1 //===--- TokenLexer.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 TokenLexer interface. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "clang/Lex/TokenLexer.h" 15 #include "clang/Basic/SourceManager.h" 16 #include "clang/Lex/LexDiagnostic.h" 17 #include "clang/Lex/MacroArgs.h" 18 #include "clang/Lex/MacroInfo.h" 19 #include "clang/Lex/Preprocessor.h" 20 #include "llvm/ADT/SmallString.h" 21 using namespace clang; 22 23 24 /// Create a TokenLexer for the specified macro with the specified actual 25 /// arguments. Note that this ctor takes ownership of the ActualArgs pointer. 26 void TokenLexer::Init(Token &Tok, SourceLocation ELEnd, MacroInfo *MI, 27 MacroArgs *Actuals) { 28 // If the client is reusing a TokenLexer, make sure to free any memory 29 // associated with it. 30 destroy(); 31 32 Macro = MI; 33 ActualArgs = Actuals; 34 CurToken = 0; 35 36 ExpandLocStart = Tok.getLocation(); 37 ExpandLocEnd = ELEnd; 38 AtStartOfLine = Tok.isAtStartOfLine(); 39 HasLeadingSpace = Tok.hasLeadingSpace(); 40 NextTokGetsSpace = false; 41 Tokens = &*Macro->tokens_begin(); 42 OwnsTokens = false; 43 DisableMacroExpansion = false; 44 NumTokens = Macro->tokens_end()-Macro->tokens_begin(); 45 MacroExpansionStart = SourceLocation(); 46 47 SourceManager &SM = PP.getSourceManager(); 48 MacroStartSLocOffset = SM.getNextLocalOffset(); 49 50 if (NumTokens > 0) { 51 assert(Tokens[0].getLocation().isValid()); 52 assert((Tokens[0].getLocation().isFileID() || Tokens[0].is(tok::comment)) && 53 "Macro defined in macro?"); 54 assert(ExpandLocStart.isValid()); 55 56 // Reserve a source location entry chunk for the length of the macro 57 // definition. Tokens that get lexed directly from the definition will 58 // have their locations pointing inside this chunk. This is to avoid 59 // creating separate source location entries for each token. 60 MacroDefStart = SM.getExpansionLoc(Tokens[0].getLocation()); 61 MacroDefLength = Macro->getDefinitionLength(SM); 62 MacroExpansionStart = SM.createExpansionLoc(MacroDefStart, 63 ExpandLocStart, 64 ExpandLocEnd, 65 MacroDefLength); 66 } 67 68 // If this is a function-like macro, expand the arguments and change 69 // Tokens to point to the expanded tokens. 70 if (Macro->isFunctionLike() && Macro->getNumArgs()) 71 ExpandFunctionArguments(); 72 73 // Mark the macro as currently disabled, so that it is not recursively 74 // expanded. The macro must be disabled only after argument pre-expansion of 75 // function-like macro arguments occurs. 76 Macro->DisableMacro(); 77 } 78 79 80 81 /// Create a TokenLexer for the specified token stream. This does not 82 /// take ownership of the specified token vector. 83 void TokenLexer::Init(const Token *TokArray, unsigned NumToks, 84 bool disableMacroExpansion, bool ownsTokens) { 85 // If the client is reusing a TokenLexer, make sure to free any memory 86 // associated with it. 87 destroy(); 88 89 Macro = nullptr; 90 ActualArgs = nullptr; 91 Tokens = TokArray; 92 OwnsTokens = ownsTokens; 93 DisableMacroExpansion = disableMacroExpansion; 94 NumTokens = NumToks; 95 CurToken = 0; 96 ExpandLocStart = ExpandLocEnd = SourceLocation(); 97 AtStartOfLine = false; 98 HasLeadingSpace = false; 99 NextTokGetsSpace = false; 100 MacroExpansionStart = SourceLocation(); 101 102 // Set HasLeadingSpace/AtStartOfLine so that the first token will be 103 // returned unmodified. 104 if (NumToks != 0) { 105 AtStartOfLine = TokArray[0].isAtStartOfLine(); 106 HasLeadingSpace = TokArray[0].hasLeadingSpace(); 107 } 108 } 109 110 111 void TokenLexer::destroy() { 112 // If this was a function-like macro that actually uses its arguments, delete 113 // the expanded tokens. 114 if (OwnsTokens) { 115 delete [] Tokens; 116 Tokens = nullptr; 117 OwnsTokens = false; 118 } 119 120 // TokenLexer owns its formal arguments. 121 if (ActualArgs) ActualArgs->destroy(PP); 122 } 123 124 bool TokenLexer::MaybeRemoveCommaBeforeVaArgs( 125 SmallVectorImpl<Token> &ResultToks, bool HasPasteOperator, MacroInfo *Macro, 126 unsigned MacroArgNo, Preprocessor &PP) { 127 // Is the macro argument __VA_ARGS__? 128 if (!Macro->isVariadic() || MacroArgNo != Macro->getNumArgs()-1) 129 return false; 130 131 // In Microsoft-compatibility mode, a comma is removed in the expansion 132 // of " ... , __VA_ARGS__ " if __VA_ARGS__ is empty. This extension is 133 // not supported by gcc. 134 if (!HasPasteOperator && !PP.getLangOpts().MSVCCompat) 135 return false; 136 137 // GCC removes the comma in the expansion of " ... , ## __VA_ARGS__ " if 138 // __VA_ARGS__ is empty, but not in strict C99 mode where there are no 139 // named arguments, where it remains. In all other modes, including C99 140 // with GNU extensions, it is removed regardless of named arguments. 141 // Microsoft also appears to support this extension, unofficially. 142 if (PP.getLangOpts().C99 && !PP.getLangOpts().GNUMode 143 && Macro->getNumArgs() < 2) 144 return false; 145 146 // Is a comma available to be removed? 147 if (ResultToks.empty() || !ResultToks.back().is(tok::comma)) 148 return false; 149 150 // Issue an extension diagnostic for the paste operator. 151 if (HasPasteOperator) 152 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 153 154 // Remove the comma. 155 ResultToks.pop_back(); 156 157 // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"), 158 // then removal of the comma should produce a placemarker token (in C99 159 // terms) which we model by popping off the previous ##, giving us a plain 160 // "X" when __VA_ARGS__ is empty. 161 if (!ResultToks.empty() && ResultToks.back().is(tok::hashhash)) 162 ResultToks.pop_back(); 163 164 // Never add a space, even if the comma, ##, or arg had a space. 165 NextTokGetsSpace = false; 166 return true; 167 } 168 169 /// Expand the arguments of a function-like macro so that we can quickly 170 /// return preexpanded tokens from Tokens. 171 void TokenLexer::ExpandFunctionArguments() { 172 173 SmallVector<Token, 128> ResultToks; 174 175 // Loop through 'Tokens', expanding them into ResultToks. Keep 176 // track of whether we change anything. If not, no need to keep them. If so, 177 // we install the newly expanded sequence as the new 'Tokens' list. 178 bool MadeChange = false; 179 180 for (unsigned i = 0, e = NumTokens; i != e; ++i) { 181 // If we found the stringify operator, get the argument stringified. The 182 // preprocessor already verified that the following token is a macro name 183 // when the #define was parsed. 184 const Token &CurTok = Tokens[i]; 185 if (i != 0 && !Tokens[i-1].is(tok::hashhash) && CurTok.hasLeadingSpace()) 186 NextTokGetsSpace = true; 187 188 if (CurTok.is(tok::hash) || CurTok.is(tok::hashat)) { 189 int ArgNo = Macro->getArgumentNum(Tokens[i+1].getIdentifierInfo()); 190 assert(ArgNo != -1 && "Token following # is not an argument?"); 191 192 SourceLocation ExpansionLocStart = 193 getExpansionLocForMacroDefLoc(CurTok.getLocation()); 194 SourceLocation ExpansionLocEnd = 195 getExpansionLocForMacroDefLoc(Tokens[i+1].getLocation()); 196 197 Token Res; 198 if (CurTok.is(tok::hash)) // Stringify 199 Res = ActualArgs->getStringifiedArgument(ArgNo, PP, 200 ExpansionLocStart, 201 ExpansionLocEnd); 202 else { 203 // 'charify': don't bother caching these. 204 Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo), 205 PP, true, 206 ExpansionLocStart, 207 ExpansionLocEnd); 208 } 209 210 // The stringified/charified string leading space flag gets set to match 211 // the #/#@ operator. 212 if (NextTokGetsSpace) 213 Res.setFlag(Token::LeadingSpace); 214 215 ResultToks.push_back(Res); 216 MadeChange = true; 217 ++i; // Skip arg name. 218 NextTokGetsSpace = false; 219 continue; 220 } 221 222 // Find out if there is a paste (##) operator before or after the token. 223 bool NonEmptyPasteBefore = 224 !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 225 bool PasteBefore = i != 0 && Tokens[i-1].is(tok::hashhash); 226 bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash); 227 assert(!NonEmptyPasteBefore || PasteBefore); 228 229 // Otherwise, if this is not an argument token, just add the token to the 230 // output buffer. 231 IdentifierInfo *II = CurTok.getIdentifierInfo(); 232 int ArgNo = II ? Macro->getArgumentNum(II) : -1; 233 if (ArgNo == -1) { 234 // This isn't an argument, just add it. 235 ResultToks.push_back(CurTok); 236 237 if (NextTokGetsSpace) { 238 ResultToks.back().setFlag(Token::LeadingSpace); 239 NextTokGetsSpace = false; 240 } else if (PasteBefore && !NonEmptyPasteBefore) 241 ResultToks.back().clearFlag(Token::LeadingSpace); 242 243 continue; 244 } 245 246 // An argument is expanded somehow, the result is different than the 247 // input. 248 MadeChange = true; 249 250 // Otherwise, this is a use of the argument. 251 252 // In Microsoft mode, remove the comma before __VA_ARGS__ to ensure there 253 // are no trailing commas if __VA_ARGS__ is empty. 254 if (!PasteBefore && ActualArgs->isVarargsElidedUse() && 255 MaybeRemoveCommaBeforeVaArgs(ResultToks, 256 /*HasPasteOperator=*/false, 257 Macro, ArgNo, PP)) 258 continue; 259 260 // If it is not the LHS/RHS of a ## operator, we must pre-expand the 261 // argument and substitute the expanded tokens into the result. This is 262 // C99 6.10.3.1p1. 263 if (!PasteBefore && !PasteAfter) { 264 const Token *ResultArgToks; 265 266 // Only preexpand the argument if it could possibly need it. This 267 // avoids some work in common cases. 268 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 269 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 270 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, Macro, PP)[0]; 271 else 272 ResultArgToks = ArgTok; // Use non-preexpanded tokens. 273 274 // If the arg token expanded into anything, append it. 275 if (ResultArgToks->isNot(tok::eof)) { 276 unsigned FirstResult = ResultToks.size(); 277 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 278 ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 279 280 // In Microsoft-compatibility mode, we follow MSVC's preprocessing 281 // behavior by not considering single commas from nested macro 282 // expansions as argument separators. Set a flag on the token so we can 283 // test for this later when the macro expansion is processed. 284 if (PP.getLangOpts().MSVCCompat && NumToks == 1 && 285 ResultToks.back().is(tok::comma)) 286 ResultToks.back().setFlag(Token::IgnoredComma); 287 288 // If the '##' came from expanding an argument, turn it into 'unknown' 289 // to avoid pasting. 290 for (unsigned i = FirstResult, e = ResultToks.size(); i != e; ++i) { 291 Token &Tok = ResultToks[i]; 292 if (Tok.is(tok::hashhash)) 293 Tok.setKind(tok::unknown); 294 } 295 296 if(ExpandLocStart.isValid()) { 297 updateLocForMacroArgTokens(CurTok.getLocation(), 298 ResultToks.begin()+FirstResult, 299 ResultToks.end()); 300 } 301 302 // If any tokens were substituted from the argument, the whitespace 303 // before the first token should match the whitespace of the arg 304 // identifier. 305 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 306 NextTokGetsSpace); 307 NextTokGetsSpace = false; 308 } 309 continue; 310 } 311 312 // Okay, we have a token that is either the LHS or RHS of a paste (##) 313 // argument. It gets substituted as its non-pre-expanded tokens. 314 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 315 unsigned NumToks = MacroArgs::getArgLength(ArgToks); 316 if (NumToks) { // Not an empty argument? 317 // If this is the GNU ", ## __VA_ARGS__" extension, and we just learned 318 // that __VA_ARGS__ expands to multiple tokens, avoid a pasting error when 319 // the expander trys to paste ',' with the first token of the __VA_ARGS__ 320 // expansion. 321 if (NonEmptyPasteBefore && ResultToks.size() >= 2 && 322 ResultToks[ResultToks.size()-2].is(tok::comma) && 323 (unsigned)ArgNo == Macro->getNumArgs()-1 && 324 Macro->isVariadic()) { 325 // Remove the paste operator, report use of the extension. 326 PP.Diag(ResultToks.pop_back_val().getLocation(), diag::ext_paste_comma); 327 } 328 329 ResultToks.append(ArgToks, ArgToks+NumToks); 330 331 // If the '##' came from expanding an argument, turn it into 'unknown' 332 // to avoid pasting. 333 for (unsigned i = ResultToks.size() - NumToks, e = ResultToks.size(); 334 i != e; ++i) { 335 Token &Tok = ResultToks[i]; 336 if (Tok.is(tok::hashhash)) 337 Tok.setKind(tok::unknown); 338 } 339 340 if (ExpandLocStart.isValid()) { 341 updateLocForMacroArgTokens(CurTok.getLocation(), 342 ResultToks.end()-NumToks, ResultToks.end()); 343 } 344 345 // If this token (the macro argument) was supposed to get leading 346 // whitespace, transfer this information onto the first token of the 347 // expansion. 348 // 349 // Do not do this if the paste operator occurs before the macro argument, 350 // as in "A ## MACROARG". In valid code, the first token will get 351 // smooshed onto the preceding one anyway (forming AMACROARG). In 352 // assembler-with-cpp mode, invalid pastes are allowed through: in this 353 // case, we do not want the extra whitespace to be added. For example, 354 // we want ". ## foo" -> ".foo" not ". foo". 355 if (NextTokGetsSpace) 356 ResultToks[ResultToks.size()-NumToks].setFlag(Token::LeadingSpace); 357 358 NextTokGetsSpace = false; 359 continue; 360 } 361 362 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 363 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 364 // implement this by eating ## operators when a LHS or RHS expands to 365 // empty. 366 if (PasteAfter) { 367 // Discard the argument token and skip (don't copy to the expansion 368 // buffer) the paste operator after it. 369 ++i; 370 continue; 371 } 372 373 // If this is on the RHS of a paste operator, we've already copied the 374 // paste operator to the ResultToks list, unless the LHS was empty too. 375 // Remove it. 376 assert(PasteBefore); 377 if (NonEmptyPasteBefore) { 378 assert(ResultToks.back().is(tok::hashhash)); 379 ResultToks.pop_back(); 380 } 381 382 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 383 // and if the macro had at least one real argument, and if the token before 384 // the ## was a comma, remove the comma. This is a GCC extension which is 385 // disabled when using -std=c99. 386 if (ActualArgs->isVarargsElidedUse()) 387 MaybeRemoveCommaBeforeVaArgs(ResultToks, 388 /*HasPasteOperator=*/true, 389 Macro, ArgNo, PP); 390 391 continue; 392 } 393 394 // If anything changed, install this as the new Tokens list. 395 if (MadeChange) { 396 assert(!OwnsTokens && "This would leak if we already own the token list"); 397 // This is deleted in the dtor. 398 NumTokens = ResultToks.size(); 399 // The tokens will be added to Preprocessor's cache and will be removed 400 // when this TokenLexer finishes lexing them. 401 Tokens = PP.cacheMacroExpandedTokens(this, ResultToks); 402 403 // The preprocessor cache of macro expanded tokens owns these tokens,not us. 404 OwnsTokens = false; 405 } 406 } 407 408 /// Lex - Lex and return a token from this macro stream. 409 /// 410 bool TokenLexer::Lex(Token &Tok) { 411 // Lexing off the end of the macro, pop this macro off the expansion stack. 412 if (isAtEnd()) { 413 // If this is a macro (not a token stream), mark the macro enabled now 414 // that it is no longer being expanded. 415 if (Macro) Macro->EnableMacro(); 416 417 Tok.startToken(); 418 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 419 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace || NextTokGetsSpace); 420 if (CurToken == 0) 421 Tok.setFlag(Token::LeadingEmptyMacro); 422 return PP.HandleEndOfTokenLexer(Tok); 423 } 424 425 SourceManager &SM = PP.getSourceManager(); 426 427 // If this is the first token of the expanded result, we inherit spacing 428 // properties later. 429 bool isFirstToken = CurToken == 0; 430 431 // Get the next token to return. 432 Tok = Tokens[CurToken++]; 433 434 bool TokenIsFromPaste = false; 435 436 // If this token is followed by a token paste (##) operator, paste the tokens! 437 // Note that ## is a normal token when not expanding a macro. 438 if (!isAtEnd() && Tokens[CurToken].is(tok::hashhash) && Macro) { 439 // When handling the microsoft /##/ extension, the final token is 440 // returned by PasteTokens, not the pasted token. 441 if (PasteTokens(Tok)) 442 return true; 443 444 TokenIsFromPaste = true; 445 } 446 447 // The token's current location indicate where the token was lexed from. We 448 // need this information to compute the spelling of the token, but any 449 // diagnostics for the expanded token should appear as if they came from 450 // ExpansionLoc. Pull this information together into a new SourceLocation 451 // that captures all of this. 452 if (ExpandLocStart.isValid() && // Don't do this for token streams. 453 // Check that the token's location was not already set properly. 454 SM.isBeforeInSLocAddrSpace(Tok.getLocation(), MacroStartSLocOffset)) { 455 SourceLocation instLoc; 456 if (Tok.is(tok::comment)) { 457 instLoc = SM.createExpansionLoc(Tok.getLocation(), 458 ExpandLocStart, 459 ExpandLocEnd, 460 Tok.getLength()); 461 } else { 462 instLoc = getExpansionLocForMacroDefLoc(Tok.getLocation()); 463 } 464 465 Tok.setLocation(instLoc); 466 } 467 468 // If this is the first token, set the lexical properties of the token to 469 // match the lexical properties of the macro identifier. 470 if (isFirstToken) { 471 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 472 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 473 } else { 474 // If this is not the first token, we may still need to pass through 475 // leading whitespace if we've expanded a macro. 476 if (AtStartOfLine) Tok.setFlag(Token::StartOfLine); 477 if (HasLeadingSpace) Tok.setFlag(Token::LeadingSpace); 478 } 479 AtStartOfLine = false; 480 HasLeadingSpace = false; 481 482 // Handle recursive expansion! 483 if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) { 484 // Change the kind of this identifier to the appropriate token kind, e.g. 485 // turning "for" into a keyword. 486 IdentifierInfo *II = Tok.getIdentifierInfo(); 487 Tok.setKind(II->getTokenID()); 488 489 // If this identifier was poisoned and from a paste, emit an error. This 490 // won't be handled by Preprocessor::HandleIdentifier because this is coming 491 // from a macro expansion. 492 if (II->isPoisoned() && TokenIsFromPaste) { 493 PP.HandlePoisonedIdentifier(Tok); 494 } 495 496 if (!DisableMacroExpansion && II->isHandleIdentifierCase()) 497 return PP.HandleIdentifier(Tok); 498 } 499 500 // Otherwise, return a normal token. 501 return true; 502 } 503 504 /// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ## 505 /// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 506 /// are more ## after it, chomp them iteratively. Return the result as Tok. 507 /// If this returns true, the caller should immediately return the token. 508 bool TokenLexer::PasteTokens(Token &Tok) { 509 SmallString<128> Buffer; 510 const char *ResultTokStrPtr = nullptr; 511 SourceLocation StartLoc = Tok.getLocation(); 512 SourceLocation PasteOpLoc; 513 do { 514 // Consume the ## operator. 515 PasteOpLoc = Tokens[CurToken].getLocation(); 516 ++CurToken; 517 assert(!isAtEnd() && "No token on the RHS of a paste operator!"); 518 519 // Get the RHS token. 520 const Token &RHS = Tokens[CurToken]; 521 522 // Allocate space for the result token. This is guaranteed to be enough for 523 // the two tokens. 524 Buffer.resize(Tok.getLength() + RHS.getLength()); 525 526 // Get the spelling of the LHS token in Buffer. 527 const char *BufPtr = &Buffer[0]; 528 bool Invalid = false; 529 unsigned LHSLen = PP.getSpelling(Tok, BufPtr, &Invalid); 530 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 531 memcpy(&Buffer[0], BufPtr, LHSLen); 532 if (Invalid) 533 return true; 534 535 BufPtr = &Buffer[LHSLen]; 536 unsigned RHSLen = PP.getSpelling(RHS, BufPtr, &Invalid); 537 if (Invalid) 538 return true; 539 if (BufPtr != &Buffer[LHSLen]) // Really, we want the chars in Buffer! 540 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 541 542 // Trim excess space. 543 Buffer.resize(LHSLen+RHSLen); 544 545 // Plop the pasted result (including the trailing newline and null) into a 546 // scratch buffer where we can lex it. 547 Token ResultTokTmp; 548 ResultTokTmp.startToken(); 549 550 // Claim that the tmp token is a string_literal so that we can get the 551 // character pointer back from CreateString in getLiteralData(). 552 ResultTokTmp.setKind(tok::string_literal); 553 PP.CreateString(Buffer, ResultTokTmp); 554 SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); 555 ResultTokStrPtr = ResultTokTmp.getLiteralData(); 556 557 // Lex the resultant pasted token into Result. 558 Token Result; 559 560 if (Tok.isAnyIdentifier() && RHS.isAnyIdentifier()) { 561 // Common paste case: identifier+identifier = identifier. Avoid creating 562 // a lexer and other overhead. 563 PP.IncrementPasteCounter(true); 564 Result.startToken(); 565 Result.setKind(tok::raw_identifier); 566 Result.setRawIdentifierData(ResultTokStrPtr); 567 Result.setLocation(ResultTokLoc); 568 Result.setLength(LHSLen+RHSLen); 569 } else { 570 PP.IncrementPasteCounter(false); 571 572 assert(ResultTokLoc.isFileID() && 573 "Should be a raw location into scratch buffer"); 574 SourceManager &SourceMgr = PP.getSourceManager(); 575 FileID LocFileID = SourceMgr.getFileID(ResultTokLoc); 576 577 bool Invalid = false; 578 const char *ScratchBufStart 579 = SourceMgr.getBufferData(LocFileID, &Invalid).data(); 580 if (Invalid) 581 return false; 582 583 // Make a lexer to lex this string from. Lex just this one token. 584 // Make a lexer object so that we lex and expand the paste result. 585 Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID), 586 PP.getLangOpts(), ScratchBufStart, 587 ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen); 588 589 // Lex a token in raw mode. This way it won't look up identifiers 590 // automatically, lexing off the end will return an eof token, and 591 // warnings are disabled. This returns true if the result token is the 592 // entire buffer. 593 bool isInvalid = !TL.LexFromRawLexer(Result); 594 595 // If we got an EOF token, we didn't form even ONE token. For example, we 596 // did "/ ## /" to get "//". 597 isInvalid |= Result.is(tok::eof); 598 599 // If pasting the two tokens didn't form a full new token, this is an 600 // error. This occurs with "x ## +" and other stuff. Return with Tok 601 // unmodified and with RHS as the next token to lex. 602 if (isInvalid) { 603 // Test for the Microsoft extension of /##/ turning into // here on the 604 // error path. 605 if (PP.getLangOpts().MicrosoftExt && Tok.is(tok::slash) && 606 RHS.is(tok::slash)) { 607 HandleMicrosoftCommentPaste(Tok); 608 return true; 609 } 610 611 // Do not emit the error when preprocessing assembler code. 612 if (!PP.getLangOpts().AsmPreprocessor) { 613 // Explicitly convert the token location to have proper expansion 614 // information so that the user knows where it came from. 615 SourceManager &SM = PP.getSourceManager(); 616 SourceLocation Loc = 617 SM.createExpansionLoc(PasteOpLoc, ExpandLocStart, ExpandLocEnd, 2); 618 // If we're in microsoft extensions mode, downgrade this from a hard 619 // error to an extension that defaults to an error. This allows 620 // disabling it. 621 PP.Diag(Loc, PP.getLangOpts().MicrosoftExt ? diag::ext_pp_bad_paste_ms 622 : diag::err_pp_bad_paste) 623 << Buffer.str(); 624 } 625 626 // An error has occurred so exit loop. 627 break; 628 } 629 630 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 631 // operator. 632 if (Result.is(tok::hashhash)) 633 Result.setKind(tok::unknown); 634 } 635 636 // Transfer properties of the LHS over the Result. 637 Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine()); 638 Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace()); 639 640 // Finally, replace LHS with the result, consume the RHS, and iterate. 641 ++CurToken; 642 Tok = Result; 643 } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)); 644 645 SourceLocation EndLoc = Tokens[CurToken - 1].getLocation(); 646 647 // The token's current location indicate where the token was lexed from. We 648 // need this information to compute the spelling of the token, but any 649 // diagnostics for the expanded token should appear as if the token was 650 // expanded from the full ## expression. Pull this information together into 651 // a new SourceLocation that captures all of this. 652 SourceManager &SM = PP.getSourceManager(); 653 if (StartLoc.isFileID()) 654 StartLoc = getExpansionLocForMacroDefLoc(StartLoc); 655 if (EndLoc.isFileID()) 656 EndLoc = getExpansionLocForMacroDefLoc(EndLoc); 657 FileID MacroFID = SM.getFileID(MacroExpansionStart); 658 while (SM.getFileID(StartLoc) != MacroFID) 659 StartLoc = SM.getImmediateExpansionRange(StartLoc).first; 660 while (SM.getFileID(EndLoc) != MacroFID) 661 EndLoc = SM.getImmediateExpansionRange(EndLoc).second; 662 663 Tok.setLocation(SM.createExpansionLoc(Tok.getLocation(), StartLoc, EndLoc, 664 Tok.getLength())); 665 666 // Now that we got the result token, it will be subject to expansion. Since 667 // token pasting re-lexes the result token in raw mode, identifier information 668 // isn't looked up. As such, if the result is an identifier, look up id info. 669 if (Tok.is(tok::raw_identifier)) { 670 // Look up the identifier info for the token. We disabled identifier lookup 671 // by saying we're skipping contents, so we need to do this manually. 672 PP.LookUpIdentifierInfo(Tok); 673 } 674 return false; 675 } 676 677 /// isNextTokenLParen - If the next token lexed will pop this macro off the 678 /// expansion stack, return 2. If the next unexpanded token is a '(', return 679 /// 1, otherwise return 0. 680 unsigned TokenLexer::isNextTokenLParen() const { 681 // Out of tokens? 682 if (isAtEnd()) 683 return 2; 684 return Tokens[CurToken].is(tok::l_paren); 685 } 686 687 /// isParsingPreprocessorDirective - Return true if we are in the middle of a 688 /// preprocessor directive. 689 bool TokenLexer::isParsingPreprocessorDirective() const { 690 return Tokens[NumTokens-1].is(tok::eod) && !isAtEnd(); 691 } 692 693 /// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 694 /// together to form a comment that comments out everything in the current 695 /// macro, other active macros, and anything left on the current physical 696 /// source line of the expanded buffer. Handle this by returning the 697 /// first token on the next line. 698 void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok) { 699 // We 'comment out' the rest of this macro by just ignoring the rest of the 700 // tokens that have not been lexed yet, if any. 701 702 // Since this must be a macro, mark the macro enabled now that it is no longer 703 // being expanded. 704 assert(Macro && "Token streams can't paste comments"); 705 Macro->EnableMacro(); 706 707 PP.HandleMicrosoftCommentPaste(Tok); 708 } 709 710 /// \brief If \arg loc is a file ID and points inside the current macro 711 /// definition, returns the appropriate source location pointing at the 712 /// macro expansion source location entry, otherwise it returns an invalid 713 /// SourceLocation. 714 SourceLocation 715 TokenLexer::getExpansionLocForMacroDefLoc(SourceLocation loc) const { 716 assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() && 717 "Not appropriate for token streams"); 718 assert(loc.isValid() && loc.isFileID()); 719 720 SourceManager &SM = PP.getSourceManager(); 721 assert(SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength) && 722 "Expected loc to come from the macro definition"); 723 724 unsigned relativeOffset = 0; 725 SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength, &relativeOffset); 726 return MacroExpansionStart.getLocWithOffset(relativeOffset); 727 } 728 729 /// \brief Finds the tokens that are consecutive (from the same FileID) 730 /// creates a single SLocEntry, and assigns SourceLocations to each token that 731 /// point to that SLocEntry. e.g for 732 /// assert(foo == bar); 733 /// There will be a single SLocEntry for the "foo == bar" chunk and locations 734 /// for the 'foo', '==', 'bar' tokens will point inside that chunk. 735 /// 736 /// \arg begin_tokens will be updated to a position past all the found 737 /// consecutive tokens. 738 static void updateConsecutiveMacroArgTokens(SourceManager &SM, 739 SourceLocation InstLoc, 740 Token *&begin_tokens, 741 Token * end_tokens) { 742 assert(begin_tokens < end_tokens); 743 744 SourceLocation FirstLoc = begin_tokens->getLocation(); 745 SourceLocation CurLoc = FirstLoc; 746 747 // Compare the source location offset of tokens and group together tokens that 748 // are close, even if their locations point to different FileIDs. e.g. 749 // 750 // |bar | foo | cake | (3 tokens from 3 consecutive FileIDs) 751 // ^ ^ 752 // |bar foo cake| (one SLocEntry chunk for all tokens) 753 // 754 // we can perform this "merge" since the token's spelling location depends 755 // on the relative offset. 756 757 Token *NextTok = begin_tokens + 1; 758 for (; NextTok < end_tokens; ++NextTok) { 759 SourceLocation NextLoc = NextTok->getLocation(); 760 if (CurLoc.isFileID() != NextLoc.isFileID()) 761 break; // Token from different kind of FileID. 762 763 int RelOffs; 764 if (!SM.isInSameSLocAddrSpace(CurLoc, NextLoc, &RelOffs)) 765 break; // Token from different local/loaded location. 766 // Check that token is not before the previous token or more than 50 767 // "characters" away. 768 if (RelOffs < 0 || RelOffs > 50) 769 break; 770 CurLoc = NextLoc; 771 } 772 773 // For the consecutive tokens, find the length of the SLocEntry to contain 774 // all of them. 775 Token &LastConsecutiveTok = *(NextTok-1); 776 int LastRelOffs = 0; 777 SM.isInSameSLocAddrSpace(FirstLoc, LastConsecutiveTok.getLocation(), 778 &LastRelOffs); 779 unsigned FullLength = LastRelOffs + LastConsecutiveTok.getLength(); 780 781 // Create a macro expansion SLocEntry that will "contain" all of the tokens. 782 SourceLocation Expansion = 783 SM.createMacroArgExpansionLoc(FirstLoc, InstLoc,FullLength); 784 785 // Change the location of the tokens from the spelling location to the new 786 // expanded location. 787 for (; begin_tokens < NextTok; ++begin_tokens) { 788 Token &Tok = *begin_tokens; 789 int RelOffs = 0; 790 SM.isInSameSLocAddrSpace(FirstLoc, Tok.getLocation(), &RelOffs); 791 Tok.setLocation(Expansion.getLocWithOffset(RelOffs)); 792 } 793 } 794 795 /// \brief Creates SLocEntries and updates the locations of macro argument 796 /// tokens to their new expanded locations. 797 /// 798 /// \param ArgIdDefLoc the location of the macro argument id inside the macro 799 /// definition. 800 /// \param Tokens the macro argument tokens to update. 801 void TokenLexer::updateLocForMacroArgTokens(SourceLocation ArgIdSpellLoc, 802 Token *begin_tokens, 803 Token *end_tokens) { 804 SourceManager &SM = PP.getSourceManager(); 805 806 SourceLocation InstLoc = 807 getExpansionLocForMacroDefLoc(ArgIdSpellLoc); 808 809 while (begin_tokens < end_tokens) { 810 // If there's only one token just create a SLocEntry for it. 811 if (end_tokens - begin_tokens == 1) { 812 Token &Tok = *begin_tokens; 813 Tok.setLocation(SM.createMacroArgExpansionLoc(Tok.getLocation(), 814 InstLoc, 815 Tok.getLength())); 816 return; 817 } 818 819 updateConsecutiveMacroArgTokens(SM, InstLoc, begin_tokens, end_tokens); 820 } 821 } 822 823 void TokenLexer::PropagateLineStartLeadingSpaceInfo(Token &Result) { 824 AtStartOfLine = Result.isAtStartOfLine(); 825 HasLeadingSpace = Result.hasLeadingSpace(); 826 } 827