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/Attributes.h"
15 #include "clang/Basic/Builtins.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/HeaderSearch.h"
27 #include "clang/Lex/LexDiagnostic.h"
28 #include "clang/Lex/LiteralSupport.h"
29 #include "clang/Lex/MacroArgs.h"
30 #include "clang/Lex/MacroInfo.h"
31 #include "clang/Lex/Preprocessor.h"
32 #include "clang/Lex/PreprocessorLexer.h"
33 #include "clang/Lex/PreprocessorOptions.h"
34 #include "clang/Lex/Token.h"
35 #include "llvm/ADT/ArrayRef.h"
36 #include "llvm/ADT/DenseMap.h"
37 #include "llvm/ADT/DenseSet.h"
38 #include "llvm/ADT/FoldingSet.h"
39 #include "llvm/ADT/None.h"
40 #include "llvm/ADT/Optional.h"
41 #include "llvm/ADT/STLExtras.h"
42 #include "llvm/ADT/SmallString.h"
43 #include "llvm/ADT/SmallVector.h"
44 #include "llvm/ADT/StringRef.h"
45 #include "llvm/ADT/StringSwitch.h"
46 #include "llvm/Support/Casting.h"
47 #include "llvm/Support/ErrorHandling.h"
48 #include "llvm/Support/Format.h"
49 #include "llvm/Support/Path.h"
50 #include "llvm/Support/raw_ostream.h"
51 #include <algorithm>
52 #include <cassert>
53 #include <cstddef>
54 #include <cstring>
55 #include <ctime>
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.find(II) == LeafModuleMacros.end())
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.find(II) == LeafModuleMacros.end())
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 LeafMacros.erase(std::remove_if(LeafMacros.begin(), LeafMacros.end(),
159 [](ModuleMacro *MM) {
160 return MM->NumOverriddenBy != 0;
161 }),
162 LeafMacros.end());
163 }
164
165 // The new macro is always a leaf macro.
166 LeafMacros.push_back(MM);
167 // The identifier now has defined macros (that may or may not be visible).
168 II->setHasMacroDefinition(true);
169
170 New = true;
171 return MM;
172 }
173
getModuleMacro(Module * Mod,const IdentifierInfo * II)174 ModuleMacro *Preprocessor::getModuleMacro(Module *Mod,
175 const IdentifierInfo *II) {
176 llvm::FoldingSetNodeID ID;
177 ModuleMacro::Profile(ID, Mod, II);
178
179 void *InsertPos;
180 return ModuleMacros.FindNodeOrInsertPos(ID, InsertPos);
181 }
182
updateModuleMacroInfo(const IdentifierInfo * II,ModuleMacroInfo & Info)183 void Preprocessor::updateModuleMacroInfo(const IdentifierInfo *II,
184 ModuleMacroInfo &Info) {
185 assert(Info.ActiveModuleMacrosGeneration !=
186 CurSubmoduleState->VisibleModules.getGeneration() &&
187 "don't need to update this macro name info");
188 Info.ActiveModuleMacrosGeneration =
189 CurSubmoduleState->VisibleModules.getGeneration();
190
191 auto Leaf = LeafModuleMacros.find(II);
192 if (Leaf == LeafModuleMacros.end()) {
193 // No imported macros at all: nothing to do.
194 return;
195 }
196
197 Info.ActiveModuleMacros.clear();
198
199 // Every macro that's locally overridden is overridden by a visible macro.
200 llvm::DenseMap<ModuleMacro *, int> NumHiddenOverrides;
201 for (auto *O : Info.OverriddenMacros)
202 NumHiddenOverrides[O] = -1;
203
204 // Collect all macros that are not overridden by a visible macro.
205 llvm::SmallVector<ModuleMacro *, 16> Worklist;
206 for (auto *LeafMM : Leaf->second) {
207 assert(LeafMM->getNumOverridingMacros() == 0 && "leaf macro overridden");
208 if (NumHiddenOverrides.lookup(LeafMM) == 0)
209 Worklist.push_back(LeafMM);
210 }
211 while (!Worklist.empty()) {
212 auto *MM = Worklist.pop_back_val();
213 if (CurSubmoduleState->VisibleModules.isVisible(MM->getOwningModule())) {
214 // We only care about collecting definitions; undefinitions only act
215 // to override other definitions.
216 if (MM->getMacroInfo())
217 Info.ActiveModuleMacros.push_back(MM);
218 } else {
219 for (auto *O : MM->overrides())
220 if ((unsigned)++NumHiddenOverrides[O] == O->getNumOverridingMacros())
221 Worklist.push_back(O);
222 }
223 }
224 // Our reverse postorder walk found the macros in reverse order.
225 std::reverse(Info.ActiveModuleMacros.begin(), Info.ActiveModuleMacros.end());
226
227 // Determine whether the macro name is ambiguous.
228 MacroInfo *MI = nullptr;
229 bool IsSystemMacro = true;
230 bool IsAmbiguous = false;
231 if (auto *MD = Info.MD) {
232 while (MD && isa<VisibilityMacroDirective>(MD))
233 MD = MD->getPrevious();
234 if (auto *DMD = dyn_cast_or_null<DefMacroDirective>(MD)) {
235 MI = DMD->getInfo();
236 IsSystemMacro &= SourceMgr.isInSystemHeader(DMD->getLocation());
237 }
238 }
239 for (auto *Active : Info.ActiveModuleMacros) {
240 auto *NewMI = Active->getMacroInfo();
241
242 // Before marking the macro as ambiguous, check if this is a case where
243 // both macros are in system headers. If so, we trust that the system
244 // did not get it wrong. This also handles cases where Clang's own
245 // headers have a different spelling of certain system macros:
246 // #define LONG_MAX __LONG_MAX__ (clang's limits.h)
247 // #define LONG_MAX 0x7fffffffffffffffL (system's limits.h)
248 //
249 // FIXME: Remove the defined-in-system-headers check. clang's limits.h
250 // overrides the system limits.h's macros, so there's no conflict here.
251 if (MI && NewMI != MI &&
252 !MI->isIdenticalTo(*NewMI, *this, /*Syntactically=*/true))
253 IsAmbiguous = true;
254 IsSystemMacro &= Active->getOwningModule()->IsSystem ||
255 SourceMgr.isInSystemHeader(NewMI->getDefinitionLoc());
256 MI = NewMI;
257 }
258 Info.IsAmbiguous = IsAmbiguous && !IsSystemMacro;
259 }
260
dumpMacroInfo(const IdentifierInfo * II)261 void Preprocessor::dumpMacroInfo(const IdentifierInfo *II) {
262 ArrayRef<ModuleMacro*> Leaf;
263 auto LeafIt = LeafModuleMacros.find(II);
264 if (LeafIt != LeafModuleMacros.end())
265 Leaf = LeafIt->second;
266 const MacroState *State = nullptr;
267 auto Pos = CurSubmoduleState->Macros.find(II);
268 if (Pos != CurSubmoduleState->Macros.end())
269 State = &Pos->second;
270
271 llvm::errs() << "MacroState " << State << " " << II->getNameStart();
272 if (State && State->isAmbiguous(*this, II))
273 llvm::errs() << " ambiguous";
274 if (State && !State->getOverriddenMacros().empty()) {
275 llvm::errs() << " overrides";
276 for (auto *O : State->getOverriddenMacros())
277 llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
278 }
279 llvm::errs() << "\n";
280
281 // Dump local macro directives.
282 for (auto *MD = State ? State->getLatest() : nullptr; MD;
283 MD = MD->getPrevious()) {
284 llvm::errs() << " ";
285 MD->dump();
286 }
287
288 // Dump module macros.
289 llvm::DenseSet<ModuleMacro*> Active;
290 for (auto *MM : State ? State->getActiveModuleMacros(*this, II) : None)
291 Active.insert(MM);
292 llvm::DenseSet<ModuleMacro*> Visited;
293 llvm::SmallVector<ModuleMacro *, 16> Worklist(Leaf.begin(), Leaf.end());
294 while (!Worklist.empty()) {
295 auto *MM = Worklist.pop_back_val();
296 llvm::errs() << " ModuleMacro " << MM << " "
297 << MM->getOwningModule()->getFullModuleName();
298 if (!MM->getMacroInfo())
299 llvm::errs() << " undef";
300
301 if (Active.count(MM))
302 llvm::errs() << " active";
303 else if (!CurSubmoduleState->VisibleModules.isVisible(
304 MM->getOwningModule()))
305 llvm::errs() << " hidden";
306 else if (MM->getMacroInfo())
307 llvm::errs() << " overridden";
308
309 if (!MM->overrides().empty()) {
310 llvm::errs() << " overrides";
311 for (auto *O : MM->overrides()) {
312 llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
313 if (Visited.insert(O).second)
314 Worklist.push_back(O);
315 }
316 }
317 llvm::errs() << "\n";
318 if (auto *MI = MM->getMacroInfo()) {
319 llvm::errs() << " ";
320 MI->dump();
321 llvm::errs() << "\n";
322 }
323 }
324 }
325
326 /// RegisterBuiltinMacro - Register the specified identifier in the identifier
327 /// table and mark it as a builtin macro to be expanded.
RegisterBuiltinMacro(Preprocessor & PP,const char * Name)328 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){
329 // Get the identifier.
330 IdentifierInfo *Id = PP.getIdentifierInfo(Name);
331
332 // Mark it as being a macro that is builtin.
333 MacroInfo *MI = PP.AllocateMacroInfo(SourceLocation());
334 MI->setIsBuiltinMacro();
335 PP.appendDefMacroDirective(Id, MI);
336 return Id;
337 }
338
339 /// RegisterBuiltinMacros - Register builtin macros, such as __LINE__ with the
340 /// identifier table.
RegisterBuiltinMacros()341 void Preprocessor::RegisterBuiltinMacros() {
342 Ident__LINE__ = RegisterBuiltinMacro(*this, "__LINE__");
343 Ident__FILE__ = RegisterBuiltinMacro(*this, "__FILE__");
344 Ident__DATE__ = RegisterBuiltinMacro(*this, "__DATE__");
345 Ident__TIME__ = RegisterBuiltinMacro(*this, "__TIME__");
346 Ident__COUNTER__ = RegisterBuiltinMacro(*this, "__COUNTER__");
347 Ident_Pragma = RegisterBuiltinMacro(*this, "_Pragma");
348
349 // C++ Standing Document Extensions.
350 if (getLangOpts().CPlusPlus)
351 Ident__has_cpp_attribute =
352 RegisterBuiltinMacro(*this, "__has_cpp_attribute");
353 else
354 Ident__has_cpp_attribute = nullptr;
355
356 // GCC Extensions.
357 Ident__BASE_FILE__ = RegisterBuiltinMacro(*this, "__BASE_FILE__");
358 Ident__INCLUDE_LEVEL__ = RegisterBuiltinMacro(*this, "__INCLUDE_LEVEL__");
359 Ident__TIMESTAMP__ = RegisterBuiltinMacro(*this, "__TIMESTAMP__");
360
361 // Microsoft Extensions.
362 if (getLangOpts().MicrosoftExt) {
363 Ident__identifier = RegisterBuiltinMacro(*this, "__identifier");
364 Ident__pragma = RegisterBuiltinMacro(*this, "__pragma");
365 } else {
366 Ident__identifier = nullptr;
367 Ident__pragma = nullptr;
368 }
369
370 // Clang Extensions.
371 Ident__FILE_NAME__ = RegisterBuiltinMacro(*this, "__FILE_NAME__");
372 Ident__has_feature = RegisterBuiltinMacro(*this, "__has_feature");
373 Ident__has_extension = RegisterBuiltinMacro(*this, "__has_extension");
374 Ident__has_builtin = RegisterBuiltinMacro(*this, "__has_builtin");
375 Ident__has_attribute = RegisterBuiltinMacro(*this, "__has_attribute");
376 if (!getLangOpts().CPlusPlus)
377 Ident__has_c_attribute = RegisterBuiltinMacro(*this, "__has_c_attribute");
378 else
379 Ident__has_c_attribute = nullptr;
380
381 Ident__has_declspec = RegisterBuiltinMacro(*this, "__has_declspec_attribute");
382 Ident__has_include = RegisterBuiltinMacro(*this, "__has_include");
383 Ident__has_include_next = RegisterBuiltinMacro(*this, "__has_include_next");
384 Ident__has_warning = RegisterBuiltinMacro(*this, "__has_warning");
385 Ident__is_identifier = RegisterBuiltinMacro(*this, "__is_identifier");
386 Ident__is_target_arch = RegisterBuiltinMacro(*this, "__is_target_arch");
387 Ident__is_target_vendor = RegisterBuiltinMacro(*this, "__is_target_vendor");
388 Ident__is_target_os = RegisterBuiltinMacro(*this, "__is_target_os");
389 Ident__is_target_environment =
390 RegisterBuiltinMacro(*this, "__is_target_environment");
391
392 // Modules.
393 Ident__building_module = RegisterBuiltinMacro(*this, "__building_module");
394 if (!getLangOpts().CurrentModule.empty())
395 Ident__MODULE__ = RegisterBuiltinMacro(*this, "__MODULE__");
396 else
397 Ident__MODULE__ = nullptr;
398 }
399
400 /// isTrivialSingleTokenExpansion - Return true if MI, which has a single token
401 /// in its expansion, currently expands to that token literally.
isTrivialSingleTokenExpansion(const MacroInfo * MI,const IdentifierInfo * MacroIdent,Preprocessor & PP)402 static bool isTrivialSingleTokenExpansion(const MacroInfo *MI,
403 const IdentifierInfo *MacroIdent,
404 Preprocessor &PP) {
405 IdentifierInfo *II = MI->getReplacementToken(0).getIdentifierInfo();
406
407 // If the token isn't an identifier, it's always literally expanded.
408 if (!II) return true;
409
410 // If the information about this identifier is out of date, update it from
411 // the external source.
412 if (II->isOutOfDate())
413 PP.getExternalSource()->updateOutOfDateIdentifier(*II);
414
415 // If the identifier is a macro, and if that macro is enabled, it may be
416 // expanded so it's not a trivial expansion.
417 if (auto *ExpansionMI = PP.getMacroInfo(II))
418 if (ExpansionMI->isEnabled() &&
419 // Fast expanding "#define X X" is ok, because X would be disabled.
420 II != MacroIdent)
421 return false;
422
423 // If this is an object-like macro invocation, it is safe to trivially expand
424 // it.
425 if (MI->isObjectLike()) return true;
426
427 // If this is a function-like macro invocation, it's safe to trivially expand
428 // as long as the identifier is not a macro argument.
429 return std::find(MI->param_begin(), MI->param_end(), II) == MI->param_end();
430 }
431
432 /// isNextPPTokenLParen - Determine whether the next preprocessor token to be
433 /// lexed is a '('. If so, consume the token and return true, if not, this
434 /// method should have no observable side-effect on the lexed tokens.
isNextPPTokenLParen()435 bool Preprocessor::isNextPPTokenLParen() {
436 // Do some quick tests for rejection cases.
437 unsigned Val;
438 if (CurLexer)
439 Val = CurLexer->isNextPPTokenLParen();
440 else
441 Val = CurTokenLexer->isNextTokenLParen();
442
443 if (Val == 2) {
444 // We have run off the end. If it's a source file we don't
445 // examine enclosing ones (C99 5.1.1.2p4). Otherwise walk up the
446 // macro stack.
447 if (CurPPLexer)
448 return false;
449 for (const IncludeStackInfo &Entry : llvm::reverse(IncludeMacroStack)) {
450 if (Entry.TheLexer)
451 Val = Entry.TheLexer->isNextPPTokenLParen();
452 else
453 Val = Entry.TheTokenLexer->isNextTokenLParen();
454
455 if (Val != 2)
456 break;
457
458 // Ran off the end of a source file?
459 if (Entry.ThePPLexer)
460 return false;
461 }
462 }
463
464 // Okay, if we know that the token is a '(', lex it and return. Otherwise we
465 // have found something that isn't a '(' or we found the end of the
466 // translation unit. In either case, return false.
467 return Val == 1;
468 }
469
470 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be
471 /// expanded as a macro, handle it and return the next token as 'Identifier'.
HandleMacroExpandedIdentifier(Token & Identifier,const MacroDefinition & M)472 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
473 const MacroDefinition &M) {
474 MacroInfo *MI = M.getMacroInfo();
475
476 // If this is a macro expansion in the "#if !defined(x)" line for the file,
477 // then the macro could expand to different things in other contexts, we need
478 // to disable the optimization in this case.
479 if (CurPPLexer) CurPPLexer->MIOpt.ExpandedMacro();
480
481 // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially.
482 if (MI->isBuiltinMacro()) {
483 if (Callbacks)
484 Callbacks->MacroExpands(Identifier, M, Identifier.getLocation(),
485 /*Args=*/nullptr);
486 ExpandBuiltinMacro(Identifier);
487 return true;
488 }
489
490 /// Args - If this is a function-like macro expansion, this contains,
491 /// for each macro argument, the list of tokens that were provided to the
492 /// invocation.
493 MacroArgs *Args = nullptr;
494
495 // Remember where the end of the expansion occurred. For an object-like
496 // macro, this is the identifier. For a function-like macro, this is the ')'.
497 SourceLocation ExpansionEnd = Identifier.getLocation();
498
499 // If this is a function-like macro, read the arguments.
500 if (MI->isFunctionLike()) {
501 // Remember that we are now parsing the arguments to a macro invocation.
502 // Preprocessor directives used inside macro arguments are not portable, and
503 // this enables the warning.
504 InMacroArgs = true;
505 ArgMacro = &Identifier;
506
507 Args = ReadMacroCallArgumentList(Identifier, MI, ExpansionEnd);
508
509 // Finished parsing args.
510 InMacroArgs = false;
511 ArgMacro = nullptr;
512
513 // If there was an error parsing the arguments, bail out.
514 if (!Args) return true;
515
516 ++NumFnMacroExpanded;
517 } else {
518 ++NumMacroExpanded;
519 }
520
521 // Notice that this macro has been used.
522 markMacroAsUsed(MI);
523
524 // Remember where the token is expanded.
525 SourceLocation ExpandLoc = Identifier.getLocation();
526 SourceRange ExpansionRange(ExpandLoc, ExpansionEnd);
527
528 if (Callbacks) {
529 if (InMacroArgs) {
530 // We can have macro expansion inside a conditional directive while
531 // reading the function macro arguments. To ensure, in that case, that
532 // MacroExpands callbacks still happen in source order, queue this
533 // callback to have it happen after the function macro callback.
534 DelayedMacroExpandsCallbacks.push_back(
535 MacroExpandsInfo(Identifier, M, ExpansionRange));
536 } else {
537 Callbacks->MacroExpands(Identifier, M, ExpansionRange, Args);
538 if (!DelayedMacroExpandsCallbacks.empty()) {
539 for (const MacroExpandsInfo &Info : DelayedMacroExpandsCallbacks) {
540 // FIXME: We lose macro args info with delayed callback.
541 Callbacks->MacroExpands(Info.Tok, Info.MD, Info.Range,
542 /*Args=*/nullptr);
543 }
544 DelayedMacroExpandsCallbacks.clear();
545 }
546 }
547 }
548
549 // If the macro definition is ambiguous, complain.
550 if (M.isAmbiguous()) {
551 Diag(Identifier, diag::warn_pp_ambiguous_macro)
552 << Identifier.getIdentifierInfo();
553 Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen)
554 << Identifier.getIdentifierInfo();
555 M.forAllDefinitions([&](const MacroInfo *OtherMI) {
556 if (OtherMI != MI)
557 Diag(OtherMI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_other)
558 << Identifier.getIdentifierInfo();
559 });
560 }
561
562 // If we started lexing a macro, enter the macro expansion body.
563
564 // If this macro expands to no tokens, don't bother to push it onto the
565 // expansion stack, only to take it right back off.
566 if (MI->getNumTokens() == 0) {
567 // No need for arg info.
568 if (Args) Args->destroy(*this);
569
570 // Propagate whitespace info as if we had pushed, then popped,
571 // a macro context.
572 Identifier.setFlag(Token::LeadingEmptyMacro);
573 PropagateLineStartLeadingSpaceInfo(Identifier);
574 ++NumFastMacroExpanded;
575 return false;
576 } else if (MI->getNumTokens() == 1 &&
577 isTrivialSingleTokenExpansion(MI, Identifier.getIdentifierInfo(),
578 *this)) {
579 // Otherwise, if this macro expands into a single trivially-expanded
580 // token: expand it now. This handles common cases like
581 // "#define VAL 42".
582
583 // No need for arg info.
584 if (Args) Args->destroy(*this);
585
586 // Propagate the isAtStartOfLine/hasLeadingSpace markers of the macro
587 // identifier to the expanded token.
588 bool isAtStartOfLine = Identifier.isAtStartOfLine();
589 bool hasLeadingSpace = Identifier.hasLeadingSpace();
590
591 // Replace the result token.
592 Identifier = MI->getReplacementToken(0);
593
594 // Restore the StartOfLine/LeadingSpace markers.
595 Identifier.setFlagValue(Token::StartOfLine , isAtStartOfLine);
596 Identifier.setFlagValue(Token::LeadingSpace, hasLeadingSpace);
597
598 // Update the tokens location to include both its expansion and physical
599 // locations.
600 SourceLocation Loc =
601 SourceMgr.createExpansionLoc(Identifier.getLocation(), ExpandLoc,
602 ExpansionEnd,Identifier.getLength());
603 Identifier.setLocation(Loc);
604
605 // If this is a disabled macro or #define X X, we must mark the result as
606 // unexpandable.
607 if (IdentifierInfo *NewII = Identifier.getIdentifierInfo()) {
608 if (MacroInfo *NewMI = getMacroInfo(NewII))
609 if (!NewMI->isEnabled() || NewMI == MI) {
610 Identifier.setFlag(Token::DisableExpand);
611 // Don't warn for "#define X X" like "#define bool bool" from
612 // stdbool.h.
613 if (NewMI != MI || MI->isFunctionLike())
614 Diag(Identifier, diag::pp_disabled_macro_expansion);
615 }
616 }
617
618 // Since this is not an identifier token, it can't be macro expanded, so
619 // we're done.
620 ++NumFastMacroExpanded;
621 return true;
622 }
623
624 // Start expanding the macro.
625 EnterMacro(Identifier, ExpansionEnd, MI, Args);
626 return false;
627 }
628
629 enum Bracket {
630 Brace,
631 Paren
632 };
633
634 /// CheckMatchedBrackets - Returns true if the braces and parentheses in the
635 /// token vector are properly nested.
CheckMatchedBrackets(const SmallVectorImpl<Token> & Tokens)636 static bool CheckMatchedBrackets(const SmallVectorImpl<Token> &Tokens) {
637 SmallVector<Bracket, 8> Brackets;
638 for (SmallVectorImpl<Token>::const_iterator I = Tokens.begin(),
639 E = Tokens.end();
640 I != E; ++I) {
641 if (I->is(tok::l_paren)) {
642 Brackets.push_back(Paren);
643 } else if (I->is(tok::r_paren)) {
644 if (Brackets.empty() || Brackets.back() == Brace)
645 return false;
646 Brackets.pop_back();
647 } else if (I->is(tok::l_brace)) {
648 Brackets.push_back(Brace);
649 } else if (I->is(tok::r_brace)) {
650 if (Brackets.empty() || Brackets.back() == Paren)
651 return false;
652 Brackets.pop_back();
653 }
654 }
655 return Brackets.empty();
656 }
657
658 /// GenerateNewArgTokens - Returns true if OldTokens can be converted to a new
659 /// vector of tokens in NewTokens. The new number of arguments will be placed
660 /// in NumArgs and the ranges which need to surrounded in parentheses will be
661 /// in ParenHints.
662 /// Returns false if the token stream cannot be changed. If this is because
663 /// of an initializer list starting a macro argument, the range of those
664 /// initializer lists will be place in InitLists.
GenerateNewArgTokens(Preprocessor & PP,SmallVectorImpl<Token> & OldTokens,SmallVectorImpl<Token> & NewTokens,unsigned & NumArgs,SmallVectorImpl<SourceRange> & ParenHints,SmallVectorImpl<SourceRange> & InitLists)665 static bool GenerateNewArgTokens(Preprocessor &PP,
666 SmallVectorImpl<Token> &OldTokens,
667 SmallVectorImpl<Token> &NewTokens,
668 unsigned &NumArgs,
669 SmallVectorImpl<SourceRange> &ParenHints,
670 SmallVectorImpl<SourceRange> &InitLists) {
671 if (!CheckMatchedBrackets(OldTokens))
672 return false;
673
674 // Once it is known that the brackets are matched, only a simple count of the
675 // braces is needed.
676 unsigned Braces = 0;
677
678 // First token of a new macro argument.
679 SmallVectorImpl<Token>::iterator ArgStartIterator = OldTokens.begin();
680
681 // First closing brace in a new macro argument. Used to generate
682 // SourceRanges for InitLists.
683 SmallVectorImpl<Token>::iterator ClosingBrace = OldTokens.end();
684 NumArgs = 0;
685 Token TempToken;
686 // Set to true when a macro separator token is found inside a braced list.
687 // If true, the fixed argument spans multiple old arguments and ParenHints
688 // will be updated.
689 bool FoundSeparatorToken = false;
690 for (SmallVectorImpl<Token>::iterator I = OldTokens.begin(),
691 E = OldTokens.end();
692 I != E; ++I) {
693 if (I->is(tok::l_brace)) {
694 ++Braces;
695 } else if (I->is(tok::r_brace)) {
696 --Braces;
697 if (Braces == 0 && ClosingBrace == E && FoundSeparatorToken)
698 ClosingBrace = I;
699 } else if (I->is(tok::eof)) {
700 // EOF token is used to separate macro arguments
701 if (Braces != 0) {
702 // Assume comma separator is actually braced list separator and change
703 // it back to a comma.
704 FoundSeparatorToken = true;
705 I->setKind(tok::comma);
706 I->setLength(1);
707 } else { // Braces == 0
708 // Separator token still separates arguments.
709 ++NumArgs;
710
711 // If the argument starts with a brace, it can't be fixed with
712 // parentheses. A different diagnostic will be given.
713 if (FoundSeparatorToken && ArgStartIterator->is(tok::l_brace)) {
714 InitLists.push_back(
715 SourceRange(ArgStartIterator->getLocation(),
716 PP.getLocForEndOfToken(ClosingBrace->getLocation())));
717 ClosingBrace = E;
718 }
719
720 // Add left paren
721 if (FoundSeparatorToken) {
722 TempToken.startToken();
723 TempToken.setKind(tok::l_paren);
724 TempToken.setLocation(ArgStartIterator->getLocation());
725 TempToken.setLength(0);
726 NewTokens.push_back(TempToken);
727 }
728
729 // Copy over argument tokens
730 NewTokens.insert(NewTokens.end(), ArgStartIterator, I);
731
732 // Add right paren and store the paren locations in ParenHints
733 if (FoundSeparatorToken) {
734 SourceLocation Loc = PP.getLocForEndOfToken((I - 1)->getLocation());
735 TempToken.startToken();
736 TempToken.setKind(tok::r_paren);
737 TempToken.setLocation(Loc);
738 TempToken.setLength(0);
739 NewTokens.push_back(TempToken);
740 ParenHints.push_back(SourceRange(ArgStartIterator->getLocation(),
741 Loc));
742 }
743
744 // Copy separator token
745 NewTokens.push_back(*I);
746
747 // Reset values
748 ArgStartIterator = I + 1;
749 FoundSeparatorToken = false;
750 }
751 }
752 }
753
754 return !ParenHints.empty() && InitLists.empty();
755 }
756
757 /// ReadFunctionLikeMacroArgs - After reading "MACRO" and knowing that the next
758 /// token is the '(' of the macro, this method is invoked to read all of the
759 /// actual arguments specified for the macro invocation. This returns null on
760 /// error.
ReadMacroCallArgumentList(Token & MacroName,MacroInfo * MI,SourceLocation & MacroEnd)761 MacroArgs *Preprocessor::ReadMacroCallArgumentList(Token &MacroName,
762 MacroInfo *MI,
763 SourceLocation &MacroEnd) {
764 // The number of fixed arguments to parse.
765 unsigned NumFixedArgsLeft = MI->getNumParams();
766 bool isVariadic = MI->isVariadic();
767
768 // Outer loop, while there are more arguments, keep reading them.
769 Token Tok;
770
771 // Read arguments as unexpanded tokens. This avoids issues, e.g., where
772 // an argument value in a macro could expand to ',' or '(' or ')'.
773 LexUnexpandedToken(Tok);
774 assert(Tok.is(tok::l_paren) && "Error computing l-paren-ness?");
775
776 // ArgTokens - Build up a list of tokens that make up each argument. Each
777 // argument is separated by an EOF token. Use a SmallVector so we can avoid
778 // heap allocations in the common case.
779 SmallVector<Token, 64> ArgTokens;
780 bool ContainsCodeCompletionTok = false;
781 bool FoundElidedComma = false;
782
783 SourceLocation TooManyArgsLoc;
784
785 unsigned NumActuals = 0;
786 while (Tok.isNot(tok::r_paren)) {
787 if (ContainsCodeCompletionTok && Tok.isOneOf(tok::eof, tok::eod))
788 break;
789
790 assert(Tok.isOneOf(tok::l_paren, tok::comma) &&
791 "only expect argument separators here");
792
793 size_t ArgTokenStart = ArgTokens.size();
794 SourceLocation ArgStartLoc = Tok.getLocation();
795
796 // C99 6.10.3p11: Keep track of the number of l_parens we have seen. Note
797 // that we already consumed the first one.
798 unsigned NumParens = 0;
799
800 while (true) {
801 // Read arguments as unexpanded tokens. This avoids issues, e.g., where
802 // an argument value in a macro could expand to ',' or '(' or ')'.
803 LexUnexpandedToken(Tok);
804
805 if (Tok.isOneOf(tok::eof, tok::eod)) { // "#if f(<eof>" & "#if f(\n"
806 if (!ContainsCodeCompletionTok) {
807 Diag(MacroName, diag::err_unterm_macro_invoc);
808 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
809 << MacroName.getIdentifierInfo();
810 // Do not lose the EOF/EOD. Return it to the client.
811 MacroName = Tok;
812 return nullptr;
813 }
814 // Do not lose the EOF/EOD.
815 auto Toks = std::make_unique<Token[]>(1);
816 Toks[0] = Tok;
817 EnterTokenStream(std::move(Toks), 1, true, /*IsReinject*/ false);
818 break;
819 } else if (Tok.is(tok::r_paren)) {
820 // If we found the ) token, the macro arg list is done.
821 if (NumParens-- == 0) {
822 MacroEnd = Tok.getLocation();
823 if (!ArgTokens.empty() &&
824 ArgTokens.back().commaAfterElided()) {
825 FoundElidedComma = true;
826 }
827 break;
828 }
829 } else if (Tok.is(tok::l_paren)) {
830 ++NumParens;
831 } else if (Tok.is(tok::comma)) {
832 // In Microsoft-compatibility mode, single commas from nested macro
833 // expansions should not be considered as argument separators. We test
834 // for this with the IgnoredComma token flag.
835 if (Tok.getFlags() & Token::IgnoredComma) {
836 // However, in MSVC's preprocessor, subsequent expansions do treat
837 // these commas as argument separators. This leads to a common
838 // workaround used in macros that need to work in both MSVC and
839 // compliant preprocessors. Therefore, the IgnoredComma flag can only
840 // apply once to any given token.
841 Tok.clearFlag(Token::IgnoredComma);
842 } else if (NumParens == 0) {
843 // Comma ends this argument if there are more fixed arguments
844 // expected. However, if this is a variadic macro, and this is part of
845 // the variadic part, then the comma is just an argument token.
846 if (!isVariadic)
847 break;
848 if (NumFixedArgsLeft > 1)
849 break;
850 }
851 } else if (Tok.is(tok::comment) && !KeepMacroComments) {
852 // If this is a comment token in the argument list and we're just in
853 // -C mode (not -CC mode), discard the comment.
854 continue;
855 } else if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) {
856 // Reading macro arguments can cause macros that we are currently
857 // expanding from to be popped off the expansion stack. Doing so causes
858 // them to be reenabled for expansion. Here we record whether any
859 // identifiers we lex as macro arguments correspond to disabled macros.
860 // If so, we mark the token as noexpand. This is a subtle aspect of
861 // C99 6.10.3.4p2.
862 if (MacroInfo *MI = getMacroInfo(Tok.getIdentifierInfo()))
863 if (!MI->isEnabled())
864 Tok.setFlag(Token::DisableExpand);
865 } else if (Tok.is(tok::code_completion)) {
866 ContainsCodeCompletionTok = true;
867 if (CodeComplete)
868 CodeComplete->CodeCompleteMacroArgument(MacroName.getIdentifierInfo(),
869 MI, NumActuals);
870 // Don't mark that we reached the code-completion point because the
871 // parser is going to handle the token and there will be another
872 // code-completion callback.
873 }
874
875 ArgTokens.push_back(Tok);
876 }
877
878 // If this was an empty argument list foo(), don't add this as an empty
879 // argument.
880 if (ArgTokens.empty() && Tok.getKind() == tok::r_paren)
881 break;
882
883 // If this is not a variadic macro, and too many args were specified, emit
884 // an error.
885 if (!isVariadic && NumFixedArgsLeft == 0 && TooManyArgsLoc.isInvalid()) {
886 if (ArgTokens.size() != ArgTokenStart)
887 TooManyArgsLoc = ArgTokens[ArgTokenStart].getLocation();
888 else
889 TooManyArgsLoc = ArgStartLoc;
890 }
891
892 // Empty arguments are standard in C99 and C++0x, and are supported as an
893 // extension in other modes.
894 if (ArgTokens.size() == ArgTokenStart && !getLangOpts().C99)
895 Diag(Tok, getLangOpts().CPlusPlus11
896 ? diag::warn_cxx98_compat_empty_fnmacro_arg
897 : diag::ext_empty_fnmacro_arg);
898
899 // Add a marker EOF token to the end of the token list for this argument.
900 Token EOFTok;
901 EOFTok.startToken();
902 EOFTok.setKind(tok::eof);
903 EOFTok.setLocation(Tok.getLocation());
904 EOFTok.setLength(0);
905 ArgTokens.push_back(EOFTok);
906 ++NumActuals;
907 if (!ContainsCodeCompletionTok && NumFixedArgsLeft != 0)
908 --NumFixedArgsLeft;
909 }
910
911 // Okay, we either found the r_paren. Check to see if we parsed too few
912 // arguments.
913 unsigned MinArgsExpected = MI->getNumParams();
914
915 // If this is not a variadic macro, and too many args were specified, emit
916 // an error.
917 if (!isVariadic && NumActuals > MinArgsExpected &&
918 !ContainsCodeCompletionTok) {
919 // Emit the diagnostic at the macro name in case there is a missing ).
920 // Emitting it at the , could be far away from the macro name.
921 Diag(TooManyArgsLoc, diag::err_too_many_args_in_macro_invoc);
922 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
923 << MacroName.getIdentifierInfo();
924
925 // Commas from braced initializer lists will be treated as argument
926 // separators inside macros. Attempt to correct for this with parentheses.
927 // TODO: See if this can be generalized to angle brackets for templates
928 // inside macro arguments.
929
930 SmallVector<Token, 4> FixedArgTokens;
931 unsigned FixedNumArgs = 0;
932 SmallVector<SourceRange, 4> ParenHints, InitLists;
933 if (!GenerateNewArgTokens(*this, ArgTokens, FixedArgTokens, FixedNumArgs,
934 ParenHints, InitLists)) {
935 if (!InitLists.empty()) {
936 DiagnosticBuilder DB =
937 Diag(MacroName,
938 diag::note_init_list_at_beginning_of_macro_argument);
939 for (SourceRange Range : InitLists)
940 DB << Range;
941 }
942 return nullptr;
943 }
944 if (FixedNumArgs != MinArgsExpected)
945 return nullptr;
946
947 DiagnosticBuilder DB = Diag(MacroName, diag::note_suggest_parens_for_macro);
948 for (SourceRange ParenLocation : ParenHints) {
949 DB << FixItHint::CreateInsertion(ParenLocation.getBegin(), "(");
950 DB << FixItHint::CreateInsertion(ParenLocation.getEnd(), ")");
951 }
952 ArgTokens.swap(FixedArgTokens);
953 NumActuals = FixedNumArgs;
954 }
955
956 // See MacroArgs instance var for description of this.
957 bool isVarargsElided = false;
958
959 if (ContainsCodeCompletionTok) {
960 // Recover from not-fully-formed macro invocation during code-completion.
961 Token EOFTok;
962 EOFTok.startToken();
963 EOFTok.setKind(tok::eof);
964 EOFTok.setLocation(Tok.getLocation());
965 EOFTok.setLength(0);
966 for (; NumActuals < MinArgsExpected; ++NumActuals)
967 ArgTokens.push_back(EOFTok);
968 }
969
970 if (NumActuals < MinArgsExpected) {
971 // There are several cases where too few arguments is ok, handle them now.
972 if (NumActuals == 0 && MinArgsExpected == 1) {
973 // #define A(X) or #define A(...) ---> A()
974
975 // If there is exactly one argument, and that argument is missing,
976 // then we have an empty "()" argument empty list. This is fine, even if
977 // the macro expects one argument (the argument is just empty).
978 isVarargsElided = MI->isVariadic();
979 } else if ((FoundElidedComma || MI->isVariadic()) &&
980 (NumActuals+1 == MinArgsExpected || // A(x, ...) -> A(X)
981 (NumActuals == 0 && MinArgsExpected == 2))) {// A(x,...) -> A()
982 // Varargs where the named vararg parameter is missing: OK as extension.
983 // #define A(x, ...)
984 // A("blah")
985 //
986 // If the macro contains the comma pasting extension, the diagnostic
987 // is suppressed; we know we'll get another diagnostic later.
988 if (!MI->hasCommaPasting()) {
989 Diag(Tok, diag::ext_missing_varargs_arg);
990 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
991 << MacroName.getIdentifierInfo();
992 }
993
994 // Remember this occurred, allowing us to elide the comma when used for
995 // cases like:
996 // #define A(x, foo...) blah(a, ## foo)
997 // #define B(x, ...) blah(a, ## __VA_ARGS__)
998 // #define C(...) blah(a, ## __VA_ARGS__)
999 // A(x) B(x) C()
1000 isVarargsElided = true;
1001 } else if (!ContainsCodeCompletionTok) {
1002 // Otherwise, emit the error.
1003 Diag(Tok, diag::err_too_few_args_in_macro_invoc);
1004 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
1005 << MacroName.getIdentifierInfo();
1006 return nullptr;
1007 }
1008
1009 // Add a marker EOF token to the end of the token list for this argument.
1010 SourceLocation EndLoc = Tok.getLocation();
1011 Tok.startToken();
1012 Tok.setKind(tok::eof);
1013 Tok.setLocation(EndLoc);
1014 Tok.setLength(0);
1015 ArgTokens.push_back(Tok);
1016
1017 // If we expect two arguments, add both as empty.
1018 if (NumActuals == 0 && MinArgsExpected == 2)
1019 ArgTokens.push_back(Tok);
1020
1021 } else if (NumActuals > MinArgsExpected && !MI->isVariadic() &&
1022 !ContainsCodeCompletionTok) {
1023 // Emit the diagnostic at the macro name in case there is a missing ).
1024 // Emitting it at the , could be far away from the macro name.
1025 Diag(MacroName, diag::err_too_many_args_in_macro_invoc);
1026 Diag(MI->getDefinitionLoc(), diag::note_macro_here)
1027 << MacroName.getIdentifierInfo();
1028 return nullptr;
1029 }
1030
1031 return MacroArgs::create(MI, ArgTokens, isVarargsElided, *this);
1032 }
1033
1034 /// Keeps macro expanded tokens for TokenLexers.
1035 //
1036 /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
1037 /// going to lex in the cache and when it finishes the tokens are removed
1038 /// from the end of the cache.
cacheMacroExpandedTokens(TokenLexer * tokLexer,ArrayRef<Token> tokens)1039 Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer,
1040 ArrayRef<Token> tokens) {
1041 assert(tokLexer);
1042 if (tokens.empty())
1043 return nullptr;
1044
1045 size_t newIndex = MacroExpandedTokens.size();
1046 bool cacheNeedsToGrow = tokens.size() >
1047 MacroExpandedTokens.capacity()-MacroExpandedTokens.size();
1048 MacroExpandedTokens.append(tokens.begin(), tokens.end());
1049
1050 if (cacheNeedsToGrow) {
1051 // Go through all the TokenLexers whose 'Tokens' pointer points in the
1052 // buffer and update the pointers to the (potential) new buffer array.
1053 for (const auto &Lexer : MacroExpandingLexersStack) {
1054 TokenLexer *prevLexer;
1055 size_t tokIndex;
1056 std::tie(prevLexer, tokIndex) = Lexer;
1057 prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex;
1058 }
1059 }
1060
1061 MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex));
1062 return MacroExpandedTokens.data() + newIndex;
1063 }
1064
removeCachedMacroExpandedTokensOfLastLexer()1065 void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() {
1066 assert(!MacroExpandingLexersStack.empty());
1067 size_t tokIndex = MacroExpandingLexersStack.back().second;
1068 assert(tokIndex < MacroExpandedTokens.size());
1069 // Pop the cached macro expanded tokens from the end.
1070 MacroExpandedTokens.resize(tokIndex);
1071 MacroExpandingLexersStack.pop_back();
1072 }
1073
1074 /// ComputeDATE_TIME - Compute the current time, enter it into the specified
1075 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of
1076 /// the identifier tokens inserted.
ComputeDATE_TIME(SourceLocation & DATELoc,SourceLocation & TIMELoc,Preprocessor & PP)1077 static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc,
1078 Preprocessor &PP) {
1079 time_t TT = time(nullptr);
1080 struct tm *TM = localtime(&TT);
1081
1082 static const char * const Months[] = {
1083 "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"
1084 };
1085
1086 {
1087 SmallString<32> TmpBuffer;
1088 llvm::raw_svector_ostream TmpStream(TmpBuffer);
1089 TmpStream << llvm::format("\"%s %2d %4d\"", Months[TM->tm_mon],
1090 TM->tm_mday, TM->tm_year + 1900);
1091 Token TmpTok;
1092 TmpTok.startToken();
1093 PP.CreateString(TmpStream.str(), TmpTok);
1094 DATELoc = TmpTok.getLocation();
1095 }
1096
1097 {
1098 SmallString<32> TmpBuffer;
1099 llvm::raw_svector_ostream TmpStream(TmpBuffer);
1100 TmpStream << llvm::format("\"%02d:%02d:%02d\"",
1101 TM->tm_hour, TM->tm_min, TM->tm_sec);
1102 Token TmpTok;
1103 TmpTok.startToken();
1104 PP.CreateString(TmpStream.str(), TmpTok);
1105 TIMELoc = TmpTok.getLocation();
1106 }
1107 }
1108
1109 /// HasFeature - Return true if we recognize and implement the feature
1110 /// specified by the identifier as a standard language feature.
HasFeature(const Preprocessor & PP,StringRef Feature)1111 static bool HasFeature(const Preprocessor &PP, StringRef Feature) {
1112 const LangOptions &LangOpts = PP.getLangOpts();
1113
1114 // Normalize the feature name, __foo__ becomes foo.
1115 if (Feature.startswith("__") && Feature.endswith("__") && Feature.size() >= 4)
1116 Feature = Feature.substr(2, Feature.size() - 4);
1117
1118 #define FEATURE(Name, Predicate) .Case(#Name, Predicate)
1119 return llvm::StringSwitch<bool>(Feature)
1120 #include "clang/Basic/Features.def"
1121 .Default(false);
1122 #undef FEATURE
1123 }
1124
1125 /// HasExtension - Return true if we recognize and implement the feature
1126 /// specified by the identifier, either as an extension or a standard language
1127 /// feature.
HasExtension(const Preprocessor & PP,StringRef Extension)1128 static bool HasExtension(const Preprocessor &PP, StringRef Extension) {
1129 if (HasFeature(PP, Extension))
1130 return true;
1131
1132 // If the use of an extension results in an error diagnostic, extensions are
1133 // effectively unavailable, so just return false here.
1134 if (PP.getDiagnostics().getExtensionHandlingBehavior() >=
1135 diag::Severity::Error)
1136 return false;
1137
1138 const LangOptions &LangOpts = PP.getLangOpts();
1139
1140 // Normalize the extension name, __foo__ becomes foo.
1141 if (Extension.startswith("__") && Extension.endswith("__") &&
1142 Extension.size() >= 4)
1143 Extension = Extension.substr(2, Extension.size() - 4);
1144
1145 // Because we inherit the feature list from HasFeature, this string switch
1146 // must be less restrictive than HasFeature's.
1147 #define EXTENSION(Name, Predicate) .Case(#Name, Predicate)
1148 return llvm::StringSwitch<bool>(Extension)
1149 #include "clang/Basic/Features.def"
1150 .Default(false);
1151 #undef EXTENSION
1152 }
1153
1154 /// EvaluateHasIncludeCommon - Process a '__has_include("path")'
1155 /// or '__has_include_next("path")' expression.
1156 /// Returns true if successful.
EvaluateHasIncludeCommon(Token & Tok,IdentifierInfo * II,Preprocessor & PP,const DirectoryLookup * LookupFrom,const FileEntry * LookupFromFile)1157 static bool EvaluateHasIncludeCommon(Token &Tok,
1158 IdentifierInfo *II, Preprocessor &PP,
1159 const DirectoryLookup *LookupFrom,
1160 const FileEntry *LookupFromFile) {
1161 // Save the location of the current token. If a '(' is later found, use
1162 // that location. If not, use the end of this location instead.
1163 SourceLocation LParenLoc = Tok.getLocation();
1164
1165 // These expressions are only allowed within a preprocessor directive.
1166 if (!PP.isParsingIfOrElifDirective()) {
1167 PP.Diag(LParenLoc, diag::err_pp_directive_required) << II;
1168 // Return a valid identifier token.
1169 assert(Tok.is(tok::identifier));
1170 Tok.setIdentifierInfo(II);
1171 return false;
1172 }
1173
1174 // Get '('. If we don't have a '(', try to form a header-name token.
1175 do {
1176 if (PP.LexHeaderName(Tok))
1177 return false;
1178 } while (Tok.getKind() == tok::comment);
1179
1180 // Ensure we have a '('.
1181 if (Tok.isNot(tok::l_paren)) {
1182 // No '(', use end of last token.
1183 LParenLoc = PP.getLocForEndOfToken(LParenLoc);
1184 PP.Diag(LParenLoc, diag::err_pp_expected_after) << II << tok::l_paren;
1185 // If the next token looks like a filename or the start of one,
1186 // assume it is and process it as such.
1187 if (Tok.isNot(tok::header_name))
1188 return false;
1189 } else {
1190 // Save '(' location for possible missing ')' message.
1191 LParenLoc = Tok.getLocation();
1192 if (PP.LexHeaderName(Tok))
1193 return false;
1194 }
1195
1196 if (Tok.isNot(tok::header_name)) {
1197 PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename);
1198 return false;
1199 }
1200
1201 // Reserve a buffer to get the spelling.
1202 SmallString<128> FilenameBuffer;
1203 bool Invalid = false;
1204 StringRef Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid);
1205 if (Invalid)
1206 return false;
1207
1208 SourceLocation FilenameLoc = Tok.getLocation();
1209
1210 // Get ')'.
1211 PP.LexNonComment(Tok);
1212
1213 // Ensure we have a trailing ).
1214 if (Tok.isNot(tok::r_paren)) {
1215 PP.Diag(PP.getLocForEndOfToken(FilenameLoc), diag::err_pp_expected_after)
1216 << II << tok::r_paren;
1217 PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1218 return false;
1219 }
1220
1221 bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename);
1222 // If GetIncludeFilenameSpelling set the start ptr to null, there was an
1223 // error.
1224 if (Filename.empty())
1225 return false;
1226
1227 // Search include directories.
1228 const DirectoryLookup *CurDir;
1229 Optional<FileEntryRef> File =
1230 PP.LookupFile(FilenameLoc, Filename, isAngled, LookupFrom, LookupFromFile,
1231 CurDir, nullptr, nullptr, nullptr, nullptr, nullptr);
1232
1233 if (PPCallbacks *Callbacks = PP.getPPCallbacks()) {
1234 SrcMgr::CharacteristicKind FileType = SrcMgr::C_User;
1235 if (File)
1236 FileType =
1237 PP.getHeaderSearchInfo().getFileDirFlavor(&File->getFileEntry());
1238 Callbacks->HasInclude(FilenameLoc, Filename, isAngled, File, FileType);
1239 }
1240
1241 // Get the result value. A result of true means the file exists.
1242 return File.hasValue();
1243 }
1244
1245 /// EvaluateHasInclude - Process a '__has_include("path")' expression.
1246 /// Returns true if successful.
EvaluateHasInclude(Token & Tok,IdentifierInfo * II,Preprocessor & PP)1247 static bool EvaluateHasInclude(Token &Tok, IdentifierInfo *II,
1248 Preprocessor &PP) {
1249 return EvaluateHasIncludeCommon(Tok, II, PP, nullptr, nullptr);
1250 }
1251
1252 /// EvaluateHasIncludeNext - Process '__has_include_next("path")' expression.
1253 /// Returns true if successful.
EvaluateHasIncludeNext(Token & Tok,IdentifierInfo * II,Preprocessor & PP)1254 static bool EvaluateHasIncludeNext(Token &Tok,
1255 IdentifierInfo *II, Preprocessor &PP) {
1256 // __has_include_next is like __has_include, except that we start
1257 // searching after the current found directory. If we can't do this,
1258 // issue a diagnostic.
1259 // FIXME: Factor out duplication with
1260 // Preprocessor::HandleIncludeNextDirective.
1261 const DirectoryLookup *Lookup = PP.GetCurDirLookup();
1262 const FileEntry *LookupFromFile = nullptr;
1263 if (PP.isInPrimaryFile() && PP.getLangOpts().IsHeaderFile) {
1264 // If the main file is a header, then it's either for PCH/AST generation,
1265 // or libclang opened it. Either way, handle it as a normal include below
1266 // and do not complain about __has_include_next.
1267 } else if (PP.isInPrimaryFile()) {
1268 Lookup = nullptr;
1269 PP.Diag(Tok, diag::pp_include_next_in_primary);
1270 } else if (PP.getCurrentLexerSubmodule()) {
1271 // Start looking up in the directory *after* the one in which the current
1272 // file would be found, if any.
1273 assert(PP.getCurrentLexer() && "#include_next directive in macro?");
1274 LookupFromFile = PP.getCurrentLexer()->getFileEntry();
1275 Lookup = nullptr;
1276 } else if (!Lookup) {
1277 PP.Diag(Tok, diag::pp_include_next_absolute_path);
1278 } else {
1279 // Start looking up in the next directory.
1280 ++Lookup;
1281 }
1282
1283 return EvaluateHasIncludeCommon(Tok, II, PP, Lookup, LookupFromFile);
1284 }
1285
1286 /// Process single-argument builtin feature-like macros that return
1287 /// integer values.
EvaluateFeatureLikeBuiltinMacro(llvm::raw_svector_ostream & OS,Token & Tok,IdentifierInfo * II,Preprocessor & PP,llvm::function_ref<int (Token & Tok,bool & HasLexedNextTok)> Op)1288 static void EvaluateFeatureLikeBuiltinMacro(llvm::raw_svector_ostream& OS,
1289 Token &Tok, IdentifierInfo *II,
1290 Preprocessor &PP,
1291 llvm::function_ref<
1292 int(Token &Tok,
1293 bool &HasLexedNextTok)> Op) {
1294 // Parse the initial '('.
1295 PP.LexUnexpandedToken(Tok);
1296 if (Tok.isNot(tok::l_paren)) {
1297 PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II
1298 << tok::l_paren;
1299
1300 // Provide a dummy '0' value on output stream to elide further errors.
1301 if (!Tok.isOneOf(tok::eof, tok::eod)) {
1302 OS << 0;
1303 Tok.setKind(tok::numeric_constant);
1304 }
1305 return;
1306 }
1307
1308 unsigned ParenDepth = 1;
1309 SourceLocation LParenLoc = Tok.getLocation();
1310 llvm::Optional<int> Result;
1311
1312 Token ResultTok;
1313 bool SuppressDiagnostic = false;
1314 while (true) {
1315 // Parse next token.
1316 PP.LexUnexpandedToken(Tok);
1317
1318 already_lexed:
1319 switch (Tok.getKind()) {
1320 case tok::eof:
1321 case tok::eod:
1322 // Don't provide even a dummy value if the eod or eof marker is
1323 // reached. Simply provide a diagnostic.
1324 PP.Diag(Tok.getLocation(), diag::err_unterm_macro_invoc);
1325 return;
1326
1327 case tok::comma:
1328 if (!SuppressDiagnostic) {
1329 PP.Diag(Tok.getLocation(), diag::err_too_many_args_in_macro_invoc);
1330 SuppressDiagnostic = true;
1331 }
1332 continue;
1333
1334 case tok::l_paren:
1335 ++ParenDepth;
1336 if (Result.hasValue())
1337 break;
1338 if (!SuppressDiagnostic) {
1339 PP.Diag(Tok.getLocation(), diag::err_pp_nested_paren) << II;
1340 SuppressDiagnostic = true;
1341 }
1342 continue;
1343
1344 case tok::r_paren:
1345 if (--ParenDepth > 0)
1346 continue;
1347
1348 // The last ')' has been reached; return the value if one found or
1349 // a diagnostic and a dummy value.
1350 if (Result.hasValue()) {
1351 OS << Result.getValue();
1352 // For strict conformance to __has_cpp_attribute rules, use 'L'
1353 // suffix for dated literals.
1354 if (Result.getValue() > 1)
1355 OS << 'L';
1356 } else {
1357 OS << 0;
1358 if (!SuppressDiagnostic)
1359 PP.Diag(Tok.getLocation(), diag::err_too_few_args_in_macro_invoc);
1360 }
1361 Tok.setKind(tok::numeric_constant);
1362 return;
1363
1364 default: {
1365 // Parse the macro argument, if one not found so far.
1366 if (Result.hasValue())
1367 break;
1368
1369 bool HasLexedNextToken = false;
1370 Result = Op(Tok, HasLexedNextToken);
1371 ResultTok = Tok;
1372 if (HasLexedNextToken)
1373 goto already_lexed;
1374 continue;
1375 }
1376 }
1377
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())
1382 Diag << LastII;
1383 else
1384 Diag << ResultTok.getKind();
1385 Diag << tok::r_paren << ResultTok.getLocation();
1386 }
1387 PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1388 SuppressDiagnostic = true;
1389 }
1390 }
1391 }
1392
1393 /// Helper function to return the IdentifierInfo structure of a Token
1394 /// or generate a diagnostic if none available.
ExpectFeatureIdentifierInfo(Token & Tok,Preprocessor & PP,signed DiagID)1395 static IdentifierInfo *ExpectFeatureIdentifierInfo(Token &Tok,
1396 Preprocessor &PP,
1397 signed DiagID) {
1398 IdentifierInfo *II;
1399 if (!Tok.isAnnotation() && (II = Tok.getIdentifierInfo()))
1400 return II;
1401
1402 PP.Diag(Tok.getLocation(), DiagID);
1403 return nullptr;
1404 }
1405
1406 /// Implements the __is_target_arch builtin macro.
isTargetArch(const TargetInfo & TI,const IdentifierInfo * II)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();
1411 if (TT.isThumb()) {
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)))
1419 return true;
1420 }
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();
1426 }
1427
1428 /// Implements the __is_target_vendor builtin macro.
isTargetVendor(const TargetInfo & TI,const IdentifierInfo * II)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());
1434 }
1435
1436 /// Implements the __is_target_os builtin macro.
isTargetOS(const TargetInfo & TI,const IdentifierInfo * II)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();
1444 }
1445 return TI.getTriple().getOS() == OS.getOS();
1446 }
1447
1448 /// Implements the __is_target_environment builtin macro.
isTargetEnvironment(const TargetInfo & TI,const IdentifierInfo * II)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();
1454 }
1455
remapMacroPath(SmallString<256> & Path,const std::map<std::string,std::string,std::greater<std::string>> & MacroPrefixMap)1456 static void remapMacroPath(
1457 SmallString<256> &Path,
1458 const std::map<std::string, std::string, std::greater<std::string>>
1459 &MacroPrefixMap) {
1460 for (const auto &Entry : MacroPrefixMap)
1461 if (llvm::sys::path::replace_path_prefix(Path, Entry.first, Entry.second))
1462 break;
1463 }
1464
1465 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded
1466 /// as a builtin macro, handle it and return the next token as 'Tok'.
ExpandBuiltinMacro(Token & Tok)1467 void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
1468 // Figure out which token this is.
1469 IdentifierInfo *II = Tok.getIdentifierInfo();
1470 assert(II && "Can't be a macro without id info!");
1471
1472 // If this is an _Pragma or Microsoft __pragma directive, expand it,
1473 // invoke the pragma handler, then lex the token after it.
1474 if (II == Ident_Pragma)
1475 return Handle_Pragma(Tok);
1476 else if (II == Ident__pragma) // in non-MS mode this is null
1477 return HandleMicrosoft__pragma(Tok);
1478
1479 ++NumBuiltinMacroExpanded;
1480
1481 SmallString<128> TmpBuffer;
1482 llvm::raw_svector_ostream OS(TmpBuffer);
1483
1484 // Set up the return result.
1485 Tok.setIdentifierInfo(nullptr);
1486 Tok.clearFlag(Token::NeedsCleaning);
1487 bool IsAtStartOfLine = Tok.isAtStartOfLine();
1488 bool HasLeadingSpace = Tok.hasLeadingSpace();
1489
1490 if (II == Ident__LINE__) {
1491 // C99 6.10.8: "__LINE__: The presumed line number (within the current
1492 // source file) of the current source line (an integer constant)". This can
1493 // be affected by #line.
1494 SourceLocation Loc = Tok.getLocation();
1495
1496 // Advance to the location of the first _, this might not be the first byte
1497 // of the token if it starts with an escaped newline.
1498 Loc = AdvanceToTokenCharacter(Loc, 0);
1499
1500 // One wrinkle here is that GCC expands __LINE__ to location of the *end* of
1501 // a macro expansion. This doesn't matter for object-like macros, but
1502 // can matter for a function-like macro that expands to contain __LINE__.
1503 // Skip down through expansion points until we find a file loc for the
1504 // end of the expansion history.
1505 Loc = SourceMgr.getExpansionRange(Loc).getEnd();
1506 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Loc);
1507
1508 // __LINE__ expands to a simple numeric value.
1509 OS << (PLoc.isValid()? PLoc.getLine() : 1);
1510 Tok.setKind(tok::numeric_constant);
1511 } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__ ||
1512 II == Ident__FILE_NAME__) {
1513 // C99 6.10.8: "__FILE__: The presumed name of the current source file (a
1514 // character string literal)". This can be affected by #line.
1515 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1516
1517 // __BASE_FILE__ is a GNU extension that returns the top of the presumed
1518 // #include stack instead of the current file.
1519 if (II == Ident__BASE_FILE__ && PLoc.isValid()) {
1520 SourceLocation NextLoc = PLoc.getIncludeLoc();
1521 while (NextLoc.isValid()) {
1522 PLoc = SourceMgr.getPresumedLoc(NextLoc);
1523 if (PLoc.isInvalid())
1524 break;
1525
1526 NextLoc = PLoc.getIncludeLoc();
1527 }
1528 }
1529
1530 // Escape this filename. Turn '\' -> '\\' '"' -> '\"'
1531 SmallString<256> FN;
1532 if (PLoc.isValid()) {
1533 // __FILE_NAME__ is a Clang-specific extension that expands to the
1534 // the last part of __FILE__.
1535 if (II == Ident__FILE_NAME__) {
1536 // Try to get the last path component, failing that return the original
1537 // presumed location.
1538 StringRef PLFileName = llvm::sys::path::filename(PLoc.getFilename());
1539 if (PLFileName != "")
1540 FN += PLFileName;
1541 else
1542 FN += PLoc.getFilename();
1543 } else {
1544 FN += PLoc.getFilename();
1545 }
1546 remapMacroPath(FN, PPOpts->MacroPrefixMap);
1547 Lexer::Stringify(FN);
1548 OS << '"' << FN << '"';
1549 }
1550 Tok.setKind(tok::string_literal);
1551 } else if (II == Ident__DATE__) {
1552 Diag(Tok.getLocation(), diag::warn_pp_date_time);
1553 if (!DATELoc.isValid())
1554 ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1555 Tok.setKind(tok::string_literal);
1556 Tok.setLength(strlen("\"Mmm dd yyyy\""));
1557 Tok.setLocation(SourceMgr.createExpansionLoc(DATELoc, Tok.getLocation(),
1558 Tok.getLocation(),
1559 Tok.getLength()));
1560 return;
1561 } else if (II == Ident__TIME__) {
1562 Diag(Tok.getLocation(), diag::warn_pp_date_time);
1563 if (!TIMELoc.isValid())
1564 ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1565 Tok.setKind(tok::string_literal);
1566 Tok.setLength(strlen("\"hh:mm:ss\""));
1567 Tok.setLocation(SourceMgr.createExpansionLoc(TIMELoc, Tok.getLocation(),
1568 Tok.getLocation(),
1569 Tok.getLength()));
1570 return;
1571 } else if (II == Ident__INCLUDE_LEVEL__) {
1572 // Compute the presumed include depth of this token. This can be affected
1573 // by GNU line markers.
1574 unsigned Depth = 0;
1575
1576 PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1577 if (PLoc.isValid()) {
1578 PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1579 for (; PLoc.isValid(); ++Depth)
1580 PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1581 }
1582
1583 // __INCLUDE_LEVEL__ expands to a simple numeric value.
1584 OS << Depth;
1585 Tok.setKind(tok::numeric_constant);
1586 } else if (II == Ident__TIMESTAMP__) {
1587 Diag(Tok.getLocation(), diag::warn_pp_date_time);
1588 // MSVC, ICC, GCC, VisualAge C++ extension. The generated string should be
1589 // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime.
1590
1591 // Get the file that we are lexing out of. If we're currently lexing from
1592 // a macro, dig into the include stack.
1593 const FileEntry *CurFile = nullptr;
1594 PreprocessorLexer *TheLexer = getCurrentFileLexer();
1595
1596 if (TheLexer)
1597 CurFile = SourceMgr.getFileEntryForID(TheLexer->getFileID());
1598
1599 const char *Result;
1600 if (CurFile) {
1601 time_t TT = CurFile->getModificationTime();
1602 struct tm *TM = localtime(&TT);
1603 Result = asctime(TM);
1604 } else {
1605 Result = "??? ??? ?? ??:??:?? ????\n";
1606 }
1607 // Surround the string with " and strip the trailing newline.
1608 OS << '"' << StringRef(Result).drop_back() << '"';
1609 Tok.setKind(tok::string_literal);
1610 } else if (II == Ident__COUNTER__) {
1611 // __COUNTER__ expands to a simple numeric value.
1612 OS << CounterValue++;
1613 Tok.setKind(tok::numeric_constant);
1614 } else if (II == Ident__has_feature) {
1615 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1616 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1617 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1618 diag::err_feature_check_malformed);
1619 return II && HasFeature(*this, II->getName());
1620 });
1621 } else if (II == Ident__has_extension) {
1622 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1623 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1624 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1625 diag::err_feature_check_malformed);
1626 return II && HasExtension(*this, II->getName());
1627 });
1628 } else if (II == Ident__has_builtin) {
1629 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1630 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1631 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1632 diag::err_feature_check_malformed);
1633 if (!II)
1634 return false;
1635 else if (II->getBuiltinID() != 0) {
1636 switch (II->getBuiltinID()) {
1637 case Builtin::BI__builtin_operator_new:
1638 case Builtin::BI__builtin_operator_delete:
1639 // denotes date of behavior change to support calling arbitrary
1640 // usual allocation and deallocation functions. Required by libc++
1641 return 201802;
1642 default:
1643 return true;
1644 }
1645 return true;
1646 } else if (II->getTokenID() != tok::identifier ||
1647 II->hasRevertedTokenIDToIdentifier()) {
1648 // Treat all keywords that introduce a custom syntax of the form
1649 //
1650 // '__some_keyword' '(' [...] ')'
1651 //
1652 // as being "builtin functions", even if the syntax isn't a valid
1653 // function call (for example, because the builtin takes a type
1654 // argument).
1655 if (II->getName().startswith("__builtin_") ||
1656 II->getName().startswith("__is_") ||
1657 II->getName().startswith("__has_"))
1658 return true;
1659 return llvm::StringSwitch<bool>(II->getName())
1660 .Case("__array_rank", true)
1661 .Case("__array_extent", true)
1662 .Case("__reference_binds_to_temporary", true)
1663 .Case("__underlying_type", true)
1664 .Default(false);
1665 } else {
1666 return llvm::StringSwitch<bool>(II->getName())
1667 // Report builtin templates as being builtins.
1668 .Case("__make_integer_seq", getLangOpts().CPlusPlus)
1669 .Case("__type_pack_element", getLangOpts().CPlusPlus)
1670 // Likewise for some builtin preprocessor macros.
1671 // FIXME: This is inconsistent; we usually suggest detecting
1672 // builtin macros via #ifdef. Don't add more cases here.
1673 .Case("__is_target_arch", true)
1674 .Case("__is_target_vendor", true)
1675 .Case("__is_target_os", true)
1676 .Case("__is_target_environment", true)
1677 .Default(false);
1678 }
1679 });
1680 } else if (II == Ident__is_identifier) {
1681 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1682 [](Token &Tok, bool &HasLexedNextToken) -> int {
1683 return Tok.is(tok::identifier);
1684 });
1685 } else if (II == Ident__has_attribute) {
1686 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1687 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1688 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1689 diag::err_feature_check_malformed);
1690 return II ? hasAttribute(AttrSyntax::GNU, nullptr, II,
1691 getTargetInfo(), getLangOpts()) : 0;
1692 });
1693 } else if (II == Ident__has_declspec) {
1694 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1695 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1696 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1697 diag::err_feature_check_malformed);
1698 if (II) {
1699 const LangOptions &LangOpts = getLangOpts();
1700 return LangOpts.DeclSpecKeyword &&
1701 hasAttribute(AttrSyntax::Declspec, nullptr, II,
1702 getTargetInfo(), LangOpts);
1703 }
1704
1705 return false;
1706 });
1707 } else if (II == Ident__has_cpp_attribute ||
1708 II == Ident__has_c_attribute) {
1709 bool IsCXX = II == Ident__has_cpp_attribute;
1710 EvaluateFeatureLikeBuiltinMacro(
1711 OS, Tok, II, *this, [&](Token &Tok, bool &HasLexedNextToken) -> int {
1712 IdentifierInfo *ScopeII = nullptr;
1713 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1714 Tok, *this, diag::err_feature_check_malformed);
1715 if (!II)
1716 return false;
1717
1718 // It is possible to receive a scope token. Read the "::", if it is
1719 // available, and the subsequent identifier.
1720 LexUnexpandedToken(Tok);
1721 if (Tok.isNot(tok::coloncolon))
1722 HasLexedNextToken = true;
1723 else {
1724 ScopeII = II;
1725 LexUnexpandedToken(Tok);
1726 II = ExpectFeatureIdentifierInfo(Tok, *this,
1727 diag::err_feature_check_malformed);
1728 }
1729
1730 AttrSyntax Syntax = IsCXX ? AttrSyntax::CXX : AttrSyntax::C;
1731 return II ? hasAttribute(Syntax, ScopeII, II, getTargetInfo(),
1732 getLangOpts())
1733 : 0;
1734 });
1735 } else if (II == Ident__has_include ||
1736 II == Ident__has_include_next) {
1737 // The argument to these two builtins should be a parenthesized
1738 // file name string literal using angle brackets (<>) or
1739 // double-quotes ("").
1740 bool Value;
1741 if (II == Ident__has_include)
1742 Value = EvaluateHasInclude(Tok, II, *this);
1743 else
1744 Value = EvaluateHasIncludeNext(Tok, II, *this);
1745
1746 if (Tok.isNot(tok::r_paren))
1747 return;
1748 OS << (int)Value;
1749 Tok.setKind(tok::numeric_constant);
1750 } else if (II == Ident__has_warning) {
1751 // The argument should be a parenthesized string literal.
1752 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1753 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1754 std::string WarningName;
1755 SourceLocation StrStartLoc = Tok.getLocation();
1756
1757 HasLexedNextToken = Tok.is(tok::string_literal);
1758 if (!FinishLexStringLiteral(Tok, WarningName, "'__has_warning'",
1759 /*AllowMacroExpansion=*/false))
1760 return false;
1761
1762 // FIXME: Should we accept "-R..." flags here, or should that be
1763 // handled by a separate __has_remark?
1764 if (WarningName.size() < 3 || WarningName[0] != '-' ||
1765 WarningName[1] != 'W') {
1766 Diag(StrStartLoc, diag::warn_has_warning_invalid_option);
1767 return false;
1768 }
1769
1770 // Finally, check if the warning flags maps to a diagnostic group.
1771 // We construct a SmallVector here to talk to getDiagnosticIDs().
1772 // Although we don't use the result, this isn't a hot path, and not
1773 // worth special casing.
1774 SmallVector<diag::kind, 10> Diags;
1775 return !getDiagnostics().getDiagnosticIDs()->
1776 getDiagnosticsInGroup(diag::Flavor::WarningOrError,
1777 WarningName.substr(2), Diags);
1778 });
1779 } else if (II == Ident__building_module) {
1780 // The argument to this builtin should be an identifier. The
1781 // builtin evaluates to 1 when that identifier names the module we are
1782 // currently building.
1783 EvaluateFeatureLikeBuiltinMacro(OS, Tok, II, *this,
1784 [this](Token &Tok, bool &HasLexedNextToken) -> int {
1785 IdentifierInfo *II = ExpectFeatureIdentifierInfo(Tok, *this,
1786 diag::err_expected_id_building_module);
1787 return getLangOpts().isCompilingModule() && II &&
1788 (II->getName() == getLangOpts().CurrentModule);
1789 });
1790 } else if (II == Ident__MODULE__) {
1791 // The current module as an identifier.
1792 OS << getLangOpts().CurrentModule;
1793 IdentifierInfo *ModuleII = getIdentifierInfo(getLangOpts().CurrentModule);
1794 Tok.setIdentifierInfo(ModuleII);
1795 Tok.setKind(ModuleII->getTokenID());
1796 } else if (II == Ident__identifier) {
1797 SourceLocation Loc = Tok.getLocation();
1798
1799 // We're expecting '__identifier' '(' identifier ')'. Try to recover
1800 // if the parens are missing.
1801 LexNonComment(Tok);
1802 if (Tok.isNot(tok::l_paren)) {
1803 // No '(', use end of last token.
1804 Diag(getLocForEndOfToken(Loc), diag::err_pp_expected_after)
1805 << II << tok::l_paren;
1806 // If the next token isn't valid as our argument, we can't recover.
1807 if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1808 Tok.setKind(tok::identifier);
1809 return;
1810 }
1811
1812 SourceLocation LParenLoc = Tok.getLocation();
1813 LexNonComment(Tok);
1814
1815 if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1816 Tok.setKind(tok::identifier);
1817 else if (Tok.is(tok::string_literal) && !Tok.hasUDSuffix()) {
1818 StringLiteralParser Literal(Tok, *this);
1819 if (Literal.hadError)
1820 return;
1821
1822 Tok.setIdentifierInfo(getIdentifierInfo(Literal.GetString()));
1823 Tok.setKind(tok::identifier);
1824 } else {
1825 Diag(Tok.getLocation(), diag::err_pp_identifier_arg_not_identifier)
1826 << Tok.getKind();
1827 // Don't walk past anything that's not a real token.
1828 if (Tok.isOneOf(tok::eof, tok::eod) || Tok.isAnnotation())
1829 return;
1830 }
1831
1832 // Discard the ')', preserving 'Tok' as our result.
1833 Token RParen;
1834 LexNonComment(RParen);
1835 if (RParen.isNot(tok::r_paren)) {
1836 Diag(getLocForEndOfToken(Tok.getLocation()), diag::err_pp_expected_after)
1837 << Tok.getKind() << tok::r_paren;
1838 Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1839 }
1840 return;
1841 } else if (II == Ident__is_target_arch) {
1842 EvaluateFeatureLikeBuiltinMacro(
1843 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1844 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1845 Tok, *this, diag::err_feature_check_malformed);
1846 return II && isTargetArch(getTargetInfo(), II);
1847 });
1848 } else if (II == Ident__is_target_vendor) {
1849 EvaluateFeatureLikeBuiltinMacro(
1850 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1851 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1852 Tok, *this, diag::err_feature_check_malformed);
1853 return II && isTargetVendor(getTargetInfo(), II);
1854 });
1855 } else if (II == Ident__is_target_os) {
1856 EvaluateFeatureLikeBuiltinMacro(
1857 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1858 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1859 Tok, *this, diag::err_feature_check_malformed);
1860 return II && isTargetOS(getTargetInfo(), II);
1861 });
1862 } else if (II == Ident__is_target_environment) {
1863 EvaluateFeatureLikeBuiltinMacro(
1864 OS, Tok, II, *this, [this](Token &Tok, bool &HasLexedNextToken) -> int {
1865 IdentifierInfo *II = ExpectFeatureIdentifierInfo(
1866 Tok, *this, diag::err_feature_check_malformed);
1867 return II && isTargetEnvironment(getTargetInfo(), II);
1868 });
1869 } else {
1870 llvm_unreachable("Unknown identifier!");
1871 }
1872 CreateString(OS.str(), Tok, Tok.getLocation(), Tok.getLocation());
1873 Tok.setFlagValue(Token::StartOfLine, IsAtStartOfLine);
1874 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace);
1875 }
1876
markMacroAsUsed(MacroInfo * MI)1877 void Preprocessor::markMacroAsUsed(MacroInfo *MI) {
1878 // If the 'used' status changed, and the macro requires 'unused' warning,
1879 // remove its SourceLocation from the warn-for-unused-macro locations.
1880 if (MI->isWarnIfUnused() && !MI->isUsed())
1881 WarnUnusedMacroLocs.erase(MI->getDefinitionLoc());
1882 MI->setIsUsed(true);
1883 }
1884