blob: c68d21c7e79aa10594f59155af11a213d99b55ee [file] [log] [blame]
Reid Spencer5f016e22007-07-11 17:01:13 +00001//===--- Lexer.cpp - C Language Family Lexer ------------------------------===//
2//
3// The LLVM Compiler Infrastructure
4//
Chris Lattner0bc735f2007-12-29 19:59:25 +00005// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
Reid Spencer5f016e22007-07-11 17:01:13 +00007//
8//===----------------------------------------------------------------------===//
9//
Chris Lattnerd2177732007-07-20 16:59:19 +000010// This file implements the Lexer and Token interfaces.
Reid Spencer5f016e22007-07-11 17:01:13 +000011//
12//===----------------------------------------------------------------------===//
13//
14// TODO: GCC Diagnostics emitted by the lexer:
15// PEDWARN: (form feed|vertical tab) in preprocessing directive
16//
17// Universal characters, unicode, char mapping:
18// WARNING: `%.*s' is not in NFKC
19// WARNING: `%.*s' is not in NFC
20//
21// Other:
22// TODO: Options to support:
23// -fexec-charset,-fwide-exec-charset
24//
25//===----------------------------------------------------------------------===//
26
27#include "clang/Lex/Lexer.h"
28#include "clang/Lex/Preprocessor.h"
Chris Lattner500d3292009-01-29 05:15:15 +000029#include "clang/Lex/LexDiagnostic.h"
Chris Lattner9dc1f532007-07-20 16:37:10 +000030#include "clang/Basic/SourceManager.h"
Chris Lattner409a0362007-07-22 18:38:25 +000031#include "llvm/Support/Compiler.h"
Reid Spencer5f016e22007-07-11 17:01:13 +000032#include "llvm/Support/MemoryBuffer.h"
33#include <cctype>
34using namespace clang;
35
36static void InitCharacterInfo();
37
Chris Lattnerdbf388b2007-10-07 08:47:24 +000038//===----------------------------------------------------------------------===//
39// Token Class Implementation
40//===----------------------------------------------------------------------===//
41
42/// isObjCAtKeyword - Return true if we have an ObjC keyword identifier.
43bool Token::isObjCAtKeyword(tok::ObjCKeywordKind objcKey) const {
Douglas Gregorbec1c9d2008-12-01 21:46:47 +000044 if (IdentifierInfo *II = getIdentifierInfo())
45 return II->getObjCKeywordID() == objcKey;
46 return false;
Chris Lattnerdbf388b2007-10-07 08:47:24 +000047}
48
49/// getObjCKeywordID - Return the ObjC keyword kind.
50tok::ObjCKeywordKind Token::getObjCKeywordID() const {
51 IdentifierInfo *specId = getIdentifierInfo();
52 return specId ? specId->getObjCKeywordID() : tok::objc_not_keyword;
53}
54
Chris Lattner53702cd2007-12-13 01:59:49 +000055
Chris Lattnerdbf388b2007-10-07 08:47:24 +000056//===----------------------------------------------------------------------===//
57// Lexer Class Implementation
58//===----------------------------------------------------------------------===//
59
Chris Lattner22d91ca2009-01-17 06:55:17 +000060void Lexer::InitLexer(const char *BufStart, const char *BufPtr,
61 const char *BufEnd) {
62 InitCharacterInfo();
63
64 BufferStart = BufStart;
65 BufferPtr = BufPtr;
66 BufferEnd = BufEnd;
67
68 assert(BufEnd[0] == 0 &&
69 "We assume that the input buffer has a null character at the end"
70 " to simplify lexing!");
71
72 Is_PragmaLexer = false;
73
74 // Start of the file is a start of line.
75 IsAtStartOfLine = true;
76
77 // We are not after parsing a #.
78 ParsingPreprocessorDirective = false;
79
80 // We are not after parsing #include.
81 ParsingFilename = false;
82
83 // We are not in raw mode. Raw mode disables diagnostics and interpretation
84 // of tokens (e.g. identifiers, thus disabling macro expansion). It is used
85 // to quickly lex the tokens of the buffer, e.g. when handling a "#if 0" block
86 // or otherwise skipping over tokens.
87 LexingRawMode = false;
88
89 // Default to not keeping comments.
90 ExtendedTokenMode = 0;
91}
92
Chris Lattner0770dab2009-01-17 07:56:59 +000093/// Lexer constructor - Create a new lexer object for the specified buffer
94/// with the specified preprocessor managing the lexing process. This lexer
95/// assumes that the associated file buffer and Preprocessor objects will
96/// outlive it, so it doesn't take ownership of either of them.
Chris Lattner88d3ac12009-01-17 08:03:42 +000097Lexer::Lexer(FileID FID, Preprocessor &PP)
98 : PreprocessorLexer(&PP, FID),
99 FileLoc(PP.getSourceManager().getLocForStartOfFile(FID)),
100 Features(PP.getLangOptions()) {
Chris Lattner0770dab2009-01-17 07:56:59 +0000101
Chris Lattner88d3ac12009-01-17 08:03:42 +0000102 const llvm::MemoryBuffer *InputFile = PP.getSourceManager().getBuffer(FID);
Chris Lattner0770dab2009-01-17 07:56:59 +0000103
104 InitLexer(InputFile->getBufferStart(), InputFile->getBufferStart(),
105 InputFile->getBufferEnd());
106
107 // Default to keeping comments if the preprocessor wants them.
108 SetCommentRetentionState(PP.getCommentRetentionState());
109}
Chris Lattnerdbf388b2007-10-07 08:47:24 +0000110
Chris Lattner168ae2d2007-10-17 20:41:00 +0000111/// Lexer constructor - Create a new raw lexer object. This object is only
Chris Lattner590f0cc2008-10-12 01:15:46 +0000112/// suitable for calls to 'LexRawToken'. This lexer assumes that the text
113/// range will outlive it, so it doesn't take ownership of it.
Chris Lattner168ae2d2007-10-17 20:41:00 +0000114Lexer::Lexer(SourceLocation fileloc, const LangOptions &features,
Chris Lattnerde96c0f2009-01-17 07:42:27 +0000115 const char *BufStart, const char *BufPtr, const char *BufEnd)
Chris Lattnerc6fe32a2009-01-17 03:48:08 +0000116 : FileLoc(fileloc), Features(features) {
Chris Lattner22d91ca2009-01-17 06:55:17 +0000117
Chris Lattner22d91ca2009-01-17 06:55:17 +0000118 InitLexer(BufStart, BufPtr, BufEnd);
Chris Lattner168ae2d2007-10-17 20:41:00 +0000119
120 // We *are* in raw mode.
121 LexingRawMode = true;
Chris Lattner168ae2d2007-10-17 20:41:00 +0000122}
123
Chris Lattner025c3a62009-01-17 07:35:14 +0000124/// Lexer constructor - Create a new raw lexer object. This object is only
125/// suitable for calls to 'LexRawToken'. This lexer assumes that the text
126/// range will outlive it, so it doesn't take ownership of it.
127Lexer::Lexer(FileID FID, const SourceManager &SM, const LangOptions &features)
128 : FileLoc(SM.getLocForStartOfFile(FID)), Features(features) {
129 const llvm::MemoryBuffer *FromFile = SM.getBuffer(FID);
130
131 InitLexer(FromFile->getBufferStart(), FromFile->getBufferStart(),
132 FromFile->getBufferEnd());
133
134 // We *are* in raw mode.
135 LexingRawMode = true;
136}
137
Chris Lattner42e00d12009-01-17 08:27:52 +0000138/// Create_PragmaLexer: Lexer constructor - Create a new lexer object for
139/// _Pragma expansion. This has a variety of magic semantics that this method
140/// sets up. It returns a new'd Lexer that must be delete'd when done.
141///
142/// On entrance to this routine, TokStartLoc is a macro location which has a
143/// spelling loc that indicates the bytes to be lexed for the token and an
144/// instantiation location that indicates where all lexed tokens should be
145/// "expanded from".
146///
147/// FIXME: It would really be nice to make _Pragma just be a wrapper around a
148/// normal lexer that remaps tokens as they fly by. This would require making
149/// Preprocessor::Lex virtual. Given that, we could just dump in a magic lexer
150/// interface that could handle this stuff. This would pull GetMappedTokenLoc
151/// out of the critical path of the lexer!
152///
Chris Lattnerbcc2a672009-01-19 06:46:35 +0000153Lexer *Lexer::Create_PragmaLexer(SourceLocation SpellingLoc,
Chris Lattnere7fb4842009-02-15 20:52:18 +0000154 SourceLocation InstantiationLocStart,
155 SourceLocation InstantiationLocEnd,
Chris Lattnerbcc2a672009-01-19 06:46:35 +0000156 unsigned TokLen, Preprocessor &PP) {
Chris Lattner42e00d12009-01-17 08:27:52 +0000157 SourceManager &SM = PP.getSourceManager();
Chris Lattner42e00d12009-01-17 08:27:52 +0000158
159 // Create the lexer as if we were going to lex the file normally.
Chris Lattnera11d6172009-01-19 07:46:45 +0000160 FileID SpellingFID = SM.getFileID(SpellingLoc);
Chris Lattnerbcc2a672009-01-19 06:46:35 +0000161 Lexer *L = new Lexer(SpellingFID, PP);
Chris Lattner42e00d12009-01-17 08:27:52 +0000162
163 // Now that the lexer is created, change the start/end locations so that we
164 // just lex the subsection of the file that we want. This is lexing from a
165 // scratch buffer.
166 const char *StrData = SM.getCharacterData(SpellingLoc);
167
168 L->BufferPtr = StrData;
169 L->BufferEnd = StrData+TokLen;
Chris Lattner1fa49532009-03-08 08:08:45 +0000170 assert(L->BufferEnd[0] == 0 && "Buffer is not nul terminated!");
Chris Lattner42e00d12009-01-17 08:27:52 +0000171
172 // Set the SourceLocation with the remapping information. This ensures that
173 // GetMappedTokenLoc will remap the tokens as they are lexed.
Chris Lattnerde7aeef2009-01-26 00:43:02 +0000174 L->FileLoc = SM.createInstantiationLoc(SM.getLocForStartOfFile(SpellingFID),
Chris Lattnere7fb4842009-02-15 20:52:18 +0000175 InstantiationLocStart,
176 InstantiationLocEnd, TokLen);
Chris Lattner42e00d12009-01-17 08:27:52 +0000177
178 // Ensure that the lexer thinks it is inside a directive, so that end \n will
179 // return an EOM token.
180 L->ParsingPreprocessorDirective = true;
181
182 // This lexer really is for _Pragma.
183 L->Is_PragmaLexer = true;
184 return L;
185}
186
Chris Lattner168ae2d2007-10-17 20:41:00 +0000187
Reid Spencer5f016e22007-07-11 17:01:13 +0000188/// Stringify - Convert the specified string into a C string, with surrounding
189/// ""'s, and with escaped \ and " characters.
190std::string Lexer::Stringify(const std::string &Str, bool Charify) {
191 std::string Result = Str;
192 char Quote = Charify ? '\'' : '"';
193 for (unsigned i = 0, e = Result.size(); i != e; ++i) {
194 if (Result[i] == '\\' || Result[i] == Quote) {
195 Result.insert(Result.begin()+i, '\\');
196 ++i; ++e;
197 }
198 }
199 return Result;
200}
201
Chris Lattnerd8e30832007-07-24 06:57:14 +0000202/// Stringify - Convert the specified string into a C string by escaping '\'
203/// and " characters. This does not add surrounding ""'s to the string.
204void Lexer::Stringify(llvm::SmallVectorImpl<char> &Str) {
205 for (unsigned i = 0, e = Str.size(); i != e; ++i) {
206 if (Str[i] == '\\' || Str[i] == '"') {
207 Str.insert(Str.begin()+i, '\\');
208 ++i; ++e;
209 }
210 }
211}
212
Reid Spencer5f016e22007-07-11 17:01:13 +0000213
Chris Lattner9a611942007-10-17 21:18:47 +0000214/// MeasureTokenLength - Relex the token at the specified location and return
215/// its length in bytes in the input file. If the token needs cleaning (e.g.
216/// includes a trigraph or an escaped newline) then this count includes bytes
217/// that are part of that.
218unsigned Lexer::MeasureTokenLength(SourceLocation Loc,
219 const SourceManager &SM) {
Chris Lattner9a611942007-10-17 21:18:47 +0000220 // TODO: this could be special cased for common tokens like identifiers, ')',
221 // etc to make this faster, if it mattered. Just look at StrData[0] to handle
222 // all obviously single-char tokens. This could use
223 // Lexer::isObviouslySimpleCharacter for example to handle identifiers or
224 // something.
Chris Lattnerde7aeef2009-01-26 00:43:02 +0000225
226 // If this comes from a macro expansion, we really do want the macro name, not
227 // the token this macro expanded to.
Chris Lattner363fdc22009-01-26 22:24:27 +0000228 Loc = SM.getInstantiationLoc(Loc);
229 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc);
Chris Lattner83503942009-01-17 08:30:10 +0000230 std::pair<const char *,const char *> Buffer = SM.getBufferData(LocInfo.first);
231 const char *StrData = Buffer.first+LocInfo.second;
232
Chris Lattner9a611942007-10-17 21:18:47 +0000233 // Create a langops struct and enable trigraphs. This is sufficient for
234 // measuring tokens.
235 LangOptions LangOpts;
236 LangOpts.Trigraphs = true;
237
238 // Create a lexer starting at the beginning of this token.
Chris Lattnerde96c0f2009-01-17 07:42:27 +0000239 Lexer TheLexer(Loc, LangOpts, Buffer.first, StrData, Buffer.second);
Chris Lattner9a611942007-10-17 21:18:47 +0000240 Token TheTok;
Chris Lattner590f0cc2008-10-12 01:15:46 +0000241 TheLexer.LexFromRawLexer(TheTok);
Chris Lattner9a611942007-10-17 21:18:47 +0000242 return TheTok.getLength();
243}
244
Reid Spencer5f016e22007-07-11 17:01:13 +0000245//===----------------------------------------------------------------------===//
246// Character information.
247//===----------------------------------------------------------------------===//
248
249static unsigned char CharInfo[256];
250
251enum {
252 CHAR_HORZ_WS = 0x01, // ' ', '\t', '\f', '\v'. Note, no '\0'
253 CHAR_VERT_WS = 0x02, // '\r', '\n'
254 CHAR_LETTER = 0x04, // a-z,A-Z
255 CHAR_NUMBER = 0x08, // 0-9
256 CHAR_UNDER = 0x10, // _
257 CHAR_PERIOD = 0x20 // .
258};
259
260static void InitCharacterInfo() {
261 static bool isInited = false;
262 if (isInited) return;
263 isInited = true;
264
265 // Intiialize the CharInfo table.
266 // TODO: statically initialize this.
267 CharInfo[(int)' '] = CharInfo[(int)'\t'] =
268 CharInfo[(int)'\f'] = CharInfo[(int)'\v'] = CHAR_HORZ_WS;
269 CharInfo[(int)'\n'] = CharInfo[(int)'\r'] = CHAR_VERT_WS;
270
271 CharInfo[(int)'_'] = CHAR_UNDER;
272 CharInfo[(int)'.'] = CHAR_PERIOD;
273 for (unsigned i = 'a'; i <= 'z'; ++i)
274 CharInfo[i] = CharInfo[i+'A'-'a'] = CHAR_LETTER;
275 for (unsigned i = '0'; i <= '9'; ++i)
276 CharInfo[i] = CHAR_NUMBER;
277}
278
279/// isIdentifierBody - Return true if this is the body character of an
280/// identifier, which is [a-zA-Z0-9_].
281static inline bool isIdentifierBody(unsigned char c) {
Hartmut Kaiser95c062b2007-10-18 12:47:01 +0000282 return (CharInfo[c] & (CHAR_LETTER|CHAR_NUMBER|CHAR_UNDER)) ? true : false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000283}
284
285/// isHorizontalWhitespace - Return true if this character is horizontal
286/// whitespace: ' ', '\t', '\f', '\v'. Note that this returns false for '\0'.
287static inline bool isHorizontalWhitespace(unsigned char c) {
Hartmut Kaiser95c062b2007-10-18 12:47:01 +0000288 return (CharInfo[c] & CHAR_HORZ_WS) ? true : false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000289}
290
291/// isWhitespace - Return true if this character is horizontal or vertical
292/// whitespace: ' ', '\t', '\f', '\v', '\n', '\r'. Note that this returns false
293/// for '\0'.
294static inline bool isWhitespace(unsigned char c) {
Hartmut Kaiser95c062b2007-10-18 12:47:01 +0000295 return (CharInfo[c] & (CHAR_HORZ_WS|CHAR_VERT_WS)) ? true : false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000296}
297
298/// isNumberBody - Return true if this is the body character of an
299/// preprocessing number, which is [a-zA-Z0-9_.].
300static inline bool isNumberBody(unsigned char c) {
Hartmut Kaiser95c062b2007-10-18 12:47:01 +0000301 return (CharInfo[c] & (CHAR_LETTER|CHAR_NUMBER|CHAR_UNDER|CHAR_PERIOD)) ?
302 true : false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000303}
304
305
306//===----------------------------------------------------------------------===//
307// Diagnostics forwarding code.
308//===----------------------------------------------------------------------===//
309
Chris Lattner409a0362007-07-22 18:38:25 +0000310/// GetMappedTokenLoc - If lexing out of a 'mapped buffer', where we pretend the
311/// lexer buffer was all instantiated at a single point, perform the mapping.
312/// This is currently only used for _Pragma implementation, so it is the slow
313/// path of the hot getSourceLocation method. Do not allow it to be inlined.
314static SourceLocation GetMappedTokenLoc(Preprocessor &PP,
315 SourceLocation FileLoc,
Chris Lattnerde7aeef2009-01-26 00:43:02 +0000316 unsigned CharNo,
317 unsigned TokLen) DISABLE_INLINE;
Chris Lattner409a0362007-07-22 18:38:25 +0000318static SourceLocation GetMappedTokenLoc(Preprocessor &PP,
319 SourceLocation FileLoc,
Chris Lattnerde7aeef2009-01-26 00:43:02 +0000320 unsigned CharNo, unsigned TokLen) {
Chris Lattnere7fb4842009-02-15 20:52:18 +0000321 assert(FileLoc.isMacroID() && "Must be an instantiation");
322
Chris Lattner409a0362007-07-22 18:38:25 +0000323 // Otherwise, we're lexing "mapped tokens". This is used for things like
324 // _Pragma handling. Combine the instantiation location of FileLoc with the
Chris Lattnerdf7c17a2009-01-16 07:00:02 +0000325 // spelling location.
Chris Lattnere7fb4842009-02-15 20:52:18 +0000326 SourceManager &SM = PP.getSourceManager();
Chris Lattner409a0362007-07-22 18:38:25 +0000327
Chris Lattnerf7cf85b2009-01-16 07:36:28 +0000328 // Create a new SLoc which is expanded from Instantiation(FileLoc) but whose
Chris Lattnerdf7c17a2009-01-16 07:00:02 +0000329 // characters come from spelling(FileLoc)+Offset.
Chris Lattnere7fb4842009-02-15 20:52:18 +0000330 SourceLocation SpellingLoc = SM.getSpellingLoc(FileLoc);
Chris Lattnerbcc2a672009-01-19 06:46:35 +0000331 SpellingLoc = SpellingLoc.getFileLocWithOffset(CharNo);
Chris Lattnere7fb4842009-02-15 20:52:18 +0000332
333 // Figure out the expansion loc range, which is the range covered by the
334 // original _Pragma(...) sequence.
335 std::pair<SourceLocation,SourceLocation> II =
336 SM.getImmediateInstantiationRange(FileLoc);
337
338 return SM.createInstantiationLoc(SpellingLoc, II.first, II.second, TokLen);
Chris Lattner409a0362007-07-22 18:38:25 +0000339}
340
Reid Spencer5f016e22007-07-11 17:01:13 +0000341/// getSourceLocation - Return a source location identifier for the specified
342/// offset in the current file.
Chris Lattnerde7aeef2009-01-26 00:43:02 +0000343SourceLocation Lexer::getSourceLocation(const char *Loc,
344 unsigned TokLen) const {
Chris Lattner448cec42007-07-22 18:44:36 +0000345 assert(Loc >= BufferStart && Loc <= BufferEnd &&
Reid Spencer5f016e22007-07-11 17:01:13 +0000346 "Location out of range for this buffer!");
Chris Lattner9dc1f532007-07-20 16:37:10 +0000347
348 // In the normal case, we're just lexing from a simple file buffer, return
349 // the file id from FileLoc with the offset specified.
Chris Lattner448cec42007-07-22 18:44:36 +0000350 unsigned CharNo = Loc-BufferStart;
Chris Lattner9dc1f532007-07-20 16:37:10 +0000351 if (FileLoc.isFileID())
Chris Lattnerbcc2a672009-01-19 06:46:35 +0000352 return FileLoc.getFileLocWithOffset(CharNo);
Chris Lattner9dc1f532007-07-20 16:37:10 +0000353
Chris Lattner2b2453a2009-01-17 06:22:33 +0000354 // Otherwise, this is the _Pragma lexer case, which pretends that all of the
355 // tokens are lexed from where the _Pragma was defined.
Chris Lattner168ae2d2007-10-17 20:41:00 +0000356 assert(PP && "This doesn't work on raw lexers");
Chris Lattnerde7aeef2009-01-26 00:43:02 +0000357 return GetMappedTokenLoc(*PP, FileLoc, CharNo, TokLen);
Reid Spencer5f016e22007-07-11 17:01:13 +0000358}
359
Reid Spencer5f016e22007-07-11 17:01:13 +0000360/// Diag - Forwarding function for diagnostics. This translate a source
361/// position in the current buffer into a SourceLocation object for rendering.
Chris Lattner3cbfe2c2008-11-22 00:59:29 +0000362DiagnosticBuilder Lexer::Diag(const char *Loc, unsigned DiagID) const {
Chris Lattner3692b092008-11-18 07:59:24 +0000363 return PP->Diag(getSourceLocation(Loc), DiagID);
Reid Spencer5f016e22007-07-11 17:01:13 +0000364}
Reid Spencer5f016e22007-07-11 17:01:13 +0000365
366//===----------------------------------------------------------------------===//
367// Trigraph and Escaped Newline Handling Code.
368//===----------------------------------------------------------------------===//
369
370/// GetTrigraphCharForLetter - Given a character that occurs after a ?? pair,
371/// return the decoded trigraph letter it corresponds to, or '\0' if nothing.
372static char GetTrigraphCharForLetter(char Letter) {
373 switch (Letter) {
374 default: return 0;
375 case '=': return '#';
376 case ')': return ']';
377 case '(': return '[';
378 case '!': return '|';
379 case '\'': return '^';
380 case '>': return '}';
381 case '/': return '\\';
382 case '<': return '{';
383 case '-': return '~';
384 }
385}
386
387/// DecodeTrigraphChar - If the specified character is a legal trigraph when
388/// prefixed with ??, emit a trigraph warning. If trigraphs are enabled,
389/// return the result character. Finally, emit a warning about trigraph use
390/// whether trigraphs are enabled or not.
391static char DecodeTrigraphChar(const char *CP, Lexer *L) {
392 char Res = GetTrigraphCharForLetter(*CP);
Chris Lattner3692b092008-11-18 07:59:24 +0000393 if (!Res || !L) return Res;
394
395 if (!L->getFeatures().Trigraphs) {
Chris Lattner74d15df2008-11-22 02:02:22 +0000396 if (!L->isLexingRawMode())
397 L->Diag(CP-2, diag::trigraph_ignored);
Chris Lattner3692b092008-11-18 07:59:24 +0000398 return 0;
Reid Spencer5f016e22007-07-11 17:01:13 +0000399 }
Chris Lattner3692b092008-11-18 07:59:24 +0000400
Chris Lattner74d15df2008-11-22 02:02:22 +0000401 if (!L->isLexingRawMode())
402 L->Diag(CP-2, diag::trigraph_converted) << std::string()+Res;
Reid Spencer5f016e22007-07-11 17:01:13 +0000403 return Res;
404}
405
406/// getCharAndSizeSlow - Peek a single 'character' from the specified buffer,
407/// get its size, and return it. This is tricky in several cases:
408/// 1. If currently at the start of a trigraph, we warn about the trigraph,
409/// then either return the trigraph (skipping 3 chars) or the '?',
410/// depending on whether trigraphs are enabled or not.
411/// 2. If this is an escaped newline (potentially with whitespace between
412/// the backslash and newline), implicitly skip the newline and return
413/// the char after it.
414/// 3. If this is a UCN, return it. FIXME: C++ UCN's?
415///
416/// This handles the slow/uncommon case of the getCharAndSize method. Here we
417/// know that we can accumulate into Size, and that we have already incremented
418/// Ptr by Size bytes.
419///
420/// NOTE: When this method is updated, getCharAndSizeSlowNoWarn (below) should
421/// be updated to match.
422///
423char Lexer::getCharAndSizeSlow(const char *Ptr, unsigned &Size,
Chris Lattnerd2177732007-07-20 16:59:19 +0000424 Token *Tok) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000425 // If we have a slash, look for an escaped newline.
426 if (Ptr[0] == '\\') {
427 ++Size;
428 ++Ptr;
429Slash:
430 // Common case, backslash-char where the char is not whitespace.
431 if (!isWhitespace(Ptr[0])) return '\\';
432
433 // See if we have optional whitespace characters followed by a newline.
434 {
435 unsigned SizeTmp = 0;
436 do {
437 ++SizeTmp;
438 if (Ptr[SizeTmp-1] == '\n' || Ptr[SizeTmp-1] == '\r') {
439 // Remember that this token needs to be cleaned.
Chris Lattnerd2177732007-07-20 16:59:19 +0000440 if (Tok) Tok->setFlag(Token::NeedsCleaning);
Reid Spencer5f016e22007-07-11 17:01:13 +0000441
442 // Warn if there was whitespace between the backslash and newline.
Chris Lattner74d15df2008-11-22 02:02:22 +0000443 if (SizeTmp != 1 && Tok && !isLexingRawMode())
Reid Spencer5f016e22007-07-11 17:01:13 +0000444 Diag(Ptr, diag::backslash_newline_space);
445
446 // If this is a \r\n or \n\r, skip the newlines.
447 if ((Ptr[SizeTmp] == '\r' || Ptr[SizeTmp] == '\n') &&
448 Ptr[SizeTmp-1] != Ptr[SizeTmp])
449 ++SizeTmp;
450
451 // Found backslash<whitespace><newline>. Parse the char after it.
452 Size += SizeTmp;
453 Ptr += SizeTmp;
454 // Use slow version to accumulate a correct size field.
455 return getCharAndSizeSlow(Ptr, Size, Tok);
456 }
457 } while (isWhitespace(Ptr[SizeTmp]));
458 }
459
460 // Otherwise, this is not an escaped newline, just return the slash.
461 return '\\';
462 }
463
464 // If this is a trigraph, process it.
465 if (Ptr[0] == '?' && Ptr[1] == '?') {
466 // If this is actually a legal trigraph (not something like "??x"), emit
467 // a trigraph warning. If so, and if trigraphs are enabled, return it.
468 if (char C = DecodeTrigraphChar(Ptr+2, Tok ? this : 0)) {
469 // Remember that this token needs to be cleaned.
Chris Lattnerd2177732007-07-20 16:59:19 +0000470 if (Tok) Tok->setFlag(Token::NeedsCleaning);
Reid Spencer5f016e22007-07-11 17:01:13 +0000471
472 Ptr += 3;
473 Size += 3;
474 if (C == '\\') goto Slash;
475 return C;
476 }
477 }
478
479 // If this is neither, return a single character.
480 ++Size;
481 return *Ptr;
482}
483
484
485/// getCharAndSizeSlowNoWarn - Handle the slow/uncommon case of the
486/// getCharAndSizeNoWarn method. Here we know that we can accumulate into Size,
487/// and that we have already incremented Ptr by Size bytes.
488///
489/// NOTE: When this method is updated, getCharAndSizeSlow (above) should
490/// be updated to match.
491char Lexer::getCharAndSizeSlowNoWarn(const char *Ptr, unsigned &Size,
492 const LangOptions &Features) {
493 // If we have a slash, look for an escaped newline.
494 if (Ptr[0] == '\\') {
495 ++Size;
496 ++Ptr;
497Slash:
498 // Common case, backslash-char where the char is not whitespace.
499 if (!isWhitespace(Ptr[0])) return '\\';
500
501 // See if we have optional whitespace characters followed by a newline.
502 {
503 unsigned SizeTmp = 0;
504 do {
505 ++SizeTmp;
506 if (Ptr[SizeTmp-1] == '\n' || Ptr[SizeTmp-1] == '\r') {
507
508 // If this is a \r\n or \n\r, skip the newlines.
509 if ((Ptr[SizeTmp] == '\r' || Ptr[SizeTmp] == '\n') &&
510 Ptr[SizeTmp-1] != Ptr[SizeTmp])
511 ++SizeTmp;
512
513 // Found backslash<whitespace><newline>. Parse the char after it.
514 Size += SizeTmp;
515 Ptr += SizeTmp;
516
517 // Use slow version to accumulate a correct size field.
518 return getCharAndSizeSlowNoWarn(Ptr, Size, Features);
519 }
520 } while (isWhitespace(Ptr[SizeTmp]));
521 }
522
523 // Otherwise, this is not an escaped newline, just return the slash.
524 return '\\';
525 }
526
527 // If this is a trigraph, process it.
528 if (Features.Trigraphs && Ptr[0] == '?' && Ptr[1] == '?') {
529 // If this is actually a legal trigraph (not something like "??x"), return
530 // it.
531 if (char C = GetTrigraphCharForLetter(Ptr[2])) {
532 Ptr += 3;
533 Size += 3;
534 if (C == '\\') goto Slash;
535 return C;
536 }
537 }
538
539 // If this is neither, return a single character.
540 ++Size;
541 return *Ptr;
542}
543
544//===----------------------------------------------------------------------===//
545// Helper methods for lexing.
546//===----------------------------------------------------------------------===//
547
Chris Lattnerd2177732007-07-20 16:59:19 +0000548void Lexer::LexIdentifier(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000549 // Match [_A-Za-z0-9]*, we have already matched [_A-Za-z$]
550 unsigned Size;
551 unsigned char C = *CurPtr++;
552 while (isIdentifierBody(C)) {
553 C = *CurPtr++;
554 }
555 --CurPtr; // Back up over the skipped character.
556
557 // Fast path, no $,\,? in identifier found. '\' might be an escaped newline
558 // or UCN, and ? might be a trigraph for '\', an escaped newline or UCN.
559 // FIXME: UCNs.
560 if (C != '\\' && C != '?' && (C != '$' || !Features.DollarIdents)) {
561FinishIdentifier:
562 const char *IdStart = BufferPtr;
Chris Lattner9e6293d2008-10-12 04:51:35 +0000563 FormTokenWithChars(Result, CurPtr, tok::identifier);
Reid Spencer5f016e22007-07-11 17:01:13 +0000564
565 // If we are in raw mode, return this identifier raw. There is no need to
566 // look up identifier information or attempt to macro expand it.
567 if (LexingRawMode) return;
568
569 // Fill in Result.IdentifierInfo, looking up the identifier in the
570 // identifier table.
Chris Lattnerd1186fa2009-01-21 07:45:14 +0000571 IdentifierInfo *II = PP->LookUpIdentifierInfo(Result, IdStart);
Reid Spencer5f016e22007-07-11 17:01:13 +0000572
Chris Lattner863c4862009-01-23 18:35:48 +0000573 // Change the kind of this identifier to the appropriate token kind, e.g.
574 // turning "for" into a keyword.
575 Result.setKind(II->getTokenID());
576
Reid Spencer5f016e22007-07-11 17:01:13 +0000577 // Finally, now that we know we have an identifier, pass this off to the
578 // preprocessor, which may macro expand it or something.
Chris Lattnerd1186fa2009-01-21 07:45:14 +0000579 if (II->isHandleIdentifierCase())
Chris Lattner6a170eb2009-01-21 07:43:11 +0000580 PP->HandleIdentifier(Result);
581 return;
Reid Spencer5f016e22007-07-11 17:01:13 +0000582 }
583
584 // Otherwise, $,\,? in identifier found. Enter slower path.
585
586 C = getCharAndSize(CurPtr, Size);
587 while (1) {
588 if (C == '$') {
589 // If we hit a $ and they are not supported in identifiers, we are done.
590 if (!Features.DollarIdents) goto FinishIdentifier;
591
592 // Otherwise, emit a diagnostic and continue.
Chris Lattner74d15df2008-11-22 02:02:22 +0000593 if (!isLexingRawMode())
594 Diag(CurPtr, diag::ext_dollar_in_identifier);
Reid Spencer5f016e22007-07-11 17:01:13 +0000595 CurPtr = ConsumeChar(CurPtr, Size, Result);
596 C = getCharAndSize(CurPtr, Size);
597 continue;
598 } else if (!isIdentifierBody(C)) { // FIXME: UCNs.
599 // Found end of identifier.
600 goto FinishIdentifier;
601 }
602
603 // Otherwise, this character is good, consume it.
604 CurPtr = ConsumeChar(CurPtr, Size, Result);
605
606 C = getCharAndSize(CurPtr, Size);
607 while (isIdentifierBody(C)) { // FIXME: UCNs.
608 CurPtr = ConsumeChar(CurPtr, Size, Result);
609 C = getCharAndSize(CurPtr, Size);
610 }
611 }
612}
613
614
Nate Begeman5253c7f2008-04-14 02:26:39 +0000615/// LexNumericConstant - Lex the remainder of a integer or floating point
Reid Spencer5f016e22007-07-11 17:01:13 +0000616/// constant. From[-1] is the first character lexed. Return the end of the
617/// constant.
Chris Lattnerd2177732007-07-20 16:59:19 +0000618void Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000619 unsigned Size;
620 char C = getCharAndSize(CurPtr, Size);
621 char PrevCh = 0;
622 while (isNumberBody(C)) { // FIXME: UCNs?
623 CurPtr = ConsumeChar(CurPtr, Size, Result);
624 PrevCh = C;
625 C = getCharAndSize(CurPtr, Size);
626 }
627
628 // If we fell out, check for a sign, due to 1e+12. If we have one, continue.
629 if ((C == '-' || C == '+') && (PrevCh == 'E' || PrevCh == 'e'))
630 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result));
631
632 // If we have a hex FP constant, continue.
Chris Lattner49842122008-11-22 07:39:03 +0000633 if ((C == '-' || C == '+') && (PrevCh == 'P' || PrevCh == 'p') &&
634 (Features.HexFloats || !Features.NoExtensions))
Reid Spencer5f016e22007-07-11 17:01:13 +0000635 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result));
636
Reid Spencer5f016e22007-07-11 17:01:13 +0000637 // Update the location of token as well as BufferPtr.
Chris Lattner47246be2009-01-26 19:29:26 +0000638 const char *TokStart = BufferPtr;
Chris Lattner9e6293d2008-10-12 04:51:35 +0000639 FormTokenWithChars(Result, CurPtr, tok::numeric_constant);
Chris Lattner47246be2009-01-26 19:29:26 +0000640 Result.setLiteralData(TokStart);
Reid Spencer5f016e22007-07-11 17:01:13 +0000641}
642
643/// LexStringLiteral - Lex the remainder of a string literal, after having lexed
644/// either " or L".
Chris Lattnerd88dc482008-10-12 04:05:48 +0000645void Lexer::LexStringLiteral(Token &Result, const char *CurPtr, bool Wide) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000646 const char *NulCharacter = 0; // Does this string contain the \0 character?
647
648 char C = getAndAdvanceChar(CurPtr, Result);
649 while (C != '"') {
650 // Skip escaped characters.
651 if (C == '\\') {
652 // Skip the escaped character.
653 C = getAndAdvanceChar(CurPtr, Result);
654 } else if (C == '\n' || C == '\r' || // Newline.
655 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
Chris Lattner74d15df2008-11-22 02:02:22 +0000656 if (!isLexingRawMode())
657 Diag(BufferPtr, diag::err_unterminated_string);
Chris Lattner9e6293d2008-10-12 04:51:35 +0000658 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
Reid Spencer5f016e22007-07-11 17:01:13 +0000659 return;
660 } else if (C == 0) {
661 NulCharacter = CurPtr-1;
662 }
663 C = getAndAdvanceChar(CurPtr, Result);
664 }
665
666 // If a nul character existed in the string, warn about it.
Chris Lattner74d15df2008-11-22 02:02:22 +0000667 if (NulCharacter && !isLexingRawMode())
668 Diag(NulCharacter, diag::null_in_string);
Reid Spencer5f016e22007-07-11 17:01:13 +0000669
Reid Spencer5f016e22007-07-11 17:01:13 +0000670 // Update the location of the token as well as the BufferPtr instance var.
Chris Lattner47246be2009-01-26 19:29:26 +0000671 const char *TokStart = BufferPtr;
Chris Lattner9e6293d2008-10-12 04:51:35 +0000672 FormTokenWithChars(Result, CurPtr,
673 Wide ? tok::wide_string_literal : tok::string_literal);
Chris Lattner47246be2009-01-26 19:29:26 +0000674 Result.setLiteralData(TokStart);
Reid Spencer5f016e22007-07-11 17:01:13 +0000675}
676
677/// LexAngledStringLiteral - Lex the remainder of an angled string literal,
678/// after having lexed the '<' character. This is used for #include filenames.
Chris Lattnerd2177732007-07-20 16:59:19 +0000679void Lexer::LexAngledStringLiteral(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000680 const char *NulCharacter = 0; // Does this string contain the \0 character?
681
682 char C = getAndAdvanceChar(CurPtr, Result);
683 while (C != '>') {
684 // Skip escaped characters.
685 if (C == '\\') {
686 // Skip the escaped character.
687 C = getAndAdvanceChar(CurPtr, Result);
688 } else if (C == '\n' || C == '\r' || // Newline.
689 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
Chris Lattner74d15df2008-11-22 02:02:22 +0000690 if (!isLexingRawMode())
Chris Lattnerb66158c2009-02-19 18:29:56 +0000691 Diag(BufferPtr, diag::err_unterminated_angled_string);
Chris Lattner9e6293d2008-10-12 04:51:35 +0000692 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
Reid Spencer5f016e22007-07-11 17:01:13 +0000693 return;
694 } else if (C == 0) {
695 NulCharacter = CurPtr-1;
696 }
697 C = getAndAdvanceChar(CurPtr, Result);
698 }
699
700 // If a nul character existed in the string, warn about it.
Chris Lattner74d15df2008-11-22 02:02:22 +0000701 if (NulCharacter && !isLexingRawMode())
702 Diag(NulCharacter, diag::null_in_string);
Reid Spencer5f016e22007-07-11 17:01:13 +0000703
Reid Spencer5f016e22007-07-11 17:01:13 +0000704 // Update the location of token as well as BufferPtr.
Chris Lattner47246be2009-01-26 19:29:26 +0000705 const char *TokStart = BufferPtr;
Chris Lattner9e6293d2008-10-12 04:51:35 +0000706 FormTokenWithChars(Result, CurPtr, tok::angle_string_literal);
Chris Lattner47246be2009-01-26 19:29:26 +0000707 Result.setLiteralData(TokStart);
Reid Spencer5f016e22007-07-11 17:01:13 +0000708}
709
710
711/// LexCharConstant - Lex the remainder of a character constant, after having
712/// lexed either ' or L'.
Chris Lattnerd2177732007-07-20 16:59:19 +0000713void Lexer::LexCharConstant(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000714 const char *NulCharacter = 0; // Does this character contain the \0 character?
715
716 // Handle the common case of 'x' and '\y' efficiently.
717 char C = getAndAdvanceChar(CurPtr, Result);
718 if (C == '\'') {
Chris Lattner74d15df2008-11-22 02:02:22 +0000719 if (!isLexingRawMode())
720 Diag(BufferPtr, diag::err_empty_character);
Chris Lattner9e6293d2008-10-12 04:51:35 +0000721 FormTokenWithChars(Result, CurPtr, tok::unknown);
Reid Spencer5f016e22007-07-11 17:01:13 +0000722 return;
723 } else if (C == '\\') {
724 // Skip the escaped character.
725 // FIXME: UCN's.
726 C = getAndAdvanceChar(CurPtr, Result);
727 }
728
729 if (C && C != '\n' && C != '\r' && CurPtr[0] == '\'') {
730 ++CurPtr;
731 } else {
732 // Fall back on generic code for embedded nulls, newlines, wide chars.
733 do {
734 // Skip escaped characters.
735 if (C == '\\') {
736 // Skip the escaped character.
737 C = getAndAdvanceChar(CurPtr, Result);
738 } else if (C == '\n' || C == '\r' || // Newline.
739 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file.
Chris Lattner74d15df2008-11-22 02:02:22 +0000740 if (!isLexingRawMode())
741 Diag(BufferPtr, diag::err_unterminated_char);
Chris Lattner9e6293d2008-10-12 04:51:35 +0000742 FormTokenWithChars(Result, CurPtr-1, tok::unknown);
Reid Spencer5f016e22007-07-11 17:01:13 +0000743 return;
744 } else if (C == 0) {
745 NulCharacter = CurPtr-1;
746 }
747 C = getAndAdvanceChar(CurPtr, Result);
748 } while (C != '\'');
749 }
750
Chris Lattner74d15df2008-11-22 02:02:22 +0000751 if (NulCharacter && !isLexingRawMode())
752 Diag(NulCharacter, diag::null_in_char);
Reid Spencer5f016e22007-07-11 17:01:13 +0000753
Reid Spencer5f016e22007-07-11 17:01:13 +0000754 // Update the location of token as well as BufferPtr.
Chris Lattner47246be2009-01-26 19:29:26 +0000755 const char *TokStart = BufferPtr;
Chris Lattner9e6293d2008-10-12 04:51:35 +0000756 FormTokenWithChars(Result, CurPtr, tok::char_constant);
Chris Lattner47246be2009-01-26 19:29:26 +0000757 Result.setLiteralData(TokStart);
Reid Spencer5f016e22007-07-11 17:01:13 +0000758}
759
760/// SkipWhitespace - Efficiently skip over a series of whitespace characters.
761/// Update BufferPtr to point to the next non-whitespace character and return.
Chris Lattnerd88dc482008-10-12 04:05:48 +0000762///
763/// This method forms a token and returns true if KeepWhitespaceMode is enabled.
764///
765bool Lexer::SkipWhitespace(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000766 // Whitespace - Skip it, then return the token after the whitespace.
767 unsigned char Char = *CurPtr; // Skip consequtive spaces efficiently.
768 while (1) {
769 // Skip horizontal whitespace very aggressively.
770 while (isHorizontalWhitespace(Char))
771 Char = *++CurPtr;
772
Daniel Dunbarddd3e8b2008-11-25 00:20:22 +0000773 // Otherwise if we have something other than whitespace, we're done.
Reid Spencer5f016e22007-07-11 17:01:13 +0000774 if (Char != '\n' && Char != '\r')
775 break;
776
777 if (ParsingPreprocessorDirective) {
778 // End of preprocessor directive line, let LexTokenInternal handle this.
779 BufferPtr = CurPtr;
Chris Lattnerd88dc482008-10-12 04:05:48 +0000780 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000781 }
782
783 // ok, but handle newline.
784 // The returned token is at the start of the line.
Chris Lattnerd2177732007-07-20 16:59:19 +0000785 Result.setFlag(Token::StartOfLine);
Reid Spencer5f016e22007-07-11 17:01:13 +0000786 // No leading whitespace seen so far.
Chris Lattnerd2177732007-07-20 16:59:19 +0000787 Result.clearFlag(Token::LeadingSpace);
Reid Spencer5f016e22007-07-11 17:01:13 +0000788 Char = *++CurPtr;
789 }
790
791 // If this isn't immediately after a newline, there is leading space.
792 char PrevChar = CurPtr[-1];
793 if (PrevChar != '\n' && PrevChar != '\r')
Chris Lattnerd2177732007-07-20 16:59:19 +0000794 Result.setFlag(Token::LeadingSpace);
Reid Spencer5f016e22007-07-11 17:01:13 +0000795
Chris Lattnerd88dc482008-10-12 04:05:48 +0000796 // If the client wants us to return whitespace, return it now.
797 if (isKeepWhitespaceMode()) {
Chris Lattner9e6293d2008-10-12 04:51:35 +0000798 FormTokenWithChars(Result, CurPtr, tok::unknown);
Chris Lattnerd88dc482008-10-12 04:05:48 +0000799 return true;
800 }
801
Reid Spencer5f016e22007-07-11 17:01:13 +0000802 BufferPtr = CurPtr;
Chris Lattnerd88dc482008-10-12 04:05:48 +0000803 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000804}
805
806// SkipBCPLComment - We have just read the // characters from input. Skip until
807// we find the newline character thats terminate the comment. Then update
Chris Lattner2d381892008-10-12 04:15:42 +0000808/// BufferPtr and return. If we're in KeepCommentMode, this will form the token
809/// and return true.
Chris Lattnerd2177732007-07-20 16:59:19 +0000810bool Lexer::SkipBCPLComment(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000811 // If BCPL comments aren't explicitly enabled for this language, emit an
812 // extension warning.
Chris Lattner74d15df2008-11-22 02:02:22 +0000813 if (!Features.BCPLComment && !isLexingRawMode()) {
Reid Spencer5f016e22007-07-11 17:01:13 +0000814 Diag(BufferPtr, diag::ext_bcpl_comment);
815
816 // Mark them enabled so we only emit one warning for this translation
817 // unit.
818 Features.BCPLComment = true;
819 }
820
821 // Scan over the body of the comment. The common case, when scanning, is that
822 // the comment contains normal ascii characters with nothing interesting in
823 // them. As such, optimize for this case with the inner loop.
824 char C;
825 do {
826 C = *CurPtr;
827 // FIXME: Speedup BCPL comment lexing. Just scan for a \n or \r character.
828 // If we find a \n character, scan backwards, checking to see if it's an
829 // escaped newline, like we do for block comments.
830
831 // Skip over characters in the fast loop.
832 while (C != 0 && // Potentially EOF.
833 C != '\\' && // Potentially escaped newline.
834 C != '?' && // Potentially trigraph.
835 C != '\n' && C != '\r') // Newline or DOS-style newline.
836 C = *++CurPtr;
837
838 // If this is a newline, we're done.
839 if (C == '\n' || C == '\r')
840 break; // Found the newline? Break out!
841
842 // Otherwise, this is a hard case. Fall back on getAndAdvanceChar to
Chris Lattnerbc3e9842008-12-12 07:34:39 +0000843 // properly decode the character. Read it in raw mode to avoid emitting
844 // diagnostics about things like trigraphs. If we see an escaped newline,
845 // we'll handle it below.
Reid Spencer5f016e22007-07-11 17:01:13 +0000846 const char *OldPtr = CurPtr;
Chris Lattnerbc3e9842008-12-12 07:34:39 +0000847 bool OldRawMode = isLexingRawMode();
848 LexingRawMode = true;
Reid Spencer5f016e22007-07-11 17:01:13 +0000849 C = getAndAdvanceChar(CurPtr, Result);
Chris Lattnerbc3e9842008-12-12 07:34:39 +0000850 LexingRawMode = OldRawMode;
Reid Spencer5f016e22007-07-11 17:01:13 +0000851
852 // If we read multiple characters, and one of those characters was a \r or
853 // \n, then we had an escaped newline within the comment. Emit diagnostic
854 // unless the next line is also a // comment.
855 if (CurPtr != OldPtr+1 && C != '/' && CurPtr[0] != '/') {
856 for (; OldPtr != CurPtr; ++OldPtr)
857 if (OldPtr[0] == '\n' || OldPtr[0] == '\r') {
858 // Okay, we found a // comment that ends in a newline, if the next
859 // line is also a // comment, but has spaces, don't emit a diagnostic.
860 if (isspace(C)) {
861 const char *ForwardPtr = CurPtr;
862 while (isspace(*ForwardPtr)) // Skip whitespace.
863 ++ForwardPtr;
864 if (ForwardPtr[0] == '/' && ForwardPtr[1] == '/')
865 break;
866 }
867
Chris Lattner74d15df2008-11-22 02:02:22 +0000868 if (!isLexingRawMode())
869 Diag(OldPtr-1, diag::ext_multi_line_bcpl_comment);
Reid Spencer5f016e22007-07-11 17:01:13 +0000870 break;
871 }
872 }
873
874 if (CurPtr == BufferEnd+1) { --CurPtr; break; }
875 } while (C != '\n' && C != '\r');
876
877 // Found but did not consume the newline.
878
879 // If we are returning comments as tokens, return this comment as a token.
Chris Lattnerfa95a012008-10-12 03:22:02 +0000880 if (inKeepCommentMode())
Reid Spencer5f016e22007-07-11 17:01:13 +0000881 return SaveBCPLComment(Result, CurPtr);
882
883 // If we are inside a preprocessor directive and we see the end of line,
884 // return immediately, so that the lexer can return this as an EOM token.
885 if (ParsingPreprocessorDirective || CurPtr == BufferEnd) {
886 BufferPtr = CurPtr;
Chris Lattner2d381892008-10-12 04:15:42 +0000887 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000888 }
889
890 // Otherwise, eat the \n character. We don't care if this is a \n\r or
Chris Lattner7a4f0042008-10-12 00:23:07 +0000891 // \r\n sequence. This is an efficiency hack (because we know the \n can't
Chris Lattnerd88dc482008-10-12 04:05:48 +0000892 // contribute to another token), it isn't needed for correctness. Note that
893 // this is ok even in KeepWhitespaceMode, because we would have returned the
894 /// comment above in that mode.
Reid Spencer5f016e22007-07-11 17:01:13 +0000895 ++CurPtr;
896
897 // The next returned token is at the start of the line.
Chris Lattnerd2177732007-07-20 16:59:19 +0000898 Result.setFlag(Token::StartOfLine);
Reid Spencer5f016e22007-07-11 17:01:13 +0000899 // No leading whitespace seen so far.
Chris Lattnerd2177732007-07-20 16:59:19 +0000900 Result.clearFlag(Token::LeadingSpace);
Reid Spencer5f016e22007-07-11 17:01:13 +0000901 BufferPtr = CurPtr;
Chris Lattner2d381892008-10-12 04:15:42 +0000902 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +0000903}
904
905/// SaveBCPLComment - If in save-comment mode, package up this BCPL comment in
906/// an appropriate way and return it.
Chris Lattnerd2177732007-07-20 16:59:19 +0000907bool Lexer::SaveBCPLComment(Token &Result, const char *CurPtr) {
Chris Lattner9e6293d2008-10-12 04:51:35 +0000908 // If we're not in a preprocessor directive, just return the // comment
909 // directly.
910 FormTokenWithChars(Result, CurPtr, tok::comment);
Reid Spencer5f016e22007-07-11 17:01:13 +0000911
Chris Lattner9e6293d2008-10-12 04:51:35 +0000912 if (!ParsingPreprocessorDirective)
913 return true;
914
915 // If this BCPL-style comment is in a macro definition, transmogrify it into
916 // a C-style block comment.
917 std::string Spelling = PP->getSpelling(Result);
918 assert(Spelling[0] == '/' && Spelling[1] == '/' && "Not bcpl comment?");
919 Spelling[1] = '*'; // Change prefix to "/*".
920 Spelling += "*/"; // add suffix.
921
922 Result.setKind(tok::comment);
Chris Lattner47246be2009-01-26 19:29:26 +0000923 PP->CreateString(&Spelling[0], Spelling.size(), Result,
924 Result.getLocation());
Chris Lattner2d381892008-10-12 04:15:42 +0000925 return true;
Reid Spencer5f016e22007-07-11 17:01:13 +0000926}
927
928/// isBlockCommentEndOfEscapedNewLine - Return true if the specified newline
929/// character (either \n or \r) is part of an escaped newline sequence. Issue a
Chris Lattner47a2b402008-12-12 07:14:34 +0000930/// diagnostic if so. We know that the newline is inside of a block comment.
Reid Spencer5f016e22007-07-11 17:01:13 +0000931static bool isEndOfBlockCommentWithEscapedNewLine(const char *CurPtr,
932 Lexer *L) {
933 assert(CurPtr[0] == '\n' || CurPtr[0] == '\r');
934
935 // Back up off the newline.
936 --CurPtr;
937
938 // If this is a two-character newline sequence, skip the other character.
939 if (CurPtr[0] == '\n' || CurPtr[0] == '\r') {
940 // \n\n or \r\r -> not escaped newline.
941 if (CurPtr[0] == CurPtr[1])
942 return false;
943 // \n\r or \r\n -> skip the newline.
944 --CurPtr;
945 }
946
947 // If we have horizontal whitespace, skip over it. We allow whitespace
948 // between the slash and newline.
949 bool HasSpace = false;
950 while (isHorizontalWhitespace(*CurPtr) || *CurPtr == 0) {
951 --CurPtr;
952 HasSpace = true;
953 }
954
955 // If we have a slash, we know this is an escaped newline.
956 if (*CurPtr == '\\') {
957 if (CurPtr[-1] != '*') return false;
958 } else {
959 // It isn't a slash, is it the ?? / trigraph?
960 if (CurPtr[0] != '/' || CurPtr[-1] != '?' || CurPtr[-2] != '?' ||
961 CurPtr[-3] != '*')
962 return false;
963
964 // This is the trigraph ending the comment. Emit a stern warning!
965 CurPtr -= 2;
966
967 // If no trigraphs are enabled, warn that we ignored this trigraph and
968 // ignore this * character.
969 if (!L->getFeatures().Trigraphs) {
Chris Lattner74d15df2008-11-22 02:02:22 +0000970 if (!L->isLexingRawMode())
971 L->Diag(CurPtr, diag::trigraph_ignored_block_comment);
Reid Spencer5f016e22007-07-11 17:01:13 +0000972 return false;
973 }
Chris Lattner74d15df2008-11-22 02:02:22 +0000974 if (!L->isLexingRawMode())
975 L->Diag(CurPtr, diag::trigraph_ends_block_comment);
Reid Spencer5f016e22007-07-11 17:01:13 +0000976 }
977
978 // Warn about having an escaped newline between the */ characters.
Chris Lattner74d15df2008-11-22 02:02:22 +0000979 if (!L->isLexingRawMode())
980 L->Diag(CurPtr, diag::escaped_newline_block_comment_end);
Reid Spencer5f016e22007-07-11 17:01:13 +0000981
982 // If there was space between the backslash and newline, warn about it.
Chris Lattner74d15df2008-11-22 02:02:22 +0000983 if (HasSpace && !L->isLexingRawMode())
984 L->Diag(CurPtr, diag::backslash_newline_space);
Reid Spencer5f016e22007-07-11 17:01:13 +0000985
986 return true;
987}
988
989#ifdef __SSE2__
990#include <emmintrin.h>
991#elif __ALTIVEC__
992#include <altivec.h>
993#undef bool
994#endif
995
996/// SkipBlockComment - We have just read the /* characters from input. Read
997/// until we find the */ characters that terminate the comment. Note that we
998/// don't bother decoding trigraphs or escaped newlines in block comments,
999/// because they cannot cause the comment to end. The only thing that can
1000/// happen is the comment could end with an escaped newline between the */ end
1001/// of comment.
Chris Lattner2d381892008-10-12 04:15:42 +00001002///
1003/// If KeepCommentMode is enabled, this forms a token from the comment and
1004/// returns true.
Chris Lattnerd2177732007-07-20 16:59:19 +00001005bool Lexer::SkipBlockComment(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +00001006 // Scan one character past where we should, looking for a '/' character. Once
1007 // we find it, check to see if it was preceeded by a *. This common
1008 // optimization helps people who like to put a lot of * characters in their
1009 // comments.
Chris Lattner8146b682007-07-21 23:43:37 +00001010
1011 // The first character we get with newlines and trigraphs skipped to handle
1012 // the degenerate /*/ case below correctly if the * has an escaped newline
1013 // after it.
1014 unsigned CharSize;
1015 unsigned char C = getCharAndSize(CurPtr, CharSize);
1016 CurPtr += CharSize;
Reid Spencer5f016e22007-07-11 17:01:13 +00001017 if (C == 0 && CurPtr == BufferEnd+1) {
Chris Lattner74d15df2008-11-22 02:02:22 +00001018 if (!isLexingRawMode())
Chris Lattner0af57422008-10-12 01:31:51 +00001019 Diag(BufferPtr, diag::err_unterminated_block_comment);
Chris Lattner31f0eca2008-10-12 04:19:49 +00001020 --CurPtr;
1021
1022 // KeepWhitespaceMode should return this broken comment as a token. Since
1023 // it isn't a well formed comment, just return it as an 'unknown' token.
1024 if (isKeepWhitespaceMode()) {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001025 FormTokenWithChars(Result, CurPtr, tok::unknown);
Chris Lattner31f0eca2008-10-12 04:19:49 +00001026 return true;
1027 }
1028
1029 BufferPtr = CurPtr;
Chris Lattner2d381892008-10-12 04:15:42 +00001030 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +00001031 }
1032
Chris Lattner8146b682007-07-21 23:43:37 +00001033 // Check to see if the first character after the '/*' is another /. If so,
1034 // then this slash does not end the block comment, it is part of it.
1035 if (C == '/')
1036 C = *CurPtr++;
1037
Reid Spencer5f016e22007-07-11 17:01:13 +00001038 while (1) {
1039 // Skip over all non-interesting characters until we find end of buffer or a
1040 // (probably ending) '/' character.
1041 if (CurPtr + 24 < BufferEnd) {
1042 // While not aligned to a 16-byte boundary.
1043 while (C != '/' && ((intptr_t)CurPtr & 0x0F) != 0)
1044 C = *CurPtr++;
1045
1046 if (C == '/') goto FoundSlash;
1047
1048#ifdef __SSE2__
1049 __m128i Slashes = _mm_set_epi8('/', '/', '/', '/', '/', '/', '/', '/',
1050 '/', '/', '/', '/', '/', '/', '/', '/');
1051 while (CurPtr+16 <= BufferEnd &&
1052 _mm_movemask_epi8(_mm_cmpeq_epi8(*(__m128i*)CurPtr, Slashes)) == 0)
1053 CurPtr += 16;
1054#elif __ALTIVEC__
1055 __vector unsigned char Slashes = {
1056 '/', '/', '/', '/', '/', '/', '/', '/',
1057 '/', '/', '/', '/', '/', '/', '/', '/'
1058 };
1059 while (CurPtr+16 <= BufferEnd &&
1060 !vec_any_eq(*(vector unsigned char*)CurPtr, Slashes))
1061 CurPtr += 16;
1062#else
1063 // Scan for '/' quickly. Many block comments are very large.
1064 while (CurPtr[0] != '/' &&
1065 CurPtr[1] != '/' &&
1066 CurPtr[2] != '/' &&
1067 CurPtr[3] != '/' &&
1068 CurPtr+4 < BufferEnd) {
1069 CurPtr += 4;
1070 }
1071#endif
1072
1073 // It has to be one of the bytes scanned, increment to it and read one.
1074 C = *CurPtr++;
1075 }
1076
1077 // Loop to scan the remainder.
1078 while (C != '/' && C != '\0')
1079 C = *CurPtr++;
1080
1081 FoundSlash:
1082 if (C == '/') {
1083 if (CurPtr[-2] == '*') // We found the final */. We're done!
1084 break;
1085
1086 if ((CurPtr[-2] == '\n' || CurPtr[-2] == '\r')) {
1087 if (isEndOfBlockCommentWithEscapedNewLine(CurPtr-2, this)) {
1088 // We found the final */, though it had an escaped newline between the
1089 // * and /. We're done!
1090 break;
1091 }
1092 }
1093 if (CurPtr[0] == '*' && CurPtr[1] != '/') {
1094 // If this is a /* inside of the comment, emit a warning. Don't do this
1095 // if this is a /*/, which will end the comment. This misses cases with
1096 // embedded escaped newlines, but oh well.
Chris Lattner74d15df2008-11-22 02:02:22 +00001097 if (!isLexingRawMode())
1098 Diag(CurPtr-1, diag::warn_nested_block_comment);
Reid Spencer5f016e22007-07-11 17:01:13 +00001099 }
1100 } else if (C == 0 && CurPtr == BufferEnd+1) {
Chris Lattner74d15df2008-11-22 02:02:22 +00001101 if (!isLexingRawMode())
1102 Diag(BufferPtr, diag::err_unterminated_block_comment);
Reid Spencer5f016e22007-07-11 17:01:13 +00001103 // Note: the user probably forgot a */. We could continue immediately
1104 // after the /*, but this would involve lexing a lot of what really is the
1105 // comment, which surely would confuse the parser.
Chris Lattner31f0eca2008-10-12 04:19:49 +00001106 --CurPtr;
1107
1108 // KeepWhitespaceMode should return this broken comment as a token. Since
1109 // it isn't a well formed comment, just return it as an 'unknown' token.
1110 if (isKeepWhitespaceMode()) {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001111 FormTokenWithChars(Result, CurPtr, tok::unknown);
Chris Lattner31f0eca2008-10-12 04:19:49 +00001112 return true;
1113 }
1114
1115 BufferPtr = CurPtr;
Chris Lattner2d381892008-10-12 04:15:42 +00001116 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +00001117 }
1118 C = *CurPtr++;
1119 }
1120
1121 // If we are returning comments as tokens, return this comment as a token.
Chris Lattnerfa95a012008-10-12 03:22:02 +00001122 if (inKeepCommentMode()) {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001123 FormTokenWithChars(Result, CurPtr, tok::comment);
Chris Lattner2d381892008-10-12 04:15:42 +00001124 return true;
Reid Spencer5f016e22007-07-11 17:01:13 +00001125 }
1126
1127 // It is common for the tokens immediately after a /**/ comment to be
1128 // whitespace. Instead of going through the big switch, handle it
Chris Lattnerd88dc482008-10-12 04:05:48 +00001129 // efficiently now. This is safe even in KeepWhitespaceMode because we would
1130 // have already returned above with the comment as a token.
Reid Spencer5f016e22007-07-11 17:01:13 +00001131 if (isHorizontalWhitespace(*CurPtr)) {
Chris Lattnerd2177732007-07-20 16:59:19 +00001132 Result.setFlag(Token::LeadingSpace);
Reid Spencer5f016e22007-07-11 17:01:13 +00001133 SkipWhitespace(Result, CurPtr+1);
Chris Lattner2d381892008-10-12 04:15:42 +00001134 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +00001135 }
1136
1137 // Otherwise, just return so that the next character will be lexed as a token.
1138 BufferPtr = CurPtr;
Chris Lattnerd2177732007-07-20 16:59:19 +00001139 Result.setFlag(Token::LeadingSpace);
Chris Lattner2d381892008-10-12 04:15:42 +00001140 return false;
Reid Spencer5f016e22007-07-11 17:01:13 +00001141}
1142
1143//===----------------------------------------------------------------------===//
1144// Primary Lexing Entry Points
1145//===----------------------------------------------------------------------===//
1146
Reid Spencer5f016e22007-07-11 17:01:13 +00001147/// ReadToEndOfLine - Read the rest of the current preprocessor line as an
1148/// uninterpreted string. This switches the lexer out of directive mode.
1149std::string Lexer::ReadToEndOfLine() {
1150 assert(ParsingPreprocessorDirective && ParsingFilename == false &&
1151 "Must be in a preprocessing directive!");
1152 std::string Result;
Chris Lattnerd2177732007-07-20 16:59:19 +00001153 Token Tmp;
Reid Spencer5f016e22007-07-11 17:01:13 +00001154
1155 // CurPtr - Cache BufferPtr in an automatic variable.
1156 const char *CurPtr = BufferPtr;
1157 while (1) {
1158 char Char = getAndAdvanceChar(CurPtr, Tmp);
1159 switch (Char) {
1160 default:
1161 Result += Char;
1162 break;
1163 case 0: // Null.
1164 // Found end of file?
1165 if (CurPtr-1 != BufferEnd) {
1166 // Nope, normal character, continue.
1167 Result += Char;
1168 break;
1169 }
1170 // FALL THROUGH.
1171 case '\r':
1172 case '\n':
1173 // Okay, we found the end of the line. First, back up past the \0, \r, \n.
1174 assert(CurPtr[-1] == Char && "Trigraphs for newline?");
1175 BufferPtr = CurPtr-1;
1176
1177 // Next, lex the character, which should handle the EOM transition.
1178 Lex(Tmp);
Chris Lattner22f6bbc2007-10-09 18:02:16 +00001179 assert(Tmp.is(tok::eom) && "Unexpected token!");
Reid Spencer5f016e22007-07-11 17:01:13 +00001180
1181 // Finally, we're done, return the string we found.
1182 return Result;
1183 }
1184 }
1185}
1186
1187/// LexEndOfFile - CurPtr points to the end of this file. Handle this
1188/// condition, reporting diagnostics and handling other edge cases as required.
1189/// This returns true if Result contains a token, false if PP.Lex should be
1190/// called again.
Chris Lattnerd2177732007-07-20 16:59:19 +00001191bool Lexer::LexEndOfFile(Token &Result, const char *CurPtr) {
Reid Spencer5f016e22007-07-11 17:01:13 +00001192 // If we hit the end of the file while parsing a preprocessor directive,
1193 // end the preprocessor directive first. The next token returned will
1194 // then be the end of file.
1195 if (ParsingPreprocessorDirective) {
1196 // Done parsing the "line".
1197 ParsingPreprocessorDirective = false;
Reid Spencer5f016e22007-07-11 17:01:13 +00001198 // Update the location of token as well as BufferPtr.
Chris Lattner9e6293d2008-10-12 04:51:35 +00001199 FormTokenWithChars(Result, CurPtr, tok::eom);
Reid Spencer5f016e22007-07-11 17:01:13 +00001200
1201 // Restore comment saving mode, in case it was disabled for directive.
Chris Lattnerf744d132008-10-12 03:27:19 +00001202 SetCommentRetentionState(PP->getCommentRetentionState());
Reid Spencer5f016e22007-07-11 17:01:13 +00001203 return true; // Have a token.
1204 }
1205
1206 // If we are in raw mode, return this event as an EOF token. Let the caller
1207 // that put us in raw mode handle the event.
Chris Lattner74d15df2008-11-22 02:02:22 +00001208 if (isLexingRawMode()) {
Reid Spencer5f016e22007-07-11 17:01:13 +00001209 Result.startToken();
1210 BufferPtr = BufferEnd;
Chris Lattner9e6293d2008-10-12 04:51:35 +00001211 FormTokenWithChars(Result, BufferEnd, tok::eof);
Reid Spencer5f016e22007-07-11 17:01:13 +00001212 return true;
1213 }
1214
1215 // Otherwise, issue diagnostics for unterminated #if and missing newline.
1216
1217 // If we are in a #if directive, emit an error.
1218 while (!ConditionalStack.empty()) {
Chris Lattner30c64762008-11-22 06:22:39 +00001219 PP->Diag(ConditionalStack.back().IfLoc,
1220 diag::err_pp_unterminated_conditional);
Reid Spencer5f016e22007-07-11 17:01:13 +00001221 ConditionalStack.pop_back();
1222 }
1223
Chris Lattnerb25e5d72008-04-12 05:54:25 +00001224 // C99 5.1.1.2p2: If the file is non-empty and didn't end in a newline, issue
1225 // a pedwarn.
1226 if (CurPtr != BufferStart && (CurPtr[-1] != '\n' && CurPtr[-1] != '\r'))
Reid Spencer5f016e22007-07-11 17:01:13 +00001227 Diag(BufferEnd, diag::ext_no_newline_eof);
1228
1229 BufferPtr = CurPtr;
1230
1231 // Finally, let the preprocessor handle this.
Chris Lattner168ae2d2007-10-17 20:41:00 +00001232 return PP->HandleEndOfFile(Result);
Reid Spencer5f016e22007-07-11 17:01:13 +00001233}
1234
1235/// isNextPPTokenLParen - Return 1 if the next unexpanded token lexed from
1236/// the specified lexer will return a tok::l_paren token, 0 if it is something
1237/// else and 2 if there are no more tokens in the buffer controlled by the
1238/// lexer.
1239unsigned Lexer::isNextPPTokenLParen() {
1240 assert(!LexingRawMode && "How can we expand a macro from a skipping buffer?");
1241
1242 // Switch to 'skipping' mode. This will ensure that we can lex a token
1243 // without emitting diagnostics, disables macro expansion, and will cause EOF
1244 // to return an EOF token instead of popping the include stack.
1245 LexingRawMode = true;
1246
1247 // Save state that can be changed while lexing so that we can restore it.
1248 const char *TmpBufferPtr = BufferPtr;
1249
Chris Lattnerd2177732007-07-20 16:59:19 +00001250 Token Tok;
Reid Spencer5f016e22007-07-11 17:01:13 +00001251 Tok.startToken();
1252 LexTokenInternal(Tok);
1253
1254 // Restore state that may have changed.
1255 BufferPtr = TmpBufferPtr;
1256
1257 // Restore the lexer back to non-skipping mode.
1258 LexingRawMode = false;
1259
Chris Lattner22f6bbc2007-10-09 18:02:16 +00001260 if (Tok.is(tok::eof))
Reid Spencer5f016e22007-07-11 17:01:13 +00001261 return 2;
Chris Lattner22f6bbc2007-10-09 18:02:16 +00001262 return Tok.is(tok::l_paren);
Reid Spencer5f016e22007-07-11 17:01:13 +00001263}
1264
1265
1266/// LexTokenInternal - This implements a simple C family lexer. It is an
1267/// extremely performance critical piece of code. This assumes that the buffer
1268/// has a null character at the end of the file. Return true if an error
1269/// occurred and compilation should terminate, false if normal. This returns a
1270/// preprocessing token, not a normal token, as such, it is an internal
1271/// interface. It assumes that the Flags of result have been cleared before
1272/// calling this.
Chris Lattnerd2177732007-07-20 16:59:19 +00001273void Lexer::LexTokenInternal(Token &Result) {
Reid Spencer5f016e22007-07-11 17:01:13 +00001274LexNextToken:
1275 // New token, can't need cleaning yet.
Chris Lattnerd2177732007-07-20 16:59:19 +00001276 Result.clearFlag(Token::NeedsCleaning);
Reid Spencer5f016e22007-07-11 17:01:13 +00001277 Result.setIdentifierInfo(0);
1278
1279 // CurPtr - Cache BufferPtr in an automatic variable.
1280 const char *CurPtr = BufferPtr;
1281
1282 // Small amounts of horizontal whitespace is very common between tokens.
1283 if ((*CurPtr == ' ') || (*CurPtr == '\t')) {
1284 ++CurPtr;
1285 while ((*CurPtr == ' ') || (*CurPtr == '\t'))
1286 ++CurPtr;
Chris Lattnerd88dc482008-10-12 04:05:48 +00001287
1288 // If we are keeping whitespace and other tokens, just return what we just
1289 // skipped. The next lexer invocation will return the token after the
1290 // whitespace.
1291 if (isKeepWhitespaceMode()) {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001292 FormTokenWithChars(Result, CurPtr, tok::unknown);
Chris Lattnerd88dc482008-10-12 04:05:48 +00001293 return;
1294 }
1295
Reid Spencer5f016e22007-07-11 17:01:13 +00001296 BufferPtr = CurPtr;
Chris Lattnerd2177732007-07-20 16:59:19 +00001297 Result.setFlag(Token::LeadingSpace);
Reid Spencer5f016e22007-07-11 17:01:13 +00001298 }
1299
1300 unsigned SizeTmp, SizeTmp2; // Temporaries for use in cases below.
1301
1302 // Read a character, advancing over it.
1303 char Char = getAndAdvanceChar(CurPtr, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001304 tok::TokenKind Kind;
1305
Reid Spencer5f016e22007-07-11 17:01:13 +00001306 switch (Char) {
1307 case 0: // Null.
1308 // Found end of file?
1309 if (CurPtr-1 == BufferEnd) {
1310 // Read the PP instance variable into an automatic variable, because
1311 // LexEndOfFile will often delete 'this'.
Chris Lattner168ae2d2007-10-17 20:41:00 +00001312 Preprocessor *PPCache = PP;
Reid Spencer5f016e22007-07-11 17:01:13 +00001313 if (LexEndOfFile(Result, CurPtr-1)) // Retreat back into the file.
1314 return; // Got a token to return.
Chris Lattner168ae2d2007-10-17 20:41:00 +00001315 assert(PPCache && "Raw buffer::LexEndOfFile should return a token");
1316 return PPCache->Lex(Result);
Reid Spencer5f016e22007-07-11 17:01:13 +00001317 }
1318
Chris Lattner74d15df2008-11-22 02:02:22 +00001319 if (!isLexingRawMode())
1320 Diag(CurPtr-1, diag::null_in_file);
Chris Lattnerd2177732007-07-20 16:59:19 +00001321 Result.setFlag(Token::LeadingSpace);
Chris Lattnerd88dc482008-10-12 04:05:48 +00001322 if (SkipWhitespace(Result, CurPtr))
1323 return; // KeepWhitespaceMode
1324
Reid Spencer5f016e22007-07-11 17:01:13 +00001325 goto LexNextToken; // GCC isn't tail call eliminating.
1326 case '\n':
1327 case '\r':
1328 // If we are inside a preprocessor directive and we see the end of line,
1329 // we know we are done with the directive, so return an EOM token.
1330 if (ParsingPreprocessorDirective) {
1331 // Done parsing the "line".
1332 ParsingPreprocessorDirective = false;
1333
1334 // Restore comment saving mode, in case it was disabled for directive.
Chris Lattnerf744d132008-10-12 03:27:19 +00001335 SetCommentRetentionState(PP->getCommentRetentionState());
Reid Spencer5f016e22007-07-11 17:01:13 +00001336
1337 // Since we consumed a newline, we are back at the start of a line.
1338 IsAtStartOfLine = true;
1339
Chris Lattner9e6293d2008-10-12 04:51:35 +00001340 Kind = tok::eom;
Reid Spencer5f016e22007-07-11 17:01:13 +00001341 break;
1342 }
1343 // The returned token is at the start of the line.
Chris Lattnerd2177732007-07-20 16:59:19 +00001344 Result.setFlag(Token::StartOfLine);
Reid Spencer5f016e22007-07-11 17:01:13 +00001345 // No leading whitespace seen so far.
Chris Lattnerd2177732007-07-20 16:59:19 +00001346 Result.clearFlag(Token::LeadingSpace);
Chris Lattnerd88dc482008-10-12 04:05:48 +00001347
1348 if (SkipWhitespace(Result, CurPtr))
1349 return; // KeepWhitespaceMode
Reid Spencer5f016e22007-07-11 17:01:13 +00001350 goto LexNextToken; // GCC isn't tail call eliminating.
1351 case ' ':
1352 case '\t':
1353 case '\f':
1354 case '\v':
Chris Lattner8133cfc2007-07-22 06:29:05 +00001355 SkipHorizontalWhitespace:
Chris Lattnerd2177732007-07-20 16:59:19 +00001356 Result.setFlag(Token::LeadingSpace);
Chris Lattnerd88dc482008-10-12 04:05:48 +00001357 if (SkipWhitespace(Result, CurPtr))
1358 return; // KeepWhitespaceMode
Chris Lattner8133cfc2007-07-22 06:29:05 +00001359
1360 SkipIgnoredUnits:
1361 CurPtr = BufferPtr;
1362
1363 // If the next token is obviously a // or /* */ comment, skip it efficiently
1364 // too (without going through the big switch stmt).
Chris Lattner8402c732009-01-16 22:39:25 +00001365 if (CurPtr[0] == '/' && CurPtr[1] == '/' && !inKeepCommentMode() &&
1366 Features.BCPLComment) {
Chris Lattner8133cfc2007-07-22 06:29:05 +00001367 SkipBCPLComment(Result, CurPtr+2);
1368 goto SkipIgnoredUnits;
Chris Lattnerfa95a012008-10-12 03:22:02 +00001369 } else if (CurPtr[0] == '/' && CurPtr[1] == '*' && !inKeepCommentMode()) {
Chris Lattner8133cfc2007-07-22 06:29:05 +00001370 SkipBlockComment(Result, CurPtr+2);
1371 goto SkipIgnoredUnits;
1372 } else if (isHorizontalWhitespace(*CurPtr)) {
1373 goto SkipHorizontalWhitespace;
1374 }
Reid Spencer5f016e22007-07-11 17:01:13 +00001375 goto LexNextToken; // GCC isn't tail call eliminating.
1376
Chris Lattner3a570772008-01-03 17:58:54 +00001377 // C99 6.4.4.1: Integer Constants.
1378 // C99 6.4.4.2: Floating Constants.
1379 case '0': case '1': case '2': case '3': case '4':
1380 case '5': case '6': case '7': case '8': case '9':
1381 // Notify MIOpt that we read a non-whitespace/non-comment token.
1382 MIOpt.ReadToken();
1383 return LexNumericConstant(Result, CurPtr);
1384
1385 case 'L': // Identifier (Loony) or wide literal (L'x' or L"xyz").
Reid Spencer5f016e22007-07-11 17:01:13 +00001386 // Notify MIOpt that we read a non-whitespace/non-comment token.
1387 MIOpt.ReadToken();
1388 Char = getCharAndSize(CurPtr, SizeTmp);
1389
1390 // Wide string literal.
1391 if (Char == '"')
1392 return LexStringLiteral(Result, ConsumeChar(CurPtr, SizeTmp, Result),
1393 true);
1394
1395 // Wide character constant.
1396 if (Char == '\'')
1397 return LexCharConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result));
1398 // FALL THROUGH, treating L like the start of an identifier.
1399
1400 // C99 6.4.2: Identifiers.
1401 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G':
1402 case 'H': case 'I': case 'J': case 'K': /*'L'*/case 'M': case 'N':
1403 case 'O': case 'P': case 'Q': case 'R': case 'S': case 'T': case 'U':
1404 case 'V': case 'W': case 'X': case 'Y': case 'Z':
1405 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g':
1406 case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n':
1407 case 'o': case 'p': case 'q': case 'r': case 's': case 't': case 'u':
1408 case 'v': case 'w': case 'x': case 'y': case 'z':
1409 case '_':
1410 // Notify MIOpt that we read a non-whitespace/non-comment token.
1411 MIOpt.ReadToken();
1412 return LexIdentifier(Result, CurPtr);
Chris Lattner3a570772008-01-03 17:58:54 +00001413
1414 case '$': // $ in identifiers.
1415 if (Features.DollarIdents) {
Chris Lattner74d15df2008-11-22 02:02:22 +00001416 if (!isLexingRawMode())
1417 Diag(CurPtr-1, diag::ext_dollar_in_identifier);
Chris Lattner3a570772008-01-03 17:58:54 +00001418 // Notify MIOpt that we read a non-whitespace/non-comment token.
1419 MIOpt.ReadToken();
1420 return LexIdentifier(Result, CurPtr);
1421 }
Reid Spencer5f016e22007-07-11 17:01:13 +00001422
Chris Lattner9e6293d2008-10-12 04:51:35 +00001423 Kind = tok::unknown;
Chris Lattner3a570772008-01-03 17:58:54 +00001424 break;
Reid Spencer5f016e22007-07-11 17:01:13 +00001425
1426 // C99 6.4.4: Character Constants.
1427 case '\'':
1428 // Notify MIOpt that we read a non-whitespace/non-comment token.
1429 MIOpt.ReadToken();
1430 return LexCharConstant(Result, CurPtr);
1431
1432 // C99 6.4.5: String Literals.
1433 case '"':
1434 // Notify MIOpt that we read a non-whitespace/non-comment token.
1435 MIOpt.ReadToken();
1436 return LexStringLiteral(Result, CurPtr, false);
1437
1438 // C99 6.4.6: Punctuators.
1439 case '?':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001440 Kind = tok::question;
Reid Spencer5f016e22007-07-11 17:01:13 +00001441 break;
1442 case '[':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001443 Kind = tok::l_square;
Reid Spencer5f016e22007-07-11 17:01:13 +00001444 break;
1445 case ']':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001446 Kind = tok::r_square;
Reid Spencer5f016e22007-07-11 17:01:13 +00001447 break;
1448 case '(':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001449 Kind = tok::l_paren;
Reid Spencer5f016e22007-07-11 17:01:13 +00001450 break;
1451 case ')':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001452 Kind = tok::r_paren;
Reid Spencer5f016e22007-07-11 17:01:13 +00001453 break;
1454 case '{':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001455 Kind = tok::l_brace;
Reid Spencer5f016e22007-07-11 17:01:13 +00001456 break;
1457 case '}':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001458 Kind = tok::r_brace;
Reid Spencer5f016e22007-07-11 17:01:13 +00001459 break;
1460 case '.':
1461 Char = getCharAndSize(CurPtr, SizeTmp);
1462 if (Char >= '0' && Char <= '9') {
1463 // Notify MIOpt that we read a non-whitespace/non-comment token.
1464 MIOpt.ReadToken();
1465
1466 return LexNumericConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result));
1467 } else if (Features.CPlusPlus && Char == '*') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001468 Kind = tok::periodstar;
Reid Spencer5f016e22007-07-11 17:01:13 +00001469 CurPtr += SizeTmp;
1470 } else if (Char == '.' &&
1471 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '.') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001472 Kind = tok::ellipsis;
Reid Spencer5f016e22007-07-11 17:01:13 +00001473 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
1474 SizeTmp2, Result);
1475 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001476 Kind = tok::period;
Reid Spencer5f016e22007-07-11 17:01:13 +00001477 }
1478 break;
1479 case '&':
1480 Char = getCharAndSize(CurPtr, SizeTmp);
1481 if (Char == '&') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001482 Kind = tok::ampamp;
Reid Spencer5f016e22007-07-11 17:01:13 +00001483 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1484 } else if (Char == '=') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001485 Kind = tok::ampequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001486 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1487 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001488 Kind = tok::amp;
Reid Spencer5f016e22007-07-11 17:01:13 +00001489 }
1490 break;
1491 case '*':
1492 if (getCharAndSize(CurPtr, SizeTmp) == '=') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001493 Kind = tok::starequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001494 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1495 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001496 Kind = tok::star;
Reid Spencer5f016e22007-07-11 17:01:13 +00001497 }
1498 break;
1499 case '+':
1500 Char = getCharAndSize(CurPtr, SizeTmp);
1501 if (Char == '+') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001502 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001503 Kind = tok::plusplus;
Reid Spencer5f016e22007-07-11 17:01:13 +00001504 } else if (Char == '=') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001505 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001506 Kind = tok::plusequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001507 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001508 Kind = tok::plus;
Reid Spencer5f016e22007-07-11 17:01:13 +00001509 }
1510 break;
1511 case '-':
1512 Char = getCharAndSize(CurPtr, SizeTmp);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001513 if (Char == '-') { // --
Reid Spencer5f016e22007-07-11 17:01:13 +00001514 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001515 Kind = tok::minusminus;
Reid Spencer5f016e22007-07-11 17:01:13 +00001516 } else if (Char == '>' && Features.CPlusPlus &&
Chris Lattner9e6293d2008-10-12 04:51:35 +00001517 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '*') { // C++ ->*
Reid Spencer5f016e22007-07-11 17:01:13 +00001518 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
1519 SizeTmp2, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001520 Kind = tok::arrowstar;
1521 } else if (Char == '>') { // ->
Reid Spencer5f016e22007-07-11 17:01:13 +00001522 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001523 Kind = tok::arrow;
1524 } else if (Char == '=') { // -=
Reid Spencer5f016e22007-07-11 17:01:13 +00001525 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001526 Kind = tok::minusequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001527 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001528 Kind = tok::minus;
Reid Spencer5f016e22007-07-11 17:01:13 +00001529 }
1530 break;
1531 case '~':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001532 Kind = tok::tilde;
Reid Spencer5f016e22007-07-11 17:01:13 +00001533 break;
1534 case '!':
1535 if (getCharAndSize(CurPtr, SizeTmp) == '=') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001536 Kind = tok::exclaimequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001537 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1538 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001539 Kind = tok::exclaim;
Reid Spencer5f016e22007-07-11 17:01:13 +00001540 }
1541 break;
1542 case '/':
1543 // 6.4.9: Comments
1544 Char = getCharAndSize(CurPtr, SizeTmp);
1545 if (Char == '/') { // BCPL comment.
Chris Lattner8402c732009-01-16 22:39:25 +00001546 // Even if BCPL comments are disabled (e.g. in C89 mode), we generally
1547 // want to lex this as a comment. There is one problem with this though,
1548 // that in one particular corner case, this can change the behavior of the
1549 // resultant program. For example, In "foo //**/ bar", C89 would lex
1550 // this as "foo / bar" and langauges with BCPL comments would lex it as
1551 // "foo". Check to see if the character after the second slash is a '*'.
1552 // If so, we will lex that as a "/" instead of the start of a comment.
1553 if (Features.BCPLComment ||
1554 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) != '*') {
1555 if (SkipBCPLComment(Result, ConsumeChar(CurPtr, SizeTmp, Result)))
1556 return; // KeepCommentMode
Chris Lattner2d381892008-10-12 04:15:42 +00001557
Chris Lattner8402c732009-01-16 22:39:25 +00001558 // It is common for the tokens immediately after a // comment to be
1559 // whitespace (indentation for the next line). Instead of going through
1560 // the big switch, handle it efficiently now.
1561 goto SkipIgnoredUnits;
1562 }
1563 }
1564
1565 if (Char == '*') { // /**/ comment.
Reid Spencer5f016e22007-07-11 17:01:13 +00001566 if (SkipBlockComment(Result, ConsumeChar(CurPtr, SizeTmp, Result)))
Chris Lattner2d381892008-10-12 04:15:42 +00001567 return; // KeepCommentMode
1568 goto LexNextToken; // GCC isn't tail call eliminating.
Chris Lattner8402c732009-01-16 22:39:25 +00001569 }
1570
1571 if (Char == '=') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001572 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001573 Kind = tok::slashequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001574 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001575 Kind = tok::slash;
Reid Spencer5f016e22007-07-11 17:01:13 +00001576 }
1577 break;
1578 case '%':
1579 Char = getCharAndSize(CurPtr, SizeTmp);
1580 if (Char == '=') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001581 Kind = tok::percentequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001582 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1583 } else if (Features.Digraphs && Char == '>') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001584 Kind = tok::r_brace; // '%>' -> '}'
Reid Spencer5f016e22007-07-11 17:01:13 +00001585 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1586 } else if (Features.Digraphs && Char == ':') {
1587 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1588 Char = getCharAndSize(CurPtr, SizeTmp);
1589 if (Char == '%' && getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == ':') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001590 Kind = tok::hashhash; // '%:%:' -> '##'
Reid Spencer5f016e22007-07-11 17:01:13 +00001591 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
1592 SizeTmp2, Result);
1593 } else if (Char == '@' && Features.Microsoft) { // %:@ -> #@ -> Charize
Reid Spencer5f016e22007-07-11 17:01:13 +00001594 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner74d15df2008-11-22 02:02:22 +00001595 if (!isLexingRawMode())
1596 Diag(BufferPtr, diag::charize_microsoft_ext);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001597 Kind = tok::hashat;
Reid Spencer5f016e22007-07-11 17:01:13 +00001598 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001599 Kind = tok::hash; // '%:' -> '#'
Reid Spencer5f016e22007-07-11 17:01:13 +00001600
1601 // We parsed a # character. If this occurs at the start of the line,
1602 // it's actually the start of a preprocessing directive. Callback to
1603 // the preprocessor to handle it.
1604 // FIXME: -fpreprocessed mode??
1605 if (Result.isAtStartOfLine() && !LexingRawMode) {
1606 BufferPtr = CurPtr;
Chris Lattner168ae2d2007-10-17 20:41:00 +00001607 PP->HandleDirective(Result);
Reid Spencer5f016e22007-07-11 17:01:13 +00001608
1609 // As an optimization, if the preprocessor didn't switch lexers, tail
1610 // recurse.
Chris Lattner168ae2d2007-10-17 20:41:00 +00001611 if (PP->isCurrentLexer(this)) {
Reid Spencer5f016e22007-07-11 17:01:13 +00001612 // Start a new token. If this is a #include or something, the PP may
1613 // want us starting at the beginning of the line again. If so, set
1614 // the StartOfLine flag.
1615 if (IsAtStartOfLine) {
Chris Lattnerd2177732007-07-20 16:59:19 +00001616 Result.setFlag(Token::StartOfLine);
Reid Spencer5f016e22007-07-11 17:01:13 +00001617 IsAtStartOfLine = false;
1618 }
1619 goto LexNextToken; // GCC isn't tail call eliminating.
1620 }
1621
Chris Lattner168ae2d2007-10-17 20:41:00 +00001622 return PP->Lex(Result);
Reid Spencer5f016e22007-07-11 17:01:13 +00001623 }
1624 }
1625 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001626 Kind = tok::percent;
Reid Spencer5f016e22007-07-11 17:01:13 +00001627 }
1628 break;
1629 case '<':
1630 Char = getCharAndSize(CurPtr, SizeTmp);
1631 if (ParsingFilename) {
1632 return LexAngledStringLiteral(Result, CurPtr+SizeTmp);
1633 } else if (Char == '<' &&
1634 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '=') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001635 Kind = tok::lesslessequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001636 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
1637 SizeTmp2, Result);
1638 } else if (Char == '<') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001639 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001640 Kind = tok::lessless;
Reid Spencer5f016e22007-07-11 17:01:13 +00001641 } else if (Char == '=') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001642 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001643 Kind = tok::lessequal;
1644 } else if (Features.Digraphs && Char == ':') { // '<:' -> '['
Reid Spencer5f016e22007-07-11 17:01:13 +00001645 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001646 Kind = tok::l_square;
1647 } else if (Features.Digraphs && Char == '%') { // '<%' -> '{'
Reid Spencer5f016e22007-07-11 17:01:13 +00001648 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001649 Kind = tok::l_brace;
Reid Spencer5f016e22007-07-11 17:01:13 +00001650 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001651 Kind = tok::less;
Reid Spencer5f016e22007-07-11 17:01:13 +00001652 }
1653 break;
1654 case '>':
1655 Char = getCharAndSize(CurPtr, SizeTmp);
1656 if (Char == '=') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001657 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001658 Kind = tok::greaterequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001659 } else if (Char == '>' &&
1660 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '=') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001661 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result),
1662 SizeTmp2, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001663 Kind = tok::greatergreaterequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001664 } else if (Char == '>') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001665 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001666 Kind = tok::greatergreater;
Reid Spencer5f016e22007-07-11 17:01:13 +00001667 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001668 Kind = tok::greater;
Reid Spencer5f016e22007-07-11 17:01:13 +00001669 }
1670 break;
1671 case '^':
1672 Char = getCharAndSize(CurPtr, SizeTmp);
1673 if (Char == '=') {
Reid Spencer5f016e22007-07-11 17:01:13 +00001674 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
Chris Lattner9e6293d2008-10-12 04:51:35 +00001675 Kind = tok::caretequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001676 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001677 Kind = tok::caret;
Reid Spencer5f016e22007-07-11 17:01:13 +00001678 }
1679 break;
1680 case '|':
1681 Char = getCharAndSize(CurPtr, SizeTmp);
1682 if (Char == '=') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001683 Kind = tok::pipeequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001684 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1685 } else if (Char == '|') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001686 Kind = tok::pipepipe;
Reid Spencer5f016e22007-07-11 17:01:13 +00001687 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1688 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001689 Kind = tok::pipe;
Reid Spencer5f016e22007-07-11 17:01:13 +00001690 }
1691 break;
1692 case ':':
1693 Char = getCharAndSize(CurPtr, SizeTmp);
1694 if (Features.Digraphs && Char == '>') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001695 Kind = tok::r_square; // ':>' -> ']'
Reid Spencer5f016e22007-07-11 17:01:13 +00001696 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1697 } else if (Features.CPlusPlus && Char == ':') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001698 Kind = tok::coloncolon;
Reid Spencer5f016e22007-07-11 17:01:13 +00001699 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1700 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001701 Kind = tok::colon;
Reid Spencer5f016e22007-07-11 17:01:13 +00001702 }
1703 break;
1704 case ';':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001705 Kind = tok::semi;
Reid Spencer5f016e22007-07-11 17:01:13 +00001706 break;
1707 case '=':
1708 Char = getCharAndSize(CurPtr, SizeTmp);
1709 if (Char == '=') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001710 Kind = tok::equalequal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001711 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1712 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001713 Kind = tok::equal;
Reid Spencer5f016e22007-07-11 17:01:13 +00001714 }
1715 break;
1716 case ',':
Chris Lattner9e6293d2008-10-12 04:51:35 +00001717 Kind = tok::comma;
Reid Spencer5f016e22007-07-11 17:01:13 +00001718 break;
1719 case '#':
1720 Char = getCharAndSize(CurPtr, SizeTmp);
1721 if (Char == '#') {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001722 Kind = tok::hashhash;
Reid Spencer5f016e22007-07-11 17:01:13 +00001723 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1724 } else if (Char == '@' && Features.Microsoft) { // #@ -> Charize
Chris Lattner9e6293d2008-10-12 04:51:35 +00001725 Kind = tok::hashat;
Chris Lattner74d15df2008-11-22 02:02:22 +00001726 if (!isLexingRawMode())
1727 Diag(BufferPtr, diag::charize_microsoft_ext);
Reid Spencer5f016e22007-07-11 17:01:13 +00001728 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result);
1729 } else {
Chris Lattner9e6293d2008-10-12 04:51:35 +00001730 Kind = tok::hash;
Reid Spencer5f016e22007-07-11 17:01:13 +00001731 // We parsed a # character. If this occurs at the start of the line,
1732 // it's actually the start of a preprocessing directive. Callback to
1733 // the preprocessor to handle it.
1734 // FIXME: -fpreprocessed mode??
1735 if (Result.isAtStartOfLine() && !LexingRawMode) {
1736 BufferPtr = CurPtr;
Chris Lattner168ae2d2007-10-17 20:41:00 +00001737 PP->HandleDirective(Result);
Reid Spencer5f016e22007-07-11 17:01:13 +00001738
1739 // As an optimization, if the preprocessor didn't switch lexers, tail
1740 // recurse.
Chris Lattner168ae2d2007-10-17 20:41:00 +00001741 if (PP->isCurrentLexer(this)) {
Reid Spencer5f016e22007-07-11 17:01:13 +00001742 // Start a new token. If this is a #include or something, the PP may
1743 // want us starting at the beginning of the line again. If so, set
1744 // the StartOfLine flag.
1745 if (IsAtStartOfLine) {
Chris Lattnerd2177732007-07-20 16:59:19 +00001746 Result.setFlag(Token::StartOfLine);
Reid Spencer5f016e22007-07-11 17:01:13 +00001747 IsAtStartOfLine = false;
1748 }
1749 goto LexNextToken; // GCC isn't tail call eliminating.
1750 }
Chris Lattner168ae2d2007-10-17 20:41:00 +00001751 return PP->Lex(Result);
Reid Spencer5f016e22007-07-11 17:01:13 +00001752 }
1753 }
1754 break;
1755
Chris Lattner3a570772008-01-03 17:58:54 +00001756 case '@':
1757 // Objective C support.
1758 if (CurPtr[-1] == '@' && Features.ObjC1)
Chris Lattner9e6293d2008-10-12 04:51:35 +00001759 Kind = tok::at;
Chris Lattner3a570772008-01-03 17:58:54 +00001760 else
Chris Lattner9e6293d2008-10-12 04:51:35 +00001761 Kind = tok::unknown;
Chris Lattner3a570772008-01-03 17:58:54 +00001762 break;
1763
Reid Spencer5f016e22007-07-11 17:01:13 +00001764 case '\\':
1765 // FIXME: UCN's.
1766 // FALL THROUGH.
1767 default:
Chris Lattner9e6293d2008-10-12 04:51:35 +00001768 Kind = tok::unknown;
Reid Spencer5f016e22007-07-11 17:01:13 +00001769 break;
1770 }
1771
1772 // Notify MIOpt that we read a non-whitespace/non-comment token.
1773 MIOpt.ReadToken();
1774
1775 // Update the location of token as well as BufferPtr.
Chris Lattner9e6293d2008-10-12 04:51:35 +00001776 FormTokenWithChars(Result, CurPtr, Kind);
Reid Spencer5f016e22007-07-11 17:01:13 +00001777}