LiteralSupport.cpp revision 201361
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) { 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 PP.Diag(Loc, diag::ext_nonstandard_escape) << "e"; 58 ResultChar = 27; 59 break; 60 case 'E': 61 PP.Diag(Loc, diag::ext_nonstandard_escape) << "E"; 62 ResultChar = 27; 63 break; 64 case 'f': 65 ResultChar = 12; 66 break; 67 case 'n': 68 ResultChar = 10; 69 break; 70 case 'r': 71 ResultChar = 13; 72 break; 73 case 't': 74 ResultChar = 9; 75 break; 76 case 'v': 77 ResultChar = 11; 78 break; 79 case 'x': { // Hex escape. 80 ResultChar = 0; 81 if (ThisTokBuf == ThisTokEnd || !isxdigit(*ThisTokBuf)) { 82 PP.Diag(Loc, diag::err_hex_escape_no_digits); 83 HadError = 1; 84 break; 85 } 86 87 // Hex escapes are a maximal series of hex digits. 88 bool Overflow = false; 89 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) { 90 int CharVal = HexDigitValue(ThisTokBuf[0]); 91 if (CharVal == -1) break; 92 // About to shift out a digit? 93 Overflow |= (ResultChar & 0xF0000000) ? true : false; 94 ResultChar <<= 4; 95 ResultChar |= CharVal; 96 } 97 98 // See if any bits will be truncated when evaluated as a character. 99 unsigned CharWidth = IsWide 100 ? PP.getTargetInfo().getWCharWidth() 101 : PP.getTargetInfo().getCharWidth(); 102 103 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 104 Overflow = true; 105 ResultChar &= ~0U >> (32-CharWidth); 106 } 107 108 // Check for overflow. 109 if (Overflow) // Too many digits to fit in 110 PP.Diag(Loc, diag::warn_hex_escape_too_large); 111 break; 112 } 113 case '0': case '1': case '2': case '3': 114 case '4': case '5': case '6': case '7': { 115 // Octal escapes. 116 --ThisTokBuf; 117 ResultChar = 0; 118 119 // Octal escapes are a series of octal digits with maximum length 3. 120 // "\0123" is a two digit sequence equal to "\012" "3". 121 unsigned NumDigits = 0; 122 do { 123 ResultChar <<= 3; 124 ResultChar |= *ThisTokBuf++ - '0'; 125 ++NumDigits; 126 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 && 127 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7'); 128 129 // Check for overflow. Reject '\777', but not L'\777'. 130 unsigned CharWidth = IsWide 131 ? PP.getTargetInfo().getWCharWidth() 132 : PP.getTargetInfo().getCharWidth(); 133 134 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 135 PP.Diag(Loc, diag::warn_octal_escape_too_large); 136 ResultChar &= ~0U >> (32-CharWidth); 137 } 138 break; 139 } 140 141 // Otherwise, these are not valid escapes. 142 case '(': case '{': case '[': case '%': 143 // GCC accepts these as extensions. We warn about them as such though. 144 PP.Diag(Loc, diag::ext_nonstandard_escape) 145 << std::string()+(char)ResultChar; 146 break; 147 default: 148 if (isgraph(ThisTokBuf[0])) 149 PP.Diag(Loc, diag::ext_unknown_escape) << std::string()+(char)ResultChar; 150 else 151 PP.Diag(Loc, diag::ext_unknown_escape) << "x"+llvm::utohexstr(ResultChar); 152 break; 153 } 154 155 return ResultChar; 156} 157 158/// ProcessUCNEscape - Read the Universal Character Name, check constraints and 159/// convert the UTF32 to UTF8. This is a subroutine of StringLiteralParser. 160/// When we decide to implement UCN's for character constants and identifiers, 161/// we will likely rework our support for UCN's. 162static void ProcessUCNEscape(const char *&ThisTokBuf, const char *ThisTokEnd, 163 char *&ResultBuf, bool &HadError, 164 SourceLocation Loc, bool IsWide, Preprocessor &PP) 165{ 166 // FIXME: Add a warning - UCN's are only valid in C++ & C99. 167 // FIXME: Handle wide strings. 168 169 // Save the beginning of the string (for error diagnostics). 170 const char *ThisTokBegin = ThisTokBuf; 171 172 // Skip the '\u' char's. 173 ThisTokBuf += 2; 174 175 if (ThisTokBuf == ThisTokEnd || !isxdigit(*ThisTokBuf)) { 176 PP.Diag(Loc, diag::err_ucn_escape_no_digits); 177 HadError = 1; 178 return; 179 } 180 typedef uint32_t UTF32; 181 182 UTF32 UcnVal = 0; 183 unsigned short UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8); 184 for (; ThisTokBuf != ThisTokEnd && UcnLen; ++ThisTokBuf, UcnLen--) { 185 int CharVal = HexDigitValue(ThisTokBuf[0]); 186 if (CharVal == -1) break; 187 UcnVal <<= 4; 188 UcnVal |= CharVal; 189 } 190 // If we didn't consume the proper number of digits, there is a problem. 191 if (UcnLen) { 192 PP.Diag(PP.AdvanceToTokenCharacter(Loc, ThisTokBuf-ThisTokBegin), 193 diag::err_ucn_escape_incomplete); 194 HadError = 1; 195 return; 196 } 197 // Check UCN constraints (C99 6.4.3p2). 198 if ((UcnVal < 0xa0 && 199 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60 )) // $, @, ` 200 || (UcnVal >= 0xD800 && UcnVal <= 0xDFFF) 201 || (UcnVal > 0x10FFFF)) /* the maximum legal UTF32 value */ { 202 PP.Diag(Loc, diag::err_ucn_escape_invalid); 203 HadError = 1; 204 return; 205 } 206 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8. 207 // The conversion below was inspired by: 208 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c 209 // First, we determine how many bytes the result will require. 210 typedef uint8_t UTF8; 211 212 unsigned short bytesToWrite = 0; 213 if (UcnVal < (UTF32)0x80) 214 bytesToWrite = 1; 215 else if (UcnVal < (UTF32)0x800) 216 bytesToWrite = 2; 217 else if (UcnVal < (UTF32)0x10000) 218 bytesToWrite = 3; 219 else 220 bytesToWrite = 4; 221 222 const unsigned byteMask = 0xBF; 223 const unsigned byteMark = 0x80; 224 225 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed 226 // into the first byte, depending on how many bytes follow. 227 static const UTF8 firstByteMark[5] = { 228 0x00, 0x00, 0xC0, 0xE0, 0xF0 229 }; 230 // Finally, we write the bytes into ResultBuf. 231 ResultBuf += bytesToWrite; 232 switch (bytesToWrite) { // note: everything falls through. 233 case 4: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 234 case 3: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 235 case 2: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 236 case 1: *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]); 237 } 238 // Update the buffer. 239 ResultBuf += bytesToWrite; 240} 241 242 243/// integer-constant: [C99 6.4.4.1] 244/// decimal-constant integer-suffix 245/// octal-constant integer-suffix 246/// hexadecimal-constant integer-suffix 247/// decimal-constant: 248/// nonzero-digit 249/// decimal-constant digit 250/// octal-constant: 251/// 0 252/// octal-constant octal-digit 253/// hexadecimal-constant: 254/// hexadecimal-prefix hexadecimal-digit 255/// hexadecimal-constant hexadecimal-digit 256/// hexadecimal-prefix: one of 257/// 0x 0X 258/// integer-suffix: 259/// unsigned-suffix [long-suffix] 260/// unsigned-suffix [long-long-suffix] 261/// long-suffix [unsigned-suffix] 262/// long-long-suffix [unsigned-sufix] 263/// nonzero-digit: 264/// 1 2 3 4 5 6 7 8 9 265/// octal-digit: 266/// 0 1 2 3 4 5 6 7 267/// hexadecimal-digit: 268/// 0 1 2 3 4 5 6 7 8 9 269/// a b c d e f 270/// A B C D E F 271/// unsigned-suffix: one of 272/// u U 273/// long-suffix: one of 274/// l L 275/// long-long-suffix: one of 276/// ll LL 277/// 278/// floating-constant: [C99 6.4.4.2] 279/// TODO: add rules... 280/// 281NumericLiteralParser:: 282NumericLiteralParser(const char *begin, const char *end, 283 SourceLocation TokLoc, Preprocessor &pp) 284 : PP(pp), ThisTokBegin(begin), ThisTokEnd(end) { 285 286 // This routine assumes that the range begin/end matches the regex for integer 287 // and FP constants (specifically, the 'pp-number' regex), and assumes that 288 // the byte at "*end" is both valid and not part of the regex. Because of 289 // this, it doesn't have to check for 'overscan' in various places. 290 assert(!isalnum(*end) && *end != '.' && *end != '_' && 291 "Lexer didn't maximally munch?"); 292 293 s = DigitsBegin = begin; 294 saw_exponent = false; 295 saw_period = false; 296 isLong = false; 297 isUnsigned = false; 298 isLongLong = false; 299 isFloat = false; 300 isImaginary = false; 301 isMicrosoftInteger = false; 302 hadError = false; 303 304 if (*s == '0') { // parse radix 305 ParseNumberStartingWithZero(TokLoc); 306 if (hadError) 307 return; 308 } else { // the first digit is non-zero 309 radix = 10; 310 s = SkipDigits(s); 311 if (s == ThisTokEnd) { 312 // Done. 313 } else if (isxdigit(*s) && !(*s == 'e' || *s == 'E')) { 314 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin), 315 diag::err_invalid_decimal_digit) << std::string(s, s+1); 316 hadError = true; 317 return; 318 } else if (*s == '.') { 319 s++; 320 saw_period = true; 321 s = SkipDigits(s); 322 } 323 if ((*s == 'e' || *s == 'E')) { // exponent 324 const char *Exponent = s; 325 s++; 326 saw_exponent = true; 327 if (*s == '+' || *s == '-') s++; // sign 328 const char *first_non_digit = SkipDigits(s); 329 if (first_non_digit != s) { 330 s = first_non_digit; 331 } else { 332 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-begin), 333 diag::err_exponent_has_no_digits); 334 hadError = true; 335 return; 336 } 337 } 338 } 339 340 SuffixBegin = s; 341 342 // Parse the suffix. At this point we can classify whether we have an FP or 343 // integer constant. 344 bool isFPConstant = isFloatingLiteral(); 345 346 // Loop over all of the characters of the suffix. If we see something bad, 347 // we break out of the loop. 348 for (; s != ThisTokEnd; ++s) { 349 switch (*s) { 350 case 'f': // FP Suffix for "float" 351 case 'F': 352 if (!isFPConstant) break; // Error for integer constant. 353 if (isFloat || isLong) break; // FF, LF invalid. 354 isFloat = true; 355 continue; // Success. 356 case 'u': 357 case 'U': 358 if (isFPConstant) break; // Error for floating constant. 359 if (isUnsigned) break; // Cannot be repeated. 360 isUnsigned = true; 361 continue; // Success. 362 case 'l': 363 case 'L': 364 if (isLong || isLongLong) break; // Cannot be repeated. 365 if (isFloat) break; // LF invalid. 366 367 // Check for long long. The L's need to be adjacent and the same case. 368 if (s+1 != ThisTokEnd && s[1] == s[0]) { 369 if (isFPConstant) break; // long long invalid for floats. 370 isLongLong = true; 371 ++s; // Eat both of them. 372 } else { 373 isLong = true; 374 } 375 continue; // Success. 376 case 'i': 377 if (PP.getLangOptions().Microsoft) { 378 if (isFPConstant || isUnsigned || isLong || isLongLong) break; 379 380 // Allow i8, i16, i32, i64, and i128. 381 if (s + 1 != ThisTokEnd) { 382 switch (s[1]) { 383 case '8': 384 s += 2; // i8 suffix 385 isMicrosoftInteger = true; 386 break; 387 case '1': 388 if (s + 2 == ThisTokEnd) break; 389 if (s[2] == '6') s += 3; // i16 suffix 390 else if (s[2] == '2') { 391 if (s + 3 == ThisTokEnd) break; 392 if (s[3] == '8') s += 4; // i128 suffix 393 } 394 isMicrosoftInteger = true; 395 break; 396 case '3': 397 if (s + 2 == ThisTokEnd) break; 398 if (s[2] == '2') s += 3; // i32 suffix 399 isMicrosoftInteger = true; 400 break; 401 case '6': 402 if (s + 2 == ThisTokEnd) break; 403 if (s[2] == '4') s += 3; // i64 suffix 404 isMicrosoftInteger = true; 405 break; 406 default: 407 break; 408 } 409 break; 410 } 411 } 412 // fall through. 413 case 'I': 414 case 'j': 415 case 'J': 416 if (isImaginary) break; // Cannot be repeated. 417 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin), 418 diag::ext_imaginary_constant); 419 isImaginary = true; 420 continue; // Success. 421 } 422 // If we reached here, there was an error. 423 break; 424 } 425 426 // Report an error if there are any. 427 if (s != ThisTokEnd) { 428 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin), 429 isFPConstant ? diag::err_invalid_suffix_float_constant : 430 diag::err_invalid_suffix_integer_constant) 431 << std::string(SuffixBegin, ThisTokEnd); 432 hadError = true; 433 return; 434 } 435} 436 437/// ParseNumberStartingWithZero - This method is called when the first character 438/// of the number is found to be a zero. This means it is either an octal 439/// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or 440/// a floating point number (01239.123e4). Eat the prefix, determining the 441/// radix etc. 442void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) { 443 assert(s[0] == '0' && "Invalid method call"); 444 s++; 445 446 // Handle a hex number like 0x1234. 447 if ((*s == 'x' || *s == 'X') && (isxdigit(s[1]) || s[1] == '.')) { 448 s++; 449 radix = 16; 450 DigitsBegin = s; 451 s = SkipHexDigits(s); 452 if (s == ThisTokEnd) { 453 // Done. 454 } else if (*s == '.') { 455 s++; 456 saw_period = true; 457 s = SkipHexDigits(s); 458 } 459 // A binary exponent can appear with or with a '.'. If dotted, the 460 // binary exponent is required. 461 if (*s == 'p' || *s == 'P') { 462 const char *Exponent = s; 463 s++; 464 saw_exponent = true; 465 if (*s == '+' || *s == '-') s++; // sign 466 const char *first_non_digit = SkipDigits(s); 467 if (first_non_digit == s) { 468 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 469 diag::err_exponent_has_no_digits); 470 hadError = true; 471 return; 472 } 473 s = first_non_digit; 474 475 if (!PP.getLangOptions().HexFloats) 476 PP.Diag(TokLoc, diag::ext_hexconstant_invalid); 477 } else if (saw_period) { 478 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 479 diag::err_hexconstant_requires_exponent); 480 hadError = true; 481 } 482 return; 483 } 484 485 // Handle simple binary numbers 0b01010 486 if (*s == 'b' || *s == 'B') { 487 // 0b101010 is a GCC extension. 488 PP.Diag(TokLoc, diag::ext_binary_literal); 489 ++s; 490 radix = 2; 491 DigitsBegin = s; 492 s = SkipBinaryDigits(s); 493 if (s == ThisTokEnd) { 494 // Done. 495 } else if (isxdigit(*s)) { 496 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 497 diag::err_invalid_binary_digit) << std::string(s, s+1); 498 hadError = true; 499 } 500 // Other suffixes will be diagnosed by the caller. 501 return; 502 } 503 504 // For now, the radix is set to 8. If we discover that we have a 505 // floating point constant, the radix will change to 10. Octal floating 506 // point constants are not permitted (only decimal and hexadecimal). 507 radix = 8; 508 DigitsBegin = s; 509 s = SkipOctalDigits(s); 510 if (s == ThisTokEnd) 511 return; // Done, simple octal number like 01234 512 513 // If we have some other non-octal digit that *is* a decimal digit, see if 514 // this is part of a floating point number like 094.123 or 09e1. 515 if (isdigit(*s)) { 516 const char *EndDecimal = SkipDigits(s); 517 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') { 518 s = EndDecimal; 519 radix = 10; 520 } 521 } 522 523 // If we have a hex digit other than 'e' (which denotes a FP exponent) then 524 // the code is using an incorrect base. 525 if (isxdigit(*s) && *s != 'e' && *s != 'E') { 526 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 527 diag::err_invalid_octal_digit) << std::string(s, s+1); 528 hadError = true; 529 return; 530 } 531 532 if (*s == '.') { 533 s++; 534 radix = 10; 535 saw_period = true; 536 s = SkipDigits(s); // Skip suffix. 537 } 538 if (*s == 'e' || *s == 'E') { // exponent 539 const char *Exponent = s; 540 s++; 541 radix = 10; 542 saw_exponent = true; 543 if (*s == '+' || *s == '-') s++; // sign 544 const char *first_non_digit = SkipDigits(s); 545 if (first_non_digit != s) { 546 s = first_non_digit; 547 } else { 548 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 549 diag::err_exponent_has_no_digits); 550 hadError = true; 551 return; 552 } 553 } 554} 555 556 557/// GetIntegerValue - Convert this numeric literal value to an APInt that 558/// matches Val's input width. If there is an overflow, set Val to the low bits 559/// of the result and return true. Otherwise, return false. 560bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) { 561 // Fast path: Compute a conservative bound on the maximum number of 562 // bits per digit in this radix. If we can't possibly overflow a 563 // uint64 based on that bound then do the simple conversion to 564 // integer. This avoids the expensive overflow checking below, and 565 // handles the common cases that matter (small decimal integers and 566 // hex/octal values which don't overflow). 567 unsigned MaxBitsPerDigit = 1; 568 while ((1U << MaxBitsPerDigit) < radix) 569 MaxBitsPerDigit += 1; 570 if ((SuffixBegin - DigitsBegin) * MaxBitsPerDigit <= 64) { 571 uint64_t N = 0; 572 for (s = DigitsBegin; s != SuffixBegin; ++s) 573 N = N*radix + HexDigitValue(*s); 574 575 // This will truncate the value to Val's input width. Simply check 576 // for overflow by comparing. 577 Val = N; 578 return Val.getZExtValue() != N; 579 } 580 581 Val = 0; 582 s = DigitsBegin; 583 584 llvm::APInt RadixVal(Val.getBitWidth(), radix); 585 llvm::APInt CharVal(Val.getBitWidth(), 0); 586 llvm::APInt OldVal = Val; 587 588 bool OverflowOccurred = false; 589 while (s < SuffixBegin) { 590 unsigned C = HexDigitValue(*s++); 591 592 // If this letter is out of bound for this radix, reject it. 593 assert(C < radix && "NumericLiteralParser ctor should have rejected this"); 594 595 CharVal = C; 596 597 // Add the digit to the value in the appropriate radix. If adding in digits 598 // made the value smaller, then this overflowed. 599 OldVal = Val; 600 601 // Multiply by radix, did overflow occur on the multiply? 602 Val *= RadixVal; 603 OverflowOccurred |= Val.udiv(RadixVal) != OldVal; 604 605 // Add value, did overflow occur on the value? 606 // (a + b) ult b <=> overflow 607 Val += CharVal; 608 OverflowOccurred |= Val.ult(CharVal); 609 } 610 return OverflowOccurred; 611} 612 613llvm::APFloat::opStatus 614NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) { 615 using llvm::APFloat; 616 using llvm::StringRef; 617 618 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin); 619 return Result.convertFromString(StringRef(ThisTokBegin, n), 620 APFloat::rmNearestTiesToEven); 621} 622 623 624CharLiteralParser::CharLiteralParser(const char *begin, const char *end, 625 SourceLocation Loc, Preprocessor &PP) { 626 // At this point we know that the character matches the regex "L?'.*'". 627 HadError = false; 628 629 // Determine if this is a wide character. 630 IsWide = begin[0] == 'L'; 631 if (IsWide) ++begin; 632 633 // Skip over the entry quote. 634 assert(begin[0] == '\'' && "Invalid token lexed"); 635 ++begin; 636 637 // FIXME: The "Value" is an uint64_t so we can handle char literals of 638 // upto 64-bits. 639 // FIXME: This extensively assumes that 'char' is 8-bits. 640 assert(PP.getTargetInfo().getCharWidth() == 8 && 641 "Assumes char is 8 bits"); 642 assert(PP.getTargetInfo().getIntWidth() <= 64 && 643 (PP.getTargetInfo().getIntWidth() & 7) == 0 && 644 "Assumes sizeof(int) on target is <= 64 and a multiple of char"); 645 assert(PP.getTargetInfo().getWCharWidth() <= 64 && 646 "Assumes sizeof(wchar) on target is <= 64"); 647 648 // This is what we will use for overflow detection 649 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0); 650 651 unsigned NumCharsSoFar = 0; 652 while (begin[0] != '\'') { 653 uint64_t ResultChar; 654 if (begin[0] != '\\') // If this is a normal character, consume it. 655 ResultChar = *begin++; 656 else // Otherwise, this is an escape character. 657 ResultChar = ProcessCharEscape(begin, end, HadError, Loc, IsWide, PP); 658 659 // If this is a multi-character constant (e.g. 'abc'), handle it. These are 660 // implementation defined (C99 6.4.4.4p10). 661 if (NumCharsSoFar) { 662 if (IsWide) { 663 // Emulate GCC's (unintentional?) behavior: L'ab' -> L'b'. 664 LitVal = 0; 665 } else { 666 // Narrow character literals act as though their value is concatenated 667 // in this implementation, but warn on overflow. 668 if (LitVal.countLeadingZeros() < 8) 669 PP.Diag(Loc, diag::warn_char_constant_too_large); 670 LitVal <<= 8; 671 } 672 } 673 674 LitVal = LitVal + ResultChar; 675 ++NumCharsSoFar; 676 } 677 678 // If this is the second character being processed, do special handling. 679 if (NumCharsSoFar > 1) { 680 // Warn about discarding the top bits for multi-char wide-character 681 // constants (L'abcd'). 682 if (IsWide) 683 PP.Diag(Loc, diag::warn_extraneous_wide_char_constant); 684 else if (NumCharsSoFar != 4) 685 PP.Diag(Loc, diag::ext_multichar_character_literal); 686 else 687 PP.Diag(Loc, diag::ext_four_char_character_literal); 688 IsMultiChar = true; 689 } else 690 IsMultiChar = false; 691 692 // Transfer the value from APInt to uint64_t 693 Value = LitVal.getZExtValue(); 694 695 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1") 696 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple 697 // character constants are not sign extended in the this implementation: 698 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC. 699 if (!IsWide && NumCharsSoFar == 1 && (Value & 128) && 700 PP.getLangOptions().CharIsSigned) 701 Value = (signed char)Value; 702} 703 704 705/// string-literal: [C99 6.4.5] 706/// " [s-char-sequence] " 707/// L" [s-char-sequence] " 708/// s-char-sequence: 709/// s-char 710/// s-char-sequence s-char 711/// s-char: 712/// any source character except the double quote ", 713/// backslash \, or newline character 714/// escape-character 715/// universal-character-name 716/// escape-character: [C99 6.4.4.4] 717/// \ escape-code 718/// universal-character-name 719/// escape-code: 720/// character-escape-code 721/// octal-escape-code 722/// hex-escape-code 723/// character-escape-code: one of 724/// n t b r f v a 725/// \ ' " ? 726/// octal-escape-code: 727/// octal-digit 728/// octal-digit octal-digit 729/// octal-digit octal-digit octal-digit 730/// hex-escape-code: 731/// x hex-digit 732/// hex-escape-code hex-digit 733/// universal-character-name: 734/// \u hex-quad 735/// \U hex-quad hex-quad 736/// hex-quad: 737/// hex-digit hex-digit hex-digit hex-digit 738/// 739StringLiteralParser:: 740StringLiteralParser(const Token *StringToks, unsigned NumStringToks, 741 Preprocessor &pp) : PP(pp) { 742 // Scan all of the string portions, remember the max individual token length, 743 // computing a bound on the concatenated string length, and see whether any 744 // piece is a wide-string. If any of the string portions is a wide-string 745 // literal, the result is a wide-string literal [C99 6.4.5p4]. 746 MaxTokenLength = StringToks[0].getLength(); 747 SizeBound = StringToks[0].getLength()-2; // -2 for "". 748 AnyWide = StringToks[0].is(tok::wide_string_literal); 749 750 hadError = false; 751 752 // Implement Translation Phase #6: concatenation of string literals 753 /// (C99 5.1.1.2p1). The common case is only one string fragment. 754 for (unsigned i = 1; i != NumStringToks; ++i) { 755 // The string could be shorter than this if it needs cleaning, but this is a 756 // reasonable bound, which is all we need. 757 SizeBound += StringToks[i].getLength()-2; // -2 for "". 758 759 // Remember maximum string piece length. 760 if (StringToks[i].getLength() > MaxTokenLength) 761 MaxTokenLength = StringToks[i].getLength(); 762 763 // Remember if we see any wide strings. 764 AnyWide |= StringToks[i].is(tok::wide_string_literal); 765 } 766 767 // Include space for the null terminator. 768 ++SizeBound; 769 770 // TODO: K&R warning: "traditional C rejects string constant concatenation" 771 772 // Get the width in bytes of wchar_t. If no wchar_t strings are used, do not 773 // query the target. As such, wchar_tByteWidth is only valid if AnyWide=true. 774 wchar_tByteWidth = ~0U; 775 if (AnyWide) { 776 wchar_tByteWidth = PP.getTargetInfo().getWCharWidth(); 777 assert((wchar_tByteWidth & 7) == 0 && "Assumes wchar_t is byte multiple!"); 778 wchar_tByteWidth /= 8; 779 } 780 781 // The output buffer size needs to be large enough to hold wide characters. 782 // This is a worst-case assumption which basically corresponds to L"" "long". 783 if (AnyWide) 784 SizeBound *= wchar_tByteWidth; 785 786 // Size the temporary buffer to hold the result string data. 787 ResultBuf.resize(SizeBound); 788 789 // Likewise, but for each string piece. 790 llvm::SmallString<512> TokenBuf; 791 TokenBuf.resize(MaxTokenLength); 792 793 // Loop over all the strings, getting their spelling, and expanding them to 794 // wide strings as appropriate. 795 ResultPtr = &ResultBuf[0]; // Next byte to fill in. 796 797 Pascal = false; 798 799 for (unsigned i = 0, e = NumStringToks; i != e; ++i) { 800 const char *ThisTokBuf = &TokenBuf[0]; 801 // Get the spelling of the token, which eliminates trigraphs, etc. We know 802 // that ThisTokBuf points to a buffer that is big enough for the whole token 803 // and 'spelled' tokens can only shrink. 804 unsigned ThisTokLen = PP.getSpelling(StringToks[i], ThisTokBuf); 805 const char *ThisTokEnd = ThisTokBuf+ThisTokLen-1; // Skip end quote. 806 807 // TODO: Input character set mapping support. 808 809 // Skip L marker for wide strings. 810 bool ThisIsWide = false; 811 if (ThisTokBuf[0] == 'L') { 812 ++ThisTokBuf; 813 ThisIsWide = true; 814 } 815 816 assert(ThisTokBuf[0] == '"' && "Expected quote, lexer broken?"); 817 ++ThisTokBuf; 818 819 // Check if this is a pascal string 820 if (pp.getLangOptions().PascalStrings && ThisTokBuf + 1 != ThisTokEnd && 821 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') { 822 823 // If the \p sequence is found in the first token, we have a pascal string 824 // Otherwise, if we already have a pascal string, ignore the first \p 825 if (i == 0) { 826 ++ThisTokBuf; 827 Pascal = true; 828 } else if (Pascal) 829 ThisTokBuf += 2; 830 } 831 832 while (ThisTokBuf != ThisTokEnd) { 833 // Is this a span of non-escape characters? 834 if (ThisTokBuf[0] != '\\') { 835 const char *InStart = ThisTokBuf; 836 do { 837 ++ThisTokBuf; 838 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\'); 839 840 // Copy the character span over. 841 unsigned Len = ThisTokBuf-InStart; 842 if (!AnyWide) { 843 memcpy(ResultPtr, InStart, Len); 844 ResultPtr += Len; 845 } else { 846 // Note: our internal rep of wide char tokens is always little-endian. 847 for (; Len; --Len, ++InStart) { 848 *ResultPtr++ = InStart[0]; 849 // Add zeros at the end. 850 for (unsigned i = 1, e = wchar_tByteWidth; i != e; ++i) 851 *ResultPtr++ = 0; 852 } 853 } 854 continue; 855 } 856 // Is this a Universal Character Name escape? 857 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') { 858 ProcessUCNEscape(ThisTokBuf, ThisTokEnd, ResultPtr, 859 hadError, StringToks[i].getLocation(), ThisIsWide, PP); 860 continue; 861 } 862 // Otherwise, this is a non-UCN escape character. Process it. 863 unsigned ResultChar = ProcessCharEscape(ThisTokBuf, ThisTokEnd, hadError, 864 StringToks[i].getLocation(), 865 ThisIsWide, PP); 866 867 // Note: our internal rep of wide char tokens is always little-endian. 868 *ResultPtr++ = ResultChar & 0xFF; 869 870 if (AnyWide) { 871 for (unsigned i = 1, e = wchar_tByteWidth; i != e; ++i) 872 *ResultPtr++ = ResultChar >> i*8; 873 } 874 } 875 } 876 877 if (Pascal) { 878 ResultBuf[0] = ResultPtr-&ResultBuf[0]-1; 879 880 // Verify that pascal strings aren't too large. 881 if (GetStringLength() > 256) { 882 PP.Diag(StringToks[0].getLocation(), diag::err_pascal_string_too_long) 883 << SourceRange(StringToks[0].getLocation(), 884 StringToks[NumStringToks-1].getLocation()); 885 hadError = 1; 886 return; 887 } 888 } 889} 890 891 892/// getOffsetOfStringByte - This function returns the offset of the 893/// specified byte of the string data represented by Token. This handles 894/// advancing over escape sequences in the string. 895unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok, 896 unsigned ByteNo, 897 Preprocessor &PP) { 898 // Get the spelling of the token. 899 llvm::SmallString<16> SpellingBuffer; 900 SpellingBuffer.resize(Tok.getLength()); 901 902 const char *SpellingPtr = &SpellingBuffer[0]; 903 unsigned TokLen = PP.getSpelling(Tok, SpellingPtr); 904 905 assert(SpellingPtr[0] != 'L' && "Doesn't handle wide strings yet"); 906 907 908 const char *SpellingStart = SpellingPtr; 909 const char *SpellingEnd = SpellingPtr+TokLen; 910 911 // Skip over the leading quote. 912 assert(SpellingPtr[0] == '"' && "Should be a string literal!"); 913 ++SpellingPtr; 914 915 // Skip over bytes until we find the offset we're looking for. 916 while (ByteNo) { 917 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!"); 918 919 // Step over non-escapes simply. 920 if (*SpellingPtr != '\\') { 921 ++SpellingPtr; 922 --ByteNo; 923 continue; 924 } 925 926 // Otherwise, this is an escape character. Advance over it. 927 bool HadError = false; 928 ProcessCharEscape(SpellingPtr, SpellingEnd, HadError, 929 Tok.getLocation(), false, PP); 930 assert(!HadError && "This method isn't valid on erroneous strings"); 931 --ByteNo; 932 } 933 934 return SpellingPtr-SpellingStart; 935} 936