1 //===--- MacroExpansion.cpp - Top level Macro Expansion -------------------===// 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 top level handling of macro expansion for the 11 // preprocessor. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "clang/Lex/Preprocessor.h" 16 #include "clang/Basic/Attributes.h" 17 #include "clang/Basic/FileManager.h" 18 #include "clang/Basic/SourceManager.h" 19 #include "clang/Basic/TargetInfo.h" 20 #include "clang/Lex/CodeCompletionHandler.h" 21 #include "clang/Lex/ExternalPreprocessorSource.h" 22 #include "clang/Lex/LexDiagnostic.h" 23 #include "clang/Lex/MacroArgs.h" 24 #include "clang/Lex/MacroInfo.h" 25 #include "llvm/ADT/STLExtras.h" 26 #include "llvm/ADT/SmallString.h" 27 #include "llvm/ADT/StringSwitch.h" 28 #include "llvm/Config/llvm-config.h" 29 #include "llvm/Support/ErrorHandling.h" 30 #include "llvm/Support/Format.h" 31 #include "llvm/Support/raw_ostream.h" 32 #include <cstdio> 33 #include <ctime> 34 using namespace clang; 35 36 MacroDirective * 37 Preprocessor::getMacroDirectiveHistory(const IdentifierInfo *II) const { 38 assert(II->hadMacroDefinition() && "Identifier has not been not a macro!"); 39 40 macro_iterator Pos = Macros.find(II); 41 assert(Pos != Macros.end() && "Identifier macro info is missing!"); 42 return Pos->second; 43 } 44 45 void Preprocessor::appendMacroDirective(IdentifierInfo *II, MacroDirective *MD){ 46 assert(MD && "MacroDirective should be non-zero!"); 47 assert(!MD->getPrevious() && "Already attached to a MacroDirective history."); 48 49 MacroDirective *&StoredMD = Macros[II]; 50 MD->setPrevious(StoredMD); 51 StoredMD = MD; 52 II->setHasMacroDefinition(MD->isDefined()); 53 bool isImportedMacro = isa<DefMacroDirective>(MD) && 54 cast<DefMacroDirective>(MD)->isImported(); 55 if (II->isFromAST() && !isImportedMacro) 56 II->setChangedSinceDeserialization(); 57 } 58 59 void Preprocessor::setLoadedMacroDirective(IdentifierInfo *II, 60 MacroDirective *MD) { 61 assert(II && MD); 62 MacroDirective *&StoredMD = Macros[II]; 63 assert(!StoredMD && 64 "the macro history was modified before initializing it from a pch"); 65 StoredMD = MD; 66 // Setup the identifier as having associated macro history. 67 II->setHasMacroDefinition(true); 68 if (!MD->isDefined()) 69 II->setHasMacroDefinition(false); 70 } 71 72 /// RegisterBuiltinMacro - Register the specified identifier in the identifier 73 /// table and mark it as a builtin macro to be expanded. 74 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){ 75 // Get the identifier. 76 IdentifierInfo *Id = PP.getIdentifierInfo(Name); 77 78 // Mark it as being a macro that is builtin. 79 MacroInfo *MI = PP.AllocateMacroInfo(SourceLocation()); 80 MI->setIsBuiltinMacro(); 81 PP.appendDefMacroDirective(Id, MI); 82 return Id; 83 } 84 85 86 /// RegisterBuiltinMacros - Register builtin macros, such as __LINE__ with the 87 /// identifier table. 88 void Preprocessor::RegisterBuiltinMacros() { 89 Ident__LINE__ = RegisterBuiltinMacro(*this, "__LINE__"); 90 Ident__FILE__ = RegisterBuiltinMacro(*this, "__FILE__"); 91 Ident__DATE__ = RegisterBuiltinMacro(*this, "__DATE__"); 92 Ident__TIME__ = RegisterBuiltinMacro(*this, "__TIME__"); 93 Ident__COUNTER__ = RegisterBuiltinMacro(*this, "__COUNTER__"); 94 Ident_Pragma = RegisterBuiltinMacro(*this, "_Pragma"); 95 96 // GCC Extensions. 97 Ident__BASE_FILE__ = RegisterBuiltinMacro(*this, "__BASE_FILE__"); 98 Ident__INCLUDE_LEVEL__ = RegisterBuiltinMacro(*this, "__INCLUDE_LEVEL__"); 99 Ident__TIMESTAMP__ = RegisterBuiltinMacro(*this, "__TIMESTAMP__"); 100 101 // Microsoft Extensions. 102 if (LangOpts.MicrosoftExt) { 103 Ident__identifier = RegisterBuiltinMacro(*this, "__identifier"); 104 Ident__pragma = RegisterBuiltinMacro(*this, "__pragma"); 105 } else { 106 Ident__identifier = 0; 107 Ident__pragma = 0; 108 } 109 110 // Clang Extensions. 111 Ident__has_feature = RegisterBuiltinMacro(*this, "__has_feature"); 112 Ident__has_extension = RegisterBuiltinMacro(*this, "__has_extension"); 113 Ident__has_builtin = RegisterBuiltinMacro(*this, "__has_builtin"); 114 Ident__has_attribute = RegisterBuiltinMacro(*this, "__has_attribute"); 115 Ident__has_include = RegisterBuiltinMacro(*this, "__has_include"); 116 Ident__has_include_next = RegisterBuiltinMacro(*this, "__has_include_next"); 117 Ident__has_warning = RegisterBuiltinMacro(*this, "__has_warning"); 118 Ident__is_identifier = RegisterBuiltinMacro(*this, "__is_identifier"); 119 120 // Modules. 121 if (LangOpts.Modules) { 122 Ident__building_module = RegisterBuiltinMacro(*this, "__building_module"); 123 124 // __MODULE__ 125 if (!LangOpts.CurrentModule.empty()) 126 Ident__MODULE__ = RegisterBuiltinMacro(*this, "__MODULE__"); 127 else 128 Ident__MODULE__ = 0; 129 } else { 130 Ident__building_module = 0; 131 Ident__MODULE__ = 0; 132 } 133 } 134 135 /// isTrivialSingleTokenExpansion - Return true if MI, which has a single token 136 /// in its expansion, currently expands to that token literally. 137 static bool isTrivialSingleTokenExpansion(const MacroInfo *MI, 138 const IdentifierInfo *MacroIdent, 139 Preprocessor &PP) { 140 IdentifierInfo *II = MI->getReplacementToken(0).getIdentifierInfo(); 141 142 // If the token isn't an identifier, it's always literally expanded. 143 if (II == 0) return true; 144 145 // If the information about this identifier is out of date, update it from 146 // the external source. 147 if (II->isOutOfDate()) 148 PP.getExternalSource()->updateOutOfDateIdentifier(*II); 149 150 // If the identifier is a macro, and if that macro is enabled, it may be 151 // expanded so it's not a trivial expansion. 152 if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled() && 153 // Fast expanding "#define X X" is ok, because X would be disabled. 154 II != MacroIdent) 155 return false; 156 157 // If this is an object-like macro invocation, it is safe to trivially expand 158 // it. 159 if (MI->isObjectLike()) return true; 160 161 // If this is a function-like macro invocation, it's safe to trivially expand 162 // as long as the identifier is not a macro argument. 163 for (MacroInfo::arg_iterator I = MI->arg_begin(), E = MI->arg_end(); 164 I != E; ++I) 165 if (*I == II) 166 return false; // Identifier is a macro argument. 167 168 return true; 169 } 170 171 172 /// isNextPPTokenLParen - Determine whether the next preprocessor token to be 173 /// lexed is a '('. If so, consume the token and return true, if not, this 174 /// method should have no observable side-effect on the lexed tokens. 175 bool Preprocessor::isNextPPTokenLParen() { 176 // Do some quick tests for rejection cases. 177 unsigned Val; 178 if (CurLexer) 179 Val = CurLexer->isNextPPTokenLParen(); 180 else if (CurPTHLexer) 181 Val = CurPTHLexer->isNextPPTokenLParen(); 182 else 183 Val = CurTokenLexer->isNextTokenLParen(); 184 185 if (Val == 2) { 186 // We have run off the end. If it's a source file we don't 187 // examine enclosing ones (C99 5.1.1.2p4). Otherwise walk up the 188 // macro stack. 189 if (CurPPLexer) 190 return false; 191 for (unsigned i = IncludeMacroStack.size(); i != 0; --i) { 192 IncludeStackInfo &Entry = IncludeMacroStack[i-1]; 193 if (Entry.TheLexer) 194 Val = Entry.TheLexer->isNextPPTokenLParen(); 195 else if (Entry.ThePTHLexer) 196 Val = Entry.ThePTHLexer->isNextPPTokenLParen(); 197 else 198 Val = Entry.TheTokenLexer->isNextTokenLParen(); 199 200 if (Val != 2) 201 break; 202 203 // Ran off the end of a source file? 204 if (Entry.ThePPLexer) 205 return false; 206 } 207 } 208 209 // Okay, if we know that the token is a '(', lex it and return. Otherwise we 210 // have found something that isn't a '(' or we found the end of the 211 // translation unit. In either case, return false. 212 return Val == 1; 213 } 214 215 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be 216 /// expanded as a macro, handle it and return the next token as 'Identifier'. 217 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier, 218 MacroDirective *MD) { 219 MacroDirective::DefInfo Def = MD->getDefinition(); 220 assert(Def.isValid()); 221 MacroInfo *MI = Def.getMacroInfo(); 222 223 // If this is a macro expansion in the "#if !defined(x)" line for the file, 224 // then the macro could expand to different things in other contexts, we need 225 // to disable the optimization in this case. 226 if (CurPPLexer) CurPPLexer->MIOpt.ExpandedMacro(); 227 228 // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially. 229 if (MI->isBuiltinMacro()) { 230 if (Callbacks) Callbacks->MacroExpands(Identifier, MD, 231 Identifier.getLocation(),/*Args=*/0); 232 ExpandBuiltinMacro(Identifier); 233 return true; 234 } 235 236 /// Args - If this is a function-like macro expansion, this contains, 237 /// for each macro argument, the list of tokens that were provided to the 238 /// invocation. 239 MacroArgs *Args = 0; 240 241 // Remember where the end of the expansion occurred. For an object-like 242 // macro, this is the identifier. For a function-like macro, this is the ')'. 243 SourceLocation ExpansionEnd = Identifier.getLocation(); 244 245 // If this is a function-like macro, read the arguments. 246 if (MI->isFunctionLike()) { 247 // Remember that we are now parsing the arguments to a macro invocation. 248 // Preprocessor directives used inside macro arguments are not portable, and 249 // this enables the warning. 250 InMacroArgs = true; 251 Args = ReadFunctionLikeMacroArgs(Identifier, MI, ExpansionEnd); 252 253 // Finished parsing args. 254 InMacroArgs = false; 255 256 // If there was an error parsing the arguments, bail out. 257 if (Args == 0) return true; 258 259 ++NumFnMacroExpanded; 260 } else { 261 ++NumMacroExpanded; 262 } 263 264 // Notice that this macro has been used. 265 markMacroAsUsed(MI); 266 267 // Remember where the token is expanded. 268 SourceLocation ExpandLoc = Identifier.getLocation(); 269 SourceRange ExpansionRange(ExpandLoc, ExpansionEnd); 270 271 if (Callbacks) { 272 if (InMacroArgs) { 273 // We can have macro expansion inside a conditional directive while 274 // reading the function macro arguments. To ensure, in that case, that 275 // MacroExpands callbacks still happen in source order, queue this 276 // callback to have it happen after the function macro callback. 277 DelayedMacroExpandsCallbacks.push_back( 278 MacroExpandsInfo(Identifier, MD, ExpansionRange)); 279 } else { 280 Callbacks->MacroExpands(Identifier, MD, ExpansionRange, Args); 281 if (!DelayedMacroExpandsCallbacks.empty()) { 282 for (unsigned i=0, e = DelayedMacroExpandsCallbacks.size(); i!=e; ++i) { 283 MacroExpandsInfo &Info = DelayedMacroExpandsCallbacks[i]; 284 // FIXME: We lose macro args info with delayed callback. 285 Callbacks->MacroExpands(Info.Tok, Info.MD, Info.Range, /*Args=*/0); 286 } 287 DelayedMacroExpandsCallbacks.clear(); 288 } 289 } 290 } 291 292 // If the macro definition is ambiguous, complain. 293 if (Def.getDirective()->isAmbiguous()) { 294 Diag(Identifier, diag::warn_pp_ambiguous_macro) 295 << Identifier.getIdentifierInfo(); 296 Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen) 297 << Identifier.getIdentifierInfo(); 298 for (MacroDirective::DefInfo PrevDef = Def.getPreviousDefinition(); 299 PrevDef && !PrevDef.isUndefined(); 300 PrevDef = PrevDef.getPreviousDefinition()) { 301 Diag(PrevDef.getMacroInfo()->getDefinitionLoc(), 302 diag::note_pp_ambiguous_macro_other) 303 << Identifier.getIdentifierInfo(); 304 if (!PrevDef.getDirective()->isAmbiguous()) 305 break; 306 } 307 } 308 309 // If we started lexing a macro, enter the macro expansion body. 310 311 // If this macro expands to no tokens, don't bother to push it onto the 312 // expansion stack, only to take it right back off. 313 if (MI->getNumTokens() == 0) { 314 // No need for arg info. 315 if (Args) Args->destroy(*this); 316 317 // Propagate whitespace info as if we had pushed, then popped, 318 // a macro context. 319 Identifier.setFlag(Token::LeadingEmptyMacro); 320 PropagateLineStartLeadingSpaceInfo(Identifier); 321 ++NumFastMacroExpanded; 322 return false; 323 } else if (MI->getNumTokens() == 1 && 324 isTrivialSingleTokenExpansion(MI, Identifier.getIdentifierInfo(), 325 *this)) { 326 // Otherwise, if this macro expands into a single trivially-expanded 327 // token: expand it now. This handles common cases like 328 // "#define VAL 42". 329 330 // No need for arg info. 331 if (Args) Args->destroy(*this); 332 333 // Propagate the isAtStartOfLine/hasLeadingSpace markers of the macro 334 // identifier to the expanded token. 335 bool isAtStartOfLine = Identifier.isAtStartOfLine(); 336 bool hasLeadingSpace = Identifier.hasLeadingSpace(); 337 338 // Replace the result token. 339 Identifier = MI->getReplacementToken(0); 340 341 // Restore the StartOfLine/LeadingSpace markers. 342 Identifier.setFlagValue(Token::StartOfLine , isAtStartOfLine); 343 Identifier.setFlagValue(Token::LeadingSpace, hasLeadingSpace); 344 345 // Update the tokens location to include both its expansion and physical 346 // locations. 347 SourceLocation Loc = 348 SourceMgr.createExpansionLoc(Identifier.getLocation(), ExpandLoc, 349 ExpansionEnd,Identifier.getLength()); 350 Identifier.setLocation(Loc); 351 352 // If this is a disabled macro or #define X X, we must mark the result as 353 // unexpandable. 354 if (IdentifierInfo *NewII = Identifier.getIdentifierInfo()) { 355 if (MacroInfo *NewMI = getMacroInfo(NewII)) 356 if (!NewMI->isEnabled() || NewMI == MI) { 357 Identifier.setFlag(Token::DisableExpand); 358 // Don't warn for "#define X X" like "#define bool bool" from 359 // stdbool.h. 360 if (NewMI != MI || MI->isFunctionLike()) 361 Diag(Identifier, diag::pp_disabled_macro_expansion); 362 } 363 } 364 365 // Since this is not an identifier token, it can't be macro expanded, so 366 // we're done. 367 ++NumFastMacroExpanded; 368 return true; 369 } 370 371 // Start expanding the macro. 372 EnterMacro(Identifier, ExpansionEnd, MI, Args); 373 return false; 374 } 375 376 enum Bracket { 377 Brace, 378 Paren 379 }; 380 381 /// CheckMatchedBrackets - Returns true if the braces and parentheses in the 382 /// token vector are properly nested. 383 static bool CheckMatchedBrackets(const SmallVectorImpl<Token> &Tokens) { 384 SmallVector<Bracket, 8> Brackets; 385 for (SmallVectorImpl<Token>::const_iterator I = Tokens.begin(), 386 E = Tokens.end(); 387 I != E; ++I) { 388 if (I->is(tok::l_paren)) { 389 Brackets.push_back(Paren); 390 } else if (I->is(tok::r_paren)) { 391 if (Brackets.empty() || Brackets.back() == Brace) 392 return false; 393 Brackets.pop_back(); 394 } else if (I->is(tok::l_brace)) { 395 Brackets.push_back(Brace); 396 } else if (I->is(tok::r_brace)) { 397 if (Brackets.empty() || Brackets.back() == Paren) 398 return false; 399 Brackets.pop_back(); 400 } 401 } 402 if (!Brackets.empty()) 403 return false; 404 return true; 405 } 406 407 /// GenerateNewArgTokens - Returns true if OldTokens can be converted to a new 408 /// vector of tokens in NewTokens. The new number of arguments will be placed 409 /// in NumArgs and the ranges which need to surrounded in parentheses will be 410 /// in ParenHints. 411 /// Returns false if the token stream cannot be changed. If this is because 412 /// of an initializer list starting a macro argument, the range of those 413 /// initializer lists will be place in InitLists. 414 static bool GenerateNewArgTokens(Preprocessor &PP, 415 SmallVectorImpl<Token> &OldTokens, 416 SmallVectorImpl<Token> &NewTokens, 417 unsigned &NumArgs, 418 SmallVectorImpl<SourceRange> &ParenHints, 419 SmallVectorImpl<SourceRange> &InitLists) { 420 if (!CheckMatchedBrackets(OldTokens)) 421 return false; 422 423 // Once it is known that the brackets are matched, only a simple count of the 424 // braces is needed. 425 unsigned Braces = 0; 426 427 // First token of a new macro argument. 428 SmallVectorImpl<Token>::iterator ArgStartIterator = OldTokens.begin(); 429 430 // First closing brace in a new macro argument. Used to generate 431 // SourceRanges for InitLists. 432 SmallVectorImpl<Token>::iterator ClosingBrace = OldTokens.end(); 433 NumArgs = 0; 434 Token TempToken; 435 // Set to true when a macro separator token is found inside a braced list. 436 // If true, the fixed argument spans multiple old arguments and ParenHints 437 // will be updated. 438 bool FoundSeparatorToken = false; 439 for (SmallVectorImpl<Token>::iterator I = OldTokens.begin(), 440 E = OldTokens.end(); 441 I != E; ++I) { 442 if (I->is(tok::l_brace)) { 443 ++Braces; 444 } else if (I->is(tok::r_brace)) { 445 --Braces; 446 if (Braces == 0 && ClosingBrace == E && FoundSeparatorToken) 447 ClosingBrace = I; 448 } else if (I->is(tok::eof)) { 449 // EOF token is used to separate macro arguments 450 if (Braces != 0) { 451 // Assume comma separator is actually braced list separator and change 452 // it back to a comma. 453 FoundSeparatorToken = true; 454 I->setKind(tok::comma); 455 I->setLength(1); 456 } else { // Braces == 0 457 // Separator token still separates arguments. 458 ++NumArgs; 459 460 // If the argument starts with a brace, it can't be fixed with 461 // parentheses. A different diagnostic will be given. 462 if (FoundSeparatorToken && ArgStartIterator->is(tok::l_brace)) { 463 InitLists.push_back( 464 SourceRange(ArgStartIterator->getLocation(), 465 PP.getLocForEndOfToken(ClosingBrace->getLocation()))); 466 ClosingBrace = E; 467 } 468 469 // Add left paren 470 if (FoundSeparatorToken) { 471 TempToken.startToken(); 472 TempToken.setKind(tok::l_paren); 473 TempToken.setLocation(ArgStartIterator->getLocation()); 474 TempToken.setLength(0); 475 NewTokens.push_back(TempToken); 476 } 477 478 // Copy over argument tokens 479 NewTokens.insert(NewTokens.end(), ArgStartIterator, I); 480 481 // Add right paren and store the paren locations in ParenHints 482 if (FoundSeparatorToken) { 483 SourceLocation Loc = PP.getLocForEndOfToken((I - 1)->getLocation()); 484 TempToken.startToken(); 485 TempToken.setKind(tok::r_paren); 486 TempToken.setLocation(Loc); 487 TempToken.setLength(0); 488 NewTokens.push_back(TempToken); 489 ParenHints.push_back(SourceRange(ArgStartIterator->getLocation(), 490 Loc)); 491 } 492 493 // Copy separator token 494 NewTokens.push_back(*I); 495 496 // Reset values 497 ArgStartIterator = I + 1; 498 FoundSeparatorToken = false; 499 } 500 } 501 } 502 503 return !ParenHints.empty() && InitLists.empty(); 504 } 505 506 /// ReadFunctionLikeMacroArgs - After reading "MACRO" and knowing that the next 507 /// token is the '(' of the macro, this method is invoked to read all of the 508 /// actual arguments specified for the macro invocation. This returns null on 509 /// error. 510 MacroArgs *Preprocessor::ReadFunctionLikeMacroArgs(Token &MacroName, 511 MacroInfo *MI, 512 SourceLocation &MacroEnd) { 513 // The number of fixed arguments to parse. 514 unsigned NumFixedArgsLeft = MI->getNumArgs(); 515 bool isVariadic = MI->isVariadic(); 516 517 // Outer loop, while there are more arguments, keep reading them. 518 Token Tok; 519 520 // Read arguments as unexpanded tokens. This avoids issues, e.g., where 521 // an argument value in a macro could expand to ',' or '(' or ')'. 522 LexUnexpandedToken(Tok); 523 assert(Tok.is(tok::l_paren) && "Error computing l-paren-ness?"); 524 525 // ArgTokens - Build up a list of tokens that make up each argument. Each 526 // argument is separated by an EOF token. Use a SmallVector so we can avoid 527 // heap allocations in the common case. 528 SmallVector<Token, 64> ArgTokens; 529 bool ContainsCodeCompletionTok = false; 530 531 SourceLocation TooManyArgsLoc; 532 533 unsigned NumActuals = 0; 534 while (Tok.isNot(tok::r_paren)) { 535 if (ContainsCodeCompletionTok && (Tok.is(tok::eof) || Tok.is(tok::eod))) 536 break; 537 538 assert((Tok.is(tok::l_paren) || Tok.is(tok::comma)) && 539 "only expect argument separators here"); 540 541 unsigned ArgTokenStart = ArgTokens.size(); 542 SourceLocation ArgStartLoc = Tok.getLocation(); 543 544 // C99 6.10.3p11: Keep track of the number of l_parens we have seen. Note 545 // that we already consumed the first one. 546 unsigned NumParens = 0; 547 548 while (1) { 549 // Read arguments as unexpanded tokens. This avoids issues, e.g., where 550 // an argument value in a macro could expand to ',' or '(' or ')'. 551 LexUnexpandedToken(Tok); 552 553 if (Tok.is(tok::eof) || Tok.is(tok::eod)) { // "#if f(<eof>" & "#if f(\n" 554 if (!ContainsCodeCompletionTok) { 555 Diag(MacroName, diag::err_unterm_macro_invoc); 556 Diag(MI->getDefinitionLoc(), diag::note_macro_here) 557 << MacroName.getIdentifierInfo(); 558 // Do not lose the EOF/EOD. Return it to the client. 559 MacroName = Tok; 560 return 0; 561 } else { 562 // Do not lose the EOF/EOD. 563 Token *Toks = new Token[1]; 564 Toks[0] = Tok; 565 EnterTokenStream(Toks, 1, true, true); 566 break; 567 } 568 } else if (Tok.is(tok::r_paren)) { 569 // If we found the ) token, the macro arg list is done. 570 if (NumParens-- == 0) { 571 MacroEnd = Tok.getLocation(); 572 break; 573 } 574 } else if (Tok.is(tok::l_paren)) { 575 ++NumParens; 576 } else if (Tok.is(tok::comma) && NumParens == 0 && 577 !(Tok.getFlags() & Token::IgnoredComma)) { 578 // In Microsoft-compatibility mode, single commas from nested macro 579 // expansions should not be considered as argument separators. We test 580 // for this with the IgnoredComma token flag above. 581 582 // Comma ends this argument if there are more fixed arguments expected. 583 // However, if this is a variadic macro, and this is part of the 584 // variadic part, then the comma is just an argument token. 585 if (!isVariadic) break; 586 if (NumFixedArgsLeft > 1) 587 break; 588 } else if (Tok.is(tok::comment) && !KeepMacroComments) { 589 // If this is a comment token in the argument list and we're just in 590 // -C mode (not -CC mode), discard the comment. 591 continue; 592 } else if (Tok.getIdentifierInfo() != 0) { 593 // Reading macro arguments can cause macros that we are currently 594 // expanding from to be popped off the expansion stack. Doing so causes 595 // them to be reenabled for expansion. Here we record whether any 596 // identifiers we lex as macro arguments correspond to disabled macros. 597 // If so, we mark the token as noexpand. This is a subtle aspect of 598 // C99 6.10.3.4p2. 599 if (MacroInfo *MI = getMacroInfo(Tok.getIdentifierInfo())) 600 if (!MI->isEnabled()) 601 Tok.setFlag(Token::DisableExpand); 602 } else if (Tok.is(tok::code_completion)) { 603 ContainsCodeCompletionTok = true; 604 if (CodeComplete) 605 CodeComplete->CodeCompleteMacroArgument(MacroName.getIdentifierInfo(), 606 MI, NumActuals); 607 // Don't mark that we reached the code-completion point because the 608 // parser is going to handle the token and there will be another 609 // code-completion callback. 610 } 611 612 ArgTokens.push_back(Tok); 613 } 614 615 // If this was an empty argument list foo(), don't add this as an empty 616 // argument. 617 if (ArgTokens.empty() && Tok.getKind() == tok::r_paren) 618 break; 619 620 // If this is not a variadic macro, and too many args were specified, emit 621 // an error. 622 if (!isVariadic && NumFixedArgsLeft == 0 && TooManyArgsLoc.isInvalid()) { 623 if (ArgTokens.size() != ArgTokenStart) 624 TooManyArgsLoc = ArgTokens[ArgTokenStart].getLocation(); 625 else 626 TooManyArgsLoc = ArgStartLoc; 627 } 628 629 // Empty arguments are standard in C99 and C++0x, and are supported as an 630 // extension in other modes. 631 if (ArgTokens.size() == ArgTokenStart && !LangOpts.C99) 632 Diag(Tok, LangOpts.CPlusPlus11 ? 633 diag::warn_cxx98_compat_empty_fnmacro_arg : 634 diag::ext_empty_fnmacro_arg); 635 636 // Add a marker EOF token to the end of the token list for this argument. 637 Token EOFTok; 638 EOFTok.startToken(); 639 EOFTok.setKind(tok::eof); 640 EOFTok.setLocation(Tok.getLocation()); 641 EOFTok.setLength(0); 642 ArgTokens.push_back(EOFTok); 643 ++NumActuals; 644 if (!ContainsCodeCompletionTok && NumFixedArgsLeft != 0) 645 --NumFixedArgsLeft; 646 } 647 648 // Okay, we either found the r_paren. Check to see if we parsed too few 649 // arguments. 650 unsigned MinArgsExpected = MI->getNumArgs(); 651 652 // If this is not a variadic macro, and too many args were specified, emit 653 // an error. 654 if (!isVariadic && NumActuals > MinArgsExpected && 655 !ContainsCodeCompletionTok) { 656 // Emit the diagnostic at the macro name in case there is a missing ). 657 // Emitting it at the , could be far away from the macro name. 658 Diag(TooManyArgsLoc, diag::err_too_many_args_in_macro_invoc); 659 Diag(MI->getDefinitionLoc(), diag::note_macro_here) 660 << MacroName.getIdentifierInfo(); 661 662 // Commas from braced initializer lists will be treated as argument 663 // separators inside macros. Attempt to correct for this with parentheses. 664 // TODO: See if this can be generalized to angle brackets for templates 665 // inside macro arguments. 666 667 SmallVector<Token, 4> FixedArgTokens; 668 unsigned FixedNumArgs = 0; 669 SmallVector<SourceRange, 4> ParenHints, InitLists; 670 if (!GenerateNewArgTokens(*this, ArgTokens, FixedArgTokens, FixedNumArgs, 671 ParenHints, InitLists)) { 672 if (!InitLists.empty()) { 673 DiagnosticBuilder DB = 674 Diag(MacroName, 675 diag::note_init_list_at_beginning_of_macro_argument); 676 for (SmallVector<SourceRange, 4>::iterator 677 Range = InitLists.begin(), RangeEnd = InitLists.end(); 678 Range != RangeEnd; ++Range) { 679 if (DB.hasMaxRanges()) 680 break; 681 DB << *Range; 682 } 683 } 684 return 0; 685 } 686 if (FixedNumArgs != MinArgsExpected) 687 return 0; 688 689 DiagnosticBuilder DB = Diag(MacroName, diag::note_suggest_parens_for_macro); 690 for (SmallVector<SourceRange, 4>::iterator 691 ParenLocation = ParenHints.begin(), ParenEnd = ParenHints.end(); 692 ParenLocation != ParenEnd; ++ParenLocation) { 693 if (DB.hasMaxFixItHints()) 694 break; 695 DB << FixItHint::CreateInsertion(ParenLocation->getBegin(), "("); 696 if (DB.hasMaxFixItHints()) 697 break; 698 DB << FixItHint::CreateInsertion(ParenLocation->getEnd(), ")"); 699 } 700 ArgTokens.swap(FixedArgTokens); 701 NumActuals = FixedNumArgs; 702 } 703 704 // See MacroArgs instance var for description of this. 705 bool isVarargsElided = false; 706 707 if (ContainsCodeCompletionTok) { 708 // Recover from not-fully-formed macro invocation during code-completion. 709 Token EOFTok; 710 EOFTok.startToken(); 711 EOFTok.setKind(tok::eof); 712 EOFTok.setLocation(Tok.getLocation()); 713 EOFTok.setLength(0); 714 for (; NumActuals < MinArgsExpected; ++NumActuals) 715 ArgTokens.push_back(EOFTok); 716 } 717 718 if (NumActuals < MinArgsExpected) { 719 // There are several cases where too few arguments is ok, handle them now. 720 if (NumActuals == 0 && MinArgsExpected == 1) { 721 // #define A(X) or #define A(...) ---> A() 722 723 // If there is exactly one argument, and that argument is missing, 724 // then we have an empty "()" argument empty list. This is fine, even if 725 // the macro expects one argument (the argument is just empty). 726 isVarargsElided = MI->isVariadic(); 727 } else if (MI->isVariadic() && 728 (NumActuals+1 == MinArgsExpected || // A(x, ...) -> A(X) 729 (NumActuals == 0 && MinArgsExpected == 2))) {// A(x,...) -> A() 730 // Varargs where the named vararg parameter is missing: OK as extension. 731 // #define A(x, ...) 732 // A("blah") 733 // 734 // If the macro contains the comma pasting extension, the diagnostic 735 // is suppressed; we know we'll get another diagnostic later. 736 if (!MI->hasCommaPasting()) { 737 Diag(Tok, diag::ext_missing_varargs_arg); 738 Diag(MI->getDefinitionLoc(), diag::note_macro_here) 739 << MacroName.getIdentifierInfo(); 740 } 741 742 // Remember this occurred, allowing us to elide the comma when used for 743 // cases like: 744 // #define A(x, foo...) blah(a, ## foo) 745 // #define B(x, ...) blah(a, ## __VA_ARGS__) 746 // #define C(...) blah(a, ## __VA_ARGS__) 747 // A(x) B(x) C() 748 isVarargsElided = true; 749 } else if (!ContainsCodeCompletionTok) { 750 // Otherwise, emit the error. 751 Diag(Tok, diag::err_too_few_args_in_macro_invoc); 752 Diag(MI->getDefinitionLoc(), diag::note_macro_here) 753 << MacroName.getIdentifierInfo(); 754 return 0; 755 } 756 757 // Add a marker EOF token to the end of the token list for this argument. 758 SourceLocation EndLoc = Tok.getLocation(); 759 Tok.startToken(); 760 Tok.setKind(tok::eof); 761 Tok.setLocation(EndLoc); 762 Tok.setLength(0); 763 ArgTokens.push_back(Tok); 764 765 // If we expect two arguments, add both as empty. 766 if (NumActuals == 0 && MinArgsExpected == 2) 767 ArgTokens.push_back(Tok); 768 769 } else if (NumActuals > MinArgsExpected && !MI->isVariadic() && 770 !ContainsCodeCompletionTok) { 771 // Emit the diagnostic at the macro name in case there is a missing ). 772 // Emitting it at the , could be far away from the macro name. 773 Diag(MacroName, diag::err_too_many_args_in_macro_invoc); 774 Diag(MI->getDefinitionLoc(), diag::note_macro_here) 775 << MacroName.getIdentifierInfo(); 776 return 0; 777 } 778 779 return MacroArgs::create(MI, ArgTokens, isVarargsElided, *this); 780 } 781 782 /// \brief Keeps macro expanded tokens for TokenLexers. 783 // 784 /// Works like a stack; a TokenLexer adds the macro expanded tokens that is 785 /// going to lex in the cache and when it finishes the tokens are removed 786 /// from the end of the cache. 787 Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer, 788 ArrayRef<Token> tokens) { 789 assert(tokLexer); 790 if (tokens.empty()) 791 return 0; 792 793 size_t newIndex = MacroExpandedTokens.size(); 794 bool cacheNeedsToGrow = tokens.size() > 795 MacroExpandedTokens.capacity()-MacroExpandedTokens.size(); 796 MacroExpandedTokens.append(tokens.begin(), tokens.end()); 797 798 if (cacheNeedsToGrow) { 799 // Go through all the TokenLexers whose 'Tokens' pointer points in the 800 // buffer and update the pointers to the (potential) new buffer array. 801 for (unsigned i = 0, e = MacroExpandingLexersStack.size(); i != e; ++i) { 802 TokenLexer *prevLexer; 803 size_t tokIndex; 804 std::tie(prevLexer, tokIndex) = MacroExpandingLexersStack[i]; 805 prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex; 806 } 807 } 808 809 MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex)); 810 return MacroExpandedTokens.data() + newIndex; 811 } 812 813 void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() { 814 assert(!MacroExpandingLexersStack.empty()); 815 size_t tokIndex = MacroExpandingLexersStack.back().second; 816 assert(tokIndex < MacroExpandedTokens.size()); 817 // Pop the cached macro expanded tokens from the end. 818 MacroExpandedTokens.resize(tokIndex); 819 MacroExpandingLexersStack.pop_back(); 820 } 821 822 /// ComputeDATE_TIME - Compute the current time, enter it into the specified 823 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of 824 /// the identifier tokens inserted. 825 static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc, 826 Preprocessor &PP) { 827 time_t TT = time(0); 828 struct tm *TM = localtime(&TT); 829 830 static const char * const Months[] = { 831 "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec" 832 }; 833 834 { 835 SmallString<32> TmpBuffer; 836 llvm::raw_svector_ostream TmpStream(TmpBuffer); 837 TmpStream << llvm::format("\"%s %2d %4d\"", Months[TM->tm_mon], 838 TM->tm_mday, TM->tm_year + 1900); 839 Token TmpTok; 840 TmpTok.startToken(); 841 PP.CreateString(TmpStream.str(), TmpTok); 842 DATELoc = TmpTok.getLocation(); 843 } 844 845 { 846 SmallString<32> TmpBuffer; 847 llvm::raw_svector_ostream TmpStream(TmpBuffer); 848 TmpStream << llvm::format("\"%02d:%02d:%02d\"", 849 TM->tm_hour, TM->tm_min, TM->tm_sec); 850 Token TmpTok; 851 TmpTok.startToken(); 852 PP.CreateString(TmpStream.str(), TmpTok); 853 TIMELoc = TmpTok.getLocation(); 854 } 855 } 856 857 858 /// HasFeature - Return true if we recognize and implement the feature 859 /// specified by the identifier as a standard language feature. 860 static bool HasFeature(const Preprocessor &PP, const IdentifierInfo *II) { 861 const LangOptions &LangOpts = PP.getLangOpts(); 862 StringRef Feature = II->getName(); 863 864 // Normalize the feature name, __foo__ becomes foo. 865 if (Feature.startswith("__") && Feature.endswith("__") && Feature.size() >= 4) 866 Feature = Feature.substr(2, Feature.size() - 4); 867 868 return llvm::StringSwitch<bool>(Feature) 869 .Case("address_sanitizer", LangOpts.Sanitize.Address) 870 .Case("attribute_analyzer_noreturn", true) 871 .Case("attribute_availability", true) 872 .Case("attribute_availability_with_message", true) 873 .Case("attribute_cf_returns_not_retained", true) 874 .Case("attribute_cf_returns_retained", true) 875 .Case("attribute_deprecated_with_message", true) 876 .Case("attribute_ext_vector_type", true) 877 .Case("attribute_ns_returns_not_retained", true) 878 .Case("attribute_ns_returns_retained", true) 879 .Case("attribute_ns_consumes_self", true) 880 .Case("attribute_ns_consumed", true) 881 .Case("attribute_cf_consumed", true) 882 .Case("attribute_objc_ivar_unused", true) 883 .Case("attribute_objc_method_family", true) 884 .Case("attribute_overloadable", true) 885 .Case("attribute_unavailable_with_message", true) 886 .Case("attribute_unused_on_fields", true) 887 .Case("blocks", LangOpts.Blocks) 888 .Case("c_thread_safety_attributes", true) 889 .Case("cxx_exceptions", LangOpts.Exceptions) 890 .Case("cxx_rtti", LangOpts.RTTI) 891 .Case("enumerator_attributes", true) 892 .Case("memory_sanitizer", LangOpts.Sanitize.Memory) 893 .Case("thread_sanitizer", LangOpts.Sanitize.Thread) 894 .Case("dataflow_sanitizer", LangOpts.Sanitize.DataFlow) 895 // Objective-C features 896 .Case("objc_arr", LangOpts.ObjCAutoRefCount) // FIXME: REMOVE? 897 .Case("objc_arc", LangOpts.ObjCAutoRefCount) 898 .Case("objc_arc_weak", LangOpts.ObjCARCWeak) 899 .Case("objc_default_synthesize_properties", LangOpts.ObjC2) 900 .Case("objc_fixed_enum", LangOpts.ObjC2) 901 .Case("objc_instancetype", LangOpts.ObjC2) 902 .Case("objc_modules", LangOpts.ObjC2 && LangOpts.Modules) 903 .Case("objc_nonfragile_abi", LangOpts.ObjCRuntime.isNonFragile()) 904 .Case("objc_property_explicit_atomic", true) // Does clang support explicit "atomic" keyword? 905 .Case("objc_protocol_qualifier_mangling", true) 906 .Case("objc_weak_class", LangOpts.ObjCRuntime.hasWeakClassImport()) 907 .Case("ownership_holds", true) 908 .Case("ownership_returns", true) 909 .Case("ownership_takes", true) 910 .Case("objc_bool", true) 911 .Case("objc_subscripting", LangOpts.ObjCRuntime.isNonFragile()) 912 .Case("objc_array_literals", LangOpts.ObjC2) 913 .Case("objc_dictionary_literals", LangOpts.ObjC2) 914 .Case("objc_boxed_expressions", LangOpts.ObjC2) 915 .Case("arc_cf_code_audited", true) 916 // C11 features 917 .Case("c_alignas", LangOpts.C11) 918 .Case("c_atomic", LangOpts.C11) 919 .Case("c_generic_selections", LangOpts.C11) 920 .Case("c_static_assert", LangOpts.C11) 921 .Case("c_thread_local", 922 LangOpts.C11 && PP.getTargetInfo().isTLSSupported()) 923 // C++11 features 924 .Case("cxx_access_control_sfinae", LangOpts.CPlusPlus11) 925 .Case("cxx_alias_templates", LangOpts.CPlusPlus11) 926 .Case("cxx_alignas", LangOpts.CPlusPlus11) 927 .Case("cxx_atomic", LangOpts.CPlusPlus11) 928 .Case("cxx_attributes", LangOpts.CPlusPlus11) 929 .Case("cxx_auto_type", LangOpts.CPlusPlus11) 930 .Case("cxx_constexpr", LangOpts.CPlusPlus11) 931 .Case("cxx_decltype", LangOpts.CPlusPlus11) 932 .Case("cxx_decltype_incomplete_return_types", LangOpts.CPlusPlus11) 933 .Case("cxx_default_function_template_args", LangOpts.CPlusPlus11) 934 .Case("cxx_defaulted_functions", LangOpts.CPlusPlus11) 935 .Case("cxx_delegating_constructors", LangOpts.CPlusPlus11) 936 .Case("cxx_deleted_functions", LangOpts.CPlusPlus11) 937 .Case("cxx_explicit_conversions", LangOpts.CPlusPlus11) 938 .Case("cxx_generalized_initializers", LangOpts.CPlusPlus11) 939 .Case("cxx_implicit_moves", LangOpts.CPlusPlus11) 940 .Case("cxx_inheriting_constructors", LangOpts.CPlusPlus11) 941 .Case("cxx_inline_namespaces", LangOpts.CPlusPlus11) 942 .Case("cxx_lambdas", LangOpts.CPlusPlus11) 943 .Case("cxx_local_type_template_args", LangOpts.CPlusPlus11) 944 .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus11) 945 .Case("cxx_noexcept", LangOpts.CPlusPlus11) 946 .Case("cxx_nullptr", LangOpts.CPlusPlus11) 947 .Case("cxx_override_control", LangOpts.CPlusPlus11) 948 .Case("cxx_range_for", LangOpts.CPlusPlus11) 949 .Case("cxx_raw_string_literals", LangOpts.CPlusPlus11) 950 .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus11) 951 .Case("cxx_rvalue_references", LangOpts.CPlusPlus11) 952 .Case("cxx_strong_enums", LangOpts.CPlusPlus11) 953 .Case("cxx_static_assert", LangOpts.CPlusPlus11) 954 .Case("cxx_thread_local", 955 LangOpts.CPlusPlus11 && PP.getTargetInfo().isTLSSupported()) 956 .Case("cxx_trailing_return", LangOpts.CPlusPlus11) 957 .Case("cxx_unicode_literals", LangOpts.CPlusPlus11) 958 .Case("cxx_unrestricted_unions", LangOpts.CPlusPlus11) 959 .Case("cxx_user_literals", LangOpts.CPlusPlus11) 960 .Case("cxx_variadic_templates", LangOpts.CPlusPlus11) 961 // C++1y features 962 .Case("cxx_aggregate_nsdmi", LangOpts.CPlusPlus1y) 963 .Case("cxx_binary_literals", LangOpts.CPlusPlus1y) 964 .Case("cxx_contextual_conversions", LangOpts.CPlusPlus1y) 965 //.Case("cxx_generic_lambdas", LangOpts.CPlusPlus1y) 966 .Case("cxx_init_captures", LangOpts.CPlusPlus1y) 967 .Case("cxx_relaxed_constexpr", LangOpts.CPlusPlus1y) 968 .Case("cxx_return_type_deduction", LangOpts.CPlusPlus1y) 969 //.Case("cxx_runtime_arrays", LangOpts.CPlusPlus1y) 970 .Case("cxx_variable_templates", LangOpts.CPlusPlus1y) 971 // Type traits 972 .Case("has_nothrow_assign", LangOpts.CPlusPlus) 973 .Case("has_nothrow_copy", LangOpts.CPlusPlus) 974 .Case("has_nothrow_constructor", LangOpts.CPlusPlus) 975 .Case("has_trivial_assign", LangOpts.CPlusPlus) 976 .Case("has_trivial_copy", LangOpts.CPlusPlus) 977 .Case("has_trivial_constructor", LangOpts.CPlusPlus) 978 .Case("has_trivial_destructor", LangOpts.CPlusPlus) 979 .Case("has_virtual_destructor", LangOpts.CPlusPlus) 980 .Case("is_abstract", LangOpts.CPlusPlus) 981 .Case("is_base_of", LangOpts.CPlusPlus) 982 .Case("is_class", LangOpts.CPlusPlus) 983 .Case("is_constructible", LangOpts.CPlusPlus) 984 .Case("is_convertible_to", LangOpts.CPlusPlus) 985 .Case("is_empty", LangOpts.CPlusPlus) 986 .Case("is_enum", LangOpts.CPlusPlus) 987 .Case("is_final", LangOpts.CPlusPlus) 988 .Case("is_literal", LangOpts.CPlusPlus) 989 .Case("is_standard_layout", LangOpts.CPlusPlus) 990 .Case("is_pod", LangOpts.CPlusPlus) 991 .Case("is_polymorphic", LangOpts.CPlusPlus) 992 .Case("is_sealed", LangOpts.MicrosoftExt) 993 .Case("is_trivial", LangOpts.CPlusPlus) 994 .Case("is_trivially_assignable", LangOpts.CPlusPlus) 995 .Case("is_trivially_constructible", LangOpts.CPlusPlus) 996 .Case("is_trivially_copyable", LangOpts.CPlusPlus) 997 .Case("is_union", LangOpts.CPlusPlus) 998 .Case("modules", LangOpts.Modules) 999 .Case("tls", PP.getTargetInfo().isTLSSupported()) 1000 .Case("underlying_type", LangOpts.CPlusPlus) 1001 .Default(false); 1002 } 1003 1004 /// HasExtension - Return true if we recognize and implement the feature 1005 /// specified by the identifier, either as an extension or a standard language 1006 /// feature. 1007 static bool HasExtension(const Preprocessor &PP, const IdentifierInfo *II) { 1008 if (HasFeature(PP, II)) 1009 return true; 1010 1011 // If the use of an extension results in an error diagnostic, extensions are 1012 // effectively unavailable, so just return false here. 1013 if (PP.getDiagnostics().getExtensionHandlingBehavior() == 1014 DiagnosticsEngine::Ext_Error) 1015 return false; 1016 1017 const LangOptions &LangOpts = PP.getLangOpts(); 1018 StringRef Extension = II->getName(); 1019 1020 // Normalize the extension name, __foo__ becomes foo. 1021 if (Extension.startswith("__") && Extension.endswith("__") && 1022 Extension.size() >= 4) 1023 Extension = Extension.substr(2, Extension.size() - 4); 1024 1025 // Because we inherit the feature list from HasFeature, this string switch 1026 // must be less restrictive than HasFeature's. 1027 return llvm::StringSwitch<bool>(Extension) 1028 // C11 features supported by other languages as extensions. 1029 .Case("c_alignas", true) 1030 .Case("c_atomic", true) 1031 .Case("c_generic_selections", true) 1032 .Case("c_static_assert", true) 1033 .Case("c_thread_local", PP.getTargetInfo().isTLSSupported()) 1034 // C++11 features supported by other languages as extensions. 1035 .Case("cxx_atomic", LangOpts.CPlusPlus) 1036 .Case("cxx_deleted_functions", LangOpts.CPlusPlus) 1037 .Case("cxx_explicit_conversions", LangOpts.CPlusPlus) 1038 .Case("cxx_inline_namespaces", LangOpts.CPlusPlus) 1039 .Case("cxx_local_type_template_args", LangOpts.CPlusPlus) 1040 .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus) 1041 .Case("cxx_override_control", LangOpts.CPlusPlus) 1042 .Case("cxx_range_for", LangOpts.CPlusPlus) 1043 .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus) 1044 .Case("cxx_rvalue_references", LangOpts.CPlusPlus) 1045 // C++1y features supported by other languages as extensions. 1046 .Case("cxx_binary_literals", true) 1047 .Case("cxx_init_captures", LangOpts.CPlusPlus11) 1048 .Case("cxx_variable_templates", LangOpts.CPlusPlus) 1049 .Default(false); 1050 } 1051 1052 /// EvaluateHasIncludeCommon - Process a '__has_include("path")' 1053 /// or '__has_include_next("path")' expression. 1054 /// Returns true if successful. 1055 static bool EvaluateHasIncludeCommon(Token &Tok, 1056 IdentifierInfo *II, Preprocessor &PP, 1057 const DirectoryLookup *LookupFrom) { 1058 // Save the location of the current token. If a '(' is later found, use 1059 // that location. If not, use the end of this location instead. 1060 SourceLocation LParenLoc = Tok.getLocation(); 1061 1062 // These expressions are only allowed within a preprocessor directive. 1063 if (!PP.isParsingIfOrElifDirective()) { 1064 PP.Diag(LParenLoc, diag::err_pp_directive_required) << II->getName(); 1065 return false; 1066 } 1067 1068 // Get '('. 1069 PP.LexNonComment(Tok); 1070 1071 // Ensure we have a '('. 1072 if (Tok.isNot(tok::l_paren)) { 1073 // No '(', use end of last token. 1074 LParenLoc = PP.getLocForEndOfToken(LParenLoc); 1075 PP.Diag(LParenLoc, diag::err_pp_expected_after) << II << tok::l_paren; 1076 // If the next token looks like a filename or the start of one, 1077 // assume it is and process it as such. 1078 if (!Tok.is(tok::angle_string_literal) && !Tok.is(tok::string_literal) && 1079 !Tok.is(tok::less)) 1080 return false; 1081 } else { 1082 // Save '(' location for possible missing ')' message. 1083 LParenLoc = Tok.getLocation(); 1084 1085 if (PP.getCurrentLexer()) { 1086 // Get the file name. 1087 PP.getCurrentLexer()->LexIncludeFilename(Tok); 1088 } else { 1089 // We're in a macro, so we can't use LexIncludeFilename; just 1090 // grab the next token. 1091 PP.Lex(Tok); 1092 } 1093 } 1094 1095 // Reserve a buffer to get the spelling. 1096 SmallString<128> FilenameBuffer; 1097 StringRef Filename; 1098 SourceLocation EndLoc; 1099 1100 switch (Tok.getKind()) { 1101 case tok::eod: 1102 // If the token kind is EOD, the error has already been diagnosed. 1103 return false; 1104 1105 case tok::angle_string_literal: 1106 case tok::string_literal: { 1107 bool Invalid = false; 1108 Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid); 1109 if (Invalid) 1110 return false; 1111 break; 1112 } 1113 1114 case tok::less: 1115 // This could be a <foo/bar.h> file coming from a macro expansion. In this 1116 // case, glue the tokens together into FilenameBuffer and interpret those. 1117 FilenameBuffer.push_back('<'); 1118 if (PP.ConcatenateIncludeName(FilenameBuffer, EndLoc)) { 1119 // Let the caller know a <eod> was found by changing the Token kind. 1120 Tok.setKind(tok::eod); 1121 return false; // Found <eod> but no ">"? Diagnostic already emitted. 1122 } 1123 Filename = FilenameBuffer.str(); 1124 break; 1125 default: 1126 PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename); 1127 return false; 1128 } 1129 1130 SourceLocation FilenameLoc = Tok.getLocation(); 1131 1132 // Get ')'. 1133 PP.LexNonComment(Tok); 1134 1135 // Ensure we have a trailing ). 1136 if (Tok.isNot(tok::r_paren)) { 1137 PP.Diag(PP.getLocForEndOfToken(FilenameLoc), diag::err_pp_expected_after) 1138 << II << tok::r_paren; 1139 PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren; 1140 return false; 1141 } 1142 1143 bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename); 1144 // If GetIncludeFilenameSpelling set the start ptr to null, there was an 1145 // error. 1146 if (Filename.empty()) 1147 return false; 1148 1149 // Search include directories. 1150 const DirectoryLookup *CurDir; 1151 const FileEntry *File = 1152 PP.LookupFile(FilenameLoc, Filename, isAngled, LookupFrom, CurDir, NULL, 1153 NULL, NULL); 1154 1155 // Get the result value. A result of true means the file exists. 1156 return File != 0; 1157 } 1158 1159 /// EvaluateHasInclude - Process a '__has_include("path")' expression. 1160 /// Returns true if successful. 1161 static bool EvaluateHasInclude(Token &Tok, IdentifierInfo *II, 1162 Preprocessor &PP) { 1163 return EvaluateHasIncludeCommon(Tok, II, PP, NULL); 1164 } 1165 1166 /// EvaluateHasIncludeNext - Process '__has_include_next("path")' expression. 1167 /// Returns true if successful. 1168 static bool EvaluateHasIncludeNext(Token &Tok, 1169 IdentifierInfo *II, Preprocessor &PP) { 1170 // __has_include_next is like __has_include, except that we start 1171 // searching after the current found directory. If we can't do this, 1172 // issue a diagnostic. 1173 const DirectoryLookup *Lookup = PP.GetCurDirLookup(); 1174 if (PP.isInPrimaryFile()) { 1175 Lookup = 0; 1176 PP.Diag(Tok, diag::pp_include_next_in_primary); 1177 } else if (Lookup == 0) { 1178 PP.Diag(Tok, diag::pp_include_next_absolute_path); 1179 } else { 1180 // Start looking up in the next directory. 1181 ++Lookup; 1182 } 1183 1184 return EvaluateHasIncludeCommon(Tok, II, PP, Lookup); 1185 } 1186 1187 /// \brief Process __building_module(identifier) expression. 1188 /// \returns true if we are building the named module, false otherwise. 1189 static bool EvaluateBuildingModule(Token &Tok, 1190 IdentifierInfo *II, Preprocessor &PP) { 1191 // Get '('. 1192 PP.LexNonComment(Tok); 1193 1194 // Ensure we have a '('. 1195 if (Tok.isNot(tok::l_paren)) { 1196 PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II 1197 << tok::l_paren; 1198 return false; 1199 } 1200 1201 // Save '(' location for possible missing ')' message. 1202 SourceLocation LParenLoc = Tok.getLocation(); 1203 1204 // Get the module name. 1205 PP.LexNonComment(Tok); 1206 1207 // Ensure that we have an identifier. 1208 if (Tok.isNot(tok::identifier)) { 1209 PP.Diag(Tok.getLocation(), diag::err_expected_id_building_module); 1210 return false; 1211 } 1212 1213 bool Result 1214 = Tok.getIdentifierInfo()->getName() == PP.getLangOpts().CurrentModule; 1215 1216 // Get ')'. 1217 PP.LexNonComment(Tok); 1218 1219 // Ensure we have a trailing ). 1220 if (Tok.isNot(tok::r_paren)) { 1221 PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II 1222 << tok::r_paren; 1223 PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren; 1224 return false; 1225 } 1226 1227 return Result; 1228 } 1229 1230 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded 1231 /// as a builtin macro, handle it and return the next token as 'Tok'. 1232 void Preprocessor::ExpandBuiltinMacro(Token &Tok) { 1233 // Figure out which token this is. 1234 IdentifierInfo *II = Tok.getIdentifierInfo(); 1235 assert(II && "Can't be a macro without id info!"); 1236 1237 // If this is an _Pragma or Microsoft __pragma directive, expand it, 1238 // invoke the pragma handler, then lex the token after it. 1239 if (II == Ident_Pragma) 1240 return Handle_Pragma(Tok); 1241 else if (II == Ident__pragma) // in non-MS mode this is null 1242 return HandleMicrosoft__pragma(Tok); 1243 1244 ++NumBuiltinMacroExpanded; 1245 1246 SmallString<128> TmpBuffer; 1247 llvm::raw_svector_ostream OS(TmpBuffer); 1248 1249 // Set up the return result. 1250 Tok.setIdentifierInfo(0); 1251 Tok.clearFlag(Token::NeedsCleaning); 1252 1253 if (II == Ident__LINE__) { 1254 // C99 6.10.8: "__LINE__: The presumed line number (within the current 1255 // source file) of the current source line (an integer constant)". This can 1256 // be affected by #line. 1257 SourceLocation Loc = Tok.getLocation(); 1258 1259 // Advance to the location of the first _, this might not be the first byte 1260 // of the token if it starts with an escaped newline. 1261 Loc = AdvanceToTokenCharacter(Loc, 0); 1262 1263 // One wrinkle here is that GCC expands __LINE__ to location of the *end* of 1264 // a macro expansion. This doesn't matter for object-like macros, but 1265 // can matter for a function-like macro that expands to contain __LINE__. 1266 // Skip down through expansion points until we find a file loc for the 1267 // end of the expansion history. 1268 Loc = SourceMgr.getExpansionRange(Loc).second; 1269 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Loc); 1270 1271 // __LINE__ expands to a simple numeric value. 1272 OS << (PLoc.isValid()? PLoc.getLine() : 1); 1273 Tok.setKind(tok::numeric_constant); 1274 } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__) { 1275 // C99 6.10.8: "__FILE__: The presumed name of the current source file (a 1276 // character string literal)". This can be affected by #line. 1277 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation()); 1278 1279 // __BASE_FILE__ is a GNU extension that returns the top of the presumed 1280 // #include stack instead of the current file. 1281 if (II == Ident__BASE_FILE__ && PLoc.isValid()) { 1282 SourceLocation NextLoc = PLoc.getIncludeLoc(); 1283 while (NextLoc.isValid()) { 1284 PLoc = SourceMgr.getPresumedLoc(NextLoc); 1285 if (PLoc.isInvalid()) 1286 break; 1287 1288 NextLoc = PLoc.getIncludeLoc(); 1289 } 1290 } 1291 1292 // Escape this filename. Turn '\' -> '\\' '"' -> '\"' 1293 SmallString<128> FN; 1294 if (PLoc.isValid()) { 1295 FN += PLoc.getFilename(); 1296 Lexer::Stringify(FN); 1297 OS << '"' << FN.str() << '"'; 1298 } 1299 Tok.setKind(tok::string_literal); 1300 } else if (II == Ident__DATE__) { 1301 if (!DATELoc.isValid()) 1302 ComputeDATE_TIME(DATELoc, TIMELoc, *this); 1303 Tok.setKind(tok::string_literal); 1304 Tok.setLength(strlen("\"Mmm dd yyyy\"")); 1305 Tok.setLocation(SourceMgr.createExpansionLoc(DATELoc, Tok.getLocation(), 1306 Tok.getLocation(), 1307 Tok.getLength())); 1308 return; 1309 } else if (II == Ident__TIME__) { 1310 if (!TIMELoc.isValid()) 1311 ComputeDATE_TIME(DATELoc, TIMELoc, *this); 1312 Tok.setKind(tok::string_literal); 1313 Tok.setLength(strlen("\"hh:mm:ss\"")); 1314 Tok.setLocation(SourceMgr.createExpansionLoc(TIMELoc, Tok.getLocation(), 1315 Tok.getLocation(), 1316 Tok.getLength())); 1317 return; 1318 } else if (II == Ident__INCLUDE_LEVEL__) { 1319 // Compute the presumed include depth of this token. This can be affected 1320 // by GNU line markers. 1321 unsigned Depth = 0; 1322 1323 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation()); 1324 if (PLoc.isValid()) { 1325 PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc()); 1326 for (; PLoc.isValid(); ++Depth) 1327 PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc()); 1328 } 1329 1330 // __INCLUDE_LEVEL__ expands to a simple numeric value. 1331 OS << Depth; 1332 Tok.setKind(tok::numeric_constant); 1333 } else if (II == Ident__TIMESTAMP__) { 1334 // MSVC, ICC, GCC, VisualAge C++ extension. The generated string should be 1335 // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime. 1336 1337 // Get the file that we are lexing out of. If we're currently lexing from 1338 // a macro, dig into the include stack. 1339 const FileEntry *CurFile = 0; 1340 PreprocessorLexer *TheLexer = getCurrentFileLexer(); 1341 1342 if (TheLexer) 1343 CurFile = SourceMgr.getFileEntryForID(TheLexer->getFileID()); 1344 1345 const char *Result; 1346 if (CurFile) { 1347 time_t TT = CurFile->getModificationTime(); 1348 struct tm *TM = localtime(&TT); 1349 Result = asctime(TM); 1350 } else { 1351 Result = "??? ??? ?? ??:??:?? ????\n"; 1352 } 1353 // Surround the string with " and strip the trailing newline. 1354 OS << '"' << StringRef(Result, strlen(Result)-1) << '"'; 1355 Tok.setKind(tok::string_literal); 1356 } else if (II == Ident__COUNTER__) { 1357 // __COUNTER__ expands to a simple numeric value. 1358 OS << CounterValue++; 1359 Tok.setKind(tok::numeric_constant); 1360 } else if (II == Ident__has_feature || 1361 II == Ident__has_extension || 1362 II == Ident__has_builtin || 1363 II == Ident__is_identifier || 1364 II == Ident__has_attribute) { 1365 // The argument to these builtins should be a parenthesized identifier. 1366 SourceLocation StartLoc = Tok.getLocation(); 1367 1368 bool IsValid = false; 1369 IdentifierInfo *FeatureII = 0; 1370 1371 // Read the '('. 1372 LexUnexpandedToken(Tok); 1373 if (Tok.is(tok::l_paren)) { 1374 // Read the identifier 1375 LexUnexpandedToken(Tok); 1376 if ((FeatureII = Tok.getIdentifierInfo())) { 1377 // Read the ')'. 1378 LexUnexpandedToken(Tok); 1379 if (Tok.is(tok::r_paren)) 1380 IsValid = true; 1381 } 1382 } 1383 1384 bool Value = false; 1385 if (!IsValid) 1386 Diag(StartLoc, diag::err_feature_check_malformed); 1387 else if (II == Ident__is_identifier) 1388 Value = FeatureII->getTokenID() == tok::identifier; 1389 else if (II == Ident__has_builtin) { 1390 // Check for a builtin is trivial. 1391 Value = FeatureII->getBuiltinID() != 0; 1392 } else if (II == Ident__has_attribute) 1393 Value = hasAttribute(AttrSyntax::Generic, nullptr, FeatureII, 1394 getTargetInfo().getTriple(), getLangOpts()); 1395 else if (II == Ident__has_extension) 1396 Value = HasExtension(*this, FeatureII); 1397 else { 1398 assert(II == Ident__has_feature && "Must be feature check"); 1399 Value = HasFeature(*this, FeatureII); 1400 } 1401 1402 OS << (int)Value; 1403 if (IsValid) 1404 Tok.setKind(tok::numeric_constant); 1405 } else if (II == Ident__has_include || 1406 II == Ident__has_include_next) { 1407 // The argument to these two builtins should be a parenthesized 1408 // file name string literal using angle brackets (<>) or 1409 // double-quotes (""). 1410 bool Value; 1411 if (II == Ident__has_include) 1412 Value = EvaluateHasInclude(Tok, II, *this); 1413 else 1414 Value = EvaluateHasIncludeNext(Tok, II, *this); 1415 OS << (int)Value; 1416 if (Tok.is(tok::r_paren)) 1417 Tok.setKind(tok::numeric_constant); 1418 } else if (II == Ident__has_warning) { 1419 // The argument should be a parenthesized string literal. 1420 // The argument to these builtins should be a parenthesized identifier. 1421 SourceLocation StartLoc = Tok.getLocation(); 1422 bool IsValid = false; 1423 bool Value = false; 1424 // Read the '('. 1425 LexUnexpandedToken(Tok); 1426 do { 1427 if (Tok.isNot(tok::l_paren)) { 1428 Diag(StartLoc, diag::err_warning_check_malformed); 1429 break; 1430 } 1431 1432 LexUnexpandedToken(Tok); 1433 std::string WarningName; 1434 SourceLocation StrStartLoc = Tok.getLocation(); 1435 if (!FinishLexStringLiteral(Tok, WarningName, "'__has_warning'", 1436 /*MacroExpansion=*/false)) { 1437 // Eat tokens until ')'. 1438 while (Tok.isNot(tok::r_paren) && Tok.isNot(tok::eod) && 1439 Tok.isNot(tok::eof)) 1440 LexUnexpandedToken(Tok); 1441 break; 1442 } 1443 1444 // Is the end a ')'? 1445 if (!(IsValid = Tok.is(tok::r_paren))) { 1446 Diag(StartLoc, diag::err_warning_check_malformed); 1447 break; 1448 } 1449 1450 if (WarningName.size() < 3 || WarningName[0] != '-' || 1451 WarningName[1] != 'W') { 1452 Diag(StrStartLoc, diag::warn_has_warning_invalid_option); 1453 break; 1454 } 1455 1456 // Finally, check if the warning flags maps to a diagnostic group. 1457 // We construct a SmallVector here to talk to getDiagnosticIDs(). 1458 // Although we don't use the result, this isn't a hot path, and not 1459 // worth special casing. 1460 SmallVector<diag::kind, 10> Diags; 1461 Value = !getDiagnostics().getDiagnosticIDs()-> 1462 getDiagnosticsInGroup(WarningName.substr(2), Diags); 1463 } while (false); 1464 1465 OS << (int)Value; 1466 if (IsValid) 1467 Tok.setKind(tok::numeric_constant); 1468 } else if (II == Ident__building_module) { 1469 // The argument to this builtin should be an identifier. The 1470 // builtin evaluates to 1 when that identifier names the module we are 1471 // currently building. 1472 OS << (int)EvaluateBuildingModule(Tok, II, *this); 1473 Tok.setKind(tok::numeric_constant); 1474 } else if (II == Ident__MODULE__) { 1475 // The current module as an identifier. 1476 OS << getLangOpts().CurrentModule; 1477 IdentifierInfo *ModuleII = getIdentifierInfo(getLangOpts().CurrentModule); 1478 Tok.setIdentifierInfo(ModuleII); 1479 Tok.setKind(ModuleII->getTokenID()); 1480 } else if (II == Ident__identifier) { 1481 SourceLocation Loc = Tok.getLocation(); 1482 1483 // We're expecting '__identifier' '(' identifier ')'. Try to recover 1484 // if the parens are missing. 1485 LexNonComment(Tok); 1486 if (Tok.isNot(tok::l_paren)) { 1487 // No '(', use end of last token. 1488 Diag(getLocForEndOfToken(Loc), diag::err_pp_expected_after) 1489 << II << tok::l_paren; 1490 // If the next token isn't valid as our argument, we can't recover. 1491 if (!Tok.isAnnotation() && Tok.getIdentifierInfo()) 1492 Tok.setKind(tok::identifier); 1493 return; 1494 } 1495 1496 SourceLocation LParenLoc = Tok.getLocation(); 1497 LexNonComment(Tok); 1498 1499 if (!Tok.isAnnotation() && Tok.getIdentifierInfo()) 1500 Tok.setKind(tok::identifier); 1501 else { 1502 Diag(Tok.getLocation(), diag::err_pp_identifier_arg_not_identifier) 1503 << Tok.getKind(); 1504 // Don't walk past anything that's not a real token. 1505 if (Tok.is(tok::eof) || Tok.is(tok::eod) || Tok.isAnnotation()) 1506 return; 1507 } 1508 1509 // Discard the ')', preserving 'Tok' as our result. 1510 Token RParen; 1511 LexNonComment(RParen); 1512 if (RParen.isNot(tok::r_paren)) { 1513 Diag(getLocForEndOfToken(Tok.getLocation()), diag::err_pp_expected_after) 1514 << Tok.getKind() << tok::r_paren; 1515 Diag(LParenLoc, diag::note_matching) << tok::l_paren; 1516 } 1517 return; 1518 } else { 1519 llvm_unreachable("Unknown identifier!"); 1520 } 1521 CreateString(OS.str(), Tok, Tok.getLocation(), Tok.getLocation()); 1522 } 1523 1524 void Preprocessor::markMacroAsUsed(MacroInfo *MI) { 1525 // If the 'used' status changed, and the macro requires 'unused' warning, 1526 // remove its SourceLocation from the warn-for-unused-macro locations. 1527 if (MI->isWarnIfUnused() && !MI->isUsed()) 1528 WarnUnusedMacroLocs.erase(MI->getDefinitionLoc()); 1529 MI->setIsUsed(true); 1530 } 1531