xref: /llvm-project/clang/lib/Basic/SourceManager.cpp (revision 8fbe98b3b6e44ca94c3edd630e39150228cc774e)
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                                                   bool *Invalid) const {
64   if (Invalid)
65     *Invalid = false;
66 
67   // Lazily create the Buffer for ContentCaches that wrap files.
68   if (!Buffer.getPointer() && Entry) {
69     std::string ErrorStr;
70     struct stat FileInfo;
71     Buffer.setPointer(MemoryBuffer::getFile(Entry->getName(), &ErrorStr,
72                                             Entry->getSize(), &FileInfo));
73     Buffer.setInt(false);
74 
75     // If we were unable to open the file, then we are in an inconsistent
76     // situation where the content cache referenced a file which no longer
77     // exists. Most likely, we were using a stat cache with an invalid entry but
78     // the file could also have been removed during processing. Since we can't
79     // really deal with this situation, just create an empty buffer.
80     //
81     // FIXME: This is definitely not ideal, but our immediate clients can't
82     // currently handle returning a null entry here. Ideally we should detect
83     // that we are in an inconsistent situation and error out as quickly as
84     // possible.
85     if (!Buffer.getPointer()) {
86       const llvm::StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
87       Buffer.setPointer(MemoryBuffer::getNewMemBuffer(Entry->getSize(),
88                                                       "<invalid>"));
89       char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
90       for (unsigned i = 0, e = Entry->getSize(); i != e; ++i)
91         Ptr[i] = FillStr[i % FillStr.size()];
92 
93       if (Diag.isDiagnosticInFlight())
94         Diag.SetDelayedDiagnostic(diag::err_cannot_open_file,
95                                   Entry->getName(), ErrorStr);
96       else
97         Diag.Report(diag::err_cannot_open_file)
98           << Entry->getName() << ErrorStr;
99 
100       Buffer.setInt(true);
101 
102     // FIXME: This conditionalization is horrible, but we see spurious failures
103     // in the test suite due to this warning and no one has had time to hunt it
104     // down. So for now, we just don't emit this diagnostic on Win32, and hope
105     // nothing bad happens.
106     //
107     // PR6812.
108 #if !defined(LLVM_ON_WIN32)
109     } else if (FileInfo.st_size != Entry->getSize() ||
110                FileInfo.st_mtime != Entry->getModificationTime()) {
111       // Check that the file's size and modification time are the same
112       // as in the file entry (which may have come from a stat cache).
113       if (Diag.isDiagnosticInFlight())
114         Diag.SetDelayedDiagnostic(diag::err_file_modified,
115                                   Entry->getName());
116       else
117         Diag.Report(diag::err_file_modified) << Entry->getName();
118 
119       Buffer.setInt(true);
120 #endif
121     }
122 
123     // If the buffer is valid, check to see if it has a UTF Byte Order Mark
124     // (BOM).  We only support UTF-8 without a BOM right now.  See
125     // http://en.wikipedia.org/wiki/Byte_order_mark for more information.
126     if (!Buffer.getInt()) {
127       llvm::StringRef BufStr = Buffer.getPointer()->getBuffer();
128       const char *BOM = 0;
129       if (BufStr.startswith("\xFE\xBB\xBF"))
130         BOM = "UTF-8";
131       else if (BufStr.startswith("\xFE\xFF"))
132         BOM = "UTF-16 (BE)";
133       else if (BufStr.startswith("\xFF\xFE"))
134         BOM = "UTF-16 (LE)";
135       else if (BufStr.startswith(llvm::StringRef("\x00\x00\xFE\xFF", 4)))
136         BOM = "UTF-32 (BE)";
137       else if (BufStr.startswith(llvm::StringRef("\xFF\xFE\x00\x00", 4)))
138         BOM = "UTF-32 (LE)";
139       else if (BufStr.startswith("\x2B\x2F\x76"))
140         BOM = "UTF-7";
141       else if (BufStr.startswith("\xF7\x64\x4C"))
142         BOM = "UTF-1";
143       else if (BufStr.startswith("\xDD\x73\x66\x73"))
144         BOM = "UTF-EBCDIC";
145       else if (BufStr.startswith("\x0E\xFE\xFF"))
146         BOM = "SDSU";
147       else if (BufStr.startswith("\xFB\xEE\x28"))
148         BOM = "BOCU-1";
149       else if (BufStr.startswith("\x84\x31\x95\x33"))
150         BOM = "BOCU-1";
151 
152       if (BOM) {
153         Diag.Report(diag::err_unsupported_bom) << BOM << Entry->getName();
154         Buffer.setInt(1);
155       }
156     }
157   }
158 
159   if (Invalid)
160     *Invalid = Buffer.getInt();
161 
162   return Buffer.getPointer();
163 }
164 
165 unsigned LineTableInfo::getLineTableFilenameID(const char *Ptr, unsigned Len) {
166   // Look up the filename in the string table, returning the pre-existing value
167   // if it exists.
168   llvm::StringMapEntry<unsigned> &Entry =
169     FilenameIDs.GetOrCreateValue(Ptr, Ptr+Len, ~0U);
170   if (Entry.getValue() != ~0U)
171     return Entry.getValue();
172 
173   // Otherwise, assign this the next available ID.
174   Entry.setValue(FilenamesByID.size());
175   FilenamesByID.push_back(&Entry);
176   return FilenamesByID.size()-1;
177 }
178 
179 /// AddLineNote - Add a line note to the line table that indicates that there
180 /// is a #line at the specified FID/Offset location which changes the presumed
181 /// location to LineNo/FilenameID.
182 void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset,
183                                 unsigned LineNo, int FilenameID) {
184   std::vector<LineEntry> &Entries = LineEntries[FID];
185 
186   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
187          "Adding line entries out of order!");
188 
189   SrcMgr::CharacteristicKind Kind = SrcMgr::C_User;
190   unsigned IncludeOffset = 0;
191 
192   if (!Entries.empty()) {
193     // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember
194     // that we are still in "foo.h".
195     if (FilenameID == -1)
196       FilenameID = Entries.back().FilenameID;
197 
198     // If we are after a line marker that switched us to system header mode, or
199     // that set #include information, preserve it.
200     Kind = Entries.back().FileKind;
201     IncludeOffset = Entries.back().IncludeOffset;
202   }
203 
204   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind,
205                                    IncludeOffset));
206 }
207 
208 /// AddLineNote This is the same as the previous version of AddLineNote, but is
209 /// used for GNU line markers.  If EntryExit is 0, then this doesn't change the
210 /// presumed #include stack.  If it is 1, this is a file entry, if it is 2 then
211 /// this is a file exit.  FileKind specifies whether this is a system header or
212 /// extern C system header.
213 void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset,
214                                 unsigned LineNo, int FilenameID,
215                                 unsigned EntryExit,
216                                 SrcMgr::CharacteristicKind FileKind) {
217   assert(FilenameID != -1 && "Unspecified filename should use other accessor");
218 
219   std::vector<LineEntry> &Entries = LineEntries[FID];
220 
221   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
222          "Adding line entries out of order!");
223 
224   unsigned IncludeOffset = 0;
225   if (EntryExit == 0) {  // No #include stack change.
226     IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset;
227   } else if (EntryExit == 1) {
228     IncludeOffset = Offset-1;
229   } else if (EntryExit == 2) {
230     assert(!Entries.empty() && Entries.back().IncludeOffset &&
231        "PPDirectives should have caught case when popping empty include stack");
232 
233     // Get the include loc of the last entries' include loc as our include loc.
234     IncludeOffset = 0;
235     if (const LineEntry *PrevEntry =
236           FindNearestLineEntry(FID, Entries.back().IncludeOffset))
237       IncludeOffset = PrevEntry->IncludeOffset;
238   }
239 
240   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind,
241                                    IncludeOffset));
242 }
243 
244 
245 /// FindNearestLineEntry - Find the line entry nearest to FID that is before
246 /// it.  If there is no line entry before Offset in FID, return null.
247 const LineEntry *LineTableInfo::FindNearestLineEntry(unsigned FID,
248                                                      unsigned Offset) {
249   const std::vector<LineEntry> &Entries = LineEntries[FID];
250   assert(!Entries.empty() && "No #line entries for this FID after all!");
251 
252   // It is very common for the query to be after the last #line, check this
253   // first.
254   if (Entries.back().FileOffset <= Offset)
255     return &Entries.back();
256 
257   // Do a binary search to find the maximal element that is still before Offset.
258   std::vector<LineEntry>::const_iterator I =
259     std::upper_bound(Entries.begin(), Entries.end(), Offset);
260   if (I == Entries.begin()) return 0;
261   return &*--I;
262 }
263 
264 /// \brief Add a new line entry that has already been encoded into
265 /// the internal representation of the line table.
266 void LineTableInfo::AddEntry(unsigned FID,
267                              const std::vector<LineEntry> &Entries) {
268   LineEntries[FID] = Entries;
269 }
270 
271 /// getLineTableFilenameID - Return the uniqued ID for the specified filename.
272 ///
273 unsigned SourceManager::getLineTableFilenameID(const char *Ptr, unsigned Len) {
274   if (LineTable == 0)
275     LineTable = new LineTableInfo();
276   return LineTable->getLineTableFilenameID(Ptr, Len);
277 }
278 
279 
280 /// AddLineNote - Add a line note to the line table for the FileID and offset
281 /// specified by Loc.  If FilenameID is -1, it is considered to be
282 /// unspecified.
283 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
284                                 int FilenameID) {
285   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
286 
287   const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile();
288 
289   // Remember that this file has #line directives now if it doesn't already.
290   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
291 
292   if (LineTable == 0)
293     LineTable = new LineTableInfo();
294   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID);
295 }
296 
297 /// AddLineNote - Add a GNU line marker to the line table.
298 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
299                                 int FilenameID, bool IsFileEntry,
300                                 bool IsFileExit, bool IsSystemHeader,
301                                 bool IsExternCHeader) {
302   // If there is no filename and no flags, this is treated just like a #line,
303   // which does not change the flags of the previous line marker.
304   if (FilenameID == -1) {
305     assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader &&
306            "Can't set flags without setting the filename!");
307     return AddLineNote(Loc, LineNo, FilenameID);
308   }
309 
310   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
311   const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile();
312 
313   // Remember that this file has #line directives now if it doesn't already.
314   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
315 
316   if (LineTable == 0)
317     LineTable = new LineTableInfo();
318 
319   SrcMgr::CharacteristicKind FileKind;
320   if (IsExternCHeader)
321     FileKind = SrcMgr::C_ExternCSystem;
322   else if (IsSystemHeader)
323     FileKind = SrcMgr::C_System;
324   else
325     FileKind = SrcMgr::C_User;
326 
327   unsigned EntryExit = 0;
328   if (IsFileEntry)
329     EntryExit = 1;
330   else if (IsFileExit)
331     EntryExit = 2;
332 
333   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID,
334                          EntryExit, FileKind);
335 }
336 
337 LineTableInfo &SourceManager::getLineTable() {
338   if (LineTable == 0)
339     LineTable = new LineTableInfo();
340   return *LineTable;
341 }
342 
343 //===----------------------------------------------------------------------===//
344 // Private 'Create' methods.
345 //===----------------------------------------------------------------------===//
346 
347 SourceManager::~SourceManager() {
348   delete LineTable;
349 
350   // Delete FileEntry objects corresponding to content caches.  Since the actual
351   // content cache objects are bump pointer allocated, we just have to run the
352   // dtors, but we call the deallocate method for completeness.
353   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) {
354     MemBufferInfos[i]->~ContentCache();
355     ContentCacheAlloc.Deallocate(MemBufferInfos[i]);
356   }
357   for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator
358        I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) {
359     I->second->~ContentCache();
360     ContentCacheAlloc.Deallocate(I->second);
361   }
362 }
363 
364 void SourceManager::clearIDTables() {
365   MainFileID = FileID();
366   SLocEntryTable.clear();
367   LastLineNoFileIDQuery = FileID();
368   LastLineNoContentCache = 0;
369   LastFileIDLookup = FileID();
370 
371   if (LineTable)
372     LineTable->clear();
373 
374   // Use up FileID #0 as an invalid instantiation.
375   NextOffset = 0;
376   createInstantiationLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1);
377 }
378 
379 /// getOrCreateContentCache - Create or return a cached ContentCache for the
380 /// specified file.
381 const ContentCache *
382 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt) {
383   assert(FileEnt && "Didn't specify a file entry to use?");
384 
385   // Do we already have information about this file?
386   ContentCache *&Entry = FileInfos[FileEnt];
387   if (Entry) return Entry;
388 
389   // Nope, create a new Cache entry.  Make sure it is at least 8-byte aligned
390   // so that FileInfo can use the low 3 bits of the pointer for its own
391   // nefarious purposes.
392   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
393   EntryAlign = std::max(8U, EntryAlign);
394   Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
395   new (Entry) ContentCache(FileEnt);
396   return Entry;
397 }
398 
399 
400 /// createMemBufferContentCache - Create a new ContentCache for the specified
401 ///  memory buffer.  This does no caching.
402 const ContentCache*
403 SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) {
404   // Add a new ContentCache to the MemBufferInfos list and return it.  Make sure
405   // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of
406   // the pointer for its own nefarious purposes.
407   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
408   EntryAlign = std::max(8U, EntryAlign);
409   ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
410   new (Entry) ContentCache();
411   MemBufferInfos.push_back(Entry);
412   Entry->setBuffer(Buffer);
413   return Entry;
414 }
415 
416 void SourceManager::PreallocateSLocEntries(ExternalSLocEntrySource *Source,
417                                            unsigned NumSLocEntries,
418                                            unsigned NextOffset) {
419   ExternalSLocEntries = Source;
420   this->NextOffset = NextOffset;
421   SLocEntryLoaded.resize(NumSLocEntries + 1);
422   SLocEntryLoaded[0] = true;
423   SLocEntryTable.resize(SLocEntryTable.size() + NumSLocEntries);
424 }
425 
426 void SourceManager::ClearPreallocatedSLocEntries() {
427   unsigned I = 0;
428   for (unsigned N = SLocEntryLoaded.size(); I != N; ++I)
429     if (!SLocEntryLoaded[I])
430       break;
431 
432   // We've already loaded all preallocated source location entries.
433   if (I == SLocEntryLoaded.size())
434     return;
435 
436   // Remove everything from location I onward.
437   SLocEntryTable.resize(I);
438   SLocEntryLoaded.clear();
439   ExternalSLocEntries = 0;
440 }
441 
442 
443 //===----------------------------------------------------------------------===//
444 // Methods to create new FileID's and instantiations.
445 //===----------------------------------------------------------------------===//
446 
447 /// createFileID - Create a new fileID for the specified ContentCache and
448 /// include position.  This works regardless of whether the ContentCache
449 /// corresponds to a file or some other input source.
450 FileID SourceManager::createFileID(const ContentCache *File,
451                                    SourceLocation IncludePos,
452                                    SrcMgr::CharacteristicKind FileCharacter,
453                                    unsigned PreallocatedID,
454                                    unsigned Offset) {
455   if (PreallocatedID) {
456     // If we're filling in a preallocated ID, just load in the file
457     // entry and return.
458     assert(PreallocatedID < SLocEntryLoaded.size() &&
459            "Preallocate ID out-of-range");
460     assert(!SLocEntryLoaded[PreallocatedID] &&
461            "Source location entry already loaded");
462     assert(Offset && "Preallocate source location cannot have zero offset");
463     SLocEntryTable[PreallocatedID]
464       = SLocEntry::get(Offset, FileInfo::get(IncludePos, File, FileCharacter));
465     SLocEntryLoaded[PreallocatedID] = true;
466     FileID FID = FileID::get(PreallocatedID);
467     return FID;
468   }
469 
470   SLocEntryTable.push_back(SLocEntry::get(NextOffset,
471                                           FileInfo::get(IncludePos, File,
472                                                         FileCharacter)));
473   unsigned FileSize = File->getSize();
474   assert(NextOffset+FileSize+1 > NextOffset && "Ran out of source locations!");
475   NextOffset += FileSize+1;
476 
477   // Set LastFileIDLookup to the newly created file.  The next getFileID call is
478   // almost guaranteed to be from that file.
479   FileID FID = FileID::get(SLocEntryTable.size()-1);
480   return LastFileIDLookup = FID;
481 }
482 
483 /// createInstantiationLoc - Return a new SourceLocation that encodes the fact
484 /// that a token from SpellingLoc should actually be referenced from
485 /// InstantiationLoc.
486 SourceLocation SourceManager::createInstantiationLoc(SourceLocation SpellingLoc,
487                                                      SourceLocation ILocStart,
488                                                      SourceLocation ILocEnd,
489                                                      unsigned TokLength,
490                                                      unsigned PreallocatedID,
491                                                      unsigned Offset) {
492   InstantiationInfo II = InstantiationInfo::get(ILocStart,ILocEnd, SpellingLoc);
493   if (PreallocatedID) {
494     // If we're filling in a preallocated ID, just load in the
495     // instantiation entry and return.
496     assert(PreallocatedID < SLocEntryLoaded.size() &&
497            "Preallocate ID out-of-range");
498     assert(!SLocEntryLoaded[PreallocatedID] &&
499            "Source location entry already loaded");
500     assert(Offset && "Preallocate source location cannot have zero offset");
501     SLocEntryTable[PreallocatedID] = SLocEntry::get(Offset, II);
502     SLocEntryLoaded[PreallocatedID] = true;
503     return SourceLocation::getMacroLoc(Offset);
504   }
505   SLocEntryTable.push_back(SLocEntry::get(NextOffset, II));
506   assert(NextOffset+TokLength+1 > NextOffset && "Ran out of source locations!");
507   NextOffset += TokLength+1;
508   return SourceLocation::getMacroLoc(NextOffset-(TokLength+1));
509 }
510 
511 const llvm::MemoryBuffer *
512 SourceManager::getMemoryBufferForFile(const FileEntry *File,
513                                       bool *Invalid) {
514   const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
515   assert(IR && "getOrCreateContentCache() cannot return NULL");
516   return IR->getBuffer(Diag, Invalid);
517 }
518 
519 bool SourceManager::overrideFileContents(const FileEntry *SourceFile,
520                                          const llvm::MemoryBuffer *Buffer) {
521   const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile);
522   if (IR == 0)
523     return true;
524 
525   const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer);
526   return false;
527 }
528 
529 llvm::StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const {
530   bool MyInvalid = false;
531   const llvm::MemoryBuffer *Buf = getBuffer(FID, &MyInvalid);
532   if (Invalid)
533     *Invalid = MyInvalid;
534 
535   if (MyInvalid)
536     return "";
537 
538   return Buf->getBuffer();
539 }
540 
541 //===----------------------------------------------------------------------===//
542 // SourceLocation manipulation methods.
543 //===----------------------------------------------------------------------===//
544 
545 /// getFileIDSlow - Return the FileID for a SourceLocation.  This is a very hot
546 /// method that is used for all SourceManager queries that start with a
547 /// SourceLocation object.  It is responsible for finding the entry in
548 /// SLocEntryTable which contains the specified location.
549 ///
550 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const {
551   assert(SLocOffset && "Invalid FileID");
552 
553   // After the first and second level caches, I see two common sorts of
554   // behavior: 1) a lot of searched FileID's are "near" the cached file location
555   // or are "near" the cached instantiation location.  2) others are just
556   // completely random and may be a very long way away.
557   //
558   // To handle this, we do a linear search for up to 8 steps to catch #1 quickly
559   // then we fall back to a less cache efficient, but more scalable, binary
560   // search to find the location.
561 
562   // See if this is near the file point - worst case we start scanning from the
563   // most newly created FileID.
564   std::vector<SrcMgr::SLocEntry>::const_iterator I;
565 
566   if (SLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) {
567     // Neither loc prunes our search.
568     I = SLocEntryTable.end();
569   } else {
570     // Perhaps it is near the file point.
571     I = SLocEntryTable.begin()+LastFileIDLookup.ID;
572   }
573 
574   // Find the FileID that contains this.  "I" is an iterator that points to a
575   // FileID whose offset is known to be larger than SLocOffset.
576   unsigned NumProbes = 0;
577   while (1) {
578     --I;
579     if (ExternalSLocEntries)
580       getSLocEntry(FileID::get(I - SLocEntryTable.begin()));
581     if (I->getOffset() <= SLocOffset) {
582 #if 0
583       printf("lin %d -> %d [%s] %d %d\n", SLocOffset,
584              I-SLocEntryTable.begin(),
585              I->isInstantiation() ? "inst" : "file",
586              LastFileIDLookup.ID,  int(SLocEntryTable.end()-I));
587 #endif
588       FileID Res = FileID::get(I-SLocEntryTable.begin());
589 
590       // If this isn't an instantiation, remember it.  We have good locality
591       // across FileID lookups.
592       if (!I->isInstantiation())
593         LastFileIDLookup = Res;
594       NumLinearScans += NumProbes+1;
595       return Res;
596     }
597     if (++NumProbes == 8)
598       break;
599   }
600 
601   // Convert "I" back into an index.  We know that it is an entry whose index is
602   // larger than the offset we are looking for.
603   unsigned GreaterIndex = I-SLocEntryTable.begin();
604   // LessIndex - This is the lower bound of the range that we're searching.
605   // We know that the offset corresponding to the FileID is is less than
606   // SLocOffset.
607   unsigned LessIndex = 0;
608   NumProbes = 0;
609   while (1) {
610     unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex;
611     unsigned MidOffset = getSLocEntry(FileID::get(MiddleIndex)).getOffset();
612 
613     ++NumProbes;
614 
615     // If the offset of the midpoint is too large, chop the high side of the
616     // range to the midpoint.
617     if (MidOffset > SLocOffset) {
618       GreaterIndex = MiddleIndex;
619       continue;
620     }
621 
622     // If the middle index contains the value, succeed and return.
623     if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) {
624 #if 0
625       printf("bin %d -> %d [%s] %d %d\n", SLocOffset,
626              I-SLocEntryTable.begin(),
627              I->isInstantiation() ? "inst" : "file",
628              LastFileIDLookup.ID, int(SLocEntryTable.end()-I));
629 #endif
630       FileID Res = FileID::get(MiddleIndex);
631 
632       // If this isn't an instantiation, remember it.  We have good locality
633       // across FileID lookups.
634       if (!I->isInstantiation())
635         LastFileIDLookup = Res;
636       NumBinaryProbes += NumProbes;
637       return Res;
638     }
639 
640     // Otherwise, move the low-side up to the middle index.
641     LessIndex = MiddleIndex;
642   }
643 }
644 
645 SourceLocation SourceManager::
646 getInstantiationLocSlowCase(SourceLocation Loc) const {
647   do {
648     // Note: If Loc indicates an offset into a token that came from a macro
649     // expansion (e.g. the 5th character of the token) we do not want to add
650     // this offset when going to the instantiation location.  The instatiation
651     // location is the macro invocation, which the offset has nothing to do
652     // with.  This is unlike when we get the spelling loc, because the offset
653     // directly correspond to the token whose spelling we're inspecting.
654     Loc = getSLocEntry(getFileID(Loc)).getInstantiation()
655                    .getInstantiationLocStart();
656   } while (!Loc.isFileID());
657 
658   return Loc;
659 }
660 
661 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
662   do {
663     std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
664     Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc();
665     Loc = Loc.getFileLocWithOffset(LocInfo.second);
666   } while (!Loc.isFileID());
667   return Loc;
668 }
669 
670 
671 std::pair<FileID, unsigned>
672 SourceManager::getDecomposedInstantiationLocSlowCase(const SrcMgr::SLocEntry *E,
673                                                      unsigned Offset) const {
674   // If this is an instantiation record, walk through all the instantiation
675   // points.
676   FileID FID;
677   SourceLocation Loc;
678   do {
679     Loc = E->getInstantiation().getInstantiationLocStart();
680 
681     FID = getFileID(Loc);
682     E = &getSLocEntry(FID);
683     Offset += Loc.getOffset()-E->getOffset();
684   } while (!Loc.isFileID());
685 
686   return std::make_pair(FID, Offset);
687 }
688 
689 std::pair<FileID, unsigned>
690 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
691                                                 unsigned Offset) const {
692   // If this is an instantiation record, walk through all the instantiation
693   // points.
694   FileID FID;
695   SourceLocation Loc;
696   do {
697     Loc = E->getInstantiation().getSpellingLoc();
698 
699     FID = getFileID(Loc);
700     E = &getSLocEntry(FID);
701     Offset += Loc.getOffset()-E->getOffset();
702   } while (!Loc.isFileID());
703 
704   return std::make_pair(FID, Offset);
705 }
706 
707 /// getImmediateSpellingLoc - Given a SourceLocation object, return the
708 /// spelling location referenced by the ID.  This is the first level down
709 /// towards the place where the characters that make up the lexed token can be
710 /// found.  This should not generally be used by clients.
711 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
712   if (Loc.isFileID()) return Loc;
713   std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
714   Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc();
715   return Loc.getFileLocWithOffset(LocInfo.second);
716 }
717 
718 
719 /// getImmediateInstantiationRange - Loc is required to be an instantiation
720 /// location.  Return the start/end of the instantiation information.
721 std::pair<SourceLocation,SourceLocation>
722 SourceManager::getImmediateInstantiationRange(SourceLocation Loc) const {
723   assert(Loc.isMacroID() && "Not an instantiation loc!");
724   const InstantiationInfo &II = getSLocEntry(getFileID(Loc)).getInstantiation();
725   return II.getInstantiationLocRange();
726 }
727 
728 /// getInstantiationRange - Given a SourceLocation object, return the
729 /// range of tokens covered by the instantiation in the ultimate file.
730 std::pair<SourceLocation,SourceLocation>
731 SourceManager::getInstantiationRange(SourceLocation Loc) const {
732   if (Loc.isFileID()) return std::make_pair(Loc, Loc);
733 
734   std::pair<SourceLocation,SourceLocation> Res =
735     getImmediateInstantiationRange(Loc);
736 
737   // Fully resolve the start and end locations to their ultimate instantiation
738   // points.
739   while (!Res.first.isFileID())
740     Res.first = getImmediateInstantiationRange(Res.first).first;
741   while (!Res.second.isFileID())
742     Res.second = getImmediateInstantiationRange(Res.second).second;
743   return Res;
744 }
745 
746 
747 
748 //===----------------------------------------------------------------------===//
749 // Queries about the code at a SourceLocation.
750 //===----------------------------------------------------------------------===//
751 
752 /// getCharacterData - Return a pointer to the start of the specified location
753 /// in the appropriate MemoryBuffer.
754 const char *SourceManager::getCharacterData(SourceLocation SL,
755                                             bool *Invalid) const {
756   // Note that this is a hot function in the getSpelling() path, which is
757   // heavily used by -E mode.
758   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
759 
760   // Note that calling 'getBuffer()' may lazily page in a source file.
761   bool CharDataInvalid = false;
762   const llvm::MemoryBuffer *Buffer
763     = getSLocEntry(LocInfo.first).getFile().getContentCache()->getBuffer(Diag,
764                                                               &CharDataInvalid);
765   if (Invalid)
766     *Invalid = CharDataInvalid;
767   return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
768 }
769 
770 
771 /// getColumnNumber - Return the column # for the specified file position.
772 /// this is significantly cheaper to compute than the line number.
773 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
774                                         bool *Invalid) const {
775   bool MyInvalid = false;
776   const char *Buf = getBuffer(FID, &MyInvalid)->getBufferStart();
777   if (Invalid)
778     *Invalid = MyInvalid;
779 
780   if (MyInvalid)
781     return 1;
782 
783   unsigned LineStart = FilePos;
784   while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
785     --LineStart;
786   return FilePos-LineStart+1;
787 }
788 
789 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
790                                                 bool *Invalid) const {
791   if (Loc.isInvalid()) return 0;
792   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
793   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
794 }
795 
796 unsigned SourceManager::getInstantiationColumnNumber(SourceLocation Loc,
797                                                      bool *Invalid) const {
798   if (Loc.isInvalid()) return 0;
799   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
800   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
801 }
802 
803 static DISABLE_INLINE void ComputeLineNumbers(Diagnostic &Diag,
804                                               ContentCache* FI,
805                                               llvm::BumpPtrAllocator &Alloc,
806                                               bool &Invalid);
807 static void ComputeLineNumbers(Diagnostic &Diag, ContentCache* FI,
808                                llvm::BumpPtrAllocator &Alloc, bool &Invalid) {
809   // Note that calling 'getBuffer()' may lazily page in the file.
810   const MemoryBuffer *Buffer = FI->getBuffer(Diag, &Invalid);
811   if (Invalid)
812     return;
813 
814   // Find the file offsets of all of the *physical* source lines.  This does
815   // not look at trigraphs, escaped newlines, or anything else tricky.
816   std::vector<unsigned> LineOffsets;
817 
818   // Line #1 starts at char 0.
819   LineOffsets.push_back(0);
820 
821   const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
822   const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
823   unsigned Offs = 0;
824   while (1) {
825     // Skip over the contents of the line.
826     // TODO: Vectorize this?  This is very performance sensitive for programs
827     // with lots of diagnostics and in -E mode.
828     const unsigned char *NextBuf = (const unsigned char *)Buf;
829     while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
830       ++NextBuf;
831     Offs += NextBuf-Buf;
832     Buf = NextBuf;
833 
834     if (Buf[0] == '\n' || Buf[0] == '\r') {
835       // If this is \n\r or \r\n, skip both characters.
836       if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1])
837         ++Offs, ++Buf;
838       ++Offs, ++Buf;
839       LineOffsets.push_back(Offs);
840     } else {
841       // Otherwise, this is a null.  If end of file, exit.
842       if (Buf == End) break;
843       // Otherwise, skip the null.
844       ++Offs, ++Buf;
845     }
846   }
847 
848   // Copy the offsets into the FileInfo structure.
849   FI->NumLines = LineOffsets.size();
850   FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
851   std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
852 }
853 
854 /// getLineNumber - Given a SourceLocation, return the spelling line number
855 /// for the position indicated.  This requires building and caching a table of
856 /// line offsets for the MemoryBuffer, so this is not cheap: use only when
857 /// about to emit a diagnostic.
858 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
859                                       bool *Invalid) const {
860   ContentCache *Content;
861   if (LastLineNoFileIDQuery == FID)
862     Content = LastLineNoContentCache;
863   else
864     Content = const_cast<ContentCache*>(getSLocEntry(FID)
865                                         .getFile().getContentCache());
866 
867   // If this is the first use of line information for this buffer, compute the
868   /// SourceLineCache for it on demand.
869   if (Content->SourceLineCache == 0) {
870     bool MyInvalid = false;
871     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, MyInvalid);
872     if (Invalid)
873       *Invalid = MyInvalid;
874     if (MyInvalid)
875       return 1;
876   } else if (Invalid)
877     *Invalid = false;
878 
879   // Okay, we know we have a line number table.  Do a binary search to find the
880   // line number that this character position lands on.
881   unsigned *SourceLineCache = Content->SourceLineCache;
882   unsigned *SourceLineCacheStart = SourceLineCache;
883   unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
884 
885   unsigned QueriedFilePos = FilePos+1;
886 
887   // FIXME: I would like to be convinced that this code is worth being as
888   // complicated as it is, binary search isn't that slow.
889   //
890   // If it is worth being optimized, then in my opinion it could be more
891   // performant, simpler, and more obviously correct by just "galloping" outward
892   // from the queried file position. In fact, this could be incorporated into a
893   // generic algorithm such as lower_bound_with_hint.
894   //
895   // If someone gives me a test case where this matters, and I will do it! - DWD
896 
897   // If the previous query was to the same file, we know both the file pos from
898   // that query and the line number returned.  This allows us to narrow the
899   // search space from the entire file to something near the match.
900   if (LastLineNoFileIDQuery == FID) {
901     if (QueriedFilePos >= LastLineNoFilePos) {
902       // FIXME: Potential overflow?
903       SourceLineCache = SourceLineCache+LastLineNoResult-1;
904 
905       // The query is likely to be nearby the previous one.  Here we check to
906       // see if it is within 5, 10 or 20 lines.  It can be far away in cases
907       // where big comment blocks and vertical whitespace eat up lines but
908       // contribute no tokens.
909       if (SourceLineCache+5 < SourceLineCacheEnd) {
910         if (SourceLineCache[5] > QueriedFilePos)
911           SourceLineCacheEnd = SourceLineCache+5;
912         else if (SourceLineCache+10 < SourceLineCacheEnd) {
913           if (SourceLineCache[10] > QueriedFilePos)
914             SourceLineCacheEnd = SourceLineCache+10;
915           else if (SourceLineCache+20 < SourceLineCacheEnd) {
916             if (SourceLineCache[20] > QueriedFilePos)
917               SourceLineCacheEnd = SourceLineCache+20;
918           }
919         }
920       }
921     } else {
922       if (LastLineNoResult < Content->NumLines)
923         SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
924     }
925   }
926 
927   // If the spread is large, do a "radix" test as our initial guess, based on
928   // the assumption that lines average to approximately the same length.
929   // NOTE: This is currently disabled, as it does not appear to be profitable in
930   // initial measurements.
931   if (0 && SourceLineCacheEnd-SourceLineCache > 20) {
932     unsigned FileLen = Content->SourceLineCache[Content->NumLines-1];
933 
934     // Take a stab at guessing where it is.
935     unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen;
936 
937     // Check for -10 and +10 lines.
938     unsigned LowerBound = std::max(int(ApproxPos-10), 0);
939     unsigned UpperBound = std::min(ApproxPos+10, FileLen);
940 
941     // If the computed lower bound is less than the query location, move it in.
942     if (SourceLineCache < SourceLineCacheStart+LowerBound &&
943         SourceLineCacheStart[LowerBound] < QueriedFilePos)
944       SourceLineCache = SourceLineCacheStart+LowerBound;
945 
946     // If the computed upper bound is greater than the query location, move it.
947     if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound &&
948         SourceLineCacheStart[UpperBound] >= QueriedFilePos)
949       SourceLineCacheEnd = SourceLineCacheStart+UpperBound;
950   }
951 
952   unsigned *Pos
953     = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
954   unsigned LineNo = Pos-SourceLineCacheStart;
955 
956   LastLineNoFileIDQuery = FID;
957   LastLineNoContentCache = Content;
958   LastLineNoFilePos = QueriedFilePos;
959   LastLineNoResult = LineNo;
960   return LineNo;
961 }
962 
963 unsigned SourceManager::getInstantiationLineNumber(SourceLocation Loc,
964                                                    bool *Invalid) const {
965   if (Loc.isInvalid()) return 0;
966   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
967   return getLineNumber(LocInfo.first, LocInfo.second);
968 }
969 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
970                                               bool *Invalid) const {
971   if (Loc.isInvalid()) return 0;
972   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
973   return getLineNumber(LocInfo.first, LocInfo.second);
974 }
975 
976 /// getFileCharacteristic - return the file characteristic of the specified
977 /// source location, indicating whether this is a normal file, a system
978 /// header, or an "implicit extern C" system header.
979 ///
980 /// This state can be modified with flags on GNU linemarker directives like:
981 ///   # 4 "foo.h" 3
982 /// which changes all source locations in the current file after that to be
983 /// considered to be from a system header.
984 SrcMgr::CharacteristicKind
985 SourceManager::getFileCharacteristic(SourceLocation Loc) const {
986   assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!");
987   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
988   const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile();
989 
990   // If there are no #line directives in this file, just return the whole-file
991   // state.
992   if (!FI.hasLineDirectives())
993     return FI.getFileCharacteristic();
994 
995   assert(LineTable && "Can't have linetable entries without a LineTable!");
996   // See if there is a #line directive before the location.
997   const LineEntry *Entry =
998     LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second);
999 
1000   // If this is before the first line marker, use the file characteristic.
1001   if (!Entry)
1002     return FI.getFileCharacteristic();
1003 
1004   return Entry->FileKind;
1005 }
1006 
1007 /// Return the filename or buffer identifier of the buffer the location is in.
1008 /// Note that this name does not respect #line directives.  Use getPresumedLoc
1009 /// for normal clients.
1010 const char *SourceManager::getBufferName(SourceLocation Loc,
1011                                          bool *Invalid) const {
1012   if (Loc.isInvalid()) return "<invalid loc>";
1013 
1014   return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
1015 }
1016 
1017 
1018 /// getPresumedLoc - This method returns the "presumed" location of a
1019 /// SourceLocation specifies.  A "presumed location" can be modified by #line
1020 /// or GNU line marker directives.  This provides a view on the data that a
1021 /// user should see in diagnostics, for example.
1022 ///
1023 /// Note that a presumed location is always given as the instantiation point
1024 /// of an instantiation location, not at the spelling location.
1025 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc) const {
1026   if (Loc.isInvalid()) return PresumedLoc();
1027 
1028   // Presumed locations are always for instantiation points.
1029   std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
1030 
1031   const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile();
1032   const SrcMgr::ContentCache *C = FI.getContentCache();
1033 
1034   // To get the source name, first consult the FileEntry (if one exists)
1035   // before the MemBuffer as this will avoid unnecessarily paging in the
1036   // MemBuffer.
1037   const char *Filename =
1038     C->Entry ? C->Entry->getName() : C->getBuffer(Diag)->getBufferIdentifier();
1039   unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second);
1040   unsigned ColNo  = getColumnNumber(LocInfo.first, LocInfo.second);
1041   SourceLocation IncludeLoc = FI.getIncludeLoc();
1042 
1043   // If we have #line directives in this file, update and overwrite the physical
1044   // location info if appropriate.
1045   if (FI.hasLineDirectives()) {
1046     assert(LineTable && "Can't have linetable entries without a LineTable!");
1047     // See if there is a #line directive before this.  If so, get it.
1048     if (const LineEntry *Entry =
1049           LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second)) {
1050       // If the LineEntry indicates a filename, use it.
1051       if (Entry->FilenameID != -1)
1052         Filename = LineTable->getFilename(Entry->FilenameID);
1053 
1054       // Use the line number specified by the LineEntry.  This line number may
1055       // be multiple lines down from the line entry.  Add the difference in
1056       // physical line numbers from the query point and the line marker to the
1057       // total.
1058       unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
1059       LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
1060 
1061       // Note that column numbers are not molested by line markers.
1062 
1063       // Handle virtual #include manipulation.
1064       if (Entry->IncludeOffset) {
1065         IncludeLoc = getLocForStartOfFile(LocInfo.first);
1066         IncludeLoc = IncludeLoc.getFileLocWithOffset(Entry->IncludeOffset);
1067       }
1068     }
1069   }
1070 
1071   return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc);
1072 }
1073 
1074 //===----------------------------------------------------------------------===//
1075 // Other miscellaneous methods.
1076 //===----------------------------------------------------------------------===//
1077 
1078 /// \brief Get the source location for the given file:line:col triplet.
1079 ///
1080 /// If the source file is included multiple times, the source location will
1081 /// be based upon the first inclusion.
1082 SourceLocation SourceManager::getLocation(const FileEntry *SourceFile,
1083                                           unsigned Line, unsigned Col) const {
1084   assert(SourceFile && "Null source file!");
1085   assert(Line && Col && "Line and column should start from 1!");
1086 
1087   fileinfo_iterator FI = FileInfos.find(SourceFile);
1088   if (FI == FileInfos.end())
1089     return SourceLocation();
1090   ContentCache *Content = FI->second;
1091 
1092   // If this is the first use of line information for this buffer, compute the
1093   /// SourceLineCache for it on demand.
1094   if (Content->SourceLineCache == 0) {
1095     bool MyInvalid = false;
1096     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, MyInvalid);
1097     if (MyInvalid)
1098       return SourceLocation();
1099   }
1100 
1101   // Find the first file ID that corresponds to the given file.
1102   FileID FirstFID;
1103 
1104   // First, check the main file ID, since it is common to look for a
1105   // location in the main file.
1106   if (!MainFileID.isInvalid()) {
1107     const SLocEntry &MainSLoc = getSLocEntry(MainFileID);
1108     if (MainSLoc.isFile() && MainSLoc.getFile().getContentCache() == Content)
1109       FirstFID = MainFileID;
1110   }
1111 
1112   if (FirstFID.isInvalid()) {
1113     // The location we're looking for isn't in the main file; look
1114     // through all of the source locations.
1115     for (unsigned I = 0, N = sloc_entry_size(); I != N; ++I) {
1116       const SLocEntry &SLoc = getSLocEntry(I);
1117       if (SLoc.isFile() && SLoc.getFile().getContentCache() == Content) {
1118         FirstFID = FileID::get(I);
1119         break;
1120       }
1121     }
1122   }
1123 
1124   if (FirstFID.isInvalid())
1125     return SourceLocation();
1126 
1127   if (Line > Content->NumLines) {
1128     unsigned Size = Content->getBuffer(Diag)->getBufferSize();
1129     if (Size > 0)
1130       --Size;
1131     return getLocForStartOfFile(FirstFID).getFileLocWithOffset(Size);
1132   }
1133 
1134   unsigned FilePos = Content->SourceLineCache[Line - 1];
1135   const char *Buf = Content->getBuffer(Diag)->getBufferStart() + FilePos;
1136   unsigned BufLength = Content->getBuffer(Diag)->getBufferEnd() - Buf;
1137   unsigned i = 0;
1138 
1139   // Check that the given column is valid.
1140   while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
1141     ++i;
1142   if (i < Col-1)
1143     return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + i);
1144 
1145   return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + Col - 1);
1146 }
1147 
1148 /// \brief Determines the order of 2 source locations in the translation unit.
1149 ///
1150 /// \returns true if LHS source location comes before RHS, false otherwise.
1151 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
1152                                               SourceLocation RHS) const {
1153   assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
1154   if (LHS == RHS)
1155     return false;
1156 
1157   std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
1158   std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
1159 
1160   // If the source locations are in the same file, just compare offsets.
1161   if (LOffs.first == ROffs.first)
1162     return LOffs.second < ROffs.second;
1163 
1164   // If we are comparing a source location with multiple locations in the same
1165   // file, we get a big win by caching the result.
1166 
1167   if (LastLFIDForBeforeTUCheck == LOffs.first &&
1168       LastRFIDForBeforeTUCheck == ROffs.first)
1169     return LastResForBeforeTUCheck;
1170 
1171   LastLFIDForBeforeTUCheck = LOffs.first;
1172   LastRFIDForBeforeTUCheck = ROffs.first;
1173 
1174   // "Traverse" the include/instantiation stacks of both locations and try to
1175   // find a common "ancestor".
1176   //
1177   // First we traverse the stack of the right location and check each level
1178   // against the level of the left location, while collecting all levels in a
1179   // "stack map".
1180 
1181   std::map<FileID, unsigned> ROffsMap;
1182   ROffsMap[ROffs.first] = ROffs.second;
1183 
1184   while (1) {
1185     SourceLocation UpperLoc;
1186     const SrcMgr::SLocEntry &Entry = getSLocEntry(ROffs.first);
1187     if (Entry.isInstantiation())
1188       UpperLoc = Entry.getInstantiation().getInstantiationLocStart();
1189     else
1190       UpperLoc = Entry.getFile().getIncludeLoc();
1191 
1192     if (UpperLoc.isInvalid())
1193       break; // We reached the top.
1194 
1195     ROffs = getDecomposedLoc(UpperLoc);
1196 
1197     if (LOffs.first == ROffs.first)
1198       return LastResForBeforeTUCheck = LOffs.second < ROffs.second;
1199 
1200     ROffsMap[ROffs.first] = ROffs.second;
1201   }
1202 
1203   // We didn't find a common ancestor. Now traverse the stack of the left
1204   // location, checking against the stack map of the right location.
1205 
1206   while (1) {
1207     SourceLocation UpperLoc;
1208     const SrcMgr::SLocEntry &Entry = getSLocEntry(LOffs.first);
1209     if (Entry.isInstantiation())
1210       UpperLoc = Entry.getInstantiation().getInstantiationLocStart();
1211     else
1212       UpperLoc = Entry.getFile().getIncludeLoc();
1213 
1214     if (UpperLoc.isInvalid())
1215       break; // We reached the top.
1216 
1217     LOffs = getDecomposedLoc(UpperLoc);
1218 
1219     std::map<FileID, unsigned>::iterator I = ROffsMap.find(LOffs.first);
1220     if (I != ROffsMap.end())
1221       return LastResForBeforeTUCheck = LOffs.second < I->second;
1222   }
1223 
1224   // There is no common ancestor, most probably because one location is in the
1225   // predefines buffer.
1226   //
1227   // FIXME: We should rearrange the external interface so this simply never
1228   // happens; it can't conceptually happen. Also see PR5662.
1229 
1230   // If exactly one location is a memory buffer, assume it preceeds the other.
1231   bool LIsMB = !getSLocEntry(LOffs.first).getFile().getContentCache()->Entry;
1232   bool RIsMB = !getSLocEntry(ROffs.first).getFile().getContentCache()->Entry;
1233   if (LIsMB != RIsMB)
1234     return LastResForBeforeTUCheck = LIsMB;
1235 
1236   // Otherwise, just assume FileIDs were created in order.
1237   return LastResForBeforeTUCheck = (LOffs.first < ROffs.first);
1238 }
1239 
1240 /// PrintStats - Print statistics to stderr.
1241 ///
1242 void SourceManager::PrintStats() const {
1243   llvm::errs() << "\n*** Source Manager Stats:\n";
1244   llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
1245                << " mem buffers mapped.\n";
1246   llvm::errs() << SLocEntryTable.size() << " SLocEntry's allocated, "
1247                << NextOffset << "B of Sloc address space used.\n";
1248 
1249   unsigned NumLineNumsComputed = 0;
1250   unsigned NumFileBytesMapped = 0;
1251   for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
1252     NumLineNumsComputed += I->second->SourceLineCache != 0;
1253     NumFileBytesMapped  += I->second->getSizeBytesMapped();
1254   }
1255 
1256   llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
1257                << NumLineNumsComputed << " files with line #'s computed.\n";
1258   llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
1259                << NumBinaryProbes << " binary.\n";
1260 }
1261 
1262 ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
1263