Coverage Report

Created: 2023-09-30 09:22

/Users/buildslave/jenkins/workspace/coverage/llvm-project/clang/lib/Serialization/GlobalModuleIndex.cpp
Line
Count
Source (jump to first uncovered line)
1
//===--- GlobalModuleIndex.cpp - Global Module Index ------------*- C++ -*-===//
2
//
3
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4
// See https://llvm.org/LICENSE.txt for license information.
5
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6
//
7
//===----------------------------------------------------------------------===//
8
//
9
// This file implements the GlobalModuleIndex class.
10
//
11
//===----------------------------------------------------------------------===//
12
13
#include "clang/Serialization/GlobalModuleIndex.h"
14
#include "ASTReaderInternals.h"
15
#include "clang/Basic/FileManager.h"
16
#include "clang/Lex/HeaderSearch.h"
17
#include "clang/Serialization/ASTBitCodes.h"
18
#include "clang/Serialization/ModuleFile.h"
19
#include "clang/Serialization/PCHContainerOperations.h"
20
#include "llvm/ADT/DenseMap.h"
21
#include "llvm/ADT/MapVector.h"
22
#include "llvm/ADT/SmallString.h"
23
#include "llvm/ADT/StringRef.h"
24
#include "llvm/Bitstream/BitstreamReader.h"
25
#include "llvm/Bitstream/BitstreamWriter.h"
26
#include "llvm/Support/DJB.h"
27
#include "llvm/Support/FileSystem.h"
28
#include "llvm/Support/LockFileManager.h"
29
#include "llvm/Support/MemoryBuffer.h"
30
#include "llvm/Support/OnDiskHashTable.h"
31
#include "llvm/Support/Path.h"
32
#include "llvm/Support/TimeProfiler.h"
33
#include "llvm/Support/raw_ostream.h"
34
#include <cstdio>
35
using namespace clang;
36
using namespace serialization;
37
38
//----------------------------------------------------------------------------//
39
// Shared constants
40
//----------------------------------------------------------------------------//
41
namespace {
42
  enum {
43
    /// The block containing the index.
44
    GLOBAL_INDEX_BLOCK_ID = llvm::bitc::FIRST_APPLICATION_BLOCKID
45
  };
46
47
  /// Describes the record types in the index.
48
  enum IndexRecordTypes {
49
    /// Contains version information and potentially other metadata,
50
    /// used to determine if we can read this global index file.
51
    INDEX_METADATA,
52
    /// Describes a module, including its file name and dependencies.
53
    MODULE,
54
    /// The index for identifiers.
55
    IDENTIFIER_INDEX
56
  };
57
}
58
59
/// The name of the global index file.
60
static const char * const IndexFileName = "modules.idx";
61
62
/// The global index file version.
63
static const unsigned CurrentVersion = 1;
64
65
//----------------------------------------------------------------------------//
66
// Global module index reader.
67
//----------------------------------------------------------------------------//
68
69
namespace {
70
71
/// Trait used to read the identifier index from the on-disk hash
72
/// table.
73
class IdentifierIndexReaderTrait {
74
public:
75
  typedef StringRef external_key_type;
76
  typedef StringRef internal_key_type;
77
  typedef SmallVector<unsigned, 2> data_type;
78
  typedef unsigned hash_value_type;
79
  typedef unsigned offset_type;
80
81
24.2k
  static bool EqualKey(const internal_key_type& a, const internal_key_type& b) {
82
24.2k
    return a == b;
83
24.2k
  }
84
85
158k
  static hash_value_type ComputeHash(const internal_key_type& a) {
86
158k
    return llvm::djbHash(a);
87
158k
  }
88
89
  static std::pair<unsigned, unsigned>
90
101k
  ReadKeyDataLength(const unsigned char*& d) {
91
101k
    using namespace llvm::support;
92
101k
    unsigned KeyLen = endian::readNext<uint16_t, little, unaligned>(d);
93
101k
    unsigned DataLen = endian::readNext<uint16_t, little, unaligned>(d);
94
101k
    return std::make_pair(KeyLen, DataLen);
95
101k
  }
96
97
  static const internal_key_type&
98
158k
  GetInternalKey(const external_key_type& x) { return x; }
99
100
  static const external_key_type&
101
2.91k
  GetExternalKey(const internal_key_type& x) { return x; }
102
103
27.2k
  static internal_key_type ReadKey(const unsigned char* d, unsigned n) {
104
27.2k
    return StringRef((const char *)d, n);
105
27.2k
  }
106
107
  static data_type ReadData(const internal_key_type& k,
108
                            const unsigned char* d,
109
24.2k
                            unsigned DataLen) {
110
24.2k
    using namespace llvm::support;
111
112
24.2k
    data_type Result;
113
113k
    while (DataLen > 0) {
114
89.1k
      unsigned ID = endian::readNext<uint32_t, little, unaligned>(d);
115
89.1k
      Result.push_back(ID);
116
89.1k
      DataLen -= 4;
117
89.1k
    }
118
119
24.2k
    return Result;
120
24.2k
  }
121
};
122
123
typedef llvm::OnDiskIterableChainedHashTable<IdentifierIndexReaderTrait>
124
    IdentifierIndexTable;
125
126
}
127
128
GlobalModuleIndex::GlobalModuleIndex(
129
    std::unique_ptr<llvm::MemoryBuffer> IndexBuffer,
130
    llvm::BitstreamCursor Cursor)
131
402
    : Buffer(std::move(IndexBuffer)), IdentifierIndex(), NumIdentifierLookups(),
132
402
      NumIdentifierLookupHits() {
133
402
  auto Fail = [&](llvm::Error &&Err) {
134
0
    report_fatal_error("Module index '" + Buffer->getBufferIdentifier() +
135
0
                       "' failed: " + toString(std::move(Err)));
136
0
  };
137
138
402
  llvm::TimeTraceScope TimeScope("Module LoadIndex");
139
  // Read the global index.
140
402
  bool InGlobalIndexBlock = false;
141
402
  bool Done = false;
142
4.19k
  while (!Done) {
143
3.78k
    llvm::BitstreamEntry Entry;
144
3.78k
    if (Expected<llvm::BitstreamEntry> Res = Cursor.advance())
145
3.78k
      Entry = Res.get();
146
0
    else
147
0
      Fail(Res.takeError());
148
149
3.78k
    switch (Entry.Kind) {
150
0
    case llvm::BitstreamEntry::Error:
151
0
      return;
152
153
402
    case llvm::BitstreamEntry::EndBlock:
154
402
      if (InGlobalIndexBlock) {
155
402
        InGlobalIndexBlock = false;
156
402
        Done = true;
157
402
        continue;
158
402
      }
159
0
      return;
160
161
162
2.58k
    case llvm::BitstreamEntry::Record:
163
      // Entries in the global index block are handled below.
164
2.58k
      if (InGlobalIndexBlock)
165
2.58k
        break;
166
167
0
      return;
168
169
804
    case llvm::BitstreamEntry::SubBlock:
170
804
      if (!InGlobalIndexBlock && Entry.ID == GLOBAL_INDEX_BLOCK_ID) {
171
402
        if (llvm::Error Err = Cursor.EnterSubBlock(GLOBAL_INDEX_BLOCK_ID))
172
0
          Fail(std::move(Err));
173
402
        InGlobalIndexBlock = true;
174
402
      } else if (llvm::Error Err = Cursor.SkipBlock())
175
0
        Fail(std::move(Err));
176
804
      continue;
177
3.78k
    }
178
179
2.58k
    SmallVector<uint64_t, 64> Record;
180
2.58k
    StringRef Blob;
181
2.58k
    Expected<unsigned> MaybeIndexRecord =
182
2.58k
        Cursor.readRecord(Entry.ID, Record, &Blob);
183
2.58k
    if (!MaybeIndexRecord)
184
0
      Fail(MaybeIndexRecord.takeError());
185
2.58k
    IndexRecordTypes IndexRecord =
186
2.58k
        static_cast<IndexRecordTypes>(MaybeIndexRecord.get());
187
2.58k
    switch (IndexRecord) {
188
402
    case INDEX_METADATA:
189
      // Make sure that the version matches.
190
402
      if (Record.size() < 1 || Record[0] != CurrentVersion)
191
0
        return;
192
402
      break;
193
194
1.77k
    case MODULE: {
195
1.77k
      unsigned Idx = 0;
196
1.77k
      unsigned ID = Record[Idx++];
197
198
      // Make room for this module's information.
199
1.77k
      if (ID == Modules.size())
200
1.77k
        Modules.push_back(ModuleInfo());
201
0
      else
202
0
        Modules.resize(ID + 1);
203
204
      // Size/modification time for this module file at the time the
205
      // global index was built.
206
1.77k
      Modules[ID].Size = Record[Idx++];
207
1.77k
      Modules[ID].ModTime = Record[Idx++];
208
209
      // File name.
210
1.77k
      unsigned NameLen = Record[Idx++];
211
1.77k
      Modules[ID].FileName.assign(Record.begin() + Idx,
212
1.77k
                                  Record.begin() + Idx + NameLen);
213
1.77k
      Idx += NameLen;
214
215
      // Dependencies
216
1.77k
      unsigned NumDeps = Record[Idx++];
217
1.77k
      Modules[ID].Dependencies.insert(Modules[ID].Dependencies.end(),
218
1.77k
                                      Record.begin() + Idx,
219
1.77k
                                      Record.begin() + Idx + NumDeps);
220
1.77k
      Idx += NumDeps;
221
222
      // Make sure we're at the end of the record.
223
1.77k
      assert(Idx == Record.size() && "More module info?");
224
225
      // Record this module as an unresolved module.
226
      // FIXME: this doesn't work correctly for module names containing path
227
      // separators.
228
1.77k
      StringRef ModuleName = llvm::sys::path::stem(Modules[ID].FileName);
229
      // Remove the -<hash of ModuleMapPath>
230
1.77k
      ModuleName = ModuleName.rsplit('-').first;
231
1.77k
      UnresolvedModules[ModuleName] = ID;
232
1.77k
      break;
233
1.77k
    }
234
235
402
    case IDENTIFIER_INDEX:
236
      // Wire up the identifier index.
237
402
      if (Record[0]) {
238
402
        IdentifierIndex = IdentifierIndexTable::Create(
239
402
            (const unsigned char *)Blob.data() + Record[0],
240
402
            (const unsigned char *)Blob.data() + sizeof(uint32_t),
241
402
            (const unsigned char *)Blob.data(), IdentifierIndexReaderTrait());
242
402
      }
243
402
      break;
244
2.58k
    }
245
2.58k
  }
246
402
}
247
248
388
GlobalModuleIndex::~GlobalModuleIndex() {
249
388
  delete static_cast<IdentifierIndexTable *>(IdentifierIndex);
250
388
}
251
252
std::pair<GlobalModuleIndex *, llvm::Error>
253
2.38k
GlobalModuleIndex::readIndex(StringRef Path) {
254
  // Load the index file, if it's there.
255
2.38k
  llvm::SmallString<128> IndexPath;
256
2.38k
  IndexPath += Path;
257
2.38k
  llvm::sys::path::append(IndexPath, IndexFileName);
258
259
2.38k
  llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> BufferOrErr =
260
2.38k
      llvm::MemoryBuffer::getFile(IndexPath.c_str());
261
2.38k
  if (!BufferOrErr)
262
1.98k
    return std::make_pair(nullptr,
263
1.98k
                          llvm::errorCodeToError(BufferOrErr.getError()));
264
402
  std::unique_ptr<llvm::MemoryBuffer> Buffer = std::move(BufferOrErr.get());
265
266
  /// The main bitstream cursor for the main block.
267
402
  llvm::BitstreamCursor Cursor(*Buffer);
268
269
  // Sniff for the signature.
270
1.60k
  for (unsigned char C : {'B', 'C', 'G', 'I'}) {
271
1.60k
    if (Expected<llvm::SimpleBitstreamCursor::word_t> Res = Cursor.Read(8)) {
272
1.60k
      if (Res.get() != C)
273
0
        return std::make_pair(
274
0
            nullptr, llvm::createStringError(std::errc::illegal_byte_sequence,
275
0
                                             "expected signature BCGI"));
276
1.60k
    } else
277
0
      return std::make_pair(nullptr, Res.takeError());
278
1.60k
  }
279
280
402
  return std::make_pair(new GlobalModuleIndex(std::move(Buffer), std::move(Cursor)),
281
402
                        llvm::Error::success());
282
402
}
283
284
void GlobalModuleIndex::getModuleDependencies(
285
       ModuleFile *File,
286
0
       SmallVectorImpl<ModuleFile *> &Dependencies) {
287
  // Look for information about this module file.
288
0
  llvm::DenseMap<ModuleFile *, unsigned>::iterator Known
289
0
    = ModulesByFile.find(File);
290
0
  if (Known == ModulesByFile.end())
291
0
    return;
292
293
  // Record dependencies.
294
0
  Dependencies.clear();
295
0
  ArrayRef<unsigned> StoredDependencies = Modules[Known->second].Dependencies;
296
0
  for (unsigned I = 0, N = StoredDependencies.size(); I != N; ++I) {
297
0
    if (ModuleFile *MF = Modules[I].File)
298
0
      Dependencies.push_back(MF);
299
0
  }
300
0
}
301
302
158k
bool GlobalModuleIndex::lookupIdentifier(StringRef Name, HitSet &Hits) {
303
158k
  Hits.clear();
304
305
  // If there's no identifier index, there is nothing we can do.
306
158k
  if (!IdentifierIndex)
307
0
    return false;
308
309
  // Look into the identifier index.
310
158k
  ++NumIdentifierLookups;
311
158k
  IdentifierIndexTable &Table
312
158k
    = *static_cast<IdentifierIndexTable *>(IdentifierIndex);
313
158k
  IdentifierIndexTable::iterator Known = Table.find(Name);
314
158k
  if (Known == Table.end()) {
315
134k
    return false;
316
134k
  }
317
318
24.2k
  SmallVector<unsigned, 2> ModuleIDs = *Known;
319
113k
  for (unsigned I = 0, N = ModuleIDs.size(); I != N; 
++I89.1k
) {
320
89.1k
    if (ModuleFile *MF = Modules[ModuleIDs[I]].File)
321
6.26k
      Hits.insert(MF);
322
89.1k
  }
323
324
24.2k
  ++NumIdentifierLookupHits;
325
24.2k
  return true;
326
158k
}
327
328
4.03k
bool GlobalModuleIndex::loadedModuleFile(ModuleFile *File) {
329
  // Look for the module in the global module index based on the module name.
330
4.03k
  StringRef Name = File->ModuleName;
331
4.03k
  llvm::StringMap<unsigned>::iterator Known = UnresolvedModules.find(Name);
332
4.03k
  if (Known == UnresolvedModules.end()) {
333
3.48k
    return true;
334
3.48k
  }
335
336
  // Rectify this module with the global module index.
337
545
  ModuleInfo &Info = Modules[Known->second];
338
339
  //  If the size and modification time match what we expected, record this
340
  // module file.
341
545
  bool Failed = true;
342
545
  if (File->File->getSize() == Info.Size &&
343
545
      
File->File->getModificationTime() == Info.ModTime512
) {
344
500
    Info.File = File;
345
500
    ModulesByFile[File] = Known->second;
346
347
500
    Failed = false;
348
500
  }
349
350
  // One way or another, we have resolved this module file.
351
545
  UnresolvedModules.erase(Known);
352
545
  return Failed;
353
4.03k
}
354
355
1
void GlobalModuleIndex::printStats() {
356
1
  std::fprintf(stderr, "*** Global Module Index Statistics:\n");
357
1
  if (NumIdentifierLookups) {
358
1
    fprintf(stderr, "  %u / %u identifier lookups succeeded (%f%%)\n",
359
1
            NumIdentifierLookupHits, NumIdentifierLookups,
360
1
            (double)NumIdentifierLookupHits*100.0/NumIdentifierLookups);
361
1
  }
362
1
  std::fprintf(stderr, "\n");
363
1
}
364
365
0
LLVM_DUMP_METHOD void GlobalModuleIndex::dump() {
366
0
  llvm::errs() << "*** Global Module Index Dump:\n";
367
0
  llvm::errs() << "Module files:\n";
368
0
  for (auto &MI : Modules) {
369
0
    llvm::errs() << "** " << MI.FileName << "\n";
370
0
    if (MI.File)
371
0
      MI.File->dump();
372
0
    else
373
0
      llvm::errs() << "\n";
374
0
  }
375
0
  llvm::errs() << "\n";
376
0
}
377
378
//----------------------------------------------------------------------------//
379
// Global module index writer.
380
//----------------------------------------------------------------------------//
381
382
namespace {
383
  /// Provides information about a specific module file.
384
  struct ModuleFileInfo {
385
    /// The numberic ID for this module file.
386
    unsigned ID;
387
388
    /// The set of modules on which this module depends. Each entry is
389
    /// a module ID.
390
    SmallVector<unsigned, 4> Dependencies;
391
    ASTFileSignature Signature;
392
  };
393
394
  struct ImportedModuleFileInfo {
395
    off_t StoredSize;
396
    time_t StoredModTime;
397
    ASTFileSignature StoredSignature;
398
    ImportedModuleFileInfo(off_t Size, time_t ModTime, ASTFileSignature Sig)
399
1.77k
        : StoredSize(Size), StoredModTime(ModTime), StoredSignature(Sig) {}
400
  };
401
402
  /// Builder that generates the global module index file.
403
  class GlobalModuleIndexBuilder {
404
    FileManager &FileMgr;
405
    const PCHContainerReader &PCHContainerRdr;
406
407
    /// Mapping from files to module file information.
408
    using ModuleFilesMap = llvm::MapVector<FileEntryRef, ModuleFileInfo>;
409
410
    /// Information about each of the known module files.
411
    ModuleFilesMap ModuleFiles;
412
413
    /// Mapping from the imported module file to the imported
414
    /// information.
415
    using ImportedModuleFilesMap =
416
        std::multimap<FileEntryRef, ImportedModuleFileInfo>;
417
418
    /// Information about each importing of a module file.
419
    ImportedModuleFilesMap ImportedModuleFiles;
420
421
    /// Mapping from identifiers to the list of module file IDs that
422
    /// consider this identifier to be interesting.
423
    typedef llvm::StringMap<SmallVector<unsigned, 2> > InterestingIdentifierMap;
424
425
    /// A mapping from all interesting identifiers to the set of module
426
    /// files in which those identifiers are considered interesting.
427
    InterestingIdentifierMap InterestingIdentifiers;
428
429
    /// Write the block-info block for the global module index file.
430
    void emitBlockInfoBlock(llvm::BitstreamWriter &Stream);
431
432
    /// Retrieve the module file information for the given file.
433
10.9k
    ModuleFileInfo &getModuleFileInfo(FileEntryRef File) {
434
10.9k
      auto Known = ModuleFiles.find(File);
435
10.9k
      if (Known != ModuleFiles.end())
436
9.02k
        return Known->second;
437
438
1.96k
      unsigned NewID = ModuleFiles.size();
439
1.96k
      ModuleFileInfo &Info = ModuleFiles[File];
440
1.96k
      Info.ID = NewID;
441
1.96k
      return Info;
442
10.9k
    }
443
444
  public:
445
    explicit GlobalModuleIndexBuilder(
446
        FileManager &FileMgr, const PCHContainerReader &PCHContainerRdr)
447
781
        : FileMgr(FileMgr), PCHContainerRdr(PCHContainerRdr) {}
448
449
    /// Load the contents of the given module file into the builder.
450
    llvm::Error loadModuleFile(FileEntryRef File);
451
452
    /// Write the index to the given bitstream.
453
    /// \returns true if an error occurred, false otherwise.
454
    bool writeIndex(llvm::BitstreamWriter &Stream);
455
  };
456
}
457
458
static void emitBlockID(unsigned ID, const char *Name,
459
                        llvm::BitstreamWriter &Stream,
460
763
                        SmallVectorImpl<uint64_t> &Record) {
461
763
  Record.clear();
462
763
  Record.push_back(ID);
463
763
  Stream.EmitRecord(llvm::bitc::BLOCKINFO_CODE_SETBID, Record);
464
465
  // Emit the block name if present.
466
763
  if (!Name || Name[0] == 0) 
return0
;
467
763
  Record.clear();
468
14.4k
  while (*Name)
469
13.7k
    Record.push_back(*Name++);
470
763
  Stream.EmitRecord(llvm::bitc::BLOCKINFO_CODE_BLOCKNAME, Record);
471
763
}
472
473
static void emitRecordID(unsigned ID, const char *Name,
474
                         llvm::BitstreamWriter &Stream,
475
2.28k
                         SmallVectorImpl<uint64_t> &Record) {
476
2.28k
  Record.clear();
477
2.28k
  Record.push_back(ID);
478
29.7k
  while (*Name)
479
27.4k
    Record.push_back(*Name++);
480
2.28k
  Stream.EmitRecord(llvm::bitc::BLOCKINFO_CODE_SETRECORDNAME, Record);
481
2.28k
}
482
483
void
484
763
GlobalModuleIndexBuilder::emitBlockInfoBlock(llvm::BitstreamWriter &Stream) {
485
763
  SmallVector<uint64_t, 64> Record;
486
763
  Stream.EnterBlockInfoBlock();
487
488
763
#define BLOCK(X) emitBlockID(X ## _ID, #X, Stream, Record)
489
2.28k
#define RECORD(X) emitRecordID(X, #X, Stream, Record)
490
763
  BLOCK(GLOBAL_INDEX_BLOCK);
491
763
  RECORD(INDEX_METADATA);
492
763
  RECORD(MODULE);
493
763
  RECORD(IDENTIFIER_INDEX);
494
763
#undef RECORD
495
763
#undef BLOCK
496
497
763
  Stream.ExitBlock();
498
763
}
499
500
namespace {
501
  class InterestingASTIdentifierLookupTrait
502
    : public serialization::reader::ASTIdentifierLookupTraitBase {
503
504
  public:
505
    /// The identifier and whether it is "interesting".
506
    typedef std::pair<StringRef, bool> data_type;
507
508
    data_type ReadData(const internal_key_type& k,
509
                       const unsigned char* d,
510
896k
                       unsigned DataLen) {
511
      // The first bit indicates whether this identifier is interesting.
512
      // That's all we care about.
513
896k
      using namespace llvm::support;
514
896k
      unsigned RawID = endian::readNext<uint32_t, little, unaligned>(d);
515
896k
      bool IsInteresting = RawID & 0x01;
516
896k
      return std::make_pair(k, IsInteresting);
517
896k
    }
518
  };
519
}
520
521
1.96k
llvm::Error GlobalModuleIndexBuilder::loadModuleFile(FileEntryRef File) {
522
  // Open the module file.
523
524
1.96k
  auto Buffer = FileMgr.getBufferForFile(File, /*isVolatile=*/true);
525
1.96k
  if (!Buffer)
526
0
    return llvm::createStringError(Buffer.getError(),
527
0
                                   "failed getting buffer for module file");
528
529
  // Initialize the input stream
530
1.96k
  llvm::BitstreamCursor InStream(PCHContainerRdr.ExtractPCH(**Buffer));
531
532
  // Sniff for the signature.
533
1.96k
  for (unsigned char C : {'C', 'P', 'C', 'H'})
534
7.85k
    if (Expected<llvm::SimpleBitstreamCursor::word_t> Res = InStream.Read(8)) {
535
7.85k
      if (Res.get() != C)
536
0
        return llvm::createStringError(std::errc::illegal_byte_sequence,
537
0
                                       "expected signature CPCH");
538
7.85k
    } else
539
0
      return Res.takeError();
540
541
  // Record this module file and assign it a unique ID (if it doesn't have
542
  // one already).
543
1.96k
  unsigned ID = getModuleFileInfo(File).ID;
544
545
  // Search for the blocks and records we care about.
546
1.96k
  enum { Other, ControlBlock, ASTBlock, DiagnosticOptionsBlock } State = Other;
547
1.96k
  bool Done = false;
548
96.4k
  while (!Done) {
549
94.4k
    Expected<llvm::BitstreamEntry> MaybeEntry = InStream.advance();
550
94.4k
    if (!MaybeEntry)
551
0
      return MaybeEntry.takeError();
552
94.4k
    llvm::BitstreamEntry Entry = MaybeEntry.get();
553
554
94.4k
    switch (Entry.Kind) {
555
1.94k
    case llvm::BitstreamEntry::Error:
556
1.94k
      Done = true;
557
1.94k
      continue;
558
559
65.1k
    case llvm::BitstreamEntry::Record:
560
      // In the 'other' state, just skip the record. We don't care.
561
65.1k
      if (State == Other) {
562
0
        if (llvm::Expected<unsigned> Skipped = InStream.skipRecord(Entry.ID))
563
0
          continue;
564
0
        else
565
0
          return Skipped.takeError();
566
0
      }
567
568
      // Handle potentially-interesting records below.
569
65.1k
      break;
570
571
65.1k
    case llvm::BitstreamEntry::SubBlock:
572
21.5k
      if (Entry.ID == CONTROL_BLOCK_ID) {
573
1.96k
        if (llvm::Error Err = InStream.EnterSubBlock(CONTROL_BLOCK_ID))
574
0
          return Err;
575
576
        // Found the control block.
577
1.96k
        State = ControlBlock;
578
1.96k
        continue;
579
1.96k
      }
580
581
19.5k
      if (Entry.ID == AST_BLOCK_ID) {
582
1.94k
        if (llvm::Error Err = InStream.EnterSubBlock(AST_BLOCK_ID))
583
0
          return Err;
584
585
        // Found the AST block.
586
1.94k
        State = ASTBlock;
587
1.94k
        continue;
588
1.94k
      }
589
590
17.6k
      if (Entry.ID == UNHASHED_CONTROL_BLOCK_ID) {
591
1.96k
        if (llvm::Error Err = InStream.EnterSubBlock(UNHASHED_CONTROL_BLOCK_ID))
592
0
          return Err;
593
594
        // Found the Diagnostic Options block.
595
1.96k
        State = DiagnosticOptionsBlock;
596
1.96k
        continue;
597
1.96k
      }
598
599
15.6k
      if (llvm::Error Err = InStream.SkipBlock())
600
0
        return Err;
601
602
15.6k
      continue;
603
604
15.6k
    case llvm::BitstreamEntry::EndBlock:
605
5.85k
      State = Other;
606
5.85k
      continue;
607
94.4k
    }
608
609
    // Read the given record.
610
65.1k
    SmallVector<uint64_t, 64> Record;
611
65.1k
    StringRef Blob;
612
65.1k
    Expected<unsigned> MaybeCode = InStream.readRecord(Entry.ID, Record, &Blob);
613
65.1k
    if (!MaybeCode)
614
0
      return MaybeCode.takeError();
615
65.1k
    unsigned Code = MaybeCode.get();
616
617
    // Handle module dependencies.
618
65.1k
    if (State == ControlBlock && 
Code == IMPORTS13.7k
) {
619
      // Load each of the imported PCH files.
620
1.96k
      unsigned Idx = 0, N = Record.size();
621
3.73k
      while (Idx < N) {
622
        // Read information about the AST file.
623
624
        // Skip the imported kind
625
1.79k
        ++Idx;
626
627
        // Skip if it is standard C++ module
628
1.79k
        ++Idx;
629
630
        // Skip the import location
631
1.79k
        ++Idx;
632
633
        // Load stored size/modification time.
634
1.79k
        off_t StoredSize = (off_t)Record[Idx++];
635
1.79k
        time_t StoredModTime = (time_t)Record[Idx++];
636
637
        // Skip the stored signature.
638
        // FIXME: we could read the signature out of the import and validate it.
639
1.79k
        auto FirstSignatureByte = Record.begin() + Idx;
640
1.79k
        ASTFileSignature StoredSignature = ASTFileSignature::create(
641
1.79k
            FirstSignatureByte, FirstSignatureByte + ASTFileSignature::size);
642
1.79k
        Idx += ASTFileSignature::size;
643
644
        // Skip the module name (currently this is only used for prebuilt
645
        // modules while here we are only dealing with cached).
646
1.79k
        Idx += Record[Idx] + 1;
647
648
        // Retrieve the imported file name.
649
1.79k
        unsigned Length = Record[Idx++];
650
1.79k
        SmallString<128> ImportedFile(Record.begin() + Idx,
651
1.79k
                                      Record.begin() + Idx + Length);
652
1.79k
        Idx += Length;
653
654
        // Find the imported module file.
655
1.79k
        auto DependsOnFile =
656
1.79k
            FileMgr.getOptionalFileRef(ImportedFile, /*OpenFile=*/false,
657
1.79k
                                       /*CacheFailure=*/false);
658
659
1.79k
        if (!DependsOnFile)
660
15
          return llvm::createStringError(std::errc::bad_file_descriptor,
661
15
                                         "imported file \"%s\" not found",
662
15
                                         ImportedFile.c_str());
663
664
        // Save the information in ImportedModuleFileInfo so we can verify after
665
        // loading all pcms.
666
1.77k
        ImportedModuleFiles.insert(std::make_pair(
667
1.77k
            *DependsOnFile, ImportedModuleFileInfo(StoredSize, StoredModTime,
668
1.77k
                                                   StoredSignature)));
669
670
        // Record the dependency.
671
1.77k
        unsigned DependsOnID = getModuleFileInfo(*DependsOnFile).ID;
672
1.77k
        getModuleFileInfo(File).Dependencies.push_back(DependsOnID);
673
1.77k
      }
674
675
1.94k
      continue;
676
1.96k
    }
677
678
    // Handle the identifier table
679
63.1k
    if (State == ASTBlock && 
Code == IDENTIFIER_TABLE39.6k
&&
Record[0] > 01.94k
) {
680
1.94k
      typedef llvm::OnDiskIterableChainedHashTable<
681
1.94k
          InterestingASTIdentifierLookupTrait> InterestingIdentifierTable;
682
1.94k
      std::unique_ptr<InterestingIdentifierTable> Table(
683
1.94k
          InterestingIdentifierTable::Create(
684
1.94k
              (const unsigned char *)Blob.data() + Record[0],
685
1.94k
              (const unsigned char *)Blob.data() + sizeof(uint32_t),
686
1.94k
              (const unsigned char *)Blob.data()));
687
1.94k
      for (InterestingIdentifierTable::data_iterator D = Table->data_begin(),
688
1.94k
                                                     DEnd = Table->data_end();
689
898k
           D != DEnd; 
++D896k
) {
690
896k
        std::pair<StringRef, bool> Ident = *D;
691
896k
        if (Ident.second)
692
572k
          InterestingIdentifiers[Ident.first].push_back(ID);
693
324k
        else
694
324k
          (void)InterestingIdentifiers[Ident.first];
695
896k
      }
696
1.94k
    }
697
698
    // Get Signature.
699
63.1k
    if (State == DiagnosticOptionsBlock && 
Code == SIGNATURE11.7k
) {
700
1.94k
      auto Signature = ASTFileSignature::create(Blob.begin(), Blob.end());
701
1.94k
      assert(Signature != ASTFileSignature::createDummy() &&
702
1.94k
             "Dummy AST file signature not backpatched in ASTWriter.");
703
1.94k
      getModuleFileInfo(File).Signature = Signature;
704
1.94k
    }
705
706
    // We don't care about this record.
707
63.1k
  }
708
709
1.94k
  return llvm::Error::success();
710
1.96k
}
711
712
namespace {
713
714
/// Trait used to generate the identifier index as an on-disk hash
715
/// table.
716
class IdentifierIndexWriterTrait {
717
public:
718
  typedef StringRef key_type;
719
  typedef StringRef key_type_ref;
720
  typedef SmallVector<unsigned, 2> data_type;
721
  typedef const SmallVector<unsigned, 2> &data_type_ref;
722
  typedef unsigned hash_value_type;
723
  typedef unsigned offset_type;
724
725
809k
  static hash_value_type ComputeHash(key_type_ref Key) {
726
809k
    return llvm::djbHash(Key);
727
809k
  }
728
729
  std::pair<unsigned,unsigned>
730
809k
  EmitKeyDataLength(raw_ostream& Out, key_type_ref Key, data_type_ref Data) {
731
809k
    using namespace llvm::support;
732
809k
    endian::Writer LE(Out, little);
733
809k
    unsigned KeyLen = Key.size();
734
809k
    unsigned DataLen = Data.size() * 4;
735
809k
    LE.write<uint16_t>(KeyLen);
736
809k
    LE.write<uint16_t>(DataLen);
737
809k
    return std::make_pair(KeyLen, DataLen);
738
809k
  }
739
740
809k
  void EmitKey(raw_ostream& Out, key_type_ref Key, unsigned KeyLen) {
741
809k
    Out.write(Key.data(), KeyLen);
742
809k
  }
743
744
  void EmitData(raw_ostream& Out, key_type_ref Key, data_type_ref Data,
745
809k
                unsigned DataLen) {
746
809k
    using namespace llvm::support;
747
1.38M
    for (unsigned I = 0, N = Data.size(); I != N; 
++I571k
)
748
571k
      endian::write<uint32_t>(Out, Data[I], little);
749
809k
  }
750
};
751
752
}
753
754
766
bool GlobalModuleIndexBuilder::writeIndex(llvm::BitstreamWriter &Stream) {
755
1.76k
  for (auto MapEntry : ImportedModuleFiles) {
756
1.76k
    auto File = MapEntry.first;
757
1.76k
    ImportedModuleFileInfo &Info = MapEntry.second;
758
1.76k
    if (getModuleFileInfo(File).Signature) {
759
1.76k
      if (getModuleFileInfo(File).Signature != Info.StoredSignature)
760
        // Verify Signature.
761
3
        return true;
762
1.76k
    } else 
if (4
Info.StoredSize != File.getSize()4
||
763
4
               Info.StoredModTime != File.getModificationTime())
764
      // Verify Size and ModTime.
765
0
      return true;
766
1.76k
  }
767
768
763
  using namespace llvm;
769
763
  llvm::TimeTraceScope TimeScope("Module WriteIndex");
770
771
  // Emit the file header.
772
763
  Stream.Emit((unsigned)'B', 8);
773
763
  Stream.Emit((unsigned)'C', 8);
774
763
  Stream.Emit((unsigned)'G', 8);
775
763
  Stream.Emit((unsigned)'I', 8);
776
777
  // Write the block-info block, which describes the records in this bitcode
778
  // file.
779
763
  emitBlockInfoBlock(Stream);
780
781
763
  Stream.EnterSubblock(GLOBAL_INDEX_BLOCK_ID, 3);
782
783
  // Write the metadata.
784
763
  SmallVector<uint64_t, 2> Record;
785
763
  Record.push_back(CurrentVersion);
786
763
  Stream.EmitRecord(INDEX_METADATA, Record);
787
788
  // Write the set of known module files.
789
763
  for (ModuleFilesMap::iterator M = ModuleFiles.begin(),
790
763
                                MEnd = ModuleFiles.end();
791
2.69k
       M != MEnd; 
++M1.92k
) {
792
1.92k
    Record.clear();
793
1.92k
    Record.push_back(M->second.ID);
794
1.92k
    Record.push_back(M->first.getSize());
795
1.92k
    Record.push_back(M->first.getModificationTime());
796
797
    // File name
798
1.92k
    StringRef Name(M->first.getName());
799
1.92k
    Record.push_back(Name.size());
800
1.92k
    Record.append(Name.begin(), Name.end());
801
802
    // Dependencies
803
1.92k
    Record.push_back(M->second.Dependencies.size());
804
1.92k
    Record.append(M->second.Dependencies.begin(), M->second.Dependencies.end());
805
1.92k
    Stream.EmitRecord(MODULE, Record);
806
1.92k
  }
807
808
  // Write the identifier -> module file mapping.
809
763
  {
810
763
    llvm::OnDiskChainedHashTableGenerator<IdentifierIndexWriterTrait> Generator;
811
763
    IdentifierIndexWriterTrait Trait;
812
813
    // Populate the hash table.
814
763
    for (InterestingIdentifierMap::iterator I = InterestingIdentifiers.begin(),
815
763
                                            IEnd = InterestingIdentifiers.end();
816
810k
         I != IEnd; 
++I809k
) {
817
809k
      Generator.insert(I->first(), I->second, Trait);
818
809k
    }
819
820
    // Create the on-disk hash table in a buffer.
821
763
    SmallString<4096> IdentifierTable;
822
763
    uint32_t BucketOffset;
823
763
    {
824
763
      using namespace llvm::support;
825
763
      llvm::raw_svector_ostream Out(IdentifierTable);
826
      // Make sure that no bucket is at offset 0
827
763
      endian::write<uint32_t>(Out, 0, little);
828
763
      BucketOffset = Generator.Emit(Out, Trait);
829
763
    }
830
831
    // Create a blob abbreviation
832
763
    auto Abbrev = std::make_shared<BitCodeAbbrev>();
833
763
    Abbrev->Add(BitCodeAbbrevOp(IDENTIFIER_INDEX));
834
763
    Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32));
835
763
    Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Blob));
836
763
    unsigned IDTableAbbrev = Stream.EmitAbbrev(std::move(Abbrev));
837
838
    // Write the identifier table
839
763
    uint64_t Record[] = {IDENTIFIER_INDEX, BucketOffset};
840
763
    Stream.EmitRecordWithBlob(IDTableAbbrev, Record, IdentifierTable);
841
763
  }
842
843
763
  Stream.ExitBlock();
844
763
  return false;
845
766
}
846
847
llvm::Error
848
GlobalModuleIndex::writeIndex(FileManager &FileMgr,
849
                              const PCHContainerReader &PCHContainerRdr,
850
809
                              StringRef Path) {
851
809
  llvm::SmallString<128> IndexPath;
852
809
  IndexPath += Path;
853
809
  llvm::sys::path::append(IndexPath, IndexFileName);
854
855
  // Coordinate building the global index file with other processes that might
856
  // try to do the same.
857
809
  llvm::LockFileManager Locked(IndexPath);
858
809
  switch (Locked) {
859
28
  case llvm::LockFileManager::LFS_Error:
860
28
    return llvm::createStringError(std::errc::io_error, "LFS error");
861
862
781
  case llvm::LockFileManager::LFS_Owned:
863
    // We're responsible for building the index ourselves. Do so below.
864
781
    break;
865
866
0
  case llvm::LockFileManager::LFS_Shared:
867
    // Someone else is responsible for building the index. We don't care
868
    // when they finish, so we're done.
869
0
    return llvm::createStringError(std::errc::device_or_resource_busy,
870
0
                                   "someone else is building the index");
871
809
  }
872
873
  // The module index builder.
874
781
  GlobalModuleIndexBuilder Builder(FileMgr, PCHContainerRdr);
875
876
  // Load each of the module files.
877
781
  std::error_code EC;
878
781
  for (llvm::sys::fs::directory_iterator D(Path, EC), DEnd;
879
4.42k
       D != DEnd && 
!EC3.66k
;
880
3.66k
       
D.increment(EC)3.64k
) {
881
    // If this isn't a module file, we don't care.
882
3.66k
    if (llvm::sys::path::extension(D->path()) != ".pcm") {
883
      // ... unless it's a .pcm.lock file, which indicates that someone is
884
      // in the process of rebuilding a module. They'll rebuild the index
885
      // at the end of that translation unit, so we don't have to.
886
1.69k
      if (llvm::sys::path::extension(D->path()) == ".pcm.lock")
887
0
        return llvm::createStringError(std::errc::device_or_resource_busy,
888
0
                                       "someone else is building the index");
889
890
1.69k
      continue;
891
1.69k
    }
892
893
    // If we can't find the module file, skip it.
894
1.96k
    auto ModuleFile = FileMgr.getOptionalFileRef(D->path());
895
1.96k
    if (!ModuleFile)
896
0
      continue;
897
898
    // Load this module file.
899
1.96k
    if (llvm::Error Err = Builder.loadModuleFile(*ModuleFile))
900
15
      return Err;
901
1.96k
  }
902
903
  // The output buffer, into which the global index will be written.
904
766
  SmallString<16> OutputBuffer;
905
766
  {
906
766
    llvm::BitstreamWriter OutputStream(OutputBuffer);
907
766
    if (Builder.writeIndex(OutputStream))
908
3
      return llvm::createStringError(std::errc::io_error,
909
3
                                     "failed writing index");
910
766
  }
911
912
763
  return llvm::writeToOutput(IndexPath, [&OutputBuffer](llvm::raw_ostream &OS) {
913
763
    OS << OutputBuffer;
914
763
    return llvm::Error::success();
915
763
  });
916
766
}
917
918
namespace {
919
  class GlobalIndexIdentifierIterator : public IdentifierIterator {
920
    /// The current position within the identifier lookup table.
921
    IdentifierIndexTable::key_iterator Current;
922
923
    /// The end position within the identifier lookup table.
924
    IdentifierIndexTable::key_iterator End;
925
926
  public:
927
11
    explicit GlobalIndexIdentifierIterator(IdentifierIndexTable &Idx) {
928
11
      Current = Idx.key_begin();
929
11
      End = Idx.key_end();
930
11
    }
931
932
2.92k
    StringRef Next() override {
933
2.92k
      if (Current == End)
934
11
        return StringRef();
935
936
2.91k
      StringRef Result = *Current;
937
2.91k
      ++Current;
938
2.91k
      return Result;
939
2.92k
    }
940
  };
941
}
942
943
11
IdentifierIterator *GlobalModuleIndex::createIdentifierIterator() const {
944
11
  IdentifierIndexTable &Table =
945
11
    *static_cast<IdentifierIndexTable *>(IdentifierIndex);
946
11
  return new GlobalIndexIdentifierIterator(Table);
947
11
}