1 //===--- SourceManager.cpp - Track and cache source files -----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the SourceManager interface.
12 //===----------------------------------------------------------------------===//
14 #include "clang/Basic/SourceManager.h"
15 #include "clang/Basic/Diagnostic.h"
16 #include "clang/Basic/FileManager.h"
17 #include "clang/Basic/SourceManagerInternals.h"
18 #include "llvm/ADT/Optional.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/ADT/StringSwitch.h"
21 #include "llvm/Support/Capacity.h"
22 #include "llvm/Support/Compiler.h"
23 #include "llvm/Support/MemoryBuffer.h"
24 #include "llvm/Support/Path.h"
25 #include "llvm/Support/raw_ostream.h"
29 using namespace clang;
30 using namespace SrcMgr;
31 using llvm::MemoryBuffer;
33 //===----------------------------------------------------------------------===//
34 // SourceManager Helper Classes
35 //===----------------------------------------------------------------------===//
37 ContentCache::~ContentCache() {
38 if (shouldFreeBuffer())
39 delete Buffer.getPointer();
42 /// getSizeBytesMapped - Returns the number of bytes actually mapped for this
43 /// ContentCache. This can be 0 if the MemBuffer was not actually expanded.
44 unsigned ContentCache::getSizeBytesMapped() const {
45 return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0;
48 /// Returns the kind of memory used to back the memory buffer for
49 /// this content cache. This is used for performance analysis.
50 llvm::MemoryBuffer::BufferKind ContentCache::getMemoryBufferKind() const {
51 assert(Buffer.getPointer());
53 // Should be unreachable, but keep for sanity.
54 if (!Buffer.getPointer())
55 return llvm::MemoryBuffer::MemoryBuffer_Malloc;
57 llvm::MemoryBuffer *buf = Buffer.getPointer();
58 return buf->getBufferKind();
61 /// getSize - Returns the size of the content encapsulated by this ContentCache.
62 /// This can be the size of the source file or the size of an arbitrary
63 /// scratch buffer. If the ContentCache encapsulates a source file, that
64 /// file is not lazily brought in from disk to satisfy this query.
65 unsigned ContentCache::getSize() const {
66 return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize()
67 : (unsigned) ContentsEntry->getSize();
70 void ContentCache::replaceBuffer(llvm::MemoryBuffer *B, bool DoNotFree) {
71 if (B && B == Buffer.getPointer()) {
72 assert(0 && "Replacing with the same buffer");
73 Buffer.setInt(DoNotFree? DoNotFreeFlag : 0);
77 if (shouldFreeBuffer())
78 delete Buffer.getPointer();
80 Buffer.setInt((B && DoNotFree) ? DoNotFreeFlag : 0);
83 llvm::MemoryBuffer *ContentCache::getBuffer(DiagnosticsEngine &Diag,
84 const SourceManager &SM,
86 bool *Invalid) const {
87 // Lazily create the Buffer for ContentCaches that wrap files. If we already
88 // computed it, just return what we have.
89 if (Buffer.getPointer() || !ContentsEntry) {
91 *Invalid = isBufferInvalid();
93 return Buffer.getPointer();
96 bool isVolatile = SM.userFilesAreVolatile() && !IsSystemFile;
98 SM.getFileManager().getBufferForFile(ContentsEntry, isVolatile);
100 // If we were unable to open the file, then we are in an inconsistent
101 // situation where the content cache referenced a file which no longer
102 // exists. Most likely, we were using a stat cache with an invalid entry but
103 // the file could also have been removed during processing. Since we can't
104 // really deal with this situation, just create an empty buffer.
106 // FIXME: This is definitely not ideal, but our immediate clients can't
107 // currently handle returning a null entry here. Ideally we should detect
108 // that we are in an inconsistent situation and error out as quickly as
110 if (!BufferOrError) {
111 StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
112 Buffer.setPointer(MemoryBuffer::getNewUninitMemBuffer(
113 ContentsEntry->getSize(), "<invalid>").release());
114 char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
115 for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i)
116 Ptr[i] = FillStr[i % FillStr.size()];
118 if (Diag.isDiagnosticInFlight())
119 Diag.SetDelayedDiagnostic(diag::err_cannot_open_file,
120 ContentsEntry->getName(),
121 BufferOrError.getError().message());
123 Diag.Report(Loc, diag::err_cannot_open_file)
124 << ContentsEntry->getName() << BufferOrError.getError().message();
126 Buffer.setInt(Buffer.getInt() | InvalidFlag);
128 if (Invalid) *Invalid = true;
129 return Buffer.getPointer();
132 Buffer.setPointer(BufferOrError->release());
134 // Check that the file's size is the same as in the file entry (which may
135 // have come from a stat cache).
136 if (getRawBuffer()->getBufferSize() != (size_t)ContentsEntry->getSize()) {
137 if (Diag.isDiagnosticInFlight())
138 Diag.SetDelayedDiagnostic(diag::err_file_modified,
139 ContentsEntry->getName());
141 Diag.Report(Loc, diag::err_file_modified)
142 << ContentsEntry->getName();
144 Buffer.setInt(Buffer.getInt() | InvalidFlag);
145 if (Invalid) *Invalid = true;
146 return Buffer.getPointer();
149 // If the buffer is valid, check to see if it has a UTF Byte Order Mark
150 // (BOM). We only support UTF-8 with and without a BOM right now. See
151 // http://en.wikipedia.org/wiki/Byte_order_mark for more information.
152 StringRef BufStr = Buffer.getPointer()->getBuffer();
153 const char *InvalidBOM = llvm::StringSwitch<const char *>(BufStr)
154 .StartsWith("\xFE\xFF", "UTF-16 (BE)")
155 .StartsWith("\xFF\xFE", "UTF-16 (LE)")
156 .StartsWith("\x00\x00\xFE\xFF", "UTF-32 (BE)")
157 .StartsWith("\xFF\xFE\x00\x00", "UTF-32 (LE)")
158 .StartsWith("\x2B\x2F\x76", "UTF-7")
159 .StartsWith("\xF7\x64\x4C", "UTF-1")
160 .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC")
161 .StartsWith("\x0E\xFE\xFF", "SDSU")
162 .StartsWith("\xFB\xEE\x28", "BOCU-1")
163 .StartsWith("\x84\x31\x95\x33", "GB-18030")
167 Diag.Report(Loc, diag::err_unsupported_bom)
168 << InvalidBOM << ContentsEntry->getName();
169 Buffer.setInt(Buffer.getInt() | InvalidFlag);
173 *Invalid = isBufferInvalid();
175 return Buffer.getPointer();
178 unsigned LineTableInfo::getLineTableFilenameID(StringRef Name) {
180 FilenameIDs.insert(std::make_pair(Name, FilenamesByID.size()));
182 FilenamesByID.push_back(&*IterBool.first);
183 return IterBool.first->second;
186 /// Add a line note to the line table that indicates that there is a \#line or
187 /// GNU line marker at the specified FID/Offset location which changes the
188 /// presumed location to LineNo/FilenameID. If EntryExit is 0, then this doesn't
189 /// change the presumed \#include stack. If it is 1, this is a file entry, if
190 /// it is 2 then this is a file exit. FileKind specifies whether this is a
191 /// system header or extern C system header.
192 void LineTableInfo::AddLineNote(FileID FID, unsigned Offset, unsigned LineNo,
193 int FilenameID, unsigned EntryExit,
194 SrcMgr::CharacteristicKind FileKind) {
195 std::vector<LineEntry> &Entries = LineEntries[FID];
197 // An unspecified FilenameID means use the last filename if available, or the
198 // main source file otherwise.
199 if (FilenameID == -1 && !Entries.empty())
200 FilenameID = Entries.back().FilenameID;
202 assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
203 "Adding line entries out of order!");
205 unsigned IncludeOffset = 0;
206 if (EntryExit == 0) { // No #include stack change.
207 IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset;
208 } else if (EntryExit == 1) {
209 IncludeOffset = Offset-1;
210 } else if (EntryExit == 2) {
211 assert(!Entries.empty() && Entries.back().IncludeOffset &&
212 "PPDirectives should have caught case when popping empty include stack");
214 // Get the include loc of the last entries' include loc as our include loc.
216 if (const LineEntry *PrevEntry =
217 FindNearestLineEntry(FID, Entries.back().IncludeOffset))
218 IncludeOffset = PrevEntry->IncludeOffset;
221 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind,
226 /// FindNearestLineEntry - Find the line entry nearest to FID that is before
227 /// it. If there is no line entry before Offset in FID, return null.
228 const LineEntry *LineTableInfo::FindNearestLineEntry(FileID FID,
230 const std::vector<LineEntry> &Entries = LineEntries[FID];
231 assert(!Entries.empty() && "No #line entries for this FID after all!");
233 // It is very common for the query to be after the last #line, check this
235 if (Entries.back().FileOffset <= Offset)
236 return &Entries.back();
238 // Do a binary search to find the maximal element that is still before Offset.
239 std::vector<LineEntry>::const_iterator I =
240 std::upper_bound(Entries.begin(), Entries.end(), Offset);
241 if (I == Entries.begin()) return nullptr;
245 /// \brief Add a new line entry that has already been encoded into
246 /// the internal representation of the line table.
247 void LineTableInfo::AddEntry(FileID FID,
248 const std::vector<LineEntry> &Entries) {
249 LineEntries[FID] = Entries;
252 /// getLineTableFilenameID - Return the uniqued ID for the specified filename.
254 unsigned SourceManager::getLineTableFilenameID(StringRef Name) {
255 return getLineTable().getLineTableFilenameID(Name);
258 /// AddLineNote - Add a line note to the line table for the FileID and offset
259 /// specified by Loc. If FilenameID is -1, it is considered to be
261 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
262 int FilenameID, bool IsFileEntry,
264 SrcMgr::CharacteristicKind FileKind) {
265 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
267 bool Invalid = false;
268 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
269 if (!Entry.isFile() || Invalid)
272 const SrcMgr::FileInfo &FileInfo = Entry.getFile();
274 // Remember that this file has #line directives now if it doesn't already.
275 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
277 (void) getLineTable();
279 unsigned EntryExit = 0;
285 LineTable->AddLineNote(LocInfo.first, LocInfo.second, LineNo, FilenameID,
286 EntryExit, FileKind);
289 LineTableInfo &SourceManager::getLineTable() {
291 LineTable = new LineTableInfo();
295 //===----------------------------------------------------------------------===//
296 // Private 'Create' methods.
297 //===----------------------------------------------------------------------===//
299 SourceManager::SourceManager(DiagnosticsEngine &Diag, FileManager &FileMgr,
300 bool UserFilesAreVolatile)
301 : Diag(Diag), FileMgr(FileMgr), OverridenFilesKeepOriginalName(true),
302 UserFilesAreVolatile(UserFilesAreVolatile), FilesAreTransient(false),
303 ExternalSLocEntries(nullptr), LineTable(nullptr), NumLinearScans(0),
306 Diag.setSourceManager(this);
309 SourceManager::~SourceManager() {
312 // Delete FileEntry objects corresponding to content caches. Since the actual
313 // content cache objects are bump pointer allocated, we just have to run the
314 // dtors, but we call the deallocate method for completeness.
315 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) {
316 if (MemBufferInfos[i]) {
317 MemBufferInfos[i]->~ContentCache();
318 ContentCacheAlloc.Deallocate(MemBufferInfos[i]);
321 for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator
322 I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) {
324 I->second->~ContentCache();
325 ContentCacheAlloc.Deallocate(I->second);
330 void SourceManager::clearIDTables() {
331 MainFileID = FileID();
332 LocalSLocEntryTable.clear();
333 LoadedSLocEntryTable.clear();
334 SLocEntryLoaded.clear();
335 LastLineNoFileIDQuery = FileID();
336 LastLineNoContentCache = nullptr;
337 LastFileIDLookup = FileID();
342 // Use up FileID #0 as an invalid expansion.
344 CurrentLoadedOffset = MaxLoadedOffset;
345 createExpansionLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1);
348 void SourceManager::initializeForReplay(const SourceManager &Old) {
349 assert(MainFileID.isInvalid() && "expected uninitialized SourceManager");
351 auto CloneContentCache = [&](const ContentCache *Cache) -> ContentCache * {
352 auto *Clone = new (ContentCacheAlloc.Allocate<ContentCache>()) ContentCache;
353 Clone->OrigEntry = Cache->OrigEntry;
354 Clone->ContentsEntry = Cache->ContentsEntry;
355 Clone->BufferOverridden = Cache->BufferOverridden;
356 Clone->IsSystemFile = Cache->IsSystemFile;
357 Clone->IsTransient = Cache->IsTransient;
358 Clone->replaceBuffer(Cache->getRawBuffer(), /*DoNotFree*/true);
362 // Ensure all SLocEntries are loaded from the external source.
363 for (unsigned I = 0, N = Old.LoadedSLocEntryTable.size(); I != N; ++I)
364 if (!Old.SLocEntryLoaded[I])
365 Old.loadSLocEntry(I, nullptr);
367 // Inherit any content cache data from the old source manager.
368 for (auto &FileInfo : Old.FileInfos) {
369 SrcMgr::ContentCache *&Slot = FileInfos[FileInfo.first];
372 Slot = CloneContentCache(FileInfo.second);
376 /// getOrCreateContentCache - Create or return a cached ContentCache for the
379 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt,
381 assert(FileEnt && "Didn't specify a file entry to use?");
383 // Do we already have information about this file?
384 ContentCache *&Entry = FileInfos[FileEnt];
385 if (Entry) return Entry;
387 // Nope, create a new Cache entry.
388 Entry = ContentCacheAlloc.Allocate<ContentCache>();
390 if (OverriddenFilesInfo) {
391 // If the file contents are overridden with contents from another file,
392 // pass that file to ContentCache.
393 llvm::DenseMap<const FileEntry *, const FileEntry *>::iterator
394 overI = OverriddenFilesInfo->OverriddenFiles.find(FileEnt);
395 if (overI == OverriddenFilesInfo->OverriddenFiles.end())
396 new (Entry) ContentCache(FileEnt);
398 new (Entry) ContentCache(OverridenFilesKeepOriginalName ? FileEnt
402 new (Entry) ContentCache(FileEnt);
405 Entry->IsSystemFile = isSystemFile;
406 Entry->IsTransient = FilesAreTransient;
412 /// createMemBufferContentCache - Create a new ContentCache for the specified
413 /// memory buffer. This does no caching.
414 const ContentCache *SourceManager::createMemBufferContentCache(
415 std::unique_ptr<llvm::MemoryBuffer> Buffer) {
416 // Add a new ContentCache to the MemBufferInfos list and return it.
417 ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>();
418 new (Entry) ContentCache();
419 MemBufferInfos.push_back(Entry);
420 Entry->setBuffer(std::move(Buffer));
424 const SrcMgr::SLocEntry &SourceManager::loadSLocEntry(unsigned Index,
425 bool *Invalid) const {
426 assert(!SLocEntryLoaded[Index]);
427 if (ExternalSLocEntries->ReadSLocEntry(-(static_cast<int>(Index) + 2))) {
430 // If the file of the SLocEntry changed we could still have loaded it.
431 if (!SLocEntryLoaded[Index]) {
432 // Try to recover; create a SLocEntry so the rest of clang can handle it.
433 LoadedSLocEntryTable[Index] = SLocEntry::get(0,
434 FileInfo::get(SourceLocation(),
435 getFakeContentCacheForRecovery(),
440 return LoadedSLocEntryTable[Index];
443 std::pair<int, unsigned>
444 SourceManager::AllocateLoadedSLocEntries(unsigned NumSLocEntries,
445 unsigned TotalSize) {
446 assert(ExternalSLocEntries && "Don't have an external sloc source");
447 // Make sure we're not about to run out of source locations.
448 if (CurrentLoadedOffset - TotalSize < NextLocalOffset)
449 return std::make_pair(0, 0);
450 LoadedSLocEntryTable.resize(LoadedSLocEntryTable.size() + NumSLocEntries);
451 SLocEntryLoaded.resize(LoadedSLocEntryTable.size());
452 CurrentLoadedOffset -= TotalSize;
453 int ID = LoadedSLocEntryTable.size();
454 return std::make_pair(-ID - 1, CurrentLoadedOffset);
457 /// \brief As part of recovering from missing or changed content, produce a
458 /// fake, non-empty buffer.
459 llvm::MemoryBuffer *SourceManager::getFakeBufferForRecovery() const {
460 if (!FakeBufferForRecovery)
461 FakeBufferForRecovery =
462 llvm::MemoryBuffer::getMemBuffer("<<<INVALID BUFFER>>");
464 return FakeBufferForRecovery.get();
467 /// \brief As part of recovering from missing or changed content, produce a
468 /// fake content cache.
469 const SrcMgr::ContentCache *
470 SourceManager::getFakeContentCacheForRecovery() const {
471 if (!FakeContentCacheForRecovery) {
472 FakeContentCacheForRecovery = llvm::make_unique<SrcMgr::ContentCache>();
473 FakeContentCacheForRecovery->replaceBuffer(getFakeBufferForRecovery(),
476 return FakeContentCacheForRecovery.get();
479 /// \brief Returns the previous in-order FileID or an invalid FileID if there
480 /// is no previous one.
481 FileID SourceManager::getPreviousFileID(FileID FID) const {
492 } else if (unsigned(-(ID-1) - 2) >= LoadedSLocEntryTable.size()) {
496 return FileID::get(ID-1);
499 /// \brief Returns the next in-order FileID or an invalid FileID if there is
501 FileID SourceManager::getNextFileID(FileID FID) const {
507 if (unsigned(ID+1) >= local_sloc_entry_size())
509 } else if (ID+1 >= -1) {
513 return FileID::get(ID+1);
516 //===----------------------------------------------------------------------===//
517 // Methods to create new FileID's and macro expansions.
518 //===----------------------------------------------------------------------===//
520 /// createFileID - Create a new FileID for the specified ContentCache and
521 /// include position. This works regardless of whether the ContentCache
522 /// corresponds to a file or some other input source.
523 FileID SourceManager::createFileID(const ContentCache *File,
524 SourceLocation IncludePos,
525 SrcMgr::CharacteristicKind FileCharacter,
526 int LoadedID, unsigned LoadedOffset) {
528 assert(LoadedID != -1 && "Loading sentinel FileID");
529 unsigned Index = unsigned(-LoadedID) - 2;
530 assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
531 assert(!SLocEntryLoaded[Index] && "FileID already loaded");
532 LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset,
533 FileInfo::get(IncludePos, File, FileCharacter));
534 SLocEntryLoaded[Index] = true;
535 return FileID::get(LoadedID);
537 LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset,
538 FileInfo::get(IncludePos, File,
540 unsigned FileSize = File->getSize();
541 assert(NextLocalOffset + FileSize + 1 > NextLocalOffset &&
542 NextLocalOffset + FileSize + 1 <= CurrentLoadedOffset &&
543 "Ran out of source locations!");
544 // We do a +1 here because we want a SourceLocation that means "the end of the
545 // file", e.g. for the "no newline at the end of the file" diagnostic.
546 NextLocalOffset += FileSize + 1;
548 // Set LastFileIDLookup to the newly created file. The next getFileID call is
549 // almost guaranteed to be from that file.
550 FileID FID = FileID::get(LocalSLocEntryTable.size()-1);
551 return LastFileIDLookup = FID;
555 SourceManager::createMacroArgExpansionLoc(SourceLocation SpellingLoc,
556 SourceLocation ExpansionLoc,
557 unsigned TokLength) {
558 ExpansionInfo Info = ExpansionInfo::createForMacroArg(SpellingLoc,
560 return createExpansionLocImpl(Info, TokLength);
564 SourceManager::createExpansionLoc(SourceLocation SpellingLoc,
565 SourceLocation ExpansionLocStart,
566 SourceLocation ExpansionLocEnd,
569 unsigned LoadedOffset) {
570 ExpansionInfo Info = ExpansionInfo::create(SpellingLoc, ExpansionLocStart,
572 return createExpansionLocImpl(Info, TokLength, LoadedID, LoadedOffset);
576 SourceManager::createExpansionLocImpl(const ExpansionInfo &Info,
579 unsigned LoadedOffset) {
581 assert(LoadedID != -1 && "Loading sentinel FileID");
582 unsigned Index = unsigned(-LoadedID) - 2;
583 assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
584 assert(!SLocEntryLoaded[Index] && "FileID already loaded");
585 LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, Info);
586 SLocEntryLoaded[Index] = true;
587 return SourceLocation::getMacroLoc(LoadedOffset);
589 LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, Info));
590 assert(NextLocalOffset + TokLength + 1 > NextLocalOffset &&
591 NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset &&
592 "Ran out of source locations!");
593 // See createFileID for that +1.
594 NextLocalOffset += TokLength + 1;
595 return SourceLocation::getMacroLoc(NextLocalOffset - (TokLength + 1));
598 llvm::MemoryBuffer *SourceManager::getMemoryBufferForFile(const FileEntry *File,
600 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
601 assert(IR && "getOrCreateContentCache() cannot return NULL");
602 return IR->getBuffer(Diag, *this, SourceLocation(), Invalid);
605 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
606 llvm::MemoryBuffer *Buffer,
608 const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile);
609 assert(IR && "getOrCreateContentCache() cannot return NULL");
611 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree);
612 const_cast<SrcMgr::ContentCache *>(IR)->BufferOverridden = true;
614 getOverriddenFilesInfo().OverriddenFilesWithBuffer.insert(SourceFile);
617 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
618 const FileEntry *NewFile) {
619 assert(SourceFile->getSize() == NewFile->getSize() &&
620 "Different sizes, use the FileManager to create a virtual file with "
622 assert(FileInfos.count(SourceFile) == 0 &&
623 "This function should be called at the initialization stage, before "
624 "any parsing occurs.");
625 getOverriddenFilesInfo().OverriddenFiles[SourceFile] = NewFile;
628 void SourceManager::disableFileContentsOverride(const FileEntry *File) {
629 if (!isFileOverridden(File))
632 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
633 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(nullptr);
634 const_cast<SrcMgr::ContentCache *>(IR)->ContentsEntry = IR->OrigEntry;
636 assert(OverriddenFilesInfo);
637 OverriddenFilesInfo->OverriddenFiles.erase(File);
638 OverriddenFilesInfo->OverriddenFilesWithBuffer.erase(File);
641 void SourceManager::setFileIsTransient(const FileEntry *File) {
642 const SrcMgr::ContentCache *CC = getOrCreateContentCache(File);
643 const_cast<SrcMgr::ContentCache *>(CC)->IsTransient = true;
646 StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const {
647 bool MyInvalid = false;
648 const SLocEntry &SLoc = getSLocEntry(FID, &MyInvalid);
649 if (!SLoc.isFile() || MyInvalid) {
652 return "<<<<<INVALID SOURCE LOCATION>>>>>";
655 llvm::MemoryBuffer *Buf = SLoc.getFile().getContentCache()->getBuffer(
656 Diag, *this, SourceLocation(), &MyInvalid);
658 *Invalid = MyInvalid;
661 return "<<<<<INVALID SOURCE LOCATION>>>>>";
663 return Buf->getBuffer();
666 //===----------------------------------------------------------------------===//
667 // SourceLocation manipulation methods.
668 //===----------------------------------------------------------------------===//
670 /// \brief Return the FileID for a SourceLocation.
672 /// This is the cache-miss path of getFileID. Not as hot as that function, but
673 /// still very important. It is responsible for finding the entry in the
674 /// SLocEntry tables that contains the specified location.
675 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const {
677 return FileID::get(0);
679 // Now it is time to search for the correct file. See where the SLocOffset
680 // sits in the global view and consult local or loaded buffers for it.
681 if (SLocOffset < NextLocalOffset)
682 return getFileIDLocal(SLocOffset);
683 return getFileIDLoaded(SLocOffset);
686 /// \brief Return the FileID for a SourceLocation with a low offset.
688 /// This function knows that the SourceLocation is in a local buffer, not a
690 FileID SourceManager::getFileIDLocal(unsigned SLocOffset) const {
691 assert(SLocOffset < NextLocalOffset && "Bad function choice");
693 // After the first and second level caches, I see two common sorts of
694 // behavior: 1) a lot of searched FileID's are "near" the cached file
695 // location or are "near" the cached expansion location. 2) others are just
696 // completely random and may be a very long way away.
698 // To handle this, we do a linear search for up to 8 steps to catch #1 quickly
699 // then we fall back to a less cache efficient, but more scalable, binary
700 // search to find the location.
702 // See if this is near the file point - worst case we start scanning from the
703 // most newly created FileID.
704 const SrcMgr::SLocEntry *I;
706 if (LastFileIDLookup.ID < 0 ||
707 LocalSLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) {
708 // Neither loc prunes our search.
709 I = LocalSLocEntryTable.end();
711 // Perhaps it is near the file point.
712 I = LocalSLocEntryTable.begin()+LastFileIDLookup.ID;
715 // Find the FileID that contains this. "I" is an iterator that points to a
716 // FileID whose offset is known to be larger than SLocOffset.
717 unsigned NumProbes = 0;
720 if (I->getOffset() <= SLocOffset) {
721 FileID Res = FileID::get(int(I - LocalSLocEntryTable.begin()));
723 // If this isn't an expansion, remember it. We have good locality across
725 if (!I->isExpansion())
726 LastFileIDLookup = Res;
727 NumLinearScans += NumProbes+1;
730 if (++NumProbes == 8)
734 // Convert "I" back into an index. We know that it is an entry whose index is
735 // larger than the offset we are looking for.
736 unsigned GreaterIndex = I - LocalSLocEntryTable.begin();
737 // LessIndex - This is the lower bound of the range that we're searching.
738 // We know that the offset corresponding to the FileID is is less than
740 unsigned LessIndex = 0;
743 bool Invalid = false;
744 unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex;
745 unsigned MidOffset = getLocalSLocEntry(MiddleIndex, &Invalid).getOffset();
747 return FileID::get(0);
751 // If the offset of the midpoint is too large, chop the high side of the
752 // range to the midpoint.
753 if (MidOffset > SLocOffset) {
754 GreaterIndex = MiddleIndex;
758 // If the middle index contains the value, succeed and return.
759 // FIXME: This could be made faster by using a function that's aware of
760 // being in the local area.
761 if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) {
762 FileID Res = FileID::get(MiddleIndex);
764 // If this isn't a macro expansion, remember it. We have good locality
765 // across FileID lookups.
766 if (!LocalSLocEntryTable[MiddleIndex].isExpansion())
767 LastFileIDLookup = Res;
768 NumBinaryProbes += NumProbes;
772 // Otherwise, move the low-side up to the middle index.
773 LessIndex = MiddleIndex;
777 /// \brief Return the FileID for a SourceLocation with a high offset.
779 /// This function knows that the SourceLocation is in a loaded buffer, not a
781 FileID SourceManager::getFileIDLoaded(unsigned SLocOffset) const {
782 // Sanity checking, otherwise a bug may lead to hanging in release build.
783 if (SLocOffset < CurrentLoadedOffset) {
784 assert(0 && "Invalid SLocOffset or bad function choice");
788 // Essentially the same as the local case, but the loaded array is sorted
789 // in the other direction.
791 // First do a linear scan from the last lookup position, if possible.
793 int LastID = LastFileIDLookup.ID;
794 if (LastID >= 0 || getLoadedSLocEntryByID(LastID).getOffset() < SLocOffset)
797 I = (-LastID - 2) + 1;
800 for (NumProbes = 0; NumProbes < 8; ++NumProbes, ++I) {
801 // Make sure the entry is loaded!
802 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(I);
803 if (E.getOffset() <= SLocOffset) {
804 FileID Res = FileID::get(-int(I) - 2);
806 if (!E.isExpansion())
807 LastFileIDLookup = Res;
808 NumLinearScans += NumProbes + 1;
813 // Linear scan failed. Do the binary search. Note the reverse sorting of the
814 // table: GreaterIndex is the one where the offset is greater, which is
815 // actually a lower index!
816 unsigned GreaterIndex = I;
817 unsigned LessIndex = LoadedSLocEntryTable.size();
821 unsigned MiddleIndex = (LessIndex - GreaterIndex) / 2 + GreaterIndex;
822 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(MiddleIndex);
823 if (E.getOffset() == 0)
824 return FileID(); // invalid entry.
828 if (E.getOffset() > SLocOffset) {
829 // Sanity checking, otherwise a bug may lead to hanging in release build.
830 if (GreaterIndex == MiddleIndex) {
831 assert(0 && "binary search missed the entry");
834 GreaterIndex = MiddleIndex;
838 if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) {
839 FileID Res = FileID::get(-int(MiddleIndex) - 2);
840 if (!E.isExpansion())
841 LastFileIDLookup = Res;
842 NumBinaryProbes += NumProbes;
846 // Sanity checking, otherwise a bug may lead to hanging in release build.
847 if (LessIndex == MiddleIndex) {
848 assert(0 && "binary search missed the entry");
851 LessIndex = MiddleIndex;
855 SourceLocation SourceManager::
856 getExpansionLocSlowCase(SourceLocation Loc) const {
858 // Note: If Loc indicates an offset into a token that came from a macro
859 // expansion (e.g. the 5th character of the token) we do not want to add
860 // this offset when going to the expansion location. The expansion
861 // location is the macro invocation, which the offset has nothing to do
862 // with. This is unlike when we get the spelling loc, because the offset
863 // directly correspond to the token whose spelling we're inspecting.
864 Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart();
865 } while (!Loc.isFileID());
870 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
872 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
873 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
874 Loc = Loc.getLocWithOffset(LocInfo.second);
875 } while (!Loc.isFileID());
879 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const {
881 if (isMacroArgExpansion(Loc))
882 Loc = getImmediateSpellingLoc(Loc);
884 Loc = getImmediateExpansionRange(Loc).first;
885 } while (!Loc.isFileID());
890 std::pair<FileID, unsigned>
891 SourceManager::getDecomposedExpansionLocSlowCase(
892 const SrcMgr::SLocEntry *E) const {
893 // If this is an expansion record, walk through all the expansion points.
898 Loc = E->getExpansion().getExpansionLocStart();
900 FID = getFileID(Loc);
901 E = &getSLocEntry(FID);
902 Offset = Loc.getOffset()-E->getOffset();
903 } while (!Loc.isFileID());
905 return std::make_pair(FID, Offset);
908 std::pair<FileID, unsigned>
909 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
910 unsigned Offset) const {
911 // If this is an expansion record, walk through all the expansion points.
915 Loc = E->getExpansion().getSpellingLoc();
916 Loc = Loc.getLocWithOffset(Offset);
918 FID = getFileID(Loc);
919 E = &getSLocEntry(FID);
920 Offset = Loc.getOffset()-E->getOffset();
921 } while (!Loc.isFileID());
923 return std::make_pair(FID, Offset);
926 /// getImmediateSpellingLoc - Given a SourceLocation object, return the
927 /// spelling location referenced by the ID. This is the first level down
928 /// towards the place where the characters that make up the lexed token can be
929 /// found. This should not generally be used by clients.
930 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
931 if (Loc.isFileID()) return Loc;
932 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
933 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
934 return Loc.getLocWithOffset(LocInfo.second);
938 /// getImmediateExpansionRange - Loc is required to be an expansion location.
939 /// Return the start/end of the expansion information.
940 std::pair<SourceLocation,SourceLocation>
941 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const {
942 assert(Loc.isMacroID() && "Not a macro expansion loc!");
943 const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion();
944 return Expansion.getExpansionLocRange();
947 /// getExpansionRange - Given a SourceLocation object, return the range of
948 /// tokens covered by the expansion in the ultimate file.
949 std::pair<SourceLocation,SourceLocation>
950 SourceManager::getExpansionRange(SourceLocation Loc) const {
951 if (Loc.isFileID()) return std::make_pair(Loc, Loc);
953 std::pair<SourceLocation,SourceLocation> Res =
954 getImmediateExpansionRange(Loc);
956 // Fully resolve the start and end locations to their ultimate expansion
958 while (!Res.first.isFileID())
959 Res.first = getImmediateExpansionRange(Res.first).first;
960 while (!Res.second.isFileID())
961 Res.second = getImmediateExpansionRange(Res.second).second;
965 bool SourceManager::isMacroArgExpansion(SourceLocation Loc,
966 SourceLocation *StartLoc) const {
967 if (!Loc.isMacroID()) return false;
969 FileID FID = getFileID(Loc);
970 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion();
971 if (!Expansion.isMacroArgExpansion()) return false;
974 *StartLoc = Expansion.getExpansionLocStart();
978 bool SourceManager::isMacroBodyExpansion(SourceLocation Loc) const {
979 if (!Loc.isMacroID()) return false;
981 FileID FID = getFileID(Loc);
982 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion();
983 return Expansion.isMacroBodyExpansion();
986 bool SourceManager::isAtStartOfImmediateMacroExpansion(SourceLocation Loc,
987 SourceLocation *MacroBegin) const {
988 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc");
990 std::pair<FileID, unsigned> DecompLoc = getDecomposedLoc(Loc);
991 if (DecompLoc.second > 0)
992 return false; // Does not point at the start of expansion range.
994 bool Invalid = false;
995 const SrcMgr::ExpansionInfo &ExpInfo =
996 getSLocEntry(DecompLoc.first, &Invalid).getExpansion();
999 SourceLocation ExpLoc = ExpInfo.getExpansionLocStart();
1001 if (ExpInfo.isMacroArgExpansion()) {
1002 // For macro argument expansions, check if the previous FileID is part of
1003 // the same argument expansion, in which case this Loc is not at the
1004 // beginning of the expansion.
1005 FileID PrevFID = getPreviousFileID(DecompLoc.first);
1006 if (!PrevFID.isInvalid()) {
1007 const SrcMgr::SLocEntry &PrevEntry = getSLocEntry(PrevFID, &Invalid);
1010 if (PrevEntry.isExpansion() &&
1011 PrevEntry.getExpansion().getExpansionLocStart() == ExpLoc)
1017 *MacroBegin = ExpLoc;
1021 bool SourceManager::isAtEndOfImmediateMacroExpansion(SourceLocation Loc,
1022 SourceLocation *MacroEnd) const {
1023 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc");
1025 FileID FID = getFileID(Loc);
1026 SourceLocation NextLoc = Loc.getLocWithOffset(1);
1027 if (isInFileID(NextLoc, FID))
1028 return false; // Does not point at the end of expansion range.
1030 bool Invalid = false;
1031 const SrcMgr::ExpansionInfo &ExpInfo =
1032 getSLocEntry(FID, &Invalid).getExpansion();
1036 if (ExpInfo.isMacroArgExpansion()) {
1037 // For macro argument expansions, check if the next FileID is part of the
1038 // same argument expansion, in which case this Loc is not at the end of the
1040 FileID NextFID = getNextFileID(FID);
1041 if (!NextFID.isInvalid()) {
1042 const SrcMgr::SLocEntry &NextEntry = getSLocEntry(NextFID, &Invalid);
1045 if (NextEntry.isExpansion() &&
1046 NextEntry.getExpansion().getExpansionLocStart() ==
1047 ExpInfo.getExpansionLocStart())
1053 *MacroEnd = ExpInfo.getExpansionLocEnd();
1058 //===----------------------------------------------------------------------===//
1059 // Queries about the code at a SourceLocation.
1060 //===----------------------------------------------------------------------===//
1062 /// getCharacterData - Return a pointer to the start of the specified location
1063 /// in the appropriate MemoryBuffer.
1064 const char *SourceManager::getCharacterData(SourceLocation SL,
1065 bool *Invalid) const {
1066 // Note that this is a hot function in the getSpelling() path, which is
1067 // heavily used by -E mode.
1068 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
1070 // Note that calling 'getBuffer()' may lazily page in a source file.
1071 bool CharDataInvalid = false;
1072 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid);
1073 if (CharDataInvalid || !Entry.isFile()) {
1077 return "<<<<INVALID BUFFER>>>>";
1079 llvm::MemoryBuffer *Buffer = Entry.getFile().getContentCache()->getBuffer(
1080 Diag, *this, SourceLocation(), &CharDataInvalid);
1082 *Invalid = CharDataInvalid;
1083 return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
1087 /// getColumnNumber - Return the column # for the specified file position.
1088 /// this is significantly cheaper to compute than the line number.
1089 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
1090 bool *Invalid) const {
1091 bool MyInvalid = false;
1092 llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid);
1094 *Invalid = MyInvalid;
1099 // It is okay to request a position just past the end of the buffer.
1100 if (FilePos > MemBuf->getBufferSize()) {
1106 const char *Buf = MemBuf->getBufferStart();
1107 // See if we just calculated the line number for this FilePos and can use
1108 // that to lookup the start of the line instead of searching for it.
1109 if (LastLineNoFileIDQuery == FID &&
1110 LastLineNoContentCache->SourceLineCache != nullptr &&
1111 LastLineNoResult < LastLineNoContentCache->NumLines) {
1112 unsigned *SourceLineCache = LastLineNoContentCache->SourceLineCache;
1113 unsigned LineStart = SourceLineCache[LastLineNoResult - 1];
1114 unsigned LineEnd = SourceLineCache[LastLineNoResult];
1115 if (FilePos >= LineStart && FilePos < LineEnd) {
1116 // LineEnd is the LineStart of the next line.
1117 // A line ends with separator LF or CR+LF on Windows.
1118 // FilePos might point to the last separator,
1119 // but we need a column number at most 1 + the last column.
1120 if (FilePos + 1 == LineEnd && FilePos > LineStart) {
1121 if (Buf[FilePos - 1] == '\r' || Buf[FilePos - 1] == '\n')
1124 return FilePos - LineStart + 1;
1128 unsigned LineStart = FilePos;
1129 while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
1131 return FilePos-LineStart+1;
1134 // isInvalid - Return the result of calling loc.isInvalid(), and
1135 // if Invalid is not null, set its value to same.
1136 template<typename LocType>
1137 static bool isInvalid(LocType Loc, bool *Invalid) {
1138 bool MyInvalid = Loc.isInvalid();
1140 *Invalid = MyInvalid;
1144 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
1145 bool *Invalid) const {
1146 if (isInvalid(Loc, Invalid)) return 0;
1147 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1148 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1151 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc,
1152 bool *Invalid) const {
1153 if (isInvalid(Loc, Invalid)) return 0;
1154 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1155 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1158 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc,
1159 bool *Invalid) const {
1160 PresumedLoc PLoc = getPresumedLoc(Loc);
1161 if (isInvalid(PLoc, Invalid)) return 0;
1162 return PLoc.getColumn();
1166 #include <emmintrin.h>
1169 static LLVM_ATTRIBUTE_NOINLINE void
1170 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1171 llvm::BumpPtrAllocator &Alloc,
1172 const SourceManager &SM, bool &Invalid);
1173 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1174 llvm::BumpPtrAllocator &Alloc,
1175 const SourceManager &SM, bool &Invalid) {
1176 // Note that calling 'getBuffer()' may lazily page in the file.
1177 MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(), &Invalid);
1181 // Find the file offsets of all of the *physical* source lines. This does
1182 // not look at trigraphs, escaped newlines, or anything else tricky.
1183 SmallVector<unsigned, 256> LineOffsets;
1185 // Line #1 starts at char 0.
1186 LineOffsets.push_back(0);
1188 const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
1189 const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
1192 // Skip over the contents of the line.
1193 const unsigned char *NextBuf = (const unsigned char *)Buf;
1196 // Try to skip to the next newline using SSE instructions. This is very
1197 // performance sensitive for programs with lots of diagnostics and in -E
1199 __m128i CRs = _mm_set1_epi8('\r');
1200 __m128i LFs = _mm_set1_epi8('\n');
1202 // First fix up the alignment to 16 bytes.
1203 while (((uintptr_t)NextBuf & 0xF) != 0) {
1204 if (*NextBuf == '\n' || *NextBuf == '\r' || *NextBuf == '\0')
1205 goto FoundSpecialChar;
1209 // Scan 16 byte chunks for '\r' and '\n'. Ignore '\0'.
1210 while (NextBuf+16 <= End) {
1211 const __m128i Chunk = *(const __m128i*)NextBuf;
1212 __m128i Cmp = _mm_or_si128(_mm_cmpeq_epi8(Chunk, CRs),
1213 _mm_cmpeq_epi8(Chunk, LFs));
1214 unsigned Mask = _mm_movemask_epi8(Cmp);
1216 // If we found a newline, adjust the pointer and jump to the handling code.
1218 NextBuf += llvm::countTrailingZeros(Mask);
1219 goto FoundSpecialChar;
1225 while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
1231 Offs += NextBuf-Buf;
1234 if (Buf[0] == '\n' || Buf[0] == '\r') {
1235 // If this is \n\r or \r\n, skip both characters.
1236 if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1]) {
1242 LineOffsets.push_back(Offs);
1244 // Otherwise, this is a null. If end of file, exit.
1245 if (Buf == End) break;
1246 // Otherwise, skip the null.
1252 // Copy the offsets into the FileInfo structure.
1253 FI->NumLines = LineOffsets.size();
1254 FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
1255 std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
1258 /// getLineNumber - Given a SourceLocation, return the spelling line number
1259 /// for the position indicated. This requires building and caching a table of
1260 /// line offsets for the MemoryBuffer, so this is not cheap: use only when
1261 /// about to emit a diagnostic.
1262 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
1263 bool *Invalid) const {
1264 if (FID.isInvalid()) {
1270 ContentCache *Content;
1271 if (LastLineNoFileIDQuery == FID)
1272 Content = LastLineNoContentCache;
1274 bool MyInvalid = false;
1275 const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid);
1276 if (MyInvalid || !Entry.isFile()) {
1282 Content = const_cast<ContentCache*>(Entry.getFile().getContentCache());
1285 // If this is the first use of line information for this buffer, compute the
1286 /// SourceLineCache for it on demand.
1287 if (!Content->SourceLineCache) {
1288 bool MyInvalid = false;
1289 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1291 *Invalid = MyInvalid;
1297 // Okay, we know we have a line number table. Do a binary search to find the
1298 // line number that this character position lands on.
1299 unsigned *SourceLineCache = Content->SourceLineCache;
1300 unsigned *SourceLineCacheStart = SourceLineCache;
1301 unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
1303 unsigned QueriedFilePos = FilePos+1;
1305 // FIXME: I would like to be convinced that this code is worth being as
1306 // complicated as it is, binary search isn't that slow.
1308 // If it is worth being optimized, then in my opinion it could be more
1309 // performant, simpler, and more obviously correct by just "galloping" outward
1310 // from the queried file position. In fact, this could be incorporated into a
1311 // generic algorithm such as lower_bound_with_hint.
1313 // If someone gives me a test case where this matters, and I will do it! - DWD
1315 // If the previous query was to the same file, we know both the file pos from
1316 // that query and the line number returned. This allows us to narrow the
1317 // search space from the entire file to something near the match.
1318 if (LastLineNoFileIDQuery == FID) {
1319 if (QueriedFilePos >= LastLineNoFilePos) {
1320 // FIXME: Potential overflow?
1321 SourceLineCache = SourceLineCache+LastLineNoResult-1;
1323 // The query is likely to be nearby the previous one. Here we check to
1324 // see if it is within 5, 10 or 20 lines. It can be far away in cases
1325 // where big comment blocks and vertical whitespace eat up lines but
1326 // contribute no tokens.
1327 if (SourceLineCache+5 < SourceLineCacheEnd) {
1328 if (SourceLineCache[5] > QueriedFilePos)
1329 SourceLineCacheEnd = SourceLineCache+5;
1330 else if (SourceLineCache+10 < SourceLineCacheEnd) {
1331 if (SourceLineCache[10] > QueriedFilePos)
1332 SourceLineCacheEnd = SourceLineCache+10;
1333 else if (SourceLineCache+20 < SourceLineCacheEnd) {
1334 if (SourceLineCache[20] > QueriedFilePos)
1335 SourceLineCacheEnd = SourceLineCache+20;
1340 if (LastLineNoResult < Content->NumLines)
1341 SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
1346 = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
1347 unsigned LineNo = Pos-SourceLineCacheStart;
1349 LastLineNoFileIDQuery = FID;
1350 LastLineNoContentCache = Content;
1351 LastLineNoFilePos = QueriedFilePos;
1352 LastLineNoResult = LineNo;
1356 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
1357 bool *Invalid) const {
1358 if (isInvalid(Loc, Invalid)) return 0;
1359 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1360 return getLineNumber(LocInfo.first, LocInfo.second);
1362 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc,
1363 bool *Invalid) const {
1364 if (isInvalid(Loc, Invalid)) return 0;
1365 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1366 return getLineNumber(LocInfo.first, LocInfo.second);
1368 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc,
1369 bool *Invalid) const {
1370 PresumedLoc PLoc = getPresumedLoc(Loc);
1371 if (isInvalid(PLoc, Invalid)) return 0;
1372 return PLoc.getLine();
1375 /// getFileCharacteristic - return the file characteristic of the specified
1376 /// source location, indicating whether this is a normal file, a system
1377 /// header, or an "implicit extern C" system header.
1379 /// This state can be modified with flags on GNU linemarker directives like:
1381 /// which changes all source locations in the current file after that to be
1382 /// considered to be from a system header.
1383 SrcMgr::CharacteristicKind
1384 SourceManager::getFileCharacteristic(SourceLocation Loc) const {
1385 assert(Loc.isValid() && "Can't get file characteristic of invalid loc!");
1386 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1387 bool Invalid = false;
1388 const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid);
1389 if (Invalid || !SEntry.isFile())
1392 const SrcMgr::FileInfo &FI = SEntry.getFile();
1394 // If there are no #line directives in this file, just return the whole-file
1396 if (!FI.hasLineDirectives())
1397 return FI.getFileCharacteristic();
1399 assert(LineTable && "Can't have linetable entries without a LineTable!");
1400 // See if there is a #line directive before the location.
1401 const LineEntry *Entry =
1402 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second);
1404 // If this is before the first line marker, use the file characteristic.
1406 return FI.getFileCharacteristic();
1408 return Entry->FileKind;
1411 /// Return the filename or buffer identifier of the buffer the location is in.
1412 /// Note that this name does not respect \#line directives. Use getPresumedLoc
1413 /// for normal clients.
1414 StringRef SourceManager::getBufferName(SourceLocation Loc,
1415 bool *Invalid) const {
1416 if (isInvalid(Loc, Invalid)) return "<invalid loc>";
1418 return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
1422 /// getPresumedLoc - This method returns the "presumed" location of a
1423 /// SourceLocation specifies. A "presumed location" can be modified by \#line
1424 /// or GNU line marker directives. This provides a view on the data that a
1425 /// user should see in diagnostics, for example.
1427 /// Note that a presumed location is always given as the expansion point of an
1428 /// expansion location, not at the spelling location.
1429 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc,
1430 bool UseLineDirectives) const {
1431 if (Loc.isInvalid()) return PresumedLoc();
1433 // Presumed locations are always for expansion points.
1434 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1436 bool Invalid = false;
1437 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
1438 if (Invalid || !Entry.isFile())
1439 return PresumedLoc();
1441 const SrcMgr::FileInfo &FI = Entry.getFile();
1442 const SrcMgr::ContentCache *C = FI.getContentCache();
1444 // To get the source name, first consult the FileEntry (if one exists)
1445 // before the MemBuffer as this will avoid unnecessarily paging in the
1449 Filename = C->OrigEntry->getName();
1451 Filename = C->getBuffer(Diag, *this)->getBufferIdentifier();
1453 unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid);
1455 return PresumedLoc();
1456 unsigned ColNo = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid);
1458 return PresumedLoc();
1460 SourceLocation IncludeLoc = FI.getIncludeLoc();
1462 // If we have #line directives in this file, update and overwrite the physical
1463 // location info if appropriate.
1464 if (UseLineDirectives && FI.hasLineDirectives()) {
1465 assert(LineTable && "Can't have linetable entries without a LineTable!");
1466 // See if there is a #line directive before this. If so, get it.
1467 if (const LineEntry *Entry =
1468 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) {
1469 // If the LineEntry indicates a filename, use it.
1470 if (Entry->FilenameID != -1)
1471 Filename = LineTable->getFilename(Entry->FilenameID);
1473 // Use the line number specified by the LineEntry. This line number may
1474 // be multiple lines down from the line entry. Add the difference in
1475 // physical line numbers from the query point and the line marker to the
1477 unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
1478 LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
1480 // Note that column numbers are not molested by line markers.
1482 // Handle virtual #include manipulation.
1483 if (Entry->IncludeOffset) {
1484 IncludeLoc = getLocForStartOfFile(LocInfo.first);
1485 IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset);
1490 return PresumedLoc(Filename.data(), LineNo, ColNo, IncludeLoc);
1493 /// \brief Returns whether the PresumedLoc for a given SourceLocation is
1494 /// in the main file.
1496 /// This computes the "presumed" location for a SourceLocation, then checks
1497 /// whether it came from a file other than the main file. This is different
1498 /// from isWrittenInMainFile() because it takes line marker directives into
1500 bool SourceManager::isInMainFile(SourceLocation Loc) const {
1501 if (Loc.isInvalid()) return false;
1503 // Presumed locations are always for expansion points.
1504 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1506 bool Invalid = false;
1507 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
1508 if (Invalid || !Entry.isFile())
1511 const SrcMgr::FileInfo &FI = Entry.getFile();
1513 // Check if there is a line directive for this location.
1514 if (FI.hasLineDirectives())
1515 if (const LineEntry *Entry =
1516 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second))
1517 if (Entry->IncludeOffset)
1520 return FI.getIncludeLoc().isInvalid();
1523 /// \brief The size of the SLocEntry that \p FID represents.
1524 unsigned SourceManager::getFileIDSize(FileID FID) const {
1525 bool Invalid = false;
1526 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1531 unsigned NextOffset;
1532 if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size()))
1533 NextOffset = getNextLocalOffset();
1534 else if (ID+1 == -1)
1535 NextOffset = MaxLoadedOffset;
1537 NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset();
1539 return NextOffset - Entry.getOffset() - 1;
1542 //===----------------------------------------------------------------------===//
1543 // Other miscellaneous methods.
1544 //===----------------------------------------------------------------------===//
1546 /// \brief Retrieve the inode for the given file entry, if possible.
1548 /// This routine involves a system call, and therefore should only be used
1549 /// in non-performance-critical code.
1550 static Optional<llvm::sys::fs::UniqueID>
1551 getActualFileUID(const FileEntry *File) {
1555 llvm::sys::fs::UniqueID ID;
1556 if (llvm::sys::fs::getUniqueID(File->getName(), ID))
1562 /// \brief Get the source location for the given file:line:col triplet.
1564 /// If the source file is included multiple times, the source location will
1565 /// be based upon an arbitrary inclusion.
1566 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile,
1568 unsigned Col) const {
1569 assert(SourceFile && "Null source file!");
1570 assert(Line && Col && "Line and column should start from 1!");
1572 FileID FirstFID = translateFile(SourceFile);
1573 return translateLineCol(FirstFID, Line, Col);
1576 /// \brief Get the FileID for the given file.
1578 /// If the source file is included multiple times, the FileID will be the
1579 /// first inclusion.
1580 FileID SourceManager::translateFile(const FileEntry *SourceFile) const {
1581 assert(SourceFile && "Null source file!");
1583 // Find the first file ID that corresponds to the given file.
1586 // First, check the main file ID, since it is common to look for a
1587 // location in the main file.
1588 Optional<llvm::sys::fs::UniqueID> SourceFileUID;
1589 Optional<StringRef> SourceFileName;
1590 if (MainFileID.isValid()) {
1591 bool Invalid = false;
1592 const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid);
1596 if (MainSLoc.isFile()) {
1597 const ContentCache *MainContentCache
1598 = MainSLoc.getFile().getContentCache();
1599 if (!MainContentCache) {
1600 // Can't do anything
1601 } else if (MainContentCache->OrigEntry == SourceFile) {
1602 FirstFID = MainFileID;
1604 // Fall back: check whether we have the same base name and inode
1605 // as the main file.
1606 const FileEntry *MainFile = MainContentCache->OrigEntry;
1607 SourceFileName = llvm::sys::path::filename(SourceFile->getName());
1608 if (*SourceFileName == llvm::sys::path::filename(MainFile->getName())) {
1609 SourceFileUID = getActualFileUID(SourceFile);
1610 if (SourceFileUID) {
1611 if (Optional<llvm::sys::fs::UniqueID> MainFileUID =
1612 getActualFileUID(MainFile)) {
1613 if (*SourceFileUID == *MainFileUID) {
1614 FirstFID = MainFileID;
1615 SourceFile = MainFile;
1624 if (FirstFID.isInvalid()) {
1625 // The location we're looking for isn't in the main file; look
1626 // through all of the local source locations.
1627 for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1628 bool Invalid = false;
1629 const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid);
1633 if (SLoc.isFile() &&
1634 SLoc.getFile().getContentCache() &&
1635 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1636 FirstFID = FileID::get(I);
1640 // If that still didn't help, try the modules.
1641 if (FirstFID.isInvalid()) {
1642 for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) {
1643 const SLocEntry &SLoc = getLoadedSLocEntry(I);
1644 if (SLoc.isFile() &&
1645 SLoc.getFile().getContentCache() &&
1646 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1647 FirstFID = FileID::get(-int(I) - 2);
1654 // If we haven't found what we want yet, try again, but this time stat()
1655 // each of the files in case the files have changed since we originally
1657 if (FirstFID.isInvalid() &&
1659 (SourceFileName = llvm::sys::path::filename(SourceFile->getName()))) &&
1660 (SourceFileUID || (SourceFileUID = getActualFileUID(SourceFile)))) {
1661 bool Invalid = false;
1662 for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1665 const SLocEntry &SLoc = getSLocEntry(IFileID, &Invalid);
1669 if (SLoc.isFile()) {
1670 const ContentCache *FileContentCache
1671 = SLoc.getFile().getContentCache();
1672 const FileEntry *Entry = FileContentCache ? FileContentCache->OrigEntry
1675 *SourceFileName == llvm::sys::path::filename(Entry->getName())) {
1676 if (Optional<llvm::sys::fs::UniqueID> EntryUID =
1677 getActualFileUID(Entry)) {
1678 if (*SourceFileUID == *EntryUID) {
1679 FirstFID = FileID::get(I);
1693 /// \brief Get the source location in \arg FID for the given line:col.
1694 /// Returns null location if \arg FID is not a file SLocEntry.
1695 SourceLocation SourceManager::translateLineCol(FileID FID,
1697 unsigned Col) const {
1698 // Lines are used as a one-based index into a zero-based array. This assert
1699 // checks for possible buffer underruns.
1700 assert(Line && Col && "Line and column should start from 1!");
1702 if (FID.isInvalid())
1703 return SourceLocation();
1705 bool Invalid = false;
1706 const SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1708 return SourceLocation();
1710 if (!Entry.isFile())
1711 return SourceLocation();
1713 SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset());
1715 if (Line == 1 && Col == 1)
1718 ContentCache *Content
1719 = const_cast<ContentCache *>(Entry.getFile().getContentCache());
1721 return SourceLocation();
1723 // If this is the first use of line information for this buffer, compute the
1724 // SourceLineCache for it on demand.
1725 if (!Content->SourceLineCache) {
1726 bool MyInvalid = false;
1727 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1729 return SourceLocation();
1732 if (Line > Content->NumLines) {
1733 unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize();
1736 return FileLoc.getLocWithOffset(Size);
1739 llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, *this);
1740 unsigned FilePos = Content->SourceLineCache[Line - 1];
1741 const char *Buf = Buffer->getBufferStart() + FilePos;
1742 unsigned BufLength = Buffer->getBufferSize() - FilePos;
1744 return FileLoc.getLocWithOffset(FilePos);
1748 // Check that the given column is valid.
1749 while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
1751 return FileLoc.getLocWithOffset(FilePos + i);
1754 /// \brief Compute a map of macro argument chunks to their expanded source
1755 /// location. Chunks that are not part of a macro argument will map to an
1756 /// invalid source location. e.g. if a file contains one macro argument at
1757 /// offset 100 with length 10, this is how the map will be formed:
1758 /// 0 -> SourceLocation()
1759 /// 100 -> Expanded macro arg location
1760 /// 110 -> SourceLocation()
1761 void SourceManager::computeMacroArgsCache(MacroArgsMap &MacroArgsCache,
1763 assert(FID.isValid());
1765 // Initially no macro argument chunk is present.
1766 MacroArgsCache.insert(std::make_pair(0, SourceLocation()));
1771 // Stop if there are no more FileIDs to check.
1773 if (unsigned(ID) >= local_sloc_entry_size())
1775 } else if (ID == -1) {
1779 bool Invalid = false;
1780 const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID, &Invalid);
1783 if (Entry.isFile()) {
1784 SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc();
1785 if (IncludeLoc.isInvalid())
1787 if (!isInFileID(IncludeLoc, FID))
1788 return; // No more files/macros that may be "contained" in this file.
1790 // Skip the files/macros of the #include'd file, we only care about macros
1791 // that lexed macro arguments from our file.
1792 if (Entry.getFile().NumCreatedFIDs)
1793 ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/;
1797 const ExpansionInfo &ExpInfo = Entry.getExpansion();
1799 if (ExpInfo.getExpansionLocStart().isFileID()) {
1800 if (!isInFileID(ExpInfo.getExpansionLocStart(), FID))
1801 return; // No more files/macros that may be "contained" in this file.
1804 if (!ExpInfo.isMacroArgExpansion())
1807 associateFileChunkWithMacroArgExp(MacroArgsCache, FID,
1808 ExpInfo.getSpellingLoc(),
1809 SourceLocation::getMacroLoc(Entry.getOffset()),
1810 getFileIDSize(FileID::get(ID)));
1814 void SourceManager::associateFileChunkWithMacroArgExp(
1815 MacroArgsMap &MacroArgsCache,
1817 SourceLocation SpellLoc,
1818 SourceLocation ExpansionLoc,
1819 unsigned ExpansionLength) const {
1820 if (!SpellLoc.isFileID()) {
1821 unsigned SpellBeginOffs = SpellLoc.getOffset();
1822 unsigned SpellEndOffs = SpellBeginOffs + ExpansionLength;
1824 // The spelling range for this macro argument expansion can span multiple
1825 // consecutive FileID entries. Go through each entry contained in the
1826 // spelling range and if one is itself a macro argument expansion, recurse
1827 // and associate the file chunk that it represents.
1829 FileID SpellFID; // Current FileID in the spelling range.
1830 unsigned SpellRelativeOffs;
1831 std::tie(SpellFID, SpellRelativeOffs) = getDecomposedLoc(SpellLoc);
1833 const SLocEntry &Entry = getSLocEntry(SpellFID);
1834 unsigned SpellFIDBeginOffs = Entry.getOffset();
1835 unsigned SpellFIDSize = getFileIDSize(SpellFID);
1836 unsigned SpellFIDEndOffs = SpellFIDBeginOffs + SpellFIDSize;
1837 const ExpansionInfo &Info = Entry.getExpansion();
1838 if (Info.isMacroArgExpansion()) {
1839 unsigned CurrSpellLength;
1840 if (SpellFIDEndOffs < SpellEndOffs)
1841 CurrSpellLength = SpellFIDSize - SpellRelativeOffs;
1843 CurrSpellLength = ExpansionLength;
1844 associateFileChunkWithMacroArgExp(MacroArgsCache, FID,
1845 Info.getSpellingLoc().getLocWithOffset(SpellRelativeOffs),
1846 ExpansionLoc, CurrSpellLength);
1849 if (SpellFIDEndOffs >= SpellEndOffs)
1850 return; // we covered all FileID entries in the spelling range.
1852 // Move to the next FileID entry in the spelling range.
1853 unsigned advance = SpellFIDSize - SpellRelativeOffs + 1;
1854 ExpansionLoc = ExpansionLoc.getLocWithOffset(advance);
1855 ExpansionLength -= advance;
1857 SpellRelativeOffs = 0;
1862 assert(SpellLoc.isFileID());
1865 if (!isInFileID(SpellLoc, FID, &BeginOffs))
1868 unsigned EndOffs = BeginOffs + ExpansionLength;
1870 // Add a new chunk for this macro argument. A previous macro argument chunk
1871 // may have been lexed again, so e.g. if the map is
1872 // 0 -> SourceLocation()
1873 // 100 -> Expanded loc #1
1874 // 110 -> SourceLocation()
1875 // and we found a new macro FileID that lexed from offet 105 with length 3,
1876 // the new map will be:
1877 // 0 -> SourceLocation()
1878 // 100 -> Expanded loc #1
1879 // 105 -> Expanded loc #2
1880 // 108 -> Expanded loc #1
1881 // 110 -> SourceLocation()
1883 // Since re-lexed macro chunks will always be the same size or less of
1884 // previous chunks, we only need to find where the ending of the new macro
1885 // chunk is mapped to and update the map with new begin/end mappings.
1887 MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs);
1889 SourceLocation EndOffsMappedLoc = I->second;
1890 MacroArgsCache[BeginOffs] = ExpansionLoc;
1891 MacroArgsCache[EndOffs] = EndOffsMappedLoc;
1894 /// \brief If \arg Loc points inside a function macro argument, the returned
1895 /// location will be the macro location in which the argument was expanded.
1896 /// If a macro argument is used multiple times, the expanded location will
1897 /// be at the first expansion of the argument.
1901 /// Passing a file location pointing at 'foo', will yield a macro location
1902 /// where 'foo' was expanded into.
1904 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const {
1905 if (Loc.isInvalid() || !Loc.isFileID())
1910 std::tie(FID, Offset) = getDecomposedLoc(Loc);
1911 if (FID.isInvalid())
1914 std::unique_ptr<MacroArgsMap> &MacroArgsCache = MacroArgsCacheMap[FID];
1915 if (!MacroArgsCache) {
1916 MacroArgsCache = llvm::make_unique<MacroArgsMap>();
1917 computeMacroArgsCache(*MacroArgsCache, FID);
1920 assert(!MacroArgsCache->empty());
1921 MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset);
1924 unsigned MacroArgBeginOffs = I->first;
1925 SourceLocation MacroArgExpandedLoc = I->second;
1926 if (MacroArgExpandedLoc.isValid())
1927 return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs);
1932 std::pair<FileID, unsigned>
1933 SourceManager::getDecomposedIncludedLoc(FileID FID) const {
1934 if (FID.isInvalid())
1935 return std::make_pair(FileID(), 0);
1937 // Uses IncludedLocMap to retrieve/cache the decomposed loc.
1939 typedef std::pair<FileID, unsigned> DecompTy;
1940 typedef llvm::DenseMap<FileID, DecompTy> MapTy;
1941 std::pair<MapTy::iterator, bool>
1942 InsertOp = IncludedLocMap.insert(std::make_pair(FID, DecompTy()));
1943 DecompTy &DecompLoc = InsertOp.first->second;
1944 if (!InsertOp.second)
1945 return DecompLoc; // already in map.
1947 SourceLocation UpperLoc;
1948 bool Invalid = false;
1949 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1951 if (Entry.isExpansion())
1952 UpperLoc = Entry.getExpansion().getExpansionLocStart();
1954 UpperLoc = Entry.getFile().getIncludeLoc();
1957 if (UpperLoc.isValid())
1958 DecompLoc = getDecomposedLoc(UpperLoc);
1963 /// Given a decomposed source location, move it up the include/expansion stack
1964 /// to the parent source location. If this is possible, return the decomposed
1965 /// version of the parent in Loc and return false. If Loc is the top-level
1966 /// entry, return true and don't modify it.
1967 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc,
1968 const SourceManager &SM) {
1969 std::pair<FileID, unsigned> UpperLoc = SM.getDecomposedIncludedLoc(Loc.first);
1970 if (UpperLoc.first.isInvalid())
1971 return true; // We reached the top.
1977 /// Return the cache entry for comparing the given file IDs
1978 /// for isBeforeInTranslationUnit.
1979 InBeforeInTUCacheEntry &SourceManager::getInBeforeInTUCache(FileID LFID,
1980 FileID RFID) const {
1981 // This is a magic number for limiting the cache size. It was experimentally
1982 // derived from a small Objective-C project (where the cache filled
1983 // out to ~250 items). We can make it larger if necessary.
1984 enum { MagicCacheSize = 300 };
1985 IsBeforeInTUCacheKey Key(LFID, RFID);
1987 // If the cache size isn't too large, do a lookup and if necessary default
1988 // construct an entry. We can then return it to the caller for direct
1989 // use. When they update the value, the cache will get automatically
1991 if (IBTUCache.size() < MagicCacheSize)
1992 return IBTUCache[Key];
1994 // Otherwise, do a lookup that will not construct a new value.
1995 InBeforeInTUCache::iterator I = IBTUCache.find(Key);
1996 if (I != IBTUCache.end())
1999 // Fall back to the overflow value.
2000 return IBTUCacheOverflow;
2003 /// \brief Determines the order of 2 source locations in the translation unit.
2005 /// \returns true if LHS source location comes before RHS, false otherwise.
2006 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
2007 SourceLocation RHS) const {
2008 assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
2012 std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
2013 std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
2015 // getDecomposedLoc may have failed to return a valid FileID because, e.g. it
2016 // is a serialized one referring to a file that was removed after we loaded
2018 if (LOffs.first.isInvalid() || ROffs.first.isInvalid())
2019 return LOffs.first.isInvalid() && !ROffs.first.isInvalid();
2021 std::pair<bool, bool> InSameTU = isInTheSameTranslationUnit(LOffs, ROffs);
2023 return InSameTU.second;
2025 // If we arrived here, the location is either in a built-ins buffer or
2026 // associated with global inline asm. PR5662 and PR22576 are examples.
2028 StringRef LB = getBuffer(LOffs.first)->getBufferIdentifier();
2029 StringRef RB = getBuffer(ROffs.first)->getBufferIdentifier();
2030 bool LIsBuiltins = LB == "<built-in>";
2031 bool RIsBuiltins = RB == "<built-in>";
2032 // Sort built-in before non-built-in.
2033 if (LIsBuiltins || RIsBuiltins) {
2034 if (LIsBuiltins != RIsBuiltins)
2036 // Both are in built-in buffers, but from different files. We just claim that
2037 // lower IDs come first.
2038 return LOffs.first < ROffs.first;
2040 bool LIsAsm = LB == "<inline asm>";
2041 bool RIsAsm = RB == "<inline asm>";
2042 // Sort assembler after built-ins, but before the rest.
2043 if (LIsAsm || RIsAsm) {
2044 if (LIsAsm != RIsAsm)
2046 assert(LOffs.first == ROffs.first);
2049 bool LIsScratch = LB == "<scratch space>";
2050 bool RIsScratch = RB == "<scratch space>";
2051 // Sort scratch after inline asm, but before the rest.
2052 if (LIsScratch || RIsScratch) {
2053 if (LIsScratch != RIsScratch)
2055 return LOffs.second < ROffs.second;
2057 llvm_unreachable("Unsortable locations found");
2060 std::pair<bool, bool> SourceManager::isInTheSameTranslationUnit(
2061 std::pair<FileID, unsigned> &LOffs,
2062 std::pair<FileID, unsigned> &ROffs) const {
2063 // If the source locations are in the same file, just compare offsets.
2064 if (LOffs.first == ROffs.first)
2065 return std::make_pair(true, LOffs.second < ROffs.second);
2067 // If we are comparing a source location with multiple locations in the same
2068 // file, we get a big win by caching the result.
2069 InBeforeInTUCacheEntry &IsBeforeInTUCache =
2070 getInBeforeInTUCache(LOffs.first, ROffs.first);
2072 // If we are comparing a source location with multiple locations in the same
2073 // file, we get a big win by caching the result.
2074 if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first))
2075 return std::make_pair(
2076 true, IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second));
2078 // Okay, we missed in the cache, start updating the cache for this query.
2079 IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first,
2080 /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID);
2082 // We need to find the common ancestor. The only way of doing this is to
2083 // build the complete include chain for one and then walking up the chain
2084 // of the other looking for a match.
2085 // We use a map from FileID to Offset to store the chain. Easier than writing
2086 // a custom set hash info that only depends on the first part of a pair.
2087 typedef llvm::SmallDenseMap<FileID, unsigned, 16> LocSet;
2090 LChain.insert(LOffs);
2091 // We catch the case where LOffs is in a file included by ROffs and
2092 // quit early. The other way round unfortunately remains suboptimal.
2093 } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this));
2095 while((I = LChain.find(ROffs.first)) == LChain.end()) {
2096 if (MoveUpIncludeHierarchy(ROffs, *this))
2097 break; // Met at topmost file.
2099 if (I != LChain.end())
2102 // If we exited because we found a nearest common ancestor, compare the
2103 // locations within the common file and cache them.
2104 if (LOffs.first == ROffs.first) {
2105 IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second);
2106 return std::make_pair(
2107 true, IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second));
2109 // Clear the lookup cache, it depends on a common location.
2110 IsBeforeInTUCache.clear();
2111 return std::make_pair(false, false);
2114 void SourceManager::PrintStats() const {
2115 llvm::errs() << "\n*** Source Manager Stats:\n";
2116 llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
2117 << " mem buffers mapped.\n";
2118 llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated ("
2119 << llvm::capacity_in_bytes(LocalSLocEntryTable)
2120 << " bytes of capacity), "
2121 << NextLocalOffset << "B of Sloc address space used.\n";
2122 llvm::errs() << LoadedSLocEntryTable.size()
2123 << " loaded SLocEntries allocated, "
2124 << MaxLoadedOffset - CurrentLoadedOffset
2125 << "B of Sloc address space used.\n";
2127 unsigned NumLineNumsComputed = 0;
2128 unsigned NumFileBytesMapped = 0;
2129 for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
2130 NumLineNumsComputed += I->second->SourceLineCache != nullptr;
2131 NumFileBytesMapped += I->second->getSizeBytesMapped();
2133 unsigned NumMacroArgsComputed = MacroArgsCacheMap.size();
2135 llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
2136 << NumLineNumsComputed << " files with line #'s computed, "
2137 << NumMacroArgsComputed << " files with macro args computed.\n";
2138 llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
2139 << NumBinaryProbes << " binary.\n";
2142 LLVM_DUMP_METHOD void SourceManager::dump() const {
2143 llvm::raw_ostream &out = llvm::errs();
2145 auto DumpSLocEntry = [&](int ID, const SrcMgr::SLocEntry &Entry,
2146 llvm::Optional<unsigned> NextStart) {
2147 out << "SLocEntry <FileID " << ID << "> " << (Entry.isFile() ? "file" : "expansion")
2148 << " <SourceLocation " << Entry.getOffset() << ":";
2150 out << *NextStart << ">\n";
2153 if (Entry.isFile()) {
2154 auto &FI = Entry.getFile();
2155 if (FI.NumCreatedFIDs)
2156 out << " covers <FileID " << ID << ":" << int(ID + FI.NumCreatedFIDs)
2158 if (FI.getIncludeLoc().isValid())
2159 out << " included from " << FI.getIncludeLoc().getOffset() << "\n";
2160 if (auto *CC = FI.getContentCache()) {
2161 out << " for " << (CC->OrigEntry ? CC->OrigEntry->getName() : "<none>")
2163 if (CC->BufferOverridden)
2164 out << " contents overridden\n";
2165 if (CC->ContentsEntry != CC->OrigEntry) {
2166 out << " contents from "
2167 << (CC->ContentsEntry ? CC->ContentsEntry->getName() : "<none>")
2172 auto &EI = Entry.getExpansion();
2173 out << " spelling from " << EI.getSpellingLoc().getOffset() << "\n";
2174 out << " macro " << (EI.isMacroArgExpansion() ? "arg" : "body")
2175 << " range <" << EI.getExpansionLocStart().getOffset() << ":"
2176 << EI.getExpansionLocEnd().getOffset() << ">\n";
2180 // Dump local SLocEntries.
2181 for (unsigned ID = 0, NumIDs = LocalSLocEntryTable.size(); ID != NumIDs; ++ID) {
2182 DumpSLocEntry(ID, LocalSLocEntryTable[ID],
2183 ID == NumIDs - 1 ? NextLocalOffset
2184 : LocalSLocEntryTable[ID + 1].getOffset());
2186 // Dump loaded SLocEntries.
2187 llvm::Optional<unsigned> NextStart;
2188 for (unsigned Index = 0; Index != LoadedSLocEntryTable.size(); ++Index) {
2189 int ID = -(int)Index - 2;
2190 if (SLocEntryLoaded[Index]) {
2191 DumpSLocEntry(ID, LoadedSLocEntryTable[Index], NextStart);
2192 NextStart = LoadedSLocEntryTable[Index].getOffset();
2199 ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
2201 /// Return the amount of memory used by memory buffers, breaking down
2202 /// by heap-backed versus mmap'ed memory.
2203 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const {
2204 size_t malloc_bytes = 0;
2205 size_t mmap_bytes = 0;
2207 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i)
2208 if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped())
2209 switch (MemBufferInfos[i]->getMemoryBufferKind()) {
2210 case llvm::MemoryBuffer::MemoryBuffer_MMap:
2211 mmap_bytes += sized_mapped;
2213 case llvm::MemoryBuffer::MemoryBuffer_Malloc:
2214 malloc_bytes += sized_mapped;
2218 return MemoryBufferSizes(malloc_bytes, mmap_bytes);
2221 size_t SourceManager::getDataStructureSizes() const {
2222 size_t size = llvm::capacity_in_bytes(MemBufferInfos)
2223 + llvm::capacity_in_bytes(LocalSLocEntryTable)
2224 + llvm::capacity_in_bytes(LoadedSLocEntryTable)
2225 + llvm::capacity_in_bytes(SLocEntryLoaded)
2226 + llvm::capacity_in_bytes(FileInfos);
2228 if (OverriddenFilesInfo)
2229 size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles);