TokenLexer.cpp revision b99bdadf9ace3e8c793e12e9bb447501f86b4f72
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/Basic/Diagnostic.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, 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 InstantiateLoc = Tok.getLocation(); 36 37 // If the instantiation loc is not already a FileID, resolve it here. If we 38 // don't do this, we end up doing it once per token lexed. 39 if (!InstantiateLoc.isFileID()) 40 InstantiateLoc = PP.getSourceManager().getInstantiationLoc(InstantiateLoc); 41 42 AtStartOfLine = Tok.isAtStartOfLine(); 43 HasLeadingSpace = Tok.hasLeadingSpace(); 44 Tokens = &*Macro->tokens_begin(); 45 OwnsTokens = false; 46 DisableMacroExpansion = false; 47 NumTokens = Macro->tokens_end()-Macro->tokens_begin(); 48 49 // If this is a function-like macro, expand the arguments and change 50 // Tokens to point to the expanded tokens. 51 if (Macro->isFunctionLike() && Macro->getNumArgs()) 52 ExpandFunctionArguments(); 53 54 // Mark the macro as currently disabled, so that it is not recursively 55 // expanded. The macro must be disabled only after argument pre-expansion of 56 // function-like macro arguments occurs. 57 Macro->DisableMacro(); 58} 59 60 61 62/// Create a TokenLexer for the specified token stream. This does not 63/// take ownership of the specified token vector. 64void TokenLexer::Init(const Token *TokArray, unsigned NumToks, 65 bool disableMacroExpansion, bool ownsTokens) { 66 // If the client is reusing a TokenLexer, make sure to free any memory 67 // associated with it. 68 destroy(); 69 70 Macro = 0; 71 ActualArgs = 0; 72 Tokens = TokArray; 73 OwnsTokens = ownsTokens; 74 DisableMacroExpansion = disableMacroExpansion; 75 NumTokens = NumToks; 76 CurToken = 0; 77 InstantiateLoc = SourceLocation(); 78 AtStartOfLine = false; 79 HasLeadingSpace = false; 80 81 // Set HasLeadingSpace/AtStartOfLine so that the first token will be 82 // returned unmodified. 83 if (NumToks != 0) { 84 AtStartOfLine = TokArray[0].isAtStartOfLine(); 85 HasLeadingSpace = TokArray[0].hasLeadingSpace(); 86 } 87} 88 89 90void TokenLexer::destroy() { 91 // If this was a function-like macro that actually uses its arguments, delete 92 // the expanded tokens. 93 if (OwnsTokens) { 94 delete [] Tokens; 95 Tokens = 0; 96 } 97 98 // TokenLexer owns its formal arguments. 99 if (ActualArgs) ActualArgs->destroy(); 100} 101 102/// Expand the arguments of a function-like macro so that we can quickly 103/// return preexpanded tokens from Tokens. 104void TokenLexer::ExpandFunctionArguments() { 105 llvm::SmallVector<Token, 128> ResultToks; 106 107 // Loop through 'Tokens', expanding them into ResultToks. Keep 108 // track of whether we change anything. If not, no need to keep them. If so, 109 // we install the newly expanded sequence as the new 'Tokens' list. 110 bool MadeChange = false; 111 112 // NextTokGetsSpace - When this is true, the next token appended to the 113 // output list will get a leading space, regardless of whether it had one to 114 // begin with or not. This is used for placemarker support. 115 bool NextTokGetsSpace = false; 116 117 for (unsigned i = 0, e = NumTokens; i != e; ++i) { 118 // If we found the stringify operator, get the argument stringified. The 119 // preprocessor already verified that the following token is a macro name 120 // when the #define was parsed. 121 const Token &CurTok = Tokens[i]; 122 if (CurTok.is(tok::hash) || CurTok.is(tok::hashat)) { 123 int ArgNo = Macro->getArgumentNum(Tokens[i+1].getIdentifierInfo()); 124 assert(ArgNo != -1 && "Token following # is not an argument?"); 125 126 Token Res; 127 if (CurTok.is(tok::hash)) // Stringify 128 Res = ActualArgs->getStringifiedArgument(ArgNo, PP); 129 else { 130 // 'charify': don't bother caching these. 131 Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo), 132 PP, true); 133 } 134 135 // The stringified/charified string leading space flag gets set to match 136 // the #/#@ operator. 137 if (CurTok.hasLeadingSpace() || NextTokGetsSpace) 138 Res.setFlag(Token::LeadingSpace); 139 140 ResultToks.push_back(Res); 141 MadeChange = true; 142 ++i; // Skip arg name. 143 NextTokGetsSpace = false; 144 continue; 145 } 146 147 // Otherwise, if this is not an argument token, just add the token to the 148 // output buffer. 149 IdentifierInfo *II = CurTok.getIdentifierInfo(); 150 int ArgNo = II ? Macro->getArgumentNum(II) : -1; 151 if (ArgNo == -1) { 152 // This isn't an argument, just add it. 153 ResultToks.push_back(CurTok); 154 155 if (NextTokGetsSpace) { 156 ResultToks.back().setFlag(Token::LeadingSpace); 157 NextTokGetsSpace = false; 158 } 159 continue; 160 } 161 162 // An argument is expanded somehow, the result is different than the 163 // input. 164 MadeChange = true; 165 166 // Otherwise, this is a use of the argument. Find out if there is a paste 167 // (##) operator before or after the argument. 168 bool PasteBefore = 169 !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 170 bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash); 171 172 // If it is not the LHS/RHS of a ## operator, we must pre-expand the 173 // argument and substitute the expanded tokens into the result. This is 174 // C99 6.10.3.1p1. 175 if (!PasteBefore && !PasteAfter) { 176 const Token *ResultArgToks; 177 178 // Only preexpand the argument if it could possibly need it. This 179 // avoids some work in common cases. 180 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 181 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 182 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, PP)[0]; 183 else 184 ResultArgToks = ArgTok; // Use non-preexpanded tokens. 185 186 // If the arg token expanded into anything, append it. 187 if (ResultArgToks->isNot(tok::eof)) { 188 unsigned FirstResult = ResultToks.size(); 189 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 190 ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 191 192 // If any tokens were substituted from the argument, the whitespace 193 // before the first token should match the whitespace of the arg 194 // identifier. 195 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 196 CurTok.hasLeadingSpace() || 197 NextTokGetsSpace); 198 NextTokGetsSpace = false; 199 } else { 200 // If this is an empty argument, and if there was whitespace before the 201 // formal token, make sure the next token gets whitespace before it. 202 NextTokGetsSpace = CurTok.hasLeadingSpace(); 203 } 204 continue; 205 } 206 207 // Okay, we have a token that is either the LHS or RHS of a paste (##) 208 // argument. It gets substituted as its non-pre-expanded tokens. 209 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 210 unsigned NumToks = MacroArgs::getArgLength(ArgToks); 211 if (NumToks) { // Not an empty argument? 212 // If this is the GNU ", ## __VA_ARG__" extension, and we just learned 213 // that __VA_ARG__ expands to multiple tokens, avoid a pasting error when 214 // the expander trys to paste ',' with the first token of the __VA_ARG__ 215 // expansion. 216 if (PasteBefore && ResultToks.size() >= 2 && 217 ResultToks[ResultToks.size()-2].is(tok::comma) && 218 (unsigned)ArgNo == Macro->getNumArgs()-1 && 219 Macro->isVariadic()) { 220 // Remove the paste operator, report use of the extension. 221 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 222 ResultToks.pop_back(); 223 } 224 225 ResultToks.append(ArgToks, ArgToks+NumToks); 226 227 // If the next token was supposed to get leading whitespace, ensure it has 228 // it now. 229 if (NextTokGetsSpace) { 230 ResultToks[ResultToks.size()-NumToks].setFlag(Token::LeadingSpace); 231 NextTokGetsSpace = false; 232 } 233 continue; 234 } 235 236 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 237 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 238 // implement this by eating ## operators when a LHS or RHS expands to 239 // empty. 240 NextTokGetsSpace |= CurTok.hasLeadingSpace(); 241 if (PasteAfter) { 242 // Discard the argument token and skip (don't copy to the expansion 243 // buffer) the paste operator after it. 244 NextTokGetsSpace |= Tokens[i+1].hasLeadingSpace(); 245 ++i; 246 continue; 247 } 248 249 // If this is on the RHS of a paste operator, we've already copied the 250 // paste operator to the ResultToks list. Remove it. 251 assert(PasteBefore && ResultToks.back().is(tok::hashhash)); 252 NextTokGetsSpace |= ResultToks.back().hasLeadingSpace(); 253 ResultToks.pop_back(); 254 255 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 256 // and if the macro had at least one real argument, and if the token before 257 // the ## was a comma, remove the comma. 258 if ((unsigned)ArgNo == Macro->getNumArgs()-1 && // is __VA_ARGS__ 259 ActualArgs->isVarargsElidedUse() && // Argument elided. 260 !ResultToks.empty() && ResultToks.back().is(tok::comma)) { 261 // Never add a space, even if the comma, ##, or arg had a space. 262 NextTokGetsSpace = false; 263 // Remove the paste operator, report use of the extension. 264 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 265 ResultToks.pop_back(); 266 } 267 continue; 268 } 269 270 // If anything changed, install this as the new Tokens list. 271 if (MadeChange) { 272 // This is deleted in the dtor. 273 NumTokens = ResultToks.size(); 274 Token *Res = new Token[ResultToks.size()]; 275 if (NumTokens) 276 memcpy(Res, &ResultToks[0], NumTokens*sizeof(Token)); 277 Tokens = Res; 278 OwnsTokens = true; 279 } 280} 281 282/// Lex - Lex and return a token from this macro stream. 283/// 284void TokenLexer::Lex(Token &Tok) { 285 // Lexing off the end of the macro, pop this macro off the expansion stack. 286 if (isAtEnd()) { 287 // If this is a macro (not a token stream), mark the macro enabled now 288 // that it is no longer being expanded. 289 if (Macro) Macro->EnableMacro(); 290 291 // Pop this context off the preprocessors lexer stack and get the next 292 // token. This will delete "this" so remember the PP instance var. 293 Preprocessor &PPCache = PP; 294 if (PP.HandleEndOfTokenLexer(Tok)) 295 return; 296 297 // HandleEndOfTokenLexer may not return a token. If it doesn't, lex 298 // whatever is next. 299 return PPCache.Lex(Tok); 300 } 301 302 // If this is the first token of the expanded result, we inherit spacing 303 // properties later. 304 bool isFirstToken = CurToken == 0; 305 306 // Get the next token to return. 307 Tok = Tokens[CurToken++]; 308 309 // If this token is followed by a token paste (##) operator, paste the tokens! 310 if (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)) 311 if (PasteTokens(Tok)) { 312 // When handling the microsoft /##/ extension, the final token is 313 // returned by PasteTokens, not the pasted token. 314 return; 315 } 316 317 // The token's current location indicate where the token was lexed from. We 318 // need this information to compute the spelling of the token, but any 319 // diagnostics for the expanded token should appear as if they came from 320 // InstantiationLoc. Pull this information together into a new SourceLocation 321 // that captures all of this. 322 if (InstantiateLoc.isValid()) { // Don't do this for token streams. 323 SourceManager &SrcMgr = PP.getSourceManager(); 324 Tok.setLocation(SrcMgr.createInstantiationLoc(Tok.getLocation(), 325 InstantiateLoc, 326 Tok.getLength())); 327 } 328 329 // If this is the first token, set the lexical properties of the token to 330 // match the lexical properties of the macro identifier. 331 if (isFirstToken) { 332 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 333 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 334 } 335 336 // Handle recursive expansion! 337 if (IdentifierInfo *II = Tok.getIdentifierInfo()) { 338 // Change the kind of this identifier to the appropriate token kind, e.g. 339 // turning "for" into a keyword. 340 Tok.setKind(II->getTokenID()); 341 342 if (!DisableMacroExpansion && II->isHandleIdentifierCase()) 343 PP.HandleIdentifier(Tok); 344 } 345 346 // Otherwise, return a normal token. 347} 348 349/// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ## 350/// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 351/// are more ## after it, chomp them iteratively. Return the result as Tok. 352/// If this returns true, the caller should immediately return the token. 353bool TokenLexer::PasteTokens(Token &Tok) { 354 llvm::SmallVector<char, 128> Buffer; 355 do { 356 // Consume the ## operator. 357 SourceLocation PasteOpLoc = Tokens[CurToken].getLocation(); 358 ++CurToken; 359 assert(!isAtEnd() && "No token on the RHS of a paste operator!"); 360 361 // Get the RHS token. 362 const Token &RHS = Tokens[CurToken]; 363 364 bool isInvalid = false; 365 366 // Allocate space for the result token. This is guaranteed to be enough for 367 // the two tokens and a null terminator. 368 Buffer.resize(Tok.getLength() + RHS.getLength() + 1); 369 370 // Get the spelling of the LHS token in Buffer. 371 const char *BufPtr = &Buffer[0]; 372 unsigned LHSLen = PP.getSpelling(Tok, BufPtr); 373 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 374 memcpy(&Buffer[0], BufPtr, LHSLen); 375 376 BufPtr = &Buffer[LHSLen]; 377 unsigned RHSLen = PP.getSpelling(RHS, BufPtr); 378 if (BufPtr != &Buffer[LHSLen]) // Really, we want the chars in Buffer! 379 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 380 381 // Add null terminator. 382 Buffer[LHSLen+RHSLen] = '\0'; 383 384 // Trim excess space. 385 Buffer.resize(LHSLen+RHSLen+1); 386 387 // Plop the pasted result (including the trailing newline and null) into a 388 // scratch buffer where we can lex it. 389 SourceLocation ResultTokLoc = PP.CreateString(&Buffer[0], Buffer.size()); 390 391 // Lex the resultant pasted token into Result. 392 Token Result; 393 394 if (Tok.is(tok::identifier) && RHS.is(tok::identifier)) { 395 // Common paste case: identifier+identifier = identifier. Avoid creating 396 // a lexer and other overhead. 397 PP.IncrementPasteCounter(true); 398 Result.startToken(); 399 Result.setKind(tok::identifier); 400 Result.setLocation(ResultTokLoc); 401 Result.setLength(LHSLen+RHSLen); 402 } else { 403 PP.IncrementPasteCounter(false); 404 405 assert(ResultTokLoc.isFileID() && 406 "Should be a raw location into scratch buffer"); 407 SourceManager &SourceMgr = PP.getSourceManager(); 408 std::pair<FileID, unsigned> LocInfo = 409 SourceMgr.getDecomposedLoc(ResultTokLoc); 410 411 const char *ScratchBufStart =SourceMgr.getBufferData(LocInfo.first).first; 412 413 // Make a lexer to lex this string from. Lex just this one token. 414 const char *ResultStrData = ScratchBufStart+LocInfo.second; 415 416 // Make a lexer object so that we lex and expand the paste result. 417 Lexer TL(SourceMgr.getLocForStartOfFile(LocInfo.first), 418 PP.getLangOptions(), 419 ScratchBufStart, 420 ResultStrData, 421 ResultStrData+LHSLen+RHSLen /*don't include null*/); 422 423 // Lex a token in raw mode. This way it won't look up identifiers 424 // automatically, lexing off the end will return an eof token, and 425 // warnings are disabled. This returns true if the result token is the 426 // entire buffer. 427 bool IsComplete = TL.LexFromRawLexer(Result); 428 429 // If we got an EOF token, we didn't form even ONE token. For example, we 430 // did "/ ## /" to get "//". 431 IsComplete &= Result.isNot(tok::eof); 432 isInvalid = !IsComplete; 433 } 434 435 // If pasting the two tokens didn't form a full new token, this is an error. 436 // This occurs with "x ## +" and other stuff. Return with Tok unmodified 437 // and with RHS as the next token to lex. 438 if (isInvalid) { 439 // Test for the Microsoft extension of /##/ turning into // here on the 440 // error path. 441 if (PP.getLangOptions().Microsoft && Tok.is(tok::slash) && 442 RHS.is(tok::slash)) { 443 HandleMicrosoftCommentPaste(Tok); 444 return true; 445 } else { 446 // TODO: If not in assembler language mode. 447 PP.Diag(PasteOpLoc, diag::err_pp_bad_paste) 448 << std::string(Buffer.begin(), Buffer.end()-1); 449 return false; 450 } 451 } 452 453 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 454 // operator. 455 if (Result.is(tok::hashhash)) 456 Result.setKind(tok::unknown); 457 // FIXME: Turn __VA_ARGS__ into "not a token"? 458 459 // Transfer properties of the LHS over the the Result. 460 Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine()); 461 Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace()); 462 463 // Finally, replace LHS with the result, consume the RHS, and iterate. 464 ++CurToken; 465 Tok = Result; 466 } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)); 467 468 // Now that we got the result token, it will be subject to expansion. Since 469 // token pasting re-lexes the result token in raw mode, identifier information 470 // isn't looked up. As such, if the result is an identifier, look up id info. 471 if (Tok.is(tok::identifier)) { 472 // Look up the identifier info for the token. We disabled identifier lookup 473 // by saying we're skipping contents, so we need to do this manually. 474 Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok)); 475 } 476 return false; 477} 478 479/// isNextTokenLParen - If the next token lexed will pop this macro off the 480/// expansion stack, return 2. If the next unexpanded token is a '(', return 481/// 1, otherwise return 0. 482unsigned TokenLexer::isNextTokenLParen() const { 483 // Out of tokens? 484 if (isAtEnd()) 485 return 2; 486 return Tokens[CurToken].is(tok::l_paren); 487} 488 489 490/// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 491/// together to form a comment that comments out everything in the current 492/// macro, other active macros, and anything left on the current physical 493/// source line of the instantiated buffer. Handle this by returning the 494/// first token on the next line. 495void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok) { 496 // We 'comment out' the rest of this macro by just ignoring the rest of the 497 // tokens that have not been lexed yet, if any. 498 499 // Since this must be a macro, mark the macro enabled now that it is no longer 500 // being expanded. 501 assert(Macro && "Token streams can't paste comments"); 502 Macro->EnableMacro(); 503 504 PP.HandleMicrosoftCommentPaste(Tok); 505} 506