1 //===--- Lexer.cpp - C Language Family Lexer ------------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the Lexer and Token interfaces.
12 //===----------------------------------------------------------------------===//
14 // TODO: GCC Diagnostics emitted by the lexer:
15 // PEDWARN: (form feed|vertical tab) in preprocessing directive
17 // Universal characters, unicode, char mapping:
18 // WARNING: `%.*s' is not in NFKC
19 // WARNING: `%.*s' is not in NFC
22 // TODO: Options to support:
23 // -fexec-charset,-fwide-exec-charset
25 //===----------------------------------------------------------------------===//
27 #include "clang/Lex/Lexer.h"
28 #include "clang/Lex/Preprocessor.h"
29 #include "clang/Lex/LexDiagnostic.h"
30 #include "clang/Lex/CodeCompletionHandler.h"
31 #include "clang/Basic/SourceManager.h"
32 #include "llvm/ADT/StringSwitch.h"
33 #include "llvm/Support/Compiler.h"
34 #include "llvm/Support/MemoryBuffer.h"
36 using namespace clang;
38 static void InitCharacterInfo();
40 //===----------------------------------------------------------------------===//
41 // Token Class Implementation
42 //===----------------------------------------------------------------------===//
44 /// isObjCAtKeyword - Return true if we have an ObjC keyword identifier.
45 bool Token::isObjCAtKeyword(tok::ObjCKeywordKind objcKey) const {
46 if (IdentifierInfo *II = getIdentifierInfo())
47 return II->getObjCKeywordID() == objcKey;
51 /// getObjCKeywordID - Return the ObjC keyword kind.
52 tok::ObjCKeywordKind Token::getObjCKeywordID() const {
53 IdentifierInfo *specId = getIdentifierInfo();
54 return specId ? specId->getObjCKeywordID() : tok::objc_not_keyword;
58 //===----------------------------------------------------------------------===//
59 // Lexer Class Implementation
60 //===----------------------------------------------------------------------===//
62 void Lexer::InitLexer(const char *BufStart, const char *BufPtr,
66 BufferStart = BufStart;
70 assert(BufEnd[0] == 0 &&
71 "We assume that the input buffer has a null character at the end"
72 " to simplify lexing!");
74 // Check whether we have a BOM in the beginning of the buffer. If yes - act
75 // accordingly. Right now we support only UTF-8 with and without BOM, so, just
76 // skip the UTF-8 BOM if it's present.
77 if (BufferStart == BufferPtr) {
78 // Determine the size of the BOM.
79 llvm::StringRef Buf(BufferStart, BufferEnd - BufferStart);
80 size_t BOMLength = llvm::StringSwitch<size_t>(Buf)
81 .StartsWith("\xEF\xBB\xBF", 3) // UTF-8 BOM
85 BufferPtr += BOMLength;
88 Is_PragmaLexer = false;
89 IsInConflictMarker = false;
91 // Start of the file is a start of line.
92 IsAtStartOfLine = true;
94 // We are not after parsing a #.
95 ParsingPreprocessorDirective = false;
97 // We are not after parsing #include.
98 ParsingFilename = false;
100 // We are not in raw mode. Raw mode disables diagnostics and interpretation
101 // of tokens (e.g. identifiers, thus disabling macro expansion). It is used
102 // to quickly lex the tokens of the buffer, e.g. when handling a "#if 0" block
103 // or otherwise skipping over tokens.
104 LexingRawMode = false;
106 // Default to not keeping comments.
107 ExtendedTokenMode = 0;
110 /// Lexer constructor - Create a new lexer object for the specified buffer
111 /// with the specified preprocessor managing the lexing process. This lexer
112 /// assumes that the associated file buffer and Preprocessor objects will
113 /// outlive it, so it doesn't take ownership of either of them.
114 Lexer::Lexer(FileID FID, const llvm::MemoryBuffer *InputFile, Preprocessor &PP)
115 : PreprocessorLexer(&PP, FID),
116 FileLoc(PP.getSourceManager().getLocForStartOfFile(FID)),
117 Features(PP.getLangOptions()) {
119 InitLexer(InputFile->getBufferStart(), InputFile->getBufferStart(),
120 InputFile->getBufferEnd());
122 // Default to keeping comments if the preprocessor wants them.
123 SetCommentRetentionState(PP.getCommentRetentionState());
126 /// Lexer constructor - Create a new raw lexer object. This object is only
127 /// suitable for calls to 'LexRawToken'. This lexer assumes that the text
128 /// range will outlive it, so it doesn't take ownership of it.
129 Lexer::Lexer(SourceLocation fileloc, const LangOptions &features,
130 const char *BufStart, const char *BufPtr, const char *BufEnd)
131 : FileLoc(fileloc), Features(features) {
133 InitLexer(BufStart, BufPtr, BufEnd);
135 // We *are* in raw mode.
136 LexingRawMode = true;
139 /// Lexer constructor - Create a new raw lexer object. This object is only
140 /// suitable for calls to 'LexRawToken'. This lexer assumes that the text
141 /// range will outlive it, so it doesn't take ownership of it.
142 Lexer::Lexer(FileID FID, const llvm::MemoryBuffer *FromFile,
143 const SourceManager &SM, const LangOptions &features)
144 : FileLoc(SM.getLocForStartOfFile(FID)), Features(features) {
146 InitLexer(FromFile->getBufferStart(), FromFile->getBufferStart(),
147 FromFile->getBufferEnd());
149 // We *are* in raw mode.
150 LexingRawMode = true;
153 /// Create_PragmaLexer: Lexer constructor - Create a new lexer object for
154 /// _Pragma expansion. This has a variety of magic semantics that this method
155 /// sets up. It returns a new'd Lexer that must be delete'd when done.
157 /// On entrance to this routine, TokStartLoc is a macro location which has a
158 /// spelling loc that indicates the bytes to be lexed for the token and an
159 /// instantiation location that indicates where all lexed tokens should be
162 /// FIXME: It would really be nice to make _Pragma just be a wrapper around a
163 /// normal lexer that remaps tokens as they fly by. This would require making
164 /// Preprocessor::Lex virtual. Given that, we could just dump in a magic lexer
165 /// interface that could handle this stuff. This would pull GetMappedTokenLoc
166 /// out of the critical path of the lexer!
168 Lexer *Lexer::Create_PragmaLexer(SourceLocation SpellingLoc,
169 SourceLocation InstantiationLocStart,
170 SourceLocation InstantiationLocEnd,
171 unsigned TokLen, Preprocessor &PP) {
172 SourceManager &SM = PP.getSourceManager();
174 // Create the lexer as if we were going to lex the file normally.
175 FileID SpellingFID = SM.getFileID(SpellingLoc);
176 const llvm::MemoryBuffer *InputFile = SM.getBuffer(SpellingFID);
177 Lexer *L = new Lexer(SpellingFID, InputFile, PP);
179 // Now that the lexer is created, change the start/end locations so that we
180 // just lex the subsection of the file that we want. This is lexing from a
182 const char *StrData = SM.getCharacterData(SpellingLoc);
184 L->BufferPtr = StrData;
185 L->BufferEnd = StrData+TokLen;
186 assert(L->BufferEnd[0] == 0 && "Buffer is not nul terminated!");
188 // Set the SourceLocation with the remapping information. This ensures that
189 // GetMappedTokenLoc will remap the tokens as they are lexed.
190 L->FileLoc = SM.createInstantiationLoc(SM.getLocForStartOfFile(SpellingFID),
191 InstantiationLocStart,
192 InstantiationLocEnd, TokLen);
194 // Ensure that the lexer thinks it is inside a directive, so that end \n will
195 // return an EOD token.
196 L->ParsingPreprocessorDirective = true;
198 // This lexer really is for _Pragma.
199 L->Is_PragmaLexer = true;
204 /// Stringify - Convert the specified string into a C string, with surrounding
205 /// ""'s, and with escaped \ and " characters.
206 std::string Lexer::Stringify(const std::string &Str, bool Charify) {
207 std::string Result = Str;
208 char Quote = Charify ? '\'' : '"';
209 for (unsigned i = 0, e = Result.size(); i != e; ++i) {
210 if (Result[i] == '\\' || Result[i] == Quote) {
211 Result.insert(Result.begin()+i, '\\');
218 /// Stringify - Convert the specified string into a C string by escaping '\'
219 /// and " characters. This does not add surrounding ""'s to the string.
220 void Lexer::Stringify(llvm::SmallVectorImpl<char> &Str) {
221 for (unsigned i = 0, e = Str.size(); i != e; ++i) {
222 if (Str[i] == '\\' || Str[i] == '"') {
223 Str.insert(Str.begin()+i, '\\');
229 //===----------------------------------------------------------------------===//
231 //===----------------------------------------------------------------------===//
233 /// getSpelling() - Return the 'spelling' of this token. The spelling of a
234 /// token are the characters used to represent the token in the source file
235 /// after trigraph expansion and escaped-newline folding. In particular, this
236 /// wants to get the true, uncanonicalized, spelling of things like digraphs
238 llvm::StringRef Lexer::getSpelling(SourceLocation loc,
239 llvm::SmallVectorImpl<char> &buffer,
240 const SourceManager &SM,
241 const LangOptions &options,
243 // Break down the source location.
244 std::pair<FileID, unsigned> locInfo = SM.getDecomposedLoc(loc);
246 // Try to the load the file buffer.
247 bool invalidTemp = false;
248 llvm::StringRef file = SM.getBufferData(locInfo.first, &invalidTemp);
250 if (invalid) *invalid = true;
251 return llvm::StringRef();
254 const char *tokenBegin = file.data() + locInfo.second;
256 // Lex from the start of the given location.
257 Lexer lexer(SM.getLocForStartOfFile(locInfo.first), options,
258 file.begin(), tokenBegin, file.end());
260 lexer.LexFromRawLexer(token);
262 unsigned length = token.getLength();
264 // Common case: no need for cleaning.
265 if (!token.needsCleaning())
266 return llvm::StringRef(tokenBegin, length);
268 // Hard case, we need to relex the characters into the string.
270 buffer.reserve(length);
272 for (const char *ti = tokenBegin, *te = ti + length; ti != te; ) {
274 buffer.push_back(Lexer::getCharAndSizeNoWarn(ti, charSize, options));
278 return llvm::StringRef(buffer.data(), buffer.size());
281 /// getSpelling() - Return the 'spelling' of this token. The spelling of a
282 /// token are the characters used to represent the token in the source file
283 /// after trigraph expansion and escaped-newline folding. In particular, this
284 /// wants to get the true, uncanonicalized, spelling of things like digraphs
286 std::string Lexer::getSpelling(const Token &Tok, const SourceManager &SourceMgr,
287 const LangOptions &Features, bool *Invalid) {
288 assert((int)Tok.getLength() >= 0 && "Token character range is bogus!");
290 // If this token contains nothing interesting, return it directly.
291 bool CharDataInvalid = false;
292 const char* TokStart = SourceMgr.getCharacterData(Tok.getLocation(),
295 *Invalid = CharDataInvalid;
297 return std::string();
299 if (!Tok.needsCleaning())
300 return std::string(TokStart, TokStart+Tok.getLength());
303 Result.reserve(Tok.getLength());
305 // Otherwise, hard case, relex the characters into the string.
306 for (const char *Ptr = TokStart, *End = TokStart+Tok.getLength();
309 Result.push_back(Lexer::getCharAndSizeNoWarn(Ptr, CharSize, Features));
312 assert(Result.size() != unsigned(Tok.getLength()) &&
313 "NeedsCleaning flag set on something that didn't need cleaning!");
317 /// getSpelling - This method is used to get the spelling of a token into a
318 /// preallocated buffer, instead of as an std::string. The caller is required
319 /// to allocate enough space for the token, which is guaranteed to be at least
320 /// Tok.getLength() bytes long. The actual length of the token is returned.
322 /// Note that this method may do two possible things: it may either fill in
323 /// the buffer specified with characters, or it may *change the input pointer*
324 /// to point to a constant buffer with the data already in it (avoiding a
325 /// copy). The caller is not allowed to modify the returned buffer pointer
326 /// if an internal buffer is returned.
327 unsigned Lexer::getSpelling(const Token &Tok, const char *&Buffer,
328 const SourceManager &SourceMgr,
329 const LangOptions &Features, bool *Invalid) {
330 assert((int)Tok.getLength() >= 0 && "Token character range is bogus!");
332 const char *TokStart = 0;
333 // NOTE: this has to be checked *before* testing for an IdentifierInfo.
334 if (Tok.is(tok::raw_identifier))
335 TokStart = Tok.getRawIdentifierData();
336 else if (const IdentifierInfo *II = Tok.getIdentifierInfo()) {
337 // Just return the string from the identifier table, which is very quick.
338 Buffer = II->getNameStart();
339 return II->getLength();
342 // NOTE: this can be checked even after testing for an IdentifierInfo.
344 TokStart = Tok.getLiteralData();
347 // Compute the start of the token in the input lexer buffer.
348 bool CharDataInvalid = false;
349 TokStart = SourceMgr.getCharacterData(Tok.getLocation(), &CharDataInvalid);
351 *Invalid = CharDataInvalid;
352 if (CharDataInvalid) {
358 // If this token contains nothing interesting, return it directly.
359 if (!Tok.needsCleaning()) {
361 return Tok.getLength();
364 // Otherwise, hard case, relex the characters into the string.
365 char *OutBuf = const_cast<char*>(Buffer);
366 for (const char *Ptr = TokStart, *End = TokStart+Tok.getLength();
369 *OutBuf++ = Lexer::getCharAndSizeNoWarn(Ptr, CharSize, Features);
372 assert(unsigned(OutBuf-Buffer) != Tok.getLength() &&
373 "NeedsCleaning flag set on something that didn't need cleaning!");
375 return OutBuf-Buffer;
380 static bool isWhitespace(unsigned char c);
382 /// MeasureTokenLength - Relex the token at the specified location and return
383 /// its length in bytes in the input file. If the token needs cleaning (e.g.
384 /// includes a trigraph or an escaped newline) then this count includes bytes
385 /// that are part of that.
386 unsigned Lexer::MeasureTokenLength(SourceLocation Loc,
387 const SourceManager &SM,
388 const LangOptions &LangOpts) {
389 // TODO: this could be special cased for common tokens like identifiers, ')',
390 // etc to make this faster, if it mattered. Just look at StrData[0] to handle
391 // all obviously single-char tokens. This could use
392 // Lexer::isObviouslySimpleCharacter for example to handle identifiers or
395 // If this comes from a macro expansion, we really do want the macro name, not
396 // the token this macro expanded to.
397 Loc = SM.getInstantiationLoc(Loc);
398 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc);
399 bool Invalid = false;
400 llvm::StringRef Buffer = SM.getBufferData(LocInfo.first, &Invalid);
404 const char *StrData = Buffer.data()+LocInfo.second;
406 if (isWhitespace(StrData[0]))
409 // Create a lexer starting at the beginning of this token.
410 Lexer TheLexer(SM.getLocForStartOfFile(LocInfo.first), LangOpts,
411 Buffer.begin(), StrData, Buffer.end());
412 TheLexer.SetCommentRetentionState(true);
414 TheLexer.LexFromRawLexer(TheTok);
415 return TheTok.getLength();
418 SourceLocation Lexer::GetBeginningOfToken(SourceLocation Loc,
419 const SourceManager &SM,
420 const LangOptions &LangOpts) {
421 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc);
422 if (LocInfo.first.isInvalid())
425 bool Invalid = false;
426 llvm::StringRef Buffer = SM.getBufferData(LocInfo.first, &Invalid);
430 // Back up from the current location until we hit the beginning of a line
431 // (or the buffer). We'll relex from that point.
432 const char *BufStart = Buffer.data();
433 if (LocInfo.second >= Buffer.size())
436 const char *StrData = BufStart+LocInfo.second;
437 if (StrData[0] == '\n' || StrData[0] == '\r')
440 const char *LexStart = StrData;
441 while (LexStart != BufStart) {
442 if (LexStart[0] == '\n' || LexStart[0] == '\r') {
450 // Create a lexer starting at the beginning of this token.
451 SourceLocation LexerStartLoc = Loc.getFileLocWithOffset(-LocInfo.second);
452 Lexer TheLexer(LexerStartLoc, LangOpts, BufStart, LexStart, Buffer.end());
453 TheLexer.SetCommentRetentionState(true);
455 // Lex tokens until we find the token that contains the source location.
458 TheLexer.LexFromRawLexer(TheTok);
460 if (TheLexer.getBufferLocation() > StrData) {
461 // Lexing this token has taken the lexer past the source location we're
462 // looking for. If the current token encompasses our source location,
463 // return the beginning of that token.
464 if (TheLexer.getBufferLocation() - TheTok.getLength() <= StrData)
465 return TheTok.getLocation();
467 // We ended up skipping over the source location entirely, which means
468 // that it points into whitespace. We're done here.
471 } while (TheTok.getKind() != tok::eof);
473 // We've passed our source location; just return the original source location.
478 enum PreambleDirectiveKind {
486 std::pair<unsigned, bool>
487 Lexer::ComputePreamble(const llvm::MemoryBuffer *Buffer, unsigned MaxLines) {
488 // Create a lexer starting at the beginning of the file. Note that we use a
489 // "fake" file source location at offset 1 so that the lexer will track our
490 // position within the file.
491 const unsigned StartOffset = 1;
492 SourceLocation StartLoc = SourceLocation::getFromRawEncoding(StartOffset);
493 LangOptions LangOpts;
494 Lexer TheLexer(StartLoc, LangOpts, Buffer->getBufferStart(),
495 Buffer->getBufferStart(), Buffer->getBufferEnd());
497 bool InPreprocessorDirective = false;
500 unsigned IfCount = 0;
504 TheLexer.LexFromRawLexer(TheTok);
506 if (InPreprocessorDirective) {
507 // If we've hit the end of the file, we're done.
508 if (TheTok.getKind() == tok::eof) {
509 InPreprocessorDirective = false;
513 // If we haven't hit the end of the preprocessor directive, skip this
515 if (!TheTok.isAtStartOfLine())
518 // We've passed the end of the preprocessor directive, and will look
519 // at this token again below.
520 InPreprocessorDirective = false;
523 // Keep track of the # of lines in the preamble.
524 if (TheTok.isAtStartOfLine()) {
527 // If we were asked to limit the number of lines in the preamble,
528 // and we're about to exceed that limit, we're done.
529 if (MaxLines && Line >= MaxLines)
533 // Comments are okay; skip over them.
534 if (TheTok.getKind() == tok::comment)
537 if (TheTok.isAtStartOfLine() && TheTok.getKind() == tok::hash) {
538 // This is the start of a preprocessor directive.
539 Token HashTok = TheTok;
540 InPreprocessorDirective = true;
542 // Figure out which direective this is. Since we're lexing raw tokens,
543 // we don't have an identifier table available. Instead, just look at
544 // the raw identifier to recognize and categorize preprocessor directives.
545 TheLexer.LexFromRawLexer(TheTok);
546 if (TheTok.getKind() == tok::raw_identifier && !TheTok.needsCleaning()) {
547 llvm::StringRef Keyword(TheTok.getRawIdentifierData(),
549 PreambleDirectiveKind PDK
550 = llvm::StringSwitch<PreambleDirectiveKind>(Keyword)
551 .Case("include", PDK_Skipped)
552 .Case("__include_macros", PDK_Skipped)
553 .Case("define", PDK_Skipped)
554 .Case("undef", PDK_Skipped)
555 .Case("line", PDK_Skipped)
556 .Case("error", PDK_Skipped)
557 .Case("pragma", PDK_Skipped)
558 .Case("import", PDK_Skipped)
559 .Case("include_next", PDK_Skipped)
560 .Case("warning", PDK_Skipped)
561 .Case("ident", PDK_Skipped)
562 .Case("sccs", PDK_Skipped)
563 .Case("assert", PDK_Skipped)
564 .Case("unassert", PDK_Skipped)
565 .Case("if", PDK_StartIf)
566 .Case("ifdef", PDK_StartIf)
567 .Case("ifndef", PDK_StartIf)
568 .Case("elif", PDK_Skipped)
569 .Case("else", PDK_Skipped)
570 .Case("endif", PDK_EndIf)
571 .Default(PDK_Unknown);
579 IfStartTok = HashTok;
585 // Mismatched #endif. The preamble ends here.
593 // We don't know what this directive is; stop at the '#'.
598 // We only end up here if we didn't recognize the preprocessor
599 // directive or it was one that can't occur in the preamble at this
600 // point. Roll back the current token to the location of the '#'.
601 InPreprocessorDirective = false;
605 // We hit a token that we don't recognize as being in the
606 // "preprocessing only" part of the file, so we're no longer in
611 SourceLocation End = IfCount? IfStartTok.getLocation() : TheTok.getLocation();
612 return std::make_pair(End.getRawEncoding() - StartLoc.getRawEncoding(),
613 IfCount? IfStartTok.isAtStartOfLine()
614 : TheTok.isAtStartOfLine());
618 /// AdvanceToTokenCharacter - Given a location that specifies the start of a
619 /// token, return a new location that specifies a character within the token.
620 SourceLocation Lexer::AdvanceToTokenCharacter(SourceLocation TokStart,
622 const SourceManager &SM,
623 const LangOptions &Features) {
624 // Figure out how many physical characters away the specified instantiation
625 // character is. This needs to take into consideration newlines and
627 bool Invalid = false;
628 const char *TokPtr = SM.getCharacterData(TokStart, &Invalid);
630 // If they request the first char of the token, we're trivially done.
631 if (Invalid || (CharNo == 0 && Lexer::isObviouslySimpleCharacter(*TokPtr)))
634 unsigned PhysOffset = 0;
636 // The usual case is that tokens don't contain anything interesting. Skip
637 // over the uninteresting characters. If a token only consists of simple
638 // chars, this method is extremely fast.
639 while (Lexer::isObviouslySimpleCharacter(*TokPtr)) {
641 return TokStart.getFileLocWithOffset(PhysOffset);
642 ++TokPtr, --CharNo, ++PhysOffset;
645 // If we have a character that may be a trigraph or escaped newline, use a
646 // lexer to parse it correctly.
647 for (; CharNo; --CharNo) {
649 Lexer::getCharAndSizeNoWarn(TokPtr, Size, Features);
654 // Final detail: if we end up on an escaped newline, we want to return the
655 // location of the actual byte of the token. For example foo\<newline>bar
656 // advanced by 3 should return the location of b, not of \\. One compounding
657 // detail of this is that the escape may be made by a trigraph.
658 if (!Lexer::isObviouslySimpleCharacter(*TokPtr))
659 PhysOffset += Lexer::SkipEscapedNewLines(TokPtr)-TokPtr;
661 return TokStart.getFileLocWithOffset(PhysOffset);
664 /// \brief Computes the source location just past the end of the
665 /// token at this source location.
667 /// This routine can be used to produce a source location that
668 /// points just past the end of the token referenced by \p Loc, and
669 /// is generally used when a diagnostic needs to point just after a
670 /// token where it expected something different that it received. If
671 /// the returned source location would not be meaningful (e.g., if
672 /// it points into a macro), this routine returns an invalid
675 /// \param Offset an offset from the end of the token, where the source
676 /// location should refer to. The default offset (0) produces a source
677 /// location pointing just past the end of the token; an offset of 1 produces
678 /// a source location pointing to the last character in the token, etc.
679 SourceLocation Lexer::getLocForEndOfToken(SourceLocation Loc, unsigned Offset,
680 const SourceManager &SM,
681 const LangOptions &Features) {
682 if (Loc.isInvalid() || !Loc.isFileID())
683 return SourceLocation();
685 unsigned Len = Lexer::MeasureTokenLength(Loc, SM, Features);
691 return Loc.getFileLocWithOffset(Len);
694 //===----------------------------------------------------------------------===//
695 // Character information.
696 //===----------------------------------------------------------------------===//
699 CHAR_HORZ_WS = 0x01, // ' ', '\t', '\f', '\v'. Note, no '\0'
700 CHAR_VERT_WS = 0x02, // '\r', '\n'
701 CHAR_LETTER = 0x04, // a-z,A-Z
702 CHAR_NUMBER = 0x08, // 0-9
703 CHAR_UNDER = 0x10, // _
704 CHAR_PERIOD = 0x20 // .
707 // Statically initialize CharInfo table based on ASCII character set
708 // Reference: FreeBSD 7.2 /usr/share/misc/ascii
709 static const unsigned char CharInfo[256] =
711 // 0 NUL 1 SOH 2 STX 3 ETX
712 // 4 EOT 5 ENQ 6 ACK 7 BEL
715 // 8 BS 9 HT 10 NL 11 VT
716 //12 NP 13 CR 14 SO 15 SI
717 0 , CHAR_HORZ_WS, CHAR_VERT_WS, CHAR_HORZ_WS,
718 CHAR_HORZ_WS, CHAR_VERT_WS, 0 , 0 ,
719 //16 DLE 17 DC1 18 DC2 19 DC3
720 //20 DC4 21 NAK 22 SYN 23 ETB
723 //24 CAN 25 EM 26 SUB 27 ESC
724 //28 FS 29 GS 30 RS 31 US
727 //32 SP 33 ! 34 " 35 #
728 //36 $ 37 % 38 & 39 '
729 CHAR_HORZ_WS, 0 , 0 , 0 ,
731 //40 ( 41 ) 42 * 43 +
732 //44 , 45 - 46 . 47 /
734 0 , 0 , CHAR_PERIOD , 0 ,
735 //48 0 49 1 50 2 51 3
736 //52 4 53 5 54 6 55 7
737 CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER ,
738 CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER ,
739 //56 8 57 9 58 : 59 ;
740 //60 < 61 = 62 > 63 ?
741 CHAR_NUMBER , CHAR_NUMBER , 0 , 0 ,
743 //64 @ 65 A 66 B 67 C
744 //68 D 69 E 70 F 71 G
745 0 , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
746 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
747 //72 H 73 I 74 J 75 K
748 //76 L 77 M 78 N 79 O
749 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
750 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
751 //80 P 81 Q 82 R 83 S
752 //84 T 85 U 86 V 87 W
753 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
754 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
755 //88 X 89 Y 90 Z 91 [
756 //92 \ 93 ] 94 ^ 95 _
757 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 0 ,
758 0 , 0 , 0 , CHAR_UNDER ,
759 //96 ` 97 a 98 b 99 c
760 //100 d 101 e 102 f 103 g
761 0 , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
762 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
763 //104 h 105 i 106 j 107 k
764 //108 l 109 m 110 n 111 o
765 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
766 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
767 //112 p 113 q 114 r 115 s
768 //116 t 117 u 118 v 119 w
769 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
770 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER ,
771 //120 x 121 y 122 z 123 {
772 //124 | 125 } 126 ~ 127 DEL
773 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 0 ,
777 static void InitCharacterInfo() {
778 static bool isInited = false;
779 if (isInited) return;
780 // check the statically-initialized CharInfo table
781 assert(CHAR_HORZ_WS == CharInfo[(int)' ']);
782 assert(CHAR_HORZ_WS == CharInfo[(int)'\t']);
783 assert(CHAR_HORZ_WS == CharInfo[(int)'\f']);
784 assert(CHAR_HORZ_WS == CharInfo[(int)'\v']);
785 assert(CHAR_VERT_WS == CharInfo[(int)'\n']);
786 assert(CHAR_VERT_WS == CharInfo[(int)'\r']);
787 assert(CHAR_UNDER == CharInfo[(int)'_']);
788 assert(CHAR_PERIOD == CharInfo[(int)'.']);
789 for (unsigned i = 'a'; i <= 'z'; ++i) {
790 assert(CHAR_LETTER == CharInfo[i]);
791 assert(CHAR_LETTER == CharInfo[i+'A'-'a']);
793 for (unsigned i = '0'; i <= '9'; ++i)
794 assert(CHAR_NUMBER == CharInfo[i]);
800 /// isIdentifierBody - Return true if this is the body character of an
801 /// identifier, which is [a-zA-Z0-9_].
802 static inline bool isIdentifierBody(unsigned char c) {
803 return (CharInfo[c] & (CHAR_LETTER|CHAR_NUMBER|CHAR_UNDER)) ? true : false;
806 /// isHorizontalWhitespace - Return true if this character is horizontal
807 /// whitespace: ' ', '\t', '\f', '\v'. Note that this returns false for '\0'.
808 static inline bool isHorizontalWhitespace(unsigned char c) {
809 return (CharInfo[c] & CHAR_HORZ_WS) ? true : false;
812 /// isWhitespace - Return true if this character is horizontal or vertical
813 /// whitespace: ' ', '\t', '\f', '\v', '\n', '\r'. Note that this returns false
815 static inline bool isWhitespace(unsigned char c) {
816 return (CharInfo[c] & (CHAR_HORZ_WS|CHAR_VERT_WS)) ? true : false;
819 /// isNumberBody - Return true if this is the body character of an
820 /// preprocessing number, which is [a-zA-Z0-9_.].
821 static inline bool isNumberBody(unsigned char c) {
822 return (CharInfo[c] & (CHAR_LETTER|CHAR_NUMBER|CHAR_UNDER|CHAR_PERIOD)) ?
827 //===----------------------------------------------------------------------===//
828 // Diagnostics forwarding code.
829 //===----------------------------------------------------------------------===//
831 /// GetMappedTokenLoc - If lexing out of a 'mapped buffer', where we pretend the
832 /// lexer buffer was all instantiated at a single point, perform the mapping.
833 /// This is currently only used for _Pragma implementation, so it is the slow
834 /// path of the hot getSourceLocation method. Do not allow it to be inlined.
835 static LLVM_ATTRIBUTE_NOINLINE SourceLocation GetMappedTokenLoc(
836 Preprocessor &PP, SourceLocation FileLoc, unsigned CharNo, unsigned TokLen);
837 static SourceLocation GetMappedTokenLoc(Preprocessor &PP,
838 SourceLocation FileLoc,
839 unsigned CharNo, unsigned TokLen) {
840 assert(FileLoc.isMacroID() && "Must be an instantiation");
842 // Otherwise, we're lexing "mapped tokens". This is used for things like
843 // _Pragma handling. Combine the instantiation location of FileLoc with the
844 // spelling location.
845 SourceManager &SM = PP.getSourceManager();
847 // Create a new SLoc which is expanded from Instantiation(FileLoc) but whose
848 // characters come from spelling(FileLoc)+Offset.
849 SourceLocation SpellingLoc = SM.getSpellingLoc(FileLoc);
850 SpellingLoc = SpellingLoc.getFileLocWithOffset(CharNo);
852 // Figure out the expansion loc range, which is the range covered by the
853 // original _Pragma(...) sequence.
854 std::pair<SourceLocation,SourceLocation> II =
855 SM.getImmediateInstantiationRange(FileLoc);
857 return SM.createInstantiationLoc(SpellingLoc, II.first, II.second, TokLen);
860 /// getSourceLocation - Return a source location identifier for the specified
861 /// offset in the current file.
862 SourceLocation Lexer::getSourceLocation(const char *Loc,
863 unsigned TokLen) const {
864 assert(Loc >= BufferStart && Loc <= BufferEnd &&
865 "Location out of range for this buffer!");
867 // In the normal case, we're just lexing from a simple file buffer, return
868 // the file id from FileLoc with the offset specified.
869 unsigned CharNo = Loc-BufferStart;
870 if (FileLoc.isFileID())
871 return FileLoc.getFileLocWithOffset(CharNo);
873 // Otherwise, this is the _Pragma lexer case, which pretends that all of the
874 // tokens are lexed from where the _Pragma was defined.
875 assert(PP && "This doesn't work on raw lexers");
876 return GetMappedTokenLoc(*PP, FileLoc, CharNo, TokLen);
879 /// Diag - Forwarding function for diagnostics. This translate a source
880 /// position in the current buffer into a SourceLocation object for rendering.
881 DiagnosticBuilder Lexer::Diag(const char *Loc, unsigned DiagID) const {
882 return PP->Diag(getSourceLocation(Loc), DiagID);
885 //===----------------------------------------------------------------------===//
886 // Trigraph and Escaped Newline Handling Code.
887 //===----------------------------------------------------------------------===//
889 /// GetTrigraphCharForLetter - Given a character that occurs after a ?? pair,
890 /// return the decoded trigraph letter it corresponds to, or '\0' if nothing.
891 static char GetTrigraphCharForLetter(char Letter) {
894 case '=': return '#';
895 case ')': return ']';
896 case '(': return '[';
897 case '!': return '|';
898 case '\'': return '^';
899 case '>': return '}';
900 case '/': return '\\';
901 case '<': return '{';
902 case '-': return '~';
906 /// DecodeTrigraphChar - If the specified character is a legal trigraph when
907 /// prefixed with ??, emit a trigraph warning. If trigraphs are enabled,
908 /// return the result character. Finally, emit a warning about trigraph use
909 /// whether trigraphs are enabled or not.
910 static char DecodeTrigraphChar(const char *CP, Lexer *L) {
911 char Res = GetTrigraphCharForLetter(*CP);
912 if (!Res || !L) return Res;
914 if (!L->getFeatures().Trigraphs) {
915 if (!L->isLexingRawMode())
916 L->Diag(CP-2, diag::trigraph_ignored);
920 if (!L->isLexingRawMode())
921 L->Diag(CP-2, diag::trigraph_converted) << llvm::StringRef(&Res, 1);
925 /// getEscapedNewLineSize - Return the size of the specified escaped newline,
926 /// or 0 if it is not an escaped newline. P[-1] is known to be a "\" or a
927 /// trigraph equivalent on entry to this function.
928 unsigned Lexer::getEscapedNewLineSize(const char *Ptr) {
930 while (isWhitespace(Ptr[Size])) {
933 if (Ptr[Size-1] != '\n' && Ptr[Size-1] != '\r')
936 // If this is a \r\n or \n\r, skip the other half.
937 if ((Ptr[Size] == '\r' || Ptr[Size] == '\n') &&
938 Ptr[Size-1] != Ptr[Size])
944 // Not an escaped newline, must be a \t or something else.
948 /// SkipEscapedNewLines - If P points to an escaped newline (or a series of
949 /// them), skip over them and return the first non-escaped-newline found,
950 /// otherwise return P.
951 const char *Lexer::SkipEscapedNewLines(const char *P) {
953 const char *AfterEscape;
956 } else if (*P == '?') {
957 // If not a trigraph for escape, bail out.
958 if (P[1] != '?' || P[2] != '/')
965 unsigned NewLineSize = Lexer::getEscapedNewLineSize(AfterEscape);
966 if (NewLineSize == 0) return P;
967 P = AfterEscape+NewLineSize;
972 /// getCharAndSizeSlow - Peek a single 'character' from the specified buffer,
973 /// get its size, and return it. This is tricky in several cases:
974 /// 1. If currently at the start of a trigraph, we warn about the trigraph,
975 /// then either return the trigraph (skipping 3 chars) or the '?',
976 /// depending on whether trigraphs are enabled or not.
977 /// 2. If this is an escaped newline (potentially with whitespace between
978 /// the backslash and newline), implicitly skip the newline and return
979 /// the char after it.
980 /// 3. If this is a UCN, return it. FIXME: C++ UCN's?
982 /// This handles the slow/uncommon case of the getCharAndSize method. Here we
983 /// know that we can accumulate into Size, and that we have already incremented
984 /// Ptr by Size bytes.
986 /// NOTE: When this method is updated, getCharAndSizeSlowNoWarn (below) should
987 /// be updated to match.
989 char Lexer::getCharAndSizeSlow(const char *Ptr, unsigned &Size,
991 // If we have a slash, look for an escaped newline.
992 if (Ptr[0] == '\\') {
996 // Common case, backslash-char where the char is not whitespace.
997 if (!isWhitespace(Ptr[0])) return '\\';
999 // See if we have optional whitespace characters between the slash and
1001 if (unsigned EscapedNewLineSize = getEscapedNewLineSize(Ptr)) {
1002 // Remember that this token needs to be cleaned.
1003 if (Tok) Tok->setFlag(Token::NeedsCleaning);
1005 // Warn if there was whitespace between the backslash and newline.
1006 if (Ptr[0] != '\n' && Ptr[0] != '\r' && Tok && !isLexingRawMode())
1007 Diag(Ptr, diag::backslash_newline_space);
1009 // Found backslash<whitespace><newline>. Parse the char after it.
1010 Size += EscapedNewLineSize;
1011 Ptr += EscapedNewLineSize;
1012 // Use slow version to accumulate a correct size field.
1013 return getCharAndSizeSlow(Ptr, Size, Tok);
1016 // Otherwise, this is not an escaped newline, just return the slash.
1020 // If this is a trigraph, process it.
1021 if (Ptr[0] == '?' && Ptr[1] == '?') {
1022 // If this is actually a legal trigraph (not something like "??x"), emit
1023 // a trigraph warning. If so, and if trigraphs are enabled, return it.
1024 if (char C = DecodeTrigraphChar(Ptr+2, Tok ? this : 0)) {
1025 // Remember that this token needs to be cleaned.
1026 if (Tok) Tok->setFlag(Token::NeedsCleaning);
1030 if (C == '\\') goto Slash;
1035 // If this is neither, return a single character.
1041 /// getCharAndSizeSlowNoWarn - Handle the slow/uncommon case of the
1042 /// getCharAndSizeNoWarn method. Here we know that we can accumulate into Size,
1043 /// and that we have already incremented Ptr by Size bytes.
1045 /// NOTE: When this method is updated, getCharAndSizeSlow (above) should
1046 /// be updated to match.
1047 char Lexer::getCharAndSizeSlowNoWarn(const char *Ptr, unsigned &Size,
1048 const LangOptions &Features) {
1049 // If we have a slash, look for an escaped newline.
1050 if (Ptr[0] == '\\') {
1054 // Common case, backslash-char where the char is not whitespace.
1055 if (!isWhitespace(Ptr[0])) return '\\';
1057 // See if we have optional whitespace characters followed by a newline.
1058 if (unsigned EscapedNewLineSize = getEscapedNewLineSize(Ptr)) {
1059 // Found backslash<whitespace><newline>. Parse the char after it.
1060 Size += EscapedNewLineSize;
1061 Ptr += EscapedNewLineSize;
1063 // Use slow version to accumulate a correct size field.
1064 return getCharAndSizeSlowNoWarn(Ptr, Size, Features);
1067 // Otherwise, this is not an escaped newline, just return the slash.
1071 // If this is a trigraph, process it.
1072 if (Features.Trigraphs && Ptr[0] == '?' && Ptr[1] == '?') {
1073 // If this is actually a legal trigraph (not something like "??x"), return
1075 if (char C = GetTrigraphCharForLetter(Ptr[2])) {
1078 if (C == '\\') goto Slash;
1083 // If this is neither, return a single character.
1088 //===----------------------------------------------------------------------===//
1089 // Helper methods for lexing.
1090 //===----------------------------------------------------------------------===//
1092 /// \brief Routine that indiscriminately skips bytes in the source file.
1093 void Lexer::SkipBytes(unsigned Bytes, bool StartOfLine) {
1095 if (BufferPtr > BufferEnd)
1096 BufferPtr = BufferEnd;
1097 IsAtStartOfLine = StartOfLine;
1100 void Lexer::LexIdentifier(Token &Result, const char *CurPtr) {
1101 // Match [_A-Za-z0-9]*, we have already matched [_A-Za-z$]
1103 unsigned char C = *CurPtr++;
1104 while (isIdentifierBody(C))
1107 --CurPtr; // Back up over the skipped character.
1109 // Fast path, no $,\,? in identifier found. '\' might be an escaped newline
1110 // or UCN, and ? might be a trigraph for '\', an escaped newline or UCN.
1113 // TODO: Could merge these checks into a CharInfo flag to make the comparison
1115 if (C != '\\' && C != '?' && (C != '$' || !Features.DollarIdents)) {
1117 const char *IdStart = BufferPtr;
1118 FormTokenWithChars(Result, CurPtr, tok::raw_identifier);
1119 Result.setRawIdentifierData(IdStart);
1121 // If we are in raw mode, return this identifier raw. There is no need to
1122 // look up identifier information or attempt to macro expand it.
1126 // Fill in Result.IdentifierInfo and update the token kind,
1127 // looking up the identifier in the identifier table.
1128 IdentifierInfo *II = PP->LookUpIdentifierInfo(Result);
1130 // Finally, now that we know we have an identifier, pass this off to the
1131 // preprocessor, which may macro expand it or something.
1132 if (II->isHandleIdentifierCase())
1133 PP->HandleIdentifier(Result);
1137 // Otherwise, $,\,? in identifier found. Enter slower path.
1139 C = getCharAndSize(CurPtr, Size);
1142 // If we hit a $ and they are not supported in identifiers, we are done.
1143 if (!Features.DollarIdents) goto FinishIdentifier;
1145 // Otherwise, emit a diagnostic and continue.
1146 if (!isLexingRawMode())
1147 Diag(CurPtr, diag::ext_dollar_in_identifier);
1148 CurPtr = ConsumeChar(CurPtr, Size, Result);
1149 C = getCharAndSize(CurPtr, Size);
1151 } else if (!isIdentifierBody(C)) { // FIXME: UCNs.
1152 // Found end of identifier.
1153 goto FinishIdentifier;
1156 // Otherwise, this character is good, consume it.
1157 CurPtr = ConsumeChar(CurPtr, Size, Result);
1159 C = getCharAndSize(CurPtr, Size);
1160 while (isIdentifierBody(C)) { // FIXME: UCNs.
1161 CurPtr = ConsumeChar(CurPtr, Size, Result);
1162 C = getCharAndSize(CurPtr, Size);
1167 /// isHexaLiteral - Return true if Start points to a hex constant.
1168 /// in microsoft mode (where this is supposed to be several different tokens).
1169 static bool isHexaLiteral(const char *Start, const LangOptions &Features) {
1171 char C1 = Lexer::getCharAndSizeNoWarn(Start, Size, Features);
1174 char C2 = Lexer::getCharAndSizeNoWarn(Start + Size, Size, Features);
1175 return (C2 == 'x' || C2 == 'X');
1178 /// LexNumericConstant - Lex the remainder of a integer or floating point
1179 /// constant. From[-1] is the first character lexed. Return the end of the
1181 void Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
1183 char C = getCharAndSize(CurPtr, Size);
1185 while (isNumberBody(C)) { // FIXME: UCNs?
1186 CurPtr = ConsumeChar(CurPtr, Size, Result);
1188 C = getCharAndSize(CurPtr, Size);
1191 // If we fell out, check for a sign, due to 1e+12. If we have one, continue.
1192 if ((C == '-' || C == '+') && (PrevCh == 'E' || PrevCh == 'e')) {
1193 // If we are in Microsoft mode, don't continue if the constant is hex.
1194 // For example, MSVC will accept the following as 3 tokens: 0x1234567e+1
1195 if (!Features.Microsoft || !isHexaLiteral(BufferPtr, Features))
1196 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result));
1199 // If we have a hex FP constant, continue.
1200 if ((C == '-' || C == '+') && (PrevCh == 'P' || PrevCh == 'p') &&
1201 !Features.CPlusPlus0x)
1202 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result));
1204 // Update the location of token as well as BufferPtr.
1205 const char *TokStart = BufferPtr;
1206 FormTokenWithChars(Result, CurPtr, tok::numeric_constant);
1207 Result.setLiteralData(TokStart);
1210 /// LexStringLiteral - Lex the remainder of a string literal, after having lexed
1212 void Lexer::LexStringLiteral(Token &Result, const char *CurPtr, bool Wide) {
1213 const char *NulCharacter = 0; // Does this string contain the \0 character?
1215 char C = getAndAdvanceChar(CurPtr, Result);
1217 // Skip escaped characters. Escaped newlines will already be processed by
1218 // getAndAdvanceChar.
1220 C = getAndAdvanceChar(CurPtr, Result);
1222 if (C == '\n' || C == '\r' || // Newline.
1223 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
1224 if (C == 0 && PP && PP->isCodeCompletionFile(FileLoc))
1225 PP->CodeCompleteNaturalLanguage();
1226 else if (!isLexingRawMode() && !Features.AsmPreprocessor)
1227 Diag(BufferPtr, diag::warn_unterminated_string);
1228 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
1233 NulCharacter = CurPtr-1;
1234 C = getAndAdvanceChar(CurPtr, Result);
1237 // If a nul character existed in the string, warn about it.
1238 if (NulCharacter && !isLexingRawMode())
1239 Diag(NulCharacter, diag::null_in_string);
1241 // Update the location of the token as well as the BufferPtr instance var.
1242 const char *TokStart = BufferPtr;
1243 FormTokenWithChars(Result, CurPtr,
1244 Wide ? tok::wide_string_literal : tok::string_literal);
1245 Result.setLiteralData(TokStart);
1248 /// LexAngledStringLiteral - Lex the remainder of an angled string literal,
1249 /// after having lexed the '<' character. This is used for #include filenames.
1250 void Lexer::LexAngledStringLiteral(Token &Result, const char *CurPtr) {
1251 const char *NulCharacter = 0; // Does this string contain the \0 character?
1252 const char *AfterLessPos = CurPtr;
1253 char C = getAndAdvanceChar(CurPtr, Result);
1255 // Skip escaped characters.
1257 // Skip the escaped character.
1258 C = getAndAdvanceChar(CurPtr, Result);
1259 } else if (C == '\n' || C == '\r' || // Newline.
1260 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
1261 // If the filename is unterminated, then it must just be a lone <
1262 // character. Return this as such.
1263 FormTokenWithChars(Result, AfterLessPos, tok::less);
1265 } else if (C == 0) {
1266 NulCharacter = CurPtr-1;
1268 C = getAndAdvanceChar(CurPtr, Result);
1271 // If a nul character existed in the string, warn about it.
1272 if (NulCharacter && !isLexingRawMode())
1273 Diag(NulCharacter, diag::null_in_string);
1275 // Update the location of token as well as BufferPtr.
1276 const char *TokStart = BufferPtr;
1277 FormTokenWithChars(Result, CurPtr, tok::angle_string_literal);
1278 Result.setLiteralData(TokStart);
1282 /// LexCharConstant - Lex the remainder of a character constant, after having
1283 /// lexed either ' or L'.
1284 void Lexer::LexCharConstant(Token &Result, const char *CurPtr) {
1285 const char *NulCharacter = 0; // Does this character contain the \0 character?
1287 char C = getAndAdvanceChar(CurPtr, Result);
1289 if (!isLexingRawMode() && !Features.AsmPreprocessor)
1290 Diag(BufferPtr, diag::err_empty_character);
1291 FormTokenWithChars(Result, CurPtr, tok::unknown);
1296 // Skip escaped characters.
1298 // Skip the escaped character.
1300 C = getAndAdvanceChar(CurPtr, Result);
1301 } else if (C == '\n' || C == '\r' || // Newline.
1302 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
1303 if (C == 0 && PP && PP->isCodeCompletionFile(FileLoc))
1304 PP->CodeCompleteNaturalLanguage();
1305 else if (!isLexingRawMode() && !Features.AsmPreprocessor)
1306 Diag(BufferPtr, diag::warn_unterminated_char);
1307 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
1309 } else if (C == 0) {
1310 NulCharacter = CurPtr-1;
1312 C = getAndAdvanceChar(CurPtr, Result);
1315 // If a nul character existed in the character, warn about it.
1316 if (NulCharacter && !isLexingRawMode())
1317 Diag(NulCharacter, diag::null_in_char);
1319 // Update the location of token as well as BufferPtr.
1320 const char *TokStart = BufferPtr;
1321 FormTokenWithChars(Result, CurPtr, tok::char_constant);
1322 Result.setLiteralData(TokStart);
1325 /// SkipWhitespace - Efficiently skip over a series of whitespace characters.
1326 /// Update BufferPtr to point to the next non-whitespace character and return.
1328 /// This method forms a token and returns true if KeepWhitespaceMode is enabled.
1330 bool Lexer::SkipWhitespace(Token &Result, const char *CurPtr) {
1331 // Whitespace - Skip it, then return the token after the whitespace.
1332 unsigned char Char = *CurPtr; // Skip consequtive spaces efficiently.
1334 // Skip horizontal whitespace very aggressively.
1335 while (isHorizontalWhitespace(Char))
1338 // Otherwise if we have something other than whitespace, we're done.
1339 if (Char != '\n' && Char != '\r')
1342 if (ParsingPreprocessorDirective) {
1343 // End of preprocessor directive line, let LexTokenInternal handle this.
1348 // ok, but handle newline.
1349 // The returned token is at the start of the line.
1350 Result.setFlag(Token::StartOfLine);
1351 // No leading whitespace seen so far.
1352 Result.clearFlag(Token::LeadingSpace);
1356 // If this isn't immediately after a newline, there is leading space.
1357 char PrevChar = CurPtr[-1];
1358 if (PrevChar != '\n' && PrevChar != '\r')
1359 Result.setFlag(Token::LeadingSpace);
1361 // If the client wants us to return whitespace, return it now.
1362 if (isKeepWhitespaceMode()) {
1363 FormTokenWithChars(Result, CurPtr, tok::unknown);
1371 // SkipBCPLComment - We have just read the // characters from input. Skip until
1372 // we find the newline character thats terminate the comment. Then update
1373 /// BufferPtr and return.
1375 /// If we're in KeepCommentMode or any CommentHandler has inserted
1376 /// some tokens, this will store the first token and return true.
1377 bool Lexer::SkipBCPLComment(Token &Result, const char *CurPtr) {
1378 // If BCPL comments aren't explicitly enabled for this language, emit an
1379 // extension warning.
1380 if (!Features.BCPLComment && !isLexingRawMode()) {
1381 Diag(BufferPtr, diag::ext_bcpl_comment);
1383 // Mark them enabled so we only emit one warning for this translation
1385 Features.BCPLComment = true;
1388 // Scan over the body of the comment. The common case, when scanning, is that
1389 // the comment contains normal ascii characters with nothing interesting in
1390 // them. As such, optimize for this case with the inner loop.
1394 // FIXME: Speedup BCPL comment lexing. Just scan for a \n or \r character.
1395 // If we find a \n character, scan backwards, checking to see if it's an
1396 // escaped newline, like we do for block comments.
1398 // Skip over characters in the fast loop.
1399 while (C != 0 && // Potentially EOF.
1400 C != '\\' && // Potentially escaped newline.
1401 C != '?' && // Potentially trigraph.
1402 C != '\n' && C != '\r') // Newline or DOS-style newline.
1405 // If this is a newline, we're done.
1406 if (C == '\n' || C == '\r')
1407 break; // Found the newline? Break out!
1409 // Otherwise, this is a hard case. Fall back on getAndAdvanceChar to
1410 // properly decode the character. Read it in raw mode to avoid emitting
1411 // diagnostics about things like trigraphs. If we see an escaped newline,
1412 // we'll handle it below.
1413 const char *OldPtr = CurPtr;
1414 bool OldRawMode = isLexingRawMode();
1415 LexingRawMode = true;
1416 C = getAndAdvanceChar(CurPtr, Result);
1417 LexingRawMode = OldRawMode;
1419 // If the char that we finally got was a \n, then we must have had something
1420 // like \<newline><newline>. We don't want to have consumed the second
1421 // newline, we want CurPtr, to end up pointing to it down below.
1422 if (C == '\n' || C == '\r') {
1424 C = 'x'; // doesn't matter what this is.
1427 // If we read multiple characters, and one of those characters was a \r or
1428 // \n, then we had an escaped newline within the comment. Emit diagnostic
1429 // unless the next line is also a // comment.
1430 if (CurPtr != OldPtr+1 && C != '/' && CurPtr[0] != '/') {
1431 for (; OldPtr != CurPtr; ++OldPtr)
1432 if (OldPtr[0] == '\n' || OldPtr[0] == '\r') {
1433 // Okay, we found a // comment that ends in a newline, if the next
1434 // line is also a // comment, but has spaces, don't emit a diagnostic.
1436 const char *ForwardPtr = CurPtr;
1437 while (isspace(*ForwardPtr)) // Skip whitespace.
1439 if (ForwardPtr[0] == '/' && ForwardPtr[1] == '/')
1443 if (!isLexingRawMode())
1444 Diag(OldPtr-1, diag::ext_multi_line_bcpl_comment);
1449 if (CurPtr == BufferEnd+1) {
1450 if (PP && PP->isCodeCompletionFile(FileLoc))
1451 PP->CodeCompleteNaturalLanguage();
1456 } while (C != '\n' && C != '\r');
1458 // Found but did not consume the newline. Notify comment handlers about the
1459 // comment unless we're in a #if 0 block.
1460 if (PP && !isLexingRawMode() &&
1461 PP->HandleComment(Result, SourceRange(getSourceLocation(BufferPtr),
1462 getSourceLocation(CurPtr)))) {
1464 return true; // A token has to be returned.
1467 // If we are returning comments as tokens, return this comment as a token.
1468 if (inKeepCommentMode())
1469 return SaveBCPLComment(Result, CurPtr);
1471 // If we are inside a preprocessor directive and we see the end of line,
1472 // return immediately, so that the lexer can return this as an EOD token.
1473 if (ParsingPreprocessorDirective || CurPtr == BufferEnd) {
1478 // Otherwise, eat the \n character. We don't care if this is a \n\r or
1479 // \r\n sequence. This is an efficiency hack (because we know the \n can't
1480 // contribute to another token), it isn't needed for correctness. Note that
1481 // this is ok even in KeepWhitespaceMode, because we would have returned the
1482 /// comment above in that mode.
1485 // The next returned token is at the start of the line.
1486 Result.setFlag(Token::StartOfLine);
1487 // No leading whitespace seen so far.
1488 Result.clearFlag(Token::LeadingSpace);
1493 /// SaveBCPLComment - If in save-comment mode, package up this BCPL comment in
1494 /// an appropriate way and return it.
1495 bool Lexer::SaveBCPLComment(Token &Result, const char *CurPtr) {
1496 // If we're not in a preprocessor directive, just return the // comment
1498 FormTokenWithChars(Result, CurPtr, tok::comment);
1500 if (!ParsingPreprocessorDirective)
1503 // If this BCPL-style comment is in a macro definition, transmogrify it into
1504 // a C-style block comment.
1505 bool Invalid = false;
1506 std::string Spelling = PP->getSpelling(Result, &Invalid);
1510 assert(Spelling[0] == '/' && Spelling[1] == '/' && "Not bcpl comment?");
1511 Spelling[1] = '*'; // Change prefix to "/*".
1512 Spelling += "*/"; // add suffix.
1514 Result.setKind(tok::comment);
1515 PP->CreateString(&Spelling[0], Spelling.size(), Result,
1516 Result.getLocation());
1520 /// isBlockCommentEndOfEscapedNewLine - Return true if the specified newline
1521 /// character (either \n or \r) is part of an escaped newline sequence. Issue a
1522 /// diagnostic if so. We know that the newline is inside of a block comment.
1523 static bool isEndOfBlockCommentWithEscapedNewLine(const char *CurPtr,
1525 assert(CurPtr[0] == '\n' || CurPtr[0] == '\r');
1527 // Back up off the newline.
1530 // If this is a two-character newline sequence, skip the other character.
1531 if (CurPtr[0] == '\n' || CurPtr[0] == '\r') {
1532 // \n\n or \r\r -> not escaped newline.
1533 if (CurPtr[0] == CurPtr[1])
1535 // \n\r or \r\n -> skip the newline.
1539 // If we have horizontal whitespace, skip over it. We allow whitespace
1540 // between the slash and newline.
1541 bool HasSpace = false;
1542 while (isHorizontalWhitespace(*CurPtr) || *CurPtr == 0) {
1547 // If we have a slash, we know this is an escaped newline.
1548 if (*CurPtr == '\\') {
1549 if (CurPtr[-1] != '*') return false;
1551 // It isn't a slash, is it the ?? / trigraph?
1552 if (CurPtr[0] != '/' || CurPtr[-1] != '?' || CurPtr[-2] != '?' ||
1556 // This is the trigraph ending the comment. Emit a stern warning!
1559 // If no trigraphs are enabled, warn that we ignored this trigraph and
1560 // ignore this * character.
1561 if (!L->getFeatures().Trigraphs) {
1562 if (!L->isLexingRawMode())
1563 L->Diag(CurPtr, diag::trigraph_ignored_block_comment);
1566 if (!L->isLexingRawMode())
1567 L->Diag(CurPtr, diag::trigraph_ends_block_comment);
1570 // Warn about having an escaped newline between the */ characters.
1571 if (!L->isLexingRawMode())
1572 L->Diag(CurPtr, diag::escaped_newline_block_comment_end);
1574 // If there was space between the backslash and newline, warn about it.
1575 if (HasSpace && !L->isLexingRawMode())
1576 L->Diag(CurPtr, diag::backslash_newline_space);
1582 #include <emmintrin.h>
1584 #include <altivec.h>
1588 /// SkipBlockComment - We have just read the /* characters from input. Read
1589 /// until we find the */ characters that terminate the comment. Note that we
1590 /// don't bother decoding trigraphs or escaped newlines in block comments,
1591 /// because they cannot cause the comment to end. The only thing that can
1592 /// happen is the comment could end with an escaped newline between the */ end
1595 /// If we're in KeepCommentMode or any CommentHandler has inserted
1596 /// some tokens, this will store the first token and return true.
1597 bool Lexer::SkipBlockComment(Token &Result, const char *CurPtr) {
1598 // Scan one character past where we should, looking for a '/' character. Once
1599 // we find it, check to see if it was preceded by a *. This common
1600 // optimization helps people who like to put a lot of * characters in their
1603 // The first character we get with newlines and trigraphs skipped to handle
1604 // the degenerate /*/ case below correctly if the * has an escaped newline
1607 unsigned char C = getCharAndSize(CurPtr, CharSize);
1609 if (C == 0 && CurPtr == BufferEnd+1) {
1610 if (!isLexingRawMode() &&
1611 !PP->isCodeCompletionFile(FileLoc))
1612 Diag(BufferPtr, diag::err_unterminated_block_comment);
1615 // KeepWhitespaceMode should return this broken comment as a token. Since
1616 // it isn't a well formed comment, just return it as an 'unknown' token.
1617 if (isKeepWhitespaceMode()) {
1618 FormTokenWithChars(Result, CurPtr, tok::unknown);
1626 // Check to see if the first character after the '/*' is another /. If so,
1627 // then this slash does not end the block comment, it is part of it.
1632 // Skip over all non-interesting characters until we find end of buffer or a
1633 // (probably ending) '/' character.
1634 if (CurPtr + 24 < BufferEnd) {
1635 // While not aligned to a 16-byte boundary.
1636 while (C != '/' && ((intptr_t)CurPtr & 0x0F) != 0)
1639 if (C == '/') goto FoundSlash;
1642 __m128i Slashes = _mm_set_epi8('/', '/', '/', '/', '/', '/', '/', '/',
1643 '/', '/', '/', '/', '/', '/', '/', '/');
1644 while (CurPtr+16 <= BufferEnd &&
1645 _mm_movemask_epi8(_mm_cmpeq_epi8(*(__m128i*)CurPtr, Slashes)) == 0)
1648 __vector unsigned char Slashes = {
1649 '/', '/', '/', '/', '/', '/', '/', '/',
1650 '/', '/', '/', '/', '/', '/', '/', '/'
1652 while (CurPtr+16 <= BufferEnd &&
1653 !vec_any_eq(*(vector unsigned char*)CurPtr, Slashes))
1656 // Scan for '/' quickly. Many block comments are very large.
1657 while (CurPtr[0] != '/' &&
1661 CurPtr+4 < BufferEnd) {
1666 // It has to be one of the bytes scanned, increment to it and read one.
1670 // Loop to scan the remainder.
1671 while (C != '/' && C != '\0')
1676 if (CurPtr[-2] == '*') // We found the final */. We're done!
1679 if ((CurPtr[-2] == '\n' || CurPtr[-2] == '\r')) {
1680 if (isEndOfBlockCommentWithEscapedNewLine(CurPtr-2, this)) {
1681 // We found the final */, though it had an escaped newline between the
1682 // * and /. We're done!
1686 if (CurPtr[0] == '*' && CurPtr[1] != '/') {
1687 // If this is a /* inside of the comment, emit a warning. Don't do this
1688 // if this is a /*/, which will end the comment. This misses cases with
1689 // embedded escaped newlines, but oh well.
1690 if (!isLexingRawMode())
1691 Diag(CurPtr-1, diag::warn_nested_block_comment);
1693 } else if (C == 0 && CurPtr == BufferEnd+1) {
1694 if (PP && PP->isCodeCompletionFile(FileLoc))
1695 PP->CodeCompleteNaturalLanguage();
1696 else if (!isLexingRawMode())
1697 Diag(BufferPtr, diag::err_unterminated_block_comment);
1698 // Note: the user probably forgot a */. We could continue immediately
1699 // after the /*, but this would involve lexing a lot of what really is the
1700 // comment, which surely would confuse the parser.
1703 // KeepWhitespaceMode should return this broken comment as a token. Since
1704 // it isn't a well formed comment, just return it as an 'unknown' token.
1705 if (isKeepWhitespaceMode()) {
1706 FormTokenWithChars(Result, CurPtr, tok::unknown);
1716 // Notify comment handlers about the comment unless we're in a #if 0 block.
1717 if (PP && !isLexingRawMode() &&
1718 PP->HandleComment(Result, SourceRange(getSourceLocation(BufferPtr),
1719 getSourceLocation(CurPtr)))) {
1721 return true; // A token has to be returned.
1724 // If we are returning comments as tokens, return this comment as a token.
1725 if (inKeepCommentMode()) {
1726 FormTokenWithChars(Result, CurPtr, tok::comment);
1730 // It is common for the tokens immediately after a /**/ comment to be
1731 // whitespace. Instead of going through the big switch, handle it
1732 // efficiently now. This is safe even in KeepWhitespaceMode because we would
1733 // have already returned above with the comment as a token.
1734 if (isHorizontalWhitespace(*CurPtr)) {
1735 Result.setFlag(Token::LeadingSpace);
1736 SkipWhitespace(Result, CurPtr+1);
1740 // Otherwise, just return so that the next character will be lexed as a token.
1742 Result.setFlag(Token::LeadingSpace);
1746 //===----------------------------------------------------------------------===//
1747 // Primary Lexing Entry Points
1748 //===----------------------------------------------------------------------===//
1750 /// ReadToEndOfLine - Read the rest of the current preprocessor line as an
1751 /// uninterpreted string. This switches the lexer out of directive mode.
1752 std::string Lexer::ReadToEndOfLine() {
1753 assert(ParsingPreprocessorDirective && ParsingFilename == false &&
1754 "Must be in a preprocessing directive!");
1758 // CurPtr - Cache BufferPtr in an automatic variable.
1759 const char *CurPtr = BufferPtr;
1761 char Char = getAndAdvanceChar(CurPtr, Tmp);
1767 // Found end of file?
1768 if (CurPtr-1 != BufferEnd) {
1769 // Nope, normal character, continue.
1776 // Okay, we found the end of the line. First, back up past the \0, \r, \n.
1777 assert(CurPtr[-1] == Char && "Trigraphs for newline?");
1778 BufferPtr = CurPtr-1;
1780 // Next, lex the character, which should handle the EOD transition.
1782 if (Tmp.is(tok::code_completion)) {
1783 if (PP && PP->getCodeCompletionHandler())
1784 PP->getCodeCompletionHandler()->CodeCompleteNaturalLanguage();
1787 assert(Tmp.is(tok::eod) && "Unexpected token!");
1789 // Finally, we're done, return the string we found.
1795 /// LexEndOfFile - CurPtr points to the end of this file. Handle this
1796 /// condition, reporting diagnostics and handling other edge cases as required.
1797 /// This returns true if Result contains a token, false if PP.Lex should be
1799 bool Lexer::LexEndOfFile(Token &Result, const char *CurPtr) {
1800 // Check if we are performing code completion.
1801 if (PP && PP->isCodeCompletionFile(FileLoc)) {
1802 // We're at the end of the file, but we've been asked to consider the
1803 // end of the file to be a code-completion token. Return the
1804 // code-completion token.
1805 Result.startToken();
1806 FormTokenWithChars(Result, CurPtr, tok::code_completion);
1808 // Only do the eof -> code_completion translation once.
1809 PP->SetCodeCompletionPoint(0, 0, 0);
1811 // Silence any diagnostics that occur once we hit the code-completion point.
1812 PP->getDiagnostics().setSuppressAllDiagnostics(true);
1816 // If we hit the end of the file while parsing a preprocessor directive,
1817 // end the preprocessor directive first. The next token returned will
1818 // then be the end of file.
1819 if (ParsingPreprocessorDirective) {
1820 // Done parsing the "line".
1821 ParsingPreprocessorDirective = false;
1822 // Update the location of token as well as BufferPtr.
1823 FormTokenWithChars(Result, CurPtr, tok::eod);
1825 // Restore comment saving mode, in case it was disabled for directive.
1826 SetCommentRetentionState(PP->getCommentRetentionState());
1827 return true; // Have a token.
1830 // If we are in raw mode, return this event as an EOF token. Let the caller
1831 // that put us in raw mode handle the event.
1832 if (isLexingRawMode()) {
1833 Result.startToken();
1834 BufferPtr = BufferEnd;
1835 FormTokenWithChars(Result, BufferEnd, tok::eof);
1839 // Issue diagnostics for unterminated #if and missing newline.
1841 // If we are in a #if directive, emit an error.
1842 while (!ConditionalStack.empty()) {
1843 if (!PP->isCodeCompletionFile(FileLoc))
1844 PP->Diag(ConditionalStack.back().IfLoc,
1845 diag::err_pp_unterminated_conditional);
1846 ConditionalStack.pop_back();
1849 // C99 5.1.1.2p2: If the file is non-empty and didn't end in a newline, issue
1851 if (CurPtr != BufferStart && (CurPtr[-1] != '\n' && CurPtr[-1] != '\r'))
1852 Diag(BufferEnd, diag::ext_no_newline_eof)
1853 << FixItHint::CreateInsertion(getSourceLocation(BufferEnd), "\n");
1857 // Finally, let the preprocessor handle this.
1858 return PP->HandleEndOfFile(Result);
1861 /// isNextPPTokenLParen - Return 1 if the next unexpanded token lexed from
1862 /// the specified lexer will return a tok::l_paren token, 0 if it is something
1863 /// else and 2 if there are no more tokens in the buffer controlled by the
1865 unsigned Lexer::isNextPPTokenLParen() {
1866 assert(!LexingRawMode && "How can we expand a macro from a skipping buffer?");
1868 // Switch to 'skipping' mode. This will ensure that we can lex a token
1869 // without emitting diagnostics, disables macro expansion, and will cause EOF
1870 // to return an EOF token instead of popping the include stack.
1871 LexingRawMode = true;
1873 // Save state that can be changed while lexing so that we can restore it.
1874 const char *TmpBufferPtr = BufferPtr;
1875 bool inPPDirectiveMode = ParsingPreprocessorDirective;
1879 LexTokenInternal(Tok);
1881 // Restore state that may have changed.
1882 BufferPtr = TmpBufferPtr;
1883 ParsingPreprocessorDirective = inPPDirectiveMode;
1885 // Restore the lexer back to non-skipping mode.
1886 LexingRawMode = false;
1888 if (Tok.is(tok::eof))
1890 return Tok.is(tok::l_paren);
1893 /// FindConflictEnd - Find the end of a version control conflict marker.
1894 static const char *FindConflictEnd(const char *CurPtr, const char *BufferEnd) {
1895 llvm::StringRef RestOfBuffer(CurPtr+7, BufferEnd-CurPtr-7);
1896 size_t Pos = RestOfBuffer.find(">>>>>>>");
1897 while (Pos != llvm::StringRef::npos) {
1898 // Must occur at start of line.
1899 if (RestOfBuffer[Pos-1] != '\r' &&
1900 RestOfBuffer[Pos-1] != '\n') {
1901 RestOfBuffer = RestOfBuffer.substr(Pos+7);
1902 Pos = RestOfBuffer.find(">>>>>>>");
1905 return RestOfBuffer.data()+Pos;
1910 /// IsStartOfConflictMarker - If the specified pointer is the start of a version
1911 /// control conflict marker like '<<<<<<<', recognize it as such, emit an error
1912 /// and recover nicely. This returns true if it is a conflict marker and false
1914 bool Lexer::IsStartOfConflictMarker(const char *CurPtr) {
1915 // Only a conflict marker if it starts at the beginning of a line.
1916 if (CurPtr != BufferStart &&
1917 CurPtr[-1] != '\n' && CurPtr[-1] != '\r')
1920 // Check to see if we have <<<<<<<.
1921 if (BufferEnd-CurPtr < 8 ||
1922 llvm::StringRef(CurPtr, 7) != "<<<<<<<")
1925 // If we have a situation where we don't care about conflict markers, ignore
1927 if (IsInConflictMarker || isLexingRawMode())
1930 // Check to see if there is a >>>>>>> somewhere in the buffer at the start of
1931 // a line to terminate this conflict marker.
1932 if (FindConflictEnd(CurPtr, BufferEnd)) {
1933 // We found a match. We are really in a conflict marker.
1934 // Diagnose this, and ignore to the end of line.
1935 Diag(CurPtr, diag::err_conflict_marker);
1936 IsInConflictMarker = true;
1938 // Skip ahead to the end of line. We know this exists because the
1939 // end-of-conflict marker starts with \r or \n.
1940 while (*CurPtr != '\r' && *CurPtr != '\n') {
1941 assert(CurPtr != BufferEnd && "Didn't find end of line");
1948 // No end of conflict marker found.
1953 /// HandleEndOfConflictMarker - If this is a '=======' or '|||||||' or '>>>>>>>'
1954 /// marker, then it is the end of a conflict marker. Handle it by ignoring up
1955 /// until the end of the line. This returns true if it is a conflict marker and
1957 bool Lexer::HandleEndOfConflictMarker(const char *CurPtr) {
1958 // Only a conflict marker if it starts at the beginning of a line.
1959 if (CurPtr != BufferStart &&
1960 CurPtr[-1] != '\n' && CurPtr[-1] != '\r')
1963 // If we have a situation where we don't care about conflict markers, ignore
1965 if (!IsInConflictMarker || isLexingRawMode())
1968 // Check to see if we have the marker (7 characters in a row).
1969 for (unsigned i = 1; i != 7; ++i)
1970 if (CurPtr[i] != CurPtr[0])
1973 // If we do have it, search for the end of the conflict marker. This could
1974 // fail if it got skipped with a '#if 0' or something. Note that CurPtr might
1975 // be the end of conflict marker.
1976 if (const char *End = FindConflictEnd(CurPtr, BufferEnd)) {
1979 // Skip ahead to the end of line.
1980 while (CurPtr != BufferEnd && *CurPtr != '\r' && *CurPtr != '\n')
1985 // No longer in the conflict marker.
1986 IsInConflictMarker = false;
1994 /// LexTokenInternal - This implements a simple C family lexer. It is an
1995 /// extremely performance critical piece of code. This assumes that the buffer
1996 /// has a null character at the end of the file. This returns a preprocessing
1997 /// token, not a normal token, as such, it is an internal interface. It assumes
1998 /// that the Flags of result have been cleared before calling this.
1999 void Lexer::LexTokenInternal(Token &Result) {
2001 // New token, can't need cleaning yet.
2002 Result.clearFlag(Token::NeedsCleaning);
2003 Result.setIdentifierInfo(0);
2005 // CurPtr - Cache BufferPtr in an automatic variable.
2006 const char *CurPtr = BufferPtr;
2008 // Small amounts of horizontal whitespace is very common between tokens.
2009 if ((*CurPtr == ' ') || (*CurPtr == '\t')) {
2011 while ((*CurPtr == ' ') || (*CurPtr == '\t'))
2014 // If we are keeping whitespace and other tokens, just return what we just
2015 // skipped. The next lexer invocation will return the token after the
2017 if (isKeepWhitespaceMode()) {
2018 FormTokenWithChars(Result, CurPtr, tok::unknown);
2023 Result.setFlag(Token::LeadingSpace);
2026 unsigned SizeTmp, SizeTmp2; // Temporaries for use in cases below.
2028 // Read a character, advancing over it.
2029 char Char = getAndAdvanceChar(CurPtr, Result);
2030 tok::TokenKind Kind;
2034 // Found end of file?
2035 if (CurPtr-1 == BufferEnd) {
2036 // Read the PP instance variable into an automatic variable, because
2037 // LexEndOfFile will often delete 'this'.
2038 Preprocessor *PPCache = PP;
2039 if (LexEndOfFile(Result, CurPtr-1)) // Retreat back into the file.
2040 return; // Got a token to return.
2041 assert(PPCache && "Raw buffer::LexEndOfFile should return a token");
2042 return PPCache->Lex(Result);
2045 if (!isLexingRawMode())
2046 Diag(CurPtr-1, diag::null_in_file);
2047 Result.setFlag(Token::LeadingSpace);
2048 if (SkipWhitespace(Result, CurPtr))
2049 return; // KeepWhitespaceMode
2051 goto LexNextToken; // GCC isn't tail call eliminating.
2053 case 26: // DOS & CP/M EOF: "^Z".
2054 // If we're in Microsoft extensions mode, treat this as end of file.
2055 if (Features.Microsoft) {
2056 // Read the PP instance variable into an automatic variable, because
2057 // LexEndOfFile will often delete 'this'.
2058 Preprocessor *PPCache = PP;
2059 if (LexEndOfFile(Result, CurPtr-1)) // Retreat back into the file.
2060 return; // Got a token to return.
2061 assert(PPCache && "Raw buffer::LexEndOfFile should return a token");
2062 return PPCache->Lex(Result);
2064 // If Microsoft extensions are disabled, this is just random garbage.
2065 Kind = tok::unknown;
2070 // If we are inside a preprocessor directive and we see the end of line,
2071 // we know we are done with the directive, so return an EOD token.
2072 if (ParsingPreprocessorDirective) {
2073 // Done parsing the "line".
2074 ParsingPreprocessorDirective = false;
2076 // Restore comment saving mode, in case it was disabled for directive.
2077 SetCommentRetentionState(PP->getCommentRetentionState());
2079 // Since we consumed a newline, we are back at the start of a line.
2080 IsAtStartOfLine = true;
2085 // The returned token is at the start of the line.
2086 Result.setFlag(Token::StartOfLine);
2087 // No leading whitespace seen so far.
2088 Result.clearFlag(Token::LeadingSpace);
2090 if (SkipWhitespace(Result, CurPtr))
2091 return; // KeepWhitespaceMode
2092 goto LexNextToken; // GCC isn't tail call eliminating.
2097 SkipHorizontalWhitespace:
2098 Result.setFlag(Token::LeadingSpace);
2099 if (SkipWhitespace(Result, CurPtr))
2100 return; // KeepWhitespaceMode
2105 // If the next token is obviously a // or /* */ comment, skip it efficiently
2106 // too (without going through the big switch stmt).
2107 if (CurPtr[0] == '/' && CurPtr[1] == '/' && !inKeepCommentMode() &&
2108 Features.BCPLComment && !Features.TraditionalCPP) {
2109 if (SkipBCPLComment(Result, CurPtr+2))
2110 return; // There is a token to return.
2111 goto SkipIgnoredUnits;
2112 } else if (CurPtr[0] == '/' && CurPtr[1] == '*' && !inKeepCommentMode()) {
2113 if (SkipBlockComment(Result, CurPtr+2))
2114 return; // There is a token to return.
2115 goto SkipIgnoredUnits;
2116 } else if (isHorizontalWhitespace(*CurPtr)) {
2117 goto SkipHorizontalWhitespace;
2119 goto LexNextToken; // GCC isn't tail call eliminating.
2121 // C99 6.4.4.1: Integer Constants.
2122 // C99 6.4.4.2: Floating Constants.
2123 case '0': case '1': case '2': case '3': case '4':
2124 case '5': case '6': case '7': case '8': case '9':
2125 // Notify MIOpt that we read a non-whitespace/non-comment token.
2127 return LexNumericConstant(Result, CurPtr);
2129 case 'L': // Identifier (Loony) or wide literal (L'x' or L"xyz").
2130 // Notify MIOpt that we read a non-whitespace/non-comment token.
2132 Char = getCharAndSize(CurPtr, SizeTmp);
2134 // Wide string literal.
2136 return LexStringLiteral(Result, ConsumeChar(CurPtr, SizeTmp, Result),
2139 // Wide character constant.
2141 return LexCharConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result));
2142 // FALL THROUGH, treating L like the start of an identifier.
2144 // C99 6.4.2: Identifiers.
2145 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G':
2146 case 'H': case 'I': case 'J': case 'K': /*'L'*/case 'M': case 'N':
2147 case 'O': case 'P': case 'Q': case 'R': case 'S': case 'T': case 'U':
2148 case 'V': case 'W': case 'X': case 'Y': case 'Z':
2149 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g':
2150 case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n':
2151 case 'o': case 'p': case 'q': case 'r': case 's': case 't': case 'u':
2152 case 'v': case 'w': case 'x': case 'y': case 'z':
2154 // Notify MIOpt that we read a non-whitespace/non-comment token.
2156 return LexIdentifier(Result, CurPtr);
2158 case '$': // $ in identifiers.
2159 if (Features.DollarIdents) {
2160 if (!isLexingRawMode())
2161 Diag(CurPtr-1, diag::ext_dollar_in_identifier);
2162 // Notify MIOpt that we read a non-whitespace/non-comment token.
2164 return LexIdentifier(Result, CurPtr);
2167 Kind = tok::unknown;
2170 // C99 6.4.4: Character Constants.
2172 // Notify MIOpt that we read a non-whitespace/non-comment token.
2174 return LexCharConstant(Result, CurPtr);
2176 // C99 6.4.5: String Literals.
2178 // Notify MIOpt that we read a non-whitespace/non-comment token.
2180 return LexStringLiteral(Result, CurPtr, false);
2182 // C99 6.4.6: Punctuators.
2184 Kind = tok::question;
2187 Kind = tok::l_square;
2190 Kind = tok::r_square;
2193 Kind = tok::l_paren;
2196 Kind = tok::r_paren;
2199 Kind = tok::l_brace;
2202 Kind = tok::r_brace;
2205 Char = getCharAndSize(CurPtr, SizeTmp);
2206 if (Char >= '0' && Char <= '9') {
2207 // Notify MIOpt that we read a non-whitespace/non-comment token.
2210 return LexNumericConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result));
2211 } else if (Features.CPlusPlus && Char == '*') {
2212 Kind = tok::periodstar;
2214 } else if (Char == '.' &&
2215 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '.') {
2216 Kind = tok::ellipsis;
2217 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
2224 Char = getCharAndSize(CurPtr, SizeTmp);
2227 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2228 } else if (Char == '=') {
2229 Kind = tok::ampequal;
2230 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2236 if (getCharAndSize(CurPtr, SizeTmp) == '=') {
2237 Kind = tok::starequal;
2238 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2244 Char = getCharAndSize(CurPtr, SizeTmp);
2246 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2247 Kind = tok::plusplus;
2248 } else if (Char == '=') {
2249 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2250 Kind = tok::plusequal;
2256 Char = getCharAndSize(CurPtr, SizeTmp);
2257 if (Char == '-') { // --
2258 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2259 Kind = tok::minusminus;
2260 } else if (Char == '>' && Features.CPlusPlus &&
2261 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '*') { // C++ ->*
2262 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
2264 Kind = tok::arrowstar;
2265 } else if (Char == '>') { // ->
2266 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2268 } else if (Char == '=') { // -=
2269 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2270 Kind = tok::minusequal;
2279 if (getCharAndSize(CurPtr, SizeTmp) == '=') {
2280 Kind = tok::exclaimequal;
2281 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2283 Kind = tok::exclaim;
2288 Char = getCharAndSize(CurPtr, SizeTmp);
2289 if (Char == '/') { // BCPL comment.
2290 // Even if BCPL comments are disabled (e.g. in C89 mode), we generally
2291 // want to lex this as a comment. There is one problem with this though,
2292 // that in one particular corner case, this can change the behavior of the
2293 // resultant program. For example, In "foo //**/ bar", C89 would lex
2294 // this as "foo / bar" and langauges with BCPL comments would lex it as
2295 // "foo". Check to see if the character after the second slash is a '*'.
2296 // If so, we will lex that as a "/" instead of the start of a comment.
2297 // However, we never do this in -traditional-cpp mode.
2298 if ((Features.BCPLComment ||
2299 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) != '*') &&
2300 !Features.TraditionalCPP) {
2301 if (SkipBCPLComment(Result, ConsumeChar(CurPtr, SizeTmp, Result)))
2302 return; // There is a token to return.
2304 // It is common for the tokens immediately after a // comment to be
2305 // whitespace (indentation for the next line). Instead of going through
2306 // the big switch, handle it efficiently now.
2307 goto SkipIgnoredUnits;
2311 if (Char == '*') { // /**/ comment.
2312 if (SkipBlockComment(Result, ConsumeChar(CurPtr, SizeTmp, Result)))
2313 return; // There is a token to return.
2314 goto LexNextToken; // GCC isn't tail call eliminating.
2318 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2319 Kind = tok::slashequal;
2325 Char = getCharAndSize(CurPtr, SizeTmp);
2327 Kind = tok::percentequal;
2328 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2329 } else if (Features.Digraphs && Char == '>') {
2330 Kind = tok::r_brace; // '%>' -> '}'
2331 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2332 } else if (Features.Digraphs && Char == ':') {
2333 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2334 Char = getCharAndSize(CurPtr, SizeTmp);
2335 if (Char == '%' && getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == ':') {
2336 Kind = tok::hashhash; // '%:%:' -> '##'
2337 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
2339 } else if (Char == '@' && Features.Microsoft) { // %:@ -> #@ -> Charize
2340 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2341 if (!isLexingRawMode())
2342 Diag(BufferPtr, diag::charize_microsoft_ext);
2344 } else { // '%:' -> '#'
2345 // We parsed a # character. If this occurs at the start of the line,
2346 // it's actually the start of a preprocessing directive. Callback to
2347 // the preprocessor to handle it.
2348 // FIXME: -fpreprocessed mode??
2349 if (Result.isAtStartOfLine() && !LexingRawMode && !Is_PragmaLexer) {
2350 FormTokenWithChars(Result, CurPtr, tok::hash);
2351 PP->HandleDirective(Result);
2353 // As an optimization, if the preprocessor didn't switch lexers, tail
2355 if (PP->isCurrentLexer(this)) {
2356 // Start a new token. If this is a #include or something, the PP may
2357 // want us starting at the beginning of the line again. If so, set
2358 // the StartOfLine flag and clear LeadingSpace.
2359 if (IsAtStartOfLine) {
2360 Result.setFlag(Token::StartOfLine);
2361 Result.clearFlag(Token::LeadingSpace);
2362 IsAtStartOfLine = false;
2364 goto LexNextToken; // GCC isn't tail call eliminating.
2367 return PP->Lex(Result);
2373 Kind = tok::percent;
2377 Char = getCharAndSize(CurPtr, SizeTmp);
2378 if (ParsingFilename) {
2379 return LexAngledStringLiteral(Result, CurPtr);
2380 } else if (Char == '<') {
2381 char After = getCharAndSize(CurPtr+SizeTmp, SizeTmp2);
2383 Kind = tok::lesslessequal;
2384 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
2386 } else if (After == '<' && IsStartOfConflictMarker(CurPtr-1)) {
2387 // If this is actually a '<<<<<<<' version control conflict marker,
2388 // recognize it as such and recover nicely.
2390 } else if (Features.CUDA && After == '<') {
2391 Kind = tok::lesslessless;
2392 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
2395 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2396 Kind = tok::lessless;
2398 } else if (Char == '=') {
2399 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2400 Kind = tok::lessequal;
2401 } else if (Features.Digraphs && Char == ':') { // '<:' -> '['
2402 if (Features.CPlusPlus0x &&
2403 getCharAndSize(CurPtr + SizeTmp, SizeTmp2) == ':') {
2404 // C++0x [lex.pptoken]p3:
2405 // Otherwise, if the next three characters are <:: and the subsequent
2406 // character is neither : nor >, the < is treated as a preprocessor
2407 // token by itself and not as the first character of the alternative
2410 char After = getCharAndSize(CurPtr + SizeTmp + SizeTmp2, SizeTmp3);
2411 if (After != ':' && After != '>') {
2417 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2418 Kind = tok::l_square;
2419 } else if (Features.Digraphs && Char == '%') { // '<%' -> '{'
2420 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2421 Kind = tok::l_brace;
2427 Char = getCharAndSize(CurPtr, SizeTmp);
2429 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2430 Kind = tok::greaterequal;
2431 } else if (Char == '>') {
2432 char After = getCharAndSize(CurPtr+SizeTmp, SizeTmp2);
2434 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
2436 Kind = tok::greatergreaterequal;
2437 } else if (After == '>' && HandleEndOfConflictMarker(CurPtr-1)) {
2438 // If this is '>>>>>>>' and we're in a conflict marker, ignore it.
2440 } else if (Features.CUDA && After == '>') {
2441 Kind = tok::greatergreatergreater;
2442 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
2445 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2446 Kind = tok::greatergreater;
2450 Kind = tok::greater;
2454 Char = getCharAndSize(CurPtr, SizeTmp);
2456 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2457 Kind = tok::caretequal;
2463 Char = getCharAndSize(CurPtr, SizeTmp);
2465 Kind = tok::pipeequal;
2466 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2467 } else if (Char == '|') {
2468 // If this is '|||||||' and we're in a conflict marker, ignore it.
2469 if (CurPtr[1] == '|' && HandleEndOfConflictMarker(CurPtr-1))
2471 Kind = tok::pipepipe;
2472 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2478 Char = getCharAndSize(CurPtr, SizeTmp);
2479 if (Features.Digraphs && Char == '>') {
2480 Kind = tok::r_square; // ':>' -> ']'
2481 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2482 } else if (Features.CPlusPlus && Char == ':') {
2483 Kind = tok::coloncolon;
2484 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2493 Char = getCharAndSize(CurPtr, SizeTmp);
2495 // If this is '=======' and we're in a conflict marker, ignore it.
2496 if (CurPtr[1] == '=' && HandleEndOfConflictMarker(CurPtr-1))
2499 Kind = tok::equalequal;
2500 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2509 Char = getCharAndSize(CurPtr, SizeTmp);
2511 Kind = tok::hashhash;
2512 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2513 } else if (Char == '@' && Features.Microsoft) { // #@ -> Charize
2515 if (!isLexingRawMode())
2516 Diag(BufferPtr, diag::charize_microsoft_ext);
2517 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
2519 // We parsed a # character. If this occurs at the start of the line,
2520 // it's actually the start of a preprocessing directive. Callback to
2521 // the preprocessor to handle it.
2522 // FIXME: -fpreprocessed mode??
2523 if (Result.isAtStartOfLine() && !LexingRawMode && !Is_PragmaLexer) {
2524 FormTokenWithChars(Result, CurPtr, tok::hash);
2525 PP->HandleDirective(Result);
2527 // As an optimization, if the preprocessor didn't switch lexers, tail
2529 if (PP->isCurrentLexer(this)) {
2530 // Start a new token. If this is a #include or something, the PP may
2531 // want us starting at the beginning of the line again. If so, set
2532 // the StartOfLine flag and clear LeadingSpace.
2533 if (IsAtStartOfLine) {
2534 Result.setFlag(Token::StartOfLine);
2535 Result.clearFlag(Token::LeadingSpace);
2536 IsAtStartOfLine = false;
2538 goto LexNextToken; // GCC isn't tail call eliminating.
2540 return PP->Lex(Result);
2548 // Objective C support.
2549 if (CurPtr[-1] == '@' && Features.ObjC1)
2552 Kind = tok::unknown;
2559 Kind = tok::unknown;
2563 // Notify MIOpt that we read a non-whitespace/non-comment token.
2566 // Update the location of token as well as BufferPtr.
2567 FormTokenWithChars(Result, CurPtr, Kind);