• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===--- GlobalModuleIndex.cpp - Global Module Index ------------*- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements the GlobalModuleIndex class.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "ASTReaderInternals.h"
15 #include "clang/Basic/FileManager.h"
16 #include "clang/Basic/OnDiskHashTable.h"
17 #include "clang/Serialization/ASTBitCodes.h"
18 #include "clang/Serialization/GlobalModuleIndex.h"
19 #include "llvm/ADT/DenseMap.h"
20 #include "llvm/ADT/MapVector.h"
21 #include "llvm/ADT/SmallString.h"
22 #include "llvm/ADT/StringExtras.h"
23 #include "llvm/Bitcode/BitstreamReader.h"
24 #include "llvm/Bitcode/BitstreamWriter.h"
25 #include "llvm/Support/FileSystem.h"
26 #include "llvm/Support/LockFileManager.h"
27 #include "llvm/Support/MemoryBuffer.h"
28 #include "llvm/Support/PathV2.h"
29 #include <cstdio>
30 using namespace clang;
31 using namespace serialization;
32 
33 //----------------------------------------------------------------------------//
34 // Shared constants
35 //----------------------------------------------------------------------------//
36 namespace {
37   enum {
38     /// \brief The block containing the index.
39     GLOBAL_INDEX_BLOCK_ID = llvm::bitc::FIRST_APPLICATION_BLOCKID
40   };
41 
42   /// \brief Describes the record types in the index.
43   enum IndexRecordTypes {
44     /// \brief Contains version information and potentially other metadata,
45     /// used to determine if we can read this global index file.
46     INDEX_METADATA,
47     /// \brief Describes a module, including its file name and dependencies.
48     MODULE,
49     /// \brief The index for identifiers.
50     IDENTIFIER_INDEX
51   };
52 }
53 
54 /// \brief The name of the global index file.
55 static const char * const IndexFileName = "modules.idx";
56 
57 /// \brief The global index file version.
58 static const unsigned CurrentVersion = 1;
59 
60 //----------------------------------------------------------------------------//
61 // Global module index reader.
62 //----------------------------------------------------------------------------//
63 
64 namespace {
65 
66 /// \brief Trait used to read the identifier index from the on-disk hash
67 /// table.
68 class IdentifierIndexReaderTrait {
69 public:
70   typedef StringRef external_key_type;
71   typedef StringRef internal_key_type;
72   typedef SmallVector<unsigned, 2> data_type;
73 
EqualKey(const internal_key_type & a,const internal_key_type & b)74   static bool EqualKey(const internal_key_type& a, const internal_key_type& b) {
75     return a == b;
76   }
77 
ComputeHash(const internal_key_type & a)78   static unsigned ComputeHash(const internal_key_type& a) {
79     return llvm::HashString(a);
80   }
81 
82   static std::pair<unsigned, unsigned>
ReadKeyDataLength(const unsigned char * & d)83   ReadKeyDataLength(const unsigned char*& d) {
84     using namespace clang::io;
85     unsigned KeyLen = ReadUnalignedLE16(d);
86     unsigned DataLen = ReadUnalignedLE16(d);
87     return std::make_pair(KeyLen, DataLen);
88   }
89 
90   static const internal_key_type&
GetInternalKey(const external_key_type & x)91   GetInternalKey(const external_key_type& x) { return x; }
92 
93   static const external_key_type&
GetExternalKey(const internal_key_type & x)94   GetExternalKey(const internal_key_type& x) { return x; }
95 
ReadKey(const unsigned char * d,unsigned n)96   static internal_key_type ReadKey(const unsigned char* d, unsigned n) {
97     return StringRef((const char *)d, n);
98   }
99 
ReadData(const internal_key_type & k,const unsigned char * d,unsigned DataLen)100   static data_type ReadData(const internal_key_type& k,
101                             const unsigned char* d,
102                             unsigned DataLen) {
103     using namespace clang::io;
104 
105     data_type Result;
106     while (DataLen > 0) {
107       unsigned ID = ReadUnalignedLE32(d);
108       Result.push_back(ID);
109       DataLen -= 4;
110     }
111 
112     return Result;
113   }
114 };
115 
116 typedef OnDiskChainedHashTable<IdentifierIndexReaderTrait> IdentifierIndexTable;
117 
118 /// \brief Module information as it was loaded from the index file.
119 struct LoadedModuleInfo {
120   const FileEntry *File;
121   SmallVector<unsigned, 2> Dependencies;
122   SmallVector<unsigned, 2> ImportedBy;
123 };
124 
125 }
126 
GlobalModuleIndex(FileManager & FileMgr,llvm::MemoryBuffer * Buffer,llvm::BitstreamCursor Cursor)127 GlobalModuleIndex::GlobalModuleIndex(FileManager &FileMgr,
128                                      llvm::MemoryBuffer *Buffer,
129                                      llvm::BitstreamCursor Cursor)
130   : Buffer(Buffer), IdentifierIndex(),
131     NumIdentifierLookups(), NumIdentifierLookupHits()
132 {
133   typedef llvm::DenseMap<unsigned, LoadedModuleInfo> LoadedModulesMap;
134   LoadedModulesMap LoadedModules;
135 
136   // Read the global index.
137   unsigned LargestID = 0;
138   bool InGlobalIndexBlock = false;
139   bool Done = false;
140   bool AnyOutOfDate = false;
141   while (!Done) {
142     llvm::BitstreamEntry Entry = Cursor.advance();
143 
144     switch (Entry.Kind) {
145     case llvm::BitstreamEntry::Error:
146       return;
147 
148     case llvm::BitstreamEntry::EndBlock:
149       if (InGlobalIndexBlock) {
150         InGlobalIndexBlock = false;
151         Done = true;
152         continue;
153       }
154       return;
155 
156 
157     case llvm::BitstreamEntry::Record:
158       // Entries in the global index block are handled below.
159       if (InGlobalIndexBlock)
160         break;
161 
162       return;
163 
164     case llvm::BitstreamEntry::SubBlock:
165       if (!InGlobalIndexBlock && Entry.ID == GLOBAL_INDEX_BLOCK_ID) {
166         if (Cursor.EnterSubBlock(GLOBAL_INDEX_BLOCK_ID))
167           return;
168 
169         InGlobalIndexBlock = true;
170       } else if (Cursor.SkipBlock()) {
171         return;
172       }
173       continue;
174     }
175 
176     SmallVector<uint64_t, 64> Record;
177     StringRef Blob;
178     switch ((IndexRecordTypes)Cursor.readRecord(Entry.ID, Record, &Blob)) {
179     case INDEX_METADATA:
180       // Make sure that the version matches.
181       if (Record.size() < 1 || Record[0] != CurrentVersion)
182         return;
183       break;
184 
185     case MODULE: {
186       unsigned Idx = 0;
187       unsigned ID = Record[Idx++];
188       if (ID > LargestID)
189         LargestID = ID;
190 
191       off_t Size = Record[Idx++];
192       time_t ModTime = Record[Idx++];
193 
194       // File name.
195       unsigned NameLen = Record[Idx++];
196       llvm::SmallString<64> FileName(Record.begin() + Idx,
197                                      Record.begin() + Idx + NameLen);
198       Idx += NameLen;
199 
200       // Dependencies
201       unsigned NumDeps = Record[Idx++];
202       llvm::SmallVector<unsigned, 2>
203         Dependencies(Record.begin() + Idx, Record.begin() + Idx + NumDeps);
204 
205       // Find the file. If we can't find it, ignore it.
206       const FileEntry *File = FileMgr.getFile(FileName, /*openFile=*/false,
207                                               /*cacheFailure=*/false);
208       if (!File) {
209         AnyOutOfDate = true;
210         break;
211       }
212 
213       // If the module file is newer than the index, ignore it.
214       if (File->getSize() != Size || File->getModificationTime() != ModTime) {
215         AnyOutOfDate = true;
216         break;
217       }
218 
219       // Record this module. The dependencies will be resolved later.
220       LoadedModuleInfo &Info = LoadedModules[ID];
221       Info.File = File;
222       Info.Dependencies.swap(Dependencies);
223       break;
224     }
225 
226     case IDENTIFIER_INDEX:
227       // Wire up the identifier index.
228       if (Record[0]) {
229         IdentifierIndex = IdentifierIndexTable::Create(
230                             (const unsigned char *)Blob.data() + Record[0],
231                             (const unsigned char *)Blob.data(),
232                             IdentifierIndexReaderTrait());
233       }
234       break;
235     }
236   }
237 
238   // If there are any modules that have gone out-of-date, prune out any modules
239   // that depend on them.
240   if (AnyOutOfDate) {
241     // First, build back links in the module dependency graph.
242     SmallVector<unsigned, 4> Stack;
243     for (LoadedModulesMap::iterator LM = LoadedModules.begin(),
244                                     LMEnd = LoadedModules.end();
245          LM != LMEnd; ++LM) {
246       unsigned ID = LM->first;
247 
248       // If this module is out-of-date, push it onto the stack.
249       if (LM->second.File == 0)
250         Stack.push_back(ID);
251 
252       for (unsigned I = 0, N = LM->second.Dependencies.size(); I != N; ++I) {
253         unsigned DepID = LM->second.Dependencies[I];
254         LoadedModulesMap::iterator Known = LoadedModules.find(DepID);
255         if (Known == LoadedModules.end() || !Known->second.File) {
256           // The dependency was out-of-date, so mark us as out of date.
257           // This is just an optimization.
258           if (LM->second.File)
259             Stack.push_back(ID);
260 
261           LM->second.File = 0;
262           continue;
263         }
264 
265         // Record this reverse dependency.
266         Known->second.ImportedBy.push_back(ID);
267       }
268     }
269 
270     // Second, walk the back links from out-of-date modules to those modules
271     // that depend on them, making those modules out-of-date as well.
272     while (!Stack.empty()) {
273       unsigned ID = Stack.back();
274       Stack.pop_back();
275 
276       LoadedModuleInfo &Info = LoadedModules[ID];
277       for (unsigned I = 0, N = Info.ImportedBy.size(); I != N; ++I) {
278         unsigned FromID = Info.ImportedBy[I];
279         if (LoadedModules[FromID].File) {
280           LoadedModules[FromID].File = 0;
281           Stack.push_back(FromID);
282         }
283       }
284     }
285   }
286 
287   // Allocate the vector containing information about all of the modules.
288   Modules.resize(LargestID + 1);
289   for (LoadedModulesMap::iterator LM = LoadedModules.begin(),
290                                   LMEnd = LoadedModules.end();
291        LM != LMEnd; ++LM) {
292     if (!LM->second.File)
293       continue;
294 
295     Modules[LM->first].File = LM->second.File;
296 
297     // Resolve dependencies. Drop any we can't resolve due to out-of-date
298     // module files.
299     for (unsigned I = 0, N = LM->second.Dependencies.size(); I != N; ++I) {
300       unsigned DepID = LM->second.Dependencies[I];
301       LoadedModulesMap::iterator Known = LoadedModules.find(DepID);
302       if (Known == LoadedModules.end() || !Known->second.File)
303         continue;
304 
305       Modules[LM->first].Dependencies.push_back(Known->second.File);
306     }
307   }
308 }
309 
~GlobalModuleIndex()310 GlobalModuleIndex::~GlobalModuleIndex() { }
311 
312 std::pair<GlobalModuleIndex *, GlobalModuleIndex::ErrorCode>
readIndex(FileManager & FileMgr,StringRef Path)313 GlobalModuleIndex::readIndex(FileManager &FileMgr, StringRef Path) {
314   // Load the index file, if it's there.
315   llvm::SmallString<128> IndexPath;
316   IndexPath += Path;
317   llvm::sys::path::append(IndexPath, IndexFileName);
318 
319   llvm::OwningPtr<llvm::MemoryBuffer> Buffer(
320                                         FileMgr.getBufferForFile(IndexPath));
321   if (!Buffer)
322     return std::make_pair((GlobalModuleIndex *)0, EC_NotFound);
323 
324   /// \brief The bitstream reader from which we'll read the AST file.
325   llvm::BitstreamReader Reader((const unsigned char *)Buffer->getBufferStart(),
326                                (const unsigned char *)Buffer->getBufferEnd());
327 
328   /// \brief The main bitstream cursor for the main block.
329   llvm::BitstreamCursor Cursor(Reader);
330 
331   // Sniff for the signature.
332   if (Cursor.Read(8) != 'B' ||
333       Cursor.Read(8) != 'C' ||
334       Cursor.Read(8) != 'G' ||
335       Cursor.Read(8) != 'I') {
336     return std::make_pair((GlobalModuleIndex *)0, EC_IOError);
337   }
338 
339   return std::make_pair(new GlobalModuleIndex(FileMgr, Buffer.take(), Cursor),
340                         EC_None);
341 }
342 
getKnownModules(SmallVectorImpl<const FileEntry * > & ModuleFiles)343 void GlobalModuleIndex::getKnownModules(
344        SmallVectorImpl<const FileEntry *> &ModuleFiles) {
345   ModuleFiles.clear();
346   for (unsigned I = 0, N = Modules.size(); I != N; ++I) {
347     if (Modules[I].File)
348       ModuleFiles.push_back(Modules[I].File);
349   }
350 }
351 
getModuleDependencies(const clang::FileEntry * ModuleFile,SmallVectorImpl<const clang::FileEntry * > & Dependencies)352 void GlobalModuleIndex::getModuleDependencies(
353        const clang::FileEntry *ModuleFile,
354        SmallVectorImpl<const clang::FileEntry *> &Dependencies) {
355   // If the file -> index mapping is empty, populate it now.
356   if (ModulesByFile.empty()) {
357     for (unsigned I = 0, N = Modules.size(); I != N; ++I) {
358       if (Modules[I].File)
359         ModulesByFile[Modules[I].File] = I;
360     }
361   }
362 
363   // Look for information about this module file.
364   llvm::DenseMap<const FileEntry *, unsigned>::iterator Known
365     = ModulesByFile.find(ModuleFile);
366   if (Known == ModulesByFile.end())
367     return;
368 
369   // Record dependencies.
370   Dependencies = Modules[Known->second].Dependencies;
371 }
372 
lookupIdentifier(StringRef Name,HitSet & Hits)373 bool GlobalModuleIndex::lookupIdentifier(StringRef Name, HitSet &Hits) {
374   Hits.clear();
375 
376   // If there's no identifier index, there is nothing we can do.
377   if (!IdentifierIndex)
378     return false;
379 
380   // Look into the identifier index.
381   ++NumIdentifierLookups;
382   IdentifierIndexTable &Table
383     = *static_cast<IdentifierIndexTable *>(IdentifierIndex);
384   IdentifierIndexTable::iterator Known = Table.find(Name);
385   if (Known == Table.end()) {
386     return true;
387   }
388 
389   SmallVector<unsigned, 2> ModuleIDs = *Known;
390   for (unsigned I = 0, N = ModuleIDs.size(); I != N; ++I) {
391     unsigned ID = ModuleIDs[I];
392     if (ID >= Modules.size() || !Modules[ID].File)
393       continue;
394 
395     Hits.insert(Modules[ID].File);
396   }
397 
398   ++NumIdentifierLookupHits;
399   return true;
400 }
401 
printStats()402 void GlobalModuleIndex::printStats() {
403   std::fprintf(stderr, "*** Global Module Index Statistics:\n");
404   if (NumIdentifierLookups) {
405     fprintf(stderr, "  %u / %u identifier lookups succeeded (%f%%)\n",
406             NumIdentifierLookupHits, NumIdentifierLookups,
407             (double)NumIdentifierLookupHits*100.0/NumIdentifierLookups);
408   }
409   std::fprintf(stderr, "\n");
410 }
411 
412 //----------------------------------------------------------------------------//
413 // Global module index writer.
414 //----------------------------------------------------------------------------//
415 
416 namespace {
417   /// \brief Provides information about a specific module file.
418   struct ModuleFileInfo {
419     /// \brief The numberic ID for this module file.
420     unsigned ID;
421 
422     /// \brief The set of modules on which this module depends. Each entry is
423     /// a module ID.
424     SmallVector<unsigned, 4> Dependencies;
425   };
426 
427   /// \brief Builder that generates the global module index file.
428   class GlobalModuleIndexBuilder {
429     FileManager &FileMgr;
430 
431     /// \brief Mapping from files to module file information.
432     typedef llvm::MapVector<const FileEntry *, ModuleFileInfo> ModuleFilesMap;
433 
434     /// \brief Information about each of the known module files.
435     ModuleFilesMap ModuleFiles;
436 
437     /// \brief Mapping from identifiers to the list of module file IDs that
438     /// consider this identifier to be interesting.
439     typedef llvm::StringMap<SmallVector<unsigned, 2> > InterestingIdentifierMap;
440 
441     /// \brief A mapping from all interesting identifiers to the set of module
442     /// files in which those identifiers are considered interesting.
443     InterestingIdentifierMap InterestingIdentifiers;
444 
445     /// \brief Write the block-info block for the global module index file.
446     void emitBlockInfoBlock(llvm::BitstreamWriter &Stream);
447 
448     /// \brief Retrieve the module file information for the given file.
getModuleFileInfo(const FileEntry * File)449     ModuleFileInfo &getModuleFileInfo(const FileEntry *File) {
450       llvm::MapVector<const FileEntry *, ModuleFileInfo>::iterator Known
451         = ModuleFiles.find(File);
452       if (Known != ModuleFiles.end())
453         return Known->second;
454 
455       unsigned NewID = ModuleFiles.size();
456       ModuleFileInfo &Info = ModuleFiles[File];
457       Info.ID = NewID;
458       return Info;
459     }
460 
461   public:
GlobalModuleIndexBuilder(FileManager & FileMgr)462     explicit GlobalModuleIndexBuilder(FileManager &FileMgr) : FileMgr(FileMgr){}
463 
464     /// \brief Load the contents of the given module file into the builder.
465     ///
466     /// \returns true if an error occurred, false otherwise.
467     bool loadModuleFile(const FileEntry *File);
468 
469     /// \brief Write the index to the given bitstream.
470     void writeIndex(llvm::BitstreamWriter &Stream);
471   };
472 }
473 
emitBlockID(unsigned ID,const char * Name,llvm::BitstreamWriter & Stream,SmallVectorImpl<uint64_t> & Record)474 static void emitBlockID(unsigned ID, const char *Name,
475                         llvm::BitstreamWriter &Stream,
476                         SmallVectorImpl<uint64_t> &Record) {
477   Record.clear();
478   Record.push_back(ID);
479   Stream.EmitRecord(llvm::bitc::BLOCKINFO_CODE_SETBID, Record);
480 
481   // Emit the block name if present.
482   if (Name == 0 || Name[0] == 0) return;
483   Record.clear();
484   while (*Name)
485     Record.push_back(*Name++);
486   Stream.EmitRecord(llvm::bitc::BLOCKINFO_CODE_BLOCKNAME, Record);
487 }
488 
emitRecordID(unsigned ID,const char * Name,llvm::BitstreamWriter & Stream,SmallVectorImpl<uint64_t> & Record)489 static void emitRecordID(unsigned ID, const char *Name,
490                          llvm::BitstreamWriter &Stream,
491                          SmallVectorImpl<uint64_t> &Record) {
492   Record.clear();
493   Record.push_back(ID);
494   while (*Name)
495     Record.push_back(*Name++);
496   Stream.EmitRecord(llvm::bitc::BLOCKINFO_CODE_SETRECORDNAME, Record);
497 }
498 
499 void
emitBlockInfoBlock(llvm::BitstreamWriter & Stream)500 GlobalModuleIndexBuilder::emitBlockInfoBlock(llvm::BitstreamWriter &Stream) {
501   SmallVector<uint64_t, 64> Record;
502   Stream.EnterSubblock(llvm::bitc::BLOCKINFO_BLOCK_ID, 3);
503 
504 #define BLOCK(X) emitBlockID(X ## _ID, #X, Stream, Record)
505 #define RECORD(X) emitRecordID(X, #X, Stream, Record)
506   BLOCK(GLOBAL_INDEX_BLOCK);
507   RECORD(INDEX_METADATA);
508   RECORD(MODULE);
509   RECORD(IDENTIFIER_INDEX);
510 #undef RECORD
511 #undef BLOCK
512 
513   Stream.ExitBlock();
514 }
515 
516 namespace {
517   class InterestingASTIdentifierLookupTrait
518     : public serialization::reader::ASTIdentifierLookupTraitBase {
519 
520   public:
521     /// \brief The identifier and whether it is "interesting".
522     typedef std::pair<StringRef, bool> data_type;
523 
ReadData(const internal_key_type & k,const unsigned char * d,unsigned DataLen)524     data_type ReadData(const internal_key_type& k,
525                        const unsigned char* d,
526                        unsigned DataLen) {
527       // The first bit indicates whether this identifier is interesting.
528       // That's all we care about.
529       using namespace clang::io;
530       unsigned RawID = ReadUnalignedLE32(d);
531       bool IsInteresting = RawID & 0x01;
532       return std::make_pair(k, IsInteresting);
533     }
534   };
535 }
536 
loadModuleFile(const FileEntry * File)537 bool GlobalModuleIndexBuilder::loadModuleFile(const FileEntry *File) {
538   // Open the module file.
539   OwningPtr<llvm::MemoryBuffer> Buffer;
540   std::string ErrorStr;
541   Buffer.reset(FileMgr.getBufferForFile(File, &ErrorStr, /*isVolatile=*/true));
542   if (!Buffer) {
543     return true;
544   }
545 
546   // Initialize the input stream
547   llvm::BitstreamReader InStreamFile;
548   llvm::BitstreamCursor InStream;
549   InStreamFile.init((const unsigned char *)Buffer->getBufferStart(),
550                   (const unsigned char *)Buffer->getBufferEnd());
551   InStream.init(InStreamFile);
552 
553   // Sniff for the signature.
554   if (InStream.Read(8) != 'C' ||
555       InStream.Read(8) != 'P' ||
556       InStream.Read(8) != 'C' ||
557       InStream.Read(8) != 'H') {
558     return true;
559   }
560 
561   // Record this module file and assign it a unique ID (if it doesn't have
562   // one already).
563   unsigned ID = getModuleFileInfo(File).ID;
564 
565   // Search for the blocks and records we care about.
566   enum { Other, ControlBlock, ASTBlock } State = Other;
567   bool Done = false;
568   while (!Done) {
569     llvm::BitstreamEntry Entry = InStream.advance();
570     switch (Entry.Kind) {
571     case llvm::BitstreamEntry::Error:
572       Done = true;
573       continue;
574 
575     case llvm::BitstreamEntry::Record:
576       // In the 'other' state, just skip the record. We don't care.
577       if (State == Other) {
578         InStream.skipRecord(Entry.ID);
579         continue;
580       }
581 
582       // Handle potentially-interesting records below.
583       break;
584 
585     case llvm::BitstreamEntry::SubBlock:
586       if (Entry.ID == CONTROL_BLOCK_ID) {
587         if (InStream.EnterSubBlock(CONTROL_BLOCK_ID))
588           return true;
589 
590         // Found the control block.
591         State = ControlBlock;
592         continue;
593       }
594 
595       if (Entry.ID == AST_BLOCK_ID) {
596         if (InStream.EnterSubBlock(AST_BLOCK_ID))
597           return true;
598 
599         // Found the AST block.
600         State = ASTBlock;
601         continue;
602       }
603 
604       if (InStream.SkipBlock())
605         return true;
606 
607       continue;
608 
609     case llvm::BitstreamEntry::EndBlock:
610       State = Other;
611       continue;
612     }
613 
614     // Read the given record.
615     SmallVector<uint64_t, 64> Record;
616     StringRef Blob;
617     unsigned Code = InStream.readRecord(Entry.ID, Record, &Blob);
618 
619     // Handle module dependencies.
620     if (State == ControlBlock && Code == IMPORTS) {
621       // Load each of the imported PCH files.
622       unsigned Idx = 0, N = Record.size();
623       while (Idx < N) {
624         // Read information about the AST file.
625 
626         // Skip the imported kind
627         ++Idx;
628 
629         // Skip the import location
630         ++Idx;
631 
632         // Retrieve the imported file name.
633         unsigned Length = Record[Idx++];
634         SmallString<128> ImportedFile(Record.begin() + Idx,
635                                       Record.begin() + Idx + Length);
636         Idx += Length;
637 
638         // Find the imported module file.
639         const FileEntry *DependsOnFile
640           = FileMgr.getFile(ImportedFile, /*openFile=*/false,
641                             /*cacheFailure=*/false);
642         if (!DependsOnFile)
643           return true;
644 
645         // Record the dependency.
646         unsigned DependsOnID = getModuleFileInfo(DependsOnFile).ID;
647         getModuleFileInfo(File).Dependencies.push_back(DependsOnID);
648       }
649 
650       continue;
651     }
652 
653     // Handle the identifier table
654     if (State == ASTBlock && Code == IDENTIFIER_TABLE && Record[0] > 0) {
655       typedef OnDiskChainedHashTable<InterestingASTIdentifierLookupTrait>
656         InterestingIdentifierTable;
657       llvm::OwningPtr<InterestingIdentifierTable>
658         Table(InterestingIdentifierTable::Create(
659                 (const unsigned char *)Blob.data() + Record[0],
660                 (const unsigned char *)Blob.data()));
661       for (InterestingIdentifierTable::data_iterator D = Table->data_begin(),
662                                                      DEnd = Table->data_end();
663            D != DEnd; ++D) {
664         std::pair<StringRef, bool> Ident = *D;
665         if (Ident.second)
666           InterestingIdentifiers[Ident.first].push_back(ID);
667         else
668           (void)InterestingIdentifiers[Ident.first];
669       }
670     }
671 
672     // We don't care about this record.
673   }
674 
675   return false;
676 }
677 
678 namespace {
679 
680 /// \brief Trait used to generate the identifier index as an on-disk hash
681 /// table.
682 class IdentifierIndexWriterTrait {
683 public:
684   typedef StringRef key_type;
685   typedef StringRef key_type_ref;
686   typedef SmallVector<unsigned, 2> data_type;
687   typedef const SmallVector<unsigned, 2> &data_type_ref;
688 
ComputeHash(key_type_ref Key)689   static unsigned ComputeHash(key_type_ref Key) {
690     return llvm::HashString(Key);
691   }
692 
693   std::pair<unsigned,unsigned>
EmitKeyDataLength(raw_ostream & Out,key_type_ref Key,data_type_ref Data)694   EmitKeyDataLength(raw_ostream& Out, key_type_ref Key, data_type_ref Data) {
695     unsigned KeyLen = Key.size();
696     unsigned DataLen = Data.size() * 4;
697     clang::io::Emit16(Out, KeyLen);
698     clang::io::Emit16(Out, DataLen);
699     return std::make_pair(KeyLen, DataLen);
700   }
701 
EmitKey(raw_ostream & Out,key_type_ref Key,unsigned KeyLen)702   void EmitKey(raw_ostream& Out, key_type_ref Key, unsigned KeyLen) {
703     Out.write(Key.data(), KeyLen);
704   }
705 
EmitData(raw_ostream & Out,key_type_ref Key,data_type_ref Data,unsigned DataLen)706   void EmitData(raw_ostream& Out, key_type_ref Key, data_type_ref Data,
707                 unsigned DataLen) {
708     for (unsigned I = 0, N = Data.size(); I != N; ++I)
709       clang::io::Emit32(Out, Data[I]);
710   }
711 };
712 
713 }
714 
writeIndex(llvm::BitstreamWriter & Stream)715 void GlobalModuleIndexBuilder::writeIndex(llvm::BitstreamWriter &Stream) {
716   using namespace llvm;
717 
718   // Emit the file header.
719   Stream.Emit((unsigned)'B', 8);
720   Stream.Emit((unsigned)'C', 8);
721   Stream.Emit((unsigned)'G', 8);
722   Stream.Emit((unsigned)'I', 8);
723 
724   // Write the block-info block, which describes the records in this bitcode
725   // file.
726   emitBlockInfoBlock(Stream);
727 
728   Stream.EnterSubblock(GLOBAL_INDEX_BLOCK_ID, 3);
729 
730   // Write the metadata.
731   SmallVector<uint64_t, 2> Record;
732   Record.push_back(CurrentVersion);
733   Stream.EmitRecord(INDEX_METADATA, Record);
734 
735   // Write the set of known module files.
736   for (ModuleFilesMap::iterator M = ModuleFiles.begin(),
737                                 MEnd = ModuleFiles.end();
738        M != MEnd; ++M) {
739     Record.clear();
740     Record.push_back(M->second.ID);
741     Record.push_back(M->first->getSize());
742     Record.push_back(M->first->getModificationTime());
743 
744     // File name
745     StringRef Name(M->first->getName());
746     Record.push_back(Name.size());
747     Record.append(Name.begin(), Name.end());
748 
749     // Dependencies
750     Record.push_back(M->second.Dependencies.size());
751     Record.append(M->second.Dependencies.begin(), M->second.Dependencies.end());
752     Stream.EmitRecord(MODULE, Record);
753   }
754 
755   // Write the identifier -> module file mapping.
756   {
757     OnDiskChainedHashTableGenerator<IdentifierIndexWriterTrait> Generator;
758     IdentifierIndexWriterTrait Trait;
759 
760     // Populate the hash table.
761     for (InterestingIdentifierMap::iterator I = InterestingIdentifiers.begin(),
762                                             IEnd = InterestingIdentifiers.end();
763          I != IEnd; ++I) {
764       Generator.insert(I->first(), I->second, Trait);
765     }
766 
767     // Create the on-disk hash table in a buffer.
768     SmallString<4096> IdentifierTable;
769     uint32_t BucketOffset;
770     {
771       llvm::raw_svector_ostream Out(IdentifierTable);
772       // Make sure that no bucket is at offset 0
773       clang::io::Emit32(Out, 0);
774       BucketOffset = Generator.Emit(Out, Trait);
775     }
776 
777     // Create a blob abbreviation
778     BitCodeAbbrev *Abbrev = new BitCodeAbbrev();
779     Abbrev->Add(BitCodeAbbrevOp(IDENTIFIER_INDEX));
780     Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32));
781     Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Blob));
782     unsigned IDTableAbbrev = Stream.EmitAbbrev(Abbrev);
783 
784     // Write the identifier table
785     Record.clear();
786     Record.push_back(IDENTIFIER_INDEX);
787     Record.push_back(BucketOffset);
788     Stream.EmitRecordWithBlob(IDTableAbbrev, Record, IdentifierTable.str());
789   }
790 
791   Stream.ExitBlock();
792 }
793 
794 GlobalModuleIndex::ErrorCode
writeIndex(FileManager & FileMgr,StringRef Path)795 GlobalModuleIndex::writeIndex(FileManager &FileMgr, StringRef Path) {
796   llvm::SmallString<128> IndexPath;
797   IndexPath += Path;
798   llvm::sys::path::append(IndexPath, IndexFileName);
799 
800   // Coordinate building the global index file with other processes that might
801   // try to do the same.
802   llvm::LockFileManager Locked(IndexPath);
803   switch (Locked) {
804   case llvm::LockFileManager::LFS_Error:
805     return EC_IOError;
806 
807   case llvm::LockFileManager::LFS_Owned:
808     // We're responsible for building the index ourselves. Do so below.
809     break;
810 
811   case llvm::LockFileManager::LFS_Shared:
812     // Someone else is responsible for building the index. We don't care
813     // when they finish, so we're done.
814     return EC_Building;
815   }
816 
817   // The module index builder.
818   GlobalModuleIndexBuilder Builder(FileMgr);
819 
820   // Load each of the module files.
821   llvm::error_code EC;
822   for (llvm::sys::fs::directory_iterator D(Path, EC), DEnd;
823        D != DEnd && !EC;
824        D.increment(EC)) {
825     // If this isn't a module file, we don't care.
826     if (llvm::sys::path::extension(D->path()) != ".pcm") {
827       // ... unless it's a .pcm.lock file, which indicates that someone is
828       // in the process of rebuilding a module. They'll rebuild the index
829       // at the end of that translation unit, so we don't have to.
830       if (llvm::sys::path::extension(D->path()) == ".pcm.lock")
831         return EC_Building;
832 
833       continue;
834     }
835 
836     // If we can't find the module file, skip it.
837     const FileEntry *ModuleFile = FileMgr.getFile(D->path());
838     if (!ModuleFile)
839       continue;
840 
841     // Load this module file.
842     if (Builder.loadModuleFile(ModuleFile))
843       return EC_IOError;
844   }
845 
846   // The output buffer, into which the global index will be written.
847   SmallVector<char, 16> OutputBuffer;
848   {
849     llvm::BitstreamWriter OutputStream(OutputBuffer);
850     Builder.writeIndex(OutputStream);
851   }
852 
853   // Write the global index file to a temporary file.
854   llvm::SmallString<128> IndexTmpPath;
855   int TmpFD;
856   if (llvm::sys::fs::unique_file(IndexPath + "-%%%%%%%%", TmpFD, IndexTmpPath))
857     return EC_IOError;
858 
859   // Open the temporary global index file for output.
860   llvm::raw_fd_ostream Out(TmpFD, true);
861   if (Out.has_error())
862     return EC_IOError;
863 
864   // Write the index.
865   Out.write(OutputBuffer.data(), OutputBuffer.size());
866   Out.close();
867   if (Out.has_error())
868     return EC_IOError;
869 
870   // Remove the old index file. It isn't relevant any more.
871   bool OldIndexExisted;
872   llvm::sys::fs::remove(IndexPath.str(), OldIndexExisted);
873 
874   // Rename the newly-written index file to the proper name.
875   if (llvm::sys::fs::rename(IndexTmpPath.str(), IndexPath.str())) {
876     // Rename failed; just remove the
877     llvm::sys::fs::remove(IndexTmpPath.str(), OldIndexExisted);
878     return EC_IOError;
879   }
880 
881   // We're done.
882   return EC_None;
883 }
884