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