xref: /llvm-project/clang/lib/Basic/SourceManager.cpp (revision ccf43ca05cb1540f3d95a73b7b329bf926808c7b)
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/SourceManagerInternals.h"
16 #include "clang/Basic/Diagnostic.h"
17 #include "clang/Basic/FileManager.h"
18 #include "llvm/ADT/StringSwitch.h"
19 #include "llvm/ADT/Optional.h"
20 #include "llvm/ADT/STLExtras.h"
21 #include "llvm/Support/Compiler.h"
22 #include "llvm/Support/MemoryBuffer.h"
23 #include "llvm/Support/raw_ostream.h"
24 #include "llvm/Support/Path.h"
25 #include "llvm/Support/Capacity.h"
26 #include <algorithm>
27 #include <string>
28 #include <cstring>
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   Buffer.setPointer(SM.getFileManager().getBufferForFile(ContentsEntry, &ErrorStr));
101 
102   // If we were unable to open the file, then we are in an inconsistent
103   // situation where the content cache referenced a file which no longer
104   // exists. Most likely, we were using a stat cache with an invalid entry but
105   // the file could also have been removed during processing. Since we can't
106   // really deal with this situation, just create an empty buffer.
107   //
108   // FIXME: This is definitely not ideal, but our immediate clients can't
109   // currently handle returning a null entry here. Ideally we should detect
110   // that we are in an inconsistent situation and error out as quickly as
111   // possible.
112   if (!Buffer.getPointer()) {
113     const StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
114     Buffer.setPointer(MemoryBuffer::getNewMemBuffer(ContentsEntry->getSize(),
115                                                     "<invalid>"));
116     char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
117     for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i)
118       Ptr[i] = FillStr[i % FillStr.size()];
119 
120     if (Diag.isDiagnosticInFlight())
121       Diag.SetDelayedDiagnostic(diag::err_cannot_open_file,
122                                 ContentsEntry->getName(), ErrorStr);
123     else
124       Diag.Report(Loc, diag::err_cannot_open_file)
125         << ContentsEntry->getName() << ErrorStr;
126 
127     Buffer.setInt(Buffer.getInt() | InvalidFlag);
128 
129     if (Invalid) *Invalid = true;
130     return Buffer.getPointer();
131   }
132 
133   // Check that the file's size is the same as in the file entry (which may
134   // have come from a stat cache).
135   if (getRawBuffer()->getBufferSize() != (size_t)ContentsEntry->getSize()) {
136     if (Diag.isDiagnosticInFlight())
137       Diag.SetDelayedDiagnostic(diag::err_file_modified,
138                                 ContentsEntry->getName());
139     else
140       Diag.Report(Loc, diag::err_file_modified)
141         << ContentsEntry->getName();
142 
143     Buffer.setInt(Buffer.getInt() | InvalidFlag);
144     if (Invalid) *Invalid = true;
145     return Buffer.getPointer();
146   }
147 
148   // If the buffer is valid, check to see if it has a UTF Byte Order Mark
149   // (BOM).  We only support UTF-8 with and without a BOM right now.  See
150   // http://en.wikipedia.org/wiki/Byte_order_mark for more information.
151   StringRef BufStr = Buffer.getPointer()->getBuffer();
152   const char *InvalidBOM = llvm::StringSwitch<const char *>(BufStr)
153     .StartsWith("\xFE\xFF", "UTF-16 (BE)")
154     .StartsWith("\xFF\xFE", "UTF-16 (LE)")
155     .StartsWith("\x00\x00\xFE\xFF", "UTF-32 (BE)")
156     .StartsWith("\xFF\xFE\x00\x00", "UTF-32 (LE)")
157     .StartsWith("\x2B\x2F\x76", "UTF-7")
158     .StartsWith("\xF7\x64\x4C", "UTF-1")
159     .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC")
160     .StartsWith("\x0E\xFE\xFF", "SDSU")
161     .StartsWith("\xFB\xEE\x28", "BOCU-1")
162     .StartsWith("\x84\x31\x95\x33", "GB-18030")
163     .Default(0);
164 
165   if (InvalidBOM) {
166     Diag.Report(Loc, diag::err_unsupported_bom)
167       << InvalidBOM << ContentsEntry->getName();
168     Buffer.setInt(Buffer.getInt() | InvalidFlag);
169   }
170 
171   if (Invalid)
172     *Invalid = isBufferInvalid();
173 
174   return Buffer.getPointer();
175 }
176 
177 unsigned LineTableInfo::getLineTableFilenameID(StringRef Name) {
178   // Look up the filename in the string table, returning the pre-existing value
179   // if it exists.
180   llvm::StringMapEntry<unsigned> &Entry =
181     FilenameIDs.GetOrCreateValue(Name, ~0U);
182   if (Entry.getValue() != ~0U)
183     return Entry.getValue();
184 
185   // Otherwise, assign this the next available ID.
186   Entry.setValue(FilenamesByID.size());
187   FilenamesByID.push_back(&Entry);
188   return FilenamesByID.size()-1;
189 }
190 
191 /// AddLineNote - Add a line note to the line table that indicates that there
192 /// is a #line at the specified FID/Offset location which changes the presumed
193 /// location to LineNo/FilenameID.
194 void LineTableInfo::AddLineNote(int FID, unsigned Offset,
195                                 unsigned LineNo, int FilenameID) {
196   std::vector<LineEntry> &Entries = LineEntries[FID];
197 
198   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
199          "Adding line entries out of order!");
200 
201   SrcMgr::CharacteristicKind Kind = SrcMgr::C_User;
202   unsigned IncludeOffset = 0;
203 
204   if (!Entries.empty()) {
205     // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember
206     // that we are still in "foo.h".
207     if (FilenameID == -1)
208       FilenameID = Entries.back().FilenameID;
209 
210     // If we are after a line marker that switched us to system header mode, or
211     // that set #include information, preserve it.
212     Kind = Entries.back().FileKind;
213     IncludeOffset = Entries.back().IncludeOffset;
214   }
215 
216   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind,
217                                    IncludeOffset));
218 }
219 
220 /// AddLineNote This is the same as the previous version of AddLineNote, but is
221 /// used for GNU line markers.  If EntryExit is 0, then this doesn't change the
222 /// presumed #include stack.  If it is 1, this is a file entry, if it is 2 then
223 /// this is a file exit.  FileKind specifies whether this is a system header or
224 /// extern C system header.
225 void LineTableInfo::AddLineNote(int FID, unsigned Offset,
226                                 unsigned LineNo, int FilenameID,
227                                 unsigned EntryExit,
228                                 SrcMgr::CharacteristicKind FileKind) {
229   assert(FilenameID != -1 && "Unspecified filename should use other accessor");
230 
231   std::vector<LineEntry> &Entries = LineEntries[FID];
232 
233   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
234          "Adding line entries out of order!");
235 
236   unsigned IncludeOffset = 0;
237   if (EntryExit == 0) {  // No #include stack change.
238     IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset;
239   } else if (EntryExit == 1) {
240     IncludeOffset = Offset-1;
241   } else if (EntryExit == 2) {
242     assert(!Entries.empty() && Entries.back().IncludeOffset &&
243        "PPDirectives should have caught case when popping empty include stack");
244 
245     // Get the include loc of the last entries' include loc as our include loc.
246     IncludeOffset = 0;
247     if (const LineEntry *PrevEntry =
248           FindNearestLineEntry(FID, Entries.back().IncludeOffset))
249       IncludeOffset = PrevEntry->IncludeOffset;
250   }
251 
252   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind,
253                                    IncludeOffset));
254 }
255 
256 
257 /// FindNearestLineEntry - Find the line entry nearest to FID that is before
258 /// it.  If there is no line entry before Offset in FID, return null.
259 const LineEntry *LineTableInfo::FindNearestLineEntry(int FID,
260                                                      unsigned Offset) {
261   const std::vector<LineEntry> &Entries = LineEntries[FID];
262   assert(!Entries.empty() && "No #line entries for this FID after all!");
263 
264   // It is very common for the query to be after the last #line, check this
265   // first.
266   if (Entries.back().FileOffset <= Offset)
267     return &Entries.back();
268 
269   // Do a binary search to find the maximal element that is still before Offset.
270   std::vector<LineEntry>::const_iterator I =
271     std::upper_bound(Entries.begin(), Entries.end(), Offset);
272   if (I == Entries.begin()) return 0;
273   return &*--I;
274 }
275 
276 /// \brief Add a new line entry that has already been encoded into
277 /// the internal representation of the line table.
278 void LineTableInfo::AddEntry(int FID,
279                              const std::vector<LineEntry> &Entries) {
280   LineEntries[FID] = Entries;
281 }
282 
283 /// getLineTableFilenameID - Return the uniqued ID for the specified filename.
284 ///
285 unsigned SourceManager::getLineTableFilenameID(StringRef Name) {
286   if (LineTable == 0)
287     LineTable = new LineTableInfo();
288   return LineTable->getLineTableFilenameID(Name);
289 }
290 
291 
292 /// AddLineNote - Add a line note to the line table for the FileID and offset
293 /// specified by Loc.  If FilenameID is -1, it is considered to be
294 /// unspecified.
295 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
296                                 int FilenameID) {
297   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
298 
299   bool Invalid = false;
300   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
301   if (!Entry.isFile() || Invalid)
302     return;
303 
304   const SrcMgr::FileInfo &FileInfo = Entry.getFile();
305 
306   // Remember that this file has #line directives now if it doesn't already.
307   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
308 
309   if (LineTable == 0)
310     LineTable = new LineTableInfo();
311   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID);
312 }
313 
314 /// AddLineNote - Add a GNU line marker to the line table.
315 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
316                                 int FilenameID, bool IsFileEntry,
317                                 bool IsFileExit, bool IsSystemHeader,
318                                 bool IsExternCHeader) {
319   // If there is no filename and no flags, this is treated just like a #line,
320   // which does not change the flags of the previous line marker.
321   if (FilenameID == -1) {
322     assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader &&
323            "Can't set flags without setting the filename!");
324     return AddLineNote(Loc, LineNo, FilenameID);
325   }
326 
327   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
328 
329   bool Invalid = false;
330   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
331   if (!Entry.isFile() || Invalid)
332     return;
333 
334   const SrcMgr::FileInfo &FileInfo = Entry.getFile();
335 
336   // Remember that this file has #line directives now if it doesn't already.
337   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
338 
339   if (LineTable == 0)
340     LineTable = new LineTableInfo();
341 
342   SrcMgr::CharacteristicKind FileKind;
343   if (IsExternCHeader)
344     FileKind = SrcMgr::C_ExternCSystem;
345   else if (IsSystemHeader)
346     FileKind = SrcMgr::C_System;
347   else
348     FileKind = SrcMgr::C_User;
349 
350   unsigned EntryExit = 0;
351   if (IsFileEntry)
352     EntryExit = 1;
353   else if (IsFileExit)
354     EntryExit = 2;
355 
356   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID,
357                          EntryExit, FileKind);
358 }
359 
360 LineTableInfo &SourceManager::getLineTable() {
361   if (LineTable == 0)
362     LineTable = new LineTableInfo();
363   return *LineTable;
364 }
365 
366 //===----------------------------------------------------------------------===//
367 // Private 'Create' methods.
368 //===----------------------------------------------------------------------===//
369 
370 SourceManager::SourceManager(DiagnosticsEngine &Diag, FileManager &FileMgr)
371   : Diag(Diag), FileMgr(FileMgr), OverridenFilesKeepOriginalName(true),
372     ExternalSLocEntries(0), LineTable(0), NumLinearScans(0),
373     NumBinaryProbes(0), FakeBufferForRecovery(0),
374     FakeContentCacheForRecovery(0) {
375   clearIDTables();
376   Diag.setSourceManager(this);
377 }
378 
379 SourceManager::~SourceManager() {
380   delete LineTable;
381 
382   // Delete FileEntry objects corresponding to content caches.  Since the actual
383   // content cache objects are bump pointer allocated, we just have to run the
384   // dtors, but we call the deallocate method for completeness.
385   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) {
386     if (MemBufferInfos[i]) {
387       MemBufferInfos[i]->~ContentCache();
388       ContentCacheAlloc.Deallocate(MemBufferInfos[i]);
389     }
390   }
391   for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator
392        I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) {
393     if (I->second) {
394       I->second->~ContentCache();
395       ContentCacheAlloc.Deallocate(I->second);
396     }
397   }
398 
399   delete FakeBufferForRecovery;
400   delete FakeContentCacheForRecovery;
401 
402   for (llvm::DenseMap<FileID, MacroArgsMap *>::iterator
403          I = MacroArgsCacheMap.begin(),E = MacroArgsCacheMap.end(); I!=E; ++I) {
404     delete I->second;
405   }
406 }
407 
408 void SourceManager::clearIDTables() {
409   MainFileID = FileID();
410   PredefinesFileID = FileID();
411   LocalSLocEntryTable.clear();
412   LoadedSLocEntryTable.clear();
413   SLocEntryLoaded.clear();
414   LastLineNoFileIDQuery = FileID();
415   LastLineNoContentCache = 0;
416   LastFileIDLookup = FileID();
417 
418   if (LineTable)
419     LineTable->clear();
420 
421   // Use up FileID #0 as an invalid expansion.
422   NextLocalOffset = 0;
423   CurrentLoadedOffset = MaxLoadedOffset;
424   createExpansionLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1);
425 }
426 
427 /// getOrCreateContentCache - Create or return a cached ContentCache for the
428 /// specified file.
429 const ContentCache *
430 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt) {
431   assert(FileEnt && "Didn't specify a file entry to use?");
432 
433   // Do we already have information about this file?
434   ContentCache *&Entry = FileInfos[FileEnt];
435   if (Entry) return Entry;
436 
437   // Nope, create a new Cache entry.  Make sure it is at least 8-byte aligned
438   // so that FileInfo can use the low 3 bits of the pointer for its own
439   // nefarious purposes.
440   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
441   EntryAlign = std::max(8U, EntryAlign);
442   Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
443 
444   if (OverriddenFilesInfo) {
445     // If the file contents are overridden with contents from another file,
446     // pass that file to ContentCache.
447     llvm::DenseMap<const FileEntry *, const FileEntry *>::iterator
448         overI = OverriddenFilesInfo->OverriddenFiles.find(FileEnt);
449     if (overI == OverriddenFilesInfo->OverriddenFiles.end())
450       new (Entry) ContentCache(FileEnt);
451     else
452       new (Entry) ContentCache(OverridenFilesKeepOriginalName ? FileEnt
453                                                               : overI->second,
454                                overI->second);
455   } else {
456     new (Entry) ContentCache(FileEnt);
457   }
458 
459   return Entry;
460 }
461 
462 
463 /// createMemBufferContentCache - Create a new ContentCache for the specified
464 ///  memory buffer.  This does no caching.
465 const ContentCache*
466 SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) {
467   // Add a new ContentCache to the MemBufferInfos list and return it.  Make sure
468   // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of
469   // the pointer for its own nefarious purposes.
470   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
471   EntryAlign = std::max(8U, EntryAlign);
472   ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
473   new (Entry) ContentCache();
474   MemBufferInfos.push_back(Entry);
475   Entry->setBuffer(Buffer);
476   return Entry;
477 }
478 
479 const SrcMgr::SLocEntry &SourceManager::loadSLocEntry(unsigned Index,
480                                                       bool *Invalid) const {
481   assert(!SLocEntryLoaded[Index]);
482   if (ExternalSLocEntries->ReadSLocEntry(-(static_cast<int>(Index) + 2))) {
483     if (Invalid)
484       *Invalid = true;
485     // If the file of the SLocEntry changed we could still have loaded it.
486     if (!SLocEntryLoaded[Index]) {
487       // Try to recover; create a SLocEntry so the rest of clang can handle it.
488       LoadedSLocEntryTable[Index] = SLocEntry::get(0,
489                                  FileInfo::get(SourceLocation(),
490                                                getFakeContentCacheForRecovery(),
491                                                SrcMgr::C_User));
492     }
493   }
494 
495   return LoadedSLocEntryTable[Index];
496 }
497 
498 std::pair<int, unsigned>
499 SourceManager::AllocateLoadedSLocEntries(unsigned NumSLocEntries,
500                                          unsigned TotalSize) {
501   assert(ExternalSLocEntries && "Don't have an external sloc source");
502   LoadedSLocEntryTable.resize(LoadedSLocEntryTable.size() + NumSLocEntries);
503   SLocEntryLoaded.resize(LoadedSLocEntryTable.size());
504   CurrentLoadedOffset -= TotalSize;
505   assert(CurrentLoadedOffset >= NextLocalOffset && "Out of source locations");
506   int ID = LoadedSLocEntryTable.size();
507   return std::make_pair(-ID - 1, CurrentLoadedOffset);
508 }
509 
510 /// \brief As part of recovering from missing or changed content, produce a
511 /// fake, non-empty buffer.
512 const llvm::MemoryBuffer *SourceManager::getFakeBufferForRecovery() const {
513   if (!FakeBufferForRecovery)
514     FakeBufferForRecovery
515       = llvm::MemoryBuffer::getMemBuffer("<<<INVALID BUFFER>>");
516 
517   return FakeBufferForRecovery;
518 }
519 
520 /// \brief As part of recovering from missing or changed content, produce a
521 /// fake content cache.
522 const SrcMgr::ContentCache *
523 SourceManager::getFakeContentCacheForRecovery() const {
524   if (!FakeContentCacheForRecovery) {
525     FakeContentCacheForRecovery = new ContentCache();
526     FakeContentCacheForRecovery->replaceBuffer(getFakeBufferForRecovery(),
527                                                /*DoNotFree=*/true);
528   }
529   return FakeContentCacheForRecovery;
530 }
531 
532 //===----------------------------------------------------------------------===//
533 // Methods to create new FileID's and macro expansions.
534 //===----------------------------------------------------------------------===//
535 
536 /// createFileID - Create a new FileID for the specified ContentCache and
537 /// include position.  This works regardless of whether the ContentCache
538 /// corresponds to a file or some other input source.
539 FileID SourceManager::createFileID(const ContentCache *File,
540                                    SourceLocation IncludePos,
541                                    SrcMgr::CharacteristicKind FileCharacter,
542                                    int LoadedID, unsigned LoadedOffset) {
543   if (LoadedID < 0) {
544     assert(LoadedID != -1 && "Loading sentinel FileID");
545     unsigned Index = unsigned(-LoadedID) - 2;
546     assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
547     assert(!SLocEntryLoaded[Index] && "FileID already loaded");
548     LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset,
549         FileInfo::get(IncludePos, File, FileCharacter));
550     SLocEntryLoaded[Index] = true;
551     return FileID::get(LoadedID);
552   }
553   LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset,
554                                                FileInfo::get(IncludePos, File,
555                                                              FileCharacter)));
556   unsigned FileSize = File->getSize();
557   assert(NextLocalOffset + FileSize + 1 > NextLocalOffset &&
558          NextLocalOffset + FileSize + 1 <= CurrentLoadedOffset &&
559          "Ran out of source locations!");
560   // We do a +1 here because we want a SourceLocation that means "the end of the
561   // file", e.g. for the "no newline at the end of the file" diagnostic.
562   NextLocalOffset += FileSize + 1;
563 
564   // Set LastFileIDLookup to the newly created file.  The next getFileID call is
565   // almost guaranteed to be from that file.
566   FileID FID = FileID::get(LocalSLocEntryTable.size()-1);
567   return LastFileIDLookup = FID;
568 }
569 
570 SourceLocation
571 SourceManager::createMacroArgExpansionLoc(SourceLocation SpellingLoc,
572                                           SourceLocation ExpansionLoc,
573                                           unsigned TokLength) {
574   ExpansionInfo Info = ExpansionInfo::createForMacroArg(SpellingLoc,
575                                                         ExpansionLoc);
576   return createExpansionLocImpl(Info, TokLength);
577 }
578 
579 SourceLocation
580 SourceManager::createExpansionLoc(SourceLocation SpellingLoc,
581                                   SourceLocation ExpansionLocStart,
582                                   SourceLocation ExpansionLocEnd,
583                                   unsigned TokLength,
584                                   int LoadedID,
585                                   unsigned LoadedOffset) {
586   ExpansionInfo Info = ExpansionInfo::create(SpellingLoc, ExpansionLocStart,
587                                              ExpansionLocEnd);
588   return createExpansionLocImpl(Info, TokLength, LoadedID, LoadedOffset);
589 }
590 
591 SourceLocation
592 SourceManager::createExpansionLocImpl(const ExpansionInfo &Info,
593                                       unsigned TokLength,
594                                       int LoadedID,
595                                       unsigned LoadedOffset) {
596   if (LoadedID < 0) {
597     assert(LoadedID != -1 && "Loading sentinel FileID");
598     unsigned Index = unsigned(-LoadedID) - 2;
599     assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
600     assert(!SLocEntryLoaded[Index] && "FileID already loaded");
601     LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, Info);
602     SLocEntryLoaded[Index] = true;
603     return SourceLocation::getMacroLoc(LoadedOffset);
604   }
605   LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, Info));
606   assert(NextLocalOffset + TokLength + 1 > NextLocalOffset &&
607          NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset &&
608          "Ran out of source locations!");
609   // See createFileID for that +1.
610   NextLocalOffset += TokLength + 1;
611   return SourceLocation::getMacroLoc(NextLocalOffset - (TokLength + 1));
612 }
613 
614 const llvm::MemoryBuffer *
615 SourceManager::getMemoryBufferForFile(const FileEntry *File,
616                                       bool *Invalid) {
617   const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
618   assert(IR && "getOrCreateContentCache() cannot return NULL");
619   return IR->getBuffer(Diag, *this, SourceLocation(), Invalid);
620 }
621 
622 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
623                                          const llvm::MemoryBuffer *Buffer,
624                                          bool DoNotFree) {
625   const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile);
626   assert(IR && "getOrCreateContentCache() cannot return NULL");
627 
628   const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree);
629   const_cast<SrcMgr::ContentCache *>(IR)->BufferOverridden = true;
630 
631   getOverriddenFilesInfo().OverriddenFilesWithBuffer.insert(SourceFile);
632 }
633 
634 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
635                                          const FileEntry *NewFile) {
636   assert(SourceFile->getSize() == NewFile->getSize() &&
637          "Different sizes, use the FileManager to create a virtual file with "
638          "the correct size");
639   assert(FileInfos.count(SourceFile) == 0 &&
640          "This function should be called at the initialization stage, before "
641          "any parsing occurs.");
642   getOverriddenFilesInfo().OverriddenFiles[SourceFile] = NewFile;
643 }
644 
645 void SourceManager::disableFileContentsOverride(const FileEntry *File) {
646   if (!isFileOverridden(File))
647     return;
648 
649   const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
650   const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(0);
651   const_cast<SrcMgr::ContentCache *>(IR)->ContentsEntry = IR->OrigEntry;
652 
653   assert(OverriddenFilesInfo);
654   OverriddenFilesInfo->OverriddenFiles.erase(File);
655   OverriddenFilesInfo->OverriddenFilesWithBuffer.erase(File);
656 }
657 
658 StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const {
659   bool MyInvalid = false;
660   const SLocEntry &SLoc = getSLocEntry(FID, &MyInvalid);
661   if (!SLoc.isFile() || MyInvalid) {
662     if (Invalid)
663       *Invalid = true;
664     return "<<<<<INVALID SOURCE LOCATION>>>>>";
665   }
666 
667   const llvm::MemoryBuffer *Buf
668     = SLoc.getFile().getContentCache()->getBuffer(Diag, *this, SourceLocation(),
669                                                   &MyInvalid);
670   if (Invalid)
671     *Invalid = MyInvalid;
672 
673   if (MyInvalid)
674     return "<<<<<INVALID SOURCE LOCATION>>>>>";
675 
676   return Buf->getBuffer();
677 }
678 
679 //===----------------------------------------------------------------------===//
680 // SourceLocation manipulation methods.
681 //===----------------------------------------------------------------------===//
682 
683 /// \brief Return the FileID for a SourceLocation.
684 ///
685 /// This is the cache-miss path of getFileID. Not as hot as that function, but
686 /// still very important. It is responsible for finding the entry in the
687 /// SLocEntry tables that contains the specified location.
688 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const {
689   if (!SLocOffset)
690     return FileID::get(0);
691 
692   // Now it is time to search for the correct file. See where the SLocOffset
693   // sits in the global view and consult local or loaded buffers for it.
694   if (SLocOffset < NextLocalOffset)
695     return getFileIDLocal(SLocOffset);
696   return getFileIDLoaded(SLocOffset);
697 }
698 
699 /// \brief Return the FileID for a SourceLocation with a low offset.
700 ///
701 /// This function knows that the SourceLocation is in a local buffer, not a
702 /// loaded one.
703 FileID SourceManager::getFileIDLocal(unsigned SLocOffset) const {
704   assert(SLocOffset < NextLocalOffset && "Bad function choice");
705 
706   // After the first and second level caches, I see two common sorts of
707   // behavior: 1) a lot of searched FileID's are "near" the cached file
708   // location or are "near" the cached expansion location. 2) others are just
709   // completely random and may be a very long way away.
710   //
711   // To handle this, we do a linear search for up to 8 steps to catch #1 quickly
712   // then we fall back to a less cache efficient, but more scalable, binary
713   // search to find the location.
714 
715   // See if this is near the file point - worst case we start scanning from the
716   // most newly created FileID.
717   std::vector<SrcMgr::SLocEntry>::const_iterator I;
718 
719   if (LastFileIDLookup.ID < 0 ||
720       LocalSLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) {
721     // Neither loc prunes our search.
722     I = LocalSLocEntryTable.end();
723   } else {
724     // Perhaps it is near the file point.
725     I = LocalSLocEntryTable.begin()+LastFileIDLookup.ID;
726   }
727 
728   // Find the FileID that contains this.  "I" is an iterator that points to a
729   // FileID whose offset is known to be larger than SLocOffset.
730   unsigned NumProbes = 0;
731   while (1) {
732     --I;
733     if (I->getOffset() <= SLocOffset) {
734       FileID Res = FileID::get(int(I - LocalSLocEntryTable.begin()));
735 
736       // If this isn't an expansion, remember it.  We have good locality across
737       // FileID lookups.
738       if (!I->isExpansion())
739         LastFileIDLookup = Res;
740       NumLinearScans += NumProbes+1;
741       return Res;
742     }
743     if (++NumProbes == 8)
744       break;
745   }
746 
747   // Convert "I" back into an index.  We know that it is an entry whose index is
748   // larger than the offset we are looking for.
749   unsigned GreaterIndex = I - LocalSLocEntryTable.begin();
750   // LessIndex - This is the lower bound of the range that we're searching.
751   // We know that the offset corresponding to the FileID is is less than
752   // SLocOffset.
753   unsigned LessIndex = 0;
754   NumProbes = 0;
755   while (1) {
756     bool Invalid = false;
757     unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex;
758     unsigned MidOffset = getLocalSLocEntry(MiddleIndex, &Invalid).getOffset();
759     if (Invalid)
760       return FileID::get(0);
761 
762     ++NumProbes;
763 
764     // If the offset of the midpoint is too large, chop the high side of the
765     // range to the midpoint.
766     if (MidOffset > SLocOffset) {
767       GreaterIndex = MiddleIndex;
768       continue;
769     }
770 
771     // If the middle index contains the value, succeed and return.
772     // FIXME: This could be made faster by using a function that's aware of
773     // being in the local area.
774     if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) {
775       FileID Res = FileID::get(MiddleIndex);
776 
777       // If this isn't a macro expansion, remember it.  We have good locality
778       // across FileID lookups.
779       if (!LocalSLocEntryTable[MiddleIndex].isExpansion())
780         LastFileIDLookup = Res;
781       NumBinaryProbes += NumProbes;
782       return Res;
783     }
784 
785     // Otherwise, move the low-side up to the middle index.
786     LessIndex = MiddleIndex;
787   }
788 }
789 
790 /// \brief Return the FileID for a SourceLocation with a high offset.
791 ///
792 /// This function knows that the SourceLocation is in a loaded buffer, not a
793 /// local one.
794 FileID SourceManager::getFileIDLoaded(unsigned SLocOffset) const {
795   // Sanity checking, otherwise a bug may lead to hanging in release build.
796   if (SLocOffset < CurrentLoadedOffset) {
797     assert(0 && "Invalid SLocOffset or bad function choice");
798     return FileID();
799   }
800 
801   // Essentially the same as the local case, but the loaded array is sorted
802   // in the other direction.
803 
804   // First do a linear scan from the last lookup position, if possible.
805   unsigned I;
806   int LastID = LastFileIDLookup.ID;
807   if (LastID >= 0 || getLoadedSLocEntryByID(LastID).getOffset() < SLocOffset)
808     I = 0;
809   else
810     I = (-LastID - 2) + 1;
811 
812   unsigned NumProbes;
813   for (NumProbes = 0; NumProbes < 8; ++NumProbes, ++I) {
814     // Make sure the entry is loaded!
815     const SrcMgr::SLocEntry &E = getLoadedSLocEntry(I);
816     if (E.getOffset() <= SLocOffset) {
817       FileID Res = FileID::get(-int(I) - 2);
818 
819       if (!E.isExpansion())
820         LastFileIDLookup = Res;
821       NumLinearScans += NumProbes + 1;
822       return Res;
823     }
824   }
825 
826   // Linear scan failed. Do the binary search. Note the reverse sorting of the
827   // table: GreaterIndex is the one where the offset is greater, which is
828   // actually a lower index!
829   unsigned GreaterIndex = I;
830   unsigned LessIndex = LoadedSLocEntryTable.size();
831   NumProbes = 0;
832   while (1) {
833     ++NumProbes;
834     unsigned MiddleIndex = (LessIndex - GreaterIndex) / 2 + GreaterIndex;
835     const SrcMgr::SLocEntry &E = getLoadedSLocEntry(MiddleIndex);
836 
837     ++NumProbes;
838 
839     if (E.getOffset() > SLocOffset) {
840       GreaterIndex = MiddleIndex;
841       continue;
842     }
843 
844     if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) {
845       FileID Res = FileID::get(-int(MiddleIndex) - 2);
846       if (!E.isExpansion())
847         LastFileIDLookup = Res;
848       NumBinaryProbes += NumProbes;
849       return Res;
850     }
851 
852     LessIndex = MiddleIndex;
853   }
854 }
855 
856 SourceLocation SourceManager::
857 getExpansionLocSlowCase(SourceLocation Loc) const {
858   do {
859     // Note: If Loc indicates an offset into a token that came from a macro
860     // expansion (e.g. the 5th character of the token) we do not want to add
861     // this offset when going to the expansion location.  The expansion
862     // location is the macro invocation, which the offset has nothing to do
863     // with.  This is unlike when we get the spelling loc, because the offset
864     // directly correspond to the token whose spelling we're inspecting.
865     Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart();
866   } while (!Loc.isFileID());
867 
868   return Loc;
869 }
870 
871 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
872   do {
873     std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
874     Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
875     Loc = Loc.getLocWithOffset(LocInfo.second);
876   } while (!Loc.isFileID());
877   return Loc;
878 }
879 
880 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const {
881   do {
882     if (isMacroArgExpansion(Loc))
883       Loc = getImmediateSpellingLoc(Loc);
884     else
885       Loc = getImmediateExpansionRange(Loc).first;
886   } while (!Loc.isFileID());
887   return Loc;
888 }
889 
890 
891 std::pair<FileID, unsigned>
892 SourceManager::getDecomposedExpansionLocSlowCase(
893                                              const SrcMgr::SLocEntry *E) const {
894   // If this is an expansion record, walk through all the expansion points.
895   FileID FID;
896   SourceLocation Loc;
897   unsigned Offset;
898   do {
899     Loc = E->getExpansion().getExpansionLocStart();
900 
901     FID = getFileID(Loc);
902     E = &getSLocEntry(FID);
903     Offset = Loc.getOffset()-E->getOffset();
904   } while (!Loc.isFileID());
905 
906   return std::make_pair(FID, Offset);
907 }
908 
909 std::pair<FileID, unsigned>
910 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
911                                                 unsigned Offset) const {
912   // If this is an expansion record, walk through all the expansion points.
913   FileID FID;
914   SourceLocation Loc;
915   do {
916     Loc = E->getExpansion().getSpellingLoc();
917     Loc = Loc.getLocWithOffset(Offset);
918 
919     FID = getFileID(Loc);
920     E = &getSLocEntry(FID);
921     Offset = Loc.getOffset()-E->getOffset();
922   } while (!Loc.isFileID());
923 
924   return std::make_pair(FID, Offset);
925 }
926 
927 /// getImmediateSpellingLoc - Given a SourceLocation object, return the
928 /// spelling location referenced by the ID.  This is the first level down
929 /// towards the place where the characters that make up the lexed token can be
930 /// found.  This should not generally be used by clients.
931 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
932   if (Loc.isFileID()) return Loc;
933   std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
934   Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
935   return Loc.getLocWithOffset(LocInfo.second);
936 }
937 
938 
939 /// getImmediateExpansionRange - Loc is required to be an expansion location.
940 /// Return the start/end of the expansion information.
941 std::pair<SourceLocation,SourceLocation>
942 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const {
943   assert(Loc.isMacroID() && "Not a macro expansion loc!");
944   const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion();
945   return Expansion.getExpansionLocRange();
946 }
947 
948 /// getExpansionRange - Given a SourceLocation object, return the range of
949 /// tokens covered by the expansion in the ultimate file.
950 std::pair<SourceLocation,SourceLocation>
951 SourceManager::getExpansionRange(SourceLocation Loc) const {
952   if (Loc.isFileID()) return std::make_pair(Loc, Loc);
953 
954   std::pair<SourceLocation,SourceLocation> Res =
955     getImmediateExpansionRange(Loc);
956 
957   // Fully resolve the start and end locations to their ultimate expansion
958   // points.
959   while (!Res.first.isFileID())
960     Res.first = getImmediateExpansionRange(Res.first).first;
961   while (!Res.second.isFileID())
962     Res.second = getImmediateExpansionRange(Res.second).second;
963   return Res;
964 }
965 
966 bool SourceManager::isMacroArgExpansion(SourceLocation Loc) const {
967   if (!Loc.isMacroID()) return false;
968 
969   FileID FID = getFileID(Loc);
970   const SrcMgr::SLocEntry *E = &getSLocEntry(FID);
971   const SrcMgr::ExpansionInfo &Expansion = E->getExpansion();
972   return Expansion.isMacroArgExpansion();
973 }
974 
975 
976 //===----------------------------------------------------------------------===//
977 // Queries about the code at a SourceLocation.
978 //===----------------------------------------------------------------------===//
979 
980 /// getCharacterData - Return a pointer to the start of the specified location
981 /// in the appropriate MemoryBuffer.
982 const char *SourceManager::getCharacterData(SourceLocation SL,
983                                             bool *Invalid) const {
984   // Note that this is a hot function in the getSpelling() path, which is
985   // heavily used by -E mode.
986   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
987 
988   // Note that calling 'getBuffer()' may lazily page in a source file.
989   bool CharDataInvalid = false;
990   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid);
991   if (CharDataInvalid || !Entry.isFile()) {
992     if (Invalid)
993       *Invalid = true;
994 
995     return "<<<<INVALID BUFFER>>>>";
996   }
997   const llvm::MemoryBuffer *Buffer
998     = Entry.getFile().getContentCache()
999                   ->getBuffer(Diag, *this, SourceLocation(), &CharDataInvalid);
1000   if (Invalid)
1001     *Invalid = CharDataInvalid;
1002   return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
1003 }
1004 
1005 
1006 /// getColumnNumber - Return the column # for the specified file position.
1007 /// this is significantly cheaper to compute than the line number.
1008 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
1009                                         bool *Invalid) const {
1010   bool MyInvalid = false;
1011   const llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid);
1012   if (Invalid)
1013     *Invalid = MyInvalid;
1014 
1015   if (MyInvalid)
1016     return 1;
1017 
1018   if (FilePos >= MemBuf->getBufferSize()) {
1019     if (Invalid)
1020       *Invalid = MyInvalid;
1021     return 1;
1022   }
1023 
1024   const char *Buf = MemBuf->getBufferStart();
1025   unsigned LineStart = FilePos;
1026   while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
1027     --LineStart;
1028   return FilePos-LineStart+1;
1029 }
1030 
1031 // isInvalid - Return the result of calling loc.isInvalid(), and
1032 // if Invalid is not null, set its value to same.
1033 static bool isInvalid(SourceLocation Loc, bool *Invalid) {
1034   bool MyInvalid = Loc.isInvalid();
1035   if (Invalid)
1036     *Invalid = MyInvalid;
1037   return MyInvalid;
1038 }
1039 
1040 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
1041                                                 bool *Invalid) const {
1042   if (isInvalid(Loc, Invalid)) return 0;
1043   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1044   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1045 }
1046 
1047 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc,
1048                                                  bool *Invalid) const {
1049   if (isInvalid(Loc, Invalid)) return 0;
1050   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1051   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1052 }
1053 
1054 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc,
1055                                                 bool *Invalid) const {
1056   if (isInvalid(Loc, Invalid)) return 0;
1057   return getPresumedLoc(Loc).getColumn();
1058 }
1059 
1060 #ifdef __SSE2__
1061 #include <emmintrin.h>
1062 #endif
1063 
1064 static LLVM_ATTRIBUTE_NOINLINE void
1065 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1066                    llvm::BumpPtrAllocator &Alloc,
1067                    const SourceManager &SM, bool &Invalid);
1068 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1069                                llvm::BumpPtrAllocator &Alloc,
1070                                const SourceManager &SM, bool &Invalid) {
1071   // Note that calling 'getBuffer()' may lazily page in the file.
1072   const MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(),
1073                                              &Invalid);
1074   if (Invalid)
1075     return;
1076 
1077   // Find the file offsets of all of the *physical* source lines.  This does
1078   // not look at trigraphs, escaped newlines, or anything else tricky.
1079   SmallVector<unsigned, 256> LineOffsets;
1080 
1081   // Line #1 starts at char 0.
1082   LineOffsets.push_back(0);
1083 
1084   const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
1085   const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
1086   unsigned Offs = 0;
1087   while (1) {
1088     // Skip over the contents of the line.
1089     const unsigned char *NextBuf = (const unsigned char *)Buf;
1090 
1091 #ifdef __SSE2__
1092     // Try to skip to the next newline using SSE instructions. This is very
1093     // performance sensitive for programs with lots of diagnostics and in -E
1094     // mode.
1095     __m128i CRs = _mm_set1_epi8('\r');
1096     __m128i LFs = _mm_set1_epi8('\n');
1097 
1098     // First fix up the alignment to 16 bytes.
1099     while (((uintptr_t)NextBuf & 0xF) != 0) {
1100       if (*NextBuf == '\n' || *NextBuf == '\r' || *NextBuf == '\0')
1101         goto FoundSpecialChar;
1102       ++NextBuf;
1103     }
1104 
1105     // Scan 16 byte chunks for '\r' and '\n'. Ignore '\0'.
1106     while (NextBuf+16 <= End) {
1107       __m128i Chunk = *(__m128i*)NextBuf;
1108       __m128i Cmp = _mm_or_si128(_mm_cmpeq_epi8(Chunk, CRs),
1109                                  _mm_cmpeq_epi8(Chunk, LFs));
1110       unsigned Mask = _mm_movemask_epi8(Cmp);
1111 
1112       // If we found a newline, adjust the pointer and jump to the handling code.
1113       if (Mask != 0) {
1114         NextBuf += llvm::CountTrailingZeros_32(Mask);
1115         goto FoundSpecialChar;
1116       }
1117       NextBuf += 16;
1118     }
1119 #endif
1120 
1121     while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
1122       ++NextBuf;
1123 
1124 #ifdef __SSE2__
1125 FoundSpecialChar:
1126 #endif
1127     Offs += NextBuf-Buf;
1128     Buf = NextBuf;
1129 
1130     if (Buf[0] == '\n' || Buf[0] == '\r') {
1131       // If this is \n\r or \r\n, skip both characters.
1132       if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1])
1133         ++Offs, ++Buf;
1134       ++Offs, ++Buf;
1135       LineOffsets.push_back(Offs);
1136     } else {
1137       // Otherwise, this is a null.  If end of file, exit.
1138       if (Buf == End) break;
1139       // Otherwise, skip the null.
1140       ++Offs, ++Buf;
1141     }
1142   }
1143 
1144   // Copy the offsets into the FileInfo structure.
1145   FI->NumLines = LineOffsets.size();
1146   FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
1147   std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
1148 }
1149 
1150 /// getLineNumber - Given a SourceLocation, return the spelling line number
1151 /// for the position indicated.  This requires building and caching a table of
1152 /// line offsets for the MemoryBuffer, so this is not cheap: use only when
1153 /// about to emit a diagnostic.
1154 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
1155                                       bool *Invalid) const {
1156   if (FID.isInvalid()) {
1157     if (Invalid)
1158       *Invalid = true;
1159     return 1;
1160   }
1161 
1162   ContentCache *Content;
1163   if (LastLineNoFileIDQuery == FID)
1164     Content = LastLineNoContentCache;
1165   else {
1166     bool MyInvalid = false;
1167     const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid);
1168     if (MyInvalid || !Entry.isFile()) {
1169       if (Invalid)
1170         *Invalid = true;
1171       return 1;
1172     }
1173 
1174     Content = const_cast<ContentCache*>(Entry.getFile().getContentCache());
1175   }
1176 
1177   // If this is the first use of line information for this buffer, compute the
1178   /// SourceLineCache for it on demand.
1179   if (Content->SourceLineCache == 0) {
1180     bool MyInvalid = false;
1181     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1182     if (Invalid)
1183       *Invalid = MyInvalid;
1184     if (MyInvalid)
1185       return 1;
1186   } else if (Invalid)
1187     *Invalid = false;
1188 
1189   // Okay, we know we have a line number table.  Do a binary search to find the
1190   // line number that this character position lands on.
1191   unsigned *SourceLineCache = Content->SourceLineCache;
1192   unsigned *SourceLineCacheStart = SourceLineCache;
1193   unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
1194 
1195   unsigned QueriedFilePos = FilePos+1;
1196 
1197   // FIXME: I would like to be convinced that this code is worth being as
1198   // complicated as it is, binary search isn't that slow.
1199   //
1200   // If it is worth being optimized, then in my opinion it could be more
1201   // performant, simpler, and more obviously correct by just "galloping" outward
1202   // from the queried file position. In fact, this could be incorporated into a
1203   // generic algorithm such as lower_bound_with_hint.
1204   //
1205   // If someone gives me a test case where this matters, and I will do it! - DWD
1206 
1207   // If the previous query was to the same file, we know both the file pos from
1208   // that query and the line number returned.  This allows us to narrow the
1209   // search space from the entire file to something near the match.
1210   if (LastLineNoFileIDQuery == FID) {
1211     if (QueriedFilePos >= LastLineNoFilePos) {
1212       // FIXME: Potential overflow?
1213       SourceLineCache = SourceLineCache+LastLineNoResult-1;
1214 
1215       // The query is likely to be nearby the previous one.  Here we check to
1216       // see if it is within 5, 10 or 20 lines.  It can be far away in cases
1217       // where big comment blocks and vertical whitespace eat up lines but
1218       // contribute no tokens.
1219       if (SourceLineCache+5 < SourceLineCacheEnd) {
1220         if (SourceLineCache[5] > QueriedFilePos)
1221           SourceLineCacheEnd = SourceLineCache+5;
1222         else if (SourceLineCache+10 < SourceLineCacheEnd) {
1223           if (SourceLineCache[10] > QueriedFilePos)
1224             SourceLineCacheEnd = SourceLineCache+10;
1225           else if (SourceLineCache+20 < SourceLineCacheEnd) {
1226             if (SourceLineCache[20] > QueriedFilePos)
1227               SourceLineCacheEnd = SourceLineCache+20;
1228           }
1229         }
1230       }
1231     } else {
1232       if (LastLineNoResult < Content->NumLines)
1233         SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
1234     }
1235   }
1236 
1237   // If the spread is large, do a "radix" test as our initial guess, based on
1238   // the assumption that lines average to approximately the same length.
1239   // NOTE: This is currently disabled, as it does not appear to be profitable in
1240   // initial measurements.
1241   if (0 && SourceLineCacheEnd-SourceLineCache > 20) {
1242     unsigned FileLen = Content->SourceLineCache[Content->NumLines-1];
1243 
1244     // Take a stab at guessing where it is.
1245     unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen;
1246 
1247     // Check for -10 and +10 lines.
1248     unsigned LowerBound = std::max(int(ApproxPos-10), 0);
1249     unsigned UpperBound = std::min(ApproxPos+10, FileLen);
1250 
1251     // If the computed lower bound is less than the query location, move it in.
1252     if (SourceLineCache < SourceLineCacheStart+LowerBound &&
1253         SourceLineCacheStart[LowerBound] < QueriedFilePos)
1254       SourceLineCache = SourceLineCacheStart+LowerBound;
1255 
1256     // If the computed upper bound is greater than the query location, move it.
1257     if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound &&
1258         SourceLineCacheStart[UpperBound] >= QueriedFilePos)
1259       SourceLineCacheEnd = SourceLineCacheStart+UpperBound;
1260   }
1261 
1262   unsigned *Pos
1263     = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
1264   unsigned LineNo = Pos-SourceLineCacheStart;
1265 
1266   LastLineNoFileIDQuery = FID;
1267   LastLineNoContentCache = Content;
1268   LastLineNoFilePos = QueriedFilePos;
1269   LastLineNoResult = LineNo;
1270   return LineNo;
1271 }
1272 
1273 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
1274                                               bool *Invalid) const {
1275   if (isInvalid(Loc, Invalid)) return 0;
1276   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1277   return getLineNumber(LocInfo.first, LocInfo.second);
1278 }
1279 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc,
1280                                                bool *Invalid) const {
1281   if (isInvalid(Loc, Invalid)) return 0;
1282   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1283   return getLineNumber(LocInfo.first, LocInfo.second);
1284 }
1285 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc,
1286                                               bool *Invalid) const {
1287   if (isInvalid(Loc, Invalid)) return 0;
1288   return getPresumedLoc(Loc).getLine();
1289 }
1290 
1291 /// getFileCharacteristic - return the file characteristic of the specified
1292 /// source location, indicating whether this is a normal file, a system
1293 /// header, or an "implicit extern C" system header.
1294 ///
1295 /// This state can be modified with flags on GNU linemarker directives like:
1296 ///   # 4 "foo.h" 3
1297 /// which changes all source locations in the current file after that to be
1298 /// considered to be from a system header.
1299 SrcMgr::CharacteristicKind
1300 SourceManager::getFileCharacteristic(SourceLocation Loc) const {
1301   assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!");
1302   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1303   bool Invalid = false;
1304   const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid);
1305   if (Invalid || !SEntry.isFile())
1306     return C_User;
1307 
1308   const SrcMgr::FileInfo &FI = SEntry.getFile();
1309 
1310   // If there are no #line directives in this file, just return the whole-file
1311   // state.
1312   if (!FI.hasLineDirectives())
1313     return FI.getFileCharacteristic();
1314 
1315   assert(LineTable && "Can't have linetable entries without a LineTable!");
1316   // See if there is a #line directive before the location.
1317   const LineEntry *Entry =
1318     LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second);
1319 
1320   // If this is before the first line marker, use the file characteristic.
1321   if (!Entry)
1322     return FI.getFileCharacteristic();
1323 
1324   return Entry->FileKind;
1325 }
1326 
1327 /// Return the filename or buffer identifier of the buffer the location is in.
1328 /// Note that this name does not respect #line directives.  Use getPresumedLoc
1329 /// for normal clients.
1330 const char *SourceManager::getBufferName(SourceLocation Loc,
1331                                          bool *Invalid) const {
1332   if (isInvalid(Loc, Invalid)) return "<invalid loc>";
1333 
1334   return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
1335 }
1336 
1337 
1338 /// getPresumedLoc - This method returns the "presumed" location of a
1339 /// SourceLocation specifies.  A "presumed location" can be modified by #line
1340 /// or GNU line marker directives.  This provides a view on the data that a
1341 /// user should see in diagnostics, for example.
1342 ///
1343 /// Note that a presumed location is always given as the expansion point of an
1344 /// expansion location, not at the spelling location.
1345 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc) const {
1346   if (Loc.isInvalid()) return PresumedLoc();
1347 
1348   // Presumed locations are always for expansion points.
1349   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1350 
1351   bool Invalid = false;
1352   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
1353   if (Invalid || !Entry.isFile())
1354     return PresumedLoc();
1355 
1356   const SrcMgr::FileInfo &FI = Entry.getFile();
1357   const SrcMgr::ContentCache *C = FI.getContentCache();
1358 
1359   // To get the source name, first consult the FileEntry (if one exists)
1360   // before the MemBuffer as this will avoid unnecessarily paging in the
1361   // MemBuffer.
1362   const char *Filename;
1363   if (C->OrigEntry)
1364     Filename = C->OrigEntry->getName();
1365   else
1366     Filename = C->getBuffer(Diag, *this)->getBufferIdentifier();
1367 
1368   unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid);
1369   if (Invalid)
1370     return PresumedLoc();
1371   unsigned ColNo  = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid);
1372   if (Invalid)
1373     return PresumedLoc();
1374 
1375   SourceLocation IncludeLoc = FI.getIncludeLoc();
1376 
1377   // If we have #line directives in this file, update and overwrite the physical
1378   // location info if appropriate.
1379   if (FI.hasLineDirectives()) {
1380     assert(LineTable && "Can't have linetable entries without a LineTable!");
1381     // See if there is a #line directive before this.  If so, get it.
1382     if (const LineEntry *Entry =
1383           LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second)) {
1384       // If the LineEntry indicates a filename, use it.
1385       if (Entry->FilenameID != -1)
1386         Filename = LineTable->getFilename(Entry->FilenameID);
1387 
1388       // Use the line number specified by the LineEntry.  This line number may
1389       // be multiple lines down from the line entry.  Add the difference in
1390       // physical line numbers from the query point and the line marker to the
1391       // total.
1392       unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
1393       LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
1394 
1395       // Note that column numbers are not molested by line markers.
1396 
1397       // Handle virtual #include manipulation.
1398       if (Entry->IncludeOffset) {
1399         IncludeLoc = getLocForStartOfFile(LocInfo.first);
1400         IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset);
1401       }
1402     }
1403   }
1404 
1405   return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc);
1406 }
1407 
1408 /// \brief The size of the SLocEnty that \arg FID represents.
1409 unsigned SourceManager::getFileIDSize(FileID FID) const {
1410   bool Invalid = false;
1411   const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1412   if (Invalid)
1413     return 0;
1414 
1415   int ID = FID.ID;
1416   unsigned NextOffset;
1417   if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size()))
1418     NextOffset = getNextLocalOffset();
1419   else if (ID+1 == -1)
1420     NextOffset = MaxLoadedOffset;
1421   else
1422     NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset();
1423 
1424   return NextOffset - Entry.getOffset() - 1;
1425 }
1426 
1427 //===----------------------------------------------------------------------===//
1428 // Other miscellaneous methods.
1429 //===----------------------------------------------------------------------===//
1430 
1431 /// \brief Retrieve the inode for the given file entry, if possible.
1432 ///
1433 /// This routine involves a system call, and therefore should only be used
1434 /// in non-performance-critical code.
1435 static llvm::Optional<ino_t> getActualFileInode(const FileEntry *File) {
1436   if (!File)
1437     return llvm::Optional<ino_t>();
1438 
1439   struct stat StatBuf;
1440   if (::stat(File->getName(), &StatBuf))
1441     return llvm::Optional<ino_t>();
1442 
1443   return StatBuf.st_ino;
1444 }
1445 
1446 /// \brief Get the source location for the given file:line:col triplet.
1447 ///
1448 /// If the source file is included multiple times, the source location will
1449 /// be based upon an arbitrary inclusion.
1450 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile,
1451                                                   unsigned Line,
1452                                                   unsigned Col) const {
1453   assert(SourceFile && "Null source file!");
1454   assert(Line && Col && "Line and column should start from 1!");
1455 
1456   FileID FirstFID = translateFile(SourceFile);
1457   return translateLineCol(FirstFID, Line, Col);
1458 }
1459 
1460 /// \brief Get the FileID for the given file.
1461 ///
1462 /// If the source file is included multiple times, the FileID will be the
1463 /// first inclusion.
1464 FileID SourceManager::translateFile(const FileEntry *SourceFile) const {
1465   assert(SourceFile && "Null source file!");
1466 
1467   // Find the first file ID that corresponds to the given file.
1468   FileID FirstFID;
1469 
1470   // First, check the main file ID, since it is common to look for a
1471   // location in the main file.
1472   llvm::Optional<ino_t> SourceFileInode;
1473   llvm::Optional<StringRef> SourceFileName;
1474   if (!MainFileID.isInvalid()) {
1475     bool Invalid = false;
1476     const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid);
1477     if (Invalid)
1478       return FileID();
1479 
1480     if (MainSLoc.isFile()) {
1481       const ContentCache *MainContentCache
1482         = MainSLoc.getFile().getContentCache();
1483       if (!MainContentCache) {
1484         // Can't do anything
1485       } else if (MainContentCache->OrigEntry == SourceFile) {
1486         FirstFID = MainFileID;
1487       } else {
1488         // Fall back: check whether we have the same base name and inode
1489         // as the main file.
1490         const FileEntry *MainFile = MainContentCache->OrigEntry;
1491         SourceFileName = llvm::sys::path::filename(SourceFile->getName());
1492         if (*SourceFileName == llvm::sys::path::filename(MainFile->getName())) {
1493           SourceFileInode = getActualFileInode(SourceFile);
1494           if (SourceFileInode) {
1495             if (llvm::Optional<ino_t> MainFileInode
1496                                                = getActualFileInode(MainFile)) {
1497               if (*SourceFileInode == *MainFileInode) {
1498                 FirstFID = MainFileID;
1499                 SourceFile = MainFile;
1500               }
1501             }
1502           }
1503         }
1504       }
1505     }
1506   }
1507 
1508   if (FirstFID.isInvalid()) {
1509     // The location we're looking for isn't in the main file; look
1510     // through all of the local source locations.
1511     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1512       bool Invalid = false;
1513       const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid);
1514       if (Invalid)
1515         return FileID();
1516 
1517       if (SLoc.isFile() &&
1518           SLoc.getFile().getContentCache() &&
1519           SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1520         FirstFID = FileID::get(I);
1521         break;
1522       }
1523     }
1524     // If that still didn't help, try the modules.
1525     if (FirstFID.isInvalid()) {
1526       for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) {
1527         const SLocEntry &SLoc = getLoadedSLocEntry(I);
1528         if (SLoc.isFile() &&
1529             SLoc.getFile().getContentCache() &&
1530             SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1531           FirstFID = FileID::get(-int(I) - 2);
1532           break;
1533         }
1534       }
1535     }
1536   }
1537 
1538   // If we haven't found what we want yet, try again, but this time stat()
1539   // each of the files in case the files have changed since we originally
1540   // parsed the file.
1541   if (FirstFID.isInvalid() &&
1542       (SourceFileName ||
1543        (SourceFileName = llvm::sys::path::filename(SourceFile->getName()))) &&
1544       (SourceFileInode ||
1545        (SourceFileInode = getActualFileInode(SourceFile)))) {
1546     bool Invalid = false;
1547     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1548       FileID IFileID;
1549       IFileID.ID = I;
1550       const SLocEntry &SLoc = getSLocEntry(IFileID, &Invalid);
1551       if (Invalid)
1552         return FileID();
1553 
1554       if (SLoc.isFile()) {
1555         const ContentCache *FileContentCache
1556           = SLoc.getFile().getContentCache();
1557       const FileEntry *Entry =FileContentCache? FileContentCache->OrigEntry : 0;
1558         if (Entry &&
1559             *SourceFileName == llvm::sys::path::filename(Entry->getName())) {
1560           if (llvm::Optional<ino_t> EntryInode = getActualFileInode(Entry)) {
1561             if (*SourceFileInode == *EntryInode) {
1562               FirstFID = FileID::get(I);
1563               SourceFile = Entry;
1564               break;
1565             }
1566           }
1567         }
1568       }
1569     }
1570   }
1571 
1572   return FirstFID;
1573 }
1574 
1575 /// \brief Get the source location in \arg FID for the given line:col.
1576 /// Returns null location if \arg FID is not a file SLocEntry.
1577 SourceLocation SourceManager::translateLineCol(FileID FID,
1578                                                unsigned Line,
1579                                                unsigned Col) const {
1580   if (FID.isInvalid())
1581     return SourceLocation();
1582 
1583   bool Invalid = false;
1584   const SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1585   if (Invalid)
1586     return SourceLocation();
1587 
1588   if (!Entry.isFile())
1589     return SourceLocation();
1590 
1591   SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset());
1592 
1593   if (Line == 1 && Col == 1)
1594     return FileLoc;
1595 
1596   ContentCache *Content
1597     = const_cast<ContentCache *>(Entry.getFile().getContentCache());
1598   if (!Content)
1599     return SourceLocation();
1600 
1601   // If this is the first use of line information for this buffer, compute the
1602   // SourceLineCache for it on demand.
1603   if (Content->SourceLineCache == 0) {
1604     bool MyInvalid = false;
1605     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1606     if (MyInvalid)
1607       return SourceLocation();
1608   }
1609 
1610   if (Line > Content->NumLines) {
1611     unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize();
1612     if (Size > 0)
1613       --Size;
1614     return FileLoc.getLocWithOffset(Size);
1615   }
1616 
1617   const llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, *this);
1618   unsigned FilePos = Content->SourceLineCache[Line - 1];
1619   const char *Buf = Buffer->getBufferStart() + FilePos;
1620   unsigned BufLength = Buffer->getBufferSize() - FilePos;
1621   if (BufLength == 0)
1622     return FileLoc.getLocWithOffset(FilePos);
1623 
1624   unsigned i = 0;
1625 
1626   // Check that the given column is valid.
1627   while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
1628     ++i;
1629   if (i < Col-1)
1630     return FileLoc.getLocWithOffset(FilePos + i);
1631 
1632   return FileLoc.getLocWithOffset(FilePos + Col - 1);
1633 }
1634 
1635 /// \brief Compute a map of macro argument chunks to their expanded source
1636 /// location. Chunks that are not part of a macro argument will map to an
1637 /// invalid source location. e.g. if a file contains one macro argument at
1638 /// offset 100 with length 10, this is how the map will be formed:
1639 ///     0   -> SourceLocation()
1640 ///     100 -> Expanded macro arg location
1641 ///     110 -> SourceLocation()
1642 void SourceManager::computeMacroArgsCache(MacroArgsMap *&CachePtr,
1643                                           FileID FID) const {
1644   assert(!FID.isInvalid());
1645   assert(!CachePtr);
1646 
1647   CachePtr = new MacroArgsMap();
1648   MacroArgsMap &MacroArgsCache = *CachePtr;
1649   // Initially no macro argument chunk is present.
1650   MacroArgsCache.insert(std::make_pair(0, SourceLocation()));
1651 
1652   int ID = FID.ID;
1653   while (1) {
1654     ++ID;
1655     // Stop if there are no more FileIDs to check.
1656     if (ID > 0) {
1657       if (unsigned(ID) >= local_sloc_entry_size())
1658         return;
1659     } else if (ID == -1) {
1660       return;
1661     }
1662 
1663     const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID);
1664     if (Entry.isFile()) {
1665       SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc();
1666       if (IncludeLoc.isInvalid())
1667         continue;
1668       if (!isInFileID(IncludeLoc, FID))
1669         return; // No more files/macros that may be "contained" in this file.
1670 
1671       // Skip the files/macros of the #include'd file, we only care about macros
1672       // that lexed macro arguments from our file.
1673       if (Entry.getFile().NumCreatedFIDs)
1674         ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/;
1675       continue;
1676     }
1677 
1678     const ExpansionInfo &ExpInfo = Entry.getExpansion();
1679 
1680     if (ExpInfo.getExpansionLocStart().isFileID()) {
1681       if (!isInFileID(ExpInfo.getExpansionLocStart(), FID))
1682         return; // No more files/macros that may be "contained" in this file.
1683     }
1684 
1685     if (!ExpInfo.isMacroArgExpansion())
1686       continue;
1687 
1688     SourceLocation SpellLoc = ExpInfo.getSpellingLoc();
1689     while (!SpellLoc.isFileID()) {
1690       std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(SpellLoc);
1691       const ExpansionInfo &Info = getSLocEntry(LocInfo.first).getExpansion();
1692       if (!Info.isMacroArgExpansion())
1693         break;
1694       SpellLoc = Info.getSpellingLoc().getLocWithOffset(LocInfo.second);
1695     }
1696     if (!SpellLoc.isFileID())
1697       continue;
1698 
1699     unsigned BeginOffs;
1700     if (!isInFileID(SpellLoc, FID, &BeginOffs))
1701       continue;
1702 
1703     unsigned EndOffs = BeginOffs + getFileIDSize(FileID::get(ID));
1704 
1705     // Add a new chunk for this macro argument. A previous macro argument chunk
1706     // may have been lexed again, so e.g. if the map is
1707     //     0   -> SourceLocation()
1708     //     100 -> Expanded loc #1
1709     //     110 -> SourceLocation()
1710     // and we found a new macro FileID that lexed from offet 105 with length 3,
1711     // the new map will be:
1712     //     0   -> SourceLocation()
1713     //     100 -> Expanded loc #1
1714     //     105 -> Expanded loc #2
1715     //     108 -> Expanded loc #1
1716     //     110 -> SourceLocation()
1717     //
1718     // Since re-lexed macro chunks will always be the same size or less of
1719     // previous chunks, we only need to find where the ending of the new macro
1720     // chunk is mapped to and update the map with new begin/end mappings.
1721 
1722     MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs);
1723     --I;
1724     SourceLocation EndOffsMappedLoc = I->second;
1725     MacroArgsCache[BeginOffs] = SourceLocation::getMacroLoc(Entry.getOffset());
1726     MacroArgsCache[EndOffs] = EndOffsMappedLoc;
1727   }
1728 }
1729 
1730 /// \brief If \arg Loc points inside a function macro argument, the returned
1731 /// location will be the macro location in which the argument was expanded.
1732 /// If a macro argument is used multiple times, the expanded location will
1733 /// be at the first expansion of the argument.
1734 /// e.g.
1735 ///   MY_MACRO(foo);
1736 ///             ^
1737 /// Passing a file location pointing at 'foo', will yield a macro location
1738 /// where 'foo' was expanded into.
1739 SourceLocation
1740 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const {
1741   if (Loc.isInvalid() || !Loc.isFileID())
1742     return Loc;
1743 
1744   FileID FID;
1745   unsigned Offset;
1746   llvm::tie(FID, Offset) = getDecomposedLoc(Loc);
1747   if (FID.isInvalid())
1748     return Loc;
1749 
1750   MacroArgsMap *&MacroArgsCache = MacroArgsCacheMap[FID];
1751   if (!MacroArgsCache)
1752     computeMacroArgsCache(MacroArgsCache, FID);
1753 
1754   assert(!MacroArgsCache->empty());
1755   MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset);
1756   --I;
1757 
1758   unsigned MacroArgBeginOffs = I->first;
1759   SourceLocation MacroArgExpandedLoc = I->second;
1760   if (MacroArgExpandedLoc.isValid())
1761     return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs);
1762 
1763   return Loc;
1764 }
1765 
1766 /// Given a decomposed source location, move it up the include/expansion stack
1767 /// to the parent source location.  If this is possible, return the decomposed
1768 /// version of the parent in Loc and return false.  If Loc is the top-level
1769 /// entry, return true and don't modify it.
1770 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc,
1771                                    const SourceManager &SM) {
1772   SourceLocation UpperLoc;
1773   const SrcMgr::SLocEntry &Entry = SM.getSLocEntry(Loc.first);
1774   if (Entry.isExpansion())
1775     UpperLoc = Entry.getExpansion().getExpansionLocStart();
1776   else
1777     UpperLoc = Entry.getFile().getIncludeLoc();
1778 
1779   if (UpperLoc.isInvalid())
1780     return true; // We reached the top.
1781 
1782   Loc = SM.getDecomposedLoc(UpperLoc);
1783   return false;
1784 }
1785 
1786 
1787 /// \brief Determines the order of 2 source locations in the translation unit.
1788 ///
1789 /// \returns true if LHS source location comes before RHS, false otherwise.
1790 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
1791                                               SourceLocation RHS) const {
1792   assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
1793   if (LHS == RHS)
1794     return false;
1795 
1796   std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
1797   std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
1798 
1799   // If the source locations are in the same file, just compare offsets.
1800   if (LOffs.first == ROffs.first)
1801     return LOffs.second < ROffs.second;
1802 
1803   // If we are comparing a source location with multiple locations in the same
1804   // file, we get a big win by caching the result.
1805   if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first))
1806     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1807 
1808   // Okay, we missed in the cache, start updating the cache for this query.
1809   IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first,
1810                           /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID);
1811 
1812   // We need to find the common ancestor. The only way of doing this is to
1813   // build the complete include chain for one and then walking up the chain
1814   // of the other looking for a match.
1815   // We use a map from FileID to Offset to store the chain. Easier than writing
1816   // a custom set hash info that only depends on the first part of a pair.
1817   typedef llvm::DenseMap<FileID, unsigned> LocSet;
1818   LocSet LChain;
1819   do {
1820     LChain.insert(LOffs);
1821     // We catch the case where LOffs is in a file included by ROffs and
1822     // quit early. The other way round unfortunately remains suboptimal.
1823   } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this));
1824   LocSet::iterator I;
1825   while((I = LChain.find(ROffs.first)) == LChain.end()) {
1826     if (MoveUpIncludeHierarchy(ROffs, *this))
1827       break; // Met at topmost file.
1828   }
1829   if (I != LChain.end())
1830     LOffs = *I;
1831 
1832   // If we exited because we found a nearest common ancestor, compare the
1833   // locations within the common file and cache them.
1834   if (LOffs.first == ROffs.first) {
1835     IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second);
1836     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1837   }
1838 
1839   // This can happen if a location is in a built-ins buffer.
1840   // But see PR5662.
1841   // Clear the lookup cache, it depends on a common location.
1842   IsBeforeInTUCache.clear();
1843   bool LIsBuiltins = strcmp("<built-in>",
1844                             getBuffer(LOffs.first)->getBufferIdentifier()) == 0;
1845   bool RIsBuiltins = strcmp("<built-in>",
1846                             getBuffer(ROffs.first)->getBufferIdentifier()) == 0;
1847   // built-in is before non-built-in
1848   if (LIsBuiltins != RIsBuiltins)
1849     return LIsBuiltins;
1850   assert(LIsBuiltins && RIsBuiltins &&
1851          "Non-built-in locations must be rooted in the main file");
1852   // Both are in built-in buffers, but from different files. We just claim that
1853   // lower IDs come first.
1854   return LOffs.first < ROffs.first;
1855 }
1856 
1857 /// PrintStats - Print statistics to stderr.
1858 ///
1859 void SourceManager::PrintStats() const {
1860   llvm::errs() << "\n*** Source Manager Stats:\n";
1861   llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
1862                << " mem buffers mapped.\n";
1863   llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated ("
1864                << llvm::capacity_in_bytes(LocalSLocEntryTable)
1865                << " bytes of capacity), "
1866                << NextLocalOffset << "B of Sloc address space used.\n";
1867   llvm::errs() << LoadedSLocEntryTable.size()
1868                << " loaded SLocEntries allocated, "
1869                << MaxLoadedOffset - CurrentLoadedOffset
1870                << "B of Sloc address space used.\n";
1871 
1872   unsigned NumLineNumsComputed = 0;
1873   unsigned NumFileBytesMapped = 0;
1874   for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
1875     NumLineNumsComputed += I->second->SourceLineCache != 0;
1876     NumFileBytesMapped  += I->second->getSizeBytesMapped();
1877   }
1878   unsigned NumMacroArgsComputed = MacroArgsCacheMap.size();
1879 
1880   llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
1881                << NumLineNumsComputed << " files with line #'s computed, "
1882                << NumMacroArgsComputed << " files with macro args computed.\n";
1883   llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
1884                << NumBinaryProbes << " binary.\n";
1885 }
1886 
1887 ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
1888 
1889 /// Return the amount of memory used by memory buffers, breaking down
1890 /// by heap-backed versus mmap'ed memory.
1891 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const {
1892   size_t malloc_bytes = 0;
1893   size_t mmap_bytes = 0;
1894 
1895   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i)
1896     if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped())
1897       switch (MemBufferInfos[i]->getMemoryBufferKind()) {
1898         case llvm::MemoryBuffer::MemoryBuffer_MMap:
1899           mmap_bytes += sized_mapped;
1900           break;
1901         case llvm::MemoryBuffer::MemoryBuffer_Malloc:
1902           malloc_bytes += sized_mapped;
1903           break;
1904       }
1905 
1906   return MemoryBufferSizes(malloc_bytes, mmap_bytes);
1907 }
1908 
1909 size_t SourceManager::getDataStructureSizes() const {
1910   size_t size = llvm::capacity_in_bytes(MemBufferInfos)
1911     + llvm::capacity_in_bytes(LocalSLocEntryTable)
1912     + llvm::capacity_in_bytes(LoadedSLocEntryTable)
1913     + llvm::capacity_in_bytes(SLocEntryLoaded)
1914     + llvm::capacity_in_bytes(FileInfos);
1915 
1916   if (OverriddenFilesInfo)
1917     size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles);
1918 
1919   return size;
1920 }
1921