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/Serialization/ASTBitCodes.h"
17#include "clang/Serialization/ModuleFile.h"
18#include "clang/Serialization/PCHContainerOperations.h"
19#include "llvm/ADT/DenseMap.h"
20#include "llvm/ADT/MapVector.h"
21#include "llvm/ADT/SmallString.h"
22#include "llvm/ADT/StringRef.h"
23#include "llvm/Bitstream/BitstreamReader.h"
24#include "llvm/Bitstream/BitstreamWriter.h"
25#include "llvm/Support/DJB.h"
26#include "llvm/Support/FileSystem.h"
27#include "llvm/Support/IOSandbox.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>
35using namespace clang;
36using namespace serialization;
37
38//----------------------------------------------------------------------------//
39// Shared constants
40//----------------------------------------------------------------------------//
41namespace {
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.
60static const char * const IndexFileName = "modules.idx";
61
62/// The global index file version.
63static const unsigned CurrentVersion = 1;
64
65//----------------------------------------------------------------------------//
66// Global module index reader.
67//----------------------------------------------------------------------------//
68
69namespace {
70
71/// Trait used to read the identifier index from the on-disk hash
72/// table.
73class IdentifierIndexReaderTrait {
74public:
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 static bool EqualKey(const internal_key_type& a, const internal_key_type& b) {
82 return a == b;
83 }
84
85 static hash_value_type ComputeHash(const internal_key_type& a) {
86 return llvm::djbHash(Buffer: a);
87 }
88
89 static std::pair<unsigned, unsigned>
90 ReadKeyDataLength(const unsigned char*& d) {
91 using namespace llvm::support;
92 unsigned KeyLen = endian::readNext<uint16_t, llvm::endianness::little>(memory&: d);
93 unsigned DataLen = endian::readNext<uint16_t, llvm::endianness::little>(memory&: d);
94 return std::make_pair(x&: KeyLen, y&: DataLen);
95 }
96
97 static const internal_key_type&
98 GetInternalKey(const external_key_type& x) { return x; }
99
100 static const external_key_type&
101 GetExternalKey(const internal_key_type& x) { return x; }
102
103 static internal_key_type ReadKey(const unsigned char* d, unsigned n) {
104 return StringRef((const char *)d, n);
105 }
106
107 static data_type ReadData(const internal_key_type& k,
108 const unsigned char* d,
109 unsigned DataLen) {
110 using namespace llvm::support;
111
112 data_type Result;
113 while (DataLen > 0) {
114 unsigned ID = endian::readNext<uint32_t, llvm::endianness::little>(memory&: d);
115 Result.push_back(Elt: ID);
116 DataLen -= 4;
117 }
118
119 return Result;
120 }
121};
122
123typedef llvm::OnDiskIterableChainedHashTable<IdentifierIndexReaderTrait>
124 IdentifierIndexTable;
125
126}
127
128GlobalModuleIndex::GlobalModuleIndex(
129 std::unique_ptr<llvm::MemoryBuffer> IndexBuffer,
130 llvm::BitstreamCursor Cursor)
131 : Buffer(std::move(IndexBuffer)), IdentifierIndex(), NumIdentifierLookups(),
132 NumIdentifierLookupHits() {
133 auto Fail = [&](llvm::Error &&Err) {
134 report_fatal_error(reason: "Module index '" + Buffer->getBufferIdentifier() +
135 "' failed: " + toString(E: std::move(Err)));
136 };
137
138 llvm::TimeTraceScope TimeScope("Module LoadIndex");
139 // Read the global index.
140 bool InGlobalIndexBlock = false;
141 bool Done = false;
142 while (!Done) {
143 llvm::BitstreamEntry Entry;
144 if (Expected<llvm::BitstreamEntry> Res = Cursor.advance())
145 Entry = Res.get();
146 else
147 Fail(Res.takeError());
148
149 switch (Entry.Kind) {
150 case llvm::BitstreamEntry::Error:
151 return;
152
153 case llvm::BitstreamEntry::EndBlock:
154 if (InGlobalIndexBlock) {
155 InGlobalIndexBlock = false;
156 Done = true;
157 continue;
158 }
159 return;
160
161
162 case llvm::BitstreamEntry::Record:
163 // Entries in the global index block are handled below.
164 if (InGlobalIndexBlock)
165 break;
166
167 return;
168
169 case llvm::BitstreamEntry::SubBlock:
170 if (!InGlobalIndexBlock && Entry.ID == GLOBAL_INDEX_BLOCK_ID) {
171 if (llvm::Error Err = Cursor.EnterSubBlock(BlockID: GLOBAL_INDEX_BLOCK_ID))
172 Fail(std::move(Err));
173 InGlobalIndexBlock = true;
174 } else if (llvm::Error Err = Cursor.SkipBlock())
175 Fail(std::move(Err));
176 continue;
177 }
178
179 SmallVector<uint64_t, 64> Record;
180 StringRef Blob;
181 Expected<unsigned> MaybeIndexRecord =
182 Cursor.readRecord(AbbrevID: Entry.ID, Vals&: Record, Blob: &Blob);
183 if (!MaybeIndexRecord)
184 Fail(MaybeIndexRecord.takeError());
185 IndexRecordTypes IndexRecord =
186 static_cast<IndexRecordTypes>(MaybeIndexRecord.get());
187 switch (IndexRecord) {
188 case INDEX_METADATA:
189 // Make sure that the version matches.
190 if (Record.size() < 1 || Record[0] != CurrentVersion)
191 return;
192 break;
193
194 case MODULE: {
195 unsigned Idx = 0;
196 unsigned ID = Record[Idx++];
197
198 // Make room for this module's information.
199 if (ID == Modules.size())
200 Modules.push_back(Elt: ModuleInfo());
201 else
202 Modules.resize(N: ID + 1);
203
204 // Size/modification time for this module file at the time the
205 // global index was built.
206 Modules[ID].Size = Record[Idx++];
207 Modules[ID].ModTime = Record[Idx++];
208
209 // File name.
210 unsigned NameLen = Record[Idx++];
211 Modules[ID].FileName.assign(first: Record.begin() + Idx,
212 last: Record.begin() + Idx + NameLen);
213 Idx += NameLen;
214
215 // Dependencies
216 unsigned NumDeps = Record[Idx++];
217 Modules[ID].Dependencies.insert(I: Modules[ID].Dependencies.end(),
218 From: Record.begin() + Idx,
219 To: Record.begin() + Idx + NumDeps);
220 Idx += NumDeps;
221
222 // Make sure we're at the end of the record.
223 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 StringRef ModuleName = llvm::sys::path::stem(path: Modules[ID].FileName);
229 // Remove the -<hash of ModuleMapPath>
230 ModuleName = ModuleName.rsplit(Separator: '-').first;
231 UnresolvedModules[ModuleName] = ID;
232 break;
233 }
234
235 case IDENTIFIER_INDEX:
236 // Wire up the identifier index.
237 if (Record[0]) {
238 IdentifierIndex = IdentifierIndexTable::Create(
239 Buckets: (const unsigned char *)Blob.data() + Record[0],
240 Payload: (const unsigned char *)Blob.data() + sizeof(uint32_t),
241 Base: (const unsigned char *)Blob.data(), InfoObj: IdentifierIndexReaderTrait());
242 }
243 break;
244 }
245 }
246}
247
248GlobalModuleIndex::~GlobalModuleIndex() {
249 delete static_cast<IdentifierIndexTable *>(IdentifierIndex);
250}
251
252std::pair<GlobalModuleIndex *, llvm::Error>
253GlobalModuleIndex::readIndex(StringRef Path) {
254 // This is a compiler-internal input/output, let's bypass the sandbox.
255 auto BypassSandbox = llvm::sys::sandbox::scopedDisable();
256
257 // Load the index file, if it's there.
258 llvm::SmallString<128> IndexPath;
259 IndexPath += Path;
260 llvm::sys::path::append(path&: IndexPath, a: IndexFileName);
261
262 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> BufferOrErr =
263 llvm::MemoryBuffer::getFile(Filename: IndexPath.c_str());
264 if (!BufferOrErr)
265 return std::make_pair(x: nullptr,
266 y: llvm::errorCodeToError(EC: BufferOrErr.getError()));
267 std::unique_ptr<llvm::MemoryBuffer> Buffer = std::move(BufferOrErr.get());
268
269 /// The main bitstream cursor for the main block.
270 llvm::BitstreamCursor Cursor(*Buffer);
271
272 // Sniff for the signature.
273 for (unsigned char C : {'B', 'C', 'G', 'I'}) {
274 if (Expected<llvm::SimpleBitstreamCursor::word_t> Res = Cursor.Read(NumBits: 8)) {
275 if (Res.get() != C)
276 return std::make_pair(
277 x: nullptr, y: llvm::createStringError(EC: std::errc::illegal_byte_sequence,
278 Fmt: "expected signature BCGI"));
279 } else
280 return std::make_pair(x: nullptr, y: Res.takeError());
281 }
282
283 return std::make_pair(x: new GlobalModuleIndex(std::move(Buffer), std::move(Cursor)),
284 y: llvm::Error::success());
285}
286
287void GlobalModuleIndex::getModuleDependencies(
288 ModuleFile *File,
289 SmallVectorImpl<ModuleFile *> &Dependencies) {
290 // Look for information about this module file.
291 llvm::DenseMap<ModuleFile *, unsigned>::iterator Known
292 = ModulesByFile.find(Val: File);
293 if (Known == ModulesByFile.end())
294 return;
295
296 // Record dependencies.
297 Dependencies.clear();
298 ArrayRef<unsigned> StoredDependencies = Modules[Known->second].Dependencies;
299 for (unsigned I = 0, N = StoredDependencies.size(); I != N; ++I) {
300 if (ModuleFile *MF = Modules[I].File)
301 Dependencies.push_back(Elt: MF);
302 }
303}
304
305bool GlobalModuleIndex::lookupIdentifier(StringRef Name, HitSet &Hits) {
306 Hits.clear();
307
308 // If there's no identifier index, there is nothing we can do.
309 if (!IdentifierIndex)
310 return false;
311
312 // Look into the identifier index.
313 ++NumIdentifierLookups;
314 IdentifierIndexTable &Table
315 = *static_cast<IdentifierIndexTable *>(IdentifierIndex);
316 IdentifierIndexTable::iterator Known = Table.find(EKey: Name);
317 if (Known == Table.end()) {
318 return false;
319 }
320
321 SmallVector<unsigned, 2> ModuleIDs = *Known;
322 for (unsigned I = 0, N = ModuleIDs.size(); I != N; ++I) {
323 if (ModuleFile *MF = Modules[ModuleIDs[I]].File)
324 Hits.insert(Ptr: MF);
325 }
326
327 ++NumIdentifierLookupHits;
328 return true;
329}
330
331bool GlobalModuleIndex::loadedModuleFile(ModuleFile *File) {
332 // Look for the module in the global module index based on the module name.
333 StringRef Name = File->ModuleName;
334 llvm::StringMap<unsigned>::iterator Known = UnresolvedModules.find(Key: Name);
335 if (Known == UnresolvedModules.end()) {
336 return true;
337 }
338
339 // Rectify this module with the global module index.
340 ModuleInfo &Info = Modules[Known->second];
341
342 // If the size and modification time match what we expected, record this
343 // module file.
344 bool Failed = true;
345 if (File->File.getSize() == Info.Size &&
346 File->File.getModificationTime() == Info.ModTime) {
347 Info.File = File;
348 ModulesByFile[File] = Known->second;
349
350 Failed = false;
351 }
352
353 // One way or another, we have resolved this module file.
354 UnresolvedModules.erase(I: Known);
355 return Failed;
356}
357
358void GlobalModuleIndex::printStats() {
359 std::fprintf(stderr, format: "*** Global Module Index Statistics:\n");
360 if (NumIdentifierLookups) {
361 fprintf(stderr, format: " %u / %u identifier lookups succeeded (%f%%)\n",
362 NumIdentifierLookupHits, NumIdentifierLookups,
363 (double)NumIdentifierLookupHits*100.0/NumIdentifierLookups);
364 }
365 std::fprintf(stderr, format: "\n");
366}
367
368LLVM_DUMP_METHOD void GlobalModuleIndex::dump() {
369 llvm::errs() << "*** Global Module Index Dump:\n";
370 llvm::errs() << "Module files:\n";
371 for (auto &MI : Modules) {
372 llvm::errs() << "** " << MI.FileName << "\n";
373 if (MI.File)
374 MI.File->dump();
375 else
376 llvm::errs() << "\n";
377 }
378 llvm::errs() << "\n";
379}
380
381//----------------------------------------------------------------------------//
382// Global module index writer.
383//----------------------------------------------------------------------------//
384
385namespace {
386 /// Provides information about a specific module file.
387 struct ModuleFileInfo {
388 /// The numberic ID for this module file.
389 unsigned ID;
390
391 /// The set of modules on which this module depends. Each entry is
392 /// a module ID.
393 SmallVector<unsigned, 4> Dependencies;
394 ASTFileSignature Signature;
395 };
396
397 struct ImportedModuleFileInfo {
398 off_t StoredSize;
399 time_t StoredModTime;
400 ASTFileSignature StoredSignature;
401 ImportedModuleFileInfo(off_t Size, time_t ModTime, ASTFileSignature Sig)
402 : StoredSize(Size), StoredModTime(ModTime), StoredSignature(Sig) {}
403 };
404
405 /// Builder that generates the global module index file.
406 class GlobalModuleIndexBuilder {
407 FileManager &FileMgr;
408 const PCHContainerReader &PCHContainerRdr;
409
410 /// Mapping from files to module file information.
411 using ModuleFilesMap = llvm::MapVector<FileEntryRef, ModuleFileInfo>;
412
413 /// Information about each of the known module files.
414 ModuleFilesMap ModuleFiles;
415
416 /// Mapping from the imported module file to the imported
417 /// information.
418 using ImportedModuleFilesMap =
419 std::multimap<FileEntryRef, ImportedModuleFileInfo>;
420
421 /// Information about each importing of a module file.
422 ImportedModuleFilesMap ImportedModuleFiles;
423
424 /// Mapping from identifiers to the list of module file IDs that
425 /// consider this identifier to be interesting.
426 typedef llvm::StringMap<SmallVector<unsigned, 2> > InterestingIdentifierMap;
427
428 /// A mapping from all interesting identifiers to the set of module
429 /// files in which those identifiers are considered interesting.
430 InterestingIdentifierMap InterestingIdentifiers;
431
432 /// Write the block-info block for the global module index file.
433 void emitBlockInfoBlock(llvm::BitstreamWriter &Stream);
434
435 /// Retrieve the module file information for the given file.
436 ModuleFileInfo &getModuleFileInfo(FileEntryRef File) {
437 auto [It, Inserted] = ModuleFiles.try_emplace(Key: File);
438 if (Inserted) {
439 unsigned NewID = ModuleFiles.size();
440 ModuleFileInfo &Info = It->second;
441 Info.ID = NewID;
442 }
443 return It->second;
444 }
445
446 public:
447 explicit GlobalModuleIndexBuilder(
448 FileManager &FileMgr, const PCHContainerReader &PCHContainerRdr)
449 : FileMgr(FileMgr), PCHContainerRdr(PCHContainerRdr) {}
450
451 /// Load the contents of the given module file into the builder.
452 llvm::Error loadModuleFile(FileEntryRef File);
453
454 /// Write the index to the given bitstream.
455 /// \returns true if an error occurred, false otherwise.
456 bool writeIndex(llvm::BitstreamWriter &Stream);
457 };
458}
459
460static void emitBlockID(unsigned ID, const char *Name,
461 llvm::BitstreamWriter &Stream,
462 SmallVectorImpl<uint64_t> &Record) {
463 Record.clear();
464 Record.push_back(Elt: ID);
465 Stream.EmitRecord(Code: llvm::bitc::BLOCKINFO_CODE_SETBID, Vals: Record);
466
467 // Emit the block name if present.
468 if (!Name || Name[0] == 0) return;
469 Record.clear();
470 while (*Name)
471 Record.push_back(Elt: *Name++);
472 Stream.EmitRecord(Code: llvm::bitc::BLOCKINFO_CODE_BLOCKNAME, Vals: Record);
473}
474
475static void emitRecordID(unsigned ID, const char *Name,
476 llvm::BitstreamWriter &Stream,
477 SmallVectorImpl<uint64_t> &Record) {
478 Record.clear();
479 Record.push_back(Elt: ID);
480 while (*Name)
481 Record.push_back(Elt: *Name++);
482 Stream.EmitRecord(Code: llvm::bitc::BLOCKINFO_CODE_SETRECORDNAME, Vals: Record);
483}
484
485void
486GlobalModuleIndexBuilder::emitBlockInfoBlock(llvm::BitstreamWriter &Stream) {
487 SmallVector<uint64_t, 64> Record;
488 Stream.EnterBlockInfoBlock();
489
490#define BLOCK(X) emitBlockID(X ## _ID, #X, Stream, Record)
491#define RECORD(X) emitRecordID(X, #X, Stream, Record)
492 BLOCK(GLOBAL_INDEX_BLOCK);
493 RECORD(INDEX_METADATA);
494 RECORD(MODULE);
495 RECORD(IDENTIFIER_INDEX);
496#undef RECORD
497#undef BLOCK
498
499 Stream.ExitBlock();
500}
501
502namespace {
503 class InterestingASTIdentifierLookupTrait
504 : public serialization::reader::ASTIdentifierLookupTraitBase {
505
506 public:
507 /// The identifier and whether it is "interesting".
508 typedef std::pair<StringRef, bool> data_type;
509
510 data_type ReadData(const internal_key_type& k,
511 const unsigned char* d,
512 unsigned DataLen) {
513 // The first bit indicates whether this identifier is interesting.
514 // That's all we care about.
515 using namespace llvm::support;
516 IdentifierID RawID =
517 endian::readNext<IdentifierID, llvm::endianness::little>(memory&: d);
518 bool IsInteresting = RawID & 0x01;
519 return std::make_pair(x: k, y&: IsInteresting);
520 }
521 };
522}
523
524llvm::Error GlobalModuleIndexBuilder::loadModuleFile(FileEntryRef File) {
525 // Open the module file.
526
527 auto Buffer = FileMgr.getBufferForFile(Entry: File, /*isVolatile=*/true);
528 if (!Buffer)
529 return llvm::createStringError(EC: Buffer.getError(),
530 S: "failed getting buffer for module file");
531
532 // Initialize the input stream
533 llvm::BitstreamCursor InStream(PCHContainerRdr.ExtractPCH(Buffer: **Buffer));
534
535 // Sniff for the signature.
536 for (unsigned char C : {'C', 'P', 'C', 'H'})
537 if (Expected<llvm::SimpleBitstreamCursor::word_t> Res = InStream.Read(NumBits: 8)) {
538 if (Res.get() != C)
539 return llvm::createStringError(EC: std::errc::illegal_byte_sequence,
540 Fmt: "expected signature CPCH");
541 } else
542 return Res.takeError();
543
544 // Record this module file and assign it a unique ID (if it doesn't have
545 // one already).
546 unsigned ID = getModuleFileInfo(File).ID;
547
548 // Search for the blocks and records we care about.
549 enum { Other, ControlBlock, ASTBlock, DiagnosticOptionsBlock } State = Other;
550 bool Done = false;
551 while (!Done) {
552 Expected<llvm::BitstreamEntry> MaybeEntry = InStream.advance();
553 if (!MaybeEntry)
554 return MaybeEntry.takeError();
555 llvm::BitstreamEntry Entry = MaybeEntry.get();
556
557 switch (Entry.Kind) {
558 case llvm::BitstreamEntry::Error:
559 Done = true;
560 continue;
561
562 case llvm::BitstreamEntry::Record:
563 // In the 'other' state, just skip the record. We don't care.
564 if (State == Other) {
565 if (llvm::Expected<unsigned> Skipped = InStream.skipRecord(AbbrevID: Entry.ID))
566 continue;
567 else
568 return Skipped.takeError();
569 }
570
571 // Handle potentially-interesting records below.
572 break;
573
574 case llvm::BitstreamEntry::SubBlock:
575 if (Entry.ID == CONTROL_BLOCK_ID) {
576 if (llvm::Error Err = InStream.EnterSubBlock(BlockID: CONTROL_BLOCK_ID))
577 return Err;
578
579 // Found the control block.
580 State = ControlBlock;
581 continue;
582 }
583
584 if (Entry.ID == AST_BLOCK_ID) {
585 if (llvm::Error Err = InStream.EnterSubBlock(BlockID: AST_BLOCK_ID))
586 return Err;
587
588 // Found the AST block.
589 State = ASTBlock;
590 continue;
591 }
592
593 if (Entry.ID == UNHASHED_CONTROL_BLOCK_ID) {
594 if (llvm::Error Err = InStream.EnterSubBlock(BlockID: UNHASHED_CONTROL_BLOCK_ID))
595 return Err;
596
597 // Found the Diagnostic Options block.
598 State = DiagnosticOptionsBlock;
599 continue;
600 }
601
602 if (llvm::Error Err = InStream.SkipBlock())
603 return Err;
604
605 continue;
606
607 case llvm::BitstreamEntry::EndBlock:
608 State = Other;
609 continue;
610 }
611
612 // Read the given record.
613 SmallVector<uint64_t, 64> Record;
614 StringRef Blob;
615 Expected<unsigned> MaybeCode = InStream.readRecord(AbbrevID: Entry.ID, Vals&: Record, Blob: &Blob);
616 if (!MaybeCode)
617 return MaybeCode.takeError();
618 unsigned Code = MaybeCode.get();
619
620 // Handle module dependencies.
621 if (State == ControlBlock && Code == IMPORT) {
622 unsigned Idx = 0;
623 // Read information about the AST file.
624
625 // Skip the imported kind
626 ++Idx;
627
628 // Skip the import location
629 ++Idx;
630
631 // Skip the module name (currently this is only used for prebuilt
632 // modules while here we are only dealing with cached).
633 Blob = Blob.substr(Start: Record[Idx++]);
634
635 // Skip if it is standard C++ module
636 ++Idx;
637
638 // Load stored size/modification time.
639 off_t StoredSize = (off_t)Record[Idx++];
640 time_t StoredModTime = (time_t)Record[Idx++];
641
642 // Skip the stored signature.
643 // FIXME: we could read the signature out of the import and validate it.
644 StringRef SignatureBytes = Blob.substr(Start: 0, N: ASTFileSignature::size);
645 auto StoredSignature = ASTFileSignature::create(First: SignatureBytes.begin(),
646 Last: SignatureBytes.end());
647 Blob = Blob.substr(Start: ASTFileSignature::size);
648
649 // Retrieve the imported file name.
650 unsigned Length = Record[Idx++];
651 StringRef ImportedFile = Blob.substr(Start: 0, N: Length);
652 Blob = Blob.substr(Start: Length);
653
654 // Find the imported module file.
655 auto DependsOnFile =
656 FileMgr.getOptionalFileRef(Filename: ImportedFile, /*OpenFile=*/false,
657 /*CacheFailure=*/false);
658
659 if (!DependsOnFile)
660 return llvm::createStringError(EC: std::errc::bad_file_descriptor,
661 Fmt: "imported file \"%s\" not found",
662 Vals: std::string(ImportedFile).c_str());
663
664 // Save the information in ImportedModuleFileInfo so we can verify after
665 // loading all pcms.
666 ImportedModuleFiles.insert(x: std::make_pair(
667 x&: *DependsOnFile, y: ImportedModuleFileInfo(StoredSize, StoredModTime,
668 StoredSignature)));
669
670 // Record the dependency.
671 unsigned DependsOnID = getModuleFileInfo(File: *DependsOnFile).ID;
672 getModuleFileInfo(File).Dependencies.push_back(Elt: DependsOnID);
673
674 continue;
675 }
676
677 // Handle the identifier table
678 if (State == ASTBlock && Code == IDENTIFIER_TABLE && Record[0] > 0) {
679 typedef llvm::OnDiskIterableChainedHashTable<
680 InterestingASTIdentifierLookupTrait> InterestingIdentifierTable;
681 std::unique_ptr<InterestingIdentifierTable> Table(
682 InterestingIdentifierTable::Create(
683 Buckets: (const unsigned char *)Blob.data() + Record[0],
684 Payload: (const unsigned char *)Blob.data() + sizeof(uint32_t),
685 Base: (const unsigned char *)Blob.data()));
686 for (InterestingIdentifierTable::data_iterator D = Table->data_begin(),
687 DEnd = Table->data_end();
688 D != DEnd; ++D) {
689 std::pair<StringRef, bool> Ident = *D;
690 if (Ident.second)
691 InterestingIdentifiers[Ident.first].push_back(Elt: ID);
692 else
693 (void)InterestingIdentifiers[Ident.first];
694 }
695 }
696
697 // Get Signature.
698 if (State == DiagnosticOptionsBlock && Code == SIGNATURE) {
699 auto Signature = ASTFileSignature::create(First: Blob.begin(), Last: Blob.end());
700 assert(Signature != ASTFileSignature::createDummy() &&
701 "Dummy AST file signature not backpatched in ASTWriter.");
702 getModuleFileInfo(File).Signature = Signature;
703 }
704
705 // We don't care about this record.
706 }
707
708 return llvm::Error::success();
709}
710
711namespace {
712
713/// Trait used to generate the identifier index as an on-disk hash
714/// table.
715class IdentifierIndexWriterTrait {
716public:
717 typedef StringRef key_type;
718 typedef StringRef key_type_ref;
719 typedef SmallVector<unsigned, 2> data_type;
720 typedef const SmallVector<unsigned, 2> &data_type_ref;
721 typedef unsigned hash_value_type;
722 typedef unsigned offset_type;
723
724 static hash_value_type ComputeHash(key_type_ref Key) {
725 return llvm::djbHash(Buffer: Key);
726 }
727
728 std::pair<unsigned,unsigned>
729 EmitKeyDataLength(raw_ostream& Out, key_type_ref Key, data_type_ref Data) {
730 using namespace llvm::support;
731 endian::Writer LE(Out, llvm::endianness::little);
732 unsigned KeyLen = Key.size();
733 unsigned DataLen = Data.size() * 4;
734 LE.write<uint16_t>(Val: KeyLen);
735 LE.write<uint16_t>(Val: DataLen);
736 return std::make_pair(x&: KeyLen, y&: DataLen);
737 }
738
739 void EmitKey(raw_ostream& Out, key_type_ref Key, unsigned KeyLen) {
740 Out.write(Ptr: Key.data(), Size: KeyLen);
741 }
742
743 void EmitData(raw_ostream& Out, key_type_ref Key, data_type_ref Data,
744 unsigned DataLen) {
745 using namespace llvm::support;
746 for (unsigned I = 0, N = Data.size(); I != N; ++I)
747 endian::write<uint32_t>(os&: Out, value: Data[I], endian: llvm::endianness::little);
748 }
749};
750
751}
752
753bool GlobalModuleIndexBuilder::writeIndex(llvm::BitstreamWriter &Stream) {
754 for (auto MapEntry : ImportedModuleFiles) {
755 auto File = MapEntry.first;
756 ImportedModuleFileInfo &Info = MapEntry.second;
757 if (getModuleFileInfo(File).Signature) {
758 if (getModuleFileInfo(File).Signature != Info.StoredSignature)
759 // Verify Signature.
760 return true;
761 } else if (Info.StoredSize != File.getSize() ||
762 Info.StoredModTime != File.getModificationTime())
763 // Verify Size and ModTime.
764 return true;
765 }
766
767 using namespace llvm;
768 llvm::TimeTraceScope TimeScope("Module WriteIndex");
769
770 // Emit the file header.
771 Stream.Emit(Val: (unsigned)'B', NumBits: 8);
772 Stream.Emit(Val: (unsigned)'C', NumBits: 8);
773 Stream.Emit(Val: (unsigned)'G', NumBits: 8);
774 Stream.Emit(Val: (unsigned)'I', NumBits: 8);
775
776 // Write the block-info block, which describes the records in this bitcode
777 // file.
778 emitBlockInfoBlock(Stream);
779
780 Stream.EnterSubblock(BlockID: GLOBAL_INDEX_BLOCK_ID, CodeLen: 3);
781
782 // Write the metadata.
783 SmallVector<uint64_t, 2> Record;
784 Record.push_back(Elt: CurrentVersion);
785 Stream.EmitRecord(Code: INDEX_METADATA, Vals: Record);
786
787 // Write the set of known module files.
788 for (ModuleFilesMap::iterator M = ModuleFiles.begin(),
789 MEnd = ModuleFiles.end();
790 M != MEnd; ++M) {
791 Record.clear();
792 Record.push_back(Elt: M->second.ID);
793 Record.push_back(Elt: M->first.getSize());
794 Record.push_back(Elt: M->first.getModificationTime());
795
796 // File name
797 StringRef Name(M->first.getName());
798 Record.push_back(Elt: Name.size());
799 Record.append(in_start: Name.begin(), in_end: Name.end());
800
801 // Dependencies
802 Record.push_back(Elt: M->second.Dependencies.size());
803 Record.append(in_start: M->second.Dependencies.begin(), in_end: M->second.Dependencies.end());
804 Stream.EmitRecord(Code: MODULE, Vals: Record);
805 }
806
807 // Write the identifier -> module file mapping.
808 {
809 llvm::OnDiskChainedHashTableGenerator<IdentifierIndexWriterTrait> Generator;
810 IdentifierIndexWriterTrait Trait;
811
812 // Populate the hash table.
813 for (InterestingIdentifierMap::iterator I = InterestingIdentifiers.begin(),
814 IEnd = InterestingIdentifiers.end();
815 I != IEnd; ++I) {
816 Generator.insert(Key: I->first(), Data: I->second, InfoObj&: Trait);
817 }
818
819 // Create the on-disk hash table in a buffer.
820 SmallString<4096> IdentifierTable;
821 uint32_t BucketOffset;
822 {
823 using namespace llvm::support;
824 llvm::raw_svector_ostream Out(IdentifierTable);
825 // Make sure that no bucket is at offset 0
826 endian::write<uint32_t>(os&: Out, value: 0, endian: llvm::endianness::little);
827 BucketOffset = Generator.Emit(Out, InfoObj&: Trait);
828 }
829
830 // Create a blob abbreviation
831 auto Abbrev = std::make_shared<BitCodeAbbrev>();
832 Abbrev->Add(OpInfo: BitCodeAbbrevOp(IDENTIFIER_INDEX));
833 Abbrev->Add(OpInfo: BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32));
834 Abbrev->Add(OpInfo: BitCodeAbbrevOp(BitCodeAbbrevOp::Blob));
835 unsigned IDTableAbbrev = Stream.EmitAbbrev(Abbv: std::move(Abbrev));
836
837 // Write the identifier table
838 uint64_t Record[] = {IDENTIFIER_INDEX, BucketOffset};
839 Stream.EmitRecordWithBlob(Abbrev: IDTableAbbrev, Vals: Record, Blob: IdentifierTable);
840 }
841
842 Stream.ExitBlock();
843 return false;
844}
845
846llvm::Error
847GlobalModuleIndex::writeIndex(FileManager &FileMgr,
848 const PCHContainerReader &PCHContainerRdr,
849 StringRef Path) {
850 // This is a compiler-internal input/output, let's bypass the sandbox.
851 auto BypassSandbox = llvm::sys::sandbox::scopedDisable();
852
853 llvm::SmallString<128> IndexPath;
854 IndexPath += Path;
855 llvm::sys::path::append(path&: IndexPath, a: IndexFileName);
856
857 // Coordinate building the global index file with other processes that might
858 // try to do the same.
859 llvm::LockFileManager Lock(IndexPath);
860 bool Owned;
861 if (llvm::Error Err = Lock.tryLock().moveInto(Value&: Owned)) {
862 llvm::consumeError(Err: std::move(Err));
863 return llvm::createStringError(EC: std::errc::io_error, Fmt: "LFS error");
864 }
865 if (!Owned) {
866 // Someone else is responsible for building the index. We don't care
867 // when they finish, so we're done.
868 return llvm::createStringError(EC: std::errc::device_or_resource_busy,
869 Fmt: "someone else is building the index");
870 }
871
872 // We're responsible for building the index ourselves.
873
874 // The module index builder.
875 GlobalModuleIndexBuilder Builder(FileMgr, PCHContainerRdr);
876
877 // Load each of the module files.
878 std::error_code EC;
879 for (llvm::sys::fs::directory_iterator D(Path, EC), DEnd;
880 D != DEnd && !EC;
881 D.increment(ec&: EC)) {
882 // If this isn't a module file, we don't care.
883 if (llvm::sys::path::extension(path: D->path()) != ".pcm") {
884 // ... unless it's a .pcm.lock file, which indicates that someone is
885 // in the process of rebuilding a module. They'll rebuild the index
886 // at the end of that translation unit, so we don't have to.
887 if (llvm::sys::path::extension(path: D->path()) == ".pcm.lock")
888 return llvm::createStringError(EC: std::errc::device_or_resource_busy,
889 Fmt: "someone else is building the index");
890
891 continue;
892 }
893
894 // If we can't find the module file, skip it.
895 auto ModuleFile = FileMgr.getOptionalFileRef(Filename: D->path());
896 if (!ModuleFile)
897 continue;
898
899 // Load this module file.
900 if (llvm::Error Err = Builder.loadModuleFile(File: *ModuleFile))
901 return Err;
902 }
903
904 // The output buffer, into which the global index will be written.
905 SmallString<16> OutputBuffer;
906 {
907 llvm::BitstreamWriter OutputStream(OutputBuffer);
908 if (Builder.writeIndex(Stream&: OutputStream))
909 return llvm::createStringError(EC: std::errc::io_error,
910 Fmt: "failed writing index");
911 }
912
913 return llvm::writeToOutput(OutputFileName: IndexPath, Write: [&OutputBuffer](llvm::raw_ostream &OS) {
914 OS << OutputBuffer;
915 return llvm::Error::success();
916 });
917}
918
919namespace {
920 class GlobalIndexIdentifierIterator : public IdentifierIterator {
921 /// The current position within the identifier lookup table.
922 IdentifierIndexTable::key_iterator Current;
923
924 /// The end position within the identifier lookup table.
925 IdentifierIndexTable::key_iterator End;
926
927 public:
928 explicit GlobalIndexIdentifierIterator(IdentifierIndexTable &Idx) {
929 Current = Idx.key_begin();
930 End = Idx.key_end();
931 }
932
933 StringRef Next() override {
934 if (Current == End)
935 return StringRef();
936
937 StringRef Result = *Current;
938 ++Current;
939 return Result;
940 }
941 };
942}
943
944IdentifierIterator *GlobalModuleIndex::createIdentifierIterator() const {
945 IdentifierIndexTable &Table =
946 *static_cast<IdentifierIndexTable *>(IdentifierIndex);
947 return new GlobalIndexIdentifierIterator(Table);
948}
949