Home | History | Annotate | Download | only in Basic
      1 //===--- SourceManager.cpp - Track and cache source files -----------------===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 //
     10 //  This file implements the SourceManager interface.
     11 //
     12 //===----------------------------------------------------------------------===//
     13 
     14 #include "clang/Basic/SourceManager.h"
     15 #include "clang/Basic/SourceManagerInternals.h"
     16 #include "clang/Basic/Diagnostic.h"
     17 #include "clang/Basic/FileManager.h"
     18 #include "llvm/ADT/StringSwitch.h"
     19 #include "llvm/ADT/Optional.h"
     20 #include "llvm/ADT/STLExtras.h"
     21 #include "llvm/Support/Compiler.h"
     22 #include "llvm/Support/MemoryBuffer.h"
     23 #include "llvm/Support/raw_ostream.h"
     24 #include "llvm/Support/Path.h"
     25 #include "llvm/Support/Capacity.h"
     26 #include <algorithm>
     27 #include <string>
     28 #include <cstring>
     29 #include <sys/stat.h>
     30 
     31 using namespace clang;
     32 using namespace SrcMgr;
     33 using llvm::MemoryBuffer;
     34 
     35 //===----------------------------------------------------------------------===//
     36 // SourceManager Helper Classes
     37 //===----------------------------------------------------------------------===//
     38 
     39 ContentCache::~ContentCache() {
     40   if (shouldFreeBuffer())
     41     delete Buffer.getPointer();
     42 }
     43 
     44 /// getSizeBytesMapped - Returns the number of bytes actually mapped for this
     45 /// ContentCache. This can be 0 if the MemBuffer was not actually expanded.
     46 unsigned ContentCache::getSizeBytesMapped() const {
     47   return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0;
     48 }
     49 
     50 /// Returns the kind of memory used to back the memory buffer for
     51 /// this content cache.  This is used for performance analysis.
     52 llvm::MemoryBuffer::BufferKind ContentCache::getMemoryBufferKind() const {
     53   assert(Buffer.getPointer());
     54 
     55   // Should be unreachable, but keep for sanity.
     56   if (!Buffer.getPointer())
     57     return llvm::MemoryBuffer::MemoryBuffer_Malloc;
     58 
     59   const llvm::MemoryBuffer *buf = Buffer.getPointer();
     60   return buf->getBufferKind();
     61 }
     62 
     63 /// getSize - Returns the size of the content encapsulated by this ContentCache.
     64 ///  This can be the size of the source file or the size of an arbitrary
     65 ///  scratch buffer.  If the ContentCache encapsulates a source file, that
     66 ///  file is not lazily brought in from disk to satisfy this query.
     67 unsigned ContentCache::getSize() const {
     68   return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize()
     69                              : (unsigned) ContentsEntry->getSize();
     70 }
     71 
     72 void ContentCache::replaceBuffer(const llvm::MemoryBuffer *B,
     73                                  bool DoNotFree) {
     74   if (B == Buffer.getPointer()) {
     75     assert(0 && "Replacing with the same buffer");
     76     Buffer.setInt(DoNotFree? DoNotFreeFlag : 0);
     77     return;
     78   }
     79 
     80   if (shouldFreeBuffer())
     81     delete Buffer.getPointer();
     82   Buffer.setPointer(B);
     83   Buffer.setInt(DoNotFree? DoNotFreeFlag : 0);
     84 }
     85 
     86 const llvm::MemoryBuffer *ContentCache::getBuffer(DiagnosticsEngine &Diag,
     87                                                   const SourceManager &SM,
     88                                                   SourceLocation Loc,
     89                                                   bool *Invalid) const {
     90   // Lazily create the Buffer for ContentCaches that wrap files.  If we already
     91   // computed it, just return what we have.
     92   if (Buffer.getPointer() || ContentsEntry == 0) {
     93     if (Invalid)
     94       *Invalid = isBufferInvalid();
     95 
     96     return Buffer.getPointer();
     97   }
     98 
     99   std::string ErrorStr;
    100   Buffer.setPointer(SM.getFileManager().getBufferForFile(ContentsEntry, &ErrorStr));
    101 
    102   // If we were unable to open the file, then we are in an inconsistent
    103   // situation where the content cache referenced a file which no longer
    104   // exists. Most likely, we were using a stat cache with an invalid entry but
    105   // the file could also have been removed during processing. Since we can't
    106   // really deal with this situation, just create an empty buffer.
    107   //
    108   // FIXME: This is definitely not ideal, but our immediate clients can't
    109   // currently handle returning a null entry here. Ideally we should detect
    110   // that we are in an inconsistent situation and error out as quickly as
    111   // possible.
    112   if (!Buffer.getPointer()) {
    113     const StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
    114     Buffer.setPointer(MemoryBuffer::getNewMemBuffer(ContentsEntry->getSize(),
    115                                                     "<invalid>"));
    116     char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
    117     for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i)
    118       Ptr[i] = FillStr[i % FillStr.size()];
    119 
    120     if (Diag.isDiagnosticInFlight())
    121       Diag.SetDelayedDiagnostic(diag::err_cannot_open_file,
    122                                 ContentsEntry->getName(), ErrorStr);
    123     else
    124       Diag.Report(Loc, diag::err_cannot_open_file)
    125         << ContentsEntry->getName() << ErrorStr;
    126 
    127     Buffer.setInt(Buffer.getInt() | InvalidFlag);
    128 
    129     if (Invalid) *Invalid = true;
    130     return Buffer.getPointer();
    131   }
    132 
    133   // Check that the file's size is the same as in the file entry (which may
    134   // have come from a stat cache).
    135   if (getRawBuffer()->getBufferSize() != (size_t)ContentsEntry->getSize()) {
    136     if (Diag.isDiagnosticInFlight())
    137       Diag.SetDelayedDiagnostic(diag::err_file_modified,
    138                                 ContentsEntry->getName());
    139     else
    140       Diag.Report(Loc, diag::err_file_modified)
    141         << ContentsEntry->getName();
    142 
    143     Buffer.setInt(Buffer.getInt() | InvalidFlag);
    144     if (Invalid) *Invalid = true;
    145     return Buffer.getPointer();
    146   }
    147 
    148   // If the buffer is valid, check to see if it has a UTF Byte Order Mark
    149   // (BOM).  We only support UTF-8 with and without a BOM right now.  See
    150   // http://en.wikipedia.org/wiki/Byte_order_mark for more information.
    151   StringRef BufStr = Buffer.getPointer()->getBuffer();
    152   const char *InvalidBOM = llvm::StringSwitch<const char *>(BufStr)
    153     .StartsWith("\xFE\xFF", "UTF-16 (BE)")
    154     .StartsWith("\xFF\xFE", "UTF-16 (LE)")
    155     .StartsWith("\x00\x00\xFE\xFF", "UTF-32 (BE)")
    156     .StartsWith("\xFF\xFE\x00\x00", "UTF-32 (LE)")
    157     .StartsWith("\x2B\x2F\x76", "UTF-7")
    158     .StartsWith("\xF7\x64\x4C", "UTF-1")
    159     .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC")
    160     .StartsWith("\x0E\xFE\xFF", "SDSU")
    161     .StartsWith("\xFB\xEE\x28", "BOCU-1")
    162     .StartsWith("\x84\x31\x95\x33", "GB-18030")
    163     .Default(0);
    164 
    165   if (InvalidBOM) {
    166     Diag.Report(Loc, diag::err_unsupported_bom)
    167       << InvalidBOM << ContentsEntry->getName();
    168     Buffer.setInt(Buffer.getInt() | InvalidFlag);
    169   }
    170 
    171   if (Invalid)
    172     *Invalid = isBufferInvalid();
    173 
    174   return Buffer.getPointer();
    175 }
    176 
    177 unsigned LineTableInfo::getLineTableFilenameID(StringRef Name) {
    178   // Look up the filename in the string table, returning the pre-existing value
    179   // if it exists.
    180   llvm::StringMapEntry<unsigned> &Entry =
    181     FilenameIDs.GetOrCreateValue(Name, ~0U);
    182   if (Entry.getValue() != ~0U)
    183     return Entry.getValue();
    184 
    185   // Otherwise, assign this the next available ID.
    186   Entry.setValue(FilenamesByID.size());
    187   FilenamesByID.push_back(&Entry);
    188   return FilenamesByID.size()-1;
    189 }
    190 
    191 /// AddLineNote - Add a line note to the line table that indicates that there
    192 /// is a #line at the specified FID/Offset location which changes the presumed
    193 /// location to LineNo/FilenameID.
    194 void LineTableInfo::AddLineNote(int FID, unsigned Offset,
    195                                 unsigned LineNo, int FilenameID) {
    196   std::vector<LineEntry> &Entries = LineEntries[FID];
    197 
    198   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
    199          "Adding line entries out of order!");
    200 
    201   SrcMgr::CharacteristicKind Kind = SrcMgr::C_User;
    202   unsigned IncludeOffset = 0;
    203 
    204   if (!Entries.empty()) {
    205     // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember
    206     // that we are still in "foo.h".
    207     if (FilenameID == -1)
    208       FilenameID = Entries.back().FilenameID;
    209 
    210     // If we are after a line marker that switched us to system header mode, or
    211     // that set #include information, preserve it.
    212     Kind = Entries.back().FileKind;
    213     IncludeOffset = Entries.back().IncludeOffset;
    214   }
    215 
    216   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind,
    217                                    IncludeOffset));
    218 }
    219 
    220 /// AddLineNote This is the same as the previous version of AddLineNote, but is
    221 /// used for GNU line markers.  If EntryExit is 0, then this doesn't change the
    222 /// presumed #include stack.  If it is 1, this is a file entry, if it is 2 then
    223 /// this is a file exit.  FileKind specifies whether this is a system header or
    224 /// extern C system header.
    225 void LineTableInfo::AddLineNote(int FID, unsigned Offset,
    226                                 unsigned LineNo, int FilenameID,
    227                                 unsigned EntryExit,
    228                                 SrcMgr::CharacteristicKind FileKind) {
    229   assert(FilenameID != -1 && "Unspecified filename should use other accessor");
    230 
    231   std::vector<LineEntry> &Entries = LineEntries[FID];
    232 
    233   assert((Entries.empty() || Entries.back().FileOffset < Offset) &&
    234          "Adding line entries out of order!");
    235 
    236   unsigned IncludeOffset = 0;
    237   if (EntryExit == 0) {  // No #include stack change.
    238     IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset;
    239   } else if (EntryExit == 1) {
    240     IncludeOffset = Offset-1;
    241   } else if (EntryExit == 2) {
    242     assert(!Entries.empty() && Entries.back().IncludeOffset &&
    243        "PPDirectives should have caught case when popping empty include stack");
    244 
    245     // Get the include loc of the last entries' include loc as our include loc.
    246     IncludeOffset = 0;
    247     if (const LineEntry *PrevEntry =
    248           FindNearestLineEntry(FID, Entries.back().IncludeOffset))
    249       IncludeOffset = PrevEntry->IncludeOffset;
    250   }
    251 
    252   Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind,
    253                                    IncludeOffset));
    254 }
    255 
    256 
    257 /// FindNearestLineEntry - Find the line entry nearest to FID that is before
    258 /// it.  If there is no line entry before Offset in FID, return null.
    259 const LineEntry *LineTableInfo::FindNearestLineEntry(int FID,
    260                                                      unsigned Offset) {
    261   const std::vector<LineEntry> &Entries = LineEntries[FID];
    262   assert(!Entries.empty() && "No #line entries for this FID after all!");
    263 
    264   // It is very common for the query to be after the last #line, check this
    265   // first.
    266   if (Entries.back().FileOffset <= Offset)
    267     return &Entries.back();
    268 
    269   // Do a binary search to find the maximal element that is still before Offset.
    270   std::vector<LineEntry>::const_iterator I =
    271     std::upper_bound(Entries.begin(), Entries.end(), Offset);
    272   if (I == Entries.begin()) return 0;
    273   return &*--I;
    274 }
    275 
    276 /// \brief Add a new line entry that has already been encoded into
    277 /// the internal representation of the line table.
    278 void LineTableInfo::AddEntry(int FID,
    279                              const std::vector<LineEntry> &Entries) {
    280   LineEntries[FID] = Entries;
    281 }
    282 
    283 /// getLineTableFilenameID - Return the uniqued ID for the specified filename.
    284 ///
    285 unsigned SourceManager::getLineTableFilenameID(StringRef Name) {
    286   if (LineTable == 0)
    287     LineTable = new LineTableInfo();
    288   return LineTable->getLineTableFilenameID(Name);
    289 }
    290 
    291 
    292 /// AddLineNote - Add a line note to the line table for the FileID and offset
    293 /// specified by Loc.  If FilenameID is -1, it is considered to be
    294 /// unspecified.
    295 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
    296                                 int FilenameID) {
    297   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
    298 
    299   bool Invalid = false;
    300   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
    301   if (!Entry.isFile() || Invalid)
    302     return;
    303 
    304   const SrcMgr::FileInfo &FileInfo = Entry.getFile();
    305 
    306   // Remember that this file has #line directives now if it doesn't already.
    307   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
    308 
    309   if (LineTable == 0)
    310     LineTable = new LineTableInfo();
    311   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID);
    312 }
    313 
    314 /// AddLineNote - Add a GNU line marker to the line table.
    315 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo,
    316                                 int FilenameID, bool IsFileEntry,
    317                                 bool IsFileExit, bool IsSystemHeader,
    318                                 bool IsExternCHeader) {
    319   // If there is no filename and no flags, this is treated just like a #line,
    320   // which does not change the flags of the previous line marker.
    321   if (FilenameID == -1) {
    322     assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader &&
    323            "Can't set flags without setting the filename!");
    324     return AddLineNote(Loc, LineNo, FilenameID);
    325   }
    326 
    327   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
    328 
    329   bool Invalid = false;
    330   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
    331   if (!Entry.isFile() || Invalid)
    332     return;
    333 
    334   const SrcMgr::FileInfo &FileInfo = Entry.getFile();
    335 
    336   // Remember that this file has #line directives now if it doesn't already.
    337   const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives();
    338 
    339   if (LineTable == 0)
    340     LineTable = new LineTableInfo();
    341 
    342   SrcMgr::CharacteristicKind FileKind;
    343   if (IsExternCHeader)
    344     FileKind = SrcMgr::C_ExternCSystem;
    345   else if (IsSystemHeader)
    346     FileKind = SrcMgr::C_System;
    347   else
    348     FileKind = SrcMgr::C_User;
    349 
    350   unsigned EntryExit = 0;
    351   if (IsFileEntry)
    352     EntryExit = 1;
    353   else if (IsFileExit)
    354     EntryExit = 2;
    355 
    356   LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID,
    357                          EntryExit, FileKind);
    358 }
    359 
    360 LineTableInfo &SourceManager::getLineTable() {
    361   if (LineTable == 0)
    362     LineTable = new LineTableInfo();
    363   return *LineTable;
    364 }
    365 
    366 //===----------------------------------------------------------------------===//
    367 // Private 'Create' methods.
    368 //===----------------------------------------------------------------------===//
    369 
    370 SourceManager::SourceManager(DiagnosticsEngine &Diag, FileManager &FileMgr)
    371   : Diag(Diag), FileMgr(FileMgr), OverridenFilesKeepOriginalName(true),
    372     ExternalSLocEntries(0), LineTable(0), NumLinearScans(0),
    373     NumBinaryProbes(0), FakeBufferForRecovery(0),
    374     FakeContentCacheForRecovery(0) {
    375   clearIDTables();
    376   Diag.setSourceManager(this);
    377 }
    378 
    379 SourceManager::~SourceManager() {
    380   delete LineTable;
    381 
    382   // Delete FileEntry objects corresponding to content caches.  Since the actual
    383   // content cache objects are bump pointer allocated, we just have to run the
    384   // dtors, but we call the deallocate method for completeness.
    385   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) {
    386     if (MemBufferInfos[i]) {
    387       MemBufferInfos[i]->~ContentCache();
    388       ContentCacheAlloc.Deallocate(MemBufferInfos[i]);
    389     }
    390   }
    391   for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator
    392        I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) {
    393     if (I->second) {
    394       I->second->~ContentCache();
    395       ContentCacheAlloc.Deallocate(I->second);
    396     }
    397   }
    398 
    399   delete FakeBufferForRecovery;
    400   delete FakeContentCacheForRecovery;
    401 
    402   for (llvm::DenseMap<FileID, MacroArgsMap *>::iterator
    403          I = MacroArgsCacheMap.begin(),E = MacroArgsCacheMap.end(); I!=E; ++I) {
    404     delete I->second;
    405   }
    406 }
    407 
    408 void SourceManager::clearIDTables() {
    409   MainFileID = FileID();
    410   LocalSLocEntryTable.clear();
    411   LoadedSLocEntryTable.clear();
    412   SLocEntryLoaded.clear();
    413   LastLineNoFileIDQuery = FileID();
    414   LastLineNoContentCache = 0;
    415   LastFileIDLookup = FileID();
    416 
    417   if (LineTable)
    418     LineTable->clear();
    419 
    420   // Use up FileID #0 as an invalid expansion.
    421   NextLocalOffset = 0;
    422   CurrentLoadedOffset = MaxLoadedOffset;
    423   createExpansionLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1);
    424 }
    425 
    426 /// getOrCreateContentCache - Create or return a cached ContentCache for the
    427 /// specified file.
    428 const ContentCache *
    429 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt) {
    430   assert(FileEnt && "Didn't specify a file entry to use?");
    431 
    432   // Do we already have information about this file?
    433   ContentCache *&Entry = FileInfos[FileEnt];
    434   if (Entry) return Entry;
    435 
    436   // Nope, create a new Cache entry.  Make sure it is at least 8-byte aligned
    437   // so that FileInfo can use the low 3 bits of the pointer for its own
    438   // nefarious purposes.
    439   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
    440   EntryAlign = std::max(8U, EntryAlign);
    441   Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
    442 
    443   // If the file contents are overridden with contents from another file,
    444   // pass that file to ContentCache.
    445   llvm::DenseMap<const FileEntry *, const FileEntry *>::iterator
    446       overI = OverriddenFiles.find(FileEnt);
    447   if (overI == OverriddenFiles.end())
    448     new (Entry) ContentCache(FileEnt);
    449   else
    450     new (Entry) ContentCache(OverridenFilesKeepOriginalName ? FileEnt
    451                                                             : overI->second,
    452                              overI->second);
    453 
    454   return Entry;
    455 }
    456 
    457 
    458 /// createMemBufferContentCache - Create a new ContentCache for the specified
    459 ///  memory buffer.  This does no caching.
    460 const ContentCache*
    461 SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) {
    462   // Add a new ContentCache to the MemBufferInfos list and return it.  Make sure
    463   // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of
    464   // the pointer for its own nefarious purposes.
    465   unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment;
    466   EntryAlign = std::max(8U, EntryAlign);
    467   ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign);
    468   new (Entry) ContentCache();
    469   MemBufferInfos.push_back(Entry);
    470   Entry->setBuffer(Buffer);
    471   return Entry;
    472 }
    473 
    474 const SrcMgr::SLocEntry &SourceManager::loadSLocEntry(unsigned Index,
    475                                                       bool *Invalid) const {
    476   assert(!SLocEntryLoaded[Index]);
    477   if (ExternalSLocEntries->ReadSLocEntry(-(static_cast<int>(Index) + 2))) {
    478     if (Invalid)
    479       *Invalid = true;
    480     // If the file of the SLocEntry changed we could still have loaded it.
    481     if (!SLocEntryLoaded[Index]) {
    482       // Try to recover; create a SLocEntry so the rest of clang can handle it.
    483       LoadedSLocEntryTable[Index] = SLocEntry::get(0,
    484                                  FileInfo::get(SourceLocation(),
    485                                                getFakeContentCacheForRecovery(),
    486                                                SrcMgr::C_User));
    487     }
    488   }
    489 
    490   return LoadedSLocEntryTable[Index];
    491 }
    492 
    493 std::pair<int, unsigned>
    494 SourceManager::AllocateLoadedSLocEntries(unsigned NumSLocEntries,
    495                                          unsigned TotalSize) {
    496   assert(ExternalSLocEntries && "Don't have an external sloc source");
    497   LoadedSLocEntryTable.resize(LoadedSLocEntryTable.size() + NumSLocEntries);
    498   SLocEntryLoaded.resize(LoadedSLocEntryTable.size());
    499   CurrentLoadedOffset -= TotalSize;
    500   assert(CurrentLoadedOffset >= NextLocalOffset && "Out of source locations");
    501   int ID = LoadedSLocEntryTable.size();
    502   return std::make_pair(-ID - 1, CurrentLoadedOffset);
    503 }
    504 
    505 /// \brief As part of recovering from missing or changed content, produce a
    506 /// fake, non-empty buffer.
    507 const llvm::MemoryBuffer *SourceManager::getFakeBufferForRecovery() const {
    508   if (!FakeBufferForRecovery)
    509     FakeBufferForRecovery
    510       = llvm::MemoryBuffer::getMemBuffer("<<<INVALID BUFFER>>");
    511 
    512   return FakeBufferForRecovery;
    513 }
    514 
    515 /// \brief As part of recovering from missing or changed content, produce a
    516 /// fake content cache.
    517 const SrcMgr::ContentCache *
    518 SourceManager::getFakeContentCacheForRecovery() const {
    519   if (!FakeContentCacheForRecovery) {
    520     FakeContentCacheForRecovery = new ContentCache();
    521     FakeContentCacheForRecovery->replaceBuffer(getFakeBufferForRecovery(),
    522                                                /*DoNotFree=*/true);
    523   }
    524   return FakeContentCacheForRecovery;
    525 }
    526 
    527 //===----------------------------------------------------------------------===//
    528 // Methods to create new FileID's and macro expansions.
    529 //===----------------------------------------------------------------------===//
    530 
    531 /// createFileID - Create a new FileID for the specified ContentCache and
    532 /// include position.  This works regardless of whether the ContentCache
    533 /// corresponds to a file or some other input source.
    534 FileID SourceManager::createFileID(const ContentCache *File,
    535                                    SourceLocation IncludePos,
    536                                    SrcMgr::CharacteristicKind FileCharacter,
    537                                    int LoadedID, unsigned LoadedOffset) {
    538   if (LoadedID < 0) {
    539     assert(LoadedID != -1 && "Loading sentinel FileID");
    540     unsigned Index = unsigned(-LoadedID) - 2;
    541     assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
    542     assert(!SLocEntryLoaded[Index] && "FileID already loaded");
    543     LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset,
    544         FileInfo::get(IncludePos, File, FileCharacter));
    545     SLocEntryLoaded[Index] = true;
    546     return FileID::get(LoadedID);
    547   }
    548   LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset,
    549                                                FileInfo::get(IncludePos, File,
    550                                                              FileCharacter)));
    551   unsigned FileSize = File->getSize();
    552   assert(NextLocalOffset + FileSize + 1 > NextLocalOffset &&
    553          NextLocalOffset + FileSize + 1 <= CurrentLoadedOffset &&
    554          "Ran out of source locations!");
    555   // We do a +1 here because we want a SourceLocation that means "the end of the
    556   // file", e.g. for the "no newline at the end of the file" diagnostic.
    557   NextLocalOffset += FileSize + 1;
    558 
    559   // Set LastFileIDLookup to the newly created file.  The next getFileID call is
    560   // almost guaranteed to be from that file.
    561   FileID FID = FileID::get(LocalSLocEntryTable.size()-1);
    562   return LastFileIDLookup = FID;
    563 }
    564 
    565 SourceLocation
    566 SourceManager::createMacroArgExpansionLoc(SourceLocation SpellingLoc,
    567                                           SourceLocation ExpansionLoc,
    568                                           unsigned TokLength) {
    569   ExpansionInfo Info = ExpansionInfo::createForMacroArg(SpellingLoc,
    570                                                         ExpansionLoc);
    571   return createExpansionLocImpl(Info, TokLength);
    572 }
    573 
    574 SourceLocation
    575 SourceManager::createExpansionLoc(SourceLocation SpellingLoc,
    576                                   SourceLocation ExpansionLocStart,
    577                                   SourceLocation ExpansionLocEnd,
    578                                   unsigned TokLength,
    579                                   int LoadedID,
    580                                   unsigned LoadedOffset) {
    581   ExpansionInfo Info = ExpansionInfo::create(SpellingLoc, ExpansionLocStart,
    582                                              ExpansionLocEnd);
    583   return createExpansionLocImpl(Info, TokLength, LoadedID, LoadedOffset);
    584 }
    585 
    586 SourceLocation
    587 SourceManager::createExpansionLocImpl(const ExpansionInfo &Info,
    588                                       unsigned TokLength,
    589                                       int LoadedID,
    590                                       unsigned LoadedOffset) {
    591   if (LoadedID < 0) {
    592     assert(LoadedID != -1 && "Loading sentinel FileID");
    593     unsigned Index = unsigned(-LoadedID) - 2;
    594     assert(Index < LoadedSLocEntryTable.size() && "FileID out of range");
    595     assert(!SLocEntryLoaded[Index] && "FileID already loaded");
    596     LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, Info);
    597     SLocEntryLoaded[Index] = true;
    598     return SourceLocation::getMacroLoc(LoadedOffset);
    599   }
    600   LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, Info));
    601   assert(NextLocalOffset + TokLength + 1 > NextLocalOffset &&
    602          NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset &&
    603          "Ran out of source locations!");
    604   // See createFileID for that +1.
    605   NextLocalOffset += TokLength + 1;
    606   return SourceLocation::getMacroLoc(NextLocalOffset - (TokLength + 1));
    607 }
    608 
    609 const llvm::MemoryBuffer *
    610 SourceManager::getMemoryBufferForFile(const FileEntry *File,
    611                                       bool *Invalid) {
    612   const SrcMgr::ContentCache *IR = getOrCreateContentCache(File);
    613   assert(IR && "getOrCreateContentCache() cannot return NULL");
    614   return IR->getBuffer(Diag, *this, SourceLocation(), Invalid);
    615 }
    616 
    617 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
    618                                          const llvm::MemoryBuffer *Buffer,
    619                                          bool DoNotFree) {
    620   const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile);
    621   assert(IR && "getOrCreateContentCache() cannot return NULL");
    622 
    623   const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree);
    624   const_cast<SrcMgr::ContentCache *>(IR)->BufferOverridden = true;
    625 }
    626 
    627 void SourceManager::overrideFileContents(const FileEntry *SourceFile,
    628                                          const FileEntry *NewFile) {
    629   assert(SourceFile->getSize() == NewFile->getSize() &&
    630          "Different sizes, use the FileManager to create a virtual file with "
    631          "the correct size");
    632   assert(FileInfos.count(SourceFile) == 0 &&
    633          "This function should be called at the initialization stage, before "
    634          "any parsing occurs.");
    635   OverriddenFiles[SourceFile] = NewFile;
    636 }
    637 
    638 StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const {
    639   bool MyInvalid = false;
    640   const SLocEntry &SLoc = getSLocEntry(FID, &MyInvalid);
    641   if (!SLoc.isFile() || MyInvalid) {
    642     if (Invalid)
    643       *Invalid = true;
    644     return "<<<<<INVALID SOURCE LOCATION>>>>>";
    645   }
    646 
    647   const llvm::MemoryBuffer *Buf
    648     = SLoc.getFile().getContentCache()->getBuffer(Diag, *this, SourceLocation(),
    649                                                   &MyInvalid);
    650   if (Invalid)
    651     *Invalid = MyInvalid;
    652 
    653   if (MyInvalid)
    654     return "<<<<<INVALID SOURCE LOCATION>>>>>";
    655 
    656   return Buf->getBuffer();
    657 }
    658 
    659 //===----------------------------------------------------------------------===//
    660 // SourceLocation manipulation methods.
    661 //===----------------------------------------------------------------------===//
    662 
    663 /// \brief Return the FileID for a SourceLocation.
    664 ///
    665 /// This is the cache-miss path of getFileID. Not as hot as that function, but
    666 /// still very important. It is responsible for finding the entry in the
    667 /// SLocEntry tables that contains the specified location.
    668 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const {
    669   if (!SLocOffset)
    670     return FileID::get(0);
    671 
    672   // Now it is time to search for the correct file. See where the SLocOffset
    673   // sits in the global view and consult local or loaded buffers for it.
    674   if (SLocOffset < NextLocalOffset)
    675     return getFileIDLocal(SLocOffset);
    676   return getFileIDLoaded(SLocOffset);
    677 }
    678 
    679 /// \brief Return the FileID for a SourceLocation with a low offset.
    680 ///
    681 /// This function knows that the SourceLocation is in a local buffer, not a
    682 /// loaded one.
    683 FileID SourceManager::getFileIDLocal(unsigned SLocOffset) const {
    684   assert(SLocOffset < NextLocalOffset && "Bad function choice");
    685 
    686   // After the first and second level caches, I see two common sorts of
    687   // behavior: 1) a lot of searched FileID's are "near" the cached file
    688   // location or are "near" the cached expansion location. 2) others are just
    689   // completely random and may be a very long way away.
    690   //
    691   // To handle this, we do a linear search for up to 8 steps to catch #1 quickly
    692   // then we fall back to a less cache efficient, but more scalable, binary
    693   // search to find the location.
    694 
    695   // See if this is near the file point - worst case we start scanning from the
    696   // most newly created FileID.
    697   std::vector<SrcMgr::SLocEntry>::const_iterator I;
    698 
    699   if (LastFileIDLookup.ID < 0 ||
    700       LocalSLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) {
    701     // Neither loc prunes our search.
    702     I = LocalSLocEntryTable.end();
    703   } else {
    704     // Perhaps it is near the file point.
    705     I = LocalSLocEntryTable.begin()+LastFileIDLookup.ID;
    706   }
    707 
    708   // Find the FileID that contains this.  "I" is an iterator that points to a
    709   // FileID whose offset is known to be larger than SLocOffset.
    710   unsigned NumProbes = 0;
    711   while (1) {
    712     --I;
    713     if (I->getOffset() <= SLocOffset) {
    714       FileID Res = FileID::get(int(I - LocalSLocEntryTable.begin()));
    715 
    716       // If this isn't an expansion, remember it.  We have good locality across
    717       // FileID lookups.
    718       if (!I->isExpansion())
    719         LastFileIDLookup = Res;
    720       NumLinearScans += NumProbes+1;
    721       return Res;
    722     }
    723     if (++NumProbes == 8)
    724       break;
    725   }
    726 
    727   // Convert "I" back into an index.  We know that it is an entry whose index is
    728   // larger than the offset we are looking for.
    729   unsigned GreaterIndex = I - LocalSLocEntryTable.begin();
    730   // LessIndex - This is the lower bound of the range that we're searching.
    731   // We know that the offset corresponding to the FileID is is less than
    732   // SLocOffset.
    733   unsigned LessIndex = 0;
    734   NumProbes = 0;
    735   while (1) {
    736     bool Invalid = false;
    737     unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex;
    738     unsigned MidOffset = getLocalSLocEntry(MiddleIndex, &Invalid).getOffset();
    739     if (Invalid)
    740       return FileID::get(0);
    741 
    742     ++NumProbes;
    743 
    744     // If the offset of the midpoint is too large, chop the high side of the
    745     // range to the midpoint.
    746     if (MidOffset > SLocOffset) {
    747       GreaterIndex = MiddleIndex;
    748       continue;
    749     }
    750 
    751     // If the middle index contains the value, succeed and return.
    752     // FIXME: This could be made faster by using a function that's aware of
    753     // being in the local area.
    754     if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) {
    755       FileID Res = FileID::get(MiddleIndex);
    756 
    757       // If this isn't a macro expansion, remember it.  We have good locality
    758       // across FileID lookups.
    759       if (!LocalSLocEntryTable[MiddleIndex].isExpansion())
    760         LastFileIDLookup = Res;
    761       NumBinaryProbes += NumProbes;
    762       return Res;
    763     }
    764 
    765     // Otherwise, move the low-side up to the middle index.
    766     LessIndex = MiddleIndex;
    767   }
    768 }
    769 
    770 /// \brief Return the FileID for a SourceLocation with a high offset.
    771 ///
    772 /// This function knows that the SourceLocation is in a loaded buffer, not a
    773 /// local one.
    774 FileID SourceManager::getFileIDLoaded(unsigned SLocOffset) const {
    775   // Sanity checking, otherwise a bug may lead to hanging in release build.
    776   if (SLocOffset < CurrentLoadedOffset) {
    777     assert(0 && "Invalid SLocOffset or bad function choice");
    778     return FileID();
    779   }
    780 
    781   // Essentially the same as the local case, but the loaded array is sorted
    782   // in the other direction.
    783 
    784   // First do a linear scan from the last lookup position, if possible.
    785   unsigned I;
    786   int LastID = LastFileIDLookup.ID;
    787   if (LastID >= 0 || getLoadedSLocEntryByID(LastID).getOffset() < SLocOffset)
    788     I = 0;
    789   else
    790     I = (-LastID - 2) + 1;
    791 
    792   unsigned NumProbes;
    793   for (NumProbes = 0; NumProbes < 8; ++NumProbes, ++I) {
    794     // Make sure the entry is loaded!
    795     const SrcMgr::SLocEntry &E = getLoadedSLocEntry(I);
    796     if (E.getOffset() <= SLocOffset) {
    797       FileID Res = FileID::get(-int(I) - 2);
    798 
    799       if (!E.isExpansion())
    800         LastFileIDLookup = Res;
    801       NumLinearScans += NumProbes + 1;
    802       return Res;
    803     }
    804   }
    805 
    806   // Linear scan failed. Do the binary search. Note the reverse sorting of the
    807   // table: GreaterIndex is the one where the offset is greater, which is
    808   // actually a lower index!
    809   unsigned GreaterIndex = I;
    810   unsigned LessIndex = LoadedSLocEntryTable.size();
    811   NumProbes = 0;
    812   while (1) {
    813     ++NumProbes;
    814     unsigned MiddleIndex = (LessIndex - GreaterIndex) / 2 + GreaterIndex;
    815     const SrcMgr::SLocEntry &E = getLoadedSLocEntry(MiddleIndex);
    816 
    817     ++NumProbes;
    818 
    819     if (E.getOffset() > SLocOffset) {
    820       GreaterIndex = MiddleIndex;
    821       continue;
    822     }
    823 
    824     if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) {
    825       FileID Res = FileID::get(-int(MiddleIndex) - 2);
    826       if (!E.isExpansion())
    827         LastFileIDLookup = Res;
    828       NumBinaryProbes += NumProbes;
    829       return Res;
    830     }
    831 
    832     LessIndex = MiddleIndex;
    833   }
    834 }
    835 
    836 SourceLocation SourceManager::
    837 getExpansionLocSlowCase(SourceLocation Loc) const {
    838   do {
    839     // Note: If Loc indicates an offset into a token that came from a macro
    840     // expansion (e.g. the 5th character of the token) we do not want to add
    841     // this offset when going to the expansion location.  The expansion
    842     // location is the macro invocation, which the offset has nothing to do
    843     // with.  This is unlike when we get the spelling loc, because the offset
    844     // directly correspond to the token whose spelling we're inspecting.
    845     Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart();
    846   } while (!Loc.isFileID());
    847 
    848   return Loc;
    849 }
    850 
    851 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const {
    852   do {
    853     std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
    854     Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
    855     Loc = Loc.getLocWithOffset(LocInfo.second);
    856   } while (!Loc.isFileID());
    857   return Loc;
    858 }
    859 
    860 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const {
    861   do {
    862     if (isMacroArgExpansion(Loc))
    863       Loc = getImmediateSpellingLoc(Loc);
    864     else
    865       Loc = getImmediateExpansionRange(Loc).first;
    866   } while (!Loc.isFileID());
    867   return Loc;
    868 }
    869 
    870 
    871 std::pair<FileID, unsigned>
    872 SourceManager::getDecomposedExpansionLocSlowCase(
    873                                              const SrcMgr::SLocEntry *E) const {
    874   // If this is an expansion record, walk through all the expansion points.
    875   FileID FID;
    876   SourceLocation Loc;
    877   unsigned Offset;
    878   do {
    879     Loc = E->getExpansion().getExpansionLocStart();
    880 
    881     FID = getFileID(Loc);
    882     E = &getSLocEntry(FID);
    883     Offset = Loc.getOffset()-E->getOffset();
    884   } while (!Loc.isFileID());
    885 
    886   return std::make_pair(FID, Offset);
    887 }
    888 
    889 std::pair<FileID, unsigned>
    890 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E,
    891                                                 unsigned Offset) const {
    892   // If this is an expansion record, walk through all the expansion points.
    893   FileID FID;
    894   SourceLocation Loc;
    895   do {
    896     Loc = E->getExpansion().getSpellingLoc();
    897     Loc = Loc.getLocWithOffset(Offset);
    898 
    899     FID = getFileID(Loc);
    900     E = &getSLocEntry(FID);
    901     Offset = Loc.getOffset()-E->getOffset();
    902   } while (!Loc.isFileID());
    903 
    904   return std::make_pair(FID, Offset);
    905 }
    906 
    907 /// getImmediateSpellingLoc - Given a SourceLocation object, return the
    908 /// spelling location referenced by the ID.  This is the first level down
    909 /// towards the place where the characters that make up the lexed token can be
    910 /// found.  This should not generally be used by clients.
    911 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{
    912   if (Loc.isFileID()) return Loc;
    913   std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc);
    914   Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc();
    915   return Loc.getLocWithOffset(LocInfo.second);
    916 }
    917 
    918 
    919 /// getImmediateExpansionRange - Loc is required to be an expansion location.
    920 /// Return the start/end of the expansion information.
    921 std::pair<SourceLocation,SourceLocation>
    922 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const {
    923   assert(Loc.isMacroID() && "Not a macro expansion loc!");
    924   const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion();
    925   return Expansion.getExpansionLocRange();
    926 }
    927 
    928 /// getExpansionRange - Given a SourceLocation object, return the range of
    929 /// tokens covered by the expansion in the ultimate file.
    930 std::pair<SourceLocation,SourceLocation>
    931 SourceManager::getExpansionRange(SourceLocation Loc) const {
    932   if (Loc.isFileID()) return std::make_pair(Loc, Loc);
    933 
    934   std::pair<SourceLocation,SourceLocation> Res =
    935     getImmediateExpansionRange(Loc);
    936 
    937   // Fully resolve the start and end locations to their ultimate expansion
    938   // points.
    939   while (!Res.first.isFileID())
    940     Res.first = getImmediateExpansionRange(Res.first).first;
    941   while (!Res.second.isFileID())
    942     Res.second = getImmediateExpansionRange(Res.second).second;
    943   return Res;
    944 }
    945 
    946 bool SourceManager::isMacroArgExpansion(SourceLocation Loc) const {
    947   if (!Loc.isMacroID()) return false;
    948 
    949   FileID FID = getFileID(Loc);
    950   const SrcMgr::SLocEntry *E = &getSLocEntry(FID);
    951   const SrcMgr::ExpansionInfo &Expansion = E->getExpansion();
    952   return Expansion.isMacroArgExpansion();
    953 }
    954 
    955 
    956 //===----------------------------------------------------------------------===//
    957 // Queries about the code at a SourceLocation.
    958 //===----------------------------------------------------------------------===//
    959 
    960 /// getCharacterData - Return a pointer to the start of the specified location
    961 /// in the appropriate MemoryBuffer.
    962 const char *SourceManager::getCharacterData(SourceLocation SL,
    963                                             bool *Invalid) const {
    964   // Note that this is a hot function in the getSpelling() path, which is
    965   // heavily used by -E mode.
    966   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL);
    967 
    968   // Note that calling 'getBuffer()' may lazily page in a source file.
    969   bool CharDataInvalid = false;
    970   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid);
    971   if (CharDataInvalid || !Entry.isFile()) {
    972     if (Invalid)
    973       *Invalid = true;
    974 
    975     return "<<<<INVALID BUFFER>>>>";
    976   }
    977   const llvm::MemoryBuffer *Buffer
    978     = Entry.getFile().getContentCache()
    979                   ->getBuffer(Diag, *this, SourceLocation(), &CharDataInvalid);
    980   if (Invalid)
    981     *Invalid = CharDataInvalid;
    982   return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second);
    983 }
    984 
    985 
    986 /// getColumnNumber - Return the column # for the specified file position.
    987 /// this is significantly cheaper to compute than the line number.
    988 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos,
    989                                         bool *Invalid) const {
    990   bool MyInvalid = false;
    991   const llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid);
    992   if (Invalid)
    993     *Invalid = MyInvalid;
    994 
    995   if (MyInvalid)
    996     return 1;
    997 
    998   if (FilePos >= MemBuf->getBufferSize()) {
    999     if (Invalid)
   1000       *Invalid = MyInvalid;
   1001     return 1;
   1002   }
   1003 
   1004   const char *Buf = MemBuf->getBufferStart();
   1005   unsigned LineStart = FilePos;
   1006   while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r')
   1007     --LineStart;
   1008   return FilePos-LineStart+1;
   1009 }
   1010 
   1011 // isInvalid - Return the result of calling loc.isInvalid(), and
   1012 // if Invalid is not null, set its value to same.
   1013 static bool isInvalid(SourceLocation Loc, bool *Invalid) {
   1014   bool MyInvalid = Loc.isInvalid();
   1015   if (Invalid)
   1016     *Invalid = MyInvalid;
   1017   return MyInvalid;
   1018 }
   1019 
   1020 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc,
   1021                                                 bool *Invalid) const {
   1022   if (isInvalid(Loc, Invalid)) return 0;
   1023   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
   1024   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
   1025 }
   1026 
   1027 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc,
   1028                                                  bool *Invalid) const {
   1029   if (isInvalid(Loc, Invalid)) return 0;
   1030   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
   1031   return getColumnNumber(LocInfo.first, LocInfo.second, Invalid);
   1032 }
   1033 
   1034 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc,
   1035                                                 bool *Invalid) const {
   1036   if (isInvalid(Loc, Invalid)) return 0;
   1037   return getPresumedLoc(Loc).getColumn();
   1038 }
   1039 
   1040 #ifdef __SSE2__
   1041 #include <emmintrin.h>
   1042 #endif
   1043 
   1044 static LLVM_ATTRIBUTE_NOINLINE void
   1045 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
   1046                    llvm::BumpPtrAllocator &Alloc,
   1047                    const SourceManager &SM, bool &Invalid);
   1048 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI,
   1049                                llvm::BumpPtrAllocator &Alloc,
   1050                                const SourceManager &SM, bool &Invalid) {
   1051   // Note that calling 'getBuffer()' may lazily page in the file.
   1052   const MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(),
   1053                                              &Invalid);
   1054   if (Invalid)
   1055     return;
   1056 
   1057   // Find the file offsets of all of the *physical* source lines.  This does
   1058   // not look at trigraphs, escaped newlines, or anything else tricky.
   1059   SmallVector<unsigned, 256> LineOffsets;
   1060 
   1061   // Line #1 starts at char 0.
   1062   LineOffsets.push_back(0);
   1063 
   1064   const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart();
   1065   const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd();
   1066   unsigned Offs = 0;
   1067   while (1) {
   1068     // Skip over the contents of the line.
   1069     const unsigned char *NextBuf = (const unsigned char *)Buf;
   1070 
   1071 #ifdef __SSE2__
   1072     // Try to skip to the next newline using SSE instructions. This is very
   1073     // performance sensitive for programs with lots of diagnostics and in -E
   1074     // mode.
   1075     __m128i CRs = _mm_set1_epi8('\r');
   1076     __m128i LFs = _mm_set1_epi8('\n');
   1077 
   1078     // First fix up the alignment to 16 bytes.
   1079     while (((uintptr_t)NextBuf & 0xF) != 0) {
   1080       if (*NextBuf == '\n' || *NextBuf == '\r' || *NextBuf == '\0')
   1081         goto FoundSpecialChar;
   1082       ++NextBuf;
   1083     }
   1084 
   1085     // Scan 16 byte chunks for '\r' and '\n'. Ignore '\0'.
   1086     while (NextBuf+16 <= End) {
   1087       __m128i Chunk = *(__m128i*)NextBuf;
   1088       __m128i Cmp = _mm_or_si128(_mm_cmpeq_epi8(Chunk, CRs),
   1089                                  _mm_cmpeq_epi8(Chunk, LFs));
   1090       unsigned Mask = _mm_movemask_epi8(Cmp);
   1091 
   1092       // If we found a newline, adjust the pointer and jump to the handling code.
   1093       if (Mask != 0) {
   1094         NextBuf += llvm::CountTrailingZeros_32(Mask);
   1095         goto FoundSpecialChar;
   1096       }
   1097       NextBuf += 16;
   1098     }
   1099 #endif
   1100 
   1101     while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0')
   1102       ++NextBuf;
   1103 
   1104 #ifdef __SSE2__
   1105 FoundSpecialChar:
   1106 #endif
   1107     Offs += NextBuf-Buf;
   1108     Buf = NextBuf;
   1109 
   1110     if (Buf[0] == '\n' || Buf[0] == '\r') {
   1111       // If this is \n\r or \r\n, skip both characters.
   1112       if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1])
   1113         ++Offs, ++Buf;
   1114       ++Offs, ++Buf;
   1115       LineOffsets.push_back(Offs);
   1116     } else {
   1117       // Otherwise, this is a null.  If end of file, exit.
   1118       if (Buf == End) break;
   1119       // Otherwise, skip the null.
   1120       ++Offs, ++Buf;
   1121     }
   1122   }
   1123 
   1124   // Copy the offsets into the FileInfo structure.
   1125   FI->NumLines = LineOffsets.size();
   1126   FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size());
   1127   std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache);
   1128 }
   1129 
   1130 /// getLineNumber - Given a SourceLocation, return the spelling line number
   1131 /// for the position indicated.  This requires building and caching a table of
   1132 /// line offsets for the MemoryBuffer, so this is not cheap: use only when
   1133 /// about to emit a diagnostic.
   1134 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos,
   1135                                       bool *Invalid) const {
   1136   if (FID.isInvalid()) {
   1137     if (Invalid)
   1138       *Invalid = true;
   1139     return 1;
   1140   }
   1141 
   1142   ContentCache *Content;
   1143   if (LastLineNoFileIDQuery == FID)
   1144     Content = LastLineNoContentCache;
   1145   else {
   1146     bool MyInvalid = false;
   1147     const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid);
   1148     if (MyInvalid || !Entry.isFile()) {
   1149       if (Invalid)
   1150         *Invalid = true;
   1151       return 1;
   1152     }
   1153 
   1154     Content = const_cast<ContentCache*>(Entry.getFile().getContentCache());
   1155   }
   1156 
   1157   // If this is the first use of line information for this buffer, compute the
   1158   /// SourceLineCache for it on demand.
   1159   if (Content->SourceLineCache == 0) {
   1160     bool MyInvalid = false;
   1161     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
   1162     if (Invalid)
   1163       *Invalid = MyInvalid;
   1164     if (MyInvalid)
   1165       return 1;
   1166   } else if (Invalid)
   1167     *Invalid = false;
   1168 
   1169   // Okay, we know we have a line number table.  Do a binary search to find the
   1170   // line number that this character position lands on.
   1171   unsigned *SourceLineCache = Content->SourceLineCache;
   1172   unsigned *SourceLineCacheStart = SourceLineCache;
   1173   unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines;
   1174 
   1175   unsigned QueriedFilePos = FilePos+1;
   1176 
   1177   // FIXME: I would like to be convinced that this code is worth being as
   1178   // complicated as it is, binary search isn't that slow.
   1179   //
   1180   // If it is worth being optimized, then in my opinion it could be more
   1181   // performant, simpler, and more obviously correct by just "galloping" outward
   1182   // from the queried file position. In fact, this could be incorporated into a
   1183   // generic algorithm such as lower_bound_with_hint.
   1184   //
   1185   // If someone gives me a test case where this matters, and I will do it! - DWD
   1186 
   1187   // If the previous query was to the same file, we know both the file pos from
   1188   // that query and the line number returned.  This allows us to narrow the
   1189   // search space from the entire file to something near the match.
   1190   if (LastLineNoFileIDQuery == FID) {
   1191     if (QueriedFilePos >= LastLineNoFilePos) {
   1192       // FIXME: Potential overflow?
   1193       SourceLineCache = SourceLineCache+LastLineNoResult-1;
   1194 
   1195       // The query is likely to be nearby the previous one.  Here we check to
   1196       // see if it is within 5, 10 or 20 lines.  It can be far away in cases
   1197       // where big comment blocks and vertical whitespace eat up lines but
   1198       // contribute no tokens.
   1199       if (SourceLineCache+5 < SourceLineCacheEnd) {
   1200         if (SourceLineCache[5] > QueriedFilePos)
   1201           SourceLineCacheEnd = SourceLineCache+5;
   1202         else if (SourceLineCache+10 < SourceLineCacheEnd) {
   1203           if (SourceLineCache[10] > QueriedFilePos)
   1204             SourceLineCacheEnd = SourceLineCache+10;
   1205           else if (SourceLineCache+20 < SourceLineCacheEnd) {
   1206             if (SourceLineCache[20] > QueriedFilePos)
   1207               SourceLineCacheEnd = SourceLineCache+20;
   1208           }
   1209         }
   1210       }
   1211     } else {
   1212       if (LastLineNoResult < Content->NumLines)
   1213         SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1;
   1214     }
   1215   }
   1216 
   1217   // If the spread is large, do a "radix" test as our initial guess, based on
   1218   // the assumption that lines average to approximately the same length.
   1219   // NOTE: This is currently disabled, as it does not appear to be profitable in
   1220   // initial measurements.
   1221   if (0 && SourceLineCacheEnd-SourceLineCache > 20) {
   1222     unsigned FileLen = Content->SourceLineCache[Content->NumLines-1];
   1223 
   1224     // Take a stab at guessing where it is.
   1225     unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen;
   1226 
   1227     // Check for -10 and +10 lines.
   1228     unsigned LowerBound = std::max(int(ApproxPos-10), 0);
   1229     unsigned UpperBound = std::min(ApproxPos+10, FileLen);
   1230 
   1231     // If the computed lower bound is less than the query location, move it in.
   1232     if (SourceLineCache < SourceLineCacheStart+LowerBound &&
   1233         SourceLineCacheStart[LowerBound] < QueriedFilePos)
   1234       SourceLineCache = SourceLineCacheStart+LowerBound;
   1235 
   1236     // If the computed upper bound is greater than the query location, move it.
   1237     if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound &&
   1238         SourceLineCacheStart[UpperBound] >= QueriedFilePos)
   1239       SourceLineCacheEnd = SourceLineCacheStart+UpperBound;
   1240   }
   1241 
   1242   unsigned *Pos
   1243     = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos);
   1244   unsigned LineNo = Pos-SourceLineCacheStart;
   1245 
   1246   LastLineNoFileIDQuery = FID;
   1247   LastLineNoContentCache = Content;
   1248   LastLineNoFilePos = QueriedFilePos;
   1249   LastLineNoResult = LineNo;
   1250   return LineNo;
   1251 }
   1252 
   1253 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc,
   1254                                               bool *Invalid) const {
   1255   if (isInvalid(Loc, Invalid)) return 0;
   1256   std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc);
   1257   return getLineNumber(LocInfo.first, LocInfo.second);
   1258 }
   1259 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc,
   1260                                                bool *Invalid) const {
   1261   if (isInvalid(Loc, Invalid)) return 0;
   1262   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
   1263   return getLineNumber(LocInfo.first, LocInfo.second);
   1264 }
   1265 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc,
   1266                                               bool *Invalid) const {
   1267   if (isInvalid(Loc, Invalid)) return 0;
   1268   return getPresumedLoc(Loc).getLine();
   1269 }
   1270 
   1271 /// getFileCharacteristic - return the file characteristic of the specified
   1272 /// source location, indicating whether this is a normal file, a system
   1273 /// header, or an "implicit extern C" system header.
   1274 ///
   1275 /// This state can be modified with flags on GNU linemarker directives like:
   1276 ///   # 4 "foo.h" 3
   1277 /// which changes all source locations in the current file after that to be
   1278 /// considered to be from a system header.
   1279 SrcMgr::CharacteristicKind
   1280 SourceManager::getFileCharacteristic(SourceLocation Loc) const {
   1281   assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!");
   1282   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
   1283   bool Invalid = false;
   1284   const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid);
   1285   if (Invalid || !SEntry.isFile())
   1286     return C_User;
   1287 
   1288   const SrcMgr::FileInfo &FI = SEntry.getFile();
   1289 
   1290   // If there are no #line directives in this file, just return the whole-file
   1291   // state.
   1292   if (!FI.hasLineDirectives())
   1293     return FI.getFileCharacteristic();
   1294 
   1295   assert(LineTable && "Can't have linetable entries without a LineTable!");
   1296   // See if there is a #line directive before the location.
   1297   const LineEntry *Entry =
   1298     LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second);
   1299 
   1300   // If this is before the first line marker, use the file characteristic.
   1301   if (!Entry)
   1302     return FI.getFileCharacteristic();
   1303 
   1304   return Entry->FileKind;
   1305 }
   1306 
   1307 /// Return the filename or buffer identifier of the buffer the location is in.
   1308 /// Note that this name does not respect #line directives.  Use getPresumedLoc
   1309 /// for normal clients.
   1310 const char *SourceManager::getBufferName(SourceLocation Loc,
   1311                                          bool *Invalid) const {
   1312   if (isInvalid(Loc, Invalid)) return "<invalid loc>";
   1313 
   1314   return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier();
   1315 }
   1316 
   1317 
   1318 /// getPresumedLoc - This method returns the "presumed" location of a
   1319 /// SourceLocation specifies.  A "presumed location" can be modified by #line
   1320 /// or GNU line marker directives.  This provides a view on the data that a
   1321 /// user should see in diagnostics, for example.
   1322 ///
   1323 /// Note that a presumed location is always given as the expansion point of an
   1324 /// expansion location, not at the spelling location.
   1325 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc) const {
   1326   if (Loc.isInvalid()) return PresumedLoc();
   1327 
   1328   // Presumed locations are always for expansion points.
   1329   std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc);
   1330 
   1331   bool Invalid = false;
   1332   const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid);
   1333   if (Invalid || !Entry.isFile())
   1334     return PresumedLoc();
   1335 
   1336   const SrcMgr::FileInfo &FI = Entry.getFile();
   1337   const SrcMgr::ContentCache *C = FI.getContentCache();
   1338 
   1339   // To get the source name, first consult the FileEntry (if one exists)
   1340   // before the MemBuffer as this will avoid unnecessarily paging in the
   1341   // MemBuffer.
   1342   const char *Filename;
   1343   if (C->OrigEntry)
   1344     Filename = C->OrigEntry->getName();
   1345   else
   1346     Filename = C->getBuffer(Diag, *this)->getBufferIdentifier();
   1347 
   1348   unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid);
   1349   if (Invalid)
   1350     return PresumedLoc();
   1351   unsigned ColNo  = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid);
   1352   if (Invalid)
   1353     return PresumedLoc();
   1354 
   1355   SourceLocation IncludeLoc = FI.getIncludeLoc();
   1356 
   1357   // If we have #line directives in this file, update and overwrite the physical
   1358   // location info if appropriate.
   1359   if (FI.hasLineDirectives()) {
   1360     assert(LineTable && "Can't have linetable entries without a LineTable!");
   1361     // See if there is a #line directive before this.  If so, get it.
   1362     if (const LineEntry *Entry =
   1363           LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second)) {
   1364       // If the LineEntry indicates a filename, use it.
   1365       if (Entry->FilenameID != -1)
   1366         Filename = LineTable->getFilename(Entry->FilenameID);
   1367 
   1368       // Use the line number specified by the LineEntry.  This line number may
   1369       // be multiple lines down from the line entry.  Add the difference in
   1370       // physical line numbers from the query point and the line marker to the
   1371       // total.
   1372       unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset);
   1373       LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1);
   1374 
   1375       // Note that column numbers are not molested by line markers.
   1376 
   1377       // Handle virtual #include manipulation.
   1378       if (Entry->IncludeOffset) {
   1379         IncludeLoc = getLocForStartOfFile(LocInfo.first);
   1380         IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset);
   1381       }
   1382     }
   1383   }
   1384 
   1385   return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc);
   1386 }
   1387 
   1388 /// \brief The size of the SLocEnty that \arg FID represents.
   1389 unsigned SourceManager::getFileIDSize(FileID FID) const {
   1390   bool Invalid = false;
   1391   const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid);
   1392   if (Invalid)
   1393     return 0;
   1394 
   1395   int ID = FID.ID;
   1396   unsigned NextOffset;
   1397   if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size()))
   1398     NextOffset = getNextLocalOffset();
   1399   else if (ID+1 == -1)
   1400     NextOffset = MaxLoadedOffset;
   1401   else
   1402     NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset();
   1403 
   1404   return NextOffset - Entry.getOffset() - 1;
   1405 }
   1406 
   1407 //===----------------------------------------------------------------------===//
   1408 // Other miscellaneous methods.
   1409 //===----------------------------------------------------------------------===//
   1410 
   1411 /// \brief Retrieve the inode for the given file entry, if possible.
   1412 ///
   1413 /// This routine involves a system call, and therefore should only be used
   1414 /// in non-performance-critical code.
   1415 static llvm::Optional<ino_t> getActualFileInode(const FileEntry *File) {
   1416   if (!File)
   1417     return llvm::Optional<ino_t>();
   1418 
   1419   struct stat StatBuf;
   1420   if (::stat(File->getName(), &StatBuf))
   1421     return llvm::Optional<ino_t>();
   1422 
   1423   return StatBuf.st_ino;
   1424 }
   1425 
   1426 /// \brief Get the source location for the given file:line:col triplet.
   1427 ///
   1428 /// If the source file is included multiple times, the source location will
   1429 /// be based upon an arbitrary inclusion.
   1430 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile,
   1431                                                   unsigned Line,
   1432                                                   unsigned Col) const {
   1433   assert(SourceFile && "Null source file!");
   1434   assert(Line && Col && "Line and column should start from 1!");
   1435 
   1436   FileID FirstFID = translateFile(SourceFile);
   1437   return translateLineCol(FirstFID, Line, Col);
   1438 }
   1439 
   1440 /// \brief Get the FileID for the given file.
   1441 ///
   1442 /// If the source file is included multiple times, the FileID will be the
   1443 /// first inclusion.
   1444 FileID SourceManager::translateFile(const FileEntry *SourceFile) const {
   1445   assert(SourceFile && "Null source file!");
   1446 
   1447   // Find the first file ID that corresponds to the given file.
   1448   FileID FirstFID;
   1449 
   1450   // First, check the main file ID, since it is common to look for a
   1451   // location in the main file.
   1452   llvm::Optional<ino_t> SourceFileInode;
   1453   llvm::Optional<StringRef> SourceFileName;
   1454   if (!MainFileID.isInvalid()) {
   1455     bool Invalid = false;
   1456     const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid);
   1457     if (Invalid)
   1458       return FileID();
   1459 
   1460     if (MainSLoc.isFile()) {
   1461       const ContentCache *MainContentCache
   1462         = MainSLoc.getFile().getContentCache();
   1463       if (!MainContentCache) {
   1464         // Can't do anything
   1465       } else if (MainContentCache->OrigEntry == SourceFile) {
   1466         FirstFID = MainFileID;
   1467       } else {
   1468         // Fall back: check whether we have the same base name and inode
   1469         // as the main file.
   1470         const FileEntry *MainFile = MainContentCache->OrigEntry;
   1471         SourceFileName = llvm::sys::path::filename(SourceFile->getName());
   1472         if (*SourceFileName == llvm::sys::path::filename(MainFile->getName())) {
   1473           SourceFileInode = getActualFileInode(SourceFile);
   1474           if (SourceFileInode) {
   1475             if (llvm::Optional<ino_t> MainFileInode
   1476                                                = getActualFileInode(MainFile)) {
   1477               if (*SourceFileInode == *MainFileInode) {
   1478                 FirstFID = MainFileID;
   1479                 SourceFile = MainFile;
   1480               }
   1481             }
   1482           }
   1483         }
   1484       }
   1485     }
   1486   }
   1487 
   1488   if (FirstFID.isInvalid()) {
   1489     // The location we're looking for isn't in the main file; look
   1490     // through all of the local source locations.
   1491     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
   1492       bool Invalid = false;
   1493       const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid);
   1494       if (Invalid)
   1495         return FileID();
   1496 
   1497       if (SLoc.isFile() &&
   1498           SLoc.getFile().getContentCache() &&
   1499           SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
   1500         FirstFID = FileID::get(I);
   1501         break;
   1502       }
   1503     }
   1504     // If that still didn't help, try the modules.
   1505     if (FirstFID.isInvalid()) {
   1506       for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) {
   1507         const SLocEntry &SLoc = getLoadedSLocEntry(I);
   1508         if (SLoc.isFile() &&
   1509             SLoc.getFile().getContentCache() &&
   1510             SLoc.getFile().getContentCache()->OrigEntry == SourceFile) {
   1511           FirstFID = FileID::get(-int(I) - 2);
   1512           break;
   1513         }
   1514       }
   1515     }
   1516   }
   1517 
   1518   // If we haven't found what we want yet, try again, but this time stat()
   1519   // each of the files in case the files have changed since we originally
   1520   // parsed the file.
   1521   if (FirstFID.isInvalid() &&
   1522       (SourceFileName ||
   1523        (SourceFileName = llvm::sys::path::filename(SourceFile->getName()))) &&
   1524       (SourceFileInode ||
   1525        (SourceFileInode = getActualFileInode(SourceFile)))) {
   1526     bool Invalid = false;
   1527     for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) {
   1528       FileID IFileID;
   1529       IFileID.ID = I;
   1530       const SLocEntry &SLoc = getSLocEntry(IFileID, &Invalid);
   1531       if (Invalid)
   1532         return FileID();
   1533 
   1534       if (SLoc.isFile()) {
   1535         const ContentCache *FileContentCache
   1536           = SLoc.getFile().getContentCache();
   1537       const FileEntry *Entry =FileContentCache? FileContentCache->OrigEntry : 0;
   1538         if (Entry &&
   1539             *SourceFileName == llvm::sys::path::filename(Entry->getName())) {
   1540           if (llvm::Optional<ino_t> EntryInode = getActualFileInode(Entry)) {
   1541             if (*SourceFileInode == *EntryInode) {
   1542               FirstFID = FileID::get(I);
   1543               SourceFile = Entry;
   1544               break;
   1545             }
   1546           }
   1547         }
   1548       }
   1549     }
   1550   }
   1551 
   1552   return FirstFID;
   1553 }
   1554 
   1555 /// \brief Get the source location in \arg FID for the given line:col.
   1556 /// Returns null location if \arg FID is not a file SLocEntry.
   1557 SourceLocation SourceManager::translateLineCol(FileID FID,
   1558                                                unsigned Line,
   1559                                                unsigned Col) const {
   1560   if (FID.isInvalid())
   1561     return SourceLocation();
   1562 
   1563   bool Invalid = false;
   1564   const SLocEntry &Entry = getSLocEntry(FID, &Invalid);
   1565   if (Invalid)
   1566     return SourceLocation();
   1567 
   1568   if (!Entry.isFile())
   1569     return SourceLocation();
   1570 
   1571   SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset());
   1572 
   1573   if (Line == 1 && Col == 1)
   1574     return FileLoc;
   1575 
   1576   ContentCache *Content
   1577     = const_cast<ContentCache *>(Entry.getFile().getContentCache());
   1578   if (!Content)
   1579     return SourceLocation();
   1580 
   1581   // If this is the first use of line information for this buffer, compute the
   1582   // SourceLineCache for it on demand.
   1583   if (Content->SourceLineCache == 0) {
   1584     bool MyInvalid = false;
   1585     ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid);
   1586     if (MyInvalid)
   1587       return SourceLocation();
   1588   }
   1589 
   1590   if (Line > Content->NumLines) {
   1591     unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize();
   1592     if (Size > 0)
   1593       --Size;
   1594     return FileLoc.getLocWithOffset(Size);
   1595   }
   1596 
   1597   const llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, *this);
   1598   unsigned FilePos = Content->SourceLineCache[Line - 1];
   1599   const char *Buf = Buffer->getBufferStart() + FilePos;
   1600   unsigned BufLength = Buffer->getBufferSize() - FilePos;
   1601   if (BufLength == 0)
   1602     return FileLoc.getLocWithOffset(FilePos);
   1603 
   1604   unsigned i = 0;
   1605 
   1606   // Check that the given column is valid.
   1607   while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r')
   1608     ++i;
   1609   if (i < Col-1)
   1610     return FileLoc.getLocWithOffset(FilePos + i);
   1611 
   1612   return FileLoc.getLocWithOffset(FilePos + Col - 1);
   1613 }
   1614 
   1615 /// \brief Compute a map of macro argument chunks to their expanded source
   1616 /// location. Chunks that are not part of a macro argument will map to an
   1617 /// invalid source location. e.g. if a file contains one macro argument at
   1618 /// offset 100 with length 10, this is how the map will be formed:
   1619 ///     0   -> SourceLocation()
   1620 ///     100 -> Expanded macro arg location
   1621 ///     110 -> SourceLocation()
   1622 void SourceManager::computeMacroArgsCache(MacroArgsMap *&CachePtr,
   1623                                           FileID FID) const {
   1624   assert(!FID.isInvalid());
   1625   assert(!CachePtr);
   1626 
   1627   CachePtr = new MacroArgsMap();
   1628   MacroArgsMap &MacroArgsCache = *CachePtr;
   1629   // Initially no macro argument chunk is present.
   1630   MacroArgsCache.insert(std::make_pair(0, SourceLocation()));
   1631 
   1632   int ID = FID.ID;
   1633   while (1) {
   1634     ++ID;
   1635     // Stop if there are no more FileIDs to check.
   1636     if (ID > 0) {
   1637       if (unsigned(ID) >= local_sloc_entry_size())
   1638         return;
   1639     } else if (ID == -1) {
   1640       return;
   1641     }
   1642 
   1643     const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID);
   1644     if (Entry.isFile()) {
   1645       SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc();
   1646       if (IncludeLoc.isInvalid())
   1647         continue;
   1648       if (!isInFileID(IncludeLoc, FID))
   1649         return; // No more files/macros that may be "contained" in this file.
   1650 
   1651       // Skip the files/macros of the #include'd file, we only care about macros
   1652       // that lexed macro arguments from our file.
   1653       if (Entry.getFile().NumCreatedFIDs)
   1654         ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/;
   1655       continue;
   1656     }
   1657 
   1658     const ExpansionInfo &ExpInfo = Entry.getExpansion();
   1659 
   1660     if (ExpInfo.getExpansionLocStart().isFileID()) {
   1661       if (!isInFileID(ExpInfo.getExpansionLocStart(), FID))
   1662         return; // No more files/macros that may be "contained" in this file.
   1663     }
   1664 
   1665     if (!ExpInfo.isMacroArgExpansion())
   1666       continue;
   1667 
   1668     SourceLocation SpellLoc = ExpInfo.getSpellingLoc();
   1669     while (!SpellLoc.isFileID()) {
   1670       std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(SpellLoc);
   1671       const ExpansionInfo &Info = getSLocEntry(LocInfo.first).getExpansion();
   1672       if (!Info.isMacroArgExpansion())
   1673         break;
   1674       SpellLoc = Info.getSpellingLoc().getLocWithOffset(LocInfo.second);
   1675     }
   1676     if (!SpellLoc.isFileID())
   1677       continue;
   1678 
   1679     unsigned BeginOffs;
   1680     if (!isInFileID(SpellLoc, FID, &BeginOffs))
   1681       continue;
   1682 
   1683     unsigned EndOffs = BeginOffs + getFileIDSize(FileID::get(ID));
   1684 
   1685     // Add a new chunk for this macro argument. A previous macro argument chunk
   1686     // may have been lexed again, so e.g. if the map is
   1687     //     0   -> SourceLocation()
   1688     //     100 -> Expanded loc #1
   1689     //     110 -> SourceLocation()
   1690     // and we found a new macro FileID that lexed from offet 105 with length 3,
   1691     // the new map will be:
   1692     //     0   -> SourceLocation()
   1693     //     100 -> Expanded loc #1
   1694     //     105 -> Expanded loc #2
   1695     //     108 -> Expanded loc #1
   1696     //     110 -> SourceLocation()
   1697     //
   1698     // Since re-lexed macro chunks will always be the same size or less of
   1699     // previous chunks, we only need to find where the ending of the new macro
   1700     // chunk is mapped to and update the map with new begin/end mappings.
   1701 
   1702     MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs);
   1703     --I;
   1704     SourceLocation EndOffsMappedLoc = I->second;
   1705     MacroArgsCache[BeginOffs] = SourceLocation::getMacroLoc(Entry.getOffset());
   1706     MacroArgsCache[EndOffs] = EndOffsMappedLoc;
   1707   }
   1708 }
   1709 
   1710 /// \brief If \arg Loc points inside a function macro argument, the returned
   1711 /// location will be the macro location in which the argument was expanded.
   1712 /// If a macro argument is used multiple times, the expanded location will
   1713 /// be at the first expansion of the argument.
   1714 /// e.g.
   1715 ///   MY_MACRO(foo);
   1716 ///             ^
   1717 /// Passing a file location pointing at 'foo', will yield a macro location
   1718 /// where 'foo' was expanded into.
   1719 SourceLocation
   1720 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const {
   1721   if (Loc.isInvalid() || !Loc.isFileID())
   1722     return Loc;
   1723 
   1724   FileID FID;
   1725   unsigned Offset;
   1726   llvm::tie(FID, Offset) = getDecomposedLoc(Loc);
   1727   if (FID.isInvalid())
   1728     return Loc;
   1729 
   1730   MacroArgsMap *&MacroArgsCache = MacroArgsCacheMap[FID];
   1731   if (!MacroArgsCache)
   1732     computeMacroArgsCache(MacroArgsCache, FID);
   1733 
   1734   assert(!MacroArgsCache->empty());
   1735   MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset);
   1736   --I;
   1737 
   1738   unsigned MacroArgBeginOffs = I->first;
   1739   SourceLocation MacroArgExpandedLoc = I->second;
   1740   if (MacroArgExpandedLoc.isValid())
   1741     return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs);
   1742 
   1743   return Loc;
   1744 }
   1745 
   1746 /// Given a decomposed source location, move it up the include/expansion stack
   1747 /// to the parent source location.  If this is possible, return the decomposed
   1748 /// version of the parent in Loc and return false.  If Loc is the top-level
   1749 /// entry, return true and don't modify it.
   1750 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc,
   1751                                    const SourceManager &SM) {
   1752   SourceLocation UpperLoc;
   1753   const SrcMgr::SLocEntry &Entry = SM.getSLocEntry(Loc.first);
   1754   if (Entry.isExpansion())
   1755     UpperLoc = Entry.getExpansion().getExpansionLocStart();
   1756   else
   1757     UpperLoc = Entry.getFile().getIncludeLoc();
   1758 
   1759   if (UpperLoc.isInvalid())
   1760     return true; // We reached the top.
   1761 
   1762   Loc = SM.getDecomposedLoc(UpperLoc);
   1763   return false;
   1764 }
   1765 
   1766 
   1767 /// \brief Determines the order of 2 source locations in the translation unit.
   1768 ///
   1769 /// \returns true if LHS source location comes before RHS, false otherwise.
   1770 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS,
   1771                                               SourceLocation RHS) const {
   1772   assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!");
   1773   if (LHS == RHS)
   1774     return false;
   1775 
   1776   std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS);
   1777   std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS);
   1778 
   1779   // If the source locations are in the same file, just compare offsets.
   1780   if (LOffs.first == ROffs.first)
   1781     return LOffs.second < ROffs.second;
   1782 
   1783   // If we are comparing a source location with multiple locations in the same
   1784   // file, we get a big win by caching the result.
   1785   if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first))
   1786     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
   1787 
   1788   // Okay, we missed in the cache, start updating the cache for this query.
   1789   IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first,
   1790                           /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID);
   1791 
   1792   // We need to find the common ancestor. The only way of doing this is to
   1793   // build the complete include chain for one and then walking up the chain
   1794   // of the other looking for a match.
   1795   // We use a map from FileID to Offset to store the chain. Easier than writing
   1796   // a custom set hash info that only depends on the first part of a pair.
   1797   typedef llvm::DenseMap<FileID, unsigned> LocSet;
   1798   LocSet LChain;
   1799   do {
   1800     LChain.insert(LOffs);
   1801     // We catch the case where LOffs is in a file included by ROffs and
   1802     // quit early. The other way round unfortunately remains suboptimal.
   1803   } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this));
   1804   LocSet::iterator I;
   1805   while((I = LChain.find(ROffs.first)) == LChain.end()) {
   1806     if (MoveUpIncludeHierarchy(ROffs, *this))
   1807       break; // Met at topmost file.
   1808   }
   1809   if (I != LChain.end())
   1810     LOffs = *I;
   1811 
   1812   // If we exited because we found a nearest common ancestor, compare the
   1813   // locations within the common file and cache them.
   1814   if (LOffs.first == ROffs.first) {
   1815     IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second);
   1816     return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second);
   1817   }
   1818 
   1819   // This can happen if a location is in a built-ins buffer.
   1820   // But see PR5662.
   1821   // Clear the lookup cache, it depends on a common location.
   1822   IsBeforeInTUCache.clear();
   1823   bool LIsBuiltins = strcmp("<built-in>",
   1824                             getBuffer(LOffs.first)->getBufferIdentifier()) == 0;
   1825   bool RIsBuiltins = strcmp("<built-in>",
   1826                             getBuffer(ROffs.first)->getBufferIdentifier()) == 0;
   1827   // built-in is before non-built-in
   1828   if (LIsBuiltins != RIsBuiltins)
   1829     return LIsBuiltins;
   1830   assert(LIsBuiltins && RIsBuiltins &&
   1831          "Non-built-in locations must be rooted in the main file");
   1832   // Both are in built-in buffers, but from different files. We just claim that
   1833   // lower IDs come first.
   1834   return LOffs.first < ROffs.first;
   1835 }
   1836 
   1837 /// PrintStats - Print statistics to stderr.
   1838 ///
   1839 void SourceManager::PrintStats() const {
   1840   llvm::errs() << "\n*** Source Manager Stats:\n";
   1841   llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size()
   1842                << " mem buffers mapped.\n";
   1843   llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated ("
   1844                << llvm::capacity_in_bytes(LocalSLocEntryTable)
   1845                << " bytes of capacity), "
   1846                << NextLocalOffset << "B of Sloc address space used.\n";
   1847   llvm::errs() << LoadedSLocEntryTable.size()
   1848                << " loaded SLocEntries allocated, "
   1849                << MaxLoadedOffset - CurrentLoadedOffset
   1850                << "B of Sloc address space used.\n";
   1851 
   1852   unsigned NumLineNumsComputed = 0;
   1853   unsigned NumFileBytesMapped = 0;
   1854   for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){
   1855     NumLineNumsComputed += I->second->SourceLineCache != 0;
   1856     NumFileBytesMapped  += I->second->getSizeBytesMapped();
   1857   }
   1858   unsigned NumMacroArgsComputed = MacroArgsCacheMap.size();
   1859 
   1860   llvm::errs() << NumFileBytesMapped << " bytes of files mapped, "
   1861                << NumLineNumsComputed << " files with line #'s computed, "
   1862                << NumMacroArgsComputed << " files with macro args computed.\n";
   1863   llvm::errs() << "FileID scans: " << NumLinearScans << " linear, "
   1864                << NumBinaryProbes << " binary.\n";
   1865 }
   1866 
   1867 ExternalSLocEntrySource::~ExternalSLocEntrySource() { }
   1868 
   1869 /// Return the amount of memory used by memory buffers, breaking down
   1870 /// by heap-backed versus mmap'ed memory.
   1871 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const {
   1872   size_t malloc_bytes = 0;
   1873   size_t mmap_bytes = 0;
   1874 
   1875   for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i)
   1876     if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped())
   1877       switch (MemBufferInfos[i]->getMemoryBufferKind()) {
   1878         case llvm::MemoryBuffer::MemoryBuffer_MMap:
   1879           mmap_bytes += sized_mapped;
   1880           break;
   1881         case llvm::MemoryBuffer::MemoryBuffer_Malloc:
   1882           malloc_bytes += sized_mapped;
   1883           break;
   1884       }
   1885 
   1886   return MemoryBufferSizes(malloc_bytes, mmap_bytes);
   1887 }
   1888 
   1889 size_t SourceManager::getDataStructureSizes() const {
   1890   return llvm::capacity_in_bytes(MemBufferInfos)
   1891     + llvm::capacity_in_bytes(LocalSLocEntryTable)
   1892     + llvm::capacity_in_bytes(LoadedSLocEntryTable)
   1893     + llvm::capacity_in_bytes(SLocEntryLoaded)
   1894     + llvm::capacity_in_bytes(FileInfos)
   1895     + llvm::capacity_in_bytes(OverriddenFiles);
   1896 }
   1897