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 expasion for the
11 // preprocessor.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "clang/Lex/Preprocessor.h"
16 #include "MacroArgs.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/MacroInfo.h"
24 #include "llvm/ADT/STLExtras.h"
25 #include "llvm/ADT/SmallString.h"
26 #include "llvm/ADT/StringSwitch.h"
27 #include "llvm/Config/llvm-config.h"
28 #include "llvm/Support/ErrorHandling.h"
29 #include "llvm/Support/Format.h"
30 #include "llvm/Support/raw_ostream.h"
31 #include <cstdio>
32 #include <ctime>
33 using namespace clang;
34 
35 MacroDirective *
36 Preprocessor::getMacroDirectiveHistory(const IdentifierInfo *II) const {
37   assert(II->hadMacroDefinition() && "Identifier has not been not a macro!");
38 
39   macro_iterator Pos = Macros.find(II);
40   assert(Pos != Macros.end() && "Identifier macro info is missing!");
41   return Pos->second;
42 }
43 
44 /// \brief Specify a macro for this identifier.
45 MacroDirective *
46 Preprocessor::setMacroDirective(IdentifierInfo *II, MacroInfo *MI,
47                                 SourceLocation Loc, bool isImported) {
48   assert(MI && "MacroInfo should be non-zero!");
49 
50   MacroDirective *MD = AllocateMacroDirective(MI, Loc, isImported);
51   MacroDirective *&StoredMD = Macros[II];
52   MD->setPrevious(StoredMD);
53   StoredMD = MD;
54   II->setHasMacroDefinition(true);
55   if (II->isFromAST())
56     II->setChangedSinceDeserialization();
57 
58   return MD;
59 }
60 
61 void Preprocessor::addLoadedMacroInfo(IdentifierInfo *II, MacroDirective *MD,
62                                       MacroDirective *Hint) {
63   assert(MD && "Missing macro?");
64   assert(MD->isImported() && "Macro is not from an AST?");
65   assert(!MD->getPrevious() && "Macro already in chain?");
66 
67   MacroDirective *&StoredMD = Macros[II];
68 
69   // Easy case: this is the first macro definition for this macro.
70   if (!StoredMD) {
71     StoredMD = MD;
72 
73     if (MD->isDefined())
74       II->setHasMacroDefinition(true);
75     return;
76   }
77 
78   // If this macro is a definition and this identifier has been neither
79   // defined nor undef'd in the current translation unit, add this macro
80   // to the end of the chain of definitions.
81   if (MD->isDefined() && StoredMD->isImported()) {
82     // Simple case: if this is the first actual definition, just put it at
83     // th beginning.
84     if (!StoredMD->isDefined()) {
85       MD->setPrevious(StoredMD);
86       StoredMD = MD;
87 
88       II->setHasMacroDefinition(true);
89       return;
90     }
91 
92     // Find the end of the definition chain.
93     MacroDirective *Prev;
94     MacroDirective *PrevPrev = StoredMD;
95     bool Ambiguous = StoredMD->isAmbiguous();
96     bool MatchedOther = false;
97     do {
98       Prev = PrevPrev;
99 
100       // If the macros are not identical, we have an ambiguity.
101       if (!Prev->getInfo()->isIdenticalTo(*MD->getInfo(), *this)) {
102         if (!Ambiguous) {
103           Ambiguous = true;
104           StoredMD->setAmbiguous(true);
105         }
106       } else {
107         MatchedOther = true;
108       }
109     } while ((PrevPrev = Prev->getPrevious()) &&
110              PrevPrev->isDefined());
111 
112     // If there are ambiguous definitions, and we didn't match any other
113     // definition, then mark us as ambiguous.
114     if (Ambiguous && !MatchedOther)
115       MD->setAmbiguous(true);
116 
117     // Wire this macro information into the chain.
118     MD->setPrevious(Prev->getPrevious());
119     Prev->setPrevious(MD);
120     return;
121   }
122 
123   // The macro is not a definition; put it at the end of the list.
124   MacroDirective *Prev = Hint? Hint : StoredMD;
125   while (Prev->getPrevious())
126     Prev = Prev->getPrevious();
127   Prev->setPrevious(MD);
128 }
129 
130 void Preprocessor::makeLoadedMacroInfoVisible(IdentifierInfo *II,
131                                               MacroDirective *MD) {
132   assert(MD->isImported() && "Macro must be from the AST");
133 
134   MacroDirective *&StoredMD = Macros[II];
135   if (StoredMD == MD) {
136     // Easy case: this is the first macro anyway.
137     II->setHasMacroDefinition(MD->isDefined());
138     return;
139   }
140 
141   // Go find the macro and pull it out of the list.
142   // FIXME: Yes, this is O(N), and making a pile of macros visible or hidden
143   // would be quadratic, but it's extremely rare.
144   MacroDirective *Prev = StoredMD;
145   while (Prev->getPrevious() != MD)
146     Prev = Prev->getPrevious();
147   Prev->setPrevious(MD->getPrevious());
148   MD->setPrevious(0);
149 
150   // Add the macro back to the list.
151   addLoadedMacroInfo(II, MD);
152 
153   II->setHasMacroDefinition(StoredMD->isDefined());
154   if (II->isFromAST())
155     II->setChangedSinceDeserialization();
156 }
157 
158 /// \brief Undefine a macro for this identifier.
159 void Preprocessor::clearMacroInfo(IdentifierInfo *II) {
160   assert(II->hasMacroDefinition() && "Macro is not defined!");
161   assert(Macros[II]->getUndefLoc().isValid() && "Macro is still defined!");
162   II->setHasMacroDefinition(false);
163   if (II->isFromAST())
164     II->setChangedSinceDeserialization();
165 }
166 
167 /// RegisterBuiltinMacro - Register the specified identifier in the identifier
168 /// table and mark it as a builtin macro to be expanded.
169 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){
170   // Get the identifier.
171   IdentifierInfo *Id = PP.getIdentifierInfo(Name);
172 
173   // Mark it as being a macro that is builtin.
174   MacroInfo *MI = PP.AllocateMacroInfo(SourceLocation());
175   MI->setIsBuiltinMacro();
176   PP.setMacroDirective(Id, MI);
177   return Id;
178 }
179 
180 
181 /// RegisterBuiltinMacros - Register builtin macros, such as __LINE__ with the
182 /// identifier table.
183 void Preprocessor::RegisterBuiltinMacros() {
184   Ident__LINE__ = RegisterBuiltinMacro(*this, "__LINE__");
185   Ident__FILE__ = RegisterBuiltinMacro(*this, "__FILE__");
186   Ident__DATE__ = RegisterBuiltinMacro(*this, "__DATE__");
187   Ident__TIME__ = RegisterBuiltinMacro(*this, "__TIME__");
188   Ident__COUNTER__ = RegisterBuiltinMacro(*this, "__COUNTER__");
189   Ident_Pragma  = RegisterBuiltinMacro(*this, "_Pragma");
190 
191   // GCC Extensions.
192   Ident__BASE_FILE__     = RegisterBuiltinMacro(*this, "__BASE_FILE__");
193   Ident__INCLUDE_LEVEL__ = RegisterBuiltinMacro(*this, "__INCLUDE_LEVEL__");
194   Ident__TIMESTAMP__     = RegisterBuiltinMacro(*this, "__TIMESTAMP__");
195 
196   // Clang Extensions.
197   Ident__has_feature      = RegisterBuiltinMacro(*this, "__has_feature");
198   Ident__has_extension    = RegisterBuiltinMacro(*this, "__has_extension");
199   Ident__has_builtin      = RegisterBuiltinMacro(*this, "__has_builtin");
200   Ident__has_attribute    = RegisterBuiltinMacro(*this, "__has_attribute");
201   Ident__has_include      = RegisterBuiltinMacro(*this, "__has_include");
202   Ident__has_include_next = RegisterBuiltinMacro(*this, "__has_include_next");
203   Ident__has_warning      = RegisterBuiltinMacro(*this, "__has_warning");
204 
205   // Modules.
206   if (LangOpts.Modules) {
207     Ident__building_module  = RegisterBuiltinMacro(*this, "__building_module");
208 
209     // __MODULE__
210     if (!LangOpts.CurrentModule.empty())
211       Ident__MODULE__ = RegisterBuiltinMacro(*this, "__MODULE__");
212     else
213       Ident__MODULE__ = 0;
214   } else {
215     Ident__building_module = 0;
216     Ident__MODULE__ = 0;
217   }
218 
219   // Microsoft Extensions.
220   if (LangOpts.MicrosoftExt)
221     Ident__pragma = RegisterBuiltinMacro(*this, "__pragma");
222   else
223     Ident__pragma = 0;
224 }
225 
226 /// isTrivialSingleTokenExpansion - Return true if MI, which has a single token
227 /// in its expansion, currently expands to that token literally.
228 static bool isTrivialSingleTokenExpansion(const MacroInfo *MI,
229                                           const IdentifierInfo *MacroIdent,
230                                           Preprocessor &PP) {
231   IdentifierInfo *II = MI->getReplacementToken(0).getIdentifierInfo();
232 
233   // If the token isn't an identifier, it's always literally expanded.
234   if (II == 0) return true;
235 
236   // If the information about this identifier is out of date, update it from
237   // the external source.
238   if (II->isOutOfDate())
239     PP.getExternalSource()->updateOutOfDateIdentifier(*II);
240 
241   // If the identifier is a macro, and if that macro is enabled, it may be
242   // expanded so it's not a trivial expansion.
243   if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled() &&
244       // Fast expanding "#define X X" is ok, because X would be disabled.
245       II != MacroIdent)
246     return false;
247 
248   // If this is an object-like macro invocation, it is safe to trivially expand
249   // it.
250   if (MI->isObjectLike()) return true;
251 
252   // If this is a function-like macro invocation, it's safe to trivially expand
253   // as long as the identifier is not a macro argument.
254   for (MacroInfo::arg_iterator I = MI->arg_begin(), E = MI->arg_end();
255        I != E; ++I)
256     if (*I == II)
257       return false;   // Identifier is a macro argument.
258 
259   return true;
260 }
261 
262 
263 /// isNextPPTokenLParen - Determine whether the next preprocessor token to be
264 /// lexed is a '('.  If so, consume the token and return true, if not, this
265 /// method should have no observable side-effect on the lexed tokens.
266 bool Preprocessor::isNextPPTokenLParen() {
267   // Do some quick tests for rejection cases.
268   unsigned Val;
269   if (CurLexer)
270     Val = CurLexer->isNextPPTokenLParen();
271   else if (CurPTHLexer)
272     Val = CurPTHLexer->isNextPPTokenLParen();
273   else
274     Val = CurTokenLexer->isNextTokenLParen();
275 
276   if (Val == 2) {
277     // We have run off the end.  If it's a source file we don't
278     // examine enclosing ones (C99 5.1.1.2p4).  Otherwise walk up the
279     // macro stack.
280     if (CurPPLexer)
281       return false;
282     for (unsigned i = IncludeMacroStack.size(); i != 0; --i) {
283       IncludeStackInfo &Entry = IncludeMacroStack[i-1];
284       if (Entry.TheLexer)
285         Val = Entry.TheLexer->isNextPPTokenLParen();
286       else if (Entry.ThePTHLexer)
287         Val = Entry.ThePTHLexer->isNextPPTokenLParen();
288       else
289         Val = Entry.TheTokenLexer->isNextTokenLParen();
290 
291       if (Val != 2)
292         break;
293 
294       // Ran off the end of a source file?
295       if (Entry.ThePPLexer)
296         return false;
297     }
298   }
299 
300   // Okay, if we know that the token is a '(', lex it and return.  Otherwise we
301   // have found something that isn't a '(' or we found the end of the
302   // translation unit.  In either case, return false.
303   return Val == 1;
304 }
305 
306 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be
307 /// expanded as a macro, handle it and return the next token as 'Identifier'.
308 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
309                                                  MacroDirective *MD) {
310   MacroInfo *MI = MD->getInfo();
311 
312   // If this is a macro expansion in the "#if !defined(x)" line for the file,
313   // then the macro could expand to different things in other contexts, we need
314   // to disable the optimization in this case.
315   if (CurPPLexer) CurPPLexer->MIOpt.ExpandedMacro();
316 
317   // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially.
318   if (MI->isBuiltinMacro()) {
319     if (Callbacks) Callbacks->MacroExpands(Identifier, MD,
320                                            Identifier.getLocation());
321     ExpandBuiltinMacro(Identifier);
322     return false;
323   }
324 
325   /// Args - If this is a function-like macro expansion, this contains,
326   /// for each macro argument, the list of tokens that were provided to the
327   /// invocation.
328   MacroArgs *Args = 0;
329 
330   // Remember where the end of the expansion occurred.  For an object-like
331   // macro, this is the identifier.  For a function-like macro, this is the ')'.
332   SourceLocation ExpansionEnd = Identifier.getLocation();
333 
334   // If this is a function-like macro, read the arguments.
335   if (MI->isFunctionLike()) {
336     // C99 6.10.3p10: If the preprocessing token immediately after the macro
337     // name isn't a '(', this macro should not be expanded.
338     if (!isNextPPTokenLParen())
339       return true;
340 
341     // Remember that we are now parsing the arguments to a macro invocation.
342     // Preprocessor directives used inside macro arguments are not portable, and
343     // this enables the warning.
344     InMacroArgs = true;
345     Args = ReadFunctionLikeMacroArgs(Identifier, MI, ExpansionEnd);
346 
347     // Finished parsing args.
348     InMacroArgs = false;
349 
350     // If there was an error parsing the arguments, bail out.
351     if (Args == 0) return false;
352 
353     ++NumFnMacroExpanded;
354   } else {
355     ++NumMacroExpanded;
356   }
357 
358   // Notice that this macro has been used.
359   markMacroAsUsed(MI);
360 
361   // Remember where the token is expanded.
362   SourceLocation ExpandLoc = Identifier.getLocation();
363   SourceRange ExpansionRange(ExpandLoc, ExpansionEnd);
364 
365   if (Callbacks) {
366     if (InMacroArgs) {
367       // We can have macro expansion inside a conditional directive while
368       // reading the function macro arguments. To ensure, in that case, that
369       // MacroExpands callbacks still happen in source order, queue this
370       // callback to have it happen after the function macro callback.
371       DelayedMacroExpandsCallbacks.push_back(
372                               MacroExpandsInfo(Identifier, MD, ExpansionRange));
373     } else {
374       Callbacks->MacroExpands(Identifier, MD, ExpansionRange);
375       if (!DelayedMacroExpandsCallbacks.empty()) {
376         for (unsigned i=0, e = DelayedMacroExpandsCallbacks.size(); i!=e; ++i) {
377           MacroExpandsInfo &Info = DelayedMacroExpandsCallbacks[i];
378           Callbacks->MacroExpands(Info.Tok, Info.MD, Info.Range);
379         }
380         DelayedMacroExpandsCallbacks.clear();
381       }
382     }
383   }
384 
385   // FIXME: Temporarily disable this warning that is currently bogus with a PCH
386   // that redefined a macro without undef'ing it first (test/PCH/macro-redef.c).
387 #if 0
388   // If the macro definition is ambiguous, complain.
389   if (MI->isAmbiguous()) {
390     Diag(Identifier, diag::warn_pp_ambiguous_macro)
391       << Identifier.getIdentifierInfo();
392     Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen)
393       << Identifier.getIdentifierInfo();
394     for (MacroInfo *PrevMI = MI->getPreviousDefinition();
395          PrevMI && PrevMI->isDefined();
396          PrevMI = PrevMI->getPreviousDefinition()) {
397       if (PrevMI->isAmbiguous()) {
398         Diag(PrevMI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_other)
399           << Identifier.getIdentifierInfo();
400       }
401     }
402   }
403 #endif
404 
405   // If we started lexing a macro, enter the macro expansion body.
406 
407   // If this macro expands to no tokens, don't bother to push it onto the
408   // expansion stack, only to take it right back off.
409   if (MI->getNumTokens() == 0) {
410     // No need for arg info.
411     if (Args) Args->destroy(*this);
412 
413     // Ignore this macro use, just return the next token in the current
414     // buffer.
415     bool HadLeadingSpace = Identifier.hasLeadingSpace();
416     bool IsAtStartOfLine = Identifier.isAtStartOfLine();
417 
418     Lex(Identifier);
419 
420     // If the identifier isn't on some OTHER line, inherit the leading
421     // whitespace/first-on-a-line property of this token.  This handles
422     // stuff like "! XX," -> "! ," and "   XX," -> "    ,", when XX is
423     // empty.
424     if (!Identifier.isAtStartOfLine()) {
425       if (IsAtStartOfLine) Identifier.setFlag(Token::StartOfLine);
426       if (HadLeadingSpace) Identifier.setFlag(Token::LeadingSpace);
427     }
428     Identifier.setFlag(Token::LeadingEmptyMacro);
429     ++NumFastMacroExpanded;
430     return false;
431 
432   } else if (MI->getNumTokens() == 1 &&
433              isTrivialSingleTokenExpansion(MI, Identifier.getIdentifierInfo(),
434                                            *this)) {
435     // Otherwise, if this macro expands into a single trivially-expanded
436     // token: expand it now.  This handles common cases like
437     // "#define VAL 42".
438 
439     // No need for arg info.
440     if (Args) Args->destroy(*this);
441 
442     // Propagate the isAtStartOfLine/hasLeadingSpace markers of the macro
443     // identifier to the expanded token.
444     bool isAtStartOfLine = Identifier.isAtStartOfLine();
445     bool hasLeadingSpace = Identifier.hasLeadingSpace();
446 
447     // Replace the result token.
448     Identifier = MI->getReplacementToken(0);
449 
450     // Restore the StartOfLine/LeadingSpace markers.
451     Identifier.setFlagValue(Token::StartOfLine , isAtStartOfLine);
452     Identifier.setFlagValue(Token::LeadingSpace, hasLeadingSpace);
453 
454     // Update the tokens location to include both its expansion and physical
455     // locations.
456     SourceLocation Loc =
457       SourceMgr.createExpansionLoc(Identifier.getLocation(), ExpandLoc,
458                                    ExpansionEnd,Identifier.getLength());
459     Identifier.setLocation(Loc);
460 
461     // If this is a disabled macro or #define X X, we must mark the result as
462     // unexpandable.
463     if (IdentifierInfo *NewII = Identifier.getIdentifierInfo()) {
464       if (MacroInfo *NewMI = getMacroInfo(NewII))
465         if (!NewMI->isEnabled() || NewMI == MI) {
466           Identifier.setFlag(Token::DisableExpand);
467           // Don't warn for "#define X X" like "#define bool bool" from
468           // stdbool.h.
469           if (NewMI != MI || MI->isFunctionLike())
470             Diag(Identifier, diag::pp_disabled_macro_expansion);
471         }
472     }
473 
474     // Since this is not an identifier token, it can't be macro expanded, so
475     // we're done.
476     ++NumFastMacroExpanded;
477     return false;
478   }
479 
480   // Start expanding the macro.
481   EnterMacro(Identifier, ExpansionEnd, MI, Args);
482 
483   // Now that the macro is at the top of the include stack, ask the
484   // preprocessor to read the next token from it.
485   Lex(Identifier);
486   return false;
487 }
488 
489 /// ReadFunctionLikeMacroArgs - After reading "MACRO" and knowing that the next
490 /// token is the '(' of the macro, this method is invoked to read all of the
491 /// actual arguments specified for the macro invocation.  This returns null on
492 /// error.
493 MacroArgs *Preprocessor::ReadFunctionLikeMacroArgs(Token &MacroName,
494                                                    MacroInfo *MI,
495                                                    SourceLocation &MacroEnd) {
496   // The number of fixed arguments to parse.
497   unsigned NumFixedArgsLeft = MI->getNumArgs();
498   bool isVariadic = MI->isVariadic();
499 
500   // Outer loop, while there are more arguments, keep reading them.
501   Token Tok;
502 
503   // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
504   // an argument value in a macro could expand to ',' or '(' or ')'.
505   LexUnexpandedToken(Tok);
506   assert(Tok.is(tok::l_paren) && "Error computing l-paren-ness?");
507 
508   // ArgTokens - Build up a list of tokens that make up each argument.  Each
509   // argument is separated by an EOF token.  Use a SmallVector so we can avoid
510   // heap allocations in the common case.
511   SmallVector<Token, 64> ArgTokens;
512   bool ContainsCodeCompletionTok = false;
513 
514   unsigned NumActuals = 0;
515   while (Tok.isNot(tok::r_paren)) {
516     if (ContainsCodeCompletionTok && (Tok.is(tok::eof) || Tok.is(tok::eod)))
517       break;
518 
519     assert((Tok.is(tok::l_paren) || Tok.is(tok::comma)) &&
520            "only expect argument separators here");
521 
522     unsigned ArgTokenStart = ArgTokens.size();
523     SourceLocation ArgStartLoc = Tok.getLocation();
524 
525     // C99 6.10.3p11: Keep track of the number of l_parens we have seen.  Note
526     // that we already consumed the first one.
527     unsigned NumParens = 0;
528 
529     while (1) {
530       // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
531       // an argument value in a macro could expand to ',' or '(' or ')'.
532       LexUnexpandedToken(Tok);
533 
534       if (Tok.is(tok::eof) || Tok.is(tok::eod)) { // "#if f(<eof>" & "#if f(\n"
535         if (!ContainsCodeCompletionTok) {
536           Diag(MacroName, diag::err_unterm_macro_invoc);
537           Diag(MI->getDefinitionLoc(), diag::note_macro_here)
538             << MacroName.getIdentifierInfo();
539           // Do not lose the EOF/EOD.  Return it to the client.
540           MacroName = Tok;
541           return 0;
542         } else {
543           // Do not lose the EOF/EOD.
544           Token *Toks = new Token[1];
545           Toks[0] = Tok;
546           EnterTokenStream(Toks, 1, true, true);
547           break;
548         }
549       } else if (Tok.is(tok::r_paren)) {
550         // If we found the ) token, the macro arg list is done.
551         if (NumParens-- == 0) {
552           MacroEnd = Tok.getLocation();
553           break;
554         }
555       } else if (Tok.is(tok::l_paren)) {
556         ++NumParens;
557       } else if (Tok.is(tok::comma) && NumParens == 0) {
558         // Comma ends this argument if there are more fixed arguments expected.
559         // However, if this is a variadic macro, and this is part of the
560         // variadic part, then the comma is just an argument token.
561         if (!isVariadic) break;
562         if (NumFixedArgsLeft > 1)
563           break;
564       } else if (Tok.is(tok::comment) && !KeepMacroComments) {
565         // If this is a comment token in the argument list and we're just in
566         // -C mode (not -CC mode), discard the comment.
567         continue;
568       } else if (Tok.getIdentifierInfo() != 0) {
569         // Reading macro arguments can cause macros that we are currently
570         // expanding from to be popped off the expansion stack.  Doing so causes
571         // them to be reenabled for expansion.  Here we record whether any
572         // identifiers we lex as macro arguments correspond to disabled macros.
573         // If so, we mark the token as noexpand.  This is a subtle aspect of
574         // C99 6.10.3.4p2.
575         if (MacroInfo *MI = getMacroInfo(Tok.getIdentifierInfo()))
576           if (!MI->isEnabled())
577             Tok.setFlag(Token::DisableExpand);
578       } else if (Tok.is(tok::code_completion)) {
579         ContainsCodeCompletionTok = true;
580         if (CodeComplete)
581           CodeComplete->CodeCompleteMacroArgument(MacroName.getIdentifierInfo(),
582                                                   MI, NumActuals);
583         // Don't mark that we reached the code-completion point because the
584         // parser is going to handle the token and there will be another
585         // code-completion callback.
586       }
587 
588       ArgTokens.push_back(Tok);
589     }
590 
591     // If this was an empty argument list foo(), don't add this as an empty
592     // argument.
593     if (ArgTokens.empty() && Tok.getKind() == tok::r_paren)
594       break;
595 
596     // If this is not a variadic macro, and too many args were specified, emit
597     // an error.
598     if (!isVariadic && NumFixedArgsLeft == 0) {
599       if (ArgTokens.size() != ArgTokenStart)
600         ArgStartLoc = ArgTokens[ArgTokenStart].getLocation();
601 
602       if (!ContainsCodeCompletionTok) {
603         // Emit the diagnostic at the macro name in case there is a missing ).
604         // Emitting it at the , could be far away from the macro name.
605         Diag(ArgStartLoc, diag::err_too_many_args_in_macro_invoc);
606         Diag(MI->getDefinitionLoc(), diag::note_macro_here)
607           << MacroName.getIdentifierInfo();
608         return 0;
609       }
610     }
611 
612     // Empty arguments are standard in C99 and C++0x, and are supported as an extension in
613     // other modes.
614     if (ArgTokens.size() == ArgTokenStart && !LangOpts.C99)
615       Diag(Tok, LangOpts.CPlusPlus11 ?
616            diag::warn_cxx98_compat_empty_fnmacro_arg :
617            diag::ext_empty_fnmacro_arg);
618 
619     // Add a marker EOF token to the end of the token list for this argument.
620     Token EOFTok;
621     EOFTok.startToken();
622     EOFTok.setKind(tok::eof);
623     EOFTok.setLocation(Tok.getLocation());
624     EOFTok.setLength(0);
625     ArgTokens.push_back(EOFTok);
626     ++NumActuals;
627     if (!ContainsCodeCompletionTok || NumFixedArgsLeft != 0) {
628       assert(NumFixedArgsLeft != 0 && "Too many arguments parsed");
629       --NumFixedArgsLeft;
630     }
631   }
632 
633   // Okay, we either found the r_paren.  Check to see if we parsed too few
634   // arguments.
635   unsigned MinArgsExpected = MI->getNumArgs();
636 
637   // See MacroArgs instance var for description of this.
638   bool isVarargsElided = false;
639 
640   if (ContainsCodeCompletionTok) {
641     // Recover from not-fully-formed macro invocation during code-completion.
642     Token EOFTok;
643     EOFTok.startToken();
644     EOFTok.setKind(tok::eof);
645     EOFTok.setLocation(Tok.getLocation());
646     EOFTok.setLength(0);
647     for (; NumActuals < MinArgsExpected; ++NumActuals)
648       ArgTokens.push_back(EOFTok);
649   }
650 
651   if (NumActuals < MinArgsExpected) {
652     // There are several cases where too few arguments is ok, handle them now.
653     if (NumActuals == 0 && MinArgsExpected == 1) {
654       // #define A(X)  or  #define A(...)   ---> A()
655 
656       // If there is exactly one argument, and that argument is missing,
657       // then we have an empty "()" argument empty list.  This is fine, even if
658       // the macro expects one argument (the argument is just empty).
659       isVarargsElided = MI->isVariadic();
660     } else if (MI->isVariadic() &&
661                (NumActuals+1 == MinArgsExpected ||  // A(x, ...) -> A(X)
662                 (NumActuals == 0 && MinArgsExpected == 2))) {// A(x,...) -> A()
663       // Varargs where the named vararg parameter is missing: OK as extension.
664       //   #define A(x, ...)
665       //   A("blah")
666       //
667       // If the macro contains the comma pasting extension, the diagnostic
668       // is suppressed; we know we'll get another diagnostic later.
669       if (!MI->hasCommaPasting()) {
670         Diag(Tok, diag::ext_missing_varargs_arg);
671         Diag(MI->getDefinitionLoc(), diag::note_macro_here)
672           << MacroName.getIdentifierInfo();
673       }
674 
675       // Remember this occurred, allowing us to elide the comma when used for
676       // cases like:
677       //   #define A(x, foo...) blah(a, ## foo)
678       //   #define B(x, ...) blah(a, ## __VA_ARGS__)
679       //   #define C(...) blah(a, ## __VA_ARGS__)
680       //  A(x) B(x) C()
681       isVarargsElided = true;
682     } else if (!ContainsCodeCompletionTok) {
683       // Otherwise, emit the error.
684       Diag(Tok, diag::err_too_few_args_in_macro_invoc);
685       Diag(MI->getDefinitionLoc(), diag::note_macro_here)
686         << MacroName.getIdentifierInfo();
687       return 0;
688     }
689 
690     // Add a marker EOF token to the end of the token list for this argument.
691     SourceLocation EndLoc = Tok.getLocation();
692     Tok.startToken();
693     Tok.setKind(tok::eof);
694     Tok.setLocation(EndLoc);
695     Tok.setLength(0);
696     ArgTokens.push_back(Tok);
697 
698     // If we expect two arguments, add both as empty.
699     if (NumActuals == 0 && MinArgsExpected == 2)
700       ArgTokens.push_back(Tok);
701 
702   } else if (NumActuals > MinArgsExpected && !MI->isVariadic() &&
703              !ContainsCodeCompletionTok) {
704     // Emit the diagnostic at the macro name in case there is a missing ).
705     // Emitting it at the , could be far away from the macro name.
706     Diag(MacroName, diag::err_too_many_args_in_macro_invoc);
707     Diag(MI->getDefinitionLoc(), diag::note_macro_here)
708       << MacroName.getIdentifierInfo();
709     return 0;
710   }
711 
712   return MacroArgs::create(MI, ArgTokens, isVarargsElided, *this);
713 }
714 
715 /// \brief Keeps macro expanded tokens for TokenLexers.
716 //
717 /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
718 /// going to lex in the cache and when it finishes the tokens are removed
719 /// from the end of the cache.
720 Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer,
721                                               ArrayRef<Token> tokens) {
722   assert(tokLexer);
723   if (tokens.empty())
724     return 0;
725 
726   size_t newIndex = MacroExpandedTokens.size();
727   bool cacheNeedsToGrow = tokens.size() >
728                       MacroExpandedTokens.capacity()-MacroExpandedTokens.size();
729   MacroExpandedTokens.append(tokens.begin(), tokens.end());
730 
731   if (cacheNeedsToGrow) {
732     // Go through all the TokenLexers whose 'Tokens' pointer points in the
733     // buffer and update the pointers to the (potential) new buffer array.
734     for (unsigned i = 0, e = MacroExpandingLexersStack.size(); i != e; ++i) {
735       TokenLexer *prevLexer;
736       size_t tokIndex;
737       llvm::tie(prevLexer, tokIndex) = MacroExpandingLexersStack[i];
738       prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex;
739     }
740   }
741 
742   MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex));
743   return MacroExpandedTokens.data() + newIndex;
744 }
745 
746 void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() {
747   assert(!MacroExpandingLexersStack.empty());
748   size_t tokIndex = MacroExpandingLexersStack.back().second;
749   assert(tokIndex < MacroExpandedTokens.size());
750   // Pop the cached macro expanded tokens from the end.
751   MacroExpandedTokens.resize(tokIndex);
752   MacroExpandingLexersStack.pop_back();
753 }
754 
755 /// ComputeDATE_TIME - Compute the current time, enter it into the specified
756 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of
757 /// the identifier tokens inserted.
758 static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc,
759                              Preprocessor &PP) {
760   time_t TT = time(0);
761   struct tm *TM = localtime(&TT);
762 
763   static const char * const Months[] = {
764     "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"
765   };
766 
767   {
768     SmallString<32> TmpBuffer;
769     llvm::raw_svector_ostream TmpStream(TmpBuffer);
770     TmpStream << llvm::format("\"%s %2d %4d\"", Months[TM->tm_mon],
771                               TM->tm_mday, TM->tm_year + 1900);
772     Token TmpTok;
773     TmpTok.startToken();
774     PP.CreateString(TmpStream.str(), TmpTok);
775     DATELoc = TmpTok.getLocation();
776   }
777 
778   {
779     SmallString<32> TmpBuffer;
780     llvm::raw_svector_ostream TmpStream(TmpBuffer);
781     TmpStream << llvm::format("\"%02d:%02d:%02d\"",
782                               TM->tm_hour, TM->tm_min, TM->tm_sec);
783     Token TmpTok;
784     TmpTok.startToken();
785     PP.CreateString(TmpStream.str(), TmpTok);
786     TIMELoc = TmpTok.getLocation();
787   }
788 }
789 
790 
791 /// HasFeature - Return true if we recognize and implement the feature
792 /// specified by the identifier as a standard language feature.
793 static bool HasFeature(const Preprocessor &PP, const IdentifierInfo *II) {
794   const LangOptions &LangOpts = PP.getLangOpts();
795   StringRef Feature = II->getName();
796 
797   // Normalize the feature name, __foo__ becomes foo.
798   if (Feature.startswith("__") && Feature.endswith("__") && Feature.size() >= 4)
799     Feature = Feature.substr(2, Feature.size() - 4);
800 
801   return llvm::StringSwitch<bool>(Feature)
802            .Case("address_sanitizer", LangOpts.Sanitize.Address)
803            .Case("attribute_analyzer_noreturn", true)
804            .Case("attribute_availability", true)
805            .Case("attribute_availability_with_message", true)
806            .Case("attribute_cf_returns_not_retained", true)
807            .Case("attribute_cf_returns_retained", true)
808            .Case("attribute_deprecated_with_message", true)
809            .Case("attribute_ext_vector_type", true)
810            .Case("attribute_ns_returns_not_retained", true)
811            .Case("attribute_ns_returns_retained", true)
812            .Case("attribute_ns_consumes_self", true)
813            .Case("attribute_ns_consumed", true)
814            .Case("attribute_cf_consumed", true)
815            .Case("attribute_objc_ivar_unused", true)
816            .Case("attribute_objc_method_family", true)
817            .Case("attribute_overloadable", true)
818            .Case("attribute_unavailable_with_message", true)
819            .Case("attribute_unused_on_fields", true)
820            .Case("blocks", LangOpts.Blocks)
821            .Case("cxx_exceptions", LangOpts.Exceptions)
822            .Case("cxx_rtti", LangOpts.RTTI)
823            .Case("enumerator_attributes", true)
824            .Case("memory_sanitizer", LangOpts.Sanitize.Memory)
825            .Case("thread_sanitizer", LangOpts.Sanitize.Thread)
826            // Objective-C features
827            .Case("objc_arr", LangOpts.ObjCAutoRefCount) // FIXME: REMOVE?
828            .Case("objc_arc", LangOpts.ObjCAutoRefCount)
829            .Case("objc_arc_weak", LangOpts.ObjCARCWeak)
830            .Case("objc_default_synthesize_properties", LangOpts.ObjC2)
831            .Case("objc_fixed_enum", LangOpts.ObjC2)
832            .Case("objc_instancetype", LangOpts.ObjC2)
833            .Case("objc_modules", LangOpts.ObjC2 && LangOpts.Modules)
834            .Case("objc_nonfragile_abi", LangOpts.ObjCRuntime.isNonFragile())
835            .Case("objc_property_explicit_atomic", true) // Does clang support explicit "atomic" keyword?
836            .Case("objc_weak_class", LangOpts.ObjCRuntime.hasWeakClassImport())
837            .Case("ownership_holds", true)
838            .Case("ownership_returns", true)
839            .Case("ownership_takes", true)
840            .Case("objc_bool", true)
841            .Case("objc_subscripting", LangOpts.ObjCRuntime.isNonFragile())
842            .Case("objc_array_literals", LangOpts.ObjC2)
843            .Case("objc_dictionary_literals", LangOpts.ObjC2)
844            .Case("objc_boxed_expressions", LangOpts.ObjC2)
845            .Case("arc_cf_code_audited", true)
846            // C11 features
847            .Case("c_alignas", LangOpts.C11)
848            .Case("c_atomic", LangOpts.C11)
849            .Case("c_generic_selections", LangOpts.C11)
850            .Case("c_static_assert", LangOpts.C11)
851            // C++11 features
852            .Case("cxx_access_control_sfinae", LangOpts.CPlusPlus11)
853            .Case("cxx_alias_templates", LangOpts.CPlusPlus11)
854            .Case("cxx_alignas", LangOpts.CPlusPlus11)
855            .Case("cxx_atomic", LangOpts.CPlusPlus11)
856            .Case("cxx_attributes", LangOpts.CPlusPlus11)
857            .Case("cxx_auto_type", LangOpts.CPlusPlus11)
858            .Case("cxx_constexpr", LangOpts.CPlusPlus11)
859            .Case("cxx_decltype", LangOpts.CPlusPlus11)
860            .Case("cxx_decltype_incomplete_return_types", LangOpts.CPlusPlus11)
861            .Case("cxx_default_function_template_args", LangOpts.CPlusPlus11)
862            .Case("cxx_defaulted_functions", LangOpts.CPlusPlus11)
863            .Case("cxx_delegating_constructors", LangOpts.CPlusPlus11)
864            .Case("cxx_deleted_functions", LangOpts.CPlusPlus11)
865            .Case("cxx_explicit_conversions", LangOpts.CPlusPlus11)
866            .Case("cxx_generalized_initializers", LangOpts.CPlusPlus11)
867            .Case("cxx_implicit_moves", LangOpts.CPlusPlus11)
868          //.Case("cxx_inheriting_constructors", false)
869            .Case("cxx_inline_namespaces", LangOpts.CPlusPlus11)
870            .Case("cxx_lambdas", LangOpts.CPlusPlus11)
871            .Case("cxx_local_type_template_args", LangOpts.CPlusPlus11)
872            .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus11)
873            .Case("cxx_noexcept", LangOpts.CPlusPlus11)
874            .Case("cxx_nullptr", LangOpts.CPlusPlus11)
875            .Case("cxx_override_control", LangOpts.CPlusPlus11)
876            .Case("cxx_range_for", LangOpts.CPlusPlus11)
877            .Case("cxx_raw_string_literals", LangOpts.CPlusPlus11)
878            .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus11)
879            .Case("cxx_rvalue_references", LangOpts.CPlusPlus11)
880            .Case("cxx_strong_enums", LangOpts.CPlusPlus11)
881            .Case("cxx_static_assert", LangOpts.CPlusPlus11)
882            .Case("cxx_trailing_return", LangOpts.CPlusPlus11)
883            .Case("cxx_unicode_literals", LangOpts.CPlusPlus11)
884            .Case("cxx_unrestricted_unions", LangOpts.CPlusPlus11)
885            .Case("cxx_user_literals", LangOpts.CPlusPlus11)
886            .Case("cxx_variadic_templates", LangOpts.CPlusPlus11)
887            // Type traits
888            .Case("has_nothrow_assign", LangOpts.CPlusPlus)
889            .Case("has_nothrow_copy", LangOpts.CPlusPlus)
890            .Case("has_nothrow_constructor", LangOpts.CPlusPlus)
891            .Case("has_trivial_assign", LangOpts.CPlusPlus)
892            .Case("has_trivial_copy", LangOpts.CPlusPlus)
893            .Case("has_trivial_constructor", LangOpts.CPlusPlus)
894            .Case("has_trivial_destructor", LangOpts.CPlusPlus)
895            .Case("has_virtual_destructor", LangOpts.CPlusPlus)
896            .Case("is_abstract", LangOpts.CPlusPlus)
897            .Case("is_base_of", LangOpts.CPlusPlus)
898            .Case("is_class", LangOpts.CPlusPlus)
899            .Case("is_convertible_to", LangOpts.CPlusPlus)
900            .Case("is_empty", LangOpts.CPlusPlus)
901            .Case("is_enum", LangOpts.CPlusPlus)
902            .Case("is_final", LangOpts.CPlusPlus)
903            .Case("is_literal", LangOpts.CPlusPlus)
904            .Case("is_standard_layout", LangOpts.CPlusPlus)
905            .Case("is_pod", LangOpts.CPlusPlus)
906            .Case("is_polymorphic", LangOpts.CPlusPlus)
907            .Case("is_trivial", LangOpts.CPlusPlus)
908            .Case("is_trivially_assignable", LangOpts.CPlusPlus)
909            .Case("is_trivially_constructible", LangOpts.CPlusPlus)
910            .Case("is_trivially_copyable", LangOpts.CPlusPlus)
911            .Case("is_union", LangOpts.CPlusPlus)
912            .Case("modules", LangOpts.Modules)
913            .Case("tls", PP.getTargetInfo().isTLSSupported())
914            .Case("underlying_type", LangOpts.CPlusPlus)
915            .Default(false);
916 }
917 
918 /// HasExtension - Return true if we recognize and implement the feature
919 /// specified by the identifier, either as an extension or a standard language
920 /// feature.
921 static bool HasExtension(const Preprocessor &PP, const IdentifierInfo *II) {
922   if (HasFeature(PP, II))
923     return true;
924 
925   // If the use of an extension results in an error diagnostic, extensions are
926   // effectively unavailable, so just return false here.
927   if (PP.getDiagnostics().getExtensionHandlingBehavior() ==
928       DiagnosticsEngine::Ext_Error)
929     return false;
930 
931   const LangOptions &LangOpts = PP.getLangOpts();
932   StringRef Extension = II->getName();
933 
934   // Normalize the extension name, __foo__ becomes foo.
935   if (Extension.startswith("__") && Extension.endswith("__") &&
936       Extension.size() >= 4)
937     Extension = Extension.substr(2, Extension.size() - 4);
938 
939   // Because we inherit the feature list from HasFeature, this string switch
940   // must be less restrictive than HasFeature's.
941   return llvm::StringSwitch<bool>(Extension)
942            // C11 features supported by other languages as extensions.
943            .Case("c_alignas", true)
944            .Case("c_atomic", true)
945            .Case("c_generic_selections", true)
946            .Case("c_static_assert", true)
947            // C++0x features supported by other languages as extensions.
948            .Case("cxx_atomic", LangOpts.CPlusPlus)
949            .Case("cxx_deleted_functions", LangOpts.CPlusPlus)
950            .Case("cxx_explicit_conversions", LangOpts.CPlusPlus)
951            .Case("cxx_inline_namespaces", LangOpts.CPlusPlus)
952            .Case("cxx_local_type_template_args", LangOpts.CPlusPlus)
953            .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus)
954            .Case("cxx_override_control", LangOpts.CPlusPlus)
955            .Case("cxx_range_for", LangOpts.CPlusPlus)
956            .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus)
957            .Case("cxx_rvalue_references", LangOpts.CPlusPlus)
958            .Default(false);
959 }
960 
961 /// HasAttribute -  Return true if we recognize and implement the attribute
962 /// specified by the given identifier.
963 static bool HasAttribute(const IdentifierInfo *II) {
964   StringRef Name = II->getName();
965   // Normalize the attribute name, __foo__ becomes foo.
966   if (Name.startswith("__") && Name.endswith("__") && Name.size() >= 4)
967     Name = Name.substr(2, Name.size() - 4);
968 
969   // FIXME: Do we need to handle namespaces here?
970   return llvm::StringSwitch<bool>(Name)
971 #include "clang/Lex/AttrSpellings.inc"
972         .Default(false);
973 }
974 
975 /// EvaluateHasIncludeCommon - Process a '__has_include("path")'
976 /// or '__has_include_next("path")' expression.
977 /// Returns true if successful.
978 static bool EvaluateHasIncludeCommon(Token &Tok,
979                                      IdentifierInfo *II, Preprocessor &PP,
980                                      const DirectoryLookup *LookupFrom) {
981   // Save the location of the current token.  If a '(' is later found, use
982   // that location.  If not, use the end of this location instead.
983   SourceLocation LParenLoc = Tok.getLocation();
984 
985   // These expressions are only allowed within a preprocessor directive.
986   if (!PP.isParsingIfOrElifDirective()) {
987     PP.Diag(LParenLoc, diag::err_pp_directive_required) << II->getName();
988     return false;
989   }
990 
991   // Get '('.
992   PP.LexNonComment(Tok);
993 
994   // Ensure we have a '('.
995   if (Tok.isNot(tok::l_paren)) {
996     // No '(', use end of last token.
997     LParenLoc = PP.getLocForEndOfToken(LParenLoc);
998     PP.Diag(LParenLoc, diag::err_pp_missing_lparen) << II->getName();
999     // If the next token looks like a filename or the start of one,
1000     // assume it is and process it as such.
1001     if (!Tok.is(tok::angle_string_literal) && !Tok.is(tok::string_literal) &&
1002         !Tok.is(tok::less))
1003       return false;
1004   } else {
1005     // Save '(' location for possible missing ')' message.
1006     LParenLoc = Tok.getLocation();
1007 
1008     if (PP.getCurrentLexer()) {
1009       // Get the file name.
1010       PP.getCurrentLexer()->LexIncludeFilename(Tok);
1011     } else {
1012       // We're in a macro, so we can't use LexIncludeFilename; just
1013       // grab the next token.
1014       PP.Lex(Tok);
1015     }
1016   }
1017 
1018   // Reserve a buffer to get the spelling.
1019   SmallString<128> FilenameBuffer;
1020   StringRef Filename;
1021   SourceLocation EndLoc;
1022 
1023   switch (Tok.getKind()) {
1024   case tok::eod:
1025     // If the token kind is EOD, the error has already been diagnosed.
1026     return false;
1027 
1028   case tok::angle_string_literal:
1029   case tok::string_literal: {
1030     bool Invalid = false;
1031     Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid);
1032     if (Invalid)
1033       return false;
1034     break;
1035   }
1036 
1037   case tok::less:
1038     // This could be a <foo/bar.h> file coming from a macro expansion.  In this
1039     // case, glue the tokens together into FilenameBuffer and interpret those.
1040     FilenameBuffer.push_back('<');
1041     if (PP.ConcatenateIncludeName(FilenameBuffer, EndLoc)) {
1042       // Let the caller know a <eod> was found by changing the Token kind.
1043       Tok.setKind(tok::eod);
1044       return false;   // Found <eod> but no ">"?  Diagnostic already emitted.
1045     }
1046     Filename = FilenameBuffer.str();
1047     break;
1048   default:
1049     PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename);
1050     return false;
1051   }
1052 
1053   SourceLocation FilenameLoc = Tok.getLocation();
1054 
1055   // Get ')'.
1056   PP.LexNonComment(Tok);
1057 
1058   // Ensure we have a trailing ).
1059   if (Tok.isNot(tok::r_paren)) {
1060     PP.Diag(PP.getLocForEndOfToken(FilenameLoc), diag::err_pp_missing_rparen)
1061         << II->getName();
1062     PP.Diag(LParenLoc, diag::note_matching) << "(";
1063     return false;
1064   }
1065 
1066   bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename);
1067   // If GetIncludeFilenameSpelling set the start ptr to null, there was an
1068   // error.
1069   if (Filename.empty())
1070     return false;
1071 
1072   // Search include directories.
1073   const DirectoryLookup *CurDir;
1074   const FileEntry *File =
1075       PP.LookupFile(Filename, isAngled, LookupFrom, CurDir, NULL, NULL, NULL);
1076 
1077   // Get the result value.  A result of true means the file exists.
1078   return File != 0;
1079 }
1080 
1081 /// EvaluateHasInclude - Process a '__has_include("path")' expression.
1082 /// Returns true if successful.
1083 static bool EvaluateHasInclude(Token &Tok, IdentifierInfo *II,
1084                                Preprocessor &PP) {
1085   return EvaluateHasIncludeCommon(Tok, II, PP, NULL);
1086 }
1087 
1088 /// EvaluateHasIncludeNext - Process '__has_include_next("path")' expression.
1089 /// Returns true if successful.
1090 static bool EvaluateHasIncludeNext(Token &Tok,
1091                                    IdentifierInfo *II, Preprocessor &PP) {
1092   // __has_include_next is like __has_include, except that we start
1093   // searching after the current found directory.  If we can't do this,
1094   // issue a diagnostic.
1095   const DirectoryLookup *Lookup = PP.GetCurDirLookup();
1096   if (PP.isInPrimaryFile()) {
1097     Lookup = 0;
1098     PP.Diag(Tok, diag::pp_include_next_in_primary);
1099   } else if (Lookup == 0) {
1100     PP.Diag(Tok, diag::pp_include_next_absolute_path);
1101   } else {
1102     // Start looking up in the next directory.
1103     ++Lookup;
1104   }
1105 
1106   return EvaluateHasIncludeCommon(Tok, II, PP, Lookup);
1107 }
1108 
1109 /// \brief Process __building_module(identifier) expression.
1110 /// \returns true if we are building the named module, false otherwise.
1111 static bool EvaluateBuildingModule(Token &Tok,
1112                                    IdentifierInfo *II, Preprocessor &PP) {
1113   // Get '('.
1114   PP.LexNonComment(Tok);
1115 
1116   // Ensure we have a '('.
1117   if (Tok.isNot(tok::l_paren)) {
1118     PP.Diag(Tok.getLocation(), diag::err_pp_missing_lparen) << II->getName();
1119     return false;
1120   }
1121 
1122   // Save '(' location for possible missing ')' message.
1123   SourceLocation LParenLoc = Tok.getLocation();
1124 
1125   // Get the module name.
1126   PP.LexNonComment(Tok);
1127 
1128   // Ensure that we have an identifier.
1129   if (Tok.isNot(tok::identifier)) {
1130     PP.Diag(Tok.getLocation(), diag::err_expected_id_building_module);
1131     return false;
1132   }
1133 
1134   bool Result
1135     = Tok.getIdentifierInfo()->getName() == PP.getLangOpts().CurrentModule;
1136 
1137   // Get ')'.
1138   PP.LexNonComment(Tok);
1139 
1140   // Ensure we have a trailing ).
1141   if (Tok.isNot(tok::r_paren)) {
1142     PP.Diag(Tok.getLocation(), diag::err_pp_missing_rparen) << II->getName();
1143     PP.Diag(LParenLoc, diag::note_matching) << "(";
1144     return false;
1145   }
1146 
1147   return Result;
1148 }
1149 
1150 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded
1151 /// as a builtin macro, handle it and return the next token as 'Tok'.
1152 void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
1153   // Figure out which token this is.
1154   IdentifierInfo *II = Tok.getIdentifierInfo();
1155   assert(II && "Can't be a macro without id info!");
1156 
1157   // If this is an _Pragma or Microsoft __pragma directive, expand it,
1158   // invoke the pragma handler, then lex the token after it.
1159   if (II == Ident_Pragma)
1160     return Handle_Pragma(Tok);
1161   else if (II == Ident__pragma) // in non-MS mode this is null
1162     return HandleMicrosoft__pragma(Tok);
1163 
1164   ++NumBuiltinMacroExpanded;
1165 
1166   SmallString<128> TmpBuffer;
1167   llvm::raw_svector_ostream OS(TmpBuffer);
1168 
1169   // Set up the return result.
1170   Tok.setIdentifierInfo(0);
1171   Tok.clearFlag(Token::NeedsCleaning);
1172 
1173   if (II == Ident__LINE__) {
1174     // C99 6.10.8: "__LINE__: The presumed line number (within the current
1175     // source file) of the current source line (an integer constant)".  This can
1176     // be affected by #line.
1177     SourceLocation Loc = Tok.getLocation();
1178 
1179     // Advance to the location of the first _, this might not be the first byte
1180     // of the token if it starts with an escaped newline.
1181     Loc = AdvanceToTokenCharacter(Loc, 0);
1182 
1183     // One wrinkle here is that GCC expands __LINE__ to location of the *end* of
1184     // a macro expansion.  This doesn't matter for object-like macros, but
1185     // can matter for a function-like macro that expands to contain __LINE__.
1186     // Skip down through expansion points until we find a file loc for the
1187     // end of the expansion history.
1188     Loc = SourceMgr.getExpansionRange(Loc).second;
1189     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Loc);
1190 
1191     // __LINE__ expands to a simple numeric value.
1192     OS << (PLoc.isValid()? PLoc.getLine() : 1);
1193     Tok.setKind(tok::numeric_constant);
1194   } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__) {
1195     // C99 6.10.8: "__FILE__: The presumed name of the current source file (a
1196     // character string literal)". This can be affected by #line.
1197     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1198 
1199     // __BASE_FILE__ is a GNU extension that returns the top of the presumed
1200     // #include stack instead of the current file.
1201     if (II == Ident__BASE_FILE__ && PLoc.isValid()) {
1202       SourceLocation NextLoc = PLoc.getIncludeLoc();
1203       while (NextLoc.isValid()) {
1204         PLoc = SourceMgr.getPresumedLoc(NextLoc);
1205         if (PLoc.isInvalid())
1206           break;
1207 
1208         NextLoc = PLoc.getIncludeLoc();
1209       }
1210     }
1211 
1212     // Escape this filename.  Turn '\' -> '\\' '"' -> '\"'
1213     SmallString<128> FN;
1214     if (PLoc.isValid()) {
1215       FN += PLoc.getFilename();
1216       Lexer::Stringify(FN);
1217       OS << '"' << FN.str() << '"';
1218     }
1219     Tok.setKind(tok::string_literal);
1220   } else if (II == Ident__DATE__) {
1221     if (!DATELoc.isValid())
1222       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1223     Tok.setKind(tok::string_literal);
1224     Tok.setLength(strlen("\"Mmm dd yyyy\""));
1225     Tok.setLocation(SourceMgr.createExpansionLoc(DATELoc, Tok.getLocation(),
1226                                                  Tok.getLocation(),
1227                                                  Tok.getLength()));
1228     return;
1229   } else if (II == Ident__TIME__) {
1230     if (!TIMELoc.isValid())
1231       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1232     Tok.setKind(tok::string_literal);
1233     Tok.setLength(strlen("\"hh:mm:ss\""));
1234     Tok.setLocation(SourceMgr.createExpansionLoc(TIMELoc, Tok.getLocation(),
1235                                                  Tok.getLocation(),
1236                                                  Tok.getLength()));
1237     return;
1238   } else if (II == Ident__INCLUDE_LEVEL__) {
1239     // Compute the presumed include depth of this token.  This can be affected
1240     // by GNU line markers.
1241     unsigned Depth = 0;
1242 
1243     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1244     if (PLoc.isValid()) {
1245       PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1246       for (; PLoc.isValid(); ++Depth)
1247         PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1248     }
1249 
1250     // __INCLUDE_LEVEL__ expands to a simple numeric value.
1251     OS << Depth;
1252     Tok.setKind(tok::numeric_constant);
1253   } else if (II == Ident__TIMESTAMP__) {
1254     // MSVC, ICC, GCC, VisualAge C++ extension.  The generated string should be
1255     // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime.
1256 
1257     // Get the file that we are lexing out of.  If we're currently lexing from
1258     // a macro, dig into the include stack.
1259     const FileEntry *CurFile = 0;
1260     PreprocessorLexer *TheLexer = getCurrentFileLexer();
1261 
1262     if (TheLexer)
1263       CurFile = SourceMgr.getFileEntryForID(TheLexer->getFileID());
1264 
1265     const char *Result;
1266     if (CurFile) {
1267       time_t TT = CurFile->getModificationTime();
1268       struct tm *TM = localtime(&TT);
1269       Result = asctime(TM);
1270     } else {
1271       Result = "??? ??? ?? ??:??:?? ????\n";
1272     }
1273     // Surround the string with " and strip the trailing newline.
1274     OS << '"' << StringRef(Result, strlen(Result)-1) << '"';
1275     Tok.setKind(tok::string_literal);
1276   } else if (II == Ident__COUNTER__) {
1277     // __COUNTER__ expands to a simple numeric value.
1278     OS << CounterValue++;
1279     Tok.setKind(tok::numeric_constant);
1280   } else if (II == Ident__has_feature   ||
1281              II == Ident__has_extension ||
1282              II == Ident__has_builtin   ||
1283              II == Ident__has_attribute) {
1284     // The argument to these builtins should be a parenthesized identifier.
1285     SourceLocation StartLoc = Tok.getLocation();
1286 
1287     bool IsValid = false;
1288     IdentifierInfo *FeatureII = 0;
1289 
1290     // Read the '('.
1291     LexUnexpandedToken(Tok);
1292     if (Tok.is(tok::l_paren)) {
1293       // Read the identifier
1294       LexUnexpandedToken(Tok);
1295       if (Tok.is(tok::identifier) || Tok.is(tok::kw_const)) {
1296         FeatureII = Tok.getIdentifierInfo();
1297 
1298         // Read the ')'.
1299         LexUnexpandedToken(Tok);
1300         if (Tok.is(tok::r_paren))
1301           IsValid = true;
1302       }
1303     }
1304 
1305     bool Value = false;
1306     if (!IsValid)
1307       Diag(StartLoc, diag::err_feature_check_malformed);
1308     else if (II == Ident__has_builtin) {
1309       // Check for a builtin is trivial.
1310       Value = FeatureII->getBuiltinID() != 0;
1311     } else if (II == Ident__has_attribute)
1312       Value = HasAttribute(FeatureII);
1313     else if (II == Ident__has_extension)
1314       Value = HasExtension(*this, FeatureII);
1315     else {
1316       assert(II == Ident__has_feature && "Must be feature check");
1317       Value = HasFeature(*this, FeatureII);
1318     }
1319 
1320     OS << (int)Value;
1321     if (IsValid)
1322       Tok.setKind(tok::numeric_constant);
1323   } else if (II == Ident__has_include ||
1324              II == Ident__has_include_next) {
1325     // The argument to these two builtins should be a parenthesized
1326     // file name string literal using angle brackets (<>) or
1327     // double-quotes ("").
1328     bool Value;
1329     if (II == Ident__has_include)
1330       Value = EvaluateHasInclude(Tok, II, *this);
1331     else
1332       Value = EvaluateHasIncludeNext(Tok, II, *this);
1333     OS << (int)Value;
1334     if (Tok.is(tok::r_paren))
1335       Tok.setKind(tok::numeric_constant);
1336   } else if (II == Ident__has_warning) {
1337     // The argument should be a parenthesized string literal.
1338     // The argument to these builtins should be a parenthesized identifier.
1339     SourceLocation StartLoc = Tok.getLocation();
1340     bool IsValid = false;
1341     bool Value = false;
1342     // Read the '('.
1343     LexUnexpandedToken(Tok);
1344     do {
1345       if (Tok.isNot(tok::l_paren)) {
1346         Diag(StartLoc, diag::err_warning_check_malformed);
1347         break;
1348       }
1349 
1350       LexUnexpandedToken(Tok);
1351       std::string WarningName;
1352       SourceLocation StrStartLoc = Tok.getLocation();
1353       if (!FinishLexStringLiteral(Tok, WarningName, "'__has_warning'",
1354                                   /*MacroExpansion=*/false)) {
1355         // Eat tokens until ')'.
1356         while (Tok.isNot(tok::r_paren) && Tok.isNot(tok::eod) &&
1357                Tok.isNot(tok::eof))
1358           LexUnexpandedToken(Tok);
1359         break;
1360       }
1361 
1362       // Is the end a ')'?
1363       if (!(IsValid = Tok.is(tok::r_paren))) {
1364         Diag(StartLoc, diag::err_warning_check_malformed);
1365         break;
1366       }
1367 
1368       if (WarningName.size() < 3 || WarningName[0] != '-' ||
1369           WarningName[1] != 'W') {
1370         Diag(StrStartLoc, diag::warn_has_warning_invalid_option);
1371         break;
1372       }
1373 
1374       // Finally, check if the warning flags maps to a diagnostic group.
1375       // We construct a SmallVector here to talk to getDiagnosticIDs().
1376       // Although we don't use the result, this isn't a hot path, and not
1377       // worth special casing.
1378       SmallVector<diag::kind, 10> Diags;
1379       Value = !getDiagnostics().getDiagnosticIDs()->
1380         getDiagnosticsInGroup(WarningName.substr(2), Diags);
1381     } while (false);
1382 
1383     OS << (int)Value;
1384     if (IsValid)
1385       Tok.setKind(tok::numeric_constant);
1386   } else if (II == Ident__building_module) {
1387     // The argument to this builtin should be an identifier. The
1388     // builtin evaluates to 1 when that identifier names the module we are
1389     // currently building.
1390     OS << (int)EvaluateBuildingModule(Tok, II, *this);
1391     Tok.setKind(tok::numeric_constant);
1392   } else if (II == Ident__MODULE__) {
1393     // The current module as an identifier.
1394     OS << getLangOpts().CurrentModule;
1395     IdentifierInfo *ModuleII = getIdentifierInfo(getLangOpts().CurrentModule);
1396     Tok.setIdentifierInfo(ModuleII);
1397     Tok.setKind(ModuleII->getTokenID());
1398   } else {
1399     llvm_unreachable("Unknown identifier!");
1400   }
1401   CreateString(OS.str(), Tok, Tok.getLocation(), Tok.getLocation());
1402 }
1403 
1404 void Preprocessor::markMacroAsUsed(MacroInfo *MI) {
1405   // If the 'used' status changed, and the macro requires 'unused' warning,
1406   // remove its SourceLocation from the warn-for-unused-macro locations.
1407   if (MI->isWarnIfUnused() && !MI->isUsed())
1408     WarnUnusedMacroLocs.erase(MI->getDefinitionLoc());
1409   MI->setIsUsed(true);
1410 }
1411