1 //===--- MacroExpansion.cpp - Top level Macro Expansion -------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements the top level handling of macro expansion for the
11 // preprocessor.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "clang/Lex/Preprocessor.h"
16 #include "clang/Basic/Attributes.h"
17 #include "clang/Basic/FileManager.h"
18 #include "clang/Basic/SourceManager.h"
19 #include "clang/Basic/TargetInfo.h"
20 #include "clang/Lex/CodeCompletionHandler.h"
21 #include "clang/Lex/ExternalPreprocessorSource.h"
22 #include "clang/Lex/LexDiagnostic.h"
23 #include "clang/Lex/MacroArgs.h"
24 #include "clang/Lex/MacroInfo.h"
25 #include "llvm/ADT/STLExtras.h"
26 #include "llvm/ADT/SmallString.h"
27 #include "llvm/ADT/StringSwitch.h"
28 #include "llvm/Config/llvm-config.h"
29 #include "llvm/Support/ErrorHandling.h"
30 #include "llvm/Support/Format.h"
31 #include "llvm/Support/raw_ostream.h"
32 #include <cstdio>
33 #include <ctime>
34 using namespace clang;
35 
36 MacroDirective *
37 Preprocessor::getMacroDirectiveHistory(const IdentifierInfo *II) const {
38   assert(II->hadMacroDefinition() && "Identifier has not been not a macro!");
39 
40   macro_iterator Pos = Macros.find(II);
41   assert(Pos != Macros.end() && "Identifier macro info is missing!");
42   return Pos->second;
43 }
44 
45 void Preprocessor::appendMacroDirective(IdentifierInfo *II, MacroDirective *MD){
46   assert(MD && "MacroDirective should be non-zero!");
47   assert(!MD->getPrevious() && "Already attached to a MacroDirective history.");
48 
49   MacroDirective *&StoredMD = Macros[II];
50   MD->setPrevious(StoredMD);
51   StoredMD = MD;
52   // Setup the identifier as having associated macro history.
53   II->setHasMacroDefinition(true);
54   if (!MD->isDefined())
55     II->setHasMacroDefinition(false);
56   bool isImportedMacro = isa<DefMacroDirective>(MD) &&
57                          cast<DefMacroDirective>(MD)->isImported();
58   if (II->isFromAST() && !isImportedMacro)
59     II->setChangedSinceDeserialization();
60 }
61 
62 void Preprocessor::setLoadedMacroDirective(IdentifierInfo *II,
63                                            MacroDirective *MD) {
64   assert(II && MD);
65   MacroDirective *&StoredMD = Macros[II];
66   assert(!StoredMD &&
67          "the macro history was modified before initializing it from a pch");
68   StoredMD = MD;
69   // Setup the identifier as having associated macro history.
70   II->setHasMacroDefinition(true);
71   if (!MD->isDefined())
72     II->setHasMacroDefinition(false);
73 }
74 
75 /// RegisterBuiltinMacro - Register the specified identifier in the identifier
76 /// table and mark it as a builtin macro to be expanded.
77 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){
78   // Get the identifier.
79   IdentifierInfo *Id = PP.getIdentifierInfo(Name);
80 
81   // Mark it as being a macro that is builtin.
82   MacroInfo *MI = PP.AllocateMacroInfo(SourceLocation());
83   MI->setIsBuiltinMacro();
84   PP.appendDefMacroDirective(Id, MI);
85   return Id;
86 }
87 
88 
89 /// RegisterBuiltinMacros - Register builtin macros, such as __LINE__ with the
90 /// identifier table.
91 void Preprocessor::RegisterBuiltinMacros() {
92   Ident__LINE__ = RegisterBuiltinMacro(*this, "__LINE__");
93   Ident__FILE__ = RegisterBuiltinMacro(*this, "__FILE__");
94   Ident__DATE__ = RegisterBuiltinMacro(*this, "__DATE__");
95   Ident__TIME__ = RegisterBuiltinMacro(*this, "__TIME__");
96   Ident__COUNTER__ = RegisterBuiltinMacro(*this, "__COUNTER__");
97   Ident_Pragma  = RegisterBuiltinMacro(*this, "_Pragma");
98 
99   // C++ Standing Document Extensions.
100   Ident__has_cpp_attribute = RegisterBuiltinMacro(*this, "__has_cpp_attribute");
101 
102   // GCC Extensions.
103   Ident__BASE_FILE__     = RegisterBuiltinMacro(*this, "__BASE_FILE__");
104   Ident__INCLUDE_LEVEL__ = RegisterBuiltinMacro(*this, "__INCLUDE_LEVEL__");
105   Ident__TIMESTAMP__     = RegisterBuiltinMacro(*this, "__TIMESTAMP__");
106 
107   // Microsoft Extensions.
108   if (LangOpts.MicrosoftExt) {
109     Ident__identifier = RegisterBuiltinMacro(*this, "__identifier");
110     Ident__pragma = RegisterBuiltinMacro(*this, "__pragma");
111   } else {
112     Ident__identifier = nullptr;
113     Ident__pragma = nullptr;
114   }
115 
116   // Clang Extensions.
117   Ident__has_feature      = RegisterBuiltinMacro(*this, "__has_feature");
118   Ident__has_extension    = RegisterBuiltinMacro(*this, "__has_extension");
119   Ident__has_builtin      = RegisterBuiltinMacro(*this, "__has_builtin");
120   Ident__has_attribute    = RegisterBuiltinMacro(*this, "__has_attribute");
121   Ident__has_include      = RegisterBuiltinMacro(*this, "__has_include");
122   Ident__has_include_next = RegisterBuiltinMacro(*this, "__has_include_next");
123   Ident__has_warning      = RegisterBuiltinMacro(*this, "__has_warning");
124   Ident__is_identifier    = RegisterBuiltinMacro(*this, "__is_identifier");
125 
126   // Modules.
127   if (LangOpts.Modules) {
128     Ident__building_module  = RegisterBuiltinMacro(*this, "__building_module");
129 
130     // __MODULE__
131     if (!LangOpts.CurrentModule.empty())
132       Ident__MODULE__ = RegisterBuiltinMacro(*this, "__MODULE__");
133     else
134       Ident__MODULE__ = nullptr;
135   } else {
136     Ident__building_module = nullptr;
137     Ident__MODULE__ = nullptr;
138   }
139 }
140 
141 /// isTrivialSingleTokenExpansion - Return true if MI, which has a single token
142 /// in its expansion, currently expands to that token literally.
143 static bool isTrivialSingleTokenExpansion(const MacroInfo *MI,
144                                           const IdentifierInfo *MacroIdent,
145                                           Preprocessor &PP) {
146   IdentifierInfo *II = MI->getReplacementToken(0).getIdentifierInfo();
147 
148   // If the token isn't an identifier, it's always literally expanded.
149   if (!II) return true;
150 
151   // If the information about this identifier is out of date, update it from
152   // the external source.
153   if (II->isOutOfDate())
154     PP.getExternalSource()->updateOutOfDateIdentifier(*II);
155 
156   // If the identifier is a macro, and if that macro is enabled, it may be
157   // expanded so it's not a trivial expansion.
158   if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled() &&
159       // Fast expanding "#define X X" is ok, because X would be disabled.
160       II != MacroIdent)
161     return false;
162 
163   // If this is an object-like macro invocation, it is safe to trivially expand
164   // it.
165   if (MI->isObjectLike()) return true;
166 
167   // If this is a function-like macro invocation, it's safe to trivially expand
168   // as long as the identifier is not a macro argument.
169   for (MacroInfo::arg_iterator I = MI->arg_begin(), E = MI->arg_end();
170        I != E; ++I)
171     if (*I == II)
172       return false;   // Identifier is a macro argument.
173 
174   return true;
175 }
176 
177 
178 /// isNextPPTokenLParen - Determine whether the next preprocessor token to be
179 /// lexed is a '('.  If so, consume the token and return true, if not, this
180 /// method should have no observable side-effect on the lexed tokens.
181 bool Preprocessor::isNextPPTokenLParen() {
182   // Do some quick tests for rejection cases.
183   unsigned Val;
184   if (CurLexer)
185     Val = CurLexer->isNextPPTokenLParen();
186   else if (CurPTHLexer)
187     Val = CurPTHLexer->isNextPPTokenLParen();
188   else
189     Val = CurTokenLexer->isNextTokenLParen();
190 
191   if (Val == 2) {
192     // We have run off the end.  If it's a source file we don't
193     // examine enclosing ones (C99 5.1.1.2p4).  Otherwise walk up the
194     // macro stack.
195     if (CurPPLexer)
196       return false;
197     for (unsigned i = IncludeMacroStack.size(); i != 0; --i) {
198       IncludeStackInfo &Entry = IncludeMacroStack[i-1];
199       if (Entry.TheLexer)
200         Val = Entry.TheLexer->isNextPPTokenLParen();
201       else if (Entry.ThePTHLexer)
202         Val = Entry.ThePTHLexer->isNextPPTokenLParen();
203       else
204         Val = Entry.TheTokenLexer->isNextTokenLParen();
205 
206       if (Val != 2)
207         break;
208 
209       // Ran off the end of a source file?
210       if (Entry.ThePPLexer)
211         return false;
212     }
213   }
214 
215   // Okay, if we know that the token is a '(', lex it and return.  Otherwise we
216   // have found something that isn't a '(' or we found the end of the
217   // translation unit.  In either case, return false.
218   return Val == 1;
219 }
220 
221 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be
222 /// expanded as a macro, handle it and return the next token as 'Identifier'.
223 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
224                                                  MacroDirective *MD) {
225   MacroDirective::DefInfo Def = MD->getDefinition();
226   assert(Def.isValid());
227   MacroInfo *MI = Def.getMacroInfo();
228 
229   // If this is a macro expansion in the "#if !defined(x)" line for the file,
230   // then the macro could expand to different things in other contexts, we need
231   // to disable the optimization in this case.
232   if (CurPPLexer) CurPPLexer->MIOpt.ExpandedMacro();
233 
234   // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially.
235   if (MI->isBuiltinMacro()) {
236     if (Callbacks) Callbacks->MacroExpands(Identifier, MD,
237                                            Identifier.getLocation(),
238                                            /*Args=*/nullptr);
239     ExpandBuiltinMacro(Identifier);
240     return true;
241   }
242 
243   /// Args - If this is a function-like macro expansion, this contains,
244   /// for each macro argument, the list of tokens that were provided to the
245   /// invocation.
246   MacroArgs *Args = nullptr;
247 
248   // Remember where the end of the expansion occurred.  For an object-like
249   // macro, this is the identifier.  For a function-like macro, this is the ')'.
250   SourceLocation ExpansionEnd = Identifier.getLocation();
251 
252   // If this is a function-like macro, read the arguments.
253   if (MI->isFunctionLike()) {
254     // Remember that we are now parsing the arguments to a macro invocation.
255     // Preprocessor directives used inside macro arguments are not portable, and
256     // this enables the warning.
257     InMacroArgs = true;
258     Args = ReadFunctionLikeMacroArgs(Identifier, MI, ExpansionEnd);
259 
260     // Finished parsing args.
261     InMacroArgs = false;
262 
263     // If there was an error parsing the arguments, bail out.
264     if (!Args) return true;
265 
266     ++NumFnMacroExpanded;
267   } else {
268     ++NumMacroExpanded;
269   }
270 
271   // Notice that this macro has been used.
272   markMacroAsUsed(MI);
273 
274   // Remember where the token is expanded.
275   SourceLocation ExpandLoc = Identifier.getLocation();
276   SourceRange ExpansionRange(ExpandLoc, ExpansionEnd);
277 
278   if (Callbacks) {
279     if (InMacroArgs) {
280       // We can have macro expansion inside a conditional directive while
281       // reading the function macro arguments. To ensure, in that case, that
282       // MacroExpands callbacks still happen in source order, queue this
283       // callback to have it happen after the function macro callback.
284       DelayedMacroExpandsCallbacks.push_back(
285                               MacroExpandsInfo(Identifier, MD, ExpansionRange));
286     } else {
287       Callbacks->MacroExpands(Identifier, MD, ExpansionRange, Args);
288       if (!DelayedMacroExpandsCallbacks.empty()) {
289         for (unsigned i=0, e = DelayedMacroExpandsCallbacks.size(); i!=e; ++i) {
290           MacroExpandsInfo &Info = DelayedMacroExpandsCallbacks[i];
291           // FIXME: We lose macro args info with delayed callback.
292           Callbacks->MacroExpands(Info.Tok, Info.MD, Info.Range,
293                                   /*Args=*/nullptr);
294         }
295         DelayedMacroExpandsCallbacks.clear();
296       }
297     }
298   }
299 
300   // If the macro definition is ambiguous, complain.
301   if (Def.getDirective()->isAmbiguous()) {
302     Diag(Identifier, diag::warn_pp_ambiguous_macro)
303       << Identifier.getIdentifierInfo();
304     Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen)
305       << Identifier.getIdentifierInfo();
306     for (MacroDirective::DefInfo PrevDef = Def.getPreviousDefinition();
307          PrevDef && !PrevDef.isUndefined();
308          PrevDef = PrevDef.getPreviousDefinition()) {
309       Diag(PrevDef.getMacroInfo()->getDefinitionLoc(),
310            diag::note_pp_ambiguous_macro_other)
311         << Identifier.getIdentifierInfo();
312       if (!PrevDef.getDirective()->isAmbiguous())
313         break;
314     }
315   }
316 
317   // If we started lexing a macro, enter the macro expansion body.
318 
319   // If this macro expands to no tokens, don't bother to push it onto the
320   // expansion stack, only to take it right back off.
321   if (MI->getNumTokens() == 0) {
322     // No need for arg info.
323     if (Args) Args->destroy(*this);
324 
325     // Propagate whitespace info as if we had pushed, then popped,
326     // a macro context.
327     Identifier.setFlag(Token::LeadingEmptyMacro);
328     PropagateLineStartLeadingSpaceInfo(Identifier);
329     ++NumFastMacroExpanded;
330     return false;
331   } else if (MI->getNumTokens() == 1 &&
332              isTrivialSingleTokenExpansion(MI, Identifier.getIdentifierInfo(),
333                                            *this)) {
334     // Otherwise, if this macro expands into a single trivially-expanded
335     // token: expand it now.  This handles common cases like
336     // "#define VAL 42".
337 
338     // No need for arg info.
339     if (Args) Args->destroy(*this);
340 
341     // Propagate the isAtStartOfLine/hasLeadingSpace markers of the macro
342     // identifier to the expanded token.
343     bool isAtStartOfLine = Identifier.isAtStartOfLine();
344     bool hasLeadingSpace = Identifier.hasLeadingSpace();
345 
346     // Replace the result token.
347     Identifier = MI->getReplacementToken(0);
348 
349     // Restore the StartOfLine/LeadingSpace markers.
350     Identifier.setFlagValue(Token::StartOfLine , isAtStartOfLine);
351     Identifier.setFlagValue(Token::LeadingSpace, hasLeadingSpace);
352 
353     // Update the tokens location to include both its expansion and physical
354     // locations.
355     SourceLocation Loc =
356       SourceMgr.createExpansionLoc(Identifier.getLocation(), ExpandLoc,
357                                    ExpansionEnd,Identifier.getLength());
358     Identifier.setLocation(Loc);
359 
360     // If this is a disabled macro or #define X X, we must mark the result as
361     // unexpandable.
362     if (IdentifierInfo *NewII = Identifier.getIdentifierInfo()) {
363       if (MacroInfo *NewMI = getMacroInfo(NewII))
364         if (!NewMI->isEnabled() || NewMI == MI) {
365           Identifier.setFlag(Token::DisableExpand);
366           // Don't warn for "#define X X" like "#define bool bool" from
367           // stdbool.h.
368           if (NewMI != MI || MI->isFunctionLike())
369             Diag(Identifier, diag::pp_disabled_macro_expansion);
370         }
371     }
372 
373     // Since this is not an identifier token, it can't be macro expanded, so
374     // we're done.
375     ++NumFastMacroExpanded;
376     return true;
377   }
378 
379   // Start expanding the macro.
380   EnterMacro(Identifier, ExpansionEnd, MI, Args);
381   return false;
382 }
383 
384 enum Bracket {
385   Brace,
386   Paren
387 };
388 
389 /// CheckMatchedBrackets - Returns true if the braces and parentheses in the
390 /// token vector are properly nested.
391 static bool CheckMatchedBrackets(const SmallVectorImpl<Token> &Tokens) {
392   SmallVector<Bracket, 8> Brackets;
393   for (SmallVectorImpl<Token>::const_iterator I = Tokens.begin(),
394                                               E = Tokens.end();
395        I != E; ++I) {
396     if (I->is(tok::l_paren)) {
397       Brackets.push_back(Paren);
398     } else if (I->is(tok::r_paren)) {
399       if (Brackets.empty() || Brackets.back() == Brace)
400         return false;
401       Brackets.pop_back();
402     } else if (I->is(tok::l_brace)) {
403       Brackets.push_back(Brace);
404     } else if (I->is(tok::r_brace)) {
405       if (Brackets.empty() || Brackets.back() == Paren)
406         return false;
407       Brackets.pop_back();
408     }
409   }
410   if (!Brackets.empty())
411     return false;
412   return true;
413 }
414 
415 /// GenerateNewArgTokens - Returns true if OldTokens can be converted to a new
416 /// vector of tokens in NewTokens.  The new number of arguments will be placed
417 /// in NumArgs and the ranges which need to surrounded in parentheses will be
418 /// in ParenHints.
419 /// Returns false if the token stream cannot be changed.  If this is because
420 /// of an initializer list starting a macro argument, the range of those
421 /// initializer lists will be place in InitLists.
422 static bool GenerateNewArgTokens(Preprocessor &PP,
423                                  SmallVectorImpl<Token> &OldTokens,
424                                  SmallVectorImpl<Token> &NewTokens,
425                                  unsigned &NumArgs,
426                                  SmallVectorImpl<SourceRange> &ParenHints,
427                                  SmallVectorImpl<SourceRange> &InitLists) {
428   if (!CheckMatchedBrackets(OldTokens))
429     return false;
430 
431   // Once it is known that the brackets are matched, only a simple count of the
432   // braces is needed.
433   unsigned Braces = 0;
434 
435   // First token of a new macro argument.
436   SmallVectorImpl<Token>::iterator ArgStartIterator = OldTokens.begin();
437 
438   // First closing brace in a new macro argument.  Used to generate
439   // SourceRanges for InitLists.
440   SmallVectorImpl<Token>::iterator ClosingBrace = OldTokens.end();
441   NumArgs = 0;
442   Token TempToken;
443   // Set to true when a macro separator token is found inside a braced list.
444   // If true, the fixed argument spans multiple old arguments and ParenHints
445   // will be updated.
446   bool FoundSeparatorToken = false;
447   for (SmallVectorImpl<Token>::iterator I = OldTokens.begin(),
448                                         E = OldTokens.end();
449        I != E; ++I) {
450     if (I->is(tok::l_brace)) {
451       ++Braces;
452     } else if (I->is(tok::r_brace)) {
453       --Braces;
454       if (Braces == 0 && ClosingBrace == E && FoundSeparatorToken)
455         ClosingBrace = I;
456     } else if (I->is(tok::eof)) {
457       // EOF token is used to separate macro arguments
458       if (Braces != 0) {
459         // Assume comma separator is actually braced list separator and change
460         // it back to a comma.
461         FoundSeparatorToken = true;
462         I->setKind(tok::comma);
463         I->setLength(1);
464       } else { // Braces == 0
465         // Separator token still separates arguments.
466         ++NumArgs;
467 
468         // If the argument starts with a brace, it can't be fixed with
469         // parentheses.  A different diagnostic will be given.
470         if (FoundSeparatorToken && ArgStartIterator->is(tok::l_brace)) {
471           InitLists.push_back(
472               SourceRange(ArgStartIterator->getLocation(),
473                           PP.getLocForEndOfToken(ClosingBrace->getLocation())));
474           ClosingBrace = E;
475         }
476 
477         // Add left paren
478         if (FoundSeparatorToken) {
479           TempToken.startToken();
480           TempToken.setKind(tok::l_paren);
481           TempToken.setLocation(ArgStartIterator->getLocation());
482           TempToken.setLength(0);
483           NewTokens.push_back(TempToken);
484         }
485 
486         // Copy over argument tokens
487         NewTokens.insert(NewTokens.end(), ArgStartIterator, I);
488 
489         // Add right paren and store the paren locations in ParenHints
490         if (FoundSeparatorToken) {
491           SourceLocation Loc = PP.getLocForEndOfToken((I - 1)->getLocation());
492           TempToken.startToken();
493           TempToken.setKind(tok::r_paren);
494           TempToken.setLocation(Loc);
495           TempToken.setLength(0);
496           NewTokens.push_back(TempToken);
497           ParenHints.push_back(SourceRange(ArgStartIterator->getLocation(),
498                                            Loc));
499         }
500 
501         // Copy separator token
502         NewTokens.push_back(*I);
503 
504         // Reset values
505         ArgStartIterator = I + 1;
506         FoundSeparatorToken = false;
507       }
508     }
509   }
510 
511   return !ParenHints.empty() && InitLists.empty();
512 }
513 
514 /// ReadFunctionLikeMacroArgs - After reading "MACRO" and knowing that the next
515 /// token is the '(' of the macro, this method is invoked to read all of the
516 /// actual arguments specified for the macro invocation.  This returns null on
517 /// error.
518 MacroArgs *Preprocessor::ReadFunctionLikeMacroArgs(Token &MacroName,
519                                                    MacroInfo *MI,
520                                                    SourceLocation &MacroEnd) {
521   // The number of fixed arguments to parse.
522   unsigned NumFixedArgsLeft = MI->getNumArgs();
523   bool isVariadic = MI->isVariadic();
524 
525   // Outer loop, while there are more arguments, keep reading them.
526   Token Tok;
527 
528   // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
529   // an argument value in a macro could expand to ',' or '(' or ')'.
530   LexUnexpandedToken(Tok);
531   assert(Tok.is(tok::l_paren) && "Error computing l-paren-ness?");
532 
533   // ArgTokens - Build up a list of tokens that make up each argument.  Each
534   // argument is separated by an EOF token.  Use a SmallVector so we can avoid
535   // heap allocations in the common case.
536   SmallVector<Token, 64> ArgTokens;
537   bool ContainsCodeCompletionTok = false;
538 
539   SourceLocation TooManyArgsLoc;
540 
541   unsigned NumActuals = 0;
542   while (Tok.isNot(tok::r_paren)) {
543     if (ContainsCodeCompletionTok && (Tok.is(tok::eof) || Tok.is(tok::eod)))
544       break;
545 
546     assert((Tok.is(tok::l_paren) || Tok.is(tok::comma)) &&
547            "only expect argument separators here");
548 
549     unsigned ArgTokenStart = ArgTokens.size();
550     SourceLocation ArgStartLoc = Tok.getLocation();
551 
552     // C99 6.10.3p11: Keep track of the number of l_parens we have seen.  Note
553     // that we already consumed the first one.
554     unsigned NumParens = 0;
555 
556     while (1) {
557       // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
558       // an argument value in a macro could expand to ',' or '(' or ')'.
559       LexUnexpandedToken(Tok);
560 
561       if (Tok.is(tok::eof) || Tok.is(tok::eod)) { // "#if f(<eof>" & "#if f(\n"
562         if (!ContainsCodeCompletionTok) {
563           Diag(MacroName, diag::err_unterm_macro_invoc);
564           Diag(MI->getDefinitionLoc(), diag::note_macro_here)
565             << MacroName.getIdentifierInfo();
566           // Do not lose the EOF/EOD.  Return it to the client.
567           MacroName = Tok;
568           return nullptr;
569         } else {
570           // Do not lose the EOF/EOD.
571           Token *Toks = new Token[1];
572           Toks[0] = Tok;
573           EnterTokenStream(Toks, 1, true, true);
574           break;
575         }
576       } else if (Tok.is(tok::r_paren)) {
577         // If we found the ) token, the macro arg list is done.
578         if (NumParens-- == 0) {
579           MacroEnd = Tok.getLocation();
580           break;
581         }
582       } else if (Tok.is(tok::l_paren)) {
583         ++NumParens;
584       } else if (Tok.is(tok::comma) && NumParens == 0 &&
585                  !(Tok.getFlags() & Token::IgnoredComma)) {
586         // In Microsoft-compatibility mode, single commas from nested macro
587         // expansions should not be considered as argument separators. We test
588         // for this with the IgnoredComma token flag above.
589 
590         // Comma ends this argument if there are more fixed arguments expected.
591         // However, if this is a variadic macro, and this is part of the
592         // variadic part, then the comma is just an argument token.
593         if (!isVariadic) break;
594         if (NumFixedArgsLeft > 1)
595           break;
596       } else if (Tok.is(tok::comment) && !KeepMacroComments) {
597         // If this is a comment token in the argument list and we're just in
598         // -C mode (not -CC mode), discard the comment.
599         continue;
600       } else if (Tok.getIdentifierInfo() != nullptr) {
601         // Reading macro arguments can cause macros that we are currently
602         // expanding from to be popped off the expansion stack.  Doing so causes
603         // them to be reenabled for expansion.  Here we record whether any
604         // identifiers we lex as macro arguments correspond to disabled macros.
605         // If so, we mark the token as noexpand.  This is a subtle aspect of
606         // C99 6.10.3.4p2.
607         if (MacroInfo *MI = getMacroInfo(Tok.getIdentifierInfo()))
608           if (!MI->isEnabled())
609             Tok.setFlag(Token::DisableExpand);
610       } else if (Tok.is(tok::code_completion)) {
611         ContainsCodeCompletionTok = true;
612         if (CodeComplete)
613           CodeComplete->CodeCompleteMacroArgument(MacroName.getIdentifierInfo(),
614                                                   MI, NumActuals);
615         // Don't mark that we reached the code-completion point because the
616         // parser is going to handle the token and there will be another
617         // code-completion callback.
618       }
619 
620       ArgTokens.push_back(Tok);
621     }
622 
623     // If this was an empty argument list foo(), don't add this as an empty
624     // argument.
625     if (ArgTokens.empty() && Tok.getKind() == tok::r_paren)
626       break;
627 
628     // If this is not a variadic macro, and too many args were specified, emit
629     // an error.
630     if (!isVariadic && NumFixedArgsLeft == 0 && TooManyArgsLoc.isInvalid()) {
631       if (ArgTokens.size() != ArgTokenStart)
632         TooManyArgsLoc = ArgTokens[ArgTokenStart].getLocation();
633       else
634         TooManyArgsLoc = ArgStartLoc;
635     }
636 
637     // Empty arguments are standard in C99 and C++0x, and are supported as an
638     // extension in other modes.
639     if (ArgTokens.size() == ArgTokenStart && !LangOpts.C99)
640       Diag(Tok, LangOpts.CPlusPlus11 ?
641            diag::warn_cxx98_compat_empty_fnmacro_arg :
642            diag::ext_empty_fnmacro_arg);
643 
644     // Add a marker EOF token to the end of the token list for this argument.
645     Token EOFTok;
646     EOFTok.startToken();
647     EOFTok.setKind(tok::eof);
648     EOFTok.setLocation(Tok.getLocation());
649     EOFTok.setLength(0);
650     ArgTokens.push_back(EOFTok);
651     ++NumActuals;
652     if (!ContainsCodeCompletionTok && NumFixedArgsLeft != 0)
653       --NumFixedArgsLeft;
654   }
655 
656   // Okay, we either found the r_paren.  Check to see if we parsed too few
657   // arguments.
658   unsigned MinArgsExpected = MI->getNumArgs();
659 
660   // If this is not a variadic macro, and too many args were specified, emit
661   // an error.
662   if (!isVariadic && NumActuals > MinArgsExpected &&
663       !ContainsCodeCompletionTok) {
664     // Emit the diagnostic at the macro name in case there is a missing ).
665     // Emitting it at the , could be far away from the macro name.
666     Diag(TooManyArgsLoc, diag::err_too_many_args_in_macro_invoc);
667     Diag(MI->getDefinitionLoc(), diag::note_macro_here)
668       << MacroName.getIdentifierInfo();
669 
670     // Commas from braced initializer lists will be treated as argument
671     // separators inside macros.  Attempt to correct for this with parentheses.
672     // TODO: See if this can be generalized to angle brackets for templates
673     // inside macro arguments.
674 
675     SmallVector<Token, 4> FixedArgTokens;
676     unsigned FixedNumArgs = 0;
677     SmallVector<SourceRange, 4> ParenHints, InitLists;
678     if (!GenerateNewArgTokens(*this, ArgTokens, FixedArgTokens, FixedNumArgs,
679                               ParenHints, InitLists)) {
680       if (!InitLists.empty()) {
681         DiagnosticBuilder DB =
682             Diag(MacroName,
683                  diag::note_init_list_at_beginning_of_macro_argument);
684         for (const SourceRange &Range : InitLists)
685           DB << Range;
686       }
687       return nullptr;
688     }
689     if (FixedNumArgs != MinArgsExpected)
690       return nullptr;
691 
692     DiagnosticBuilder DB = Diag(MacroName, diag::note_suggest_parens_for_macro);
693     for (const SourceRange &ParenLocation : ParenHints) {
694       DB << FixItHint::CreateInsertion(ParenLocation.getBegin(), "(");
695       DB << FixItHint::CreateInsertion(ParenLocation.getEnd(), ")");
696     }
697     ArgTokens.swap(FixedArgTokens);
698     NumActuals = FixedNumArgs;
699   }
700 
701   // See MacroArgs instance var for description of this.
702   bool isVarargsElided = false;
703 
704   if (ContainsCodeCompletionTok) {
705     // Recover from not-fully-formed macro invocation during code-completion.
706     Token EOFTok;
707     EOFTok.startToken();
708     EOFTok.setKind(tok::eof);
709     EOFTok.setLocation(Tok.getLocation());
710     EOFTok.setLength(0);
711     for (; NumActuals < MinArgsExpected; ++NumActuals)
712       ArgTokens.push_back(EOFTok);
713   }
714 
715   if (NumActuals < MinArgsExpected) {
716     // There are several cases where too few arguments is ok, handle them now.
717     if (NumActuals == 0 && MinArgsExpected == 1) {
718       // #define A(X)  or  #define A(...)   ---> A()
719 
720       // If there is exactly one argument, and that argument is missing,
721       // then we have an empty "()" argument empty list.  This is fine, even if
722       // the macro expects one argument (the argument is just empty).
723       isVarargsElided = MI->isVariadic();
724     } else if (MI->isVariadic() &&
725                (NumActuals+1 == MinArgsExpected ||  // A(x, ...) -> A(X)
726                 (NumActuals == 0 && MinArgsExpected == 2))) {// A(x,...) -> A()
727       // Varargs where the named vararg parameter is missing: OK as extension.
728       //   #define A(x, ...)
729       //   A("blah")
730       //
731       // If the macro contains the comma pasting extension, the diagnostic
732       // is suppressed; we know we'll get another diagnostic later.
733       if (!MI->hasCommaPasting()) {
734         Diag(Tok, diag::ext_missing_varargs_arg);
735         Diag(MI->getDefinitionLoc(), diag::note_macro_here)
736           << MacroName.getIdentifierInfo();
737       }
738 
739       // Remember this occurred, allowing us to elide the comma when used for
740       // cases like:
741       //   #define A(x, foo...) blah(a, ## foo)
742       //   #define B(x, ...) blah(a, ## __VA_ARGS__)
743       //   #define C(...) blah(a, ## __VA_ARGS__)
744       //  A(x) B(x) C()
745       isVarargsElided = true;
746     } else if (!ContainsCodeCompletionTok) {
747       // Otherwise, emit the error.
748       Diag(Tok, diag::err_too_few_args_in_macro_invoc);
749       Diag(MI->getDefinitionLoc(), diag::note_macro_here)
750         << MacroName.getIdentifierInfo();
751       return nullptr;
752     }
753 
754     // Add a marker EOF token to the end of the token list for this argument.
755     SourceLocation EndLoc = Tok.getLocation();
756     Tok.startToken();
757     Tok.setKind(tok::eof);
758     Tok.setLocation(EndLoc);
759     Tok.setLength(0);
760     ArgTokens.push_back(Tok);
761 
762     // If we expect two arguments, add both as empty.
763     if (NumActuals == 0 && MinArgsExpected == 2)
764       ArgTokens.push_back(Tok);
765 
766   } else if (NumActuals > MinArgsExpected && !MI->isVariadic() &&
767              !ContainsCodeCompletionTok) {
768     // Emit the diagnostic at the macro name in case there is a missing ).
769     // Emitting it at the , could be far away from the macro name.
770     Diag(MacroName, diag::err_too_many_args_in_macro_invoc);
771     Diag(MI->getDefinitionLoc(), diag::note_macro_here)
772       << MacroName.getIdentifierInfo();
773     return nullptr;
774   }
775 
776   return MacroArgs::create(MI, ArgTokens, isVarargsElided, *this);
777 }
778 
779 /// \brief Keeps macro expanded tokens for TokenLexers.
780 //
781 /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
782 /// going to lex in the cache and when it finishes the tokens are removed
783 /// from the end of the cache.
784 Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer,
785                                               ArrayRef<Token> tokens) {
786   assert(tokLexer);
787   if (tokens.empty())
788     return nullptr;
789 
790   size_t newIndex = MacroExpandedTokens.size();
791   bool cacheNeedsToGrow = tokens.size() >
792                       MacroExpandedTokens.capacity()-MacroExpandedTokens.size();
793   MacroExpandedTokens.append(tokens.begin(), tokens.end());
794 
795   if (cacheNeedsToGrow) {
796     // Go through all the TokenLexers whose 'Tokens' pointer points in the
797     // buffer and update the pointers to the (potential) new buffer array.
798     for (unsigned i = 0, e = MacroExpandingLexersStack.size(); i != e; ++i) {
799       TokenLexer *prevLexer;
800       size_t tokIndex;
801       std::tie(prevLexer, tokIndex) = MacroExpandingLexersStack[i];
802       prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex;
803     }
804   }
805 
806   MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex));
807   return MacroExpandedTokens.data() + newIndex;
808 }
809 
810 void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() {
811   assert(!MacroExpandingLexersStack.empty());
812   size_t tokIndex = MacroExpandingLexersStack.back().second;
813   assert(tokIndex < MacroExpandedTokens.size());
814   // Pop the cached macro expanded tokens from the end.
815   MacroExpandedTokens.resize(tokIndex);
816   MacroExpandingLexersStack.pop_back();
817 }
818 
819 /// ComputeDATE_TIME - Compute the current time, enter it into the specified
820 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of
821 /// the identifier tokens inserted.
822 static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc,
823                              Preprocessor &PP) {
824   time_t TT = time(nullptr);
825   struct tm *TM = localtime(&TT);
826 
827   static const char * const Months[] = {
828     "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"
829   };
830 
831   {
832     SmallString<32> TmpBuffer;
833     llvm::raw_svector_ostream TmpStream(TmpBuffer);
834     TmpStream << llvm::format("\"%s %2d %4d\"", Months[TM->tm_mon],
835                               TM->tm_mday, TM->tm_year + 1900);
836     Token TmpTok;
837     TmpTok.startToken();
838     PP.CreateString(TmpStream.str(), TmpTok);
839     DATELoc = TmpTok.getLocation();
840   }
841 
842   {
843     SmallString<32> TmpBuffer;
844     llvm::raw_svector_ostream TmpStream(TmpBuffer);
845     TmpStream << llvm::format("\"%02d:%02d:%02d\"",
846                               TM->tm_hour, TM->tm_min, TM->tm_sec);
847     Token TmpTok;
848     TmpTok.startToken();
849     PP.CreateString(TmpStream.str(), TmpTok);
850     TIMELoc = TmpTok.getLocation();
851   }
852 }
853 
854 
855 /// HasFeature - Return true if we recognize and implement the feature
856 /// specified by the identifier as a standard language feature.
857 static bool HasFeature(const Preprocessor &PP, const IdentifierInfo *II) {
858   const LangOptions &LangOpts = PP.getLangOpts();
859   StringRef Feature = II->getName();
860 
861   // Normalize the feature name, __foo__ becomes foo.
862   if (Feature.startswith("__") && Feature.endswith("__") && Feature.size() >= 4)
863     Feature = Feature.substr(2, Feature.size() - 4);
864 
865   return llvm::StringSwitch<bool>(Feature)
866       .Case("address_sanitizer", LangOpts.Sanitize.has(SanitizerKind::Address))
867       .Case("attribute_analyzer_noreturn", true)
868       .Case("attribute_availability", true)
869       .Case("attribute_availability_with_message", true)
870       .Case("attribute_cf_returns_not_retained", true)
871       .Case("attribute_cf_returns_retained", true)
872       .Case("attribute_deprecated_with_message", true)
873       .Case("attribute_ext_vector_type", true)
874       .Case("attribute_ns_returns_not_retained", true)
875       .Case("attribute_ns_returns_retained", true)
876       .Case("attribute_ns_consumes_self", true)
877       .Case("attribute_ns_consumed", true)
878       .Case("attribute_cf_consumed", true)
879       .Case("attribute_objc_ivar_unused", true)
880       .Case("attribute_objc_method_family", true)
881       .Case("attribute_overloadable", true)
882       .Case("attribute_unavailable_with_message", true)
883       .Case("attribute_unused_on_fields", true)
884       .Case("blocks", LangOpts.Blocks)
885       .Case("c_thread_safety_attributes", true)
886       .Case("cxx_exceptions", LangOpts.CXXExceptions)
887       .Case("cxx_rtti", LangOpts.RTTI)
888       .Case("enumerator_attributes", true)
889       .Case("memory_sanitizer", LangOpts.Sanitize.has(SanitizerKind::Memory))
890       .Case("thread_sanitizer", LangOpts.Sanitize.has(SanitizerKind::Thread))
891       .Case("dataflow_sanitizer", LangOpts.Sanitize.has(SanitizerKind::DataFlow))
892       // Objective-C features
893       .Case("objc_arr", LangOpts.ObjCAutoRefCount) // FIXME: REMOVE?
894       .Case("objc_arc", LangOpts.ObjCAutoRefCount)
895       .Case("objc_arc_weak", LangOpts.ObjCARCWeak)
896       .Case("objc_default_synthesize_properties", LangOpts.ObjC2)
897       .Case("objc_fixed_enum", LangOpts.ObjC2)
898       .Case("objc_instancetype", LangOpts.ObjC2)
899       .Case("objc_modules", LangOpts.ObjC2 && LangOpts.Modules)
900       .Case("objc_nonfragile_abi", LangOpts.ObjCRuntime.isNonFragile())
901       .Case("objc_property_explicit_atomic",
902             true) // Does clang support explicit "atomic" keyword?
903       .Case("objc_protocol_qualifier_mangling", true)
904       .Case("objc_weak_class", LangOpts.ObjCRuntime.hasWeakClassImport())
905       .Case("ownership_holds", true)
906       .Case("ownership_returns", true)
907       .Case("ownership_takes", true)
908       .Case("objc_bool", true)
909       .Case("objc_subscripting", LangOpts.ObjCRuntime.isNonFragile())
910       .Case("objc_array_literals", LangOpts.ObjC2)
911       .Case("objc_dictionary_literals", LangOpts.ObjC2)
912       .Case("objc_boxed_expressions", LangOpts.ObjC2)
913       .Case("arc_cf_code_audited", true)
914       // C11 features
915       .Case("c_alignas", LangOpts.C11)
916       .Case("c_atomic", LangOpts.C11)
917       .Case("c_generic_selections", LangOpts.C11)
918       .Case("c_static_assert", LangOpts.C11)
919       .Case("c_thread_local",
920             LangOpts.C11 && PP.getTargetInfo().isTLSSupported())
921       // C++11 features
922       .Case("cxx_access_control_sfinae", LangOpts.CPlusPlus11)
923       .Case("cxx_alias_templates", LangOpts.CPlusPlus11)
924       .Case("cxx_alignas", LangOpts.CPlusPlus11)
925       .Case("cxx_atomic", LangOpts.CPlusPlus11)
926       .Case("cxx_attributes", LangOpts.CPlusPlus11)
927       .Case("cxx_auto_type", LangOpts.CPlusPlus11)
928       .Case("cxx_constexpr", LangOpts.CPlusPlus11)
929       .Case("cxx_decltype", LangOpts.CPlusPlus11)
930       .Case("cxx_decltype_incomplete_return_types", LangOpts.CPlusPlus11)
931       .Case("cxx_default_function_template_args", LangOpts.CPlusPlus11)
932       .Case("cxx_defaulted_functions", LangOpts.CPlusPlus11)
933       .Case("cxx_delegating_constructors", LangOpts.CPlusPlus11)
934       .Case("cxx_deleted_functions", LangOpts.CPlusPlus11)
935       .Case("cxx_explicit_conversions", LangOpts.CPlusPlus11)
936       .Case("cxx_generalized_initializers", LangOpts.CPlusPlus11)
937       .Case("cxx_implicit_moves", LangOpts.CPlusPlus11)
938       .Case("cxx_inheriting_constructors", LangOpts.CPlusPlus11)
939       .Case("cxx_inline_namespaces", LangOpts.CPlusPlus11)
940       .Case("cxx_lambdas", LangOpts.CPlusPlus11)
941       .Case("cxx_local_type_template_args", LangOpts.CPlusPlus11)
942       .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus11)
943       .Case("cxx_noexcept", LangOpts.CPlusPlus11)
944       .Case("cxx_nullptr", LangOpts.CPlusPlus11)
945       .Case("cxx_override_control", LangOpts.CPlusPlus11)
946       .Case("cxx_range_for", LangOpts.CPlusPlus11)
947       .Case("cxx_raw_string_literals", LangOpts.CPlusPlus11)
948       .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus11)
949       .Case("cxx_rvalue_references", LangOpts.CPlusPlus11)
950       .Case("cxx_strong_enums", LangOpts.CPlusPlus11)
951       .Case("cxx_static_assert", LangOpts.CPlusPlus11)
952       .Case("cxx_thread_local",
953             LangOpts.CPlusPlus11 && PP.getTargetInfo().isTLSSupported())
954       .Case("cxx_trailing_return", LangOpts.CPlusPlus11)
955       .Case("cxx_unicode_literals", LangOpts.CPlusPlus11)
956       .Case("cxx_unrestricted_unions", LangOpts.CPlusPlus11)
957       .Case("cxx_user_literals", LangOpts.CPlusPlus11)
958       .Case("cxx_variadic_templates", LangOpts.CPlusPlus11)
959       // C++1y features
960       .Case("cxx_aggregate_nsdmi", LangOpts.CPlusPlus14)
961       .Case("cxx_binary_literals", LangOpts.CPlusPlus14)
962       .Case("cxx_contextual_conversions", LangOpts.CPlusPlus14)
963       .Case("cxx_decltype_auto", LangOpts.CPlusPlus14)
964       .Case("cxx_generic_lambdas", LangOpts.CPlusPlus14)
965       .Case("cxx_init_captures", LangOpts.CPlusPlus14)
966       .Case("cxx_relaxed_constexpr", LangOpts.CPlusPlus14)
967       .Case("cxx_return_type_deduction", LangOpts.CPlusPlus14)
968       .Case("cxx_variable_templates", LangOpts.CPlusPlus14)
969       // C++ TSes
970       //.Case("cxx_runtime_arrays", LangOpts.CPlusPlusTSArrays)
971       //.Case("cxx_concepts", LangOpts.CPlusPlusTSConcepts)
972       // FIXME: Should this be __has_feature or __has_extension?
973       //.Case("raw_invocation_type", LangOpts.CPlusPlus)
974       // Type traits
975       .Case("has_nothrow_assign", LangOpts.CPlusPlus)
976       .Case("has_nothrow_copy", LangOpts.CPlusPlus)
977       .Case("has_nothrow_constructor", LangOpts.CPlusPlus)
978       .Case("has_trivial_assign", LangOpts.CPlusPlus)
979       .Case("has_trivial_copy", LangOpts.CPlusPlus)
980       .Case("has_trivial_constructor", LangOpts.CPlusPlus)
981       .Case("has_trivial_destructor", LangOpts.CPlusPlus)
982       .Case("has_virtual_destructor", LangOpts.CPlusPlus)
983       .Case("is_abstract", LangOpts.CPlusPlus)
984       .Case("is_base_of", LangOpts.CPlusPlus)
985       .Case("is_class", LangOpts.CPlusPlus)
986       .Case("is_constructible", LangOpts.CPlusPlus)
987       .Case("is_convertible_to", LangOpts.CPlusPlus)
988       .Case("is_empty", LangOpts.CPlusPlus)
989       .Case("is_enum", LangOpts.CPlusPlus)
990       .Case("is_final", LangOpts.CPlusPlus)
991       .Case("is_literal", LangOpts.CPlusPlus)
992       .Case("is_standard_layout", LangOpts.CPlusPlus)
993       .Case("is_pod", LangOpts.CPlusPlus)
994       .Case("is_polymorphic", LangOpts.CPlusPlus)
995       .Case("is_sealed", LangOpts.MicrosoftExt)
996       .Case("is_trivial", LangOpts.CPlusPlus)
997       .Case("is_trivially_assignable", LangOpts.CPlusPlus)
998       .Case("is_trivially_constructible", LangOpts.CPlusPlus)
999       .Case("is_trivially_copyable", LangOpts.CPlusPlus)
1000       .Case("is_union", LangOpts.CPlusPlus)
1001       .Case("modules", LangOpts.Modules)
1002       .Case("tls", PP.getTargetInfo().isTLSSupported())
1003       .Case("underlying_type", LangOpts.CPlusPlus)
1004       .Default(false);
1005 }
1006 
1007 /// HasExtension - Return true if we recognize and implement the feature
1008 /// specified by the identifier, either as an extension or a standard language
1009 /// feature.
1010 static bool HasExtension(const Preprocessor &PP, const IdentifierInfo *II) {
1011   if (HasFeature(PP, II))
1012     return true;
1013 
1014   // If the use of an extension results in an error diagnostic, extensions are
1015   // effectively unavailable, so just return false here.
1016   if (PP.getDiagnostics().getExtensionHandlingBehavior() >=
1017       diag::Severity::Error)
1018     return false;
1019 
1020   const LangOptions &LangOpts = PP.getLangOpts();
1021   StringRef Extension = II->getName();
1022 
1023   // Normalize the extension name, __foo__ becomes foo.
1024   if (Extension.startswith("__") && Extension.endswith("__") &&
1025       Extension.size() >= 4)
1026     Extension = Extension.substr(2, Extension.size() - 4);
1027 
1028   // Because we inherit the feature list from HasFeature, this string switch
1029   // must be less restrictive than HasFeature's.
1030   return llvm::StringSwitch<bool>(Extension)
1031            // C11 features supported by other languages as extensions.
1032            .Case("c_alignas", true)
1033            .Case("c_atomic", true)
1034            .Case("c_generic_selections", true)
1035            .Case("c_static_assert", true)
1036            .Case("c_thread_local", PP.getTargetInfo().isTLSSupported())
1037            // C++11 features supported by other languages as extensions.
1038            .Case("cxx_atomic", LangOpts.CPlusPlus)
1039            .Case("cxx_deleted_functions", LangOpts.CPlusPlus)
1040            .Case("cxx_explicit_conversions", LangOpts.CPlusPlus)
1041            .Case("cxx_inline_namespaces", LangOpts.CPlusPlus)
1042            .Case("cxx_local_type_template_args", LangOpts.CPlusPlus)
1043            .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus)
1044            .Case("cxx_override_control", LangOpts.CPlusPlus)
1045            .Case("cxx_range_for", LangOpts.CPlusPlus)
1046            .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus)
1047            .Case("cxx_rvalue_references", LangOpts.CPlusPlus)
1048            // C++1y features supported by other languages as extensions.
1049            .Case("cxx_binary_literals", true)
1050            .Case("cxx_init_captures", LangOpts.CPlusPlus11)
1051            .Case("cxx_variable_templates", LangOpts.CPlusPlus)
1052            .Default(false);
1053 }
1054 
1055 /// EvaluateHasIncludeCommon - Process a '__has_include("path")'
1056 /// or '__has_include_next("path")' expression.
1057 /// Returns true if successful.
1058 static bool EvaluateHasIncludeCommon(Token &Tok,
1059                                      IdentifierInfo *II, Preprocessor &PP,
1060                                      const DirectoryLookup *LookupFrom,
1061                                      const FileEntry *LookupFromFile) {
1062   // Save the location of the current token.  If a '(' is later found, use
1063   // that location.  If not, use the end of this location instead.
1064   SourceLocation LParenLoc = Tok.getLocation();
1065 
1066   // These expressions are only allowed within a preprocessor directive.
1067   if (!PP.isParsingIfOrElifDirective()) {
1068     PP.Diag(LParenLoc, diag::err_pp_directive_required) << II->getName();
1069     return false;
1070   }
1071 
1072   // Get '('.
1073   PP.LexNonComment(Tok);
1074 
1075   // Ensure we have a '('.
1076   if (Tok.isNot(tok::l_paren)) {
1077     // No '(', use end of last token.
1078     LParenLoc = PP.getLocForEndOfToken(LParenLoc);
1079     PP.Diag(LParenLoc, diag::err_pp_expected_after) << II << tok::l_paren;
1080     // If the next token looks like a filename or the start of one,
1081     // assume it is and process it as such.
1082     if (!Tok.is(tok::angle_string_literal) && !Tok.is(tok::string_literal) &&
1083         !Tok.is(tok::less))
1084       return false;
1085   } else {
1086     // Save '(' location for possible missing ')' message.
1087     LParenLoc = Tok.getLocation();
1088 
1089     if (PP.getCurrentLexer()) {
1090       // Get the file name.
1091       PP.getCurrentLexer()->LexIncludeFilename(Tok);
1092     } else {
1093       // We're in a macro, so we can't use LexIncludeFilename; just
1094       // grab the next token.
1095       PP.Lex(Tok);
1096     }
1097   }
1098 
1099   // Reserve a buffer to get the spelling.
1100   SmallString<128> FilenameBuffer;
1101   StringRef Filename;
1102   SourceLocation EndLoc;
1103 
1104   switch (Tok.getKind()) {
1105   case tok::eod:
1106     // If the token kind is EOD, the error has already been diagnosed.
1107     return false;
1108 
1109   case tok::angle_string_literal:
1110   case tok::string_literal: {
1111     bool Invalid = false;
1112     Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid);
1113     if (Invalid)
1114       return false;
1115     break;
1116   }
1117 
1118   case tok::less:
1119     // This could be a <foo/bar.h> file coming from a macro expansion.  In this
1120     // case, glue the tokens together into FilenameBuffer and interpret those.
1121     FilenameBuffer.push_back('<');
1122     if (PP.ConcatenateIncludeName(FilenameBuffer, EndLoc)) {
1123       // Let the caller know a <eod> was found by changing the Token kind.
1124       Tok.setKind(tok::eod);
1125       return false;   // Found <eod> but no ">"?  Diagnostic already emitted.
1126     }
1127     Filename = FilenameBuffer.str();
1128     break;
1129   default:
1130     PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename);
1131     return false;
1132   }
1133 
1134   SourceLocation FilenameLoc = Tok.getLocation();
1135 
1136   // Get ')'.
1137   PP.LexNonComment(Tok);
1138 
1139   // Ensure we have a trailing ).
1140   if (Tok.isNot(tok::r_paren)) {
1141     PP.Diag(PP.getLocForEndOfToken(FilenameLoc), diag::err_pp_expected_after)
1142         << II << tok::r_paren;
1143     PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1144     return false;
1145   }
1146 
1147   bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename);
1148   // If GetIncludeFilenameSpelling set the start ptr to null, there was an
1149   // error.
1150   if (Filename.empty())
1151     return false;
1152 
1153   // Search include directories.
1154   const DirectoryLookup *CurDir;
1155   const FileEntry *File =
1156       PP.LookupFile(FilenameLoc, Filename, isAngled, LookupFrom, LookupFromFile,
1157                     CurDir, nullptr, nullptr, nullptr);
1158 
1159   // Get the result value.  A result of true means the file exists.
1160   return File != nullptr;
1161 }
1162 
1163 /// EvaluateHasInclude - Process a '__has_include("path")' expression.
1164 /// Returns true if successful.
1165 static bool EvaluateHasInclude(Token &Tok, IdentifierInfo *II,
1166                                Preprocessor &PP) {
1167   return EvaluateHasIncludeCommon(Tok, II, PP, nullptr, nullptr);
1168 }
1169 
1170 /// EvaluateHasIncludeNext - Process '__has_include_next("path")' expression.
1171 /// Returns true if successful.
1172 static bool EvaluateHasIncludeNext(Token &Tok,
1173                                    IdentifierInfo *II, Preprocessor &PP) {
1174   // __has_include_next is like __has_include, except that we start
1175   // searching after the current found directory.  If we can't do this,
1176   // issue a diagnostic.
1177   // FIXME: Factor out duplication wiht
1178   // Preprocessor::HandleIncludeNextDirective.
1179   const DirectoryLookup *Lookup = PP.GetCurDirLookup();
1180   const FileEntry *LookupFromFile = nullptr;
1181   if (PP.isInPrimaryFile()) {
1182     Lookup = nullptr;
1183     PP.Diag(Tok, diag::pp_include_next_in_primary);
1184   } else if (PP.getCurrentSubmodule()) {
1185     // Start looking up in the directory *after* the one in which the current
1186     // file would be found, if any.
1187     assert(PP.getCurrentLexer() && "#include_next directive in macro?");
1188     LookupFromFile = PP.getCurrentLexer()->getFileEntry();
1189     Lookup = nullptr;
1190   } else if (!Lookup) {
1191     PP.Diag(Tok, diag::pp_include_next_absolute_path);
1192   } else {
1193     // Start looking up in the next directory.
1194     ++Lookup;
1195   }
1196 
1197   return EvaluateHasIncludeCommon(Tok, II, PP, Lookup, LookupFromFile);
1198 }
1199 
1200 /// \brief Process __building_module(identifier) expression.
1201 /// \returns true if we are building the named module, false otherwise.
1202 static bool EvaluateBuildingModule(Token &Tok,
1203                                    IdentifierInfo *II, Preprocessor &PP) {
1204   // Get '('.
1205   PP.LexNonComment(Tok);
1206 
1207   // Ensure we have a '('.
1208   if (Tok.isNot(tok::l_paren)) {
1209     PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II
1210                                                             << tok::l_paren;
1211     return false;
1212   }
1213 
1214   // Save '(' location for possible missing ')' message.
1215   SourceLocation LParenLoc = Tok.getLocation();
1216 
1217   // Get the module name.
1218   PP.LexNonComment(Tok);
1219 
1220   // Ensure that we have an identifier.
1221   if (Tok.isNot(tok::identifier)) {
1222     PP.Diag(Tok.getLocation(), diag::err_expected_id_building_module);
1223     return false;
1224   }
1225 
1226   bool Result
1227     = Tok.getIdentifierInfo()->getName() == PP.getLangOpts().CurrentModule;
1228 
1229   // Get ')'.
1230   PP.LexNonComment(Tok);
1231 
1232   // Ensure we have a trailing ).
1233   if (Tok.isNot(tok::r_paren)) {
1234     PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II
1235                                                             << tok::r_paren;
1236     PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1237     return false;
1238   }
1239 
1240   return Result;
1241 }
1242 
1243 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded
1244 /// as a builtin macro, handle it and return the next token as 'Tok'.
1245 void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
1246   // Figure out which token this is.
1247   IdentifierInfo *II = Tok.getIdentifierInfo();
1248   assert(II && "Can't be a macro without id info!");
1249 
1250   // If this is an _Pragma or Microsoft __pragma directive, expand it,
1251   // invoke the pragma handler, then lex the token after it.
1252   if (II == Ident_Pragma)
1253     return Handle_Pragma(Tok);
1254   else if (II == Ident__pragma) // in non-MS mode this is null
1255     return HandleMicrosoft__pragma(Tok);
1256 
1257   ++NumBuiltinMacroExpanded;
1258 
1259   SmallString<128> TmpBuffer;
1260   llvm::raw_svector_ostream OS(TmpBuffer);
1261 
1262   // Set up the return result.
1263   Tok.setIdentifierInfo(nullptr);
1264   Tok.clearFlag(Token::NeedsCleaning);
1265 
1266   if (II == Ident__LINE__) {
1267     // C99 6.10.8: "__LINE__: The presumed line number (within the current
1268     // source file) of the current source line (an integer constant)".  This can
1269     // be affected by #line.
1270     SourceLocation Loc = Tok.getLocation();
1271 
1272     // Advance to the location of the first _, this might not be the first byte
1273     // of the token if it starts with an escaped newline.
1274     Loc = AdvanceToTokenCharacter(Loc, 0);
1275 
1276     // One wrinkle here is that GCC expands __LINE__ to location of the *end* of
1277     // a macro expansion.  This doesn't matter for object-like macros, but
1278     // can matter for a function-like macro that expands to contain __LINE__.
1279     // Skip down through expansion points until we find a file loc for the
1280     // end of the expansion history.
1281     Loc = SourceMgr.getExpansionRange(Loc).second;
1282     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Loc);
1283 
1284     // __LINE__ expands to a simple numeric value.
1285     OS << (PLoc.isValid()? PLoc.getLine() : 1);
1286     Tok.setKind(tok::numeric_constant);
1287   } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__) {
1288     // C99 6.10.8: "__FILE__: The presumed name of the current source file (a
1289     // character string literal)". This can be affected by #line.
1290     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1291 
1292     // __BASE_FILE__ is a GNU extension that returns the top of the presumed
1293     // #include stack instead of the current file.
1294     if (II == Ident__BASE_FILE__ && PLoc.isValid()) {
1295       SourceLocation NextLoc = PLoc.getIncludeLoc();
1296       while (NextLoc.isValid()) {
1297         PLoc = SourceMgr.getPresumedLoc(NextLoc);
1298         if (PLoc.isInvalid())
1299           break;
1300 
1301         NextLoc = PLoc.getIncludeLoc();
1302       }
1303     }
1304 
1305     // Escape this filename.  Turn '\' -> '\\' '"' -> '\"'
1306     SmallString<128> FN;
1307     if (PLoc.isValid()) {
1308       FN += PLoc.getFilename();
1309       Lexer::Stringify(FN);
1310       OS << '"' << FN.str() << '"';
1311     }
1312     Tok.setKind(tok::string_literal);
1313   } else if (II == Ident__DATE__) {
1314     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1315     if (!DATELoc.isValid())
1316       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1317     Tok.setKind(tok::string_literal);
1318     Tok.setLength(strlen("\"Mmm dd yyyy\""));
1319     Tok.setLocation(SourceMgr.createExpansionLoc(DATELoc, Tok.getLocation(),
1320                                                  Tok.getLocation(),
1321                                                  Tok.getLength()));
1322     return;
1323   } else if (II == Ident__TIME__) {
1324     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1325     if (!TIMELoc.isValid())
1326       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1327     Tok.setKind(tok::string_literal);
1328     Tok.setLength(strlen("\"hh:mm:ss\""));
1329     Tok.setLocation(SourceMgr.createExpansionLoc(TIMELoc, Tok.getLocation(),
1330                                                  Tok.getLocation(),
1331                                                  Tok.getLength()));
1332     return;
1333   } else if (II == Ident__INCLUDE_LEVEL__) {
1334     // Compute the presumed include depth of this token.  This can be affected
1335     // by GNU line markers.
1336     unsigned Depth = 0;
1337 
1338     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1339     if (PLoc.isValid()) {
1340       PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1341       for (; PLoc.isValid(); ++Depth)
1342         PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1343     }
1344 
1345     // __INCLUDE_LEVEL__ expands to a simple numeric value.
1346     OS << Depth;
1347     Tok.setKind(tok::numeric_constant);
1348   } else if (II == Ident__TIMESTAMP__) {
1349     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1350     // MSVC, ICC, GCC, VisualAge C++ extension.  The generated string should be
1351     // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime.
1352 
1353     // Get the file that we are lexing out of.  If we're currently lexing from
1354     // a macro, dig into the include stack.
1355     const FileEntry *CurFile = nullptr;
1356     PreprocessorLexer *TheLexer = getCurrentFileLexer();
1357 
1358     if (TheLexer)
1359       CurFile = SourceMgr.getFileEntryForID(TheLexer->getFileID());
1360 
1361     const char *Result;
1362     if (CurFile) {
1363       time_t TT = CurFile->getModificationTime();
1364       struct tm *TM = localtime(&TT);
1365       Result = asctime(TM);
1366     } else {
1367       Result = "??? ??? ?? ??:??:?? ????\n";
1368     }
1369     // Surround the string with " and strip the trailing newline.
1370     OS << '"' << StringRef(Result).drop_back() << '"';
1371     Tok.setKind(tok::string_literal);
1372   } else if (II == Ident__COUNTER__) {
1373     // __COUNTER__ expands to a simple numeric value.
1374     OS << CounterValue++;
1375     Tok.setKind(tok::numeric_constant);
1376   } else if (II == Ident__has_feature   ||
1377              II == Ident__has_extension ||
1378              II == Ident__has_builtin   ||
1379              II == Ident__is_identifier ||
1380              II == Ident__has_attribute ||
1381              II == Ident__has_cpp_attribute) {
1382     // The argument to these builtins should be a parenthesized identifier.
1383     SourceLocation StartLoc = Tok.getLocation();
1384 
1385     bool IsValid = false;
1386     IdentifierInfo *FeatureII = nullptr;
1387     IdentifierInfo *ScopeII = nullptr;
1388 
1389     // Read the '('.
1390     LexUnexpandedToken(Tok);
1391     if (Tok.is(tok::l_paren)) {
1392       // Read the identifier
1393       LexUnexpandedToken(Tok);
1394       if ((FeatureII = Tok.getIdentifierInfo())) {
1395         // If we're checking __has_cpp_attribute, it is possible to receive a
1396         // scope token. Read the "::", if it's available.
1397         LexUnexpandedToken(Tok);
1398         bool IsScopeValid = true;
1399         if (II == Ident__has_cpp_attribute && Tok.is(tok::coloncolon)) {
1400           LexUnexpandedToken(Tok);
1401           // The first thing we read was not the feature, it was the scope.
1402           ScopeII = FeatureII;
1403           if ((FeatureII = Tok.getIdentifierInfo()))
1404             LexUnexpandedToken(Tok);
1405           else
1406             IsScopeValid = false;
1407         }
1408         // Read the closing paren.
1409         if (IsScopeValid && Tok.is(tok::r_paren))
1410           IsValid = true;
1411       }
1412     }
1413 
1414     int Value = 0;
1415     if (!IsValid)
1416       Diag(StartLoc, diag::err_feature_check_malformed);
1417     else if (II == Ident__is_identifier)
1418       Value = FeatureII->getTokenID() == tok::identifier;
1419     else if (II == Ident__has_builtin) {
1420       // Check for a builtin is trivial.
1421       Value = FeatureII->getBuiltinID() != 0;
1422     } else if (II == Ident__has_attribute)
1423       Value = hasAttribute(AttrSyntax::Generic, nullptr, FeatureII,
1424                            getTargetInfo().getTriple(), getLangOpts());
1425     else if (II == Ident__has_cpp_attribute)
1426       Value = hasAttribute(AttrSyntax::CXX, ScopeII, FeatureII,
1427                            getTargetInfo().getTriple(), getLangOpts());
1428     else if (II == Ident__has_extension)
1429       Value = HasExtension(*this, FeatureII);
1430     else {
1431       assert(II == Ident__has_feature && "Must be feature check");
1432       Value = HasFeature(*this, FeatureII);
1433     }
1434 
1435     OS << Value;
1436     if (IsValid)
1437       Tok.setKind(tok::numeric_constant);
1438   } else if (II == Ident__has_include ||
1439              II == Ident__has_include_next) {
1440     // The argument to these two builtins should be a parenthesized
1441     // file name string literal using angle brackets (<>) or
1442     // double-quotes ("").
1443     bool Value;
1444     if (II == Ident__has_include)
1445       Value = EvaluateHasInclude(Tok, II, *this);
1446     else
1447       Value = EvaluateHasIncludeNext(Tok, II, *this);
1448     OS << (int)Value;
1449     if (Tok.is(tok::r_paren))
1450       Tok.setKind(tok::numeric_constant);
1451   } else if (II == Ident__has_warning) {
1452     // The argument should be a parenthesized string literal.
1453     // The argument to these builtins should be a parenthesized identifier.
1454     SourceLocation StartLoc = Tok.getLocation();
1455     bool IsValid = false;
1456     bool Value = false;
1457     // Read the '('.
1458     LexUnexpandedToken(Tok);
1459     do {
1460       if (Tok.isNot(tok::l_paren)) {
1461         Diag(StartLoc, diag::err_warning_check_malformed);
1462         break;
1463       }
1464 
1465       LexUnexpandedToken(Tok);
1466       std::string WarningName;
1467       SourceLocation StrStartLoc = Tok.getLocation();
1468       if (!FinishLexStringLiteral(Tok, WarningName, "'__has_warning'",
1469                                   /*MacroExpansion=*/false)) {
1470         // Eat tokens until ')'.
1471         while (Tok.isNot(tok::r_paren) && Tok.isNot(tok::eod) &&
1472                Tok.isNot(tok::eof))
1473           LexUnexpandedToken(Tok);
1474         break;
1475       }
1476 
1477       // Is the end a ')'?
1478       if (!(IsValid = Tok.is(tok::r_paren))) {
1479         Diag(StartLoc, diag::err_warning_check_malformed);
1480         break;
1481       }
1482 
1483       // FIXME: Should we accept "-R..." flags here, or should that be handled
1484       // by a separate __has_remark?
1485       if (WarningName.size() < 3 || WarningName[0] != '-' ||
1486           WarningName[1] != 'W') {
1487         Diag(StrStartLoc, diag::warn_has_warning_invalid_option);
1488         break;
1489       }
1490 
1491       // Finally, check if the warning flags maps to a diagnostic group.
1492       // We construct a SmallVector here to talk to getDiagnosticIDs().
1493       // Although we don't use the result, this isn't a hot path, and not
1494       // worth special casing.
1495       SmallVector<diag::kind, 10> Diags;
1496       Value = !getDiagnostics().getDiagnosticIDs()->
1497         getDiagnosticsInGroup(diag::Flavor::WarningOrError,
1498                               WarningName.substr(2), Diags);
1499     } while (false);
1500 
1501     OS << (int)Value;
1502     if (IsValid)
1503       Tok.setKind(tok::numeric_constant);
1504   } else if (II == Ident__building_module) {
1505     // The argument to this builtin should be an identifier. The
1506     // builtin evaluates to 1 when that identifier names the module we are
1507     // currently building.
1508     OS << (int)EvaluateBuildingModule(Tok, II, *this);
1509     Tok.setKind(tok::numeric_constant);
1510   } else if (II == Ident__MODULE__) {
1511     // The current module as an identifier.
1512     OS << getLangOpts().CurrentModule;
1513     IdentifierInfo *ModuleII = getIdentifierInfo(getLangOpts().CurrentModule);
1514     Tok.setIdentifierInfo(ModuleII);
1515     Tok.setKind(ModuleII->getTokenID());
1516   } else if (II == Ident__identifier) {
1517     SourceLocation Loc = Tok.getLocation();
1518 
1519     // We're expecting '__identifier' '(' identifier ')'. Try to recover
1520     // if the parens are missing.
1521     LexNonComment(Tok);
1522     if (Tok.isNot(tok::l_paren)) {
1523       // No '(', use end of last token.
1524       Diag(getLocForEndOfToken(Loc), diag::err_pp_expected_after)
1525         << II << tok::l_paren;
1526       // If the next token isn't valid as our argument, we can't recover.
1527       if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1528         Tok.setKind(tok::identifier);
1529       return;
1530     }
1531 
1532     SourceLocation LParenLoc = Tok.getLocation();
1533     LexNonComment(Tok);
1534 
1535     if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1536       Tok.setKind(tok::identifier);
1537     else {
1538       Diag(Tok.getLocation(), diag::err_pp_identifier_arg_not_identifier)
1539         << Tok.getKind();
1540       // Don't walk past anything that's not a real token.
1541       if (Tok.is(tok::eof) || Tok.is(tok::eod) || Tok.isAnnotation())
1542         return;
1543     }
1544 
1545     // Discard the ')', preserving 'Tok' as our result.
1546     Token RParen;
1547     LexNonComment(RParen);
1548     if (RParen.isNot(tok::r_paren)) {
1549       Diag(getLocForEndOfToken(Tok.getLocation()), diag::err_pp_expected_after)
1550         << Tok.getKind() << tok::r_paren;
1551       Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1552     }
1553     return;
1554   } else {
1555     llvm_unreachable("Unknown identifier!");
1556   }
1557   CreateString(OS.str(), Tok, Tok.getLocation(), Tok.getLocation());
1558 }
1559 
1560 void Preprocessor::markMacroAsUsed(MacroInfo *MI) {
1561   // If the 'used' status changed, and the macro requires 'unused' warning,
1562   // remove its SourceLocation from the warn-for-unused-macro locations.
1563   if (MI->isWarnIfUnused() && !MI->isUsed())
1564     WarnUnusedMacroLocs.erase(MI->getDefinitionLoc());
1565   MI->setIsUsed(true);
1566 }
1567