TokenLexer.cpp revision 263508
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/Lex/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" 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/// Remove comma ahead of __VA_ARGS__, if present, according to compiler dialect 123/// settings. Returns true if the comma is removed. 124static bool MaybeRemoveCommaBeforeVaArgs(SmallVectorImpl<Token> &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. 173void 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 NonEmptyPasteBefore = 248 !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 249 bool PasteBefore = i != 0 && Tokens[i-1].is(tok::hashhash); 250 bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash); 251 assert(!NonEmptyPasteBefore || PasteBefore); 252 253 // In Microsoft mode, remove the comma before __VA_ARGS__ to ensure there 254 // are no trailing commas if __VA_ARGS__ is empty. 255 if (!PasteBefore && ActualArgs->isVarargsElidedUse() && 256 MaybeRemoveCommaBeforeVaArgs(ResultToks, NextTokGetsSpace, 257 /*HasPasteOperator=*/false, 258 Macro, ArgNo, PP)) 259 continue; 260 261 // If it is not the LHS/RHS of a ## operator, we must pre-expand the 262 // argument and substitute the expanded tokens into the result. This is 263 // C99 6.10.3.1p1. 264 if (!PasteBefore && !PasteAfter) { 265 const Token *ResultArgToks; 266 267 // Only preexpand the argument if it could possibly need it. This 268 // avoids some work in common cases. 269 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 270 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 271 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, Macro, PP)[0]; 272 else 273 ResultArgToks = ArgTok; // Use non-preexpanded tokens. 274 275 // If the arg token expanded into anything, append it. 276 if (ResultArgToks->isNot(tok::eof)) { 277 unsigned FirstResult = ResultToks.size(); 278 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 279 ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 280 281 // In Microsoft-compatibility mode, we follow MSVC's preprocessing 282 // behavior by not considering single commas from nested macro 283 // expansions as argument separators. Set a flag on the token so we can 284 // test for this later when the macro expansion is processed. 285 if (PP.getLangOpts().MicrosoftMode && NumToks == 1 && 286 ResultToks.back().is(tok::comma)) 287 ResultToks.back().setFlag(Token::IgnoredComma); 288 289 // If the '##' came from expanding an argument, turn it into 'unknown' 290 // to avoid pasting. 291 for (unsigned i = FirstResult, e = ResultToks.size(); i != e; ++i) { 292 Token &Tok = ResultToks[i]; 293 if (Tok.is(tok::hashhash)) 294 Tok.setKind(tok::unknown); 295 } 296 297 if(ExpandLocStart.isValid()) { 298 updateLocForMacroArgTokens(CurTok.getLocation(), 299 ResultToks.begin()+FirstResult, 300 ResultToks.end()); 301 } 302 303 // If any tokens were substituted from the argument, the whitespace 304 // before the first token should match the whitespace of the arg 305 // identifier. 306 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 307 CurTok.hasLeadingSpace() || 308 NextTokGetsSpace); 309 NextTokGetsSpace = false; 310 } else { 311 // If this is an empty argument, and if there was whitespace before the 312 // formal token, make sure the next token gets whitespace before it. 313 NextTokGetsSpace = CurTok.hasLeadingSpace(); 314 } 315 continue; 316 } 317 318 // Okay, we have a token that is either the LHS or RHS of a paste (##) 319 // argument. It gets substituted as its non-pre-expanded tokens. 320 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 321 unsigned NumToks = MacroArgs::getArgLength(ArgToks); 322 if (NumToks) { // Not an empty argument? 323 // If this is the GNU ", ## __VA_ARGS__" extension, and we just learned 324 // that __VA_ARGS__ expands to multiple tokens, avoid a pasting error when 325 // the expander trys to paste ',' with the first token of the __VA_ARGS__ 326 // expansion. 327 if (NonEmptyPasteBefore && ResultToks.size() >= 2 && 328 ResultToks[ResultToks.size()-2].is(tok::comma) && 329 (unsigned)ArgNo == Macro->getNumArgs()-1 && 330 Macro->isVariadic()) { 331 // Remove the paste operator, report use of the extension. 332 PP.Diag(ResultToks.pop_back_val().getLocation(), diag::ext_paste_comma); 333 } 334 335 ResultToks.append(ArgToks, ArgToks+NumToks); 336 337 // If the '##' came from expanding an argument, turn it into 'unknown' 338 // to avoid pasting. 339 for (unsigned i = ResultToks.size() - NumToks, e = ResultToks.size(); 340 i != e; ++i) { 341 Token &Tok = ResultToks[i]; 342 if (Tok.is(tok::hashhash)) 343 Tok.setKind(tok::unknown); 344 } 345 346 if (ExpandLocStart.isValid()) { 347 updateLocForMacroArgTokens(CurTok.getLocation(), 348 ResultToks.end()-NumToks, ResultToks.end()); 349 } 350 351 // If this token (the macro argument) was supposed to get leading 352 // whitespace, transfer this information onto the first token of the 353 // expansion. 354 // 355 // Do not do this if the paste operator occurs before the macro argument, 356 // as in "A ## MACROARG". In valid code, the first token will get 357 // smooshed onto the preceding one anyway (forming AMACROARG). In 358 // assembler-with-cpp mode, invalid pastes are allowed through: in this 359 // case, we do not want the extra whitespace to be added. For example, 360 // we want ". ## foo" -> ".foo" not ". foo". 361 if ((CurTok.hasLeadingSpace() || NextTokGetsSpace) && 362 !NonEmptyPasteBefore) 363 ResultToks[ResultToks.size()-NumToks].setFlag(Token::LeadingSpace); 364 365 NextTokGetsSpace = false; 366 continue; 367 } 368 369 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 370 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 371 // implement this by eating ## operators when a LHS or RHS expands to 372 // empty. 373 NextTokGetsSpace |= CurTok.hasLeadingSpace(); 374 if (PasteAfter) { 375 // Discard the argument token and skip (don't copy to the expansion 376 // buffer) the paste operator after it. 377 NextTokGetsSpace |= Tokens[i+1].hasLeadingSpace(); 378 ++i; 379 continue; 380 } 381 382 // If this is on the RHS of a paste operator, we've already copied the 383 // paste operator to the ResultToks list, unless the LHS was empty too. 384 // Remove it. 385 assert(PasteBefore); 386 if (NonEmptyPasteBefore) { 387 assert(ResultToks.back().is(tok::hashhash)); 388 NextTokGetsSpace |= ResultToks.pop_back_val().hasLeadingSpace(); 389 } 390 391 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 392 // and if the macro had at least one real argument, and if the token before 393 // the ## was a comma, remove the comma. This is a GCC extension which is 394 // disabled when using -std=c99. 395 if (ActualArgs->isVarargsElidedUse()) 396 MaybeRemoveCommaBeforeVaArgs(ResultToks, NextTokGetsSpace, 397 /*HasPasteOperator=*/true, 398 Macro, ArgNo, PP); 399 400 continue; 401 } 402 403 // If anything changed, install this as the new Tokens list. 404 if (MadeChange) { 405 assert(!OwnsTokens && "This would leak if we already own the token list"); 406 // This is deleted in the dtor. 407 NumTokens = ResultToks.size(); 408 // The tokens will be added to Preprocessor's cache and will be removed 409 // when this TokenLexer finishes lexing them. 410 Tokens = PP.cacheMacroExpandedTokens(this, ResultToks); 411 412 // The preprocessor cache of macro expanded tokens owns these tokens,not us. 413 OwnsTokens = false; 414 } 415} 416 417/// Lex - Lex and return a token from this macro stream. 418/// 419bool TokenLexer::Lex(Token &Tok) { 420 // Lexing off the end of the macro, pop this macro off the expansion stack. 421 if (isAtEnd()) { 422 // If this is a macro (not a token stream), mark the macro enabled now 423 // that it is no longer being expanded. 424 if (Macro) Macro->EnableMacro(); 425 426 Tok.startToken(); 427 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 428 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 429 if (CurToken == 0) 430 Tok.setFlag(Token::LeadingEmptyMacro); 431 return PP.HandleEndOfTokenLexer(Tok); 432 } 433 434 SourceManager &SM = PP.getSourceManager(); 435 436 // If this is the first token of the expanded result, we inherit spacing 437 // properties later. 438 bool isFirstToken = CurToken == 0; 439 440 // Get the next token to return. 441 Tok = Tokens[CurToken++]; 442 443 bool TokenIsFromPaste = false; 444 445 // If this token is followed by a token paste (##) operator, paste the tokens! 446 // Note that ## is a normal token when not expanding a macro. 447 if (!isAtEnd() && Tokens[CurToken].is(tok::hashhash) && Macro) { 448 // When handling the microsoft /##/ extension, the final token is 449 // returned by PasteTokens, not the pasted token. 450 if (PasteTokens(Tok)) 451 return true; 452 453 TokenIsFromPaste = true; 454 } 455 456 // The token's current location indicate where the token was lexed from. We 457 // need this information to compute the spelling of the token, but any 458 // diagnostics for the expanded token should appear as if they came from 459 // ExpansionLoc. Pull this information together into a new SourceLocation 460 // that captures all of this. 461 if (ExpandLocStart.isValid() && // Don't do this for token streams. 462 // Check that the token's location was not already set properly. 463 SM.isBeforeInSLocAddrSpace(Tok.getLocation(), MacroStartSLocOffset)) { 464 SourceLocation instLoc; 465 if (Tok.is(tok::comment)) { 466 instLoc = SM.createExpansionLoc(Tok.getLocation(), 467 ExpandLocStart, 468 ExpandLocEnd, 469 Tok.getLength()); 470 } else { 471 instLoc = getExpansionLocForMacroDefLoc(Tok.getLocation()); 472 } 473 474 Tok.setLocation(instLoc); 475 } 476 477 // If this is the first token, set the lexical properties of the token to 478 // match the lexical properties of the macro identifier. 479 if (isFirstToken) { 480 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 481 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 482 AtStartOfLine = false; 483 HasLeadingSpace = false; 484 } 485 486 // Handle recursive expansion! 487 if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != 0) { 488 // Change the kind of this identifier to the appropriate token kind, e.g. 489 // turning "for" into a keyword. 490 IdentifierInfo *II = Tok.getIdentifierInfo(); 491 Tok.setKind(II->getTokenID()); 492 493 // If this identifier was poisoned and from a paste, emit an error. This 494 // won't be handled by Preprocessor::HandleIdentifier because this is coming 495 // from a macro expansion. 496 if (II->isPoisoned() && TokenIsFromPaste) { 497 PP.HandlePoisonedIdentifier(Tok); 498 } 499 500 if (!DisableMacroExpansion && II->isHandleIdentifierCase()) 501 return PP.HandleIdentifier(Tok); 502 } 503 504 // Otherwise, return a normal token. 505 return true; 506} 507 508/// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ## 509/// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 510/// are more ## after it, chomp them iteratively. Return the result as Tok. 511/// If this returns true, the caller should immediately return the token. 512bool TokenLexer::PasteTokens(Token &Tok) { 513 SmallString<128> Buffer; 514 const char *ResultTokStrPtr = 0; 515 SourceLocation StartLoc = Tok.getLocation(); 516 SourceLocation PasteOpLoc; 517 do { 518 // Consume the ## operator. 519 PasteOpLoc = Tokens[CurToken].getLocation(); 520 ++CurToken; 521 assert(!isAtEnd() && "No token on the RHS of a paste operator!"); 522 523 // Get the RHS token. 524 const Token &RHS = Tokens[CurToken]; 525 526 // Allocate space for the result token. This is guaranteed to be enough for 527 // the two tokens. 528 Buffer.resize(Tok.getLength() + RHS.getLength()); 529 530 // Get the spelling of the LHS token in Buffer. 531 const char *BufPtr = &Buffer[0]; 532 bool Invalid = false; 533 unsigned LHSLen = PP.getSpelling(Tok, BufPtr, &Invalid); 534 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 535 memcpy(&Buffer[0], BufPtr, LHSLen); 536 if (Invalid) 537 return true; 538 539 BufPtr = &Buffer[LHSLen]; 540 unsigned RHSLen = PP.getSpelling(RHS, BufPtr, &Invalid); 541 if (Invalid) 542 return true; 543 if (BufPtr != &Buffer[LHSLen]) // Really, we want the chars in Buffer! 544 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 545 546 // Trim excess space. 547 Buffer.resize(LHSLen+RHSLen); 548 549 // Plop the pasted result (including the trailing newline and null) into a 550 // scratch buffer where we can lex it. 551 Token ResultTokTmp; 552 ResultTokTmp.startToken(); 553 554 // Claim that the tmp token is a string_literal so that we can get the 555 // character pointer back from CreateString in getLiteralData(). 556 ResultTokTmp.setKind(tok::string_literal); 557 PP.CreateString(Buffer, ResultTokTmp); 558 SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); 559 ResultTokStrPtr = ResultTokTmp.getLiteralData(); 560 561 // Lex the resultant pasted token into Result. 562 Token Result; 563 564 if (Tok.isAnyIdentifier() && RHS.isAnyIdentifier()) { 565 // Common paste case: identifier+identifier = identifier. Avoid creating 566 // a lexer and other overhead. 567 PP.IncrementPasteCounter(true); 568 Result.startToken(); 569 Result.setKind(tok::raw_identifier); 570 Result.setRawIdentifierData(ResultTokStrPtr); 571 Result.setLocation(ResultTokLoc); 572 Result.setLength(LHSLen+RHSLen); 573 } else { 574 PP.IncrementPasteCounter(false); 575 576 assert(ResultTokLoc.isFileID() && 577 "Should be a raw location into scratch buffer"); 578 SourceManager &SourceMgr = PP.getSourceManager(); 579 FileID LocFileID = SourceMgr.getFileID(ResultTokLoc); 580 581 bool Invalid = false; 582 const char *ScratchBufStart 583 = SourceMgr.getBufferData(LocFileID, &Invalid).data(); 584 if (Invalid) 585 return false; 586 587 // Make a lexer to lex this string from. Lex just this one token. 588 // Make a lexer object so that we lex and expand the paste result. 589 Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID), 590 PP.getLangOpts(), ScratchBufStart, 591 ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen); 592 593 // Lex a token in raw mode. This way it won't look up identifiers 594 // automatically, lexing off the end will return an eof token, and 595 // warnings are disabled. This returns true if the result token is the 596 // entire buffer. 597 bool isInvalid = !TL.LexFromRawLexer(Result); 598 599 // If we got an EOF token, we didn't form even ONE token. For example, we 600 // did "/ ## /" to get "//". 601 isInvalid |= Result.is(tok::eof); 602 603 // If pasting the two tokens didn't form a full new token, this is an 604 // error. This occurs with "x ## +" and other stuff. Return with Tok 605 // unmodified and with RHS as the next token to lex. 606 if (isInvalid) { 607 // Test for the Microsoft extension of /##/ turning into // here on the 608 // error path. 609 if (PP.getLangOpts().MicrosoftExt && Tok.is(tok::slash) && 610 RHS.is(tok::slash)) { 611 HandleMicrosoftCommentPaste(Tok); 612 return true; 613 } 614 615 // Do not emit the error when preprocessing assembler code. 616 if (!PP.getLangOpts().AsmPreprocessor) { 617 // Explicitly convert the token location to have proper expansion 618 // information so that the user knows where it came from. 619 SourceManager &SM = PP.getSourceManager(); 620 SourceLocation Loc = 621 SM.createExpansionLoc(PasteOpLoc, ExpandLocStart, ExpandLocEnd, 2); 622 // If we're in microsoft extensions mode, downgrade this from a hard 623 // error to a warning that defaults to an error. This allows 624 // disabling it. 625 PP.Diag(Loc, 626 PP.getLangOpts().MicrosoftExt ? diag::err_pp_bad_paste_ms 627 : diag::err_pp_bad_paste) 628 << Buffer.str(); 629 } 630 631 // An error has occurred so exit loop. 632 break; 633 } 634 635 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 636 // operator. 637 if (Result.is(tok::hashhash)) 638 Result.setKind(tok::unknown); 639 } 640 641 // Transfer properties of the LHS over the Result. 642 Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine()); 643 Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace()); 644 645 // Finally, replace LHS with the result, consume the RHS, and iterate. 646 ++CurToken; 647 Tok = Result; 648 } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)); 649 650 SourceLocation EndLoc = Tokens[CurToken - 1].getLocation(); 651 652 // The token's current location indicate where the token was lexed from. We 653 // need this information to compute the spelling of the token, but any 654 // diagnostics for the expanded token should appear as if the token was 655 // expanded from the full ## expression. Pull this information together into 656 // a new SourceLocation that captures all of this. 657 SourceManager &SM = PP.getSourceManager(); 658 if (StartLoc.isFileID()) 659 StartLoc = getExpansionLocForMacroDefLoc(StartLoc); 660 if (EndLoc.isFileID()) 661 EndLoc = getExpansionLocForMacroDefLoc(EndLoc); 662 FileID MacroFID = SM.getFileID(MacroExpansionStart); 663 while (SM.getFileID(StartLoc) != MacroFID) 664 StartLoc = SM.getImmediateExpansionRange(StartLoc).first; 665 while (SM.getFileID(EndLoc) != MacroFID) 666 EndLoc = SM.getImmediateExpansionRange(EndLoc).second; 667 668 Tok.setLocation(SM.createExpansionLoc(Tok.getLocation(), StartLoc, EndLoc, 669 Tok.getLength())); 670 671 // Now that we got the result token, it will be subject to expansion. Since 672 // token pasting re-lexes the result token in raw mode, identifier information 673 // isn't looked up. As such, if the result is an identifier, look up id info. 674 if (Tok.is(tok::raw_identifier)) { 675 // Look up the identifier info for the token. We disabled identifier lookup 676 // by saying we're skipping contents, so we need to do this manually. 677 PP.LookUpIdentifierInfo(Tok); 678 } 679 return false; 680} 681 682/// isNextTokenLParen - If the next token lexed will pop this macro off the 683/// expansion stack, return 2. If the next unexpanded token is a '(', return 684/// 1, otherwise return 0. 685unsigned TokenLexer::isNextTokenLParen() const { 686 // Out of tokens? 687 if (isAtEnd()) 688 return 2; 689 return Tokens[CurToken].is(tok::l_paren); 690} 691 692/// isParsingPreprocessorDirective - Return true if we are in the middle of a 693/// preprocessor directive. 694bool TokenLexer::isParsingPreprocessorDirective() const { 695 return Tokens[NumTokens-1].is(tok::eod) && !isAtEnd(); 696} 697 698/// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 699/// together to form a comment that comments out everything in the current 700/// macro, other active macros, and anything left on the current physical 701/// source line of the expanded buffer. Handle this by returning the 702/// first token on the next line. 703void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok) { 704 // We 'comment out' the rest of this macro by just ignoring the rest of the 705 // tokens that have not been lexed yet, if any. 706 707 // Since this must be a macro, mark the macro enabled now that it is no longer 708 // being expanded. 709 assert(Macro && "Token streams can't paste comments"); 710 Macro->EnableMacro(); 711 712 PP.HandleMicrosoftCommentPaste(Tok); 713} 714 715/// \brief If \arg loc is a file ID and points inside the current macro 716/// definition, returns the appropriate source location pointing at the 717/// macro expansion source location entry, otherwise it returns an invalid 718/// SourceLocation. 719SourceLocation 720TokenLexer::getExpansionLocForMacroDefLoc(SourceLocation loc) const { 721 assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() && 722 "Not appropriate for token streams"); 723 assert(loc.isValid() && loc.isFileID()); 724 725 SourceManager &SM = PP.getSourceManager(); 726 assert(SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength) && 727 "Expected loc to come from the macro definition"); 728 729 unsigned relativeOffset = 0; 730 SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength, &relativeOffset); 731 return MacroExpansionStart.getLocWithOffset(relativeOffset); 732} 733 734/// \brief Finds the tokens that are consecutive (from the same FileID) 735/// creates a single SLocEntry, and assigns SourceLocations to each token that 736/// point to that SLocEntry. e.g for 737/// assert(foo == bar); 738/// There will be a single SLocEntry for the "foo == bar" chunk and locations 739/// for the 'foo', '==', 'bar' tokens will point inside that chunk. 740/// 741/// \arg begin_tokens will be updated to a position past all the found 742/// consecutive tokens. 743static void updateConsecutiveMacroArgTokens(SourceManager &SM, 744 SourceLocation InstLoc, 745 Token *&begin_tokens, 746 Token * end_tokens) { 747 assert(begin_tokens < end_tokens); 748 749 SourceLocation FirstLoc = begin_tokens->getLocation(); 750 SourceLocation CurLoc = FirstLoc; 751 752 // Compare the source location offset of tokens and group together tokens that 753 // are close, even if their locations point to different FileIDs. e.g. 754 // 755 // |bar | foo | cake | (3 tokens from 3 consecutive FileIDs) 756 // ^ ^ 757 // |bar foo cake| (one SLocEntry chunk for all tokens) 758 // 759 // we can perform this "merge" since the token's spelling location depends 760 // on the relative offset. 761 762 Token *NextTok = begin_tokens + 1; 763 for (; NextTok < end_tokens; ++NextTok) { 764 SourceLocation NextLoc = NextTok->getLocation(); 765 if (CurLoc.isFileID() != NextLoc.isFileID()) 766 break; // Token from different kind of FileID. 767 768 int RelOffs; 769 if (!SM.isInSameSLocAddrSpace(CurLoc, NextLoc, &RelOffs)) 770 break; // Token from different local/loaded location. 771 // Check that token is not before the previous token or more than 50 772 // "characters" away. 773 if (RelOffs < 0 || RelOffs > 50) 774 break; 775 CurLoc = NextLoc; 776 } 777 778 // For the consecutive tokens, find the length of the SLocEntry to contain 779 // all of them. 780 Token &LastConsecutiveTok = *(NextTok-1); 781 int LastRelOffs = 0; 782 SM.isInSameSLocAddrSpace(FirstLoc, LastConsecutiveTok.getLocation(), 783 &LastRelOffs); 784 unsigned FullLength = LastRelOffs + LastConsecutiveTok.getLength(); 785 786 // Create a macro expansion SLocEntry that will "contain" all of the tokens. 787 SourceLocation Expansion = 788 SM.createMacroArgExpansionLoc(FirstLoc, InstLoc,FullLength); 789 790 // Change the location of the tokens from the spelling location to the new 791 // expanded location. 792 for (; begin_tokens < NextTok; ++begin_tokens) { 793 Token &Tok = *begin_tokens; 794 int RelOffs = 0; 795 SM.isInSameSLocAddrSpace(FirstLoc, Tok.getLocation(), &RelOffs); 796 Tok.setLocation(Expansion.getLocWithOffset(RelOffs)); 797 } 798} 799 800/// \brief Creates SLocEntries and updates the locations of macro argument 801/// tokens to their new expanded locations. 802/// 803/// \param ArgIdDefLoc the location of the macro argument id inside the macro 804/// definition. 805/// \param Tokens the macro argument tokens to update. 806void TokenLexer::updateLocForMacroArgTokens(SourceLocation ArgIdSpellLoc, 807 Token *begin_tokens, 808 Token *end_tokens) { 809 SourceManager &SM = PP.getSourceManager(); 810 811 SourceLocation InstLoc = 812 getExpansionLocForMacroDefLoc(ArgIdSpellLoc); 813 814 while (begin_tokens < end_tokens) { 815 // If there's only one token just create a SLocEntry for it. 816 if (end_tokens - begin_tokens == 1) { 817 Token &Tok = *begin_tokens; 818 Tok.setLocation(SM.createMacroArgExpansionLoc(Tok.getLocation(), 819 InstLoc, 820 Tok.getLength())); 821 return; 822 } 823 824 updateConsecutiveMacroArgTokens(SM, InstLoc, begin_tokens, end_tokens); 825 } 826} 827 828void TokenLexer::PropagateLineStartLeadingSpaceInfo(Token &Result) { 829 AtStartOfLine = Result.isAtStartOfLine(); 830 HasLeadingSpace = Result.hasLeadingSpace(); 831} 832