1 //===--- Lexer.cpp - C Language Family Lexer ------------------------------===//
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 Lexer and Token interfaces.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "clang/Lex/Lexer.h"
15 #include "UnicodeCharSets.h"
16 #include "clang/Basic/CharInfo.h"
17 #include "clang/Basic/SourceManager.h"
18 #include "clang/Lex/CodeCompletionHandler.h"
19 #include "clang/Lex/LexDiagnostic.h"
20 #include "clang/Lex/LiteralSupport.h"
21 #include "clang/Lex/Preprocessor.h"
22 #include "llvm/ADT/STLExtras.h"
23 #include "llvm/ADT/StringExtras.h"
24 #include "llvm/ADT/StringSwitch.h"
25 #include "llvm/Support/Compiler.h"
26 #include "llvm/Support/ConvertUTF.h"
27 #include "llvm/Support/MemoryBuffer.h"
28 #include <cstring>
29 using namespace clang;
30
31 //===----------------------------------------------------------------------===//
32 // Token Class Implementation
33 //===----------------------------------------------------------------------===//
34
35 /// isObjCAtKeyword - Return true if we have an ObjC keyword identifier.
isObjCAtKeyword(tok::ObjCKeywordKind objcKey) const36 bool Token::isObjCAtKeyword(tok::ObjCKeywordKind objcKey) const {
37 if (IdentifierInfo *II = getIdentifierInfo())
38 return II->getObjCKeywordID() == objcKey;
39 return false;
40 }
41
42 /// getObjCKeywordID - Return the ObjC keyword kind.
getObjCKeywordID() const43 tok::ObjCKeywordKind Token::getObjCKeywordID() const {
44 IdentifierInfo *specId = getIdentifierInfo();
45 return specId ? specId->getObjCKeywordID() : tok::objc_not_keyword;
46 }
47
48
49 //===----------------------------------------------------------------------===//
50 // Lexer Class Implementation
51 //===----------------------------------------------------------------------===//
52
anchor()53 void Lexer::anchor() { }
54
InitLexer(const char * BufStart,const char * BufPtr,const char * BufEnd)55 void Lexer::InitLexer(const char *BufStart, const char *BufPtr,
56 const char *BufEnd) {
57 BufferStart = BufStart;
58 BufferPtr = BufPtr;
59 BufferEnd = BufEnd;
60
61 assert(BufEnd[0] == 0 &&
62 "We assume that the input buffer has a null character at the end"
63 " to simplify lexing!");
64
65 // Check whether we have a BOM in the beginning of the buffer. If yes - act
66 // accordingly. Right now we support only UTF-8 with and without BOM, so, just
67 // skip the UTF-8 BOM if it's present.
68 if (BufferStart == BufferPtr) {
69 // Determine the size of the BOM.
70 StringRef Buf(BufferStart, BufferEnd - BufferStart);
71 size_t BOMLength = llvm::StringSwitch<size_t>(Buf)
72 .StartsWith("\xEF\xBB\xBF", 3) // UTF-8 BOM
73 .Default(0);
74
75 // Skip the BOM.
76 BufferPtr += BOMLength;
77 }
78
79 Is_PragmaLexer = false;
80 CurrentConflictMarkerState = CMK_None;
81
82 // Start of the file is a start of line.
83 IsAtStartOfLine = true;
84 IsAtPhysicalStartOfLine = true;
85
86 HasLeadingSpace = false;
87 HasLeadingEmptyMacro = false;
88
89 // We are not after parsing a #.
90 ParsingPreprocessorDirective = false;
91
92 // We are not after parsing #include.
93 ParsingFilename = false;
94
95 // We are not in raw mode. Raw mode disables diagnostics and interpretation
96 // of tokens (e.g. identifiers, thus disabling macro expansion). It is used
97 // to quickly lex the tokens of the buffer, e.g. when handling a "#if 0" block
98 // or otherwise skipping over tokens.
99 LexingRawMode = false;
100
101 // Default to not keeping comments.
102 ExtendedTokenMode = 0;
103 }
104
105 /// Lexer constructor - Create a new lexer object for the specified buffer
106 /// with the specified preprocessor managing the lexing process. This lexer
107 /// assumes that the associated file buffer and Preprocessor objects will
108 /// outlive it, so it doesn't take ownership of either of them.
Lexer(FileID FID,const llvm::MemoryBuffer * InputFile,Preprocessor & PP)109 Lexer::Lexer(FileID FID, const llvm::MemoryBuffer *InputFile, Preprocessor &PP)
110 : PreprocessorLexer(&PP, FID),
111 FileLoc(PP.getSourceManager().getLocForStartOfFile(FID)),
112 LangOpts(PP.getLangOpts()) {
113
114 InitLexer(InputFile->getBufferStart(), InputFile->getBufferStart(),
115 InputFile->getBufferEnd());
116
117 resetExtendedTokenMode();
118 }
119
resetExtendedTokenMode()120 void Lexer::resetExtendedTokenMode() {
121 assert(PP && "Cannot reset token mode without a preprocessor");
122 if (LangOpts.TraditionalCPP)
123 SetKeepWhitespaceMode(true);
124 else
125 SetCommentRetentionState(PP->getCommentRetentionState());
126 }
127
128 /// Lexer constructor - Create a new raw lexer object. This object is only
129 /// suitable for calls to 'LexFromRawLexer'. This lexer assumes that the text
130 /// range will outlive it, so it doesn't take ownership of it.
Lexer(SourceLocation fileloc,const LangOptions & langOpts,const char * BufStart,const char * BufPtr,const char * BufEnd)131 Lexer::Lexer(SourceLocation fileloc, const LangOptions &langOpts,
132 const char *BufStart, const char *BufPtr, const char *BufEnd)
133 : FileLoc(fileloc), LangOpts(langOpts) {
134
135 InitLexer(BufStart, BufPtr, BufEnd);
136
137 // We *are* in raw mode.
138 LexingRawMode = true;
139 }
140
141 /// Lexer constructor - Create a new raw lexer object. This object is only
142 /// suitable for calls to 'LexFromRawLexer'. This lexer assumes that the text
143 /// range will outlive it, so it doesn't take ownership of it.
Lexer(FileID FID,const llvm::MemoryBuffer * FromFile,const SourceManager & SM,const LangOptions & langOpts)144 Lexer::Lexer(FileID FID, const llvm::MemoryBuffer *FromFile,
145 const SourceManager &SM, const LangOptions &langOpts)
146 : FileLoc(SM.getLocForStartOfFile(FID)), LangOpts(langOpts) {
147
148 InitLexer(FromFile->getBufferStart(), FromFile->getBufferStart(),
149 FromFile->getBufferEnd());
150
151 // We *are* in raw mode.
152 LexingRawMode = true;
153 }
154
155 /// Create_PragmaLexer: Lexer constructor - Create a new lexer object for
156 /// _Pragma expansion. This has a variety of magic semantics that this method
157 /// sets up. It returns a new'd Lexer that must be delete'd when done.
158 ///
159 /// On entrance to this routine, TokStartLoc is a macro location which has a
160 /// spelling loc that indicates the bytes to be lexed for the token and an
161 /// expansion location that indicates where all lexed tokens should be
162 /// "expanded from".
163 ///
164 /// TODO: It would really be nice to make _Pragma just be a wrapper around a
165 /// normal lexer that remaps tokens as they fly by. This would require making
166 /// Preprocessor::Lex virtual. Given that, we could just dump in a magic lexer
167 /// interface that could handle this stuff. This would pull GetMappedTokenLoc
168 /// out of the critical path of the lexer!
169 ///
Create_PragmaLexer(SourceLocation SpellingLoc,SourceLocation ExpansionLocStart,SourceLocation ExpansionLocEnd,unsigned TokLen,Preprocessor & PP)170 Lexer *Lexer::Create_PragmaLexer(SourceLocation SpellingLoc,
171 SourceLocation ExpansionLocStart,
172 SourceLocation ExpansionLocEnd,
173 unsigned TokLen, Preprocessor &PP) {
174 SourceManager &SM = PP.getSourceManager();
175
176 // Create the lexer as if we were going to lex the file normally.
177 FileID SpellingFID = SM.getFileID(SpellingLoc);
178 const llvm::MemoryBuffer *InputFile = SM.getBuffer(SpellingFID);
179 Lexer *L = new Lexer(SpellingFID, InputFile, PP);
180
181 // Now that the lexer is created, change the start/end locations so that we
182 // just lex the subsection of the file that we want. This is lexing from a
183 // scratch buffer.
184 const char *StrData = SM.getCharacterData(SpellingLoc);
185
186 L->BufferPtr = StrData;
187 L->BufferEnd = StrData+TokLen;
188 assert(L->BufferEnd[0] == 0 && "Buffer is not nul terminated!");
189
190 // Set the SourceLocation with the remapping information. This ensures that
191 // GetMappedTokenLoc will remap the tokens as they are lexed.
192 L->FileLoc = SM.createExpansionLoc(SM.getLocForStartOfFile(SpellingFID),
193 ExpansionLocStart,
194 ExpansionLocEnd, TokLen);
195
196 // Ensure that the lexer thinks it is inside a directive, so that end \n will
197 // return an EOD token.
198 L->ParsingPreprocessorDirective = true;
199
200 // This lexer really is for _Pragma.
201 L->Is_PragmaLexer = true;
202 return L;
203 }
204
205
206 /// Stringify - Convert the specified string into a C string, with surrounding
207 /// ""'s, and with escaped \ and " characters.
Stringify(const std::string & Str,bool Charify)208 std::string Lexer::Stringify(const std::string &Str, bool Charify) {
209 std::string Result = Str;
210 char Quote = Charify ? '\'' : '"';
211 for (unsigned i = 0, e = Result.size(); i != e; ++i) {
212 if (Result[i] == '\\' || Result[i] == Quote) {
213 Result.insert(Result.begin()+i, '\\');
214 ++i; ++e;
215 }
216 }
217 return Result;
218 }
219
220 /// Stringify - Convert the specified string into a C string by escaping '\'
221 /// and " characters. This does not add surrounding ""'s to the string.
Stringify(SmallVectorImpl<char> & Str)222 void Lexer::Stringify(SmallVectorImpl<char> &Str) {
223 for (unsigned i = 0, e = Str.size(); i != e; ++i) {
224 if (Str[i] == '\\' || Str[i] == '"') {
225 Str.insert(Str.begin()+i, '\\');
226 ++i; ++e;
227 }
228 }
229 }
230
231 //===----------------------------------------------------------------------===//
232 // Token Spelling
233 //===----------------------------------------------------------------------===//
234
235 /// \brief Slow case of getSpelling. Extract the characters comprising the
236 /// spelling of this token from the provided input buffer.
getSpellingSlow(const Token & Tok,const char * BufPtr,const LangOptions & LangOpts,char * Spelling)237 static size_t getSpellingSlow(const Token &Tok, const char *BufPtr,
238 const LangOptions &LangOpts, char *Spelling) {
239 assert(Tok.needsCleaning() && "getSpellingSlow called on simple token");
240
241 size_t Length = 0;
242 const char *BufEnd = BufPtr + Tok.getLength();
243
244 if (Tok.is(tok::string_literal)) {
245 // Munch the encoding-prefix and opening double-quote.
246 while (BufPtr < BufEnd) {
247 unsigned Size;
248 Spelling[Length++] = Lexer::getCharAndSizeNoWarn(BufPtr, Size, LangOpts);
249 BufPtr += Size;
250
251 if (Spelling[Length - 1] == '"')
252 break;
253 }
254
255 // Raw string literals need special handling; trigraph expansion and line
256 // splicing do not occur within their d-char-sequence nor within their
257 // r-char-sequence.
258 if (Length >= 2 &&
259 Spelling[Length - 2] == 'R' && Spelling[Length - 1] == '"') {
260 // Search backwards from the end of the token to find the matching closing
261 // quote.
262 const char *RawEnd = BufEnd;
263 do --RawEnd; while (*RawEnd != '"');
264 size_t RawLength = RawEnd - BufPtr + 1;
265
266 // Everything between the quotes is included verbatim in the spelling.
267 memcpy(Spelling + Length, BufPtr, RawLength);
268 Length += RawLength;
269 BufPtr += RawLength;
270
271 // The rest of the token is lexed normally.
272 }
273 }
274
275 while (BufPtr < BufEnd) {
276 unsigned Size;
277 Spelling[Length++] = Lexer::getCharAndSizeNoWarn(BufPtr, Size, LangOpts);
278 BufPtr += Size;
279 }
280
281 assert(Length < Tok.getLength() &&
282 "NeedsCleaning flag set on token that didn't need cleaning!");
283 return Length;
284 }
285
286 /// getSpelling() - Return the 'spelling' of this token. The spelling of a
287 /// token are the characters used to represent the token in the source file
288 /// after trigraph expansion and escaped-newline folding. In particular, this
289 /// wants to get the true, uncanonicalized, spelling of things like digraphs
290 /// UCNs, etc.
getSpelling(SourceLocation loc,SmallVectorImpl<char> & buffer,const SourceManager & SM,const LangOptions & options,bool * invalid)291 StringRef Lexer::getSpelling(SourceLocation loc,
292 SmallVectorImpl<char> &buffer,
293 const SourceManager &SM,
294 const LangOptions &options,
295 bool *invalid) {
296 // Break down the source location.
297 std::pair<FileID, unsigned> locInfo = SM.getDecomposedLoc(loc);
298
299 // Try to the load the file buffer.
300 bool invalidTemp = false;
301 StringRef file = SM.getBufferData(locInfo.first, &invalidTemp);
302 if (invalidTemp) {
303 if (invalid) *invalid = true;
304 return StringRef();
305 }
306
307 const char *tokenBegin = file.data() + locInfo.second;
308
309 // Lex from the start of the given location.
310 Lexer lexer(SM.getLocForStartOfFile(locInfo.first), options,
311 file.begin(), tokenBegin, file.end());
312 Token token;
313 lexer.LexFromRawLexer(token);
314
315 unsigned length = token.getLength();
316
317 // Common case: no need for cleaning.
318 if (!token.needsCleaning())
319 return StringRef(tokenBegin, length);
320
321 // Hard case, we need to relex the characters into the string.
322 buffer.resize(length);
323 buffer.resize(getSpellingSlow(token, tokenBegin, options, buffer.data()));
324 return StringRef(buffer.data(), buffer.size());
325 }
326
327 /// getSpelling() - Return the 'spelling' of this token. The spelling of a
328 /// token are the characters used to represent the token in the source file
329 /// after trigraph expansion and escaped-newline folding. In particular, this
330 /// wants to get the true, uncanonicalized, spelling of things like digraphs
331 /// UCNs, etc.
getSpelling(const Token & Tok,const SourceManager & SourceMgr,const LangOptions & LangOpts,bool * Invalid)332 std::string Lexer::getSpelling(const Token &Tok, const SourceManager &SourceMgr,
333 const LangOptions &LangOpts, bool *Invalid) {
334 assert((int)Tok.getLength() >= 0 && "Token character range is bogus!");
335
336 bool CharDataInvalid = false;
337 const char *TokStart = SourceMgr.getCharacterData(Tok.getLocation(),
338 &CharDataInvalid);
339 if (Invalid)
340 *Invalid = CharDataInvalid;
341 if (CharDataInvalid)
342 return std::string();
343
344 // If this token contains nothing interesting, return it directly.
345 if (!Tok.needsCleaning())
346 return std::string(TokStart, TokStart + Tok.getLength());
347
348 std::string Result;
349 Result.resize(Tok.getLength());
350 Result.resize(getSpellingSlow(Tok, TokStart, LangOpts, &*Result.begin()));
351 return Result;
352 }
353
354 /// getSpelling - This method is used to get the spelling of a token into a
355 /// preallocated buffer, instead of as an std::string. The caller is required
356 /// to allocate enough space for the token, which is guaranteed to be at least
357 /// Tok.getLength() bytes long. The actual length of the token is returned.
358 ///
359 /// Note that this method may do two possible things: it may either fill in
360 /// the buffer specified with characters, or it may *change the input pointer*
361 /// to point to a constant buffer with the data already in it (avoiding a
362 /// copy). The caller is not allowed to modify the returned buffer pointer
363 /// if an internal buffer is returned.
getSpelling(const Token & Tok,const char * & Buffer,const SourceManager & SourceMgr,const LangOptions & LangOpts,bool * Invalid)364 unsigned Lexer::getSpelling(const Token &Tok, const char *&Buffer,
365 const SourceManager &SourceMgr,
366 const LangOptions &LangOpts, bool *Invalid) {
367 assert((int)Tok.getLength() >= 0 && "Token character range is bogus!");
368
369 const char *TokStart = nullptr;
370 // NOTE: this has to be checked *before* testing for an IdentifierInfo.
371 if (Tok.is(tok::raw_identifier))
372 TokStart = Tok.getRawIdentifier().data();
373 else if (!Tok.hasUCN()) {
374 if (const IdentifierInfo *II = Tok.getIdentifierInfo()) {
375 // Just return the string from the identifier table, which is very quick.
376 Buffer = II->getNameStart();
377 return II->getLength();
378 }
379 }
380
381 // NOTE: this can be checked even after testing for an IdentifierInfo.
382 if (Tok.isLiteral())
383 TokStart = Tok.getLiteralData();
384
385 if (!TokStart) {
386 // Compute the start of the token in the input lexer buffer.
387 bool CharDataInvalid = false;
388 TokStart = SourceMgr.getCharacterData(Tok.getLocation(), &CharDataInvalid);
389 if (Invalid)
390 *Invalid = CharDataInvalid;
391 if (CharDataInvalid) {
392 Buffer = "";
393 return 0;
394 }
395 }
396
397 // If this token contains nothing interesting, return it directly.
398 if (!Tok.needsCleaning()) {
399 Buffer = TokStart;
400 return Tok.getLength();
401 }
402
403 // Otherwise, hard case, relex the characters into the string.
404 return getSpellingSlow(Tok, TokStart, LangOpts, const_cast<char*>(Buffer));
405 }
406
407
408 /// MeasureTokenLength - Relex the token at the specified location and return
409 /// its length in bytes in the input file. If the token needs cleaning (e.g.
410 /// includes a trigraph or an escaped newline) then this count includes bytes
411 /// that are part of that.
MeasureTokenLength(SourceLocation Loc,const SourceManager & SM,const LangOptions & LangOpts)412 unsigned Lexer::MeasureTokenLength(SourceLocation Loc,
413 const SourceManager &SM,
414 const LangOptions &LangOpts) {
415 Token TheTok;
416 if (getRawToken(Loc, TheTok, SM, LangOpts))
417 return 0;
418 return TheTok.getLength();
419 }
420
421 /// \brief Relex the token at the specified location.
422 /// \returns true if there was a failure, false on success.
getRawToken(SourceLocation Loc,Token & Result,const SourceManager & SM,const LangOptions & LangOpts,bool IgnoreWhiteSpace)423 bool Lexer::getRawToken(SourceLocation Loc, Token &Result,
424 const SourceManager &SM,
425 const LangOptions &LangOpts,
426 bool IgnoreWhiteSpace) {
427 // TODO: this could be special cased for common tokens like identifiers, ')',
428 // etc to make this faster, if it mattered. Just look at StrData[0] to handle
429 // all obviously single-char tokens. This could use
430 // Lexer::isObviouslySimpleCharacter for example to handle identifiers or
431 // something.
432
433 // If this comes from a macro expansion, we really do want the macro name, not
434 // the token this macro expanded to.
435 Loc = SM.getExpansionLoc(Loc);
436 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc);
437 bool Invalid = false;
438 StringRef Buffer = SM.getBufferData(LocInfo.first, &Invalid);
439 if (Invalid)
440 return true;
441
442 const char *StrData = Buffer.data()+LocInfo.second;
443
444 if (!IgnoreWhiteSpace && isWhitespace(StrData[0]))
445 return true;
446
447 // Create a lexer starting at the beginning of this token.
448 Lexer TheLexer(SM.getLocForStartOfFile(LocInfo.first), LangOpts,
449 Buffer.begin(), StrData, Buffer.end());
450 TheLexer.SetCommentRetentionState(true);
451 TheLexer.LexFromRawLexer(Result);
452 return false;
453 }
454
getBeginningOfFileToken(SourceLocation Loc,const SourceManager & SM,const LangOptions & LangOpts)455 static SourceLocation getBeginningOfFileToken(SourceLocation Loc,
456 const SourceManager &SM,
457 const LangOptions &LangOpts) {
458 assert(Loc.isFileID());
459 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc);
460 if (LocInfo.first.isInvalid())
461 return Loc;
462
463 bool Invalid = false;
464 StringRef Buffer = SM.getBufferData(LocInfo.first, &Invalid);
465 if (Invalid)
466 return Loc;
467
468 // Back up from the current location until we hit the beginning of a line
469 // (or the buffer). We'll relex from that point.
470 const char *BufStart = Buffer.data();
471 if (LocInfo.second >= Buffer.size())
472 return Loc;
473
474 const char *StrData = BufStart+LocInfo.second;
475 if (StrData[0] == '\n' || StrData[0] == '\r')
476 return Loc;
477
478 const char *LexStart = StrData;
479 while (LexStart != BufStart) {
480 if (LexStart[0] == '\n' || LexStart[0] == '\r') {
481 ++LexStart;
482 break;
483 }
484
485 --LexStart;
486 }
487
488 // Create a lexer starting at the beginning of this token.
489 SourceLocation LexerStartLoc = Loc.getLocWithOffset(-LocInfo.second);
490 Lexer TheLexer(LexerStartLoc, LangOpts, BufStart, LexStart, Buffer.end());
491 TheLexer.SetCommentRetentionState(true);
492
493 // Lex tokens until we find the token that contains the source location.
494 Token TheTok;
495 do {
496 TheLexer.LexFromRawLexer(TheTok);
497
498 if (TheLexer.getBufferLocation() > StrData) {
499 // Lexing this token has taken the lexer past the source location we're
500 // looking for. If the current token encompasses our source location,
501 // return the beginning of that token.
502 if (TheLexer.getBufferLocation() - TheTok.getLength() <= StrData)
503 return TheTok.getLocation();
504
505 // We ended up skipping over the source location entirely, which means
506 // that it points into whitespace. We're done here.
507 break;
508 }
509 } while (TheTok.getKind() != tok::eof);
510
511 // We've passed our source location; just return the original source location.
512 return Loc;
513 }
514
GetBeginningOfToken(SourceLocation Loc,const SourceManager & SM,const LangOptions & LangOpts)515 SourceLocation Lexer::GetBeginningOfToken(SourceLocation Loc,
516 const SourceManager &SM,
517 const LangOptions &LangOpts) {
518 if (Loc.isFileID())
519 return getBeginningOfFileToken(Loc, SM, LangOpts);
520
521 if (!SM.isMacroArgExpansion(Loc))
522 return Loc;
523
524 SourceLocation FileLoc = SM.getSpellingLoc(Loc);
525 SourceLocation BeginFileLoc = getBeginningOfFileToken(FileLoc, SM, LangOpts);
526 std::pair<FileID, unsigned> FileLocInfo = SM.getDecomposedLoc(FileLoc);
527 std::pair<FileID, unsigned> BeginFileLocInfo
528 = SM.getDecomposedLoc(BeginFileLoc);
529 assert(FileLocInfo.first == BeginFileLocInfo.first &&
530 FileLocInfo.second >= BeginFileLocInfo.second);
531 return Loc.getLocWithOffset(BeginFileLocInfo.second - FileLocInfo.second);
532 }
533
534 namespace {
535 enum PreambleDirectiveKind {
536 PDK_Skipped,
537 PDK_StartIf,
538 PDK_EndIf,
539 PDK_Unknown
540 };
541 }
542
ComputePreamble(StringRef Buffer,const LangOptions & LangOpts,unsigned MaxLines)543 std::pair<unsigned, bool> Lexer::ComputePreamble(StringRef Buffer,
544 const LangOptions &LangOpts,
545 unsigned MaxLines) {
546 // Create a lexer starting at the beginning of the file. Note that we use a
547 // "fake" file source location at offset 1 so that the lexer will track our
548 // position within the file.
549 const unsigned StartOffset = 1;
550 SourceLocation FileLoc = SourceLocation::getFromRawEncoding(StartOffset);
551 Lexer TheLexer(FileLoc, LangOpts, Buffer.begin(), Buffer.begin(),
552 Buffer.end());
553 TheLexer.SetCommentRetentionState(true);
554
555 // StartLoc will differ from FileLoc if there is a BOM that was skipped.
556 SourceLocation StartLoc = TheLexer.getSourceLocation();
557
558 bool InPreprocessorDirective = false;
559 Token TheTok;
560 Token IfStartTok;
561 unsigned IfCount = 0;
562 SourceLocation ActiveCommentLoc;
563
564 unsigned MaxLineOffset = 0;
565 if (MaxLines) {
566 const char *CurPtr = Buffer.begin();
567 unsigned CurLine = 0;
568 while (CurPtr != Buffer.end()) {
569 char ch = *CurPtr++;
570 if (ch == '\n') {
571 ++CurLine;
572 if (CurLine == MaxLines)
573 break;
574 }
575 }
576 if (CurPtr != Buffer.end())
577 MaxLineOffset = CurPtr - Buffer.begin();
578 }
579
580 do {
581 TheLexer.LexFromRawLexer(TheTok);
582
583 if (InPreprocessorDirective) {
584 // If we've hit the end of the file, we're done.
585 if (TheTok.getKind() == tok::eof) {
586 break;
587 }
588
589 // If we haven't hit the end of the preprocessor directive, skip this
590 // token.
591 if (!TheTok.isAtStartOfLine())
592 continue;
593
594 // We've passed the end of the preprocessor directive, and will look
595 // at this token again below.
596 InPreprocessorDirective = false;
597 }
598
599 // Keep track of the # of lines in the preamble.
600 if (TheTok.isAtStartOfLine()) {
601 unsigned TokOffset = TheTok.getLocation().getRawEncoding() - StartOffset;
602
603 // If we were asked to limit the number of lines in the preamble,
604 // and we're about to exceed that limit, we're done.
605 if (MaxLineOffset && TokOffset >= MaxLineOffset)
606 break;
607 }
608
609 // Comments are okay; skip over them.
610 if (TheTok.getKind() == tok::comment) {
611 if (ActiveCommentLoc.isInvalid())
612 ActiveCommentLoc = TheTok.getLocation();
613 continue;
614 }
615
616 if (TheTok.isAtStartOfLine() && TheTok.getKind() == tok::hash) {
617 // This is the start of a preprocessor directive.
618 Token HashTok = TheTok;
619 InPreprocessorDirective = true;
620 ActiveCommentLoc = SourceLocation();
621
622 // Figure out which directive this is. Since we're lexing raw tokens,
623 // we don't have an identifier table available. Instead, just look at
624 // the raw identifier to recognize and categorize preprocessor directives.
625 TheLexer.LexFromRawLexer(TheTok);
626 if (TheTok.getKind() == tok::raw_identifier && !TheTok.needsCleaning()) {
627 StringRef Keyword = TheTok.getRawIdentifier();
628 PreambleDirectiveKind PDK
629 = llvm::StringSwitch<PreambleDirectiveKind>(Keyword)
630 .Case("include", PDK_Skipped)
631 .Case("__include_macros", PDK_Skipped)
632 .Case("define", PDK_Skipped)
633 .Case("undef", PDK_Skipped)
634 .Case("line", PDK_Skipped)
635 .Case("error", PDK_Skipped)
636 .Case("pragma", PDK_Skipped)
637 .Case("import", PDK_Skipped)
638 .Case("include_next", PDK_Skipped)
639 .Case("warning", PDK_Skipped)
640 .Case("ident", PDK_Skipped)
641 .Case("sccs", PDK_Skipped)
642 .Case("assert", PDK_Skipped)
643 .Case("unassert", PDK_Skipped)
644 .Case("if", PDK_StartIf)
645 .Case("ifdef", PDK_StartIf)
646 .Case("ifndef", PDK_StartIf)
647 .Case("elif", PDK_Skipped)
648 .Case("else", PDK_Skipped)
649 .Case("endif", PDK_EndIf)
650 .Default(PDK_Unknown);
651
652 switch (PDK) {
653 case PDK_Skipped:
654 continue;
655
656 case PDK_StartIf:
657 if (IfCount == 0)
658 IfStartTok = HashTok;
659
660 ++IfCount;
661 continue;
662
663 case PDK_EndIf:
664 // Mismatched #endif. The preamble ends here.
665 if (IfCount == 0)
666 break;
667
668 --IfCount;
669 continue;
670
671 case PDK_Unknown:
672 // We don't know what this directive is; stop at the '#'.
673 break;
674 }
675 }
676
677 // We only end up here if we didn't recognize the preprocessor
678 // directive or it was one that can't occur in the preamble at this
679 // point. Roll back the current token to the location of the '#'.
680 InPreprocessorDirective = false;
681 TheTok = HashTok;
682 }
683
684 // We hit a token that we don't recognize as being in the
685 // "preprocessing only" part of the file, so we're no longer in
686 // the preamble.
687 break;
688 } while (true);
689
690 SourceLocation End;
691 if (IfCount)
692 End = IfStartTok.getLocation();
693 else if (ActiveCommentLoc.isValid())
694 End = ActiveCommentLoc; // don't truncate a decl comment.
695 else
696 End = TheTok.getLocation();
697
698 return std::make_pair(End.getRawEncoding() - StartLoc.getRawEncoding(),
699 IfCount? IfStartTok.isAtStartOfLine()
700 : TheTok.isAtStartOfLine());
701 }
702
703
704 /// AdvanceToTokenCharacter - Given a location that specifies the start of a
705 /// token, return a new location that specifies a character within the token.
AdvanceToTokenCharacter(SourceLocation TokStart,unsigned CharNo,const SourceManager & SM,const LangOptions & LangOpts)706 SourceLocation Lexer::AdvanceToTokenCharacter(SourceLocation TokStart,
707 unsigned CharNo,
708 const SourceManager &SM,
709 const LangOptions &LangOpts) {
710 // Figure out how many physical characters away the specified expansion
711 // character is. This needs to take into consideration newlines and
712 // trigraphs.
713 bool Invalid = false;
714 const char *TokPtr = SM.getCharacterData(TokStart, &Invalid);
715
716 // If they request the first char of the token, we're trivially done.
717 if (Invalid || (CharNo == 0 && Lexer::isObviouslySimpleCharacter(*TokPtr)))
718 return TokStart;
719
720 unsigned PhysOffset = 0;
721
722 // The usual case is that tokens don't contain anything interesting. Skip
723 // over the uninteresting characters. If a token only consists of simple
724 // chars, this method is extremely fast.
725 while (Lexer::isObviouslySimpleCharacter(*TokPtr)) {
726 if (CharNo == 0)
727 return TokStart.getLocWithOffset(PhysOffset);
728 ++TokPtr, --CharNo, ++PhysOffset;
729 }
730
731 // If we have a character that may be a trigraph or escaped newline, use a
732 // lexer to parse it correctly.
733 for (; CharNo; --CharNo) {
734 unsigned Size;
735 Lexer::getCharAndSizeNoWarn(TokPtr, Size, LangOpts);
736 TokPtr += Size;
737 PhysOffset += Size;
738 }
739
740 // Final detail: if we end up on an escaped newline, we want to return the
741 // location of the actual byte of the token. For example foo\<newline>bar
742 // advanced by 3 should return the location of b, not of \\. One compounding
743 // detail of this is that the escape may be made by a trigraph.
744 if (!Lexer::isObviouslySimpleCharacter(*TokPtr))
745 PhysOffset += Lexer::SkipEscapedNewLines(TokPtr)-TokPtr;
746
747 return TokStart.getLocWithOffset(PhysOffset);
748 }
749
750 /// \brief Computes the source location just past the end of the
751 /// token at this source location.
752 ///
753 /// This routine can be used to produce a source location that
754 /// points just past the end of the token referenced by \p Loc, and
755 /// is generally used when a diagnostic needs to point just after a
756 /// token where it expected something different that it received. If
757 /// the returned source location would not be meaningful (e.g., if
758 /// it points into a macro), this routine returns an invalid
759 /// source location.
760 ///
761 /// \param Offset an offset from the end of the token, where the source
762 /// location should refer to. The default offset (0) produces a source
763 /// location pointing just past the end of the token; an offset of 1 produces
764 /// a source location pointing to the last character in the token, etc.
getLocForEndOfToken(SourceLocation Loc,unsigned Offset,const SourceManager & SM,const LangOptions & LangOpts)765 SourceLocation Lexer::getLocForEndOfToken(SourceLocation Loc, unsigned Offset,
766 const SourceManager &SM,
767 const LangOptions &LangOpts) {
768 if (Loc.isInvalid())
769 return SourceLocation();
770
771 if (Loc.isMacroID()) {
772 if (Offset > 0 || !isAtEndOfMacroExpansion(Loc, SM, LangOpts, &Loc))
773 return SourceLocation(); // Points inside the macro expansion.
774 }
775
776 unsigned Len = Lexer::MeasureTokenLength(Loc, SM, LangOpts);
777 if (Len > Offset)
778 Len = Len - Offset;
779 else
780 return Loc;
781
782 return Loc.getLocWithOffset(Len);
783 }
784
785 /// \brief Returns true if the given MacroID location points at the first
786 /// token of the macro expansion.
isAtStartOfMacroExpansion(SourceLocation loc,const SourceManager & SM,const LangOptions & LangOpts,SourceLocation * MacroBegin)787 bool Lexer::isAtStartOfMacroExpansion(SourceLocation loc,
788 const SourceManager &SM,
789 const LangOptions &LangOpts,
790 SourceLocation *MacroBegin) {
791 assert(loc.isValid() && loc.isMacroID() && "Expected a valid macro loc");
792
793 SourceLocation expansionLoc;
794 if (!SM.isAtStartOfImmediateMacroExpansion(loc, &expansionLoc))
795 return false;
796
797 if (expansionLoc.isFileID()) {
798 // No other macro expansions, this is the first.
799 if (MacroBegin)
800 *MacroBegin = expansionLoc;
801 return true;
802 }
803
804 return isAtStartOfMacroExpansion(expansionLoc, SM, LangOpts, MacroBegin);
805 }
806
807 /// \brief Returns true if the given MacroID location points at the last
808 /// token of the macro expansion.
isAtEndOfMacroExpansion(SourceLocation loc,const SourceManager & SM,const LangOptions & LangOpts,SourceLocation * MacroEnd)809 bool Lexer::isAtEndOfMacroExpansion(SourceLocation loc,
810 const SourceManager &SM,
811 const LangOptions &LangOpts,
812 SourceLocation *MacroEnd) {
813 assert(loc.isValid() && loc.isMacroID() && "Expected a valid macro loc");
814
815 SourceLocation spellLoc = SM.getSpellingLoc(loc);
816 unsigned tokLen = MeasureTokenLength(spellLoc, SM, LangOpts);
817 if (tokLen == 0)
818 return false;
819
820 SourceLocation afterLoc = loc.getLocWithOffset(tokLen);
821 SourceLocation expansionLoc;
822 if (!SM.isAtEndOfImmediateMacroExpansion(afterLoc, &expansionLoc))
823 return false;
824
825 if (expansionLoc.isFileID()) {
826 // No other macro expansions.
827 if (MacroEnd)
828 *MacroEnd = expansionLoc;
829 return true;
830 }
831
832 return isAtEndOfMacroExpansion(expansionLoc, SM, LangOpts, MacroEnd);
833 }
834
makeRangeFromFileLocs(CharSourceRange Range,const SourceManager & SM,const LangOptions & LangOpts)835 static CharSourceRange makeRangeFromFileLocs(CharSourceRange Range,
836 const SourceManager &SM,
837 const LangOptions &LangOpts) {
838 SourceLocation Begin = Range.getBegin();
839 SourceLocation End = Range.getEnd();
840 assert(Begin.isFileID() && End.isFileID());
841 if (Range.isTokenRange()) {
842 End = Lexer::getLocForEndOfToken(End, 0, SM,LangOpts);
843 if (End.isInvalid())
844 return CharSourceRange();
845 }
846
847 // Break down the source locations.
848 FileID FID;
849 unsigned BeginOffs;
850 std::tie(FID, BeginOffs) = SM.getDecomposedLoc(Begin);
851 if (FID.isInvalid())
852 return CharSourceRange();
853
854 unsigned EndOffs;
855 if (!SM.isInFileID(End, FID, &EndOffs) ||
856 BeginOffs > EndOffs)
857 return CharSourceRange();
858
859 return CharSourceRange::getCharRange(Begin, End);
860 }
861
makeFileCharRange(CharSourceRange Range,const SourceManager & SM,const LangOptions & LangOpts)862 CharSourceRange Lexer::makeFileCharRange(CharSourceRange Range,
863 const SourceManager &SM,
864 const LangOptions &LangOpts) {
865 SourceLocation Begin = Range.getBegin();
866 SourceLocation End = Range.getEnd();
867 if (Begin.isInvalid() || End.isInvalid())
868 return CharSourceRange();
869
870 if (Begin.isFileID() && End.isFileID())
871 return makeRangeFromFileLocs(Range, SM, LangOpts);
872
873 if (Begin.isMacroID() && End.isFileID()) {
874 if (!isAtStartOfMacroExpansion(Begin, SM, LangOpts, &Begin))
875 return CharSourceRange();
876 Range.setBegin(Begin);
877 return makeRangeFromFileLocs(Range, SM, LangOpts);
878 }
879
880 if (Begin.isFileID() && End.isMacroID()) {
881 if ((Range.isTokenRange() && !isAtEndOfMacroExpansion(End, SM, LangOpts,
882 &End)) ||
883 (Range.isCharRange() && !isAtStartOfMacroExpansion(End, SM, LangOpts,
884 &End)))
885 return CharSourceRange();
886 Range.setEnd(End);
887 return makeRangeFromFileLocs(Range, SM, LangOpts);
888 }
889
890 assert(Begin.isMacroID() && End.isMacroID());
891 SourceLocation MacroBegin, MacroEnd;
892 if (isAtStartOfMacroExpansion(Begin, SM, LangOpts, &MacroBegin) &&
893 ((Range.isTokenRange() && isAtEndOfMacroExpansion(End, SM, LangOpts,
894 &MacroEnd)) ||
895 (Range.isCharRange() && isAtStartOfMacroExpansion(End, SM, LangOpts,
896 &MacroEnd)))) {
897 Range.setBegin(MacroBegin);
898 Range.setEnd(MacroEnd);
899 return makeRangeFromFileLocs(Range, SM, LangOpts);
900 }
901
902 bool Invalid = false;
903 const SrcMgr::SLocEntry &BeginEntry = SM.getSLocEntry(SM.getFileID(Begin),
904 &Invalid);
905 if (Invalid)
906 return CharSourceRange();
907
908 if (BeginEntry.getExpansion().isMacroArgExpansion()) {
909 const SrcMgr::SLocEntry &EndEntry = SM.getSLocEntry(SM.getFileID(End),
910 &Invalid);
911 if (Invalid)
912 return CharSourceRange();
913
914 if (EndEntry.getExpansion().isMacroArgExpansion() &&
915 BeginEntry.getExpansion().getExpansionLocStart() ==
916 EndEntry.getExpansion().getExpansionLocStart()) {
917 Range.setBegin(SM.getImmediateSpellingLoc(Begin));
918 Range.setEnd(SM.getImmediateSpellingLoc(End));
919 return makeFileCharRange(Range, SM, LangOpts);
920 }
921 }
922
923 return CharSourceRange();
924 }
925
getSourceText(CharSourceRange Range,const SourceManager & SM,const LangOptions & LangOpts,bool * Invalid)926 StringRef Lexer::getSourceText(CharSourceRange Range,
927 const SourceManager &SM,
928 const LangOptions &LangOpts,
929 bool *Invalid) {
930 Range = makeFileCharRange(Range, SM, LangOpts);
931 if (Range.isInvalid()) {
932 if (Invalid) *Invalid = true;
933 return StringRef();
934 }
935
936 // Break down the source location.
937 std::pair<FileID, unsigned> beginInfo = SM.getDecomposedLoc(Range.getBegin());
938 if (beginInfo.first.isInvalid()) {
939 if (Invalid) *Invalid = true;
940 return StringRef();
941 }
942
943 unsigned EndOffs;
944 if (!SM.isInFileID(Range.getEnd(), beginInfo.first, &EndOffs) ||
945 beginInfo.second > EndOffs) {
946 if (Invalid) *Invalid = true;
947 return StringRef();
948 }
949
950 // Try to the load the file buffer.
951 bool invalidTemp = false;
952 StringRef file = SM.getBufferData(beginInfo.first, &invalidTemp);
953 if (invalidTemp) {
954 if (Invalid) *Invalid = true;
955 return StringRef();
956 }
957
958 if (Invalid) *Invalid = false;
959 return file.substr(beginInfo.second, EndOffs - beginInfo.second);
960 }
961
getImmediateMacroName(SourceLocation Loc,const SourceManager & SM,const LangOptions & LangOpts)962 StringRef Lexer::getImmediateMacroName(SourceLocation Loc,
963 const SourceManager &SM,
964 const LangOptions &LangOpts) {
965 assert(Loc.isMacroID() && "Only reasonble to call this on macros");
966
967 // Find the location of the immediate macro expansion.
968 while (1) {
969 FileID FID = SM.getFileID(Loc);
970 const SrcMgr::SLocEntry *E = &SM.getSLocEntry(FID);
971 const SrcMgr::ExpansionInfo &Expansion = E->getExpansion();
972 Loc = Expansion.getExpansionLocStart();
973 if (!Expansion.isMacroArgExpansion())
974 break;
975
976 // For macro arguments we need to check that the argument did not come
977 // from an inner macro, e.g: "MAC1( MAC2(foo) )"
978
979 // Loc points to the argument id of the macro definition, move to the
980 // macro expansion.
981 Loc = SM.getImmediateExpansionRange(Loc).first;
982 SourceLocation SpellLoc = Expansion.getSpellingLoc();
983 if (SpellLoc.isFileID())
984 break; // No inner macro.
985
986 // If spelling location resides in the same FileID as macro expansion
987 // location, it means there is no inner macro.
988 FileID MacroFID = SM.getFileID(Loc);
989 if (SM.isInFileID(SpellLoc, MacroFID))
990 break;
991
992 // Argument came from inner macro.
993 Loc = SpellLoc;
994 }
995
996 // Find the spelling location of the start of the non-argument expansion
997 // range. This is where the macro name was spelled in order to begin
998 // expanding this macro.
999 Loc = SM.getSpellingLoc(Loc);
1000
1001 // Dig out the buffer where the macro name was spelled and the extents of the
1002 // name so that we can render it into the expansion note.
1003 std::pair<FileID, unsigned> ExpansionInfo = SM.getDecomposedLoc(Loc);
1004 unsigned MacroTokenLength = Lexer::MeasureTokenLength(Loc, SM, LangOpts);
1005 StringRef ExpansionBuffer = SM.getBufferData(ExpansionInfo.first);
1006 return ExpansionBuffer.substr(ExpansionInfo.second, MacroTokenLength);
1007 }
1008
isIdentifierBodyChar(char c,const LangOptions & LangOpts)1009 bool Lexer::isIdentifierBodyChar(char c, const LangOptions &LangOpts) {
1010 return isIdentifierBody(c, LangOpts.DollarIdents);
1011 }
1012
1013
1014 //===----------------------------------------------------------------------===//
1015 // Diagnostics forwarding code.
1016 //===----------------------------------------------------------------------===//
1017
1018 /// GetMappedTokenLoc - If lexing out of a 'mapped buffer', where we pretend the
1019 /// lexer buffer was all expanded at a single point, perform the mapping.
1020 /// This is currently only used for _Pragma implementation, so it is the slow
1021 /// path of the hot getSourceLocation method. Do not allow it to be inlined.
1022 static LLVM_ATTRIBUTE_NOINLINE SourceLocation GetMappedTokenLoc(
1023 Preprocessor &PP, SourceLocation FileLoc, unsigned CharNo, unsigned TokLen);
GetMappedTokenLoc(Preprocessor & PP,SourceLocation FileLoc,unsigned CharNo,unsigned TokLen)1024 static SourceLocation GetMappedTokenLoc(Preprocessor &PP,
1025 SourceLocation FileLoc,
1026 unsigned CharNo, unsigned TokLen) {
1027 assert(FileLoc.isMacroID() && "Must be a macro expansion");
1028
1029 // Otherwise, we're lexing "mapped tokens". This is used for things like
1030 // _Pragma handling. Combine the expansion location of FileLoc with the
1031 // spelling location.
1032 SourceManager &SM = PP.getSourceManager();
1033
1034 // Create a new SLoc which is expanded from Expansion(FileLoc) but whose
1035 // characters come from spelling(FileLoc)+Offset.
1036 SourceLocation SpellingLoc = SM.getSpellingLoc(FileLoc);
1037 SpellingLoc = SpellingLoc.getLocWithOffset(CharNo);
1038
1039 // Figure out the expansion loc range, which is the range covered by the
1040 // original _Pragma(...) sequence.
1041 std::pair<SourceLocation,SourceLocation> II =
1042 SM.getImmediateExpansionRange(FileLoc);
1043
1044 return SM.createExpansionLoc(SpellingLoc, II.first, II.second, TokLen);
1045 }
1046
1047 /// getSourceLocation - Return a source location identifier for the specified
1048 /// offset in the current file.
getSourceLocation(const char * Loc,unsigned TokLen) const1049 SourceLocation Lexer::getSourceLocation(const char *Loc,
1050 unsigned TokLen) const {
1051 assert(Loc >= BufferStart && Loc <= BufferEnd &&
1052 "Location out of range for this buffer!");
1053
1054 // In the normal case, we're just lexing from a simple file buffer, return
1055 // the file id from FileLoc with the offset specified.
1056 unsigned CharNo = Loc-BufferStart;
1057 if (FileLoc.isFileID())
1058 return FileLoc.getLocWithOffset(CharNo);
1059
1060 // Otherwise, this is the _Pragma lexer case, which pretends that all of the
1061 // tokens are lexed from where the _Pragma was defined.
1062 assert(PP && "This doesn't work on raw lexers");
1063 return GetMappedTokenLoc(*PP, FileLoc, CharNo, TokLen);
1064 }
1065
1066 /// Diag - Forwarding function for diagnostics. This translate a source
1067 /// position in the current buffer into a SourceLocation object for rendering.
Diag(const char * Loc,unsigned DiagID) const1068 DiagnosticBuilder Lexer::Diag(const char *Loc, unsigned DiagID) const {
1069 return PP->Diag(getSourceLocation(Loc), DiagID);
1070 }
1071
1072 //===----------------------------------------------------------------------===//
1073 // Trigraph and Escaped Newline Handling Code.
1074 //===----------------------------------------------------------------------===//
1075
1076 /// GetTrigraphCharForLetter - Given a character that occurs after a ?? pair,
1077 /// return the decoded trigraph letter it corresponds to, or '\0' if nothing.
GetTrigraphCharForLetter(char Letter)1078 static char GetTrigraphCharForLetter(char Letter) {
1079 switch (Letter) {
1080 default: return 0;
1081 case '=': return '#';
1082 case ')': return ']';
1083 case '(': return '[';
1084 case '!': return '|';
1085 case '\'': return '^';
1086 case '>': return '}';
1087 case '/': return '\\';
1088 case '<': return '{';
1089 case '-': return '~';
1090 }
1091 }
1092
1093 /// DecodeTrigraphChar - If the specified character is a legal trigraph when
1094 /// prefixed with ??, emit a trigraph warning. If trigraphs are enabled,
1095 /// return the result character. Finally, emit a warning about trigraph use
1096 /// whether trigraphs are enabled or not.
DecodeTrigraphChar(const char * CP,Lexer * L)1097 static char DecodeTrigraphChar(const char *CP, Lexer *L) {
1098 char Res = GetTrigraphCharForLetter(*CP);
1099 if (!Res || !L) return Res;
1100
1101 if (!L->getLangOpts().Trigraphs) {
1102 if (!L->isLexingRawMode())
1103 L->Diag(CP-2, diag::trigraph_ignored);
1104 return 0;
1105 }
1106
1107 if (!L->isLexingRawMode())
1108 L->Diag(CP-2, diag::trigraph_converted) << StringRef(&Res, 1);
1109 return Res;
1110 }
1111
1112 /// getEscapedNewLineSize - Return the size of the specified escaped newline,
1113 /// or 0 if it is not an escaped newline. P[-1] is known to be a "\" or a
1114 /// trigraph equivalent on entry to this function.
getEscapedNewLineSize(const char * Ptr)1115 unsigned Lexer::getEscapedNewLineSize(const char *Ptr) {
1116 unsigned Size = 0;
1117 while (isWhitespace(Ptr[Size])) {
1118 ++Size;
1119
1120 if (Ptr[Size-1] != '\n' && Ptr[Size-1] != '\r')
1121 continue;
1122
1123 // If this is a \r\n or \n\r, skip the other half.
1124 if ((Ptr[Size] == '\r' || Ptr[Size] == '\n') &&
1125 Ptr[Size-1] != Ptr[Size])
1126 ++Size;
1127
1128 return Size;
1129 }
1130
1131 // Not an escaped newline, must be a \t or something else.
1132 return 0;
1133 }
1134
1135 /// SkipEscapedNewLines - If P points to an escaped newline (or a series of
1136 /// them), skip over them and return the first non-escaped-newline found,
1137 /// otherwise return P.
SkipEscapedNewLines(const char * P)1138 const char *Lexer::SkipEscapedNewLines(const char *P) {
1139 while (1) {
1140 const char *AfterEscape;
1141 if (*P == '\\') {
1142 AfterEscape = P+1;
1143 } else if (*P == '?') {
1144 // If not a trigraph for escape, bail out.
1145 if (P[1] != '?' || P[2] != '/')
1146 return P;
1147 AfterEscape = P+3;
1148 } else {
1149 return P;
1150 }
1151
1152 unsigned NewLineSize = Lexer::getEscapedNewLineSize(AfterEscape);
1153 if (NewLineSize == 0) return P;
1154 P = AfterEscape+NewLineSize;
1155 }
1156 }
1157
1158 /// \brief Checks that the given token is the first token that occurs after the
1159 /// given location (this excludes comments and whitespace). Returns the location
1160 /// immediately after the specified token. If the token is not found or the
1161 /// location is inside a macro, the returned source location will be invalid.
findLocationAfterToken(SourceLocation Loc,tok::TokenKind TKind,const SourceManager & SM,const LangOptions & LangOpts,bool SkipTrailingWhitespaceAndNewLine)1162 SourceLocation Lexer::findLocationAfterToken(SourceLocation Loc,
1163 tok::TokenKind TKind,
1164 const SourceManager &SM,
1165 const LangOptions &LangOpts,
1166 bool SkipTrailingWhitespaceAndNewLine) {
1167 if (Loc.isMacroID()) {
1168 if (!Lexer::isAtEndOfMacroExpansion(Loc, SM, LangOpts, &Loc))
1169 return SourceLocation();
1170 }
1171 Loc = Lexer::getLocForEndOfToken(Loc, 0, SM, LangOpts);
1172
1173 // Break down the source location.
1174 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc);
1175
1176 // Try to load the file buffer.
1177 bool InvalidTemp = false;
1178 StringRef File = SM.getBufferData(LocInfo.first, &InvalidTemp);
1179 if (InvalidTemp)
1180 return SourceLocation();
1181
1182 const char *TokenBegin = File.data() + LocInfo.second;
1183
1184 // Lex from the start of the given location.
1185 Lexer lexer(SM.getLocForStartOfFile(LocInfo.first), LangOpts, File.begin(),
1186 TokenBegin, File.end());
1187 // Find the token.
1188 Token Tok;
1189 lexer.LexFromRawLexer(Tok);
1190 if (Tok.isNot(TKind))
1191 return SourceLocation();
1192 SourceLocation TokenLoc = Tok.getLocation();
1193
1194 // Calculate how much whitespace needs to be skipped if any.
1195 unsigned NumWhitespaceChars = 0;
1196 if (SkipTrailingWhitespaceAndNewLine) {
1197 const char *TokenEnd = SM.getCharacterData(TokenLoc) +
1198 Tok.getLength();
1199 unsigned char C = *TokenEnd;
1200 while (isHorizontalWhitespace(C)) {
1201 C = *(++TokenEnd);
1202 NumWhitespaceChars++;
1203 }
1204
1205 // Skip \r, \n, \r\n, or \n\r
1206 if (C == '\n' || C == '\r') {
1207 char PrevC = C;
1208 C = *(++TokenEnd);
1209 NumWhitespaceChars++;
1210 if ((C == '\n' || C == '\r') && C != PrevC)
1211 NumWhitespaceChars++;
1212 }
1213 }
1214
1215 return TokenLoc.getLocWithOffset(Tok.getLength() + NumWhitespaceChars);
1216 }
1217
1218 /// getCharAndSizeSlow - Peek a single 'character' from the specified buffer,
1219 /// get its size, and return it. This is tricky in several cases:
1220 /// 1. If currently at the start of a trigraph, we warn about the trigraph,
1221 /// then either return the trigraph (skipping 3 chars) or the '?',
1222 /// depending on whether trigraphs are enabled or not.
1223 /// 2. If this is an escaped newline (potentially with whitespace between
1224 /// the backslash and newline), implicitly skip the newline and return
1225 /// the char after it.
1226 ///
1227 /// This handles the slow/uncommon case of the getCharAndSize method. Here we
1228 /// know that we can accumulate into Size, and that we have already incremented
1229 /// Ptr by Size bytes.
1230 ///
1231 /// NOTE: When this method is updated, getCharAndSizeSlowNoWarn (below) should
1232 /// be updated to match.
1233 ///
getCharAndSizeSlow(const char * Ptr,unsigned & Size,Token * Tok)1234 char Lexer::getCharAndSizeSlow(const char *Ptr, unsigned &Size,
1235 Token *Tok) {
1236 // If we have a slash, look for an escaped newline.
1237 if (Ptr[0] == '\\') {
1238 ++Size;
1239 ++Ptr;
1240 Slash:
1241 // Common case, backslash-char where the char is not whitespace.
1242 if (!isWhitespace(Ptr[0])) return '\\';
1243
1244 // See if we have optional whitespace characters between the slash and
1245 // newline.
1246 if (unsigned EscapedNewLineSize = getEscapedNewLineSize(Ptr)) {
1247 // Remember that this token needs to be cleaned.
1248 if (Tok) Tok->setFlag(Token::NeedsCleaning);
1249
1250 // Warn if there was whitespace between the backslash and newline.
1251 if (Ptr[0] != '\n' && Ptr[0] != '\r' && Tok && !isLexingRawMode())
1252 Diag(Ptr, diag::backslash_newline_space);
1253
1254 // Found backslash<whitespace><newline>. Parse the char after it.
1255 Size += EscapedNewLineSize;
1256 Ptr += EscapedNewLineSize;
1257
1258 // If the char that we finally got was a \n, then we must have had
1259 // something like \<newline><newline>. We don't want to consume the
1260 // second newline.
1261 if (*Ptr == '\n' || *Ptr == '\r' || *Ptr == '\0')
1262 return ' ';
1263
1264 // Use slow version to accumulate a correct size field.
1265 return getCharAndSizeSlow(Ptr, Size, Tok);
1266 }
1267
1268 // Otherwise, this is not an escaped newline, just return the slash.
1269 return '\\';
1270 }
1271
1272 // If this is a trigraph, process it.
1273 if (Ptr[0] == '?' && Ptr[1] == '?') {
1274 // If this is actually a legal trigraph (not something like "??x"), emit
1275 // a trigraph warning. If so, and if trigraphs are enabled, return it.
1276 if (char C = DecodeTrigraphChar(Ptr+2, Tok ? this : nullptr)) {
1277 // Remember that this token needs to be cleaned.
1278 if (Tok) Tok->setFlag(Token::NeedsCleaning);
1279
1280 Ptr += 3;
1281 Size += 3;
1282 if (C == '\\') goto Slash;
1283 return C;
1284 }
1285 }
1286
1287 // If this is neither, return a single character.
1288 ++Size;
1289 return *Ptr;
1290 }
1291
1292
1293 /// getCharAndSizeSlowNoWarn - Handle the slow/uncommon case of the
1294 /// getCharAndSizeNoWarn method. Here we know that we can accumulate into Size,
1295 /// and that we have already incremented Ptr by Size bytes.
1296 ///
1297 /// NOTE: When this method is updated, getCharAndSizeSlow (above) should
1298 /// be updated to match.
getCharAndSizeSlowNoWarn(const char * Ptr,unsigned & Size,const LangOptions & LangOpts)1299 char Lexer::getCharAndSizeSlowNoWarn(const char *Ptr, unsigned &Size,
1300 const LangOptions &LangOpts) {
1301 // If we have a slash, look for an escaped newline.
1302 if (Ptr[0] == '\\') {
1303 ++Size;
1304 ++Ptr;
1305 Slash:
1306 // Common case, backslash-char where the char is not whitespace.
1307 if (!isWhitespace(Ptr[0])) return '\\';
1308
1309 // See if we have optional whitespace characters followed by a newline.
1310 if (unsigned EscapedNewLineSize = getEscapedNewLineSize(Ptr)) {
1311 // Found backslash<whitespace><newline>. Parse the char after it.
1312 Size += EscapedNewLineSize;
1313 Ptr += EscapedNewLineSize;
1314
1315 // If the char that we finally got was a \n, then we must have had
1316 // something like \<newline><newline>. We don't want to consume the
1317 // second newline.
1318 if (*Ptr == '\n' || *Ptr == '\r' || *Ptr == '\0')
1319 return ' ';
1320
1321 // Use slow version to accumulate a correct size field.
1322 return getCharAndSizeSlowNoWarn(Ptr, Size, LangOpts);
1323 }
1324
1325 // Otherwise, this is not an escaped newline, just return the slash.
1326 return '\\';
1327 }
1328
1329 // If this is a trigraph, process it.
1330 if (LangOpts.Trigraphs && Ptr[0] == '?' && Ptr[1] == '?') {
1331 // If this is actually a legal trigraph (not something like "??x"), return
1332 // it.
1333 if (char C = GetTrigraphCharForLetter(Ptr[2])) {
1334 Ptr += 3;
1335 Size += 3;
1336 if (C == '\\') goto Slash;
1337 return C;
1338 }
1339 }
1340
1341 // If this is neither, return a single character.
1342 ++Size;
1343 return *Ptr;
1344 }
1345
1346 //===----------------------------------------------------------------------===//
1347 // Helper methods for lexing.
1348 //===----------------------------------------------------------------------===//
1349
1350 /// \brief Routine that indiscriminately skips bytes in the source file.
SkipBytes(unsigned Bytes,bool StartOfLine)1351 void Lexer::SkipBytes(unsigned Bytes, bool StartOfLine) {
1352 BufferPtr += Bytes;
1353 if (BufferPtr > BufferEnd)
1354 BufferPtr = BufferEnd;
1355 // FIXME: What exactly does the StartOfLine bit mean? There are two
1356 // possible meanings for the "start" of the line: the first token on the
1357 // unexpanded line, or the first token on the expanded line.
1358 IsAtStartOfLine = StartOfLine;
1359 IsAtPhysicalStartOfLine = StartOfLine;
1360 }
1361
isAllowedIDChar(uint32_t C,const LangOptions & LangOpts)1362 static bool isAllowedIDChar(uint32_t C, const LangOptions &LangOpts) {
1363 if (LangOpts.CPlusPlus11 || LangOpts.C11) {
1364 static const llvm::sys::UnicodeCharSet C11AllowedIDChars(
1365 C11AllowedIDCharRanges);
1366 return C11AllowedIDChars.contains(C);
1367 } else if (LangOpts.CPlusPlus) {
1368 static const llvm::sys::UnicodeCharSet CXX03AllowedIDChars(
1369 CXX03AllowedIDCharRanges);
1370 return CXX03AllowedIDChars.contains(C);
1371 } else {
1372 static const llvm::sys::UnicodeCharSet C99AllowedIDChars(
1373 C99AllowedIDCharRanges);
1374 return C99AllowedIDChars.contains(C);
1375 }
1376 }
1377
isAllowedInitiallyIDChar(uint32_t C,const LangOptions & LangOpts)1378 static bool isAllowedInitiallyIDChar(uint32_t C, const LangOptions &LangOpts) {
1379 assert(isAllowedIDChar(C, LangOpts));
1380 if (LangOpts.CPlusPlus11 || LangOpts.C11) {
1381 static const llvm::sys::UnicodeCharSet C11DisallowedInitialIDChars(
1382 C11DisallowedInitialIDCharRanges);
1383 return !C11DisallowedInitialIDChars.contains(C);
1384 } else if (LangOpts.CPlusPlus) {
1385 return true;
1386 } else {
1387 static const llvm::sys::UnicodeCharSet C99DisallowedInitialIDChars(
1388 C99DisallowedInitialIDCharRanges);
1389 return !C99DisallowedInitialIDChars.contains(C);
1390 }
1391 }
1392
makeCharRange(Lexer & L,const char * Begin,const char * End)1393 static inline CharSourceRange makeCharRange(Lexer &L, const char *Begin,
1394 const char *End) {
1395 return CharSourceRange::getCharRange(L.getSourceLocation(Begin),
1396 L.getSourceLocation(End));
1397 }
1398
maybeDiagnoseIDCharCompat(DiagnosticsEngine & Diags,uint32_t C,CharSourceRange Range,bool IsFirst)1399 static void maybeDiagnoseIDCharCompat(DiagnosticsEngine &Diags, uint32_t C,
1400 CharSourceRange Range, bool IsFirst) {
1401 // Check C99 compatibility.
1402 if (!Diags.isIgnored(diag::warn_c99_compat_unicode_id, Range.getBegin())) {
1403 enum {
1404 CannotAppearInIdentifier = 0,
1405 CannotStartIdentifier
1406 };
1407
1408 static const llvm::sys::UnicodeCharSet C99AllowedIDChars(
1409 C99AllowedIDCharRanges);
1410 static const llvm::sys::UnicodeCharSet C99DisallowedInitialIDChars(
1411 C99DisallowedInitialIDCharRanges);
1412 if (!C99AllowedIDChars.contains(C)) {
1413 Diags.Report(Range.getBegin(), diag::warn_c99_compat_unicode_id)
1414 << Range
1415 << CannotAppearInIdentifier;
1416 } else if (IsFirst && C99DisallowedInitialIDChars.contains(C)) {
1417 Diags.Report(Range.getBegin(), diag::warn_c99_compat_unicode_id)
1418 << Range
1419 << CannotStartIdentifier;
1420 }
1421 }
1422
1423 // Check C++98 compatibility.
1424 if (!Diags.isIgnored(diag::warn_cxx98_compat_unicode_id, Range.getBegin())) {
1425 static const llvm::sys::UnicodeCharSet CXX03AllowedIDChars(
1426 CXX03AllowedIDCharRanges);
1427 if (!CXX03AllowedIDChars.contains(C)) {
1428 Diags.Report(Range.getBegin(), diag::warn_cxx98_compat_unicode_id)
1429 << Range;
1430 }
1431 }
1432 }
1433
tryConsumeIdentifierUCN(const char * & CurPtr,unsigned Size,Token & Result)1434 bool Lexer::tryConsumeIdentifierUCN(const char *&CurPtr, unsigned Size,
1435 Token &Result) {
1436 const char *UCNPtr = CurPtr + Size;
1437 uint32_t CodePoint = tryReadUCN(UCNPtr, CurPtr, /*Token=*/nullptr);
1438 if (CodePoint == 0 || !isAllowedIDChar(CodePoint, LangOpts))
1439 return false;
1440
1441 if (!isLexingRawMode())
1442 maybeDiagnoseIDCharCompat(PP->getDiagnostics(), CodePoint,
1443 makeCharRange(*this, CurPtr, UCNPtr),
1444 /*IsFirst=*/false);
1445
1446 Result.setFlag(Token::HasUCN);
1447 if ((UCNPtr - CurPtr == 6 && CurPtr[1] == 'u') ||
1448 (UCNPtr - CurPtr == 10 && CurPtr[1] == 'U'))
1449 CurPtr = UCNPtr;
1450 else
1451 while (CurPtr != UCNPtr)
1452 (void)getAndAdvanceChar(CurPtr, Result);
1453 return true;
1454 }
1455
tryConsumeIdentifierUTF8Char(const char * & CurPtr)1456 bool Lexer::tryConsumeIdentifierUTF8Char(const char *&CurPtr) {
1457 const char *UnicodePtr = CurPtr;
1458 UTF32 CodePoint;
1459 ConversionResult Result =
1460 llvm::convertUTF8Sequence((const UTF8 **)&UnicodePtr,
1461 (const UTF8 *)BufferEnd,
1462 &CodePoint,
1463 strictConversion);
1464 if (Result != conversionOK ||
1465 !isAllowedIDChar(static_cast<uint32_t>(CodePoint), LangOpts))
1466 return false;
1467
1468 if (!isLexingRawMode())
1469 maybeDiagnoseIDCharCompat(PP->getDiagnostics(), CodePoint,
1470 makeCharRange(*this, CurPtr, UnicodePtr),
1471 /*IsFirst=*/false);
1472
1473 CurPtr = UnicodePtr;
1474 return true;
1475 }
1476
LexIdentifier(Token & Result,const char * CurPtr)1477 bool Lexer::LexIdentifier(Token &Result, const char *CurPtr) {
1478 // Match [_A-Za-z0-9]*, we have already matched [_A-Za-z$]
1479 unsigned Size;
1480 unsigned char C = *CurPtr++;
1481 while (isIdentifierBody(C))
1482 C = *CurPtr++;
1483
1484 --CurPtr; // Back up over the skipped character.
1485
1486 // Fast path, no $,\,? in identifier found. '\' might be an escaped newline
1487 // or UCN, and ? might be a trigraph for '\', an escaped newline or UCN.
1488 //
1489 // TODO: Could merge these checks into an InfoTable flag to make the
1490 // comparison cheaper
1491 if (isASCII(C) && C != '\\' && C != '?' &&
1492 (C != '$' || !LangOpts.DollarIdents)) {
1493 FinishIdentifier:
1494 const char *IdStart = BufferPtr;
1495 FormTokenWithChars(Result, CurPtr, tok::raw_identifier);
1496 Result.setRawIdentifierData(IdStart);
1497
1498 // If we are in raw mode, return this identifier raw. There is no need to
1499 // look up identifier information or attempt to macro expand it.
1500 if (LexingRawMode)
1501 return true;
1502
1503 // Fill in Result.IdentifierInfo and update the token kind,
1504 // looking up the identifier in the identifier table.
1505 IdentifierInfo *II = PP->LookUpIdentifierInfo(Result);
1506
1507 // Finally, now that we know we have an identifier, pass this off to the
1508 // preprocessor, which may macro expand it or something.
1509 if (II->isHandleIdentifierCase())
1510 return PP->HandleIdentifier(Result);
1511
1512 return true;
1513 }
1514
1515 // Otherwise, $,\,? in identifier found. Enter slower path.
1516
1517 C = getCharAndSize(CurPtr, Size);
1518 while (1) {
1519 if (C == '$') {
1520 // If we hit a $ and they are not supported in identifiers, we are done.
1521 if (!LangOpts.DollarIdents) goto FinishIdentifier;
1522
1523 // Otherwise, emit a diagnostic and continue.
1524 if (!isLexingRawMode())
1525 Diag(CurPtr, diag::ext_dollar_in_identifier);
1526 CurPtr = ConsumeChar(CurPtr, Size, Result);
1527 C = getCharAndSize(CurPtr, Size);
1528 continue;
1529
1530 } else if (C == '\\' && tryConsumeIdentifierUCN(CurPtr, Size, Result)) {
1531 C = getCharAndSize(CurPtr, Size);
1532 continue;
1533 } else if (!isASCII(C) && tryConsumeIdentifierUTF8Char(CurPtr)) {
1534 C = getCharAndSize(CurPtr, Size);
1535 continue;
1536 } else if (!isIdentifierBody(C)) {
1537 goto FinishIdentifier;
1538 }
1539
1540 // Otherwise, this character is good, consume it.
1541 CurPtr = ConsumeChar(CurPtr, Size, Result);
1542
1543 C = getCharAndSize(CurPtr, Size);
1544 while (isIdentifierBody(C)) {
1545 CurPtr = ConsumeChar(CurPtr, Size, Result);
1546 C = getCharAndSize(CurPtr, Size);
1547 }
1548 }
1549 }
1550
1551 /// isHexaLiteral - Return true if Start points to a hex constant.
1552 /// in microsoft mode (where this is supposed to be several different tokens).
isHexaLiteral(const char * Start,const LangOptions & LangOpts)1553 bool Lexer::isHexaLiteral(const char *Start, const LangOptions &LangOpts) {
1554 unsigned Size;
1555 char C1 = Lexer::getCharAndSizeNoWarn(Start, Size, LangOpts);
1556 if (C1 != '0')
1557 return false;
1558 char C2 = Lexer::getCharAndSizeNoWarn(Start + Size, Size, LangOpts);
1559 return (C2 == 'x' || C2 == 'X');
1560 }
1561
1562 /// LexNumericConstant - Lex the remainder of a integer or floating point
1563 /// constant. From[-1] is the first character lexed. Return the end of the
1564 /// constant.
LexNumericConstant(Token & Result,const char * CurPtr)1565 bool Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
1566 unsigned Size;
1567 char C = getCharAndSize(CurPtr, Size);
1568 char PrevCh = 0;
1569 while (isPreprocessingNumberBody(C)) {
1570 CurPtr = ConsumeChar(CurPtr, Size, Result);
1571 PrevCh = C;
1572 C = getCharAndSize(CurPtr, Size);
1573 }
1574
1575 // If we fell out, check for a sign, due to 1e+12. If we have one, continue.
1576 if ((C == '-' || C == '+') && (PrevCh == 'E' || PrevCh == 'e')) {
1577 // If we are in Microsoft mode, don't continue if the constant is hex.
1578 // For example, MSVC will accept the following as 3 tokens: 0x1234567e+1
1579 if (!LangOpts.MicrosoftExt || !isHexaLiteral(BufferPtr, LangOpts))
1580 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result));
1581 }
1582
1583 // If we have a hex FP constant, continue.
1584 if ((C == '-' || C == '+') && (PrevCh == 'P' || PrevCh == 'p')) {
1585 // Outside C99, we accept hexadecimal floating point numbers as a
1586 // not-quite-conforming extension. Only do so if this looks like it's
1587 // actually meant to be a hexfloat, and not if it has a ud-suffix.
1588 bool IsHexFloat = true;
1589 if (!LangOpts.C99) {
1590 if (!isHexaLiteral(BufferPtr, LangOpts))
1591 IsHexFloat = false;
1592 else if (std::find(BufferPtr, CurPtr, '_') != CurPtr)
1593 IsHexFloat = false;
1594 }
1595 if (IsHexFloat)
1596 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result));
1597 }
1598
1599 // If we have a digit separator, continue.
1600 if (C == '\'' && getLangOpts().CPlusPlus14) {
1601 unsigned NextSize;
1602 char Next = getCharAndSizeNoWarn(CurPtr + Size, NextSize, getLangOpts());
1603 if (isIdentifierBody(Next)) {
1604 if (!isLexingRawMode())
1605 Diag(CurPtr, diag::warn_cxx11_compat_digit_separator);
1606 CurPtr = ConsumeChar(CurPtr, Size, Result);
1607 CurPtr = ConsumeChar(CurPtr, NextSize, Result);
1608 return LexNumericConstant(Result, CurPtr);
1609 }
1610 }
1611
1612 // If we have a UCN or UTF-8 character (perhaps in a ud-suffix), continue.
1613 if (C == '\\' && tryConsumeIdentifierUCN(CurPtr, Size, Result))
1614 return LexNumericConstant(Result, CurPtr);
1615 if (!isASCII(C) && tryConsumeIdentifierUTF8Char(CurPtr))
1616 return LexNumericConstant(Result, CurPtr);
1617
1618 // Update the location of token as well as BufferPtr.
1619 const char *TokStart = BufferPtr;
1620 FormTokenWithChars(Result, CurPtr, tok::numeric_constant);
1621 Result.setLiteralData(TokStart);
1622 return true;
1623 }
1624
1625 /// LexUDSuffix - Lex the ud-suffix production for user-defined literal suffixes
1626 /// in C++11, or warn on a ud-suffix in C++98.
LexUDSuffix(Token & Result,const char * CurPtr,bool IsStringLiteral)1627 const char *Lexer::LexUDSuffix(Token &Result, const char *CurPtr,
1628 bool IsStringLiteral) {
1629 assert(getLangOpts().CPlusPlus);
1630
1631 // Maximally munch an identifier.
1632 unsigned Size;
1633 char C = getCharAndSize(CurPtr, Size);
1634 bool Consumed = false;
1635
1636 if (!isIdentifierHead(C)) {
1637 if (C == '\\' && tryConsumeIdentifierUCN(CurPtr, Size, Result))
1638 Consumed = true;
1639 else if (!isASCII(C) && tryConsumeIdentifierUTF8Char(CurPtr))
1640 Consumed = true;
1641 else
1642 return CurPtr;
1643 }
1644
1645 if (!getLangOpts().CPlusPlus11) {
1646 if (!isLexingRawMode())
1647 Diag(CurPtr,
1648 C == '_' ? diag::warn_cxx11_compat_user_defined_literal
1649 : diag::warn_cxx11_compat_reserved_user_defined_literal)
1650 << FixItHint::CreateInsertion(getSourceLocation(CurPtr), " ");
1651 return CurPtr;
1652 }
1653
1654 // C++11 [lex.ext]p10, [usrlit.suffix]p1: A program containing a ud-suffix
1655 // that does not start with an underscore is ill-formed. As a conforming
1656 // extension, we treat all such suffixes as if they had whitespace before
1657 // them. We assume a suffix beginning with a UCN or UTF-8 character is more
1658 // likely to be a ud-suffix than a macro, however, and accept that.
1659 if (!Consumed) {
1660 bool IsUDSuffix = false;
1661 if (C == '_')
1662 IsUDSuffix = true;
1663 else if (IsStringLiteral && getLangOpts().CPlusPlus14) {
1664 // In C++1y, we need to look ahead a few characters to see if this is a
1665 // valid suffix for a string literal or a numeric literal (this could be
1666 // the 'operator""if' defining a numeric literal operator).
1667 const unsigned MaxStandardSuffixLength = 3;
1668 char Buffer[MaxStandardSuffixLength] = { C };
1669 unsigned Consumed = Size;
1670 unsigned Chars = 1;
1671 while (true) {
1672 unsigned NextSize;
1673 char Next = getCharAndSizeNoWarn(CurPtr + Consumed, NextSize,
1674 getLangOpts());
1675 if (!isIdentifierBody(Next)) {
1676 // End of suffix. Check whether this is on the whitelist.
1677 IsUDSuffix = (Chars == 1 && Buffer[0] == 's') ||
1678 NumericLiteralParser::isValidUDSuffix(
1679 getLangOpts(), StringRef(Buffer, Chars));
1680 break;
1681 }
1682
1683 if (Chars == MaxStandardSuffixLength)
1684 // Too long: can't be a standard suffix.
1685 break;
1686
1687 Buffer[Chars++] = Next;
1688 Consumed += NextSize;
1689 }
1690 }
1691
1692 if (!IsUDSuffix) {
1693 if (!isLexingRawMode())
1694 Diag(CurPtr, getLangOpts().MSVCCompat
1695 ? diag::ext_ms_reserved_user_defined_literal
1696 : diag::ext_reserved_user_defined_literal)
1697 << FixItHint::CreateInsertion(getSourceLocation(CurPtr), " ");
1698 return CurPtr;
1699 }
1700
1701 CurPtr = ConsumeChar(CurPtr, Size, Result);
1702 }
1703
1704 Result.setFlag(Token::HasUDSuffix);
1705 while (true) {
1706 C = getCharAndSize(CurPtr, Size);
1707 if (isIdentifierBody(C)) { CurPtr = ConsumeChar(CurPtr, Size, Result); }
1708 else if (C == '\\' && tryConsumeIdentifierUCN(CurPtr, Size, Result)) {}
1709 else if (!isASCII(C) && tryConsumeIdentifierUTF8Char(CurPtr)) {}
1710 else break;
1711 }
1712
1713 return CurPtr;
1714 }
1715
1716 /// LexStringLiteral - Lex the remainder of a string literal, after having lexed
1717 /// either " or L" or u8" or u" or U".
LexStringLiteral(Token & Result,const char * CurPtr,tok::TokenKind Kind)1718 bool Lexer::LexStringLiteral(Token &Result, const char *CurPtr,
1719 tok::TokenKind Kind) {
1720 // Does this string contain the \0 character?
1721 const char *NulCharacter = nullptr;
1722
1723 if (!isLexingRawMode() &&
1724 (Kind == tok::utf8_string_literal ||
1725 Kind == tok::utf16_string_literal ||
1726 Kind == tok::utf32_string_literal))
1727 Diag(BufferPtr, getLangOpts().CPlusPlus
1728 ? diag::warn_cxx98_compat_unicode_literal
1729 : diag::warn_c99_compat_unicode_literal);
1730
1731 char C = getAndAdvanceChar(CurPtr, Result);
1732 while (C != '"') {
1733 // Skip escaped characters. Escaped newlines will already be processed by
1734 // getAndAdvanceChar.
1735 if (C == '\\')
1736 C = getAndAdvanceChar(CurPtr, Result);
1737
1738 if (C == '\n' || C == '\r' || // Newline.
1739 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
1740 if (!isLexingRawMode() && !LangOpts.AsmPreprocessor)
1741 Diag(BufferPtr, diag::ext_unterminated_string);
1742 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
1743 return true;
1744 }
1745
1746 if (C == 0) {
1747 if (isCodeCompletionPoint(CurPtr-1)) {
1748 PP->CodeCompleteNaturalLanguage();
1749 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
1750 cutOffLexing();
1751 return true;
1752 }
1753
1754 NulCharacter = CurPtr-1;
1755 }
1756 C = getAndAdvanceChar(CurPtr, Result);
1757 }
1758
1759 // If we are in C++11, lex the optional ud-suffix.
1760 if (getLangOpts().CPlusPlus)
1761 CurPtr = LexUDSuffix(Result, CurPtr, true);
1762
1763 // If a nul character existed in the string, warn about it.
1764 if (NulCharacter && !isLexingRawMode())
1765 Diag(NulCharacter, diag::null_in_string);
1766
1767 // Update the location of the token as well as the BufferPtr instance var.
1768 const char *TokStart = BufferPtr;
1769 FormTokenWithChars(Result, CurPtr, Kind);
1770 Result.setLiteralData(TokStart);
1771 return true;
1772 }
1773
1774 /// LexRawStringLiteral - Lex the remainder of a raw string literal, after
1775 /// having lexed R", LR", u8R", uR", or UR".
LexRawStringLiteral(Token & Result,const char * CurPtr,tok::TokenKind Kind)1776 bool Lexer::LexRawStringLiteral(Token &Result, const char *CurPtr,
1777 tok::TokenKind Kind) {
1778 // This function doesn't use getAndAdvanceChar because C++0x [lex.pptoken]p3:
1779 // Between the initial and final double quote characters of the raw string,
1780 // any transformations performed in phases 1 and 2 (trigraphs,
1781 // universal-character-names, and line splicing) are reverted.
1782
1783 if (!isLexingRawMode())
1784 Diag(BufferPtr, diag::warn_cxx98_compat_raw_string_literal);
1785
1786 unsigned PrefixLen = 0;
1787
1788 while (PrefixLen != 16 && isRawStringDelimBody(CurPtr[PrefixLen]))
1789 ++PrefixLen;
1790
1791 // If the last character was not a '(', then we didn't lex a valid delimiter.
1792 if (CurPtr[PrefixLen] != '(') {
1793 if (!isLexingRawMode()) {
1794 const char *PrefixEnd = &CurPtr[PrefixLen];
1795 if (PrefixLen == 16) {
1796 Diag(PrefixEnd, diag::err_raw_delim_too_long);
1797 } else {
1798 Diag(PrefixEnd, diag::err_invalid_char_raw_delim)
1799 << StringRef(PrefixEnd, 1);
1800 }
1801 }
1802
1803 // Search for the next '"' in hopes of salvaging the lexer. Unfortunately,
1804 // it's possible the '"' was intended to be part of the raw string, but
1805 // there's not much we can do about that.
1806 while (1) {
1807 char C = *CurPtr++;
1808
1809 if (C == '"')
1810 break;
1811 if (C == 0 && CurPtr-1 == BufferEnd) {
1812 --CurPtr;
1813 break;
1814 }
1815 }
1816
1817 FormTokenWithChars(Result, CurPtr, tok::unknown);
1818 return true;
1819 }
1820
1821 // Save prefix and move CurPtr past it
1822 const char *Prefix = CurPtr;
1823 CurPtr += PrefixLen + 1; // skip over prefix and '('
1824
1825 while (1) {
1826 char C = *CurPtr++;
1827
1828 if (C == ')') {
1829 // Check for prefix match and closing quote.
1830 if (strncmp(CurPtr, Prefix, PrefixLen) == 0 && CurPtr[PrefixLen] == '"') {
1831 CurPtr += PrefixLen + 1; // skip over prefix and '"'
1832 break;
1833 }
1834 } else if (C == 0 && CurPtr-1 == BufferEnd) { // End of file.
1835 if (!isLexingRawMode())
1836 Diag(BufferPtr, diag::err_unterminated_raw_string)
1837 << StringRef(Prefix, PrefixLen);
1838 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
1839 return true;
1840 }
1841 }
1842
1843 // If we are in C++11, lex the optional ud-suffix.
1844 if (getLangOpts().CPlusPlus)
1845 CurPtr = LexUDSuffix(Result, CurPtr, true);
1846
1847 // Update the location of token as well as BufferPtr.
1848 const char *TokStart = BufferPtr;
1849 FormTokenWithChars(Result, CurPtr, Kind);
1850 Result.setLiteralData(TokStart);
1851 return true;
1852 }
1853
1854 /// LexAngledStringLiteral - Lex the remainder of an angled string literal,
1855 /// after having lexed the '<' character. This is used for #include filenames.
LexAngledStringLiteral(Token & Result,const char * CurPtr)1856 bool Lexer::LexAngledStringLiteral(Token &Result, const char *CurPtr) {
1857 // Does this string contain the \0 character?
1858 const char *NulCharacter = nullptr;
1859 const char *AfterLessPos = CurPtr;
1860 char C = getAndAdvanceChar(CurPtr, Result);
1861 while (C != '>') {
1862 // Skip escaped characters.
1863 if (C == '\\') {
1864 // Skip the escaped character.
1865 getAndAdvanceChar(CurPtr, Result);
1866 } else if (C == '\n' || C == '\r' || // Newline.
1867 (C == 0 && (CurPtr-1 == BufferEnd || // End of file.
1868 isCodeCompletionPoint(CurPtr-1)))) {
1869 // If the filename is unterminated, then it must just be a lone <
1870 // character. Return this as such.
1871 FormTokenWithChars(Result, AfterLessPos, tok::less);
1872 return true;
1873 } else if (C == 0) {
1874 NulCharacter = CurPtr-1;
1875 }
1876 C = getAndAdvanceChar(CurPtr, Result);
1877 }
1878
1879 // If a nul character existed in the string, warn about it.
1880 if (NulCharacter && !isLexingRawMode())
1881 Diag(NulCharacter, diag::null_in_string);
1882
1883 // Update the location of token as well as BufferPtr.
1884 const char *TokStart = BufferPtr;
1885 FormTokenWithChars(Result, CurPtr, tok::angle_string_literal);
1886 Result.setLiteralData(TokStart);
1887 return true;
1888 }
1889
1890
1891 /// LexCharConstant - Lex the remainder of a character constant, after having
1892 /// lexed either ' or L' or u8' or u' or U'.
LexCharConstant(Token & Result,const char * CurPtr,tok::TokenKind Kind)1893 bool Lexer::LexCharConstant(Token &Result, const char *CurPtr,
1894 tok::TokenKind Kind) {
1895 // Does this character contain the \0 character?
1896 const char *NulCharacter = nullptr;
1897
1898 if (!isLexingRawMode()) {
1899 if (Kind == tok::utf16_char_constant || Kind == tok::utf32_char_constant)
1900 Diag(BufferPtr, getLangOpts().CPlusPlus
1901 ? diag::warn_cxx98_compat_unicode_literal
1902 : diag::warn_c99_compat_unicode_literal);
1903 else if (Kind == tok::utf8_char_constant)
1904 Diag(BufferPtr, diag::warn_cxx14_compat_u8_character_literal);
1905 }
1906
1907 char C = getAndAdvanceChar(CurPtr, Result);
1908 if (C == '\'') {
1909 if (!isLexingRawMode() && !LangOpts.AsmPreprocessor)
1910 Diag(BufferPtr, diag::ext_empty_character);
1911 FormTokenWithChars(Result, CurPtr, tok::unknown);
1912 return true;
1913 }
1914
1915 while (C != '\'') {
1916 // Skip escaped characters.
1917 if (C == '\\')
1918 C = getAndAdvanceChar(CurPtr, Result);
1919
1920 if (C == '\n' || C == '\r' || // Newline.
1921 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
1922 if (!isLexingRawMode() && !LangOpts.AsmPreprocessor)
1923 Diag(BufferPtr, diag::ext_unterminated_char);
1924 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
1925 return true;
1926 }
1927
1928 if (C == 0) {
1929 if (isCodeCompletionPoint(CurPtr-1)) {
1930 PP->CodeCompleteNaturalLanguage();
1931 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
1932 cutOffLexing();
1933 return true;
1934 }
1935
1936 NulCharacter = CurPtr-1;
1937 }
1938 C = getAndAdvanceChar(CurPtr, Result);
1939 }
1940
1941 // If we are in C++11, lex the optional ud-suffix.
1942 if (getLangOpts().CPlusPlus)
1943 CurPtr = LexUDSuffix(Result, CurPtr, false);
1944
1945 // If a nul character existed in the character, warn about it.
1946 if (NulCharacter && !isLexingRawMode())
1947 Diag(NulCharacter, diag::null_in_char);
1948
1949 // Update the location of token as well as BufferPtr.
1950 const char *TokStart = BufferPtr;
1951 FormTokenWithChars(Result, CurPtr, Kind);
1952 Result.setLiteralData(TokStart);
1953 return true;
1954 }
1955
1956 /// SkipWhitespace - Efficiently skip over a series of whitespace characters.
1957 /// Update BufferPtr to point to the next non-whitespace character and return.
1958 ///
1959 /// This method forms a token and returns true if KeepWhitespaceMode is enabled.
1960 ///
SkipWhitespace(Token & Result,const char * CurPtr,bool & TokAtPhysicalStartOfLine)1961 bool Lexer::SkipWhitespace(Token &Result, const char *CurPtr,
1962 bool &TokAtPhysicalStartOfLine) {
1963 // Whitespace - Skip it, then return the token after the whitespace.
1964 bool SawNewline = isVerticalWhitespace(CurPtr[-1]);
1965
1966 unsigned char Char = *CurPtr;
1967
1968 // Skip consecutive spaces efficiently.
1969 while (1) {
1970 // Skip horizontal whitespace very aggressively.
1971 while (isHorizontalWhitespace(Char))
1972 Char = *++CurPtr;
1973
1974 // Otherwise if we have something other than whitespace, we're done.
1975 if (!isVerticalWhitespace(Char))
1976 break;
1977
1978 if (ParsingPreprocessorDirective) {
1979 // End of preprocessor directive line, let LexTokenInternal handle this.
1980 BufferPtr = CurPtr;
1981 return false;
1982 }
1983
1984 // OK, but handle newline.
1985 SawNewline = true;
1986 Char = *++CurPtr;
1987 }
1988
1989 // If the client wants us to return whitespace, return it now.
1990 if (isKeepWhitespaceMode()) {
1991 FormTokenWithChars(Result, CurPtr, tok::unknown);
1992 if (SawNewline) {
1993 IsAtStartOfLine = true;
1994 IsAtPhysicalStartOfLine = true;
1995 }
1996 // FIXME: The next token will not have LeadingSpace set.
1997 return true;
1998 }
1999
2000 // If this isn't immediately after a newline, there is leading space.
2001 char PrevChar = CurPtr[-1];
2002 bool HasLeadingSpace = !isVerticalWhitespace(PrevChar);
2003
2004 Result.setFlagValue(Token::LeadingSpace, HasLeadingSpace);
2005 if (SawNewline) {
2006 Result.setFlag(Token::StartOfLine);
2007 TokAtPhysicalStartOfLine = true;
2008 }
2009
2010 BufferPtr = CurPtr;
2011 return false;
2012 }
2013
2014 /// We have just read the // characters from input. Skip until we find the
2015 /// newline character thats terminate the comment. Then update BufferPtr and
2016 /// return.
2017 ///
2018 /// If we're in KeepCommentMode or any CommentHandler has inserted
2019 /// some tokens, this will store the first token and return true.
SkipLineComment(Token & Result,const char * CurPtr,bool & TokAtPhysicalStartOfLine)2020 bool Lexer::SkipLineComment(Token &Result, const char *CurPtr,
2021 bool &TokAtPhysicalStartOfLine) {
2022 // If Line comments aren't explicitly enabled for this language, emit an
2023 // extension warning.
2024 if (!LangOpts.LineComment && !isLexingRawMode()) {
2025 Diag(BufferPtr, diag::ext_line_comment);
2026
2027 // Mark them enabled so we only emit one warning for this translation
2028 // unit.
2029 LangOpts.LineComment = true;
2030 }
2031
2032 // Scan over the body of the comment. The common case, when scanning, is that
2033 // the comment contains normal ascii characters with nothing interesting in
2034 // them. As such, optimize for this case with the inner loop.
2035 char C;
2036 do {
2037 C = *CurPtr;
2038 // Skip over characters in the fast loop.
2039 while (C != 0 && // Potentially EOF.
2040 C != '\n' && C != '\r') // Newline or DOS-style newline.
2041 C = *++CurPtr;
2042
2043 const char *NextLine = CurPtr;
2044 if (C != 0) {
2045 // We found a newline, see if it's escaped.
2046 const char *EscapePtr = CurPtr-1;
2047 bool HasSpace = false;
2048 while (isHorizontalWhitespace(*EscapePtr)) { // Skip whitespace.
2049 --EscapePtr;
2050 HasSpace = true;
2051 }
2052
2053 if (*EscapePtr == '\\') // Escaped newline.
2054 CurPtr = EscapePtr;
2055 else if (EscapePtr[0] == '/' && EscapePtr[-1] == '?' &&
2056 EscapePtr[-2] == '?') // Trigraph-escaped newline.
2057 CurPtr = EscapePtr-2;
2058 else
2059 break; // This is a newline, we're done.
2060
2061 // If there was space between the backslash and newline, warn about it.
2062 if (HasSpace && !isLexingRawMode())
2063 Diag(EscapePtr, diag::backslash_newline_space);
2064 }
2065
2066 // Otherwise, this is a hard case. Fall back on getAndAdvanceChar to
2067 // properly decode the character. Read it in raw mode to avoid emitting
2068 // diagnostics about things like trigraphs. If we see an escaped newline,
2069 // we'll handle it below.
2070 const char *OldPtr = CurPtr;
2071 bool OldRawMode = isLexingRawMode();
2072 LexingRawMode = true;
2073 C = getAndAdvanceChar(CurPtr, Result);
2074 LexingRawMode = OldRawMode;
2075
2076 // If we only read only one character, then no special handling is needed.
2077 // We're done and can skip forward to the newline.
2078 if (C != 0 && CurPtr == OldPtr+1) {
2079 CurPtr = NextLine;
2080 break;
2081 }
2082
2083 // If we read multiple characters, and one of those characters was a \r or
2084 // \n, then we had an escaped newline within the comment. Emit diagnostic
2085 // unless the next line is also a // comment.
2086 if (CurPtr != OldPtr+1 && C != '/' && CurPtr[0] != '/') {
2087 for (; OldPtr != CurPtr; ++OldPtr)
2088 if (OldPtr[0] == '\n' || OldPtr[0] == '\r') {
2089 // Okay, we found a // comment that ends in a newline, if the next
2090 // line is also a // comment, but has spaces, don't emit a diagnostic.
2091 if (isWhitespace(C)) {
2092 const char *ForwardPtr = CurPtr;
2093 while (isWhitespace(*ForwardPtr)) // Skip whitespace.
2094 ++ForwardPtr;
2095 if (ForwardPtr[0] == '/' && ForwardPtr[1] == '/')
2096 break;
2097 }
2098
2099 if (!isLexingRawMode())
2100 Diag(OldPtr-1, diag::ext_multi_line_line_comment);
2101 break;
2102 }
2103 }
2104
2105 if (CurPtr == BufferEnd+1) {
2106 --CurPtr;
2107 break;
2108 }
2109
2110 if (C == '\0' && isCodeCompletionPoint(CurPtr-1)) {
2111 PP->CodeCompleteNaturalLanguage();
2112 cutOffLexing();
2113 return false;
2114 }
2115
2116 } while (C != '\n' && C != '\r');
2117
2118 // Found but did not consume the newline. Notify comment handlers about the
2119 // comment unless we're in a #if 0 block.
2120 if (PP && !isLexingRawMode() &&
2121 PP->HandleComment(Result, SourceRange(getSourceLocation(BufferPtr),
2122 getSourceLocation(CurPtr)))) {
2123 BufferPtr = CurPtr;
2124 return true; // A token has to be returned.
2125 }
2126
2127 // If we are returning comments as tokens, return this comment as a token.
2128 if (inKeepCommentMode())
2129 return SaveLineComment(Result, CurPtr);
2130
2131 // If we are inside a preprocessor directive and we see the end of line,
2132 // return immediately, so that the lexer can return this as an EOD token.
2133 if (ParsingPreprocessorDirective || CurPtr == BufferEnd) {
2134 BufferPtr = CurPtr;
2135 return false;
2136 }
2137
2138 // Otherwise, eat the \n character. We don't care if this is a \n\r or
2139 // \r\n sequence. This is an efficiency hack (because we know the \n can't
2140 // contribute to another token), it isn't needed for correctness. Note that
2141 // this is ok even in KeepWhitespaceMode, because we would have returned the
2142 /// comment above in that mode.
2143 ++CurPtr;
2144
2145 // The next returned token is at the start of the line.
2146 Result.setFlag(Token::StartOfLine);
2147 TokAtPhysicalStartOfLine = true;
2148 // No leading whitespace seen so far.
2149 Result.clearFlag(Token::LeadingSpace);
2150 BufferPtr = CurPtr;
2151 return false;
2152 }
2153
2154 /// If in save-comment mode, package up this Line comment in an appropriate
2155 /// way and return it.
SaveLineComment(Token & Result,const char * CurPtr)2156 bool Lexer::SaveLineComment(Token &Result, const char *CurPtr) {
2157 // If we're not in a preprocessor directive, just return the // comment
2158 // directly.
2159 FormTokenWithChars(Result, CurPtr, tok::comment);
2160
2161 if (!ParsingPreprocessorDirective || LexingRawMode)
2162 return true;
2163
2164 // If this Line-style comment is in a macro definition, transmogrify it into
2165 // a C-style block comment.
2166 bool Invalid = false;
2167 std::string Spelling = PP->getSpelling(Result, &Invalid);
2168 if (Invalid)
2169 return true;
2170
2171 assert(Spelling[0] == '/' && Spelling[1] == '/' && "Not line comment?");
2172 Spelling[1] = '*'; // Change prefix to "/*".
2173 Spelling += "*/"; // add suffix.
2174
2175 Result.setKind(tok::comment);
2176 PP->CreateString(Spelling, Result,
2177 Result.getLocation(), Result.getLocation());
2178 return true;
2179 }
2180
2181 /// isBlockCommentEndOfEscapedNewLine - Return true if the specified newline
2182 /// character (either \\n or \\r) is part of an escaped newline sequence. Issue
2183 /// a diagnostic if so. We know that the newline is inside of a block comment.
isEndOfBlockCommentWithEscapedNewLine(const char * CurPtr,Lexer * L)2184 static bool isEndOfBlockCommentWithEscapedNewLine(const char *CurPtr,
2185 Lexer *L) {
2186 assert(CurPtr[0] == '\n' || CurPtr[0] == '\r');
2187
2188 // Back up off the newline.
2189 --CurPtr;
2190
2191 // If this is a two-character newline sequence, skip the other character.
2192 if (CurPtr[0] == '\n' || CurPtr[0] == '\r') {
2193 // \n\n or \r\r -> not escaped newline.
2194 if (CurPtr[0] == CurPtr[1])
2195 return false;
2196 // \n\r or \r\n -> skip the newline.
2197 --CurPtr;
2198 }
2199
2200 // If we have horizontal whitespace, skip over it. We allow whitespace
2201 // between the slash and newline.
2202 bool HasSpace = false;
2203 while (isHorizontalWhitespace(*CurPtr) || *CurPtr == 0) {
2204 --CurPtr;
2205 HasSpace = true;
2206 }
2207
2208 // If we have a slash, we know this is an escaped newline.
2209 if (*CurPtr == '\\') {
2210 if (CurPtr[-1] != '*') return false;
2211 } else {
2212 // It isn't a slash, is it the ?? / trigraph?
2213 if (CurPtr[0] != '/' || CurPtr[-1] != '?' || CurPtr[-2] != '?' ||
2214 CurPtr[-3] != '*')
2215 return false;
2216
2217 // This is the trigraph ending the comment. Emit a stern warning!
2218 CurPtr -= 2;
2219
2220 // If no trigraphs are enabled, warn that we ignored this trigraph and
2221 // ignore this * character.
2222 if (!L->getLangOpts().Trigraphs) {
2223 if (!L->isLexingRawMode())
2224 L->Diag(CurPtr, diag::trigraph_ignored_block_comment);
2225 return false;
2226 }
2227 if (!L->isLexingRawMode())
2228 L->Diag(CurPtr, diag::trigraph_ends_block_comment);
2229 }
2230
2231 // Warn about having an escaped newline between the */ characters.
2232 if (!L->isLexingRawMode())
2233 L->Diag(CurPtr, diag::escaped_newline_block_comment_end);
2234
2235 // If there was space between the backslash and newline, warn about it.
2236 if (HasSpace && !L->isLexingRawMode())
2237 L->Diag(CurPtr, diag::backslash_newline_space);
2238
2239 return true;
2240 }
2241
2242 #ifdef __SSE2__
2243 #include <emmintrin.h>
2244 #elif __ALTIVEC__
2245 #include <altivec.h>
2246 #undef bool
2247 #endif
2248
2249 /// We have just read from input the / and * characters that started a comment.
2250 /// Read until we find the * and / characters that terminate the comment.
2251 /// Note that we don't bother decoding trigraphs or escaped newlines in block
2252 /// comments, because they cannot cause the comment to end. The only thing
2253 /// that can happen is the comment could end with an escaped newline between
2254 /// the terminating * and /.
2255 ///
2256 /// If we're in KeepCommentMode or any CommentHandler has inserted
2257 /// some tokens, this will store the first token and return true.
SkipBlockComment(Token & Result,const char * CurPtr,bool & TokAtPhysicalStartOfLine)2258 bool Lexer::SkipBlockComment(Token &Result, const char *CurPtr,
2259 bool &TokAtPhysicalStartOfLine) {
2260 // Scan one character past where we should, looking for a '/' character. Once
2261 // we find it, check to see if it was preceded by a *. This common
2262 // optimization helps people who like to put a lot of * characters in their
2263 // comments.
2264
2265 // The first character we get with newlines and trigraphs skipped to handle
2266 // the degenerate /*/ case below correctly if the * has an escaped newline
2267 // after it.
2268 unsigned CharSize;
2269 unsigned char C = getCharAndSize(CurPtr, CharSize);
2270 CurPtr += CharSize;
2271 if (C == 0 && CurPtr == BufferEnd+1) {
2272 if (!isLexingRawMode())
2273 Diag(BufferPtr, diag::err_unterminated_block_comment);
2274 --CurPtr;
2275
2276 // KeepWhitespaceMode should return this broken comment as a token. Since
2277 // it isn't a well formed comment, just return it as an 'unknown' token.
2278 if (isKeepWhitespaceMode()) {
2279 FormTokenWithChars(Result, CurPtr, tok::unknown);
2280 return true;
2281 }
2282
2283 BufferPtr = CurPtr;
2284 return false;
2285 }
2286
2287 // Check to see if the first character after the '/*' is another /. If so,
2288 // then this slash does not end the block comment, it is part of it.
2289 if (C == '/')
2290 C = *CurPtr++;
2291
2292 while (1) {
2293 // Skip over all non-interesting characters until we find end of buffer or a
2294 // (probably ending) '/' character.
2295 if (CurPtr + 24 < BufferEnd &&
2296 // If there is a code-completion point avoid the fast scan because it
2297 // doesn't check for '\0'.
2298 !(PP && PP->getCodeCompletionFileLoc() == FileLoc)) {
2299 // While not aligned to a 16-byte boundary.
2300 while (C != '/' && ((intptr_t)CurPtr & 0x0F) != 0)
2301 C = *CurPtr++;
2302
2303 if (C == '/') goto FoundSlash;
2304
2305 #ifdef __SSE2__
2306 __m128i Slashes = _mm_set1_epi8('/');
2307 while (CurPtr+16 <= BufferEnd) {
2308 int cmp = _mm_movemask_epi8(_mm_cmpeq_epi8(*(const __m128i*)CurPtr,
2309 Slashes));
2310 if (cmp != 0) {
2311 // Adjust the pointer to point directly after the first slash. It's
2312 // not necessary to set C here, it will be overwritten at the end of
2313 // the outer loop.
2314 CurPtr += llvm::countTrailingZeros<unsigned>(cmp) + 1;
2315 goto FoundSlash;
2316 }
2317 CurPtr += 16;
2318 }
2319 #elif __ALTIVEC__
2320 __vector unsigned char Slashes = {
2321 '/', '/', '/', '/', '/', '/', '/', '/',
2322 '/', '/', '/', '/', '/', '/', '/', '/'
2323 };
2324 while (CurPtr+16 <= BufferEnd &&
2325 !vec_any_eq(*(const vector unsigned char*)CurPtr, Slashes))
2326 CurPtr += 16;
2327 #else
2328 // Scan for '/' quickly. Many block comments are very large.
2329 while (CurPtr[0] != '/' &&
2330 CurPtr[1] != '/' &&
2331 CurPtr[2] != '/' &&
2332 CurPtr[3] != '/' &&
2333 CurPtr+4 < BufferEnd) {
2334 CurPtr += 4;
2335 }
2336 #endif
2337
2338 // It has to be one of the bytes scanned, increment to it and read one.
2339 C = *CurPtr++;
2340 }
2341
2342 // Loop to scan the remainder.
2343 while (C != '/' && C != '\0')
2344 C = *CurPtr++;
2345
2346 if (C == '/') {
2347 FoundSlash:
2348 if (CurPtr[-2] == '*') // We found the final */. We're done!
2349 break;
2350
2351 if ((CurPtr[-2] == '\n' || CurPtr[-2] == '\r')) {
2352 if (isEndOfBlockCommentWithEscapedNewLine(CurPtr-2, this)) {
2353 // We found the final */, though it had an escaped newline between the
2354 // * and /. We're done!
2355 break;
2356 }
2357 }
2358 if (CurPtr[0] == '*' && CurPtr[1] != '/') {
2359 // If this is a /* inside of the comment, emit a warning. Don't do this
2360 // if this is a /*/, which will end the comment. This misses cases with
2361 // embedded escaped newlines, but oh well.
2362 if (!isLexingRawMode())
2363 Diag(CurPtr-1, diag::warn_nested_block_comment);
2364 }
2365 } else if (C == 0 && CurPtr == BufferEnd+1) {
2366 if (!isLexingRawMode())
2367 Diag(BufferPtr, diag::err_unterminated_block_comment);
2368 // Note: the user probably forgot a */. We could continue immediately
2369 // after the /*, but this would involve lexing a lot of what really is the
2370 // comment, which surely would confuse the parser.
2371 --CurPtr;
2372
2373 // KeepWhitespaceMode should return this broken comment as a token. Since
2374 // it isn't a well formed comment, just return it as an 'unknown' token.
2375 if (isKeepWhitespaceMode()) {
2376 FormTokenWithChars(Result, CurPtr, tok::unknown);
2377 return true;
2378 }
2379
2380 BufferPtr = CurPtr;
2381 return false;
2382 } else if (C == '\0' && isCodeCompletionPoint(CurPtr-1)) {
2383 PP->CodeCompleteNaturalLanguage();
2384 cutOffLexing();
2385 return false;
2386 }
2387
2388 C = *CurPtr++;
2389 }
2390
2391 // Notify comment handlers about the comment unless we're in a #if 0 block.
2392 if (PP && !isLexingRawMode() &&
2393 PP->HandleComment(Result, SourceRange(getSourceLocation(BufferPtr),
2394 getSourceLocation(CurPtr)))) {
2395 BufferPtr = CurPtr;
2396 return true; // A token has to be returned.
2397 }
2398
2399 // If we are returning comments as tokens, return this comment as a token.
2400 if (inKeepCommentMode()) {
2401 FormTokenWithChars(Result, CurPtr, tok::comment);
2402 return true;
2403 }
2404
2405 // It is common for the tokens immediately after a /**/ comment to be
2406 // whitespace. Instead of going through the big switch, handle it
2407 // efficiently now. This is safe even in KeepWhitespaceMode because we would
2408 // have already returned above with the comment as a token.
2409 if (isHorizontalWhitespace(*CurPtr)) {
2410 SkipWhitespace(Result, CurPtr+1, TokAtPhysicalStartOfLine);
2411 return false;
2412 }
2413
2414 // Otherwise, just return so that the next character will be lexed as a token.
2415 BufferPtr = CurPtr;
2416 Result.setFlag(Token::LeadingSpace);
2417 return false;
2418 }
2419
2420 //===----------------------------------------------------------------------===//
2421 // Primary Lexing Entry Points
2422 //===----------------------------------------------------------------------===//
2423
2424 /// ReadToEndOfLine - Read the rest of the current preprocessor line as an
2425 /// uninterpreted string. This switches the lexer out of directive mode.
ReadToEndOfLine(SmallVectorImpl<char> * Result)2426 void Lexer::ReadToEndOfLine(SmallVectorImpl<char> *Result) {
2427 assert(ParsingPreprocessorDirective && ParsingFilename == false &&
2428 "Must be in a preprocessing directive!");
2429 Token Tmp;
2430
2431 // CurPtr - Cache BufferPtr in an automatic variable.
2432 const char *CurPtr = BufferPtr;
2433 while (1) {
2434 char Char = getAndAdvanceChar(CurPtr, Tmp);
2435 switch (Char) {
2436 default:
2437 if (Result)
2438 Result->push_back(Char);
2439 break;
2440 case 0: // Null.
2441 // Found end of file?
2442 if (CurPtr-1 != BufferEnd) {
2443 if (isCodeCompletionPoint(CurPtr-1)) {
2444 PP->CodeCompleteNaturalLanguage();
2445 cutOffLexing();
2446 return;
2447 }
2448
2449 // Nope, normal character, continue.
2450 if (Result)
2451 Result->push_back(Char);
2452 break;
2453 }
2454 // FALL THROUGH.
2455 case '\r':
2456 case '\n':
2457 // Okay, we found the end of the line. First, back up past the \0, \r, \n.
2458 assert(CurPtr[-1] == Char && "Trigraphs for newline?");
2459 BufferPtr = CurPtr-1;
2460
2461 // Next, lex the character, which should handle the EOD transition.
2462 Lex(Tmp);
2463 if (Tmp.is(tok::code_completion)) {
2464 if (PP)
2465 PP->CodeCompleteNaturalLanguage();
2466 Lex(Tmp);
2467 }
2468 assert(Tmp.is(tok::eod) && "Unexpected token!");
2469
2470 // Finally, we're done;
2471 return;
2472 }
2473 }
2474 }
2475
2476 /// LexEndOfFile - CurPtr points to the end of this file. Handle this
2477 /// condition, reporting diagnostics and handling other edge cases as required.
2478 /// This returns true if Result contains a token, false if PP.Lex should be
2479 /// called again.
LexEndOfFile(Token & Result,const char * CurPtr)2480 bool Lexer::LexEndOfFile(Token &Result, const char *CurPtr) {
2481 // If we hit the end of the file while parsing a preprocessor directive,
2482 // end the preprocessor directive first. The next token returned will
2483 // then be the end of file.
2484 if (ParsingPreprocessorDirective) {
2485 // Done parsing the "line".
2486 ParsingPreprocessorDirective = false;
2487 // Update the location of token as well as BufferPtr.
2488 FormTokenWithChars(Result, CurPtr, tok::eod);
2489
2490 // Restore comment saving mode, in case it was disabled for directive.
2491 if (PP)
2492 resetExtendedTokenMode();
2493 return true; // Have a token.
2494 }
2495
2496 // If we are in raw mode, return this event as an EOF token. Let the caller
2497 // that put us in raw mode handle the event.
2498 if (isLexingRawMode()) {
2499 Result.startToken();
2500 BufferPtr = BufferEnd;
2501 FormTokenWithChars(Result, BufferEnd, tok::eof);
2502 return true;
2503 }
2504
2505 // Issue diagnostics for unterminated #if and missing newline.
2506
2507 // If we are in a #if directive, emit an error.
2508 while (!ConditionalStack.empty()) {
2509 if (PP->getCodeCompletionFileLoc() != FileLoc)
2510 PP->Diag(ConditionalStack.back().IfLoc,
2511 diag::err_pp_unterminated_conditional);
2512 ConditionalStack.pop_back();
2513 }
2514
2515 // C99 5.1.1.2p2: If the file is non-empty and didn't end in a newline, issue
2516 // a pedwarn.
2517 if (CurPtr != BufferStart && (CurPtr[-1] != '\n' && CurPtr[-1] != '\r')) {
2518 DiagnosticsEngine &Diags = PP->getDiagnostics();
2519 SourceLocation EndLoc = getSourceLocation(BufferEnd);
2520 unsigned DiagID;
2521
2522 if (LangOpts.CPlusPlus11) {
2523 // C++11 [lex.phases] 2.2 p2
2524 // Prefer the C++98 pedantic compatibility warning over the generic,
2525 // non-extension, user-requested "missing newline at EOF" warning.
2526 if (!Diags.isIgnored(diag::warn_cxx98_compat_no_newline_eof, EndLoc)) {
2527 DiagID = diag::warn_cxx98_compat_no_newline_eof;
2528 } else {
2529 DiagID = diag::warn_no_newline_eof;
2530 }
2531 } else {
2532 DiagID = diag::ext_no_newline_eof;
2533 }
2534
2535 Diag(BufferEnd, DiagID)
2536 << FixItHint::CreateInsertion(EndLoc, "\n");
2537 }
2538
2539 BufferPtr = CurPtr;
2540
2541 // Finally, let the preprocessor handle this.
2542 return PP->HandleEndOfFile(Result, isPragmaLexer());
2543 }
2544
2545 /// isNextPPTokenLParen - Return 1 if the next unexpanded token lexed from
2546 /// the specified lexer will return a tok::l_paren token, 0 if it is something
2547 /// else and 2 if there are no more tokens in the buffer controlled by the
2548 /// lexer.
isNextPPTokenLParen()2549 unsigned Lexer::isNextPPTokenLParen() {
2550 assert(!LexingRawMode && "How can we expand a macro from a skipping buffer?");
2551
2552 // Switch to 'skipping' mode. This will ensure that we can lex a token
2553 // without emitting diagnostics, disables macro expansion, and will cause EOF
2554 // to return an EOF token instead of popping the include stack.
2555 LexingRawMode = true;
2556
2557 // Save state that can be changed while lexing so that we can restore it.
2558 const char *TmpBufferPtr = BufferPtr;
2559 bool inPPDirectiveMode = ParsingPreprocessorDirective;
2560 bool atStartOfLine = IsAtStartOfLine;
2561 bool atPhysicalStartOfLine = IsAtPhysicalStartOfLine;
2562 bool leadingSpace = HasLeadingSpace;
2563
2564 Token Tok;
2565 Lex(Tok);
2566
2567 // Restore state that may have changed.
2568 BufferPtr = TmpBufferPtr;
2569 ParsingPreprocessorDirective = inPPDirectiveMode;
2570 HasLeadingSpace = leadingSpace;
2571 IsAtStartOfLine = atStartOfLine;
2572 IsAtPhysicalStartOfLine = atPhysicalStartOfLine;
2573
2574 // Restore the lexer back to non-skipping mode.
2575 LexingRawMode = false;
2576
2577 if (Tok.is(tok::eof))
2578 return 2;
2579 return Tok.is(tok::l_paren);
2580 }
2581
2582 /// \brief Find the end of a version control conflict marker.
FindConflictEnd(const char * CurPtr,const char * BufferEnd,ConflictMarkerKind CMK)2583 static const char *FindConflictEnd(const char *CurPtr, const char *BufferEnd,
2584 ConflictMarkerKind CMK) {
2585 const char *Terminator = CMK == CMK_Perforce ? "<<<<\n" : ">>>>>>>";
2586 size_t TermLen = CMK == CMK_Perforce ? 5 : 7;
2587 StringRef RestOfBuffer(CurPtr+TermLen, BufferEnd-CurPtr-TermLen);
2588 size_t Pos = RestOfBuffer.find(Terminator);
2589 while (Pos != StringRef::npos) {
2590 // Must occur at start of line.
2591 if (Pos == 0 ||
2592 (RestOfBuffer[Pos - 1] != '\r' && RestOfBuffer[Pos - 1] != '\n')) {
2593 RestOfBuffer = RestOfBuffer.substr(Pos+TermLen);
2594 Pos = RestOfBuffer.find(Terminator);
2595 continue;
2596 }
2597 return RestOfBuffer.data()+Pos;
2598 }
2599 return nullptr;
2600 }
2601
2602 /// IsStartOfConflictMarker - If the specified pointer is the start of a version
2603 /// control conflict marker like '<<<<<<<', recognize it as such, emit an error
2604 /// and recover nicely. This returns true if it is a conflict marker and false
2605 /// if not.
IsStartOfConflictMarker(const char * CurPtr)2606 bool Lexer::IsStartOfConflictMarker(const char *CurPtr) {
2607 // Only a conflict marker if it starts at the beginning of a line.
2608 if (CurPtr != BufferStart &&
2609 CurPtr[-1] != '\n' && CurPtr[-1] != '\r')
2610 return false;
2611
2612 // Check to see if we have <<<<<<< or >>>>.
2613 if ((BufferEnd-CurPtr < 8 || StringRef(CurPtr, 7) != "<<<<<<<") &&
2614 (BufferEnd-CurPtr < 6 || StringRef(CurPtr, 5) != ">>>> "))
2615 return false;
2616
2617 // If we have a situation where we don't care about conflict markers, ignore
2618 // it.
2619 if (CurrentConflictMarkerState || isLexingRawMode())
2620 return false;
2621
2622 ConflictMarkerKind Kind = *CurPtr == '<' ? CMK_Normal : CMK_Perforce;
2623
2624 // Check to see if there is an ending marker somewhere in the buffer at the
2625 // start of a line to terminate this conflict marker.
2626 if (FindConflictEnd(CurPtr, BufferEnd, Kind)) {
2627 // We found a match. We are really in a conflict marker.
2628 // Diagnose this, and ignore to the end of line.
2629 Diag(CurPtr, diag::err_conflict_marker);
2630 CurrentConflictMarkerState = Kind;
2631
2632 // Skip ahead to the end of line. We know this exists because the
2633 // end-of-conflict marker starts with \r or \n.
2634 while (*CurPtr != '\r' && *CurPtr != '\n') {
2635 assert(CurPtr != BufferEnd && "Didn't find end of line");
2636 ++CurPtr;
2637 }
2638 BufferPtr = CurPtr;
2639 return true;
2640 }
2641
2642 // No end of conflict marker found.
2643 return false;
2644 }
2645
2646
2647 /// HandleEndOfConflictMarker - If this is a '====' or '||||' or '>>>>', or if
2648 /// it is '<<<<' and the conflict marker started with a '>>>>' marker, then it
2649 /// is the end of a conflict marker. Handle it by ignoring up until the end of
2650 /// the line. This returns true if it is a conflict marker and false if not.
HandleEndOfConflictMarker(const char * CurPtr)2651 bool Lexer::HandleEndOfConflictMarker(const char *CurPtr) {
2652 // Only a conflict marker if it starts at the beginning of a line.
2653 if (CurPtr != BufferStart &&
2654 CurPtr[-1] != '\n' && CurPtr[-1] != '\r')
2655 return false;
2656
2657 // If we have a situation where we don't care about conflict markers, ignore
2658 // it.
2659 if (!CurrentConflictMarkerState || isLexingRawMode())
2660 return false;
2661
2662 // Check to see if we have the marker (4 characters in a row).
2663 for (unsigned i = 1; i != 4; ++i)
2664 if (CurPtr[i] != CurPtr[0])
2665 return false;
2666
2667 // If we do have it, search for the end of the conflict marker. This could
2668 // fail if it got skipped with a '#if 0' or something. Note that CurPtr might
2669 // be the end of conflict marker.
2670 if (const char *End = FindConflictEnd(CurPtr, BufferEnd,
2671 CurrentConflictMarkerState)) {
2672 CurPtr = End;
2673
2674 // Skip ahead to the end of line.
2675 while (CurPtr != BufferEnd && *CurPtr != '\r' && *CurPtr != '\n')
2676 ++CurPtr;
2677
2678 BufferPtr = CurPtr;
2679
2680 // No longer in the conflict marker.
2681 CurrentConflictMarkerState = CMK_None;
2682 return true;
2683 }
2684
2685 return false;
2686 }
2687
isCodeCompletionPoint(const char * CurPtr) const2688 bool Lexer::isCodeCompletionPoint(const char *CurPtr) const {
2689 if (PP && PP->isCodeCompletionEnabled()) {
2690 SourceLocation Loc = FileLoc.getLocWithOffset(CurPtr-BufferStart);
2691 return Loc == PP->getCodeCompletionLoc();
2692 }
2693
2694 return false;
2695 }
2696
tryReadUCN(const char * & StartPtr,const char * SlashLoc,Token * Result)2697 uint32_t Lexer::tryReadUCN(const char *&StartPtr, const char *SlashLoc,
2698 Token *Result) {
2699 unsigned CharSize;
2700 char Kind = getCharAndSize(StartPtr, CharSize);
2701
2702 unsigned NumHexDigits;
2703 if (Kind == 'u')
2704 NumHexDigits = 4;
2705 else if (Kind == 'U')
2706 NumHexDigits = 8;
2707 else
2708 return 0;
2709
2710 if (!LangOpts.CPlusPlus && !LangOpts.C99) {
2711 if (Result && !isLexingRawMode())
2712 Diag(SlashLoc, diag::warn_ucn_not_valid_in_c89);
2713 return 0;
2714 }
2715
2716 const char *CurPtr = StartPtr + CharSize;
2717 const char *KindLoc = &CurPtr[-1];
2718
2719 uint32_t CodePoint = 0;
2720 for (unsigned i = 0; i < NumHexDigits; ++i) {
2721 char C = getCharAndSize(CurPtr, CharSize);
2722
2723 unsigned Value = llvm::hexDigitValue(C);
2724 if (Value == -1U) {
2725 if (Result && !isLexingRawMode()) {
2726 if (i == 0) {
2727 Diag(BufferPtr, diag::warn_ucn_escape_no_digits)
2728 << StringRef(KindLoc, 1);
2729 } else {
2730 Diag(BufferPtr, diag::warn_ucn_escape_incomplete);
2731
2732 // If the user wrote \U1234, suggest a fixit to \u.
2733 if (i == 4 && NumHexDigits == 8) {
2734 CharSourceRange URange = makeCharRange(*this, KindLoc, KindLoc + 1);
2735 Diag(KindLoc, diag::note_ucn_four_not_eight)
2736 << FixItHint::CreateReplacement(URange, "u");
2737 }
2738 }
2739 }
2740
2741 return 0;
2742 }
2743
2744 CodePoint <<= 4;
2745 CodePoint += Value;
2746
2747 CurPtr += CharSize;
2748 }
2749
2750 if (Result) {
2751 Result->setFlag(Token::HasUCN);
2752 if (CurPtr - StartPtr == (ptrdiff_t)NumHexDigits + 2)
2753 StartPtr = CurPtr;
2754 else
2755 while (StartPtr != CurPtr)
2756 (void)getAndAdvanceChar(StartPtr, *Result);
2757 } else {
2758 StartPtr = CurPtr;
2759 }
2760
2761 // Don't apply C family restrictions to UCNs in assembly mode
2762 if (LangOpts.AsmPreprocessor)
2763 return CodePoint;
2764
2765 // C99 6.4.3p2: A universal character name shall not specify a character whose
2766 // short identifier is less than 00A0 other than 0024 ($), 0040 (@), or
2767 // 0060 (`), nor one in the range D800 through DFFF inclusive.)
2768 // C++11 [lex.charset]p2: If the hexadecimal value for a
2769 // universal-character-name corresponds to a surrogate code point (in the
2770 // range 0xD800-0xDFFF, inclusive), the program is ill-formed. Additionally,
2771 // if the hexadecimal value for a universal-character-name outside the
2772 // c-char-sequence, s-char-sequence, or r-char-sequence of a character or
2773 // string literal corresponds to a control character (in either of the
2774 // ranges 0x00-0x1F or 0x7F-0x9F, both inclusive) or to a character in the
2775 // basic source character set, the program is ill-formed.
2776 if (CodePoint < 0xA0) {
2777 if (CodePoint == 0x24 || CodePoint == 0x40 || CodePoint == 0x60)
2778 return CodePoint;
2779
2780 // We don't use isLexingRawMode() here because we need to warn about bad
2781 // UCNs even when skipping preprocessing tokens in a #if block.
2782 if (Result && PP) {
2783 if (CodePoint < 0x20 || CodePoint >= 0x7F)
2784 Diag(BufferPtr, diag::err_ucn_control_character);
2785 else {
2786 char C = static_cast<char>(CodePoint);
2787 Diag(BufferPtr, diag::err_ucn_escape_basic_scs) << StringRef(&C, 1);
2788 }
2789 }
2790
2791 return 0;
2792
2793 } else if (CodePoint >= 0xD800 && CodePoint <= 0xDFFF) {
2794 // C++03 allows UCNs representing surrogate characters. C99 and C++11 don't.
2795 // We don't use isLexingRawMode() here because we need to diagnose bad
2796 // UCNs even when skipping preprocessing tokens in a #if block.
2797 if (Result && PP) {
2798 if (LangOpts.CPlusPlus && !LangOpts.CPlusPlus11)
2799 Diag(BufferPtr, diag::warn_ucn_escape_surrogate);
2800 else
2801 Diag(BufferPtr, diag::err_ucn_escape_invalid);
2802 }
2803 return 0;
2804 }
2805
2806 return CodePoint;
2807 }
2808
CheckUnicodeWhitespace(Token & Result,uint32_t C,const char * CurPtr)2809 bool Lexer::CheckUnicodeWhitespace(Token &Result, uint32_t C,
2810 const char *CurPtr) {
2811 static const llvm::sys::UnicodeCharSet UnicodeWhitespaceChars(
2812 UnicodeWhitespaceCharRanges);
2813 if (!isLexingRawMode() && !PP->isPreprocessedOutput() &&
2814 UnicodeWhitespaceChars.contains(C)) {
2815 Diag(BufferPtr, diag::ext_unicode_whitespace)
2816 << makeCharRange(*this, BufferPtr, CurPtr);
2817
2818 Result.setFlag(Token::LeadingSpace);
2819 return true;
2820 }
2821 return false;
2822 }
2823
LexUnicode(Token & Result,uint32_t C,const char * CurPtr)2824 bool Lexer::LexUnicode(Token &Result, uint32_t C, const char *CurPtr) {
2825 if (isAllowedIDChar(C, LangOpts) && isAllowedInitiallyIDChar(C, LangOpts)) {
2826 if (!isLexingRawMode() && !ParsingPreprocessorDirective &&
2827 !PP->isPreprocessedOutput()) {
2828 maybeDiagnoseIDCharCompat(PP->getDiagnostics(), C,
2829 makeCharRange(*this, BufferPtr, CurPtr),
2830 /*IsFirst=*/true);
2831 }
2832
2833 MIOpt.ReadToken();
2834 return LexIdentifier(Result, CurPtr);
2835 }
2836
2837 if (!isLexingRawMode() && !ParsingPreprocessorDirective &&
2838 !PP->isPreprocessedOutput() &&
2839 !isASCII(*BufferPtr) && !isAllowedIDChar(C, LangOpts)) {
2840 // Non-ASCII characters tend to creep into source code unintentionally.
2841 // Instead of letting the parser complain about the unknown token,
2842 // just drop the character.
2843 // Note that we can /only/ do this when the non-ASCII character is actually
2844 // spelled as Unicode, not written as a UCN. The standard requires that
2845 // we not throw away any possible preprocessor tokens, but there's a
2846 // loophole in the mapping of Unicode characters to basic character set
2847 // characters that allows us to map these particular characters to, say,
2848 // whitespace.
2849 Diag(BufferPtr, diag::err_non_ascii)
2850 << FixItHint::CreateRemoval(makeCharRange(*this, BufferPtr, CurPtr));
2851
2852 BufferPtr = CurPtr;
2853 return false;
2854 }
2855
2856 // Otherwise, we have an explicit UCN or a character that's unlikely to show
2857 // up by accident.
2858 MIOpt.ReadToken();
2859 FormTokenWithChars(Result, CurPtr, tok::unknown);
2860 return true;
2861 }
2862
PropagateLineStartLeadingSpaceInfo(Token & Result)2863 void Lexer::PropagateLineStartLeadingSpaceInfo(Token &Result) {
2864 IsAtStartOfLine = Result.isAtStartOfLine();
2865 HasLeadingSpace = Result.hasLeadingSpace();
2866 HasLeadingEmptyMacro = Result.hasLeadingEmptyMacro();
2867 // Note that this doesn't affect IsAtPhysicalStartOfLine.
2868 }
2869
Lex(Token & Result)2870 bool Lexer::Lex(Token &Result) {
2871 // Start a new token.
2872 Result.startToken();
2873
2874 // Set up misc whitespace flags for LexTokenInternal.
2875 if (IsAtStartOfLine) {
2876 Result.setFlag(Token::StartOfLine);
2877 IsAtStartOfLine = false;
2878 }
2879
2880 if (HasLeadingSpace) {
2881 Result.setFlag(Token::LeadingSpace);
2882 HasLeadingSpace = false;
2883 }
2884
2885 if (HasLeadingEmptyMacro) {
2886 Result.setFlag(Token::LeadingEmptyMacro);
2887 HasLeadingEmptyMacro = false;
2888 }
2889
2890 bool atPhysicalStartOfLine = IsAtPhysicalStartOfLine;
2891 IsAtPhysicalStartOfLine = false;
2892 bool isRawLex = isLexingRawMode();
2893 (void) isRawLex;
2894 bool returnedToken = LexTokenInternal(Result, atPhysicalStartOfLine);
2895 // (After the LexTokenInternal call, the lexer might be destroyed.)
2896 assert((returnedToken || !isRawLex) && "Raw lex must succeed");
2897 return returnedToken;
2898 }
2899
2900 /// LexTokenInternal - This implements a simple C family lexer. It is an
2901 /// extremely performance critical piece of code. This assumes that the buffer
2902 /// has a null character at the end of the file. This returns a preprocessing
2903 /// token, not a normal token, as such, it is an internal interface. It assumes
2904 /// that the Flags of result have been cleared before calling this.
LexTokenInternal(Token & Result,bool TokAtPhysicalStartOfLine)2905 bool Lexer::LexTokenInternal(Token &Result, bool TokAtPhysicalStartOfLine) {
2906 LexNextToken:
2907 // New token, can't need cleaning yet.
2908 Result.clearFlag(Token::NeedsCleaning);
2909 Result.setIdentifierInfo(nullptr);
2910
2911 // CurPtr - Cache BufferPtr in an automatic variable.
2912 const char *CurPtr = BufferPtr;
2913
2914 // Small amounts of horizontal whitespace is very common between tokens.
2915 if ((*CurPtr == ' ') || (*CurPtr == '\t')) {
2916 ++CurPtr;
2917 while ((*CurPtr == ' ') || (*CurPtr == '\t'))
2918 ++CurPtr;
2919
2920 // If we are keeping whitespace and other tokens, just return what we just
2921 // skipped. The next lexer invocation will return the token after the
2922 // whitespace.
2923 if (isKeepWhitespaceMode()) {
2924 FormTokenWithChars(Result, CurPtr, tok::unknown);
2925 // FIXME: The next token will not have LeadingSpace set.
2926 return true;
2927 }
2928
2929 BufferPtr = CurPtr;
2930 Result.setFlag(Token::LeadingSpace);
2931 }
2932
2933 unsigned SizeTmp, SizeTmp2; // Temporaries for use in cases below.
2934
2935 // Read a character, advancing over it.
2936 char Char = getAndAdvanceChar(CurPtr, Result);
2937 tok::TokenKind Kind;
2938
2939 switch (Char) {
2940 case 0: // Null.
2941 // Found end of file?
2942 if (CurPtr-1 == BufferEnd)
2943 return LexEndOfFile(Result, CurPtr-1);
2944
2945 // Check if we are performing code completion.
2946 if (isCodeCompletionPoint(CurPtr-1)) {
2947 // Return the code-completion token.
2948 Result.startToken();
2949 FormTokenWithChars(Result, CurPtr, tok::code_completion);
2950 return true;
2951 }
2952
2953 if (!isLexingRawMode())
2954 Diag(CurPtr-1, diag::null_in_file);
2955 Result.setFlag(Token::LeadingSpace);
2956 if (SkipWhitespace(Result, CurPtr, TokAtPhysicalStartOfLine))
2957 return true; // KeepWhitespaceMode
2958
2959 // We know the lexer hasn't changed, so just try again with this lexer.
2960 // (We manually eliminate the tail call to avoid recursion.)
2961 goto LexNextToken;
2962
2963 case 26: // DOS & CP/M EOF: "^Z".
2964 // If we're in Microsoft extensions mode, treat this as end of file.
2965 if (LangOpts.MicrosoftExt)
2966 return LexEndOfFile(Result, CurPtr-1);
2967
2968 // If Microsoft extensions are disabled, this is just random garbage.
2969 Kind = tok::unknown;
2970 break;
2971
2972 case '\n':
2973 case '\r':
2974 // If we are inside a preprocessor directive and we see the end of line,
2975 // we know we are done with the directive, so return an EOD token.
2976 if (ParsingPreprocessorDirective) {
2977 // Done parsing the "line".
2978 ParsingPreprocessorDirective = false;
2979
2980 // Restore comment saving mode, in case it was disabled for directive.
2981 if (PP)
2982 resetExtendedTokenMode();
2983
2984 // Since we consumed a newline, we are back at the start of a line.
2985 IsAtStartOfLine = true;
2986 IsAtPhysicalStartOfLine = true;
2987
2988 Kind = tok::eod;
2989 break;
2990 }
2991
2992 // No leading whitespace seen so far.
2993 Result.clearFlag(Token::LeadingSpace);
2994
2995 if (SkipWhitespace(Result, CurPtr, TokAtPhysicalStartOfLine))
2996 return true; // KeepWhitespaceMode
2997
2998 // We only saw whitespace, so just try again with this lexer.
2999 // (We manually eliminate the tail call to avoid recursion.)
3000 goto LexNextToken;
3001 case ' ':
3002 case '\t':
3003 case '\f':
3004 case '\v':
3005 SkipHorizontalWhitespace:
3006 Result.setFlag(Token::LeadingSpace);
3007 if (SkipWhitespace(Result, CurPtr, TokAtPhysicalStartOfLine))
3008 return true; // KeepWhitespaceMode
3009
3010 SkipIgnoredUnits:
3011 CurPtr = BufferPtr;
3012
3013 // If the next token is obviously a // or /* */ comment, skip it efficiently
3014 // too (without going through the big switch stmt).
3015 if (CurPtr[0] == '/' && CurPtr[1] == '/' && !inKeepCommentMode() &&
3016 LangOpts.LineComment &&
3017 (LangOpts.CPlusPlus || !LangOpts.TraditionalCPP)) {
3018 if (SkipLineComment(Result, CurPtr+2, TokAtPhysicalStartOfLine))
3019 return true; // There is a token to return.
3020 goto SkipIgnoredUnits;
3021 } else if (CurPtr[0] == '/' && CurPtr[1] == '*' && !inKeepCommentMode()) {
3022 if (SkipBlockComment(Result, CurPtr+2, TokAtPhysicalStartOfLine))
3023 return true; // There is a token to return.
3024 goto SkipIgnoredUnits;
3025 } else if (isHorizontalWhitespace(*CurPtr)) {
3026 goto SkipHorizontalWhitespace;
3027 }
3028 // We only saw whitespace, so just try again with this lexer.
3029 // (We manually eliminate the tail call to avoid recursion.)
3030 goto LexNextToken;
3031
3032 // C99 6.4.4.1: Integer Constants.
3033 // C99 6.4.4.2: Floating Constants.
3034 case '0': case '1': case '2': case '3': case '4':
3035 case '5': case '6': case '7': case '8': case '9':
3036 // Notify MIOpt that we read a non-whitespace/non-comment token.
3037 MIOpt.ReadToken();
3038 return LexNumericConstant(Result, CurPtr);
3039
3040 case 'u': // Identifier (uber) or C11/C++11 UTF-8 or UTF-16 string literal
3041 // Notify MIOpt that we read a non-whitespace/non-comment token.
3042 MIOpt.ReadToken();
3043
3044 if (LangOpts.CPlusPlus11 || LangOpts.C11) {
3045 Char = getCharAndSize(CurPtr, SizeTmp);
3046
3047 // UTF-16 string literal
3048 if (Char == '"')
3049 return LexStringLiteral(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3050 tok::utf16_string_literal);
3051
3052 // UTF-16 character constant
3053 if (Char == '\'')
3054 return LexCharConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3055 tok::utf16_char_constant);
3056
3057 // UTF-16 raw string literal
3058 if (Char == 'R' && LangOpts.CPlusPlus11 &&
3059 getCharAndSize(CurPtr + SizeTmp, SizeTmp2) == '"')
3060 return LexRawStringLiteral(Result,
3061 ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3062 SizeTmp2, Result),
3063 tok::utf16_string_literal);
3064
3065 if (Char == '8') {
3066 char Char2 = getCharAndSize(CurPtr + SizeTmp, SizeTmp2);
3067
3068 // UTF-8 string literal
3069 if (Char2 == '"')
3070 return LexStringLiteral(Result,
3071 ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3072 SizeTmp2, Result),
3073 tok::utf8_string_literal);
3074 if (Char2 == '\'' && LangOpts.CPlusPlus1z)
3075 return LexCharConstant(
3076 Result, ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3077 SizeTmp2, Result),
3078 tok::utf8_char_constant);
3079
3080 if (Char2 == 'R' && LangOpts.CPlusPlus11) {
3081 unsigned SizeTmp3;
3082 char Char3 = getCharAndSize(CurPtr + SizeTmp + SizeTmp2, SizeTmp3);
3083 // UTF-8 raw string literal
3084 if (Char3 == '"') {
3085 return LexRawStringLiteral(Result,
3086 ConsumeChar(ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3087 SizeTmp2, Result),
3088 SizeTmp3, Result),
3089 tok::utf8_string_literal);
3090 }
3091 }
3092 }
3093 }
3094
3095 // treat u like the start of an identifier.
3096 return LexIdentifier(Result, CurPtr);
3097
3098 case 'U': // Identifier (Uber) or C11/C++11 UTF-32 string literal
3099 // Notify MIOpt that we read a non-whitespace/non-comment token.
3100 MIOpt.ReadToken();
3101
3102 if (LangOpts.CPlusPlus11 || LangOpts.C11) {
3103 Char = getCharAndSize(CurPtr, SizeTmp);
3104
3105 // UTF-32 string literal
3106 if (Char == '"')
3107 return LexStringLiteral(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3108 tok::utf32_string_literal);
3109
3110 // UTF-32 character constant
3111 if (Char == '\'')
3112 return LexCharConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3113 tok::utf32_char_constant);
3114
3115 // UTF-32 raw string literal
3116 if (Char == 'R' && LangOpts.CPlusPlus11 &&
3117 getCharAndSize(CurPtr + SizeTmp, SizeTmp2) == '"')
3118 return LexRawStringLiteral(Result,
3119 ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3120 SizeTmp2, Result),
3121 tok::utf32_string_literal);
3122 }
3123
3124 // treat U like the start of an identifier.
3125 return LexIdentifier(Result, CurPtr);
3126
3127 case 'R': // Identifier or C++0x raw string literal
3128 // Notify MIOpt that we read a non-whitespace/non-comment token.
3129 MIOpt.ReadToken();
3130
3131 if (LangOpts.CPlusPlus11) {
3132 Char = getCharAndSize(CurPtr, SizeTmp);
3133
3134 if (Char == '"')
3135 return LexRawStringLiteral(Result,
3136 ConsumeChar(CurPtr, SizeTmp, Result),
3137 tok::string_literal);
3138 }
3139
3140 // treat R like the start of an identifier.
3141 return LexIdentifier(Result, CurPtr);
3142
3143 case 'L': // Identifier (Loony) or wide literal (L'x' or L"xyz").
3144 // Notify MIOpt that we read a non-whitespace/non-comment token.
3145 MIOpt.ReadToken();
3146 Char = getCharAndSize(CurPtr, SizeTmp);
3147
3148 // Wide string literal.
3149 if (Char == '"')
3150 return LexStringLiteral(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3151 tok::wide_string_literal);
3152
3153 // Wide raw string literal.
3154 if (LangOpts.CPlusPlus11 && Char == 'R' &&
3155 getCharAndSize(CurPtr + SizeTmp, SizeTmp2) == '"')
3156 return LexRawStringLiteral(Result,
3157 ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3158 SizeTmp2, Result),
3159 tok::wide_string_literal);
3160
3161 // Wide character constant.
3162 if (Char == '\'')
3163 return LexCharConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3164 tok::wide_char_constant);
3165 // FALL THROUGH, treating L like the start of an identifier.
3166
3167 // C99 6.4.2: Identifiers.
3168 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G':
3169 case 'H': case 'I': case 'J': case 'K': /*'L'*/case 'M': case 'N':
3170 case 'O': case 'P': case 'Q': /*'R'*/case 'S': case 'T': /*'U'*/
3171 case 'V': case 'W': case 'X': case 'Y': case 'Z':
3172 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g':
3173 case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n':
3174 case 'o': case 'p': case 'q': case 'r': case 's': case 't': /*'u'*/
3175 case 'v': case 'w': case 'x': case 'y': case 'z':
3176 case '_':
3177 // Notify MIOpt that we read a non-whitespace/non-comment token.
3178 MIOpt.ReadToken();
3179 return LexIdentifier(Result, CurPtr);
3180
3181 case '$': // $ in identifiers.
3182 if (LangOpts.DollarIdents) {
3183 if (!isLexingRawMode())
3184 Diag(CurPtr-1, diag::ext_dollar_in_identifier);
3185 // Notify MIOpt that we read a non-whitespace/non-comment token.
3186 MIOpt.ReadToken();
3187 return LexIdentifier(Result, CurPtr);
3188 }
3189
3190 Kind = tok::unknown;
3191 break;
3192
3193 // C99 6.4.4: Character Constants.
3194 case '\'':
3195 // Notify MIOpt that we read a non-whitespace/non-comment token.
3196 MIOpt.ReadToken();
3197 return LexCharConstant(Result, CurPtr, tok::char_constant);
3198
3199 // C99 6.4.5: String Literals.
3200 case '"':
3201 // Notify MIOpt that we read a non-whitespace/non-comment token.
3202 MIOpt.ReadToken();
3203 return LexStringLiteral(Result, CurPtr, tok::string_literal);
3204
3205 // C99 6.4.6: Punctuators.
3206 case '?':
3207 Kind = tok::question;
3208 break;
3209 case '[':
3210 Kind = tok::l_square;
3211 break;
3212 case ']':
3213 Kind = tok::r_square;
3214 break;
3215 case '(':
3216 Kind = tok::l_paren;
3217 break;
3218 case ')':
3219 Kind = tok::r_paren;
3220 break;
3221 case '{':
3222 Kind = tok::l_brace;
3223 break;
3224 case '}':
3225 Kind = tok::r_brace;
3226 break;
3227 case '.':
3228 Char = getCharAndSize(CurPtr, SizeTmp);
3229 if (Char >= '0' && Char <= '9') {
3230 // Notify MIOpt that we read a non-whitespace/non-comment token.
3231 MIOpt.ReadToken();
3232
3233 return LexNumericConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result));
3234 } else if (LangOpts.CPlusPlus && Char == '*') {
3235 Kind = tok::periodstar;
3236 CurPtr += SizeTmp;
3237 } else if (Char == '.' &&
3238 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '.') {
3239 Kind = tok::ellipsis;
3240 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3241 SizeTmp2, Result);
3242 } else {
3243 Kind = tok::period;
3244 }
3245 break;
3246 case '&':
3247 Char = getCharAndSize(CurPtr, SizeTmp);
3248 if (Char == '&') {
3249 Kind = tok::ampamp;
3250 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3251 } else if (Char == '=') {
3252 Kind = tok::ampequal;
3253 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3254 } else {
3255 Kind = tok::amp;
3256 }
3257 break;
3258 case '*':
3259 if (getCharAndSize(CurPtr, SizeTmp) == '=') {
3260 Kind = tok::starequal;
3261 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3262 } else {
3263 Kind = tok::star;
3264 }
3265 break;
3266 case '+':
3267 Char = getCharAndSize(CurPtr, SizeTmp);
3268 if (Char == '+') {
3269 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3270 Kind = tok::plusplus;
3271 } else if (Char == '=') {
3272 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3273 Kind = tok::plusequal;
3274 } else {
3275 Kind = tok::plus;
3276 }
3277 break;
3278 case '-':
3279 Char = getCharAndSize(CurPtr, SizeTmp);
3280 if (Char == '-') { // --
3281 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3282 Kind = tok::minusminus;
3283 } else if (Char == '>' && LangOpts.CPlusPlus &&
3284 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '*') { // C++ ->*
3285 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3286 SizeTmp2, Result);
3287 Kind = tok::arrowstar;
3288 } else if (Char == '>') { // ->
3289 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3290 Kind = tok::arrow;
3291 } else if (Char == '=') { // -=
3292 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3293 Kind = tok::minusequal;
3294 } else {
3295 Kind = tok::minus;
3296 }
3297 break;
3298 case '~':
3299 Kind = tok::tilde;
3300 break;
3301 case '!':
3302 if (getCharAndSize(CurPtr, SizeTmp) == '=') {
3303 Kind = tok::exclaimequal;
3304 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3305 } else {
3306 Kind = tok::exclaim;
3307 }
3308 break;
3309 case '/':
3310 // 6.4.9: Comments
3311 Char = getCharAndSize(CurPtr, SizeTmp);
3312 if (Char == '/') { // Line comment.
3313 // Even if Line comments are disabled (e.g. in C89 mode), we generally
3314 // want to lex this as a comment. There is one problem with this though,
3315 // that in one particular corner case, this can change the behavior of the
3316 // resultant program. For example, In "foo //**/ bar", C89 would lex
3317 // this as "foo / bar" and langauges with Line comments would lex it as
3318 // "foo". Check to see if the character after the second slash is a '*'.
3319 // If so, we will lex that as a "/" instead of the start of a comment.
3320 // However, we never do this if we are just preprocessing.
3321 bool TreatAsComment = LangOpts.LineComment &&
3322 (LangOpts.CPlusPlus || !LangOpts.TraditionalCPP);
3323 if (!TreatAsComment)
3324 if (!(PP && PP->isPreprocessedOutput()))
3325 TreatAsComment = getCharAndSize(CurPtr+SizeTmp, SizeTmp2) != '*';
3326
3327 if (TreatAsComment) {
3328 if (SkipLineComment(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3329 TokAtPhysicalStartOfLine))
3330 return true; // There is a token to return.
3331
3332 // It is common for the tokens immediately after a // comment to be
3333 // whitespace (indentation for the next line). Instead of going through
3334 // the big switch, handle it efficiently now.
3335 goto SkipIgnoredUnits;
3336 }
3337 }
3338
3339 if (Char == '*') { // /**/ comment.
3340 if (SkipBlockComment(Result, ConsumeChar(CurPtr, SizeTmp, Result),
3341 TokAtPhysicalStartOfLine))
3342 return true; // There is a token to return.
3343
3344 // We only saw whitespace, so just try again with this lexer.
3345 // (We manually eliminate the tail call to avoid recursion.)
3346 goto LexNextToken;
3347 }
3348
3349 if (Char == '=') {
3350 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3351 Kind = tok::slashequal;
3352 } else {
3353 Kind = tok::slash;
3354 }
3355 break;
3356 case '%':
3357 Char = getCharAndSize(CurPtr, SizeTmp);
3358 if (Char == '=') {
3359 Kind = tok::percentequal;
3360 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3361 } else if (LangOpts.Digraphs && Char == '>') {
3362 Kind = tok::r_brace; // '%>' -> '}'
3363 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3364 } else if (LangOpts.Digraphs && Char == ':') {
3365 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3366 Char = getCharAndSize(CurPtr, SizeTmp);
3367 if (Char == '%' && getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == ':') {
3368 Kind = tok::hashhash; // '%:%:' -> '##'
3369 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3370 SizeTmp2, Result);
3371 } else if (Char == '@' && LangOpts.MicrosoftExt) {// %:@ -> #@ -> Charize
3372 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3373 if (!isLexingRawMode())
3374 Diag(BufferPtr, diag::ext_charize_microsoft);
3375 Kind = tok::hashat;
3376 } else { // '%:' -> '#'
3377 // We parsed a # character. If this occurs at the start of the line,
3378 // it's actually the start of a preprocessing directive. Callback to
3379 // the preprocessor to handle it.
3380 // TODO: -fpreprocessed mode??
3381 if (TokAtPhysicalStartOfLine && !LexingRawMode && !Is_PragmaLexer)
3382 goto HandleDirective;
3383
3384 Kind = tok::hash;
3385 }
3386 } else {
3387 Kind = tok::percent;
3388 }
3389 break;
3390 case '<':
3391 Char = getCharAndSize(CurPtr, SizeTmp);
3392 if (ParsingFilename) {
3393 return LexAngledStringLiteral(Result, CurPtr);
3394 } else if (Char == '<') {
3395 char After = getCharAndSize(CurPtr+SizeTmp, SizeTmp2);
3396 if (After == '=') {
3397 Kind = tok::lesslessequal;
3398 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3399 SizeTmp2, Result);
3400 } else if (After == '<' && IsStartOfConflictMarker(CurPtr-1)) {
3401 // If this is actually a '<<<<<<<' version control conflict marker,
3402 // recognize it as such and recover nicely.
3403 goto LexNextToken;
3404 } else if (After == '<' && HandleEndOfConflictMarker(CurPtr-1)) {
3405 // If this is '<<<<' and we're in a Perforce-style conflict marker,
3406 // ignore it.
3407 goto LexNextToken;
3408 } else if (LangOpts.CUDA && After == '<') {
3409 Kind = tok::lesslessless;
3410 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3411 SizeTmp2, Result);
3412 } else {
3413 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3414 Kind = tok::lessless;
3415 }
3416 } else if (Char == '=') {
3417 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3418 Kind = tok::lessequal;
3419 } else if (LangOpts.Digraphs && Char == ':') { // '<:' -> '['
3420 if (LangOpts.CPlusPlus11 &&
3421 getCharAndSize(CurPtr + SizeTmp, SizeTmp2) == ':') {
3422 // C++0x [lex.pptoken]p3:
3423 // Otherwise, if the next three characters are <:: and the subsequent
3424 // character is neither : nor >, the < is treated as a preprocessor
3425 // token by itself and not as the first character of the alternative
3426 // token <:.
3427 unsigned SizeTmp3;
3428 char After = getCharAndSize(CurPtr + SizeTmp + SizeTmp2, SizeTmp3);
3429 if (After != ':' && After != '>') {
3430 Kind = tok::less;
3431 if (!isLexingRawMode())
3432 Diag(BufferPtr, diag::warn_cxx98_compat_less_colon_colon);
3433 break;
3434 }
3435 }
3436
3437 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3438 Kind = tok::l_square;
3439 } else if (LangOpts.Digraphs && Char == '%') { // '<%' -> '{'
3440 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3441 Kind = tok::l_brace;
3442 } else {
3443 Kind = tok::less;
3444 }
3445 break;
3446 case '>':
3447 Char = getCharAndSize(CurPtr, SizeTmp);
3448 if (Char == '=') {
3449 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3450 Kind = tok::greaterequal;
3451 } else if (Char == '>') {
3452 char After = getCharAndSize(CurPtr+SizeTmp, SizeTmp2);
3453 if (After == '=') {
3454 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3455 SizeTmp2, Result);
3456 Kind = tok::greatergreaterequal;
3457 } else if (After == '>' && IsStartOfConflictMarker(CurPtr-1)) {
3458 // If this is actually a '>>>>' conflict marker, recognize it as such
3459 // and recover nicely.
3460 goto LexNextToken;
3461 } else if (After == '>' && HandleEndOfConflictMarker(CurPtr-1)) {
3462 // If this is '>>>>>>>' and we're in a conflict marker, ignore it.
3463 goto LexNextToken;
3464 } else if (LangOpts.CUDA && After == '>') {
3465 Kind = tok::greatergreatergreater;
3466 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
3467 SizeTmp2, Result);
3468 } else {
3469 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3470 Kind = tok::greatergreater;
3471 }
3472
3473 } else {
3474 Kind = tok::greater;
3475 }
3476 break;
3477 case '^':
3478 Char = getCharAndSize(CurPtr, SizeTmp);
3479 if (Char == '=') {
3480 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3481 Kind = tok::caretequal;
3482 } else {
3483 Kind = tok::caret;
3484 }
3485 break;
3486 case '|':
3487 Char = getCharAndSize(CurPtr, SizeTmp);
3488 if (Char == '=') {
3489 Kind = tok::pipeequal;
3490 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3491 } else if (Char == '|') {
3492 // If this is '|||||||' and we're in a conflict marker, ignore it.
3493 if (CurPtr[1] == '|' && HandleEndOfConflictMarker(CurPtr-1))
3494 goto LexNextToken;
3495 Kind = tok::pipepipe;
3496 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3497 } else {
3498 Kind = tok::pipe;
3499 }
3500 break;
3501 case ':':
3502 Char = getCharAndSize(CurPtr, SizeTmp);
3503 if (LangOpts.Digraphs && Char == '>') {
3504 Kind = tok::r_square; // ':>' -> ']'
3505 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3506 } else if (LangOpts.CPlusPlus && Char == ':') {
3507 Kind = tok::coloncolon;
3508 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3509 } else {
3510 Kind = tok::colon;
3511 }
3512 break;
3513 case ';':
3514 Kind = tok::semi;
3515 break;
3516 case '=':
3517 Char = getCharAndSize(CurPtr, SizeTmp);
3518 if (Char == '=') {
3519 // If this is '====' and we're in a conflict marker, ignore it.
3520 if (CurPtr[1] == '=' && HandleEndOfConflictMarker(CurPtr-1))
3521 goto LexNextToken;
3522
3523 Kind = tok::equalequal;
3524 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3525 } else {
3526 Kind = tok::equal;
3527 }
3528 break;
3529 case ',':
3530 Kind = tok::comma;
3531 break;
3532 case '#':
3533 Char = getCharAndSize(CurPtr, SizeTmp);
3534 if (Char == '#') {
3535 Kind = tok::hashhash;
3536 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3537 } else if (Char == '@' && LangOpts.MicrosoftExt) { // #@ -> Charize
3538 Kind = tok::hashat;
3539 if (!isLexingRawMode())
3540 Diag(BufferPtr, diag::ext_charize_microsoft);
3541 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
3542 } else {
3543 // We parsed a # character. If this occurs at the start of the line,
3544 // it's actually the start of a preprocessing directive. Callback to
3545 // the preprocessor to handle it.
3546 // TODO: -fpreprocessed mode??
3547 if (TokAtPhysicalStartOfLine && !LexingRawMode && !Is_PragmaLexer)
3548 goto HandleDirective;
3549
3550 Kind = tok::hash;
3551 }
3552 break;
3553
3554 case '@':
3555 // Objective C support.
3556 if (CurPtr[-1] == '@' && LangOpts.ObjC1)
3557 Kind = tok::at;
3558 else
3559 Kind = tok::unknown;
3560 break;
3561
3562 // UCNs (C99 6.4.3, C++11 [lex.charset]p2)
3563 case '\\':
3564 if (uint32_t CodePoint = tryReadUCN(CurPtr, BufferPtr, &Result)) {
3565 if (CheckUnicodeWhitespace(Result, CodePoint, CurPtr)) {
3566 if (SkipWhitespace(Result, CurPtr, TokAtPhysicalStartOfLine))
3567 return true; // KeepWhitespaceMode
3568
3569 // We only saw whitespace, so just try again with this lexer.
3570 // (We manually eliminate the tail call to avoid recursion.)
3571 goto LexNextToken;
3572 }
3573
3574 return LexUnicode(Result, CodePoint, CurPtr);
3575 }
3576
3577 Kind = tok::unknown;
3578 break;
3579
3580 default: {
3581 if (isASCII(Char)) {
3582 Kind = tok::unknown;
3583 break;
3584 }
3585
3586 UTF32 CodePoint;
3587
3588 // We can't just reset CurPtr to BufferPtr because BufferPtr may point to
3589 // an escaped newline.
3590 --CurPtr;
3591 ConversionResult Status =
3592 llvm::convertUTF8Sequence((const UTF8 **)&CurPtr,
3593 (const UTF8 *)BufferEnd,
3594 &CodePoint,
3595 strictConversion);
3596 if (Status == conversionOK) {
3597 if (CheckUnicodeWhitespace(Result, CodePoint, CurPtr)) {
3598 if (SkipWhitespace(Result, CurPtr, TokAtPhysicalStartOfLine))
3599 return true; // KeepWhitespaceMode
3600
3601 // We only saw whitespace, so just try again with this lexer.
3602 // (We manually eliminate the tail call to avoid recursion.)
3603 goto LexNextToken;
3604 }
3605 return LexUnicode(Result, CodePoint, CurPtr);
3606 }
3607
3608 if (isLexingRawMode() || ParsingPreprocessorDirective ||
3609 PP->isPreprocessedOutput()) {
3610 ++CurPtr;
3611 Kind = tok::unknown;
3612 break;
3613 }
3614
3615 // Non-ASCII characters tend to creep into source code unintentionally.
3616 // Instead of letting the parser complain about the unknown token,
3617 // just diagnose the invalid UTF-8, then drop the character.
3618 Diag(CurPtr, diag::err_invalid_utf8);
3619
3620 BufferPtr = CurPtr+1;
3621 // We're pretending the character didn't exist, so just try again with
3622 // this lexer.
3623 // (We manually eliminate the tail call to avoid recursion.)
3624 goto LexNextToken;
3625 }
3626 }
3627
3628 // Notify MIOpt that we read a non-whitespace/non-comment token.
3629 MIOpt.ReadToken();
3630
3631 // Update the location of token as well as BufferPtr.
3632 FormTokenWithChars(Result, CurPtr, Kind);
3633 return true;
3634
3635 HandleDirective:
3636 // We parsed a # character and it's the start of a preprocessing directive.
3637
3638 FormTokenWithChars(Result, CurPtr, tok::hash);
3639 PP->HandleDirective(Result);
3640
3641 if (PP->hadModuleLoaderFatalFailure()) {
3642 // With a fatal failure in the module loader, we abort parsing.
3643 assert(Result.is(tok::eof) && "Preprocessor did not set tok:eof");
3644 return true;
3645 }
3646
3647 // We parsed the directive; lex a token with the new state.
3648 return false;
3649 }
3650