CacheTokens.cpp revision b09f6e15c59b89d5820db8ef40598eb1d1323c1f
1//===--- CacheTokens.cpp - Caching of lexer tokens for PTH support --------===// 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 provides a possible implementation of PTH support for Clang that is 11// based on caching lexed tokens and identifiers. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Frontend/Utils.h" 16#include "clang/Basic/FileManager.h" 17#include "clang/Basic/SourceManager.h" 18#include "clang/Basic/IdentifierTable.h" 19#include "clang/Basic/Diagnostic.h" 20#include "clang/Basic/OnDiskHashTable.h" 21#include "clang/Lex/Lexer.h" 22#include "clang/Lex/Preprocessor.h" 23#include "llvm/ADT/StringMap.h" 24#include "llvm/Support/MemoryBuffer.h" 25#include "llvm/Support/raw_ostream.h" 26#include "llvm/System/Path.h" 27#include "llvm/Support/Compiler.h" 28#include "llvm/Support/Streams.h" 29 30// FIXME: put this somewhere else? 31#ifndef S_ISDIR 32#define S_ISDIR(x) (((x)&_S_IFDIR)!=0) 33#endif 34 35using namespace clang; 36using namespace clang::io; 37 38//===----------------------------------------------------------------------===// 39// PTH-specific stuff. 40//===----------------------------------------------------------------------===// 41 42namespace { 43class VISIBILITY_HIDDEN PTHEntry { 44 Offset TokenData, PPCondData; 45 46public: 47 PTHEntry() {} 48 49 PTHEntry(Offset td, Offset ppcd) 50 : TokenData(td), PPCondData(ppcd) {} 51 52 Offset getTokenOffset() const { return TokenData; } 53 Offset getPPCondTableOffset() const { return PPCondData; } 54}; 55 56 57class VISIBILITY_HIDDEN PTHEntryKeyVariant { 58 union { const FileEntry* FE; const char* Path; }; 59 enum { IsFE = 0x1, IsDE = 0x2, IsNoExist = 0x0 } Kind; 60 struct stat *StatBuf; 61public: 62 PTHEntryKeyVariant(const FileEntry *fe) 63 : FE(fe), Kind(IsFE), StatBuf(0) {} 64 65 PTHEntryKeyVariant(struct stat* statbuf, const char* path) 66 : Path(path), Kind(IsDE), StatBuf(new struct stat(*statbuf)) {} 67 68 PTHEntryKeyVariant(const char* path) 69 : Path(path), Kind(IsNoExist), StatBuf(0) {} 70 71 bool isFile() const { return Kind == IsFE; } 72 73 const char* getCString() const { 74 return Kind == IsFE ? FE->getName() : Path; 75 } 76 77 unsigned getKind() const { return (unsigned) Kind; } 78 79 void EmitData(llvm::raw_ostream& Out) { 80 switch (Kind) { 81 case IsFE: 82 // Emit stat information. 83 ::Emit32(Out, FE->getInode()); 84 ::Emit32(Out, FE->getDevice()); 85 ::Emit16(Out, FE->getFileMode()); 86 ::Emit64(Out, FE->getModificationTime()); 87 ::Emit64(Out, FE->getSize()); 88 break; 89 case IsDE: 90 // Emit stat information. 91 ::Emit32(Out, (uint32_t) StatBuf->st_ino); 92 ::Emit32(Out, (uint32_t) StatBuf->st_dev); 93 ::Emit16(Out, (uint16_t) StatBuf->st_mode); 94 ::Emit64(Out, (uint64_t) StatBuf->st_mtime); 95 ::Emit64(Out, (uint64_t) StatBuf->st_size); 96 delete StatBuf; 97 break; 98 default: 99 break; 100 } 101 } 102 103 unsigned getRepresentationLength() const { 104 return Kind == IsNoExist ? 0 : 4 + 4 + 2 + 8 + 8; 105 } 106}; 107 108class VISIBILITY_HIDDEN FileEntryPTHEntryInfo { 109public: 110 typedef PTHEntryKeyVariant key_type; 111 typedef key_type key_type_ref; 112 113 typedef PTHEntry data_type; 114 typedef const PTHEntry& data_type_ref; 115 116 static unsigned ComputeHash(PTHEntryKeyVariant V) { 117 return BernsteinHash(V.getCString()); 118 } 119 120 static std::pair<unsigned,unsigned> 121 EmitKeyDataLength(llvm::raw_ostream& Out, PTHEntryKeyVariant V, 122 const PTHEntry& E) { 123 124 unsigned n = strlen(V.getCString()) + 1 + 1; 125 ::Emit16(Out, n); 126 127 unsigned m = V.getRepresentationLength() + (V.isFile() ? 4 + 4 : 0); 128 ::Emit8(Out, m); 129 130 return std::make_pair(n, m); 131 } 132 133 static void EmitKey(llvm::raw_ostream& Out, PTHEntryKeyVariant V, unsigned n){ 134 // Emit the entry kind. 135 ::Emit8(Out, (unsigned) V.getKind()); 136 // Emit the string. 137 Out.write(V.getCString(), n - 1); 138 } 139 140 static void EmitData(llvm::raw_ostream& Out, PTHEntryKeyVariant V, 141 const PTHEntry& E, unsigned) { 142 143 144 // For file entries emit the offsets into the PTH file for token data 145 // and the preprocessor blocks table. 146 if (V.isFile()) { 147 ::Emit32(Out, E.getTokenOffset()); 148 ::Emit32(Out, E.getPPCondTableOffset()); 149 } 150 151 // Emit any other data associated with the key (i.e., stat information). 152 V.EmitData(Out); 153 } 154}; 155 156class OffsetOpt { 157 bool valid; 158 Offset off; 159public: 160 OffsetOpt() : valid(false) {} 161 bool hasOffset() const { return valid; } 162 Offset getOffset() const { assert(valid); return off; } 163 void setOffset(Offset o) { off = o; valid = true; } 164}; 165} // end anonymous namespace 166 167typedef OnDiskChainedHashTableGenerator<FileEntryPTHEntryInfo> PTHMap; 168typedef llvm::DenseMap<const IdentifierInfo*,uint32_t> IDMap; 169typedef llvm::StringMap<OffsetOpt, llvm::BumpPtrAllocator> CachedStrsTy; 170 171namespace { 172class VISIBILITY_HIDDEN PTHWriter { 173 IDMap IM; 174 llvm::raw_fd_ostream& Out; 175 Preprocessor& PP; 176 uint32_t idcount; 177 PTHMap PM; 178 CachedStrsTy CachedStrs; 179 Offset CurStrOffset; 180 std::vector<llvm::StringMapEntry<OffsetOpt>*> StrEntries; 181 182 //// Get the persistent id for the given IdentifierInfo*. 183 uint32_t ResolveID(const IdentifierInfo* II); 184 185 /// Emit a token to the PTH file. 186 void EmitToken(const Token& T); 187 188 void Emit8(uint32_t V) { 189 Out << (unsigned char)(V); 190 } 191 192 void Emit16(uint32_t V) { ::Emit16(Out, V); } 193 194 void Emit24(uint32_t V) { 195 Out << (unsigned char)(V); 196 Out << (unsigned char)(V >> 8); 197 Out << (unsigned char)(V >> 16); 198 assert((V >> 24) == 0); 199 } 200 201 void Emit32(uint32_t V) { ::Emit32(Out, V); } 202 203 void EmitBuf(const char *Ptr, unsigned NumBytes) { 204 Out.write(Ptr, NumBytes); 205 } 206 207 /// EmitIdentifierTable - Emits two tables to the PTH file. The first is 208 /// a hashtable mapping from identifier strings to persistent IDs. 209 /// The second is a straight table mapping from persistent IDs to string data 210 /// (the keys of the first table). 211 std::pair<Offset, Offset> EmitIdentifierTable(); 212 213 /// EmitFileTable - Emit a table mapping from file name strings to PTH 214 /// token data. 215 Offset EmitFileTable() { return PM.Emit(Out); } 216 217 PTHEntry LexTokens(Lexer& L); 218 Offset EmitCachedSpellings(); 219 220public: 221 PTHWriter(llvm::raw_fd_ostream& out, Preprocessor& pp) 222 : Out(out), PP(pp), idcount(0), CurStrOffset(0) {} 223 224 PTHMap &getPM() { return PM; } 225 void GeneratePTH(const std::string *MainFile = 0); 226}; 227} // end anonymous namespace 228 229uint32_t PTHWriter::ResolveID(const IdentifierInfo* II) { 230 // Null IdentifierInfo's map to the persistent ID 0. 231 if (!II) 232 return 0; 233 234 IDMap::iterator I = IM.find(II); 235 if (I != IM.end()) 236 return I->second; // We've already added 1. 237 238 IM[II] = ++idcount; // Pre-increment since '0' is reserved for NULL. 239 return idcount; 240} 241 242void PTHWriter::EmitToken(const Token& T) { 243 // Emit the token kind, flags, and length. 244 Emit32(((uint32_t) T.getKind()) | ((((uint32_t) T.getFlags())) << 8)| 245 (((uint32_t) T.getLength()) << 16)); 246 247 if (!T.isLiteral()) { 248 Emit32(ResolveID(T.getIdentifierInfo())); 249 } else { 250 // We cache *un-cleaned* spellings. This gives us 100% fidelity with the 251 // source code. 252 const char* s = T.getLiteralData(); 253 unsigned len = T.getLength(); 254 255 // Get the string entry. 256 llvm::StringMapEntry<OffsetOpt> *E = &CachedStrs.GetOrCreateValue(s, s+len); 257 258 // If this is a new string entry, bump the PTH offset. 259 if (!E->getValue().hasOffset()) { 260 E->getValue().setOffset(CurStrOffset); 261 StrEntries.push_back(E); 262 CurStrOffset += len + 1; 263 } 264 265 // Emit the relative offset into the PTH file for the spelling string. 266 Emit32(E->getValue().getOffset()); 267 } 268 269 // Emit the offset into the original source file of this token so that we 270 // can reconstruct its SourceLocation. 271 Emit32(PP.getSourceManager().getFileOffset(T.getLocation())); 272} 273 274PTHEntry PTHWriter::LexTokens(Lexer& L) { 275 // Pad 0's so that we emit tokens to a 4-byte alignment. 276 // This speed up reading them back in. 277 Pad(Out, 4); 278 Offset off = (Offset) Out.tell(); 279 280 // Keep track of matching '#if' ... '#endif'. 281 typedef std::vector<std::pair<Offset, unsigned> > PPCondTable; 282 PPCondTable PPCond; 283 std::vector<unsigned> PPStartCond; 284 bool ParsingPreprocessorDirective = false; 285 Token Tok; 286 287 do { 288 L.LexFromRawLexer(Tok); 289 NextToken: 290 291 if ((Tok.isAtStartOfLine() || Tok.is(tok::eof)) && 292 ParsingPreprocessorDirective) { 293 // Insert an eom token into the token cache. It has the same 294 // position as the next token that is not on the same line as the 295 // preprocessor directive. Observe that we continue processing 296 // 'Tok' when we exit this branch. 297 Token Tmp = Tok; 298 Tmp.setKind(tok::eom); 299 Tmp.clearFlag(Token::StartOfLine); 300 Tmp.setIdentifierInfo(0); 301 EmitToken(Tmp); 302 ParsingPreprocessorDirective = false; 303 } 304 305 if (Tok.is(tok::identifier)) { 306 Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok)); 307 EmitToken(Tok); 308 continue; 309 } 310 311 if (Tok.is(tok::hash) && Tok.isAtStartOfLine()) { 312 // Special processing for #include. Store the '#' token and lex 313 // the next token. 314 assert(!ParsingPreprocessorDirective); 315 Offset HashOff = (Offset) Out.tell(); 316 EmitToken(Tok); 317 318 // Get the next token. 319 L.LexFromRawLexer(Tok); 320 321 // If we see the start of line, then we had a null directive "#". 322 if (Tok.isAtStartOfLine()) 323 goto NextToken; 324 325 // Did we see 'include'/'import'/'include_next'? 326 if (Tok.isNot(tok::identifier)) { 327 EmitToken(Tok); 328 continue; 329 } 330 331 IdentifierInfo* II = PP.LookUpIdentifierInfo(Tok); 332 Tok.setIdentifierInfo(II); 333 tok::PPKeywordKind K = II->getPPKeywordID(); 334 335 ParsingPreprocessorDirective = true; 336 337 switch (K) { 338 case tok::pp_not_keyword: 339 // Invalid directives "#foo" can occur in #if 0 blocks etc, just pass 340 // them through. 341 default: 342 break; 343 344 case tok::pp_include: 345 case tok::pp_import: 346 case tok::pp_include_next: { 347 // Save the 'include' token. 348 EmitToken(Tok); 349 // Lex the next token as an include string. 350 L.setParsingPreprocessorDirective(true); 351 L.LexIncludeFilename(Tok); 352 L.setParsingPreprocessorDirective(false); 353 assert(!Tok.isAtStartOfLine()); 354 if (Tok.is(tok::identifier)) 355 Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok)); 356 357 break; 358 } 359 case tok::pp_if: 360 case tok::pp_ifdef: 361 case tok::pp_ifndef: { 362 // Add an entry for '#if' and friends. We initially set the target 363 // index to 0. This will get backpatched when we hit #endif. 364 PPStartCond.push_back(PPCond.size()); 365 PPCond.push_back(std::make_pair(HashOff, 0U)); 366 break; 367 } 368 case tok::pp_endif: { 369 // Add an entry for '#endif'. We set the target table index to itself. 370 // This will later be set to zero when emitting to the PTH file. We 371 // use 0 for uninitialized indices because that is easier to debug. 372 unsigned index = PPCond.size(); 373 // Backpatch the opening '#if' entry. 374 assert(!PPStartCond.empty()); 375 assert(PPCond.size() > PPStartCond.back()); 376 assert(PPCond[PPStartCond.back()].second == 0); 377 PPCond[PPStartCond.back()].second = index; 378 PPStartCond.pop_back(); 379 // Add the new entry to PPCond. 380 PPCond.push_back(std::make_pair(HashOff, index)); 381 EmitToken(Tok); 382 383 // Some files have gibberish on the same line as '#endif'. 384 // Discard these tokens. 385 do 386 L.LexFromRawLexer(Tok); 387 while (Tok.isNot(tok::eof) && !Tok.isAtStartOfLine()); 388 // We have the next token in hand. 389 // Don't immediately lex the next one. 390 goto NextToken; 391 } 392 case tok::pp_elif: 393 case tok::pp_else: { 394 // Add an entry for #elif or #else. 395 // This serves as both a closing and opening of a conditional block. 396 // This means that its entry will get backpatched later. 397 unsigned index = PPCond.size(); 398 // Backpatch the previous '#if' entry. 399 assert(!PPStartCond.empty()); 400 assert(PPCond.size() > PPStartCond.back()); 401 assert(PPCond[PPStartCond.back()].second == 0); 402 PPCond[PPStartCond.back()].second = index; 403 PPStartCond.pop_back(); 404 // Now add '#elif' as a new block opening. 405 PPCond.push_back(std::make_pair(HashOff, 0U)); 406 PPStartCond.push_back(index); 407 break; 408 } 409 } 410 } 411 412 EmitToken(Tok); 413 } 414 while (Tok.isNot(tok::eof)); 415 416 assert(PPStartCond.empty() && "Error: imblanced preprocessor conditionals."); 417 418 // Next write out PPCond. 419 Offset PPCondOff = (Offset) Out.tell(); 420 421 // Write out the size of PPCond so that clients can identifer empty tables. 422 Emit32(PPCond.size()); 423 424 for (unsigned i = 0, e = PPCond.size(); i!=e; ++i) { 425 Emit32(PPCond[i].first - off); 426 uint32_t x = PPCond[i].second; 427 assert(x != 0 && "PPCond entry not backpatched."); 428 // Emit zero for #endifs. This allows us to do checking when 429 // we read the PTH file back in. 430 Emit32(x == i ? 0 : x); 431 } 432 433 return PTHEntry(off, PPCondOff); 434} 435 436Offset PTHWriter::EmitCachedSpellings() { 437 // Write each cached strings to the PTH file. 438 Offset SpellingsOff = Out.tell(); 439 440 for (std::vector<llvm::StringMapEntry<OffsetOpt>*>::iterator 441 I = StrEntries.begin(), E = StrEntries.end(); I!=E; ++I) 442 EmitBuf((*I)->getKeyData(), (*I)->getKeyLength()+1 /*nul included*/); 443 444 return SpellingsOff; 445} 446 447void PTHWriter::GeneratePTH(const std::string *MainFile) { 448 // Generate the prologue. 449 Out << "cfe-pth"; 450 Emit32(PTHManager::Version); 451 452 // Leave 4 words for the prologue. 453 Offset PrologueOffset = Out.tell(); 454 for (unsigned i = 0; i < 4; ++i) 455 Emit32(0); 456 457 // Write the name of the MainFile. 458 if (MainFile && !MainFile->empty()) { 459 Emit16(MainFile->length()); 460 EmitBuf(MainFile->data(), MainFile->length()); 461 } else { 462 // String with 0 bytes. 463 Emit16(0); 464 } 465 Emit8(0); 466 467 // Iterate over all the files in SourceManager. Create a lexer 468 // for each file and cache the tokens. 469 SourceManager &SM = PP.getSourceManager(); 470 const LangOptions &LOpts = PP.getLangOptions(); 471 472 for (SourceManager::fileinfo_iterator I = SM.fileinfo_begin(), 473 E = SM.fileinfo_end(); I != E; ++I) { 474 const SrcMgr::ContentCache &C = *I->second; 475 const FileEntry *FE = C.Entry; 476 477 // FIXME: Handle files with non-absolute paths. 478 llvm::sys::Path P(FE->getName()); 479 if (!P.isAbsolute()) 480 continue; 481 482 const llvm::MemoryBuffer *B = C.getBuffer(); 483 if (!B) continue; 484 485 FileID FID = SM.createFileID(FE, SourceLocation(), SrcMgr::C_User); 486 Lexer L(FID, SM, LOpts); 487 PM.insert(FE, LexTokens(L)); 488 } 489 490 // Write out the identifier table. 491 const std::pair<Offset,Offset> &IdTableOff = EmitIdentifierTable(); 492 493 // Write out the cached strings table. 494 Offset SpellingOff = EmitCachedSpellings(); 495 496 // Write out the file table. 497 Offset FileTableOff = EmitFileTable(); 498 499 // Finally, write the prologue. 500 Out.seek(PrologueOffset); 501 Emit32(IdTableOff.first); 502 Emit32(IdTableOff.second); 503 Emit32(FileTableOff); 504 Emit32(SpellingOff); 505} 506 507namespace { 508/// StatListener - A simple "interpose" object used to monitor stat calls 509/// invoked by FileManager while processing the original sources used 510/// as input to PTH generation. StatListener populates the PTHWriter's 511/// file map with stat information for directories as well as negative stats. 512/// Stat information for files are populated elsewhere. 513class StatListener : public StatSysCallCache { 514 PTHMap &PM; 515public: 516 StatListener(PTHMap &pm) : PM(pm) {} 517 ~StatListener() {} 518 519 int stat(const char *path, struct stat *buf) { 520 int result = ::stat(path, buf); 521 522 if (result != 0) // Failed 'stat'. 523 PM.insert(path, PTHEntry()); 524 else if (S_ISDIR(buf->st_mode)) { 525 // Only cache directories with absolute paths. 526 if (!llvm::sys::Path(path).isAbsolute()) 527 return result; 528 529 PM.insert(PTHEntryKeyVariant(buf, path), PTHEntry()); 530 } 531 532 return result; 533 } 534}; 535} // end anonymous namespace 536 537 538void clang::CacheTokens(Preprocessor &PP, llvm::raw_fd_ostream* OS) { 539 // Get the name of the main file. 540 const SourceManager &SrcMgr = PP.getSourceManager(); 541 const FileEntry *MainFile = SrcMgr.getFileEntryForID(SrcMgr.getMainFileID()); 542 llvm::sys::Path MainFilePath(MainFile->getName()); 543 std::string MainFileName; 544 545 if (!MainFilePath.isAbsolute()) { 546 llvm::sys::Path P = llvm::sys::Path::GetCurrentDirectory(); 547 P.appendComponent(MainFilePath.toString()); 548 MainFileName = P.toString(); 549 } else { 550 MainFileName = MainFilePath.toString(); 551 } 552 553 // Create the PTHWriter. 554 PTHWriter PW(*OS, PP); 555 556 // Install the 'stat' system call listener in the FileManager. 557 PP.getFileManager().setStatCache(new StatListener(PW.getPM())); 558 559 // Lex through the entire file. This will populate SourceManager with 560 // all of the header information. 561 Token Tok; 562 PP.EnterMainSourceFile(); 563 do { PP.Lex(Tok); } while (Tok.isNot(tok::eof)); 564 565 // Generate the PTH file. 566 PP.getFileManager().setStatCache(0); 567 PW.GeneratePTH(&MainFileName); 568} 569 570//===----------------------------------------------------------------------===// 571 572class PTHIdKey { 573public: 574 const IdentifierInfo* II; 575 uint32_t FileOffset; 576}; 577 578namespace { 579class VISIBILITY_HIDDEN PTHIdentifierTableTrait { 580public: 581 typedef PTHIdKey* key_type; 582 typedef key_type key_type_ref; 583 584 typedef uint32_t data_type; 585 typedef data_type data_type_ref; 586 587 static unsigned ComputeHash(PTHIdKey* key) { 588 return BernsteinHash(key->II->getName()); 589 } 590 591 static std::pair<unsigned,unsigned> 592 EmitKeyDataLength(llvm::raw_ostream& Out, const PTHIdKey* key, uint32_t) { 593 unsigned n = strlen(key->II->getName()) + 1; 594 ::Emit16(Out, n); 595 return std::make_pair(n, sizeof(uint32_t)); 596 } 597 598 static void EmitKey(llvm::raw_ostream& Out, PTHIdKey* key, unsigned n) { 599 // Record the location of the key data. This is used when generating 600 // the mapping from persistent IDs to strings. 601 key->FileOffset = Out.tell(); 602 Out.write(key->II->getName(), n); 603 } 604 605 static void EmitData(llvm::raw_ostream& Out, PTHIdKey*, uint32_t pID, 606 unsigned) { 607 ::Emit32(Out, pID); 608 } 609}; 610} // end anonymous namespace 611 612/// EmitIdentifierTable - Emits two tables to the PTH file. The first is 613/// a hashtable mapping from identifier strings to persistent IDs. The second 614/// is a straight table mapping from persistent IDs to string data (the 615/// keys of the first table). 616/// 617std::pair<Offset,Offset> PTHWriter::EmitIdentifierTable() { 618 // Build two maps: 619 // (1) an inverse map from persistent IDs -> (IdentifierInfo*,Offset) 620 // (2) a map from (IdentifierInfo*, Offset)* -> persistent IDs 621 622 // Note that we use 'calloc', so all the bytes are 0. 623 PTHIdKey *IIDMap = (PTHIdKey*)calloc(idcount, sizeof(PTHIdKey)); 624 625 // Create the hashtable. 626 OnDiskChainedHashTableGenerator<PTHIdentifierTableTrait> IIOffMap; 627 628 // Generate mapping from persistent IDs -> IdentifierInfo*. 629 for (IDMap::iterator I = IM.begin(), E = IM.end(); I != E; ++I) { 630 // Decrement by 1 because we are using a vector for the lookup and 631 // 0 is reserved for NULL. 632 assert(I->second > 0); 633 assert(I->second-1 < idcount); 634 unsigned idx = I->second-1; 635 636 // Store the mapping from persistent ID to IdentifierInfo* 637 IIDMap[idx].II = I->first; 638 639 // Store the reverse mapping in a hashtable. 640 IIOffMap.insert(&IIDMap[idx], I->second); 641 } 642 643 // Write out the inverse map first. This causes the PCIDKey entries to 644 // record PTH file offsets for the string data. This is used to write 645 // the second table. 646 Offset StringTableOffset = IIOffMap.Emit(Out); 647 648 // Now emit the table mapping from persistent IDs to PTH file offsets. 649 Offset IDOff = Out.tell(); 650 Emit32(idcount); // Emit the number of identifiers. 651 for (unsigned i = 0 ; i < idcount; ++i) 652 Emit32(IIDMap[i].FileOffset); 653 654 // Finally, release the inverse map. 655 free(IIDMap); 656 657 return std::make_pair(IDOff, StringTableOffset); 658} 659