• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===--- TokenLexer.cpp - Lex from a token stream -------------------------===//
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 TokenLexer interface.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "clang/Lex/TokenLexer.h"
15 #include "MacroArgs.h"
16 #include "clang/Lex/MacroInfo.h"
17 #include "clang/Lex/Preprocessor.h"
18 #include "clang/Basic/SourceManager.h"
19 #include "clang/Lex/LexDiagnostic.h"
20 #include "llvm/ADT/SmallVector.h"
21 using namespace clang;
22 
23 
24 /// Create a TokenLexer for the specified macro with the specified actual
25 /// arguments.  Note that this ctor takes ownership of the ActualArgs pointer.
Init(Token & Tok,SourceLocation ELEnd,MacroArgs * Actuals)26 void TokenLexer::Init(Token &Tok, SourceLocation ELEnd, MacroArgs *Actuals) {
27   // If the client is reusing a TokenLexer, make sure to free any memory
28   // associated with it.
29   destroy();
30 
31   Macro = PP.getMacroInfo(Tok.getIdentifierInfo());
32   ActualArgs = Actuals;
33   CurToken = 0;
34 
35   ExpandLocStart = Tok.getLocation();
36   ExpandLocEnd = ELEnd;
37   AtStartOfLine = Tok.isAtStartOfLine();
38   HasLeadingSpace = Tok.hasLeadingSpace();
39   Tokens = &*Macro->tokens_begin();
40   OwnsTokens = false;
41   DisableMacroExpansion = false;
42   NumTokens = Macro->tokens_end()-Macro->tokens_begin();
43   MacroExpansionStart = SourceLocation();
44 
45   SourceManager &SM = PP.getSourceManager();
46   MacroStartSLocOffset = SM.getNextLocalOffset();
47 
48   if (NumTokens > 0) {
49     assert(Tokens[0].getLocation().isValid());
50     assert((Tokens[0].getLocation().isFileID() || Tokens[0].is(tok::comment)) &&
51            "Macro defined in macro?");
52     assert(ExpandLocStart.isValid());
53 
54     // Reserve a source location entry chunk for the length of the macro
55     // definition. Tokens that get lexed directly from the definition will
56     // have their locations pointing inside this chunk. This is to avoid
57     // creating separate source location entries for each token.
58     SourceLocation macroStart = SM.getInstantiationLoc(Tokens[0].getLocation());
59     MacroDefStartInfo = SM.getDecomposedLoc(macroStart);
60     MacroExpansionStart = SM.createInstantiationLoc(macroStart,
61                                               ExpandLocStart,
62                                               ExpandLocEnd,
63                                               Macro->getDefinitionLength(SM));
64   }
65 
66   // If this is a function-like macro, expand the arguments and change
67   // Tokens to point to the expanded tokens.
68   if (Macro->isFunctionLike() && Macro->getNumArgs())
69     ExpandFunctionArguments();
70 
71   // Mark the macro as currently disabled, so that it is not recursively
72   // expanded.  The macro must be disabled only after argument pre-expansion of
73   // function-like macro arguments occurs.
74   Macro->DisableMacro();
75 }
76 
77 
78 
79 /// Create a TokenLexer for the specified token stream.  This does not
80 /// take ownership of the specified token vector.
Init(const Token * TokArray,unsigned NumToks,bool disableMacroExpansion,bool ownsTokens)81 void TokenLexer::Init(const Token *TokArray, unsigned NumToks,
82                       bool disableMacroExpansion, bool ownsTokens) {
83   // If the client is reusing a TokenLexer, make sure to free any memory
84   // associated with it.
85   destroy();
86 
87   Macro = 0;
88   ActualArgs = 0;
89   Tokens = TokArray;
90   OwnsTokens = ownsTokens;
91   DisableMacroExpansion = disableMacroExpansion;
92   NumTokens = NumToks;
93   CurToken = 0;
94   ExpandLocStart = ExpandLocEnd = SourceLocation();
95   AtStartOfLine = false;
96   HasLeadingSpace = false;
97   MacroExpansionStart = SourceLocation();
98 
99   // Set HasLeadingSpace/AtStartOfLine so that the first token will be
100   // returned unmodified.
101   if (NumToks != 0) {
102     AtStartOfLine   = TokArray[0].isAtStartOfLine();
103     HasLeadingSpace = TokArray[0].hasLeadingSpace();
104   }
105 }
106 
107 
destroy()108 void TokenLexer::destroy() {
109   // If this was a function-like macro that actually uses its arguments, delete
110   // the expanded tokens.
111   if (OwnsTokens) {
112     delete [] Tokens;
113     Tokens = 0;
114     OwnsTokens = false;
115   }
116 
117   // TokenLexer owns its formal arguments.
118   if (ActualArgs) ActualArgs->destroy(PP);
119 }
120 
121 /// Expand the arguments of a function-like macro so that we can quickly
122 /// return preexpanded tokens from Tokens.
ExpandFunctionArguments()123 void TokenLexer::ExpandFunctionArguments() {
124   SourceManager &SM = PP.getSourceManager();
125 
126   llvm::SmallVector<Token, 128> ResultToks;
127 
128   // Loop through 'Tokens', expanding them into ResultToks.  Keep
129   // track of whether we change anything.  If not, no need to keep them.  If so,
130   // we install the newly expanded sequence as the new 'Tokens' list.
131   bool MadeChange = false;
132 
133   // NextTokGetsSpace - When this is true, the next token appended to the
134   // output list will get a leading space, regardless of whether it had one to
135   // begin with or not.  This is used for placemarker support.
136   bool NextTokGetsSpace = false;
137 
138   for (unsigned i = 0, e = NumTokens; i != e; ++i) {
139     // If we found the stringify operator, get the argument stringified.  The
140     // preprocessor already verified that the following token is a macro name
141     // when the #define was parsed.
142     const Token &CurTok = Tokens[i];
143     if (CurTok.is(tok::hash) || CurTok.is(tok::hashat)) {
144       int ArgNo = Macro->getArgumentNum(Tokens[i+1].getIdentifierInfo());
145       assert(ArgNo != -1 && "Token following # is not an argument?");
146 
147       SourceLocation hashInstLoc;
148       if(ExpandLocStart.isValid()) {
149         hashInstLoc = getMacroExpansionLocation(CurTok.getLocation());
150         assert(hashInstLoc.isValid() && "Expected '#' to come from definition");
151       }
152 
153       Token Res;
154       if (CurTok.is(tok::hash))  // Stringify
155         Res = ActualArgs->getStringifiedArgument(ArgNo, PP, hashInstLoc);
156       else {
157         // 'charify': don't bother caching these.
158         Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo),
159                                            PP, true, hashInstLoc);
160       }
161 
162       // The stringified/charified string leading space flag gets set to match
163       // the #/#@ operator.
164       if (CurTok.hasLeadingSpace() || NextTokGetsSpace)
165         Res.setFlag(Token::LeadingSpace);
166 
167       ResultToks.push_back(Res);
168       MadeChange = true;
169       ++i;  // Skip arg name.
170       NextTokGetsSpace = false;
171       continue;
172     }
173 
174     // Otherwise, if this is not an argument token, just add the token to the
175     // output buffer.
176     IdentifierInfo *II = CurTok.getIdentifierInfo();
177     int ArgNo = II ? Macro->getArgumentNum(II) : -1;
178     if (ArgNo == -1) {
179       // This isn't an argument, just add it.
180       ResultToks.push_back(CurTok);
181 
182       if (NextTokGetsSpace) {
183         ResultToks.back().setFlag(Token::LeadingSpace);
184         NextTokGetsSpace = false;
185       }
186       continue;
187     }
188 
189     // An argument is expanded somehow, the result is different than the
190     // input.
191     MadeChange = true;
192 
193     // Otherwise, this is a use of the argument.  Find out if there is a paste
194     // (##) operator before or after the argument.
195     bool PasteBefore =
196       !ResultToks.empty() && ResultToks.back().is(tok::hashhash);
197     bool PasteAfter = i+1 != e && Tokens[i+1].is(tok::hashhash);
198 
199     // If it is not the LHS/RHS of a ## operator, we must pre-expand the
200     // argument and substitute the expanded tokens into the result.  This is
201     // C99 6.10.3.1p1.
202     if (!PasteBefore && !PasteAfter) {
203       const Token *ResultArgToks;
204 
205       // Only preexpand the argument if it could possibly need it.  This
206       // avoids some work in common cases.
207       const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo);
208       if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP))
209         ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, Macro, PP)[0];
210       else
211         ResultArgToks = ArgTok;  // Use non-preexpanded tokens.
212 
213       // If the arg token expanded into anything, append it.
214       if (ResultArgToks->isNot(tok::eof)) {
215         unsigned FirstResult = ResultToks.size();
216         unsigned NumToks = MacroArgs::getArgLength(ResultArgToks);
217         ResultToks.append(ResultArgToks, ResultArgToks+NumToks);
218 
219         // If the '##' came from expanding an argument, turn it into 'unknown'
220         // to avoid pasting.
221         for (unsigned i = FirstResult, e = ResultToks.size(); i != e; ++i) {
222           Token &Tok = ResultToks[i];
223           if (Tok.is(tok::hashhash))
224             Tok.setKind(tok::unknown);
225         }
226 
227         if(ExpandLocStart.isValid()) {
228           SourceLocation curInst =
229               getMacroExpansionLocation(CurTok.getLocation());
230           assert(curInst.isValid() &&
231                  "Expected arg identifier to come from definition");
232           for (unsigned i = FirstResult, e = ResultToks.size(); i != e; ++i) {
233             Token &Tok = ResultToks[i];
234             Tok.setLocation(SM.createMacroArgInstantiationLoc(Tok.getLocation(),
235                                                               curInst,
236                                                               Tok.getLength()));
237           }
238         }
239 
240         // If any tokens were substituted from the argument, the whitespace
241         // before the first token should match the whitespace of the arg
242         // identifier.
243         ResultToks[FirstResult].setFlagValue(Token::LeadingSpace,
244                                              CurTok.hasLeadingSpace() ||
245                                              NextTokGetsSpace);
246         NextTokGetsSpace = false;
247       } else {
248         // If this is an empty argument, and if there was whitespace before the
249         // formal token, make sure the next token gets whitespace before it.
250         NextTokGetsSpace = CurTok.hasLeadingSpace();
251       }
252       continue;
253     }
254 
255     // Okay, we have a token that is either the LHS or RHS of a paste (##)
256     // argument.  It gets substituted as its non-pre-expanded tokens.
257     const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo);
258     unsigned NumToks = MacroArgs::getArgLength(ArgToks);
259     if (NumToks) {  // Not an empty argument?
260       // If this is the GNU ", ## __VA_ARG__" extension, and we just learned
261       // that __VA_ARG__ expands to multiple tokens, avoid a pasting error when
262       // the expander trys to paste ',' with the first token of the __VA_ARG__
263       // expansion.
264       if (PasteBefore && ResultToks.size() >= 2 &&
265           ResultToks[ResultToks.size()-2].is(tok::comma) &&
266           (unsigned)ArgNo == Macro->getNumArgs()-1 &&
267           Macro->isVariadic()) {
268         // Remove the paste operator, report use of the extension.
269         PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma);
270         ResultToks.pop_back();
271       }
272 
273       ResultToks.append(ArgToks, ArgToks+NumToks);
274 
275       // If the '##' came from expanding an argument, turn it into 'unknown'
276       // to avoid pasting.
277       for (unsigned i = ResultToks.size() - NumToks, e = ResultToks.size();
278              i != e; ++i) {
279         Token &Tok = ResultToks[i];
280         if (Tok.is(tok::hashhash))
281           Tok.setKind(tok::unknown);
282       }
283 
284       if (ExpandLocStart.isValid()) {
285         SourceLocation curInst =
286             getMacroExpansionLocation(CurTok.getLocation());
287         assert(curInst.isValid() &&
288                "Expected arg identifier to come from definition");
289         for (unsigned i = ResultToks.size() - NumToks, e = ResultToks.size();
290                i != e; ++i) {
291           Token &Tok = ResultToks[i];
292           Tok.setLocation(SM.createMacroArgInstantiationLoc(Tok.getLocation(),
293                                                             curInst,
294                                                             Tok.getLength()));
295         }
296       }
297 
298       // If this token (the macro argument) was supposed to get leading
299       // whitespace, transfer this information onto the first token of the
300       // expansion.
301       //
302       // Do not do this if the paste operator occurs before the macro argument,
303       // as in "A ## MACROARG".  In valid code, the first token will get
304       // smooshed onto the preceding one anyway (forming AMACROARG).  In
305       // assembler-with-cpp mode, invalid pastes are allowed through: in this
306       // case, we do not want the extra whitespace to be added.  For example,
307       // we want ". ## foo" -> ".foo" not ". foo".
308       if ((CurTok.hasLeadingSpace() || NextTokGetsSpace) &&
309           !PasteBefore)
310         ResultToks[ResultToks.size()-NumToks].setFlag(Token::LeadingSpace);
311 
312       NextTokGetsSpace = false;
313       continue;
314     }
315 
316     // If an empty argument is on the LHS or RHS of a paste, the standard (C99
317     // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur.  We
318     // implement this by eating ## operators when a LHS or RHS expands to
319     // empty.
320     NextTokGetsSpace |= CurTok.hasLeadingSpace();
321     if (PasteAfter) {
322       // Discard the argument token and skip (don't copy to the expansion
323       // buffer) the paste operator after it.
324       NextTokGetsSpace |= Tokens[i+1].hasLeadingSpace();
325       ++i;
326       continue;
327     }
328 
329     // If this is on the RHS of a paste operator, we've already copied the
330     // paste operator to the ResultToks list.  Remove it.
331     assert(PasteBefore && ResultToks.back().is(tok::hashhash));
332     NextTokGetsSpace |= ResultToks.back().hasLeadingSpace();
333     ResultToks.pop_back();
334 
335     // If this is the __VA_ARGS__ token, and if the argument wasn't provided,
336     // and if the macro had at least one real argument, and if the token before
337     // the ## was a comma, remove the comma.
338     if ((unsigned)ArgNo == Macro->getNumArgs()-1 && // is __VA_ARGS__
339         ActualArgs->isVarargsElidedUse() &&       // Argument elided.
340         !ResultToks.empty() && ResultToks.back().is(tok::comma)) {
341       // Never add a space, even if the comma, ##, or arg had a space.
342       NextTokGetsSpace = false;
343       // Remove the paste operator, report use of the extension.
344       PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma);
345       ResultToks.pop_back();
346 
347       // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"),
348       // then removal of the comma should produce a placemarker token (in C99
349       // terms) which we model by popping off the previous ##, giving us a plain
350       // "X" when __VA_ARGS__ is empty.
351       if (!ResultToks.empty() && ResultToks.back().is(tok::hashhash))
352         ResultToks.pop_back();
353     }
354     continue;
355   }
356 
357   // If anything changed, install this as the new Tokens list.
358   if (MadeChange) {
359     assert(!OwnsTokens && "This would leak if we already own the token list");
360     // This is deleted in the dtor.
361     NumTokens = ResultToks.size();
362     // The tokens will be added to Preprocessor's cache and will be removed
363     // when this TokenLexer finishes lexing them.
364     Tokens = PP.cacheMacroExpandedTokens(this, ResultToks);
365 
366     // The preprocessor cache of macro expanded tokens owns these tokens,not us.
367     OwnsTokens = false;
368   }
369 }
370 
371 /// Lex - Lex and return a token from this macro stream.
372 ///
Lex(Token & Tok)373 void TokenLexer::Lex(Token &Tok) {
374   // Lexing off the end of the macro, pop this macro off the expansion stack.
375   if (isAtEnd()) {
376     // If this is a macro (not a token stream), mark the macro enabled now
377     // that it is no longer being expanded.
378     if (Macro) Macro->EnableMacro();
379 
380     // Pop this context off the preprocessors lexer stack and get the next
381     // token.  This will delete "this" so remember the PP instance var.
382     Preprocessor &PPCache = PP;
383     if (PP.HandleEndOfTokenLexer(Tok))
384       return;
385 
386     // HandleEndOfTokenLexer may not return a token.  If it doesn't, lex
387     // whatever is next.
388     return PPCache.Lex(Tok);
389   }
390 
391   SourceManager &SM = PP.getSourceManager();
392 
393   // If this is the first token of the expanded result, we inherit spacing
394   // properties later.
395   bool isFirstToken = CurToken == 0;
396 
397   // Get the next token to return.
398   Tok = Tokens[CurToken++];
399 
400   bool TokenIsFromPaste = false;
401 
402   // If this token is followed by a token paste (##) operator, paste the tokens!
403   // Note that ## is a normal token when not expanding a macro.
404   if (!isAtEnd() && Tokens[CurToken].is(tok::hashhash) && Macro) {
405     // When handling the microsoft /##/ extension, the final token is
406     // returned by PasteTokens, not the pasted token.
407     if (PasteTokens(Tok))
408       return;
409 
410     TokenIsFromPaste = true;
411   }
412 
413   // The token's current location indicate where the token was lexed from.  We
414   // need this information to compute the spelling of the token, but any
415   // diagnostics for the expanded token should appear as if they came from
416   // ExpansionLoc.  Pull this information together into a new SourceLocation
417   // that captures all of this.
418   if (ExpandLocStart.isValid() &&   // Don't do this for token streams.
419       // Check that the token's location was not already set properly.
420       SM.isBeforeInSourceLocationOffset(Tok.getLocation(),
421                                         MacroStartSLocOffset)) {
422     SourceLocation instLoc;
423     if (Tok.is(tok::comment)) {
424       instLoc = SM.createInstantiationLoc(Tok.getLocation(),
425                                           ExpandLocStart,
426                                           ExpandLocEnd,
427                                           Tok.getLength());
428     } else {
429       instLoc = getMacroExpansionLocation(Tok.getLocation());
430       assert(instLoc.isValid() &&
431              "Location for token not coming from definition was not set!");
432     }
433 
434     Tok.setLocation(instLoc);
435   }
436 
437   // If this is the first token, set the lexical properties of the token to
438   // match the lexical properties of the macro identifier.
439   if (isFirstToken) {
440     Tok.setFlagValue(Token::StartOfLine , AtStartOfLine);
441     Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace);
442   }
443 
444   // Handle recursive expansion!
445   if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != 0) {
446     // Change the kind of this identifier to the appropriate token kind, e.g.
447     // turning "for" into a keyword.
448     IdentifierInfo *II = Tok.getIdentifierInfo();
449     Tok.setKind(II->getTokenID());
450 
451     // If this identifier was poisoned and from a paste, emit an error.  This
452     // won't be handled by Preprocessor::HandleIdentifier because this is coming
453     // from a macro expansion.
454     if (II->isPoisoned() && TokenIsFromPaste) {
455       PP.HandlePoisonedIdentifier(Tok);
456     }
457 
458     if (!DisableMacroExpansion && II->isHandleIdentifierCase())
459       PP.HandleIdentifier(Tok);
460   }
461 
462   // Otherwise, return a normal token.
463 }
464 
465 /// PasteTokens - Tok is the LHS of a ## operator, and CurToken is the ##
466 /// operator.  Read the ## and RHS, and paste the LHS/RHS together.  If there
467 /// are more ## after it, chomp them iteratively.  Return the result as Tok.
468 /// If this returns true, the caller should immediately return the token.
PasteTokens(Token & Tok)469 bool TokenLexer::PasteTokens(Token &Tok) {
470   llvm::SmallString<128> Buffer;
471   const char *ResultTokStrPtr = 0;
472   SourceLocation PasteOpLoc;
473   do {
474     // Consume the ## operator.
475     PasteOpLoc = Tokens[CurToken].getLocation();
476     ++CurToken;
477     assert(!isAtEnd() && "No token on the RHS of a paste operator!");
478 
479     // Get the RHS token.
480     const Token &RHS = Tokens[CurToken];
481 
482     // Allocate space for the result token.  This is guaranteed to be enough for
483     // the two tokens.
484     Buffer.resize(Tok.getLength() + RHS.getLength());
485 
486     // Get the spelling of the LHS token in Buffer.
487     const char *BufPtr = &Buffer[0];
488     bool Invalid = false;
489     unsigned LHSLen = PP.getSpelling(Tok, BufPtr, &Invalid);
490     if (BufPtr != &Buffer[0])   // Really, we want the chars in Buffer!
491       memcpy(&Buffer[0], BufPtr, LHSLen);
492     if (Invalid)
493       return true;
494 
495     BufPtr = &Buffer[LHSLen];
496     unsigned RHSLen = PP.getSpelling(RHS, BufPtr, &Invalid);
497     if (Invalid)
498       return true;
499     if (BufPtr != &Buffer[LHSLen])   // Really, we want the chars in Buffer!
500       memcpy(&Buffer[LHSLen], BufPtr, RHSLen);
501 
502     // Trim excess space.
503     Buffer.resize(LHSLen+RHSLen);
504 
505     // Plop the pasted result (including the trailing newline and null) into a
506     // scratch buffer where we can lex it.
507     Token ResultTokTmp;
508     ResultTokTmp.startToken();
509 
510     // Claim that the tmp token is a string_literal so that we can get the
511     // character pointer back from CreateString in getLiteralData().
512     ResultTokTmp.setKind(tok::string_literal);
513     PP.CreateString(&Buffer[0], Buffer.size(), ResultTokTmp);
514     SourceLocation ResultTokLoc = ResultTokTmp.getLocation();
515     ResultTokStrPtr = ResultTokTmp.getLiteralData();
516 
517     // Lex the resultant pasted token into Result.
518     Token Result;
519 
520     if (Tok.isAnyIdentifier() && RHS.isAnyIdentifier()) {
521       // Common paste case: identifier+identifier = identifier.  Avoid creating
522       // a lexer and other overhead.
523       PP.IncrementPasteCounter(true);
524       Result.startToken();
525       Result.setKind(tok::raw_identifier);
526       Result.setRawIdentifierData(ResultTokStrPtr);
527       Result.setLocation(ResultTokLoc);
528       Result.setLength(LHSLen+RHSLen);
529     } else {
530       PP.IncrementPasteCounter(false);
531 
532       assert(ResultTokLoc.isFileID() &&
533              "Should be a raw location into scratch buffer");
534       SourceManager &SourceMgr = PP.getSourceManager();
535       FileID LocFileID = SourceMgr.getFileID(ResultTokLoc);
536 
537       bool Invalid = false;
538       const char *ScratchBufStart
539         = SourceMgr.getBufferData(LocFileID, &Invalid).data();
540       if (Invalid)
541         return false;
542 
543       // Make a lexer to lex this string from.  Lex just this one token.
544       // Make a lexer object so that we lex and expand the paste result.
545       Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID),
546                PP.getLangOptions(), ScratchBufStart,
547                ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen);
548 
549       // Lex a token in raw mode.  This way it won't look up identifiers
550       // automatically, lexing off the end will return an eof token, and
551       // warnings are disabled.  This returns true if the result token is the
552       // entire buffer.
553       bool isInvalid = !TL.LexFromRawLexer(Result);
554 
555       // If we got an EOF token, we didn't form even ONE token.  For example, we
556       // did "/ ## /" to get "//".
557       isInvalid |= Result.is(tok::eof);
558 
559       // If pasting the two tokens didn't form a full new token, this is an
560       // error.  This occurs with "x ## +"  and other stuff.  Return with Tok
561       // unmodified and with RHS as the next token to lex.
562       if (isInvalid) {
563         // Test for the Microsoft extension of /##/ turning into // here on the
564         // error path.
565         if (PP.getLangOptions().Microsoft && Tok.is(tok::slash) &&
566             RHS.is(tok::slash)) {
567           HandleMicrosoftCommentPaste(Tok);
568           return true;
569         }
570 
571         // Do not emit the error when preprocessing assembler code.
572         if (!PP.getLangOptions().AsmPreprocessor) {
573           // Explicitly convert the token location to have proper expansion
574           // information so that the user knows where it came from.
575           SourceManager &SM = PP.getSourceManager();
576           SourceLocation Loc =
577             SM.createInstantiationLoc(PasteOpLoc, ExpandLocStart,
578                                       ExpandLocEnd, 2);
579           // If we're in microsoft extensions mode, downgrade this from a hard
580           // error to a warning that defaults to an error.  This allows
581           // disabling it.
582           PP.Diag(Loc,
583                   PP.getLangOptions().Microsoft ? diag::err_pp_bad_paste_ms
584                                                 : diag::err_pp_bad_paste)
585             << Buffer.str();
586         }
587 
588         // Do not consume the RHS.
589         --CurToken;
590       }
591 
592       // Turn ## into 'unknown' to avoid # ## # from looking like a paste
593       // operator.
594       if (Result.is(tok::hashhash))
595         Result.setKind(tok::unknown);
596     }
597 
598     // Transfer properties of the LHS over the the Result.
599     Result.setFlagValue(Token::StartOfLine , Tok.isAtStartOfLine());
600     Result.setFlagValue(Token::LeadingSpace, Tok.hasLeadingSpace());
601 
602     // Finally, replace LHS with the result, consume the RHS, and iterate.
603     ++CurToken;
604     Tok = Result;
605   } while (!isAtEnd() && Tokens[CurToken].is(tok::hashhash));
606 
607   // The token's current location indicate where the token was lexed from.  We
608   // need this information to compute the spelling of the token, but any
609   // diagnostics for the expanded token should appear as if the token was
610   // expanded from the (##) operator. Pull this information together into
611   // a new SourceLocation that captures all of this.
612   if (ExpandLocStart.isValid()) {
613     SourceManager &SM = PP.getSourceManager();
614     SourceLocation pasteLocInst =
615         getMacroExpansionLocation(PasteOpLoc);
616     assert(pasteLocInst.isValid() &&
617            "Expected '##' to come from definition");
618 
619     Tok.setLocation(SM.createInstantiationLoc(Tok.getLocation(),
620                                               pasteLocInst,
621                                               pasteLocInst,
622                                               Tok.getLength()));
623   }
624 
625   // Now that we got the result token, it will be subject to expansion.  Since
626   // token pasting re-lexes the result token in raw mode, identifier information
627   // isn't looked up.  As such, if the result is an identifier, look up id info.
628   if (Tok.is(tok::raw_identifier)) {
629     // Look up the identifier info for the token.  We disabled identifier lookup
630     // by saying we're skipping contents, so we need to do this manually.
631     PP.LookUpIdentifierInfo(Tok);
632   }
633   return false;
634 }
635 
636 /// isNextTokenLParen - If the next token lexed will pop this macro off the
637 /// expansion stack, return 2.  If the next unexpanded token is a '(', return
638 /// 1, otherwise return 0.
isNextTokenLParen() const639 unsigned TokenLexer::isNextTokenLParen() const {
640   // Out of tokens?
641   if (isAtEnd())
642     return 2;
643   return Tokens[CurToken].is(tok::l_paren);
644 }
645 
646 /// isParsingPreprocessorDirective - Return true if we are in the middle of a
647 /// preprocessor directive.
isParsingPreprocessorDirective() const648 bool TokenLexer::isParsingPreprocessorDirective() const {
649   return Tokens[NumTokens-1].is(tok::eod) && !isAtEnd();
650 }
651 
652 /// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes
653 /// together to form a comment that comments out everything in the current
654 /// macro, other active macros, and anything left on the current physical
655 /// source line of the expanded buffer.  Handle this by returning the
656 /// first token on the next line.
HandleMicrosoftCommentPaste(Token & Tok)657 void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok) {
658   // We 'comment out' the rest of this macro by just ignoring the rest of the
659   // tokens that have not been lexed yet, if any.
660 
661   // Since this must be a macro, mark the macro enabled now that it is no longer
662   // being expanded.
663   assert(Macro && "Token streams can't paste comments");
664   Macro->EnableMacro();
665 
666   PP.HandleMicrosoftCommentPaste(Tok);
667 }
668 
669 /// \brief If \arg loc is a FileID and points inside the current macro
670 /// definition, returns the appropriate source location pointing at the
671 /// macro expansion source location entry.
getMacroExpansionLocation(SourceLocation loc) const672 SourceLocation TokenLexer::getMacroExpansionLocation(SourceLocation loc) const {
673   assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() &&
674          "Not appropriate for token streams");
675   assert(loc.isValid());
676 
677   SourceManager &SM = PP.getSourceManager();
678   unsigned relativeOffset;
679   if (loc.isFileID() &&
680       SM.isInFileID(loc,
681                     MacroDefStartInfo.first, MacroDefStartInfo.second,
682                     Macro->getDefinitionLength(SM), &relativeOffset)) {
683     return MacroExpansionStart.getFileLocWithOffset(relativeOffset);
684   }
685 
686   return SourceLocation();
687 }
688