Lexer.cpp revision 2d804c4325fcf3893386e16970b82fd0f9af1d7c
1//===--- Lexer.cpp - C Language Family Lexer ------------------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements the Lexer and Token interfaces. 11// 12//===----------------------------------------------------------------------===// 13// 14// 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" 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" 35#include <cctype> 36using namespace clang; 37 38static void InitCharacterInfo(); 39 40//===----------------------------------------------------------------------===// 41// Token Class Implementation 42//===----------------------------------------------------------------------===// 43 44/// isObjCAtKeyword - Return true if we have an ObjC keyword identifier. 45bool Token::isObjCAtKeyword(tok::ObjCKeywordKind objcKey) const { 46 if (IdentifierInfo *II = getIdentifierInfo()) 47 return II->getObjCKeywordID() == objcKey; 48 return false; 49} 50 51/// getObjCKeywordID - Return the ObjC keyword kind. 52tok::ObjCKeywordKind Token::getObjCKeywordID() const { 53 IdentifierInfo *specId = getIdentifierInfo(); 54 return specId ? specId->getObjCKeywordID() : tok::objc_not_keyword; 55} 56 57 58//===----------------------------------------------------------------------===// 59// Lexer Class Implementation 60//===----------------------------------------------------------------------===// 61 62void Lexer::InitLexer(const char *BufStart, const char *BufPtr, 63 const char *BufEnd) { 64 InitCharacterInfo(); 65 66 BufferStart = BufStart; 67 BufferPtr = BufPtr; 68 BufferEnd = BufEnd; 69 70 assert(BufEnd[0] == 0 && 71 "We assume that the input buffer has a null character at the end" 72 " to simplify lexing!"); 73 74 Is_PragmaLexer = false; 75 IsInConflictMarker = false; 76 77 // Start of the file is a start of line. 78 IsAtStartOfLine = true; 79 80 // We are not after parsing a #. 81 ParsingPreprocessorDirective = false; 82 83 // We are not after parsing #include. 84 ParsingFilename = false; 85 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; 91 92 // Default to not keeping comments. 93 ExtendedTokenMode = 0; 94} 95 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. 100Lexer::Lexer(FileID FID, const llvm::MemoryBuffer *InputFile, Preprocessor &PP) 101 : PreprocessorLexer(&PP, FID), 102 FileLoc(PP.getSourceManager().getLocForStartOfFile(FID)), 103 Features(PP.getLangOptions()) { 104 105 InitLexer(InputFile->getBufferStart(), InputFile->getBufferStart(), 106 InputFile->getBufferEnd()); 107 108 // Default to keeping comments if the preprocessor wants them. 109 SetCommentRetentionState(PP.getCommentRetentionState()); 110} 111 112/// 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. 115Lexer::Lexer(SourceLocation fileloc, const LangOptions &features, 116 const char *BufStart, const char *BufPtr, const char *BufEnd) 117 : FileLoc(fileloc), Features(features) { 118 119 InitLexer(BufStart, BufPtr, BufEnd); 120 121 // We *are* in raw mode. 122 LexingRawMode = true; 123} 124 125/// 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. 128Lexer::Lexer(FileID FID, const llvm::MemoryBuffer *FromFile, 129 const SourceManager &SM, const LangOptions &features) 130 : FileLoc(SM.getLocForStartOfFile(FID)), Features(features) { 131 132 InitLexer(FromFile->getBufferStart(), FromFile->getBufferStart(), 133 FromFile->getBufferEnd()); 134 135 // We *are* in raw mode. 136 LexingRawMode = true; 137} 138 139/// 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. 142/// 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 146/// "expanded from". 147/// 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! 153/// 154Lexer *Lexer::Create_PragmaLexer(SourceLocation SpellingLoc, 155 SourceLocation InstantiationLocStart, 156 SourceLocation InstantiationLocEnd, 157 unsigned TokLen, Preprocessor &PP) { 158 SourceManager &SM = PP.getSourceManager(); 159 160 // 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); 164 165 // 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 167 // scratch buffer. 168 const char *StrData = SM.getCharacterData(SpellingLoc); 169 170 L->BufferPtr = StrData; 171 L->BufferEnd = StrData+TokLen; 172 assert(L->BufferEnd[0] == 0 && "Buffer is not nul terminated!"); 173 174 // 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); 179 180 // Ensure that the lexer thinks it is inside a directive, so that end \n will 181 // return an EOM token. 182 L->ParsingPreprocessorDirective = true; 183 184 // This lexer really is for _Pragma. 185 L->Is_PragmaLexer = true; 186 return L; 187} 188 189 190/// Stringify - Convert the specified string into a C string, with surrounding 191/// ""'s, and with escaped \ and " characters. 192std::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, '\\'); 198 ++i; ++e; 199 } 200 } 201 return Result; 202} 203 204/// Stringify - Convert the specified string into a C string by escaping '\' 205/// and " characters. This does not add surrounding ""'s to the string. 206void 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, '\\'); 210 ++i; ++e; 211 } 212 } 213} 214 215static bool isWhitespace(unsigned char c); 216 217/// MeasureTokenLength - Relex the token at the specified location and return 218/// its length in bytes in the input file. If the token needs cleaning (e.g. 219/// includes a trigraph or an escaped newline) then this count includes bytes 220/// that are part of that. 221unsigned Lexer::MeasureTokenLength(SourceLocation Loc, 222 const SourceManager &SM, 223 const LangOptions &LangOpts) { 224 // TODO: this could be special cased for common tokens like identifiers, ')', 225 // etc to make this faster, if it mattered. Just look at StrData[0] to handle 226 // all obviously single-char tokens. This could use 227 // Lexer::isObviouslySimpleCharacter for example to handle identifiers or 228 // something. 229 230 // If this comes from a macro expansion, we really do want the macro name, not 231 // the token this macro expanded to. 232 Loc = SM.getInstantiationLoc(Loc); 233 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc); 234 bool Invalid = false; 235 llvm::StringRef Buffer = SM.getBufferData(LocInfo.first, &Invalid); 236 if (Invalid) 237 return 0; 238 239 const char *StrData = Buffer.data()+LocInfo.second; 240 241 if (isWhitespace(StrData[0])) 242 return 0; 243 244 // Create a lexer starting at the beginning of this token. 245 Lexer TheLexer(Loc, LangOpts, Buffer.begin(), StrData, Buffer.end()); 246 TheLexer.SetCommentRetentionState(true); 247 Token TheTok; 248 TheLexer.LexFromRawLexer(TheTok); 249 return TheTok.getLength(); 250} 251 252SourceLocation Lexer::GetBeginningOfToken(SourceLocation Loc, 253 const SourceManager &SM, 254 const LangOptions &LangOpts) { 255 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc); 256 bool Invalid = false; 257 llvm::StringRef Buffer = SM.getBufferData(LocInfo.first, &Invalid); 258 if (Invalid) 259 return Loc; 260 261 // Back up from the current location until we hit the beginning of a line 262 // (or the buffer). We'll relex from that point. 263 const char *BufStart = Buffer.data(); 264 const char *StrData = BufStart+LocInfo.second; 265 if (StrData[0] == '\n' || StrData[0] == '\r') 266 return Loc; 267 268 const char *LexStart = StrData; 269 while (LexStart != BufStart) { 270 if (LexStart[0] == '\n' || LexStart[0] == '\r') { 271 ++LexStart; 272 break; 273 } 274 275 --LexStart; 276 } 277 278 // Create a lexer starting at the beginning of this token. 279 SourceLocation LexerStartLoc = Loc.getFileLocWithOffset(-LocInfo.second); 280 Lexer TheLexer(LexerStartLoc, LangOpts, BufStart, LexStart, Buffer.end()); 281 TheLexer.SetCommentRetentionState(true); 282 283 // Lex tokens until we find the token that contains the source location. 284 Token TheTok; 285 do { 286 TheLexer.LexFromRawLexer(TheTok); 287 288 if (TheLexer.getBufferLocation() > StrData) { 289 // Lexing this token has taken the lexer past the source location we're 290 // looking for. If the current token encompasses our source location, 291 // return the beginning of that token. 292 if (TheLexer.getBufferLocation() - TheTok.getLength() <= StrData) 293 return TheTok.getLocation(); 294 295 // We ended up skipping over the source location entirely, which means 296 // that it points into whitespace. We're done here. 297 break; 298 } 299 } while (TheTok.getKind() != tok::eof); 300 301 // We've passed our source location; just return the original source location. 302 return Loc; 303} 304 305namespace { 306 enum PreambleDirectiveKind { 307 PDK_Skipped, 308 PDK_StartIf, 309 PDK_EndIf, 310 PDK_Unknown 311 }; 312} 313 314std::pair<unsigned, bool> 315Lexer::ComputePreamble(const llvm::MemoryBuffer *Buffer, unsigned MaxLines) { 316 // Create a lexer starting at the beginning of the file. Note that we use a 317 // "fake" file source location at offset 1 so that the lexer will track our 318 // position within the file. 319 const unsigned StartOffset = 1; 320 SourceLocation StartLoc = SourceLocation::getFromRawEncoding(StartOffset); 321 LangOptions LangOpts; 322 Lexer TheLexer(StartLoc, LangOpts, Buffer->getBufferStart(), 323 Buffer->getBufferStart(), Buffer->getBufferEnd()); 324 325 bool InPreprocessorDirective = false; 326 Token TheTok; 327 Token IfStartTok; 328 unsigned IfCount = 0; 329 unsigned Line = 0; 330 331 do { 332 TheLexer.LexFromRawLexer(TheTok); 333 334 if (InPreprocessorDirective) { 335 // If we've hit the end of the file, we're done. 336 if (TheTok.getKind() == tok::eof) { 337 InPreprocessorDirective = false; 338 break; 339 } 340 341 // If we haven't hit the end of the preprocessor directive, skip this 342 // token. 343 if (!TheTok.isAtStartOfLine()) 344 continue; 345 346 // We've passed the end of the preprocessor directive, and will look 347 // at this token again below. 348 InPreprocessorDirective = false; 349 } 350 351 // Keep track of the # of lines in the preamble. 352 if (TheTok.isAtStartOfLine()) { 353 ++Line; 354 355 // If we were asked to limit the number of lines in the preamble, 356 // and we're about to exceed that limit, we're done. 357 if (MaxLines && Line >= MaxLines) 358 break; 359 } 360 361 // Comments are okay; skip over them. 362 if (TheTok.getKind() == tok::comment) 363 continue; 364 365 if (TheTok.isAtStartOfLine() && TheTok.getKind() == tok::hash) { 366 // This is the start of a preprocessor directive. 367 Token HashTok = TheTok; 368 InPreprocessorDirective = true; 369 370 // Figure out which direective this is. Since we're lexing raw tokens, 371 // we don't have an identifier table available. Instead, just look at 372 // the raw identifier to recognize and categorize preprocessor directives. 373 TheLexer.LexFromRawLexer(TheTok); 374 if (TheTok.getKind() == tok::identifier && !TheTok.needsCleaning()) { 375 const char *IdStart = Buffer->getBufferStart() 376 + TheTok.getLocation().getRawEncoding() - 1; 377 llvm::StringRef Keyword(IdStart, TheTok.getLength()); 378 PreambleDirectiveKind PDK 379 = llvm::StringSwitch<PreambleDirectiveKind>(Keyword) 380 .Case("include", PDK_Skipped) 381 .Case("__include_macros", PDK_Skipped) 382 .Case("define", PDK_Skipped) 383 .Case("undef", PDK_Skipped) 384 .Case("line", PDK_Skipped) 385 .Case("error", PDK_Skipped) 386 .Case("pragma", PDK_Skipped) 387 .Case("import", PDK_Skipped) 388 .Case("include_next", PDK_Skipped) 389 .Case("warning", PDK_Skipped) 390 .Case("ident", PDK_Skipped) 391 .Case("sccs", PDK_Skipped) 392 .Case("assert", PDK_Skipped) 393 .Case("unassert", PDK_Skipped) 394 .Case("if", PDK_StartIf) 395 .Case("ifdef", PDK_StartIf) 396 .Case("ifndef", PDK_StartIf) 397 .Case("elif", PDK_Skipped) 398 .Case("else", PDK_Skipped) 399 .Case("endif", PDK_EndIf) 400 .Default(PDK_Unknown); 401 402 switch (PDK) { 403 case PDK_Skipped: 404 continue; 405 406 case PDK_StartIf: 407 if (IfCount == 0) 408 IfStartTok = HashTok; 409 410 ++IfCount; 411 continue; 412 413 case PDK_EndIf: 414 // Mismatched #endif. The preamble ends here. 415 if (IfCount == 0) 416 break; 417 418 --IfCount; 419 continue; 420 421 case PDK_Unknown: 422 // We don't know what this directive is; stop at the '#'. 423 break; 424 } 425 } 426 427 // We only end up here if we didn't recognize the preprocessor 428 // directive or it was one that can't occur in the preamble at this 429 // point. Roll back the current token to the location of the '#'. 430 InPreprocessorDirective = false; 431 TheTok = HashTok; 432 } 433 434 // We hit a token that we don't recognize as being in the 435 // "preprocessing only" part of the file, so we're no longer in 436 // the preamble. 437 break; 438 } while (true); 439 440 SourceLocation End = IfCount? IfStartTok.getLocation() : TheTok.getLocation(); 441 return std::make_pair(End.getRawEncoding() - StartLoc.getRawEncoding(), 442 IfCount? IfStartTok.isAtStartOfLine() 443 : TheTok.isAtStartOfLine()); 444} 445 446//===----------------------------------------------------------------------===// 447// Character information. 448//===----------------------------------------------------------------------===// 449 450enum { 451 CHAR_HORZ_WS = 0x01, // ' ', '\t', '\f', '\v'. Note, no '\0' 452 CHAR_VERT_WS = 0x02, // '\r', '\n' 453 CHAR_LETTER = 0x04, // a-z,A-Z 454 CHAR_NUMBER = 0x08, // 0-9 455 CHAR_UNDER = 0x10, // _ 456 CHAR_PERIOD = 0x20 // . 457}; 458 459// Statically initialize CharInfo table based on ASCII character set 460// Reference: FreeBSD 7.2 /usr/share/misc/ascii 461static const unsigned char CharInfo[256] = 462{ 463// 0 NUL 1 SOH 2 STX 3 ETX 464// 4 EOT 5 ENQ 6 ACK 7 BEL 465 0 , 0 , 0 , 0 , 466 0 , 0 , 0 , 0 , 467// 8 BS 9 HT 10 NL 11 VT 468//12 NP 13 CR 14 SO 15 SI 469 0 , CHAR_HORZ_WS, CHAR_VERT_WS, CHAR_HORZ_WS, 470 CHAR_HORZ_WS, CHAR_VERT_WS, 0 , 0 , 471//16 DLE 17 DC1 18 DC2 19 DC3 472//20 DC4 21 NAK 22 SYN 23 ETB 473 0 , 0 , 0 , 0 , 474 0 , 0 , 0 , 0 , 475//24 CAN 25 EM 26 SUB 27 ESC 476//28 FS 29 GS 30 RS 31 US 477 0 , 0 , 0 , 0 , 478 0 , 0 , 0 , 0 , 479//32 SP 33 ! 34 " 35 # 480//36 $ 37 % 38 & 39 ' 481 CHAR_HORZ_WS, 0 , 0 , 0 , 482 0 , 0 , 0 , 0 , 483//40 ( 41 ) 42 * 43 + 484//44 , 45 - 46 . 47 / 485 0 , 0 , 0 , 0 , 486 0 , 0 , CHAR_PERIOD , 0 , 487//48 0 49 1 50 2 51 3 488//52 4 53 5 54 6 55 7 489 CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER , 490 CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER , CHAR_NUMBER , 491//56 8 57 9 58 : 59 ; 492//60 < 61 = 62 > 63 ? 493 CHAR_NUMBER , CHAR_NUMBER , 0 , 0 , 494 0 , 0 , 0 , 0 , 495//64 @ 65 A 66 B 67 C 496//68 D 69 E 70 F 71 G 497 0 , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 498 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 499//72 H 73 I 74 J 75 K 500//76 L 77 M 78 N 79 O 501 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 502 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 503//80 P 81 Q 82 R 83 S 504//84 T 85 U 86 V 87 W 505 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 506 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 507//88 X 89 Y 90 Z 91 [ 508//92 \ 93 ] 94 ^ 95 _ 509 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 0 , 510 0 , 0 , 0 , CHAR_UNDER , 511//96 ` 97 a 98 b 99 c 512//100 d 101 e 102 f 103 g 513 0 , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 514 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 515//104 h 105 i 106 j 107 k 516//108 l 109 m 110 n 111 o 517 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 518 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 519//112 p 113 q 114 r 115 s 520//116 t 117 u 118 v 119 w 521 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 522 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 523//120 x 121 y 122 z 123 { 524//124 | 125 } 126 ~ 127 DEL 525 CHAR_LETTER , CHAR_LETTER , CHAR_LETTER , 0 , 526 0 , 0 , 0 , 0 527}; 528 529static void InitCharacterInfo() { 530 static bool isInited = false; 531 if (isInited) return; 532 // check the statically-initialized CharInfo table 533 assert(CHAR_HORZ_WS == CharInfo[(int)' ']); 534 assert(CHAR_HORZ_WS == CharInfo[(int)'\t']); 535 assert(CHAR_HORZ_WS == CharInfo[(int)'\f']); 536 assert(CHAR_HORZ_WS == CharInfo[(int)'\v']); 537 assert(CHAR_VERT_WS == CharInfo[(int)'\n']); 538 assert(CHAR_VERT_WS == CharInfo[(int)'\r']); 539 assert(CHAR_UNDER == CharInfo[(int)'_']); 540 assert(CHAR_PERIOD == CharInfo[(int)'.']); 541 for (unsigned i = 'a'; i <= 'z'; ++i) { 542 assert(CHAR_LETTER == CharInfo[i]); 543 assert(CHAR_LETTER == CharInfo[i+'A'-'a']); 544 } 545 for (unsigned i = '0'; i <= '9'; ++i) 546 assert(CHAR_NUMBER == CharInfo[i]); 547 548 isInited = true; 549} 550 551 552/// isIdentifierBody - Return true if this is the body character of an 553/// identifier, which is [a-zA-Z0-9_]. 554static inline bool isIdentifierBody(unsigned char c) { 555 return (CharInfo[c] & (CHAR_LETTER|CHAR_NUMBER|CHAR_UNDER)) ? true : false; 556} 557 558/// isHorizontalWhitespace - Return true if this character is horizontal 559/// whitespace: ' ', '\t', '\f', '\v'. Note that this returns false for '\0'. 560static inline bool isHorizontalWhitespace(unsigned char c) { 561 return (CharInfo[c] & CHAR_HORZ_WS) ? true : false; 562} 563 564/// isWhitespace - Return true if this character is horizontal or vertical 565/// whitespace: ' ', '\t', '\f', '\v', '\n', '\r'. Note that this returns false 566/// for '\0'. 567static inline bool isWhitespace(unsigned char c) { 568 return (CharInfo[c] & (CHAR_HORZ_WS|CHAR_VERT_WS)) ? true : false; 569} 570 571/// isNumberBody - Return true if this is the body character of an 572/// preprocessing number, which is [a-zA-Z0-9_.]. 573static inline bool isNumberBody(unsigned char c) { 574 return (CharInfo[c] & (CHAR_LETTER|CHAR_NUMBER|CHAR_UNDER|CHAR_PERIOD)) ? 575 true : false; 576} 577 578 579//===----------------------------------------------------------------------===// 580// Diagnostics forwarding code. 581//===----------------------------------------------------------------------===// 582 583/// GetMappedTokenLoc - If lexing out of a 'mapped buffer', where we pretend the 584/// lexer buffer was all instantiated at a single point, perform the mapping. 585/// This is currently only used for _Pragma implementation, so it is the slow 586/// path of the hot getSourceLocation method. Do not allow it to be inlined. 587static DISABLE_INLINE SourceLocation GetMappedTokenLoc(Preprocessor &PP, 588 SourceLocation FileLoc, 589 unsigned CharNo, 590 unsigned TokLen); 591static SourceLocation GetMappedTokenLoc(Preprocessor &PP, 592 SourceLocation FileLoc, 593 unsigned CharNo, unsigned TokLen) { 594 assert(FileLoc.isMacroID() && "Must be an instantiation"); 595 596 // Otherwise, we're lexing "mapped tokens". This is used for things like 597 // _Pragma handling. Combine the instantiation location of FileLoc with the 598 // spelling location. 599 SourceManager &SM = PP.getSourceManager(); 600 601 // Create a new SLoc which is expanded from Instantiation(FileLoc) but whose 602 // characters come from spelling(FileLoc)+Offset. 603 SourceLocation SpellingLoc = SM.getSpellingLoc(FileLoc); 604 SpellingLoc = SpellingLoc.getFileLocWithOffset(CharNo); 605 606 // Figure out the expansion loc range, which is the range covered by the 607 // original _Pragma(...) sequence. 608 std::pair<SourceLocation,SourceLocation> II = 609 SM.getImmediateInstantiationRange(FileLoc); 610 611 return SM.createInstantiationLoc(SpellingLoc, II.first, II.second, TokLen); 612} 613 614/// getSourceLocation - Return a source location identifier for the specified 615/// offset in the current file. 616SourceLocation Lexer::getSourceLocation(const char *Loc, 617 unsigned TokLen) const { 618 assert(Loc >= BufferStart && Loc <= BufferEnd && 619 "Location out of range for this buffer!"); 620 621 // In the normal case, we're just lexing from a simple file buffer, return 622 // the file id from FileLoc with the offset specified. 623 unsigned CharNo = Loc-BufferStart; 624 if (FileLoc.isFileID()) 625 return FileLoc.getFileLocWithOffset(CharNo); 626 627 // Otherwise, this is the _Pragma lexer case, which pretends that all of the 628 // tokens are lexed from where the _Pragma was defined. 629 assert(PP && "This doesn't work on raw lexers"); 630 return GetMappedTokenLoc(*PP, FileLoc, CharNo, TokLen); 631} 632 633/// Diag - Forwarding function for diagnostics. This translate a source 634/// position in the current buffer into a SourceLocation object for rendering. 635DiagnosticBuilder Lexer::Diag(const char *Loc, unsigned DiagID) const { 636 return PP->Diag(getSourceLocation(Loc), DiagID); 637} 638 639//===----------------------------------------------------------------------===// 640// Trigraph and Escaped Newline Handling Code. 641//===----------------------------------------------------------------------===// 642 643/// GetTrigraphCharForLetter - Given a character that occurs after a ?? pair, 644/// return the decoded trigraph letter it corresponds to, or '\0' if nothing. 645static char GetTrigraphCharForLetter(char Letter) { 646 switch (Letter) { 647 default: return 0; 648 case '=': return '#'; 649 case ')': return ']'; 650 case '(': return '['; 651 case '!': return '|'; 652 case '\'': return '^'; 653 case '>': return '}'; 654 case '/': return '\\'; 655 case '<': return '{'; 656 case '-': return '~'; 657 } 658} 659 660/// DecodeTrigraphChar - If the specified character is a legal trigraph when 661/// prefixed with ??, emit a trigraph warning. If trigraphs are enabled, 662/// return the result character. Finally, emit a warning about trigraph use 663/// whether trigraphs are enabled or not. 664static char DecodeTrigraphChar(const char *CP, Lexer *L) { 665 char Res = GetTrigraphCharForLetter(*CP); 666 if (!Res || !L) return Res; 667 668 if (!L->getFeatures().Trigraphs) { 669 if (!L->isLexingRawMode()) 670 L->Diag(CP-2, diag::trigraph_ignored); 671 return 0; 672 } 673 674 if (!L->isLexingRawMode()) 675 L->Diag(CP-2, diag::trigraph_converted) << llvm::StringRef(&Res, 1); 676 return Res; 677} 678 679/// getEscapedNewLineSize - Return the size of the specified escaped newline, 680/// or 0 if it is not an escaped newline. P[-1] is known to be a "\" or a 681/// trigraph equivalent on entry to this function. 682unsigned Lexer::getEscapedNewLineSize(const char *Ptr) { 683 unsigned Size = 0; 684 while (isWhitespace(Ptr[Size])) { 685 ++Size; 686 687 if (Ptr[Size-1] != '\n' && Ptr[Size-1] != '\r') 688 continue; 689 690 // If this is a \r\n or \n\r, skip the other half. 691 if ((Ptr[Size] == '\r' || Ptr[Size] == '\n') && 692 Ptr[Size-1] != Ptr[Size]) 693 ++Size; 694 695 return Size; 696 } 697 698 // Not an escaped newline, must be a \t or something else. 699 return 0; 700} 701 702/// SkipEscapedNewLines - If P points to an escaped newline (or a series of 703/// them), skip over them and return the first non-escaped-newline found, 704/// otherwise return P. 705const char *Lexer::SkipEscapedNewLines(const char *P) { 706 while (1) { 707 const char *AfterEscape; 708 if (*P == '\\') { 709 AfterEscape = P+1; 710 } else if (*P == '?') { 711 // If not a trigraph for escape, bail out. 712 if (P[1] != '?' || P[2] != '/') 713 return P; 714 AfterEscape = P+3; 715 } else { 716 return P; 717 } 718 719 unsigned NewLineSize = Lexer::getEscapedNewLineSize(AfterEscape); 720 if (NewLineSize == 0) return P; 721 P = AfterEscape+NewLineSize; 722 } 723} 724 725 726/// getCharAndSizeSlow - Peek a single 'character' from the specified buffer, 727/// get its size, and return it. This is tricky in several cases: 728/// 1. If currently at the start of a trigraph, we warn about the trigraph, 729/// then either return the trigraph (skipping 3 chars) or the '?', 730/// depending on whether trigraphs are enabled or not. 731/// 2. If this is an escaped newline (potentially with whitespace between 732/// the backslash and newline), implicitly skip the newline and return 733/// the char after it. 734/// 3. If this is a UCN, return it. FIXME: C++ UCN's? 735/// 736/// This handles the slow/uncommon case of the getCharAndSize method. Here we 737/// know that we can accumulate into Size, and that we have already incremented 738/// Ptr by Size bytes. 739/// 740/// NOTE: When this method is updated, getCharAndSizeSlowNoWarn (below) should 741/// be updated to match. 742/// 743char Lexer::getCharAndSizeSlow(const char *Ptr, unsigned &Size, 744 Token *Tok) { 745 // If we have a slash, look for an escaped newline. 746 if (Ptr[0] == '\\') { 747 ++Size; 748 ++Ptr; 749Slash: 750 // Common case, backslash-char where the char is not whitespace. 751 if (!isWhitespace(Ptr[0])) return '\\'; 752 753 // See if we have optional whitespace characters between the slash and 754 // newline. 755 if (unsigned EscapedNewLineSize = getEscapedNewLineSize(Ptr)) { 756 // Remember that this token needs to be cleaned. 757 if (Tok) Tok->setFlag(Token::NeedsCleaning); 758 759 // Warn if there was whitespace between the backslash and newline. 760 if (Ptr[0] != '\n' && Ptr[0] != '\r' && Tok && !isLexingRawMode()) 761 Diag(Ptr, diag::backslash_newline_space); 762 763 // Found backslash<whitespace><newline>. Parse the char after it. 764 Size += EscapedNewLineSize; 765 Ptr += EscapedNewLineSize; 766 // Use slow version to accumulate a correct size field. 767 return getCharAndSizeSlow(Ptr, Size, Tok); 768 } 769 770 // Otherwise, this is not an escaped newline, just return the slash. 771 return '\\'; 772 } 773 774 // If this is a trigraph, process it. 775 if (Ptr[0] == '?' && Ptr[1] == '?') { 776 // If this is actually a legal trigraph (not something like "??x"), emit 777 // a trigraph warning. If so, and if trigraphs are enabled, return it. 778 if (char C = DecodeTrigraphChar(Ptr+2, Tok ? this : 0)) { 779 // Remember that this token needs to be cleaned. 780 if (Tok) Tok->setFlag(Token::NeedsCleaning); 781 782 Ptr += 3; 783 Size += 3; 784 if (C == '\\') goto Slash; 785 return C; 786 } 787 } 788 789 // If this is neither, return a single character. 790 ++Size; 791 return *Ptr; 792} 793 794 795/// getCharAndSizeSlowNoWarn - Handle the slow/uncommon case of the 796/// getCharAndSizeNoWarn method. Here we know that we can accumulate into Size, 797/// and that we have already incremented Ptr by Size bytes. 798/// 799/// NOTE: When this method is updated, getCharAndSizeSlow (above) should 800/// be updated to match. 801char Lexer::getCharAndSizeSlowNoWarn(const char *Ptr, unsigned &Size, 802 const LangOptions &Features) { 803 // If we have a slash, look for an escaped newline. 804 if (Ptr[0] == '\\') { 805 ++Size; 806 ++Ptr; 807Slash: 808 // Common case, backslash-char where the char is not whitespace. 809 if (!isWhitespace(Ptr[0])) return '\\'; 810 811 // See if we have optional whitespace characters followed by a newline. 812 if (unsigned EscapedNewLineSize = getEscapedNewLineSize(Ptr)) { 813 // Found backslash<whitespace><newline>. Parse the char after it. 814 Size += EscapedNewLineSize; 815 Ptr += EscapedNewLineSize; 816 817 // Use slow version to accumulate a correct size field. 818 return getCharAndSizeSlowNoWarn(Ptr, Size, Features); 819 } 820 821 // Otherwise, this is not an escaped newline, just return the slash. 822 return '\\'; 823 } 824 825 // If this is a trigraph, process it. 826 if (Features.Trigraphs && Ptr[0] == '?' && Ptr[1] == '?') { 827 // If this is actually a legal trigraph (not something like "??x"), return 828 // it. 829 if (char C = GetTrigraphCharForLetter(Ptr[2])) { 830 Ptr += 3; 831 Size += 3; 832 if (C == '\\') goto Slash; 833 return C; 834 } 835 } 836 837 // If this is neither, return a single character. 838 ++Size; 839 return *Ptr; 840} 841 842//===----------------------------------------------------------------------===// 843// Helper methods for lexing. 844//===----------------------------------------------------------------------===// 845 846/// \brief Routine that indiscriminately skips bytes in the source file. 847void Lexer::SkipBytes(unsigned Bytes, bool StartOfLine) { 848 BufferPtr += Bytes; 849 if (BufferPtr > BufferEnd) 850 BufferPtr = BufferEnd; 851 IsAtStartOfLine = StartOfLine; 852} 853 854void Lexer::LexIdentifier(Token &Result, const char *CurPtr) { 855 // Match [_A-Za-z0-9]*, we have already matched [_A-Za-z$] 856 unsigned Size; 857 unsigned char C = *CurPtr++; 858 while (isIdentifierBody(C)) 859 C = *CurPtr++; 860 861 --CurPtr; // Back up over the skipped character. 862 863 // Fast path, no $,\,? in identifier found. '\' might be an escaped newline 864 // or UCN, and ? might be a trigraph for '\', an escaped newline or UCN. 865 // FIXME: UCNs. 866 // 867 // TODO: Could merge these checks into a CharInfo flag to make the comparison 868 // cheaper 869 if (C != '\\' && C != '?' && (C != '$' || !Features.DollarIdents)) { 870FinishIdentifier: 871 const char *IdStart = BufferPtr; 872 FormTokenWithChars(Result, CurPtr, tok::identifier); 873 874 // If we are in raw mode, return this identifier raw. There is no need to 875 // look up identifier information or attempt to macro expand it. 876 if (LexingRawMode) return; 877 878 // Fill in Result.IdentifierInfo, looking up the identifier in the 879 // identifier table. 880 IdentifierInfo *II = PP->LookUpIdentifierInfo(Result, IdStart); 881 882 // Change the kind of this identifier to the appropriate token kind, e.g. 883 // turning "for" into a keyword. 884 Result.setKind(II->getTokenID()); 885 886 // Finally, now that we know we have an identifier, pass this off to the 887 // preprocessor, which may macro expand it or something. 888 if (II->isHandleIdentifierCase()) 889 PP->HandleIdentifier(Result); 890 return; 891 } 892 893 // Otherwise, $,\,? in identifier found. Enter slower path. 894 895 C = getCharAndSize(CurPtr, Size); 896 while (1) { 897 if (C == '$') { 898 // If we hit a $ and they are not supported in identifiers, we are done. 899 if (!Features.DollarIdents) goto FinishIdentifier; 900 901 // Otherwise, emit a diagnostic and continue. 902 if (!isLexingRawMode()) 903 Diag(CurPtr, diag::ext_dollar_in_identifier); 904 CurPtr = ConsumeChar(CurPtr, Size, Result); 905 C = getCharAndSize(CurPtr, Size); 906 continue; 907 } else if (!isIdentifierBody(C)) { // FIXME: UCNs. 908 // Found end of identifier. 909 goto FinishIdentifier; 910 } 911 912 // Otherwise, this character is good, consume it. 913 CurPtr = ConsumeChar(CurPtr, Size, Result); 914 915 C = getCharAndSize(CurPtr, Size); 916 while (isIdentifierBody(C)) { // FIXME: UCNs. 917 CurPtr = ConsumeChar(CurPtr, Size, Result); 918 C = getCharAndSize(CurPtr, Size); 919 } 920 } 921} 922 923/// isHexaLiteral - Return true if Start points to a hex constant. 924/// FIXME: This isn't correct, it will mislex: 925/// 0\ <- escaped newline. 926/// x1234e+1 927/// in microsoft mode (where this is supposed to be several different tokens). 928static inline bool isHexaLiteral(const char *Start, const char *End) { 929 return ((End - Start > 2) && Start[0] == '0' && 930 (Start[1] == 'x' || Start[1] == 'X')); 931} 932 933/// LexNumericConstant - Lex the remainder of a integer or floating point 934/// constant. From[-1] is the first character lexed. Return the end of the 935/// constant. 936void Lexer::LexNumericConstant(Token &Result, const char *CurPtr) { 937 unsigned Size; 938 char C = getCharAndSize(CurPtr, Size); 939 char PrevCh = 0; 940 while (isNumberBody(C)) { // FIXME: UCNs? 941 CurPtr = ConsumeChar(CurPtr, Size, Result); 942 PrevCh = C; 943 C = getCharAndSize(CurPtr, Size); 944 } 945 946 // If we fell out, check for a sign, due to 1e+12. If we have one, continue. 947 if ((C == '-' || C == '+') && (PrevCh == 'E' || PrevCh == 'e')) { 948 // If we are in Microsoft mode, don't continue if the constant is hex. 949 // For example, MSVC will accept the following as 3 tokens: 0x1234567e+1 950 if (!Features.Microsoft || !isHexaLiteral(BufferPtr, CurPtr)) 951 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result)); 952 } 953 954 // If we have a hex FP constant, continue. 955 if ((C == '-' || C == '+') && (PrevCh == 'P' || PrevCh == 'p') && 956 !Features.CPlusPlus0x) 957 return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result)); 958 959 // Update the location of token as well as BufferPtr. 960 const char *TokStart = BufferPtr; 961 FormTokenWithChars(Result, CurPtr, tok::numeric_constant); 962 Result.setLiteralData(TokStart); 963} 964 965/// LexStringLiteral - Lex the remainder of a string literal, after having lexed 966/// either " or L". 967void Lexer::LexStringLiteral(Token &Result, const char *CurPtr, bool Wide) { 968 const char *NulCharacter = 0; // Does this string contain the \0 character? 969 970 char C = getAndAdvanceChar(CurPtr, Result); 971 while (C != '"') { 972 // Skip escaped characters. Escaped newlines will already be processed by 973 // getAndAdvanceChar. 974 if (C == '\\') 975 C = getAndAdvanceChar(CurPtr, Result); 976 977 if (C == '\n' || C == '\r' || // Newline. 978 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file. 979 if (C == 0 && PP && PP->isCodeCompletionFile(FileLoc)) 980 PP->CodeCompleteNaturalLanguage(); 981 else if (!isLexingRawMode() && !Features.AsmPreprocessor) 982 Diag(BufferPtr, diag::err_unterminated_string); 983 FormTokenWithChars(Result, CurPtr-1, tok::unknown); 984 return; 985 } 986 987 if (C == 0) 988 NulCharacter = CurPtr-1; 989 C = getAndAdvanceChar(CurPtr, Result); 990 } 991 992 // If a nul character existed in the string, warn about it. 993 if (NulCharacter && !isLexingRawMode()) 994 Diag(NulCharacter, diag::null_in_string); 995 996 // Update the location of the token as well as the BufferPtr instance var. 997 const char *TokStart = BufferPtr; 998 FormTokenWithChars(Result, CurPtr, 999 Wide ? tok::wide_string_literal : tok::string_literal); 1000 Result.setLiteralData(TokStart); 1001} 1002 1003/// LexAngledStringLiteral - Lex the remainder of an angled string literal, 1004/// after having lexed the '<' character. This is used for #include filenames. 1005void Lexer::LexAngledStringLiteral(Token &Result, const char *CurPtr) { 1006 const char *NulCharacter = 0; // Does this string contain the \0 character? 1007 const char *AfterLessPos = CurPtr; 1008 char C = getAndAdvanceChar(CurPtr, Result); 1009 while (C != '>') { 1010 // Skip escaped characters. 1011 if (C == '\\') { 1012 // Skip the escaped character. 1013 C = getAndAdvanceChar(CurPtr, Result); 1014 } else if (C == '\n' || C == '\r' || // Newline. 1015 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file. 1016 // If the filename is unterminated, then it must just be a lone < 1017 // character. Return this as such. 1018 FormTokenWithChars(Result, AfterLessPos, tok::less); 1019 return; 1020 } else if (C == 0) { 1021 NulCharacter = CurPtr-1; 1022 } 1023 C = getAndAdvanceChar(CurPtr, Result); 1024 } 1025 1026 // If a nul character existed in the string, warn about it. 1027 if (NulCharacter && !isLexingRawMode()) 1028 Diag(NulCharacter, diag::null_in_string); 1029 1030 // Update the location of token as well as BufferPtr. 1031 const char *TokStart = BufferPtr; 1032 FormTokenWithChars(Result, CurPtr, tok::angle_string_literal); 1033 Result.setLiteralData(TokStart); 1034} 1035 1036 1037/// LexCharConstant - Lex the remainder of a character constant, after having 1038/// lexed either ' or L'. 1039void Lexer::LexCharConstant(Token &Result, const char *CurPtr) { 1040 const char *NulCharacter = 0; // Does this character contain the \0 character? 1041 1042 char C = getAndAdvanceChar(CurPtr, Result); 1043 if (C == '\'') { 1044 if (!isLexingRawMode() && !Features.AsmPreprocessor) 1045 Diag(BufferPtr, diag::err_empty_character); 1046 FormTokenWithChars(Result, CurPtr, tok::unknown); 1047 return; 1048 } 1049 1050 while (C != '\'') { 1051 // Skip escaped characters. 1052 if (C == '\\') { 1053 // Skip the escaped character. 1054 // FIXME: UCN's 1055 C = getAndAdvanceChar(CurPtr, Result); 1056 } else if (C == '\n' || C == '\r' || // Newline. 1057 (C == 0 && CurPtr-1 == BufferEnd)) { // End of file. 1058 if (C == 0 && PP && PP->isCodeCompletionFile(FileLoc)) 1059 PP->CodeCompleteNaturalLanguage(); 1060 else if (!isLexingRawMode() && !Features.AsmPreprocessor) 1061 Diag(BufferPtr, diag::err_unterminated_char); 1062 FormTokenWithChars(Result, CurPtr-1, tok::unknown); 1063 return; 1064 } else if (C == 0) { 1065 NulCharacter = CurPtr-1; 1066 } 1067 C = getAndAdvanceChar(CurPtr, Result); 1068 } 1069 1070 // If a nul character existed in the character, warn about it. 1071 if (NulCharacter && !isLexingRawMode()) 1072 Diag(NulCharacter, diag::null_in_char); 1073 1074 // Update the location of token as well as BufferPtr. 1075 const char *TokStart = BufferPtr; 1076 FormTokenWithChars(Result, CurPtr, tok::char_constant); 1077 Result.setLiteralData(TokStart); 1078} 1079 1080/// SkipWhitespace - Efficiently skip over a series of whitespace characters. 1081/// Update BufferPtr to point to the next non-whitespace character and return. 1082/// 1083/// This method forms a token and returns true if KeepWhitespaceMode is enabled. 1084/// 1085bool Lexer::SkipWhitespace(Token &Result, const char *CurPtr) { 1086 // Whitespace - Skip it, then return the token after the whitespace. 1087 unsigned char Char = *CurPtr; // Skip consequtive spaces efficiently. 1088 while (1) { 1089 // Skip horizontal whitespace very aggressively. 1090 while (isHorizontalWhitespace(Char)) 1091 Char = *++CurPtr; 1092 1093 // Otherwise if we have something other than whitespace, we're done. 1094 if (Char != '\n' && Char != '\r') 1095 break; 1096 1097 if (ParsingPreprocessorDirective) { 1098 // End of preprocessor directive line, let LexTokenInternal handle this. 1099 BufferPtr = CurPtr; 1100 return false; 1101 } 1102 1103 // ok, but handle newline. 1104 // The returned token is at the start of the line. 1105 Result.setFlag(Token::StartOfLine); 1106 // No leading whitespace seen so far. 1107 Result.clearFlag(Token::LeadingSpace); 1108 Char = *++CurPtr; 1109 } 1110 1111 // If this isn't immediately after a newline, there is leading space. 1112 char PrevChar = CurPtr[-1]; 1113 if (PrevChar != '\n' && PrevChar != '\r') 1114 Result.setFlag(Token::LeadingSpace); 1115 1116 // If the client wants us to return whitespace, return it now. 1117 if (isKeepWhitespaceMode()) { 1118 FormTokenWithChars(Result, CurPtr, tok::unknown); 1119 return true; 1120 } 1121 1122 BufferPtr = CurPtr; 1123 return false; 1124} 1125 1126// SkipBCPLComment - We have just read the // characters from input. Skip until 1127// we find the newline character thats terminate the comment. Then update 1128/// BufferPtr and return. 1129/// 1130/// If we're in KeepCommentMode or any CommentHandler has inserted 1131/// some tokens, this will store the first token and return true. 1132bool Lexer::SkipBCPLComment(Token &Result, const char *CurPtr) { 1133 // If BCPL comments aren't explicitly enabled for this language, emit an 1134 // extension warning. 1135 if (!Features.BCPLComment && !isLexingRawMode()) { 1136 Diag(BufferPtr, diag::ext_bcpl_comment); 1137 1138 // Mark them enabled so we only emit one warning for this translation 1139 // unit. 1140 Features.BCPLComment = true; 1141 } 1142 1143 // Scan over the body of the comment. The common case, when scanning, is that 1144 // the comment contains normal ascii characters with nothing interesting in 1145 // them. As such, optimize for this case with the inner loop. 1146 char C; 1147 do { 1148 C = *CurPtr; 1149 // FIXME: Speedup BCPL comment lexing. Just scan for a \n or \r character. 1150 // If we find a \n character, scan backwards, checking to see if it's an 1151 // escaped newline, like we do for block comments. 1152 1153 // Skip over characters in the fast loop. 1154 while (C != 0 && // Potentially EOF. 1155 C != '\\' && // Potentially escaped newline. 1156 C != '?' && // Potentially trigraph. 1157 C != '\n' && C != '\r') // Newline or DOS-style newline. 1158 C = *++CurPtr; 1159 1160 // If this is a newline, we're done. 1161 if (C == '\n' || C == '\r') 1162 break; // Found the newline? Break out! 1163 1164 // Otherwise, this is a hard case. Fall back on getAndAdvanceChar to 1165 // properly decode the character. Read it in raw mode to avoid emitting 1166 // diagnostics about things like trigraphs. If we see an escaped newline, 1167 // we'll handle it below. 1168 const char *OldPtr = CurPtr; 1169 bool OldRawMode = isLexingRawMode(); 1170 LexingRawMode = true; 1171 C = getAndAdvanceChar(CurPtr, Result); 1172 LexingRawMode = OldRawMode; 1173 1174 // If the char that we finally got was a \n, then we must have had something 1175 // like \<newline><newline>. We don't want to have consumed the second 1176 // newline, we want CurPtr, to end up pointing to it down below. 1177 if (C == '\n' || C == '\r') { 1178 --CurPtr; 1179 C = 'x'; // doesn't matter what this is. 1180 } 1181 1182 // If we read multiple characters, and one of those characters was a \r or 1183 // \n, then we had an escaped newline within the comment. Emit diagnostic 1184 // unless the next line is also a // comment. 1185 if (CurPtr != OldPtr+1 && C != '/' && CurPtr[0] != '/') { 1186 for (; OldPtr != CurPtr; ++OldPtr) 1187 if (OldPtr[0] == '\n' || OldPtr[0] == '\r') { 1188 // Okay, we found a // comment that ends in a newline, if the next 1189 // line is also a // comment, but has spaces, don't emit a diagnostic. 1190 if (isspace(C)) { 1191 const char *ForwardPtr = CurPtr; 1192 while (isspace(*ForwardPtr)) // Skip whitespace. 1193 ++ForwardPtr; 1194 if (ForwardPtr[0] == '/' && ForwardPtr[1] == '/') 1195 break; 1196 } 1197 1198 if (!isLexingRawMode()) 1199 Diag(OldPtr-1, diag::ext_multi_line_bcpl_comment); 1200 break; 1201 } 1202 } 1203 1204 if (CurPtr == BufferEnd+1) { 1205 if (PP && PP->isCodeCompletionFile(FileLoc)) 1206 PP->CodeCompleteNaturalLanguage(); 1207 1208 --CurPtr; 1209 break; 1210 } 1211 } while (C != '\n' && C != '\r'); 1212 1213 // Found but did not consume the newline. Notify comment handlers about the 1214 // comment unless we're in a #if 0 block. 1215 if (PP && !isLexingRawMode() && 1216 PP->HandleComment(Result, SourceRange(getSourceLocation(BufferPtr), 1217 getSourceLocation(CurPtr)))) { 1218 BufferPtr = CurPtr; 1219 return true; // A token has to be returned. 1220 } 1221 1222 // If we are returning comments as tokens, return this comment as a token. 1223 if (inKeepCommentMode()) 1224 return SaveBCPLComment(Result, CurPtr); 1225 1226 // If we are inside a preprocessor directive and we see the end of line, 1227 // return immediately, so that the lexer can return this as an EOM token. 1228 if (ParsingPreprocessorDirective || CurPtr == BufferEnd) { 1229 BufferPtr = CurPtr; 1230 return false; 1231 } 1232 1233 // Otherwise, eat the \n character. We don't care if this is a \n\r or 1234 // \r\n sequence. This is an efficiency hack (because we know the \n can't 1235 // contribute to another token), it isn't needed for correctness. Note that 1236 // this is ok even in KeepWhitespaceMode, because we would have returned the 1237 /// comment above in that mode. 1238 ++CurPtr; 1239 1240 // The next returned token is at the start of the line. 1241 Result.setFlag(Token::StartOfLine); 1242 // No leading whitespace seen so far. 1243 Result.clearFlag(Token::LeadingSpace); 1244 BufferPtr = CurPtr; 1245 return false; 1246} 1247 1248/// SaveBCPLComment - If in save-comment mode, package up this BCPL comment in 1249/// an appropriate way and return it. 1250bool Lexer::SaveBCPLComment(Token &Result, const char *CurPtr) { 1251 // If we're not in a preprocessor directive, just return the // comment 1252 // directly. 1253 FormTokenWithChars(Result, CurPtr, tok::comment); 1254 1255 if (!ParsingPreprocessorDirective) 1256 return true; 1257 1258 // If this BCPL-style comment is in a macro definition, transmogrify it into 1259 // a C-style block comment. 1260 bool Invalid = false; 1261 std::string Spelling = PP->getSpelling(Result, &Invalid); 1262 if (Invalid) 1263 return true; 1264 1265 assert(Spelling[0] == '/' && Spelling[1] == '/' && "Not bcpl comment?"); 1266 Spelling[1] = '*'; // Change prefix to "/*". 1267 Spelling += "*/"; // add suffix. 1268 1269 Result.setKind(tok::comment); 1270 PP->CreateString(&Spelling[0], Spelling.size(), Result, 1271 Result.getLocation()); 1272 return true; 1273} 1274 1275/// isBlockCommentEndOfEscapedNewLine - Return true if the specified newline 1276/// character (either \n or \r) is part of an escaped newline sequence. Issue a 1277/// diagnostic if so. We know that the newline is inside of a block comment. 1278static bool isEndOfBlockCommentWithEscapedNewLine(const char *CurPtr, 1279 Lexer *L) { 1280 assert(CurPtr[0] == '\n' || CurPtr[0] == '\r'); 1281 1282 // Back up off the newline. 1283 --CurPtr; 1284 1285 // If this is a two-character newline sequence, skip the other character. 1286 if (CurPtr[0] == '\n' || CurPtr[0] == '\r') { 1287 // \n\n or \r\r -> not escaped newline. 1288 if (CurPtr[0] == CurPtr[1]) 1289 return false; 1290 // \n\r or \r\n -> skip the newline. 1291 --CurPtr; 1292 } 1293 1294 // If we have horizontal whitespace, skip over it. We allow whitespace 1295 // between the slash and newline. 1296 bool HasSpace = false; 1297 while (isHorizontalWhitespace(*CurPtr) || *CurPtr == 0) { 1298 --CurPtr; 1299 HasSpace = true; 1300 } 1301 1302 // If we have a slash, we know this is an escaped newline. 1303 if (*CurPtr == '\\') { 1304 if (CurPtr[-1] != '*') return false; 1305 } else { 1306 // It isn't a slash, is it the ?? / trigraph? 1307 if (CurPtr[0] != '/' || CurPtr[-1] != '?' || CurPtr[-2] != '?' || 1308 CurPtr[-3] != '*') 1309 return false; 1310 1311 // This is the trigraph ending the comment. Emit a stern warning! 1312 CurPtr -= 2; 1313 1314 // If no trigraphs are enabled, warn that we ignored this trigraph and 1315 // ignore this * character. 1316 if (!L->getFeatures().Trigraphs) { 1317 if (!L->isLexingRawMode()) 1318 L->Diag(CurPtr, diag::trigraph_ignored_block_comment); 1319 return false; 1320 } 1321 if (!L->isLexingRawMode()) 1322 L->Diag(CurPtr, diag::trigraph_ends_block_comment); 1323 } 1324 1325 // Warn about having an escaped newline between the */ characters. 1326 if (!L->isLexingRawMode()) 1327 L->Diag(CurPtr, diag::escaped_newline_block_comment_end); 1328 1329 // If there was space between the backslash and newline, warn about it. 1330 if (HasSpace && !L->isLexingRawMode()) 1331 L->Diag(CurPtr, diag::backslash_newline_space); 1332 1333 return true; 1334} 1335 1336#ifdef __SSE2__ 1337#include <emmintrin.h> 1338#elif __ALTIVEC__ 1339#include <altivec.h> 1340#undef bool 1341#endif 1342 1343/// SkipBlockComment - We have just read the /* characters from input. Read 1344/// until we find the */ characters that terminate the comment. Note that we 1345/// don't bother decoding trigraphs or escaped newlines in block comments, 1346/// because they cannot cause the comment to end. The only thing that can 1347/// happen is the comment could end with an escaped newline between the */ end 1348/// of comment. 1349/// 1350/// If we're in KeepCommentMode or any CommentHandler has inserted 1351/// some tokens, this will store the first token and return true. 1352bool Lexer::SkipBlockComment(Token &Result, const char *CurPtr) { 1353 // Scan one character past where we should, looking for a '/' character. Once 1354 // we find it, check to see if it was preceeded by a *. This common 1355 // optimization helps people who like to put a lot of * characters in their 1356 // comments. 1357 1358 // The first character we get with newlines and trigraphs skipped to handle 1359 // the degenerate /*/ case below correctly if the * has an escaped newline 1360 // after it. 1361 unsigned CharSize; 1362 unsigned char C = getCharAndSize(CurPtr, CharSize); 1363 CurPtr += CharSize; 1364 if (C == 0 && CurPtr == BufferEnd+1) { 1365 if (!isLexingRawMode() && 1366 !PP->isCodeCompletionFile(FileLoc)) 1367 Diag(BufferPtr, diag::err_unterminated_block_comment); 1368 --CurPtr; 1369 1370 // KeepWhitespaceMode should return this broken comment as a token. Since 1371 // it isn't a well formed comment, just return it as an 'unknown' token. 1372 if (isKeepWhitespaceMode()) { 1373 FormTokenWithChars(Result, CurPtr, tok::unknown); 1374 return true; 1375 } 1376 1377 BufferPtr = CurPtr; 1378 return false; 1379 } 1380 1381 // Check to see if the first character after the '/*' is another /. If so, 1382 // then this slash does not end the block comment, it is part of it. 1383 if (C == '/') 1384 C = *CurPtr++; 1385 1386 while (1) { 1387 // Skip over all non-interesting characters until we find end of buffer or a 1388 // (probably ending) '/' character. 1389 if (CurPtr + 24 < BufferEnd) { 1390 // While not aligned to a 16-byte boundary. 1391 while (C != '/' && ((intptr_t)CurPtr & 0x0F) != 0) 1392 C = *CurPtr++; 1393 1394 if (C == '/') goto FoundSlash; 1395 1396#ifdef __SSE2__ 1397 __m128i Slashes = _mm_set_epi8('/', '/', '/', '/', '/', '/', '/', '/', 1398 '/', '/', '/', '/', '/', '/', '/', '/'); 1399 while (CurPtr+16 <= BufferEnd && 1400 _mm_movemask_epi8(_mm_cmpeq_epi8(*(__m128i*)CurPtr, Slashes)) == 0) 1401 CurPtr += 16; 1402#elif __ALTIVEC__ 1403 __vector unsigned char Slashes = { 1404 '/', '/', '/', '/', '/', '/', '/', '/', 1405 '/', '/', '/', '/', '/', '/', '/', '/' 1406 }; 1407 while (CurPtr+16 <= BufferEnd && 1408 !vec_any_eq(*(vector unsigned char*)CurPtr, Slashes)) 1409 CurPtr += 16; 1410#else 1411 // Scan for '/' quickly. Many block comments are very large. 1412 while (CurPtr[0] != '/' && 1413 CurPtr[1] != '/' && 1414 CurPtr[2] != '/' && 1415 CurPtr[3] != '/' && 1416 CurPtr+4 < BufferEnd) { 1417 CurPtr += 4; 1418 } 1419#endif 1420 1421 // It has to be one of the bytes scanned, increment to it and read one. 1422 C = *CurPtr++; 1423 } 1424 1425 // Loop to scan the remainder. 1426 while (C != '/' && C != '\0') 1427 C = *CurPtr++; 1428 1429 FoundSlash: 1430 if (C == '/') { 1431 if (CurPtr[-2] == '*') // We found the final */. We're done! 1432 break; 1433 1434 if ((CurPtr[-2] == '\n' || CurPtr[-2] == '\r')) { 1435 if (isEndOfBlockCommentWithEscapedNewLine(CurPtr-2, this)) { 1436 // We found the final */, though it had an escaped newline between the 1437 // * and /. We're done! 1438 break; 1439 } 1440 } 1441 if (CurPtr[0] == '*' && CurPtr[1] != '/') { 1442 // If this is a /* inside of the comment, emit a warning. Don't do this 1443 // if this is a /*/, which will end the comment. This misses cases with 1444 // embedded escaped newlines, but oh well. 1445 if (!isLexingRawMode()) 1446 Diag(CurPtr-1, diag::warn_nested_block_comment); 1447 } 1448 } else if (C == 0 && CurPtr == BufferEnd+1) { 1449 if (PP && PP->isCodeCompletionFile(FileLoc)) 1450 PP->CodeCompleteNaturalLanguage(); 1451 else if (!isLexingRawMode()) 1452 Diag(BufferPtr, diag::err_unterminated_block_comment); 1453 // Note: the user probably forgot a */. We could continue immediately 1454 // after the /*, but this would involve lexing a lot of what really is the 1455 // comment, which surely would confuse the parser. 1456 --CurPtr; 1457 1458 // KeepWhitespaceMode should return this broken comment as a token. Since 1459 // it isn't a well formed comment, just return it as an 'unknown' token. 1460 if (isKeepWhitespaceMode()) { 1461 FormTokenWithChars(Result, CurPtr, tok::unknown); 1462 return true; 1463 } 1464 1465 BufferPtr = CurPtr; 1466 return false; 1467 } 1468 C = *CurPtr++; 1469 } 1470 1471 // Notify comment handlers about the comment unless we're in a #if 0 block. 1472 if (PP && !isLexingRawMode() && 1473 PP->HandleComment(Result, SourceRange(getSourceLocation(BufferPtr), 1474 getSourceLocation(CurPtr)))) { 1475 BufferPtr = CurPtr; 1476 return true; // A token has to be returned. 1477 } 1478 1479 // If we are returning comments as tokens, return this comment as a token. 1480 if (inKeepCommentMode()) { 1481 FormTokenWithChars(Result, CurPtr, tok::comment); 1482 return true; 1483 } 1484 1485 // It is common for the tokens immediately after a /**/ comment to be 1486 // whitespace. Instead of going through the big switch, handle it 1487 // efficiently now. This is safe even in KeepWhitespaceMode because we would 1488 // have already returned above with the comment as a token. 1489 if (isHorizontalWhitespace(*CurPtr)) { 1490 Result.setFlag(Token::LeadingSpace); 1491 SkipWhitespace(Result, CurPtr+1); 1492 return false; 1493 } 1494 1495 // Otherwise, just return so that the next character will be lexed as a token. 1496 BufferPtr = CurPtr; 1497 Result.setFlag(Token::LeadingSpace); 1498 return false; 1499} 1500 1501//===----------------------------------------------------------------------===// 1502// Primary Lexing Entry Points 1503//===----------------------------------------------------------------------===// 1504 1505/// ReadToEndOfLine - Read the rest of the current preprocessor line as an 1506/// uninterpreted string. This switches the lexer out of directive mode. 1507std::string Lexer::ReadToEndOfLine() { 1508 assert(ParsingPreprocessorDirective && ParsingFilename == false && 1509 "Must be in a preprocessing directive!"); 1510 std::string Result; 1511 Token Tmp; 1512 1513 // CurPtr - Cache BufferPtr in an automatic variable. 1514 const char *CurPtr = BufferPtr; 1515 while (1) { 1516 char Char = getAndAdvanceChar(CurPtr, Tmp); 1517 switch (Char) { 1518 default: 1519 Result += Char; 1520 break; 1521 case 0: // Null. 1522 // Found end of file? 1523 if (CurPtr-1 != BufferEnd) { 1524 // Nope, normal character, continue. 1525 Result += Char; 1526 break; 1527 } 1528 // FALL THROUGH. 1529 case '\r': 1530 case '\n': 1531 // Okay, we found the end of the line. First, back up past the \0, \r, \n. 1532 assert(CurPtr[-1] == Char && "Trigraphs for newline?"); 1533 BufferPtr = CurPtr-1; 1534 1535 // Next, lex the character, which should handle the EOM transition. 1536 Lex(Tmp); 1537 if (Tmp.is(tok::code_completion)) { 1538 if (PP && PP->getCodeCompletionHandler()) 1539 PP->getCodeCompletionHandler()->CodeCompleteNaturalLanguage(); 1540 Lex(Tmp); 1541 } 1542 assert(Tmp.is(tok::eom) && "Unexpected token!"); 1543 1544 // Finally, we're done, return the string we found. 1545 return Result; 1546 } 1547 } 1548} 1549 1550/// LexEndOfFile - CurPtr points to the end of this file. Handle this 1551/// condition, reporting diagnostics and handling other edge cases as required. 1552/// This returns true if Result contains a token, false if PP.Lex should be 1553/// called again. 1554bool Lexer::LexEndOfFile(Token &Result, const char *CurPtr) { 1555 // Check if we are performing code completion. 1556 if (PP && PP->isCodeCompletionFile(FileLoc)) { 1557 // We're at the end of the file, but we've been asked to consider the 1558 // end of the file to be a code-completion token. Return the 1559 // code-completion token. 1560 Result.startToken(); 1561 FormTokenWithChars(Result, CurPtr, tok::code_completion); 1562 1563 // Only do the eof -> code_completion translation once. 1564 PP->SetCodeCompletionPoint(0, 0, 0); 1565 1566 // Silence any diagnostics that occur once we hit the code-completion point. 1567 PP->getDiagnostics().setSuppressAllDiagnostics(true); 1568 return true; 1569 } 1570 1571 // If we hit the end of the file while parsing a preprocessor directive, 1572 // end the preprocessor directive first. The next token returned will 1573 // then be the end of file. 1574 if (ParsingPreprocessorDirective) { 1575 // Done parsing the "line". 1576 ParsingPreprocessorDirective = false; 1577 // Update the location of token as well as BufferPtr. 1578 FormTokenWithChars(Result, CurPtr, tok::eom); 1579 1580 // Restore comment saving mode, in case it was disabled for directive. 1581 SetCommentRetentionState(PP->getCommentRetentionState()); 1582 return true; // Have a token. 1583 } 1584 1585 // If we are in raw mode, return this event as an EOF token. Let the caller 1586 // that put us in raw mode handle the event. 1587 if (isLexingRawMode()) { 1588 Result.startToken(); 1589 BufferPtr = BufferEnd; 1590 FormTokenWithChars(Result, BufferEnd, tok::eof); 1591 return true; 1592 } 1593 1594 // Issue diagnostics for unterminated #if and missing newline. 1595 1596 // If we are in a #if directive, emit an error. 1597 while (!ConditionalStack.empty()) { 1598 if (!PP->isCodeCompletionFile(FileLoc)) 1599 PP->Diag(ConditionalStack.back().IfLoc, 1600 diag::err_pp_unterminated_conditional); 1601 ConditionalStack.pop_back(); 1602 } 1603 1604 // C99 5.1.1.2p2: If the file is non-empty and didn't end in a newline, issue 1605 // a pedwarn. 1606 if (CurPtr != BufferStart && (CurPtr[-1] != '\n' && CurPtr[-1] != '\r')) 1607 Diag(BufferEnd, diag::ext_no_newline_eof) 1608 << FixItHint::CreateInsertion(getSourceLocation(BufferEnd), "\n"); 1609 1610 BufferPtr = CurPtr; 1611 1612 // Finally, let the preprocessor handle this. 1613 return PP->HandleEndOfFile(Result); 1614} 1615 1616/// isNextPPTokenLParen - Return 1 if the next unexpanded token lexed from 1617/// the specified lexer will return a tok::l_paren token, 0 if it is something 1618/// else and 2 if there are no more tokens in the buffer controlled by the 1619/// lexer. 1620unsigned Lexer::isNextPPTokenLParen() { 1621 assert(!LexingRawMode && "How can we expand a macro from a skipping buffer?"); 1622 1623 // Switch to 'skipping' mode. This will ensure that we can lex a token 1624 // without emitting diagnostics, disables macro expansion, and will cause EOF 1625 // to return an EOF token instead of popping the include stack. 1626 LexingRawMode = true; 1627 1628 // Save state that can be changed while lexing so that we can restore it. 1629 const char *TmpBufferPtr = BufferPtr; 1630 bool inPPDirectiveMode = ParsingPreprocessorDirective; 1631 1632 Token Tok; 1633 Tok.startToken(); 1634 LexTokenInternal(Tok); 1635 1636 // Restore state that may have changed. 1637 BufferPtr = TmpBufferPtr; 1638 ParsingPreprocessorDirective = inPPDirectiveMode; 1639 1640 // Restore the lexer back to non-skipping mode. 1641 LexingRawMode = false; 1642 1643 if (Tok.is(tok::eof)) 1644 return 2; 1645 return Tok.is(tok::l_paren); 1646} 1647 1648/// FindConflictEnd - Find the end of a version control conflict marker. 1649static const char *FindConflictEnd(const char *CurPtr, const char *BufferEnd) { 1650 llvm::StringRef RestOfBuffer(CurPtr+7, BufferEnd-CurPtr-7); 1651 size_t Pos = RestOfBuffer.find(">>>>>>>"); 1652 while (Pos != llvm::StringRef::npos) { 1653 // Must occur at start of line. 1654 if (RestOfBuffer[Pos-1] != '\r' && 1655 RestOfBuffer[Pos-1] != '\n') { 1656 RestOfBuffer = RestOfBuffer.substr(Pos+7); 1657 Pos = RestOfBuffer.find(">>>>>>>"); 1658 continue; 1659 } 1660 return RestOfBuffer.data()+Pos; 1661 } 1662 return 0; 1663} 1664 1665/// IsStartOfConflictMarker - If the specified pointer is the start of a version 1666/// control conflict marker like '<<<<<<<', recognize it as such, emit an error 1667/// and recover nicely. This returns true if it is a conflict marker and false 1668/// if not. 1669bool Lexer::IsStartOfConflictMarker(const char *CurPtr) { 1670 // Only a conflict marker if it starts at the beginning of a line. 1671 if (CurPtr != BufferStart && 1672 CurPtr[-1] != '\n' && CurPtr[-1] != '\r') 1673 return false; 1674 1675 // Check to see if we have <<<<<<<. 1676 if (BufferEnd-CurPtr < 8 || 1677 llvm::StringRef(CurPtr, 7) != "<<<<<<<") 1678 return false; 1679 1680 // If we have a situation where we don't care about conflict markers, ignore 1681 // it. 1682 if (IsInConflictMarker || isLexingRawMode()) 1683 return false; 1684 1685 // Check to see if there is a >>>>>>> somewhere in the buffer at the start of 1686 // a line to terminate this conflict marker. 1687 if (FindConflictEnd(CurPtr, BufferEnd)) { 1688 // We found a match. We are really in a conflict marker. 1689 // Diagnose this, and ignore to the end of line. 1690 Diag(CurPtr, diag::err_conflict_marker); 1691 IsInConflictMarker = true; 1692 1693 // Skip ahead to the end of line. We know this exists because the 1694 // end-of-conflict marker starts with \r or \n. 1695 while (*CurPtr != '\r' && *CurPtr != '\n') { 1696 assert(CurPtr != BufferEnd && "Didn't find end of line"); 1697 ++CurPtr; 1698 } 1699 BufferPtr = CurPtr; 1700 return true; 1701 } 1702 1703 // No end of conflict marker found. 1704 return false; 1705} 1706 1707 1708/// HandleEndOfConflictMarker - If this is a '=======' or '|||||||' or '>>>>>>>' 1709/// marker, then it is the end of a conflict marker. Handle it by ignoring up 1710/// until the end of the line. This returns true if it is a conflict marker and 1711/// false if not. 1712bool Lexer::HandleEndOfConflictMarker(const char *CurPtr) { 1713 // Only a conflict marker if it starts at the beginning of a line. 1714 if (CurPtr != BufferStart && 1715 CurPtr[-1] != '\n' && CurPtr[-1] != '\r') 1716 return false; 1717 1718 // If we have a situation where we don't care about conflict markers, ignore 1719 // it. 1720 if (!IsInConflictMarker || isLexingRawMode()) 1721 return false; 1722 1723 // Check to see if we have the marker (7 characters in a row). 1724 for (unsigned i = 1; i != 7; ++i) 1725 if (CurPtr[i] != CurPtr[0]) 1726 return false; 1727 1728 // If we do have it, search for the end of the conflict marker. This could 1729 // fail if it got skipped with a '#if 0' or something. Note that CurPtr might 1730 // be the end of conflict marker. 1731 if (const char *End = FindConflictEnd(CurPtr, BufferEnd)) { 1732 CurPtr = End; 1733 1734 // Skip ahead to the end of line. 1735 while (CurPtr != BufferEnd && *CurPtr != '\r' && *CurPtr != '\n') 1736 ++CurPtr; 1737 1738 BufferPtr = CurPtr; 1739 1740 // No longer in the conflict marker. 1741 IsInConflictMarker = false; 1742 return true; 1743 } 1744 1745 return false; 1746} 1747 1748 1749/// LexTokenInternal - This implements a simple C family lexer. It is an 1750/// extremely performance critical piece of code. This assumes that the buffer 1751/// has a null character at the end of the file. This returns a preprocessing 1752/// token, not a normal token, as such, it is an internal interface. It assumes 1753/// that the Flags of result have been cleared before calling this. 1754void Lexer::LexTokenInternal(Token &Result) { 1755LexNextToken: 1756 // New token, can't need cleaning yet. 1757 Result.clearFlag(Token::NeedsCleaning); 1758 Result.setIdentifierInfo(0); 1759 1760 // CurPtr - Cache BufferPtr in an automatic variable. 1761 const char *CurPtr = BufferPtr; 1762 1763 // Small amounts of horizontal whitespace is very common between tokens. 1764 if ((*CurPtr == ' ') || (*CurPtr == '\t')) { 1765 ++CurPtr; 1766 while ((*CurPtr == ' ') || (*CurPtr == '\t')) 1767 ++CurPtr; 1768 1769 // If we are keeping whitespace and other tokens, just return what we just 1770 // skipped. The next lexer invocation will return the token after the 1771 // whitespace. 1772 if (isKeepWhitespaceMode()) { 1773 FormTokenWithChars(Result, CurPtr, tok::unknown); 1774 return; 1775 } 1776 1777 BufferPtr = CurPtr; 1778 Result.setFlag(Token::LeadingSpace); 1779 } 1780 1781 unsigned SizeTmp, SizeTmp2; // Temporaries for use in cases below. 1782 1783 // Read a character, advancing over it. 1784 char Char = getAndAdvanceChar(CurPtr, Result); 1785 tok::TokenKind Kind; 1786 1787 switch (Char) { 1788 case 0: // Null. 1789 // Found end of file? 1790 if (CurPtr-1 == BufferEnd) { 1791 // Read the PP instance variable into an automatic variable, because 1792 // LexEndOfFile will often delete 'this'. 1793 Preprocessor *PPCache = PP; 1794 if (LexEndOfFile(Result, CurPtr-1)) // Retreat back into the file. 1795 return; // Got a token to return. 1796 assert(PPCache && "Raw buffer::LexEndOfFile should return a token"); 1797 return PPCache->Lex(Result); 1798 } 1799 1800 if (!isLexingRawMode()) 1801 Diag(CurPtr-1, diag::null_in_file); 1802 Result.setFlag(Token::LeadingSpace); 1803 if (SkipWhitespace(Result, CurPtr)) 1804 return; // KeepWhitespaceMode 1805 1806 goto LexNextToken; // GCC isn't tail call eliminating. 1807 1808 case 26: // DOS & CP/M EOF: "^Z". 1809 // If we're in Microsoft extensions mode, treat this as end of file. 1810 if (Features.Microsoft) { 1811 // Read the PP instance variable into an automatic variable, because 1812 // LexEndOfFile will often delete 'this'. 1813 Preprocessor *PPCache = PP; 1814 if (LexEndOfFile(Result, CurPtr-1)) // Retreat back into the file. 1815 return; // Got a token to return. 1816 assert(PPCache && "Raw buffer::LexEndOfFile should return a token"); 1817 return PPCache->Lex(Result); 1818 } 1819 // If Microsoft extensions are disabled, this is just random garbage. 1820 Kind = tok::unknown; 1821 break; 1822 1823 case '\n': 1824 case '\r': 1825 // If we are inside a preprocessor directive and we see the end of line, 1826 // we know we are done with the directive, so return an EOM token. 1827 if (ParsingPreprocessorDirective) { 1828 // Done parsing the "line". 1829 ParsingPreprocessorDirective = false; 1830 1831 // Restore comment saving mode, in case it was disabled for directive. 1832 SetCommentRetentionState(PP->getCommentRetentionState()); 1833 1834 // Since we consumed a newline, we are back at the start of a line. 1835 IsAtStartOfLine = true; 1836 1837 Kind = tok::eom; 1838 break; 1839 } 1840 // The returned token is at the start of the line. 1841 Result.setFlag(Token::StartOfLine); 1842 // No leading whitespace seen so far. 1843 Result.clearFlag(Token::LeadingSpace); 1844 1845 if (SkipWhitespace(Result, CurPtr)) 1846 return; // KeepWhitespaceMode 1847 goto LexNextToken; // GCC isn't tail call eliminating. 1848 case ' ': 1849 case '\t': 1850 case '\f': 1851 case '\v': 1852 SkipHorizontalWhitespace: 1853 Result.setFlag(Token::LeadingSpace); 1854 if (SkipWhitespace(Result, CurPtr)) 1855 return; // KeepWhitespaceMode 1856 1857 SkipIgnoredUnits: 1858 CurPtr = BufferPtr; 1859 1860 // If the next token is obviously a // or /* */ comment, skip it efficiently 1861 // too (without going through the big switch stmt). 1862 if (CurPtr[0] == '/' && CurPtr[1] == '/' && !inKeepCommentMode() && 1863 Features.BCPLComment) { 1864 if (SkipBCPLComment(Result, CurPtr+2)) 1865 return; // There is a token to return. 1866 goto SkipIgnoredUnits; 1867 } else if (CurPtr[0] == '/' && CurPtr[1] == '*' && !inKeepCommentMode()) { 1868 if (SkipBlockComment(Result, CurPtr+2)) 1869 return; // There is a token to return. 1870 goto SkipIgnoredUnits; 1871 } else if (isHorizontalWhitespace(*CurPtr)) { 1872 goto SkipHorizontalWhitespace; 1873 } 1874 goto LexNextToken; // GCC isn't tail call eliminating. 1875 1876 // C99 6.4.4.1: Integer Constants. 1877 // C99 6.4.4.2: Floating Constants. 1878 case '0': case '1': case '2': case '3': case '4': 1879 case '5': case '6': case '7': case '8': case '9': 1880 // Notify MIOpt that we read a non-whitespace/non-comment token. 1881 MIOpt.ReadToken(); 1882 return LexNumericConstant(Result, CurPtr); 1883 1884 case 'L': // Identifier (Loony) or wide literal (L'x' or L"xyz"). 1885 // Notify MIOpt that we read a non-whitespace/non-comment token. 1886 MIOpt.ReadToken(); 1887 Char = getCharAndSize(CurPtr, SizeTmp); 1888 1889 // Wide string literal. 1890 if (Char == '"') 1891 return LexStringLiteral(Result, ConsumeChar(CurPtr, SizeTmp, Result), 1892 true); 1893 1894 // Wide character constant. 1895 if (Char == '\'') 1896 return LexCharConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result)); 1897 // FALL THROUGH, treating L like the start of an identifier. 1898 1899 // C99 6.4.2: Identifiers. 1900 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G': 1901 case 'H': case 'I': case 'J': case 'K': /*'L'*/case 'M': case 'N': 1902 case 'O': case 'P': case 'Q': case 'R': case 'S': case 'T': case 'U': 1903 case 'V': case 'W': case 'X': case 'Y': case 'Z': 1904 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g': 1905 case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n': 1906 case 'o': case 'p': case 'q': case 'r': case 's': case 't': case 'u': 1907 case 'v': case 'w': case 'x': case 'y': case 'z': 1908 case '_': 1909 // Notify MIOpt that we read a non-whitespace/non-comment token. 1910 MIOpt.ReadToken(); 1911 return LexIdentifier(Result, CurPtr); 1912 1913 case '$': // $ in identifiers. 1914 if (Features.DollarIdents) { 1915 if (!isLexingRawMode()) 1916 Diag(CurPtr-1, diag::ext_dollar_in_identifier); 1917 // Notify MIOpt that we read a non-whitespace/non-comment token. 1918 MIOpt.ReadToken(); 1919 return LexIdentifier(Result, CurPtr); 1920 } 1921 1922 Kind = tok::unknown; 1923 break; 1924 1925 // C99 6.4.4: Character Constants. 1926 case '\'': 1927 // Notify MIOpt that we read a non-whitespace/non-comment token. 1928 MIOpt.ReadToken(); 1929 return LexCharConstant(Result, CurPtr); 1930 1931 // C99 6.4.5: String Literals. 1932 case '"': 1933 // Notify MIOpt that we read a non-whitespace/non-comment token. 1934 MIOpt.ReadToken(); 1935 return LexStringLiteral(Result, CurPtr, false); 1936 1937 // C99 6.4.6: Punctuators. 1938 case '?': 1939 Kind = tok::question; 1940 break; 1941 case '[': 1942 Kind = tok::l_square; 1943 break; 1944 case ']': 1945 Kind = tok::r_square; 1946 break; 1947 case '(': 1948 Kind = tok::l_paren; 1949 break; 1950 case ')': 1951 Kind = tok::r_paren; 1952 break; 1953 case '{': 1954 Kind = tok::l_brace; 1955 break; 1956 case '}': 1957 Kind = tok::r_brace; 1958 break; 1959 case '.': 1960 Char = getCharAndSize(CurPtr, SizeTmp); 1961 if (Char >= '0' && Char <= '9') { 1962 // Notify MIOpt that we read a non-whitespace/non-comment token. 1963 MIOpt.ReadToken(); 1964 1965 return LexNumericConstant(Result, ConsumeChar(CurPtr, SizeTmp, Result)); 1966 } else if (Features.CPlusPlus && Char == '*') { 1967 Kind = tok::periodstar; 1968 CurPtr += SizeTmp; 1969 } else if (Char == '.' && 1970 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '.') { 1971 Kind = tok::ellipsis; 1972 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result), 1973 SizeTmp2, Result); 1974 } else { 1975 Kind = tok::period; 1976 } 1977 break; 1978 case '&': 1979 Char = getCharAndSize(CurPtr, SizeTmp); 1980 if (Char == '&') { 1981 Kind = tok::ampamp; 1982 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 1983 } else if (Char == '=') { 1984 Kind = tok::ampequal; 1985 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 1986 } else { 1987 Kind = tok::amp; 1988 } 1989 break; 1990 case '*': 1991 if (getCharAndSize(CurPtr, SizeTmp) == '=') { 1992 Kind = tok::starequal; 1993 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 1994 } else { 1995 Kind = tok::star; 1996 } 1997 break; 1998 case '+': 1999 Char = getCharAndSize(CurPtr, SizeTmp); 2000 if (Char == '+') { 2001 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2002 Kind = tok::plusplus; 2003 } else if (Char == '=') { 2004 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2005 Kind = tok::plusequal; 2006 } else { 2007 Kind = tok::plus; 2008 } 2009 break; 2010 case '-': 2011 Char = getCharAndSize(CurPtr, SizeTmp); 2012 if (Char == '-') { // -- 2013 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2014 Kind = tok::minusminus; 2015 } else if (Char == '>' && Features.CPlusPlus && 2016 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == '*') { // C++ ->* 2017 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result), 2018 SizeTmp2, Result); 2019 Kind = tok::arrowstar; 2020 } else if (Char == '>') { // -> 2021 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2022 Kind = tok::arrow; 2023 } else if (Char == '=') { // -= 2024 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2025 Kind = tok::minusequal; 2026 } else { 2027 Kind = tok::minus; 2028 } 2029 break; 2030 case '~': 2031 Kind = tok::tilde; 2032 break; 2033 case '!': 2034 if (getCharAndSize(CurPtr, SizeTmp) == '=') { 2035 Kind = tok::exclaimequal; 2036 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2037 } else { 2038 Kind = tok::exclaim; 2039 } 2040 break; 2041 case '/': 2042 // 6.4.9: Comments 2043 Char = getCharAndSize(CurPtr, SizeTmp); 2044 if (Char == '/') { // BCPL comment. 2045 // Even if BCPL comments are disabled (e.g. in C89 mode), we generally 2046 // want to lex this as a comment. There is one problem with this though, 2047 // that in one particular corner case, this can change the behavior of the 2048 // resultant program. For example, In "foo //**/ bar", C89 would lex 2049 // this as "foo / bar" and langauges with BCPL comments would lex it as 2050 // "foo". Check to see if the character after the second slash is a '*'. 2051 // If so, we will lex that as a "/" instead of the start of a comment. 2052 if (Features.BCPLComment || 2053 getCharAndSize(CurPtr+SizeTmp, SizeTmp2) != '*') { 2054 if (SkipBCPLComment(Result, ConsumeChar(CurPtr, SizeTmp, Result))) 2055 return; // There is a token to return. 2056 2057 // It is common for the tokens immediately after a // comment to be 2058 // whitespace (indentation for the next line). Instead of going through 2059 // the big switch, handle it efficiently now. 2060 goto SkipIgnoredUnits; 2061 } 2062 } 2063 2064 if (Char == '*') { // /**/ comment. 2065 if (SkipBlockComment(Result, ConsumeChar(CurPtr, SizeTmp, Result))) 2066 return; // There is a token to return. 2067 goto LexNextToken; // GCC isn't tail call eliminating. 2068 } 2069 2070 if (Char == '=') { 2071 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2072 Kind = tok::slashequal; 2073 } else { 2074 Kind = tok::slash; 2075 } 2076 break; 2077 case '%': 2078 Char = getCharAndSize(CurPtr, SizeTmp); 2079 if (Char == '=') { 2080 Kind = tok::percentequal; 2081 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2082 } else if (Features.Digraphs && Char == '>') { 2083 Kind = tok::r_brace; // '%>' -> '}' 2084 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2085 } else if (Features.Digraphs && Char == ':') { 2086 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2087 Char = getCharAndSize(CurPtr, SizeTmp); 2088 if (Char == '%' && getCharAndSize(CurPtr+SizeTmp, SizeTmp2) == ':') { 2089 Kind = tok::hashhash; // '%:%:' -> '##' 2090 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result), 2091 SizeTmp2, Result); 2092 } else if (Char == '@' && Features.Microsoft) { // %:@ -> #@ -> Charize 2093 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2094 if (!isLexingRawMode()) 2095 Diag(BufferPtr, diag::charize_microsoft_ext); 2096 Kind = tok::hashat; 2097 } else { // '%:' -> '#' 2098 // We parsed a # character. If this occurs at the start of the line, 2099 // it's actually the start of a preprocessing directive. Callback to 2100 // the preprocessor to handle it. 2101 // FIXME: -fpreprocessed mode?? 2102 if (Result.isAtStartOfLine() && !LexingRawMode && !Is_PragmaLexer) { 2103 FormTokenWithChars(Result, CurPtr, tok::hash); 2104 PP->HandleDirective(Result); 2105 2106 // As an optimization, if the preprocessor didn't switch lexers, tail 2107 // recurse. 2108 if (PP->isCurrentLexer(this)) { 2109 // Start a new token. If this is a #include or something, the PP may 2110 // want us starting at the beginning of the line again. If so, set 2111 // the StartOfLine flag and clear LeadingSpace. 2112 if (IsAtStartOfLine) { 2113 Result.setFlag(Token::StartOfLine); 2114 Result.clearFlag(Token::LeadingSpace); 2115 IsAtStartOfLine = false; 2116 } 2117 goto LexNextToken; // GCC isn't tail call eliminating. 2118 } 2119 2120 return PP->Lex(Result); 2121 } 2122 2123 Kind = tok::hash; 2124 } 2125 } else { 2126 Kind = tok::percent; 2127 } 2128 break; 2129 case '<': 2130 Char = getCharAndSize(CurPtr, SizeTmp); 2131 if (ParsingFilename) { 2132 return LexAngledStringLiteral(Result, CurPtr); 2133 } else if (Char == '<') { 2134 char After = getCharAndSize(CurPtr+SizeTmp, SizeTmp2); 2135 if (After == '=') { 2136 Kind = tok::lesslessequal; 2137 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result), 2138 SizeTmp2, Result); 2139 } else if (After == '<' && IsStartOfConflictMarker(CurPtr-1)) { 2140 // If this is actually a '<<<<<<<' version control conflict marker, 2141 // recognize it as such and recover nicely. 2142 goto LexNextToken; 2143 } else { 2144 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2145 Kind = tok::lessless; 2146 } 2147 } else if (Char == '=') { 2148 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2149 Kind = tok::lessequal; 2150 } else if (Features.Digraphs && Char == ':') { // '<:' -> '[' 2151 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2152 Kind = tok::l_square; 2153 } else if (Features.Digraphs && Char == '%') { // '<%' -> '{' 2154 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2155 Kind = tok::l_brace; 2156 } else { 2157 Kind = tok::less; 2158 } 2159 break; 2160 case '>': 2161 Char = getCharAndSize(CurPtr, SizeTmp); 2162 if (Char == '=') { 2163 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2164 Kind = tok::greaterequal; 2165 } else if (Char == '>') { 2166 char After = getCharAndSize(CurPtr+SizeTmp, SizeTmp2); 2167 if (After == '=') { 2168 CurPtr = ConsumeChar(ConsumeChar(CurPtr, SizeTmp, Result), 2169 SizeTmp2, Result); 2170 Kind = tok::greatergreaterequal; 2171 } else if (After == '>' && HandleEndOfConflictMarker(CurPtr-1)) { 2172 // If this is '>>>>>>>' and we're in a conflict marker, ignore it. 2173 goto LexNextToken; 2174 } else { 2175 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2176 Kind = tok::greatergreater; 2177 } 2178 2179 } else { 2180 Kind = tok::greater; 2181 } 2182 break; 2183 case '^': 2184 Char = getCharAndSize(CurPtr, SizeTmp); 2185 if (Char == '=') { 2186 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2187 Kind = tok::caretequal; 2188 } else { 2189 Kind = tok::caret; 2190 } 2191 break; 2192 case '|': 2193 Char = getCharAndSize(CurPtr, SizeTmp); 2194 if (Char == '=') { 2195 Kind = tok::pipeequal; 2196 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2197 } else if (Char == '|') { 2198 // If this is '|||||||' and we're in a conflict marker, ignore it. 2199 if (CurPtr[1] == '|' && HandleEndOfConflictMarker(CurPtr-1)) 2200 goto LexNextToken; 2201 Kind = tok::pipepipe; 2202 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2203 } else { 2204 Kind = tok::pipe; 2205 } 2206 break; 2207 case ':': 2208 Char = getCharAndSize(CurPtr, SizeTmp); 2209 if (Features.Digraphs && Char == '>') { 2210 Kind = tok::r_square; // ':>' -> ']' 2211 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2212 } else if (Features.CPlusPlus && Char == ':') { 2213 Kind = tok::coloncolon; 2214 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2215 } else { 2216 Kind = tok::colon; 2217 } 2218 break; 2219 case ';': 2220 Kind = tok::semi; 2221 break; 2222 case '=': 2223 Char = getCharAndSize(CurPtr, SizeTmp); 2224 if (Char == '=') { 2225 // If this is '=======' and we're in a conflict marker, ignore it. 2226 if (CurPtr[1] == '=' && HandleEndOfConflictMarker(CurPtr-1)) 2227 goto LexNextToken; 2228 2229 Kind = tok::equalequal; 2230 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2231 } else { 2232 Kind = tok::equal; 2233 } 2234 break; 2235 case ',': 2236 Kind = tok::comma; 2237 break; 2238 case '#': 2239 Char = getCharAndSize(CurPtr, SizeTmp); 2240 if (Char == '#') { 2241 Kind = tok::hashhash; 2242 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2243 } else if (Char == '@' && Features.Microsoft) { // #@ -> Charize 2244 Kind = tok::hashat; 2245 if (!isLexingRawMode()) 2246 Diag(BufferPtr, diag::charize_microsoft_ext); 2247 CurPtr = ConsumeChar(CurPtr, SizeTmp, Result); 2248 } else { 2249 // We parsed a # character. If this occurs at the start of the line, 2250 // it's actually the start of a preprocessing directive. Callback to 2251 // the preprocessor to handle it. 2252 // FIXME: -fpreprocessed mode?? 2253 if (Result.isAtStartOfLine() && !LexingRawMode && !Is_PragmaLexer) { 2254 FormTokenWithChars(Result, CurPtr, tok::hash); 2255 PP->HandleDirective(Result); 2256 2257 // As an optimization, if the preprocessor didn't switch lexers, tail 2258 // recurse. 2259 if (PP->isCurrentLexer(this)) { 2260 // Start a new token. If this is a #include or something, the PP may 2261 // want us starting at the beginning of the line again. If so, set 2262 // the StartOfLine flag and clear LeadingSpace. 2263 if (IsAtStartOfLine) { 2264 Result.setFlag(Token::StartOfLine); 2265 Result.clearFlag(Token::LeadingSpace); 2266 IsAtStartOfLine = false; 2267 } 2268 goto LexNextToken; // GCC isn't tail call eliminating. 2269 } 2270 return PP->Lex(Result); 2271 } 2272 2273 Kind = tok::hash; 2274 } 2275 break; 2276 2277 case '@': 2278 // Objective C support. 2279 if (CurPtr[-1] == '@' && Features.ObjC1) 2280 Kind = tok::at; 2281 else 2282 Kind = tok::unknown; 2283 break; 2284 2285 case '\\': 2286 // FIXME: UCN's. 2287 // FALL THROUGH. 2288 default: 2289 Kind = tok::unknown; 2290 break; 2291 } 2292 2293 // Notify MIOpt that we read a non-whitespace/non-comment token. 2294 MIOpt.ReadToken(); 2295 2296 // Update the location of token as well as BufferPtr. 2297 FormTokenWithChars(Result, CurPtr, Kind); 2298} 2299