• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
~ContentCache()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.
getSizeBytesMapped() const46 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.
getMemoryBufferKind() const52 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.
getSize() const67 unsigned ContentCache::getSize() const {
68   return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize()
69                              : (unsigned) ContentsEntry->getSize();
70 }
71 
replaceBuffer(const llvm::MemoryBuffer * B,bool DoNotFree)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 
getBuffer(DiagnosticsEngine & Diag,const SourceManager & SM,SourceLocation Loc,bool * Invalid) const86 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 
getLineTableFilenameID(StringRef Name)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.
AddLineNote(FileID FID,unsigned Offset,unsigned LineNo,int 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.
AddLineNote(FileID FID,unsigned Offset,unsigned LineNo,int FilenameID,unsigned EntryExit,SrcMgr::CharacteristicKind FileKind)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.
FindNearestLineEntry(FileID FID,unsigned Offset)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.
AddEntry(FileID FID,const std::vector<LineEntry> & Entries)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 ///
getLineTableFilenameID(StringRef Name)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.
AddLineNote(SourceLocation Loc,unsigned LineNo,int FilenameID)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.
AddLineNote(SourceLocation Loc,unsigned LineNo,int FilenameID,bool IsFileEntry,bool IsFileExit,bool IsSystemHeader,bool IsExternCHeader)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 
getLineTable()363 LineTableInfo &SourceManager::getLineTable() {
364   if (LineTable == 0)
365     LineTable = new LineTableInfo();
366   return *LineTable;
367 }
368 
369 //===----------------------------------------------------------------------===//
370 // Private 'Create' methods.
371 //===----------------------------------------------------------------------===//
372 
SourceManager(DiagnosticsEngine & Diag,FileManager & FileMgr,bool UserFilesAreVolatile)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 
~SourceManager()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 
clearIDTables()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 *
getOrCreateContentCache(const FileEntry * FileEnt,bool isSystemFile)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*
createMemBufferContentCache(const MemoryBuffer * Buffer)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 
loadSLocEntry(unsigned Index,bool * Invalid) const486 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>
AllocateLoadedSLocEntries(unsigned NumSLocEntries,unsigned TotalSize)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.
getFakeBufferForRecovery() const519 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 *
getFakeContentCacheForRecovery() const530 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.
createFileID(const ContentCache * File,SourceLocation IncludePos,SrcMgr::CharacteristicKind FileCharacter,int LoadedID,unsigned LoadedOffset)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
createMacroArgExpansionLoc(SourceLocation SpellingLoc,SourceLocation ExpansionLoc,unsigned TokLength)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
createExpansionLoc(SourceLocation SpellingLoc,SourceLocation ExpansionLocStart,SourceLocation ExpansionLocEnd,unsigned TokLength,int LoadedID,unsigned LoadedOffset)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
createExpansionLocImpl(const ExpansionInfo & Info,unsigned TokLength,int LoadedID,unsigned LoadedOffset)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 *
getMemoryBufferForFile(const FileEntry * File,bool * Invalid)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 
overrideFileContents(const FileEntry * SourceFile,const llvm::MemoryBuffer * Buffer,bool DoNotFree)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 
overrideFileContents(const FileEntry * SourceFile,const FileEntry * NewFile)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 
disableFileContentsOverride(const FileEntry * File)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 
getBufferData(FileID FID,bool * Invalid) const665 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.
getFileIDSlow(unsigned SLocOffset) const695 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.
getFileIDLocal(unsigned SLocOffset) const710 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   const SrcMgr::SLocEntry *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.
getFileIDLoaded(unsigned SLocOffset) const801 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     if (E.getOffset() == 0)
844       return FileID(); // invalid entry.
845 
846     ++NumProbes;
847 
848     if (E.getOffset() > SLocOffset) {
849       // Sanity checking, otherwise a bug may lead to hanging in release build.
850       if (GreaterIndex == MiddleIndex) {
851         assert(0 && "binary search missed the entry");
852         return FileID();
853       }
854       GreaterIndex = MiddleIndex;
855       continue;
856     }
857 
858     if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) {
859       FileID Res = FileID::get(-int(MiddleIndex) - 2);
860       if (!E.isExpansion())
861         LastFileIDLookup = Res;
862       NumBinaryProbes += NumProbes;
863       return Res;
864     }
865 
866     // Sanity checking, otherwise a bug may lead to hanging in release build.
867     if (LessIndex == MiddleIndex) {
868       assert(0 && "binary search missed the entry");
869       return FileID();
870     }
871     LessIndex = MiddleIndex;
872   }
873 }
874 
875 SourceLocation SourceManager::
getExpansionLocSlowCase(SourceLocation Loc) const876 getExpansionLocSlowCase(SourceLocation Loc) const {
877   do {
878     // Note: If Loc indicates an offset into a token that came from a macro
879     // expansion (e.g. the 5th character of the token) we do not want to add
880     // this offset when going to the expansion location.  The expansion
881     // location is the macro invocation, which the offset has nothing to do
882     // with.  This is unlike when we get the spelling loc, because the offset
883     // directly correspond to the token whose spelling we're inspecting.
884     Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart();
885   } while (!Loc.isFileID());
886 
887   return Loc;
888 }
889 
getSpellingLocSlowCase(SourceLocation Loc) const890 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
891   do {
892     std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
893     Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
894     Loc = Loc.getLocWithOffset(LocInfo.second);
895   } while (!Loc.isFileID());
896   return Loc;
897 }
898 
getFileLocSlowCase(SourceLocation Loc) const899 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const {
900   do {
901     if (isMacroArgExpansion(Loc))
902       Loc = getImmediateSpellingLoc(Loc);
903     else
904       Loc = getImmediateExpansionRange(Loc).first;
905   } while (!Loc.isFileID());
906   return Loc;
907 }
908 
909 
910 std::pair<FileID, unsigned>
getDecomposedExpansionLocSlowCase(const SrcMgr::SLocEntry * E) const911 SourceManager::getDecomposedExpansionLocSlowCase(
912                                              const SrcMgr::SLocEntry *E) const {
913   // If this is an expansion record, walk through all the expansion points.
914   FileID FID;
915   SourceLocation Loc;
916   unsigned Offset;
917   do {
918     Loc = E->getExpansion().getExpansionLocStart();
919 
920     FID = getFileID(Loc);
921     E = &getSLocEntry(FID);
922     Offset = Loc.getOffset()-E->getOffset();
923   } while (!Loc.isFileID());
924 
925   return std::make_pair(FID, Offset);
926 }
927 
928 std::pair<FileID, unsigned>
getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry * E,unsigned Offset) const929 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
930                                                 unsigned Offset) const {
931   // If this is an expansion record, walk through all the expansion points.
932   FileID FID;
933   SourceLocation Loc;
934   do {
935     Loc = E->getExpansion().getSpellingLoc();
936     Loc = Loc.getLocWithOffset(Offset);
937 
938     FID = getFileID(Loc);
939     E = &getSLocEntry(FID);
940     Offset = Loc.getOffset()-E->getOffset();
941   } while (!Loc.isFileID());
942 
943   return std::make_pair(FID, Offset);
944 }
945 
946 /// getImmediateSpellingLoc - Given a SourceLocation object, return the
947 /// spelling location referenced by the ID.  This is the first level down
948 /// towards the place where the characters that make up the lexed token can be
949 /// found.  This should not generally be used by clients.
getImmediateSpellingLoc(SourceLocation Loc) const950 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
951   if (Loc.isFileID()) return Loc;
952   std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
953   Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
954   return Loc.getLocWithOffset(LocInfo.second);
955 }
956 
957 
958 /// getImmediateExpansionRange - Loc is required to be an expansion location.
959 /// Return the start/end of the expansion information.
960 std::pair<SourceLocation,SourceLocation>
getImmediateExpansionRange(SourceLocation Loc) const961 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const {
962   assert(Loc.isMacroID() && "Not a macro expansion loc!");
963   const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion();
964   return Expansion.getExpansionLocRange();
965 }
966 
967 /// getExpansionRange - Given a SourceLocation object, return the range of
968 /// tokens covered by the expansion in the ultimate file.
969 std::pair<SourceLocation,SourceLocation>
getExpansionRange(SourceLocation Loc) const970 SourceManager::getExpansionRange(SourceLocation Loc) const {
971   if (Loc.isFileID()) return std::make_pair(Loc, Loc);
972 
973   std::pair<SourceLocation,SourceLocation> Res =
974     getImmediateExpansionRange(Loc);
975 
976   // Fully resolve the start and end locations to their ultimate expansion
977   // points.
978   while (!Res.first.isFileID())
979     Res.first = getImmediateExpansionRange(Res.first).first;
980   while (!Res.second.isFileID())
981     Res.second = getImmediateExpansionRange(Res.second).second;
982   return Res;
983 }
984 
isMacroArgExpansion(SourceLocation Loc) const985 bool SourceManager::isMacroArgExpansion(SourceLocation Loc) const {
986   if (!Loc.isMacroID()) return false;
987 
988   FileID FID = getFileID(Loc);
989   const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion();
990   return Expansion.isMacroArgExpansion();
991 }
992 
isMacroBodyExpansion(SourceLocation Loc) const993 bool SourceManager::isMacroBodyExpansion(SourceLocation Loc) const {
994   if (!Loc.isMacroID()) return false;
995 
996   FileID FID = getFileID(Loc);
997   const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion();
998   return Expansion.isMacroBodyExpansion();
999 }
1000 
1001 
1002 //===----------------------------------------------------------------------===//
1003 // Queries about the code at a SourceLocation.
1004 //===----------------------------------------------------------------------===//
1005 
1006 /// getCharacterData - Return a pointer to the start of the specified location
1007 /// in the appropriate MemoryBuffer.
getCharacterData(SourceLocation SL,bool * Invalid) const1008 const char *SourceManager::getCharacterData(SourceLocation SL,
1009                                             bool *Invalid) const {
1010   // Note that this is a hot function in the getSpelling() path, which is
1011   // heavily used by -E mode.
1012   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
1013 
1014   // Note that calling 'getBuffer()' may lazily page in a source file.
1015   bool CharDataInvalid = false;
1016   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid);
1017   if (CharDataInvalid || !Entry.isFile()) {
1018     if (Invalid)
1019       *Invalid = true;
1020 
1021     return "<<<<INVALID BUFFER>>>>";
1022   }
1023   const llvm::MemoryBuffer *Buffer
1024     = Entry.getFile().getContentCache()
1025                   ->getBuffer(Diag, *this, SourceLocation(), &CharDataInvalid);
1026   if (Invalid)
1027     *Invalid = CharDataInvalid;
1028   return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
1029 }
1030 
1031 
1032 /// getColumnNumber - Return the column # for the specified file position.
1033 /// this is significantly cheaper to compute than the line number.
getColumnNumber(FileID FID,unsigned FilePos,bool * Invalid) const1034 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
1035                                         bool *Invalid) const {
1036   bool MyInvalid = false;
1037   const llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid);
1038   if (Invalid)
1039     *Invalid = MyInvalid;
1040 
1041   if (MyInvalid)
1042     return 1;
1043 
1044   // It is okay to request a position just past the end of the buffer.
1045   if (FilePos > MemBuf->getBufferSize()) {
1046     if (Invalid)
1047       *Invalid = true;
1048     return 1;
1049   }
1050 
1051   // See if we just calculated the line number for this FilePos and can use
1052   // that to lookup the start of the line instead of searching for it.
1053   if (LastLineNoFileIDQuery == FID &&
1054       LastLineNoContentCache->SourceLineCache != 0 &&
1055       LastLineNoResult < LastLineNoContentCache->NumLines) {
1056     unsigned *SourceLineCache = LastLineNoContentCache->SourceLineCache;
1057     unsigned LineStart = SourceLineCache[LastLineNoResult - 1];
1058     unsigned LineEnd = SourceLineCache[LastLineNoResult];
1059     if (FilePos >= LineStart && FilePos < LineEnd)
1060       return FilePos - LineStart + 1;
1061   }
1062 
1063   const char *Buf = MemBuf->getBufferStart();
1064   unsigned LineStart = FilePos;
1065   while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
1066     --LineStart;
1067   return FilePos-LineStart+1;
1068 }
1069 
1070 // isInvalid - Return the result of calling loc.isInvalid(), and
1071 // if Invalid is not null, set its value to same.
isInvalid(SourceLocation Loc,bool * Invalid)1072 static bool isInvalid(SourceLocation Loc, bool *Invalid) {
1073   bool MyInvalid = Loc.isInvalid();
1074   if (Invalid)
1075     *Invalid = MyInvalid;
1076   return MyInvalid;
1077 }
1078 
getSpellingColumnNumber(SourceLocation Loc,bool * Invalid) const1079 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
1080                                                 bool *Invalid) const {
1081   if (isInvalid(Loc, Invalid)) return 0;
1082   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1083   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1084 }
1085 
getExpansionColumnNumber(SourceLocation Loc,bool * Invalid) const1086 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc,
1087                                                  bool *Invalid) const {
1088   if (isInvalid(Loc, Invalid)) return 0;
1089   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1090   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
1091 }
1092 
getPresumedColumnNumber(SourceLocation Loc,bool * Invalid) const1093 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc,
1094                                                 bool *Invalid) const {
1095   if (isInvalid(Loc, Invalid)) return 0;
1096   return getPresumedLoc(Loc).getColumn();
1097 }
1098 
1099 #ifdef __SSE2__
1100 #include <emmintrin.h>
1101 #endif
1102 
1103 static LLVM_ATTRIBUTE_NOINLINE void
1104 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1105                    llvm::BumpPtrAllocator &Alloc,
1106                    const SourceManager &SM, bool &Invalid);
ComputeLineNumbers(DiagnosticsEngine & Diag,ContentCache * FI,llvm::BumpPtrAllocator & Alloc,const SourceManager & SM,bool & Invalid)1107 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
1108                                llvm::BumpPtrAllocator &Alloc,
1109                                const SourceManager &SM, bool &Invalid) {
1110   // Note that calling 'getBuffer()' may lazily page in the file.
1111   const MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(),
1112                                              &Invalid);
1113   if (Invalid)
1114     return;
1115 
1116   // Find the file offsets of all of the *physical* source lines.  This does
1117   // not look at trigraphs, escaped newlines, or anything else tricky.
1118   SmallVector<unsigned, 256> LineOffsets;
1119 
1120   // Line #1 starts at char 0.
1121   LineOffsets.push_back(0);
1122 
1123   const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
1124   const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
1125   unsigned Offs = 0;
1126   while (1) {
1127     // Skip over the contents of the line.
1128     const unsigned char *NextBuf = (const unsigned char *)Buf;
1129 
1130 #ifdef __SSE2__
1131     // Try to skip to the next newline using SSE instructions. This is very
1132     // performance sensitive for programs with lots of diagnostics and in -E
1133     // mode.
1134     __m128i CRs = _mm_set1_epi8('\r');
1135     __m128i LFs = _mm_set1_epi8('\n');
1136 
1137     // First fix up the alignment to 16 bytes.
1138     while (((uintptr_t)NextBuf & 0xF) != 0) {
1139       if (*NextBuf == '\n' || *NextBuf == '\r' || *NextBuf == '\0')
1140         goto FoundSpecialChar;
1141       ++NextBuf;
1142     }
1143 
1144     // Scan 16 byte chunks for '\r' and '\n'. Ignore '\0'.
1145     while (NextBuf+16 <= End) {
1146       const __m128i Chunk = *(const __m128i*)NextBuf;
1147       __m128i Cmp = _mm_or_si128(_mm_cmpeq_epi8(Chunk, CRs),
1148                                  _mm_cmpeq_epi8(Chunk, LFs));
1149       unsigned Mask = _mm_movemask_epi8(Cmp);
1150 
1151       // If we found a newline, adjust the pointer and jump to the handling code.
1152       if (Mask != 0) {
1153         NextBuf += llvm::CountTrailingZeros_32(Mask);
1154         goto FoundSpecialChar;
1155       }
1156       NextBuf += 16;
1157     }
1158 #endif
1159 
1160     while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
1161       ++NextBuf;
1162 
1163 #ifdef __SSE2__
1164 FoundSpecialChar:
1165 #endif
1166     Offs += NextBuf-Buf;
1167     Buf = NextBuf;
1168 
1169     if (Buf[0] == '\n' || Buf[0] == '\r') {
1170       // If this is \n\r or \r\n, skip both characters.
1171       if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1])
1172         ++Offs, ++Buf;
1173       ++Offs, ++Buf;
1174       LineOffsets.push_back(Offs);
1175     } else {
1176       // Otherwise, this is a null.  If end of file, exit.
1177       if (Buf == End) break;
1178       // Otherwise, skip the null.
1179       ++Offs, ++Buf;
1180     }
1181   }
1182 
1183   // Copy the offsets into the FileInfo structure.
1184   FI->NumLines = LineOffsets.size();
1185   FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
1186   std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
1187 }
1188 
1189 /// getLineNumber - Given a SourceLocation, return the spelling line number
1190 /// for the position indicated.  This requires building and caching a table of
1191 /// line offsets for the MemoryBuffer, so this is not cheap: use only when
1192 /// about to emit a diagnostic.
getLineNumber(FileID FID,unsigned FilePos,bool * Invalid) const1193 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
1194                                       bool *Invalid) const {
1195   if (FID.isInvalid()) {
1196     if (Invalid)
1197       *Invalid = true;
1198     return 1;
1199   }
1200 
1201   ContentCache *Content;
1202   if (LastLineNoFileIDQuery == FID)
1203     Content = LastLineNoContentCache;
1204   else {
1205     bool MyInvalid = false;
1206     const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid);
1207     if (MyInvalid || !Entry.isFile()) {
1208       if (Invalid)
1209         *Invalid = true;
1210       return 1;
1211     }
1212 
1213     Content = const_cast<ContentCache*>(Entry.getFile().getContentCache());
1214   }
1215 
1216   // If this is the first use of line information for this buffer, compute the
1217   /// SourceLineCache for it on demand.
1218   if (Content->SourceLineCache == 0) {
1219     bool MyInvalid = false;
1220     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1221     if (Invalid)
1222       *Invalid = MyInvalid;
1223     if (MyInvalid)
1224       return 1;
1225   } else if (Invalid)
1226     *Invalid = false;
1227 
1228   // Okay, we know we have a line number table.  Do a binary search to find the
1229   // line number that this character position lands on.
1230   unsigned *SourceLineCache = Content->SourceLineCache;
1231   unsigned *SourceLineCacheStart = SourceLineCache;
1232   unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
1233 
1234   unsigned QueriedFilePos = FilePos+1;
1235 
1236   // FIXME: I would like to be convinced that this code is worth being as
1237   // complicated as it is, binary search isn't that slow.
1238   //
1239   // If it is worth being optimized, then in my opinion it could be more
1240   // performant, simpler, and more obviously correct by just "galloping" outward
1241   // from the queried file position. In fact, this could be incorporated into a
1242   // generic algorithm such as lower_bound_with_hint.
1243   //
1244   // If someone gives me a test case where this matters, and I will do it! - DWD
1245 
1246   // If the previous query was to the same file, we know both the file pos from
1247   // that query and the line number returned.  This allows us to narrow the
1248   // search space from the entire file to something near the match.
1249   if (LastLineNoFileIDQuery == FID) {
1250     if (QueriedFilePos >= LastLineNoFilePos) {
1251       // FIXME: Potential overflow?
1252       SourceLineCache = SourceLineCache+LastLineNoResult-1;
1253 
1254       // The query is likely to be nearby the previous one.  Here we check to
1255       // see if it is within 5, 10 or 20 lines.  It can be far away in cases
1256       // where big comment blocks and vertical whitespace eat up lines but
1257       // contribute no tokens.
1258       if (SourceLineCache+5 < SourceLineCacheEnd) {
1259         if (SourceLineCache[5] > QueriedFilePos)
1260           SourceLineCacheEnd = SourceLineCache+5;
1261         else if (SourceLineCache+10 < SourceLineCacheEnd) {
1262           if (SourceLineCache[10] > QueriedFilePos)
1263             SourceLineCacheEnd = SourceLineCache+10;
1264           else if (SourceLineCache+20 < SourceLineCacheEnd) {
1265             if (SourceLineCache[20] > QueriedFilePos)
1266               SourceLineCacheEnd = SourceLineCache+20;
1267           }
1268         }
1269       }
1270     } else {
1271       if (LastLineNoResult < Content->NumLines)
1272         SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
1273     }
1274   }
1275 
1276   // If the spread is large, do a "radix" test as our initial guess, based on
1277   // the assumption that lines average to approximately the same length.
1278   // NOTE: This is currently disabled, as it does not appear to be profitable in
1279   // initial measurements.
1280   if (0 && SourceLineCacheEnd-SourceLineCache > 20) {
1281     unsigned FileLen = Content->SourceLineCache[Content->NumLines-1];
1282 
1283     // Take a stab at guessing where it is.
1284     unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen;
1285 
1286     // Check for -10 and +10 lines.
1287     unsigned LowerBound = std::max(int(ApproxPos-10), 0);
1288     unsigned UpperBound = std::min(ApproxPos+10, FileLen);
1289 
1290     // If the computed lower bound is less than the query location, move it in.
1291     if (SourceLineCache < SourceLineCacheStart+LowerBound &&
1292         SourceLineCacheStart[LowerBound] < QueriedFilePos)
1293       SourceLineCache = SourceLineCacheStart+LowerBound;
1294 
1295     // If the computed upper bound is greater than the query location, move it.
1296     if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound &&
1297         SourceLineCacheStart[UpperBound] >= QueriedFilePos)
1298       SourceLineCacheEnd = SourceLineCacheStart+UpperBound;
1299   }
1300 
1301   unsigned *Pos
1302     = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
1303   unsigned LineNo = Pos-SourceLineCacheStart;
1304 
1305   LastLineNoFileIDQuery = FID;
1306   LastLineNoContentCache = Content;
1307   LastLineNoFilePos = QueriedFilePos;
1308   LastLineNoResult = LineNo;
1309   return LineNo;
1310 }
1311 
getSpellingLineNumber(SourceLocation Loc,bool * Invalid) const1312 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
1313                                               bool *Invalid) const {
1314   if (isInvalid(Loc, Invalid)) return 0;
1315   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
1316   return getLineNumber(LocInfo.first, LocInfo.second);
1317 }
getExpansionLineNumber(SourceLocation Loc,bool * Invalid) const1318 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc,
1319                                                bool *Invalid) const {
1320   if (isInvalid(Loc, Invalid)) return 0;
1321   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1322   return getLineNumber(LocInfo.first, LocInfo.second);
1323 }
getPresumedLineNumber(SourceLocation Loc,bool * Invalid) const1324 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc,
1325                                               bool *Invalid) const {
1326   if (isInvalid(Loc, Invalid)) return 0;
1327   return getPresumedLoc(Loc).getLine();
1328 }
1329 
1330 /// getFileCharacteristic - return the file characteristic of the specified
1331 /// source location, indicating whether this is a normal file, a system
1332 /// header, or an "implicit extern C" system header.
1333 ///
1334 /// This state can be modified with flags on GNU linemarker directives like:
1335 ///   # 4 "foo.h" 3
1336 /// which changes all source locations in the current file after that to be
1337 /// considered to be from a system header.
1338 SrcMgr::CharacteristicKind
getFileCharacteristic(SourceLocation Loc) const1339 SourceManager::getFileCharacteristic(SourceLocation Loc) const {
1340   assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!");
1341   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1342   bool Invalid = false;
1343   const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid);
1344   if (Invalid || !SEntry.isFile())
1345     return C_User;
1346 
1347   const SrcMgr::FileInfo &FI = SEntry.getFile();
1348 
1349   // If there are no #line directives in this file, just return the whole-file
1350   // state.
1351   if (!FI.hasLineDirectives())
1352     return FI.getFileCharacteristic();
1353 
1354   assert(LineTable && "Can't have linetable entries without a LineTable!");
1355   // See if there is a #line directive before the location.
1356   const LineEntry *Entry =
1357     LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second);
1358 
1359   // If this is before the first line marker, use the file characteristic.
1360   if (!Entry)
1361     return FI.getFileCharacteristic();
1362 
1363   return Entry->FileKind;
1364 }
1365 
1366 /// Return the filename or buffer identifier of the buffer the location is in.
1367 /// Note that this name does not respect \#line directives.  Use getPresumedLoc
1368 /// for normal clients.
getBufferName(SourceLocation Loc,bool * Invalid) const1369 const char *SourceManager::getBufferName(SourceLocation Loc,
1370                                          bool *Invalid) const {
1371   if (isInvalid(Loc, Invalid)) return "<invalid loc>";
1372 
1373   return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
1374 }
1375 
1376 
1377 /// getPresumedLoc - This method returns the "presumed" location of a
1378 /// SourceLocation specifies.  A "presumed location" can be modified by \#line
1379 /// or GNU line marker directives.  This provides a view on the data that a
1380 /// user should see in diagnostics, for example.
1381 ///
1382 /// Note that a presumed location is always given as the expansion point of an
1383 /// expansion location, not at the spelling location.
getPresumedLoc(SourceLocation Loc,bool UseLineDirectives) const1384 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc,
1385                                           bool UseLineDirectives) const {
1386   if (Loc.isInvalid()) return PresumedLoc();
1387 
1388   // Presumed locations are always for expansion points.
1389   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
1390 
1391   bool Invalid = false;
1392   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
1393   if (Invalid || !Entry.isFile())
1394     return PresumedLoc();
1395 
1396   const SrcMgr::FileInfo &FI = Entry.getFile();
1397   const SrcMgr::ContentCache *C = FI.getContentCache();
1398 
1399   // To get the source name, first consult the FileEntry (if one exists)
1400   // before the MemBuffer as this will avoid unnecessarily paging in the
1401   // MemBuffer.
1402   const char *Filename;
1403   if (C->OrigEntry)
1404     Filename = C->OrigEntry->getName();
1405   else
1406     Filename = C->getBuffer(Diag, *this)->getBufferIdentifier();
1407 
1408   unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid);
1409   if (Invalid)
1410     return PresumedLoc();
1411   unsigned ColNo  = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid);
1412   if (Invalid)
1413     return PresumedLoc();
1414 
1415   SourceLocation IncludeLoc = FI.getIncludeLoc();
1416 
1417   // If we have #line directives in this file, update and overwrite the physical
1418   // location info if appropriate.
1419   if (UseLineDirectives && FI.hasLineDirectives()) {
1420     assert(LineTable && "Can't have linetable entries without a LineTable!");
1421     // See if there is a #line directive before this.  If so, get it.
1422     if (const LineEntry *Entry =
1423           LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) {
1424       // If the LineEntry indicates a filename, use it.
1425       if (Entry->FilenameID != -1)
1426         Filename = LineTable->getFilename(Entry->FilenameID);
1427 
1428       // Use the line number specified by the LineEntry.  This line number may
1429       // be multiple lines down from the line entry.  Add the difference in
1430       // physical line numbers from the query point and the line marker to the
1431       // total.
1432       unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
1433       LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
1434 
1435       // Note that column numbers are not molested by line markers.
1436 
1437       // Handle virtual #include manipulation.
1438       if (Entry->IncludeOffset) {
1439         IncludeLoc = getLocForStartOfFile(LocInfo.first);
1440         IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset);
1441       }
1442     }
1443   }
1444 
1445   return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc);
1446 }
1447 
1448 /// \brief The size of the SLocEnty that \arg FID represents.
getFileIDSize(FileID FID) const1449 unsigned SourceManager::getFileIDSize(FileID FID) const {
1450   bool Invalid = false;
1451   const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1452   if (Invalid)
1453     return 0;
1454 
1455   int ID = FID.ID;
1456   unsigned NextOffset;
1457   if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size()))
1458     NextOffset = getNextLocalOffset();
1459   else if (ID+1 == -1)
1460     NextOffset = MaxLoadedOffset;
1461   else
1462     NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset();
1463 
1464   return NextOffset - Entry.getOffset() - 1;
1465 }
1466 
1467 //===----------------------------------------------------------------------===//
1468 // Other miscellaneous methods.
1469 //===----------------------------------------------------------------------===//
1470 
1471 /// \brief Retrieve the inode for the given file entry, if possible.
1472 ///
1473 /// This routine involves a system call, and therefore should only be used
1474 /// in non-performance-critical code.
getActualFileInode(const FileEntry * File)1475 static Optional<ino_t> getActualFileInode(const FileEntry *File) {
1476   if (!File)
1477     return None;
1478 
1479   struct stat StatBuf;
1480   if (::stat(File->getName(), &StatBuf))
1481     return None;
1482 
1483   return StatBuf.st_ino;
1484 }
1485 
1486 /// \brief Get the source location for the given file:line:col triplet.
1487 ///
1488 /// If the source file is included multiple times, the source location will
1489 /// be based upon an arbitrary inclusion.
translateFileLineCol(const FileEntry * SourceFile,unsigned Line,unsigned Col) const1490 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile,
1491                                                   unsigned Line,
1492                                                   unsigned Col) const {
1493   assert(SourceFile && "Null source file!");
1494   assert(Line && Col && "Line and column should start from 1!");
1495 
1496   FileID FirstFID = translateFile(SourceFile);
1497   return translateLineCol(FirstFID, Line, Col);
1498 }
1499 
1500 /// \brief Get the FileID for the given file.
1501 ///
1502 /// If the source file is included multiple times, the FileID will be the
1503 /// first inclusion.
translateFile(const FileEntry * SourceFile) const1504 FileID SourceManager::translateFile(const FileEntry *SourceFile) const {
1505   assert(SourceFile && "Null source file!");
1506 
1507   // Find the first file ID that corresponds to the given file.
1508   FileID FirstFID;
1509 
1510   // First, check the main file ID, since it is common to look for a
1511   // location in the main file.
1512   Optional<ino_t> SourceFileInode;
1513   Optional<StringRef> SourceFileName;
1514   if (!MainFileID.isInvalid()) {
1515     bool Invalid = false;
1516     const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid);
1517     if (Invalid)
1518       return FileID();
1519 
1520     if (MainSLoc.isFile()) {
1521       const ContentCache *MainContentCache
1522         = MainSLoc.getFile().getContentCache();
1523       if (!MainContentCache) {
1524         // Can't do anything
1525       } else if (MainContentCache->OrigEntry == SourceFile) {
1526         FirstFID = MainFileID;
1527       } else {
1528         // Fall back: check whether we have the same base name and inode
1529         // as the main file.
1530         const FileEntry *MainFile = MainContentCache->OrigEntry;
1531         SourceFileName = llvm::sys::path::filename(SourceFile->getName());
1532         if (*SourceFileName == llvm::sys::path::filename(MainFile->getName())) {
1533           SourceFileInode = getActualFileInode(SourceFile);
1534           if (SourceFileInode) {
1535             if (Optional<ino_t> MainFileInode = getActualFileInode(MainFile)) {
1536               if (*SourceFileInode == *MainFileInode) {
1537                 FirstFID = MainFileID;
1538                 SourceFile = MainFile;
1539               }
1540             }
1541           }
1542         }
1543       }
1544     }
1545   }
1546 
1547   if (FirstFID.isInvalid()) {
1548     // The location we're looking for isn't in the main file; look
1549     // through all of the local source locations.
1550     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1551       bool Invalid = false;
1552       const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid);
1553       if (Invalid)
1554         return FileID();
1555 
1556       if (SLoc.isFile() &&
1557           SLoc.getFile().getContentCache() &&
1558           SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1559         FirstFID = FileID::get(I);
1560         break;
1561       }
1562     }
1563     // If that still didn't help, try the modules.
1564     if (FirstFID.isInvalid()) {
1565       for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) {
1566         const SLocEntry &SLoc = getLoadedSLocEntry(I);
1567         if (SLoc.isFile() &&
1568             SLoc.getFile().getContentCache() &&
1569             SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
1570           FirstFID = FileID::get(-int(I) - 2);
1571           break;
1572         }
1573       }
1574     }
1575   }
1576 
1577   // If we haven't found what we want yet, try again, but this time stat()
1578   // each of the files in case the files have changed since we originally
1579   // parsed the file.
1580   if (FirstFID.isInvalid() &&
1581       (SourceFileName ||
1582        (SourceFileName = llvm::sys::path::filename(SourceFile->getName()))) &&
1583       (SourceFileInode ||
1584        (SourceFileInode = getActualFileInode(SourceFile)))) {
1585     bool Invalid = false;
1586     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
1587       FileID IFileID;
1588       IFileID.ID = I;
1589       const SLocEntry &SLoc = getSLocEntry(IFileID, &Invalid);
1590       if (Invalid)
1591         return FileID();
1592 
1593       if (SLoc.isFile()) {
1594         const ContentCache *FileContentCache
1595           = SLoc.getFile().getContentCache();
1596       const FileEntry *Entry =FileContentCache? FileContentCache->OrigEntry : 0;
1597         if (Entry &&
1598             *SourceFileName == llvm::sys::path::filename(Entry->getName())) {
1599           if (Optional<ino_t> EntryInode = getActualFileInode(Entry)) {
1600             if (*SourceFileInode == *EntryInode) {
1601               FirstFID = FileID::get(I);
1602               SourceFile = Entry;
1603               break;
1604             }
1605           }
1606         }
1607       }
1608     }
1609   }
1610 
1611   (void) SourceFile;
1612   return FirstFID;
1613 }
1614 
1615 /// \brief Get the source location in \arg FID for the given line:col.
1616 /// Returns null location if \arg FID is not a file SLocEntry.
translateLineCol(FileID FID,unsigned Line,unsigned Col) const1617 SourceLocation SourceManager::translateLineCol(FileID FID,
1618                                                unsigned Line,
1619                                                unsigned Col) const {
1620   if (FID.isInvalid())
1621     return SourceLocation();
1622 
1623   bool Invalid = false;
1624   const SLocEntry &Entry = getSLocEntry(FID, &Invalid);
1625   if (Invalid)
1626     return SourceLocation();
1627 
1628   if (!Entry.isFile())
1629     return SourceLocation();
1630 
1631   SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset());
1632 
1633   if (Line == 1 && Col == 1)
1634     return FileLoc;
1635 
1636   ContentCache *Content
1637     = const_cast<ContentCache *>(Entry.getFile().getContentCache());
1638   if (!Content)
1639     return SourceLocation();
1640 
1641   // If this is the first use of line information for this buffer, compute the
1642   // SourceLineCache for it on demand.
1643   if (Content->SourceLineCache == 0) {
1644     bool MyInvalid = false;
1645     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1646     if (MyInvalid)
1647       return SourceLocation();
1648   }
1649 
1650   if (Line > Content->NumLines) {
1651     unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize();
1652     if (Size > 0)
1653       --Size;
1654     return FileLoc.getLocWithOffset(Size);
1655   }
1656 
1657   const llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, *this);
1658   unsigned FilePos = Content->SourceLineCache[Line - 1];
1659   const char *Buf = Buffer->getBufferStart() + FilePos;
1660   unsigned BufLength = Buffer->getBufferSize() - FilePos;
1661   if (BufLength == 0)
1662     return FileLoc.getLocWithOffset(FilePos);
1663 
1664   unsigned i = 0;
1665 
1666   // Check that the given column is valid.
1667   while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
1668     ++i;
1669   if (i < Col-1)
1670     return FileLoc.getLocWithOffset(FilePos + i);
1671 
1672   return FileLoc.getLocWithOffset(FilePos + Col - 1);
1673 }
1674 
1675 /// \brief Compute a map of macro argument chunks to their expanded source
1676 /// location. Chunks that are not part of a macro argument will map to an
1677 /// invalid source location. e.g. if a file contains one macro argument at
1678 /// offset 100 with length 10, this is how the map will be formed:
1679 ///     0   -> SourceLocation()
1680 ///     100 -> Expanded macro arg location
1681 ///     110 -> SourceLocation()
computeMacroArgsCache(MacroArgsMap * & CachePtr,FileID FID) const1682 void SourceManager::computeMacroArgsCache(MacroArgsMap *&CachePtr,
1683                                           FileID FID) const {
1684   assert(!FID.isInvalid());
1685   assert(!CachePtr);
1686 
1687   CachePtr = new MacroArgsMap();
1688   MacroArgsMap &MacroArgsCache = *CachePtr;
1689   // Initially no macro argument chunk is present.
1690   MacroArgsCache.insert(std::make_pair(0, SourceLocation()));
1691 
1692   int ID = FID.ID;
1693   while (1) {
1694     ++ID;
1695     // Stop if there are no more FileIDs to check.
1696     if (ID > 0) {
1697       if (unsigned(ID) >= local_sloc_entry_size())
1698         return;
1699     } else if (ID == -1) {
1700       return;
1701     }
1702 
1703     const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID);
1704     if (Entry.isFile()) {
1705       SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc();
1706       if (IncludeLoc.isInvalid())
1707         continue;
1708       if (!isInFileID(IncludeLoc, FID))
1709         return; // No more files/macros that may be "contained" in this file.
1710 
1711       // Skip the files/macros of the #include'd file, we only care about macros
1712       // that lexed macro arguments from our file.
1713       if (Entry.getFile().NumCreatedFIDs)
1714         ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/;
1715       continue;
1716     }
1717 
1718     const ExpansionInfo &ExpInfo = Entry.getExpansion();
1719 
1720     if (ExpInfo.getExpansionLocStart().isFileID()) {
1721       if (!isInFileID(ExpInfo.getExpansionLocStart(), FID))
1722         return; // No more files/macros that may be "contained" in this file.
1723     }
1724 
1725     if (!ExpInfo.isMacroArgExpansion())
1726       continue;
1727 
1728     associateFileChunkWithMacroArgExp(MacroArgsCache, FID,
1729                                  ExpInfo.getSpellingLoc(),
1730                                  SourceLocation::getMacroLoc(Entry.getOffset()),
1731                                  getFileIDSize(FileID::get(ID)));
1732   }
1733 }
1734 
associateFileChunkWithMacroArgExp(MacroArgsMap & MacroArgsCache,FileID FID,SourceLocation SpellLoc,SourceLocation ExpansionLoc,unsigned ExpansionLength) const1735 void SourceManager::associateFileChunkWithMacroArgExp(
1736                                          MacroArgsMap &MacroArgsCache,
1737                                          FileID FID,
1738                                          SourceLocation SpellLoc,
1739                                          SourceLocation ExpansionLoc,
1740                                          unsigned ExpansionLength) const {
1741   if (!SpellLoc.isFileID()) {
1742     unsigned SpellBeginOffs = SpellLoc.getOffset();
1743     unsigned SpellEndOffs = SpellBeginOffs + ExpansionLength;
1744 
1745     // The spelling range for this macro argument expansion can span multiple
1746     // consecutive FileID entries. Go through each entry contained in the
1747     // spelling range and if one is itself a macro argument expansion, recurse
1748     // and associate the file chunk that it represents.
1749 
1750     FileID SpellFID; // Current FileID in the spelling range.
1751     unsigned SpellRelativeOffs;
1752     llvm::tie(SpellFID, SpellRelativeOffs) = getDecomposedLoc(SpellLoc);
1753     while (1) {
1754       const SLocEntry &Entry = getSLocEntry(SpellFID);
1755       unsigned SpellFIDBeginOffs = Entry.getOffset();
1756       unsigned SpellFIDSize = getFileIDSize(SpellFID);
1757       unsigned SpellFIDEndOffs = SpellFIDBeginOffs + SpellFIDSize;
1758       const ExpansionInfo &Info = Entry.getExpansion();
1759       if (Info.isMacroArgExpansion()) {
1760         unsigned CurrSpellLength;
1761         if (SpellFIDEndOffs < SpellEndOffs)
1762           CurrSpellLength = SpellFIDSize - SpellRelativeOffs;
1763         else
1764           CurrSpellLength = ExpansionLength;
1765         associateFileChunkWithMacroArgExp(MacroArgsCache, FID,
1766                       Info.getSpellingLoc().getLocWithOffset(SpellRelativeOffs),
1767                       ExpansionLoc, CurrSpellLength);
1768       }
1769 
1770       if (SpellFIDEndOffs >= SpellEndOffs)
1771         return; // we covered all FileID entries in the spelling range.
1772 
1773       // Move to the next FileID entry in the spelling range.
1774       unsigned advance = SpellFIDSize - SpellRelativeOffs + 1;
1775       ExpansionLoc = ExpansionLoc.getLocWithOffset(advance);
1776       ExpansionLength -= advance;
1777       ++SpellFID.ID;
1778       SpellRelativeOffs = 0;
1779     }
1780 
1781   }
1782 
1783   assert(SpellLoc.isFileID());
1784 
1785   unsigned BeginOffs;
1786   if (!isInFileID(SpellLoc, FID, &BeginOffs))
1787     return;
1788 
1789   unsigned EndOffs = BeginOffs + ExpansionLength;
1790 
1791   // Add a new chunk for this macro argument. A previous macro argument chunk
1792   // may have been lexed again, so e.g. if the map is
1793   //     0   -> SourceLocation()
1794   //     100 -> Expanded loc #1
1795   //     110 -> SourceLocation()
1796   // and we found a new macro FileID that lexed from offet 105 with length 3,
1797   // the new map will be:
1798   //     0   -> SourceLocation()
1799   //     100 -> Expanded loc #1
1800   //     105 -> Expanded loc #2
1801   //     108 -> Expanded loc #1
1802   //     110 -> SourceLocation()
1803   //
1804   // Since re-lexed macro chunks will always be the same size or less of
1805   // previous chunks, we only need to find where the ending of the new macro
1806   // chunk is mapped to and update the map with new begin/end mappings.
1807 
1808   MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs);
1809   --I;
1810   SourceLocation EndOffsMappedLoc = I->second;
1811   MacroArgsCache[BeginOffs] = ExpansionLoc;
1812   MacroArgsCache[EndOffs] = EndOffsMappedLoc;
1813 }
1814 
1815 /// \brief If \arg Loc points inside a function macro argument, the returned
1816 /// location will be the macro location in which the argument was expanded.
1817 /// If a macro argument is used multiple times, the expanded location will
1818 /// be at the first expansion of the argument.
1819 /// e.g.
1820 ///   MY_MACRO(foo);
1821 ///             ^
1822 /// Passing a file location pointing at 'foo', will yield a macro location
1823 /// where 'foo' was expanded into.
1824 SourceLocation
getMacroArgExpandedLocation(SourceLocation Loc) const1825 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const {
1826   if (Loc.isInvalid() || !Loc.isFileID())
1827     return Loc;
1828 
1829   FileID FID;
1830   unsigned Offset;
1831   llvm::tie(FID, Offset) = getDecomposedLoc(Loc);
1832   if (FID.isInvalid())
1833     return Loc;
1834 
1835   MacroArgsMap *&MacroArgsCache = MacroArgsCacheMap[FID];
1836   if (!MacroArgsCache)
1837     computeMacroArgsCache(MacroArgsCache, FID);
1838 
1839   assert(!MacroArgsCache->empty());
1840   MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset);
1841   --I;
1842 
1843   unsigned MacroArgBeginOffs = I->first;
1844   SourceLocation MacroArgExpandedLoc = I->second;
1845   if (MacroArgExpandedLoc.isValid())
1846     return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs);
1847 
1848   return Loc;
1849 }
1850 
1851 /// Given a decomposed source location, move it up the include/expansion stack
1852 /// to the parent source location.  If this is possible, return the decomposed
1853 /// version of the parent in Loc and return false.  If Loc is the top-level
1854 /// entry, return true and don't modify it.
MoveUpIncludeHierarchy(std::pair<FileID,unsigned> & Loc,const SourceManager & SM)1855 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc,
1856                                    const SourceManager &SM) {
1857   SourceLocation UpperLoc;
1858   const SrcMgr::SLocEntry &Entry = SM.getSLocEntry(Loc.first);
1859   if (Entry.isExpansion())
1860     UpperLoc = Entry.getExpansion().getExpansionLocStart();
1861   else
1862     UpperLoc = Entry.getFile().getIncludeLoc();
1863 
1864   if (UpperLoc.isInvalid())
1865     return true; // We reached the top.
1866 
1867   Loc = SM.getDecomposedLoc(UpperLoc);
1868   return false;
1869 }
1870 
1871 /// Return the cache entry for comparing the given file IDs
1872 /// for isBeforeInTranslationUnit.
getInBeforeInTUCache(FileID LFID,FileID RFID) const1873 InBeforeInTUCacheEntry &SourceManager::getInBeforeInTUCache(FileID LFID,
1874                                                             FileID RFID) const {
1875   // This is a magic number for limiting the cache size.  It was experimentally
1876   // derived from a small Objective-C project (where the cache filled
1877   // out to ~250 items).  We can make it larger if necessary.
1878   enum { MagicCacheSize = 300 };
1879   IsBeforeInTUCacheKey Key(LFID, RFID);
1880 
1881   // If the cache size isn't too large, do a lookup and if necessary default
1882   // construct an entry.  We can then return it to the caller for direct
1883   // use.  When they update the value, the cache will get automatically
1884   // updated as well.
1885   if (IBTUCache.size() < MagicCacheSize)
1886     return IBTUCache[Key];
1887 
1888   // Otherwise, do a lookup that will not construct a new value.
1889   InBeforeInTUCache::iterator I = IBTUCache.find(Key);
1890   if (I != IBTUCache.end())
1891     return I->second;
1892 
1893   // Fall back to the overflow value.
1894   return IBTUCacheOverflow;
1895 }
1896 
1897 /// \brief Determines the order of 2 source locations in the translation unit.
1898 ///
1899 /// \returns true if LHS source location comes before RHS, false otherwise.
isBeforeInTranslationUnit(SourceLocation LHS,SourceLocation RHS) const1900 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
1901                                               SourceLocation RHS) const {
1902   assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
1903   if (LHS == RHS)
1904     return false;
1905 
1906   std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
1907   std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
1908 
1909   // If the source locations are in the same file, just compare offsets.
1910   if (LOffs.first == ROffs.first)
1911     return LOffs.second < ROffs.second;
1912 
1913   // If we are comparing a source location with multiple locations in the same
1914   // file, we get a big win by caching the result.
1915   InBeforeInTUCacheEntry &IsBeforeInTUCache =
1916     getInBeforeInTUCache(LOffs.first, ROffs.first);
1917 
1918   // If we are comparing a source location with multiple locations in the same
1919   // file, we get a big win by caching the result.
1920   if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first))
1921     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1922 
1923   // Okay, we missed in the cache, start updating the cache for this query.
1924   IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first,
1925                           /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID);
1926 
1927   // We need to find the common ancestor. The only way of doing this is to
1928   // build the complete include chain for one and then walking up the chain
1929   // of the other looking for a match.
1930   // We use a map from FileID to Offset to store the chain. Easier than writing
1931   // a custom set hash info that only depends on the first part of a pair.
1932   typedef llvm::DenseMap<FileID, unsigned> LocSet;
1933   LocSet LChain;
1934   do {
1935     LChain.insert(LOffs);
1936     // We catch the case where LOffs is in a file included by ROffs and
1937     // quit early. The other way round unfortunately remains suboptimal.
1938   } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this));
1939   LocSet::iterator I;
1940   while((I = LChain.find(ROffs.first)) == LChain.end()) {
1941     if (MoveUpIncludeHierarchy(ROffs, *this))
1942       break; // Met at topmost file.
1943   }
1944   if (I != LChain.end())
1945     LOffs = *I;
1946 
1947   // If we exited because we found a nearest common ancestor, compare the
1948   // locations within the common file and cache them.
1949   if (LOffs.first == ROffs.first) {
1950     IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second);
1951     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1952   }
1953 
1954   // This can happen if a location is in a built-ins buffer.
1955   // But see PR5662.
1956   // Clear the lookup cache, it depends on a common location.
1957   IsBeforeInTUCache.clear();
1958   bool LIsBuiltins = strcmp("<built-in>",
1959                             getBuffer(LOffs.first)->getBufferIdentifier()) == 0;
1960   bool RIsBuiltins = strcmp("<built-in>",
1961                             getBuffer(ROffs.first)->getBufferIdentifier()) == 0;
1962   // built-in is before non-built-in
1963   if (LIsBuiltins != RIsBuiltins)
1964     return LIsBuiltins;
1965   assert(LIsBuiltins && RIsBuiltins &&
1966          "Non-built-in locations must be rooted in the main file");
1967   // Both are in built-in buffers, but from different files. We just claim that
1968   // lower IDs come first.
1969   return LOffs.first < ROffs.first;
1970 }
1971 
PrintStats() const1972 void SourceManager::PrintStats() const {
1973   llvm::errs() << "\n*** Source Manager Stats:\n";
1974   llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
1975                << " mem buffers mapped.\n";
1976   llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated ("
1977                << llvm::capacity_in_bytes(LocalSLocEntryTable)
1978                << " bytes of capacity), "
1979                << NextLocalOffset << "B of Sloc address space used.\n";
1980   llvm::errs() << LoadedSLocEntryTable.size()
1981                << " loaded SLocEntries allocated, "
1982                << MaxLoadedOffset - CurrentLoadedOffset
1983                << "B of Sloc address space used.\n";
1984 
1985   unsigned NumLineNumsComputed = 0;
1986   unsigned NumFileBytesMapped = 0;
1987   for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
1988     NumLineNumsComputed += I->second->SourceLineCache != 0;
1989     NumFileBytesMapped  += I->second->getSizeBytesMapped();
1990   }
1991   unsigned NumMacroArgsComputed = MacroArgsCacheMap.size();
1992 
1993   llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
1994                << NumLineNumsComputed << " files with line #'s computed, "
1995                << NumMacroArgsComputed << " files with macro args computed.\n";
1996   llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
1997                << NumBinaryProbes << " binary.\n";
1998 }
1999 
~ExternalSLocEntrySource()2000 ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
2001 
2002 /// Return the amount of memory used by memory buffers, breaking down
2003 /// by heap-backed versus mmap'ed memory.
getMemoryBufferSizes() const2004 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const {
2005   size_t malloc_bytes = 0;
2006   size_t mmap_bytes = 0;
2007 
2008   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i)
2009     if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped())
2010       switch (MemBufferInfos[i]->getMemoryBufferKind()) {
2011         case llvm::MemoryBuffer::MemoryBuffer_MMap:
2012           mmap_bytes += sized_mapped;
2013           break;
2014         case llvm::MemoryBuffer::MemoryBuffer_Malloc:
2015           malloc_bytes += sized_mapped;
2016           break;
2017       }
2018 
2019   return MemoryBufferSizes(malloc_bytes, mmap_bytes);
2020 }
2021 
getDataStructureSizes() const2022 size_t SourceManager::getDataStructureSizes() const {
2023   size_t size = llvm::capacity_in_bytes(MemBufferInfos)
2024     + llvm::capacity_in_bytes(LocalSLocEntryTable)
2025     + llvm::capacity_in_bytes(LoadedSLocEntryTable)
2026     + llvm::capacity_in_bytes(SLocEntryLoaded)
2027     + llvm::capacity_in_bytes(FileInfos);
2028 
2029   if (OverriddenFilesInfo)
2030     size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles);
2031 
2032   return size;
2033 }
2034