xref: /llvm-project/clang/lib/Basic/SourceManager.cpp (revision 3a02247dc9e39bc62c5ca6bbe6a3c34d746b5407)
1 //===--- SourceManager.cpp - Track and cache source files -----------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 //  This file implements the SourceManager interface.
11 //
12 //===----------------------------------------------------------------------===//
13 
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"
26 #include <algorithm>
27 #include <cstring>
28 #include <string>
29 #include <sys/stat.h>
30 
31 using namespace clang;
32 using namespace SrcMgr;
33 using llvm::MemoryBuffer;
34 
35 //===----------------------------------------------------------------------===//
36 // SourceManager Helper Classes
37 //===----------------------------------------------------------------------===//
38 
39 ContentCache::~ContentCache() {
40   if (shouldFreeBuffer())
41     delete Buffer.getPointer();
42 }
43 
44 /// getSizeBytesMapped - Returns the number of bytes actually mapped for this
45 /// ContentCache. This can be 0 if the MemBuffer was not actually expanded.
46 unsigned ContentCache::getSizeBytesMapped() const {
47   return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0;
48 }
49 
50 /// Returns the kind of memory used to back the memory buffer for
51 /// this content cache.  This is used for performance analysis.
52 llvm::MemoryBuffer::BufferKind ContentCache::getMemoryBufferKind() const {
53   assert(Buffer.getPointer());
54 
55   // Should be unreachable, but keep for sanity.
56   if (!Buffer.getPointer())
57     return llvm::MemoryBuffer::MemoryBuffer_Malloc;
58 
59   const llvm::MemoryBuffer *buf = Buffer.getPointer();
60   return buf->getBufferKind();
61 }
62 
63 /// getSize - Returns the size of the content encapsulated by this ContentCache.
64 ///  This can be the size of the source file or the size of an arbitrary
65 ///  scratch buffer.  If the ContentCache encapsulates a source file, that
66 ///  file is not lazily brought in from disk to satisfy this query.
67 unsigned ContentCache::getSize() const {
68   return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize()
69                              : (unsigned) ContentsEntry->getSize();
70 }
71 
72 void ContentCache::replaceBuffer(const llvm::MemoryBuffer *B,
73                                  bool DoNotFree) {
74   if (B && B == Buffer.getPointer()) {
75     assert(0 && "Replacing with the same buffer");
76     Buffer.setInt(DoNotFree? DoNotFreeFlag : 0);
77     return;
78   }
79 
80   if (shouldFreeBuffer())
81     delete Buffer.getPointer();
82   Buffer.setPointer(B);
83   Buffer.setInt(DoNotFree? DoNotFreeFlag : 0);
84 }
85 
86 const llvm::MemoryBuffer *ContentCache::getBuffer(DiagnosticsEngine &Diag,
87                                                   const SourceManager &SM,
88                                                   SourceLocation Loc,
89                                                   bool *Invalid) const {
90   // Lazily create the Buffer for ContentCaches that wrap files.  If we already
91   // computed it, just return what we have.
92   if (Buffer.getPointer() || ContentsEntry == 0) {
93     if (Invalid)
94       *Invalid = isBufferInvalid();
95 
96     return Buffer.getPointer();
97   }
98 
99   std::string ErrorStr;
100   bool isVolatile = SM.userFilesAreVolatile() && !IsSystemFile;
101   Buffer.setPointer(SM.getFileManager().getBufferForFile(ContentsEntry,
102                                                          &ErrorStr,
103                                                          isVolatile));
104 
105   // If we were unable to open the file, then we are in an inconsistent
106   // situation where the content cache referenced a file which no longer
107   // exists. Most likely, we were using a stat cache with an invalid entry but
108   // the file could also have been removed during processing. Since we can't
109   // really deal with this situation, just create an empty buffer.
110   //
111   // FIXME: This is definitely not ideal, but our immediate clients can't
112   // currently handle returning a null entry here. Ideally we should detect
113   // that we are in an inconsistent situation and error out as quickly as
114   // possible.
115   if (!Buffer.getPointer()) {
116     const StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
117     Buffer.setPointer(MemoryBuffer::getNewMemBuffer(ContentsEntry->getSize(),
118                                                     "<invalid>"));
119     char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
120     for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i)
121       Ptr[i] = FillStr[i % FillStr.size()];
122 
123     if (Diag.isDiagnosticInFlight())
124       Diag.SetDelayedDiagnostic(diag::err_cannot_open_file,
125                                 ContentsEntry->getName(), ErrorStr);
126     else
127       Diag.Report(Loc, diag::err_cannot_open_file)
128         << ContentsEntry->getName() << ErrorStr;
129 
130     Buffer.setInt(Buffer.getInt() | InvalidFlag);
131 
132     if (Invalid) *Invalid = true;
133     return Buffer.getPointer();
134   }
135 
136   // Check that the file's size is the same as in the file entry (which may
137   // have come from a stat cache).
138   if (getRawBuffer()->getBufferSize() != (size_t)ContentsEntry->getSize()) {
139     if (Diag.isDiagnosticInFlight())
140       Diag.SetDelayedDiagnostic(diag::err_file_modified,
141                                 ContentsEntry->getName());
142     else
143       Diag.Report(Loc, diag::err_file_modified)
144         << ContentsEntry->getName();
145 
146     Buffer.setInt(Buffer.getInt() | InvalidFlag);
147     if (Invalid) *Invalid = true;
148     return Buffer.getPointer();
149   }
150 
151   // If the buffer is valid, check to see if it has a UTF Byte Order Mark
152   // (BOM).  We only support UTF-8 with and without a BOM right now.  See
153   // http://en.wikipedia.org/wiki/Byte_order_mark for more information.
154   StringRef BufStr = Buffer.getPointer()->getBuffer();
155   const char *InvalidBOM = llvm::StringSwitch<const char *>(BufStr)
156     .StartsWith("\xFE\xFF", "UTF-16 (BE)")
157     .StartsWith("\xFF\xFE", "UTF-16 (LE)")
158     .StartsWith("\x00\x00\xFE\xFF", "UTF-32 (BE)")
159     .StartsWith("\xFF\xFE\x00\x00", "UTF-32 (LE)")
160     .StartsWith("\x2B\x2F\x76", "UTF-7")
161     .StartsWith("\xF7\x64\x4C", "UTF-1")
162     .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC")
163     .StartsWith("\x0E\xFE\xFF", "SDSU")
164     .StartsWith("\xFB\xEE\x28", "BOCU-1")
165     .StartsWith("\x84\x31\x95\x33", "GB-18030")
166     .Default(0);
167 
168   if (InvalidBOM) {
169     Diag.Report(Loc, diag::err_unsupported_bom)
170       << InvalidBOM << ContentsEntry->getName();
171     Buffer.setInt(Buffer.getInt() | InvalidFlag);
172   }
173 
174   if (Invalid)
175     *Invalid = isBufferInvalid();
176 
177   return Buffer.getPointer();
178 }
179 
180 unsigned LineTableInfo::getLineTableFilenameID(StringRef Name) {
181   // Look up the filename in the string table, returning the pre-existing value
182   // if it exists.
183   llvm::StringMapEntry<unsigned> &Entry =
184     FilenameIDs.GetOrCreateValue(Name, ~0U);
185   if (Entry.getValue() != ~0U)
186     return Entry.getValue();
187 
188   // Otherwise, assign this the next available ID.
189   Entry.setValue(FilenamesByID.size());
190   FilenamesByID.push_back(&Entry);
191   return FilenamesByID.size()-1;
192 }
193 
194 /// AddLineNote - Add a line note to the line table that indicates that there
195 /// is a \#line at the specified FID/Offset location which changes the presumed
196 /// location to LineNo/FilenameID.
197 void LineTableInfo::AddLineNote(FileID FID, unsigned Offset,
198                                 unsigned LineNo, int FilenameID) {
199   std::vector<LineEntry> &Entries = LineEntries[FID];
200 
201   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
202          "Adding line entries out of order!");
203 
204   SrcMgr::CharacteristicKind Kind = SrcMgr::C_User;
205   unsigned IncludeOffset = 0;
206 
207   if (!Entries.empty()) {
208     // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember
209     // that we are still in "foo.h".
210     if (FilenameID == -1)
211       FilenameID = Entries.back().FilenameID;
212 
213     // If we are after a line marker that switched us to system header mode, or
214     // that set #include information, preserve it.
215     Kind = Entries.back().FileKind;
216     IncludeOffset = Entries.back().IncludeOffset;
217   }
218 
219   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind,
220                                    IncludeOffset));
221 }
222 
223 /// AddLineNote This is the same as the previous version of AddLineNote, but is
224 /// used for GNU line markers.  If EntryExit is 0, then this doesn't change the
225 /// presumed \#include stack.  If it is 1, this is a file entry, if it is 2 then
226 /// this is a file exit.  FileKind specifies whether this is a system header or
227 /// extern C system header.
228 void LineTableInfo::AddLineNote(FileID FID, unsigned Offset,
229                                 unsigned LineNo, int FilenameID,
230                                 unsigned EntryExit,
231                                 SrcMgr::CharacteristicKind FileKind) {
232   assert(FilenameID != -1 && "Unspecified filename should use other accessor");
233 
234   std::vector<LineEntry> &Entries = LineEntries[FID];
235 
236   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
237          "Adding line entries out of order!");
238 
239   unsigned IncludeOffset = 0;
240   if (EntryExit == 0) {  // No #include stack change.
241     IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset;
242   } else if (EntryExit == 1) {
243     IncludeOffset = Offset-1;
244   } else if (EntryExit == 2) {
245     assert(!Entries.empty() && Entries.back().IncludeOffset &&
246        "PPDirectives should have caught case when popping empty include stack");
247 
248     // Get the include loc of the last entries' include loc as our include loc.
249     IncludeOffset = 0;
250     if (const LineEntry *PrevEntry =
251           FindNearestLineEntry(FID, Entries.back().IncludeOffset))
252       IncludeOffset = PrevEntry->IncludeOffset;
253   }
254 
255   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind,
256                                    IncludeOffset));
257 }
258 
259 
260 /// FindNearestLineEntry - Find the line entry nearest to FID that is before
261 /// it.  If there is no line entry before Offset in FID, return null.
262 const LineEntry *LineTableInfo::FindNearestLineEntry(FileID FID,
263                                                      unsigned Offset) {
264   const std::vector<LineEntry> &Entries = LineEntries[FID];
265   assert(!Entries.empty() && "No #line entries for this FID after all!");
266 
267   // It is very common for the query to be after the last #line, check this
268   // first.
269   if (Entries.back().FileOffset <= Offset)
270     return &Entries.back();
271 
272   // Do a binary search to find the maximal element that is still before Offset.
273   std::vector<LineEntry>::const_iterator I =
274     std::upper_bound(Entries.begin(), Entries.end(), Offset);
275   if (I == Entries.begin()) return 0;
276   return &*--I;
277 }
278 
279 /// \brief Add a new line entry that has already been encoded into
280 /// the internal representation of the line table.
281 void LineTableInfo::AddEntry(FileID FID,
282                              const std::vector<LineEntry> &Entries) {
283   LineEntries[FID] = Entries;
284 }
285 
286 /// getLineTableFilenameID - Return the uniqued ID for the specified filename.
287 ///
288 unsigned SourceManager::getLineTableFilenameID(StringRef Name) {
289   if (LineTable == 0)
290     LineTable = new LineTableInfo();
291   return LineTable->getLineTableFilenameID(Name);
292 }
293 
294 
295 /// AddLineNote - Add a line note to the line table for the FileID and offset
296 /// specified by Loc.  If FilenameID is -1, it is considered to be
297 /// unspecified.
298 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
299                                 int FilenameID) {
300   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
301 
302   bool Invalid = false;
303   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
304   if (!Entry.isFile() || Invalid)
305     return;
306 
307   const SrcMgr::FileInfo &FileInfo = Entry.getFile();
308 
309   // Remember that this file has #line directives now if it doesn't already.
310   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
311 
312   if (LineTable == 0)
313     LineTable = new LineTableInfo();
314   LineTable->AddLineNote(LocInfo.first, LocInfo.second, LineNo, FilenameID);
315 }
316 
317 /// AddLineNote - Add a GNU line marker to the line table.
318 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
319                                 int FilenameID, bool IsFileEntry,
320                                 bool IsFileExit, bool IsSystemHeader,
321                                 bool IsExternCHeader) {
322   // If there is no filename and no flags, this is treated just like a #line,
323   // which does not change the flags of the previous line marker.
324   if (FilenameID == -1) {
325     assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader &&
326            "Can't set flags without setting the filename!");
327     return AddLineNote(Loc, LineNo, FilenameID);
328   }
329 
330   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
331 
332   bool Invalid = false;
333   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
334   if (!Entry.isFile() || Invalid)
335     return;
336 
337   const SrcMgr::FileInfo &FileInfo = Entry.getFile();
338 
339   // Remember that this file has #line directives now if it doesn't already.
340   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
341 
342   if (LineTable == 0)
343     LineTable = new LineTableInfo();
344 
345   SrcMgr::CharacteristicKind FileKind;
346   if (IsExternCHeader)
347     FileKind = SrcMgr::C_ExternCSystem;
348   else if (IsSystemHeader)
349     FileKind = SrcMgr::C_System;
350   else
351     FileKind = SrcMgr::C_User;
352 
353   unsigned EntryExit = 0;
354   if (IsFileEntry)
355     EntryExit = 1;
356   else if (IsFileExit)
357     EntryExit = 2;
358 
359   LineTable->AddLineNote(LocInfo.first, LocInfo.second, LineNo, FilenameID,
360                          EntryExit, FileKind);
361 }
362 
363 LineTableInfo &SourceManager::getLineTable() {
364   if (LineTable == 0)
365     LineTable = new LineTableInfo();
366   return *LineTable;
367 }
368 
369 //===----------------------------------------------------------------------===//
370 // Private 'Create' methods.
371 //===----------------------------------------------------------------------===//
372 
373 SourceManager::SourceManager(DiagnosticsEngine &Diag, FileManager &FileMgr,
374                              bool UserFilesAreVolatile)
375   : Diag(Diag), FileMgr(FileMgr), OverridenFilesKeepOriginalName(true),
376     UserFilesAreVolatile(UserFilesAreVolatile),
377     ExternalSLocEntries(0), LineTable(0), NumLinearScans(0),
378     NumBinaryProbes(0), FakeBufferForRecovery(0),
379     FakeContentCacheForRecovery(0) {
380   clearIDTables();
381   Diag.setSourceManager(this);
382 }
383 
384 SourceManager::~SourceManager() {
385   delete LineTable;
386 
387   // Delete FileEntry objects corresponding to content caches.  Since the actual
388   // content cache objects are bump pointer allocated, we just have to run the
389   // dtors, but we call the deallocate method for completeness.
390   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) {
391     if (MemBufferInfos[i]) {
392       MemBufferInfos[i]->~ContentCache();
393       ContentCacheAlloc.Deallocate(MemBufferInfos[i]);
394     }
395   }
396   for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator
397        I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) {
398     if (I->second) {
399       I->second->~ContentCache();
400       ContentCacheAlloc.Deallocate(I->second);
401     }
402   }
403 
404   delete FakeBufferForRecovery;
405   delete FakeContentCacheForRecovery;
406 
407   for (llvm::DenseMap<FileID, MacroArgsMap *>::iterator
408          I = MacroArgsCacheMap.begin(),E = MacroArgsCacheMap.end(); I!=E; ++I) {
409     delete I->second;
410   }
411 }
412 
413 void SourceManager::clearIDTables() {
414   MainFileID = FileID();
415   LocalSLocEntryTable.clear();
416   LoadedSLocEntryTable.clear();
417   SLocEntryLoaded.clear();
418   LastLineNoFileIDQuery = FileID();
419   LastLineNoContentCache = 0;
420   LastFileIDLookup = FileID();
421 
422   if (LineTable)
423     LineTable->clear();
424 
425   // Use up FileID #0 as an invalid expansion.
426   NextLocalOffset = 0;
427   CurrentLoadedOffset = MaxLoadedOffset;
428   createExpansionLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1);
429 }
430 
431 /// getOrCreateContentCache - Create or return a cached ContentCache for the
432 /// specified file.
433 const ContentCache *
434 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt,
435                                        bool isSystemFile) {
436   assert(FileEnt && "Didn't specify a file entry to use?");
437 
438   // Do we already have information about this file?
439   ContentCache *&Entry = FileInfos[FileEnt];
440   if (Entry) return Entry;
441 
442   // Nope, create a new Cache entry.  Make sure it is at least 8-byte aligned
443   // so that FileInfo can use the low 3 bits of the pointer for its own
444   // nefarious purposes.
445   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
446   EntryAlign = std::max(8U, EntryAlign);
447   Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
448 
449   if (OverriddenFilesInfo) {
450     // If the file contents are overridden with contents from another file,
451     // pass that file to ContentCache.
452     llvm::DenseMap<const FileEntry *, const FileEntry *>::iterator
453         overI = OverriddenFilesInfo->OverriddenFiles.find(FileEnt);
454     if (overI == OverriddenFilesInfo->OverriddenFiles.end())
455       new (Entry) ContentCache(FileEnt);
456     else
457       new (Entry) ContentCache(OverridenFilesKeepOriginalName ? FileEnt
458                                                               : overI->second,
459                                overI->second);
460   } else {
461     new (Entry) ContentCache(FileEnt);
462   }
463 
464   Entry->IsSystemFile = isSystemFile;
465 
466   return Entry;
467 }
468 
469 
470 /// createMemBufferContentCache - Create a new ContentCache for the specified
471 ///  memory buffer.  This does no caching.
472 const ContentCache*
473 SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) {
474   // Add a new ContentCache to the MemBufferInfos list and return it.  Make sure
475   // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of
476   // the pointer for its own nefarious purposes.
477   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
478   EntryAlign = std::max(8U, EntryAlign);
479   ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
480   new (Entry) ContentCache();
481   MemBufferInfos.push_back(Entry);
482   Entry->setBuffer(Buffer);
483   return Entry;
484 }
485 
486 const SrcMgr::SLocEntry &SourceManager::loadSLocEntry(unsigned Index,
487                                                       bool *Invalid) const {
488   assert(!SLocEntryLoaded[Index]);
489   if (ExternalSLocEntries->ReadSLocEntry(-(static_cast<int>(Index) + 2))) {
490     if (Invalid)
491       *Invalid = true;
492     // If the file of the SLocEntry changed we could still have loaded it.
493     if (!SLocEntryLoaded[Index]) {
494       // Try to recover; create a SLocEntry so the rest of clang can handle it.
495       LoadedSLocEntryTable[Index] = SLocEntry::get(0,
496                                  FileInfo::get(SourceLocation(),
497                                                getFakeContentCacheForRecovery(),
498                                                SrcMgr::C_User));
499     }
500   }
501 
502   return LoadedSLocEntryTable[Index];
503 }
504 
505 std::pair<int, unsigned>
506 SourceManager::AllocateLoadedSLocEntries(unsigned NumSLocEntries,
507                                          unsigned TotalSize) {
508   assert(ExternalSLocEntries && "Don't have an external sloc source");
509   LoadedSLocEntryTable.resize(LoadedSLocEntryTable.size() + NumSLocEntries);
510   SLocEntryLoaded.resize(LoadedSLocEntryTable.size());
511   CurrentLoadedOffset -= TotalSize;
512   assert(CurrentLoadedOffset >= NextLocalOffset && "Out of source locations");
513   int ID = LoadedSLocEntryTable.size();
514   return std::make_pair(-ID - 1, CurrentLoadedOffset);
515 }
516 
517 /// \brief As part of recovering from missing or changed content, produce a
518 /// fake, non-empty buffer.
519 const llvm::MemoryBuffer *SourceManager::getFakeBufferForRecovery() const {
520   if (!FakeBufferForRecovery)
521     FakeBufferForRecovery
522       = llvm::MemoryBuffer::getMemBuffer("<<<INVALID BUFFER>>");
523 
524   return FakeBufferForRecovery;
525 }
526 
527 /// \brief As part of recovering from missing or changed content, produce a
528 /// fake content cache.
529 const SrcMgr::ContentCache *
530 SourceManager::getFakeContentCacheForRecovery() const {
531   if (!FakeContentCacheForRecovery) {
532     FakeContentCacheForRecovery = new ContentCache();
533     FakeContentCacheForRecovery->replaceBuffer(getFakeBufferForRecovery(),
534                                                /*DoNotFree=*/true);
535   }
536   return FakeContentCacheForRecovery;
537 }
538 
539 //===----------------------------------------------------------------------===//
540 // Methods to create new FileID's and macro expansions.
541 //===----------------------------------------------------------------------===//
542 
543 /// createFileID - Create a new FileID for the specified ContentCache and
544 /// include position.  This works regardless of whether the ContentCache
545 /// corresponds to a file or some other input source.
546 FileID SourceManager::createFileID(const ContentCache *File,
547                                    SourceLocation IncludePos,
548                                    SrcMgr::CharacteristicKind FileCharacter,
549                                    int LoadedID, unsigned LoadedOffset) {
550   if (LoadedID < 0) {
551     assert(LoadedID != -1 && "Loading sentinel FileID");
552     unsigned Index = unsigned(-LoadedID) - 2;
553     assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
554     assert(!SLocEntryLoaded[Index] && "FileID already loaded");
555     LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset,
556         FileInfo::get(IncludePos, File, FileCharacter));
557     SLocEntryLoaded[Index] = true;
558     return FileID::get(LoadedID);
559   }
560   LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset,
561                                                FileInfo::get(IncludePos, File,
562                                                              FileCharacter)));
563   unsigned FileSize = File->getSize();
564   assert(NextLocalOffset + FileSize + 1 > NextLocalOffset &&
565          NextLocalOffset + FileSize + 1 <= CurrentLoadedOffset &&
566          "Ran out of source locations!");
567   // We do a +1 here because we want a SourceLocation that means "the end of the
568   // file", e.g. for the "no newline at the end of the file" diagnostic.
569   NextLocalOffset += FileSize + 1;
570 
571   // Set LastFileIDLookup to the newly created file.  The next getFileID call is
572   // almost guaranteed to be from that file.
573   FileID FID = FileID::get(LocalSLocEntryTable.size()-1);
574   return LastFileIDLookup = FID;
575 }
576 
577 SourceLocation
578 SourceManager::createMacroArgExpansionLoc(SourceLocation SpellingLoc,
579                                           SourceLocation ExpansionLoc,
580                                           unsigned TokLength) {
581   ExpansionInfo Info = ExpansionInfo::createForMacroArg(SpellingLoc,
582                                                         ExpansionLoc);
583   return createExpansionLocImpl(Info, TokLength);
584 }
585 
586 SourceLocation
587 SourceManager::createExpansionLoc(SourceLocation SpellingLoc,
588                                   SourceLocation ExpansionLocStart,
589                                   SourceLocation ExpansionLocEnd,
590                                   unsigned TokLength,
591                                   int LoadedID,
592                                   unsigned LoadedOffset) {
593   ExpansionInfo Info = ExpansionInfo::create(SpellingLoc, ExpansionLocStart,
594                                              ExpansionLocEnd);
595   return createExpansionLocImpl(Info, TokLength, LoadedID, LoadedOffset);
596 }
597 
598 SourceLocation
599 SourceManager::createExpansionLocImpl(const ExpansionInfo &Info,
600                                       unsigned TokLength,
601                                       int LoadedID,
602                                       unsigned LoadedOffset) {
603   if (LoadedID < 0) {
604     assert(LoadedID != -1 && "Loading sentinel FileID");
605     unsigned Index = unsigned(-LoadedID) - 2;
606     assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
607     assert(!SLocEntryLoaded[Index] && "FileID already loaded");
608     LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, Info);
609     SLocEntryLoaded[Index] = true;
610     return SourceLocation::getMacroLoc(LoadedOffset);
611   }
612   LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, Info));
613   assert(NextLocalOffset + TokLength + 1 > NextLocalOffset &&
614          NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset &&
615          "Ran out of source locations!");
616   // See createFileID for that +1.
617   NextLocalOffset += TokLength + 1;
618   return SourceLocation::getMacroLoc(NextLocalOffset - (TokLength + 1));
619 }
620 
621 const llvm::MemoryBuffer *
622 SourceManager::getMemoryBufferForFile(const FileEntry *File,
623                                       bool *Invalid) {
624   const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
625   assert(IR && "getOrCreateContentCache() cannot return NULL");
626   return IR->getBuffer(Diag, *this, SourceLocation(), Invalid);
627 }
628 
629 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
630                                          const llvm::MemoryBuffer *Buffer,
631                                          bool DoNotFree) {
632   const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile);
633   assert(IR && "getOrCreateContentCache() cannot return NULL");
634 
635   const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree);
636   const_cast<SrcMgr::ContentCache *>(IR)->BufferOverridden = true;
637 
638   getOverriddenFilesInfo().OverriddenFilesWithBuffer.insert(SourceFile);
639 }
640 
641 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
642                                          const FileEntry *NewFile) {
643   assert(SourceFile->getSize() == NewFile->getSize() &&
644          "Different sizes, use the FileManager to create a virtual file with "
645          "the correct size");
646   assert(FileInfos.count(SourceFile) == 0 &&
647          "This function should be called at the initialization stage, before "
648          "any parsing occurs.");
649   getOverriddenFilesInfo().OverriddenFiles[SourceFile] = NewFile;
650 }
651 
652 void SourceManager::disableFileContentsOverride(const FileEntry *File) {
653   if (!isFileOverridden(File))
654     return;
655 
656   const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
657   const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(0);
658   const_cast<SrcMgr::ContentCache *>(IR)->ContentsEntry = IR->OrigEntry;
659 
660   assert(OverriddenFilesInfo);
661   OverriddenFilesInfo->OverriddenFiles.erase(File);
662   OverriddenFilesInfo->OverriddenFilesWithBuffer.erase(File);
663 }
664 
665 StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const {
666   bool MyInvalid = false;
667   const SLocEntry &SLoc = getSLocEntry(FID, &MyInvalid);
668   if (!SLoc.isFile() || MyInvalid) {
669     if (Invalid)
670       *Invalid = true;
671     return "<<<<<INVALID SOURCE LOCATION>>>>>";
672   }
673 
674   const llvm::MemoryBuffer *Buf
675     = SLoc.getFile().getContentCache()->getBuffer(Diag, *this, SourceLocation(),
676                                                   &MyInvalid);
677   if (Invalid)
678     *Invalid = MyInvalid;
679 
680   if (MyInvalid)
681     return "<<<<<INVALID SOURCE LOCATION>>>>>";
682 
683   return Buf->getBuffer();
684 }
685 
686 //===----------------------------------------------------------------------===//
687 // SourceLocation manipulation methods.
688 //===----------------------------------------------------------------------===//
689 
690 /// \brief Return the FileID for a SourceLocation.
691 ///
692 /// This is the cache-miss path of getFileID. Not as hot as that function, but
693 /// still very important. It is responsible for finding the entry in the
694 /// SLocEntry tables that contains the specified location.
695 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const {
696   if (!SLocOffset)
697     return FileID::get(0);
698 
699   // Now it is time to search for the correct file. See where the SLocOffset
700   // sits in the global view and consult local or loaded buffers for it.
701   if (SLocOffset < NextLocalOffset)
702     return getFileIDLocal(SLocOffset);
703   return getFileIDLoaded(SLocOffset);
704 }
705 
706 /// \brief Return the FileID for a SourceLocation with a low offset.
707 ///
708 /// This function knows that the SourceLocation is in a local buffer, not a
709 /// loaded one.
710 FileID SourceManager::getFileIDLocal(unsigned SLocOffset) const {
711   assert(SLocOffset < NextLocalOffset && "Bad function choice");
712 
713   // After the first and second level caches, I see two common sorts of
714   // behavior: 1) a lot of searched FileID's are "near" the cached file
715   // location or are "near" the cached expansion location. 2) others are just
716   // completely random and may be a very long way away.
717   //
718   // To handle this, we do a linear search for up to 8 steps to catch #1 quickly
719   // then we fall back to a less cache efficient, but more scalable, binary
720   // search to find the location.
721 
722   // See if this is near the file point - worst case we start scanning from the
723   // most newly created FileID.
724   std::vector<SrcMgr::SLocEntry>::const_iterator I;
725 
726   if (LastFileIDLookup.ID < 0 ||
727       LocalSLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) {
728     // Neither loc prunes our search.
729     I = LocalSLocEntryTable.end();
730   } else {
731     // Perhaps it is near the file point.
732     I = LocalSLocEntryTable.begin()+LastFileIDLookup.ID;
733   }
734 
735   // Find the FileID that contains this.  "I" is an iterator that points to a
736   // FileID whose offset is known to be larger than SLocOffset.
737   unsigned NumProbes = 0;
738   while (1) {
739     --I;
740     if (I->getOffset() <= SLocOffset) {
741       FileID Res = FileID::get(int(I - LocalSLocEntryTable.begin()));
742 
743       // If this isn't an expansion, remember it.  We have good locality across
744       // FileID lookups.
745       if (!I->isExpansion())
746         LastFileIDLookup = Res;
747       NumLinearScans += NumProbes+1;
748       return Res;
749     }
750     if (++NumProbes == 8)
751       break;
752   }
753 
754   // Convert "I" back into an index.  We know that it is an entry whose index is
755   // larger than the offset we are looking for.
756   unsigned GreaterIndex = I - LocalSLocEntryTable.begin();
757   // LessIndex - This is the lower bound of the range that we're searching.
758   // We know that the offset corresponding to the FileID is is less than
759   // SLocOffset.
760   unsigned LessIndex = 0;
761   NumProbes = 0;
762   while (1) {
763     bool Invalid = false;
764     unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex;
765     unsigned MidOffset = getLocalSLocEntry(MiddleIndex, &Invalid).getOffset();
766     if (Invalid)
767       return FileID::get(0);
768 
769     ++NumProbes;
770 
771     // If the offset of the midpoint is too large, chop the high side of the
772     // range to the midpoint.
773     if (MidOffset > SLocOffset) {
774       GreaterIndex = MiddleIndex;
775       continue;
776     }
777 
778     // If the middle index contains the value, succeed and return.
779     // FIXME: This could be made faster by using a function that's aware of
780     // being in the local area.
781     if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) {
782       FileID Res = FileID::get(MiddleIndex);
783 
784       // If this isn't a macro expansion, remember it.  We have good locality
785       // across FileID lookups.
786       if (!LocalSLocEntryTable[MiddleIndex].isExpansion())
787         LastFileIDLookup = Res;
788       NumBinaryProbes += NumProbes;
789       return Res;
790     }
791 
792     // Otherwise, move the low-side up to the middle index.
793     LessIndex = MiddleIndex;
794   }
795 }
796 
797 /// \brief Return the FileID for a SourceLocation with a high offset.
798 ///
799 /// This function knows that the SourceLocation is in a loaded buffer, not a
800 /// local one.
801 FileID SourceManager::getFileIDLoaded(unsigned SLocOffset) const {
802   // Sanity checking, otherwise a bug may lead to hanging in release build.
803   if (SLocOffset < CurrentLoadedOffset) {
804     assert(0 && "Invalid SLocOffset or bad function choice");
805     return FileID();
806   }
807 
808   // Essentially the same as the local case, but the loaded array is sorted
809   // in the other direction.
810 
811   // First do a linear scan from the last lookup position, if possible.
812   unsigned I;
813   int LastID = LastFileIDLookup.ID;
814   if (LastID >= 0 || getLoadedSLocEntryByID(LastID).getOffset() < SLocOffset)
815     I = 0;
816   else
817     I = (-LastID - 2) + 1;
818 
819   unsigned NumProbes;
820   for (NumProbes = 0; NumProbes < 8; ++NumProbes, ++I) {
821     // Make sure the entry is loaded!
822     const SrcMgr::SLocEntry &E = getLoadedSLocEntry(I);
823     if (E.getOffset() <= SLocOffset) {
824       FileID Res = FileID::get(-int(I) - 2);
825 
826       if (!E.isExpansion())
827         LastFileIDLookup = Res;
828       NumLinearScans += NumProbes + 1;
829       return Res;
830     }
831   }
832 
833   // Linear scan failed. Do the binary search. Note the reverse sorting of the
834   // table: GreaterIndex is the one where the offset is greater, which is
835   // actually a lower index!
836   unsigned GreaterIndex = I;
837   unsigned LessIndex = LoadedSLocEntryTable.size();
838   NumProbes = 0;
839   while (1) {
840     ++NumProbes;
841     unsigned MiddleIndex = (LessIndex - GreaterIndex) / 2 + GreaterIndex;
842     const SrcMgr::SLocEntry &E = getLoadedSLocEntry(MiddleIndex);
843 
844     ++NumProbes;
845 
846     if (E.getOffset() > SLocOffset) {
847       GreaterIndex = MiddleIndex;
848       continue;
849     }
850 
851     if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) {
852       FileID Res = FileID::get(-int(MiddleIndex) - 2);
853       if (!E.isExpansion())
854         LastFileIDLookup = Res;
855       NumBinaryProbes += NumProbes;
856       return Res;
857     }
858 
859     LessIndex = MiddleIndex;
860   }
861 }
862 
863 SourceLocation SourceManager::
864 getExpansionLocSlowCase(SourceLocation Loc) const {
865   do {
866     // Note: If Loc indicates an offset into a token that came from a macro
867     // expansion (e.g. the 5th character of the token) we do not want to add
868     // this offset when going to the expansion location.  The expansion
869     // location is the macro invocation, which the offset has nothing to do
870     // with.  This is unlike when we get the spelling loc, because the offset
871     // directly correspond to the token whose spelling we're inspecting.
872     Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart();
873   } while (!Loc.isFileID());
874 
875   return Loc;
876 }
877 
878 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
879   do {
880     std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
881     Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
882     Loc = Loc.getLocWithOffset(LocInfo.second);
883   } while (!Loc.isFileID());
884   return Loc;
885 }
886 
887 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const {
888   do {
889     if (isMacroArgExpansion(Loc))
890       Loc = getImmediateSpellingLoc(Loc);
891     else
892       Loc = getImmediateExpansionRange(Loc).first;
893   } while (!Loc.isFileID());
894   return Loc;
895 }
896 
897 
898 std::pair<FileID, unsigned>
899 SourceManager::getDecomposedExpansionLocSlowCase(
900                                              const SrcMgr::SLocEntry *E) const {
901   // If this is an expansion record, walk through all the expansion points.
902   FileID FID;
903   SourceLocation Loc;
904   unsigned Offset;
905   do {
906     Loc = E->getExpansion().getExpansionLocStart();
907 
908     FID = getFileID(Loc);
909     E = &getSLocEntry(FID);
910     Offset = Loc.getOffset()-E->getOffset();
911   } while (!Loc.isFileID());
912 
913   return std::make_pair(FID, Offset);
914 }
915 
916 std::pair<FileID, unsigned>
917 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
918                                                 unsigned Offset) const {
919   // If this is an expansion record, walk through all the expansion points.
920   FileID FID;
921   SourceLocation Loc;
922   do {
923     Loc = E->getExpansion().getSpellingLoc();
924     Loc = Loc.getLocWithOffset(Offset);
925 
926     FID = getFileID(Loc);
927     E = &getSLocEntry(FID);
928     Offset = Loc.getOffset()-E->getOffset();
929   } while (!Loc.isFileID());
930 
931   return std::make_pair(FID, Offset);
932 }
933 
934 /// getImmediateSpellingLoc - Given a SourceLocation object, return the
935 /// spelling location referenced by the ID.  This is the first level down
936 /// towards the place where the characters that make up the lexed token can be
937 /// found.  This should not generally be used by clients.
938 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
939   if (Loc.isFileID()) return Loc;
940   std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
941   Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
942   return Loc.getLocWithOffset(LocInfo.second);
943 }
944 
945 
946 /// getImmediateExpansionRange - Loc is required to be an expansion location.
947 /// Return the start/end of the expansion information.
948 std::pair<SourceLocation,SourceLocation>
949 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const {
950   assert(Loc.isMacroID() && "Not a macro expansion loc!");
951   const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion();
952   return Expansion.getExpansionLocRange();
953 }
954 
955 /// getExpansionRange - Given a SourceLocation object, return the range of
956 /// tokens covered by the expansion in the ultimate file.
957 std::pair<SourceLocation,SourceLocation>
958 SourceManager::getExpansionRange(SourceLocation Loc) const {
959   if (Loc.isFileID()) return std::make_pair(Loc, Loc);
960 
961   std::pair<SourceLocation,SourceLocation> Res =
962     getImmediateExpansionRange(Loc);
963 
964   // Fully resolve the start and end locations to their ultimate expansion
965   // points.
966   while (!Res.first.isFileID())
967     Res.first = getImmediateExpansionRange(Res.first).first;
968   while (!Res.second.isFileID())
969     Res.second = getImmediateExpansionRange(Res.second).second;
970   return Res;
971 }
972 
973 bool SourceManager::isMacroArgExpansion(SourceLocation Loc) const {
974   if (!Loc.isMacroID()) return false;
975 
976   FileID FID = getFileID(Loc);
977   const SrcMgr::SLocEntry *E = &getSLocEntry(FID);
978   const SrcMgr::ExpansionInfo &Expansion = E->getExpansion();
979   return Expansion.isMacroArgExpansion();
980 }
981 
982 
983 //===----------------------------------------------------------------------===//
984 // Queries about the code at a SourceLocation.
985 //===----------------------------------------------------------------------===//
986 
987 /// getCharacterData - Return a pointer to the start of the specified location
988 /// in the appropriate MemoryBuffer.
989 const char *SourceManager::getCharacterData(SourceLocation SL,
990                                             bool *Invalid) const {
991   // Note that this is a hot function in the getSpelling() path, which is
992   // heavily used by -E mode.
993   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
994 
995   // Note that calling 'getBuffer()' may lazily page in a source file.
996   bool CharDataInvalid = false;
997   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid);
998   if (CharDataInvalid || !Entry.isFile()) {
999     if (Invalid)
1000       *Invalid = true;
1001 
1002     return "<<<<INVALID BUFFER>>>>";
1003   }
1004   const llvm::MemoryBuffer *Buffer
1005     = Entry.getFile().getContentCache()
1006                   ->getBuffer(Diag, *this, SourceLocation(), &CharDataInvalid);
1007   if (Invalid)
1008     *Invalid = CharDataInvalid;
1009   return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
1010 }
1011 
1012 
1013 /// getColumnNumber - Return the column # for the specified file position.
1014 /// this is significantly cheaper to compute than the line number.
1015 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
1016                                         bool *Invalid) const {
1017   bool MyInvalid = false;
1018   const llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid);
1019   if (Invalid)
1020     *Invalid = MyInvalid;
1021 
1022   if (MyInvalid)
1023     return 1;
1024 
1025   // It is okay to request a position just past the end of the buffer.
1026   if (FilePos > MemBuf->getBufferSize()) {
1027     if (Invalid)
1028       *Invalid = true;
1029     return 1;
1030   }
1031 
1032   // See if we just calculated the line number for this FilePos and can use
1033   // that to lookup the start of the line instead of searching for it.
1034   if (LastLineNoFileIDQuery == FID &&
1035       LastLineNoContentCache->SourceLineCache != 0) {
1036     unsigned *SourceLineCache = LastLineNoContentCache->SourceLineCache;
1037     unsigned LineStart = SourceLineCache[LastLineNoResult - 1];
1038     unsigned LineEnd = SourceLineCache[LastLineNoResult];
1039     if (FilePos >= LineStart && FilePos < LineEnd)
1040       return FilePos - LineStart + 1;
1041   }
1042 
1043   const char *Buf = MemBuf->getBufferStart();
1044   unsigned LineStart = FilePos;
1045   while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
1046     --LineStart;
1047   return FilePos-LineStart+1;
1048 }
1049 
1050 // isInvalid - Return the result of calling loc.isInvalid(), and
1051 // if Invalid is not null, set its value to same.
1052 static bool isInvalid(SourceLocation Loc, bool *Invalid) {
1053   bool MyInvalid = Loc.isInvalid();
1054   if (Invalid)
1055     *Invalid = MyInvalid;
1056   return MyInvalid;
1057 }
1058 
1059 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
1060                                                 bool *Invalid) const {
1061   if (isInvalid(Loc, Invalid)) return 0;
1062   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1063   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1064 }
1065 
1066 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc,
1067                                                  bool *Invalid) const {
1068   if (isInvalid(Loc, Invalid)) return 0;
1069   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1070   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1071 }
1072 
1073 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc,
1074                                                 bool *Invalid) const {
1075   if (isInvalid(Loc, Invalid)) return 0;
1076   return getPresumedLoc(Loc).getColumn();
1077 }
1078 
1079 #ifdef __SSE2__
1080 #include <emmintrin.h>
1081 #endif
1082 
1083 static LLVM_ATTRIBUTE_NOINLINE void
1084 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1085                    llvm::BumpPtrAllocator &Alloc,
1086                    const SourceManager &SM, bool &Invalid);
1087 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1088                                llvm::BumpPtrAllocator &Alloc,
1089                                const SourceManager &SM, bool &Invalid) {
1090   // Note that calling 'getBuffer()' may lazily page in the file.
1091   const MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(),
1092                                              &Invalid);
1093   if (Invalid)
1094     return;
1095 
1096   // Find the file offsets of all of the *physical* source lines.  This does
1097   // not look at trigraphs, escaped newlines, or anything else tricky.
1098   SmallVector<unsigned, 256> LineOffsets;
1099 
1100   // Line #1 starts at char 0.
1101   LineOffsets.push_back(0);
1102 
1103   const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
1104   const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
1105   unsigned Offs = 0;
1106   while (1) {
1107     // Skip over the contents of the line.
1108     const unsigned char *NextBuf = (const unsigned char *)Buf;
1109 
1110 #ifdef __SSE2__
1111     // Try to skip to the next newline using SSE instructions. This is very
1112     // performance sensitive for programs with lots of diagnostics and in -E
1113     // mode.
1114     __m128i CRs = _mm_set1_epi8('\r');
1115     __m128i LFs = _mm_set1_epi8('\n');
1116 
1117     // First fix up the alignment to 16 bytes.
1118     while (((uintptr_t)NextBuf & 0xF) != 0) {
1119       if (*NextBuf == '\n' || *NextBuf == '\r' || *NextBuf == '\0')
1120         goto FoundSpecialChar;
1121       ++NextBuf;
1122     }
1123 
1124     // Scan 16 byte chunks for '\r' and '\n'. Ignore '\0'.
1125     while (NextBuf+16 <= End) {
1126       const __m128i Chunk = *(const __m128i*)NextBuf;
1127       __m128i Cmp = _mm_or_si128(_mm_cmpeq_epi8(Chunk, CRs),
1128                                  _mm_cmpeq_epi8(Chunk, LFs));
1129       unsigned Mask = _mm_movemask_epi8(Cmp);
1130 
1131       // If we found a newline, adjust the pointer and jump to the handling code.
1132       if (Mask != 0) {
1133         NextBuf += llvm::CountTrailingZeros_32(Mask);
1134         goto FoundSpecialChar;
1135       }
1136       NextBuf += 16;
1137     }
1138 #endif
1139 
1140     while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
1141       ++NextBuf;
1142 
1143 #ifdef __SSE2__
1144 FoundSpecialChar:
1145 #endif
1146     Offs += NextBuf-Buf;
1147     Buf = NextBuf;
1148 
1149     if (Buf[0] == '\n' || Buf[0] == '\r') {
1150       // If this is \n\r or \r\n, skip both characters.
1151       if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1])
1152         ++Offs, ++Buf;
1153       ++Offs, ++Buf;
1154       LineOffsets.push_back(Offs);
1155     } else {
1156       // Otherwise, this is a null.  If end of file, exit.
1157       if (Buf == End) break;
1158       // Otherwise, skip the null.
1159       ++Offs, ++Buf;
1160     }
1161   }
1162 
1163   // Copy the offsets into the FileInfo structure.
1164   FI->NumLines = LineOffsets.size();
1165   FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
1166   std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
1167 }
1168 
1169 /// getLineNumber - Given a SourceLocation, return the spelling line number
1170 /// for the position indicated.  This requires building and caching a table of
1171 /// line offsets for the MemoryBuffer, so this is not cheap: use only when
1172 /// about to emit a diagnostic.
1173 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
1174                                       bool *Invalid) const {
1175   if (FID.isInvalid()) {
1176     if (Invalid)
1177       *Invalid = true;
1178     return 1;
1179   }
1180 
1181   ContentCache *Content;
1182   if (LastLineNoFileIDQuery == FID)
1183     Content = LastLineNoContentCache;
1184   else {
1185     bool MyInvalid = false;
1186     const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid);
1187     if (MyInvalid || !Entry.isFile()) {
1188       if (Invalid)
1189         *Invalid = true;
1190       return 1;
1191     }
1192 
1193     Content = const_cast<ContentCache*>(Entry.getFile().getContentCache());
1194   }
1195 
1196   // If this is the first use of line information for this buffer, compute the
1197   /// SourceLineCache for it on demand.
1198   if (Content->SourceLineCache == 0) {
1199     bool MyInvalid = false;
1200     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1201     if (Invalid)
1202       *Invalid = MyInvalid;
1203     if (MyInvalid)
1204       return 1;
1205   } else if (Invalid)
1206     *Invalid = false;
1207 
1208   // Okay, we know we have a line number table.  Do a binary search to find the
1209   // line number that this character position lands on.
1210   unsigned *SourceLineCache = Content->SourceLineCache;
1211   unsigned *SourceLineCacheStart = SourceLineCache;
1212   unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
1213 
1214   unsigned QueriedFilePos = FilePos+1;
1215 
1216   // FIXME: I would like to be convinced that this code is worth being as
1217   // complicated as it is, binary search isn't that slow.
1218   //
1219   // If it is worth being optimized, then in my opinion it could be more
1220   // performant, simpler, and more obviously correct by just "galloping" outward
1221   // from the queried file position. In fact, this could be incorporated into a
1222   // generic algorithm such as lower_bound_with_hint.
1223   //
1224   // If someone gives me a test case where this matters, and I will do it! - DWD
1225 
1226   // If the previous query was to the same file, we know both the file pos from
1227   // that query and the line number returned.  This allows us to narrow the
1228   // search space from the entire file to something near the match.
1229   if (LastLineNoFileIDQuery == FID) {
1230     if (QueriedFilePos >= LastLineNoFilePos) {
1231       // FIXME: Potential overflow?
1232       SourceLineCache = SourceLineCache+LastLineNoResult-1;
1233 
1234       // The query is likely to be nearby the previous one.  Here we check to
1235       // see if it is within 5, 10 or 20 lines.  It can be far away in cases
1236       // where big comment blocks and vertical whitespace eat up lines but
1237       // contribute no tokens.
1238       if (SourceLineCache+5 < SourceLineCacheEnd) {
1239         if (SourceLineCache[5] > QueriedFilePos)
1240           SourceLineCacheEnd = SourceLineCache+5;
1241         else if (SourceLineCache+10 < SourceLineCacheEnd) {
1242           if (SourceLineCache[10] > QueriedFilePos)
1243             SourceLineCacheEnd = SourceLineCache+10;
1244           else if (SourceLineCache+20 < SourceLineCacheEnd) {
1245             if (SourceLineCache[20] > QueriedFilePos)
1246               SourceLineCacheEnd = SourceLineCache+20;
1247           }
1248         }
1249       }
1250     } else {
1251       if (LastLineNoResult < Content->NumLines)
1252         SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
1253     }
1254   }
1255 
1256   // If the spread is large, do a "radix" test as our initial guess, based on
1257   // the assumption that lines average to approximately the same length.
1258   // NOTE: This is currently disabled, as it does not appear to be profitable in
1259   // initial measurements.
1260   if (0 && SourceLineCacheEnd-SourceLineCache > 20) {
1261     unsigned FileLen = Content->SourceLineCache[Content->NumLines-1];
1262 
1263     // Take a stab at guessing where it is.
1264     unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen;
1265 
1266     // Check for -10 and +10 lines.
1267     unsigned LowerBound = std::max(int(ApproxPos-10), 0);
1268     unsigned UpperBound = std::min(ApproxPos+10, FileLen);
1269 
1270     // If the computed lower bound is less than the query location, move it in.
1271     if (SourceLineCache < SourceLineCacheStart+LowerBound &&
1272         SourceLineCacheStart[LowerBound] < QueriedFilePos)
1273       SourceLineCache = SourceLineCacheStart+LowerBound;
1274 
1275     // If the computed upper bound is greater than the query location, move it.
1276     if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound &&
1277         SourceLineCacheStart[UpperBound] >= QueriedFilePos)
1278       SourceLineCacheEnd = SourceLineCacheStart+UpperBound;
1279   }
1280 
1281   unsigned *Pos
1282     = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
1283   unsigned LineNo = Pos-SourceLineCacheStart;
1284 
1285   LastLineNoFileIDQuery = FID;
1286   LastLineNoContentCache = Content;
1287   LastLineNoFilePos = QueriedFilePos;
1288   LastLineNoResult = LineNo;
1289   return LineNo;
1290 }
1291 
1292 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
1293                                               bool *Invalid) const {
1294   if (isInvalid(Loc, Invalid)) return 0;
1295   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1296   return getLineNumber(LocInfo.first, LocInfo.second);
1297 }
1298 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc,
1299                                                bool *Invalid) const {
1300   if (isInvalid(Loc, Invalid)) return 0;
1301   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1302   return getLineNumber(LocInfo.first, LocInfo.second);
1303 }
1304 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc,
1305                                               bool *Invalid) const {
1306   if (isInvalid(Loc, Invalid)) return 0;
1307   return getPresumedLoc(Loc).getLine();
1308 }
1309 
1310 /// getFileCharacteristic - return the file characteristic of the specified
1311 /// source location, indicating whether this is a normal file, a system
1312 /// header, or an "implicit extern C" system header.
1313 ///
1314 /// This state can be modified with flags on GNU linemarker directives like:
1315 ///   # 4 "foo.h" 3
1316 /// which changes all source locations in the current file after that to be
1317 /// considered to be from a system header.
1318 SrcMgr::CharacteristicKind
1319 SourceManager::getFileCharacteristic(SourceLocation Loc) const {
1320   assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!");
1321   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1322   bool Invalid = false;
1323   const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid);
1324   if (Invalid || !SEntry.isFile())
1325     return C_User;
1326 
1327   const SrcMgr::FileInfo &FI = SEntry.getFile();
1328 
1329   // If there are no #line directives in this file, just return the whole-file
1330   // state.
1331   if (!FI.hasLineDirectives())
1332     return FI.getFileCharacteristic();
1333 
1334   assert(LineTable && "Can't have linetable entries without a LineTable!");
1335   // See if there is a #line directive before the location.
1336   const LineEntry *Entry =
1337     LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second);
1338 
1339   // If this is before the first line marker, use the file characteristic.
1340   if (!Entry)
1341     return FI.getFileCharacteristic();
1342 
1343   return Entry->FileKind;
1344 }
1345 
1346 /// Return the filename or buffer identifier of the buffer the location is in.
1347 /// Note that this name does not respect \#line directives.  Use getPresumedLoc
1348 /// for normal clients.
1349 const char *SourceManager::getBufferName(SourceLocation Loc,
1350                                          bool *Invalid) const {
1351   if (isInvalid(Loc, Invalid)) return "<invalid loc>";
1352 
1353   return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
1354 }
1355 
1356 
1357 /// getPresumedLoc - This method returns the "presumed" location of a
1358 /// SourceLocation specifies.  A "presumed location" can be modified by \#line
1359 /// or GNU line marker directives.  This provides a view on the data that a
1360 /// user should see in diagnostics, for example.
1361 ///
1362 /// Note that a presumed location is always given as the expansion point of an
1363 /// expansion location, not at the spelling location.
1364 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc,
1365                                           bool UseLineDirectives) const {
1366   if (Loc.isInvalid()) return PresumedLoc();
1367 
1368   // Presumed locations are always for expansion points.
1369   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1370 
1371   bool Invalid = false;
1372   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
1373   if (Invalid || !Entry.isFile())
1374     return PresumedLoc();
1375 
1376   const SrcMgr::FileInfo &FI = Entry.getFile();
1377   const SrcMgr::ContentCache *C = FI.getContentCache();
1378 
1379   // To get the source name, first consult the FileEntry (if one exists)
1380   // before the MemBuffer as this will avoid unnecessarily paging in the
1381   // MemBuffer.
1382   const char *Filename;
1383   if (C->OrigEntry)
1384     Filename = C->OrigEntry->getName();
1385   else
1386     Filename = C->getBuffer(Diag, *this)->getBufferIdentifier();
1387 
1388   unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid);
1389   if (Invalid)
1390     return PresumedLoc();
1391   unsigned ColNo  = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid);
1392   if (Invalid)
1393     return PresumedLoc();
1394 
1395   SourceLocation IncludeLoc = FI.getIncludeLoc();
1396 
1397   // If we have #line directives in this file, update and overwrite the physical
1398   // location info if appropriate.
1399   if (UseLineDirectives && FI.hasLineDirectives()) {
1400     assert(LineTable && "Can't have linetable entries without a LineTable!");
1401     // See if there is a #line directive before this.  If so, get it.
1402     if (const LineEntry *Entry =
1403           LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) {
1404       // If the LineEntry indicates a filename, use it.
1405       if (Entry->FilenameID != -1)
1406         Filename = LineTable->getFilename(Entry->FilenameID);
1407 
1408       // Use the line number specified by the LineEntry.  This line number may
1409       // be multiple lines down from the line entry.  Add the difference in
1410       // physical line numbers from the query point and the line marker to the
1411       // total.
1412       unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
1413       LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
1414 
1415       // Note that column numbers are not molested by line markers.
1416 
1417       // Handle virtual #include manipulation.
1418       if (Entry->IncludeOffset) {
1419         IncludeLoc = getLocForStartOfFile(LocInfo.first);
1420         IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset);
1421       }
1422     }
1423   }
1424 
1425   return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc);
1426 }
1427 
1428 /// \brief The size of the SLocEnty that \arg FID represents.
1429 unsigned SourceManager::getFileIDSize(FileID FID) const {
1430   bool Invalid = false;
1431   const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1432   if (Invalid)
1433     return 0;
1434 
1435   int ID = FID.ID;
1436   unsigned NextOffset;
1437   if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size()))
1438     NextOffset = getNextLocalOffset();
1439   else if (ID+1 == -1)
1440     NextOffset = MaxLoadedOffset;
1441   else
1442     NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset();
1443 
1444   return NextOffset - Entry.getOffset() - 1;
1445 }
1446 
1447 //===----------------------------------------------------------------------===//
1448 // Other miscellaneous methods.
1449 //===----------------------------------------------------------------------===//
1450 
1451 /// \brief Retrieve the inode for the given file entry, if possible.
1452 ///
1453 /// This routine involves a system call, and therefore should only be used
1454 /// in non-performance-critical code.
1455 static llvm::Optional<ino_t> getActualFileInode(const FileEntry *File) {
1456   if (!File)
1457     return llvm::Optional<ino_t>();
1458 
1459   struct stat StatBuf;
1460   if (::stat(File->getName(), &StatBuf))
1461     return llvm::Optional<ino_t>();
1462 
1463   return StatBuf.st_ino;
1464 }
1465 
1466 /// \brief Get the source location for the given file:line:col triplet.
1467 ///
1468 /// If the source file is included multiple times, the source location will
1469 /// be based upon an arbitrary inclusion.
1470 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile,
1471                                                   unsigned Line,
1472                                                   unsigned Col) const {
1473   assert(SourceFile && "Null source file!");
1474   assert(Line && Col && "Line and column should start from 1!");
1475 
1476   FileID FirstFID = translateFile(SourceFile);
1477   return translateLineCol(FirstFID, Line, Col);
1478 }
1479 
1480 /// \brief Get the FileID for the given file.
1481 ///
1482 /// If the source file is included multiple times, the FileID will be the
1483 /// first inclusion.
1484 FileID SourceManager::translateFile(const FileEntry *SourceFile) const {
1485   assert(SourceFile && "Null source file!");
1486 
1487   // Find the first file ID that corresponds to the given file.
1488   FileID FirstFID;
1489 
1490   // First, check the main file ID, since it is common to look for a
1491   // location in the main file.
1492   llvm::Optional<ino_t> SourceFileInode;
1493   llvm::Optional<StringRef> SourceFileName;
1494   if (!MainFileID.isInvalid()) {
1495     bool Invalid = false;
1496     const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid);
1497     if (Invalid)
1498       return FileID();
1499 
1500     if (MainSLoc.isFile()) {
1501       const ContentCache *MainContentCache
1502         = MainSLoc.getFile().getContentCache();
1503       if (!MainContentCache) {
1504         // Can't do anything
1505       } else if (MainContentCache->OrigEntry == SourceFile) {
1506         FirstFID = MainFileID;
1507       } else {
1508         // Fall back: check whether we have the same base name and inode
1509         // as the main file.
1510         const FileEntry *MainFile = MainContentCache->OrigEntry;
1511         SourceFileName = llvm::sys::path::filename(SourceFile->getName());
1512         if (*SourceFileName == llvm::sys::path::filename(MainFile->getName())) {
1513           SourceFileInode = getActualFileInode(SourceFile);
1514           if (SourceFileInode) {
1515             if (llvm::Optional<ino_t> MainFileInode
1516                                                = getActualFileInode(MainFile)) {
1517               if (*SourceFileInode == *MainFileInode) {
1518                 FirstFID = MainFileID;
1519                 SourceFile = MainFile;
1520               }
1521             }
1522           }
1523         }
1524       }
1525     }
1526   }
1527 
1528   if (FirstFID.isInvalid()) {
1529     // The location we're looking for isn't in the main file; look
1530     // through all of the local source locations.
1531     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1532       bool Invalid = false;
1533       const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid);
1534       if (Invalid)
1535         return FileID();
1536 
1537       if (SLoc.isFile() &&
1538           SLoc.getFile().getContentCache() &&
1539           SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1540         FirstFID = FileID::get(I);
1541         break;
1542       }
1543     }
1544     // If that still didn't help, try the modules.
1545     if (FirstFID.isInvalid()) {
1546       for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) {
1547         const SLocEntry &SLoc = getLoadedSLocEntry(I);
1548         if (SLoc.isFile() &&
1549             SLoc.getFile().getContentCache() &&
1550             SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1551           FirstFID = FileID::get(-int(I) - 2);
1552           break;
1553         }
1554       }
1555     }
1556   }
1557 
1558   // If we haven't found what we want yet, try again, but this time stat()
1559   // each of the files in case the files have changed since we originally
1560   // parsed the file.
1561   if (FirstFID.isInvalid() &&
1562       (SourceFileName ||
1563        (SourceFileName = llvm::sys::path::filename(SourceFile->getName()))) &&
1564       (SourceFileInode ||
1565        (SourceFileInode = getActualFileInode(SourceFile)))) {
1566     bool Invalid = false;
1567     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1568       FileID IFileID;
1569       IFileID.ID = I;
1570       const SLocEntry &SLoc = getSLocEntry(IFileID, &Invalid);
1571       if (Invalid)
1572         return FileID();
1573 
1574       if (SLoc.isFile()) {
1575         const ContentCache *FileContentCache
1576           = SLoc.getFile().getContentCache();
1577       const FileEntry *Entry =FileContentCache? FileContentCache->OrigEntry : 0;
1578         if (Entry &&
1579             *SourceFileName == llvm::sys::path::filename(Entry->getName())) {
1580           if (llvm::Optional<ino_t> EntryInode = getActualFileInode(Entry)) {
1581             if (*SourceFileInode == *EntryInode) {
1582               FirstFID = FileID::get(I);
1583               SourceFile = Entry;
1584               break;
1585             }
1586           }
1587         }
1588       }
1589     }
1590   }
1591 
1592   (void) SourceFile;
1593   return FirstFID;
1594 }
1595 
1596 /// \brief Get the source location in \arg FID for the given line:col.
1597 /// Returns null location if \arg FID is not a file SLocEntry.
1598 SourceLocation SourceManager::translateLineCol(FileID FID,
1599                                                unsigned Line,
1600                                                unsigned Col) const {
1601   if (FID.isInvalid())
1602     return SourceLocation();
1603 
1604   bool Invalid = false;
1605   const SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1606   if (Invalid)
1607     return SourceLocation();
1608 
1609   if (!Entry.isFile())
1610     return SourceLocation();
1611 
1612   SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset());
1613 
1614   if (Line == 1 && Col == 1)
1615     return FileLoc;
1616 
1617   ContentCache *Content
1618     = const_cast<ContentCache *>(Entry.getFile().getContentCache());
1619   if (!Content)
1620     return SourceLocation();
1621 
1622   // If this is the first use of line information for this buffer, compute the
1623   // SourceLineCache for it on demand.
1624   if (Content->SourceLineCache == 0) {
1625     bool MyInvalid = false;
1626     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1627     if (MyInvalid)
1628       return SourceLocation();
1629   }
1630 
1631   if (Line > Content->NumLines) {
1632     unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize();
1633     if (Size > 0)
1634       --Size;
1635     return FileLoc.getLocWithOffset(Size);
1636   }
1637 
1638   const llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, *this);
1639   unsigned FilePos = Content->SourceLineCache[Line - 1];
1640   const char *Buf = Buffer->getBufferStart() + FilePos;
1641   unsigned BufLength = Buffer->getBufferSize() - FilePos;
1642   if (BufLength == 0)
1643     return FileLoc.getLocWithOffset(FilePos);
1644 
1645   unsigned i = 0;
1646 
1647   // Check that the given column is valid.
1648   while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
1649     ++i;
1650   if (i < Col-1)
1651     return FileLoc.getLocWithOffset(FilePos + i);
1652 
1653   return FileLoc.getLocWithOffset(FilePos + Col - 1);
1654 }
1655 
1656 /// \brief Compute a map of macro argument chunks to their expanded source
1657 /// location. Chunks that are not part of a macro argument will map to an
1658 /// invalid source location. e.g. if a file contains one macro argument at
1659 /// offset 100 with length 10, this is how the map will be formed:
1660 ///     0   -> SourceLocation()
1661 ///     100 -> Expanded macro arg location
1662 ///     110 -> SourceLocation()
1663 void SourceManager::computeMacroArgsCache(MacroArgsMap *&CachePtr,
1664                                           FileID FID) const {
1665   assert(!FID.isInvalid());
1666   assert(!CachePtr);
1667 
1668   CachePtr = new MacroArgsMap();
1669   MacroArgsMap &MacroArgsCache = *CachePtr;
1670   // Initially no macro argument chunk is present.
1671   MacroArgsCache.insert(std::make_pair(0, SourceLocation()));
1672 
1673   int ID = FID.ID;
1674   while (1) {
1675     ++ID;
1676     // Stop if there are no more FileIDs to check.
1677     if (ID > 0) {
1678       if (unsigned(ID) >= local_sloc_entry_size())
1679         return;
1680     } else if (ID == -1) {
1681       return;
1682     }
1683 
1684     const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID);
1685     if (Entry.isFile()) {
1686       SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc();
1687       if (IncludeLoc.isInvalid())
1688         continue;
1689       if (!isInFileID(IncludeLoc, FID))
1690         return; // No more files/macros that may be "contained" in this file.
1691 
1692       // Skip the files/macros of the #include'd file, we only care about macros
1693       // that lexed macro arguments from our file.
1694       if (Entry.getFile().NumCreatedFIDs)
1695         ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/;
1696       continue;
1697     }
1698 
1699     const ExpansionInfo &ExpInfo = Entry.getExpansion();
1700 
1701     if (ExpInfo.getExpansionLocStart().isFileID()) {
1702       if (!isInFileID(ExpInfo.getExpansionLocStart(), FID))
1703         return; // No more files/macros that may be "contained" in this file.
1704     }
1705 
1706     if (!ExpInfo.isMacroArgExpansion())
1707       continue;
1708 
1709     associateFileChunkWithMacroArgExp(MacroArgsCache, FID,
1710                                  ExpInfo.getSpellingLoc(),
1711                                  SourceLocation::getMacroLoc(Entry.getOffset()),
1712                                  getFileIDSize(FileID::get(ID)));
1713   }
1714 }
1715 
1716 void SourceManager::associateFileChunkWithMacroArgExp(
1717                                          MacroArgsMap &MacroArgsCache,
1718                                          FileID FID,
1719                                          SourceLocation SpellLoc,
1720                                          SourceLocation ExpansionLoc,
1721                                          unsigned ExpansionLength) const {
1722   if (!SpellLoc.isFileID()) {
1723     unsigned SpellBeginOffs = SpellLoc.getOffset();
1724     unsigned SpellEndOffs = SpellBeginOffs + ExpansionLength;
1725 
1726     // The spelling range for this macro argument expansion can span multiple
1727     // consecutive FileID entries. Go through each entry contained in the
1728     // spelling range and if one is itself a macro argument expansion, recurse
1729     // and associate the file chunk that it represents.
1730 
1731     FileID SpellFID; // Current FileID in the spelling range.
1732     unsigned SpellRelativeOffs;
1733     llvm::tie(SpellFID, SpellRelativeOffs) = getDecomposedLoc(SpellLoc);
1734     while (1) {
1735       const SLocEntry &Entry = getSLocEntry(SpellFID);
1736       unsigned SpellFIDBeginOffs = Entry.getOffset();
1737       unsigned SpellFIDSize = getFileIDSize(SpellFID);
1738       unsigned SpellFIDEndOffs = SpellFIDBeginOffs + SpellFIDSize;
1739       const ExpansionInfo &Info = Entry.getExpansion();
1740       if (Info.isMacroArgExpansion()) {
1741         unsigned CurrSpellLength;
1742         if (SpellFIDEndOffs < SpellEndOffs)
1743           CurrSpellLength = SpellFIDSize - SpellRelativeOffs;
1744         else
1745           CurrSpellLength = ExpansionLength;
1746         associateFileChunkWithMacroArgExp(MacroArgsCache, FID,
1747                       Info.getSpellingLoc().getLocWithOffset(SpellRelativeOffs),
1748                       ExpansionLoc, CurrSpellLength);
1749       }
1750 
1751       if (SpellFIDEndOffs >= SpellEndOffs)
1752         return; // we covered all FileID entries in the spelling range.
1753 
1754       // Move to the next FileID entry in the spelling range.
1755       unsigned advance = SpellFIDSize - SpellRelativeOffs + 1;
1756       ExpansionLoc = ExpansionLoc.getLocWithOffset(advance);
1757       ExpansionLength -= advance;
1758       ++SpellFID.ID;
1759       SpellRelativeOffs = 0;
1760     }
1761 
1762   }
1763 
1764   assert(SpellLoc.isFileID());
1765 
1766   unsigned BeginOffs;
1767   if (!isInFileID(SpellLoc, FID, &BeginOffs))
1768     return;
1769 
1770   unsigned EndOffs = BeginOffs + ExpansionLength;
1771 
1772   // Add a new chunk for this macro argument. A previous macro argument chunk
1773   // may have been lexed again, so e.g. if the map is
1774   //     0   -> SourceLocation()
1775   //     100 -> Expanded loc #1
1776   //     110 -> SourceLocation()
1777   // and we found a new macro FileID that lexed from offet 105 with length 3,
1778   // the new map will be:
1779   //     0   -> SourceLocation()
1780   //     100 -> Expanded loc #1
1781   //     105 -> Expanded loc #2
1782   //     108 -> Expanded loc #1
1783   //     110 -> SourceLocation()
1784   //
1785   // Since re-lexed macro chunks will always be the same size or less of
1786   // previous chunks, we only need to find where the ending of the new macro
1787   // chunk is mapped to and update the map with new begin/end mappings.
1788 
1789   MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs);
1790   --I;
1791   SourceLocation EndOffsMappedLoc = I->second;
1792   MacroArgsCache[BeginOffs] = ExpansionLoc;
1793   MacroArgsCache[EndOffs] = EndOffsMappedLoc;
1794 }
1795 
1796 /// \brief If \arg Loc points inside a function macro argument, the returned
1797 /// location will be the macro location in which the argument was expanded.
1798 /// If a macro argument is used multiple times, the expanded location will
1799 /// be at the first expansion of the argument.
1800 /// e.g.
1801 ///   MY_MACRO(foo);
1802 ///             ^
1803 /// Passing a file location pointing at 'foo', will yield a macro location
1804 /// where 'foo' was expanded into.
1805 SourceLocation
1806 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const {
1807   if (Loc.isInvalid() || !Loc.isFileID())
1808     return Loc;
1809 
1810   FileID FID;
1811   unsigned Offset;
1812   llvm::tie(FID, Offset) = getDecomposedLoc(Loc);
1813   if (FID.isInvalid())
1814     return Loc;
1815 
1816   MacroArgsMap *&MacroArgsCache = MacroArgsCacheMap[FID];
1817   if (!MacroArgsCache)
1818     computeMacroArgsCache(MacroArgsCache, FID);
1819 
1820   assert(!MacroArgsCache->empty());
1821   MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset);
1822   --I;
1823 
1824   unsigned MacroArgBeginOffs = I->first;
1825   SourceLocation MacroArgExpandedLoc = I->second;
1826   if (MacroArgExpandedLoc.isValid())
1827     return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs);
1828 
1829   return Loc;
1830 }
1831 
1832 /// Given a decomposed source location, move it up the include/expansion stack
1833 /// to the parent source location.  If this is possible, return the decomposed
1834 /// version of the parent in Loc and return false.  If Loc is the top-level
1835 /// entry, return true and don't modify it.
1836 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc,
1837                                    const SourceManager &SM) {
1838   SourceLocation UpperLoc;
1839   const SrcMgr::SLocEntry &Entry = SM.getSLocEntry(Loc.first);
1840   if (Entry.isExpansion())
1841     UpperLoc = Entry.getExpansion().getExpansionLocStart();
1842   else
1843     UpperLoc = Entry.getFile().getIncludeLoc();
1844 
1845   if (UpperLoc.isInvalid())
1846     return true; // We reached the top.
1847 
1848   Loc = SM.getDecomposedLoc(UpperLoc);
1849   return false;
1850 }
1851 
1852 
1853 /// \brief Determines the order of 2 source locations in the translation unit.
1854 ///
1855 /// \returns true if LHS source location comes before RHS, false otherwise.
1856 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
1857                                               SourceLocation RHS) const {
1858   assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
1859   if (LHS == RHS)
1860     return false;
1861 
1862   std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
1863   std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
1864 
1865   // If the source locations are in the same file, just compare offsets.
1866   if (LOffs.first == ROffs.first)
1867     return LOffs.second < ROffs.second;
1868 
1869   // If we are comparing a source location with multiple locations in the same
1870   // file, we get a big win by caching the result.
1871   if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first))
1872     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1873 
1874   // Okay, we missed in the cache, start updating the cache for this query.
1875   IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first,
1876                           /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID);
1877 
1878   // We need to find the common ancestor. The only way of doing this is to
1879   // build the complete include chain for one and then walking up the chain
1880   // of the other looking for a match.
1881   // We use a map from FileID to Offset to store the chain. Easier than writing
1882   // a custom set hash info that only depends on the first part of a pair.
1883   typedef llvm::DenseMap<FileID, unsigned> LocSet;
1884   LocSet LChain;
1885   do {
1886     LChain.insert(LOffs);
1887     // We catch the case where LOffs is in a file included by ROffs and
1888     // quit early. The other way round unfortunately remains suboptimal.
1889   } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this));
1890   LocSet::iterator I;
1891   while((I = LChain.find(ROffs.first)) == LChain.end()) {
1892     if (MoveUpIncludeHierarchy(ROffs, *this))
1893       break; // Met at topmost file.
1894   }
1895   if (I != LChain.end())
1896     LOffs = *I;
1897 
1898   // If we exited because we found a nearest common ancestor, compare the
1899   // locations within the common file and cache them.
1900   if (LOffs.first == ROffs.first) {
1901     IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second);
1902     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1903   }
1904 
1905   // This can happen if a location is in a built-ins buffer.
1906   // But see PR5662.
1907   // Clear the lookup cache, it depends on a common location.
1908   IsBeforeInTUCache.clear();
1909   bool LIsBuiltins = strcmp("<built-in>",
1910                             getBuffer(LOffs.first)->getBufferIdentifier()) == 0;
1911   bool RIsBuiltins = strcmp("<built-in>",
1912                             getBuffer(ROffs.first)->getBufferIdentifier()) == 0;
1913   // built-in is before non-built-in
1914   if (LIsBuiltins != RIsBuiltins)
1915     return LIsBuiltins;
1916   assert(LIsBuiltins && RIsBuiltins &&
1917          "Non-built-in locations must be rooted in the main file");
1918   // Both are in built-in buffers, but from different files. We just claim that
1919   // lower IDs come first.
1920   return LOffs.first < ROffs.first;
1921 }
1922 
1923 void SourceManager::PrintStats() const {
1924   llvm::errs() << "\n*** Source Manager Stats:\n";
1925   llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
1926                << " mem buffers mapped.\n";
1927   llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated ("
1928                << llvm::capacity_in_bytes(LocalSLocEntryTable)
1929                << " bytes of capacity), "
1930                << NextLocalOffset << "B of Sloc address space used.\n";
1931   llvm::errs() << LoadedSLocEntryTable.size()
1932                << " loaded SLocEntries allocated, "
1933                << MaxLoadedOffset - CurrentLoadedOffset
1934                << "B of Sloc address space used.\n";
1935 
1936   unsigned NumLineNumsComputed = 0;
1937   unsigned NumFileBytesMapped = 0;
1938   for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
1939     NumLineNumsComputed += I->second->SourceLineCache != 0;
1940     NumFileBytesMapped  += I->second->getSizeBytesMapped();
1941   }
1942   unsigned NumMacroArgsComputed = MacroArgsCacheMap.size();
1943 
1944   llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
1945                << NumLineNumsComputed << " files with line #'s computed, "
1946                << NumMacroArgsComputed << " files with macro args computed.\n";
1947   llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
1948                << NumBinaryProbes << " binary.\n";
1949 }
1950 
1951 ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
1952 
1953 /// Return the amount of memory used by memory buffers, breaking down
1954 /// by heap-backed versus mmap'ed memory.
1955 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const {
1956   size_t malloc_bytes = 0;
1957   size_t mmap_bytes = 0;
1958 
1959   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i)
1960     if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped())
1961       switch (MemBufferInfos[i]->getMemoryBufferKind()) {
1962         case llvm::MemoryBuffer::MemoryBuffer_MMap:
1963           mmap_bytes += sized_mapped;
1964           break;
1965         case llvm::MemoryBuffer::MemoryBuffer_Malloc:
1966           malloc_bytes += sized_mapped;
1967           break;
1968       }
1969 
1970   return MemoryBufferSizes(malloc_bytes, mmap_bytes);
1971 }
1972 
1973 size_t SourceManager::getDataStructureSizes() const {
1974   size_t size = llvm::capacity_in_bytes(MemBufferInfos)
1975     + llvm::capacity_in_bytes(LocalSLocEntryTable)
1976     + llvm::capacity_in_bytes(LoadedSLocEntryTable)
1977     + llvm::capacity_in_bytes(SLocEntryLoaded)
1978     + llvm::capacity_in_bytes(FileInfos);
1979 
1980   if (OverriddenFilesInfo)
1981     size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles);
1982 
1983   return size;
1984 }
1985