• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Protocol Buffers - Google's data interchange format
2 // Copyright 2008 Google Inc.  All rights reserved.
3 // http://code.google.com/p/protobuf/
4 //
5 // Redistribution and use in source and binary forms, with or without
6 // modification, are permitted provided that the following conditions are
7 // met:
8 //
9 //     * Redistributions of source code must retain the above copyright
10 // notice, this list of conditions and the following disclaimer.
11 //     * Redistributions in binary form must reproduce the above
12 // copyright notice, this list of conditions and the following disclaimer
13 // in the documentation and/or other materials provided with the
14 // distribution.
15 //     * Neither the name of Google Inc. nor the names of its
16 // contributors may be used to endorse or promote products derived from
17 // this software without specific prior written permission.
18 //
19 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 
31 // Author: kenton@google.com (Kenton Varda)
32 //  Based on original Protocol Buffers design by
33 //  Sanjay Ghemawat, Jeff Dean, and others.
34 
35 #include <google/protobuf/stubs/hash.h>
36 #include <map>
37 #include <set>
38 #include <vector>
39 #include <algorithm>
40 #include <limits>
41 
42 #include <google/protobuf/descriptor.h>
43 #include <google/protobuf/descriptor_database.h>
44 #include <google/protobuf/descriptor.pb.h>
45 #include <google/protobuf/text_format.h>
46 #include <google/protobuf/unknown_field_set.h>
47 #include <google/protobuf/wire_format.h>
48 #include <google/protobuf/io/coded_stream.h>
49 #include <google/protobuf/io/zero_copy_stream_impl.h>
50 #include <google/protobuf/stubs/common.h>
51 #include <google/protobuf/stubs/once.h>
52 #include <google/protobuf/stubs/strutil.h>
53 #include <google/protobuf/stubs/substitute.h>
54 #include <google/protobuf/stubs/map-util.h>
55 #include <google/protobuf/stubs/stl_util-inl.h>
56 
57 #undef PACKAGE  // autoheader #defines this.  :(
58 
59 namespace google {
60 namespace protobuf {
61 
62 const FieldDescriptor::CppType
63 FieldDescriptor::kTypeToCppTypeMap[MAX_TYPE + 1] = {
64   static_cast<CppType>(0),  // 0 is reserved for errors
65 
66   CPPTYPE_DOUBLE,   // TYPE_DOUBLE
67   CPPTYPE_FLOAT,    // TYPE_FLOAT
68   CPPTYPE_INT64,    // TYPE_INT64
69   CPPTYPE_UINT64,   // TYPE_UINT64
70   CPPTYPE_INT32,    // TYPE_INT32
71   CPPTYPE_UINT64,   // TYPE_FIXED64
72   CPPTYPE_UINT32,   // TYPE_FIXED32
73   CPPTYPE_BOOL,     // TYPE_BOOL
74   CPPTYPE_STRING,   // TYPE_STRING
75   CPPTYPE_MESSAGE,  // TYPE_GROUP
76   CPPTYPE_MESSAGE,  // TYPE_MESSAGE
77   CPPTYPE_STRING,   // TYPE_BYTES
78   CPPTYPE_UINT32,   // TYPE_UINT32
79   CPPTYPE_ENUM,     // TYPE_ENUM
80   CPPTYPE_INT32,    // TYPE_SFIXED32
81   CPPTYPE_INT64,    // TYPE_SFIXED64
82   CPPTYPE_INT32,    // TYPE_SINT32
83   CPPTYPE_INT64,    // TYPE_SINT64
84 };
85 
86 const char * const FieldDescriptor::kTypeToName[MAX_TYPE + 1] = {
87   "ERROR",     // 0 is reserved for errors
88 
89   "double",    // TYPE_DOUBLE
90   "float",     // TYPE_FLOAT
91   "int64",     // TYPE_INT64
92   "uint64",    // TYPE_UINT64
93   "int32",     // TYPE_INT32
94   "fixed64",   // TYPE_FIXED64
95   "fixed32",   // TYPE_FIXED32
96   "bool",      // TYPE_BOOL
97   "string",    // TYPE_STRING
98   "group",     // TYPE_GROUP
99   "message",   // TYPE_MESSAGE
100   "bytes",     // TYPE_BYTES
101   "uint32",    // TYPE_UINT32
102   "enum",      // TYPE_ENUM
103   "sfixed32",  // TYPE_SFIXED32
104   "sfixed64",  // TYPE_SFIXED64
105   "sint32",    // TYPE_SINT32
106   "sint64",    // TYPE_SINT64
107 };
108 
109 const char * const FieldDescriptor::kLabelToName[MAX_LABEL + 1] = {
110   "ERROR",     // 0 is reserved for errors
111 
112   "optional",  // LABEL_OPTIONAL
113   "required",  // LABEL_REQUIRED
114   "repeated",  // LABEL_REPEATED
115 };
116 
117 #ifndef _MSC_VER  // MSVC doesn't need these and won't even accept them.
118 const int FieldDescriptor::kMaxNumber;
119 const int FieldDescriptor::kFirstReservedNumber;
120 const int FieldDescriptor::kLastReservedNumber;
121 #endif
122 
123 namespace {
124 
125 const string kEmptyString;
126 
ToCamelCase(const string & input)127 string ToCamelCase(const string& input) {
128   bool capitalize_next = false;
129   string result;
130   result.reserve(input.size());
131 
132   for (int i = 0; i < input.size(); i++) {
133     if (input[i] == '_') {
134       capitalize_next = true;
135     } else if (capitalize_next) {
136       // Note:  I distrust ctype.h due to locales.
137       if ('a' <= input[i] && input[i] <= 'z') {
138         result.push_back(input[i] - 'a' + 'A');
139       } else {
140         result.push_back(input[i]);
141       }
142       capitalize_next = false;
143     } else {
144       result.push_back(input[i]);
145     }
146   }
147 
148   // Lower-case the first letter.
149   if (!result.empty() && 'A' <= result[0] && result[0] <= 'Z') {
150     result[0] = result[0] - 'A' + 'a';
151   }
152 
153   return result;
154 }
155 
156 // A DescriptorPool contains a bunch of hash_maps to implement the
157 // various Find*By*() methods.  Since hashtable lookups are O(1), it's
158 // most efficient to construct a fixed set of large hash_maps used by
159 // all objects in the pool rather than construct one or more small
160 // hash_maps for each object.
161 //
162 // The keys to these hash_maps are (parent, name) or (parent, number)
163 // pairs.  Unfortunately STL doesn't provide hash functions for pair<>,
164 // so we must invent our own.
165 //
166 // TODO(kenton):  Use StringPiece rather than const char* in keys?  It would
167 //   be a lot cleaner but we'd just have to convert it back to const char*
168 //   for the open source release.
169 
170 typedef pair<const void*, const char*> PointerStringPair;
171 
172 struct PointerStringPairEqual {
operator ()google::protobuf::__anona61a9b310111::PointerStringPairEqual173   inline bool operator()(const PointerStringPair& a,
174                          const PointerStringPair& b) const {
175     return a.first == b.first && strcmp(a.second, b.second) == 0;
176   }
177 };
178 
179 template<typename PairType>
180 struct PointerIntegerPairHash {
operator ()google::protobuf::__anona61a9b310111::PointerIntegerPairHash181   size_t operator()(const PairType& p) const {
182     // FIXME(kenton):  What is the best way to compute this hash?  I have
183     // no idea!  This seems a bit better than an XOR.
184     return reinterpret_cast<intptr_t>(p.first) * ((1 << 16) - 1) + p.second;
185   }
186 
187   // Used only by MSVC and platforms where hash_map is not available.
188   static const size_t bucket_size = 4;
189   static const size_t min_buckets = 8;
operator ()google::protobuf::__anona61a9b310111::PointerIntegerPairHash190   inline bool operator()(const PairType& a, const PairType& b) const {
191     return a.first < b.first ||
192           (a.first == b.first && a.second < b.second);
193   }
194 };
195 
196 typedef pair<const Descriptor*, int> DescriptorIntPair;
197 typedef pair<const EnumDescriptor*, int> EnumIntPair;
198 
199 struct PointerStringPairHash {
operator ()google::protobuf::__anona61a9b310111::PointerStringPairHash200   size_t operator()(const PointerStringPair& p) const {
201     // FIXME(kenton):  What is the best way to compute this hash?  I have
202     // no idea!  This seems a bit better than an XOR.
203     hash<const char*> cstring_hash;
204     return reinterpret_cast<intptr_t>(p.first) * ((1 << 16) - 1) +
205            cstring_hash(p.second);
206   }
207 
208   // Used only by MSVC and platforms where hash_map is not available.
209   static const size_t bucket_size = 4;
210   static const size_t min_buckets = 8;
operator ()google::protobuf::__anona61a9b310111::PointerStringPairHash211   inline bool operator()(const PointerStringPair& a,
212                          const PointerStringPair& b) const {
213     if (a.first < b.first) return true;
214     if (a.first > b.first) return false;
215     return strcmp(a.second, b.second) < 0;
216   }
217 };
218 
219 
220 struct Symbol {
221   enum Type {
222     NULL_SYMBOL, MESSAGE, FIELD, ENUM, ENUM_VALUE, SERVICE, METHOD, PACKAGE
223   };
224   Type type;
225   union {
226     const Descriptor* descriptor;
227     const FieldDescriptor* field_descriptor;
228     const EnumDescriptor* enum_descriptor;
229     const EnumValueDescriptor* enum_value_descriptor;
230     const ServiceDescriptor* service_descriptor;
231     const MethodDescriptor* method_descriptor;
232     const FileDescriptor* package_file_descriptor;
233   };
234 
Symbolgoogle::protobuf::__anona61a9b310111::Symbol235   inline Symbol() : type(NULL_SYMBOL) { descriptor = NULL; }
IsNullgoogle::protobuf::__anona61a9b310111::Symbol236   inline bool IsNull() const { return type == NULL_SYMBOL; }
IsTypegoogle::protobuf::__anona61a9b310111::Symbol237   inline bool IsType() const {
238     return type == MESSAGE || type == ENUM;
239   }
IsAggregategoogle::protobuf::__anona61a9b310111::Symbol240   inline bool IsAggregate() const {
241     return type == MESSAGE || type == PACKAGE
242         || type == ENUM || type == SERVICE;
243   }
244 
245 #define CONSTRUCTOR(TYPE, TYPE_CONSTANT, FIELD)  \
246   inline explicit Symbol(const TYPE* value) {    \
247     type = TYPE_CONSTANT;                        \
248     this->FIELD = value;                         \
249   }
250 
CONSTRUCTORgoogle::protobuf::__anona61a9b310111::Symbol251   CONSTRUCTOR(Descriptor         , MESSAGE   , descriptor             )
252   CONSTRUCTOR(FieldDescriptor    , FIELD     , field_descriptor       )
253   CONSTRUCTOR(EnumDescriptor     , ENUM      , enum_descriptor        )
254   CONSTRUCTOR(EnumValueDescriptor, ENUM_VALUE, enum_value_descriptor  )
255   CONSTRUCTOR(ServiceDescriptor  , SERVICE   , service_descriptor     )
256   CONSTRUCTOR(MethodDescriptor   , METHOD    , method_descriptor      )
257   CONSTRUCTOR(FileDescriptor     , PACKAGE   , package_file_descriptor)
258 #undef CONSTRUCTOR
259 
260   const FileDescriptor* GetFile() const {
261     switch (type) {
262       case NULL_SYMBOL: return NULL;
263       case MESSAGE    : return descriptor           ->file();
264       case FIELD      : return field_descriptor     ->file();
265       case ENUM       : return enum_descriptor      ->file();
266       case ENUM_VALUE : return enum_value_descriptor->type()->file();
267       case SERVICE    : return service_descriptor   ->file();
268       case METHOD     : return method_descriptor    ->service()->file();
269       case PACKAGE    : return package_file_descriptor;
270     }
271     return NULL;
272   }
273 };
274 
275 const Symbol kNullSymbol;
276 
277 typedef hash_map<const char*, Symbol,
278                  hash<const char*>, streq>
279   SymbolsByNameMap;
280 typedef hash_map<PointerStringPair, Symbol,
281                  PointerStringPairHash, PointerStringPairEqual>
282   SymbolsByParentMap;
283 typedef hash_map<const char*, const FileDescriptor*,
284                  hash<const char*>, streq>
285   FilesByNameMap;
286 typedef hash_map<PointerStringPair, const FieldDescriptor*,
287                  PointerStringPairHash, PointerStringPairEqual>
288   FieldsByNameMap;
289 typedef hash_map<DescriptorIntPair, const FieldDescriptor*,
290                  PointerIntegerPairHash<DescriptorIntPair> >
291   FieldsByNumberMap;
292 typedef hash_map<EnumIntPair, const EnumValueDescriptor*,
293                  PointerIntegerPairHash<EnumIntPair> >
294   EnumValuesByNumberMap;
295 // This is a map rather than a hash_map, since we use it to iterate
296 // through all the extensions that extend a given Descriptor, and an
297 // ordered data structure that implements lower_bound is convenient
298 // for that.
299 typedef map<DescriptorIntPair, const FieldDescriptor*>
300   ExtensionsGroupedByDescriptorMap;
301 
302 }  // anonymous namespace
303 
304 // ===================================================================
305 // DescriptorPool::Tables
306 
307 class DescriptorPool::Tables {
308  public:
309   Tables();
310   ~Tables();
311 
312   // Checkpoint the state of the tables.  Future calls to Rollback() will
313   // return the Tables to this state.  This is used when building files, since
314   // some kinds of validation errors cannot be detected until the file's
315   // descriptors have already been added to the tables.  BuildFile() calls
316   // Checkpoint() before it starts building and Rollback() if it encounters
317   // an error.
318   void Checkpoint();
319 
320   // Roll back the Tables to the state of the last Checkpoint(), removing
321   // everything that was added after that point.
322   void Rollback();
323 
324   // The stack of files which are currently being built.  Used to detect
325   // cyclic dependencies when loading files from a DescriptorDatabase.  Not
326   // used when fallback_database_ == NULL.
327   vector<string> pending_files_;
328 
329   // A set of files which we have tried to load from the fallback database
330   // and encountered errors.  We will not attempt to load them again.
331   // Not used when fallback_database_ == NULL.
332   hash_set<string> known_bad_files_;
333 
334   // The set of descriptors for which we've already loaded the full
335   // set of extensions numbers from fallback_database_.
336   hash_set<const Descriptor*> extensions_loaded_from_db_;
337 
338   // -----------------------------------------------------------------
339   // Finding items.
340 
341   // Find symbols.  This returns a null Symbol (symbol.IsNull() is true)
342   // if not found.
343   inline Symbol FindSymbol(const string& key) const;
344 
345   // This implements the body of DescriptorPool::Find*ByName().  It should
346   // really be a private method of DescriptorPool, but that would require
347   // declaring Symbol in descriptor.h, which would drag all kinds of other
348   // stuff into the header.  Yay C++.
349   Symbol FindByNameHelper(
350     const DescriptorPool* pool, const string& name) const;
351 
352   // These return NULL if not found.
353   inline const FileDescriptor* FindFile(const string& key) const;
354   inline const FieldDescriptor* FindExtension(const Descriptor* extendee,
355                                               int number);
356   inline void FindAllExtensions(const Descriptor* extendee,
357                                 vector<const FieldDescriptor*>* out) const;
358 
359   // -----------------------------------------------------------------
360   // Adding items.
361 
362   // These add items to the corresponding tables.  They return false if
363   // the key already exists in the table.  For AddSymbol(), the string passed
364   // in must be one that was constructed using AllocateString(), as it will
365   // be used as a key in the symbols_by_name_ map without copying.
366   bool AddSymbol(const string& full_name, Symbol symbol);
367   bool AddFile(const FileDescriptor* file);
368   bool AddExtension(const FieldDescriptor* field);
369 
370   // -----------------------------------------------------------------
371   // Allocating memory.
372 
373   // Allocate an object which will be reclaimed when the pool is
374   // destroyed.  Note that the object's destructor will never be called,
375   // so its fields must be plain old data (primitive data types and
376   // pointers).  All of the descriptor types are such objects.
377   template<typename Type> Type* Allocate();
378 
379   // Allocate an array of objects which will be reclaimed when the
380   // pool in destroyed.  Again, destructors are never called.
381   template<typename Type> Type* AllocateArray(int count);
382 
383   // Allocate a string which will be destroyed when the pool is destroyed.
384   // The string is initialized to the given value for convenience.
385   string* AllocateString(const string& value);
386 
387   // Allocate a protocol message object.  Some older versions of GCC have
388   // trouble understanding explicit template instantiations in some cases, so
389   // in those cases we have to pass a dummy pointer of the right type as the
390   // parameter instead of specifying the type explicitly.
391   template<typename Type> Type* AllocateMessage(Type* dummy = NULL);
392 
393   // Allocate a FileDescriptorTables object.
394   FileDescriptorTables* AllocateFileTables();
395 
396  private:
397   vector<string*> strings_;    // All strings in the pool.
398   vector<Message*> messages_;  // All messages in the pool.
399   vector<FileDescriptorTables*> file_tables_;  // All file tables in the pool.
400   vector<void*> allocations_;  // All other memory allocated in the pool.
401 
402   SymbolsByNameMap      symbols_by_name_;
403   FilesByNameMap        files_by_name_;
404   ExtensionsGroupedByDescriptorMap extensions_;
405 
406   int strings_before_checkpoint_;
407   int messages_before_checkpoint_;
408   int file_tables_before_checkpoint_;
409   int allocations_before_checkpoint_;
410   vector<const char*      > symbols_after_checkpoint_;
411   vector<const char*      > files_after_checkpoint_;
412   vector<DescriptorIntPair> extensions_after_checkpoint_;
413 
414   // Allocate some bytes which will be reclaimed when the pool is
415   // destroyed.
416   void* AllocateBytes(int size);
417 };
418 
419 // Contains tables specific to a particular file.  These tables are not
420 // modified once the file has been constructed, so they need not be
421 // protected by a mutex.  This makes operations that depend only on the
422 // contents of a single file -- e.g. Descriptor::FindFieldByName() --
423 // lock-free.
424 //
425 // For historical reasons, the definitions of the methods of
426 // FileDescriptorTables and DescriptorPool::Tables are interleaved below.
427 // These used to be a single class.
428 class FileDescriptorTables {
429  public:
430   FileDescriptorTables();
431   ~FileDescriptorTables();
432 
433   // Empty table, used with placeholder files.
434   static const FileDescriptorTables kEmpty;
435 
436   // -----------------------------------------------------------------
437   // Finding items.
438 
439   // Find symbols.  These return a null Symbol (symbol.IsNull() is true)
440   // if not found.
441   inline Symbol FindNestedSymbol(const void* parent,
442                                  const string& name) const;
443   inline Symbol FindNestedSymbolOfType(const void* parent,
444                                        const string& name,
445                                        const Symbol::Type type) const;
446 
447   // These return NULL if not found.
448   inline const FieldDescriptor* FindFieldByNumber(
449     const Descriptor* parent, int number) const;
450   inline const FieldDescriptor* FindFieldByLowercaseName(
451     const void* parent, const string& lowercase_name) const;
452   inline const FieldDescriptor* FindFieldByCamelcaseName(
453     const void* parent, const string& camelcase_name) const;
454   inline const EnumValueDescriptor* FindEnumValueByNumber(
455     const EnumDescriptor* parent, int number) const;
456 
457   // -----------------------------------------------------------------
458   // Adding items.
459 
460   // These add items to the corresponding tables.  They return false if
461   // the key already exists in the table.  For AddAliasUnderParent(), the
462   // string passed in must be one that was constructed using AllocateString(),
463   // as it will be used as a key in the symbols_by_parent_ map without copying.
464   bool AddAliasUnderParent(const void* parent, const string& name,
465                            Symbol symbol);
466   bool AddFieldByNumber(const FieldDescriptor* field);
467   bool AddEnumValueByNumber(const EnumValueDescriptor* value);
468 
469   // Adds the field to the lowercase_name and camelcase_name maps.  Never
470   // fails because we allow duplicates; the first field by the name wins.
471   void AddFieldByStylizedNames(const FieldDescriptor* field);
472 
473  private:
474   SymbolsByParentMap    symbols_by_parent_;
475   FieldsByNameMap       fields_by_lowercase_name_;
476   FieldsByNameMap       fields_by_camelcase_name_;
477   FieldsByNumberMap     fields_by_number_;       // Not including extensions.
478   EnumValuesByNumberMap enum_values_by_number_;
479 };
480 
Tables()481 DescriptorPool::Tables::Tables()
482   : strings_before_checkpoint_(0),
483     messages_before_checkpoint_(0),
484     allocations_before_checkpoint_(0) {}
485 
~Tables()486 DescriptorPool::Tables::~Tables() {
487   // Note that the deletion order is important, since the destructors of some
488   // messages may refer to objects in allocations_.
489   STLDeleteElements(&messages_);
490   for (int i = 0; i < allocations_.size(); i++) {
491     operator delete(allocations_[i]);
492   }
493   STLDeleteElements(&strings_);
494   STLDeleteElements(&file_tables_);
495 }
496 
FileDescriptorTables()497 FileDescriptorTables::FileDescriptorTables() {}
~FileDescriptorTables()498 FileDescriptorTables::~FileDescriptorTables() {}
499 
500 const FileDescriptorTables FileDescriptorTables::kEmpty;
501 
Checkpoint()502 void DescriptorPool::Tables::Checkpoint() {
503   strings_before_checkpoint_ = strings_.size();
504   messages_before_checkpoint_ = messages_.size();
505   file_tables_before_checkpoint_ = file_tables_.size();
506   allocations_before_checkpoint_ = allocations_.size();
507 
508   symbols_after_checkpoint_.clear();
509   files_after_checkpoint_.clear();
510   extensions_after_checkpoint_.clear();
511 }
512 
Rollback()513 void DescriptorPool::Tables::Rollback() {
514   for (int i = 0; i < symbols_after_checkpoint_.size(); i++) {
515     symbols_by_name_.erase(symbols_after_checkpoint_[i]);
516   }
517   for (int i = 0; i < files_after_checkpoint_.size(); i++) {
518     files_by_name_.erase(files_after_checkpoint_[i]);
519   }
520   for (int i = 0; i < extensions_after_checkpoint_.size(); i++) {
521     extensions_.erase(extensions_after_checkpoint_[i]);
522   }
523 
524   symbols_after_checkpoint_.clear();
525   files_after_checkpoint_.clear();
526   extensions_after_checkpoint_.clear();
527 
528   STLDeleteContainerPointers(
529     strings_.begin() + strings_before_checkpoint_, strings_.end());
530   STLDeleteContainerPointers(
531     messages_.begin() + messages_before_checkpoint_, messages_.end());
532   STLDeleteContainerPointers(
533     file_tables_.begin() + file_tables_before_checkpoint_, file_tables_.end());
534   for (int i = allocations_before_checkpoint_; i < allocations_.size(); i++) {
535     operator delete(allocations_[i]);
536   }
537 
538   strings_.resize(strings_before_checkpoint_);
539   messages_.resize(messages_before_checkpoint_);
540   file_tables_.resize(file_tables_before_checkpoint_);
541   allocations_.resize(allocations_before_checkpoint_);
542 }
543 
544 // -------------------------------------------------------------------
545 
FindSymbol(const string & key) const546 inline Symbol DescriptorPool::Tables::FindSymbol(const string& key) const {
547   const Symbol* result = FindOrNull(symbols_by_name_, key.c_str());
548   if (result == NULL) {
549     return kNullSymbol;
550   } else {
551     return *result;
552   }
553 }
554 
FindNestedSymbol(const void * parent,const string & name) const555 inline Symbol FileDescriptorTables::FindNestedSymbol(
556     const void* parent, const string& name) const {
557   const Symbol* result =
558     FindOrNull(symbols_by_parent_, PointerStringPair(parent, name.c_str()));
559   if (result == NULL) {
560     return kNullSymbol;
561   } else {
562     return *result;
563   }
564 }
565 
FindNestedSymbolOfType(const void * parent,const string & name,const Symbol::Type type) const566 inline Symbol FileDescriptorTables::FindNestedSymbolOfType(
567     const void* parent, const string& name, const Symbol::Type type) const {
568   Symbol result = FindNestedSymbol(parent, name);
569   if (result.type != type) return kNullSymbol;
570   return result;
571 }
572 
FindByNameHelper(const DescriptorPool * pool,const string & name) const573 Symbol DescriptorPool::Tables::FindByNameHelper(
574     const DescriptorPool* pool, const string& name) const {
575   MutexLockMaybe lock(pool->mutex_);
576   Symbol result = FindSymbol(name);
577 
578   if (result.IsNull() && pool->underlay_ != NULL) {
579     // Symbol not found; check the underlay.
580     result =
581       pool->underlay_->tables_->FindByNameHelper(pool->underlay_, name);
582   }
583 
584   if (result.IsNull()) {
585     // Symbol still not found, so check fallback database.
586     if (pool->TryFindSymbolInFallbackDatabase(name)) {
587       result = FindSymbol(name);
588     }
589   }
590 
591   return result;
592 }
593 
FindFile(const string & key) const594 inline const FileDescriptor* DescriptorPool::Tables::FindFile(
595     const string& key) const {
596   return FindPtrOrNull(files_by_name_, key.c_str());
597 }
598 
FindFieldByNumber(const Descriptor * parent,int number) const599 inline const FieldDescriptor* FileDescriptorTables::FindFieldByNumber(
600     const Descriptor* parent, int number) const {
601   return FindPtrOrNull(fields_by_number_, make_pair(parent, number));
602 }
603 
FindFieldByLowercaseName(const void * parent,const string & lowercase_name) const604 inline const FieldDescriptor* FileDescriptorTables::FindFieldByLowercaseName(
605     const void* parent, const string& lowercase_name) const {
606   return FindPtrOrNull(fields_by_lowercase_name_,
607                        PointerStringPair(parent, lowercase_name.c_str()));
608 }
609 
FindFieldByCamelcaseName(const void * parent,const string & camelcase_name) const610 inline const FieldDescriptor* FileDescriptorTables::FindFieldByCamelcaseName(
611     const void* parent, const string& camelcase_name) const {
612   return FindPtrOrNull(fields_by_camelcase_name_,
613                        PointerStringPair(parent, camelcase_name.c_str()));
614 }
615 
FindEnumValueByNumber(const EnumDescriptor * parent,int number) const616 inline const EnumValueDescriptor* FileDescriptorTables::FindEnumValueByNumber(
617     const EnumDescriptor* parent, int number) const {
618   return FindPtrOrNull(enum_values_by_number_, make_pair(parent, number));
619 }
620 
FindExtension(const Descriptor * extendee,int number)621 inline const FieldDescriptor* DescriptorPool::Tables::FindExtension(
622     const Descriptor* extendee, int number) {
623   return FindPtrOrNull(extensions_, make_pair(extendee, number));
624 }
625 
FindAllExtensions(const Descriptor * extendee,vector<const FieldDescriptor * > * out) const626 inline void DescriptorPool::Tables::FindAllExtensions(
627     const Descriptor* extendee, vector<const FieldDescriptor*>* out) const {
628   ExtensionsGroupedByDescriptorMap::const_iterator it =
629       extensions_.lower_bound(make_pair(extendee, 0));
630   for (; it != extensions_.end() && it->first.first == extendee; ++it) {
631     out->push_back(it->second);
632   }
633 }
634 
635 // -------------------------------------------------------------------
636 
AddSymbol(const string & full_name,Symbol symbol)637 bool DescriptorPool::Tables::AddSymbol(
638     const string& full_name, Symbol symbol) {
639   if (InsertIfNotPresent(&symbols_by_name_, full_name.c_str(), symbol)) {
640     symbols_after_checkpoint_.push_back(full_name.c_str());
641     return true;
642   } else {
643     return false;
644   }
645 }
646 
AddAliasUnderParent(const void * parent,const string & name,Symbol symbol)647 bool FileDescriptorTables::AddAliasUnderParent(
648     const void* parent, const string& name, Symbol symbol) {
649   PointerStringPair by_parent_key(parent, name.c_str());
650   return InsertIfNotPresent(&symbols_by_parent_, by_parent_key, symbol);
651 }
652 
AddFile(const FileDescriptor * file)653 bool DescriptorPool::Tables::AddFile(const FileDescriptor* file) {
654   if (InsertIfNotPresent(&files_by_name_, file->name().c_str(), file)) {
655     files_after_checkpoint_.push_back(file->name().c_str());
656     return true;
657   } else {
658     return false;
659   }
660 }
661 
AddFieldByStylizedNames(const FieldDescriptor * field)662 void FileDescriptorTables::AddFieldByStylizedNames(
663     const FieldDescriptor* field) {
664   const void* parent;
665   if (field->is_extension()) {
666     if (field->extension_scope() == NULL) {
667       parent = field->file();
668     } else {
669       parent = field->extension_scope();
670     }
671   } else {
672     parent = field->containing_type();
673   }
674 
675   PointerStringPair lowercase_key(parent, field->lowercase_name().c_str());
676   InsertIfNotPresent(&fields_by_lowercase_name_, lowercase_key, field);
677 
678   PointerStringPair camelcase_key(parent, field->camelcase_name().c_str());
679   InsertIfNotPresent(&fields_by_camelcase_name_, camelcase_key, field);
680 }
681 
AddFieldByNumber(const FieldDescriptor * field)682 bool FileDescriptorTables::AddFieldByNumber(const FieldDescriptor* field) {
683   DescriptorIntPair key(field->containing_type(), field->number());
684   return InsertIfNotPresent(&fields_by_number_, key, field);
685 }
686 
AddEnumValueByNumber(const EnumValueDescriptor * value)687 bool FileDescriptorTables::AddEnumValueByNumber(
688     const EnumValueDescriptor* value) {
689   EnumIntPair key(value->type(), value->number());
690   return InsertIfNotPresent(&enum_values_by_number_, key, value);
691 }
692 
AddExtension(const FieldDescriptor * field)693 bool DescriptorPool::Tables::AddExtension(const FieldDescriptor* field) {
694   DescriptorIntPair key(field->containing_type(), field->number());
695   if (InsertIfNotPresent(&extensions_, key, field)) {
696     extensions_after_checkpoint_.push_back(key);
697     return true;
698   } else {
699     return false;
700   }
701 }
702 
703 // -------------------------------------------------------------------
704 
705 template<typename Type>
Allocate()706 Type* DescriptorPool::Tables::Allocate() {
707   return reinterpret_cast<Type*>(AllocateBytes(sizeof(Type)));
708 }
709 
710 template<typename Type>
AllocateArray(int count)711 Type* DescriptorPool::Tables::AllocateArray(int count) {
712   return reinterpret_cast<Type*>(AllocateBytes(sizeof(Type) * count));
713 }
714 
AllocateString(const string & value)715 string* DescriptorPool::Tables::AllocateString(const string& value) {
716   string* result = new string(value);
717   strings_.push_back(result);
718   return result;
719 }
720 
721 template<typename Type>
AllocateMessage(Type * dummy)722 Type* DescriptorPool::Tables::AllocateMessage(Type* dummy) {
723   Type* result = new Type;
724   messages_.push_back(result);
725   return result;
726 }
727 
AllocateFileTables()728 FileDescriptorTables* DescriptorPool::Tables::AllocateFileTables() {
729   FileDescriptorTables* result = new FileDescriptorTables;
730   file_tables_.push_back(result);
731   return result;
732 }
733 
AllocateBytes(int size)734 void* DescriptorPool::Tables::AllocateBytes(int size) {
735   // TODO(kenton):  Would it be worthwhile to implement this in some more
736   // sophisticated way?  Probably not for the open source release, but for
737   // internal use we could easily plug in one of our existing memory pool
738   // allocators...
739   if (size == 0) return NULL;
740 
741   void* result = operator new(size);
742   allocations_.push_back(result);
743   return result;
744 }
745 
746 // ===================================================================
747 // DescriptorPool
748 
~ErrorCollector()749 DescriptorPool::ErrorCollector::~ErrorCollector() {}
750 
DescriptorPool()751 DescriptorPool::DescriptorPool()
752   : mutex_(NULL),
753     fallback_database_(NULL),
754     default_error_collector_(NULL),
755     underlay_(NULL),
756     tables_(new Tables),
757     enforce_dependencies_(true),
758     allow_unknown_(false) {}
759 
DescriptorPool(DescriptorDatabase * fallback_database,ErrorCollector * error_collector)760 DescriptorPool::DescriptorPool(DescriptorDatabase* fallback_database,
761                                ErrorCollector* error_collector)
762   : mutex_(new Mutex),
763     fallback_database_(fallback_database),
764     default_error_collector_(error_collector),
765     underlay_(NULL),
766     tables_(new Tables),
767     enforce_dependencies_(true),
768     allow_unknown_(false) {
769 }
770 
DescriptorPool(const DescriptorPool * underlay)771 DescriptorPool::DescriptorPool(const DescriptorPool* underlay)
772   : mutex_(NULL),
773     fallback_database_(NULL),
774     default_error_collector_(NULL),
775     underlay_(underlay),
776     tables_(new Tables),
777     enforce_dependencies_(true),
778     allow_unknown_(false) {}
779 
~DescriptorPool()780 DescriptorPool::~DescriptorPool() {
781   if (mutex_ != NULL) delete mutex_;
782 }
783 
784 // DescriptorPool::BuildFile() defined later.
785 // DescriptorPool::BuildFileCollectingErrors() defined later.
786 
InternalDontEnforceDependencies()787 void DescriptorPool::InternalDontEnforceDependencies() {
788   enforce_dependencies_ = false;
789 }
790 
InternalIsFileLoaded(const string & filename) const791 bool DescriptorPool::InternalIsFileLoaded(const string& filename) const {
792   MutexLockMaybe lock(mutex_);
793   return tables_->FindFile(filename) != NULL;
794 }
795 
796 // generated_pool ====================================================
797 
798 namespace {
799 
800 
801 EncodedDescriptorDatabase* generated_database_ = NULL;
802 DescriptorPool* generated_pool_ = NULL;
803 GOOGLE_PROTOBUF_DECLARE_ONCE(generated_pool_init_);
804 
DeleteGeneratedPool()805 void DeleteGeneratedPool() {
806   delete generated_database_;
807   generated_database_ = NULL;
808   delete generated_pool_;
809   generated_pool_ = NULL;
810 }
811 
InitGeneratedPool()812 void InitGeneratedPool() {
813   generated_database_ = new EncodedDescriptorDatabase;
814   generated_pool_ = new DescriptorPool(generated_database_);
815 
816   internal::OnShutdown(&DeleteGeneratedPool);
817 }
818 
InitGeneratedPoolOnce()819 inline void InitGeneratedPoolOnce() {
820   ::google::protobuf::GoogleOnceInit(&generated_pool_init_, &InitGeneratedPool);
821 }
822 
823 }  // anonymous namespace
824 
generated_pool()825 const DescriptorPool* DescriptorPool::generated_pool() {
826   InitGeneratedPoolOnce();
827   return generated_pool_;
828 }
829 
internal_generated_pool()830 DescriptorPool* DescriptorPool::internal_generated_pool() {
831   InitGeneratedPoolOnce();
832   return generated_pool_;
833 }
834 
InternalAddGeneratedFile(const void * encoded_file_descriptor,int size)835 void DescriptorPool::InternalAddGeneratedFile(
836     const void* encoded_file_descriptor, int size) {
837   // So, this function is called in the process of initializing the
838   // descriptors for generated proto classes.  Each generated .pb.cc file
839   // has an internal procedure called AddDescriptors() which is called at
840   // process startup, and that function calls this one in order to register
841   // the raw bytes of the FileDescriptorProto representing the file.
842   //
843   // We do not actually construct the descriptor objects right away.  We just
844   // hang on to the bytes until they are actually needed.  We actually construct
845   // the descriptor the first time one of the following things happens:
846   // * Someone calls a method like descriptor(), GetDescriptor(), or
847   //   GetReflection() on the generated types, which requires returning the
848   //   descriptor or an object based on it.
849   // * Someone looks up the descriptor in DescriptorPool::generated_pool().
850   //
851   // Once one of these happens, the DescriptorPool actually parses the
852   // FileDescriptorProto and generates a FileDescriptor (and all its children)
853   // based on it.
854   //
855   // Note that FileDescriptorProto is itself a generated protocol message.
856   // Therefore, when we parse one, we have to be very careful to avoid using
857   // any descriptor-based operations, since this might cause infinite recursion
858   // or deadlock.
859   InitGeneratedPoolOnce();
860   GOOGLE_CHECK(generated_database_->Add(encoded_file_descriptor, size));
861 }
862 
863 
864 // Find*By* methods ==================================================
865 
866 // TODO(kenton):  There's a lot of repeated code here, but I'm not sure if
867 //   there's any good way to factor it out.  Think about this some time when
868 //   there's nothing more important to do (read: never).
869 
FindFileByName(const string & name) const870 const FileDescriptor* DescriptorPool::FindFileByName(const string& name) const {
871   MutexLockMaybe lock(mutex_);
872   const FileDescriptor* result = tables_->FindFile(name);
873   if (result != NULL) return result;
874   if (underlay_ != NULL) {
875     const FileDescriptor* result = underlay_->FindFileByName(name);
876     if (result != NULL) return result;
877   }
878   if (TryFindFileInFallbackDatabase(name)) {
879     const FileDescriptor* result = tables_->FindFile(name);
880     if (result != NULL) return result;
881   }
882   return NULL;
883 }
884 
FindFileContainingSymbol(const string & symbol_name) const885 const FileDescriptor* DescriptorPool::FindFileContainingSymbol(
886     const string& symbol_name) const {
887   MutexLockMaybe lock(mutex_);
888   Symbol result = tables_->FindSymbol(symbol_name);
889   if (!result.IsNull()) return result.GetFile();
890   if (underlay_ != NULL) {
891     const FileDescriptor* result =
892       underlay_->FindFileContainingSymbol(symbol_name);
893     if (result != NULL) return result;
894   }
895   if (TryFindSymbolInFallbackDatabase(symbol_name)) {
896     Symbol result = tables_->FindSymbol(symbol_name);
897     if (!result.IsNull()) return result.GetFile();
898   }
899   return NULL;
900 }
901 
FindMessageTypeByName(const string & name) const902 const Descriptor* DescriptorPool::FindMessageTypeByName(
903     const string& name) const {
904   Symbol result = tables_->FindByNameHelper(this, name);
905   return (result.type == Symbol::MESSAGE) ? result.descriptor : NULL;
906 }
907 
FindFieldByName(const string & name) const908 const FieldDescriptor* DescriptorPool::FindFieldByName(
909     const string& name) const {
910   Symbol result = tables_->FindByNameHelper(this, name);
911   if (result.type == Symbol::FIELD &&
912       !result.field_descriptor->is_extension()) {
913     return result.field_descriptor;
914   } else {
915     return NULL;
916   }
917 }
918 
FindExtensionByName(const string & name) const919 const FieldDescriptor* DescriptorPool::FindExtensionByName(
920     const string& name) const {
921   Symbol result = tables_->FindByNameHelper(this, name);
922   if (result.type == Symbol::FIELD &&
923       result.field_descriptor->is_extension()) {
924     return result.field_descriptor;
925   } else {
926     return NULL;
927   }
928 }
929 
FindEnumTypeByName(const string & name) const930 const EnumDescriptor* DescriptorPool::FindEnumTypeByName(
931     const string& name) const {
932   Symbol result = tables_->FindByNameHelper(this, name);
933   return (result.type == Symbol::ENUM) ? result.enum_descriptor : NULL;
934 }
935 
FindEnumValueByName(const string & name) const936 const EnumValueDescriptor* DescriptorPool::FindEnumValueByName(
937     const string& name) const {
938   Symbol result = tables_->FindByNameHelper(this, name);
939   return (result.type == Symbol::ENUM_VALUE) ?
940     result.enum_value_descriptor : NULL;
941 }
942 
FindServiceByName(const string & name) const943 const ServiceDescriptor* DescriptorPool::FindServiceByName(
944     const string& name) const {
945   Symbol result = tables_->FindByNameHelper(this, name);
946   return (result.type == Symbol::SERVICE) ? result.service_descriptor : NULL;
947 }
948 
FindMethodByName(const string & name) const949 const MethodDescriptor* DescriptorPool::FindMethodByName(
950     const string& name) const {
951   Symbol result = tables_->FindByNameHelper(this, name);
952   return (result.type == Symbol::METHOD) ? result.method_descriptor : NULL;
953 }
954 
FindExtensionByNumber(const Descriptor * extendee,int number) const955 const FieldDescriptor* DescriptorPool::FindExtensionByNumber(
956     const Descriptor* extendee, int number) const {
957   MutexLockMaybe lock(mutex_);
958   const FieldDescriptor* result = tables_->FindExtension(extendee, number);
959   if (result != NULL) {
960     return result;
961   }
962   if (underlay_ != NULL) {
963     const FieldDescriptor* result =
964       underlay_->FindExtensionByNumber(extendee, number);
965     if (result != NULL) return result;
966   }
967   if (TryFindExtensionInFallbackDatabase(extendee, number)) {
968     const FieldDescriptor* result = tables_->FindExtension(extendee, number);
969     if (result != NULL) {
970       return result;
971     }
972   }
973   return NULL;
974 }
975 
FindAllExtensions(const Descriptor * extendee,vector<const FieldDescriptor * > * out) const976 void DescriptorPool::FindAllExtensions(
977     const Descriptor* extendee, vector<const FieldDescriptor*>* out) const {
978   MutexLockMaybe lock(mutex_);
979 
980   // Initialize tables_->extensions_ from the fallback database first
981   // (but do this only once per descriptor).
982   if (fallback_database_ != NULL &&
983       tables_->extensions_loaded_from_db_.count(extendee) == 0) {
984     vector<int> numbers;
985     if (fallback_database_->FindAllExtensionNumbers(extendee->full_name(),
986                                                     &numbers)) {
987       for (int i = 0; i < numbers.size(); ++i) {
988         int number = numbers[i];
989         if (tables_->FindExtension(extendee, number) == NULL) {
990           TryFindExtensionInFallbackDatabase(extendee, number);
991         }
992       }
993       tables_->extensions_loaded_from_db_.insert(extendee);
994     }
995   }
996 
997   tables_->FindAllExtensions(extendee, out);
998   if (underlay_ != NULL) {
999     underlay_->FindAllExtensions(extendee, out);
1000   }
1001 }
1002 
1003 // -------------------------------------------------------------------
1004 
1005 const FieldDescriptor*
FindFieldByNumber(int key) const1006 Descriptor::FindFieldByNumber(int key) const {
1007   const FieldDescriptor* result =
1008     file()->tables_->FindFieldByNumber(this, key);
1009   if (result == NULL || result->is_extension()) {
1010     return NULL;
1011   } else {
1012     return result;
1013   }
1014 }
1015 
1016 const FieldDescriptor*
FindFieldByLowercaseName(const string & key) const1017 Descriptor::FindFieldByLowercaseName(const string& key) const {
1018   const FieldDescriptor* result =
1019     file()->tables_->FindFieldByLowercaseName(this, key);
1020   if (result == NULL || result->is_extension()) {
1021     return NULL;
1022   } else {
1023     return result;
1024   }
1025 }
1026 
1027 const FieldDescriptor*
FindFieldByCamelcaseName(const string & key) const1028 Descriptor::FindFieldByCamelcaseName(const string& key) const {
1029   const FieldDescriptor* result =
1030     file()->tables_->FindFieldByCamelcaseName(this, key);
1031   if (result == NULL || result->is_extension()) {
1032     return NULL;
1033   } else {
1034     return result;
1035   }
1036 }
1037 
1038 const FieldDescriptor*
FindFieldByName(const string & key) const1039 Descriptor::FindFieldByName(const string& key) const {
1040   Symbol result =
1041     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1042   if (!result.IsNull() && !result.field_descriptor->is_extension()) {
1043     return result.field_descriptor;
1044   } else {
1045     return NULL;
1046   }
1047 }
1048 
1049 const FieldDescriptor*
FindExtensionByName(const string & key) const1050 Descriptor::FindExtensionByName(const string& key) const {
1051   Symbol result =
1052     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1053   if (!result.IsNull() && result.field_descriptor->is_extension()) {
1054     return result.field_descriptor;
1055   } else {
1056     return NULL;
1057   }
1058 }
1059 
1060 const FieldDescriptor*
FindExtensionByLowercaseName(const string & key) const1061 Descriptor::FindExtensionByLowercaseName(const string& key) const {
1062   const FieldDescriptor* result =
1063     file()->tables_->FindFieldByLowercaseName(this, key);
1064   if (result == NULL || !result->is_extension()) {
1065     return NULL;
1066   } else {
1067     return result;
1068   }
1069 }
1070 
1071 const FieldDescriptor*
FindExtensionByCamelcaseName(const string & key) const1072 Descriptor::FindExtensionByCamelcaseName(const string& key) const {
1073   const FieldDescriptor* result =
1074     file()->tables_->FindFieldByCamelcaseName(this, key);
1075   if (result == NULL || !result->is_extension()) {
1076     return NULL;
1077   } else {
1078     return result;
1079   }
1080 }
1081 
1082 const Descriptor*
FindNestedTypeByName(const string & key) const1083 Descriptor::FindNestedTypeByName(const string& key) const {
1084   Symbol result =
1085     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::MESSAGE);
1086   if (!result.IsNull()) {
1087     return result.descriptor;
1088   } else {
1089     return NULL;
1090   }
1091 }
1092 
1093 const EnumDescriptor*
FindEnumTypeByName(const string & key) const1094 Descriptor::FindEnumTypeByName(const string& key) const {
1095   Symbol result =
1096     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM);
1097   if (!result.IsNull()) {
1098     return result.enum_descriptor;
1099   } else {
1100     return NULL;
1101   }
1102 }
1103 
1104 const EnumValueDescriptor*
FindEnumValueByName(const string & key) const1105 Descriptor::FindEnumValueByName(const string& key) const {
1106   Symbol result =
1107     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1108   if (!result.IsNull()) {
1109     return result.enum_value_descriptor;
1110   } else {
1111     return NULL;
1112   }
1113 }
1114 
1115 const EnumValueDescriptor*
FindValueByName(const string & key) const1116 EnumDescriptor::FindValueByName(const string& key) const {
1117   Symbol result =
1118     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1119   if (!result.IsNull()) {
1120     return result.enum_value_descriptor;
1121   } else {
1122     return NULL;
1123   }
1124 }
1125 
1126 const EnumValueDescriptor*
FindValueByNumber(int key) const1127 EnumDescriptor::FindValueByNumber(int key) const {
1128   return file()->tables_->FindEnumValueByNumber(this, key);
1129 }
1130 
1131 const MethodDescriptor*
FindMethodByName(const string & key) const1132 ServiceDescriptor::FindMethodByName(const string& key) const {
1133   Symbol result =
1134     file()->tables_->FindNestedSymbolOfType(this, key, Symbol::METHOD);
1135   if (!result.IsNull()) {
1136     return result.method_descriptor;
1137   } else {
1138     return NULL;
1139   }
1140 }
1141 
1142 const Descriptor*
FindMessageTypeByName(const string & key) const1143 FileDescriptor::FindMessageTypeByName(const string& key) const {
1144   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::MESSAGE);
1145   if (!result.IsNull()) {
1146     return result.descriptor;
1147   } else {
1148     return NULL;
1149   }
1150 }
1151 
1152 const EnumDescriptor*
FindEnumTypeByName(const string & key) const1153 FileDescriptor::FindEnumTypeByName(const string& key) const {
1154   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM);
1155   if (!result.IsNull()) {
1156     return result.enum_descriptor;
1157   } else {
1158     return NULL;
1159   }
1160 }
1161 
1162 const EnumValueDescriptor*
FindEnumValueByName(const string & key) const1163 FileDescriptor::FindEnumValueByName(const string& key) const {
1164   Symbol result =
1165     tables_->FindNestedSymbolOfType(this, key, Symbol::ENUM_VALUE);
1166   if (!result.IsNull()) {
1167     return result.enum_value_descriptor;
1168   } else {
1169     return NULL;
1170   }
1171 }
1172 
1173 const ServiceDescriptor*
FindServiceByName(const string & key) const1174 FileDescriptor::FindServiceByName(const string& key) const {
1175   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::SERVICE);
1176   if (!result.IsNull()) {
1177     return result.service_descriptor;
1178   } else {
1179     return NULL;
1180   }
1181 }
1182 
1183 const FieldDescriptor*
FindExtensionByName(const string & key) const1184 FileDescriptor::FindExtensionByName(const string& key) const {
1185   Symbol result = tables_->FindNestedSymbolOfType(this, key, Symbol::FIELD);
1186   if (!result.IsNull() && result.field_descriptor->is_extension()) {
1187     return result.field_descriptor;
1188   } else {
1189     return NULL;
1190   }
1191 }
1192 
1193 const FieldDescriptor*
FindExtensionByLowercaseName(const string & key) const1194 FileDescriptor::FindExtensionByLowercaseName(const string& key) const {
1195   const FieldDescriptor* result = tables_->FindFieldByLowercaseName(this, key);
1196   if (result == NULL || !result->is_extension()) {
1197     return NULL;
1198   } else {
1199     return result;
1200   }
1201 }
1202 
1203 const FieldDescriptor*
FindExtensionByCamelcaseName(const string & key) const1204 FileDescriptor::FindExtensionByCamelcaseName(const string& key) const {
1205   const FieldDescriptor* result = tables_->FindFieldByCamelcaseName(this, key);
1206   if (result == NULL || !result->is_extension()) {
1207     return NULL;
1208   } else {
1209     return result;
1210   }
1211 }
1212 
IsExtensionNumber(int number) const1213 bool Descriptor::IsExtensionNumber(int number) const {
1214   // Linear search should be fine because we don't expect a message to have
1215   // more than a couple extension ranges.
1216   for (int i = 0; i < extension_range_count(); i++) {
1217     if (number >= extension_range(i)->start &&
1218         number <  extension_range(i)->end) {
1219       return true;
1220     }
1221   }
1222   return false;
1223 }
1224 
1225 // -------------------------------------------------------------------
1226 
TryFindFileInFallbackDatabase(const string & name) const1227 bool DescriptorPool::TryFindFileInFallbackDatabase(const string& name) const {
1228   if (fallback_database_ == NULL) return false;
1229 
1230   if (tables_->known_bad_files_.count(name) > 0) return false;
1231 
1232   FileDescriptorProto file_proto;
1233   if (!fallback_database_->FindFileByName(name, &file_proto) ||
1234       BuildFileFromDatabase(file_proto) == NULL) {
1235     tables_->known_bad_files_.insert(name);
1236     return false;
1237   }
1238 
1239   return true;
1240 }
1241 
TryFindSymbolInFallbackDatabase(const string & name) const1242 bool DescriptorPool::TryFindSymbolInFallbackDatabase(const string& name) const {
1243   if (fallback_database_ == NULL) return false;
1244 
1245   FileDescriptorProto file_proto;
1246   if (!fallback_database_->FindFileContainingSymbol(name, &file_proto)) {
1247     return false;
1248   }
1249 
1250   if (tables_->FindFile(file_proto.name()) != NULL) {
1251     // We've already loaded this file, and it apparently doesn't contain the
1252     // symbol we're looking for.  Some DescriptorDatabases return false
1253     // positives.
1254     return false;
1255   }
1256 
1257   if (BuildFileFromDatabase(file_proto) == NULL) {
1258     return false;
1259   }
1260 
1261   return true;
1262 }
1263 
TryFindExtensionInFallbackDatabase(const Descriptor * containing_type,int field_number) const1264 bool DescriptorPool::TryFindExtensionInFallbackDatabase(
1265     const Descriptor* containing_type, int field_number) const {
1266   if (fallback_database_ == NULL) return false;
1267 
1268   FileDescriptorProto file_proto;
1269   if (!fallback_database_->FindFileContainingExtension(
1270         containing_type->full_name(), field_number, &file_proto)) {
1271     return false;
1272   }
1273 
1274   if (tables_->FindFile(file_proto.name()) != NULL) {
1275     // We've already loaded this file, and it apparently doesn't contain the
1276     // extension we're looking for.  Some DescriptorDatabases return false
1277     // positives.
1278     return false;
1279   }
1280 
1281   if (BuildFileFromDatabase(file_proto) == NULL) {
1282     return false;
1283   }
1284 
1285   return true;
1286 }
1287 
1288 // ===================================================================
1289 
DefaultValueAsString(bool quote_string_type) const1290 string FieldDescriptor::DefaultValueAsString(bool quote_string_type) const {
1291   GOOGLE_CHECK(has_default_value()) << "No default value";
1292   switch (cpp_type()) {
1293     case CPPTYPE_INT32:
1294       return SimpleItoa(default_value_int32());
1295       break;
1296     case CPPTYPE_INT64:
1297       return SimpleItoa(default_value_int64());
1298       break;
1299     case CPPTYPE_UINT32:
1300       return SimpleItoa(default_value_uint32());
1301       break;
1302     case CPPTYPE_UINT64:
1303       return SimpleItoa(default_value_uint64());
1304       break;
1305     case CPPTYPE_FLOAT:
1306       return SimpleFtoa(default_value_float());
1307       break;
1308     case CPPTYPE_DOUBLE:
1309       return SimpleDtoa(default_value_double());
1310       break;
1311     case CPPTYPE_BOOL:
1312       return default_value_bool() ? "true" : "false";
1313       break;
1314     case CPPTYPE_STRING:
1315       if (quote_string_type) {
1316         return "\"" + CEscape(default_value_string()) + "\"";
1317       } else {
1318         if (type() == TYPE_BYTES) {
1319           return CEscape(default_value_string());
1320         } else {
1321           return default_value_string();
1322         }
1323       }
1324       break;
1325     case CPPTYPE_ENUM:
1326       return default_value_enum()->name();
1327       break;
1328     case CPPTYPE_MESSAGE:
1329       GOOGLE_LOG(DFATAL) << "Messages can't have default values!";
1330       break;
1331   }
1332   GOOGLE_LOG(FATAL) << "Can't get here: failed to get default value as string";
1333   return "";
1334 }
1335 
1336 // CopyTo methods ====================================================
1337 
CopyTo(FileDescriptorProto * proto) const1338 void FileDescriptor::CopyTo(FileDescriptorProto* proto) const {
1339   proto->set_name(name());
1340   if (!package().empty()) proto->set_package(package());
1341 
1342   for (int i = 0; i < dependency_count(); i++) {
1343     proto->add_dependency(dependency(i)->name());
1344   }
1345 
1346   for (int i = 0; i < message_type_count(); i++) {
1347     message_type(i)->CopyTo(proto->add_message_type());
1348   }
1349   for (int i = 0; i < enum_type_count(); i++) {
1350     enum_type(i)->CopyTo(proto->add_enum_type());
1351   }
1352   for (int i = 0; i < service_count(); i++) {
1353     service(i)->CopyTo(proto->add_service());
1354   }
1355   for (int i = 0; i < extension_count(); i++) {
1356     extension(i)->CopyTo(proto->add_extension());
1357   }
1358 
1359   if (&options() != &FileOptions::default_instance()) {
1360     proto->mutable_options()->CopyFrom(options());
1361   }
1362 }
1363 
CopyTo(DescriptorProto * proto) const1364 void Descriptor::CopyTo(DescriptorProto* proto) const {
1365   proto->set_name(name());
1366 
1367   for (int i = 0; i < field_count(); i++) {
1368     field(i)->CopyTo(proto->add_field());
1369   }
1370   for (int i = 0; i < nested_type_count(); i++) {
1371     nested_type(i)->CopyTo(proto->add_nested_type());
1372   }
1373   for (int i = 0; i < enum_type_count(); i++) {
1374     enum_type(i)->CopyTo(proto->add_enum_type());
1375   }
1376   for (int i = 0; i < extension_range_count(); i++) {
1377     DescriptorProto::ExtensionRange* range = proto->add_extension_range();
1378     range->set_start(extension_range(i)->start);
1379     range->set_end(extension_range(i)->end);
1380   }
1381   for (int i = 0; i < extension_count(); i++) {
1382     extension(i)->CopyTo(proto->add_extension());
1383   }
1384 
1385   if (&options() != &MessageOptions::default_instance()) {
1386     proto->mutable_options()->CopyFrom(options());
1387   }
1388 }
1389 
CopyTo(FieldDescriptorProto * proto) const1390 void FieldDescriptor::CopyTo(FieldDescriptorProto* proto) const {
1391   proto->set_name(name());
1392   proto->set_number(number());
1393 
1394   // Some compilers do not allow static_cast directly between two enum types,
1395   // so we must cast to int first.
1396   proto->set_label(static_cast<FieldDescriptorProto::Label>(
1397                      implicit_cast<int>(label())));
1398   proto->set_type(static_cast<FieldDescriptorProto::Type>(
1399                     implicit_cast<int>(type())));
1400 
1401   if (is_extension()) {
1402     if (!containing_type()->is_unqualified_placeholder_) {
1403       proto->set_extendee(".");
1404     }
1405     proto->mutable_extendee()->append(containing_type()->full_name());
1406   }
1407 
1408   if (cpp_type() == CPPTYPE_MESSAGE) {
1409     if (message_type()->is_placeholder_) {
1410       // We don't actually know if the type is a message type.  It could be
1411       // an enum.
1412       proto->clear_type();
1413     }
1414 
1415     if (!message_type()->is_unqualified_placeholder_) {
1416       proto->set_type_name(".");
1417     }
1418     proto->mutable_type_name()->append(message_type()->full_name());
1419   } else if (cpp_type() == CPPTYPE_ENUM) {
1420     if (!enum_type()->is_unqualified_placeholder_) {
1421       proto->set_type_name(".");
1422     }
1423     proto->mutable_type_name()->append(enum_type()->full_name());
1424   }
1425 
1426   if (has_default_value()) {
1427     proto->set_default_value(DefaultValueAsString(false));
1428   }
1429 
1430   if (&options() != &FieldOptions::default_instance()) {
1431     proto->mutable_options()->CopyFrom(options());
1432   }
1433 }
1434 
CopyTo(EnumDescriptorProto * proto) const1435 void EnumDescriptor::CopyTo(EnumDescriptorProto* proto) const {
1436   proto->set_name(name());
1437 
1438   for (int i = 0; i < value_count(); i++) {
1439     value(i)->CopyTo(proto->add_value());
1440   }
1441 
1442   if (&options() != &EnumOptions::default_instance()) {
1443     proto->mutable_options()->CopyFrom(options());
1444   }
1445 }
1446 
CopyTo(EnumValueDescriptorProto * proto) const1447 void EnumValueDescriptor::CopyTo(EnumValueDescriptorProto* proto) const {
1448   proto->set_name(name());
1449   proto->set_number(number());
1450 
1451   if (&options() != &EnumValueOptions::default_instance()) {
1452     proto->mutable_options()->CopyFrom(options());
1453   }
1454 }
1455 
CopyTo(ServiceDescriptorProto * proto) const1456 void ServiceDescriptor::CopyTo(ServiceDescriptorProto* proto) const {
1457   proto->set_name(name());
1458 
1459   for (int i = 0; i < method_count(); i++) {
1460     method(i)->CopyTo(proto->add_method());
1461   }
1462 
1463   if (&options() != &ServiceOptions::default_instance()) {
1464     proto->mutable_options()->CopyFrom(options());
1465   }
1466 }
1467 
CopyTo(MethodDescriptorProto * proto) const1468 void MethodDescriptor::CopyTo(MethodDescriptorProto* proto) const {
1469   proto->set_name(name());
1470 
1471   if (!input_type()->is_unqualified_placeholder_) {
1472     proto->set_input_type(".");
1473   }
1474   proto->mutable_input_type()->append(input_type()->full_name());
1475 
1476   if (!output_type()->is_unqualified_placeholder_) {
1477     proto->set_output_type(".");
1478   }
1479   proto->mutable_output_type()->append(output_type()->full_name());
1480 
1481   if (&options() != &MethodOptions::default_instance()) {
1482     proto->mutable_options()->CopyFrom(options());
1483   }
1484 }
1485 
1486 // DebugString methods ===============================================
1487 
1488 namespace {
1489 
1490 // Used by each of the option formatters.
RetrieveOptions(const Message & options,vector<string> * option_entries)1491 bool RetrieveOptions(const Message &options, vector<string> *option_entries) {
1492   option_entries->clear();
1493   const Reflection* reflection = options.GetReflection();
1494   vector<const FieldDescriptor*> fields;
1495   reflection->ListFields(options, &fields);
1496   for (int i = 0; i < fields.size(); i++) {
1497     // Doesn't make sense to have message type fields here
1498     if (fields[i]->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
1499       continue;
1500     }
1501     int count = 1;
1502     bool repeated = false;
1503     if (fields[i]->is_repeated()) {
1504       count = reflection->FieldSize(options, fields[i]);
1505       repeated = true;
1506     }
1507     for (int j = 0; j < count; j++) {
1508       string fieldval;
1509       TextFormat::PrintFieldValueToString(options, fields[i],
1510                                           repeated ? count : -1, &fieldval);
1511       option_entries->push_back(fields[i]->name() + " = " + fieldval);
1512     }
1513   }
1514   return !option_entries->empty();
1515 }
1516 
1517 // Formats options that all appear together in brackets. Does not include
1518 // brackets.
FormatBracketedOptions(const Message & options,string * output)1519 bool FormatBracketedOptions(const Message &options, string *output) {
1520   vector<string> all_options;
1521   if (RetrieveOptions(options, &all_options)) {
1522     output->append(JoinStrings(all_options, ", "));
1523   }
1524   return !all_options.empty();
1525 }
1526 
1527 // Formats options one per line
FormatLineOptions(int depth,const Message & options,string * output)1528 bool FormatLineOptions(int depth, const Message &options, string *output) {
1529   string prefix(depth * 2, ' ');
1530   vector<string> all_options;
1531   if (RetrieveOptions(options, &all_options)) {
1532     for (int i = 0; i < all_options.size(); i++) {
1533       strings::SubstituteAndAppend(output, "$0option $1;\n",
1534                                    prefix, all_options[i]);
1535     }
1536   }
1537   return !all_options.empty();
1538 }
1539 
1540 }  // anonymous namespace
1541 
DebugString() const1542 string FileDescriptor::DebugString() const {
1543   string contents = "syntax = \"proto2\";\n\n";
1544 
1545   for (int i = 0; i < dependency_count(); i++) {
1546     strings::SubstituteAndAppend(&contents, "import \"$0\";\n",
1547                                  dependency(i)->name());
1548   }
1549 
1550   if (!package().empty()) {
1551     strings::SubstituteAndAppend(&contents, "package $0;\n\n", package());
1552   }
1553 
1554   if (FormatLineOptions(0, options(), &contents)) {
1555     contents.append("\n");  // add some space if we had options
1556   }
1557 
1558   for (int i = 0; i < enum_type_count(); i++) {
1559     enum_type(i)->DebugString(0, &contents);
1560     contents.append("\n");
1561   }
1562 
1563   // Find all the 'group' type extensions; we will not output their nested
1564   // definitions (those will be done with their group field descriptor).
1565   set<const Descriptor*> groups;
1566   for (int i = 0; i < extension_count(); i++) {
1567     if (extension(i)->type() == FieldDescriptor::TYPE_GROUP) {
1568       groups.insert(extension(i)->message_type());
1569     }
1570   }
1571 
1572   for (int i = 0; i < message_type_count(); i++) {
1573     if (groups.count(message_type(i)) == 0) {
1574       strings::SubstituteAndAppend(&contents, "message $0",
1575                                    message_type(i)->name());
1576       message_type(i)->DebugString(0, &contents);
1577       contents.append("\n");
1578     }
1579   }
1580 
1581   for (int i = 0; i < service_count(); i++) {
1582     service(i)->DebugString(&contents);
1583     contents.append("\n");
1584   }
1585 
1586   const Descriptor* containing_type = NULL;
1587   for (int i = 0; i < extension_count(); i++) {
1588     if (extension(i)->containing_type() != containing_type) {
1589       if (i > 0) contents.append("}\n\n");
1590       containing_type = extension(i)->containing_type();
1591       strings::SubstituteAndAppend(&contents, "extend .$0 {\n",
1592                                    containing_type->full_name());
1593     }
1594     extension(i)->DebugString(1, &contents);
1595   }
1596   if (extension_count() > 0) contents.append("}\n\n");
1597 
1598   return contents;
1599 }
1600 
DebugString() const1601 string Descriptor::DebugString() const {
1602   string contents;
1603   strings::SubstituteAndAppend(&contents, "message $0", name());
1604   DebugString(0, &contents);
1605   return contents;
1606 }
1607 
DebugString(int depth,string * contents) const1608 void Descriptor::DebugString(int depth, string *contents) const {
1609   string prefix(depth * 2, ' ');
1610   ++depth;
1611   contents->append(" {\n");
1612 
1613   FormatLineOptions(depth, options(), contents);
1614 
1615   // Find all the 'group' types for fields and extensions; we will not output
1616   // their nested definitions (those will be done with their group field
1617   // descriptor).
1618   set<const Descriptor*> groups;
1619   for (int i = 0; i < field_count(); i++) {
1620     if (field(i)->type() == FieldDescriptor::TYPE_GROUP) {
1621       groups.insert(field(i)->message_type());
1622     }
1623   }
1624   for (int i = 0; i < extension_count(); i++) {
1625     if (extension(i)->type() == FieldDescriptor::TYPE_GROUP) {
1626       groups.insert(extension(i)->message_type());
1627     }
1628   }
1629 
1630   for (int i = 0; i < nested_type_count(); i++) {
1631     if (groups.count(nested_type(i)) == 0) {
1632       strings::SubstituteAndAppend(contents, "$0  message $1",
1633                                    prefix, nested_type(i)->name());
1634       nested_type(i)->DebugString(depth, contents);
1635     }
1636   }
1637   for (int i = 0; i < enum_type_count(); i++) {
1638     enum_type(i)->DebugString(depth, contents);
1639   }
1640   for (int i = 0; i < field_count(); i++) {
1641     field(i)->DebugString(depth, contents);
1642   }
1643 
1644   for (int i = 0; i < extension_range_count(); i++) {
1645     strings::SubstituteAndAppend(contents, "$0  extensions $1 to $2;\n",
1646                                  prefix,
1647                                  extension_range(i)->start,
1648                                  extension_range(i)->end - 1);
1649   }
1650 
1651   // Group extensions by what they extend, so they can be printed out together.
1652   const Descriptor* containing_type = NULL;
1653   for (int i = 0; i < extension_count(); i++) {
1654     if (extension(i)->containing_type() != containing_type) {
1655       if (i > 0) strings::SubstituteAndAppend(contents, "$0  }\n", prefix);
1656       containing_type = extension(i)->containing_type();
1657       strings::SubstituteAndAppend(contents, "$0  extend .$1 {\n",
1658                                    prefix, containing_type->full_name());
1659     }
1660     extension(i)->DebugString(depth + 1, contents);
1661   }
1662   if (extension_count() > 0)
1663     strings::SubstituteAndAppend(contents, "$0  }\n", prefix);
1664 
1665   strings::SubstituteAndAppend(contents, "$0}\n", prefix);
1666 }
1667 
DebugString() const1668 string FieldDescriptor::DebugString() const {
1669   string contents;
1670   int depth = 0;
1671   if (is_extension()) {
1672     strings::SubstituteAndAppend(&contents, "extend .$0 {\n",
1673                                  containing_type()->full_name());
1674     depth = 1;
1675   }
1676   DebugString(depth, &contents);
1677   if (is_extension()) {
1678      contents.append("}\n");
1679   }
1680   return contents;
1681 }
1682 
DebugString(int depth,string * contents) const1683 void FieldDescriptor::DebugString(int depth, string *contents) const {
1684   string prefix(depth * 2, ' ');
1685   string field_type;
1686   switch (type()) {
1687     case TYPE_MESSAGE:
1688       field_type = "." + message_type()->full_name();
1689       break;
1690     case TYPE_ENUM:
1691       field_type = "." + enum_type()->full_name();
1692       break;
1693     default:
1694       field_type = kTypeToName[type()];
1695   }
1696 
1697   strings::SubstituteAndAppend(contents, "$0$1 $2 $3 = $4",
1698                                prefix,
1699                                kLabelToName[label()],
1700                                field_type,
1701                                type() == TYPE_GROUP ? message_type()->name() :
1702                                                       name(),
1703                                number());
1704 
1705   bool bracketed = false;
1706   if (has_default_value()) {
1707     bracketed = true;
1708     strings::SubstituteAndAppend(contents, " [default = $0",
1709                                  DefaultValueAsString(true));
1710   }
1711 
1712   string formatted_options;
1713   if (FormatBracketedOptions(options(), &formatted_options)) {
1714     contents->append(bracketed ? ", " : " [");
1715     bracketed = true;
1716     contents->append(formatted_options);
1717   }
1718 
1719   if (bracketed) {
1720     contents->append("]");
1721   }
1722 
1723   if (type() == TYPE_GROUP) {
1724     message_type()->DebugString(depth, contents);
1725   } else {
1726     contents->append(";\n");
1727   }
1728 }
1729 
DebugString() const1730 string EnumDescriptor::DebugString() const {
1731   string contents;
1732   DebugString(0, &contents);
1733   return contents;
1734 }
1735 
DebugString(int depth,string * contents) const1736 void EnumDescriptor::DebugString(int depth, string *contents) const {
1737   string prefix(depth * 2, ' ');
1738   ++depth;
1739   strings::SubstituteAndAppend(contents, "$0enum $1 {\n",
1740                                prefix, name());
1741 
1742   FormatLineOptions(depth, options(), contents);
1743 
1744   for (int i = 0; i < value_count(); i++) {
1745     value(i)->DebugString(depth, contents);
1746   }
1747   strings::SubstituteAndAppend(contents, "$0}\n", prefix);
1748 }
1749 
DebugString() const1750 string EnumValueDescriptor::DebugString() const {
1751   string contents;
1752   DebugString(0, &contents);
1753   return contents;
1754 }
1755 
DebugString(int depth,string * contents) const1756 void EnumValueDescriptor::DebugString(int depth, string *contents) const {
1757   string prefix(depth * 2, ' ');
1758   strings::SubstituteAndAppend(contents, "$0$1 = $2",
1759                                prefix, name(), number());
1760 
1761   string formatted_options;
1762   if (FormatBracketedOptions(options(), &formatted_options)) {
1763     strings::SubstituteAndAppend(contents, " [$0]", formatted_options);
1764   }
1765   contents->append(";\n");
1766 }
1767 
DebugString() const1768 string ServiceDescriptor::DebugString() const {
1769   string contents;
1770   DebugString(&contents);
1771   return contents;
1772 }
1773 
DebugString(string * contents) const1774 void ServiceDescriptor::DebugString(string *contents) const {
1775   strings::SubstituteAndAppend(contents, "service $0 {\n", name());
1776 
1777   FormatLineOptions(1, options(), contents);
1778 
1779   for (int i = 0; i < method_count(); i++) {
1780     method(i)->DebugString(1, contents);
1781   }
1782 
1783   contents->append("}\n");
1784 }
1785 
DebugString() const1786 string MethodDescriptor::DebugString() const {
1787   string contents;
1788   DebugString(0, &contents);
1789   return contents;
1790 }
1791 
DebugString(int depth,string * contents) const1792 void MethodDescriptor::DebugString(int depth, string *contents) const {
1793   string prefix(depth * 2, ' ');
1794   ++depth;
1795   strings::SubstituteAndAppend(contents, "$0rpc $1(.$2) returns (.$3)",
1796                                prefix, name(),
1797                                input_type()->full_name(),
1798                                output_type()->full_name());
1799 
1800   string formatted_options;
1801   if (FormatLineOptions(depth, options(), &formatted_options)) {
1802     strings::SubstituteAndAppend(contents, " {\n$0$1}\n",
1803                                  formatted_options, prefix);
1804   } else {
1805     contents->append(";\n");
1806   }
1807 }
1808 // ===================================================================
1809 
1810 namespace {
1811 
1812 // Represents an options message to interpret. Extension names in the option
1813 // name are respolved relative to name_scope. element_name and orig_opt are
1814 // used only for error reporting (since the parser records locations against
1815 // pointers in the original options, not the mutable copy). The Message must be
1816 // one of the Options messages in descriptor.proto.
1817 struct OptionsToInterpret {
OptionsToInterpretgoogle::protobuf::__anona61a9b310511::OptionsToInterpret1818   OptionsToInterpret(const string& ns,
1819                      const string& el,
1820                      const Message* orig_opt,
1821                      Message* opt)
1822       : name_scope(ns),
1823         element_name(el),
1824         original_options(orig_opt),
1825         options(opt) {
1826   }
1827   string name_scope;
1828   string element_name;
1829   const Message* original_options;
1830   Message* options;
1831 };
1832 
1833 }  // namespace
1834 
1835 class DescriptorBuilder {
1836  public:
1837   DescriptorBuilder(const DescriptorPool* pool,
1838                     DescriptorPool::Tables* tables,
1839                     DescriptorPool::ErrorCollector* error_collector);
1840   ~DescriptorBuilder();
1841 
1842   const FileDescriptor* BuildFile(const FileDescriptorProto& proto);
1843 
1844  private:
1845   friend class OptionInterpreter;
1846 
1847   const DescriptorPool* pool_;
1848   DescriptorPool::Tables* tables_;  // for convenience
1849   DescriptorPool::ErrorCollector* error_collector_;
1850 
1851   // As we build descriptors we store copies of the options messages in
1852   // them. We put pointers to those copies in this vector, as we build, so we
1853   // can later (after cross-linking) interpret those options.
1854   vector<OptionsToInterpret> options_to_interpret_;
1855 
1856   bool had_errors_;
1857   string filename_;
1858   FileDescriptor* file_;
1859   FileDescriptorTables* file_tables_;
1860 
1861   // If LookupSymbol() finds a symbol that is in a file which is not a declared
1862   // dependency of this file, it will fail, but will set
1863   // possible_undeclared_dependency_ to point at that file.  This is only used
1864   // by AddNotDefinedError() to report a more useful error message.
1865   // possible_undeclared_dependency_name_ is the name of the symbol that was
1866   // actually found in possible_undeclared_dependency_, which may be a parent
1867   // of the symbol actually looked for.
1868   const FileDescriptor* possible_undeclared_dependency_;
1869   string possible_undeclared_dependency_name_;
1870 
1871   void AddError(const string& element_name,
1872                 const Message& descriptor,
1873                 DescriptorPool::ErrorCollector::ErrorLocation location,
1874                 const string& error);
1875 
1876   // Adds an error indicating that undefined_symbol was not defined.  Must
1877   // only be called after LookupSymbol() fails.
1878   void AddNotDefinedError(
1879     const string& element_name,
1880     const Message& descriptor,
1881     DescriptorPool::ErrorCollector::ErrorLocation location,
1882     const string& undefined_symbol);
1883 
1884   // Silly helper which determines if the given file is in the given package.
1885   // I.e., either file->package() == package_name or file->package() is a
1886   // nested package within package_name.
1887   bool IsInPackage(const FileDescriptor* file, const string& package_name);
1888 
1889   // Like tables_->FindSymbol(), but additionally:
1890   // - Search the pool's underlay if not found in tables_.
1891   // - Insure that the resulting Symbol is from one of the file's declared
1892   //   dependencies.
1893   Symbol FindSymbol(const string& name);
1894 
1895   // Like FindSymbol() but does not require that the symbol is in one of the
1896   // file's declared dependencies.
1897   Symbol FindSymbolNotEnforcingDeps(const string& name);
1898 
1899   // Like FindSymbol(), but looks up the name relative to some other symbol
1900   // name.  This first searches siblings of relative_to, then siblings of its
1901   // parents, etc.  For example, LookupSymbol("foo.bar", "baz.qux.corge") makes
1902   // the following calls, returning the first non-null result:
1903   // FindSymbol("baz.qux.foo.bar"), FindSymbol("baz.foo.bar"),
1904   // FindSymbol("foo.bar").  If AllowUnknownDependencies() has been called
1905   // on the DescriptorPool, this will generate a placeholder type if
1906   // the name is not found (unless the name itself is malformed).  The
1907   // placeholder_type parameter indicates what kind of placeholder should be
1908   // constructed in this case.  The resolve_mode parameter determines whether
1909   // any symbol is returned, or only symbols that are types.  Note, however,
1910   // that LookupSymbol may still return a non-type symbol in LOOKUP_TYPES mode,
1911   // if it believes that's all it could refer to.  The caller should always
1912   // check that it receives the type of symbol it was expecting.
1913   enum PlaceholderType {
1914     PLACEHOLDER_MESSAGE,
1915     PLACEHOLDER_ENUM,
1916     PLACEHOLDER_EXTENDABLE_MESSAGE
1917   };
1918   enum ResolveMode {
1919     LOOKUP_ALL, LOOKUP_TYPES
1920   };
1921   Symbol LookupSymbol(const string& name, const string& relative_to,
1922                       PlaceholderType placeholder_type = PLACEHOLDER_MESSAGE,
1923                       ResolveMode resolve_mode = LOOKUP_ALL);
1924 
1925   // Like LookupSymbol() but will not return a placeholder even if
1926   // AllowUnknownDependencies() has been used.
1927   Symbol LookupSymbolNoPlaceholder(const string& name,
1928                                    const string& relative_to,
1929                                    ResolveMode resolve_mode = LOOKUP_ALL);
1930 
1931   // Creates a placeholder type suitable for return from LookupSymbol().  May
1932   // return kNullSymbol if the name is not a valid type name.
1933   Symbol NewPlaceholder(const string& name, PlaceholderType placeholder_type);
1934 
1935   // Creates a placeholder file.  Never returns NULL.  This is used when an
1936   // import is not found and AllowUnknownDependencies() is enabled.
1937   const FileDescriptor* NewPlaceholderFile(const string& name);
1938 
1939   // Calls tables_->AddSymbol() and records an error if it fails.  Returns
1940   // true if successful or false if failed, though most callers can ignore
1941   // the return value since an error has already been recorded.
1942   bool AddSymbol(const string& full_name,
1943                  const void* parent, const string& name,
1944                  const Message& proto, Symbol symbol);
1945 
1946   // Like AddSymbol(), but succeeds if the symbol is already defined as long
1947   // as the existing definition is also a package (because it's OK to define
1948   // the same package in two different files).  Also adds all parents of the
1949   // packgae to the symbol table (e.g. AddPackage("foo.bar", ...) will add
1950   // "foo.bar" and "foo" to the table).
1951   void AddPackage(const string& name, const Message& proto,
1952                   const FileDescriptor* file);
1953 
1954   // Checks that the symbol name contains only alphanumeric characters and
1955   // underscores.  Records an error otherwise.
1956   void ValidateSymbolName(const string& name, const string& full_name,
1957                           const Message& proto);
1958 
1959   // Like ValidateSymbolName(), but the name is allowed to contain periods and
1960   // an error is indicated by returning false (not recording the error).
1961   bool ValidateQualifiedName(const string& name);
1962 
1963   // Used by BUILD_ARRAY macro (below) to avoid having to have the type
1964   // specified as a macro parameter.
1965   template <typename Type>
AllocateArray(int size,Type ** output)1966   inline void AllocateArray(int size, Type** output) {
1967     *output = tables_->AllocateArray<Type>(size);
1968   }
1969 
1970   // Allocates a copy of orig_options in tables_ and stores it in the
1971   // descriptor. Remembers its uninterpreted options, to be interpreted
1972   // later. DescriptorT must be one of the Descriptor messages from
1973   // descriptor.proto.
1974   template<class DescriptorT> void AllocateOptions(
1975       const typename DescriptorT::OptionsType& orig_options,
1976       DescriptorT* descriptor);
1977   // Specialization for FileOptions.
1978   void AllocateOptions(const FileOptions& orig_options,
1979                        FileDescriptor* descriptor);
1980 
1981   // Implementation for AllocateOptions(). Don't call this directly.
1982   template<class DescriptorT> void AllocateOptionsImpl(
1983       const string& name_scope,
1984       const string& element_name,
1985       const typename DescriptorT::OptionsType& orig_options,
1986       DescriptorT* descriptor);
1987 
1988   // These methods all have the same signature for the sake of the BUILD_ARRAY
1989   // macro, below.
1990   void BuildMessage(const DescriptorProto& proto,
1991                     const Descriptor* parent,
1992                     Descriptor* result);
1993   void BuildFieldOrExtension(const FieldDescriptorProto& proto,
1994                              const Descriptor* parent,
1995                              FieldDescriptor* result,
1996                              bool is_extension);
BuildField(const FieldDescriptorProto & proto,const Descriptor * parent,FieldDescriptor * result)1997   void BuildField(const FieldDescriptorProto& proto,
1998                   const Descriptor* parent,
1999                   FieldDescriptor* result) {
2000     BuildFieldOrExtension(proto, parent, result, false);
2001   }
BuildExtension(const FieldDescriptorProto & proto,const Descriptor * parent,FieldDescriptor * result)2002   void BuildExtension(const FieldDescriptorProto& proto,
2003                       const Descriptor* parent,
2004                       FieldDescriptor* result) {
2005     BuildFieldOrExtension(proto, parent, result, true);
2006   }
2007   void BuildExtensionRange(const DescriptorProto::ExtensionRange& proto,
2008                            const Descriptor* parent,
2009                            Descriptor::ExtensionRange* result);
2010   void BuildEnum(const EnumDescriptorProto& proto,
2011                  const Descriptor* parent,
2012                  EnumDescriptor* result);
2013   void BuildEnumValue(const EnumValueDescriptorProto& proto,
2014                       const EnumDescriptor* parent,
2015                       EnumValueDescriptor* result);
2016   void BuildService(const ServiceDescriptorProto& proto,
2017                     const void* dummy,
2018                     ServiceDescriptor* result);
2019   void BuildMethod(const MethodDescriptorProto& proto,
2020                    const ServiceDescriptor* parent,
2021                    MethodDescriptor* result);
2022 
2023   // Must be run only after building.
2024   //
2025   // NOTE: Options will not be available during cross-linking, as they
2026   // have not yet been interpreted. Defer any handling of options to the
2027   // Validate*Options methods.
2028   void CrossLinkFile(FileDescriptor* file, const FileDescriptorProto& proto);
2029   void CrossLinkMessage(Descriptor* message, const DescriptorProto& proto);
2030   void CrossLinkField(FieldDescriptor* field,
2031                       const FieldDescriptorProto& proto);
2032   void CrossLinkEnum(EnumDescriptor* enum_type,
2033                      const EnumDescriptorProto& proto);
2034   void CrossLinkEnumValue(EnumValueDescriptor* enum_value,
2035                           const EnumValueDescriptorProto& proto);
2036   void CrossLinkService(ServiceDescriptor* service,
2037                         const ServiceDescriptorProto& proto);
2038   void CrossLinkMethod(MethodDescriptor* method,
2039                        const MethodDescriptorProto& proto);
2040 
2041   // Must be run only after cross-linking.
2042   void InterpretOptions();
2043 
2044   // A helper class for interpreting options.
2045   class OptionInterpreter {
2046    public:
2047     // Creates an interpreter that operates in the context of the pool of the
2048     // specified builder, which must not be NULL. We don't take ownership of the
2049     // builder.
2050     explicit OptionInterpreter(DescriptorBuilder* builder);
2051 
2052     ~OptionInterpreter();
2053 
2054     // Interprets the uninterpreted options in the specified Options message.
2055     // On error, calls AddError() on the underlying builder and returns false.
2056     // Otherwise returns true.
2057     bool InterpretOptions(OptionsToInterpret* options_to_interpret);
2058 
2059    private:
2060     // Interprets uninterpreted_option_ on the specified message, which
2061     // must be the mutable copy of the original options message to which
2062     // uninterpreted_option_ belongs.
2063     bool InterpretSingleOption(Message* options);
2064 
2065     // Adds the uninterpreted_option to the given options message verbatim.
2066     // Used when AllowUnknownDependencies() is in effect and we can't find
2067     // the option's definition.
2068     void AddWithoutInterpreting(const UninterpretedOption& uninterpreted_option,
2069                                 Message* options);
2070 
2071     // A recursive helper function that drills into the intermediate fields
2072     // in unknown_fields to check if field innermost_field is set on the
2073     // innermost message. Returns false and sets an error if so.
2074     bool ExamineIfOptionIsSet(
2075         vector<const FieldDescriptor*>::const_iterator intermediate_fields_iter,
2076         vector<const FieldDescriptor*>::const_iterator intermediate_fields_end,
2077         const FieldDescriptor* innermost_field, const string& debug_msg_name,
2078         const UnknownFieldSet& unknown_fields);
2079 
2080     // Validates the value for the option field of the currently interpreted
2081     // option and then sets it on the unknown_field.
2082     bool SetOptionValue(const FieldDescriptor* option_field,
2083                         UnknownFieldSet* unknown_fields);
2084 
2085     // Convenience functions to set an int field the right way, depending on
2086     // its wire type (a single int CppType can represent multiple wire types).
2087     void SetInt32(int number, int32 value, FieldDescriptor::Type type,
2088                   UnknownFieldSet* unknown_fields);
2089     void SetInt64(int number, int64 value, FieldDescriptor::Type type,
2090                   UnknownFieldSet* unknown_fields);
2091     void SetUInt32(int number, uint32 value, FieldDescriptor::Type type,
2092                    UnknownFieldSet* unknown_fields);
2093     void SetUInt64(int number, uint64 value, FieldDescriptor::Type type,
2094                    UnknownFieldSet* unknown_fields);
2095 
2096     // A helper function that adds an error at the specified location of the
2097     // option we're currently interpreting, and returns false.
AddOptionError(DescriptorPool::ErrorCollector::ErrorLocation location,const string & msg)2098     bool AddOptionError(DescriptorPool::ErrorCollector::ErrorLocation location,
2099                         const string& msg) {
2100       builder_->AddError(options_to_interpret_->element_name,
2101                          *uninterpreted_option_, location, msg);
2102       return false;
2103     }
2104 
2105     // A helper function that adds an error at the location of the option name
2106     // and returns false.
AddNameError(const string & msg)2107     bool AddNameError(const string& msg) {
2108       return AddOptionError(DescriptorPool::ErrorCollector::OPTION_NAME, msg);
2109     }
2110 
2111     // A helper function that adds an error at the location of the option name
2112     // and returns false.
AddValueError(const string & msg)2113     bool AddValueError(const string& msg) {
2114       return AddOptionError(DescriptorPool::ErrorCollector::OPTION_VALUE, msg);
2115     }
2116 
2117     // We interpret against this builder's pool. Is never NULL. We don't own
2118     // this pointer.
2119     DescriptorBuilder* builder_;
2120 
2121     // The options we're currently interpreting, or NULL if we're not in a call
2122     // to InterpretOptions.
2123     const OptionsToInterpret* options_to_interpret_;
2124 
2125     // The option we're currently interpreting within options_to_interpret_, or
2126     // NULL if we're not in a call to InterpretOptions(). This points to a
2127     // submessage of the original option, not the mutable copy. Therefore we
2128     // can use it to find locations recorded by the parser.
2129     const UninterpretedOption* uninterpreted_option_;
2130 
2131     GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(OptionInterpreter);
2132   };
2133 
2134   // Work-around for broken compilers:  According to the C++ standard,
2135   // OptionInterpreter should have access to the private members of any class
2136   // which has declared DescriptorBuilder as a friend.  Unfortunately some old
2137   // versions of GCC and other compilers do not implement this correctly.  So,
2138   // we have to have these intermediate methods to provide access.  We also
2139   // redundantly declare OptionInterpreter a friend just to make things extra
2140   // clear for these bad compilers.
2141   friend class OptionInterpreter;
get_allow_unknown(const DescriptorPool * pool)2142   static inline bool get_allow_unknown(const DescriptorPool* pool) {
2143     return pool->allow_unknown_;
2144   }
get_is_placeholder(const Descriptor * descriptor)2145   static inline bool get_is_placeholder(const Descriptor* descriptor) {
2146     return descriptor->is_placeholder_;
2147   }
2148 
2149   // Must be run only after options have been interpreted.
2150   //
2151   // NOTE: Validation code must only reference the options in the mutable
2152   // descriptors, which are the ones that have been interpreted. The const
2153   // proto references are passed in only so they can be provided to calls to
2154   // AddError(). Do not look at their options, which have not been interpreted.
2155   void ValidateFileOptions(FileDescriptor* file,
2156                            const FileDescriptorProto& proto);
2157   void ValidateMessageOptions(Descriptor* message,
2158                               const DescriptorProto& proto);
2159   void ValidateFieldOptions(FieldDescriptor* field,
2160                             const FieldDescriptorProto& proto);
2161   void ValidateEnumOptions(EnumDescriptor* enm,
2162                            const EnumDescriptorProto& proto);
2163   void ValidateEnumValueOptions(EnumValueDescriptor* enum_value,
2164                                 const EnumValueDescriptorProto& proto);
2165   void ValidateServiceOptions(ServiceDescriptor* service,
2166                               const ServiceDescriptorProto& proto);
2167   void ValidateMethodOptions(MethodDescriptor* method,
2168                              const MethodDescriptorProto& proto);
2169 
2170   void ValidateMapKey(FieldDescriptor* field,
2171                       const FieldDescriptorProto& proto);
2172 };
2173 
BuildFile(const FileDescriptorProto & proto)2174 const FileDescriptor* DescriptorPool::BuildFile(
2175     const FileDescriptorProto& proto) {
2176   GOOGLE_CHECK(fallback_database_ == NULL)
2177     << "Cannot call BuildFile on a DescriptorPool that uses a "
2178        "DescriptorDatabase.  You must instead find a way to get your file "
2179        "into the underlying database.";
2180   GOOGLE_CHECK(mutex_ == NULL);   // Implied by the above GOOGLE_CHECK.
2181   return DescriptorBuilder(this, tables_.get(), NULL).BuildFile(proto);
2182 }
2183 
BuildFileCollectingErrors(const FileDescriptorProto & proto,ErrorCollector * error_collector)2184 const FileDescriptor* DescriptorPool::BuildFileCollectingErrors(
2185     const FileDescriptorProto& proto,
2186     ErrorCollector* error_collector) {
2187   GOOGLE_CHECK(fallback_database_ == NULL)
2188     << "Cannot call BuildFile on a DescriptorPool that uses a "
2189        "DescriptorDatabase.  You must instead find a way to get your file "
2190        "into the underlying database.";
2191   GOOGLE_CHECK(mutex_ == NULL);   // Implied by the above GOOGLE_CHECK.
2192   return DescriptorBuilder(this, tables_.get(),
2193                            error_collector).BuildFile(proto);
2194 }
2195 
BuildFileFromDatabase(const FileDescriptorProto & proto) const2196 const FileDescriptor* DescriptorPool::BuildFileFromDatabase(
2197     const FileDescriptorProto& proto) const {
2198   mutex_->AssertHeld();
2199   return DescriptorBuilder(this, tables_.get(),
2200                            default_error_collector_).BuildFile(proto);
2201 }
2202 
DescriptorBuilder(const DescriptorPool * pool,DescriptorPool::Tables * tables,DescriptorPool::ErrorCollector * error_collector)2203 DescriptorBuilder::DescriptorBuilder(
2204     const DescriptorPool* pool,
2205     DescriptorPool::Tables* tables,
2206     DescriptorPool::ErrorCollector* error_collector)
2207   : pool_(pool),
2208     tables_(tables),
2209     error_collector_(error_collector),
2210     had_errors_(false),
2211     possible_undeclared_dependency_(NULL) {}
2212 
~DescriptorBuilder()2213 DescriptorBuilder::~DescriptorBuilder() {}
2214 
AddError(const string & element_name,const Message & descriptor,DescriptorPool::ErrorCollector::ErrorLocation location,const string & error)2215 void DescriptorBuilder::AddError(
2216     const string& element_name,
2217     const Message& descriptor,
2218     DescriptorPool::ErrorCollector::ErrorLocation location,
2219     const string& error) {
2220   if (error_collector_ == NULL) {
2221     if (!had_errors_) {
2222       GOOGLE_LOG(ERROR) << "Invalid proto descriptor for file \"" << filename_
2223                  << "\":";
2224     }
2225     GOOGLE_LOG(ERROR) << "  " << element_name << ": " << error;
2226   } else {
2227     error_collector_->AddError(filename_, element_name,
2228                                &descriptor, location, error);
2229   }
2230   had_errors_ = true;
2231 }
2232 
AddNotDefinedError(const string & element_name,const Message & descriptor,DescriptorPool::ErrorCollector::ErrorLocation location,const string & undefined_symbol)2233 void DescriptorBuilder::AddNotDefinedError(
2234     const string& element_name,
2235     const Message& descriptor,
2236     DescriptorPool::ErrorCollector::ErrorLocation location,
2237     const string& undefined_symbol) {
2238   if (possible_undeclared_dependency_ == NULL) {
2239     AddError(element_name, descriptor, location,
2240              "\"" + undefined_symbol + "\" is not defined.");
2241   } else {
2242     AddError(element_name, descriptor, location,
2243              "\"" + possible_undeclared_dependency_name_ +
2244              "\" seems to be defined in \"" +
2245              possible_undeclared_dependency_->name() + "\", which is not "
2246              "imported by \"" + filename_ + "\".  To use it here, please "
2247              "add the necessary import.");
2248   }
2249 }
2250 
IsInPackage(const FileDescriptor * file,const string & package_name)2251 bool DescriptorBuilder::IsInPackage(const FileDescriptor* file,
2252                                     const string& package_name) {
2253   return HasPrefixString(file->package(), package_name) &&
2254            (file->package().size() == package_name.size() ||
2255             file->package()[package_name.size()] == '.');
2256 }
2257 
FindSymbolNotEnforcingDeps(const string & name)2258 Symbol DescriptorBuilder::FindSymbolNotEnforcingDeps(const string& name) {
2259   Symbol result;
2260 
2261   // We need to search our pool and all its underlays.
2262   const DescriptorPool* pool = pool_;
2263   while (true) {
2264     // If we are looking at an underlay, we must lock its mutex_, since we are
2265     // accessing the underlay's tables_ dircetly.
2266     MutexLockMaybe lock((pool == pool_) ? NULL : pool->mutex_);
2267 
2268     // Note that we don't have to check fallback_database_ here because the
2269     // symbol has to be in one of its file's direct dependencies, and we have
2270     // already loaded those by the time we get here.
2271     result = pool->tables_->FindSymbol(name);
2272     if (!result.IsNull()) break;
2273     if (pool->underlay_ == NULL) return kNullSymbol;
2274     pool = pool->underlay_;
2275   }
2276 
2277   return result;
2278 }
2279 
FindSymbol(const string & name)2280 Symbol DescriptorBuilder::FindSymbol(const string& name) {
2281   Symbol result = FindSymbolNotEnforcingDeps(name);
2282 
2283   if (!pool_->enforce_dependencies_) {
2284     // Hack for CompilerUpgrader.
2285     return result;
2286   }
2287 
2288   // Only find symbols which were defined in this file or one of its
2289   // dependencies.
2290   const FileDescriptor* file = result.GetFile();
2291   if (file == file_) return result;
2292   for (int i = 0; i < file_->dependency_count(); i++) {
2293     if (file == file_->dependency(i)) return result;
2294   }
2295 
2296   if (result.type == Symbol::PACKAGE) {
2297     // Arg, this is overcomplicated.  The symbol is a package name.  It could
2298     // be that the package was defined in multiple files.  result.GetFile()
2299     // returns the first file we saw that used this package.  We've determined
2300     // that that file is not a direct dependency of the file we are currently
2301     // building, but it could be that some other file which *is* a direct
2302     // dependency also defines the same package.  We can't really rule out this
2303     // symbol unless none of the dependencies define it.
2304     if (IsInPackage(file_, name)) return result;
2305     for (int i = 0; i < file_->dependency_count(); i++) {
2306       // Note:  A dependency may be NULL if it was not found or had errors.
2307       if (file_->dependency(i) != NULL &&
2308           IsInPackage(file_->dependency(i), name)) {
2309         return result;
2310       }
2311     }
2312   }
2313 
2314   possible_undeclared_dependency_ = file;
2315   possible_undeclared_dependency_name_ = name;
2316   return kNullSymbol;
2317 }
2318 
LookupSymbolNoPlaceholder(const string & name,const string & relative_to,ResolveMode resolve_mode)2319 Symbol DescriptorBuilder::LookupSymbolNoPlaceholder(
2320     const string& name, const string& relative_to, ResolveMode resolve_mode) {
2321   possible_undeclared_dependency_ = NULL;
2322 
2323   if (name.size() > 0 && name[0] == '.') {
2324     // Fully-qualified name.
2325     return FindSymbol(name.substr(1));
2326   }
2327 
2328   // If name is something like "Foo.Bar.baz", and symbols named "Foo" are
2329   // defined in multiple parent scopes, we only want to find "Bar.baz" in the
2330   // innermost one.  E.g., the following should produce an error:
2331   //   message Bar { message Baz {} }
2332   //   message Foo {
2333   //     message Bar {
2334   //     }
2335   //     optional Bar.Baz baz = 1;
2336   //   }
2337   // So, we look for just "Foo" first, then look for "Bar.baz" within it if
2338   // found.
2339   int name_dot_pos = name.find_first_of('.');
2340   string first_part_of_name;
2341   if (name_dot_pos == string::npos) {
2342     first_part_of_name = name;
2343   } else {
2344     first_part_of_name = name.substr(0, name_dot_pos);
2345   }
2346 
2347   string scope_to_try(relative_to);
2348 
2349   while (true) {
2350     // Chop off the last component of the scope.
2351     string::size_type dot_pos = scope_to_try.find_last_of('.');
2352     if (dot_pos == string::npos) {
2353       return FindSymbol(name);
2354     } else {
2355       scope_to_try.erase(dot_pos);
2356     }
2357 
2358     // Append ".first_part_of_name" and try to find.
2359     string::size_type old_size = scope_to_try.size();
2360     scope_to_try.append(1, '.');
2361     scope_to_try.append(first_part_of_name);
2362     Symbol result = FindSymbol(scope_to_try);
2363     if (!result.IsNull()) {
2364       if (first_part_of_name.size() < name.size()) {
2365         // name is a compound symbol, of which we only found the first part.
2366         // Now try to look up the rest of it.
2367         if (result.IsAggregate()) {
2368           scope_to_try.append(name, first_part_of_name.size(),
2369                               name.size() - first_part_of_name.size());
2370           return FindSymbol(scope_to_try);
2371         } else {
2372           // We found a symbol but it's not an aggregate.  Continue the loop.
2373         }
2374       } else {
2375         if (resolve_mode == LOOKUP_TYPES && !result.IsType()) {
2376           // We found a symbol but it's not a type.  Continue the loop.
2377         } else {
2378           return result;
2379         }
2380       }
2381     }
2382 
2383     // Not found.  Remove the name so we can try again.
2384     scope_to_try.erase(old_size);
2385   }
2386 }
2387 
LookupSymbol(const string & name,const string & relative_to,PlaceholderType placeholder_type,ResolveMode resolve_mode)2388 Symbol DescriptorBuilder::LookupSymbol(
2389     const string& name, const string& relative_to,
2390     PlaceholderType placeholder_type, ResolveMode resolve_mode) {
2391   Symbol result = LookupSymbolNoPlaceholder(
2392       name, relative_to, resolve_mode);
2393   if (result.IsNull() && pool_->allow_unknown_) {
2394     // Not found, but AllowUnknownDependencies() is enabled.  Return a
2395     // placeholder instead.
2396     result = NewPlaceholder(name, placeholder_type);
2397   }
2398   return result;
2399 }
2400 
NewPlaceholder(const string & name,PlaceholderType placeholder_type)2401 Symbol DescriptorBuilder::NewPlaceholder(const string& name,
2402                                          PlaceholderType placeholder_type) {
2403   // Compute names.
2404   const string* placeholder_full_name;
2405   const string* placeholder_name;
2406   const string* placeholder_package;
2407 
2408   if (!ValidateQualifiedName(name)) return kNullSymbol;
2409   if (name[0] == '.') {
2410     // Fully-qualified.
2411     placeholder_full_name = tables_->AllocateString(name.substr(1));
2412   } else {
2413     placeholder_full_name = tables_->AllocateString(name);
2414   }
2415 
2416   string::size_type dotpos = placeholder_full_name->find_last_of('.');
2417   if (dotpos != string::npos) {
2418     placeholder_package = tables_->AllocateString(
2419       placeholder_full_name->substr(0, dotpos));
2420     placeholder_name = tables_->AllocateString(
2421       placeholder_full_name->substr(dotpos + 1));
2422   } else {
2423     placeholder_package = &kEmptyString;
2424     placeholder_name = placeholder_full_name;
2425   }
2426 
2427   // Create the placeholders.
2428   FileDescriptor* placeholder_file = tables_->Allocate<FileDescriptor>();
2429   memset(placeholder_file, 0, sizeof(*placeholder_file));
2430 
2431   placeholder_file->name_ =
2432     tables_->AllocateString(*placeholder_full_name + ".placeholder.proto");
2433   placeholder_file->package_ = placeholder_package;
2434   placeholder_file->pool_ = pool_;
2435   placeholder_file->options_ = &FileOptions::default_instance();
2436   placeholder_file->tables_ = &FileDescriptorTables::kEmpty;
2437   // All other fields are zero or NULL.
2438 
2439   if (placeholder_type == PLACEHOLDER_ENUM) {
2440     placeholder_file->enum_type_count_ = 1;
2441     placeholder_file->enum_types_ =
2442       tables_->AllocateArray<EnumDescriptor>(1);
2443 
2444     EnumDescriptor* placeholder_enum = &placeholder_file->enum_types_[0];
2445     memset(placeholder_enum, 0, sizeof(*placeholder_enum));
2446 
2447     placeholder_enum->full_name_ = placeholder_full_name;
2448     placeholder_enum->name_ = placeholder_name;
2449     placeholder_enum->file_ = placeholder_file;
2450     placeholder_enum->options_ = &EnumOptions::default_instance();
2451     placeholder_enum->is_placeholder_ = true;
2452     placeholder_enum->is_unqualified_placeholder_ = (name[0] != '.');
2453 
2454     // Enums must have at least one value.
2455     placeholder_enum->value_count_ = 1;
2456     placeholder_enum->values_ = tables_->AllocateArray<EnumValueDescriptor>(1);
2457 
2458     EnumValueDescriptor* placeholder_value = &placeholder_enum->values_[0];
2459     memset(placeholder_value, 0, sizeof(*placeholder_value));
2460 
2461     placeholder_value->name_ = tables_->AllocateString("PLACEHOLDER_VALUE");
2462     // Note that enum value names are siblings of their type, not children.
2463     placeholder_value->full_name_ =
2464       placeholder_package->empty() ? placeholder_value->name_ :
2465         tables_->AllocateString(*placeholder_package + ".PLACEHOLDER_VALUE");
2466 
2467     placeholder_value->number_ = 0;
2468     placeholder_value->type_ = placeholder_enum;
2469     placeholder_value->options_ = &EnumValueOptions::default_instance();
2470 
2471     return Symbol(placeholder_enum);
2472   } else {
2473     placeholder_file->message_type_count_ = 1;
2474     placeholder_file->message_types_ =
2475       tables_->AllocateArray<Descriptor>(1);
2476 
2477     Descriptor* placeholder_message = &placeholder_file->message_types_[0];
2478     memset(placeholder_message, 0, sizeof(*placeholder_message));
2479 
2480     placeholder_message->full_name_ = placeholder_full_name;
2481     placeholder_message->name_ = placeholder_name;
2482     placeholder_message->file_ = placeholder_file;
2483     placeholder_message->options_ = &MessageOptions::default_instance();
2484     placeholder_message->is_placeholder_ = true;
2485     placeholder_message->is_unqualified_placeholder_ = (name[0] != '.');
2486 
2487     if (placeholder_type == PLACEHOLDER_EXTENDABLE_MESSAGE) {
2488       placeholder_message->extension_range_count_ = 1;
2489       placeholder_message->extension_ranges_ =
2490         tables_->AllocateArray<Descriptor::ExtensionRange>(1);
2491       placeholder_message->extension_ranges_->start = 1;
2492       // kMaxNumber + 1 because ExtensionRange::end is exclusive.
2493       placeholder_message->extension_ranges_->end =
2494         FieldDescriptor::kMaxNumber + 1;
2495     }
2496 
2497     return Symbol(placeholder_message);
2498   }
2499 }
2500 
NewPlaceholderFile(const string & name)2501 const FileDescriptor* DescriptorBuilder::NewPlaceholderFile(
2502     const string& name) {
2503   FileDescriptor* placeholder = tables_->Allocate<FileDescriptor>();
2504   memset(placeholder, 0, sizeof(*placeholder));
2505 
2506   placeholder->name_ = tables_->AllocateString(name);
2507   placeholder->package_ = &kEmptyString;
2508   placeholder->pool_ = pool_;
2509   placeholder->options_ = &FileOptions::default_instance();
2510   placeholder->tables_ = &FileDescriptorTables::kEmpty;
2511   // All other fields are zero or NULL.
2512 
2513   return placeholder;
2514 }
2515 
AddSymbol(const string & full_name,const void * parent,const string & name,const Message & proto,Symbol symbol)2516 bool DescriptorBuilder::AddSymbol(
2517     const string& full_name, const void* parent, const string& name,
2518     const Message& proto, Symbol symbol) {
2519   // If the caller passed NULL for the parent, the symbol is at file scope.
2520   // Use its file as the parent instead.
2521   if (parent == NULL) parent = file_;
2522 
2523   if (tables_->AddSymbol(full_name, symbol)) {
2524     if (!file_tables_->AddAliasUnderParent(parent, name, symbol)) {
2525       GOOGLE_LOG(DFATAL) << "\"" << full_name << "\" not previously defined in "
2526                      "symbols_by_name_, but was defined in symbols_by_parent_; "
2527                      "this shouldn't be possible.";
2528       return false;
2529     }
2530     return true;
2531   } else {
2532     const FileDescriptor* other_file = tables_->FindSymbol(full_name).GetFile();
2533     if (other_file == file_) {
2534       string::size_type dot_pos = full_name.find_last_of('.');
2535       if (dot_pos == string::npos) {
2536         AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2537                  "\"" + full_name + "\" is already defined.");
2538       } else {
2539         AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2540                  "\"" + full_name.substr(dot_pos + 1) +
2541                  "\" is already defined in \"" +
2542                  full_name.substr(0, dot_pos) + "\".");
2543       }
2544     } else {
2545       // Symbol seems to have been defined in a different file.
2546       AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2547                "\"" + full_name + "\" is already defined in file \"" +
2548                other_file->name() + "\".");
2549     }
2550     return false;
2551   }
2552 }
2553 
AddPackage(const string & name,const Message & proto,const FileDescriptor * file)2554 void DescriptorBuilder::AddPackage(
2555     const string& name, const Message& proto, const FileDescriptor* file) {
2556   if (tables_->AddSymbol(name, Symbol(file))) {
2557     // Success.  Also add parent package, if any.
2558     string::size_type dot_pos = name.find_last_of('.');
2559     if (dot_pos == string::npos) {
2560       // No parents.
2561       ValidateSymbolName(name, name, proto);
2562     } else {
2563       // Has parent.
2564       string* parent_name = tables_->AllocateString(name.substr(0, dot_pos));
2565       AddPackage(*parent_name, proto, file);
2566       ValidateSymbolName(name.substr(dot_pos + 1), name, proto);
2567     }
2568   } else {
2569     Symbol existing_symbol = tables_->FindSymbol(name);
2570     // It's OK to redefine a package.
2571     if (existing_symbol.type != Symbol::PACKAGE) {
2572       // Symbol seems to have been defined in a different file.
2573       AddError(name, proto, DescriptorPool::ErrorCollector::NAME,
2574                "\"" + name + "\" is already defined (as something other than "
2575                "a package) in file \"" + existing_symbol.GetFile()->name() +
2576                "\".");
2577     }
2578   }
2579 }
2580 
ValidateSymbolName(const string & name,const string & full_name,const Message & proto)2581 void DescriptorBuilder::ValidateSymbolName(
2582     const string& name, const string& full_name, const Message& proto) {
2583   if (name.empty()) {
2584     AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2585              "Missing name.");
2586   } else {
2587     for (int i = 0; i < name.size(); i++) {
2588       // I don't trust isalnum() due to locales.  :(
2589       if ((name[i] < 'a' || 'z' < name[i]) &&
2590           (name[i] < 'A' || 'Z' < name[i]) &&
2591           (name[i] < '0' || '9' < name[i]) &&
2592           (name[i] != '_')) {
2593         AddError(full_name, proto, DescriptorPool::ErrorCollector::NAME,
2594                  "\"" + name + "\" is not a valid identifier.");
2595       }
2596     }
2597   }
2598 }
2599 
ValidateQualifiedName(const string & name)2600 bool DescriptorBuilder::ValidateQualifiedName(const string& name) {
2601   bool last_was_period = false;
2602 
2603   for (int i = 0; i < name.size(); i++) {
2604     // I don't trust isalnum() due to locales.  :(
2605     if (('a' <= name[i] && name[i] <= 'z') ||
2606         ('A' <= name[i] && name[i] <= 'Z') ||
2607         ('0' <= name[i] && name[i] <= '9') ||
2608         (name[i] == '_')) {
2609       last_was_period = false;
2610     } else if (name[i] == '.') {
2611       if (last_was_period) return false;
2612       last_was_period = true;
2613     } else {
2614       return false;
2615     }
2616   }
2617 
2618   return !name.empty() && !last_was_period;
2619 }
2620 
2621 // -------------------------------------------------------------------
2622 
2623 // This generic implementation is good for all descriptors except
2624 // FileDescriptor.
AllocateOptions(const typename DescriptorT::OptionsType & orig_options,DescriptorT * descriptor)2625 template<class DescriptorT> void DescriptorBuilder::AllocateOptions(
2626     const typename DescriptorT::OptionsType& orig_options,
2627     DescriptorT* descriptor) {
2628   AllocateOptionsImpl(descriptor->full_name(), descriptor->full_name(),
2629                       orig_options, descriptor);
2630 }
2631 
2632 // We specialize for FileDescriptor.
AllocateOptions(const FileOptions & orig_options,FileDescriptor * descriptor)2633 void DescriptorBuilder::AllocateOptions(const FileOptions& orig_options,
2634                                         FileDescriptor* descriptor) {
2635   // We add the dummy token so that LookupSymbol does the right thing.
2636   AllocateOptionsImpl(descriptor->package() + ".dummy", descriptor->name(),
2637                       orig_options, descriptor);
2638 }
2639 
AllocateOptionsImpl(const string & name_scope,const string & element_name,const typename DescriptorT::OptionsType & orig_options,DescriptorT * descriptor)2640 template<class DescriptorT> void DescriptorBuilder::AllocateOptionsImpl(
2641     const string& name_scope,
2642     const string& element_name,
2643     const typename DescriptorT::OptionsType& orig_options,
2644     DescriptorT* descriptor) {
2645   // We need to use a dummy pointer to work around a bug in older versions of
2646   // GCC.  Otherwise, the following two lines could be replaced with:
2647   //   typename DescriptorT::OptionsType* options =
2648   //       tables_->AllocateMessage<typename DescriptorT::OptionsType>();
2649   typename DescriptorT::OptionsType* const dummy = NULL;
2650   typename DescriptorT::OptionsType* options = tables_->AllocateMessage(dummy);
2651   options->CopyFrom(orig_options);
2652   descriptor->options_ = options;
2653 
2654   // Don't add to options_to_interpret_ unless there were uninterpreted
2655   // options.  This not only avoids unnecessary work, but prevents a
2656   // bootstrapping problem when building descriptors for descriptor.proto.
2657   // descriptor.proto does not contain any uninterpreted options, but
2658   // attempting to interpret options anyway will cause
2659   // OptionsType::GetDescriptor() to be called which may then deadlock since
2660   // we're still trying to build it.
2661   if (options->uninterpreted_option_size() > 0) {
2662     options_to_interpret_.push_back(
2663         OptionsToInterpret(name_scope, element_name, &orig_options, options));
2664   }
2665 }
2666 
2667 
2668 // A common pattern:  We want to convert a repeated field in the descriptor
2669 // to an array of values, calling some method to build each value.
2670 #define BUILD_ARRAY(INPUT, OUTPUT, NAME, METHOD, PARENT)             \
2671   OUTPUT->NAME##_count_ = INPUT.NAME##_size();                       \
2672   AllocateArray(INPUT.NAME##_size(), &OUTPUT->NAME##s_);             \
2673   for (int i = 0; i < INPUT.NAME##_size(); i++) {                    \
2674     METHOD(INPUT.NAME(i), PARENT, OUTPUT->NAME##s_ + i);             \
2675   }
2676 
BuildFile(const FileDescriptorProto & proto)2677 const FileDescriptor* DescriptorBuilder::BuildFile(
2678     const FileDescriptorProto& proto) {
2679   filename_ = proto.name();
2680 
2681   // Check if the file already exists and is identical to the one being built.
2682   // Note:  This only works if the input is canonical -- that is, it
2683   //   fully-qualifies all type names, has no UninterpretedOptions, etc.
2684   //   This is fine, because this idempotency "feature" really only exists to
2685   //   accomodate one hack in the proto1->proto2 migration layer.
2686   const FileDescriptor* existing_file = tables_->FindFile(filename_);
2687   if (existing_file != NULL) {
2688     // File already in pool.  Compare the existing one to the input.
2689     FileDescriptorProto existing_proto;
2690     existing_file->CopyTo(&existing_proto);
2691     if (existing_proto.SerializeAsString() == proto.SerializeAsString()) {
2692       // They're identical.  Return the existing descriptor.
2693       return existing_file;
2694     }
2695 
2696     // Not a match.  The error will be detected and handled later.
2697   }
2698 
2699   // Check to see if this file is already on the pending files list.
2700   // TODO(kenton):  Allow recursive imports?  It may not work with some
2701   //   (most?) programming languages.  E.g., in C++, a forward declaration
2702   //   of a type is not sufficient to allow it to be used even in a
2703   //   generated header file due to inlining.  This could perhaps be
2704   //   worked around using tricks involving inserting #include statements
2705   //   mid-file, but that's pretty ugly, and I'm pretty sure there are
2706   //   some languages out there that do not allow recursive dependencies
2707   //   at all.
2708   for (int i = 0; i < tables_->pending_files_.size(); i++) {
2709     if (tables_->pending_files_[i] == proto.name()) {
2710       string error_message("File recursively imports itself: ");
2711       for (; i < tables_->pending_files_.size(); i++) {
2712         error_message.append(tables_->pending_files_[i]);
2713         error_message.append(" -> ");
2714       }
2715       error_message.append(proto.name());
2716 
2717       AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
2718                error_message);
2719       return NULL;
2720     }
2721   }
2722 
2723   // If we have a fallback_database_, attempt to load all dependencies now,
2724   // before checkpointing tables_.  This avoids confusion with recursive
2725   // checkpoints.
2726   if (pool_->fallback_database_ != NULL) {
2727     tables_->pending_files_.push_back(proto.name());
2728     for (int i = 0; i < proto.dependency_size(); i++) {
2729       if (tables_->FindFile(proto.dependency(i)) == NULL &&
2730           (pool_->underlay_ == NULL ||
2731            pool_->underlay_->FindFileByName(proto.dependency(i)) == NULL)) {
2732         // We don't care what this returns since we'll find out below anyway.
2733         pool_->TryFindFileInFallbackDatabase(proto.dependency(i));
2734       }
2735     }
2736     tables_->pending_files_.pop_back();
2737   }
2738 
2739   // Checkpoint the tables so that we can roll back if something goes wrong.
2740   tables_->Checkpoint();
2741 
2742   FileDescriptor* result = tables_->Allocate<FileDescriptor>();
2743   file_ = result;
2744 
2745   file_tables_ = tables_->AllocateFileTables();
2746   file_->tables_ = file_tables_;
2747 
2748   if (!proto.has_name()) {
2749     AddError("", proto, DescriptorPool::ErrorCollector::OTHER,
2750              "Missing field: FileDescriptorProto.name.");
2751   }
2752 
2753   result->name_ = tables_->AllocateString(proto.name());
2754   if (proto.has_package()) {
2755     result->package_ = tables_->AllocateString(proto.package());
2756   } else {
2757     // We cannot rely on proto.package() returning a valid string if
2758     // proto.has_package() is false, because we might be running at static
2759     // initialization time, in which case default values have not yet been
2760     // initialized.
2761     result->package_ = tables_->AllocateString("");
2762   }
2763   result->pool_ = pool_;
2764 
2765   // Add to tables.
2766   if (!tables_->AddFile(result)) {
2767     AddError(proto.name(), proto, DescriptorPool::ErrorCollector::OTHER,
2768              "A file with this name is already in the pool.");
2769     // Bail out early so that if this is actually the exact same file, we
2770     // don't end up reporting that every single symbol is already defined.
2771     tables_->Rollback();
2772     return NULL;
2773   }
2774   if (!result->package().empty()) {
2775     AddPackage(result->package(), proto, result);
2776   }
2777 
2778   // Make sure all dependencies are loaded.
2779   set<string> seen_dependencies;
2780   result->dependency_count_ = proto.dependency_size();
2781   result->dependencies_ =
2782     tables_->AllocateArray<const FileDescriptor*>(proto.dependency_size());
2783   for (int i = 0; i < proto.dependency_size(); i++) {
2784     if (!seen_dependencies.insert(proto.dependency(i)).second) {
2785       AddError(proto.name(), proto,
2786                DescriptorPool::ErrorCollector::OTHER,
2787                "Import \"" + proto.dependency(i) + "\" was listed twice.");
2788     }
2789 
2790     const FileDescriptor* dependency = tables_->FindFile(proto.dependency(i));
2791     if (dependency == NULL && pool_->underlay_ != NULL) {
2792       dependency = pool_->underlay_->FindFileByName(proto.dependency(i));
2793     }
2794 
2795     if (dependency == NULL) {
2796       if (pool_->allow_unknown_) {
2797         dependency = NewPlaceholderFile(proto.dependency(i));
2798       } else {
2799         string message;
2800         if (pool_->fallback_database_ == NULL) {
2801           message = "Import \"" + proto.dependency(i) +
2802                     "\" has not been loaded.";
2803         } else {
2804           message = "Import \"" + proto.dependency(i) +
2805                     "\" was not found or had errors.";
2806         }
2807         AddError(proto.name(), proto,
2808                  DescriptorPool::ErrorCollector::OTHER,
2809                  message);
2810       }
2811     }
2812 
2813     result->dependencies_[i] = dependency;
2814   }
2815 
2816   // Convert children.
2817   BUILD_ARRAY(proto, result, message_type, BuildMessage  , NULL);
2818   BUILD_ARRAY(proto, result, enum_type   , BuildEnum     , NULL);
2819   BUILD_ARRAY(proto, result, service     , BuildService  , NULL);
2820   BUILD_ARRAY(proto, result, extension   , BuildExtension, NULL);
2821 
2822   // Copy options.
2823   if (!proto.has_options()) {
2824     result->options_ = NULL;  // Will set to default_instance later.
2825   } else {
2826     AllocateOptions(proto.options(), result);
2827   }
2828 
2829   // Note that the following steps must occur in exactly the specified order.
2830 
2831   // Cross-link.
2832   CrossLinkFile(result, proto);
2833 
2834   // Interpret any remaining uninterpreted options gathered into
2835   // options_to_interpret_ during descriptor building.  Cross-linking has made
2836   // extension options known, so all interpretations should now succeed.
2837   if (!had_errors_) {
2838     OptionInterpreter option_interpreter(this);
2839     for (vector<OptionsToInterpret>::iterator iter =
2840              options_to_interpret_.begin();
2841          iter != options_to_interpret_.end(); ++iter) {
2842       option_interpreter.InterpretOptions(&(*iter));
2843     }
2844     options_to_interpret_.clear();
2845   }
2846 
2847   // Validate options.
2848   if (!had_errors_) {
2849     ValidateFileOptions(result, proto);
2850   }
2851 
2852   if (had_errors_) {
2853     tables_->Rollback();
2854     return NULL;
2855   } else {
2856     tables_->Checkpoint();
2857     return result;
2858   }
2859 }
2860 
BuildMessage(const DescriptorProto & proto,const Descriptor * parent,Descriptor * result)2861 void DescriptorBuilder::BuildMessage(const DescriptorProto& proto,
2862                                      const Descriptor* parent,
2863                                      Descriptor* result) {
2864   const string& scope = (parent == NULL) ?
2865     file_->package() : parent->full_name();
2866   string* full_name = tables_->AllocateString(scope);
2867   if (!full_name->empty()) full_name->append(1, '.');
2868   full_name->append(proto.name());
2869 
2870   ValidateSymbolName(proto.name(), *full_name, proto);
2871 
2872   result->name_            = tables_->AllocateString(proto.name());
2873   result->full_name_       = full_name;
2874   result->file_            = file_;
2875   result->containing_type_ = parent;
2876   result->is_placeholder_  = false;
2877   result->is_unqualified_placeholder_ = false;
2878 
2879   BUILD_ARRAY(proto, result, field          , BuildField         , result);
2880   BUILD_ARRAY(proto, result, nested_type    , BuildMessage       , result);
2881   BUILD_ARRAY(proto, result, enum_type      , BuildEnum          , result);
2882   BUILD_ARRAY(proto, result, extension_range, BuildExtensionRange, result);
2883   BUILD_ARRAY(proto, result, extension      , BuildExtension     , result);
2884 
2885   // Copy options.
2886   if (!proto.has_options()) {
2887     result->options_ = NULL;  // Will set to default_instance later.
2888   } else {
2889     AllocateOptions(proto.options(), result);
2890   }
2891 
2892   AddSymbol(result->full_name(), parent, result->name(),
2893             proto, Symbol(result));
2894 
2895   // Check that no fields have numbers in extension ranges.
2896   for (int i = 0; i < result->field_count(); i++) {
2897     const FieldDescriptor* field = result->field(i);
2898     for (int j = 0; j < result->extension_range_count(); j++) {
2899       const Descriptor::ExtensionRange* range = result->extension_range(j);
2900       if (range->start <= field->number() && field->number() < range->end) {
2901         AddError(field->full_name(), proto.extension_range(j),
2902                  DescriptorPool::ErrorCollector::NUMBER,
2903                  strings::Substitute(
2904                    "Extension range $0 to $1 includes field \"$2\" ($3).",
2905                    range->start, range->end - 1,
2906                    field->name(), field->number()));
2907       }
2908     }
2909   }
2910 
2911   // Check that extension ranges don't overlap.
2912   for (int i = 0; i < result->extension_range_count(); i++) {
2913     const Descriptor::ExtensionRange* range1 = result->extension_range(i);
2914     for (int j = i + 1; j < result->extension_range_count(); j++) {
2915       const Descriptor::ExtensionRange* range2 = result->extension_range(j);
2916       if (range1->end > range2->start && range2->end > range1->start) {
2917         AddError(result->full_name(), proto.extension_range(j),
2918                  DescriptorPool::ErrorCollector::NUMBER,
2919                  strings::Substitute("Extension range $0 to $1 overlaps with "
2920                                      "already-defined range $2 to $3.",
2921                                      range2->start, range2->end - 1,
2922                                      range1->start, range1->end - 1));
2923       }
2924     }
2925   }
2926 }
2927 
BuildFieldOrExtension(const FieldDescriptorProto & proto,const Descriptor * parent,FieldDescriptor * result,bool is_extension)2928 void DescriptorBuilder::BuildFieldOrExtension(const FieldDescriptorProto& proto,
2929                                               const Descriptor* parent,
2930                                               FieldDescriptor* result,
2931                                               bool is_extension) {
2932   const string& scope = (parent == NULL) ?
2933     file_->package() : parent->full_name();
2934   string* full_name = tables_->AllocateString(scope);
2935   if (!full_name->empty()) full_name->append(1, '.');
2936   full_name->append(proto.name());
2937 
2938   ValidateSymbolName(proto.name(), *full_name, proto);
2939 
2940   result->name_         = tables_->AllocateString(proto.name());
2941   result->full_name_    = full_name;
2942   result->file_         = file_;
2943   result->number_       = proto.number();
2944   result->is_extension_ = is_extension;
2945 
2946   // If .proto files follow the style guide then the name should already be
2947   // lower-cased.  If that's the case we can just reuse the string we already
2948   // allocated rather than allocate a new one.
2949   string lowercase_name(proto.name());
2950   LowerString(&lowercase_name);
2951   if (lowercase_name == proto.name()) {
2952     result->lowercase_name_ = result->name_;
2953   } else {
2954     result->lowercase_name_ = tables_->AllocateString(lowercase_name);
2955   }
2956 
2957   // Don't bother with the above optimization for camel-case names since
2958   // .proto files that follow the guide shouldn't be using names in this
2959   // format, so the optimization wouldn't help much.
2960   result->camelcase_name_ = tables_->AllocateString(ToCamelCase(proto.name()));
2961 
2962   // Some compilers do not allow static_cast directly between two enum types,
2963   // so we must cast to int first.
2964   result->type_  = static_cast<FieldDescriptor::Type>(
2965                      implicit_cast<int>(proto.type()));
2966   result->label_ = static_cast<FieldDescriptor::Label>(
2967                      implicit_cast<int>(proto.label()));
2968 
2969   // Some of these may be filled in when cross-linking.
2970   result->containing_type_ = NULL;
2971   result->extension_scope_ = NULL;
2972   result->experimental_map_key_ = NULL;
2973   result->message_type_ = NULL;
2974   result->enum_type_ = NULL;
2975 
2976   result->has_default_value_ = proto.has_default_value();
2977   if (proto.has_default_value() && result->is_repeated()) {
2978     AddError(result->full_name(), proto,
2979              DescriptorPool::ErrorCollector::DEFAULT_VALUE,
2980              "Repeated fields can't have default values.");
2981   }
2982 
2983   if (proto.has_type()) {
2984     if (proto.has_default_value()) {
2985       char* end_pos = NULL;
2986       switch (result->cpp_type()) {
2987         case FieldDescriptor::CPPTYPE_INT32:
2988           result->default_value_int32_ =
2989             strtol(proto.default_value().c_str(), &end_pos, 0);
2990           break;
2991         case FieldDescriptor::CPPTYPE_INT64:
2992           result->default_value_int64_ =
2993             strto64(proto.default_value().c_str(), &end_pos, 0);
2994           break;
2995         case FieldDescriptor::CPPTYPE_UINT32:
2996           result->default_value_uint32_ =
2997             strtoul(proto.default_value().c_str(), &end_pos, 0);
2998           break;
2999         case FieldDescriptor::CPPTYPE_UINT64:
3000           result->default_value_uint64_ =
3001             strtou64(proto.default_value().c_str(), &end_pos, 0);
3002           break;
3003         case FieldDescriptor::CPPTYPE_FLOAT:
3004           if (proto.default_value() == "inf") {
3005             result->default_value_float_ = numeric_limits<float>::infinity();
3006           } else if (proto.default_value() == "-inf") {
3007             result->default_value_float_ = -numeric_limits<float>::infinity();
3008           } else if (proto.default_value() == "nan") {
3009             result->default_value_float_ = numeric_limits<float>::quiet_NaN();
3010           } else  {
3011             result->default_value_float_ =
3012               NoLocaleStrtod(proto.default_value().c_str(), &end_pos);
3013           }
3014           break;
3015         case FieldDescriptor::CPPTYPE_DOUBLE:
3016           if (proto.default_value() == "inf") {
3017             result->default_value_double_ = numeric_limits<double>::infinity();
3018           } else if (proto.default_value() == "-inf") {
3019             result->default_value_double_ = -numeric_limits<double>::infinity();
3020           } else if (proto.default_value() == "nan") {
3021             result->default_value_double_ = numeric_limits<double>::quiet_NaN();
3022           } else  {
3023             result->default_value_double_ =
3024               NoLocaleStrtod(proto.default_value().c_str(), &end_pos);
3025           }
3026           break;
3027         case FieldDescriptor::CPPTYPE_BOOL:
3028           if (proto.default_value() == "true") {
3029             result->default_value_bool_ = true;
3030           } else if (proto.default_value() == "false") {
3031             result->default_value_bool_ = false;
3032           } else {
3033             AddError(result->full_name(), proto,
3034                      DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3035                      "Boolean default must be true or false.");
3036           }
3037           break;
3038         case FieldDescriptor::CPPTYPE_ENUM:
3039           // This will be filled in when cross-linking.
3040           result->default_value_enum_ = NULL;
3041           break;
3042         case FieldDescriptor::CPPTYPE_STRING:
3043           if (result->type() == FieldDescriptor::TYPE_BYTES) {
3044             result->default_value_string_ = tables_->AllocateString(
3045               UnescapeCEscapeString(proto.default_value()));
3046           } else {
3047             result->default_value_string_ =
3048               tables_->AllocateString(proto.default_value());
3049           }
3050           break;
3051         case FieldDescriptor::CPPTYPE_MESSAGE:
3052           AddError(result->full_name(), proto,
3053                    DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3054                    "Messages can't have default values.");
3055           result->has_default_value_ = false;
3056           break;
3057       }
3058 
3059       if (end_pos != NULL) {
3060         // end_pos is only set non-NULL by the parsers for numeric types, above.
3061         // This checks that the default was non-empty and had no extra junk
3062         // after the end of the number.
3063         if (proto.default_value().empty() || *end_pos != '\0') {
3064           AddError(result->full_name(), proto,
3065                    DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3066                    "Couldn't parse default value.");
3067         }
3068       }
3069     } else {
3070       // No explicit default value
3071       switch (result->cpp_type()) {
3072         case FieldDescriptor::CPPTYPE_INT32:
3073           result->default_value_int32_ = 0;
3074           break;
3075         case FieldDescriptor::CPPTYPE_INT64:
3076           result->default_value_int64_ = 0;
3077           break;
3078         case FieldDescriptor::CPPTYPE_UINT32:
3079           result->default_value_uint32_ = 0;
3080           break;
3081         case FieldDescriptor::CPPTYPE_UINT64:
3082           result->default_value_uint64_ = 0;
3083           break;
3084         case FieldDescriptor::CPPTYPE_FLOAT:
3085           result->default_value_float_ = 0.0f;
3086           break;
3087         case FieldDescriptor::CPPTYPE_DOUBLE:
3088           result->default_value_double_ = 0.0;
3089           break;
3090         case FieldDescriptor::CPPTYPE_BOOL:
3091           result->default_value_bool_ = false;
3092           break;
3093         case FieldDescriptor::CPPTYPE_ENUM:
3094           // This will be filled in when cross-linking.
3095           result->default_value_enum_ = NULL;
3096           break;
3097         case FieldDescriptor::CPPTYPE_STRING:
3098           result->default_value_string_ = &kEmptyString;
3099           break;
3100         case FieldDescriptor::CPPTYPE_MESSAGE:
3101           break;
3102       }
3103     }
3104   }
3105 
3106   if (result->number() <= 0) {
3107     AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
3108              "Field numbers must be positive integers.");
3109   } else if (result->number() > FieldDescriptor::kMaxNumber) {
3110     AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
3111              strings::Substitute("Field numbers cannot be greater than $0.",
3112                                  FieldDescriptor::kMaxNumber));
3113   } else if (result->number() >= FieldDescriptor::kFirstReservedNumber &&
3114              result->number() <= FieldDescriptor::kLastReservedNumber) {
3115     AddError(result->full_name(), proto, DescriptorPool::ErrorCollector::NUMBER,
3116              strings::Substitute(
3117                "Field numbers $0 through $1 are reserved for the protocol "
3118                "buffer library implementation.",
3119                FieldDescriptor::kFirstReservedNumber,
3120                FieldDescriptor::kLastReservedNumber));
3121   }
3122 
3123   if (is_extension) {
3124     if (!proto.has_extendee()) {
3125       AddError(result->full_name(), proto,
3126                DescriptorPool::ErrorCollector::EXTENDEE,
3127                "FieldDescriptorProto.extendee not set for extension field.");
3128     }
3129 
3130     result->extension_scope_ = parent;
3131   } else {
3132     if (proto.has_extendee()) {
3133       AddError(result->full_name(), proto,
3134                DescriptorPool::ErrorCollector::EXTENDEE,
3135                "FieldDescriptorProto.extendee set for non-extension field.");
3136     }
3137 
3138     result->containing_type_ = parent;
3139   }
3140 
3141   // Copy options.
3142   if (!proto.has_options()) {
3143     result->options_ = NULL;  // Will set to default_instance later.
3144   } else {
3145     AllocateOptions(proto.options(), result);
3146   }
3147 
3148   AddSymbol(result->full_name(), parent, result->name(),
3149             proto, Symbol(result));
3150 }
3151 
BuildExtensionRange(const DescriptorProto::ExtensionRange & proto,const Descriptor * parent,Descriptor::ExtensionRange * result)3152 void DescriptorBuilder::BuildExtensionRange(
3153     const DescriptorProto::ExtensionRange& proto,
3154     const Descriptor* parent,
3155     Descriptor::ExtensionRange* result) {
3156   result->start = proto.start();
3157   result->end = proto.end();
3158   if (result->start <= 0) {
3159     AddError(parent->full_name(), proto,
3160              DescriptorPool::ErrorCollector::NUMBER,
3161              "Extension numbers must be positive integers.");
3162   }
3163 
3164   if (result->end > FieldDescriptor::kMaxNumber + 1) {
3165     AddError(parent->full_name(), proto,
3166              DescriptorPool::ErrorCollector::NUMBER,
3167              strings::Substitute("Extension numbers cannot be greater than $0.",
3168                                  FieldDescriptor::kMaxNumber));
3169   }
3170 
3171   if (result->start >= result->end) {
3172     AddError(parent->full_name(), proto,
3173              DescriptorPool::ErrorCollector::NUMBER,
3174              "Extension range end number must be greater than start number.");
3175   }
3176 }
3177 
BuildEnum(const EnumDescriptorProto & proto,const Descriptor * parent,EnumDescriptor * result)3178 void DescriptorBuilder::BuildEnum(const EnumDescriptorProto& proto,
3179                                   const Descriptor* parent,
3180                                   EnumDescriptor* result) {
3181   const string& scope = (parent == NULL) ?
3182     file_->package() : parent->full_name();
3183   string* full_name = tables_->AllocateString(scope);
3184   if (!full_name->empty()) full_name->append(1, '.');
3185   full_name->append(proto.name());
3186 
3187   ValidateSymbolName(proto.name(), *full_name, proto);
3188 
3189   result->name_            = tables_->AllocateString(proto.name());
3190   result->full_name_       = full_name;
3191   result->file_            = file_;
3192   result->containing_type_ = parent;
3193   result->is_placeholder_  = false;
3194   result->is_unqualified_placeholder_ = false;
3195 
3196   if (proto.value_size() == 0) {
3197     // We cannot allow enums with no values because this would mean there
3198     // would be no valid default value for fields of this type.
3199     AddError(result->full_name(), proto,
3200              DescriptorPool::ErrorCollector::NAME,
3201              "Enums must contain at least one value.");
3202   }
3203 
3204   BUILD_ARRAY(proto, result, value, BuildEnumValue, result);
3205 
3206   // Copy options.
3207   if (!proto.has_options()) {
3208     result->options_ = NULL;  // Will set to default_instance later.
3209   } else {
3210     AllocateOptions(proto.options(), result);
3211   }
3212 
3213   AddSymbol(result->full_name(), parent, result->name(),
3214             proto, Symbol(result));
3215 }
3216 
BuildEnumValue(const EnumValueDescriptorProto & proto,const EnumDescriptor * parent,EnumValueDescriptor * result)3217 void DescriptorBuilder::BuildEnumValue(const EnumValueDescriptorProto& proto,
3218                                        const EnumDescriptor* parent,
3219                                        EnumValueDescriptor* result) {
3220   result->name_   = tables_->AllocateString(proto.name());
3221   result->number_ = proto.number();
3222   result->type_   = parent;
3223 
3224   // Note:  full_name for enum values is a sibling to the parent's name, not a
3225   //   child of it.
3226   string* full_name = tables_->AllocateString(*parent->full_name_);
3227   full_name->resize(full_name->size() - parent->name_->size());
3228   full_name->append(*result->name_);
3229   result->full_name_ = full_name;
3230 
3231   ValidateSymbolName(proto.name(), *full_name, proto);
3232 
3233   // Copy options.
3234   if (!proto.has_options()) {
3235     result->options_ = NULL;  // Will set to default_instance later.
3236   } else {
3237     AllocateOptions(proto.options(), result);
3238   }
3239 
3240   // Again, enum values are weird because we makes them appear as siblings
3241   // of the enum type instead of children of it.  So, we use
3242   // parent->containing_type() as the value's parent.
3243   bool added_to_outer_scope =
3244     AddSymbol(result->full_name(), parent->containing_type(), result->name(),
3245               proto, Symbol(result));
3246 
3247   // However, we also want to be able to search for values within a single
3248   // enum type, so we add it as a child of the enum type itself, too.
3249   // Note:  This could fail, but if it does, the error has already been
3250   //   reported by the above AddSymbol() call, so we ignore the return code.
3251   bool added_to_inner_scope =
3252     file_tables_->AddAliasUnderParent(parent, result->name(), Symbol(result));
3253 
3254   if (added_to_inner_scope && !added_to_outer_scope) {
3255     // This value did not conflict with any values defined in the same enum,
3256     // but it did conflict with some other symbol defined in the enum type's
3257     // scope.  Let's print an additional error to explain this.
3258     string outer_scope;
3259     if (parent->containing_type() == NULL) {
3260       outer_scope = file_->package();
3261     } else {
3262       outer_scope = parent->containing_type()->full_name();
3263     }
3264 
3265     if (outer_scope.empty()) {
3266       outer_scope = "the global scope";
3267     } else {
3268       outer_scope = "\"" + outer_scope + "\"";
3269     }
3270 
3271     AddError(result->full_name(), proto,
3272              DescriptorPool::ErrorCollector::NAME,
3273              "Note that enum values use C++ scoping rules, meaning that "
3274              "enum values are siblings of their type, not children of it.  "
3275              "Therefore, \"" + result->name() + "\" must be unique within "
3276              + outer_scope + ", not just within \"" + parent->name() + "\".");
3277   }
3278 
3279   // An enum is allowed to define two numbers that refer to the same value.
3280   // FindValueByNumber() should return the first such value, so we simply
3281   // ignore AddEnumValueByNumber()'s return code.
3282   file_tables_->AddEnumValueByNumber(result);
3283 }
3284 
BuildService(const ServiceDescriptorProto & proto,const void * dummy,ServiceDescriptor * result)3285 void DescriptorBuilder::BuildService(const ServiceDescriptorProto& proto,
3286                                      const void* dummy,
3287                                      ServiceDescriptor* result) {
3288   string* full_name = tables_->AllocateString(file_->package());
3289   if (!full_name->empty()) full_name->append(1, '.');
3290   full_name->append(proto.name());
3291 
3292   ValidateSymbolName(proto.name(), *full_name, proto);
3293 
3294   result->name_      = tables_->AllocateString(proto.name());
3295   result->full_name_ = full_name;
3296   result->file_      = file_;
3297 
3298   BUILD_ARRAY(proto, result, method, BuildMethod, result);
3299 
3300   // Copy options.
3301   if (!proto.has_options()) {
3302     result->options_ = NULL;  // Will set to default_instance later.
3303   } else {
3304     AllocateOptions(proto.options(), result);
3305   }
3306 
3307   AddSymbol(result->full_name(), NULL, result->name(),
3308             proto, Symbol(result));
3309 }
3310 
BuildMethod(const MethodDescriptorProto & proto,const ServiceDescriptor * parent,MethodDescriptor * result)3311 void DescriptorBuilder::BuildMethod(const MethodDescriptorProto& proto,
3312                                     const ServiceDescriptor* parent,
3313                                     MethodDescriptor* result) {
3314   result->name_    = tables_->AllocateString(proto.name());
3315   result->service_ = parent;
3316 
3317   string* full_name = tables_->AllocateString(parent->full_name());
3318   full_name->append(1, '.');
3319   full_name->append(*result->name_);
3320   result->full_name_ = full_name;
3321 
3322   ValidateSymbolName(proto.name(), *full_name, proto);
3323 
3324   // These will be filled in when cross-linking.
3325   result->input_type_ = NULL;
3326   result->output_type_ = NULL;
3327 
3328   // Copy options.
3329   if (!proto.has_options()) {
3330     result->options_ = NULL;  // Will set to default_instance later.
3331   } else {
3332     AllocateOptions(proto.options(), result);
3333   }
3334 
3335   AddSymbol(result->full_name(), parent, result->name(),
3336             proto, Symbol(result));
3337 }
3338 
3339 #undef BUILD_ARRAY
3340 
3341 // -------------------------------------------------------------------
3342 
CrossLinkFile(FileDescriptor * file,const FileDescriptorProto & proto)3343 void DescriptorBuilder::CrossLinkFile(
3344     FileDescriptor* file, const FileDescriptorProto& proto) {
3345   if (file->options_ == NULL) {
3346     file->options_ = &FileOptions::default_instance();
3347   }
3348 
3349   for (int i = 0; i < file->message_type_count(); i++) {
3350     CrossLinkMessage(&file->message_types_[i], proto.message_type(i));
3351   }
3352 
3353   for (int i = 0; i < file->extension_count(); i++) {
3354     CrossLinkField(&file->extensions_[i], proto.extension(i));
3355   }
3356 
3357   for (int i = 0; i < file->enum_type_count(); i++) {
3358     CrossLinkEnum(&file->enum_types_[i], proto.enum_type(i));
3359   }
3360 
3361   for (int i = 0; i < file->service_count(); i++) {
3362     CrossLinkService(&file->services_[i], proto.service(i));
3363   }
3364 }
3365 
CrossLinkMessage(Descriptor * message,const DescriptorProto & proto)3366 void DescriptorBuilder::CrossLinkMessage(
3367     Descriptor* message, const DescriptorProto& proto) {
3368   if (message->options_ == NULL) {
3369     message->options_ = &MessageOptions::default_instance();
3370   }
3371 
3372   for (int i = 0; i < message->nested_type_count(); i++) {
3373     CrossLinkMessage(&message->nested_types_[i], proto.nested_type(i));
3374   }
3375 
3376   for (int i = 0; i < message->enum_type_count(); i++) {
3377     CrossLinkEnum(&message->enum_types_[i], proto.enum_type(i));
3378   }
3379 
3380   for (int i = 0; i < message->field_count(); i++) {
3381     CrossLinkField(&message->fields_[i], proto.field(i));
3382   }
3383 
3384   for (int i = 0; i < message->extension_count(); i++) {
3385     CrossLinkField(&message->extensions_[i], proto.extension(i));
3386   }
3387 }
3388 
CrossLinkField(FieldDescriptor * field,const FieldDescriptorProto & proto)3389 void DescriptorBuilder::CrossLinkField(
3390     FieldDescriptor* field, const FieldDescriptorProto& proto) {
3391   if (field->options_ == NULL) {
3392     field->options_ = &FieldOptions::default_instance();
3393   }
3394 
3395   if (proto.has_extendee()) {
3396     Symbol extendee = LookupSymbol(proto.extendee(), field->full_name(),
3397                                    PLACEHOLDER_EXTENDABLE_MESSAGE);
3398     if (extendee.IsNull()) {
3399       AddNotDefinedError(field->full_name(), proto,
3400                          DescriptorPool::ErrorCollector::EXTENDEE,
3401                          proto.extendee());
3402       return;
3403     } else if (extendee.type != Symbol::MESSAGE) {
3404       AddError(field->full_name(), proto,
3405                DescriptorPool::ErrorCollector::EXTENDEE,
3406                "\"" + proto.extendee() + "\" is not a message type.");
3407       return;
3408     }
3409     field->containing_type_ = extendee.descriptor;
3410 
3411     if (!field->containing_type()->IsExtensionNumber(field->number())) {
3412       AddError(field->full_name(), proto,
3413                DescriptorPool::ErrorCollector::NUMBER,
3414                strings::Substitute("\"$0\" does not declare $1 as an "
3415                                    "extension number.",
3416                                    field->containing_type()->full_name(),
3417                                    field->number()));
3418     }
3419   }
3420 
3421   if (proto.has_type_name()) {
3422     // Assume we are expecting a message type unless the proto contains some
3423     // evidence that it expects an enum type.  This only makes a difference if
3424     // we end up creating a placeholder.
3425     bool expecting_enum = (proto.type() == FieldDescriptorProto::TYPE_ENUM) ||
3426                           proto.has_default_value();
3427 
3428     Symbol type =
3429       LookupSymbol(proto.type_name(), field->full_name(),
3430                    expecting_enum ? PLACEHOLDER_ENUM : PLACEHOLDER_MESSAGE,
3431                    LOOKUP_TYPES);
3432 
3433     if (type.IsNull()) {
3434       AddNotDefinedError(field->full_name(), proto,
3435                          DescriptorPool::ErrorCollector::TYPE,
3436                          proto.type_name());
3437       return;
3438     }
3439 
3440     if (!proto.has_type()) {
3441       // Choose field type based on symbol.
3442       if (type.type == Symbol::MESSAGE) {
3443         field->type_ = FieldDescriptor::TYPE_MESSAGE;
3444       } else if (type.type == Symbol::ENUM) {
3445         field->type_ = FieldDescriptor::TYPE_ENUM;
3446       } else {
3447         AddError(field->full_name(), proto,
3448                  DescriptorPool::ErrorCollector::TYPE,
3449                  "\"" + proto.type_name() + "\" is not a type.");
3450         return;
3451       }
3452     }
3453 
3454     if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
3455       if (type.type != Symbol::MESSAGE) {
3456         AddError(field->full_name(), proto,
3457                  DescriptorPool::ErrorCollector::TYPE,
3458                  "\"" + proto.type_name() + "\" is not a message type.");
3459         return;
3460       }
3461       field->message_type_ = type.descriptor;
3462 
3463       if (field->has_default_value()) {
3464         AddError(field->full_name(), proto,
3465                  DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3466                  "Messages can't have default values.");
3467       }
3468     } else if (field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM) {
3469       if (type.type != Symbol::ENUM) {
3470         AddError(field->full_name(), proto,
3471                  DescriptorPool::ErrorCollector::TYPE,
3472                  "\"" + proto.type_name() + "\" is not an enum type.");
3473         return;
3474       }
3475       field->enum_type_ = type.enum_descriptor;
3476 
3477       if (field->enum_type()->is_placeholder_) {
3478         // We can't look up default values for placeholder types.  We'll have
3479         // to just drop them.
3480         field->has_default_value_ = false;
3481       }
3482 
3483       if (field->has_default_value()) {
3484         // We can't just use field->enum_type()->FindValueByName() here
3485         // because that locks the pool's mutex, which we have already locked
3486         // at this point.
3487         Symbol default_value =
3488           LookupSymbolNoPlaceholder(proto.default_value(),
3489                                     field->enum_type()->full_name());
3490 
3491         if (default_value.type == Symbol::ENUM_VALUE &&
3492             default_value.enum_value_descriptor->type() == field->enum_type()) {
3493           field->default_value_enum_ = default_value.enum_value_descriptor;
3494         } else {
3495           AddError(field->full_name(), proto,
3496                    DescriptorPool::ErrorCollector::DEFAULT_VALUE,
3497                    "Enum type \"" + field->enum_type()->full_name() +
3498                    "\" has no value named \"" + proto.default_value() + "\".");
3499         }
3500       } else if (field->enum_type()->value_count() > 0) {
3501         // All enums must have at least one value, or we would have reported
3502         // an error elsewhere.  We use the first defined value as the default
3503         // if a default is not explicitly defined.
3504         field->default_value_enum_ = field->enum_type()->value(0);
3505       }
3506     } else {
3507       AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3508                "Field with primitive type has type_name.");
3509     }
3510   } else {
3511     if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE ||
3512         field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM) {
3513       AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3514                "Field with message or enum type missing type_name.");
3515     }
3516   }
3517 
3518   // Add the field to the fields-by-number table.
3519   // Note:  We have to do this *after* cross-linking because extensions do not
3520   //   know their containing type until now.
3521   if (!file_tables_->AddFieldByNumber(field)) {
3522     const FieldDescriptor* conflicting_field =
3523       file_tables_->FindFieldByNumber(field->containing_type(),
3524                                       field->number());
3525     if (field->is_extension()) {
3526       AddError(field->full_name(), proto,
3527                DescriptorPool::ErrorCollector::NUMBER,
3528                strings::Substitute("Extension number $0 has already been used "
3529                                    "in \"$1\" by extension \"$2\".",
3530                                    field->number(),
3531                                    field->containing_type()->full_name(),
3532                                    conflicting_field->full_name()));
3533     } else {
3534       AddError(field->full_name(), proto,
3535                DescriptorPool::ErrorCollector::NUMBER,
3536                strings::Substitute("Field number $0 has already been used in "
3537                                    "\"$1\" by field \"$2\".",
3538                                    field->number(),
3539                                    field->containing_type()->full_name(),
3540                                    conflicting_field->name()));
3541     }
3542   }
3543 
3544   if (field->is_extension()) {
3545     // No need for error checking: if the extension number collided,
3546     // we've already been informed of it by the if() above.
3547     tables_->AddExtension(field);
3548   }
3549 
3550   // Add the field to the lowercase-name and camelcase-name tables.
3551   file_tables_->AddFieldByStylizedNames(field);
3552 }
3553 
CrossLinkEnum(EnumDescriptor * enum_type,const EnumDescriptorProto & proto)3554 void DescriptorBuilder::CrossLinkEnum(
3555     EnumDescriptor* enum_type, const EnumDescriptorProto& proto) {
3556   if (enum_type->options_ == NULL) {
3557     enum_type->options_ = &EnumOptions::default_instance();
3558   }
3559 
3560   for (int i = 0; i < enum_type->value_count(); i++) {
3561     CrossLinkEnumValue(&enum_type->values_[i], proto.value(i));
3562   }
3563 }
3564 
CrossLinkEnumValue(EnumValueDescriptor * enum_value,const EnumValueDescriptorProto & proto)3565 void DescriptorBuilder::CrossLinkEnumValue(
3566     EnumValueDescriptor* enum_value, const EnumValueDescriptorProto& proto) {
3567   if (enum_value->options_ == NULL) {
3568     enum_value->options_ = &EnumValueOptions::default_instance();
3569   }
3570 }
3571 
CrossLinkService(ServiceDescriptor * service,const ServiceDescriptorProto & proto)3572 void DescriptorBuilder::CrossLinkService(
3573     ServiceDescriptor* service, const ServiceDescriptorProto& proto) {
3574   if (service->options_ == NULL) {
3575     service->options_ = &ServiceOptions::default_instance();
3576   }
3577 
3578   for (int i = 0; i < service->method_count(); i++) {
3579     CrossLinkMethod(&service->methods_[i], proto.method(i));
3580   }
3581 }
3582 
CrossLinkMethod(MethodDescriptor * method,const MethodDescriptorProto & proto)3583 void DescriptorBuilder::CrossLinkMethod(
3584     MethodDescriptor* method, const MethodDescriptorProto& proto) {
3585   if (method->options_ == NULL) {
3586     method->options_ = &MethodOptions::default_instance();
3587   }
3588 
3589   Symbol input_type = LookupSymbol(proto.input_type(), method->full_name());
3590   if (input_type.IsNull()) {
3591     AddNotDefinedError(method->full_name(), proto,
3592                        DescriptorPool::ErrorCollector::INPUT_TYPE,
3593                        proto.input_type());
3594   } else if (input_type.type != Symbol::MESSAGE) {
3595     AddError(method->full_name(), proto,
3596              DescriptorPool::ErrorCollector::INPUT_TYPE,
3597              "\"" + proto.input_type() + "\" is not a message type.");
3598   } else {
3599     method->input_type_ = input_type.descriptor;
3600   }
3601 
3602   Symbol output_type = LookupSymbol(proto.output_type(), method->full_name());
3603   if (output_type.IsNull()) {
3604     AddNotDefinedError(method->full_name(), proto,
3605                        DescriptorPool::ErrorCollector::OUTPUT_TYPE,
3606                        proto.output_type());
3607   } else if (output_type.type != Symbol::MESSAGE) {
3608     AddError(method->full_name(), proto,
3609              DescriptorPool::ErrorCollector::OUTPUT_TYPE,
3610              "\"" + proto.output_type() + "\" is not a message type.");
3611   } else {
3612     method->output_type_ = output_type.descriptor;
3613   }
3614 }
3615 
3616 // -------------------------------------------------------------------
3617 
3618 #define VALIDATE_OPTIONS_FROM_ARRAY(descriptor, array_name, type)  \
3619   for (int i = 0; i < descriptor->array_name##_count(); ++i) {     \
3620     Validate##type##Options(descriptor->array_name##s_ + i,        \
3621                             proto.array_name(i));                  \
3622   }
3623 
3624 // Determine if the file uses optimize_for = LITE_RUNTIME, being careful to
3625 // avoid problems that exist at init time.
IsLite(const FileDescriptor * file)3626 static bool IsLite(const FileDescriptor* file) {
3627   // TODO(kenton):  I don't even remember how many of these conditions are
3628   //   actually possible.  I'm just being super-safe.
3629   return file != NULL &&
3630          &file->options() != NULL &&
3631          &file->options() != &FileOptions::default_instance() &&
3632          file->options().optimize_for() == FileOptions::LITE_RUNTIME;
3633 }
3634 
ValidateFileOptions(FileDescriptor * file,const FileDescriptorProto & proto)3635 void DescriptorBuilder::ValidateFileOptions(FileDescriptor* file,
3636                                             const FileDescriptorProto& proto) {
3637   VALIDATE_OPTIONS_FROM_ARRAY(file, message_type, Message);
3638   VALIDATE_OPTIONS_FROM_ARRAY(file, enum_type, Enum);
3639   VALIDATE_OPTIONS_FROM_ARRAY(file, service, Service);
3640   VALIDATE_OPTIONS_FROM_ARRAY(file, extension, Field);
3641 
3642   // Lite files can only be imported by other Lite files.
3643   if (!IsLite(file)) {
3644     for (int i = 0; i < file->dependency_count(); i++) {
3645       if (IsLite(file->dependency(i))) {
3646         AddError(
3647           file->name(), proto,
3648           DescriptorPool::ErrorCollector::OTHER,
3649           "Files that do not use optimize_for = LITE_RUNTIME cannot import "
3650           "files which do use this option.  This file is not lite, but it "
3651           "imports \"" + file->dependency(i)->name() + "\" which is.");
3652         break;
3653       }
3654     }
3655   }
3656 }
3657 
ValidateMessageOptions(Descriptor * message,const DescriptorProto & proto)3658 void DescriptorBuilder::ValidateMessageOptions(Descriptor* message,
3659                                                const DescriptorProto& proto) {
3660   VALIDATE_OPTIONS_FROM_ARRAY(message, field, Field);
3661   VALIDATE_OPTIONS_FROM_ARRAY(message, nested_type, Message);
3662   VALIDATE_OPTIONS_FROM_ARRAY(message, enum_type, Enum);
3663   VALIDATE_OPTIONS_FROM_ARRAY(message, extension, Field);
3664 }
3665 
ValidateFieldOptions(FieldDescriptor * field,const FieldDescriptorProto & proto)3666 void DescriptorBuilder::ValidateFieldOptions(FieldDescriptor* field,
3667     const FieldDescriptorProto& proto) {
3668   if (field->options().has_experimental_map_key()) {
3669     ValidateMapKey(field, proto);
3670   }
3671 
3672   // Only repeated primitive fields may be packed.
3673   if (field->options().packed() && !field->is_packable()) {
3674     AddError(
3675       field->full_name(), proto,
3676       DescriptorPool::ErrorCollector::TYPE,
3677       "[packed = true] can only be specified for repeated primitive fields.");
3678   }
3679 
3680   // Note:  Default instance may not yet be initialized here, so we have to
3681   //   avoid reading from it.
3682   if (field->containing_type_ != NULL &&
3683       &field->containing_type()->options() !=
3684       &MessageOptions::default_instance() &&
3685       field->containing_type()->options().message_set_wire_format()) {
3686     if (field->is_extension()) {
3687       if (!field->is_optional() ||
3688           field->type() != FieldDescriptor::TYPE_MESSAGE) {
3689         AddError(field->full_name(), proto,
3690                  DescriptorPool::ErrorCollector::TYPE,
3691                  "Extensions of MessageSets must be optional messages.");
3692       }
3693     } else {
3694       AddError(field->full_name(), proto,
3695                DescriptorPool::ErrorCollector::NAME,
3696                "MessageSets cannot have fields, only extensions.");
3697     }
3698   }
3699 
3700   // Lite extensions can only be of Lite types.
3701   if (IsLite(field->file()) &&
3702       field->containing_type_ != NULL &&
3703       !IsLite(field->containing_type()->file())) {
3704     AddError(field->full_name(), proto,
3705              DescriptorPool::ErrorCollector::EXTENDEE,
3706              "Extensions to non-lite types can only be declared in non-lite "
3707              "files.  Note that you cannot extend a non-lite type to contain "
3708              "a lite type, but the reverse is allowed.");
3709   }
3710 }
3711 
ValidateEnumOptions(EnumDescriptor * enm,const EnumDescriptorProto & proto)3712 void DescriptorBuilder::ValidateEnumOptions(EnumDescriptor* enm,
3713                                             const EnumDescriptorProto& proto) {
3714   VALIDATE_OPTIONS_FROM_ARRAY(enm, value, EnumValue);
3715 }
3716 
ValidateEnumValueOptions(EnumValueDescriptor * enum_value,const EnumValueDescriptorProto & proto)3717 void DescriptorBuilder::ValidateEnumValueOptions(
3718     EnumValueDescriptor* enum_value, const EnumValueDescriptorProto& proto) {
3719   // Nothing to do so far.
3720 }
ValidateServiceOptions(ServiceDescriptor * service,const ServiceDescriptorProto & proto)3721 void DescriptorBuilder::ValidateServiceOptions(ServiceDescriptor* service,
3722     const ServiceDescriptorProto& proto) {
3723   if (IsLite(service->file())) {
3724     AddError(service->full_name(), proto,
3725              DescriptorPool::ErrorCollector::NAME,
3726              "Files with optimize_for = LITE_RUNTIME cannot define services.");
3727   }
3728 
3729   VALIDATE_OPTIONS_FROM_ARRAY(service, method, Method);
3730 }
3731 
ValidateMethodOptions(MethodDescriptor * method,const MethodDescriptorProto & proto)3732 void DescriptorBuilder::ValidateMethodOptions(MethodDescriptor* method,
3733     const MethodDescriptorProto& proto) {
3734   // Nothing to do so far.
3735 }
3736 
ValidateMapKey(FieldDescriptor * field,const FieldDescriptorProto & proto)3737 void DescriptorBuilder::ValidateMapKey(FieldDescriptor* field,
3738                                        const FieldDescriptorProto& proto) {
3739   if (!field->is_repeated()) {
3740     AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3741              "map type is only allowed for repeated fields.");
3742     return;
3743   }
3744 
3745   if (field->cpp_type() != FieldDescriptor::CPPTYPE_MESSAGE) {
3746     AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3747              "map type is only allowed for fields with a message type.");
3748     return;
3749   }
3750 
3751   const Descriptor* item_type = field->message_type();
3752   if (item_type == NULL) {
3753     AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3754              "Could not find field type.");
3755     return;
3756   }
3757 
3758   // Find the field in item_type named by "experimental_map_key"
3759   const string& key_name = field->options().experimental_map_key();
3760   const Symbol key_symbol = LookupSymbol(
3761       key_name,
3762       // We append ".key_name" to the containing type's name since
3763       // LookupSymbol() searches for peers of the supplied name, not
3764       // children of the supplied name.
3765       item_type->full_name() + "." + key_name);
3766 
3767   if (key_symbol.IsNull() || key_symbol.field_descriptor->is_extension()) {
3768     AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3769              "Could not find field named \"" + key_name + "\" in type \"" +
3770              item_type->full_name() + "\".");
3771     return;
3772   }
3773   const FieldDescriptor* key_field = key_symbol.field_descriptor;
3774 
3775   if (key_field->is_repeated()) {
3776     AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3777              "map_key must not name a repeated field.");
3778     return;
3779   }
3780 
3781   if (key_field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
3782     AddError(field->full_name(), proto, DescriptorPool::ErrorCollector::TYPE,
3783              "map key must name a scalar or string field.");
3784     return;
3785   }
3786 
3787   field->experimental_map_key_ = key_field;
3788 }
3789 
3790 #undef VALIDATE_OPTIONS_FROM_ARRAY
3791 
3792 // -------------------------------------------------------------------
3793 
OptionInterpreter(DescriptorBuilder * builder)3794 DescriptorBuilder::OptionInterpreter::OptionInterpreter(
3795     DescriptorBuilder* builder) : builder_(builder) {
3796   GOOGLE_CHECK(builder_);
3797 }
3798 
~OptionInterpreter()3799 DescriptorBuilder::OptionInterpreter::~OptionInterpreter() {
3800 }
3801 
InterpretOptions(OptionsToInterpret * options_to_interpret)3802 bool DescriptorBuilder::OptionInterpreter::InterpretOptions(
3803     OptionsToInterpret* options_to_interpret) {
3804   // Note that these may be in different pools, so we can't use the same
3805   // descriptor and reflection objects on both.
3806   Message* options = options_to_interpret->options;
3807   const Message* original_options = options_to_interpret->original_options;
3808 
3809   bool failed = false;
3810   options_to_interpret_ = options_to_interpret;
3811 
3812   // Find the uninterpreted_option field in the mutable copy of the options
3813   // and clear them, since we're about to interpret them.
3814   const FieldDescriptor* uninterpreted_options_field =
3815       options->GetDescriptor()->FindFieldByName("uninterpreted_option");
3816   GOOGLE_CHECK(uninterpreted_options_field != NULL)
3817       << "No field named \"uninterpreted_option\" in the Options proto.";
3818   options->GetReflection()->ClearField(options, uninterpreted_options_field);
3819 
3820   // Find the uninterpreted_option field in the original options.
3821   const FieldDescriptor* original_uninterpreted_options_field =
3822       original_options->GetDescriptor()->
3823           FindFieldByName("uninterpreted_option");
3824   GOOGLE_CHECK(original_uninterpreted_options_field != NULL)
3825       << "No field named \"uninterpreted_option\" in the Options proto.";
3826 
3827   const int num_uninterpreted_options = original_options->GetReflection()->
3828       FieldSize(*original_options, original_uninterpreted_options_field);
3829   for (int i = 0; i < num_uninterpreted_options; ++i) {
3830     uninterpreted_option_ = down_cast<const UninterpretedOption*>(
3831         &original_options->GetReflection()->GetRepeatedMessage(
3832             *original_options, original_uninterpreted_options_field, i));
3833     if (!InterpretSingleOption(options)) {
3834       // Error already added by InterpretSingleOption().
3835       failed = true;
3836       break;
3837     }
3838   }
3839   // Reset these, so we don't have any dangling pointers.
3840   uninterpreted_option_ = NULL;
3841   options_to_interpret_ = NULL;
3842 
3843   if (!failed) {
3844     // InterpretSingleOption() added the interpreted options in the
3845     // UnknownFieldSet, in case the option isn't yet known to us.  Now we
3846     // serialize the options message and deserialize it back.  That way, any
3847     // option fields that we do happen to know about will get moved from the
3848     // UnknownFieldSet into the real fields, and thus be available right away.
3849     // If they are not known, that's OK too. They will get reparsed into the
3850     // UnknownFieldSet and wait there until the message is parsed by something
3851     // that does know about the options.
3852     string buf;
3853     options->AppendToString(&buf);
3854     GOOGLE_CHECK(options->ParseFromString(buf))
3855         << "Protocol message serialized itself in invalid fashion.";
3856   }
3857 
3858   return !failed;
3859 }
3860 
InterpretSingleOption(Message * options)3861 bool DescriptorBuilder::OptionInterpreter::InterpretSingleOption(
3862     Message* options) {
3863   // First do some basic validation.
3864   if (uninterpreted_option_->name_size() == 0) {
3865     // This should never happen unless the parser has gone seriously awry or
3866     // someone has manually created the uninterpreted option badly.
3867     return AddNameError("Option must have a name.");
3868   }
3869   if (uninterpreted_option_->name(0).name_part() == "uninterpreted_option") {
3870     return AddNameError("Option must not use reserved name "
3871                         "\"uninterpreted_option\".");
3872   }
3873 
3874   const Descriptor* options_descriptor = NULL;
3875   // Get the options message's descriptor from the builder's pool, so that we
3876   // get the version that knows about any extension options declared in the
3877   // file we're currently building. The descriptor should be there as long as
3878   // the file we're building imported "google/protobuf/descriptors.proto".
3879 
3880   // Note that we use DescriptorBuilder::FindSymbol(), not
3881   // DescriptorPool::FindMessageTypeByName() because we're already holding the
3882   // pool's mutex, and the latter method locks it again.
3883   Symbol symbol = builder_->FindSymbolNotEnforcingDeps(
3884     options->GetDescriptor()->full_name());
3885   if (!symbol.IsNull() && symbol.type == Symbol::MESSAGE) {
3886     options_descriptor = symbol.descriptor;
3887   } else {
3888     // The options message's descriptor was not in the builder's pool, so use
3889     // the standard version from the generated pool. We're not holding the
3890     // generated pool's mutex, so we can search it the straightforward way.
3891     options_descriptor = options->GetDescriptor();
3892   }
3893   GOOGLE_CHECK(options_descriptor);
3894 
3895   // We iterate over the name parts to drill into the submessages until we find
3896   // the leaf field for the option. As we drill down we remember the current
3897   // submessage's descriptor in |descriptor| and the next field in that
3898   // submessage in |field|. We also track the fields we're drilling down
3899   // through in |intermediate_fields|. As we go, we reconstruct the full option
3900   // name in |debug_msg_name|, for use in error messages.
3901   const Descriptor* descriptor = options_descriptor;
3902   const FieldDescriptor* field = NULL;
3903   vector<const FieldDescriptor*> intermediate_fields;
3904   string debug_msg_name = "";
3905 
3906   for (int i = 0; i < uninterpreted_option_->name_size(); ++i) {
3907     const string& name_part = uninterpreted_option_->name(i).name_part();
3908     if (debug_msg_name.size() > 0) {
3909       debug_msg_name += ".";
3910     }
3911     if (uninterpreted_option_->name(i).is_extension()) {
3912       debug_msg_name += "(" + name_part + ")";
3913       // Search for the extension's descriptor as an extension in the builder's
3914       // pool. Note that we use DescriptorBuilder::LookupSymbol(), not
3915       // DescriptorPool::FindExtensionByName(), for two reasons: 1) It allows
3916       // relative lookups, and 2) because we're already holding the pool's
3917       // mutex, and the latter method locks it again.
3918       Symbol symbol = builder_->LookupSymbol(name_part,
3919                                              options_to_interpret_->name_scope);
3920       if (!symbol.IsNull() && symbol.type == Symbol::FIELD) {
3921         field = symbol.field_descriptor;
3922       }
3923       // If we don't find the field then the field's descriptor was not in the
3924       // builder's pool, but there's no point in looking in the generated
3925       // pool. We require that you import the file that defines any extensions
3926       // you use, so they must be present in the builder's pool.
3927     } else {
3928       debug_msg_name += name_part;
3929       // Search for the field's descriptor as a regular field.
3930       field = descriptor->FindFieldByName(name_part);
3931     }
3932 
3933     if (field == NULL) {
3934       if (get_allow_unknown(builder_->pool_)) {
3935         // We can't find the option, but AllowUnknownDependencies() is enabled,
3936         // so we will just leave it as uninterpreted.
3937         AddWithoutInterpreting(*uninterpreted_option_, options);
3938         return true;
3939       } else {
3940         return AddNameError("Option \"" + debug_msg_name + "\" unknown.");
3941       }
3942     } else if (field->containing_type() != descriptor) {
3943       if (get_is_placeholder(field->containing_type())) {
3944         // The field is an extension of a placeholder type, so we can't
3945         // reliably verify whether it is a valid extension to use here (e.g.
3946         // we don't know if it is an extension of the correct *Options message,
3947         // or if it has a valid field number, etc.).  Just leave it as
3948         // uninterpreted instead.
3949         AddWithoutInterpreting(*uninterpreted_option_, options);
3950         return true;
3951       } else {
3952         // This can only happen if, due to some insane misconfiguration of the
3953         // pools, we find the options message in one pool but the field in
3954         // another. This would probably imply a hefty bug somewhere.
3955         return AddNameError("Option field \"" + debug_msg_name +
3956                             "\" is not a field or extension of message \"" +
3957                             descriptor->name() + "\".");
3958       }
3959     } else if (field->is_repeated()) {
3960       return AddNameError("Option field \"" + debug_msg_name +
3961                           "\" is repeated. Repeated options are not "
3962                           "supported.");
3963     } else if (i < uninterpreted_option_->name_size() - 1) {
3964       if (field->cpp_type() != FieldDescriptor::CPPTYPE_MESSAGE) {
3965         return AddNameError("Option \"" +  debug_msg_name +
3966                             "\" is an atomic type, not a message.");
3967       } else {
3968         // Drill down into the submessage.
3969         intermediate_fields.push_back(field);
3970         descriptor = field->message_type();
3971       }
3972     } else if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) {
3973       return AddNameError("Option field \"" + debug_msg_name +
3974                           "\" cannot be of message type.");
3975     }
3976   }
3977 
3978   // We've found the leaf field. Now we use UnknownFieldSets to set its value
3979   // on the options message. We do so because the message may not yet know
3980   // about its extension fields, so we may not be able to set the fields
3981   // directly. But the UnknownFieldSets will serialize to the same wire-format
3982   // message, so reading that message back in once the extension fields are
3983   // known will populate them correctly.
3984 
3985   // First see if the option is already set.
3986   if (!ExamineIfOptionIsSet(
3987           intermediate_fields.begin(),
3988           intermediate_fields.end(),
3989           field, debug_msg_name,
3990           options->GetReflection()->GetUnknownFields(*options))) {
3991     return false;  // ExamineIfOptionIsSet() already added the error.
3992   }
3993 
3994 
3995   // First set the value on the UnknownFieldSet corresponding to the
3996   // innermost message.
3997   scoped_ptr<UnknownFieldSet> unknown_fields(new UnknownFieldSet());
3998   if (!SetOptionValue(field, unknown_fields.get())) {
3999     return false;  // SetOptionValue() already added the error.
4000   }
4001 
4002   // Now wrap the UnknownFieldSet with UnknownFieldSets corresponding to all
4003   // the intermediate messages.
4004   for (vector<const FieldDescriptor*>::reverse_iterator iter =
4005            intermediate_fields.rbegin();
4006        iter != intermediate_fields.rend(); ++iter) {
4007     scoped_ptr<UnknownFieldSet> parent_unknown_fields(new UnknownFieldSet());
4008     switch ((*iter)->type()) {
4009       case FieldDescriptor::TYPE_MESSAGE: {
4010         io::StringOutputStream outstr(
4011             parent_unknown_fields->AddLengthDelimited((*iter)->number()));
4012         io::CodedOutputStream out(&outstr);
4013         internal::WireFormat::SerializeUnknownFields(*unknown_fields, &out);
4014         GOOGLE_CHECK(!out.HadError())
4015             << "Unexpected failure while serializing option submessage "
4016             << debug_msg_name << "\".";
4017         break;
4018       }
4019 
4020       case FieldDescriptor::TYPE_GROUP: {
4021          parent_unknown_fields->AddGroup((*iter)->number())
4022                               ->MergeFrom(*unknown_fields);
4023         break;
4024       }
4025 
4026       default:
4027         GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_MESSAGE: "
4028                    << (*iter)->type();
4029         return false;
4030     }
4031     unknown_fields.reset(parent_unknown_fields.release());
4032   }
4033 
4034   // Now merge the UnknownFieldSet corresponding to the top-level message into
4035   // the options message.
4036   options->GetReflection()->MutableUnknownFields(options)->MergeFrom(
4037       *unknown_fields);
4038 
4039   return true;
4040 }
4041 
AddWithoutInterpreting(const UninterpretedOption & uninterpreted_option,Message * options)4042 void DescriptorBuilder::OptionInterpreter::AddWithoutInterpreting(
4043     const UninterpretedOption& uninterpreted_option, Message* options) {
4044   const FieldDescriptor* field =
4045     options->GetDescriptor()->FindFieldByName("uninterpreted_option");
4046   GOOGLE_CHECK(field != NULL);
4047 
4048   options->GetReflection()->AddMessage(options, field)
4049     ->CopyFrom(uninterpreted_option);
4050 }
4051 
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)4052 bool DescriptorBuilder::OptionInterpreter::ExamineIfOptionIsSet(
4053     vector<const FieldDescriptor*>::const_iterator intermediate_fields_iter,
4054     vector<const FieldDescriptor*>::const_iterator intermediate_fields_end,
4055     const FieldDescriptor* innermost_field, const string& debug_msg_name,
4056     const UnknownFieldSet& unknown_fields) {
4057   // We do linear searches of the UnknownFieldSet and its sub-groups.  This
4058   // should be fine since it's unlikely that any one options structure will
4059   // contain more than a handful of options.
4060 
4061   if (intermediate_fields_iter == intermediate_fields_end) {
4062     // We're at the innermost submessage.
4063     for (int i = 0; i < unknown_fields.field_count(); i++) {
4064       if (unknown_fields.field(i).number() == innermost_field->number()) {
4065         return AddNameError("Option \"" + debug_msg_name +
4066                             "\" was already set.");
4067       }
4068     }
4069     return true;
4070   }
4071 
4072   for (int i = 0; i < unknown_fields.field_count(); i++) {
4073     if (unknown_fields.field(i).number() ==
4074         (*intermediate_fields_iter)->number()) {
4075       const UnknownField* unknown_field = &unknown_fields.field(i);
4076       FieldDescriptor::Type type = (*intermediate_fields_iter)->type();
4077       // Recurse into the next submessage.
4078       switch (type) {
4079         case FieldDescriptor::TYPE_MESSAGE:
4080           if (unknown_field->type() == UnknownField::TYPE_LENGTH_DELIMITED) {
4081             UnknownFieldSet intermediate_unknown_fields;
4082             if (intermediate_unknown_fields.ParseFromString(
4083                     unknown_field->length_delimited()) &&
4084                 !ExamineIfOptionIsSet(intermediate_fields_iter + 1,
4085                                       intermediate_fields_end,
4086                                       innermost_field, debug_msg_name,
4087                                       intermediate_unknown_fields)) {
4088               return false;  // Error already added.
4089             }
4090           }
4091           break;
4092 
4093         case FieldDescriptor::TYPE_GROUP:
4094           if (unknown_field->type() == UnknownField::TYPE_GROUP) {
4095             if (!ExamineIfOptionIsSet(intermediate_fields_iter + 1,
4096                                       intermediate_fields_end,
4097                                       innermost_field, debug_msg_name,
4098                                       unknown_field->group())) {
4099               return false;  // Error already added.
4100             }
4101           }
4102           break;
4103 
4104         default:
4105           GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_MESSAGE: " << type;
4106           return false;
4107       }
4108     }
4109   }
4110   return true;
4111 }
4112 
SetOptionValue(const FieldDescriptor * option_field,UnknownFieldSet * unknown_fields)4113 bool DescriptorBuilder::OptionInterpreter::SetOptionValue(
4114     const FieldDescriptor* option_field,
4115     UnknownFieldSet* unknown_fields) {
4116   // We switch on the CppType to validate.
4117   switch (option_field->cpp_type()) {
4118 
4119     case FieldDescriptor::CPPTYPE_INT32:
4120       if (uninterpreted_option_->has_positive_int_value()) {
4121         if (uninterpreted_option_->positive_int_value() >
4122             static_cast<uint64>(kint32max)) {
4123           return AddValueError("Value out of range for int32 option \"" +
4124                                option_field->full_name() + "\".");
4125         } else {
4126           SetInt32(option_field->number(),
4127                    uninterpreted_option_->positive_int_value(),
4128                    option_field->type(), unknown_fields);
4129         }
4130       } else if (uninterpreted_option_->has_negative_int_value()) {
4131         if (uninterpreted_option_->negative_int_value() <
4132             static_cast<int64>(kint32min)) {
4133           return AddValueError("Value out of range for int32 option \"" +
4134                                option_field->full_name() + "\".");
4135         } else {
4136           SetInt32(option_field->number(),
4137                    uninterpreted_option_->negative_int_value(),
4138                    option_field->type(), unknown_fields);
4139         }
4140       } else {
4141         return AddValueError("Value must be integer for int32 option \"" +
4142                              option_field->full_name() + "\".");
4143       }
4144       break;
4145 
4146     case FieldDescriptor::CPPTYPE_INT64:
4147       if (uninterpreted_option_->has_positive_int_value()) {
4148         if (uninterpreted_option_->positive_int_value() >
4149             static_cast<uint64>(kint64max)) {
4150           return AddValueError("Value out of range for int64 option \"" +
4151                                option_field->full_name() + "\".");
4152         } else {
4153           SetInt64(option_field->number(),
4154                    uninterpreted_option_->positive_int_value(),
4155                    option_field->type(), unknown_fields);
4156         }
4157       } else if (uninterpreted_option_->has_negative_int_value()) {
4158         SetInt64(option_field->number(),
4159                  uninterpreted_option_->negative_int_value(),
4160                  option_field->type(), unknown_fields);
4161       } else {
4162         return AddValueError("Value must be integer for int64 option \"" +
4163                              option_field->full_name() + "\".");
4164       }
4165       break;
4166 
4167     case FieldDescriptor::CPPTYPE_UINT32:
4168       if (uninterpreted_option_->has_positive_int_value()) {
4169         if (uninterpreted_option_->positive_int_value() > kuint32max) {
4170           return AddValueError("Value out of range for uint32 option \"" +
4171                                option_field->name() + "\".");
4172         } else {
4173           SetUInt32(option_field->number(),
4174                     uninterpreted_option_->positive_int_value(),
4175                     option_field->type(), unknown_fields);
4176         }
4177       } else {
4178         return AddValueError("Value must be non-negative integer for uint32 "
4179                              "option \"" + option_field->full_name() + "\".");
4180       }
4181       break;
4182 
4183     case FieldDescriptor::CPPTYPE_UINT64:
4184       if (uninterpreted_option_->has_positive_int_value()) {
4185         SetUInt64(option_field->number(),
4186                   uninterpreted_option_->positive_int_value(),
4187                   option_field->type(), unknown_fields);
4188       } else {
4189         return AddValueError("Value must be non-negative integer for uint64 "
4190                              "option \"" + option_field->full_name() + "\".");
4191       }
4192       break;
4193 
4194     case FieldDescriptor::CPPTYPE_FLOAT: {
4195       float value;
4196       if (uninterpreted_option_->has_double_value()) {
4197         value = uninterpreted_option_->double_value();
4198       } else if (uninterpreted_option_->has_positive_int_value()) {
4199         value = uninterpreted_option_->positive_int_value();
4200       } else if (uninterpreted_option_->has_negative_int_value()) {
4201         value = uninterpreted_option_->negative_int_value();
4202       } else {
4203         return AddValueError("Value must be number for float option \"" +
4204                              option_field->full_name() + "\".");
4205       }
4206       unknown_fields->AddFixed32(option_field->number(),
4207           google::protobuf::internal::WireFormatLite::EncodeFloat(value));
4208       break;
4209     }
4210 
4211     case FieldDescriptor::CPPTYPE_DOUBLE: {
4212       double value;
4213       if (uninterpreted_option_->has_double_value()) {
4214         value = uninterpreted_option_->double_value();
4215       } else if (uninterpreted_option_->has_positive_int_value()) {
4216         value = uninterpreted_option_->positive_int_value();
4217       } else if (uninterpreted_option_->has_negative_int_value()) {
4218         value = uninterpreted_option_->negative_int_value();
4219       } else {
4220         return AddValueError("Value must be number for double option \"" +
4221                              option_field->full_name() + "\".");
4222       }
4223       unknown_fields->AddFixed64(option_field->number(),
4224           google::protobuf::internal::WireFormatLite::EncodeDouble(value));
4225       break;
4226     }
4227 
4228     case FieldDescriptor::CPPTYPE_BOOL:
4229       uint64 value;
4230       if (!uninterpreted_option_->has_identifier_value()) {
4231         return AddValueError("Value must be identifier for boolean option "
4232                              "\"" + option_field->full_name() + "\".");
4233       }
4234       if (uninterpreted_option_->identifier_value() == "true") {
4235         value = 1;
4236       } else if (uninterpreted_option_->identifier_value() == "false") {
4237         value = 0;
4238       } else {
4239         return AddValueError("Value must be \"true\" or \"false\" for boolean "
4240                              "option \"" + option_field->full_name() + "\".");
4241       }
4242       unknown_fields->AddVarint(option_field->number(), value);
4243       break;
4244 
4245     case FieldDescriptor::CPPTYPE_ENUM: {
4246       if (!uninterpreted_option_->has_identifier_value()) {
4247         return AddValueError("Value must be identifier for enum-valued option "
4248                              "\"" + option_field->full_name() + "\".");
4249       }
4250       const EnumDescriptor* enum_type = option_field->enum_type();
4251       const string& value_name = uninterpreted_option_->identifier_value();
4252       const EnumValueDescriptor* enum_value = NULL;
4253 
4254       if (enum_type->file()->pool() != DescriptorPool::generated_pool()) {
4255         // Note that the enum value's fully-qualified name is a sibling of the
4256         // enum's name, not a child of it.
4257         string fully_qualified_name = enum_type->full_name();
4258         fully_qualified_name.resize(fully_qualified_name.size() -
4259                                     enum_type->name().size());
4260         fully_qualified_name += value_name;
4261 
4262         // Search for the enum value's descriptor in the builder's pool. Note
4263         // that we use DescriptorBuilder::FindSymbolNotEnforcingDeps(), not
4264         // DescriptorPool::FindEnumValueByName() because we're already holding
4265         // the pool's mutex, and the latter method locks it again.
4266         Symbol symbol =
4267           builder_->FindSymbolNotEnforcingDeps(fully_qualified_name);
4268         if (!symbol.IsNull() && symbol.type == Symbol::ENUM_VALUE) {
4269           if (symbol.enum_value_descriptor->type() != enum_type) {
4270             return AddValueError("Enum type \"" + enum_type->full_name() +
4271                 "\" has no value named \"" + value_name + "\" for option \"" +
4272                 option_field->full_name() +
4273                 "\". This appears to be a value from a sibling type.");
4274           } else {
4275             enum_value = symbol.enum_value_descriptor;
4276           }
4277         }
4278       } else {
4279         // The enum type is in the generated pool, so we can search for the
4280         // value there.
4281         enum_value = enum_type->FindValueByName(value_name);
4282       }
4283 
4284       if (enum_value == NULL) {
4285         return AddValueError("Enum type \"" +
4286                              option_field->enum_type()->full_name() +
4287                              "\" has no value named \"" + value_name + "\" for "
4288                              "option \"" + option_field->full_name() + "\".");
4289       } else {
4290         // Sign-extension is not a problem, since we cast directly from int32 to
4291         // uint64, without first going through uint32.
4292         unknown_fields->AddVarint(option_field->number(),
4293           static_cast<uint64>(static_cast<int64>(enum_value->number())));
4294       }
4295       break;
4296     }
4297 
4298     case FieldDescriptor::CPPTYPE_STRING:
4299       if (!uninterpreted_option_->has_string_value()) {
4300         return AddValueError("Value must be quoted string for string option "
4301                              "\"" + option_field->full_name() + "\".");
4302       }
4303       // The string has already been unquoted and unescaped by the parser.
4304       unknown_fields->AddLengthDelimited(option_field->number(),
4305           uninterpreted_option_->string_value());
4306       break;
4307 
4308     case FieldDescriptor::CPPTYPE_MESSAGE:
4309       // We don't currently support defining a message-typed option, so we
4310       // should never actually get here.
4311       return AddValueError("Option \"" + option_field->full_name() +
4312                            "\" is a message.  To set fields within it, use "
4313                            "syntax like \"" + option_field->name() +
4314                            ".foo = value\".");
4315       break;
4316   }
4317 
4318   return true;
4319 }
4320 
SetInt32(int number,int32 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)4321 void DescriptorBuilder::OptionInterpreter::SetInt32(int number, int32 value,
4322     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4323   switch (type) {
4324     case FieldDescriptor::TYPE_INT32:
4325       unknown_fields->AddVarint(number,
4326         static_cast<uint64>(static_cast<int64>(value)));
4327       break;
4328 
4329     case FieldDescriptor::TYPE_SFIXED32:
4330       unknown_fields->AddFixed32(number, static_cast<uint32>(value));
4331       break;
4332 
4333     case FieldDescriptor::TYPE_SINT32:
4334       unknown_fields->AddVarint(number,
4335           google::protobuf::internal::WireFormatLite::ZigZagEncode32(value));
4336       break;
4337 
4338     default:
4339       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_INT32: " << type;
4340       break;
4341   }
4342 }
4343 
SetInt64(int number,int64 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)4344 void DescriptorBuilder::OptionInterpreter::SetInt64(int number, int64 value,
4345     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4346   switch (type) {
4347     case FieldDescriptor::TYPE_INT64:
4348       unknown_fields->AddVarint(number, static_cast<uint64>(value));
4349       break;
4350 
4351     case FieldDescriptor::TYPE_SFIXED64:
4352       unknown_fields->AddFixed64(number, static_cast<uint64>(value));
4353       break;
4354 
4355     case FieldDescriptor::TYPE_SINT64:
4356       unknown_fields->AddVarint(number,
4357           google::protobuf::internal::WireFormatLite::ZigZagEncode64(value));
4358       break;
4359 
4360     default:
4361       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_INT64: " << type;
4362       break;
4363   }
4364 }
4365 
SetUInt32(int number,uint32 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)4366 void DescriptorBuilder::OptionInterpreter::SetUInt32(int number, uint32 value,
4367     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4368   switch (type) {
4369     case FieldDescriptor::TYPE_UINT32:
4370       unknown_fields->AddVarint(number, static_cast<uint64>(value));
4371       break;
4372 
4373     case FieldDescriptor::TYPE_FIXED32:
4374       unknown_fields->AddFixed32(number, static_cast<uint32>(value));
4375       break;
4376 
4377     default:
4378       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_UINT32: " << type;
4379       break;
4380   }
4381 }
4382 
SetUInt64(int number,uint64 value,FieldDescriptor::Type type,UnknownFieldSet * unknown_fields)4383 void DescriptorBuilder::OptionInterpreter::SetUInt64(int number, uint64 value,
4384     FieldDescriptor::Type type, UnknownFieldSet* unknown_fields) {
4385   switch (type) {
4386     case FieldDescriptor::TYPE_UINT64:
4387       unknown_fields->AddVarint(number, value);
4388       break;
4389 
4390     case FieldDescriptor::TYPE_FIXED64:
4391       unknown_fields->AddFixed64(number, value);
4392       break;
4393 
4394     default:
4395       GOOGLE_LOG(FATAL) << "Invalid wire type for CPPTYPE_UINT64: " << type;
4396       break;
4397   }
4398 }
4399 
4400 }  // namespace protobuf
4401 }  // namespace google
4402