lib/Lex/PTHLexer.cpp

   1 //===--- PTHLexer.cpp - Lex from a token stream ---------------------------===//
   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 PTHLexer interface.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "clang/Basic/TokenKinds.h"
  15 #include "clang/Basic/FileManager.h"
  16 #include "clang/Basic/FileSystemStatCache.h"
  17 #include "clang/Basic/IdentifierTable.h"
  18 #include "clang/Basic/OnDiskHashTable.h"
  19 #include "clang/Lex/LexDiagnostic.h"
  20 #include "clang/Lex/PTHLexer.h"
  21 #include "clang/Lex/Preprocessor.h"
  22 #include "clang/Lex/PTHManager.h"
  23 #include "clang/Lex/Token.h"
  24 #include "clang/Lex/Preprocessor.h"
  25 #include "llvm/ADT/OwningPtr.h"
  26 #include "llvm/ADT/StringExtras.h"
  27 #include "llvm/ADT/StringMap.h"
  28 #include "llvm/Support/MemoryBuffer.h"
  29 #include "llvm/Support/system_error.h"
  30 using namespace clang;
  31 using namespace clang::io;
  32
  33 #define DISK_TOKEN_SIZE (1+1+2+4+4)
  34
  35 //===----------------------------------------------------------------------===//
  36 // PTHLexer methods.
  37 //===----------------------------------------------------------------------===//
  38
  39 PTHLexer::PTHLexer(Preprocessor &PP, FileID FID, const unsigned char *D,
  40                    const unsigned char *ppcond, PTHManager &PM)
  41   : PreprocessorLexer(&PP, FID), TokBuf(D), CurPtr(D), LastHashTokPtr(0),
  42     PPCond(ppcond), CurPPCondPtr(ppcond), PTHMgr(PM) {
  43
  44   FileStartLoc = PP.getSourceManager().getLocForStartOfFile(FID);
  45 }
  46
  47 void PTHLexer::Lex(Token& Tok) {
  48 LexNextToken:
  49
  50   //===--------------------------------------==//
  51   // Read the raw token data.
  52   //===--------------------------------------==//
  53
  54   // Shadow CurPtr into an automatic variable.
  55   const unsigned char *CurPtrShadow = CurPtr;
  56
  57   // Read in the data for the token.
  58   unsigned Word0 = ReadLE32(CurPtrShadow);
  59   uint32_t IdentifierID = ReadLE32(CurPtrShadow);
  60   uint32_t FileOffset = ReadLE32(CurPtrShadow);
  61
  62   tok::TokenKind TKind = (tok::TokenKind) (Word0 & 0xFF);
  63   Token::TokenFlags TFlags = (Token::TokenFlags) ((Word0 >> 8) & 0xFF);
  64   uint32_t Len = Word0 >> 16;
  65
  66   CurPtr = CurPtrShadow;
  67
  68   //===--------------------------------------==//
  69   // Construct the token itself.
  70   //===--------------------------------------==//
  71
  72   Tok.startToken();
  73   Tok.setKind(TKind);
  74   Tok.setFlag(TFlags);
  75   assert(!LexingRawMode);
  76   Tok.setLocation(FileStartLoc.getFileLocWithOffset(FileOffset));
  77   Tok.setLength(Len);
  78
  79   // Handle identifiers.
  80   if (Tok.isLiteral()) {
  81     Tok.setLiteralData((const char*) (PTHMgr.SpellingBase + IdentifierID));
  82   }
  83   else if (IdentifierID) {
  84     MIOpt.ReadToken();
  85     IdentifierInfo *II = PTHMgr.GetIdentifierInfo(IdentifierID-1);
  86
  87     Tok.setIdentifierInfo(II);
  88
  89     // Change the kind of this identifier to the appropriate token kind, e.g.
  90     // turning "for" into a keyword.
  91     Tok.setKind(II->getTokenID());
  92
  93     if (II->isHandleIdentifierCase())
  94       PP->HandleIdentifier(Tok);
  95     return;
  96   }
  97
  98   //===--------------------------------------==//
  99   // Process the token.
 100   //===--------------------------------------==//
 101   if (TKind == tok::eof) {
 102     // Save the end-of-file token.
 103     EofToken = Tok;
 104
 105     // Save 'PP' to 'PPCache' as LexEndOfFile can delete 'this'.
 106     Preprocessor *PPCache = PP;
 107
 108     assert(!ParsingPreprocessorDirective);
 109     assert(!LexingRawMode);
 110
 111     if (LexEndOfFile(Tok))
 112       return;
 113
 114     return PPCache->Lex(Tok);
 115   }
 116
 117   if (TKind == tok::hash && Tok.isAtStartOfLine()) {
 118     LastHashTokPtr = CurPtr - DISK_TOKEN_SIZE;
 119     assert(!LexingRawMode);
 120     PP->HandleDirective(Tok);
 121
 122     if (PP->isCurrentLexer(this))
 123       goto LexNextToken;
 124
 125     return PP->Lex(Tok);
 126   }
 127
 128   if (TKind == tok::eom) {
 129     assert(ParsingPreprocessorDirective);
 130     ParsingPreprocessorDirective = false;
 131     return;
 132   }
 133
 134   MIOpt.ReadToken();
 135 }
 136
 137 bool PTHLexer::LexEndOfFile(Token &Result) {
 138   // If we hit the end of the file while parsing a preprocessor directive,
 139   // end the preprocessor directive first.  The next token returned will
 140   // then be the end of file.
 141   if (ParsingPreprocessorDirective) {
 142     ParsingPreprocessorDirective = false; // Done parsing the "line".
 143     return true;  // Have a token.
 144   }
 145
 146   assert(!LexingRawMode);
 147
 148   // If we are in a #if directive, emit an error.
 149   while (!ConditionalStack.empty()) {
 150     if (!PP->isCodeCompletionFile(FileStartLoc))
 151       PP->Diag(ConditionalStack.back().IfLoc,
 152                diag::err_pp_unterminated_conditional);
 153     ConditionalStack.pop_back();
 154   }
 155
 156   // Finally, let the preprocessor handle this.
 157   return PP->HandleEndOfFile(Result);
 158 }
 159
 160 // FIXME: We can just grab the last token instead of storing a copy
 161 // into EofToken.
 162 void PTHLexer::getEOF(Token& Tok) {
 163   assert(EofToken.is(tok::eof));
 164   Tok = EofToken;
 165 }
 166
 167 void PTHLexer::DiscardToEndOfLine() {
 168   assert(ParsingPreprocessorDirective && ParsingFilename == false &&
 169          "Must be in a preprocessing directive!");
 170
 171   // We assume that if the preprocessor wishes to discard to the end of
 172   // the line that it also means to end the current preprocessor directive.
 173   ParsingPreprocessorDirective = false;
 174
 175   // Skip tokens by only peeking at their token kind and the flags.
 176   // We don't need to actually reconstruct full tokens from the token buffer.
 177   // This saves some copies and it also reduces IdentifierInfo* lookup.
 178   const unsigned char* p = CurPtr;
 179   while (1) {
 180     // Read the token kind.  Are we at the end of the file?
 181     tok::TokenKind x = (tok::TokenKind) (uint8_t) *p;
 182     if (x == tok::eof) break;
 183
 184     // Read the token flags.  Are we at the start of the next line?
 185     Token::TokenFlags y = (Token::TokenFlags) (uint8_t) p[1];
 186     if (y & Token::StartOfLine) break;
 187
 188     // Skip to the next token.
 189     p += DISK_TOKEN_SIZE;
 190   }
 191
 192   CurPtr = p;
 193 }
 194
 195 /// SkipBlock - Used by Preprocessor to skip the current conditional block.
 196 bool PTHLexer::SkipBlock() {
 197   assert(CurPPCondPtr && "No cached PP conditional information.");
 198   assert(LastHashTokPtr && "No known '#' token.");
 199
 200   const unsigned char* HashEntryI = 0;
 201   uint32_t Offset;
 202   uint32_t TableIdx;
 203
 204   do {
 205     // Read the token offset from the side-table.
 206     Offset = ReadLE32(CurPPCondPtr);
 207
 208     // Read the target table index from the side-table.
 209     TableIdx = ReadLE32(CurPPCondPtr);
 210
 211     // Compute the actual memory address of the '#' token data for this entry.
 212     HashEntryI = TokBuf + Offset;
 213
 214     // Optmization: "Sibling jumping".  #if...#else...#endif blocks can
 215     //  contain nested blocks.  In the side-table we can jump over these
 216     //  nested blocks instead of doing a linear search if the next "sibling"
 217     //  entry is not at a location greater than LastHashTokPtr.
 218     if (HashEntryI < LastHashTokPtr && TableIdx) {
 219       // In the side-table we are still at an entry for a '#' token that
 220       // is earlier than the last one we saw.  Check if the location we would
 221       // stride gets us closer.
 222       const unsigned char* NextPPCondPtr =
 223         PPCond + TableIdx*(sizeof(uint32_t)*2);
 224       assert(NextPPCondPtr >= CurPPCondPtr);
 225       // Read where we should jump to.
 226       uint32_t TmpOffset = ReadLE32(NextPPCondPtr);
 227       const unsigned char* HashEntryJ = TokBuf + TmpOffset;
 228
 229       if (HashEntryJ <= LastHashTokPtr) {
 230         // Jump directly to the next entry in the side table.
 231         HashEntryI = HashEntryJ;
 232         Offset = TmpOffset;
 233         TableIdx = ReadLE32(NextPPCondPtr);
 234         CurPPCondPtr = NextPPCondPtr;
 235       }
 236     }
 237   }
 238   while (HashEntryI < LastHashTokPtr);
 239   assert(HashEntryI == LastHashTokPtr && "No PP-cond entry found for '#'");
 240   assert(TableIdx && "No jumping from #endifs.");
 241
 242   // Update our side-table iterator.
 243   const unsigned char* NextPPCondPtr = PPCond + TableIdx*(sizeof(uint32_t)*2);
 244   assert(NextPPCondPtr >= CurPPCondPtr);
 245   CurPPCondPtr = NextPPCondPtr;
 246
 247   // Read where we should jump to.
 248   HashEntryI = TokBuf + ReadLE32(NextPPCondPtr);
 249   uint32_t NextIdx = ReadLE32(NextPPCondPtr);
 250
 251   // By construction NextIdx will be zero if this is a #endif.  This is useful
 252   // to know to obviate lexing another token.
 253   bool isEndif = NextIdx == 0;
 254
 255   // This case can occur when we see something like this:
 256   //
 257   //  #if ...
 258   //   /* a comment or nothing */
 259   //  #elif
 260   //
 261   // If we are skipping the first #if block it will be the case that CurPtr
 262   // already points 'elif'.  Just return.
 263
 264   if (CurPtr > HashEntryI) {
 265     assert(CurPtr == HashEntryI + DISK_TOKEN_SIZE);
 266     // Did we reach a #endif?  If so, go ahead and consume that token as well.
 267     if (isEndif)
 268       CurPtr += DISK_TOKEN_SIZE*2;
 269     else
 270       LastHashTokPtr = HashEntryI;
 271
 272     return isEndif;
 273   }
 274
 275   // Otherwise, we need to advance.  Update CurPtr to point to the '#' token.
 276   CurPtr = HashEntryI;
 277
 278   // Update the location of the last observed '#'.  This is useful if we
 279   // are skipping multiple blocks.
 280   LastHashTokPtr = CurPtr;
 281
 282   // Skip the '#' token.
 283   assert(((tok::TokenKind)*CurPtr) == tok::hash);
 284   CurPtr += DISK_TOKEN_SIZE;
 285
 286   // Did we reach a #endif?  If so, go ahead and consume that token as well.
 287   if (isEndif) { CurPtr += DISK_TOKEN_SIZE*2; }
 288
 289   return isEndif;
 290 }
 291
 292 SourceLocation PTHLexer::getSourceLocation() {
 293   // getSourceLocation is not on the hot path.  It is used to get the location
 294   // of the next token when transitioning back to this lexer when done
 295   // handling a #included file.  Just read the necessary data from the token
 296   // data buffer to construct the SourceLocation object.
 297   // NOTE: This is a virtual function; hence it is defined out-of-line.
 298   const unsigned char *OffsetPtr = CurPtr + (DISK_TOKEN_SIZE - 4);
 299   uint32_t Offset = ReadLE32(OffsetPtr);
 300   return FileStartLoc.getFileLocWithOffset(Offset);
 301 }
 302
 303 //===----------------------------------------------------------------------===//
 304 // PTH file lookup: map from strings to file data.
 305 //===----------------------------------------------------------------------===//
 306
 307 /// PTHFileLookup - This internal data structure is used by the PTHManager
 308 ///  to map from FileEntry objects managed by FileManager to offsets within
 309 ///  the PTH file.
 310 namespace {
 311 class PTHFileData {
 312   const uint32_t TokenOff;
 313   const uint32_t PPCondOff;
 314 public:
 315   PTHFileData(uint32_t tokenOff, uint32_t ppCondOff)
 316     : TokenOff(tokenOff), PPCondOff(ppCondOff) {}
 317
 318   uint32_t getTokenOffset() const { return TokenOff; }
 319   uint32_t getPPCondOffset() const { return PPCondOff; }
 320 };
 321
 322
 323 class PTHFileLookupCommonTrait {
 324 public:
 325   typedef std::pair<unsigned char, const char*> internal_key_type;
 326
 327   static unsigned ComputeHash(internal_key_type x) {
 328     return llvm::HashString(x.second);
 329   }
 330
 331   static std::pair<unsigned, unsigned>
 332   ReadKeyDataLength(const unsigned char*& d) {
 333     unsigned keyLen = (unsigned) ReadUnalignedLE16(d);
 334     unsigned dataLen = (unsigned) *(d++);
 335     return std::make_pair(keyLen, dataLen);
 336   }
 337
 338   static internal_key_type ReadKey(const unsigned char* d, unsigned) {
 339     unsigned char k = *(d++); // Read the entry kind.
 340     return std::make_pair(k, (const char*) d);
 341   }
 342 };
 343
 344 class PTHFileLookupTrait : public PTHFileLookupCommonTrait {
 345 public:
 346   typedef const FileEntry* external_key_type;
 347   typedef PTHFileData      data_type;
 348
 349   static internal_key_type GetInternalKey(const FileEntry* FE) {
 350     return std::make_pair((unsigned char) 0x1, FE->getName());
 351   }
 352
 353   static bool EqualKey(internal_key_type a, internal_key_type b) {
 354     return a.first == b.first && strcmp(a.second, b.second) == 0;
 355   }
 356
 357   static PTHFileData ReadData(const internal_key_type& k,
 358                               const unsigned char* d, unsigned) {
 359     assert(k.first == 0x1 && "Only file lookups can match!");
 360     uint32_t x = ::ReadUnalignedLE32(d);
 361     uint32_t y = ::ReadUnalignedLE32(d);
 362     return PTHFileData(x, y);
 363   }
 364 };
 365
 366 class PTHStringLookupTrait {
 367 public:
 368   typedef uint32_t
 369           data_type;
 370
 371   typedef const std::pair<const char*, unsigned>
 372           external_key_type;
 373
 374   typedef external_key_type internal_key_type;
 375
 376   static bool EqualKey(const internal_key_type& a,
 377                        const internal_key_type& b) {
 378     return (a.second == b.second) ? memcmp(a.first, b.first, a.second) == 0
 379                                   : false;
 380   }
 381
 382   static unsigned ComputeHash(const internal_key_type& a) {
 383     return llvm::HashString(llvm::StringRef(a.first, a.second));
 384   }
 385
 386   // This hopefully will just get inlined and removed by the optimizer.
 387   static const internal_key_type&
 388   GetInternalKey(const external_key_type& x) { return x; }
 389
 390   static std::pair<unsigned, unsigned>
 391   ReadKeyDataLength(const unsigned char*& d) {
 392     return std::make_pair((unsigned) ReadUnalignedLE16(d), sizeof(uint32_t));
 393   }
 394
 395   static std::pair<const char*, unsigned>
 396   ReadKey(const unsigned char* d, unsigned n) {
 397       assert(n >= 2 && d[n-1] == '\0');
 398       return std::make_pair((const char*) d, n-1);
 399     }
 400
 401   static uint32_t ReadData(const internal_key_type& k, const unsigned char* d,
 402                            unsigned) {
 403     return ::ReadUnalignedLE32(d);
 404   }
 405 };
 406
 407 } // end anonymous namespace
 408
 409 typedef OnDiskChainedHashTable<PTHFileLookupTrait>   PTHFileLookup;
 410 typedef OnDiskChainedHashTable<PTHStringLookupTrait> PTHStringIdLookup;
 411
 412 //===----------------------------------------------------------------------===//
 413 // PTHManager methods.
 414 //===----------------------------------------------------------------------===//
 415
 416 PTHManager::PTHManager(const llvm::MemoryBuffer* buf, void* fileLookup,
 417                        const unsigned char* idDataTable,
 418                        IdentifierInfo** perIDCache,
 419                        void* stringIdLookup, unsigned numIds,
 420                        const unsigned char* spellingBase,
 421                        const char* originalSourceFile)
 422 : Buf(buf), PerIDCache(perIDCache), FileLookup(fileLookup),
 423   IdDataTable(idDataTable), StringIdLookup(stringIdLookup),
 424   NumIds(numIds), PP(0), SpellingBase(spellingBase),
 425   OriginalSourceFile(originalSourceFile) {}
 426
 427 PTHManager::~PTHManager() {
 428   delete Buf;
 429   delete (PTHFileLookup*) FileLookup;
 430   delete (PTHStringIdLookup*) StringIdLookup;
 431   free(PerIDCache);
 432 }
 433
 434 static void InvalidPTH(Diagnostic &Diags, const char *Msg) {
 435   Diags.Report(Diags.getCustomDiagID(Diagnostic::Error, Msg));
 436 }
 437
 438 PTHManager *PTHManager::Create(const std::string &file, Diagnostic &Diags) {
 439   // Memory map the PTH file.
 440   llvm::OwningPtr<llvm::MemoryBuffer> File;
 441
 442   if (llvm::MemoryBuffer::getFile(file, File)) {
 443     // FIXME: Add ec.message() to this diag.
 444     Diags.Report(diag::err_invalid_pth_file) << file;
 445     return 0;
 446   }
 447
 448   // Get the buffer ranges and check if there are at least three 32-bit
 449   // words at the end of the file.
 450   const unsigned char *BufBeg = (unsigned char*)File->getBufferStart();
 451   const unsigned char *BufEnd = (unsigned char*)File->getBufferEnd();
 452
 453   // Check the prologue of the file.
 454   if ((BufEnd - BufBeg) < (signed)(sizeof("cfe-pth") + 3 + 4) ||
 455       memcmp(BufBeg, "cfe-pth", sizeof("cfe-pth") - 1) != 0) {
 456     Diags.Report(diag::err_invalid_pth_file) << file;
 457     return 0;
 458   }
 459
 460   // Read the PTH version.
 461   const unsigned char *p = BufBeg + (sizeof("cfe-pth") - 1);
 462   unsigned Version = ReadLE32(p);
 463
 464   if (Version < PTHManager::Version) {
 465     InvalidPTH(Diags,
 466         Version < PTHManager::Version
 467         ? "PTH file uses an older PTH format that is no longer supported"
 468         : "PTH file uses a newer PTH format that cannot be read");
 469     return 0;
 470   }
 471
 472   // Compute the address of the index table at the end of the PTH file.
 473   const unsigned char *PrologueOffset = p;
 474
 475   if (PrologueOffset >= BufEnd) {
 476     Diags.Report(diag::err_invalid_pth_file) << file;
 477     return 0;
 478   }
 479
 480   // Construct the file lookup table.  This will be used for mapping from
 481   // FileEntry*'s to cached tokens.
 482   const unsigned char* FileTableOffset = PrologueOffset + sizeof(uint32_t)*2;
 483   const unsigned char* FileTable = BufBeg + ReadLE32(FileTableOffset);
 484
 485   if (!(FileTable > BufBeg && FileTable < BufEnd)) {
 486     Diags.Report(diag::err_invalid_pth_file) << file;
 487     return 0; // FIXME: Proper error diagnostic?
 488   }
 489
 490   llvm::OwningPtr<PTHFileLookup> FL(PTHFileLookup::Create(FileTable, BufBeg));
 491
 492   // Warn if the PTH file is empty.  We still want to create a PTHManager
 493   // as the PTH could be used with -include-pth.
 494   if (FL->isEmpty())
 495     InvalidPTH(Diags, "PTH file contains no cached source data");
 496
 497   // Get the location of the table mapping from persistent ids to the
 498   // data needed to reconstruct identifiers.
 499   const unsigned char* IDTableOffset = PrologueOffset + sizeof(uint32_t)*0;
 500   const unsigned char* IData = BufBeg + ReadLE32(IDTableOffset);
 501
 502   if (!(IData >= BufBeg && IData < BufEnd)) {
 503     Diags.Report(diag::err_invalid_pth_file) << file;
 504     return 0;
 505   }
 506
 507   // Get the location of the hashtable mapping between strings and
 508   // persistent IDs.
 509   const unsigned char* StringIdTableOffset = PrologueOffset + sizeof(uint32_t)*1;
 510   const unsigned char* StringIdTable = BufBeg + ReadLE32(StringIdTableOffset);
 511   if (!(StringIdTable >= BufBeg && StringIdTable < BufEnd)) {
 512     Diags.Report(diag::err_invalid_pth_file) << file;
 513     return 0;
 514   }
 515
 516   llvm::OwningPtr<PTHStringIdLookup> SL(PTHStringIdLookup::Create(StringIdTable,
 517                                                                   BufBeg));
 518
 519   // Get the location of the spelling cache.
 520   const unsigned char* spellingBaseOffset = PrologueOffset + sizeof(uint32_t)*3;
 521   const unsigned char* spellingBase = BufBeg + ReadLE32(spellingBaseOffset);
 522   if (!(spellingBase >= BufBeg && spellingBase < BufEnd)) {
 523     Diags.Report(diag::err_invalid_pth_file) << file;
 524     return 0;
 525   }
 526
 527   // Get the number of IdentifierInfos and pre-allocate the identifier cache.
 528   uint32_t NumIds = ReadLE32(IData);
 529
 530   // Pre-allocate the peristent ID -> IdentifierInfo* cache.  We use calloc()
 531   // so that we in the best case only zero out memory once when the OS returns
 532   // us new pages.
 533   IdentifierInfo** PerIDCache = 0;
 534
 535   if (NumIds) {
 536     PerIDCache = (IdentifierInfo**)calloc(NumIds, sizeof(*PerIDCache));
 537     if (!PerIDCache) {
 538       InvalidPTH(Diags, "Could not allocate memory for processing PTH file");
 539       return 0;
 540     }
 541   }
 542
 543   // Compute the address of the original source file.
 544   const unsigned char* originalSourceBase = PrologueOffset + sizeof(uint32_t)*4;
 545   unsigned len = ReadUnalignedLE16(originalSourceBase);
 546   if (!len) originalSourceBase = 0;
 547
 548   // Create the new PTHManager.
 549   return new PTHManager(File.take(), FL.take(), IData, PerIDCache,
 550                         SL.take(), NumIds, spellingBase,
 551                         (const char*) originalSourceBase);
 552 }
 553
 554 IdentifierInfo* PTHManager::LazilyCreateIdentifierInfo(unsigned PersistentID) {
 555   // Look in the PTH file for the string data for the IdentifierInfo object.
 556   const unsigned char* TableEntry = IdDataTable + sizeof(uint32_t)*PersistentID;
 557   const unsigned char* IDData =
 558     (const unsigned char*)Buf->getBufferStart() + ReadLE32(TableEntry);
 559   assert(IDData < (const unsigned char*)Buf->getBufferEnd());
 560
 561   // Allocate the object.
 562   std::pair<IdentifierInfo,const unsigned char*> *Mem =
 563     Alloc.Allocate<std::pair<IdentifierInfo,const unsigned char*> >();
 564
 565   Mem->second = IDData;
 566   assert(IDData[0] != '\0');
 567   IdentifierInfo *II = new ((void*) Mem) IdentifierInfo();
 568
 569   // Store the new IdentifierInfo in the cache.
 570   PerIDCache[PersistentID] = II;
 571   assert(II->getNameStart() && II->getNameStart()[0] != '\0');
 572   return II;
 573 }
 574
 575 IdentifierInfo* PTHManager::get(llvm::StringRef Name) {
 576   PTHStringIdLookup& SL = *((PTHStringIdLookup*)StringIdLookup);
 577   // Double check our assumption that the last character isn't '\0'.
 578   assert(Name.empty() || Name.data()[Name.size()-1] != '\0');
 579   PTHStringIdLookup::iterator I = SL.find(std::make_pair(Name.data(),
 580                                                          Name.size()));
 581   if (I == SL.end()) // No identifier found?
 582     return 0;
 583
 584   // Match found.  Return the identifier!
 585   assert(*I > 0);
 586   return GetIdentifierInfo(*I-1);
 587 }
 588
 589 PTHLexer *PTHManager::CreateLexer(FileID FID) {
 590   const FileEntry *FE = PP->getSourceManager().getFileEntryForID(FID);
 591   if (!FE)
 592     return 0;
 593
 594   // Lookup the FileEntry object in our file lookup data structure.  It will
 595   // return a variant that indicates whether or not there is an offset within
 596   // the PTH file that contains cached tokens.
 597   PTHFileLookup& PFL = *((PTHFileLookup*)FileLookup);
 598   PTHFileLookup::iterator I = PFL.find(FE);
 599
 600   if (I == PFL.end()) // No tokens available?
 601     return 0;
 602
 603   const PTHFileData& FileData = *I;
 604
 605   const unsigned char *BufStart = (const unsigned char *)Buf->getBufferStart();
 606   // Compute the offset of the token data within the buffer.
 607   const unsigned char* data = BufStart + FileData.getTokenOffset();
 608
 609   // Get the location of pp-conditional table.
 610   const unsigned char* ppcond = BufStart + FileData.getPPCondOffset();
 611   uint32_t Len = ReadLE32(ppcond);
 612   if (Len == 0) ppcond = 0;
 613
 614   assert(PP && "No preprocessor set yet!");
 615   return new PTHLexer(*PP, FID, data, ppcond, *this);
 616 }
 617
 618 //===----------------------------------------------------------------------===//
 619 // 'stat' caching.
 620 //===----------------------------------------------------------------------===//
 621
 622 namespace {
 623 class PTHStatData {
 624 public:
 625   const bool hasStat;
 626   const ino_t ino;
 627   const dev_t dev;
 628   const mode_t mode;
 629   const time_t mtime;
 630   const off_t size;
 631
 632   PTHStatData(ino_t i, dev_t d, mode_t mo, time_t m, off_t s)
 633   : hasStat(true), ino(i), dev(d), mode(mo), mtime(m), size(s) {}
 634
 635   PTHStatData()
 636     : hasStat(false), ino(0), dev(0), mode(0), mtime(0), size(0) {}
 637 };
 638
 639 class PTHStatLookupTrait : public PTHFileLookupCommonTrait {
 640 public:
 641   typedef const char* external_key_type;  // const char*
 642   typedef PTHStatData data_type;
 643
 644   static internal_key_type GetInternalKey(const char *path) {
 645     // The key 'kind' doesn't matter here because it is ignored in EqualKey.
 646     return std::make_pair((unsigned char) 0x0, path);
 647   }
 648
 649   static bool EqualKey(internal_key_type a, internal_key_type b) {
 650     // When doing 'stat' lookups we don't care about the kind of 'a' and 'b',
 651     // just the paths.
 652     return strcmp(a.second, b.second) == 0;
 653   }
 654
 655   static data_type ReadData(const internal_key_type& k, const unsigned char* d,
 656                             unsigned) {
 657
 658     if (k.first /* File or Directory */) {
 659       if (k.first == 0x1 /* File */) d += 4 * 2; // Skip the first 2 words.
 660       ino_t ino = (ino_t) ReadUnalignedLE32(d);
 661       dev_t dev = (dev_t) ReadUnalignedLE32(d);
 662       mode_t mode = (mode_t) ReadUnalignedLE16(d);
 663       time_t mtime = (time_t) ReadUnalignedLE64(d);
 664       return data_type(ino, dev, mode, mtime, (off_t) ReadUnalignedLE64(d));
 665     }
 666
 667     // Negative stat.  Don't read anything.
 668     return data_type();
 669   }
 670 };
 671
 672 class PTHStatCache : public FileSystemStatCache {
 673   typedef OnDiskChainedHashTable<PTHStatLookupTrait> CacheTy;
 674   CacheTy Cache;
 675
 676 public:
 677   PTHStatCache(PTHFileLookup &FL) :
 678     Cache(FL.getNumBuckets(), FL.getNumEntries(), FL.getBuckets(),
 679           FL.getBase()) {}
 680
 681   ~PTHStatCache() {}
 682
 683   LookupResult getStat(const char *Path, struct stat &StatBuf,
 684                        int *FileDescriptor) {
 685     // Do the lookup for the file's data in the PTH file.
 686     CacheTy::iterator I = Cache.find(Path);
 687
 688     // If we don't get a hit in the PTH file just forward to 'stat'.
 689     if (I == Cache.end())
 690       return statChained(Path, StatBuf, FileDescriptor);
 691
 692     const PTHStatData &Data = *I;
 693
 694     if (!Data.hasStat)
 695       return CacheMissing;
 696
 697     StatBuf.st_ino = Data.ino;
 698     StatBuf.st_dev = Data.dev;
 699     StatBuf.st_mtime = Data.mtime;
 700     StatBuf.st_mode = Data.mode;
 701     StatBuf.st_size = Data.size;
 702     return CacheExists;
 703   }
 704 };
 705 } // end anonymous namespace
 706
 707 FileSystemStatCache *PTHManager::createStatCache() {
 708   return new PTHStatCache(*((PTHFileLookup*) FileLookup));
 709 }