr <<
+"Usage: " << executable_name_ << " [OPTION] PROTO_FILE\n"
+"Parse PROTO_FILE and generate output based on the options given:\n"
+"  -IPATH, --proto_path=PATH   Specify the directory in which to search for\n"
+"                              imports.  May be specified multiple times;\n"
+"                              directories will be searched in order.  If not\n"
+"                              given, the current working directory is used.\n"
+"  --version                   Show version info and exit.\n"
+"  -h, --help                  Show this text and exit." << endl;
+
+  for (GeneratorMap::iterator iter = generators_.begin();
+       iter != generators_.end(); ++iter) {
+    // FIXME(kenton):  If the text is long enough it will wrap, which is ugly,
+    //   but fixing this nicely (e.g. splitting on spaces) is probably more
+    //   trouble than it's worth.
+    cerr << "  " << iter->first << "=OUT_DIR "
+         << string(19 - iter->first.size(), ' ')  // Spaces for alignment.
+         << iter->second.help_text << endl;
+  }
+}
+
+bool CommandLineInterface::GenerateOutput(
+    const FileDescriptor* parsed_file,
+    const OutputDirective& output_directive) {
+  // Create the output directory.
+  DiskOutputDirectory output_directory(output_directive.output_location);
+  if (!output_directory.VerifyExistence()) {
+    return false;
+  }
+
+  // Opened successfully.  Write it.
+
+  // Call the generator.
+  string error;
+  if (!output_directive.generator->Generate(
+      parsed_file, output_directive.parameter, &output_directory, &error)) {
+    // Generator returned an error.
+    cerr << output_directive.name << ": " << error << endl;
+    return false;
+  }
+
+  // Check for write errors.
+  if (output_directory.had_error()) {
+    return false;
+  }
+
+  return true;
+}
+
+
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/command_line_interface.h b/src/google/protobuf/compiler/command_line_interface.h
new file mode 100644
index 00000000..d3cae75e
--- /dev/null
+++ b/src/google/protobuf/compiler/command_line_interface.h
@@ -0,0 +1,210 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// Implements the Protocol Compiler front-end such that it may be reused by
+// custom compilers written to support other languages.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_COMMAND_LINE_INTERFACE_H__
+#define GOOGLE_PROTOBUF_COMPILER_COMMAND_LINE_INTERFACE_H__
+
+#include <google/protobuf/stubs/common.h>
+#include <string>
+#include <vector>
+#include <map>
+#include <set>
+#include <utility>
+
+namespace google {
+namespace protobuf {
+
+class FileDescriptor;        // descriptor.h
+
+namespace compiler {
+
+class CodeGenerator;        // code_generator.h
+class DiskSourceTree;       // importer.h
+
+// This class implements the command-line interface to the protocol compiler.
+// It is designed to make it very easy to create a custom protocol compiler
+// supporting the languages of your choice.  For example, if you wanted to
+// create a custom protocol compiler binary which includes both the regular
+// C++ support plus support for your own custom output "Foo", you would
+// write a class "FooGenerator" which implements the CodeGenerator interface,
+// then write a main() procedure like this:
+//
+//   int main(int argc, char* argv[]) {
+//     google::protobuf::compiler::CommandLineInterface cli;
+//
+//     // Support generation of C++ source and headers.
+//     google::protobuf::compiler::cpp::CppGenerator cpp_generator;
+//     cli.RegisterGenerator("--cpp_out", &cpp_generator,
+//       "Generate C++ source and header.");
+//
+//     // Support generation of Foo code.
+//     FooGenerator foo_generator;
+//     cli.RegisterGenerator("--foo_out", &foo_generator,
+//       "Generate Foo file.");
+//
+//     return cli.Run(argc, argv);
+//   }
+//
+// The compiler is invoked with syntax like:
+//   protoc --cpp_out=outdir --foo_out=outdir --proto_path=src foo.proto
+//
+// For a full description of the command-line syntax, invoke it with --help.
+class LIBPROTOC_EXPORT CommandLineInterface {
+ public:
+  CommandLineInterface();
+  ~CommandLineInterface();
+
+  // Register a code generator for a language.
+  //
+  // Parameters:
+  // * flag_name: The command-line flag used to specify an output file of
+  //   this type.  The name must start with a '-'.  If the name is longer
+  //   than one letter, it must start with two '-'s.
+  // * generator: The CodeGenerator which will be called to generate files
+  //   of this type.
+  // * help_text: Text describing this flag in the --help output.
+  //
+  // Some generators accept extra parameters.  You can specify this parameter
+  // on the command-line by placing it before the output directory, separated
+  // by a colon:
+  //   protoc --foo_out=enable_bar:outdir
+  // The text before the colon is passed to CodeGenerator::Generate() as the
+  // "parameter".
+  void RegisterGenerator(const string& flag_name,
+                         CodeGenerator* generator,
+                         const string& help_text);
+
+  // Run the Protocol Compiler with the given command-line parameters.
+  // Returns the error code which should be returned by main().
+  //
+  // It may not be safe to call Run() in a multi-threaded environment because
+  // it calls strerror().  I'm not sure why you'd want to do this anyway.
+  int Run(int argc, const char* const argv[]);
+
+  // Call SetInputsAreCwdRelative(true) if the input files given on the command
+  // line should be interpreted relative to the proto import path specified
+  // using --proto_path or -I flags.  Otherwise, input file names will be
+  // interpreted relative to the current working directory (or as absolute
+  // paths if they start with '/'), though they must still reside inside
+  // a directory given by --proto_path or the compiler will fail.  The latter
+  // mode is generally more intuitive and easier to use, especially e.g. when
+  // defining implicit rules in Makefiles.
+  void SetInputsAreProtoPathRelative(bool enable) {
+    inputs_are_proto_path_relative_ = enable;
+  }
+
+  // Provides some text which will be printed when the --version flag is
+  // used.  The version of libprotoc will also be printed on the next line
+  // after this text.
+  void SetVersionInfo(const string& text) {
+    version_info_ = text;
+  }
+
+
+ private:
+  // -----------------------------------------------------------------
+
+  class ErrorPrinter;
+  class DiskOutputDirectory;
+  class ErrorReportingFileOutput;
+
+  // Clear state from previous Run().
+  void Clear();
+
+  // Remaps each file in input_files_ so that it is relative to one of the
+  // directories in proto_path_.  Returns false if an error occurred.  This
+  // is only used if inputs_are_proto_path_relative_ is false.
+  bool MakeInputsBeProtoPathRelative(
+    DiskSourceTree* source_tree);
+
+  // Parse all command-line arguments.
+  bool ParseArguments(int argc, const char* const argv[]);
+
+  // Parses a command-line argument into a name/value pair.  Returns
+  // true if the next argument in the argv should be used as the value,
+  // false otherwise.
+  //
+  // Exmaples:
+  //   "-Isrc/protos" ->
+  //     name = "-I", value = "src/protos"
+  //   "--cpp_out=src/foo.pb2.cc" ->
+  //     name = "--cpp_out", value = "src/foo.pb2.cc"
+  //   "foo.proto" ->
+  //     name = "", value = "foo.proto"
+  bool ParseArgument(const char* arg, string* name, string* value);
+
+  // Interprets arguments parsed with ParseArgument.
+  bool InterpretArgument(const string& name, const string& value);
+
+  // Print the --help text to stderr.
+  void PrintHelpText();
+
+  // Generate the given output file from the given input.
+  struct OutputDirective;  // see below
+  bool GenerateOutput(const FileDescriptor* proto_file,
+                      const OutputDirective& output_directive);
+
+  // -----------------------------------------------------------------
+
+  // The name of the executable as invoked (i.e. argv[0]).
+  string executable_name_;
+
+  // Version info set with SetVersionInfo().
+  string version_info_;
+
+  // Map from flag names to registered generators.
+  struct GeneratorInfo {
+    CodeGenerator* generator;
+    string help_text;
+  };
+  typedef map<string, GeneratorInfo> GeneratorMap;
+  GeneratorMap generators_;
+
+  // Stuff parsed from command line.
+  vector<pair<string, string> > proto_path_;  // Search path for proto files.
+  vector<string> input_files_;                // Names of the input proto files.
+
+  // output_directives_ lists all the files we are supposed to output and what
+  // generator to use for each.
+  struct OutputDirective {
+    string name;
+    CodeGenerator* generator;
+    string parameter;
+    string output_location;
+  };
+  vector<OutputDirective> output_directives_;
+
+  // Was the --disallow_services flag used?
+  bool disallow_services_;
+
+  // See SetInputsAreProtoPathRelative().
+  bool inputs_are_proto_path_relative_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(CommandLineInterface);
+};
+
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_COMMAND_LINE_INTERFACE_H__
diff --git a/src/google/protobuf/compiler/command_line_interface_unittest.cc b/src/google/protobuf/compiler/command_line_interface_unittest.cc
new file mode 100644
index 00000000..1b1458de
--- /dev/null
+++ b/src/google/protobuf/compiler/command_line_interface_unittest.cc
@@ -0,0 +1,964 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <vector>
+
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/io/zero_copy_stream.h>
+#include <google/protobuf/compiler/command_line_interface.h>
+#include <google/protobuf/compiler/code_generator.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/testing/file.h>
+#include <google/protobuf/stubs/strutil.h>
+
+#include <google/protobuf/testing/googletest.h>
+#include <gtest/gtest.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+
+namespace {
+
+class CommandLineInterfaceTest : public testing::Test {
+ protected:
+  virtual void SetUp();
+  virtual void TearDown();
+
+  // Runs the CommandLineInterface with the given command line.  The
+  // command is automatically split on spaces, and the string "$tmpdir"
+  // is replaced with TestTempDir().
+  void Run(const string& command);
+
+  // -----------------------------------------------------------------
+  // Methods to set up the test (called before Run()).
+
+  class MockCodeGenerator;
+
+  // Registers a MockCodeGenerator with the given name.
+  MockCodeGenerator* RegisterGenerator(const string& generator_name,
+                                       const string& flag_name,
+                                       const string& filename,
+                                       const string& help_text);
+  MockCodeGenerator* RegisterErrorGenerator(const string& generator_name,
+                                            const string& error_text,
+                                            const string& flag_name,
+                                            const string& filename,
+                                            const string& help_text);
+
+  // Create a temp file within temp_directory_ with the given name.
+  // The containing directory is also created if necessary.
+  void CreateTempFile(const string& name, const string& contents);
+
+  void SetInputsAreProtoPathRelative(bool enable) {
+    cli_.SetInputsAreProtoPathRelative(enable);
+  }
+
+  // -----------------------------------------------------------------
+  // Methods to check the test results (called after Run()).
+
+  // Checks that no text was written to stderr during Run(), and Run()
+  // returned 0.
+  void ExpectNoErrors();
+
+  // Checks that Run() returned non-zero and the stderr output is exactly
+  // the text given.  expected_test may contain references to "$tmpdir",
+  // which will be replaced by the temporary directory path.
+  void ExpectErrorText(const string& expected_text);
+
+  // Checks that Run() returned non-zero and the stderr contains the given
+  // substring.
+  void ExpectErrorSubstring(const string& expected_substring);
+
+  // Returns true if ExpectErrorSubstring(expected_substring) would pass, but
+  // does not fail otherwise.
+  bool HasAlternateErrorSubstring(const string& expected_substring);
+
+  // Checks that MockCodeGenerator::Generate() was called in the given
+  // context.  That is, this tests if the generator with the given name
+  // was called with the given parameter and proto file and produced the
+  // given output file.  This is checked by reading the output file and
+  // checking that it contains the content that MockCodeGenerator would
+  // generate given these inputs.  message_name is the name of the first
+  // message that appeared in the proto file; this is just to make extra
+  // sure that the correct file was parsed.
+  void ExpectGenerated(const string& generator_name,
+                       const string& parameter,
+                       const string& proto_name,
+                       const string& message_name,
+                       const string& output_file);
+
+ private:
+  // The object we are testing.
+  CommandLineInterface cli_;
+
+  // We create a directory within TestTempDir() in order to add extra
+  // protection against accidentally deleting user files (since we recursively
+  // delete this directory during the test).  This is the full path of that
+  // directory.
+  string temp_directory_;
+
+  // The result of Run().
+  int return_code_;
+
+  // The captured stderr output.
+  string error_text_;
+
+  // Pointers which need to be deleted later.
+  vector<MockCodeGenerator*> mock_generators_to_delete_;
+};
+
+// A mock CodeGenerator which outputs information about the context in which
+// it was called, which can then be checked.  Output is written to a filename
+// constructed by concatenating the filename_prefix (given to the constructor)
+// with the proto file name, separated by a '.'.
+class CommandLineInterfaceTest::MockCodeGenerator : public CodeGenerator {
+ public:
+  // Create a MockCodeGenerator whose Generate() method returns true.
+  MockCodeGenerator(const string& name, const string& filename_prefix);
+
+  // Create a MockCodeGenerator whose Generate() method returns false
+  // and sets the error string to the given string.
+  MockCodeGenerator(const string& name, const string& filename_prefix,
+                    const string& error);
+
+  ~MockCodeGenerator();
+
+  void set_expect_write_error(bool value) {
+    expect_write_error_ = value;
+  }
+
+  // implements CodeGenerator ----------------------------------------
+  bool Generate(const FileDescriptor* file,
+                const string& parameter,
+                OutputDirectory* output_directory,
+                string* error) const;
+
+ private:
+  string name_;
+  string filename_prefix_;
+  bool return_error_;
+  string error_;
+  bool expect_write_error_;
+};
+
+// ===================================================================
+
+void CommandLineInterfaceTest::SetUp() {
+  // Most of these tests were written before this option was added, so we
+  // run with the option on (which used to be the only way) except in certain
+  // tests where we turn it off.
+  cli_.SetInputsAreProtoPathRelative(true);
+
+  temp_directory_ = TestTempDir() + "/proto2_cli_test_temp";
+
+  // If the temp directory already exists, it must be left over from a
+  // previous run.  Delete it.
+  if (File::Exists(temp_directory_)) {
+    File::DeleteRecursively(temp_directory_, NULL, NULL);
+  }
+
+  // Create the temp directory.
+  GOOGLE_CHECK(File::CreateDir(temp_directory_.c_str(), DEFAULT_FILE_MODE));
+}
+
+void CommandLineInterfaceTest::TearDown() {
+  // Delete the temp directory.
+  File::DeleteRecursively(temp_directory_, NULL, NULL);
+
+  // Delete all the MockCodeGenerators.
+  for (int i = 0; i < mock_generators_to_delete_.size(); i++) {
+    delete mock_generators_to_delete_[i];
+  }
+  mock_generators_to_delete_.clear();
+}
+
+void CommandLineInterfaceTest::Run(const string& command) {
+  vector<string> args;
+  SplitStringUsing(command, " ", &args);
+
+  scoped_array<const char*> argv(new const char*[args.size()]);
+
+  for (int i = 0; i < args.size(); i++) {
+    args[i] = StringReplace(args[i], "$tmpdir", temp_directory_, true);
+    argv[i] = args[i].c_str();
+  }
+
+  CaptureTestStderr();
+
+  return_code_ = cli_.Run(args.size(), argv.get());
+
+  error_text_ = GetCapturedTestStderr();
+}
+
+// -------------------------------------------------------------------
+
+CommandLineInterfaceTest::MockCodeGenerator*
+CommandLineInterfaceTest::RegisterGenerator(
+    const string& generator_name,
+    const string& flag_name,
+    const string& filename,
+    const string& help_text) {
+  MockCodeGenerator* generator =
+    new MockCodeGenerator(generator_name, filename);
+  mock_generators_to_delete_.push_back(generator);
+
+  cli_.RegisterGenerator(flag_name, generator, help_text);
+  return generator;
+}
+
+CommandLineInterfaceTest::MockCodeGenerator*
+CommandLineInterfaceTest::RegisterErrorGenerator(
+    const string& generator_name,
+    const string& error_text,
+    const string& flag_name,
+    const string& filename_prefix,
+    const string& help_text) {
+  MockCodeGenerator* generator =
+    new MockCodeGenerator(generator_name, filename_prefix, error_text);
+  mock_generators_to_delete_.push_back(generator);
+
+  cli_.RegisterGenerator(flag_name, generator, help_text);
+  return generator;
+}
+
+void CommandLineInterfaceTest::CreateTempFile(
+    const string& name,
+    const string& contents) {
+  // Create parent directory, if necessary.
+  string::size_type slash_pos = name.find_last_of('/');
+  if (slash_pos != string::npos) {
+    string dir = name.substr(0, slash_pos);
+    File::RecursivelyCreateDir(temp_directory_ + "/" + dir, 0777);
+  }
+
+  // Write file.
+  string full_name = temp_directory_ + "/" + name;
+  File::WriteStringToFileOrDie(contents, full_name);
+}
+
+// -------------------------------------------------------------------
+
+void CommandLineInterfaceTest::ExpectNoErrors() {
+  EXPECT_EQ(0, return_code_);
+  EXPECT_EQ("", error_text_);
+}
+
+void CommandLineInterfaceTest::ExpectErrorText(const string& expected_text) {
+  EXPECT_NE(0, return_code_);
+  EXPECT_EQ(StringReplace(expected_text, "$tmpdir", temp_directory_, true),
+            error_text_);
+}
+
+void CommandLineInterfaceTest::ExpectErrorSubstring(
+    const string& expected_substring) {
+  EXPECT_NE(0, return_code_);
+  EXPECT_PRED_FORMAT2(testing::IsSubstring, expected_substring, error_text_);
+}
+
+bool CommandLineInterfaceTest::HasAlternateErrorSubstring(
+    const string& expected_substring) {
+  EXPECT_NE(0, return_code_);
+  return error_text_.find(expected_substring) != string::npos;
+}
+
+void CommandLineInterfaceTest::ExpectGenerated(
+    const string& generator_name,
+    const string& parameter,
+    const string& proto_name,
+    const string& message_name,
+    const string& output_file_prefix) {
+  // Open and read the file.
+  string output_file = output_file_prefix + "." + proto_name;
+  string file_contents;
+  ASSERT_TRUE(File::ReadFileToString(temp_directory_ + "/" + output_file,
+                                     &file_contents))
+    << "Failed to open file: " + output_file;
+
+  // Check that the contents are as we expect.
+  string expected_contents =
+    generator_name + ": " + parameter + ", " + proto_name + ", " +
+    message_name + "\n";
+  EXPECT_EQ(expected_contents, file_contents)
+    << "Output file did not have expected contents: " + output_file;
+}
+
+// ===================================================================
+
+CommandLineInterfaceTest::MockCodeGenerator::MockCodeGenerator(
+    const string& name, const string& filename_prefix)
+  : name_(name),
+    filename_prefix_(filename_prefix),
+    return_error_(false),
+    expect_write_error_(false) {
+}
+
+CommandLineInterfaceTest::MockCodeGenerator::MockCodeGenerator(
+    const string& name, const string& filename_prefix, const string& error)
+  : name_(name),
+    filename_prefix_(filename_prefix),
+    return_error_(true),
+    error_(error),
+    expect_write_error_(false) {
+}
+
+CommandLineInterfaceTest::MockCodeGenerator::~MockCodeGenerator() {}
+
+bool CommandLineInterfaceTest::MockCodeGenerator::Generate(
+    const FileDescriptor* file,
+    const string& parameter,
+    OutputDirectory* output_directory,
+    string* error) const {
+  scoped_ptr<io::ZeroCopyOutputStream> output(
+    output_directory->Open(filename_prefix_ + "." + file->name()));
+  io::Printer printer(output.get(), '$');
+  map<string, string> vars;
+  vars["name"] = name_;
+  vars["parameter"] = parameter;
+  vars["proto_name"] = file->name();
+  vars["message_name"] = file->message_type_count() > 0 ?
+    file->message_type(0)->full_name().c_str() : "(none)";
+
+  printer.Print(vars, "$name$: $parameter$, $proto_name$, $message_name$\n");
+
+  if (expect_write_error_) {
+    EXPECT_TRUE(printer.failed());
+  } else {
+    EXPECT_FALSE(printer.failed());
+  }
+
+  *error = error_;
+  return !return_error_;
+}
+
+// ===================================================================
+
+TEST_F(CommandLineInterfaceTest, BasicOutput) {
+  // Test that the common case works.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, MultipleInputs) {
+  // Test parsing multiple input files.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+  CreateTempFile("bar.proto",
+    "syntax = \"proto2\";\n"
+    "message Bar {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto bar.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+  ExpectGenerated("test_generator", "", "bar.proto", "Bar", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, CreateDirectory) {
+  // Test that when we output to a sub-directory, it is created.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "bar/baz/output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "",
+                  "foo.proto", "Foo", "bar/baz/output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, GeneratorParameters) {
+  // Test that generator parameters are correctly parsed from the command line.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out=TestParameter:$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "TestParameter",
+                  "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, PathLookup) {
+  // Test that specifying multiple directories in the proto search path works.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("b/bar.proto",
+    "syntax = \"proto2\";\n"
+    "message Bar {}\n");
+  CreateTempFile("a/foo.proto",
+    "syntax = \"proto2\";\n"
+    "import \"bar.proto\";\n"
+    "message Foo {\n"
+    "  optional Bar a = 1;\n"
+    "}\n");
+  CreateTempFile("b/foo.proto", "this should not be parsed\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir/a --proto_path=$tmpdir/b foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, ColonDelimitedPath) {
+  // Same as PathLookup, but we provide the proto_path in a single flag.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("b/bar.proto",
+    "syntax = \"proto2\";\n"
+    "message Bar {}\n");
+  CreateTempFile("a/foo.proto",
+    "syntax = \"proto2\";\n"
+    "import \"bar.proto\";\n"
+    "message Foo {\n"
+    "  optional Bar a = 1;\n"
+    "}\n");
+  CreateTempFile("b/foo.proto", "this should not be parsed\n");
+
+#undef PATH_SEPARATOR
+#if defined(_WIN32)
+#define PATH_SEPARATOR ";"
+#else
+#define PATH_SEPARATOR ":"
+#endif
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir/a"PATH_SEPARATOR"$tmpdir/b foo.proto");
+
+#undef PATH_SEPARATOR
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, NonRootMapping) {
+  // Test setting up a search path mapping a directory to a non-root location.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=bar=$tmpdir bar/foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "bar/foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, MultipleGenerators) {
+  // Test that we can have multiple generators and use both in one invocation,
+  // each with a different output directory.
+
+  RegisterGenerator("test_generator_1", "--test1_out",
+                    "output1.test", "Test output 1.");
+  RegisterGenerator("test_generator_2", "--test2_out",
+                    "output2.test", "Test output 2.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+  // Create the "a" and "b" sub-directories.
+  CreateTempFile("a/dummy", "");
+  CreateTempFile("b/dummy", "");
+
+  Run("protocol_compiler "
+      "--test1_out=$tmpdir/a "
+      "--test2_out=$tmpdir/b "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator_1", "", "foo.proto", "Foo", "a/output1.test");
+  ExpectGenerated("test_generator_2", "", "foo.proto", "Foo", "b/output2.test");
+}
+
+TEST_F(CommandLineInterfaceTest, DisallowServicesNoServices) {
+  // Test that --disallow_services doesn't cause a problem when there are no
+  // services.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --disallow_services --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, DisallowServicesHasService) {
+  // Test that --disallow_services produces an error when there are services.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n"
+    "service Bar {}\n");
+
+  Run("protocol_compiler --disallow_services --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectErrorSubstring("foo.proto: This file contains services");
+}
+
+TEST_F(CommandLineInterfaceTest, AllowServicesHasService) {
+  // Test that services work fine as long as --disallow_services is not used.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n"
+    "service Bar {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, CwdRelativeInputs) {
+  // Test that we can accept working-directory-relative input files.
+
+  SetInputsAreProtoPathRelative(false);
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir $tmpdir/foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+// -------------------------------------------------------------------
+
+TEST_F(CommandLineInterfaceTest, ParseErrors) {
+  // Test that parse errors are reported.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "badsyntax\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectErrorText(
+    "foo.proto:2:1: Expected top-level statement (e.g. \"message\").\n");
+}
+
+TEST_F(CommandLineInterfaceTest, ParseErrorsMultipleFiles) {
+  // Test that parse errors are reported from multiple files.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  // We set up files such that foo.proto actually depends on bar.proto in
+  // two ways:  Directly and through baz.proto.  bar.proto's errors should
+  // only be reported once.
+  CreateTempFile("bar.proto",
+    "syntax = \"proto2\";\n"
+    "badsyntax\n");
+  CreateTempFile("baz.proto",
+    "syntax = \"proto2\";\n"
+    "import \"bar.proto\";\n");
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "import \"bar.proto\";\n"
+    "import \"baz.proto\";\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectErrorText(
+    "bar.proto:2:1: Expected top-level statement (e.g. \"message\").\n"
+    "baz.proto: Import \"bar.proto\" was not found or had errors.\n"
+    "foo.proto: Import \"bar.proto\" was not found or had errors.\n"
+    "foo.proto: Import \"baz.proto\" was not found or had errors.\n");
+}
+
+TEST_F(CommandLineInterfaceTest, InputNotFoundError) {
+  // Test what happens if the input file is not found.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectErrorText(
+    "foo.proto: File not found.\n");
+}
+
+TEST_F(CommandLineInterfaceTest, CwdRelativeInputNotFoundError) {
+  // Test what happens when a working-directory-relative input file is not
+  // found.
+
+  SetInputsAreProtoPathRelative(false);
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir $tmpdir/foo.proto");
+
+  ExpectErrorText(
+    "$tmpdir/foo.proto: No such file or directory\n");
+}
+
+TEST_F(CommandLineInterfaceTest, CwdRelativeInputNotMappedError) {
+  // Test what happens when a working-directory-relative input file is not
+  // mapped to a virtual path.
+
+  SetInputsAreProtoPathRelative(false);
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  // Create a directory called "bar" so that we can point --proto_path at it.
+  CreateTempFile("bar/dummy", "");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir/bar $tmpdir/foo.proto");
+
+  ExpectErrorText(
+    "$tmpdir/foo.proto: File does not reside within any path "
+      "specified using --proto_path (or -I).  You must specify a "
+      "--proto_path which encompasses this file.\n");
+}
+
+TEST_F(CommandLineInterfaceTest, CwdRelativeInputNotFoundAndNotMappedError) {
+  // Check what happens if the input file is not found *and* is not mapped
+  // in the proto_path.
+
+  SetInputsAreProtoPathRelative(false);
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  // Create a directory called "bar" so that we can point --proto_path at it.
+  CreateTempFile("bar/dummy", "");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir/bar $tmpdir/foo.proto");
+
+  ExpectErrorText(
+    "$tmpdir/foo.proto: No such file or directory\n");
+}
+
+TEST_F(CommandLineInterfaceTest, CwdRelativeInputShadowedError) {
+  // Test what happens when a working-directory-relative input file is shadowed
+  // by another file in the virtual path.
+
+  SetInputsAreProtoPathRelative(false);
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo/foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+  CreateTempFile("bar/foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Bar {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir/foo --proto_path=$tmpdir/bar "
+      "$tmpdir/bar/foo.proto");
+
+  ExpectErrorText(
+    "$tmpdir/bar/foo.proto: Input is shadowed in the --proto_path "
+    "by \"$tmpdir/foo/foo.proto\".  Either use the latter "
+    "file as your input or reorder the --proto_path so that the "
+    "former file's location comes first.\n");
+}
+
+TEST_F(CommandLineInterfaceTest, ProtoPathNotFoundError) {
+  // Test what happens if the input file is not found.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir/foo foo.proto");
+
+  ExpectErrorText(
+    "$tmpdir/foo: warning: directory does not exist.\n"
+    "foo.proto: File not found.\n");
+}
+
+TEST_F(CommandLineInterfaceTest, MissingInputError) {
+  // Test that we get an error if no inputs are given.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir");
+
+  ExpectErrorText("Missing input file.\n");
+}
+
+TEST_F(CommandLineInterfaceTest, MissingOutputError) {
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --proto_path=$tmpdir foo.proto");
+
+  ExpectErrorText("Missing output directives.\n");
+}
+
+TEST_F(CommandLineInterfaceTest, OutputWriteError) {
+  MockCodeGenerator* generator =
+    RegisterGenerator("test_generator", "--test_out",
+                      "output.test", "Test output.");
+  generator->set_expect_write_error(true);
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  // Create a directory blocking our output location.
+  CreateTempFile("output.test.foo.proto/foo", "");
+
+  Run("protocol_compiler --test_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+#if defined(_WIN32) && !defined(__CYGWIN__)
+  // Windows with MSVCRT.dll produces EPERM instead of EISDIR.
+  if (HasAlternateErrorSubstring("output.test.foo.proto: Permission denied")) {
+    return;
+  }
+#endif
+
+  ExpectErrorSubstring("output.test.foo.proto: Is a directory");
+}
+
+TEST_F(CommandLineInterfaceTest, OutputDirectoryNotFoundError) {
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir/nosuchdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectErrorSubstring("nosuchdir/: "
+                       "No such file or directory");
+}
+
+TEST_F(CommandLineInterfaceTest, OutputDirectoryIsFileError) {
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out=$tmpdir/foo.proto "
+      "--proto_path=$tmpdir foo.proto");
+
+#if defined(_WIN32) && !defined(__CYGWIN__)
+  // Windows with MSVCRT.dll produces EINVAL instead of ENOTDIR.
+  if (HasAlternateErrorSubstring("foo.proto/: Invalid argument")) {
+    return;
+  }
+#endif
+
+  ExpectErrorSubstring("foo.proto/: Not a directory");
+}
+
+TEST_F(CommandLineInterfaceTest, GeneratorError) {
+  RegisterErrorGenerator("error_generator", "Test error message.",
+                         "--error_out", "output.test", "Test error output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --error_out=$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectErrorSubstring("--error_out: Test error message.");
+}
+
+TEST_F(CommandLineInterfaceTest, HelpText) {
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+  RegisterErrorGenerator("error_generator", "Test error message.",
+                         "--error_out", "output.test", "Test error output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("test_exec_name --help");
+
+  ExpectErrorSubstring("Usage: test_exec_name ");
+  ExpectErrorSubstring("--test_out=OUT_DIR");
+  ExpectErrorSubstring("Test output.");
+  ExpectErrorSubstring("--error_out=OUT_DIR");
+  ExpectErrorSubstring("Test error output.");
+}
+
+// -------------------------------------------------------------------
+// Flag parsing tests
+
+TEST_F(CommandLineInterfaceTest, ParseSingleCharacterFlag) {
+  // Test that a single-character flag works.
+
+  RegisterGenerator("test_generator", "-o",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler -o$tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, ParseSpaceDelimitedValue) {
+  // Test that separating the flag value with a space works.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler --test_out $tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, ParseSingleCharacterSpaceDelimitedValue) {
+  // Test that separating the flag value with a space works for
+  // single-character flags.
+
+  RegisterGenerator("test_generator", "-o",
+                    "output.test", "Test output.");
+
+  CreateTempFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  Run("protocol_compiler -o $tmpdir "
+      "--proto_path=$tmpdir foo.proto");
+
+  ExpectNoErrors();
+  ExpectGenerated("test_generator", "", "foo.proto", "Foo", "output.test");
+}
+
+TEST_F(CommandLineInterfaceTest, MissingValueError) {
+  // Test that we get an error if a flag is missing its value.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  Run("protocol_compiler --test_out --proto_path=$tmpdir foo.proto");
+
+  ExpectErrorText("Missing value for flag: --test_out\n");
+}
+
+TEST_F(CommandLineInterfaceTest, MissingValueAtEndError) {
+  // Test that we get an error if the last argument is a flag requiring a
+  // value.
+
+  RegisterGenerator("test_generator", "--test_out",
+                    "output.test", "Test output.");
+
+  Run("protocol_compiler --test_out");
+
+  ExpectErrorText("Missing value for flag: --test_out\n");
+}
+
+}  // anonymous namespace
+
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_bootstrap_unittest.cc b/src/google/protobuf/compiler/cpp/cpp_bootstrap_unittest.cc
new file mode 100644
index 00000000..daa66c8c
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_bootstrap_unittest.cc
@@ -0,0 +1,135 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// This test insures that google/protobuf/descriptor.pb.{h,cc} match exactly
+// what would be generated by the protocol compiler.  These files are not
+// generated automatically at build time because they are compiled into the
+// protocol compiler itself.  So, if they were auto-generated, you'd have a
+// chicken-and-egg problem.
+//
+// If this test fails, run the script
+// "generate_descriptor_proto.sh" and add
+// descriptor.pb.{h,cc} to your changelist.
+
+#include <map>
+
+#include <google/protobuf/compiler/cpp/cpp_generator.h>
+#include <google/protobuf/compiler/importer.h>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/io/zero_copy_stream_impl.h>
+#include <google/protobuf/stubs/stl_util-inl.h>
+#include <google/protobuf/stubs/map-util.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/stubs/substitute.h>
+
+#include <google/protobuf/testing/file.h>
+#include <google/protobuf/testing/googletest.h>
+#include <gtest/gtest.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+namespace {
+
+class MockErrorCollector : public MultiFileErrorCollector {
+ public:
+  MockErrorCollector() {}
+  ~MockErrorCollector() {}
+
+  string text_;
+
+  // implements ErrorCollector ---------------------------------------
+  void AddError(const string& filename, int line, int column,
+                const string& message) {
+    strings::SubstituteAndAppend(&text_, "$0:$1:$2: $3\n",
+                                 filename, line, column, message);
+  }
+};
+
+class MockOutputDirectory : public OutputDirectory {
+ public:
+  MockOutputDirectory() {}
+  ~MockOutputDirectory() {
+    STLDeleteValues(&files_);
+  }
+
+  void ExpectFileMatches(const string& virtual_filename,
+                         const string& physical_filename) {
+    string* expected_contents = FindPtrOrNull(files_, virtual_filename);
+    ASSERT_TRUE(expected_contents != NULL)
+      << "Generator failed to generate file: " << virtual_filename;
+
+    string actual_contents;
+    File::ReadFileToStringOrDie(
+      TestSourceDir() + "/" + physical_filename,
+      &actual_contents);
+    EXPECT_TRUE(actual_contents == *expected_contents)
+      << physical_filename << " needs to be regenerated.  Please run "
+         "generate_descriptor_proto.sh and add this file "
+         "to your CL.";
+  }
+
+  // implements OutputDirectory --------------------------------------
+
+  virtual io::ZeroCopyOutputStream* Open(const string& filename) {
+    string** map_slot = &files_[filename];
+    if (*map_slot != NULL) delete *map_slot;
+    *map_slot = new string;
+
+    return new io::StringOutputStream(*map_slot);
+  }
+
+ private:
+  map<string, string*> files_;
+};
+
+TEST(BootstrapTest, GeneratedDescriptorMatches) {
+  MockErrorCollector error_collector;
+  DiskSourceTree source_tree;
+  source_tree.MapPath("", TestSourceDir());
+  Importer importer(&source_tree, &error_collector);
+  const FileDescriptor* proto_file =
+    importer.Import("google/protobuf/descriptor.proto");
+  EXPECT_EQ("", error_collector.text_);
+  ASSERT_TRUE(proto_file != NULL);
+
+  CppGenerator generator;
+  MockOutputDirectory output_directory;
+  string error;
+  string parameter;
+  parameter = "dllexport_decl=LIBPROTOBUF_EXPORT";
+  ASSERT_TRUE(generator.Generate(proto_file, parameter,
+                                 &output_directory, &error));
+
+  output_directory.ExpectFileMatches("google/protobuf/descriptor.pb.h",
+                                     "google/protobuf/descriptor.pb.h");
+  output_directory.ExpectFileMatches("google/protobuf/descriptor.pb.cc",
+                                     "google/protobuf/descriptor.pb.cc");
+}
+
+}  // namespace
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_enum.cc b/src/google/protobuf/compiler/cpp/cpp_enum.cc
new file mode 100644
index 00000000..f78d60d8
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_enum.cc
@@ -0,0 +1,196 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <set>
+#include <map>
+
+#include <google/protobuf/compiler/cpp/cpp_enum.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+EnumGenerator::EnumGenerator(const EnumDescriptor* descriptor,
+                             const string& dllexport_decl)
+  : descriptor_(descriptor),
+    classname_(ClassName(descriptor, false)),
+    dllexport_decl_(dllexport_decl) {
+}
+
+EnumGenerator::~EnumGenerator() {}
+
+void EnumGenerator::GenerateDefinition(io::Printer* printer) {
+  map<string, string> vars;
+  vars["classname"] = classname_;
+  vars["short_name"] = descriptor_->name();
+
+  printer->Print(vars, "enum $classname$ {\n");
+  printer->Indent();
+
+  const EnumValueDescriptor* min_value = descriptor_->value(0);
+  const EnumValueDescriptor* max_value = descriptor_->value(0);
+
+  for (int i = 0; i < descriptor_->value_count(); i++) {
+    vars["name"] = descriptor_->value(i)->name();
+    vars["number"] = SimpleItoa(descriptor_->value(i)->number());
+    vars["prefix"] = (descriptor_->containing_type() == NULL) ?
+      "" : classname_ + "_";
+
+    printer->Print(vars, "$prefix$$name$ = $number$,\n");
+
+    if (descriptor_->value(i)->number() < min_value->number()) {
+      min_value = descriptor_->value(i);
+    }
+    if (descriptor_->value(i)->number() > max_value->number()) {
+      max_value = descriptor_->value(i);
+    }
+  }
+
+  printer->Outdent();
+  printer->Print("};\n");
+
+  vars["min_name"] = min_value->name();
+  vars["max_name"] = max_value->name();
+
+  if (dllexport_decl_.empty()) {
+    vars["dllexport"] = "";
+  } else {
+    vars["dllexport"] = dllexport_decl_ + " ";
+  }
+
+  printer->Print(vars,
+    "$dllexport$const ::google::protobuf::EnumDescriptor* $classname$_descriptor();\n"
+    "$dllexport$bool $classname$_IsValid(int value);\n"
+    "const $classname$ $prefix$$short_name$_MIN = $prefix$$min_name$;\n"
+    "const $classname$ $prefix$$short_name$_MAX = $prefix$$max_name$;\n"
+    "\n");
+}
+
+void EnumGenerator::GenerateSymbolImports(io::Printer* printer) {
+  map<string, string> vars;
+  vars["nested_name"] = descriptor_->name();
+  vars["classname"] = classname_;
+  printer->Print(vars, "typedef $classname$ $nested_name$;\n");
+
+  for (int j = 0; j < descriptor_->value_count(); j++) {
+    vars["tag"] = descriptor_->value(j)->name();
+    printer->Print(vars,
+      "static const $nested_name$ $tag$ = $classname$_$tag$;\n");
+  }
+
+  printer->Print(vars,
+    "static inline const ::google::protobuf::EnumDescriptor*\n"
+    "$nested_name$_descriptor() {\n"
+    "  return $classname$_descriptor();\n"
+    "}\n"
+    "static inline bool $nested_name$_IsValid(int value) {\n"
+    "  return $classname$_IsValid(value);\n"
+    "}\n"
+    "static const $nested_name$ $nested_name$_MIN =\n"
+    "  $classname$_$nested_name$_MIN;\n"
+    "static const $nested_name$ $nested_name$_MAX =\n"
+    "  $classname$_$nested_name$_MAX;\n");
+}
+
+void EnumGenerator::GenerateDescriptorInitializer(
+    io::Printer* printer, int index) {
+  map<string, string> vars;
+  vars["classname"] = classname_;
+  vars["index"] = SimpleItoa(index);
+
+  if (descriptor_->containing_type() == NULL) {
+    printer->Print(vars,
+      "$classname$_descriptor_ = file->enum_type($index$);\n");
+  } else {
+    vars["parent"] = ClassName(descriptor_->containing_type(), false);
+    printer->Print(vars,
+      "$classname$_descriptor_ = $parent$_descriptor_->enum_type($index$);\n");
+  }
+}
+
+void EnumGenerator::GenerateMethods(io::Printer* printer) {
+  map<string, string> vars;
+  vars["classname"] = classname_;
+  vars["builddescriptorsname"] =
+      GlobalBuildDescriptorsName(descriptor_->file()->name());
+
+  printer->Print(vars,
+    "const ::google::protobuf::EnumDescriptor* $classname$_descriptor() {\n"
+    "  if ($classname$_descriptor_ == NULL) $builddescriptorsname$();\n"
+    "  return $classname$_descriptor_;\n"
+    "}\n"
+    "bool $classname$_IsValid(int value) {\n"
+    "  switch(value) {\n");
+
+  // Multiple values may have the same number.  Make sure we only cover
+  // each number once by first constructing a set containing all valid
+  // numbers, then printing a case statement for each element.
+
+  set<int> numbers;
+  for (int j = 0; j < descriptor_->value_count(); j++) {
+    const EnumValueDescriptor* value = descriptor_->value(j);
+    numbers.insert(value->number());
+  }
+
+  for (set<int>::iterator iter = numbers.begin();
+       iter != numbers.end(); ++iter) {
+    printer->Print(
+      "    case $number$:\n",
+      "number", SimpleItoa(*iter));
+  }
+
+  printer->Print(vars,
+    "      return true;\n"
+    "    default:\n"
+    "      return false;\n"
+    "  }\n"
+    "}\n"
+    "\n");
+
+  if (descriptor_->containing_type() != NULL) {
+    // We need to "define" the static constants which were declared in the
+    // header, to give the linker a place to put them.  Or at least the C++
+    // standard says we have to.  MSVC actually insists tha we do _not_ define
+    // them again in the .cc file.
+    printer->Print("#ifndef _MSC_VER\n");
+
+    vars["parent"] = ClassName(descriptor_->containing_type(), false);
+    vars["nested_name"] = descriptor_->name();
+    for (int i = 0; i < descriptor_->value_count(); i++) {
+      vars["value"] = descriptor_->value(i)->name();
+      printer->Print(vars,
+        "const $classname$ $parent$::$value$;\n");
+    }
+    printer->Print(vars,
+      "const $classname$ $parent$::$nested_name$_MIN;\n"
+      "const $classname$ $parent$::$nested_name$_MAX;\n");
+
+    printer->Print("#endif  // _MSC_VER\n");
+  }
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_enum.h b/src/google/protobuf/compiler/cpp/cpp_enum.h
new file mode 100644
index 00000000..b30997c9
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_enum.h
@@ -0,0 +1,81 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_ENUM_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_ENUM_H__
+
+#include <string>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class EnumGenerator {
+ public:
+  // See generator.cc for the meaning of dllexport_decl.
+  explicit EnumGenerator(const EnumDescriptor* descriptor,
+                         const string& dllexport_decl);
+  ~EnumGenerator();
+
+  // Header stuff.
+
+  // Generate header code defining the enum.  This code should be placed
+  // within the enum's package namespace, but NOT within any class, even for
+  // nested enums.
+  void GenerateDefinition(io::Printer* printer);
+
+  // For enums nested within a message, generate code to import all the enum's
+  // symbols (e.g. the enum type name, all its values, etc.) into the class's
+  // namespace.  This should be placed inside the class definition in the
+  // header.
+  void GenerateSymbolImports(io::Printer* printer);
+
+  // Source file stuff.
+
+  // Generate code that initializes the global variable storing the enum's
+  // descriptor.
+  void GenerateDescriptorInitializer(io::Printer* printer, int index);
+
+  // Generate non-inline methods related to the enum, such as IsValidValue().
+  // Goes in the .cc file.
+  void GenerateMethods(io::Printer* printer);
+
+ private:
+  const EnumDescriptor* descriptor_;
+  string classname_;
+  string dllexport_decl_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(EnumGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_ENUM_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_enum_field.cc b/src/google/protobuf/compiler/cpp/cpp_enum_field.cc
new file mode 100644
index 00000000..e02d7d8a
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_enum_field.cc
@@ -0,0 +1,226 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_enum_field.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/wire_format_inl.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+using internal::WireFormat;
+
+namespace {
+
+// TODO(kenton):  Factor out a "SetCommonFieldVariables()" to get rid of
+//   repeat code between this and the other field types.
+void SetEnumVariables(const FieldDescriptor* descriptor,
+                      map<string, string>* variables) {
+  const EnumValueDescriptor* default_value = descriptor->default_value_enum();
+
+  (*variables)["name"] = FieldName(descriptor);
+  (*variables)["type"] = ClassName(descriptor->enum_type(), true);
+  (*variables)["default"] = SimpleItoa(default_value->number());
+  (*variables)["index"] = SimpleItoa(descriptor->index());
+  (*variables)["number"] = SimpleItoa(descriptor->number());
+  (*variables)["classname"] = ClassName(FieldScope(descriptor), false);
+  (*variables)["tag_size"] = SimpleItoa(
+    WireFormat::TagSize(descriptor->number(), descriptor->type()));
+}
+
+}  // namespace
+
+// ===================================================================
+
+EnumFieldGenerator::
+EnumFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetEnumVariables(descriptor, &variables_);
+}
+
+EnumFieldGenerator::~EnumFieldGenerator() {}
+
+void EnumFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_, "int $name$_;\n");
+}
+
+void EnumFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline $type$ $name$() const;\n"
+    "inline void set_$name$($type$ value);\n");
+}
+
+void EnumFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline $type$ $classname$::$name$() const {\n"
+    "  return static_cast< $type$ >($name$_);\n"
+    "}\n"
+    "inline void $classname$::set_$name$($type$ value) {\n"
+    "  GOOGLE_DCHECK($type$_IsValid(value));\n"
+    "  _set_bit($index$);\n"
+    "  $name$_ = value;\n"
+    "}\n");
+}
+
+void EnumFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_ = $default$;\n");
+}
+
+void EnumFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_, "set_$name$(from.$name$());\n");
+}
+
+void EnumFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  printer->Print(variables_, ",\n$name$_($default$)");
+}
+
+void EnumFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  printer->Print(variables_,
+    "int value;\n"
+    "DO_(::google::protobuf::internal::WireFormat::ReadEnum(input, &value));\n"
+    "if ($type$_IsValid(value)) {\n"
+    "  set_$name$(static_cast< $type$ >(value));\n"
+    "} else {\n"
+    "  mutable_unknown_fields()->AddField($number$)->add_varint(value);\n"
+    "}\n");
+}
+
+void EnumFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::WriteEnum("
+      "$number$, this->$name$(), output));\n");
+}
+
+void EnumFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  printer->Print(variables_,
+    "total_size += $tag_size$ +\n"
+    "  ::google::protobuf::internal::WireFormat::EnumSize(this->$name$());\n");
+}
+
+// ===================================================================
+
+RepeatedEnumFieldGenerator::
+RepeatedEnumFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetEnumVariables(descriptor, &variables_);
+}
+
+RepeatedEnumFieldGenerator::~RepeatedEnumFieldGenerator() {}
+
+void RepeatedEnumFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_, "::google::protobuf::RepeatedField<int> $name$_;\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedField<int>& $name$() const;\n"
+    "inline ::google::protobuf::RepeatedField<int>* mutable_$name$();\n"
+    "inline $type$ $name$(int index) const;\n"
+    "inline void set_$name$(int index, $type$ value);\n"
+    "inline void add_$name$($type$ value);\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedField<int>&\n"
+    "$classname$::$name$() const {\n"
+    "  return $name$_;\n"
+    "}\n"
+    "inline ::google::protobuf::RepeatedField<int>*\n"
+    "$classname$::mutable_$name$() {\n"
+    "  return &$name$_;\n"
+    "}\n"
+    "inline $type$ $classname$::$name$(int index) const {\n"
+    "  return static_cast< $type$ >($name$_.Get(index));\n"
+    "}\n"
+    "inline void $classname$::set_$name$(int index, $type$ value) {\n"
+    "  GOOGLE_DCHECK($type$_IsValid(value));\n"
+    "  $name$_.Set(index, value);\n"
+    "}\n"
+    "inline void $classname$::add_$name$($type$ value) {\n"
+    "  GOOGLE_DCHECK($type$_IsValid(value));\n"
+    "  $name$_.Add(value);\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.Clear();\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.MergeFrom(from.$name$_);\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  // Not needed for repeated fields.
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  printer->Print(variables_,
+    "int value;\n"
+    "DO_(::google::protobuf::internal::WireFormat::ReadEnum(input, &value));\n"
+    "if ($type$_IsValid(value)) {\n"
+    "  add_$name$(static_cast< $type$ >(value));\n"
+    "} else {\n"
+    "  mutable_unknown_fields()->AddField($number$)->add_varint(value);\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::WriteEnum("
+      "$number$, this->$name$(i), output));\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  printer->Print(variables_,
+    "total_size += $tag_size$ * $name$_size();\n"
+    "for (int i = 0; i < $name$_size(); i++) {\n"
+    "  total_size += ::google::protobuf::internal::WireFormat::EnumSize(\n"
+    "    this->$name$(i));\n"
+    "}\n");
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_enum_field.h b/src/google/protobuf/compiler/cpp/cpp_enum_field.h
new file mode 100644
index 00000000..a297e961
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_enum_field.h
@@ -0,0 +1,84 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_ENUM_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_ENUM_FIELD_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/compiler/cpp/cpp_field.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class EnumFieldGenerator : public FieldGenerator {
+ public:
+  explicit EnumFieldGenerator(const FieldDescriptor* descriptor);
+  ~EnumFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(EnumFieldGenerator);
+};
+
+class RepeatedEnumFieldGenerator : public FieldGenerator {
+ public:
+  explicit RepeatedEnumFieldGenerator(const FieldDescriptor* descriptor);
+  ~RepeatedEnumFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(RepeatedEnumFieldGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_ENUM_FIELD_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_extension.cc b/src/google/protobuf/compiler/cpp/cpp_extension.cc
new file mode 100644
index 00000000..87da63d7
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_extension.cc
@@ -0,0 +1,104 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_extension.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/io/printer.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+ExtensionGenerator::ExtensionGenerator(const FieldDescriptor* descriptor,
+                                       const string& dllexport_decl)
+  : descriptor_(descriptor),
+    dllexport_decl_(dllexport_decl) {
+  // Construct type_traits_.
+  if (descriptor_->is_repeated()) {
+    type_traits_ = "Repeated";
+  }
+
+  switch (descriptor_->cpp_type()) {
+    case FieldDescriptor::CPPTYPE_ENUM:
+      type_traits_.append("EnumTypeTraits< ");
+      type_traits_.append(ClassName(descriptor_->enum_type(), true));
+      type_traits_.append(" >");
+      break;
+    case FieldDescriptor::CPPTYPE_STRING:
+      type_traits_.append("StringTypeTraits");
+      break;
+    case FieldDescriptor::CPPTYPE_MESSAGE:
+      type_traits_.append("MessageTypeTraits< ");
+      type_traits_.append(ClassName(descriptor_->message_type(), true));
+      type_traits_.append(" >");
+      break;
+    default:
+      type_traits_.append("PrimitiveTypeTraits< ");
+      type_traits_.append(PrimitiveTypeName(descriptor_->cpp_type()));
+      type_traits_.append(" >");
+      break;
+  }
+}
+
+ExtensionGenerator::~ExtensionGenerator() {}
+
+void ExtensionGenerator::GenerateDeclaration(io::Printer* printer) {
+  map<string, string> vars;
+  vars["extendee"   ] = ClassName(descriptor_->containing_type(), true);
+  vars["type_traits"] = type_traits_;
+  vars["name"       ] = descriptor_->name();
+
+  // If this is a class member, it needs to be declared "static".  Otherwise,
+  // it needs to be "extern".
+  vars["qualifier"] =
+    (descriptor_->extension_scope() == NULL) ? "extern" : "static";
+
+  if (!dllexport_decl_.empty()) {
+    vars["qualifier"] = dllexport_decl_ + " " + vars["qualifier"];
+  }
+
+  printer->Print(vars,
+    "$qualifier$ ::google::protobuf::internal::ExtensionIdentifier< $extendee$,\n"
+    "  ::google::protobuf::internal::$type_traits$ > $name$;\n");
+}
+
+void ExtensionGenerator::GenerateDefinition(io::Printer* printer) {
+  map<string, string> vars;
+  vars["extendee"   ] = ClassName(descriptor_->containing_type(), true);
+  vars["number"     ] = SimpleItoa(descriptor_->number());
+  vars["type_traits"] = type_traits_;
+  vars["name"       ] = descriptor_->name();
+
+  // If this is a class member, it needs to be declared in its class scope.
+  vars["scope"] = (descriptor_->extension_scope() == NULL) ? "" :
+    ClassName(descriptor_->extension_scope(), false) + "::";
+
+  printer->Print(vars,
+    "::google::protobuf::internal::ExtensionIdentifier< $extendee$,\n"
+    "  ::google::protobuf::internal::$type_traits$ > $scope$$name$($number$);\n");
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_extension.h b/src/google/protobuf/compiler/cpp/cpp_extension.h
new file mode 100644
index 00000000..149dbca9
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_extension.h
@@ -0,0 +1,68 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_EXTENSION_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_EXTENSION_H__
+
+#include <string>
+#include <google/protobuf/stubs/common.h>
+
+namespace google {
+namespace protobuf {
+  class FieldDescriptor;       // descriptor.h
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+// Generates code for an extension, which may be within the scope of some
+// message or may be at file scope.  This is much simpler than FieldGenerator
+// since extensions are just simple identifiers with interesting types.
+class ExtensionGenerator {
+ public:
+  // See generator.cc for the meaning of dllexport_decl.
+  explicit ExtensionGenerator(const FieldDescriptor* descriptor,
+                              const string& dllexport_decl);
+  ~ExtensionGenerator();
+
+  // Header stuff.
+  void GenerateDeclaration(io::Printer* printer);
+
+  // Source file stuff.
+  void GenerateDefinition(io::Printer* printer);
+
+ private:
+  const FieldDescriptor* descriptor_;
+  string type_traits_;
+  string dllexport_decl_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(ExtensionGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_MESSAGE_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_field.cc b/src/google/protobuf/compiler/cpp/cpp_field.cc
new file mode 100644
index 00000000..2b1041be
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_field.cc
@@ -0,0 +1,83 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_field.h>
+#include <google/protobuf/compiler/cpp/cpp_primitive_field.h>
+#include <google/protobuf/compiler/cpp/cpp_string_field.h>
+#include <google/protobuf/compiler/cpp/cpp_enum_field.h>
+#include <google/protobuf/compiler/cpp/cpp_message_field.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/common.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+FieldGenerator::~FieldGenerator() {}
+
+FieldGeneratorMap::FieldGeneratorMap(const Descriptor* descriptor)
+  : descriptor_(descriptor),
+    field_generators_(
+      new scoped_ptr<FieldGenerator>[descriptor->field_count()]) {
+  // Construct all the FieldGenerators.
+  for (int i = 0; i < descriptor->field_count(); i++) {
+    field_generators_[i].reset(MakeGenerator(descriptor->field(i)));
+  }
+}
+
+FieldGenerator* FieldGeneratorMap::MakeGenerator(const FieldDescriptor* field) {
+  if (field->is_repeated()) {
+    switch (field->cpp_type()) {
+      case FieldDescriptor::CPPTYPE_MESSAGE:
+        return new RepeatedMessageFieldGenerator(field);
+      case FieldDescriptor::CPPTYPE_STRING:
+          return new RepeatedStringFieldGenerator(field);
+      case FieldDescriptor::CPPTYPE_ENUM:
+        return new RepeatedEnumFieldGenerator(field);
+      default:
+        return new RepeatedPrimitiveFieldGenerator(field);
+    }
+  } else {
+    switch (field->cpp_type()) {
+      case FieldDescriptor::CPPTYPE_MESSAGE:
+        return new MessageFieldGenerator(field);
+      case FieldDescriptor::CPPTYPE_STRING:
+          return new StringFieldGenerator(field);
+      case FieldDescriptor::CPPTYPE_ENUM:
+        return new EnumFieldGenerator(field);
+      default:
+        return new PrimitiveFieldGenerator(field);
+    }
+  }
+}
+
+FieldGeneratorMap::~FieldGeneratorMap() {}
+
+const FieldGenerator& FieldGeneratorMap::get(
+    const FieldDescriptor* field) const {
+  GOOGLE_CHECK_EQ(field->containing_type(), descriptor_);
+  return *field_generators_[field->index()];
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_field.h b/src/google/protobuf/compiler/cpp/cpp_field.h
new file mode 100644
index 00000000..d37eb962
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_field.h
@@ -0,0 +1,127 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_FIELD_H__
+
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class FieldGenerator {
+ public:
+  FieldGenerator() {}
+  virtual ~FieldGenerator();
+
+  // Generate lines of code declaring members fields of the message class
+  // needed to represent this field.  These are placed inside the message
+  // class.
+  virtual void GeneratePrivateMembers(io::Printer* printer) const = 0;
+
+  // Generate prototypes for all of the accessor functions related to this
+  // field.  These are placed inside the class definition.
+  virtual void GenerateAccessorDeclarations(io::Printer* printer) const = 0;
+
+  // Generate inline definitions of accessor functions for this field.
+  // These are placed inside the header after all class definitions.
+  virtual void GenerateInlineAccessorDefinitions(
+    io::Printer* printer) const = 0;
+
+  // Generate definitions of accessors that aren't inlined.  These are
+  // placed somewhere in the .cc file.
+  // Most field types don't need this, so the default implementation is empty.
+  virtual void GenerateNonInlineAccessorDefinitions(
+    io::Printer* printer) const {}
+
+  // Generate lines of code (statements, not declarations) which clear the
+  // field.  This is used to define the clear_$name$() method as well as
+  // the Clear() method for the whole message.
+  virtual void GenerateClearingCode(io::Printer* printer) const = 0;
+
+  // Generate lines of code (statements, not declarations) which merges the
+  // contents of the field from the current message to the target message,
+  // which is stored in the generated code variable "from".
+  // This is used to fill in the MergeFrom method for the whole message.
+  // Details of this usage can be found in message.cc under the
+  // GenerateMergeFrom method.
+  virtual void GenerateMergingCode(io::Printer* printer) const = 0;
+
+  // Generate any initializers needed for the private members declared by
+  // GeneratePrivateMembers().  These go into the message class's
+  // constructor's initializer list.  For each initializer, this method
+  // must print the comma and newline separating it from the *previous*
+  // initializer, not the *next* initailizer.  That is, print a ",\n" first,
+  // e.g.:
+  //   printer->Print(",\n$name$_($default$)");
+  virtual void GenerateInitializer(io::Printer* printer) const = 0;
+
+  // Generate any code that needs to go in the class's destructor.
+  // Most field types don't need this, so the default implementation is empty.
+  virtual void GenerateDestructorCode(io::Printer* printer) const {}
+
+  // Generate lines to decode this field, which will be placed inside the
+  // message's MergeFromCodedStream() method.
+  virtual void GenerateMergeFromCodedStream(io::Printer* printer) const = 0;
+
+  // Generate lines to serialize this field, which are placed within the
+  // message's SerializeWithCachedSizes() method.
+  virtual void GenerateSerializeWithCachedSizes(io::Printer* printer) const = 0;
+
+  // Generate lines to compute the serialized size of this field, which
+  // are placed in the message's ByteSize() method.
+  virtual void GenerateByteSize(io::Printer* printer) const = 0;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(FieldGenerator);
+};
+
+// Convenience class which constructs FieldGenerators for a Descriptor.
+class FieldGeneratorMap {
+ public:
+  explicit FieldGeneratorMap(const Descriptor* descriptor);
+  ~FieldGeneratorMap();
+
+  const FieldGenerator& get(const FieldDescriptor* field) const;
+
+ private:
+  const Descriptor* descriptor_;
+  scoped_array<scoped_ptr<FieldGenerator> > field_generators_;
+
+  static FieldGenerator* MakeGenerator(const FieldDescriptor* field);
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(FieldGeneratorMap);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_FIELD_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_file.cc b/src/google/protobuf/compiler/cpp/cpp_file.cc
new file mode 100644
index 00000000..aea3a4b2
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_file.cc
@@ -0,0 +1,404 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_file.h>
+#include <google/protobuf/compiler/cpp/cpp_enum.h>
+#include <google/protobuf/compiler/cpp/cpp_service.h>
+#include <google/protobuf/compiler/cpp/cpp_extension.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/compiler/cpp/cpp_message.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+// ===================================================================
+
+FileGenerator::FileGenerator(const FileDescriptor* file,
+                             const string& dllexport_decl)
+  : file_(file),
+    message_generators_(
+      new scoped_ptr<MessageGenerator>[file->message_type_count()]),
+    enum_generators_(
+      new scoped_ptr<EnumGenerator>[file->enum_type_count()]),
+    service_generators_(
+      new scoped_ptr<ServiceGenerator>[file->service_count()]),
+    extension_generators_(
+      new scoped_ptr<ExtensionGenerator>[file->extension_count()]) {
+
+  for (int i = 0; i < file->message_type_count(); i++) {
+    message_generators_[i].reset(
+      new MessageGenerator(file->message_type(i), dllexport_decl));
+  }
+
+  for (int i = 0; i < file->enum_type_count(); i++) {
+    enum_generators_[i].reset(
+      new EnumGenerator(file->enum_type(i), dllexport_decl));
+  }
+
+  for (int i = 0; i < file->service_count(); i++) {
+    service_generators_[i].reset(
+      new ServiceGenerator(file->service(i), dllexport_decl));
+  }
+
+  for (int i = 0; i < file->extension_count(); i++) {
+    extension_generators_[i].reset(
+      new ExtensionGenerator(file->extension(i), dllexport_decl));
+  }
+
+  SplitStringUsing(file_->package(), ".", &package_parts_);
+}
+
+FileGenerator::~FileGenerator() {}
+
+void FileGenerator::GenerateHeader(io::Printer* printer) {
+  string filename_identifier = FilenameIdentifier(file_->name());
+
+  // Generate top of header.
+  printer->Print(
+    "// Generated by the protocol buffer compiler.  DO NOT EDIT!\n"
+    "\n"
+    "#ifndef PROTOBUF_$filename_identifier$__INCLUDED\n"
+    "#define PROTOBUF_$filename_identifier$__INCLUDED\n"
+    "\n"
+    "#include <string>\n"
+    "\n",
+    "filename_identifier", filename_identifier);
+
+  printer->Print(
+    "#include <google/protobuf/stubs/common.h>\n"
+    "\n");
+
+  // Verify the protobuf library header version is compatible with the protoc
+  // version before going any further.
+  printer->Print(
+    "#if GOOGLE_PROTOBUF_VERSION < $min_header_version$\n"
+    "#error This file was generated by a newer version of protoc which is\n"
+    "#error incompatible with your Protocol Buffer headers.  Please update\n"
+    "#error your headers.\n"
+    "#endif\n"
+    "#if $protoc_version$ < GOOGLE_PROTOBUF_MIN_PROTOC_VERSION\n"
+    "#error This file was generated by an older version of protoc which is\n"
+    "#error incompatible with your Protocol Buffer headers.  Please\n"
+    "#error regenerate this file with a newer version of protoc.\n"
+    "#endif\n"
+    "\n",
+    "min_header_version",
+      SimpleItoa(protobuf::internal::kMinHeaderVersionForProtoc),
+    "protoc_version", SimpleItoa(GOOGLE_PROTOBUF_VERSION));
+
+  // OK, it's now safe to #include other files.
+  printer->Print(
+    "#include <google/protobuf/generated_message_reflection.h>\n"
+    "#include <google/protobuf/repeated_field.h>\n"
+    "#include <google/protobuf/extension_set.h>\n");
+
+  if (file_->service_count() > 0) {
+    printer->Print(
+      "#include <google/protobuf/service.h>\n");
+  }
+
+  for (int i = 0; i < file_->dependency_count(); i++) {
+    printer->Print(
+      "#include \"$dependency$.pb.h\"\n",
+      "dependency", StripProto(file_->dependency(i)->name()));
+  }
+
+  // Open namespace.
+  GenerateNamespaceOpeners(printer);
+
+  printer->Print("\n");
+
+  // Generate forward declarations of classes.
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    message_generators_[i]->GenerateForwardDeclaration(printer);
+  }
+
+  printer->Print("\n");
+
+  // Generate enum definitions.
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    message_generators_[i]->GenerateEnumDefinitions(printer);
+  }
+  for (int i = 0; i < file_->enum_type_count(); i++) {
+    enum_generators_[i]->GenerateDefinition(printer);
+  }
+
+  printer->Print(kThickSeparator);
+  printer->Print("\n");
+
+  // Generate class definitions.
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    if (i > 0) {
+      printer->Print("\n");
+      printer->Print(kThinSeparator);
+      printer->Print("\n");
+    }
+    message_generators_[i]->GenerateClassDefinition(printer);
+  }
+
+  printer->Print("\n");
+  printer->Print(kThickSeparator);
+  printer->Print("\n");
+
+  // Generate service definitions.
+  for (int i = 0; i < file_->service_count(); i++) {
+    if (i > 0) {
+      printer->Print("\n");
+      printer->Print(kThinSeparator);
+      printer->Print("\n");
+    }
+    service_generators_[i]->GenerateDeclarations(printer);
+  }
+
+  printer->Print("\n");
+  printer->Print(kThickSeparator);
+  printer->Print("\n");
+
+  // Declare extension identifiers.
+  for (int i = 0; i < file_->extension_count(); i++) {
+    extension_generators_[i]->GenerateDeclaration(printer);
+  }
+
+  printer->Print("\n");
+  printer->Print(kThickSeparator);
+  printer->Print("\n");
+
+  // Generate class inline methods.
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    if (i > 0) {
+      printer->Print(kThinSeparator);
+      printer->Print("\n");
+    }
+    message_generators_[i]->GenerateInlineMethods(printer);
+  }
+
+  // Close up namespace.
+  GenerateNamespaceClosers(printer);
+
+  printer->Print(
+    "#endif  // PROTOBUF_$filename_identifier$__INCLUDED\n",
+    "filename_identifier", filename_identifier);
+}
+
+void FileGenerator::GenerateSource(io::Printer* printer) {
+  printer->Print(
+    "// Generated by the protocol buffer compiler.  DO NOT EDIT!\n"
+    "\n"
+    "#include \"$basename$.pb.h\"\n"
+    "#include <google/protobuf/descriptor.h>\n"
+    "#include <google/protobuf/io/coded_stream.h>\n"
+    "#include <google/protobuf/reflection_ops.h>\n"
+    "#include <google/protobuf/wire_format_inl.h>\n",
+    "basename", StripProto(file_->name()));
+
+  // For each dependency, write a prototype for that dependency's
+  // BuildDescriptors() function.  We don't expose these in the header because
+  // they are internal implementation details, and since this is generated code
+  // we don't have the usual risks involved with declaring external functions
+  // within a .cc file.
+  for (int i = 0; i < file_->dependency_count(); i++) {
+    const FileDescriptor* dependency = file_->dependency(i);
+    // Open the dependency's namespace.
+    vector<string> dependency_package_parts;
+    SplitStringUsing(dependency->package(), ".", &dependency_package_parts);
+    for (int i = 0; i < dependency_package_parts.size(); i++) {
+      printer->Print("namespace $name$ { ",
+                     "name", dependency_package_parts[i]);
+    }
+    // Declare its BuildDescriptors() function.
+    printer->Print(
+      "void $function$();",
+      "function", GlobalBuildDescriptorsName(dependency->name()));
+    // Close the namespace.
+    for (int i = 0; i < dependency_package_parts.size(); i++) {
+      printer->Print(" }");
+    }
+    printer->Print("\n");
+  }
+
+  GenerateNamespaceOpeners(printer);
+
+  printer->Print(
+    "\n"
+    "namespace {\n"
+    "\n");
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    message_generators_[i]->GenerateDescriptorDeclarations(printer);
+  }
+  for (int i = 0; i < file_->enum_type_count(); i++) {
+    printer->Print(
+      "const ::google::protobuf::EnumDescriptor* $name$_descriptor_ = NULL;\n",
+      "name", ClassName(file_->enum_type(i), false));
+  }
+  for (int i = 0; i < file_->service_count(); i++) {
+    printer->Print(
+      "const ::google::protobuf::ServiceDescriptor* $name$_descriptor_ = NULL;\n",
+      "name", file_->service(i)->name());
+  }
+
+  printer->Print(
+    "\n"
+    "}  // namespace\n"
+    "\n");
+
+  // Define our externally-visible BuildDescriptors() function.
+  GenerateBuildDescriptors(printer);
+
+  // Generate enums.
+  for (int i = 0; i < file_->enum_type_count(); i++) {
+    enum_generators_[i]->GenerateMethods(printer);
+  }
+
+  // Generate classes.
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    printer->Print("\n");
+    printer->Print(kThickSeparator);
+    printer->Print("\n");
+    message_generators_[i]->GenerateClassMethods(printer);
+  }
+
+  // Generate services.
+  for (int i = 0; i < file_->service_count(); i++) {
+    if (i == 0) printer->Print("\n");
+    printer->Print(kThickSeparator);
+    printer->Print("\n");
+    service_generators_[i]->GenerateImplementation(printer);
+  }
+
+  // Define extensions.
+  for (int i = 0; i < file_->extension_count(); i++) {
+    extension_generators_[i]->GenerateDefinition(printer);
+  }
+
+  GenerateNamespaceClosers(printer);
+}
+
+void FileGenerator::GenerateBuildDescriptors(io::Printer* printer) {
+  // BuildDescriptors() is a file-level procedure which initializes all of
+  // the Descriptor objects for this file.  It runs the first time one of the
+  // descriptors is accessed.  This will always be at static initialization
+  // time, because every message has a statically-initialized default instance,
+  // and the constructor for a message class accesses its descriptor.  See the
+  // constructor and the descriptor() method of message classes.
+  printer->Print(
+    "\n"
+    "void $builddescriptorsname$() {\n"
+    "  static bool already_here = false;\n"
+    "  if (already_here) return;\n"
+    "  already_here = true;\n"
+    "  GOOGLE_PROTOBUF_VERIFY_VERSION;\n"
+    "  ::google::protobuf::DescriptorPool* pool =\n"
+    "    ::google::protobuf::DescriptorPool::internal_generated_pool();\n"
+    "\n",
+    "builddescriptorsname", GlobalBuildDescriptorsName(file_->name()));
+  printer->Indent();
+
+  // Call the BuildDescriptors() methods for all of our dependencies, to make
+  // sure they get initialized first.
+  for (int i = 0; i < file_->dependency_count(); i++) {
+    const FileDescriptor* dependency = file_->dependency(i);
+    // Print the namespace prefix for the dependency.
+    vector<string> dependency_package_parts;
+    SplitStringUsing(dependency->package(), ".", &dependency_package_parts);
+    printer->Print("::");
+    for (int i = 0; i < dependency_package_parts.size(); i++) {
+      printer->Print("$name$::",
+                     "name", dependency_package_parts[i]);
+    }
+    // Call its BuildDescriptors function.
+    printer->Print(
+      "$name$();\n",
+      "name", GlobalBuildDescriptorsName(dependency->name()));
+  }
+
+  // Embed the descriptor.  We simply serialize the entire FileDescriptorProto
+  // and embed it as a string literal, which is parsed and built into real
+  // descriptors at initialization time.
+  FileDescriptorProto file_proto;
+  file_->CopyTo(&file_proto);
+  string file_data;
+  file_proto.SerializeToString(&file_data);
+
+  printer->Print(
+    "const ::google::protobuf::FileDescriptor* file = pool->InternalBuildGeneratedFile(");
+
+  // Only write 40 bytes per line.
+  static const int kBytesPerLine = 40;
+  for (int i = 0; i < file_data.size(); i += kBytesPerLine) {
+    printer->Print("\n  \"$data$\"",
+      "data", CEscape(file_data.substr(i, kBytesPerLine)));
+  }
+  printer->Print(
+    ", $size$);\n",
+    "size", SimpleItoa(file_data.size()));
+
+  // Assign all global descriptors.
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    message_generators_[i]->GenerateDescriptorInitializer(printer, i);
+  }
+  for (int i = 0; i < file_->enum_type_count(); i++) {
+    enum_generators_[i]->GenerateDescriptorInitializer(printer, i);
+  }
+  for (int i = 0; i < file_->service_count(); i++) {
+    service_generators_[i]->GenerateDescriptorInitializer(printer, i);
+  }
+
+  printer->Outdent();
+  printer->Print(
+    "}\n"
+    "\n"
+    "// Force BuildDescriptors() to be called at static initialization time.\n"
+    "struct StaticDescriptorInitializer_$filename$ {\n"
+    "  StaticDescriptorInitializer_$filename$() {\n"
+    "    $builddescriptorsname$();\n"
+    "  }\n"
+    "} static_descriptor_initializer_$filename$_;\n"
+    "\n",
+    "builddescriptorsname", GlobalBuildDescriptorsName(file_->name()),
+    "filename", FilenameIdentifier(file_->name()));
+}
+
+void FileGenerator::GenerateNamespaceOpeners(io::Printer* printer) {
+  if (package_parts_.size() > 0) printer->Print("\n");
+
+  for (int i = 0; i < package_parts_.size(); i++) {
+    printer->Print("namespace $part$ {\n",
+                   "part", package_parts_[i]);
+  }
+}
+
+void FileGenerator::GenerateNamespaceClosers(io::Printer* printer) {
+  if (package_parts_.size() > 0) printer->Print("\n");
+
+  for (int i = package_parts_.size() - 1; i >= 0; i--) {
+    printer->Print("}  // namespace $part$\n",
+                   "part", package_parts_[i]);
+  }
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_file.h b/src/google/protobuf/compiler/cpp/cpp_file.h
new file mode 100644
index 00000000..f2255ee1
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_file.h
@@ -0,0 +1,82 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_FILE_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_FILE_H__
+
+#include <string>
+#include <vector>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/compiler/cpp/cpp_field.h>
+
+namespace google {
+namespace protobuf {
+  class FileDescriptor;        // descriptor.h
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class EnumGenerator;           // enum.h
+class MessageGenerator;        // message.h
+class ServiceGenerator;        // service.h
+class ExtensionGenerator;      // extension.h
+
+class FileGenerator {
+ public:
+  // See generator.cc for the meaning of dllexport_decl.
+  explicit FileGenerator(const FileDescriptor* file,
+                         const string& dllexport_decl);
+  ~FileGenerator();
+
+  void GenerateHeader(io::Printer* printer);
+  void GenerateSource(io::Printer* printer);
+
+ private:
+  // Generate the BuildDescriptors() procedure, which builds all descriptors
+  // for types defined in the file.
+  void GenerateBuildDescriptors(io::Printer* printer);
+
+  void GenerateNamespaceOpeners(io::Printer* printer);
+  void GenerateNamespaceClosers(io::Printer* printer);
+
+  const FileDescriptor* file_;
+
+  scoped_array<scoped_ptr<MessageGenerator> > message_generators_;
+  scoped_array<scoped_ptr<EnumGenerator> > enum_generators_;
+  scoped_array<scoped_ptr<ServiceGenerator> > service_generators_;
+  scoped_array<scoped_ptr<ExtensionGenerator> > extension_generators_;
+
+  // E.g. if the package is foo.bar, package_parts_ is {"foo", "bar"}.
+  vector<string> package_parts_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(FileGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_FILE_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_generator.cc b/src/google/protobuf/compiler/cpp/cpp_generator.cc
new file mode 100644
index 00000000..200f6358
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_generator.cc
@@ -0,0 +1,135 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_generator.h>
+
+#include <vector>
+#include <utility>
+
+#include <google/protobuf/compiler/cpp/cpp_file.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/io/zero_copy_stream.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+namespace {
+
+// Parses a set of comma-delimited name/value pairs, e.g.:
+//   "foo=bar,baz,qux=corge"
+// parses to the pairs:
+//   ("foo", "bar"), ("baz", ""), ("qux", "corge")
+void ParseOptions(const string& text, vector<pair<string, string> >* output) {
+  vector<string> parts;
+  SplitStringUsing(text, ",", &parts);
+
+  for (int i = 0; i < parts.size(); i++) {
+    string::size_type equals_pos = parts[i].find_first_of('=');
+    pair<string, string> value;
+    if (equals_pos == string::npos) {
+      value.first = parts[i];
+      value.second = "";
+    } else {
+      value.first = parts[i].substr(0, equals_pos);
+      value.second = parts[i].substr(equals_pos + 1);
+    }
+    output->push_back(value);
+  }
+}
+
+}  // namespace
+
+CppGenerator::CppGenerator() {}
+CppGenerator::~CppGenerator() {}
+
+bool CppGenerator::Generate(const FileDescriptor* file,
+                            const string& parameter,
+                            OutputDirectory* output_directory,
+                            string* error) const {
+  vector<pair<string, string> > options;
+  ParseOptions(parameter, &options);
+
+  // -----------------------------------------------------------------
+  // parse generator options
+
+  // TODO(kenton):  If we ever have more options, we may want to create a
+  //   class that encapsulates them which we can pass down to all the
+  //   generator classes.  Currently we pass dllexport_decl down to all of
+  //   them via the constructors, but we don't want to have to add another
+  //   constructor parameter for every option.
+
+  // If the dllexport_decl option is passed to the compiler, we need to write
+  // it in front of every symbol that should be exported if this .proto is
+  // compiled into a Windows DLL.  E.g., if the user invokes the protocol
+  // compiler as:
+  //   protoc --cpp_out=dllexport_decl=FOO_EXPORT:outdir foo.proto
+  // then we'll define classes like this:
+  //   class FOO_EXPORT Foo {
+  //     ...
+  //   }
+  // FOO_EXPORT is a macro which should expand to __declspec(dllexport) or
+  // __declspec(dllimport) depending on what is being compiled.
+  string dllexport_decl;
+
+  for (int i = 0; i < options.size(); i++) {
+    if (options[i].first == "dllexport_decl") {
+      dllexport_decl = options[i].second;
+    } else {
+      *error = "Unknown generator option: " + options[i].first;
+      return false;
+    }
+  }
+
+  // -----------------------------------------------------------------
+
+
+  string basename = StripProto(file->name());
+  basename.append(".pb");
+
+  FileGenerator file_generator(file, dllexport_decl);
+
+  // Generate header.
+  {
+    scoped_ptr<io::ZeroCopyOutputStream> output(
+      output_directory->Open(basename + ".h"));
+    io::Printer printer(output.get(), '$');
+    file_generator.GenerateHeader(&printer);
+  }
+
+  // Generate cc file.
+  {
+    scoped_ptr<io::ZeroCopyOutputStream> output(
+      output_directory->Open(basename + ".cc"));
+    io::Printer printer(output.get(), '$');
+    file_generator.GenerateSource(&printer);
+  }
+
+  return true;
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_generator.h b/src/google/protobuf/compiler/cpp/cpp_generator.h
new file mode 100644
index 00000000..26fb8e97
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_generator.h
@@ -0,0 +1,58 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// Generates C++ code for a given .proto file.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_GENERATOR_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_GENERATOR_H__
+
+#include <string>
+#include <google/protobuf/compiler/code_generator.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+// CodeGenerator implementation which generates a C++ source file and
+// header.  If you create your own protocol compiler binary and you want
+// it to support C++ output, you can do so by registering an instance of this
+// CodeGenerator with the CommandLineInterface in your main() function.
+class LIBPROTOC_EXPORT CppGenerator : public CodeGenerator {
+ public:
+  CppGenerator();
+  ~CppGenerator();
+
+  // implements CodeGenerator ----------------------------------------
+  bool Generate(const FileDescriptor* file,
+                const string& parameter,
+                OutputDirectory* output_directory,
+                string* error) const;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(CppGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_GENERATOR_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_helpers.cc b/src/google/protobuf/compiler/cpp/cpp_helpers.cc
new file mode 100644
index 00000000..21de816c
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_helpers.cc
@@ -0,0 +1,197 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <vector>
+#include <google/protobuf/stubs/hash.h>
+
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+namespace {
+
+string DotsToUnderscores(const string& name) {
+  return StringReplace(name, ".", "_", true);
+}
+
+string DotsToColons(const string& name) {
+  return StringReplace(name, ".", "::", true);
+}
+
+const char* const kKeywordList[] = {
+  "and", "and_eq", "asm", "auto", "bitand", "bitor", "bool", "break", "case",
+  "catch", "char", "class", "compl", "const", "const_cast", "continue",
+  "default", "delete", "do", "double", "dynamic_cast", "else", "enum",
+  "explicit", "extern", "false", "float", "for", "friend", "goto", "if",
+  "inline", "int", "long", "mutable", "namespace", "new", "not", "not_eq",
+  "operator", "or", "or_eq", "private", "protected", "public", "register",
+  "reinterpret_cast", "return", "short", "signed", "sizeof", "static",
+  "static_cast", "struct", "switch", "template", "this", "throw", "true", "try",
+  "typedef", "typeid", "typename", "union", "unsigned", "using", "virtual",
+  "void", "volatile", "wchar_t", "while", "xor", "xor_eq"
+};
+
+hash_set<string> MakeKeywordsMap() {
+  hash_set<string> result;
+  for (int i = 0; i < GOOGLE_ARRAYSIZE(kKeywordList); i++) {
+    result.insert(kKeywordList[i]);
+  }
+  return result;
+}
+
+hash_set<string> kKeywords = MakeKeywordsMap();
+
+}  // namespace
+
+const char kThickSeparator[] =
+  "// ===================================================================\n";
+const char kThinSeparator[] =
+  "// -------------------------------------------------------------------\n";
+
+string ClassName(const Descriptor* descriptor, bool qualified) {
+  // Find "outer", the descriptor of the top-level message in which
+  // "descriptor" is embedded.
+  const Descriptor* outer = descriptor;
+  while (outer->containing_type() != NULL) outer = outer->containing_type();
+
+  const string& outer_name = outer->full_name();
+  string inner_name = descriptor->full_name().substr(outer_name.size());
+
+  if (qualified) {
+    return "::" + DotsToColons(outer_name) + DotsToUnderscores(inner_name);
+  } else {
+    return outer->name() + DotsToUnderscores(inner_name);
+  }
+}
+
+string ClassName(const EnumDescriptor* enum_descriptor, bool qualified) {
+  if (enum_descriptor->containing_type() == NULL) {
+    if (qualified) {
+      return DotsToColons(enum_descriptor->full_name());
+    } else {
+      return enum_descriptor->name();
+    }
+  } else {
+    string result = ClassName(enum_descriptor->containing_type(), qualified);
+    result += '_';
+    result += enum_descriptor->name();
+    return result;
+  }
+}
+
+string FieldName(const FieldDescriptor* field) {
+  string result = field->name();
+  LowerString(&result);
+  if (kKeywords.count(result) > 0) {
+    result.append("_");
+  }
+  return result;
+}
+
+string StripProto(const string& filename) {
+  if (HasSuffixString(filename, ".protodevel")) {
+    return StripSuffixString(filename, ".protodevel");
+  } else {
+    return StripSuffixString(filename, ".proto");
+  }
+}
+
+const char* PrimitiveTypeName(FieldDescriptor::CppType type) {
+  switch (type) {
+    case FieldDescriptor::CPPTYPE_INT32  : return "::google::protobuf::int32";
+    case FieldDescriptor::CPPTYPE_INT64  : return "::google::protobuf::int64";
+    case FieldDescriptor::CPPTYPE_UINT32 : return "::google::protobuf::uint32";
+    case FieldDescriptor::CPPTYPE_UINT64 : return "::google::protobuf::uint64";
+    case FieldDescriptor::CPPTYPE_DOUBLE : return "double";
+    case FieldDescriptor::CPPTYPE_FLOAT  : return "float";
+    case FieldDescriptor::CPPTYPE_BOOL   : return "bool";
+    case FieldDescriptor::CPPTYPE_ENUM   : return "int";
+    case FieldDescriptor::CPPTYPE_STRING : return "::std::string";
+    case FieldDescriptor::CPPTYPE_MESSAGE: return NULL;
+
+    // No default because we want the compiler to complain if any new
+    // CppTypes are added.
+  }
+
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return NULL;
+}
+
+const char* DeclaredTypeMethodName(FieldDescriptor::Type type) {
+  switch (type) {
+    case FieldDescriptor::TYPE_INT32   : return "Int32";
+    case FieldDescriptor::TYPE_INT64   : return "Int64";
+    case FieldDescriptor::TYPE_UINT32  : return "UInt32";
+    case FieldDescriptor::TYPE_UINT64  : return "UInt64";
+    case FieldDescriptor::TYPE_SINT32  : return "SInt32";
+    case FieldDescriptor::TYPE_SINT64  : return "SInt64";
+    case FieldDescriptor::TYPE_FIXED32 : return "Fixed32";
+    case FieldDescriptor::TYPE_FIXED64 : return "Fixed64";
+    case FieldDescriptor::TYPE_SFIXED32: return "SFixed32";
+    case FieldDescriptor::TYPE_SFIXED64: return "SFixed64";
+    case FieldDescriptor::TYPE_FLOAT   : return "Float";
+    case FieldDescriptor::TYPE_DOUBLE  : return "Double";
+
+    case FieldDescriptor::TYPE_BOOL    : return "Bool";
+    case FieldDescriptor::TYPE_ENUM    : return "Enum";
+
+    case FieldDescriptor::TYPE_STRING  : return "String";
+    case FieldDescriptor::TYPE_BYTES   : return "Bytes";
+    case FieldDescriptor::TYPE_GROUP   : return "Group";
+    case FieldDescriptor::TYPE_MESSAGE : return "Message";
+
+    // No default because we want the compiler to complain if any new
+    // types are added.
+  }
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return "";
+}
+
+// Convert a file name into a valid identifier.
+string FilenameIdentifier(const string& filename) {
+  string result;
+  for (int i = 0; i < filename.size(); i++) {
+    if (ascii_isalnum(filename[i])) {
+      result.push_back(filename[i]);
+    } else {
+      // Not alphanumeric.  To avoid any possibility of name conflicts we
+      // use the hex code for the character.
+      result.push_back('_');
+      char buffer[kFastToBufferSize];
+      result.append(FastHexToBuffer(static_cast<uint8>(filename[i]), buffer));
+    }
+  }
+  return result;
+}
+
+// Return the name of the BuildDescriptors() function for a given file.
+string GlobalBuildDescriptorsName(const string& filename) {
+  return "proto_BuildDescriptors_" + FilenameIdentifier(filename);
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_helpers.h b/src/google/protobuf/compiler/cpp/cpp_helpers.h
new file mode 100644
index 00000000..7f57d694
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_helpers.h
@@ -0,0 +1,86 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_HELPERS_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_HELPERS_H__
+
+#include <string>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+// Commonly-used separator comments.  Thick is a line of '=', thin is a line
+// of '-'.
+extern const char kThickSeparator[];
+extern const char kThinSeparator[];
+
+// Returns the non-nested type name for the given type.  If "qualified" is
+// true, prefix the type with the full namespace.  For example, if you had:
+//   package foo.bar;
+//   message Baz { message Qux {} }
+// Then the qualified ClassName for Qux would be:
+//   ::foo::bar::Baz_Qux
+// While the non-qualified version would be:
+//   Baz_Qux
+string ClassName(const Descriptor* descriptor, bool qualified);
+string ClassName(const EnumDescriptor* enum_descriptor, bool qualified);
+
+// Get the (unqualified) name that should be used for this field in C++ code.
+// The name is coerced to lower-case to emulate proto1 behavior.  People
+// should be using lowercase-with-underscores style for proto field names
+// anyway, so normally this just returns field->name().
+string FieldName(const FieldDescriptor* field);
+
+// Returns the scope where the field was defined (for extensions, this is
+// different from the message type to which the field applies).
+inline const Descriptor* FieldScope(const FieldDescriptor* field) {
+  return field->is_extension() ?
+    field->extension_scope() : field->containing_type();
+}
+
+// Strips ".proto" or ".protodevel" from the end of a filename.
+string StripProto(const string& filename);
+
+// Get the C++ type name for a primitive type (e.g. "double", "::google::protobuf::int32", etc.).
+// Note:  non-built-in type names will be qualified, meaning they will start
+// with a ::.  If you are using the type as a template parameter, you will
+// need to insure there is a space between the < and the ::, because the
+// ridiculous C++ standard defines "<:" to be a synonym for "[".
+const char* PrimitiveTypeName(FieldDescriptor::CppType type);
+
+// Get the declared type name in CamelCase format, as is used e.g. for the
+// methods of WireFormat.  For example, TYPE_INT32 becomes "Int32".
+const char* DeclaredTypeMethodName(FieldDescriptor::Type type);
+
+// Convert a file name into a valid identifier.
+string FilenameIdentifier(const string& filename);
+
+// Return the name of the BuildDescriptors() function for a given file.
+string GlobalBuildDescriptorsName(const string& filename);
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_HELPERS_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_message.cc b/src/google/protobuf/compiler/cpp/cpp_message.cc
new file mode 100644
index 00000000..002b0ad2
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_message.cc
@@ -0,0 +1,1445 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <algorithm>
+#include <google/protobuf/stubs/hash.h>
+#include <google/protobuf/compiler/cpp/cpp_message.h>
+#include <google/protobuf/compiler/cpp/cpp_enum.h>
+#include <google/protobuf/compiler/cpp/cpp_extension.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/io/coded_stream.h>
+#include <google/protobuf/wire_format.h>
+#include <google/protobuf/descriptor.pb.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+using internal::WireFormat;
+
+namespace {
+
+void PrintFieldComment(io::Printer* printer, const FieldDescriptor* field) {
+  // Print the field's proto-syntax definition as a comment.  We don't want to
+  // print group bodies so we cut off after the first line.
+  string def = field->DebugString();
+  printer->Print("// $def$\n",
+    "def", def.substr(0, def.find_first_of('\n')));
+}
+
+struct FieldOrderingByNumber {
+  inline bool operator()(const FieldDescriptor* a,
+                         const FieldDescriptor* b) const {
+    return a->number() < b->number();
+  }
+};
+
+const char* kWireTypeNames[] = {
+  "VARINT",
+  "FIXED64",
+  "LENGTH_DELIMITED",
+  "START_GROUP",
+  "END_GROUP",
+  "FIXED32",
+};
+
+// Sort the fields of the given Descriptor by number into a new[]'d array
+// and return it.
+const FieldDescriptor** SortFieldsByNumber(const Descriptor* descriptor) {
+  const FieldDescriptor** fields =
+    new const FieldDescriptor*[descriptor->field_count()];
+  for (int i = 0; i < descriptor->field_count(); i++) {
+    fields[i] = descriptor->field(i);
+  }
+  sort(fields, fields + descriptor->field_count(),
+       FieldOrderingByNumber());
+  return fields;
+}
+
+// Functor for sorting extension ranges by their "start" field number.
+struct ExtensionRangeSorter {
+  bool operator()(const Descriptor::ExtensionRange* left,
+                  const Descriptor::ExtensionRange* right) const {
+    return left->start < right->start;
+  }
+};
+
+// Returns true if the message type has any required fields.  If it doesn't,
+// we can optimize out calls to its IsInitialized() method.
+//
+// already_seen is used to avoid checking the same type multiple times
+// (and also to protect against recursion).
+static bool HasRequiredFields(
+    const Descriptor* type,
+    hash_set<const Descriptor*>* already_seen) {
+  if (already_seen->count(type) > 0) {
+    // Since the first occurrence of a required field causes the whole
+    // function to return true, we can assume that if the type is already
+    // in the cache it didn't have any required fields.
+    return false;
+  }
+  already_seen->insert(type);
+
+  // If the type has extensions, an extension with message type could contain
+  // required fields, so we have to be conservative and assume such an
+  // extension exists.
+  if (type->extension_range_count() > 0) return true;
+
+  for (int i = 0; i < type->field_count(); i++) {
+    const FieldDescriptor* field = type->field(i);
+    if (field->is_required()) {
+      return true;
+    }
+    if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
+      if (HasRequiredFields(field->message_type(), already_seen)) {
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
+static bool HasRequiredFields(const Descriptor* type) {
+  hash_set<const Descriptor*> already_seen;
+  return HasRequiredFields(type, &already_seen);
+}
+
+}
+
+// ===================================================================
+
+MessageGenerator::MessageGenerator(const Descriptor* descriptor,
+                                   const string& dllexport_decl)
+  : descriptor_(descriptor),
+    classname_(ClassName(descriptor, false)),
+    dllexport_decl_(dllexport_decl),
+    field_generators_(descriptor),
+    nested_generators_(new scoped_ptr<MessageGenerator>[
+      descriptor->nested_type_count()]),
+    enum_generators_(new scoped_ptr<EnumGenerator>[
+      descriptor->enum_type_count()]),
+    extension_generators_(new scoped_ptr<ExtensionGenerator>[
+      descriptor->extension_count()]) {
+
+  for (int i = 0; i < descriptor->nested_type_count(); i++) {
+    nested_generators_[i].reset(
+      new MessageGenerator(descriptor->nested_type(i), dllexport_decl));
+  }
+
+  for (int i = 0; i < descriptor->enum_type_count(); i++) {
+    enum_generators_[i].reset(
+      new EnumGenerator(descriptor->enum_type(i), dllexport_decl));
+  }
+
+  for (int i = 0; i < descriptor->extension_count(); i++) {
+    extension_generators_[i].reset(
+      new ExtensionGenerator(descriptor->extension(i), dllexport_decl));
+  }
+}
+
+MessageGenerator::~MessageGenerator() {}
+
+void MessageGenerator::
+GenerateForwardDeclaration(io::Printer* printer) {
+  printer->Print("class $classname$;\n",
+                 "classname", classname_);
+
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    nested_generators_[i]->GenerateForwardDeclaration(printer);
+  }
+}
+
+void MessageGenerator::
+GenerateEnumDefinitions(io::Printer* printer) {
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    nested_generators_[i]->GenerateEnumDefinitions(printer);
+  }
+
+  for (int i = 0; i < descriptor_->enum_type_count(); i++) {
+    enum_generators_[i]->GenerateDefinition(printer);
+  }
+}
+
+void MessageGenerator::
+GenerateFieldAccessorDeclarations(io::Printer* printer) {
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    PrintFieldComment(printer, field);
+
+    map<string, string> vars;
+    vars["name"] = FieldName(field);
+
+    if (field->is_repeated()) {
+      printer->Print(vars, "inline int $name$_size() const;\n");
+    } else {
+      printer->Print(vars, "inline bool has_$name$() const;\n");
+    }
+
+    printer->Print(vars, "inline void clear_$name$();\n");
+
+    // Generate type-specific accessor declarations.
+    field_generators_.get(field).GenerateAccessorDeclarations(printer);
+
+    printer->Print("\n");
+  }
+
+  if (descriptor_->extension_range_count() > 0) {
+    // Generate accessors for extensions.
+
+    // Normally I'd generate prototypes here and generate the actual
+    // definitions of these methods in GenerateFieldAccessorDefinitions, but
+    // the prototypes for these silly methods are so absurdly complicated that
+    // it meant way too much repitition.
+    //
+    // We use "_proto_TypeTraits" as a type name below because "TypeTraits"
+    // causes problems if the class has a nested message or enum type with that
+    // name and "_TypeTraits" is technically reserved for the C++ library since
+    // it starts with an underscore followed by a capital letter.
+    printer->Print(
+      // Has, Size, Clear
+      "template <typename _proto_TypeTraits>\n"
+      "inline bool HasExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id) const {\n"
+      "  return _extensions_.Has(id.number());\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline void ClearExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id) {\n"
+      "  _extensions_.ClearExtension(id.number());\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline int ExtensionSize(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id) const {\n"
+      "  return _extensions_.ExtensionSize(id.number());\n"
+      "}\n"
+      "\n"
+
+      // Singular accessors
+      "template <typename _proto_TypeTraits>\n"
+      "inline typename _proto_TypeTraits::ConstType GetExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id) const {\n"
+      "  return _proto_TypeTraits::Get(id.number(), _extensions_);\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline typename _proto_TypeTraits::MutableType MutableExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id) {\n"
+      "  return _proto_TypeTraits::Mutable(id.number(), &_extensions_);\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline void SetExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id,\n"
+      "    typename _proto_TypeTraits::ConstType value) {\n"
+      "  _proto_TypeTraits::Set(id.number(), value, &_extensions_);\n"
+      "}\n"
+      "\n"
+
+      // Repeated accessors
+      "template <typename _proto_TypeTraits>\n"
+      "inline typename _proto_TypeTraits::ConstType GetExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id,\n"
+      "    int index) const {\n"
+      "  return _proto_TypeTraits::Get(id.number(), _extensions_, index);\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline typename _proto_TypeTraits::MutableType MutableExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id,\n"
+      "    int index) {\n"
+      "  return _proto_TypeTraits::Mutable(id.number(),index,&_extensions_);\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline void SetExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id,\n"
+      "    int index, typename _proto_TypeTraits::ConstType value) {\n"
+      "  _proto_TypeTraits::Set(id.number(), index, value, &_extensions_);\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline typename _proto_TypeTraits::MutableType AddExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id) {\n"
+      "  return _proto_TypeTraits::Add(id.number(), &_extensions_);\n"
+      "}\n"
+      "\n"
+      "template <typename _proto_TypeTraits>\n"
+      "inline void AddExtension(\n"
+      "    const ::google::protobuf::internal::ExtensionIdentifier<\n"
+      "      $classname$, _proto_TypeTraits>& id,\n"
+      "    typename _proto_TypeTraits::ConstType value) {\n"
+      "  _proto_TypeTraits::Add(id.number(), value, &_extensions_);\n"
+      "}\n",
+      "classname", classname_);
+  }
+}
+
+void MessageGenerator::
+GenerateFieldAccessorDefinitions(io::Printer* printer) {
+  printer->Print("// $classname$\n\n", "classname", classname_);
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    PrintFieldComment(printer, field);
+
+    map<string, string> vars;
+    vars["name"] = FieldName(field);
+    vars["index"] = SimpleItoa(field->index());
+    vars["classname"] = classname_;
+
+    // Generate has_$name$() or $name$_size().
+    if (field->is_repeated()) {
+      printer->Print(vars,
+        "inline int $classname$::$name$_size() const {\n"
+        "  return $name$_.size();\n"
+        "}\n");
+    } else {
+      // Singular field.
+      printer->Print(vars,
+        "inline bool $classname$::has_$name$() const {\n"
+        "  return _has_bit($index$);\n"
+        "}\n");
+    }
+
+    // Generate clear_$name$()
+    printer->Print(vars,
+      "inline void $classname$::clear_$name$() {\n");
+
+    printer->Indent();
+    field_generators_.get(field).GenerateClearingCode(printer);
+    printer->Outdent();
+
+    if (!field->is_repeated()) {
+      printer->Print(vars, "  _clear_bit($index$);\n");
+    }
+
+    printer->Print("}\n");
+
+    // Generate type-specific accessors.
+    field_generators_.get(field).GenerateInlineAccessorDefinitions(printer);
+
+    printer->Print("\n");
+  }
+}
+
+void MessageGenerator::
+GenerateClassDefinition(io::Printer* printer) {
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    nested_generators_[i]->GenerateClassDefinition(printer);
+    printer->Print("\n");
+    printer->Print(kThinSeparator);
+    printer->Print("\n");
+  }
+
+  map<string, string> vars;
+  vars["classname"] = classname_;
+  vars["field_count"] = SimpleItoa(descriptor_->field_count());
+  if (dllexport_decl_.empty()) {
+    vars["dllexport"] = "";
+  } else {
+    vars["dllexport"] = dllexport_decl_ + " ";
+  }
+
+  printer->Print(vars,
+    "class $dllexport$$classname$ : public ::google::protobuf::Message {\n"
+    " public:\n");
+  printer->Indent();
+
+  printer->Print(vars,
+    "$classname$();\n"
+    "virtual ~$classname$();\n"
+    "\n"
+    "$classname$(const $classname$& from);\n"
+    "\n"
+    "inline $classname$& operator=(const $classname$& from) {\n"
+    "  CopyFrom(from);\n"
+    "  return *this;\n"
+    "}\n"
+    "\n"
+    "inline static const $classname$& default_instance() {\n"
+    "  return default_instance_;\n"
+    "}\n"
+    "\n"
+    "inline const ::google::protobuf::UnknownFieldSet& unknown_fields() const {\n"
+    "  return _reflection_.unknown_fields();\n"
+    "}\n"
+    "\n"
+    "inline ::google::protobuf::UnknownFieldSet* mutable_unknown_fields() {\n"
+    "  return _reflection_.mutable_unknown_fields();\n"
+    "}\n"
+    "\n"
+    "static const ::google::protobuf::Descriptor* descriptor();\n"
+    "\n"
+    "// implements Message ----------------------------------------------\n"
+    "\n"
+    "$classname$* New() const;\n");
+
+  if (descriptor_->file()->options().optimize_for() == FileOptions::SPEED) {
+    printer->Print(vars,
+      "void CopyFrom(const ::google::protobuf::Message& from);\n"
+      "void MergeFrom(const ::google::protobuf::Message& from);\n"
+      "void CopyFrom(const $classname$& from);\n"
+      "void MergeFrom(const $classname$& from);\n"
+      "void Clear();\n"
+      "bool IsInitialized() const;\n"
+      "int ByteSize() const;\n"
+      "\n"
+      "bool MergePartialFromCodedStream(\n"
+      "    ::google::protobuf::io::CodedInputStream* input);\n"
+      "bool SerializeWithCachedSizes(\n"
+      "    ::google::protobuf::io::CodedOutputStream* output) const;\n");
+  }
+
+  printer->Print(vars,
+    "int GetCachedSize() const { return _cached_size_; }\n"
+    "private:\n"
+    "void SetCachedSize(int size) const { _cached_size_ = size; }\n"
+    "public:\n"
+    "\n"
+    "const ::google::protobuf::Descriptor* GetDescriptor() const;\n"
+    "const ::google::protobuf::Message::Reflection* GetReflection() const;\n"
+    "::google::protobuf::Message::Reflection* GetReflection();\n"
+    "\n"
+    "// nested types ----------------------------------------------------\n"
+    "\n");
+
+  // Import all nested message classes into this class's scope with typedefs.
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    const Descriptor* nested_type = descriptor_->nested_type(i);
+    printer->Print("typedef $nested_full_name$ $nested_name$;\n",
+                   "nested_name", nested_type->name(),
+                   "nested_full_name", ClassName(nested_type, false));
+  }
+
+  if (descriptor_->nested_type_count() > 0) {
+    printer->Print("\n");
+  }
+
+  // Import all nested enums and their values into this class's scope with
+  // typedefs and constants.
+  for (int i = 0; i < descriptor_->enum_type_count(); i++) {
+    enum_generators_[i]->GenerateSymbolImports(printer);
+    printer->Print("\n");
+  }
+
+  printer->Print(
+    "// accessors -------------------------------------------------------\n"
+    "\n");
+
+  // Generate accessor methods for all fields.
+  GenerateFieldAccessorDeclarations(printer);
+
+  // Declare extension identifiers.
+  for (int i = 0; i < descriptor_->extension_count(); i++) {
+    extension_generators_[i]->GenerateDeclaration(printer);
+  }
+
+  // Generate private members for fields.
+  printer->Outdent();
+  printer->Print(" private:\n");
+  printer->Indent();
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "::google::protobuf::internal::ExtensionSet _extensions_;\n");
+  }
+
+  // TODO(kenton):  Make _cached_size_ an atomic<int> when C++ supports it.
+  printer->Print(
+    "::google::protobuf::internal::GeneratedMessageReflection _reflection_;\n"
+    "mutable int _cached_size_;\n"
+    "\n");
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    field_generators_.get(descriptor_->field(i))
+                     .GeneratePrivateMembers(printer);
+  }
+
+  // Generate offsets and _has_bits_ boilerplate.
+  printer->Print(vars,
+    "\n"
+    "static const $classname$ default_instance_;\n");
+
+  if (descriptor_->field_count() > 0) {
+    printer->Print(vars,
+      "static const int _offsets_[$field_count$];\n"
+      "\n"
+      "::google::protobuf::uint32 _has_bits_[($field_count$ + 31) / 32];\n");
+  } else {
+    // Zero-size arrays aren't technically allowed, and MSVC in particular
+    // doesn't like them.  We still need to declare these arrays to make
+    // other code compile.  Since this is an uncommon case, we'll just declare
+    // them with size 1 and waste some space.  Oh well.
+    printer->Print(
+      "static const int _offsets_[1];\n"
+      "\n"
+      "::google::protobuf::uint32 _has_bits_[1];\n");
+  }
+
+  printer->Print(
+    "\n"
+    "// WHY DOES & HAVE LOWER PRECEDENCE THAN != !?\n"
+    "inline bool _has_bit(int index) const {\n"
+    "  return (_has_bits_[index / 32] & (1u << (index % 32))) != 0;\n"
+    "}\n"
+    "inline void _set_bit(int index) {\n"
+    "  _has_bits_[index / 32] |= (1u << (index % 32));\n"
+    "}\n"
+    "inline void _clear_bit(int index) {\n"
+    "  _has_bits_[index / 32] &= ~(1u << (index % 32));\n"
+    "}\n");
+
+  printer->Outdent();
+  printer->Print(vars, "};");
+}
+
+void MessageGenerator::
+GenerateInlineMethods(io::Printer* printer) {
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    nested_generators_[i]->GenerateInlineMethods(printer);
+    printer->Print(kThinSeparator);
+    printer->Print("\n");
+  }
+
+  GenerateFieldAccessorDefinitions(printer);
+}
+
+void MessageGenerator::
+GenerateDescriptorDeclarations(io::Printer* printer) {
+  printer->Print("const ::google::protobuf::Descriptor* $name$_descriptor_ = NULL;\n",
+                 "name", classname_);
+
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    nested_generators_[i]->GenerateDescriptorDeclarations(printer);
+  }
+
+  for (int i = 0; i < descriptor_->enum_type_count(); i++) {
+    printer->Print(
+      "const ::google::protobuf::EnumDescriptor* $name$_descriptor_ = NULL;\n",
+      "name", ClassName(descriptor_->enum_type(i), false));
+  }
+}
+
+void MessageGenerator::
+GenerateDescriptorInitializer(io::Printer* printer, int index) {
+  // TODO(kenton):  Passing the index to this method is redundant; just use
+  //   descriptor_->index() instead.
+  map<string, string> vars;
+  vars["classname"] = classname_;
+  vars["index"] = SimpleItoa(index);
+
+  if (descriptor_->containing_type() == NULL) {
+    printer->Print(vars,
+      "$classname$_descriptor_ = file->message_type($index$);\n");
+  } else {
+    vars["parent"] = ClassName(descriptor_->containing_type(), false);
+    printer->Print(vars,
+      "$classname$_descriptor_ = "
+        "$parent$_descriptor_->nested_type($index$);\n");
+  }
+
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    nested_generators_[i]->GenerateDescriptorInitializer(printer, i);
+  }
+
+  for (int i = 0; i < descriptor_->enum_type_count(); i++) {
+    enum_generators_[i]->GenerateDescriptorInitializer(printer, i);
+  }
+
+  // Register this message type with the message factory.
+  printer->Print(vars,
+    "::google::protobuf::MessageFactory::InternalRegisterGeneratedMessage(\n"
+    "  $classname$_descriptor_, &$classname$::default_instance());\n");
+}
+
+void MessageGenerator::
+GenerateClassMethods(io::Printer* printer) {
+  for (int i = 0; i < descriptor_->enum_type_count(); i++) {
+    enum_generators_[i]->GenerateMethods(printer);
+  }
+
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    nested_generators_[i]->GenerateClassMethods(printer);
+    printer->Print("\n");
+    printer->Print(kThinSeparator);
+    printer->Print("\n");
+  }
+
+  printer->Print(
+    "const $classname$ $classname$::default_instance_;\n"
+    "\n",
+    "classname", classname_);
+
+  // Generate non-inline field definitions.
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    field_generators_.get(descriptor_->field(i))
+                     .GenerateNonInlineAccessorDefinitions(printer);
+    printer->Print("\n");
+  }
+
+  // Define extension identifiers.
+  for (int i = 0; i < descriptor_->extension_count(); i++) {
+    extension_generators_[i]->GenerateDefinition(printer);
+  }
+
+  GenerateOffsets(printer);
+  printer->Print("\n");
+
+  GenerateStructors(printer);
+  printer->Print("\n");
+
+  if (descriptor_->file()->options().optimize_for() == FileOptions::SPEED) {
+    GenerateClear(printer);
+    printer->Print("\n");
+
+    GenerateMergeFromCodedStream(printer);
+    printer->Print("\n");
+
+    GenerateSerializeWithCachedSizes(printer);
+    printer->Print("\n");
+
+    GenerateByteSize(printer);
+    printer->Print("\n");
+
+    GenerateMergeFrom(printer);
+    printer->Print("\n");
+
+    GenerateCopyFrom(printer);
+    printer->Print("\n");
+
+    GenerateIsInitialized(printer);
+    printer->Print("\n");
+  }
+
+  printer->Print(
+    "const ::google::protobuf::Descriptor* $classname$::GetDescriptor() const {\n"
+    "  return descriptor();\n"
+    "}\n"
+    "\n"
+    "const ::google::protobuf::Message::Reflection*\n"
+    "$classname$::GetReflection() const {\n"
+    "  return &_reflection_;\n"
+    "}\n"
+    "\n"
+    "::google::protobuf::Message::Reflection* $classname$::GetReflection() {\n"
+    "  return &_reflection_;\n"
+    "}\n",
+    "classname", classname_);
+}
+
+void MessageGenerator::
+GenerateOffsets(io::Printer* printer) {
+  printer->Print(
+    "const int $classname$::_offsets_[$field_count$] = {\n",
+    "classname", classname_,
+    "field_count", SimpleItoa(max(1, descriptor_->field_count())));
+  printer->Indent();
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+    printer->Print(
+      "GOOGLE_PROTOBUF_GENERATED_MESSAGE_FIELD_OFFSET($classname$, $name$_),\n",
+      "classname", classname_,
+      "name", FieldName(field));
+  }
+
+  printer->Outdent();
+  printer->Print("};\n");
+}
+
+void MessageGenerator::
+GenerateInitializerList(io::Printer* printer) {
+  printer->Indent();
+  printer->Indent();
+
+  bool has_extensions = descriptor_->extension_range_count() > 0;
+  if (has_extensions) {
+    printer->Print(
+      "_extensions_(descriptor(),\n"
+      "             ::google::protobuf::DescriptorPool::generated_pool(),\n"
+      "             ::google::protobuf::MessageFactory::generated_factory()),\n");
+  }
+
+  printer->Print(
+    "_reflection_(descriptor(),\n"
+    "             this, &default_instance_,\n"
+    "             _offsets_, _has_bits_, $extensions$),\n"
+    "_cached_size_(0)",
+    "extensions", has_extensions ? "&_extensions_" : "NULL");
+
+  // Write the initializers for each field.
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    field_generators_.get(descriptor_->field(i))
+                     .GenerateInitializer(printer);
+  }
+
+  printer->Outdent();
+  printer->Outdent();
+}
+
+void MessageGenerator::
+GenerateStructors(io::Printer* printer) {
+  // Generate the default constructor.
+  printer->Print(
+      "$classname$::$classname$()\n"
+      "  : ",
+      "classname", classname_);
+  GenerateInitializerList(printer);
+  printer->Print(" {\n"
+    "  ::memset(_has_bits_, 0, sizeof(_has_bits_));\n"
+    "  if (this == &default_instance_) {\n");
+
+  // The default instance needs all of its embedded message pointers
+  // cross-linked to other default instances.
+  // TODO(kenton):  Maybe all message fields (even for non-default messages)
+  //   should be initialized to point at default instances rather than NULL?
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (!field->is_repeated() &&
+        field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
+      printer->Print(
+          "    $name$_ = const_cast< $type$*>(&$type$::default_instance());\n",
+          "name", FieldName(field),
+          "type", ClassName(field->message_type(), true));
+    }
+  }
+  printer->Print(
+    "  }\n"
+    "}\n"
+    "\n");
+
+  // Generate the copy constructor.
+  printer->Print(
+      "$classname$::$classname$(const $classname$& from)\n"
+      "  : ",
+      "classname", classname_);
+  GenerateInitializerList(printer);
+  printer->Print(" {\n"
+    "  ::memset(_has_bits_, 0, sizeof(_has_bits_));\n"
+    "  MergeFrom(from);\n"
+    "}\n"
+    "\n");
+
+  // Generate the destructor.
+  printer->Print(
+    "$classname$::~$classname$() {\n",
+    "classname", classname_);
+
+  printer->Indent();
+
+  // Write the destructors for each field.
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    field_generators_.get(descriptor_->field(i))
+                     .GenerateDestructorCode(printer);
+  }
+
+  printer->Print(
+    "if (this != &default_instance_) {\n");
+
+  // We need to delete all embedded messages.
+  // TODO(kenton):  If we make unset messages point at default instances
+  //   instead of NULL, then it would make sense to move this code into
+  //   MessageFieldGenerator::GenerateDestructorCode().
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (!field->is_repeated() &&
+        field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
+      printer->Print("  delete $name$_;\n",
+                     "name", FieldName(field));
+    }
+  }
+
+  printer->Outdent();
+
+  printer->Print(
+    "  }\n"
+    "}\n"
+    "\n"
+    "const ::google::protobuf::Descriptor* $classname$::descriptor() {\n"
+    "  if ($classname$_descriptor_ == NULL) $builddescriptorsname$();\n"
+    "  return $classname$_descriptor_;\n"
+    "}\n"
+    "\n"
+    "$classname$* $classname$::New() const {\n"
+    "  return new $classname$;\n"
+    "}\n",
+    "classname", classname_,
+    "builddescriptorsname",
+    GlobalBuildDescriptorsName(descriptor_->file()->name()));
+}
+
+void MessageGenerator::
+GenerateClear(io::Printer* printer) {
+  printer->Print("void $classname$::Clear() {\n",
+                 "classname", classname_);
+  printer->Indent();
+
+  int last_index = -1;
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print("_extensions_.Clear();\n");
+  }
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (!field->is_repeated()) {
+      map<string, string> vars;
+      vars["index"] = SimpleItoa(field->index());
+
+      // We can use the fact that _has_bits_ is a giant bitfield to our
+      // advantage:  We can check up to 32 bits at a time for equality to
+      // zero, and skip the whole range if so.  This can improve the speed
+      // of Clear() for messages which contain a very large number of
+      // optional fields of which only a few are used at a time.  Here,
+      // we've chosen to check 8 bits at a time rather than 32.
+      if (i / 8 != last_index / 8 || last_index < 0) {
+        if (last_index >= 0) {
+          printer->Outdent();
+          printer->Print("}\n");
+        }
+        printer->Print(vars,
+          "if (_has_bits_[$index$ / 32] & (0xffu << ($index$ % 32))) {\n");
+        printer->Indent();
+      }
+      last_index = i;
+
+      // It's faster to just overwrite primitive types, but we should
+      // only clear strings and messages if they were set.
+      // TODO(kenton):  Let the CppFieldGenerator decide this somehow.
+      bool should_check_bit =
+        field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE ||
+        field->cpp_type() == FieldDescriptor::CPPTYPE_STRING;
+
+      if (should_check_bit) {
+        printer->Print(vars, "if (_has_bit($index$)) {\n");
+        printer->Indent();
+      }
+
+      field_generators_.get(field).GenerateClearingCode(printer);
+
+      if (should_check_bit) {
+        printer->Outdent();
+        printer->Print("}\n");
+      }
+    }
+  }
+
+  if (last_index >= 0) {
+    printer->Outdent();
+    printer->Print("}\n");
+  }
+
+  // Repeated fields don't use _has_bits_ so we clear them in a separate
+  // pass.
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (field->is_repeated()) {
+      field_generators_.get(field).GenerateClearingCode(printer);
+    }
+  }
+
+  printer->Print(
+    "::memset(_has_bits_, 0, sizeof(_has_bits_));\n"
+    "mutable_unknown_fields()->Clear();\n");
+
+  printer->Outdent();
+  printer->Print("}\n");
+}
+
+void MessageGenerator::
+GenerateMergeFrom(io::Printer* printer) {
+  // Generate the generalized MergeFrom (aka that which takes in the Message
+  // base class as a parameter).
+  printer->Print(
+    "void $classname$::MergeFrom(const ::google::protobuf::Message& from) {\n"
+    "  GOOGLE_CHECK_NE(&from, this);\n",
+    "classname", classname_);
+  printer->Indent();
+
+  if (descriptor_->field_count() > 0) {
+    // Cast the message to the proper type. If we find that the message is
+    // *not* of the proper type, we can still call Merge via the reflection
+    // system, as the GOOGLE_CHECK above ensured that we have the same descriptor
+    // for each message.
+    printer->Print(
+      "const $classname$* source =\n"
+      "  ::google::protobuf::internal::dynamic_cast_if_available<const $classname$*>(\n"
+      "    &from);\n"
+      "if (source == NULL) {\n"
+      "  ::google::protobuf::internal::ReflectionOps::Merge(\n"
+      "    descriptor(), *from.GetReflection(), &_reflection_);\n"
+      "} else {\n"
+      "  MergeFrom(*source);\n"
+      "}\n",
+      "classname", classname_);
+  }
+
+  printer->Outdent();
+  printer->Print("}\n\n");
+
+  // Generate the class-specific MergeFrom, which avoids the GOOGLE_CHECK and cast.
+  printer->Print(
+    "void $classname$::MergeFrom(const $classname$& from) {\n"
+    "  GOOGLE_CHECK_NE(&from, this);\n",
+    "classname", classname_);
+  printer->Indent();
+
+  // Merge Repeated fields. These fields do not require a
+  // check as we can simply iterate over them.
+  for (int i = 0; i < descriptor_->field_count(); ++i) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (field->is_repeated()) {
+      field_generators_.get(field).GenerateMergingCode(printer);
+    }
+  }
+
+  // Merge Optional and Required fields (after a _has_bit check).
+  int last_index = -1;
+
+  for (int i = 0; i < descriptor_->field_count(); ++i) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (!field->is_repeated()) {
+      map<string, string> vars;
+      vars["index"] = SimpleItoa(field->index());
+
+      // See above in GenerateClear for an explanation of this.
+      if (i / 8 != last_index / 8 || last_index < 0) {
+        if (last_index >= 0) {
+          printer->Outdent();
+          printer->Print("}\n");
+        }
+        printer->Print(vars,
+          "if (from._has_bits_[$index$ / 32] & (0xffu << ($index$ % 32))) {\n");
+        printer->Indent();
+      }
+
+      last_index = i;
+
+      printer->Print(vars,
+        "if (from._has_bit($index$)) {\n");
+      printer->Indent();
+
+      field_generators_.get(field).GenerateMergingCode(printer);
+
+      printer->Outdent();
+      printer->Print("}\n");
+    }
+  }
+
+  if (last_index >= 0) {
+    printer->Outdent();
+    printer->Print("}\n");
+  }
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print("_extensions_.MergeFrom(from._extensions_);\n");
+  }
+
+  printer->Print(
+    "mutable_unknown_fields()->MergeFrom(from.unknown_fields());\n");
+
+  printer->Outdent();
+  printer->Print("}\n");
+}
+
+void MessageGenerator::
+GenerateCopyFrom(io::Printer* printer) {
+  // Generate the generalized CopyFrom (aka that which takes in the Message
+  // base class as a parameter).
+  printer->Print(
+    "void $classname$::CopyFrom(const ::google::protobuf::Message& from) {\n",
+    "classname", classname_);
+  printer->Indent();
+
+  printer->Print(
+    "if (&from == this) return;\n"
+    "Clear();\n"
+    "MergeFrom(from);\n");
+
+  printer->Outdent();
+  printer->Print("}\n\n");
+
+  // Generate the class-specific CopyFrom.
+  printer->Print(
+    "void $classname$::CopyFrom(const $classname$& from) {\n",
+    "classname", classname_);
+  printer->Indent();
+
+  printer->Print(
+    "if (&from == this) return;\n"
+    "Clear();\n"
+    "MergeFrom(from);\n");
+
+  printer->Outdent();
+  printer->Print("}\n");
+}
+
+void MessageGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) {
+  if (descriptor_->options().message_set_wire_format()) {
+    // For message_set_wire_format, we don't generate a parser, for two
+    // reasons:
+    // - WireFormat already needs to special-case this, and we'd like to
+    //   avoid having multiple implementations of MessageSet wire format
+    //   lying around the code base.
+    // - All fields are extensions, and extension parsing falls back to
+    //   reflection anyway, so it wouldn't be any faster.
+    printer->Print(
+      "bool $classname$::MergePartialFromCodedStream(\n"
+      "    ::google::protobuf::io::CodedInputStream* input) {\n"
+      "  return ::google::protobuf::internal::WireFormat::ParseAndMergePartial(\n"
+      "    descriptor(), input, &_reflection_);\n"
+      "}\n",
+      "classname", classname_);
+    return;
+  }
+
+  printer->Print(
+    "bool $classname$::MergePartialFromCodedStream(\n"
+    "    ::google::protobuf::io::CodedInputStream* input) {\n"
+    "#define DO_(EXPRESSION) if (!(EXPRESSION)) return false\n"
+    "  ::google::protobuf::uint32 tag;\n"
+    "  while ((tag = input->ReadTag()) != 0) {\n",
+    "classname", classname_);
+
+  printer->Indent();
+  printer->Indent();
+
+  if (descriptor_->field_count() > 0) {
+    // We don't even want to print the switch() if we have no fields because
+    // MSVC dislikes switch() statements that contain only a default value.
+
+    // Note:  If we just switched on the tag rather than the field number, we
+    // could avoid the need for the if() to check the wire type at the beginning
+    // of each case.  However, this is actually a bit slower in practice as it
+    // creates a jump table that is 8x larger and sparser, and meanwhile the
+    // if()s are highly predictable.
+    printer->Print(
+      "switch (::google::protobuf::internal::WireFormat::GetTagFieldNumber(tag)) {\n");
+
+    printer->Indent();
+
+    scoped_array<const FieldDescriptor*> ordered_fields(
+      SortFieldsByNumber(descriptor_));
+
+    for (int i = 0; i < descriptor_->field_count(); i++) {
+      const FieldDescriptor* field = ordered_fields[i];
+
+      PrintFieldComment(printer, field);
+
+      printer->Print(
+        "case $number$: {\n"
+        "  if (::google::protobuf::internal::WireFormat::GetTagWireType(tag) !=\n"
+        "      ::google::protobuf::internal::WireFormat::WIRETYPE_$wiretype$) {\n"
+        "    goto handle_uninterpreted;\n"
+        "  }\n",
+        "number", SimpleItoa(field->number()),
+        "wiretype", kWireTypeNames[
+          WireFormat::WireTypeForFieldType(field->type())]);
+
+      if (i > 0 || field->is_repeated()) {
+        printer->Print(
+          " parse_$name$:\n",
+          "name", field->name());
+      }
+
+      printer->Indent();
+
+      field_generators_.get(field).GenerateMergeFromCodedStream(printer);
+
+      // switch() is slow since it can't be predicted well.  Insert some if()s
+      // here that attempt to predict the next tag.
+      if (field->is_repeated()) {
+        // Expect repeats of this field.
+        printer->Print(
+          "if (input->ExpectTag($tag$)) goto parse_$name$;\n",
+          "tag", SimpleItoa(WireFormat::MakeTag(field)),
+          "name", field->name());
+      }
+
+      if (i + 1 < descriptor_->field_count()) {
+        // Expect the next field in order.
+        const FieldDescriptor* next_field = ordered_fields[i + 1];
+        printer->Print(
+          "if (input->ExpectTag($next_tag$)) goto parse_$next_name$;\n",
+          "next_tag", SimpleItoa(WireFormat::MakeTag(next_field)),
+          "next_name", next_field->name());
+      } else {
+        // Expect EOF.
+        // TODO(kenton):  Expect group end-tag?
+        printer->Print(
+          "if (input->ExpectAtEnd()) return true;\n");
+      }
+
+      printer->Print(
+        "break;\n");
+
+      printer->Outdent();
+      printer->Print("}\n\n");
+    }
+
+    printer->Print(
+      "default: {\n"
+      "handle_uninterpreted:\n");
+    printer->Indent();
+  }
+
+  // Is this an end-group tag?  If so, this must be the end of the message.
+  printer->Print(
+    "if (::google::protobuf::internal::WireFormat::GetTagWireType(tag) ==\n"
+    "    ::google::protobuf::internal::WireFormat::WIRETYPE_END_GROUP) {\n"
+    "  return true;\n"
+    "}\n");
+
+  // Handle extension ranges.
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "if (");
+    for (int i = 0; i < descriptor_->extension_range_count(); i++) {
+      const Descriptor::ExtensionRange* range =
+        descriptor_->extension_range(i);
+      if (i > 0) printer->Print(" &&\n    ");
+
+      uint32 start_tag = WireFormat::MakeTag(
+        range->start, static_cast<WireFormat::WireType>(0));
+      uint32 end_tag = WireFormat::MakeTag(
+        range->end, static_cast<WireFormat::WireType>(0));
+
+      if (range->end > FieldDescriptor::kMaxNumber) {
+        printer->Print(
+          "($start$u <= tag)",
+          "start", SimpleItoa(start_tag));
+      } else {
+        printer->Print(
+          "($start$u <= tag && tag < $end$u)",
+          "start", SimpleItoa(start_tag),
+          "end", SimpleItoa(end_tag));
+      }
+    }
+    printer->Print(") {\n"
+      "  DO_(_extensions_.ParseField(tag, input, &_reflection_));\n"
+      "  continue;\n"
+      "}\n");
+  }
+
+  // We really don't recognize this tag.  Skip it.
+  printer->Print(
+    "DO_(::google::protobuf::internal::WireFormat::SkipField(\n"
+    "      input, tag, mutable_unknown_fields()));\n");
+
+  if (descriptor_->field_count() > 0) {
+    printer->Print("break;\n");
+    printer->Outdent();
+    printer->Print("}\n");    // default:
+    printer->Outdent();
+    printer->Print("}\n");    // switch
+  }
+
+  printer->Outdent();
+  printer->Outdent();
+  printer->Print(
+    "  }\n"                   // while
+    "  return true;\n"
+    "#undef DO_\n"
+    "}\n");
+}
+
+void MessageGenerator::GenerateSerializeOneField(
+    io::Printer* printer, const FieldDescriptor* field) {
+  PrintFieldComment(printer, field);
+
+  if (field->is_repeated()) {
+    printer->Print(
+      "for (int i = 0; i < $name$_.size(); i++) {\n",
+      "name", FieldName(field));
+  } else {
+    printer->Print(
+      "if (_has_bit($index$)) {\n",
+      "index", SimpleItoa(field->index()));
+  }
+
+  printer->Indent();
+
+  field_generators_.get(field).GenerateSerializeWithCachedSizes(printer);
+
+  printer->Outdent();
+  printer->Print("}\n\n");
+}
+
+void MessageGenerator::GenerateSerializeOneExtensionRange(
+    io::Printer* printer, const Descriptor::ExtensionRange* range) {
+  map<string, string> vars;
+  vars["start"] = SimpleItoa(range->start);
+  vars["end"] = SimpleItoa(range->end);
+  printer->Print(vars,
+    "// Extension range [$start$, $end$)\n"
+    "DO_(_extensions_.SerializeWithCachedSizes(\n"
+    "    $start$, $end$, &_reflection_, output));\n\n");
+}
+
+void MessageGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) {
+  printer->Print(
+    "bool $classname$::SerializeWithCachedSizes(\n"
+    "    ::google::protobuf::io::CodedOutputStream* output) const {\n"
+    "#define DO_(EXPRESSION) if (!(EXPRESSION)) return false\n",
+    "classname", classname_);
+  printer->Indent();
+
+  scoped_array<const FieldDescriptor*> ordered_fields(
+    SortFieldsByNumber(descriptor_));
+
+  vector<const Descriptor::ExtensionRange*> sorted_extensions;
+  for (int i = 0; i < descriptor_->extension_range_count(); ++i) {
+    sorted_extensions.push_back(descriptor_->extension_range(i));
+  }
+  sort(sorted_extensions.begin(), sorted_extensions.end(),
+       ExtensionRangeSorter());
+
+  // Merge the fields and the extension ranges, both sorted by field number.
+  int i, j;
+  for (i = 0, j = 0;
+       i < descriptor_->field_count() || j < sorted_extensions.size();
+       ) {
+    if (i == descriptor_->field_count()) {
+      GenerateSerializeOneExtensionRange(printer, sorted_extensions[j++]);
+    } else if (j == sorted_extensions.size()) {
+      GenerateSerializeOneField(printer, ordered_fields[i++]);
+    } else if (ordered_fields[i]->number() < sorted_extensions[j]->start) {
+      GenerateSerializeOneField(printer, ordered_fields[i++]);
+    } else {
+      GenerateSerializeOneExtensionRange(printer, sorted_extensions[j++]);
+    }
+  }
+
+  printer->Print("if (!unknown_fields().empty()) {\n");
+  printer->Indent();
+  if (descriptor_->options().message_set_wire_format()) {
+    printer->Print(
+      "DO_(::google::protobuf::internal::WireFormat::SerializeUnknownMessageSetItems(\n"
+      "    unknown_fields(), output));\n");
+  } else {
+    printer->Print(
+      "DO_(::google::protobuf::internal::WireFormat::SerializeUnknownFields(\n"
+      "    unknown_fields(), output));\n");
+  }
+  printer->Outdent();
+  printer->Print(
+    "}\n"
+    "return true;\n");
+
+  printer->Outdent();
+  printer->Print(
+    "#undef DO_\n"
+    "}\n");
+}
+
+void MessageGenerator::
+GenerateByteSize(io::Printer* printer) {
+  printer->Print(
+    "int $classname$::ByteSize() const {\n",
+    "classname", classname_);
+  printer->Indent();
+  printer->Print(
+    "int total_size = 0;\n"
+    "\n");
+
+  int last_index = -1;
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (!field->is_repeated()) {
+      // See above in GenerateClear for an explanation of this.
+      // TODO(kenton):  Share code?  Unclear how to do so without
+      //   over-engineering.
+      if ((i / 8) != (last_index / 8) ||
+          last_index < 0) {
+        if (last_index >= 0) {
+          printer->Outdent();
+          printer->Print("}\n");
+        }
+        printer->Print(
+          "if (_has_bits_[$index$ / 32] & (0xffu << ($index$ % 32))) {\n",
+          "index", SimpleItoa(field->index()));
+        printer->Indent();
+      }
+      last_index = i;
+
+      PrintFieldComment(printer, field);
+
+      printer->Print(
+        "if (has_$name$()) {\n",
+        "name", FieldName(field));
+      printer->Indent();
+
+      field_generators_.get(field).GenerateByteSize(printer);
+
+      printer->Outdent();
+      printer->Print(
+        "}\n"
+        "\n");
+    }
+  }
+
+  if (last_index >= 0) {
+    printer->Outdent();
+    printer->Print("}\n");
+  }
+
+  // Repeated fields don't use _has_bits_ so we count them in a separate
+  // pass.
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (field->is_repeated()) {
+      PrintFieldComment(printer, field);
+      field_generators_.get(field).GenerateByteSize(printer);
+      printer->Print("\n");
+    }
+  }
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "total_size += _extensions_.ByteSize(&_reflection_);\n"
+      "\n");
+  }
+
+  printer->Print("if (!unknown_fields().empty()) {\n");
+  printer->Indent();
+  if (descriptor_->options().message_set_wire_format()) {
+    printer->Print(
+      "total_size +=\n"
+      "  ::google::protobuf::internal::WireFormat::ComputeUnknownMessageSetItemsSize(\n"
+      "    unknown_fields());\n");
+  } else {
+    printer->Print(
+      "total_size +=\n"
+      "  ::google::protobuf::internal::WireFormat::ComputeUnknownFieldsSize(\n"
+      "    unknown_fields());\n");
+  }
+  printer->Outdent();
+  printer->Print("}\n");
+
+  // We update _cached_size_ even though this is a const method.  In theory,
+  // this is not thread-compatible, because concurrent writes have undefined
+  // results.  In practice, since any concurrent writes will be writing the
+  // exact same value, it works on all common processors.  In a future version
+  // of C++, _cached_size_ should be made into an atomic<int>.
+  printer->Print(
+    "_cached_size_ = total_size;\n"
+    "return total_size;\n");
+
+  printer->Outdent();
+  printer->Print("}\n");
+}
+
+void MessageGenerator::
+GenerateIsInitialized(io::Printer* printer) {
+  printer->Print(
+    "bool $classname$::IsInitialized() const {\n",
+    "classname", classname_);
+  printer->Indent();
+
+  // Check that all required fields in this message are set.  We can do this
+  // most efficiently by checking 32 "has bits" at a time.
+  int has_bits_array_size = (descriptor_->field_count() + 31) / 32;
+  for (int i = 0; i < has_bits_array_size; i++) {
+    uint32 mask = 0;
+    for (int bit = 0; bit < 32; bit++) {
+      int index = i * 32 + bit;
+      if (index >= descriptor_->field_count()) break;
+      const FieldDescriptor* field = descriptor_->field(index);
+
+      if (field->is_required()) {
+        mask |= 1 << bit;
+      }
+    }
+
+    if (mask != 0) {
+      char buffer[kFastToBufferSize];
+      printer->Print(
+        "if ((_has_bits_[$i$] & 0x$mask$) != 0x$mask$) return false;\n",
+        "i", SimpleItoa(i),
+        "mask", FastHex32ToBuffer(mask, buffer));
+    }
+  }
+
+  // Now check that all embedded messages are initialized.
+  printer->Print("\n");
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+    if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE &&
+        HasRequiredFields(field->message_type())) {
+      if (field->is_repeated()) {
+        printer->Print(
+          "for (int i = 0; i < $name$_size(); i++) {\n"
+          "  if (!this->$name$(i).IsInitialized()) return false;\n"
+          "}\n",
+          "name", FieldName(field));
+      } else {
+        printer->Print(
+          "if (has_$name$()) {\n"
+          "  if (!this->$name$().IsInitialized()) return false;\n"
+          "}\n",
+          "name", FieldName(field));
+      }
+    }
+  }
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "\n"
+      "if (!_extensions_.IsInitialized()) return false;");
+  }
+
+  printer->Outdent();
+  printer->Print(
+    "  return true;\n"
+    "}\n");
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_message.h b/src/google/protobuf/compiler/cpp/cpp_message.h
new file mode 100644
index 00000000..904e0487
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_message.h
@@ -0,0 +1,125 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_MESSAGE_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_MESSAGE_H__
+
+#include <string>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/compiler/cpp/cpp_field.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class EnumGenerator;           // enum.h
+class ExtensionGenerator;      // extension.h
+
+class MessageGenerator {
+ public:
+  // See generator.cc for the meaning of dllexport_decl.
+  explicit MessageGenerator(const Descriptor* descriptor,
+                            const string& dllexport_decl);
+  ~MessageGenerator();
+
+  // Header stuff.
+
+  // Generate foward declarations for this class and all its nested types.
+  void GenerateForwardDeclaration(io::Printer* printer);
+
+  // Generate definitions of all nested enums (must come before class
+  // definitions because those classes use the enums definitions).
+  void GenerateEnumDefinitions(io::Printer* printer);
+
+  // Generate definitions for this class and all its nested types.
+  void GenerateClassDefinition(io::Printer* printer);
+
+  // Generate definitions of inline methods (placed at the end of the header
+  // file).
+  void GenerateInlineMethods(io::Printer* printer);
+
+  // Source file stuff.
+
+  // Generate code which declares all the global descriptor pointers which
+  // will be initialized by the methods below.
+  void GenerateDescriptorDeclarations(io::Printer* printer);
+
+  // Generate code that initializes the global variable storing the message's
+  // descriptor.
+  void GenerateDescriptorInitializer(io::Printer* printer, int index);
+
+  // Generate all non-inline methods for this class.
+  void GenerateClassMethods(io::Printer* printer);
+
+ private:
+  // Generate declarations and definitions of accessors for fields.
+  void GenerateFieldAccessorDeclarations(io::Printer* printer);
+  void GenerateFieldAccessorDefinitions(io::Printer* printer);
+
+  // Generate the field offsets array.
+  void GenerateOffsets(io::Printer* printer);
+
+  // Generate constructors and destructor.
+  void GenerateStructors(io::Printer* printer);
+
+  // Generate the member initializer list for the constructors. The member
+  // initializer list is shared between the default constructor and the copy
+  // constructor.
+  void GenerateInitializerList(io::Printer* printer);
+
+  // Generate standard Message methods.
+  void GenerateClear(io::Printer* printer);
+  void GenerateMergeFromCodedStream(io::Printer* printer);
+  void GenerateSerializeWithCachedSizes(io::Printer* printer);
+  void GenerateByteSize(io::Printer* printer);
+  void GenerateMergeFrom(io::Printer* printer);
+  void GenerateCopyFrom(io::Printer* printer);
+  void GenerateIsInitialized(io::Printer* printer);
+
+  // Helpers for GenerateSerializeWithCachedSizes().
+  void GenerateSerializeOneField(io::Printer* printer,
+                                 const FieldDescriptor* field);
+  void GenerateSerializeOneExtensionRange(
+      io::Printer* printer, const Descriptor::ExtensionRange* range);
+
+  const Descriptor* descriptor_;
+  string classname_;
+  string dllexport_decl_;
+  FieldGeneratorMap field_generators_;
+  scoped_array<scoped_ptr<MessageGenerator> > nested_generators_;
+  scoped_array<scoped_ptr<EnumGenerator> > enum_generators_;
+  scoped_array<scoped_ptr<ExtensionGenerator> > extension_generators_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(MessageGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_MESSAGE_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_message_field.cc b/src/google/protobuf/compiler/cpp/cpp_message_field.cc
new file mode 100644
index 00000000..501ca569
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_message_field.cc
@@ -0,0 +1,229 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_message_field.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/wire_format_inl.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+using internal::WireFormat;
+
+namespace {
+
+// TODO(kenton):  Factor out a "SetCommonFieldVariables()" to get rid of
+//   repeat code between this and the other field types.
+void SetMessageVariables(const FieldDescriptor* descriptor,
+                         map<string, string>* variables) {
+  (*variables)["name"] = FieldName(descriptor);
+  (*variables)["type"] = ClassName(descriptor->message_type(), true);
+  (*variables)["index"] = SimpleItoa(descriptor->index());
+  (*variables)["number"] = SimpleItoa(descriptor->number());
+  (*variables)["classname"] = ClassName(FieldScope(descriptor), false);
+  (*variables)["declared_type"] = DeclaredTypeMethodName(descriptor->type());
+  (*variables)["tag_size"] = SimpleItoa(
+    WireFormat::TagSize(descriptor->number(), descriptor->type()));
+}
+
+}  // namespace
+
+// ===================================================================
+
+MessageFieldGenerator::
+MessageFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetMessageVariables(descriptor, &variables_);
+}
+
+MessageFieldGenerator::~MessageFieldGenerator() {}
+
+void MessageFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_, "$type$* $name$_;\n");
+}
+
+void MessageFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const $type$& $name$() const;\n"
+    "inline $type$* mutable_$name$();\n");
+}
+
+void MessageFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const $type$& $classname$::$name$() const {\n"
+    "  return $name$_ != NULL ? *$name$_ : *default_instance_.$name$_;\n"
+    "}\n"
+    "inline $type$* $classname$::mutable_$name$() {\n"
+    "  _set_bit($index$);\n"
+    "  if ($name$_ == NULL) $name$_ = new $type$;\n"
+    "  return $name$_;\n"
+    "}\n");
+}
+
+void MessageFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if ($name$_ != NULL) $name$_->$type$::Clear();\n");
+}
+
+void MessageFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "mutable_$name$()->$type$::MergeFrom(from.$name$());\n");
+}
+
+void MessageFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  printer->Print(variables_, ",\n$name$_(NULL)");
+}
+
+void MessageFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  if (descriptor_->type() == FieldDescriptor::TYPE_MESSAGE) {
+    printer->Print(variables_,
+      "DO_(::google::protobuf::internal::WireFormat::ReadMessageNoVirtual(\n"
+      "     input, mutable_$name$()));\n");
+  } else {
+    printer->Print(variables_,
+      "DO_(::google::protobuf::internal::WireFormat::ReadGroupNoVirtual("
+        "$number$, input, mutable_$name$()));\n");
+  }
+}
+
+void MessageFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Write$declared_type$NoVirtual("
+      "$number$, this->$name$(), output));\n");
+}
+
+void MessageFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  printer->Print(variables_,
+    "total_size += $tag_size$ +\n"
+    "  ::google::protobuf::internal::WireFormat::$declared_type$SizeNoVirtual(\n"
+    "    this->$name$());\n");
+}
+
+// ===================================================================
+
+RepeatedMessageFieldGenerator::
+RepeatedMessageFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetMessageVariables(descriptor, &variables_);
+}
+
+RepeatedMessageFieldGenerator::~RepeatedMessageFieldGenerator() {}
+
+void RepeatedMessageFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "::google::protobuf::RepeatedPtrField< $type$ > $name$_;\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedPtrField< $type$ >& $name$() const;\n"
+    "inline ::google::protobuf::RepeatedPtrField< $type$ >* mutable_$name$();\n"
+    "inline const $type$& $name$(int index) const;\n"
+    "inline $type$* mutable_$name$(int index);\n"
+    "inline $type$* add_$name$();\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedPtrField< $type$ >&\n"
+    "$classname$::$name$() const {\n"
+    "  return $name$_;\n"
+    "}\n"
+    "inline ::google::protobuf::RepeatedPtrField< $type$ >*\n"
+    "$classname$::mutable_$name$() {\n"
+    "  return &$name$_;\n"
+    "}\n"
+    "inline const $type$& $classname$::$name$(int index) const {\n"
+    "  return $name$_.Get(index);\n"
+    "}\n"
+    "inline $type$* $classname$::mutable_$name$(int index) {\n"
+    "  return $name$_.Mutable(index);\n"
+    "}\n"
+    "inline $type$* $classname$::add_$name$() {\n"
+    "  return $name$_.Add();\n"
+    "}\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.Clear();\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.MergeFrom(from.$name$_);\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  // Not needed for repeated fields.
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  if (descriptor_->type() == FieldDescriptor::TYPE_MESSAGE) {
+    printer->Print(variables_,
+      "DO_(::google::protobuf::internal::WireFormat::ReadMessageNoVirtual(\n"
+      "     input, add_$name$()));\n");
+  } else {
+    printer->Print(variables_,
+      "DO_(::google::protobuf::internal::WireFormat::ReadGroupNoVirtual("
+        "$number$, input, add_$name$()));\n");
+  }
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Write$declared_type$NoVirtual("
+      "$number$, this->$name$(i), output));\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  printer->Print(variables_,
+    "total_size += $tag_size$ * $name$_size();\n"
+    "for (int i = 0; i < $name$_size(); i++) {\n"
+    "  total_size +=\n"
+    "    ::google::protobuf::internal::WireFormat::$declared_type$SizeNoVirtual(\n"
+    "      this->$name$(i));\n"
+    "}\n");
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_message_field.h b/src/google/protobuf/compiler/cpp/cpp_message_field.h
new file mode 100644
index 00000000..a2be6b65
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_message_field.h
@@ -0,0 +1,84 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_MESSAGE_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_MESSAGE_FIELD_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/compiler/cpp/cpp_field.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class MessageFieldGenerator : public FieldGenerator {
+ public:
+  explicit MessageFieldGenerator(const FieldDescriptor* descriptor);
+  ~MessageFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(MessageFieldGenerator);
+};
+
+class RepeatedMessageFieldGenerator : public FieldGenerator {
+ public:
+  explicit RepeatedMessageFieldGenerator(const FieldDescriptor* descriptor);
+  ~RepeatedMessageFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(RepeatedMessageFieldGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_MESSAGE_FIELD_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_primitive_field.cc b/src/google/protobuf/compiler/cpp/cpp_primitive_field.cc
new file mode 100644
index 00000000..312ed031
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_primitive_field.cc
@@ -0,0 +1,294 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_primitive_field.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/wire_format_inl.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+using internal::WireFormat;
+
+namespace {
+
+// For encodings with fixed sizes, returns that size in bytes.  Otherwise
+// returns -1.
+int FixedSize(FieldDescriptor::Type type) {
+  switch (type) {
+    case FieldDescriptor::TYPE_INT32   : return -1;
+    case FieldDescriptor::TYPE_INT64   : return -1;
+    case FieldDescriptor::TYPE_UINT32  : return -1;
+    case FieldDescriptor::TYPE_UINT64  : return -1;
+    case FieldDescriptor::TYPE_SINT32  : return -1;
+    case FieldDescriptor::TYPE_SINT64  : return -1;
+    case FieldDescriptor::TYPE_FIXED32 : return WireFormat::kFixed32Size;
+    case FieldDescriptor::TYPE_FIXED64 : return WireFormat::kFixed64Size;
+    case FieldDescriptor::TYPE_SFIXED32: return WireFormat::kSFixed32Size;
+    case FieldDescriptor::TYPE_SFIXED64: return WireFormat::kSFixed64Size;
+    case FieldDescriptor::TYPE_FLOAT   : return WireFormat::kFloatSize;
+    case FieldDescriptor::TYPE_DOUBLE  : return WireFormat::kDoubleSize;
+
+    case FieldDescriptor::TYPE_BOOL    : return WireFormat::kBoolSize;
+    case FieldDescriptor::TYPE_ENUM    : return -1;
+
+    case FieldDescriptor::TYPE_STRING  : return -1;
+    case FieldDescriptor::TYPE_BYTES   : return -1;
+    case FieldDescriptor::TYPE_GROUP   : return -1;
+    case FieldDescriptor::TYPE_MESSAGE : return -1;
+
+    // No default because we want the compiler to complain if any new
+    // types are added.
+  }
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return -1;
+}
+
+string DefaultValue(const FieldDescriptor* field) {
+  switch (field->cpp_type()) {
+    case FieldDescriptor::CPPTYPE_INT32:
+      return SimpleItoa(field->default_value_int32());
+    case FieldDescriptor::CPPTYPE_UINT32:
+      return SimpleItoa(field->default_value_uint32()) + "u";
+    case FieldDescriptor::CPPTYPE_INT64:
+      return "GOOGLE_LONGLONG(" + SimpleItoa(field->default_value_int64()) + ")";
+    case FieldDescriptor::CPPTYPE_UINT64:
+      return "GOOGLE_ULONGLONG(" + SimpleItoa(field->default_value_uint64())+ ")";
+    case FieldDescriptor::CPPTYPE_DOUBLE:
+      return SimpleDtoa(field->default_value_double());
+    case FieldDescriptor::CPPTYPE_FLOAT:
+      return SimpleFtoa(field->default_value_float());
+    case FieldDescriptor::CPPTYPE_BOOL:
+      return field->default_value_bool() ? "true" : "false";
+
+    case FieldDescriptor::CPPTYPE_ENUM:
+    case FieldDescriptor::CPPTYPE_STRING:
+    case FieldDescriptor::CPPTYPE_MESSAGE:
+      GOOGLE_LOG(FATAL) << "Shouldn't get here.";
+      return "";
+  }
+  // Can't actually get here; make compiler happy.  (We could add a default
+  // case above but then we wouldn't get the nice compiler warning when a
+  // new type is added.)
+  return "";
+}
+
+// TODO(kenton):  Factor out a "SetCommonFieldVariables()" to get rid of
+//   repeat code between this and the other field types.
+void SetPrimitiveVariables(const FieldDescriptor* descriptor,
+                           map<string, string>* variables) {
+  (*variables)["name"] = FieldName(descriptor);
+  (*variables)["type"] = PrimitiveTypeName(descriptor->cpp_type());
+  (*variables)["default"] = DefaultValue(descriptor);
+  (*variables)["index"] = SimpleItoa(descriptor->index());
+  (*variables)["number"] = SimpleItoa(descriptor->number());
+  (*variables)["classname"] = ClassName(FieldScope(descriptor), false);
+  (*variables)["declared_type"] = DeclaredTypeMethodName(descriptor->type());
+  (*variables)["tag_size"] = SimpleItoa(
+    WireFormat::TagSize(descriptor->number(), descriptor->type()));
+
+  int fixed_size = FixedSize(descriptor->type());
+  if (fixed_size != -1) {
+    (*variables)["fixed_size"] = SimpleItoa(fixed_size);
+  }
+}
+
+}  // namespace
+
+// ===================================================================
+
+PrimitiveFieldGenerator::
+PrimitiveFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetPrimitiveVariables(descriptor, &variables_);
+}
+
+PrimitiveFieldGenerator::~PrimitiveFieldGenerator() {}
+
+void PrimitiveFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_, "$type$ $name$_;\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline $type$ $name$() const;\n"
+    "inline void set_$name$($type$ value);\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline $type$ $classname$::$name$() const {\n"
+    "  return $name$_;\n"
+    "}\n"
+    "inline void $classname$::set_$name$($type$ value) {\n"
+    "  _set_bit($index$);\n"
+    "  $name$_ = value;\n"
+    "}\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_ = $default$;\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_, "set_$name$(from.$name$());\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  printer->Print(variables_, ",\n$name$_($default$)");
+}
+
+void PrimitiveFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Read$declared_type$(\n"
+    "      input, &$name$_));\n"
+    "_set_bit($index$);\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Write$declared_type$("
+      "$number$, this->$name$(), output));\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  int fixed_size = FixedSize(descriptor_->type());
+  if (fixed_size == -1) {
+    printer->Print(variables_,
+      "total_size += $tag_size$ +\n"
+      "  ::google::protobuf::internal::WireFormat::$declared_type$Size(\n"
+      "    this->$name$());\n");
+  } else {
+    printer->Print(variables_,
+      "total_size += $tag_size$ + $fixed_size$;\n");
+  }
+}
+
+// ===================================================================
+
+RepeatedPrimitiveFieldGenerator::
+RepeatedPrimitiveFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetPrimitiveVariables(descriptor, &variables_);
+}
+
+RepeatedPrimitiveFieldGenerator::~RepeatedPrimitiveFieldGenerator() {}
+
+void RepeatedPrimitiveFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "::google::protobuf::RepeatedField< $type$ > $name$_;\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedField< $type$ >& $name$() const;\n"
+    "inline ::google::protobuf::RepeatedField< $type$ >* mutable_$name$();\n"
+    "inline $type$ $name$(int index) const;\n"
+    "inline void set_$name$(int index, $type$ value);\n"
+    "inline void add_$name$($type$ value);\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedField< $type$ >&\n"
+    "$classname$::$name$() const {\n"
+    "  return $name$_;\n"
+    "}\n"
+    "inline ::google::protobuf::RepeatedField< $type$ >*\n"
+    "$classname$::mutable_$name$() {\n"
+    "  return &$name$_;\n"
+    "}\n"
+    "inline $type$ $classname$::$name$(int index) const {\n"
+    "  return $name$_.Get(index);\n"
+    "}\n"
+    "inline void $classname$::set_$name$(int index, $type$ value) {\n"
+    "  $name$_.Set(index, value);\n"
+    "}\n"
+    "inline void $classname$::add_$name$($type$ value) {\n"
+    "  $name$_.Add(value);\n"
+    "}\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.Clear();\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.MergeFrom(from.$name$_);\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  // Not needed for repeated fields.
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  printer->Print(variables_,
+    "$type$ value;\n"
+    "DO_(::google::protobuf::internal::WireFormat::Read$declared_type$(input, &value));\n"
+    "add_$name$(value);\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Write$declared_type$("
+      "$number$, this->$name$(i), output));\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  int fixed_size = FixedSize(descriptor_->type());
+  if (fixed_size == -1) {
+    printer->Print(variables_,
+      "total_size += $tag_size$ * $name$_size();\n"
+      "for (int i = 0; i < $name$_size(); i++) {\n"
+      "  total_size += ::google::protobuf::internal::WireFormat::$declared_type$Size(\n"
+      "    this->$name$(i));\n"
+      "}\n");
+  } else {
+    printer->Print(variables_,
+      "total_size += ($tag_size$ + $fixed_size$) * $name$_size();\n");
+  }
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_primitive_field.h b/src/google/protobuf/compiler/cpp/cpp_primitive_field.h
new file mode 100644
index 00000000..832b2411
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_primitive_field.h
@@ -0,0 +1,84 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_PRIMITIVE_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_PRIMITIVE_FIELD_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/compiler/cpp/cpp_field.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class PrimitiveFieldGenerator : public FieldGenerator {
+ public:
+  explicit PrimitiveFieldGenerator(const FieldDescriptor* descriptor);
+  ~PrimitiveFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(PrimitiveFieldGenerator);
+};
+
+class RepeatedPrimitiveFieldGenerator : public FieldGenerator {
+ public:
+  explicit RepeatedPrimitiveFieldGenerator(const FieldDescriptor* descriptor);
+  ~RepeatedPrimitiveFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(RepeatedPrimitiveFieldGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_PRIMITIVE_FIELD_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_service.cc b/src/google/protobuf/compiler/cpp/cpp_service.cc
new file mode 100644
index 00000000..124f3b43
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_service.cc
@@ -0,0 +1,318 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_service.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+ServiceGenerator::ServiceGenerator(const ServiceDescriptor* descriptor,
+                                   const string& dllexport_decl)
+  : descriptor_(descriptor) {
+  vars_["classname"] = descriptor_->name();
+  vars_["full_name"] = descriptor_->full_name();
+  if (dllexport_decl.empty()) {
+    vars_["dllexport"] = "";
+  } else {
+    vars_["dllexport"] = dllexport_decl + " ";
+  }
+}
+
+ServiceGenerator::~ServiceGenerator() {}
+
+void ServiceGenerator::GenerateDeclarations(io::Printer* printer) {
+  // Forward-declare the stub type.
+  printer->Print(vars_,
+    "class $classname$_Stub;\n"
+    "\n");
+
+  GenerateInterface(printer);
+  GenerateStubDefinition(printer);
+}
+
+void ServiceGenerator::GenerateInterface(io::Printer* printer) {
+  printer->Print(vars_,
+    "class $dllexport$$classname$ : public ::google::protobuf::Service {\n"
+    " protected:\n"
+    "  // This class should be treated as an abstract interface.\n"
+    "  inline $classname$() {};\n"
+    " public:\n"
+    "  virtual ~$classname$();\n");
+  printer->Indent();
+
+  printer->Print(vars_,
+    "\n"
+    "typedef $classname$_Stub Stub;\n"
+    "\n"
+    "static const ::google::protobuf::ServiceDescriptor* descriptor();\n"
+    "\n");
+
+  GenerateMethodSignatures(VIRTUAL, printer);
+
+  printer->Print(
+    "\n"
+    "// implements Service ----------------------------------------------\n"
+    "\n"
+    "const ::google::protobuf::ServiceDescriptor* GetDescriptor();\n"
+    "void CallMethod(const ::google::protobuf::MethodDescriptor* method,\n"
+    "                ::google::protobuf::RpcController* controller,\n"
+    "                const ::google::protobuf::Message* request,\n"
+    "                ::google::protobuf::Message* response,\n"
+    "                ::google::protobuf::Closure* done);\n"
+    "const ::google::protobuf::Message& GetRequestPrototype(\n"
+    "  const ::google::protobuf::MethodDescriptor* method) const;\n"
+    "const ::google::protobuf::Message& GetResponsePrototype(\n"
+    "  const ::google::protobuf::MethodDescriptor* method) const;\n");
+
+  printer->Outdent();
+  printer->Print(vars_,
+    "\n"
+    " private:\n"
+    "  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS($classname$);\n"
+    "};\n"
+    "\n");
+}
+
+void ServiceGenerator::GenerateStubDefinition(io::Printer* printer) {
+  printer->Print(vars_,
+    "class $dllexport$$classname$_Stub : public $classname$ {\n"
+    " public:\n");
+
+  printer->Indent();
+
+  printer->Print(vars_,
+    "$classname$_Stub(::google::protobuf::RpcChannel* channel);\n"
+    "$classname$_Stub(::google::protobuf::RpcChannel* channel,\n"
+    "                 ::google::protobuf::Service::ChannelOwnership ownership);\n"
+    "~$classname$_Stub();\n"
+    "\n"
+    "inline ::google::protobuf::RpcChannel* channel() { return channel_; }\n"
+    "\n"
+    "// implements $classname$ ------------------------------------------\n"
+    "\n");
+
+  GenerateMethodSignatures(NON_VIRTUAL, printer);
+
+  printer->Outdent();
+  printer->Print(vars_,
+    " private:\n"
+    "  ::google::protobuf::RpcChannel* channel_;\n"
+    "  bool owns_channel_;\n"
+    "  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS($classname$_Stub);\n"
+    "};\n"
+    "\n");
+}
+
+void ServiceGenerator::GenerateMethodSignatures(
+    VirtualOrNon virtual_or_non, io::Printer* printer) {
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> sub_vars;
+    sub_vars["name"] = method->name();
+    sub_vars["input_type"] = ClassName(method->input_type(), true);
+    sub_vars["output_type"] = ClassName(method->output_type(), true);
+    sub_vars["virtual"] = virtual_or_non == VIRTUAL ? "virtual " : "";
+
+    printer->Print(sub_vars,
+      "$virtual$void $name$(::google::protobuf::RpcController* controller,\n"
+      "                     const $input_type$* request,\n"
+      "                     $output_type$* response,\n"
+      "                     ::google::protobuf::Closure* done);\n");
+  }
+}
+
+// ===================================================================
+
+void ServiceGenerator::GenerateDescriptorInitializer(
+    io::Printer* printer, int index) {
+  map<string, string> vars;
+  vars["classname"] = descriptor_->name();
+  vars["index"] = SimpleItoa(index);
+
+  printer->Print(vars,
+    "$classname$_descriptor_ = file->service($index$);\n");
+}
+
+// ===================================================================
+
+void ServiceGenerator::GenerateImplementation(io::Printer* printer) {
+  printer->Print(vars_,
+    "$classname$::~$classname$() {}\n"
+    "\n"
+    "const ::google::protobuf::ServiceDescriptor* $classname$::descriptor() {\n"
+    "  return $classname$_descriptor_;\n"
+    "}\n"
+    "\n"
+    "const ::google::protobuf::ServiceDescriptor* $classname$::GetDescriptor() {\n"
+    "  return $classname$_descriptor_;\n"
+    "}\n"
+    "\n");
+
+  // Generate methods of the interface.
+  GenerateNotImplementedMethods(printer);
+  GenerateCallMethod(printer);
+  GenerateGetPrototype(REQUEST, printer);
+  GenerateGetPrototype(RESPONSE, printer);
+
+  // Generate stub implementation.
+  printer->Print(vars_,
+    "$classname$_Stub::$classname$_Stub(::google::protobuf::RpcChannel* channel)\n"
+    "  : channel_(channel), owns_channel_(false) {}\n"
+    "$classname$_Stub::$classname$_Stub(\n"
+    "    ::google::protobuf::RpcChannel* channel,\n"
+    "    ::google::protobuf::Service::ChannelOwnership ownership)\n"
+    "  : channel_(channel),\n"
+    "    owns_channel_(ownership == ::google::protobuf::Service::STUB_OWNS_CHANNEL) {}\n"
+    "$classname$_Stub::~$classname$_Stub() {\n"
+    "  if (owns_channel_) delete channel_;\n"
+    "}\n"
+    "\n");
+
+  GenerateStubMethods(printer);
+}
+
+void ServiceGenerator::GenerateNotImplementedMethods(io::Printer* printer) {
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> sub_vars;
+    sub_vars["classname"] = descriptor_->name();
+    sub_vars["name"] = method->name();
+    sub_vars["index"] = SimpleItoa(i);
+    sub_vars["input_type"] = ClassName(method->input_type(), true);
+    sub_vars["output_type"] = ClassName(method->output_type(), true);
+
+    printer->Print(sub_vars,
+      "void $classname$::$name$(::google::protobuf::RpcController* controller,\n"
+      "                         const $input_type$* request,\n"
+      "                         $output_type$* response,\n"
+      "                         ::google::protobuf::Closure* done) {\n"
+      "  controller->SetFailed(\"Method $name$() not implemented.\");\n"
+      "  done->Run();\n"
+      "}\n"
+      "\n");
+  }
+}
+
+void ServiceGenerator::GenerateCallMethod(io::Printer* printer) {
+  printer->Print(vars_,
+    "void $classname$::CallMethod(const ::google::protobuf::MethodDescriptor* method,\n"
+    "                             ::google::protobuf::RpcController* controller,\n"
+    "                             const ::google::protobuf::Message* request,\n"
+    "                             ::google::protobuf::Message* response,\n"
+    "                             ::google::protobuf::Closure* done) {\n"
+    "  GOOGLE_DCHECK_EQ(method->service(), $classname$_descriptor_);\n"
+    "  switch(method->index()) {\n");
+
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> sub_vars;
+    sub_vars["name"] = method->name();
+    sub_vars["index"] = SimpleItoa(i);
+    sub_vars["input_type"] = ClassName(method->input_type(), true);
+    sub_vars["output_type"] = ClassName(method->output_type(), true);
+
+    // Note:  ::google::protobuf::down_cast does not work here because it only works on pointers,
+    //   not references.
+    printer->Print(sub_vars,
+      "    case $index$:\n"
+      "      $name$(controller,\n"
+      "             ::google::protobuf::down_cast<const $input_type$*>(request),\n"
+      "             ::google::protobuf::down_cast< $output_type$*>(response),\n"
+      "             done);\n"
+      "      break;\n");
+  }
+
+  printer->Print(vars_,
+    "    default:\n"
+    "      GOOGLE_LOG(FATAL) << \"Bad method index; this should never happen.\";\n"
+    "      break;\n"
+    "  }\n"
+    "}\n"
+    "\n");
+}
+
+void ServiceGenerator::GenerateGetPrototype(RequestOrResponse which,
+                                            io::Printer* printer) {
+  if (which == REQUEST) {
+    printer->Print(vars_,
+      "const ::google::protobuf::Message& $classname$::GetRequestPrototype(\n");
+  } else {
+    printer->Print(vars_,
+      "const ::google::protobuf::Message& $classname$::GetResponsePrototype(\n");
+  }
+
+  printer->Print(vars_,
+    "    const ::google::protobuf::MethodDescriptor* method) const {\n"
+    "  GOOGLE_DCHECK_EQ(method->service(), $classname$_descriptor_);\n"
+    "  switch(method->index()) {\n");
+
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    const Descriptor* type =
+      (which == REQUEST) ? method->input_type() : method->output_type();
+
+    map<string, string> sub_vars;
+    sub_vars["index"] = SimpleItoa(i);
+    sub_vars["type"] = ClassName(type, true);
+
+    printer->Print(sub_vars,
+      "    case $index$:\n"
+      "      return $type$::default_instance();\n");
+  }
+
+  printer->Print(vars_,
+    "    default:\n"
+    "      GOOGLE_LOG(FATAL) << \"Bad method index; this should never happen.\";\n"
+    "      return *reinterpret_cast< ::google::protobuf::Message*>(NULL);\n"
+    "  }\n"
+    "}\n"
+    "\n");
+}
+
+void ServiceGenerator::GenerateStubMethods(io::Printer* printer) {
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> sub_vars;
+    sub_vars["classname"] = descriptor_->name();
+    sub_vars["name"] = method->name();
+    sub_vars["index"] = SimpleItoa(i);
+    sub_vars["input_type"] = ClassName(method->input_type(), true);
+    sub_vars["output_type"] = ClassName(method->output_type(), true);
+
+    printer->Print(sub_vars,
+      "void $classname$_Stub::$name$(::google::protobuf::RpcController* controller,\n"
+      "                              const $input_type$* request,\n"
+      "                              $output_type$* response,\n"
+      "                              ::google::protobuf::Closure* done) {\n"
+      "  channel_->CallMethod($classname$_descriptor_->method($index$),\n"
+      "                       controller, request, response, done);\n"
+      "}\n");
+  }
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_service.h b/src/google/protobuf/compiler/cpp/cpp_service.h
new file mode 100644
index 00000000..bccb64e4
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_service.h
@@ -0,0 +1,104 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_SERVICE_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_SERVICE_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class ServiceGenerator {
+ public:
+  // See generator.cc for the meaning of dllexport_decl.
+  explicit ServiceGenerator(const ServiceDescriptor* descriptor,
+                            const string& dllexport_decl);
+  ~ServiceGenerator();
+
+  // Header stuff.
+
+  // Generate the class definitions for the service's interface and the
+  // stub implementation.
+  void GenerateDeclarations(io::Printer* printer);
+
+  // Source file stuff.
+
+  // Generate code that initializes the global variable storing the service's
+  // descriptor.
+  void GenerateDescriptorInitializer(io::Printer* printer, int index);
+
+  // Generate implementations of everything declared by GenerateDeclarations().
+  void GenerateImplementation(io::Printer* printer);
+
+ private:
+  enum RequestOrResponse { REQUEST, RESPONSE };
+  enum VirtualOrNon { VIRTUAL, NON_VIRTUAL };
+
+  // Header stuff.
+
+  // Generate the service abstract interface.
+  void GenerateInterface(io::Printer* printer);
+
+  // Generate the stub class definition.
+  void GenerateStubDefinition(io::Printer* printer);
+
+  // Prints signatures for all methods in the
+  void GenerateMethodSignatures(VirtualOrNon virtual_or_non,
+                                io::Printer* printer);
+
+  // Source file stuff.
+
+  // Generate the default implementations of the service methods, which
+  // produce a "not implemented" error.
+  void GenerateNotImplementedMethods(io::Printer* printer);
+
+  // Generate the CallMethod() method of the service.
+  void GenerateCallMethod(io::Printer* printer);
+
+  // Generate the Get{Request,Response}Prototype() methods.
+  void GenerateGetPrototype(RequestOrResponse which, io::Printer* printer);
+
+  // Generate the stub's implementations of the service methods.
+  void GenerateStubMethods(io::Printer* printer);
+
+  const ServiceDescriptor* descriptor_;
+  map<string, string> vars_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(ServiceGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_SERVICE_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_string_field.cc b/src/google/protobuf/compiler/cpp/cpp_string_field.cc
new file mode 100644
index 00000000..de59ac87
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_string_field.cc
@@ -0,0 +1,336 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/cpp/cpp_string_field.h>
+#include <google/protobuf/compiler/cpp/cpp_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/wire_format_inl.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+using internal::WireFormat;
+
+namespace {
+
+// TODO(kenton):  Factor out a "SetCommonFieldVariables()" to get rid of
+//   repeat code between this and the other field types.
+void SetStringVariables(const FieldDescriptor* descriptor,
+                        map<string, string>* variables) {
+  (*variables)["name"] = FieldName(descriptor);
+  (*variables)["default"] =
+    "\"" + CEscape(descriptor->default_value_string()) + "\"";
+  (*variables)["index"] = SimpleItoa(descriptor->index());
+  (*variables)["number"] = SimpleItoa(descriptor->number());
+  (*variables)["classname"] = ClassName(FieldScope(descriptor), false);
+  (*variables)["declared_type"] = DeclaredTypeMethodName(descriptor->type());
+  (*variables)["tag_size"] = SimpleItoa(
+    WireFormat::TagSize(descriptor->number(), descriptor->type()));
+}
+
+}  // namespace
+
+// ===================================================================
+
+StringFieldGenerator::
+StringFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetStringVariables(descriptor, &variables_);
+}
+
+StringFieldGenerator::~StringFieldGenerator() {}
+
+void StringFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "::std::string* $name$_;\n"
+    "static const ::std::string _default_$name$_;\n");
+}
+
+void StringFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  // If we're using StringFieldGenerator for a field with a ctype, it's
+  // because that ctype isn't actually implemented.  In particular, this is
+  // true of ctype=CORD and ctype=STRING_PIECE in the open source release.
+  // We aren't releasing Cord because it has too many Google-specific
+  // dependencies and we aren't releasing StringPiece because it's hardly
+  // useful outside of Google and because it would get confusing to have
+  // multiple instances of the StringPiece class in different libraries (PCRE
+  // already includes it for their C++ bindings, which came from Google).
+  //
+  // In any case, we make all the accessors private while still actually
+  // using a string to represent the field internally.  This way, we can
+  // guarantee that if we do ever implement the ctype, it won't break any
+  // existing users who might be -- for whatever reason -- already using .proto
+  // files that applied the ctype.  The field can still be accessed via the
+  // reflection interface since the reflection interface is independent of
+  // the string's underlying representation.
+  if (descriptor_->options().has_ctype()) {
+    printer->Outdent();
+    printer->Print(
+      " private:\n"
+      "  // Hidden due to unknown ctype option.\n");
+    printer->Indent();
+  }
+
+  printer->Print(variables_,
+    "inline const ::std::string& $name$() const;\n"
+    "inline void set_$name$(const ::std::string& value);\n"
+    "inline void set_$name$(const char* value);\n");
+
+  printer->Print(variables_,
+    "inline ::std::string* mutable_$name$();\n");
+
+  if (descriptor_->options().has_ctype()) {
+    printer->Outdent();
+    printer->Print(" public:\n");
+    printer->Indent();
+  }
+}
+
+void StringFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::std::string& $classname$::$name$() const {\n"
+    "  return *$name$_;\n"
+    "}\n"
+    "inline void $classname$::set_$name$(const ::std::string& value) {\n"
+    "  _set_bit($index$);\n"
+    "  if ($name$_ == &_default_$name$_) {\n"
+    "    $name$_ = new ::std::string;\n"
+    "  }\n"
+    "  $name$_->assign(value);\n"
+    "}\n"
+    "inline void $classname$::set_$name$(const char* value) {\n"
+    "  _set_bit($index$);\n"
+    "  if ($name$_ == &_default_$name$_) {\n"
+    "    $name$_ = new ::std::string;\n"
+    "  }\n"
+    "  $name$_->assign(value);\n"
+    "}\n");
+  printer->Print(variables_,
+    "inline ::std::string* $classname$::mutable_$name$() {\n"
+    "  _set_bit($index$);\n"
+    "  if ($name$_ == &_default_$name$_) {\n");
+  if (descriptor_->has_default_value()) {
+    printer->Print(variables_,
+      "    $name$_ = new ::std::string(_default_$name$_);\n");
+  } else {
+    printer->Print(variables_,
+      "    $name$_ = new ::std::string;\n");
+  }
+  printer->Print(variables_,
+    "  }\n"
+    "  return $name$_;\n"
+    "}\n");
+}
+
+void StringFieldGenerator::
+GenerateNonInlineAccessorDefinitions(io::Printer* printer) const {
+  if (descriptor_->has_default_value()) {
+    printer->Print(variables_,
+      "const ::std::string $classname$::_default_$name$_($default$);");
+  } else {
+    printer->Print(variables_,
+      "const ::std::string $classname$::_default_$name$_;");
+  }
+}
+
+void StringFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  if (descriptor_->has_default_value()) {
+    printer->Print(variables_,
+      "if ($name$_ != &_default_$name$_) {\n"
+      "  $name$_->assign(_default_$name$_);\n"
+      "}\n");
+  } else {
+    printer->Print(variables_,
+      "if ($name$_ != &_default_$name$_) {\n"
+      "  $name$_->clear();\n"
+      "}\n");
+  }
+}
+
+void StringFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_, "set_$name$(from.$name$());\n");
+}
+
+void StringFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  printer->Print(variables_,
+    ",\n$name$_(const_cast< ::std::string*>(&_default_$name$_))");
+}
+
+void StringFieldGenerator::
+GenerateDestructorCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if ($name$_ != &_default_$name$_) {\n"
+    "  delete $name$_;\n"
+    "}\n");
+}
+
+void StringFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Read$declared_type$("
+      "input, mutable_$name$()));\n");
+}
+
+void StringFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Write$declared_type$("
+      "$number$, this->$name$(), output));\n");
+}
+
+void StringFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  printer->Print(variables_,
+    "total_size += $tag_size$ +\n"
+    "  ::google::protobuf::internal::WireFormat::$declared_type$Size(this->$name$());\n");
+}
+
+// ===================================================================
+
+RepeatedStringFieldGenerator::
+RepeatedStringFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetStringVariables(descriptor, &variables_);
+}
+
+RepeatedStringFieldGenerator::~RepeatedStringFieldGenerator() {}
+
+void RepeatedStringFieldGenerator::
+GeneratePrivateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "::google::protobuf::RepeatedPtrField< ::std::string> $name$_;\n");
+}
+
+void RepeatedStringFieldGenerator::
+GenerateAccessorDeclarations(io::Printer* printer) const {
+  // See comment above about unknown ctypes.
+  if (descriptor_->options().has_ctype()) {
+    printer->Outdent();
+    printer->Print(
+      " private:\n"
+      "  // Hidden due to unknown ctype option.\n");
+    printer->Indent();
+  }
+
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedPtrField< ::std::string>& $name$() const;\n"
+    "inline ::google::protobuf::RepeatedPtrField< ::std::string>* mutable_$name$();\n"
+    "inline const ::std::string& $name$(int index) const;\n"
+    "inline ::std::string* mutable_$name$(int index);\n"
+    "inline void set_$name$(int index, const ::std::string& value);\n"
+    "inline void set_$name$(int index, const char* value);\n"
+    "inline ::std::string* add_$name$();\n"
+    "inline void add_$name$(const ::std::string& value);\n"
+    "inline void add_$name$(const char* value);\n");
+
+  if (descriptor_->options().has_ctype()) {
+    printer->Outdent();
+    printer->Print(" public:\n");
+    printer->Indent();
+  }
+}
+
+void RepeatedStringFieldGenerator::
+GenerateInlineAccessorDefinitions(io::Printer* printer) const {
+  printer->Print(variables_,
+    "inline const ::google::protobuf::RepeatedPtrField< ::std::string>&\n"
+    "$classname$::$name$() const {\n"
+    "  return $name$_;\n"
+    "}\n"
+    "inline ::google::protobuf::RepeatedPtrField< ::std::string>*\n"
+    "$classname$::mutable_$name$() {\n"
+    "  return &$name$_;\n"
+    "}\n"
+    "inline const ::std::string& $classname$::$name$(int index) const {\n"
+    "  return $name$_.Get(index);\n"
+    "}\n"
+    "inline ::std::string* $classname$::mutable_$name$(int index) {\n"
+    "  return $name$_.Mutable(index);\n"
+    "}\n"
+    "inline void $classname$::set_$name$(int index, const ::std::string& value) {\n"
+    "  $name$_.Mutable(index)->assign(value);\n"
+    "}\n"
+    "inline void $classname$::set_$name$(int index, const char* value) {\n"
+    "  $name$_.Mutable(index)->assign(value);\n"
+    "}\n"
+    "inline ::std::string* $classname$::add_$name$() {\n"
+    "  return $name$_.Add();\n"
+    "}\n"
+    "inline void $classname$::add_$name$(const ::std::string& value) {\n"
+    "  $name$_.Add()->assign(value);\n"
+    "}\n"
+    "inline void $classname$::add_$name$(const char* value) {\n"
+    "  $name$_.Add()->assign(value);\n"
+    "}\n");
+}
+
+void RepeatedStringFieldGenerator::
+GenerateClearingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.Clear();\n");
+}
+
+void RepeatedStringFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_, "$name$_.MergeFrom(from.$name$_);\n");
+}
+
+void RepeatedStringFieldGenerator::
+GenerateInitializer(io::Printer* printer) const {
+  // Not needed for repeated fields.
+}
+
+void RepeatedStringFieldGenerator::
+GenerateMergeFromCodedStream(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Read$declared_type$(\n"
+    "     input, add_$name$()));\n");
+}
+
+void RepeatedStringFieldGenerator::
+GenerateSerializeWithCachedSizes(io::Printer* printer) const {
+  printer->Print(variables_,
+    "DO_(::google::protobuf::internal::WireFormat::Write$declared_type$("
+      "$number$, this->$name$(i), output));\n");
+}
+
+void RepeatedStringFieldGenerator::
+GenerateByteSize(io::Printer* printer) const {
+  printer->Print(variables_,
+    "total_size += $tag_size$ * $name$_size();\n"
+    "for (int i = 0; i < $name$_size(); i++) {\n"
+    "  total_size += ::google::protobuf::internal::WireFormat::$declared_type$Size(\n"
+    "    this->$name$(i));\n"
+    "}\n");
+}
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/cpp/cpp_string_field.h b/src/google/protobuf/compiler/cpp/cpp_string_field.h
new file mode 100644
index 00000000..44ffd18c
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_string_field.h
@@ -0,0 +1,86 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_CPP_STRING_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_CPP_STRING_FIELD_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/compiler/cpp/cpp_field.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+class StringFieldGenerator : public FieldGenerator {
+ public:
+  explicit StringFieldGenerator(const FieldDescriptor* descriptor);
+  ~StringFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateNonInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateDestructorCode(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(StringFieldGenerator);
+};
+
+class RepeatedStringFieldGenerator : public FieldGenerator {
+ public:
+  explicit RepeatedStringFieldGenerator(const FieldDescriptor* descriptor);
+  ~RepeatedStringFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GeneratePrivateMembers(io::Printer* printer) const;
+  void GenerateAccessorDeclarations(io::Printer* printer) const;
+  void GenerateInlineAccessorDefinitions(io::Printer* printer) const;
+  void GenerateClearingCode(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateInitializer(io::Printer* printer) const;
+  void GenerateMergeFromCodedStream(io::Printer* printer) const;
+  void GenerateSerializeWithCachedSizes(io::Printer* printer) const;
+  void GenerateByteSize(io::Printer* printer) const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(RepeatedStringFieldGenerator);
+};
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_CPP_STRING_FIELD_H__
diff --git a/src/google/protobuf/compiler/cpp/cpp_test_bad_identifiers.proto b/src/google/protobuf/compiler/cpp/cpp_test_bad_identifiers.proto
new file mode 100644
index 00000000..ee0499bf
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_test_bad_identifiers.proto
@@ -0,0 +1,87 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// This file tests that various identifiers work as field and type names even
+// though the same identifiers are used internally by the C++ code generator.
+
+
+// We don't put this in a package within proto2 because we need to make sure
+// that the generated code doesn't depend on being in the proto2 namespace.
+package protobuf_unittest;
+
+// Test that fields can have names like "input" and "i" which are also used
+// internally by the code generator for local variables.
+message TestConflictingSymbolNames {
+  message BuildDescriptors {}
+  message TypeTraits {}
+
+  optional int32 input = 1;
+  optional int32 output = 2;
+  optional string length = 3;
+  repeated int32 i = 4;
+  repeated string new_element = 5 [ctype=STRING_PIECE];
+  optional int32 total_size = 6;
+  optional int32 tag = 7;
+
+  optional int32 source = 8;
+  optional int32 value = 9;
+  optional int32 file = 10;
+  optional int32 from = 11;
+  optional int32 handle_uninterpreted = 12;
+  repeated int32 index = 13;
+  optional int32 controller = 14;
+  optional int32 already_here = 15;
+
+  optional uint32 uint32 = 16;
+  optional uint64 uint64 = 17;
+  optional string string = 18;
+  optional int32 memset = 19;
+  optional int32 int32 = 20;
+  optional int64 int64 = 21;
+
+  optional uint32 cached_size = 22;
+  optional uint32 extensions = 23;
+  optional uint32 bit = 24;
+  optional uint32 bits = 25;
+  optional uint32 offsets = 26;
+  optional uint32 reflection = 27;
+
+  message Cord {}
+  optional string some_cord = 28 [ctype=CORD];
+
+  message StringPiece {}
+  optional string some_string_piece = 29 [ctype=STRING_PIECE];
+
+  // Some keywords.
+  optional uint32 int = 30;
+  optional uint32 friend = 31;
+
+  // The generator used to #define a macro called "DO" inside the .cc file.
+  message DO {}
+  optional DO do = 32;
+
+  extensions 1000 to max;
+}
+
+message DummyMessage {}
+
+service TestConflictingMethodNames {
+  rpc Closure(DummyMessage) returns (DummyMessage);
+}
diff --git a/src/google/protobuf/compiler/cpp/cpp_unittest.cc b/src/google/protobuf/compiler/cpp/cpp_unittest.cc
new file mode 100644
index 00000000..8253242b
--- /dev/null
+++ b/src/google/protobuf/compiler/cpp/cpp_unittest.cc
@@ -0,0 +1,835 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// To test the code generator, we actually use it to generate code for
+// google/protobuf/unittest.proto, then test that.  This means that we
+// are actually testing the parser and other parts of the system at the same
+// time, and that problems in the generator may show up as compile-time errors
+// rather than unittest failures, which may be surprising.  However, testing
+// the output of the C++ generator directly would be very hard.  We can't very
+// well just check it against golden files since those files would have to be
+// updated for any small change; such a test would be very brittle and probably
+// not very helpful.  What we really want to test is that the code compiles
+// correctly and produces the interfaces we expect, which is why this test
+// is written this way.
+
+#include <vector>
+
+#include <google/protobuf/unittest.pb.h>
+#include <google/protobuf/unittest_optimize_for.pb.h>
+#include <google/protobuf/unittest_embed_optimize_for.pb.h>
+#include <google/protobuf/test_util.h>
+#include <google/protobuf/compiler/cpp/cpp_test_bad_identifiers.pb.h>
+#include <google/protobuf/compiler/importer.h>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/dynamic_message.h>
+
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/stubs/substitute.h>
+#include <google/protobuf/testing/googletest.h>
+#include <gtest/gtest.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace cpp {
+
+namespace {
+
+
+class MockErrorCollector : public MultiFileErrorCollector {
+ public:
+  MockErrorCollector() {}
+  ~MockErrorCollector() {}
+
+  string text_;
+
+  // implements ErrorCollector ---------------------------------------
+  void AddError(const string& filename, int line, int column,
+                const string& message) {
+    strings::SubstituteAndAppend(&text_, "$0:$1:$2: $3\n",
+                                 filename, line, column, message);
+  }
+};
+
+// Test that generated code has proper descriptors:
+// Parse a descriptor directly (using google::protobuf::compiler::Importer) and
+// compare it to the one that was produced by generated code.
+TEST(GeneratedDescriptorTest, IdenticalDescriptors) {
+  const FileDescriptor* generated_descriptor =
+    unittest::TestAllTypes::descriptor()->file();
+
+  // Set up the Importer.
+  MockErrorCollector error_collector;
+  DiskSourceTree source_tree;
+  source_tree.MapPath("", TestSourceDir());
+  Importer importer(&source_tree, &error_collector);
+
+  // Import (parse) unittest.proto.
+  const FileDescriptor* parsed_descriptor =
+    importer.Import("google/protobuf/unittest.proto");
+  EXPECT_EQ("", error_collector.text_);
+  ASSERT_TRUE(parsed_descriptor != NULL);
+
+  // Test that descriptors are generated correctly by converting them to
+  // FileDescriptorProtos and comparing.
+  FileDescriptorProto generated_decsriptor_proto, parsed_descriptor_proto;
+  generated_descriptor->CopyTo(&generated_decsriptor_proto);
+  parsed_descriptor->CopyTo(&parsed_descriptor_proto);
+
+  EXPECT_EQ(parsed_descriptor_proto.DebugString(),
+            generated_decsriptor_proto.DebugString());
+}
+
+// ===================================================================
+
+TEST(GeneratedMessageTest, Defaults) {
+  // Check that all default values are set correctly in the initial message.
+  unittest::TestAllTypes message;
+
+  TestUtil::ExpectClear(message);
+
+  // Messages should return pointers to default instances until first use.
+  // (This is not checked by ExpectClear() since it is not actually true after
+  // the fields have been set and then cleared.)
+  EXPECT_EQ(&unittest::TestAllTypes::OptionalGroup::default_instance(),
+            &message.optionalgroup());
+  EXPECT_EQ(&unittest::TestAllTypes::NestedMessage::default_instance(),
+            &message.optional_nested_message());
+  EXPECT_EQ(&unittest::ForeignMessage::default_instance(),
+            &message.optional_foreign_message());
+  EXPECT_EQ(&unittest_import::ImportMessage::default_instance(),
+            &message.optional_import_message());
+}
+
+TEST(GeneratedMessageTest, Accessors) {
+  // Set every field to a unique value then go back and check all those
+  // values.
+  unittest::TestAllTypes message;
+
+  TestUtil::SetAllFields(&message);
+  TestUtil::ExpectAllFieldsSet(message);
+
+  TestUtil::ModifyRepeatedFields(&message);
+  TestUtil::ExpectRepeatedFieldsModified(message);
+}
+
+TEST(GeneratedMessageTest, MutableStringDefault) {
+  // mutable_foo() for a string should return a string initialized to its
+  // default value.
+  unittest::TestAllTypes message;
+
+  EXPECT_EQ("hello", *message.mutable_default_string());
+
+  // Note that the first time we call mutable_foo(), we get a newly-allocated
+  // string, but if we clear it and call it again, we get the same object again.
+  // We should verify that it has its default value in both cases.
+  message.set_default_string("blah");
+  message.Clear();
+
+  EXPECT_EQ("hello", *message.mutable_default_string());
+}
+
+TEST(GeneratedMessageTest, Clear) {
+  // Set every field to a unique value, clear the message, then check that
+  // it is cleared.
+  unittest::TestAllTypes message;
+
+  TestUtil::SetAllFields(&message);
+  message.Clear();
+  TestUtil::ExpectClear(message);
+
+  // Unlike with the defaults test, we do NOT expect that requesting embedded
+  // messages will return a pointer to the default instance.  Instead, they
+  // should return the objects that were created when mutable_blah() was
+  // called.
+  EXPECT_NE(&unittest::TestAllTypes::OptionalGroup::default_instance(),
+            &message.optionalgroup());
+  EXPECT_NE(&unittest::TestAllTypes::NestedMessage::default_instance(),
+            &message.optional_nested_message());
+  EXPECT_NE(&unittest::ForeignMessage::default_instance(),
+            &message.optional_foreign_message());
+  EXPECT_NE(&unittest_import::ImportMessage::default_instance(),
+            &message.optional_import_message());
+}
+
+TEST(GeneratedMessageTest, ClearOneField) {
+  // Set every field to a unique value, then clear one value and insure that
+  // only that one value is cleared.
+  unittest::TestAllTypes message;
+
+  TestUtil::SetAllFields(&message);
+  int64 original_value = message.optional_int64();
+
+  // Clear the field and make sure it shows up as cleared.
+  message.clear_optional_int64();
+  EXPECT_FALSE(message.has_optional_int64());
+  EXPECT_EQ(0, message.optional_int64());
+
+  // Other adjacent fields should not be cleared.
+  EXPECT_TRUE(message.has_optional_int32());
+  EXPECT_TRUE(message.has_optional_uint32());
+
+  // Make sure if we set it again, then all fields are set.
+  message.set_optional_int64(original_value);
+  TestUtil::ExpectAllFieldsSet(message);
+}
+
+
+TEST(GeneratedMessageTest, CopyFrom) {
+  unittest::TestAllTypes message1, message2;
+  string data;
+
+  TestUtil::SetAllFields(&message1);
+  message2.CopyFrom(message1);
+  TestUtil::ExpectAllFieldsSet(message2);
+
+  // Copying from self should be a no-op.
+  message2.CopyFrom(message2);
+  TestUtil::ExpectAllFieldsSet(message2);
+}
+
+TEST(GeneratedMessageTest, CopyConstructor) {
+  unittest::TestAllTypes message1;
+  TestUtil::SetAllFields(&message1);
+
+  unittest::TestAllTypes message2(message1);
+  TestUtil::ExpectAllFieldsSet(message2);
+}
+
+TEST(GeneratedMessageTest, CopyAssignmentOperator) {
+  unittest::TestAllTypes message1;
+  TestUtil::SetAllFields(&message1);
+
+  unittest::TestAllTypes message2;
+  message2 = message1;
+  TestUtil::ExpectAllFieldsSet(message2);
+
+  // Make sure that self-assignment does something sane.
+  message2 = message2;
+  TestUtil::ExpectAllFieldsSet(message2);
+}
+
+TEST(GeneratedMessageTest, UpcastCopyFrom) {
+  // Test the CopyFrom method that takes in the generic const Message&
+  // parameter.
+  unittest::TestAllTypes message1, message2;
+
+  TestUtil::SetAllFields(&message1);
+
+  const Message* source = implicit_cast<const Message*>(&message1);
+  message2.CopyFrom(*source);
+
+  TestUtil::ExpectAllFieldsSet(message2);
+}
+
+TEST(GeneratedMessageTest, DynamicMessageCopyFrom) {
+  // Test copying from a DynamicMessage, which must fall back to using
+  // reflection.
+  unittest::TestAllTypes message2;
+
+  // Construct a new version of the dynamic message via the factory.
+  DynamicMessageFactory factory;
+  scoped_ptr<Message> message1;
+  message1.reset(factory.GetPrototype(
+                     unittest::TestAllTypes::descriptor())->New());
+
+  TestUtil::ReflectionTester reflection_tester(
+    unittest::TestAllTypes::descriptor());
+  reflection_tester.SetAllFieldsViaReflection(message1->GetReflection());
+
+  message2.CopyFrom(*message1);
+
+  TestUtil::ExpectAllFieldsSet(message2);
+}
+
+TEST(GeneratedMessageTest, NonEmptyMergeFrom) {
+  // Test merging with a non-empty message. Code is a modified form
+  // of that found in google/protobuf/reflection_ops_unittest.cc.
+  unittest::TestAllTypes message1, message2;
+
+  TestUtil::SetAllFields(&message1);
+
+  // This field will test merging into an empty spot.
+  message2.set_optional_int32(message1.optional_int32());
+  message1.clear_optional_int32();
+
+  // This tests overwriting.
+  message2.set_optional_string(message1.optional_string());
+  message1.set_optional_string("something else");
+
+  // This tests concatenating.
+  message2.add_repeated_int32(message1.repeated_int32(1));
+  int32 i = message1.repeated_int32(0);
+  message1.clear_repeated_int32();
+  message1.add_repeated_int32(i);
+
+  message1.MergeFrom(message2);
+
+  TestUtil::ExpectAllFieldsSet(message1);
+}
+
+#ifdef GTEST_HAS_DEATH_TEST
+
+TEST(GeneratedMessageTest, MergeFromSelf) {
+  unittest::TestAllTypes message;
+  EXPECT_DEATH(message.MergeFrom(message), "&from");
+  EXPECT_DEATH(message.MergeFrom(implicit_cast<const Message&>(message)),
+               "&from");
+}
+
+#endif  // GTEST_HAS_DEATH_TEST
+
+TEST(GeneratedMessageTest, Serialization) {
+  unittest::TestAllTypes message1, message2;
+  string data;
+
+  TestUtil::SetAllFields(&message1);
+  message1.SerializeToString(&data);
+  EXPECT_TRUE(message2.ParseFromString(data));
+  TestUtil::ExpectAllFieldsSet(message2);
+
+}
+
+
+TEST(GeneratedMessageTest, Required) {
+  // Test that IsInitialized() returns false if required fields are missing.
+  unittest::TestRequired message;
+
+  EXPECT_FALSE(message.IsInitialized());
+  message.set_a(1);
+  EXPECT_FALSE(message.IsInitialized());
+  message.set_b(2);
+  EXPECT_FALSE(message.IsInitialized());
+  message.set_c(3);
+  EXPECT_TRUE(message.IsInitialized());
+}
+
+TEST(GeneratedMessageTest, RequiredForeign) {
+  // Test that IsInitialized() returns false if required fields in nested
+  // messages are missing.
+  unittest::TestRequiredForeign message;
+
+  EXPECT_TRUE(message.IsInitialized());
+
+  message.mutable_optional_message();
+  EXPECT_FALSE(message.IsInitialized());
+
+  message.mutable_optional_message()->set_a(1);
+  message.mutable_optional_message()->set_b(2);
+  message.mutable_optional_message()->set_c(3);
+  EXPECT_TRUE(message.IsInitialized());
+
+  message.add_repeated_message();
+  EXPECT_FALSE(message.IsInitialized());
+
+  message.mutable_repeated_message(0)->set_a(1);
+  message.mutable_repeated_message(0)->set_b(2);
+  message.mutable_repeated_message(0)->set_c(3);
+  EXPECT_TRUE(message.IsInitialized());
+}
+
+TEST(GeneratedMessageTest, ForeignNested) {
+  // Test that TestAllTypes::NestedMessage can be embedded directly into
+  // another message.
+  unittest::TestForeignNested message;
+
+  // If this compiles and runs without crashing, it must work.  We have
+  // nothing more to test.
+  unittest::TestAllTypes::NestedMessage* nested =
+    message.mutable_foreign_nested();
+  nested->set_bb(1);
+}
+
+TEST(GeneratedMessageTest, ReallyLargeTagNumber) {
+  // Test that really large tag numbers don't break anything.
+  unittest::TestReallyLargeTagNumber message1, message2;
+  string data;
+
+  // For the most part, if this compiles and runs then we're probably good.
+  // (The most likely cause for failure would be if something were attempting
+  // to allocate a lookup table of some sort using tag numbers as the index.)
+  // We'll try serializing just for fun.
+  message1.set_a(1234);
+  message1.set_bb(5678);
+  message1.SerializeToString(&data);
+  EXPECT_TRUE(message2.ParseFromString(data));
+  EXPECT_EQ(1234, message2.a());
+  EXPECT_EQ(5678, message2.bb());
+}
+
+TEST(GeneratedMessageTest, MutualRecursion) {
+  // Test that mutually-recursive message types work.
+  unittest::TestMutualRecursionA message;
+  unittest::TestMutualRecursionA* nested = message.mutable_bb()->mutable_a();
+  unittest::TestMutualRecursionA* nested2 = nested->mutable_bb()->mutable_a();
+
+  // Again, if the above compiles and runs, that's all we really have to
+  // test, but just for run we'll check that the system didn't somehow come
+  // up with a pointer loop...
+  EXPECT_NE(&message, nested);
+  EXPECT_NE(&message, nested2);
+  EXPECT_NE(nested, nested2);
+}
+
+TEST(GeneratedMessageTest, CamelCaseFieldNames) {
+  // This test is mainly checking that the following compiles, which verifies
+  // that the field names were coerced to lower-case.
+  //
+  // Protocol buffers standard style is to use lowercase-with-underscores for
+  // field names.  Some old proto1 .protos unfortunately used camel-case field
+  // names.  In proto1, these names were forced to lower-case.  So, we do the
+  // same thing in proto2.
+
+  unittest::TestCamelCaseFieldNames message;
+
+  message.set_primitivefield(2);
+  message.set_stringfield("foo");
+  message.set_enumfield(unittest::FOREIGN_FOO);
+  message.mutable_messagefield()->set_c(6);
+
+  message.add_repeatedprimitivefield(8);
+  message.add_repeatedstringfield("qux");
+  message.add_repeatedenumfield(unittest::FOREIGN_BAR);
+  message.add_repeatedmessagefield()->set_c(15);
+
+  EXPECT_EQ(2, message.primitivefield());
+  EXPECT_EQ("foo", message.stringfield());
+  EXPECT_EQ(unittest::FOREIGN_FOO, message.enumfield());
+  EXPECT_EQ(6, message.messagefield().c());
+
+  EXPECT_EQ(8, message.repeatedprimitivefield(0));
+  EXPECT_EQ("qux", message.repeatedstringfield(0));
+  EXPECT_EQ(unittest::FOREIGN_BAR, message.repeatedenumfield(0));
+  EXPECT_EQ(15, message.repeatedmessagefield(0).c());
+}
+
+TEST(GeneratedMessageTest, TestConflictingSymbolNames) {
+  // test_bad_identifiers.proto successfully compiled, then it works.  The
+  // following is just a token usage to insure that the code is, in fact,
+  // being compiled and linked.
+
+  protobuf_unittest::TestConflictingSymbolNames message;
+  message.set_uint32(1);
+  EXPECT_EQ(3, message.ByteSize());
+
+  message.set_friend_(5);
+  EXPECT_EQ(5, message.friend_());
+}
+
+TEST(GeneratedMessageTest, TestOptimizedForSize) {
+  // We rely on the tests in reflection_ops_unittest and wire_format_unittest
+  // to really test that reflection-based methods work.  Here we are mostly
+  // just making sure that TestOptimizedForSize actually builds and seems to
+  // function.
+
+  protobuf_unittest::TestOptimizedForSize message, message2;
+  message.set_i(1);
+  message.mutable_msg()->set_c(2);
+  message2.CopyFrom(message);
+  EXPECT_EQ(1, message2.i());
+  EXPECT_EQ(2, message2.msg().c());
+}
+
+TEST(GeneratedMessageTest, TestEmbedOptimizedForSize) {
+  // Verifies that something optimized for speed can contain something optimized
+  // for size.
+
+  protobuf_unittest::TestEmbedOptimizedForSize message, message2;
+  message.mutable_optional_message()->set_i(1);
+  message.add_repeated_message()->mutable_msg()->set_c(2);
+  string data;
+  message.SerializeToString(&data);
+  ASSERT_TRUE(message2.ParseFromString(data));
+  EXPECT_EQ(1, message2.optional_message().i());
+  EXPECT_EQ(2, message2.repeated_message(0).msg().c());
+}
+
+// ===================================================================
+
+TEST(GeneratedEnumTest, EnumValuesAsSwitchCases) {
+  // Test that our nested enum values can be used as switch cases.  This test
+  // doesn't actually do anything, the proof that it works is that it
+  // compiles.
+  int i =0;
+  unittest::TestAllTypes::NestedEnum a = unittest::TestAllTypes::BAR;
+  switch (a) {
+    case unittest::TestAllTypes::FOO:
+      i = 1;
+      break;
+    case unittest::TestAllTypes::BAR:
+      i = 2;
+      break;
+    case unittest::TestAllTypes::BAZ:
+      i = 3;
+      break;
+    // no default case:  We want to make sure the compiler recognizes that
+    //   all cases are covered.  (GCC warns if you do not cover all cases of
+    //   an enum in a switch.)
+  }
+
+  // Token check just for fun.
+  EXPECT_EQ(2, i);
+}
+
+TEST(GeneratedEnumTest, IsValidValue) {
+  // Test enum IsValidValue.
+  EXPECT_TRUE(unittest::TestAllTypes::NestedEnum_IsValid(1));
+  EXPECT_TRUE(unittest::TestAllTypes::NestedEnum_IsValid(2));
+  EXPECT_TRUE(unittest::TestAllTypes::NestedEnum_IsValid(3));
+
+  EXPECT_FALSE(unittest::TestAllTypes::NestedEnum_IsValid(0));
+  EXPECT_FALSE(unittest::TestAllTypes::NestedEnum_IsValid(4));
+
+  // Make sure it also works when there are dups.
+  EXPECT_TRUE(unittest::TestEnumWithDupValue_IsValid(1));
+  EXPECT_TRUE(unittest::TestEnumWithDupValue_IsValid(2));
+  EXPECT_TRUE(unittest::TestEnumWithDupValue_IsValid(3));
+
+  EXPECT_FALSE(unittest::TestEnumWithDupValue_IsValid(0));
+  EXPECT_FALSE(unittest::TestEnumWithDupValue_IsValid(4));
+}
+
+TEST(GeneratedEnumTest, MinAndMax) {
+  EXPECT_EQ(unittest::TestAllTypes::FOO,unittest::TestAllTypes::NestedEnum_MIN);
+  EXPECT_EQ(unittest::TestAllTypes::BAZ,unittest::TestAllTypes::NestedEnum_MAX);
+
+  EXPECT_EQ(unittest::FOREIGN_FOO, unittest::ForeignEnum_MIN);
+  EXPECT_EQ(unittest::FOREIGN_BAZ, unittest::ForeignEnum_MAX);
+
+  EXPECT_EQ(1, unittest::TestEnumWithDupValue_MIN);
+  EXPECT_EQ(3, unittest::TestEnumWithDupValue_MAX);
+
+  EXPECT_EQ(unittest::SPARSE_E, unittest::TestSparseEnum_MIN);
+  EXPECT_EQ(unittest::SPARSE_C, unittest::TestSparseEnum_MAX);
+
+  // Make sure we can use _MIN and _MAX as switch cases.
+  switch(unittest::SPARSE_A) {
+    case unittest::TestSparseEnum_MIN:
+    case unittest::TestSparseEnum_MAX:
+      break;
+    default:
+      break;
+  }
+}
+
+// ===================================================================
+
+// Support code for testing services.
+class GeneratedServiceTest : public testing::Test {
+ protected:
+  class MockTestService : public unittest::TestService {
+   public:
+    MockTestService()
+      : called_(false),
+        method_(""),
+        controller_(NULL),
+        request_(NULL),
+        response_(NULL),
+        done_(NULL) {}
+
+    ~MockTestService() {}
+
+    void Reset() { called_ = false; }
+
+    // implements TestService ----------------------------------------
+
+    void Foo(RpcController* controller,
+             const unittest::FooRequest* request,
+             unittest::FooResponse* response,
+             Closure* done) {
+      ASSERT_FALSE(called_);
+      called_ = true;
+      method_ = "Foo";
+      controller_ = controller;
+      request_ = request;
+      response_ = response;
+      done_ = done;
+    }
+
+    void Bar(RpcController* controller,
+             const unittest::BarRequest* request,
+             unittest::BarResponse* response,
+             Closure* done) {
+      ASSERT_FALSE(called_);
+      called_ = true;
+      method_ = "Bar";
+      controller_ = controller;
+      request_ = request;
+      response_ = response;
+      done_ = done;
+    }
+
+    // ---------------------------------------------------------------
+
+    bool called_;
+    string method_;
+    RpcController* controller_;
+    const Message* request_;
+    Message* response_;
+    Closure* done_;
+  };
+
+  class MockRpcChannel : public RpcChannel {
+   public:
+    MockRpcChannel()
+      : called_(false),
+        method_(NULL),
+        controller_(NULL),
+        request_(NULL),
+        response_(NULL),
+        done_(NULL),
+        destroyed_(NULL) {}
+
+    ~MockRpcChannel() {
+      if (destroyed_ != NULL) *destroyed_ = true;
+    }
+
+    void Reset() { called_ = false; }
+
+    // implements TestService ----------------------------------------
+
+    void CallMethod(const MethodDescriptor* method,
+                    RpcController* controller,
+                    const Message* request,
+                    Message* response,
+                    Closure* done) {
+      ASSERT_FALSE(called_);
+      called_ = true;
+      method_ = method;
+      controller_ = controller;
+      request_ = request;
+      response_ = response;
+      done_ = done;
+    }
+
+    // ---------------------------------------------------------------
+
+    bool called_;
+    const MethodDescriptor* method_;
+    RpcController* controller_;
+    const Message* request_;
+    Message* response_;
+    Closure* done_;
+    bool* destroyed_;
+  };
+
+  class MockController : public RpcController {
+   public:
+    void Reset() {
+      ADD_FAILURE() << "Reset() not expected during this test.";
+    }
+    bool Failed() const {
+      ADD_FAILURE() << "Failed() not expected during this test.";
+      return false;
+    }
+    string ErrorText() const {
+      ADD_FAILURE() << "ErrorText() not expected during this test.";
+      return "";
+    }
+    void StartCancel() {
+      ADD_FAILURE() << "StartCancel() not expected during this test.";
+    }
+    void SetFailed(const string& reason) {
+      ADD_FAILURE() << "SetFailed() not expected during this test.";
+    }
+    bool IsCanceled() const {
+      ADD_FAILURE() << "IsCanceled() not expected during this test.";
+      return false;
+    }
+    void NotifyOnCancel(Closure* callback) {
+      ADD_FAILURE() << "NotifyOnCancel() not expected during this test.";
+    }
+  };
+
+  GeneratedServiceTest()
+    : descriptor_(unittest::TestService::descriptor()),
+      foo_(descriptor_->FindMethodByName("Foo")),
+      bar_(descriptor_->FindMethodByName("Bar")),
+      stub_(&mock_channel_),
+      done_(NewPermanentCallback(&DoNothing)) {}
+
+  virtual void SetUp() {
+    ASSERT_TRUE(foo_ != NULL);
+    ASSERT_TRUE(bar_ != NULL);
+  }
+
+  const ServiceDescriptor* descriptor_;
+  const MethodDescriptor* foo_;
+  const MethodDescriptor* bar_;
+
+  MockTestService mock_service_;
+  MockController mock_controller_;
+
+  MockRpcChannel mock_channel_;
+  unittest::TestService::Stub stub_;
+
+  // Just so we don't have to re-define these with every test.
+  unittest::FooRequest foo_request_;
+  unittest::FooResponse foo_response_;
+  unittest::BarRequest bar_request_;
+  unittest::BarResponse bar_response_;
+  scoped_ptr<Closure> done_;
+};
+
+TEST_F(GeneratedServiceTest, GetDescriptor) {
+  // Test that GetDescriptor() works.
+
+  EXPECT_EQ(descriptor_, mock_service_.GetDescriptor());
+}
+
+TEST_F(GeneratedServiceTest, GetChannel) {
+  EXPECT_EQ(&mock_channel_, stub_.channel());
+}
+
+TEST_F(GeneratedServiceTest, OwnsChannel) {
+  MockRpcChannel* channel = new MockRpcChannel;
+  bool destroyed = false;
+  channel->destroyed_ = &destroyed;
+
+  {
+    unittest::TestService::Stub owning_stub(channel,
+                                            Service::STUB_OWNS_CHANNEL);
+    EXPECT_FALSE(destroyed);
+  }
+
+  EXPECT_TRUE(destroyed);
+}
+
+TEST_F(GeneratedServiceTest, CallMethod) {
+  // Test that CallMethod() works.
+
+  // Call Foo() via CallMethod().
+  mock_service_.CallMethod(foo_, &mock_controller_,
+                           &foo_request_, &foo_response_, done_.get());
+
+  ASSERT_TRUE(mock_service_.called_);
+
+  EXPECT_EQ("Foo"            , mock_service_.method_    );
+  EXPECT_EQ(&mock_controller_, mock_service_.controller_);
+  EXPECT_EQ(&foo_request_    , mock_service_.request_   );
+  EXPECT_EQ(&foo_response_   , mock_service_.response_  );
+  EXPECT_EQ(done_.get()      , mock_service_.done_      );
+
+  // Try again, but call Bar() instead.
+  mock_service_.Reset();
+  mock_service_.CallMethod(bar_, &mock_controller_,
+                           &bar_request_, &bar_response_, done_.get());
+
+  ASSERT_TRUE(mock_service_.called_);
+  EXPECT_EQ("Bar", mock_service_.method_);
+}
+
+TEST_F(GeneratedServiceTest, CallMethodTypeFailure) {
+  // Verify death if we call Foo() with Bar's message types.
+
+#ifdef GTEST_HAS_DEATH_TEST  // death tests do not work on Windows yet
+  EXPECT_DEBUG_DEATH(
+    mock_service_.CallMethod(foo_, &mock_controller_,
+                             &foo_request_, &bar_response_, done_.get()),
+    "dynamic_cast");
+
+  mock_service_.Reset();
+  EXPECT_DEBUG_DEATH(
+    mock_service_.CallMethod(foo_, &mock_controller_,
+                             &bar_request_, &foo_response_, done_.get()),
+    "dynamic_cast");
+#endif  // GTEST_HAS_DEATH_TEST
+}
+
+TEST_F(GeneratedServiceTest, GetPrototypes) {
+  // Test Get{Request,Response}Prototype() methods.
+
+  EXPECT_EQ(&unittest::FooRequest::default_instance(),
+            &mock_service_.GetRequestPrototype(foo_));
+  EXPECT_EQ(&unittest::BarRequest::default_instance(),
+            &mock_service_.GetRequestPrototype(bar_));
+
+  EXPECT_EQ(&unittest::FooResponse::default_instance(),
+            &mock_service_.GetResponsePrototype(foo_));
+  EXPECT_EQ(&unittest::BarResponse::default_instance(),
+            &mock_service_.GetResponsePrototype(bar_));
+}
+
+TEST_F(GeneratedServiceTest, Stub) {
+  // Test that the stub class works.
+
+  // Call Foo() via the stub.
+  stub_.Foo(&mock_controller_, &foo_request_, &foo_response_, done_.get());
+
+  ASSERT_TRUE(mock_channel_.called_);
+
+  EXPECT_EQ(foo_             , mock_channel_.method_    );
+  EXPECT_EQ(&mock_controller_, mock_channel_.controller_);
+  EXPECT_EQ(&foo_request_    , mock_channel_.request_   );
+  EXPECT_EQ(&foo_response_   , mock_channel_.response_  );
+  EXPECT_EQ(done_.get()      , mock_channel_.done_      );
+
+  // Call Bar() via the stub.
+  mock_channel_.Reset();
+  stub_.Bar(&mock_controller_, &bar_request_, &bar_response_, done_.get());
+
+  ASSERT_TRUE(mock_channel_.called_);
+  EXPECT_EQ(bar_, mock_channel_.method_);
+}
+
+TEST_F(GeneratedServiceTest, NotImplemented) {
+  // Test that failing to implement a method of a service causes it to fail
+  // with a "not implemented" error message.
+
+  // A service which doesn't implement any methods.
+  class UnimplementedService : public unittest::TestService {
+   public:
+    UnimplementedService() {}
+  };
+
+  UnimplementedService unimplemented_service;
+
+  // And a controller which expects to get a "not implemented" error.
+  class ExpectUnimplementedController : public MockController {
+   public:
+    ExpectUnimplementedController() : called_(false) {}
+
+    void SetFailed(const string& reason) {
+      EXPECT_FALSE(called_);
+      called_ = true;
+      EXPECT_EQ("Method Foo() not implemented.", reason);
+    }
+
+    bool called_;
+  };
+
+  ExpectUnimplementedController controller;
+
+  // Call Foo.
+  unimplemented_service.Foo(&controller, &foo_request_, &foo_response_,
+                            done_.get());
+
+  EXPECT_TRUE(controller.called_);
+}
+
+}  // namespace
+
+}  // namespace cpp
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/importer.cc b/src/google/protobuf/compiler/importer.cc
new file mode 100644
index 00000000..61182933
--- /dev/null
+++ b/src/google/protobuf/compiler/importer.cc
@@ -0,0 +1,398 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifdef _MSC_VER
+#include <io.h>
+#else
+#include <unistd.h>
+#endif
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <errno.h>
+
+#include <algorithm>
+
+#include <google/protobuf/compiler/importer.h>
+
+#include <google/protobuf/compiler/parser.h>
+#include <google/protobuf/io/tokenizer.h>
+#include <google/protobuf/io/zero_copy_stream_impl.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+
+#ifdef _WIN32
+#ifndef F_OK
+#define F_OK 00  // not defined by MSVC for whatever reason
+#endif
+#endif
+
+MultiFileErrorCollector::~MultiFileErrorCollector() {}
+
+// This class serves two purposes:
+// - It implements the ErrorCollector interface (used by Tokenizer and Parser)
+//   in terms of MultiFileErrorCollector, using a particular filename.
+// - It lets us check if any errors have occurred.
+class SourceTreeDescriptorDatabase::SingleFileErrorCollector
+    : public io::ErrorCollector {
+ public:
+  SingleFileErrorCollector(const string& filename,
+                           MultiFileErrorCollector* multi_file_error_collector)
+    : filename_(filename),
+      multi_file_error_collector_(multi_file_error_collector),
+      had_errors_(false) {}
+  ~SingleFileErrorCollector() {}
+
+  bool had_errors() { return had_errors_; }
+
+  // implements ErrorCollector ---------------------------------------
+  void AddError(int line, int column, const string& message) {
+    if (multi_file_error_collector_ != NULL) {
+      multi_file_error_collector_->AddError(filename_, line, column, message);
+    }
+    had_errors_ = true;
+  }
+
+ private:
+  string filename_;
+  MultiFileErrorCollector* multi_file_error_collector_;
+  bool had_errors_;
+};
+
+// ===================================================================
+
+SourceTreeDescriptorDatabase::SourceTreeDescriptorDatabase(
+    SourceTree* source_tree)
+  : source_tree_(source_tree),
+    error_collector_(NULL),
+    using_validation_error_collector_(false),
+    validation_error_collector_(this) {}
+
+SourceTreeDescriptorDatabase::~SourceTreeDescriptorDatabase() {}
+
+bool SourceTreeDescriptorDatabase::FindFileByName(
+    const string& filename, FileDescriptorProto* output) {
+  scoped_ptr<io::ZeroCopyInputStream> input(source_tree_->Open(filename));
+  if (input == NULL) {
+    if (error_collector_ != NULL) {
+      error_collector_->AddError(filename, -1, 0, "File not found.");
+    }
+    return false;
+  }
+
+  // Set up the tokenizer and parser.
+  SingleFileErrorCollector file_error_collector(filename, error_collector_);
+  io::Tokenizer tokenizer(input.get(), &file_error_collector);
+
+  Parser parser;
+  if (error_collector_ != NULL) {
+    parser.RecordErrorsTo(&file_error_collector);
+  }
+  if (using_validation_error_collector_) {
+    parser.RecordSourceLocationsTo(&source_locations_);
+  }
+
+  // Parse it.
+  output->set_name(filename);
+  return parser.Parse(&tokenizer, output) &&
+         !file_error_collector.had_errors();
+}
+
+bool SourceTreeDescriptorDatabase::FindFileContainingSymbol(
+    const string& symbol_name, FileDescriptorProto* output) {
+  return false;
+}
+
+bool SourceTreeDescriptorDatabase::FindFileContainingExtension(
+    const string& containing_type, int field_number,
+    FileDescriptorProto* output) {
+  return false;
+}
+
+// -------------------------------------------------------------------
+
+SourceTreeDescriptorDatabase::ValidationErrorCollector::
+ValidationErrorCollector(SourceTreeDescriptorDatabase* owner)
+  : owner_(owner) {}
+
+SourceTreeDescriptorDatabase::ValidationErrorCollector::
+~ValidationErrorCollector() {}
+
+void SourceTreeDescriptorDatabase::ValidationErrorCollector::AddError(
+    const string& filename,
+    const string& element_name,
+    const Message* descriptor,
+    ErrorLocation location,
+    const string& message) {
+  if (owner_->error_collector_ == NULL) return;
+
+  int line, column;
+  owner_->source_locations_.Find(descriptor, location, &line, &column);
+  owner_->error_collector_->AddError(filename, line, column, message);
+}
+
+// ===================================================================
+
+Importer::Importer(SourceTree* source_tree,
+                   MultiFileErrorCollector* error_collector)
+  : database_(source_tree),
+    pool_(&database_, database_.GetValidationErrorCollector()) {
+  database_.RecordErrorsTo(error_collector);
+}
+
+Importer::~Importer() {}
+
+const FileDescriptor* Importer::Import(const string& filename) {
+  return pool_.FindFileByName(filename);
+}
+
+// ===================================================================
+
+SourceTree::~SourceTree() {}
+
+DiskSourceTree::DiskSourceTree() {}
+
+DiskSourceTree::~DiskSourceTree() {}
+
+static inline char LastChar(const string& str) {
+  return str[str.size() - 1];
+}
+
+// Given a path, returns an equivalent path with these changes:
+// - On Windows, any backslashes are replaced with forward slashes.
+// - Any instances of the directory "." are removed.
+// - Any consecutive '/'s are collapsed into a single slash.
+// Note that the resulting string may be empty.
+//
+// TODO(kenton):  It would be nice to handle "..", e.g. so that we can figure
+//   out that "foo/bar.proto" is inside "baz/../foo".  However, if baz is a
+//   symlink or doesn't exist, then things get complicated, and we can't
+//   actually determine this without investigating the filesystem, probably
+//   in non-portable ways.  So, we punt.
+//
+// TODO(kenton):  It would be nice to use realpath() here except that it
+//   resolves symbolic links.  This could cause problems if people place
+//   symbolic links in their source tree.  For example, if you executed:
+//     protoc --proto_path=foo foo/bar/baz.proto
+//   then if foo/bar is a symbolic link, foo/bar/baz.proto will canonicalize
+//   to a path which does not appear to be under foo, and thus the compiler
+//   will complain that baz.proto is not inside the --proto_path.
+static string CanonicalizePath(string path) {
+#ifdef _WIN32
+  // The Win32 API accepts forward slashes as a path delimiter even though
+  // backslashes are standard.  Let's avoid confusion and use only forward
+  // slashes.
+  path = StringReplace(path, "\\", "/", true);
+#endif
+
+  vector<string> parts;
+  vector<string> canonical_parts;
+  SplitStringUsing(path, "/", &parts);  // Note:  Removes empty parts.
+  for (int i = 0; i < parts.size(); i++) {
+    if (parts[i] == ".") {
+      // Ignore.
+    } else {
+      canonical_parts.push_back(parts[i]);
+    }
+  }
+  string result = JoinStrings(canonical_parts, "/");
+  if (!path.empty() && path[0] == '/') {
+    // Restore leading slash.
+    result = '/' + result;
+  }
+  if (!path.empty() && LastChar(path) == '/' &&
+      !result.empty() && LastChar(result) != '/') {
+    // Restore trailing slash.
+    result += '/';
+  }
+  return result;
+}
+
+static inline bool ContainsParentReference(const string& path) {
+  return path == ".." ||
+         HasPrefixString(path, "../") ||
+         HasSuffixString(path, "/..") ||
+         path.find("/../") != string::npos;
+}
+
+// Maps a file from an old location to a new one.  Typically, old_prefix is
+// a virtual path and new_prefix is its corresponding disk path.  Returns
+// false if the filename did not start with old_prefix, otherwise replaces
+// old_prefix with new_prefix and stores the result in *result.  Examples:
+//   string result;
+//   assert(ApplyMapping("foo/bar", "", "baz", &result));
+//   assert(result == "baz/foo/bar");
+//
+//   assert(ApplyMapping("foo/bar", "foo", "baz", &result));
+//   assert(result == "baz/bar");
+//
+//   assert(ApplyMapping("foo", "foo", "bar", &result));
+//   assert(result == "bar");
+//
+//   assert(!ApplyMapping("foo/bar", "baz", "qux", &result));
+//   assert(!ApplyMapping("foo/bar", "baz", "qux", &result));
+//   assert(!ApplyMapping("foobar", "foo", "baz", &result));
+static bool ApplyMapping(const string& filename,
+                         const string& old_prefix,
+                         const string& new_prefix,
+                         string* result) {
+  if (old_prefix.empty()) {
+    // old_prefix matches any relative path.
+    if (ContainsParentReference(filename)) {
+      // We do not allow the file name to use "..".
+      return false;
+    }
+    if (HasPrefixString(filename, "/")) {
+      // This is an absolute path, so it isn't matched by the empty string.
+      return false;
+    }
+    result->assign(new_prefix);
+    if (!result->empty()) result->push_back('/');
+    result->append(filename);
+    return true;
+  } else if (HasPrefixString(filename, old_prefix)) {
+    // old_prefix is a prefix of the filename.  Is it the whole filename?
+    if (filename.size() == old_prefix.size()) {
+      // Yep, it's an exact match.
+      *result = new_prefix;
+      return true;
+    } else {
+      // Not an exact match.  Is the next character a '/'?  Otherwise,
+      // this isn't actually a match at all.  E.g. the prefix "foo/bar"
+      // does not match the filename "foo/barbaz".
+      if (filename[old_prefix.size()] == '/') {
+        // Yep.  So the prefixes are directories and the filename is a file
+        // inside them.
+        string after_prefix = filename.substr(old_prefix.size() + 1);
+        if (ContainsParentReference(after_prefix)) {
+          // We do not allow the file name to use "..".
+          return false;
+        }
+        result->assign(new_prefix);
+        if (!result->empty()) result->push_back('/');
+        result->append(after_prefix);
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
+void DiskSourceTree::MapPath(const string& virtual_path,
+                             const string& disk_path) {
+  mappings_.push_back(Mapping(virtual_path, CanonicalizePath(disk_path)));
+}
+
+DiskSourceTree::DiskFileToVirtualFileResult
+DiskSourceTree::DiskFileToVirtualFile(
+    const string& disk_file,
+    string* virtual_file,
+    string* shadowing_disk_file) {
+  int mapping_index = -1;
+  string canonical_disk_file = CanonicalizePath(disk_file);
+
+  for (int i = 0; i < mappings_.size(); i++) {
+    // Apply the mapping in reverse.
+    if (ApplyMapping(canonical_disk_file, mappings_[i].disk_path,
+                     mappings_[i].virtual_path, virtual_file)) {
+      // Success.
+      mapping_index = i;
+      break;
+    }
+  }
+
+  if (mapping_index == -1) {
+    return NO_MAPPING;
+  }
+
+  // Iterate through all mappings with higher precedence and verify that none
+  // of them map this file to some other existing file.
+  for (int i = 0; i < mapping_index; i++) {
+    if (ApplyMapping(*virtual_file, mappings_[i].virtual_path,
+                     mappings_[i].disk_path, shadowing_disk_file)) {
+      if (access(shadowing_disk_file->c_str(), F_OK) >= 0) {
+        // File exists.
+        return SHADOWED;
+      }
+    }
+  }
+  shadowing_disk_file->clear();
+
+  // Verify that we can open the file.  Note that this also has the side-effect
+  // of verifying that we are not canonicalizing away any non-existent
+  // directories.
+  scoped_ptr<io::ZeroCopyInputStream> stream(OpenDiskFile(disk_file));
+  if (stream == NULL) {
+    return CANNOT_OPEN;
+  }
+
+  return SUCCESS;
+}
+
+io::ZeroCopyInputStream* DiskSourceTree::Open(const string& filename) {
+  if (filename != CanonicalizePath(filename) ||
+      ContainsParentReference(filename)) {
+    // We do not allow importing of paths containing things like ".." or
+    // consecutive slashes since the compiler expects files to be uniquely
+    // identified by file name.
+    return NULL;
+  }
+
+  for (int i = 0; i < mappings_.size(); i++) {
+    string disk_file;
+    if (ApplyMapping(filename, mappings_[i].virtual_path,
+                     mappings_[i].disk_path, &disk_file)) {
+      io::ZeroCopyInputStream* stream = OpenDiskFile(disk_file);
+      if (stream != NULL) return stream;
+
+      if (errno == EACCES) {
+        // The file exists but is not readable.
+        // TODO(kenton):  Find a way to report this more nicely.
+        GOOGLE_LOG(WARNING) << "Read access is denied for file: " << disk_file;
+        return NULL;
+      }
+    }
+  }
+
+  return NULL;
+}
+
+io::ZeroCopyInputStream* DiskSourceTree::OpenDiskFile(
+    const string& filename) {
+  int file_descriptor;
+  do {
+    file_descriptor = open(filename.c_str(), O_RDONLY);
+  } while (file_descriptor < 0 && errno == EINTR);
+  if (file_descriptor >= 0) {
+    io::FileInputStream* result = new io::FileInputStream(file_descriptor);
+    result->SetCloseOnDelete(true);
+    return result;
+  } else {
+    return NULL;
+  }
+}
+
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/importer.h b/src/google/protobuf/compiler/importer.h
new file mode 100644
index 00000000..2e3d0df4
--- /dev/null
+++ b/src/google/protobuf/compiler/importer.h
@@ -0,0 +1,279 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// This file is the public interface to the .proto file parser.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_IMPORTER_H__
+#define GOOGLE_PROTOBUF_COMPILER_IMPORTER_H__
+
+#include <string>
+#include <vector>
+#include <set>
+#include <utility>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/descriptor_database.h>
+#include <google/protobuf/compiler/parser.h>
+
+namespace google {
+namespace protobuf {
+
+namespace io { class ZeroCopyInputStream; }
+
+namespace compiler {
+
+// Defined in this file.
+class Importer;
+class MultiFileErrorCollector;
+class SourceTree;
+class DiskSourceTree;
+
+// TODO(kenton):  Move all SourceTree stuff to a separate file?
+
+// An implementation of DescriptorDatabase which loads files from a SourceTree
+// and parses them.
+//
+// Note:  This class is not thread-safe since it maintains a table of source
+//   code locations for error reporting.  However, when a DescriptorPool wraps
+//   a DescriptorDatabase, it uses mutex locking to make sure only one method
+//   of the database is called at a time, even if the DescriptorPool is used
+//   from multiple threads.  Therefore, there is only a problem if you create
+//   multiple DescriptorPools wrapping the same SourceTreeDescriptorDatabase
+//   and use them from multiple threads.
+//
+// Note:  This class does not implement FindFileContainingSymbol() or
+//   FindFileContainingExtension(); these will always return false.
+class LIBPROTOBUF_EXPORT SourceTreeDescriptorDatabase : public DescriptorDatabase {
+ public:
+  SourceTreeDescriptorDatabase(SourceTree* source_tree);
+  ~SourceTreeDescriptorDatabase();
+
+  // Instructs the SourceTreeDescriptorDatabase to report any parse errors
+  // to the given MultiFileErrorCollector.  This should be called before
+  // parsing.  error_collector must remain valid until either this method
+  // is called again or the SourceTreeDescriptorDatabase is destroyed.
+  void RecordErrorsTo(MultiFileErrorCollector* error_collector) {
+    error_collector_ = error_collector;
+  }
+
+  // Gets a DescriptorPool::ErrorCollector which records errors to the
+  // MultiFileErrorCollector specified with RecordErrorsTo().  This collector
+  // has the ability to determine exact line and column numbers of errors
+  // from the information given to it by the DescriptorPool.
+  DescriptorPool::ErrorCollector* GetValidationErrorCollector() {
+    using_validation_error_collector_ = true;
+    return &validation_error_collector_;
+  }
+
+  // implements DescriptorDatabase -----------------------------------
+  bool FindFileByName(const string& filename, FileDescriptorProto* output);
+  bool FindFileContainingSymbol(const string& symbol_name,
+                                FileDescriptorProto* output);
+  bool FindFileContainingExtension(const string& containing_type,
+                                   int field_number,
+                                   FileDescriptorProto* output);
+
+ private:
+  class SingleFileErrorCollector;
+
+  SourceTree* source_tree_;
+  MultiFileErrorCollector* error_collector_;
+
+  class LIBPROTOBUF_EXPORT ValidationErrorCollector : public DescriptorPool::ErrorCollector {
+   public:
+    ValidationErrorCollector(SourceTreeDescriptorDatabase* owner);
+    ~ValidationErrorCollector();
+
+    // implements ErrorCollector ---------------------------------------
+    void AddError(const string& filename,
+                  const string& element_name,
+                  const Message* descriptor,
+                  ErrorLocation location,
+                  const string& message);
+
+   private:
+    SourceTreeDescriptorDatabase* owner_;
+  };
+  friend class ValidationErrorCollector;
+
+  bool using_validation_error_collector_;
+  SourceLocationTable source_locations_;
+  ValidationErrorCollector validation_error_collector_;
+};
+
+// Simple interface for parsing .proto files.  This wraps the process
+// of opening the file, parsing it with a Parser, recursively parsing all its
+// imports, and then cross-linking the results to produce a FileDescriptor.
+//
+// This is really just a thin wrapper around SourceTreeDescriptorDatabase.
+// You may find that SourceTreeDescriptorDatabase is more flexible.
+//
+// TODO(kenton):  I feel like this class is not well-named.
+class LIBPROTOBUF_EXPORT Importer {
+ public:
+  Importer(SourceTree* source_tree,
+           MultiFileErrorCollector* error_collector);
+  ~Importer();
+
+  // Import the given file and build a FileDescriptor representing it.  If
+  // the file is already in the DescriptorPool, the existing FileDescriptor
+  // will be returned.  The FileDescriptor is property of the DescriptorPool,
+  // and will remain valid until it is destroyed.  If any errors occur, they
+  // will be reported using the error collector and Import() will return NULL.
+  //
+  // A particular Importer object will only report errors for a particular
+  // file once.  All future attempts to import the same file will return NULL
+  // without reporting any errors.  The idea is that you might want to import
+  // a lot of files without seeing the same errors over and over again.  If
+  // you want to see errors for the same files repeatedly, you can use a
+  // separate Importer object to import each one (but use the same
+  // DescriptorPool so that they can be cross-linked).
+  const FileDescriptor* Import(const string& filename);
+
+  // The DescriptorPool in which all imported FileDescriptors and their
+  // contents are stored.
+  inline const DescriptorPool* pool() const {
+    return &pool_;
+  }
+
+ private:
+  SourceTreeDescriptorDatabase database_;
+  DescriptorPool pool_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(Importer);
+};
+
+// If the importer encounters problems while trying to import the proto files,
+// it reports them to a MultiFileErrorCollector.
+class LIBPROTOBUF_EXPORT MultiFileErrorCollector {
+ public:
+  inline MultiFileErrorCollector() {}
+  virtual ~MultiFileErrorCollector();
+
+  // Line and column numbers are zero-based.  A line number of -1 indicates
+  // an error with the entire file (e.g. "not found").
+  virtual void AddError(const string& filename, int line, int column,
+                        const string& message) = 0;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(MultiFileErrorCollector);
+};
+
+// Abstract interface which represents a directory tree containing proto files.
+// Used by the default implementation of Importer to resolve import statements
+// Most users will probably want to use the DiskSourceTree implementation,
+// below.
+class LIBPROTOBUF_EXPORT SourceTree {
+ public:
+  inline SourceTree() {}
+  virtual ~SourceTree();
+
+  // Open the given file and return a stream that reads it, or NULL if not
+  // found.  The caller takes ownership of the returned object.  The filename
+  // must be a path relative to the root of the source tree and must not
+  // contain "." or ".." components.
+  virtual io::ZeroCopyInputStream* Open(const string& filename) = 0;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(SourceTree);
+};
+
+// An implementation of SourceTree which loads files from locations on disk.
+// Multiple mappings can be set up to map locations in the DiskSourceTree to
+// locations in the physical filesystem.
+class LIBPROTOBUF_EXPORT DiskSourceTree : public SourceTree {
+ public:
+  DiskSourceTree();
+  ~DiskSourceTree();
+
+  // Map a path on disk to a location in the SourceTree.  The path may be
+  // either a file or a directory.  If it is a directory, the entire tree
+  // under it will be mapped to the given virtual location.  To map a directory
+  // to the root of the source tree, pass an empty string for virtual_path.
+  //
+  // If multiple mapped paths apply when opening a file, they will be searched
+  // in order.  For example, if you do:
+  //   MapPath("bar", "foo/bar");
+  //   MapPath("", "baz");
+  // and then you do:
+  //   Open("bar/qux");
+  // the DiskSourceTree will first try to open foo/bar/qux, then baz/bar/qux,
+  // returning the first one that opens successfuly.
+  //
+  // disk_path may be an absolute path or relative to the current directory,
+  // just like a path you'd pass to open().
+  void MapPath(const string& virtual_path, const string& disk_path);
+
+  // Return type for DiskFileToVirtualFile().
+  enum DiskFileToVirtualFileResult {
+    SUCCESS,
+    SHADOWED,
+    CANNOT_OPEN,
+    NO_MAPPING
+  };
+
+  // Given a path to a file on disk, find a virtual path mapping to that
+  // file.  The first mapping created with MapPath() whose disk_path contains
+  // the filename is used.  However, that virtual path may not actually be
+  // usable to open the given file.  Possible return values are:
+  // * SUCCESS: The mapping was found.  *virtual_file is filled in so that
+  //   calling Open(*virtual_file) will open the file named by disk_file.
+  // * SHADOWED: A mapping was found, but using Open() to open this virtual
+  //   path will end up returning some different file.  This is because some
+  //   other mapping with a higher precedence also matches this virtual path
+  //   and maps it to a different file that exists on disk.  *virtual_file
+  //   is filled in as it would be in the SUCCESS case.  *shadowing_disk_file
+  //   is filled in with the disk path of the file which would be opened if
+  //   you were to call Open(*virtual_file).
+  // * CANNOT_OPEN: The mapping was found and was not shadowed, but the
+  //   file specified cannot be opened.  When this value is returned,
+  //   errno will indicate the reason the file cannot be opened.  *virtual_file
+  //   will be set to the virtual path as in the SUCCESS case, even though
+  //   it is not useful.
+  // * NO_MAPPING: Indicates that no mapping was found which contains this
+  //   file.
+  DiskFileToVirtualFileResult
+    DiskFileToVirtualFile(const string& disk_file,
+                          string* virtual_file,
+                          string* shadowing_disk_file);
+
+  // implements SourceTree -------------------------------------------
+  io::ZeroCopyInputStream* Open(const string& filename);
+
+ private:
+  struct Mapping {
+    string virtual_path;
+    string disk_path;
+
+    inline Mapping(const string& virtual_path, const string& disk_path)
+      : virtual_path(virtual_path), disk_path(disk_path) {}
+  };
+  vector<Mapping> mappings_;
+
+  // Like Open() but given the actual on-disk path.
+  io::ZeroCopyInputStream* OpenDiskFile(const string& filename);
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(DiskSourceTree);
+};
+
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_IMPORTER_H__
diff --git a/src/google/protobuf/compiler/importer_unittest.cc b/src/google/protobuf/compiler/importer_unittest.cc
new file mode 100644
index 00000000..5b5d2831
--- /dev/null
+++ b/src/google/protobuf/compiler/importer_unittest.cc
@@ -0,0 +1,539 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/stubs/hash.h>
+
+#include <google/protobuf/compiler/importer.h>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/io/zero_copy_stream_impl.h>
+
+#include <google/protobuf/stubs/map-util.h>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/testing/file.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/stubs/substitute.h>
+#include <google/protobuf/testing/googletest.h>
+#include <gtest/gtest.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+
+namespace {
+
+#define EXPECT_SUBSTRING(needle, haystack) \
+  EXPECT_PRED_FORMAT2(testing::IsSubstring, (needle), (haystack))
+
+class MockErrorCollector : public MultiFileErrorCollector {
+ public:
+  MockErrorCollector() {}
+  ~MockErrorCollector() {}
+
+  string text_;
+
+  // implements ErrorCollector ---------------------------------------
+  void AddError(const string& filename, int line, int column,
+                const string& message) {
+    strings::SubstituteAndAppend(&text_, "$0:$1:$2: $3\n",
+                                 filename, line, column, message);
+  }
+};
+
+// -------------------------------------------------------------------
+
+// A dummy implementation of SourceTree backed by a simple map.
+class MockSourceTree : public SourceTree {
+ public:
+  MockSourceTree() {}
+  ~MockSourceTree() {}
+
+  void AddFile(const string& name, const char* contents) {
+    files_[name] = contents;
+  }
+
+  // implements SourceTree -------------------------------------------
+  io::ZeroCopyInputStream* Open(const string& filename) {
+    const char* contents = FindPtrOrNull(files_, filename);
+    if (contents == NULL) {
+      return NULL;
+    } else {
+      return new io::ArrayInputStream(contents, strlen(contents));
+    }
+  }
+
+ private:
+  hash_map<string, const char*> files_;
+};
+
+// ===================================================================
+
+class ImporterTest : public testing::Test {
+ protected:
+  ImporterTest()
+    : importer_(&source_tree_, &error_collector_) {}
+
+  void AddFile(const string& filename, const char* text) {
+    source_tree_.AddFile(filename, text);
+  }
+
+  // Return the collected error text
+  string error() const { return error_collector_.text_; }
+
+  MockErrorCollector error_collector_;
+  MockSourceTree source_tree_;
+  Importer importer_;
+};
+
+TEST_F(ImporterTest, Import) {
+  // Test normal importing.
+  AddFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "message Foo {}\n");
+
+  const FileDescriptor* file = importer_.Import("foo.proto");
+  EXPECT_EQ("", error_collector_.text_);
+  ASSERT_TRUE(file != NULL);
+
+  ASSERT_EQ(1, file->message_type_count());
+  EXPECT_EQ("Foo", file->message_type(0)->name());
+
+  // Importing again should return same object.
+  EXPECT_EQ(file, importer_.Import("foo.proto"));
+}
+
+TEST_F(ImporterTest, ImportNested) {
+  // Test that importing a file which imports another file works.
+  AddFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "import \"bar.proto\";\n"
+    "message Foo {\n"
+    "  optional Bar bar = 1;\n"
+    "}\n");
+  AddFile("bar.proto",
+    "syntax = \"proto2\";\n"
+    "message Bar {}\n");
+
+  // Note that both files are actually parsed by the first call to Import()
+  // here, since foo.proto imports bar.proto.  The second call just returns
+  // the same ProtoFile for bar.proto which was constructed while importing
+  // foo.proto.  We test that this is the case below by checking that bar
+  // is among foo's dependencies (by pointer).
+  const FileDescriptor* foo = importer_.Import("foo.proto");
+  const FileDescriptor* bar = importer_.Import("bar.proto");
+  EXPECT_EQ("", error_collector_.text_);
+  ASSERT_TRUE(foo != NULL);
+  ASSERT_TRUE(bar != NULL);
+
+  // Check that foo's dependency is the same object as bar.
+  ASSERT_EQ(1, foo->dependency_count());
+  EXPECT_EQ(bar, foo->dependency(0));
+
+  // Check that foo properly cross-links bar.
+  ASSERT_EQ(1, foo->message_type_count());
+  ASSERT_EQ(1, bar->message_type_count());
+  ASSERT_EQ(1, foo->message_type(0)->field_count());
+  ASSERT_EQ(FieldDescriptor::TYPE_MESSAGE,
+            foo->message_type(0)->field(0)->type());
+  EXPECT_EQ(bar->message_type(0),
+            foo->message_type(0)->field(0)->message_type());
+}
+
+TEST_F(ImporterTest, FileNotFound) {
+  // Error:  Parsing a file that doesn't exist.
+  EXPECT_TRUE(importer_.Import("foo.proto") == NULL);
+  EXPECT_EQ(
+    "foo.proto:-1:0: File not found.\n",
+    error_collector_.text_);
+}
+
+TEST_F(ImporterTest, ImportNotFound) {
+  // Error:  Importing a file that doesn't exist.
+  AddFile("foo.proto",
+    "syntax = \"proto2\";\n"
+    "import \"bar.proto\";\n");
+
+  EXPECT_TRUE(importer_.Import("foo.proto") == NULL);
+  EXPECT_EQ(
+    "bar.proto:-1:0: File not found.\n"
+    "foo.proto:-1:0: Import \"bar.proto\" was not found or had errors.\n",
+    error_collector_.text_);
+}
+
+TEST_F(ImporterTest, RecursiveImport) {
+  // Error:  Recursive import.
+  AddFile("recursive1.proto",
+    "syntax = \"proto2\";\n"
+    "import \"recursive2.proto\";\n");
+  AddFile("recursive2.proto",
+    "syntax = \"proto2\";\n"
+    "import \"recursive1.proto\";\n");
+
+  EXPECT_TRUE(importer_.Import("recursive1.proto") == NULL);
+  EXPECT_EQ(
+    "recursive1.proto:-1:0: File recursively imports itself: recursive1.proto "
+      "-> recursive2.proto -> recursive1.proto\n"
+    "recursive2.proto:-1:0: Import \"recursive1.proto\" was not found "
+      "or had errors.\n"
+    "recursive1.proto:-1:0: Import \"recursive2.proto\" was not found "
+      "or had errors.\n",
+    error_collector_.text_);
+}
+
+// TODO(sanjay): The MapField tests below more properly belong in
+// descriptor_unittest, but are more convenient to test here.
+TEST_F(ImporterTest, MapFieldValid) {
+  AddFile(
+      "map.proto",
+      "syntax = \"proto2\";\n"
+      "message Item {\n"
+      "  required string key = 1;\n"
+      "}\n"
+      "message Map {\n"
+      "  repeated Item items = 1 [experimental_map_key = \"key\"];\n"
+      "}\n"
+      );
+  const FileDescriptor* file = importer_.Import("map.proto");
+  ASSERT_TRUE(file != NULL) << error_collector_.text_;
+  EXPECT_EQ("", error_collector_.text_);
+
+  // Check that Map::items points to Item::key
+  const Descriptor* item_type = file->FindMessageTypeByName("Item");
+  ASSERT_TRUE(item_type != NULL);
+  const Descriptor* map_type = file->FindMessageTypeByName("Map");
+  ASSERT_TRUE(map_type != NULL);
+  const FieldDescriptor* key_field = item_type->FindFieldByName("key");
+  ASSERT_TRUE(key_field != NULL);
+  const FieldDescriptor* items_field = map_type->FindFieldByName("items");
+  ASSERT_TRUE(items_field != NULL);
+  EXPECT_EQ(items_field->experimental_map_key(), key_field);
+}
+
+TEST_F(ImporterTest, MapFieldNotRepeated) {
+  AddFile(
+      "map.proto",
+      "syntax = \"proto2\";\n"
+      "message Item {\n"
+      "  required string key = 1;\n"
+      "}\n"
+      "message Map {\n"
+      "  required Item items = 1 [experimental_map_key = \"key\"];\n"
+      "}\n"
+      );
+  EXPECT_TRUE(importer_.Import("map.proto") == NULL);
+  EXPECT_SUBSTRING("only allowed for repeated fields", error());
+}
+
+TEST_F(ImporterTest, MapFieldNotMessageType) {
+  AddFile(
+      "map.proto",
+      "syntax = \"proto2\";\n"
+      "message Map {\n"
+      "  repeated int32 items = 1 [experimental_map_key = \"key\"];\n"
+      "}\n"
+      );
+  EXPECT_TRUE(importer_.Import("map.proto") == NULL);
+  EXPECT_SUBSTRING("only allowed for fields with a message type", error());
+}
+
+TEST_F(ImporterTest, MapFieldTypeNotFound) {
+  AddFile(
+      "map.proto",
+      "syntax = \"proto2\";\n"
+      "message Map {\n"
+      "  repeated Unknown items = 1 [experimental_map_key = \"key\"];\n"
+      "}\n"
+      );
+  EXPECT_TRUE(importer_.Import("map.proto") == NULL);
+  EXPECT_SUBSTRING("not defined", error());
+}
+
+TEST_F(ImporterTest, MapFieldKeyNotFound) {
+  AddFile(
+      "map.proto",
+      "syntax = \"proto2\";\n"
+      "message Item {\n"
+      "  required string key = 1;\n"
+      "}\n"
+      "message Map {\n"
+      "  repeated Item items = 1 [experimental_map_key = \"badkey\"];\n"
+      "}\n"
+      );
+  EXPECT_TRUE(importer_.Import("map.proto") == NULL);
+  EXPECT_SUBSTRING("Could not find field", error());
+}
+
+TEST_F(ImporterTest, MapFieldKeyRepeated) {
+  AddFile(
+      "map.proto",
+      "syntax = \"proto2\";\n"
+      "message Item {\n"
+      "  repeated string key = 1;\n"
+      "}\n"
+      "message Map {\n"
+      "  repeated Item items = 1 [experimental_map_key = \"key\"];\n"
+      "}\n"
+      );
+  EXPECT_TRUE(importer_.Import("map.proto") == NULL);
+  EXPECT_SUBSTRING("must not name a repeated field", error());
+}
+
+TEST_F(ImporterTest, MapFieldKeyNotScalar) {
+  AddFile(
+      "map.proto",
+      "syntax = \"proto2\";\n"
+      "message ItemKey { }\n"
+      "message Item {\n"
+      "  required ItemKey key = 1;\n"
+      "}\n"
+      "message Map {\n"
+      "  repeated Item items = 1 [experimental_map_key = \"key\"];\n"
+      "}\n"
+      );
+  EXPECT_TRUE(importer_.Import("map.proto") == NULL);
+  EXPECT_SUBSTRING("must name a scalar or string", error());
+}
+
+// ===================================================================
+
+class DiskSourceTreeTest : public testing::Test {
+ protected:
+  virtual void SetUp() {
+    dirnames_.push_back(TestTempDir() + "/test_proto2_import_path_1");
+    dirnames_.push_back(TestTempDir() + "/test_proto2_import_path_2");
+
+    for (int i = 0; i < dirnames_.size(); i++) {
+      if (File::Exists(dirnames_[i])) {
+        File::DeleteRecursively(dirnames_[i], NULL, NULL);
+      }
+      GOOGLE_CHECK(File::CreateDir(dirnames_[i].c_str(), DEFAULT_FILE_MODE));
+    }
+  }
+
+  virtual void TearDown() {
+    for (int i = 0; i < dirnames_.size(); i++) {
+      File::DeleteRecursively(dirnames_[i], NULL, NULL);
+    }
+  }
+
+  void AddFile(const string& filename, const char* contents) {
+    File::WriteStringToFileOrDie(contents, filename);
+  }
+
+  void AddSubdir(const string& dirname) {
+    GOOGLE_CHECK(File::CreateDir(dirname.c_str(), DEFAULT_FILE_MODE));
+  }
+
+  void ExpectFileContents(const string& filename,
+                          const char* expected_contents) {
+    scoped_ptr<io::ZeroCopyInputStream> input(source_tree_.Open(filename));
+
+    ASSERT_FALSE(input == NULL);
+
+    // Read all the data from the file.
+    string file_contents;
+    const void* data;
+    int size;
+    while (input->Next(&data, &size)) {
+      file_contents.append(reinterpret_cast<const char*>(data), size);
+    }
+
+    EXPECT_EQ(expected_contents, file_contents);
+  }
+
+  void ExpectFileNotFound(const string& filename) {
+    scoped_ptr<io::ZeroCopyInputStream> input(source_tree_.Open(filename));
+    EXPECT_TRUE(input == NULL);
+  }
+
+  DiskSourceTree source_tree_;
+
+  // Paths of two on-disk directories to use during the test.
+  vector<string> dirnames_;
+};
+
+TEST_F(DiskSourceTreeTest, MapRoot) {
+  // Test opening a file in a directory that is mapped to the root of the
+  // source tree.
+  AddFile(dirnames_[0] + "/foo", "Hello World!");
+  source_tree_.MapPath("", dirnames_[0]);
+
+  ExpectFileContents("foo", "Hello World!");
+  ExpectFileNotFound("bar");
+}
+
+TEST_F(DiskSourceTreeTest, MapDirectory) {
+  // Test opening a file in a directory that is mapped to somewhere other
+  // than the root of the source tree.
+
+  AddFile(dirnames_[0] + "/foo", "Hello World!");
+  source_tree_.MapPath("baz", dirnames_[0]);
+
+  ExpectFileContents("baz/foo", "Hello World!");
+  ExpectFileNotFound("baz/bar");
+  ExpectFileNotFound("foo");
+  ExpectFileNotFound("bar");
+
+  // Non-canonical file names should not work.
+  ExpectFileNotFound("baz//foo");
+  ExpectFileNotFound("baz/../baz/foo");
+  ExpectFileNotFound("baz/./foo");
+  ExpectFileNotFound("baz/foo/");
+}
+
+TEST_F(DiskSourceTreeTest, NoParent) {
+  // Test that we cannot open files in a parent of a mapped directory.
+
+  AddFile(dirnames_[0] + "/foo", "Hello World!");
+  AddSubdir(dirnames_[0] + "/bar");
+  AddFile(dirnames_[0] + "/bar/baz", "Blah.");
+  source_tree_.MapPath("", dirnames_[0] + "/bar");
+
+  ExpectFileContents("baz", "Blah.");
+  ExpectFileNotFound("../foo");
+  ExpectFileNotFound("../bar/baz");
+}
+
+TEST_F(DiskSourceTreeTest, MapFile) {
+  // Test opening a file that is mapped directly into the source tree.
+
+  AddFile(dirnames_[0] + "/foo", "Hello World!");
+  source_tree_.MapPath("foo", dirnames_[0] + "/foo");
+
+  ExpectFileContents("foo", "Hello World!");
+  ExpectFileNotFound("bar");
+}
+
+TEST_F(DiskSourceTreeTest, SearchMultipleDirectories) {
+  // Test mapping and searching multiple directories.
+
+  AddFile(dirnames_[0] + "/foo", "Hello World!");
+  AddFile(dirnames_[1] + "/foo", "This file should be hidden.");
+  AddFile(dirnames_[1] + "/bar", "Goodbye World!");
+  source_tree_.MapPath("", dirnames_[0]);
+  source_tree_.MapPath("", dirnames_[1]);
+
+  ExpectFileContents("foo", "Hello World!");
+  ExpectFileContents("bar", "Goodbye World!");
+  ExpectFileNotFound("baz");
+}
+
+TEST_F(DiskSourceTreeTest, OrderingTrumpsSpecificity) {
+  // Test that directories are always searched in order, even when a latter
+  // directory is more-specific than a former one.
+
+  // Create the "bar" directory so we can put a file in it.
+  ASSERT_TRUE(File::CreateDir((dirnames_[0] + "/bar").c_str(),
+                              DEFAULT_FILE_MODE));
+
+  // Add files and map paths.
+  AddFile(dirnames_[0] + "/bar/foo", "Hello World!");
+  AddFile(dirnames_[1] + "/foo", "This file should be hidden.");
+  source_tree_.MapPath("", dirnames_[0]);
+  source_tree_.MapPath("bar", dirnames_[1]);
+
+  // Check.
+  ExpectFileContents("bar/foo", "Hello World!");
+}
+
+TEST_F(DiskSourceTreeTest, DiskFileToVirtualFile) {
+  // Test DiskFileToVirtualFile.
+
+  AddFile(dirnames_[0] + "/foo", "Hello World!");
+  AddFile(dirnames_[1] + "/foo", "This file should be hidden.");
+  source_tree_.MapPath("bar", dirnames_[0]);
+  source_tree_.MapPath("bar", dirnames_[1]);
+
+  string virtual_file;
+  string shadowing_disk_file;
+
+  EXPECT_EQ(DiskSourceTree::NO_MAPPING,
+    source_tree_.DiskFileToVirtualFile(
+      "/foo", &virtual_file, &shadowing_disk_file));
+
+  EXPECT_EQ(DiskSourceTree::SHADOWED,
+    source_tree_.DiskFileToVirtualFile(
+      dirnames_[1] + "/foo", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("bar/foo", virtual_file);
+  EXPECT_EQ(dirnames_[0] + "/foo", shadowing_disk_file);
+
+  EXPECT_EQ(DiskSourceTree::CANNOT_OPEN,
+    source_tree_.DiskFileToVirtualFile(
+      dirnames_[1] + "/baz", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("bar/baz", virtual_file);
+
+  EXPECT_EQ(DiskSourceTree::SUCCESS,
+    source_tree_.DiskFileToVirtualFile(
+      dirnames_[0] + "/foo", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("bar/foo", virtual_file);
+}
+
+TEST_F(DiskSourceTreeTest, DiskFileToVirtualFileCanonicalization) {
+  // Test handling of "..", ".", etc. in DiskFileToVirtualFile().
+
+  source_tree_.MapPath("dir1", "..");
+  source_tree_.MapPath("dir2", "../../foo");
+  source_tree_.MapPath("dir3", "./foo/bar/.");
+  source_tree_.MapPath("dir4", ".");
+  source_tree_.MapPath("", "/qux");
+
+  string virtual_file;
+  string shadowing_disk_file;
+
+  // "../.." should not be considered to be under "..".
+  EXPECT_EQ(DiskSourceTree::NO_MAPPING,
+    source_tree_.DiskFileToVirtualFile(
+      "../../baz", &virtual_file, &shadowing_disk_file));
+
+  // But "../baz" should be.
+  EXPECT_EQ(DiskSourceTree::CANNOT_OPEN,
+    source_tree_.DiskFileToVirtualFile(
+      "../baz", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("dir1/baz", virtual_file);
+
+  // "../../foo/baz" is under "../../foo".
+  EXPECT_EQ(DiskSourceTree::CANNOT_OPEN,
+    source_tree_.DiskFileToVirtualFile(
+      "../../foo/baz", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("dir2/baz", virtual_file);
+
+  // "foo/./bar/baz" is under "./foo/bar/.".
+  EXPECT_EQ(DiskSourceTree::CANNOT_OPEN,
+    source_tree_.DiskFileToVirtualFile(
+      "foo/bar/baz", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("dir3/baz", virtual_file);
+
+  // "bar" is under ".".
+  EXPECT_EQ(DiskSourceTree::CANNOT_OPEN,
+    source_tree_.DiskFileToVirtualFile(
+      "bar", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("dir4/bar", virtual_file);
+
+  // "/qux/baz" is under "/qux".
+  EXPECT_EQ(DiskSourceTree::CANNOT_OPEN,
+    source_tree_.DiskFileToVirtualFile(
+      "/qux/baz", &virtual_file, &shadowing_disk_file));
+  EXPECT_EQ("baz", virtual_file);
+}
+
+}  // namespace
+
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_enum.cc b/src/google/protobuf/compiler/java/java_enum.cc
new file mode 100644
index 00000000..aa15a468
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_enum.cc
@@ -0,0 +1,189 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <map>
+#include <string>
+
+#include <google/protobuf/compiler/java/java_enum.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+EnumGenerator::EnumGenerator(const EnumDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  for (int i = 0; i < descriptor_->value_count(); i++) {
+    const EnumValueDescriptor* value = descriptor_->value(i);
+    const EnumValueDescriptor* canonical_value =
+      descriptor_->FindValueByNumber(value->number());
+
+    if (value == canonical_value) {
+      canonical_values_.push_back(value);
+    } else {
+      Alias alias;
+      alias.value = value;
+      alias.canonical_value = canonical_value;
+      aliases_.push_back(alias);
+    }
+  }
+}
+
+EnumGenerator::~EnumGenerator() {}
+
+void EnumGenerator::Generate(io::Printer* printer) {
+  bool is_own_file =
+    descriptor_->containing_type() == NULL &&
+    descriptor_->file()->options().java_multiple_files();
+  printer->Print(
+    "public $static$ enum $classname$ {\n",
+    "static", is_own_file ? "" : "static",
+    "classname", descriptor_->name());
+  printer->Indent();
+
+  for (int i = 0; i < canonical_values_.size(); i++) {
+    map<string, string> vars;
+    vars["name"] = canonical_values_[i]->name();
+    vars["index"] = SimpleItoa(canonical_values_[i]->index());
+    vars["number"] = SimpleItoa(canonical_values_[i]->number());
+    printer->Print(vars,
+      "$name$($index$, $number$),\n");
+  }
+
+  printer->Print(
+    ";\n"
+    "\n");
+
+  // -----------------------------------------------------------------
+
+  for (int i = 0; i < aliases_.size(); i++) {
+    map<string, string> vars;
+    vars["classname"] = descriptor_->name();
+    vars["name"] = aliases_[i].value->name();
+    vars["canonical_name"] = aliases_[i].canonical_value->name();
+    printer->Print(vars,
+      "public static final $classname$ $name$ = $canonical_name$;\n");
+  }
+
+  // -----------------------------------------------------------------
+
+  printer->Print(
+    "\n"
+    "public final int getNumber() { return value; }\n"
+    "\n"
+    "public static $classname$ valueOf(int value) {\n"
+    "  switch (value) {\n",
+    "classname", descriptor_->name());
+  printer->Indent();
+  printer->Indent();
+
+  for (int i = 0; i < canonical_values_.size(); i++) {
+    printer->Print(
+      "case $number$: return $name$;\n",
+      "name", canonical_values_[i]->name(),
+      "number", SimpleItoa(canonical_values_[i]->number()));
+  }
+
+  printer->Outdent();
+  printer->Outdent();
+  printer->Print(
+    "    default: return null;\n"
+    "  }\n"
+    "}\n"
+    "\n");
+
+  // -----------------------------------------------------------------
+  // Reflection
+
+  printer->Print(
+    "public final com.google.protobuf.Descriptors.EnumValueDescriptor\n"
+    "    getValueDescriptor() {\n"
+    "  return getDescriptor().getValues().get(index);\n"
+    "}\n"
+    "public final com.google.protobuf.Descriptors.EnumDescriptor\n"
+    "    getDescriptorForType() {\n"
+    "  return getDescriptor();\n"
+    "}\n"
+    "public static final com.google.protobuf.Descriptors.EnumDescriptor\n"
+    "    getDescriptor() {\n");
+
+  // TODO(kenton):  Cache statically?  Note that we can't access descriptors
+  //   at module init time because it wouldn't work with descriptor.proto, but
+  //   we can cache the value the first time getDescriptor() is called.
+  if (descriptor_->containing_type() == NULL) {
+    printer->Print(
+      "  return $file$.getDescriptor().getEnumTypes().get($index$);\n",
+      "file", ClassName(descriptor_->file()),
+      "index", SimpleItoa(descriptor_->index()));
+  } else {
+    printer->Print(
+      "  return $parent$.getDescriptor().getEnumTypes().get($index$);\n",
+      "parent", ClassName(descriptor_->containing_type()),
+      "index", SimpleItoa(descriptor_->index()));
+  }
+
+  printer->Print(
+    "}\n"
+    "\n"
+    "private static final $classname$[] VALUES = {\n"
+    "  ",
+    "classname", descriptor_->name());
+
+  for (int i = 0; i < descriptor_->value_count(); i++) {
+    printer->Print("$name$, ",
+      "name", descriptor_->value(i)->name());
+  }
+
+  printer->Print(
+    "\n"
+    "};\n"
+    "public static $classname$ valueOf(\n"
+    "    com.google.protobuf.Descriptors.EnumValueDescriptor desc) {\n"
+    "  if (desc.getType() != getDescriptor()) {\n"
+    "    throw new java.lang.IllegalArgumentException(\n"
+    "      \"EnumValueDescriptor is not for this type.\");\n"
+    "  }\n"
+    "  return VALUES[desc.getIndex()];\n"
+    "}\n",
+    "classname", descriptor_->name());
+
+  // -----------------------------------------------------------------
+
+  printer->Print(
+    "private final int index;\n"
+    "private final int value;\n"
+    "private $classname$(int index, int value) {\n"
+    "  this.index = index;\n"
+    "  this.value = value;\n"
+    "}\n",
+    "classname", descriptor_->name());
+
+  printer->Outdent();
+  printer->Print("}\n\n");
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_enum.h b/src/google/protobuf/compiler/java/java_enum.h
new file mode 100644
index 00000000..f0a16036
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_enum.h
@@ -0,0 +1,70 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_ENUM_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_ENUM_H__
+
+#include <string>
+#include <vector>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class EnumGenerator {
+ public:
+  explicit EnumGenerator(const EnumDescriptor* descriptor);
+  ~EnumGenerator();
+
+  void Generate(io::Printer* printer);
+
+ private:
+  const EnumDescriptor* descriptor_;
+
+  // The proto language allows multiple enum constants to have the same numeric
+  // value.  Java, however, does not allow multiple enum constants to be
+  // considered equivalent.  We treat the first defined constant for any
+  // given numeric value as "canonical" and the rest as aliases of that
+  // canonical value.
+  vector<const EnumValueDescriptor*> canonical_values_;
+
+  struct Alias {
+    const EnumValueDescriptor* value;
+    const EnumValueDescriptor* canonical_value;
+  };
+  vector<Alias> aliases_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(EnumGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_ENUM_H__
diff --git a/src/google/protobuf/compiler/java/java_enum_field.cc b/src/google/protobuf/compiler/java/java_enum_field.cc
new file mode 100644
index 00000000..cb80c4b8
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_enum_field.cc
@@ -0,0 +1,264 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <map>
+#include <string>
+
+#include <google/protobuf/compiler/java/java_enum_field.h>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+namespace {
+
+// TODO(kenton):  Factor out a "SetCommonFieldVariables()" to get rid of
+//   repeat code between this and the other field types.
+void SetEnumVariables(const FieldDescriptor* descriptor,
+                      map<string, string>* variables) {
+  const EnumValueDescriptor* default_value;
+  default_value = descriptor->default_value_enum();
+
+  string type = ClassName(descriptor->enum_type());
+
+  (*variables)["name"] =
+    UnderscoresToCamelCase(descriptor);
+  (*variables)["capitalized_name"] =
+    UnderscoresToCapitalizedCamelCase(descriptor);
+  (*variables)["number"] = SimpleItoa(descriptor->number());
+  (*variables)["type"] = type;
+  (*variables)["default"] = type + "." + default_value->name();
+}
+
+}  // namespace
+
+// ===================================================================
+
+EnumFieldGenerator::
+EnumFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetEnumVariables(descriptor, &variables_);
+}
+
+EnumFieldGenerator::~EnumFieldGenerator() {}
+
+void EnumFieldGenerator::
+GenerateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "private boolean has$capitalized_name$;\n"
+    "private $type$ $name$_ = $default$;\n"
+    "public boolean has$capitalized_name$() { return has$capitalized_name$; }\n"
+    "public $type$ get$capitalized_name$() { return $name$_; }\n");
+}
+
+void EnumFieldGenerator::
+GenerateBuilderMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "public boolean has$capitalized_name$() {\n"
+    "  return result.has$capitalized_name$();\n"
+    "}\n"
+    "public $type$ get$capitalized_name$() {\n"
+    "  return result.get$capitalized_name$();\n"
+    "}\n"
+    "public Builder set$capitalized_name$($type$ value) {\n"
+    "  result.has$capitalized_name$ = true;\n"
+    "  result.$name$_ = value;\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder clear$capitalized_name$() {\n"
+    "  result.has$capitalized_name$ = false;\n"
+    "  result.$name$_ = $default$;\n"
+    "  return this;\n"
+    "}\n");
+}
+
+void EnumFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (other.has$capitalized_name$()) {\n"
+    "  set$capitalized_name$(other.get$capitalized_name$());\n"
+    "}\n");
+}
+
+void EnumFieldGenerator::
+GenerateBuildingCode(io::Printer* printer) const {
+  // Nothing to do here for enum types.
+}
+
+void EnumFieldGenerator::
+GenerateParsingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "int rawValue = input.readEnum();\n"
+    "$type$ value = $type$.valueOf(rawValue);\n"
+    "if (value == null) {\n"
+    "  unknownFields.mergeVarintField($number$, rawValue);\n"
+    "} else {\n"
+    "  set$capitalized_name$(value);\n"
+    "}\n");
+}
+
+void EnumFieldGenerator::
+GenerateSerializationCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (has$capitalized_name$()) {\n"
+    "  output.writeEnum($number$, get$capitalized_name$().getNumber());\n"
+    "}\n");
+}
+
+void EnumFieldGenerator::
+GenerateSerializedSizeCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (has$capitalized_name$()) {\n"
+    "  size += com.google.protobuf.CodedOutputStream\n"
+    "    .computeEnumSize($number$, get$capitalized_name$().getNumber());\n"
+    "}\n");
+}
+
+string EnumFieldGenerator::GetBoxedType() const {
+  return ClassName(descriptor_->enum_type());
+}
+
+// ===================================================================
+
+RepeatedEnumFieldGenerator::
+RepeatedEnumFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetEnumVariables(descriptor, &variables_);
+}
+
+RepeatedEnumFieldGenerator::~RepeatedEnumFieldGenerator() {}
+
+void RepeatedEnumFieldGenerator::
+GenerateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "private java.util.List<$type$> $name$_ =\n"
+    "  java.util.Collections.emptyList();\n"
+    "public java.util.List<$type$> get$capitalized_name$List() {\n"
+    "  return $name$_;\n"   // note:  unmodifiable list
+    "}\n"
+    "public int get$capitalized_name$Count() { return $name$_.size(); }\n"
+    "public $type$ get$capitalized_name$(int index) {\n"
+    "  return $name$_.get(index);\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateBuilderMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    // Note:  We return an unmodifiable list because otherwise the caller
+    //   could hold on to the returned list and modify it after the message
+    //   has been built, thus mutating the message which is supposed to be
+    //   immutable.
+    "public java.util.List<$type$> get$capitalized_name$List() {\n"
+    "  return java.util.Collections.unmodifiableList(result.$name$_);\n"
+    "}\n"
+    "public int get$capitalized_name$Count() {\n"
+    "  return result.get$capitalized_name$Count();\n"
+    "}\n"
+    "public $type$ get$capitalized_name$(int index) {\n"
+    "  return result.get$capitalized_name$(index);\n"
+    "}\n"
+    "public Builder set$capitalized_name$(int index, $type$ value) {\n"
+    "  result.$name$_.set(index, value);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder add$capitalized_name$($type$ value) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$type$>();\n"
+    "  }\n"
+    "  result.$name$_.add(value);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder addAll$capitalized_name$(\n"
+    "    java.lang.Iterable<? extends $type$> values) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$type$>();\n"
+    "  }\n"
+    "  super.addAll(values, result.$name$_);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder clear$capitalized_name$() {\n"
+    "  result.$name$_ = java.util.Collections.emptyList();\n"
+    "  return this;\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (!other.$name$_.isEmpty()) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$type$>();\n"
+    "  }\n"
+    "  result.$name$_.addAll(other.$name$_);\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateBuildingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (result.$name$_ != java.util.Collections.EMPTY_LIST) {\n"
+    "  result.$name$_ =\n"
+    "    java.util.Collections.unmodifiableList(result.$name$_);\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateParsingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "int rawValue = input.readEnum();\n"
+    "$type$ value = $type$.valueOf(rawValue);\n"
+    "if (value == null) {\n"
+    "  unknownFields.mergeVarintField($number$, rawValue);\n"
+    "} else {\n"
+    "  add$capitalized_name$(value);\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateSerializationCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "for ($type$ element : get$capitalized_name$List()) {\n"
+    "  output.writeEnum($number$, element.getNumber());\n"
+    "}\n");
+}
+
+void RepeatedEnumFieldGenerator::
+GenerateSerializedSizeCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "for ($type$ element : get$capitalized_name$List()) {\n"
+    "  size += com.google.protobuf.CodedOutputStream\n"
+    "    .computeEnumSize($number$, element.getNumber());\n"
+    "}\n");
+}
+
+string RepeatedEnumFieldGenerator::GetBoxedType() const {
+  return ClassName(descriptor_->enum_type());
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_enum_field.h b/src/google/protobuf/compiler/java/java_enum_field.h
new file mode 100644
index 00000000..473ba617
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_enum_field.h
@@ -0,0 +1,84 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_ENUM_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_ENUM_FIELD_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/compiler/java/java_field.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class EnumFieldGenerator : public FieldGenerator {
+ public:
+  explicit EnumFieldGenerator(const FieldDescriptor* descriptor);
+  ~EnumFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GenerateMembers(io::Printer* printer) const;
+  void GenerateBuilderMembers(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateBuildingCode(io::Printer* printer) const;
+  void GenerateParsingCode(io::Printer* printer) const;
+  void GenerateSerializationCode(io::Printer* printer) const;
+  void GenerateSerializedSizeCode(io::Printer* printer) const;
+
+  string GetBoxedType() const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(EnumFieldGenerator);
+};
+
+class RepeatedEnumFieldGenerator : public FieldGenerator {
+ public:
+  explicit RepeatedEnumFieldGenerator(const FieldDescriptor* descriptor);
+  ~RepeatedEnumFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GenerateMembers(io::Printer* printer) const;
+  void GenerateBuilderMembers(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateBuildingCode(io::Printer* printer) const;
+  void GenerateParsingCode(io::Printer* printer) const;
+  void GenerateSerializationCode(io::Printer* printer) const;
+  void GenerateSerializedSizeCode(io::Printer* printer) const;
+
+  string GetBoxedType() const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(RepeatedEnumFieldGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_ENUM_FIELD_H__
diff --git a/src/google/protobuf/compiler/java/java_extension.cc b/src/google/protobuf/compiler/java/java_extension.cc
new file mode 100644
index 00000000..1b637fab
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_extension.cc
@@ -0,0 +1,82 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/java/java_extension.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/io/printer.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+ExtensionGenerator::ExtensionGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {}
+
+ExtensionGenerator::~ExtensionGenerator() {}
+
+void ExtensionGenerator::Generate(io::Printer* printer) {
+  map<string, string> vars;
+  vars["name"] = UnderscoresToCamelCase(descriptor_);
+  vars["containing_type"] = ClassName(descriptor_->containing_type());
+  vars["index"] = SimpleItoa(descriptor_->index());
+
+  JavaType java_type = GetJavaType(descriptor_);
+  string singular_type;
+  switch (java_type) {
+    case JAVATYPE_MESSAGE:
+      vars["type"] = ClassName(descriptor_->message_type());
+      break;
+    case JAVATYPE_ENUM:
+      vars["type"] = ClassName(descriptor_->enum_type());
+      break;
+    default:
+      vars["type"] = BoxedPrimitiveTypeName(java_type);
+      break;
+  }
+
+  if (descriptor_->is_repeated()) {
+    printer->Print(vars,
+      "public static final\n"
+      "  com.google.protobuf.GeneratedMessage.GeneratedExtension<\n"
+      "    $containing_type$,\n"
+      "    java.util.List<$type$>> $name$ =\n"
+      "      com.google.protobuf.GeneratedMessage\n"
+      "        .newRepeatedGeneratedExtension(\n"
+      "          getDescriptor().getExtensions().get($index$),\n"
+      "          $type$.class);\n");
+  } else {
+    printer->Print(vars,
+      "public static final\n"
+      "  com.google.protobuf.GeneratedMessage.GeneratedExtension<\n"
+      "    $containing_type$,\n"
+      "    $type$> $name$ =\n"
+      "      com.google.protobuf.GeneratedMessage.newGeneratedExtension(\n"
+      "        getDescriptor().getExtensions().get($index$),\n"
+      "        $type$.class);\n");
+  }
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_extension.h b/src/google/protobuf/compiler/java/java_extension.h
new file mode 100644
index 00000000..9088fecd
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_extension.h
@@ -0,0 +1,58 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_EXTENSION_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_EXTENSION_H__
+
+#include <google/protobuf/stubs/common.h>
+
+namespace google {
+namespace protobuf {
+  class FieldDescriptor;       // descriptor.h
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+// Generates code for an extension, which may be within the scope of some
+// message or may be at file scope.  This is much simpler than FieldGenerator
+// since extensions are just simple identifiers with interesting types.
+class ExtensionGenerator {
+ public:
+  explicit ExtensionGenerator(const FieldDescriptor* descriptor);
+  ~ExtensionGenerator();
+
+  void Generate(io::Printer* printer);
+
+ private:
+  const FieldDescriptor* descriptor_;
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(ExtensionGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_MESSAGE_H__
diff --git a/src/google/protobuf/compiler/java/java_field.cc b/src/google/protobuf/compiler/java/java_field.cc
new file mode 100644
index 00000000..b7197cac
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_field.cc
@@ -0,0 +1,88 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/java/java_field.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/compiler/java/java_primitive_field.h>
+#include <google/protobuf/compiler/java/java_enum_field.h>
+#include <google/protobuf/compiler/java/java_message_field.h>
+#include <google/protobuf/stubs/common.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+FieldGenerator::~FieldGenerator() {}
+
+FieldGeneratorMap::FieldGeneratorMap(const Descriptor* descriptor)
+  : descriptor_(descriptor),
+    field_generators_(
+      new scoped_ptr<FieldGenerator>[descriptor->field_count()]),
+    extension_generators_(
+      new scoped_ptr<FieldGenerator>[descriptor->extension_count()]) {
+
+  // Construct all the FieldGenerators.
+  for (int i = 0; i < descriptor->field_count(); i++) {
+    field_generators_[i].reset(MakeGenerator(descriptor->field(i)));
+  }
+  for (int i = 0; i < descriptor->extension_count(); i++) {
+    extension_generators_[i].reset(MakeGenerator(descriptor->extension(i)));
+  }
+}
+
+FieldGenerator* FieldGeneratorMap::MakeGenerator(const FieldDescriptor* field) {
+  if (field->is_repeated()) {
+    switch (GetJavaType(field)) {
+      case JAVATYPE_MESSAGE:
+        return new RepeatedMessageFieldGenerator(field);
+      case JAVATYPE_ENUM:
+        return new RepeatedEnumFieldGenerator(field);
+      default:
+        return new RepeatedPrimitiveFieldGenerator(field);
+    }
+  } else {
+    switch (GetJavaType(field)) {
+      case JAVATYPE_MESSAGE:
+        return new MessageFieldGenerator(field);
+      case JAVATYPE_ENUM:
+        return new EnumFieldGenerator(field);
+      default:
+        return new PrimitiveFieldGenerator(field);
+    }
+  }
+}
+
+FieldGeneratorMap::~FieldGeneratorMap() {}
+
+const FieldGenerator& FieldGeneratorMap::get(
+    const FieldDescriptor* field) const {
+  GOOGLE_CHECK_EQ(field->containing_type(), descriptor_);
+  return *field_generators_[field->index()];
+}
+
+const FieldGenerator& FieldGeneratorMap::get_extension(int index) const {
+  return *extension_generators_[index];
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_field.h b/src/google/protobuf/compiler/java/java_field.h
new file mode 100644
index 00000000..ef47c735
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_field.h
@@ -0,0 +1,82 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_FIELD_H__
+
+#include <string>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class FieldGenerator {
+ public:
+  FieldGenerator() {}
+  virtual ~FieldGenerator();
+
+  virtual void GenerateMembers(io::Printer* printer) const = 0;
+  virtual void GenerateBuilderMembers(io::Printer* printer) const = 0;
+  virtual void GenerateMergingCode(io::Printer* printer) const = 0;
+  virtual void GenerateBuildingCode(io::Printer* printer) const = 0;
+  virtual void GenerateParsingCode(io::Printer* printer) const = 0;
+  virtual void GenerateSerializationCode(io::Printer* printer) const = 0;
+  virtual void GenerateSerializedSizeCode(io::Printer* printer) const = 0;
+
+  virtual string GetBoxedType() const = 0;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(FieldGenerator);
+};
+
+// Convenience class which constructs FieldGenerators for a Descriptor.
+class FieldGeneratorMap {
+ public:
+  explicit FieldGeneratorMap(const Descriptor* descriptor);
+  ~FieldGeneratorMap();
+
+  const FieldGenerator& get(const FieldDescriptor* field) const;
+  const FieldGenerator& get_extension(int index) const;
+
+ private:
+  const Descriptor* descriptor_;
+  scoped_array<scoped_ptr<FieldGenerator> > field_generators_;
+  scoped_array<scoped_ptr<FieldGenerator> > extension_generators_;
+
+  static FieldGenerator* MakeGenerator(const FieldDescriptor* field);
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(FieldGeneratorMap);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_FIELD_H__
diff --git a/src/google/protobuf/compiler/java/java_file.cc b/src/google/protobuf/compiler/java/java_file.cc
new file mode 100644
index 00000000..e7e8618d
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_file.cc
@@ -0,0 +1,249 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/java/java_file.h>
+#include <google/protobuf/compiler/java/java_enum.h>
+#include <google/protobuf/compiler/java/java_service.h>
+#include <google/protobuf/compiler/java/java_extension.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/compiler/java/java_message.h>
+#include <google/protobuf/compiler/code_generator.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/io/zero_copy_stream.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+FileGenerator::FileGenerator(const FileDescriptor* file)
+  : file_(file),
+    java_package_(FileJavaPackage(file)),
+    classname_(FileClassName(file)) {}
+
+FileGenerator::~FileGenerator() {}
+
+bool FileGenerator::Validate(string* error) {
+  // Check that no class name matches the file's class name.  This is a common
+  // problem that leads to Java compile errors that can be hard to understand.
+  // It's especially bad when using the java_multiple_files, since we would
+  // end up overwriting the outer class with one of the inner ones.
+
+  bool found_conflict = false;
+  for (int i = 0; i < file_->enum_type_count() && !found_conflict; i++) {
+    if (file_->enum_type(i)->name() == classname_) {
+      found_conflict = true;
+    }
+  }
+  for (int i = 0; i < file_->message_type_count() && !found_conflict; i++) {
+    if (file_->message_type(i)->name() == classname_) {
+      found_conflict = true;
+    }
+  }
+  for (int i = 0; i < file_->service_count() && !found_conflict; i++) {
+    if (file_->service(i)->name() == classname_) {
+      found_conflict = true;
+    }
+  }
+
+  if (found_conflict) {
+    error->assign(file_->name());
+    error->append(
+      ": Cannot generate Java output because the file's outer class name, \"");
+    error->append(classname_);
+    error->append(
+      "\", matches the name of one of the types declared inside it.  "
+      "Please either rename the type or use the java_outer_classname "
+      "option to specify a different outer class name for the .proto file.");
+    return false;
+  }
+
+  return true;
+}
+
+void FileGenerator::Generate(io::Printer* printer) {
+  // We don't import anything because we refer to all classes by their
+  // fully-qualified names in the generated source.
+  printer->Print(
+    "// Generated by the protocol buffer compiler.  DO NOT EDIT!\n"
+    "\n");
+  if (!java_package_.empty()) {
+    printer->Print(
+      "package $package$;\n"
+      "\n",
+      "package", java_package_);
+  }
+  printer->Print(
+    "public final class $classname$ {\n"
+    "  private $classname$() {}\n",
+    "classname", classname_);
+  printer->Indent();
+
+  // -----------------------------------------------------------------
+
+  // Embed the descriptor.  We simply serialize the entire FileDescriptorProto
+  // and embed it as a string literal, which is parsed and built into real
+  // descriptors at initialization time.  We unfortunately have to put it in
+  // a string literal, not a byte array, because apparently using a literal
+  // byte array causes the Java compiler to generate *instructions* to
+  // initialize each and every byte of the array, e.g. as if you typed:
+  //   b[0] = 123; b[1] = 456; b[2] = 789;
+  // This makes huge bytecode files and can easily hit the compiler's internal
+  // code size limits (error "code to large").  String literals are apparently
+  // embedded raw, which is what we want.
+  FileDescriptorProto file_proto;
+  file_->CopyTo(&file_proto);
+  string file_data;
+  file_proto.SerializeToString(&file_data);
+
+  printer->Print(
+    "public static com.google.protobuf.Descriptors.FileDescriptor\n"
+    "    getDescriptor() {\n"
+    "  return descriptor;\n"
+    "}\n"
+    "private static final com.google.protobuf.Descriptors.FileDescriptor\n"
+    "    descriptor = buildDescriptor();\n"
+    "private static\n"
+    "    com.google.protobuf.Descriptors.FileDescriptor\n"
+    "    buildDescriptor() {\n"
+    "  java.lang.String descriptorData =\n");
+  printer->Indent();
+  printer->Indent();
+
+  // Only write 40 bytes per line.
+  static const int kBytesPerLine = 40;
+  for (int i = 0; i < file_data.size(); i += kBytesPerLine) {
+    if (i > 0) printer->Print(" +\n");
+    printer->Print("\"$data$\"",
+      "data", CEscape(file_data.substr(i, kBytesPerLine)));
+  }
+  printer->Print(";\n");
+
+  printer->Outdent();
+  printer->Print(
+    "try {\n"
+    "  return com.google.protobuf.Descriptors.FileDescriptor\n"
+    "    .internalBuildGeneratedFileFrom(descriptorData,\n"
+    "      new com.google.protobuf.Descriptors.FileDescriptor[] {\n");
+
+  for (int i = 0; i < file_->dependency_count(); i++) {
+    printer->Print(
+      "        $dependency$.getDescriptor(),\n",
+      "dependency", ClassName(file_->dependency(i)));
+  }
+
+  printer->Print(
+    "      });\n"
+    "} catch (Exception e) {\n"
+    "  throw new RuntimeException(\n"
+    "    \"Failed to parse protocol buffer descriptor for \" +\n"
+    "    \"\\\"$filename$\\\".\", e);\n"
+    "}\n",
+    "filename", file_->name());
+
+  printer->Outdent();
+  printer->Print(
+    "}\n"
+    "\n");
+
+  // -----------------------------------------------------------------
+
+  if (!file_->options().java_multiple_files()) {
+    for (int i = 0; i < file_->enum_type_count(); i++) {
+      EnumGenerator(file_->enum_type(i)).Generate(printer);
+    }
+    for (int i = 0; i < file_->message_type_count(); i++) {
+      MessageGenerator(file_->message_type(i)).Generate(printer);
+    }
+    for (int i = 0; i < file_->service_count(); i++) {
+      ServiceGenerator(file_->service(i)).Generate(printer);
+    }
+  }
+
+  // Extensions must be generated in the outer class since they are values,
+  // not classes.
+  for (int i = 0; i < file_->extension_count(); i++) {
+    ExtensionGenerator(file_->extension(i)).Generate(printer);
+  }
+
+  // Static variables.
+  for (int i = 0; i < file_->message_type_count(); i++) {
+    // TODO(kenton):  Reuse MessageGenerator objects?
+    MessageGenerator(file_->message_type(i)).GenerateStaticVariables(printer);
+  }
+
+  printer->Outdent();
+  printer->Print("}\n");
+}
+
+template<typename GeneratorClass, typename DescriptorClass>
+static void GenerateSibling(const string& package_dir,
+                            const string& java_package,
+                            const DescriptorClass* descriptor,
+                            OutputDirectory* output_directory,
+                            vector<string>* file_list) {
+  string filename = package_dir + descriptor->name() + ".java";
+  file_list->push_back(filename);
+
+  scoped_ptr<io::ZeroCopyOutputStream> output(
+    output_directory->Open(filename));
+  io::Printer printer(output.get(), '$');
+
+  printer.Print(
+    "// Generated by the protocol buffer compiler.  DO NOT EDIT!\n"
+    "\n");
+  if (!java_package.empty()) {
+    printer.Print(
+      "package $package$;\n"
+      "\n",
+      "package", java_package);
+  }
+
+  GeneratorClass(descriptor).Generate(&printer);
+}
+
+void FileGenerator::GenerateSiblings(const string& package_dir,
+                                     OutputDirectory* output_directory,
+                                     vector<string>* file_list) {
+  if (file_->options().java_multiple_files()) {
+    for (int i = 0; i < file_->enum_type_count(); i++) {
+      GenerateSibling<EnumGenerator>(package_dir, java_package_,
+                                     file_->enum_type(i),
+                                     output_directory, file_list);
+    }
+    for (int i = 0; i < file_->message_type_count(); i++) {
+      GenerateSibling<MessageGenerator>(package_dir, java_package_,
+                                        file_->message_type(i),
+                                        output_directory, file_list);
+    }
+    for (int i = 0; i < file_->service_count(); i++) {
+      GenerateSibling<ServiceGenerator>(package_dir, java_package_,
+                                        file_->service(i),
+                                        output_directory, file_list);
+    }
+  }
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_file.h b/src/google/protobuf/compiler/java/java_file.h
new file mode 100644
index 00000000..9eeec253
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_file.h
@@ -0,0 +1,78 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_FILE_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_FILE_H__
+
+#include <string>
+#include <vector>
+#include <google/protobuf/stubs/common.h>
+
+namespace google {
+namespace protobuf {
+  class FileDescriptor;        // descriptor.h
+  namespace io {
+    class Printer;             // printer.h
+  }
+  namespace compiler {
+    class OutputDirectory;     // code_generator.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class FileGenerator {
+ public:
+  explicit FileGenerator(const FileDescriptor* file);
+  ~FileGenerator();
+
+  // Checks for problems that would otherwise lead to cryptic compile errors.
+  // Returns true if there are no problems, or writes an error description to
+  // the given string and returns false otherwise.
+  bool Validate(string* error);
+
+  void Generate(io::Printer* printer);
+
+  // If we aren't putting everything into one file, this will write all the
+  // files other than the outer file (i.e. one for each message, enum, and
+  // service type).
+  void GenerateSiblings(const string& package_dir,
+                        OutputDirectory* output_directory,
+                        vector<string>* file_list);
+
+  const string& java_package() { return java_package_; }
+  const string& classname()    { return classname_;    }
+
+ private:
+  const FileDescriptor* file_;
+  string java_package_;
+  string classname_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(FileGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_FILE_H__
diff --git a/src/google/protobuf/compiler/java/java_generator.cc b/src/google/protobuf/compiler/java/java_generator.cc
new file mode 100644
index 00000000..89a32da4
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_generator.cc
@@ -0,0 +1,133 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/java/java_generator.h>
+#include <google/protobuf/compiler/java/java_file.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/io/zero_copy_stream.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+namespace {
+
+// Parses a set of comma-delimited name/value pairs, e.g.:
+//   "foo=bar,baz,qux=corge"
+// parses to the pairs:
+//   ("foo", "bar"), ("baz", ""), ("qux", "corge")
+void ParseOptions(const string& text, vector<pair<string, string> >* output) {
+  vector<string> parts;
+  SplitStringUsing(text, ",", &parts);
+
+  for (int i = 0; i < parts.size(); i++) {
+    string::size_type equals_pos = parts[i].find_first_of('=');
+    pair<string, string> value;
+    if (equals_pos == string::npos) {
+      value.first = parts[i];
+      value.second = "";
+    } else {
+      value.first = parts[i].substr(0, equals_pos);
+      value.second = parts[i].substr(equals_pos + 1);
+    }
+    output->push_back(value);
+  }
+}
+
+}  // namespace
+
+JavaGenerator::JavaGenerator() {}
+JavaGenerator::~JavaGenerator() {}
+
+bool JavaGenerator::Generate(const FileDescriptor* file,
+                             const string& parameter,
+                             OutputDirectory* output_directory,
+                             string* error) const {
+  vector<pair<string, string> > options;
+  ParseOptions(parameter, &options);
+
+  // -----------------------------------------------------------------
+  // parse generator options
+
+  // Name a file where we will write a list of generated file names, one
+  // per line.
+  string output_list_file;
+
+  for (int i = 0; i < options.size(); i++) {
+    if (options[i].first == "output_list_file") {
+      output_list_file = options[i].second;
+    } else {
+      *error = "Unknown generator option: " + options[i].first;
+      return false;
+    }
+  }
+
+
+  // -----------------------------------------------------------------
+
+
+  FileGenerator file_generator(file);
+  if (!file_generator.Validate(error)) {
+    return false;
+  }
+
+  string package_dir =
+    StringReplace(file_generator.java_package(), ".", "/", true);
+  if (!package_dir.empty()) package_dir += "/";
+
+  vector<string> all_files;
+
+  string java_filename = package_dir;
+  java_filename += file_generator.classname();
+  java_filename += ".java";
+  all_files.push_back(java_filename);
+
+  // Generate main java file.
+  scoped_ptr<io::ZeroCopyOutputStream> output(
+    output_directory->Open(java_filename));
+  io::Printer printer(output.get(), '$');
+  file_generator.Generate(&printer);
+
+  // Generate sibling files.
+  file_generator.GenerateSiblings(package_dir, output_directory, &all_files);
+
+  // Generate output list if requested.
+  if (!output_list_file.empty()) {
+    // Generate output list.  This is just a simple text file placed in a
+    // deterministic location which lists the .java files being generated.
+    scoped_ptr<io::ZeroCopyOutputStream> srclist_raw_output(
+      output_directory->Open(output_list_file));
+    io::Printer srclist_printer(srclist_raw_output.get(), '$');
+    for (int i = 0; i < all_files.size(); i++) {
+      srclist_printer.Print("$filename$\n", "filename", all_files[i]);
+    }
+  }
+
+  return true;
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_generator.h b/src/google/protobuf/compiler/java/java_generator.h
new file mode 100644
index 00000000..11886aa0
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_generator.h
@@ -0,0 +1,58 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// Generates Java code for a given .proto file.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_GENERATOR_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_GENERATOR_H__
+
+#include <string>
+#include <google/protobuf/compiler/code_generator.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+// CodeGenerator implementation which generates Java code.  If you create your
+// own protocol compiler binary and you want it to support Java output, you
+// can do so by registering an instance of this CodeGenerator with the
+// CommandLineInterface in your main() function.
+class LIBPROTOC_EXPORT JavaGenerator : public CodeGenerator {
+ public:
+  JavaGenerator();
+  ~JavaGenerator();
+
+  // implements CodeGenerator ----------------------------------------
+  bool Generate(const FileDescriptor* file,
+                const string& parameter,
+                OutputDirectory* output_directory,
+                string* error) const;
+
+ private:
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(JavaGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_GENERATOR_H__
diff --git a/src/google/protobuf/compiler/java/java_helpers.cc b/src/google/protobuf/compiler/java/java_helpers.cc
new file mode 100644
index 00000000..4ba82c2a
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_helpers.cc
@@ -0,0 +1,229 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <vector>
+
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+const char kThickSeparator[] =
+  "// ===================================================================\n";
+const char kThinSeparator[] =
+  "// -------------------------------------------------------------------\n";
+
+namespace {
+
+const char* kDefaultPackage = "";
+
+const string& FieldName(const FieldDescriptor* field) {
+  // Groups are hacky:  The name of the field is just the lower-cased name
+  // of the group type.  In Java, though, we would like to retain the original
+  // capitalization of the type name.
+  if (field->type() == FieldDescriptor::TYPE_GROUP) {
+    return field->message_type()->name();
+  } else {
+    return field->name();
+  }
+}
+
+string UnderscoresToCamelCaseImpl(const string& input, bool cap_next_letter) {
+  string result;
+  // Note:  I distrust ctype.h due to locales.
+  for (int i = 0; i < input.size(); i++) {
+    if ('a' <= input[i] && input[i] <= 'z') {
+      if (cap_next_letter) {
+        result += input[i] + ('A' - 'a');
+      } else {
+        result += input[i];
+      }
+      cap_next_letter = false;
+    } else if ('A' <= input[i] && input[i] <= 'Z') {
+      if (i == 0 && !cap_next_letter) {
+        // Force first letter to lower-case unless explicitly told to
+        // capitalize it.
+        result += input[i] + ('a' - 'A');
+      } else {
+        // Capital letters after the first are left as-is.
+        result += input[i];
+      }
+      cap_next_letter = false;
+    } else if ('0' <= input[i] && input[i] <= '9') {
+      result += input[i];
+      cap_next_letter = true;
+    } else {
+      cap_next_letter = true;
+    }
+  }
+  return result;
+}
+
+}  // namespace
+
+string UnderscoresToCamelCase(const FieldDescriptor* field) {
+  return UnderscoresToCamelCaseImpl(FieldName(field), false);
+}
+
+string UnderscoresToCapitalizedCamelCase(const FieldDescriptor* field) {
+  return UnderscoresToCamelCaseImpl(FieldName(field), true);
+}
+
+string UnderscoresToCamelCase(const MethodDescriptor* method) {
+  return UnderscoresToCamelCaseImpl(method->name(), false);
+}
+
+string StripProto(const string& filename) {
+  if (HasSuffixString(filename, ".protodevel")) {
+    return StripSuffixString(filename, ".protodevel");
+  } else {
+    return StripSuffixString(filename, ".proto");
+  }
+}
+
+string FileClassName(const FileDescriptor* file) {
+  if (file->options().has_java_outer_classname()) {
+    return file->options().java_outer_classname();
+  } else {
+    string basename;
+    string::size_type last_slash = file->name().find_last_of('/');
+    if (last_slash == string::npos) {
+      basename = file->name();
+    } else {
+      basename = file->name().substr(last_slash + 1);
+    }
+    return UnderscoresToCamelCaseImpl(StripProto(basename), true);
+  }
+}
+
+string FileJavaPackage(const FileDescriptor* file) {
+  if (file->options().has_java_package()) {
+    return file->options().java_package();
+  } else {
+    string result = kDefaultPackage;
+    if (!file->package().empty()) {
+      if (!result.empty()) result += '.';
+      result += file->package();
+    }
+    return result;
+  }
+}
+
+string ToJavaName(const string& full_name, const FileDescriptor* file) {
+  string result;
+  if (file->options().java_multiple_files()) {
+    result = FileJavaPackage(file);
+  } else {
+    result = ClassName(file);
+  }
+  if (!result.empty()) {
+    result += '.';
+  }
+  if (file->package().empty()) {
+    result += full_name;
+  } else {
+    // Strip the proto package from full_name since we've replaced it with
+    // the Java package.
+    result += full_name.substr(file->package().size() + 1);
+  }
+  return result;
+}
+
+string ClassName(const FileDescriptor* descriptor) {
+  string result = FileJavaPackage(descriptor);
+  if (!result.empty()) result += '.';
+  result += FileClassName(descriptor);
+  return result;
+}
+
+JavaType GetJavaType(FieldDescriptor::Type field_type) {
+  switch (field_type) {
+    case FieldDescriptor::TYPE_INT32:
+    case FieldDescriptor::TYPE_UINT32:
+    case FieldDescriptor::TYPE_SINT32:
+    case FieldDescriptor::TYPE_FIXED32:
+    case FieldDescriptor::TYPE_SFIXED32:
+      return JAVATYPE_INT;
+
+    case FieldDescriptor::TYPE_INT64:
+    case FieldDescriptor::TYPE_UINT64:
+    case FieldDescriptor::TYPE_SINT64:
+    case FieldDescriptor::TYPE_FIXED64:
+    case FieldDescriptor::TYPE_SFIXED64:
+      return JAVATYPE_LONG;
+
+    case FieldDescriptor::TYPE_FLOAT:
+      return JAVATYPE_FLOAT;
+
+    case FieldDescriptor::TYPE_DOUBLE:
+      return JAVATYPE_DOUBLE;
+
+    case FieldDescriptor::TYPE_BOOL:
+      return JAVATYPE_BOOLEAN;
+
+    case FieldDescriptor::TYPE_STRING:
+      return JAVATYPE_STRING;
+
+    case FieldDescriptor::TYPE_BYTES:
+      return JAVATYPE_BYTES;
+
+    case FieldDescriptor::TYPE_ENUM:
+      return JAVATYPE_ENUM;
+
+    case FieldDescriptor::TYPE_GROUP:
+    case FieldDescriptor::TYPE_MESSAGE:
+      return JAVATYPE_MESSAGE;
+
+    // No default because we want the compiler to complain if any new
+    // types are added.
+  }
+
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return JAVATYPE_INT;
+}
+
+const char* BoxedPrimitiveTypeName(JavaType type) {
+  switch (type) {
+    case JAVATYPE_INT    : return "java.lang.Integer";
+    case JAVATYPE_LONG   : return "java.lang.Long";
+    case JAVATYPE_FLOAT  : return "java.lang.Float";
+    case JAVATYPE_DOUBLE : return "java.lang.Double";
+    case JAVATYPE_BOOLEAN: return "java.lang.Boolean";
+    case JAVATYPE_STRING : return "java.lang.String";
+    case JAVATYPE_BYTES  : return "com.google.protobuf.ByteString";
+    case JAVATYPE_ENUM   : return NULL;
+    case JAVATYPE_MESSAGE: return NULL;
+
+    // No default because we want the compiler to complain if any new
+    // JavaTypes are added.
+  }
+
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return NULL;
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_helpers.h b/src/google/protobuf/compiler/java/java_helpers.h
new file mode 100644
index 00000000..74376330
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_helpers.h
@@ -0,0 +1,102 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_HELPERS_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_HELPERS_H__
+
+#include <string>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+// Commonly-used separator comments.  Thick is a line of '=', thin is a line
+// of '-'.
+extern const char kThickSeparator[];
+extern const char kThinSeparator[];
+
+// Converts the field's name to camel-case, e.g. "foo_bar_baz" becomes
+// "fooBarBaz" or "FooBarBaz", respectively.
+string UnderscoresToCamelCase(const FieldDescriptor* field);
+string UnderscoresToCapitalizedCamelCase(const FieldDescriptor* field);
+
+// Similar, but for method names.  (Typically, this merely has the effect
+// of lower-casing the first letter of the name.)
+string UnderscoresToCamelCase(const MethodDescriptor* method);
+
+// Strips ".proto" or ".protodevel" from the end of a filename.
+string StripProto(const string& filename);
+
+// Gets the unqualified class name for the file.  Each .proto file becomes a
+// single Java class, with all its contents nested in that class.
+string FileClassName(const FileDescriptor* file);
+
+// Returns the file's Java package name.
+string FileJavaPackage(const FileDescriptor* file);
+
+// Converts the given fully-qualified name in the proto namespace to its
+// fully-qualified name in the Java namespace, given that it is in the given
+// file.
+string ToJavaName(const string& full_name, const FileDescriptor* file);
+
+// These return the fully-qualified class name corresponding to the given
+// descriptor.
+inline string ClassName(const Descriptor* descriptor) {
+  return ToJavaName(descriptor->full_name(), descriptor->file());
+}
+inline string ClassName(const EnumDescriptor* descriptor) {
+  return ToJavaName(descriptor->full_name(), descriptor->file());
+}
+inline string ClassName(const ServiceDescriptor* descriptor) {
+  return ToJavaName(descriptor->full_name(), descriptor->file());
+}
+string ClassName(const FileDescriptor* descriptor);
+
+enum JavaType {
+  JAVATYPE_INT,
+  JAVATYPE_LONG,
+  JAVATYPE_FLOAT,
+  JAVATYPE_DOUBLE,
+  JAVATYPE_BOOLEAN,
+  JAVATYPE_STRING,
+  JAVATYPE_BYTES,
+  JAVATYPE_ENUM,
+  JAVATYPE_MESSAGE
+};
+
+JavaType GetJavaType(FieldDescriptor::Type field_type);
+
+inline JavaType GetJavaType(const FieldDescriptor* field) {
+  return GetJavaType(field->type());
+}
+
+// Get the fully-qualified class name for a boxed primitive type, e.g.
+// "java.lang.Integer" for JAVATYPE_INT.  Returns NULL for enum and message
+// types.
+const char* BoxedPrimitiveTypeName(JavaType type);
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_HELPERS_H__
diff --git a/src/google/protobuf/compiler/java/java_message.cc b/src/google/protobuf/compiler/java/java_message.cc
new file mode 100644
index 00000000..ebba1f7b
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_message.cc
@@ -0,0 +1,722 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <algorithm>
+#include <google/protobuf/stubs/hash.h>
+#include <google/protobuf/compiler/java/java_message.h>
+#include <google/protobuf/compiler/java/java_enum.h>
+#include <google/protobuf/compiler/java/java_extension.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/io/coded_stream.h>
+#include <google/protobuf/wire_format.h>
+#include <google/protobuf/descriptor.pb.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+using internal::WireFormat;
+
+namespace {
+
+void PrintFieldComment(io::Printer* printer, const FieldDescriptor* field) {
+  // Print the field's proto-syntax definition as a comment.  We don't want to
+  // print group bodies so we cut off after the first line.
+  string def = field->DebugString();
+  printer->Print("// $def$\n",
+    "def", def.substr(0, def.find_first_of('\n')));
+}
+
+struct FieldOrderingByNumber {
+  inline bool operator()(const FieldDescriptor* a,
+                         const FieldDescriptor* b) const {
+    return a->number() < b->number();
+  }
+};
+
+struct ExtensionRangeOrdering {
+  bool operator()(const Descriptor::ExtensionRange* a,
+                  const Descriptor::ExtensionRange* b) const {
+    return a->start < b->start;
+  }
+};
+
+// Sort the fields of the given Descriptor by number into a new[]'d array
+// and return it.
+const FieldDescriptor** SortFieldsByNumber(const Descriptor* descriptor) {
+  const FieldDescriptor** fields =
+    new const FieldDescriptor*[descriptor->field_count()];
+  for (int i = 0; i < descriptor->field_count(); i++) {
+    fields[i] = descriptor->field(i);
+  }
+  sort(fields, fields + descriptor->field_count(),
+       FieldOrderingByNumber());
+  return fields;
+}
+
+// Get an identifier that uniquely identifies this type within the file.
+// This is used to declare static variables related to this type at the
+// outermost file scope.
+string UniqueFileScopeIdentifier(const Descriptor* descriptor) {
+  return "static_" + StringReplace(descriptor->full_name(), ".", "_", true);
+}
+
+// Returns true if the message type has any required fields.  If it doesn't,
+// we can optimize out calls to its isInitialized() method.
+//
+// already_seen is used to avoid checking the same type multiple times
+// (and also to protect against recursion).
+static bool HasRequiredFields(
+    const Descriptor* type,
+    hash_set<const Descriptor*>* already_seen) {
+  if (already_seen->count(type) > 0) {
+    // The type is already in cache.  This means that either:
+    // a. The type has no required fields.
+    // b. We are in the midst of checking if the type has required fields,
+    //    somewhere up the stack.  In this case, we know that if the type
+    //    has any required fields, they'll be found when we return to it,
+    //    and the whole call to HasRequiredFields() will return true.
+    //    Therefore, we don't have to check if this type has required fields
+    //    here.
+    return false;
+  }
+  already_seen->insert(type);
+
+  // If the type has extensions, an extension with message type could contain
+  // required fields, so we have to be conservative and assume such an
+  // extension exists.
+  if (type->extension_range_count() > 0) return true;
+
+  for (int i = 0; i < type->field_count(); i++) {
+    const FieldDescriptor* field = type->field(i);
+    if (field->is_required()) {
+      return true;
+    }
+    if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
+      if (HasRequiredFields(field->message_type(), already_seen)) {
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
+static bool HasRequiredFields(const Descriptor* type) {
+  hash_set<const Descriptor*> already_seen;
+  return HasRequiredFields(type, &already_seen);
+}
+
+}  // namespace
+
+// ===================================================================
+
+MessageGenerator::MessageGenerator(const Descriptor* descriptor)
+  : descriptor_(descriptor),
+    field_generators_(descriptor) {
+}
+
+MessageGenerator::~MessageGenerator() {}
+
+void MessageGenerator::GenerateStaticVariables(io::Printer* printer) {
+  // Because descriptor.proto (com.google.protobuf.DescriptorProtos) is
+  // used in the construction of descriptors, we have a tricky bootstrapping
+  // problem.  To help control static initialization order, we make sure all
+  // descriptors and other static data that depends on them are members of
+  // the outermost class in the file.  This way, they will be initialized in
+  // a deterministic order.
+
+  map<string, string> vars;
+  vars["identifier"] = UniqueFileScopeIdentifier(descriptor_);
+  vars["index"] = SimpleItoa(descriptor_->index());
+  vars["classname"] = ClassName(descriptor_);
+  if (descriptor_->containing_type() != NULL) {
+    vars["parent"] = UniqueFileScopeIdentifier(descriptor_->containing_type());
+  }
+  if (descriptor_->file()->options().java_multiple_files()) {
+    // We can only make these package-private since the classes that use them
+    // are in separate files.
+    vars["private"] = "";
+  } else {
+    vars["private"] = "private ";
+  }
+
+  // The descriptor for this type.
+  if (descriptor_->containing_type() == NULL) {
+    printer->Print(vars,
+      "$private$static final com.google.protobuf.Descriptors.Descriptor\n"
+      "  internal_$identifier$_descriptor =\n"
+      "    getDescriptor().getMessageTypes().get($index$);\n");
+  } else {
+    printer->Print(vars,
+      "$private$static final com.google.protobuf.Descriptors.Descriptor\n"
+      "  internal_$identifier$_descriptor =\n"
+      "    internal_$parent$_descriptor.getNestedTypes().get($index$);\n");
+  }
+
+  // And the FieldAccessorTable.
+  printer->Print(vars,
+    "$private$static\n"
+    "  com.google.protobuf.GeneratedMessage.FieldAccessorTable\n"
+    "    internal_$identifier$_fieldAccessorTable = new\n"
+    "      com.google.protobuf.GeneratedMessage.FieldAccessorTable(\n"
+    "        internal_$identifier$_descriptor,\n"
+    "        new java.lang.String[] { ");
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    printer->Print(
+      "\"$field_name$\", ",
+      "field_name",
+        UnderscoresToCapitalizedCamelCase(descriptor_->field(i)));
+  }
+  printer->Print("},\n"
+    "        $classname$.class,\n"
+    "        $classname$.Builder.class);\n",
+    "classname", ClassName(descriptor_));
+
+  // Generate static members for all nested types.
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    // TODO(kenton):  Reuse MessageGenerator objects?
+    MessageGenerator(descriptor_->nested_type(i))
+      .GenerateStaticVariables(printer);
+  }
+}
+
+void MessageGenerator::Generate(io::Printer* printer) {
+  bool is_own_file =
+    descriptor_->containing_type() == NULL &&
+    descriptor_->file()->options().java_multiple_files();
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "public $static$ final class $classname$ extends\n"
+      "    com.google.protobuf.GeneratedMessage.ExtendableMessage<\n"
+      "      $classname$> {\n",
+      "static", is_own_file ? "" : "static",
+      "classname", descriptor_->name());
+  } else {
+    printer->Print(
+      "public $static$ final class $classname$ extends\n"
+      "    com.google.protobuf.GeneratedMessage {\n",
+      "static", is_own_file ? "" : "static",
+      "classname", descriptor_->name());
+  }
+  printer->Indent();
+  printer->Print(
+    "// Use $classname$.newBuilder() to construct.\n"
+    "private $classname$() {}\n"
+    "\n"
+    "private static final $classname$ defaultInstance = new $classname$();\n"
+    "public static $classname$ getDefaultInstance() {\n"
+    "  return defaultInstance;\n"
+    "}\n"
+    "\n"
+    "public $classname$ getDefaultInstanceForType() {\n"
+    "  return defaultInstance;\n"
+    "}\n"
+    "\n",
+    "classname", descriptor_->name());
+  printer->Print(
+    "public static final com.google.protobuf.Descriptors.Descriptor\n"
+    "    getDescriptor() {\n"
+    "  return $fileclass$.internal_$identifier$_descriptor;\n"
+    "}\n"
+    "\n"
+    "protected com.google.protobuf.GeneratedMessage.FieldAccessorTable\n"
+    "    internalGetFieldAccessorTable() {\n"
+    "  return $fileclass$.internal_$identifier$_fieldAccessorTable;\n"
+    "}\n"
+    "\n",
+    "fileclass", ClassName(descriptor_->file()),
+    "identifier", UniqueFileScopeIdentifier(descriptor_));
+
+  // Nested types and extensions
+  for (int i = 0; i < descriptor_->enum_type_count(); i++) {
+    EnumGenerator(descriptor_->enum_type(i)).Generate(printer);
+  }
+
+  for (int i = 0; i < descriptor_->nested_type_count(); i++) {
+    MessageGenerator(descriptor_->nested_type(i)).Generate(printer);
+  }
+
+  for (int i = 0; i < descriptor_->extension_count(); i++) {
+    ExtensionGenerator(descriptor_->extension(i)).Generate(printer);
+  }
+
+  // Fields
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    PrintFieldComment(printer, descriptor_->field(i));
+    field_generators_.get(descriptor_->field(i)).GenerateMembers(printer);
+    printer->Print("\n");
+  }
+
+  if (descriptor_->file()->options().optimize_for() == FileOptions::SPEED) {
+    GenerateIsInitialized(printer);
+    GenerateMessageSerializationMethods(printer);
+  }
+
+  GenerateParseFromMethods(printer);
+  GenerateBuilder(printer);
+}
+
+// ===================================================================
+
+void MessageGenerator::
+GenerateMessageSerializationMethods(io::Printer* printer) {
+  scoped_array<const FieldDescriptor*> sorted_fields(
+    SortFieldsByNumber(descriptor_));
+
+  vector<const Descriptor::ExtensionRange*> sorted_extensions;
+  for (int i = 0; i < descriptor_->extension_range_count(); ++i) {
+    sorted_extensions.push_back(descriptor_->extension_range(i));
+  }
+  sort(sorted_extensions.begin(), sorted_extensions.end(),
+       ExtensionRangeOrdering());
+
+  printer->Print(
+    "public void writeTo(com.google.protobuf.CodedOutputStream output)\n"
+    "                    throws java.io.IOException {\n");
+  printer->Indent();
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "com.google.protobuf.GeneratedMessage.ExtendableMessage\n"
+      "  .ExtensionWriter extensionWriter = newExtensionWriter();\n");
+  }
+
+  // Merge the fields and the extension ranges, both sorted by field number.
+  for (int i = 0, j = 0;
+       i < descriptor_->field_count() || j < sorted_extensions.size();
+       ) {
+    if (i == descriptor_->field_count()) {
+      GenerateSerializeOneExtensionRange(printer, sorted_extensions[j++]);
+    } else if (j == sorted_extensions.size()) {
+      GenerateSerializeOneField(printer, sorted_fields[i++]);
+    } else if (sorted_fields[i]->number() < sorted_extensions[j]->start) {
+      GenerateSerializeOneField(printer, sorted_fields[i++]);
+    } else {
+      GenerateSerializeOneExtensionRange(printer, sorted_extensions[j++]);
+    }
+  }
+
+  if (descriptor_->options().message_set_wire_format()) {
+    printer->Print(
+      "getUnknownFields().writeAsMessageSetTo(output);\n");
+  } else {
+    printer->Print(
+      "getUnknownFields().writeTo(output);\n");
+  }
+
+  printer->Outdent();
+  printer->Print(
+    "}\n"
+    "\n"
+    "private int memoizedSerializedSize = -1;\n"
+    "public int getSerializedSize() {\n"
+    "  int size = memoizedSerializedSize;\n"
+    "  if (size != -1) return size;\n"
+    "\n"
+    "  size = 0;\n");
+  printer->Indent();
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    field_generators_.get(sorted_fields[i]).GenerateSerializedSizeCode(printer);
+  }
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "size += extensionsSerializedSize();\n");
+  }
+
+  if (descriptor_->options().message_set_wire_format()) {
+    printer->Print(
+      "size += getUnknownFields().getSerializedSizeAsMessageSet();\n");
+  } else {
+    printer->Print(
+      "size += getUnknownFields().getSerializedSize();\n");
+  }
+
+  printer->Outdent();
+  printer->Print(
+    "  memoizedSerializedSize = size;\n"
+    "  return size;\n"
+    "}\n"
+    "\n");
+}
+
+void MessageGenerator::
+GenerateParseFromMethods(io::Printer* printer) {
+  // Note:  These are separate from GenerateMessageSerializationMethods()
+  //   because they need to be generated even for messages that are optimized
+  //   for code size.
+  printer->Print(
+    "public static $classname$ parseFrom(\n"
+    "    com.google.protobuf.ByteString data)\n"
+    "    throws com.google.protobuf.InvalidProtocolBufferException {\n"
+    "  return newBuilder().mergeFrom(data).buildParsed();\n"
+    "}\n"
+    "public static $classname$ parseFrom(\n"
+    "    com.google.protobuf.ByteString data,\n"
+    "    com.google.protobuf.ExtensionRegistry extensionRegistry)\n"
+    "    throws com.google.protobuf.InvalidProtocolBufferException {\n"
+    "  return newBuilder().mergeFrom(data, extensionRegistry)\n"
+    "           .buildParsed();\n"
+    "}\n"
+    "public static $classname$ parseFrom(byte[] data)\n"
+    "    throws com.google.protobuf.InvalidProtocolBufferException {\n"
+    "  return newBuilder().mergeFrom(data).buildParsed();\n"
+    "}\n"
+    "public static $classname$ parseFrom(\n"
+    "    byte[] data,\n"
+    "    com.google.protobuf.ExtensionRegistry extensionRegistry)\n"
+    "    throws com.google.protobuf.InvalidProtocolBufferException {\n"
+    "  return newBuilder().mergeFrom(data, extensionRegistry)\n"
+    "           .buildParsed();\n"
+    "}\n"
+    "public static $classname$ parseFrom(java.io.InputStream input)\n"
+    "    throws java.io.IOException {\n"
+    "  return newBuilder().mergeFrom(input).buildParsed();\n"
+    "}\n"
+    "public static $classname$ parseFrom(\n"
+    "    java.io.InputStream input,\n"
+    "    com.google.protobuf.ExtensionRegistry extensionRegistry)\n"
+    "    throws java.io.IOException {\n"
+    "  return newBuilder().mergeFrom(input, extensionRegistry)\n"
+    "           .buildParsed();\n"
+    "}\n"
+    "public static $classname$ parseFrom(\n"
+    "    com.google.protobuf.CodedInputStream input)\n"
+    "    throws java.io.IOException {\n"
+    "  return newBuilder().mergeFrom(input).buildParsed();\n"
+    "}\n"
+    "public static $classname$ parseFrom(\n"
+    "    com.google.protobuf.CodedInputStream input,\n"
+    "    com.google.protobuf.ExtensionRegistry extensionRegistry)\n"
+    "    throws java.io.IOException {\n"
+    "  return newBuilder().mergeFrom(input, extensionRegistry)\n"
+    "           .buildParsed();\n"
+    "}\n"
+    "\n",
+    "classname", ClassName(descriptor_));
+}
+
+void MessageGenerator::GenerateSerializeOneField(
+    io::Printer* printer, const FieldDescriptor* field) {
+  field_generators_.get(field).GenerateSerializationCode(printer);
+}
+
+void MessageGenerator::GenerateSerializeOneExtensionRange(
+    io::Printer* printer, const Descriptor::ExtensionRange* range) {
+  printer->Print(
+    "extensionWriter.writeUntil($end$, output);\n",
+    "end", SimpleItoa(range->end));
+}
+
+// ===================================================================
+
+void MessageGenerator::GenerateBuilder(io::Printer* printer) {
+  printer->Print(
+    "public static Builder newBuilder() { return new Builder(); }\n"
+    "public Builder newBuilderForType() { return new Builder(); }\n"
+    "public static Builder newBuilder($classname$ prototype) {\n"
+    "  return new Builder().mergeFrom(prototype);\n"
+    "}\n"
+    "\n",
+    "classname", ClassName(descriptor_));
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "public static final class Builder extends\n"
+      "    com.google.protobuf.GeneratedMessage.ExtendableBuilder<\n"
+      "      $classname$, Builder> {\n",
+      "classname", ClassName(descriptor_));
+  } else {
+    printer->Print(
+      "public static final class Builder extends\n"
+      "    com.google.protobuf.GeneratedMessage.Builder<Builder> {\n",
+      "classname", ClassName(descriptor_));
+  }
+  printer->Indent();
+
+  GenerateCommonBuilderMethods(printer);
+
+  if (descriptor_->file()->options().optimize_for() == FileOptions::SPEED) {
+    GenerateBuilderParsingMethods(printer);
+  }
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    printer->Print("\n");
+    PrintFieldComment(printer, descriptor_->field(i));
+    field_generators_.get(descriptor_->field(i))
+                     .GenerateBuilderMembers(printer);
+  }
+
+  printer->Outdent();
+  printer->Print("}\n");
+  printer->Outdent();
+  printer->Print("}\n\n");
+}
+
+// ===================================================================
+
+void MessageGenerator::GenerateCommonBuilderMethods(io::Printer* printer) {
+  printer->Print(
+    "// Construct using $classname$.newBuilder()\n"
+    "private Builder() {}\n"
+    "\n"
+    "$classname$ result = new $classname$();\n"
+    "\n"
+    "protected $classname$ internalGetResult() {\n"
+    "  return result;\n"
+    "}\n"
+    "\n"
+    "public Builder clear() {\n"
+    "  result = new $classname$();\n"
+    "  return this;\n"
+    "}\n"
+    "\n"
+    "public Builder clone() {\n"
+    "  return new Builder().mergeFrom(result);\n"
+    "}\n"
+    "\n"
+    "public com.google.protobuf.Descriptors.Descriptor\n"
+    "    getDescriptorForType() {\n"
+    "  return $classname$.getDescriptor();\n"
+    "}\n"
+    "\n"
+    "public $classname$ getDefaultInstanceForType() {\n"
+    "  return $classname$.getDefaultInstance();\n"
+    "}\n"
+    "\n",
+    "classname", ClassName(descriptor_));
+
+  // -----------------------------------------------------------------
+
+  printer->Print(
+    "public $classname$ build() {\n"
+    "  if (!isInitialized()) {\n"
+    "    throw new com.google.protobuf.UninitializedMessageException(\n"
+    "      result);\n"
+    "  }\n"
+    "  return buildPartial();\n"
+    "}\n"
+    "\n"
+    "private $classname$ buildParsed()\n"
+    "    throws com.google.protobuf.InvalidProtocolBufferException {\n"
+    "  if (!isInitialized()) {\n"
+    "    throw new com.google.protobuf.UninitializedMessageException(\n"
+    "      result).asInvalidProtocolBufferException();\n"
+    "  }\n"
+    "  return buildPartial();\n"
+    "}\n"
+    "\n"
+    "public $classname$ buildPartial() {\n",
+    "classname", ClassName(descriptor_));
+  printer->Indent();
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    field_generators_.get(descriptor_->field(i)).GenerateBuildingCode(printer);
+  }
+
+  printer->Outdent();
+  printer->Print(
+    "  $classname$ returnMe = result;\n"
+    "  result = null;\n"
+    "  return returnMe;\n"
+    "}\n"
+    "\n",
+    "classname", ClassName(descriptor_));
+
+  // -----------------------------------------------------------------
+
+  if (descriptor_->file()->options().optimize_for() == FileOptions::SPEED) {
+    printer->Print(
+      "public Builder mergeFrom(com.google.protobuf.Message other) {\n"
+      "  if (other instanceof $classname$) {\n"
+      "    return mergeFrom(($classname$)other);\n"
+      "  } else {\n"
+      "    super.mergeFrom(other);\n"
+      "    return this;\n"
+      "  }\n"
+      "}\n"
+      "\n"
+      "public Builder mergeFrom($classname$ other) {\n"
+      // Optimization:  If other is the default instance, we know none of its
+      //   fields are set so we can skip the merge.
+      "  if (other == $classname$.getDefaultInstance()) return this;\n",
+      "classname", ClassName(descriptor_));
+    printer->Indent();
+
+    for (int i = 0; i < descriptor_->field_count(); i++) {
+      field_generators_.get(descriptor_->field(i)).GenerateMergingCode(printer);
+    }
+
+    printer->Outdent();
+    printer->Print(
+      "  this.mergeUnknownFields(other.getUnknownFields());\n"
+      "  return this;\n"
+      "}\n"
+      "\n");
+  }
+}
+
+// ===================================================================
+
+void MessageGenerator::GenerateBuilderParsingMethods(io::Printer* printer) {
+  scoped_array<const FieldDescriptor*> sorted_fields(
+    SortFieldsByNumber(descriptor_));
+
+  printer->Print(
+    "public Builder mergeFrom(\n"
+    "    com.google.protobuf.CodedInputStream input)\n"
+    "    throws java.io.IOException {\n"
+    "  return mergeFrom(input,\n"
+    "    com.google.protobuf.ExtensionRegistry.getEmptyRegistry());\n"
+    "}\n"
+    "\n"
+    "public Builder mergeFrom(\n"
+    "    com.google.protobuf.CodedInputStream input,\n"
+    "    com.google.protobuf.ExtensionRegistry extensionRegistry)\n"
+    "    throws java.io.IOException {\n");
+  printer->Indent();
+
+  printer->Print(
+    "com.google.protobuf.UnknownFieldSet.Builder unknownFields =\n"
+    "  com.google.protobuf.UnknownFieldSet.newBuilder(\n"
+    "    this.getUnknownFields());\n"
+    "while (true) {\n");
+  printer->Indent();
+
+  printer->Print(
+    "int tag = input.readTag();\n"
+    "switch (tag) {\n");
+  printer->Indent();
+
+  printer->Print(
+    "case 0:\n"          // zero signals EOF / limit reached
+    "  this.setUnknownFields(unknownFields.build());\n"
+    "  return this;\n"
+    "default: {\n"
+    "  if (!parseUnknownField(input, unknownFields,\n"
+    "                         extensionRegistry, tag)) {\n"
+    "    this.setUnknownFields(unknownFields.build());\n"
+    "    return this;\n"   // it's an endgroup tag
+    "  }\n"
+    "  break;\n"
+    "}\n");
+
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = sorted_fields[i];
+    uint32 tag = WireFormat::MakeTag(field->number(),
+      WireFormat::WireTypeForFieldType(field->type()));
+
+    printer->Print(
+      "case $tag$: {\n",
+      "tag", SimpleItoa(tag));
+    printer->Indent();
+
+    field_generators_.get(field).GenerateParsingCode(printer);
+
+    printer->Outdent();
+    printer->Print(
+      "  break;\n"
+      "}\n");
+  }
+
+  printer->Outdent();
+  printer->Outdent();
+  printer->Outdent();
+  printer->Print(
+    "    }\n"     // switch (tag)
+    "  }\n"       // while (true)
+    "}\n"
+    "\n");
+}
+
+// ===================================================================
+
+void MessageGenerator::GenerateIsInitialized(io::Printer* printer) {
+  printer->Print(
+    "public final boolean isInitialized() {\n");
+  printer->Indent();
+
+  // Check that all required fields in this message are set.
+  // TODO(kenton):  We can optimize this when we switch to putting all the
+  //   "has" fields into a single bitfield.
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+
+    if (field->is_required()) {
+      printer->Print(
+        "if (!has$name$) return false;\n",
+        "name", UnderscoresToCapitalizedCamelCase(field));
+    }
+  }
+
+  // Now check that all embedded messages are initialized.
+  for (int i = 0; i < descriptor_->field_count(); i++) {
+    const FieldDescriptor* field = descriptor_->field(i);
+    if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE &&
+        HasRequiredFields(field->message_type())) {
+      switch (field->label()) {
+        case FieldDescriptor::LABEL_REQUIRED:
+          printer->Print(
+            "if (!get$name$().isInitialized()) return false;\n",
+            "type", ClassName(field->message_type()),
+            "name", UnderscoresToCapitalizedCamelCase(field));
+          break;
+        case FieldDescriptor::LABEL_OPTIONAL:
+          printer->Print(
+            "if (has$name$()) {\n"
+            "  if (!get$name$().isInitialized()) return false;\n"
+            "}\n",
+            "type", ClassName(field->message_type()),
+            "name", UnderscoresToCapitalizedCamelCase(field));
+          break;
+        case FieldDescriptor::LABEL_REPEATED:
+          printer->Print(
+            "for ($type$ element : get$name$List()) {\n"
+            "  if (!element.isInitialized()) return false;\n"
+            "}\n",
+            "type", ClassName(field->message_type()),
+            "name", UnderscoresToCapitalizedCamelCase(field));
+          break;
+      }
+    }
+  }
+
+  if (descriptor_->extension_range_count() > 0) {
+    printer->Print(
+      "if (!extensionsAreInitialized()) return false;\n");
+  }
+
+  printer->Outdent();
+  printer->Print(
+    "  return true;\n"
+    "}\n"
+    "\n");
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_message.h b/src/google/protobuf/compiler/java/java_message.h
new file mode 100644
index 00000000..d9f8798e
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_message.h
@@ -0,0 +1,76 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_MESSAGE_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_MESSAGE_H__
+
+#include <string>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/compiler/java/java_field.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class MessageGenerator {
+ public:
+  explicit MessageGenerator(const Descriptor* descriptor);
+  ~MessageGenerator();
+
+  // All static variables have to be declared at the top-level of the file
+  // so that we can control initialization order, which is important for
+  // DescriptorProto bootstrapping to work.
+  void GenerateStaticVariables(io::Printer* printer);
+
+  // Generate the class itself.
+  void Generate(io::Printer* printer);
+
+ private:
+  void GenerateMessageSerializationMethods(io::Printer* printer);
+  void GenerateParseFromMethods(io::Printer* printer);
+  void GenerateSerializeOneField(io::Printer* printer,
+                                 const FieldDescriptor* field);
+  void GenerateSerializeOneExtensionRange(
+      io::Printer* printer, const Descriptor::ExtensionRange* range);
+
+  void GenerateBuilder(io::Printer* printer);
+  void GenerateCommonBuilderMethods(io::Printer* printer);
+  void GenerateBuilderParsingMethods(io::Printer* printer);
+  void GenerateIsInitialized(io::Printer* printer);
+
+  const Descriptor* descriptor_;
+  FieldGeneratorMap field_generators_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(MessageGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_MESSAGE_H__
diff --git a/src/google/protobuf/compiler/java/java_message_field.cc b/src/google/protobuf/compiler/java/java_message_field.cc
new file mode 100644
index 00000000..16ddb0d6
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_message_field.cc
@@ -0,0 +1,302 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <map>
+#include <string>
+
+#include <google/protobuf/compiler/java/java_message_field.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/wire_format.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+namespace {
+
+// TODO(kenton):  Factor out a "SetCommonFieldVariables()" to get rid of
+//   repeat code between this and the other field types.
+void SetMessageVariables(const FieldDescriptor* descriptor,
+                         map<string, string>* variables) {
+  (*variables)["name"] =
+    UnderscoresToCamelCase(descriptor);
+  (*variables)["capitalized_name"] =
+    UnderscoresToCapitalizedCamelCase(descriptor);
+  (*variables)["number"] = SimpleItoa(descriptor->number());
+  (*variables)["type"] = ClassName(descriptor->message_type());
+  (*variables)["group_or_message"] =
+    (descriptor->type() == FieldDescriptor::TYPE_GROUP) ?
+    "Group" : "Message";
+}
+
+}  // namespace
+
+// ===================================================================
+
+MessageFieldGenerator::
+MessageFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetMessageVariables(descriptor, &variables_);
+}
+
+MessageFieldGenerator::~MessageFieldGenerator() {}
+
+void MessageFieldGenerator::
+GenerateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "private boolean has$capitalized_name$;\n"
+    "private $type$ $name$_ = $type$.getDefaultInstance();\n"
+    "public boolean has$capitalized_name$() { return has$capitalized_name$; }\n"
+    "public $type$ get$capitalized_name$() { return $name$_; }\n");
+}
+
+void MessageFieldGenerator::
+GenerateBuilderMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "public boolean has$capitalized_name$() {\n"
+    "  return result.has$capitalized_name$();\n"
+    "}\n"
+    "public $type$ get$capitalized_name$() {\n"
+    "  return result.get$capitalized_name$();\n"
+    "}\n"
+    "public Builder set$capitalized_name$($type$ value) {\n"
+    "  result.has$capitalized_name$ = true;\n"
+    "  result.$name$_ = value;\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder set$capitalized_name$($type$.Builder builderForValue) {\n"
+    "  result.has$capitalized_name$ = true;\n"
+    "  result.$name$_ = builderForValue.build();\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder merge$capitalized_name$($type$ value) {\n"
+    "  if (result.has$capitalized_name$() &&\n"
+    "      result.$name$_ != $type$.getDefaultInstance()) {\n"
+    "    result.$name$_ =\n"
+    "      $type$.newBuilder(result.$name$_).mergeFrom(value).buildPartial();\n"
+    "  } else {\n"
+    "    result.$name$_ = value;\n"
+    "  }\n"
+    "  result.has$capitalized_name$ = true;\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder clear$capitalized_name$() {\n"
+    "  result.has$capitalized_name$ = false;\n"
+    "  result.$name$_ = $type$.getDefaultInstance();\n"
+    "  return this;\n"
+    "}\n");
+}
+
+void MessageFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (other.has$capitalized_name$()) {\n"
+    "  merge$capitalized_name$(other.get$capitalized_name$());\n"
+    "}\n");
+}
+
+void MessageFieldGenerator::
+GenerateBuildingCode(io::Printer* printer) const {
+  // Nothing to do for singular fields.
+}
+
+void MessageFieldGenerator::
+GenerateParsingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "$type$.Builder subBuilder = $type$.newBuilder();\n"
+    "if (has$capitalized_name$()) {\n"
+    "  subBuilder.mergeFrom(get$capitalized_name$());\n"
+    "}\n");
+
+  if (descriptor_->type() == FieldDescriptor::TYPE_GROUP) {
+    printer->Print(variables_,
+      "input.readGroup($number$, subBuilder, extensionRegistry);\n");
+  } else {
+    printer->Print(variables_,
+      "input.readMessage(subBuilder, extensionRegistry);\n");
+  }
+
+  printer->Print(variables_,
+    "set$capitalized_name$(subBuilder.buildPartial());\n");
+}
+
+void MessageFieldGenerator::
+GenerateSerializationCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (has$capitalized_name$()) {\n"
+    "  output.write$group_or_message$($number$, get$capitalized_name$());\n"
+    "}\n");
+}
+
+void MessageFieldGenerator::
+GenerateSerializedSizeCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (has$capitalized_name$()) {\n"
+    "  size += com.google.protobuf.CodedOutputStream\n"
+    "    .compute$group_or_message$Size($number$, get$capitalized_name$());\n"
+    "}\n");
+}
+
+string MessageFieldGenerator::GetBoxedType() const {
+  return ClassName(descriptor_->message_type());
+}
+
+// ===================================================================
+
+RepeatedMessageFieldGenerator::
+RepeatedMessageFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetMessageVariables(descriptor, &variables_);
+}
+
+RepeatedMessageFieldGenerator::~RepeatedMessageFieldGenerator() {}
+
+void RepeatedMessageFieldGenerator::
+GenerateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "private java.util.List<$type$> $name$_ =\n"
+    "  java.util.Collections.emptyList();\n"
+    "public java.util.List<$type$> get$capitalized_name$List() {\n"
+    "  return $name$_;\n"   // note:  unmodifiable list
+    "}\n"
+    "public int get$capitalized_name$Count() { return $name$_.size(); }\n"
+    "public $type$ get$capitalized_name$(int index) {\n"
+    "  return $name$_.get(index);\n"
+    "}\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateBuilderMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    // Note:  We return an unmodifiable list because otherwise the caller
+    //   could hold on to the returned list and modify it after the message
+    //   has been built, thus mutating the message which is supposed to be
+    //   immutable.
+    "public java.util.List<$type$> get$capitalized_name$List() {\n"
+    "  return java.util.Collections.unmodifiableList(result.$name$_);\n"
+    "}\n"
+    "public int get$capitalized_name$Count() {\n"
+    "  return result.get$capitalized_name$Count();\n"
+    "}\n"
+    "public $type$ get$capitalized_name$(int index) {\n"
+    "  return result.get$capitalized_name$(index);\n"
+    "}\n"
+    "public Builder set$capitalized_name$(int index, $type$ value) {\n"
+    "  result.$name$_.set(index, value);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder set$capitalized_name$(int index, "
+      "$type$.Builder builderForValue) {\n"
+    "  result.$name$_.set(index, builderForValue.build());\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder add$capitalized_name$($type$ value) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$type$>();\n"
+    "  }\n"
+    "  result.$name$_.add(value);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder add$capitalized_name$($type$.Builder builderForValue) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$type$>();\n"
+    "  }\n"
+    "  result.$name$_.add(builderForValue.build());\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder addAll$capitalized_name$(\n"
+    "    java.lang.Iterable<? extends $type$> values) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$type$>();\n"
+    "  }\n"
+    "  super.addAll(values, result.$name$_);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder clear$capitalized_name$() {\n"
+    "  result.$name$_ = java.util.Collections.emptyList();\n"
+    "  return this;\n"
+    "}\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (!other.$name$_.isEmpty()) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$type$>();\n"
+    "  }\n"
+    "  result.$name$_.addAll(other.$name$_);\n"
+    "}\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateBuildingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (result.$name$_ != java.util.Collections.EMPTY_LIST) {\n"
+    "  result.$name$_ =\n"
+    "    java.util.Collections.unmodifiableList(result.$name$_);\n"
+    "}\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateParsingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "$type$.Builder subBuilder = $type$.newBuilder();\n");
+
+  if (descriptor_->type() == FieldDescriptor::TYPE_GROUP) {
+    printer->Print(variables_,
+      "input.readGroup($number$, subBuilder, extensionRegistry);\n");
+  } else {
+    printer->Print(variables_,
+      "input.readMessage(subBuilder, extensionRegistry);\n");
+  }
+
+  printer->Print(variables_,
+    "add$capitalized_name$(subBuilder.buildPartial());\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateSerializationCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "for ($type$ element : get$capitalized_name$List()) {\n"
+    "  output.write$group_or_message$($number$, element);\n"
+    "}\n");
+}
+
+void RepeatedMessageFieldGenerator::
+GenerateSerializedSizeCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "for ($type$ element : get$capitalized_name$List()) {\n"
+    "  size += com.google.protobuf.CodedOutputStream\n"
+    "    .compute$group_or_message$Size($number$, element);\n"
+    "}\n");
+}
+
+string RepeatedMessageFieldGenerator::GetBoxedType() const {
+  return ClassName(descriptor_->message_type());
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_message_field.h b/src/google/protobuf/compiler/java/java_message_field.h
new file mode 100644
index 00000000..52cf7d15
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_message_field.h
@@ -0,0 +1,84 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_MESSAGE_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_MESSAGE_FIELD_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/compiler/java/java_field.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class MessageFieldGenerator : public FieldGenerator {
+ public:
+  explicit MessageFieldGenerator(const FieldDescriptor* descriptor);
+  ~MessageFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GenerateMembers(io::Printer* printer) const;
+  void GenerateBuilderMembers(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateBuildingCode(io::Printer* printer) const;
+  void GenerateParsingCode(io::Printer* printer) const;
+  void GenerateSerializationCode(io::Printer* printer) const;
+  void GenerateSerializedSizeCode(io::Printer* printer) const;
+
+  string GetBoxedType() const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(MessageFieldGenerator);
+};
+
+class RepeatedMessageFieldGenerator : public FieldGenerator {
+ public:
+  explicit RepeatedMessageFieldGenerator(const FieldDescriptor* descriptor);
+  ~RepeatedMessageFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GenerateMembers(io::Printer* printer) const;
+  void GenerateBuilderMembers(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateBuildingCode(io::Printer* printer) const;
+  void GenerateParsingCode(io::Printer* printer) const;
+  void GenerateSerializationCode(io::Printer* printer) const;
+  void GenerateSerializedSizeCode(io::Printer* printer) const;
+
+  string GetBoxedType() const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(RepeatedMessageFieldGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_MESSAGE_FIELD_H__
diff --git a/src/google/protobuf/compiler/java/java_primitive_field.cc b/src/google/protobuf/compiler/java/java_primitive_field.cc
new file mode 100644
index 00000000..9138f2c9
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_primitive_field.cc
@@ -0,0 +1,365 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <map>
+#include <string>
+
+#include <google/protobuf/compiler/java/java_primitive_field.h>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/stubs/substitute.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+namespace {
+
+const char* PrimitiveTypeName(JavaType type) {
+  switch (type) {
+    case JAVATYPE_INT    : return "int";
+    case JAVATYPE_LONG   : return "long";
+    case JAVATYPE_FLOAT  : return "float";
+    case JAVATYPE_DOUBLE : return "double";
+    case JAVATYPE_BOOLEAN: return "boolean";
+    case JAVATYPE_STRING : return "java.lang.String";
+    case JAVATYPE_BYTES  : return "com.google.protobuf.ByteString";
+    case JAVATYPE_ENUM   : return NULL;
+    case JAVATYPE_MESSAGE: return NULL;
+
+    // No default because we want the compiler to complain if any new
+    // JavaTypes are added.
+  }
+
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return NULL;
+}
+
+const char* GetCapitalizedType(const FieldDescriptor* field) {
+  switch (field->type()) {
+    case FieldDescriptor::TYPE_INT32   : return "Int32"   ;
+    case FieldDescriptor::TYPE_UINT32  : return "UInt32"  ;
+    case FieldDescriptor::TYPE_SINT32  : return "SInt32"  ;
+    case FieldDescriptor::TYPE_FIXED32 : return "Fixed32" ;
+    case FieldDescriptor::TYPE_SFIXED32: return "SFixed32";
+    case FieldDescriptor::TYPE_INT64   : return "Int64"   ;
+    case FieldDescriptor::TYPE_UINT64  : return "UInt64"  ;
+    case FieldDescriptor::TYPE_SINT64  : return "SInt64"  ;
+    case FieldDescriptor::TYPE_FIXED64 : return "Fixed64" ;
+    case FieldDescriptor::TYPE_SFIXED64: return "SFixed64";
+    case FieldDescriptor::TYPE_FLOAT   : return "Float"   ;
+    case FieldDescriptor::TYPE_DOUBLE  : return "Double"  ;
+    case FieldDescriptor::TYPE_BOOL    : return "Bool"    ;
+    case FieldDescriptor::TYPE_STRING  : return "String"  ;
+    case FieldDescriptor::TYPE_BYTES   : return "Bytes"   ;
+    case FieldDescriptor::TYPE_ENUM    : return "Enum"    ;
+    case FieldDescriptor::TYPE_GROUP   : return "Group"   ;
+    case FieldDescriptor::TYPE_MESSAGE : return "Message" ;
+
+    // No default because we want the compiler to complain if any new
+    // types are added.
+  }
+
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return NULL;
+}
+
+bool AllPrintableAscii(const string& text) {
+  // Cannot use isprint() because it's locale-specific.  :(
+  for (int i = 0; i < text.size(); i++) {
+    if ((text[i] < 0x20) || text[i] >= 0x7F) {
+      return false;
+    }
+  }
+  return true;
+}
+
+string DefaultValue(const FieldDescriptor* field) {
+  // Switch on cpp_type since we need to know which default_value_* method
+  // of FieldDescriptor to call.
+  switch (field->cpp_type()) {
+    case FieldDescriptor::CPPTYPE_INT32:
+      return SimpleItoa(field->default_value_int32());
+    case FieldDescriptor::CPPTYPE_UINT32:
+      // Need to print as a signed int since Java has no unsigned.
+      return SimpleItoa(static_cast<int32>(field->default_value_uint32()));
+    case FieldDescriptor::CPPTYPE_INT64:
+      return SimpleItoa(field->default_value_int64()) + "L";
+    case FieldDescriptor::CPPTYPE_UINT64:
+      return SimpleItoa(static_cast<int64>(field->default_value_uint64())) +
+             "L";
+    case FieldDescriptor::CPPTYPE_DOUBLE:
+      return SimpleDtoa(field->default_value_double()) + "D";
+    case FieldDescriptor::CPPTYPE_FLOAT:
+      return SimpleFtoa(field->default_value_float()) + "F";
+    case FieldDescriptor::CPPTYPE_BOOL:
+      return field->default_value_bool() ? "true" : "false";
+    case FieldDescriptor::CPPTYPE_STRING: {
+      bool isBytes = field->type() == FieldDescriptor::TYPE_BYTES;
+
+      if (!isBytes && AllPrintableAscii(field->default_value_string())) {
+        // All chars are ASCII and printable.  In this case CEscape() works
+        // fine (it will only escape quotes and backslashes).
+        // Note:  If this "optimization" is removed, DescriptorProtos will
+        //   no longer be able to initialize itself due to bootstrapping
+        //   problems.
+        return "\"" + CEscape(field->default_value_string()) + "\"";
+      }
+
+      if (isBytes && !field->has_default_value()) {
+        return "com.google.protobuf.ByteString.EMPTY";
+      }
+
+      // Escaping strings correctly for Java and generating efficient
+      // initializers for ByteStrings are both tricky.  We can sidestep the
+      // whole problem by just grabbing the default value from the descriptor.
+      return strings::Substitute(
+        "(($0) $1.getDescriptor().getFields().get($2).getDefaultValue())",
+        isBytes ? "com.google.protobuf.ByteString" : "java.lang.String",
+        ClassName(field->containing_type()), field->index());
+    }
+
+    case FieldDescriptor::CPPTYPE_ENUM:
+    case FieldDescriptor::CPPTYPE_MESSAGE:
+      GOOGLE_LOG(FATAL) << "Can't get here.";
+      return "";
+
+    // No default because we want the compiler to complain if any new
+    // types are added.
+  }
+
+  GOOGLE_LOG(FATAL) << "Can't get here.";
+  return "";
+}
+
+void SetPrimitiveVariables(const FieldDescriptor* descriptor,
+                           map<string, string>* variables) {
+  (*variables)["name"] =
+    UnderscoresToCamelCase(descriptor);
+  (*variables)["capitalized_name"] =
+    UnderscoresToCapitalizedCamelCase(descriptor);
+  (*variables)["number"] = SimpleItoa(descriptor->number());
+  (*variables)["type"] = PrimitiveTypeName(GetJavaType(descriptor));
+  (*variables)["boxed_type"] = BoxedPrimitiveTypeName(GetJavaType(descriptor));
+  (*variables)["default"] = DefaultValue(descriptor);
+  (*variables)["capitalized_type"] = GetCapitalizedType(descriptor);
+}
+
+}  // namespace
+
+// ===================================================================
+
+PrimitiveFieldGenerator::
+PrimitiveFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetPrimitiveVariables(descriptor, &variables_);
+}
+
+PrimitiveFieldGenerator::~PrimitiveFieldGenerator() {}
+
+void PrimitiveFieldGenerator::
+GenerateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "private boolean has$capitalized_name$;\n"
+    "private $type$ $name$_ = $default$;\n"
+    "public boolean has$capitalized_name$() { return has$capitalized_name$; }\n"
+    "public $type$ get$capitalized_name$() { return $name$_; }\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateBuilderMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "public boolean has$capitalized_name$() {\n"
+    "  return result.has$capitalized_name$();\n"
+    "}\n"
+    "public $type$ get$capitalized_name$() {\n"
+    "  return result.get$capitalized_name$();\n"
+    "}\n"
+    "public Builder set$capitalized_name$($type$ value) {\n"
+    "  result.has$capitalized_name$ = true;\n"
+    "  result.$name$_ = value;\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder clear$capitalized_name$() {\n"
+    "  result.has$capitalized_name$ = false;\n"
+    "  result.$name$_ = $default$;\n"
+    "  return this;\n"
+    "}\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (other.has$capitalized_name$()) {\n"
+    "  set$capitalized_name$(other.get$capitalized_name$());\n"
+    "}\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateBuildingCode(io::Printer* printer) const {
+  // Nothing to do here for primitive types.
+}
+
+void PrimitiveFieldGenerator::
+GenerateParsingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "set$capitalized_name$(input.read$capitalized_type$());\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateSerializationCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (has$capitalized_name$()) {\n"
+    "  output.write$capitalized_type$($number$, get$capitalized_name$());\n"
+    "}\n");
+}
+
+void PrimitiveFieldGenerator::
+GenerateSerializedSizeCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (has$capitalized_name$()) {\n"
+    "  size += com.google.protobuf.CodedOutputStream\n"
+    "    .compute$capitalized_type$Size($number$, get$capitalized_name$());\n"
+    "}\n");
+}
+
+string PrimitiveFieldGenerator::GetBoxedType() const {
+  return BoxedPrimitiveTypeName(GetJavaType(descriptor_));
+}
+
+// ===================================================================
+
+RepeatedPrimitiveFieldGenerator::
+RepeatedPrimitiveFieldGenerator(const FieldDescriptor* descriptor)
+  : descriptor_(descriptor) {
+  SetPrimitiveVariables(descriptor, &variables_);
+}
+
+RepeatedPrimitiveFieldGenerator::~RepeatedPrimitiveFieldGenerator() {}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    "private java.util.List<$boxed_type$> $name$_ =\n"
+    "  java.util.Collections.emptyList();\n"
+    "public java.util.List<$boxed_type$> get$capitalized_name$List() {\n"
+    "  return $name$_;\n"   // note:  unmodifiable list
+    "}\n"
+    "public int get$capitalized_name$Count() { return $name$_.size(); }\n"
+    "public $type$ get$capitalized_name$(int index) {\n"
+    "  return $name$_.get(index);\n"
+    "}\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateBuilderMembers(io::Printer* printer) const {
+  printer->Print(variables_,
+    // Note:  We return an unmodifiable list because otherwise the caller
+    //   could hold on to the returned list and modify it after the message
+    //   has been built, thus mutating the message which is supposed to be
+    //   immutable.
+    "public java.util.List<$boxed_type$> get$capitalized_name$List() {\n"
+    "  return java.util.Collections.unmodifiableList(result.$name$_);\n"
+    "}\n"
+    "public int get$capitalized_name$Count() {\n"
+    "  return result.get$capitalized_name$Count();\n"
+    "}\n"
+    "public $type$ get$capitalized_name$(int index) {\n"
+    "  return result.get$capitalized_name$(index);\n"
+    "}\n"
+    "public Builder set$capitalized_name$(int index, $type$ value) {\n"
+    "  result.$name$_.set(index, value);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder add$capitalized_name$($type$ value) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$boxed_type$>();\n"
+    "  }\n"
+    "  result.$name$_.add(value);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder addAll$capitalized_name$(\n"
+    "    java.lang.Iterable<? extends $boxed_type$> values) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$boxed_type$>();\n"
+    "  }\n"
+    "  super.addAll(values, result.$name$_);\n"
+    "  return this;\n"
+    "}\n"
+    "public Builder clear$capitalized_name$() {\n"
+    "  result.$name$_ = java.util.Collections.emptyList();\n"
+    "  return this;\n"
+    "}\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateMergingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (!other.$name$_.isEmpty()) {\n"
+    "  if (result.$name$_.isEmpty()) {\n"
+    "    result.$name$_ = new java.util.ArrayList<$boxed_type$>();\n"
+    "  }\n"
+    "  result.$name$_.addAll(other.$name$_);\n"
+    "}\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateBuildingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "if (result.$name$_ != java.util.Collections.EMPTY_LIST) {\n"
+    "  result.$name$_ =\n"
+    "    java.util.Collections.unmodifiableList(result.$name$_);\n"
+    "}\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateParsingCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "add$capitalized_name$(input.read$capitalized_type$());\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateSerializationCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "for ($type$ element : get$capitalized_name$List()) {\n"
+    "  output.write$capitalized_type$($number$, element);\n"
+    "}\n");
+}
+
+void RepeatedPrimitiveFieldGenerator::
+GenerateSerializedSizeCode(io::Printer* printer) const {
+  printer->Print(variables_,
+    "for ($type$ element : get$capitalized_name$List()) {\n"
+    "  size += com.google.protobuf.CodedOutputStream\n"
+    "    .compute$capitalized_type$Size($number$, element);\n"
+    "}\n");
+}
+
+string RepeatedPrimitiveFieldGenerator::GetBoxedType() const {
+  return BoxedPrimitiveTypeName(GetJavaType(descriptor_));
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_primitive_field.h b/src/google/protobuf/compiler/java/java_primitive_field.h
new file mode 100644
index 00000000..6fe9177a
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_primitive_field.h
@@ -0,0 +1,84 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_PRIMITIVE_FIELD_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_PRIMITIVE_FIELD_H__
+
+#include <map>
+#include <string>
+#include <google/protobuf/compiler/java/java_field.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class PrimitiveFieldGenerator : public FieldGenerator {
+ public:
+  explicit PrimitiveFieldGenerator(const FieldDescriptor* descriptor);
+  ~PrimitiveFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GenerateMembers(io::Printer* printer) const;
+  void GenerateBuilderMembers(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateBuildingCode(io::Printer* printer) const;
+  void GenerateParsingCode(io::Printer* printer) const;
+  void GenerateSerializationCode(io::Printer* printer) const;
+  void GenerateSerializedSizeCode(io::Printer* printer) const;
+
+  string GetBoxedType() const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(PrimitiveFieldGenerator);
+};
+
+class RepeatedPrimitiveFieldGenerator : public FieldGenerator {
+ public:
+  explicit RepeatedPrimitiveFieldGenerator(const FieldDescriptor* descriptor);
+  ~RepeatedPrimitiveFieldGenerator();
+
+  // implements FieldGenerator ---------------------------------------
+  void GenerateMembers(io::Printer* printer) const;
+  void GenerateBuilderMembers(io::Printer* printer) const;
+  void GenerateMergingCode(io::Printer* printer) const;
+  void GenerateBuildingCode(io::Printer* printer) const;
+  void GenerateParsingCode(io::Printer* printer) const;
+  void GenerateSerializationCode(io::Printer* printer) const;
+  void GenerateSerializedSizeCode(io::Printer* printer) const;
+
+  string GetBoxedType() const;
+
+ private:
+  const FieldDescriptor* descriptor_;
+  map<string, string> variables_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(RepeatedPrimitiveFieldGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_JAVA_PRIMITIVE_FIELD_H__
diff --git a/src/google/protobuf/compiler/java/java_service.cc b/src/google/protobuf/compiler/java/java_service.cc
new file mode 100644
index 00000000..8cb06270
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_service.cc
@@ -0,0 +1,225 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <google/protobuf/compiler/java/java_service.h>
+#include <google/protobuf/compiler/java/java_helpers.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+ServiceGenerator::ServiceGenerator(const ServiceDescriptor* descriptor)
+  : descriptor_(descriptor) {}
+
+ServiceGenerator::~ServiceGenerator() {}
+
+void ServiceGenerator::Generate(io::Printer* printer) {
+  bool is_own_file = descriptor_->file()->options().java_multiple_files();
+  printer->Print(
+    "public $static$ abstract class $classname$\n"
+    "    implements com.google.protobuf.Service {\n",
+    "static", is_own_file ? "" : "static",
+    "classname", descriptor_->name());
+  printer->Indent();
+
+  // Generate abstract method declarations.
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> vars;
+    vars["name"] = UnderscoresToCamelCase(method);
+    vars["input"] = ClassName(method->input_type());
+    vars["output"] = ClassName(method->output_type());
+    printer->Print(vars,
+      "public abstract void $name$(\n"
+      "    com.google.protobuf.RpcController controller,\n"
+      "    $input$ request,\n"
+      "    com.google.protobuf.RpcCallback<$output$> done);\n");
+  }
+
+  // Generate getDescriptor() and getDescriptorForType().
+  printer->Print(
+    "\n"
+    "public static final\n"
+    "    com.google.protobuf.Descriptors.ServiceDescriptor\n"
+    "    getDescriptor() {\n"
+    "  return $file$.getDescriptor().getServices().get($index$);\n"
+    "}\n"
+    "public final com.google.protobuf.Descriptors.ServiceDescriptor\n"
+    "    getDescriptorForType() {\n"
+    "  return getDescriptor();\n"
+    "}\n",
+    "file", ClassName(descriptor_->file()),
+    "index", SimpleItoa(descriptor_->index()));
+
+  // Generate more stuff.
+  GenerateCallMethod(printer);
+  GenerateGetPrototype(REQUEST, printer);
+  GenerateGetPrototype(RESPONSE, printer);
+  GenerateStub(printer);
+
+  printer->Outdent();
+  printer->Print("}\n\n");
+}
+
+void ServiceGenerator::GenerateCallMethod(io::Printer* printer) {
+  printer->Print(
+    "\n"
+    "public final void callMethod(\n"
+    "    com.google.protobuf.Descriptors.MethodDescriptor method,\n"
+    "    com.google.protobuf.RpcController controller,\n"
+    "    com.google.protobuf.Message request,\n"
+    "    com.google.protobuf.RpcCallback<\n"
+    "      com.google.protobuf.Message> done) {\n"
+    "  if (method.getService() != getDescriptor()) {\n"
+    "    throw new java.lang.IllegalArgumentException(\n"
+    "      \"Service.callMethod() given method descriptor for wrong \" +\n"
+    "      \"service type.\");\n"
+    "  }\n"
+    "  switch(method.getIndex()) {\n");
+  printer->Indent();
+  printer->Indent();
+
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> vars;
+    vars["index"] = SimpleItoa(i);
+    vars["method"] = UnderscoresToCamelCase(method);
+    vars["input"] = ClassName(method->input_type());
+    vars["output"] = ClassName(method->output_type());
+    printer->Print(vars,
+      "case $index$:\n"
+      "  this.$method$(controller, ($input$)request,\n"
+      "    com.google.protobuf.RpcUtil.<$output$>specializeCallback(\n"
+      "      done));\n"
+      "  return;\n");
+  }
+
+  printer->Print(
+    "default:\n"
+    "  throw new java.lang.RuntimeException(\"Can't get here.\");\n");
+
+  printer->Outdent();
+  printer->Outdent();
+
+  printer->Print(
+    "  }\n"
+    "}\n"
+    "\n");
+}
+
+void ServiceGenerator::GenerateGetPrototype(RequestOrResponse which,
+                                            io::Printer* printer) {
+  printer->Print(
+    "public final com.google.protobuf.Message\n"
+    "    get$request_or_response$Prototype(\n"
+    "    com.google.protobuf.Descriptors.MethodDescriptor method) {\n"
+    "  if (method.getService() != getDescriptor()) {\n"
+    "    throw new java.lang.IllegalArgumentException(\n"
+    "      \"Service.get$request_or_response$Prototype() given method \" +\n"
+    "      \"descriptor for wrong service type.\");\n"
+    "  }\n"
+    "  switch(method.getIndex()) {\n",
+    "request_or_response", (which == REQUEST) ? "Request" : "Response");
+  printer->Indent();
+  printer->Indent();
+
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> vars;
+    vars["index"] = SimpleItoa(i);
+    vars["type"] = ClassName(
+      (which == REQUEST) ? method->input_type() : method->output_type());
+    printer->Print(vars,
+      "case $index$:\n"
+      "  return $type$.getDefaultInstance();\n");
+  }
+
+  printer->Print(
+    "default:\n"
+    "  throw new java.lang.RuntimeException(\"Can't get here.\");\n");
+
+  printer->Outdent();
+  printer->Outdent();
+
+  printer->Print(
+    "  }\n"
+    "}\n"
+    "\n");
+}
+
+void ServiceGenerator::GenerateStub(io::Printer* printer) {
+  printer->Print(
+    "public static Stub newStub(\n"
+    "    com.google.protobuf.RpcChannel channel) {\n"
+    "  return new Stub(channel);\n"
+    "}\n"
+    "\n"
+    "public static final class Stub extends $classname$ {\n",
+    "classname", ClassName(descriptor_));
+  printer->Indent();
+
+  printer->Print(
+    "private Stub(com.google.protobuf.RpcChannel channel) {\n"
+    "  this.channel = channel;\n"
+    "}\n"
+    "\n"
+    "private final com.google.protobuf.RpcChannel channel;\n"
+    "\n"
+    "public com.google.protobuf.RpcChannel getChannel() {\n"
+    "  return channel;\n"
+    "}\n");
+
+  for (int i = 0; i < descriptor_->method_count(); i++) {
+    const MethodDescriptor* method = descriptor_->method(i);
+    map<string, string> vars;
+    vars["index"] = SimpleItoa(i);
+    vars["method"] = UnderscoresToCamelCase(method);
+    vars["input"] = ClassName(method->input_type());
+    vars["output"] = ClassName(method->output_type());
+    printer->Print(vars,
+      "\n"
+      "public void $method$(\n"
+      "    com.google.protobuf.RpcController controller,\n"
+      "    $input$ request,\n"
+      "    com.google.protobuf.RpcCallback<$output$> done) {\n"
+      "  channel.callMethod(\n"
+      "    getDescriptor().getMethods().get($index$),\n"
+      "    controller,\n"
+      "    request,\n"
+      "    $output$.getDefaultInstance(),\n"
+      "    com.google.protobuf.RpcUtil.generalizeCallback(\n"
+      "      done,\n"
+      "      $output$.class,\n"
+      "      $output$.getDefaultInstance()));\n"
+      "}\n");
+  }
+
+  printer->Outdent();
+  printer->Print("}\n");
+}
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/java/java_service.h b/src/google/protobuf/compiler/java/java_service.h
new file mode 100644
index 00000000..adf3dfd4
--- /dev/null
+++ b/src/google/protobuf/compiler/java/java_service.h
@@ -0,0 +1,66 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_SERVICE_H__
+#define GOOGLE_PROTOBUF_COMPILER_JAVA_SERVICE_H__
+
+#include <map>
+#include <google/protobuf/descriptor.h>
+
+namespace google {
+namespace protobuf {
+  namespace io {
+    class Printer;             // printer.h
+  }
+}
+
+namespace protobuf {
+namespace compiler {
+namespace java {
+
+class ServiceGenerator {
+ public:
+  explicit ServiceGenerator(const ServiceDescriptor* descriptor);
+  ~ServiceGenerator();
+
+  void Generate(io::Printer* printer);
+
+ private:
+  // Generate the implementation of Service.callMethod().
+  void GenerateCallMethod(io::Printer* printer);
+
+  // Generate the implementations of Service.get{Request,Response}Prototype().
+  enum RequestOrResponse { REQUEST, RESPONSE };
+  void GenerateGetPrototype(RequestOrResponse which, io::Printer* printer);
+
+  // Generate a stub implementation of the service.
+  void GenerateStub(io::Printer* printer);
+
+  const ServiceDescriptor* descriptor_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(ServiceGenerator);
+};
+
+}  // namespace java
+}  // namespace compiler
+}  // namespace protobuf
+
+#endif  // NET_PROTO2_COMPILER_JAVA_SERVICE_H__
+}  // namespace google
diff --git a/src/google/protobuf/compiler/main.cc b/src/google/protobuf/compiler/main.cc
new file mode 100644
index 00000000..a5a28349
--- /dev/null
+++ b/src/google/protobuf/compiler/main.cc
@@ -0,0 +1,46 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+
+#include <google/protobuf/compiler/command_line_interface.h>
+#include <google/protobuf/compiler/cpp/cpp_generator.h>
+#include <google/protobuf/compiler/python/python_generator.h>
+#include <google/protobuf/compiler/java/java_generator.h>
+
+
+int main(int argc, char* argv[]) {
+
+  google::protobuf::compiler::CommandLineInterface cli;
+
+  // Proto2 C++
+  google::protobuf::compiler::cpp::CppGenerator cpp_generator;
+  cli.RegisterGenerator("--cpp_out", &cpp_generator,
+                        "Generate C++ header and source.");
+
+  // Proto2 Java
+  google::protobuf::compiler::java::JavaGenerator java_generator;
+  cli.RegisterGenerator("--java_out", &java_generator,
+                        "Generate Java source file.");
+
+
+  // Proto2 Python
+  google::protobuf::compiler::python::Generator py_generator;
+  cli.RegisterGenerator("--python_out", &py_generator,
+                        "Generate Python source file.");
+
+  return cli.Run(argc, argv);
+}
diff --git a/src/google/protobuf/compiler/package_info.h b/src/google/protobuf/compiler/package_info.h
new file mode 100644
index 00000000..aa0c2823
--- /dev/null
+++ b/src/google/protobuf/compiler/package_info.h
@@ -0,0 +1,50 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// This file exists solely to document the google::protobuf::compiler namespace.
+// It is not compiled into anything, but it may be read by an automated
+// documentation generator.
+
+namespace google {
+
+namespace protobuf {
+
+// Implementation of the Protocol Buffer compiler.
+//
+// This package contains code for parsing .proto files and generating code
+// based on them.  There are two reasons you might be interested in this
+// package:
+// - You want to parse .proto files at runtime.  In this case, you should
+//   look at importer.h.  Since this functionality is widely useful, it is
+//   included in the libprotobuf base library; you do not have to link against
+//   libprotoc.
+// - You want to write a custom protocol compiler which generates different
+//   kinds of code, e.g. code in a different language which is not supported
+//   by the official compiler.  For this purpose, command_line_interface.h
+//   provides you with a complete compiler front-end, so all you need to do
+//   is write a custom implementation of CodeGenerator and a trivial main()
+//   function.  You can even make your compiler support the official languages
+//   in addition to your own.  Since this functionality is only useful to those
+//   writing custom compilers, it is in a separate library called "libprotoc"
+//   which you will have to link against.
+namespace compiler {}
+
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/parser.cc b/src/google/protobuf/compiler/parser.cc
new file mode 100644
index 00000000..622895ff
--- /dev/null
+++ b/src/google/protobuf/compiler/parser.cc
@@ -0,0 +1,1105 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// Recursive descent FTW.
+
+#include <google/protobuf/stubs/hash.h>
+#include <float.h>
+
+
+#include <google/protobuf/compiler/parser.h>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/wire_format.h>
+#include <google/protobuf/io/tokenizer.h>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/stubs/map-util.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+
+using internal::WireFormat;
+
+namespace {
+
+typedef hash_map<string, FieldDescriptorProto::Type> TypeNameMap;
+
+TypeNameMap MakeTypeNameTable() {
+  TypeNameMap result;
+
+  result["double"  ] = FieldDescriptorProto::TYPE_DOUBLE;
+  result["float"   ] = FieldDescriptorProto::TYPE_FLOAT;
+  result["uint64"  ] = FieldDescriptorProto::TYPE_UINT64;
+  result["fixed64" ] = FieldDescriptorProto::TYPE_FIXED64;
+  result["fixed32" ] = FieldDescriptorProto::TYPE_FIXED32;
+  result["bool"    ] = FieldDescriptorProto::TYPE_BOOL;
+  result["string"  ] = FieldDescriptorProto::TYPE_STRING;
+  result["group"   ] = FieldDescriptorProto::TYPE_GROUP;
+
+  result["bytes"   ] = FieldDescriptorProto::TYPE_BYTES;
+  result["uint32"  ] = FieldDescriptorProto::TYPE_UINT32;
+  result["sfixed32"] = FieldDescriptorProto::TYPE_SFIXED32;
+  result["sfixed64"] = FieldDescriptorProto::TYPE_SFIXED64;
+  result["int32"   ] = FieldDescriptorProto::TYPE_INT32;
+  result["int64"   ] = FieldDescriptorProto::TYPE_INT64;
+  result["sint32"  ] = FieldDescriptorProto::TYPE_SINT32;
+  result["sint64"  ] = FieldDescriptorProto::TYPE_SINT64;
+
+  return result;
+}
+
+const TypeNameMap kTypeNames = MakeTypeNameTable();
+
+}  // anonymous namespace
+
+// Makes code slightly more readable.  The meaning of "DO(foo)" is
+// "Execute foo and fail if it fails.", where failure is indicated by
+// returning false.
+#define DO(STATEMENT) if (STATEMENT) {} else return false
+
+// ===================================================================
+
+Parser::Parser()
+  : input_(NULL),
+    error_collector_(NULL),
+    source_location_table_(NULL),
+    had_errors_(false),
+    require_syntax_identifier_(false) {
+}
+
+Parser::~Parser() {
+}
+
+// ===================================================================
+
+inline bool Parser::LookingAt(const char* text) {
+  return input_->current().text == text;
+}
+
+inline bool Parser::LookingAtType(io::Tokenizer::TokenType token_type) {
+  return input_->current().type == token_type;
+}
+
+inline bool Parser::AtEnd() {
+  return LookingAtType(io::Tokenizer::TYPE_END);
+}
+
+bool Parser::TryConsume(const char* text) {
+  if (LookingAt(text)) {
+    input_->Next();
+    return true;
+  } else {
+    return false;
+  }
+}
+
+bool Parser::Consume(const char* text, const char* error) {
+  if (TryConsume(text)) {
+    return true;
+  } else {
+    AddError(error);
+    return false;
+  }
+}
+
+bool Parser::Consume(const char* text) {
+  if (TryConsume(text)) {
+    return true;
+  } else {
+    AddError("Expected \"" + string(text) + "\".");
+    return false;
+  }
+}
+
+bool Parser::ConsumeIdentifier(string* output, const char* error) {
+  if (LookingAtType(io::Tokenizer::TYPE_IDENTIFIER)) {
+    *output = input_->current().text;
+    input_->Next();
+    return true;
+  } else {
+    AddError(error);
+    return false;
+  }
+}
+
+bool Parser::ConsumeInteger(int* output, const char* error) {
+  if (LookingAtType(io::Tokenizer::TYPE_INTEGER)) {
+    uint64 value = 0;
+    if (!io::Tokenizer::ParseInteger(input_->current().text,
+                                     kint32max, &value)) {
+      AddError("Integer out of range.");
+      // We still return true because we did, in fact, parse an integer.
+    }
+    *output = value;
+    input_->Next();
+    return true;
+  } else {
+    AddError(error);
+    return false;
+  }
+}
+
+bool Parser::ConsumeInteger64(uint64 max_value, uint64* output,
+                              const char* error) {
+  if (LookingAtType(io::Tokenizer::TYPE_INTEGER)) {
+    if (!io::Tokenizer::ParseInteger(input_->current().text, max_value,
+                                     output)) {
+      AddError("Integer out of range.");
+      // We still return true because we did, in fact, parse an integer.
+      *output = 0;
+    }
+    input_->Next();
+    return true;
+  } else {
+    AddError(error);
+    return false;
+  }
+}
+
+bool Parser::ConsumeNumber(double* output, const char* error) {
+  if (LookingAtType(io::Tokenizer::TYPE_FLOAT)) {
+    *output = io::Tokenizer::ParseFloat(input_->current().text);
+    input_->Next();
+    return true;
+  } else if (LookingAtType(io::Tokenizer::TYPE_INTEGER)) {
+    // Also accept integers.
+    uint64 value = 0;
+    if (!io::Tokenizer::ParseInteger(input_->current().text,
+                                     kuint64max, &value)) {
+      AddError("Integer out of range.");
+      // We still return true because we did, in fact, parse a number.
+    }
+    *output = value;
+    input_->Next();
+    return true;
+  } else {
+    AddError(error);
+    return false;
+  }
+}
+
+bool Parser::ConsumeString(string* output, const char* error) {
+  if (LookingAtType(io::Tokenizer::TYPE_STRING)) {
+    io::Tokenizer::ParseString(input_->current().text, output);
+    input_->Next();
+    return true;
+  } else {
+    AddError(error);
+    return false;
+  }
+}
+
+// -------------------------------------------------------------------
+
+void Parser::AddError(int line, int column, const string& error) {
+  if (error_collector_ != NULL) {
+    error_collector_->AddError(line, column, error);
+  }
+  had_errors_ = true;
+}
+
+void Parser::AddError(const string& error) {
+  AddError(input_->current().line, input_->current().column, error);
+}
+
+void Parser::RecordLocation(
+    const Message* descriptor,
+    DescriptorPool::ErrorCollector::ErrorLocation location,
+    int line, int column) {
+  if (source_location_table_ != NULL) {
+    source_location_table_->Add(descriptor, location, line, column);
+  }
+}
+
+void Parser::RecordLocation(
+    const Message* descriptor,
+    DescriptorPool::ErrorCollector::ErrorLocation location) {
+  RecordLocation(descriptor, location,
+                 input_->current().line, input_->current().column);
+}
+
+// -------------------------------------------------------------------
+
+void Parser::SkipStatement() {
+  while (true) {
+    if (AtEnd()) {
+      return;
+    } else if (LookingAtType(io::Tokenizer::TYPE_SYMBOL)) {
+      if (TryConsume(";")) {
+        return;
+      } else if (TryConsume("{")) {
+        SkipRestOfBlock();
+        return;
+      } else if (LookingAt("}")) {
+        return;
+      }
+    }
+    input_->Next();
+  }
+}
+
+void Parser::SkipRestOfBlock() {
+  while (true) {
+    if (AtEnd()) {
+      return;
+    } else if (LookingAtType(io::Tokenizer::TYPE_SYMBOL)) {
+      if (TryConsume("}")) {
+        return;
+      } else if (TryConsume("{")) {
+        SkipRestOfBlock();
+      }
+    }
+    input_->Next();
+  }
+}
+
+// ===================================================================
+
+bool Parser::Parse(io::Tokenizer* input, FileDescriptorProto* file) {
+  input_ = input;
+  had_errors_ = false;
+  syntax_identifier_.clear();
+
+  if (LookingAtType(io::Tokenizer::TYPE_START)) {
+    // Advance to first token.
+    input_->Next();
+  }
+
+  if (require_syntax_identifier_ || LookingAt("syntax")) {
+    if (!ParseSyntaxIdentifier()) {
+      // Don't attempt to parse the file if we didn't recognize the syntax
+      // identifier.
+      return false;
+    }
+  } else {
+    syntax_identifier_ = "proto2";
+  }
+
+  // Repeatedly parse statemetns until we reach the end of the file.
+  while (!AtEnd()) {
+    if (!ParseTopLevelStatement(file)) {
+      // This statement failed to parse.  Skip it, but keep looping to parse
+      // other statements.
+      SkipStatement();
+
+      if (LookingAt("}")) {
+        AddError("Unmatched \"}\".");
+        input_->Next();
+      }
+    }
+  }
+
+  input_ = NULL;
+  return !had_errors_;
+}
+
+bool Parser::ParseSyntaxIdentifier() {
+  DO(Consume("syntax", "File must begin with 'syntax = \"proto2\";'."));
+  DO(Consume("="));
+  io::Tokenizer::Token syntax_token = input_->current();
+  string syntax;
+  DO(ConsumeString(&syntax, "Expected syntax identifier."));
+  DO(Consume(";"));
+
+  syntax_identifier_ = syntax;
+
+  if (syntax != "proto2") {
+    AddError(syntax_token.line, syntax_token.column,
+      "Unrecognized syntax identifier \"" + syntax + "\".  This parser "
+      "only recognizes \"proto2\".");
+    return false;
+  }
+
+  return true;
+}
+
+bool Parser::ParseTopLevelStatement(FileDescriptorProto* file) {
+  if (TryConsume(";")) {
+    // empty statement; ignore
+    return true;
+  } else if (LookingAt("message")) {
+    return ParseMessageDefinition(file->add_message_type());
+  } else if (LookingAt("enum")) {
+    return ParseEnumDefinition(file->add_enum_type());
+  } else if (LookingAt("service")) {
+    return ParseServiceDefinition(file->add_service());
+  } else if (LookingAt("extend")) {
+    return ParseExtend(file->mutable_extension(),
+                       file->mutable_message_type());
+  } else if (LookingAt("import")) {
+    return ParseImport(file->add_dependency());
+  } else if (LookingAt("package")) {
+    return ParsePackage(file);
+  } else if (LookingAt("option")) {
+    return ParseOption(file->mutable_options());
+  } else {
+    AddError("Expected top-level statement (e.g. \"message\").");
+    return false;
+  }
+}
+
+// -------------------------------------------------------------------
+// Messages
+
+bool Parser::ParseMessageDefinition(DescriptorProto* message) {
+  DO(Consume("message"));
+  RecordLocation(message, DescriptorPool::ErrorCollector::NAME);
+  DO(ConsumeIdentifier(message->mutable_name(), "Expected message name."));
+  DO(ParseMessageBlock(message));
+  return true;
+}
+
+bool Parser::ParseMessageBlock(DescriptorProto* message) {
+  DO(Consume("{"));
+
+  while (!TryConsume("}")) {
+    if (AtEnd()) {
+      AddError("Reached end of input in message definition (missing '}').");
+      return false;
+    }
+
+    if (!ParseMessageStatement(message)) {
+      // This statement failed to parse.  Skip it, but keep looping to parse
+      // other statements.
+      SkipStatement();
+    }
+  }
+
+  return true;
+}
+
+bool Parser::ParseMessageStatement(DescriptorProto* message) {
+  if (TryConsume(";")) {
+    // empty statement; ignore
+    return true;
+  } else if (LookingAt("message")) {
+    return ParseMessageDefinition(message->add_nested_type());
+  } else if (LookingAt("enum")) {
+    return ParseEnumDefinition(message->add_enum_type());
+  } else if (LookingAt("extensions")) {
+    return ParseExtensions(message);
+  } else if (LookingAt("extend")) {
+    return ParseExtend(message->mutable_extension(),
+                       message->mutable_nested_type());
+  } else if (LookingAt("option")) {
+    return ParseOption(message->mutable_options());
+  } else {
+    return ParseMessageField(message->add_field(),
+                             message->mutable_nested_type());
+  }
+}
+
+bool Parser::ParseMessageField(FieldDescriptorProto* field,
+                               RepeatedPtrField<DescriptorProto>* messages) {
+  // Parse label and type.
+  FieldDescriptorProto::Label label;
+  DO(ParseLabel(&label));
+  field->set_label(label);
+
+  RecordLocation(field, DescriptorPool::ErrorCollector::TYPE);
+  FieldDescriptorProto::Type type = FieldDescriptorProto::TYPE_INT32;
+  string type_name;
+  DO(ParseType(&type, &type_name));
+  if (type_name.empty()) {
+    field->set_type(type);
+  } else {
+    field->set_type_name(type_name);
+  }
+
+  // Parse name and '='.
+  RecordLocation(field, DescriptorPool::ErrorCollector::NAME);
+  io::Tokenizer::Token name_token = input_->current();
+  DO(ConsumeIdentifier(field->mutable_name(), "Expected field name."));
+  DO(Consume("=", "Missing field number."));
+
+  // Parse field number.
+  RecordLocation(field, DescriptorPool::ErrorCollector::NUMBER);
+  int number;
+  DO(ConsumeInteger(&number, "Expected field number."));
+  field->set_number(number);
+
+  // Parse options.
+  DO(ParseFieldOptions(field));
+
+  // Deal with groups.
+  if (type_name.empty() && type == FieldDescriptorProto::TYPE_GROUP) {
+    DescriptorProto* group = messages->Add();
+    group->set_name(field->name());
+    // Record name location to match the field name's location.
+    RecordLocation(group, DescriptorPool::ErrorCollector::NAME,
+                   name_token.line, name_token.column);
+
+    // As a hack for backwards-compatibility, we force the group name to start
+    // with a capital letter and lower-case the field name.  New code should
+    // not use groups; it should use nested messages.
+    if (group->name()[0] < 'A' || 'Z' < group->name()[0]) {
+      AddError(name_token.line, name_token.column,
+        "Group names must start with a capital letter.");
+    }
+    LowerString(field->mutable_name());
+
+    field->set_type_name(group->name());
+    if (LookingAt("{")) {
+      DO(ParseMessageBlock(group));
+    } else {
+      AddError("Missing group body.");
+      return false;
+    }
+  } else {
+    DO(Consume(";"));
+  }
+
+  return true;
+}
+
+bool Parser::ParseFieldOptions(FieldDescriptorProto* field) {
+  if (!TryConsume("[")) return true;
+
+  // Parse field options.
+  do {
+    if (LookingAt("default")) {
+      DO(ParseDefaultAssignment(field));
+    } else {
+      DO(ParseOptionAssignment(field->mutable_options()));
+    }
+  } while (TryConsume(","));
+
+  DO(Consume("]"));
+  return true;
+}
+
+bool Parser::ParseDefaultAssignment(FieldDescriptorProto* field) {
+  if (field->has_default_value()) {
+    AddError("Already set option \"default\".");
+    field->clear_default_value();
+  }
+
+  DO(Consume("default"));
+  DO(Consume("="));
+
+  RecordLocation(field, DescriptorPool::ErrorCollector::DEFAULT_VALUE);
+  string* default_value = field->mutable_default_value();
+
+  if (!field->has_type()) {
+    // The field has a type name, but we don't know if it is a message or an
+    // enum yet.  Assume an enum for now.
+    DO(ConsumeIdentifier(default_value, "Expected identifier."));
+    return true;
+  }
+
+  switch (field->type()) {
+    case FieldDescriptorProto::TYPE_INT32:
+    case FieldDescriptorProto::TYPE_INT64:
+    case FieldDescriptorProto::TYPE_SINT32:
+    case FieldDescriptorProto::TYPE_SINT64:
+    case FieldDescriptorProto::TYPE_SFIXED32:
+    case FieldDescriptorProto::TYPE_SFIXED64: {
+      uint64 max_value = kint64max;
+      if (field->type() == FieldDescriptorProto::TYPE_INT32 ||
+          field->type() == FieldDescriptorProto::TYPE_SINT32 ||
+          field->type() == FieldDescriptorProto::TYPE_SFIXED32) {
+        max_value = kint32max;
+      }
+
+      // These types can be negative.
+      if (TryConsume("-")) {
+        default_value->append("-");
+        // Two's complement always has one more negative value than positive.
+        ++max_value;
+      }
+      // Parse the integer to verify that it is not out-of-range.
+      uint64 value;
+      DO(ConsumeInteger64(max_value, &value, "Expected integer."));
+      // And stringify it again.
+      default_value->append(SimpleItoa(value));
+      break;
+    }
+
+    case FieldDescriptorProto::TYPE_UINT32:
+    case FieldDescriptorProto::TYPE_UINT64:
+    case FieldDescriptorProto::TYPE_FIXED32:
+    case FieldDescriptorProto::TYPE_FIXED64: {
+      uint64 max_value = kuint64max;
+      if (field->type() == FieldDescriptorProto::TYPE_UINT32 ||
+          field->type() == FieldDescriptorProto::TYPE_FIXED32) {
+        max_value = kuint32max;
+      }
+
+      // Numeric, not negative.
+      if (TryConsume("-")) {
+        AddError("Unsigned field can't have negative default value.");
+      }
+      // Parse the integer to verify that it is not out-of-range.
+      uint64 value;
+      DO(ConsumeInteger64(max_value, &value, "Expected integer."));
+      // And stringify it again.
+      default_value->append(SimpleItoa(value));
+      break;
+    }
+
+    case FieldDescriptorProto::TYPE_FLOAT:
+    case FieldDescriptorProto::TYPE_DOUBLE:
+      // These types can be negative.
+      if (TryConsume("-")) {
+        default_value->append("-");
+      }
+      // Parse the integer because we have to convert hex integers to decimal
+      // floats.
+      double value;
+      DO(ConsumeNumber(&value, "Expected number."));
+      // And stringify it again.
+      default_value->append(SimpleDtoa(value));
+      break;
+
+    case FieldDescriptorProto::TYPE_BOOL:
+      if (TryConsume("true")) {
+        default_value->assign("true");
+      } else if (TryConsume("false")) {
+        default_value->assign("false");
+      } else {
+        AddError("Expected \"true\" or \"false\".");
+        return false;
+      }
+      break;
+
+    case FieldDescriptorProto::TYPE_STRING:
+      DO(ConsumeString(default_value, "Expected string."));
+      break;
+
+    case FieldDescriptorProto::TYPE_BYTES:
+      DO(ConsumeString(default_value, "Expected string."));
+      *default_value = CEscape(*default_value);
+      break;
+
+    case FieldDescriptorProto::TYPE_ENUM:
+      DO(ConsumeIdentifier(default_value, "Expected identifier."));
+      break;
+
+    case FieldDescriptorProto::TYPE_MESSAGE:
+    case FieldDescriptorProto::TYPE_GROUP:
+      AddError("Messages can't have default values.");
+      return false;
+  }
+
+  return true;
+}
+
+bool Parser::ParseOptionAssignment(Message* options) {
+  Message::Reflection* reflection = options->GetReflection();
+  const Descriptor* descriptor = options->GetDescriptor();
+
+  // Parse name.
+  string name;
+  int line = input_->current().line;
+  int column = input_->current().column;
+  DO(ConsumeIdentifier(&name, "Expected option name."));
+
+  // Is it valid?
+  const FieldDescriptor* field = descriptor->FindFieldByName(name);
+  if (field == NULL) {
+    AddError(line, column, "Unknown option: " + name);
+    return false;
+  }
+  if (field->is_repeated()) {
+    AddError(line, column, "Not implemented: repeated options.");
+    return false;
+  }
+  if (reflection->HasField(field)) {
+    AddError(line, column, "Option \"" + name + "\" was already set.");
+    return false;
+  }
+
+  // Are we trying to assign a member of a message?
+  if (LookingAt(".")) {
+    if (field->cpp_type() != FieldDescriptor::CPPTYPE_MESSAGE) {
+      AddError("Option \"" + name + "\" is an atomic type, not a message.");
+      return false;
+    }
+    DO(Consume("."));
+
+    // This field is a message/group.  The user must identify a field within
+    // it to set.
+    return ParseOptionAssignment(reflection->MutableMessage(field));
+  }
+
+  DO(Consume("="));
+
+  // Parse the option value.
+  switch (field->cpp_type()) {
+    case FieldDescriptor::CPPTYPE_INT32: {
+      uint64 value;
+      bool is_negative = TryConsume("-");
+      uint64 max_value = kint32max;
+      if (is_negative) ++max_value;
+      DO(ConsumeInteger64(max_value, &value, "Expected integer."));
+      reflection->SetInt32(field, is_negative ? -value : value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_INT64: {
+      uint64 value;
+      bool is_negative = TryConsume("-");
+      uint64 max_value = kint64max;
+      if (is_negative) ++max_value;
+      DO(ConsumeInteger64(max_value, &value, "Expected integer."));
+      reflection->SetInt64(field, is_negative ? -value : value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_UINT32: {
+      uint64 value;
+      DO(ConsumeInteger64(kuint32max, &value, "Expected integer."));
+      reflection->SetUInt32(field, value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_UINT64: {
+      uint64 value;
+      DO(ConsumeInteger64(kuint64max, &value, "Expected integer."));
+      reflection->SetUInt64(field, value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_DOUBLE: {
+      double value;
+      bool is_negative = TryConsume("-");
+      DO(ConsumeNumber(&value, "Expected number."));
+      reflection->SetDouble(field, is_negative ? -value : value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_FLOAT: {
+      double value;
+      bool is_negative = TryConsume("-");
+      DO(ConsumeNumber(&value, "Expected number."));
+      reflection->SetFloat(field, is_negative ? -value : value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_BOOL:
+      if (TryConsume("true")) {
+        reflection->SetBool(field, true);
+      } else if (TryConsume("false")) {
+        reflection->SetBool(field, false);
+      } else {
+        AddError("Expected \"true\" or \"false\".");
+        return false;
+      }
+      break;
+
+    case FieldDescriptor::CPPTYPE_ENUM: {
+      string value_name;
+      int value_line = input_->current().line;
+      int value_column = input_->current().column;
+      DO(ConsumeIdentifier(&value_name, "Expected enum value."));
+      const EnumValueDescriptor* value =
+        field->enum_type()->FindValueByName(value_name);
+      if (value == NULL) {
+        AddError(value_line, value_column,
+          "Enum type \"" + field->enum_type()->full_name() + "\" has no value "
+          "named \"" + value_name + "\".");
+        return false;
+      }
+      reflection->SetEnum(field, value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_STRING: {
+      string value;
+      DO(ConsumeString(&value, "Expected string."));
+      reflection->SetString(field, value);
+      break;
+    }
+
+    case FieldDescriptor::CPPTYPE_MESSAGE: {
+      // TODO(kenton):  Allow use of protocol buffer text format here?
+      AddError("\"" + name + "\" is a message.  To set fields within it, use "
+               "syntax like \"" + name + ".foo = value\".");
+      return false;
+      break;
+    }
+  }
+
+  return true;
+}
+
+bool Parser::ParseExtensions(DescriptorProto* message) {
+  // Parse the declaration.
+  DO(Consume("extensions"));
+
+  do {
+    DescriptorProto::ExtensionRange* range = message->add_extension_range();
+    RecordLocation(range, DescriptorPool::ErrorCollector::NUMBER);
+
+    int start, end;
+    DO(ConsumeInteger(&start, "Expected field number range."));
+
+    if (TryConsume("to")) {
+      if (TryConsume("max")) {
+        end = FieldDescriptor::kMaxNumber;
+      } else {
+        DO(ConsumeInteger(&end, "Expected integer."));
+      }
+    } else {
+      end = start;
+    }
+
+    // Users like to specify inclusive ranges, but in code we like the end
+    // number to be exclusive.
+    ++end;
+
+    range->set_start(start);
+    range->set_end(end);
+  } while (TryConsume(","));
+
+  DO(Consume(";"));
+  return true;
+}
+
+bool Parser::ParseExtend(RepeatedPtrField<FieldDescriptorProto>* extensions,
+                         RepeatedPtrField<DescriptorProto>* messages) {
+  DO(Consume("extend"));
+
+  // We expect to see at least one extension field defined in the extend block.
+  // We need to create it now so we can record the extendee's location.
+  FieldDescriptorProto* first_field = extensions->Add();
+
+  // Parse the extendee type.
+  RecordLocation(first_field, DescriptorPool::ErrorCollector::EXTENDEE);
+  DO(ParseUserDefinedType(first_field->mutable_extendee()));
+
+  // Parse the block.
+  DO(Consume("{"));
+
+  bool is_first = true;
+
+  do {
+    if (AtEnd()) {
+      AddError("Reached end of input in extend definition (missing '}').");
+      return false;
+    }
+
+    FieldDescriptorProto* field;
+    if (is_first) {
+      field = first_field;
+      is_first = false;
+    } else {
+      field = extensions->Add();
+      field->set_extendee(first_field->extendee());
+    }
+
+    if (!ParseMessageField(field, messages)) {
+      // This statement failed to parse.  Skip it, but keep looping to parse
+      // other statements.
+      SkipStatement();
+    }
+  } while(!TryConsume("}"));
+
+  return true;
+}
+
+// -------------------------------------------------------------------
+// Enums
+
+bool Parser::ParseEnumDefinition(EnumDescriptorProto* enum_type) {
+  DO(Consume("enum"));
+  RecordLocation(enum_type, DescriptorPool::ErrorCollector::NAME);
+  DO(ConsumeIdentifier(enum_type->mutable_name(), "Expected enum name."));
+  DO(ParseEnumBlock(enum_type));
+  return true;
+}
+
+bool Parser::ParseEnumBlock(EnumDescriptorProto* enum_type) {
+  DO(Consume("{"));
+
+  while (!TryConsume("}")) {
+    if (AtEnd()) {
+      AddError("Reached end of input in enum definition (missing '}').");
+      return false;
+    }
+
+    if (!ParseEnumStatement(enum_type)) {
+      // This statement failed to parse.  Skip it, but keep looping to parse
+      // other statements.
+      SkipStatement();
+    }
+  }
+
+  return true;
+}
+
+bool Parser::ParseEnumStatement(EnumDescriptorProto* enum_type) {
+  if (TryConsume(";")) {
+    // empty statement; ignore
+    return true;
+  } else if (LookingAt("option")) {
+    return ParseOption(enum_type->mutable_options());
+  } else {
+    return ParseEnumConstant(enum_type->add_value());
+  }
+}
+
+bool Parser::ParseEnumConstant(EnumValueDescriptorProto* enum_value) {
+  RecordLocation(enum_value, DescriptorPool::ErrorCollector::NAME);
+  DO(ConsumeIdentifier(enum_value->mutable_name(),
+                       "Expected enum constant name."));
+  DO(Consume("=", "Missing numeric value for enum constant."));
+
+  bool is_negative = TryConsume("-");
+  int number;
+  DO(ConsumeInteger(&number, "Expected integer."));
+  if (is_negative) number *= -1;
+  enum_value->set_number(number);
+
+  // TODO(kenton):  Options for enum values?
+
+  DO(Consume(";"));
+
+  return true;
+}
+
+// -------------------------------------------------------------------
+// Services
+
+bool Parser::ParseServiceDefinition(ServiceDescriptorProto* service) {
+  DO(Consume("service"));
+  RecordLocation(service, DescriptorPool::ErrorCollector::NAME);
+  DO(ConsumeIdentifier(service->mutable_name(), "Expected service name."));
+  DO(ParseServiceBlock(service));
+  return true;
+}
+
+bool Parser::ParseServiceBlock(ServiceDescriptorProto* service) {
+  DO(Consume("{"));
+
+  while (!TryConsume("}")) {
+    if (AtEnd()) {
+      AddError("Reached end of input in service definition (missing '}').");
+      return false;
+    }
+
+    if (!ParseServiceStatement(service)) {
+      // This statement failed to parse.  Skip it, but keep looping to parse
+      // other statements.
+      SkipStatement();
+    }
+  }
+
+  return true;
+}
+
+bool Parser::ParseServiceStatement(ServiceDescriptorProto* service) {
+  if (TryConsume(";")) {
+    // empty statement; ignore
+    return true;
+  } else if (LookingAt("option")) {
+    return ParseOption(service->mutable_options());
+  } else {
+    return ParseServiceMethod(service->add_method());
+  }
+}
+
+bool Parser::ParseServiceMethod(MethodDescriptorProto* method) {
+  DO(Consume("rpc"));
+  RecordLocation(method, DescriptorPool::ErrorCollector::NAME);
+  DO(ConsumeIdentifier(method->mutable_name(), "Expected method name."));
+
+  // Parse input type.
+  DO(Consume("("));
+  RecordLocation(method, DescriptorPool::ErrorCollector::INPUT_TYPE);
+  DO(ParseUserDefinedType(method->mutable_input_type()));
+  DO(Consume(")"));
+
+  // Parse output type.
+  DO(Consume("returns"));
+  DO(Consume("("));
+  RecordLocation(method, DescriptorPool::ErrorCollector::OUTPUT_TYPE);
+  DO(ParseUserDefinedType(method->mutable_output_type()));
+  DO(Consume(")"));
+
+  if (TryConsume("{")) {
+    // Options!
+    while (!TryConsume("}")) {
+      if (AtEnd()) {
+        AddError("Reached end of input in method options (missing '}').");
+        return false;
+      }
+
+      if (TryConsume(";")) {
+        // empty statement; ignore
+      } else {
+        if (!ParseOption(method->mutable_options())) {
+          // This statement failed to parse.  Skip it, but keep looping to
+          // parse other statements.
+          SkipStatement();
+        }
+      }
+    }
+  } else {
+    DO(Consume(";"));
+  }
+
+  return true;
+}
+
+// -------------------------------------------------------------------
+
+bool Parser::ParseLabel(FieldDescriptorProto::Label* label) {
+  if (TryConsume("optional")) {
+    *label = FieldDescriptorProto::LABEL_OPTIONAL;
+    return true;
+  } else if (TryConsume("repeated")) {
+    *label = FieldDescriptorProto::LABEL_REPEATED;
+    return true;
+  } else if (TryConsume("required")) {
+    *label = FieldDescriptorProto::LABEL_REQUIRED;
+    return true;
+  } else {
+    AddError("Expected \"required\", \"optional\", or \"repeated\".");
+    // We can actually reasonably recover here by just assuming the user
+    // forgot the label altogether.
+    *label = FieldDescriptorProto::LABEL_OPTIONAL;
+    return true;
+  }
+}
+
+bool Parser::ParseType(FieldDescriptorProto::Type* type,
+                       string* type_name) {
+  TypeNameMap::const_iterator iter = kTypeNames.find(input_->current().text);
+  if (iter != kTypeNames.end()) {
+    *type = iter->second;
+    input_->Next();
+  } else {
+    DO(ParseUserDefinedType(type_name));
+  }
+  return true;
+}
+
+bool Parser::ParseUserDefinedType(string* type_name) {
+  type_name->clear();
+
+  TypeNameMap::const_iterator iter = kTypeNames.find(input_->current().text);
+  if (iter != kTypeNames.end()) {
+    // Note:  The only place enum types are allowed is for field types, but
+    //   if we are parsing a field type then we would not get here because
+    //   primitives are allowed there as well.  So this error message doesn't
+    //   need to account for enums.
+    AddError("Expected message type.");
+
+    // Pretend to accept this type so that we can go on parsing.
+    *type_name = input_->current().text;
+    input_->Next();
+    return true;
+  }
+
+  // A leading "." means the name is fully-qualified.
+  if (TryConsume(".")) type_name->append(".");
+
+  // Consume the first part of the name.
+  string identifier;
+  DO(ConsumeIdentifier(&identifier, "Expected type name."));
+  type_name->append(identifier);
+
+  // Consume more parts.
+  while (TryConsume(".")) {
+    type_name->append(".");
+    DO(ConsumeIdentifier(&identifier, "Expected identifier."));
+    type_name->append(identifier);
+  }
+
+  return true;
+}
+
+// ===================================================================
+
+bool Parser::ParsePackage(FileDescriptorProto* file) {
+  if (file->has_package()) {
+    AddError("Multiple package definitions.");
+  }
+
+  DO(Consume("package"));
+
+  RecordLocation(file, DescriptorPool::ErrorCollector::NAME);
+
+  while (true) {
+    string identifier;
+    DO(ConsumeIdentifier(&identifier, "Expected identifier."));
+    file->mutable_package()->append(identifier);
+    if (!TryConsume(".")) break;
+    file->mutable_package()->append(".");
+  }
+
+  DO(Consume(";"));
+  return true;
+}
+
+bool Parser::ParseImport(string* import_filename) {
+  DO(Consume("import"));
+  DO(ConsumeString(import_filename,
+    "Expected a string naming the file to import."));
+  DO(Consume(";"));
+  return true;
+}
+
+bool Parser::ParseOption(Message* options) {
+  DO(Consume("option"));
+  DO(ParseOptionAssignment(options));
+  DO(Consume(";"));
+  return true;
+}
+
+// ===================================================================
+
+SourceLocationTable::SourceLocationTable() {}
+SourceLocationTable::~SourceLocationTable() {}
+
+bool SourceLocationTable::Find(
+    const Message* descriptor,
+    DescriptorPool::ErrorCollector::ErrorLocation location,
+    int* line, int* column) const {
+  const pair<int, int>* result =
+    FindOrNull(location_map_, make_pair(descriptor, location));
+  if (result == NULL) {
+    *line   = -1;
+    *column = 0;
+    return false;
+  } else {
+    *line   = result->first;
+    *column = result->second;
+    return true;
+  }
+}
+
+void SourceLocationTable::Add(
+    const Message* descriptor,
+    DescriptorPool::ErrorCollector::ErrorLocation location,
+    int line, int column) {
+  location_map_[make_pair(descriptor, location)] = make_pair(line, column);
+}
+
+void SourceLocationTable::Clear() {
+  location_map_.clear();
+}
+
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/parser.h b/src/google/protobuf/compiler/parser.h
new file mode 100644
index 00000000..adf6e9b1
--- /dev/null
+++ b/src/google/protobuf/compiler/parser.h
@@ -0,0 +1,301 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+//
+// Implements parsing of .proto files to FileDescriptorProtos.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_PARSER_H__
+#define GOOGLE_PROTOBUF_COMPILER_PARSER_H__
+
+#include <map>
+#include <string>
+#include <utility>
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/repeated_field.h>
+#include <google/protobuf/io/tokenizer.h>
+
+namespace google {
+namespace protobuf { class Message; }
+
+namespace protobuf {
+namespace compiler {
+
+// Defined in this file.
+class Parser;
+class SourceLocationTable;
+
+// Implements parsing of protocol definitions (such as .proto files).
+//
+// Note that most users will be more interested in the Importer class.
+// Parser is a lower-level class which simply converts a single .proto file
+// to a FileDescriptorProto.  It does not resolve import directives or perform
+// many other kinds of validation needed to construct a complete
+// FileDescriptor.
+class LIBPROTOBUF_EXPORT Parser {
+ public:
+  Parser();
+  ~Parser();
+
+  // Parse the entire input and construct a FileDescriptorProto representing
+  // it.  Returns true if no errors occurred, false otherwise.
+  bool Parse(io::Tokenizer* input, FileDescriptorProto* file);
+
+  // Optional fetaures:
+
+  // Requests that locations of certain definitions be recorded to the given
+  // SourceLocationTable while parsing.  This can be used to look up exact line
+  // and column numbers for errors reported by DescriptorPool during validation.
+  // Set to NULL (the default) to discard source location information.
+  void RecordSourceLocationsTo(SourceLocationTable* location_table) {
+    source_location_table_ = location_table;
+  }
+
+  // Requsets that errors be recorded to the given ErrorCollector while
+  // parsing.  Set to NULL (the default) to discard error messages.
+  void RecordErrorsTo(io::ErrorCollector* error_collector) {
+    error_collector_ = error_collector;
+  }
+
+  // Returns the identifier used in the "syntax = " declaration, if one was
+  // seen during the last call to Parse(), or the empty string otherwise.
+  const string& GetSyntaxIndentifier() { return syntax_identifier_; }
+
+  // If set true, input files will be required to begin with a syntax
+  // identifier.  Otherwise, files may omit this.  If a syntax identifier
+  // is provided, it must be 'syntax = "proto2";' and must appear at the
+  // top of this file regardless of whether or not it was required.
+  void SetRequireSyntaxIdentifier(bool value) {
+    require_syntax_identifier_ = value;
+  }
+
+ private:
+  // =================================================================
+  // Error recovery helpers
+
+  // Consume the rest of the current statement.  This consumes tokens
+  // until it sees one of:
+  //   ';'  Consumes the token and returns.
+  //   '{'  Consumes the brace then calls SkipRestOfBlock().
+  //   '}'  Returns without consuming.
+  //   EOF  Returns (can't consume).
+  // The Parser often calls SkipStatement() after encountering a syntax
+  // error.  This allows it to go on parsing the following lines, allowing
+  // it to report more than just one error in the file.
+  void SkipStatement();
+
+  // Consume the rest of the current block, including nested blocks,
+  // ending after the closing '}' is encountered and consumed, or at EOF.
+  void SkipRestOfBlock();
+
+  // -----------------------------------------------------------------
+  // Single-token consuming helpers
+  //
+  // These make parsing code more readable.
+
+  // True if the current token is TYPE_END.
+  inline bool AtEnd();
+
+  // True if the next token matches the given text.
+  inline bool LookingAt(const char* text);
+  // True if the next token is of the given type.
+  inline bool LookingAtType(io::Tokenizer::TokenType token_type);
+
+  // If the next token exactly matches the text given, consume it and return
+  // true.  Otherwise, return false without logging an error.
+  bool TryConsume(const char* text);
+
+  // These attempt to read some kind of token from the input.  If successful,
+  // they return true.  Otherwise they return false and add the given error
+  // to the error list.
+
+  // Consume a token with the exact text given.
+  bool Consume(const char* text, const char* error);
+  // Same as above, but automatically generates the error "Expected \"text\".",
+  // where "text" is the expected token text.
+  bool Consume(const char* text);
+  // Consume a token of type IDENTIFIER and store its text in "output".
+  bool ConsumeIdentifier(string* output, const char* error);
+  // Consume an integer and store its value in "output".
+  bool ConsumeInteger(int* output, const char* error);
+  // Consume a 64-bit integer and store its value in "output".  If the value
+  // is greater than max_value, an error will be reported.
+  bool ConsumeInteger64(uint64 max_value, uint64* output, const char* error);
+  // Consume a number and store its value in "output".  This will accept
+  // tokens of either INTEGER or FLOAT type.
+  bool ConsumeNumber(double* output, const char* error);
+  // Consume a string literal and store its (unescaped) value in "output".
+  bool ConsumeString(string* output, const char* error);
+
+  // -----------------------------------------------------------------
+  // Error logging helpers
+
+  // Invokes error_collector_->AddError(), if error_collector_ is not NULL.
+  void AddError(int line, int column, const string& error);
+
+  // Invokes error_collector_->AddError() with the line and column number
+  // of the current token.
+  void AddError(const string& error);
+
+  // Record the given line and column and associate it with this descriptor
+  // in the SourceLocationTable.
+  void RecordLocation(const Message* descriptor,
+                      DescriptorPool::ErrorCollector::ErrorLocation location,
+                      int line, int column);
+
+  // Record the current line and column and associate it with this descriptor
+  // in the SourceLocationTable.
+  void RecordLocation(const Message* descriptor,
+                      DescriptorPool::ErrorCollector::ErrorLocation location);
+
+  // =================================================================
+  // Parsers for various language constructs
+
+  // Parses the "syntax = \"proto2\";" line at the top of the file.  Returns
+  // false if it failed to parse or if the syntax identifier was not
+  // recognized.
+  bool ParseSyntaxIdentifier();
+
+  // These methods parse various individual bits of code.  They return
+  // false if they completely fail to parse the construct.  In this case,
+  // it is probably necessary to skip the rest of the statement to recover.
+  // However, if these methods return true, it does NOT mean that there
+  // were no errors; only that there were no *syntax* errors.  For instance,
+  // if a service method is defined using proper syntax but uses a primitive
+  // type as its input or output, ParseMethodField() still returns true
+  // and only reports the error by calling AddError().  In practice, this
+  // makes logic much simpler for the caller.
+
+  // Parse a top-level message, enum, service, etc.
+  bool ParseTopLevelStatement(FileDescriptorProto* file);
+
+  // Parse various language high-level language construrcts.
+  bool ParseMessageDefinition(DescriptorProto* message);
+  bool ParseEnumDefinition(EnumDescriptorProto* enum_type);
+  bool ParseServiceDefinition(ServiceDescriptorProto* service);
+  bool ParsePackage(FileDescriptorProto* file);
+  bool ParseImport(string* import_filename);
+  bool ParseOption(Message* options);
+
+  // These methods parse the contents of a message, enum, or service type and
+  // add them to the given object.  They consume the entire block including
+  // the beginning and ending brace.
+  bool ParseMessageBlock(DescriptorProto* message);
+  bool ParseEnumBlock(EnumDescriptorProto* enum_type);
+  bool ParseServiceBlock(ServiceDescriptorProto* service);
+
+  // Parse one statement within a message, enum, or service block, inclunding
+  // final semicolon.
+  bool ParseMessageStatement(DescriptorProto* message);
+  bool ParseEnumStatement(EnumDescriptorProto* message);
+  bool ParseServiceStatement(ServiceDescriptorProto* message);
+
+  // Parse a field of a message.  If the field is a group, its type will be
+  // added to "messages".
+  bool ParseMessageField(FieldDescriptorProto* field,
+                         RepeatedPtrField<DescriptorProto>* messages);
+
+  // Parse an "extensions" declaration.
+  bool ParseExtensions(DescriptorProto* message);
+
+  // Parse an "extend" declaration.
+  bool ParseExtend(RepeatedPtrField<FieldDescriptorProto>* extensions,
+                   RepeatedPtrField<DescriptorProto>* messages);
+
+  // Parse a single enum value within an enum block.
+  bool ParseEnumConstant(EnumValueDescriptorProto* enum_value);
+
+  // Parse a single method within a service definition.
+  bool ParseServiceMethod(MethodDescriptorProto* method);
+
+  // Parse "required", "optional", or "repeated" and fill in "label"
+  // with the value.
+  bool ParseLabel(FieldDescriptorProto::Label* label);
+
+  // Parse a type name and fill in "type" (if it is a primitive) or
+  // "type_name" (if it is not) with the type parsed.
+  bool ParseType(FieldDescriptorProto::Type* type,
+                 string* type_name);
+  // Parse a user-defined type and fill in "type_name" with the name.
+  // If a primitive type is named, it is treated as an error.
+  bool ParseUserDefinedType(string* type_name);
+
+  // Parses field options, i.e. the stuff in square brackets at the end
+  // of a field definition.  Also parses default value.
+  bool ParseFieldOptions(FieldDescriptorProto* field);
+
+  // Parse the "default" option.  This needs special handling because its
+  // type is the field's type.
+  bool ParseDefaultAssignment(FieldDescriptorProto* field);
+
+  // Parse a single option name/value pair, e.g. "ctype = CORD".  The name
+  // identifies a field of the given Message, and the value of that field
+  // is set to the parsed value.
+  bool ParseOptionAssignment(Message* options);
+
+  // =================================================================
+
+  io::Tokenizer* input_;
+  io::ErrorCollector* error_collector_;
+  SourceLocationTable* source_location_table_;
+  bool had_errors_;
+  bool require_syntax_identifier_;
+  string syntax_identifier_;
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(Parser);
+};
+
+// A table mapping (descriptor, ErrorLocation) pairs -- as reported by
+// DescriptorPool when validating descriptors -- to line and column numbers
+// within the original source code.
+class LIBPROTOBUF_EXPORT SourceLocationTable {
+ public:
+  SourceLocationTable();
+  ~SourceLocationTable();
+
+  // Finds the precise location of the given error and fills in *line and
+  // *column with the line and column numbers.  If not found, sets *line to
+  // -1 and *column to 0 (since line = -1 is used to mean "error has no exact
+  // location" in the ErrorCollector interface).  Returns true if found, false
+  // otherwise.
+  bool Find(const Message* descriptor,
+            DescriptorPool::ErrorCollector::ErrorLocation location,
+            int* line, int* column) const;
+
+  // Adds a location to the table.
+  void Add(const Message* descriptor,
+           DescriptorPool::ErrorCollector::ErrorLocation location,
+           int line, int column);
+
+  // Clears the contents of the table.
+  void Clear();
+
+ private:
+  typedef map<
+    pair<const Message*, DescriptorPool::ErrorCollector::ErrorLocation>,
+    pair<int, int> > LocationMap;
+  LocationMap location_map_;
+};
+
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_PARSER_H__
diff --git a/src/google/protobuf/compiler/parser_unittest.cc b/src/google/protobuf/compiler/parser_unittest.cc
new file mode 100644
index 00000000..489106b0
--- /dev/null
+++ b/src/google/protobuf/compiler/parser_unittest.cc
@@ -0,0 +1,1142 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: kenton@google.com (Kenton Varda)
+//  Based on original Protocol Buffers design by
+//  Sanjay Ghemawat, Jeff Dean, and others.
+
+#include <vector>
+#include <algorithm>
+
+#include <google/protobuf/compiler/parser.h>
+
+#include <google/protobuf/io/tokenizer.h>
+#include <google/protobuf/io/zero_copy_stream_impl.h>
+#include <google/protobuf/descriptor.pb.h>
+#include <google/protobuf/wire_format.h>
+#include <google/protobuf/text_format.h>
+#include <google/protobuf/unittest.pb.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/stubs/substitute.h>
+
+#include <google/protobuf/testing/googletest.h>
+#include <gtest/gtest.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+
+namespace {
+
+class MockErrorCollector : public io::ErrorCollector {
+ public:
+  MockErrorCollector() {}
+  ~MockErrorCollector() {}
+
+  string text_;
+
+  // implements ErrorCollector ---------------------------------------
+  void AddError(int line, int column, const string& message) {
+    strings::SubstituteAndAppend(&text_, "$0:$1: $2\n",
+                                 line, column, message);
+  }
+};
+
+class MockValidationErrorCollector : public DescriptorPool::ErrorCollector {
+ public:
+  MockValidationErrorCollector(const SourceLocationTable& source_locations,
+                               io::ErrorCollector* wrapped_collector)
+    : source_locations_(source_locations),
+      wrapped_collector_(wrapped_collector) {}
+  ~MockValidationErrorCollector() {}
+
+  // implements ErrorCollector ---------------------------------------
+  void AddError(const string& filename,
+                const string& element_name,
+                const Message* descriptor,
+                ErrorLocation location,
+                const string& message) {
+    int line, column;
+    source_locations_.Find(descriptor, location, &line, &column);
+    wrapped_collector_->AddError(line, column, message);
+  }
+
+ private:
+  const SourceLocationTable& source_locations_;
+  io::ErrorCollector* wrapped_collector_;
+};
+
+class ParserTest : public testing::Test {
+ protected:
+  ParserTest()
+    : require_syntax_identifier_(false) {}
+
+  // Set up the parser to parse the given text.
+  void SetupParser(const char* text) {
+    raw_input_.reset(new io::ArrayInputStream(text, strlen(text)));
+    input_.reset(new io::Tokenizer(raw_input_.get(), &error_collector_));
+    parser_.reset(new Parser());
+    parser_->RecordErrorsTo(&error_collector_);
+    parser_->SetRequireSyntaxIdentifier(require_syntax_identifier_);
+  }
+
+  // Parse the input and expect that the resulting FileDescriptorProto matches
+  // the given output.  The output is a FileDescriptorProto in protocol buffer
+  // text format.
+  void ExpectParsesTo(const char* input, const char* output) {
+    SetupParser(input);
+    FileDescriptorProto actual, expected;
+
+    parser_->Parse(input_.get(), &actual);
+    EXPECT_EQ(io::Tokenizer::TYPE_END, input_->current().type);
+    ASSERT_EQ("", error_collector_.text_);
+
+    // Parse the ASCII representation in order to canonicalize it.  We could
+    // just compare directly to actual.DebugString(), but that would require
+    // that the caller precisely match the formatting that DebugString()
+    // produces.
+    ASSERT_TRUE(TextFormat::ParseFromString(output, &expected));
+
+    // Compare by comparing debug strings.
+    // TODO(kenton):  Use differencer, once it is available.
+    EXPECT_EQ(expected.DebugString(), actual.DebugString());
+  }
+
+  // Parse the text and expect that the given errors are reported.
+  void ExpectHasErrors(const char* text, const char* expected_errors) {
+    ExpectHasEarlyExitErrors(text, expected_errors);
+    EXPECT_EQ(io::Tokenizer::TYPE_END, input_->current().type);
+  }
+
+  // Same as above but does not expect that the parser parses the complete
+  // input.
+  void ExpectHasEarlyExitErrors(const char* text, const char* expected_errors) {
+    SetupParser(text);
+    FileDescriptorProto file;
+    parser_->Parse(input_.get(), &file);
+    EXPECT_EQ(expected_errors, error_collector_.text_);
+  }
+
+  // Parse the text as a file and validate it (with a DescriptorPool), and
+  // expect that the validation step reports the given errors.
+  void ExpectHasValidationErrors(const char* text,
+                                 const char* expected_errors) {
+    SetupParser(text);
+    SourceLocationTable source_locations;
+    parser_->RecordSourceLocationsTo(&source_locations);
+
+    FileDescriptorProto file;
+    file.set_name("foo.proto");
+    parser_->Parse(input_.get(), &file);
+    EXPECT_EQ(io::Tokenizer::TYPE_END, input_->current().type);
+    ASSERT_EQ("", error_collector_.text_);
+
+    MockValidationErrorCollector validation_error_collector(
+      source_locations, &error_collector_);
+    EXPECT_TRUE(pool_.BuildFileCollectingErrors(
+      file, &validation_error_collector) == NULL);
+    EXPECT_EQ(expected_errors, error_collector_.text_);
+  }
+
+  MockErrorCollector error_collector_;
+  DescriptorPool pool_;
+
+  scoped_ptr<io::ZeroCopyInputStream> raw_input_;
+  scoped_ptr<io::Tokenizer> input_;
+  scoped_ptr<Parser> parser_;
+  bool require_syntax_identifier_;
+};
+
+// ===================================================================
+
+typedef ParserTest ParseMessageTest;
+
+TEST_F(ParseMessageTest, SimpleMessage) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  required int32 foo = 1;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_INT32 number:1 }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, ImplicitSyntaxIdentifier) {
+  require_syntax_identifier_ = false;
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  required int32 foo = 1;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_INT32 number:1 }"
+    "}");
+  EXPECT_EQ("proto2", parser_->GetSyntaxIndentifier());
+}
+
+TEST_F(ParseMessageTest, ExplicitSyntaxIdentifier) {
+  ExpectParsesTo(
+    "syntax = \"proto2\";\n"
+    "message TestMessage {\n"
+    "  required int32 foo = 1;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_INT32 number:1 }"
+    "}");
+  EXPECT_EQ("proto2", parser_->GetSyntaxIndentifier());
+}
+
+TEST_F(ParseMessageTest, ExplicitRequiredSyntaxIdentifier) {
+  require_syntax_identifier_ = true;
+  ExpectParsesTo(
+    "syntax = \"proto2\";\n"
+    "message TestMessage {\n"
+    "  required int32 foo = 1;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_INT32 number:1 }"
+    "}");
+  EXPECT_EQ("proto2", parser_->GetSyntaxIndentifier());
+}
+
+TEST_F(ParseMessageTest, SimpleFields) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  required int32 foo = 15;\n"
+    "  optional int32 bar = 34;\n"
+    "  repeated int32 baz = 3;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_INT32 number:15 }"
+    "  field { name:\"bar\" label:LABEL_OPTIONAL type:TYPE_INT32 number:34 }"
+    "  field { name:\"baz\" label:LABEL_REPEATED type:TYPE_INT32 number:3  }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, PrimitiveFieldTypes) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  required int32    foo = 1;\n"
+    "  required int64    foo = 1;\n"
+    "  required uint32   foo = 1;\n"
+    "  required uint64   foo = 1;\n"
+    "  required sint32   foo = 1;\n"
+    "  required sint64   foo = 1;\n"
+    "  required fixed32  foo = 1;\n"
+    "  required fixed64  foo = 1;\n"
+    "  required sfixed32 foo = 1;\n"
+    "  required sfixed64 foo = 1;\n"
+    "  required float    foo = 1;\n"
+    "  required double   foo = 1;\n"
+    "  required string   foo = 1;\n"
+    "  required bytes    foo = 1;\n"
+    "  required bool     foo = 1;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_INT32    number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_INT64    number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_UINT32   number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_UINT64   number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_SINT32   number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_SINT64   number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_FIXED32  number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_FIXED64  number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_SFIXED32 number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_SFIXED64 number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_FLOAT    number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_DOUBLE   number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_STRING   number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_BYTES    number:1 }"
+    "  field { name:\"foo\" label:LABEL_REQUIRED type:TYPE_BOOL     number:1 }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, FieldDefaults) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  required int32  foo = 1 [default=  1  ];\n"
+    "  required int32  foo = 1 [default= -2  ];\n"
+    "  required int64  foo = 1 [default=  3  ];\n"
+    "  required int64  foo = 1 [default= -4  ];\n"
+    "  required uint32 foo = 1 [default=  5  ];\n"
+    "  required uint64 foo = 1 [default=  6  ];\n"
+    "  required float  foo = 1 [default=  7.5];\n"
+    "  required float  foo = 1 [default= -8.5];\n"
+    "  required float  foo = 1 [default=  9  ];\n"
+    "  required double foo = 1 [default= 10.5];\n"
+    "  required double foo = 1 [default=-11.5];\n"
+    "  required double foo = 1 [default= 12  ];\n"
+    "  required string foo = 1 [default='13\\001'];\n"
+    "  required bytes  foo = 1 [default='14\\002'];\n"
+    "  required bool   foo = 1 [default=true ];\n"
+    "  required Foo    foo = 1 [default=FOO  ];\n"
+
+    "  required int32  foo = 1 [default= 0x7FFFFFFF];\n"
+    "  required int32  foo = 1 [default=-0x80000000];\n"
+    "  required uint32 foo = 1 [default= 0xFFFFFFFF];\n"
+    "  required int64  foo = 1 [default= 0x7FFFFFFFFFFFFFFF];\n"
+    "  required int64  foo = 1 [default=-0x8000000000000000];\n"
+    "  required uint64 foo = 1 [default= 0xFFFFFFFFFFFFFFFF];\n"
+    "  required double foo = 1 [default= 0xabcd];\n"
+    "}\n",
+
+#define ETC "name:\"foo\" label:LABEL_REQUIRED number:1"
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { type:TYPE_INT32   default_value:\"1\"         "ETC" }"
+    "  field { type:TYPE_INT32   default_value:\"-2\"        "ETC" }"
+    "  field { type:TYPE_INT64   default_value:\"3\"         "ETC" }"
+    "  field { type:TYPE_INT64   default_value:\"-4\"        "ETC" }"
+    "  field { type:TYPE_UINT32  default_value:\"5\"         "ETC" }"
+    "  field { type:TYPE_UINT64  default_value:\"6\"         "ETC" }"
+    "  field { type:TYPE_FLOAT   default_value:\"7.5\"       "ETC" }"
+    "  field { type:TYPE_FLOAT   default_value:\"-8.5\"      "ETC" }"
+    "  field { type:TYPE_FLOAT   default_value:\"9\"         "ETC" }"
+    "  field { type:TYPE_DOUBLE  default_value:\"10.5\"      "ETC" }"
+    "  field { type:TYPE_DOUBLE  default_value:\"-11.5\"     "ETC" }"
+    "  field { type:TYPE_DOUBLE  default_value:\"12\"        "ETC" }"
+    "  field { type:TYPE_STRING  default_value:\"13\\001\"   "ETC" }"
+    "  field { type:TYPE_BYTES   default_value:\"14\\\\002\" "ETC" }"
+    "  field { type:TYPE_BOOL    default_value:\"true\"      "ETC" }"
+    "  field { type_name:\"Foo\" default_value:\"FOO\"       "ETC" }"
+
+    "  field { type:TYPE_INT32   default_value:\"2147483647\"           "ETC" }"
+    "  field { type:TYPE_INT32   default_value:\"-2147483648\"          "ETC" }"
+    "  field { type:TYPE_UINT32  default_value:\"4294967295\"           "ETC" }"
+    "  field { type:TYPE_INT64   default_value:\"9223372036854775807\"  "ETC" }"
+    "  field { type:TYPE_INT64   default_value:\"-9223372036854775808\" "ETC" }"
+    "  field { type:TYPE_UINT64  default_value:\"18446744073709551615\" "ETC" }"
+    "  field { type:TYPE_DOUBLE  default_value:\"43981\"                "ETC" }"
+    "}");
+#undef ETC
+}
+
+TEST_F(ParseMessageTest, FieldOptions) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  optional string foo = 1 [ctype=CORD];\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  field { name:\"foo\" label:LABEL_OPTIONAL type:TYPE_STRING number:1"
+    "          options { ctype:CORD } }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, Group) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  optional group TestGroup = 1 {};\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  nested_type { name: \"TestGroup\" }"
+    "  field { name:\"testgroup\" label:LABEL_OPTIONAL number:1"
+    "          type:TYPE_GROUP type_name: \"TestGroup\" }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, NestedMessage) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  message Nested {}\n"
+    "  optional Nested test_nested = 1;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  nested_type { name: \"Nested\" }"
+    "  field { name:\"test_nested\" label:LABEL_OPTIONAL number:1"
+    "          type_name: \"Nested\" }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, NestedEnum) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  enum NestedEnum {}\n"
+    "  optional NestedEnum test_enum = 1;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  enum_type { name: \"NestedEnum\" }"
+    "  field { name:\"test_enum\" label:LABEL_OPTIONAL number:1"
+    "          type_name: \"NestedEnum\" }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, ExtensionRange) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  extensions 10 to 19;\n"
+    "  extensions 30 to max;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  extension_range { start:10 end:20        }"
+    "  extension_range { start:30 end:536870912 }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, CompoundExtensionRange) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  extensions 2, 15, 9 to 11, 100 to max, 3;\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  extension_range { start:2   end:3         }"
+    "  extension_range { start:15  end:16        }"
+    "  extension_range { start:9   end:12        }"
+    "  extension_range { start:100 end:536870912 }"
+    "  extension_range { start:3   end:4         }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, Extensions) {
+  ExpectParsesTo(
+    "extend Extendee1 { optional int32 foo = 12; }\n"
+    "extend Extendee2 { repeated TestMessage bar = 22; }\n",
+
+    "extension { name:\"foo\" label:LABEL_OPTIONAL type:TYPE_INT32 number:12"
+    "            extendee: \"Extendee1\" } "
+    "extension { name:\"bar\" label:LABEL_REPEATED number:22"
+    "            type_name:\"TestMessage\" extendee: \"Extendee2\" }");
+}
+
+TEST_F(ParseMessageTest, ExtensionsInMessageScope) {
+  ExpectParsesTo(
+    "message TestMessage {\n"
+    "  extend Extendee1 { optional int32 foo = 12; }\n"
+    "  extend Extendee2 { repeated TestMessage bar = 22; }\n"
+    "}\n",
+
+    "message_type {"
+    "  name: \"TestMessage\""
+    "  extension { name:\"foo\" label:LABEL_OPTIONAL type:TYPE_INT32 number:12"
+    "              extendee: \"Extendee1\" }"
+    "  extension { name:\"bar\" label:LABEL_REPEATED number:22"
+    "              type_name:\"TestMessage\" extendee: \"Extendee2\" }"
+    "}");
+}
+
+TEST_F(ParseMessageTest, MultipleExtensionsOneExtendee) {
+  ExpectParsesTo(
+    "extend Extendee1 {\n"
+    "  optional int32 foo = 12;\n"
+    "  repeated TestMessage bar = 22;\n"
+    "}\n",
+
+    "extension { name:\"foo\" label:LABEL_OPTIONAL type:TYPE_INT32 number:12"
+    "            extendee: \"Extendee1\" } "
+    "extension { name:\"bar\" label:LABEL_REPEATED number:22"
+    "            type_name:\"TestMessage\" extendee: \"Extendee1\" }");
+}
+
+// ===================================================================
+
+typedef ParserTest ParseEnumTest;
+
+TEST_F(ParseEnumTest, SimpleEnum) {
+  ExpectParsesTo(
+    "enum TestEnum {\n"
+    "  FOO = 0;\n"
+    "}\n",
+
+    "enum_type {"
+    "  name: \"TestEnum\""
+    "  value { name:\"FOO\" number:0 }"
+    "}");
+}
+
+TEST_F(ParseEnumTest, Values) {
+  ExpectParsesTo(
+    "enum TestEnum {\n"
+    "  FOO = 13;\n"
+    "  BAR = -10;\n"
+    "  BAZ = 500;\n"
+    "}\n",
+
+    "enum_type {"
+    "  name: \"TestEnum\""
+    "  value { name:\"FOO\" number:13 }"
+    "  value { name:\"BAR\" number:-10 }"
+    "  value { name:\"BAZ\" number:500 }"
+    "}");
+}
+
+// ===================================================================
+
+typedef ParserTest ParseServiceTest;
+
+TEST_F(ParseServiceTest, SimpleService) {
+  ExpectParsesTo(
+    "service TestService {\n"
+    "  rpc Foo(In) returns (Out);\n"
+    "}\n",
+
+    "service {"
+    "  name: \"TestService\""
+    "  method { name:\"Foo\" input_type:\"In\" output_type:\"Out\" }"
+    "}");
+}
+
+TEST_F(ParseServiceTest, Methods) {
+  ExpectParsesTo(
+    "service TestService {\n"
+    "  rpc Foo(In1) returns (Out1);\n"
+    "  rpc Bar(In2) returns (Out2);\n"
+    "  rpc Baz(In3) returns (Out3);\n"
+    "}\n",
+
+    "service {"
+    "  name: \"TestService\""
+    "  method { name:\"Foo\" input_type:\"In1\" output_type:\"Out1\" }"
+    "  method { name:\"Bar\" input_type:\"In2\" output_type:\"Out2\" }"
+    "  method { name:\"Baz\" input_type:\"In3\" output_type:\"Out3\" }"
+    "}");
+}
+
+// ===================================================================
+// imports and packages
+
+typedef ParserTest ParseMiscTest;
+
+TEST_F(ParseMiscTest, ParseImport) {
+  ExpectParsesTo(
+    "import \"foo/bar/baz.proto\";\n",
+    "dependency: \"foo/bar/baz.proto\"");
+}
+
+TEST_F(ParseMiscTest, ParseMultipleImports) {
+  ExpectParsesTo(
+    "import \"foo.proto\";\n"
+    "import \"bar.proto\";\n"
+    "import \"baz.proto\";\n",
+    "dependency: \"foo.proto\""
+    "dependency: \"bar.proto\""
+    "dependency: \"baz.proto\"");
+}
+
+TEST_F(ParseMiscTest, ParsePackage) {
+  ExpectParsesTo(
+    "package foo.bar.baz;\n",
+    "package: \"foo.bar.baz\"");
+}
+
+TEST_F(ParseMiscTest, ParsePackageWithSpaces) {
+  ExpectParsesTo(
+    "package foo   .   bar.  \n"
+    "  baz;\n",
+    "package: \"foo.bar.baz\"");
+}
+
+// ===================================================================
+// options
+
+TEST_F(ParseMiscTest, ParseFileOptions) {
+  ExpectParsesTo(
+    "option java_package = \"com.google.foo\";\n"
+    "option optimize_for = CODE_SIZE;",
+
+    "options {"
+    "  java_package: \"com.google.foo\""
+    "  optimize_for: CODE_SIZE"
+    "}");
+}
+
+// TODO(kenton):  We'd like to be able to test all possible option types,
+//   but we are unable to do so here because we can only test the options
+//   that actually exist, which currently doesn't cover everything.  Perhaps
+//   we can solve this in the future by allowing options to be extended, then
+//   defining extensions of every type?
+
+// ===================================================================
+// Error tests
+//
+// There are a very large number of possible errors that the parser could
+// report, so it's infeasible to test every single one of them.  Instead,
+// we test each unique call to AddError() in parser.h.  This does not mean
+// we are testing every possible error that Parser can generate because
+// each variant of the Consume() helper only counts as one unique call to
+// AddError().
+
+typedef ParserTest ParseErrorTest;
+
+TEST_F(ParseErrorTest, MissingSyntaxIdentifier) {
+  require_syntax_identifier_ = true;
+  ExpectHasEarlyExitErrors(
+    "message TestMessage {}",
+    "0:0: File must begin with 'syntax = \"proto2\";'.\n");
+  EXPECT_EQ("", parser_->GetSyntaxIndentifier());
+}
+
+TEST_F(ParseErrorTest, UnknownSyntaxIdentifier) {
+  ExpectHasEarlyExitErrors(
+    "syntax = \"no_such_syntax\";",
+    "0:9: Unrecognized syntax identifier \"no_such_syntax\".  This parser "
+      "only recognizes \"proto2\".\n");
+  EXPECT_EQ("no_such_syntax", parser_->GetSyntaxIndentifier());
+}
+
+TEST_F(ParseErrorTest, SimpleSyntaxError) {
+  ExpectHasErrors(
+    "message TestMessage @#$ { blah }",
+    "0:20: Expected \"{\".\n");
+  EXPECT_EQ("proto2", parser_->GetSyntaxIndentifier());
+}
+
+TEST_F(ParseErrorTest, ExpectedTopLevel) {
+  ExpectHasErrors(
+    "blah;",
+    "0:0: Expected top-level statement (e.g. \"message\").\n");
+}
+
+TEST_F(ParseErrorTest, UnmatchedCloseBrace) {
+  // This used to cause an infinite loop.  Doh.
+  ExpectHasErrors(
+    "}",
+    "0:0: Expected top-level statement (e.g. \"message\").\n"
+    "0:0: Unmatched \"}\".\n");
+}
+
+// -------------------------------------------------------------------
+// Message errors
+
+TEST_F(ParseErrorTest, MessageMissingName) {
+  ExpectHasErrors(
+    "message {}",
+    "0:8: Expected message name.\n");
+}
+
+TEST_F(ParseErrorTest, MessageMissingBody) {
+  ExpectHasErrors(
+    "message TestMessage;",
+    "0:19: Expected \"{\".\n");
+}
+
+TEST_F(ParseErrorTest, EofInMessage) {
+  ExpectHasErrors(
+    "message TestMessage {",
+    "0:21: Reached end of input in message definition (missing '}').\n");
+}
+
+TEST_F(ParseErrorTest, MissingFieldNumber) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional int32 foo;\n"
+    "}\n",
+    "1:20: Missing field number.\n");
+}
+
+TEST_F(ParseErrorTest, ExpectedFieldNumber) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional int32 foo = ;\n"
+    "}\n",
+    "1:23: Expected field number.\n");
+}
+
+TEST_F(ParseErrorTest, FieldNumberOutOfRange) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional int32 foo = 0x100000000;\n"
+    "}\n",
+    "1:23: Integer out of range.\n");
+}
+
+TEST_F(ParseErrorTest, MissingLabel) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  int32 foo = 1;\n"
+    "}\n",
+    "1:2: Expected \"required\", \"optional\", or \"repeated\".\n");
+}
+
+TEST_F(ParseErrorTest, ExpectedOptionName) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [];\n"
+    "}\n",
+    "1:27: Expected option name.\n");
+}
+
+TEST_F(ParseErrorTest, UnknownOption) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [nosuchoption=5];\n"
+    "}\n",
+    "1:27: Unknown option: nosuchoption\n");
+}
+
+TEST_F(ParseErrorTest, DefaultValueTypeMismatch) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [default=true];\n"
+    "}\n",
+    "1:35: Expected integer.\n");
+}
+
+TEST_F(ParseErrorTest, DefaultValueNotBoolean) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional bool foo = 1 [default=blah];\n"
+    "}\n",
+    "1:33: Expected \"true\" or \"false\".\n");
+}
+
+TEST_F(ParseErrorTest, DefaultValueNotString) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional string foo = 1 [default=1];\n"
+    "}\n",
+    "1:35: Expected string.\n");
+}
+
+TEST_F(ParseErrorTest, DefaultValueUnsignedNegative) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [default=-1];\n"
+    "}\n",
+    "1:36: Unsigned field can't have negative default value.\n");
+}
+
+TEST_F(ParseErrorTest, DefaultValueTooLarge) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional int32  foo = 1 [default= 0x80000000];\n"
+    "  optional int32  foo = 1 [default=-0x80000001];\n"
+    "  optional uint32 foo = 1 [default= 0x100000000];\n"
+    "  optional int64  foo = 1 [default= 0x80000000000000000];\n"
+    "  optional int64  foo = 1 [default=-0x80000000000000001];\n"
+    "  optional uint64 foo = 1 [default= 0x100000000000000000];\n"
+    "}\n",
+    "1:36: Integer out of range.\n"
+    "2:36: Integer out of range.\n"
+    "3:36: Integer out of range.\n"
+    "4:36: Integer out of range.\n"
+    "5:36: Integer out of range.\n"
+    "6:36: Integer out of range.\n");
+}
+
+TEST_F(ParseErrorTest, DefaultValueMissing) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [default=];\n"
+    "}\n",
+    "1:35: Expected integer.\n");
+}
+
+TEST_F(ParseErrorTest, DefaultValueForGroup) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional group Foo = 1 [default=blah] {}\n"
+    "}\n",
+    "1:34: Messages can't have default values.\n");
+}
+
+TEST_F(ParseErrorTest, DuplicateDefaultValue) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [default=1,default=2];\n"
+    "}\n",
+    "1:37: Already set option \"default\".\n");
+}
+
+TEST_F(ParseErrorTest, GroupNotCapitalized) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional group foo = 1 {}\n"
+    "}\n",
+    "1:17: Group names must start with a capital letter.\n");
+}
+
+TEST_F(ParseErrorTest, GroupMissingBody) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional group Foo = 1;\n"
+    "}\n",
+    "1:24: Missing group body.\n");
+}
+
+TEST_F(ParseErrorTest, ExtendingPrimitive) {
+  ExpectHasErrors(
+    "extend int32 { optional string foo = 4; }\n",
+    "0:7: Expected message type.\n");
+}
+
+TEST_F(ParseErrorTest, ErrorInExtension) {
+  ExpectHasErrors(
+    "message Foo { extensions 100 to 199; }\n"
+    "extend Foo { optional string foo; }\n",
+    "1:32: Missing field number.\n");
+}
+
+TEST_F(ParseErrorTest, MultipleParseErrors) {
+  // When a statement has a parse error, the parser should be able to continue
+  // parsing at the next statement.
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional int32 foo;\n"
+    "  !invalid statement ending in a block { blah blah { blah } blah }\n"
+    "  optional int32 bar = 3 {}\n"
+    "}\n",
+    "1:20: Missing field number.\n"
+    "2:2: Expected \"required\", \"optional\", or \"repeated\".\n"
+    "2:2: Expected type name.\n"
+    "3:25: Expected \";\".\n");
+}
+
+// -------------------------------------------------------------------
+// Enum errors
+
+TEST_F(ParseErrorTest, EofInEnum) {
+  ExpectHasErrors(
+    "enum TestEnum {",
+    "0:15: Reached end of input in enum definition (missing '}').\n");
+}
+
+TEST_F(ParseErrorTest, EnumValueMissingNumber) {
+  ExpectHasErrors(
+    "enum TestEnum {\n"
+    "  FOO;\n"
+    "}\n",
+    "1:5: Missing numeric value for enum constant.\n");
+}
+
+// -------------------------------------------------------------------
+// Service errors
+
+TEST_F(ParseErrorTest, EofInService) {
+  ExpectHasErrors(
+    "service TestService {",
+    "0:21: Reached end of input in service definition (missing '}').\n");
+}
+
+TEST_F(ParseErrorTest, ServiceMethodPrimitiveParams) {
+  ExpectHasErrors(
+    "service TestService {\n"
+    "  rpc Foo(int32) returns (string);\n"
+    "}\n",
+    "1:10: Expected message type.\n"
+    "1:26: Expected message type.\n");
+}
+
+TEST_F(ParseErrorTest, EofInMethodOptions) {
+  ExpectHasErrors(
+    "service TestService {\n"
+    "  rpc Foo(Bar) returns(Bar) {",
+    "1:29: Reached end of input in method options (missing '}').\n"
+    "1:29: Reached end of input in service definition (missing '}').\n");
+}
+
+TEST_F(ParseErrorTest, PrimitiveMethodInput) {
+  ExpectHasErrors(
+    "service TestService {\n"
+    "  rpc Foo(int32) returns(Bar);\n"
+    "}\n",
+    "1:10: Expected message type.\n");
+}
+
+TEST_F(ParseErrorTest, MethodOptionTypeError) {
+  // This used to cause an infinite loop.
+  ExpectHasErrors(
+    "message Baz {}\n"
+    "service Foo {\n"
+    "  rpc Bar(Baz) returns(Baz) { option invalid syntax; }\n"
+    "}\n",
+    "2:37: Unknown option: invalid\n");
+}
+
+// -------------------------------------------------------------------
+// Import and package errors
+
+TEST_F(ParseErrorTest, ImportNotQuoted) {
+  ExpectHasErrors(
+    "import foo;\n",
+    "0:7: Expected a string naming the file to import.\n");
+}
+
+TEST_F(ParseErrorTest, MultiplePackagesInFile) {
+  ExpectHasErrors(
+    "package foo;\n"
+    "package bar;\n",
+    "1:0: Multiple package definitions.\n");
+}
+
+// -------------------------------------------------------------------
+// Option errors
+
+TEST_F(ParseErrorTest, OptionWrongType) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional string foo = 1 [ctype=1];\n"
+    "}\n",
+    "1:33: Expected enum value.\n");
+}
+
+TEST_F(ParseErrorTest, DupOption) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [ctype=CORD,ctype=CORD];\n"
+    "}\n",
+    "1:38: Option \"ctype\" was already set.\n");
+}
+
+TEST_F(ParseErrorTest, NotMessageOption) {
+  ExpectHasErrors(
+    "message TestMessage {\n"
+    "  optional uint32 foo = 1 [ctype.blah=1];\n"
+    "}\n",
+    "1:32: Option \"ctype\" is an atomic type, not a message.\n");
+}
+
+// TODO(kenton):  Test errors for all possible option types (see TODO above,
+//   under the option parsing tests).
+
+// ===================================================================
+// Test that errors detected by DescriptorPool correctly report line and
+// column numbers.  We have one test for every call to RecordLocation() in
+// parser.cc.
+
+typedef ParserTest ParserValidationErrorTest;
+
+TEST_F(ParserValidationErrorTest, PackageNameError) {
+  // Create another file which defines symbol "foo".
+  FileDescriptorProto other_file;
+  other_file.set_name("bar.proto");
+  other_file.add_message_type()->set_name("foo");
+  EXPECT_TRUE(pool_.BuildFile(other_file) != NULL);
+
+  // Now try to define it as a package.
+  ExpectHasValidationErrors(
+    "package foo.bar;",
+    "0:8: \"foo\" is already defined (as something other than a package) "
+      "in file \"bar.proto\".\n");
+}
+
+TEST_F(ParserValidationErrorTest, MessageNameError) {
+  ExpectHasValidationErrors(
+    "message Foo {}\n"
+    "message Foo {}\n",
+    "1:8: \"Foo\" is already defined.\n");
+}
+
+TEST_F(ParserValidationErrorTest, FieldNameError) {
+  ExpectHasValidationErrors(
+    "message Foo {\n"
+    "  optional int32 bar = 1;\n"
+    "  optional int32 bar = 2;\n"
+    "}\n",
+    "2:17: \"bar\" is already defined in \"Foo\".\n");
+}
+
+TEST_F(ParserValidationErrorTest, FieldTypeError) {
+  ExpectHasValidationErrors(
+    "message Foo {\n"
+    "  optional Baz bar = 1;\n"
+    "}\n",
+    "1:11: \"Baz\" is not defined.\n");
+}
+
+TEST_F(ParserValidationErrorTest, FieldNumberError) {
+  ExpectHasValidationErrors(
+    "message Foo {\n"
+    "  optional int32 bar = 0;\n"
+    "}\n",
+    "1:23: Field numbers must be positive integers.\n");
+}
+
+TEST_F(ParserValidationErrorTest, FieldExtendeeError) {
+  ExpectHasValidationErrors(
+    "extend Baz { optional int32 bar = 1; }\n",
+    "0:7: \"Baz\" is not defined.\n");
+}
+
+TEST_F(ParserValidationErrorTest, FieldDefaultValueError) {
+  ExpectHasValidationErrors(
+    "enum Baz { QUX = 1; }\n"
+    "message Foo {\n"
+    "  optional Baz bar = 1 [default=NO_SUCH_VALUE];\n"
+    "}\n",
+    "2:32: Enum type \"Baz\" has no value named \"NO_SUCH_VALUE\".\n");
+}
+
+TEST_F(ParserValidationErrorTest, ExtensionRangeNumberError) {
+  ExpectHasValidationErrors(
+    "message Foo {\n"
+    "  extensions 0;\n"
+    "}\n",
+    "1:13: Extension numbers must be positive integers.\n");
+}
+
+TEST_F(ParserValidationErrorTest, EnumNameError) {
+  ExpectHasValidationErrors(
+    "enum Foo {A = 1;}\n"
+    "enum Foo {B = 1;}\n",
+    "1:5: \"Foo\" is already defined.\n");
+}
+
+TEST_F(ParserValidationErrorTest, EnumValueNameError) {
+  ExpectHasValidationErrors(
+    "enum Foo {\n"
+    "  BAR = 1;\n"
+    "  BAR = 1;\n"
+    "}\n",
+    "2:2: \"BAR\" is already defined.\n");
+}
+
+TEST_F(ParserValidationErrorTest, ServiceNameError) {
+  ExpectHasValidationErrors(
+    "service Foo {}\n"
+    "service Foo {}\n",
+    "1:8: \"Foo\" is already defined.\n");
+}
+
+TEST_F(ParserValidationErrorTest, MethodNameError) {
+  ExpectHasValidationErrors(
+    "message Baz {}\n"
+    "service Foo {\n"
+    "  rpc Bar(Baz) returns(Baz);\n"
+    "  rpc Bar(Baz) returns(Baz);\n"
+    "}\n",
+    "3:6: \"Bar\" is already defined in \"Foo\".\n");
+}
+
+TEST_F(ParserValidationErrorTest, MethodInputTypeError) {
+  ExpectHasValidationErrors(
+    "message Baz {}\n"
+    "service Foo {\n"
+    "  rpc Bar(Qux) returns(Baz);\n"
+    "}\n",
+    "2:10: \"Qux\" is not defined.\n");
+}
+
+TEST_F(ParserValidationErrorTest, MethodOutputTypeError) {
+  ExpectHasValidationErrors(
+    "message Baz {}\n"
+    "service Foo {\n"
+    "  rpc Bar(Baz) returns(Qux);\n"
+    "}\n",
+    "2:23: \"Qux\" is not defined.\n");
+}
+
+// ===================================================================
+// Test that the output from FileDescriptor::DebugString() (and all other
+// descriptor types) is parseable, and results in the same Descriptor
+// definitions again afoter parsing (not, however, that the order of messages
+// cannot be guaranteed to be the same)
+
+typedef ParserTest ParseDecriptorDebugTest;
+
+class CompareDescriptorNames {
+ public:
+  bool operator()(const DescriptorProto* left, const DescriptorProto* right) {
+    return left->name() < right->name();
+  }
+};
+
+// Sorts nested DescriptorProtos of a DescriptoProto, by name.
+void SortMessages(DescriptorProto *descriptor_proto) {
+  int size = descriptor_proto->nested_type_size();
+  // recursively sort; we can't guarantee the order of nested messages either
+  for (int i = 0; i < size; ++i) {
+    SortMessages(descriptor_proto->mutable_nested_type(i));
+  }
+  DescriptorProto **data =
+    descriptor_proto->mutable_nested_type()->mutable_data();
+  sort(data, data + size, CompareDescriptorNames());
+}
+
+// Sorts DescriptorProtos belonging to a FileDescriptorProto, by name.
+void SortMessages(FileDescriptorProto *file_descriptor_proto) {
+  int size = file_descriptor_proto->message_type_size();
+  // recursively sort; we can't guarantee the order of nested messages either
+  for (int i = 0; i < size; ++i) {
+    SortMessages(file_descriptor_proto->mutable_message_type(i));
+  }
+  DescriptorProto **data =
+    file_descriptor_proto->mutable_message_type()->mutable_data();
+  sort(data, data + size, CompareDescriptorNames());
+}
+
+TEST_F(ParseDecriptorDebugTest, TestAllDescriptorTypes) {
+  const FileDescriptor* original_file =
+     protobuf_unittest::TestAllTypes::descriptor()->file();
+  FileDescriptorProto expected;
+  original_file->CopyTo(&expected);
+
+  // Get the DebugString of the unittest.proto FileDecriptor, which includes
+  // all other descriptor types
+  string debug_string = original_file->DebugString();
+
+  // Parse the debug string
+  SetupParser(debug_string.c_str());
+  FileDescriptorProto parsed;
+  parser_->Parse(input_.get(), &parsed);
+  EXPECT_EQ(io::Tokenizer::TYPE_END, input_->current().type);
+  ASSERT_EQ("", error_collector_.text_);
+
+  // We now have a FileDescriptorProto, but to compare with the expected we
+  // need to link to a FileDecriptor, then output back to a proto. We'll
+  // also need to give it the same name as the original.
+  parsed.set_name("google/protobuf/unittest.proto");
+  // We need the imported dependency before we can build our parsed proto
+  const FileDescriptor* import =
+       protobuf_unittest_import::ImportMessage::descriptor()->file();
+  FileDescriptorProto import_proto;
+  import->CopyTo(&import_proto);
+  ASSERT_TRUE(pool_.BuildFile(import_proto) != NULL);
+  const FileDescriptor* actual = pool_.BuildFile(parsed);
+  parsed.Clear();
+  actual->CopyTo(&parsed);
+  ASSERT_TRUE(actual != NULL);
+
+  // The messages might be in different orders, making them hard to compare.
+  // So, sort the messages in the descriptor protos (including nested messages,
+  // recursively).
+  SortMessages(&expected);
+  SortMessages(&parsed);
+
+  // I really wanted to use StringDiff here for the debug output on fail,
+  // but the strings are too long for it, and if I increase its max size,
+  // we get a memory allocation failure :(
+  EXPECT_EQ(expected.DebugString(), parsed.DebugString());
+}
+
+// ===================================================================
+
+}  // anonymous namespace
+
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/python/python_generator.cc b/src/google/protobuf/compiler/python/python_generator.cc
new file mode 100644
index 00000000..e1171382
--- /dev/null
+++ b/src/google/protobuf/compiler/python/python_generator.cc
@@ -0,0 +1,794 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: robinson@google.com (Will Robinson)
+//
+// This module outputs pure-Python protocol message classes that will
+// largely be constructed at runtime via the metaclass in reflection.py.
+// In other words, our job is basically to output a Python equivalent
+// of the C++ *Descriptor objects, and fix up all circular references
+// within these objects.
+//
+// Note that the runtime performance of protocol message classes created in
+// this way is expected to be lousy.  The plan is to create an alternate
+// generator that outputs a Python/C extension module that lets
+// performance-minded Python code leverage the fast C++ implementation
+// directly.
+
+#include <utility>
+#include <map>
+#include <string>
+#include <vector>
+
+#include <google/protobuf/compiler/python/python_generator.h>
+#include <google/protobuf/descriptor.pb.h>
+
+#include <google/protobuf/stubs/common.h>
+#include <google/protobuf/io/printer.h>
+#include <google/protobuf/descriptor.h>
+#include <google/protobuf/io/zero_copy_stream.h>
+#include <google/protobuf/stubs/strutil.h>
+#include <google/protobuf/stubs/substitute.h>
+
+namespace google {
+namespace protobuf {
+namespace compiler {
+namespace python {
+
+namespace {
+
+// Returns a copy of |filename| with any trailing ".protodevel" or ".proto
+// suffix stripped.
+// TODO(robinson): Unify with copy in compiler/cpp/internal/helpers.cc.
+string StripProto(const string& filename) {
+  const char* suffix = HasSuffixString(filename, ".protodevel")
+      ? ".protodevel" : ".proto";
+  return StripSuffixString(filename, suffix);
+}
+
+
+// Returns the Python module name expected for a given .proto filename.
+string ModuleName(const string& filename) {
+  string basename = StripProto(filename);
+  StripString(&basename, "-", '_');
+  StripString(&basename, "/", '.');
+  return basename + "_pb2";
+}
+
+
+// Returns the name of all containing types for descriptor,
+// in order from outermost to innermost, followed by descriptor's
+// own name.  Each name is separated by |separator|.
+template <typename DescriptorT>
+string NamePrefixedWithNestedTypes(const DescriptorT& descriptor,
+                                   const string& separator) {
+  string name = descriptor.name();
+  for (const Descriptor* current = descriptor.containing_type();
+       current != NULL; current = current->containing_type()) {
+    name = current->name() + separator + name;
+  }
+  return name;
+}
+
+
+// Name of the class attribute where we store the Python
+// descriptor.Descriptor instance for the generated class.
+// Must stay consistent with the _DESCRIPTOR_KEY constant
+// in proto2/public/reflection.py.
+const char kDescriptorKey[] = "DESCRIPTOR";
+
+
+// Prints the common boilerplate needed at the top of every .py
+// file output by this generator.
+void PrintTopBoilerplate(
+    io::Printer* printer, const FileDescriptor* file, bool descriptor_proto) {
+  // TODO(robinson): Allow parameterization of Python version?
+  printer->Print(
+      "#!/usr/bin/python2.4\n"
+      "# Generated by the protocol buffer compiler.  DO NOT EDIT!\n"
+      "\n"
+      "from google.protobuf import descriptor\n"
+      "from google.protobuf import message\n"
+      "from google.protobuf import reflection\n"
+      "from google.protobuf import service\n"
+      "from google.protobuf import service_reflection\n");
+  // Avoid circular imports if this module is descriptor_pb2.
+  if (!descriptor_proto) {
+    printer->Print(
+        "from google.protobuf import descriptor_pb2\n");
+  }
+}
+
+
+// Returns a Python literal giving the default value for a field.
+// If the field specifies no explicit default value, we'll return
+// the default default value for the field type (zero for numbers,
+// empty string for strings, empty list for repeated fields, and
+// None for non-repeated, composite fields).
+//
+// TODO(robinson): Unify with code from
+// //compiler/cpp/internal/primitive_field.cc
+// //compiler/cpp/internal/enum_field.cc
+// //compiler/cpp/internal/string_field.cc
+string StringifyDefaultValue(const FieldDescriptor& field) {
+  if (field.is_repeated()) {
+    return "[]";
+  }
+
+  switch (field.cpp_type()) {
+    case FieldDescriptor::CPPTYPE_INT32:
+      return SimpleItoa(field.default_value_int32());
+    case FieldDescriptor::CPPTYPE_UINT32:
+      return SimpleItoa(field.default_value_uint32());
+    case FieldDescriptor::CPPTYPE_INT64:
+      return SimpleItoa(field.default_value_int64());
+    case FieldDescriptor::CPPTYPE_UINT64:
+      return SimpleItoa(field.default_value_uint64());
+    case FieldDescriptor::CPPTYPE_DOUBLE:
+      return SimpleDtoa(field.default_value_double());
+    case FieldDescriptor::CPPTYPE_FLOAT:
+      return SimpleFtoa(field.default_value_float());
+    case FieldDescriptor::CPPTYPE_BOOL:
+      return field.default_value_bool() ? "True" : "False";
+    case FieldDescriptor::CPPTYPE_ENUM:
+      return SimpleItoa(field.default_value_enum()->number());
+    case FieldDescriptor::CPPTYPE_STRING:
+      return "\"" + CEscape(field.default_value_string()) + "\"";
+    case FieldDescriptor::CPPTYPE_MESSAGE:
+      return "None";
+  }
+  // (We could add a default case above but then we wouldn't get the nice
+  // compiler warning when a new type is added.)
+  GOOGLE_LOG(FATAL) << "Not reached.";
+  return "";
+}
+
+
+
+}  // namespace
+
+
+Generator::Generator() : file_(NULL) {
+}
+
+Generator::~Generator() {
+}
+
+bool Generator::Generate(const FileDescriptor* file,
+                         const string& parameter,
+                         OutputDirectory* output_directory,
+                         string* error) const {
+
+  // Completely serialize all Generate() calls on this instance.  The
+  // thread-safety constraints of the CodeGenerator interface aren't clear so
+  // just be as conservative as possible.  It's easier to relax this later if
+  // we need to, but I doubt it will be an issue.
+  // TODO(kenton):  The proper thing to do would be to allocate any state on
+  //   the stack and use that, so that the Generator class itself does not need
+  //   to have any mutable members.  Then it is implicitly thread-safe.
+  MutexLock lock(&mutex_);
+  file_ = file;
+  string module_name = ModuleName(file->name());
+  string filename = module_name;
+  StripString(&filename, ".", '/');
+  filename += ".py";
+
+
+  scoped_ptr<io::ZeroCopyOutputStream> output(output_directory->Open(filename));
+  GOOGLE_CHECK(output.get());
+  io::Printer printer(output.get(), '$');
+  printer_ = &printer;
+
+  PrintTopBoilerplate(printer_, file_, GeneratingDescriptorProto());
+  PrintTopLevelEnums();
+  PrintTopLevelExtensions();
+  PrintAllNestedEnumsInFile();
+  PrintMessageDescriptors();
+  // We have to print the imports after the descriptors, so that mutually
+  // recursive protos in separate files can successfully reference each other.
+  PrintImports();
+  FixForeignFieldsInDescriptors();
+  PrintMessages();
+  // We have to fix up the extensions after the message classes themselves,
+  // since they need to call static RegisterExtension() methods on these
+  // classes.
+  FixForeignFieldsInExtensions();
+  PrintServices();
+  return !printer.failed();
+}
+
+// Prints Python imports for all modules imported by |file|.
+void Generator::PrintImports() const {
+  for (int i = 0; i < file_->dependency_count(); ++i) {
+    string module_name = ModuleName(file_->dependency(i)->name());
+    printer_->Print("import $module$\n", "module",
+                    module_name);
+  }
+  printer_->Print("\n");
+}
+
+// Prints descriptors and module-level constants for all top-level
+// enums defined in |file|.
+void Generator::PrintTopLevelEnums() const {
+  vector<pair<string, int> > top_level_enum_values;
+  for (int i = 0; i < file_->enum_type_count(); ++i) {
+    const EnumDescriptor& enum_descriptor = *file_->enum_type(i);
+    PrintEnum(enum_descriptor);
+    printer_->Print("\n");
+
+    for (int j = 0; j < enum_descriptor.value_count(); ++j) {
+      const EnumValueDescriptor& value_descriptor = *enum_descriptor.value(j);
+      top_level_enum_values.push_back(
+          make_pair(value_descriptor.name(), value_descriptor.number()));
+    }
+  }
+
+  for (int i = 0; i < top_level_enum_values.size(); ++i) {
+    printer_->Print("$name$ = $value$\n",
+                    "name", top_level_enum_values[i].first,
+                    "value", SimpleItoa(top_level_enum_values[i].second));
+  }
+  printer_->Print("\n");
+}
+
+// Prints all enums contained in all message types in |file|.
+void Generator::PrintAllNestedEnumsInFile() const {
+  for (int i = 0; i < file_->message_type_count(); ++i) {
+    PrintNestedEnums(*file_->message_type(i));
+  }
+}
+
+// Prints a Python statement assigning the appropriate module-level
+// enum name to a Python EnumDescriptor object equivalent to
+// enum_descriptor.
+void Generator::PrintEnum(const EnumDescriptor& enum_descriptor) const {
+  map<string, string> m;
+  m["descriptor_name"] = ModuleLevelDescriptorName(enum_descriptor);
+  m["name"] = enum_descriptor.name();
+  m["full_name"] = enum_descriptor.full_name();
+  m["filename"] = enum_descriptor.name();
+  const char enum_descriptor_template[] =
+      "$descriptor_name$ = descriptor.EnumDescriptor(\n"
+      "  name='$name$',\n"
+      "  full_name='$full_name$',\n"
+      "  filename='$filename$',\n"
+      "  values=[\n";
+  string options_string;
+  enum_descriptor.options().SerializeToString(&options_string);
+  printer_->Print(m, enum_descriptor_template);
+  printer_->Indent();
+  printer_->Indent();
+  for (int i = 0; i < enum_descriptor.value_count(); ++i) {
+    PrintEnumValueDescriptor(*enum_descriptor.value(i));
+    printer_->Print(",\n");
+  }
+  printer_->Outdent();
+  printer_->Print("],\n");
+  printer_->Print("options=$options_value$,\n",
+                  "options_value",
+                  OptionsValue("EnumOptions", CEscape(options_string)));
+  printer_->Outdent();
+  printer_->Print(")\n");
+  printer_->Print("\n");
+}
+
+// Recursively prints enums in nested types within descriptor, then
+// prints enums contained at the top level in descriptor.
+void Generator::PrintNestedEnums(const Descriptor& descriptor) const {
+  for (int i = 0; i < descriptor.nested_type_count(); ++i) {
+    PrintNestedEnums(*descriptor.nested_type(i));
+  }
+
+  for (int i = 0; i < descriptor.enum_type_count(); ++i) {
+    PrintEnum(*descriptor.enum_type(i));
+  }
+}
+
+void Generator::PrintTopLevelExtensions() const {
+  const bool is_extension = true;
+  for (int i = 0; i < file_->extension_count(); ++i) {
+    const FieldDescriptor& extension_field = *file_->extension(i);
+    printer_->Print("$name$ = ", "name", extension_field.name());
+    PrintFieldDescriptor(extension_field, is_extension);
+    printer_->Print("\n");
+  }
+  printer_->Print("\n");
+}
+
+// Prints Python equivalents of all Descriptors in |file|.
+void Generator::PrintMessageDescriptors() const {
+  for (int i = 0; i < file_->message_type_count(); ++i) {
+    PrintDescriptor(*file_->message_type(i));
+    printer_->Print("\n");
+  }
+}
+
+void Generator::PrintServices() const {
+  for (int i = 0; i < file_->service_count(); ++i) {
+    PrintServiceDescriptor(*file_->service(i));
+    PrintServiceClass(*file_->service(i));
+    PrintServiceStub(*file_->service(i));
+    printer_->Print("\n");
+  }
+}
+
+void Generator::PrintServiceDescriptor(
+    const ServiceDescriptor& descriptor) const {
+  printer_->Print("\n");
+  string service_name = ModuleLevelServiceDescriptorName(descriptor);
+  string options_string;
+  descriptor.options().SerializeToString(&options_string);
+
+  printer_->Print(
+      "$service_name$ = descriptor.ServiceDescriptor(\n",
+      "service_name", service_name);
+  printer_->Indent();
+  map<string, string> m;
+  m["name"] = descriptor.name();
+  m["full_name"] = descriptor.full_name();
+  m["index"] = SimpleItoa(descriptor.index());
+  m["options_value"] = OptionsValue("ServiceOptions", options_string);
+  const char required_function_arguments[] =
+      "name='$name$',\n"
+      "full_name='$full_name$',\n"
+      "index=$index$,\n"
+      "options=$options_value$,\n"
+      "methods=[\n";
+  printer_->Print(m, required_function_arguments);
+  for (int i = 0; i < descriptor.method_count(); ++i) {
+    const MethodDescriptor* method = descriptor.method(i);
+    string options_string;
+    method->options().SerializeToString(&options_string);
+
+    m.clear();
+    m["name"] = method->name();
+    m["full_name"] = method->full_name();
+    m["index"] = SimpleItoa(method->index());
+    m["serialized_options"] = CEscape(options_string);
+    m["input_type"] = ModuleLevelDescriptorName(*(method->input_type()));
+    m["output_type"] = ModuleLevelDescriptorName(*(method->output_type()));
+    m["options_value"] = OptionsValue("MethodOptions", options_string);
+    printer_->Print("descriptor.MethodDescriptor(\n");
+    printer_->Indent();
+    printer_->Print(
+        m,
+        "name='$name$',\n"
+        "full_name='$full_name$',\n"
+        "index=$index$,\n"
+        "containing_service=None,\n"
+        "input_type=$input_type$,\n"
+        "output_type=$output_type$,\n"
+        "options=$options_value$,\n");
+    printer_->Outdent();
+    printer_->Print("),\n");
+  }
+
+  printer_->Outdent();
+  printer_->Print("])\n\n");
+}
+
+void Generator::PrintServiceClass(const ServiceDescriptor& descriptor) const {
+  // Print the service.
+  printer_->Print("class $class_name$(service.Service):\n",
+                  "class_name", descriptor.name());
+  printer_->Indent();
+  printer_->Print(
+      "__metaclass__ = service_reflection.GeneratedServiceType\n"
+      "$descriptor_key$ = $descriptor_name$\n",
+      "descriptor_key", kDescriptorKey,
+      "descriptor_name", ModuleLevelServiceDescriptorName(descriptor));
+  printer_->Outdent();
+}
+
+void Generator::PrintServiceStub(const ServiceDescriptor& descriptor) const {
+  // Print the service stub.
+  printer_->Print("class $class_name$_Stub($class_name$):\n",
+                  "class_name", descriptor.name());
+  printer_->Indent();
+  printer_->Print(
+      "__metaclass__ = service_reflection.GeneratedServiceStubType\n"
+      "$descriptor_key$ = $descriptor_name$\n",
+      "descriptor_key", kDescriptorKey,
+      "descriptor_name", ModuleLevelServiceDescriptorName(descriptor));
+  printer_->Outdent();
+}
+
+// Prints statement assigning ModuleLevelDescriptorName(message_descriptor)
+// to a Python Descriptor object for message_descriptor.
+//
+// Mutually recursive with PrintNestedDescriptors().
+void Generator::PrintDescriptor(const Descriptor& message_descriptor) const {
+  PrintNestedDescriptors(message_descriptor);
+
+  printer_->Print("\n");
+  printer_->Print("$descriptor_name$ = descriptor.Descriptor(\n",
+                  "descriptor_name",
+                  ModuleLevelDescriptorName(message_descriptor));
+  printer_->Indent();
+  map<string, string> m;
+  m["name"] = message_descriptor.name();
+  m["full_name"] = message_descriptor.full_name();
+  m["filename"] = message_descriptor.file()->name();
+  const char required_function_arguments[] =
+      "name='$name$',\n"
+      "full_name='$full_name$',\n"
+      "filename='$filename$',\n"
+      "containing_type=None,\n";  // TODO(robinson): Implement containing_type.
+  printer_->Print(m, required_function_arguments);
+  PrintFieldsInDescriptor(message_descriptor);
+  PrintExtensionsInDescriptor(message_descriptor);
+  // TODO(robinson): implement printing of nested_types.
+  printer_->Print("nested_types=[],  # TODO(robinson): Implement.\n");
+  printer_->Print("enum_types=[\n");
+  printer_->Indent();
+  for (int i = 0; i < message_descriptor.enum_type_count(); ++i) {
+    const string descriptor_name = ModuleLevelDescriptorName(
+        *message_descriptor.enum_type(i));
+    printer_->Print(descriptor_name.c_str());
+    printer_->Print(",\n");
+  }
+  printer_->Outdent();
+  printer_->Print("],\n");
+  string options_string;
+  message_descriptor.options().SerializeToString(&options_string);
+  printer_->Print(
+      "options=$options_value$",
+      "options_value", OptionsValue("MessageOptions", options_string));
+  printer_->Outdent();
+  printer_->Print(")\n");
+}
+
+// Prints Python Descriptor objects for all nested types contained in
+// message_descriptor.
+//
+// Mutually recursive with PrintDescriptor().
+void Generator::PrintNestedDescriptors(
+    const Descriptor& containing_descriptor) const {
+  for (int i = 0; i < containing_descriptor.nested_type_count(); ++i) {
+    PrintDescriptor(*containing_descriptor.nested_type(i));
+  }
+}
+
+// Prints all messages in |file|.
+void Generator::PrintMessages() const {
+  for (int i = 0; i < file_->message_type_count(); ++i) {
+    PrintMessage(*file_->message_type(i));
+    printer_->Print("\n");
+  }
+}
+
+// Prints a Python class for the given message descriptor.  We defer to the
+// metaclass to do almost all of the work of actually creating a useful class.
+// The purpose of this function and its many helper functions above is merely
+// to output a Python version of the descriptors, which the metaclass in
+// reflection.py will use to construct the meat of the class itself.
+//
+// Mutually recursive with PrintNestedMessages().
+void Generator::PrintMessage(
+    const Descriptor& message_descriptor) const {
+  printer_->Print("class $name$(message.Message):\n", "name",
+                  message_descriptor.name());
+  printer_->Indent();
+  printer_->Print("__metaclass__ = reflection.GeneratedProtocolMessageType\n");
+  PrintNestedMessages(message_descriptor);
+  map<string, string> m;
+  m["descriptor_key"] = kDescriptorKey;
+  m["descriptor_name"] = ModuleLevelDescriptorName(message_descriptor);
+  printer_->Print(m, "$descriptor_key$ = $descriptor_name$\n");
+  printer_->Outdent();
+}
+
+// Prints all nested messages within |containing_descriptor|.
+// Mutually recursive with PrintMessage().
+void Generator::PrintNestedMessages(
+    const Descriptor& containing_descriptor) const {
+  for (int i = 0; i < containing_descriptor.nested_type_count(); ++i) {
+    printer_->Print("\n");
+    PrintMessage(*containing_descriptor.nested_type(i));
+  }
+}
+
+// Recursively fixes foreign fields in all nested types in |descriptor|, then
+// sets the message_type and enum_type of all message and enum fields to point
+// to their respective descriptors.
+void Generator::FixForeignFieldsInDescriptor(
+    const Descriptor& descriptor) const {
+  for (int i = 0; i < descriptor.nested_type_count(); ++i) {
+    FixForeignFieldsInDescriptor(*descriptor.nested_type(i));
+  }
+
+  for (int i = 0; i < descriptor.field_count(); ++i) {
+    const FieldDescriptor& field_descriptor = *descriptor.field(i);
+    FixForeignFieldsInField(&descriptor, field_descriptor, "fields_by_name");
+  }
+}
+
+// Sets any necessary message_type and enum_type attributes
+// for the Python version of |field|.
+//
+// containing_type may be NULL, in which case this is a module-level field.
+//
+// python_dict_name is the name of the Python dict where we should
+// look the field up in the containing type.  (e.g., fields_by_name
+// or extensions_by_name).  We ignore python_dict_name if containing_type
+// is NULL.
+void Generator::FixForeignFieldsInField(const Descriptor* containing_type,
+                                        const FieldDescriptor& field,
+                                        const string& python_dict_name) const {
+  const string field_referencing_expression = FieldReferencingExpression(
+      containing_type, field, python_dict_name);
+  map<string, string> m;
+  m["field_ref"] = field_referencing_expression;
+  const Descriptor* foreign_message_type = field.message_type();
+  if (foreign_message_type) {
+    m["foreign_type"] = ModuleLevelDescriptorName(*foreign_message_type);
+    printer_->Print(m, "$field_ref$.message_type = $foreign_type$\n");
+  }
+  const EnumDescriptor* enum_type = field.enum_type();
+  if (enum_type) {
+    m["enum_type"] = ModuleLevelDescriptorName(*enum_type);
+    printer_->Print(m, "$field_ref$.enum_type = $enum_type$\n");
+  }
+}
+
+// Returns the module-level expression for the given FieldDescriptor.
+// Only works for fields in the .proto file this Generator is generating for.
+//
+// containing_type may be NULL, in which case this is a module-level field.
+//
+// python_dict_name is the name of the Python dict where we should
+// look the field up in the containing type.  (e.g., fields_by_name
+// or extensions_by_name).  We ignore python_dict_name if containing_type
+// is NULL.
+string Generator::FieldReferencingExpression(
+    const Descriptor* containing_type,
+    const FieldDescriptor& field,
+    const string& python_dict_name) const {
+  // We should only ever be looking up fields in the current file.
+  // The only things we refer to from other files are message descriptors.
+  GOOGLE_CHECK_EQ(field.file(), file_) << field.file()->name() << " vs. "
+                                << file_->name();
+  if (!containing_type) {
+    return field.name();
+  }
+  return strings::Substitute(
+      "$0.$1['$2']",
+      ModuleLevelDescriptorName(*containing_type),
+      python_dict_name, field.name());
+}
+
+// Prints statements setting the message_type and enum_type fields in the
+// Python descriptor objects we've already output in ths file.  We must
+// do this in a separate step due to circular references (otherwise, we'd
+// just set everything in the initial assignment statements).
+void Generator::FixForeignFieldsInDescriptors() const {
+  for (int i = 0; i < file_->message_type_count(); ++i) {
+    FixForeignFieldsInDescriptor(*file_->message_type(i));
+  }
+  printer_->Print("\n");
+}
+
+// We need to not only set any necessary message_type fields, but
+// also need to call RegisterExtension() on each message we're
+// extending.
+void Generator::FixForeignFieldsInExtensions() const {
+  // Top-level extensions.
+  for (int i = 0; i < file_->extension_count(); ++i) {
+    FixForeignFieldsInExtension(*file_->extension(i));
+  }
+  // Nested extensions.
+  for (int i = 0; i < file_->message_type_count(); ++i) {
+    FixForeignFieldsInNestedExtensions(*file_->message_type(i));
+  }
+}
+
+void Generator::FixForeignFieldsInExtension(
+    const FieldDescriptor& extension_field) const {
+  GOOGLE_CHECK(extension_field.is_extension());
+  // extension_scope() will be NULL for top-level extensions, which is
+  // exactly what FixForeignFieldsInField() wants.
+  FixForeignFieldsInField(extension_field.extension_scope(), extension_field,
+                          "extensions_by_name");
+
+  map<string, string> m;
+  // Confusingly, for FieldDescriptors that happen to be extensions,
+  // containing_type() means "extended type."
+  // On the other hand, extension_scope() will give us what we normally
+  // mean by containing_type().
+  m["extended_message_class"] = ModuleLevelMessageName(
+      *extension_field.containing_type());
+  m["field"] = FieldReferencingExpression(extension_field.extension_scope(),
+                                          extension_field,
+                                          "extensions_by_name");
+  printer_->Print(m, "$extended_message_class$.RegisterExtension($field$)\n");
+}
+
+void Generator::FixForeignFieldsInNestedExtensions(
+    const Descriptor& descriptor) const {
+  // Recursively fix up extensions in all nested types.
+  for (int i = 0; i < descriptor.nested_type_count(); ++i) {
+    FixForeignFieldsInNestedExtensions(*descriptor.nested_type(i));
+  }
+  // Fix up extensions directly contained within this type.
+  for (int i = 0; i < descriptor.extension_count(); ++i) {
+    FixForeignFieldsInExtension(*descriptor.extension(i));
+  }
+}
+
+// Returns a Python expression that instantiates a Python EnumValueDescriptor
+// object for the given C++ descriptor.
+void Generator::PrintEnumValueDescriptor(
+    const EnumValueDescriptor& descriptor) const {
+  // TODO(robinson): Fix up EnumValueDescriptor "type" fields.
+  // More circular references.  ::sigh::
+  string options_string;
+  descriptor.options().SerializeToString(&options_string);
+  map<string, string> m;
+  m["name"] = descriptor.name();
+  m["index"] = SimpleItoa(descriptor.index());
+  m["number"] = SimpleItoa(descriptor.number());
+  m["options"] = OptionsValue("EnumValueOptions", options_string);
+  printer_->Print(
+      m,
+      "descriptor.EnumValueDescriptor(\n"
+      "  name='$name$', index=$index$, number=$number$,\n"
+      "  options=$options$,\n"
+      "  type=None)");
+}
+
+string Generator::OptionsValue(
+    const string& class_name, const string& serialized_options) const {
+  if (serialized_options.length() == 0 || GeneratingDescriptorProto()) {
+    return "None";
+  } else {
+    string full_class_name = "descriptor_pb2." + class_name;
+    return "descriptor._ParseOptions(" + full_class_name + "(), '"
+        + CEscape(serialized_options)+ "')";
+  }
+}
+
+// Prints an expression for a Python FieldDescriptor for |field|.
+void Generator::PrintFieldDescriptor(
+    const FieldDescriptor& field, bool is_extension) const {
+  string options_string;
+  field.options().SerializeToString(&options_string);
+  map<string, string> m;
+  m["name"] = field.name();
+  m["full_name"] = field.full_name();
+  m["index"] = SimpleItoa(field.index());
+  m["number"] = SimpleItoa(field.number());
+  m["type"] = SimpleItoa(field.type());
+  m["cpp_type"] = SimpleItoa(field.cpp_type());
+  m["label"] = SimpleItoa(field.label());
+  m["default_value"] = StringifyDefaultValue(field);
+  m["is_extension"] = is_extension ? "True" : "False";
+  m["options"] = OptionsValue("FieldOptions", options_string);
+  // We always set message_type and enum_type to None at this point, and then
+  // these fields in correctly after all referenced descriptors have been
+  // defined and/or imported (see FixForeignFieldsInDescriptors()).
+  const char field_descriptor_decl[] =
+      "descriptor.FieldDescriptor(\n"
+      "  name='$name$', full_name='$full_name$', index=$index$,\n"
+      "  number=$number$, type=$type$, cpp_type=$cpp_type$, label=$label$,\n"
+      "  default_value=$default_value$,\n"
+      "  message_type=None, enum_type=None, containing_type=None,\n"
+      "  is_extension=$is_extension$, extension_scope=None,\n"
+      "  options=$options$)";
+  printer_->Print(m, field_descriptor_decl);
+}
+
+// Helper for Print{Fields,Extensions}InDescriptor().
+void Generator::PrintFieldDescriptorsInDescriptor(
+    const Descriptor& message_descriptor,
+    bool is_extension,
+    const string& list_variable_name,
+    int (Descriptor::*CountFn)() const,
+    const FieldDescriptor* (Descriptor::*GetterFn)(int) const) const {
+  printer_->Print("$list$=[\n", "list", list_variable_name);
+  printer_->Indent();
+  for (int i = 0; i < (message_descriptor.*CountFn)(); ++i) {
+    PrintFieldDescriptor(*(message_descriptor.*GetterFn)(i),
+                         is_extension);
+    printer_->Print(",\n");
+  }
+  printer_->Outdent();
+  printer_->Print("],\n");
+}
+
+// Prints a statement assigning "fields" to a list of Python FieldDescriptors,
+// one for each field present in message_descriptor.
+void Generator::PrintFieldsInDescriptor(
+    const Descriptor& message_descriptor) const {
+  const bool is_extension = false;
+  PrintFieldDescriptorsInDescriptor(
+      message_descriptor, is_extension, "fields",
+      &Descriptor::field_count, &Descriptor::field);
+}
+
+// Prints a statement assigning "extensions" to a list of Python
+// FieldDescriptors, one for each extension present in message_descriptor.
+void Generator::PrintExtensionsInDescriptor(
+    const Descriptor& message_descriptor) const {
+  const bool is_extension = true;
+  PrintFieldDescriptorsInDescriptor(
+      message_descriptor, is_extension, "extensions",
+      &Descriptor::extension_count, &Descriptor::extension);
+}
+
+bool Generator::GeneratingDescriptorProto() const {
+  return file_->name() == "google/protobuf/descriptor.proto";
+}
+
+// Returns the unique Python module-level identifier given to a descriptor.
+// This name is module-qualified iff the given descriptor describes an
+// entity that doesn't come from the current file.
+template <typename DescriptorT>
+string Generator::ModuleLevelDescriptorName(
+    const DescriptorT& descriptor) const {
+  // FIXME(robinson):
+  // We currently don't worry about collisions with underscores in the type
+  // names, so these would collide in nasty ways if found in the same file:
+  //   OuterProto.ProtoA.ProtoB
+  //   OuterProto_ProtoA.ProtoB  # Underscore instead of period.
+  // As would these:
+  //   OuterProto.ProtoA_.ProtoB
+  //   OuterProto.ProtoA._ProtoB  # Leading vs. trailing underscore.
+  // (Contrived, but certainly possible).
+  //
+  // The C++ implementation doesn't guard against this either.  Leaving
+  // it for now...
+  string name = NamePrefixedWithNestedTypes(descriptor, "_");
+  UpperString(&name);
+  // Module-private for now.  Easy to make public later; almost impossible
+  // to make private later.
+  name = "_" + name;
+  // We now have the name relative to its own module.  Also qualify with
+  // the module name iff this descriptor is from a different .proto file.
+  if (descriptor.file() != file_) {
+    name = ModuleName(descriptor.file()->name()) + "." + name;
+  }
+  return name;
+}
+
+// Returns the name of the message class itself, not the descriptor.
+// Like ModuleLevelDescriptorName(), module-qualifies the name iff
+// the given descriptor describes an entity that doesn't come from
+// the current file.
+string Generator::ModuleLevelMessageName(const Descriptor& descriptor) const {
+  string name = NamePrefixedWithNestedTypes(descriptor, ".");
+  if (descriptor.file() != file_) {
+    name = ModuleName(descriptor.file()->name()) + "." + name;
+  }
+  return name;
+}
+
+// Returns the unique Python module-level identifier given to a service
+// descriptor.
+string Generator::ModuleLevelServiceDescriptorName(
+    const ServiceDescriptor& descriptor) const {
+  string name = descriptor.name();
+  UpperString(&name);
+  name = "_" + name;
+  if (descriptor.file() != file_) {
+    name = ModuleName(descriptor.file()->name()) + "." + name;
+  }
+  return name;
+}
+
+}  // namespace python
+}  // namespace compiler
+}  // namespace protobuf
+}  // namespace google
diff --git a/src/google/protobuf/compiler/python/python_generator.h b/src/google/protobuf/compiler/python/python_generator.h
new file mode 100644
index 00000000..98ee0c6d
--- /dev/null
+++ b/src/google/protobuf/compiler/python/python_generator.h
@@ -0,0 +1,129 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.
+// http://code.google.com/p/protobuf/
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Author: robinson@google.com (Will Robinson)
+//
+// Generates Python code for a given .proto file.
+
+#ifndef GOOGLE_PROTOBUF_COMPILER_PYTHON_GENERATOR_H__
+#define GOOGLE_PROTOBUF_COMPILER_PYTHON_GENERATOR_H__
+
+#include <string>
+
+#include <google/protobuf/compiler/code_generator.h>
+#include <google/protobuf/stubs/common.h>
+
+namespace google {
+namespace protobuf {
+
+class Descriptor;
+class EnumDescriptor;
+class EnumValueDescriptor;
+class FieldDescriptor;
+class ServiceDescriptor;
+
+namespace io { class Printer; }
+
+namespace compiler {
+namespace python {
+
+// CodeGenerator implementation for generated Python protocol buffer classes.
+// If you create your own protocol compiler binary and you want it to support
+// Python output, you can do so by registering an instance of this
+// CodeGenerator with the CommandLineInterface in your main() function.
+class LIBPROTOC_EXPORT Generator : public CodeGenerator {
+ public:
+  Generator();
+  virtual ~Generator();
+
+  // CodeGenerator methods.
+  virtual bool Generate(const FileDescriptor* file,
+                        const string& parameter,
+                        OutputDirectory* output_directory,
+                        string* error) const;
+
+ private:
+  void PrintImports() const;
+  void PrintTopLevelEnums() const;
+  void PrintAllNestedEnumsInFile() const;
+  void PrintNestedEnums(const Descriptor& descriptor) const;
+  void PrintEnum(const EnumDescriptor& enum_descriptor) const;
+
+  void PrintTopLevelExtensions() const;
+
+  void PrintFieldDescriptor(
+      const FieldDescriptor& field, bool is_extension) const;
+  void PrintFieldDescriptorsInDescriptor(
+      const Descriptor& message_descriptor,
+      bool is_extension,
+      const string& list_variable_name,
+      int (Descriptor::*CountFn)() const,
+      const FieldDescriptor* (Descriptor::*GetterFn)(int) const) const;
+  void PrintFieldsInDescriptor(const Descriptor& message_descriptor) const;
+  void PrintExtensionsInDescriptor(const Descriptor& message_descriptor) const;
+  void PrintMessageDescriptors() const;
+  void PrintDescriptor(const Descriptor& message_descriptor) const;
+  void PrintNestedDescriptors(const Descriptor& containing_descriptor) const;
+
+  void PrintMessages() const;
+  void PrintMessage(const Descriptor& message_descriptor) const;
+  void PrintNestedMessages(const Descriptor& containing_descriptor) const;
+
+  void FixForeignFieldsInDescriptors() const;
+  void FixForeignFieldsInDescriptor(const Descriptor& descriptor) const;
+  void FixForeignFieldsInField(const Descriptor* containing_type,
+                               const FieldDescriptor& field,
+                               const string& python_dict_name) const;
+  string FieldReferencingExpression(const Descriptor* containing_type,
+                                    const FieldDescriptor& field,
+                                    const string& python_dict_name) const;
+
+  void FixForeignFieldsInExtensions() const;
+  void FixForeignFieldsInExtension(
+      const FieldDescriptor& extension_field) const;
+  void FixForeignFieldsInNestedExtensions(const Descriptor& descriptor) const;
+
+  void PrintServices() const;
+  void PrintServiceDescriptor(const ServiceDescriptor& descriptor) const;
+  void PrintServiceClass(const ServiceDescriptor& descriptor) const;
+  void PrintServiceStub(const ServiceDescriptor& descriptor) const;
+
+  void PrintEnumValueDescriptor(const EnumValueDescriptor& descriptor) const;
+  string OptionsValue(const string& class_name,
+                      const string& serialized_options) const;
+  bool GeneratingDescriptorProto() const;
+
+  template <typename DescriptorT>
+  string ModuleLevelDescriptorName(const DescriptorT& descriptor) const;
+  string ModuleLevelMessageName(const Descriptor& descriptor) const;
+  string ModuleLevelServiceDescriptorName(
+      const ServiceDescriptor& descriptor) const;
+
+  // Very coarse-grained lock to ensure that Generate() is reentrant.
+  // Guards file_ and printer_.
+  mutable Mutex mutex_;
+  mutable const FileDescriptor* file_;  // Set in Generate().  Under mutex_.
+  mutable io::Printer* printer_;  // Set in Generate().  Under mutex_.
+
+  GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(Generator);
+};
+
+}  // namespace python
+}  // namespace compiler
+}  // namespace protobuf
+
+}  // namespace google
+#endif  // GOOGLE_PROTOBUF_COMPILER_PYTHON_GENERATOR_H__