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