xref: /llvm-project/clang/lib/Basic/SourceManager.cpp (revision fb24a3a4ec3d625981bb822e021dc89a0cf3fde1)
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/Support/Compiler.h"
19 #include "llvm/Support/MemoryBuffer.h"
20 #include "llvm/Support/raw_ostream.h"
21 #include "llvm/System/Path.h"
22 #include <algorithm>
23 #include <string>
24 #include <cstring>
25 
26 using namespace clang;
27 using namespace SrcMgr;
28 using llvm::MemoryBuffer;
29 
30 //===----------------------------------------------------------------------===//
31 // SourceManager Helper Classes
32 //===----------------------------------------------------------------------===//
33 
34 ContentCache::~ContentCache() {
35   delete Buffer.getPointer();
36 }
37 
38 /// getSizeBytesMapped - Returns the number of bytes actually mapped for
39 ///  this ContentCache.  This can be 0 if the MemBuffer was not actually
40 ///  instantiated.
41 unsigned ContentCache::getSizeBytesMapped() const {
42   return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0;
43 }
44 
45 /// getSize - Returns the size of the content encapsulated by this ContentCache.
46 ///  This can be the size of the source file or the size of an arbitrary
47 ///  scratch buffer.  If the ContentCache encapsulates a source file, that
48 ///  file is not lazily brought in from disk to satisfy this query.
49 unsigned ContentCache::getSize() const {
50   return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize()
51                              : (unsigned) Entry->getSize();
52 }
53 
54 void ContentCache::replaceBuffer(const llvm::MemoryBuffer *B) {
55   assert(B != Buffer.getPointer());
56 
57   delete Buffer.getPointer();
58   Buffer.setPointer(B);
59   Buffer.setInt(false);
60 }
61 
62 const llvm::MemoryBuffer *ContentCache::getBuffer(Diagnostic &Diag,
63                                                   const SourceManager &sm,
64                                                   SourceLocation Loc,
65                                                   bool *Invalid) const {
66   if (Invalid)
67     *Invalid = false;
68 
69   // Lazily create the Buffer for ContentCaches that wrap files.
70   if (!Buffer.getPointer() && Entry) {
71     // FIXME:
72     SourceManager &SM = const_cast<SourceManager &>(sm);
73 
74     std::string ErrorStr;
75     struct stat FileInfo;
76     Buffer.setPointer(MemoryBuffer::getFile(Entry->getName(), &ErrorStr,
77                                             Entry->getSize(), &FileInfo));
78     Buffer.setInt(false);
79 
80     // If we were unable to open the file, then we are in an inconsistent
81     // situation where the content cache referenced a file which no longer
82     // exists. Most likely, we were using a stat cache with an invalid entry but
83     // the file could also have been removed during processing. Since we can't
84     // really deal with this situation, just create an empty buffer.
85     //
86     // FIXME: This is definitely not ideal, but our immediate clients can't
87     // currently handle returning a null entry here. Ideally we should detect
88     // that we are in an inconsistent situation and error out as quickly as
89     // possible.
90     if (!Buffer.getPointer()) {
91       const llvm::StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
92       Buffer.setPointer(MemoryBuffer::getNewMemBuffer(Entry->getSize(),
93                                                       "<invalid>"));
94       char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
95       for (unsigned i = 0, e = Entry->getSize(); i != e; ++i)
96         Ptr[i] = FillStr[i % FillStr.size()];
97 
98       if (Diag.isDiagnosticInFlight())
99         Diag.SetDelayedDiagnostic(diag::err_cannot_open_file,
100                                   Entry->getName(), ErrorStr);
101       else
102         Diag.Report(FullSourceLoc(Loc, SM), diag::err_cannot_open_file)
103           << Entry->getName() << ErrorStr;
104 
105       Buffer.setInt(true);
106 
107     // FIXME: This conditionalization is horrible, but we see spurious failures
108     // in the test suite due to this warning and no one has had time to hunt it
109     // down. So for now, we just don't emit this diagnostic on Win32, and hope
110     // nothing bad happens.
111     //
112     // PR6812.
113 #if !defined(LLVM_ON_WIN32)
114     } else if (FileInfo.st_size != Entry->getSize() ||
115                FileInfo.st_mtime != Entry->getModificationTime()) {
116       // Check that the file's size and modification time are the same
117       // as in the file entry (which may have come from a stat cache).
118       if (Diag.isDiagnosticInFlight())
119         Diag.SetDelayedDiagnostic(diag::err_file_modified,
120                                   Entry->getName());
121       else
122         Diag.Report(FullSourceLoc(Loc, SM), diag::err_file_modified)
123           << Entry->getName();
124 
125       Buffer.setInt(true);
126 #endif
127     }
128 
129     // If the buffer is valid, check to see if it has a UTF Byte Order Mark
130     // (BOM).  We only support UTF-8 without a BOM right now.  See
131     // http://en.wikipedia.org/wiki/Byte_order_mark for more information.
132     if (!Buffer.getInt()) {
133       llvm::StringRef BufStr = Buffer.getPointer()->getBuffer();
134       const char *BOM = 0;
135       if (BufStr.startswith("\xFE\xBB\xBF"))
136         BOM = "UTF-8";
137       else if (BufStr.startswith("\xFE\xFF"))
138         BOM = "UTF-16 (BE)";
139       else if (BufStr.startswith("\xFF\xFE"))
140         BOM = "UTF-16 (LE)";
141       else if (BufStr.startswith(llvm::StringRef("\x00\x00\xFE\xFF", 4)))
142         BOM = "UTF-32 (BE)";
143       else if (BufStr.startswith(llvm::StringRef("\xFF\xFE\x00\x00", 4)))
144         BOM = "UTF-32 (LE)";
145       else if (BufStr.startswith("\x2B\x2F\x76"))
146         BOM = "UTF-7";
147       else if (BufStr.startswith("\xF7\x64\x4C"))
148         BOM = "UTF-1";
149       else if (BufStr.startswith("\xDD\x73\x66\x73"))
150         BOM = "UTF-EBCDIC";
151       else if (BufStr.startswith("\x0E\xFE\xFF"))
152         BOM = "SDSU";
153       else if (BufStr.startswith("\xFB\xEE\x28"))
154         BOM = "BOCU-1";
155       else if (BufStr.startswith("\x84\x31\x95\x33"))
156         BOM = "BOCU-1";
157 
158       if (BOM) {
159         Diag.Report(FullSourceLoc(Loc, SM), diag::err_unsupported_bom)
160           << BOM << Entry->getName();
161         Buffer.setInt(1);
162       }
163     }
164   }
165 
166   if (Invalid)
167     *Invalid = Buffer.getInt();
168 
169   return Buffer.getPointer();
170 }
171 
172 unsigned LineTableInfo::getLineTableFilenameID(const char *Ptr, unsigned Len) {
173   // Look up the filename in the string table, returning the pre-existing value
174   // if it exists.
175   llvm::StringMapEntry<unsigned> &Entry =
176     FilenameIDs.GetOrCreateValue(Ptr, Ptr+Len, ~0U);
177   if (Entry.getValue() != ~0U)
178     return Entry.getValue();
179 
180   // Otherwise, assign this the next available ID.
181   Entry.setValue(FilenamesByID.size());
182   FilenamesByID.push_back(&Entry);
183   return FilenamesByID.size()-1;
184 }
185 
186 /// AddLineNote - Add a line note to the line table that indicates that there
187 /// is a #line at the specified FID/Offset location which changes the presumed
188 /// location to LineNo/FilenameID.
189 void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset,
190                                 unsigned LineNo, int FilenameID) {
191   std::vector<LineEntry> &Entries = LineEntries[FID];
192 
193   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
194          "Adding line entries out of order!");
195 
196   SrcMgr::CharacteristicKind Kind = SrcMgr::C_User;
197   unsigned IncludeOffset = 0;
198 
199   if (!Entries.empty()) {
200     // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember
201     // that we are still in "foo.h".
202     if (FilenameID == -1)
203       FilenameID = Entries.back().FilenameID;
204 
205     // If we are after a line marker that switched us to system header mode, or
206     // that set #include information, preserve it.
207     Kind = Entries.back().FileKind;
208     IncludeOffset = Entries.back().IncludeOffset;
209   }
210 
211   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind,
212                                    IncludeOffset));
213 }
214 
215 /// AddLineNote This is the same as the previous version of AddLineNote, but is
216 /// used for GNU line markers.  If EntryExit is 0, then this doesn't change the
217 /// presumed #include stack.  If it is 1, this is a file entry, if it is 2 then
218 /// this is a file exit.  FileKind specifies whether this is a system header or
219 /// extern C system header.
220 void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset,
221                                 unsigned LineNo, int FilenameID,
222                                 unsigned EntryExit,
223                                 SrcMgr::CharacteristicKind FileKind) {
224   assert(FilenameID != -1 && "Unspecified filename should use other accessor");
225 
226   std::vector<LineEntry> &Entries = LineEntries[FID];
227 
228   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
229          "Adding line entries out of order!");
230 
231   unsigned IncludeOffset = 0;
232   if (EntryExit == 0) {  // No #include stack change.
233     IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset;
234   } else if (EntryExit == 1) {
235     IncludeOffset = Offset-1;
236   } else if (EntryExit == 2) {
237     assert(!Entries.empty() && Entries.back().IncludeOffset &&
238        "PPDirectives should have caught case when popping empty include stack");
239 
240     // Get the include loc of the last entries' include loc as our include loc.
241     IncludeOffset = 0;
242     if (const LineEntry *PrevEntry =
243           FindNearestLineEntry(FID, Entries.back().IncludeOffset))
244       IncludeOffset = PrevEntry->IncludeOffset;
245   }
246 
247   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind,
248                                    IncludeOffset));
249 }
250 
251 
252 /// FindNearestLineEntry - Find the line entry nearest to FID that is before
253 /// it.  If there is no line entry before Offset in FID, return null.
254 const LineEntry *LineTableInfo::FindNearestLineEntry(unsigned FID,
255                                                      unsigned Offset) {
256   const std::vector<LineEntry> &Entries = LineEntries[FID];
257   assert(!Entries.empty() && "No #line entries for this FID after all!");
258 
259   // It is very common for the query to be after the last #line, check this
260   // first.
261   if (Entries.back().FileOffset <= Offset)
262     return &Entries.back();
263 
264   // Do a binary search to find the maximal element that is still before Offset.
265   std::vector<LineEntry>::const_iterator I =
266     std::upper_bound(Entries.begin(), Entries.end(), Offset);
267   if (I == Entries.begin()) return 0;
268   return &*--I;
269 }
270 
271 /// \brief Add a new line entry that has already been encoded into
272 /// the internal representation of the line table.
273 void LineTableInfo::AddEntry(unsigned FID,
274                              const std::vector<LineEntry> &Entries) {
275   LineEntries[FID] = Entries;
276 }
277 
278 /// getLineTableFilenameID - Return the uniqued ID for the specified filename.
279 ///
280 unsigned SourceManager::getLineTableFilenameID(const char *Ptr, unsigned Len) {
281   if (LineTable == 0)
282     LineTable = new LineTableInfo();
283   return LineTable->getLineTableFilenameID(Ptr, Len);
284 }
285 
286 
287 /// AddLineNote - Add a line note to the line table for the FileID and offset
288 /// specified by Loc.  If FilenameID is -1, it is considered to be
289 /// unspecified.
290 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
291                                 int FilenameID) {
292   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
293 
294   const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile();
295 
296   // Remember that this file has #line directives now if it doesn't already.
297   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
298 
299   if (LineTable == 0)
300     LineTable = new LineTableInfo();
301   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID);
302 }
303 
304 /// AddLineNote - Add a GNU line marker to the line table.
305 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
306                                 int FilenameID, bool IsFileEntry,
307                                 bool IsFileExit, bool IsSystemHeader,
308                                 bool IsExternCHeader) {
309   // If there is no filename and no flags, this is treated just like a #line,
310   // which does not change the flags of the previous line marker.
311   if (FilenameID == -1) {
312     assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader &&
313            "Can't set flags without setting the filename!");
314     return AddLineNote(Loc, LineNo, FilenameID);
315   }
316 
317   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
318   const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile();
319 
320   // Remember that this file has #line directives now if it doesn't already.
321   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
322 
323   if (LineTable == 0)
324     LineTable = new LineTableInfo();
325 
326   SrcMgr::CharacteristicKind FileKind;
327   if (IsExternCHeader)
328     FileKind = SrcMgr::C_ExternCSystem;
329   else if (IsSystemHeader)
330     FileKind = SrcMgr::C_System;
331   else
332     FileKind = SrcMgr::C_User;
333 
334   unsigned EntryExit = 0;
335   if (IsFileEntry)
336     EntryExit = 1;
337   else if (IsFileExit)
338     EntryExit = 2;
339 
340   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID,
341                          EntryExit, FileKind);
342 }
343 
344 LineTableInfo &SourceManager::getLineTable() {
345   if (LineTable == 0)
346     LineTable = new LineTableInfo();
347   return *LineTable;
348 }
349 
350 //===----------------------------------------------------------------------===//
351 // Private 'Create' methods.
352 //===----------------------------------------------------------------------===//
353 
354 SourceManager::~SourceManager() {
355   delete LineTable;
356 
357   // Delete FileEntry objects corresponding to content caches.  Since the actual
358   // content cache objects are bump pointer allocated, we just have to run the
359   // dtors, but we call the deallocate method for completeness.
360   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) {
361     MemBufferInfos[i]->~ContentCache();
362     ContentCacheAlloc.Deallocate(MemBufferInfos[i]);
363   }
364   for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator
365        I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) {
366     I->second->~ContentCache();
367     ContentCacheAlloc.Deallocate(I->second);
368   }
369 }
370 
371 void SourceManager::clearIDTables() {
372   MainFileID = FileID();
373   SLocEntryTable.clear();
374   LastLineNoFileIDQuery = FileID();
375   LastLineNoContentCache = 0;
376   LastFileIDLookup = FileID();
377 
378   if (LineTable)
379     LineTable->clear();
380 
381   // Use up FileID #0 as an invalid instantiation.
382   NextOffset = 0;
383   createInstantiationLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1);
384 }
385 
386 /// getOrCreateContentCache - Create or return a cached ContentCache for the
387 /// specified file.
388 const ContentCache *
389 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt) {
390   assert(FileEnt && "Didn't specify a file entry to use?");
391 
392   // Do we already have information about this file?
393   ContentCache *&Entry = FileInfos[FileEnt];
394   if (Entry) return Entry;
395 
396   // Nope, create a new Cache entry.  Make sure it is at least 8-byte aligned
397   // so that FileInfo can use the low 3 bits of the pointer for its own
398   // nefarious purposes.
399   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
400   EntryAlign = std::max(8U, EntryAlign);
401   Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
402   new (Entry) ContentCache(FileEnt);
403   return Entry;
404 }
405 
406 
407 /// createMemBufferContentCache - Create a new ContentCache for the specified
408 ///  memory buffer.  This does no caching.
409 const ContentCache*
410 SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) {
411   // Add a new ContentCache to the MemBufferInfos list and return it.  Make sure
412   // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of
413   // the pointer for its own nefarious purposes.
414   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
415   EntryAlign = std::max(8U, EntryAlign);
416   ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
417   new (Entry) ContentCache();
418   MemBufferInfos.push_back(Entry);
419   Entry->setBuffer(Buffer);
420   return Entry;
421 }
422 
423 void SourceManager::PreallocateSLocEntries(ExternalSLocEntrySource *Source,
424                                            unsigned NumSLocEntries,
425                                            unsigned NextOffset) {
426   ExternalSLocEntries = Source;
427   this->NextOffset = NextOffset;
428   SLocEntryLoaded.resize(NumSLocEntries + 1);
429   SLocEntryLoaded[0] = true;
430   SLocEntryTable.resize(SLocEntryTable.size() + NumSLocEntries);
431 }
432 
433 void SourceManager::ClearPreallocatedSLocEntries() {
434   unsigned I = 0;
435   for (unsigned N = SLocEntryLoaded.size(); I != N; ++I)
436     if (!SLocEntryLoaded[I])
437       break;
438 
439   // We've already loaded all preallocated source location entries.
440   if (I == SLocEntryLoaded.size())
441     return;
442 
443   // Remove everything from location I onward.
444   SLocEntryTable.resize(I);
445   SLocEntryLoaded.clear();
446   ExternalSLocEntries = 0;
447 }
448 
449 
450 //===----------------------------------------------------------------------===//
451 // Methods to create new FileID's and instantiations.
452 //===----------------------------------------------------------------------===//
453 
454 /// createFileID - Create a new fileID for the specified ContentCache and
455 /// include position.  This works regardless of whether the ContentCache
456 /// corresponds to a file or some other input source.
457 FileID SourceManager::createFileID(const ContentCache *File,
458                                    SourceLocation IncludePos,
459                                    SrcMgr::CharacteristicKind FileCharacter,
460                                    unsigned PreallocatedID,
461                                    unsigned Offset) {
462   if (PreallocatedID) {
463     // If we're filling in a preallocated ID, just load in the file
464     // entry and return.
465     assert(PreallocatedID < SLocEntryLoaded.size() &&
466            "Preallocate ID out-of-range");
467     assert(!SLocEntryLoaded[PreallocatedID] &&
468            "Source location entry already loaded");
469     assert(Offset && "Preallocate source location cannot have zero offset");
470     SLocEntryTable[PreallocatedID]
471       = SLocEntry::get(Offset, FileInfo::get(IncludePos, File, FileCharacter));
472     SLocEntryLoaded[PreallocatedID] = true;
473     FileID FID = FileID::get(PreallocatedID);
474     return FID;
475   }
476 
477   SLocEntryTable.push_back(SLocEntry::get(NextOffset,
478                                           FileInfo::get(IncludePos, File,
479                                                         FileCharacter)));
480   unsigned FileSize = File->getSize();
481   assert(NextOffset+FileSize+1 > NextOffset && "Ran out of source locations!");
482   NextOffset += FileSize+1;
483 
484   // Set LastFileIDLookup to the newly created file.  The next getFileID call is
485   // almost guaranteed to be from that file.
486   FileID FID = FileID::get(SLocEntryTable.size()-1);
487   return LastFileIDLookup = FID;
488 }
489 
490 /// createInstantiationLoc - Return a new SourceLocation that encodes the fact
491 /// that a token from SpellingLoc should actually be referenced from
492 /// InstantiationLoc.
493 SourceLocation SourceManager::createInstantiationLoc(SourceLocation SpellingLoc,
494                                                      SourceLocation ILocStart,
495                                                      SourceLocation ILocEnd,
496                                                      unsigned TokLength,
497                                                      unsigned PreallocatedID,
498                                                      unsigned Offset) {
499   InstantiationInfo II = InstantiationInfo::get(ILocStart,ILocEnd, SpellingLoc);
500   if (PreallocatedID) {
501     // If we're filling in a preallocated ID, just load in the
502     // instantiation entry and return.
503     assert(PreallocatedID < SLocEntryLoaded.size() &&
504            "Preallocate ID out-of-range");
505     assert(!SLocEntryLoaded[PreallocatedID] &&
506            "Source location entry already loaded");
507     assert(Offset && "Preallocate source location cannot have zero offset");
508     SLocEntryTable[PreallocatedID] = SLocEntry::get(Offset, II);
509     SLocEntryLoaded[PreallocatedID] = true;
510     return SourceLocation::getMacroLoc(Offset);
511   }
512   SLocEntryTable.push_back(SLocEntry::get(NextOffset, II));
513   assert(NextOffset+TokLength+1 > NextOffset && "Ran out of source locations!");
514   NextOffset += TokLength+1;
515   return SourceLocation::getMacroLoc(NextOffset-(TokLength+1));
516 }
517 
518 const llvm::MemoryBuffer *
519 SourceManager::getMemoryBufferForFile(const FileEntry *File,
520                                       bool *Invalid) {
521   const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
522   assert(IR && "getOrCreateContentCache() cannot return NULL");
523   return IR->getBuffer(Diag, *this, SourceLocation(), Invalid);
524 }
525 
526 bool SourceManager::overrideFileContents(const FileEntry *SourceFile,
527                                          const llvm::MemoryBuffer *Buffer) {
528   const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile);
529   if (IR == 0)
530     return true;
531 
532   const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer);
533   return false;
534 }
535 
536 llvm::StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const {
537   bool MyInvalid = false;
538   const llvm::MemoryBuffer *Buf = getBuffer(FID, &MyInvalid);
539   if (Invalid)
540     *Invalid = MyInvalid;
541 
542   if (MyInvalid)
543     return "";
544 
545   return Buf->getBuffer();
546 }
547 
548 //===----------------------------------------------------------------------===//
549 // SourceLocation manipulation methods.
550 //===----------------------------------------------------------------------===//
551 
552 /// getFileIDSlow - Return the FileID for a SourceLocation.  This is a very hot
553 /// method that is used for all SourceManager queries that start with a
554 /// SourceLocation object.  It is responsible for finding the entry in
555 /// SLocEntryTable which contains the specified location.
556 ///
557 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const {
558   assert(SLocOffset && "Invalid FileID");
559 
560   // After the first and second level caches, I see two common sorts of
561   // behavior: 1) a lot of searched FileID's are "near" the cached file location
562   // or are "near" the cached instantiation location.  2) others are just
563   // completely random and may be a very long way away.
564   //
565   // To handle this, we do a linear search for up to 8 steps to catch #1 quickly
566   // then we fall back to a less cache efficient, but more scalable, binary
567   // search to find the location.
568 
569   // See if this is near the file point - worst case we start scanning from the
570   // most newly created FileID.
571   std::vector<SrcMgr::SLocEntry>::const_iterator I;
572 
573   if (SLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) {
574     // Neither loc prunes our search.
575     I = SLocEntryTable.end();
576   } else {
577     // Perhaps it is near the file point.
578     I = SLocEntryTable.begin()+LastFileIDLookup.ID;
579   }
580 
581   // Find the FileID that contains this.  "I" is an iterator that points to a
582   // FileID whose offset is known to be larger than SLocOffset.
583   unsigned NumProbes = 0;
584   while (1) {
585     --I;
586     if (ExternalSLocEntries)
587       getSLocEntry(FileID::get(I - SLocEntryTable.begin()));
588     if (I->getOffset() <= SLocOffset) {
589 #if 0
590       printf("lin %d -> %d [%s] %d %d\n", SLocOffset,
591              I-SLocEntryTable.begin(),
592              I->isInstantiation() ? "inst" : "file",
593              LastFileIDLookup.ID,  int(SLocEntryTable.end()-I));
594 #endif
595       FileID Res = FileID::get(I-SLocEntryTable.begin());
596 
597       // If this isn't an instantiation, remember it.  We have good locality
598       // across FileID lookups.
599       if (!I->isInstantiation())
600         LastFileIDLookup = Res;
601       NumLinearScans += NumProbes+1;
602       return Res;
603     }
604     if (++NumProbes == 8)
605       break;
606   }
607 
608   // Convert "I" back into an index.  We know that it is an entry whose index is
609   // larger than the offset we are looking for.
610   unsigned GreaterIndex = I-SLocEntryTable.begin();
611   // LessIndex - This is the lower bound of the range that we're searching.
612   // We know that the offset corresponding to the FileID is is less than
613   // SLocOffset.
614   unsigned LessIndex = 0;
615   NumProbes = 0;
616   while (1) {
617     unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex;
618     unsigned MidOffset = getSLocEntry(FileID::get(MiddleIndex)).getOffset();
619 
620     ++NumProbes;
621 
622     // If the offset of the midpoint is too large, chop the high side of the
623     // range to the midpoint.
624     if (MidOffset > SLocOffset) {
625       GreaterIndex = MiddleIndex;
626       continue;
627     }
628 
629     // If the middle index contains the value, succeed and return.
630     if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) {
631 #if 0
632       printf("bin %d -> %d [%s] %d %d\n", SLocOffset,
633              I-SLocEntryTable.begin(),
634              I->isInstantiation() ? "inst" : "file",
635              LastFileIDLookup.ID, int(SLocEntryTable.end()-I));
636 #endif
637       FileID Res = FileID::get(MiddleIndex);
638 
639       // If this isn't an instantiation, remember it.  We have good locality
640       // across FileID lookups.
641       if (!I->isInstantiation())
642         LastFileIDLookup = Res;
643       NumBinaryProbes += NumProbes;
644       return Res;
645     }
646 
647     // Otherwise, move the low-side up to the middle index.
648     LessIndex = MiddleIndex;
649   }
650 }
651 
652 SourceLocation SourceManager::
653 getInstantiationLocSlowCase(SourceLocation Loc) const {
654   do {
655     // Note: If Loc indicates an offset into a token that came from a macro
656     // expansion (e.g. the 5th character of the token) we do not want to add
657     // this offset when going to the instantiation location.  The instatiation
658     // location is the macro invocation, which the offset has nothing to do
659     // with.  This is unlike when we get the spelling loc, because the offset
660     // directly correspond to the token whose spelling we're inspecting.
661     Loc = getSLocEntry(getFileID(Loc)).getInstantiation()
662                    .getInstantiationLocStart();
663   } while (!Loc.isFileID());
664 
665   return Loc;
666 }
667 
668 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
669   do {
670     std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
671     Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc();
672     Loc = Loc.getFileLocWithOffset(LocInfo.second);
673   } while (!Loc.isFileID());
674   return Loc;
675 }
676 
677 
678 std::pair<FileID, unsigned>
679 SourceManager::getDecomposedInstantiationLocSlowCase(const SrcMgr::SLocEntry *E,
680                                                      unsigned Offset) const {
681   // If this is an instantiation record, walk through all the instantiation
682   // points.
683   FileID FID;
684   SourceLocation Loc;
685   do {
686     Loc = E->getInstantiation().getInstantiationLocStart();
687 
688     FID = getFileID(Loc);
689     E = &getSLocEntry(FID);
690     Offset += Loc.getOffset()-E->getOffset();
691   } while (!Loc.isFileID());
692 
693   return std::make_pair(FID, Offset);
694 }
695 
696 std::pair<FileID, unsigned>
697 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
698                                                 unsigned Offset) const {
699   // If this is an instantiation record, walk through all the instantiation
700   // points.
701   FileID FID;
702   SourceLocation Loc;
703   do {
704     Loc = E->getInstantiation().getSpellingLoc();
705 
706     FID = getFileID(Loc);
707     E = &getSLocEntry(FID);
708     Offset += Loc.getOffset()-E->getOffset();
709   } while (!Loc.isFileID());
710 
711   return std::make_pair(FID, Offset);
712 }
713 
714 /// getImmediateSpellingLoc - Given a SourceLocation object, return the
715 /// spelling location referenced by the ID.  This is the first level down
716 /// towards the place where the characters that make up the lexed token can be
717 /// found.  This should not generally be used by clients.
718 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
719   if (Loc.isFileID()) return Loc;
720   std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
721   Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc();
722   return Loc.getFileLocWithOffset(LocInfo.second);
723 }
724 
725 
726 /// getImmediateInstantiationRange - Loc is required to be an instantiation
727 /// location.  Return the start/end of the instantiation information.
728 std::pair<SourceLocation,SourceLocation>
729 SourceManager::getImmediateInstantiationRange(SourceLocation Loc) const {
730   assert(Loc.isMacroID() && "Not an instantiation loc!");
731   const InstantiationInfo &II = getSLocEntry(getFileID(Loc)).getInstantiation();
732   return II.getInstantiationLocRange();
733 }
734 
735 /// getInstantiationRange - Given a SourceLocation object, return the
736 /// range of tokens covered by the instantiation in the ultimate file.
737 std::pair<SourceLocation,SourceLocation>
738 SourceManager::getInstantiationRange(SourceLocation Loc) const {
739   if (Loc.isFileID()) return std::make_pair(Loc, Loc);
740 
741   std::pair<SourceLocation,SourceLocation> Res =
742     getImmediateInstantiationRange(Loc);
743 
744   // Fully resolve the start and end locations to their ultimate instantiation
745   // points.
746   while (!Res.first.isFileID())
747     Res.first = getImmediateInstantiationRange(Res.first).first;
748   while (!Res.second.isFileID())
749     Res.second = getImmediateInstantiationRange(Res.second).second;
750   return Res;
751 }
752 
753 
754 
755 //===----------------------------------------------------------------------===//
756 // Queries about the code at a SourceLocation.
757 //===----------------------------------------------------------------------===//
758 
759 /// getCharacterData - Return a pointer to the start of the specified location
760 /// in the appropriate MemoryBuffer.
761 const char *SourceManager::getCharacterData(SourceLocation SL,
762                                             bool *Invalid) const {
763   // Note that this is a hot function in the getSpelling() path, which is
764   // heavily used by -E mode.
765   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
766 
767   // Note that calling 'getBuffer()' may lazily page in a source file.
768   bool CharDataInvalid = false;
769   const llvm::MemoryBuffer *Buffer
770     = getSLocEntry(LocInfo.first).getFile().getContentCache()
771     ->getBuffer(Diag, *this, SourceLocation(), &CharDataInvalid);
772   if (Invalid)
773     *Invalid = CharDataInvalid;
774   return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
775 }
776 
777 
778 /// getColumnNumber - Return the column # for the specified file position.
779 /// this is significantly cheaper to compute than the line number.
780 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
781                                         bool *Invalid) const {
782   bool MyInvalid = false;
783   const char *Buf = getBuffer(FID, &MyInvalid)->getBufferStart();
784   if (Invalid)
785     *Invalid = MyInvalid;
786 
787   if (MyInvalid)
788     return 1;
789 
790   unsigned LineStart = FilePos;
791   while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
792     --LineStart;
793   return FilePos-LineStart+1;
794 }
795 
796 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
797                                                 bool *Invalid) const {
798   if (Loc.isInvalid()) return 0;
799   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
800   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
801 }
802 
803 unsigned SourceManager::getInstantiationColumnNumber(SourceLocation Loc,
804                                                      bool *Invalid) const {
805   if (Loc.isInvalid()) return 0;
806   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
807   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
808 }
809 
810 static DISABLE_INLINE void
811 ComputeLineNumbers(Diagnostic &Diag, ContentCache *FI,
812                    llvm::BumpPtrAllocator &Alloc,
813                    const SourceManager &SM, bool &Invalid);
814 static void ComputeLineNumbers(Diagnostic &Diag, ContentCache *FI,
815                                llvm::BumpPtrAllocator &Alloc,
816                                const SourceManager &SM, bool &Invalid) {
817   // Note that calling 'getBuffer()' may lazily page in the file.
818   const MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(),
819                                              &Invalid);
820   if (Invalid)
821     return;
822 
823   // Find the file offsets of all of the *physical* source lines.  This does
824   // not look at trigraphs, escaped newlines, or anything else tricky.
825   std::vector<unsigned> LineOffsets;
826 
827   // Line #1 starts at char 0.
828   LineOffsets.push_back(0);
829 
830   const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
831   const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
832   unsigned Offs = 0;
833   while (1) {
834     // Skip over the contents of the line.
835     // TODO: Vectorize this?  This is very performance sensitive for programs
836     // with lots of diagnostics and in -E mode.
837     const unsigned char *NextBuf = (const unsigned char *)Buf;
838     while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
839       ++NextBuf;
840     Offs += NextBuf-Buf;
841     Buf = NextBuf;
842 
843     if (Buf[0] == '\n' || Buf[0] == '\r') {
844       // If this is \n\r or \r\n, skip both characters.
845       if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1])
846         ++Offs, ++Buf;
847       ++Offs, ++Buf;
848       LineOffsets.push_back(Offs);
849     } else {
850       // Otherwise, this is a null.  If end of file, exit.
851       if (Buf == End) break;
852       // Otherwise, skip the null.
853       ++Offs, ++Buf;
854     }
855   }
856 
857   // Copy the offsets into the FileInfo structure.
858   FI->NumLines = LineOffsets.size();
859   FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
860   std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
861 }
862 
863 /// getLineNumber - Given a SourceLocation, return the spelling line number
864 /// for the position indicated.  This requires building and caching a table of
865 /// line offsets for the MemoryBuffer, so this is not cheap: use only when
866 /// about to emit a diagnostic.
867 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
868                                       bool *Invalid) const {
869   ContentCache *Content;
870   if (LastLineNoFileIDQuery == FID)
871     Content = LastLineNoContentCache;
872   else
873     Content = const_cast<ContentCache*>(getSLocEntry(FID)
874                                         .getFile().getContentCache());
875 
876   // If this is the first use of line information for this buffer, compute the
877   /// SourceLineCache for it on demand.
878   if (Content->SourceLineCache == 0) {
879     bool MyInvalid = false;
880     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
881     if (Invalid)
882       *Invalid = MyInvalid;
883     if (MyInvalid)
884       return 1;
885   } else if (Invalid)
886     *Invalid = false;
887 
888   // Okay, we know we have a line number table.  Do a binary search to find the
889   // line number that this character position lands on.
890   unsigned *SourceLineCache = Content->SourceLineCache;
891   unsigned *SourceLineCacheStart = SourceLineCache;
892   unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
893 
894   unsigned QueriedFilePos = FilePos+1;
895 
896   // FIXME: I would like to be convinced that this code is worth being as
897   // complicated as it is, binary search isn't that slow.
898   //
899   // If it is worth being optimized, then in my opinion it could be more
900   // performant, simpler, and more obviously correct by just "galloping" outward
901   // from the queried file position. In fact, this could be incorporated into a
902   // generic algorithm such as lower_bound_with_hint.
903   //
904   // If someone gives me a test case where this matters, and I will do it! - DWD
905 
906   // If the previous query was to the same file, we know both the file pos from
907   // that query and the line number returned.  This allows us to narrow the
908   // search space from the entire file to something near the match.
909   if (LastLineNoFileIDQuery == FID) {
910     if (QueriedFilePos >= LastLineNoFilePos) {
911       // FIXME: Potential overflow?
912       SourceLineCache = SourceLineCache+LastLineNoResult-1;
913 
914       // The query is likely to be nearby the previous one.  Here we check to
915       // see if it is within 5, 10 or 20 lines.  It can be far away in cases
916       // where big comment blocks and vertical whitespace eat up lines but
917       // contribute no tokens.
918       if (SourceLineCache+5 < SourceLineCacheEnd) {
919         if (SourceLineCache[5] > QueriedFilePos)
920           SourceLineCacheEnd = SourceLineCache+5;
921         else if (SourceLineCache+10 < SourceLineCacheEnd) {
922           if (SourceLineCache[10] > QueriedFilePos)
923             SourceLineCacheEnd = SourceLineCache+10;
924           else if (SourceLineCache+20 < SourceLineCacheEnd) {
925             if (SourceLineCache[20] > QueriedFilePos)
926               SourceLineCacheEnd = SourceLineCache+20;
927           }
928         }
929       }
930     } else {
931       if (LastLineNoResult < Content->NumLines)
932         SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
933     }
934   }
935 
936   // If the spread is large, do a "radix" test as our initial guess, based on
937   // the assumption that lines average to approximately the same length.
938   // NOTE: This is currently disabled, as it does not appear to be profitable in
939   // initial measurements.
940   if (0 && SourceLineCacheEnd-SourceLineCache > 20) {
941     unsigned FileLen = Content->SourceLineCache[Content->NumLines-1];
942 
943     // Take a stab at guessing where it is.
944     unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen;
945 
946     // Check for -10 and +10 lines.
947     unsigned LowerBound = std::max(int(ApproxPos-10), 0);
948     unsigned UpperBound = std::min(ApproxPos+10, FileLen);
949 
950     // If the computed lower bound is less than the query location, move it in.
951     if (SourceLineCache < SourceLineCacheStart+LowerBound &&
952         SourceLineCacheStart[LowerBound] < QueriedFilePos)
953       SourceLineCache = SourceLineCacheStart+LowerBound;
954 
955     // If the computed upper bound is greater than the query location, move it.
956     if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound &&
957         SourceLineCacheStart[UpperBound] >= QueriedFilePos)
958       SourceLineCacheEnd = SourceLineCacheStart+UpperBound;
959   }
960 
961   unsigned *Pos
962     = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
963   unsigned LineNo = Pos-SourceLineCacheStart;
964 
965   LastLineNoFileIDQuery = FID;
966   LastLineNoContentCache = Content;
967   LastLineNoFilePos = QueriedFilePos;
968   LastLineNoResult = LineNo;
969   return LineNo;
970 }
971 
972 unsigned SourceManager::getInstantiationLineNumber(SourceLocation Loc,
973                                                    bool *Invalid) const {
974   if (Loc.isInvalid()) return 0;
975   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
976   return getLineNumber(LocInfo.first, LocInfo.second);
977 }
978 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
979                                               bool *Invalid) const {
980   if (Loc.isInvalid()) return 0;
981   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
982   return getLineNumber(LocInfo.first, LocInfo.second);
983 }
984 
985 /// getFileCharacteristic - return the file characteristic of the specified
986 /// source location, indicating whether this is a normal file, a system
987 /// header, or an "implicit extern C" system header.
988 ///
989 /// This state can be modified with flags on GNU linemarker directives like:
990 ///   # 4 "foo.h" 3
991 /// which changes all source locations in the current file after that to be
992 /// considered to be from a system header.
993 SrcMgr::CharacteristicKind
994 SourceManager::getFileCharacteristic(SourceLocation Loc) const {
995   assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!");
996   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
997   const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile();
998 
999   // If there are no #line directives in this file, just return the whole-file
1000   // state.
1001   if (!FI.hasLineDirectives())
1002     return FI.getFileCharacteristic();
1003 
1004   assert(LineTable && "Can't have linetable entries without a LineTable!");
1005   // See if there is a #line directive before the location.
1006   const LineEntry *Entry =
1007     LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second);
1008 
1009   // If this is before the first line marker, use the file characteristic.
1010   if (!Entry)
1011     return FI.getFileCharacteristic();
1012 
1013   return Entry->FileKind;
1014 }
1015 
1016 /// Return the filename or buffer identifier of the buffer the location is in.
1017 /// Note that this name does not respect #line directives.  Use getPresumedLoc
1018 /// for normal clients.
1019 const char *SourceManager::getBufferName(SourceLocation Loc,
1020                                          bool *Invalid) const {
1021   if (Loc.isInvalid()) return "<invalid loc>";
1022 
1023   return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
1024 }
1025 
1026 
1027 /// getPresumedLoc - This method returns the "presumed" location of a
1028 /// SourceLocation specifies.  A "presumed location" can be modified by #line
1029 /// or GNU line marker directives.  This provides a view on the data that a
1030 /// user should see in diagnostics, for example.
1031 ///
1032 /// Note that a presumed location is always given as the instantiation point
1033 /// of an instantiation location, not at the spelling location.
1034 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc) const {
1035   if (Loc.isInvalid()) return PresumedLoc();
1036 
1037   // Presumed locations are always for instantiation points.
1038   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
1039 
1040   const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile();
1041   const SrcMgr::ContentCache *C = FI.getContentCache();
1042 
1043   // To get the source name, first consult the FileEntry (if one exists)
1044   // before the MemBuffer as this will avoid unnecessarily paging in the
1045   // MemBuffer.
1046   const char *Filename;
1047   if (C->Entry)
1048     Filename = C->Entry->getName();
1049   else
1050     Filename = C->getBuffer(Diag, *this)->getBufferIdentifier();
1051   unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second);
1052   unsigned ColNo  = getColumnNumber(LocInfo.first, LocInfo.second);
1053   SourceLocation IncludeLoc = FI.getIncludeLoc();
1054 
1055   // If we have #line directives in this file, update and overwrite the physical
1056   // location info if appropriate.
1057   if (FI.hasLineDirectives()) {
1058     assert(LineTable && "Can't have linetable entries without a LineTable!");
1059     // See if there is a #line directive before this.  If so, get it.
1060     if (const LineEntry *Entry =
1061           LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second)) {
1062       // If the LineEntry indicates a filename, use it.
1063       if (Entry->FilenameID != -1)
1064         Filename = LineTable->getFilename(Entry->FilenameID);
1065 
1066       // Use the line number specified by the LineEntry.  This line number may
1067       // be multiple lines down from the line entry.  Add the difference in
1068       // physical line numbers from the query point and the line marker to the
1069       // total.
1070       unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
1071       LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
1072 
1073       // Note that column numbers are not molested by line markers.
1074 
1075       // Handle virtual #include manipulation.
1076       if (Entry->IncludeOffset) {
1077         IncludeLoc = getLocForStartOfFile(LocInfo.first);
1078         IncludeLoc = IncludeLoc.getFileLocWithOffset(Entry->IncludeOffset);
1079       }
1080     }
1081   }
1082 
1083   return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc);
1084 }
1085 
1086 //===----------------------------------------------------------------------===//
1087 // Other miscellaneous methods.
1088 //===----------------------------------------------------------------------===//
1089 
1090 /// \brief Get the source location for the given file:line:col triplet.
1091 ///
1092 /// If the source file is included multiple times, the source location will
1093 /// be based upon the first inclusion.
1094 SourceLocation SourceManager::getLocation(const FileEntry *SourceFile,
1095                                           unsigned Line, unsigned Col) const {
1096   assert(SourceFile && "Null source file!");
1097   assert(Line && Col && "Line and column should start from 1!");
1098 
1099   fileinfo_iterator FI = FileInfos.find(SourceFile);
1100   if (FI == FileInfos.end())
1101     return SourceLocation();
1102   ContentCache *Content = FI->second;
1103 
1104   // If this is the first use of line information for this buffer, compute the
1105   /// SourceLineCache for it on demand.
1106   if (Content->SourceLineCache == 0) {
1107     bool MyInvalid = false;
1108     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1109     if (MyInvalid)
1110       return SourceLocation();
1111   }
1112 
1113   // Find the first file ID that corresponds to the given file.
1114   FileID FirstFID;
1115 
1116   // First, check the main file ID, since it is common to look for a
1117   // location in the main file.
1118   if (!MainFileID.isInvalid()) {
1119     const SLocEntry &MainSLoc = getSLocEntry(MainFileID);
1120     if (MainSLoc.isFile() && MainSLoc.getFile().getContentCache() == Content)
1121       FirstFID = MainFileID;
1122   }
1123 
1124   if (FirstFID.isInvalid()) {
1125     // The location we're looking for isn't in the main file; look
1126     // through all of the source locations.
1127     for (unsigned I = 0, N = sloc_entry_size(); I != N; ++I) {
1128       const SLocEntry &SLoc = getSLocEntry(I);
1129       if (SLoc.isFile() && SLoc.getFile().getContentCache() == Content) {
1130         FirstFID = FileID::get(I);
1131         break;
1132       }
1133     }
1134   }
1135 
1136   if (FirstFID.isInvalid())
1137     return SourceLocation();
1138 
1139   if (Line > Content->NumLines) {
1140     unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize();
1141     if (Size > 0)
1142       --Size;
1143     return getLocForStartOfFile(FirstFID).getFileLocWithOffset(Size);
1144   }
1145 
1146   unsigned FilePos = Content->SourceLineCache[Line - 1];
1147   const char *Buf = Content->getBuffer(Diag, *this)->getBufferStart() + FilePos;
1148   unsigned BufLength = Content->getBuffer(Diag, *this)->getBufferEnd() - Buf;
1149   unsigned i = 0;
1150 
1151   // Check that the given column is valid.
1152   while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
1153     ++i;
1154   if (i < Col-1)
1155     return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + i);
1156 
1157   return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + Col - 1);
1158 }
1159 
1160 /// \brief Determines the order of 2 source locations in the translation unit.
1161 ///
1162 /// \returns true if LHS source location comes before RHS, false otherwise.
1163 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
1164                                               SourceLocation RHS) const {
1165   assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
1166   if (LHS == RHS)
1167     return false;
1168 
1169   std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
1170   std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
1171 
1172   // If the source locations are in the same file, just compare offsets.
1173   if (LOffs.first == ROffs.first)
1174     return LOffs.second < ROffs.second;
1175 
1176   // If we are comparing a source location with multiple locations in the same
1177   // file, we get a big win by caching the result.
1178 
1179   if (LastLFIDForBeforeTUCheck == LOffs.first &&
1180       LastRFIDForBeforeTUCheck == ROffs.first)
1181     return LastResForBeforeTUCheck;
1182 
1183   LastLFIDForBeforeTUCheck = LOffs.first;
1184   LastRFIDForBeforeTUCheck = ROffs.first;
1185 
1186   // "Traverse" the include/instantiation stacks of both locations and try to
1187   // find a common "ancestor".
1188   //
1189   // First we traverse the stack of the right location and check each level
1190   // against the level of the left location, while collecting all levels in a
1191   // "stack map".
1192 
1193   std::map<FileID, unsigned> ROffsMap;
1194   ROffsMap[ROffs.first] = ROffs.second;
1195 
1196   while (1) {
1197     SourceLocation UpperLoc;
1198     const SrcMgr::SLocEntry &Entry = getSLocEntry(ROffs.first);
1199     if (Entry.isInstantiation())
1200       UpperLoc = Entry.getInstantiation().getInstantiationLocStart();
1201     else
1202       UpperLoc = Entry.getFile().getIncludeLoc();
1203 
1204     if (UpperLoc.isInvalid())
1205       break; // We reached the top.
1206 
1207     ROffs = getDecomposedLoc(UpperLoc);
1208 
1209     if (LOffs.first == ROffs.first)
1210       return LastResForBeforeTUCheck = LOffs.second < ROffs.second;
1211 
1212     ROffsMap[ROffs.first] = ROffs.second;
1213   }
1214 
1215   // We didn't find a common ancestor. Now traverse the stack of the left
1216   // location, checking against the stack map of the right location.
1217 
1218   while (1) {
1219     SourceLocation UpperLoc;
1220     const SrcMgr::SLocEntry &Entry = getSLocEntry(LOffs.first);
1221     if (Entry.isInstantiation())
1222       UpperLoc = Entry.getInstantiation().getInstantiationLocStart();
1223     else
1224       UpperLoc = Entry.getFile().getIncludeLoc();
1225 
1226     if (UpperLoc.isInvalid())
1227       break; // We reached the top.
1228 
1229     LOffs = getDecomposedLoc(UpperLoc);
1230 
1231     std::map<FileID, unsigned>::iterator I = ROffsMap.find(LOffs.first);
1232     if (I != ROffsMap.end())
1233       return LastResForBeforeTUCheck = LOffs.second < I->second;
1234   }
1235 
1236   // There is no common ancestor, most probably because one location is in the
1237   // predefines buffer.
1238   //
1239   // FIXME: We should rearrange the external interface so this simply never
1240   // happens; it can't conceptually happen. Also see PR5662.
1241 
1242   // If exactly one location is a memory buffer, assume it preceeds the other.
1243   bool LIsMB = !getSLocEntry(LOffs.first).getFile().getContentCache()->Entry;
1244   bool RIsMB = !getSLocEntry(ROffs.first).getFile().getContentCache()->Entry;
1245   if (LIsMB != RIsMB)
1246     return LastResForBeforeTUCheck = LIsMB;
1247 
1248   // Otherwise, just assume FileIDs were created in order.
1249   return LastResForBeforeTUCheck = (LOffs.first < ROffs.first);
1250 }
1251 
1252 /// PrintStats - Print statistics to stderr.
1253 ///
1254 void SourceManager::PrintStats() const {
1255   llvm::errs() << "\n*** Source Manager Stats:\n";
1256   llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
1257                << " mem buffers mapped.\n";
1258   llvm::errs() << SLocEntryTable.size() << " SLocEntry's allocated, "
1259                << NextOffset << "B of Sloc address space used.\n";
1260 
1261   unsigned NumLineNumsComputed = 0;
1262   unsigned NumFileBytesMapped = 0;
1263   for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
1264     NumLineNumsComputed += I->second->SourceLineCache != 0;
1265     NumFileBytesMapped  += I->second->getSizeBytesMapped();
1266   }
1267 
1268   llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
1269                << NumLineNumsComputed << " files with line #'s computed.\n";
1270   llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
1271                << NumBinaryProbes << " binary.\n";
1272 }
1273 
1274 ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
1275