1 //===--- LiteralSupport.cpp - Code to parse and process literals ----------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements the NumericLiteralParser, CharLiteralParser, and 11 // StringLiteralParser interfaces. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "clang/Lex/LiteralSupport.h" 16 #include "clang/Basic/CharInfo.h" 17 #include "clang/Basic/TargetInfo.h" 18 #include "clang/Lex/LexDiagnostic.h" 19 #include "clang/Lex/Preprocessor.h" 20 #include "llvm/ADT/StringExtras.h" 21 #include "llvm/Support/ConvertUTF.h" 22 #include "llvm/Support/ErrorHandling.h" 23 24 using namespace clang; 25 26 static unsigned getCharWidth(tok::TokenKind kind, const TargetInfo &Target) { 27 switch (kind) { 28 default: llvm_unreachable("Unknown token type!"); 29 case tok::char_constant: 30 case tok::string_literal: 31 case tok::utf8_char_constant: 32 case tok::utf8_string_literal: 33 return Target.getCharWidth(); 34 case tok::wide_char_constant: 35 case tok::wide_string_literal: 36 return Target.getWCharWidth(); 37 case tok::utf16_char_constant: 38 case tok::utf16_string_literal: 39 return Target.getChar16Width(); 40 case tok::utf32_char_constant: 41 case tok::utf32_string_literal: 42 return Target.getChar32Width(); 43 } 44 } 45 46 static CharSourceRange MakeCharSourceRange(const LangOptions &Features, 47 FullSourceLoc TokLoc, 48 const char *TokBegin, 49 const char *TokRangeBegin, 50 const char *TokRangeEnd) { 51 SourceLocation Begin = 52 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin, 53 TokLoc.getManager(), Features); 54 SourceLocation End = 55 Lexer::AdvanceToTokenCharacter(Begin, TokRangeEnd - TokRangeBegin, 56 TokLoc.getManager(), Features); 57 return CharSourceRange::getCharRange(Begin, End); 58 } 59 60 /// \brief Produce a diagnostic highlighting some portion of a literal. 61 /// 62 /// Emits the diagnostic \p DiagID, highlighting the range of characters from 63 /// \p TokRangeBegin (inclusive) to \p TokRangeEnd (exclusive), which must be 64 /// a substring of a spelling buffer for the token beginning at \p TokBegin. 65 static DiagnosticBuilder Diag(DiagnosticsEngine *Diags, 66 const LangOptions &Features, FullSourceLoc TokLoc, 67 const char *TokBegin, const char *TokRangeBegin, 68 const char *TokRangeEnd, unsigned DiagID) { 69 SourceLocation Begin = 70 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin, 71 TokLoc.getManager(), Features); 72 return Diags->Report(Begin, DiagID) << 73 MakeCharSourceRange(Features, TokLoc, TokBegin, TokRangeBegin, TokRangeEnd); 74 } 75 76 /// ProcessCharEscape - Parse a standard C escape sequence, which can occur in 77 /// either a character or a string literal. 78 static unsigned ProcessCharEscape(const char *ThisTokBegin, 79 const char *&ThisTokBuf, 80 const char *ThisTokEnd, bool &HadError, 81 FullSourceLoc Loc, unsigned CharWidth, 82 DiagnosticsEngine *Diags, 83 const LangOptions &Features) { 84 const char *EscapeBegin = ThisTokBuf; 85 86 // Skip the '\' char. 87 ++ThisTokBuf; 88 89 // We know that this character can't be off the end of the buffer, because 90 // that would have been \", which would not have been the end of string. 91 unsigned ResultChar = *ThisTokBuf++; 92 switch (ResultChar) { 93 // These map to themselves. 94 case '\\': case '\'': case '"': case '?': break; 95 96 // These have fixed mappings. 97 case 'a': 98 // TODO: K&R: the meaning of '\\a' is different in traditional C 99 ResultChar = 7; 100 break; 101 case 'b': 102 ResultChar = 8; 103 break; 104 case 'e': 105 if (Diags) 106 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 107 diag::ext_nonstandard_escape) << "e"; 108 ResultChar = 27; 109 break; 110 case 'E': 111 if (Diags) 112 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 113 diag::ext_nonstandard_escape) << "E"; 114 ResultChar = 27; 115 break; 116 case 'f': 117 ResultChar = 12; 118 break; 119 case 'n': 120 ResultChar = 10; 121 break; 122 case 'r': 123 ResultChar = 13; 124 break; 125 case 't': 126 ResultChar = 9; 127 break; 128 case 'v': 129 ResultChar = 11; 130 break; 131 case 'x': { // Hex escape. 132 ResultChar = 0; 133 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) { 134 if (Diags) 135 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 136 diag::err_hex_escape_no_digits) << "x"; 137 HadError = 1; 138 break; 139 } 140 141 // Hex escapes are a maximal series of hex digits. 142 bool Overflow = false; 143 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) { 144 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]); 145 if (CharVal == -1) break; 146 // About to shift out a digit? 147 if (ResultChar & 0xF0000000) 148 Overflow = true; 149 ResultChar <<= 4; 150 ResultChar |= CharVal; 151 } 152 153 // See if any bits will be truncated when evaluated as a character. 154 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 155 Overflow = true; 156 ResultChar &= ~0U >> (32-CharWidth); 157 } 158 159 // Check for overflow. 160 if (Overflow && Diags) // Too many digits to fit in 161 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 162 diag::err_hex_escape_too_large); 163 break; 164 } 165 case '0': case '1': case '2': case '3': 166 case '4': case '5': case '6': case '7': { 167 // Octal escapes. 168 --ThisTokBuf; 169 ResultChar = 0; 170 171 // Octal escapes are a series of octal digits with maximum length 3. 172 // "\0123" is a two digit sequence equal to "\012" "3". 173 unsigned NumDigits = 0; 174 do { 175 ResultChar <<= 3; 176 ResultChar |= *ThisTokBuf++ - '0'; 177 ++NumDigits; 178 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 && 179 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7'); 180 181 // Check for overflow. Reject '\777', but not L'\777'. 182 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 183 if (Diags) 184 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 185 diag::err_octal_escape_too_large); 186 ResultChar &= ~0U >> (32-CharWidth); 187 } 188 break; 189 } 190 191 // Otherwise, these are not valid escapes. 192 case '(': case '{': case '[': case '%': 193 // GCC accepts these as extensions. We warn about them as such though. 194 if (Diags) 195 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 196 diag::ext_nonstandard_escape) 197 << std::string(1, ResultChar); 198 break; 199 default: 200 if (!Diags) 201 break; 202 203 if (isPrintable(ResultChar)) 204 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 205 diag::ext_unknown_escape) 206 << std::string(1, ResultChar); 207 else 208 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 209 diag::ext_unknown_escape) 210 << "x" + llvm::utohexstr(ResultChar); 211 break; 212 } 213 214 return ResultChar; 215 } 216 217 static void appendCodePoint(unsigned Codepoint, 218 llvm::SmallVectorImpl<char> &Str) { 219 char ResultBuf[4]; 220 char *ResultPtr = ResultBuf; 221 bool Res = llvm::ConvertCodePointToUTF8(Codepoint, ResultPtr); 222 (void)Res; 223 assert(Res && "Unexpected conversion failure"); 224 Str.append(ResultBuf, ResultPtr); 225 } 226 227 void clang::expandUCNs(SmallVectorImpl<char> &Buf, StringRef Input) { 228 for (StringRef::iterator I = Input.begin(), E = Input.end(); I != E; ++I) { 229 if (*I != '\\') { 230 Buf.push_back(*I); 231 continue; 232 } 233 234 ++I; 235 assert(*I == 'u' || *I == 'U'); 236 237 unsigned NumHexDigits; 238 if (*I == 'u') 239 NumHexDigits = 4; 240 else 241 NumHexDigits = 8; 242 243 assert(I + NumHexDigits <= E); 244 245 uint32_t CodePoint = 0; 246 for (++I; NumHexDigits != 0; ++I, --NumHexDigits) { 247 unsigned Value = llvm::hexDigitValue(*I); 248 assert(Value != -1U); 249 250 CodePoint <<= 4; 251 CodePoint += Value; 252 } 253 254 appendCodePoint(CodePoint, Buf); 255 --I; 256 } 257 } 258 259 /// ProcessUCNEscape - Read the Universal Character Name, check constraints and 260 /// return the UTF32. 261 static bool ProcessUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 262 const char *ThisTokEnd, 263 uint32_t &UcnVal, unsigned short &UcnLen, 264 FullSourceLoc Loc, DiagnosticsEngine *Diags, 265 const LangOptions &Features, 266 bool in_char_string_literal = false) { 267 const char *UcnBegin = ThisTokBuf; 268 269 // Skip the '\u' char's. 270 ThisTokBuf += 2; 271 272 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) { 273 if (Diags) 274 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 275 diag::err_hex_escape_no_digits) << StringRef(&ThisTokBuf[-1], 1); 276 return false; 277 } 278 UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8); 279 unsigned short UcnLenSave = UcnLen; 280 for (; ThisTokBuf != ThisTokEnd && UcnLenSave; ++ThisTokBuf, UcnLenSave--) { 281 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]); 282 if (CharVal == -1) break; 283 UcnVal <<= 4; 284 UcnVal |= CharVal; 285 } 286 // If we didn't consume the proper number of digits, there is a problem. 287 if (UcnLenSave) { 288 if (Diags) 289 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 290 diag::err_ucn_escape_incomplete); 291 return false; 292 } 293 294 // Check UCN constraints (C99 6.4.3p2) [C++11 lex.charset p2] 295 if ((0xD800 <= UcnVal && UcnVal <= 0xDFFF) || // surrogate codepoints 296 UcnVal > 0x10FFFF) { // maximum legal UTF32 value 297 if (Diags) 298 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 299 diag::err_ucn_escape_invalid); 300 return false; 301 } 302 303 // C++11 allows UCNs that refer to control characters and basic source 304 // characters inside character and string literals 305 if (UcnVal < 0xa0 && 306 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60)) { // $, @, ` 307 bool IsError = (!Features.CPlusPlus11 || !in_char_string_literal); 308 if (Diags) { 309 char BasicSCSChar = UcnVal; 310 if (UcnVal >= 0x20 && UcnVal < 0x7f) 311 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 312 IsError ? diag::err_ucn_escape_basic_scs : 313 diag::warn_cxx98_compat_literal_ucn_escape_basic_scs) 314 << StringRef(&BasicSCSChar, 1); 315 else 316 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 317 IsError ? diag::err_ucn_control_character : 318 diag::warn_cxx98_compat_literal_ucn_control_character); 319 } 320 if (IsError) 321 return false; 322 } 323 324 if (!Features.CPlusPlus && !Features.C99 && Diags) 325 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 326 diag::warn_ucn_not_valid_in_c89_literal); 327 328 return true; 329 } 330 331 /// MeasureUCNEscape - Determine the number of bytes within the resulting string 332 /// which this UCN will occupy. 333 static int MeasureUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 334 const char *ThisTokEnd, unsigned CharByteWidth, 335 const LangOptions &Features, bool &HadError) { 336 // UTF-32: 4 bytes per escape. 337 if (CharByteWidth == 4) 338 return 4; 339 340 uint32_t UcnVal = 0; 341 unsigned short UcnLen = 0; 342 FullSourceLoc Loc; 343 344 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, 345 UcnLen, Loc, nullptr, Features, true)) { 346 HadError = true; 347 return 0; 348 } 349 350 // UTF-16: 2 bytes for BMP, 4 bytes otherwise. 351 if (CharByteWidth == 2) 352 return UcnVal <= 0xFFFF ? 2 : 4; 353 354 // UTF-8. 355 if (UcnVal < 0x80) 356 return 1; 357 if (UcnVal < 0x800) 358 return 2; 359 if (UcnVal < 0x10000) 360 return 3; 361 return 4; 362 } 363 364 /// EncodeUCNEscape - Read the Universal Character Name, check constraints and 365 /// convert the UTF32 to UTF8 or UTF16. This is a subroutine of 366 /// StringLiteralParser. When we decide to implement UCN's for identifiers, 367 /// we will likely rework our support for UCN's. 368 static void EncodeUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 369 const char *ThisTokEnd, 370 char *&ResultBuf, bool &HadError, 371 FullSourceLoc Loc, unsigned CharByteWidth, 372 DiagnosticsEngine *Diags, 373 const LangOptions &Features) { 374 typedef uint32_t UTF32; 375 UTF32 UcnVal = 0; 376 unsigned short UcnLen = 0; 377 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, UcnLen, 378 Loc, Diags, Features, true)) { 379 HadError = true; 380 return; 381 } 382 383 assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth == 4) && 384 "only character widths of 1, 2, or 4 bytes supported"); 385 386 (void)UcnLen; 387 assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported"); 388 389 if (CharByteWidth == 4) { 390 // FIXME: Make the type of the result buffer correct instead of 391 // using reinterpret_cast. 392 UTF32 *ResultPtr = reinterpret_cast<UTF32*>(ResultBuf); 393 *ResultPtr = UcnVal; 394 ResultBuf += 4; 395 return; 396 } 397 398 if (CharByteWidth == 2) { 399 // FIXME: Make the type of the result buffer correct instead of 400 // using reinterpret_cast. 401 UTF16 *ResultPtr = reinterpret_cast<UTF16*>(ResultBuf); 402 403 if (UcnVal <= (UTF32)0xFFFF) { 404 *ResultPtr = UcnVal; 405 ResultBuf += 2; 406 return; 407 } 408 409 // Convert to UTF16. 410 UcnVal -= 0x10000; 411 *ResultPtr = 0xD800 + (UcnVal >> 10); 412 *(ResultPtr+1) = 0xDC00 + (UcnVal & 0x3FF); 413 ResultBuf += 4; 414 return; 415 } 416 417 assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters"); 418 419 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8. 420 // The conversion below was inspired by: 421 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c 422 // First, we determine how many bytes the result will require. 423 typedef uint8_t UTF8; 424 425 unsigned short bytesToWrite = 0; 426 if (UcnVal < (UTF32)0x80) 427 bytesToWrite = 1; 428 else if (UcnVal < (UTF32)0x800) 429 bytesToWrite = 2; 430 else if (UcnVal < (UTF32)0x10000) 431 bytesToWrite = 3; 432 else 433 bytesToWrite = 4; 434 435 const unsigned byteMask = 0xBF; 436 const unsigned byteMark = 0x80; 437 438 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed 439 // into the first byte, depending on how many bytes follow. 440 static const UTF8 firstByteMark[5] = { 441 0x00, 0x00, 0xC0, 0xE0, 0xF0 442 }; 443 // Finally, we write the bytes into ResultBuf. 444 ResultBuf += bytesToWrite; 445 switch (bytesToWrite) { // note: everything falls through. 446 case 4: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 447 case 3: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 448 case 2: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 449 case 1: *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]); 450 } 451 // Update the buffer. 452 ResultBuf += bytesToWrite; 453 } 454 455 456 /// integer-constant: [C99 6.4.4.1] 457 /// decimal-constant integer-suffix 458 /// octal-constant integer-suffix 459 /// hexadecimal-constant integer-suffix 460 /// binary-literal integer-suffix [GNU, C++1y] 461 /// user-defined-integer-literal: [C++11 lex.ext] 462 /// decimal-literal ud-suffix 463 /// octal-literal ud-suffix 464 /// hexadecimal-literal ud-suffix 465 /// binary-literal ud-suffix [GNU, C++1y] 466 /// decimal-constant: 467 /// nonzero-digit 468 /// decimal-constant digit 469 /// octal-constant: 470 /// 0 471 /// octal-constant octal-digit 472 /// hexadecimal-constant: 473 /// hexadecimal-prefix hexadecimal-digit 474 /// hexadecimal-constant hexadecimal-digit 475 /// hexadecimal-prefix: one of 476 /// 0x 0X 477 /// binary-literal: 478 /// 0b binary-digit 479 /// 0B binary-digit 480 /// binary-literal binary-digit 481 /// integer-suffix: 482 /// unsigned-suffix [long-suffix] 483 /// unsigned-suffix [long-long-suffix] 484 /// long-suffix [unsigned-suffix] 485 /// long-long-suffix [unsigned-sufix] 486 /// nonzero-digit: 487 /// 1 2 3 4 5 6 7 8 9 488 /// octal-digit: 489 /// 0 1 2 3 4 5 6 7 490 /// hexadecimal-digit: 491 /// 0 1 2 3 4 5 6 7 8 9 492 /// a b c d e f 493 /// A B C D E F 494 /// binary-digit: 495 /// 0 496 /// 1 497 /// unsigned-suffix: one of 498 /// u U 499 /// long-suffix: one of 500 /// l L 501 /// long-long-suffix: one of 502 /// ll LL 503 /// 504 /// floating-constant: [C99 6.4.4.2] 505 /// TODO: add rules... 506 /// 507 NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling, 508 SourceLocation TokLoc, 509 Preprocessor &PP) 510 : PP(PP), ThisTokBegin(TokSpelling.begin()), ThisTokEnd(TokSpelling.end()) { 511 512 // This routine assumes that the range begin/end matches the regex for integer 513 // and FP constants (specifically, the 'pp-number' regex), and assumes that 514 // the byte at "*end" is both valid and not part of the regex. Because of 515 // this, it doesn't have to check for 'overscan' in various places. 516 assert(!isPreprocessingNumberBody(*ThisTokEnd) && "didn't maximally munch?"); 517 518 s = DigitsBegin = ThisTokBegin; 519 saw_exponent = false; 520 saw_period = false; 521 saw_ud_suffix = false; 522 isLong = false; 523 isUnsigned = false; 524 isLongLong = false; 525 isFloat = false; 526 isImaginary = false; 527 MicrosoftInteger = 0; 528 hadError = false; 529 530 if (*s == '0') { // parse radix 531 ParseNumberStartingWithZero(TokLoc); 532 if (hadError) 533 return; 534 } else { // the first digit is non-zero 535 radix = 10; 536 s = SkipDigits(s); 537 if (s == ThisTokEnd) { 538 // Done. 539 } else if (isHexDigit(*s) && !(*s == 'e' || *s == 'E')) { 540 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 541 diag::err_invalid_decimal_digit) << StringRef(s, 1); 542 hadError = true; 543 return; 544 } else if (*s == '.') { 545 checkSeparator(TokLoc, s, CSK_AfterDigits); 546 s++; 547 saw_period = true; 548 checkSeparator(TokLoc, s, CSK_BeforeDigits); 549 s = SkipDigits(s); 550 } 551 if ((*s == 'e' || *s == 'E')) { // exponent 552 checkSeparator(TokLoc, s, CSK_AfterDigits); 553 const char *Exponent = s; 554 s++; 555 saw_exponent = true; 556 if (*s == '+' || *s == '-') s++; // sign 557 checkSeparator(TokLoc, s, CSK_BeforeDigits); 558 const char *first_non_digit = SkipDigits(s); 559 if (first_non_digit != s) { 560 s = first_non_digit; 561 } else { 562 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent - ThisTokBegin), 563 diag::err_exponent_has_no_digits); 564 hadError = true; 565 return; 566 } 567 } 568 } 569 570 SuffixBegin = s; 571 checkSeparator(TokLoc, s, CSK_AfterDigits); 572 573 // Parse the suffix. At this point we can classify whether we have an FP or 574 // integer constant. 575 bool isFPConstant = isFloatingLiteral(); 576 const char *ImaginarySuffixLoc = nullptr; 577 578 // Loop over all of the characters of the suffix. If we see something bad, 579 // we break out of the loop. 580 for (; s != ThisTokEnd; ++s) { 581 switch (*s) { 582 case 'f': // FP Suffix for "float" 583 case 'F': 584 if (!isFPConstant) break; // Error for integer constant. 585 if (isFloat || isLong) break; // FF, LF invalid. 586 isFloat = true; 587 continue; // Success. 588 case 'u': 589 case 'U': 590 if (isFPConstant) break; // Error for floating constant. 591 if (isUnsigned) break; // Cannot be repeated. 592 isUnsigned = true; 593 continue; // Success. 594 case 'l': 595 case 'L': 596 if (isLong || isLongLong) break; // Cannot be repeated. 597 if (isFloat) break; // LF invalid. 598 599 // Check for long long. The L's need to be adjacent and the same case. 600 if (s[1] == s[0]) { 601 assert(s + 1 < ThisTokEnd && "didn't maximally munch?"); 602 if (isFPConstant) break; // long long invalid for floats. 603 isLongLong = true; 604 ++s; // Eat both of them. 605 } else { 606 isLong = true; 607 } 608 continue; // Success. 609 case 'i': 610 case 'I': 611 if (PP.getLangOpts().MicrosoftExt) { 612 if (isLong || isLongLong || MicrosoftInteger) 613 break; 614 615 if (!isFPConstant) { 616 // Allow i8, i16, i32, i64, and i128. 617 switch (s[1]) { 618 case '8': 619 s += 2; // i8 suffix 620 MicrosoftInteger = 8; 621 break; 622 case '1': 623 if (s[2] == '6') { 624 s += 3; // i16 suffix 625 MicrosoftInteger = 16; 626 } else if (s[2] == '2' && s[3] == '8') { 627 s += 4; // i128 suffix 628 MicrosoftInteger = 128; 629 } 630 break; 631 case '3': 632 if (s[2] == '2') { 633 s += 3; // i32 suffix 634 MicrosoftInteger = 32; 635 } 636 break; 637 case '6': 638 if (s[2] == '4') { 639 s += 3; // i64 suffix 640 MicrosoftInteger = 64; 641 } 642 break; 643 default: 644 break; 645 } 646 } 647 if (MicrosoftInteger) { 648 assert(s <= ThisTokEnd && "didn't maximally munch?"); 649 break; 650 } 651 } 652 // "i", "if", and "il" are user-defined suffixes in C++1y. 653 if (*s == 'i' && PP.getLangOpts().CPlusPlus14) 654 break; 655 // fall through. 656 case 'j': 657 case 'J': 658 if (isImaginary) break; // Cannot be repeated. 659 isImaginary = true; 660 ImaginarySuffixLoc = s; 661 continue; // Success. 662 } 663 // If we reached here, there was an error or a ud-suffix. 664 break; 665 } 666 667 if (s != ThisTokEnd) { 668 // FIXME: Don't bother expanding UCNs if !tok.hasUCN(). 669 expandUCNs(UDSuffixBuf, StringRef(SuffixBegin, ThisTokEnd - SuffixBegin)); 670 if (isValidUDSuffix(PP.getLangOpts(), UDSuffixBuf)) { 671 // Any suffix pieces we might have parsed are actually part of the 672 // ud-suffix. 673 isLong = false; 674 isUnsigned = false; 675 isLongLong = false; 676 isFloat = false; 677 isImaginary = false; 678 MicrosoftInteger = 0; 679 680 saw_ud_suffix = true; 681 return; 682 } 683 684 // Report an error if there are any. 685 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin), 686 isFPConstant ? diag::err_invalid_suffix_float_constant : 687 diag::err_invalid_suffix_integer_constant) 688 << StringRef(SuffixBegin, ThisTokEnd-SuffixBegin); 689 hadError = true; 690 return; 691 } 692 693 if (isImaginary) { 694 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, 695 ImaginarySuffixLoc - ThisTokBegin), 696 diag::ext_imaginary_constant); 697 } 698 } 699 700 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved 701 /// suffixes as ud-suffixes, because the diagnostic experience is better if we 702 /// treat it as an invalid suffix. 703 bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts, 704 StringRef Suffix) { 705 if (!LangOpts.CPlusPlus11 || Suffix.empty()) 706 return false; 707 708 // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid. 709 if (Suffix[0] == '_') 710 return true; 711 712 // In C++11, there are no library suffixes. 713 if (!LangOpts.CPlusPlus14) 714 return false; 715 716 // In C++1y, "s", "h", "min", "ms", "us", and "ns" are used in the library. 717 // Per tweaked N3660, "il", "i", and "if" are also used in the library. 718 return llvm::StringSwitch<bool>(Suffix) 719 .Cases("h", "min", "s", true) 720 .Cases("ms", "us", "ns", true) 721 .Cases("il", "i", "if", true) 722 .Default(false); 723 } 724 725 void NumericLiteralParser::checkSeparator(SourceLocation TokLoc, 726 const char *Pos, 727 CheckSeparatorKind IsAfterDigits) { 728 if (IsAfterDigits == CSK_AfterDigits) { 729 if (Pos == ThisTokBegin) 730 return; 731 --Pos; 732 } else if (Pos == ThisTokEnd) 733 return; 734 735 if (isDigitSeparator(*Pos)) 736 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin), 737 diag::err_digit_separator_not_between_digits) 738 << IsAfterDigits; 739 } 740 741 /// ParseNumberStartingWithZero - This method is called when the first character 742 /// of the number is found to be a zero. This means it is either an octal 743 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or 744 /// a floating point number (01239.123e4). Eat the prefix, determining the 745 /// radix etc. 746 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) { 747 assert(s[0] == '0' && "Invalid method call"); 748 s++; 749 750 int c1 = s[0]; 751 752 // Handle a hex number like 0x1234. 753 if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) { 754 s++; 755 assert(s < ThisTokEnd && "didn't maximally munch?"); 756 radix = 16; 757 DigitsBegin = s; 758 s = SkipHexDigits(s); 759 bool noSignificand = (s == DigitsBegin); 760 if (s == ThisTokEnd) { 761 // Done. 762 } else if (*s == '.') { 763 s++; 764 saw_period = true; 765 const char *floatDigitsBegin = s; 766 checkSeparator(TokLoc, s, CSK_BeforeDigits); 767 s = SkipHexDigits(s); 768 noSignificand &= (floatDigitsBegin == s); 769 } 770 771 if (noSignificand) { 772 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 773 diag::err_hexconstant_requires_digits); 774 hadError = true; 775 return; 776 } 777 778 // A binary exponent can appear with or with a '.'. If dotted, the 779 // binary exponent is required. 780 if (*s == 'p' || *s == 'P') { 781 checkSeparator(TokLoc, s, CSK_AfterDigits); 782 const char *Exponent = s; 783 s++; 784 saw_exponent = true; 785 if (*s == '+' || *s == '-') s++; // sign 786 const char *first_non_digit = SkipDigits(s); 787 if (first_non_digit == s) { 788 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 789 diag::err_exponent_has_no_digits); 790 hadError = true; 791 return; 792 } 793 checkSeparator(TokLoc, s, CSK_BeforeDigits); 794 s = first_non_digit; 795 796 if (!PP.getLangOpts().HexFloats) 797 PP.Diag(TokLoc, diag::ext_hexconstant_invalid); 798 } else if (saw_period) { 799 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 800 diag::err_hexconstant_requires_exponent); 801 hadError = true; 802 } 803 return; 804 } 805 806 // Handle simple binary numbers 0b01010 807 if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) { 808 // 0b101010 is a C++1y / GCC extension. 809 PP.Diag(TokLoc, 810 PP.getLangOpts().CPlusPlus14 811 ? diag::warn_cxx11_compat_binary_literal 812 : PP.getLangOpts().CPlusPlus 813 ? diag::ext_binary_literal_cxx14 814 : diag::ext_binary_literal); 815 ++s; 816 assert(s < ThisTokEnd && "didn't maximally munch?"); 817 radix = 2; 818 DigitsBegin = s; 819 s = SkipBinaryDigits(s); 820 if (s == ThisTokEnd) { 821 // Done. 822 } else if (isHexDigit(*s)) { 823 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 824 diag::err_invalid_binary_digit) << StringRef(s, 1); 825 hadError = true; 826 } 827 // Other suffixes will be diagnosed by the caller. 828 return; 829 } 830 831 // For now, the radix is set to 8. If we discover that we have a 832 // floating point constant, the radix will change to 10. Octal floating 833 // point constants are not permitted (only decimal and hexadecimal). 834 radix = 8; 835 DigitsBegin = s; 836 s = SkipOctalDigits(s); 837 if (s == ThisTokEnd) 838 return; // Done, simple octal number like 01234 839 840 // If we have some other non-octal digit that *is* a decimal digit, see if 841 // this is part of a floating point number like 094.123 or 09e1. 842 if (isDigit(*s)) { 843 const char *EndDecimal = SkipDigits(s); 844 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') { 845 s = EndDecimal; 846 radix = 10; 847 } 848 } 849 850 // If we have a hex digit other than 'e' (which denotes a FP exponent) then 851 // the code is using an incorrect base. 852 if (isHexDigit(*s) && *s != 'e' && *s != 'E') { 853 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 854 diag::err_invalid_octal_digit) << StringRef(s, 1); 855 hadError = true; 856 return; 857 } 858 859 if (*s == '.') { 860 s++; 861 radix = 10; 862 saw_period = true; 863 checkSeparator(TokLoc, s, CSK_BeforeDigits); 864 s = SkipDigits(s); // Skip suffix. 865 } 866 if (*s == 'e' || *s == 'E') { // exponent 867 checkSeparator(TokLoc, s, CSK_AfterDigits); 868 const char *Exponent = s; 869 s++; 870 radix = 10; 871 saw_exponent = true; 872 if (*s == '+' || *s == '-') s++; // sign 873 const char *first_non_digit = SkipDigits(s); 874 if (first_non_digit != s) { 875 checkSeparator(TokLoc, s, CSK_BeforeDigits); 876 s = first_non_digit; 877 } else { 878 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 879 diag::err_exponent_has_no_digits); 880 hadError = true; 881 return; 882 } 883 } 884 } 885 886 static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) { 887 switch (Radix) { 888 case 2: 889 return NumDigits <= 64; 890 case 8: 891 return NumDigits <= 64 / 3; // Digits are groups of 3 bits. 892 case 10: 893 return NumDigits <= 19; // floor(log10(2^64)) 894 case 16: 895 return NumDigits <= 64 / 4; // Digits are groups of 4 bits. 896 default: 897 llvm_unreachable("impossible Radix"); 898 } 899 } 900 901 /// GetIntegerValue - Convert this numeric literal value to an APInt that 902 /// matches Val's input width. If there is an overflow, set Val to the low bits 903 /// of the result and return true. Otherwise, return false. 904 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) { 905 // Fast path: Compute a conservative bound on the maximum number of 906 // bits per digit in this radix. If we can't possibly overflow a 907 // uint64 based on that bound then do the simple conversion to 908 // integer. This avoids the expensive overflow checking below, and 909 // handles the common cases that matter (small decimal integers and 910 // hex/octal values which don't overflow). 911 const unsigned NumDigits = SuffixBegin - DigitsBegin; 912 if (alwaysFitsInto64Bits(radix, NumDigits)) { 913 uint64_t N = 0; 914 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr) 915 if (!isDigitSeparator(*Ptr)) 916 N = N * radix + llvm::hexDigitValue(*Ptr); 917 918 // This will truncate the value to Val's input width. Simply check 919 // for overflow by comparing. 920 Val = N; 921 return Val.getZExtValue() != N; 922 } 923 924 Val = 0; 925 const char *Ptr = DigitsBegin; 926 927 llvm::APInt RadixVal(Val.getBitWidth(), radix); 928 llvm::APInt CharVal(Val.getBitWidth(), 0); 929 llvm::APInt OldVal = Val; 930 931 bool OverflowOccurred = false; 932 while (Ptr < SuffixBegin) { 933 if (isDigitSeparator(*Ptr)) { 934 ++Ptr; 935 continue; 936 } 937 938 unsigned C = llvm::hexDigitValue(*Ptr++); 939 940 // If this letter is out of bound for this radix, reject it. 941 assert(C < radix && "NumericLiteralParser ctor should have rejected this"); 942 943 CharVal = C; 944 945 // Add the digit to the value in the appropriate radix. If adding in digits 946 // made the value smaller, then this overflowed. 947 OldVal = Val; 948 949 // Multiply by radix, did overflow occur on the multiply? 950 Val *= RadixVal; 951 OverflowOccurred |= Val.udiv(RadixVal) != OldVal; 952 953 // Add value, did overflow occur on the value? 954 // (a + b) ult b <=> overflow 955 Val += CharVal; 956 OverflowOccurred |= Val.ult(CharVal); 957 } 958 return OverflowOccurred; 959 } 960 961 llvm::APFloat::opStatus 962 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) { 963 using llvm::APFloat; 964 965 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin); 966 967 llvm::SmallString<16> Buffer; 968 StringRef Str(ThisTokBegin, n); 969 if (Str.find('\'') != StringRef::npos) { 970 Buffer.reserve(n); 971 std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer), 972 &isDigitSeparator); 973 Str = Buffer; 974 } 975 976 return Result.convertFromString(Str, APFloat::rmNearestTiesToEven); 977 } 978 979 980 /// \verbatim 981 /// user-defined-character-literal: [C++11 lex.ext] 982 /// character-literal ud-suffix 983 /// ud-suffix: 984 /// identifier 985 /// character-literal: [C++11 lex.ccon] 986 /// ' c-char-sequence ' 987 /// u' c-char-sequence ' 988 /// U' c-char-sequence ' 989 /// L' c-char-sequence ' 990 /// c-char-sequence: 991 /// c-char 992 /// c-char-sequence c-char 993 /// c-char: 994 /// any member of the source character set except the single-quote ', 995 /// backslash \, or new-line character 996 /// escape-sequence 997 /// universal-character-name 998 /// escape-sequence: 999 /// simple-escape-sequence 1000 /// octal-escape-sequence 1001 /// hexadecimal-escape-sequence 1002 /// simple-escape-sequence: 1003 /// one of \' \" \? \\ \a \b \f \n \r \t \v 1004 /// octal-escape-sequence: 1005 /// \ octal-digit 1006 /// \ octal-digit octal-digit 1007 /// \ octal-digit octal-digit octal-digit 1008 /// hexadecimal-escape-sequence: 1009 /// \x hexadecimal-digit 1010 /// hexadecimal-escape-sequence hexadecimal-digit 1011 /// universal-character-name: [C++11 lex.charset] 1012 /// \u hex-quad 1013 /// \U hex-quad hex-quad 1014 /// hex-quad: 1015 /// hex-digit hex-digit hex-digit hex-digit 1016 /// \endverbatim 1017 /// 1018 CharLiteralParser::CharLiteralParser(const char *begin, const char *end, 1019 SourceLocation Loc, Preprocessor &PP, 1020 tok::TokenKind kind) { 1021 // At this point we know that the character matches the regex "(L|u|U)?'.*'". 1022 HadError = false; 1023 1024 Kind = kind; 1025 1026 const char *TokBegin = begin; 1027 1028 // Skip over wide character determinant. 1029 if (Kind != tok::char_constant) 1030 ++begin; 1031 if (Kind == tok::utf8_char_constant) 1032 ++begin; 1033 1034 // Skip over the entry quote. 1035 assert(begin[0] == '\'' && "Invalid token lexed"); 1036 ++begin; 1037 1038 // Remove an optional ud-suffix. 1039 if (end[-1] != '\'') { 1040 const char *UDSuffixEnd = end; 1041 do { 1042 --end; 1043 } while (end[-1] != '\''); 1044 // FIXME: Don't bother with this if !tok.hasUCN(). 1045 expandUCNs(UDSuffixBuf, StringRef(end, UDSuffixEnd - end)); 1046 UDSuffixOffset = end - TokBegin; 1047 } 1048 1049 // Trim the ending quote. 1050 assert(end != begin && "Invalid token lexed"); 1051 --end; 1052 1053 // FIXME: The "Value" is an uint64_t so we can handle char literals of 1054 // up to 64-bits. 1055 // FIXME: This extensively assumes that 'char' is 8-bits. 1056 assert(PP.getTargetInfo().getCharWidth() == 8 && 1057 "Assumes char is 8 bits"); 1058 assert(PP.getTargetInfo().getIntWidth() <= 64 && 1059 (PP.getTargetInfo().getIntWidth() & 7) == 0 && 1060 "Assumes sizeof(int) on target is <= 64 and a multiple of char"); 1061 assert(PP.getTargetInfo().getWCharWidth() <= 64 && 1062 "Assumes sizeof(wchar) on target is <= 64"); 1063 1064 SmallVector<uint32_t, 4> codepoint_buffer; 1065 codepoint_buffer.resize(end - begin); 1066 uint32_t *buffer_begin = &codepoint_buffer.front(); 1067 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size(); 1068 1069 // Unicode escapes representing characters that cannot be correctly 1070 // represented in a single code unit are disallowed in character literals 1071 // by this implementation. 1072 uint32_t largest_character_for_kind; 1073 if (tok::wide_char_constant == Kind) { 1074 largest_character_for_kind = 1075 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth()); 1076 } else if (tok::utf8_char_constant == Kind) { 1077 largest_character_for_kind = 0x7F; 1078 } else if (tok::utf16_char_constant == Kind) { 1079 largest_character_for_kind = 0xFFFF; 1080 } else if (tok::utf32_char_constant == Kind) { 1081 largest_character_for_kind = 0x10FFFF; 1082 } else { 1083 largest_character_for_kind = 0x7Fu; 1084 } 1085 1086 while (begin != end) { 1087 // Is this a span of non-escape characters? 1088 if (begin[0] != '\\') { 1089 char const *start = begin; 1090 do { 1091 ++begin; 1092 } while (begin != end && *begin != '\\'); 1093 1094 char const *tmp_in_start = start; 1095 uint32_t *tmp_out_start = buffer_begin; 1096 ConversionResult res = 1097 ConvertUTF8toUTF32(reinterpret_cast<UTF8 const **>(&start), 1098 reinterpret_cast<UTF8 const *>(begin), 1099 &buffer_begin, buffer_end, strictConversion); 1100 if (res != conversionOK) { 1101 // If we see bad encoding for unprefixed character literals, warn and 1102 // simply copy the byte values, for compatibility with gcc and 1103 // older versions of clang. 1104 bool NoErrorOnBadEncoding = isAscii(); 1105 unsigned Msg = diag::err_bad_character_encoding; 1106 if (NoErrorOnBadEncoding) 1107 Msg = diag::warn_bad_character_encoding; 1108 PP.Diag(Loc, Msg); 1109 if (NoErrorOnBadEncoding) { 1110 start = tmp_in_start; 1111 buffer_begin = tmp_out_start; 1112 for (; start != begin; ++start, ++buffer_begin) 1113 *buffer_begin = static_cast<uint8_t>(*start); 1114 } else { 1115 HadError = true; 1116 } 1117 } else { 1118 for (; tmp_out_start < buffer_begin; ++tmp_out_start) { 1119 if (*tmp_out_start > largest_character_for_kind) { 1120 HadError = true; 1121 PP.Diag(Loc, diag::err_character_too_large); 1122 } 1123 } 1124 } 1125 1126 continue; 1127 } 1128 // Is this a Universal Character Name escape? 1129 if (begin[1] == 'u' || begin[1] == 'U') { 1130 unsigned short UcnLen = 0; 1131 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen, 1132 FullSourceLoc(Loc, PP.getSourceManager()), 1133 &PP.getDiagnostics(), PP.getLangOpts(), true)) { 1134 HadError = true; 1135 } else if (*buffer_begin > largest_character_for_kind) { 1136 HadError = true; 1137 PP.Diag(Loc, diag::err_character_too_large); 1138 } 1139 1140 ++buffer_begin; 1141 continue; 1142 } 1143 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo()); 1144 uint64_t result = 1145 ProcessCharEscape(TokBegin, begin, end, HadError, 1146 FullSourceLoc(Loc,PP.getSourceManager()), 1147 CharWidth, &PP.getDiagnostics(), PP.getLangOpts()); 1148 *buffer_begin++ = result; 1149 } 1150 1151 unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front(); 1152 1153 if (NumCharsSoFar > 1) { 1154 if (isWide()) 1155 PP.Diag(Loc, diag::warn_extraneous_char_constant); 1156 else if (isAscii() && NumCharsSoFar == 4) 1157 PP.Diag(Loc, diag::ext_four_char_character_literal); 1158 else if (isAscii()) 1159 PP.Diag(Loc, diag::ext_multichar_character_literal); 1160 else 1161 PP.Diag(Loc, diag::err_multichar_utf_character_literal); 1162 IsMultiChar = true; 1163 } else { 1164 IsMultiChar = false; 1165 } 1166 1167 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0); 1168 1169 // Narrow character literals act as though their value is concatenated 1170 // in this implementation, but warn on overflow. 1171 bool multi_char_too_long = false; 1172 if (isAscii() && isMultiChar()) { 1173 LitVal = 0; 1174 for (size_t i = 0; i < NumCharsSoFar; ++i) { 1175 // check for enough leading zeros to shift into 1176 multi_char_too_long |= (LitVal.countLeadingZeros() < 8); 1177 LitVal <<= 8; 1178 LitVal = LitVal + (codepoint_buffer[i] & 0xFF); 1179 } 1180 } else if (NumCharsSoFar > 0) { 1181 // otherwise just take the last character 1182 LitVal = buffer_begin[-1]; 1183 } 1184 1185 if (!HadError && multi_char_too_long) { 1186 PP.Diag(Loc, diag::warn_char_constant_too_large); 1187 } 1188 1189 // Transfer the value from APInt to uint64_t 1190 Value = LitVal.getZExtValue(); 1191 1192 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1") 1193 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple 1194 // character constants are not sign extended in the this implementation: 1195 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC. 1196 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) && 1197 PP.getLangOpts().CharIsSigned) 1198 Value = (signed char)Value; 1199 } 1200 1201 /// \verbatim 1202 /// string-literal: [C++0x lex.string] 1203 /// encoding-prefix " [s-char-sequence] " 1204 /// encoding-prefix R raw-string 1205 /// encoding-prefix: 1206 /// u8 1207 /// u 1208 /// U 1209 /// L 1210 /// s-char-sequence: 1211 /// s-char 1212 /// s-char-sequence s-char 1213 /// s-char: 1214 /// any member of the source character set except the double-quote ", 1215 /// backslash \, or new-line character 1216 /// escape-sequence 1217 /// universal-character-name 1218 /// raw-string: 1219 /// " d-char-sequence ( r-char-sequence ) d-char-sequence " 1220 /// r-char-sequence: 1221 /// r-char 1222 /// r-char-sequence r-char 1223 /// r-char: 1224 /// any member of the source character set, except a right parenthesis ) 1225 /// followed by the initial d-char-sequence (which may be empty) 1226 /// followed by a double quote ". 1227 /// d-char-sequence: 1228 /// d-char 1229 /// d-char-sequence d-char 1230 /// d-char: 1231 /// any member of the basic source character set except: 1232 /// space, the left parenthesis (, the right parenthesis ), 1233 /// the backslash \, and the control characters representing horizontal 1234 /// tab, vertical tab, form feed, and newline. 1235 /// escape-sequence: [C++0x lex.ccon] 1236 /// simple-escape-sequence 1237 /// octal-escape-sequence 1238 /// hexadecimal-escape-sequence 1239 /// simple-escape-sequence: 1240 /// one of \' \" \? \\ \a \b \f \n \r \t \v 1241 /// octal-escape-sequence: 1242 /// \ octal-digit 1243 /// \ octal-digit octal-digit 1244 /// \ octal-digit octal-digit octal-digit 1245 /// hexadecimal-escape-sequence: 1246 /// \x hexadecimal-digit 1247 /// hexadecimal-escape-sequence hexadecimal-digit 1248 /// universal-character-name: 1249 /// \u hex-quad 1250 /// \U hex-quad hex-quad 1251 /// hex-quad: 1252 /// hex-digit hex-digit hex-digit hex-digit 1253 /// \endverbatim 1254 /// 1255 StringLiteralParser:: 1256 StringLiteralParser(ArrayRef<Token> StringToks, 1257 Preprocessor &PP, bool Complain) 1258 : SM(PP.getSourceManager()), Features(PP.getLangOpts()), 1259 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() :nullptr), 1260 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown), 1261 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) { 1262 init(StringToks); 1263 } 1264 1265 void StringLiteralParser::init(ArrayRef<Token> StringToks){ 1266 // The literal token may have come from an invalid source location (e.g. due 1267 // to a PCH error), in which case the token length will be 0. 1268 if (StringToks.empty() || StringToks[0].getLength() < 2) 1269 return DiagnoseLexingError(SourceLocation()); 1270 1271 // Scan all of the string portions, remember the max individual token length, 1272 // computing a bound on the concatenated string length, and see whether any 1273 // piece is a wide-string. If any of the string portions is a wide-string 1274 // literal, the result is a wide-string literal [C99 6.4.5p4]. 1275 assert(!StringToks.empty() && "expected at least one token"); 1276 MaxTokenLength = StringToks[0].getLength(); 1277 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!"); 1278 SizeBound = StringToks[0].getLength()-2; // -2 for "". 1279 Kind = StringToks[0].getKind(); 1280 1281 hadError = false; 1282 1283 // Implement Translation Phase #6: concatenation of string literals 1284 /// (C99 5.1.1.2p1). The common case is only one string fragment. 1285 for (unsigned i = 1; i != StringToks.size(); ++i) { 1286 if (StringToks[i].getLength() < 2) 1287 return DiagnoseLexingError(StringToks[i].getLocation()); 1288 1289 // The string could be shorter than this if it needs cleaning, but this is a 1290 // reasonable bound, which is all we need. 1291 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!"); 1292 SizeBound += StringToks[i].getLength()-2; // -2 for "". 1293 1294 // Remember maximum string piece length. 1295 if (StringToks[i].getLength() > MaxTokenLength) 1296 MaxTokenLength = StringToks[i].getLength(); 1297 1298 // Remember if we see any wide or utf-8/16/32 strings. 1299 // Also check for illegal concatenations. 1300 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) { 1301 if (isAscii()) { 1302 Kind = StringToks[i].getKind(); 1303 } else { 1304 if (Diags) 1305 Diags->Report(StringToks[i].getLocation(), 1306 diag::err_unsupported_string_concat); 1307 hadError = true; 1308 } 1309 } 1310 } 1311 1312 // Include space for the null terminator. 1313 ++SizeBound; 1314 1315 // TODO: K&R warning: "traditional C rejects string constant concatenation" 1316 1317 // Get the width in bytes of char/wchar_t/char16_t/char32_t 1318 CharByteWidth = getCharWidth(Kind, Target); 1319 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple"); 1320 CharByteWidth /= 8; 1321 1322 // The output buffer size needs to be large enough to hold wide characters. 1323 // This is a worst-case assumption which basically corresponds to L"" "long". 1324 SizeBound *= CharByteWidth; 1325 1326 // Size the temporary buffer to hold the result string data. 1327 ResultBuf.resize(SizeBound); 1328 1329 // Likewise, but for each string piece. 1330 SmallString<512> TokenBuf; 1331 TokenBuf.resize(MaxTokenLength); 1332 1333 // Loop over all the strings, getting their spelling, and expanding them to 1334 // wide strings as appropriate. 1335 ResultPtr = &ResultBuf[0]; // Next byte to fill in. 1336 1337 Pascal = false; 1338 1339 SourceLocation UDSuffixTokLoc; 1340 1341 for (unsigned i = 0, e = StringToks.size(); i != e; ++i) { 1342 const char *ThisTokBuf = &TokenBuf[0]; 1343 // Get the spelling of the token, which eliminates trigraphs, etc. We know 1344 // that ThisTokBuf points to a buffer that is big enough for the whole token 1345 // and 'spelled' tokens can only shrink. 1346 bool StringInvalid = false; 1347 unsigned ThisTokLen = 1348 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features, 1349 &StringInvalid); 1350 if (StringInvalid) 1351 return DiagnoseLexingError(StringToks[i].getLocation()); 1352 1353 const char *ThisTokBegin = ThisTokBuf; 1354 const char *ThisTokEnd = ThisTokBuf+ThisTokLen; 1355 1356 // Remove an optional ud-suffix. 1357 if (ThisTokEnd[-1] != '"') { 1358 const char *UDSuffixEnd = ThisTokEnd; 1359 do { 1360 --ThisTokEnd; 1361 } while (ThisTokEnd[-1] != '"'); 1362 1363 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd); 1364 1365 if (UDSuffixBuf.empty()) { 1366 if (StringToks[i].hasUCN()) 1367 expandUCNs(UDSuffixBuf, UDSuffix); 1368 else 1369 UDSuffixBuf.assign(UDSuffix); 1370 UDSuffixToken = i; 1371 UDSuffixOffset = ThisTokEnd - ThisTokBuf; 1372 UDSuffixTokLoc = StringToks[i].getLocation(); 1373 } else { 1374 SmallString<32> ExpandedUDSuffix; 1375 if (StringToks[i].hasUCN()) { 1376 expandUCNs(ExpandedUDSuffix, UDSuffix); 1377 UDSuffix = ExpandedUDSuffix; 1378 } 1379 1380 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the 1381 // result of a concatenation involving at least one user-defined-string- 1382 // literal, all the participating user-defined-string-literals shall 1383 // have the same ud-suffix. 1384 if (UDSuffixBuf != UDSuffix) { 1385 if (Diags) { 1386 SourceLocation TokLoc = StringToks[i].getLocation(); 1387 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix) 1388 << UDSuffixBuf << UDSuffix 1389 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc) 1390 << SourceRange(TokLoc, TokLoc); 1391 } 1392 hadError = true; 1393 } 1394 } 1395 } 1396 1397 // Strip the end quote. 1398 --ThisTokEnd; 1399 1400 // TODO: Input character set mapping support. 1401 1402 // Skip marker for wide or unicode strings. 1403 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') { 1404 ++ThisTokBuf; 1405 // Skip 8 of u8 marker for utf8 strings. 1406 if (ThisTokBuf[0] == '8') 1407 ++ThisTokBuf; 1408 } 1409 1410 // Check for raw string 1411 if (ThisTokBuf[0] == 'R') { 1412 ThisTokBuf += 2; // skip R" 1413 1414 const char *Prefix = ThisTokBuf; 1415 while (ThisTokBuf[0] != '(') 1416 ++ThisTokBuf; 1417 ++ThisTokBuf; // skip '(' 1418 1419 // Remove same number of characters from the end 1420 ThisTokEnd -= ThisTokBuf - Prefix; 1421 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal"); 1422 1423 // Copy the string over 1424 if (CopyStringFragment(StringToks[i], ThisTokBegin, 1425 StringRef(ThisTokBuf, ThisTokEnd - ThisTokBuf))) 1426 hadError = true; 1427 } else { 1428 if (ThisTokBuf[0] != '"') { 1429 // The file may have come from PCH and then changed after loading the 1430 // PCH; Fail gracefully. 1431 return DiagnoseLexingError(StringToks[i].getLocation()); 1432 } 1433 ++ThisTokBuf; // skip " 1434 1435 // Check if this is a pascal string 1436 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd && 1437 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') { 1438 1439 // If the \p sequence is found in the first token, we have a pascal string 1440 // Otherwise, if we already have a pascal string, ignore the first \p 1441 if (i == 0) { 1442 ++ThisTokBuf; 1443 Pascal = true; 1444 } else if (Pascal) 1445 ThisTokBuf += 2; 1446 } 1447 1448 while (ThisTokBuf != ThisTokEnd) { 1449 // Is this a span of non-escape characters? 1450 if (ThisTokBuf[0] != '\\') { 1451 const char *InStart = ThisTokBuf; 1452 do { 1453 ++ThisTokBuf; 1454 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\'); 1455 1456 // Copy the character span over. 1457 if (CopyStringFragment(StringToks[i], ThisTokBegin, 1458 StringRef(InStart, ThisTokBuf - InStart))) 1459 hadError = true; 1460 continue; 1461 } 1462 // Is this a Universal Character Name escape? 1463 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') { 1464 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, 1465 ResultPtr, hadError, 1466 FullSourceLoc(StringToks[i].getLocation(), SM), 1467 CharByteWidth, Diags, Features); 1468 continue; 1469 } 1470 // Otherwise, this is a non-UCN escape character. Process it. 1471 unsigned ResultChar = 1472 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError, 1473 FullSourceLoc(StringToks[i].getLocation(), SM), 1474 CharByteWidth*8, Diags, Features); 1475 1476 if (CharByteWidth == 4) { 1477 // FIXME: Make the type of the result buffer correct instead of 1478 // using reinterpret_cast. 1479 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultPtr); 1480 *ResultWidePtr = ResultChar; 1481 ResultPtr += 4; 1482 } else if (CharByteWidth == 2) { 1483 // FIXME: Make the type of the result buffer correct instead of 1484 // using reinterpret_cast. 1485 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultPtr); 1486 *ResultWidePtr = ResultChar & 0xFFFF; 1487 ResultPtr += 2; 1488 } else { 1489 assert(CharByteWidth == 1 && "Unexpected char width"); 1490 *ResultPtr++ = ResultChar & 0xFF; 1491 } 1492 } 1493 } 1494 } 1495 1496 if (Pascal) { 1497 if (CharByteWidth == 4) { 1498 // FIXME: Make the type of the result buffer correct instead of 1499 // using reinterpret_cast. 1500 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultBuf.data()); 1501 ResultWidePtr[0] = GetNumStringChars() - 1; 1502 } else if (CharByteWidth == 2) { 1503 // FIXME: Make the type of the result buffer correct instead of 1504 // using reinterpret_cast. 1505 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultBuf.data()); 1506 ResultWidePtr[0] = GetNumStringChars() - 1; 1507 } else { 1508 assert(CharByteWidth == 1 && "Unexpected char width"); 1509 ResultBuf[0] = GetNumStringChars() - 1; 1510 } 1511 1512 // Verify that pascal strings aren't too large. 1513 if (GetStringLength() > 256) { 1514 if (Diags) 1515 Diags->Report(StringToks.front().getLocation(), 1516 diag::err_pascal_string_too_long) 1517 << SourceRange(StringToks.front().getLocation(), 1518 StringToks.back().getLocation()); 1519 hadError = true; 1520 return; 1521 } 1522 } else if (Diags) { 1523 // Complain if this string literal has too many characters. 1524 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509; 1525 1526 if (GetNumStringChars() > MaxChars) 1527 Diags->Report(StringToks.front().getLocation(), 1528 diag::ext_string_too_long) 1529 << GetNumStringChars() << MaxChars 1530 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0) 1531 << SourceRange(StringToks.front().getLocation(), 1532 StringToks.back().getLocation()); 1533 } 1534 } 1535 1536 static const char *resyncUTF8(const char *Err, const char *End) { 1537 if (Err == End) 1538 return End; 1539 End = Err + std::min<unsigned>(getNumBytesForUTF8(*Err), End-Err); 1540 while (++Err != End && (*Err & 0xC0) == 0x80) 1541 ; 1542 return Err; 1543 } 1544 1545 /// \brief This function copies from Fragment, which is a sequence of bytes 1546 /// within Tok's contents (which begin at TokBegin) into ResultPtr. 1547 /// Performs widening for multi-byte characters. 1548 bool StringLiteralParser::CopyStringFragment(const Token &Tok, 1549 const char *TokBegin, 1550 StringRef Fragment) { 1551 const UTF8 *ErrorPtrTmp; 1552 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp)) 1553 return false; 1554 1555 // If we see bad encoding for unprefixed string literals, warn and 1556 // simply copy the byte values, for compatibility with gcc and older 1557 // versions of clang. 1558 bool NoErrorOnBadEncoding = isAscii(); 1559 if (NoErrorOnBadEncoding) { 1560 memcpy(ResultPtr, Fragment.data(), Fragment.size()); 1561 ResultPtr += Fragment.size(); 1562 } 1563 1564 if (Diags) { 1565 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1566 1567 FullSourceLoc SourceLoc(Tok.getLocation(), SM); 1568 const DiagnosticBuilder &Builder = 1569 Diag(Diags, Features, SourceLoc, TokBegin, 1570 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()), 1571 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding 1572 : diag::err_bad_string_encoding); 1573 1574 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1575 StringRef NextFragment(NextStart, Fragment.end()-NextStart); 1576 1577 // Decode into a dummy buffer. 1578 SmallString<512> Dummy; 1579 Dummy.reserve(Fragment.size() * CharByteWidth); 1580 char *Ptr = Dummy.data(); 1581 1582 while (!ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) { 1583 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1584 NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1585 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin, 1586 ErrorPtr, NextStart); 1587 NextFragment = StringRef(NextStart, Fragment.end()-NextStart); 1588 } 1589 } 1590 return !NoErrorOnBadEncoding; 1591 } 1592 1593 void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) { 1594 hadError = true; 1595 if (Diags) 1596 Diags->Report(Loc, diag::err_lexing_string); 1597 } 1598 1599 /// getOffsetOfStringByte - This function returns the offset of the 1600 /// specified byte of the string data represented by Token. This handles 1601 /// advancing over escape sequences in the string. 1602 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok, 1603 unsigned ByteNo) const { 1604 // Get the spelling of the token. 1605 SmallString<32> SpellingBuffer; 1606 SpellingBuffer.resize(Tok.getLength()); 1607 1608 bool StringInvalid = false; 1609 const char *SpellingPtr = &SpellingBuffer[0]; 1610 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features, 1611 &StringInvalid); 1612 if (StringInvalid) 1613 return 0; 1614 1615 const char *SpellingStart = SpellingPtr; 1616 const char *SpellingEnd = SpellingPtr+TokLen; 1617 1618 // Handle UTF-8 strings just like narrow strings. 1619 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8') 1620 SpellingPtr += 2; 1621 1622 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' && 1623 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet"); 1624 1625 // For raw string literals, this is easy. 1626 if (SpellingPtr[0] == 'R') { 1627 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!"); 1628 // Skip 'R"'. 1629 SpellingPtr += 2; 1630 while (*SpellingPtr != '(') { 1631 ++SpellingPtr; 1632 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal"); 1633 } 1634 // Skip '('. 1635 ++SpellingPtr; 1636 return SpellingPtr - SpellingStart + ByteNo; 1637 } 1638 1639 // Skip over the leading quote 1640 assert(SpellingPtr[0] == '"' && "Should be a string literal!"); 1641 ++SpellingPtr; 1642 1643 // Skip over bytes until we find the offset we're looking for. 1644 while (ByteNo) { 1645 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!"); 1646 1647 // Step over non-escapes simply. 1648 if (*SpellingPtr != '\\') { 1649 ++SpellingPtr; 1650 --ByteNo; 1651 continue; 1652 } 1653 1654 // Otherwise, this is an escape character. Advance over it. 1655 bool HadError = false; 1656 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') { 1657 const char *EscapePtr = SpellingPtr; 1658 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd, 1659 1, Features, HadError); 1660 if (Len > ByteNo) { 1661 // ByteNo is somewhere within the escape sequence. 1662 SpellingPtr = EscapePtr; 1663 break; 1664 } 1665 ByteNo -= Len; 1666 } else { 1667 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError, 1668 FullSourceLoc(Tok.getLocation(), SM), 1669 CharByteWidth*8, Diags, Features); 1670 --ByteNo; 1671 } 1672 assert(!HadError && "This method isn't valid on erroneous strings"); 1673 } 1674 1675 return SpellingPtr-SpellingStart; 1676 } 1677