• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 *
getMacroDirectiveHistory(const IdentifierInfo * II) const37 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 
appendMacroDirective(IdentifierInfo * II,MacroDirective * MD)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   II->setHasMacroDefinition(MD->isDefined());
53   bool isImportedMacro = isa<DefMacroDirective>(MD) &&
54                          cast<DefMacroDirective>(MD)->isImported();
55   if (II->isFromAST() && !isImportedMacro)
56     II->setChangedSinceDeserialization();
57 }
58 
setLoadedMacroDirective(IdentifierInfo * II,MacroDirective * MD)59 void Preprocessor::setLoadedMacroDirective(IdentifierInfo *II,
60                                            MacroDirective *MD) {
61   assert(II && MD);
62   MacroDirective *&StoredMD = Macros[II];
63   assert(!StoredMD &&
64          "the macro history was modified before initializing it from a pch");
65   StoredMD = MD;
66   // Setup the identifier as having associated macro history.
67   II->setHasMacroDefinition(true);
68   if (!MD->isDefined())
69     II->setHasMacroDefinition(false);
70 }
71 
72 /// RegisterBuiltinMacro - Register the specified identifier in the identifier
73 /// table and mark it as a builtin macro to be expanded.
RegisterBuiltinMacro(Preprocessor & PP,const char * Name)74 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){
75   // Get the identifier.
76   IdentifierInfo *Id = PP.getIdentifierInfo(Name);
77 
78   // Mark it as being a macro that is builtin.
79   MacroInfo *MI = PP.AllocateMacroInfo(SourceLocation());
80   MI->setIsBuiltinMacro();
81   PP.appendDefMacroDirective(Id, MI);
82   return Id;
83 }
84 
85 
86 /// RegisterBuiltinMacros - Register builtin macros, such as __LINE__ with the
87 /// identifier table.
RegisterBuiltinMacros()88 void Preprocessor::RegisterBuiltinMacros() {
89   Ident__LINE__ = RegisterBuiltinMacro(*this, "__LINE__");
90   Ident__FILE__ = RegisterBuiltinMacro(*this, "__FILE__");
91   Ident__DATE__ = RegisterBuiltinMacro(*this, "__DATE__");
92   Ident__TIME__ = RegisterBuiltinMacro(*this, "__TIME__");
93   Ident__COUNTER__ = RegisterBuiltinMacro(*this, "__COUNTER__");
94   Ident_Pragma  = RegisterBuiltinMacro(*this, "_Pragma");
95 
96   // GCC Extensions.
97   Ident__BASE_FILE__     = RegisterBuiltinMacro(*this, "__BASE_FILE__");
98   Ident__INCLUDE_LEVEL__ = RegisterBuiltinMacro(*this, "__INCLUDE_LEVEL__");
99   Ident__TIMESTAMP__     = RegisterBuiltinMacro(*this, "__TIMESTAMP__");
100 
101   // Microsoft Extensions.
102   if (LangOpts.MicrosoftExt) {
103     Ident__identifier = RegisterBuiltinMacro(*this, "__identifier");
104     Ident__pragma = RegisterBuiltinMacro(*this, "__pragma");
105   } else {
106     Ident__identifier = nullptr;
107     Ident__pragma = nullptr;
108   }
109 
110   // Clang Extensions.
111   Ident__has_feature      = RegisterBuiltinMacro(*this, "__has_feature");
112   Ident__has_extension    = RegisterBuiltinMacro(*this, "__has_extension");
113   Ident__has_builtin      = RegisterBuiltinMacro(*this, "__has_builtin");
114   Ident__has_attribute    = RegisterBuiltinMacro(*this, "__has_attribute");
115   Ident__has_include      = RegisterBuiltinMacro(*this, "__has_include");
116   Ident__has_include_next = RegisterBuiltinMacro(*this, "__has_include_next");
117   Ident__has_warning      = RegisterBuiltinMacro(*this, "__has_warning");
118   Ident__is_identifier    = RegisterBuiltinMacro(*this, "__is_identifier");
119 
120   // Modules.
121   if (LangOpts.Modules) {
122     Ident__building_module  = RegisterBuiltinMacro(*this, "__building_module");
123 
124     // __MODULE__
125     if (!LangOpts.CurrentModule.empty())
126       Ident__MODULE__ = RegisterBuiltinMacro(*this, "__MODULE__");
127     else
128       Ident__MODULE__ = nullptr;
129   } else {
130     Ident__building_module = nullptr;
131     Ident__MODULE__ = nullptr;
132   }
133 }
134 
135 /// isTrivialSingleTokenExpansion - Return true if MI, which has a single token
136 /// in its expansion, currently expands to that token literally.
isTrivialSingleTokenExpansion(const MacroInfo * MI,const IdentifierInfo * MacroIdent,Preprocessor & PP)137 static bool isTrivialSingleTokenExpansion(const MacroInfo *MI,
138                                           const IdentifierInfo *MacroIdent,
139                                           Preprocessor &PP) {
140   IdentifierInfo *II = MI->getReplacementToken(0).getIdentifierInfo();
141 
142   // If the token isn't an identifier, it's always literally expanded.
143   if (!II) return true;
144 
145   // If the information about this identifier is out of date, update it from
146   // the external source.
147   if (II->isOutOfDate())
148     PP.getExternalSource()->updateOutOfDateIdentifier(*II);
149 
150   // If the identifier is a macro, and if that macro is enabled, it may be
151   // expanded so it's not a trivial expansion.
152   if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled() &&
153       // Fast expanding "#define X X" is ok, because X would be disabled.
154       II != MacroIdent)
155     return false;
156 
157   // If this is an object-like macro invocation, it is safe to trivially expand
158   // it.
159   if (MI->isObjectLike()) return true;
160 
161   // If this is a function-like macro invocation, it's safe to trivially expand
162   // as long as the identifier is not a macro argument.
163   for (MacroInfo::arg_iterator I = MI->arg_begin(), E = MI->arg_end();
164        I != E; ++I)
165     if (*I == II)
166       return false;   // Identifier is a macro argument.
167 
168   return true;
169 }
170 
171 
172 /// isNextPPTokenLParen - Determine whether the next preprocessor token to be
173 /// lexed is a '('.  If so, consume the token and return true, if not, this
174 /// method should have no observable side-effect on the lexed tokens.
isNextPPTokenLParen()175 bool Preprocessor::isNextPPTokenLParen() {
176   // Do some quick tests for rejection cases.
177   unsigned Val;
178   if (CurLexer)
179     Val = CurLexer->isNextPPTokenLParen();
180   else if (CurPTHLexer)
181     Val = CurPTHLexer->isNextPPTokenLParen();
182   else
183     Val = CurTokenLexer->isNextTokenLParen();
184 
185   if (Val == 2) {
186     // We have run off the end.  If it's a source file we don't
187     // examine enclosing ones (C99 5.1.1.2p4).  Otherwise walk up the
188     // macro stack.
189     if (CurPPLexer)
190       return false;
191     for (unsigned i = IncludeMacroStack.size(); i != 0; --i) {
192       IncludeStackInfo &Entry = IncludeMacroStack[i-1];
193       if (Entry.TheLexer)
194         Val = Entry.TheLexer->isNextPPTokenLParen();
195       else if (Entry.ThePTHLexer)
196         Val = Entry.ThePTHLexer->isNextPPTokenLParen();
197       else
198         Val = Entry.TheTokenLexer->isNextTokenLParen();
199 
200       if (Val != 2)
201         break;
202 
203       // Ran off the end of a source file?
204       if (Entry.ThePPLexer)
205         return false;
206     }
207   }
208 
209   // Okay, if we know that the token is a '(', lex it and return.  Otherwise we
210   // have found something that isn't a '(' or we found the end of the
211   // translation unit.  In either case, return false.
212   return Val == 1;
213 }
214 
215 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be
216 /// expanded as a macro, handle it and return the next token as 'Identifier'.
HandleMacroExpandedIdentifier(Token & Identifier,MacroDirective * MD)217 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
218                                                  MacroDirective *MD) {
219   MacroDirective::DefInfo Def = MD->getDefinition();
220   assert(Def.isValid());
221   MacroInfo *MI = Def.getMacroInfo();
222 
223   // If this is a macro expansion in the "#if !defined(x)" line for the file,
224   // then the macro could expand to different things in other contexts, we need
225   // to disable the optimization in this case.
226   if (CurPPLexer) CurPPLexer->MIOpt.ExpandedMacro();
227 
228   // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially.
229   if (MI->isBuiltinMacro()) {
230     if (Callbacks) Callbacks->MacroExpands(Identifier, MD,
231                                            Identifier.getLocation(),
232                                            /*Args=*/nullptr);
233     ExpandBuiltinMacro(Identifier);
234     return true;
235   }
236 
237   /// Args - If this is a function-like macro expansion, this contains,
238   /// for each macro argument, the list of tokens that were provided to the
239   /// invocation.
240   MacroArgs *Args = nullptr;
241 
242   // Remember where the end of the expansion occurred.  For an object-like
243   // macro, this is the identifier.  For a function-like macro, this is the ')'.
244   SourceLocation ExpansionEnd = Identifier.getLocation();
245 
246   // If this is a function-like macro, read the arguments.
247   if (MI->isFunctionLike()) {
248     // Remember that we are now parsing the arguments to a macro invocation.
249     // Preprocessor directives used inside macro arguments are not portable, and
250     // this enables the warning.
251     InMacroArgs = true;
252     Args = ReadFunctionLikeMacroArgs(Identifier, MI, ExpansionEnd);
253 
254     // Finished parsing args.
255     InMacroArgs = false;
256 
257     // If there was an error parsing the arguments, bail out.
258     if (!Args) return true;
259 
260     ++NumFnMacroExpanded;
261   } else {
262     ++NumMacroExpanded;
263   }
264 
265   // Notice that this macro has been used.
266   markMacroAsUsed(MI);
267 
268   // Remember where the token is expanded.
269   SourceLocation ExpandLoc = Identifier.getLocation();
270   SourceRange ExpansionRange(ExpandLoc, ExpansionEnd);
271 
272   if (Callbacks) {
273     if (InMacroArgs) {
274       // We can have macro expansion inside a conditional directive while
275       // reading the function macro arguments. To ensure, in that case, that
276       // MacroExpands callbacks still happen in source order, queue this
277       // callback to have it happen after the function macro callback.
278       DelayedMacroExpandsCallbacks.push_back(
279                               MacroExpandsInfo(Identifier, MD, ExpansionRange));
280     } else {
281       Callbacks->MacroExpands(Identifier, MD, ExpansionRange, Args);
282       if (!DelayedMacroExpandsCallbacks.empty()) {
283         for (unsigned i=0, e = DelayedMacroExpandsCallbacks.size(); i!=e; ++i) {
284           MacroExpandsInfo &Info = DelayedMacroExpandsCallbacks[i];
285           // FIXME: We lose macro args info with delayed callback.
286           Callbacks->MacroExpands(Info.Tok, Info.MD, Info.Range,
287                                   /*Args=*/nullptr);
288         }
289         DelayedMacroExpandsCallbacks.clear();
290       }
291     }
292   }
293 
294   // If the macro definition is ambiguous, complain.
295   if (Def.getDirective()->isAmbiguous()) {
296     Diag(Identifier, diag::warn_pp_ambiguous_macro)
297       << Identifier.getIdentifierInfo();
298     Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen)
299       << Identifier.getIdentifierInfo();
300     for (MacroDirective::DefInfo PrevDef = Def.getPreviousDefinition();
301          PrevDef && !PrevDef.isUndefined();
302          PrevDef = PrevDef.getPreviousDefinition()) {
303       Diag(PrevDef.getMacroInfo()->getDefinitionLoc(),
304            diag::note_pp_ambiguous_macro_other)
305         << Identifier.getIdentifierInfo();
306       if (!PrevDef.getDirective()->isAmbiguous())
307         break;
308     }
309   }
310 
311   // If we started lexing a macro, enter the macro expansion body.
312 
313   // If this macro expands to no tokens, don't bother to push it onto the
314   // expansion stack, only to take it right back off.
315   if (MI->getNumTokens() == 0) {
316     // No need for arg info.
317     if (Args) Args->destroy(*this);
318 
319     // Propagate whitespace info as if we had pushed, then popped,
320     // a macro context.
321     Identifier.setFlag(Token::LeadingEmptyMacro);
322     PropagateLineStartLeadingSpaceInfo(Identifier);
323     ++NumFastMacroExpanded;
324     return false;
325   } else if (MI->getNumTokens() == 1 &&
326              isTrivialSingleTokenExpansion(MI, Identifier.getIdentifierInfo(),
327                                            *this)) {
328     // Otherwise, if this macro expands into a single trivially-expanded
329     // token: expand it now.  This handles common cases like
330     // "#define VAL 42".
331 
332     // No need for arg info.
333     if (Args) Args->destroy(*this);
334 
335     // Propagate the isAtStartOfLine/hasLeadingSpace markers of the macro
336     // identifier to the expanded token.
337     bool isAtStartOfLine = Identifier.isAtStartOfLine();
338     bool hasLeadingSpace = Identifier.hasLeadingSpace();
339 
340     // Replace the result token.
341     Identifier = MI->getReplacementToken(0);
342 
343     // Restore the StartOfLine/LeadingSpace markers.
344     Identifier.setFlagValue(Token::StartOfLine , isAtStartOfLine);
345     Identifier.setFlagValue(Token::LeadingSpace, hasLeadingSpace);
346 
347     // Update the tokens location to include both its expansion and physical
348     // locations.
349     SourceLocation Loc =
350       SourceMgr.createExpansionLoc(Identifier.getLocation(), ExpandLoc,
351                                    ExpansionEnd,Identifier.getLength());
352     Identifier.setLocation(Loc);
353 
354     // If this is a disabled macro or #define X X, we must mark the result as
355     // unexpandable.
356     if (IdentifierInfo *NewII = Identifier.getIdentifierInfo()) {
357       if (MacroInfo *NewMI = getMacroInfo(NewII))
358         if (!NewMI->isEnabled() || NewMI == MI) {
359           Identifier.setFlag(Token::DisableExpand);
360           // Don't warn for "#define X X" like "#define bool bool" from
361           // stdbool.h.
362           if (NewMI != MI || MI->isFunctionLike())
363             Diag(Identifier, diag::pp_disabled_macro_expansion);
364         }
365     }
366 
367     // Since this is not an identifier token, it can't be macro expanded, so
368     // we're done.
369     ++NumFastMacroExpanded;
370     return true;
371   }
372 
373   // Start expanding the macro.
374   EnterMacro(Identifier, ExpansionEnd, MI, Args);
375   return false;
376 }
377 
378 enum Bracket {
379   Brace,
380   Paren
381 };
382 
383 /// CheckMatchedBrackets - Returns true if the braces and parentheses in the
384 /// token vector are properly nested.
CheckMatchedBrackets(const SmallVectorImpl<Token> & Tokens)385 static bool CheckMatchedBrackets(const SmallVectorImpl<Token> &Tokens) {
386   SmallVector<Bracket, 8> Brackets;
387   for (SmallVectorImpl<Token>::const_iterator I = Tokens.begin(),
388                                               E = Tokens.end();
389        I != E; ++I) {
390     if (I->is(tok::l_paren)) {
391       Brackets.push_back(Paren);
392     } else if (I->is(tok::r_paren)) {
393       if (Brackets.empty() || Brackets.back() == Brace)
394         return false;
395       Brackets.pop_back();
396     } else if (I->is(tok::l_brace)) {
397       Brackets.push_back(Brace);
398     } else if (I->is(tok::r_brace)) {
399       if (Brackets.empty() || Brackets.back() == Paren)
400         return false;
401       Brackets.pop_back();
402     }
403   }
404   if (!Brackets.empty())
405     return false;
406   return true;
407 }
408 
409 /// GenerateNewArgTokens - Returns true if OldTokens can be converted to a new
410 /// vector of tokens in NewTokens.  The new number of arguments will be placed
411 /// in NumArgs and the ranges which need to surrounded in parentheses will be
412 /// in ParenHints.
413 /// Returns false if the token stream cannot be changed.  If this is because
414 /// of an initializer list starting a macro argument, the range of those
415 /// initializer lists will be place in InitLists.
GenerateNewArgTokens(Preprocessor & PP,SmallVectorImpl<Token> & OldTokens,SmallVectorImpl<Token> & NewTokens,unsigned & NumArgs,SmallVectorImpl<SourceRange> & ParenHints,SmallVectorImpl<SourceRange> & InitLists)416 static bool GenerateNewArgTokens(Preprocessor &PP,
417                                  SmallVectorImpl<Token> &OldTokens,
418                                  SmallVectorImpl<Token> &NewTokens,
419                                  unsigned &NumArgs,
420                                  SmallVectorImpl<SourceRange> &ParenHints,
421                                  SmallVectorImpl<SourceRange> &InitLists) {
422   if (!CheckMatchedBrackets(OldTokens))
423     return false;
424 
425   // Once it is known that the brackets are matched, only a simple count of the
426   // braces is needed.
427   unsigned Braces = 0;
428 
429   // First token of a new macro argument.
430   SmallVectorImpl<Token>::iterator ArgStartIterator = OldTokens.begin();
431 
432   // First closing brace in a new macro argument.  Used to generate
433   // SourceRanges for InitLists.
434   SmallVectorImpl<Token>::iterator ClosingBrace = OldTokens.end();
435   NumArgs = 0;
436   Token TempToken;
437   // Set to true when a macro separator token is found inside a braced list.
438   // If true, the fixed argument spans multiple old arguments and ParenHints
439   // will be updated.
440   bool FoundSeparatorToken = false;
441   for (SmallVectorImpl<Token>::iterator I = OldTokens.begin(),
442                                         E = OldTokens.end();
443        I != E; ++I) {
444     if (I->is(tok::l_brace)) {
445       ++Braces;
446     } else if (I->is(tok::r_brace)) {
447       --Braces;
448       if (Braces == 0 && ClosingBrace == E && FoundSeparatorToken)
449         ClosingBrace = I;
450     } else if (I->is(tok::eof)) {
451       // EOF token is used to separate macro arguments
452       if (Braces != 0) {
453         // Assume comma separator is actually braced list separator and change
454         // it back to a comma.
455         FoundSeparatorToken = true;
456         I->setKind(tok::comma);
457         I->setLength(1);
458       } else { // Braces == 0
459         // Separator token still separates arguments.
460         ++NumArgs;
461 
462         // If the argument starts with a brace, it can't be fixed with
463         // parentheses.  A different diagnostic will be given.
464         if (FoundSeparatorToken && ArgStartIterator->is(tok::l_brace)) {
465           InitLists.push_back(
466               SourceRange(ArgStartIterator->getLocation(),
467                           PP.getLocForEndOfToken(ClosingBrace->getLocation())));
468           ClosingBrace = E;
469         }
470 
471         // Add left paren
472         if (FoundSeparatorToken) {
473           TempToken.startToken();
474           TempToken.setKind(tok::l_paren);
475           TempToken.setLocation(ArgStartIterator->getLocation());
476           TempToken.setLength(0);
477           NewTokens.push_back(TempToken);
478         }
479 
480         // Copy over argument tokens
481         NewTokens.insert(NewTokens.end(), ArgStartIterator, I);
482 
483         // Add right paren and store the paren locations in ParenHints
484         if (FoundSeparatorToken) {
485           SourceLocation Loc = PP.getLocForEndOfToken((I - 1)->getLocation());
486           TempToken.startToken();
487           TempToken.setKind(tok::r_paren);
488           TempToken.setLocation(Loc);
489           TempToken.setLength(0);
490           NewTokens.push_back(TempToken);
491           ParenHints.push_back(SourceRange(ArgStartIterator->getLocation(),
492                                            Loc));
493         }
494 
495         // Copy separator token
496         NewTokens.push_back(*I);
497 
498         // Reset values
499         ArgStartIterator = I + 1;
500         FoundSeparatorToken = false;
501       }
502     }
503   }
504 
505   return !ParenHints.empty() && InitLists.empty();
506 }
507 
508 /// ReadFunctionLikeMacroArgs - After reading "MACRO" and knowing that the next
509 /// token is the '(' of the macro, this method is invoked to read all of the
510 /// actual arguments specified for the macro invocation.  This returns null on
511 /// error.
ReadFunctionLikeMacroArgs(Token & MacroName,MacroInfo * MI,SourceLocation & MacroEnd)512 MacroArgs *Preprocessor::ReadFunctionLikeMacroArgs(Token &MacroName,
513                                                    MacroInfo *MI,
514                                                    SourceLocation &MacroEnd) {
515   // The number of fixed arguments to parse.
516   unsigned NumFixedArgsLeft = MI->getNumArgs();
517   bool isVariadic = MI->isVariadic();
518 
519   // Outer loop, while there are more arguments, keep reading them.
520   Token Tok;
521 
522   // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
523   // an argument value in a macro could expand to ',' or '(' or ')'.
524   LexUnexpandedToken(Tok);
525   assert(Tok.is(tok::l_paren) && "Error computing l-paren-ness?");
526 
527   // ArgTokens - Build up a list of tokens that make up each argument.  Each
528   // argument is separated by an EOF token.  Use a SmallVector so we can avoid
529   // heap allocations in the common case.
530   SmallVector<Token, 64> ArgTokens;
531   bool ContainsCodeCompletionTok = false;
532 
533   SourceLocation TooManyArgsLoc;
534 
535   unsigned NumActuals = 0;
536   while (Tok.isNot(tok::r_paren)) {
537     if (ContainsCodeCompletionTok && (Tok.is(tok::eof) || Tok.is(tok::eod)))
538       break;
539 
540     assert((Tok.is(tok::l_paren) || Tok.is(tok::comma)) &&
541            "only expect argument separators here");
542 
543     unsigned ArgTokenStart = ArgTokens.size();
544     SourceLocation ArgStartLoc = Tok.getLocation();
545 
546     // C99 6.10.3p11: Keep track of the number of l_parens we have seen.  Note
547     // that we already consumed the first one.
548     unsigned NumParens = 0;
549 
550     while (1) {
551       // Read arguments as unexpanded tokens.  This avoids issues, e.g., where
552       // an argument value in a macro could expand to ',' or '(' or ')'.
553       LexUnexpandedToken(Tok);
554 
555       if (Tok.is(tok::eof) || Tok.is(tok::eod)) { // "#if f(<eof>" & "#if f(\n"
556         if (!ContainsCodeCompletionTok) {
557           Diag(MacroName, diag::err_unterm_macro_invoc);
558           Diag(MI->getDefinitionLoc(), diag::note_macro_here)
559             << MacroName.getIdentifierInfo();
560           // Do not lose the EOF/EOD.  Return it to the client.
561           MacroName = Tok;
562           return nullptr;
563         } else {
564           // Do not lose the EOF/EOD.
565           Token *Toks = new Token[1];
566           Toks[0] = Tok;
567           EnterTokenStream(Toks, 1, true, true);
568           break;
569         }
570       } else if (Tok.is(tok::r_paren)) {
571         // If we found the ) token, the macro arg list is done.
572         if (NumParens-- == 0) {
573           MacroEnd = Tok.getLocation();
574           break;
575         }
576       } else if (Tok.is(tok::l_paren)) {
577         ++NumParens;
578       } else if (Tok.is(tok::comma) && NumParens == 0 &&
579                  !(Tok.getFlags() & Token::IgnoredComma)) {
580         // In Microsoft-compatibility mode, single commas from nested macro
581         // expansions should not be considered as argument separators. We test
582         // for this with the IgnoredComma token flag above.
583 
584         // Comma ends this argument if there are more fixed arguments expected.
585         // However, if this is a variadic macro, and this is part of the
586         // variadic part, then the comma is just an argument token.
587         if (!isVariadic) break;
588         if (NumFixedArgsLeft > 1)
589           break;
590       } else if (Tok.is(tok::comment) && !KeepMacroComments) {
591         // If this is a comment token in the argument list and we're just in
592         // -C mode (not -CC mode), discard the comment.
593         continue;
594       } else if (Tok.getIdentifierInfo() != nullptr) {
595         // Reading macro arguments can cause macros that we are currently
596         // expanding from to be popped off the expansion stack.  Doing so causes
597         // them to be reenabled for expansion.  Here we record whether any
598         // identifiers we lex as macro arguments correspond to disabled macros.
599         // If so, we mark the token as noexpand.  This is a subtle aspect of
600         // C99 6.10.3.4p2.
601         if (MacroInfo *MI = getMacroInfo(Tok.getIdentifierInfo()))
602           if (!MI->isEnabled())
603             Tok.setFlag(Token::DisableExpand);
604       } else if (Tok.is(tok::code_completion)) {
605         ContainsCodeCompletionTok = true;
606         if (CodeComplete)
607           CodeComplete->CodeCompleteMacroArgument(MacroName.getIdentifierInfo(),
608                                                   MI, NumActuals);
609         // Don't mark that we reached the code-completion point because the
610         // parser is going to handle the token and there will be another
611         // code-completion callback.
612       }
613 
614       ArgTokens.push_back(Tok);
615     }
616 
617     // If this was an empty argument list foo(), don't add this as an empty
618     // argument.
619     if (ArgTokens.empty() && Tok.getKind() == tok::r_paren)
620       break;
621 
622     // If this is not a variadic macro, and too many args were specified, emit
623     // an error.
624     if (!isVariadic && NumFixedArgsLeft == 0 && TooManyArgsLoc.isInvalid()) {
625       if (ArgTokens.size() != ArgTokenStart)
626         TooManyArgsLoc = ArgTokens[ArgTokenStart].getLocation();
627       else
628         TooManyArgsLoc = ArgStartLoc;
629     }
630 
631     // Empty arguments are standard in C99 and C++0x, and are supported as an
632     // extension in other modes.
633     if (ArgTokens.size() == ArgTokenStart && !LangOpts.C99)
634       Diag(Tok, LangOpts.CPlusPlus11 ?
635            diag::warn_cxx98_compat_empty_fnmacro_arg :
636            diag::ext_empty_fnmacro_arg);
637 
638     // Add a marker EOF token to the end of the token list for this argument.
639     Token EOFTok;
640     EOFTok.startToken();
641     EOFTok.setKind(tok::eof);
642     EOFTok.setLocation(Tok.getLocation());
643     EOFTok.setLength(0);
644     ArgTokens.push_back(EOFTok);
645     ++NumActuals;
646     if (!ContainsCodeCompletionTok && NumFixedArgsLeft != 0)
647       --NumFixedArgsLeft;
648   }
649 
650   // Okay, we either found the r_paren.  Check to see if we parsed too few
651   // arguments.
652   unsigned MinArgsExpected = MI->getNumArgs();
653 
654   // If this is not a variadic macro, and too many args were specified, emit
655   // an error.
656   if (!isVariadic && NumActuals > MinArgsExpected &&
657       !ContainsCodeCompletionTok) {
658     // Emit the diagnostic at the macro name in case there is a missing ).
659     // Emitting it at the , could be far away from the macro name.
660     Diag(TooManyArgsLoc, diag::err_too_many_args_in_macro_invoc);
661     Diag(MI->getDefinitionLoc(), diag::note_macro_here)
662       << MacroName.getIdentifierInfo();
663 
664     // Commas from braced initializer lists will be treated as argument
665     // separators inside macros.  Attempt to correct for this with parentheses.
666     // TODO: See if this can be generalized to angle brackets for templates
667     // inside macro arguments.
668 
669     SmallVector<Token, 4> FixedArgTokens;
670     unsigned FixedNumArgs = 0;
671     SmallVector<SourceRange, 4> ParenHints, InitLists;
672     if (!GenerateNewArgTokens(*this, ArgTokens, FixedArgTokens, FixedNumArgs,
673                               ParenHints, InitLists)) {
674       if (!InitLists.empty()) {
675         DiagnosticBuilder DB =
676             Diag(MacroName,
677                  diag::note_init_list_at_beginning_of_macro_argument);
678         for (const SourceRange &Range : InitLists)
679           DB << Range;
680       }
681       return nullptr;
682     }
683     if (FixedNumArgs != MinArgsExpected)
684       return nullptr;
685 
686     DiagnosticBuilder DB = Diag(MacroName, diag::note_suggest_parens_for_macro);
687     for (const SourceRange &ParenLocation : ParenHints) {
688       DB << FixItHint::CreateInsertion(ParenLocation.getBegin(), "(");
689       DB << FixItHint::CreateInsertion(ParenLocation.getEnd(), ")");
690     }
691     ArgTokens.swap(FixedArgTokens);
692     NumActuals = FixedNumArgs;
693   }
694 
695   // See MacroArgs instance var for description of this.
696   bool isVarargsElided = false;
697 
698   if (ContainsCodeCompletionTok) {
699     // Recover from not-fully-formed macro invocation during code-completion.
700     Token EOFTok;
701     EOFTok.startToken();
702     EOFTok.setKind(tok::eof);
703     EOFTok.setLocation(Tok.getLocation());
704     EOFTok.setLength(0);
705     for (; NumActuals < MinArgsExpected; ++NumActuals)
706       ArgTokens.push_back(EOFTok);
707   }
708 
709   if (NumActuals < MinArgsExpected) {
710     // There are several cases where too few arguments is ok, handle them now.
711     if (NumActuals == 0 && MinArgsExpected == 1) {
712       // #define A(X)  or  #define A(...)   ---> A()
713 
714       // If there is exactly one argument, and that argument is missing,
715       // then we have an empty "()" argument empty list.  This is fine, even if
716       // the macro expects one argument (the argument is just empty).
717       isVarargsElided = MI->isVariadic();
718     } else if (MI->isVariadic() &&
719                (NumActuals+1 == MinArgsExpected ||  // A(x, ...) -> A(X)
720                 (NumActuals == 0 && MinArgsExpected == 2))) {// A(x,...) -> A()
721       // Varargs where the named vararg parameter is missing: OK as extension.
722       //   #define A(x, ...)
723       //   A("blah")
724       //
725       // If the macro contains the comma pasting extension, the diagnostic
726       // is suppressed; we know we'll get another diagnostic later.
727       if (!MI->hasCommaPasting()) {
728         Diag(Tok, diag::ext_missing_varargs_arg);
729         Diag(MI->getDefinitionLoc(), diag::note_macro_here)
730           << MacroName.getIdentifierInfo();
731       }
732 
733       // Remember this occurred, allowing us to elide the comma when used for
734       // cases like:
735       //   #define A(x, foo...) blah(a, ## foo)
736       //   #define B(x, ...) blah(a, ## __VA_ARGS__)
737       //   #define C(...) blah(a, ## __VA_ARGS__)
738       //  A(x) B(x) C()
739       isVarargsElided = true;
740     } else if (!ContainsCodeCompletionTok) {
741       // Otherwise, emit the error.
742       Diag(Tok, diag::err_too_few_args_in_macro_invoc);
743       Diag(MI->getDefinitionLoc(), diag::note_macro_here)
744         << MacroName.getIdentifierInfo();
745       return nullptr;
746     }
747 
748     // Add a marker EOF token to the end of the token list for this argument.
749     SourceLocation EndLoc = Tok.getLocation();
750     Tok.startToken();
751     Tok.setKind(tok::eof);
752     Tok.setLocation(EndLoc);
753     Tok.setLength(0);
754     ArgTokens.push_back(Tok);
755 
756     // If we expect two arguments, add both as empty.
757     if (NumActuals == 0 && MinArgsExpected == 2)
758       ArgTokens.push_back(Tok);
759 
760   } else if (NumActuals > MinArgsExpected && !MI->isVariadic() &&
761              !ContainsCodeCompletionTok) {
762     // Emit the diagnostic at the macro name in case there is a missing ).
763     // Emitting it at the , could be far away from the macro name.
764     Diag(MacroName, diag::err_too_many_args_in_macro_invoc);
765     Diag(MI->getDefinitionLoc(), diag::note_macro_here)
766       << MacroName.getIdentifierInfo();
767     return nullptr;
768   }
769 
770   return MacroArgs::create(MI, ArgTokens, isVarargsElided, *this);
771 }
772 
773 /// \brief Keeps macro expanded tokens for TokenLexers.
774 //
775 /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
776 /// going to lex in the cache and when it finishes the tokens are removed
777 /// from the end of the cache.
cacheMacroExpandedTokens(TokenLexer * tokLexer,ArrayRef<Token> tokens)778 Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer,
779                                               ArrayRef<Token> tokens) {
780   assert(tokLexer);
781   if (tokens.empty())
782     return nullptr;
783 
784   size_t newIndex = MacroExpandedTokens.size();
785   bool cacheNeedsToGrow = tokens.size() >
786                       MacroExpandedTokens.capacity()-MacroExpandedTokens.size();
787   MacroExpandedTokens.append(tokens.begin(), tokens.end());
788 
789   if (cacheNeedsToGrow) {
790     // Go through all the TokenLexers whose 'Tokens' pointer points in the
791     // buffer and update the pointers to the (potential) new buffer array.
792     for (unsigned i = 0, e = MacroExpandingLexersStack.size(); i != e; ++i) {
793       TokenLexer *prevLexer;
794       size_t tokIndex;
795       std::tie(prevLexer, tokIndex) = MacroExpandingLexersStack[i];
796       prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex;
797     }
798   }
799 
800   MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex));
801   return MacroExpandedTokens.data() + newIndex;
802 }
803 
removeCachedMacroExpandedTokensOfLastLexer()804 void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() {
805   assert(!MacroExpandingLexersStack.empty());
806   size_t tokIndex = MacroExpandingLexersStack.back().second;
807   assert(tokIndex < MacroExpandedTokens.size());
808   // Pop the cached macro expanded tokens from the end.
809   MacroExpandedTokens.resize(tokIndex);
810   MacroExpandingLexersStack.pop_back();
811 }
812 
813 /// ComputeDATE_TIME - Compute the current time, enter it into the specified
814 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of
815 /// the identifier tokens inserted.
ComputeDATE_TIME(SourceLocation & DATELoc,SourceLocation & TIMELoc,Preprocessor & PP)816 static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc,
817                              Preprocessor &PP) {
818   time_t TT = time(nullptr);
819   struct tm *TM = localtime(&TT);
820 
821   static const char * const Months[] = {
822     "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"
823   };
824 
825   {
826     SmallString<32> TmpBuffer;
827     llvm::raw_svector_ostream TmpStream(TmpBuffer);
828     TmpStream << llvm::format("\"%s %2d %4d\"", Months[TM->tm_mon],
829                               TM->tm_mday, TM->tm_year + 1900);
830     Token TmpTok;
831     TmpTok.startToken();
832     PP.CreateString(TmpStream.str(), TmpTok);
833     DATELoc = TmpTok.getLocation();
834   }
835 
836   {
837     SmallString<32> TmpBuffer;
838     llvm::raw_svector_ostream TmpStream(TmpBuffer);
839     TmpStream << llvm::format("\"%02d:%02d:%02d\"",
840                               TM->tm_hour, TM->tm_min, TM->tm_sec);
841     Token TmpTok;
842     TmpTok.startToken();
843     PP.CreateString(TmpStream.str(), TmpTok);
844     TIMELoc = TmpTok.getLocation();
845   }
846 }
847 
848 
849 /// HasFeature - Return true if we recognize and implement the feature
850 /// specified by the identifier as a standard language feature.
HasFeature(const Preprocessor & PP,const IdentifierInfo * II)851 static bool HasFeature(const Preprocessor &PP, const IdentifierInfo *II) {
852   const LangOptions &LangOpts = PP.getLangOpts();
853   StringRef Feature = II->getName();
854 
855   // Normalize the feature name, __foo__ becomes foo.
856   if (Feature.startswith("__") && Feature.endswith("__") && Feature.size() >= 4)
857     Feature = Feature.substr(2, Feature.size() - 4);
858 
859   return llvm::StringSwitch<bool>(Feature)
860            .Case("address_sanitizer", LangOpts.Sanitize.Address)
861            .Case("attribute_analyzer_noreturn", true)
862            .Case("attribute_availability", true)
863            .Case("attribute_availability_with_message", true)
864            .Case("attribute_cf_returns_not_retained", true)
865            .Case("attribute_cf_returns_retained", true)
866            .Case("attribute_deprecated_with_message", true)
867            .Case("attribute_ext_vector_type", true)
868            .Case("attribute_ns_returns_not_retained", true)
869            .Case("attribute_ns_returns_retained", true)
870            .Case("attribute_ns_consumes_self", true)
871            .Case("attribute_ns_consumed", true)
872            .Case("attribute_cf_consumed", true)
873            .Case("attribute_objc_ivar_unused", true)
874            .Case("attribute_objc_method_family", true)
875            .Case("attribute_overloadable", true)
876            .Case("attribute_unavailable_with_message", true)
877            .Case("attribute_unused_on_fields", true)
878            .Case("blocks", LangOpts.Blocks)
879            .Case("c_thread_safety_attributes", true)
880            .Case("cxx_exceptions", LangOpts.CXXExceptions)
881            .Case("cxx_rtti", LangOpts.RTTI)
882            .Case("enumerator_attributes", true)
883            .Case("memory_sanitizer", LangOpts.Sanitize.Memory)
884            .Case("thread_sanitizer", LangOpts.Sanitize.Thread)
885            .Case("dataflow_sanitizer", LangOpts.Sanitize.DataFlow)
886            // Objective-C features
887            .Case("objc_arr", LangOpts.ObjCAutoRefCount) // FIXME: REMOVE?
888            .Case("objc_arc", LangOpts.ObjCAutoRefCount)
889            .Case("objc_arc_weak", LangOpts.ObjCARCWeak)
890            .Case("objc_default_synthesize_properties", LangOpts.ObjC2)
891            .Case("objc_fixed_enum", LangOpts.ObjC2)
892            .Case("objc_instancetype", LangOpts.ObjC2)
893            .Case("objc_modules", LangOpts.ObjC2 && LangOpts.Modules)
894            .Case("objc_nonfragile_abi", LangOpts.ObjCRuntime.isNonFragile())
895            .Case("objc_property_explicit_atomic", true) // Does clang support explicit "atomic" keyword?
896            .Case("objc_protocol_qualifier_mangling", true)
897            .Case("objc_weak_class", LangOpts.ObjCRuntime.hasWeakClassImport())
898            .Case("ownership_holds", true)
899            .Case("ownership_returns", true)
900            .Case("ownership_takes", true)
901            .Case("objc_bool", true)
902            .Case("objc_subscripting", LangOpts.ObjCRuntime.isNonFragile())
903            .Case("objc_array_literals", LangOpts.ObjC2)
904            .Case("objc_dictionary_literals", LangOpts.ObjC2)
905            .Case("objc_boxed_expressions", LangOpts.ObjC2)
906            .Case("arc_cf_code_audited", true)
907            // C11 features
908            .Case("c_alignas", LangOpts.C11)
909            .Case("c_atomic", LangOpts.C11)
910            .Case("c_generic_selections", LangOpts.C11)
911            .Case("c_static_assert", LangOpts.C11)
912            .Case("c_thread_local",
913                  LangOpts.C11 && PP.getTargetInfo().isTLSSupported())
914            // C++11 features
915            .Case("cxx_access_control_sfinae", LangOpts.CPlusPlus11)
916            .Case("cxx_alias_templates", LangOpts.CPlusPlus11)
917            .Case("cxx_alignas", LangOpts.CPlusPlus11)
918            .Case("cxx_atomic", LangOpts.CPlusPlus11)
919            .Case("cxx_attributes", LangOpts.CPlusPlus11)
920            .Case("cxx_auto_type", LangOpts.CPlusPlus11)
921            .Case("cxx_constexpr", LangOpts.CPlusPlus11)
922            .Case("cxx_decltype", LangOpts.CPlusPlus11)
923            .Case("cxx_decltype_incomplete_return_types", LangOpts.CPlusPlus11)
924            .Case("cxx_default_function_template_args", LangOpts.CPlusPlus11)
925            .Case("cxx_defaulted_functions", LangOpts.CPlusPlus11)
926            .Case("cxx_delegating_constructors", LangOpts.CPlusPlus11)
927            .Case("cxx_deleted_functions", LangOpts.CPlusPlus11)
928            .Case("cxx_explicit_conversions", LangOpts.CPlusPlus11)
929            .Case("cxx_generalized_initializers", LangOpts.CPlusPlus11)
930            .Case("cxx_implicit_moves", LangOpts.CPlusPlus11)
931            .Case("cxx_inheriting_constructors", LangOpts.CPlusPlus11)
932            .Case("cxx_inline_namespaces", LangOpts.CPlusPlus11)
933            .Case("cxx_lambdas", LangOpts.CPlusPlus11)
934            .Case("cxx_local_type_template_args", LangOpts.CPlusPlus11)
935            .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus11)
936            .Case("cxx_noexcept", LangOpts.CPlusPlus11)
937            .Case("cxx_nullptr", LangOpts.CPlusPlus11)
938            .Case("cxx_override_control", LangOpts.CPlusPlus11)
939            .Case("cxx_range_for", LangOpts.CPlusPlus11)
940            .Case("cxx_raw_string_literals", LangOpts.CPlusPlus11)
941            .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus11)
942            .Case("cxx_rvalue_references", LangOpts.CPlusPlus11)
943            .Case("cxx_strong_enums", LangOpts.CPlusPlus11)
944            .Case("cxx_static_assert", LangOpts.CPlusPlus11)
945            .Case("cxx_thread_local",
946                  LangOpts.CPlusPlus11 && PP.getTargetInfo().isTLSSupported())
947            .Case("cxx_trailing_return", LangOpts.CPlusPlus11)
948            .Case("cxx_unicode_literals", LangOpts.CPlusPlus11)
949            .Case("cxx_unrestricted_unions", LangOpts.CPlusPlus11)
950            .Case("cxx_user_literals", LangOpts.CPlusPlus11)
951            .Case("cxx_variadic_templates", LangOpts.CPlusPlus11)
952            // C++1y features
953            .Case("cxx_aggregate_nsdmi", LangOpts.CPlusPlus1y)
954            .Case("cxx_binary_literals", LangOpts.CPlusPlus1y)
955            .Case("cxx_contextual_conversions", LangOpts.CPlusPlus1y)
956            .Case("cxx_decltype_auto", LangOpts.CPlusPlus1y)
957            .Case("cxx_generic_lambdas", LangOpts.CPlusPlus1y)
958            .Case("cxx_init_captures", LangOpts.CPlusPlus1y)
959            .Case("cxx_relaxed_constexpr", LangOpts.CPlusPlus1y)
960            .Case("cxx_return_type_deduction", LangOpts.CPlusPlus1y)
961            .Case("cxx_variable_templates", LangOpts.CPlusPlus1y)
962            // C++ TSes
963            //.Case("cxx_runtime_arrays", LangOpts.CPlusPlusTSArrays)
964            //.Case("cxx_concepts", LangOpts.CPlusPlusTSConcepts)
965            // FIXME: Should this be __has_feature or __has_extension?
966            //.Case("raw_invocation_type", LangOpts.CPlusPlus)
967            // Type traits
968            .Case("has_nothrow_assign", LangOpts.CPlusPlus)
969            .Case("has_nothrow_copy", LangOpts.CPlusPlus)
970            .Case("has_nothrow_constructor", LangOpts.CPlusPlus)
971            .Case("has_trivial_assign", LangOpts.CPlusPlus)
972            .Case("has_trivial_copy", LangOpts.CPlusPlus)
973            .Case("has_trivial_constructor", LangOpts.CPlusPlus)
974            .Case("has_trivial_destructor", LangOpts.CPlusPlus)
975            .Case("has_virtual_destructor", LangOpts.CPlusPlus)
976            .Case("is_abstract", LangOpts.CPlusPlus)
977            .Case("is_base_of", LangOpts.CPlusPlus)
978            .Case("is_class", LangOpts.CPlusPlus)
979            .Case("is_constructible", LangOpts.CPlusPlus)
980            .Case("is_convertible_to", LangOpts.CPlusPlus)
981            .Case("is_empty", LangOpts.CPlusPlus)
982            .Case("is_enum", LangOpts.CPlusPlus)
983            .Case("is_final", LangOpts.CPlusPlus)
984            .Case("is_literal", LangOpts.CPlusPlus)
985            .Case("is_standard_layout", LangOpts.CPlusPlus)
986            .Case("is_pod", LangOpts.CPlusPlus)
987            .Case("is_polymorphic", LangOpts.CPlusPlus)
988            .Case("is_sealed", LangOpts.MicrosoftExt)
989            .Case("is_trivial", LangOpts.CPlusPlus)
990            .Case("is_trivially_assignable", LangOpts.CPlusPlus)
991            .Case("is_trivially_constructible", LangOpts.CPlusPlus)
992            .Case("is_trivially_copyable", LangOpts.CPlusPlus)
993            .Case("is_union", LangOpts.CPlusPlus)
994            .Case("modules", LangOpts.Modules)
995            .Case("tls", PP.getTargetInfo().isTLSSupported())
996            .Case("underlying_type", LangOpts.CPlusPlus)
997            .Default(false);
998 }
999 
1000 /// HasExtension - Return true if we recognize and implement the feature
1001 /// specified by the identifier, either as an extension or a standard language
1002 /// feature.
HasExtension(const Preprocessor & PP,const IdentifierInfo * II)1003 static bool HasExtension(const Preprocessor &PP, const IdentifierInfo *II) {
1004   if (HasFeature(PP, II))
1005     return true;
1006 
1007   // If the use of an extension results in an error diagnostic, extensions are
1008   // effectively unavailable, so just return false here.
1009   if (PP.getDiagnostics().getExtensionHandlingBehavior() >=
1010       diag::Severity::Error)
1011     return false;
1012 
1013   const LangOptions &LangOpts = PP.getLangOpts();
1014   StringRef Extension = II->getName();
1015 
1016   // Normalize the extension name, __foo__ becomes foo.
1017   if (Extension.startswith("__") && Extension.endswith("__") &&
1018       Extension.size() >= 4)
1019     Extension = Extension.substr(2, Extension.size() - 4);
1020 
1021   // Because we inherit the feature list from HasFeature, this string switch
1022   // must be less restrictive than HasFeature's.
1023   return llvm::StringSwitch<bool>(Extension)
1024            // C11 features supported by other languages as extensions.
1025            .Case("c_alignas", true)
1026            .Case("c_atomic", true)
1027            .Case("c_generic_selections", true)
1028            .Case("c_static_assert", true)
1029            .Case("c_thread_local", PP.getTargetInfo().isTLSSupported())
1030            // C++11 features supported by other languages as extensions.
1031            .Case("cxx_atomic", LangOpts.CPlusPlus)
1032            .Case("cxx_deleted_functions", LangOpts.CPlusPlus)
1033            .Case("cxx_explicit_conversions", LangOpts.CPlusPlus)
1034            .Case("cxx_inline_namespaces", LangOpts.CPlusPlus)
1035            .Case("cxx_local_type_template_args", LangOpts.CPlusPlus)
1036            .Case("cxx_nonstatic_member_init", LangOpts.CPlusPlus)
1037            .Case("cxx_override_control", LangOpts.CPlusPlus)
1038            .Case("cxx_range_for", LangOpts.CPlusPlus)
1039            .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus)
1040            .Case("cxx_rvalue_references", LangOpts.CPlusPlus)
1041            // C++1y features supported by other languages as extensions.
1042            .Case("cxx_binary_literals", true)
1043            .Case("cxx_init_captures", LangOpts.CPlusPlus11)
1044            .Case("cxx_variable_templates", LangOpts.CPlusPlus)
1045            .Default(false);
1046 }
1047 
1048 /// EvaluateHasIncludeCommon - Process a '__has_include("path")'
1049 /// or '__has_include_next("path")' expression.
1050 /// Returns true if successful.
EvaluateHasIncludeCommon(Token & Tok,IdentifierInfo * II,Preprocessor & PP,const DirectoryLookup * LookupFrom)1051 static bool EvaluateHasIncludeCommon(Token &Tok,
1052                                      IdentifierInfo *II, Preprocessor &PP,
1053                                      const DirectoryLookup *LookupFrom) {
1054   // Save the location of the current token.  If a '(' is later found, use
1055   // that location.  If not, use the end of this location instead.
1056   SourceLocation LParenLoc = Tok.getLocation();
1057 
1058   // These expressions are only allowed within a preprocessor directive.
1059   if (!PP.isParsingIfOrElifDirective()) {
1060     PP.Diag(LParenLoc, diag::err_pp_directive_required) << II->getName();
1061     return false;
1062   }
1063 
1064   // Get '('.
1065   PP.LexNonComment(Tok);
1066 
1067   // Ensure we have a '('.
1068   if (Tok.isNot(tok::l_paren)) {
1069     // No '(', use end of last token.
1070     LParenLoc = PP.getLocForEndOfToken(LParenLoc);
1071     PP.Diag(LParenLoc, diag::err_pp_expected_after) << II << tok::l_paren;
1072     // If the next token looks like a filename or the start of one,
1073     // assume it is and process it as such.
1074     if (!Tok.is(tok::angle_string_literal) && !Tok.is(tok::string_literal) &&
1075         !Tok.is(tok::less))
1076       return false;
1077   } else {
1078     // Save '(' location for possible missing ')' message.
1079     LParenLoc = Tok.getLocation();
1080 
1081     if (PP.getCurrentLexer()) {
1082       // Get the file name.
1083       PP.getCurrentLexer()->LexIncludeFilename(Tok);
1084     } else {
1085       // We're in a macro, so we can't use LexIncludeFilename; just
1086       // grab the next token.
1087       PP.Lex(Tok);
1088     }
1089   }
1090 
1091   // Reserve a buffer to get the spelling.
1092   SmallString<128> FilenameBuffer;
1093   StringRef Filename;
1094   SourceLocation EndLoc;
1095 
1096   switch (Tok.getKind()) {
1097   case tok::eod:
1098     // If the token kind is EOD, the error has already been diagnosed.
1099     return false;
1100 
1101   case tok::angle_string_literal:
1102   case tok::string_literal: {
1103     bool Invalid = false;
1104     Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid);
1105     if (Invalid)
1106       return false;
1107     break;
1108   }
1109 
1110   case tok::less:
1111     // This could be a <foo/bar.h> file coming from a macro expansion.  In this
1112     // case, glue the tokens together into FilenameBuffer and interpret those.
1113     FilenameBuffer.push_back('<');
1114     if (PP.ConcatenateIncludeName(FilenameBuffer, EndLoc)) {
1115       // Let the caller know a <eod> was found by changing the Token kind.
1116       Tok.setKind(tok::eod);
1117       return false;   // Found <eod> but no ">"?  Diagnostic already emitted.
1118     }
1119     Filename = FilenameBuffer.str();
1120     break;
1121   default:
1122     PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename);
1123     return false;
1124   }
1125 
1126   SourceLocation FilenameLoc = Tok.getLocation();
1127 
1128   // Get ')'.
1129   PP.LexNonComment(Tok);
1130 
1131   // Ensure we have a trailing ).
1132   if (Tok.isNot(tok::r_paren)) {
1133     PP.Diag(PP.getLocForEndOfToken(FilenameLoc), diag::err_pp_expected_after)
1134         << II << tok::r_paren;
1135     PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1136     return false;
1137   }
1138 
1139   bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename);
1140   // If GetIncludeFilenameSpelling set the start ptr to null, there was an
1141   // error.
1142   if (Filename.empty())
1143     return false;
1144 
1145   // Search include directories.
1146   const DirectoryLookup *CurDir;
1147   const FileEntry *File =
1148       PP.LookupFile(FilenameLoc, Filename, isAngled, LookupFrom, CurDir,
1149                     nullptr, nullptr, nullptr);
1150 
1151   // Get the result value.  A result of true means the file exists.
1152   return File != nullptr;
1153 }
1154 
1155 /// EvaluateHasInclude - Process a '__has_include("path")' expression.
1156 /// Returns true if successful.
EvaluateHasInclude(Token & Tok,IdentifierInfo * II,Preprocessor & PP)1157 static bool EvaluateHasInclude(Token &Tok, IdentifierInfo *II,
1158                                Preprocessor &PP) {
1159   return EvaluateHasIncludeCommon(Tok, II, PP, nullptr);
1160 }
1161 
1162 /// EvaluateHasIncludeNext - Process '__has_include_next("path")' expression.
1163 /// Returns true if successful.
EvaluateHasIncludeNext(Token & Tok,IdentifierInfo * II,Preprocessor & PP)1164 static bool EvaluateHasIncludeNext(Token &Tok,
1165                                    IdentifierInfo *II, Preprocessor &PP) {
1166   // __has_include_next is like __has_include, except that we start
1167   // searching after the current found directory.  If we can't do this,
1168   // issue a diagnostic.
1169   const DirectoryLookup *Lookup = PP.GetCurDirLookup();
1170   if (PP.isInPrimaryFile()) {
1171     Lookup = nullptr;
1172     PP.Diag(Tok, diag::pp_include_next_in_primary);
1173   } else if (!Lookup) {
1174     PP.Diag(Tok, diag::pp_include_next_absolute_path);
1175   } else {
1176     // Start looking up in the next directory.
1177     ++Lookup;
1178   }
1179 
1180   return EvaluateHasIncludeCommon(Tok, II, PP, Lookup);
1181 }
1182 
1183 /// \brief Process __building_module(identifier) expression.
1184 /// \returns true if we are building the named module, false otherwise.
EvaluateBuildingModule(Token & Tok,IdentifierInfo * II,Preprocessor & PP)1185 static bool EvaluateBuildingModule(Token &Tok,
1186                                    IdentifierInfo *II, Preprocessor &PP) {
1187   // Get '('.
1188   PP.LexNonComment(Tok);
1189 
1190   // Ensure we have a '('.
1191   if (Tok.isNot(tok::l_paren)) {
1192     PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II
1193                                                             << tok::l_paren;
1194     return false;
1195   }
1196 
1197   // Save '(' location for possible missing ')' message.
1198   SourceLocation LParenLoc = Tok.getLocation();
1199 
1200   // Get the module name.
1201   PP.LexNonComment(Tok);
1202 
1203   // Ensure that we have an identifier.
1204   if (Tok.isNot(tok::identifier)) {
1205     PP.Diag(Tok.getLocation(), diag::err_expected_id_building_module);
1206     return false;
1207   }
1208 
1209   bool Result
1210     = Tok.getIdentifierInfo()->getName() == PP.getLangOpts().CurrentModule;
1211 
1212   // Get ')'.
1213   PP.LexNonComment(Tok);
1214 
1215   // Ensure we have a trailing ).
1216   if (Tok.isNot(tok::r_paren)) {
1217     PP.Diag(Tok.getLocation(), diag::err_pp_expected_after) << II
1218                                                             << tok::r_paren;
1219     PP.Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1220     return false;
1221   }
1222 
1223   return Result;
1224 }
1225 
1226 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded
1227 /// as a builtin macro, handle it and return the next token as 'Tok'.
ExpandBuiltinMacro(Token & Tok)1228 void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
1229   // Figure out which token this is.
1230   IdentifierInfo *II = Tok.getIdentifierInfo();
1231   assert(II && "Can't be a macro without id info!");
1232 
1233   // If this is an _Pragma or Microsoft __pragma directive, expand it,
1234   // invoke the pragma handler, then lex the token after it.
1235   if (II == Ident_Pragma)
1236     return Handle_Pragma(Tok);
1237   else if (II == Ident__pragma) // in non-MS mode this is null
1238     return HandleMicrosoft__pragma(Tok);
1239 
1240   ++NumBuiltinMacroExpanded;
1241 
1242   SmallString<128> TmpBuffer;
1243   llvm::raw_svector_ostream OS(TmpBuffer);
1244 
1245   // Set up the return result.
1246   Tok.setIdentifierInfo(nullptr);
1247   Tok.clearFlag(Token::NeedsCleaning);
1248 
1249   if (II == Ident__LINE__) {
1250     // C99 6.10.8: "__LINE__: The presumed line number (within the current
1251     // source file) of the current source line (an integer constant)".  This can
1252     // be affected by #line.
1253     SourceLocation Loc = Tok.getLocation();
1254 
1255     // Advance to the location of the first _, this might not be the first byte
1256     // of the token if it starts with an escaped newline.
1257     Loc = AdvanceToTokenCharacter(Loc, 0);
1258 
1259     // One wrinkle here is that GCC expands __LINE__ to location of the *end* of
1260     // a macro expansion.  This doesn't matter for object-like macros, but
1261     // can matter for a function-like macro that expands to contain __LINE__.
1262     // Skip down through expansion points until we find a file loc for the
1263     // end of the expansion history.
1264     Loc = SourceMgr.getExpansionRange(Loc).second;
1265     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Loc);
1266 
1267     // __LINE__ expands to a simple numeric value.
1268     OS << (PLoc.isValid()? PLoc.getLine() : 1);
1269     Tok.setKind(tok::numeric_constant);
1270   } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__) {
1271     // C99 6.10.8: "__FILE__: The presumed name of the current source file (a
1272     // character string literal)". This can be affected by #line.
1273     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1274 
1275     // __BASE_FILE__ is a GNU extension that returns the top of the presumed
1276     // #include stack instead of the current file.
1277     if (II == Ident__BASE_FILE__ && PLoc.isValid()) {
1278       SourceLocation NextLoc = PLoc.getIncludeLoc();
1279       while (NextLoc.isValid()) {
1280         PLoc = SourceMgr.getPresumedLoc(NextLoc);
1281         if (PLoc.isInvalid())
1282           break;
1283 
1284         NextLoc = PLoc.getIncludeLoc();
1285       }
1286     }
1287 
1288     // Escape this filename.  Turn '\' -> '\\' '"' -> '\"'
1289     SmallString<128> FN;
1290     if (PLoc.isValid()) {
1291       FN += PLoc.getFilename();
1292       Lexer::Stringify(FN);
1293       OS << '"' << FN.str() << '"';
1294     }
1295     Tok.setKind(tok::string_literal);
1296   } else if (II == Ident__DATE__) {
1297     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1298     if (!DATELoc.isValid())
1299       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1300     Tok.setKind(tok::string_literal);
1301     Tok.setLength(strlen("\"Mmm dd yyyy\""));
1302     Tok.setLocation(SourceMgr.createExpansionLoc(DATELoc, Tok.getLocation(),
1303                                                  Tok.getLocation(),
1304                                                  Tok.getLength()));
1305     return;
1306   } else if (II == Ident__TIME__) {
1307     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1308     if (!TIMELoc.isValid())
1309       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
1310     Tok.setKind(tok::string_literal);
1311     Tok.setLength(strlen("\"hh:mm:ss\""));
1312     Tok.setLocation(SourceMgr.createExpansionLoc(TIMELoc, Tok.getLocation(),
1313                                                  Tok.getLocation(),
1314                                                  Tok.getLength()));
1315     return;
1316   } else if (II == Ident__INCLUDE_LEVEL__) {
1317     // Compute the presumed include depth of this token.  This can be affected
1318     // by GNU line markers.
1319     unsigned Depth = 0;
1320 
1321     PresumedLoc PLoc = SourceMgr.getPresumedLoc(Tok.getLocation());
1322     if (PLoc.isValid()) {
1323       PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1324       for (; PLoc.isValid(); ++Depth)
1325         PLoc = SourceMgr.getPresumedLoc(PLoc.getIncludeLoc());
1326     }
1327 
1328     // __INCLUDE_LEVEL__ expands to a simple numeric value.
1329     OS << Depth;
1330     Tok.setKind(tok::numeric_constant);
1331   } else if (II == Ident__TIMESTAMP__) {
1332     Diag(Tok.getLocation(), diag::warn_pp_date_time);
1333     // MSVC, ICC, GCC, VisualAge C++ extension.  The generated string should be
1334     // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime.
1335 
1336     // Get the file that we are lexing out of.  If we're currently lexing from
1337     // a macro, dig into the include stack.
1338     const FileEntry *CurFile = nullptr;
1339     PreprocessorLexer *TheLexer = getCurrentFileLexer();
1340 
1341     if (TheLexer)
1342       CurFile = SourceMgr.getFileEntryForID(TheLexer->getFileID());
1343 
1344     const char *Result;
1345     if (CurFile) {
1346       time_t TT = CurFile->getModificationTime();
1347       struct tm *TM = localtime(&TT);
1348       Result = asctime(TM);
1349     } else {
1350       Result = "??? ??? ?? ??:??:?? ????\n";
1351     }
1352     // Surround the string with " and strip the trailing newline.
1353     OS << '"' << StringRef(Result).drop_back() << '"';
1354     Tok.setKind(tok::string_literal);
1355   } else if (II == Ident__COUNTER__) {
1356     // __COUNTER__ expands to a simple numeric value.
1357     OS << CounterValue++;
1358     Tok.setKind(tok::numeric_constant);
1359   } else if (II == Ident__has_feature   ||
1360              II == Ident__has_extension ||
1361              II == Ident__has_builtin   ||
1362              II == Ident__is_identifier ||
1363              II == Ident__has_attribute) {
1364     // The argument to these builtins should be a parenthesized identifier.
1365     SourceLocation StartLoc = Tok.getLocation();
1366 
1367     bool IsValid = false;
1368     IdentifierInfo *FeatureII = nullptr;
1369 
1370     // Read the '('.
1371     LexUnexpandedToken(Tok);
1372     if (Tok.is(tok::l_paren)) {
1373       // Read the identifier
1374       LexUnexpandedToken(Tok);
1375       if ((FeatureII = Tok.getIdentifierInfo())) {
1376         // Read the ')'.
1377         LexUnexpandedToken(Tok);
1378         if (Tok.is(tok::r_paren))
1379           IsValid = true;
1380       }
1381     }
1382 
1383     bool Value = false;
1384     if (!IsValid)
1385       Diag(StartLoc, diag::err_feature_check_malformed);
1386     else if (II == Ident__is_identifier)
1387       Value = FeatureII->getTokenID() == tok::identifier;
1388     else if (II == Ident__has_builtin) {
1389       // Check for a builtin is trivial.
1390       Value = FeatureII->getBuiltinID() != 0;
1391     } else if (II == Ident__has_attribute)
1392       Value = hasAttribute(AttrSyntax::Generic, nullptr, FeatureII,
1393                            getTargetInfo().getTriple(), getLangOpts());
1394     else if (II == Ident__has_extension)
1395       Value = HasExtension(*this, FeatureII);
1396     else {
1397       assert(II == Ident__has_feature && "Must be feature check");
1398       Value = HasFeature(*this, FeatureII);
1399     }
1400 
1401     OS << (int)Value;
1402     if (IsValid)
1403       Tok.setKind(tok::numeric_constant);
1404   } else if (II == Ident__has_include ||
1405              II == Ident__has_include_next) {
1406     // The argument to these two builtins should be a parenthesized
1407     // file name string literal using angle brackets (<>) or
1408     // double-quotes ("").
1409     bool Value;
1410     if (II == Ident__has_include)
1411       Value = EvaluateHasInclude(Tok, II, *this);
1412     else
1413       Value = EvaluateHasIncludeNext(Tok, II, *this);
1414     OS << (int)Value;
1415     if (Tok.is(tok::r_paren))
1416       Tok.setKind(tok::numeric_constant);
1417   } else if (II == Ident__has_warning) {
1418     // The argument should be a parenthesized string literal.
1419     // The argument to these builtins should be a parenthesized identifier.
1420     SourceLocation StartLoc = Tok.getLocation();
1421     bool IsValid = false;
1422     bool Value = false;
1423     // Read the '('.
1424     LexUnexpandedToken(Tok);
1425     do {
1426       if (Tok.isNot(tok::l_paren)) {
1427         Diag(StartLoc, diag::err_warning_check_malformed);
1428         break;
1429       }
1430 
1431       LexUnexpandedToken(Tok);
1432       std::string WarningName;
1433       SourceLocation StrStartLoc = Tok.getLocation();
1434       if (!FinishLexStringLiteral(Tok, WarningName, "'__has_warning'",
1435                                   /*MacroExpansion=*/false)) {
1436         // Eat tokens until ')'.
1437         while (Tok.isNot(tok::r_paren) && Tok.isNot(tok::eod) &&
1438                Tok.isNot(tok::eof))
1439           LexUnexpandedToken(Tok);
1440         break;
1441       }
1442 
1443       // Is the end a ')'?
1444       if (!(IsValid = Tok.is(tok::r_paren))) {
1445         Diag(StartLoc, diag::err_warning_check_malformed);
1446         break;
1447       }
1448 
1449       if (WarningName.size() < 3 || WarningName[0] != '-' ||
1450           WarningName[1] != 'W') {
1451         Diag(StrStartLoc, diag::warn_has_warning_invalid_option);
1452         break;
1453       }
1454 
1455       // Finally, check if the warning flags maps to a diagnostic group.
1456       // We construct a SmallVector here to talk to getDiagnosticIDs().
1457       // Although we don't use the result, this isn't a hot path, and not
1458       // worth special casing.
1459       SmallVector<diag::kind, 10> Diags;
1460       Value = !getDiagnostics().getDiagnosticIDs()->
1461         getDiagnosticsInGroup(WarningName.substr(2), Diags);
1462     } while (false);
1463 
1464     OS << (int)Value;
1465     if (IsValid)
1466       Tok.setKind(tok::numeric_constant);
1467   } else if (II == Ident__building_module) {
1468     // The argument to this builtin should be an identifier. The
1469     // builtin evaluates to 1 when that identifier names the module we are
1470     // currently building.
1471     OS << (int)EvaluateBuildingModule(Tok, II, *this);
1472     Tok.setKind(tok::numeric_constant);
1473   } else if (II == Ident__MODULE__) {
1474     // The current module as an identifier.
1475     OS << getLangOpts().CurrentModule;
1476     IdentifierInfo *ModuleII = getIdentifierInfo(getLangOpts().CurrentModule);
1477     Tok.setIdentifierInfo(ModuleII);
1478     Tok.setKind(ModuleII->getTokenID());
1479   } else if (II == Ident__identifier) {
1480     SourceLocation Loc = Tok.getLocation();
1481 
1482     // We're expecting '__identifier' '(' identifier ')'. Try to recover
1483     // if the parens are missing.
1484     LexNonComment(Tok);
1485     if (Tok.isNot(tok::l_paren)) {
1486       // No '(', use end of last token.
1487       Diag(getLocForEndOfToken(Loc), diag::err_pp_expected_after)
1488         << II << tok::l_paren;
1489       // If the next token isn't valid as our argument, we can't recover.
1490       if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1491         Tok.setKind(tok::identifier);
1492       return;
1493     }
1494 
1495     SourceLocation LParenLoc = Tok.getLocation();
1496     LexNonComment(Tok);
1497 
1498     if (!Tok.isAnnotation() && Tok.getIdentifierInfo())
1499       Tok.setKind(tok::identifier);
1500     else {
1501       Diag(Tok.getLocation(), diag::err_pp_identifier_arg_not_identifier)
1502         << Tok.getKind();
1503       // Don't walk past anything that's not a real token.
1504       if (Tok.is(tok::eof) || Tok.is(tok::eod) || Tok.isAnnotation())
1505         return;
1506     }
1507 
1508     // Discard the ')', preserving 'Tok' as our result.
1509     Token RParen;
1510     LexNonComment(RParen);
1511     if (RParen.isNot(tok::r_paren)) {
1512       Diag(getLocForEndOfToken(Tok.getLocation()), diag::err_pp_expected_after)
1513         << Tok.getKind() << tok::r_paren;
1514       Diag(LParenLoc, diag::note_matching) << tok::l_paren;
1515     }
1516     return;
1517   } else {
1518     llvm_unreachable("Unknown identifier!");
1519   }
1520   CreateString(OS.str(), Tok, Tok.getLocation(), Tok.getLocation());
1521 }
1522 
markMacroAsUsed(MacroInfo * MI)1523 void Preprocessor::markMacroAsUsed(MacroInfo *MI) {
1524   // If the 'used' status changed, and the macro requires 'unused' warning,
1525   // remove its SourceLocation from the warn-for-unused-macro locations.
1526   if (MI->isWarnIfUnused() && !MI->isUsed())
1527     WarnUnusedMacroLocs.erase(MI->getDefinitionLoc());
1528   MI->setIsUsed(true);
1529 }
1530