LiteralSupport.cpp revision 202879
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 || 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') && !PP.getLangOptions().CPlusPlus0x) { 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 // In C++0x, we cannot support hexadecmial floating literals because 476 // they conflict with user-defined literals, so we warn in previous 477 // versions of C++ by default. 478 if (PP.getLangOptions().CPlusPlus) 479 PP.Diag(TokLoc, diag::ext_hexconstant_cplusplus); 480 else if (!PP.getLangOptions().HexFloats) 481 PP.Diag(TokLoc, diag::ext_hexconstant_invalid); 482 } else if (saw_period) { 483 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 484 diag::err_hexconstant_requires_exponent); 485 hadError = true; 486 } 487 return; 488 } 489 490 // Handle simple binary numbers 0b01010 491 if (*s == 'b' || *s == 'B') { 492 // 0b101010 is a GCC extension. 493 PP.Diag(TokLoc, diag::ext_binary_literal); 494 ++s; 495 radix = 2; 496 DigitsBegin = s; 497 s = SkipBinaryDigits(s); 498 if (s == ThisTokEnd) { 499 // Done. 500 } else if (isxdigit(*s)) { 501 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 502 diag::err_invalid_binary_digit) << std::string(s, s+1); 503 hadError = true; 504 } 505 // Other suffixes will be diagnosed by the caller. 506 return; 507 } 508 509 // For now, the radix is set to 8. If we discover that we have a 510 // floating point constant, the radix will change to 10. Octal floating 511 // point constants are not permitted (only decimal and hexadecimal). 512 radix = 8; 513 DigitsBegin = s; 514 s = SkipOctalDigits(s); 515 if (s == ThisTokEnd) 516 return; // Done, simple octal number like 01234 517 518 // If we have some other non-octal digit that *is* a decimal digit, see if 519 // this is part of a floating point number like 094.123 or 09e1. 520 if (isdigit(*s)) { 521 const char *EndDecimal = SkipDigits(s); 522 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') { 523 s = EndDecimal; 524 radix = 10; 525 } 526 } 527 528 // If we have a hex digit other than 'e' (which denotes a FP exponent) then 529 // the code is using an incorrect base. 530 if (isxdigit(*s) && *s != 'e' && *s != 'E') { 531 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 532 diag::err_invalid_octal_digit) << std::string(s, s+1); 533 hadError = true; 534 return; 535 } 536 537 if (*s == '.') { 538 s++; 539 radix = 10; 540 saw_period = true; 541 s = SkipDigits(s); // Skip suffix. 542 } 543 if (*s == 'e' || *s == 'E') { // exponent 544 const char *Exponent = s; 545 s++; 546 radix = 10; 547 saw_exponent = true; 548 if (*s == '+' || *s == '-') s++; // sign 549 const char *first_non_digit = SkipDigits(s); 550 if (first_non_digit != s) { 551 s = first_non_digit; 552 } else { 553 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 554 diag::err_exponent_has_no_digits); 555 hadError = true; 556 return; 557 } 558 } 559} 560 561 562/// GetIntegerValue - Convert this numeric literal value to an APInt that 563/// matches Val's input width. If there is an overflow, set Val to the low bits 564/// of the result and return true. Otherwise, return false. 565bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) { 566 // Fast path: Compute a conservative bound on the maximum number of 567 // bits per digit in this radix. If we can't possibly overflow a 568 // uint64 based on that bound then do the simple conversion to 569 // integer. This avoids the expensive overflow checking below, and 570 // handles the common cases that matter (small decimal integers and 571 // hex/octal values which don't overflow). 572 unsigned MaxBitsPerDigit = 1; 573 while ((1U << MaxBitsPerDigit) < radix) 574 MaxBitsPerDigit += 1; 575 if ((SuffixBegin - DigitsBegin) * MaxBitsPerDigit <= 64) { 576 uint64_t N = 0; 577 for (s = DigitsBegin; s != SuffixBegin; ++s) 578 N = N*radix + HexDigitValue(*s); 579 580 // This will truncate the value to Val's input width. Simply check 581 // for overflow by comparing. 582 Val = N; 583 return Val.getZExtValue() != N; 584 } 585 586 Val = 0; 587 s = DigitsBegin; 588 589 llvm::APInt RadixVal(Val.getBitWidth(), radix); 590 llvm::APInt CharVal(Val.getBitWidth(), 0); 591 llvm::APInt OldVal = Val; 592 593 bool OverflowOccurred = false; 594 while (s < SuffixBegin) { 595 unsigned C = HexDigitValue(*s++); 596 597 // If this letter is out of bound for this radix, reject it. 598 assert(C < radix && "NumericLiteralParser ctor should have rejected this"); 599 600 CharVal = C; 601 602 // Add the digit to the value in the appropriate radix. If adding in digits 603 // made the value smaller, then this overflowed. 604 OldVal = Val; 605 606 // Multiply by radix, did overflow occur on the multiply? 607 Val *= RadixVal; 608 OverflowOccurred |= Val.udiv(RadixVal) != OldVal; 609 610 // Add value, did overflow occur on the value? 611 // (a + b) ult b <=> overflow 612 Val += CharVal; 613 OverflowOccurred |= Val.ult(CharVal); 614 } 615 return OverflowOccurred; 616} 617 618llvm::APFloat::opStatus 619NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) { 620 using llvm::APFloat; 621 using llvm::StringRef; 622 623 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin); 624 return Result.convertFromString(StringRef(ThisTokBegin, n), 625 APFloat::rmNearestTiesToEven); 626} 627 628 629CharLiteralParser::CharLiteralParser(const char *begin, const char *end, 630 SourceLocation Loc, Preprocessor &PP) { 631 // At this point we know that the character matches the regex "L?'.*'". 632 HadError = false; 633 634 // Determine if this is a wide character. 635 IsWide = begin[0] == 'L'; 636 if (IsWide) ++begin; 637 638 // Skip over the entry quote. 639 assert(begin[0] == '\'' && "Invalid token lexed"); 640 ++begin; 641 642 // FIXME: The "Value" is an uint64_t so we can handle char literals of 643 // upto 64-bits. 644 // FIXME: This extensively assumes that 'char' is 8-bits. 645 assert(PP.getTargetInfo().getCharWidth() == 8 && 646 "Assumes char is 8 bits"); 647 assert(PP.getTargetInfo().getIntWidth() <= 64 && 648 (PP.getTargetInfo().getIntWidth() & 7) == 0 && 649 "Assumes sizeof(int) on target is <= 64 and a multiple of char"); 650 assert(PP.getTargetInfo().getWCharWidth() <= 64 && 651 "Assumes sizeof(wchar) on target is <= 64"); 652 653 // This is what we will use for overflow detection 654 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0); 655 656 unsigned NumCharsSoFar = 0; 657 while (begin[0] != '\'') { 658 uint64_t ResultChar; 659 if (begin[0] != '\\') // If this is a normal character, consume it. 660 ResultChar = *begin++; 661 else // Otherwise, this is an escape character. 662 ResultChar = ProcessCharEscape(begin, end, HadError, Loc, IsWide, PP); 663 664 // If this is a multi-character constant (e.g. 'abc'), handle it. These are 665 // implementation defined (C99 6.4.4.4p10). 666 if (NumCharsSoFar) { 667 if (IsWide) { 668 // Emulate GCC's (unintentional?) behavior: L'ab' -> L'b'. 669 LitVal = 0; 670 } else { 671 // Narrow character literals act as though their value is concatenated 672 // in this implementation, but warn on overflow. 673 if (LitVal.countLeadingZeros() < 8) 674 PP.Diag(Loc, diag::warn_char_constant_too_large); 675 LitVal <<= 8; 676 } 677 } 678 679 LitVal = LitVal + ResultChar; 680 ++NumCharsSoFar; 681 } 682 683 // If this is the second character being processed, do special handling. 684 if (NumCharsSoFar > 1) { 685 // Warn about discarding the top bits for multi-char wide-character 686 // constants (L'abcd'). 687 if (IsWide) 688 PP.Diag(Loc, diag::warn_extraneous_wide_char_constant); 689 else if (NumCharsSoFar != 4) 690 PP.Diag(Loc, diag::ext_multichar_character_literal); 691 else 692 PP.Diag(Loc, diag::ext_four_char_character_literal); 693 IsMultiChar = true; 694 } else 695 IsMultiChar = false; 696 697 // Transfer the value from APInt to uint64_t 698 Value = LitVal.getZExtValue(); 699 700 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1") 701 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple 702 // character constants are not sign extended in the this implementation: 703 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC. 704 if (!IsWide && NumCharsSoFar == 1 && (Value & 128) && 705 PP.getLangOptions().CharIsSigned) 706 Value = (signed char)Value; 707} 708 709 710/// string-literal: [C99 6.4.5] 711/// " [s-char-sequence] " 712/// L" [s-char-sequence] " 713/// s-char-sequence: 714/// s-char 715/// s-char-sequence s-char 716/// s-char: 717/// any source character except the double quote ", 718/// backslash \, or newline character 719/// escape-character 720/// universal-character-name 721/// escape-character: [C99 6.4.4.4] 722/// \ escape-code 723/// universal-character-name 724/// escape-code: 725/// character-escape-code 726/// octal-escape-code 727/// hex-escape-code 728/// character-escape-code: one of 729/// n t b r f v a 730/// \ ' " ? 731/// octal-escape-code: 732/// octal-digit 733/// octal-digit octal-digit 734/// octal-digit octal-digit octal-digit 735/// hex-escape-code: 736/// x hex-digit 737/// hex-escape-code hex-digit 738/// universal-character-name: 739/// \u hex-quad 740/// \U hex-quad hex-quad 741/// hex-quad: 742/// hex-digit hex-digit hex-digit hex-digit 743/// 744StringLiteralParser:: 745StringLiteralParser(const Token *StringToks, unsigned NumStringToks, 746 Preprocessor &pp) : PP(pp) { 747 // Scan all of the string portions, remember the max individual token length, 748 // computing a bound on the concatenated string length, and see whether any 749 // piece is a wide-string. If any of the string portions is a wide-string 750 // literal, the result is a wide-string literal [C99 6.4.5p4]. 751 MaxTokenLength = StringToks[0].getLength(); 752 SizeBound = StringToks[0].getLength()-2; // -2 for "". 753 AnyWide = StringToks[0].is(tok::wide_string_literal); 754 755 hadError = false; 756 757 // Implement Translation Phase #6: concatenation of string literals 758 /// (C99 5.1.1.2p1). The common case is only one string fragment. 759 for (unsigned i = 1; i != NumStringToks; ++i) { 760 // The string could be shorter than this if it needs cleaning, but this is a 761 // reasonable bound, which is all we need. 762 SizeBound += StringToks[i].getLength()-2; // -2 for "". 763 764 // Remember maximum string piece length. 765 if (StringToks[i].getLength() > MaxTokenLength) 766 MaxTokenLength = StringToks[i].getLength(); 767 768 // Remember if we see any wide strings. 769 AnyWide |= StringToks[i].is(tok::wide_string_literal); 770 } 771 772 // Include space for the null terminator. 773 ++SizeBound; 774 775 // TODO: K&R warning: "traditional C rejects string constant concatenation" 776 777 // Get the width in bytes of wchar_t. If no wchar_t strings are used, do not 778 // query the target. As such, wchar_tByteWidth is only valid if AnyWide=true. 779 wchar_tByteWidth = ~0U; 780 if (AnyWide) { 781 wchar_tByteWidth = PP.getTargetInfo().getWCharWidth(); 782 assert((wchar_tByteWidth & 7) == 0 && "Assumes wchar_t is byte multiple!"); 783 wchar_tByteWidth /= 8; 784 } 785 786 // The output buffer size needs to be large enough to hold wide characters. 787 // This is a worst-case assumption which basically corresponds to L"" "long". 788 if (AnyWide) 789 SizeBound *= wchar_tByteWidth; 790 791 // Size the temporary buffer to hold the result string data. 792 ResultBuf.resize(SizeBound); 793 794 // Likewise, but for each string piece. 795 llvm::SmallString<512> TokenBuf; 796 TokenBuf.resize(MaxTokenLength); 797 798 // Loop over all the strings, getting their spelling, and expanding them to 799 // wide strings as appropriate. 800 ResultPtr = &ResultBuf[0]; // Next byte to fill in. 801 802 Pascal = false; 803 804 for (unsigned i = 0, e = NumStringToks; i != e; ++i) { 805 const char *ThisTokBuf = &TokenBuf[0]; 806 // Get the spelling of the token, which eliminates trigraphs, etc. We know 807 // that ThisTokBuf points to a buffer that is big enough for the whole token 808 // and 'spelled' tokens can only shrink. 809 unsigned ThisTokLen = PP.getSpelling(StringToks[i], ThisTokBuf); 810 const char *ThisTokEnd = ThisTokBuf+ThisTokLen-1; // Skip end quote. 811 812 // TODO: Input character set mapping support. 813 814 // Skip L marker for wide strings. 815 bool ThisIsWide = false; 816 if (ThisTokBuf[0] == 'L') { 817 ++ThisTokBuf; 818 ThisIsWide = true; 819 } 820 821 assert(ThisTokBuf[0] == '"' && "Expected quote, lexer broken?"); 822 ++ThisTokBuf; 823 824 // Check if this is a pascal string 825 if (pp.getLangOptions().PascalStrings && ThisTokBuf + 1 != ThisTokEnd && 826 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') { 827 828 // If the \p sequence is found in the first token, we have a pascal string 829 // Otherwise, if we already have a pascal string, ignore the first \p 830 if (i == 0) { 831 ++ThisTokBuf; 832 Pascal = true; 833 } else if (Pascal) 834 ThisTokBuf += 2; 835 } 836 837 while (ThisTokBuf != ThisTokEnd) { 838 // Is this a span of non-escape characters? 839 if (ThisTokBuf[0] != '\\') { 840 const char *InStart = ThisTokBuf; 841 do { 842 ++ThisTokBuf; 843 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\'); 844 845 // Copy the character span over. 846 unsigned Len = ThisTokBuf-InStart; 847 if (!AnyWide) { 848 memcpy(ResultPtr, InStart, Len); 849 ResultPtr += Len; 850 } else { 851 // Note: our internal rep of wide char tokens is always little-endian. 852 for (; Len; --Len, ++InStart) { 853 *ResultPtr++ = InStart[0]; 854 // Add zeros at the end. 855 for (unsigned i = 1, e = wchar_tByteWidth; i != e; ++i) 856 *ResultPtr++ = 0; 857 } 858 } 859 continue; 860 } 861 // Is this a Universal Character Name escape? 862 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') { 863 ProcessUCNEscape(ThisTokBuf, ThisTokEnd, ResultPtr, 864 hadError, StringToks[i].getLocation(), ThisIsWide, PP); 865 continue; 866 } 867 // Otherwise, this is a non-UCN escape character. Process it. 868 unsigned ResultChar = ProcessCharEscape(ThisTokBuf, ThisTokEnd, hadError, 869 StringToks[i].getLocation(), 870 ThisIsWide, PP); 871 872 // Note: our internal rep of wide char tokens is always little-endian. 873 *ResultPtr++ = ResultChar & 0xFF; 874 875 if (AnyWide) { 876 for (unsigned i = 1, e = wchar_tByteWidth; i != e; ++i) 877 *ResultPtr++ = ResultChar >> i*8; 878 } 879 } 880 } 881 882 if (Pascal) { 883 ResultBuf[0] = ResultPtr-&ResultBuf[0]-1; 884 885 // Verify that pascal strings aren't too large. 886 if (GetStringLength() > 256) { 887 PP.Diag(StringToks[0].getLocation(), diag::err_pascal_string_too_long) 888 << SourceRange(StringToks[0].getLocation(), 889 StringToks[NumStringToks-1].getLocation()); 890 hadError = 1; 891 return; 892 } 893 } 894} 895 896 897/// getOffsetOfStringByte - This function returns the offset of the 898/// specified byte of the string data represented by Token. This handles 899/// advancing over escape sequences in the string. 900unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok, 901 unsigned ByteNo, 902 Preprocessor &PP) { 903 // Get the spelling of the token. 904 llvm::SmallString<16> SpellingBuffer; 905 SpellingBuffer.resize(Tok.getLength()); 906 907 const char *SpellingPtr = &SpellingBuffer[0]; 908 unsigned TokLen = PP.getSpelling(Tok, SpellingPtr); 909 910 assert(SpellingPtr[0] != 'L' && "Doesn't handle wide strings yet"); 911 912 913 const char *SpellingStart = SpellingPtr; 914 const char *SpellingEnd = SpellingPtr+TokLen; 915 916 // Skip over the leading quote. 917 assert(SpellingPtr[0] == '"' && "Should be a string literal!"); 918 ++SpellingPtr; 919 920 // Skip over bytes until we find the offset we're looking for. 921 while (ByteNo) { 922 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!"); 923 924 // Step over non-escapes simply. 925 if (*SpellingPtr != '\\') { 926 ++SpellingPtr; 927 --ByteNo; 928 continue; 929 } 930 931 // Otherwise, this is an escape character. Advance over it. 932 bool HadError = false; 933 ProcessCharEscape(SpellingPtr, SpellingEnd, HadError, 934 Tok.getLocation(), false, PP); 935 assert(!HadError && "This method isn't valid on erroneous strings"); 936 --ByteNo; 937 } 938 939 return SpellingPtr-SpellingStart; 940} 941