TokenLexer.cpp revision 496af57256566c0db39b6a538bf7e4c6634afb2a
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/SmallVector.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 ILEnd, MacroArgs *Actuals) { 27 // If the client is reusing a TokenLexer, make sure to free any memory 28 // associated with it. 29 destroy(); 30 31 Macro = PP.getMacroInfo(Tok.getIdentifierInfo()); 32 ActualArgs = Actuals; 33 CurToken = 0; 34 35 InstantiateLocStart = Tok.getLocation(); 36 InstantiateLocEnd = ILEnd; 37 AtStartOfLine = Tok.isAtStartOfLine(); 38 HasLeadingSpace = Tok.hasLeadingSpace(); 39 Tokens = &*Macro->tokens_begin(); 40 OwnsTokens = false; 41 DisableMacroExpansion = false; 42 NumTokens = Macro->tokens_end()-Macro->tokens_begin(); 43 44 // If this is a function-like macro, expand the arguments and change 45 // Tokens to point to the expanded tokens. 46 if (Macro->isFunctionLike() && Macro->getNumArgs()) 47 ExpandFunctionArguments(); 48 49 // Mark the macro as currently disabled, so that it is not recursively 50 // expanded. The macro must be disabled only after argument pre-expansion of 51 // function-like macro arguments occurs. 52 Macro->DisableMacro(); 53} 54 55 56 57/// Create a TokenLexer for the specified token stream. This does not 58/// take ownership of the specified token vector. 59void TokenLexer::Init(const Token *TokArray, unsigned NumToks, 60 bool disableMacroExpansion, bool ownsTokens) { 61 // If the client is reusing a TokenLexer, make sure to free any memory 62 // associated with it. 63 destroy(); 64 65 Macro = 0; 66 ActualArgs = 0; 67 Tokens = TokArray; 68 OwnsTokens = ownsTokens; 69 DisableMacroExpansion = disableMacroExpansion; 70 NumTokens = NumToks; 71 CurToken = 0; 72 InstantiateLocStart = InstantiateLocEnd = SourceLocation(); 73 AtStartOfLine = false; 74 HasLeadingSpace = false; 75 76 // Set HasLeadingSpace/AtStartOfLine so that the first token will be 77 // returned unmodified. 78 if (NumToks != 0) { 79 AtStartOfLine = TokArray[0].isAtStartOfLine(); 80 HasLeadingSpace = TokArray[0].hasLeadingSpace(); 81 } 82} 83 84 85void TokenLexer::destroy() { 86 // If this was a function-like macro that actually uses its arguments, delete 87 // the expanded tokens. 88 if (OwnsTokens) { 89 delete [] Tokens; 90 Tokens = 0; 91 OwnsTokens = false; 92 } 93 94 // TokenLexer owns its formal arguments. 95 if (ActualArgs) ActualArgs->destroy(PP); 96} 97 98/// Expand the arguments of a function-like macro so that we can quickly 99/// return preexpanded tokens from Tokens. 100void TokenLexer::ExpandFunctionArguments() { 101 llvm::SmallVector<Token, 128> ResultToks; 102 103 // Loop through 'Tokens', expanding them into ResultToks. Keep 104 // track of whether we change anything. If not, no need to keep them. If so, 105 // we install the newly expanded sequence as the new 'Tokens' list. 106 bool MadeChange = false; 107 108 // NextTokGetsSpace - When this is true, the next token appended to the 109 // output list will get a leading space, regardless of whether it had one to 110 // begin with or not. This is used for placemarker support. 111 bool NextTokGetsSpace = false; 112 113 for (unsigned i = 0, e = NumTokens; i != e; ++i) { 114 // If we found the stringify operator, get the argument stringified. The 115 // preprocessor already verified that the following token is a macro name 116 // when the #define was parsed. 117 const Token &CurTok = Tokens[i]; 118 if (CurTok.is(tok::hash) || CurTok.is(tok::hashat)) { 119 int ArgNo = Macro->getArgumentNum(Tokens[i+1].getIdentifierInfo()); 120 assert(ArgNo != -1 && "Token following # is not an argument?"); 121 122 Token Res; 123 if (CurTok.is(tok::hash)) // Stringify 124 Res = ActualArgs->getStringifiedArgument(ArgNo, PP); 125 else { 126 // 'charify': don't bother caching these. 127 Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo), 128 PP, true); 129 } 130 131 // The stringified/charified string leading space flag gets set to match 132 // the #/#@ operator. 133 if (CurTok.hasLeadingSpace() || NextTokGetsSpace) 134 Res.setFlag(Token::LeadingSpace); 135 136 ResultToks.push_back(Res); 137 MadeChange = true; 138 ++i; // Skip arg name. 139 NextTokGetsSpace = false; 140 continue; 141 } 142 143 // Otherwise, if this is not an argument token, just add the token to the 144 // output buffer. 145 IdentifierInfo *II = CurTok.getIdentifierInfo(); 146 int ArgNo = II ? Macro->getArgumentNum(II) : -1; 147 if (ArgNo == -1) { 148 // This isn't an argument, just add it. 149 ResultToks.push_back(CurTok); 150 151 if (NextTokGetsSpace) { 152 ResultToks.back().setFlag(Token::LeadingSpace); 153 NextTokGetsSpace = false; 154 } 155 continue; 156 } 157 158 // An argument is expanded somehow, the result is different than the 159 // input. 160 MadeChange = true; 161 162 // Otherwise, this is a use of the argument. Find out if there is a paste 163 // (##) operator before or after the argument. 164 bool PasteBefore = 165 !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 166 bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash); 167 168 // If it is not the LHS/RHS of a ## operator, we must pre-expand the 169 // argument and substitute the expanded tokens into the result. This is 170 // C99 6.10.3.1p1. 171 if (!PasteBefore && !PasteAfter) { 172 const Token *ResultArgToks; 173 174 // Only preexpand the argument if it could possibly need it. This 175 // avoids some work in common cases. 176 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 177 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 178 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, Macro, PP)[0]; 179 else 180 ResultArgToks = ArgTok; // Use non-preexpanded tokens. 181 182 // If the arg token expanded into anything, append it. 183 if (ResultArgToks->isNot(tok::eof)) { 184 unsigned FirstResult = ResultToks.size(); 185 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 186 ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 187 188 // If any tokens were substituted from the argument, the whitespace 189 // before the first token should match the whitespace of the arg 190 // identifier. 191 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 192 CurTok.hasLeadingSpace() || 193 NextTokGetsSpace); 194 NextTokGetsSpace = false; 195 } else { 196 // If this is an empty argument, and if there was whitespace before the 197 // formal token, make sure the next token gets whitespace before it. 198 NextTokGetsSpace = CurTok.hasLeadingSpace(); 199 } 200 continue; 201 } 202 203 // Okay, we have a token that is either the LHS or RHS of a paste (##) 204 // argument. It gets substituted as its non-pre-expanded tokens. 205 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 206 unsigned NumToks = MacroArgs::getArgLength(ArgToks); 207 if (NumToks) { // Not an empty argument? 208 // If this is the GNU ", ## __VA_ARG__" extension, and we just learned 209 // that __VA_ARG__ expands to multiple tokens, avoid a pasting error when 210 // the expander trys to paste ',' with the first token of the __VA_ARG__ 211 // expansion. 212 if (PasteBefore && ResultToks.size() >= 2 && 213 ResultToks[ResultToks.size()-2].is(tok::comma) && 214 (unsigned)ArgNo == Macro->getNumArgs()-1 && 215 Macro->isVariadic()) { 216 // Remove the paste operator, report use of the extension. 217 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 218 ResultToks.pop_back(); 219 } 220 221 ResultToks.append(ArgToks, ArgToks+NumToks); 222 223 // If this token (the macro argument) was supposed to get leading 224 // whitespace, transfer this information onto the first token of the 225 // expansion. 226 // 227 // Do not do this if the paste operator occurs before the macro argument, 228 // as in "A ## MACROARG". In valid code, the first token will get 229 // smooshed onto the preceding one anyway (forming AMACROARG). In 230 // assembler-with-cpp mode, invalid pastes are allowed through: in this 231 // case, we do not want the extra whitespace to be added. For example, 232 // we want ". ## foo" -> ".foo" not ". foo". 233 if ((CurTok.hasLeadingSpace() || NextTokGetsSpace) && 234 !PasteBefore) 235 ResultToks[ResultToks.size()-NumToks].setFlag(Token::LeadingSpace); 236 237 NextTokGetsSpace = false; 238 continue; 239 } 240 241 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 242 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 243 // implement this by eating ## operators when a LHS or RHS expands to 244 // empty. 245 NextTokGetsSpace |= CurTok.hasLeadingSpace(); 246 if (PasteAfter) { 247 // Discard the argument token and skip (don't copy to the expansion 248 // buffer) the paste operator after it. 249 NextTokGetsSpace |= Tokens[i+1].hasLeadingSpace(); 250 ++i; 251 continue; 252 } 253 254 // If this is on the RHS of a paste operator, we've already copied the 255 // paste operator to the ResultToks list. Remove it. 256 assert(PasteBefore && ResultToks.back().is(tok::hashhash)); 257 NextTokGetsSpace |= ResultToks.back().hasLeadingSpace(); 258 ResultToks.pop_back(); 259 260 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 261 // and if the macro had at least one real argument, and if the token before 262 // the ## was a comma, remove the comma. 263 if ((unsigned)ArgNo == Macro->getNumArgs()-1 && // is __VA_ARGS__ 264 ActualArgs->isVarargsElidedUse() && // Argument elided. 265 !ResultToks.empty() && ResultToks.back().is(tok::comma)) { 266 // Never add a space, even if the comma, ##, or arg had a space. 267 NextTokGetsSpace = false; 268 // Remove the paste operator, report use of the extension. 269 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 270 ResultToks.pop_back(); 271 272 // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"), 273 // then removal of the comma should produce a placemarker token (in C99 274 // terms) which we model by popping off the previous ##, giving us a plain 275 // "X" when __VA_ARGS__ is empty. 276 if (!ResultToks.empty() && ResultToks.back().is(tok::hashhash)) 277 ResultToks.pop_back(); 278 } 279 continue; 280 } 281 282 // If anything changed, install this as the new Tokens list. 283 if (MadeChange) { 284 assert(!OwnsTokens && "This would leak if we already own the token list"); 285 // This is deleted in the dtor. 286 NumTokens = ResultToks.size(); 287 llvm::BumpPtrAllocator &Alloc = PP.getPreprocessorAllocator(); 288 Token *Res = 289 static_cast<Token *>(Alloc.Allocate(sizeof(Token)*ResultToks.size(), 290 llvm::alignof<Token>())); 291 if (NumTokens) 292 memcpy(Res, &ResultToks[0], NumTokens*sizeof(Token)); 293 Tokens = Res; 294 295 // The preprocessor bump pointer owns these tokens, not us. 296 OwnsTokens = false; 297 } 298} 299 300/// Lex - Lex and return a token from this macro stream. 301/// 302void TokenLexer::Lex(Token &Tok) { 303 // Lexing off the end of the macro, pop this macro off the expansion stack. 304 if (isAtEnd()) { 305 // If this is a macro (not a token stream), mark the macro enabled now 306 // that it is no longer being expanded. 307 if (Macro) Macro->EnableMacro(); 308 309 // Pop this context off the preprocessors lexer stack and get the next 310 // token. This will delete "this" so remember the PP instance var. 311 Preprocessor &PPCache = PP; 312 if (PP.HandleEndOfTokenLexer(Tok)) 313 return; 314 315 // HandleEndOfTokenLexer may not return a token. If it doesn't, lex 316 // whatever is next. 317 return PPCache.Lex(Tok); 318 } 319 320 // If this is the first token of the expanded result, we inherit spacing 321 // properties later. 322 bool isFirstToken = CurToken == 0; 323 324 // Get the next token to return. 325 Tok = Tokens[CurToken++]; 326 327 bool TokenIsFromPaste = false; 328 329 // If this token is followed by a token paste (##) operator, paste the tokens! 330 if (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)) { 331 // When handling the microsoft /##/ extension, the final token is 332 // returned by PasteTokens, not the pasted token. 333 if (PasteTokens(Tok)) 334 return; 335 336 TokenIsFromPaste = true; 337 } 338 339 // The token's current location indicate where the token was lexed from. We 340 // need this information to compute the spelling of the token, but any 341 // diagnostics for the expanded token should appear as if they came from 342 // InstantiationLoc. Pull this information together into a new SourceLocation 343 // that captures all of this. 344 if (InstantiateLocStart.isValid()) { // Don't do this for token streams. 345 SourceManager &SM = PP.getSourceManager(); 346 Tok.setLocation(SM.createInstantiationLoc(Tok.getLocation(), 347 InstantiateLocStart, 348 InstantiateLocEnd, 349 Tok.getLength())); 350 } 351 352 // If this is the first token, set the lexical properties of the token to 353 // match the lexical properties of the macro identifier. 354 if (isFirstToken) { 355 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 356 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 357 } 358 359 // Handle recursive expansion! 360 if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != 0) { 361 // Change the kind of this identifier to the appropriate token kind, e.g. 362 // turning "for" into a keyword. 363 IdentifierInfo *II = Tok.getIdentifierInfo(); 364 Tok.setKind(II->getTokenID()); 365 366 // If this identifier was poisoned and from a paste, emit an error. This 367 // won't be handled by Preprocessor::HandleIdentifier because this is coming 368 // from a macro expansion. 369 if (II->isPoisoned() && TokenIsFromPaste) { 370 // We warn about __VA_ARGS__ with poisoning. 371 if (II->isStr("__VA_ARGS__")) 372 PP.Diag(Tok, diag::ext_pp_bad_vaargs_use); 373 else 374 PP.Diag(Tok, diag::err_pp_used_poisoned_id); 375 } 376 377 if (!DisableMacroExpansion && II->isHandleIdentifierCase()) 378 PP.HandleIdentifier(Tok); 379 } 380 381 // Otherwise, return a normal token. 382} 383 384/// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ## 385/// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 386/// are more ## after it, chomp them iteratively. Return the result as Tok. 387/// If this returns true, the caller should immediately return the token. 388bool TokenLexer::PasteTokens(Token &Tok) { 389 llvm::SmallString<128> Buffer; 390 const char *ResultTokStrPtr = 0; 391 do { 392 // Consume the ## operator. 393 SourceLocation PasteOpLoc = Tokens[CurToken].getLocation(); 394 ++CurToken; 395 assert(!isAtEnd() && "No token on the RHS of a paste operator!"); 396 397 // Get the RHS token. 398 const Token &RHS = Tokens[CurToken]; 399 400 // Allocate space for the result token. This is guaranteed to be enough for 401 // the two tokens. 402 Buffer.resize(Tok.getLength() + RHS.getLength()); 403 404 // Get the spelling of the LHS token in Buffer. 405 const char *BufPtr = &Buffer[0]; 406 bool Invalid = false; 407 unsigned LHSLen = PP.getSpelling(Tok, BufPtr, &Invalid); 408 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 409 memcpy(&Buffer[0], BufPtr, LHSLen); 410 if (Invalid) 411 return true; 412 413 BufPtr = &Buffer[LHSLen]; 414 unsigned RHSLen = PP.getSpelling(RHS, BufPtr, &Invalid); 415 if (Invalid) 416 return true; 417 if (BufPtr != &Buffer[LHSLen]) // Really, we want the chars in Buffer! 418 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 419 420 // Trim excess space. 421 Buffer.resize(LHSLen+RHSLen); 422 423 // Plop the pasted result (including the trailing newline and null) into a 424 // scratch buffer where we can lex it. 425 Token ResultTokTmp; 426 ResultTokTmp.startToken(); 427 428 // Claim that the tmp token is a string_literal so that we can get the 429 // character pointer back from CreateString in getLiteralData(). 430 ResultTokTmp.setKind(tok::string_literal); 431 PP.CreateString(&Buffer[0], Buffer.size(), ResultTokTmp); 432 SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); 433 ResultTokStrPtr = ResultTokTmp.getLiteralData(); 434 435 // Lex the resultant pasted token into Result. 436 Token Result; 437 438 if (Tok.is(tok::identifier) && RHS.is(tok::identifier)) { 439 // Common paste case: identifier+identifier = identifier. Avoid creating 440 // a lexer and other overhead. 441 PP.IncrementPasteCounter(true); 442 Result.startToken(); 443 Result.setKind(tok::identifier); 444 Result.setLocation(ResultTokLoc); 445 Result.setLength(LHSLen+RHSLen); 446 } else { 447 PP.IncrementPasteCounter(false); 448 449 assert(ResultTokLoc.isFileID() && 450 "Should be a raw location into scratch buffer"); 451 SourceManager &SourceMgr = PP.getSourceManager(); 452 FileID LocFileID = SourceMgr.getFileID(ResultTokLoc); 453 454 bool Invalid = false; 455 const char *ScratchBufStart 456 = SourceMgr.getBufferData(LocFileID, &Invalid).data(); 457 if (Invalid) 458 return false; 459 460 // Make a lexer to lex this string from. Lex just this one token. 461 // Make a lexer object so that we lex and expand the paste result. 462 Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID), 463 PP.getLangOptions(), ScratchBufStart, 464 ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen); 465 466 // Lex a token in raw mode. This way it won't look up identifiers 467 // automatically, lexing off the end will return an eof token, and 468 // warnings are disabled. This returns true if the result token is the 469 // entire buffer. 470 bool isInvalid = !TL.LexFromRawLexer(Result); 471 472 // If we got an EOF token, we didn't form even ONE token. For example, we 473 // did "/ ## /" to get "//". 474 isInvalid |= Result.is(tok::eof); 475 476 // If pasting the two tokens didn't form a full new token, this is an 477 // error. This occurs with "x ## +" and other stuff. Return with Tok 478 // unmodified and with RHS as the next token to lex. 479 if (isInvalid) { 480 // Test for the Microsoft extension of /##/ turning into // here on the 481 // error path. 482 if (PP.getLangOptions().Microsoft && Tok.is(tok::slash) && 483 RHS.is(tok::slash)) { 484 HandleMicrosoftCommentPaste(Tok); 485 return true; 486 } 487 488 // Do not emit the error when preprocessing assembler code. 489 if (!PP.getLangOptions().AsmPreprocessor) { 490 // Explicitly convert the token location to have proper instantiation 491 // information so that the user knows where it came from. 492 SourceManager &SM = PP.getSourceManager(); 493 SourceLocation Loc = 494 SM.createInstantiationLoc(PasteOpLoc, InstantiateLocStart, 495 InstantiateLocEnd, 2); 496 // If we're in microsoft extensions mode, downgrade this from a hard 497 // error to a warning that defaults to an error. This allows 498 // disabling it. 499 PP.Diag(Loc, 500 PP.getLangOptions().Microsoft ? diag::err_pp_bad_paste_ms 501 : diag::err_pp_bad_paste) 502 << Buffer.str(); 503 } 504 505 // Do not consume the RHS. 506 --CurToken; 507 } 508 509 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 510 // operator. 511 if (Result.is(tok::hashhash)) 512 Result.setKind(tok::unknown); 513 } 514 515 // Transfer properties of the LHS over the the Result. 516 Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine()); 517 Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace()); 518 519 // Finally, replace LHS with the result, consume the RHS, and iterate. 520 ++CurToken; 521 Tok = Result; 522 } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)); 523 524 // Now that we got the result token, it will be subject to expansion. Since 525 // token pasting re-lexes the result token in raw mode, identifier information 526 // isn't looked up. As such, if the result is an identifier, look up id info. 527 if (Tok.is(tok::identifier)) { 528 // Look up the identifier info for the token. We disabled identifier lookup 529 // by saying we're skipping contents, so we need to do this manually. 530 PP.LookUpIdentifierInfo(Tok, ResultTokStrPtr); 531 } 532 return false; 533} 534 535/// isNextTokenLParen - If the next token lexed will pop this macro off the 536/// expansion stack, return 2. If the next unexpanded token is a '(', return 537/// 1, otherwise return 0. 538unsigned TokenLexer::isNextTokenLParen() const { 539 // Out of tokens? 540 if (isAtEnd()) 541 return 2; 542 return Tokens[CurToken].is(tok::l_paren); 543} 544 545 546/// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 547/// together to form a comment that comments out everything in the current 548/// macro, other active macros, and anything left on the current physical 549/// source line of the instantiated buffer. Handle this by returning the 550/// first token on the next line. 551void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok) { 552 // We 'comment out' the rest of this macro by just ignoring the rest of the 553 // tokens that have not been lexed yet, if any. 554 555 // Since this must be a macro, mark the macro enabled now that it is no longer 556 // being expanded. 557 assert(Macro && "Token streams can't paste comments"); 558 Macro->EnableMacro(); 559 560 PP.HandleMicrosoftCommentPaste(Tok); 561} 562