1 //===--- MacroExpansion.cpp - Top level Macro Expansion -------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the top level handling of macro expansion for the
13 //===----------------------------------------------------------------------===//
15 #include "clang/Basic/Attributes.h"
16 #include "clang/Basic/FileManager.h"
17 #include "clang/Basic/IdentifierTable.h"
18 #include "clang/Basic/LLVM.h"
19 #include "clang/Basic/LangOptions.h"
20 #include "clang/Basic/ObjCRuntime.h"
21 #include "clang/Basic/SourceLocation.h"
22 #include "clang/Basic/TargetInfo.h"
23 #include "clang/Lex/CodeCompletionHandler.h"
24 #include "clang/Lex/DirectoryLookup.h"
25 #include "clang/Lex/ExternalPreprocessorSource.h"
26 #include "clang/Lex/LexDiagnostic.h"
27 #include "clang/Lex/MacroArgs.h"
28 #include "clang/Lex/MacroInfo.h"
29 #include "clang/Lex/Preprocessor.h"
30 #include "clang/Lex/PreprocessorLexer.h"
31 #include "clang/Lex/PTHLexer.h"
32 #include "clang/Lex/Token.h"
33 #include "llvm/ADT/ArrayRef.h"
34 #include "llvm/ADT/DenseMap.h"
35 #include "llvm/ADT/DenseSet.h"
36 #include "llvm/ADT/FoldingSet.h"
37 #include "llvm/ADT/None.h"
38 #include "llvm/ADT/Optional.h"
39 #include "llvm/ADT/SmallString.h"
40 #include "llvm/ADT/SmallVector.h"
41 #include "llvm/ADT/STLExtras.h"
42 #include "llvm/ADT/StringRef.h"
43 #include "llvm/ADT/StringSwitch.h"
44 #include "llvm/Support/Casting.h"
45 #include "llvm/Support/ErrorHandling.h"
46 #include "llvm/Support/Format.h"
47 #include "llvm/Support/raw_ostream.h"
57 using namespace clang;
60 Preprocessor::getLocalMacroDirectiveHistory(const IdentifierInfo *II) const {
61 if (!II->hadMacroDefinition())
63 auto Pos = CurSubmoduleState->Macros.find(II);
64 return Pos == CurSubmoduleState->Macros.end() ? nullptr
65 : Pos->second.getLatest();
68 void Preprocessor::appendMacroDirective(IdentifierInfo *II, MacroDirective *MD){
69 assert(MD && "MacroDirective should be non-zero!");
70 assert(!MD->getPrevious() && "Already attached to a MacroDirective history.");
72 MacroState &StoredMD = CurSubmoduleState->Macros[II];
73 auto *OldMD = StoredMD.getLatest();
74 MD->setPrevious(OldMD);
75 StoredMD.setLatest(MD);
76 StoredMD.overrideActiveModuleMacros(*this, II);
78 if (needModuleMacros()) {
79 // Track that we created a new macro directive, so we know we should
80 // consider building a ModuleMacro for it when we get to the end of
82 PendingModuleMacroNames.push_back(II);
85 // Set up the identifier as having associated macro history.
86 II->setHasMacroDefinition(true);
87 if (!MD->isDefined() && LeafModuleMacros.find(II) == LeafModuleMacros.end())
88 II->setHasMacroDefinition(false);
90 II->setChangedSinceDeserialization();
93 void Preprocessor::setLoadedMacroDirective(IdentifierInfo *II,
96 // Normally, when a macro is defined, it goes through appendMacroDirective()
97 // above, which chains a macro to previous defines, undefs, etc.
98 // However, in a pch, the whole macro history up to the end of the pch is
99 // stored, so ASTReader goes through this function instead.
100 // However, built-in macros are already registered in the Preprocessor
101 // ctor, and ASTWriter stops writing the macro chain at built-in macros,
102 // so in that case the chain from the pch needs to be spliced to the existing
106 MacroState &StoredMD = CurSubmoduleState->Macros[II];
108 if (auto *OldMD = StoredMD.getLatest()) {
109 // shouldIgnoreMacro() in ASTWriter also stops at macros from the
110 // predefines buffer in module builds. However, in module builds, modules
111 // are loaded completely before predefines are processed, so StoredMD
112 // will be nullptr for them when they're loaded. StoredMD should only be
113 // non-nullptr for builtins read from a pch file.
114 assert(OldMD->getMacroInfo()->isBuiltinMacro() &&
115 "only built-ins should have an entry here");
116 assert(!OldMD->getPrevious() && "builtin should only have a single entry");
117 ED->setPrevious(OldMD);
118 StoredMD.setLatest(MD);
123 // Setup the identifier as having associated macro history.
124 II->setHasMacroDefinition(true);
125 if (!MD->isDefined() && LeafModuleMacros.find(II) == LeafModuleMacros.end())
126 II->setHasMacroDefinition(false);
129 ModuleMacro *Preprocessor::addModuleMacro(Module *Mod, IdentifierInfo *II,
131 ArrayRef<ModuleMacro *> Overrides,
133 llvm::FoldingSetNodeID ID;
134 ModuleMacro::Profile(ID, Mod, II);
137 if (auto *MM = ModuleMacros.FindNodeOrInsertPos(ID, InsertPos)) {
142 auto *MM = ModuleMacro::create(*this, Mod, II, Macro, Overrides);
143 ModuleMacros.InsertNode(MM, InsertPos);
145 // Each overridden macro is now overridden by one more macro.
147 for (auto *O : Overrides) {
148 HidAny |= (O->NumOverriddenBy == 0);
149 ++O->NumOverriddenBy;
152 // If we were the first overrider for any macro, it's no longer a leaf.
153 auto &LeafMacros = LeafModuleMacros[II];
155 LeafMacros.erase(std::remove_if(LeafMacros.begin(), LeafMacros.end(),
156 [](ModuleMacro *MM) {
157 return MM->NumOverriddenBy != 0;
162 // The new macro is always a leaf macro.
163 LeafMacros.push_back(MM);
164 // The identifier now has defined macros (that may or may not be visible).
165 II->setHasMacroDefinition(true);
171 ModuleMacro *Preprocessor::getModuleMacro(Module *Mod, IdentifierInfo *II) {
172 llvm::FoldingSetNodeID ID;
173 ModuleMacro::Profile(ID, Mod, II);
176 return ModuleMacros.FindNodeOrInsertPos(ID, InsertPos);
179 void Preprocessor::updateModuleMacroInfo(const IdentifierInfo *II,
180 ModuleMacroInfo &Info) {
181 assert(Info.ActiveModuleMacrosGeneration !=
182 CurSubmoduleState->VisibleModules.getGeneration() &&
183 "don't need to update this macro name info");
184 Info.ActiveModuleMacrosGeneration =
185 CurSubmoduleState->VisibleModules.getGeneration();
187 auto Leaf = LeafModuleMacros.find(II);
188 if (Leaf == LeafModuleMacros.end()) {
189 // No imported macros at all: nothing to do.
193 Info.ActiveModuleMacros.clear();
195 // Every macro that's locally overridden is overridden by a visible macro.
196 llvm::DenseMap<ModuleMacro *, int> NumHiddenOverrides;
197 for (auto *O : Info.OverriddenMacros)
198 NumHiddenOverrides[O] = -1;
200 // Collect all macros that are not overridden by a visible macro.
201 llvm::SmallVector<ModuleMacro *, 16> Worklist;
202 for (auto *LeafMM : Leaf->second) {
203 assert(LeafMM->getNumOverridingMacros() == 0 && "leaf macro overridden");
204 if (NumHiddenOverrides.lookup(LeafMM) == 0)
205 Worklist.push_back(LeafMM);
207 while (!Worklist.empty()) {
208 auto *MM = Worklist.pop_back_val();
209 if (CurSubmoduleState->VisibleModules.isVisible(MM->getOwningModule())) {
210 // We only care about collecting definitions; undefinitions only act
211 // to override other definitions.
212 if (MM->getMacroInfo())
213 Info.ActiveModuleMacros.push_back(MM);
215 for (auto *O : MM->overrides())
216 if ((unsigned)++NumHiddenOverrides[O] == O->getNumOverridingMacros())
217 Worklist.push_back(O);
220 // Our reverse postorder walk found the macros in reverse order.
221 std::reverse(Info.ActiveModuleMacros.begin(), Info.ActiveModuleMacros.end());
223 // Determine whether the macro name is ambiguous.
224 MacroInfo *MI = nullptr;
225 bool IsSystemMacro = true;
226 bool IsAmbiguous = false;
227 if (auto *MD = Info.MD) {
228 while (MD && isa<VisibilityMacroDirective>(MD))
229 MD = MD->getPrevious();
230 if (auto *DMD = dyn_cast_or_null<DefMacroDirective>(MD)) {
232 IsSystemMacro &= SourceMgr.isInSystemHeader(DMD->getLocation());
235 for (auto *Active : Info.ActiveModuleMacros) {
236 auto *NewMI = Active->getMacroInfo();
238 // Before marking the macro as ambiguous, check if this is a case where
239 // both macros are in system headers. If so, we trust that the system
240 // did not get it wrong. This also handles cases where Clang's own
241 // headers have a different spelling of certain system macros:
242 // #define LONG_MAX __LONG_MAX__ (clang's limits.h)
243 // #define LONG_MAX 0x7fffffffffffffffL (system's limits.h)
245 // FIXME: Remove the defined-in-system-headers check. clang's limits.h
246 // overrides the system limits.h's macros, so there's no conflict here.
247 if (MI && NewMI != MI &&
248 !MI->isIdenticalTo(*NewMI, *this, /*Syntactically=*/true))
250 IsSystemMacro &= Active->getOwningModule()->IsSystem ||
251 SourceMgr.isInSystemHeader(NewMI->getDefinitionLoc());
254 Info.IsAmbiguous = IsAmbiguous && !IsSystemMacro;
257 void Preprocessor::dumpMacroInfo(const IdentifierInfo *II) {
258 ArrayRef<ModuleMacro*> Leaf;
259 auto LeafIt = LeafModuleMacros.find(II);
260 if (LeafIt != LeafModuleMacros.end())
261 Leaf = LeafIt->second;
262 const MacroState *State = nullptr;
263 auto Pos = CurSubmoduleState->Macros.find(II);
264 if (Pos != CurSubmoduleState->Macros.end())
265 State = &Pos->second;
267 llvm::errs() << "MacroState " << State << " " << II->getNameStart();
268 if (State && State->isAmbiguous(*this, II))
269 llvm::errs() << " ambiguous";
270 if (State && !State->getOverriddenMacros().empty()) {
271 llvm::errs() << " overrides";
272 for (auto *O : State->getOverriddenMacros())
273 llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
275 llvm::errs() << "\n";
277 // Dump local macro directives.
278 for (auto *MD = State ? State->getLatest() : nullptr; MD;
279 MD = MD->getPrevious()) {
284 // Dump module macros.
285 llvm::DenseSet<ModuleMacro*> Active;
286 for (auto *MM : State ? State->getActiveModuleMacros(*this, II) : None)
288 llvm::DenseSet<ModuleMacro*> Visited;
289 llvm::SmallVector<ModuleMacro *, 16> Worklist(Leaf.begin(), Leaf.end());
290 while (!Worklist.empty()) {
291 auto *MM = Worklist.pop_back_val();
292 llvm::errs() << " ModuleMacro " << MM << " "
293 << MM->getOwningModule()->getFullModuleName();
294 if (!MM->getMacroInfo())
295 llvm::errs() << " undef";
297 if (Active.count(MM))
298 llvm::errs() << " active";
299 else if (!CurSubmoduleState->VisibleModules.isVisible(
300 MM->getOwningModule()))
301 llvm::errs() << " hidden";
302 else if (MM->getMacroInfo())
303 llvm::errs() << " overridden";
305 if (!MM->overrides().empty()) {
306 llvm::errs() << " overrides";
307 for (auto *O : MM->overrides()) {
308 llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
309 if (Visited.insert(O).second)
310 Worklist.push_back(O);
313 llvm::errs() << "\n";
314 if (auto *MI = MM->getMacroInfo()) {
317 llvm::errs() << "\n";
322 /// RegisterBuiltinMacro - Register the specified identifier in the identifier
323 /// table and mark it as a builtin macro to be expanded.
324 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){
325 // Get the identifier.
326 IdentifierInfo *Id = PP.getIdentifierInfo(Name);
328 // Mark it as being a macro that is builtin.
329 MacroInfo *MI = PP.AllocateMacroInfo(SourceLocation());
330 MI->setIsBuiltinMacro();
331 PP.appendDefMacroDirective(Id, MI);
335 /// RegisterBuiltinMacros - Register builtin macros, such as __LINE__ with the
336 /// identifier table.
337 void Preprocessor::RegisterBuiltinMacros() {
338 Ident__LINE__ = RegisterBuiltinMacro(*this, "__LINE__");
339 Ident__FILE__ = RegisterBuiltinMacro(*this, "__FILE__");
340 Ident__DATE__ = RegisterBuiltinMacro(*this, "__DATE__");
341 Ident__TIME__ = RegisterBuiltinMacro(*this, "__TIME__");
342 Ident__COUNTER__ = RegisterBuiltinMacro(*this, "__COUNTER__");
343 Ident_Pragma = RegisterBuiltinMacro(*this, "_Pragma");
345 // C++ Standing Document Extensions.
346 if (LangOpts.CPlusPlus)
347 Ident__has_cpp_attribute =
348 RegisterBuiltinMacro(*this, "__has_cpp_attribute");
350 Ident__has_cpp_attribute = nullptr;
353 Ident__BASE_FILE__ = RegisterBuiltinMacro(*this, "__BASE_FILE__");
354 Ident__INCLUDE_LEVEL__ = RegisterBuiltinMacro(*this, "__INCLUDE_LEVEL__");
355 Ident__TIMESTAMP__ = RegisterBuiltinMacro(*this, "__TIMESTAMP__");
357 // Microsoft Extensions.
358 if (LangOpts.MicrosoftExt) {
359 Ident__identifier = RegisterBuiltinMacro(*this, "__identifier");
360 Ident__pragma = RegisterBuiltinMacro(*this, "__pragma");
362 Ident__identifier = nullptr;
363 Ident__pragma = nullptr;
367 Ident__has_feature = RegisterBuiltinMacro(*this, "__has_feature");
368 Ident__has_extension = RegisterBuiltinMacro(*this, "__has_extension");
369 Ident__has_builtin = RegisterBuiltinMacro(*this, "__has_builtin");
370 Ident__has_attribute = RegisterBuiltinMacro(*this, "__has_attribute");
371 Ident__has_c_attribute = RegisterBuiltinMacro(*this, "__has_c_attribute");
372 Ident__has_declspec = RegisterBuiltinMacro(*this, "__has_declspec_attribute");
373 Ident__has_include = RegisterBuiltinMacro(*this, "__has_include");
374 Ident__has_include_next = RegisterBuiltinMacro(*this, "__has_include_next");
375 Ident__has_warning = RegisterBuiltinMacro(*this, "__has_warning");
376 Ident__is_identifier = RegisterBuiltinMacro(*this, "__is_identifier");
377 Ident__is_target_arch = RegisterBuiltinMacro(*this, "__is_target_arch");
378 Ident__is_target_vendor = RegisterBuiltinMacro(*this, "__is_target_vendor");
379 Ident__is_target_os = RegisterBuiltinMacro(*this, "__is_target_os");
380 Ident__is_target_environment =
381 RegisterBuiltinMacro(*this, "__is_target_environment");
384 Ident__building_module = RegisterBuiltinMacro(*this, "__building_module");
385 if (!LangOpts.CurrentModule.empty())
386 Ident__MODULE__ = RegisterBuiltinMacro(*this, "__MODULE__");
388 Ident__MODULE__ = nullptr;
391 /// isTrivialSingleTokenExpansion - Return true if MI, which has a single token
392 /// in its expansion, currently expands to that token literally.
393 static bool isTrivialSingleTokenExpansion(const MacroInfo *MI,
394 const IdentifierInfo *MacroIdent,
396 IdentifierInfo *II = MI->getReplacementToken(0).getIdentifierInfo();
398 // If the token isn't an identifier, it's always literally expanded.
399 if (!II) return true;
401 // If the information about this identifier is out of date, update it from
402 // the external source.
403 if (II->isOutOfDate())
404 PP.getExternalSource()->updateOutOfDateIdentifier(*II);
406 // If the identifier is a macro, and if that macro is enabled, it may be
407 // expanded so it's not a trivial expansion.
408 if (auto *ExpansionMI = PP.getMacroInfo(II))
409 if (ExpansionMI->isEnabled() &&
410 // Fast expanding "#define X X" is ok, because X would be disabled.
414 // If this is an object-like macro invocation, it is safe to trivially expand
416 if (MI->isObjectLike()) return true;
418 // If this is a function-like macro invocation, it's safe to trivially expand
419 // as long as the identifier is not a macro argument.
420 return std::find(MI->param_begin(), MI->param_end(), II) == MI->param_end();
423 /// isNextPPTokenLParen - Determine whether the next preprocessor token to be
424 /// lexed is a '('. If so, consume the token and return true, if not, this
425 /// method should have no observable side-effect on the lexed tokens.
426 bool Preprocessor::isNextPPTokenLParen() {
427 // Do some quick tests for rejection cases.
430 Val = CurLexer->isNextPPTokenLParen();
431 else if (CurPTHLexer)
432 Val = CurPTHLexer->isNextPPTokenLParen();
434 Val = CurTokenLexer->isNextTokenLParen();
437 // We have run off the end. If it's a source file we don't
438 // examine enclosing ones (C99 5.1.1.2p4). Otherwise walk up the
442 for (const IncludeStackInfo &Entry : llvm::reverse(IncludeMacroStack)) {
444 Val = Entry.TheLexer->isNextPPTokenLParen();
445 else if (Entry.ThePTHLexer)
446 Val = Entry.ThePTHLexer->isNextPPTokenLParen();
448 Val = Entry.TheTokenLexer->isNextTokenLParen();
453 // Ran off the end of a source file?
454 if (Entry.ThePPLexer)
459 // Okay, if we know that the token is a '(', lex it and return. Otherwise we
460 // have found something that isn't a '(' or we found the end of the
461 // translation unit. In either case, return false.
465 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be
466 /// expanded as a macro, handle it and return the next token as 'Identifier'.
467 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
468 const MacroDefinition &M) {
469 MacroInfo *MI = M.getMacroInfo();
471 // If this is a macro expansion in the "#if !defined(x)" line for the file,
472 // then the macro could expand to different things in other contexts, we need
473 // to disable the optimization in this case.
474 if (CurPPLexer) CurPPLexer->MIOpt.ExpandedMacro();
476 // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially.
477 if (MI->isBuiltinMacro()) {
479 Callbacks->MacroExpands(Identifier, M, Identifier.getLocation(),
481 ExpandBuiltinMacro(Identifier);
485 /// Args - If this is a function-like macro expansion, this contains,
486 /// for each macro argument, the list of tokens that were provided to the
488 MacroArgs *Args = nullptr;
490 // Remember where the end of the expansion occurred. For an object-like
491 // macro, this is the identifier. For a function-like macro, this is the ')'.
492 SourceLocation ExpansionEnd = Identifier.getLocation();
494 // If this is a function-like macro, read the arguments.
495 if (MI->isFunctionLike()) {
496 // Remember that we are now parsing the arguments to a macro invocation.
497 // Preprocessor directives used inside macro arguments are not portable, and
498 // this enables the warning.
500 Args = ReadMacroCallArgumentList(Identifier, MI, ExpansionEnd);
502 // Finished parsing args.
505 // If there was an error parsing the arguments, bail out.
506 if (!Args) return true;
508 ++NumFnMacroExpanded;
513 // Notice that this macro has been used.
516 // Remember where the token is expanded.
517 SourceLocation ExpandLoc = Identifier.getLocation();
518 SourceRange ExpansionRange(ExpandLoc, ExpansionEnd);
522 // We can have macro expansion inside a conditional directive while
523 // reading the function macro arguments. To ensure, in that case, that
524 // MacroExpands callbacks still happen in source order, queue this
525 // callback to have it happen after the function macro callback.
526 DelayedMacroExpandsCallbacks.push_back(
527 MacroExpandsInfo(Identifier, M, ExpansionRange));
529 Callbacks->MacroExpands(Identifier, M, ExpansionRange, Args);
530 if (!DelayedMacroExpandsCallbacks.empty()) {
531 for (const MacroExpandsInfo &Info : DelayedMacroExpandsCallbacks) {
532 // FIXME: We lose macro args info with delayed callback.
533 Callbacks->MacroExpands(Info.Tok, Info.MD, Info.Range,
536 DelayedMacroExpandsCallbacks.clear();
541 // If the macro definition is ambiguous, complain.
542 if (M.isAmbiguous()) {
543 Diag(Identifier, diag::warn_pp_ambiguous_macro)
544 << Identifier.getIdentifierInfo();
545 Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen)
546 << Identifier.getIdentifierInfo();
547 M.forAllDefinitions([&](const MacroInfo *OtherMI) {
549 Diag(OtherMI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_other)
550 << Identifier.getIdentifierInfo();
554 // If we started lexing a macro, enter the macro expansion body.
556 // If this macro expands to no tokens, don't bother to push it onto the
557 // expansion stack, only to take it right back off.
558 if (MI->getNumTokens() == 0) {
559 // No need for arg info.
560 if (Args) Args->destroy(*this);
562 // Propagate whitespace info as if we had pushed, then popped,
564 Identifier.setFlag(Token::LeadingEmptyMacro);
565 PropagateLineStartLeadingSpaceInfo(Identifier);
566 ++NumFastMacroExpanded;
568 } else if (MI->getNumTokens() == 1 &&
569 isTrivialSingleTokenExpansion(MI, Identifier.getIdentifierInfo(),
571 // Otherwise, if this macro expands into a single trivially-expanded
572 // token: expand it now. This handles common cases like
575 // No need for arg info.
576 if (Args) Args->destroy(*this);
578 // Propagate the isAtStartOfLine/hasLeadingSpace markers of the macro
579 // identifier to the expanded token.
580 bool isAtStartOfLine = Identifier.isAtStartOfLine();
581 bool hasLeadingSpace = Identifier.hasLeadingSpace();
583 // Replace the result token.
584 Identifier = MI->getReplacementToken(0);
586 // Restore the StartOfLine/LeadingSpace markers.
587 Identifier.setFlagValue(Token::StartOfLine , isAtStartOfLine);
588 Identifier.setFlagValue(Token::LeadingSpace, hasLeadingSpace);
590 // Update the tokens location to include both its expansion and physical
593 SourceMgr.createExpansionLoc(Identifier.getLocation(), ExpandLoc,
594 ExpansionEnd,Identifier.getLength());
595 Identifier.setLocation(Loc);
597 // If this is a disabled macro or #define X X, we must mark the result as
599 if (IdentifierInfo *NewII = Identifier.getIdentifierInfo()) {
600 if (MacroInfo *NewMI = getMacroInfo(NewII))
601 if (!NewMI->isEnabled() || NewMI == MI) {
602 Identifier.setFlag(Token::DisableExpand);
603 // Don't warn for "#define X X" like "#define bool bool" from
605 if (NewMI != MI || MI->isFunctionLike())
606 Diag(Identifier, diag::pp_disabled_macro_expansion);
610 // Since this is not an identifier token, it can't be macro expanded, so
612 ++NumFastMacroExpanded;
616 // Start expanding the macro.
617 EnterMacro(Identifier, ExpansionEnd, MI, Args);
626 /// CheckMatchedBrackets - Returns true if the braces and parentheses in the
627 /// token vector are properly nested.
628 static bool CheckMatchedBrackets(const SmallVectorImpl<Token> &Tokens) {
629 SmallVector<Bracket, 8> Brackets;
630 for (SmallVectorImpl<Token>::const_iterator I = Tokens.begin(),
633 if (I->is(tok::l_paren)) {
634 Brackets.push_back(Paren);
635 } else if (I->is(tok::r_paren)) {
636 if (Brackets.empty() || Brackets.back() == Brace)
639 } else if (I->is(tok::l_brace)) {
640 Brackets.push_back(Brace);
641 } else if (I->is(tok::r_brace)) {
642 if (Brackets.empty() || Brackets.back() == Paren)
647 return Brackets.empty();
650 /// GenerateNewArgTokens - Returns true if OldTokens can be converted to a new
651 /// vector of tokens in NewTokens. The new number of arguments will be placed
652 /// in NumArgs and the ranges which need to surrounded in parentheses will be
654 /// Returns false if the token stream cannot be changed. If this is because
655 /// of an initializer list starting a macro argument, the range of those
656 /// initializer lists will be place in InitLists.
657 static bool GenerateNewArgTokens(Preprocessor &PP,
658 SmallVectorImpl<Token> &OldTokens,
659 SmallVectorImpl<Token> &NewTokens,
661 SmallVectorImpl<SourceRange> &ParenHints,
662 SmallVectorImpl<SourceRange> &InitLists) {
663 if (!CheckMatchedBrackets(OldTokens))
666 // Once it is known that the brackets are matched, only a simple count of the
670 // First token of a new macro argument.
671 SmallVectorImpl<Token>::iterator ArgStartIterator = OldTokens.begin();
673 // First closing brace in a new macro argument. Used to generate
674 // SourceRanges for InitLists.
675 SmallVectorImpl<Token>::iterator ClosingBrace = OldTokens.end();
678 // Set to true when a macro separator token is found inside a braced list.
679 // If true, the fixed argument spans multiple old arguments and ParenHints
681 bool FoundSeparatorToken = false;
682 for (SmallVectorImpl<Token>::iterator I = OldTokens.begin(),
685 if (I->is(tok::l_brace)) {
687 } else if (I->is(tok::r_brace)) {
689 if (Braces == 0 && ClosingBrace == E && FoundSeparatorToken)
691 } else if (I->is(tok::eof)) {
692 // EOF token is used to separate macro arguments
694 // Assume comma separator is actually braced list separator and change
695 // it back to a comma.
696 FoundSeparatorToken = true;
697 I->setKind(tok::comma);
699 } else { // Braces == 0
700 // Separator token still separates arguments.
703 // If the argument starts with a brace, it can't be fixed with
704 // parentheses. A different diagnostic will be given.
705 if (FoundSeparatorToken && ArgStartIterator->is(tok::l_brace)) {
707 SourceRange(ArgStartIterator->getLocation(),
708 PP.getLocForEndOfToken(ClosingBrace->getLocation())));
713 if (FoundSeparatorToken) {
714 TempToken.startToken();
715 TempToken.setKind(tok::l_paren);
716 TempToken.setLocation(ArgStartIterator->getLocation());
717 TempToken.setLength(0);
718 NewTokens.push_back(TempToken);
721 // Copy over argument tokens
722 NewTokens.insert(NewTokens.end(), ArgStartIterator, I);
724 // Add right paren and store the paren locations in ParenHints
725 if (FoundSeparatorToken) {
726 SourceLocation Loc = PP.getLocForEndOfToken((I - 1)->getLocation());
727 TempToken.startToken();
728 TempToken.setKind(tok::r_paren);
729 TempToken.setLocation(Loc);
730 TempToken.setLength(0);
731 NewTokens.push_back(TempToken);
732 ParenHints.push_back(SourceRange(ArgStartIterator->getLocation(),
736 // Copy separator token
737 NewTokens.push_back(*I);
740 ArgStartIterator = I + 1;
741 FoundSeparatorToken = false;
746 return !ParenHints.empty() && InitLists.empty();
749 /// ReadFunctionLikeMacroArgs - After reading "MACRO" and knowing that the next
750 /// token is the '(' of the macro, this method is invoked to read all of the
751 /// actual arguments specified for the macro invocation. This returns null on
753 MacroArgs *Preprocessor::ReadMacroCallArgumentList(Token &MacroName,
755 SourceLocation &MacroEnd) {
756 // The number of fixed arguments to parse.
757 unsigned NumFixedArgsLeft = MI->getNumParams();
758 bool isVariadic = MI->isVariadic();
760 // Outer loop, while there are more arguments, keep reading them.
763 // Read arguments as unexpanded tokens. This avoids issues, e.g., where
764 // an argument value in a macro could expand to ',' or '(' or ')'.
765 LexUnexpandedToken(Tok);
766 assert(Tok.is(tok::l_paren) && "Error computing l-paren-ness?");
768 // ArgTokens - Build up a list of tokens that make up each argument. Each
769 // argument is separated by an EOF token. Use a SmallVector so we can avoid
770 // heap allocations in the common case.
771 SmallVector<Token, 64> ArgTokens;
772 bool ContainsCodeCompletionTok = false;
773 bool FoundElidedComma = false;
775 SourceLocation TooManyArgsLoc;
777 unsigned NumActuals = 0;
778 while (Tok.isNot(tok::r_paren)) {
779 if (ContainsCodeCompletionTok && Tok.isOneOf(tok::eof, tok::eod))
782 assert(Tok.isOneOf(tok::l_paren, tok::comma) &&
783 "only expect argument separators here");
785 size_t ArgTokenStart = ArgTokens.size();
786 SourceLocation ArgStartLoc = Tok.getLocation();
788 // C99 6.10.3p11: Keep track of the number of l_parens we have seen. Note
789 // that we already consumed the first one.
790 unsigned NumParens = 0;
793 // Read arguments as unexpanded tokens. This avoids issues, e.g., where
794 // an argument value in a macro could expand to ',' or '(' or ')'.
795 LexUnexpandedToken(Tok);
797 if (Tok.isOneOf(tok::eof, tok::eod)) { // "#if f(<eof>" & "#if f(\n"
798 if (!ContainsCodeCompletionTok) {
799 Diag(MacroName, diag::err_unterm_macro_invoc);
800 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
801 << MacroName.getIdentifierInfo();
802 // Do not lose the EOF/EOD. Return it to the client.
806 // Do not lose the EOF/EOD.
807 auto Toks = llvm::make_unique<Token[]>(1);
809 EnterTokenStream(std::move(Toks), 1, true);
811 } else if (Tok.is(tok::r_paren)) {
812 // If we found the ) token, the macro arg list is done.
813 if (NumParens-- == 0) {
814 MacroEnd = Tok.getLocation();
815 if (!ArgTokens.empty() &&
816 ArgTokens.back().commaAfterElided()) {
817 FoundElidedComma = true;
821 } else if (Tok.is(tok::l_paren)) {
823 } else if (Tok.is(tok::comma) && NumParens == 0 &&
824 !(Tok.getFlags() & Token::IgnoredComma)) {
825 // In Microsoft-compatibility mode, single commas from nested macro
826 // expansions should not be considered as argument separators. We test
827 // for this with the IgnoredComma token flag above.
829 // Comma ends this argument if there are more fixed arguments expected.
830 // However, if this is a variadic macro, and this is part of the
831 // variadic part, then the comma is just an argument token.
832 if (!isVariadic) break;
833 if (NumFixedArgsLeft > 1)
835 } else if (Tok.is(tok::comment) && !KeepMacroComments) {
836 // If this is a comment token in the argument list and we're just in
837 // -C mode (not -CC mode), discard the comment.
839 } else if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) {
840 // Reading macro arguments can cause macros that we are currently
841 // expanding from to be popped off the expansion stack. Doing so causes
842 // them to be reenabled for expansion. Here we record whether any
843 // identifiers we lex as macro arguments correspond to disabled macros.
844 // If so, we mark the token as noexpand. This is a subtle aspect of
846 if (MacroInfo *MI = getMacroInfo(Tok.getIdentifierInfo()))
847 if (!MI->isEnabled())
848 Tok.setFlag(Token::DisableExpand);
849 } else if (Tok.is(tok::code_completion)) {
850 ContainsCodeCompletionTok = true;
852 CodeComplete->CodeCompleteMacroArgument(MacroName.getIdentifierInfo(),
854 // Don't mark that we reached the code-completion point because the
855 // parser is going to handle the token and there will be another
856 // code-completion callback.
859 ArgTokens.push_back(Tok);
862 // If this was an empty argument list foo(), don't add this as an empty
864 if (ArgTokens.empty() && Tok.getKind() == tok::r_paren)
867 // If this is not a variadic macro, and too many args were specified, emit
869 if (!isVariadic && NumFixedArgsLeft == 0 && TooManyArgsLoc.isInvalid()) {
870 if (ArgTokens.size() != ArgTokenStart)
871 TooManyArgsLoc = ArgTokens[ArgTokenStart].getLocation();
873 TooManyArgsLoc = ArgStartLoc;
876 // Empty arguments are standard in C99 and C++0x, and are supported as an
877 // extension in other modes.
878 if (ArgTokens.size() == ArgTokenStart && !LangOpts.C99)
879 Diag(Tok, LangOpts.CPlusPlus11 ?
880 diag::warn_cxx98_compat_empty_fnmacro_arg :
881 diag::ext_empty_fnmacro_arg);
883 // Add a marker EOF token to the end of the token list for this argument.
886 EOFTok.setKind(tok::eof);
887 EOFTok.setLocation(Tok.getLocation());
889 ArgTokens.push_back(EOFTok);
891 if (!ContainsCodeCompletionTok && NumFixedArgsLeft != 0)
895 // Okay, we either found the r_paren. Check to see if we parsed too few
897 unsigned MinArgsExpected = MI->getNumParams();
899 // If this is not a variadic macro, and too many args were specified, emit
901 if (!isVariadic && NumActuals > MinArgsExpected &&
902 !ContainsCodeCompletionTok) {
903 // Emit the diagnostic at the macro name in case there is a missing ).
904 // Emitting it at the , could be far away from the macro name.
905 Diag(TooManyArgsLoc, diag::err_too_many_args_in_macro_invoc);
906 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
907 << MacroName.getIdentifierInfo();
909 // Commas from braced initializer lists will be treated as argument
910 // separators inside macros. Attempt to correct for this with parentheses.
911 // TODO: See if this can be generalized to angle brackets for templates
912 // inside macro arguments.
914 SmallVector<Token, 4> FixedArgTokens;
915 unsigned FixedNumArgs = 0;
916 SmallVector<SourceRange, 4> ParenHints, InitLists;
917 if (!GenerateNewArgTokens(*this, ArgTokens, FixedArgTokens, FixedNumArgs,
918 ParenHints, InitLists)) {
919 if (!InitLists.empty()) {
920 DiagnosticBuilder DB =
922 diag::note_init_list_at_beginning_of_macro_argument);
923 for (SourceRange Range : InitLists)
928 if (FixedNumArgs != MinArgsExpected)
931 DiagnosticBuilder DB = Diag(MacroName, diag::note_suggest_parens_for_macro);
932 for (SourceRange ParenLocation : ParenHints) {
933 DB << FixItHint::CreateInsertion(ParenLocation.getBegin(), "(");
934 DB << FixItHint::CreateInsertion(ParenLocation.getEnd(), ")");
936 ArgTokens.swap(FixedArgTokens);
937 NumActuals = FixedNumArgs;
940 // See MacroArgs instance var for description of this.
941 bool isVarargsElided = false;
943 if (ContainsCodeCompletionTok) {
944 // Recover from not-fully-formed macro invocation during code-completion.
947 EOFTok.setKind(tok::eof);
948 EOFTok.setLocation(Tok.getLocation());
950 for (; NumActuals < MinArgsExpected; ++NumActuals)
951 ArgTokens.push_back(EOFTok);
954 if (NumActuals < MinArgsExpected) {
955 // There are several cases where too few arguments is ok, handle them now.
956 if (NumActuals == 0 && MinArgsExpected == 1) {
957 // #define A(X) or #define A(...) ---> A()
959 // If there is exactly one argument, and that argument is missing,
960 // then we have an empty "()" argument empty list. This is fine, even if
961 // the macro expects one argument (the argument is just empty).
962 isVarargsElided = MI->isVariadic();
963 } else if ((FoundElidedComma || MI->isVariadic()) &&
964 (NumActuals+1 == MinArgsExpected || // A(x, ...) -> A(X)
965 (NumActuals == 0 && MinArgsExpected == 2))) {// A(x,...) -> A()
966 // Varargs where the named vararg parameter is missing: OK as extension.
970 // If the macro contains the comma pasting extension, the diagnostic
971 // is suppressed; we know we'll get another diagnostic later.
972 if (!MI->hasCommaPasting()) {
973 Diag(Tok, diag::ext_missing_varargs_arg);
974 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
975 << MacroName.getIdentifierInfo();
978 // Remember this occurred, allowing us to elide the comma when used for
980 // #define A(x, foo...) blah(a, ## foo)
981 // #define B(x, ...) blah(a, ## __VA_ARGS__)
982 // #define C(...) blah(a, ## __VA_ARGS__)
984 isVarargsElided = true;
985 } else if (!ContainsCodeCompletionTok) {
986 // Otherwise, emit the error.
987 Diag(Tok, diag::err_too_few_args_in_macro_invoc);
988 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
989 << MacroName.getIdentifierInfo();
993 // Add a marker EOF token to the end of the token list for this argument.
994 SourceLocation EndLoc = Tok.getLocation();
996 Tok.setKind(tok::eof);
997 Tok.setLocation(EndLoc);
999 ArgTokens.push_back(Tok);
1001 // If we expect two arguments, add both as empty.
1002 if (NumActuals == 0 && MinArgsExpected == 2)
1003 ArgTokens.push_back(Tok);
1005 } else if (NumActuals > MinArgsExpected && !MI->isVariadic() &&
1006 !ContainsCodeCompletionTok) {
1007 // Emit the diagnostic at the macro name in case there is a missing ).
1008 // Emitting it at the , could be far away from the macro name.
1009 Diag(MacroName, diag::err_too_many_args_in_macro_invoc);
1010 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
1011 << MacroName.getIdentifierInfo();
1015 return MacroArgs::create(MI, ArgTokens, isVarargsElided, *this);
1018 /// Keeps macro expanded tokens for TokenLexers.
1020 /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
1021 /// going to lex in the cache and when it finishes the tokens are removed
1022 /// from the end of the cache.
1023 Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer,
1024 ArrayRef<Token> tokens) {
1029 size_t newIndex = MacroExpandedTokens.size();
1030 bool cacheNeedsToGrow = tokens.size() >
1031 MacroExpandedTokens.capacity()-MacroExpandedTokens.size();
1032 MacroExpandedTokens.append(tokens.begin(), tokens.end());
1034 if (cacheNeedsToGrow) {
1035 // Go through all the TokenLexers whose 'Tokens' pointer points in the
1036 // buffer and update the pointers to the (potential) new buffer array.
1037 for (const auto &Lexer : MacroExpandingLexersStack) {
1038 TokenLexer *prevLexer;
1040 std::tie(prevLexer, tokIndex) = Lexer;
1041 prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex;
1045 MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex));
1046 return MacroExpandedTokens.data() + newIndex;
1049 void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() {
1050 assert(!MacroExpandingLexersStack.empty());
1051 size_t tokIndex = MacroExpandingLexersStack.back().second;
1052 assert(tokIndex < MacroExpandedTokens.size());
1053 // Pop the cached macro expanded tokens from the end.
1054 MacroExpandedTokens.resize(tokIndex);
1055 MacroExpandingLexersStack.pop_back();
1058 /// ComputeDATE_TIME - Compute the current time, enter it into the specified
1059 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of
1060 /// the identifier tokens inserted.
1061 static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc,
1063 time_t TT = time(nullptr);
1064 struct tm *TM = localtime(&TT);
1066 static const char * const Months[] = {
1067 "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"
1071 SmallString<32> TmpBuffer;
1072 llvm::raw_svector_ostream TmpStream(TmpBuffer);
1073 TmpStream << llvm::format("\"%s %2d %4d\"", Months[TM->tm_mon],
1074 TM->tm_mday, TM->tm_year + 1900);
1076 TmpTok.startToken();
1077 PP.CreateString(TmpStream.str(), TmpTok);
1078 DATELoc = TmpTok.getLocation();
1082 SmallString<32> TmpBuffer;
1083 llvm::raw_svector_ostream TmpStream(TmpBuffer);
1084 TmpStream << llvm::format("\"%02d:%02d:%02d\"",
1085 TM->tm_hour, TM->tm_min, TM->tm_sec);
1087 TmpTok.startToken();
1088 PP.CreateString(TmpStream.str(), TmpTok);
1089 TIMELoc = TmpTok.getLocation();
1093 /// HasFeature - Return true if we recognize and implement the feature
1094 /// specified by the identifier as a standard language feature.
1095 static bool HasFeature(const Preprocessor &PP, StringRef Feature) {
1096 const LangOptions &LangOpts = PP.getLangOpts();
1098 // Normalize the feature name, __foo__ becomes foo.
1099 if (Feature.startswith("__") && Feature.endswith("__") && Feature.size() >= 4)
1100 Feature = Feature.substr(2, Feature.size() - 4);
1102 #define FEATURE(Name, Predicate) .Case(#Name, Predicate)
1103 return llvm::StringSwitch<bool>(Feature)
1104 #include "clang/Basic/Features.def"
1109 /// HasExtension - Return true if we recognize and implement the feature
1110 /// specified by the identifier, either as an extension or a standard language
1112 static bool HasExtension(const Preprocessor &PP, StringRef Extension) {
1113 if (HasFeature(PP, Extension))
1116 // If the use of an extension results in an error diagnostic, extensions are
1117 // effectively unavailable, so just return false here.
1118 if (PP.getDiagnostics().getExtensionHandlingBehavior() >=
1119 diag::Severity::Error)
1122 const LangOptions &LangOpts = PP.getLangOpts();
1124 // Normalize the extension name, __foo__ becomes foo.
1125 if (Extension.startswith("__") && Extension.endswith("__") &&
1126 Extension.size() >= 4)
1127 Extension = Extension.substr(2, Extension.size() - 4);
1129 // Because we inherit the feature list from HasFeature, this string switch
1130 // must be less restrictive than HasFeature's.
1131 #define EXTENSION(Name, Predicate) .Case(#Name, Predicate)
1132 return llvm::StringSwitch<bool>(Extension)
1133 #include "clang/Basic/Features.def"
1138 /// EvaluateHasIncludeCommon - Process a '__has_include("path")'
1139 /// or '__has_include_next("path")' expression.
1140 /// Returns true if successful.
1141 static bool EvaluateHasIncludeCommon(Token &Tok,
1142 IdentifierInfo *II, Preprocessor &PP,
1143 const DirectoryLookup *LookupFrom,
1144 const FileEntry *LookupFromFile) {
1145 // Save the location of the current token. If a '(' is later found, use
1146 // that location. If not, use the end of this location instead.
1147 SourceLocation LParenLoc = Tok.getLocation();
1149 // These expressions are only allowed within a preprocessor directive.
1150 if (!PP.isParsingIfOrElifDirective()) {
1151 PP.Diag(LParenLoc, diag::err_pp_directive_required) << II;
1152 // Return a valid identifier token.
1153 assert(Tok.is(tok::identifier));
1154 Tok.setIdentifierInfo(II);
1159 PP.LexNonComment(Tok);
1161 // Ensure we have a '('.
1162 if (Tok.isNot(tok::l_paren)) {
1163 // No '(', use end of last token.
1164 LParenLoc = PP.getLocForEndOfToken(LParenLoc);
1165 PP.Diag(LParenLoc, diag::err_pp_expected_after) << II << tok::l_paren;
1166 // If the next token looks like a filename or the start of one,
1167 // assume it is and process it as such.
1168 if (!Tok.is(tok::angle_string_literal) && !Tok.is(tok::string_literal) &&
1172 // Save '(' location for possible missing ')' message.
1173 LParenLoc = Tok.getLocation();
1175 if (PP.getCurrentLexer()) {
1176 // Get the file name.
1177 PP.getCurrentLexer()->LexIncludeFilename(Tok);
1179 // We're in a macro, so we can't use LexIncludeFilename; just
1180 // grab the next token.
1185 // Reserve a buffer to get the spelling.
1186 SmallString<128> FilenameBuffer;
1188 SourceLocation EndLoc;
1190 switch (Tok.getKind()) {
1192 // If the token kind is EOD, the error has already been diagnosed.
1195 case tok::angle_string_literal:
1196 case tok::string_literal: {
1197 bool Invalid = false;
1198 Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid);
1205 // This could be a <foo/bar.h> file coming from a macro expansion. In this
1206 // case, glue the tokens together into FilenameBuffer and interpret those.
1207 FilenameBuffer.push_back('<');
1208 if (PP.ConcatenateIncludeName(FilenameBuffer, EndLoc)) {
1209 // Let the caller know a <eod> was found by changing the Token kind.
1210 Tok.setKind(tok::eod);
1211 return false; // Found <eod> but no ">"? Diagnostic already emitted.
1213 Filename = FilenameBuffer;
1216 PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename);
1220 SourceLocation FilenameLoc = Tok.getLocation();
1223 PP.LexNonComment(Tok);
1225 // Ensure we have a trailing ).
1226 if (Tok.isNot(tok::r_paren)) {
1227 PP.Diag(PP.getLocForEndOfToken(FilenameLoc), diag::err_pp_expected_after)
1228 << II << tok::r_paren;
1229 PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1233 bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename);
1234 // If GetIncludeFilenameSpelling set the start ptr to null, there was an
1236 if (Filename.empty())
1239 // Search include directories.
1240 const DirectoryLookup *CurDir;
1241 const FileEntry *File =
1242 PP.LookupFile(FilenameLoc, Filename, isAngled, LookupFrom, LookupFromFile,
1243 CurDir, nullptr, nullptr, nullptr, nullptr);
1245 // Get the result value. A result of true means the file exists.
1246 return File != nullptr;
1249 /// EvaluateHasInclude - Process a '__has_include("path")' expression.
1250 /// Returns true if successful.
1251 static bool EvaluateHasInclude(Token &Tok, IdentifierInfo *II,
1253 return EvaluateHasIncludeCommon(Tok, II, PP, nullptr, nullptr);
1256 /// EvaluateHasIncludeNext - Process '__has_include_next("path")' expression.
1257 /// Returns true if successful.
1258 static bool EvaluateHasIncludeNext(Token &Tok,
1259 IdentifierInfo *II, Preprocessor &PP) {
1260 // __has_include_next is like __has_include, except that we start
1261 // searching after the current found directory. If we can't do this,
1262 // issue a diagnostic.
1263 // FIXME: Factor out duplication with
1264 // Preprocessor::HandleIncludeNextDirective.
1265 const DirectoryLookup *Lookup = PP.GetCurDirLookup();
1266 const FileEntry *LookupFromFile = nullptr;
1267 if (PP.isInPrimaryFile() && PP.getLangOpts().IsHeaderFile) {
1268 // If the main file is a header, then it's either for PCH/AST generation,
1269 // or libclang opened it. Either way, handle it as a normal include below
1270 // and do not complain about __has_include_next.
1271 } else if (PP.isInPrimaryFile()) {
1273 PP.Diag(Tok, diag::pp_include_next_in_primary);
1274 } else if (PP.getCurrentLexerSubmodule()) {
1275 // Start looking up in the directory *after* the one in which the current
1276 // file would be found, if any.
1277 assert(PP.getCurrentLexer() && "#include_next directive in macro?");
1278 LookupFromFile = PP.getCurrentLexer()->getFileEntry();
1280 } else if (!Lookup) {
1281 PP.Diag(Tok, diag::pp_include_next_absolute_path);
1283 // Start looking up in the next directory.
1287 return EvaluateHasIncludeCommon(Tok, II, PP, Lookup, LookupFromFile);
1290 /// Process single-argument builtin feature-like macros that return
1292 static void EvaluateFeatureLikeBuiltinMacro(llvm::raw_svector_ostream& OS,
1293 Token &Tok, IdentifierInfo *II,
1297 bool &HasLexedNextTok)> Op) {
1298 // Parse the initial '('.
1299 PP.LexUnexpandedToken(Tok);
1300 if (Tok.isNot(tok::l_paren)) {
1301 PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II
1304 // Provide a dummy '0' value on output stream to elide further errors.
1305 if (!Tok.isOneOf(tok::eof, tok::eod)) {
1307 Tok.setKind(tok::numeric_constant);
1312 unsigned ParenDepth = 1;
1313 SourceLocation LParenLoc = Tok.getLocation();
1314 llvm::Optional<int> Result;
1317 bool SuppressDiagnostic = false;
1319 // Parse next token.
1320 PP.LexUnexpandedToken(Tok);
1323 switch (Tok.getKind()) {
1326 // Don't provide even a dummy value if the eod or eof marker is
1327 // reached. Simply provide a diagnostic.
1328 PP.Diag(Tok.getLocation(), diag::err_unterm_macro_invoc);
1332 if (!SuppressDiagnostic) {
1333 PP.Diag(Tok.getLocation(), diag::err_too_many_args_in_macro_invoc);
1334 SuppressDiagnostic = true;
1340 if (Result.hasValue())
1342 if (!SuppressDiagnostic) {
1343 PP.Diag(Tok.getLocation(), diag::err_pp_nested_paren) << II;
1344 SuppressDiagnostic = true;
1349 if (--ParenDepth > 0)
1352 // The last ')' has been reached; return the value if one found or
1353 // a diagnostic and a dummy value.
1354 if (Result.hasValue())
1355 OS << Result.getValue();
1358 if (!SuppressDiagnostic)
1359 PP.Diag(Tok.getLocation(), diag::err_too_few_args_in_macro_invoc);
1361 Tok.setKind(tok::numeric_constant);
1365 // Parse the macro argument, if one not found so far.
1366 if (Result.hasValue())
1369 bool HasLexedNextToken = false;
1370 Result = Op(Tok, HasLexedNextToken);
1372 if (HasLexedNextToken)
1378 // Diagnose missing ')'.
1379 if (!SuppressDiagnostic) {
1380 if (auto Diag = PP.Diag(Tok.getLocation(), diag::err_pp_expected_after)) {
1381 if (IdentifierInfo *LastII = ResultTok.getIdentifierInfo())
1384 Diag << ResultTok.getKind();
1385 Diag << tok::r_paren << ResultTok.getLocation();
1387 PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1388 SuppressDiagnostic = true;
1393 /// Helper function to return the IdentifierInfo structure of a Token
1394 /// or generate a diagnostic if none available.
1395 static IdentifierInfo *ExpectFeatureIdentifierInfo(Token &Tok,
1399 if (!Tok.isAnnotation() && (II = Tok.getIdentifierInfo()))
1402 PP.Diag(Tok.getLocation(), DiagID);
1406 /// Implements the __is_target_arch builtin macro.
1407 static bool isTargetArch(const TargetInfo &TI, const IdentifierInfo *II) {
1408 std::string ArchName = II->getName().lower() + "--";
1409 llvm::Triple Arch(ArchName);
1410 const llvm::Triple &TT = TI.getTriple();
1412 // arm matches thumb or thumbv7. armv7 matches thumbv7.
1413 if ((Arch.getSubArch() == llvm::Triple::NoSubArch ||
1414 Arch.getSubArch() == TT.getSubArch()) &&
1415 ((TT.getArch() == llvm::Triple::thumb &&
1416 Arch.getArch() == llvm::Triple::arm) ||
1417 (TT.getArch() == llvm::Triple::thumbeb &&
1418 Arch.getArch() == llvm::Triple::armeb)))
1421 // Check the parsed arch when it has no sub arch to allow Clang to
1422 // match thumb to thumbv7 but to prohibit matching thumbv6 to thumbv7.
1423 return (Arch.getSubArch() == llvm::Triple::NoSubArch ||
1424 Arch.getSubArch() == TT.getSubArch()) &&
1425 Arch.getArch() == TT.getArch();
1428 /// Implements the __is_target_vendor builtin macro.
1429 static bool isTargetVendor(const TargetInfo &TI, const IdentifierInfo *II) {
1430 StringRef VendorName = TI.getTriple().getVendorName();
1431 if (VendorName.empty())
1432 VendorName = "unknown";
1433 return VendorName.equals_lower(II->getName());
1436 /// Implements the __is_target_os builtin macro.
1437 static bool isTargetOS(const TargetInfo &TI, const IdentifierInfo *II) {
1438 std::string OSName =
1439 (llvm::Twine("unknown-unknown-") + II->getName().lower()).str();
1440 llvm::Triple OS(OSName);
1441 if (OS.getOS() == llvm::Triple::Darwin) {
1442 // Darwin matches macos, ios, etc.
1443 return TI.getTriple().isOSDarwin();
1445 return TI.getTriple().getOS() == OS.getOS();
1448 /// Implements the __is_target_environment builtin macro.
1449 static bool isTargetEnvironment(const TargetInfo &TI,
1450 const IdentifierInfo *II) {
1451 std::string EnvName = (llvm::Twine("---") + II->getName().lower()).str();
1452 llvm::Triple Env(EnvName);
1453 return TI.getTriple().getEnvironment() == Env.getEnvironment();
1456 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded
1457 /// as a builtin macro, handle it and return the next token as 'Tok'.
1458 void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
1459 // Figure out which token this is.
1460 IdentifierInfo *II = Tok.getIdentifierInfo();
1461 assert(II && "Can't be a macro without id info!");
1463 // If this is an _Pragma or Microsoft __pragma directive, expand it,
1464 // invoke the pragma handler, then lex the token after it.
1465 if (II == Ident_Pragma)
1466 return Handle_Pragma(Tok);
1467 else if (II == Ident__pragma) // in non-MS mode this is null
1468 return HandleMicrosoft__pragma(Tok);
1470 ++NumBuiltinMacroExpanded;
1472 SmallString<128> TmpBuffer;
1473 llvm::raw_svector_ostream OS(TmpBuffer);
1475 // Set up the return result.
1476 Tok.setIdentifierInfo(nullptr);
1477 Tok.clearFlag(Token::NeedsCleaning);
1479 if (II == Ident__LINE__) {
1480 // C99 6.10.8: "__LINE__: The presumed line number (within the current
1481 // source file) of the current source line (an integer constant)". This can
1482 // be affected by #line.
1483 SourceLocation Loc = Tok.getLocation();
1485 // Advance to the location of the first _, this might not be the first byte
1486 // of the token if it starts with an escaped newline.
1487 Loc = AdvanceToTokenCharacter(Loc, 0);
1489 // One wrinkle here is that GCC expands __LINE__ to location of the *end* of
1490 // a macro expansion. This doesn't matter for object-like macros, but
1491 // can matter for a function-like macro that expands to contain __LINE__.
1492 // Skip down through expansion points until we find a file loc for the
1493 // end of the expansion history.
1494 Loc = SourceMgr.getExpansionRange(Loc).getEnd();
1495 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Loc);
1497 // __LINE__ expands to a simple numeric value.
1498 OS << (PLoc.isValid()? PLoc.getLine() : 1);
1499 Tok.setKind(tok::numeric_constant);
1500 } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__) {
1501 // C99 6.10.8: "__FILE__: The presumed name of the current source file (a
1502 // character string literal)". This can be affected by #line.
1503 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1505 // __BASE_FILE__ is a GNU extension that returns the top of the presumed
1506 // #include stack instead of the current file.
1507 if (II == Ident__BASE_FILE__ && PLoc.isValid()) {
1508 SourceLocation NextLoc = PLoc.getIncludeLoc();
1509 while (NextLoc.isValid()) {
1510 PLoc = SourceMgr.getPresumedLoc(NextLoc);
1511 if (PLoc.isInvalid())
1514 NextLoc = PLoc.getIncludeLoc();
1518 // Escape this filename. Turn '\' -> '\\' '"' -> '\"'
1519 SmallString<128> FN;
1520 if (PLoc.isValid()) {
1521 FN += PLoc.getFilename();
1522 Lexer::Stringify(FN);
1523 OS << '"' << FN << '"';
1525 Tok.setKind(tok::string_literal);
1526 } else if (II == Ident__DATE__) {
1527 Diag(Tok.getLocation(), diag::warn_pp_date_time);
1528 if (!DATELoc.isValid())
1529 ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1530 Tok.setKind(tok::string_literal);
1531 Tok.setLength(strlen("\"Mmm dd yyyy\""));
1532 Tok.setLocation(SourceMgr.createExpansionLoc(DATELoc, Tok.getLocation(),
1536 } else if (II == Ident__TIME__) {
1537 Diag(Tok.getLocation(), diag::warn_pp_date_time);
1538 if (!TIMELoc.isValid())
1539 ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1540 Tok.setKind(tok::string_literal);
1541 Tok.setLength(strlen("\"hh:mm:ss\""));
1542 Tok.setLocation(SourceMgr.createExpansionLoc(TIMELoc, Tok.getLocation(),
1546 } else if (II == Ident__INCLUDE_LEVEL__) {
1547 // Compute the presumed include depth of this token. This can be affected
1548 // by GNU line markers.
1551 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1552 if (PLoc.isValid()) {
1553 PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1554 for (; PLoc.isValid(); ++Depth)
1555 PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1558 // __INCLUDE_LEVEL__ expands to a simple numeric value.
1560 Tok.setKind(tok::numeric_constant);
1561 } else if (II == Ident__TIMESTAMP__) {
1562 Diag(Tok.getLocation(), diag::warn_pp_date_time);
1563 // MSVC, ICC, GCC, VisualAge C++ extension. The generated string should be
1564 // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime.
1566 // Get the file that we are lexing out of. If we're currently lexing from
1567 // a macro, dig into the include stack.
1568 const FileEntry *CurFile = nullptr;
1569 PreprocessorLexer *TheLexer = getCurrentFileLexer();
1572 CurFile = SourceMgr.getFileEntryForID(TheLexer->getFileID());
1576 time_t TT = CurFile->getModificationTime();
1577 struct tm *TM = localtime(&TT);
1578 Result = asctime(TM);
1580 Result = "??? ??? ?? ??:??:?? ????\n";
1582 // Surround the string with " and strip the trailing newline.
1583 OS << '"' << StringRef(Result).drop_back() << '"';
1584 Tok.setKind(tok::string_literal);
1585 } else if (II == Ident__COUNTER__) {
1586 // __COUNTER__ expands to a simple numeric value.
1587 OS << CounterValue++;
1588 Tok.setKind(tok::numeric_constant);
1589 } else if (II == Ident__has_feature) {
1590 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1591 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1592 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1593 diag::err_feature_check_malformed);
1594 return II && HasFeature(*this, II->getName());
1596 } else if (II == Ident__has_extension) {
1597 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1598 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1599 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1600 diag::err_feature_check_malformed);
1601 return II && HasExtension(*this, II->getName());
1603 } else if (II == Ident__has_builtin) {
1604 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1605 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1606 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1607 diag::err_feature_check_malformed);
1608 const LangOptions &LangOpts = getLangOpts();
1611 else if (II->getBuiltinID() != 0) {
1612 switch (II->getBuiltinID()) {
1613 case Builtin::BI__builtin_operator_new:
1614 case Builtin::BI__builtin_operator_delete:
1615 // denotes date of behavior change to support calling arbitrary
1616 // usual allocation and deallocation functions. Required by libc++
1623 return llvm::StringSwitch<bool>(II->getName())
1624 .Case("__make_integer_seq", LangOpts.CPlusPlus)
1625 .Case("__type_pack_element", LangOpts.CPlusPlus)
1626 .Case("__builtin_available", true)
1627 .Case("__is_target_arch", true)
1628 .Case("__is_target_vendor", true)
1629 .Case("__is_target_os", true)
1630 .Case("__is_target_environment", true)
1634 } else if (II == Ident__is_identifier) {
1635 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1636 [](Token &Tok, bool &HasLexedNextToken) -> int {
1637 return Tok.is(tok::identifier);
1639 } else if (II == Ident__has_attribute) {
1640 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1641 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1642 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1643 diag::err_feature_check_malformed);
1644 return II ? hasAttribute(AttrSyntax::GNU, nullptr, II,
1645 getTargetInfo(), getLangOpts()) : 0;
1647 } else if (II == Ident__has_declspec) {
1648 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1649 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1650 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1651 diag::err_feature_check_malformed);
1652 return II ? hasAttribute(AttrSyntax::Declspec, nullptr, II,
1653 getTargetInfo(), getLangOpts()) : 0;
1655 } else if (II == Ident__has_cpp_attribute ||
1656 II == Ident__has_c_attribute) {
1657 bool IsCXX = II == Ident__has_cpp_attribute;
1658 EvaluateFeatureLikeBuiltinMacro(
1659 OS, Tok, II, *this, [&](Token &Tok, bool &HasLexedNextToken) -> int {
1660 IdentifierInfo *ScopeII = nullptr;
1661 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1662 Tok, *this, diag::err_feature_check_malformed);
1666 // It is possible to receive a scope token. Read the "::", if it is
1667 // available, and the subsequent identifier.
1668 LexUnexpandedToken(Tok);
1669 if (Tok.isNot(tok::coloncolon))
1670 HasLexedNextToken = true;
1673 LexUnexpandedToken(Tok);
1674 II = ExpectFeatureIdentifierInfo(Tok, *this,
1675 diag::err_feature_check_malformed);
1678 AttrSyntax Syntax = IsCXX ? AttrSyntax::CXX : AttrSyntax::C;
1679 return II ? hasAttribute(Syntax, ScopeII, II, getTargetInfo(),
1683 } else if (II == Ident__has_include ||
1684 II == Ident__has_include_next) {
1685 // The argument to these two builtins should be a parenthesized
1686 // file name string literal using angle brackets (<>) or
1687 // double-quotes ("").
1689 if (II == Ident__has_include)
1690 Value = EvaluateHasInclude(Tok, II, *this);
1692 Value = EvaluateHasIncludeNext(Tok, II, *this);
1694 if (Tok.isNot(tok::r_paren))
1697 Tok.setKind(tok::numeric_constant);
1698 } else if (II == Ident__has_warning) {
1699 // The argument should be a parenthesized string literal.
1700 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1701 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1702 std::string WarningName;
1703 SourceLocation StrStartLoc = Tok.getLocation();
1705 HasLexedNextToken = Tok.is(tok::string_literal);
1706 if (!FinishLexStringLiteral(Tok, WarningName, "'__has_warning'",
1707 /*MacroExpansion=*/false))
1710 // FIXME: Should we accept "-R..." flags here, or should that be
1711 // handled by a separate __has_remark?
1712 if (WarningName.size() < 3 || WarningName[0] != '-' ||
1713 WarningName[1] != 'W') {
1714 Diag(StrStartLoc, diag::warn_has_warning_invalid_option);
1718 // Finally, check if the warning flags maps to a diagnostic group.
1719 // We construct a SmallVector here to talk to getDiagnosticIDs().
1720 // Although we don't use the result, this isn't a hot path, and not
1721 // worth special casing.
1722 SmallVector<diag::kind, 10> Diags;
1723 return !getDiagnostics().getDiagnosticIDs()->
1724 getDiagnosticsInGroup(diag::Flavor::WarningOrError,
1725 WarningName.substr(2), Diags);
1727 } else if (II == Ident__building_module) {
1728 // The argument to this builtin should be an identifier. The
1729 // builtin evaluates to 1 when that identifier names the module we are
1730 // currently building.
1731 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1732 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1733 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1734 diag::err_expected_id_building_module);
1735 return getLangOpts().isCompilingModule() && II &&
1736 (II->getName() == getLangOpts().CurrentModule);
1738 } else if (II == Ident__MODULE__) {
1739 // The current module as an identifier.
1740 OS << getLangOpts().CurrentModule;
1741 IdentifierInfo *ModuleII = getIdentifierInfo(getLangOpts().CurrentModule);
1742 Tok.setIdentifierInfo(ModuleII);
1743 Tok.setKind(ModuleII->getTokenID());
1744 } else if (II == Ident__identifier) {
1745 SourceLocation Loc = Tok.getLocation();
1747 // We're expecting '__identifier' '(' identifier ')'. Try to recover
1748 // if the parens are missing.
1750 if (Tok.isNot(tok::l_paren)) {
1751 // No '(', use end of last token.
1752 Diag(getLocForEndOfToken(Loc), diag::err_pp_expected_after)
1753 << II << tok::l_paren;
1754 // If the next token isn't valid as our argument, we can't recover.
1755 if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1756 Tok.setKind(tok::identifier);
1760 SourceLocation LParenLoc = Tok.getLocation();
1763 if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1764 Tok.setKind(tok::identifier);
1766 Diag(Tok.getLocation(), diag::err_pp_identifier_arg_not_identifier)
1768 // Don't walk past anything that's not a real token.
1769 if (Tok.isOneOf(tok::eof, tok::eod) || Tok.isAnnotation())
1773 // Discard the ')', preserving 'Tok' as our result.
1775 LexNonComment(RParen);
1776 if (RParen.isNot(tok::r_paren)) {
1777 Diag(getLocForEndOfToken(Tok.getLocation()), diag::err_pp_expected_after)
1778 << Tok.getKind() << tok::r_paren;
1779 Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1782 } else if (II == Ident__is_target_arch) {
1783 EvaluateFeatureLikeBuiltinMacro(
1784 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1785 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1786 Tok, *this, diag::err_feature_check_malformed);
1787 return II && isTargetArch(getTargetInfo(), II);
1789 } else if (II == Ident__is_target_vendor) {
1790 EvaluateFeatureLikeBuiltinMacro(
1791 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1792 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1793 Tok, *this, diag::err_feature_check_malformed);
1794 return II && isTargetVendor(getTargetInfo(), II);
1796 } else if (II == Ident__is_target_os) {
1797 EvaluateFeatureLikeBuiltinMacro(
1798 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1799 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1800 Tok, *this, diag::err_feature_check_malformed);
1801 return II && isTargetOS(getTargetInfo(), II);
1803 } else if (II == Ident__is_target_environment) {
1804 EvaluateFeatureLikeBuiltinMacro(
1805 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1806 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1807 Tok, *this, diag::err_feature_check_malformed);
1808 return II && isTargetEnvironment(getTargetInfo(), II);
1811 llvm_unreachable("Unknown identifier!");
1813 CreateString(OS.str(), Tok, Tok.getLocation(), Tok.getLocation());
1816 void Preprocessor::markMacroAsUsed(MacroInfo *MI) {
1817 // If the 'used' status changed, and the macro requires 'unused' warning,
1818 // remove its SourceLocation from the warn-for-unused-macro locations.
1819 if (MI->isWarnIfUnused() && !MI->isUsed())
1820 WarnUnusedMacroLocs.erase(MI->getDefinitionLoc());
1821 MI->setIsUsed(true);