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 Is_PragmaLexer
= false;
75 IsInConflictMarker
= false;
77 // Start of the file is a start of line.
78 IsAtStartOfLine
= true;
80 // We are not after parsing a #.
81 ParsingPreprocessorDirective
= false;
83 // We are not after parsing #include.
84 ParsingFilename
= false;
86 // We are not in raw mode. Raw mode disables diagnostics and interpretation
87 // of tokens (e.g. identifiers, thus disabling macro expansion). It is used
88 // to quickly lex the tokens of the buffer, e.g. when handling a "#if 0" block
89 // or otherwise skipping over tokens.
90 LexingRawMode
= false;
92 // Default to not keeping comments.
93 ExtendedTokenMode
= 0;
96 /// Lexer constructor - Create a new lexer object for the specified buffer
97 /// with the specified preprocessor managing the lexing process. This lexer
98 /// assumes that the associated file buffer and Preprocessor objects will
99 /// outlive it, so it doesn't take ownership of either of them.
100 Lexer::Lexer(FileID FID
, const llvm::MemoryBuffer
*InputFile
, Preprocessor
&PP
)
101 : PreprocessorLexer(&PP
, FID
),
102 FileLoc(PP
.getSourceManager().getLocForStartOfFile(FID
)),
103 Features(PP
.getLangOptions()) {
105 InitLexer(InputFile
->getBufferStart(), InputFile
->getBufferStart(),
106 InputFile
->getBufferEnd());
108 // Default to keeping comments if the preprocessor wants them.
109 SetCommentRetentionState(PP
.getCommentRetentionState());
112 /// Lexer constructor - Create a new raw lexer object. This object is only
113 /// suitable for calls to 'LexRawToken'. This lexer assumes that the text
114 /// range will outlive it, so it doesn't take ownership of it.
115 Lexer::Lexer(SourceLocation fileloc
, const LangOptions
&features
,
116 const char *BufStart
, const char *BufPtr
, const char *BufEnd
)
117 : FileLoc(fileloc
), Features(features
) {
119 InitLexer(BufStart
, BufPtr
, BufEnd
);
121 // We *are* in raw mode.
122 LexingRawMode
= true;
125 /// Lexer constructor - Create a new raw lexer object. This object is only
126 /// suitable for calls to 'LexRawToken'. This lexer assumes that the text
127 /// range will outlive it, so it doesn't take ownership of it.
128 Lexer::Lexer(FileID FID
, const llvm::MemoryBuffer
*FromFile
,
129 const SourceManager
&SM
, const LangOptions
&features
)
130 : FileLoc(SM
.getLocForStartOfFile(FID
)), Features(features
) {
132 InitLexer(FromFile
->getBufferStart(), FromFile
->getBufferStart(),
133 FromFile
->getBufferEnd());
135 // We *are* in raw mode.
136 LexingRawMode
= true;
139 /// Create_PragmaLexer: Lexer constructor - Create a new lexer object for
140 /// _Pragma expansion. This has a variety of magic semantics that this method
141 /// sets up. It returns a new'd Lexer that must be delete'd when done.
143 /// On entrance to this routine, TokStartLoc is a macro location which has a
144 /// spelling loc that indicates the bytes to be lexed for the token and an
145 /// instantiation location that indicates where all lexed tokens should be
148 /// FIXME: It would really be nice to make _Pragma just be a wrapper around a
149 /// normal lexer that remaps tokens as they fly by. This would require making
150 /// Preprocessor::Lex virtual. Given that, we could just dump in a magic lexer
151 /// interface that could handle this stuff. This would pull GetMappedTokenLoc
152 /// out of the critical path of the lexer!
154 Lexer
*Lexer::Create_PragmaLexer(SourceLocation SpellingLoc
,
155 SourceLocation InstantiationLocStart
,
156 SourceLocation InstantiationLocEnd
,
157 unsigned TokLen
, Preprocessor
&PP
) {
158 SourceManager
&SM
= PP
.getSourceManager();
160 // Create the lexer as if we were going to lex the file normally.
161 FileID SpellingFID
= SM
.getFileID(SpellingLoc
);
162 const llvm::MemoryBuffer
*InputFile
= SM
.getBuffer(SpellingFID
);
163 Lexer
*L
= new Lexer(SpellingFID
, InputFile
, PP
);
165 // Now that the lexer is created, change the start/end locations so that we
166 // just lex the subsection of the file that we want. This is lexing from a
168 const char *StrData
= SM
.getCharacterData(SpellingLoc
);
170 L
->BufferPtr
= StrData
;
171 L
->BufferEnd
= StrData
+TokLen
;
172 assert(L
->BufferEnd
[0] == 0 && "Buffer is not nul terminated!");
174 // Set the SourceLocation with the remapping information. This ensures that
175 // GetMappedTokenLoc will remap the tokens as they are lexed.
176 L
->FileLoc
= SM
.createInstantiationLoc(SM
.getLocForStartOfFile(SpellingFID
),
177 InstantiationLocStart
,
178 InstantiationLocEnd
, TokLen
);
180 // Ensure that the lexer thinks it is inside a directive, so that end \n will
181 // return an EOM token.
182 L
->ParsingPreprocessorDirective
= true;
184 // This lexer really is for _Pragma.
185 L
->Is_PragmaLexer
= true;
190 /// Stringify - Convert the specified string into a C string, with surrounding
191 /// ""'s, and with escaped \ and " characters.
192 std::string
Lexer::Stringify(const std::string
&Str
, bool Charify
) {
193 std::string Result
= Str
;
194 char Quote
= Charify
? '\'' : '"';
195 for (unsigned i
= 0, e
= Result
.size(); i
!= e
; ++i
) {
196 if (Result
[i
] == '\\' || Result
[i
] == Quote
) {
197 Result
.insert(Result
.begin()+i
, '\\');
204 /// Stringify - Convert the specified string into a C string by escaping '\'
205 /// and " characters. This does not add surrounding ""'s to the string.
206 void Lexer::Stringify(llvm::SmallVectorImpl
<char> &Str
) {
207 for (unsigned i
= 0, e
= Str
.size(); i
!= e
; ++i
) {
208 if (Str
[i
] == '\\' || Str
[i
] == '"') {
209 Str
.insert(Str
.begin()+i
, '\\');
215 //===----------------------------------------------------------------------===//
217 //===----------------------------------------------------------------------===//
219 /// getSpelling() - Return the 'spelling' of this token. The spelling of a
220 /// token are the characters used to represent the token in the source file
221 /// after trigraph expansion and escaped-newline folding. In particular, this
222 /// wants to get the true, uncanonicalized, spelling of things like digraphs
224 std::string
Lexer::getSpelling(const Token
&Tok
, const SourceManager
&SourceMgr
,
225 const LangOptions
&Features
, bool *Invalid
) {
226 assert((int)Tok
.getLength() >= 0 && "Token character range is bogus!");
228 // If this token contains nothing interesting, return it directly.
229 bool CharDataInvalid
= false;
230 const char* TokStart
= SourceMgr
.getCharacterData(Tok
.getLocation(),
233 *Invalid
= CharDataInvalid
;
235 return std::string();
237 if (!Tok
.needsCleaning())
238 return std::string(TokStart
, TokStart
+Tok
.getLength());
241 Result
.reserve(Tok
.getLength());
243 // Otherwise, hard case, relex the characters into the string.
244 for (const char *Ptr
= TokStart
, *End
= TokStart
+Tok
.getLength();
247 Result
.push_back(Lexer::getCharAndSizeNoWarn(Ptr
, CharSize
, Features
));
250 assert(Result
.size() != unsigned(Tok
.getLength()) &&
251 "NeedsCleaning flag set on something that didn't need cleaning!");
255 /// getSpelling - This method is used to get the spelling of a token into a
256 /// preallocated buffer, instead of as an std::string. The caller is required
257 /// to allocate enough space for the token, which is guaranteed to be at least
258 /// Tok.getLength() bytes long. The actual length of the token is returned.
260 /// Note that this method may do two possible things: it may either fill in
261 /// the buffer specified with characters, or it may *change the input pointer*
262 /// to point to a constant buffer with the data already in it (avoiding a
263 /// copy). The caller is not allowed to modify the returned buffer pointer
264 /// if an internal buffer is returned.
265 unsigned Lexer::getSpelling(const Token
&Tok
, const char *&Buffer
,
266 const SourceManager
&SourceMgr
,
267 const LangOptions
&Features
, bool *Invalid
) {
268 assert((int)Tok
.getLength() >= 0 && "Token character range is bogus!");
270 const char *TokStart
= 0;
271 // NOTE: this has to be checked *before* testing for an IdentifierInfo.
272 if (Tok
.is(tok::raw_identifier
))
273 TokStart
= Tok
.getRawIdentifierData();
274 else if (const IdentifierInfo
*II
= Tok
.getIdentifierInfo()) {
275 // Just return the string from the identifier table, which is very quick.
276 Buffer
= II
->getNameStart();
277 return II
->getLength();
280 // NOTE: this can be checked even after testing for an IdentifierInfo.
282 TokStart
= Tok
.getLiteralData();
285 // Compute the start of the token in the input lexer buffer.
286 bool CharDataInvalid
= false;
287 TokStart
= SourceMgr
.getCharacterData(Tok
.getLocation(), &CharDataInvalid
);
289 *Invalid
= CharDataInvalid
;
290 if (CharDataInvalid
) {
296 // If this token contains nothing interesting, return it directly.
297 if (!Tok
.needsCleaning()) {
299 return Tok
.getLength();
302 // Otherwise, hard case, relex the characters into the string.
303 char *OutBuf
= const_cast<char*>(Buffer
);
304 for (const char *Ptr
= TokStart
, *End
= TokStart
+Tok
.getLength();
307 *OutBuf
++ = Lexer::getCharAndSizeNoWarn(Ptr
, CharSize
, Features
);
310 assert(unsigned(OutBuf
-Buffer
) != Tok
.getLength() &&
311 "NeedsCleaning flag set on something that didn't need cleaning!");
313 return OutBuf
-Buffer
;
318 static bool isWhitespace(unsigned char c
);
320 /// MeasureTokenLength - Relex the token at the specified location and return
321 /// its length in bytes in the input file. If the token needs cleaning (e.g.
322 /// includes a trigraph or an escaped newline) then this count includes bytes
323 /// that are part of that.
324 unsigned Lexer::MeasureTokenLength(SourceLocation Loc
,
325 const SourceManager
&SM
,
326 const LangOptions
&LangOpts
) {
327 // TODO: this could be special cased for common tokens like identifiers, ')',
328 // etc to make this faster, if it mattered. Just look at StrData[0] to handle
329 // all obviously single-char tokens. This could use
330 // Lexer::isObviouslySimpleCharacter for example to handle identifiers or
333 // If this comes from a macro expansion, we really do want the macro name, not
334 // the token this macro expanded to.
335 Loc
= SM
.getInstantiationLoc(Loc
);
336 std::pair
<FileID
, unsigned> LocInfo
= SM
.getDecomposedLoc(Loc
);
337 bool Invalid
= false;
338 llvm::StringRef Buffer
= SM
.getBufferData(LocInfo
.first
, &Invalid
);
342 const char *StrData
= Buffer
.data()+LocInfo
.second
;
344 if (isWhitespace(StrData
[0]))
347 // Create a lexer starting at the beginning of this token.
348 Lexer
TheLexer(SM
.getLocForStartOfFile(LocInfo
.first
), LangOpts
,
349 Buffer
.begin(), StrData
, Buffer
.end());
350 TheLexer
.SetCommentRetentionState(true);
352 TheLexer
.LexFromRawLexer(TheTok
);
353 return TheTok
.getLength();
356 SourceLocation
Lexer::GetBeginningOfToken(SourceLocation Loc
,
357 const SourceManager
&SM
,
358 const LangOptions
&LangOpts
) {
359 std::pair
<FileID
, unsigned> LocInfo
= SM
.getDecomposedLoc(Loc
);
360 if (LocInfo
.first
.isInvalid())
363 bool Invalid
= false;
364 llvm::StringRef Buffer
= SM
.getBufferData(LocInfo
.first
, &Invalid
);
368 // Back up from the current location until we hit the beginning of a line
369 // (or the buffer). We'll relex from that point.
370 const char *BufStart
= Buffer
.data();
371 if (LocInfo
.second
>= Buffer
.size())
374 const char *StrData
= BufStart
+LocInfo
.second
;
375 if (StrData
[0] == '\n' || StrData
[0] == '\r')
378 const char *LexStart
= StrData
;
379 while (LexStart
!= BufStart
) {
380 if (LexStart
[0] == '\n' || LexStart
[0] == '\r') {
388 // Create a lexer starting at the beginning of this token.
389 SourceLocation LexerStartLoc
= Loc
.getFileLocWithOffset(-LocInfo
.second
);
390 Lexer
TheLexer(LexerStartLoc
, LangOpts
, BufStart
, LexStart
, Buffer
.end());
391 TheLexer
.SetCommentRetentionState(true);
393 // Lex tokens until we find the token that contains the source location.
396 TheLexer
.LexFromRawLexer(TheTok
);
398 if (TheLexer
.getBufferLocation() > StrData
) {
399 // Lexing this token has taken the lexer past the source location we're
400 // looking for. If the current token encompasses our source location,
401 // return the beginning of that token.
402 if (TheLexer
.getBufferLocation() - TheTok
.getLength() <= StrData
)
403 return TheTok
.getLocation();
405 // We ended up skipping over the source location entirely, which means
406 // that it points into whitespace. We're done here.
409 } while (TheTok
.getKind() != tok::eof
);
411 // We've passed our source location; just return the original source location.
416 enum PreambleDirectiveKind
{
424 std::pair
<unsigned, bool>
425 Lexer::ComputePreamble(const llvm::MemoryBuffer
*Buffer
, unsigned MaxLines
) {
426 // Create a lexer starting at the beginning of the file. Note that we use a
427 // "fake" file source location at offset 1 so that the lexer will track our
428 // position within the file.
429 const unsigned StartOffset
= 1;
430 SourceLocation StartLoc
= SourceLocation::getFromRawEncoding(StartOffset
);
431 LangOptions LangOpts
;
432 Lexer
TheLexer(StartLoc
, LangOpts
, Buffer
->getBufferStart(),
433 Buffer
->getBufferStart(), Buffer
->getBufferEnd());
435 bool InPreprocessorDirective
= false;
438 unsigned IfCount
= 0;
442 TheLexer
.LexFromRawLexer(TheTok
);
444 if (InPreprocessorDirective
) {
445 // If we've hit the end of the file, we're done.
446 if (TheTok
.getKind() == tok::eof
) {
447 InPreprocessorDirective
= false;
451 // If we haven't hit the end of the preprocessor directive, skip this
453 if (!TheTok
.isAtStartOfLine())
456 // We've passed the end of the preprocessor directive, and will look
457 // at this token again below.
458 InPreprocessorDirective
= false;
461 // Keep track of the # of lines in the preamble.
462 if (TheTok
.isAtStartOfLine()) {
465 // If we were asked to limit the number of lines in the preamble,
466 // and we're about to exceed that limit, we're done.
467 if (MaxLines
&& Line
>= MaxLines
)
471 // Comments are okay; skip over them.
472 if (TheTok
.getKind() == tok::comment
)
475 if (TheTok
.isAtStartOfLine() && TheTok
.getKind() == tok::hash
) {
476 // This is the start of a preprocessor directive.
477 Token HashTok
= TheTok
;
478 InPreprocessorDirective
= true;
480 // Figure out which direective this is. Since we're lexing raw tokens,
481 // we don't have an identifier table available. Instead, just look at
482 // the raw identifier to recognize and categorize preprocessor directives.
483 TheLexer
.LexFromRawLexer(TheTok
);
484 if (TheTok
.getKind() == tok::raw_identifier
&& !TheTok
.needsCleaning()) {
485 llvm::StringRef
Keyword(TheTok
.getRawIdentifierData(),
487 PreambleDirectiveKind PDK
488 = llvm::StringSwitch
<PreambleDirectiveKind
>(Keyword
)
489 .Case("include", PDK_Skipped
)
490 .Case("__include_macros", PDK_Skipped
)
491 .Case("define", PDK_Skipped
)
492 .Case("undef", PDK_Skipped
)
493 .Case("line", PDK_Skipped
)
494 .Case("error", PDK_Skipped
)
495 .Case("pragma", PDK_Skipped
)
496 .Case("import", PDK_Skipped
)
497 .Case("include_next", PDK_Skipped
)
498 .Case("warning", PDK_Skipped
)
499 .Case("ident", PDK_Skipped
)
500 .Case("sccs", PDK_Skipped
)
501 .Case("assert", PDK_Skipped
)
502 .Case("unassert", PDK_Skipped
)
503 .Case("if", PDK_StartIf
)
504 .Case("ifdef", PDK_StartIf
)
505 .Case("ifndef", PDK_StartIf
)
506 .Case("elif", PDK_Skipped
)
507 .Case("else", PDK_Skipped
)
508 .Case("endif", PDK_EndIf
)
509 .Default(PDK_Unknown
);
517 IfStartTok
= HashTok
;
523 // Mismatched #endif. The preamble ends here.
531 // We don't know what this directive is; stop at the '#'.
536 // We only end up here if we didn't recognize the preprocessor
537 // directive or it was one that can't occur in the preamble at this
538 // point. Roll back the current token to the location of the '#'.
539 InPreprocessorDirective
= false;
543 // We hit a token that we don't recognize as being in the
544 // "preprocessing only" part of the file, so we're no longer in
549 SourceLocation End
= IfCount
? IfStartTok
.getLocation() : TheTok
.getLocation();
550 return std::make_pair(End
.getRawEncoding() - StartLoc
.getRawEncoding(),
551 IfCount
? IfStartTok
.isAtStartOfLine()
552 : TheTok
.isAtStartOfLine());
556 /// AdvanceToTokenCharacter - Given a location that specifies the start of a
557 /// token, return a new location that specifies a character within the token.
558 SourceLocation
Lexer::AdvanceToTokenCharacter(SourceLocation TokStart
,
560 const SourceManager
&SM
,
561 const LangOptions
&Features
) {
562 // Figure out how many physical characters away the specified instantiation
563 // character is. This needs to take into consideration newlines and
565 bool Invalid
= false;
566 const char *TokPtr
= SM
.getCharacterData(TokStart
, &Invalid
);
568 // If they request the first char of the token, we're trivially done.
569 if (Invalid
|| (CharNo
== 0 && Lexer::isObviouslySimpleCharacter(*TokPtr
)))
572 unsigned PhysOffset
= 0;
574 // The usual case is that tokens don't contain anything interesting. Skip
575 // over the uninteresting characters. If a token only consists of simple
576 // chars, this method is extremely fast.
577 while (Lexer::isObviouslySimpleCharacter(*TokPtr
)) {
579 return TokStart
.getFileLocWithOffset(PhysOffset
);
580 ++TokPtr
, --CharNo
, ++PhysOffset
;
583 // If we have a character that may be a trigraph or escaped newline, use a
584 // lexer to parse it correctly.
585 for (; CharNo
; --CharNo
) {
587 Lexer::getCharAndSizeNoWarn(TokPtr
, Size
, Features
);
592 // Final detail: if we end up on an escaped newline, we want to return the
593 // location of the actual byte of the token. For example foo\<newline>bar
594 // advanced by 3 should return the location of b, not of \\. One compounding
595 // detail of this is that the escape may be made by a trigraph.
596 if (!Lexer::isObviouslySimpleCharacter(*TokPtr
))
597 PhysOffset
+= Lexer::SkipEscapedNewLines(TokPtr
)-TokPtr
;
599 return TokStart
.getFileLocWithOffset(PhysOffset
);
602 /// \brief Computes the source location just past the end of the
603 /// token at this source location.
605 /// This routine can be used to produce a source location that
606 /// points just past the end of the token referenced by \p Loc, and
607 /// is generally used when a diagnostic needs to point just after a
608 /// token where it expected something different that it received. If
609 /// the returned source location would not be meaningful (e.g., if
610 /// it points into a macro), this routine returns an invalid
613 /// \param Offset an offset from the end of the token, where the source
614 /// location should refer to. The default offset (0) produces a source
615 /// location pointing just past the end of the token; an offset of 1 produces
616 /// a source location pointing to the last character in the token, etc.
617 SourceLocation
Lexer::getLocForEndOfToken(SourceLocation Loc
, unsigned Offset
,
618 const SourceManager
&SM
,
619 const LangOptions
&Features
) {
620 if (Loc
.isInvalid() || !Loc
.isFileID())
621 return SourceLocation();
623 unsigned Len
= Lexer::MeasureTokenLength(Loc
, SM
, Features
);
629 return AdvanceToTokenCharacter(Loc
, Len
, SM
, Features
);
632 //===----------------------------------------------------------------------===//
633 // Character information.
634 //===----------------------------------------------------------------------===//
637 CHAR_HORZ_WS
= 0x01, // ' ', '\t', '\f', '\v'. Note, no '\0'
638 CHAR_VERT_WS
= 0x02, // '\r', '\n'
639 CHAR_LETTER
= 0x04, // a-z,A-Z
640 CHAR_NUMBER
= 0x08, // 0-9
641 CHAR_UNDER
= 0x10, // _
642 CHAR_PERIOD
= 0x20 // .
645 // Statically initialize CharInfo table based on ASCII character set
646 // Reference: FreeBSD 7.2 /usr/share/misc/ascii
647 static const unsigned char CharInfo
[256] =
649 // 0 NUL 1 SOH 2 STX 3 ETX
650 // 4 EOT 5 ENQ 6 ACK 7 BEL
653 // 8 BS 9 HT 10 NL 11 VT
654 //12 NP 13 CR 14 SO 15 SI
655 0 , CHAR_HORZ_WS
, CHAR_VERT_WS
, CHAR_HORZ_WS
,
656 CHAR_HORZ_WS
, CHAR_VERT_WS
, 0 , 0 ,
657 //16 DLE 17 DC1 18 DC2 19 DC3
658 //20 DC4 21 NAK 22 SYN 23 ETB
661 //24 CAN 25 EM 26 SUB 27 ESC
662 //28 FS 29 GS 30 RS 31 US
665 //32 SP 33 ! 34 " 35 #
666 //36 $ 37 % 38 & 39 '
667 CHAR_HORZ_WS
, 0 , 0 , 0 ,
669 //40 ( 41 ) 42 * 43 +
670 //44 , 45 - 46 . 47 /
672 0 , 0 , CHAR_PERIOD
, 0 ,
673 //48 0 49 1 50 2 51 3
674 //52 4 53 5 54 6 55 7
675 CHAR_NUMBER
, CHAR_NUMBER
, CHAR_NUMBER
, CHAR_NUMBER
,
676 CHAR_NUMBER
, CHAR_NUMBER
, CHAR_NUMBER
, CHAR_NUMBER
,
677 //56 8 57 9 58 : 59 ;
678 //60 < 61 = 62 > 63 ?
679 CHAR_NUMBER
, CHAR_NUMBER
, 0 , 0 ,
681 //64 @ 65 A 66 B 67 C
682 //68 D 69 E 70 F 71 G
683 0 , CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
684 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
685 //72 H 73 I 74 J 75 K
686 //76 L 77 M 78 N 79 O
687 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
688 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
689 //80 P 81 Q 82 R 83 S
690 //84 T 85 U 86 V 87 W
691 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
692 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
693 //88 X 89 Y 90 Z 91 [
694 //92 \ 93 ] 94 ^ 95 _
695 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, 0 ,
696 0 , 0 , 0 , CHAR_UNDER
,
697 //96 ` 97 a 98 b 99 c
698 //100 d 101 e 102 f 103 g
699 0 , CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
700 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
701 //104 h 105 i 106 j 107 k
702 //108 l 109 m 110 n 111 o
703 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
704 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
705 //112 p 113 q 114 r 115 s
706 //116 t 117 u 118 v 119 w
707 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
708 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
,
709 //120 x 121 y 122 z 123 {
710 //124 | 125 } 126 ~ 127 DEL
711 CHAR_LETTER
, CHAR_LETTER
, CHAR_LETTER
, 0 ,
715 static void InitCharacterInfo() {
716 static bool isInited
= false;
717 if (isInited
) return;
718 // check the statically-initialized CharInfo table
719 assert(CHAR_HORZ_WS
== CharInfo
[(int)' ']);
720 assert(CHAR_HORZ_WS
== CharInfo
[(int)'\t']);
721 assert(CHAR_HORZ_WS
== CharInfo
[(int)'\f']);
722 assert(CHAR_HORZ_WS
== CharInfo
[(int)'\v']);
723 assert(CHAR_VERT_WS
== CharInfo
[(int)'\n']);
724 assert(CHAR_VERT_WS
== CharInfo
[(int)'\r']);
725 assert(CHAR_UNDER
== CharInfo
[(int)'_']);
726 assert(CHAR_PERIOD
== CharInfo
[(int)'.']);
727 for (unsigned i
= 'a'; i
<= 'z'; ++i
) {
728 assert(CHAR_LETTER
== CharInfo
[i
]);
729 assert(CHAR_LETTER
== CharInfo
[i
+'A'-'a']);
731 for (unsigned i
= '0'; i
<= '9'; ++i
)
732 assert(CHAR_NUMBER
== CharInfo
[i
]);
738 /// isIdentifierBody - Return true if this is the body character of an
739 /// identifier, which is [a-zA-Z0-9_].
740 static inline bool isIdentifierBody(unsigned char c
) {
741 return (CharInfo
[c
] & (CHAR_LETTER
|CHAR_NUMBER
|CHAR_UNDER
)) ? true : false;
744 /// isHorizontalWhitespace - Return true if this character is horizontal
745 /// whitespace: ' ', '\t', '\f', '\v'. Note that this returns false for '\0'.
746 static inline bool isHorizontalWhitespace(unsigned char c
) {
747 return (CharInfo
[c
] & CHAR_HORZ_WS
) ? true : false;
750 /// isWhitespace - Return true if this character is horizontal or vertical
751 /// whitespace: ' ', '\t', '\f', '\v', '\n', '\r'. Note that this returns false
753 static inline bool isWhitespace(unsigned char c
) {
754 return (CharInfo
[c
] & (CHAR_HORZ_WS
|CHAR_VERT_WS
)) ? true : false;
757 /// isNumberBody - Return true if this is the body character of an
758 /// preprocessing number, which is [a-zA-Z0-9_.].
759 static inline bool isNumberBody(unsigned char c
) {
760 return (CharInfo
[c
] & (CHAR_LETTER
|CHAR_NUMBER
|CHAR_UNDER
|CHAR_PERIOD
)) ?
765 //===----------------------------------------------------------------------===//
766 // Diagnostics forwarding code.
767 //===----------------------------------------------------------------------===//
769 /// GetMappedTokenLoc - If lexing out of a 'mapped buffer', where we pretend the
770 /// lexer buffer was all instantiated at a single point, perform the mapping.
771 /// This is currently only used for _Pragma implementation, so it is the slow
772 /// path of the hot getSourceLocation method. Do not allow it to be inlined.
773 static LLVM_ATTRIBUTE_NOINLINE SourceLocation
GetMappedTokenLoc(
774 Preprocessor
&PP
, SourceLocation FileLoc
, unsigned CharNo
, unsigned TokLen
);
775 static SourceLocation
GetMappedTokenLoc(Preprocessor
&PP
,
776 SourceLocation FileLoc
,
777 unsigned CharNo
, unsigned TokLen
) {
778 assert(FileLoc
.isMacroID() && "Must be an instantiation");
780 // Otherwise, we're lexing "mapped tokens". This is used for things like
781 // _Pragma handling. Combine the instantiation location of FileLoc with the
782 // spelling location.
783 SourceManager
&SM
= PP
.getSourceManager();
785 // Create a new SLoc which is expanded from Instantiation(FileLoc) but whose
786 // characters come from spelling(FileLoc)+Offset.
787 SourceLocation SpellingLoc
= SM
.getSpellingLoc(FileLoc
);
788 SpellingLoc
= SpellingLoc
.getFileLocWithOffset(CharNo
);
790 // Figure out the expansion loc range, which is the range covered by the
791 // original _Pragma(...) sequence.
792 std::pair
<SourceLocation
,SourceLocation
> II
=
793 SM
.getImmediateInstantiationRange(FileLoc
);
795 return SM
.createInstantiationLoc(SpellingLoc
, II
.first
, II
.second
, TokLen
);
798 /// getSourceLocation - Return a source location identifier for the specified
799 /// offset in the current file.
800 SourceLocation
Lexer::getSourceLocation(const char *Loc
,
801 unsigned TokLen
) const {
802 assert(Loc
>= BufferStart
&& Loc
<= BufferEnd
&&
803 "Location out of range for this buffer!");
805 // In the normal case, we're just lexing from a simple file buffer, return
806 // the file id from FileLoc with the offset specified.
807 unsigned CharNo
= Loc
-BufferStart
;
808 if (FileLoc
.isFileID())
809 return FileLoc
.getFileLocWithOffset(CharNo
);
811 // Otherwise, this is the _Pragma lexer case, which pretends that all of the
812 // tokens are lexed from where the _Pragma was defined.
813 assert(PP
&& "This doesn't work on raw lexers");
814 return GetMappedTokenLoc(*PP
, FileLoc
, CharNo
, TokLen
);
817 /// Diag - Forwarding function for diagnostics. This translate a source
818 /// position in the current buffer into a SourceLocation object for rendering.
819 DiagnosticBuilder
Lexer::Diag(const char *Loc
, unsigned DiagID
) const {
820 return PP
->Diag(getSourceLocation(Loc
), DiagID
);
823 //===----------------------------------------------------------------------===//
824 // Trigraph and Escaped Newline Handling Code.
825 //===----------------------------------------------------------------------===//
827 /// GetTrigraphCharForLetter - Given a character that occurs after a ?? pair,
828 /// return the decoded trigraph letter it corresponds to, or '\0' if nothing.
829 static char GetTrigraphCharForLetter(char Letter
) {
832 case '=': return '#';
833 case ')': return ']';
834 case '(': return '[';
835 case '!': return '|';
836 case '\'': return '^';
837 case '>': return '}';
838 case '/': return '\\';
839 case '<': return '{';
840 case '-': return '~';
844 /// DecodeTrigraphChar - If the specified character is a legal trigraph when
845 /// prefixed with ??, emit a trigraph warning. If trigraphs are enabled,
846 /// return the result character. Finally, emit a warning about trigraph use
847 /// whether trigraphs are enabled or not.
848 static char DecodeTrigraphChar(const char *CP
, Lexer
*L
) {
849 char Res
= GetTrigraphCharForLetter(*CP
);
850 if (!Res
|| !L
) return Res
;
852 if (!L
->getFeatures().Trigraphs
) {
853 if (!L
->isLexingRawMode())
854 L
->Diag(CP
-2, diag::trigraph_ignored
);
858 if (!L
->isLexingRawMode())
859 L
->Diag(CP
-2, diag::trigraph_converted
) << llvm::StringRef(&Res
, 1);
863 /// getEscapedNewLineSize - Return the size of the specified escaped newline,
864 /// or 0 if it is not an escaped newline. P[-1] is known to be a "\" or a
865 /// trigraph equivalent on entry to this function.
866 unsigned Lexer::getEscapedNewLineSize(const char *Ptr
) {
868 while (isWhitespace(Ptr
[Size
])) {
871 if (Ptr
[Size
-1] != '\n' && Ptr
[Size
-1] != '\r')
874 // If this is a \r\n or \n\r, skip the other half.
875 if ((Ptr
[Size
] == '\r' || Ptr
[Size
] == '\n') &&
876 Ptr
[Size
-1] != Ptr
[Size
])
882 // Not an escaped newline, must be a \t or something else.
886 /// SkipEscapedNewLines - If P points to an escaped newline (or a series of
887 /// them), skip over them and return the first non-escaped-newline found,
888 /// otherwise return P.
889 const char *Lexer::SkipEscapedNewLines(const char *P
) {
891 const char *AfterEscape
;
894 } else if (*P
== '?') {
895 // If not a trigraph for escape, bail out.
896 if (P
[1] != '?' || P
[2] != '/')
903 unsigned NewLineSize
= Lexer::getEscapedNewLineSize(AfterEscape
);
904 if (NewLineSize
== 0) return P
;
905 P
= AfterEscape
+NewLineSize
;
910 /// getCharAndSizeSlow - Peek a single 'character' from the specified buffer,
911 /// get its size, and return it. This is tricky in several cases:
912 /// 1. If currently at the start of a trigraph, we warn about the trigraph,
913 /// then either return the trigraph (skipping 3 chars) or the '?',
914 /// depending on whether trigraphs are enabled or not.
915 /// 2. If this is an escaped newline (potentially with whitespace between
916 /// the backslash and newline), implicitly skip the newline and return
917 /// the char after it.
918 /// 3. If this is a UCN, return it. FIXME: C++ UCN's?
920 /// This handles the slow/uncommon case of the getCharAndSize method. Here we
921 /// know that we can accumulate into Size, and that we have already incremented
922 /// Ptr by Size bytes.
924 /// NOTE: When this method is updated, getCharAndSizeSlowNoWarn (below) should
925 /// be updated to match.
927 char Lexer::getCharAndSizeSlow(const char *Ptr
, unsigned &Size
,
929 // If we have a slash, look for an escaped newline.
930 if (Ptr
[0] == '\\') {
934 // Common case, backslash-char where the char is not whitespace.
935 if (!isWhitespace(Ptr
[0])) return '\\';
937 // See if we have optional whitespace characters between the slash and
939 if (unsigned EscapedNewLineSize
= getEscapedNewLineSize(Ptr
)) {
940 // Remember that this token needs to be cleaned.
941 if (Tok
) Tok
->setFlag(Token::NeedsCleaning
);
943 // Warn if there was whitespace between the backslash and newline.
944 if (Ptr
[0] != '\n' && Ptr
[0] != '\r' && Tok
&& !isLexingRawMode())
945 Diag(Ptr
, diag::backslash_newline_space
);
947 // Found backslash<whitespace><newline>. Parse the char after it.
948 Size
+= EscapedNewLineSize
;
949 Ptr
+= EscapedNewLineSize
;
950 // Use slow version to accumulate a correct size field.
951 return getCharAndSizeSlow(Ptr
, Size
, Tok
);
954 // Otherwise, this is not an escaped newline, just return the slash.
958 // If this is a trigraph, process it.
959 if (Ptr
[0] == '?' && Ptr
[1] == '?') {
960 // If this is actually a legal trigraph (not something like "??x"), emit
961 // a trigraph warning. If so, and if trigraphs are enabled, return it.
962 if (char C
= DecodeTrigraphChar(Ptr
+2, Tok
? this : 0)) {
963 // Remember that this token needs to be cleaned.
964 if (Tok
) Tok
->setFlag(Token::NeedsCleaning
);
968 if (C
== '\\') goto Slash
;
973 // If this is neither, return a single character.
979 /// getCharAndSizeSlowNoWarn - Handle the slow/uncommon case of the
980 /// getCharAndSizeNoWarn method. Here we know that we can accumulate into Size,
981 /// and that we have already incremented Ptr by Size bytes.
983 /// NOTE: When this method is updated, getCharAndSizeSlow (above) should
984 /// be updated to match.
985 char Lexer::getCharAndSizeSlowNoWarn(const char *Ptr
, unsigned &Size
,
986 const LangOptions
&Features
) {
987 // If we have a slash, look for an escaped newline.
988 if (Ptr
[0] == '\\') {
992 // Common case, backslash-char where the char is not whitespace.
993 if (!isWhitespace(Ptr
[0])) return '\\';
995 // See if we have optional whitespace characters followed by a newline.
996 if (unsigned EscapedNewLineSize
= getEscapedNewLineSize(Ptr
)) {
997 // Found backslash<whitespace><newline>. Parse the char after it.
998 Size
+= EscapedNewLineSize
;
999 Ptr
+= EscapedNewLineSize
;
1001 // Use slow version to accumulate a correct size field.
1002 return getCharAndSizeSlowNoWarn(Ptr
, Size
, Features
);
1005 // Otherwise, this is not an escaped newline, just return the slash.
1009 // If this is a trigraph, process it.
1010 if (Features
.Trigraphs
&& Ptr
[0] == '?' && Ptr
[1] == '?') {
1011 // If this is actually a legal trigraph (not something like "??x"), return
1013 if (char C
= GetTrigraphCharForLetter(Ptr
[2])) {
1016 if (C
== '\\') goto Slash
;
1021 // If this is neither, return a single character.
1026 //===----------------------------------------------------------------------===//
1027 // Helper methods for lexing.
1028 //===----------------------------------------------------------------------===//
1030 /// \brief Routine that indiscriminately skips bytes in the source file.
1031 void Lexer::SkipBytes(unsigned Bytes
, bool StartOfLine
) {
1033 if (BufferPtr
> BufferEnd
)
1034 BufferPtr
= BufferEnd
;
1035 IsAtStartOfLine
= StartOfLine
;
1038 void Lexer::LexIdentifier(Token
&Result
, const char *CurPtr
) {
1039 // Match [_A-Za-z0-9]*, we have already matched [_A-Za-z$]
1041 unsigned char C
= *CurPtr
++;
1042 while (isIdentifierBody(C
))
1045 --CurPtr
; // Back up over the skipped character.
1047 // Fast path, no $,\,? in identifier found. '\' might be an escaped newline
1048 // or UCN, and ? might be a trigraph for '\', an escaped newline or UCN.
1051 // TODO: Could merge these checks into a CharInfo flag to make the comparison
1053 if (C
!= '\\' && C
!= '?' && (C
!= '$' || !Features
.DollarIdents
)) {
1055 const char *IdStart
= BufferPtr
;
1056 FormTokenWithChars(Result
, CurPtr
, tok::raw_identifier
);
1057 Result
.setRawIdentifierData(IdStart
);
1059 // If we are in raw mode, return this identifier raw. There is no need to
1060 // look up identifier information or attempt to macro expand it.
1064 // Fill in Result.IdentifierInfo and update the token kind,
1065 // looking up the identifier in the identifier table.
1066 IdentifierInfo
*II
= PP
->LookUpIdentifierInfo(Result
);
1068 // Finally, now that we know we have an identifier, pass this off to the
1069 // preprocessor, which may macro expand it or something.
1070 if (II
->isHandleIdentifierCase())
1071 PP
->HandleIdentifier(Result
);
1075 // Otherwise, $,\,? in identifier found. Enter slower path.
1077 C
= getCharAndSize(CurPtr
, Size
);
1080 // If we hit a $ and they are not supported in identifiers, we are done.
1081 if (!Features
.DollarIdents
) goto FinishIdentifier
;
1083 // Otherwise, emit a diagnostic and continue.
1084 if (!isLexingRawMode())
1085 Diag(CurPtr
, diag::ext_dollar_in_identifier
);
1086 CurPtr
= ConsumeChar(CurPtr
, Size
, Result
);
1087 C
= getCharAndSize(CurPtr
, Size
);
1089 } else if (!isIdentifierBody(C
)) { // FIXME: UCNs.
1090 // Found end of identifier.
1091 goto FinishIdentifier
;
1094 // Otherwise, this character is good, consume it.
1095 CurPtr
= ConsumeChar(CurPtr
, Size
, Result
);
1097 C
= getCharAndSize(CurPtr
, Size
);
1098 while (isIdentifierBody(C
)) { // FIXME: UCNs.
1099 CurPtr
= ConsumeChar(CurPtr
, Size
, Result
);
1100 C
= getCharAndSize(CurPtr
, Size
);
1105 /// isHexaLiteral - Return true if Start points to a hex constant.
1106 /// in microsoft mode (where this is supposed to be several different tokens).
1107 static bool isHexaLiteral(const char *Start
, const LangOptions
&Features
) {
1109 char C1
= Lexer::getCharAndSizeNoWarn(Start
, Size
, Features
);
1112 char C2
= Lexer::getCharAndSizeNoWarn(Start
+ Size
, Size
, Features
);
1113 return (C2
== 'x' || C2
== 'X');
1116 /// LexNumericConstant - Lex the remainder of a integer or floating point
1117 /// constant. From[-1] is the first character lexed. Return the end of the
1119 void Lexer::LexNumericConstant(Token
&Result
, const char *CurPtr
) {
1121 char C
= getCharAndSize(CurPtr
, Size
);
1123 while (isNumberBody(C
)) { // FIXME: UCNs?
1124 CurPtr
= ConsumeChar(CurPtr
, Size
, Result
);
1126 C
= getCharAndSize(CurPtr
, Size
);
1129 // If we fell out, check for a sign, due to 1e+12. If we have one, continue.
1130 if ((C
== '-' || C
== '+') && (PrevCh
== 'E' || PrevCh
== 'e')) {
1131 // If we are in Microsoft mode, don't continue if the constant is hex.
1132 // For example, MSVC will accept the following as 3 tokens: 0x1234567e+1
1133 if (!Features
.Microsoft
|| !isHexaLiteral(BufferPtr
, Features
))
1134 return LexNumericConstant(Result
, ConsumeChar(CurPtr
, Size
, Result
));
1137 // If we have a hex FP constant, continue.
1138 if ((C
== '-' || C
== '+') && (PrevCh
== 'P' || PrevCh
== 'p') &&
1139 !Features
.CPlusPlus0x
)
1140 return LexNumericConstant(Result
, ConsumeChar(CurPtr
, Size
, Result
));
1142 // Update the location of token as well as BufferPtr.
1143 const char *TokStart
= BufferPtr
;
1144 FormTokenWithChars(Result
, CurPtr
, tok::numeric_constant
);
1145 Result
.setLiteralData(TokStart
);
1148 /// LexStringLiteral - Lex the remainder of a string literal, after having lexed
1150 void Lexer::LexStringLiteral(Token
&Result
, const char *CurPtr
, bool Wide
) {
1151 const char *NulCharacter
= 0; // Does this string contain the \0 character?
1153 char C
= getAndAdvanceChar(CurPtr
, Result
);
1155 // Skip escaped characters. Escaped newlines will already be processed by
1156 // getAndAdvanceChar.
1158 C
= getAndAdvanceChar(CurPtr
, Result
);
1160 if (C
== '\n' || C
== '\r' || // Newline.
1161 (C
== 0 && CurPtr
-1 == BufferEnd
)) { // End of file.
1162 if (C
== 0 && PP
&& PP
->isCodeCompletionFile(FileLoc
))
1163 PP
->CodeCompleteNaturalLanguage();
1164 else if (!isLexingRawMode() && !Features
.AsmPreprocessor
)
1165 Diag(BufferPtr
, diag::warn_unterminated_string
);
1166 FormTokenWithChars(Result
, CurPtr
-1, tok::unknown
);
1171 NulCharacter
= CurPtr
-1;
1172 C
= getAndAdvanceChar(CurPtr
, Result
);
1175 // If a nul character existed in the string, warn about it.
1176 if (NulCharacter
&& !isLexingRawMode())
1177 Diag(NulCharacter
, diag::null_in_string
);
1179 // Update the location of the token as well as the BufferPtr instance var.
1180 const char *TokStart
= BufferPtr
;
1181 FormTokenWithChars(Result
, CurPtr
,
1182 Wide
? tok::wide_string_literal
: tok::string_literal
);
1183 Result
.setLiteralData(TokStart
);
1186 /// LexAngledStringLiteral - Lex the remainder of an angled string literal,
1187 /// after having lexed the '<' character. This is used for #include filenames.
1188 void Lexer::LexAngledStringLiteral(Token
&Result
, const char *CurPtr
) {
1189 const char *NulCharacter
= 0; // Does this string contain the \0 character?
1190 const char *AfterLessPos
= CurPtr
;
1191 char C
= getAndAdvanceChar(CurPtr
, Result
);
1193 // Skip escaped characters.
1195 // Skip the escaped character.
1196 C
= getAndAdvanceChar(CurPtr
, Result
);
1197 } else if (C
== '\n' || C
== '\r' || // Newline.
1198 (C
== 0 && CurPtr
-1 == BufferEnd
)) { // End of file.
1199 // If the filename is unterminated, then it must just be a lone <
1200 // character. Return this as such.
1201 FormTokenWithChars(Result
, AfterLessPos
, tok::less
);
1203 } else if (C
== 0) {
1204 NulCharacter
= CurPtr
-1;
1206 C
= getAndAdvanceChar(CurPtr
, Result
);
1209 // If a nul character existed in the string, warn about it.
1210 if (NulCharacter
&& !isLexingRawMode())
1211 Diag(NulCharacter
, diag::null_in_string
);
1213 // Update the location of token as well as BufferPtr.
1214 const char *TokStart
= BufferPtr
;
1215 FormTokenWithChars(Result
, CurPtr
, tok::angle_string_literal
);
1216 Result
.setLiteralData(TokStart
);
1220 /// LexCharConstant - Lex the remainder of a character constant, after having
1221 /// lexed either ' or L'.
1222 void Lexer::LexCharConstant(Token
&Result
, const char *CurPtr
) {
1223 const char *NulCharacter
= 0; // Does this character contain the \0 character?
1225 char C
= getAndAdvanceChar(CurPtr
, Result
);
1227 if (!isLexingRawMode() && !Features
.AsmPreprocessor
)
1228 Diag(BufferPtr
, diag::err_empty_character
);
1229 FormTokenWithChars(Result
, CurPtr
, tok::unknown
);
1234 // Skip escaped characters.
1236 // Skip the escaped character.
1238 C
= getAndAdvanceChar(CurPtr
, Result
);
1239 } else if (C
== '\n' || C
== '\r' || // Newline.
1240 (C
== 0 && CurPtr
-1 == BufferEnd
)) { // End of file.
1241 if (C
== 0 && PP
&& PP
->isCodeCompletionFile(FileLoc
))
1242 PP
->CodeCompleteNaturalLanguage();
1243 else if (!isLexingRawMode() && !Features
.AsmPreprocessor
)
1244 Diag(BufferPtr
, diag::warn_unterminated_char
);
1245 FormTokenWithChars(Result
, CurPtr
-1, tok::unknown
);
1247 } else if (C
== 0) {
1248 NulCharacter
= CurPtr
-1;
1250 C
= getAndAdvanceChar(CurPtr
, Result
);
1253 // If a nul character existed in the character, warn about it.
1254 if (NulCharacter
&& !isLexingRawMode())
1255 Diag(NulCharacter
, diag::null_in_char
);
1257 // Update the location of token as well as BufferPtr.
1258 const char *TokStart
= BufferPtr
;
1259 FormTokenWithChars(Result
, CurPtr
, tok::char_constant
);
1260 Result
.setLiteralData(TokStart
);
1263 /// SkipWhitespace - Efficiently skip over a series of whitespace characters.
1264 /// Update BufferPtr to point to the next non-whitespace character and return.
1266 /// This method forms a token and returns true if KeepWhitespaceMode is enabled.
1268 bool Lexer::SkipWhitespace(Token
&Result
, const char *CurPtr
) {
1269 // Whitespace - Skip it, then return the token after the whitespace.
1270 unsigned char Char
= *CurPtr
; // Skip consequtive spaces efficiently.
1272 // Skip horizontal whitespace very aggressively.
1273 while (isHorizontalWhitespace(Char
))
1276 // Otherwise if we have something other than whitespace, we're done.
1277 if (Char
!= '\n' && Char
!= '\r')
1280 if (ParsingPreprocessorDirective
) {
1281 // End of preprocessor directive line, let LexTokenInternal handle this.
1286 // ok, but handle newline.
1287 // The returned token is at the start of the line.
1288 Result
.setFlag(Token::StartOfLine
);
1289 // No leading whitespace seen so far.
1290 Result
.clearFlag(Token::LeadingSpace
);
1294 // If this isn't immediately after a newline, there is leading space.
1295 char PrevChar
= CurPtr
[-1];
1296 if (PrevChar
!= '\n' && PrevChar
!= '\r')
1297 Result
.setFlag(Token::LeadingSpace
);
1299 // If the client wants us to return whitespace, return it now.
1300 if (isKeepWhitespaceMode()) {
1301 FormTokenWithChars(Result
, CurPtr
, tok::unknown
);
1309 // SkipBCPLComment - We have just read the // characters from input. Skip until
1310 // we find the newline character thats terminate the comment. Then update
1311 /// BufferPtr and return.
1313 /// If we're in KeepCommentMode or any CommentHandler has inserted
1314 /// some tokens, this will store the first token and return true.
1315 bool Lexer::SkipBCPLComment(Token
&Result
, const char *CurPtr
) {
1316 // If BCPL comments aren't explicitly enabled for this language, emit an
1317 // extension warning.
1318 if (!Features
.BCPLComment
&& !isLexingRawMode()) {
1319 Diag(BufferPtr
, diag::ext_bcpl_comment
);
1321 // Mark them enabled so we only emit one warning for this translation
1323 Features
.BCPLComment
= true;
1326 // Scan over the body of the comment. The common case, when scanning, is that
1327 // the comment contains normal ascii characters with nothing interesting in
1328 // them. As such, optimize for this case with the inner loop.
1332 // FIXME: Speedup BCPL comment lexing. Just scan for a \n or \r character.
1333 // If we find a \n character, scan backwards, checking to see if it's an
1334 // escaped newline, like we do for block comments.
1336 // Skip over characters in the fast loop.
1337 while (C
!= 0 && // Potentially EOF.
1338 C
!= '\\' && // Potentially escaped newline.
1339 C
!= '?' && // Potentially trigraph.
1340 C
!= '\n' && C
!= '\r') // Newline or DOS-style newline.
1343 // If this is a newline, we're done.
1344 if (C
== '\n' || C
== '\r')
1345 break; // Found the newline? Break out!
1347 // Otherwise, this is a hard case. Fall back on getAndAdvanceChar to
1348 // properly decode the character. Read it in raw mode to avoid emitting
1349 // diagnostics about things like trigraphs. If we see an escaped newline,
1350 // we'll handle it below.
1351 const char *OldPtr
= CurPtr
;
1352 bool OldRawMode
= isLexingRawMode();
1353 LexingRawMode
= true;
1354 C
= getAndAdvanceChar(CurPtr
, Result
);
1355 LexingRawMode
= OldRawMode
;
1357 // If the char that we finally got was a \n, then we must have had something
1358 // like \<newline><newline>. We don't want to have consumed the second
1359 // newline, we want CurPtr, to end up pointing to it down below.
1360 if (C
== '\n' || C
== '\r') {
1362 C
= 'x'; // doesn't matter what this is.
1365 // If we read multiple characters, and one of those characters was a \r or
1366 // \n, then we had an escaped newline within the comment. Emit diagnostic
1367 // unless the next line is also a // comment.
1368 if (CurPtr
!= OldPtr
+1 && C
!= '/' && CurPtr
[0] != '/') {
1369 for (; OldPtr
!= CurPtr
; ++OldPtr
)
1370 if (OldPtr
[0] == '\n' || OldPtr
[0] == '\r') {
1371 // Okay, we found a // comment that ends in a newline, if the next
1372 // line is also a // comment, but has spaces, don't emit a diagnostic.
1374 const char *ForwardPtr
= CurPtr
;
1375 while (isspace(*ForwardPtr
)) // Skip whitespace.
1377 if (ForwardPtr
[0] == '/' && ForwardPtr
[1] == '/')
1381 if (!isLexingRawMode())
1382 Diag(OldPtr
-1, diag::ext_multi_line_bcpl_comment
);
1387 if (CurPtr
== BufferEnd
+1) {
1388 if (PP
&& PP
->isCodeCompletionFile(FileLoc
))
1389 PP
->CodeCompleteNaturalLanguage();
1394 } while (C
!= '\n' && C
!= '\r');
1396 // Found but did not consume the newline. Notify comment handlers about the
1397 // comment unless we're in a #if 0 block.
1398 if (PP
&& !isLexingRawMode() &&
1399 PP
->HandleComment(Result
, SourceRange(getSourceLocation(BufferPtr
),
1400 getSourceLocation(CurPtr
)))) {
1402 return true; // A token has to be returned.
1405 // If we are returning comments as tokens, return this comment as a token.
1406 if (inKeepCommentMode())
1407 return SaveBCPLComment(Result
, CurPtr
);
1409 // If we are inside a preprocessor directive and we see the end of line,
1410 // return immediately, so that the lexer can return this as an EOM token.
1411 if (ParsingPreprocessorDirective
|| CurPtr
== BufferEnd
) {
1416 // Otherwise, eat the \n character. We don't care if this is a \n\r or
1417 // \r\n sequence. This is an efficiency hack (because we know the \n can't
1418 // contribute to another token), it isn't needed for correctness. Note that
1419 // this is ok even in KeepWhitespaceMode, because we would have returned the
1420 /// comment above in that mode.
1423 // The next returned token is at the start of the line.
1424 Result
.setFlag(Token::StartOfLine
);
1425 // No leading whitespace seen so far.
1426 Result
.clearFlag(Token::LeadingSpace
);
1431 /// SaveBCPLComment - If in save-comment mode, package up this BCPL comment in
1432 /// an appropriate way and return it.
1433 bool Lexer::SaveBCPLComment(Token
&Result
, const char *CurPtr
) {
1434 // If we're not in a preprocessor directive, just return the // comment
1436 FormTokenWithChars(Result
, CurPtr
, tok::comment
);
1438 if (!ParsingPreprocessorDirective
)
1441 // If this BCPL-style comment is in a macro definition, transmogrify it into
1442 // a C-style block comment.
1443 bool Invalid
= false;
1444 std::string Spelling
= PP
->getSpelling(Result
, &Invalid
);
1448 assert(Spelling
[0] == '/' && Spelling
[1] == '/' && "Not bcpl comment?");
1449 Spelling
[1] = '*'; // Change prefix to "/*".
1450 Spelling
+= "*/"; // add suffix.
1452 Result
.setKind(tok::comment
);
1453 PP
->CreateString(&Spelling
[0], Spelling
.size(), Result
,
1454 Result
.getLocation());
1458 /// isBlockCommentEndOfEscapedNewLine - Return true if the specified newline
1459 /// character (either \n or \r) is part of an escaped newline sequence. Issue a
1460 /// diagnostic if so. We know that the newline is inside of a block comment.
1461 static bool isEndOfBlockCommentWithEscapedNewLine(const char *CurPtr
,
1463 assert(CurPtr
[0] == '\n' || CurPtr
[0] == '\r');
1465 // Back up off the newline.
1468 // If this is a two-character newline sequence, skip the other character.
1469 if (CurPtr
[0] == '\n' || CurPtr
[0] == '\r') {
1470 // \n\n or \r\r -> not escaped newline.
1471 if (CurPtr
[0] == CurPtr
[1])
1473 // \n\r or \r\n -> skip the newline.
1477 // If we have horizontal whitespace, skip over it. We allow whitespace
1478 // between the slash and newline.
1479 bool HasSpace
= false;
1480 while (isHorizontalWhitespace(*CurPtr
) || *CurPtr
== 0) {
1485 // If we have a slash, we know this is an escaped newline.
1486 if (*CurPtr
== '\\') {
1487 if (CurPtr
[-1] != '*') return false;
1489 // It isn't a slash, is it the ?? / trigraph?
1490 if (CurPtr
[0] != '/' || CurPtr
[-1] != '?' || CurPtr
[-2] != '?' ||
1494 // This is the trigraph ending the comment. Emit a stern warning!
1497 // If no trigraphs are enabled, warn that we ignored this trigraph and
1498 // ignore this * character.
1499 if (!L
->getFeatures().Trigraphs
) {
1500 if (!L
->isLexingRawMode())
1501 L
->Diag(CurPtr
, diag::trigraph_ignored_block_comment
);
1504 if (!L
->isLexingRawMode())
1505 L
->Diag(CurPtr
, diag::trigraph_ends_block_comment
);
1508 // Warn about having an escaped newline between the */ characters.
1509 if (!L
->isLexingRawMode())
1510 L
->Diag(CurPtr
, diag::escaped_newline_block_comment_end
);
1512 // If there was space between the backslash and newline, warn about it.
1513 if (HasSpace
&& !L
->isLexingRawMode())
1514 L
->Diag(CurPtr
, diag::backslash_newline_space
);
1520 #include <emmintrin.h>
1522 #include <altivec.h>
1526 /// SkipBlockComment - We have just read the /* characters from input. Read
1527 /// until we find the */ characters that terminate the comment. Note that we
1528 /// don't bother decoding trigraphs or escaped newlines in block comments,
1529 /// because they cannot cause the comment to end. The only thing that can
1530 /// happen is the comment could end with an escaped newline between the */ end
1533 /// If we're in KeepCommentMode or any CommentHandler has inserted
1534 /// some tokens, this will store the first token and return true.
1535 bool Lexer::SkipBlockComment(Token
&Result
, const char *CurPtr
) {
1536 // Scan one character past where we should, looking for a '/' character. Once
1537 // we find it, check to see if it was preceeded by a *. This common
1538 // optimization helps people who like to put a lot of * characters in their
1541 // The first character we get with newlines and trigraphs skipped to handle
1542 // the degenerate /*/ case below correctly if the * has an escaped newline
1545 unsigned char C
= getCharAndSize(CurPtr
, CharSize
);
1547 if (C
== 0 && CurPtr
== BufferEnd
+1) {
1548 if (!isLexingRawMode() &&
1549 !PP
->isCodeCompletionFile(FileLoc
))
1550 Diag(BufferPtr
, diag::err_unterminated_block_comment
);
1553 // KeepWhitespaceMode should return this broken comment as a token. Since
1554 // it isn't a well formed comment, just return it as an 'unknown' token.
1555 if (isKeepWhitespaceMode()) {
1556 FormTokenWithChars(Result
, CurPtr
, tok::unknown
);
1564 // Check to see if the first character after the '/*' is another /. If so,
1565 // then this slash does not end the block comment, it is part of it.
1570 // Skip over all non-interesting characters until we find end of buffer or a
1571 // (probably ending) '/' character.
1572 if (CurPtr
+ 24 < BufferEnd
) {
1573 // While not aligned to a 16-byte boundary.
1574 while (C
!= '/' && ((intptr_t)CurPtr
& 0x0F) != 0)
1577 if (C
== '/') goto FoundSlash
;
1580 __m128i Slashes
= _mm_set_epi8('/', '/', '/', '/', '/', '/', '/', '/',
1581 '/', '/', '/', '/', '/', '/', '/', '/');
1582 while (CurPtr
+16 <= BufferEnd
&&
1583 _mm_movemask_epi8(_mm_cmpeq_epi8(*(__m128i
*)CurPtr
, Slashes
)) == 0)
1586 __vector
unsigned char Slashes
= {
1587 '/', '/', '/', '/', '/', '/', '/', '/',
1588 '/', '/', '/', '/', '/', '/', '/', '/'
1590 while (CurPtr
+16 <= BufferEnd
&&
1591 !vec_any_eq(*(vector
unsigned char*)CurPtr
, Slashes
))
1594 // Scan for '/' quickly. Many block comments are very large.
1595 while (CurPtr
[0] != '/' &&
1599 CurPtr
+4 < BufferEnd
) {
1604 // It has to be one of the bytes scanned, increment to it and read one.
1608 // Loop to scan the remainder.
1609 while (C
!= '/' && C
!= '\0')
1614 if (CurPtr
[-2] == '*') // We found the final */. We're done!
1617 if ((CurPtr
[-2] == '\n' || CurPtr
[-2] == '\r')) {
1618 if (isEndOfBlockCommentWithEscapedNewLine(CurPtr
-2, this)) {
1619 // We found the final */, though it had an escaped newline between the
1620 // * and /. We're done!
1624 if (CurPtr
[0] == '*' && CurPtr
[1] != '/') {
1625 // If this is a /* inside of the comment, emit a warning. Don't do this
1626 // if this is a /*/, which will end the comment. This misses cases with
1627 // embedded escaped newlines, but oh well.
1628 if (!isLexingRawMode())
1629 Diag(CurPtr
-1, diag::warn_nested_block_comment
);
1631 } else if (C
== 0 && CurPtr
== BufferEnd
+1) {
1632 if (PP
&& PP
->isCodeCompletionFile(FileLoc
))
1633 PP
->CodeCompleteNaturalLanguage();
1634 else if (!isLexingRawMode())
1635 Diag(BufferPtr
, diag::err_unterminated_block_comment
);
1636 // Note: the user probably forgot a */. We could continue immediately
1637 // after the /*, but this would involve lexing a lot of what really is the
1638 // comment, which surely would confuse the parser.
1641 // KeepWhitespaceMode should return this broken comment as a token. Since
1642 // it isn't a well formed comment, just return it as an 'unknown' token.
1643 if (isKeepWhitespaceMode()) {
1644 FormTokenWithChars(Result
, CurPtr
, tok::unknown
);
1654 // Notify comment handlers about the comment unless we're in a #if 0 block.
1655 if (PP
&& !isLexingRawMode() &&
1656 PP
->HandleComment(Result
, SourceRange(getSourceLocation(BufferPtr
),
1657 getSourceLocation(CurPtr
)))) {
1659 return true; // A token has to be returned.
1662 // If we are returning comments as tokens, return this comment as a token.
1663 if (inKeepCommentMode()) {
1664 FormTokenWithChars(Result
, CurPtr
, tok::comment
);
1668 // It is common for the tokens immediately after a /**/ comment to be
1669 // whitespace. Instead of going through the big switch, handle it
1670 // efficiently now. This is safe even in KeepWhitespaceMode because we would
1671 // have already returned above with the comment as a token.
1672 if (isHorizontalWhitespace(*CurPtr
)) {
1673 Result
.setFlag(Token::LeadingSpace
);
1674 SkipWhitespace(Result
, CurPtr
+1);
1678 // Otherwise, just return so that the next character will be lexed as a token.
1680 Result
.setFlag(Token::LeadingSpace
);
1684 //===----------------------------------------------------------------------===//
1685 // Primary Lexing Entry Points
1686 //===----------------------------------------------------------------------===//
1688 /// ReadToEndOfLine - Read the rest of the current preprocessor line as an
1689 /// uninterpreted string. This switches the lexer out of directive mode.
1690 std::string
Lexer::ReadToEndOfLine() {
1691 assert(ParsingPreprocessorDirective
&& ParsingFilename
== false &&
1692 "Must be in a preprocessing directive!");
1696 // CurPtr - Cache BufferPtr in an automatic variable.
1697 const char *CurPtr
= BufferPtr
;
1699 char Char
= getAndAdvanceChar(CurPtr
, Tmp
);
1705 // Found end of file?
1706 if (CurPtr
-1 != BufferEnd
) {
1707 // Nope, normal character, continue.
1714 // Okay, we found the end of the line. First, back up past the \0, \r, \n.
1715 assert(CurPtr
[-1] == Char
&& "Trigraphs for newline?");
1716 BufferPtr
= CurPtr
-1;
1718 // Next, lex the character, which should handle the EOM transition.
1720 if (Tmp
.is(tok::code_completion
)) {
1721 if (PP
&& PP
->getCodeCompletionHandler())
1722 PP
->getCodeCompletionHandler()->CodeCompleteNaturalLanguage();
1725 assert(Tmp
.is(tok::eom
) && "Unexpected token!");
1727 // Finally, we're done, return the string we found.
1733 /// LexEndOfFile - CurPtr points to the end of this file. Handle this
1734 /// condition, reporting diagnostics and handling other edge cases as required.
1735 /// This returns true if Result contains a token, false if PP.Lex should be
1737 bool Lexer::LexEndOfFile(Token
&Result
, const char *CurPtr
) {
1738 // Check if we are performing code completion.
1739 if (PP
&& PP
->isCodeCompletionFile(FileLoc
)) {
1740 // We're at the end of the file, but we've been asked to consider the
1741 // end of the file to be a code-completion token. Return the
1742 // code-completion token.
1743 Result
.startToken();
1744 FormTokenWithChars(Result
, CurPtr
, tok::code_completion
);
1746 // Only do the eof -> code_completion translation once.
1747 PP
->SetCodeCompletionPoint(0, 0, 0);
1749 // Silence any diagnostics that occur once we hit the code-completion point.
1750 PP
->getDiagnostics().setSuppressAllDiagnostics(true);
1754 // If we hit the end of the file while parsing a preprocessor directive,
1755 // end the preprocessor directive first. The next token returned will
1756 // then be the end of file.
1757 if (ParsingPreprocessorDirective
) {
1758 // Done parsing the "line".
1759 ParsingPreprocessorDirective
= false;
1760 // Update the location of token as well as BufferPtr.
1761 FormTokenWithChars(Result
, CurPtr
, tok::eom
);
1763 // Restore comment saving mode, in case it was disabled for directive.
1764 SetCommentRetentionState(PP
->getCommentRetentionState());
1765 return true; // Have a token.
1768 // If we are in raw mode, return this event as an EOF token. Let the caller
1769 // that put us in raw mode handle the event.
1770 if (isLexingRawMode()) {
1771 Result
.startToken();
1772 BufferPtr
= BufferEnd
;
1773 FormTokenWithChars(Result
, BufferEnd
, tok::eof
);
1777 // Issue diagnostics for unterminated #if and missing newline.
1779 // If we are in a #if directive, emit an error.
1780 while (!ConditionalStack
.empty()) {
1781 if (!PP
->isCodeCompletionFile(FileLoc
))
1782 PP
->Diag(ConditionalStack
.back().IfLoc
,
1783 diag::err_pp_unterminated_conditional
);
1784 ConditionalStack
.pop_back();
1787 // C99 5.1.1.2p2: If the file is non-empty and didn't end in a newline, issue
1789 if (CurPtr
!= BufferStart
&& (CurPtr
[-1] != '\n' && CurPtr
[-1] != '\r'))
1790 Diag(BufferEnd
, diag::ext_no_newline_eof
)
1791 << FixItHint::CreateInsertion(getSourceLocation(BufferEnd
), "\n");
1795 // Finally, let the preprocessor handle this.
1796 return PP
->HandleEndOfFile(Result
);
1799 /// isNextPPTokenLParen - Return 1 if the next unexpanded token lexed from
1800 /// the specified lexer will return a tok::l_paren token, 0 if it is something
1801 /// else and 2 if there are no more tokens in the buffer controlled by the
1803 unsigned Lexer::isNextPPTokenLParen() {
1804 assert(!LexingRawMode
&& "How can we expand a macro from a skipping buffer?");
1806 // Switch to 'skipping' mode. This will ensure that we can lex a token
1807 // without emitting diagnostics, disables macro expansion, and will cause EOF
1808 // to return an EOF token instead of popping the include stack.
1809 LexingRawMode
= true;
1811 // Save state that can be changed while lexing so that we can restore it.
1812 const char *TmpBufferPtr
= BufferPtr
;
1813 bool inPPDirectiveMode
= ParsingPreprocessorDirective
;
1817 LexTokenInternal(Tok
);
1819 // Restore state that may have changed.
1820 BufferPtr
= TmpBufferPtr
;
1821 ParsingPreprocessorDirective
= inPPDirectiveMode
;
1823 // Restore the lexer back to non-skipping mode.
1824 LexingRawMode
= false;
1826 if (Tok
.is(tok::eof
))
1828 return Tok
.is(tok::l_paren
);
1831 /// FindConflictEnd - Find the end of a version control conflict marker.
1832 static const char *FindConflictEnd(const char *CurPtr
, const char *BufferEnd
) {
1833 llvm::StringRef
RestOfBuffer(CurPtr
+7, BufferEnd
-CurPtr
-7);
1834 size_t Pos
= RestOfBuffer
.find(">>>>>>>");
1835 while (Pos
!= llvm::StringRef::npos
) {
1836 // Must occur at start of line.
1837 if (RestOfBuffer
[Pos
-1] != '\r' &&
1838 RestOfBuffer
[Pos
-1] != '\n') {
1839 RestOfBuffer
= RestOfBuffer
.substr(Pos
+7);
1840 Pos
= RestOfBuffer
.find(">>>>>>>");
1843 return RestOfBuffer
.data()+Pos
;
1848 /// IsStartOfConflictMarker - If the specified pointer is the start of a version
1849 /// control conflict marker like '<<<<<<<', recognize it as such, emit an error
1850 /// and recover nicely. This returns true if it is a conflict marker and false
1852 bool Lexer::IsStartOfConflictMarker(const char *CurPtr
) {
1853 // Only a conflict marker if it starts at the beginning of a line.
1854 if (CurPtr
!= BufferStart
&&
1855 CurPtr
[-1] != '\n' && CurPtr
[-1] != '\r')
1858 // Check to see if we have <<<<<<<.
1859 if (BufferEnd
-CurPtr
< 8 ||
1860 llvm::StringRef(CurPtr
, 7) != "<<<<<<<")
1863 // If we have a situation where we don't care about conflict markers, ignore
1865 if (IsInConflictMarker
|| isLexingRawMode())
1868 // Check to see if there is a >>>>>>> somewhere in the buffer at the start of
1869 // a line to terminate this conflict marker.
1870 if (FindConflictEnd(CurPtr
, BufferEnd
)) {
1871 // We found a match. We are really in a conflict marker.
1872 // Diagnose this, and ignore to the end of line.
1873 Diag(CurPtr
, diag::err_conflict_marker
);
1874 IsInConflictMarker
= true;
1876 // Skip ahead to the end of line. We know this exists because the
1877 // end-of-conflict marker starts with \r or \n.
1878 while (*CurPtr
!= '\r' && *CurPtr
!= '\n') {
1879 assert(CurPtr
!= BufferEnd
&& "Didn't find end of line");
1886 // No end of conflict marker found.
1891 /// HandleEndOfConflictMarker - If this is a '=======' or '|||||||' or '>>>>>>>'
1892 /// marker, then it is the end of a conflict marker. Handle it by ignoring up
1893 /// until the end of the line. This returns true if it is a conflict marker and
1895 bool Lexer::HandleEndOfConflictMarker(const char *CurPtr
) {
1896 // Only a conflict marker if it starts at the beginning of a line.
1897 if (CurPtr
!= BufferStart
&&
1898 CurPtr
[-1] != '\n' && CurPtr
[-1] != '\r')
1901 // If we have a situation where we don't care about conflict markers, ignore
1903 if (!IsInConflictMarker
|| isLexingRawMode())
1906 // Check to see if we have the marker (7 characters in a row).
1907 for (unsigned i
= 1; i
!= 7; ++i
)
1908 if (CurPtr
[i
] != CurPtr
[0])
1911 // If we do have it, search for the end of the conflict marker. This could
1912 // fail if it got skipped with a '#if 0' or something. Note that CurPtr might
1913 // be the end of conflict marker.
1914 if (const char *End
= FindConflictEnd(CurPtr
, BufferEnd
)) {
1917 // Skip ahead to the end of line.
1918 while (CurPtr
!= BufferEnd
&& *CurPtr
!= '\r' && *CurPtr
!= '\n')
1923 // No longer in the conflict marker.
1924 IsInConflictMarker
= false;
1932 /// LexTokenInternal - This implements a simple C family lexer. It is an
1933 /// extremely performance critical piece of code. This assumes that the buffer
1934 /// has a null character at the end of the file. This returns a preprocessing
1935 /// token, not a normal token, as such, it is an internal interface. It assumes
1936 /// that the Flags of result have been cleared before calling this.
1937 void Lexer::LexTokenInternal(Token
&Result
) {
1939 // New token, can't need cleaning yet.
1940 Result
.clearFlag(Token::NeedsCleaning
);
1941 Result
.setIdentifierInfo(0);
1943 // CurPtr - Cache BufferPtr in an automatic variable.
1944 const char *CurPtr
= BufferPtr
;
1946 // Small amounts of horizontal whitespace is very common between tokens.
1947 if ((*CurPtr
== ' ') || (*CurPtr
== '\t')) {
1949 while ((*CurPtr
== ' ') || (*CurPtr
== '\t'))
1952 // If we are keeping whitespace and other tokens, just return what we just
1953 // skipped. The next lexer invocation will return the token after the
1955 if (isKeepWhitespaceMode()) {
1956 FormTokenWithChars(Result
, CurPtr
, tok::unknown
);
1961 Result
.setFlag(Token::LeadingSpace
);
1964 unsigned SizeTmp
, SizeTmp2
; // Temporaries for use in cases below.
1966 // Read a character, advancing over it.
1967 char Char
= getAndAdvanceChar(CurPtr
, Result
);
1968 tok::TokenKind Kind
;
1972 // Found end of file?
1973 if (CurPtr
-1 == BufferEnd
) {
1974 // Read the PP instance variable into an automatic variable, because
1975 // LexEndOfFile will often delete 'this'.
1976 Preprocessor
*PPCache
= PP
;
1977 if (LexEndOfFile(Result
, CurPtr
-1)) // Retreat back into the file.
1978 return; // Got a token to return.
1979 assert(PPCache
&& "Raw buffer::LexEndOfFile should return a token");
1980 return PPCache
->Lex(Result
);
1983 if (!isLexingRawMode())
1984 Diag(CurPtr
-1, diag::null_in_file
);
1985 Result
.setFlag(Token::LeadingSpace
);
1986 if (SkipWhitespace(Result
, CurPtr
))
1987 return; // KeepWhitespaceMode
1989 goto LexNextToken
; // GCC isn't tail call eliminating.
1991 case 26: // DOS & CP/M EOF: "^Z".
1992 // If we're in Microsoft extensions mode, treat this as end of file.
1993 if (Features
.Microsoft
) {
1994 // Read the PP instance variable into an automatic variable, because
1995 // LexEndOfFile will often delete 'this'.
1996 Preprocessor
*PPCache
= PP
;
1997 if (LexEndOfFile(Result
, CurPtr
-1)) // Retreat back into the file.
1998 return; // Got a token to return.
1999 assert(PPCache
&& "Raw buffer::LexEndOfFile should return a token");
2000 return PPCache
->Lex(Result
);
2002 // If Microsoft extensions are disabled, this is just random garbage.
2003 Kind
= tok::unknown
;
2008 // If we are inside a preprocessor directive and we see the end of line,
2009 // we know we are done with the directive, so return an EOM token.
2010 if (ParsingPreprocessorDirective
) {
2011 // Done parsing the "line".
2012 ParsingPreprocessorDirective
= false;
2014 // Restore comment saving mode, in case it was disabled for directive.
2015 SetCommentRetentionState(PP
->getCommentRetentionState());
2017 // Since we consumed a newline, we are back at the start of a line.
2018 IsAtStartOfLine
= true;
2023 // The returned token is at the start of the line.
2024 Result
.setFlag(Token::StartOfLine
);
2025 // No leading whitespace seen so far.
2026 Result
.clearFlag(Token::LeadingSpace
);
2028 if (SkipWhitespace(Result
, CurPtr
))
2029 return; // KeepWhitespaceMode
2030 goto LexNextToken
; // GCC isn't tail call eliminating.
2035 SkipHorizontalWhitespace
:
2036 Result
.setFlag(Token::LeadingSpace
);
2037 if (SkipWhitespace(Result
, CurPtr
))
2038 return; // KeepWhitespaceMode
2043 // If the next token is obviously a // or /* */ comment, skip it efficiently
2044 // too (without going through the big switch stmt).
2045 if (CurPtr
[0] == '/' && CurPtr
[1] == '/' && !inKeepCommentMode() &&
2046 Features
.BCPLComment
) {
2047 if (SkipBCPLComment(Result
, CurPtr
+2))
2048 return; // There is a token to return.
2049 goto SkipIgnoredUnits
;
2050 } else if (CurPtr
[0] == '/' && CurPtr
[1] == '*' && !inKeepCommentMode()) {
2051 if (SkipBlockComment(Result
, CurPtr
+2))
2052 return; // There is a token to return.
2053 goto SkipIgnoredUnits
;
2054 } else if (isHorizontalWhitespace(*CurPtr
)) {
2055 goto SkipHorizontalWhitespace
;
2057 goto LexNextToken
; // GCC isn't tail call eliminating.
2059 // C99 6.4.4.1: Integer Constants.
2060 // C99 6.4.4.2: Floating Constants.
2061 case '0': case '1': case '2': case '3': case '4':
2062 case '5': case '6': case '7': case '8': case '9':
2063 // Notify MIOpt that we read a non-whitespace/non-comment token.
2065 return LexNumericConstant(Result
, CurPtr
);
2067 case 'L': // Identifier (Loony) or wide literal (L'x' or L"xyz").
2068 // Notify MIOpt that we read a non-whitespace/non-comment token.
2070 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2072 // Wide string literal.
2074 return LexStringLiteral(Result
, ConsumeChar(CurPtr
, SizeTmp
, Result
),
2077 // Wide character constant.
2079 return LexCharConstant(Result
, ConsumeChar(CurPtr
, SizeTmp
, Result
));
2080 // FALL THROUGH, treating L like the start of an identifier.
2082 // C99 6.4.2: Identifiers.
2083 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G':
2084 case 'H': case 'I': case 'J': case 'K': /*'L'*/case 'M': case 'N':
2085 case 'O': case 'P': case 'Q': case 'R': case 'S': case 'T': case 'U':
2086 case 'V': case 'W': case 'X': case 'Y': case 'Z':
2087 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g':
2088 case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n':
2089 case 'o': case 'p': case 'q': case 'r': case 's': case 't': case 'u':
2090 case 'v': case 'w': case 'x': case 'y': case 'z':
2092 // Notify MIOpt that we read a non-whitespace/non-comment token.
2094 return LexIdentifier(Result
, CurPtr
);
2096 case '$': // $ in identifiers.
2097 if (Features
.DollarIdents
) {
2098 if (!isLexingRawMode())
2099 Diag(CurPtr
-1, diag::ext_dollar_in_identifier
);
2100 // Notify MIOpt that we read a non-whitespace/non-comment token.
2102 return LexIdentifier(Result
, CurPtr
);
2105 Kind
= tok::unknown
;
2108 // C99 6.4.4: Character Constants.
2110 // Notify MIOpt that we read a non-whitespace/non-comment token.
2112 return LexCharConstant(Result
, CurPtr
);
2114 // C99 6.4.5: String Literals.
2116 // Notify MIOpt that we read a non-whitespace/non-comment token.
2118 return LexStringLiteral(Result
, CurPtr
, false);
2120 // C99 6.4.6: Punctuators.
2122 Kind
= tok::question
;
2125 Kind
= tok::l_square
;
2128 Kind
= tok::r_square
;
2131 Kind
= tok::l_paren
;
2134 Kind
= tok::r_paren
;
2137 Kind
= tok::l_brace
;
2140 Kind
= tok::r_brace
;
2143 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2144 if (Char
>= '0' && Char
<= '9') {
2145 // Notify MIOpt that we read a non-whitespace/non-comment token.
2148 return LexNumericConstant(Result
, ConsumeChar(CurPtr
, SizeTmp
, Result
));
2149 } else if (Features
.CPlusPlus
&& Char
== '*') {
2150 Kind
= tok::periodstar
;
2152 } else if (Char
== '.' &&
2153 getCharAndSize(CurPtr
+SizeTmp
, SizeTmp2
) == '.') {
2154 Kind
= tok::ellipsis
;
2155 CurPtr
= ConsumeChar(ConsumeChar(CurPtr
, SizeTmp
, Result
),
2162 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2165 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2166 } else if (Char
== '=') {
2167 Kind
= tok::ampequal
;
2168 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2174 if (getCharAndSize(CurPtr
, SizeTmp
) == '=') {
2175 Kind
= tok::starequal
;
2176 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2182 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2184 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2185 Kind
= tok::plusplus
;
2186 } else if (Char
== '=') {
2187 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2188 Kind
= tok::plusequal
;
2194 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2195 if (Char
== '-') { // --
2196 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2197 Kind
= tok::minusminus
;
2198 } else if (Char
== '>' && Features
.CPlusPlus
&&
2199 getCharAndSize(CurPtr
+SizeTmp
, SizeTmp2
) == '*') { // C++ ->*
2200 CurPtr
= ConsumeChar(ConsumeChar(CurPtr
, SizeTmp
, Result
),
2202 Kind
= tok::arrowstar
;
2203 } else if (Char
== '>') { // ->
2204 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2206 } else if (Char
== '=') { // -=
2207 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2208 Kind
= tok::minusequal
;
2217 if (getCharAndSize(CurPtr
, SizeTmp
) == '=') {
2218 Kind
= tok::exclaimequal
;
2219 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2221 Kind
= tok::exclaim
;
2226 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2227 if (Char
== '/') { // BCPL comment.
2228 // Even if BCPL comments are disabled (e.g. in C89 mode), we generally
2229 // want to lex this as a comment. There is one problem with this though,
2230 // that in one particular corner case, this can change the behavior of the
2231 // resultant program. For example, In "foo //**/ bar", C89 would lex
2232 // this as "foo / bar" and langauges with BCPL comments would lex it as
2233 // "foo". Check to see if the character after the second slash is a '*'.
2234 // If so, we will lex that as a "/" instead of the start of a comment.
2235 if (Features
.BCPLComment
||
2236 getCharAndSize(CurPtr
+SizeTmp
, SizeTmp2
) != '*') {
2237 if (SkipBCPLComment(Result
, ConsumeChar(CurPtr
, SizeTmp
, Result
)))
2238 return; // There is a token to return.
2240 // It is common for the tokens immediately after a // comment to be
2241 // whitespace (indentation for the next line). Instead of going through
2242 // the big switch, handle it efficiently now.
2243 goto SkipIgnoredUnits
;
2247 if (Char
== '*') { // /**/ comment.
2248 if (SkipBlockComment(Result
, ConsumeChar(CurPtr
, SizeTmp
, Result
)))
2249 return; // There is a token to return.
2250 goto LexNextToken
; // GCC isn't tail call eliminating.
2254 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2255 Kind
= tok::slashequal
;
2261 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2263 Kind
= tok::percentequal
;
2264 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2265 } else if (Features
.Digraphs
&& Char
== '>') {
2266 Kind
= tok::r_brace
; // '%>' -> '}'
2267 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2268 } else if (Features
.Digraphs
&& Char
== ':') {
2269 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2270 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2271 if (Char
== '%' && getCharAndSize(CurPtr
+SizeTmp
, SizeTmp2
) == ':') {
2272 Kind
= tok::hashhash
; // '%:%:' -> '##'
2273 CurPtr
= ConsumeChar(ConsumeChar(CurPtr
, SizeTmp
, Result
),
2275 } else if (Char
== '@' && Features
.Microsoft
) { // %:@ -> #@ -> Charize
2276 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2277 if (!isLexingRawMode())
2278 Diag(BufferPtr
, diag::charize_microsoft_ext
);
2280 } else { // '%:' -> '#'
2281 // We parsed a # character. If this occurs at the start of the line,
2282 // it's actually the start of a preprocessing directive. Callback to
2283 // the preprocessor to handle it.
2284 // FIXME: -fpreprocessed mode??
2285 if (Result
.isAtStartOfLine() && !LexingRawMode
&& !Is_PragmaLexer
) {
2286 FormTokenWithChars(Result
, CurPtr
, tok::hash
);
2287 PP
->HandleDirective(Result
);
2289 // As an optimization, if the preprocessor didn't switch lexers, tail
2291 if (PP
->isCurrentLexer(this)) {
2292 // Start a new token. If this is a #include or something, the PP may
2293 // want us starting at the beginning of the line again. If so, set
2294 // the StartOfLine flag and clear LeadingSpace.
2295 if (IsAtStartOfLine
) {
2296 Result
.setFlag(Token::StartOfLine
);
2297 Result
.clearFlag(Token::LeadingSpace
);
2298 IsAtStartOfLine
= false;
2300 goto LexNextToken
; // GCC isn't tail call eliminating.
2303 return PP
->Lex(Result
);
2309 Kind
= tok::percent
;
2313 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2314 if (ParsingFilename
) {
2315 return LexAngledStringLiteral(Result
, CurPtr
);
2316 } else if (Char
== '<') {
2317 char After
= getCharAndSize(CurPtr
+SizeTmp
, SizeTmp2
);
2319 Kind
= tok::lesslessequal
;
2320 CurPtr
= ConsumeChar(ConsumeChar(CurPtr
, SizeTmp
, Result
),
2322 } else if (After
== '<' && IsStartOfConflictMarker(CurPtr
-1)) {
2323 // If this is actually a '<<<<<<<' version control conflict marker,
2324 // recognize it as such and recover nicely.
2326 } else if (Features
.CUDA
&& After
== '<') {
2327 Kind
= tok::lesslessless
;
2328 CurPtr
= ConsumeChar(ConsumeChar(CurPtr
, SizeTmp
, Result
),
2331 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2332 Kind
= tok::lessless
;
2334 } else if (Char
== '=') {
2335 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2336 Kind
= tok::lessequal
;
2337 } else if (Features
.Digraphs
&& Char
== ':') { // '<:' -> '['
2338 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2339 Kind
= tok::l_square
;
2340 } else if (Features
.Digraphs
&& Char
== '%') { // '<%' -> '{'
2341 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2342 Kind
= tok::l_brace
;
2348 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2350 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2351 Kind
= tok::greaterequal
;
2352 } else if (Char
== '>') {
2353 char After
= getCharAndSize(CurPtr
+SizeTmp
, SizeTmp2
);
2355 CurPtr
= ConsumeChar(ConsumeChar(CurPtr
, SizeTmp
, Result
),
2357 Kind
= tok::greatergreaterequal
;
2358 } else if (After
== '>' && HandleEndOfConflictMarker(CurPtr
-1)) {
2359 // If this is '>>>>>>>' and we're in a conflict marker, ignore it.
2361 } else if (Features
.CUDA
&& After
== '>') {
2362 Kind
= tok::greatergreatergreater
;
2363 CurPtr
= ConsumeChar(ConsumeChar(CurPtr
, SizeTmp
, Result
),
2366 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2367 Kind
= tok::greatergreater
;
2371 Kind
= tok::greater
;
2375 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2377 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2378 Kind
= tok::caretequal
;
2384 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2386 Kind
= tok::pipeequal
;
2387 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2388 } else if (Char
== '|') {
2389 // If this is '|||||||' and we're in a conflict marker, ignore it.
2390 if (CurPtr
[1] == '|' && HandleEndOfConflictMarker(CurPtr
-1))
2392 Kind
= tok::pipepipe
;
2393 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2399 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2400 if (Features
.Digraphs
&& Char
== '>') {
2401 Kind
= tok::r_square
; // ':>' -> ']'
2402 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2403 } else if (Features
.CPlusPlus
&& Char
== ':') {
2404 Kind
= tok::coloncolon
;
2405 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2414 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2416 // If this is '=======' and we're in a conflict marker, ignore it.
2417 if (CurPtr
[1] == '=' && HandleEndOfConflictMarker(CurPtr
-1))
2420 Kind
= tok::equalequal
;
2421 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2430 Char
= getCharAndSize(CurPtr
, SizeTmp
);
2432 Kind
= tok::hashhash
;
2433 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2434 } else if (Char
== '@' && Features
.Microsoft
) { // #@ -> Charize
2436 if (!isLexingRawMode())
2437 Diag(BufferPtr
, diag::charize_microsoft_ext
);
2438 CurPtr
= ConsumeChar(CurPtr
, SizeTmp
, Result
);
2440 // We parsed a # character. If this occurs at the start of the line,
2441 // it's actually the start of a preprocessing directive. Callback to
2442 // the preprocessor to handle it.
2443 // FIXME: -fpreprocessed mode??
2444 if (Result
.isAtStartOfLine() && !LexingRawMode
&& !Is_PragmaLexer
) {
2445 FormTokenWithChars(Result
, CurPtr
, tok::hash
);
2446 PP
->HandleDirective(Result
);
2448 // As an optimization, if the preprocessor didn't switch lexers, tail
2450 if (PP
->isCurrentLexer(this)) {
2451 // Start a new token. If this is a #include or something, the PP may
2452 // want us starting at the beginning of the line again. If so, set
2453 // the StartOfLine flag and clear LeadingSpace.
2454 if (IsAtStartOfLine
) {
2455 Result
.setFlag(Token::StartOfLine
);
2456 Result
.clearFlag(Token::LeadingSpace
);
2457 IsAtStartOfLine
= false;
2459 goto LexNextToken
; // GCC isn't tail call eliminating.
2461 return PP
->Lex(Result
);
2469 // Objective C support.
2470 if (CurPtr
[-1] == '@' && Features
.ObjC1
)
2473 Kind
= tok::unknown
;
2480 Kind
= tok::unknown
;
2484 // Notify MIOpt that we read a non-whitespace/non-comment token.
2487 // Update the location of token as well as BufferPtr.
2488 FormTokenWithChars(Result
, CurPtr
, Kind
);