• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Protocol Buffers - Google's data interchange format
2 // Copyright 2008 Google Inc.  All rights reserved.
3 // https://developers.google.com/protocol-buffers/
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 <memory>
38 #ifndef _SHARED_PTR_H
39 #include <google/protobuf/stubs/shared_ptr.h>
40 #endif
41 #include <set>
42 #include <string>
43 #include <vector>
44 #include <algorithm>
45 #include <limits>
46 
47 #include <google/protobuf/descriptor.h>
48 #include <google/protobuf/descriptor_database.h>
49 #include <google/protobuf/descriptor.pb.h>
50 #include <google/protobuf/dynamic_message.h>
51 #include <google/protobuf/generated_message_util.h>
52 #include <google/protobuf/text_format.h>
53 #include <google/protobuf/unknown_field_set.h>
54 #include <google/protobuf/wire_format.h>
55 #include <google/protobuf/io/strtod.h>
56 #include <google/protobuf/io/coded_stream.h>
57 #include <google/protobuf/io/tokenizer.h>
58 #include <google/protobuf/io/zero_copy_stream_impl.h>
59 #include <google/protobuf/stubs/common.h>
60 #include <google/protobuf/stubs/logging.h>
61 #include <google/protobuf/stubs/mutex.h>
62 #include <google/protobuf/stubs/once.h>
63 #include <google/protobuf/stubs/stringprintf.h>
64 #include <google/protobuf/stubs/strutil.h>
65 #include <google/protobuf/stubs/substitute.h>
66 #include <google/protobuf/stubs/map_util.h>
67 #include <google/protobuf/stubs/stl_util.h>
68 
69 #undef PACKAGE  // autoheader #defines this.  :(
70 
71 namespace google {
72 namespace protobuf {
73 
74 const FieldDescriptor::CppType
75 FieldDescriptor::kTypeToCppTypeMap[MAX_TYPE + 1] = {
76   static_cast<CppType>(0),  // 0 is reserved for errors
77 
78   CPPTYPE_DOUBLE,   // TYPE_DOUBLE
79   CPPTYPE_FLOAT,    // TYPE_FLOAT
80   CPPTYPE_INT64,    // TYPE_INT64
81   CPPTYPE_UINT64,   // TYPE_UINT64
82   CPPTYPE_INT32,    // TYPE_INT32
83   CPPTYPE_UINT64,   // TYPE_FIXED64
84   CPPTYPE_UINT32,   // TYPE_FIXED32
85   CPPTYPE_BOOL,     // TYPE_BOOL
86   CPPTYPE_STRING,   // TYPE_STRING
87   CPPTYPE_MESSAGE,  // TYPE_GROUP
88   CPPTYPE_MESSAGE,  // TYPE_MESSAGE
89   CPPTYPE_STRING,   // TYPE_BYTES
90   CPPTYPE_UINT32,   // TYPE_UINT32
91   CPPTYPE_ENUM,     // TYPE_ENUM
92   CPPTYPE_INT32,    // TYPE_SFIXED32
93   CPPTYPE_INT64,    // TYPE_SFIXED64
94   CPPTYPE_INT32,    // TYPE_SINT32
95   CPPTYPE_INT64,    // TYPE_SINT64
96 };
97 
98 const char * const FieldDescriptor::kTypeToName[MAX_TYPE + 1] = {
99   "ERROR",     // 0 is reserved for errors
100 
101   "double",    // TYPE_DOUBLE
102   "float",     // TYPE_FLOAT
103   "int64",     // TYPE_INT64
104   "uint64",    // TYPE_UINT64
105   "int32",     // TYPE_INT32
106   "fixed64",   // TYPE_FIXED64
107   "fixed32",   // TYPE_FIXED32
108   "bool",      // TYPE_BOOL
109   "string",    // TYPE_STRING
110   "group",     // TYPE_GROUP
111   "message",   // TYPE_MESSAGE
112   "bytes",     // TYPE_BYTES
113   "uint32",    // TYPE_UINT32
114   "enum",      // TYPE_ENUM
115   "sfixed32",  // TYPE_SFIXED32
116   "sfixed64",  // TYPE_SFIXED64
117   "sint32",    // TYPE_SINT32
118   "sint64",    // TYPE_SINT64
119 };
120 
121 const char * const FieldDescriptor::kCppTypeToName[MAX_CPPTYPE + 1] = {
122   "ERROR",     // 0 is reserved for errors
123 
124   "int32",     // CPPTYPE_INT32
125   "int64",     // CPPTYPE_INT64
126   "uint32",    // CPPTYPE_UINT32
127   "uint64",    // CPPTYPE_UINT64
128   "double",    // CPPTYPE_DOUBLE
129   "float",     // CPPTYPE_FLOAT
130   "bool",      // CPPTYPE_BOOL
131   "enum",      // CPPTYPE_ENUM
132   "string",    // CPPTYPE_STRING
133   "message",   // CPPTYPE_MESSAGE
134 };
135 
136 const char * const FieldDescriptor::kLabelToName[MAX_LABEL + 1] = {
137   "ERROR",     // 0 is reserved for errors
138 
139   "optional",  // LABEL_OPTIONAL
140   "required",  // LABEL_REQUIRED
141   "repeated",  // LABEL_REPEATED
142 };
143 
SyntaxName(FileDescriptor::Syntax syntax)144 const char* FileDescriptor::SyntaxName(FileDescriptor::Syntax syntax) {
145   switch (syntax) {
146     case SYNTAX_PROTO2:
147       return "proto2";
148     case SYNTAX_PROTO3:
149       return "proto3";
150     case SYNTAX_UNKNOWN:
151       return "unknown";
152   }
153   GOOGLE_LOG(FATAL) << "can't reach here.";
154   return NULL;
155 }
156 
157 static const char * const kNonLinkedWeakMessageReplacementName = "google.protobuf.Empty";
158 
159 #if !defined(_MSC_VER) || _MSC_VER >= 1900
160 const int FieldDescriptor::kMaxNumber;
161 const int FieldDescriptor::kFirstReservedNumber;
162 const int FieldDescriptor::kLastReservedNumber;
163 #endif
164 
165 namespace {
166 
ToCamelCase(const string & input,bool lower_first)167 string ToCamelCase(const string& input, bool lower_first) {
168   bool capitalize_next = !lower_first;
169   string result;
170   result.reserve(input.size());
171 
172   for (int i = 0; i < input.size(); i++) {
173     if (input[i] == '_') {
174       capitalize_next = true;
175     } else if (capitalize_next) {
176       // Note:  I distrust ctype.h due to locales.
177       if ('a' <= input[i] && input[i] <= 'z') {
178         result.push_back(input[i] - 'a' + 'A');
179       } else {
180         result.push_back(input[i]);
181       }
182       capitalize_next = false;
183     } else {
184       result.push_back(input[i]);
185     }
186   }
187 
188   // Lower-case the first letter.
189   if (lower_first && !result.empty() && 'A' <= result[0] && result[0] <= 'Z') {
190       result[0] = result[0] - 'A' + 'a';
191   }
192 
193   return result;
194 }
195 
196 // A DescriptorPool contains a bunch of hash_maps to implement the
197 // various Find*By*() methods.  Since hashtable lookups are O(1), it's
198 // most efficient to construct a fixed set of large hash_maps used by
199 // all objects in the pool rather than construct one or more small
200 // hash_maps for each object.
201 //
202 // The keys to these hash_maps are (parent, name) or (parent, number)
203 // pairs.  Unfortunately STL doesn't provide hash functions for pair<>,
204 // so we must invent our own.
205 //
206 // TODO(kenton):  Use StringPiece rather than const char* in keys?  It would
207 //   be a lot cleaner but we'd just have to convert it back to const char*
208 //   for the open source release.
209 
210 typedef pair<const void*, const char*> PointerStringPair;
211 
212 struct PointerStringPairEqual {
operator ()google::protobuf::__anoncbf42d300111::PointerStringPairEqual213   inline bool operator()(const PointerStringPair& a,
214                          const PointerStringPair& b) const {
215     return a.first == b.first && strcmp(a.second, b.second) == 0;
216   }
217 };
218 
219 template<typename PairType>
220 struct PointerIntegerPairHash {
operator ()google::protobuf::__anoncbf42d300111::PointerIntegerPairHash221   size_t operator()(const PairType& p) const {
222     // FIXME(kenton):  What is the best way to compute this hash?  I have
223     // no idea!  This seems a bit better than an XOR.
224     return reinterpret_cast<intptr_t>(p.first) * ((1 << 16) - 1) + p.second;
225   }
226 
227 #ifdef _MSC_VER
228   // Used only by MSVC and platforms where hash_map is not available.
229   static const size_t bucket_size = 4;
230   static const size_t min_buckets = 8;
231 #endif
operator ()google::protobuf::__anoncbf42d300111::PointerIntegerPairHash232   inline bool operator()(const PairType& a, const PairType& b) const {
233     return a.first < b.first ||
234           (a.first == b.first && a.second < b.second);
235   }
236 };
237 
238 typedef pair<const Descriptor*, int> DescriptorIntPair;
239 typedef pair<const EnumDescriptor*, int> EnumIntPair;
240 
241 struct PointerStringPairHash {
operator ()google::protobuf::__anoncbf42d300111::PointerStringPairHash242   size_t operator()(const PointerStringPair& p) const {
243     // FIXME(kenton):  What is the best way to compute this hash?  I have
244     // no idea!  This seems a bit better than an XOR.
245     hash<const char*> cstring_hash;
246     return reinterpret_cast<intptr_t>(p.first) * ((1 << 16) - 1) +
247            cstring_hash(p.second);
248   }
249 
250 #ifdef _MSC_VER
251   // Used only by MSVC and platforms where hash_map is not available.
252   static const size_t bucket_size = 4;
253   static const size_t min_buckets = 8;
254 #endif
operator ()google::protobuf::__anoncbf42d300111::PointerStringPairHash255   inline bool operator()(const PointerStringPair& a,
256                          const PointerStringPair& b) const {
257     if (a.first < b.first) return true;
258     if (a.first > b.first) return false;
259     return strcmp(a.second, b.second) < 0;
260   }
261 };
262 
263 
264 struct Symbol {
265   enum Type {
266     NULL_SYMBOL, MESSAGE, FIELD, ONEOF, ENUM, ENUM_VALUE, SERVICE, METHOD,
267     PACKAGE
268   };
269   Type type;
270   union {
271     const Descriptor* descriptor;
272     const FieldDescriptor* field_descriptor;
273     const OneofDescriptor* oneof_descriptor;
274     const EnumDescriptor* enum_descriptor;
275     const EnumValueDescriptor* enum_value_descriptor;
276     const ServiceDescriptor* service_descriptor;
277     const MethodDescriptor* method_descriptor;
278     const FileDescriptor* package_file_descriptor;
279   };
280 
Symbolgoogle::protobuf::__anoncbf42d300111::Symbol281   inline Symbol() : type(NULL_SYMBOL) { descriptor = NULL; }
IsNullgoogle::protobuf::__anoncbf42d300111::Symbol282   inline bool IsNull() const { return type == NULL_SYMBOL; }
IsTypegoogle::protobuf::__anoncbf42d300111::Symbol283   inline bool IsType() const {
284     return type == MESSAGE || type == ENUM;
285   }
IsAggregategoogle::protobuf::__anoncbf42d300111::Symbol286   inline bool IsAggregate() const {
287     return type == MESSAGE || type == PACKAGE
288         || type == ENUM || type == SERVICE;
289   }
290 
291 #define CONSTRUCTOR(TYPE, TYPE_CONSTANT, FIELD)  \
292   inline explicit Symbol(const TYPE* value) {    \
293     type = TYPE_CONSTANT;                        \
294     this->FIELD = value;                         \
295   }
296 
CONSTRUCTORgoogle::protobuf::__anoncbf42d300111::Symbol297   CONSTRUCTOR(Descriptor         , MESSAGE   , descriptor             )
298   CONSTRUCTOR(FieldDescriptor    , FIELD     , field_descriptor       )
299   CONSTRUCTOR(OneofDescriptor    , ONEOF     , oneof_descriptor       )
300   CONSTRUCTOR(EnumDescriptor     , ENUM      , enum_descriptor        )
301   CONSTRUCTOR(EnumValueDescriptor, ENUM_VALUE, enum_value_descriptor  )
302   CONSTRUCTOR(ServiceDescriptor  , SERVICE   , service_descriptor     )
303   CONSTRUCTOR(MethodDescriptor   , METHOD    , method_descriptor      )
304   CONSTRUCTOR(FileDescriptor     , PACKAGE   , package_file_descriptor)
305 #undef CONSTRUCTOR
306 
307   const FileDescriptor* GetFile() const {
308     switch (type) {
309       case NULL_SYMBOL: return NULL;
310       case MESSAGE    : return descriptor           ->file();
311       case FIELD      : return field_descriptor     ->file();
312       case ONEOF      : return oneof_descriptor     ->containing_type()->file();
313       case ENUM       : return enum_descriptor      ->file();
314       case ENUM_VALUE : return enum_value_descriptor->type()->file();
315       case SERVICE    : return service_descriptor   ->file();
316       case METHOD     : return method_descriptor    ->service()->file();
317       case PACKAGE    : return package_file_descriptor;
318     }
319     return NULL;
320   }
321 };
322 
323 const Symbol kNullSymbol;
324 
325 typedef hash_map<const char*, Symbol,
326                  hash<const char*>, streq>
327   SymbolsByNameMap;
328 typedef hash_map<PointerStringPair, Symbol,
329                  PointerStringPairHash, PointerStringPairEqual>
330   SymbolsByParentMap;
331 typedef hash_map<const char*, const FileDescriptor*,
332                  hash<const char*>, streq>
333   FilesByNameMap;
334 typedef hash_map<PointerStringPair, const FieldDescriptor*,
335                  PointerStringPairHash, PointerStringPairEqual>
336   FieldsByNameMap;
337 typedef hash_map<DescriptorIntPair, const FieldDescriptor*,
338                  PointerIntegerPairHash<DescriptorIntPair> >
339   FieldsByNumberMap;
340 typedef hash_map<EnumIntPair, const EnumValueDescriptor*,
341                  PointerIntegerPairHash<EnumIntPair> >
342   EnumValuesByNumberMap;
343 // This is a map rather than a hash_map, since we use it to iterate
344 // through all the extensions that extend a given Descriptor, and an
345 // ordered data structure that implements lower_bound is convenient
346 // for that.
347 typedef map<DescriptorIntPair, const FieldDescriptor*>
348   ExtensionsGroupedByDescriptorMap;
349 typedef hash_map<string, const SourceCodeInfo_Location*> LocationsByPathMap;
350 
351 set<string>* allowed_proto3_extendees_ = NULL;
352 GOOGLE_PROTOBUF_DECLARE_ONCE(allowed_proto3_extendees_init_);
353 
DeleteAllowedProto3Extendee()354 void DeleteAllowedProto3Extendee() {
355   delete allowed_proto3_extendees_;
356 }
357 
InitAllowedProto3Extendee()358 void InitAllowedProto3Extendee() {
359   allowed_proto3_extendees_ = new set<string>;
360   const char* kOptionNames[] = {
361       "FileOptions",      "MessageOptions", "FieldOptions", "EnumOptions",
362       "EnumValueOptions", "ServiceOptions", "MethodOptions"};
363   for (int i = 0; i < GOOGLE_ARRAYSIZE(kOptionNames); ++i) {
364     // descriptor.proto has a different package name in opensource. We allow
365     // both so the opensource protocol compiler can also compile internal
366     // proto3 files with custom options. See: b/27567912
367     allowed_proto3_extendees_->insert(string("google.protobuf.") +
368                                       kOptionNames[i]);
369     // Split the word to trick the opensource processing scripts so they
370     // will keep the origial package name.
371     allowed_proto3_extendees_->insert(string("proto") + "2." + kOptionNames[i]);
372   }
373 
374   google::protobuf::internal::OnShutdown(&DeleteAllowedProto3Extendee);
375 }
376 
377 // Checks whether the extendee type is allowed in proto3.
378 // Only extensions to descriptor options are allowed. We use name comparison
379 // instead of comparing the descriptor directly because the extensions may be
380 // defined in a different pool.
AllowedExtendeeInProto3(const string & name)381 bool AllowedExtendeeInProto3(const string& name) {
382   ::google::protobuf::GoogleOnceInit(&allowed_proto3_extendees_init_, &InitAllowedProto3Extendee);
383   return allowed_proto3_extendees_->find(name) !=
384          allowed_proto3_extendees_->end();
385 }
386 
387 }  // anonymous namespace
388 
389 // ===================================================================
390 // DescriptorPool::Tables
391 
392 class DescriptorPool::Tables {
393  public:
394   Tables();
395   ~Tables();
396 
397   // Record the current state of the tables to the stack of checkpoints.
398   // Each call to AddCheckpoint() must be paired with exactly one call to either
399   // ClearLastCheckpoint() or RollbackToLastCheckpoint().
400   //
401   // This is used when building files, since some kinds of validation errors
402   // cannot be detected until the file's descriptors have already been added to
403   // the tables.
404   //
405   // This supports recursive checkpoints, since building a file may trigger
406   // recursive building of other files. Note that recursive checkpoints are not
407   // normally necessary; explicit dependencies are built prior to checkpointing.
408   // So although we recursively build transitive imports, there is at most one
409   // checkpoint in the stack during dependency building.
410   //
411   // Recursive checkpoints only arise during cross-linking of the descriptors.
412   // Symbol references must be resolved, via DescriptorBuilder::FindSymbol and
413   // friends. If the pending file references an unknown symbol
414   // (e.g., it is not defined in the pending file's explicit dependencies), and
415   // the pool is using a fallback database, and that database contains a file
416   // defining that symbol, and that file has not yet been built by the pool,
417   // the pool builds the file during cross-linking, leading to another
418   // checkpoint.
419   void AddCheckpoint();
420 
421   // Mark the last checkpoint as having cleared successfully, removing it from
422   // the stack. If the stack is empty, all pending symbols will be committed.
423   //
424   // Note that this does not guarantee that the symbols added since the last
425   // checkpoint won't be rolled back: if a checkpoint gets rolled back,
426   // everything past that point gets rolled back, including symbols added after
427   // checkpoints that were pushed onto the stack after it and marked as cleared.
428   void ClearLastCheckpoint();
429 
430   // Roll back the Tables to the state of the checkpoint at the top of the
431   // stack, removing everything that was added after that point.
432   void RollbackToLastCheckpoint();
433 
434   // The stack of files which are currently being built.  Used to detect
435   // cyclic dependencies when loading files from a DescriptorDatabase.  Not
436   // used when fallback_database_ == NULL.
437   vector<string> pending_files_;
438 
439   // A set of files which we have tried to load from the fallback database
440   // and encountered errors.  We will not attempt to load them again during
441   // execution of the current public API call, but for compatibility with
442   // legacy clients, this is cleared at the beginning of each public API call.
443   // Not used when fallback_database_ == NULL.
444   hash_set<string> known_bad_files_;
445 
446   // A set of symbols which we have tried to load from the fallback database
447   // and encountered errors. We will not attempt to load them again during
448   // execution of the current public API call, but for compatibility with
449   // legacy clients, this is cleared at the beginning of each public API call.
450   hash_set<string> known_bad_symbols_;
451 
452   // The set of descriptors for which we've already loaded the full
453   // set of extensions numbers from fallback_database_.
454   hash_set<const Descriptor*> extensions_loaded_from_db_;
455 
456   // -----------------------------------------------------------------
457   // Finding items.
458 
459   // Find symbols.  This returns a null Symbol (symbol.IsNull() is true)
460   // if not found.
461   inline Symbol FindSymbol(const string& key) const;
462 
463   // This implements the body of DescriptorPool::Find*ByName().  It should
464   // really be a private method of DescriptorPool, but that would require
465   // declaring Symbol in descriptor.h, which would drag all kinds of other
466   // stuff into the header.  Yay C++.
467   Symbol FindByNameHelper(
468     const DescriptorPool* pool, const string& name);
469 
470   // These return NULL if not found.
471   inline const FileDescriptor* FindFile(const string& key) const;
472   inline const FieldDescriptor* FindExtension(const Descriptor* extendee,
473                                               int number);
474   inline void FindAllExtensions(const Descriptor* extendee,
475                                 vector<const FieldDescriptor*>* out) const;
476 
477   // -----------------------------------------------------------------
478   // Adding items.
479 
480   // These add items to the corresponding tables.  They return false if
481   // the key already exists in the table.  For AddSymbol(), the string passed
482   // in must be one that was constructed using AllocateString(), as it will
483   // be used as a key in the symbols_by_name_ map without copying.
484   bool AddSymbol(const string& full_name, Symbol symbol);
485   bool AddFile(const FileDescriptor* file);
486   bool AddExtension(const FieldDescriptor* field);
487 
488   // -----------------------------------------------------------------
489   // Allocating memory.
490 
491   // Allocate an object which will be reclaimed when the pool is
492   // destroyed.  Note that the object's destructor will never be called,
493   // so its fields must be plain old data (primitive data types and
494   // pointers).  All of the descriptor types are such objects.
495   template<typename Type> Type* Allocate();
496 
497   // Allocate an array of objects which will be reclaimed when the
498   // pool in destroyed.  Again, destructors are never called.
499   template<typename Type> Type* AllocateArray(int count);
500 
501   // Allocate a string which will be destroyed when the pool is destroyed.
502   // The string is initialized to the given value for convenience.
503   string* AllocateString(const string& value);
504 
505   // Allocate a protocol message object.  Some older versions of GCC have
506   // trouble understanding explicit template instantiations in some cases, so
507   // in those cases we have to pass a dummy pointer of the right type as the
508   // parameter instead of specifying the type explicitly.
509   template<typename Type> Type* AllocateMessage(Type* dummy = NULL);
510 
511   // Allocate a FileDescriptorTables object.
512   FileDescriptorTables* AllocateFileTables();
513 
514  private:
515   vector<string*> strings_;    // All strings in the pool.
516   vector<Message*> messages_;  // All messages in the pool.
517   vector<FileDescriptorTables*> file_tables_;  // All file tables in the pool.
518   vector<void*> allocations_;  // All other memory allocated in the pool.
519 
520   SymbolsByNameMap      symbols_by_name_;
521   FilesByNameMap        files_by_name_;
522   ExtensionsGroupedByDescriptorMap extensions_;
523 
524   struct CheckPoint {
CheckPointgoogle::protobuf::DescriptorPool::Tables::CheckPoint525     explicit CheckPoint(const Tables* tables)
526       : strings_before_checkpoint(tables->strings_.size()),
527         messages_before_checkpoint(tables->messages_.size()),
528         file_tables_before_checkpoint(tables->file_tables_.size()),
529         allocations_before_checkpoint(tables->allocations_.size()),
530         pending_symbols_before_checkpoint(
531             tables->symbols_after_checkpoint_.size()),
532         pending_files_before_checkpoint(
533             tables->files_after_checkpoint_.size()),
534         pending_extensions_before_checkpoint(
535             tables->extensions_after_checkpoint_.size()) {
536     }
537     int strings_before_checkpoint;
538     int messages_before_checkpoint;
539     int file_tables_before_checkpoint;
540     int allocations_before_checkpoint;
541     int pending_symbols_before_checkpoint;
542     int pending_files_before_checkpoint;
543     int pending_extensions_before_checkpoint;
544   };
545   vector<CheckPoint> checkpoints_;
546   vector<const char*      > symbols_after_checkpoint_;
547   vector<const char*      > files_after_checkpoint_;
548   vector<DescriptorIntPair> extensions_after_checkpoint_;
549 
550   // Allocate some bytes which will be reclaimed when the pool is
551   // destroyed.
552   void* AllocateBytes(int size);
553 };
554 
555 // Contains tables specific to a particular file.  These tables are not
556 // modified once the file has been constructed, so they need not be
557 // protected by a mutex.  This makes operations that depend only on the
558 // contents of a single file -- e.g. Descriptor::FindFieldByName() --
559 // lock-free.
560 //
561 // For historical reasons, the definitions of the methods of
562 // FileDescriptorTables and DescriptorPool::Tables are interleaved below.
563 // These used to be a single class.
564 class FileDescriptorTables {
565  public:
566   FileDescriptorTables();
567   ~FileDescriptorTables();
568 
569   // Empty table, used with placeholder files.
570   inline static const FileDescriptorTables& GetEmptyInstance();
571 
572   // -----------------------------------------------------------------
573   // Finding items.
574 
575   // Find symbols.  These return a null Symbol (symbol.IsNull() is true)
576   // if not found.
577   inline Symbol FindNestedSymbol(const void* parent,
578                                  const string& name) const;
579   inline Symbol FindNestedSymbolOfType(const void* parent,
580                                        const string& name,
581                                        const Symbol::Type type) const;
582 
583   // These return NULL if not found.
584   inline const FieldDescriptor* FindFieldByNumber(
585     const Descriptor* parent, int number) const;
586   inline const FieldDescriptor* FindFieldByLowercaseName(
587     const void* parent, const string& lowercase_name) const;
588   inline const FieldDescriptor* FindFieldByCamelcaseName(
589     const void* parent, const string& camelcase_name) const;
590   inline const EnumValueDescriptor* FindEnumValueByNumber(
591     const EnumDescriptor* parent, int number) const;
592   // This creates a new EnumValueDescriptor if not found, in a thread-safe way.
593   inline const EnumValueDescriptor* FindEnumValueByNumberCreatingIfUnknown(
594       const EnumDescriptor* parent, int number) const;
595 
596   // -----------------------------------------------------------------
597   // Adding items.
598 
599   // These add items to the corresponding tables.  They return false if
600   // the key already exists in the table.  For AddAliasUnderParent(), the
601   // string passed in must be one that was constructed using AllocateString(),
602   // as it will be used as a key in the symbols_by_parent_ map without copying.
603   bool AddAliasUnderParent(const void* parent, const string& name,
604                            Symbol symbol);
605   bool AddFieldByNumber(const FieldDescriptor* field);
606   bool AddEnumValueByNumber(const EnumValueDescriptor* value);
607 
608   // Adds the field to the lowercase_name and camelcase_name maps.  Never
609   // fails because we allow duplicates; the first field by the name wins.
610   void AddFieldByStylizedNames(const FieldDescriptor* field);
611 
612   // Populates p->first->locations_by_path_ from p->second.
613   // Unusual signature dictated by GoogleOnceDynamic.
614   static void BuildLocationsByPath(
615       pair<const FileDescriptorTables*, const SourceCodeInfo*>* p);
616 
617   // Returns the location denoted by the specified path through info,
618   // or NULL if not found.
619   // The value of info must be that of the corresponding FileDescriptor.
620   // (Conceptually a pure function, but stateful as an optimisation.)
621   const SourceCodeInfo_Location* GetSourceLocation(
622       const vector<int>& path, const SourceCodeInfo* info) const;
623 
624  private:
625   SymbolsByParentMap    symbols_by_parent_;
626   FieldsByNameMap       fields_by_lowercase_name_;
627   FieldsByNameMap       fields_by_camelcase_name_;
628   FieldsByNumberMap     fields_by_number_;       // Not including extensions.
629   EnumValuesByNumberMap enum_values_by_number_;
630   mutable EnumValuesByNumberMap unknown_enum_values_by_number_
631       GOOGLE_GUARDED_BY(unknown_enum_values_mu_);
632 
633   // Populated on first request to save space, hence constness games.
634   mutable GoogleOnceDynamic locations_by_path_once_;
635   mutable LocationsByPathMap locations_by_path_;
636 
637   // Mutex to protect the unknown-enum-value map due to dynamic
638   // EnumValueDescriptor creation on unknown values.
639   mutable Mutex unknown_enum_values_mu_;
640 };
641 
Tables()642 DescriptorPool::Tables::Tables()
643     // Start some hash_map and hash_set objects with a small # of buckets
644     : known_bad_files_(3),
645       known_bad_symbols_(3),
646       extensions_loaded_from_db_(3),
647       symbols_by_name_(3),
648       files_by_name_(3) {}
649 
650 
~Tables()651 DescriptorPool::Tables::~Tables() {
652   GOOGLE_DCHECK(checkpoints_.empty());
653   // Note that the deletion order is important, since the destructors of some
654   // messages may refer to objects in allocations_.
655   STLDeleteElements(&messages_);
656   for (int i = 0; i < allocations_.size(); i++) {
657     operator delete(allocations_[i]);
658   }
659   STLDeleteElements(&strings_);
660   STLDeleteElements(&file_tables_);
661 }
662 
FileDescriptorTables()663 FileDescriptorTables::FileDescriptorTables()
664     // Initialize all the hash tables to start out with a small # of buckets
665     : symbols_by_parent_(3),
666       fields_by_lowercase_name_(3),
667       fields_by_camelcase_name_(3),
668       fields_by_number_(3),
669       enum_values_by_number_(3),
670       unknown_enum_values_by_number_(3) {
671 }
672 
~FileDescriptorTables()673 FileDescriptorTables::~FileDescriptorTables() {}
674 
675 namespace {
676 
677 FileDescriptorTables* file_descriptor_tables_ = NULL;
678 GOOGLE_PROTOBUF_DECLARE_ONCE(file_descriptor_tables_once_init_);
679 
DeleteFileDescriptorTables()680 void DeleteFileDescriptorTables() {
681   delete file_descriptor_tables_;
682   file_descriptor_tables_ = NULL;
683 }
684 
InitFileDescriptorTables()685 void InitFileDescriptorTables() {
686   file_descriptor_tables_ = new FileDescriptorTables();
687   internal::OnShutdown(&DeleteFileDescriptorTables);
688 }
689 
InitFileDescriptorTablesOnce()690 inline void InitFileDescriptorTablesOnce() {
691   ::google::protobuf::GoogleOnceInit(
692       &file_descriptor_tables_once_init_, &InitFileDescriptorTables);
693 }
694 
695 }  // anonymous namespace
696 
GetEmptyInstance()697 inline const FileDescriptorTables& FileDescriptorTables::GetEmptyInstance() {
698   InitFileDescriptorTablesOnce();
699   return *file_descriptor_tables_;
700 }
701 
AddCheckpoint()702 void DescriptorPool::Tables::AddCheckpoint() {
703   checkpoints_.push_back(CheckPoint(this));
704 }
705 
ClearLastCheckpoint()706 void DescriptorPool::Tables::ClearLastCheckpoint() {
707   GOOGLE_DCHECK(!checkpoints_.empty());
708   checkpoints_.pop_back();
709   if (checkpoints_.empty()) {
710     // All checkpoints have been cleared: we can now commit all of the pending
711     // data.
712     symbols_after_checkpoint_.clear();
713     files_after_checkpoint_.clear();
714     extensions_after_checkpoint_.clear();
715   }
716 }
717 
RollbackToLastCheckpoint()718 void DescriptorPool::Tables::RollbackToLastCheckpoint() {
719   GOOGLE_DCHECK(!checkpoints_.empty());
720   const CheckPoint& checkpoint = checkpoints_.back();
721 
722   for (int i = checkpoint.pending_symbols_before_checkpoint;
723        i < symbols_after_checkpoint_.size();
724        i++) {
725     symbols_by_name_.erase(symbols_after_checkpoint_[i]);
726   }
727   for (int i = checkpoint.pending_files_before_checkpoint;
728        i < files_after_checkpoint_.size();
729        i++) {
730     files_by_name_.erase(files_after_checkpoint_[i]);
731   }
732   for (int i = checkpoint.pending_extensions_before_checkpoint;
733        i < extensions_after_checkpoint_.size();
734        i++) {
735     extensions_.erase(extensions_after_checkpoint_[i]);
736   }
737 
738   symbols_after_checkpoint_.resize(
739       checkpoint.pending_symbols_before_checkpoint);
740   files_after_checkpoint_.resize(checkpoint.pending_files_before_checkpoint);
741   extensions_after_checkpoint_.resize(
742       checkpoint.pending_extensions_before_checkpoint);
743 
744   STLDeleteContainerPointers(
745       strings_.begin() + checkpoint.strings_before_checkpoint, strings_.end());
746   STLDeleteContainerPointers(
747       messages_.begin() + checkpoint.messages_before_checkpoint,
748       messages_.end());
749   STLDeleteContainerPointers(
750       file_tables_.begin() + checkpoint.file_tables_before_checkpoint,
751       file_tables_.end());
752   for (int i = checkpoint.allocations_before_checkpoint;
753        i < allocations_.size();
754        i++) {
755     operator delete(allocations_[i]);
756   }
757 
758   strings_.resize(checkpoint.strings_before_checkpoint);
759   messages_.resize(checkpoint.messages_before_checkpoint);
760   file_tables_.resize(checkpoint.file_tables_before_checkpoint);
761   allocations_.resize(checkpoint.allocations_before_checkpoint);
762   checkpoints_.pop_back();
763 }
764 
765 // -------------------------------------------------------------------
766 
FindSymbol(const string & key) const767 inline Symbol DescriptorPool::Tables::FindSymbol(const string& key) const {
768   const Symbol* result = FindOrNull(symbols_by_name_, key.c_str());
769   if (result == NULL) {
770     return kNullSymbol;
771   } else {
772     return *result;
773   }
774 }
775 
FindNestedSymbol(const void * parent,const string & name) const776 inline Symbol FileDescriptorTables::FindNestedSymbol(
777     const void* parent, const string& name) const {
778   const Symbol* result =
779     FindOrNull(symbols_by_parent_, PointerStringPair(parent, name.c_str()));
780   if (result == NULL) {
781     return kNullSymbol;
782   } else {
783     return *result;
784   }
785 }
786 
FindNestedSymbolOfType(const void * parent,const string & name,const Symbol::Type type) const787 inline Symbol FileDescriptorTables::FindNestedSymbolOfType(
788     const void* parent, const string& name, const Symbol::Type type) const {
789   Symbol result = FindNestedSymbol(parent, name);
790   if (result.type != type) return kNullSymbol;
791   return result;
792 }
793 
FindByNameHelper(const DescriptorPool * pool,const string & name)794 Symbol DescriptorPool::Tables::FindByNameHelper(
795     const DescriptorPool* pool, const string& name) {
796   MutexLockMaybe lock(pool->mutex_);
797   known_bad_symbols_.clear();
798   known_bad_files_.clear();
799   Symbol result = FindSymbol(name);
800 
801   if (result.IsNull() && pool->underlay_ != NULL) {
802     // Symbol not found; check the underlay.
803     result =
804       pool->underlay_->tables_->FindByNameHelper(pool->underlay_, name);
805   }
806 
807   if (result.IsNull()) {
808     // Symbol still not found, so check fallback database.
809     if (pool->TryFindSymbolInFallbackDatabase(name)) {
810       result = FindSymbol(name);
811     }
812   }
813 
814   return result;
815 }
816 
FindFile(const string & key) const817 inline const FileDescriptor* DescriptorPool::Tables::FindFile(
818     const string& key) const {
819   return FindPtrOrNull(files_by_name_, key.c_str());
820 }
821 
FindFieldByNumber(const Descriptor * parent,int number) const822 inline const FieldDescriptor* FileDescriptorTables::FindFieldByNumber(
823     const Descriptor* parent, int number) const {
824   return FindPtrOrNull(fields_by_number_, std::make_pair(parent, number));
825 }
826 
FindFieldByLowercaseName(const void * parent,const string & lowercase_name) const827 inline const FieldDescriptor* FileDescriptorTables::FindFieldByLowercaseName(
828     const void* parent, const string& lowercase_name) const {
829   return FindPtrOrNull(fields_by_lowercase_name_,
830                        PointerStringPair(parent, lowercase_name.c_str()));
831 }
832 
FindFieldByCamelcaseName(const void * parent,const string & camelcase_name) const833 inline const FieldDescriptor* FileDescriptorTables::FindFieldByCamelcaseName(
834     const void* parent, const string& camelcase_name) const {
835   return FindPtrOrNull(fields_by_camelcase_name_,
836                        PointerStringPair(parent, camelcase_name.c_str()));
837 }
838 
FindEnumValueByNumber(const EnumDescriptor * parent,int number) const839 inline const EnumValueDescriptor* FileDescriptorTables::FindEnumValueByNumber(
840     const EnumDescriptor* parent, int number) const {
841   return FindPtrOrNull(enum_values_by_number_, std::make_pair(parent, number));
842 }
843 
844 inline const EnumValueDescriptor*
FindEnumValueByNumberCreatingIfUnknown(const EnumDescriptor * parent,int number) const845 FileDescriptorTables::FindEnumValueByNumberCreatingIfUnknown(
846     const EnumDescriptor* parent, int number) const {
847   // First try, with map of compiled-in values.
848   {
849     const EnumValueDescriptor* desc =
850         FindPtrOrNull(enum_values_by_number_, std::make_pair(parent, number));
851     if (desc != NULL) {
852       return desc;
853     }
854   }
855   // Second try, with reader lock held on unknown enum values: common case.
856   {
857     ReaderMutexLock l(&unknown_enum_values_mu_);
858     const EnumValueDescriptor* desc = FindPtrOrNull(
859         unknown_enum_values_by_number_, std::make_pair(parent, number));
860     if (desc != NULL) {
861       return desc;
862     }
863   }
864   // If not found, try again with writer lock held, and create new descriptor if
865   // necessary.
866   {
867     WriterMutexLock l(&unknown_enum_values_mu_);
868     const EnumValueDescriptor* desc = FindPtrOrNull(
869         unknown_enum_values_by_number_, std::make_pair(parent, number));
870     if (desc != NULL) {
871       return desc;
872     }
873 
874     // Create an EnumValueDescriptor dynamically. We don't insert it into the
875     // EnumDescriptor (it's not a part of the enum as originally defined), but
876     // we do insert it into the table so that we can return the same pointer
877     // later.
878     string enum_value_name = StringPrintf(
879         "UNKNOWN_ENUM_VALUE_%s_%d", parent->name().c_str(), number);
880     DescriptorPool::Tables* tables =
881         const_cast<DescriptorPool::Tables*>(DescriptorPool::generated_pool()->
882                                             tables_.get());
883     EnumValueDescriptor* result = tables->Allocate<EnumValueDescriptor>();
884     result->name_ = tables->AllocateString(enum_value_name);
885     result->full_name_ = tables->AllocateString(parent->full_name() +
886                                                 "." + enum_value_name);
887     result->number_ = number;
888     result->type_ = parent;
889     result->options_ = &EnumValueOptions::default_instance();
890     InsertIfNotPresent(&unknown_enum_values_by_number_,
891                        std::make_pair(parent, number), result);
892     return result;
893   }
894 }
895 
896 
FindExtension(const Descriptor * extendee,int number)897 inline const FieldDescriptor* DescriptorPool::Tables::FindExtension(
898     const Descriptor* extendee, int number) {
899   return FindPtrOrNull(extensions_, std::make_pair(extendee, number));
900 }
901 
FindAllExtensions(const Descriptor * extendee,vector<const FieldDescriptor * > * out) const902 inline void DescriptorPool::Tables::FindAllExtensions(
903     const Descriptor* extendee, vector<const FieldDescriptor*>* out) const {
904   ExtensionsGroupedByDescriptorMap::const_iterator it =
905       extensions_.lower_bound(std::make_pair(extendee, 0));
906   for (; it != extensions_.end() && it->first.first == extendee; ++it) {
907     out->push_back(it->second);
908   }
909 }
910 
911 // -------------------------------------------------------------------
912 
AddSymbol(const string & full_name,Symbol symbol)913 bool DescriptorPool::Tables::AddSymbol(
914     const string& full_name, Symbol symbol) {
915   if (InsertIfNotPresent(&symbols_by_name_, full_name.c_str(), symbol)) {
916     symbols_after_checkpoint_.push_back(full_name.c_str());
917     return true;
918   } else {
919     return false;
920   }
921 }
922 
AddAliasUnderParent(const void * parent,const string & name,Symbol symbol)923 bool FileDescriptorTables::AddAliasUnderParent(
924     const void* parent, const string& name, Symbol symbol) {
925   PointerStringPair by_parent_key(parent, name.c_str());
926   return InsertIfNotPresent(&symbols_by_parent_, by_parent_key, symbol);
927 }
928 
AddFile(const FileDescriptor * file)929 bool DescriptorPool::Tables::AddFile(const FileDescriptor* file) {
930   if (InsertIfNotPresent(&files_by_name_, file->name().c_str(), file)) {
931     files_after_checkpoint_.push_back(file->name().c_str());
932     return true;
933   } else {
934     return false;
935   }
936 }
937 
AddFieldByStylizedNames(const FieldDescriptor * field)938 void FileDescriptorTables::AddFieldByStylizedNames(
939     const FieldDescriptor* field) {
940   const void* parent;
941   if (field->is_extension()) {
942     if (field->extension_scope() == NULL) {
943       parent = field->file();
944     } else {
945       parent = field->extension_scope();
946     }
947   } else {
948     parent = field->containing_type();
949   }
950 
951   PointerStringPair lowercase_key(parent, field->lowercase_name().c_str());
952   InsertIfNotPresent(&fields_by_lowercase_name_, lowercase_key, field);
953 
954   PointerStringPair camelcase_key(parent, field->camelcase_name().c_str());
955   InsertIfNotPresent(&fields_by_camelcase_name_, camelcase_key, field);
956 }
957 
AddFieldByNumber(const FieldDescriptor * field)958 bool FileDescriptorTables::AddFieldByNumber(const FieldDescriptor* field) {
959   DescriptorIntPair key(field->containing_type(), field->number());
960   return InsertIfNotPresent(&fields_by_number_, key, field);
961 }
962 
AddEnumValueByNumber(const EnumValueDescriptor * value)963 bool FileDescriptorTables::AddEnumValueByNumber(
964     const EnumValueDescriptor* value) {
965   EnumIntPair key(value->type(), value->number());
966   return InsertIfNotPresent(&enum_values_by_number_, key, value);
967 }
968 
AddExtension(const FieldDescriptor * field)969 bool DescriptorPool::Tables::AddExtension(const FieldDescriptor* field) {
970   DescriptorIntPair key(field->containing_type(), field->number());
971   if (InsertIfNotPresent(&extensions_, key, field)) {
972     extensions_after_checkpoint_.push_back(key);
973     return true;
974   } else {
975     return false;
976   }
977 }
978 
979 // -------------------------------------------------------------------
980 
981 template<typename Type>
Allocate()982 Type* DescriptorPool::Tables::Allocate() {
983   return reinterpret_cast<Type*>(AllocateBytes(sizeof(Type)));
984 }
985 
986 template<typename Type>
AllocateArray(int count)987 Type* DescriptorPool::Tables::AllocateArray(int count) {
988   return reinterpret_cast<Type*>(AllocateBytes(sizeof(Type) * count));
989 }
990 
AllocateString(const string & value)991 string* DescriptorPool::Tables::AllocateString(const string& value) {
992   string* result = new string(value);
993   strings_.push_back(result);
994   return result;
995 }
996 
997 template<typename Type>
AllocateMessage(Type *)998 Type* DescriptorPool::Tables::AllocateMessage(Type* /* dummy */) {
999   Type* result = new Type;
1000   messages_.push_back(result);
1001   return result;
1002 }
1003 
AllocateFileTables()1004 FileDescriptorTables* DescriptorPool::Tables::AllocateFileTables() {
1005   FileDescriptorTables* result = new FileDescriptorTables;
1006   file_tables_.push_back(result);
1007   return result;
1008 }
1009 
AllocateBytes(int size)1010 void* DescriptorPool::Tables::AllocateBytes(int size) {
1011   // TODO(kenton):  Would it be worthwhile to implement this in some more
1012   // sophisticated way?  Probably not for the open source release, but for
1013   // internal use we could easily plug in one of our existing memory pool
1014   // allocators...
1015   if (size == 0) return NULL;
1016 
1017   void* result = operator new(size);
1018   allocations_.push_back(result);
1019   return result;
1020 }
1021 
BuildLocationsByPath(pair<const FileDescriptorTables *,const SourceCodeInfo * > * p)1022 void FileDescriptorTables::BuildLocationsByPath(
1023     pair<const FileDescriptorTables*, const SourceCodeInfo*>* p) {
1024   for (int i = 0, len = p->second->location_size(); i < len; ++i) {
1025     const SourceCodeInfo_Location* loc = &p->second->location().Get(i);
1026     p->first->locations_by_path_[Join(loc->path(), ",")] = loc;
1027   }
1028 }
1029 
GetSourceLocation(const vector<int> & path,const SourceCodeInfo * info) const1030 const SourceCodeInfo_Location* FileDescriptorTables::GetSourceLocation(
1031     const vector<int>& path, const SourceCodeInfo* info) const {
1032   pair<const FileDescriptorTables*, const SourceCodeInfo*> p(
1033       std::make_pair(this, info));
1034   locations_by_path_once_.Init(&FileDescriptorTables::BuildLocationsByPath, &p);
1035   return FindPtrOrNull(locations_by_path_, Join(path, ","));
1036 }
1037 
1038 // ===================================================================
1039 // DescriptorPool
1040 
~ErrorCollector()1041 DescriptorPool::ErrorCollector::~ErrorCollector() {}
1042 
DescriptorPool()1043 DescriptorPool::DescriptorPool()
1044   : mutex_(NULL),
1045     fallback_database_(NULL),
1046     default_error_collector_(NULL),
1047     underlay_(NULL),
1048     tables_(new Tables),
1049     enforce_dependencies_(true),
1050     allow_unknown_(false),
1051     enforce_weak_(false) {}
1052 
DescriptorPool(DescriptorDatabase * fallback_database,ErrorCollector * error_collector)1053 DescriptorPool::DescriptorPool(DescriptorDatabase* fallback_database,
1054                                ErrorCollector* error_collector)
1055   : mutex_(new Mutex),
1056     fallback_database_(fallback_database),
1057     default_error_collector_(error_collector),
1058     underlay_(NULL),
1059     tables_(new Tables),
1060     enforce_dependencies_(true),
1061     allow_unknown_(false),
1062     enforce_weak_(false) {
1063 }
1064 
DescriptorPool(const DescriptorPool * underlay)1065 DescriptorPool::DescriptorPool(const DescriptorPool* underlay)
1066   : mutex_(NULL),
1067     fallback_database_(NULL),
1068     default_error_collector_(NULL),
1069     underlay_(underlay),
1070     tables_(new Tables),
1071     enforce_dependencies_(true),
1072     allow_unknown_(false),
1073     enforce_weak_(false) {}
1074 
~DescriptorPool()1075 DescriptorPool::~DescriptorPool() {
1076   if (mutex_ != NULL) delete mutex_;
1077 }
1078 
1079 // DescriptorPool::BuildFile() defined later.
1080 // DescriptorPool::BuildFileCollectingErrors() defined later.
1081 
InternalDontEnforceDependencies()1082 void DescriptorPool::InternalDontEnforceDependencies() {
1083   enforce_dependencies_ = false;
1084 }
1085 
AddUnusedImportTrackFile(const string & file_name)1086 void DescriptorPool::AddUnusedImportTrackFile(const string& file_name) {
1087   unused_import_track_files_.insert(file_name);
1088 }
1089 
ClearUnusedImportTrackFiles()1090 void DescriptorPool::ClearUnusedImportTrackFiles() {
1091   unused_import_track_files_.clear();
1092 }
1093 
InternalIsFileLoaded(const string & filename) const1094 bool DescriptorPool::InternalIsFileLoaded(const string& filename) const {
1095   MutexLockMaybe lock(mutex_);
1096   return tables_->FindFile(filename) != NULL;
1097 }
1098 
1099 // generated_pool ====================================================
1100 
1101 namespace {
1102 
1103 
1104 EncodedDescriptorDatabase* generated_database_ = NULL;
1105 DescriptorPool* generated_pool_ = NULL;
1106 GOOGLE_PROTOBUF_DECLARE_ONCE(generated_pool_init_);
1107 
DeleteGeneratedPool()1108 void DeleteGeneratedPool() {
1109   delete generated_database_;
1110   generated_database_ = NULL;
1111   delete generated_pool_;
1112   generated_pool_ = NULL;
1113 }
1114 
InitGeneratedPool()1115 static void InitGeneratedPool() {
1116   generated_database_ = new EncodedDescriptorDatabase;
1117   generated_pool_ = new DescriptorPool(generated_database_);
1118 
1119   internal::OnShutdown(&DeleteGeneratedPool);
1120 }
1121 
InitGeneratedPoolOnce()1122 inline void InitGeneratedPoolOnce() {
1123   ::google::protobuf::GoogleOnceInit(&generated_pool_init_, &InitGeneratedPool);
1124 }
1125 
1126 }  // anonymous namespace
1127 
generated_pool()1128 const DescriptorPool* DescriptorPool::generated_pool() {
1129   InitGeneratedPoolOnce();
1130   return generated_pool_;
1131 }
1132 
1133 
internal_generated_pool()1134 DescriptorPool* DescriptorPool::internal_generated_pool() {
1135   InitGeneratedPoolOnce();
1136   return generated_pool_;
1137 }
1138 
InternalAddGeneratedFile(const void * encoded_file_descriptor,int size)1139 void DescriptorPool::InternalAddGeneratedFile(
1140     const void* encoded_file_descriptor, int size) {
1141   // So, this function is called in the process of initializing the
1142   // descriptors for generated proto classes.  Each generated .pb.cc file
1143   // has an internal procedure called AddDescriptors() which is called at
1144   // process startup, and that function calls this one in order to register
1145   // the raw bytes of the FileDescriptorProto representing the file.
1146   //
1147   // We do not actually construct the descriptor objects right away.  We just
1148   // hang on to the bytes until they are actually needed.  We actually construct
1149   // the descriptor the first time one of the following things happens:
1150   // * Someone calls a method like descriptor(), GetDescriptor(), or
1151   //   GetReflection() on the generated types, which requires returning the
1152   //   descriptor or an object based on it.
1153   // * Someone looks up the descriptor in DescriptorPool::generated_pool().
1154   //
1155   // Once one of these happens, the DescriptorPool actually parses the
1156   // FileDescriptorProto and generates a FileDescriptor (and all its children)
1157   // based on it.
1158   //
1159   // Note that FileDescriptorProto is itself a generated protocol message.
1160   // Therefore, when we parse one, we have to be very careful to avoid using
1161   // any descriptor-based operations, since this might cause infinite recursion
1162   // or deadlock.
1163   InitGeneratedPoolOnce();
1164   GOOGLE_CHECK(generated_database_->Add(encoded_file_descriptor, size));
1165 }
1166 
1167 
1168 // Find*By* methods ==================================================
1169 
1170 // TODO(kenton):  There's a lot of repeated code here, but I'm not sure if
1171 //   there's any good way to factor it out.  Think about this some time when
1172 //   there's nothing more important to do (read: never).
1173 
FindFileByName(const string & name) const1174 const FileDescriptor* DescriptorPool::FindFileByName(const string& name) const {
1175   MutexLockMaybe lock(mutex_);
1176   tables_->known_bad_symbols_.clear();
1177   tables_->known_bad_files_.clear();
1178   const FileDescriptor* result = tables_->FindFile(name);
1179   if (result != NULL) return result;
1180   if (underlay_ != NULL) {
1181     result = underlay_->FindFileByName(name);
1182     if (result != NULL) return result;
1183   }
1184   if (TryFindFileInFallbackDatabase(name)) {
1185     result = tables_->FindFile(name);
1186     if (result != NULL) return result;
1187   }
1188   return NULL;
1189 }
1190 
FindFileContainingSymbol(const string & symbol_name) const1191 const FileDescriptor* DescriptorPool::FindFileContainingSymbol(
1192     const string& symbol_name) const {
1193   MutexLockMaybe lock(mutex_);
1194   tables_->known_bad_symbols_.clear();
1195   tables_->known_bad_files_.clear();
1196   Symbol result = tables_->FindSymbol(symbol_name);
1197   if (!result.IsNull()) return result.GetFile();
1198   if (underlay_ != NULL) {
1199     const FileDescriptor* file_result =
1200       underlay_->FindFileContainingSymbol(symbol_name);
1201     if (file_result != NULL) return file_result;
1202   }
1203   if (TryFindSymbolInFallbackDatabase(symbol_name)) {
1204     result = tables_->FindSymbol(symbol_name);
1205     if (!result.IsNull()) return result.GetFile();
1206   }
1207   return NULL;
1208 }
1209 
FindMessageTypeByName(const string & name) const1210 const Descriptor* DescriptorPool::FindMessageTypeByName(
1211     const string& name) const {
1212   Symbol result = tables_->FindByNameHelper(this, name);
1213   return (result.type == Symbol::MESSAGE) ? result.descriptor : NULL;
1214 }
1215 
FindFieldByName(const string & name) const1216 const FieldDescriptor* DescriptorPool::FindFieldByName(
1217     const string& name) const {
1218   Symbol result = tables_->FindByNameHelper(this, name);
1219   if (result.type == Symbol::FIELD &&
1220       !result.field_descriptor->is_extension()) {
1221     return result.field_descriptor;
1222   } else {
1223     return NULL;
1224   }
1225 }
1226 
FindExtensionByName(const string & name) const1227 const FieldDescriptor* DescriptorPool::FindExtensionByName(
1228     const string& name) const {
1229   Symbol result = tables_->FindByNameHelper(this, name);
1230   if (result.type == Symbol::FIELD &&
1231       result.field_descriptor->is_extension()) {
1232     return result.field_descriptor;
1233   } else {
1234     return NULL;
1235   }
1236 }
1237 
FindOneofByName(const string & name) const1238 const OneofDescriptor* DescriptorPool::FindOneofByName(
1239     const string& name) const {
1240   Symbol result = tables_->FindByNameHelper(this, name);
1241   return (result.type == Symbol::ONEOF) ? result.oneof_descriptor : NULL;
1242 }
1243 
FindEnumTypeByName(const string & name) const1244 const EnumDescriptor* DescriptorPool::FindEnumTypeByName(
1245     const string& name) const {
1246   Symbol result = tables_->FindByNameHelper(this, name);
1247   return (result.type == Symbol::ENUM) ? result.enum_descriptor : NULL;
1248 }
1249 
FindEnumValueByName(const string & name) const1250 const EnumValueDescriptor* DescriptorPool::FindEnumValueByName(
1251     const string& name) const {
1252   Symbol result = tables_->FindByNameHelper(this, name);
1253   return (result.type == Symbol::ENUM_VALUE) ?
1254     result.enum_value_descriptor : NULL;
1255 }
1256 
FindServiceByName(const string & name) const1257 const ServiceDescriptor* DescriptorPool::FindServiceByName(
1258     const string& name) const {
1259   Symbol result = tables_->FindByNameHelper(this, name);
1260   return (result.type == Symbol::SERVICE) ? result.service_descriptor : NULL;
1261 }
1262 
FindMethodByName(const string & name) const1263 const MethodDescriptor* DescriptorPool::FindMethodByName(
1264     const string& name) const {
1265   Symbol result = tables_->FindByNameHelper(this, name);
1266   return (result.type == Symbol::METHOD) ? result.method_descriptor : NULL;
1267 }
1268 
FindExtensionByNumber(const Descriptor * extendee,int number) const1269 const FieldDescriptor* DescriptorPool::FindExtensionByNumber(
1270     const Descriptor* extendee, int number) const {
1271   MutexLockMaybe lock(mutex_);
1272   tables_->known_bad_symbols_.clear();
1273   tables_->known_bad_files_.clear();
1274   const FieldDescriptor* result = tables_->FindExtension(extendee, number);
1275   if (result != NULL) {
1276     return result;
1277   }
1278   if (underlay_ != NULL) {
1279     result = underlay_->FindExtensionByNumber(extendee, number);
1280     if (result != NULL) return result;
1281   }
1282   if (TryFindExtensionInFallbackDatabase(extendee, number)) {
1283     result = tables_->FindExtension(extendee, number);
1284     if (result != NULL) {
1285       return result;
1286     }
1287   }
1288   return NULL;
1289 }
1290 
FindAllExtensions(const Descriptor * extendee,vector<const FieldDescriptor * > * out) const1291 void DescriptorPool::FindAllExtensions(
1292     const Descriptor* extendee, vector<const FieldDescriptor*>* out) const {
1293   MutexLockMaybe lock(mutex_);
1294   tables_->known_bad_symbols_.clear();
1295   tables_->known_bad_files_.clear();
1296 
1297   // Initialize tables_->extensions_ from the fallback database first
1298   // (but do this only once per descriptor).
1299   if (fallback_database_ != NULL &&
1300       tables_->extensions_loaded_from_db_.count(extendee) == 0) {
1301     vector<int> numbers;
1302     if (fallback_database_->FindAllExtensionNumbers(extendee->full_name(),
1303                                                     &numbers)) {
1304       for (int i = 0; i < numbers.size(); ++i) {
1305         int number = numbers[i];
1306         if (tables_->FindExtension(extendee, number) == NULL) {
1307           TryFindExtensionInFallbackDatabase(extendee, number);
1308         }
1309       }
1310       tables_->extensions_loaded_from_db_.insert(extendee);
1311     }
1312   }
1313 
1314   tables_->FindAllExtensions(extendee, out);
1315   if (underlay_ != NULL) {
1316     underlay_->FindAllExtensions(extendee, out);
1317   }
1318 }
1319 
1320 
1321 // -------------------------------------------------------------------
1322 
1323 const FieldDescriptor*
FindFieldByNumber(int key) const1324 Descriptor::FindFieldByNumber(int key) const {
1325   const FieldDescriptor* result =
1326     file()->tables_->FindFieldByNumber(this, key);
1327   if (result == NULL || result->is_extension()) {
1328     return NULL;
1329   } else {
1330     return result;
1331   }
1332 }
1333 
1334 const FieldDescriptor*
FindFieldByLowercaseName(const string & key) const1335 Descriptor::FindFieldByLowercaseName(const string& key) const {
1336   const FieldDescriptor* result =
1337     file()->tables_->FindFieldByLowercaseName(this, key);
1338   if (result == NULL || result->is_extension()) {
1339     return NULL;
1340   } else {
1341     return result;
1342   }
1343 }
1344 
1345 const FieldDescriptor*
FindFieldByCamelcaseName(const string & key) const1346 Descriptor::FindFieldByCamelcaseName(const string& key) const {
1347   const FieldDescriptor* result =
1348     file()->tables_->FindFieldByCamelcaseName(this, key);
1349   if (result == NULL || result->is_extension()) {
1350     return NULL;
1351   } else {
1352     return result;
1353   }
1354 }
1355 
1356 const FieldDescriptor*
FindFieldByName(const string & key) const1357 Descriptor::FindFieldByName(const string& key) const {
1358   Symbol result =
1359     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1360   if (!result.IsNull() && !result.field_descriptor->is_extension()) {
1361     return result.field_descriptor;
1362   } else {
1363     return NULL;
1364   }
1365 }
1366 
1367 const OneofDescriptor*
FindOneofByName(const string & key) const1368 Descriptor::FindOneofByName(const string& key) const {
1369   Symbol result =
1370     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ONEOF);
1371   if (!result.IsNull()) {
1372     return result.oneof_descriptor;
1373   } else {
1374     return NULL;
1375   }
1376 }
1377 
1378 const FieldDescriptor*
FindExtensionByName(const string & key) const1379 Descriptor::FindExtensionByName(const string& key) const {
1380   Symbol result =
1381     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1382   if (!result.IsNull() && result.field_descriptor->is_extension()) {
1383     return result.field_descriptor;
1384   } else {
1385     return NULL;
1386   }
1387 }
1388 
1389 const FieldDescriptor*
FindExtensionByLowercaseName(const string & key) const1390 Descriptor::FindExtensionByLowercaseName(const string& key) const {
1391   const FieldDescriptor* result =
1392     file()->tables_->FindFieldByLowercaseName(this, key);
1393   if (result == NULL || !result->is_extension()) {
1394     return NULL;
1395   } else {
1396     return result;
1397   }
1398 }
1399 
1400 const FieldDescriptor*
FindExtensionByCamelcaseName(const string & key) const1401 Descriptor::FindExtensionByCamelcaseName(const string& key) const {
1402   const FieldDescriptor* result =
1403     file()->tables_->FindFieldByCamelcaseName(this, key);
1404   if (result == NULL || !result->is_extension()) {
1405     return NULL;
1406   } else {
1407     return result;
1408   }
1409 }
1410 
1411 const Descriptor*
FindNestedTypeByName(const string & key) const1412 Descriptor::FindNestedTypeByName(const string& key) const {
1413   Symbol result =
1414     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::MESSAGE);
1415   if (!result.IsNull()) {
1416     return result.descriptor;
1417   } else {
1418     return NULL;
1419   }
1420 }
1421 
1422 const EnumDescriptor*
FindEnumTypeByName(const string & key) const1423 Descriptor::FindEnumTypeByName(const string& key) const {
1424   Symbol result =
1425     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM);
1426   if (!result.IsNull()) {
1427     return result.enum_descriptor;
1428   } else {
1429     return NULL;
1430   }
1431 }
1432 
1433 const EnumValueDescriptor*
FindEnumValueByName(const string & key) const1434 Descriptor::FindEnumValueByName(const string& key) const {
1435   Symbol result =
1436     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1437   if (!result.IsNull()) {
1438     return result.enum_value_descriptor;
1439   } else {
1440     return NULL;
1441   }
1442 }
1443 
1444 const EnumValueDescriptor*
FindValueByName(const string & key) const1445 EnumDescriptor::FindValueByName(const string& key) const {
1446   Symbol result =
1447     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1448   if (!result.IsNull()) {
1449     return result.enum_value_descriptor;
1450   } else {
1451     return NULL;
1452   }
1453 }
1454 
1455 const EnumValueDescriptor*
FindValueByNumber(int key) const1456 EnumDescriptor::FindValueByNumber(int key) const {
1457   return file()->tables_->FindEnumValueByNumber(this, key);
1458 }
1459 
1460 const EnumValueDescriptor*
FindValueByNumberCreatingIfUnknown(int key) const1461 EnumDescriptor::FindValueByNumberCreatingIfUnknown(int key) const {
1462   return file()->tables_->FindEnumValueByNumberCreatingIfUnknown(this, key);
1463 }
1464 
1465 const MethodDescriptor*
FindMethodByName(const string & key) const1466 ServiceDescriptor::FindMethodByName(const string& key) const {
1467   Symbol result =
1468     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::METHOD);
1469   if (!result.IsNull()) {
1470     return result.method_descriptor;
1471   } else {
1472     return NULL;
1473   }
1474 }
1475 
1476 const Descriptor*
FindMessageTypeByName(const string & key) const1477 FileDescriptor::FindMessageTypeByName(const string& key) const {
1478   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::MESSAGE);
1479   if (!result.IsNull()) {
1480     return result.descriptor;
1481   } else {
1482     return NULL;
1483   }
1484 }
1485 
1486 const EnumDescriptor*
FindEnumTypeByName(const string & key) const1487 FileDescriptor::FindEnumTypeByName(const string& key) const {
1488   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM);
1489   if (!result.IsNull()) {
1490     return result.enum_descriptor;
1491   } else {
1492     return NULL;
1493   }
1494 }
1495 
1496 const EnumValueDescriptor*
FindEnumValueByName(const string & key) const1497 FileDescriptor::FindEnumValueByName(const string& key) const {
1498   Symbol result =
1499     tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1500   if (!result.IsNull()) {
1501     return result.enum_value_descriptor;
1502   } else {
1503     return NULL;
1504   }
1505 }
1506 
1507 const ServiceDescriptor*
FindServiceByName(const string & key) const1508 FileDescriptor::FindServiceByName(const string& key) const {
1509   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::SERVICE);
1510   if (!result.IsNull()) {
1511     return result.service_descriptor;
1512   } else {
1513     return NULL;
1514   }
1515 }
1516 
1517 const FieldDescriptor*
FindExtensionByName(const string & key) const1518 FileDescriptor::FindExtensionByName(const string& key) const {
1519   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1520   if (!result.IsNull() && result.field_descriptor->is_extension()) {
1521     return result.field_descriptor;
1522   } else {
1523     return NULL;
1524   }
1525 }
1526 
1527 const FieldDescriptor*
FindExtensionByLowercaseName(const string & key) const1528 FileDescriptor::FindExtensionByLowercaseName(const string& key) const {
1529   const FieldDescriptor* result = tables_->FindFieldByLowercaseName(this, key);
1530   if (result == NULL || !result->is_extension()) {
1531     return NULL;
1532   } else {
1533     return result;
1534   }
1535 }
1536 
1537 const FieldDescriptor*
FindExtensionByCamelcaseName(const string & key) const1538 FileDescriptor::FindExtensionByCamelcaseName(const string& key) const {
1539   const FieldDescriptor* result = tables_->FindFieldByCamelcaseName(this, key);
1540   if (result == NULL || !result->is_extension()) {
1541     return NULL;
1542   } else {
1543     return result;
1544   }
1545 }
1546 
1547 const Descriptor::ExtensionRange*
FindExtensionRangeContainingNumber(int number) const1548 Descriptor::FindExtensionRangeContainingNumber(int number) const {
1549   // Linear search should be fine because we don't expect a message to have
1550   // more than a couple extension ranges.
1551   for (int i = 0; i < extension_range_count(); i++) {
1552     if (number >= extension_range(i)->start &&
1553         number <  extension_range(i)->end) {
1554       return extension_range(i);
1555     }
1556   }
1557   return NULL;
1558 }
1559 
1560 const Descriptor::ReservedRange*
FindReservedRangeContainingNumber(int number) const1561 Descriptor::FindReservedRangeContainingNumber(int number) const {
1562   // TODO(chrisn): Consider a non-linear search.
1563   for (int i = 0; i < reserved_range_count(); i++) {
1564     if (number >= reserved_range(i)->start &&
1565         number <  reserved_range(i)->end) {
1566       return reserved_range(i);
1567     }
1568   }
1569   return NULL;
1570 }
1571 
1572 // -------------------------------------------------------------------
1573 
TryFindFileInFallbackDatabase(const string & name) const1574 bool DescriptorPool::TryFindFileInFallbackDatabase(const string& name) const {
1575   if (fallback_database_ == NULL) return false;
1576 
1577   if (tables_->known_bad_files_.count(name) > 0) return false;
1578 
1579   FileDescriptorProto file_proto;
1580   if (!fallback_database_->FindFileByName(name, &file_proto) ||
1581       BuildFileFromDatabase(file_proto) == NULL) {
1582     tables_->known_bad_files_.insert(name);
1583     return false;
1584   }
1585   return true;
1586 }
1587 
IsSubSymbolOfBuiltType(const string & name) const1588 bool DescriptorPool::IsSubSymbolOfBuiltType(const string& name) const {
1589   string prefix = name;
1590   for (;;) {
1591     string::size_type dot_pos = prefix.find_last_of('.');
1592     if (dot_pos == string::npos) {
1593       break;
1594     }
1595     prefix = prefix.substr(0, dot_pos);
1596     Symbol symbol = tables_->FindSymbol(prefix);
1597     // If the symbol type is anything other than PACKAGE, then its complete
1598     // definition is already known.
1599     if (!symbol.IsNull() && symbol.type != Symbol::PACKAGE) {
1600       return true;
1601     }
1602   }
1603   if (underlay_ != NULL) {
1604     // Check to see if any prefix of this symbol exists in the underlay.
1605     return underlay_->IsSubSymbolOfBuiltType(name);
1606   }
1607   return false;
1608 }
1609 
TryFindSymbolInFallbackDatabase(const string & name) const1610 bool DescriptorPool::TryFindSymbolInFallbackDatabase(const string& name) const {
1611   if (fallback_database_ == NULL) return false;
1612 
1613   if (tables_->known_bad_symbols_.count(name) > 0) return false;
1614 
1615   FileDescriptorProto file_proto;
1616   if (// We skip looking in the fallback database if the name is a sub-symbol
1617       // of any descriptor that already exists in the descriptor pool (except
1618       // for package descriptors).  This is valid because all symbols except
1619       // for packages are defined in a single file, so if the symbol exists
1620       // then we should already have its definition.
1621       //
1622       // The other reason to do this is to support "overriding" type
1623       // definitions by merging two databases that define the same type.  (Yes,
1624       // people do this.)  The main difficulty with making this work is that
1625       // FindFileContainingSymbol() is allowed to return both false positives
1626       // (e.g., SimpleDescriptorDatabase, UpgradedDescriptorDatabase) and false
1627       // negatives (e.g. ProtoFileParser, SourceTreeDescriptorDatabase).
1628       // When two such databases are merged, looking up a non-existent
1629       // sub-symbol of a type that already exists in the descriptor pool can
1630       // result in an attempt to load multiple definitions of the same type.
1631       // The check below avoids this.
1632       IsSubSymbolOfBuiltType(name)
1633 
1634       // Look up file containing this symbol in fallback database.
1635       || !fallback_database_->FindFileContainingSymbol(name, &file_proto)
1636 
1637       // Check if we've already built this file. If so, it apparently doesn't
1638       // contain the symbol we're looking for.  Some DescriptorDatabases
1639       // return false positives.
1640       || tables_->FindFile(file_proto.name()) != NULL
1641 
1642       // Build the file.
1643       || BuildFileFromDatabase(file_proto) == NULL) {
1644     tables_->known_bad_symbols_.insert(name);
1645     return false;
1646   }
1647 
1648   return true;
1649 }
1650 
TryFindExtensionInFallbackDatabase(const Descriptor * containing_type,int field_number) const1651 bool DescriptorPool::TryFindExtensionInFallbackDatabase(
1652     const Descriptor* containing_type, int field_number) const {
1653   if (fallback_database_ == NULL) return false;
1654 
1655   FileDescriptorProto file_proto;
1656   if (!fallback_database_->FindFileContainingExtension(
1657         containing_type->full_name(), field_number, &file_proto)) {
1658     return false;
1659   }
1660 
1661   if (tables_->FindFile(file_proto.name()) != NULL) {
1662     // We've already loaded this file, and it apparently doesn't contain the
1663     // extension we're looking for.  Some DescriptorDatabases return false
1664     // positives.
1665     return false;
1666   }
1667 
1668   if (BuildFileFromDatabase(file_proto) == NULL) {
1669     return false;
1670   }
1671 
1672   return true;
1673 }
1674 
1675 // ===================================================================
1676 
is_map() const1677 bool FieldDescriptor::is_map() const {
1678   return type() == TYPE_MESSAGE && message_type()->options().map_entry();
1679 }
1680 
DefaultValueAsString(bool quote_string_type) const1681 string FieldDescriptor::DefaultValueAsString(bool quote_string_type) const {
1682   GOOGLE_CHECK(has_default_value()) << "No default value";
1683   switch (cpp_type()) {
1684     case CPPTYPE_INT32:
1685       return SimpleItoa(default_value_int32());
1686       break;
1687     case CPPTYPE_INT64:
1688       return SimpleItoa(default_value_int64());
1689       break;
1690     case CPPTYPE_UINT32:
1691       return SimpleItoa(default_value_uint32());
1692       break;
1693     case CPPTYPE_UINT64:
1694       return SimpleItoa(default_value_uint64());
1695       break;
1696     case CPPTYPE_FLOAT:
1697       return SimpleFtoa(default_value_float());
1698       break;
1699     case CPPTYPE_DOUBLE:
1700       return SimpleDtoa(default_value_double());
1701       break;
1702     case CPPTYPE_BOOL:
1703       return default_value_bool() ? "true" : "false";
1704       break;
1705     case CPPTYPE_STRING:
1706       if (quote_string_type) {
1707         return "\"" + CEscape(default_value_string()) + "\"";
1708       } else {
1709         if (type() == TYPE_BYTES) {
1710           return CEscape(default_value_string());
1711         } else {
1712           return default_value_string();
1713         }
1714       }
1715       break;
1716     case CPPTYPE_ENUM:
1717       return default_value_enum()->name();
1718       break;
1719     case CPPTYPE_MESSAGE:
1720       GOOGLE_LOG(DFATAL) << "Messages can't have default values!";
1721       break;
1722   }
1723   GOOGLE_LOG(FATAL) << "Can't get here: failed to get default value as string";
1724   return "";
1725 }
1726 
1727 // CopyTo methods ====================================================
1728 
CopyTo(FileDescriptorProto * proto) const1729 void FileDescriptor::CopyTo(FileDescriptorProto* proto) const {
1730   proto->set_name(name());
1731   if (!package().empty()) proto->set_package(package());
1732   // TODO(liujisi): Also populate when syntax="proto2".
1733   if (syntax() == SYNTAX_PROTO3) proto->set_syntax(SyntaxName(syntax()));
1734 
1735   for (int i = 0; i < dependency_count(); i++) {
1736     proto->add_dependency(dependency(i)->name());
1737   }
1738 
1739   for (int i = 0; i < public_dependency_count(); i++) {
1740     proto->add_public_dependency(public_dependencies_[i]);
1741   }
1742 
1743   for (int i = 0; i < weak_dependency_count(); i++) {
1744     proto->add_weak_dependency(weak_dependencies_[i]);
1745   }
1746 
1747   for (int i = 0; i < message_type_count(); i++) {
1748     message_type(i)->CopyTo(proto->add_message_type());
1749   }
1750   for (int i = 0; i < enum_type_count(); i++) {
1751     enum_type(i)->CopyTo(proto->add_enum_type());
1752   }
1753   for (int i = 0; i < service_count(); i++) {
1754     service(i)->CopyTo(proto->add_service());
1755   }
1756   for (int i = 0; i < extension_count(); i++) {
1757     extension(i)->CopyTo(proto->add_extension());
1758   }
1759 
1760   if (&options() != &FileOptions::default_instance()) {
1761     proto->mutable_options()->CopyFrom(options());
1762   }
1763 }
1764 
CopyJsonNameTo(FileDescriptorProto * proto) const1765 void FileDescriptor::CopyJsonNameTo(FileDescriptorProto* proto) const {
1766   if (message_type_count() != proto->message_type_size() ||
1767       extension_count() != proto->extension_size()) {
1768     GOOGLE_LOG(ERROR) << "Cannot copy json_name to a proto of a different size.";
1769     return;
1770   }
1771   for (int i = 0; i < message_type_count(); i++) {
1772     message_type(i)->CopyJsonNameTo(proto->mutable_message_type(i));
1773   }
1774   for (int i = 0; i < extension_count(); i++) {
1775     extension(i)->CopyJsonNameTo(proto->mutable_extension(i));
1776   }
1777 }
1778 
CopySourceCodeInfoTo(FileDescriptorProto * proto) const1779 void FileDescriptor::CopySourceCodeInfoTo(FileDescriptorProto* proto) const {
1780   if (source_code_info_ &&
1781       source_code_info_ != &SourceCodeInfo::default_instance()) {
1782     proto->mutable_source_code_info()->CopyFrom(*source_code_info_);
1783   }
1784 }
1785 
CopyTo(DescriptorProto * proto) const1786 void Descriptor::CopyTo(DescriptorProto* proto) const {
1787   proto->set_name(name());
1788 
1789   for (int i = 0; i < field_count(); i++) {
1790     field(i)->CopyTo(proto->add_field());
1791   }
1792   for (int i = 0; i < oneof_decl_count(); i++) {
1793     oneof_decl(i)->CopyTo(proto->add_oneof_decl());
1794   }
1795   for (int i = 0; i < nested_type_count(); i++) {
1796     nested_type(i)->CopyTo(proto->add_nested_type());
1797   }
1798   for (int i = 0; i < enum_type_count(); i++) {
1799     enum_type(i)->CopyTo(proto->add_enum_type());
1800   }
1801   for (int i = 0; i < extension_range_count(); i++) {
1802     DescriptorProto::ExtensionRange* range = proto->add_extension_range();
1803     range->set_start(extension_range(i)->start);
1804     range->set_end(extension_range(i)->end);
1805   }
1806   for (int i = 0; i < extension_count(); i++) {
1807     extension(i)->CopyTo(proto->add_extension());
1808   }
1809   for (int i = 0; i < reserved_range_count(); i++) {
1810     DescriptorProto::ReservedRange* range = proto->add_reserved_range();
1811     range->set_start(reserved_range(i)->start);
1812     range->set_end(reserved_range(i)->end);
1813   }
1814   for (int i = 0; i < reserved_name_count(); i++) {
1815     proto->add_reserved_name(reserved_name(i));
1816   }
1817 
1818   if (&options() != &MessageOptions::default_instance()) {
1819     proto->mutable_options()->CopyFrom(options());
1820   }
1821 }
1822 
CopyJsonNameTo(DescriptorProto * proto) const1823 void Descriptor::CopyJsonNameTo(DescriptorProto* proto) const {
1824   if (field_count() != proto->field_size() ||
1825       nested_type_count() != proto->nested_type_size() ||
1826       extension_count() != proto->extension_size()) {
1827     GOOGLE_LOG(ERROR) << "Cannot copy json_name to a proto of a different size.";
1828     return;
1829   }
1830   for (int i = 0; i < field_count(); i++) {
1831     field(i)->CopyJsonNameTo(proto->mutable_field(i));
1832   }
1833   for (int i = 0; i < nested_type_count(); i++) {
1834     nested_type(i)->CopyJsonNameTo(proto->mutable_nested_type(i));
1835   }
1836   for (int i = 0; i < extension_count(); i++) {
1837     extension(i)->CopyJsonNameTo(proto->mutable_extension(i));
1838   }
1839 }
1840 
CopyTo(FieldDescriptorProto * proto) const1841 void FieldDescriptor::CopyTo(FieldDescriptorProto* proto) const {
1842   proto->set_name(name());
1843   proto->set_number(number());
1844   if (has_json_name_) {
1845     proto->set_json_name(json_name());
1846   }
1847 
1848   // Some compilers do not allow static_cast directly between two enum types,
1849   // so we must cast to int first.
1850   proto->set_label(static_cast<FieldDescriptorProto::Label>(
1851                      implicit_cast<int>(label())));
1852   proto->set_type(static_cast<FieldDescriptorProto::Type>(
1853                     implicit_cast<int>(type())));
1854 
1855   if (is_extension()) {
1856     if (!containing_type()->is_unqualified_placeholder_) {
1857       proto->set_extendee(".");
1858     }
1859     proto->mutable_extendee()->append(containing_type()->full_name());
1860   }
1861 
1862   if (cpp_type() == CPPTYPE_MESSAGE) {
1863     if (message_type()->is_placeholder_) {
1864       // We don't actually know if the type is a message type.  It could be
1865       // an enum.
1866       proto->clear_type();
1867     }
1868 
1869     if (!message_type()->is_unqualified_placeholder_) {
1870       proto->set_type_name(".");
1871     }
1872     proto->mutable_type_name()->append(message_type()->full_name());
1873   } else if (cpp_type() == CPPTYPE_ENUM) {
1874     if (!enum_type()->is_unqualified_placeholder_) {
1875       proto->set_type_name(".");
1876     }
1877     proto->mutable_type_name()->append(enum_type()->full_name());
1878   }
1879 
1880   if (has_default_value()) {
1881     proto->set_default_value(DefaultValueAsString(false));
1882   }
1883 
1884   if (containing_oneof() != NULL && !is_extension()) {
1885     proto->set_oneof_index(containing_oneof()->index());
1886   }
1887 
1888   if (&options() != &FieldOptions::default_instance()) {
1889     proto->mutable_options()->CopyFrom(options());
1890   }
1891 }
1892 
CopyJsonNameTo(FieldDescriptorProto * proto) const1893 void FieldDescriptor::CopyJsonNameTo(FieldDescriptorProto* proto) const {
1894   proto->set_json_name(json_name());
1895 }
1896 
CopyTo(OneofDescriptorProto * proto) const1897 void OneofDescriptor::CopyTo(OneofDescriptorProto* proto) const {
1898   proto->set_name(name());
1899   if (&options() != &OneofOptions::default_instance()) {
1900     proto->mutable_options()->CopyFrom(options());
1901   }
1902 }
1903 
CopyTo(EnumDescriptorProto * proto) const1904 void EnumDescriptor::CopyTo(EnumDescriptorProto* proto) const {
1905   proto->set_name(name());
1906 
1907   for (int i = 0; i < value_count(); i++) {
1908     value(i)->CopyTo(proto->add_value());
1909   }
1910 
1911   if (&options() != &EnumOptions::default_instance()) {
1912     proto->mutable_options()->CopyFrom(options());
1913   }
1914 }
1915 
CopyTo(EnumValueDescriptorProto * proto) const1916 void EnumValueDescriptor::CopyTo(EnumValueDescriptorProto* proto) const {
1917   proto->set_name(name());
1918   proto->set_number(number());
1919 
1920   if (&options() != &EnumValueOptions::default_instance()) {
1921     proto->mutable_options()->CopyFrom(options());
1922   }
1923 }
1924 
CopyTo(ServiceDescriptorProto * proto) const1925 void ServiceDescriptor::CopyTo(ServiceDescriptorProto* proto) const {
1926   proto->set_name(name());
1927 
1928   for (int i = 0; i < method_count(); i++) {
1929     method(i)->CopyTo(proto->add_method());
1930   }
1931 
1932   if (&options() != &ServiceOptions::default_instance()) {
1933     proto->mutable_options()->CopyFrom(options());
1934   }
1935 }
1936 
CopyTo(MethodDescriptorProto * proto) const1937 void MethodDescriptor::CopyTo(MethodDescriptorProto* proto) const {
1938   proto->set_name(name());
1939 
1940   if (!input_type()->is_unqualified_placeholder_) {
1941     proto->set_input_type(".");
1942   }
1943   proto->mutable_input_type()->append(input_type()->full_name());
1944 
1945   if (!output_type()->is_unqualified_placeholder_) {
1946     proto->set_output_type(".");
1947   }
1948   proto->mutable_output_type()->append(output_type()->full_name());
1949 
1950   if (&options() != &MethodOptions::default_instance()) {
1951     proto->mutable_options()->CopyFrom(options());
1952   }
1953 
1954   if (client_streaming_) {
1955     proto->set_client_streaming(true);
1956   }
1957   if (server_streaming_) {
1958     proto->set_server_streaming(true);
1959   }
1960 }
1961 
1962 // DebugString methods ===============================================
1963 
1964 namespace {
1965 
1966 // Used by each of the option formatters.
RetrieveOptions(int depth,const Message & options,vector<string> * option_entries)1967 bool RetrieveOptions(int depth,
1968                      const Message &options,
1969                      vector<string> *option_entries) {
1970   option_entries->clear();
1971   const Reflection* reflection = options.GetReflection();
1972   vector<const FieldDescriptor*> fields;
1973   reflection->ListFields(options, &fields);
1974   for (int i = 0; i < fields.size(); i++) {
1975     int count = 1;
1976     bool repeated = false;
1977     if (fields[i]->is_repeated()) {
1978       count = reflection->FieldSize(options, fields[i]);
1979       repeated = true;
1980     }
1981     for (int j = 0; j < count; j++) {
1982       string fieldval;
1983       if (fields[i]->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
1984         string tmp;
1985         TextFormat::Printer printer;
1986         printer.SetInitialIndentLevel(depth + 1);
1987         printer.PrintFieldValueToString(options, fields[i],
1988                                         repeated ? j : -1, &tmp);
1989         fieldval.append("{\n");
1990         fieldval.append(tmp);
1991         fieldval.append(depth * 2, ' ');
1992         fieldval.append("}");
1993       } else {
1994         TextFormat::PrintFieldValueToString(options, fields[i],
1995                                             repeated ? j : -1, &fieldval);
1996       }
1997       string name;
1998       if (fields[i]->is_extension()) {
1999         name = "(." + fields[i]->full_name() + ")";
2000       } else {
2001         name = fields[i]->name();
2002       }
2003       option_entries->push_back(name + " = " + fieldval);
2004     }
2005   }
2006   return !option_entries->empty();
2007 }
2008 
2009 // Formats options that all appear together in brackets. Does not include
2010 // brackets.
FormatBracketedOptions(int depth,const Message & options,string * output)2011 bool FormatBracketedOptions(int depth, const Message &options, string *output) {
2012   vector<string> all_options;
2013   if (RetrieveOptions(depth, options, &all_options)) {
2014     output->append(Join(all_options, ", "));
2015   }
2016   return !all_options.empty();
2017 }
2018 
2019 // Formats options one per line
FormatLineOptions(int depth,const Message & options,string * output)2020 bool FormatLineOptions(int depth, const Message &options, string *output) {
2021   string prefix(depth * 2, ' ');
2022   vector<string> all_options;
2023   if (RetrieveOptions(depth, options, &all_options)) {
2024     for (int i = 0; i < all_options.size(); i++) {
2025       strings::SubstituteAndAppend(output, "$0option $1;\n",
2026                                    prefix, all_options[i]);
2027     }
2028   }
2029   return !all_options.empty();
2030 }
2031 
2032 class SourceLocationCommentPrinter {
2033  public:
2034   template<typename DescType>
SourceLocationCommentPrinter(const DescType * desc,const string & prefix,const DebugStringOptions & options)2035   SourceLocationCommentPrinter(const DescType* desc,
2036                                const string& prefix,
2037                                const DebugStringOptions& options)
2038       : options_(options), prefix_(prefix) {
2039     // Perform the SourceLocation lookup only if we're including user comments,
2040     // because the lookup is fairly expensive.
2041     have_source_loc_ = options.include_comments &&
2042         desc->GetSourceLocation(&source_loc_);
2043   }
SourceLocationCommentPrinter(const FileDescriptor * file,const vector<int> & path,const string & prefix,const DebugStringOptions & options)2044   SourceLocationCommentPrinter(const FileDescriptor* file,
2045                                const vector<int>& path,
2046                                const string& prefix,
2047                                const DebugStringOptions& options)
2048       : options_(options), prefix_(prefix) {
2049     // Perform the SourceLocation lookup only if we're including user comments,
2050     // because the lookup is fairly expensive.
2051     have_source_loc_ = options.include_comments &&
2052         file->GetSourceLocation(path, &source_loc_);
2053   }
AddPreComment(string * output)2054   void AddPreComment(string* output) {
2055     if (have_source_loc_) {
2056       // Detached leading comments.
2057       for (int i = 0 ; i < source_loc_.leading_detached_comments.size(); ++i) {
2058         *output += FormatComment(source_loc_.leading_detached_comments[i]);
2059         *output += "\n";
2060       }
2061       // Attached leading comments.
2062       if (!source_loc_.leading_comments.empty()) {
2063         *output += FormatComment(source_loc_.leading_comments);
2064       }
2065     }
2066   }
AddPostComment(string * output)2067   void AddPostComment(string* output) {
2068     if (have_source_loc_ && source_loc_.trailing_comments.size() > 0) {
2069       *output += FormatComment(source_loc_.trailing_comments);
2070     }
2071   }
2072 
2073   // Format comment such that each line becomes a full-line C++-style comment in
2074   // the DebugString() output.
FormatComment(const string & comment_text)2075   string FormatComment(const string& comment_text) {
2076     string stripped_comment = comment_text;
2077     StripWhitespace(&stripped_comment);
2078     vector<string> lines = Split(stripped_comment, "\n");
2079     string output;
2080     for (int i = 0; i < lines.size(); ++i) {
2081       const string& line = lines[i];
2082       strings::SubstituteAndAppend(&output, "$0// $1\n", prefix_, line);
2083     }
2084     return output;
2085   }
2086 
2087  private:
2088 
2089   bool have_source_loc_;
2090   SourceLocation source_loc_;
2091   DebugStringOptions options_;
2092   string prefix_;
2093 };
2094 
2095 }  // anonymous namespace
2096 
DebugString() const2097 string FileDescriptor::DebugString() const {
2098   DebugStringOptions options;  // default options
2099   return DebugStringWithOptions(options);
2100 }
2101 
DebugStringWithOptions(const DebugStringOptions & debug_string_options) const2102 string FileDescriptor::DebugStringWithOptions(
2103     const DebugStringOptions& debug_string_options) const {
2104   string contents;
2105   {
2106     vector<int> path;
2107     path.push_back(FileDescriptorProto::kSyntaxFieldNumber);
2108     SourceLocationCommentPrinter syntax_comment(
2109         this, path, "", debug_string_options);
2110     syntax_comment.AddPreComment(&contents);
2111     strings::SubstituteAndAppend(&contents, "syntax = \"$0\";\n\n",
2112                                  SyntaxName(syntax()));
2113     syntax_comment.AddPostComment(&contents);
2114   }
2115 
2116   SourceLocationCommentPrinter
2117       comment_printer(this, "", debug_string_options);
2118   comment_printer.AddPreComment(&contents);
2119 
2120   set<int> public_dependencies;
2121   set<int> weak_dependencies;
2122   public_dependencies.insert(public_dependencies_,
2123                              public_dependencies_ + public_dependency_count_);
2124   weak_dependencies.insert(weak_dependencies_,
2125                            weak_dependencies_ + weak_dependency_count_);
2126 
2127   for (int i = 0; i < dependency_count(); i++) {
2128     if (public_dependencies.count(i) > 0) {
2129       strings::SubstituteAndAppend(&contents, "import public \"$0\";\n",
2130                                    dependency(i)->name());
2131     } else if (weak_dependencies.count(i) > 0) {
2132       strings::SubstituteAndAppend(&contents, "import weak \"$0\";\n",
2133                                    dependency(i)->name());
2134     } else {
2135       strings::SubstituteAndAppend(&contents, "import \"$0\";\n",
2136                                    dependency(i)->name());
2137     }
2138   }
2139 
2140   if (!package().empty()) {
2141     vector<int> path;
2142     path.push_back(FileDescriptorProto::kPackageFieldNumber);
2143     SourceLocationCommentPrinter package_comment(
2144         this, path, "", debug_string_options);
2145     package_comment.AddPreComment(&contents);
2146     strings::SubstituteAndAppend(&contents, "package $0;\n\n", package());
2147     package_comment.AddPostComment(&contents);
2148   }
2149 
2150   if (FormatLineOptions(0, options(), &contents)) {
2151     contents.append("\n");  // add some space if we had options
2152   }
2153 
2154   for (int i = 0; i < enum_type_count(); i++) {
2155     enum_type(i)->DebugString(0, &contents, debug_string_options);
2156     contents.append("\n");
2157   }
2158 
2159   // Find all the 'group' type extensions; we will not output their nested
2160   // definitions (those will be done with their group field descriptor).
2161   set<const Descriptor*> groups;
2162   for (int i = 0; i < extension_count(); i++) {
2163     if (extension(i)->type() == FieldDescriptor::TYPE_GROUP) {
2164       groups.insert(extension(i)->message_type());
2165     }
2166   }
2167 
2168   for (int i = 0; i < message_type_count(); i++) {
2169     if (groups.count(message_type(i)) == 0) {
2170       message_type(i)->DebugString(0, &contents, debug_string_options,
2171                                    /* include_opening_clause */ true);
2172       contents.append("\n");
2173     }
2174   }
2175 
2176   for (int i = 0; i < service_count(); i++) {
2177     service(i)->DebugString(&contents, debug_string_options);
2178     contents.append("\n");
2179   }
2180 
2181   const Descriptor* containing_type = NULL;
2182   for (int i = 0; i < extension_count(); i++) {
2183     if (extension(i)->containing_type() != containing_type) {
2184       if (i > 0) contents.append("}\n\n");
2185       containing_type = extension(i)->containing_type();
2186       strings::SubstituteAndAppend(&contents, "extend .$0 {\n",
2187                                    containing_type->full_name());
2188     }
2189     extension(i)->DebugString(1, FieldDescriptor::PRINT_LABEL, &contents,
2190                               debug_string_options);
2191   }
2192   if (extension_count() > 0) contents.append("}\n\n");
2193 
2194   comment_printer.AddPostComment(&contents);
2195 
2196   return contents;
2197 }
2198 
DebugString() const2199 string Descriptor::DebugString() const {
2200   DebugStringOptions options;  // default options
2201   return DebugStringWithOptions(options);
2202 }
2203 
DebugStringWithOptions(const DebugStringOptions & options) const2204 string Descriptor::DebugStringWithOptions(
2205     const DebugStringOptions& options) const {
2206   string contents;
2207   DebugString(0, &contents, options, /* include_opening_clause */ true);
2208   return contents;
2209 }
2210 
DebugString(int depth,string * contents,const DebugStringOptions & debug_string_options,bool include_opening_clause) const2211 void Descriptor::DebugString(int depth, string *contents,
2212                              const DebugStringOptions&
2213                              debug_string_options,
2214                              bool include_opening_clause) const {
2215   if (options().map_entry()) {
2216     // Do not generate debug string for auto-generated map-entry type.
2217     return;
2218   }
2219   string prefix(depth * 2, ' ');
2220   ++depth;
2221 
2222   SourceLocationCommentPrinter
2223       comment_printer(this, prefix, debug_string_options);
2224   comment_printer.AddPreComment(contents);
2225 
2226   if (include_opening_clause) {
2227     strings::SubstituteAndAppend(contents, "$0message $1", prefix, name());
2228   }
2229   contents->append(" {\n");
2230 
2231   FormatLineOptions(depth, options(), contents);
2232 
2233   // Find all the 'group' types for fields and extensions; we will not output
2234   // their nested definitions (those will be done with their group field
2235   // descriptor).
2236   set<const Descriptor*> groups;
2237   for (int i = 0; i < field_count(); i++) {
2238     if (field(i)->type() == FieldDescriptor::TYPE_GROUP) {
2239       groups.insert(field(i)->message_type());
2240     }
2241   }
2242   for (int i = 0; i < extension_count(); i++) {
2243     if (extension(i)->type() == FieldDescriptor::TYPE_GROUP) {
2244       groups.insert(extension(i)->message_type());
2245     }
2246   }
2247 
2248   for (int i = 0; i < nested_type_count(); i++) {
2249     if (groups.count(nested_type(i)) == 0) {
2250       nested_type(i)->DebugString(depth, contents, debug_string_options,
2251                                   /* include_opening_clause */ true);
2252     }
2253   }
2254   for (int i = 0; i < enum_type_count(); i++) {
2255     enum_type(i)->DebugString(depth, contents, debug_string_options);
2256   }
2257   for (int i = 0; i < field_count(); i++) {
2258     if (field(i)->containing_oneof() == NULL) {
2259       field(i)->DebugString(depth, FieldDescriptor::PRINT_LABEL, contents,
2260                             debug_string_options);
2261     } else if (field(i)->containing_oneof()->field(0) == field(i)) {
2262       // This is the first field in this oneof, so print the whole oneof.
2263       field(i)->containing_oneof()->DebugString(depth, contents,
2264                                                 debug_string_options);
2265     }
2266   }
2267 
2268   for (int i = 0; i < extension_range_count(); i++) {
2269     strings::SubstituteAndAppend(contents, "$0  extensions $1 to $2;\n",
2270                                  prefix,
2271                                  extension_range(i)->start,
2272                                  extension_range(i)->end - 1);
2273   }
2274 
2275   // Group extensions by what they extend, so they can be printed out together.
2276   const Descriptor* containing_type = NULL;
2277   for (int i = 0; i < extension_count(); i++) {
2278     if (extension(i)->containing_type() != containing_type) {
2279       if (i > 0) strings::SubstituteAndAppend(contents, "$0  }\n", prefix);
2280       containing_type = extension(i)->containing_type();
2281       strings::SubstituteAndAppend(contents, "$0  extend .$1 {\n",
2282                                    prefix, containing_type->full_name());
2283     }
2284     extension(i)->DebugString(
2285         depth + 1, FieldDescriptor::PRINT_LABEL, contents,
2286         debug_string_options);
2287   }
2288   if (extension_count() > 0)
2289     strings::SubstituteAndAppend(contents, "$0  }\n", prefix);
2290 
2291   if (reserved_range_count() > 0) {
2292     strings::SubstituteAndAppend(contents, "$0  reserved ", prefix);
2293     for (int i = 0; i < reserved_range_count(); i++) {
2294       const Descriptor::ReservedRange* range = reserved_range(i);
2295       if (range->end == range->start + 1) {
2296         strings::SubstituteAndAppend(contents, "$0, ", range->start);
2297       } else {
2298         strings::SubstituteAndAppend(contents, "$0 to $1, ",
2299                                      range->start, range->end - 1);
2300       }
2301     }
2302     contents->replace(contents->size() - 2, 2, ";\n");
2303   }
2304 
2305   if (reserved_name_count() > 0) {
2306     strings::SubstituteAndAppend(contents, "$0  reserved ", prefix);
2307     for (int i = 0; i < reserved_name_count(); i++) {
2308       strings::SubstituteAndAppend(contents, "\"$0\", ",
2309                                    CEscape(reserved_name(i)));
2310     }
2311     contents->replace(contents->size() - 2, 2, ";\n");
2312   }
2313 
2314   strings::SubstituteAndAppend(contents, "$0}\n", prefix);
2315   comment_printer.AddPostComment(contents);
2316 }
2317 
DebugString() const2318 string FieldDescriptor::DebugString() const {
2319   DebugStringOptions options;  // default options
2320   return DebugStringWithOptions(options);
2321 }
2322 
DebugStringWithOptions(const DebugStringOptions & debug_string_options) const2323 string FieldDescriptor::DebugStringWithOptions(
2324     const DebugStringOptions& debug_string_options) const {
2325   string contents;
2326   int depth = 0;
2327   if (is_extension()) {
2328     strings::SubstituteAndAppend(&contents, "extend .$0 {\n",
2329                                  containing_type()->full_name());
2330     depth = 1;
2331   }
2332   DebugString(depth, PRINT_LABEL, &contents, debug_string_options);
2333   if (is_extension()) {
2334     contents.append("}\n");
2335   }
2336   return contents;
2337 }
2338 
2339 // The field type string used in FieldDescriptor::DebugString()
FieldTypeNameDebugString() const2340 string FieldDescriptor::FieldTypeNameDebugString() const {
2341   switch(type()) {
2342     case TYPE_MESSAGE:
2343       return "." + message_type()->full_name();
2344     case TYPE_ENUM:
2345       return "." + enum_type()->full_name();
2346     default:
2347       return kTypeToName[type()];
2348   }
2349 }
2350 
DebugString(int depth,PrintLabelFlag print_label_flag,string * contents,const DebugStringOptions & debug_string_options) const2351 void FieldDescriptor::DebugString(int depth,
2352                                   PrintLabelFlag print_label_flag,
2353                                   string *contents,
2354                                   const DebugStringOptions&
2355                                   debug_string_options) const {
2356   string prefix(depth * 2, ' ');
2357   string field_type;
2358 
2359   // Special case map fields.
2360   if (is_map()) {
2361     strings::SubstituteAndAppend(
2362         &field_type, "map<$0, $1>",
2363         message_type()->field(0)->FieldTypeNameDebugString(),
2364         message_type()->field(1)->FieldTypeNameDebugString());
2365   } else {
2366     field_type = FieldTypeNameDebugString();
2367   }
2368 
2369   string label;
2370   if (print_label_flag == PRINT_LABEL && !is_map()) {
2371     label = kLabelToName[this->label()];
2372     label.push_back(' ');
2373   }
2374 
2375   SourceLocationCommentPrinter
2376       comment_printer(this, prefix, debug_string_options);
2377   comment_printer.AddPreComment(contents);
2378 
2379   strings::SubstituteAndAppend(contents, "$0$1$2 $3 = $4",
2380                                prefix,
2381                                label,
2382                                field_type,
2383                                type() == TYPE_GROUP ? message_type()->name() :
2384                                                       name(),
2385                                number());
2386 
2387   bool bracketed = false;
2388   if (has_default_value()) {
2389     bracketed = true;
2390     strings::SubstituteAndAppend(contents, " [default = $0",
2391                                  DefaultValueAsString(true));
2392   }
2393 
2394   string formatted_options;
2395   if (FormatBracketedOptions(depth, options(), &formatted_options)) {
2396     contents->append(bracketed ? ", " : " [");
2397     bracketed = true;
2398     contents->append(formatted_options);
2399   }
2400 
2401   if (bracketed) {
2402     contents->append("]");
2403   }
2404 
2405   if (type() == TYPE_GROUP) {
2406     if (debug_string_options.elide_group_body) {
2407       contents->append(" { ... };\n");
2408     } else {
2409       message_type()->DebugString(depth, contents, debug_string_options,
2410                                   /* include_opening_clause */ false);
2411     }
2412   } else {
2413     contents->append(";\n");
2414   }
2415 
2416   comment_printer.AddPostComment(contents);
2417 }
2418 
DebugString() const2419 string OneofDescriptor::DebugString() const {
2420   DebugStringOptions options;  // default values
2421   return DebugStringWithOptions(options);
2422 }
2423 
DebugStringWithOptions(const DebugStringOptions & options) const2424 string OneofDescriptor::DebugStringWithOptions(
2425     const DebugStringOptions& options) const {
2426   string contents;
2427   DebugString(0, &contents, options);
2428   return contents;
2429 }
2430 
DebugString(int depth,string * contents,const DebugStringOptions & debug_string_options) const2431 void OneofDescriptor::DebugString(int depth, string* contents,
2432                                   const DebugStringOptions&
2433                                   debug_string_options) const {
2434   string prefix(depth * 2, ' ');
2435   ++depth;
2436   SourceLocationCommentPrinter
2437       comment_printer(this, prefix, debug_string_options);
2438   comment_printer.AddPreComment(contents);
2439   strings::SubstituteAndAppend(
2440       contents, "$0 oneof $1 {", prefix, name());
2441 
2442   FormatLineOptions(depth, options(), contents);
2443 
2444   if (debug_string_options.elide_oneof_body) {
2445     contents->append(" ... }\n");
2446   } else {
2447     for (int i = 0; i < field_count(); i++) {
2448       field(i)->DebugString(depth, FieldDescriptor::OMIT_LABEL, contents,
2449                             debug_string_options);
2450     }
2451     strings::SubstituteAndAppend(contents, "$0}\n", prefix);
2452   }
2453   comment_printer.AddPostComment(contents);
2454 }
2455 
DebugString() const2456 string EnumDescriptor::DebugString() const {
2457   DebugStringOptions options;  // default values
2458   return DebugStringWithOptions(options);
2459 }
2460 
DebugStringWithOptions(const DebugStringOptions & options) const2461 string EnumDescriptor::DebugStringWithOptions(
2462     const DebugStringOptions& options) const {
2463   string contents;
2464   DebugString(0, &contents, options);
2465   return contents;
2466 }
2467 
DebugString(int depth,string * contents,const DebugStringOptions & debug_string_options) const2468 void EnumDescriptor::DebugString(int depth, string *contents,
2469                                  const DebugStringOptions&
2470                                  debug_string_options) const {
2471   string prefix(depth * 2, ' ');
2472   ++depth;
2473 
2474   SourceLocationCommentPrinter
2475       comment_printer(this, prefix, debug_string_options);
2476   comment_printer.AddPreComment(contents);
2477 
2478   strings::SubstituteAndAppend(contents, "$0enum $1 {\n",
2479                                prefix, name());
2480 
2481   FormatLineOptions(depth, options(), contents);
2482 
2483   for (int i = 0; i < value_count(); i++) {
2484     value(i)->DebugString(depth, contents, debug_string_options);
2485   }
2486   strings::SubstituteAndAppend(contents, "$0}\n", prefix);
2487 
2488   comment_printer.AddPostComment(contents);
2489 }
2490 
DebugString() const2491 string EnumValueDescriptor::DebugString() const {
2492   DebugStringOptions options;  // default values
2493   return DebugStringWithOptions(options);
2494 }
2495 
DebugStringWithOptions(const DebugStringOptions & options) const2496 string EnumValueDescriptor::DebugStringWithOptions(
2497     const DebugStringOptions& options) const {
2498   string contents;
2499   DebugString(0, &contents, options);
2500   return contents;
2501 }
2502 
DebugString(int depth,string * contents,const DebugStringOptions & debug_string_options) const2503 void EnumValueDescriptor::DebugString(int depth, string *contents,
2504                                       const DebugStringOptions&
2505                                       debug_string_options) const {
2506   string prefix(depth * 2, ' ');
2507 
2508   SourceLocationCommentPrinter
2509       comment_printer(this, prefix, debug_string_options);
2510   comment_printer.AddPreComment(contents);
2511 
2512   strings::SubstituteAndAppend(contents, "$0$1 = $2",
2513                                prefix, name(), number());
2514 
2515   string formatted_options;
2516   if (FormatBracketedOptions(depth, options(), &formatted_options)) {
2517     strings::SubstituteAndAppend(contents, " [$0]", formatted_options);
2518   }
2519   contents->append(";\n");
2520 
2521   comment_printer.AddPostComment(contents);
2522 }
2523 
DebugString() const2524 string ServiceDescriptor::DebugString() const {
2525   DebugStringOptions options;  // default values
2526   return DebugStringWithOptions(options);
2527 }
2528 
DebugStringWithOptions(const DebugStringOptions & options) const2529 string ServiceDescriptor::DebugStringWithOptions(
2530     const DebugStringOptions& options) const {
2531   string contents;
2532   DebugString(&contents, options);
2533   return contents;
2534 }
2535 
DebugString(string * contents,const DebugStringOptions & debug_string_options) const2536 void ServiceDescriptor::DebugString(string *contents,
2537                                     const DebugStringOptions&
2538                                     debug_string_options) const {
2539   SourceLocationCommentPrinter
2540       comment_printer(this, /* prefix */ "", debug_string_options);
2541   comment_printer.AddPreComment(contents);
2542 
2543   strings::SubstituteAndAppend(contents, "service $0 {\n", name());
2544 
2545   FormatLineOptions(1, options(), contents);
2546 
2547   for (int i = 0; i < method_count(); i++) {
2548     method(i)->DebugString(1, contents, debug_string_options);
2549   }
2550 
2551   contents->append("}\n");
2552 
2553   comment_printer.AddPostComment(contents);
2554 }
2555 
DebugString() const2556 string MethodDescriptor::DebugString() const {
2557   DebugStringOptions options;  // default values
2558   return DebugStringWithOptions(options);
2559 }
2560 
DebugStringWithOptions(const DebugStringOptions & options) const2561 string MethodDescriptor::DebugStringWithOptions(
2562     const DebugStringOptions& options) const {
2563   string contents;
2564   DebugString(0, &contents, options);
2565   return contents;
2566 }
2567 
DebugString(int depth,string * contents,const DebugStringOptions & debug_string_options) const2568 void MethodDescriptor::DebugString(int depth, string *contents,
2569                                    const DebugStringOptions&
2570                                    debug_string_options) const {
2571   string prefix(depth * 2, ' ');
2572   ++depth;
2573 
2574   SourceLocationCommentPrinter
2575       comment_printer(this, prefix, debug_string_options);
2576   comment_printer.AddPreComment(contents);
2577 
2578   strings::SubstituteAndAppend(contents, "$0rpc $1($4.$2) returns ($5.$3)",
2579                                prefix, name(),
2580                                input_type()->full_name(),
2581                                output_type()->full_name(),
2582                                client_streaming() ? "stream " : "",
2583                                server_streaming() ? "stream " : "");
2584 
2585   string formatted_options;
2586   if (FormatLineOptions(depth, options(), &formatted_options)) {
2587     strings::SubstituteAndAppend(contents, " {\n$0$1}\n",
2588                                  formatted_options, prefix);
2589   } else {
2590     contents->append(";\n");
2591   }
2592 
2593   comment_printer.AddPostComment(contents);
2594 }
2595 
2596 
2597 // Location methods ===============================================
2598 
GetSourceLocation(const vector<int> & path,SourceLocation * out_location) const2599 bool FileDescriptor::GetSourceLocation(const vector<int>& path,
2600                                        SourceLocation* out_location) const {
2601   GOOGLE_CHECK_NOTNULL(out_location);
2602   if (source_code_info_) {
2603     if (const SourceCodeInfo_Location* loc =
2604         tables_->GetSourceLocation(path, source_code_info_)) {
2605       const RepeatedField<int32>& span = loc->span();
2606       if (span.size() == 3 || span.size() == 4) {
2607         out_location->start_line   = span.Get(0);
2608         out_location->start_column = span.Get(1);
2609         out_location->end_line     = span.Get(span.size() == 3 ? 0 : 2);
2610         out_location->end_column   = span.Get(span.size() - 1);
2611 
2612         out_location->leading_comments = loc->leading_comments();
2613         out_location->trailing_comments = loc->trailing_comments();
2614         out_location->leading_detached_comments.assign(
2615             loc->leading_detached_comments().begin(),
2616             loc->leading_detached_comments().end());
2617         return true;
2618       }
2619     }
2620   }
2621   return false;
2622 }
2623 
GetSourceLocation(SourceLocation * out_location) const2624 bool FileDescriptor::GetSourceLocation(SourceLocation* out_location) const {
2625   vector<int> path;  // empty path for root FileDescriptor
2626   return GetSourceLocation(path, out_location);
2627 }
2628 
is_packed() const2629 bool FieldDescriptor::is_packed() const {
2630   if (!is_packable()) return false;
2631   if (file_->syntax() == FileDescriptor::SYNTAX_PROTO2) {
2632     return (options_ != NULL) && options_->packed();
2633   } else {
2634     return options_ == NULL || !options_->has_packed() || options_->packed();
2635   }
2636 }
2637 
GetSourceLocation(SourceLocation * out_location) const2638 bool Descriptor::GetSourceLocation(SourceLocation* out_location) const {
2639   vector<int> path;
2640   GetLocationPath(&path);
2641   return file()->GetSourceLocation(path, out_location);
2642 }
2643 
GetSourceLocation(SourceLocation * out_location) const2644 bool FieldDescriptor::GetSourceLocation(SourceLocation* out_location) const {
2645   vector<int> path;
2646   GetLocationPath(&path);
2647   return file()->GetSourceLocation(path, out_location);
2648 }
2649 
GetSourceLocation(SourceLocation * out_location) const2650 bool OneofDescriptor::GetSourceLocation(SourceLocation* out_location) const {
2651   vector<int> path;
2652   GetLocationPath(&path);
2653   return containing_type()->file()->GetSourceLocation(path, out_location);
2654 }
2655 
GetSourceLocation(SourceLocation * out_location) const2656 bool EnumDescriptor::GetSourceLocation(SourceLocation* out_location) const {
2657   vector<int> path;
2658   GetLocationPath(&path);
2659   return file()->GetSourceLocation(path, out_location);
2660 }
2661 
GetSourceLocation(SourceLocation * out_location) const2662 bool MethodDescriptor::GetSourceLocation(SourceLocation* out_location) const {
2663   vector<int> path;
2664   GetLocationPath(&path);
2665   return service()->file()->GetSourceLocation(path, out_location);
2666 }
2667 
GetSourceLocation(SourceLocation * out_location) const2668 bool ServiceDescriptor::GetSourceLocation(SourceLocation* out_location) const {
2669   vector<int> path;
2670   GetLocationPath(&path);
2671   return file()->GetSourceLocation(path, out_location);
2672 }
2673 
GetSourceLocation(SourceLocation * out_location) const2674 bool EnumValueDescriptor::GetSourceLocation(
2675     SourceLocation* out_location) const {
2676   vector<int> path;
2677   GetLocationPath(&path);
2678   return type()->file()->GetSourceLocation(path, out_location);
2679 }
2680 
GetLocationPath(vector<int> * output) const2681 void Descriptor::GetLocationPath(vector<int>* output) const {
2682   if (containing_type()) {
2683     containing_type()->GetLocationPath(output);
2684     output->push_back(DescriptorProto::kNestedTypeFieldNumber);
2685     output->push_back(index());
2686   } else {
2687     output->push_back(FileDescriptorProto::kMessageTypeFieldNumber);
2688     output->push_back(index());
2689   }
2690 }
2691 
GetLocationPath(vector<int> * output) const2692 void FieldDescriptor::GetLocationPath(vector<int>* output) const {
2693   if (is_extension()) {
2694     if (extension_scope() == NULL) {
2695       output->push_back(FileDescriptorProto::kExtensionFieldNumber);
2696       output->push_back(index());
2697     } else {
2698       extension_scope()->GetLocationPath(output);
2699       output->push_back(DescriptorProto::kExtensionFieldNumber);
2700       output->push_back(index());
2701     }
2702   } else {
2703     containing_type()->GetLocationPath(output);
2704     output->push_back(DescriptorProto::kFieldFieldNumber);
2705     output->push_back(index());
2706   }
2707 }
2708 
GetLocationPath(vector<int> * output) const2709 void OneofDescriptor::GetLocationPath(vector<int>* output) const {
2710   containing_type()->GetLocationPath(output);
2711   output->push_back(DescriptorProto::kOneofDeclFieldNumber);
2712   output->push_back(index());
2713 }
2714 
GetLocationPath(vector<int> * output) const2715 void EnumDescriptor::GetLocationPath(vector<int>* output) const {
2716   if (containing_type()) {
2717     containing_type()->GetLocationPath(output);
2718     output->push_back(DescriptorProto::kEnumTypeFieldNumber);
2719     output->push_back(index());
2720   } else {
2721     output->push_back(FileDescriptorProto::kEnumTypeFieldNumber);
2722     output->push_back(index());
2723   }
2724 }
2725 
GetLocationPath(vector<int> * output) const2726 void EnumValueDescriptor::GetLocationPath(vector<int>* output) const {
2727   type()->GetLocationPath(output);
2728   output->push_back(EnumDescriptorProto::kValueFieldNumber);
2729   output->push_back(index());
2730 }
2731 
GetLocationPath(vector<int> * output) const2732 void ServiceDescriptor::GetLocationPath(vector<int>* output) const {
2733   output->push_back(FileDescriptorProto::kServiceFieldNumber);
2734   output->push_back(index());
2735 }
2736 
GetLocationPath(vector<int> * output) const2737 void MethodDescriptor::GetLocationPath(vector<int>* output) const {
2738   service()->GetLocationPath(output);
2739   output->push_back(ServiceDescriptorProto::kMethodFieldNumber);
2740   output->push_back(index());
2741 }
2742 
2743 // ===================================================================
2744 
2745 namespace {
2746 
2747 // Represents an options message to interpret. Extension names in the option
2748 // name are resolved relative to name_scope. element_name and orig_opt are
2749 // used only for error reporting (since the parser records locations against
2750 // pointers in the original options, not the mutable copy). The Message must be
2751 // one of the Options messages in descriptor.proto.
2752 struct OptionsToInterpret {
OptionsToInterpretgoogle::protobuf::__anoncbf42d300611::OptionsToInterpret2753   OptionsToInterpret(const string& ns,
2754                      const string& el,
2755                      const Message* orig_opt,
2756                      Message* opt)
2757       : name_scope(ns),
2758         element_name(el),
2759         original_options(orig_opt),
2760         options(opt) {
2761   }
2762   string name_scope;
2763   string element_name;
2764   const Message* original_options;
2765   Message* options;
2766 };
2767 
2768 }  // namespace
2769 
2770 class DescriptorBuilder {
2771  public:
2772   DescriptorBuilder(const DescriptorPool* pool,
2773                     DescriptorPool::Tables* tables,
2774                     DescriptorPool::ErrorCollector* error_collector);
2775   ~DescriptorBuilder();
2776 
2777   const FileDescriptor* BuildFile(const FileDescriptorProto& proto);
2778 
2779  private:
2780   friend class OptionInterpreter;
2781 
2782   // Non-recursive part of BuildFile functionality.
2783   const FileDescriptor* BuildFileImpl(const FileDescriptorProto& proto);
2784 
2785   const DescriptorPool* pool_;
2786   DescriptorPool::Tables* tables_;  // for convenience
2787   DescriptorPool::ErrorCollector* error_collector_;
2788 
2789   // As we build descriptors we store copies of the options messages in
2790   // them. We put pointers to those copies in this vector, as we build, so we
2791   // can later (after cross-linking) interpret those options.
2792   vector<OptionsToInterpret> options_to_interpret_;
2793 
2794   bool had_errors_;
2795   string filename_;
2796   FileDescriptor* file_;
2797   FileDescriptorTables* file_tables_;
2798   set<const FileDescriptor*> dependencies_;
2799 
2800   // unused_dependency_ is used to record the unused imported files.
2801   // Note: public import is not considered.
2802   set<const FileDescriptor*> unused_dependency_;
2803 
2804   // If LookupSymbol() finds a symbol that is in a file which is not a declared
2805   // dependency of this file, it will fail, but will set
2806   // possible_undeclared_dependency_ to point at that file.  This is only used
2807   // by AddNotDefinedError() to report a more useful error message.
2808   // possible_undeclared_dependency_name_ is the name of the symbol that was
2809   // actually found in possible_undeclared_dependency_, which may be a parent
2810   // of the symbol actually looked for.
2811   const FileDescriptor* possible_undeclared_dependency_;
2812   string possible_undeclared_dependency_name_;
2813 
2814   // If LookupSymbol() could resolve a symbol which is not defined,
2815   // record the resolved name.  This is only used by AddNotDefinedError()
2816   // to report a more useful error message.
2817   string undefine_resolved_name_;
2818 
2819   void AddError(const string& element_name,
2820                 const Message& descriptor,
2821                 DescriptorPool::ErrorCollector::ErrorLocation location,
2822                 const string& error);
2823   void AddError(const string& element_name,
2824                 const Message& descriptor,
2825                 DescriptorPool::ErrorCollector::ErrorLocation location,
2826                 const char* error);
2827   void AddRecursiveImportError(const FileDescriptorProto& proto, int from_here);
2828   void AddTwiceListedError(const FileDescriptorProto& proto, int index);
2829   void AddImportError(const FileDescriptorProto& proto, int index);
2830 
2831   // Adds an error indicating that undefined_symbol was not defined.  Must
2832   // only be called after LookupSymbol() fails.
2833   void AddNotDefinedError(
2834     const string& element_name,
2835     const Message& descriptor,
2836     DescriptorPool::ErrorCollector::ErrorLocation location,
2837     const string& undefined_symbol);
2838 
2839   void AddWarning(const string& element_name, const Message& descriptor,
2840                   DescriptorPool::ErrorCollector::ErrorLocation location,
2841                   const string& error);
2842 
2843   // Silly helper which determines if the given file is in the given package.
2844   // I.e., either file->package() == package_name or file->package() is a
2845   // nested package within package_name.
2846   bool IsInPackage(const FileDescriptor* file, const string& package_name);
2847 
2848   // Helper function which finds all public dependencies of the given file, and
2849   // stores the them in the dependencies_ set in the builder.
2850   void RecordPublicDependencies(const FileDescriptor* file);
2851 
2852   // Like tables_->FindSymbol(), but additionally:
2853   // - Search the pool's underlay if not found in tables_.
2854   // - Insure that the resulting Symbol is from one of the file's declared
2855   //   dependencies.
2856   Symbol FindSymbol(const string& name);
2857 
2858   // Like FindSymbol() but does not require that the symbol is in one of the
2859   // file's declared dependencies.
2860   Symbol FindSymbolNotEnforcingDeps(const string& name);
2861 
2862   // This implements the body of FindSymbolNotEnforcingDeps().
2863   Symbol FindSymbolNotEnforcingDepsHelper(const DescriptorPool* pool,
2864                                           const string& name);
2865 
2866   // Like FindSymbol(), but looks up the name relative to some other symbol
2867   // name.  This first searches siblings of relative_to, then siblings of its
2868   // parents, etc.  For example, LookupSymbol("foo.bar", "baz.qux.corge") makes
2869   // the following calls, returning the first non-null result:
2870   // FindSymbol("baz.qux.foo.bar"), FindSymbol("baz.foo.bar"),
2871   // FindSymbol("foo.bar").  If AllowUnknownDependencies() has been called
2872   // on the DescriptorPool, this will generate a placeholder type if
2873   // the name is not found (unless the name itself is malformed).  The
2874   // placeholder_type parameter indicates what kind of placeholder should be
2875   // constructed in this case.  The resolve_mode parameter determines whether
2876   // any symbol is returned, or only symbols that are types.  Note, however,
2877   // that LookupSymbol may still return a non-type symbol in LOOKUP_TYPES mode,
2878   // if it believes that's all it could refer to.  The caller should always
2879   // check that it receives the type of symbol it was expecting.
2880   enum PlaceholderType {
2881     PLACEHOLDER_MESSAGE,
2882     PLACEHOLDER_ENUM,
2883     PLACEHOLDER_EXTENDABLE_MESSAGE
2884   };
2885   enum ResolveMode {
2886     LOOKUP_ALL, LOOKUP_TYPES
2887   };
2888   Symbol LookupSymbol(const string& name, const string& relative_to,
2889                       PlaceholderType placeholder_type = PLACEHOLDER_MESSAGE,
2890                       ResolveMode resolve_mode = LOOKUP_ALL);
2891 
2892   // Like LookupSymbol() but will not return a placeholder even if
2893   // AllowUnknownDependencies() has been used.
2894   Symbol LookupSymbolNoPlaceholder(const string& name,
2895                                    const string& relative_to,
2896                                    ResolveMode resolve_mode = LOOKUP_ALL);
2897 
2898   // Creates a placeholder type suitable for return from LookupSymbol().  May
2899   // return kNullSymbol if the name is not a valid type name.
2900   Symbol NewPlaceholder(const string& name, PlaceholderType placeholder_type);
2901 
2902   // Creates a placeholder file.  Never returns NULL.  This is used when an
2903   // import is not found and AllowUnknownDependencies() is enabled.
2904   FileDescriptor* NewPlaceholderFile(const string& name);
2905 
2906   // Calls tables_->AddSymbol() and records an error if it fails.  Returns
2907   // true if successful or false if failed, though most callers can ignore
2908   // the return value since an error has already been recorded.
2909   bool AddSymbol(const string& full_name,
2910                  const void* parent, const string& name,
2911                  const Message& proto, Symbol symbol);
2912 
2913   // Like AddSymbol(), but succeeds if the symbol is already defined as long
2914   // as the existing definition is also a package (because it's OK to define
2915   // the same package in two different files).  Also adds all parents of the
2916   // packgae to the symbol table (e.g. AddPackage("foo.bar", ...) will add
2917   // "foo.bar" and "foo" to the table).
2918   void AddPackage(const string& name, const Message& proto,
2919                   const FileDescriptor* file);
2920 
2921   // Checks that the symbol name contains only alphanumeric characters and
2922   // underscores.  Records an error otherwise.
2923   void ValidateSymbolName(const string& name, const string& full_name,
2924                           const Message& proto);
2925 
2926   // Like ValidateSymbolName(), but the name is allowed to contain periods and
2927   // an error is indicated by returning false (not recording the error).
2928   bool ValidateQualifiedName(const string& name);
2929 
2930   // Used by BUILD_ARRAY macro (below) to avoid having to have the type
2931   // specified as a macro parameter.
2932   template <typename Type>
AllocateArray(int size,Type ** output)2933   inline void AllocateArray(int size, Type** output) {
2934     *output = tables_->AllocateArray<Type>(size);
2935   }
2936 
2937   // Allocates a copy of orig_options in tables_ and stores it in the
2938   // descriptor. Remembers its uninterpreted options, to be interpreted
2939   // later. DescriptorT must be one of the Descriptor messages from
2940   // descriptor.proto.
2941   template<class DescriptorT> void AllocateOptions(
2942       const typename DescriptorT::OptionsType& orig_options,
2943       DescriptorT* descriptor);
2944   // Specialization for FileOptions.
2945   void AllocateOptions(const FileOptions& orig_options,
2946                        FileDescriptor* descriptor);
2947 
2948   // Implementation for AllocateOptions(). Don't call this directly.
2949   template<class DescriptorT> void AllocateOptionsImpl(
2950       const string& name_scope,
2951       const string& element_name,
2952       const typename DescriptorT::OptionsType& orig_options,
2953       DescriptorT* descriptor);
2954 
2955   // These methods all have the same signature for the sake of the BUILD_ARRAY
2956   // macro, below.
2957   void BuildMessage(const DescriptorProto& proto,
2958                     const Descriptor* parent,
2959                     Descriptor* result);
2960   void BuildFieldOrExtension(const FieldDescriptorProto& proto,
2961                              const Descriptor* parent,
2962                              FieldDescriptor* result,
2963                              bool is_extension);
BuildField(const FieldDescriptorProto & proto,const Descriptor * parent,FieldDescriptor * result)2964   void BuildField(const FieldDescriptorProto& proto,
2965                   const Descriptor* parent,
2966                   FieldDescriptor* result) {
2967     BuildFieldOrExtension(proto, parent, result, false);
2968   }
BuildExtension(const FieldDescriptorProto & proto,const Descriptor * parent,FieldDescriptor * result)2969   void BuildExtension(const FieldDescriptorProto& proto,
2970                       const Descriptor* parent,
2971                       FieldDescriptor* result) {
2972     BuildFieldOrExtension(proto, parent, result, true);
2973   }
2974   void BuildExtensionRange(const DescriptorProto::ExtensionRange& proto,
2975                            const Descriptor* parent,
2976                            Descriptor::ExtensionRange* result);
2977   void BuildReservedRange(const DescriptorProto::ReservedRange& proto,
2978                            const Descriptor* parent,
2979                            Descriptor::ReservedRange* result);
2980   void BuildOneof(const OneofDescriptorProto& proto,
2981                   Descriptor* parent,
2982                   OneofDescriptor* result);
2983   void BuildEnum(const EnumDescriptorProto& proto,
2984                  const Descriptor* parent,
2985                  EnumDescriptor* result);
2986   void BuildEnumValue(const EnumValueDescriptorProto& proto,
2987                       const EnumDescriptor* parent,
2988                       EnumValueDescriptor* result);
2989   void BuildService(const ServiceDescriptorProto& proto,
2990                     const void* dummy,
2991                     ServiceDescriptor* result);
2992   void BuildMethod(const MethodDescriptorProto& proto,
2993                    const ServiceDescriptor* parent,
2994                    MethodDescriptor* result);
2995 
2996   void LogUnusedDependency(const FileDescriptorProto& proto,
2997                            const FileDescriptor* result);
2998 
2999   // Must be run only after building.
3000   //
3001   // NOTE: Options will not be available during cross-linking, as they
3002   // have not yet been interpreted. Defer any handling of options to the
3003   // Validate*Options methods.
3004   void CrossLinkFile(FileDescriptor* file, const FileDescriptorProto& proto);
3005   void CrossLinkMessage(Descriptor* message, const DescriptorProto& proto);
3006   void CrossLinkField(FieldDescriptor* field,
3007                       const FieldDescriptorProto& proto);
3008   void CrossLinkEnum(EnumDescriptor* enum_type,
3009                      const EnumDescriptorProto& proto);
3010   void CrossLinkEnumValue(EnumValueDescriptor* enum_value,
3011                           const EnumValueDescriptorProto& proto);
3012   void CrossLinkService(ServiceDescriptor* service,
3013                         const ServiceDescriptorProto& proto);
3014   void CrossLinkMethod(MethodDescriptor* method,
3015                        const MethodDescriptorProto& proto);
3016 
3017   // Must be run only after cross-linking.
3018   void InterpretOptions();
3019 
3020   // A helper class for interpreting options.
3021   class OptionInterpreter {
3022    public:
3023     // Creates an interpreter that operates in the context of the pool of the
3024     // specified builder, which must not be NULL. We don't take ownership of the
3025     // builder.
3026     explicit OptionInterpreter(DescriptorBuilder* builder);
3027 
3028     ~OptionInterpreter();
3029 
3030     // Interprets the uninterpreted options in the specified Options message.
3031     // On error, calls AddError() on the underlying builder and returns false.
3032     // Otherwise returns true.
3033     bool InterpretOptions(OptionsToInterpret* options_to_interpret);
3034 
3035     class AggregateOptionFinder;
3036 
3037    private:
3038     // Interprets uninterpreted_option_ on the specified message, which
3039     // must be the mutable copy of the original options message to which
3040     // uninterpreted_option_ belongs.
3041     bool InterpretSingleOption(Message* options);
3042 
3043     // Adds the uninterpreted_option to the given options message verbatim.
3044     // Used when AllowUnknownDependencies() is in effect and we can't find
3045     // the option's definition.
3046     void AddWithoutInterpreting(const UninterpretedOption& uninterpreted_option,
3047                                 Message* options);
3048 
3049     // A recursive helper function that drills into the intermediate fields
3050     // in unknown_fields to check if field innermost_field is set on the
3051     // innermost message. Returns false and sets an error if so.
3052     bool ExamineIfOptionIsSet(
3053         vector<const FieldDescriptor*>::const_iterator intermediate_fields_iter,
3054         vector<const FieldDescriptor*>::const_iterator intermediate_fields_end,
3055         const FieldDescriptor* innermost_field, const string& debug_msg_name,
3056         const UnknownFieldSet& unknown_fields);
3057 
3058     // Validates the value for the option field of the currently interpreted
3059     // option and then sets it on the unknown_field.
3060     bool SetOptionValue(const FieldDescriptor* option_field,
3061                         UnknownFieldSet* unknown_fields);
3062 
3063     // Parses an aggregate value for a CPPTYPE_MESSAGE option and
3064     // saves it into *unknown_fields.
3065     bool SetAggregateOption(const FieldDescriptor* option_field,
3066                             UnknownFieldSet* unknown_fields);
3067 
3068     // Convenience functions to set an int field the right way, depending on
3069     // its wire type (a single int CppType can represent multiple wire types).
3070     void SetInt32(int number, int32 value, FieldDescriptor::Type type,
3071                   UnknownFieldSet* unknown_fields);
3072     void SetInt64(int number, int64 value, FieldDescriptor::Type type,
3073                   UnknownFieldSet* unknown_fields);
3074     void SetUInt32(int number, uint32 value, FieldDescriptor::Type type,
3075                    UnknownFieldSet* unknown_fields);
3076     void SetUInt64(int number, uint64 value, FieldDescriptor::Type type,
3077                    UnknownFieldSet* unknown_fields);
3078 
3079     // A helper function that adds an error at the specified location of the
3080     // option we're currently interpreting, and returns false.
AddOptionError(DescriptorPool::ErrorCollector::ErrorLocation location,const string & msg)3081     bool AddOptionError(DescriptorPool::ErrorCollector::ErrorLocation location,
3082                         const string& msg) {
3083       builder_->AddError(options_to_interpret_->element_name,
3084                          *uninterpreted_option_, location, msg);
3085       return false;
3086     }
3087 
3088     // A helper function that adds an error at the location of the option name
3089     // and returns false.
AddNameError(const string & msg)3090     bool AddNameError(const string& msg) {
3091       return AddOptionError(DescriptorPool::ErrorCollector::OPTION_NAME, msg);
3092     }
3093 
3094     // A helper function that adds an error at the location of the option name
3095     // and returns false.
AddValueError(const string & msg)3096     bool AddValueError(const string& msg) {
3097       return AddOptionError(DescriptorPool::ErrorCollector::OPTION_VALUE, msg);
3098     }
3099 
3100     // We interpret against this builder's pool. Is never NULL. We don't own
3101     // this pointer.
3102     DescriptorBuilder* builder_;
3103 
3104     // The options we're currently interpreting, or NULL if we're not in a call
3105     // to InterpretOptions.
3106     const OptionsToInterpret* options_to_interpret_;
3107 
3108     // The option we're currently interpreting within options_to_interpret_, or
3109     // NULL if we're not in a call to InterpretOptions(). This points to a
3110     // submessage of the original option, not the mutable copy. Therefore we
3111     // can use it to find locations recorded by the parser.
3112     const UninterpretedOption* uninterpreted_option_;
3113 
3114     // Factory used to create the dynamic messages we need to parse
3115     // any aggregate option values we encounter.
3116     DynamicMessageFactory dynamic_factory_;
3117 
3118     GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(OptionInterpreter);
3119   };
3120 
3121   // Work-around for broken compilers:  According to the C++ standard,
3122   // OptionInterpreter should have access to the private members of any class
3123   // which has declared DescriptorBuilder as a friend.  Unfortunately some old
3124   // versions of GCC and other compilers do not implement this correctly.  So,
3125   // we have to have these intermediate methods to provide access.  We also
3126   // redundantly declare OptionInterpreter a friend just to make things extra
3127   // clear for these bad compilers.
3128   friend class OptionInterpreter;
3129   friend class OptionInterpreter::AggregateOptionFinder;
3130 
get_allow_unknown(const DescriptorPool * pool)3131   static inline bool get_allow_unknown(const DescriptorPool* pool) {
3132     return pool->allow_unknown_;
3133   }
get_enforce_weak(const DescriptorPool * pool)3134   static inline bool get_enforce_weak(const DescriptorPool* pool) {
3135     return pool->enforce_weak_;
3136   }
get_is_placeholder(const Descriptor * descriptor)3137   static inline bool get_is_placeholder(const Descriptor* descriptor) {
3138     return descriptor->is_placeholder_;
3139   }
assert_mutex_held(const DescriptorPool * pool)3140   static inline void assert_mutex_held(const DescriptorPool* pool) {
3141     if (pool->mutex_ != NULL) {
3142       pool->mutex_->AssertHeld();
3143     }
3144   }
3145 
3146   // Must be run only after options have been interpreted.
3147   //
3148   // NOTE: Validation code must only reference the options in the mutable
3149   // descriptors, which are the ones that have been interpreted. The const
3150   // proto references are passed in only so they can be provided to calls to
3151   // AddError(). Do not look at their options, which have not been interpreted.
3152   void ValidateFileOptions(FileDescriptor* file,
3153                            const FileDescriptorProto& proto);
3154   void ValidateMessageOptions(Descriptor* message,
3155                               const DescriptorProto& proto);
3156   void ValidateFieldOptions(FieldDescriptor* field,
3157                             const FieldDescriptorProto& proto);
3158   void ValidateEnumOptions(EnumDescriptor* enm,
3159                            const EnumDescriptorProto& proto);
3160   void ValidateEnumValueOptions(EnumValueDescriptor* enum_value,
3161                                 const EnumValueDescriptorProto& proto);
3162   void ValidateServiceOptions(ServiceDescriptor* service,
3163                               const ServiceDescriptorProto& proto);
3164   void ValidateMethodOptions(MethodDescriptor* method,
3165                              const MethodDescriptorProto& proto);
3166   void ValidateProto3(FileDescriptor* file,
3167                       const FileDescriptorProto& proto);
3168   void ValidateProto3Message(Descriptor* message,
3169                              const DescriptorProto& proto);
3170   void ValidateProto3Field(FieldDescriptor* field,
3171                            const FieldDescriptorProto& proto);
3172   void ValidateProto3Enum(EnumDescriptor* enm,
3173                           const EnumDescriptorProto& proto);
3174 
3175   // Returns true if the map entry message is compatible with the
3176   // auto-generated entry message from map fields syntax.
3177   bool ValidateMapEntry(FieldDescriptor* field,
3178                         const FieldDescriptorProto& proto);
3179 
3180   // Recursively detects naming conflicts with map entry types for a
3181   // better error message.
3182   void DetectMapConflicts(const Descriptor* message,
3183                           const DescriptorProto& proto);
3184 
3185 };
3186 
BuildFile(const FileDescriptorProto & proto)3187 const FileDescriptor* DescriptorPool::BuildFile(
3188     const FileDescriptorProto& proto) {
3189   GOOGLE_CHECK(fallback_database_ == NULL)
3190     << "Cannot call BuildFile on a DescriptorPool that uses a "
3191        "DescriptorDatabase.  You must instead find a way to get your file "
3192        "into the underlying database.";
3193   GOOGLE_CHECK(mutex_ == NULL);   // Implied by the above GOOGLE_CHECK.
3194   tables_->known_bad_symbols_.clear();
3195   tables_->known_bad_files_.clear();
3196   return DescriptorBuilder(this, tables_.get(), NULL).BuildFile(proto);
3197 }
3198 
BuildFileCollectingErrors(const FileDescriptorProto & proto,ErrorCollector * error_collector)3199 const FileDescriptor* DescriptorPool::BuildFileCollectingErrors(
3200     const FileDescriptorProto& proto,
3201     ErrorCollector* error_collector) {
3202   GOOGLE_CHECK(fallback_database_ == NULL)
3203     << "Cannot call BuildFile on a DescriptorPool that uses a "
3204        "DescriptorDatabase.  You must instead find a way to get your file "
3205        "into the underlying database.";
3206   GOOGLE_CHECK(mutex_ == NULL);   // Implied by the above GOOGLE_CHECK.
3207   tables_->known_bad_symbols_.clear();
3208   tables_->known_bad_files_.clear();
3209   return DescriptorBuilder(this, tables_.get(),
3210                            error_collector).BuildFile(proto);
3211 }
3212 
BuildFileFromDatabase(const FileDescriptorProto & proto) const3213 const FileDescriptor* DescriptorPool::BuildFileFromDatabase(
3214     const FileDescriptorProto& proto) const {
3215   mutex_->AssertHeld();
3216   if (tables_->known_bad_files_.count(proto.name()) > 0) {
3217     return NULL;
3218   }
3219   const FileDescriptor* result =
3220       DescriptorBuilder(this, tables_.get(),
3221                         default_error_collector_).BuildFile(proto);
3222   if (result == NULL) {
3223     tables_->known_bad_files_.insert(proto.name());
3224   }
3225   return result;
3226 }
3227 
DescriptorBuilder(const DescriptorPool * pool,DescriptorPool::Tables * tables,DescriptorPool::ErrorCollector * error_collector)3228 DescriptorBuilder::DescriptorBuilder(
3229     const DescriptorPool* pool,
3230     DescriptorPool::Tables* tables,
3231     DescriptorPool::ErrorCollector* error_collector)
3232   : pool_(pool),
3233     tables_(tables),
3234     error_collector_(error_collector),
3235     had_errors_(false),
3236     possible_undeclared_dependency_(NULL),
3237     undefine_resolved_name_("") {}
3238 
~DescriptorBuilder()3239 DescriptorBuilder::~DescriptorBuilder() {}
3240 
AddError(const string & element_name,const Message & descriptor,DescriptorPool::ErrorCollector::ErrorLocation location,const string & error)3241 void DescriptorBuilder::AddError(
3242     const string& element_name,
3243     const Message& descriptor,
3244     DescriptorPool::ErrorCollector::ErrorLocation location,
3245     const string& error) {
3246   if (error_collector_ == NULL) {
3247     if (!had_errors_) {
3248       GOOGLE_LOG(ERROR) << "Invalid proto descriptor for file \"" << filename_
3249                  << "\":";
3250     }
3251     GOOGLE_LOG(ERROR) << "  " << element_name << ": " << error;
3252   } else {
3253     error_collector_->AddError(filename_, element_name,
3254                                &descriptor, location, error);
3255   }
3256   had_errors_ = true;
3257 }
3258 
AddError(const string & element_name,const Message & descriptor,DescriptorPool::ErrorCollector::ErrorLocation location,const char * error)3259 void DescriptorBuilder::AddError(
3260     const string& element_name,
3261     const Message& descriptor,
3262     DescriptorPool::ErrorCollector::ErrorLocation location,
3263     const char* error) {
3264   AddError(element_name, descriptor, location, string(error));
3265 }
3266 
AddNotDefinedError(const string & element_name,const Message & descriptor,DescriptorPool::ErrorCollector::ErrorLocation location,const string & undefined_symbol)3267 void DescriptorBuilder::AddNotDefinedError(
3268     const string& element_name,
3269     const Message& descriptor,
3270     DescriptorPool::ErrorCollector::ErrorLocation location,
3271     const string& undefined_symbol) {
3272   if (possible_undeclared_dependency_ == NULL &&
3273       undefine_resolved_name_.empty()) {
3274     AddError(element_name, descriptor, location,
3275              "\"" + undefined_symbol + "\" is not defined.");
3276   } else {
3277     if (possible_undeclared_dependency_ != NULL) {
3278       AddError(element_name, descriptor, location,
3279                "\"" + possible_undeclared_dependency_name_ +
3280                "\" seems to be defined in \"" +
3281                possible_undeclared_dependency_->name() + "\", which is not "
3282                "imported by \"" + filename_ + "\".  To use it here, please "
3283                "add the necessary import.");
3284     }
3285     if (!undefine_resolved_name_.empty()) {
3286       AddError(element_name, descriptor, location,
3287                "\"" + undefined_symbol + "\" is resolved to \"" +
3288                undefine_resolved_name_ + "\", which is not defined. "
3289                "The innermost scope is searched first in name resolution. "
3290                "Consider using a leading '.'(i.e., \"."
3291                + undefined_symbol +
3292                "\") to start from the outermost scope.");
3293     }
3294   }
3295 }
3296 
AddWarning(const string & element_name,const Message & descriptor,DescriptorPool::ErrorCollector::ErrorLocation location,const string & error)3297 void DescriptorBuilder::AddWarning(
3298     const string& element_name, const Message& descriptor,
3299     DescriptorPool::ErrorCollector::ErrorLocation location,
3300     const string& error) {
3301   if (error_collector_ == NULL) {
3302     GOOGLE_LOG(WARNING) << filename_ << " " << element_name << ": " << error;
3303   } else {
3304     error_collector_->AddWarning(filename_, element_name, &descriptor, location,
3305                                  error);
3306   }
3307 }
3308 
IsInPackage(const FileDescriptor * file,const string & package_name)3309 bool DescriptorBuilder::IsInPackage(const FileDescriptor* file,
3310                                     const string& package_name) {
3311   return HasPrefixString(file->package(), package_name) &&
3312            (file->package().size() == package_name.size() ||
3313             file->package()[package_name.size()] == '.');
3314 }
3315 
RecordPublicDependencies(const FileDescriptor * file)3316 void DescriptorBuilder::RecordPublicDependencies(const FileDescriptor* file) {
3317   if (file == NULL || !dependencies_.insert(file).second) return;
3318   for (int i = 0; file != NULL && i < file->public_dependency_count(); i++) {
3319     RecordPublicDependencies(file->public_dependency(i));
3320   }
3321 }
3322 
FindSymbolNotEnforcingDepsHelper(const DescriptorPool * pool,const string & name)3323 Symbol DescriptorBuilder::FindSymbolNotEnforcingDepsHelper(
3324     const DescriptorPool* pool, const string& name) {
3325   // If we are looking at an underlay, we must lock its mutex_, since we are
3326   // accessing the underlay's tables_ directly.
3327   MutexLockMaybe lock((pool == pool_) ? NULL : pool->mutex_);
3328 
3329   Symbol result = pool->tables_->FindSymbol(name);
3330   if (result.IsNull() && pool->underlay_ != NULL) {
3331     // Symbol not found; check the underlay.
3332     result = FindSymbolNotEnforcingDepsHelper(pool->underlay_, name);
3333   }
3334 
3335   if (result.IsNull()) {
3336     // In theory, we shouldn't need to check fallback_database_ because the
3337     // symbol should be in one of its file's direct dependencies, and we have
3338     // already loaded those by the time we get here.  But we check anyway so
3339     // that we can generate better error message when dependencies are missing
3340     // (i.e., "missing dependency" rather than "type is not defined").
3341     if (pool->TryFindSymbolInFallbackDatabase(name)) {
3342       result = pool->tables_->FindSymbol(name);
3343     }
3344   }
3345 
3346   return result;
3347 }
3348 
FindSymbolNotEnforcingDeps(const string & name)3349 Symbol DescriptorBuilder::FindSymbolNotEnforcingDeps(const string& name) {
3350   return FindSymbolNotEnforcingDepsHelper(pool_, name);
3351 }
3352 
FindSymbol(const string & name)3353 Symbol DescriptorBuilder::FindSymbol(const string& name) {
3354   Symbol result = FindSymbolNotEnforcingDeps(name);
3355 
3356   if (result.IsNull()) return result;
3357 
3358   if (!pool_->enforce_dependencies_) {
3359     // Hack for CompilerUpgrader.
3360     return result;
3361   }
3362 
3363   // Only find symbols which were defined in this file or one of its
3364   // dependencies.
3365   const FileDescriptor* file = result.GetFile();
3366   if (file == file_ || dependencies_.count(file) > 0) {
3367     unused_dependency_.erase(file);
3368     return result;
3369   }
3370 
3371   if (result.type == Symbol::PACKAGE) {
3372     // Arg, this is overcomplicated.  The symbol is a package name.  It could
3373     // be that the package was defined in multiple files.  result.GetFile()
3374     // returns the first file we saw that used this package.  We've determined
3375     // that that file is not a direct dependency of the file we are currently
3376     // building, but it could be that some other file which *is* a direct
3377     // dependency also defines the same package.  We can't really rule out this
3378     // symbol unless none of the dependencies define it.
3379     if (IsInPackage(file_, name)) return result;
3380     for (set<const FileDescriptor*>::const_iterator it = dependencies_.begin();
3381          it != dependencies_.end(); ++it) {
3382       // Note:  A dependency may be NULL if it was not found or had errors.
3383       if (*it != NULL && IsInPackage(*it, name)) return result;
3384     }
3385   }
3386 
3387   possible_undeclared_dependency_ = file;
3388   possible_undeclared_dependency_name_ = name;
3389   return kNullSymbol;
3390 }
3391 
LookupSymbolNoPlaceholder(const string & name,const string & relative_to,ResolveMode resolve_mode)3392 Symbol DescriptorBuilder::LookupSymbolNoPlaceholder(
3393     const string& name, const string& relative_to, ResolveMode resolve_mode) {
3394   possible_undeclared_dependency_ = NULL;
3395   undefine_resolved_name_.clear();
3396 
3397   if (name.size() > 0 && name[0] == '.') {
3398     // Fully-qualified name.
3399     return FindSymbol(name.substr(1));
3400   }
3401 
3402   // If name is something like "Foo.Bar.baz", and symbols named "Foo" are
3403   // defined in multiple parent scopes, we only want to find "Bar.baz" in the
3404   // innermost one.  E.g., the following should produce an error:
3405   //   message Bar { message Baz {} }
3406   //   message Foo {
3407   //     message Bar {
3408   //     }
3409   //     optional Bar.Baz baz = 1;
3410   //   }
3411   // So, we look for just "Foo" first, then look for "Bar.baz" within it if
3412   // found.
3413   string::size_type name_dot_pos = name.find_first_of('.');
3414   string first_part_of_name;
3415   if (name_dot_pos == string::npos) {
3416     first_part_of_name = name;
3417   } else {
3418     first_part_of_name = name.substr(0, name_dot_pos);
3419   }
3420 
3421   string scope_to_try(relative_to);
3422 
3423   while (true) {
3424     // Chop off the last component of the scope.
3425     string::size_type dot_pos = scope_to_try.find_last_of('.');
3426     if (dot_pos == string::npos) {
3427       return FindSymbol(name);
3428     } else {
3429       scope_to_try.erase(dot_pos);
3430     }
3431 
3432     // Append ".first_part_of_name" and try to find.
3433     string::size_type old_size = scope_to_try.size();
3434     scope_to_try.append(1, '.');
3435     scope_to_try.append(first_part_of_name);
3436     Symbol result = FindSymbol(scope_to_try);
3437     if (!result.IsNull()) {
3438       if (first_part_of_name.size() < name.size()) {
3439         // name is a compound symbol, of which we only found the first part.
3440         // Now try to look up the rest of it.
3441         if (result.IsAggregate()) {
3442           scope_to_try.append(name, first_part_of_name.size(),
3443                               name.size() - first_part_of_name.size());
3444           result = FindSymbol(scope_to_try);
3445           if (result.IsNull()) {
3446             undefine_resolved_name_ = scope_to_try;
3447           }
3448           return result;
3449         } else {
3450           // We found a symbol but it's not an aggregate.  Continue the loop.
3451         }
3452       } else {
3453         if (resolve_mode == LOOKUP_TYPES && !result.IsType()) {
3454           // We found a symbol but it's not a type.  Continue the loop.
3455         } else {
3456           return result;
3457         }
3458       }
3459     }
3460 
3461     // Not found.  Remove the name so we can try again.
3462     scope_to_try.erase(old_size);
3463   }
3464 }
3465 
LookupSymbol(const string & name,const string & relative_to,PlaceholderType placeholder_type,ResolveMode resolve_mode)3466 Symbol DescriptorBuilder::LookupSymbol(
3467     const string& name, const string& relative_to,
3468     PlaceholderType placeholder_type, ResolveMode resolve_mode) {
3469   Symbol result = LookupSymbolNoPlaceholder(
3470       name, relative_to, resolve_mode);
3471   if (result.IsNull() && pool_->allow_unknown_) {
3472     // Not found, but AllowUnknownDependencies() is enabled.  Return a
3473     // placeholder instead.
3474     result = NewPlaceholder(name, placeholder_type);
3475   }
3476   return result;
3477 }
3478 
NewPlaceholder(const string & name,PlaceholderType placeholder_type)3479 Symbol DescriptorBuilder::NewPlaceholder(const string& name,
3480                                          PlaceholderType placeholder_type) {
3481   // Compute names.
3482   const string* placeholder_full_name;
3483   const string* placeholder_name;
3484   const string* placeholder_package;
3485 
3486   if (!ValidateQualifiedName(name)) return kNullSymbol;
3487   if (name[0] == '.') {
3488     // Fully-qualified.
3489     placeholder_full_name = tables_->AllocateString(name.substr(1));
3490   } else {
3491     placeholder_full_name = tables_->AllocateString(name);
3492   }
3493 
3494   string::size_type dotpos = placeholder_full_name->find_last_of('.');
3495   if (dotpos != string::npos) {
3496     placeholder_package = tables_->AllocateString(
3497       placeholder_full_name->substr(0, dotpos));
3498     placeholder_name = tables_->AllocateString(
3499       placeholder_full_name->substr(dotpos + 1));
3500   } else {
3501     placeholder_package = &internal::GetEmptyString();
3502     placeholder_name = placeholder_full_name;
3503   }
3504 
3505   // Create the placeholders.
3506   FileDescriptor* placeholder_file = NewPlaceholderFile(
3507       *placeholder_full_name + ".placeholder.proto");
3508   placeholder_file->package_ = placeholder_package;
3509 
3510   if (placeholder_type == PLACEHOLDER_ENUM) {
3511     placeholder_file->enum_type_count_ = 1;
3512     placeholder_file->enum_types_ =
3513       tables_->AllocateArray<EnumDescriptor>(1);
3514 
3515     EnumDescriptor* placeholder_enum = &placeholder_file->enum_types_[0];
3516     memset(placeholder_enum, 0, sizeof(*placeholder_enum));
3517 
3518     placeholder_enum->full_name_ = placeholder_full_name;
3519     placeholder_enum->name_ = placeholder_name;
3520     placeholder_enum->file_ = placeholder_file;
3521     placeholder_enum->options_ = &EnumOptions::default_instance();
3522     placeholder_enum->is_placeholder_ = true;
3523     placeholder_enum->is_unqualified_placeholder_ = (name[0] != '.');
3524 
3525     // Enums must have at least one value.
3526     placeholder_enum->value_count_ = 1;
3527     placeholder_enum->values_ = tables_->AllocateArray<EnumValueDescriptor>(1);
3528 
3529     EnumValueDescriptor* placeholder_value = &placeholder_enum->values_[0];
3530     memset(placeholder_value, 0, sizeof(*placeholder_value));
3531 
3532     placeholder_value->name_ = tables_->AllocateString("PLACEHOLDER_VALUE");
3533     // Note that enum value names are siblings of their type, not children.
3534     placeholder_value->full_name_ =
3535       placeholder_package->empty() ? placeholder_value->name_ :
3536         tables_->AllocateString(*placeholder_package + ".PLACEHOLDER_VALUE");
3537 
3538     placeholder_value->number_ = 0;
3539     placeholder_value->type_ = placeholder_enum;
3540     placeholder_value->options_ = &EnumValueOptions::default_instance();
3541 
3542     return Symbol(placeholder_enum);
3543   } else {
3544     placeholder_file->message_type_count_ = 1;
3545     placeholder_file->message_types_ =
3546       tables_->AllocateArray<Descriptor>(1);
3547 
3548     Descriptor* placeholder_message = &placeholder_file->message_types_[0];
3549     memset(placeholder_message, 0, sizeof(*placeholder_message));
3550 
3551     placeholder_message->full_name_ = placeholder_full_name;
3552     placeholder_message->name_ = placeholder_name;
3553     placeholder_message->file_ = placeholder_file;
3554     placeholder_message->options_ = &MessageOptions::default_instance();
3555     placeholder_message->is_placeholder_ = true;
3556     placeholder_message->is_unqualified_placeholder_ = (name[0] != '.');
3557 
3558     if (placeholder_type == PLACEHOLDER_EXTENDABLE_MESSAGE) {
3559       placeholder_message->extension_range_count_ = 1;
3560       placeholder_message->extension_ranges_ =
3561         tables_->AllocateArray<Descriptor::ExtensionRange>(1);
3562       placeholder_message->extension_ranges_->start = 1;
3563       // kMaxNumber + 1 because ExtensionRange::end is exclusive.
3564       placeholder_message->extension_ranges_->end =
3565         FieldDescriptor::kMaxNumber + 1;
3566     }
3567 
3568     return Symbol(placeholder_message);
3569   }
3570 }
3571 
NewPlaceholderFile(const string & name)3572 FileDescriptor* DescriptorBuilder::NewPlaceholderFile(
3573     const string& name) {
3574   FileDescriptor* placeholder = tables_->Allocate<FileDescriptor>();
3575   memset(placeholder, 0, sizeof(*placeholder));
3576 
3577   placeholder->name_ = tables_->AllocateString(name);
3578   placeholder->package_ = &internal::GetEmptyString();
3579   placeholder->pool_ = pool_;
3580   placeholder->options_ = &FileOptions::default_instance();
3581   placeholder->tables_ = &FileDescriptorTables::GetEmptyInstance();
3582   placeholder->source_code_info_ = &SourceCodeInfo::default_instance();
3583   placeholder->is_placeholder_ = true;
3584   placeholder->syntax_ = FileDescriptor::SYNTAX_PROTO2;
3585   // All other fields are zero or NULL.
3586 
3587   return placeholder;
3588 }
3589 
AddSymbol(const string & full_name,const void * parent,const string & name,const Message & proto,Symbol symbol)3590 bool DescriptorBuilder::AddSymbol(
3591     const string& full_name, const void* parent, const string& name,
3592     const Message& proto, Symbol symbol) {
3593   // If the caller passed NULL for the parent, the symbol is at file scope.
3594   // Use its file as the parent instead.
3595   if (parent == NULL) parent = file_;
3596 
3597   if (tables_->AddSymbol(full_name, symbol)) {
3598     if (!file_tables_->AddAliasUnderParent(parent, name, symbol)) {
3599       GOOGLE_LOG(DFATAL) << "\"" << full_name << "\" not previously defined in "
3600                      "symbols_by_name_, but was defined in symbols_by_parent_; "
3601                      "this shouldn't be possible.";
3602       return false;
3603     }
3604     return true;
3605   } else {
3606     const FileDescriptor* other_file = tables_->FindSymbol(full_name).GetFile();
3607     if (other_file == file_) {
3608       string::size_type dot_pos = full_name.find_last_of('.');
3609       if (dot_pos == string::npos) {
3610         AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
3611                  "\"" + full_name + "\" is already defined.");
3612       } else {
3613         AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
3614                  "\"" + full_name.substr(dot_pos + 1) +
3615                  "\" is already defined in \"" +
3616                  full_name.substr(0, dot_pos) + "\".");
3617       }
3618     } else {
3619       // Symbol seems to have been defined in a different file.
3620       AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
3621                "\"" + full_name + "\" is already defined in file \"" +
3622                other_file->name() + "\".");
3623     }
3624     return false;
3625   }
3626 }
3627 
AddPackage(const string & name,const Message & proto,const FileDescriptor * file)3628 void DescriptorBuilder::AddPackage(
3629     const string& name, const Message& proto, const FileDescriptor* file) {
3630   if (tables_->AddSymbol(name, Symbol(file))) {
3631     // Success.  Also add parent package, if any.
3632     string::size_type dot_pos = name.find_last_of('.');
3633     if (dot_pos == string::npos) {
3634       // No parents.
3635       ValidateSymbolName(name, name, proto);
3636     } else {
3637       // Has parent.
3638       string* parent_name = tables_->AllocateString(name.substr(0, dot_pos));
3639       AddPackage(*parent_name, proto, file);
3640       ValidateSymbolName(name.substr(dot_pos + 1), name, proto);
3641     }
3642   } else {
3643     Symbol existing_symbol = tables_->FindSymbol(name);
3644     // It's OK to redefine a package.
3645     if (existing_symbol.type != Symbol::PACKAGE) {
3646       // Symbol seems to have been defined in a different file.
3647       AddError(name, proto, DescriptorPool::ErrorCollector::NAME,
3648                "\"" + name + "\" is already defined (as something other than "
3649                "a package) in file \"" + existing_symbol.GetFile()->name() +
3650                "\".");
3651     }
3652   }
3653 }
3654 
ValidateSymbolName(const string & name,const string & full_name,const Message & proto)3655 void DescriptorBuilder::ValidateSymbolName(
3656     const string& name, const string& full_name, const Message& proto) {
3657   if (name.empty()) {
3658     AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
3659              "Missing name.");
3660   } else {
3661     for (int i = 0; i < name.size(); i++) {
3662       // I don't trust isalnum() due to locales.  :(
3663       if ((name[i] < 'a' || 'z' < name[i]) &&
3664           (name[i] < 'A' || 'Z' < name[i]) &&
3665           (name[i] < '0' || '9' < name[i]) &&
3666           (name[i] != '_')) {
3667         AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
3668                  "\"" + name + "\" is not a valid identifier.");
3669       }
3670     }
3671   }
3672 }
3673 
ValidateQualifiedName(const string & name)3674 bool DescriptorBuilder::ValidateQualifiedName(const string& name) {
3675   bool last_was_period = false;
3676 
3677   for (int i = 0; i < name.size(); i++) {
3678     // I don't trust isalnum() due to locales.  :(
3679     if (('a' <= name[i] && name[i] <= 'z') ||
3680         ('A' <= name[i] && name[i] <= 'Z') ||
3681         ('0' <= name[i] && name[i] <= '9') ||
3682         (name[i] == '_')) {
3683       last_was_period = false;
3684     } else if (name[i] == '.') {
3685       if (last_was_period) return false;
3686       last_was_period = true;
3687     } else {
3688       return false;
3689     }
3690   }
3691 
3692   return !name.empty() && !last_was_period;
3693 }
3694 
3695 // -------------------------------------------------------------------
3696 
3697 // This generic implementation is good for all descriptors except
3698 // FileDescriptor.
AllocateOptions(const typename DescriptorT::OptionsType & orig_options,DescriptorT * descriptor)3699 template<class DescriptorT> void DescriptorBuilder::AllocateOptions(
3700     const typename DescriptorT::OptionsType& orig_options,
3701     DescriptorT* descriptor) {
3702   AllocateOptionsImpl(descriptor->full_name(), descriptor->full_name(),
3703                       orig_options, descriptor);
3704 }
3705 
3706 // We specialize for FileDescriptor.
AllocateOptions(const FileOptions & orig_options,FileDescriptor * descriptor)3707 void DescriptorBuilder::AllocateOptions(const FileOptions& orig_options,
3708                                         FileDescriptor* descriptor) {
3709   // We add the dummy token so that LookupSymbol does the right thing.
3710   AllocateOptionsImpl(descriptor->package() + ".dummy", descriptor->name(),
3711                       orig_options, descriptor);
3712 }
3713 
AllocateOptionsImpl(const string & name_scope,const string & element_name,const typename DescriptorT::OptionsType & orig_options,DescriptorT * descriptor)3714 template<class DescriptorT> void DescriptorBuilder::AllocateOptionsImpl(
3715     const string& name_scope,
3716     const string& element_name,
3717     const typename DescriptorT::OptionsType& orig_options,
3718     DescriptorT* descriptor) {
3719   // We need to use a dummy pointer to work around a bug in older versions of
3720   // GCC.  Otherwise, the following two lines could be replaced with:
3721   //   typename DescriptorT::OptionsType* options =
3722   //       tables_->AllocateMessage<typename DescriptorT::OptionsType>();
3723   typename DescriptorT::OptionsType* const dummy = NULL;
3724   typename DescriptorT::OptionsType* options = tables_->AllocateMessage(dummy);
3725   // Avoid using MergeFrom()/CopyFrom() in this class to make it -fno-rtti
3726   // friendly. Without RTTI, MergeFrom() and CopyFrom() will fallback to the
3727   // reflection based method, which requires the Descriptor. However, we are in
3728   // the middle of building the descriptors, thus the deadlock.
3729   options->ParseFromString(orig_options.SerializeAsString());
3730   descriptor->options_ = options;
3731 
3732   // Don't add to options_to_interpret_ unless there were uninterpreted
3733   // options.  This not only avoids unnecessary work, but prevents a
3734   // bootstrapping problem when building descriptors for descriptor.proto.
3735   // descriptor.proto does not contain any uninterpreted options, but
3736   // attempting to interpret options anyway will cause
3737   // OptionsType::GetDescriptor() to be called which may then deadlock since
3738   // we're still trying to build it.
3739   if (options->uninterpreted_option_size() > 0) {
3740     options_to_interpret_.push_back(
3741         OptionsToInterpret(name_scope, element_name, &orig_options, options));
3742   }
3743 }
3744 
3745 
3746 // A common pattern:  We want to convert a repeated field in the descriptor
3747 // to an array of values, calling some method to build each value.
3748 #define BUILD_ARRAY(INPUT, OUTPUT, NAME, METHOD, PARENT)             \
3749   OUTPUT->NAME##_count_ = INPUT.NAME##_size();                       \
3750   AllocateArray(INPUT.NAME##_size(), &OUTPUT->NAME##s_);             \
3751   for (int i = 0; i < INPUT.NAME##_size(); i++) {                    \
3752     METHOD(INPUT.NAME(i), PARENT, OUTPUT->NAME##s_ + i);             \
3753   }
3754 
AddRecursiveImportError(const FileDescriptorProto & proto,int from_here)3755 void DescriptorBuilder::AddRecursiveImportError(
3756     const FileDescriptorProto& proto, int from_here) {
3757   string error_message("File recursively imports itself: ");
3758   for (int i = from_here; i < tables_->pending_files_.size(); i++) {
3759     error_message.append(tables_->pending_files_[i]);
3760     error_message.append(" -> ");
3761   }
3762   error_message.append(proto.name());
3763 
3764   AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
3765            error_message);
3766 }
3767 
AddTwiceListedError(const FileDescriptorProto & proto,int index)3768 void DescriptorBuilder::AddTwiceListedError(const FileDescriptorProto& proto,
3769                                             int index) {
3770   AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
3771            "Import \"" + proto.dependency(index) + "\" was listed twice.");
3772 }
3773 
AddImportError(const FileDescriptorProto & proto,int index)3774 void DescriptorBuilder::AddImportError(const FileDescriptorProto& proto,
3775                                        int index) {
3776   string message;
3777   if (pool_->fallback_database_ == NULL) {
3778     message = "Import \"" + proto.dependency(index) +
3779               "\" has not been loaded.";
3780   } else {
3781     message = "Import \"" + proto.dependency(index) +
3782               "\" was not found or had errors.";
3783   }
3784   AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER, message);
3785 }
3786 
ExistingFileMatchesProto(const FileDescriptor * existing_file,const FileDescriptorProto & proto)3787 static bool ExistingFileMatchesProto(const FileDescriptor* existing_file,
3788                                      const FileDescriptorProto& proto) {
3789   FileDescriptorProto existing_proto;
3790   existing_file->CopyTo(&existing_proto);
3791   // TODO(liujisi): Remove it when CopyTo supports copying syntax params when
3792   // syntax="proto2".
3793   if (existing_file->syntax() == FileDescriptor::SYNTAX_PROTO2 &&
3794       proto.has_syntax()) {
3795     existing_proto.set_syntax(
3796         existing_file->SyntaxName(existing_file->syntax()));
3797   }
3798 
3799   return existing_proto.SerializeAsString() == proto.SerializeAsString();
3800 }
3801 
BuildFile(const FileDescriptorProto & proto)3802 const FileDescriptor* DescriptorBuilder::BuildFile(
3803     const FileDescriptorProto& proto) {
3804   filename_ = proto.name();
3805 
3806   // Check if the file already exists and is identical to the one being built.
3807   // Note:  This only works if the input is canonical -- that is, it
3808   //   fully-qualifies all type names, has no UninterpretedOptions, etc.
3809   //   This is fine, because this idempotency "feature" really only exists to
3810   //   accommodate one hack in the proto1->proto2 migration layer.
3811   const FileDescriptor* existing_file = tables_->FindFile(filename_);
3812   if (existing_file != NULL) {
3813     // File already in pool.  Compare the existing one to the input.
3814     if (ExistingFileMatchesProto(existing_file, proto)) {
3815       // They're identical.  Return the existing descriptor.
3816       return existing_file;
3817     }
3818 
3819     // Not a match.  The error will be detected and handled later.
3820   }
3821 
3822   // Check to see if this file is already on the pending files list.
3823   // TODO(kenton):  Allow recursive imports?  It may not work with some
3824   //   (most?) programming languages.  E.g., in C++, a forward declaration
3825   //   of a type is not sufficient to allow it to be used even in a
3826   //   generated header file due to inlining.  This could perhaps be
3827   //   worked around using tricks involving inserting #include statements
3828   //   mid-file, but that's pretty ugly, and I'm pretty sure there are
3829   //   some languages out there that do not allow recursive dependencies
3830   //   at all.
3831   for (int i = 0; i < tables_->pending_files_.size(); i++) {
3832     if (tables_->pending_files_[i] == proto.name()) {
3833       AddRecursiveImportError(proto, i);
3834       return NULL;
3835     }
3836   }
3837 
3838   // If we have a fallback_database_, attempt to load all dependencies now,
3839   // before checkpointing tables_.  This avoids confusion with recursive
3840   // checkpoints.
3841   if (pool_->fallback_database_ != NULL) {
3842     tables_->pending_files_.push_back(proto.name());
3843     for (int i = 0; i < proto.dependency_size(); i++) {
3844       if (tables_->FindFile(proto.dependency(i)) == NULL &&
3845           (pool_->underlay_ == NULL ||
3846            pool_->underlay_->FindFileByName(proto.dependency(i)) == NULL)) {
3847         // We don't care what this returns since we'll find out below anyway.
3848         pool_->TryFindFileInFallbackDatabase(proto.dependency(i));
3849       }
3850     }
3851     tables_->pending_files_.pop_back();
3852   }
3853   return BuildFileImpl(proto);
3854 }
3855 
BuildFileImpl(const FileDescriptorProto & proto)3856 const FileDescriptor* DescriptorBuilder::BuildFileImpl(
3857     const FileDescriptorProto& proto) {
3858   // Checkpoint the tables so that we can roll back if something goes wrong.
3859   tables_->AddCheckpoint();
3860 
3861   FileDescriptor* result = tables_->Allocate<FileDescriptor>();
3862   file_ = result;
3863 
3864   result->is_placeholder_ = false;
3865   if (proto.has_source_code_info()) {
3866     SourceCodeInfo *info = tables_->AllocateMessage<SourceCodeInfo>();
3867     info->CopyFrom(proto.source_code_info());
3868     result->source_code_info_ = info;
3869   } else {
3870     result->source_code_info_ = &SourceCodeInfo::default_instance();
3871   }
3872 
3873   file_tables_ = tables_->AllocateFileTables();
3874   file_->tables_ = file_tables_;
3875 
3876   if (!proto.has_name()) {
3877     AddError("", proto, DescriptorPool::ErrorCollector::OTHER,
3878              "Missing field: FileDescriptorProto.name.");
3879   }
3880 
3881   // TODO(liujisi): Report error when the syntax is empty after all the protos
3882   // have added the syntax statement.
3883   if (proto.syntax().empty() || proto.syntax() == "proto2") {
3884     file_->syntax_ = FileDescriptor::SYNTAX_PROTO2;
3885   } else if (proto.syntax() == "proto3") {
3886     file_->syntax_ = FileDescriptor::SYNTAX_PROTO3;
3887   } else {
3888     file_->syntax_ = FileDescriptor::SYNTAX_UNKNOWN;
3889     AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
3890              "Unrecognized syntax: " + proto.syntax());
3891   }
3892 
3893   result->name_ = tables_->AllocateString(proto.name());
3894   if (proto.has_package()) {
3895     result->package_ = tables_->AllocateString(proto.package());
3896   } else {
3897     // We cannot rely on proto.package() returning a valid string if
3898     // proto.has_package() is false, because we might be running at static
3899     // initialization time, in which case default values have not yet been
3900     // initialized.
3901     result->package_ = tables_->AllocateString("");
3902   }
3903   result->pool_ = pool_;
3904 
3905   // Add to tables.
3906   if (!tables_->AddFile(result)) {
3907     AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
3908              "A file with this name is already in the pool.");
3909     // Bail out early so that if this is actually the exact same file, we
3910     // don't end up reporting that every single symbol is already defined.
3911     tables_->RollbackToLastCheckpoint();
3912     return NULL;
3913   }
3914   if (!result->package().empty()) {
3915     AddPackage(result->package(), proto, result);
3916   }
3917 
3918   // Make sure all dependencies are loaded.
3919   set<string> seen_dependencies;
3920   result->dependency_count_ = proto.dependency_size();
3921   result->dependencies_ =
3922     tables_->AllocateArray<const FileDescriptor*>(proto.dependency_size());
3923   unused_dependency_.clear();
3924   set<int> weak_deps;
3925   for (int i = 0; i < proto.weak_dependency_size(); ++i) {
3926     weak_deps.insert(proto.weak_dependency(i));
3927   }
3928   for (int i = 0; i < proto.dependency_size(); i++) {
3929     if (!seen_dependencies.insert(proto.dependency(i)).second) {
3930       AddTwiceListedError(proto, i);
3931     }
3932 
3933     const FileDescriptor* dependency = tables_->FindFile(proto.dependency(i));
3934     if (dependency == NULL && pool_->underlay_ != NULL) {
3935       dependency = pool_->underlay_->FindFileByName(proto.dependency(i));
3936     }
3937 
3938     if (dependency == NULL) {
3939       if (pool_->allow_unknown_ ||
3940           (!pool_->enforce_weak_ && weak_deps.find(i) != weak_deps.end())) {
3941         dependency = NewPlaceholderFile(proto.dependency(i));
3942       } else {
3943         AddImportError(proto, i);
3944       }
3945     } else {
3946       // Add to unused_dependency_ to track unused imported files.
3947       // Note: do not track unused imported files for public import.
3948       if (pool_->enforce_dependencies_ &&
3949           (pool_->unused_import_track_files_.find(proto.name()) !=
3950            pool_->unused_import_track_files_.end()) &&
3951           (dependency->public_dependency_count() == 0)) {
3952         unused_dependency_.insert(dependency);
3953       }
3954     }
3955 
3956     result->dependencies_[i] = dependency;
3957   }
3958 
3959   // Check public dependencies.
3960   int public_dependency_count = 0;
3961   result->public_dependencies_ = tables_->AllocateArray<int>(
3962       proto.public_dependency_size());
3963   for (int i = 0; i < proto.public_dependency_size(); i++) {
3964     // Only put valid public dependency indexes.
3965     int index = proto.public_dependency(i);
3966     if (index >= 0 && index < proto.dependency_size()) {
3967       result->public_dependencies_[public_dependency_count++] = index;
3968       // Do not track unused imported files for public import.
3969       unused_dependency_.erase(result->dependency(index));
3970     } else {
3971       AddError(proto.name(), proto,
3972                DescriptorPool::ErrorCollector::OTHER,
3973                "Invalid public dependency index.");
3974     }
3975   }
3976   result->public_dependency_count_ = public_dependency_count;
3977 
3978   // Build dependency set
3979   dependencies_.clear();
3980   for (int i = 0; i < result->dependency_count(); i++) {
3981     RecordPublicDependencies(result->dependency(i));
3982   }
3983 
3984   // Check weak dependencies.
3985   int weak_dependency_count = 0;
3986   result->weak_dependencies_ = tables_->AllocateArray<int>(
3987       proto.weak_dependency_size());
3988   for (int i = 0; i < proto.weak_dependency_size(); i++) {
3989     int index = proto.weak_dependency(i);
3990     if (index >= 0 && index < proto.dependency_size()) {
3991       result->weak_dependencies_[weak_dependency_count++] = index;
3992     } else {
3993       AddError(proto.name(), proto,
3994                DescriptorPool::ErrorCollector::OTHER,
3995                "Invalid weak dependency index.");
3996     }
3997   }
3998   result->weak_dependency_count_ = weak_dependency_count;
3999 
4000   // Convert children.
4001   BUILD_ARRAY(proto, result, message_type, BuildMessage  , NULL);
4002   BUILD_ARRAY(proto, result, enum_type   , BuildEnum     , NULL);
4003   BUILD_ARRAY(proto, result, service     , BuildService  , NULL);
4004   BUILD_ARRAY(proto, result, extension   , BuildExtension, NULL);
4005 
4006   // Copy options.
4007   if (!proto.has_options()) {
4008     result->options_ = NULL;  // Will set to default_instance later.
4009   } else {
4010     AllocateOptions(proto.options(), result);
4011   }
4012 
4013   // Note that the following steps must occur in exactly the specified order.
4014 
4015   // Cross-link.
4016   CrossLinkFile(result, proto);
4017 
4018   // Interpret any remaining uninterpreted options gathered into
4019   // options_to_interpret_ during descriptor building.  Cross-linking has made
4020   // extension options known, so all interpretations should now succeed.
4021   if (!had_errors_) {
4022     OptionInterpreter option_interpreter(this);
4023     for (vector<OptionsToInterpret>::iterator iter =
4024              options_to_interpret_.begin();
4025          iter != options_to_interpret_.end(); ++iter) {
4026       option_interpreter.InterpretOptions(&(*iter));
4027     }
4028     options_to_interpret_.clear();
4029   }
4030 
4031   // Validate options.
4032   if (!had_errors_) {
4033     ValidateFileOptions(result, proto);
4034   }
4035 
4036   // Additional naming conflict check for map entry types. Only need to check
4037   // this if there are already errors.
4038   if (had_errors_) {
4039     for (int i = 0; i < proto.message_type_size(); ++i) {
4040       DetectMapConflicts(result->message_type(i), proto.message_type(i));
4041     }
4042   }
4043 
4044 
4045   if (!unused_dependency_.empty()) {
4046     LogUnusedDependency(proto, result);
4047   }
4048 
4049   if (had_errors_) {
4050     tables_->RollbackToLastCheckpoint();
4051     return NULL;
4052   } else {
4053     tables_->ClearLastCheckpoint();
4054     return result;
4055   }
4056 }
4057 
BuildMessage(const DescriptorProto & proto,const Descriptor * parent,Descriptor * result)4058 void DescriptorBuilder::BuildMessage(const DescriptorProto& proto,
4059                                      const Descriptor* parent,
4060                                      Descriptor* result) {
4061   const string& scope = (parent == NULL) ?
4062     file_->package() : parent->full_name();
4063   string* full_name = tables_->AllocateString(scope);
4064   if (!full_name->empty()) full_name->append(1, '.');
4065   full_name->append(proto.name());
4066 
4067   ValidateSymbolName(proto.name(), *full_name, proto);
4068 
4069   result->name_            = tables_->AllocateString(proto.name());
4070   result->full_name_       = full_name;
4071   result->file_            = file_;
4072   result->containing_type_ = parent;
4073   result->is_placeholder_  = false;
4074   result->is_unqualified_placeholder_ = false;
4075 
4076   // Build oneofs first so that fields and extension ranges can refer to them.
4077   BUILD_ARRAY(proto, result, oneof_decl     , BuildOneof         , result);
4078   BUILD_ARRAY(proto, result, field          , BuildField         , result);
4079   BUILD_ARRAY(proto, result, nested_type    , BuildMessage       , result);
4080   BUILD_ARRAY(proto, result, enum_type      , BuildEnum          , result);
4081   BUILD_ARRAY(proto, result, extension_range, BuildExtensionRange, result);
4082   BUILD_ARRAY(proto, result, extension      , BuildExtension     , result);
4083   BUILD_ARRAY(proto, result, reserved_range , BuildReservedRange , result);
4084 
4085   // Copy reserved names.
4086   int reserved_name_count = proto.reserved_name_size();
4087   result->reserved_name_count_ = reserved_name_count;
4088   result->reserved_names_ =
4089       tables_->AllocateArray<const string*>(reserved_name_count);
4090   for (int i = 0; i < reserved_name_count; ++i) {
4091     result->reserved_names_[i] =
4092         tables_->AllocateString(proto.reserved_name(i));
4093   }
4094 
4095   // Copy options.
4096   if (!proto.has_options()) {
4097     result->options_ = NULL;  // Will set to default_instance later.
4098   } else {
4099     AllocateOptions(proto.options(), result);
4100   }
4101 
4102   AddSymbol(result->full_name(), parent, result->name(),
4103             proto, Symbol(result));
4104 
4105   for (int i = 0; i < proto.reserved_range_size(); i++) {
4106     const DescriptorProto_ReservedRange& range1 = proto.reserved_range(i);
4107     for (int j = i + 1; j < proto.reserved_range_size(); j++) {
4108       const DescriptorProto_ReservedRange& range2 = proto.reserved_range(j);
4109       if (range1.end() > range2.start() && range2.end() > range1.start()) {
4110         AddError(result->full_name(), proto.reserved_range(i),
4111                  DescriptorPool::ErrorCollector::NUMBER,
4112                  strings::Substitute("Reserved range $0 to $1 overlaps with "
4113                                      "already-defined range $2 to $3.",
4114                                      range2.start(), range2.end() - 1,
4115                                      range1.start(), range1.end() - 1));
4116       }
4117     }
4118   }
4119 
4120   hash_set<string> reserved_name_set;
4121   for (int i = 0; i < proto.reserved_name_size(); i++) {
4122     const string& name = proto.reserved_name(i);
4123     if (reserved_name_set.find(name) == reserved_name_set.end()) {
4124       reserved_name_set.insert(name);
4125     } else {
4126       AddError(name, proto, DescriptorPool::ErrorCollector::NAME,
4127                strings::Substitute(
4128                  "Field name \"$0\" is reserved multiple times.",
4129                  name));
4130     }
4131   }
4132 
4133   for (int i = 0; i < result->field_count(); i++) {
4134     const FieldDescriptor* field = result->field(i);
4135     for (int j = 0; j < result->extension_range_count(); j++) {
4136       const Descriptor::ExtensionRange* range = result->extension_range(j);
4137       if (range->start <= field->number() && field->number() < range->end) {
4138         AddError(field->full_name(), proto.extension_range(j),
4139                  DescriptorPool::ErrorCollector::NUMBER,
4140                  strings::Substitute(
4141                    "Extension range $0 to $1 includes field \"$2\" ($3).",
4142                    range->start, range->end - 1,
4143                    field->name(), field->number()));
4144       }
4145     }
4146     for (int j = 0; j < result->reserved_range_count(); j++) {
4147       const Descriptor::ReservedRange* range = result->reserved_range(j);
4148       if (range->start <= field->number() && field->number() < range->end) {
4149         AddError(field->full_name(), proto.reserved_range(j),
4150                  DescriptorPool::ErrorCollector::NUMBER,
4151                  strings::Substitute(
4152                    "Field \"$0\" uses reserved number $1.",
4153                    field->name(), field->number()));
4154       }
4155     }
4156     if (reserved_name_set.find(field->name()) != reserved_name_set.end()) {
4157       AddError(field->full_name(), proto.field(i),
4158                DescriptorPool::ErrorCollector::NAME,
4159                strings::Substitute(
4160                  "Field name \"$0\" is reserved.", field->name()));
4161     }
4162   }
4163 
4164   // Check that extension ranges don't overlap and don't include
4165   // reserved field numbers.
4166   for (int i = 0; i < result->extension_range_count(); i++) {
4167     const Descriptor::ExtensionRange* range1 = result->extension_range(i);
4168     for (int j = 0; j < result->reserved_range_count(); j++) {
4169       const Descriptor::ReservedRange* range2 = result->reserved_range(j);
4170       if (range1->end > range2->start && range2->end > range1->start) {
4171         AddError(result->full_name(), proto.extension_range(j),
4172                  DescriptorPool::ErrorCollector::NUMBER,
4173                  strings::Substitute("Extension range $0 to $1 overlaps with "
4174                                      "reserved range $2 to $3.",
4175                                      range1->start, range1->end - 1,
4176                                      range2->start, range2->end - 1));
4177       }
4178     }
4179     for (int j = i + 1; j < result->extension_range_count(); j++) {
4180       const Descriptor::ExtensionRange* range2 = result->extension_range(j);
4181       if (range1->end > range2->start && range2->end > range1->start) {
4182         AddError(result->full_name(), proto.extension_range(j),
4183                  DescriptorPool::ErrorCollector::NUMBER,
4184                  strings::Substitute("Extension range $0 to $1 overlaps with "
4185                                      "already-defined range $2 to $3.",
4186                                      range2->start, range2->end - 1,
4187                                      range1->start, range1->end - 1));
4188       }
4189     }
4190   }
4191 }
4192 
4193 
BuildFieldOrExtension(const FieldDescriptorProto & proto,const Descriptor * parent,FieldDescriptor * result,bool is_extension)4194 void DescriptorBuilder::BuildFieldOrExtension(const FieldDescriptorProto& proto,
4195                                               const Descriptor* parent,
4196                                               FieldDescriptor* result,
4197                                               bool is_extension) {
4198   const string& scope = (parent == NULL) ?
4199     file_->package() : parent->full_name();
4200   string* full_name = tables_->AllocateString(scope);
4201   if (!full_name->empty()) full_name->append(1, '.');
4202   full_name->append(proto.name());
4203 
4204   ValidateSymbolName(proto.name(), *full_name, proto);
4205 
4206   result->name_         = tables_->AllocateString(proto.name());
4207   result->full_name_    = full_name;
4208   result->file_         = file_;
4209   result->number_       = proto.number();
4210   result->is_extension_ = is_extension;
4211 
4212   // If .proto files follow the style guide then the name should already be
4213   // lower-cased.  If that's the case we can just reuse the string we already
4214   // allocated rather than allocate a new one.
4215   string lowercase_name(proto.name());
4216   LowerString(&lowercase_name);
4217   if (lowercase_name == proto.name()) {
4218     result->lowercase_name_ = result->name_;
4219   } else {
4220     result->lowercase_name_ = tables_->AllocateString(lowercase_name);
4221   }
4222 
4223   // Don't bother with the above optimization for camel-case names since
4224   // .proto files that follow the guide shouldn't be using names in this
4225   // format, so the optimization wouldn't help much.
4226   result->camelcase_name_ =
4227       tables_->AllocateString(ToCamelCase(proto.name(),
4228                                           /* lower_first = */ true));
4229 
4230   if (proto.has_json_name()) {
4231     result->has_json_name_ = true;
4232     result->json_name_ = tables_->AllocateString(proto.json_name());
4233   } else {
4234     result->has_json_name_ = false;
4235     result->json_name_ = result->camelcase_name_;
4236   }
4237 
4238   // Some compilers do not allow static_cast directly between two enum types,
4239   // so we must cast to int first.
4240   result->type_  = static_cast<FieldDescriptor::Type>(
4241                      implicit_cast<int>(proto.type()));
4242   result->label_ = static_cast<FieldDescriptor::Label>(
4243                      implicit_cast<int>(proto.label()));
4244 
4245   // An extension cannot have a required field (b/13365836).
4246   if (result->is_extension_ &&
4247       result->label_ == FieldDescriptor::LABEL_REQUIRED) {
4248     AddError(result->full_name(), proto,
4249              // Error location `TYPE`: we would really like to indicate
4250              // `LABEL`, but the `ErrorLocation` enum has no entry for this, and
4251              // we don't necessarily know about all implementations of the
4252              // `ErrorCollector` interface to extend them to handle the new
4253              // error location type properly.
4254              DescriptorPool::ErrorCollector::TYPE,
4255              "Message extensions cannot have required fields.");
4256   }
4257 
4258   // Some of these may be filled in when cross-linking.
4259   result->containing_type_ = NULL;
4260   result->extension_scope_ = NULL;
4261   result->message_type_ = NULL;
4262   result->enum_type_ = NULL;
4263 
4264   result->has_default_value_ = proto.has_default_value();
4265   if (proto.has_default_value() && result->is_repeated()) {
4266     AddError(result->full_name(), proto,
4267              DescriptorPool::ErrorCollector::DEFAULT_VALUE,
4268              "Repeated fields can't have default values.");
4269   }
4270 
4271   if (proto.has_type()) {
4272     if (proto.has_default_value()) {
4273       char* end_pos = NULL;
4274       switch (result->cpp_type()) {
4275         case FieldDescriptor::CPPTYPE_INT32:
4276           result->default_value_int32_ =
4277             strtol(proto.default_value().c_str(), &end_pos, 0);
4278           break;
4279         case FieldDescriptor::CPPTYPE_INT64:
4280           result->default_value_int64_ =
4281             strto64(proto.default_value().c_str(), &end_pos, 0);
4282           break;
4283         case FieldDescriptor::CPPTYPE_UINT32:
4284           result->default_value_uint32_ =
4285             strtoul(proto.default_value().c_str(), &end_pos, 0);
4286           break;
4287         case FieldDescriptor::CPPTYPE_UINT64:
4288           result->default_value_uint64_ =
4289             strtou64(proto.default_value().c_str(), &end_pos, 0);
4290           break;
4291         case FieldDescriptor::CPPTYPE_FLOAT:
4292           if (proto.default_value() == "inf") {
4293             result->default_value_float_ = numeric_limits<float>::infinity();
4294           } else if (proto.default_value() == "-inf") {
4295             result->default_value_float_ = -numeric_limits<float>::infinity();
4296           } else if (proto.default_value() == "nan") {
4297             result->default_value_float_ = numeric_limits<float>::quiet_NaN();
4298           } else  {
4299             result->default_value_float_ = io::SafeDoubleToFloat(
4300                 io::NoLocaleStrtod(proto.default_value().c_str(), &end_pos));
4301           }
4302           break;
4303         case FieldDescriptor::CPPTYPE_DOUBLE:
4304           if (proto.default_value() == "inf") {
4305             result->default_value_double_ = numeric_limits<double>::infinity();
4306           } else if (proto.default_value() == "-inf") {
4307             result->default_value_double_ = -numeric_limits<double>::infinity();
4308           } else if (proto.default_value() == "nan") {
4309             result->default_value_double_ = numeric_limits<double>::quiet_NaN();
4310           } else  {
4311             result->default_value_double_ =
4312                 io::NoLocaleStrtod(proto.default_value().c_str(), &end_pos);
4313           }
4314           break;
4315         case FieldDescriptor::CPPTYPE_BOOL:
4316           if (proto.default_value() == "true") {
4317             result->default_value_bool_ = true;
4318           } else if (proto.default_value() == "false") {
4319             result->default_value_bool_ = false;
4320           } else {
4321             AddError(result->full_name(), proto,
4322                      DescriptorPool::ErrorCollector::DEFAULT_VALUE,
4323                      "Boolean default must be true or false.");
4324           }
4325           break;
4326         case FieldDescriptor::CPPTYPE_ENUM:
4327           // This will be filled in when cross-linking.
4328           result->default_value_enum_ = NULL;
4329           break;
4330         case FieldDescriptor::CPPTYPE_STRING:
4331           if (result->type() == FieldDescriptor::TYPE_BYTES) {
4332             result->default_value_string_ = tables_->AllocateString(
4333               UnescapeCEscapeString(proto.default_value()));
4334           } else {
4335             result->default_value_string_ =
4336                 tables_->AllocateString(proto.default_value());
4337           }
4338           break;
4339         case FieldDescriptor::CPPTYPE_MESSAGE:
4340           AddError(result->full_name(), proto,
4341                    DescriptorPool::ErrorCollector::DEFAULT_VALUE,
4342                    "Messages can't have default values.");
4343           result->has_default_value_ = false;
4344           break;
4345       }
4346 
4347       if (end_pos != NULL) {
4348         // end_pos is only set non-NULL by the parsers for numeric types, above.
4349         // This checks that the default was non-empty and had no extra junk
4350         // after the end of the number.
4351         if (proto.default_value().empty() || *end_pos != '\0') {
4352           AddError(result->full_name(), proto,
4353                    DescriptorPool::ErrorCollector::DEFAULT_VALUE,
4354                    "Couldn't parse default value \"" + proto.default_value() +
4355                    "\".");
4356         }
4357       }
4358     } else {
4359       // No explicit default value
4360       switch (result->cpp_type()) {
4361         case FieldDescriptor::CPPTYPE_INT32:
4362           result->default_value_int32_ = 0;
4363           break;
4364         case FieldDescriptor::CPPTYPE_INT64:
4365           result->default_value_int64_ = 0;
4366           break;
4367         case FieldDescriptor::CPPTYPE_UINT32:
4368           result->default_value_uint32_ = 0;
4369           break;
4370         case FieldDescriptor::CPPTYPE_UINT64:
4371           result->default_value_uint64_ = 0;
4372           break;
4373         case FieldDescriptor::CPPTYPE_FLOAT:
4374           result->default_value_float_ = 0.0f;
4375           break;
4376         case FieldDescriptor::CPPTYPE_DOUBLE:
4377           result->default_value_double_ = 0.0;
4378           break;
4379         case FieldDescriptor::CPPTYPE_BOOL:
4380           result->default_value_bool_ = false;
4381           break;
4382         case FieldDescriptor::CPPTYPE_ENUM:
4383           // This will be filled in when cross-linking.
4384           result->default_value_enum_ = NULL;
4385           break;
4386         case FieldDescriptor::CPPTYPE_STRING:
4387           result->default_value_string_ = &internal::GetEmptyString();
4388           break;
4389         case FieldDescriptor::CPPTYPE_MESSAGE:
4390           break;
4391       }
4392     }
4393   }
4394 
4395   if (result->number() <= 0) {
4396     AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
4397              "Field numbers must be positive integers.");
4398   } else if (!is_extension && result->number() > FieldDescriptor::kMaxNumber) {
4399     // Only validate that the number is within the valid field range if it is
4400     // not an extension. Since extension numbers are validated with the
4401     // extendee's valid set of extension numbers, and those are in turn
4402     // validated against the max allowed number, the check is unnecessary for
4403     // extension fields.
4404     // This avoids cross-linking issues that arise when attempting to check if
4405     // the extendee is a message_set_wire_format message, which has a higher max
4406     // on extension numbers.
4407     AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
4408              strings::Substitute("Field numbers cannot be greater than $0.",
4409                                  FieldDescriptor::kMaxNumber));
4410   } else if (result->number() >= FieldDescriptor::kFirstReservedNumber &&
4411              result->number() <= FieldDescriptor::kLastReservedNumber) {
4412     AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
4413              strings::Substitute(
4414                "Field numbers $0 through $1 are reserved for the protocol "
4415                "buffer library implementation.",
4416                FieldDescriptor::kFirstReservedNumber,
4417                FieldDescriptor::kLastReservedNumber));
4418   }
4419 
4420   if (is_extension) {
4421     if (!proto.has_extendee()) {
4422       AddError(result->full_name(), proto,
4423                DescriptorPool::ErrorCollector::EXTENDEE,
4424                "FieldDescriptorProto.extendee not set for extension field.");
4425     }
4426 
4427     result->extension_scope_ = parent;
4428 
4429     if (proto.has_oneof_index()) {
4430       AddError(result->full_name(), proto,
4431                DescriptorPool::ErrorCollector::OTHER,
4432                "FieldDescriptorProto.oneof_index should not be set for "
4433                "extensions.");
4434     }
4435 
4436     // Fill in later (maybe).
4437     result->containing_oneof_ = NULL;
4438   } else {
4439     if (proto.has_extendee()) {
4440       AddError(result->full_name(), proto,
4441                DescriptorPool::ErrorCollector::EXTENDEE,
4442                "FieldDescriptorProto.extendee set for non-extension field.");
4443     }
4444 
4445     result->containing_type_ = parent;
4446 
4447     if (proto.has_oneof_index()) {
4448       if (proto.oneof_index() < 0 ||
4449           proto.oneof_index() >= parent->oneof_decl_count()) {
4450         AddError(result->full_name(), proto,
4451                  DescriptorPool::ErrorCollector::OTHER,
4452                  strings::Substitute("FieldDescriptorProto.oneof_index $0 is "
4453                                      "out of range for type \"$1\".",
4454                                      proto.oneof_index(),
4455                                      parent->name()));
4456         result->containing_oneof_ = NULL;
4457       } else {
4458         result->containing_oneof_ = parent->oneof_decl(proto.oneof_index());
4459       }
4460     } else {
4461       result->containing_oneof_ = NULL;
4462     }
4463   }
4464 
4465   // Copy options.
4466   if (!proto.has_options()) {
4467     result->options_ = NULL;  // Will set to default_instance later.
4468   } else {
4469     AllocateOptions(proto.options(), result);
4470   }
4471 
4472 
4473   AddSymbol(result->full_name(), parent, result->name(),
4474             proto, Symbol(result));
4475 }
4476 
BuildExtensionRange(const DescriptorProto::ExtensionRange & proto,const Descriptor * parent,Descriptor::ExtensionRange * result)4477 void DescriptorBuilder::BuildExtensionRange(
4478     const DescriptorProto::ExtensionRange& proto,
4479     const Descriptor* parent,
4480     Descriptor::ExtensionRange* result) {
4481   result->start = proto.start();
4482   result->end = proto.end();
4483   if (result->start <= 0) {
4484     AddError(parent->full_name(), proto,
4485              DescriptorPool::ErrorCollector::NUMBER,
4486              "Extension numbers must be positive integers.");
4487   }
4488 
4489   // Checking of the upper bound of the extension range is deferred until after
4490   // options interpreting. This allows messages with message_set_wire_format to
4491   // have extensions beyond FieldDescriptor::kMaxNumber, since the extension
4492   // numbers are actually used as int32s in the message_set_wire_format.
4493 
4494   if (result->start >= result->end) {
4495     AddError(parent->full_name(), proto,
4496              DescriptorPool::ErrorCollector::NUMBER,
4497              "Extension range end number must be greater than start number.");
4498   }
4499 }
4500 
BuildReservedRange(const DescriptorProto::ReservedRange & proto,const Descriptor * parent,Descriptor::ReservedRange * result)4501 void DescriptorBuilder::BuildReservedRange(
4502     const DescriptorProto::ReservedRange& proto,
4503     const Descriptor* parent,
4504     Descriptor::ReservedRange* result) {
4505   result->start = proto.start();
4506   result->end = proto.end();
4507   if (result->start <= 0) {
4508     AddError(parent->full_name(), proto,
4509              DescriptorPool::ErrorCollector::NUMBER,
4510              "Reserved numbers must be positive integers.");
4511   }
4512 }
4513 
BuildOneof(const OneofDescriptorProto & proto,Descriptor * parent,OneofDescriptor * result)4514 void DescriptorBuilder::BuildOneof(const OneofDescriptorProto& proto,
4515                                    Descriptor* parent,
4516                                    OneofDescriptor* result) {
4517   string* full_name = tables_->AllocateString(parent->full_name());
4518   full_name->append(1, '.');
4519   full_name->append(proto.name());
4520 
4521   ValidateSymbolName(proto.name(), *full_name, proto);
4522 
4523   result->name_ = tables_->AllocateString(proto.name());
4524   result->full_name_ = full_name;
4525 
4526   result->containing_type_ = parent;
4527 
4528   // We need to fill these in later.
4529   result->field_count_ = 0;
4530   result->fields_ = NULL;
4531 
4532   // Copy options.
4533   if (!proto.has_options()) {
4534     result->options_ = NULL;  // Will set to default_instance later.
4535   } else {
4536     AllocateOptions(proto.options(), result);
4537   }
4538 
4539   AddSymbol(result->full_name(), parent, result->name(),
4540             proto, Symbol(result));
4541 }
4542 
BuildEnum(const EnumDescriptorProto & proto,const Descriptor * parent,EnumDescriptor * result)4543 void DescriptorBuilder::BuildEnum(const EnumDescriptorProto& proto,
4544                                   const Descriptor* parent,
4545                                   EnumDescriptor* result) {
4546   const string& scope = (parent == NULL) ?
4547     file_->package() : parent->full_name();
4548   string* full_name = tables_->AllocateString(scope);
4549   if (!full_name->empty()) full_name->append(1, '.');
4550   full_name->append(proto.name());
4551 
4552   ValidateSymbolName(proto.name(), *full_name, proto);
4553 
4554   result->name_            = tables_->AllocateString(proto.name());
4555   result->full_name_       = full_name;
4556   result->file_            = file_;
4557   result->containing_type_ = parent;
4558   result->is_placeholder_  = false;
4559   result->is_unqualified_placeholder_ = false;
4560 
4561   if (proto.value_size() == 0) {
4562     // We cannot allow enums with no values because this would mean there
4563     // would be no valid default value for fields of this type.
4564     AddError(result->full_name(), proto,
4565              DescriptorPool::ErrorCollector::NAME,
4566              "Enums must contain at least one value.");
4567   }
4568 
4569   BUILD_ARRAY(proto, result, value, BuildEnumValue, result);
4570 
4571   // Copy options.
4572   if (!proto.has_options()) {
4573     result->options_ = NULL;  // Will set to default_instance later.
4574   } else {
4575     AllocateOptions(proto.options(), result);
4576   }
4577 
4578   AddSymbol(result->full_name(), parent, result->name(),
4579             proto, Symbol(result));
4580 }
4581 
BuildEnumValue(const EnumValueDescriptorProto & proto,const EnumDescriptor * parent,EnumValueDescriptor * result)4582 void DescriptorBuilder::BuildEnumValue(const EnumValueDescriptorProto& proto,
4583                                        const EnumDescriptor* parent,
4584                                        EnumValueDescriptor* result) {
4585   result->name_   = tables_->AllocateString(proto.name());
4586   result->number_ = proto.number();
4587   result->type_   = parent;
4588 
4589   // Note:  full_name for enum values is a sibling to the parent's name, not a
4590   //   child of it.
4591   string* full_name = tables_->AllocateString(*parent->full_name_);
4592   full_name->resize(full_name->size() - parent->name_->size());
4593   full_name->append(*result->name_);
4594   result->full_name_ = full_name;
4595 
4596   ValidateSymbolName(proto.name(), *full_name, proto);
4597 
4598   // Copy options.
4599   if (!proto.has_options()) {
4600     result->options_ = NULL;  // Will set to default_instance later.
4601   } else {
4602     AllocateOptions(proto.options(), result);
4603   }
4604 
4605   // Again, enum values are weird because we makes them appear as siblings
4606   // of the enum type instead of children of it.  So, we use
4607   // parent->containing_type() as the value's parent.
4608   bool added_to_outer_scope =
4609     AddSymbol(result->full_name(), parent->containing_type(), result->name(),
4610               proto, Symbol(result));
4611 
4612   // However, we also want to be able to search for values within a single
4613   // enum type, so we add it as a child of the enum type itself, too.
4614   // Note:  This could fail, but if it does, the error has already been
4615   //   reported by the above AddSymbol() call, so we ignore the return code.
4616   bool added_to_inner_scope =
4617     file_tables_->AddAliasUnderParent(parent, result->name(), Symbol(result));
4618 
4619   if (added_to_inner_scope && !added_to_outer_scope) {
4620     // This value did not conflict with any values defined in the same enum,
4621     // but it did conflict with some other symbol defined in the enum type's
4622     // scope.  Let's print an additional error to explain this.
4623     string outer_scope;
4624     if (parent->containing_type() == NULL) {
4625       outer_scope = file_->package();
4626     } else {
4627       outer_scope = parent->containing_type()->full_name();
4628     }
4629 
4630     if (outer_scope.empty()) {
4631       outer_scope = "the global scope";
4632     } else {
4633       outer_scope = "\"" + outer_scope + "\"";
4634     }
4635 
4636     AddError(result->full_name(), proto,
4637              DescriptorPool::ErrorCollector::NAME,
4638              "Note that enum values use C++ scoping rules, meaning that "
4639              "enum values are siblings of their type, not children of it.  "
4640              "Therefore, \"" + result->name() + "\" must be unique within "
4641              + outer_scope + ", not just within \"" + parent->name() + "\".");
4642   }
4643 
4644   // An enum is allowed to define two numbers that refer to the same value.
4645   // FindValueByNumber() should return the first such value, so we simply
4646   // ignore AddEnumValueByNumber()'s return code.
4647   file_tables_->AddEnumValueByNumber(result);
4648 }
4649 
BuildService(const ServiceDescriptorProto & proto,const void *,ServiceDescriptor * result)4650 void DescriptorBuilder::BuildService(const ServiceDescriptorProto& proto,
4651                                      const void* /* dummy */,
4652                                      ServiceDescriptor* result) {
4653   string* full_name = tables_->AllocateString(file_->package());
4654   if (!full_name->empty()) full_name->append(1, '.');
4655   full_name->append(proto.name());
4656 
4657   ValidateSymbolName(proto.name(), *full_name, proto);
4658 
4659   result->name_      = tables_->AllocateString(proto.name());
4660   result->full_name_ = full_name;
4661   result->file_      = file_;
4662 
4663   BUILD_ARRAY(proto, result, method, BuildMethod, result);
4664 
4665   // Copy options.
4666   if (!proto.has_options()) {
4667     result->options_ = NULL;  // Will set to default_instance later.
4668   } else {
4669     AllocateOptions(proto.options(), result);
4670   }
4671 
4672   AddSymbol(result->full_name(), NULL, result->name(),
4673             proto, Symbol(result));
4674 }
4675 
BuildMethod(const MethodDescriptorProto & proto,const ServiceDescriptor * parent,MethodDescriptor * result)4676 void DescriptorBuilder::BuildMethod(const MethodDescriptorProto& proto,
4677                                     const ServiceDescriptor* parent,
4678                                     MethodDescriptor* result) {
4679   result->name_    = tables_->AllocateString(proto.name());
4680   result->service_ = parent;
4681 
4682   string* full_name = tables_->AllocateString(parent->full_name());
4683   full_name->append(1, '.');
4684   full_name->append(*result->name_);
4685   result->full_name_ = full_name;
4686 
4687   ValidateSymbolName(proto.name(), *full_name, proto);
4688 
4689   // These will be filled in when cross-linking.
4690   result->input_type_ = NULL;
4691   result->output_type_ = NULL;
4692 
4693   // Copy options.
4694   if (!proto.has_options()) {
4695     result->options_ = NULL;  // Will set to default_instance later.
4696   } else {
4697     AllocateOptions(proto.options(), result);
4698   }
4699 
4700   result->client_streaming_ = proto.client_streaming();
4701   result->server_streaming_ = proto.server_streaming();
4702 
4703   AddSymbol(result->full_name(), parent, result->name(),
4704             proto, Symbol(result));
4705 }
4706 
4707 #undef BUILD_ARRAY
4708 
4709 // -------------------------------------------------------------------
4710 
CrossLinkFile(FileDescriptor * file,const FileDescriptorProto & proto)4711 void DescriptorBuilder::CrossLinkFile(
4712     FileDescriptor* file, const FileDescriptorProto& proto) {
4713   if (file->options_ == NULL) {
4714     file->options_ = &FileOptions::default_instance();
4715   }
4716 
4717   for (int i = 0; i < file->message_type_count(); i++) {
4718     CrossLinkMessage(&file->message_types_[i], proto.message_type(i));
4719   }
4720 
4721   for (int i = 0; i < file->extension_count(); i++) {
4722     CrossLinkField(&file->extensions_[i], proto.extension(i));
4723   }
4724 
4725   for (int i = 0; i < file->enum_type_count(); i++) {
4726     CrossLinkEnum(&file->enum_types_[i], proto.enum_type(i));
4727   }
4728 
4729   for (int i = 0; i < file->service_count(); i++) {
4730     CrossLinkService(&file->services_[i], proto.service(i));
4731   }
4732 }
4733 
CrossLinkMessage(Descriptor * message,const DescriptorProto & proto)4734 void DescriptorBuilder::CrossLinkMessage(
4735     Descriptor* message, const DescriptorProto& proto) {
4736   if (message->options_ == NULL) {
4737     message->options_ = &MessageOptions::default_instance();
4738   }
4739 
4740   for (int i = 0; i < message->nested_type_count(); i++) {
4741     CrossLinkMessage(&message->nested_types_[i], proto.nested_type(i));
4742   }
4743 
4744   for (int i = 0; i < message->enum_type_count(); i++) {
4745     CrossLinkEnum(&message->enum_types_[i], proto.enum_type(i));
4746   }
4747 
4748   for (int i = 0; i < message->field_count(); i++) {
4749     CrossLinkField(&message->fields_[i], proto.field(i));
4750   }
4751 
4752   for (int i = 0; i < message->extension_count(); i++) {
4753     CrossLinkField(&message->extensions_[i], proto.extension(i));
4754   }
4755 
4756   // Set up field array for each oneof.
4757 
4758   // First count the number of fields per oneof.
4759   for (int i = 0; i < message->field_count(); i++) {
4760     const OneofDescriptor* oneof_decl = message->field(i)->containing_oneof();
4761     if (oneof_decl != NULL) {
4762       // Make sure fields belonging to the same oneof are defined consecutively.
4763       // This enables optimizations in codegens and reflection libraries to
4764       // skip fields in the oneof group, as only one of the field can be set.
4765       // Note that field_count() returns how many fields in this oneof we have
4766       // seen so far. field_count() > 0 guarantees that i > 0, so field(i-1) is
4767       // safe.
4768       if (oneof_decl->field_count() > 0 &&
4769           message->field(i - 1)->containing_oneof() != oneof_decl) {
4770         AddError(
4771             message->full_name() + "." + message->field(i - 1)->name(),
4772             proto.field(i - 1), DescriptorPool::ErrorCollector::OTHER,
4773             strings::Substitute(
4774                 "Fields in the same oneof must be defined consecutively. "
4775                 "\"$0\" cannot be defined before the completion of the "
4776                 "\"$1\" oneof definition.",
4777                 message->field(i - 1)->name(), oneof_decl->name()));
4778       }
4779       // Must go through oneof_decls_ array to get a non-const version of the
4780       // OneofDescriptor.
4781       ++message->oneof_decls_[oneof_decl->index()].field_count_;
4782     }
4783   }
4784 
4785   // Then allocate the arrays.
4786   for (int i = 0; i < message->oneof_decl_count(); i++) {
4787     OneofDescriptor* oneof_decl = &message->oneof_decls_[i];
4788 
4789     if (oneof_decl->field_count() == 0) {
4790       AddError(message->full_name() + "." + oneof_decl->name(),
4791                proto.oneof_decl(i),
4792                DescriptorPool::ErrorCollector::NAME,
4793                "Oneof must have at least one field.");
4794     }
4795 
4796     oneof_decl->fields_ =
4797       tables_->AllocateArray<const FieldDescriptor*>(oneof_decl->field_count_);
4798     oneof_decl->field_count_ = 0;
4799 
4800     if (oneof_decl->options_ == NULL) {
4801       oneof_decl->options_ = &OneofOptions::default_instance();
4802     }
4803   }
4804 
4805   // Then fill them in.
4806   for (int i = 0; i < message->field_count(); i++) {
4807     const OneofDescriptor* oneof_decl = message->field(i)->containing_oneof();
4808     if (oneof_decl != NULL) {
4809       OneofDescriptor* mutable_oneof_decl =
4810           &message->oneof_decls_[oneof_decl->index()];
4811       message->fields_[i].index_in_oneof_ = mutable_oneof_decl->field_count_;
4812       mutable_oneof_decl->fields_[mutable_oneof_decl->field_count_++] =
4813           message->field(i);
4814     }
4815   }
4816 }
4817 
CrossLinkField(FieldDescriptor * field,const FieldDescriptorProto & proto)4818 void DescriptorBuilder::CrossLinkField(
4819     FieldDescriptor* field, const FieldDescriptorProto& proto) {
4820   if (field->options_ == NULL) {
4821     field->options_ = &FieldOptions::default_instance();
4822   }
4823 
4824   if (proto.has_extendee()) {
4825     Symbol extendee = LookupSymbol(proto.extendee(), field->full_name(),
4826                                    PLACEHOLDER_EXTENDABLE_MESSAGE);
4827     if (extendee.IsNull()) {
4828       AddNotDefinedError(field->full_name(), proto,
4829                          DescriptorPool::ErrorCollector::EXTENDEE,
4830                          proto.extendee());
4831       return;
4832     } else if (extendee.type != Symbol::MESSAGE) {
4833       AddError(field->full_name(), proto,
4834                DescriptorPool::ErrorCollector::EXTENDEE,
4835                "\"" + proto.extendee() + "\" is not a message type.");
4836       return;
4837     }
4838     field->containing_type_ = extendee.descriptor;
4839 
4840     const Descriptor::ExtensionRange* extension_range = field->containing_type()
4841         ->FindExtensionRangeContainingNumber(field->number());
4842 
4843     if (extension_range == NULL) {
4844       AddError(field->full_name(), proto,
4845                DescriptorPool::ErrorCollector::NUMBER,
4846                strings::Substitute("\"$0\" does not declare $1 as an "
4847                                    "extension number.",
4848                                    field->containing_type()->full_name(),
4849                                    field->number()));
4850     }
4851   }
4852 
4853   if (field->containing_oneof() != NULL) {
4854     if (field->label() != FieldDescriptor::LABEL_OPTIONAL) {
4855       // Note that this error will never happen when parsing .proto files.
4856       // It can only happen if you manually construct a FileDescriptorProto
4857       // that is incorrect.
4858       AddError(field->full_name(), proto,
4859                DescriptorPool::ErrorCollector::NAME,
4860                "Fields of oneofs must themselves have label LABEL_OPTIONAL.");
4861     }
4862   }
4863 
4864   if (proto.has_type_name()) {
4865     // Assume we are expecting a message type unless the proto contains some
4866     // evidence that it expects an enum type.  This only makes a difference if
4867     // we end up creating a placeholder.
4868     bool expecting_enum = (proto.type() == FieldDescriptorProto::TYPE_ENUM) ||
4869                           proto.has_default_value();
4870 
4871     Symbol type =
4872       LookupSymbol(proto.type_name(), field->full_name(),
4873                    expecting_enum ? PLACEHOLDER_ENUM : PLACEHOLDER_MESSAGE,
4874                    LOOKUP_TYPES);
4875 
4876     // If the type is a weak type, we change the type to a google.protobuf.Empty field.
4877     if (type.IsNull() && !pool_->enforce_weak_ && proto.options().weak()) {
4878       type = FindSymbol(kNonLinkedWeakMessageReplacementName);
4879     }
4880 
4881     if (type.IsNull()) {
4882       AddNotDefinedError(field->full_name(), proto,
4883                          DescriptorPool::ErrorCollector::TYPE,
4884                          proto.type_name());
4885       return;
4886     }
4887 
4888     if (!proto.has_type()) {
4889       // Choose field type based on symbol.
4890       if (type.type == Symbol::MESSAGE) {
4891         field->type_ = FieldDescriptor::TYPE_MESSAGE;
4892       } else if (type.type == Symbol::ENUM) {
4893         field->type_ = FieldDescriptor::TYPE_ENUM;
4894       } else {
4895         AddError(field->full_name(), proto,
4896                  DescriptorPool::ErrorCollector::TYPE,
4897                  "\"" + proto.type_name() + "\" is not a type.");
4898         return;
4899       }
4900     }
4901 
4902     if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
4903       if (type.type != Symbol::MESSAGE) {
4904         AddError(field->full_name(), proto,
4905                  DescriptorPool::ErrorCollector::TYPE,
4906                  "\"" + proto.type_name() + "\" is not a message type.");
4907         return;
4908       }
4909       field->message_type_ = type.descriptor;
4910 
4911       if (field->has_default_value()) {
4912         AddError(field->full_name(), proto,
4913                  DescriptorPool::ErrorCollector::DEFAULT_VALUE,
4914                  "Messages can't have default values.");
4915       }
4916     } else if (field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM) {
4917       if (type.type != Symbol::ENUM) {
4918         AddError(field->full_name(), proto,
4919                  DescriptorPool::ErrorCollector::TYPE,
4920                  "\"" + proto.type_name() + "\" is not an enum type.");
4921         return;
4922       }
4923       field->enum_type_ = type.enum_descriptor;
4924 
4925       if (field->enum_type()->is_placeholder_) {
4926         // We can't look up default values for placeholder types.  We'll have
4927         // to just drop them.
4928         field->has_default_value_ = false;
4929       }
4930 
4931       if (field->has_default_value()) {
4932         // Ensure that the default value is an identifier. Parser cannot always
4933         // verify this because it does not have complete type information.
4934         // N.B. that this check yields better error messages but is not
4935         // necessary for correctness (an enum symbol must be a valid identifier
4936         // anyway), only for better errors.
4937         if (!io::Tokenizer::IsIdentifier(proto.default_value())) {
4938           AddError(field->full_name(), proto,
4939                    DescriptorPool::ErrorCollector::DEFAULT_VALUE,
4940                    "Default value for an enum field must be an identifier.");
4941         } else {
4942           // We can't just use field->enum_type()->FindValueByName() here
4943           // because that locks the pool's mutex, which we have already locked
4944           // at this point.
4945           Symbol default_value =
4946             LookupSymbolNoPlaceholder(proto.default_value(),
4947                                       field->enum_type()->full_name());
4948 
4949           if (default_value.type == Symbol::ENUM_VALUE &&
4950               default_value.enum_value_descriptor->type() ==
4951               field->enum_type()) {
4952             field->default_value_enum_ = default_value.enum_value_descriptor;
4953           } else {
4954             AddError(field->full_name(), proto,
4955                      DescriptorPool::ErrorCollector::DEFAULT_VALUE,
4956                      "Enum type \"" + field->enum_type()->full_name() +
4957                      "\" has no value named \"" + proto.default_value() +
4958                      "\".");
4959           }
4960         }
4961       } else if (field->enum_type()->value_count() > 0) {
4962         // All enums must have at least one value, or we would have reported
4963         // an error elsewhere.  We use the first defined value as the default
4964         // if a default is not explicitly defined.
4965         field->default_value_enum_ = field->enum_type()->value(0);
4966       }
4967     } else {
4968       AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
4969                "Field with primitive type has type_name.");
4970     }
4971   } else {
4972     if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE ||
4973         field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM) {
4974       AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
4975                "Field with message or enum type missing type_name.");
4976     }
4977   }
4978 
4979   // Add the field to the fields-by-number table.
4980   // Note:  We have to do this *after* cross-linking because extensions do not
4981   //   know their containing type until now.
4982   if (!file_tables_->AddFieldByNumber(field)) {
4983     const FieldDescriptor* conflicting_field =
4984       file_tables_->FindFieldByNumber(field->containing_type(),
4985                                       field->number());
4986     if (field->is_extension()) {
4987       AddError(field->full_name(), proto,
4988                DescriptorPool::ErrorCollector::NUMBER,
4989                strings::Substitute("Extension number $0 has already been used "
4990                                    "in \"$1\" by extension \"$2\".",
4991                                    field->number(),
4992                                    field->containing_type()->full_name(),
4993                                    conflicting_field->full_name()));
4994     } else {
4995       AddError(field->full_name(), proto,
4996                DescriptorPool::ErrorCollector::NUMBER,
4997                strings::Substitute("Field number $0 has already been used in "
4998                                    "\"$1\" by field \"$2\".",
4999                                    field->number(),
5000                                    field->containing_type()->full_name(),
5001                                    conflicting_field->name()));
5002     }
5003   } else {
5004     if (field->is_extension()) {
5005       if (!tables_->AddExtension(field)) {
5006         const FieldDescriptor* conflicting_field =
5007             tables_->FindExtension(field->containing_type(), field->number());
5008         string error_msg = strings::Substitute(
5009             "Extension number $0 has already been used in \"$1\" by extension "
5010             "\"$2\" defined in $3.",
5011             field->number(),
5012             field->containing_type()->full_name(),
5013             conflicting_field->full_name(),
5014             conflicting_field->file()->name());
5015         // Conflicting extension numbers should be an error. However, before
5016         // turning this into an error we need to fix all existing broken
5017         // protos first.
5018         // TODO(xiaofeng): Change this to an error.
5019         AddWarning(field->full_name(), proto,
5020                    DescriptorPool::ErrorCollector::NUMBER, error_msg);
5021       }
5022     }
5023   }
5024 
5025   // Add the field to the lowercase-name and camelcase-name tables.
5026   file_tables_->AddFieldByStylizedNames(field);
5027 }
5028 
CrossLinkEnum(EnumDescriptor * enum_type,const EnumDescriptorProto & proto)5029 void DescriptorBuilder::CrossLinkEnum(
5030     EnumDescriptor* enum_type, const EnumDescriptorProto& proto) {
5031   if (enum_type->options_ == NULL) {
5032     enum_type->options_ = &EnumOptions::default_instance();
5033   }
5034 
5035   for (int i = 0; i < enum_type->value_count(); i++) {
5036     CrossLinkEnumValue(&enum_type->values_[i], proto.value(i));
5037   }
5038 }
5039 
CrossLinkEnumValue(EnumValueDescriptor * enum_value,const EnumValueDescriptorProto &)5040 void DescriptorBuilder::CrossLinkEnumValue(
5041     EnumValueDescriptor* enum_value,
5042     const EnumValueDescriptorProto& /* proto */) {
5043   if (enum_value->options_ == NULL) {
5044     enum_value->options_ = &EnumValueOptions::default_instance();
5045   }
5046 }
5047 
CrossLinkService(ServiceDescriptor * service,const ServiceDescriptorProto & proto)5048 void DescriptorBuilder::CrossLinkService(
5049     ServiceDescriptor* service, const ServiceDescriptorProto& proto) {
5050   if (service->options_ == NULL) {
5051     service->options_ = &ServiceOptions::default_instance();
5052   }
5053 
5054   for (int i = 0; i < service->method_count(); i++) {
5055     CrossLinkMethod(&service->methods_[i], proto.method(i));
5056   }
5057 }
5058 
CrossLinkMethod(MethodDescriptor * method,const MethodDescriptorProto & proto)5059 void DescriptorBuilder::CrossLinkMethod(
5060     MethodDescriptor* method, const MethodDescriptorProto& proto) {
5061   if (method->options_ == NULL) {
5062     method->options_ = &MethodOptions::default_instance();
5063   }
5064 
5065   Symbol input_type = LookupSymbol(proto.input_type(), method->full_name());
5066   if (input_type.IsNull()) {
5067     AddNotDefinedError(method->full_name(), proto,
5068                        DescriptorPool::ErrorCollector::INPUT_TYPE,
5069                        proto.input_type());
5070   } else if (input_type.type != Symbol::MESSAGE) {
5071     AddError(method->full_name(), proto,
5072              DescriptorPool::ErrorCollector::INPUT_TYPE,
5073              "\"" + proto.input_type() + "\" is not a message type.");
5074   } else {
5075     method->input_type_ = input_type.descriptor;
5076   }
5077 
5078   Symbol output_type = LookupSymbol(proto.output_type(), method->full_name());
5079   if (output_type.IsNull()) {
5080     AddNotDefinedError(method->full_name(), proto,
5081                        DescriptorPool::ErrorCollector::OUTPUT_TYPE,
5082                        proto.output_type());
5083   } else if (output_type.type != Symbol::MESSAGE) {
5084     AddError(method->full_name(), proto,
5085              DescriptorPool::ErrorCollector::OUTPUT_TYPE,
5086              "\"" + proto.output_type() + "\" is not a message type.");
5087   } else {
5088     method->output_type_ = output_type.descriptor;
5089   }
5090 }
5091 
5092 // -------------------------------------------------------------------
5093 
5094 #define VALIDATE_OPTIONS_FROM_ARRAY(descriptor, array_name, type)  \
5095   for (int i = 0; i < descriptor->array_name##_count(); ++i) {     \
5096     Validate##type##Options(descriptor->array_name##s_ + i,        \
5097                             proto.array_name(i));                  \
5098   }
5099 
5100 // Determine if the file uses optimize_for = LITE_RUNTIME, being careful to
5101 // avoid problems that exist at init time.
IsLite(const FileDescriptor * file)5102 static bool IsLite(const FileDescriptor* file) {
5103   // TODO(kenton):  I don't even remember how many of these conditions are
5104   //   actually possible.  I'm just being super-safe.
5105   return file != NULL &&
5106          &file->options() != &FileOptions::default_instance() &&
5107          file->options().optimize_for() == FileOptions::LITE_RUNTIME;
5108 }
5109 
ValidateFileOptions(FileDescriptor * file,const FileDescriptorProto & proto)5110 void DescriptorBuilder::ValidateFileOptions(FileDescriptor* file,
5111                                             const FileDescriptorProto& proto) {
5112   VALIDATE_OPTIONS_FROM_ARRAY(file, message_type, Message);
5113   VALIDATE_OPTIONS_FROM_ARRAY(file, enum_type, Enum);
5114   VALIDATE_OPTIONS_FROM_ARRAY(file, service, Service);
5115   VALIDATE_OPTIONS_FROM_ARRAY(file, extension, Field);
5116 
5117   // Lite files can only be imported by other Lite files.
5118   if (!IsLite(file)) {
5119     for (int i = 0; i < file->dependency_count(); i++) {
5120       if (IsLite(file->dependency(i))) {
5121         AddError(
5122           file->name(), proto,
5123           DescriptorPool::ErrorCollector::OTHER,
5124           "Files that do not use optimize_for = LITE_RUNTIME cannot import "
5125           "files which do use this option.  This file is not lite, but it "
5126           "imports \"" + file->dependency(i)->name() + "\" which is.");
5127         break;
5128       }
5129     }
5130   }
5131   if (file->syntax() == FileDescriptor::SYNTAX_PROTO3) {
5132     ValidateProto3(file, proto);
5133   }
5134 }
5135 
ValidateProto3(FileDescriptor * file,const FileDescriptorProto & proto)5136 void DescriptorBuilder::ValidateProto3(
5137     FileDescriptor* file, const FileDescriptorProto& proto) {
5138   for (int i = 0; i < file->extension_count(); ++i) {
5139     ValidateProto3Field(file->extensions_ + i, proto.extension(i));
5140   }
5141   for (int i = 0; i < file->message_type_count(); ++i) {
5142     ValidateProto3Message(file->message_types_ + i, proto.message_type(i));
5143   }
5144   for (int i = 0; i < file->enum_type_count(); ++i) {
5145     ValidateProto3Enum(file->enum_types_ + i, proto.enum_type(i));
5146   }
5147 }
5148 
ToLowercaseWithoutUnderscores(const string & name)5149 static string ToLowercaseWithoutUnderscores(const string& name) {
5150   string result;
5151   for (int i = 0; i < name.size(); ++i) {
5152     if (name[i] != '_') {
5153       if (name[i] >= 'A' && name[i] <= 'Z') {
5154         result.push_back(name[i] - 'A' + 'a');
5155       } else {
5156         result.push_back(name[i]);
5157       }
5158     }
5159   }
5160   return result;
5161 }
5162 
ValidateProto3Message(Descriptor * message,const DescriptorProto & proto)5163 void DescriptorBuilder::ValidateProto3Message(
5164     Descriptor* message, const DescriptorProto& proto) {
5165   for (int i = 0; i < message->nested_type_count(); ++i) {
5166     ValidateProto3Message(message->nested_types_ + i,
5167                                  proto.nested_type(i));
5168   }
5169   for (int i = 0; i < message->enum_type_count(); ++i) {
5170     ValidateProto3Enum(message->enum_types_ + i,
5171                               proto.enum_type(i));
5172   }
5173   for (int i = 0; i < message->field_count(); ++i) {
5174     ValidateProto3Field(message->fields_ + i, proto.field(i));
5175   }
5176   for (int i = 0; i < message->extension_count(); ++i) {
5177     ValidateProto3Field(message->extensions_ +i, proto.extension(i));
5178   }
5179   if (message->extension_range_count() > 0) {
5180     AddError(message->full_name(), proto,
5181              DescriptorPool::ErrorCollector::OTHER,
5182              "Extension ranges are not allowed in proto3.");
5183   }
5184   if (message->options().message_set_wire_format()) {
5185     // Using MessageSet doesn't make sense since we disallow extensions.
5186     AddError(message->full_name(), proto,
5187              DescriptorPool::ErrorCollector::OTHER,
5188              "MessageSet is not supported in proto3.");
5189   }
5190 
5191   // In proto3, we reject field names if they conflict in camelCase.
5192   // Note that we currently enforce a stricter rule: Field names must be
5193   // unique after being converted to lowercase with underscores removed.
5194   map<string, const FieldDescriptor*> name_to_field;
5195   for (int i = 0; i < message->field_count(); ++i) {
5196     string lowercase_name = ToLowercaseWithoutUnderscores(
5197         message->field(i)->name());
5198     if (name_to_field.find(lowercase_name) != name_to_field.end()) {
5199       AddError(message->full_name(), proto,
5200                DescriptorPool::ErrorCollector::OTHER,
5201                "The JSON camel-case name of field \"" +
5202                message->field(i)->name() + "\" conflicts with field \"" +
5203                name_to_field[lowercase_name]->name() + "\". This is not " +
5204                "allowed in proto3.");
5205     } else {
5206       name_to_field[lowercase_name] = message->field(i);
5207     }
5208   }
5209 }
5210 
ValidateProto3Field(FieldDescriptor * field,const FieldDescriptorProto & proto)5211 void DescriptorBuilder::ValidateProto3Field(
5212     FieldDescriptor* field, const FieldDescriptorProto& proto) {
5213   if (field->is_extension() &&
5214       !AllowedExtendeeInProto3(field->containing_type()->full_name())) {
5215     AddError(field->full_name(), proto,
5216              DescriptorPool::ErrorCollector::OTHER,
5217              "Extensions in proto3 are only allowed for defining options.");
5218   }
5219   if (field->is_required()) {
5220     AddError(field->full_name(), proto,
5221              DescriptorPool::ErrorCollector::OTHER,
5222              "Required fields are not allowed in proto3.");
5223   }
5224   if (field->has_default_value()) {
5225     AddError(
5226         field->full_name(), proto, DescriptorPool::ErrorCollector::OTHER,
5227         "Explicit default values are not allowed in proto3.");
5228   }
5229   if (field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM &&
5230       field->enum_type() &&
5231       field->enum_type()->file()->syntax() != FileDescriptor::SYNTAX_PROTO3) {
5232     // Proto3 messages can only use Proto3 enum types; otherwise we can't
5233     // guarantee that the default value is zero.
5234     AddError(field->full_name(), proto,
5235              DescriptorPool::ErrorCollector::TYPE,
5236              "Enum type \"" + field->enum_type()->full_name() +
5237              "\" is not a proto3 enum, but is used in \"" +
5238              field->containing_type()->full_name() +
5239              "\" which is a proto3 message type.");
5240   }
5241   if (field->type() == FieldDescriptor::TYPE_GROUP) {
5242     AddError(field->full_name(), proto,
5243              DescriptorPool::ErrorCollector::TYPE,
5244              "Groups are not supported in proto3 syntax.");
5245   }
5246 }
5247 
ValidateProto3Enum(EnumDescriptor * enm,const EnumDescriptorProto & proto)5248 void DescriptorBuilder::ValidateProto3Enum(
5249     EnumDescriptor* enm, const EnumDescriptorProto& proto) {
5250   if (enm->value_count() > 0 && enm->value(0)->number() != 0) {
5251     AddError(
5252         enm->full_name(), proto, DescriptorPool::ErrorCollector::OTHER,
5253         "The first enum value must be zero in proto3.");
5254   }
5255 }
5256 
5257 
ValidateMessageOptions(Descriptor * message,const DescriptorProto & proto)5258 void DescriptorBuilder::ValidateMessageOptions(Descriptor* message,
5259                                                const DescriptorProto& proto) {
5260   VALIDATE_OPTIONS_FROM_ARRAY(message, field, Field);
5261   VALIDATE_OPTIONS_FROM_ARRAY(message, nested_type, Message);
5262   VALIDATE_OPTIONS_FROM_ARRAY(message, enum_type, Enum);
5263   VALIDATE_OPTIONS_FROM_ARRAY(message, extension, Field);
5264 
5265   const int64 max_extension_range =
5266       static_cast<int64>(message->options().message_set_wire_format() ?
5267                          kint32max :
5268                          FieldDescriptor::kMaxNumber);
5269   for (int i = 0; i < message->extension_range_count(); ++i) {
5270     if (message->extension_range(i)->end > max_extension_range + 1) {
5271       AddError(
5272           message->full_name(), proto.extension_range(i),
5273           DescriptorPool::ErrorCollector::NUMBER,
5274           strings::Substitute("Extension numbers cannot be greater than $0.",
5275                               max_extension_range));
5276     }
5277   }
5278 
5279 }
5280 
ValidateFieldOptions(FieldDescriptor * field,const FieldDescriptorProto & proto)5281 void DescriptorBuilder::ValidateFieldOptions(FieldDescriptor* field,
5282     const FieldDescriptorProto& proto) {
5283   // Only message type fields may be lazy.
5284   if (field->options().lazy()) {
5285     if (field->type() != FieldDescriptor::TYPE_MESSAGE) {
5286       AddError(field->full_name(), proto,
5287                DescriptorPool::ErrorCollector::TYPE,
5288                "[lazy = true] can only be specified for submessage fields.");
5289     }
5290   }
5291 
5292   // Only repeated primitive fields may be packed.
5293   if (field->options().packed() && !field->is_packable()) {
5294     AddError(
5295       field->full_name(), proto,
5296       DescriptorPool::ErrorCollector::TYPE,
5297       "[packed = true] can only be specified for repeated primitive fields.");
5298   }
5299 
5300   // Note:  Default instance may not yet be initialized here, so we have to
5301   //   avoid reading from it.
5302   if (field->containing_type_ != NULL &&
5303       &field->containing_type()->options() !=
5304       &MessageOptions::default_instance() &&
5305       field->containing_type()->options().message_set_wire_format()) {
5306     if (field->is_extension()) {
5307       if (!field->is_optional() ||
5308           field->type() != FieldDescriptor::TYPE_MESSAGE) {
5309         AddError(field->full_name(), proto,
5310                  DescriptorPool::ErrorCollector::TYPE,
5311                  "Extensions of MessageSets must be optional messages.");
5312       }
5313     } else {
5314       AddError(field->full_name(), proto,
5315                DescriptorPool::ErrorCollector::NAME,
5316                "MessageSets cannot have fields, only extensions.");
5317     }
5318   }
5319 
5320   // Lite extensions can only be of Lite types.
5321   if (IsLite(field->file()) &&
5322       field->containing_type_ != NULL &&
5323       !IsLite(field->containing_type()->file())) {
5324     AddError(field->full_name(), proto,
5325              DescriptorPool::ErrorCollector::EXTENDEE,
5326              "Extensions to non-lite types can only be declared in non-lite "
5327              "files.  Note that you cannot extend a non-lite type to contain "
5328              "a lite type, but the reverse is allowed.");
5329   }
5330 
5331   // Validate map types.
5332   if (field->is_map()) {
5333     if (!ValidateMapEntry(field, proto)) {
5334       AddError(field->full_name(), proto,
5335                DescriptorPool::ErrorCollector::OTHER,
5336                "map_entry should not be set explicitly. Use map<KeyType, "
5337                "ValueType> instead.");
5338     }
5339   }
5340 
5341 }
5342 
ValidateEnumOptions(EnumDescriptor * enm,const EnumDescriptorProto & proto)5343 void DescriptorBuilder::ValidateEnumOptions(EnumDescriptor* enm,
5344                                             const EnumDescriptorProto& proto) {
5345   VALIDATE_OPTIONS_FROM_ARRAY(enm, value, EnumValue);
5346   if (!enm->options().has_allow_alias() || !enm->options().allow_alias()) {
5347     map<int, string> used_values;
5348     for (int i = 0; i < enm->value_count(); ++i) {
5349       const EnumValueDescriptor* enum_value = enm->value(i);
5350       if (used_values.find(enum_value->number()) != used_values.end()) {
5351         string error =
5352             "\"" + enum_value->full_name() +
5353             "\" uses the same enum value as \"" +
5354             used_values[enum_value->number()] + "\". If this is intended, set "
5355             "'option allow_alias = true;' to the enum definition.";
5356         if (!enm->options().allow_alias()) {
5357           // Generate error if duplicated enum values are explicitly disallowed.
5358           AddError(enm->full_name(), proto,
5359                    DescriptorPool::ErrorCollector::NUMBER,
5360                    error);
5361         } else {
5362           // Generate warning if duplicated values are found but the option
5363           // isn't set.
5364           GOOGLE_LOG(ERROR) << error;
5365         }
5366       } else {
5367         used_values[enum_value->number()] = enum_value->full_name();
5368       }
5369     }
5370   }
5371 }
5372 
ValidateEnumValueOptions(EnumValueDescriptor *,const EnumValueDescriptorProto &)5373 void DescriptorBuilder::ValidateEnumValueOptions(
5374     EnumValueDescriptor* /* enum_value */,
5375     const EnumValueDescriptorProto& /* proto */) {
5376   // Nothing to do so far.
5377 }
ValidateServiceOptions(ServiceDescriptor * service,const ServiceDescriptorProto & proto)5378 void DescriptorBuilder::ValidateServiceOptions(ServiceDescriptor* service,
5379     const ServiceDescriptorProto& proto) {
5380   if (IsLite(service->file()) &&
5381       (service->file()->options().cc_generic_services() ||
5382        service->file()->options().java_generic_services())) {
5383     AddError(service->full_name(), proto,
5384              DescriptorPool::ErrorCollector::NAME,
5385              "Files with optimize_for = LITE_RUNTIME cannot define services "
5386              "unless you set both options cc_generic_services and "
5387              "java_generic_sevices to false.");
5388   }
5389 
5390   VALIDATE_OPTIONS_FROM_ARRAY(service, method, Method);
5391 }
5392 
ValidateMethodOptions(MethodDescriptor *,const MethodDescriptorProto &)5393 void DescriptorBuilder::ValidateMethodOptions(MethodDescriptor* /* method */,
5394     const MethodDescriptorProto& /* proto */) {
5395   // Nothing to do so far.
5396 }
5397 
ValidateMapEntry(FieldDescriptor * field,const FieldDescriptorProto & proto)5398 bool DescriptorBuilder::ValidateMapEntry(FieldDescriptor* field,
5399                                          const FieldDescriptorProto& proto) {
5400   const Descriptor* message = field->message_type();
5401   if (// Must not contain extensions, extension range or nested message or
5402       // enums
5403       message->extension_count() != 0 ||
5404       field->label() != FieldDescriptor::LABEL_REPEATED ||
5405       message->extension_range_count() != 0 ||
5406       message->nested_type_count() != 0 || message->enum_type_count() != 0 ||
5407       // Must contain exactly two fields
5408       message->field_count() != 2 ||
5409       // Field name and message name must match
5410       message->name() != ToCamelCase(field->name(), false) + "Entry" ||
5411       // Entry message must be in the same containing type of the field.
5412       field->containing_type() != message->containing_type()) {
5413     return false;
5414   }
5415 
5416   const FieldDescriptor* key = message->field(0);
5417   const FieldDescriptor* value = message->field(1);
5418   if (key->label() != FieldDescriptor::LABEL_OPTIONAL || key->number() != 1 ||
5419       key->name() != "key") {
5420     return false;
5421   }
5422   if (value->label() != FieldDescriptor::LABEL_OPTIONAL ||
5423       value->number() != 2 || value->name() != "value") {
5424     return false;
5425   }
5426 
5427   // Check key types are legal.
5428   switch (key->type()) {
5429     case FieldDescriptor::TYPE_ENUM:
5430       AddError(
5431           field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
5432           "Key in map fields cannot be enum types.");
5433       break;
5434     case FieldDescriptor::TYPE_FLOAT:
5435     case FieldDescriptor::TYPE_DOUBLE:
5436     case FieldDescriptor::TYPE_MESSAGE:
5437     case FieldDescriptor::TYPE_GROUP:
5438     case FieldDescriptor::TYPE_BYTES:
5439       AddError(
5440           field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
5441           "Key in map fields cannot be float/double, bytes or message types.");
5442       break;
5443     case FieldDescriptor::TYPE_BOOL:
5444     case FieldDescriptor::TYPE_INT32:
5445     case FieldDescriptor::TYPE_INT64:
5446     case FieldDescriptor::TYPE_SINT32:
5447     case FieldDescriptor::TYPE_SINT64:
5448     case FieldDescriptor::TYPE_STRING:
5449     case FieldDescriptor::TYPE_UINT32:
5450     case FieldDescriptor::TYPE_UINT64:
5451     case FieldDescriptor::TYPE_FIXED32:
5452     case FieldDescriptor::TYPE_FIXED64:
5453     case FieldDescriptor::TYPE_SFIXED32:
5454     case FieldDescriptor::TYPE_SFIXED64:
5455       // Legal cases
5456       break;
5457     // Do not add a default, so that the compiler will complain when new types
5458     // are added.
5459   }
5460 
5461   if (value->type() == FieldDescriptor::TYPE_ENUM) {
5462     if (value->enum_type()->value(0)->number() != 0) {
5463       AddError(
5464           field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
5465           "Enum value in map must define 0 as the first value.");
5466     }
5467   }
5468 
5469   return true;
5470 }
5471 
DetectMapConflicts(const Descriptor * message,const DescriptorProto & proto)5472 void DescriptorBuilder::DetectMapConflicts(const Descriptor* message,
5473                                            const DescriptorProto& proto) {
5474   map<string, const Descriptor*> seen_types;
5475   for (int i = 0; i < message->nested_type_count(); ++i) {
5476     const Descriptor* nested = message->nested_type(i);
5477     pair<map<string, const Descriptor*>::iterator, bool> result =
5478         seen_types.insert(std::make_pair(nested->name(), nested));
5479     if (!result.second) {
5480       if (result.first->second->options().map_entry() ||
5481           nested->options().map_entry()) {
5482         AddError(message->full_name(), proto,
5483                  DescriptorPool::ErrorCollector::NAME,
5484                  "Expanded map entry type " + nested->name() +
5485                  " conflicts with an existing nested message type.");
5486       }
5487     }
5488     // Recursively test on the nested types.
5489     DetectMapConflicts(message->nested_type(i), proto.nested_type(i));
5490   }
5491   // Check for conflicted field names.
5492   for (int i = 0; i < message->field_count(); ++i) {
5493     const FieldDescriptor* field = message->field(i);
5494     map<string, const Descriptor*>::iterator iter =
5495         seen_types.find(field->name());
5496     if (iter != seen_types.end() && iter->second->options().map_entry()) {
5497       AddError(message->full_name(), proto,
5498                DescriptorPool::ErrorCollector::NAME,
5499                "Expanded map entry type " + iter->second->name() +
5500                " conflicts with an existing field.");
5501     }
5502   }
5503   // Check for conflicted enum names.
5504   for (int i = 0; i < message->enum_type_count(); ++i) {
5505     const EnumDescriptor* enum_desc = message->enum_type(i);
5506     map<string, const Descriptor*>::iterator iter =
5507         seen_types.find(enum_desc->name());
5508     if (iter != seen_types.end() && iter->second->options().map_entry()) {
5509       AddError(message->full_name(), proto,
5510                DescriptorPool::ErrorCollector::NAME,
5511                "Expanded map entry type " + iter->second->name() +
5512                " conflicts with an existing enum type.");
5513     }
5514   }
5515   // Check for conflicted oneof names.
5516   for (int i = 0; i < message->oneof_decl_count(); ++i) {
5517     const OneofDescriptor* oneof_desc = message->oneof_decl(i);
5518     map<string, const Descriptor*>::iterator iter =
5519         seen_types.find(oneof_desc->name());
5520     if (iter != seen_types.end() && iter->second->options().map_entry()) {
5521       AddError(message->full_name(), proto,
5522                DescriptorPool::ErrorCollector::NAME,
5523                "Expanded map entry type " + iter->second->name() +
5524                " conflicts with an existing oneof type.");
5525     }
5526   }
5527 }
5528 
5529 
5530 #undef VALIDATE_OPTIONS_FROM_ARRAY
5531 
5532 // -------------------------------------------------------------------
5533 
OptionInterpreter(DescriptorBuilder * builder)5534 DescriptorBuilder::OptionInterpreter::OptionInterpreter(
5535     DescriptorBuilder* builder) : builder_(builder) {
5536   GOOGLE_CHECK(builder_);
5537 }
5538 
~OptionInterpreter()5539 DescriptorBuilder::OptionInterpreter::~OptionInterpreter() {
5540 }
5541 
InterpretOptions(OptionsToInterpret * options_to_interpret)5542 bool DescriptorBuilder::OptionInterpreter::InterpretOptions(
5543     OptionsToInterpret* options_to_interpret) {
5544   // Note that these may be in different pools, so we can't use the same
5545   // descriptor and reflection objects on both.
5546   Message* options = options_to_interpret->options;
5547   const Message* original_options = options_to_interpret->original_options;
5548 
5549   bool failed = false;
5550   options_to_interpret_ = options_to_interpret;
5551 
5552   // Find the uninterpreted_option field in the mutable copy of the options
5553   // and clear them, since we're about to interpret them.
5554   const FieldDescriptor* uninterpreted_options_field =
5555       options->GetDescriptor()->FindFieldByName("uninterpreted_option");
5556   GOOGLE_CHECK(uninterpreted_options_field != NULL)
5557       << "No field named \"uninterpreted_option\" in the Options proto.";
5558   options->GetReflection()->ClearField(options, uninterpreted_options_field);
5559 
5560   // Find the uninterpreted_option field in the original options.
5561   const FieldDescriptor* original_uninterpreted_options_field =
5562       original_options->GetDescriptor()->
5563           FindFieldByName("uninterpreted_option");
5564   GOOGLE_CHECK(original_uninterpreted_options_field != NULL)
5565       << "No field named \"uninterpreted_option\" in the Options proto.";
5566 
5567   const int num_uninterpreted_options = original_options->GetReflection()->
5568       FieldSize(*original_options, original_uninterpreted_options_field);
5569   for (int i = 0; i < num_uninterpreted_options; ++i) {
5570     uninterpreted_option_ = down_cast<const UninterpretedOption*>(
5571         &original_options->GetReflection()->GetRepeatedMessage(
5572             *original_options, original_uninterpreted_options_field, i));
5573     if (!InterpretSingleOption(options)) {
5574       // Error already added by InterpretSingleOption().
5575       failed = true;
5576       break;
5577     }
5578   }
5579   // Reset these, so we don't have any dangling pointers.
5580   uninterpreted_option_ = NULL;
5581   options_to_interpret_ = NULL;
5582 
5583   if (!failed) {
5584     // InterpretSingleOption() added the interpreted options in the
5585     // UnknownFieldSet, in case the option isn't yet known to us.  Now we
5586     // serialize the options message and deserialize it back.  That way, any
5587     // option fields that we do happen to know about will get moved from the
5588     // UnknownFieldSet into the real fields, and thus be available right away.
5589     // If they are not known, that's OK too. They will get reparsed into the
5590     // UnknownFieldSet and wait there until the message is parsed by something
5591     // that does know about the options.
5592     string buf;
5593     GOOGLE_CHECK(options->AppendPartialToString(&buf))
5594         << "Protocol message could not be serialized.";
5595     GOOGLE_CHECK(options->ParsePartialFromString(buf))
5596         << "Protocol message serialized itself in invalid fashion.";
5597     if (!options->IsInitialized()) {
5598       builder_->AddWarning(
5599           options_to_interpret->element_name, *original_options,
5600           DescriptorPool::ErrorCollector::OTHER,
5601           "Options could not be fully parsed using the proto descriptors "
5602           "compiled into this binary. Missing required fields: " +
5603           options->InitializationErrorString());
5604     }
5605   }
5606   return !failed;
5607 }
5608 
InterpretSingleOption(Message * options)5609 bool DescriptorBuilder::OptionInterpreter::InterpretSingleOption(
5610     Message* options) {
5611   // First do some basic validation.
5612   if (uninterpreted_option_->name_size() == 0) {
5613     // This should never happen unless the parser has gone seriously awry or
5614     // someone has manually created the uninterpreted option badly.
5615     return AddNameError("Option must have a name.");
5616   }
5617   if (uninterpreted_option_->name(0).name_part() == "uninterpreted_option") {
5618     return AddNameError("Option must not use reserved name "
5619                         "\"uninterpreted_option\".");
5620   }
5621 
5622   const Descriptor* options_descriptor = NULL;
5623   // Get the options message's descriptor from the builder's pool, so that we
5624   // get the version that knows about any extension options declared in the
5625   // file we're currently building. The descriptor should be there as long as
5626   // the file we're building imported "google/protobuf/descriptors.proto".
5627 
5628   // Note that we use DescriptorBuilder::FindSymbolNotEnforcingDeps(), not
5629   // DescriptorPool::FindMessageTypeByName() because we're already holding the
5630   // pool's mutex, and the latter method locks it again.  We don't use
5631   // FindSymbol() because files that use custom options only need to depend on
5632   // the file that defines the option, not descriptor.proto itself.
5633   Symbol symbol = builder_->FindSymbolNotEnforcingDeps(
5634     options->GetDescriptor()->full_name());
5635   if (!symbol.IsNull() && symbol.type == Symbol::MESSAGE) {
5636     options_descriptor = symbol.descriptor;
5637   } else {
5638     // The options message's descriptor was not in the builder's pool, so use
5639     // the standard version from the generated pool. We're not holding the
5640     // generated pool's mutex, so we can search it the straightforward way.
5641     options_descriptor = options->GetDescriptor();
5642   }
5643   GOOGLE_CHECK(options_descriptor);
5644 
5645   // We iterate over the name parts to drill into the submessages until we find
5646   // the leaf field for the option. As we drill down we remember the current
5647   // submessage's descriptor in |descriptor| and the next field in that
5648   // submessage in |field|. We also track the fields we're drilling down
5649   // through in |intermediate_fields|. As we go, we reconstruct the full option
5650   // name in |debug_msg_name|, for use in error messages.
5651   const Descriptor* descriptor = options_descriptor;
5652   const FieldDescriptor* field = NULL;
5653   vector<const FieldDescriptor*> intermediate_fields;
5654   string debug_msg_name = "";
5655 
5656   for (int i = 0; i < uninterpreted_option_->name_size(); ++i) {
5657     const string& name_part = uninterpreted_option_->name(i).name_part();
5658     if (debug_msg_name.size() > 0) {
5659       debug_msg_name += ".";
5660     }
5661     if (uninterpreted_option_->name(i).is_extension()) {
5662       debug_msg_name += "(" + name_part + ")";
5663       // Search for the extension's descriptor as an extension in the builder's
5664       // pool. Note that we use DescriptorBuilder::LookupSymbol(), not
5665       // DescriptorPool::FindExtensionByName(), for two reasons: 1) It allows
5666       // relative lookups, and 2) because we're already holding the pool's
5667       // mutex, and the latter method locks it again.
5668       symbol = builder_->LookupSymbol(name_part,
5669                                       options_to_interpret_->name_scope);
5670       if (!symbol.IsNull() && symbol.type == Symbol::FIELD) {
5671         field = symbol.field_descriptor;
5672       }
5673       // If we don't find the field then the field's descriptor was not in the
5674       // builder's pool, but there's no point in looking in the generated
5675       // pool. We require that you import the file that defines any extensions
5676       // you use, so they must be present in the builder's pool.
5677     } else {
5678       debug_msg_name += name_part;
5679       // Search for the field's descriptor as a regular field.
5680       field = descriptor->FindFieldByName(name_part);
5681     }
5682 
5683     if (field == NULL) {
5684       if (get_allow_unknown(builder_->pool_)) {
5685         // We can't find the option, but AllowUnknownDependencies() is enabled,
5686         // so we will just leave it as uninterpreted.
5687         AddWithoutInterpreting(*uninterpreted_option_, options);
5688         return true;
5689       } else if (!(builder_->undefine_resolved_name_).empty()) {
5690         // Option is resolved to a name which is not defined.
5691         return AddNameError(
5692             "Option \"" + debug_msg_name + "\" is resolved to \"(" +
5693             builder_->undefine_resolved_name_ +
5694             ")\", which is not defined. The innermost scope is searched first "
5695             "in name resolution. Consider using a leading '.'(i.e., \"(." +
5696             debug_msg_name.substr(1) +
5697             "\") to start from the outermost scope.");
5698       } else {
5699         return AddNameError("Option \"" + debug_msg_name + "\" unknown.");
5700       }
5701     } else if (field->containing_type() != descriptor) {
5702       if (get_is_placeholder(field->containing_type())) {
5703         // The field is an extension of a placeholder type, so we can't
5704         // reliably verify whether it is a valid extension to use here (e.g.
5705         // we don't know if it is an extension of the correct *Options message,
5706         // or if it has a valid field number, etc.).  Just leave it as
5707         // uninterpreted instead.
5708         AddWithoutInterpreting(*uninterpreted_option_, options);
5709         return true;
5710       } else {
5711         // This can only happen if, due to some insane misconfiguration of the
5712         // pools, we find the options message in one pool but the field in
5713         // another. This would probably imply a hefty bug somewhere.
5714         return AddNameError("Option field \"" + debug_msg_name +
5715                             "\" is not a field or extension of message \"" +
5716                             descriptor->name() + "\".");
5717       }
5718     } else if (i < uninterpreted_option_->name_size() - 1) {
5719       if (field->cpp_type() != FieldDescriptor::CPPTYPE_MESSAGE) {
5720         return AddNameError("Option \"" +  debug_msg_name +
5721                             "\" is an atomic type, not a message.");
5722       } else if (field->is_repeated()) {
5723         return AddNameError("Option field \"" + debug_msg_name +
5724                             "\" is a repeated message. Repeated message "
5725                             "options must be initialized using an "
5726                             "aggregate value.");
5727       } else {
5728         // Drill down into the submessage.
5729         intermediate_fields.push_back(field);
5730         descriptor = field->message_type();
5731       }
5732     }
5733   }
5734 
5735   // We've found the leaf field. Now we use UnknownFieldSets to set its value
5736   // on the options message. We do so because the message may not yet know
5737   // about its extension fields, so we may not be able to set the fields
5738   // directly. But the UnknownFieldSets will serialize to the same wire-format
5739   // message, so reading that message back in once the extension fields are
5740   // known will populate them correctly.
5741 
5742   // First see if the option is already set.
5743   if (!field->is_repeated() && !ExamineIfOptionIsSet(
5744           intermediate_fields.begin(),
5745           intermediate_fields.end(),
5746           field, debug_msg_name,
5747           options->GetReflection()->GetUnknownFields(*options))) {
5748     return false;  // ExamineIfOptionIsSet() already added the error.
5749   }
5750 
5751 
5752   // First set the value on the UnknownFieldSet corresponding to the
5753   // innermost message.
5754   google::protobuf::scoped_ptr<UnknownFieldSet> unknown_fields(new UnknownFieldSet());
5755   if (!SetOptionValue(field, unknown_fields.get())) {
5756     return false;  // SetOptionValue() already added the error.
5757   }
5758 
5759   // Now wrap the UnknownFieldSet with UnknownFieldSets corresponding to all
5760   // the intermediate messages.
5761   for (vector<const FieldDescriptor*>::reverse_iterator iter =
5762            intermediate_fields.rbegin();
5763        iter != intermediate_fields.rend(); ++iter) {
5764     google::protobuf::scoped_ptr<UnknownFieldSet> parent_unknown_fields(
5765         new UnknownFieldSet());
5766     switch ((*iter)->type()) {
5767       case FieldDescriptor::TYPE_MESSAGE: {
5768         io::StringOutputStream outstr(
5769             parent_unknown_fields->AddLengthDelimited((*iter)->number()));
5770         io::CodedOutputStream out(&outstr);
5771         internal::WireFormat::SerializeUnknownFields(*unknown_fields, &out);
5772         GOOGLE_CHECK(!out.HadError())
5773             << "Unexpected failure while serializing option submessage "
5774             << debug_msg_name << "\".";
5775         break;
5776       }
5777 
5778       case FieldDescriptor::TYPE_GROUP: {
5779          parent_unknown_fields->AddGroup((*iter)->number())
5780                               ->MergeFrom(*unknown_fields);
5781         break;
5782       }
5783 
5784       default:
5785         GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_MESSAGE: "
5786                    << (*iter)->type();
5787         return false;
5788     }
5789     unknown_fields.reset(parent_unknown_fields.release());
5790   }
5791 
5792   // Now merge the UnknownFieldSet corresponding to the top-level message into
5793   // the options message.
5794   options->GetReflection()->MutableUnknownFields(options)->MergeFrom(
5795       *unknown_fields);
5796 
5797   return true;
5798 }
5799 
AddWithoutInterpreting(const UninterpretedOption & uninterpreted_option,Message * options)5800 void DescriptorBuilder::OptionInterpreter::AddWithoutInterpreting(
5801     const UninterpretedOption& uninterpreted_option, Message* options) {
5802   const FieldDescriptor* field =
5803     options->GetDescriptor()->FindFieldByName("uninterpreted_option");
5804   GOOGLE_CHECK(field != NULL);
5805 
5806   options->GetReflection()->AddMessage(options, field)
5807     ->CopyFrom(uninterpreted_option);
5808 }
5809 
ExamineIfOptionIsSet(vector<const FieldDescriptor * >::const_iterator intermediate_fields_iter,vector<const FieldDescriptor * >::const_iterator intermediate_fields_end,const FieldDescriptor * innermost_field,const string & debug_msg_name,const UnknownFieldSet & unknown_fields)5810 bool DescriptorBuilder::OptionInterpreter::ExamineIfOptionIsSet(
5811     vector<const FieldDescriptor*>::const_iterator intermediate_fields_iter,
5812     vector<const FieldDescriptor*>::const_iterator intermediate_fields_end,
5813     const FieldDescriptor* innermost_field, const string& debug_msg_name,
5814     const UnknownFieldSet& unknown_fields) {
5815   // We do linear searches of the UnknownFieldSet and its sub-groups.  This
5816   // should be fine since it's unlikely that any one options structure will
5817   // contain more than a handful of options.
5818 
5819   if (intermediate_fields_iter == intermediate_fields_end) {
5820     // We're at the innermost submessage.
5821     for (int i = 0; i < unknown_fields.field_count(); i++) {
5822       if (unknown_fields.field(i).number() == innermost_field->number()) {
5823         return AddNameError("Option \"" + debug_msg_name +
5824                             "\" was already set.");
5825       }
5826     }
5827     return true;
5828   }
5829 
5830   for (int i = 0; i < unknown_fields.field_count(); i++) {
5831     if (unknown_fields.field(i).number() ==
5832         (*intermediate_fields_iter)->number()) {
5833       const UnknownField* unknown_field = &unknown_fields.field(i);
5834       FieldDescriptor::Type type = (*intermediate_fields_iter)->type();
5835       // Recurse into the next submessage.
5836       switch (type) {
5837         case FieldDescriptor::TYPE_MESSAGE:
5838           if (unknown_field->type() == UnknownField::TYPE_LENGTH_DELIMITED) {
5839             UnknownFieldSet intermediate_unknown_fields;
5840             if (intermediate_unknown_fields.ParseFromString(
5841                     unknown_field->length_delimited()) &&
5842                 !ExamineIfOptionIsSet(intermediate_fields_iter + 1,
5843                                       intermediate_fields_end,
5844                                       innermost_field, debug_msg_name,
5845                                       intermediate_unknown_fields)) {
5846               return false;  // Error already added.
5847             }
5848           }
5849           break;
5850 
5851         case FieldDescriptor::TYPE_GROUP:
5852           if (unknown_field->type() == UnknownField::TYPE_GROUP) {
5853             if (!ExamineIfOptionIsSet(intermediate_fields_iter + 1,
5854                                       intermediate_fields_end,
5855                                       innermost_field, debug_msg_name,
5856                                       unknown_field->group())) {
5857               return false;  // Error already added.
5858             }
5859           }
5860           break;
5861 
5862         default:
5863           GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_MESSAGE: " << type;
5864           return false;
5865       }
5866     }
5867   }
5868   return true;
5869 }
5870 
SetOptionValue(const FieldDescriptor * option_field,UnknownFieldSet * unknown_fields)5871 bool DescriptorBuilder::OptionInterpreter::SetOptionValue(
5872     const FieldDescriptor* option_field,
5873     UnknownFieldSet* unknown_fields) {
5874   // We switch on the CppType to validate.
5875   switch (option_field->cpp_type()) {
5876 
5877     case FieldDescriptor::CPPTYPE_INT32:
5878       if (uninterpreted_option_->has_positive_int_value()) {
5879         if (uninterpreted_option_->positive_int_value() >
5880             static_cast<uint64>(kint32max)) {
5881           return AddValueError("Value out of range for int32 option \"" +
5882                                option_field->full_name() + "\".");
5883         } else {
5884           SetInt32(option_field->number(),
5885                    uninterpreted_option_->positive_int_value(),
5886                    option_field->type(), unknown_fields);
5887         }
5888       } else if (uninterpreted_option_->has_negative_int_value()) {
5889         if (uninterpreted_option_->negative_int_value() <
5890             static_cast<int64>(kint32min)) {
5891           return AddValueError("Value out of range for int32 option \"" +
5892                                option_field->full_name() + "\".");
5893         } else {
5894           SetInt32(option_field->number(),
5895                    uninterpreted_option_->negative_int_value(),
5896                    option_field->type(), unknown_fields);
5897         }
5898       } else {
5899         return AddValueError("Value must be integer for int32 option \"" +
5900                              option_field->full_name() + "\".");
5901       }
5902       break;
5903 
5904     case FieldDescriptor::CPPTYPE_INT64:
5905       if (uninterpreted_option_->has_positive_int_value()) {
5906         if (uninterpreted_option_->positive_int_value() >
5907             static_cast<uint64>(kint64max)) {
5908           return AddValueError("Value out of range for int64 option \"" +
5909                                option_field->full_name() + "\".");
5910         } else {
5911           SetInt64(option_field->number(),
5912                    uninterpreted_option_->positive_int_value(),
5913                    option_field->type(), unknown_fields);
5914         }
5915       } else if (uninterpreted_option_->has_negative_int_value()) {
5916         SetInt64(option_field->number(),
5917                  uninterpreted_option_->negative_int_value(),
5918                  option_field->type(), unknown_fields);
5919       } else {
5920         return AddValueError("Value must be integer for int64 option \"" +
5921                              option_field->full_name() + "\".");
5922       }
5923       break;
5924 
5925     case FieldDescriptor::CPPTYPE_UINT32:
5926       if (uninterpreted_option_->has_positive_int_value()) {
5927         if (uninterpreted_option_->positive_int_value() > kuint32max) {
5928           return AddValueError("Value out of range for uint32 option \"" +
5929                                option_field->name() + "\".");
5930         } else {
5931           SetUInt32(option_field->number(),
5932                     uninterpreted_option_->positive_int_value(),
5933                     option_field->type(), unknown_fields);
5934         }
5935       } else {
5936         return AddValueError("Value must be non-negative integer for uint32 "
5937                              "option \"" + option_field->full_name() + "\".");
5938       }
5939       break;
5940 
5941     case FieldDescriptor::CPPTYPE_UINT64:
5942       if (uninterpreted_option_->has_positive_int_value()) {
5943         SetUInt64(option_field->number(),
5944                   uninterpreted_option_->positive_int_value(),
5945                   option_field->type(), unknown_fields);
5946       } else {
5947         return AddValueError("Value must be non-negative integer for uint64 "
5948                              "option \"" + option_field->full_name() + "\".");
5949       }
5950       break;
5951 
5952     case FieldDescriptor::CPPTYPE_FLOAT: {
5953       float value;
5954       if (uninterpreted_option_->has_double_value()) {
5955         value = uninterpreted_option_->double_value();
5956       } else if (uninterpreted_option_->has_positive_int_value()) {
5957         value = uninterpreted_option_->positive_int_value();
5958       } else if (uninterpreted_option_->has_negative_int_value()) {
5959         value = uninterpreted_option_->negative_int_value();
5960       } else {
5961         return AddValueError("Value must be number for float option \"" +
5962                              option_field->full_name() + "\".");
5963       }
5964       unknown_fields->AddFixed32(option_field->number(),
5965           google::protobuf::internal::WireFormatLite::EncodeFloat(value));
5966       break;
5967     }
5968 
5969     case FieldDescriptor::CPPTYPE_DOUBLE: {
5970       double value;
5971       if (uninterpreted_option_->has_double_value()) {
5972         value = uninterpreted_option_->double_value();
5973       } else if (uninterpreted_option_->has_positive_int_value()) {
5974         value = uninterpreted_option_->positive_int_value();
5975       } else if (uninterpreted_option_->has_negative_int_value()) {
5976         value = uninterpreted_option_->negative_int_value();
5977       } else {
5978         return AddValueError("Value must be number for double option \"" +
5979                              option_field->full_name() + "\".");
5980       }
5981       unknown_fields->AddFixed64(option_field->number(),
5982           google::protobuf::internal::WireFormatLite::EncodeDouble(value));
5983       break;
5984     }
5985 
5986     case FieldDescriptor::CPPTYPE_BOOL:
5987       uint64 value;
5988       if (!uninterpreted_option_->has_identifier_value()) {
5989         return AddValueError("Value must be identifier for boolean option "
5990                              "\"" + option_field->full_name() + "\".");
5991       }
5992       if (uninterpreted_option_->identifier_value() == "true") {
5993         value = 1;
5994       } else if (uninterpreted_option_->identifier_value() == "false") {
5995         value = 0;
5996       } else {
5997         return AddValueError("Value must be \"true\" or \"false\" for boolean "
5998                              "option \"" + option_field->full_name() + "\".");
5999       }
6000       unknown_fields->AddVarint(option_field->number(), value);
6001       break;
6002 
6003     case FieldDescriptor::CPPTYPE_ENUM: {
6004       if (!uninterpreted_option_->has_identifier_value()) {
6005         return AddValueError("Value must be identifier for enum-valued option "
6006                              "\"" + option_field->full_name() + "\".");
6007       }
6008       const EnumDescriptor* enum_type = option_field->enum_type();
6009       const string& value_name = uninterpreted_option_->identifier_value();
6010       const EnumValueDescriptor* enum_value = NULL;
6011 
6012       if (enum_type->file()->pool() != DescriptorPool::generated_pool()) {
6013         // Note that the enum value's fully-qualified name is a sibling of the
6014         // enum's name, not a child of it.
6015         string fully_qualified_name = enum_type->full_name();
6016         fully_qualified_name.resize(fully_qualified_name.size() -
6017                                     enum_type->name().size());
6018         fully_qualified_name += value_name;
6019 
6020         // Search for the enum value's descriptor in the builder's pool. Note
6021         // that we use DescriptorBuilder::FindSymbolNotEnforcingDeps(), not
6022         // DescriptorPool::FindEnumValueByName() because we're already holding
6023         // the pool's mutex, and the latter method locks it again.
6024         Symbol symbol =
6025           builder_->FindSymbolNotEnforcingDeps(fully_qualified_name);
6026         if (!symbol.IsNull() && symbol.type == Symbol::ENUM_VALUE) {
6027           if (symbol.enum_value_descriptor->type() != enum_type) {
6028             return AddValueError("Enum type \"" + enum_type->full_name() +
6029                 "\" has no value named \"" + value_name + "\" for option \"" +
6030                 option_field->full_name() +
6031                 "\". This appears to be a value from a sibling type.");
6032           } else {
6033             enum_value = symbol.enum_value_descriptor;
6034           }
6035         }
6036       } else {
6037         // The enum type is in the generated pool, so we can search for the
6038         // value there.
6039         enum_value = enum_type->FindValueByName(value_name);
6040       }
6041 
6042       if (enum_value == NULL) {
6043         return AddValueError("Enum type \"" +
6044                              option_field->enum_type()->full_name() +
6045                              "\" has no value named \"" + value_name + "\" for "
6046                              "option \"" + option_field->full_name() + "\".");
6047       } else {
6048         // Sign-extension is not a problem, since we cast directly from int32 to
6049         // uint64, without first going through uint32.
6050         unknown_fields->AddVarint(option_field->number(),
6051           static_cast<uint64>(static_cast<int64>(enum_value->number())));
6052       }
6053       break;
6054     }
6055 
6056     case FieldDescriptor::CPPTYPE_STRING:
6057       if (!uninterpreted_option_->has_string_value()) {
6058         return AddValueError("Value must be quoted string for string option "
6059                              "\"" + option_field->full_name() + "\".");
6060       }
6061       // The string has already been unquoted and unescaped by the parser.
6062       unknown_fields->AddLengthDelimited(option_field->number(),
6063           uninterpreted_option_->string_value());
6064       break;
6065 
6066     case FieldDescriptor::CPPTYPE_MESSAGE:
6067       if (!SetAggregateOption(option_field, unknown_fields)) {
6068         return false;
6069       }
6070       break;
6071   }
6072 
6073   return true;
6074 }
6075 
6076 class DescriptorBuilder::OptionInterpreter::AggregateOptionFinder
6077     : public TextFormat::Finder {
6078  public:
6079   DescriptorBuilder* builder_;
6080 
FindExtension(Message * message,const string & name) const6081   virtual const FieldDescriptor* FindExtension(
6082       Message* message, const string& name) const {
6083     assert_mutex_held(builder_->pool_);
6084     const Descriptor* descriptor = message->GetDescriptor();
6085     Symbol result = builder_->LookupSymbolNoPlaceholder(
6086         name, descriptor->full_name());
6087     if (result.type == Symbol::FIELD &&
6088         result.field_descriptor->is_extension()) {
6089       return result.field_descriptor;
6090     } else if (result.type == Symbol::MESSAGE &&
6091                descriptor->options().message_set_wire_format()) {
6092       const Descriptor* foreign_type = result.descriptor;
6093       // The text format allows MessageSet items to be specified using
6094       // the type name, rather than the extension identifier. If the symbol
6095       // lookup returned a Message, and the enclosing Message has
6096       // message_set_wire_format = true, then return the message set
6097       // extension, if one exists.
6098       for (int i = 0; i < foreign_type->extension_count(); i++) {
6099         const FieldDescriptor* extension = foreign_type->extension(i);
6100         if (extension->containing_type() == descriptor &&
6101             extension->type() == FieldDescriptor::TYPE_MESSAGE &&
6102             extension->is_optional() &&
6103             extension->message_type() == foreign_type) {
6104           // Found it.
6105           return extension;
6106         }
6107       }
6108     }
6109     return NULL;
6110   }
6111 };
6112 
6113 // A custom error collector to record any text-format parsing errors
6114 namespace {
6115 class AggregateErrorCollector : public io::ErrorCollector {
6116  public:
6117   string error_;
6118 
AddError(int,int,const string & message)6119   virtual void AddError(int /* line */, int /* column */,
6120                         const string& message) {
6121     if (!error_.empty()) {
6122       error_ += "; ";
6123     }
6124     error_ += message;
6125   }
6126 
AddWarning(int,int,const string &)6127   virtual void AddWarning(int /* line */, int /* column */,
6128                           const string& /* message */) {
6129     // Ignore warnings
6130   }
6131 };
6132 }
6133 
6134 // We construct a dynamic message of the type corresponding to
6135 // option_field, parse the supplied text-format string into this
6136 // message, and serialize the resulting message to produce the value.
SetAggregateOption(const FieldDescriptor * option_field,UnknownFieldSet * unknown_fields)6137 bool DescriptorBuilder::OptionInterpreter::SetAggregateOption(
6138     const FieldDescriptor* option_field,
6139     UnknownFieldSet* unknown_fields) {
6140   if (!uninterpreted_option_->has_aggregate_value()) {
6141     return AddValueError("Option \"" + option_field->full_name() +
6142                          "\" is a message. To set the entire message, use "
6143                          "syntax like \"" + option_field->name() +
6144                          " = { <proto text format> }\". "
6145                          "To set fields within it, use "
6146                          "syntax like \"" + option_field->name() +
6147                          ".foo = value\".");
6148   }
6149 
6150   const Descriptor* type = option_field->message_type();
6151   google::protobuf::scoped_ptr<Message> dynamic(dynamic_factory_.GetPrototype(type)->New());
6152   GOOGLE_CHECK(dynamic.get() != NULL)
6153       << "Could not create an instance of " << option_field->DebugString();
6154 
6155   AggregateErrorCollector collector;
6156   AggregateOptionFinder finder;
6157   finder.builder_ = builder_;
6158   TextFormat::Parser parser;
6159   parser.RecordErrorsTo(&collector);
6160   parser.SetFinder(&finder);
6161   if (!parser.ParseFromString(uninterpreted_option_->aggregate_value(),
6162                               dynamic.get())) {
6163     AddValueError("Error while parsing option value for \"" +
6164                   option_field->name() + "\": " + collector.error_);
6165     return false;
6166   } else {
6167     string serial;
6168     dynamic->SerializeToString(&serial);  // Never fails
6169     if (option_field->type() == FieldDescriptor::TYPE_MESSAGE) {
6170       unknown_fields->AddLengthDelimited(option_field->number(), serial);
6171     } else {
6172       GOOGLE_CHECK_EQ(option_field->type(),  FieldDescriptor::TYPE_GROUP);
6173       UnknownFieldSet* group = unknown_fields->AddGroup(option_field->number());
6174       group->ParseFromString(serial);
6175     }
6176     return true;
6177   }
6178 }
6179 
SetInt32(int number,int32 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)6180 void DescriptorBuilder::OptionInterpreter::SetInt32(int number, int32 value,
6181     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
6182   switch (type) {
6183     case FieldDescriptor::TYPE_INT32:
6184       unknown_fields->AddVarint(number,
6185         static_cast<uint64>(static_cast<int64>(value)));
6186       break;
6187 
6188     case FieldDescriptor::TYPE_SFIXED32:
6189       unknown_fields->AddFixed32(number, static_cast<uint32>(value));
6190       break;
6191 
6192     case FieldDescriptor::TYPE_SINT32:
6193       unknown_fields->AddVarint(number,
6194           google::protobuf::internal::WireFormatLite::ZigZagEncode32(value));
6195       break;
6196 
6197     default:
6198       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_INT32: " << type;
6199       break;
6200   }
6201 }
6202 
SetInt64(int number,int64 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)6203 void DescriptorBuilder::OptionInterpreter::SetInt64(int number, int64 value,
6204     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
6205   switch (type) {
6206     case FieldDescriptor::TYPE_INT64:
6207       unknown_fields->AddVarint(number, static_cast<uint64>(value));
6208       break;
6209 
6210     case FieldDescriptor::TYPE_SFIXED64:
6211       unknown_fields->AddFixed64(number, static_cast<uint64>(value));
6212       break;
6213 
6214     case FieldDescriptor::TYPE_SINT64:
6215       unknown_fields->AddVarint(number,
6216           google::protobuf::internal::WireFormatLite::ZigZagEncode64(value));
6217       break;
6218 
6219     default:
6220       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_INT64: " << type;
6221       break;
6222   }
6223 }
6224 
SetUInt32(int number,uint32 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)6225 void DescriptorBuilder::OptionInterpreter::SetUInt32(int number, uint32 value,
6226     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
6227   switch (type) {
6228     case FieldDescriptor::TYPE_UINT32:
6229       unknown_fields->AddVarint(number, static_cast<uint64>(value));
6230       break;
6231 
6232     case FieldDescriptor::TYPE_FIXED32:
6233       unknown_fields->AddFixed32(number, static_cast<uint32>(value));
6234       break;
6235 
6236     default:
6237       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_UINT32: " << type;
6238       break;
6239   }
6240 }
6241 
SetUInt64(int number,uint64 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)6242 void DescriptorBuilder::OptionInterpreter::SetUInt64(int number, uint64 value,
6243     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
6244   switch (type) {
6245     case FieldDescriptor::TYPE_UINT64:
6246       unknown_fields->AddVarint(number, value);
6247       break;
6248 
6249     case FieldDescriptor::TYPE_FIXED64:
6250       unknown_fields->AddFixed64(number, value);
6251       break;
6252 
6253     default:
6254       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_UINT64: " << type;
6255       break;
6256   }
6257 }
6258 
LogUnusedDependency(const FileDescriptorProto & proto,const FileDescriptor * result)6259 void DescriptorBuilder::LogUnusedDependency(const FileDescriptorProto& proto,
6260                                             const FileDescriptor* result) {
6261 
6262   if (!unused_dependency_.empty()) {
6263     std::set<string> annotation_extensions;
6264     annotation_extensions.insert("google.protobuf.MessageOptions");
6265     annotation_extensions.insert("google.protobuf.FileOptions");
6266     annotation_extensions.insert("google.protobuf.FieldOptions");
6267     annotation_extensions.insert("google.protobuf.EnumOptions");
6268     annotation_extensions.insert("google.protobuf.EnumValueOptions");
6269     annotation_extensions.insert("google.protobuf.ServiceOptions");
6270     annotation_extensions.insert("google.protobuf.MethodOptions");
6271     annotation_extensions.insert("google.protobuf.StreamOptions");
6272     for (set<const FileDescriptor*>::const_iterator
6273              it = unused_dependency_.begin();
6274          it != unused_dependency_.end(); ++it) {
6275       // Do not log warnings for proto files which extend annotations.
6276       int i;
6277       for (i = 0 ; i < (*it)->extension_count(); ++i) {
6278         if (annotation_extensions.find(
6279                 (*it)->extension(i)->containing_type()->full_name())
6280             != annotation_extensions.end()) {
6281           break;
6282         }
6283       }
6284       // Log warnings for unused imported files.
6285       if (i == (*it)->extension_count()) {
6286         string error_message = "Import " + (*it)->name() + " but not used.";
6287         AddWarning((*it)->name(), proto, DescriptorPool::ErrorCollector::OTHER,
6288                    error_message);
6289       }
6290     }
6291   }
6292 }
6293 
6294 }  // namespace protobuf
6295 }  // namespace google
6296