TokenLexer.cpp revision 6a170eb3ea6d6319277becabef68eb1a26bf8766
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 InstantiateLoc = Tok.getLocation(); 35 AtStartOfLine = Tok.isAtStartOfLine(); 36 HasLeadingSpace = Tok.hasLeadingSpace(); 37 Tokens = &*Macro->tokens_begin(); 38 OwnsTokens = false; 39 DisableMacroExpansion = false; 40 NumTokens = Macro->tokens_end()-Macro->tokens_begin(); 41 42 // If this is a function-like macro, expand the arguments and change 43 // Tokens to point to the expanded tokens. 44 if (Macro->isFunctionLike() && Macro->getNumArgs()) 45 ExpandFunctionArguments(); 46 47 // Mark the macro as currently disabled, so that it is not recursively 48 // expanded. The macro must be disabled only after argument pre-expansion of 49 // function-like macro arguments occurs. 50 Macro->DisableMacro(); 51} 52 53 54 55/// Create a TokenLexer for the specified token stream. This does not 56/// take ownership of the specified token vector. 57void TokenLexer::Init(const Token *TokArray, unsigned NumToks, 58 bool disableMacroExpansion, bool ownsTokens) { 59 // If the client is reusing a TokenLexer, make sure to free any memory 60 // associated with it. 61 destroy(); 62 63 Macro = 0; 64 ActualArgs = 0; 65 Tokens = TokArray; 66 OwnsTokens = ownsTokens; 67 DisableMacroExpansion = disableMacroExpansion; 68 NumTokens = NumToks; 69 CurToken = 0; 70 InstantiateLoc = SourceLocation(); 71 AtStartOfLine = false; 72 HasLeadingSpace = false; 73 74 // Set HasLeadingSpace/AtStartOfLine so that the first token will be 75 // returned unmodified. 76 if (NumToks != 0) { 77 AtStartOfLine = TokArray[0].isAtStartOfLine(); 78 HasLeadingSpace = TokArray[0].hasLeadingSpace(); 79 } 80} 81 82 83void TokenLexer::destroy() { 84 // If this was a function-like macro that actually uses its arguments, delete 85 // the expanded tokens. 86 if (OwnsTokens) { 87 delete [] Tokens; 88 Tokens = 0; 89 } 90 91 // TokenLexer owns its formal arguments. 92 if (ActualArgs) ActualArgs->destroy(); 93} 94 95/// Expand the arguments of a function-like macro so that we can quickly 96/// return preexpanded tokens from Tokens. 97void TokenLexer::ExpandFunctionArguments() { 98 llvm::SmallVector<Token, 128> ResultToks; 99 100 // Loop through 'Tokens', expanding them into ResultToks. Keep 101 // track of whether we change anything. If not, no need to keep them. If so, 102 // we install the newly expanded sequence as the new 'Tokens' list. 103 bool MadeChange = false; 104 105 // NextTokGetsSpace - When this is true, the next token appended to the 106 // output list will get a leading space, regardless of whether it had one to 107 // begin with or not. This is used for placemarker support. 108 bool NextTokGetsSpace = false; 109 110 for (unsigned i = 0, e = NumTokens; i != e; ++i) { 111 // If we found the stringify operator, get the argument stringified. The 112 // preprocessor already verified that the following token is a macro name 113 // when the #define was parsed. 114 const Token &CurTok = Tokens[i]; 115 if (CurTok.is(tok::hash) || CurTok.is(tok::hashat)) { 116 int ArgNo = Macro->getArgumentNum(Tokens[i+1].getIdentifierInfo()); 117 assert(ArgNo != -1 && "Token following # is not an argument?"); 118 119 Token Res; 120 if (CurTok.is(tok::hash)) // Stringify 121 Res = ActualArgs->getStringifiedArgument(ArgNo, PP); 122 else { 123 // 'charify': don't bother caching these. 124 Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo), 125 PP, true); 126 } 127 128 // The stringified/charified string leading space flag gets set to match 129 // the #/#@ operator. 130 if (CurTok.hasLeadingSpace() || NextTokGetsSpace) 131 Res.setFlag(Token::LeadingSpace); 132 133 ResultToks.push_back(Res); 134 MadeChange = true; 135 ++i; // Skip arg name. 136 NextTokGetsSpace = false; 137 continue; 138 } 139 140 // Otherwise, if this is not an argument token, just add the token to the 141 // output buffer. 142 IdentifierInfo *II = CurTok.getIdentifierInfo(); 143 int ArgNo = II ? Macro->getArgumentNum(II) : -1; 144 if (ArgNo == -1) { 145 // This isn't an argument, just add it. 146 ResultToks.push_back(CurTok); 147 148 if (NextTokGetsSpace) { 149 ResultToks.back().setFlag(Token::LeadingSpace); 150 NextTokGetsSpace = false; 151 } 152 continue; 153 } 154 155 // An argument is expanded somehow, the result is different than the 156 // input. 157 MadeChange = true; 158 159 // Otherwise, this is a use of the argument. Find out if there is a paste 160 // (##) operator before or after the argument. 161 bool PasteBefore = 162 !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 163 bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash); 164 165 // If it is not the LHS/RHS of a ## operator, we must pre-expand the 166 // argument and substitute the expanded tokens into the result. This is 167 // C99 6.10.3.1p1. 168 if (!PasteBefore && !PasteAfter) { 169 const Token *ResultArgToks; 170 171 // Only preexpand the argument if it could possibly need it. This 172 // avoids some work in common cases. 173 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 174 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 175 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, PP)[0]; 176 else 177 ResultArgToks = ArgTok; // Use non-preexpanded tokens. 178 179 // If the arg token expanded into anything, append it. 180 if (ResultArgToks->isNot(tok::eof)) { 181 unsigned FirstResult = ResultToks.size(); 182 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 183 ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 184 185 // If any tokens were substituted from the argument, the whitespace 186 // before the first token should match the whitespace of the arg 187 // identifier. 188 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 189 CurTok.hasLeadingSpace() || 190 NextTokGetsSpace); 191 NextTokGetsSpace = false; 192 } else { 193 // If this is an empty argument, and if there was whitespace before the 194 // formal token, make sure the next token gets whitespace before it. 195 NextTokGetsSpace = CurTok.hasLeadingSpace(); 196 } 197 continue; 198 } 199 200 // Okay, we have a token that is either the LHS or RHS of a paste (##) 201 // argument. It gets substituted as its non-pre-expanded tokens. 202 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 203 unsigned NumToks = MacroArgs::getArgLength(ArgToks); 204 if (NumToks) { // Not an empty argument? 205 // If this is the GNU ", ## __VA_ARG__" extension, and we just learned 206 // that __VA_ARG__ expands to multiple tokens, avoid a pasting error when 207 // the expander trys to paste ',' with the first token of the __VA_ARG__ 208 // expansion. 209 if (PasteBefore && ResultToks.size() >= 2 && 210 ResultToks[ResultToks.size()-2].is(tok::comma) && 211 (unsigned)ArgNo == Macro->getNumArgs()-1 && 212 Macro->isVariadic()) { 213 // Remove the paste operator, report use of the extension. 214 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 215 ResultToks.pop_back(); 216 } 217 218 ResultToks.append(ArgToks, ArgToks+NumToks); 219 220 // If the next token was supposed to get leading whitespace, ensure it has 221 // it now. 222 if (NextTokGetsSpace) { 223 ResultToks[ResultToks.size()-NumToks].setFlag(Token::LeadingSpace); 224 NextTokGetsSpace = false; 225 } 226 continue; 227 } 228 229 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 230 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 231 // implement this by eating ## operators when a LHS or RHS expands to 232 // empty. 233 NextTokGetsSpace |= CurTok.hasLeadingSpace(); 234 if (PasteAfter) { 235 // Discard the argument token and skip (don't copy to the expansion 236 // buffer) the paste operator after it. 237 NextTokGetsSpace |= Tokens[i+1].hasLeadingSpace(); 238 ++i; 239 continue; 240 } 241 242 // If this is on the RHS of a paste operator, we've already copied the 243 // paste operator to the ResultToks list. Remove it. 244 assert(PasteBefore && ResultToks.back().is(tok::hashhash)); 245 NextTokGetsSpace |= ResultToks.back().hasLeadingSpace(); 246 ResultToks.pop_back(); 247 248 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 249 // and if the macro had at least one real argument, and if the token before 250 // the ## was a comma, remove the comma. 251 if ((unsigned)ArgNo == Macro->getNumArgs()-1 && // is __VA_ARGS__ 252 ActualArgs->isVarargsElidedUse() && // Argument elided. 253 !ResultToks.empty() && ResultToks.back().is(tok::comma)) { 254 // Never add a space, even if the comma, ##, or arg had a space. 255 NextTokGetsSpace = false; 256 // Remove the paste operator, report use of the extension. 257 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 258 ResultToks.pop_back(); 259 } 260 continue; 261 } 262 263 // If anything changed, install this as the new Tokens list. 264 if (MadeChange) { 265 // This is deleted in the dtor. 266 NumTokens = ResultToks.size(); 267 Token *Res = new Token[ResultToks.size()]; 268 if (NumTokens) 269 memcpy(Res, &ResultToks[0], NumTokens*sizeof(Token)); 270 Tokens = Res; 271 OwnsTokens = true; 272 } 273} 274 275/// Lex - Lex and return a token from this macro stream. 276/// 277void TokenLexer::Lex(Token &Tok) { 278 // Lexing off the end of the macro, pop this macro off the expansion stack. 279 if (isAtEnd()) { 280 // If this is a macro (not a token stream), mark the macro enabled now 281 // that it is no longer being expanded. 282 if (Macro) Macro->EnableMacro(); 283 284 // Pop this context off the preprocessors lexer stack and get the next 285 // token. This will delete "this" so remember the PP instance var. 286 Preprocessor &PPCache = PP; 287 if (PP.HandleEndOfTokenLexer(Tok)) 288 return; 289 290 // HandleEndOfTokenLexer may not return a token. If it doesn't, lex 291 // whatever is next. 292 return PPCache.Lex(Tok); 293 } 294 295 // If this is the first token of the expanded result, we inherit spacing 296 // properties later. 297 bool isFirstToken = CurToken == 0; 298 299 // Get the next token to return. 300 Tok = Tokens[CurToken++]; 301 302 // If this token is followed by a token paste (##) operator, paste the tokens! 303 if (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)) 304 if (PasteTokens(Tok)) { 305 // When handling the microsoft /##/ extension, the final token is 306 // returned by PasteTokens, not the pasted token. 307 return; 308 } 309 310 // The token's current location indicate where the token was lexed from. We 311 // need this information to compute the spelling of the token, but any 312 // diagnostics for the expanded token should appear as if they came from 313 // InstantiationLoc. Pull this information together into a new SourceLocation 314 // that captures all of this. 315 if (InstantiateLoc.isValid()) { // Don't do this for token streams. 316 SourceManager &SrcMgr = PP.getSourceManager(); 317 Tok.setLocation(SrcMgr.getInstantiationLoc(Tok.getLocation(), 318 InstantiateLoc)); 319 } 320 321 // If this is the first token, set the lexical properties of the token to 322 // match the lexical properties of the macro identifier. 323 if (isFirstToken) { 324 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 325 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 326 } 327 328 // Handle recursive expansion! 329 if (Tok.getIdentifierInfo() && !DisableMacroExpansion && 330 Tok.getIdentifierInfo()->isHandleIdentifierCase()) 331 PP.HandleIdentifier(Tok); 332 333 // Otherwise, return a normal token. 334} 335 336/// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ## 337/// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 338/// are more ## after it, chomp them iteratively. Return the result as Tok. 339/// If this returns true, the caller should immediately return the token. 340bool TokenLexer::PasteTokens(Token &Tok) { 341 llvm::SmallVector<char, 128> Buffer; 342 do { 343 // Consume the ## operator. 344 SourceLocation PasteOpLoc = Tokens[CurToken].getLocation(); 345 ++CurToken; 346 assert(!isAtEnd() && "No token on the RHS of a paste operator!"); 347 348 // Get the RHS token. 349 const Token &RHS = Tokens[CurToken]; 350 351 bool isInvalid = false; 352 353 // Allocate space for the result token. This is guaranteed to be enough for 354 // the two tokens and a null terminator. 355 Buffer.resize(Tok.getLength() + RHS.getLength() + 1); 356 357 // Get the spelling of the LHS token in Buffer. 358 const char *BufPtr = &Buffer[0]; 359 unsigned LHSLen = PP.getSpelling(Tok, BufPtr); 360 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 361 memcpy(&Buffer[0], BufPtr, LHSLen); 362 363 BufPtr = &Buffer[LHSLen]; 364 unsigned RHSLen = PP.getSpelling(RHS, BufPtr); 365 if (BufPtr != &Buffer[LHSLen]) // Really, we want the chars in Buffer! 366 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 367 368 // Add null terminator. 369 Buffer[LHSLen+RHSLen] = '\0'; 370 371 // Trim excess space. 372 Buffer.resize(LHSLen+RHSLen+1); 373 374 // Plop the pasted result (including the trailing newline and null) into a 375 // scratch buffer where we can lex it. 376 SourceLocation ResultTokLoc = PP.CreateString(&Buffer[0], Buffer.size()); 377 378 // Lex the resultant pasted token into Result. 379 Token Result; 380 381 if (Tok.is(tok::identifier) && RHS.is(tok::identifier)) { 382 // Common paste case: identifier+identifier = identifier. Avoid creating 383 // a lexer and other overhead. 384 PP.IncrementPasteCounter(true); 385 Result.startToken(); 386 Result.setKind(tok::identifier); 387 Result.setLocation(ResultTokLoc); 388 Result.setLength(LHSLen+RHSLen); 389 } else { 390 PP.IncrementPasteCounter(false); 391 392 assert(ResultTokLoc.isFileID() && 393 "Should be a raw location into scratch buffer"); 394 SourceManager &SourceMgr = PP.getSourceManager(); 395 std::pair<FileID, unsigned> LocInfo = 396 SourceMgr.getDecomposedFileLoc(ResultTokLoc); 397 398 const char *ScratchBufStart =SourceMgr.getBufferData(LocInfo.first).first; 399 400 // Make a lexer to lex this string from. Lex just this one token. 401 const char *ResultStrData = ScratchBufStart+LocInfo.second; 402 403 // Make a lexer object so that we lex and expand the paste result. 404 Lexer TL(SourceMgr.getLocForStartOfFile(LocInfo.first), 405 PP.getLangOptions(), 406 ScratchBufStart, 407 ResultStrData, 408 ResultStrData+LHSLen+RHSLen /*don't include null*/); 409 410 // Lex a token in raw mode. This way it won't look up identifiers 411 // automatically, lexing off the end will return an eof token, and 412 // warnings are disabled. This returns true if the result token is the 413 // entire buffer. 414 bool IsComplete = TL.LexFromRawLexer(Result); 415 416 // If we got an EOF token, we didn't form even ONE token. For example, we 417 // did "/ ## /" to get "//". 418 IsComplete &= Result.isNot(tok::eof); 419 isInvalid = !IsComplete; 420 } 421 422 // If pasting the two tokens didn't form a full new token, this is an error. 423 // This occurs with "x ## +" and other stuff. Return with Tok unmodified 424 // and with RHS as the next token to lex. 425 if (isInvalid) { 426 // Test for the Microsoft extension of /##/ turning into // here on the 427 // error path. 428 if (PP.getLangOptions().Microsoft && Tok.is(tok::slash) && 429 RHS.is(tok::slash)) { 430 HandleMicrosoftCommentPaste(Tok); 431 return true; 432 } else { 433 // TODO: If not in assembler language mode. 434 PP.Diag(PasteOpLoc, diag::err_pp_bad_paste) 435 << std::string(Buffer.begin(), Buffer.end()-1); 436 return false; 437 } 438 } 439 440 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 441 // operator. 442 if (Result.is(tok::hashhash)) 443 Result.setKind(tok::unknown); 444 // FIXME: Turn __VA_ARGS__ into "not a token"? 445 446 // Transfer properties of the LHS over the the Result. 447 Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine()); 448 Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace()); 449 450 // Finally, replace LHS with the result, consume the RHS, and iterate. 451 ++CurToken; 452 Tok = Result; 453 } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash)); 454 455 // Now that we got the result token, it will be subject to expansion. Since 456 // token pasting re-lexes the result token in raw mode, identifier information 457 // isn't looked up. As such, if the result is an identifier, look up id info. 458 if (Tok.is(tok::identifier)) { 459 // Look up the identifier info for the token. We disabled identifier lookup 460 // by saying we're skipping contents, so we need to do this manually. 461 Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok)); 462 } 463 return false; 464} 465 466/// isNextTokenLParen - If the next token lexed will pop this macro off the 467/// expansion stack, return 2. If the next unexpanded token is a '(', return 468/// 1, otherwise return 0. 469unsigned TokenLexer::isNextTokenLParen() const { 470 // Out of tokens? 471 if (isAtEnd()) 472 return 2; 473 return Tokens[CurToken].is(tok::l_paren); 474} 475 476 477/// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 478/// together to form a comment that comments out everything in the current 479/// macro, other active macros, and anything left on the current physical 480/// source line of the instantiated buffer. Handle this by returning the 481/// first token on the next line. 482void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok) { 483 // We 'comment out' the rest of this macro by just ignoring the rest of the 484 // tokens that have not been lexed yet, if any. 485 486 // Since this must be a macro, mark the macro enabled now that it is no longer 487 // being expanded. 488 assert(Macro && "Token streams can't paste comments"); 489 Macro->EnableMacro(); 490 491 PP.HandleMicrosoftCommentPaste(Tok); 492} 493