1 //===--- LiteralSupport.cpp - Code to parse and process literals ----------===// 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 NumericLiteralParser, CharLiteralParser, and 11 // StringLiteralParser interfaces. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "clang/Lex/LiteralSupport.h" 16 #include "clang/Basic/CharInfo.h" 17 #include "clang/Basic/TargetInfo.h" 18 #include "clang/Lex/LexDiagnostic.h" 19 #include "clang/Lex/Preprocessor.h" 20 #include "llvm/ADT/StringExtras.h" 21 #include "llvm/Support/ConvertUTF.h" 22 #include "llvm/Support/ErrorHandling.h" 23 24 using namespace clang; 25 26 static unsigned getCharWidth(tok::TokenKind kind, const TargetInfo &Target) { 27 switch (kind) { 28 default: llvm_unreachable("Unknown token type!"); 29 case tok::char_constant: 30 case tok::string_literal: 31 case tok::utf8_string_literal: 32 return Target.getCharWidth(); 33 case tok::wide_char_constant: 34 case tok::wide_string_literal: 35 return Target.getWCharWidth(); 36 case tok::utf16_char_constant: 37 case tok::utf16_string_literal: 38 return Target.getChar16Width(); 39 case tok::utf32_char_constant: 40 case tok::utf32_string_literal: 41 return Target.getChar32Width(); 42 } 43 } 44 45 static CharSourceRange MakeCharSourceRange(const LangOptions &Features, 46 FullSourceLoc TokLoc, 47 const char *TokBegin, 48 const char *TokRangeBegin, 49 const char *TokRangeEnd) { 50 SourceLocation Begin = 51 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin, 52 TokLoc.getManager(), Features); 53 SourceLocation End = 54 Lexer::AdvanceToTokenCharacter(Begin, TokRangeEnd - TokRangeBegin, 55 TokLoc.getManager(), Features); 56 return CharSourceRange::getCharRange(Begin, End); 57 } 58 59 /// \brief Produce a diagnostic highlighting some portion of a literal. 60 /// 61 /// Emits the diagnostic \p DiagID, highlighting the range of characters from 62 /// \p TokRangeBegin (inclusive) to \p TokRangeEnd (exclusive), which must be 63 /// a substring of a spelling buffer for the token beginning at \p TokBegin. 64 static DiagnosticBuilder Diag(DiagnosticsEngine *Diags, 65 const LangOptions &Features, FullSourceLoc TokLoc, 66 const char *TokBegin, const char *TokRangeBegin, 67 const char *TokRangeEnd, unsigned DiagID) { 68 SourceLocation Begin = 69 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin, 70 TokLoc.getManager(), Features); 71 return Diags->Report(Begin, DiagID) << 72 MakeCharSourceRange(Features, TokLoc, TokBegin, TokRangeBegin, TokRangeEnd); 73 } 74 75 /// ProcessCharEscape - Parse a standard C escape sequence, which can occur in 76 /// either a character or a string literal. 77 static unsigned ProcessCharEscape(const char *ThisTokBegin, 78 const char *&ThisTokBuf, 79 const char *ThisTokEnd, bool &HadError, 80 FullSourceLoc Loc, unsigned CharWidth, 81 DiagnosticsEngine *Diags, 82 const LangOptions &Features) { 83 const char *EscapeBegin = ThisTokBuf; 84 85 // Skip the '\' char. 86 ++ThisTokBuf; 87 88 // We know that this character can't be off the end of the buffer, because 89 // that would have been \", which would not have been the end of string. 90 unsigned ResultChar = *ThisTokBuf++; 91 switch (ResultChar) { 92 // These map to themselves. 93 case '\\': case '\'': case '"': case '?': break; 94 95 // These have fixed mappings. 96 case 'a': 97 // TODO: K&R: the meaning of '\\a' is different in traditional C 98 ResultChar = 7; 99 break; 100 case 'b': 101 ResultChar = 8; 102 break; 103 case 'e': 104 if (Diags) 105 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 106 diag::ext_nonstandard_escape) << "e"; 107 ResultChar = 27; 108 break; 109 case 'E': 110 if (Diags) 111 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 112 diag::ext_nonstandard_escape) << "E"; 113 ResultChar = 27; 114 break; 115 case 'f': 116 ResultChar = 12; 117 break; 118 case 'n': 119 ResultChar = 10; 120 break; 121 case 'r': 122 ResultChar = 13; 123 break; 124 case 't': 125 ResultChar = 9; 126 break; 127 case 'v': 128 ResultChar = 11; 129 break; 130 case 'x': { // Hex escape. 131 ResultChar = 0; 132 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) { 133 if (Diags) 134 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 135 diag::err_hex_escape_no_digits) << "x"; 136 HadError = 1; 137 break; 138 } 139 140 // Hex escapes are a maximal series of hex digits. 141 bool Overflow = false; 142 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) { 143 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]); 144 if (CharVal == -1) break; 145 // About to shift out a digit? 146 Overflow |= (ResultChar & 0xF0000000) ? true : false; 147 ResultChar <<= 4; 148 ResultChar |= CharVal; 149 } 150 151 // See if any bits will be truncated when evaluated as a character. 152 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 153 Overflow = true; 154 ResultChar &= ~0U >> (32-CharWidth); 155 } 156 157 // Check for overflow. 158 if (Overflow && Diags) // Too many digits to fit in 159 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 160 diag::warn_hex_escape_too_large); 161 break; 162 } 163 case '0': case '1': case '2': case '3': 164 case '4': case '5': case '6': case '7': { 165 // Octal escapes. 166 --ThisTokBuf; 167 ResultChar = 0; 168 169 // Octal escapes are a series of octal digits with maximum length 3. 170 // "\0123" is a two digit sequence equal to "\012" "3". 171 unsigned NumDigits = 0; 172 do { 173 ResultChar <<= 3; 174 ResultChar |= *ThisTokBuf++ - '0'; 175 ++NumDigits; 176 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 && 177 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7'); 178 179 // Check for overflow. Reject '\777', but not L'\777'. 180 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 181 if (Diags) 182 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 183 diag::warn_octal_escape_too_large); 184 ResultChar &= ~0U >> (32-CharWidth); 185 } 186 break; 187 } 188 189 // Otherwise, these are not valid escapes. 190 case '(': case '{': case '[': case '%': 191 // GCC accepts these as extensions. We warn about them as such though. 192 if (Diags) 193 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 194 diag::ext_nonstandard_escape) 195 << std::string(1, ResultChar); 196 break; 197 default: 198 if (Diags == 0) 199 break; 200 201 if (isPrintable(ResultChar)) 202 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 203 diag::ext_unknown_escape) 204 << std::string(1, ResultChar); 205 else 206 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 207 diag::ext_unknown_escape) 208 << "x" + llvm::utohexstr(ResultChar); 209 break; 210 } 211 212 return ResultChar; 213 } 214 215 /// ProcessUCNEscape - Read the Universal Character Name, check constraints and 216 /// return the UTF32. 217 static bool ProcessUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 218 const char *ThisTokEnd, 219 uint32_t &UcnVal, unsigned short &UcnLen, 220 FullSourceLoc Loc, DiagnosticsEngine *Diags, 221 const LangOptions &Features, 222 bool in_char_string_literal = false) { 223 const char *UcnBegin = ThisTokBuf; 224 225 // Skip the '\u' char's. 226 ThisTokBuf += 2; 227 228 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) { 229 if (Diags) 230 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 231 diag::err_hex_escape_no_digits) << StringRef(&ThisTokBuf[-1], 1); 232 return false; 233 } 234 UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8); 235 unsigned short UcnLenSave = UcnLen; 236 for (; ThisTokBuf != ThisTokEnd && UcnLenSave; ++ThisTokBuf, UcnLenSave--) { 237 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]); 238 if (CharVal == -1) break; 239 UcnVal <<= 4; 240 UcnVal |= CharVal; 241 } 242 // If we didn't consume the proper number of digits, there is a problem. 243 if (UcnLenSave) { 244 if (Diags) 245 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 246 diag::err_ucn_escape_incomplete); 247 return false; 248 } 249 250 // Check UCN constraints (C99 6.4.3p2) [C++11 lex.charset p2] 251 if ((0xD800 <= UcnVal && UcnVal <= 0xDFFF) || // surrogate codepoints 252 UcnVal > 0x10FFFF) { // maximum legal UTF32 value 253 if (Diags) 254 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 255 diag::err_ucn_escape_invalid); 256 return false; 257 } 258 259 // C++11 allows UCNs that refer to control characters and basic source 260 // characters inside character and string literals 261 if (UcnVal < 0xa0 && 262 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60)) { // $, @, ` 263 bool IsError = (!Features.CPlusPlus11 || !in_char_string_literal); 264 if (Diags) { 265 char BasicSCSChar = UcnVal; 266 if (UcnVal >= 0x20 && UcnVal < 0x7f) 267 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 268 IsError ? diag::err_ucn_escape_basic_scs : 269 diag::warn_cxx98_compat_literal_ucn_escape_basic_scs) 270 << StringRef(&BasicSCSChar, 1); 271 else 272 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 273 IsError ? diag::err_ucn_control_character : 274 diag::warn_cxx98_compat_literal_ucn_control_character); 275 } 276 if (IsError) 277 return false; 278 } 279 280 if (!Features.CPlusPlus && !Features.C99 && Diags) 281 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 282 diag::warn_ucn_not_valid_in_c89_literal); 283 284 return true; 285 } 286 287 /// MeasureUCNEscape - Determine the number of bytes within the resulting string 288 /// which this UCN will occupy. 289 static int MeasureUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 290 const char *ThisTokEnd, unsigned CharByteWidth, 291 const LangOptions &Features, bool &HadError) { 292 // UTF-32: 4 bytes per escape. 293 if (CharByteWidth == 4) 294 return 4; 295 296 uint32_t UcnVal = 0; 297 unsigned short UcnLen = 0; 298 FullSourceLoc Loc; 299 300 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, 301 UcnLen, Loc, 0, Features, true)) { 302 HadError = true; 303 return 0; 304 } 305 306 // UTF-16: 2 bytes for BMP, 4 bytes otherwise. 307 if (CharByteWidth == 2) 308 return UcnVal <= 0xFFFF ? 2 : 4; 309 310 // UTF-8. 311 if (UcnVal < 0x80) 312 return 1; 313 if (UcnVal < 0x800) 314 return 2; 315 if (UcnVal < 0x10000) 316 return 3; 317 return 4; 318 } 319 320 /// EncodeUCNEscape - Read the Universal Character Name, check constraints and 321 /// convert the UTF32 to UTF8 or UTF16. This is a subroutine of 322 /// StringLiteralParser. When we decide to implement UCN's for identifiers, 323 /// we will likely rework our support for UCN's. 324 static void EncodeUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 325 const char *ThisTokEnd, 326 char *&ResultBuf, bool &HadError, 327 FullSourceLoc Loc, unsigned CharByteWidth, 328 DiagnosticsEngine *Diags, 329 const LangOptions &Features) { 330 typedef uint32_t UTF32; 331 UTF32 UcnVal = 0; 332 unsigned short UcnLen = 0; 333 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, UcnLen, 334 Loc, Diags, Features, true)) { 335 HadError = true; 336 return; 337 } 338 339 assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth) && 340 "only character widths of 1, 2, or 4 bytes supported"); 341 342 (void)UcnLen; 343 assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported"); 344 345 if (CharByteWidth == 4) { 346 // FIXME: Make the type of the result buffer correct instead of 347 // using reinterpret_cast. 348 UTF32 *ResultPtr = reinterpret_cast<UTF32*>(ResultBuf); 349 *ResultPtr = UcnVal; 350 ResultBuf += 4; 351 return; 352 } 353 354 if (CharByteWidth == 2) { 355 // FIXME: Make the type of the result buffer correct instead of 356 // using reinterpret_cast. 357 UTF16 *ResultPtr = reinterpret_cast<UTF16*>(ResultBuf); 358 359 if (UcnVal <= (UTF32)0xFFFF) { 360 *ResultPtr = UcnVal; 361 ResultBuf += 2; 362 return; 363 } 364 365 // Convert to UTF16. 366 UcnVal -= 0x10000; 367 *ResultPtr = 0xD800 + (UcnVal >> 10); 368 *(ResultPtr+1) = 0xDC00 + (UcnVal & 0x3FF); 369 ResultBuf += 4; 370 return; 371 } 372 373 assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters"); 374 375 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8. 376 // The conversion below was inspired by: 377 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c 378 // First, we determine how many bytes the result will require. 379 typedef uint8_t UTF8; 380 381 unsigned short bytesToWrite = 0; 382 if (UcnVal < (UTF32)0x80) 383 bytesToWrite = 1; 384 else if (UcnVal < (UTF32)0x800) 385 bytesToWrite = 2; 386 else if (UcnVal < (UTF32)0x10000) 387 bytesToWrite = 3; 388 else 389 bytesToWrite = 4; 390 391 const unsigned byteMask = 0xBF; 392 const unsigned byteMark = 0x80; 393 394 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed 395 // into the first byte, depending on how many bytes follow. 396 static const UTF8 firstByteMark[5] = { 397 0x00, 0x00, 0xC0, 0xE0, 0xF0 398 }; 399 // Finally, we write the bytes into ResultBuf. 400 ResultBuf += bytesToWrite; 401 switch (bytesToWrite) { // note: everything falls through. 402 case 4: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 403 case 3: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 404 case 2: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 405 case 1: *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]); 406 } 407 // Update the buffer. 408 ResultBuf += bytesToWrite; 409 } 410 411 412 /// integer-constant: [C99 6.4.4.1] 413 /// decimal-constant integer-suffix 414 /// octal-constant integer-suffix 415 /// hexadecimal-constant integer-suffix 416 /// user-defined-integer-literal: [C++11 lex.ext] 417 /// decimal-literal ud-suffix 418 /// octal-literal ud-suffix 419 /// hexadecimal-literal ud-suffix 420 /// decimal-constant: 421 /// nonzero-digit 422 /// decimal-constant digit 423 /// octal-constant: 424 /// 0 425 /// octal-constant octal-digit 426 /// hexadecimal-constant: 427 /// hexadecimal-prefix hexadecimal-digit 428 /// hexadecimal-constant hexadecimal-digit 429 /// hexadecimal-prefix: one of 430 /// 0x 0X 431 /// integer-suffix: 432 /// unsigned-suffix [long-suffix] 433 /// unsigned-suffix [long-long-suffix] 434 /// long-suffix [unsigned-suffix] 435 /// long-long-suffix [unsigned-sufix] 436 /// nonzero-digit: 437 /// 1 2 3 4 5 6 7 8 9 438 /// octal-digit: 439 /// 0 1 2 3 4 5 6 7 440 /// hexadecimal-digit: 441 /// 0 1 2 3 4 5 6 7 8 9 442 /// a b c d e f 443 /// A B C D E F 444 /// unsigned-suffix: one of 445 /// u U 446 /// long-suffix: one of 447 /// l L 448 /// long-long-suffix: one of 449 /// ll LL 450 /// 451 /// floating-constant: [C99 6.4.4.2] 452 /// TODO: add rules... 453 /// 454 NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling, 455 SourceLocation TokLoc, 456 Preprocessor &PP) 457 : PP(PP), ThisTokBegin(TokSpelling.begin()), ThisTokEnd(TokSpelling.end()) { 458 459 // This routine assumes that the range begin/end matches the regex for integer 460 // and FP constants (specifically, the 'pp-number' regex), and assumes that 461 // the byte at "*end" is both valid and not part of the regex. Because of 462 // this, it doesn't have to check for 'overscan' in various places. 463 assert(!isPreprocessingNumberBody(*ThisTokEnd) && "didn't maximally munch?"); 464 465 s = DigitsBegin = ThisTokBegin; 466 saw_exponent = false; 467 saw_period = false; 468 saw_ud_suffix = false; 469 isLong = false; 470 isUnsigned = false; 471 isLongLong = false; 472 isFloat = false; 473 isImaginary = false; 474 isMicrosoftInteger = false; 475 hadError = false; 476 477 if (*s == '0') { // parse radix 478 ParseNumberStartingWithZero(TokLoc); 479 if (hadError) 480 return; 481 } else { // the first digit is non-zero 482 radix = 10; 483 s = SkipDigits(s); 484 if (s == ThisTokEnd) { 485 // Done. 486 } else if (isHexDigit(*s) && !(*s == 'e' || *s == 'E')) { 487 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 488 diag::err_invalid_decimal_digit) << StringRef(s, 1); 489 hadError = true; 490 return; 491 } else if (*s == '.') { 492 s++; 493 saw_period = true; 494 s = SkipDigits(s); 495 } 496 if ((*s == 'e' || *s == 'E')) { // exponent 497 const char *Exponent = s; 498 s++; 499 saw_exponent = true; 500 if (*s == '+' || *s == '-') s++; // sign 501 const char *first_non_digit = SkipDigits(s); 502 if (first_non_digit != s) { 503 s = first_non_digit; 504 } else { 505 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent - ThisTokBegin), 506 diag::err_exponent_has_no_digits); 507 hadError = true; 508 return; 509 } 510 } 511 } 512 513 SuffixBegin = s; 514 515 // Parse the suffix. At this point we can classify whether we have an FP or 516 // integer constant. 517 bool isFPConstant = isFloatingLiteral(); 518 519 // Loop over all of the characters of the suffix. If we see something bad, 520 // we break out of the loop. 521 for (; s != ThisTokEnd; ++s) { 522 switch (*s) { 523 case 'f': // FP Suffix for "float" 524 case 'F': 525 if (!isFPConstant) break; // Error for integer constant. 526 if (isFloat || isLong) break; // FF, LF invalid. 527 isFloat = true; 528 continue; // Success. 529 case 'u': 530 case 'U': 531 if (isFPConstant) break; // Error for floating constant. 532 if (isUnsigned) break; // Cannot be repeated. 533 isUnsigned = true; 534 continue; // Success. 535 case 'l': 536 case 'L': 537 if (isLong || isLongLong) break; // Cannot be repeated. 538 if (isFloat) break; // LF invalid. 539 540 // Check for long long. The L's need to be adjacent and the same case. 541 if (s+1 != ThisTokEnd && s[1] == s[0]) { 542 if (isFPConstant) break; // long long invalid for floats. 543 isLongLong = true; 544 ++s; // Eat both of them. 545 } else { 546 isLong = true; 547 } 548 continue; // Success. 549 case 'i': 550 case 'I': 551 if (PP.getLangOpts().MicrosoftExt) { 552 if (isFPConstant || isLong || isLongLong) break; 553 554 // Allow i8, i16, i32, i64, and i128. 555 if (s + 1 != ThisTokEnd) { 556 switch (s[1]) { 557 case '8': 558 s += 2; // i8 suffix 559 isMicrosoftInteger = true; 560 break; 561 case '1': 562 if (s + 2 == ThisTokEnd) break; 563 if (s[2] == '6') { 564 s += 3; // i16 suffix 565 isMicrosoftInteger = true; 566 } 567 else if (s[2] == '2') { 568 if (s + 3 == ThisTokEnd) break; 569 if (s[3] == '8') { 570 s += 4; // i128 suffix 571 isMicrosoftInteger = true; 572 } 573 } 574 break; 575 case '3': 576 if (s + 2 == ThisTokEnd) break; 577 if (s[2] == '2') { 578 s += 3; // i32 suffix 579 isLong = true; 580 isMicrosoftInteger = true; 581 } 582 break; 583 case '6': 584 if (s + 2 == ThisTokEnd) break; 585 if (s[2] == '4') { 586 s += 3; // i64 suffix 587 isLongLong = true; 588 isMicrosoftInteger = true; 589 } 590 break; 591 default: 592 break; 593 } 594 break; 595 } 596 } 597 // fall through. 598 case 'j': 599 case 'J': 600 if (isImaginary) break; // Cannot be repeated. 601 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 602 diag::ext_imaginary_constant); 603 isImaginary = true; 604 continue; // Success. 605 } 606 // If we reached here, there was an error or a ud-suffix. 607 break; 608 } 609 610 if (s != ThisTokEnd) { 611 if (PP.getLangOpts().CPlusPlus11 && s == SuffixBegin && *s == '_') { 612 // We have a ud-suffix! By C++11 [lex.ext]p10, ud-suffixes not starting 613 // with an '_' are ill-formed. 614 saw_ud_suffix = true; 615 return; 616 } 617 618 // Report an error if there are any. 619 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin), 620 isFPConstant ? diag::err_invalid_suffix_float_constant : 621 diag::err_invalid_suffix_integer_constant) 622 << StringRef(SuffixBegin, ThisTokEnd-SuffixBegin); 623 hadError = true; 624 return; 625 } 626 } 627 628 /// ParseNumberStartingWithZero - This method is called when the first character 629 /// of the number is found to be a zero. This means it is either an octal 630 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or 631 /// a floating point number (01239.123e4). Eat the prefix, determining the 632 /// radix etc. 633 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) { 634 assert(s[0] == '0' && "Invalid method call"); 635 s++; 636 637 // Handle a hex number like 0x1234. 638 if ((*s == 'x' || *s == 'X') && (isHexDigit(s[1]) || s[1] == '.')) { 639 s++; 640 radix = 16; 641 DigitsBegin = s; 642 s = SkipHexDigits(s); 643 bool noSignificand = (s == DigitsBegin); 644 if (s == ThisTokEnd) { 645 // Done. 646 } else if (*s == '.') { 647 s++; 648 saw_period = true; 649 const char *floatDigitsBegin = s; 650 s = SkipHexDigits(s); 651 noSignificand &= (floatDigitsBegin == s); 652 } 653 654 if (noSignificand) { 655 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 656 diag::err_hexconstant_requires_digits); 657 hadError = true; 658 return; 659 } 660 661 // A binary exponent can appear with or with a '.'. If dotted, the 662 // binary exponent is required. 663 if (*s == 'p' || *s == 'P') { 664 const char *Exponent = s; 665 s++; 666 saw_exponent = true; 667 if (*s == '+' || *s == '-') s++; // sign 668 const char *first_non_digit = SkipDigits(s); 669 if (first_non_digit == s) { 670 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 671 diag::err_exponent_has_no_digits); 672 hadError = true; 673 return; 674 } 675 s = first_non_digit; 676 677 if (!PP.getLangOpts().HexFloats) 678 PP.Diag(TokLoc, diag::ext_hexconstant_invalid); 679 } else if (saw_period) { 680 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 681 diag::err_hexconstant_requires_exponent); 682 hadError = true; 683 } 684 return; 685 } 686 687 // Handle simple binary numbers 0b01010 688 if (*s == 'b' || *s == 'B') { 689 // 0b101010 is a GCC extension. 690 PP.Diag(TokLoc, diag::ext_binary_literal); 691 ++s; 692 radix = 2; 693 DigitsBegin = s; 694 s = SkipBinaryDigits(s); 695 if (s == ThisTokEnd) { 696 // Done. 697 } else if (isHexDigit(*s)) { 698 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 699 diag::err_invalid_binary_digit) << StringRef(s, 1); 700 hadError = true; 701 } 702 // Other suffixes will be diagnosed by the caller. 703 return; 704 } 705 706 // For now, the radix is set to 8. If we discover that we have a 707 // floating point constant, the radix will change to 10. Octal floating 708 // point constants are not permitted (only decimal and hexadecimal). 709 radix = 8; 710 DigitsBegin = s; 711 s = SkipOctalDigits(s); 712 if (s == ThisTokEnd) 713 return; // Done, simple octal number like 01234 714 715 // If we have some other non-octal digit that *is* a decimal digit, see if 716 // this is part of a floating point number like 094.123 or 09e1. 717 if (isDigit(*s)) { 718 const char *EndDecimal = SkipDigits(s); 719 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') { 720 s = EndDecimal; 721 radix = 10; 722 } 723 } 724 725 // If we have a hex digit other than 'e' (which denotes a FP exponent) then 726 // the code is using an incorrect base. 727 if (isHexDigit(*s) && *s != 'e' && *s != 'E') { 728 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 729 diag::err_invalid_octal_digit) << StringRef(s, 1); 730 hadError = true; 731 return; 732 } 733 734 if (*s == '.') { 735 s++; 736 radix = 10; 737 saw_period = true; 738 s = SkipDigits(s); // Skip suffix. 739 } 740 if (*s == 'e' || *s == 'E') { // exponent 741 const char *Exponent = s; 742 s++; 743 radix = 10; 744 saw_exponent = true; 745 if (*s == '+' || *s == '-') s++; // sign 746 const char *first_non_digit = SkipDigits(s); 747 if (first_non_digit != s) { 748 s = first_non_digit; 749 } else { 750 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 751 diag::err_exponent_has_no_digits); 752 hadError = true; 753 return; 754 } 755 } 756 } 757 758 static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) { 759 switch (Radix) { 760 case 2: 761 return NumDigits <= 64; 762 case 8: 763 return NumDigits <= 64 / 3; // Digits are groups of 3 bits. 764 case 10: 765 return NumDigits <= 19; // floor(log10(2^64)) 766 case 16: 767 return NumDigits <= 64 / 4; // Digits are groups of 4 bits. 768 default: 769 llvm_unreachable("impossible Radix"); 770 } 771 } 772 773 /// GetIntegerValue - Convert this numeric literal value to an APInt that 774 /// matches Val's input width. If there is an overflow, set Val to the low bits 775 /// of the result and return true. Otherwise, return false. 776 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) { 777 // Fast path: Compute a conservative bound on the maximum number of 778 // bits per digit in this radix. If we can't possibly overflow a 779 // uint64 based on that bound then do the simple conversion to 780 // integer. This avoids the expensive overflow checking below, and 781 // handles the common cases that matter (small decimal integers and 782 // hex/octal values which don't overflow). 783 const unsigned NumDigits = SuffixBegin - DigitsBegin; 784 if (alwaysFitsInto64Bits(radix, NumDigits)) { 785 uint64_t N = 0; 786 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr) 787 N = N * radix + llvm::hexDigitValue(*Ptr); 788 789 // This will truncate the value to Val's input width. Simply check 790 // for overflow by comparing. 791 Val = N; 792 return Val.getZExtValue() != N; 793 } 794 795 Val = 0; 796 const char *Ptr = DigitsBegin; 797 798 llvm::APInt RadixVal(Val.getBitWidth(), radix); 799 llvm::APInt CharVal(Val.getBitWidth(), 0); 800 llvm::APInt OldVal = Val; 801 802 bool OverflowOccurred = false; 803 while (Ptr < SuffixBegin) { 804 unsigned C = llvm::hexDigitValue(*Ptr++); 805 806 // If this letter is out of bound for this radix, reject it. 807 assert(C < radix && "NumericLiteralParser ctor should have rejected this"); 808 809 CharVal = C; 810 811 // Add the digit to the value in the appropriate radix. If adding in digits 812 // made the value smaller, then this overflowed. 813 OldVal = Val; 814 815 // Multiply by radix, did overflow occur on the multiply? 816 Val *= RadixVal; 817 OverflowOccurred |= Val.udiv(RadixVal) != OldVal; 818 819 // Add value, did overflow occur on the value? 820 // (a + b) ult b <=> overflow 821 Val += CharVal; 822 OverflowOccurred |= Val.ult(CharVal); 823 } 824 return OverflowOccurred; 825 } 826 827 llvm::APFloat::opStatus 828 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) { 829 using llvm::APFloat; 830 831 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin); 832 return Result.convertFromString(StringRef(ThisTokBegin, n), 833 APFloat::rmNearestTiesToEven); 834 } 835 836 837 /// \verbatim 838 /// user-defined-character-literal: [C++11 lex.ext] 839 /// character-literal ud-suffix 840 /// ud-suffix: 841 /// identifier 842 /// character-literal: [C++11 lex.ccon] 843 /// ' c-char-sequence ' 844 /// u' c-char-sequence ' 845 /// U' c-char-sequence ' 846 /// L' c-char-sequence ' 847 /// c-char-sequence: 848 /// c-char 849 /// c-char-sequence c-char 850 /// c-char: 851 /// any member of the source character set except the single-quote ', 852 /// backslash \, or new-line character 853 /// escape-sequence 854 /// universal-character-name 855 /// escape-sequence: 856 /// simple-escape-sequence 857 /// octal-escape-sequence 858 /// hexadecimal-escape-sequence 859 /// simple-escape-sequence: 860 /// one of \' \" \? \\ \a \b \f \n \r \t \v 861 /// octal-escape-sequence: 862 /// \ octal-digit 863 /// \ octal-digit octal-digit 864 /// \ octal-digit octal-digit octal-digit 865 /// hexadecimal-escape-sequence: 866 /// \x hexadecimal-digit 867 /// hexadecimal-escape-sequence hexadecimal-digit 868 /// universal-character-name: [C++11 lex.charset] 869 /// \u hex-quad 870 /// \U hex-quad hex-quad 871 /// hex-quad: 872 /// hex-digit hex-digit hex-digit hex-digit 873 /// \endverbatim 874 /// 875 CharLiteralParser::CharLiteralParser(const char *begin, const char *end, 876 SourceLocation Loc, Preprocessor &PP, 877 tok::TokenKind kind) { 878 // At this point we know that the character matches the regex "(L|u|U)?'.*'". 879 HadError = false; 880 881 Kind = kind; 882 883 const char *TokBegin = begin; 884 885 // Skip over wide character determinant. 886 if (Kind != tok::char_constant) { 887 ++begin; 888 } 889 890 // Skip over the entry quote. 891 assert(begin[0] == '\'' && "Invalid token lexed"); 892 ++begin; 893 894 // Remove an optional ud-suffix. 895 if (end[-1] != '\'') { 896 const char *UDSuffixEnd = end; 897 do { 898 --end; 899 } while (end[-1] != '\''); 900 UDSuffixBuf.assign(end, UDSuffixEnd); 901 UDSuffixOffset = end - TokBegin; 902 } 903 904 // Trim the ending quote. 905 assert(end != begin && "Invalid token lexed"); 906 --end; 907 908 // FIXME: The "Value" is an uint64_t so we can handle char literals of 909 // up to 64-bits. 910 // FIXME: This extensively assumes that 'char' is 8-bits. 911 assert(PP.getTargetInfo().getCharWidth() == 8 && 912 "Assumes char is 8 bits"); 913 assert(PP.getTargetInfo().getIntWidth() <= 64 && 914 (PP.getTargetInfo().getIntWidth() & 7) == 0 && 915 "Assumes sizeof(int) on target is <= 64 and a multiple of char"); 916 assert(PP.getTargetInfo().getWCharWidth() <= 64 && 917 "Assumes sizeof(wchar) on target is <= 64"); 918 919 SmallVector<uint32_t,4> codepoint_buffer; 920 codepoint_buffer.resize(end-begin); 921 uint32_t *buffer_begin = &codepoint_buffer.front(); 922 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size(); 923 924 // Unicode escapes representing characters that cannot be correctly 925 // represented in a single code unit are disallowed in character literals 926 // by this implementation. 927 uint32_t largest_character_for_kind; 928 if (tok::wide_char_constant == Kind) { 929 largest_character_for_kind = 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth()); 930 } else if (tok::utf16_char_constant == Kind) { 931 largest_character_for_kind = 0xFFFF; 932 } else if (tok::utf32_char_constant == Kind) { 933 largest_character_for_kind = 0x10FFFF; 934 } else { 935 largest_character_for_kind = 0x7Fu; 936 } 937 938 while (begin!=end) { 939 // Is this a span of non-escape characters? 940 if (begin[0] != '\\') { 941 char const *start = begin; 942 do { 943 ++begin; 944 } while (begin != end && *begin != '\\'); 945 946 char const *tmp_in_start = start; 947 uint32_t *tmp_out_start = buffer_begin; 948 ConversionResult res = 949 ConvertUTF8toUTF32(reinterpret_cast<UTF8 const **>(&start), 950 reinterpret_cast<UTF8 const *>(begin), 951 &buffer_begin,buffer_end,strictConversion); 952 if (res!=conversionOK) { 953 // If we see bad encoding for unprefixed character literals, warn and 954 // simply copy the byte values, for compatibility with gcc and 955 // older versions of clang. 956 bool NoErrorOnBadEncoding = isAscii(); 957 unsigned Msg = diag::err_bad_character_encoding; 958 if (NoErrorOnBadEncoding) 959 Msg = diag::warn_bad_character_encoding; 960 PP.Diag(Loc, Msg); 961 if (NoErrorOnBadEncoding) { 962 start = tmp_in_start; 963 buffer_begin = tmp_out_start; 964 for ( ; start != begin; ++start, ++buffer_begin) 965 *buffer_begin = static_cast<uint8_t>(*start); 966 } else { 967 HadError = true; 968 } 969 } else { 970 for (; tmp_out_start <buffer_begin; ++tmp_out_start) { 971 if (*tmp_out_start > largest_character_for_kind) { 972 HadError = true; 973 PP.Diag(Loc, diag::err_character_too_large); 974 } 975 } 976 } 977 978 continue; 979 } 980 // Is this a Universal Character Name excape? 981 if (begin[1] == 'u' || begin[1] == 'U') { 982 unsigned short UcnLen = 0; 983 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen, 984 FullSourceLoc(Loc, PP.getSourceManager()), 985 &PP.getDiagnostics(), PP.getLangOpts(), 986 true)) 987 { 988 HadError = true; 989 } else if (*buffer_begin > largest_character_for_kind) { 990 HadError = true; 991 PP.Diag(Loc, diag::err_character_too_large); 992 } 993 994 ++buffer_begin; 995 continue; 996 } 997 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo()); 998 uint64_t result = 999 ProcessCharEscape(TokBegin, begin, end, HadError, 1000 FullSourceLoc(Loc,PP.getSourceManager()), 1001 CharWidth, &PP.getDiagnostics(), PP.getLangOpts()); 1002 *buffer_begin++ = result; 1003 } 1004 1005 unsigned NumCharsSoFar = buffer_begin-&codepoint_buffer.front(); 1006 1007 if (NumCharsSoFar > 1) { 1008 if (isWide()) 1009 PP.Diag(Loc, diag::warn_extraneous_char_constant); 1010 else if (isAscii() && NumCharsSoFar == 4) 1011 PP.Diag(Loc, diag::ext_four_char_character_literal); 1012 else if (isAscii()) 1013 PP.Diag(Loc, diag::ext_multichar_character_literal); 1014 else 1015 PP.Diag(Loc, diag::err_multichar_utf_character_literal); 1016 IsMultiChar = true; 1017 } else 1018 IsMultiChar = false; 1019 1020 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0); 1021 1022 // Narrow character literals act as though their value is concatenated 1023 // in this implementation, but warn on overflow. 1024 bool multi_char_too_long = false; 1025 if (isAscii() && isMultiChar()) { 1026 LitVal = 0; 1027 for (size_t i=0;i<NumCharsSoFar;++i) { 1028 // check for enough leading zeros to shift into 1029 multi_char_too_long |= (LitVal.countLeadingZeros() < 8); 1030 LitVal <<= 8; 1031 LitVal = LitVal + (codepoint_buffer[i] & 0xFF); 1032 } 1033 } else if (NumCharsSoFar > 0) { 1034 // otherwise just take the last character 1035 LitVal = buffer_begin[-1]; 1036 } 1037 1038 if (!HadError && multi_char_too_long) { 1039 PP.Diag(Loc,diag::warn_char_constant_too_large); 1040 } 1041 1042 // Transfer the value from APInt to uint64_t 1043 Value = LitVal.getZExtValue(); 1044 1045 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1") 1046 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple 1047 // character constants are not sign extended in the this implementation: 1048 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC. 1049 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) && 1050 PP.getLangOpts().CharIsSigned) 1051 Value = (signed char)Value; 1052 } 1053 1054 /// \verbatim 1055 /// string-literal: [C++0x lex.string] 1056 /// encoding-prefix " [s-char-sequence] " 1057 /// encoding-prefix R raw-string 1058 /// encoding-prefix: 1059 /// u8 1060 /// u 1061 /// U 1062 /// L 1063 /// s-char-sequence: 1064 /// s-char 1065 /// s-char-sequence s-char 1066 /// s-char: 1067 /// any member of the source character set except the double-quote ", 1068 /// backslash \, or new-line character 1069 /// escape-sequence 1070 /// universal-character-name 1071 /// raw-string: 1072 /// " d-char-sequence ( r-char-sequence ) d-char-sequence " 1073 /// r-char-sequence: 1074 /// r-char 1075 /// r-char-sequence r-char 1076 /// r-char: 1077 /// any member of the source character set, except a right parenthesis ) 1078 /// followed by the initial d-char-sequence (which may be empty) 1079 /// followed by a double quote ". 1080 /// d-char-sequence: 1081 /// d-char 1082 /// d-char-sequence d-char 1083 /// d-char: 1084 /// any member of the basic source character set except: 1085 /// space, the left parenthesis (, the right parenthesis ), 1086 /// the backslash \, and the control characters representing horizontal 1087 /// tab, vertical tab, form feed, and newline. 1088 /// escape-sequence: [C++0x lex.ccon] 1089 /// simple-escape-sequence 1090 /// octal-escape-sequence 1091 /// hexadecimal-escape-sequence 1092 /// simple-escape-sequence: 1093 /// one of \' \" \? \\ \a \b \f \n \r \t \v 1094 /// octal-escape-sequence: 1095 /// \ octal-digit 1096 /// \ octal-digit octal-digit 1097 /// \ octal-digit octal-digit octal-digit 1098 /// hexadecimal-escape-sequence: 1099 /// \x hexadecimal-digit 1100 /// hexadecimal-escape-sequence hexadecimal-digit 1101 /// universal-character-name: 1102 /// \u hex-quad 1103 /// \U hex-quad hex-quad 1104 /// hex-quad: 1105 /// hex-digit hex-digit hex-digit hex-digit 1106 /// \endverbatim 1107 /// 1108 StringLiteralParser:: 1109 StringLiteralParser(const Token *StringToks, unsigned NumStringToks, 1110 Preprocessor &PP, bool Complain) 1111 : SM(PP.getSourceManager()), Features(PP.getLangOpts()), 1112 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() : 0), 1113 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown), 1114 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) { 1115 init(StringToks, NumStringToks); 1116 } 1117 1118 void StringLiteralParser::init(const Token *StringToks, unsigned NumStringToks){ 1119 // The literal token may have come from an invalid source location (e.g. due 1120 // to a PCH error), in which case the token length will be 0. 1121 if (NumStringToks == 0 || StringToks[0].getLength() < 2) 1122 return DiagnoseLexingError(SourceLocation()); 1123 1124 // Scan all of the string portions, remember the max individual token length, 1125 // computing a bound on the concatenated string length, and see whether any 1126 // piece is a wide-string. If any of the string portions is a wide-string 1127 // literal, the result is a wide-string literal [C99 6.4.5p4]. 1128 assert(NumStringToks && "expected at least one token"); 1129 MaxTokenLength = StringToks[0].getLength(); 1130 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!"); 1131 SizeBound = StringToks[0].getLength()-2; // -2 for "". 1132 Kind = StringToks[0].getKind(); 1133 1134 hadError = false; 1135 1136 // Implement Translation Phase #6: concatenation of string literals 1137 /// (C99 5.1.1.2p1). The common case is only one string fragment. 1138 for (unsigned i = 1; i != NumStringToks; ++i) { 1139 if (StringToks[i].getLength() < 2) 1140 return DiagnoseLexingError(StringToks[i].getLocation()); 1141 1142 // The string could be shorter than this if it needs cleaning, but this is a 1143 // reasonable bound, which is all we need. 1144 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!"); 1145 SizeBound += StringToks[i].getLength()-2; // -2 for "". 1146 1147 // Remember maximum string piece length. 1148 if (StringToks[i].getLength() > MaxTokenLength) 1149 MaxTokenLength = StringToks[i].getLength(); 1150 1151 // Remember if we see any wide or utf-8/16/32 strings. 1152 // Also check for illegal concatenations. 1153 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) { 1154 if (isAscii()) { 1155 Kind = StringToks[i].getKind(); 1156 } else { 1157 if (Diags) 1158 Diags->Report(StringToks[i].getLocation(), 1159 diag::err_unsupported_string_concat); 1160 hadError = true; 1161 } 1162 } 1163 } 1164 1165 // Include space for the null terminator. 1166 ++SizeBound; 1167 1168 // TODO: K&R warning: "traditional C rejects string constant concatenation" 1169 1170 // Get the width in bytes of char/wchar_t/char16_t/char32_t 1171 CharByteWidth = getCharWidth(Kind, Target); 1172 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple"); 1173 CharByteWidth /= 8; 1174 1175 // The output buffer size needs to be large enough to hold wide characters. 1176 // This is a worst-case assumption which basically corresponds to L"" "long". 1177 SizeBound *= CharByteWidth; 1178 1179 // Size the temporary buffer to hold the result string data. 1180 ResultBuf.resize(SizeBound); 1181 1182 // Likewise, but for each string piece. 1183 SmallString<512> TokenBuf; 1184 TokenBuf.resize(MaxTokenLength); 1185 1186 // Loop over all the strings, getting their spelling, and expanding them to 1187 // wide strings as appropriate. 1188 ResultPtr = &ResultBuf[0]; // Next byte to fill in. 1189 1190 Pascal = false; 1191 1192 SourceLocation UDSuffixTokLoc; 1193 1194 for (unsigned i = 0, e = NumStringToks; i != e; ++i) { 1195 const char *ThisTokBuf = &TokenBuf[0]; 1196 // Get the spelling of the token, which eliminates trigraphs, etc. We know 1197 // that ThisTokBuf points to a buffer that is big enough for the whole token 1198 // and 'spelled' tokens can only shrink. 1199 bool StringInvalid = false; 1200 unsigned ThisTokLen = 1201 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features, 1202 &StringInvalid); 1203 if (StringInvalid) 1204 return DiagnoseLexingError(StringToks[i].getLocation()); 1205 1206 const char *ThisTokBegin = ThisTokBuf; 1207 const char *ThisTokEnd = ThisTokBuf+ThisTokLen; 1208 1209 // Remove an optional ud-suffix. 1210 if (ThisTokEnd[-1] != '"') { 1211 const char *UDSuffixEnd = ThisTokEnd; 1212 do { 1213 --ThisTokEnd; 1214 } while (ThisTokEnd[-1] != '"'); 1215 1216 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd); 1217 1218 if (UDSuffixBuf.empty()) { 1219 UDSuffixBuf.assign(UDSuffix); 1220 UDSuffixToken = i; 1221 UDSuffixOffset = ThisTokEnd - ThisTokBuf; 1222 UDSuffixTokLoc = StringToks[i].getLocation(); 1223 } else if (!UDSuffixBuf.equals(UDSuffix)) { 1224 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the 1225 // result of a concatenation involving at least one user-defined-string- 1226 // literal, all the participating user-defined-string-literals shall 1227 // have the same ud-suffix. 1228 if (Diags) { 1229 SourceLocation TokLoc = StringToks[i].getLocation(); 1230 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix) 1231 << UDSuffixBuf << UDSuffix 1232 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc) 1233 << SourceRange(TokLoc, TokLoc); 1234 } 1235 hadError = true; 1236 } 1237 } 1238 1239 // Strip the end quote. 1240 --ThisTokEnd; 1241 1242 // TODO: Input character set mapping support. 1243 1244 // Skip marker for wide or unicode strings. 1245 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') { 1246 ++ThisTokBuf; 1247 // Skip 8 of u8 marker for utf8 strings. 1248 if (ThisTokBuf[0] == '8') 1249 ++ThisTokBuf; 1250 } 1251 1252 // Check for raw string 1253 if (ThisTokBuf[0] == 'R') { 1254 ThisTokBuf += 2; // skip R" 1255 1256 const char *Prefix = ThisTokBuf; 1257 while (ThisTokBuf[0] != '(') 1258 ++ThisTokBuf; 1259 ++ThisTokBuf; // skip '(' 1260 1261 // Remove same number of characters from the end 1262 ThisTokEnd -= ThisTokBuf - Prefix; 1263 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal"); 1264 1265 // Copy the string over 1266 if (CopyStringFragment(StringToks[i], ThisTokBegin, 1267 StringRef(ThisTokBuf, ThisTokEnd - ThisTokBuf))) 1268 hadError = true; 1269 } else { 1270 if (ThisTokBuf[0] != '"') { 1271 // The file may have come from PCH and then changed after loading the 1272 // PCH; Fail gracefully. 1273 return DiagnoseLexingError(StringToks[i].getLocation()); 1274 } 1275 ++ThisTokBuf; // skip " 1276 1277 // Check if this is a pascal string 1278 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd && 1279 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') { 1280 1281 // If the \p sequence is found in the first token, we have a pascal string 1282 // Otherwise, if we already have a pascal string, ignore the first \p 1283 if (i == 0) { 1284 ++ThisTokBuf; 1285 Pascal = true; 1286 } else if (Pascal) 1287 ThisTokBuf += 2; 1288 } 1289 1290 while (ThisTokBuf != ThisTokEnd) { 1291 // Is this a span of non-escape characters? 1292 if (ThisTokBuf[0] != '\\') { 1293 const char *InStart = ThisTokBuf; 1294 do { 1295 ++ThisTokBuf; 1296 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\'); 1297 1298 // Copy the character span over. 1299 if (CopyStringFragment(StringToks[i], ThisTokBegin, 1300 StringRef(InStart, ThisTokBuf - InStart))) 1301 hadError = true; 1302 continue; 1303 } 1304 // Is this a Universal Character Name escape? 1305 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') { 1306 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, 1307 ResultPtr, hadError, 1308 FullSourceLoc(StringToks[i].getLocation(), SM), 1309 CharByteWidth, Diags, Features); 1310 continue; 1311 } 1312 // Otherwise, this is a non-UCN escape character. Process it. 1313 unsigned ResultChar = 1314 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError, 1315 FullSourceLoc(StringToks[i].getLocation(), SM), 1316 CharByteWidth*8, Diags, Features); 1317 1318 if (CharByteWidth == 4) { 1319 // FIXME: Make the type of the result buffer correct instead of 1320 // using reinterpret_cast. 1321 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultPtr); 1322 *ResultWidePtr = ResultChar; 1323 ResultPtr += 4; 1324 } else if (CharByteWidth == 2) { 1325 // FIXME: Make the type of the result buffer correct instead of 1326 // using reinterpret_cast. 1327 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultPtr); 1328 *ResultWidePtr = ResultChar & 0xFFFF; 1329 ResultPtr += 2; 1330 } else { 1331 assert(CharByteWidth == 1 && "Unexpected char width"); 1332 *ResultPtr++ = ResultChar & 0xFF; 1333 } 1334 } 1335 } 1336 } 1337 1338 if (Pascal) { 1339 if (CharByteWidth == 4) { 1340 // FIXME: Make the type of the result buffer correct instead of 1341 // using reinterpret_cast. 1342 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultBuf.data()); 1343 ResultWidePtr[0] = GetNumStringChars() - 1; 1344 } else if (CharByteWidth == 2) { 1345 // FIXME: Make the type of the result buffer correct instead of 1346 // using reinterpret_cast. 1347 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultBuf.data()); 1348 ResultWidePtr[0] = GetNumStringChars() - 1; 1349 } else { 1350 assert(CharByteWidth == 1 && "Unexpected char width"); 1351 ResultBuf[0] = GetNumStringChars() - 1; 1352 } 1353 1354 // Verify that pascal strings aren't too large. 1355 if (GetStringLength() > 256) { 1356 if (Diags) 1357 Diags->Report(StringToks[0].getLocation(), 1358 diag::err_pascal_string_too_long) 1359 << SourceRange(StringToks[0].getLocation(), 1360 StringToks[NumStringToks-1].getLocation()); 1361 hadError = true; 1362 return; 1363 } 1364 } else if (Diags) { 1365 // Complain if this string literal has too many characters. 1366 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509; 1367 1368 if (GetNumStringChars() > MaxChars) 1369 Diags->Report(StringToks[0].getLocation(), 1370 diag::ext_string_too_long) 1371 << GetNumStringChars() << MaxChars 1372 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0) 1373 << SourceRange(StringToks[0].getLocation(), 1374 StringToks[NumStringToks-1].getLocation()); 1375 } 1376 } 1377 1378 static const char *resyncUTF8(const char *Err, const char *End) { 1379 if (Err == End) 1380 return End; 1381 End = Err + std::min<unsigned>(getNumBytesForUTF8(*Err), End-Err); 1382 while (++Err != End && (*Err & 0xC0) == 0x80) 1383 ; 1384 return Err; 1385 } 1386 1387 /// \brief This function copies from Fragment, which is a sequence of bytes 1388 /// within Tok's contents (which begin at TokBegin) into ResultPtr. 1389 /// Performs widening for multi-byte characters. 1390 bool StringLiteralParser::CopyStringFragment(const Token &Tok, 1391 const char *TokBegin, 1392 StringRef Fragment) { 1393 const UTF8 *ErrorPtrTmp; 1394 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp)) 1395 return false; 1396 1397 // If we see bad encoding for unprefixed string literals, warn and 1398 // simply copy the byte values, for compatibility with gcc and older 1399 // versions of clang. 1400 bool NoErrorOnBadEncoding = isAscii(); 1401 if (NoErrorOnBadEncoding) { 1402 memcpy(ResultPtr, Fragment.data(), Fragment.size()); 1403 ResultPtr += Fragment.size(); 1404 } 1405 1406 if (Diags) { 1407 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1408 1409 FullSourceLoc SourceLoc(Tok.getLocation(), SM); 1410 const DiagnosticBuilder &Builder = 1411 Diag(Diags, Features, SourceLoc, TokBegin, 1412 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()), 1413 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding 1414 : diag::err_bad_string_encoding); 1415 1416 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1417 StringRef NextFragment(NextStart, Fragment.end()-NextStart); 1418 1419 // Decode into a dummy buffer. 1420 SmallString<512> Dummy; 1421 Dummy.reserve(Fragment.size() * CharByteWidth); 1422 char *Ptr = Dummy.data(); 1423 1424 while (!Builder.hasMaxRanges() && 1425 !ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) { 1426 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1427 NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1428 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin, 1429 ErrorPtr, NextStart); 1430 NextFragment = StringRef(NextStart, Fragment.end()-NextStart); 1431 } 1432 } 1433 return !NoErrorOnBadEncoding; 1434 } 1435 1436 void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) { 1437 hadError = true; 1438 if (Diags) 1439 Diags->Report(Loc, diag::err_lexing_string); 1440 } 1441 1442 /// getOffsetOfStringByte - This function returns the offset of the 1443 /// specified byte of the string data represented by Token. This handles 1444 /// advancing over escape sequences in the string. 1445 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok, 1446 unsigned ByteNo) const { 1447 // Get the spelling of the token. 1448 SmallString<32> SpellingBuffer; 1449 SpellingBuffer.resize(Tok.getLength()); 1450 1451 bool StringInvalid = false; 1452 const char *SpellingPtr = &SpellingBuffer[0]; 1453 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features, 1454 &StringInvalid); 1455 if (StringInvalid) 1456 return 0; 1457 1458 const char *SpellingStart = SpellingPtr; 1459 const char *SpellingEnd = SpellingPtr+TokLen; 1460 1461 // Handle UTF-8 strings just like narrow strings. 1462 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8') 1463 SpellingPtr += 2; 1464 1465 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' && 1466 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet"); 1467 1468 // For raw string literals, this is easy. 1469 if (SpellingPtr[0] == 'R') { 1470 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!"); 1471 // Skip 'R"'. 1472 SpellingPtr += 2; 1473 while (*SpellingPtr != '(') { 1474 ++SpellingPtr; 1475 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal"); 1476 } 1477 // Skip '('. 1478 ++SpellingPtr; 1479 return SpellingPtr - SpellingStart + ByteNo; 1480 } 1481 1482 // Skip over the leading quote 1483 assert(SpellingPtr[0] == '"' && "Should be a string literal!"); 1484 ++SpellingPtr; 1485 1486 // Skip over bytes until we find the offset we're looking for. 1487 while (ByteNo) { 1488 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!"); 1489 1490 // Step over non-escapes simply. 1491 if (*SpellingPtr != '\\') { 1492 ++SpellingPtr; 1493 --ByteNo; 1494 continue; 1495 } 1496 1497 // Otherwise, this is an escape character. Advance over it. 1498 bool HadError = false; 1499 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') { 1500 const char *EscapePtr = SpellingPtr; 1501 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd, 1502 1, Features, HadError); 1503 if (Len > ByteNo) { 1504 // ByteNo is somewhere within the escape sequence. 1505 SpellingPtr = EscapePtr; 1506 break; 1507 } 1508 ByteNo -= Len; 1509 } else { 1510 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError, 1511 FullSourceLoc(Tok.getLocation(), SM), 1512 CharByteWidth*8, Diags, Features); 1513 --ByteNo; 1514 } 1515 assert(!HadError && "This method isn't valid on erroneous strings"); 1516 } 1517 1518 return SpellingPtr-SpellingStart; 1519 } 1520