SourceManager.cpp revision 669a67760c601a20787d375eb73047763d5adaaa
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
26using namespace clang;
27using namespace SrcMgr;
28using llvm::MemoryBuffer;
29
30//===----------------------------------------------------------------------===//
31// SourceManager Helper Classes
32//===----------------------------------------------------------------------===//
33
34ContentCache::~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.
41unsigned 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.
49unsigned ContentCache::getSize() const {
50  return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize()
51                             : (unsigned) Entry->getSize();
52}
53
54void 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
62const 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    std::string ErrorStr;
72    struct stat FileInfo;
73    Buffer.setPointer(MemoryBuffer::getFile(Entry->getName(), &ErrorStr,
74                                            Entry->getSize(), &FileInfo));
75    Buffer.setInt(false);
76
77    // If we were unable to open the file, then we are in an inconsistent
78    // situation where the content cache referenced a file which no longer
79    // exists. Most likely, we were using a stat cache with an invalid entry but
80    // the file could also have been removed during processing. Since we can't
81    // really deal with this situation, just create an empty buffer.
82    //
83    // FIXME: This is definitely not ideal, but our immediate clients can't
84    // currently handle returning a null entry here. Ideally we should detect
85    // that we are in an inconsistent situation and error out as quickly as
86    // possible.
87    if (!Buffer.getPointer()) {
88      const llvm::StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
89      Buffer.setPointer(MemoryBuffer::getNewMemBuffer(Entry->getSize(),
90                                                      "<invalid>"));
91      char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
92      for (unsigned i = 0, e = Entry->getSize(); i != e; ++i)
93        Ptr[i] = FillStr[i % FillStr.size()];
94
95      if (Diag.isDiagnosticInFlight())
96        Diag.SetDelayedDiagnostic(diag::err_cannot_open_file,
97                                  Entry->getName(), ErrorStr);
98      else
99        Diag.Report(FullSourceLoc(Loc, SM), diag::err_cannot_open_file)
100          << Entry->getName() << ErrorStr;
101
102      Buffer.setInt(true);
103
104    // FIXME: This conditionalization is horrible, but we see spurious failures
105    // in the test suite due to this warning and no one has had time to hunt it
106    // down. So for now, we just don't emit this diagnostic on Win32, and hope
107    // nothing bad happens.
108    //
109    // PR6812.
110#if !defined(LLVM_ON_WIN32)
111    } else if (FileInfo.st_size != Entry->getSize() ||
112               FileInfo.st_mtime != Entry->getModificationTime()) {
113      // Check that the file's size and modification time are the same
114      // as in the file entry (which may have come from a stat cache).
115      if (Diag.isDiagnosticInFlight())
116        Diag.SetDelayedDiagnostic(diag::err_file_modified,
117                                  Entry->getName());
118      else
119        Diag.Report(FullSourceLoc(Loc, SM), diag::err_file_modified)
120          << Entry->getName();
121
122      Buffer.setInt(true);
123#endif
124    }
125
126    // If the buffer is valid, check to see if it has a UTF Byte Order Mark
127    // (BOM).  We only support UTF-8 without a BOM right now.  See
128    // http://en.wikipedia.org/wiki/Byte_order_mark for more information.
129    if (!Buffer.getInt()) {
130      llvm::StringRef BufStr = Buffer.getPointer()->getBuffer();
131      const char *BOM = 0;
132      if (BufStr.startswith("\xFE\xBB\xBF"))
133        BOM = "UTF-8";
134      else if (BufStr.startswith("\xFE\xFF"))
135        BOM = "UTF-16 (BE)";
136      else if (BufStr.startswith("\xFF\xFE"))
137        BOM = "UTF-16 (LE)";
138      else if (BufStr.startswith(llvm::StringRef("\x00\x00\xFE\xFF", 4)))
139        BOM = "UTF-32 (BE)";
140      else if (BufStr.startswith(llvm::StringRef("\xFF\xFE\x00\x00", 4)))
141        BOM = "UTF-32 (LE)";
142      else if (BufStr.startswith("\x2B\x2F\x76"))
143        BOM = "UTF-7";
144      else if (BufStr.startswith("\xF7\x64\x4C"))
145        BOM = "UTF-1";
146      else if (BufStr.startswith("\xDD\x73\x66\x73"))
147        BOM = "UTF-EBCDIC";
148      else if (BufStr.startswith("\x0E\xFE\xFF"))
149        BOM = "SDSU";
150      else if (BufStr.startswith("\xFB\xEE\x28"))
151        BOM = "BOCU-1";
152      else if (BufStr.startswith("\x84\x31\x95\x33"))
153        BOM = "BOCU-1";
154
155      if (BOM) {
156        Diag.Report(FullSourceLoc(Loc, SM), diag::err_unsupported_bom)
157          << BOM << Entry->getName();
158        Buffer.setInt(1);
159      }
160    }
161  }
162
163  if (Invalid)
164    *Invalid = Buffer.getInt();
165
166  return Buffer.getPointer();
167}
168
169unsigned LineTableInfo::getLineTableFilenameID(const char *Ptr, unsigned Len) {
170  // Look up the filename in the string table, returning the pre-existing value
171  // if it exists.
172  llvm::StringMapEntry<unsigned> &Entry =
173    FilenameIDs.GetOrCreateValue(Ptr, Ptr+Len, ~0U);
174  if (Entry.getValue() != ~0U)
175    return Entry.getValue();
176
177  // Otherwise, assign this the next available ID.
178  Entry.setValue(FilenamesByID.size());
179  FilenamesByID.push_back(&Entry);
180  return FilenamesByID.size()-1;
181}
182
183/// AddLineNote - Add a line note to the line table that indicates that there
184/// is a #line at the specified FID/Offset location which changes the presumed
185/// location to LineNo/FilenameID.
186void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset,
187                                unsigned LineNo, int FilenameID) {
188  std::vector<LineEntry> &Entries = LineEntries[FID];
189
190  assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
191         "Adding line entries out of order!");
192
193  SrcMgr::CharacteristicKind Kind = SrcMgr::C_User;
194  unsigned IncludeOffset = 0;
195
196  if (!Entries.empty()) {
197    // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember
198    // that we are still in "foo.h".
199    if (FilenameID == -1)
200      FilenameID = Entries.back().FilenameID;
201
202    // If we are after a line marker that switched us to system header mode, or
203    // that set #include information, preserve it.
204    Kind = Entries.back().FileKind;
205    IncludeOffset = Entries.back().IncludeOffset;
206  }
207
208  Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind,
209                                   IncludeOffset));
210}
211
212/// AddLineNote This is the same as the previous version of AddLineNote, but is
213/// used for GNU line markers.  If EntryExit is 0, then this doesn't change the
214/// presumed #include stack.  If it is 1, this is a file entry, if it is 2 then
215/// this is a file exit.  FileKind specifies whether this is a system header or
216/// extern C system header.
217void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset,
218                                unsigned LineNo, int FilenameID,
219                                unsigned EntryExit,
220                                SrcMgr::CharacteristicKind FileKind) {
221  assert(FilenameID != -1 && "Unspecified filename should use other accessor");
222
223  std::vector<LineEntry> &Entries = LineEntries[FID];
224
225  assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
226         "Adding line entries out of order!");
227
228  unsigned IncludeOffset = 0;
229  if (EntryExit == 0) {  // No #include stack change.
230    IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset;
231  } else if (EntryExit == 1) {
232    IncludeOffset = Offset-1;
233  } else if (EntryExit == 2) {
234    assert(!Entries.empty() && Entries.back().IncludeOffset &&
235       "PPDirectives should have caught case when popping empty include stack");
236
237    // Get the include loc of the last entries' include loc as our include loc.
238    IncludeOffset = 0;
239    if (const LineEntry *PrevEntry =
240          FindNearestLineEntry(FID, Entries.back().IncludeOffset))
241      IncludeOffset = PrevEntry->IncludeOffset;
242  }
243
244  Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind,
245                                   IncludeOffset));
246}
247
248
249/// FindNearestLineEntry - Find the line entry nearest to FID that is before
250/// it.  If there is no line entry before Offset in FID, return null.
251const LineEntry *LineTableInfo::FindNearestLineEntry(unsigned FID,
252                                                     unsigned Offset) {
253  const std::vector<LineEntry> &Entries = LineEntries[FID];
254  assert(!Entries.empty() && "No #line entries for this FID after all!");
255
256  // It is very common for the query to be after the last #line, check this
257  // first.
258  if (Entries.back().FileOffset <= Offset)
259    return &Entries.back();
260
261  // Do a binary search to find the maximal element that is still before Offset.
262  std::vector<LineEntry>::const_iterator I =
263    std::upper_bound(Entries.begin(), Entries.end(), Offset);
264  if (I == Entries.begin()) return 0;
265  return &*--I;
266}
267
268/// \brief Add a new line entry that has already been encoded into
269/// the internal representation of the line table.
270void LineTableInfo::AddEntry(unsigned FID,
271                             const std::vector<LineEntry> &Entries) {
272  LineEntries[FID] = Entries;
273}
274
275/// getLineTableFilenameID - Return the uniqued ID for the specified filename.
276///
277unsigned SourceManager::getLineTableFilenameID(const char *Ptr, unsigned Len) {
278  if (LineTable == 0)
279    LineTable = new LineTableInfo();
280  return LineTable->getLineTableFilenameID(Ptr, Len);
281}
282
283
284/// AddLineNote - Add a line note to the line table for the FileID and offset
285/// specified by Loc.  If FilenameID is -1, it is considered to be
286/// unspecified.
287void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
288                                int FilenameID) {
289  std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
290
291  const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile();
292
293  // Remember that this file has #line directives now if it doesn't already.
294  const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
295
296  if (LineTable == 0)
297    LineTable = new LineTableInfo();
298  LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID);
299}
300
301/// AddLineNote - Add a GNU line marker to the line table.
302void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
303                                int FilenameID, bool IsFileEntry,
304                                bool IsFileExit, bool IsSystemHeader,
305                                bool IsExternCHeader) {
306  // If there is no filename and no flags, this is treated just like a #line,
307  // which does not change the flags of the previous line marker.
308  if (FilenameID == -1) {
309    assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader &&
310           "Can't set flags without setting the filename!");
311    return AddLineNote(Loc, LineNo, FilenameID);
312  }
313
314  std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
315  const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile();
316
317  // Remember that this file has #line directives now if it doesn't already.
318  const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
319
320  if (LineTable == 0)
321    LineTable = new LineTableInfo();
322
323  SrcMgr::CharacteristicKind FileKind;
324  if (IsExternCHeader)
325    FileKind = SrcMgr::C_ExternCSystem;
326  else if (IsSystemHeader)
327    FileKind = SrcMgr::C_System;
328  else
329    FileKind = SrcMgr::C_User;
330
331  unsigned EntryExit = 0;
332  if (IsFileEntry)
333    EntryExit = 1;
334  else if (IsFileExit)
335    EntryExit = 2;
336
337  LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID,
338                         EntryExit, FileKind);
339}
340
341LineTableInfo &SourceManager::getLineTable() {
342  if (LineTable == 0)
343    LineTable = new LineTableInfo();
344  return *LineTable;
345}
346
347//===----------------------------------------------------------------------===//
348// Private 'Create' methods.
349//===----------------------------------------------------------------------===//
350
351SourceManager::~SourceManager() {
352  delete LineTable;
353
354  // Delete FileEntry objects corresponding to content caches.  Since the actual
355  // content cache objects are bump pointer allocated, we just have to run the
356  // dtors, but we call the deallocate method for completeness.
357  for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) {
358    MemBufferInfos[i]->~ContentCache();
359    ContentCacheAlloc.Deallocate(MemBufferInfos[i]);
360  }
361  for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator
362       I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) {
363    I->second->~ContentCache();
364    ContentCacheAlloc.Deallocate(I->second);
365  }
366}
367
368void SourceManager::clearIDTables() {
369  MainFileID = FileID();
370  SLocEntryTable.clear();
371  LastLineNoFileIDQuery = FileID();
372  LastLineNoContentCache = 0;
373  LastFileIDLookup = FileID();
374
375  if (LineTable)
376    LineTable->clear();
377
378  // Use up FileID #0 as an invalid instantiation.
379  NextOffset = 0;
380  createInstantiationLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1);
381}
382
383/// getOrCreateContentCache - Create or return a cached ContentCache for the
384/// specified file.
385const ContentCache *
386SourceManager::getOrCreateContentCache(const FileEntry *FileEnt) {
387  assert(FileEnt && "Didn't specify a file entry to use?");
388
389  // Do we already have information about this file?
390  ContentCache *&Entry = FileInfos[FileEnt];
391  if (Entry) return Entry;
392
393  // Nope, create a new Cache entry.  Make sure it is at least 8-byte aligned
394  // so that FileInfo can use the low 3 bits of the pointer for its own
395  // nefarious purposes.
396  unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
397  EntryAlign = std::max(8U, EntryAlign);
398  Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
399  new (Entry) ContentCache(FileEnt);
400  return Entry;
401}
402
403
404/// createMemBufferContentCache - Create a new ContentCache for the specified
405///  memory buffer.  This does no caching.
406const ContentCache*
407SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) {
408  // Add a new ContentCache to the MemBufferInfos list and return it.  Make sure
409  // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of
410  // the pointer for its own nefarious purposes.
411  unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
412  EntryAlign = std::max(8U, EntryAlign);
413  ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
414  new (Entry) ContentCache();
415  MemBufferInfos.push_back(Entry);
416  Entry->setBuffer(Buffer);
417  return Entry;
418}
419
420void SourceManager::PreallocateSLocEntries(ExternalSLocEntrySource *Source,
421                                           unsigned NumSLocEntries,
422                                           unsigned NextOffset) {
423  ExternalSLocEntries = Source;
424  this->NextOffset = NextOffset;
425  SLocEntryLoaded.resize(NumSLocEntries + 1);
426  SLocEntryLoaded[0] = true;
427  SLocEntryTable.resize(SLocEntryTable.size() + NumSLocEntries);
428}
429
430void SourceManager::ClearPreallocatedSLocEntries() {
431  unsigned I = 0;
432  for (unsigned N = SLocEntryLoaded.size(); I != N; ++I)
433    if (!SLocEntryLoaded[I])
434      break;
435
436  // We've already loaded all preallocated source location entries.
437  if (I == SLocEntryLoaded.size())
438    return;
439
440  // Remove everything from location I onward.
441  SLocEntryTable.resize(I);
442  SLocEntryLoaded.clear();
443  ExternalSLocEntries = 0;
444}
445
446
447//===----------------------------------------------------------------------===//
448// Methods to create new FileID's and instantiations.
449//===----------------------------------------------------------------------===//
450
451/// createFileID - Create a new fileID for the specified ContentCache and
452/// include position.  This works regardless of whether the ContentCache
453/// corresponds to a file or some other input source.
454FileID SourceManager::createFileID(const ContentCache *File,
455                                   SourceLocation IncludePos,
456                                   SrcMgr::CharacteristicKind FileCharacter,
457                                   unsigned PreallocatedID,
458                                   unsigned Offset) {
459  if (PreallocatedID) {
460    // If we're filling in a preallocated ID, just load in the file
461    // entry and return.
462    assert(PreallocatedID < SLocEntryLoaded.size() &&
463           "Preallocate ID out-of-range");
464    assert(!SLocEntryLoaded[PreallocatedID] &&
465           "Source location entry already loaded");
466    assert(Offset && "Preallocate source location cannot have zero offset");
467    SLocEntryTable[PreallocatedID]
468      = SLocEntry::get(Offset, FileInfo::get(IncludePos, File, FileCharacter));
469    SLocEntryLoaded[PreallocatedID] = true;
470    FileID FID = FileID::get(PreallocatedID);
471    return FID;
472  }
473
474  SLocEntryTable.push_back(SLocEntry::get(NextOffset,
475                                          FileInfo::get(IncludePos, File,
476                                                        FileCharacter)));
477  unsigned FileSize = File->getSize();
478  assert(NextOffset+FileSize+1 > NextOffset && "Ran out of source locations!");
479  NextOffset += FileSize+1;
480
481  // Set LastFileIDLookup to the newly created file.  The next getFileID call is
482  // almost guaranteed to be from that file.
483  FileID FID = FileID::get(SLocEntryTable.size()-1);
484  return LastFileIDLookup = FID;
485}
486
487/// createInstantiationLoc - Return a new SourceLocation that encodes the fact
488/// that a token from SpellingLoc should actually be referenced from
489/// InstantiationLoc.
490SourceLocation SourceManager::createInstantiationLoc(SourceLocation SpellingLoc,
491                                                     SourceLocation ILocStart,
492                                                     SourceLocation ILocEnd,
493                                                     unsigned TokLength,
494                                                     unsigned PreallocatedID,
495                                                     unsigned Offset) {
496  InstantiationInfo II = InstantiationInfo::get(ILocStart,ILocEnd, SpellingLoc);
497  if (PreallocatedID) {
498    // If we're filling in a preallocated ID, just load in the
499    // instantiation entry and return.
500    assert(PreallocatedID < SLocEntryLoaded.size() &&
501           "Preallocate ID out-of-range");
502    assert(!SLocEntryLoaded[PreallocatedID] &&
503           "Source location entry already loaded");
504    assert(Offset && "Preallocate source location cannot have zero offset");
505    SLocEntryTable[PreallocatedID] = SLocEntry::get(Offset, II);
506    SLocEntryLoaded[PreallocatedID] = true;
507    return SourceLocation::getMacroLoc(Offset);
508  }
509  SLocEntryTable.push_back(SLocEntry::get(NextOffset, II));
510  assert(NextOffset+TokLength+1 > NextOffset && "Ran out of source locations!");
511  NextOffset += TokLength+1;
512  return SourceLocation::getMacroLoc(NextOffset-(TokLength+1));
513}
514
515const llvm::MemoryBuffer *
516SourceManager::getMemoryBufferForFile(const FileEntry *File,
517                                      bool *Invalid) {
518  const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
519  assert(IR && "getOrCreateContentCache() cannot return NULL");
520  return IR->getBuffer(Diag, *this, SourceLocation(), Invalid);
521}
522
523bool SourceManager::overrideFileContents(const FileEntry *SourceFile,
524                                         const llvm::MemoryBuffer *Buffer) {
525  const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile);
526  if (IR == 0)
527    return true;
528
529  const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer);
530  return false;
531}
532
533llvm::StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const {
534  bool MyInvalid = false;
535  const llvm::MemoryBuffer *Buf = getBuffer(FID, &MyInvalid);
536  if (Invalid)
537    *Invalid = MyInvalid;
538
539  if (MyInvalid)
540    return "";
541
542  return Buf->getBuffer();
543}
544
545//===----------------------------------------------------------------------===//
546// SourceLocation manipulation methods.
547//===----------------------------------------------------------------------===//
548
549/// getFileIDSlow - Return the FileID for a SourceLocation.  This is a very hot
550/// method that is used for all SourceManager queries that start with a
551/// SourceLocation object.  It is responsible for finding the entry in
552/// SLocEntryTable which contains the specified location.
553///
554FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const {
555  assert(SLocOffset && "Invalid FileID");
556
557  // After the first and second level caches, I see two common sorts of
558  // behavior: 1) a lot of searched FileID's are "near" the cached file location
559  // or are "near" the cached instantiation location.  2) others are just
560  // completely random and may be a very long way away.
561  //
562  // To handle this, we do a linear search for up to 8 steps to catch #1 quickly
563  // then we fall back to a less cache efficient, but more scalable, binary
564  // search to find the location.
565
566  // See if this is near the file point - worst case we start scanning from the
567  // most newly created FileID.
568  std::vector<SrcMgr::SLocEntry>::const_iterator I;
569
570  if (SLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) {
571    // Neither loc prunes our search.
572    I = SLocEntryTable.end();
573  } else {
574    // Perhaps it is near the file point.
575    I = SLocEntryTable.begin()+LastFileIDLookup.ID;
576  }
577
578  // Find the FileID that contains this.  "I" is an iterator that points to a
579  // FileID whose offset is known to be larger than SLocOffset.
580  unsigned NumProbes = 0;
581  while (1) {
582    --I;
583    if (ExternalSLocEntries)
584      getSLocEntry(FileID::get(I - SLocEntryTable.begin()));
585    if (I->getOffset() <= SLocOffset) {
586#if 0
587      printf("lin %d -> %d [%s] %d %d\n", SLocOffset,
588             I-SLocEntryTable.begin(),
589             I->isInstantiation() ? "inst" : "file",
590             LastFileIDLookup.ID,  int(SLocEntryTable.end()-I));
591#endif
592      FileID Res = FileID::get(I-SLocEntryTable.begin());
593
594      // If this isn't an instantiation, remember it.  We have good locality
595      // across FileID lookups.
596      if (!I->isInstantiation())
597        LastFileIDLookup = Res;
598      NumLinearScans += NumProbes+1;
599      return Res;
600    }
601    if (++NumProbes == 8)
602      break;
603  }
604
605  // Convert "I" back into an index.  We know that it is an entry whose index is
606  // larger than the offset we are looking for.
607  unsigned GreaterIndex = I-SLocEntryTable.begin();
608  // LessIndex - This is the lower bound of the range that we're searching.
609  // We know that the offset corresponding to the FileID is is less than
610  // SLocOffset.
611  unsigned LessIndex = 0;
612  NumProbes = 0;
613  while (1) {
614    unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex;
615    unsigned MidOffset = getSLocEntry(FileID::get(MiddleIndex)).getOffset();
616
617    ++NumProbes;
618
619    // If the offset of the midpoint is too large, chop the high side of the
620    // range to the midpoint.
621    if (MidOffset > SLocOffset) {
622      GreaterIndex = MiddleIndex;
623      continue;
624    }
625
626    // If the middle index contains the value, succeed and return.
627    if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) {
628#if 0
629      printf("bin %d -> %d [%s] %d %d\n", SLocOffset,
630             I-SLocEntryTable.begin(),
631             I->isInstantiation() ? "inst" : "file",
632             LastFileIDLookup.ID, int(SLocEntryTable.end()-I));
633#endif
634      FileID Res = FileID::get(MiddleIndex);
635
636      // If this isn't an instantiation, remember it.  We have good locality
637      // across FileID lookups.
638      if (!I->isInstantiation())
639        LastFileIDLookup = Res;
640      NumBinaryProbes += NumProbes;
641      return Res;
642    }
643
644    // Otherwise, move the low-side up to the middle index.
645    LessIndex = MiddleIndex;
646  }
647}
648
649SourceLocation SourceManager::
650getInstantiationLocSlowCase(SourceLocation Loc) const {
651  do {
652    // Note: If Loc indicates an offset into a token that came from a macro
653    // expansion (e.g. the 5th character of the token) we do not want to add
654    // this offset when going to the instantiation location.  The instatiation
655    // location is the macro invocation, which the offset has nothing to do
656    // with.  This is unlike when we get the spelling loc, because the offset
657    // directly correspond to the token whose spelling we're inspecting.
658    Loc = getSLocEntry(getFileID(Loc)).getInstantiation()
659                   .getInstantiationLocStart();
660  } while (!Loc.isFileID());
661
662  return Loc;
663}
664
665SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
666  do {
667    std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
668    Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc();
669    Loc = Loc.getFileLocWithOffset(LocInfo.second);
670  } while (!Loc.isFileID());
671  return Loc;
672}
673
674
675std::pair<FileID, unsigned>
676SourceManager::getDecomposedInstantiationLocSlowCase(const SrcMgr::SLocEntry *E,
677                                                     unsigned Offset) const {
678  // If this is an instantiation record, walk through all the instantiation
679  // points.
680  FileID FID;
681  SourceLocation Loc;
682  do {
683    Loc = E->getInstantiation().getInstantiationLocStart();
684
685    FID = getFileID(Loc);
686    E = &getSLocEntry(FID);
687    Offset += Loc.getOffset()-E->getOffset();
688  } while (!Loc.isFileID());
689
690  return std::make_pair(FID, Offset);
691}
692
693std::pair<FileID, unsigned>
694SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
695                                                unsigned Offset) const {
696  // If this is an instantiation record, walk through all the instantiation
697  // points.
698  FileID FID;
699  SourceLocation Loc;
700  do {
701    Loc = E->getInstantiation().getSpellingLoc();
702
703    FID = getFileID(Loc);
704    E = &getSLocEntry(FID);
705    Offset += Loc.getOffset()-E->getOffset();
706  } while (!Loc.isFileID());
707
708  return std::make_pair(FID, Offset);
709}
710
711/// getImmediateSpellingLoc - Given a SourceLocation object, return the
712/// spelling location referenced by the ID.  This is the first level down
713/// towards the place where the characters that make up the lexed token can be
714/// found.  This should not generally be used by clients.
715SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
716  if (Loc.isFileID()) return Loc;
717  std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
718  Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc();
719  return Loc.getFileLocWithOffset(LocInfo.second);
720}
721
722
723/// getImmediateInstantiationRange - Loc is required to be an instantiation
724/// location.  Return the start/end of the instantiation information.
725std::pair<SourceLocation,SourceLocation>
726SourceManager::getImmediateInstantiationRange(SourceLocation Loc) const {
727  assert(Loc.isMacroID() && "Not an instantiation loc!");
728  const InstantiationInfo &II = getSLocEntry(getFileID(Loc)).getInstantiation();
729  return II.getInstantiationLocRange();
730}
731
732/// getInstantiationRange - Given a SourceLocation object, return the
733/// range of tokens covered by the instantiation in the ultimate file.
734std::pair<SourceLocation,SourceLocation>
735SourceManager::getInstantiationRange(SourceLocation Loc) const {
736  if (Loc.isFileID()) return std::make_pair(Loc, Loc);
737
738  std::pair<SourceLocation,SourceLocation> Res =
739    getImmediateInstantiationRange(Loc);
740
741  // Fully resolve the start and end locations to their ultimate instantiation
742  // points.
743  while (!Res.first.isFileID())
744    Res.first = getImmediateInstantiationRange(Res.first).first;
745  while (!Res.second.isFileID())
746    Res.second = getImmediateInstantiationRange(Res.second).second;
747  return Res;
748}
749
750
751
752//===----------------------------------------------------------------------===//
753// Queries about the code at a SourceLocation.
754//===----------------------------------------------------------------------===//
755
756/// getCharacterData - Return a pointer to the start of the specified location
757/// in the appropriate MemoryBuffer.
758const char *SourceManager::getCharacterData(SourceLocation SL,
759                                            bool *Invalid) const {
760  // Note that this is a hot function in the getSpelling() path, which is
761  // heavily used by -E mode.
762  std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
763
764  // Note that calling 'getBuffer()' may lazily page in a source file.
765  bool CharDataInvalid = false;
766  const llvm::MemoryBuffer *Buffer
767    = getSLocEntry(LocInfo.first).getFile().getContentCache()
768    ->getBuffer(Diag, *this, SourceLocation(), &CharDataInvalid);
769  if (Invalid)
770    *Invalid = CharDataInvalid;
771  return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
772}
773
774
775/// getColumnNumber - Return the column # for the specified file position.
776/// this is significantly cheaper to compute than the line number.
777unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
778                                        bool *Invalid) const {
779  bool MyInvalid = false;
780  const char *Buf = getBuffer(FID, &MyInvalid)->getBufferStart();
781  if (Invalid)
782    *Invalid = MyInvalid;
783
784  if (MyInvalid)
785    return 1;
786
787  unsigned LineStart = FilePos;
788  while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
789    --LineStart;
790  return FilePos-LineStart+1;
791}
792
793unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
794                                                bool *Invalid) const {
795  if (Loc.isInvalid()) return 0;
796  std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
797  return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
798}
799
800unsigned SourceManager::getInstantiationColumnNumber(SourceLocation Loc,
801                                                     bool *Invalid) const {
802  if (Loc.isInvalid()) return 0;
803  std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
804  return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
805}
806
807static DISABLE_INLINE void
808ComputeLineNumbers(Diagnostic &Diag, ContentCache *FI,
809                   llvm::BumpPtrAllocator &Alloc,
810                   const SourceManager &SM, bool &Invalid);
811static void ComputeLineNumbers(Diagnostic &Diag, ContentCache *FI,
812                               llvm::BumpPtrAllocator &Alloc,
813                               const SourceManager &SM, bool &Invalid) {
814  // Note that calling 'getBuffer()' may lazily page in the file.
815  const MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(),
816                                             &Invalid);
817  if (Invalid)
818    return;
819
820  // Find the file offsets of all of the *physical* source lines.  This does
821  // not look at trigraphs, escaped newlines, or anything else tricky.
822  std::vector<unsigned> LineOffsets;
823
824  // Line #1 starts at char 0.
825  LineOffsets.push_back(0);
826
827  const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
828  const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
829  unsigned Offs = 0;
830  while (1) {
831    // Skip over the contents of the line.
832    // TODO: Vectorize this?  This is very performance sensitive for programs
833    // with lots of diagnostics and in -E mode.
834    const unsigned char *NextBuf = (const unsigned char *)Buf;
835    while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
836      ++NextBuf;
837    Offs += NextBuf-Buf;
838    Buf = NextBuf;
839
840    if (Buf[0] == '\n' || Buf[0] == '\r') {
841      // If this is \n\r or \r\n, skip both characters.
842      if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1])
843        ++Offs, ++Buf;
844      ++Offs, ++Buf;
845      LineOffsets.push_back(Offs);
846    } else {
847      // Otherwise, this is a null.  If end of file, exit.
848      if (Buf == End) break;
849      // Otherwise, skip the null.
850      ++Offs, ++Buf;
851    }
852  }
853
854  // Copy the offsets into the FileInfo structure.
855  FI->NumLines = LineOffsets.size();
856  FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
857  std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
858}
859
860/// getLineNumber - Given a SourceLocation, return the spelling line number
861/// for the position indicated.  This requires building and caching a table of
862/// line offsets for the MemoryBuffer, so this is not cheap: use only when
863/// about to emit a diagnostic.
864unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
865                                      bool *Invalid) const {
866  ContentCache *Content;
867  if (LastLineNoFileIDQuery == FID)
868    Content = LastLineNoContentCache;
869  else
870    Content = const_cast<ContentCache*>(getSLocEntry(FID)
871                                        .getFile().getContentCache());
872
873  // If this is the first use of line information for this buffer, compute the
874  /// SourceLineCache for it on demand.
875  if (Content->SourceLineCache == 0) {
876    bool MyInvalid = false;
877    ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
878    if (Invalid)
879      *Invalid = MyInvalid;
880    if (MyInvalid)
881      return 1;
882  } else if (Invalid)
883    *Invalid = false;
884
885  // Okay, we know we have a line number table.  Do a binary search to find the
886  // line number that this character position lands on.
887  unsigned *SourceLineCache = Content->SourceLineCache;
888  unsigned *SourceLineCacheStart = SourceLineCache;
889  unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
890
891  unsigned QueriedFilePos = FilePos+1;
892
893  // FIXME: I would like to be convinced that this code is worth being as
894  // complicated as it is, binary search isn't that slow.
895  //
896  // If it is worth being optimized, then in my opinion it could be more
897  // performant, simpler, and more obviously correct by just "galloping" outward
898  // from the queried file position. In fact, this could be incorporated into a
899  // generic algorithm such as lower_bound_with_hint.
900  //
901  // If someone gives me a test case where this matters, and I will do it! - DWD
902
903  // If the previous query was to the same file, we know both the file pos from
904  // that query and the line number returned.  This allows us to narrow the
905  // search space from the entire file to something near the match.
906  if (LastLineNoFileIDQuery == FID) {
907    if (QueriedFilePos >= LastLineNoFilePos) {
908      // FIXME: Potential overflow?
909      SourceLineCache = SourceLineCache+LastLineNoResult-1;
910
911      // The query is likely to be nearby the previous one.  Here we check to
912      // see if it is within 5, 10 or 20 lines.  It can be far away in cases
913      // where big comment blocks and vertical whitespace eat up lines but
914      // contribute no tokens.
915      if (SourceLineCache+5 < SourceLineCacheEnd) {
916        if (SourceLineCache[5] > QueriedFilePos)
917          SourceLineCacheEnd = SourceLineCache+5;
918        else if (SourceLineCache+10 < SourceLineCacheEnd) {
919          if (SourceLineCache[10] > QueriedFilePos)
920            SourceLineCacheEnd = SourceLineCache+10;
921          else if (SourceLineCache+20 < SourceLineCacheEnd) {
922            if (SourceLineCache[20] > QueriedFilePos)
923              SourceLineCacheEnd = SourceLineCache+20;
924          }
925        }
926      }
927    } else {
928      if (LastLineNoResult < Content->NumLines)
929        SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
930    }
931  }
932
933  // If the spread is large, do a "radix" test as our initial guess, based on
934  // the assumption that lines average to approximately the same length.
935  // NOTE: This is currently disabled, as it does not appear to be profitable in
936  // initial measurements.
937  if (0 && SourceLineCacheEnd-SourceLineCache > 20) {
938    unsigned FileLen = Content->SourceLineCache[Content->NumLines-1];
939
940    // Take a stab at guessing where it is.
941    unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen;
942
943    // Check for -10 and +10 lines.
944    unsigned LowerBound = std::max(int(ApproxPos-10), 0);
945    unsigned UpperBound = std::min(ApproxPos+10, FileLen);
946
947    // If the computed lower bound is less than the query location, move it in.
948    if (SourceLineCache < SourceLineCacheStart+LowerBound &&
949        SourceLineCacheStart[LowerBound] < QueriedFilePos)
950      SourceLineCache = SourceLineCacheStart+LowerBound;
951
952    // If the computed upper bound is greater than the query location, move it.
953    if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound &&
954        SourceLineCacheStart[UpperBound] >= QueriedFilePos)
955      SourceLineCacheEnd = SourceLineCacheStart+UpperBound;
956  }
957
958  unsigned *Pos
959    = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
960  unsigned LineNo = Pos-SourceLineCacheStart;
961
962  LastLineNoFileIDQuery = FID;
963  LastLineNoContentCache = Content;
964  LastLineNoFilePos = QueriedFilePos;
965  LastLineNoResult = LineNo;
966  return LineNo;
967}
968
969unsigned SourceManager::getInstantiationLineNumber(SourceLocation Loc,
970                                                   bool *Invalid) const {
971  if (Loc.isInvalid()) return 0;
972  std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
973  return getLineNumber(LocInfo.first, LocInfo.second);
974}
975unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
976                                              bool *Invalid) const {
977  if (Loc.isInvalid()) return 0;
978  std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
979  return getLineNumber(LocInfo.first, LocInfo.second);
980}
981
982/// getFileCharacteristic - return the file characteristic of the specified
983/// source location, indicating whether this is a normal file, a system
984/// header, or an "implicit extern C" system header.
985///
986/// This state can be modified with flags on GNU linemarker directives like:
987///   # 4 "foo.h" 3
988/// which changes all source locations in the current file after that to be
989/// considered to be from a system header.
990SrcMgr::CharacteristicKind
991SourceManager::getFileCharacteristic(SourceLocation Loc) const {
992  assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!");
993  std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
994  const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile();
995
996  // If there are no #line directives in this file, just return the whole-file
997  // state.
998  if (!FI.hasLineDirectives())
999    return FI.getFileCharacteristic();
1000
1001  assert(LineTable && "Can't have linetable entries without a LineTable!");
1002  // See if there is a #line directive before the location.
1003  const LineEntry *Entry =
1004    LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second);
1005
1006  // If this is before the first line marker, use the file characteristic.
1007  if (!Entry)
1008    return FI.getFileCharacteristic();
1009
1010  return Entry->FileKind;
1011}
1012
1013/// Return the filename or buffer identifier of the buffer the location is in.
1014/// Note that this name does not respect #line directives.  Use getPresumedLoc
1015/// for normal clients.
1016const char *SourceManager::getBufferName(SourceLocation Loc,
1017                                         bool *Invalid) const {
1018  if (Loc.isInvalid()) return "<invalid loc>";
1019
1020  return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
1021}
1022
1023
1024/// getPresumedLoc - This method returns the "presumed" location of a
1025/// SourceLocation specifies.  A "presumed location" can be modified by #line
1026/// or GNU line marker directives.  This provides a view on the data that a
1027/// user should see in diagnostics, for example.
1028///
1029/// Note that a presumed location is always given as the instantiation point
1030/// of an instantiation location, not at the spelling location.
1031PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc) const {
1032  if (Loc.isInvalid()) return PresumedLoc();
1033
1034  // Presumed locations are always for instantiation points.
1035  std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc);
1036
1037  const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile();
1038  const SrcMgr::ContentCache *C = FI.getContentCache();
1039
1040  // To get the source name, first consult the FileEntry (if one exists)
1041  // before the MemBuffer as this will avoid unnecessarily paging in the
1042  // MemBuffer.
1043  const char *Filename;
1044  if (C->Entry)
1045    Filename = C->Entry->getName();
1046  else
1047    Filename = C->getBuffer(Diag, *this)->getBufferIdentifier();
1048  unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second);
1049  unsigned ColNo  = getColumnNumber(LocInfo.first, LocInfo.second);
1050  SourceLocation IncludeLoc = FI.getIncludeLoc();
1051
1052  // If we have #line directives in this file, update and overwrite the physical
1053  // location info if appropriate.
1054  if (FI.hasLineDirectives()) {
1055    assert(LineTable && "Can't have linetable entries without a LineTable!");
1056    // See if there is a #line directive before this.  If so, get it.
1057    if (const LineEntry *Entry =
1058          LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second)) {
1059      // If the LineEntry indicates a filename, use it.
1060      if (Entry->FilenameID != -1)
1061        Filename = LineTable->getFilename(Entry->FilenameID);
1062
1063      // Use the line number specified by the LineEntry.  This line number may
1064      // be multiple lines down from the line entry.  Add the difference in
1065      // physical line numbers from the query point and the line marker to the
1066      // total.
1067      unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
1068      LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
1069
1070      // Note that column numbers are not molested by line markers.
1071
1072      // Handle virtual #include manipulation.
1073      if (Entry->IncludeOffset) {
1074        IncludeLoc = getLocForStartOfFile(LocInfo.first);
1075        IncludeLoc = IncludeLoc.getFileLocWithOffset(Entry->IncludeOffset);
1076      }
1077    }
1078  }
1079
1080  return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc);
1081}
1082
1083//===----------------------------------------------------------------------===//
1084// Other miscellaneous methods.
1085//===----------------------------------------------------------------------===//
1086
1087/// \brief Get the source location for the given file:line:col triplet.
1088///
1089/// If the source file is included multiple times, the source location will
1090/// be based upon the first inclusion.
1091SourceLocation SourceManager::getLocation(const FileEntry *SourceFile,
1092                                          unsigned Line, unsigned Col) const {
1093  assert(SourceFile && "Null source file!");
1094  assert(Line && Col && "Line and column should start from 1!");
1095
1096  fileinfo_iterator FI = FileInfos.find(SourceFile);
1097  if (FI == FileInfos.end())
1098    return SourceLocation();
1099  ContentCache *Content = FI->second;
1100
1101  // If this is the first use of line information for this buffer, compute the
1102  /// SourceLineCache for it on demand.
1103  if (Content->SourceLineCache == 0) {
1104    bool MyInvalid = false;
1105    ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
1106    if (MyInvalid)
1107      return SourceLocation();
1108  }
1109
1110  // Find the first file ID that corresponds to the given file.
1111  FileID FirstFID;
1112
1113  // First, check the main file ID, since it is common to look for a
1114  // location in the main file.
1115  if (!MainFileID.isInvalid()) {
1116    const SLocEntry &MainSLoc = getSLocEntry(MainFileID);
1117    if (MainSLoc.isFile() && MainSLoc.getFile().getContentCache() == Content)
1118      FirstFID = MainFileID;
1119  }
1120
1121  if (FirstFID.isInvalid()) {
1122    // The location we're looking for isn't in the main file; look
1123    // through all of the source locations.
1124    for (unsigned I = 0, N = sloc_entry_size(); I != N; ++I) {
1125      const SLocEntry &SLoc = getSLocEntry(I);
1126      if (SLoc.isFile() && SLoc.getFile().getContentCache() == Content) {
1127        FirstFID = FileID::get(I);
1128        break;
1129      }
1130    }
1131  }
1132
1133  if (FirstFID.isInvalid())
1134    return SourceLocation();
1135
1136  if (Line > Content->NumLines) {
1137    unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize();
1138    if (Size > 0)
1139      --Size;
1140    return getLocForStartOfFile(FirstFID).getFileLocWithOffset(Size);
1141  }
1142
1143  unsigned FilePos = Content->SourceLineCache[Line - 1];
1144  const char *Buf = Content->getBuffer(Diag, *this)->getBufferStart() + FilePos;
1145  unsigned BufLength = Content->getBuffer(Diag, *this)->getBufferEnd() - Buf;
1146  unsigned i = 0;
1147
1148  // Check that the given column is valid.
1149  while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
1150    ++i;
1151  if (i < Col-1)
1152    return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + i);
1153
1154  return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + Col - 1);
1155}
1156
1157/// \brief Determines the order of 2 source locations in the translation unit.
1158///
1159/// \returns true if LHS source location comes before RHS, false otherwise.
1160bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
1161                                              SourceLocation RHS) const {
1162  assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
1163  if (LHS == RHS)
1164    return false;
1165
1166  std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
1167  std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
1168
1169  // If the source locations are in the same file, just compare offsets.
1170  if (LOffs.first == ROffs.first)
1171    return LOffs.second < ROffs.second;
1172
1173  // If we are comparing a source location with multiple locations in the same
1174  // file, we get a big win by caching the result.
1175  if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first))
1176    return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1177
1178  // Okay, we missed in the cache, start updating the cache for this query.
1179  IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first);
1180
1181  // "Traverse" the include/instantiation stacks of both locations and try to
1182  // find a common "ancestor".  FileIDs build a tree-like structure that
1183  // reflects the #include hierarchy, and this algorithm needs to find the
1184  // nearest common ancestor between the two locations.  For example, if you
1185  // have a.c that includes b.h and c.h, and are comparing a location in b.h to
1186  // a location in c.h, we need to find that their nearest common ancestor is
1187  // a.c, and compare the locations of the two #includes to find their relative
1188  // ordering.
1189  //
1190  // SourceManager assigns FileIDs in order of parsing.  This means that an
1191  // includee always has a larger FileID than an includer.  While you might
1192  // think that we could just compare the FileID's here, that doesn't work to
1193  // compare a point at the end of a.c with a point within c.h.  Though c.h has
1194  // a larger FileID, we have to compare the include point of c.h to the
1195  // location in a.c.
1196  //
1197  // Despite not being able to directly compare FileID's, we can tell that a
1198  // larger FileID is necessarily more deeply nested than a lower one and use
1199  // this information to walk up the tree to the nearest common ancestor.
1200  do {
1201    // If LOffs is larger than ROffs, then LOffs must be more deeply nested than
1202    // ROffs, walk up the #include chain.
1203    if (LOffs.first.ID > ROffs.first.ID) {
1204      SourceLocation UpperLoc;
1205      const SrcMgr::SLocEntry &Entry = getSLocEntry(LOffs.first);
1206      if (Entry.isInstantiation())
1207        UpperLoc = Entry.getInstantiation().getInstantiationLocStart();
1208      else
1209        UpperLoc = Entry.getFile().getIncludeLoc();
1210
1211      if (UpperLoc.isInvalid())
1212        break; // We reached the top.
1213
1214      LOffs = getDecomposedLoc(UpperLoc);
1215    } else {
1216      // Otherwise, ROffs is larger than LOffs, so ROffs must be more deeply
1217      // nested than LOffs, walk up the #include chain.
1218      SourceLocation UpperLoc;
1219      const SrcMgr::SLocEntry &Entry = getSLocEntry(ROffs.first);
1220      if (Entry.isInstantiation())
1221        UpperLoc = Entry.getInstantiation().getInstantiationLocStart();
1222      else
1223        UpperLoc = Entry.getFile().getIncludeLoc();
1224
1225      if (UpperLoc.isInvalid())
1226        break; // We reached the top.
1227
1228      ROffs = getDecomposedLoc(UpperLoc);
1229    }
1230  } while (LOffs.first != ROffs.first);
1231
1232  // If we exited because we found a nearest common ancestor, compare the
1233  // locations within the common file and cache them.
1234  if (LOffs.first == ROffs.first) {
1235    IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second);
1236    return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
1237  }
1238
1239  // There is no common ancestor, most probably because one location is in the
1240  // predefines buffer.
1241  IsBeforeInTUCache.setQueryFIDs(FileID(), FileID()); // Don't try caching.
1242
1243  // FIXME: We should rearrange the external interface so this simply never
1244  // happens; it can't conceptually happen. Also see PR5662.
1245
1246  // If exactly one location is a memory buffer, assume it preceeds the other.
1247  bool LIsMB = !getSLocEntry(LOffs.first).getFile().getContentCache()->Entry;
1248  bool RIsMB = !getSLocEntry(ROffs.first).getFile().getContentCache()->Entry;
1249  if (LIsMB != RIsMB)
1250    return LIsMB;
1251
1252  // Otherwise, just assume FileIDs were created in order.
1253  return LOffs.first < ROffs.first;
1254}
1255
1256/// PrintStats - Print statistics to stderr.
1257///
1258void SourceManager::PrintStats() const {
1259  llvm::errs() << "\n*** Source Manager Stats:\n";
1260  llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
1261               << " mem buffers mapped.\n";
1262  llvm::errs() << SLocEntryTable.size() << " SLocEntry's allocated, "
1263               << NextOffset << "B of Sloc address space used.\n";
1264
1265  unsigned NumLineNumsComputed = 0;
1266  unsigned NumFileBytesMapped = 0;
1267  for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
1268    NumLineNumsComputed += I->second->SourceLineCache != 0;
1269    NumFileBytesMapped  += I->second->getSizeBytesMapped();
1270  }
1271
1272  llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
1273               << NumLineNumsComputed << " files with line #'s computed.\n";
1274  llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
1275               << NumBinaryProbes << " binary.\n";
1276}
1277
1278ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
1279