TokenLexer.cpp revision e7689886d6a68507ae90ca330bd24bf89d4d6df1
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 AtStartOfLine = Tok.isAtStartOfLine(); 37 HasLeadingSpace = Tok.hasLeadingSpace(); 38 Tokens = &*Macro->tokens_begin(); 39 OwnsTokens = false; 40 DisableMacroExpansion = false; 41 NumTokens = Macro->tokens_end()-Macro->tokens_begin(); 42 43 // If this is a function-like macro, expand the arguments and change 44 // Tokens to point to the expanded tokens. 45 if (Macro->isFunctionLike() && Macro->getNumArgs()) 46 ExpandFunctionArguments(); 47 48 // Mark the macro as currently disabled, so that it is not recursively 49 // expanded. The macro must be disabled only after argument pre-expansion of 50 // function-like macro arguments occurs. 51 Macro->DisableMacro(); 52} 53 54 55 56/// Create a TokenLexer for the specified token stream. This does not 57/// take ownership of the specified token vector. 58void TokenLexer::Init(const Token *TokArray, unsigned NumToks, 59 bool disableMacroExpansion, bool ownsTokens) { 60 // If the client is reusing a TokenLexer, make sure to free any memory 61 // associated with it. 62 destroy(); 63 64 Macro = 0; 65 ActualArgs = 0; 66 Tokens = TokArray; 67 OwnsTokens = ownsTokens; 68 DisableMacroExpansion = disableMacroExpansion; 69 NumTokens = NumToks; 70 CurToken = 0; 71 InstantiateLoc = SourceLocation(); 72 AtStartOfLine = false; 73 HasLeadingSpace = false; 74 75 // Set HasLeadingSpace/AtStartOfLine so that the first token will be 76 // returned unmodified. 77 if (NumToks != 0) { 78 AtStartOfLine = TokArray[0].isAtStartOfLine(); 79 HasLeadingSpace = TokArray[0].hasLeadingSpace(); 80 } 81} 82 83 84void TokenLexer::destroy() { 85 // If this was a function-like macro that actually uses its arguments, delete 86 // the expanded tokens. 87 if (OwnsTokens) { 88 delete [] Tokens; 89 Tokens = 0; 90 } 91 92 // TokenLexer owns its formal arguments. 93 if (ActualArgs) ActualArgs->destroy(); 94} 95 96/// Expand the arguments of a function-like macro so that we can quickly 97/// return preexpanded tokens from Tokens. 98void TokenLexer::ExpandFunctionArguments() { 99 llvm::SmallVector<Token, 128> ResultToks; 100 101 // Loop through 'Tokens', expanding them into ResultToks. Keep 102 // track of whether we change anything. If not, no need to keep them. If so, 103 // we install the newly expanded sequence as the new 'Tokens' list. 104 bool MadeChange = false; 105 106 // NextTokGetsSpace - When this is true, the next token appended to the 107 // output list will get a leading space, regardless of whether it had one to 108 // begin with or not. This is used for placemarker support. 109 bool NextTokGetsSpace = false; 110 111 for (unsigned i = 0, e = NumTokens; i != e; ++i) { 112 // If we found the stringify operator, get the argument stringified. The 113 // preprocessor already verified that the following token is a macro name 114 // when the #define was parsed. 115 const Token &CurTok = Tokens[i]; 116 if (CurTok.is(tok::hash) || CurTok.is(tok::hashat)) { 117 int ArgNo = Macro->getArgumentNum(Tokens[i+1].getIdentifierInfo()); 118 assert(ArgNo != -1 && "Token following # is not an argument?"); 119 120 Token Res; 121 if (CurTok.is(tok::hash)) // Stringify 122 Res = ActualArgs->getStringifiedArgument(ArgNo, PP); 123 else { 124 // 'charify': don't bother caching these. 125 Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo), 126 PP, true); 127 } 128 129 // The stringified/charified string leading space flag gets set to match 130 // the #/#@ operator. 131 if (CurTok.hasLeadingSpace() || NextTokGetsSpace) 132 Res.setFlag(Token::LeadingSpace); 133 134 ResultToks.push_back(Res); 135 MadeChange = true; 136 ++i; // Skip arg name. 137 NextTokGetsSpace = false; 138 continue; 139 } 140 141 // Otherwise, if this is not an argument token, just add the token to the 142 // output buffer. 143 IdentifierInfo *II = CurTok.getIdentifierInfo(); 144 int ArgNo = II ? Macro->getArgumentNum(II) : -1; 145 if (ArgNo == -1) { 146 // This isn't an argument, just add it. 147 ResultToks.push_back(CurTok); 148 149 if (NextTokGetsSpace) { 150 ResultToks.back().setFlag(Token::LeadingSpace); 151 NextTokGetsSpace = false; 152 } 153 continue; 154 } 155 156 // An argument is expanded somehow, the result is different than the 157 // input. 158 MadeChange = true; 159 160 // Otherwise, this is a use of the argument. Find out if there is a paste 161 // (##) operator before or after the argument. 162 bool PasteBefore = 163 !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 164 bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash); 165 166 // If it is not the LHS/RHS of a ## operator, we must pre-expand the 167 // argument and substitute the expanded tokens into the result. This is 168 // C99 6.10.3.1p1. 169 if (!PasteBefore && !PasteAfter) { 170 const Token *ResultArgToks; 171 172 // Only preexpand the argument if it could possibly need it. This 173 // avoids some work in common cases. 174 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 175 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 176 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, PP)[0]; 177 else 178 ResultArgToks = ArgTok; // Use non-preexpanded tokens. 179 180 // If the arg token expanded into anything, append it. 181 if (ResultArgToks->isNot(tok::eof)) { 182 unsigned FirstResult = ResultToks.size(); 183 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 184 ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 185 186 // If any tokens were substituted from the argument, the whitespace 187 // before the first token should match the whitespace of the arg 188 // identifier. 189 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 190 CurTok.hasLeadingSpace() || 191 NextTokGetsSpace); 192 NextTokGetsSpace = false; 193 } else { 194 // If this is an empty argument, and if there was whitespace before the 195 // formal token, make sure the next token gets whitespace before it. 196 NextTokGetsSpace = CurTok.hasLeadingSpace(); 197 } 198 continue; 199 } 200 201 // Okay, we have a token that is either the LHS or RHS of a paste (##) 202 // argument. It gets substituted as its non-pre-expanded tokens. 203 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 204 unsigned NumToks = MacroArgs::getArgLength(ArgToks); 205 if (NumToks) { // Not an empty argument? 206 // If this is the GNU ", ## __VA_ARG__" extension, and we just learned 207 // that __VA_ARG__ expands to multiple tokens, avoid a pasting error when 208 // the expander trys to paste ',' with the first token of the __VA_ARG__ 209 // expansion. 210 if (PasteBefore && ResultToks.size() >= 2 && 211 ResultToks[ResultToks.size()-2].is(tok::comma) && 212 (unsigned)ArgNo == Macro->getNumArgs()-1 && 213 Macro->isVariadic()) { 214 // Remove the paste operator, report use of the extension. 215 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 216 ResultToks.pop_back(); 217 } 218 219 ResultToks.append(ArgToks, ArgToks+NumToks); 220 221 // If the next token was supposed to get leading whitespace, ensure it has 222 // it now. 223 if (NextTokGetsSpace) { 224 ResultToks[ResultToks.size()-NumToks].setFlag(Token::LeadingSpace); 225 NextTokGetsSpace = false; 226 } 227 continue; 228 } 229 230 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 231 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 232 // implement this by eating ## operators when a LHS or RHS expands to 233 // empty. 234 NextTokGetsSpace |= CurTok.hasLeadingSpace(); 235 if (PasteAfter) { 236 // Discard the argument token and skip (don't copy to the expansion 237 // buffer) the paste operator after it. 238 NextTokGetsSpace |= Tokens[i+1].hasLeadingSpace(); 239 ++i; 240 continue; 241 } 242 243 // If this is on the RHS of a paste operator, we've already copied the 244 // paste operator to the ResultToks list. Remove it. 245 assert(PasteBefore && ResultToks.back().is(tok::hashhash)); 246 NextTokGetsSpace |= ResultToks.back().hasLeadingSpace(); 247 ResultToks.pop_back(); 248 249 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 250 // and if the macro had at least one real argument, and if the token before 251 // the ## was a comma, remove the comma. 252 if ((unsigned)ArgNo == Macro->getNumArgs()-1 && // is __VA_ARGS__ 253 ActualArgs->isVarargsElidedUse() && // Argument elided. 254 !ResultToks.empty() && ResultToks.back().is(tok::comma)) { 255 // Never add a space, even if the comma, ##, or arg had a space. 256 NextTokGetsSpace = false; 257 // Remove the paste operator, report use of the extension. 258 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 259 ResultToks.pop_back(); 260 } 261 continue; 262 } 263 264 // If anything changed, install this as the new Tokens list. 265 if (MadeChange) { 266 // This is deleted in the dtor. 267 NumTokens = ResultToks.size(); 268 Token *Res = new Token[ResultToks.size()]; 269 if (NumTokens) 270 memcpy(Res, &ResultToks[0], NumTokens*sizeof(Token)); 271 Tokens = Res; 272 OwnsTokens = true; 273 } 274} 275 276/// Lex - Lex and return a token from this macro stream. 277/// 278void TokenLexer::Lex(Token &Tok) { 279 // Lexing off the end of the macro, pop this macro off the expansion stack. 280 if (isAtEnd()) { 281 // If this is a macro (not a token stream), mark the macro enabled now 282 // that it is no longer being expanded. 283 if (Macro) Macro->EnableMacro(); 284 285 // Pop this context off the preprocessors lexer stack and get the next 286 // token. This will delete "this" so remember the PP instance var. 287 Preprocessor &PPCache = PP; 288 if (PP.HandleEndOfTokenLexer(Tok)) 289 return; 290 291 // HandleEndOfTokenLexer may not return a token. If it doesn't, lex 292 // whatever is next. 293 return PPCache.Lex(Tok); 294 } 295 296 // If this is the first token of the expanded result, we inherit spacing 297 // properties later. 298 bool isFirstToken = CurToken == 0; 299 300 // Get the next token to return. 301 Tok = Tokens[CurToken++]; 302 303 // If this token is followed by a token paste (##) operator, paste the tokens! 304 if (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)) 305 if (PasteTokens(Tok)) { 306 // When handling the microsoft /##/ extension, the final token is 307 // returned by PasteTokens, not the pasted token. 308 return; 309 } 310 311 // The token's current location indicate where the token was lexed from. We 312 // need this information to compute the spelling of the token, but any 313 // diagnostics for the expanded token should appear as if they came from 314 // InstantiationLoc. Pull this information together into a new SourceLocation 315 // that captures all of this. 316 if (InstantiateLoc.isValid()) { // Don't do this for token streams. 317 SourceManager &SrcMgr = PP.getSourceManager(); 318 Tok.setLocation(SrcMgr.createInstantiationLoc(Tok.getLocation(), 319 InstantiateLoc, 320 Tok.getLength())); 321 } 322 323 // If this is the first token, set the lexical properties of the token to 324 // match the lexical properties of the macro identifier. 325 if (isFirstToken) { 326 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 327 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 328 } 329 330 // Handle recursive expansion! 331 if (IdentifierInfo *II = Tok.getIdentifierInfo()) { 332 // Change the kind of this identifier to the appropriate token kind, e.g. 333 // turning "for" into a keyword. 334 Tok.setKind(II->getTokenID()); 335 336 if (!DisableMacroExpansion && II->isHandleIdentifierCase()) 337 PP.HandleIdentifier(Tok); 338 } 339 340 // Otherwise, return a normal token. 341} 342 343/// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ## 344/// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 345/// are more ## after it, chomp them iteratively. Return the result as Tok. 346/// If this returns true, the caller should immediately return the token. 347bool TokenLexer::PasteTokens(Token &Tok) { 348 llvm::SmallVector<char, 128> Buffer; 349 const char *ResultTokStrPtr = 0; 350 do { 351 // Consume the ## operator. 352 SourceLocation PasteOpLoc = Tokens[CurToken].getLocation(); 353 ++CurToken; 354 assert(!isAtEnd() && "No token on the RHS of a paste operator!"); 355 356 // Get the RHS token. 357 const Token &RHS = Tokens[CurToken]; 358 359 bool isInvalid = false; 360 361 // Allocate space for the result token. This is guaranteed to be enough for 362 // the two tokens and a null terminator. 363 Buffer.resize(Tok.getLength() + RHS.getLength() + 1); 364 365 // Get the spelling of the LHS token in Buffer. 366 const char *BufPtr = &Buffer[0]; 367 unsigned LHSLen = PP.getSpelling(Tok, BufPtr); 368 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 369 memcpy(&Buffer[0], BufPtr, LHSLen); 370 371 BufPtr = &Buffer[LHSLen]; 372 unsigned RHSLen = PP.getSpelling(RHS, BufPtr); 373 if (BufPtr != &Buffer[LHSLen]) // Really, we want the chars in Buffer! 374 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 375 376 // Add null terminator. 377 Buffer[LHSLen+RHSLen] = '\0'; 378 379 // Trim excess space. 380 Buffer.resize(LHSLen+RHSLen+1); 381 382 // Plop the pasted result (including the trailing newline and null) into a 383 // scratch buffer where we can lex it. 384 Token ResultTokTmp; 385 ResultTokTmp.startToken(); 386 387 // Claim that the tmp token is a string_literal so that we can get the 388 // character pointer back from CreateString. 389 ResultTokTmp.setKind(tok::string_literal); 390 PP.CreateString(&Buffer[0], Buffer.size(), ResultTokTmp); 391 SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); 392 ResultTokStrPtr = ResultTokTmp.getLiteralData(); 393 394 // Lex the resultant pasted token into Result. 395 Token Result; 396 397 if (Tok.is(tok::identifier) && RHS.is(tok::identifier)) { 398 // Common paste case: identifier+identifier = identifier. Avoid creating 399 // a lexer and other overhead. 400 PP.IncrementPasteCounter(true); 401 Result.startToken(); 402 Result.setKind(tok::identifier); 403 Result.setLocation(ResultTokLoc); 404 Result.setLength(LHSLen+RHSLen); 405 } else { 406 PP.IncrementPasteCounter(false); 407 408 assert(ResultTokLoc.isFileID() && 409 "Should be a raw location into scratch buffer"); 410 SourceManager &SourceMgr = PP.getSourceManager(); 411 FileID LocFileID = SourceMgr.getFileID(ResultTokLoc); 412 413 const char *ScratchBufStart = SourceMgr.getBufferData(LocFileID).first; 414 415 // Make a lexer to lex this string from. Lex just this one token. 416 // Make a lexer object so that we lex and expand the paste result. 417 Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID), 418 PP.getLangOptions(), ScratchBufStart, 419 ResultTokStrPtr, 420 ResultTokStrPtr+LHSLen+RHSLen /*don't include null*/); 421 422 // Lex a token in raw mode. This way it won't look up identifiers 423 // automatically, lexing off the end will return an eof token, and 424 // warnings are disabled. This returns true if the result token is the 425 // entire buffer. 426 bool IsComplete = TL.LexFromRawLexer(Result); 427 428 // If we got an EOF token, we didn't form even ONE token. For example, we 429 // did "/ ## /" to get "//". 430 IsComplete &= Result.isNot(tok::eof); 431 isInvalid = !IsComplete; 432 } 433 434 // If pasting the two tokens didn't form a full new token, this is an error. 435 // This occurs with "x ## +" and other stuff. Return with Tok unmodified 436 // and with RHS as the next token to lex. 437 if (isInvalid) { 438 // Test for the Microsoft extension of /##/ turning into // here on the 439 // error path. 440 if (PP.getLangOptions().Microsoft && Tok.is(tok::slash) && 441 RHS.is(tok::slash)) { 442 HandleMicrosoftCommentPaste(Tok); 443 return true; 444 } 445 446 // 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 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 453 // operator. 454 if (Result.is(tok::hashhash)) 455 Result.setKind(tok::unknown); 456 // FIXME: Turn __VA_ARGS__ into "not a token"? 457 458 // Transfer properties of the LHS over the the Result. 459 Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine()); 460 Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace()); 461 462 // Finally, replace LHS with the result, consume the RHS, and iterate. 463 ++CurToken; 464 Tok = Result; 465 } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)); 466 467 // Now that we got the result token, it will be subject to expansion. Since 468 // token pasting re-lexes the result token in raw mode, identifier information 469 // isn't looked up. As such, if the result is an identifier, look up id info. 470 if (Tok.is(tok::identifier)) { 471 // Look up the identifier info for the token. We disabled identifier lookup 472 // by saying we're skipping contents, so we need to do this manually. 473 Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok, ResultTokStrPtr)); 474 } 475 return false; 476} 477 478/// isNextTokenLParen - If the next token lexed will pop this macro off the 479/// expansion stack, return 2. If the next unexpanded token is a '(', return 480/// 1, otherwise return 0. 481unsigned TokenLexer::isNextTokenLParen() const { 482 // Out of tokens? 483 if (isAtEnd()) 484 return 2; 485 return Tokens[CurToken].is(tok::l_paren); 486} 487 488 489/// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 490/// together to form a comment that comments out everything in the current 491/// macro, other active macros, and anything left on the current physical 492/// source line of the instantiated buffer. Handle this by returning the 493/// first token on the next line. 494void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok) { 495 // We 'comment out' the rest of this macro by just ignoring the rest of the 496 // tokens that have not been lexed yet, if any. 497 498 // Since this must be a macro, mark the macro enabled now that it is no longer 499 // being expanded. 500 assert(Macro && "Token streams can't paste comments"); 501 Macro->EnableMacro(); 502 503 PP.HandleMicrosoftCommentPaste(Tok); 504} 505