CacheTokens.cpp revision 6cb7c1a43b0c8f739d1f54b7fdae5ede86033496
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/Support/Compiler.h" 24#include "llvm/Support/MemoryBuffer.h" 25#include "llvm/Support/raw_ostream.h" 26#include "llvm/System/Path.h" 27#include "llvm/ADT/StringMap.h" 28 29// FIXME: put this somewhere else? 30#ifndef S_ISDIR 31#define S_ISDIR(x) (((x)&_S_IFDIR)!=0) 32#endif 33 34using namespace clang; 35using namespace clang::io; 36 37//===----------------------------------------------------------------------===// 38// PTH-specific stuff. 39//===----------------------------------------------------------------------===// 40 41namespace { 42class VISIBILITY_HIDDEN PTHEntry { 43 Offset TokenData, PPCondData; 44 45public: 46 PTHEntry() {} 47 48 PTHEntry(Offset td, Offset ppcd) 49 : TokenData(td), PPCondData(ppcd) {} 50 51 Offset getTokenOffset() const { return TokenData; } 52 Offset getPPCondTableOffset() const { return PPCondData; } 53}; 54 55 56class VISIBILITY_HIDDEN PTHEntryKeyVariant { 57 union { const FileEntry* FE; const char* Path; }; 58 enum { IsFE = 0x1, IsDE = 0x2, IsNoExist = 0x0 } Kind; 59 struct stat *StatBuf; 60public: 61 PTHEntryKeyVariant(const FileEntry *fe) 62 : FE(fe), Kind(IsFE), StatBuf(0) {} 63 64 PTHEntryKeyVariant(struct stat* statbuf, const char* path) 65 : Path(path), Kind(IsDE), StatBuf(new struct stat(*statbuf)) {} 66 67 PTHEntryKeyVariant(const char* path) 68 : Path(path), Kind(IsNoExist), StatBuf(0) {} 69 70 bool isFile() const { return Kind == IsFE; } 71 72 const char* getCString() const { 73 return Kind == IsFE ? FE->getName() : Path; 74 } 75 76 unsigned getKind() const { return (unsigned) Kind; } 77 78 void EmitData(llvm::raw_ostream& Out) { 79 switch (Kind) { 80 case IsFE: 81 // Emit stat information. 82 ::Emit32(Out, FE->getInode()); 83 ::Emit32(Out, FE->getDevice()); 84 ::Emit16(Out, FE->getFileMode()); 85 ::Emit64(Out, FE->getModificationTime()); 86 ::Emit64(Out, FE->getSize()); 87 break; 88 case IsDE: 89 // Emit stat information. 90 ::Emit32(Out, (uint32_t) StatBuf->st_ino); 91 ::Emit32(Out, (uint32_t) StatBuf->st_dev); 92 ::Emit16(Out, (uint16_t) StatBuf->st_mode); 93 ::Emit64(Out, (uint64_t) StatBuf->st_mtime); 94 ::Emit64(Out, (uint64_t) StatBuf->st_size); 95 delete StatBuf; 96 break; 97 default: 98 break; 99 } 100 } 101 102 unsigned getRepresentationLength() const { 103 return Kind == IsNoExist ? 0 : 4 + 4 + 2 + 8 + 8; 104 } 105}; 106 107class VISIBILITY_HIDDEN FileEntryPTHEntryInfo { 108public: 109 typedef PTHEntryKeyVariant key_type; 110 typedef key_type key_type_ref; 111 112 typedef PTHEntry data_type; 113 typedef const PTHEntry& data_type_ref; 114 115 static unsigned ComputeHash(PTHEntryKeyVariant V) { 116 return BernsteinHash(V.getCString()); 117 } 118 119 static std::pair<unsigned,unsigned> 120 EmitKeyDataLength(llvm::raw_ostream& Out, PTHEntryKeyVariant V, 121 const PTHEntry& E) { 122 123 unsigned n = strlen(V.getCString()) + 1 + 1; 124 ::Emit16(Out, n); 125 126 unsigned m = V.getRepresentationLength() + (V.isFile() ? 4 + 4 : 0); 127 ::Emit8(Out, m); 128 129 return std::make_pair(n, m); 130 } 131 132 static void EmitKey(llvm::raw_ostream& Out, PTHEntryKeyVariant V, unsigned n){ 133 // Emit the entry kind. 134 ::Emit8(Out, (unsigned) V.getKind()); 135 // Emit the string. 136 Out.write(V.getCString(), n - 1); 137 } 138 139 static void EmitData(llvm::raw_ostream& Out, PTHEntryKeyVariant V, 140 const PTHEntry& E, unsigned) { 141 142 143 // For file entries emit the offsets into the PTH file for token data 144 // and the preprocessor blocks table. 145 if (V.isFile()) { 146 ::Emit32(Out, E.getTokenOffset()); 147 ::Emit32(Out, E.getPPCondTableOffset()); 148 } 149 150 // Emit any other data associated with the key (i.e., stat information). 151 V.EmitData(Out); 152 } 153}; 154 155class OffsetOpt { 156 bool valid; 157 Offset off; 158public: 159 OffsetOpt() : valid(false) {} 160 bool hasOffset() const { return valid; } 161 Offset getOffset() const { assert(valid); return off; } 162 void setOffset(Offset o) { off = o; valid = true; } 163}; 164} // end anonymous namespace 165 166typedef OnDiskChainedHashTableGenerator<FileEntryPTHEntryInfo> PTHMap; 167typedef llvm::DenseMap<const IdentifierInfo*,uint32_t> IDMap; 168typedef llvm::StringMap<OffsetOpt, llvm::BumpPtrAllocator> CachedStrsTy; 169 170namespace { 171class VISIBILITY_HIDDEN PTHWriter { 172 IDMap IM; 173 llvm::raw_fd_ostream& Out; 174 Preprocessor& PP; 175 uint32_t idcount; 176 PTHMap PM; 177 CachedStrsTy CachedStrs; 178 Offset CurStrOffset; 179 std::vector<llvm::StringMapEntry<OffsetOpt>*> StrEntries; 180 181 //// Get the persistent id for the given IdentifierInfo*. 182 uint32_t ResolveID(const IdentifierInfo* II); 183 184 /// Emit a token to the PTH file. 185 void EmitToken(const Token& T); 186 187 void Emit8(uint32_t V) { 188 Out << (unsigned char)(V); 189 } 190 191 void Emit16(uint32_t V) { ::Emit16(Out, V); } 192 193 void Emit24(uint32_t V) { 194 Out << (unsigned char)(V); 195 Out << (unsigned char)(V >> 8); 196 Out << (unsigned char)(V >> 16); 197 assert((V >> 24) == 0); 198 } 199 200 void Emit32(uint32_t V) { ::Emit32(Out, V); } 201 202 void EmitBuf(const char *Ptr, unsigned NumBytes) { 203 Out.write(Ptr, NumBytes); 204 } 205 206 /// EmitIdentifierTable - Emits two tables to the PTH file. The first is 207 /// a hashtable mapping from identifier strings to persistent IDs. 208 /// The second is a straight table mapping from persistent IDs to string data 209 /// (the keys of the first table). 210 std::pair<Offset, Offset> EmitIdentifierTable(); 211 212 /// EmitFileTable - Emit a table mapping from file name strings to PTH 213 /// token data. 214 Offset EmitFileTable() { return PM.Emit(Out); } 215 216 PTHEntry LexTokens(Lexer& L); 217 Offset EmitCachedSpellings(); 218 219public: 220 PTHWriter(llvm::raw_fd_ostream& out, Preprocessor& pp) 221 : Out(out), PP(pp), idcount(0), CurStrOffset(0) {} 222 223 PTHMap &getPM() { return PM; } 224 void GeneratePTH(const std::string *MainFile = 0); 225}; 226} // end anonymous namespace 227 228uint32_t PTHWriter::ResolveID(const IdentifierInfo* II) { 229 // Null IdentifierInfo's map to the persistent ID 0. 230 if (!II) 231 return 0; 232 233 IDMap::iterator I = IM.find(II); 234 if (I != IM.end()) 235 return I->second; // We've already added 1. 236 237 IM[II] = ++idcount; // Pre-increment since '0' is reserved for NULL. 238 return idcount; 239} 240 241void PTHWriter::EmitToken(const Token& T) { 242 // Emit the token kind, flags, and length. 243 Emit32(((uint32_t) T.getKind()) | ((((uint32_t) T.getFlags())) << 8)| 244 (((uint32_t) T.getLength()) << 16)); 245 246 if (!T.isLiteral()) { 247 Emit32(ResolveID(T.getIdentifierInfo())); 248 } else { 249 // We cache *un-cleaned* spellings. This gives us 100% fidelity with the 250 // source code. 251 const char* s = T.getLiteralData(); 252 unsigned len = T.getLength(); 253 254 // Get the string entry. 255 llvm::StringMapEntry<OffsetOpt> *E = &CachedStrs.GetOrCreateValue(s, s+len); 256 257 // If this is a new string entry, bump the PTH offset. 258 if (!E->getValue().hasOffset()) { 259 E->getValue().setOffset(CurStrOffset); 260 StrEntries.push_back(E); 261 CurStrOffset += len + 1; 262 } 263 264 // Emit the relative offset into the PTH file for the spelling string. 265 Emit32(E->getValue().getOffset()); 266 } 267 268 // Emit the offset into the original source file of this token so that we 269 // can reconstruct its SourceLocation. 270 Emit32(PP.getSourceManager().getFileOffset(T.getLocation())); 271} 272 273PTHEntry PTHWriter::LexTokens(Lexer& L) { 274 // Pad 0's so that we emit tokens to a 4-byte alignment. 275 // This speed up reading them back in. 276 Pad(Out, 4); 277 Offset off = (Offset) Out.tell(); 278 279 // Keep track of matching '#if' ... '#endif'. 280 typedef std::vector<std::pair<Offset, unsigned> > PPCondTable; 281 PPCondTable PPCond; 282 std::vector<unsigned> PPStartCond; 283 bool ParsingPreprocessorDirective = false; 284 Token Tok; 285 286 do { 287 L.LexFromRawLexer(Tok); 288 NextToken: 289 290 if ((Tok.isAtStartOfLine() || Tok.is(tok::eof)) && 291 ParsingPreprocessorDirective) { 292 // Insert an eom token into the token cache. It has the same 293 // position as the next token that is not on the same line as the 294 // preprocessor directive. Observe that we continue processing 295 // 'Tok' when we exit this branch. 296 Token Tmp = Tok; 297 Tmp.setKind(tok::eom); 298 Tmp.clearFlag(Token::StartOfLine); 299 Tmp.setIdentifierInfo(0); 300 EmitToken(Tmp); 301 ParsingPreprocessorDirective = false; 302 } 303 304 if (Tok.is(tok::identifier)) { 305 Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok)); 306 EmitToken(Tok); 307 continue; 308 } 309 310 if (Tok.is(tok::hash) && Tok.isAtStartOfLine()) { 311 // Special processing for #include. Store the '#' token and lex 312 // the next token. 313 assert(!ParsingPreprocessorDirective); 314 Offset HashOff = (Offset) Out.tell(); 315 EmitToken(Tok); 316 317 // Get the next token. 318 L.LexFromRawLexer(Tok); 319 320 // If we see the start of line, then we had a null directive "#". 321 if (Tok.isAtStartOfLine()) 322 goto NextToken; 323 324 // Did we see 'include'/'import'/'include_next'? 325 if (Tok.isNot(tok::identifier)) { 326 EmitToken(Tok); 327 continue; 328 } 329 330 IdentifierInfo* II = PP.LookUpIdentifierInfo(Tok); 331 Tok.setIdentifierInfo(II); 332 tok::PPKeywordKind K = II->getPPKeywordID(); 333 334 ParsingPreprocessorDirective = true; 335 336 switch (K) { 337 case tok::pp_not_keyword: 338 // Invalid directives "#foo" can occur in #if 0 blocks etc, just pass 339 // them through. 340 default: 341 break; 342 343 case tok::pp_include: 344 case tok::pp_import: 345 case tok::pp_include_next: { 346 // Save the 'include' token. 347 EmitToken(Tok); 348 // Lex the next token as an include string. 349 L.setParsingPreprocessorDirective(true); 350 L.LexIncludeFilename(Tok); 351 L.setParsingPreprocessorDirective(false); 352 assert(!Tok.isAtStartOfLine()); 353 if (Tok.is(tok::identifier)) 354 Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok)); 355 356 break; 357 } 358 case tok::pp_if: 359 case tok::pp_ifdef: 360 case tok::pp_ifndef: { 361 // Add an entry for '#if' and friends. We initially set the target 362 // index to 0. This will get backpatched when we hit #endif. 363 PPStartCond.push_back(PPCond.size()); 364 PPCond.push_back(std::make_pair(HashOff, 0U)); 365 break; 366 } 367 case tok::pp_endif: { 368 // Add an entry for '#endif'. We set the target table index to itself. 369 // This will later be set to zero when emitting to the PTH file. We 370 // use 0 for uninitialized indices because that is easier to debug. 371 unsigned index = PPCond.size(); 372 // Backpatch the opening '#if' entry. 373 assert(!PPStartCond.empty()); 374 assert(PPCond.size() > PPStartCond.back()); 375 assert(PPCond[PPStartCond.back()].second == 0); 376 PPCond[PPStartCond.back()].second = index; 377 PPStartCond.pop_back(); 378 // Add the new entry to PPCond. 379 PPCond.push_back(std::make_pair(HashOff, index)); 380 EmitToken(Tok); 381 382 // Some files have gibberish on the same line as '#endif'. 383 // Discard these tokens. 384 do 385 L.LexFromRawLexer(Tok); 386 while (Tok.isNot(tok::eof) && !Tok.isAtStartOfLine()); 387 // We have the next token in hand. 388 // Don't immediately lex the next one. 389 goto NextToken; 390 } 391 case tok::pp_elif: 392 case tok::pp_else: { 393 // Add an entry for #elif or #else. 394 // This serves as both a closing and opening of a conditional block. 395 // This means that its entry will get backpatched later. 396 unsigned index = PPCond.size(); 397 // Backpatch the previous '#if' entry. 398 assert(!PPStartCond.empty()); 399 assert(PPCond.size() > PPStartCond.back()); 400 assert(PPCond[PPStartCond.back()].second == 0); 401 PPCond[PPStartCond.back()].second = index; 402 PPStartCond.pop_back(); 403 // Now add '#elif' as a new block opening. 404 PPCond.push_back(std::make_pair(HashOff, 0U)); 405 PPStartCond.push_back(index); 406 break; 407 } 408 } 409 } 410 411 EmitToken(Tok); 412 } 413 while (Tok.isNot(tok::eof)); 414 415 assert(PPStartCond.empty() && "Error: imblanced preprocessor conditionals."); 416 417 // Next write out PPCond. 418 Offset PPCondOff = (Offset) Out.tell(); 419 420 // Write out the size of PPCond so that clients can identifer empty tables. 421 Emit32(PPCond.size()); 422 423 for (unsigned i = 0, e = PPCond.size(); i!=e; ++i) { 424 Emit32(PPCond[i].first - off); 425 uint32_t x = PPCond[i].second; 426 assert(x != 0 && "PPCond entry not backpatched."); 427 // Emit zero for #endifs. This allows us to do checking when 428 // we read the PTH file back in. 429 Emit32(x == i ? 0 : x); 430 } 431 432 return PTHEntry(off, PPCondOff); 433} 434 435Offset PTHWriter::EmitCachedSpellings() { 436 // Write each cached strings to the PTH file. 437 Offset SpellingsOff = Out.tell(); 438 439 for (std::vector<llvm::StringMapEntry<OffsetOpt>*>::iterator 440 I = StrEntries.begin(), E = StrEntries.end(); I!=E; ++I) 441 EmitBuf((*I)->getKeyData(), (*I)->getKeyLength()+1 /*nul included*/); 442 443 return SpellingsOff; 444} 445 446void PTHWriter::GeneratePTH(const std::string *MainFile) { 447 // Generate the prologue. 448 Out << "cfe-pth"; 449 Emit32(PTHManager::Version); 450 451 // Leave 4 words for the prologue. 452 Offset PrologueOffset = Out.tell(); 453 for (unsigned i = 0; i < 4; ++i) 454 Emit32(0); 455 456 // Write the name of the MainFile. 457 if (MainFile && !MainFile->empty()) { 458 Emit16(MainFile->length()); 459 EmitBuf(MainFile->data(), MainFile->length()); 460 } else { 461 // String with 0 bytes. 462 Emit16(0); 463 } 464 Emit8(0); 465 466 // Iterate over all the files in SourceManager. Create a lexer 467 // for each file and cache the tokens. 468 SourceManager &SM = PP.getSourceManager(); 469 const LangOptions &LOpts = PP.getLangOptions(); 470 471 for (SourceManager::fileinfo_iterator I = SM.fileinfo_begin(), 472 E = SM.fileinfo_end(); I != E; ++I) { 473 const SrcMgr::ContentCache &C = *I->second; 474 const FileEntry *FE = C.Entry; 475 476 // FIXME: Handle files with non-absolute paths. 477 llvm::sys::Path P(FE->getName()); 478 if (!P.isAbsolute()) 479 continue; 480 481 const llvm::MemoryBuffer *B = C.getBuffer(); 482 if (!B) continue; 483 484 FileID FID = SM.createFileID(FE, SourceLocation(), SrcMgr::C_User); 485 Lexer L(FID, SM, LOpts); 486 PM.insert(FE, LexTokens(L)); 487 } 488 489 // Write out the identifier table. 490 const std::pair<Offset,Offset> &IdTableOff = EmitIdentifierTable(); 491 492 // Write out the cached strings table. 493 Offset SpellingOff = EmitCachedSpellings(); 494 495 // Write out the file table. 496 Offset FileTableOff = EmitFileTable(); 497 498 // Finally, write the prologue. 499 Out.seek(PrologueOffset); 500 Emit32(IdTableOff.first); 501 Emit32(IdTableOff.second); 502 Emit32(FileTableOff); 503 Emit32(SpellingOff); 504} 505 506namespace { 507/// StatListener - A simple "interpose" object used to monitor stat calls 508/// invoked by FileManager while processing the original sources used 509/// as input to PTH generation. StatListener populates the PTHWriter's 510/// file map with stat information for directories as well as negative stats. 511/// Stat information for files are populated elsewhere. 512class StatListener : public StatSysCallCache { 513 PTHMap &PM; 514public: 515 StatListener(PTHMap &pm) : PM(pm) {} 516 ~StatListener() {} 517 518 int stat(const char *path, struct stat *buf) { 519 int result = ::stat(path, buf); 520 521 if (result != 0) // Failed 'stat'. 522 PM.insert(path, PTHEntry()); 523 else if (S_ISDIR(buf->st_mode)) { 524 // Only cache directories with absolute paths. 525 if (!llvm::sys::Path(path).isAbsolute()) 526 return result; 527 528 PM.insert(PTHEntryKeyVariant(buf, path), PTHEntry()); 529 } 530 531 return result; 532 } 533}; 534} // end anonymous namespace 535 536 537void clang::CacheTokens(Preprocessor &PP, llvm::raw_fd_ostream* OS) { 538 // Get the name of the main file. 539 const SourceManager &SrcMgr = PP.getSourceManager(); 540 const FileEntry *MainFile = SrcMgr.getFileEntryForID(SrcMgr.getMainFileID()); 541 llvm::sys::Path MainFilePath(MainFile->getName()); 542 std::string MainFileName; 543 544 if (!MainFilePath.isAbsolute()) { 545 llvm::sys::Path P = llvm::sys::Path::GetCurrentDirectory(); 546 P.appendComponent(MainFilePath.toString()); 547 MainFileName = P.toString(); 548 } else { 549 MainFileName = MainFilePath.toString(); 550 } 551 552 // Create the PTHWriter. 553 PTHWriter PW(*OS, PP); 554 555 // Install the 'stat' system call listener in the FileManager. 556 PP.getFileManager().setStatCache(new StatListener(PW.getPM())); 557 558 // Lex through the entire file. This will populate SourceManager with 559 // all of the header information. 560 Token Tok; 561 PP.EnterMainSourceFile(); 562 do { PP.Lex(Tok); } while (Tok.isNot(tok::eof)); 563 564 // Generate the PTH file. 565 PP.getFileManager().setStatCache(0); 566 PW.GeneratePTH(&MainFileName); 567} 568 569//===----------------------------------------------------------------------===// 570 571class PTHIdKey { 572public: 573 const IdentifierInfo* II; 574 uint32_t FileOffset; 575}; 576 577namespace { 578class VISIBILITY_HIDDEN PTHIdentifierTableTrait { 579public: 580 typedef PTHIdKey* key_type; 581 typedef key_type key_type_ref; 582 583 typedef uint32_t data_type; 584 typedef data_type data_type_ref; 585 586 static unsigned ComputeHash(PTHIdKey* key) { 587 return BernsteinHash(key->II->getName()); 588 } 589 590 static std::pair<unsigned,unsigned> 591 EmitKeyDataLength(llvm::raw_ostream& Out, const PTHIdKey* key, uint32_t) { 592 unsigned n = strlen(key->II->getName()) + 1; 593 ::Emit16(Out, n); 594 return std::make_pair(n, sizeof(uint32_t)); 595 } 596 597 static void EmitKey(llvm::raw_ostream& Out, PTHIdKey* key, unsigned n) { 598 // Record the location of the key data. This is used when generating 599 // the mapping from persistent IDs to strings. 600 key->FileOffset = Out.tell(); 601 Out.write(key->II->getName(), n); 602 } 603 604 static void EmitData(llvm::raw_ostream& Out, PTHIdKey*, uint32_t pID, 605 unsigned) { 606 ::Emit32(Out, pID); 607 } 608}; 609} // end anonymous namespace 610 611/// EmitIdentifierTable - Emits two tables to the PTH file. The first is 612/// a hashtable mapping from identifier strings to persistent IDs. The second 613/// is a straight table mapping from persistent IDs to string data (the 614/// keys of the first table). 615/// 616std::pair<Offset,Offset> PTHWriter::EmitIdentifierTable() { 617 // Build two maps: 618 // (1) an inverse map from persistent IDs -> (IdentifierInfo*,Offset) 619 // (2) a map from (IdentifierInfo*, Offset)* -> persistent IDs 620 621 // Note that we use 'calloc', so all the bytes are 0. 622 PTHIdKey *IIDMap = (PTHIdKey*)calloc(idcount, sizeof(PTHIdKey)); 623 624 // Create the hashtable. 625 OnDiskChainedHashTableGenerator<PTHIdentifierTableTrait> IIOffMap; 626 627 // Generate mapping from persistent IDs -> IdentifierInfo*. 628 for (IDMap::iterator I = IM.begin(), E = IM.end(); I != E; ++I) { 629 // Decrement by 1 because we are using a vector for the lookup and 630 // 0 is reserved for NULL. 631 assert(I->second > 0); 632 assert(I->second-1 < idcount); 633 unsigned idx = I->second-1; 634 635 // Store the mapping from persistent ID to IdentifierInfo* 636 IIDMap[idx].II = I->first; 637 638 // Store the reverse mapping in a hashtable. 639 IIOffMap.insert(&IIDMap[idx], I->second); 640 } 641 642 // Write out the inverse map first. This causes the PCIDKey entries to 643 // record PTH file offsets for the string data. This is used to write 644 // the second table. 645 Offset StringTableOffset = IIOffMap.Emit(Out); 646 647 // Now emit the table mapping from persistent IDs to PTH file offsets. 648 Offset IDOff = Out.tell(); 649 Emit32(idcount); // Emit the number of identifiers. 650 for (unsigned i = 0 ; i < idcount; ++i) 651 Emit32(IIDMap[i].FileOffset); 652 653 // Finally, release the inverse map. 654 free(IIDMap); 655 656 return std::make_pair(IDOff, StringTableOffset); 657} 658