descriptor.cc revision fbaaef999ba563838ebd00874ed8a1c01fbf286d
1// Protocol Buffers - Google's data interchange format
2// Copyright 2008 Google Inc.  All rights reserved.
3// http://code.google.com/p/protobuf/
4//
5// Redistribution and use in source and binary forms, with or without
6// modification, are permitted provided that the following conditions are
7// met:
8//
9//     * Redistributions of source code must retain the above copyright
10// notice, this list of conditions and the following disclaimer.
11//     * Redistributions in binary form must reproduce the above
12// copyright notice, this list of conditions and the following disclaimer
13// in the documentation and/or other materials provided with the
14// distribution.
15//     * Neither the name of Google Inc. nor the names of its
16// contributors may be used to endorse or promote products derived from
17// this software without specific prior written permission.
18//
19// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30
31// Author: kenton@google.com (Kenton Varda)
32//  Based on original Protocol Buffers design by
33//  Sanjay Ghemawat, Jeff Dean, and others.
34
35#include <google/protobuf/stubs/hash.h>
36#include <map>
37#include <set>
38#include <vector>
39#include <algorithm>
40
41#include <google/protobuf/descriptor.h>
42#include <google/protobuf/descriptor_database.h>
43#include <google/protobuf/descriptor.pb.h>
44#include <google/protobuf/text_format.h>
45#include <google/protobuf/unknown_field_set.h>
46#include <google/protobuf/wire_format.h>
47#include <google/protobuf/io/coded_stream.h>
48#include <google/protobuf/io/zero_copy_stream_impl.h>
49#include <google/protobuf/stubs/common.h>
50#include <google/protobuf/stubs/once.h>
51#include <google/protobuf/stubs/strutil.h>
52#include <google/protobuf/stubs/substitute.h>
53#include <google/protobuf/stubs/map-util.h>
54#include <google/protobuf/stubs/stl_util-inl.h>
55
56#undef PACKAGE  // autoheader #defines this.  :(
57
58namespace google {
59namespace protobuf {
60
61const FieldDescriptor::CppType
62FieldDescriptor::kTypeToCppTypeMap[MAX_TYPE + 1] = {
63  static_cast<CppType>(0),  // 0 is reserved for errors
64
65  CPPTYPE_DOUBLE,   // TYPE_DOUBLE
66  CPPTYPE_FLOAT,    // TYPE_FLOAT
67  CPPTYPE_INT64,    // TYPE_INT64
68  CPPTYPE_UINT64,   // TYPE_UINT64
69  CPPTYPE_INT32,    // TYPE_INT32
70  CPPTYPE_UINT64,   // TYPE_FIXED64
71  CPPTYPE_UINT32,   // TYPE_FIXED32
72  CPPTYPE_BOOL,     // TYPE_BOOL
73  CPPTYPE_STRING,   // TYPE_STRING
74  CPPTYPE_MESSAGE,  // TYPE_GROUP
75  CPPTYPE_MESSAGE,  // TYPE_MESSAGE
76  CPPTYPE_STRING,   // TYPE_BYTES
77  CPPTYPE_UINT32,   // TYPE_UINT32
78  CPPTYPE_ENUM,     // TYPE_ENUM
79  CPPTYPE_INT32,    // TYPE_SFIXED32
80  CPPTYPE_INT64,    // TYPE_SFIXED64
81  CPPTYPE_INT32,    // TYPE_SINT32
82  CPPTYPE_INT64,    // TYPE_SINT64
83};
84
85const char * const FieldDescriptor::kTypeToName[MAX_TYPE + 1] = {
86  "ERROR",     // 0 is reserved for errors
87
88  "double",    // TYPE_DOUBLE
89  "float",     // TYPE_FLOAT
90  "int64",     // TYPE_INT64
91  "uint64",    // TYPE_UINT64
92  "int32",     // TYPE_INT32
93  "fixed64",   // TYPE_FIXED64
94  "fixed32",   // TYPE_FIXED32
95  "bool",      // TYPE_BOOL
96  "string",    // TYPE_STRING
97  "group",     // TYPE_GROUP
98  "message",   // TYPE_MESSAGE
99  "bytes",     // TYPE_BYTES
100  "uint32",    // TYPE_UINT32
101  "enum",      // TYPE_ENUM
102  "sfixed32",  // TYPE_SFIXED32
103  "sfixed64",  // TYPE_SFIXED64
104  "sint32",    // TYPE_SINT32
105  "sint64",    // TYPE_SINT64
106};
107
108const char * const FieldDescriptor::kLabelToName[MAX_LABEL + 1] = {
109  "ERROR",     // 0 is reserved for errors
110
111  "optional",  // LABEL_OPTIONAL
112  "required",  // LABEL_REQUIRED
113  "repeated",  // LABEL_REPEATED
114};
115
116#ifndef _MSC_VER  // MSVC doesn't need these and won't even accept them.
117const int FieldDescriptor::kMaxNumber;
118const int FieldDescriptor::kFirstReservedNumber;
119const int FieldDescriptor::kLastReservedNumber;
120#endif
121
122namespace {
123
124const string kEmptyString;
125
126string ToCamelCase(const string& input) {
127  bool capitalize_next = false;
128  string result;
129  result.reserve(input.size());
130
131  for (int i = 0; i < input.size(); i++) {
132    if (input[i] == '_') {
133      capitalize_next = true;
134    } else if (capitalize_next) {
135      // Note:  I distrust ctype.h due to locales.
136      if ('a' <= input[i] && input[i] <= 'z') {
137        result.push_back(input[i] - 'a' + 'A');
138      } else {
139        result.push_back(input[i]);
140      }
141      capitalize_next = false;
142    } else {
143      result.push_back(input[i]);
144    }
145  }
146
147  // Lower-case the first letter.
148  if (!result.empty() && 'A' <= result[0] && result[0] <= 'Z') {
149    result[0] = result[0] - 'A' + 'a';
150  }
151
152  return result;
153}
154
155// A DescriptorPool contains a bunch of hash_maps to implement the
156// various Find*By*() methods.  Since hashtable lookups are O(1), it's
157// most efficient to construct a fixed set of large hash_maps used by
158// all objects in the pool rather than construct one or more small
159// hash_maps for each object.
160//
161// The keys to these hash_maps are (parent, name) or (parent, number)
162// pairs.  Unfortunately STL doesn't provide hash functions for pair<>,
163// so we must invent our own.
164//
165// TODO(kenton):  Use StringPiece rather than const char* in keys?  It would
166//   be a lot cleaner but we'd just have to convert it back to const char*
167//   for the open source release.
168
169typedef pair<const void*, const char*> PointerStringPair;
170
171struct PointerStringPairEqual {
172  inline bool operator()(const PointerStringPair& a,
173                         const PointerStringPair& b) const {
174    return a.first == b.first && strcmp(a.second, b.second) == 0;
175  }
176};
177
178template<typename PairType>
179struct PointerIntegerPairHash {
180  size_t operator()(const PairType& p) const {
181    // FIXME(kenton):  What is the best way to compute this hash?  I have
182    // no idea!  This seems a bit better than an XOR.
183    return reinterpret_cast<intptr_t>(p.first) * ((1 << 16) - 1) + p.second;
184  }
185
186  // Used only by MSVC and platforms where hash_map is not available.
187  static const size_t bucket_size = 4;
188  static const size_t min_buckets = 8;
189  inline bool operator()(const PairType& a, const PairType& b) const {
190    return a.first < b.first ||
191          (a.first == b.first && a.second < b.second);
192  }
193};
194
195typedef pair<const Descriptor*, int> DescriptorIntPair;
196typedef pair<const EnumDescriptor*, int> EnumIntPair;
197
198struct PointerStringPairHash {
199  size_t operator()(const PointerStringPair& p) const {
200    // FIXME(kenton):  What is the best way to compute this hash?  I have
201    // no idea!  This seems a bit better than an XOR.
202    hash<const char*> cstring_hash;
203    return reinterpret_cast<intptr_t>(p.first) * ((1 << 16) - 1) +
204           cstring_hash(p.second);
205  }
206
207  // Used only by MSVC and platforms where hash_map is not available.
208  static const size_t bucket_size = 4;
209  static const size_t min_buckets = 8;
210  inline bool operator()(const PointerStringPair& a,
211                         const PointerStringPair& b) const {
212    if (a.first < b.first) return true;
213    if (a.first > b.first) return false;
214    return strcmp(a.second, b.second) < 0;
215  }
216};
217
218
219struct Symbol {
220  enum Type {
221    NULL_SYMBOL, MESSAGE, FIELD, ENUM, ENUM_VALUE, SERVICE, METHOD, PACKAGE
222  };
223  Type type;
224  union {
225    const Descriptor* descriptor;
226    const FieldDescriptor* field_descriptor;
227    const EnumDescriptor* enum_descriptor;
228    const EnumValueDescriptor* enum_value_descriptor;
229    const ServiceDescriptor* service_descriptor;
230    const MethodDescriptor* method_descriptor;
231    const FileDescriptor* package_file_descriptor;
232  };
233
234  inline Symbol() : type(NULL_SYMBOL) { descriptor = NULL; }
235  inline bool IsNull() const { return type == NULL_SYMBOL; }
236  inline bool IsType() const {
237    return type == MESSAGE || type == ENUM;
238  }
239  inline bool IsAggregate() const {
240    return type == MESSAGE || type == PACKAGE
241        || type == ENUM || type == SERVICE;
242  }
243
244#define CONSTRUCTOR(TYPE, TYPE_CONSTANT, FIELD)  \
245  inline explicit Symbol(const TYPE* value) {    \
246    type = TYPE_CONSTANT;                        \
247    this->FIELD = value;                         \
248  }
249
250  CONSTRUCTOR(Descriptor         , MESSAGE   , descriptor             )
251  CONSTRUCTOR(FieldDescriptor    , FIELD     , field_descriptor       )
252  CONSTRUCTOR(EnumDescriptor     , ENUM      , enum_descriptor        )
253  CONSTRUCTOR(EnumValueDescriptor, ENUM_VALUE, enum_value_descriptor  )
254  CONSTRUCTOR(ServiceDescriptor  , SERVICE   , service_descriptor     )
255  CONSTRUCTOR(MethodDescriptor   , METHOD    , method_descriptor      )
256  CONSTRUCTOR(FileDescriptor     , PACKAGE   , package_file_descriptor)
257#undef CONSTRUCTOR
258
259  const FileDescriptor* GetFile() const {
260    switch (type) {
261      case NULL_SYMBOL: return NULL;
262      case MESSAGE    : return descriptor           ->file();
263      case FIELD      : return field_descriptor     ->file();
264      case ENUM       : return enum_descriptor      ->file();
265      case ENUM_VALUE : return enum_value_descriptor->type()->file();
266      case SERVICE    : return service_descriptor   ->file();
267      case METHOD     : return method_descriptor    ->service()->file();
268      case PACKAGE    : return package_file_descriptor;
269    }
270    return NULL;
271  }
272};
273
274const Symbol kNullSymbol;
275
276typedef hash_map<const char*, Symbol,
277                 hash<const char*>, streq>
278  SymbolsByNameMap;
279typedef hash_map<PointerStringPair, Symbol,
280                 PointerStringPairHash, PointerStringPairEqual>
281  SymbolsByParentMap;
282typedef hash_map<const char*, const FileDescriptor*,
283                 hash<const char*>, streq>
284  FilesByNameMap;
285typedef hash_map<PointerStringPair, const FieldDescriptor*,
286                 PointerStringPairHash, PointerStringPairEqual>
287  FieldsByNameMap;
288typedef hash_map<DescriptorIntPair, const FieldDescriptor*,
289                 PointerIntegerPairHash<DescriptorIntPair> >
290  FieldsByNumberMap;
291typedef hash_map<EnumIntPair, const EnumValueDescriptor*,
292                 PointerIntegerPairHash<EnumIntPair> >
293  EnumValuesByNumberMap;
294// This is a map rather than a hash_map, since we use it to iterate
295// through all the extensions that extend a given Descriptor, and an
296// ordered data structure that implements lower_bound is convenient
297// for that.
298typedef map<DescriptorIntPair, const FieldDescriptor*>
299  ExtensionsGroupedByDescriptorMap;
300
301}  // anonymous namespace
302
303// ===================================================================
304// DescriptorPool::Tables
305
306class DescriptorPool::Tables {
307 public:
308  Tables();
309  ~Tables();
310
311  // Checkpoint the state of the tables.  Future calls to Rollback() will
312  // return the Tables to this state.  This is used when building files, since
313  // some kinds of validation errors cannot be detected until the file's
314  // descriptors have already been added to the tables.  BuildFile() calls
315  // Checkpoint() before it starts building and Rollback() if it encounters
316  // an error.
317  void Checkpoint();
318
319  // Roll back the Tables to the state of the last Checkpoint(), removing
320  // everything that was added after that point.
321  void Rollback();
322
323  // The stack of files which are currently being built.  Used to detect
324  // cyclic dependencies when loading files from a DescriptorDatabase.  Not
325  // used when fallback_database_ == NULL.
326  vector<string> pending_files_;
327
328  // A set of files which we have tried to load from the fallback database
329  // and encountered errors.  We will not attempt to load them again.
330  // Not used when fallback_database_ == NULL.
331  hash_set<string> known_bad_files_;
332
333  // The set of descriptors for which we've already loaded the full
334  // set of extensions numbers from fallback_database_.
335  hash_set<const Descriptor*> extensions_loaded_from_db_;
336
337  // -----------------------------------------------------------------
338  // Finding items.
339
340  // Find symbols.  This returns a null Symbol (symbol.IsNull() is true)
341  // if not found.
342  inline Symbol FindSymbol(const string& key) const;
343
344  // This implements the body of DescriptorPool::Find*ByName().  It should
345  // really be a private method of DescriptorPool, but that would require
346  // declaring Symbol in descriptor.h, which would drag all kinds of other
347  // stuff into the header.  Yay C++.
348  Symbol FindByNameHelper(
349    const DescriptorPool* pool, const string& name) const;
350
351  // These return NULL if not found.
352  inline const FileDescriptor* FindFile(const string& key) const;
353  inline const FieldDescriptor* FindExtension(const Descriptor* extendee,
354                                              int number);
355  inline void FindAllExtensions(const Descriptor* extendee,
356                                vector<const FieldDescriptor*>* out) const;
357
358  // -----------------------------------------------------------------
359  // Adding items.
360
361  // These add items to the corresponding tables.  They return false if
362  // the key already exists in the table.  For AddSymbol(), the string passed
363  // in must be one that was constructed using AllocateString(), as it will
364  // be used as a key in the symbols_by_name_ map without copying.
365  bool AddSymbol(const string& full_name, Symbol symbol);
366  bool AddFile(const FileDescriptor* file);
367  bool AddExtension(const FieldDescriptor* field);
368
369  // -----------------------------------------------------------------
370  // Allocating memory.
371
372  // Allocate an object which will be reclaimed when the pool is
373  // destroyed.  Note that the object's destructor will never be called,
374  // so its fields must be plain old data (primitive data types and
375  // pointers).  All of the descriptor types are such objects.
376  template<typename Type> Type* Allocate();
377
378  // Allocate an array of objects which will be reclaimed when the
379  // pool in destroyed.  Again, destructors are never called.
380  template<typename Type> Type* AllocateArray(int count);
381
382  // Allocate a string which will be destroyed when the pool is destroyed.
383  // The string is initialized to the given value for convenience.
384  string* AllocateString(const string& value);
385
386  // Allocate a protocol message object.  Some older versions of GCC have
387  // trouble understanding explicit template instantiations in some cases, so
388  // in those cases we have to pass a dummy pointer of the right type as the
389  // parameter instead of specifying the type explicitly.
390  template<typename Type> Type* AllocateMessage(Type* dummy = NULL);
391
392  // Allocate a FileDescriptorTables object.
393  FileDescriptorTables* AllocateFileTables();
394
395 private:
396  vector<string*> strings_;    // All strings in the pool.
397  vector<Message*> messages_;  // All messages in the pool.
398  vector<FileDescriptorTables*> file_tables_;  // All file tables in the pool.
399  vector<void*> allocations_;  // All other memory allocated in the pool.
400
401  SymbolsByNameMap      symbols_by_name_;
402  FilesByNameMap        files_by_name_;
403  ExtensionsGroupedByDescriptorMap extensions_;
404
405  int strings_before_checkpoint_;
406  int messages_before_checkpoint_;
407  int file_tables_before_checkpoint_;
408  int allocations_before_checkpoint_;
409  vector<const char*      > symbols_after_checkpoint_;
410  vector<const char*      > files_after_checkpoint_;
411  vector<DescriptorIntPair> extensions_after_checkpoint_;
412
413  // Allocate some bytes which will be reclaimed when the pool is
414  // destroyed.
415  void* AllocateBytes(int size);
416};
417
418// Contains tables specific to a particular file.  These tables are not
419// modified once the file has been constructed, so they need not be
420// protected by a mutex.  This makes operations that depend only on the
421// contents of a single file -- e.g. Descriptor::FindFieldByName() --
422// lock-free.
423//
424// For historical reasons, the definitions of the methods of
425// FileDescriptorTables and DescriptorPool::Tables are interleaved below.
426// These used to be a single class.
427class FileDescriptorTables {
428 public:
429  FileDescriptorTables();
430  ~FileDescriptorTables();
431
432  // Empty table, used with placeholder files.
433  static const FileDescriptorTables kEmpty;
434
435  // -----------------------------------------------------------------
436  // Finding items.
437
438  // Find symbols.  These return a null Symbol (symbol.IsNull() is true)
439  // if not found.
440  inline Symbol FindNestedSymbol(const void* parent,
441                                 const string& name) const;
442  inline Symbol FindNestedSymbolOfType(const void* parent,
443                                       const string& name,
444                                       const Symbol::Type type) const;
445
446  // These return NULL if not found.
447  inline const FieldDescriptor* FindFieldByNumber(
448    const Descriptor* parent, int number) const;
449  inline const FieldDescriptor* FindFieldByLowercaseName(
450    const void* parent, const string& lowercase_name) const;
451  inline const FieldDescriptor* FindFieldByCamelcaseName(
452    const void* parent, const string& camelcase_name) const;
453  inline const EnumValueDescriptor* FindEnumValueByNumber(
454    const EnumDescriptor* parent, int number) const;
455
456  // -----------------------------------------------------------------
457  // Adding items.
458
459  // These add items to the corresponding tables.  They return false if
460  // the key already exists in the table.  For AddAliasUnderParent(), the
461  // string passed in must be one that was constructed using AllocateString(),
462  // as it will be used as a key in the symbols_by_parent_ map without copying.
463  bool AddAliasUnderParent(const void* parent, const string& name,
464                           Symbol symbol);
465  bool AddFieldByNumber(const FieldDescriptor* field);
466  bool AddEnumValueByNumber(const EnumValueDescriptor* value);
467
468  // Adds the field to the lowercase_name and camelcase_name maps.  Never
469  // fails because we allow duplicates; the first field by the name wins.
470  void AddFieldByStylizedNames(const FieldDescriptor* field);
471
472 private:
473  SymbolsByParentMap    symbols_by_parent_;
474  FieldsByNameMap       fields_by_lowercase_name_;
475  FieldsByNameMap       fields_by_camelcase_name_;
476  FieldsByNumberMap     fields_by_number_;       // Not including extensions.
477  EnumValuesByNumberMap enum_values_by_number_;
478};
479
480DescriptorPool::Tables::Tables()
481  : strings_before_checkpoint_(0),
482    messages_before_checkpoint_(0),
483    allocations_before_checkpoint_(0) {}
484
485DescriptorPool::Tables::~Tables() {
486  // Note that the deletion order is important, since the destructors of some
487  // messages may refer to objects in allocations_.
488  STLDeleteElements(&messages_);
489  for (int i = 0; i < allocations_.size(); i++) {
490    operator delete(allocations_[i]);
491  }
492  STLDeleteElements(&strings_);
493  STLDeleteElements(&file_tables_);
494}
495
496FileDescriptorTables::FileDescriptorTables() {}
497FileDescriptorTables::~FileDescriptorTables() {}
498
499const FileDescriptorTables FileDescriptorTables::kEmpty;
500
501void DescriptorPool::Tables::Checkpoint() {
502  strings_before_checkpoint_ = strings_.size();
503  messages_before_checkpoint_ = messages_.size();
504  file_tables_before_checkpoint_ = file_tables_.size();
505  allocations_before_checkpoint_ = allocations_.size();
506
507  symbols_after_checkpoint_.clear();
508  files_after_checkpoint_.clear();
509  extensions_after_checkpoint_.clear();
510}
511
512void DescriptorPool::Tables::Rollback() {
513  for (int i = 0; i < symbols_after_checkpoint_.size(); i++) {
514    symbols_by_name_.erase(symbols_after_checkpoint_[i]);
515  }
516  for (int i = 0; i < files_after_checkpoint_.size(); i++) {
517    files_by_name_.erase(files_after_checkpoint_[i]);
518  }
519  for (int i = 0; i < extensions_after_checkpoint_.size(); i++) {
520    extensions_.erase(extensions_after_checkpoint_[i]);
521  }
522
523  symbols_after_checkpoint_.clear();
524  files_after_checkpoint_.clear();
525  extensions_after_checkpoint_.clear();
526
527  STLDeleteContainerPointers(
528    strings_.begin() + strings_before_checkpoint_, strings_.end());
529  STLDeleteContainerPointers(
530    messages_.begin() + messages_before_checkpoint_, messages_.end());
531  STLDeleteContainerPointers(
532    file_tables_.begin() + file_tables_before_checkpoint_, file_tables_.end());
533  for (int i = allocations_before_checkpoint_; i < allocations_.size(); i++) {
534    operator delete(allocations_[i]);
535  }
536
537  strings_.resize(strings_before_checkpoint_);
538  messages_.resize(messages_before_checkpoint_);
539  file_tables_.resize(file_tables_before_checkpoint_);
540  allocations_.resize(allocations_before_checkpoint_);
541}
542
543// -------------------------------------------------------------------
544
545inline Symbol DescriptorPool::Tables::FindSymbol(const string& key) const {
546  const Symbol* result = FindOrNull(symbols_by_name_, key.c_str());
547  if (result == NULL) {
548    return kNullSymbol;
549  } else {
550    return *result;
551  }
552}
553
554inline Symbol FileDescriptorTables::FindNestedSymbol(
555    const void* parent, const string& name) const {
556  const Symbol* result =
557    FindOrNull(symbols_by_parent_, PointerStringPair(parent, name.c_str()));
558  if (result == NULL) {
559    return kNullSymbol;
560  } else {
561    return *result;
562  }
563}
564
565inline Symbol FileDescriptorTables::FindNestedSymbolOfType(
566    const void* parent, const string& name, const Symbol::Type type) const {
567  Symbol result = FindNestedSymbol(parent, name);
568  if (result.type != type) return kNullSymbol;
569  return result;
570}
571
572Symbol DescriptorPool::Tables::FindByNameHelper(
573    const DescriptorPool* pool, const string& name) const {
574  MutexLockMaybe lock(pool->mutex_);
575  Symbol result = FindSymbol(name);
576
577  if (result.IsNull() && pool->underlay_ != NULL) {
578    // Symbol not found; check the underlay.
579    result =
580      pool->underlay_->tables_->FindByNameHelper(pool->underlay_, name);
581  }
582
583  if (result.IsNull()) {
584    // Symbol still not found, so check fallback database.
585    if (pool->TryFindSymbolInFallbackDatabase(name)) {
586      result = FindSymbol(name);
587    }
588  }
589
590  return result;
591}
592
593inline const FileDescriptor* DescriptorPool::Tables::FindFile(
594    const string& key) const {
595  return FindPtrOrNull(files_by_name_, key.c_str());
596}
597
598inline const FieldDescriptor* FileDescriptorTables::FindFieldByNumber(
599    const Descriptor* parent, int number) const {
600  return FindPtrOrNull(fields_by_number_, make_pair(parent, number));
601}
602
603inline const FieldDescriptor* FileDescriptorTables::FindFieldByLowercaseName(
604    const void* parent, const string& lowercase_name) const {
605  return FindPtrOrNull(fields_by_lowercase_name_,
606                       PointerStringPair(parent, lowercase_name.c_str()));
607}
608
609inline const FieldDescriptor* FileDescriptorTables::FindFieldByCamelcaseName(
610    const void* parent, const string& camelcase_name) const {
611  return FindPtrOrNull(fields_by_camelcase_name_,
612                       PointerStringPair(parent, camelcase_name.c_str()));
613}
614
615inline const EnumValueDescriptor* FileDescriptorTables::FindEnumValueByNumber(
616    const EnumDescriptor* parent, int number) const {
617  return FindPtrOrNull(enum_values_by_number_, make_pair(parent, number));
618}
619
620inline const FieldDescriptor* DescriptorPool::Tables::FindExtension(
621    const Descriptor* extendee, int number) {
622  return FindPtrOrNull(extensions_, make_pair(extendee, number));
623}
624
625inline void DescriptorPool::Tables::FindAllExtensions(
626    const Descriptor* extendee, vector<const FieldDescriptor*>* out) const {
627  ExtensionsGroupedByDescriptorMap::const_iterator it =
628      extensions_.lower_bound(make_pair(extendee, 0));
629  for (; it != extensions_.end() && it->first.first == extendee; ++it) {
630    out->push_back(it->second);
631  }
632}
633
634// -------------------------------------------------------------------
635
636bool DescriptorPool::Tables::AddSymbol(
637    const string& full_name, Symbol symbol) {
638  if (InsertIfNotPresent(&symbols_by_name_, full_name.c_str(), symbol)) {
639    symbols_after_checkpoint_.push_back(full_name.c_str());
640    return true;
641  } else {
642    return false;
643  }
644}
645
646bool FileDescriptorTables::AddAliasUnderParent(
647    const void* parent, const string& name, Symbol symbol) {
648  PointerStringPair by_parent_key(parent, name.c_str());
649  return InsertIfNotPresent(&symbols_by_parent_, by_parent_key, symbol);
650}
651
652bool DescriptorPool::Tables::AddFile(const FileDescriptor* file) {
653  if (InsertIfNotPresent(&files_by_name_, file->name().c_str(), file)) {
654    files_after_checkpoint_.push_back(file->name().c_str());
655    return true;
656  } else {
657    return false;
658  }
659}
660
661void FileDescriptorTables::AddFieldByStylizedNames(
662    const FieldDescriptor* field) {
663  const void* parent;
664  if (field->is_extension()) {
665    if (field->extension_scope() == NULL) {
666      parent = field->file();
667    } else {
668      parent = field->extension_scope();
669    }
670  } else {
671    parent = field->containing_type();
672  }
673
674  PointerStringPair lowercase_key(parent, field->lowercase_name().c_str());
675  InsertIfNotPresent(&fields_by_lowercase_name_, lowercase_key, field);
676
677  PointerStringPair camelcase_key(parent, field->camelcase_name().c_str());
678  InsertIfNotPresent(&fields_by_camelcase_name_, camelcase_key, field);
679}
680
681bool FileDescriptorTables::AddFieldByNumber(const FieldDescriptor* field) {
682  DescriptorIntPair key(field->containing_type(), field->number());
683  return InsertIfNotPresent(&fields_by_number_, key, field);
684}
685
686bool FileDescriptorTables::AddEnumValueByNumber(
687    const EnumValueDescriptor* value) {
688  EnumIntPair key(value->type(), value->number());
689  return InsertIfNotPresent(&enum_values_by_number_, key, value);
690}
691
692bool DescriptorPool::Tables::AddExtension(const FieldDescriptor* field) {
693  DescriptorIntPair key(field->containing_type(), field->number());
694  if (InsertIfNotPresent(&extensions_, key, field)) {
695    extensions_after_checkpoint_.push_back(key);
696    return true;
697  } else {
698    return false;
699  }
700}
701
702// -------------------------------------------------------------------
703
704template<typename Type>
705Type* DescriptorPool::Tables::Allocate() {
706  return reinterpret_cast<Type*>(AllocateBytes(sizeof(Type)));
707}
708
709template<typename Type>
710Type* DescriptorPool::Tables::AllocateArray(int count) {
711  return reinterpret_cast<Type*>(AllocateBytes(sizeof(Type) * count));
712}
713
714string* DescriptorPool::Tables::AllocateString(const string& value) {
715  string* result = new string(value);
716  strings_.push_back(result);
717  return result;
718}
719
720template<typename Type>
721Type* DescriptorPool::Tables::AllocateMessage(Type* dummy) {
722  Type* result = new Type;
723  messages_.push_back(result);
724  return result;
725}
726
727FileDescriptorTables* DescriptorPool::Tables::AllocateFileTables() {
728  FileDescriptorTables* result = new FileDescriptorTables;
729  file_tables_.push_back(result);
730  return result;
731}
732
733void* DescriptorPool::Tables::AllocateBytes(int size) {
734  // TODO(kenton):  Would it be worthwhile to implement this in some more
735  // sophisticated way?  Probably not for the open source release, but for
736  // internal use we could easily plug in one of our existing memory pool
737  // allocators...
738  if (size == 0) return NULL;
739
740  void* result = operator new(size);
741  allocations_.push_back(result);
742  return result;
743}
744
745// ===================================================================
746// DescriptorPool
747
748DescriptorPool::ErrorCollector::~ErrorCollector() {}
749
750DescriptorPool::DescriptorPool()
751  : mutex_(NULL),
752    fallback_database_(NULL),
753    default_error_collector_(NULL),
754    underlay_(NULL),
755    tables_(new Tables),
756    enforce_dependencies_(true),
757    allow_unknown_(false) {}
758
759DescriptorPool::DescriptorPool(DescriptorDatabase* fallback_database,
760                               ErrorCollector* error_collector)
761  : mutex_(new Mutex),
762    fallback_database_(fallback_database),
763    default_error_collector_(error_collector),
764    underlay_(NULL),
765    tables_(new Tables),
766    enforce_dependencies_(true),
767    allow_unknown_(false) {
768}
769
770DescriptorPool::DescriptorPool(const DescriptorPool* underlay)
771  : mutex_(NULL),
772    fallback_database_(NULL),
773    default_error_collector_(NULL),
774    underlay_(underlay),
775    tables_(new Tables),
776    enforce_dependencies_(true),
777    allow_unknown_(false) {}
778
779DescriptorPool::~DescriptorPool() {
780  if (mutex_ != NULL) delete mutex_;
781}
782
783// DescriptorPool::BuildFile() defined later.
784// DescriptorPool::BuildFileCollectingErrors() defined later.
785
786void DescriptorPool::InternalDontEnforceDependencies() {
787  enforce_dependencies_ = false;
788}
789
790bool DescriptorPool::InternalIsFileLoaded(const string& filename) const {
791  MutexLockMaybe lock(mutex_);
792  return tables_->FindFile(filename) != NULL;
793}
794
795// generated_pool ====================================================
796
797namespace {
798
799EncodedDescriptorDatabase* generated_database_ = NULL;
800DescriptorPool* generated_pool_ = NULL;
801GOOGLE_PROTOBUF_DECLARE_ONCE(generated_pool_init_);
802
803void DeleteGeneratedPool() {
804  delete generated_database_;
805  generated_database_ = NULL;
806  delete generated_pool_;
807  generated_pool_ = NULL;
808}
809
810void InitGeneratedPool() {
811  generated_database_ = new EncodedDescriptorDatabase;
812  generated_pool_ = new DescriptorPool(generated_database_);
813  internal::OnShutdown(&DeleteGeneratedPool);
814}
815
816inline void InitGeneratedPoolOnce() {
817  ::google::protobuf::GoogleOnceInit(&generated_pool_init_, &InitGeneratedPool);
818}
819
820}  // anonymous namespace
821
822const DescriptorPool* DescriptorPool::generated_pool() {
823  InitGeneratedPoolOnce();
824  return generated_pool_;
825}
826
827DescriptorPool* DescriptorPool::internal_generated_pool() {
828  InitGeneratedPoolOnce();
829  return generated_pool_;
830}
831
832void DescriptorPool::InternalAddGeneratedFile(
833    const void* encoded_file_descriptor, int size) {
834  // So, this function is called in the process of initializing the
835  // descriptors for generated proto classes.  Each generated .pb.cc file
836  // has an internal procedure called AddDescriptors() which is called at
837  // process startup, and that function calls this one in order to register
838  // the raw bytes of the FileDescriptorProto representing the file.
839  //
840  // We do not actually construct the descriptor objects right away.  We just
841  // hang on to the bytes until they are actually needed.  We actually construct
842  // the descriptor the first time one of the following things happens:
843  // * Someone calls a method like descriptor(), GetDescriptor(), or
844  //   GetReflection() on the generated types, which requires returning the
845  //   descriptor or an object based on it.
846  // * Someone looks up the descriptor in DescriptorPool::generated_pool().
847  //
848  // Once one of these happens, the DescriptorPool actually parses the
849  // FileDescriptorProto and generates a FileDescriptor (and all its children)
850  // based on it.
851  //
852  // Note that FileDescriptorProto is itself a generated protocol message.
853  // Therefore, when we parse one, we have to be very careful to avoid using
854  // any descriptor-based operations, since this might cause infinite recursion
855  // or deadlock.
856  InitGeneratedPoolOnce();
857  GOOGLE_CHECK(generated_database_->Add(encoded_file_descriptor, size));
858}
859
860
861// Find*By* methods ==================================================
862
863// TODO(kenton):  There's a lot of repeated code here, but I'm not sure if
864//   there's any good way to factor it out.  Think about this some time when
865//   there's nothing more important to do (read: never).
866
867const FileDescriptor* DescriptorPool::FindFileByName(const string& name) const {
868  MutexLockMaybe lock(mutex_);
869  const FileDescriptor* result = tables_->FindFile(name);
870  if (result != NULL) return result;
871  if (underlay_ != NULL) {
872    const FileDescriptor* result = underlay_->FindFileByName(name);
873    if (result != NULL) return result;
874  }
875  if (TryFindFileInFallbackDatabase(name)) {
876    const FileDescriptor* result = tables_->FindFile(name);
877    if (result != NULL) return result;
878  }
879  return NULL;
880}
881
882const FileDescriptor* DescriptorPool::FindFileContainingSymbol(
883    const string& symbol_name) const {
884  MutexLockMaybe lock(mutex_);
885  Symbol result = tables_->FindSymbol(symbol_name);
886  if (!result.IsNull()) return result.GetFile();
887  if (underlay_ != NULL) {
888    const FileDescriptor* result =
889      underlay_->FindFileContainingSymbol(symbol_name);
890    if (result != NULL) return result;
891  }
892  if (TryFindSymbolInFallbackDatabase(symbol_name)) {
893    Symbol result = tables_->FindSymbol(symbol_name);
894    if (!result.IsNull()) return result.GetFile();
895  }
896  return NULL;
897}
898
899const Descriptor* DescriptorPool::FindMessageTypeByName(
900    const string& name) const {
901  Symbol result = tables_->FindByNameHelper(this, name);
902  return (result.type == Symbol::MESSAGE) ? result.descriptor : NULL;
903}
904
905const FieldDescriptor* DescriptorPool::FindFieldByName(
906    const string& name) const {
907  Symbol result = tables_->FindByNameHelper(this, name);
908  if (result.type == Symbol::FIELD &&
909      !result.field_descriptor->is_extension()) {
910    return result.field_descriptor;
911  } else {
912    return NULL;
913  }
914}
915
916const FieldDescriptor* DescriptorPool::FindExtensionByName(
917    const string& name) const {
918  Symbol result = tables_->FindByNameHelper(this, name);
919  if (result.type == Symbol::FIELD &&
920      result.field_descriptor->is_extension()) {
921    return result.field_descriptor;
922  } else {
923    return NULL;
924  }
925}
926
927const EnumDescriptor* DescriptorPool::FindEnumTypeByName(
928    const string& name) const {
929  Symbol result = tables_->FindByNameHelper(this, name);
930  return (result.type == Symbol::ENUM) ? result.enum_descriptor : NULL;
931}
932
933const EnumValueDescriptor* DescriptorPool::FindEnumValueByName(
934    const string& name) const {
935  Symbol result = tables_->FindByNameHelper(this, name);
936  return (result.type == Symbol::ENUM_VALUE) ?
937    result.enum_value_descriptor : NULL;
938}
939
940const ServiceDescriptor* DescriptorPool::FindServiceByName(
941    const string& name) const {
942  Symbol result = tables_->FindByNameHelper(this, name);
943  return (result.type == Symbol::SERVICE) ? result.service_descriptor : NULL;
944}
945
946const MethodDescriptor* DescriptorPool::FindMethodByName(
947    const string& name) const {
948  Symbol result = tables_->FindByNameHelper(this, name);
949  return (result.type == Symbol::METHOD) ? result.method_descriptor : NULL;
950}
951
952const FieldDescriptor* DescriptorPool::FindExtensionByNumber(
953    const Descriptor* extendee, int number) const {
954  MutexLockMaybe lock(mutex_);
955  const FieldDescriptor* result = tables_->FindExtension(extendee, number);
956  if (result != NULL) {
957    return result;
958  }
959  if (underlay_ != NULL) {
960    const FieldDescriptor* result =
961      underlay_->FindExtensionByNumber(extendee, number);
962    if (result != NULL) return result;
963  }
964  if (TryFindExtensionInFallbackDatabase(extendee, number)) {
965    const FieldDescriptor* result = tables_->FindExtension(extendee, number);
966    if (result != NULL) {
967      return result;
968    }
969  }
970  return NULL;
971}
972
973void DescriptorPool::FindAllExtensions(
974    const Descriptor* extendee, vector<const FieldDescriptor*>* out) const {
975  MutexLockMaybe lock(mutex_);
976
977  // Initialize tables_->extensions_ from the fallback database first
978  // (but do this only once per descriptor).
979  if (fallback_database_ != NULL &&
980      tables_->extensions_loaded_from_db_.count(extendee) == 0) {
981    vector<int> numbers;
982    if (fallback_database_->FindAllExtensionNumbers(extendee->full_name(),
983                                                    &numbers)) {
984      for (int i = 0; i < numbers.size(); ++i) {
985        int number = numbers[i];
986        if (tables_->FindExtension(extendee, number) == NULL) {
987          TryFindExtensionInFallbackDatabase(extendee, number);
988        }
989      }
990      tables_->extensions_loaded_from_db_.insert(extendee);
991    }
992  }
993
994  tables_->FindAllExtensions(extendee, out);
995  if (underlay_ != NULL) {
996    underlay_->FindAllExtensions(extendee, out);
997  }
998}
999
1000// -------------------------------------------------------------------
1001
1002const FieldDescriptor*
1003Descriptor::FindFieldByNumber(int key) const {
1004  const FieldDescriptor* result =
1005    file()->tables_->FindFieldByNumber(this, key);
1006  if (result == NULL || result->is_extension()) {
1007    return NULL;
1008  } else {
1009    return result;
1010  }
1011}
1012
1013const FieldDescriptor*
1014Descriptor::FindFieldByLowercaseName(const string& key) const {
1015  const FieldDescriptor* result =
1016    file()->tables_->FindFieldByLowercaseName(this, key);
1017  if (result == NULL || result->is_extension()) {
1018    return NULL;
1019  } else {
1020    return result;
1021  }
1022}
1023
1024const FieldDescriptor*
1025Descriptor::FindFieldByCamelcaseName(const string& key) const {
1026  const FieldDescriptor* result =
1027    file()->tables_->FindFieldByCamelcaseName(this, key);
1028  if (result == NULL || result->is_extension()) {
1029    return NULL;
1030  } else {
1031    return result;
1032  }
1033}
1034
1035const FieldDescriptor*
1036Descriptor::FindFieldByName(const string& key) const {
1037  Symbol result =
1038    file()->tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1039  if (!result.IsNull() && !result.field_descriptor->is_extension()) {
1040    return result.field_descriptor;
1041  } else {
1042    return NULL;
1043  }
1044}
1045
1046const FieldDescriptor*
1047Descriptor::FindExtensionByName(const string& key) const {
1048  Symbol result =
1049    file()->tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1050  if (!result.IsNull() && result.field_descriptor->is_extension()) {
1051    return result.field_descriptor;
1052  } else {
1053    return NULL;
1054  }
1055}
1056
1057const FieldDescriptor*
1058Descriptor::FindExtensionByLowercaseName(const string& key) const {
1059  const FieldDescriptor* result =
1060    file()->tables_->FindFieldByLowercaseName(this, key);
1061  if (result == NULL || !result->is_extension()) {
1062    return NULL;
1063  } else {
1064    return result;
1065  }
1066}
1067
1068const FieldDescriptor*
1069Descriptor::FindExtensionByCamelcaseName(const string& key) const {
1070  const FieldDescriptor* result =
1071    file()->tables_->FindFieldByCamelcaseName(this, key);
1072  if (result == NULL || !result->is_extension()) {
1073    return NULL;
1074  } else {
1075    return result;
1076  }
1077}
1078
1079const Descriptor*
1080Descriptor::FindNestedTypeByName(const string& key) const {
1081  Symbol result =
1082    file()->tables_->FindNestedSymbolOfType(this, key, Symbol::MESSAGE);
1083  if (!result.IsNull()) {
1084    return result.descriptor;
1085  } else {
1086    return NULL;
1087  }
1088}
1089
1090const EnumDescriptor*
1091Descriptor::FindEnumTypeByName(const string& key) const {
1092  Symbol result =
1093    file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM);
1094  if (!result.IsNull()) {
1095    return result.enum_descriptor;
1096  } else {
1097    return NULL;
1098  }
1099}
1100
1101const EnumValueDescriptor*
1102Descriptor::FindEnumValueByName(const string& key) const {
1103  Symbol result =
1104    file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1105  if (!result.IsNull()) {
1106    return result.enum_value_descriptor;
1107  } else {
1108    return NULL;
1109  }
1110}
1111
1112const EnumValueDescriptor*
1113EnumDescriptor::FindValueByName(const string& key) const {
1114  Symbol result =
1115    file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1116  if (!result.IsNull()) {
1117    return result.enum_value_descriptor;
1118  } else {
1119    return NULL;
1120  }
1121}
1122
1123const EnumValueDescriptor*
1124EnumDescriptor::FindValueByNumber(int key) const {
1125  return file()->tables_->FindEnumValueByNumber(this, key);
1126}
1127
1128const MethodDescriptor*
1129ServiceDescriptor::FindMethodByName(const string& key) const {
1130  Symbol result =
1131    file()->tables_->FindNestedSymbolOfType(this, key, Symbol::METHOD);
1132  if (!result.IsNull()) {
1133    return result.method_descriptor;
1134  } else {
1135    return NULL;
1136  }
1137}
1138
1139const Descriptor*
1140FileDescriptor::FindMessageTypeByName(const string& key) const {
1141  Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::MESSAGE);
1142  if (!result.IsNull()) {
1143    return result.descriptor;
1144  } else {
1145    return NULL;
1146  }
1147}
1148
1149const EnumDescriptor*
1150FileDescriptor::FindEnumTypeByName(const string& key) const {
1151  Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM);
1152  if (!result.IsNull()) {
1153    return result.enum_descriptor;
1154  } else {
1155    return NULL;
1156  }
1157}
1158
1159const EnumValueDescriptor*
1160FileDescriptor::FindEnumValueByName(const string& key) const {
1161  Symbol result =
1162    tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1163  if (!result.IsNull()) {
1164    return result.enum_value_descriptor;
1165  } else {
1166    return NULL;
1167  }
1168}
1169
1170const ServiceDescriptor*
1171FileDescriptor::FindServiceByName(const string& key) const {
1172  Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::SERVICE);
1173  if (!result.IsNull()) {
1174    return result.service_descriptor;
1175  } else {
1176    return NULL;
1177  }
1178}
1179
1180const FieldDescriptor*
1181FileDescriptor::FindExtensionByName(const string& key) const {
1182  Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1183  if (!result.IsNull() && result.field_descriptor->is_extension()) {
1184    return result.field_descriptor;
1185  } else {
1186    return NULL;
1187  }
1188}
1189
1190const FieldDescriptor*
1191FileDescriptor::FindExtensionByLowercaseName(const string& key) const {
1192  const FieldDescriptor* result = tables_->FindFieldByLowercaseName(this, key);
1193  if (result == NULL || !result->is_extension()) {
1194    return NULL;
1195  } else {
1196    return result;
1197  }
1198}
1199
1200const FieldDescriptor*
1201FileDescriptor::FindExtensionByCamelcaseName(const string& key) const {
1202  const FieldDescriptor* result = tables_->FindFieldByCamelcaseName(this, key);
1203  if (result == NULL || !result->is_extension()) {
1204    return NULL;
1205  } else {
1206    return result;
1207  }
1208}
1209
1210bool Descriptor::IsExtensionNumber(int number) const {
1211  // Linear search should be fine because we don't expect a message to have
1212  // more than a couple extension ranges.
1213  for (int i = 0; i < extension_range_count(); i++) {
1214    if (number >= extension_range(i)->start &&
1215        number <  extension_range(i)->end) {
1216      return true;
1217    }
1218  }
1219  return false;
1220}
1221
1222// -------------------------------------------------------------------
1223
1224bool DescriptorPool::TryFindFileInFallbackDatabase(const string& name) const {
1225  if (fallback_database_ == NULL) return false;
1226
1227  if (tables_->known_bad_files_.count(name) > 0) return false;
1228
1229  FileDescriptorProto file_proto;
1230  if (!fallback_database_->FindFileByName(name, &file_proto) ||
1231      BuildFileFromDatabase(file_proto) == NULL) {
1232    tables_->known_bad_files_.insert(name);
1233    return false;
1234  }
1235
1236  return true;
1237}
1238
1239bool DescriptorPool::TryFindSymbolInFallbackDatabase(const string& name) const {
1240  if (fallback_database_ == NULL) return false;
1241
1242  FileDescriptorProto file_proto;
1243  if (!fallback_database_->FindFileContainingSymbol(name, &file_proto)) {
1244    return false;
1245  }
1246
1247  if (tables_->FindFile(file_proto.name()) != NULL) {
1248    // We've already loaded this file, and it apparently doesn't contain the
1249    // symbol we're looking for.  Some DescriptorDatabases return false
1250    // positives.
1251    return false;
1252  }
1253
1254  if (BuildFileFromDatabase(file_proto) == NULL) {
1255    return false;
1256  }
1257
1258  return true;
1259}
1260
1261bool DescriptorPool::TryFindExtensionInFallbackDatabase(
1262    const Descriptor* containing_type, int field_number) const {
1263  if (fallback_database_ == NULL) return false;
1264
1265  FileDescriptorProto file_proto;
1266  if (!fallback_database_->FindFileContainingExtension(
1267        containing_type->full_name(), field_number, &file_proto)) {
1268    return false;
1269  }
1270
1271  if (tables_->FindFile(file_proto.name()) != NULL) {
1272    // We've already loaded this file, and it apparently doesn't contain the
1273    // extension we're looking for.  Some DescriptorDatabases return false
1274    // positives.
1275    return false;
1276  }
1277
1278  if (BuildFileFromDatabase(file_proto) == NULL) {
1279    return false;
1280  }
1281
1282  return true;
1283}
1284
1285// ===================================================================
1286
1287string FieldDescriptor::DefaultValueAsString(bool quote_string_type) const {
1288  GOOGLE_CHECK(has_default_value()) << "No default value";
1289  switch (cpp_type()) {
1290    case CPPTYPE_INT32:
1291      return SimpleItoa(default_value_int32());
1292      break;
1293    case CPPTYPE_INT64:
1294      return SimpleItoa(default_value_int64());
1295      break;
1296    case CPPTYPE_UINT32:
1297      return SimpleItoa(default_value_uint32());
1298      break;
1299    case CPPTYPE_UINT64:
1300      return SimpleItoa(default_value_uint64());
1301      break;
1302    case CPPTYPE_FLOAT:
1303      return SimpleFtoa(default_value_float());
1304      break;
1305    case CPPTYPE_DOUBLE:
1306      return SimpleDtoa(default_value_double());
1307      break;
1308    case CPPTYPE_BOOL:
1309      return default_value_bool() ? "true" : "false";
1310      break;
1311    case CPPTYPE_STRING:
1312      if (quote_string_type) {
1313        return "\"" + CEscape(default_value_string()) + "\"";
1314      } else {
1315        if (type() == TYPE_BYTES) {
1316          return CEscape(default_value_string());
1317        } else {
1318          return default_value_string();
1319        }
1320      }
1321      break;
1322    case CPPTYPE_ENUM:
1323      return default_value_enum()->name();
1324      break;
1325    case CPPTYPE_MESSAGE:
1326      GOOGLE_LOG(DFATAL) << "Messages can't have default values!";
1327      break;
1328  }
1329  GOOGLE_LOG(FATAL) << "Can't get here: failed to get default value as string";
1330  return "";
1331}
1332
1333// CopyTo methods ====================================================
1334
1335void FileDescriptor::CopyTo(FileDescriptorProto* proto) const {
1336  proto->set_name(name());
1337  if (!package().empty()) proto->set_package(package());
1338
1339  for (int i = 0; i < dependency_count(); i++) {
1340    proto->add_dependency(dependency(i)->name());
1341  }
1342
1343  for (int i = 0; i < message_type_count(); i++) {
1344    message_type(i)->CopyTo(proto->add_message_type());
1345  }
1346  for (int i = 0; i < enum_type_count(); i++) {
1347    enum_type(i)->CopyTo(proto->add_enum_type());
1348  }
1349  for (int i = 0; i < service_count(); i++) {
1350    service(i)->CopyTo(proto->add_service());
1351  }
1352  for (int i = 0; i < extension_count(); i++) {
1353    extension(i)->CopyTo(proto->add_extension());
1354  }
1355
1356  if (&options() != &FileOptions::default_instance()) {
1357    proto->mutable_options()->CopyFrom(options());
1358  }
1359}
1360
1361void Descriptor::CopyTo(DescriptorProto* proto) const {
1362  proto->set_name(name());
1363
1364  for (int i = 0; i < field_count(); i++) {
1365    field(i)->CopyTo(proto->add_field());
1366  }
1367  for (int i = 0; i < nested_type_count(); i++) {
1368    nested_type(i)->CopyTo(proto->add_nested_type());
1369  }
1370  for (int i = 0; i < enum_type_count(); i++) {
1371    enum_type(i)->CopyTo(proto->add_enum_type());
1372  }
1373  for (int i = 0; i < extension_range_count(); i++) {
1374    DescriptorProto::ExtensionRange* range = proto->add_extension_range();
1375    range->set_start(extension_range(i)->start);
1376    range->set_end(extension_range(i)->end);
1377  }
1378  for (int i = 0; i < extension_count(); i++) {
1379    extension(i)->CopyTo(proto->add_extension());
1380  }
1381
1382  if (&options() != &MessageOptions::default_instance()) {
1383    proto->mutable_options()->CopyFrom(options());
1384  }
1385}
1386
1387void FieldDescriptor::CopyTo(FieldDescriptorProto* proto) const {
1388  proto->set_name(name());
1389  proto->set_number(number());
1390
1391  // Some compilers do not allow static_cast directly between two enum types,
1392  // so we must cast to int first.
1393  proto->set_label(static_cast<FieldDescriptorProto::Label>(
1394                     implicit_cast<int>(label())));
1395  proto->set_type(static_cast<FieldDescriptorProto::Type>(
1396                    implicit_cast<int>(type())));
1397
1398  if (is_extension()) {
1399    if (!containing_type()->is_unqualified_placeholder_) {
1400      proto->set_extendee(".");
1401    }
1402    proto->mutable_extendee()->append(containing_type()->full_name());
1403  }
1404
1405  if (cpp_type() == CPPTYPE_MESSAGE) {
1406    if (message_type()->is_placeholder_) {
1407      // We don't actually know if the type is a message type.  It could be
1408      // an enum.
1409      proto->clear_type();
1410    }
1411
1412    if (!message_type()->is_unqualified_placeholder_) {
1413      proto->set_type_name(".");
1414    }
1415    proto->mutable_type_name()->append(message_type()->full_name());
1416  } else if (cpp_type() == CPPTYPE_ENUM) {
1417    if (!enum_type()->is_unqualified_placeholder_) {
1418      proto->set_type_name(".");
1419    }
1420    proto->mutable_type_name()->append(enum_type()->full_name());
1421  }
1422
1423  if (has_default_value()) {
1424    proto->set_default_value(DefaultValueAsString(false));
1425  }
1426
1427  if (&options() != &FieldOptions::default_instance()) {
1428    proto->mutable_options()->CopyFrom(options());
1429  }
1430}
1431
1432void EnumDescriptor::CopyTo(EnumDescriptorProto* proto) const {
1433  proto->set_name(name());
1434
1435  for (int i = 0; i < value_count(); i++) {
1436    value(i)->CopyTo(proto->add_value());
1437  }
1438
1439  if (&options() != &EnumOptions::default_instance()) {
1440    proto->mutable_options()->CopyFrom(options());
1441  }
1442}
1443
1444void EnumValueDescriptor::CopyTo(EnumValueDescriptorProto* proto) const {
1445  proto->set_name(name());
1446  proto->set_number(number());
1447
1448  if (&options() != &EnumValueOptions::default_instance()) {
1449    proto->mutable_options()->CopyFrom(options());
1450  }
1451}
1452
1453void ServiceDescriptor::CopyTo(ServiceDescriptorProto* proto) const {
1454  proto->set_name(name());
1455
1456  for (int i = 0; i < method_count(); i++) {
1457    method(i)->CopyTo(proto->add_method());
1458  }
1459
1460  if (&options() != &ServiceOptions::default_instance()) {
1461    proto->mutable_options()->CopyFrom(options());
1462  }
1463}
1464
1465void MethodDescriptor::CopyTo(MethodDescriptorProto* proto) const {
1466  proto->set_name(name());
1467
1468  if (!input_type()->is_unqualified_placeholder_) {
1469    proto->set_input_type(".");
1470  }
1471  proto->mutable_input_type()->append(input_type()->full_name());
1472
1473  if (!output_type()->is_unqualified_placeholder_) {
1474    proto->set_output_type(".");
1475  }
1476  proto->mutable_output_type()->append(output_type()->full_name());
1477
1478  if (&options() != &MethodOptions::default_instance()) {
1479    proto->mutable_options()->CopyFrom(options());
1480  }
1481}
1482
1483// DebugString methods ===============================================
1484
1485namespace {
1486
1487// Used by each of the option formatters.
1488bool RetrieveOptions(const Message &options, vector<string> *option_entries) {
1489  option_entries->clear();
1490  const Reflection* reflection = options.GetReflection();
1491  vector<const FieldDescriptor*> fields;
1492  reflection->ListFields(options, &fields);
1493  for (int i = 0; i < fields.size(); i++) {
1494    // Doesn't make sense to have message type fields here
1495    if (fields[i]->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
1496      continue;
1497    }
1498    int count = 1;
1499    bool repeated = false;
1500    if (fields[i]->is_repeated()) {
1501      count = reflection->FieldSize(options, fields[i]);
1502      repeated = true;
1503    }
1504    for (int j = 0; j < count; j++) {
1505      string fieldval;
1506      TextFormat::PrintFieldValueToString(options, fields[i],
1507                                          repeated ? count : -1, &fieldval);
1508      option_entries->push_back(fields[i]->name() + " = " + fieldval);
1509    }
1510  }
1511  return !option_entries->empty();
1512}
1513
1514// Formats options that all appear together in brackets. Does not include
1515// brackets.
1516bool FormatBracketedOptions(const Message &options, string *output) {
1517  vector<string> all_options;
1518  if (RetrieveOptions(options, &all_options)) {
1519    output->append(JoinStrings(all_options, ", "));
1520  }
1521  return !all_options.empty();
1522}
1523
1524// Formats options one per line
1525bool FormatLineOptions(int depth, const Message &options, string *output) {
1526  string prefix(depth * 2, ' ');
1527  vector<string> all_options;
1528  if (RetrieveOptions(options, &all_options)) {
1529    for (int i = 0; i < all_options.size(); i++) {
1530      strings::SubstituteAndAppend(output, "$0option $1;\n",
1531                                   prefix, all_options[i]);
1532    }
1533  }
1534  return !all_options.empty();
1535}
1536
1537}  // anonymous namespace
1538
1539string FileDescriptor::DebugString() const {
1540  string contents = "syntax = \"proto2\";\n\n";
1541
1542  for (int i = 0; i < dependency_count(); i++) {
1543    strings::SubstituteAndAppend(&contents, "import \"$0\";\n",
1544                                 dependency(i)->name());
1545  }
1546
1547  if (!package().empty()) {
1548    strings::SubstituteAndAppend(&contents, "package $0;\n\n", package());
1549  }
1550
1551  if (FormatLineOptions(0, options(), &contents)) {
1552    contents.append("\n");  // add some space if we had options
1553  }
1554
1555  for (int i = 0; i < enum_type_count(); i++) {
1556    enum_type(i)->DebugString(0, &contents);
1557    contents.append("\n");
1558  }
1559
1560  // Find all the 'group' type extensions; we will not output their nested
1561  // definitions (those will be done with their group field descriptor).
1562  set<const Descriptor*> groups;
1563  for (int i = 0; i < extension_count(); i++) {
1564    if (extension(i)->type() == FieldDescriptor::TYPE_GROUP) {
1565      groups.insert(extension(i)->message_type());
1566    }
1567  }
1568
1569  for (int i = 0; i < message_type_count(); i++) {
1570    if (groups.count(message_type(i)) == 0) {
1571      strings::SubstituteAndAppend(&contents, "message $0",
1572                                   message_type(i)->name());
1573      message_type(i)->DebugString(0, &contents);
1574      contents.append("\n");
1575    }
1576  }
1577
1578  for (int i = 0; i < service_count(); i++) {
1579    service(i)->DebugString(&contents);
1580    contents.append("\n");
1581  }
1582
1583  const Descriptor* containing_type = NULL;
1584  for (int i = 0; i < extension_count(); i++) {
1585    if (extension(i)->containing_type() != containing_type) {
1586      if (i > 0) contents.append("}\n\n");
1587      containing_type = extension(i)->containing_type();
1588      strings::SubstituteAndAppend(&contents, "extend .$0 {\n",
1589                                   containing_type->full_name());
1590    }
1591    extension(i)->DebugString(1, &contents);
1592  }
1593  if (extension_count() > 0) contents.append("}\n\n");
1594
1595  return contents;
1596}
1597
1598string Descriptor::DebugString() const {
1599  string contents;
1600  strings::SubstituteAndAppend(&contents, "message $0", name());
1601  DebugString(0, &contents);
1602  return contents;
1603}
1604
1605void Descriptor::DebugString(int depth, string *contents) const {
1606  string prefix(depth * 2, ' ');
1607  ++depth;
1608  contents->append(" {\n");
1609
1610  FormatLineOptions(depth, options(), contents);
1611
1612  // Find all the 'group' types for fields and extensions; we will not output
1613  // their nested definitions (those will be done with their group field
1614  // descriptor).
1615  set<const Descriptor*> groups;
1616  for (int i = 0; i < field_count(); i++) {
1617    if (field(i)->type() == FieldDescriptor::TYPE_GROUP) {
1618      groups.insert(field(i)->message_type());
1619    }
1620  }
1621  for (int i = 0; i < extension_count(); i++) {
1622    if (extension(i)->type() == FieldDescriptor::TYPE_GROUP) {
1623      groups.insert(extension(i)->message_type());
1624    }
1625  }
1626
1627  for (int i = 0; i < nested_type_count(); i++) {
1628    if (groups.count(nested_type(i)) == 0) {
1629      strings::SubstituteAndAppend(contents, "$0  message $1",
1630                                   prefix, nested_type(i)->name());
1631      nested_type(i)->DebugString(depth, contents);
1632    }
1633  }
1634  for (int i = 0; i < enum_type_count(); i++) {
1635    enum_type(i)->DebugString(depth, contents);
1636  }
1637  for (int i = 0; i < field_count(); i++) {
1638    field(i)->DebugString(depth, contents);
1639  }
1640
1641  for (int i = 0; i < extension_range_count(); i++) {
1642    strings::SubstituteAndAppend(contents, "$0  extensions $1 to $2;\n",
1643                                 prefix,
1644                                 extension_range(i)->start,
1645                                 extension_range(i)->end - 1);
1646  }
1647
1648  // Group extensions by what they extend, so they can be printed out together.
1649  const Descriptor* containing_type = NULL;
1650  for (int i = 0; i < extension_count(); i++) {
1651    if (extension(i)->containing_type() != containing_type) {
1652      if (i > 0) strings::SubstituteAndAppend(contents, "$0  }\n", prefix);
1653      containing_type = extension(i)->containing_type();
1654      strings::SubstituteAndAppend(contents, "$0  extend .$1 {\n",
1655                                   prefix, containing_type->full_name());
1656    }
1657    extension(i)->DebugString(depth + 1, contents);
1658  }
1659  if (extension_count() > 0)
1660    strings::SubstituteAndAppend(contents, "$0  }\n", prefix);
1661
1662  strings::SubstituteAndAppend(contents, "$0}\n", prefix);
1663}
1664
1665string FieldDescriptor::DebugString() const {
1666  string contents;
1667  int depth = 0;
1668  if (is_extension()) {
1669    strings::SubstituteAndAppend(&contents, "extend .$0 {\n",
1670                                 containing_type()->full_name());
1671    depth = 1;
1672  }
1673  DebugString(depth, &contents);
1674  if (is_extension()) {
1675     contents.append("}\n");
1676  }
1677  return contents;
1678}
1679
1680void FieldDescriptor::DebugString(int depth, string *contents) const {
1681  string prefix(depth * 2, ' ');
1682  string field_type;
1683  switch (type()) {
1684    case TYPE_MESSAGE:
1685      field_type = "." + message_type()->full_name();
1686      break;
1687    case TYPE_ENUM:
1688      field_type = "." + enum_type()->full_name();
1689      break;
1690    default:
1691      field_type = kTypeToName[type()];
1692  }
1693
1694  strings::SubstituteAndAppend(contents, "$0$1 $2 $3 = $4",
1695                               prefix,
1696                               kLabelToName[label()],
1697                               field_type,
1698                               type() == TYPE_GROUP ? message_type()->name() :
1699                                                      name(),
1700                               number());
1701
1702  bool bracketed = false;
1703  if (has_default_value()) {
1704    bracketed = true;
1705    strings::SubstituteAndAppend(contents, " [default = $0",
1706                                 DefaultValueAsString(true));
1707  }
1708
1709  string formatted_options;
1710  if (FormatBracketedOptions(options(), &formatted_options)) {
1711    contents->append(bracketed ? ", " : " [");
1712    bracketed = true;
1713    contents->append(formatted_options);
1714  }
1715
1716  if (bracketed) {
1717    contents->append("]");
1718  }
1719
1720  if (type() == TYPE_GROUP) {
1721    message_type()->DebugString(depth, contents);
1722  } else {
1723    contents->append(";\n");
1724  }
1725}
1726
1727string EnumDescriptor::DebugString() const {
1728  string contents;
1729  DebugString(0, &contents);
1730  return contents;
1731}
1732
1733void EnumDescriptor::DebugString(int depth, string *contents) const {
1734  string prefix(depth * 2, ' ');
1735  ++depth;
1736  strings::SubstituteAndAppend(contents, "$0enum $1 {\n",
1737                               prefix, name());
1738
1739  FormatLineOptions(depth, options(), contents);
1740
1741  for (int i = 0; i < value_count(); i++) {
1742    value(i)->DebugString(depth, contents);
1743  }
1744  strings::SubstituteAndAppend(contents, "$0}\n", prefix);
1745}
1746
1747string EnumValueDescriptor::DebugString() const {
1748  string contents;
1749  DebugString(0, &contents);
1750  return contents;
1751}
1752
1753void EnumValueDescriptor::DebugString(int depth, string *contents) const {
1754  string prefix(depth * 2, ' ');
1755  strings::SubstituteAndAppend(contents, "$0$1 = $2",
1756                               prefix, name(), number());
1757
1758  string formatted_options;
1759  if (FormatBracketedOptions(options(), &formatted_options)) {
1760    strings::SubstituteAndAppend(contents, " [$0]", formatted_options);
1761  }
1762  contents->append(";\n");
1763}
1764
1765string ServiceDescriptor::DebugString() const {
1766  string contents;
1767  DebugString(&contents);
1768  return contents;
1769}
1770
1771void ServiceDescriptor::DebugString(string *contents) const {
1772  strings::SubstituteAndAppend(contents, "service $0 {\n", name());
1773
1774  FormatLineOptions(1, options(), contents);
1775
1776  for (int i = 0; i < method_count(); i++) {
1777    method(i)->DebugString(1, contents);
1778  }
1779
1780  contents->append("}\n");
1781}
1782
1783string MethodDescriptor::DebugString() const {
1784  string contents;
1785  DebugString(0, &contents);
1786  return contents;
1787}
1788
1789void MethodDescriptor::DebugString(int depth, string *contents) const {
1790  string prefix(depth * 2, ' ');
1791  ++depth;
1792  strings::SubstituteAndAppend(contents, "$0rpc $1(.$2) returns (.$3)",
1793                               prefix, name(),
1794                               input_type()->full_name(),
1795                               output_type()->full_name());
1796
1797  string formatted_options;
1798  if (FormatLineOptions(depth, options(), &formatted_options)) {
1799    strings::SubstituteAndAppend(contents, " {\n$0$1}\n",
1800                                 formatted_options, prefix);
1801  } else {
1802    contents->append(";\n");
1803  }
1804}
1805// ===================================================================
1806
1807namespace {
1808
1809// Represents an options message to interpret. Extension names in the option
1810// name are respolved relative to name_scope. element_name and orig_opt are
1811// used only for error reporting (since the parser records locations against
1812// pointers in the original options, not the mutable copy). The Message must be
1813// one of the Options messages in descriptor.proto.
1814struct OptionsToInterpret {
1815  OptionsToInterpret(const string& ns,
1816                     const string& el,
1817                     const Message* orig_opt,
1818                     Message* opt)
1819      : name_scope(ns),
1820        element_name(el),
1821        original_options(orig_opt),
1822        options(opt) {
1823  }
1824  string name_scope;
1825  string element_name;
1826  const Message* original_options;
1827  Message* options;
1828};
1829
1830}  // namespace
1831
1832class DescriptorBuilder {
1833 public:
1834  DescriptorBuilder(const DescriptorPool* pool,
1835                    DescriptorPool::Tables* tables,
1836                    DescriptorPool::ErrorCollector* error_collector);
1837  ~DescriptorBuilder();
1838
1839  const FileDescriptor* BuildFile(const FileDescriptorProto& proto);
1840
1841 private:
1842  friend class OptionInterpreter;
1843
1844  const DescriptorPool* pool_;
1845  DescriptorPool::Tables* tables_;  // for convenience
1846  DescriptorPool::ErrorCollector* error_collector_;
1847
1848  // As we build descriptors we store copies of the options messages in
1849  // them. We put pointers to those copies in this vector, as we build, so we
1850  // can later (after cross-linking) interpret those options.
1851  vector<OptionsToInterpret> options_to_interpret_;
1852
1853  bool had_errors_;
1854  string filename_;
1855  FileDescriptor* file_;
1856  FileDescriptorTables* file_tables_;
1857
1858  // If LookupSymbol() finds a symbol that is in a file which is not a declared
1859  // dependency of this file, it will fail, but will set
1860  // possible_undeclared_dependency_ to point at that file.  This is only used
1861  // by AddNotDefinedError() to report a more useful error message.
1862  // possible_undeclared_dependency_name_ is the name of the symbol that was
1863  // actually found in possible_undeclared_dependency_, which may be a parent
1864  // of the symbol actually looked for.
1865  const FileDescriptor* possible_undeclared_dependency_;
1866  string possible_undeclared_dependency_name_;
1867
1868  void AddError(const string& element_name,
1869                const Message& descriptor,
1870                DescriptorPool::ErrorCollector::ErrorLocation location,
1871                const string& error);
1872
1873  // Adds an error indicating that undefined_symbol was not defined.  Must
1874  // only be called after LookupSymbol() fails.
1875  void AddNotDefinedError(
1876    const string& element_name,
1877    const Message& descriptor,
1878    DescriptorPool::ErrorCollector::ErrorLocation location,
1879    const string& undefined_symbol);
1880
1881  // Silly helper which determines if the given file is in the given package.
1882  // I.e., either file->package() == package_name or file->package() is a
1883  // nested package within package_name.
1884  bool IsInPackage(const FileDescriptor* file, const string& package_name);
1885
1886  // Like tables_->FindSymbol(), but additionally:
1887  // - Search the pool's underlay if not found in tables_.
1888  // - Insure that the resulting Symbol is from one of the file's declared
1889  //   dependencies.
1890  Symbol FindSymbol(const string& name);
1891
1892  // Like FindSymbol() but does not require that the symbol is in one of the
1893  // file's declared dependencies.
1894  Symbol FindSymbolNotEnforcingDeps(const string& name);
1895
1896  // Like FindSymbol(), but looks up the name relative to some other symbol
1897  // name.  This first searches siblings of relative_to, then siblings of its
1898  // parents, etc.  For example, LookupSymbol("foo.bar", "baz.qux.corge") makes
1899  // the following calls, returning the first non-null result:
1900  // FindSymbol("baz.qux.foo.bar"), FindSymbol("baz.foo.bar"),
1901  // FindSymbol("foo.bar").  If AllowUnknownDependencies() has been called
1902  // on the DescriptorPool, this will generate a placeholder type if
1903  // the name is not found (unless the name itself is malformed).  The
1904  // placeholder_type parameter indicates what kind of placeholder should be
1905  // constructed in this case.  The resolve_mode parameter determines whether
1906  // any symbol is returned, or only symbols that are types.  Note, however,
1907  // that LookupSymbol may still return a non-type symbol in LOOKUP_TYPES mode,
1908  // if it believes that's all it could refer to.  The caller should always
1909  // check that it receives the type of symbol it was expecting.
1910  enum PlaceholderType {
1911    PLACEHOLDER_MESSAGE,
1912    PLACEHOLDER_ENUM,
1913    PLACEHOLDER_EXTENDABLE_MESSAGE
1914  };
1915  enum ResolveMode {
1916    LOOKUP_ALL, LOOKUP_TYPES
1917  };
1918  Symbol LookupSymbol(const string& name, const string& relative_to,
1919                      PlaceholderType placeholder_type = PLACEHOLDER_MESSAGE,
1920                      ResolveMode resolve_mode = LOOKUP_ALL);
1921
1922  // Like LookupSymbol() but will not return a placeholder even if
1923  // AllowUnknownDependencies() has been used.
1924  Symbol LookupSymbolNoPlaceholder(const string& name,
1925                                   const string& relative_to,
1926                                   ResolveMode resolve_mode = LOOKUP_ALL);
1927
1928  // Creates a placeholder type suitable for return from LookupSymbol().  May
1929  // return kNullSymbol if the name is not a valid type name.
1930  Symbol NewPlaceholder(const string& name, PlaceholderType placeholder_type);
1931
1932  // Creates a placeholder file.  Never returns NULL.  This is used when an
1933  // import is not found and AllowUnknownDependencies() is enabled.
1934  const FileDescriptor* NewPlaceholderFile(const string& name);
1935
1936  // Calls tables_->AddSymbol() and records an error if it fails.  Returns
1937  // true if successful or false if failed, though most callers can ignore
1938  // the return value since an error has already been recorded.
1939  bool AddSymbol(const string& full_name,
1940                 const void* parent, const string& name,
1941                 const Message& proto, Symbol symbol);
1942
1943  // Like AddSymbol(), but succeeds if the symbol is already defined as long
1944  // as the existing definition is also a package (because it's OK to define
1945  // the same package in two different files).  Also adds all parents of the
1946  // packgae to the symbol table (e.g. AddPackage("foo.bar", ...) will add
1947  // "foo.bar" and "foo" to the table).
1948  void AddPackage(const string& name, const Message& proto,
1949                  const FileDescriptor* file);
1950
1951  // Checks that the symbol name contains only alphanumeric characters and
1952  // underscores.  Records an error otherwise.
1953  void ValidateSymbolName(const string& name, const string& full_name,
1954                          const Message& proto);
1955
1956  // Like ValidateSymbolName(), but the name is allowed to contain periods and
1957  // an error is indicated by returning false (not recording the error).
1958  bool ValidateQualifiedName(const string& name);
1959
1960  // Used by BUILD_ARRAY macro (below) to avoid having to have the type
1961  // specified as a macro parameter.
1962  template <typename Type>
1963  inline void AllocateArray(int size, Type** output) {
1964    *output = tables_->AllocateArray<Type>(size);
1965  }
1966
1967  // Allocates a copy of orig_options in tables_ and stores it in the
1968  // descriptor. Remembers its uninterpreted options, to be interpreted
1969  // later. DescriptorT must be one of the Descriptor messages from
1970  // descriptor.proto.
1971  template<class DescriptorT> void AllocateOptions(
1972      const typename DescriptorT::OptionsType& orig_options,
1973      DescriptorT* descriptor);
1974  // Specialization for FileOptions.
1975  void AllocateOptions(const FileOptions& orig_options,
1976                       FileDescriptor* descriptor);
1977
1978  // Implementation for AllocateOptions(). Don't call this directly.
1979  template<class DescriptorT> void AllocateOptionsImpl(
1980      const string& name_scope,
1981      const string& element_name,
1982      const typename DescriptorT::OptionsType& orig_options,
1983      DescriptorT* descriptor);
1984
1985  // These methods all have the same signature for the sake of the BUILD_ARRAY
1986  // macro, below.
1987  void BuildMessage(const DescriptorProto& proto,
1988                    const Descriptor* parent,
1989                    Descriptor* result);
1990  void BuildFieldOrExtension(const FieldDescriptorProto& proto,
1991                             const Descriptor* parent,
1992                             FieldDescriptor* result,
1993                             bool is_extension);
1994  void BuildField(const FieldDescriptorProto& proto,
1995                  const Descriptor* parent,
1996                  FieldDescriptor* result) {
1997    BuildFieldOrExtension(proto, parent, result, false);
1998  }
1999  void BuildExtension(const FieldDescriptorProto& proto,
2000                      const Descriptor* parent,
2001                      FieldDescriptor* result) {
2002    BuildFieldOrExtension(proto, parent, result, true);
2003  }
2004  void BuildExtensionRange(const DescriptorProto::ExtensionRange& proto,
2005                           const Descriptor* parent,
2006                           Descriptor::ExtensionRange* result);
2007  void BuildEnum(const EnumDescriptorProto& proto,
2008                 const Descriptor* parent,
2009                 EnumDescriptor* result);
2010  void BuildEnumValue(const EnumValueDescriptorProto& proto,
2011                      const EnumDescriptor* parent,
2012                      EnumValueDescriptor* result);
2013  void BuildService(const ServiceDescriptorProto& proto,
2014                    const void* dummy,
2015                    ServiceDescriptor* result);
2016  void BuildMethod(const MethodDescriptorProto& proto,
2017                   const ServiceDescriptor* parent,
2018                   MethodDescriptor* result);
2019
2020  // Must be run only after building.
2021  //
2022  // NOTE: Options will not be available during cross-linking, as they
2023  // have not yet been interpreted. Defer any handling of options to the
2024  // Validate*Options methods.
2025  void CrossLinkFile(FileDescriptor* file, const FileDescriptorProto& proto);
2026  void CrossLinkMessage(Descriptor* message, const DescriptorProto& proto);
2027  void CrossLinkField(FieldDescriptor* field,
2028                      const FieldDescriptorProto& proto);
2029  void CrossLinkEnum(EnumDescriptor* enum_type,
2030                     const EnumDescriptorProto& proto);
2031  void CrossLinkEnumValue(EnumValueDescriptor* enum_value,
2032                          const EnumValueDescriptorProto& proto);
2033  void CrossLinkService(ServiceDescriptor* service,
2034                        const ServiceDescriptorProto& proto);
2035  void CrossLinkMethod(MethodDescriptor* method,
2036                       const MethodDescriptorProto& proto);
2037
2038  // Must be run only after cross-linking.
2039  void InterpretOptions();
2040
2041  // A helper class for interpreting options.
2042  class OptionInterpreter {
2043   public:
2044    // Creates an interpreter that operates in the context of the pool of the
2045    // specified builder, which must not be NULL. We don't take ownership of the
2046    // builder.
2047    explicit OptionInterpreter(DescriptorBuilder* builder);
2048
2049    ~OptionInterpreter();
2050
2051    // Interprets the uninterpreted options in the specified Options message.
2052    // On error, calls AddError() on the underlying builder and returns false.
2053    // Otherwise returns true.
2054    bool InterpretOptions(OptionsToInterpret* options_to_interpret);
2055
2056   private:
2057    // Interprets uninterpreted_option_ on the specified message, which
2058    // must be the mutable copy of the original options message to which
2059    // uninterpreted_option_ belongs.
2060    bool InterpretSingleOption(Message* options);
2061
2062    // Adds the uninterpreted_option to the given options message verbatim.
2063    // Used when AllowUnknownDependencies() is in effect and we can't find
2064    // the option's definition.
2065    void AddWithoutInterpreting(const UninterpretedOption& uninterpreted_option,
2066                                Message* options);
2067
2068    // A recursive helper function that drills into the intermediate fields
2069    // in unknown_fields to check if field innermost_field is set on the
2070    // innermost message. Returns false and sets an error if so.
2071    bool ExamineIfOptionIsSet(
2072        vector<const FieldDescriptor*>::const_iterator intermediate_fields_iter,
2073        vector<const FieldDescriptor*>::const_iterator intermediate_fields_end,
2074        const FieldDescriptor* innermost_field, const string& debug_msg_name,
2075        const UnknownFieldSet& unknown_fields);
2076
2077    // Validates the value for the option field of the currently interpreted
2078    // option and then sets it on the unknown_field.
2079    bool SetOptionValue(const FieldDescriptor* option_field,
2080                        UnknownFieldSet* unknown_fields);
2081
2082    // Convenience functions to set an int field the right way, depending on
2083    // its wire type (a single int CppType can represent multiple wire types).
2084    void SetInt32(int number, int32 value, FieldDescriptor::Type type,
2085                  UnknownFieldSet* unknown_fields);
2086    void SetInt64(int number, int64 value, FieldDescriptor::Type type,
2087                  UnknownFieldSet* unknown_fields);
2088    void SetUInt32(int number, uint32 value, FieldDescriptor::Type type,
2089                   UnknownFieldSet* unknown_fields);
2090    void SetUInt64(int number, uint64 value, FieldDescriptor::Type type,
2091                   UnknownFieldSet* unknown_fields);
2092
2093    // A helper function that adds an error at the specified location of the
2094    // option we're currently interpreting, and returns false.
2095    bool AddOptionError(DescriptorPool::ErrorCollector::ErrorLocation location,
2096                        const string& msg) {
2097      builder_->AddError(options_to_interpret_->element_name,
2098                         *uninterpreted_option_, location, msg);
2099      return false;
2100    }
2101
2102    // A helper function that adds an error at the location of the option name
2103    // and returns false.
2104    bool AddNameError(const string& msg) {
2105      return AddOptionError(DescriptorPool::ErrorCollector::OPTION_NAME, msg);
2106    }
2107
2108    // A helper function that adds an error at the location of the option name
2109    // and returns false.
2110    bool AddValueError(const string& msg) {
2111      return AddOptionError(DescriptorPool::ErrorCollector::OPTION_VALUE, msg);
2112    }
2113
2114    // We interpret against this builder's pool. Is never NULL. We don't own
2115    // this pointer.
2116    DescriptorBuilder* builder_;
2117
2118    // The options we're currently interpreting, or NULL if we're not in a call
2119    // to InterpretOptions.
2120    const OptionsToInterpret* options_to_interpret_;
2121
2122    // The option we're currently interpreting within options_to_interpret_, or
2123    // NULL if we're not in a call to InterpretOptions(). This points to a
2124    // submessage of the original option, not the mutable copy. Therefore we
2125    // can use it to find locations recorded by the parser.
2126    const UninterpretedOption* uninterpreted_option_;
2127
2128    GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(OptionInterpreter);
2129  };
2130
2131  // Work-around for broken compilers:  According to the C++ standard,
2132  // OptionInterpreter should have access to the private members of any class
2133  // which has declared DescriptorBuilder as a friend.  Unfortunately some old
2134  // versions of GCC and other compilers do not implement this correctly.  So,
2135  // we have to have these intermediate methods to provide access.  We also
2136  // redundantly declare OptionInterpreter a friend just to make things extra
2137  // clear for these bad compilers.
2138  friend class OptionInterpreter;
2139  static inline bool get_allow_unknown(const DescriptorPool* pool) {
2140    return pool->allow_unknown_;
2141  }
2142  static inline bool get_is_placeholder(const Descriptor* descriptor) {
2143    return descriptor->is_placeholder_;
2144  }
2145
2146  // Must be run only after options have been interpreted.
2147  //
2148  // NOTE: Validation code must only reference the options in the mutable
2149  // descriptors, which are the ones that have been interpreted. The const
2150  // proto references are passed in only so they can be provided to calls to
2151  // AddError(). Do not look at their options, which have not been interpreted.
2152  void ValidateFileOptions(FileDescriptor* file,
2153                           const FileDescriptorProto& proto);
2154  void ValidateMessageOptions(Descriptor* message,
2155                              const DescriptorProto& proto);
2156  void ValidateFieldOptions(FieldDescriptor* field,
2157                            const FieldDescriptorProto& proto);
2158  void ValidateEnumOptions(EnumDescriptor* enm,
2159                           const EnumDescriptorProto& proto);
2160  void ValidateEnumValueOptions(EnumValueDescriptor* enum_value,
2161                                const EnumValueDescriptorProto& proto);
2162  void ValidateServiceOptions(ServiceDescriptor* service,
2163                              const ServiceDescriptorProto& proto);
2164  void ValidateMethodOptions(MethodDescriptor* method,
2165                             const MethodDescriptorProto& proto);
2166
2167  void ValidateMapKey(FieldDescriptor* field,
2168                      const FieldDescriptorProto& proto);
2169};
2170
2171const FileDescriptor* DescriptorPool::BuildFile(
2172    const FileDescriptorProto& proto) {
2173  GOOGLE_CHECK(fallback_database_ == NULL)
2174    << "Cannot call BuildFile on a DescriptorPool that uses a "
2175       "DescriptorDatabase.  You must instead find a way to get your file "
2176       "into the underlying database.";
2177  GOOGLE_CHECK(mutex_ == NULL);   // Implied by the above GOOGLE_CHECK.
2178  return DescriptorBuilder(this, tables_.get(), NULL).BuildFile(proto);
2179}
2180
2181const FileDescriptor* DescriptorPool::BuildFileCollectingErrors(
2182    const FileDescriptorProto& proto,
2183    ErrorCollector* error_collector) {
2184  GOOGLE_CHECK(fallback_database_ == NULL)
2185    << "Cannot call BuildFile on a DescriptorPool that uses a "
2186       "DescriptorDatabase.  You must instead find a way to get your file "
2187       "into the underlying database.";
2188  GOOGLE_CHECK(mutex_ == NULL);   // Implied by the above GOOGLE_CHECK.
2189  return DescriptorBuilder(this, tables_.get(),
2190                           error_collector).BuildFile(proto);
2191}
2192
2193const FileDescriptor* DescriptorPool::BuildFileFromDatabase(
2194    const FileDescriptorProto& proto) const {
2195  mutex_->AssertHeld();
2196  return DescriptorBuilder(this, tables_.get(),
2197                           default_error_collector_).BuildFile(proto);
2198}
2199
2200DescriptorBuilder::DescriptorBuilder(
2201    const DescriptorPool* pool,
2202    DescriptorPool::Tables* tables,
2203    DescriptorPool::ErrorCollector* error_collector)
2204  : pool_(pool),
2205    tables_(tables),
2206    error_collector_(error_collector),
2207    had_errors_(false),
2208    possible_undeclared_dependency_(NULL) {}
2209
2210DescriptorBuilder::~DescriptorBuilder() {}
2211
2212void DescriptorBuilder::AddError(
2213    const string& element_name,
2214    const Message& descriptor,
2215    DescriptorPool::ErrorCollector::ErrorLocation location,
2216    const string& error) {
2217  if (error_collector_ == NULL) {
2218    if (!had_errors_) {
2219      GOOGLE_LOG(ERROR) << "Invalid proto descriptor for file \"" << filename_
2220                 << "\":";
2221    }
2222    GOOGLE_LOG(ERROR) << "  " << element_name << ": " << error;
2223  } else {
2224    error_collector_->AddError(filename_, element_name,
2225                               &descriptor, location, error);
2226  }
2227  had_errors_ = true;
2228}
2229
2230void DescriptorBuilder::AddNotDefinedError(
2231    const string& element_name,
2232    const Message& descriptor,
2233    DescriptorPool::ErrorCollector::ErrorLocation location,
2234    const string& undefined_symbol) {
2235  if (possible_undeclared_dependency_ == NULL) {
2236    AddError(element_name, descriptor, location,
2237             "\"" + undefined_symbol + "\" is not defined.");
2238  } else {
2239    AddError(element_name, descriptor, location,
2240             "\"" + possible_undeclared_dependency_name_ +
2241             "\" seems to be defined in \"" +
2242             possible_undeclared_dependency_->name() + "\", which is not "
2243             "imported by \"" + filename_ + "\".  To use it here, please "
2244             "add the necessary import.");
2245  }
2246}
2247
2248bool DescriptorBuilder::IsInPackage(const FileDescriptor* file,
2249                                    const string& package_name) {
2250  return HasPrefixString(file->package(), package_name) &&
2251           (file->package().size() == package_name.size() ||
2252            file->package()[package_name.size()] == '.');
2253}
2254
2255Symbol DescriptorBuilder::FindSymbolNotEnforcingDeps(const string& name) {
2256  Symbol result;
2257
2258  // We need to search our pool and all its underlays.
2259  const DescriptorPool* pool = pool_;
2260  while (true) {
2261    // If we are looking at an underlay, we must lock its mutex_, since we are
2262    // accessing the underlay's tables_ dircetly.
2263    MutexLockMaybe lock((pool == pool_) ? NULL : pool->mutex_);
2264
2265    // Note that we don't have to check fallback_database_ here because the
2266    // symbol has to be in one of its file's direct dependencies, and we have
2267    // already loaded those by the time we get here.
2268    result = pool->tables_->FindSymbol(name);
2269    if (!result.IsNull()) break;
2270    if (pool->underlay_ == NULL) return kNullSymbol;
2271    pool = pool->underlay_;
2272  }
2273
2274  return result;
2275}
2276
2277Symbol DescriptorBuilder::FindSymbol(const string& name) {
2278  Symbol result = FindSymbolNotEnforcingDeps(name);
2279
2280  if (!pool_->enforce_dependencies_) {
2281    // Hack for CompilerUpgrader.
2282    return result;
2283  }
2284
2285  // Only find symbols which were defined in this file or one of its
2286  // dependencies.
2287  const FileDescriptor* file = result.GetFile();
2288  if (file == file_) return result;
2289  for (int i = 0; i < file_->dependency_count(); i++) {
2290    if (file == file_->dependency(i)) return result;
2291  }
2292
2293  if (result.type == Symbol::PACKAGE) {
2294    // Arg, this is overcomplicated.  The symbol is a package name.  It could
2295    // be that the package was defined in multiple files.  result.GetFile()
2296    // returns the first file we saw that used this package.  We've determined
2297    // that that file is not a direct dependency of the file we are currently
2298    // building, but it could be that some other file which *is* a direct
2299    // dependency also defines the same package.  We can't really rule out this
2300    // symbol unless none of the dependencies define it.
2301    if (IsInPackage(file_, name)) return result;
2302    for (int i = 0; i < file_->dependency_count(); i++) {
2303      // Note:  A dependency may be NULL if it was not found or had errors.
2304      if (file_->dependency(i) != NULL &&
2305          IsInPackage(file_->dependency(i), name)) {
2306        return result;
2307      }
2308    }
2309  }
2310
2311  possible_undeclared_dependency_ = file;
2312  possible_undeclared_dependency_name_ = name;
2313  return kNullSymbol;
2314}
2315
2316Symbol DescriptorBuilder::LookupSymbolNoPlaceholder(
2317    const string& name, const string& relative_to, ResolveMode resolve_mode) {
2318  possible_undeclared_dependency_ = NULL;
2319
2320  if (name.size() > 0 && name[0] == '.') {
2321    // Fully-qualified name.
2322    return FindSymbol(name.substr(1));
2323  }
2324
2325  // If name is something like "Foo.Bar.baz", and symbols named "Foo" are
2326  // defined in multiple parent scopes, we only want to find "Bar.baz" in the
2327  // innermost one.  E.g., the following should produce an error:
2328  //   message Bar { message Baz {} }
2329  //   message Foo {
2330  //     message Bar {
2331  //     }
2332  //     optional Bar.Baz baz = 1;
2333  //   }
2334  // So, we look for just "Foo" first, then look for "Bar.baz" within it if
2335  // found.
2336  int name_dot_pos = name.find_first_of('.');
2337  string first_part_of_name;
2338  if (name_dot_pos == string::npos) {
2339    first_part_of_name = name;
2340  } else {
2341    first_part_of_name = name.substr(0, name_dot_pos);
2342  }
2343
2344  string scope_to_try(relative_to);
2345
2346  while (true) {
2347    // Chop off the last component of the scope.
2348    string::size_type dot_pos = scope_to_try.find_last_of('.');
2349    if (dot_pos == string::npos) {
2350      return FindSymbol(name);
2351    } else {
2352      scope_to_try.erase(dot_pos);
2353    }
2354
2355    // Append ".first_part_of_name" and try to find.
2356    string::size_type old_size = scope_to_try.size();
2357    scope_to_try.append(1, '.');
2358    scope_to_try.append(first_part_of_name);
2359    Symbol result = FindSymbol(scope_to_try);
2360    if (!result.IsNull()) {
2361      if (first_part_of_name.size() < name.size()) {
2362        // name is a compound symbol, of which we only found the first part.
2363        // Now try to look up the rest of it.
2364        if (result.IsAggregate()) {
2365          scope_to_try.append(name, first_part_of_name.size(),
2366                              name.size() - first_part_of_name.size());
2367          return FindSymbol(scope_to_try);
2368        } else {
2369          // We found a symbol but it's not an aggregate.  Continue the loop.
2370        }
2371      } else {
2372        if (resolve_mode == LOOKUP_TYPES && !result.IsType()) {
2373          // We found a symbol but it's not a type.  Continue the loop.
2374        } else {
2375          return result;
2376        }
2377      }
2378    }
2379
2380    // Not found.  Remove the name so we can try again.
2381    scope_to_try.erase(old_size);
2382  }
2383}
2384
2385Symbol DescriptorBuilder::LookupSymbol(
2386    const string& name, const string& relative_to,
2387    PlaceholderType placeholder_type, ResolveMode resolve_mode) {
2388  Symbol result = LookupSymbolNoPlaceholder(
2389      name, relative_to, resolve_mode);
2390  if (result.IsNull() && pool_->allow_unknown_) {
2391    // Not found, but AllowUnknownDependencies() is enabled.  Return a
2392    // placeholder instead.
2393    result = NewPlaceholder(name, placeholder_type);
2394  }
2395  return result;
2396}
2397
2398Symbol DescriptorBuilder::NewPlaceholder(const string& name,
2399                                         PlaceholderType placeholder_type) {
2400  // Compute names.
2401  const string* placeholder_full_name;
2402  const string* placeholder_name;
2403  const string* placeholder_package;
2404
2405  if (!ValidateQualifiedName(name)) return kNullSymbol;
2406  if (name[0] == '.') {
2407    // Fully-qualified.
2408    placeholder_full_name = tables_->AllocateString(name.substr(1));
2409  } else {
2410    placeholder_full_name = tables_->AllocateString(name);
2411  }
2412
2413  string::size_type dotpos = placeholder_full_name->find_last_of('.');
2414  if (dotpos != string::npos) {
2415    placeholder_package = tables_->AllocateString(
2416      placeholder_full_name->substr(0, dotpos));
2417    placeholder_name = tables_->AllocateString(
2418      placeholder_full_name->substr(dotpos + 1));
2419  } else {
2420    placeholder_package = &kEmptyString;
2421    placeholder_name = placeholder_full_name;
2422  }
2423
2424  // Create the placeholders.
2425  FileDescriptor* placeholder_file = tables_->Allocate<FileDescriptor>();
2426  memset(placeholder_file, 0, sizeof(*placeholder_file));
2427
2428  placeholder_file->name_ =
2429    tables_->AllocateString(*placeholder_full_name + ".placeholder.proto");
2430  placeholder_file->package_ = placeholder_package;
2431  placeholder_file->pool_ = pool_;
2432  placeholder_file->options_ = &FileOptions::default_instance();
2433  placeholder_file->tables_ = &FileDescriptorTables::kEmpty;
2434  // All other fields are zero or NULL.
2435
2436  if (placeholder_type == PLACEHOLDER_ENUM) {
2437    placeholder_file->enum_type_count_ = 1;
2438    placeholder_file->enum_types_ =
2439      tables_->AllocateArray<EnumDescriptor>(1);
2440
2441    EnumDescriptor* placeholder_enum = &placeholder_file->enum_types_[0];
2442    memset(placeholder_enum, 0, sizeof(*placeholder_enum));
2443
2444    placeholder_enum->full_name_ = placeholder_full_name;
2445    placeholder_enum->name_ = placeholder_name;
2446    placeholder_enum->file_ = placeholder_file;
2447    placeholder_enum->options_ = &EnumOptions::default_instance();
2448    placeholder_enum->is_placeholder_ = true;
2449    placeholder_enum->is_unqualified_placeholder_ = (name[0] != '.');
2450
2451    // Enums must have at least one value.
2452    placeholder_enum->value_count_ = 1;
2453    placeholder_enum->values_ = tables_->AllocateArray<EnumValueDescriptor>(1);
2454
2455    EnumValueDescriptor* placeholder_value = &placeholder_enum->values_[0];
2456    memset(placeholder_value, 0, sizeof(*placeholder_value));
2457
2458    placeholder_value->name_ = tables_->AllocateString("PLACEHOLDER_VALUE");
2459    // Note that enum value names are siblings of their type, not children.
2460    placeholder_value->full_name_ =
2461      placeholder_package->empty() ? placeholder_value->name_ :
2462        tables_->AllocateString(*placeholder_package + ".PLACEHOLDER_VALUE");
2463
2464    placeholder_value->number_ = 0;
2465    placeholder_value->type_ = placeholder_enum;
2466    placeholder_value->options_ = &EnumValueOptions::default_instance();
2467
2468    return Symbol(placeholder_enum);
2469  } else {
2470    placeholder_file->message_type_count_ = 1;
2471    placeholder_file->message_types_ =
2472      tables_->AllocateArray<Descriptor>(1);
2473
2474    Descriptor* placeholder_message = &placeholder_file->message_types_[0];
2475    memset(placeholder_message, 0, sizeof(*placeholder_message));
2476
2477    placeholder_message->full_name_ = placeholder_full_name;
2478    placeholder_message->name_ = placeholder_name;
2479    placeholder_message->file_ = placeholder_file;
2480    placeholder_message->options_ = &MessageOptions::default_instance();
2481    placeholder_message->is_placeholder_ = true;
2482    placeholder_message->is_unqualified_placeholder_ = (name[0] != '.');
2483
2484    if (placeholder_type == PLACEHOLDER_EXTENDABLE_MESSAGE) {
2485      placeholder_message->extension_range_count_ = 1;
2486      placeholder_message->extension_ranges_ =
2487        tables_->AllocateArray<Descriptor::ExtensionRange>(1);
2488      placeholder_message->extension_ranges_->start = 1;
2489      // kMaxNumber + 1 because ExtensionRange::end is exclusive.
2490      placeholder_message->extension_ranges_->end =
2491        FieldDescriptor::kMaxNumber + 1;
2492    }
2493
2494    return Symbol(placeholder_message);
2495  }
2496}
2497
2498const FileDescriptor* DescriptorBuilder::NewPlaceholderFile(
2499    const string& name) {
2500  FileDescriptor* placeholder = tables_->Allocate<FileDescriptor>();
2501  memset(placeholder, 0, sizeof(*placeholder));
2502
2503  placeholder->name_ = tables_->AllocateString(name);
2504  placeholder->package_ = &kEmptyString;
2505  placeholder->pool_ = pool_;
2506  placeholder->options_ = &FileOptions::default_instance();
2507  placeholder->tables_ = &FileDescriptorTables::kEmpty;
2508  // All other fields are zero or NULL.
2509
2510  return placeholder;
2511}
2512
2513bool DescriptorBuilder::AddSymbol(
2514    const string& full_name, const void* parent, const string& name,
2515    const Message& proto, Symbol symbol) {
2516  // If the caller passed NULL for the parent, the symbol is at file scope.
2517  // Use its file as the parent instead.
2518  if (parent == NULL) parent = file_;
2519
2520  if (tables_->AddSymbol(full_name, symbol)) {
2521    if (!file_tables_->AddAliasUnderParent(parent, name, symbol)) {
2522      GOOGLE_LOG(DFATAL) << "\"" << full_name << "\" not previously defined in "
2523                     "symbols_by_name_, but was defined in symbols_by_parent_; "
2524                     "this shouldn't be possible.";
2525      return false;
2526    }
2527    return true;
2528  } else {
2529    const FileDescriptor* other_file = tables_->FindSymbol(full_name).GetFile();
2530    if (other_file == file_) {
2531      string::size_type dot_pos = full_name.find_last_of('.');
2532      if (dot_pos == string::npos) {
2533        AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2534                 "\"" + full_name + "\" is already defined.");
2535      } else {
2536        AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2537                 "\"" + full_name.substr(dot_pos + 1) +
2538                 "\" is already defined in \"" +
2539                 full_name.substr(0, dot_pos) + "\".");
2540      }
2541    } else {
2542      // Symbol seems to have been defined in a different file.
2543      AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2544               "\"" + full_name + "\" is already defined in file \"" +
2545               other_file->name() + "\".");
2546    }
2547    return false;
2548  }
2549}
2550
2551void DescriptorBuilder::AddPackage(
2552    const string& name, const Message& proto, const FileDescriptor* file) {
2553  if (tables_->AddSymbol(name, Symbol(file))) {
2554    // Success.  Also add parent package, if any.
2555    string::size_type dot_pos = name.find_last_of('.');
2556    if (dot_pos == string::npos) {
2557      // No parents.
2558      ValidateSymbolName(name, name, proto);
2559    } else {
2560      // Has parent.
2561      string* parent_name = tables_->AllocateString(name.substr(0, dot_pos));
2562      AddPackage(*parent_name, proto, file);
2563      ValidateSymbolName(name.substr(dot_pos + 1), name, proto);
2564    }
2565  } else {
2566    Symbol existing_symbol = tables_->FindSymbol(name);
2567    // It's OK to redefine a package.
2568    if (existing_symbol.type != Symbol::PACKAGE) {
2569      // Symbol seems to have been defined in a different file.
2570      AddError(name, proto, DescriptorPool::ErrorCollector::NAME,
2571               "\"" + name + "\" is already defined (as something other than "
2572               "a package) in file \"" + existing_symbol.GetFile()->name() +
2573               "\".");
2574    }
2575  }
2576}
2577
2578void DescriptorBuilder::ValidateSymbolName(
2579    const string& name, const string& full_name, const Message& proto) {
2580  if (name.empty()) {
2581    AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2582             "Missing name.");
2583  } else {
2584    for (int i = 0; i < name.size(); i++) {
2585      // I don't trust isalnum() due to locales.  :(
2586      if ((name[i] < 'a' || 'z' < name[i]) &&
2587          (name[i] < 'A' || 'Z' < name[i]) &&
2588          (name[i] < '0' || '9' < name[i]) &&
2589          (name[i] != '_')) {
2590        AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2591                 "\"" + name + "\" is not a valid identifier.");
2592      }
2593    }
2594  }
2595}
2596
2597bool DescriptorBuilder::ValidateQualifiedName(const string& name) {
2598  bool last_was_period = false;
2599
2600  for (int i = 0; i < name.size(); i++) {
2601    // I don't trust isalnum() due to locales.  :(
2602    if (('a' <= name[i] && name[i] <= 'z') ||
2603        ('A' <= name[i] && name[i] <= 'Z') ||
2604        ('0' <= name[i] && name[i] <= '9') ||
2605        (name[i] == '_')) {
2606      last_was_period = false;
2607    } else if (name[i] == '.') {
2608      if (last_was_period) return false;
2609      last_was_period = true;
2610    } else {
2611      return false;
2612    }
2613  }
2614
2615  return !name.empty() && !last_was_period;
2616}
2617
2618// -------------------------------------------------------------------
2619
2620// This generic implementation is good for all descriptors except
2621// FileDescriptor.
2622template<class DescriptorT> void DescriptorBuilder::AllocateOptions(
2623    const typename DescriptorT::OptionsType& orig_options,
2624    DescriptorT* descriptor) {
2625  AllocateOptionsImpl(descriptor->full_name(), descriptor->full_name(),
2626                      orig_options, descriptor);
2627}
2628
2629// We specialize for FileDescriptor.
2630void DescriptorBuilder::AllocateOptions(const FileOptions& orig_options,
2631                                        FileDescriptor* descriptor) {
2632  // We add the dummy token so that LookupSymbol does the right thing.
2633  AllocateOptionsImpl(descriptor->package() + ".dummy", descriptor->name(),
2634                      orig_options, descriptor);
2635}
2636
2637template<class DescriptorT> void DescriptorBuilder::AllocateOptionsImpl(
2638    const string& name_scope,
2639    const string& element_name,
2640    const typename DescriptorT::OptionsType& orig_options,
2641    DescriptorT* descriptor) {
2642  // We need to use a dummy pointer to work around a bug in older versions of
2643  // GCC.  Otherwise, the following two lines could be replaced with:
2644  //   typename DescriptorT::OptionsType* options =
2645  //       tables_->AllocateMessage<typename DescriptorT::OptionsType>();
2646  typename DescriptorT::OptionsType* const dummy = NULL;
2647  typename DescriptorT::OptionsType* options = tables_->AllocateMessage(dummy);
2648  options->CopyFrom(orig_options);
2649  descriptor->options_ = options;
2650
2651  // Don't add to options_to_interpret_ unless there were uninterpreted
2652  // options.  This not only avoids unnecessary work, but prevents a
2653  // bootstrapping problem when building descriptors for descriptor.proto.
2654  // descriptor.proto does not contain any uninterpreted options, but
2655  // attempting to interpret options anyway will cause
2656  // OptionsType::GetDescriptor() to be called which may then deadlock since
2657  // we're still trying to build it.
2658  if (options->uninterpreted_option_size() > 0) {
2659    options_to_interpret_.push_back(
2660        OptionsToInterpret(name_scope, element_name, &orig_options, options));
2661  }
2662}
2663
2664
2665// A common pattern:  We want to convert a repeated field in the descriptor
2666// to an array of values, calling some method to build each value.
2667#define BUILD_ARRAY(INPUT, OUTPUT, NAME, METHOD, PARENT)             \
2668  OUTPUT->NAME##_count_ = INPUT.NAME##_size();                       \
2669  AllocateArray(INPUT.NAME##_size(), &OUTPUT->NAME##s_);             \
2670  for (int i = 0; i < INPUT.NAME##_size(); i++) {                    \
2671    METHOD(INPUT.NAME(i), PARENT, OUTPUT->NAME##s_ + i);             \
2672  }
2673
2674const FileDescriptor* DescriptorBuilder::BuildFile(
2675    const FileDescriptorProto& proto) {
2676  filename_ = proto.name();
2677
2678  // Check if the file already exists and is identical to the one being built.
2679  // Note:  This only works if the input is canonical -- that is, it
2680  //   fully-qualifies all type names, has no UninterpretedOptions, etc.
2681  //   This is fine, because this idempotency "feature" really only exists to
2682  //   accomodate one hack in the proto1->proto2 migration layer.
2683  const FileDescriptor* existing_file = tables_->FindFile(filename_);
2684  if (existing_file != NULL) {
2685    // File already in pool.  Compare the existing one to the input.
2686    FileDescriptorProto existing_proto;
2687    existing_file->CopyTo(&existing_proto);
2688    if (existing_proto.SerializeAsString() == proto.SerializeAsString()) {
2689      // They're identical.  Return the existing descriptor.
2690      return existing_file;
2691    }
2692
2693    // Not a match.  The error will be detected and handled later.
2694  }
2695
2696  // Check to see if this file is already on the pending files list.
2697  // TODO(kenton):  Allow recursive imports?  It may not work with some
2698  //   (most?) programming languages.  E.g., in C++, a forward declaration
2699  //   of a type is not sufficient to allow it to be used even in a
2700  //   generated header file due to inlining.  This could perhaps be
2701  //   worked around using tricks involving inserting #include statements
2702  //   mid-file, but that's pretty ugly, and I'm pretty sure there are
2703  //   some languages out there that do not allow recursive dependencies
2704  //   at all.
2705  for (int i = 0; i < tables_->pending_files_.size(); i++) {
2706    if (tables_->pending_files_[i] == proto.name()) {
2707      string error_message("File recursively imports itself: ");
2708      for (; i < tables_->pending_files_.size(); i++) {
2709        error_message.append(tables_->pending_files_[i]);
2710        error_message.append(" -> ");
2711      }
2712      error_message.append(proto.name());
2713
2714      AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
2715               error_message);
2716      return NULL;
2717    }
2718  }
2719
2720  // If we have a fallback_database_, attempt to load all dependencies now,
2721  // before checkpointing tables_.  This avoids confusion with recursive
2722  // checkpoints.
2723  if (pool_->fallback_database_ != NULL) {
2724    tables_->pending_files_.push_back(proto.name());
2725    for (int i = 0; i < proto.dependency_size(); i++) {
2726      if (tables_->FindFile(proto.dependency(i)) == NULL &&
2727          (pool_->underlay_ == NULL ||
2728           pool_->underlay_->FindFileByName(proto.dependency(i)) == NULL)) {
2729        // We don't care what this returns since we'll find out below anyway.
2730        pool_->TryFindFileInFallbackDatabase(proto.dependency(i));
2731      }
2732    }
2733    tables_->pending_files_.pop_back();
2734  }
2735
2736  // Checkpoint the tables so that we can roll back if something goes wrong.
2737  tables_->Checkpoint();
2738
2739  FileDescriptor* result = tables_->Allocate<FileDescriptor>();
2740  file_ = result;
2741
2742  file_tables_ = tables_->AllocateFileTables();
2743  file_->tables_ = file_tables_;
2744
2745  if (!proto.has_name()) {
2746    AddError("", proto, DescriptorPool::ErrorCollector::OTHER,
2747             "Missing field: FileDescriptorProto.name.");
2748  }
2749
2750  result->name_ = tables_->AllocateString(proto.name());
2751  if (proto.has_package()) {
2752    result->package_ = tables_->AllocateString(proto.package());
2753  } else {
2754    // We cannot rely on proto.package() returning a valid string if
2755    // proto.has_package() is false, because we might be running at static
2756    // initialization time, in which case default values have not yet been
2757    // initialized.
2758    result->package_ = tables_->AllocateString("");
2759  }
2760  result->pool_ = pool_;
2761
2762  // Add to tables.
2763  if (!tables_->AddFile(result)) {
2764    AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
2765             "A file with this name is already in the pool.");
2766    // Bail out early so that if this is actually the exact same file, we
2767    // don't end up reporting that every single symbol is already defined.
2768    tables_->Rollback();
2769    return NULL;
2770  }
2771  if (!result->package().empty()) {
2772    AddPackage(result->package(), proto, result);
2773  }
2774
2775  // Make sure all dependencies are loaded.
2776  set<string> seen_dependencies;
2777  result->dependency_count_ = proto.dependency_size();
2778  result->dependencies_ =
2779    tables_->AllocateArray<const FileDescriptor*>(proto.dependency_size());
2780  for (int i = 0; i < proto.dependency_size(); i++) {
2781    if (!seen_dependencies.insert(proto.dependency(i)).second) {
2782      AddError(proto.name(), proto,
2783               DescriptorPool::ErrorCollector::OTHER,
2784               "Import \"" + proto.dependency(i) + "\" was listed twice.");
2785    }
2786
2787    const FileDescriptor* dependency = tables_->FindFile(proto.dependency(i));
2788    if (dependency == NULL && pool_->underlay_ != NULL) {
2789      dependency = pool_->underlay_->FindFileByName(proto.dependency(i));
2790    }
2791
2792    if (dependency == NULL) {
2793      if (pool_->allow_unknown_) {
2794        dependency = NewPlaceholderFile(proto.dependency(i));
2795      } else {
2796        string message;
2797        if (pool_->fallback_database_ == NULL) {
2798          message = "Import \"" + proto.dependency(i) +
2799                    "\" has not been loaded.";
2800        } else {
2801          message = "Import \"" + proto.dependency(i) +
2802                    "\" was not found or had errors.";
2803        }
2804        AddError(proto.name(), proto,
2805                 DescriptorPool::ErrorCollector::OTHER,
2806                 message);
2807      }
2808    }
2809
2810    result->dependencies_[i] = dependency;
2811  }
2812
2813  // Convert children.
2814  BUILD_ARRAY(proto, result, message_type, BuildMessage  , NULL);
2815  BUILD_ARRAY(proto, result, enum_type   , BuildEnum     , NULL);
2816  BUILD_ARRAY(proto, result, service     , BuildService  , NULL);
2817  BUILD_ARRAY(proto, result, extension   , BuildExtension, NULL);
2818
2819  // Copy options.
2820  if (!proto.has_options()) {
2821    result->options_ = NULL;  // Will set to default_instance later.
2822  } else {
2823    AllocateOptions(proto.options(), result);
2824  }
2825
2826  // Note that the following steps must occur in exactly the specified order.
2827
2828  // Cross-link.
2829  CrossLinkFile(result, proto);
2830
2831  // Interpret any remaining uninterpreted options gathered into
2832  // options_to_interpret_ during descriptor building.  Cross-linking has made
2833  // extension options known, so all interpretations should now succeed.
2834  if (!had_errors_) {
2835    OptionInterpreter option_interpreter(this);
2836    for (vector<OptionsToInterpret>::iterator iter =
2837             options_to_interpret_.begin();
2838         iter != options_to_interpret_.end(); ++iter) {
2839      option_interpreter.InterpretOptions(&(*iter));
2840    }
2841    options_to_interpret_.clear();
2842  }
2843
2844  // Validate options.
2845  if (!had_errors_) {
2846    ValidateFileOptions(result, proto);
2847  }
2848
2849  if (had_errors_) {
2850    tables_->Rollback();
2851    return NULL;
2852  } else {
2853    tables_->Checkpoint();
2854    return result;
2855  }
2856}
2857
2858void DescriptorBuilder::BuildMessage(const DescriptorProto& proto,
2859                                     const Descriptor* parent,
2860                                     Descriptor* result) {
2861  const string& scope = (parent == NULL) ?
2862    file_->package() : parent->full_name();
2863  string* full_name = tables_->AllocateString(scope);
2864  if (!full_name->empty()) full_name->append(1, '.');
2865  full_name->append(proto.name());
2866
2867  ValidateSymbolName(proto.name(), *full_name, proto);
2868
2869  result->name_            = tables_->AllocateString(proto.name());
2870  result->full_name_       = full_name;
2871  result->file_            = file_;
2872  result->containing_type_ = parent;
2873  result->is_placeholder_  = false;
2874  result->is_unqualified_placeholder_ = false;
2875
2876  BUILD_ARRAY(proto, result, field          , BuildField         , result);
2877  BUILD_ARRAY(proto, result, nested_type    , BuildMessage       , result);
2878  BUILD_ARRAY(proto, result, enum_type      , BuildEnum          , result);
2879  BUILD_ARRAY(proto, result, extension_range, BuildExtensionRange, result);
2880  BUILD_ARRAY(proto, result, extension      , BuildExtension     , result);
2881
2882  // Copy options.
2883  if (!proto.has_options()) {
2884    result->options_ = NULL;  // Will set to default_instance later.
2885  } else {
2886    AllocateOptions(proto.options(), result);
2887  }
2888
2889  AddSymbol(result->full_name(), parent, result->name(),
2890            proto, Symbol(result));
2891
2892  // Check that no fields have numbers in extension ranges.
2893  for (int i = 0; i < result->field_count(); i++) {
2894    const FieldDescriptor* field = result->field(i);
2895    for (int j = 0; j < result->extension_range_count(); j++) {
2896      const Descriptor::ExtensionRange* range = result->extension_range(j);
2897      if (range->start <= field->number() && field->number() < range->end) {
2898        AddError(field->full_name(), proto.extension_range(j),
2899                 DescriptorPool::ErrorCollector::NUMBER,
2900                 strings::Substitute(
2901                   "Extension range $0 to $1 includes field \"$2\" ($3).",
2902                   range->start, range->end - 1,
2903                   field->name(), field->number()));
2904      }
2905    }
2906  }
2907
2908  // Check that extension ranges don't overlap.
2909  for (int i = 0; i < result->extension_range_count(); i++) {
2910    const Descriptor::ExtensionRange* range1 = result->extension_range(i);
2911    for (int j = i + 1; j < result->extension_range_count(); j++) {
2912      const Descriptor::ExtensionRange* range2 = result->extension_range(j);
2913      if (range1->end > range2->start && range2->end > range1->start) {
2914        AddError(result->full_name(), proto.extension_range(j),
2915                 DescriptorPool::ErrorCollector::NUMBER,
2916                 strings::Substitute("Extension range $0 to $1 overlaps with "
2917                                     "already-defined range $2 to $3.",
2918                                     range2->start, range2->end - 1,
2919                                     range1->start, range1->end - 1));
2920      }
2921    }
2922  }
2923}
2924
2925void DescriptorBuilder::BuildFieldOrExtension(const FieldDescriptorProto& proto,
2926                                              const Descriptor* parent,
2927                                              FieldDescriptor* result,
2928                                              bool is_extension) {
2929  const string& scope = (parent == NULL) ?
2930    file_->package() : parent->full_name();
2931  string* full_name = tables_->AllocateString(scope);
2932  if (!full_name->empty()) full_name->append(1, '.');
2933  full_name->append(proto.name());
2934
2935  ValidateSymbolName(proto.name(), *full_name, proto);
2936
2937  result->name_         = tables_->AllocateString(proto.name());
2938  result->full_name_    = full_name;
2939  result->file_         = file_;
2940  result->number_       = proto.number();
2941  result->is_extension_ = is_extension;
2942
2943  // If .proto files follow the style guide then the name should already be
2944  // lower-cased.  If that's the case we can just reuse the string we already
2945  // allocated rather than allocate a new one.
2946  string lowercase_name(proto.name());
2947  LowerString(&lowercase_name);
2948  if (lowercase_name == proto.name()) {
2949    result->lowercase_name_ = result->name_;
2950  } else {
2951    result->lowercase_name_ = tables_->AllocateString(lowercase_name);
2952  }
2953
2954  // Don't bother with the above optimization for camel-case names since
2955  // .proto files that follow the guide shouldn't be using names in this
2956  // format, so the optimization wouldn't help much.
2957  result->camelcase_name_ = tables_->AllocateString(ToCamelCase(proto.name()));
2958
2959  // Some compilers do not allow static_cast directly between two enum types,
2960  // so we must cast to int first.
2961  result->type_  = static_cast<FieldDescriptor::Type>(
2962                     implicit_cast<int>(proto.type()));
2963  result->label_ = static_cast<FieldDescriptor::Label>(
2964                     implicit_cast<int>(proto.label()));
2965
2966  // Some of these may be filled in when cross-linking.
2967  result->containing_type_ = NULL;
2968  result->extension_scope_ = NULL;
2969  result->experimental_map_key_ = NULL;
2970  result->message_type_ = NULL;
2971  result->enum_type_ = NULL;
2972
2973  result->has_default_value_ = proto.has_default_value();
2974  if (proto.has_default_value() && result->is_repeated()) {
2975    AddError(result->full_name(), proto,
2976             DescriptorPool::ErrorCollector::DEFAULT_VALUE,
2977             "Repeated fields can't have default values.");
2978  }
2979
2980  if (proto.has_type()) {
2981    if (proto.has_default_value()) {
2982      char* end_pos = NULL;
2983      switch (result->cpp_type()) {
2984        case FieldDescriptor::CPPTYPE_INT32:
2985          result->default_value_int32_ =
2986            strtol(proto.default_value().c_str(), &end_pos, 0);
2987          break;
2988        case FieldDescriptor::CPPTYPE_INT64:
2989          result->default_value_int64_ =
2990            strto64(proto.default_value().c_str(), &end_pos, 0);
2991          break;
2992        case FieldDescriptor::CPPTYPE_UINT32:
2993          result->default_value_uint32_ =
2994            strtoul(proto.default_value().c_str(), &end_pos, 0);
2995          break;
2996        case FieldDescriptor::CPPTYPE_UINT64:
2997          result->default_value_uint64_ =
2998            strtou64(proto.default_value().c_str(), &end_pos, 0);
2999          break;
3000        case FieldDescriptor::CPPTYPE_FLOAT:
3001          result->default_value_float_ =
3002            NoLocaleStrtod(proto.default_value().c_str(), &end_pos);
3003          break;
3004        case FieldDescriptor::CPPTYPE_DOUBLE:
3005          result->default_value_double_ =
3006            NoLocaleStrtod(proto.default_value().c_str(), &end_pos);
3007          break;
3008        case FieldDescriptor::CPPTYPE_BOOL:
3009          if (proto.default_value() == "true") {
3010            result->default_value_bool_ = true;
3011          } else if (proto.default_value() == "false") {
3012            result->default_value_bool_ = false;
3013          } else {
3014            AddError(result->full_name(), proto,
3015                     DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3016                     "Boolean default must be true or false.");
3017          }
3018          break;
3019        case FieldDescriptor::CPPTYPE_ENUM:
3020          // This will be filled in when cross-linking.
3021          result->default_value_enum_ = NULL;
3022          break;
3023        case FieldDescriptor::CPPTYPE_STRING:
3024          if (result->type() == FieldDescriptor::TYPE_BYTES) {
3025            result->default_value_string_ = tables_->AllocateString(
3026              UnescapeCEscapeString(proto.default_value()));
3027          } else {
3028            result->default_value_string_ =
3029              tables_->AllocateString(proto.default_value());
3030          }
3031          break;
3032        case FieldDescriptor::CPPTYPE_MESSAGE:
3033          AddError(result->full_name(), proto,
3034                   DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3035                   "Messages can't have default values.");
3036          result->has_default_value_ = false;
3037          break;
3038      }
3039
3040      if (end_pos != NULL) {
3041        // end_pos is only set non-NULL by the parsers for numeric types, above.
3042        // This checks that the default was non-empty and had no extra junk
3043        // after the end of the number.
3044        if (proto.default_value().empty() || *end_pos != '\0') {
3045          AddError(result->full_name(), proto,
3046                   DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3047                   "Couldn't parse default value.");
3048        }
3049      }
3050    } else {
3051      // No explicit default value
3052      switch (result->cpp_type()) {
3053        case FieldDescriptor::CPPTYPE_INT32:
3054          result->default_value_int32_ = 0;
3055          break;
3056        case FieldDescriptor::CPPTYPE_INT64:
3057          result->default_value_int64_ = 0;
3058          break;
3059        case FieldDescriptor::CPPTYPE_UINT32:
3060          result->default_value_uint32_ = 0;
3061          break;
3062        case FieldDescriptor::CPPTYPE_UINT64:
3063          result->default_value_uint64_ = 0;
3064          break;
3065        case FieldDescriptor::CPPTYPE_FLOAT:
3066          result->default_value_float_ = 0.0f;
3067          break;
3068        case FieldDescriptor::CPPTYPE_DOUBLE:
3069          result->default_value_double_ = 0.0;
3070          break;
3071        case FieldDescriptor::CPPTYPE_BOOL:
3072          result->default_value_bool_ = false;
3073          break;
3074        case FieldDescriptor::CPPTYPE_ENUM:
3075          // This will be filled in when cross-linking.
3076          result->default_value_enum_ = NULL;
3077          break;
3078        case FieldDescriptor::CPPTYPE_STRING:
3079          result->default_value_string_ = &kEmptyString;
3080          break;
3081        case FieldDescriptor::CPPTYPE_MESSAGE:
3082          break;
3083      }
3084    }
3085  }
3086
3087  if (result->number() <= 0) {
3088    AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
3089             "Field numbers must be positive integers.");
3090  } else if (result->number() > FieldDescriptor::kMaxNumber) {
3091    AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
3092             strings::Substitute("Field numbers cannot be greater than $0.",
3093                                 FieldDescriptor::kMaxNumber));
3094  } else if (result->number() >= FieldDescriptor::kFirstReservedNumber &&
3095             result->number() <= FieldDescriptor::kLastReservedNumber) {
3096    AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
3097             strings::Substitute(
3098               "Field numbers $0 through $1 are reserved for the protocol "
3099               "buffer library implementation.",
3100               FieldDescriptor::kFirstReservedNumber,
3101               FieldDescriptor::kLastReservedNumber));
3102  }
3103
3104  if (is_extension) {
3105    if (!proto.has_extendee()) {
3106      AddError(result->full_name(), proto,
3107               DescriptorPool::ErrorCollector::EXTENDEE,
3108               "FieldDescriptorProto.extendee not set for extension field.");
3109    }
3110
3111    result->extension_scope_ = parent;
3112  } else {
3113    if (proto.has_extendee()) {
3114      AddError(result->full_name(), proto,
3115               DescriptorPool::ErrorCollector::EXTENDEE,
3116               "FieldDescriptorProto.extendee set for non-extension field.");
3117    }
3118
3119    result->containing_type_ = parent;
3120  }
3121
3122  // Copy options.
3123  if (!proto.has_options()) {
3124    result->options_ = NULL;  // Will set to default_instance later.
3125  } else {
3126    AllocateOptions(proto.options(), result);
3127  }
3128
3129  AddSymbol(result->full_name(), parent, result->name(),
3130            proto, Symbol(result));
3131}
3132
3133void DescriptorBuilder::BuildExtensionRange(
3134    const DescriptorProto::ExtensionRange& proto,
3135    const Descriptor* parent,
3136    Descriptor::ExtensionRange* result) {
3137  result->start = proto.start();
3138  result->end = proto.end();
3139  if (result->start <= 0) {
3140    AddError(parent->full_name(), proto,
3141             DescriptorPool::ErrorCollector::NUMBER,
3142             "Extension numbers must be positive integers.");
3143  }
3144
3145  if (result->end > FieldDescriptor::kMaxNumber + 1) {
3146    AddError(parent->full_name(), proto,
3147             DescriptorPool::ErrorCollector::NUMBER,
3148             strings::Substitute("Extension numbers cannot be greater than $0.",
3149                                 FieldDescriptor::kMaxNumber));
3150  }
3151
3152  if (result->start >= result->end) {
3153    AddError(parent->full_name(), proto,
3154             DescriptorPool::ErrorCollector::NUMBER,
3155             "Extension range end number must be greater than start number.");
3156  }
3157}
3158
3159void DescriptorBuilder::BuildEnum(const EnumDescriptorProto& proto,
3160                                  const Descriptor* parent,
3161                                  EnumDescriptor* result) {
3162  const string& scope = (parent == NULL) ?
3163    file_->package() : parent->full_name();
3164  string* full_name = tables_->AllocateString(scope);
3165  if (!full_name->empty()) full_name->append(1, '.');
3166  full_name->append(proto.name());
3167
3168  ValidateSymbolName(proto.name(), *full_name, proto);
3169
3170  result->name_            = tables_->AllocateString(proto.name());
3171  result->full_name_       = full_name;
3172  result->file_            = file_;
3173  result->containing_type_ = parent;
3174  result->is_placeholder_  = false;
3175  result->is_unqualified_placeholder_ = false;
3176
3177  if (proto.value_size() == 0) {
3178    // We cannot allow enums with no values because this would mean there
3179    // would be no valid default value for fields of this type.
3180    AddError(result->full_name(), proto,
3181             DescriptorPool::ErrorCollector::NAME,
3182             "Enums must contain at least one value.");
3183  }
3184
3185  BUILD_ARRAY(proto, result, value, BuildEnumValue, result);
3186
3187  // Copy options.
3188  if (!proto.has_options()) {
3189    result->options_ = NULL;  // Will set to default_instance later.
3190  } else {
3191    AllocateOptions(proto.options(), result);
3192  }
3193
3194  AddSymbol(result->full_name(), parent, result->name(),
3195            proto, Symbol(result));
3196}
3197
3198void DescriptorBuilder::BuildEnumValue(const EnumValueDescriptorProto& proto,
3199                                       const EnumDescriptor* parent,
3200                                       EnumValueDescriptor* result) {
3201  result->name_   = tables_->AllocateString(proto.name());
3202  result->number_ = proto.number();
3203  result->type_   = parent;
3204
3205  // Note:  full_name for enum values is a sibling to the parent's name, not a
3206  //   child of it.
3207  string* full_name = tables_->AllocateString(*parent->full_name_);
3208  full_name->resize(full_name->size() - parent->name_->size());
3209  full_name->append(*result->name_);
3210  result->full_name_ = full_name;
3211
3212  ValidateSymbolName(proto.name(), *full_name, proto);
3213
3214  // Copy options.
3215  if (!proto.has_options()) {
3216    result->options_ = NULL;  // Will set to default_instance later.
3217  } else {
3218    AllocateOptions(proto.options(), result);
3219  }
3220
3221  // Again, enum values are weird because we makes them appear as siblings
3222  // of the enum type instead of children of it.  So, we use
3223  // parent->containing_type() as the value's parent.
3224  bool added_to_outer_scope =
3225    AddSymbol(result->full_name(), parent->containing_type(), result->name(),
3226              proto, Symbol(result));
3227
3228  // However, we also want to be able to search for values within a single
3229  // enum type, so we add it as a child of the enum type itself, too.
3230  // Note:  This could fail, but if it does, the error has already been
3231  //   reported by the above AddSymbol() call, so we ignore the return code.
3232  bool added_to_inner_scope =
3233    file_tables_->AddAliasUnderParent(parent, result->name(), Symbol(result));
3234
3235  if (added_to_inner_scope && !added_to_outer_scope) {
3236    // This value did not conflict with any values defined in the same enum,
3237    // but it did conflict with some other symbol defined in the enum type's
3238    // scope.  Let's print an additional error to explain this.
3239    string outer_scope;
3240    if (parent->containing_type() == NULL) {
3241      outer_scope = file_->package();
3242    } else {
3243      outer_scope = parent->containing_type()->full_name();
3244    }
3245
3246    if (outer_scope.empty()) {
3247      outer_scope = "the global scope";
3248    } else {
3249      outer_scope = "\"" + outer_scope + "\"";
3250    }
3251
3252    AddError(result->full_name(), proto,
3253             DescriptorPool::ErrorCollector::NAME,
3254             "Note that enum values use C++ scoping rules, meaning that "
3255             "enum values are siblings of their type, not children of it.  "
3256             "Therefore, \"" + result->name() + "\" must be unique within "
3257             + outer_scope + ", not just within \"" + parent->name() + "\".");
3258  }
3259
3260  // An enum is allowed to define two numbers that refer to the same value.
3261  // FindValueByNumber() should return the first such value, so we simply
3262  // ignore AddEnumValueByNumber()'s return code.
3263  file_tables_->AddEnumValueByNumber(result);
3264}
3265
3266void DescriptorBuilder::BuildService(const ServiceDescriptorProto& proto,
3267                                     const void* dummy,
3268                                     ServiceDescriptor* result) {
3269  string* full_name = tables_->AllocateString(file_->package());
3270  if (!full_name->empty()) full_name->append(1, '.');
3271  full_name->append(proto.name());
3272
3273  ValidateSymbolName(proto.name(), *full_name, proto);
3274
3275  result->name_      = tables_->AllocateString(proto.name());
3276  result->full_name_ = full_name;
3277  result->file_      = file_;
3278
3279  BUILD_ARRAY(proto, result, method, BuildMethod, result);
3280
3281  // Copy options.
3282  if (!proto.has_options()) {
3283    result->options_ = NULL;  // Will set to default_instance later.
3284  } else {
3285    AllocateOptions(proto.options(), result);
3286  }
3287
3288  AddSymbol(result->full_name(), NULL, result->name(),
3289            proto, Symbol(result));
3290}
3291
3292void DescriptorBuilder::BuildMethod(const MethodDescriptorProto& proto,
3293                                    const ServiceDescriptor* parent,
3294                                    MethodDescriptor* result) {
3295  result->name_    = tables_->AllocateString(proto.name());
3296  result->service_ = parent;
3297
3298  string* full_name = tables_->AllocateString(parent->full_name());
3299  full_name->append(1, '.');
3300  full_name->append(*result->name_);
3301  result->full_name_ = full_name;
3302
3303  ValidateSymbolName(proto.name(), *full_name, proto);
3304
3305  // These will be filled in when cross-linking.
3306  result->input_type_ = NULL;
3307  result->output_type_ = NULL;
3308
3309  // Copy options.
3310  if (!proto.has_options()) {
3311    result->options_ = NULL;  // Will set to default_instance later.
3312  } else {
3313    AllocateOptions(proto.options(), result);
3314  }
3315
3316  AddSymbol(result->full_name(), parent, result->name(),
3317            proto, Symbol(result));
3318}
3319
3320#undef BUILD_ARRAY
3321
3322// -------------------------------------------------------------------
3323
3324void DescriptorBuilder::CrossLinkFile(
3325    FileDescriptor* file, const FileDescriptorProto& proto) {
3326  if (file->options_ == NULL) {
3327    file->options_ = &FileOptions::default_instance();
3328  }
3329
3330  for (int i = 0; i < file->message_type_count(); i++) {
3331    CrossLinkMessage(&file->message_types_[i], proto.message_type(i));
3332  }
3333
3334  for (int i = 0; i < file->extension_count(); i++) {
3335    CrossLinkField(&file->extensions_[i], proto.extension(i));
3336  }
3337
3338  for (int i = 0; i < file->enum_type_count(); i++) {
3339    CrossLinkEnum(&file->enum_types_[i], proto.enum_type(i));
3340  }
3341
3342  for (int i = 0; i < file->service_count(); i++) {
3343    CrossLinkService(&file->services_[i], proto.service(i));
3344  }
3345}
3346
3347void DescriptorBuilder::CrossLinkMessage(
3348    Descriptor* message, const DescriptorProto& proto) {
3349  if (message->options_ == NULL) {
3350    message->options_ = &MessageOptions::default_instance();
3351  }
3352
3353  for (int i = 0; i < message->nested_type_count(); i++) {
3354    CrossLinkMessage(&message->nested_types_[i], proto.nested_type(i));
3355  }
3356
3357  for (int i = 0; i < message->enum_type_count(); i++) {
3358    CrossLinkEnum(&message->enum_types_[i], proto.enum_type(i));
3359  }
3360
3361  for (int i = 0; i < message->field_count(); i++) {
3362    CrossLinkField(&message->fields_[i], proto.field(i));
3363  }
3364
3365  for (int i = 0; i < message->extension_count(); i++) {
3366    CrossLinkField(&message->extensions_[i], proto.extension(i));
3367  }
3368}
3369
3370void DescriptorBuilder::CrossLinkField(
3371    FieldDescriptor* field, const FieldDescriptorProto& proto) {
3372  if (field->options_ == NULL) {
3373    field->options_ = &FieldOptions::default_instance();
3374  }
3375
3376  if (proto.has_extendee()) {
3377    Symbol extendee = LookupSymbol(proto.extendee(), field->full_name(),
3378                                   PLACEHOLDER_EXTENDABLE_MESSAGE);
3379    if (extendee.IsNull()) {
3380      AddNotDefinedError(field->full_name(), proto,
3381                         DescriptorPool::ErrorCollector::EXTENDEE,
3382                         proto.extendee());
3383      return;
3384    } else if (extendee.type != Symbol::MESSAGE) {
3385      AddError(field->full_name(), proto,
3386               DescriptorPool::ErrorCollector::EXTENDEE,
3387               "\"" + proto.extendee() + "\" is not a message type.");
3388      return;
3389    }
3390    field->containing_type_ = extendee.descriptor;
3391
3392    if (!field->containing_type()->IsExtensionNumber(field->number())) {
3393      AddError(field->full_name(), proto,
3394               DescriptorPool::ErrorCollector::NUMBER,
3395               strings::Substitute("\"$0\" does not declare $1 as an "
3396                                   "extension number.",
3397                                   field->containing_type()->full_name(),
3398                                   field->number()));
3399    }
3400  }
3401
3402  if (proto.has_type_name()) {
3403    // Assume we are expecting a message type unless the proto contains some
3404    // evidence that it expects an enum type.  This only makes a difference if
3405    // we end up creating a placeholder.
3406    bool expecting_enum = (proto.type() == FieldDescriptorProto::TYPE_ENUM) ||
3407                          proto.has_default_value();
3408
3409    Symbol type =
3410      LookupSymbol(proto.type_name(), field->full_name(),
3411                   expecting_enum ? PLACEHOLDER_ENUM : PLACEHOLDER_MESSAGE,
3412                   LOOKUP_TYPES);
3413
3414    if (type.IsNull()) {
3415      AddNotDefinedError(field->full_name(), proto,
3416                         DescriptorPool::ErrorCollector::TYPE,
3417                         proto.type_name());
3418      return;
3419    }
3420
3421    if (!proto.has_type()) {
3422      // Choose field type based on symbol.
3423      if (type.type == Symbol::MESSAGE) {
3424        field->type_ = FieldDescriptor::TYPE_MESSAGE;
3425      } else if (type.type == Symbol::ENUM) {
3426        field->type_ = FieldDescriptor::TYPE_ENUM;
3427      } else {
3428        AddError(field->full_name(), proto,
3429                 DescriptorPool::ErrorCollector::TYPE,
3430                 "\"" + proto.type_name() + "\" is not a type.");
3431        return;
3432      }
3433    }
3434
3435    if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
3436      if (type.type != Symbol::MESSAGE) {
3437        AddError(field->full_name(), proto,
3438                 DescriptorPool::ErrorCollector::TYPE,
3439                 "\"" + proto.type_name() + "\" is not a message type.");
3440        return;
3441      }
3442      field->message_type_ = type.descriptor;
3443
3444      if (field->has_default_value()) {
3445        AddError(field->full_name(), proto,
3446                 DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3447                 "Messages can't have default values.");
3448      }
3449    } else if (field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM) {
3450      if (type.type != Symbol::ENUM) {
3451        AddError(field->full_name(), proto,
3452                 DescriptorPool::ErrorCollector::TYPE,
3453                 "\"" + proto.type_name() + "\" is not an enum type.");
3454        return;
3455      }
3456      field->enum_type_ = type.enum_descriptor;
3457
3458      if (field->enum_type()->is_placeholder_) {
3459        // We can't look up default values for placeholder types.  We'll have
3460        // to just drop them.
3461        field->has_default_value_ = false;
3462      }
3463
3464      if (field->has_default_value()) {
3465        // We can't just use field->enum_type()->FindValueByName() here
3466        // because that locks the pool's mutex, which we have already locked
3467        // at this point.
3468        Symbol default_value =
3469          LookupSymbolNoPlaceholder(proto.default_value(),
3470                                    field->enum_type()->full_name());
3471
3472        if (default_value.type == Symbol::ENUM_VALUE &&
3473            default_value.enum_value_descriptor->type() == field->enum_type()) {
3474          field->default_value_enum_ = default_value.enum_value_descriptor;
3475        } else {
3476          AddError(field->full_name(), proto,
3477                   DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3478                   "Enum type \"" + field->enum_type()->full_name() +
3479                   "\" has no value named \"" + proto.default_value() + "\".");
3480        }
3481      } else if (field->enum_type()->value_count() > 0) {
3482        // All enums must have at least one value, or we would have reported
3483        // an error elsewhere.  We use the first defined value as the default
3484        // if a default is not explicitly defined.
3485        field->default_value_enum_ = field->enum_type()->value(0);
3486      }
3487    } else {
3488      AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3489               "Field with primitive type has type_name.");
3490    }
3491  } else {
3492    if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE ||
3493        field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM) {
3494      AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3495               "Field with message or enum type missing type_name.");
3496    }
3497  }
3498
3499  // Add the field to the fields-by-number table.
3500  // Note:  We have to do this *after* cross-linking because extensions do not
3501  //   know their containing type until now.
3502  if (!file_tables_->AddFieldByNumber(field)) {
3503    const FieldDescriptor* conflicting_field =
3504      file_tables_->FindFieldByNumber(field->containing_type(),
3505                                      field->number());
3506    if (field->is_extension()) {
3507      AddError(field->full_name(), proto,
3508               DescriptorPool::ErrorCollector::NUMBER,
3509               strings::Substitute("Extension number $0 has already been used "
3510                                   "in \"$1\" by extension \"$2\".",
3511                                   field->number(),
3512                                   field->containing_type()->full_name(),
3513                                   conflicting_field->full_name()));
3514    } else {
3515      AddError(field->full_name(), proto,
3516               DescriptorPool::ErrorCollector::NUMBER,
3517               strings::Substitute("Field number $0 has already been used in "
3518                                   "\"$1\" by field \"$2\".",
3519                                   field->number(),
3520                                   field->containing_type()->full_name(),
3521                                   conflicting_field->name()));
3522    }
3523  }
3524
3525  if (field->is_extension()) {
3526    // No need for error checking: if the extension number collided,
3527    // we've already been informed of it by the if() above.
3528    tables_->AddExtension(field);
3529  }
3530
3531  // Add the field to the lowercase-name and camelcase-name tables.
3532  file_tables_->AddFieldByStylizedNames(field);
3533}
3534
3535void DescriptorBuilder::CrossLinkEnum(
3536    EnumDescriptor* enum_type, const EnumDescriptorProto& proto) {
3537  if (enum_type->options_ == NULL) {
3538    enum_type->options_ = &EnumOptions::default_instance();
3539  }
3540
3541  for (int i = 0; i < enum_type->value_count(); i++) {
3542    CrossLinkEnumValue(&enum_type->values_[i], proto.value(i));
3543  }
3544}
3545
3546void DescriptorBuilder::CrossLinkEnumValue(
3547    EnumValueDescriptor* enum_value, const EnumValueDescriptorProto& proto) {
3548  if (enum_value->options_ == NULL) {
3549    enum_value->options_ = &EnumValueOptions::default_instance();
3550  }
3551}
3552
3553void DescriptorBuilder::CrossLinkService(
3554    ServiceDescriptor* service, const ServiceDescriptorProto& proto) {
3555  if (service->options_ == NULL) {
3556    service->options_ = &ServiceOptions::default_instance();
3557  }
3558
3559  for (int i = 0; i < service->method_count(); i++) {
3560    CrossLinkMethod(&service->methods_[i], proto.method(i));
3561  }
3562}
3563
3564void DescriptorBuilder::CrossLinkMethod(
3565    MethodDescriptor* method, const MethodDescriptorProto& proto) {
3566  if (method->options_ == NULL) {
3567    method->options_ = &MethodOptions::default_instance();
3568  }
3569
3570  Symbol input_type = LookupSymbol(proto.input_type(), method->full_name());
3571  if (input_type.IsNull()) {
3572    AddNotDefinedError(method->full_name(), proto,
3573                       DescriptorPool::ErrorCollector::INPUT_TYPE,
3574                       proto.input_type());
3575  } else if (input_type.type != Symbol::MESSAGE) {
3576    AddError(method->full_name(), proto,
3577             DescriptorPool::ErrorCollector::INPUT_TYPE,
3578             "\"" + proto.input_type() + "\" is not a message type.");
3579  } else {
3580    method->input_type_ = input_type.descriptor;
3581  }
3582
3583  Symbol output_type = LookupSymbol(proto.output_type(), method->full_name());
3584  if (output_type.IsNull()) {
3585    AddNotDefinedError(method->full_name(), proto,
3586                       DescriptorPool::ErrorCollector::OUTPUT_TYPE,
3587                       proto.output_type());
3588  } else if (output_type.type != Symbol::MESSAGE) {
3589    AddError(method->full_name(), proto,
3590             DescriptorPool::ErrorCollector::OUTPUT_TYPE,
3591             "\"" + proto.output_type() + "\" is not a message type.");
3592  } else {
3593    method->output_type_ = output_type.descriptor;
3594  }
3595}
3596
3597// -------------------------------------------------------------------
3598
3599#define VALIDATE_OPTIONS_FROM_ARRAY(descriptor, array_name, type)  \
3600  for (int i = 0; i < descriptor->array_name##_count(); ++i) {     \
3601    Validate##type##Options(descriptor->array_name##s_ + i,        \
3602                            proto.array_name(i));                  \
3603  }
3604
3605// Determine if the file uses optimize_for = LITE_RUNTIME, being careful to
3606// avoid problems that exist at init time.
3607static bool IsLite(const FileDescriptor* file) {
3608  // TODO(kenton):  I don't even remember how many of these conditions are
3609  //   actually possible.  I'm just being super-safe.
3610  return file != NULL &&
3611         &file->options() != NULL &&
3612         &file->options() != &FileOptions::default_instance() &&
3613         file->options().optimize_for() == FileOptions::LITE_RUNTIME;
3614}
3615
3616void DescriptorBuilder::ValidateFileOptions(FileDescriptor* file,
3617                                            const FileDescriptorProto& proto) {
3618  VALIDATE_OPTIONS_FROM_ARRAY(file, message_type, Message);
3619  VALIDATE_OPTIONS_FROM_ARRAY(file, enum_type, Enum);
3620  VALIDATE_OPTIONS_FROM_ARRAY(file, service, Service);
3621  VALIDATE_OPTIONS_FROM_ARRAY(file, extension, Field);
3622
3623  // Lite files can only be imported by other Lite files.
3624  if (!IsLite(file)) {
3625    for (int i = 0; i < file->dependency_count(); i++) {
3626      if (IsLite(file->dependency(i))) {
3627        AddError(
3628          file->name(), proto,
3629          DescriptorPool::ErrorCollector::OTHER,
3630          "Files that do not use optimize_for = LITE_RUNTIME cannot import "
3631          "files which do use this option.  This file is not lite, but it "
3632          "imports \"" + file->dependency(i)->name() + "\" which is.");
3633        break;
3634      }
3635    }
3636  }
3637}
3638
3639void DescriptorBuilder::ValidateMessageOptions(Descriptor* message,
3640                                               const DescriptorProto& proto) {
3641  VALIDATE_OPTIONS_FROM_ARRAY(message, field, Field);
3642  VALIDATE_OPTIONS_FROM_ARRAY(message, nested_type, Message);
3643  VALIDATE_OPTIONS_FROM_ARRAY(message, enum_type, Enum);
3644  VALIDATE_OPTIONS_FROM_ARRAY(message, extension, Field);
3645}
3646
3647void DescriptorBuilder::ValidateFieldOptions(FieldDescriptor* field,
3648    const FieldDescriptorProto& proto) {
3649  if (field->options().has_experimental_map_key()) {
3650    ValidateMapKey(field, proto);
3651  }
3652
3653  // Only repeated primitive fields may be packed.
3654  if (field->options().packed()) {
3655    if (!field->is_repeated() ||
3656        field->type() == FieldDescriptor::TYPE_STRING ||
3657        field->type() == FieldDescriptor::TYPE_GROUP ||
3658        field->type() == FieldDescriptor::TYPE_MESSAGE ||
3659        field->type() == FieldDescriptor::TYPE_BYTES) {
3660      AddError(
3661        field->full_name(), proto,
3662        DescriptorPool::ErrorCollector::TYPE,
3663        "[packed = true] can only be specified for repeated primitive fields.");
3664    }
3665  }
3666
3667  // Note:  Default instance may not yet be initialized here, so we have to
3668  //   avoid reading from it.
3669  if (field->containing_type_ != NULL &&
3670      &field->containing_type()->options() !=
3671      &MessageOptions::default_instance() &&
3672      field->containing_type()->options().message_set_wire_format()) {
3673    if (field->is_extension()) {
3674      if (!field->is_optional() ||
3675          field->type() != FieldDescriptor::TYPE_MESSAGE) {
3676        AddError(field->full_name(), proto,
3677                 DescriptorPool::ErrorCollector::TYPE,
3678                 "Extensions of MessageSets must be optional messages.");
3679      }
3680    } else {
3681      AddError(field->full_name(), proto,
3682               DescriptorPool::ErrorCollector::NAME,
3683               "MessageSets cannot have fields, only extensions.");
3684    }
3685  }
3686
3687  // Lite extensions can only be of Lite types.
3688  if (IsLite(field->file()) &&
3689      field->containing_type_ != NULL &&
3690      !IsLite(field->containing_type()->file())) {
3691    AddError(field->full_name(), proto,
3692             DescriptorPool::ErrorCollector::EXTENDEE,
3693             "Extensions to non-lite types can only be declared in non-lite "
3694             "files.  Note that you cannot extend a non-lite type to contain "
3695             "a lite type, but the reverse is allowed.");
3696  }
3697}
3698
3699void DescriptorBuilder::ValidateEnumOptions(EnumDescriptor* enm,
3700                                            const EnumDescriptorProto& proto) {
3701  VALIDATE_OPTIONS_FROM_ARRAY(enm, value, EnumValue);
3702}
3703
3704void DescriptorBuilder::ValidateEnumValueOptions(
3705    EnumValueDescriptor* enum_value, const EnumValueDescriptorProto& proto) {
3706  // Nothing to do so far.
3707}
3708void DescriptorBuilder::ValidateServiceOptions(ServiceDescriptor* service,
3709    const ServiceDescriptorProto& proto) {
3710  if (IsLite(service->file())) {
3711    AddError(service->full_name(), proto,
3712             DescriptorPool::ErrorCollector::NAME,
3713             "Files with optimize_for = LITE_RUNTIME cannot define services.");
3714  }
3715
3716  VALIDATE_OPTIONS_FROM_ARRAY(service, method, Method);
3717}
3718
3719void DescriptorBuilder::ValidateMethodOptions(MethodDescriptor* method,
3720    const MethodDescriptorProto& proto) {
3721  // Nothing to do so far.
3722}
3723
3724void DescriptorBuilder::ValidateMapKey(FieldDescriptor* field,
3725                                       const FieldDescriptorProto& proto) {
3726  if (!field->is_repeated()) {
3727    AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3728             "map type is only allowed for repeated fields.");
3729    return;
3730  }
3731
3732  if (field->cpp_type() != FieldDescriptor::CPPTYPE_MESSAGE) {
3733    AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3734             "map type is only allowed for fields with a message type.");
3735    return;
3736  }
3737
3738  const Descriptor* item_type = field->message_type();
3739  if (item_type == NULL) {
3740    AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3741             "Could not find field type.");
3742    return;
3743  }
3744
3745  // Find the field in item_type named by "experimental_map_key"
3746  const string& key_name = field->options().experimental_map_key();
3747  const Symbol key_symbol = LookupSymbol(
3748      key_name,
3749      // We append ".key_name" to the containing type's name since
3750      // LookupSymbol() searches for peers of the supplied name, not
3751      // children of the supplied name.
3752      item_type->full_name() + "." + key_name);
3753
3754  if (key_symbol.IsNull() || key_symbol.field_descriptor->is_extension()) {
3755    AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3756             "Could not find field named \"" + key_name + "\" in type \"" +
3757             item_type->full_name() + "\".");
3758    return;
3759  }
3760  const FieldDescriptor* key_field = key_symbol.field_descriptor;
3761
3762  if (key_field->is_repeated()) {
3763    AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3764             "map_key must not name a repeated field.");
3765    return;
3766  }
3767
3768  if (key_field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
3769    AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3770             "map key must name a scalar or string field.");
3771    return;
3772  }
3773
3774  field->experimental_map_key_ = key_field;
3775}
3776
3777#undef VALIDATE_OPTIONS_FROM_ARRAY
3778
3779// -------------------------------------------------------------------
3780
3781DescriptorBuilder::OptionInterpreter::OptionInterpreter(
3782    DescriptorBuilder* builder) : builder_(builder) {
3783  GOOGLE_CHECK(builder_);
3784}
3785
3786DescriptorBuilder::OptionInterpreter::~OptionInterpreter() {
3787}
3788
3789bool DescriptorBuilder::OptionInterpreter::InterpretOptions(
3790    OptionsToInterpret* options_to_interpret) {
3791  // Note that these may be in different pools, so we can't use the same
3792  // descriptor and reflection objects on both.
3793  Message* options = options_to_interpret->options;
3794  const Message* original_options = options_to_interpret->original_options;
3795
3796  bool failed = false;
3797  options_to_interpret_ = options_to_interpret;
3798
3799  // Find the uninterpreted_option field in the mutable copy of the options
3800  // and clear them, since we're about to interpret them.
3801  const FieldDescriptor* uninterpreted_options_field =
3802      options->GetDescriptor()->FindFieldByName("uninterpreted_option");
3803  GOOGLE_CHECK(uninterpreted_options_field != NULL)
3804      << "No field named \"uninterpreted_option\" in the Options proto.";
3805  options->GetReflection()->ClearField(options, uninterpreted_options_field);
3806
3807  // Find the uninterpreted_option field in the original options.
3808  const FieldDescriptor* original_uninterpreted_options_field =
3809      original_options->GetDescriptor()->
3810          FindFieldByName("uninterpreted_option");
3811  GOOGLE_CHECK(original_uninterpreted_options_field != NULL)
3812      << "No field named \"uninterpreted_option\" in the Options proto.";
3813
3814  const int num_uninterpreted_options = original_options->GetReflection()->
3815      FieldSize(*original_options, original_uninterpreted_options_field);
3816  for (int i = 0; i < num_uninterpreted_options; ++i) {
3817    uninterpreted_option_ = down_cast<const UninterpretedOption*>(
3818        &original_options->GetReflection()->GetRepeatedMessage(
3819            *original_options, original_uninterpreted_options_field, i));
3820    if (!InterpretSingleOption(options)) {
3821      // Error already added by InterpretSingleOption().
3822      failed = true;
3823      break;
3824    }
3825  }
3826  // Reset these, so we don't have any dangling pointers.
3827  uninterpreted_option_ = NULL;
3828  options_to_interpret_ = NULL;
3829
3830  if (!failed) {
3831    // InterpretSingleOption() added the interpreted options in the
3832    // UnknownFieldSet, in case the option isn't yet known to us.  Now we
3833    // serialize the options message and deserialize it back.  That way, any
3834    // option fields that we do happen to know about will get moved from the
3835    // UnknownFieldSet into the real fields, and thus be available right away.
3836    // If they are not known, that's OK too. They will get reparsed into the
3837    // UnknownFieldSet and wait there until the message is parsed by something
3838    // that does know about the options.
3839    string buf;
3840    options->AppendToString(&buf);
3841    GOOGLE_CHECK(options->ParseFromString(buf))
3842        << "Protocol message serialized itself in invalid fashion.";
3843  }
3844
3845  return !failed;
3846}
3847
3848bool DescriptorBuilder::OptionInterpreter::InterpretSingleOption(
3849    Message* options) {
3850  // First do some basic validation.
3851  if (uninterpreted_option_->name_size() == 0) {
3852    // This should never happen unless the parser has gone seriously awry or
3853    // someone has manually created the uninterpreted option badly.
3854    return AddNameError("Option must have a name.");
3855  }
3856  if (uninterpreted_option_->name(0).name_part() == "uninterpreted_option") {
3857    return AddNameError("Option must not use reserved name "
3858                        "\"uninterpreted_option\".");
3859  }
3860
3861  const Descriptor* options_descriptor = NULL;
3862  // Get the options message's descriptor from the builder's pool, so that we
3863  // get the version that knows about any extension options declared in the
3864  // file we're currently building. The descriptor should be there as long as
3865  // the file we're building imported "google/protobuf/descriptors.proto".
3866
3867  // Note that we use DescriptorBuilder::FindSymbol(), not
3868  // DescriptorPool::FindMessageTypeByName() because we're already holding the
3869  // pool's mutex, and the latter method locks it again.
3870  Symbol symbol = builder_->FindSymbolNotEnforcingDeps(
3871    options->GetDescriptor()->full_name());
3872  if (!symbol.IsNull() && symbol.type == Symbol::MESSAGE) {
3873    options_descriptor = symbol.descriptor;
3874  } else {
3875    // The options message's descriptor was not in the builder's pool, so use
3876    // the standard version from the generated pool. We're not holding the
3877    // generated pool's mutex, so we can search it the straightforward way.
3878    options_descriptor = options->GetDescriptor();
3879  }
3880  GOOGLE_CHECK(options_descriptor);
3881
3882  // We iterate over the name parts to drill into the submessages until we find
3883  // the leaf field for the option. As we drill down we remember the current
3884  // submessage's descriptor in |descriptor| and the next field in that
3885  // submessage in |field|. We also track the fields we're drilling down
3886  // through in |intermediate_fields|. As we go, we reconstruct the full option
3887  // name in |debug_msg_name|, for use in error messages.
3888  const Descriptor* descriptor = options_descriptor;
3889  const FieldDescriptor* field = NULL;
3890  vector<const FieldDescriptor*> intermediate_fields;
3891  string debug_msg_name = "";
3892
3893  for (int i = 0; i < uninterpreted_option_->name_size(); ++i) {
3894    const string& name_part = uninterpreted_option_->name(i).name_part();
3895    if (debug_msg_name.size() > 0) {
3896      debug_msg_name += ".";
3897    }
3898    if (uninterpreted_option_->name(i).is_extension()) {
3899      debug_msg_name += "(" + name_part + ")";
3900      // Search for the extension's descriptor as an extension in the builder's
3901      // pool. Note that we use DescriptorBuilder::LookupSymbol(), not
3902      // DescriptorPool::FindExtensionByName(), for two reasons: 1) It allows
3903      // relative lookups, and 2) because we're already holding the pool's
3904      // mutex, and the latter method locks it again.
3905      Symbol symbol = builder_->LookupSymbol(name_part,
3906                                             options_to_interpret_->name_scope);
3907      if (!symbol.IsNull() && symbol.type == Symbol::FIELD) {
3908        field = symbol.field_descriptor;
3909      }
3910      // If we don't find the field then the field's descriptor was not in the
3911      // builder's pool, but there's no point in looking in the generated
3912      // pool. We require that you import the file that defines any extensions
3913      // you use, so they must be present in the builder's pool.
3914    } else {
3915      debug_msg_name += name_part;
3916      // Search for the field's descriptor as a regular field.
3917      field = descriptor->FindFieldByName(name_part);
3918    }
3919
3920    if (field == NULL) {
3921      if (get_allow_unknown(builder_->pool_)) {
3922        // We can't find the option, but AllowUnknownDependencies() is enabled,
3923        // so we will just leave it as uninterpreted.
3924        AddWithoutInterpreting(*uninterpreted_option_, options);
3925        return true;
3926      } else {
3927        return AddNameError("Option \"" + debug_msg_name + "\" unknown.");
3928      }
3929    } else if (field->containing_type() != descriptor) {
3930      if (get_is_placeholder(field->containing_type())) {
3931        // The field is an extension of a placeholder type, so we can't
3932        // reliably verify whether it is a valid extension to use here (e.g.
3933        // we don't know if it is an extension of the correct *Options message,
3934        // or if it has a valid field number, etc.).  Just leave it as
3935        // uninterpreted instead.
3936        AddWithoutInterpreting(*uninterpreted_option_, options);
3937        return true;
3938      } else {
3939        // This can only happen if, due to some insane misconfiguration of the
3940        // pools, we find the options message in one pool but the field in
3941        // another. This would probably imply a hefty bug somewhere.
3942        return AddNameError("Option field \"" + debug_msg_name +
3943                            "\" is not a field or extension of message \"" +
3944                            descriptor->name() + "\".");
3945      }
3946    } else if (field->is_repeated()) {
3947      return AddNameError("Option field \"" + debug_msg_name +
3948                          "\" is repeated. Repeated options are not "
3949                          "supported.");
3950    } else if (i < uninterpreted_option_->name_size() - 1) {
3951      if (field->cpp_type() != FieldDescriptor::CPPTYPE_MESSAGE) {
3952        return AddNameError("Option \"" +  debug_msg_name +
3953                            "\" is an atomic type, not a message.");
3954      } else {
3955        // Drill down into the submessage.
3956        intermediate_fields.push_back(field);
3957        descriptor = field->message_type();
3958      }
3959    } else if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
3960      return AddNameError("Option field \"" + debug_msg_name +
3961                          "\" cannot be of message type.");
3962    }
3963  }
3964
3965  // We've found the leaf field. Now we use UnknownFieldSets to set its value
3966  // on the options message. We do so because the message may not yet know
3967  // about its extension fields, so we may not be able to set the fields
3968  // directly. But the UnknownFieldSets will serialize to the same wire-format
3969  // message, so reading that message back in once the extension fields are
3970  // known will populate them correctly.
3971
3972  // First see if the option is already set.
3973  if (!ExamineIfOptionIsSet(
3974          intermediate_fields.begin(),
3975          intermediate_fields.end(),
3976          field, debug_msg_name,
3977          options->GetReflection()->GetUnknownFields(*options))) {
3978    return false;  // ExamineIfOptionIsSet() already added the error.
3979  }
3980
3981
3982  // First set the value on the UnknownFieldSet corresponding to the
3983  // innermost message.
3984  scoped_ptr<UnknownFieldSet> unknown_fields(new UnknownFieldSet());
3985  if (!SetOptionValue(field, unknown_fields.get())) {
3986    return false;  // SetOptionValue() already added the error.
3987  }
3988
3989  // Now wrap the UnknownFieldSet with UnknownFieldSets corresponding to all
3990  // the intermediate messages.
3991  for (vector<const FieldDescriptor*>::reverse_iterator iter =
3992           intermediate_fields.rbegin();
3993       iter != intermediate_fields.rend(); ++iter) {
3994    scoped_ptr<UnknownFieldSet> parent_unknown_fields(new UnknownFieldSet());
3995    switch ((*iter)->type()) {
3996      case FieldDescriptor::TYPE_MESSAGE: {
3997        io::StringOutputStream outstr(
3998            parent_unknown_fields->AddLengthDelimited((*iter)->number()));
3999        io::CodedOutputStream out(&outstr);
4000        internal::WireFormat::SerializeUnknownFields(*unknown_fields, &out);
4001        GOOGLE_CHECK(!out.HadError())
4002            << "Unexpected failure while serializing option submessage "
4003            << debug_msg_name << "\".";
4004        break;
4005      }
4006
4007      case FieldDescriptor::TYPE_GROUP: {
4008         parent_unknown_fields->AddGroup((*iter)->number())
4009                              ->MergeFrom(*unknown_fields);
4010        break;
4011      }
4012
4013      default:
4014        GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_MESSAGE: "
4015                   << (*iter)->type();
4016        return false;
4017    }
4018    unknown_fields.reset(parent_unknown_fields.release());
4019  }
4020
4021  // Now merge the UnknownFieldSet corresponding to the top-level message into
4022  // the options message.
4023  options->GetReflection()->MutableUnknownFields(options)->MergeFrom(
4024      *unknown_fields);
4025
4026  return true;
4027}
4028
4029void DescriptorBuilder::OptionInterpreter::AddWithoutInterpreting(
4030    const UninterpretedOption& uninterpreted_option, Message* options) {
4031  const FieldDescriptor* field =
4032    options->GetDescriptor()->FindFieldByName("uninterpreted_option");
4033  GOOGLE_CHECK(field != NULL);
4034
4035  options->GetReflection()->AddMessage(options, field)
4036    ->CopyFrom(uninterpreted_option);
4037}
4038
4039bool DescriptorBuilder::OptionInterpreter::ExamineIfOptionIsSet(
4040    vector<const FieldDescriptor*>::const_iterator intermediate_fields_iter,
4041    vector<const FieldDescriptor*>::const_iterator intermediate_fields_end,
4042    const FieldDescriptor* innermost_field, const string& debug_msg_name,
4043    const UnknownFieldSet& unknown_fields) {
4044  // We do linear searches of the UnknownFieldSet and its sub-groups.  This
4045  // should be fine since it's unlikely that any one options structure will
4046  // contain more than a handful of options.
4047
4048  if (intermediate_fields_iter == intermediate_fields_end) {
4049    // We're at the innermost submessage.
4050    for (int i = 0; i < unknown_fields.field_count(); i++) {
4051      if (unknown_fields.field(i).number() == innermost_field->number()) {
4052        return AddNameError("Option \"" + debug_msg_name +
4053                            "\" was already set.");
4054      }
4055    }
4056    return true;
4057  }
4058
4059  for (int i = 0; i < unknown_fields.field_count(); i++) {
4060    if (unknown_fields.field(i).number() ==
4061        (*intermediate_fields_iter)->number()) {
4062      const UnknownField* unknown_field = &unknown_fields.field(i);
4063      FieldDescriptor::Type type = (*intermediate_fields_iter)->type();
4064      // Recurse into the next submessage.
4065      switch (type) {
4066        case FieldDescriptor::TYPE_MESSAGE:
4067          if (unknown_field->type() == UnknownField::TYPE_LENGTH_DELIMITED) {
4068            UnknownFieldSet intermediate_unknown_fields;
4069            if (intermediate_unknown_fields.ParseFromString(
4070                    unknown_field->length_delimited()) &&
4071                !ExamineIfOptionIsSet(intermediate_fields_iter + 1,
4072                                      intermediate_fields_end,
4073                                      innermost_field, debug_msg_name,
4074                                      intermediate_unknown_fields)) {
4075              return false;  // Error already added.
4076            }
4077          }
4078          break;
4079
4080        case FieldDescriptor::TYPE_GROUP:
4081          if (unknown_field->type() == UnknownField::TYPE_GROUP) {
4082            if (!ExamineIfOptionIsSet(intermediate_fields_iter + 1,
4083                                      intermediate_fields_end,
4084                                      innermost_field, debug_msg_name,
4085                                      unknown_field->group())) {
4086              return false;  // Error already added.
4087            }
4088          }
4089          break;
4090
4091        default:
4092          GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_MESSAGE: " << type;
4093          return false;
4094      }
4095    }
4096  }
4097  return true;
4098}
4099
4100bool DescriptorBuilder::OptionInterpreter::SetOptionValue(
4101    const FieldDescriptor* option_field,
4102    UnknownFieldSet* unknown_fields) {
4103  // We switch on the CppType to validate.
4104  switch (option_field->cpp_type()) {
4105
4106    case FieldDescriptor::CPPTYPE_INT32:
4107      if (uninterpreted_option_->has_positive_int_value()) {
4108        if (uninterpreted_option_->positive_int_value() >
4109            static_cast<uint64>(kint32max)) {
4110          return AddValueError("Value out of range for int32 option \"" +
4111                               option_field->full_name() + "\".");
4112        } else {
4113          SetInt32(option_field->number(),
4114                   uninterpreted_option_->positive_int_value(),
4115                   option_field->type(), unknown_fields);
4116        }
4117      } else if (uninterpreted_option_->has_negative_int_value()) {
4118        if (uninterpreted_option_->negative_int_value() <
4119            static_cast<int64>(kint32min)) {
4120          return AddValueError("Value out of range for int32 option \"" +
4121                               option_field->full_name() + "\".");
4122        } else {
4123          SetInt32(option_field->number(),
4124                   uninterpreted_option_->negative_int_value(),
4125                   option_field->type(), unknown_fields);
4126        }
4127      } else {
4128        return AddValueError("Value must be integer for int32 option \"" +
4129                             option_field->full_name() + "\".");
4130      }
4131      break;
4132
4133    case FieldDescriptor::CPPTYPE_INT64:
4134      if (uninterpreted_option_->has_positive_int_value()) {
4135        if (uninterpreted_option_->positive_int_value() >
4136            static_cast<uint64>(kint64max)) {
4137          return AddValueError("Value out of range for int64 option \"" +
4138                               option_field->full_name() + "\".");
4139        } else {
4140          SetInt64(option_field->number(),
4141                   uninterpreted_option_->positive_int_value(),
4142                   option_field->type(), unknown_fields);
4143        }
4144      } else if (uninterpreted_option_->has_negative_int_value()) {
4145        SetInt64(option_field->number(),
4146                 uninterpreted_option_->negative_int_value(),
4147                 option_field->type(), unknown_fields);
4148      } else {
4149        return AddValueError("Value must be integer for int64 option \"" +
4150                             option_field->full_name() + "\".");
4151      }
4152      break;
4153
4154    case FieldDescriptor::CPPTYPE_UINT32:
4155      if (uninterpreted_option_->has_positive_int_value()) {
4156        if (uninterpreted_option_->positive_int_value() > kuint32max) {
4157          return AddValueError("Value out of range for uint32 option \"" +
4158                               option_field->name() + "\".");
4159        } else {
4160          SetUInt32(option_field->number(),
4161                    uninterpreted_option_->positive_int_value(),
4162                    option_field->type(), unknown_fields);
4163        }
4164      } else {
4165        return AddValueError("Value must be non-negative integer for uint32 "
4166                             "option \"" + option_field->full_name() + "\".");
4167      }
4168      break;
4169
4170    case FieldDescriptor::CPPTYPE_UINT64:
4171      if (uninterpreted_option_->has_positive_int_value()) {
4172        SetUInt64(option_field->number(),
4173                  uninterpreted_option_->positive_int_value(),
4174                  option_field->type(), unknown_fields);
4175      } else {
4176        return AddValueError("Value must be non-negative integer for uint64 "
4177                             "option \"" + option_field->full_name() + "\".");
4178      }
4179      break;
4180
4181    case FieldDescriptor::CPPTYPE_FLOAT: {
4182      float value;
4183      if (uninterpreted_option_->has_double_value()) {
4184        value = uninterpreted_option_->double_value();
4185      } else if (uninterpreted_option_->has_positive_int_value()) {
4186        value = uninterpreted_option_->positive_int_value();
4187      } else if (uninterpreted_option_->has_negative_int_value()) {
4188        value = uninterpreted_option_->negative_int_value();
4189      } else {
4190        return AddValueError("Value must be number for float option \"" +
4191                             option_field->full_name() + "\".");
4192      }
4193      unknown_fields->AddFixed32(option_field->number(),
4194          google::protobuf::internal::WireFormatLite::EncodeFloat(value));
4195      break;
4196    }
4197
4198    case FieldDescriptor::CPPTYPE_DOUBLE: {
4199      double value;
4200      if (uninterpreted_option_->has_double_value()) {
4201        value = uninterpreted_option_->double_value();
4202      } else if (uninterpreted_option_->has_positive_int_value()) {
4203        value = uninterpreted_option_->positive_int_value();
4204      } else if (uninterpreted_option_->has_negative_int_value()) {
4205        value = uninterpreted_option_->negative_int_value();
4206      } else {
4207        return AddValueError("Value must be number for double option \"" +
4208                             option_field->full_name() + "\".");
4209      }
4210      unknown_fields->AddFixed64(option_field->number(),
4211          google::protobuf::internal::WireFormatLite::EncodeDouble(value));
4212      break;
4213    }
4214
4215    case FieldDescriptor::CPPTYPE_BOOL:
4216      uint64 value;
4217      if (!uninterpreted_option_->has_identifier_value()) {
4218        return AddValueError("Value must be identifier for boolean option "
4219                             "\"" + option_field->full_name() + "\".");
4220      }
4221      if (uninterpreted_option_->identifier_value() == "true") {
4222        value = 1;
4223      } else if (uninterpreted_option_->identifier_value() == "false") {
4224        value = 0;
4225      } else {
4226        return AddValueError("Value must be \"true\" or \"false\" for boolean "
4227                             "option \"" + option_field->full_name() + "\".");
4228      }
4229      unknown_fields->AddVarint(option_field->number(), value);
4230      break;
4231
4232    case FieldDescriptor::CPPTYPE_ENUM: {
4233      if (!uninterpreted_option_->has_identifier_value()) {
4234        return AddValueError("Value must be identifier for enum-valued option "
4235                             "\"" + option_field->full_name() + "\".");
4236      }
4237      const EnumDescriptor* enum_type = option_field->enum_type();
4238      const string& value_name = uninterpreted_option_->identifier_value();
4239      const EnumValueDescriptor* enum_value = NULL;
4240
4241      if (enum_type->file()->pool() != DescriptorPool::generated_pool()) {
4242        // Note that the enum value's fully-qualified name is a sibling of the
4243        // enum's name, not a child of it.
4244        string fully_qualified_name = enum_type->full_name();
4245        fully_qualified_name.resize(fully_qualified_name.size() -
4246                                    enum_type->name().size());
4247        fully_qualified_name += value_name;
4248
4249        // Search for the enum value's descriptor in the builder's pool. Note
4250        // that we use DescriptorBuilder::FindSymbolNotEnforcingDeps(), not
4251        // DescriptorPool::FindEnumValueByName() because we're already holding
4252        // the pool's mutex, and the latter method locks it again.
4253        Symbol symbol =
4254          builder_->FindSymbolNotEnforcingDeps(fully_qualified_name);
4255        if (!symbol.IsNull() && symbol.type == Symbol::ENUM_VALUE) {
4256          if (symbol.enum_value_descriptor->type() != enum_type) {
4257            return AddValueError("Enum type \"" + enum_type->full_name() +
4258                "\" has no value named \"" + value_name + "\" for option \"" +
4259                option_field->full_name() +
4260                "\". This appears to be a value from a sibling type.");
4261          } else {
4262            enum_value = symbol.enum_value_descriptor;
4263          }
4264        }
4265      } else {
4266        // The enum type is in the generated pool, so we can search for the
4267        // value there.
4268        enum_value = enum_type->FindValueByName(value_name);
4269      }
4270
4271      if (enum_value == NULL) {
4272        return AddValueError("Enum type \"" +
4273                             option_field->enum_type()->full_name() +
4274                             "\" has no value named \"" + value_name + "\" for "
4275                             "option \"" + option_field->full_name() + "\".");
4276      } else {
4277        // Sign-extension is not a problem, since we cast directly from int32 to
4278        // uint64, without first going through uint32.
4279        unknown_fields->AddVarint(option_field->number(),
4280          static_cast<uint64>(static_cast<int64>(enum_value->number())));
4281      }
4282      break;
4283    }
4284
4285    case FieldDescriptor::CPPTYPE_STRING:
4286      if (!uninterpreted_option_->has_string_value()) {
4287        return AddValueError("Value must be quoted string for string option "
4288                             "\"" + option_field->full_name() + "\".");
4289      }
4290      // The string has already been unquoted and unescaped by the parser.
4291      unknown_fields->AddLengthDelimited(option_field->number(),
4292          uninterpreted_option_->string_value());
4293      break;
4294
4295    case FieldDescriptor::CPPTYPE_MESSAGE:
4296      // We don't currently support defining a message-typed option, so we
4297      // should never actually get here.
4298      return AddValueError("Option \"" + option_field->full_name() +
4299                           "\" is a message.  To set fields within it, use "
4300                           "syntax like \"" + option_field->name() +
4301                           ".foo = value\".");
4302      break;
4303  }
4304
4305  return true;
4306}
4307
4308void DescriptorBuilder::OptionInterpreter::SetInt32(int number, int32 value,
4309    FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4310  switch (type) {
4311    case FieldDescriptor::TYPE_INT32:
4312      unknown_fields->AddVarint(number,
4313        static_cast<uint64>(static_cast<int64>(value)));
4314      break;
4315
4316    case FieldDescriptor::TYPE_SFIXED32:
4317      unknown_fields->AddFixed32(number, static_cast<uint32>(value));
4318      break;
4319
4320    case FieldDescriptor::TYPE_SINT32:
4321      unknown_fields->AddVarint(number,
4322          google::protobuf::internal::WireFormatLite::ZigZagEncode32(value));
4323      break;
4324
4325    default:
4326      GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_INT32: " << type;
4327      break;
4328  }
4329}
4330
4331void DescriptorBuilder::OptionInterpreter::SetInt64(int number, int64 value,
4332    FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4333  switch (type) {
4334    case FieldDescriptor::TYPE_INT64:
4335      unknown_fields->AddVarint(number, static_cast<uint64>(value));
4336      break;
4337
4338    case FieldDescriptor::TYPE_SFIXED64:
4339      unknown_fields->AddFixed64(number, static_cast<uint64>(value));
4340      break;
4341
4342    case FieldDescriptor::TYPE_SINT64:
4343      unknown_fields->AddVarint(number,
4344          google::protobuf::internal::WireFormatLite::ZigZagEncode64(value));
4345      break;
4346
4347    default:
4348      GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_INT64: " << type;
4349      break;
4350  }
4351}
4352
4353void DescriptorBuilder::OptionInterpreter::SetUInt32(int number, uint32 value,
4354    FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4355  switch (type) {
4356    case FieldDescriptor::TYPE_UINT32:
4357      unknown_fields->AddVarint(number, static_cast<uint64>(value));
4358      break;
4359
4360    case FieldDescriptor::TYPE_FIXED32:
4361      unknown_fields->AddFixed32(number, static_cast<uint32>(value));
4362      break;
4363
4364    default:
4365      GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_UINT32: " << type;
4366      break;
4367  }
4368}
4369
4370void DescriptorBuilder::OptionInterpreter::SetUInt64(int number, uint64 value,
4371    FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4372  switch (type) {
4373    case FieldDescriptor::TYPE_UINT64:
4374      unknown_fields->AddVarint(number, value);
4375      break;
4376
4377    case FieldDescriptor::TYPE_FIXED64:
4378      unknown_fields->AddFixed64(number, value);
4379      break;
4380
4381    default:
4382      GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_UINT64: " << type;
4383      break;
4384  }
4385}
4386
4387}  // namespace protobuf
4388}  // namespace google
4389