TokenLexer.cpp revision 321369
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 22using namespace clang; 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. 26void 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->getNumParams()) 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/// Create a TokenLexer for the specified token stream. This does not 80/// take ownership of the specified token vector. 81void TokenLexer::Init(const Token *TokArray, unsigned NumToks, 82 bool disableMacroExpansion, bool ownsTokens) { 83 // If the client is reusing a TokenLexer, make sure to free any memory 84 // associated with it. 85 destroy(); 86 87 Macro = nullptr; 88 ActualArgs = nullptr; 89 Tokens = TokArray; 90 OwnsTokens = ownsTokens; 91 DisableMacroExpansion = disableMacroExpansion; 92 NumTokens = NumToks; 93 CurToken = 0; 94 ExpandLocStart = ExpandLocEnd = SourceLocation(); 95 AtStartOfLine = false; 96 HasLeadingSpace = false; 97 NextTokGetsSpace = 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 108void TokenLexer::destroy() { 109 // If this was a function-like macro that actually uses its arguments, delete 110 // the expanded tokens. 111 if (OwnsTokens) { 112 delete [] Tokens; 113 Tokens = nullptr; 114 OwnsTokens = false; 115 } 116 117 // TokenLexer owns its formal arguments. 118 if (ActualArgs) ActualArgs->destroy(PP); 119} 120 121bool TokenLexer::MaybeRemoveCommaBeforeVaArgs( 122 SmallVectorImpl<Token> &ResultToks, bool HasPasteOperator, MacroInfo *Macro, 123 unsigned MacroArgNo, Preprocessor &PP) { 124 // Is the macro argument __VA_ARGS__? 125 if (!Macro->isVariadic() || MacroArgNo != Macro->getNumParams()-1) 126 return false; 127 128 // In Microsoft-compatibility mode, a comma is removed in the expansion 129 // of " ... , __VA_ARGS__ " if __VA_ARGS__ is empty. This extension is 130 // not supported by gcc. 131 if (!HasPasteOperator && !PP.getLangOpts().MSVCCompat) 132 return false; 133 134 // GCC removes the comma in the expansion of " ... , ## __VA_ARGS__ " if 135 // __VA_ARGS__ is empty, but not in strict C99 mode where there are no 136 // named arguments, where it remains. In all other modes, including C99 137 // with GNU extensions, it is removed regardless of named arguments. 138 // Microsoft also appears to support this extension, unofficially. 139 if (PP.getLangOpts().C99 && !PP.getLangOpts().GNUMode 140 && Macro->getNumParams() < 2) 141 return false; 142 143 // Is a comma available to be removed? 144 if (ResultToks.empty() || !ResultToks.back().is(tok::comma)) 145 return false; 146 147 // Issue an extension diagnostic for the paste operator. 148 if (HasPasteOperator) 149 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 150 151 // Remove the comma. 152 ResultToks.pop_back(); 153 154 if (!ResultToks.empty()) { 155 // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"), 156 // then removal of the comma should produce a placemarker token (in C99 157 // terms) which we model by popping off the previous ##, giving us a plain 158 // "X" when __VA_ARGS__ is empty. 159 if (ResultToks.back().is(tok::hashhash)) 160 ResultToks.pop_back(); 161 162 // Remember that this comma was elided. 163 ResultToks.back().setFlag(Token::CommaAfterElided); 164 } 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. 173void TokenLexer::ExpandFunctionArguments() { 174 SmallVector<Token, 128> ResultToks; 175 176 // Loop through 'Tokens', expanding them into ResultToks. Keep 177 // track of whether we change anything. If not, no need to keep them. If so, 178 // we install the newly expanded sequence as the new 'Tokens' list. 179 bool MadeChange = false; 180 181 for (unsigned i = 0, e = NumTokens; i != e; ++i) { 182 // If we found the stringify operator, get the argument stringified. The 183 // preprocessor already verified that the following token is a macro name 184 // when the #define was parsed. 185 const Token &CurTok = Tokens[i]; 186 // We don't want a space for the next token after a paste 187 // operator. In valid code, the token will get smooshed onto the 188 // preceding one anyway. In assembler-with-cpp mode, invalid 189 // pastes are allowed through: in this case, we do not want the 190 // extra whitespace to be added. For example, we want ". ## foo" 191 // -> ".foo" not ". foo". 192 if (i != 0 && !Tokens[i-1].is(tok::hashhash) && CurTok.hasLeadingSpace()) 193 NextTokGetsSpace = true; 194 195 if (CurTok.isOneOf(tok::hash, tok::hashat)) { 196 int ArgNo = Macro->getParameterNum(Tokens[i+1].getIdentifierInfo()); 197 assert(ArgNo != -1 && "Token following # is not an argument?"); 198 199 SourceLocation ExpansionLocStart = 200 getExpansionLocForMacroDefLoc(CurTok.getLocation()); 201 SourceLocation ExpansionLocEnd = 202 getExpansionLocForMacroDefLoc(Tokens[i+1].getLocation()); 203 204 Token Res; 205 if (CurTok.is(tok::hash)) // Stringify 206 Res = ActualArgs->getStringifiedArgument(ArgNo, PP, 207 ExpansionLocStart, 208 ExpansionLocEnd); 209 else { 210 // 'charify': don't bother caching these. 211 Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo), 212 PP, true, 213 ExpansionLocStart, 214 ExpansionLocEnd); 215 } 216 Res.setFlag(Token::StringifiedInMacro); 217 218 // The stringified/charified string leading space flag gets set to match 219 // the #/#@ operator. 220 if (NextTokGetsSpace) 221 Res.setFlag(Token::LeadingSpace); 222 223 ResultToks.push_back(Res); 224 MadeChange = true; 225 ++i; // Skip arg name. 226 NextTokGetsSpace = false; 227 continue; 228 } 229 230 // Find out if there is a paste (##) operator before or after the token. 231 bool NonEmptyPasteBefore = 232 !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 233 bool PasteBefore = i != 0 && Tokens[i-1].is(tok::hashhash); 234 bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash); 235 assert(!NonEmptyPasteBefore || PasteBefore); 236 237 // Otherwise, if this is not an argument token, just add the token to the 238 // output buffer. 239 IdentifierInfo *II = CurTok.getIdentifierInfo(); 240 int ArgNo = II ? Macro->getParameterNum(II) : -1; 241 if (ArgNo == -1) { 242 // This isn't an argument, just add it. 243 ResultToks.push_back(CurTok); 244 245 if (NextTokGetsSpace) { 246 ResultToks.back().setFlag(Token::LeadingSpace); 247 NextTokGetsSpace = false; 248 } else if (PasteBefore && !NonEmptyPasteBefore) 249 ResultToks.back().clearFlag(Token::LeadingSpace); 250 251 continue; 252 } 253 254 // An argument is expanded somehow, the result is different than the 255 // input. 256 MadeChange = true; 257 258 // Otherwise, this is a use of the argument. 259 260 // In Microsoft mode, remove the comma before __VA_ARGS__ to ensure there 261 // are no trailing commas if __VA_ARGS__ is empty. 262 if (!PasteBefore && ActualArgs->isVarargsElidedUse() && 263 MaybeRemoveCommaBeforeVaArgs(ResultToks, 264 /*HasPasteOperator=*/false, 265 Macro, ArgNo, PP)) 266 continue; 267 268 // If it is not the LHS/RHS of a ## operator, we must pre-expand the 269 // argument and substitute the expanded tokens into the result. This is 270 // C99 6.10.3.1p1. 271 if (!PasteBefore && !PasteAfter) { 272 const Token *ResultArgToks; 273 274 // Only preexpand the argument if it could possibly need it. This 275 // avoids some work in common cases. 276 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 277 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 278 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, Macro, PP)[0]; 279 else 280 ResultArgToks = ArgTok; // Use non-preexpanded tokens. 281 282 // If the arg token expanded into anything, append it. 283 if (ResultArgToks->isNot(tok::eof)) { 284 size_t FirstResult = ResultToks.size(); 285 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 286 ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 287 288 // In Microsoft-compatibility mode, we follow MSVC's preprocessing 289 // behavior by not considering single commas from nested macro 290 // expansions as argument separators. Set a flag on the token so we can 291 // test for this later when the macro expansion is processed. 292 if (PP.getLangOpts().MSVCCompat && NumToks == 1 && 293 ResultToks.back().is(tok::comma)) 294 ResultToks.back().setFlag(Token::IgnoredComma); 295 296 // If the '##' came from expanding an argument, turn it into 'unknown' 297 // to avoid pasting. 298 for (Token &Tok : llvm::make_range(ResultToks.begin() + FirstResult, 299 ResultToks.end())) { 300 if (Tok.is(tok::hashhash)) 301 Tok.setKind(tok::unknown); 302 } 303 304 if(ExpandLocStart.isValid()) { 305 updateLocForMacroArgTokens(CurTok.getLocation(), 306 ResultToks.begin()+FirstResult, 307 ResultToks.end()); 308 } 309 310 // If any tokens were substituted from the argument, the whitespace 311 // before the first token should match the whitespace of the arg 312 // identifier. 313 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 314 NextTokGetsSpace); 315 ResultToks[FirstResult].setFlagValue(Token::StartOfLine, false); 316 NextTokGetsSpace = false; 317 } 318 continue; 319 } 320 321 // Okay, we have a token that is either the LHS or RHS of a paste (##) 322 // argument. It gets substituted as its non-pre-expanded tokens. 323 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 324 unsigned NumToks = MacroArgs::getArgLength(ArgToks); 325 if (NumToks) { // Not an empty argument? 326 bool VaArgsPseudoPaste = false; 327 // If this is the GNU ", ## __VA_ARGS__" extension, and we just learned 328 // that __VA_ARGS__ expands to multiple tokens, avoid a pasting error when 329 // the expander trys to paste ',' with the first token of the __VA_ARGS__ 330 // expansion. 331 if (NonEmptyPasteBefore && ResultToks.size() >= 2 && 332 ResultToks[ResultToks.size()-2].is(tok::comma) && 333 (unsigned)ArgNo == Macro->getNumParams()-1 && 334 Macro->isVariadic()) { 335 VaArgsPseudoPaste = true; 336 // Remove the paste operator, report use of the extension. 337 PP.Diag(ResultToks.pop_back_val().getLocation(), diag::ext_paste_comma); 338 } 339 340 ResultToks.append(ArgToks, ArgToks+NumToks); 341 342 // If the '##' came from expanding an argument, turn it into 'unknown' 343 // to avoid pasting. 344 for (Token &Tok : llvm::make_range(ResultToks.end() - NumToks, 345 ResultToks.end())) { 346 if (Tok.is(tok::hashhash)) 347 Tok.setKind(tok::unknown); 348 } 349 350 if (ExpandLocStart.isValid()) { 351 updateLocForMacroArgTokens(CurTok.getLocation(), 352 ResultToks.end()-NumToks, ResultToks.end()); 353 } 354 355 // Transfer the leading whitespace information from the token 356 // (the macro argument) onto the first token of the 357 // expansion. Note that we don't do this for the GNU 358 // pseudo-paste extension ", ## __VA_ARGS__". 359 if (!VaArgsPseudoPaste) { 360 ResultToks[ResultToks.size() - NumToks].setFlagValue(Token::StartOfLine, 361 false); 362 ResultToks[ResultToks.size() - NumToks].setFlagValue( 363 Token::LeadingSpace, NextTokGetsSpace); 364 } 365 366 NextTokGetsSpace = false; 367 continue; 368 } 369 370 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 371 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 372 // implement this by eating ## operators when a LHS or RHS expands to 373 // empty. 374 if (PasteAfter) { 375 // Discard the argument token and skip (don't copy to the expansion 376 // buffer) the paste operator after it. 377 ++i; 378 continue; 379 } 380 381 // If this is on the RHS of a paste operator, we've already copied the 382 // paste operator to the ResultToks list, unless the LHS was empty too. 383 // Remove it. 384 assert(PasteBefore); 385 if (NonEmptyPasteBefore) { 386 assert(ResultToks.back().is(tok::hashhash)); 387 ResultToks.pop_back(); 388 } 389 390 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 391 // and if the macro had at least one real argument, and if the token before 392 // the ## was a comma, remove the comma. This is a GCC extension which is 393 // disabled when using -std=c99. 394 if (ActualArgs->isVarargsElidedUse()) 395 MaybeRemoveCommaBeforeVaArgs(ResultToks, 396 /*HasPasteOperator=*/true, 397 Macro, ArgNo, PP); 398 } 399 400 // If anything changed, install this as the new Tokens list. 401 if (MadeChange) { 402 assert(!OwnsTokens && "This would leak if we already own the token list"); 403 // This is deleted in the dtor. 404 NumTokens = ResultToks.size(); 405 // The tokens will be added to Preprocessor's cache and will be removed 406 // when this TokenLexer finishes lexing them. 407 Tokens = PP.cacheMacroExpandedTokens(this, ResultToks); 408 409 // The preprocessor cache of macro expanded tokens owns these tokens,not us. 410 OwnsTokens = false; 411 } 412} 413 414/// \brief Checks if two tokens form wide string literal. 415static bool isWideStringLiteralFromMacro(const Token &FirstTok, 416 const Token &SecondTok) { 417 return FirstTok.is(tok::identifier) && 418 FirstTok.getIdentifierInfo()->isStr("L") && SecondTok.isLiteral() && 419 SecondTok.stringifiedInMacro(); 420} 421 422/// Lex - Lex and return a token from this macro stream. 423/// 424bool TokenLexer::Lex(Token &Tok) { 425 // Lexing off the end of the macro, pop this macro off the expansion stack. 426 if (isAtEnd()) { 427 // If this is a macro (not a token stream), mark the macro enabled now 428 // that it is no longer being expanded. 429 if (Macro) Macro->EnableMacro(); 430 431 Tok.startToken(); 432 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 433 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace || NextTokGetsSpace); 434 if (CurToken == 0) 435 Tok.setFlag(Token::LeadingEmptyMacro); 436 return PP.HandleEndOfTokenLexer(Tok); 437 } 438 439 SourceManager &SM = PP.getSourceManager(); 440 441 // If this is the first token of the expanded result, we inherit spacing 442 // properties later. 443 bool isFirstToken = CurToken == 0; 444 445 // Get the next token to return. 446 Tok = Tokens[CurToken++]; 447 448 bool TokenIsFromPaste = false; 449 450 // If this token is followed by a token paste (##) operator, paste the tokens! 451 // Note that ## is a normal token when not expanding a macro. 452 if (!isAtEnd() && Macro && 453 (Tokens[CurToken].is(tok::hashhash) || 454 // Special processing of L#x macros in -fms-compatibility mode. 455 // Microsoft compiler is able to form a wide string literal from 456 // 'L#macro_arg' construct in a function-like macro. 457 (PP.getLangOpts().MSVCCompat && 458 isWideStringLiteralFromMacro(Tok, Tokens[CurToken])))) { 459 // When handling the microsoft /##/ extension, the final token is 460 // returned by PasteTokens, not the pasted token. 461 if (PasteTokens(Tok)) 462 return true; 463 464 TokenIsFromPaste = true; 465 } 466 467 // The token's current location indicate where the token was lexed from. We 468 // need this information to compute the spelling of the token, but any 469 // diagnostics for the expanded token should appear as if they came from 470 // ExpansionLoc. Pull this information together into a new SourceLocation 471 // that captures all of this. 472 if (ExpandLocStart.isValid() && // Don't do this for token streams. 473 // Check that the token's location was not already set properly. 474 SM.isBeforeInSLocAddrSpace(Tok.getLocation(), MacroStartSLocOffset)) { 475 SourceLocation instLoc; 476 if (Tok.is(tok::comment)) { 477 instLoc = SM.createExpansionLoc(Tok.getLocation(), 478 ExpandLocStart, 479 ExpandLocEnd, 480 Tok.getLength()); 481 } else { 482 instLoc = getExpansionLocForMacroDefLoc(Tok.getLocation()); 483 } 484 485 Tok.setLocation(instLoc); 486 } 487 488 // If this is the first token, set the lexical properties of the token to 489 // match the lexical properties of the macro identifier. 490 if (isFirstToken) { 491 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 492 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 493 } else { 494 // If this is not the first token, we may still need to pass through 495 // leading whitespace if we've expanded a macro. 496 if (AtStartOfLine) Tok.setFlag(Token::StartOfLine); 497 if (HasLeadingSpace) Tok.setFlag(Token::LeadingSpace); 498 } 499 AtStartOfLine = false; 500 HasLeadingSpace = false; 501 502 // Handle recursive expansion! 503 if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) { 504 // Change the kind of this identifier to the appropriate token kind, e.g. 505 // turning "for" into a keyword. 506 IdentifierInfo *II = Tok.getIdentifierInfo(); 507 Tok.setKind(II->getTokenID()); 508 509 // If this identifier was poisoned and from a paste, emit an error. This 510 // won't be handled by Preprocessor::HandleIdentifier because this is coming 511 // from a macro expansion. 512 if (II->isPoisoned() && TokenIsFromPaste) { 513 PP.HandlePoisonedIdentifier(Tok); 514 } 515 516 if (!DisableMacroExpansion && II->isHandleIdentifierCase()) 517 return PP.HandleIdentifier(Tok); 518 } 519 520 // Otherwise, return a normal token. 521 return true; 522} 523 524/// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ## 525/// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 526/// are more ## after it, chomp them iteratively. Return the result as Tok. 527/// If this returns true, the caller should immediately return the token. 528bool TokenLexer::PasteTokens(Token &Tok) { 529 // MSVC: If previous token was pasted, this must be a recovery from an invalid 530 // paste operation. Ignore spaces before this token to mimic MSVC output. 531 // Required for generating valid UUID strings in some MS headers. 532 if (PP.getLangOpts().MicrosoftExt && (CurToken >= 2) && 533 Tokens[CurToken - 2].is(tok::hashhash)) 534 Tok.clearFlag(Token::LeadingSpace); 535 536 SmallString<128> Buffer; 537 const char *ResultTokStrPtr = nullptr; 538 SourceLocation StartLoc = Tok.getLocation(); 539 SourceLocation PasteOpLoc; 540 do { 541 // Consume the ## operator if any. 542 PasteOpLoc = Tokens[CurToken].getLocation(); 543 if (Tokens[CurToken].is(tok::hashhash)) 544 ++CurToken; 545 assert(!isAtEnd() && "No token on the RHS of a paste operator!"); 546 547 // Get the RHS token. 548 const Token &RHS = Tokens[CurToken]; 549 550 // Allocate space for the result token. This is guaranteed to be enough for 551 // the two tokens. 552 Buffer.resize(Tok.getLength() + RHS.getLength()); 553 554 // Get the spelling of the LHS token in Buffer. 555 const char *BufPtr = &Buffer[0]; 556 bool Invalid = false; 557 unsigned LHSLen = PP.getSpelling(Tok, BufPtr, &Invalid); 558 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 559 memcpy(&Buffer[0], BufPtr, LHSLen); 560 if (Invalid) 561 return true; 562 563 BufPtr = Buffer.data() + LHSLen; 564 unsigned RHSLen = PP.getSpelling(RHS, BufPtr, &Invalid); 565 if (Invalid) 566 return true; 567 if (RHSLen && BufPtr != &Buffer[LHSLen]) 568 // Really, we want the chars in Buffer! 569 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 570 571 // Trim excess space. 572 Buffer.resize(LHSLen+RHSLen); 573 574 // Plop the pasted result (including the trailing newline and null) into a 575 // scratch buffer where we can lex it. 576 Token ResultTokTmp; 577 ResultTokTmp.startToken(); 578 579 // Claim that the tmp token is a string_literal so that we can get the 580 // character pointer back from CreateString in getLiteralData(). 581 ResultTokTmp.setKind(tok::string_literal); 582 PP.CreateString(Buffer, ResultTokTmp); 583 SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); 584 ResultTokStrPtr = ResultTokTmp.getLiteralData(); 585 586 // Lex the resultant pasted token into Result. 587 Token Result; 588 589 if (Tok.isAnyIdentifier() && RHS.isAnyIdentifier()) { 590 // Common paste case: identifier+identifier = identifier. Avoid creating 591 // a lexer and other overhead. 592 PP.IncrementPasteCounter(true); 593 Result.startToken(); 594 Result.setKind(tok::raw_identifier); 595 Result.setRawIdentifierData(ResultTokStrPtr); 596 Result.setLocation(ResultTokLoc); 597 Result.setLength(LHSLen+RHSLen); 598 } else { 599 PP.IncrementPasteCounter(false); 600 601 assert(ResultTokLoc.isFileID() && 602 "Should be a raw location into scratch buffer"); 603 SourceManager &SourceMgr = PP.getSourceManager(); 604 FileID LocFileID = SourceMgr.getFileID(ResultTokLoc); 605 606 bool Invalid = false; 607 const char *ScratchBufStart 608 = SourceMgr.getBufferData(LocFileID, &Invalid).data(); 609 if (Invalid) 610 return false; 611 612 // Make a lexer to lex this string from. Lex just this one token. 613 // Make a lexer object so that we lex and expand the paste result. 614 Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID), 615 PP.getLangOpts(), ScratchBufStart, 616 ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen); 617 618 // Lex a token in raw mode. This way it won't look up identifiers 619 // automatically, lexing off the end will return an eof token, and 620 // warnings are disabled. This returns true if the result token is the 621 // entire buffer. 622 bool isInvalid = !TL.LexFromRawLexer(Result); 623 624 // If we got an EOF token, we didn't form even ONE token. For example, we 625 // did "/ ## /" to get "//". 626 isInvalid |= Result.is(tok::eof); 627 628 // If pasting the two tokens didn't form a full new token, this is an 629 // error. This occurs with "x ## +" and other stuff. Return with Tok 630 // unmodified and with RHS as the next token to lex. 631 if (isInvalid) { 632 // Explicitly convert the token location to have proper expansion 633 // information so that the user knows where it came from. 634 SourceManager &SM = PP.getSourceManager(); 635 SourceLocation Loc = 636 SM.createExpansionLoc(PasteOpLoc, ExpandLocStart, ExpandLocEnd, 2); 637 638 // Test for the Microsoft extension of /##/ turning into // here on the 639 // error path. 640 if (PP.getLangOpts().MicrosoftExt && Tok.is(tok::slash) && 641 RHS.is(tok::slash)) { 642 HandleMicrosoftCommentPaste(Tok, Loc); 643 return true; 644 } 645 646 // Do not emit the error when preprocessing assembler code. 647 if (!PP.getLangOpts().AsmPreprocessor) { 648 // If we're in microsoft extensions mode, downgrade this from a hard 649 // error to an extension that defaults to an error. This allows 650 // disabling it. 651 PP.Diag(Loc, PP.getLangOpts().MicrosoftExt ? diag::ext_pp_bad_paste_ms 652 : diag::err_pp_bad_paste) 653 << Buffer; 654 } 655 656 // An error has occurred so exit loop. 657 break; 658 } 659 660 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 661 // operator. 662 if (Result.is(tok::hashhash)) 663 Result.setKind(tok::unknown); 664 } 665 666 // Transfer properties of the LHS over the Result. 667 Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine()); 668 Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace()); 669 670 // Finally, replace LHS with the result, consume the RHS, and iterate. 671 ++CurToken; 672 Tok = Result; 673 } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)); 674 675 SourceLocation EndLoc = Tokens[CurToken - 1].getLocation(); 676 677 // The token's current location indicate where the token was lexed from. We 678 // need this information to compute the spelling of the token, but any 679 // diagnostics for the expanded token should appear as if the token was 680 // expanded from the full ## expression. Pull this information together into 681 // a new SourceLocation that captures all of this. 682 SourceManager &SM = PP.getSourceManager(); 683 if (StartLoc.isFileID()) 684 StartLoc = getExpansionLocForMacroDefLoc(StartLoc); 685 if (EndLoc.isFileID()) 686 EndLoc = getExpansionLocForMacroDefLoc(EndLoc); 687 FileID MacroFID = SM.getFileID(MacroExpansionStart); 688 while (SM.getFileID(StartLoc) != MacroFID) 689 StartLoc = SM.getImmediateExpansionRange(StartLoc).first; 690 while (SM.getFileID(EndLoc) != MacroFID) 691 EndLoc = SM.getImmediateExpansionRange(EndLoc).second; 692 693 Tok.setLocation(SM.createExpansionLoc(Tok.getLocation(), StartLoc, EndLoc, 694 Tok.getLength())); 695 696 // Now that we got the result token, it will be subject to expansion. Since 697 // token pasting re-lexes the result token in raw mode, identifier information 698 // isn't looked up. As such, if the result is an identifier, look up id info. 699 if (Tok.is(tok::raw_identifier)) { 700 // Look up the identifier info for the token. We disabled identifier lookup 701 // by saying we're skipping contents, so we need to do this manually. 702 PP.LookUpIdentifierInfo(Tok); 703 } 704 return false; 705} 706 707/// isNextTokenLParen - If the next token lexed will pop this macro off the 708/// expansion stack, return 2. If the next unexpanded token is a '(', return 709/// 1, otherwise return 0. 710unsigned TokenLexer::isNextTokenLParen() const { 711 // Out of tokens? 712 if (isAtEnd()) 713 return 2; 714 return Tokens[CurToken].is(tok::l_paren); 715} 716 717/// isParsingPreprocessorDirective - Return true if we are in the middle of a 718/// preprocessor directive. 719bool TokenLexer::isParsingPreprocessorDirective() const { 720 return Tokens[NumTokens-1].is(tok::eod) && !isAtEnd(); 721} 722 723/// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 724/// together to form a comment that comments out everything in the current 725/// macro, other active macros, and anything left on the current physical 726/// source line of the expanded buffer. Handle this by returning the 727/// first token on the next line. 728void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok, SourceLocation OpLoc) { 729 PP.Diag(OpLoc, diag::ext_comment_paste_microsoft); 730 731 // We 'comment out' the rest of this macro by just ignoring the rest of the 732 // tokens that have not been lexed yet, if any. 733 734 // Since this must be a macro, mark the macro enabled now that it is no longer 735 // being expanded. 736 assert(Macro && "Token streams can't paste comments"); 737 Macro->EnableMacro(); 738 739 PP.HandleMicrosoftCommentPaste(Tok); 740} 741 742/// \brief If \arg loc is a file ID and points inside the current macro 743/// definition, returns the appropriate source location pointing at the 744/// macro expansion source location entry, otherwise it returns an invalid 745/// SourceLocation. 746SourceLocation 747TokenLexer::getExpansionLocForMacroDefLoc(SourceLocation loc) const { 748 assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() && 749 "Not appropriate for token streams"); 750 assert(loc.isValid() && loc.isFileID()); 751 752 SourceManager &SM = PP.getSourceManager(); 753 assert(SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength) && 754 "Expected loc to come from the macro definition"); 755 756 unsigned relativeOffset = 0; 757 SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength, &relativeOffset); 758 return MacroExpansionStart.getLocWithOffset(relativeOffset); 759} 760 761/// \brief Finds the tokens that are consecutive (from the same FileID) 762/// creates a single SLocEntry, and assigns SourceLocations to each token that 763/// point to that SLocEntry. e.g for 764/// assert(foo == bar); 765/// There will be a single SLocEntry for the "foo == bar" chunk and locations 766/// for the 'foo', '==', 'bar' tokens will point inside that chunk. 767/// 768/// \arg begin_tokens will be updated to a position past all the found 769/// consecutive tokens. 770static void updateConsecutiveMacroArgTokens(SourceManager &SM, 771 SourceLocation InstLoc, 772 Token *&begin_tokens, 773 Token * end_tokens) { 774 assert(begin_tokens < end_tokens); 775 776 SourceLocation FirstLoc = begin_tokens->getLocation(); 777 SourceLocation CurLoc = FirstLoc; 778 779 // Compare the source location offset of tokens and group together tokens that 780 // are close, even if their locations point to different FileIDs. e.g. 781 // 782 // |bar | foo | cake | (3 tokens from 3 consecutive FileIDs) 783 // ^ ^ 784 // |bar foo cake| (one SLocEntry chunk for all tokens) 785 // 786 // we can perform this "merge" since the token's spelling location depends 787 // on the relative offset. 788 789 Token *NextTok = begin_tokens + 1; 790 for (; NextTok < end_tokens; ++NextTok) { 791 SourceLocation NextLoc = NextTok->getLocation(); 792 if (CurLoc.isFileID() != NextLoc.isFileID()) 793 break; // Token from different kind of FileID. 794 795 int RelOffs; 796 if (!SM.isInSameSLocAddrSpace(CurLoc, NextLoc, &RelOffs)) 797 break; // Token from different local/loaded location. 798 // Check that token is not before the previous token or more than 50 799 // "characters" away. 800 if (RelOffs < 0 || RelOffs > 50) 801 break; 802 803 if (CurLoc.isMacroID() && !SM.isWrittenInSameFile(CurLoc, NextLoc)) 804 break; // Token from a different macro. 805 806 CurLoc = NextLoc; 807 } 808 809 // For the consecutive tokens, find the length of the SLocEntry to contain 810 // all of them. 811 Token &LastConsecutiveTok = *(NextTok-1); 812 int LastRelOffs = 0; 813 SM.isInSameSLocAddrSpace(FirstLoc, LastConsecutiveTok.getLocation(), 814 &LastRelOffs); 815 unsigned FullLength = LastRelOffs + LastConsecutiveTok.getLength(); 816 817 // Create a macro expansion SLocEntry that will "contain" all of the tokens. 818 SourceLocation Expansion = 819 SM.createMacroArgExpansionLoc(FirstLoc, InstLoc,FullLength); 820 821 // Change the location of the tokens from the spelling location to the new 822 // expanded location. 823 for (; begin_tokens < NextTok; ++begin_tokens) { 824 Token &Tok = *begin_tokens; 825 int RelOffs = 0; 826 SM.isInSameSLocAddrSpace(FirstLoc, Tok.getLocation(), &RelOffs); 827 Tok.setLocation(Expansion.getLocWithOffset(RelOffs)); 828 } 829} 830 831/// \brief Creates SLocEntries and updates the locations of macro argument 832/// tokens to their new expanded locations. 833/// 834/// \param ArgIdDefLoc the location of the macro argument id inside the macro 835/// definition. 836/// \param Tokens the macro argument tokens to update. 837void TokenLexer::updateLocForMacroArgTokens(SourceLocation ArgIdSpellLoc, 838 Token *begin_tokens, 839 Token *end_tokens) { 840 SourceManager &SM = PP.getSourceManager(); 841 842 SourceLocation InstLoc = 843 getExpansionLocForMacroDefLoc(ArgIdSpellLoc); 844 845 while (begin_tokens < end_tokens) { 846 // If there's only one token just create a SLocEntry for it. 847 if (end_tokens - begin_tokens == 1) { 848 Token &Tok = *begin_tokens; 849 Tok.setLocation(SM.createMacroArgExpansionLoc(Tok.getLocation(), 850 InstLoc, 851 Tok.getLength())); 852 return; 853 } 854 855 updateConsecutiveMacroArgTokens(SM, InstLoc, begin_tokens, end_tokens); 856 } 857} 858 859void TokenLexer::PropagateLineStartLeadingSpaceInfo(Token &Result) { 860 AtStartOfLine = Result.isAtStartOfLine(); 861 HasLeadingSpace = Result.hasLeadingSpace(); 862} 863