LiteralSupport.cpp revision 208600
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/Lex/Preprocessor.h" 17#include "clang/Lex/LexDiagnostic.h" 18#include "clang/Basic/TargetInfo.h" 19#include "llvm/ADT/StringRef.h" 20#include "llvm/ADT/StringExtras.h" 21using namespace clang; 22 23/// HexDigitValue - Return the value of the specified hex digit, or -1 if it's 24/// not valid. 25static int HexDigitValue(char C) { 26 if (C >= '0' && C <= '9') return C-'0'; 27 if (C >= 'a' && C <= 'f') return C-'a'+10; 28 if (C >= 'A' && C <= 'F') return C-'A'+10; 29 return -1; 30} 31 32/// ProcessCharEscape - Parse a standard C escape sequence, which can occur in 33/// either a character or a string literal. 34static unsigned ProcessCharEscape(const char *&ThisTokBuf, 35 const char *ThisTokEnd, bool &HadError, 36 SourceLocation Loc, bool IsWide, 37 Preprocessor &PP, bool Complain) { 38 // Skip the '\' char. 39 ++ThisTokBuf; 40 41 // We know that this character can't be off the end of the buffer, because 42 // that would have been \", which would not have been the end of string. 43 unsigned ResultChar = *ThisTokBuf++; 44 switch (ResultChar) { 45 // These map to themselves. 46 case '\\': case '\'': case '"': case '?': break; 47 48 // These have fixed mappings. 49 case 'a': 50 // TODO: K&R: the meaning of '\\a' is different in traditional C 51 ResultChar = 7; 52 break; 53 case 'b': 54 ResultChar = 8; 55 break; 56 case 'e': 57 if (Complain) 58 PP.Diag(Loc, diag::ext_nonstandard_escape) << "e"; 59 ResultChar = 27; 60 break; 61 case 'E': 62 if (Complain) 63 PP.Diag(Loc, diag::ext_nonstandard_escape) << "E"; 64 ResultChar = 27; 65 break; 66 case 'f': 67 ResultChar = 12; 68 break; 69 case 'n': 70 ResultChar = 10; 71 break; 72 case 'r': 73 ResultChar = 13; 74 break; 75 case 't': 76 ResultChar = 9; 77 break; 78 case 'v': 79 ResultChar = 11; 80 break; 81 case 'x': { // Hex escape. 82 ResultChar = 0; 83 if (ThisTokBuf == ThisTokEnd || !isxdigit(*ThisTokBuf)) { 84 if (Complain) 85 PP.Diag(Loc, diag::err_hex_escape_no_digits); 86 HadError = 1; 87 break; 88 } 89 90 // Hex escapes are a maximal series of hex digits. 91 bool Overflow = false; 92 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) { 93 int CharVal = HexDigitValue(ThisTokBuf[0]); 94 if (CharVal == -1) break; 95 // About to shift out a digit? 96 Overflow |= (ResultChar & 0xF0000000) ? true : false; 97 ResultChar <<= 4; 98 ResultChar |= CharVal; 99 } 100 101 // See if any bits will be truncated when evaluated as a character. 102 unsigned CharWidth = IsWide 103 ? PP.getTargetInfo().getWCharWidth() 104 : PP.getTargetInfo().getCharWidth(); 105 106 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 107 Overflow = true; 108 ResultChar &= ~0U >> (32-CharWidth); 109 } 110 111 // Check for overflow. 112 if (Overflow && Complain) // Too many digits to fit in 113 PP.Diag(Loc, diag::warn_hex_escape_too_large); 114 break; 115 } 116 case '0': case '1': case '2': case '3': 117 case '4': case '5': case '6': case '7': { 118 // Octal escapes. 119 --ThisTokBuf; 120 ResultChar = 0; 121 122 // Octal escapes are a series of octal digits with maximum length 3. 123 // "\0123" is a two digit sequence equal to "\012" "3". 124 unsigned NumDigits = 0; 125 do { 126 ResultChar <<= 3; 127 ResultChar |= *ThisTokBuf++ - '0'; 128 ++NumDigits; 129 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 && 130 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7'); 131 132 // Check for overflow. Reject '\777', but not L'\777'. 133 unsigned CharWidth = IsWide 134 ? PP.getTargetInfo().getWCharWidth() 135 : PP.getTargetInfo().getCharWidth(); 136 137 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 138 if (Complain) 139 PP.Diag(Loc, diag::warn_octal_escape_too_large); 140 ResultChar &= ~0U >> (32-CharWidth); 141 } 142 break; 143 } 144 145 // Otherwise, these are not valid escapes. 146 case '(': case '{': case '[': case '%': 147 // GCC accepts these as extensions. We warn about them as such though. 148 if (Complain) 149 PP.Diag(Loc, diag::ext_nonstandard_escape) 150 << std::string()+(char)ResultChar; 151 break; 152 default: 153 if (!Complain) 154 break; 155 156 if (isgraph(ThisTokBuf[0])) 157 PP.Diag(Loc, diag::ext_unknown_escape) << std::string()+(char)ResultChar; 158 else 159 PP.Diag(Loc, diag::ext_unknown_escape) << "x"+llvm::utohexstr(ResultChar); 160 break; 161 } 162 163 return ResultChar; 164} 165 166/// ProcessUCNEscape - Read the Universal Character Name, check constraints and 167/// convert the UTF32 to UTF8. This is a subroutine of StringLiteralParser. 168/// When we decide to implement UCN's for character constants and identifiers, 169/// we will likely rework our support for UCN's. 170static void ProcessUCNEscape(const char *&ThisTokBuf, const char *ThisTokEnd, 171 char *&ResultBuf, bool &HadError, 172 SourceLocation Loc, bool IsWide, Preprocessor &PP, 173 bool Complain) 174{ 175 // FIXME: Add a warning - UCN's are only valid in C++ & C99. 176 // FIXME: Handle wide strings. 177 178 // Save the beginning of the string (for error diagnostics). 179 const char *ThisTokBegin = ThisTokBuf; 180 181 // Skip the '\u' char's. 182 ThisTokBuf += 2; 183 184 if (ThisTokBuf == ThisTokEnd || !isxdigit(*ThisTokBuf)) { 185 if (Complain) 186 PP.Diag(Loc, diag::err_ucn_escape_no_digits); 187 HadError = 1; 188 return; 189 } 190 typedef uint32_t UTF32; 191 192 UTF32 UcnVal = 0; 193 unsigned short UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8); 194 for (; ThisTokBuf != ThisTokEnd && UcnLen; ++ThisTokBuf, UcnLen--) { 195 int CharVal = HexDigitValue(ThisTokBuf[0]); 196 if (CharVal == -1) break; 197 UcnVal <<= 4; 198 UcnVal |= CharVal; 199 } 200 // If we didn't consume the proper number of digits, there is a problem. 201 if (UcnLen) { 202 if (Complain) 203 PP.Diag(PP.AdvanceToTokenCharacter(Loc, ThisTokBuf-ThisTokBegin), 204 diag::err_ucn_escape_incomplete); 205 HadError = 1; 206 return; 207 } 208 // Check UCN constraints (C99 6.4.3p2). 209 if ((UcnVal < 0xa0 && 210 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60 )) // $, @, ` 211 || (UcnVal >= 0xD800 && UcnVal <= 0xDFFF) 212 || (UcnVal > 0x10FFFF)) /* the maximum legal UTF32 value */ { 213 if (Complain) 214 PP.Diag(Loc, diag::err_ucn_escape_invalid); 215 HadError = 1; 216 return; 217 } 218 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8. 219 // The conversion below was inspired by: 220 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c 221 // First, we determine how many bytes the result will require. 222 typedef uint8_t UTF8; 223 224 unsigned short bytesToWrite = 0; 225 if (UcnVal < (UTF32)0x80) 226 bytesToWrite = 1; 227 else if (UcnVal < (UTF32)0x800) 228 bytesToWrite = 2; 229 else if (UcnVal < (UTF32)0x10000) 230 bytesToWrite = 3; 231 else 232 bytesToWrite = 4; 233 234 const unsigned byteMask = 0xBF; 235 const unsigned byteMark = 0x80; 236 237 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed 238 // into the first byte, depending on how many bytes follow. 239 static const UTF8 firstByteMark[5] = { 240 0x00, 0x00, 0xC0, 0xE0, 0xF0 241 }; 242 // Finally, we write the bytes into ResultBuf. 243 ResultBuf += bytesToWrite; 244 switch (bytesToWrite) { // note: everything falls through. 245 case 4: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 246 case 3: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 247 case 2: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 248 case 1: *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]); 249 } 250 // Update the buffer. 251 ResultBuf += bytesToWrite; 252} 253 254 255/// integer-constant: [C99 6.4.4.1] 256/// decimal-constant integer-suffix 257/// octal-constant integer-suffix 258/// hexadecimal-constant integer-suffix 259/// decimal-constant: 260/// nonzero-digit 261/// decimal-constant digit 262/// octal-constant: 263/// 0 264/// octal-constant octal-digit 265/// hexadecimal-constant: 266/// hexadecimal-prefix hexadecimal-digit 267/// hexadecimal-constant hexadecimal-digit 268/// hexadecimal-prefix: one of 269/// 0x 0X 270/// integer-suffix: 271/// unsigned-suffix [long-suffix] 272/// unsigned-suffix [long-long-suffix] 273/// long-suffix [unsigned-suffix] 274/// long-long-suffix [unsigned-sufix] 275/// nonzero-digit: 276/// 1 2 3 4 5 6 7 8 9 277/// octal-digit: 278/// 0 1 2 3 4 5 6 7 279/// hexadecimal-digit: 280/// 0 1 2 3 4 5 6 7 8 9 281/// a b c d e f 282/// A B C D E F 283/// unsigned-suffix: one of 284/// u U 285/// long-suffix: one of 286/// l L 287/// long-long-suffix: one of 288/// ll LL 289/// 290/// floating-constant: [C99 6.4.4.2] 291/// TODO: add rules... 292/// 293NumericLiteralParser:: 294NumericLiteralParser(const char *begin, const char *end, 295 SourceLocation TokLoc, Preprocessor &pp) 296 : PP(pp), ThisTokBegin(begin), ThisTokEnd(end) { 297 298 // This routine assumes that the range begin/end matches the regex for integer 299 // and FP constants (specifically, the 'pp-number' regex), and assumes that 300 // the byte at "*end" is both valid and not part of the regex. Because of 301 // this, it doesn't have to check for 'overscan' in various places. 302 assert(!isalnum(*end) && *end != '.' && *end != '_' && 303 "Lexer didn't maximally munch?"); 304 305 s = DigitsBegin = begin; 306 saw_exponent = false; 307 saw_period = false; 308 isLong = false; 309 isUnsigned = false; 310 isLongLong = false; 311 isFloat = false; 312 isImaginary = false; 313 isMicrosoftInteger = false; 314 hadError = false; 315 316 if (*s == '0') { // parse radix 317 ParseNumberStartingWithZero(TokLoc); 318 if (hadError) 319 return; 320 } else { // the first digit is non-zero 321 radix = 10; 322 s = SkipDigits(s); 323 if (s == ThisTokEnd) { 324 // Done. 325 } else if (isxdigit(*s) && !(*s == 'e' || *s == 'E')) { 326 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin), 327 diag::err_invalid_decimal_digit) << std::string(s, s+1); 328 hadError = true; 329 return; 330 } else if (*s == '.') { 331 s++; 332 saw_period = true; 333 s = SkipDigits(s); 334 } 335 if ((*s == 'e' || *s == 'E')) { // exponent 336 const char *Exponent = s; 337 s++; 338 saw_exponent = true; 339 if (*s == '+' || *s == '-') s++; // sign 340 const char *first_non_digit = SkipDigits(s); 341 if (first_non_digit != s) { 342 s = first_non_digit; 343 } else { 344 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-begin), 345 diag::err_exponent_has_no_digits); 346 hadError = true; 347 return; 348 } 349 } 350 } 351 352 SuffixBegin = s; 353 354 // Parse the suffix. At this point we can classify whether we have an FP or 355 // integer constant. 356 bool isFPConstant = isFloatingLiteral(); 357 358 // Loop over all of the characters of the suffix. If we see something bad, 359 // we break out of the loop. 360 for (; s != ThisTokEnd; ++s) { 361 switch (*s) { 362 case 'f': // FP Suffix for "float" 363 case 'F': 364 if (!isFPConstant) break; // Error for integer constant. 365 if (isFloat || isLong) break; // FF, LF invalid. 366 isFloat = true; 367 continue; // Success. 368 case 'u': 369 case 'U': 370 if (isFPConstant) break; // Error for floating constant. 371 if (isUnsigned) break; // Cannot be repeated. 372 isUnsigned = true; 373 continue; // Success. 374 case 'l': 375 case 'L': 376 if (isLong || isLongLong) break; // Cannot be repeated. 377 if (isFloat) break; // LF invalid. 378 379 // Check for long long. The L's need to be adjacent and the same case. 380 if (s+1 != ThisTokEnd && s[1] == s[0]) { 381 if (isFPConstant) break; // long long invalid for floats. 382 isLongLong = true; 383 ++s; // Eat both of them. 384 } else { 385 isLong = true; 386 } 387 continue; // Success. 388 case 'i': 389 if (PP.getLangOptions().Microsoft) { 390 if (isFPConstant || isLong || isLongLong) break; 391 392 // Allow i8, i16, i32, i64, and i128. 393 if (s + 1 != ThisTokEnd) { 394 switch (s[1]) { 395 case '8': 396 s += 2; // i8 suffix 397 isMicrosoftInteger = true; 398 break; 399 case '1': 400 if (s + 2 == ThisTokEnd) break; 401 if (s[2] == '6') s += 3; // i16 suffix 402 else if (s[2] == '2') { 403 if (s + 3 == ThisTokEnd) break; 404 if (s[3] == '8') s += 4; // i128 suffix 405 } 406 isMicrosoftInteger = true; 407 break; 408 case '3': 409 if (s + 2 == ThisTokEnd) break; 410 if (s[2] == '2') s += 3; // i32 suffix 411 isMicrosoftInteger = true; 412 break; 413 case '6': 414 if (s + 2 == ThisTokEnd) break; 415 if (s[2] == '4') s += 3; // i64 suffix 416 isMicrosoftInteger = true; 417 break; 418 default: 419 break; 420 } 421 break; 422 } 423 } 424 // fall through. 425 case 'I': 426 case 'j': 427 case 'J': 428 if (isImaginary) break; // Cannot be repeated. 429 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin), 430 diag::ext_imaginary_constant); 431 isImaginary = true; 432 continue; // Success. 433 } 434 // If we reached here, there was an error. 435 break; 436 } 437 438 // Report an error if there are any. 439 if (s != ThisTokEnd) { 440 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin), 441 isFPConstant ? diag::err_invalid_suffix_float_constant : 442 diag::err_invalid_suffix_integer_constant) 443 << std::string(SuffixBegin, ThisTokEnd); 444 hadError = true; 445 return; 446 } 447} 448 449/// ParseNumberStartingWithZero - This method is called when the first character 450/// of the number is found to be a zero. This means it is either an octal 451/// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or 452/// a floating point number (01239.123e4). Eat the prefix, determining the 453/// radix etc. 454void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) { 455 assert(s[0] == '0' && "Invalid method call"); 456 s++; 457 458 // Handle a hex number like 0x1234. 459 if ((*s == 'x' || *s == 'X') && (isxdigit(s[1]) || s[1] == '.')) { 460 s++; 461 radix = 16; 462 DigitsBegin = s; 463 s = SkipHexDigits(s); 464 if (s == ThisTokEnd) { 465 // Done. 466 } else if (*s == '.') { 467 s++; 468 saw_period = true; 469 s = SkipHexDigits(s); 470 } 471 // A binary exponent can appear with or with a '.'. If dotted, the 472 // binary exponent is required. 473 if ((*s == 'p' || *s == 'P') && !PP.getLangOptions().CPlusPlus0x) { 474 const char *Exponent = s; 475 s++; 476 saw_exponent = true; 477 if (*s == '+' || *s == '-') s++; // sign 478 const char *first_non_digit = SkipDigits(s); 479 if (first_non_digit == s) { 480 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 481 diag::err_exponent_has_no_digits); 482 hadError = true; 483 return; 484 } 485 s = first_non_digit; 486 487 // In C++0x, we cannot support hexadecmial floating literals because 488 // they conflict with user-defined literals, so we warn in previous 489 // versions of C++ by default. 490 if (PP.getLangOptions().CPlusPlus) 491 PP.Diag(TokLoc, diag::ext_hexconstant_cplusplus); 492 else if (!PP.getLangOptions().HexFloats) 493 PP.Diag(TokLoc, diag::ext_hexconstant_invalid); 494 } else if (saw_period) { 495 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 496 diag::err_hexconstant_requires_exponent); 497 hadError = true; 498 } 499 return; 500 } 501 502 // Handle simple binary numbers 0b01010 503 if (*s == 'b' || *s == 'B') { 504 // 0b101010 is a GCC extension. 505 PP.Diag(TokLoc, diag::ext_binary_literal); 506 ++s; 507 radix = 2; 508 DigitsBegin = s; 509 s = SkipBinaryDigits(s); 510 if (s == ThisTokEnd) { 511 // Done. 512 } else if (isxdigit(*s)) { 513 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 514 diag::err_invalid_binary_digit) << std::string(s, s+1); 515 hadError = true; 516 } 517 // Other suffixes will be diagnosed by the caller. 518 return; 519 } 520 521 // For now, the radix is set to 8. If we discover that we have a 522 // floating point constant, the radix will change to 10. Octal floating 523 // point constants are not permitted (only decimal and hexadecimal). 524 radix = 8; 525 DigitsBegin = s; 526 s = SkipOctalDigits(s); 527 if (s == ThisTokEnd) 528 return; // Done, simple octal number like 01234 529 530 // If we have some other non-octal digit that *is* a decimal digit, see if 531 // this is part of a floating point number like 094.123 or 09e1. 532 if (isdigit(*s)) { 533 const char *EndDecimal = SkipDigits(s); 534 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') { 535 s = EndDecimal; 536 radix = 10; 537 } 538 } 539 540 // If we have a hex digit other than 'e' (which denotes a FP exponent) then 541 // the code is using an incorrect base. 542 if (isxdigit(*s) && *s != 'e' && *s != 'E') { 543 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 544 diag::err_invalid_octal_digit) << std::string(s, s+1); 545 hadError = true; 546 return; 547 } 548 549 if (*s == '.') { 550 s++; 551 radix = 10; 552 saw_period = true; 553 s = SkipDigits(s); // Skip suffix. 554 } 555 if (*s == 'e' || *s == 'E') { // exponent 556 const char *Exponent = s; 557 s++; 558 radix = 10; 559 saw_exponent = true; 560 if (*s == '+' || *s == '-') s++; // sign 561 const char *first_non_digit = SkipDigits(s); 562 if (first_non_digit != s) { 563 s = first_non_digit; 564 } else { 565 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 566 diag::err_exponent_has_no_digits); 567 hadError = true; 568 return; 569 } 570 } 571} 572 573 574/// GetIntegerValue - Convert this numeric literal value to an APInt that 575/// matches Val's input width. If there is an overflow, set Val to the low bits 576/// of the result and return true. Otherwise, return false. 577bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) { 578 // Fast path: Compute a conservative bound on the maximum number of 579 // bits per digit in this radix. If we can't possibly overflow a 580 // uint64 based on that bound then do the simple conversion to 581 // integer. This avoids the expensive overflow checking below, and 582 // handles the common cases that matter (small decimal integers and 583 // hex/octal values which don't overflow). 584 unsigned MaxBitsPerDigit = 1; 585 while ((1U << MaxBitsPerDigit) < radix) 586 MaxBitsPerDigit += 1; 587 if ((SuffixBegin - DigitsBegin) * MaxBitsPerDigit <= 64) { 588 uint64_t N = 0; 589 for (s = DigitsBegin; s != SuffixBegin; ++s) 590 N = N*radix + HexDigitValue(*s); 591 592 // This will truncate the value to Val's input width. Simply check 593 // for overflow by comparing. 594 Val = N; 595 return Val.getZExtValue() != N; 596 } 597 598 Val = 0; 599 s = DigitsBegin; 600 601 llvm::APInt RadixVal(Val.getBitWidth(), radix); 602 llvm::APInt CharVal(Val.getBitWidth(), 0); 603 llvm::APInt OldVal = Val; 604 605 bool OverflowOccurred = false; 606 while (s < SuffixBegin) { 607 unsigned C = HexDigitValue(*s++); 608 609 // If this letter is out of bound for this radix, reject it. 610 assert(C < radix && "NumericLiteralParser ctor should have rejected this"); 611 612 CharVal = C; 613 614 // Add the digit to the value in the appropriate radix. If adding in digits 615 // made the value smaller, then this overflowed. 616 OldVal = Val; 617 618 // Multiply by radix, did overflow occur on the multiply? 619 Val *= RadixVal; 620 OverflowOccurred |= Val.udiv(RadixVal) != OldVal; 621 622 // Add value, did overflow occur on the value? 623 // (a + b) ult b <=> overflow 624 Val += CharVal; 625 OverflowOccurred |= Val.ult(CharVal); 626 } 627 return OverflowOccurred; 628} 629 630llvm::APFloat::opStatus 631NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) { 632 using llvm::APFloat; 633 using llvm::StringRef; 634 635 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin); 636 return Result.convertFromString(StringRef(ThisTokBegin, n), 637 APFloat::rmNearestTiesToEven); 638} 639 640 641CharLiteralParser::CharLiteralParser(const char *begin, const char *end, 642 SourceLocation Loc, Preprocessor &PP) { 643 // At this point we know that the character matches the regex "L?'.*'". 644 HadError = false; 645 646 // Determine if this is a wide character. 647 IsWide = begin[0] == 'L'; 648 if (IsWide) ++begin; 649 650 // Skip over the entry quote. 651 assert(begin[0] == '\'' && "Invalid token lexed"); 652 ++begin; 653 654 // FIXME: The "Value" is an uint64_t so we can handle char literals of 655 // upto 64-bits. 656 // FIXME: This extensively assumes that 'char' is 8-bits. 657 assert(PP.getTargetInfo().getCharWidth() == 8 && 658 "Assumes char is 8 bits"); 659 assert(PP.getTargetInfo().getIntWidth() <= 64 && 660 (PP.getTargetInfo().getIntWidth() & 7) == 0 && 661 "Assumes sizeof(int) on target is <= 64 and a multiple of char"); 662 assert(PP.getTargetInfo().getWCharWidth() <= 64 && 663 "Assumes sizeof(wchar) on target is <= 64"); 664 665 // This is what we will use for overflow detection 666 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0); 667 668 unsigned NumCharsSoFar = 0; 669 bool Warned = false; 670 while (begin[0] != '\'') { 671 uint64_t ResultChar; 672 if (begin[0] != '\\') // If this is a normal character, consume it. 673 ResultChar = *begin++; 674 else // Otherwise, this is an escape character. 675 ResultChar = ProcessCharEscape(begin, end, HadError, Loc, IsWide, PP, 676 /*Complain=*/true); 677 678 // If this is a multi-character constant (e.g. 'abc'), handle it. These are 679 // implementation defined (C99 6.4.4.4p10). 680 if (NumCharsSoFar) { 681 if (IsWide) { 682 // Emulate GCC's (unintentional?) behavior: L'ab' -> L'b'. 683 LitVal = 0; 684 } else { 685 // Narrow character literals act as though their value is concatenated 686 // in this implementation, but warn on overflow. 687 if (LitVal.countLeadingZeros() < 8 && !Warned) { 688 PP.Diag(Loc, diag::warn_char_constant_too_large); 689 Warned = true; 690 } 691 LitVal <<= 8; 692 } 693 } 694 695 LitVal = LitVal + ResultChar; 696 ++NumCharsSoFar; 697 } 698 699 // If this is the second character being processed, do special handling. 700 if (NumCharsSoFar > 1) { 701 // Warn about discarding the top bits for multi-char wide-character 702 // constants (L'abcd'). 703 if (IsWide) 704 PP.Diag(Loc, diag::warn_extraneous_wide_char_constant); 705 else if (NumCharsSoFar != 4) 706 PP.Diag(Loc, diag::ext_multichar_character_literal); 707 else 708 PP.Diag(Loc, diag::ext_four_char_character_literal); 709 IsMultiChar = true; 710 } else 711 IsMultiChar = false; 712 713 // Transfer the value from APInt to uint64_t 714 Value = LitVal.getZExtValue(); 715 716 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1") 717 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple 718 // character constants are not sign extended in the this implementation: 719 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC. 720 if (!IsWide && NumCharsSoFar == 1 && (Value & 128) && 721 PP.getLangOptions().CharIsSigned) 722 Value = (signed char)Value; 723} 724 725 726/// string-literal: [C99 6.4.5] 727/// " [s-char-sequence] " 728/// L" [s-char-sequence] " 729/// s-char-sequence: 730/// s-char 731/// s-char-sequence s-char 732/// s-char: 733/// any source character except the double quote ", 734/// backslash \, or newline character 735/// escape-character 736/// universal-character-name 737/// escape-character: [C99 6.4.4.4] 738/// \ escape-code 739/// universal-character-name 740/// escape-code: 741/// character-escape-code 742/// octal-escape-code 743/// hex-escape-code 744/// character-escape-code: one of 745/// n t b r f v a 746/// \ ' " ? 747/// octal-escape-code: 748/// octal-digit 749/// octal-digit octal-digit 750/// octal-digit octal-digit octal-digit 751/// hex-escape-code: 752/// x hex-digit 753/// hex-escape-code hex-digit 754/// universal-character-name: 755/// \u hex-quad 756/// \U hex-quad hex-quad 757/// hex-quad: 758/// hex-digit hex-digit hex-digit hex-digit 759/// 760StringLiteralParser:: 761StringLiteralParser(const Token *StringToks, unsigned NumStringToks, 762 Preprocessor &pp, bool Complain) : PP(pp) { 763 // Scan all of the string portions, remember the max individual token length, 764 // computing a bound on the concatenated string length, and see whether any 765 // piece is a wide-string. If any of the string portions is a wide-string 766 // literal, the result is a wide-string literal [C99 6.4.5p4]. 767 MaxTokenLength = StringToks[0].getLength(); 768 SizeBound = StringToks[0].getLength()-2; // -2 for "". 769 AnyWide = StringToks[0].is(tok::wide_string_literal); 770 771 hadError = false; 772 773 // Implement Translation Phase #6: concatenation of string literals 774 /// (C99 5.1.1.2p1). The common case is only one string fragment. 775 for (unsigned i = 1; i != NumStringToks; ++i) { 776 // The string could be shorter than this if it needs cleaning, but this is a 777 // reasonable bound, which is all we need. 778 SizeBound += StringToks[i].getLength()-2; // -2 for "". 779 780 // Remember maximum string piece length. 781 if (StringToks[i].getLength() > MaxTokenLength) 782 MaxTokenLength = StringToks[i].getLength(); 783 784 // Remember if we see any wide strings. 785 AnyWide |= StringToks[i].is(tok::wide_string_literal); 786 } 787 788 // Include space for the null terminator. 789 ++SizeBound; 790 791 // TODO: K&R warning: "traditional C rejects string constant concatenation" 792 793 // Get the width in bytes of wchar_t. If no wchar_t strings are used, do not 794 // query the target. As such, wchar_tByteWidth is only valid if AnyWide=true. 795 wchar_tByteWidth = ~0U; 796 if (AnyWide) { 797 wchar_tByteWidth = PP.getTargetInfo().getWCharWidth(); 798 assert((wchar_tByteWidth & 7) == 0 && "Assumes wchar_t is byte multiple!"); 799 wchar_tByteWidth /= 8; 800 } 801 802 // The output buffer size needs to be large enough to hold wide characters. 803 // This is a worst-case assumption which basically corresponds to L"" "long". 804 if (AnyWide) 805 SizeBound *= wchar_tByteWidth; 806 807 // Size the temporary buffer to hold the result string data. 808 ResultBuf.resize(SizeBound); 809 810 // Likewise, but for each string piece. 811 llvm::SmallString<512> TokenBuf; 812 TokenBuf.resize(MaxTokenLength); 813 814 // Loop over all the strings, getting their spelling, and expanding them to 815 // wide strings as appropriate. 816 ResultPtr = &ResultBuf[0]; // Next byte to fill in. 817 818 Pascal = false; 819 820 for (unsigned i = 0, e = NumStringToks; i != e; ++i) { 821 const char *ThisTokBuf = &TokenBuf[0]; 822 // Get the spelling of the token, which eliminates trigraphs, etc. We know 823 // that ThisTokBuf points to a buffer that is big enough for the whole token 824 // and 'spelled' tokens can only shrink. 825 bool StringInvalid = false; 826 unsigned ThisTokLen = PP.getSpelling(StringToks[i], ThisTokBuf, 827 &StringInvalid); 828 if (StringInvalid) { 829 hadError = 1; 830 continue; 831 } 832 833 const char *ThisTokEnd = ThisTokBuf+ThisTokLen-1; // Skip end quote. 834 835 // TODO: Input character set mapping support. 836 837 // Skip L marker for wide strings. 838 bool ThisIsWide = false; 839 if (ThisTokBuf[0] == 'L') { 840 ++ThisTokBuf; 841 ThisIsWide = true; 842 } 843 844 assert(ThisTokBuf[0] == '"' && "Expected quote, lexer broken?"); 845 ++ThisTokBuf; 846 847 // Check if this is a pascal string 848 if (pp.getLangOptions().PascalStrings && ThisTokBuf + 1 != ThisTokEnd && 849 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') { 850 851 // If the \p sequence is found in the first token, we have a pascal string 852 // Otherwise, if we already have a pascal string, ignore the first \p 853 if (i == 0) { 854 ++ThisTokBuf; 855 Pascal = true; 856 } else if (Pascal) 857 ThisTokBuf += 2; 858 } 859 860 while (ThisTokBuf != ThisTokEnd) { 861 // Is this a span of non-escape characters? 862 if (ThisTokBuf[0] != '\\') { 863 const char *InStart = ThisTokBuf; 864 do { 865 ++ThisTokBuf; 866 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\'); 867 868 // Copy the character span over. 869 unsigned Len = ThisTokBuf-InStart; 870 if (!AnyWide) { 871 memcpy(ResultPtr, InStart, Len); 872 ResultPtr += Len; 873 } else { 874 // Note: our internal rep of wide char tokens is always little-endian. 875 for (; Len; --Len, ++InStart) { 876 *ResultPtr++ = InStart[0]; 877 // Add zeros at the end. 878 for (unsigned i = 1, e = wchar_tByteWidth; i != e; ++i) 879 *ResultPtr++ = 0; 880 } 881 } 882 continue; 883 } 884 // Is this a Universal Character Name escape? 885 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') { 886 ProcessUCNEscape(ThisTokBuf, ThisTokEnd, ResultPtr, 887 hadError, StringToks[i].getLocation(), ThisIsWide, PP, 888 Complain); 889 continue; 890 } 891 // Otherwise, this is a non-UCN escape character. Process it. 892 unsigned ResultChar = ProcessCharEscape(ThisTokBuf, ThisTokEnd, hadError, 893 StringToks[i].getLocation(), 894 ThisIsWide, PP, Complain); 895 896 // Note: our internal rep of wide char tokens is always little-endian. 897 *ResultPtr++ = ResultChar & 0xFF; 898 899 if (AnyWide) { 900 for (unsigned i = 1, e = wchar_tByteWidth; i != e; ++i) 901 *ResultPtr++ = ResultChar >> i*8; 902 } 903 } 904 } 905 906 if (Pascal) { 907 ResultBuf[0] = ResultPtr-&ResultBuf[0]-1; 908 909 // Verify that pascal strings aren't too large. 910 if (GetStringLength() > 256 && Complain) { 911 PP.Diag(StringToks[0].getLocation(), diag::err_pascal_string_too_long) 912 << SourceRange(StringToks[0].getLocation(), 913 StringToks[NumStringToks-1].getLocation()); 914 hadError = 1; 915 return; 916 } 917 } 918} 919 920 921/// getOffsetOfStringByte - This function returns the offset of the 922/// specified byte of the string data represented by Token. This handles 923/// advancing over escape sequences in the string. 924unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok, 925 unsigned ByteNo, 926 Preprocessor &PP, 927 bool Complain) { 928 // Get the spelling of the token. 929 llvm::SmallString<16> SpellingBuffer; 930 SpellingBuffer.resize(Tok.getLength()); 931 932 bool StringInvalid = false; 933 const char *SpellingPtr = &SpellingBuffer[0]; 934 unsigned TokLen = PP.getSpelling(Tok, SpellingPtr, &StringInvalid); 935 if (StringInvalid) { 936 return 0; 937 } 938 939 assert(SpellingPtr[0] != 'L' && "Doesn't handle wide strings yet"); 940 941 942 const char *SpellingStart = SpellingPtr; 943 const char *SpellingEnd = SpellingPtr+TokLen; 944 945 // Skip over the leading quote. 946 assert(SpellingPtr[0] == '"' && "Should be a string literal!"); 947 ++SpellingPtr; 948 949 // Skip over bytes until we find the offset we're looking for. 950 while (ByteNo) { 951 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!"); 952 953 // Step over non-escapes simply. 954 if (*SpellingPtr != '\\') { 955 ++SpellingPtr; 956 --ByteNo; 957 continue; 958 } 959 960 // Otherwise, this is an escape character. Advance over it. 961 bool HadError = false; 962 ProcessCharEscape(SpellingPtr, SpellingEnd, HadError, 963 Tok.getLocation(), false, PP, Complain); 964 assert(!HadError && "This method isn't valid on erroneous strings"); 965 --ByteNo; 966 } 967 968 return SpellingPtr-SpellingStart; 969} 970