1 //===--- PPMacroExpansion.cpp - Top level Macro Expansion -----------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements the top level handling of macro expansion for the
10 // preprocessor.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "clang/Basic/AttributeCommonInfo.h"
15 #include "clang/Basic/Attributes.h"
16 #include "clang/Basic/Builtins.h"
17 #include "clang/Basic/FileManager.h"
18 #include "clang/Basic/IdentifierTable.h"
19 #include "clang/Basic/LLVM.h"
20 #include "clang/Basic/LangOptions.h"
21 #include "clang/Basic/ObjCRuntime.h"
22 #include "clang/Basic/SourceLocation.h"
23 #include "clang/Basic/TargetInfo.h"
24 #include "clang/Lex/CodeCompletionHandler.h"
25 #include "clang/Lex/DirectoryLookup.h"
26 #include "clang/Lex/ExternalPreprocessorSource.h"
27 #include "clang/Lex/HeaderSearch.h"
28 #include "clang/Lex/LexDiagnostic.h"
29 #include "clang/Lex/LiteralSupport.h"
30 #include "clang/Lex/MacroArgs.h"
31 #include "clang/Lex/MacroInfo.h"
32 #include "clang/Lex/Preprocessor.h"
33 #include "clang/Lex/PreprocessorLexer.h"
34 #include "clang/Lex/PreprocessorOptions.h"
35 #include "clang/Lex/Token.h"
36 #include "llvm/ADT/ArrayRef.h"
37 #include "llvm/ADT/DenseMap.h"
38 #include "llvm/ADT/DenseSet.h"
39 #include "llvm/ADT/FoldingSet.h"
40 #include "llvm/ADT/STLExtras.h"
41 #include "llvm/ADT/SmallString.h"
42 #include "llvm/ADT/SmallVector.h"
43 #include "llvm/ADT/StringRef.h"
44 #include "llvm/ADT/StringSwitch.h"
45 #include "llvm/Support/Casting.h"
46 #include "llvm/Support/ErrorHandling.h"
47 #include "llvm/Support/Format.h"
48 #include "llvm/Support/Path.h"
49 #include "llvm/Support/raw_ostream.h"
50 #include <algorithm>
51 #include <cassert>
52 #include <cstddef>
53 #include <cstring>
54 #include <ctime>
55 #include <optional>
56 #include <string>
57 #include <tuple>
58 #include <utility>
59 
60 using namespace clang;
61 
62 MacroDirective *
getLocalMacroDirectiveHistory(const IdentifierInfo * II) const63 Preprocessor::getLocalMacroDirectiveHistory(const IdentifierInfo *II) const {
64   if (!II->hadMacroDefinition())
65     return nullptr;
66   auto Pos = CurSubmoduleState->Macros.find(II);
67   return Pos == CurSubmoduleState->Macros.end() ? nullptr
68                                                 : Pos->second.getLatest();
69 }
70 
appendMacroDirective(IdentifierInfo * II,MacroDirective * MD)71 void Preprocessor::appendMacroDirective(IdentifierInfo *II, MacroDirective *MD){
72   assert(MD && "MacroDirective should be non-zero!");
73   assert(!MD->getPrevious() && "Already attached to a MacroDirective history.");
74 
75   MacroState &StoredMD = CurSubmoduleState->Macros[II];
76   auto *OldMD = StoredMD.getLatest();
77   MD->setPrevious(OldMD);
78   StoredMD.setLatest(MD);
79   StoredMD.overrideActiveModuleMacros(*this, II);
80 
81   if (needModuleMacros()) {
82     // Track that we created a new macro directive, so we know we should
83     // consider building a ModuleMacro for it when we get to the end of
84     // the module.
85     PendingModuleMacroNames.push_back(II);
86   }
87 
88   // Set up the identifier as having associated macro history.
89   II->setHasMacroDefinition(true);
90   if (!MD->isDefined() && !LeafModuleMacros.contains(II))
91     II->setHasMacroDefinition(false);
92   if (II->isFromAST())
93     II->setChangedSinceDeserialization();
94 }
95 
setLoadedMacroDirective(IdentifierInfo * II,MacroDirective * ED,MacroDirective * MD)96 void Preprocessor::setLoadedMacroDirective(IdentifierInfo *II,
97                                            MacroDirective *ED,
98                                            MacroDirective *MD) {
99   // Normally, when a macro is defined, it goes through appendMacroDirective()
100   // above, which chains a macro to previous defines, undefs, etc.
101   // However, in a pch, the whole macro history up to the end of the pch is
102   // stored, so ASTReader goes through this function instead.
103   // However, built-in macros are already registered in the Preprocessor
104   // ctor, and ASTWriter stops writing the macro chain at built-in macros,
105   // so in that case the chain from the pch needs to be spliced to the existing
106   // built-in.
107 
108   assert(II && MD);
109   MacroState &StoredMD = CurSubmoduleState->Macros[II];
110 
111   if (auto *OldMD = StoredMD.getLatest()) {
112     // shouldIgnoreMacro() in ASTWriter also stops at macros from the
113     // predefines buffer in module builds. However, in module builds, modules
114     // are loaded completely before predefines are processed, so StoredMD
115     // will be nullptr for them when they're loaded. StoredMD should only be
116     // non-nullptr for builtins read from a pch file.
117     assert(OldMD->getMacroInfo()->isBuiltinMacro() &&
118            "only built-ins should have an entry here");
119     assert(!OldMD->getPrevious() && "builtin should only have a single entry");
120     ED->setPrevious(OldMD);
121     StoredMD.setLatest(MD);
122   } else {
123     StoredMD = MD;
124   }
125 
126   // Setup the identifier as having associated macro history.
127   II->setHasMacroDefinition(true);
128   if (!MD->isDefined() && !LeafModuleMacros.contains(II))
129     II->setHasMacroDefinition(false);
130 }
131 
addModuleMacro(Module * Mod,IdentifierInfo * II,MacroInfo * Macro,ArrayRef<ModuleMacro * > Overrides,bool & New)132 ModuleMacro *Preprocessor::addModuleMacro(Module *Mod, IdentifierInfo *II,
133                                           MacroInfo *Macro,
134                                           ArrayRef<ModuleMacro *> Overrides,
135                                           bool &New) {
136   llvm::FoldingSetNodeID ID;
137   ModuleMacro::Profile(ID, Mod, II);
138 
139   void *InsertPos;
140   if (auto *MM = ModuleMacros.FindNodeOrInsertPos(ID, InsertPos)) {
141     New = false;
142     return MM;
143   }
144 
145   auto *MM = ModuleMacro::create(*this, Mod, II, Macro, Overrides);
146   ModuleMacros.InsertNode(MM, InsertPos);
147 
148   // Each overridden macro is now overridden by one more macro.
149   bool HidAny = false;
150   for (auto *O : Overrides) {
151     HidAny |= (O->NumOverriddenBy == 0);
152     ++O->NumOverriddenBy;
153   }
154 
155   // If we were the first overrider for any macro, it's no longer a leaf.
156   auto &LeafMacros = LeafModuleMacros[II];
157   if (HidAny) {
158     llvm::erase_if(LeafMacros,
159                    [](ModuleMacro *MM) { return MM->NumOverriddenBy != 0; });
160   }
161 
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);
166 
167   New = true;
168   return MM;
169 }
170 
getModuleMacro(Module * Mod,const IdentifierInfo * II)171 ModuleMacro *Preprocessor::getModuleMacro(Module *Mod,
172                                           const IdentifierInfo *II) {
173   llvm::FoldingSetNodeID ID;
174   ModuleMacro::Profile(ID, Mod, II);
175 
176   void *InsertPos;
177   return ModuleMacros.FindNodeOrInsertPos(ID, InsertPos);
178 }
179 
updateModuleMacroInfo(const IdentifierInfo * II,ModuleMacroInfo & Info)180 void Preprocessor::updateModuleMacroInfo(const IdentifierInfo *II,
181                                          ModuleMacroInfo &Info) {
182   assert(Info.ActiveModuleMacrosGeneration !=
183              CurSubmoduleState->VisibleModules.getGeneration() &&
184          "don't need to update this macro name info");
185   Info.ActiveModuleMacrosGeneration =
186       CurSubmoduleState->VisibleModules.getGeneration();
187 
188   auto Leaf = LeafModuleMacros.find(II);
189   if (Leaf == LeafModuleMacros.end()) {
190     // No imported macros at all: nothing to do.
191     return;
192   }
193 
194   Info.ActiveModuleMacros.clear();
195 
196   // Every macro that's locally overridden is overridden by a visible macro.
197   llvm::DenseMap<ModuleMacro *, int> NumHiddenOverrides;
198   for (auto *O : Info.OverriddenMacros)
199     NumHiddenOverrides[O] = -1;
200 
201   // Collect all macros that are not overridden by a visible macro.
202   llvm::SmallVector<ModuleMacro *, 16> Worklist;
203   for (auto *LeafMM : Leaf->second) {
204     assert(LeafMM->getNumOverridingMacros() == 0 && "leaf macro overridden");
205     if (NumHiddenOverrides.lookup(LeafMM) == 0)
206       Worklist.push_back(LeafMM);
207   }
208   while (!Worklist.empty()) {
209     auto *MM = Worklist.pop_back_val();
210     if (CurSubmoduleState->VisibleModules.isVisible(MM->getOwningModule())) {
211       // We only care about collecting definitions; undefinitions only act
212       // to override other definitions.
213       if (MM->getMacroInfo())
214         Info.ActiveModuleMacros.push_back(MM);
215     } else {
216       for (auto *O : MM->overrides())
217         if ((unsigned)++NumHiddenOverrides[O] == O->getNumOverridingMacros())
218           Worklist.push_back(O);
219     }
220   }
221   // Our reverse postorder walk found the macros in reverse order.
222   std::reverse(Info.ActiveModuleMacros.begin(), Info.ActiveModuleMacros.end());
223 
224   // Determine whether the macro name is ambiguous.
225   MacroInfo *MI = nullptr;
226   bool IsSystemMacro = true;
227   bool IsAmbiguous = false;
228   if (auto *MD = Info.MD) {
229     while (MD && isa<VisibilityMacroDirective>(MD))
230       MD = MD->getPrevious();
231     if (auto *DMD = dyn_cast_or_null<DefMacroDirective>(MD)) {
232       MI = DMD->getInfo();
233       IsSystemMacro &= SourceMgr.isInSystemHeader(DMD->getLocation());
234     }
235   }
236   for (auto *Active : Info.ActiveModuleMacros) {
237     auto *NewMI = Active->getMacroInfo();
238 
239     // Before marking the macro as ambiguous, check if this is a case where
240     // both macros are in system headers. If so, we trust that the system
241     // did not get it wrong. This also handles cases where Clang's own
242     // headers have a different spelling of certain system macros:
243     //   #define LONG_MAX __LONG_MAX__ (clang's limits.h)
244     //   #define LONG_MAX 0x7fffffffffffffffL (system's limits.h)
245     //
246     // FIXME: Remove the defined-in-system-headers check. clang's limits.h
247     // overrides the system limits.h's macros, so there's no conflict here.
248     if (MI && NewMI != MI &&
249         !MI->isIdenticalTo(*NewMI, *this, /*Syntactically=*/true))
250       IsAmbiguous = true;
251     IsSystemMacro &= Active->getOwningModule()->IsSystem ||
252                      SourceMgr.isInSystemHeader(NewMI->getDefinitionLoc());
253     MI = NewMI;
254   }
255   Info.IsAmbiguous = IsAmbiguous && !IsSystemMacro;
256 }
257 
dumpMacroInfo(const IdentifierInfo * II)258 void Preprocessor::dumpMacroInfo(const IdentifierInfo *II) {
259   ArrayRef<ModuleMacro*> Leaf;
260   auto LeafIt = LeafModuleMacros.find(II);
261   if (LeafIt != LeafModuleMacros.end())
262     Leaf = LeafIt->second;
263   const MacroState *State = nullptr;
264   auto Pos = CurSubmoduleState->Macros.find(II);
265   if (Pos != CurSubmoduleState->Macros.end())
266     State = &Pos->second;
267 
268   llvm::errs() << "MacroState " << State << " " << II->getNameStart();
269   if (State && State->isAmbiguous(*this, II))
270     llvm::errs() << " ambiguous";
271   if (State && !State->getOverriddenMacros().empty()) {
272     llvm::errs() << " overrides";
273     for (auto *O : State->getOverriddenMacros())
274       llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
275   }
276   llvm::errs() << "\n";
277 
278   // Dump local macro directives.
279   for (auto *MD = State ? State->getLatest() : nullptr; MD;
280        MD = MD->getPrevious()) {
281     llvm::errs() << " ";
282     MD->dump();
283   }
284 
285   // Dump module macros.
286   llvm::DenseSet<ModuleMacro*> Active;
287   for (auto *MM :
288        State ? State->getActiveModuleMacros(*this, II) : std::nullopt)
289     Active.insert(MM);
290   llvm::DenseSet<ModuleMacro*> Visited;
291   llvm::SmallVector<ModuleMacro *, 16> Worklist(Leaf.begin(), Leaf.end());
292   while (!Worklist.empty()) {
293     auto *MM = Worklist.pop_back_val();
294     llvm::errs() << " ModuleMacro " << MM << " "
295                  << MM->getOwningModule()->getFullModuleName();
296     if (!MM->getMacroInfo())
297       llvm::errs() << " undef";
298 
299     if (Active.count(MM))
300       llvm::errs() << " active";
301     else if (!CurSubmoduleState->VisibleModules.isVisible(
302                  MM->getOwningModule()))
303       llvm::errs() << " hidden";
304     else if (MM->getMacroInfo())
305       llvm::errs() << " overridden";
306 
307     if (!MM->overrides().empty()) {
308       llvm::errs() << " overrides";
309       for (auto *O : MM->overrides()) {
310         llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
311         if (Visited.insert(O).second)
312           Worklist.push_back(O);
313       }
314     }
315     llvm::errs() << "\n";
316     if (auto *MI = MM->getMacroInfo()) {
317       llvm::errs() << "  ";
318       MI->dump();
319       llvm::errs() << "\n";
320     }
321   }
322 }
323 
324 /// RegisterBuiltinMacro - Register the specified identifier in the identifier
325 /// table and mark it as a builtin macro to be expanded.
RegisterBuiltinMacro(Preprocessor & PP,const char * Name)326 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){
327   // Get the identifier.
328   IdentifierInfo *Id = PP.getIdentifierInfo(Name);
329 
330   // Mark it as being a macro that is builtin.
331   MacroInfo *MI = PP.AllocateMacroInfo(SourceLocation());
332   MI->setIsBuiltinMacro();
333   PP.appendDefMacroDirective(Id, MI);
334   return Id;
335 }
336 
337 /// RegisterBuiltinMacros - Register builtin macros, such as __LINE__ with the
338 /// identifier table.
RegisterBuiltinMacros()339 void Preprocessor::RegisterBuiltinMacros() {
340   Ident__LINE__ = RegisterBuiltinMacro(*this, "__LINE__");
341   Ident__FILE__ = RegisterBuiltinMacro(*this, "__FILE__");
342   Ident__DATE__ = RegisterBuiltinMacro(*this, "__DATE__");
343   Ident__TIME__ = RegisterBuiltinMacro(*this, "__TIME__");
344   Ident__COUNTER__ = RegisterBuiltinMacro(*this, "__COUNTER__");
345   Ident_Pragma  = RegisterBuiltinMacro(*this, "_Pragma");
346   Ident__FLT_EVAL_METHOD__ = RegisterBuiltinMacro(*this, "__FLT_EVAL_METHOD__");
347 
348   // C++ Standing Document Extensions.
349   if (getLangOpts().CPlusPlus)
350     Ident__has_cpp_attribute =
351         RegisterBuiltinMacro(*this, "__has_cpp_attribute");
352   else
353     Ident__has_cpp_attribute = nullptr;
354 
355   // GCC Extensions.
356   Ident__BASE_FILE__     = RegisterBuiltinMacro(*this, "__BASE_FILE__");
357   Ident__INCLUDE_LEVEL__ = RegisterBuiltinMacro(*this, "__INCLUDE_LEVEL__");
358   Ident__TIMESTAMP__     = RegisterBuiltinMacro(*this, "__TIMESTAMP__");
359 
360   // Microsoft Extensions.
361   if (getLangOpts().MicrosoftExt) {
362     Ident__identifier = RegisterBuiltinMacro(*this, "__identifier");
363     Ident__pragma = RegisterBuiltinMacro(*this, "__pragma");
364   } else {
365     Ident__identifier = nullptr;
366     Ident__pragma = nullptr;
367   }
368 
369   // Clang Extensions.
370   Ident__FILE_NAME__      = RegisterBuiltinMacro(*this, "__FILE_NAME__");
371   Ident__has_feature      = RegisterBuiltinMacro(*this, "__has_feature");
372   Ident__has_extension    = RegisterBuiltinMacro(*this, "__has_extension");
373   Ident__has_builtin      = RegisterBuiltinMacro(*this, "__has_builtin");
374   Ident__has_constexpr_builtin =
375       RegisterBuiltinMacro(*this, "__has_constexpr_builtin");
376   Ident__has_attribute    = RegisterBuiltinMacro(*this, "__has_attribute");
377   if (!getLangOpts().CPlusPlus)
378     Ident__has_c_attribute = RegisterBuiltinMacro(*this, "__has_c_attribute");
379   else
380     Ident__has_c_attribute = nullptr;
381 
382   Ident__has_declspec = RegisterBuiltinMacro(*this, "__has_declspec_attribute");
383   Ident__has_include      = RegisterBuiltinMacro(*this, "__has_include");
384   Ident__has_include_next = RegisterBuiltinMacro(*this, "__has_include_next");
385   Ident__has_warning      = RegisterBuiltinMacro(*this, "__has_warning");
386   Ident__is_identifier    = RegisterBuiltinMacro(*this, "__is_identifier");
387   Ident__is_target_arch   = RegisterBuiltinMacro(*this, "__is_target_arch");
388   Ident__is_target_vendor = RegisterBuiltinMacro(*this, "__is_target_vendor");
389   Ident__is_target_os     = RegisterBuiltinMacro(*this, "__is_target_os");
390   Ident__is_target_environment =
391       RegisterBuiltinMacro(*this, "__is_target_environment");
392   Ident__is_target_variant_os =
393       RegisterBuiltinMacro(*this, "__is_target_variant_os");
394   Ident__is_target_variant_environment =
395       RegisterBuiltinMacro(*this, "__is_target_variant_environment");
396 
397   // Modules.
398   Ident__building_module  = RegisterBuiltinMacro(*this, "__building_module");
399   if (!getLangOpts().CurrentModule.empty())
400     Ident__MODULE__ = RegisterBuiltinMacro(*this, "__MODULE__");
401   else
402     Ident__MODULE__ = nullptr;
403 }
404 
405 /// isTrivialSingleTokenExpansion - Return true if MI, which has a single token
406 /// in its expansion, currently expands to that token literally.
isTrivialSingleTokenExpansion(const MacroInfo * MI,const IdentifierInfo * MacroIdent,Preprocessor & PP)407 static bool isTrivialSingleTokenExpansion(const MacroInfo *MI,
408                                           const IdentifierInfo *MacroIdent,
409                                           Preprocessor &PP) {
410   IdentifierInfo *II = MI->getReplacementToken(0).getIdentifierInfo();
411 
412   // If the token isn't an identifier, it's always literally expanded.
413   if (!II) return true;
414 
415   // If the information about this identifier is out of date, update it from
416   // the external source.
417   if (II->isOutOfDate())
418     PP.getExternalSource()->updateOutOfDateIdentifier(*II);
419 
420   // If the identifier is a macro, and if that macro is enabled, it may be
421   // expanded so it's not a trivial expansion.
422   if (auto *ExpansionMI = PP.getMacroInfo(II))
423     if (ExpansionMI->isEnabled() &&
424         // Fast expanding "#define X X" is ok, because X would be disabled.
425         II != MacroIdent)
426       return false;
427 
428   // If this is an object-like macro invocation, it is safe to trivially expand
429   // it.
430   if (MI->isObjectLike()) return true;
431 
432   // If this is a function-like macro invocation, it's safe to trivially expand
433   // as long as the identifier is not a macro argument.
434   return !llvm::is_contained(MI->params(), II);
435 }
436 
437 /// isNextPPTokenLParen - Determine whether the next preprocessor token to be
438 /// lexed is a '('.  If so, consume the token and return true, if not, this
439 /// method should have no observable side-effect on the lexed tokens.
isNextPPTokenLParen()440 bool Preprocessor::isNextPPTokenLParen() {
441   // Do some quick tests for rejection cases.
442   unsigned Val;
443   if (CurLexer)
444     Val = CurLexer->isNextPPTokenLParen();
445   else
446     Val = CurTokenLexer->isNextTokenLParen();
447 
448   if (Val == 2) {
449     // We have run off the end.  If it's a source file we don't
450     // examine enclosing ones (C99 5.1.1.2p4).  Otherwise walk up the
451     // macro stack.
452     if (CurPPLexer)
453       return false;
454     for (const IncludeStackInfo &Entry : llvm::reverse(IncludeMacroStack)) {
455       if (Entry.TheLexer)
456         Val = Entry.TheLexer->isNextPPTokenLParen();
457       else
458         Val = Entry.TheTokenLexer->isNextTokenLParen();
459 
460       if (Val != 2)
461         break;
462 
463       // Ran off the end of a source file?
464       if (Entry.ThePPLexer)
465         return false;
466     }
467   }
468 
469   // Okay, if we know that the token is a '(', lex it and return.  Otherwise we
470   // have found something that isn't a '(' or we found the end of the
471   // translation unit.  In either case, return false.
472   return Val == 1;
473 }
474 
475 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be
476 /// expanded as a macro, handle it and return the next token as 'Identifier'.
HandleMacroExpandedIdentifier(Token & Identifier,const MacroDefinition & M)477 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
478                                                  const MacroDefinition &M) {
479   emitMacroExpansionWarnings(Identifier);
480 
481   MacroInfo *MI = M.getMacroInfo();
482 
483   // If this is a macro expansion in the "#if !defined(x)" line for the file,
484   // then the macro could expand to different things in other contexts, we need
485   // to disable the optimization in this case.
486   if (CurPPLexer) CurPPLexer->MIOpt.ExpandedMacro();
487 
488   // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially.
489   if (MI->isBuiltinMacro()) {
490     if (Callbacks)
491       Callbacks->MacroExpands(Identifier, M, Identifier.getLocation(),
492                               /*Args=*/nullptr);
493     ExpandBuiltinMacro(Identifier);
494     return true;
495   }
496 
497   /// Args - If this is a function-like macro expansion, this contains,
498   /// for each macro argument, the list of tokens that were provided to the
499   /// invocation.
500   MacroArgs *Args = nullptr;
501 
502   // Remember where the end of the expansion occurred.  For an object-like
503   // macro, this is the identifier.  For a function-like macro, this is the ')'.
504   SourceLocation ExpansionEnd = Identifier.getLocation();
505 
506   // If this is a function-like macro, read the arguments.
507   if (MI->isFunctionLike()) {
508     // Remember that we are now parsing the arguments to a macro invocation.
509     // Preprocessor directives used inside macro arguments are not portable, and
510     // this enables the warning.
511     InMacroArgs = true;
512     ArgMacro = &Identifier;
513 
514     Args = ReadMacroCallArgumentList(Identifier, MI, ExpansionEnd);
515 
516     // Finished parsing args.
517     InMacroArgs = false;
518     ArgMacro = nullptr;
519 
520     // If there was an error parsing the arguments, bail out.
521     if (!Args) return true;
522 
523     ++NumFnMacroExpanded;
524   } else {
525     ++NumMacroExpanded;
526   }
527 
528   // Notice that this macro has been used.
529   markMacroAsUsed(MI);
530 
531   // Remember where the token is expanded.
532   SourceLocation ExpandLoc = Identifier.getLocation();
533   SourceRange ExpansionRange(ExpandLoc, ExpansionEnd);
534 
535   if (Callbacks) {
536     if (InMacroArgs) {
537       // We can have macro expansion inside a conditional directive while
538       // reading the function macro arguments. To ensure, in that case, that
539       // MacroExpands callbacks still happen in source order, queue this
540       // callback to have it happen after the function macro callback.
541       DelayedMacroExpandsCallbacks.push_back(
542           MacroExpandsInfo(Identifier, M, ExpansionRange));
543     } else {
544       Callbacks->MacroExpands(Identifier, M, ExpansionRange, Args);
545       if (!DelayedMacroExpandsCallbacks.empty()) {
546         for (const MacroExpandsInfo &Info : DelayedMacroExpandsCallbacks) {
547           // FIXME: We lose macro args info with delayed callback.
548           Callbacks->MacroExpands(Info.Tok, Info.MD, Info.Range,
549                                   /*Args=*/nullptr);
550         }
551         DelayedMacroExpandsCallbacks.clear();
552       }
553     }
554   }
555 
556   // If the macro definition is ambiguous, complain.
557   if (M.isAmbiguous()) {
558     Diag(Identifier, diag::warn_pp_ambiguous_macro)
559       << Identifier.getIdentifierInfo();
560     Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen)
561       << Identifier.getIdentifierInfo();
562     M.forAllDefinitions([&](const MacroInfo *OtherMI) {
563       if (OtherMI != MI)
564         Diag(OtherMI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_other)
565           << Identifier.getIdentifierInfo();
566     });
567   }
568 
569   // If we started lexing a macro, enter the macro expansion body.
570 
571   // If this macro expands to no tokens, don't bother to push it onto the
572   // expansion stack, only to take it right back off.
573   if (MI->getNumTokens() == 0) {
574     // No need for arg info.
575     if (Args) Args->destroy(*this);
576 
577     // Propagate whitespace info as if we had pushed, then popped,
578     // a macro context.
579     Identifier.setFlag(Token::LeadingEmptyMacro);
580     PropagateLineStartLeadingSpaceInfo(Identifier);
581     ++NumFastMacroExpanded;
582     return false;
583   } else if (MI->getNumTokens() == 1 &&
584              isTrivialSingleTokenExpansion(MI, Identifier.getIdentifierInfo(),
585                                            *this)) {
586     // Otherwise, if this macro expands into a single trivially-expanded
587     // token: expand it now.  This handles common cases like
588     // "#define VAL 42".
589 
590     // No need for arg info.
591     if (Args) Args->destroy(*this);
592 
593     // Propagate the isAtStartOfLine/hasLeadingSpace markers of the macro
594     // identifier to the expanded token.
595     bool isAtStartOfLine = Identifier.isAtStartOfLine();
596     bool hasLeadingSpace = Identifier.hasLeadingSpace();
597 
598     // Replace the result token.
599     Identifier = MI->getReplacementToken(0);
600 
601     // Restore the StartOfLine/LeadingSpace markers.
602     Identifier.setFlagValue(Token::StartOfLine , isAtStartOfLine);
603     Identifier.setFlagValue(Token::LeadingSpace, hasLeadingSpace);
604 
605     // Update the tokens location to include both its expansion and physical
606     // locations.
607     SourceLocation Loc =
608       SourceMgr.createExpansionLoc(Identifier.getLocation(), ExpandLoc,
609                                    ExpansionEnd,Identifier.getLength());
610     Identifier.setLocation(Loc);
611 
612     // If this is a disabled macro or #define X X, we must mark the result as
613     // unexpandable.
614     if (IdentifierInfo *NewII = Identifier.getIdentifierInfo()) {
615       if (MacroInfo *NewMI = getMacroInfo(NewII))
616         if (!NewMI->isEnabled() || NewMI == MI) {
617           Identifier.setFlag(Token::DisableExpand);
618           // Don't warn for "#define X X" like "#define bool bool" from
619           // stdbool.h.
620           if (NewMI != MI || MI->isFunctionLike())
621             Diag(Identifier, diag::pp_disabled_macro_expansion);
622         }
623     }
624 
625     // Since this is not an identifier token, it can't be macro expanded, so
626     // we're done.
627     ++NumFastMacroExpanded;
628     return true;
629   }
630 
631   // Start expanding the macro.
632   EnterMacro(Identifier, ExpansionEnd, MI, Args);
633   return false;
634 }
635 
636 enum Bracket {
637   Brace,
638   Paren
639 };
640 
641 /// CheckMatchedBrackets - Returns true if the braces and parentheses in the
642 /// token vector are properly nested.
CheckMatchedBrackets(const SmallVectorImpl<Token> & Tokens)643 static bool CheckMatchedBrackets(const SmallVectorImpl<Token> &Tokens) {
644   SmallVector<Bracket, 8> Brackets;
645   for (SmallVectorImpl<Token>::const_iterator I = Tokens.begin(),
646                                               E = Tokens.end();
647        I != E; ++I) {
648     if (I->is(tok::l_paren)) {
649       Brackets.push_back(Paren);
650     } else if (I->is(tok::r_paren)) {
651       if (Brackets.empty() || Brackets.back() == Brace)
652         return false;
653       Brackets.pop_back();
654     } else if (I->is(tok::l_brace)) {
655       Brackets.push_back(Brace);
656     } else if (I->is(tok::r_brace)) {
657       if (Brackets.empty() || Brackets.back() == Paren)
658         return false;
659       Brackets.pop_back();
660     }
661   }
662   return Brackets.empty();
663 }
664 
665 /// GenerateNewArgTokens - Returns true if OldTokens can be converted to a new
666 /// vector of tokens in NewTokens.  The new number of arguments will be placed
667 /// in NumArgs and the ranges which need to surrounded in parentheses will be
668 /// in ParenHints.
669 /// Returns false if the token stream cannot be changed.  If this is because
670 /// of an initializer list starting a macro argument, the range of those
671 /// initializer lists will be place in InitLists.
GenerateNewArgTokens(Preprocessor & PP,SmallVectorImpl<Token> & OldTokens,SmallVectorImpl<Token> & NewTokens,unsigned & NumArgs,SmallVectorImpl<SourceRange> & ParenHints,SmallVectorImpl<SourceRange> & InitLists)672 static bool GenerateNewArgTokens(Preprocessor &PP,
673                                  SmallVectorImpl<Token> &OldTokens,
674                                  SmallVectorImpl<Token> &NewTokens,
675                                  unsigned &NumArgs,
676                                  SmallVectorImpl<SourceRange> &ParenHints,
677                                  SmallVectorImpl<SourceRange> &InitLists) {
678   if (!CheckMatchedBrackets(OldTokens))
679     return false;
680 
681   // Once it is known that the brackets are matched, only a simple count of the
682   // braces is needed.
683   unsigned Braces = 0;
684 
685   // First token of a new macro argument.
686   SmallVectorImpl<Token>::iterator ArgStartIterator = OldTokens.begin();
687 
688   // First closing brace in a new macro argument.  Used to generate
689   // SourceRanges for InitLists.
690   SmallVectorImpl<Token>::iterator ClosingBrace = OldTokens.end();
691   NumArgs = 0;
692   Token TempToken;
693   // Set to true when a macro separator token is found inside a braced list.
694   // If true, the fixed argument spans multiple old arguments and ParenHints
695   // will be updated.
696   bool FoundSeparatorToken = false;
697   for (SmallVectorImpl<Token>::iterator I = OldTokens.begin(),
698                                         E = OldTokens.end();
699        I != E; ++I) {
700     if (I->is(tok::l_brace)) {
701       ++Braces;
702     } else if (I->is(tok::r_brace)) {
703       --Braces;
704       if (Braces == 0 && ClosingBrace == E && FoundSeparatorToken)
705         ClosingBrace = I;
706     } else if (I->is(tok::eof)) {
707       // EOF token is used to separate macro arguments
708       if (Braces != 0) {
709         // Assume comma separator is actually braced list separator and change
710         // it back to a comma.
711         FoundSeparatorToken = true;
712         I->setKind(tok::comma);
713         I->setLength(1);
714       } else { // Braces == 0
715         // Separator token still separates arguments.
716         ++NumArgs;
717 
718         // If the argument starts with a brace, it can't be fixed with
719         // parentheses.  A different diagnostic will be given.
720         if (FoundSeparatorToken && ArgStartIterator->is(tok::l_brace)) {
721           InitLists.push_back(
722               SourceRange(ArgStartIterator->getLocation(),
723                           PP.getLocForEndOfToken(ClosingBrace->getLocation())));
724           ClosingBrace = E;
725         }
726 
727         // Add left paren
728         if (FoundSeparatorToken) {
729           TempToken.startToken();
730           TempToken.setKind(tok::l_paren);
731           TempToken.setLocation(ArgStartIterator->getLocation());
732           TempToken.setLength(0);
733           NewTokens.push_back(TempToken);
734         }
735 
736         // Copy over argument tokens
737         NewTokens.insert(NewTokens.end(), ArgStartIterator, I);
738 
739         // Add right paren and store the paren locations in ParenHints
740         if (FoundSeparatorToken) {
741           SourceLocation Loc = PP.getLocForEndOfToken((I - 1)->getLocation());
742           TempToken.startToken();
743           TempToken.setKind(tok::r_paren);
744           TempToken.setLocation(Loc);
745           TempToken.setLength(0);
746           NewTokens.push_back(TempToken);
747           ParenHints.push_back(SourceRange(ArgStartIterator->getLocation(),
748                                            Loc));
749         }
750 
751         // Copy separator token
752         NewTokens.push_back(*I);
753 
754         // Reset values
755         ArgStartIterator = I + 1;
756         FoundSeparatorToken = false;
757       }
758     }
759   }
760 
761   return !ParenHints.empty() && InitLists.empty();
762 }
763 
764 /// ReadFunctionLikeMacroArgs - After reading "MACRO" and knowing that the next
765 /// token is the '(' of the macro, this method is invoked to read all of the
766 /// actual arguments specified for the macro invocation.  This returns null on
767 /// error.
ReadMacroCallArgumentList(Token & MacroName,MacroInfo * MI,SourceLocation & MacroEnd)768 MacroArgs *Preprocessor::ReadMacroCallArgumentList(Token &MacroName,
769                                                    MacroInfo *MI,
770                                                    SourceLocation &MacroEnd) {
771   // The number of fixed arguments to parse.
772   unsigned NumFixedArgsLeft = MI->getNumParams();
773   bool isVariadic = MI->isVariadic();
774 
775   // Outer loop, while there are more arguments, keep reading them.
776   Token Tok;
777 
778   // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
779   // an argument value in a macro could expand to ',' or '(' or ')'.
780   LexUnexpandedToken(Tok);
781   assert(Tok.is(tok::l_paren) && "Error computing l-paren-ness?");
782 
783   // ArgTokens - Build up a list of tokens that make up each argument.  Each
784   // argument is separated by an EOF token.  Use a SmallVector so we can avoid
785   // heap allocations in the common case.
786   SmallVector<Token, 64> ArgTokens;
787   bool ContainsCodeCompletionTok = false;
788   bool FoundElidedComma = false;
789 
790   SourceLocation TooManyArgsLoc;
791 
792   unsigned NumActuals = 0;
793   while (Tok.isNot(tok::r_paren)) {
794     if (ContainsCodeCompletionTok && Tok.isOneOf(tok::eof, tok::eod))
795       break;
796 
797     assert(Tok.isOneOf(tok::l_paren, tok::comma) &&
798            "only expect argument separators here");
799 
800     size_t ArgTokenStart = ArgTokens.size();
801     SourceLocation ArgStartLoc = Tok.getLocation();
802 
803     // C99 6.10.3p11: Keep track of the number of l_parens we have seen.  Note
804     // that we already consumed the first one.
805     unsigned NumParens = 0;
806 
807     while (true) {
808       // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
809       // an argument value in a macro could expand to ',' or '(' or ')'.
810       LexUnexpandedToken(Tok);
811 
812       if (Tok.isOneOf(tok::eof, tok::eod)) { // "#if f(<eof>" & "#if f(\n"
813         if (!ContainsCodeCompletionTok) {
814           Diag(MacroName, diag::err_unterm_macro_invoc);
815           Diag(MI->getDefinitionLoc(), diag::note_macro_here)
816             << MacroName.getIdentifierInfo();
817           // Do not lose the EOF/EOD.  Return it to the client.
818           MacroName = Tok;
819           return nullptr;
820         }
821         // Do not lose the EOF/EOD.
822         auto Toks = std::make_unique<Token[]>(1);
823         Toks[0] = Tok;
824         EnterTokenStream(std::move(Toks), 1, true, /*IsReinject*/ false);
825         break;
826       } else if (Tok.is(tok::r_paren)) {
827         // If we found the ) token, the macro arg list is done.
828         if (NumParens-- == 0) {
829           MacroEnd = Tok.getLocation();
830           if (!ArgTokens.empty() &&
831               ArgTokens.back().commaAfterElided()) {
832             FoundElidedComma = true;
833           }
834           break;
835         }
836       } else if (Tok.is(tok::l_paren)) {
837         ++NumParens;
838       } else if (Tok.is(tok::comma)) {
839         // In Microsoft-compatibility mode, single commas from nested macro
840         // expansions should not be considered as argument separators. We test
841         // for this with the IgnoredComma token flag.
842         if (Tok.getFlags() & Token::IgnoredComma) {
843           // However, in MSVC's preprocessor, subsequent expansions do treat
844           // these commas as argument separators. This leads to a common
845           // workaround used in macros that need to work in both MSVC and
846           // compliant preprocessors. Therefore, the IgnoredComma flag can only
847           // apply once to any given token.
848           Tok.clearFlag(Token::IgnoredComma);
849         } else if (NumParens == 0) {
850           // Comma ends this argument if there are more fixed arguments
851           // expected. However, if this is a variadic macro, and this is part of
852           // the variadic part, then the comma is just an argument token.
853           if (!isVariadic)
854             break;
855           if (NumFixedArgsLeft > 1)
856             break;
857         }
858       } else if (Tok.is(tok::comment) && !KeepMacroComments) {
859         // If this is a comment token in the argument list and we're just in
860         // -C mode (not -CC mode), discard the comment.
861         continue;
862       } else if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) {
863         // Reading macro arguments can cause macros that we are currently
864         // expanding from to be popped off the expansion stack.  Doing so causes
865         // them to be reenabled for expansion.  Here we record whether any
866         // identifiers we lex as macro arguments correspond to disabled macros.
867         // If so, we mark the token as noexpand.  This is a subtle aspect of
868         // C99 6.10.3.4p2.
869         if (MacroInfo *MI = getMacroInfo(Tok.getIdentifierInfo()))
870           if (!MI->isEnabled())
871             Tok.setFlag(Token::DisableExpand);
872       } else if (Tok.is(tok::code_completion)) {
873         ContainsCodeCompletionTok = true;
874         if (CodeComplete)
875           CodeComplete->CodeCompleteMacroArgument(MacroName.getIdentifierInfo(),
876                                                   MI, NumActuals);
877         // Don't mark that we reached the code-completion point because the
878         // parser is going to handle the token and there will be another
879         // code-completion callback.
880       }
881 
882       ArgTokens.push_back(Tok);
883     }
884 
885     // If this was an empty argument list foo(), don't add this as an empty
886     // argument.
887     if (ArgTokens.empty() && Tok.getKind() == tok::r_paren)
888       break;
889 
890     // If this is not a variadic macro, and too many args were specified, emit
891     // an error.
892     if (!isVariadic && NumFixedArgsLeft == 0 && TooManyArgsLoc.isInvalid()) {
893       if (ArgTokens.size() != ArgTokenStart)
894         TooManyArgsLoc = ArgTokens[ArgTokenStart].getLocation();
895       else
896         TooManyArgsLoc = ArgStartLoc;
897     }
898 
899     // Empty arguments are standard in C99 and C++0x, and are supported as an
900     // extension in other modes.
901     if (ArgTokens.size() == ArgTokenStart && !getLangOpts().C99)
902       Diag(Tok, getLangOpts().CPlusPlus11
903                     ? diag::warn_cxx98_compat_empty_fnmacro_arg
904                     : diag::ext_empty_fnmacro_arg);
905 
906     // Add a marker EOF token to the end of the token list for this argument.
907     Token EOFTok;
908     EOFTok.startToken();
909     EOFTok.setKind(tok::eof);
910     EOFTok.setLocation(Tok.getLocation());
911     EOFTok.setLength(0);
912     ArgTokens.push_back(EOFTok);
913     ++NumActuals;
914     if (!ContainsCodeCompletionTok && NumFixedArgsLeft != 0)
915       --NumFixedArgsLeft;
916   }
917 
918   // Okay, we either found the r_paren.  Check to see if we parsed too few
919   // arguments.
920   unsigned MinArgsExpected = MI->getNumParams();
921 
922   // If this is not a variadic macro, and too many args were specified, emit
923   // an error.
924   if (!isVariadic && NumActuals > MinArgsExpected &&
925       !ContainsCodeCompletionTok) {
926     // Emit the diagnostic at the macro name in case there is a missing ).
927     // Emitting it at the , could be far away from the macro name.
928     Diag(TooManyArgsLoc, diag::err_too_many_args_in_macro_invoc);
929     Diag(MI->getDefinitionLoc(), diag::note_macro_here)
930       << MacroName.getIdentifierInfo();
931 
932     // Commas from braced initializer lists will be treated as argument
933     // separators inside macros.  Attempt to correct for this with parentheses.
934     // TODO: See if this can be generalized to angle brackets for templates
935     // inside macro arguments.
936 
937     SmallVector<Token, 4> FixedArgTokens;
938     unsigned FixedNumArgs = 0;
939     SmallVector<SourceRange, 4> ParenHints, InitLists;
940     if (!GenerateNewArgTokens(*this, ArgTokens, FixedArgTokens, FixedNumArgs,
941                               ParenHints, InitLists)) {
942       if (!InitLists.empty()) {
943         DiagnosticBuilder DB =
944             Diag(MacroName,
945                  diag::note_init_list_at_beginning_of_macro_argument);
946         for (SourceRange Range : InitLists)
947           DB << Range;
948       }
949       return nullptr;
950     }
951     if (FixedNumArgs != MinArgsExpected)
952       return nullptr;
953 
954     DiagnosticBuilder DB = Diag(MacroName, diag::note_suggest_parens_for_macro);
955     for (SourceRange ParenLocation : ParenHints) {
956       DB << FixItHint::CreateInsertion(ParenLocation.getBegin(), "(");
957       DB << FixItHint::CreateInsertion(ParenLocation.getEnd(), ")");
958     }
959     ArgTokens.swap(FixedArgTokens);
960     NumActuals = FixedNumArgs;
961   }
962 
963   // See MacroArgs instance var for description of this.
964   bool isVarargsElided = false;
965 
966   if (ContainsCodeCompletionTok) {
967     // Recover from not-fully-formed macro invocation during code-completion.
968     Token EOFTok;
969     EOFTok.startToken();
970     EOFTok.setKind(tok::eof);
971     EOFTok.setLocation(Tok.getLocation());
972     EOFTok.setLength(0);
973     for (; NumActuals < MinArgsExpected; ++NumActuals)
974       ArgTokens.push_back(EOFTok);
975   }
976 
977   if (NumActuals < MinArgsExpected) {
978     // There are several cases where too few arguments is ok, handle them now.
979     if (NumActuals == 0 && MinArgsExpected == 1) {
980       // #define A(X)  or  #define A(...)   ---> A()
981 
982       // If there is exactly one argument, and that argument is missing,
983       // then we have an empty "()" argument empty list.  This is fine, even if
984       // the macro expects one argument (the argument is just empty).
985       isVarargsElided = MI->isVariadic();
986     } else if ((FoundElidedComma || MI->isVariadic()) &&
987                (NumActuals+1 == MinArgsExpected ||  // A(x, ...) -> A(X)
988                 (NumActuals == 0 && MinArgsExpected == 2))) {// A(x,...) -> A()
989       // Varargs where the named vararg parameter is missing: OK as extension.
990       //   #define A(x, ...)
991       //   A("blah")
992       //
993       // If the macro contains the comma pasting extension, the diagnostic
994       // is suppressed; we know we'll get another diagnostic later.
995       if (!MI->hasCommaPasting()) {
996         // C++20 allows this construct, but standards before C++20 and all C
997         // standards do not allow the construct (we allow it as an extension).
998         Diag(Tok, getLangOpts().CPlusPlus20
999                       ? diag::warn_cxx17_compat_missing_varargs_arg
1000                       : diag::ext_missing_varargs_arg);
1001         Diag(MI->getDefinitionLoc(), diag::note_macro_here)
1002           << MacroName.getIdentifierInfo();
1003       }
1004 
1005       // Remember this occurred, allowing us to elide the comma when used for
1006       // cases like:
1007       //   #define A(x, foo...) blah(a, ## foo)
1008       //   #define B(x, ...) blah(a, ## __VA_ARGS__)
1009       //   #define C(...) blah(a, ## __VA_ARGS__)
1010       //  A(x) B(x) C()
1011       isVarargsElided = true;
1012     } else if (!ContainsCodeCompletionTok) {
1013       // Otherwise, emit the error.
1014       Diag(Tok, diag::err_too_few_args_in_macro_invoc);
1015       Diag(MI->getDefinitionLoc(), diag::note_macro_here)
1016         << MacroName.getIdentifierInfo();
1017       return nullptr;
1018     }
1019 
1020     // Add a marker EOF token to the end of the token list for this argument.
1021     SourceLocation EndLoc = Tok.getLocation();
1022     Tok.startToken();
1023     Tok.setKind(tok::eof);
1024     Tok.setLocation(EndLoc);
1025     Tok.setLength(0);
1026     ArgTokens.push_back(Tok);
1027 
1028     // If we expect two arguments, add both as empty.
1029     if (NumActuals == 0 && MinArgsExpected == 2)
1030       ArgTokens.push_back(Tok);
1031 
1032   } else if (NumActuals > MinArgsExpected && !MI->isVariadic() &&
1033              !ContainsCodeCompletionTok) {
1034     // Emit the diagnostic at the macro name in case there is a missing ).
1035     // Emitting it at the , could be far away from the macro name.
1036     Diag(MacroName, diag::err_too_many_args_in_macro_invoc);
1037     Diag(MI->getDefinitionLoc(), diag::note_macro_here)
1038       << MacroName.getIdentifierInfo();
1039     return nullptr;
1040   }
1041 
1042   return MacroArgs::create(MI, ArgTokens, isVarargsElided, *this);
1043 }
1044 
1045 /// Keeps macro expanded tokens for TokenLexers.
1046 //
1047 /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
1048 /// going to lex in the cache and when it finishes the tokens are removed
1049 /// from the end of the cache.
cacheMacroExpandedTokens(TokenLexer * tokLexer,ArrayRef<Token> tokens)1050 Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer,
1051                                               ArrayRef<Token> tokens) {
1052   assert(tokLexer);
1053   if (tokens.empty())
1054     return nullptr;
1055 
1056   size_t newIndex = MacroExpandedTokens.size();
1057   bool cacheNeedsToGrow = tokens.size() >
1058                       MacroExpandedTokens.capacity()-MacroExpandedTokens.size();
1059   MacroExpandedTokens.append(tokens.begin(), tokens.end());
1060 
1061   if (cacheNeedsToGrow) {
1062     // Go through all the TokenLexers whose 'Tokens' pointer points in the
1063     // buffer and update the pointers to the (potential) new buffer array.
1064     for (const auto &Lexer : MacroExpandingLexersStack) {
1065       TokenLexer *prevLexer;
1066       size_t tokIndex;
1067       std::tie(prevLexer, tokIndex) = Lexer;
1068       prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex;
1069     }
1070   }
1071 
1072   MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex));
1073   return MacroExpandedTokens.data() + newIndex;
1074 }
1075 
removeCachedMacroExpandedTokensOfLastLexer()1076 void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() {
1077   assert(!MacroExpandingLexersStack.empty());
1078   size_t tokIndex = MacroExpandingLexersStack.back().second;
1079   assert(tokIndex < MacroExpandedTokens.size());
1080   // Pop the cached macro expanded tokens from the end.
1081   MacroExpandedTokens.resize(tokIndex);
1082   MacroExpandingLexersStack.pop_back();
1083 }
1084 
1085 /// ComputeDATE_TIME - Compute the current time, enter it into the specified
1086 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of
1087 /// the identifier tokens inserted.
ComputeDATE_TIME(SourceLocation & DATELoc,SourceLocation & TIMELoc,Preprocessor & PP)1088 static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc,
1089                              Preprocessor &PP) {
1090   time_t TT;
1091   std::tm *TM;
1092   if (PP.getPreprocessorOpts().SourceDateEpoch) {
1093     TT = *PP.getPreprocessorOpts().SourceDateEpoch;
1094     TM = std::gmtime(&TT);
1095   } else {
1096     TT = std::time(nullptr);
1097     TM = std::localtime(&TT);
1098   }
1099 
1100   static const char * const Months[] = {
1101     "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"
1102   };
1103 
1104   {
1105     SmallString<32> TmpBuffer;
1106     llvm::raw_svector_ostream TmpStream(TmpBuffer);
1107     if (TM)
1108       TmpStream << llvm::format("\"%s %2d %4d\"", Months[TM->tm_mon],
1109                                 TM->tm_mday, TM->tm_year + 1900);
1110     else
1111       TmpStream << "??? ?? ????";
1112     Token TmpTok;
1113     TmpTok.startToken();
1114     PP.CreateString(TmpStream.str(), TmpTok);
1115     DATELoc = TmpTok.getLocation();
1116   }
1117 
1118   {
1119     SmallString<32> TmpBuffer;
1120     llvm::raw_svector_ostream TmpStream(TmpBuffer);
1121     if (TM)
1122       TmpStream << llvm::format("\"%02d:%02d:%02d\"", TM->tm_hour, TM->tm_min,
1123                                 TM->tm_sec);
1124     else
1125       TmpStream << "??:??:??";
1126     Token TmpTok;
1127     TmpTok.startToken();
1128     PP.CreateString(TmpStream.str(), TmpTok);
1129     TIMELoc = TmpTok.getLocation();
1130   }
1131 }
1132 
1133 /// HasFeature - Return true if we recognize and implement the feature
1134 /// specified by the identifier as a standard language feature.
HasFeature(const Preprocessor & PP,StringRef Feature)1135 static bool HasFeature(const Preprocessor &PP, StringRef Feature) {
1136   const LangOptions &LangOpts = PP.getLangOpts();
1137 
1138   // Normalize the feature name, __foo__ becomes foo.
1139   if (Feature.starts_with("__") && Feature.ends_with("__") &&
1140       Feature.size() >= 4)
1141     Feature = Feature.substr(2, Feature.size() - 4);
1142 
1143 #define FEATURE(Name, Predicate) .Case(#Name, Predicate)
1144   return llvm::StringSwitch<bool>(Feature)
1145 #include "clang/Basic/Features.def"
1146       .Default(false);
1147 #undef FEATURE
1148 }
1149 
1150 /// HasExtension - Return true if we recognize and implement the feature
1151 /// specified by the identifier, either as an extension or a standard language
1152 /// feature.
HasExtension(const Preprocessor & PP,StringRef Extension)1153 static bool HasExtension(const Preprocessor &PP, StringRef Extension) {
1154   if (HasFeature(PP, Extension))
1155     return true;
1156 
1157   // If the use of an extension results in an error diagnostic, extensions are
1158   // effectively unavailable, so just return false here.
1159   if (PP.getDiagnostics().getExtensionHandlingBehavior() >=
1160       diag::Severity::Error)
1161     return false;
1162 
1163   const LangOptions &LangOpts = PP.getLangOpts();
1164 
1165   // Normalize the extension name, __foo__ becomes foo.
1166   if (Extension.starts_with("__") && Extension.ends_with("__") &&
1167       Extension.size() >= 4)
1168     Extension = Extension.substr(2, Extension.size() - 4);
1169 
1170     // Because we inherit the feature list from HasFeature, this string switch
1171     // must be less restrictive than HasFeature's.
1172 #define EXTENSION(Name, Predicate) .Case(#Name, Predicate)
1173   return llvm::StringSwitch<bool>(Extension)
1174 #include "clang/Basic/Features.def"
1175       .Default(false);
1176 #undef EXTENSION
1177 }
1178 
1179 /// EvaluateHasIncludeCommon - Process a '__has_include("path")'
1180 /// or '__has_include_next("path")' expression.
1181 /// Returns true if successful.
EvaluateHasIncludeCommon(Token & Tok,IdentifierInfo * II,Preprocessor & PP,ConstSearchDirIterator LookupFrom,const FileEntry * LookupFromFile)1182 static bool EvaluateHasIncludeCommon(Token &Tok, IdentifierInfo *II,
1183                                      Preprocessor &PP,
1184                                      ConstSearchDirIterator LookupFrom,
1185                                      const FileEntry *LookupFromFile) {
1186   // Save the location of the current token.  If a '(' is later found, use
1187   // that location.  If not, use the end of this location instead.
1188   SourceLocation LParenLoc = Tok.getLocation();
1189 
1190   // These expressions are only allowed within a preprocessor directive.
1191   if (!PP.isParsingIfOrElifDirective()) {
1192     PP.Diag(LParenLoc, diag::err_pp_directive_required) << II;
1193     // Return a valid identifier token.
1194     assert(Tok.is(tok::identifier));
1195     Tok.setIdentifierInfo(II);
1196     return false;
1197   }
1198 
1199   // Get '('. If we don't have a '(', try to form a header-name token.
1200   do {
1201     if (PP.LexHeaderName(Tok))
1202       return false;
1203   } while (Tok.getKind() == tok::comment);
1204 
1205   // Ensure we have a '('.
1206   if (Tok.isNot(tok::l_paren)) {
1207     // No '(', use end of last token.
1208     LParenLoc = PP.getLocForEndOfToken(LParenLoc);
1209     PP.Diag(LParenLoc, diag::err_pp_expected_after) << II << tok::l_paren;
1210     // If the next token looks like a filename or the start of one,
1211     // assume it is and process it as such.
1212     if (Tok.isNot(tok::header_name))
1213       return false;
1214   } else {
1215     // Save '(' location for possible missing ')' message.
1216     LParenLoc = Tok.getLocation();
1217     if (PP.LexHeaderName(Tok))
1218       return false;
1219   }
1220 
1221   if (Tok.isNot(tok::header_name)) {
1222     PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename);
1223     return false;
1224   }
1225 
1226   // Reserve a buffer to get the spelling.
1227   SmallString<128> FilenameBuffer;
1228   bool Invalid = false;
1229   StringRef Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid);
1230   if (Invalid)
1231     return false;
1232 
1233   SourceLocation FilenameLoc = Tok.getLocation();
1234 
1235   // Get ')'.
1236   PP.LexNonComment(Tok);
1237 
1238   // Ensure we have a trailing ).
1239   if (Tok.isNot(tok::r_paren)) {
1240     PP.Diag(PP.getLocForEndOfToken(FilenameLoc), diag::err_pp_expected_after)
1241         << II << tok::r_paren;
1242     PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1243     return false;
1244   }
1245 
1246   bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename);
1247   // If GetIncludeFilenameSpelling set the start ptr to null, there was an
1248   // error.
1249   if (Filename.empty())
1250     return false;
1251 
1252   // Passing this to LookupFile forces header search to check whether the found
1253   // file belongs to a module. Skipping that check could incorrectly mark
1254   // modular header as textual, causing issues down the line.
1255   ModuleMap::KnownHeader KH;
1256 
1257   // Search include directories.
1258   OptionalFileEntryRef File =
1259       PP.LookupFile(FilenameLoc, Filename, isAngled, LookupFrom, LookupFromFile,
1260                     nullptr, nullptr, nullptr, &KH, nullptr, nullptr);
1261 
1262   if (PPCallbacks *Callbacks = PP.getPPCallbacks()) {
1263     SrcMgr::CharacteristicKind FileType = SrcMgr::C_User;
1264     if (File)
1265       FileType = PP.getHeaderSearchInfo().getFileDirFlavor(*File);
1266     Callbacks->HasInclude(FilenameLoc, Filename, isAngled, File, FileType);
1267   }
1268 
1269   // Get the result value.  A result of true means the file exists.
1270   return File.has_value();
1271 }
1272 
EvaluateHasInclude(Token & Tok,IdentifierInfo * II)1273 bool Preprocessor::EvaluateHasInclude(Token &Tok, IdentifierInfo *II) {
1274   return EvaluateHasIncludeCommon(Tok, II, *this, nullptr, nullptr);
1275 }
1276 
EvaluateHasIncludeNext(Token & Tok,IdentifierInfo * II)1277 bool Preprocessor::EvaluateHasIncludeNext(Token &Tok, IdentifierInfo *II) {
1278   ConstSearchDirIterator Lookup = nullptr;
1279   const FileEntry *LookupFromFile;
1280   std::tie(Lookup, LookupFromFile) = getIncludeNextStart(Tok);
1281 
1282   return EvaluateHasIncludeCommon(Tok, II, *this, Lookup, LookupFromFile);
1283 }
1284 
1285 /// Process single-argument builtin feature-like macros that return
1286 /// integer values.
EvaluateFeatureLikeBuiltinMacro(llvm::raw_svector_ostream & OS,Token & Tok,IdentifierInfo * II,Preprocessor & PP,bool ExpandArgs,llvm::function_ref<int (Token & Tok,bool & HasLexedNextTok)> Op)1287 static void EvaluateFeatureLikeBuiltinMacro(llvm::raw_svector_ostream& OS,
1288                                             Token &Tok, IdentifierInfo *II,
1289                                             Preprocessor &PP, bool ExpandArgs,
1290                                             llvm::function_ref<
1291                                               int(Token &Tok,
1292                                                   bool &HasLexedNextTok)> Op) {
1293   // Parse the initial '('.
1294   PP.LexUnexpandedToken(Tok);
1295   if (Tok.isNot(tok::l_paren)) {
1296     PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II
1297                                                             << tok::l_paren;
1298 
1299     // Provide a dummy '0' value on output stream to elide further errors.
1300     if (!Tok.isOneOf(tok::eof, tok::eod)) {
1301       OS << 0;
1302       Tok.setKind(tok::numeric_constant);
1303     }
1304     return;
1305   }
1306 
1307   unsigned ParenDepth = 1;
1308   SourceLocation LParenLoc = Tok.getLocation();
1309   std::optional<int> Result;
1310 
1311   Token ResultTok;
1312   bool SuppressDiagnostic = false;
1313   while (true) {
1314     // Parse next token.
1315     if (ExpandArgs)
1316       PP.Lex(Tok);
1317     else
1318       PP.LexUnexpandedToken(Tok);
1319 
1320 already_lexed:
1321     switch (Tok.getKind()) {
1322       case tok::eof:
1323       case tok::eod:
1324         // Don't provide even a dummy value if the eod or eof marker is
1325         // reached.  Simply provide a diagnostic.
1326         PP.Diag(Tok.getLocation(), diag::err_unterm_macro_invoc);
1327         return;
1328 
1329       case tok::comma:
1330         if (!SuppressDiagnostic) {
1331           PP.Diag(Tok.getLocation(), diag::err_too_many_args_in_macro_invoc);
1332           SuppressDiagnostic = true;
1333         }
1334         continue;
1335 
1336       case tok::l_paren:
1337         ++ParenDepth;
1338         if (Result)
1339           break;
1340         if (!SuppressDiagnostic) {
1341           PP.Diag(Tok.getLocation(), diag::err_pp_nested_paren) << II;
1342           SuppressDiagnostic = true;
1343         }
1344         continue;
1345 
1346       case tok::r_paren:
1347         if (--ParenDepth > 0)
1348           continue;
1349 
1350         // The last ')' has been reached; return the value if one found or
1351         // a diagnostic and a dummy value.
1352         if (Result) {
1353           OS << *Result;
1354           // For strict conformance to __has_cpp_attribute rules, use 'L'
1355           // suffix for dated literals.
1356           if (*Result > 1)
1357             OS << 'L';
1358         } else {
1359           OS << 0;
1360           if (!SuppressDiagnostic)
1361             PP.Diag(Tok.getLocation(), diag::err_too_few_args_in_macro_invoc);
1362         }
1363         Tok.setKind(tok::numeric_constant);
1364         return;
1365 
1366       default: {
1367         // Parse the macro argument, if one not found so far.
1368         if (Result)
1369           break;
1370 
1371         bool HasLexedNextToken = false;
1372         Result = Op(Tok, HasLexedNextToken);
1373         ResultTok = Tok;
1374         if (HasLexedNextToken)
1375           goto already_lexed;
1376         continue;
1377       }
1378     }
1379 
1380     // Diagnose missing ')'.
1381     if (!SuppressDiagnostic) {
1382       if (auto Diag = PP.Diag(Tok.getLocation(), diag::err_pp_expected_after)) {
1383         if (IdentifierInfo *LastII = ResultTok.getIdentifierInfo())
1384           Diag << LastII;
1385         else
1386           Diag << ResultTok.getKind();
1387         Diag << tok::r_paren << ResultTok.getLocation();
1388       }
1389       PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1390       SuppressDiagnostic = true;
1391     }
1392   }
1393 }
1394 
1395 /// Helper function to return the IdentifierInfo structure of a Token
1396 /// or generate a diagnostic if none available.
ExpectFeatureIdentifierInfo(Token & Tok,Preprocessor & PP,signed DiagID)1397 static IdentifierInfo *ExpectFeatureIdentifierInfo(Token &Tok,
1398                                                    Preprocessor &PP,
1399                                                    signed DiagID) {
1400   IdentifierInfo *II;
1401   if (!Tok.isAnnotation() && (II = Tok.getIdentifierInfo()))
1402     return II;
1403 
1404   PP.Diag(Tok.getLocation(), DiagID);
1405   return nullptr;
1406 }
1407 
1408 /// Implements the __is_target_arch builtin macro.
isTargetArch(const TargetInfo & TI,const IdentifierInfo * II)1409 static bool isTargetArch(const TargetInfo &TI, const IdentifierInfo *II) {
1410   std::string ArchName = II->getName().lower() + "--";
1411   llvm::Triple Arch(ArchName);
1412   const llvm::Triple &TT = TI.getTriple();
1413   if (TT.isThumb()) {
1414     // arm matches thumb or thumbv7. armv7 matches thumbv7.
1415     if ((Arch.getSubArch() == llvm::Triple::NoSubArch ||
1416          Arch.getSubArch() == TT.getSubArch()) &&
1417         ((TT.getArch() == llvm::Triple::thumb &&
1418           Arch.getArch() == llvm::Triple::arm) ||
1419          (TT.getArch() == llvm::Triple::thumbeb &&
1420           Arch.getArch() == llvm::Triple::armeb)))
1421       return true;
1422   }
1423   // Check the parsed arch when it has no sub arch to allow Clang to
1424   // match thumb to thumbv7 but to prohibit matching thumbv6 to thumbv7.
1425   return (Arch.getSubArch() == llvm::Triple::NoSubArch ||
1426           Arch.getSubArch() == TT.getSubArch()) &&
1427          Arch.getArch() == TT.getArch();
1428 }
1429 
1430 /// Implements the __is_target_vendor builtin macro.
isTargetVendor(const TargetInfo & TI,const IdentifierInfo * II)1431 static bool isTargetVendor(const TargetInfo &TI, const IdentifierInfo *II) {
1432   StringRef VendorName = TI.getTriple().getVendorName();
1433   if (VendorName.empty())
1434     VendorName = "unknown";
1435   return VendorName.equals_insensitive(II->getName());
1436 }
1437 
1438 /// Implements the __is_target_os builtin macro.
isTargetOS(const TargetInfo & TI,const IdentifierInfo * II)1439 static bool isTargetOS(const TargetInfo &TI, const IdentifierInfo *II) {
1440   std::string OSName =
1441       (llvm::Twine("unknown-unknown-") + II->getName().lower()).str();
1442   llvm::Triple OS(OSName);
1443   if (OS.getOS() == llvm::Triple::Darwin) {
1444     // Darwin matches macos, ios, etc.
1445     return TI.getTriple().isOSDarwin();
1446   }
1447   return TI.getTriple().getOS() == OS.getOS();
1448 }
1449 
1450 /// Implements the __is_target_environment builtin macro.
isTargetEnvironment(const TargetInfo & TI,const IdentifierInfo * II)1451 static bool isTargetEnvironment(const TargetInfo &TI,
1452                                 const IdentifierInfo *II) {
1453   std::string EnvName = (llvm::Twine("---") + II->getName().lower()).str();
1454   llvm::Triple Env(EnvName);
1455   // The unknown environment is matched only if
1456   // '__is_target_environment(unknown)' is used.
1457   if (Env.getEnvironment() == llvm::Triple::UnknownEnvironment &&
1458       EnvName != "---unknown")
1459     return false;
1460   return TI.getTriple().getEnvironment() == Env.getEnvironment();
1461 }
1462 
1463 /// Implements the __is_target_variant_os builtin macro.
isTargetVariantOS(const TargetInfo & TI,const IdentifierInfo * II)1464 static bool isTargetVariantOS(const TargetInfo &TI, const IdentifierInfo *II) {
1465   if (TI.getTriple().isOSDarwin()) {
1466     const llvm::Triple *VariantTriple = TI.getDarwinTargetVariantTriple();
1467     if (!VariantTriple)
1468       return false;
1469 
1470     std::string OSName =
1471         (llvm::Twine("unknown-unknown-") + II->getName().lower()).str();
1472     llvm::Triple OS(OSName);
1473     if (OS.getOS() == llvm::Triple::Darwin) {
1474       // Darwin matches macos, ios, etc.
1475       return VariantTriple->isOSDarwin();
1476     }
1477     return VariantTriple->getOS() == OS.getOS();
1478   }
1479   return false;
1480 }
1481 
1482 /// Implements the __is_target_variant_environment builtin macro.
isTargetVariantEnvironment(const TargetInfo & TI,const IdentifierInfo * II)1483 static bool isTargetVariantEnvironment(const TargetInfo &TI,
1484                                 const IdentifierInfo *II) {
1485   if (TI.getTriple().isOSDarwin()) {
1486     const llvm::Triple *VariantTriple = TI.getDarwinTargetVariantTriple();
1487     if (!VariantTriple)
1488       return false;
1489     std::string EnvName = (llvm::Twine("---") + II->getName().lower()).str();
1490     llvm::Triple Env(EnvName);
1491     return VariantTriple->getEnvironment() == Env.getEnvironment();
1492   }
1493   return false;
1494 }
1495 
1496 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded
1497 /// as a builtin macro, handle it and return the next token as 'Tok'.
ExpandBuiltinMacro(Token & Tok)1498 void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
1499   // Figure out which token this is.
1500   IdentifierInfo *II = Tok.getIdentifierInfo();
1501   assert(II && "Can't be a macro without id info!");
1502 
1503   // If this is an _Pragma or Microsoft __pragma directive, expand it,
1504   // invoke the pragma handler, then lex the token after it.
1505   if (II == Ident_Pragma)
1506     return Handle_Pragma(Tok);
1507   else if (II == Ident__pragma) // in non-MS mode this is null
1508     return HandleMicrosoft__pragma(Tok);
1509 
1510   ++NumBuiltinMacroExpanded;
1511 
1512   SmallString<128> TmpBuffer;
1513   llvm::raw_svector_ostream OS(TmpBuffer);
1514 
1515   // Set up the return result.
1516   Tok.setIdentifierInfo(nullptr);
1517   Tok.clearFlag(Token::NeedsCleaning);
1518   bool IsAtStartOfLine = Tok.isAtStartOfLine();
1519   bool HasLeadingSpace = Tok.hasLeadingSpace();
1520 
1521   if (II == Ident__LINE__) {
1522     // C99 6.10.8: "__LINE__: The presumed line number (within the current
1523     // source file) of the current source line (an integer constant)".  This can
1524     // be affected by #line.
1525     SourceLocation Loc = Tok.getLocation();
1526 
1527     // Advance to the location of the first _, this might not be the first byte
1528     // of the token if it starts with an escaped newline.
1529     Loc = AdvanceToTokenCharacter(Loc, 0);
1530 
1531     // One wrinkle here is that GCC expands __LINE__ to location of the *end* of
1532     // a macro expansion.  This doesn't matter for object-like macros, but
1533     // can matter for a function-like macro that expands to contain __LINE__.
1534     // Skip down through expansion points until we find a file loc for the
1535     // end of the expansion history.
1536     Loc = SourceMgr.getExpansionRange(Loc).getEnd();
1537     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Loc);
1538 
1539     // __LINE__ expands to a simple numeric value.
1540     OS << (PLoc.isValid()? PLoc.getLine() : 1);
1541     Tok.setKind(tok::numeric_constant);
1542   } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__ ||
1543              II == Ident__FILE_NAME__) {
1544     // C99 6.10.8: "__FILE__: The presumed name of the current source file (a
1545     // character string literal)". This can be affected by #line.
1546     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1547 
1548     // __BASE_FILE__ is a GNU extension that returns the top of the presumed
1549     // #include stack instead of the current file.
1550     if (II == Ident__BASE_FILE__ && PLoc.isValid()) {
1551       SourceLocation NextLoc = PLoc.getIncludeLoc();
1552       while (NextLoc.isValid()) {
1553         PLoc = SourceMgr.getPresumedLoc(NextLoc);
1554         if (PLoc.isInvalid())
1555           break;
1556 
1557         NextLoc = PLoc.getIncludeLoc();
1558       }
1559     }
1560 
1561     // Escape this filename.  Turn '\' -> '\\' '"' -> '\"'
1562     SmallString<256> FN;
1563     if (PLoc.isValid()) {
1564       // __FILE_NAME__ is a Clang-specific extension that expands to the
1565       // the last part of __FILE__.
1566       if (II == Ident__FILE_NAME__) {
1567         processPathToFileName(FN, PLoc, getLangOpts(), getTargetInfo());
1568       } else {
1569         FN += PLoc.getFilename();
1570         processPathForFileMacro(FN, getLangOpts(), getTargetInfo());
1571       }
1572       Lexer::Stringify(FN);
1573       OS << '"' << FN << '"';
1574     }
1575     Tok.setKind(tok::string_literal);
1576   } else if (II == Ident__DATE__) {
1577     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1578     if (!DATELoc.isValid())
1579       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1580     Tok.setKind(tok::string_literal);
1581     Tok.setLength(strlen("\"Mmm dd yyyy\""));
1582     Tok.setLocation(SourceMgr.createExpansionLoc(DATELoc, Tok.getLocation(),
1583                                                  Tok.getLocation(),
1584                                                  Tok.getLength()));
1585     return;
1586   } else if (II == Ident__TIME__) {
1587     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1588     if (!TIMELoc.isValid())
1589       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1590     Tok.setKind(tok::string_literal);
1591     Tok.setLength(strlen("\"hh:mm:ss\""));
1592     Tok.setLocation(SourceMgr.createExpansionLoc(TIMELoc, Tok.getLocation(),
1593                                                  Tok.getLocation(),
1594                                                  Tok.getLength()));
1595     return;
1596   } else if (II == Ident__INCLUDE_LEVEL__) {
1597     // Compute the presumed include depth of this token.  This can be affected
1598     // by GNU line markers.
1599     unsigned Depth = 0;
1600 
1601     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1602     if (PLoc.isValid()) {
1603       PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1604       for (; PLoc.isValid(); ++Depth)
1605         PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1606     }
1607 
1608     // __INCLUDE_LEVEL__ expands to a simple numeric value.
1609     OS << Depth;
1610     Tok.setKind(tok::numeric_constant);
1611   } else if (II == Ident__TIMESTAMP__) {
1612     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1613     // MSVC, ICC, GCC, VisualAge C++ extension.  The generated string should be
1614     // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime.
1615     const char *Result;
1616     if (getPreprocessorOpts().SourceDateEpoch) {
1617       time_t TT = *getPreprocessorOpts().SourceDateEpoch;
1618       std::tm *TM = std::gmtime(&TT);
1619       Result = asctime(TM);
1620     } else {
1621       // Get the file that we are lexing out of.  If we're currently lexing from
1622       // a macro, dig into the include stack.
1623       const FileEntry *CurFile = nullptr;
1624       if (PreprocessorLexer *TheLexer = getCurrentFileLexer())
1625         CurFile = SourceMgr.getFileEntryForID(TheLexer->getFileID());
1626       if (CurFile) {
1627         time_t TT = CurFile->getModificationTime();
1628         struct tm *TM = localtime(&TT);
1629         Result = asctime(TM);
1630       } else {
1631         Result = "??? ??? ?? ??:??:?? ????\n";
1632       }
1633     }
1634     // Surround the string with " and strip the trailing newline.
1635     OS << '"' << StringRef(Result).drop_back() << '"';
1636     Tok.setKind(tok::string_literal);
1637   } else if (II == Ident__FLT_EVAL_METHOD__) {
1638     // __FLT_EVAL_METHOD__ is set to the default value.
1639     OS << getTUFPEvalMethod();
1640     // __FLT_EVAL_METHOD__ expands to a simple numeric value.
1641     Tok.setKind(tok::numeric_constant);
1642     if (getLastFPEvalPragmaLocation().isValid()) {
1643       // The program is ill-formed. The value of __FLT_EVAL_METHOD__ is altered
1644       // by the pragma.
1645       Diag(Tok, diag::err_illegal_use_of_flt_eval_macro);
1646       Diag(getLastFPEvalPragmaLocation(), diag::note_pragma_entered_here);
1647     }
1648   } else if (II == Ident__COUNTER__) {
1649     // __COUNTER__ expands to a simple numeric value.
1650     OS << CounterValue++;
1651     Tok.setKind(tok::numeric_constant);
1652   } else if (II == Ident__has_feature) {
1653     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, false,
1654       [this](Token &Tok, bool &HasLexedNextToken) -> int {
1655         IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1656                                            diag::err_feature_check_malformed);
1657         return II && HasFeature(*this, II->getName());
1658       });
1659   } else if (II == Ident__has_extension) {
1660     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, false,
1661       [this](Token &Tok, bool &HasLexedNextToken) -> int {
1662         IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1663                                            diag::err_feature_check_malformed);
1664         return II && HasExtension(*this, II->getName());
1665       });
1666   } else if (II == Ident__has_builtin) {
1667     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, false,
1668       [this](Token &Tok, bool &HasLexedNextToken) -> int {
1669         IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1670                                            diag::err_feature_check_malformed);
1671         if (!II)
1672           return false;
1673         else if (II->getBuiltinID() != 0) {
1674           switch (II->getBuiltinID()) {
1675           case Builtin::BI__builtin_operator_new:
1676           case Builtin::BI__builtin_operator_delete:
1677             // denotes date of behavior change to support calling arbitrary
1678             // usual allocation and deallocation functions. Required by libc++
1679             return 201802;
1680           default:
1681             return Builtin::evaluateRequiredTargetFeatures(
1682                 getBuiltinInfo().getRequiredFeatures(II->getBuiltinID()),
1683                 getTargetInfo().getTargetOpts().FeatureMap);
1684           }
1685           return true;
1686         } else if (II->getTokenID() != tok::identifier ||
1687                    II->hasRevertedTokenIDToIdentifier()) {
1688           // Treat all keywords that introduce a custom syntax of the form
1689           //
1690           //   '__some_keyword' '(' [...] ')'
1691           //
1692           // as being "builtin functions", even if the syntax isn't a valid
1693           // function call (for example, because the builtin takes a type
1694           // argument).
1695           if (II->getName().starts_with("__builtin_") ||
1696               II->getName().starts_with("__is_") ||
1697               II->getName().starts_with("__has_"))
1698             return true;
1699           return llvm::StringSwitch<bool>(II->getName())
1700               .Case("__array_rank", true)
1701               .Case("__array_extent", true)
1702               .Case("__reference_binds_to_temporary", true)
1703               .Case("__reference_constructs_from_temporary", true)
1704 #define TRANSFORM_TYPE_TRAIT_DEF(_, Trait) .Case("__" #Trait, true)
1705 #include "clang/Basic/TransformTypeTraits.def"
1706               .Default(false);
1707         } else {
1708           return llvm::StringSwitch<bool>(II->getName())
1709               // Report builtin templates as being builtins.
1710               .Case("__make_integer_seq", getLangOpts().CPlusPlus)
1711               .Case("__type_pack_element", getLangOpts().CPlusPlus)
1712               // Likewise for some builtin preprocessor macros.
1713               // FIXME: This is inconsistent; we usually suggest detecting
1714               // builtin macros via #ifdef. Don't add more cases here.
1715               .Case("__is_target_arch", true)
1716               .Case("__is_target_vendor", true)
1717               .Case("__is_target_os", true)
1718               .Case("__is_target_environment", true)
1719               .Case("__is_target_variant_os", true)
1720               .Case("__is_target_variant_environment", true)
1721               .Default(false);
1722         }
1723       });
1724   } else if (II == Ident__has_constexpr_builtin) {
1725     EvaluateFeatureLikeBuiltinMacro(
1726         OS, Tok, II, *this, false,
1727         [this](Token &Tok, bool &HasLexedNextToken) -> int {
1728           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1729               Tok, *this, diag::err_feature_check_malformed);
1730           if (!II)
1731             return false;
1732           unsigned BuiltinOp = II->getBuiltinID();
1733           return BuiltinOp != 0 &&
1734                  this->getBuiltinInfo().isConstantEvaluated(BuiltinOp);
1735         });
1736   } else if (II == Ident__is_identifier) {
1737     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, false,
1738       [](Token &Tok, bool &HasLexedNextToken) -> int {
1739         return Tok.is(tok::identifier);
1740       });
1741   } else if (II == Ident__has_attribute) {
1742     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, true,
1743       [this](Token &Tok, bool &HasLexedNextToken) -> int {
1744         IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1745                                            diag::err_feature_check_malformed);
1746         return II ? hasAttribute(AttributeCommonInfo::Syntax::AS_GNU, nullptr,
1747                                  II, getTargetInfo(), getLangOpts())
1748                   : 0;
1749       });
1750   } else if (II == Ident__has_declspec) {
1751     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, true,
1752       [this](Token &Tok, bool &HasLexedNextToken) -> int {
1753         IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1754                                            diag::err_feature_check_malformed);
1755         if (II) {
1756           const LangOptions &LangOpts = getLangOpts();
1757           return LangOpts.DeclSpecKeyword &&
1758                  hasAttribute(AttributeCommonInfo::Syntax::AS_Declspec, nullptr,
1759                               II, getTargetInfo(), LangOpts);
1760         }
1761 
1762         return false;
1763       });
1764   } else if (II == Ident__has_cpp_attribute ||
1765              II == Ident__has_c_attribute) {
1766     bool IsCXX = II == Ident__has_cpp_attribute;
1767     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, true,
1768         [&](Token &Tok, bool &HasLexedNextToken) -> int {
1769           IdentifierInfo *ScopeII = nullptr;
1770           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1771               Tok, *this, diag::err_feature_check_malformed);
1772           if (!II)
1773             return false;
1774 
1775           // It is possible to receive a scope token.  Read the "::", if it is
1776           // available, and the subsequent identifier.
1777           LexUnexpandedToken(Tok);
1778           if (Tok.isNot(tok::coloncolon))
1779             HasLexedNextToken = true;
1780           else {
1781             ScopeII = II;
1782             // Lex an expanded token for the attribute name.
1783             Lex(Tok);
1784             II = ExpectFeatureIdentifierInfo(Tok, *this,
1785                                              diag::err_feature_check_malformed);
1786           }
1787 
1788           AttributeCommonInfo::Syntax Syntax =
1789               IsCXX ? AttributeCommonInfo::Syntax::AS_CXX11
1790                     : AttributeCommonInfo::Syntax::AS_C23;
1791           return II ? hasAttribute(Syntax, ScopeII, II, getTargetInfo(),
1792                                    getLangOpts())
1793                     : 0;
1794         });
1795   } else if (II == Ident__has_include ||
1796              II == Ident__has_include_next) {
1797     // The argument to these two builtins should be a parenthesized
1798     // file name string literal using angle brackets (<>) or
1799     // double-quotes ("").
1800     bool Value;
1801     if (II == Ident__has_include)
1802       Value = EvaluateHasInclude(Tok, II);
1803     else
1804       Value = EvaluateHasIncludeNext(Tok, II);
1805 
1806     if (Tok.isNot(tok::r_paren))
1807       return;
1808     OS << (int)Value;
1809     Tok.setKind(tok::numeric_constant);
1810   } else if (II == Ident__has_warning) {
1811     // The argument should be a parenthesized string literal.
1812     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, false,
1813       [this](Token &Tok, bool &HasLexedNextToken) -> int {
1814         std::string WarningName;
1815         SourceLocation StrStartLoc = Tok.getLocation();
1816 
1817         HasLexedNextToken = Tok.is(tok::string_literal);
1818         if (!FinishLexStringLiteral(Tok, WarningName, "'__has_warning'",
1819                                     /*AllowMacroExpansion=*/false))
1820           return false;
1821 
1822         // FIXME: Should we accept "-R..." flags here, or should that be
1823         // handled by a separate __has_remark?
1824         if (WarningName.size() < 3 || WarningName[0] != '-' ||
1825             WarningName[1] != 'W') {
1826           Diag(StrStartLoc, diag::warn_has_warning_invalid_option);
1827           return false;
1828         }
1829 
1830         // Finally, check if the warning flags maps to a diagnostic group.
1831         // We construct a SmallVector here to talk to getDiagnosticIDs().
1832         // Although we don't use the result, this isn't a hot path, and not
1833         // worth special casing.
1834         SmallVector<diag::kind, 10> Diags;
1835         return !getDiagnostics().getDiagnosticIDs()->
1836                 getDiagnosticsInGroup(diag::Flavor::WarningOrError,
1837                                       WarningName.substr(2), Diags);
1838       });
1839   } else if (II == Ident__building_module) {
1840     // The argument to this builtin should be an identifier. The
1841     // builtin evaluates to 1 when that identifier names the module we are
1842     // currently building.
1843     EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this, false,
1844       [this](Token &Tok, bool &HasLexedNextToken) -> int {
1845         IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1846                                        diag::err_expected_id_building_module);
1847         return getLangOpts().isCompilingModule() && II &&
1848                (II->getName() == getLangOpts().CurrentModule);
1849       });
1850   } else if (II == Ident__MODULE__) {
1851     // The current module as an identifier.
1852     OS << getLangOpts().CurrentModule;
1853     IdentifierInfo *ModuleII = getIdentifierInfo(getLangOpts().CurrentModule);
1854     Tok.setIdentifierInfo(ModuleII);
1855     Tok.setKind(ModuleII->getTokenID());
1856   } else if (II == Ident__identifier) {
1857     SourceLocation Loc = Tok.getLocation();
1858 
1859     // We're expecting '__identifier' '(' identifier ')'. Try to recover
1860     // if the parens are missing.
1861     LexNonComment(Tok);
1862     if (Tok.isNot(tok::l_paren)) {
1863       // No '(', use end of last token.
1864       Diag(getLocForEndOfToken(Loc), diag::err_pp_expected_after)
1865         << II << tok::l_paren;
1866       // If the next token isn't valid as our argument, we can't recover.
1867       if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1868         Tok.setKind(tok::identifier);
1869       return;
1870     }
1871 
1872     SourceLocation LParenLoc = Tok.getLocation();
1873     LexNonComment(Tok);
1874 
1875     if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1876       Tok.setKind(tok::identifier);
1877     else if (Tok.is(tok::string_literal) && !Tok.hasUDSuffix()) {
1878       StringLiteralParser Literal(Tok, *this,
1879                                   StringLiteralEvalMethod::Unevaluated);
1880       if (Literal.hadError)
1881         return;
1882 
1883       Tok.setIdentifierInfo(getIdentifierInfo(Literal.GetString()));
1884       Tok.setKind(tok::identifier);
1885     } else {
1886       Diag(Tok.getLocation(), diag::err_pp_identifier_arg_not_identifier)
1887         << Tok.getKind();
1888       // Don't walk past anything that's not a real token.
1889       if (Tok.isOneOf(tok::eof, tok::eod) || Tok.isAnnotation())
1890         return;
1891     }
1892 
1893     // Discard the ')', preserving 'Tok' as our result.
1894     Token RParen;
1895     LexNonComment(RParen);
1896     if (RParen.isNot(tok::r_paren)) {
1897       Diag(getLocForEndOfToken(Tok.getLocation()), diag::err_pp_expected_after)
1898         << Tok.getKind() << tok::r_paren;
1899       Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1900     }
1901     return;
1902   } else if (II == Ident__is_target_arch) {
1903     EvaluateFeatureLikeBuiltinMacro(
1904         OS, Tok, II, *this, false,
1905         [this](Token &Tok, bool &HasLexedNextToken) -> int {
1906           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1907               Tok, *this, diag::err_feature_check_malformed);
1908           return II && isTargetArch(getTargetInfo(), II);
1909         });
1910   } else if (II == Ident__is_target_vendor) {
1911     EvaluateFeatureLikeBuiltinMacro(
1912         OS, Tok, II, *this, false,
1913         [this](Token &Tok, bool &HasLexedNextToken) -> int {
1914           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1915               Tok, *this, diag::err_feature_check_malformed);
1916           return II && isTargetVendor(getTargetInfo(), II);
1917         });
1918   } else if (II == Ident__is_target_os) {
1919     EvaluateFeatureLikeBuiltinMacro(
1920         OS, Tok, II, *this, false,
1921         [this](Token &Tok, bool &HasLexedNextToken) -> int {
1922           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1923               Tok, *this, diag::err_feature_check_malformed);
1924           return II && isTargetOS(getTargetInfo(), II);
1925         });
1926   } else if (II == Ident__is_target_environment) {
1927     EvaluateFeatureLikeBuiltinMacro(
1928         OS, Tok, II, *this, false,
1929         [this](Token &Tok, bool &HasLexedNextToken) -> int {
1930           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1931               Tok, *this, diag::err_feature_check_malformed);
1932           return II && isTargetEnvironment(getTargetInfo(), II);
1933         });
1934   } else if (II == Ident__is_target_variant_os) {
1935     EvaluateFeatureLikeBuiltinMacro(
1936         OS, Tok, II, *this, false,
1937         [this](Token &Tok, bool &HasLexedNextToken) -> int {
1938           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1939               Tok, *this, diag::err_feature_check_malformed);
1940           return II && isTargetVariantOS(getTargetInfo(), II);
1941         });
1942   } else if (II == Ident__is_target_variant_environment) {
1943     EvaluateFeatureLikeBuiltinMacro(
1944         OS, Tok, II, *this, false,
1945         [this](Token &Tok, bool &HasLexedNextToken) -> int {
1946           IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1947               Tok, *this, diag::err_feature_check_malformed);
1948           return II && isTargetVariantEnvironment(getTargetInfo(), II);
1949         });
1950   } else {
1951     llvm_unreachable("Unknown identifier!");
1952   }
1953   CreateString(OS.str(), Tok, Tok.getLocation(), Tok.getLocation());
1954   Tok.setFlagValue(Token::StartOfLine, IsAtStartOfLine);
1955   Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace);
1956 }
1957 
markMacroAsUsed(MacroInfo * MI)1958 void Preprocessor::markMacroAsUsed(MacroInfo *MI) {
1959   // If the 'used' status changed, and the macro requires 'unused' warning,
1960   // remove its SourceLocation from the warn-for-unused-macro locations.
1961   if (MI->isWarnIfUnused() && !MI->isUsed())
1962     WarnUnusedMacroLocs.erase(MI->getDefinitionLoc());
1963   MI->setIsUsed(true);
1964 }
1965 
processPathForFileMacro(SmallVectorImpl<char> & Path,const LangOptions & LangOpts,const TargetInfo & TI)1966 void Preprocessor::processPathForFileMacro(SmallVectorImpl<char> &Path,
1967                                            const LangOptions &LangOpts,
1968                                            const TargetInfo &TI) {
1969   LangOpts.remapPathPrefix(Path);
1970   if (LangOpts.UseTargetPathSeparator) {
1971     if (TI.getTriple().isOSWindows())
1972       llvm::sys::path::remove_dots(Path, false,
1973                                    llvm::sys::path::Style::windows_backslash);
1974     else
1975       llvm::sys::path::remove_dots(Path, false, llvm::sys::path::Style::posix);
1976   }
1977 }
1978 
processPathToFileName(SmallVectorImpl<char> & FileName,const PresumedLoc & PLoc,const LangOptions & LangOpts,const TargetInfo & TI)1979 void Preprocessor::processPathToFileName(SmallVectorImpl<char> &FileName,
1980                                          const PresumedLoc &PLoc,
1981                                          const LangOptions &LangOpts,
1982                                          const TargetInfo &TI) {
1983   // Try to get the last path component, failing that return the original
1984   // presumed location.
1985   StringRef PLFileName = llvm::sys::path::filename(PLoc.getFilename());
1986   if (PLFileName.empty())
1987     PLFileName = PLoc.getFilename();
1988   FileName.append(PLFileName.begin(), PLFileName.end());
1989   processPathForFileMacro(FileName, LangOpts, TI);
1990 }
1991