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_escape_too_large) << 0; 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_escape_too_large) << 1; 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 isHalf = false; 526 isFloat = false; 527 isImaginary = false; 528 isFloat128 = false; 529 MicrosoftInteger = 0; 530 hadError = false; 531 532 if (*s == '0') { // parse radix 533 ParseNumberStartingWithZero(TokLoc); 534 if (hadError) 535 return; 536 } else { // the first digit is non-zero 537 radix = 10; 538 s = SkipDigits(s); 539 if (s == ThisTokEnd) { 540 // Done. 541 } else { 542 ParseDecimalOrOctalCommon(TokLoc); 543 if (hadError) 544 return; 545 } 546 } 547 548 SuffixBegin = s; 549 checkSeparator(TokLoc, s, CSK_AfterDigits); 550 551 // Parse the suffix. At this point we can classify whether we have an FP or 552 // integer constant. 553 bool isFPConstant = isFloatingLiteral(); 554 const char *ImaginarySuffixLoc = nullptr; 555 556 // Loop over all of the characters of the suffix. If we see something bad, 557 // we break out of the loop. 558 for (; s != ThisTokEnd; ++s) { 559 switch (*s) { 560 case 'h': // FP Suffix for "half". 561 case 'H': 562 // OpenCL Extension v1.2 s9.5 - h or H suffix for half type. 563 if (!PP.getLangOpts().Half) break; 564 if (!isFPConstant) break; // Error for integer constant. 565 if (isHalf || isFloat || isLong) break; // HH, FH, LH invalid. 566 isHalf = true; 567 continue; // Success. 568 case 'f': // FP Suffix for "float" 569 case 'F': 570 if (!isFPConstant) break; // Error for integer constant. 571 if (isHalf || isFloat || isLong || isFloat128) 572 break; // HF, FF, LF, QF invalid. 573 isFloat = true; 574 continue; // Success. 575 case 'q': // FP Suffix for "__float128" 576 case 'Q': 577 if (!isFPConstant) break; // Error for integer constant. 578 if (isHalf || isFloat || isLong || isFloat128) 579 break; // HQ, FQ, LQ, QQ invalid. 580 isFloat128 = true; 581 continue; // Success. 582 case 'u': 583 case 'U': 584 if (isFPConstant) break; // Error for floating constant. 585 if (isUnsigned) break; // Cannot be repeated. 586 isUnsigned = true; 587 continue; // Success. 588 case 'l': 589 case 'L': 590 if (isLong || isLongLong) break; // Cannot be repeated. 591 if (isHalf || isFloat || isFloat128) break; // LH, LF, LQ invalid. 592 593 // Check for long long. The L's need to be adjacent and the same case. 594 if (s[1] == s[0]) { 595 assert(s + 1 < ThisTokEnd && "didn't maximally munch?"); 596 if (isFPConstant) break; // long long invalid for floats. 597 isLongLong = true; 598 ++s; // Eat both of them. 599 } else { 600 isLong = true; 601 } 602 continue; // Success. 603 case 'i': 604 case 'I': 605 if (PP.getLangOpts().MicrosoftExt) { 606 if (isLong || isLongLong || MicrosoftInteger) 607 break; 608 609 if (!isFPConstant) { 610 // Allow i8, i16, i32, and i64. 611 switch (s[1]) { 612 case '8': 613 s += 2; // i8 suffix 614 MicrosoftInteger = 8; 615 break; 616 case '1': 617 if (s[2] == '6') { 618 s += 3; // i16 suffix 619 MicrosoftInteger = 16; 620 } 621 break; 622 case '3': 623 if (s[2] == '2') { 624 s += 3; // i32 suffix 625 MicrosoftInteger = 32; 626 } 627 break; 628 case '6': 629 if (s[2] == '4') { 630 s += 3; // i64 suffix 631 MicrosoftInteger = 64; 632 } 633 break; 634 default: 635 break; 636 } 637 } 638 if (MicrosoftInteger) { 639 assert(s <= ThisTokEnd && "didn't maximally munch?"); 640 break; 641 } 642 } 643 // "i", "if", and "il" are user-defined suffixes in C++1y. 644 if (*s == 'i' && PP.getLangOpts().CPlusPlus14) 645 break; 646 // fall through. 647 case 'j': 648 case 'J': 649 if (isImaginary) break; // Cannot be repeated. 650 isImaginary = true; 651 ImaginarySuffixLoc = s; 652 continue; // Success. 653 } 654 // If we reached here, there was an error or a ud-suffix. 655 break; 656 } 657 658 if (s != ThisTokEnd) { 659 // FIXME: Don't bother expanding UCNs if !tok.hasUCN(). 660 expandUCNs(UDSuffixBuf, StringRef(SuffixBegin, ThisTokEnd - SuffixBegin)); 661 if (isValidUDSuffix(PP.getLangOpts(), UDSuffixBuf)) { 662 // Any suffix pieces we might have parsed are actually part of the 663 // ud-suffix. 664 isLong = false; 665 isUnsigned = false; 666 isLongLong = false; 667 isFloat = false; 668 isHalf = false; 669 isImaginary = false; 670 MicrosoftInteger = 0; 671 672 saw_ud_suffix = true; 673 return; 674 } 675 676 // Report an error if there are any. 677 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin), 678 diag::err_invalid_suffix_constant) 679 << StringRef(SuffixBegin, ThisTokEnd-SuffixBegin) << isFPConstant; 680 hadError = true; 681 return; 682 } 683 684 if (isImaginary) { 685 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, 686 ImaginarySuffixLoc - ThisTokBegin), 687 diag::ext_imaginary_constant); 688 } 689 } 690 691 /// ParseDecimalOrOctalCommon - This method is called for decimal or octal 692 /// numbers. It issues an error for illegal digits, and handles floating point 693 /// parsing. If it detects a floating point number, the radix is set to 10. 694 void NumericLiteralParser::ParseDecimalOrOctalCommon(SourceLocation TokLoc){ 695 assert((radix == 8 || radix == 10) && "Unexpected radix"); 696 697 // If we have a hex digit other than 'e' (which denotes a FP exponent) then 698 // the code is using an incorrect base. 699 if (isHexDigit(*s) && *s != 'e' && *s != 'E') { 700 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 701 diag::err_invalid_digit) << StringRef(s, 1) << (radix == 8 ? 1 : 0); 702 hadError = true; 703 return; 704 } 705 706 if (*s == '.') { 707 checkSeparator(TokLoc, s, CSK_AfterDigits); 708 s++; 709 radix = 10; 710 saw_period = true; 711 checkSeparator(TokLoc, s, CSK_BeforeDigits); 712 s = SkipDigits(s); // Skip suffix. 713 } 714 if (*s == 'e' || *s == 'E') { // exponent 715 checkSeparator(TokLoc, s, CSK_AfterDigits); 716 const char *Exponent = s; 717 s++; 718 radix = 10; 719 saw_exponent = true; 720 if (*s == '+' || *s == '-') s++; // sign 721 const char *first_non_digit = SkipDigits(s); 722 if (containsDigits(s, first_non_digit)) { 723 checkSeparator(TokLoc, s, CSK_BeforeDigits); 724 s = first_non_digit; 725 } else { 726 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 727 diag::err_exponent_has_no_digits); 728 hadError = true; 729 return; 730 } 731 } 732 } 733 734 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved 735 /// suffixes as ud-suffixes, because the diagnostic experience is better if we 736 /// treat it as an invalid suffix. 737 bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts, 738 StringRef Suffix) { 739 if (!LangOpts.CPlusPlus11 || Suffix.empty()) 740 return false; 741 742 // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid. 743 if (Suffix[0] == '_') 744 return true; 745 746 // In C++11, there are no library suffixes. 747 if (!LangOpts.CPlusPlus14) 748 return false; 749 750 // In C++1y, "s", "h", "min", "ms", "us", and "ns" are used in the library. 751 // Per tweaked N3660, "il", "i", and "if" are also used in the library. 752 return llvm::StringSwitch<bool>(Suffix) 753 .Cases("h", "min", "s", true) 754 .Cases("ms", "us", "ns", true) 755 .Cases("il", "i", "if", true) 756 .Default(false); 757 } 758 759 void NumericLiteralParser::checkSeparator(SourceLocation TokLoc, 760 const char *Pos, 761 CheckSeparatorKind IsAfterDigits) { 762 if (IsAfterDigits == CSK_AfterDigits) { 763 if (Pos == ThisTokBegin) 764 return; 765 --Pos; 766 } else if (Pos == ThisTokEnd) 767 return; 768 769 if (isDigitSeparator(*Pos)) 770 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin), 771 diag::err_digit_separator_not_between_digits) 772 << IsAfterDigits; 773 } 774 775 /// ParseNumberStartingWithZero - This method is called when the first character 776 /// of the number is found to be a zero. This means it is either an octal 777 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or 778 /// a floating point number (01239.123e4). Eat the prefix, determining the 779 /// radix etc. 780 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) { 781 assert(s[0] == '0' && "Invalid method call"); 782 s++; 783 784 int c1 = s[0]; 785 786 // Handle a hex number like 0x1234. 787 if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) { 788 s++; 789 assert(s < ThisTokEnd && "didn't maximally munch?"); 790 radix = 16; 791 DigitsBegin = s; 792 s = SkipHexDigits(s); 793 bool HasSignificandDigits = containsDigits(DigitsBegin, s); 794 if (s == ThisTokEnd) { 795 // Done. 796 } else if (*s == '.') { 797 s++; 798 saw_period = true; 799 const char *floatDigitsBegin = s; 800 s = SkipHexDigits(s); 801 if (containsDigits(floatDigitsBegin, s)) 802 HasSignificandDigits = true; 803 if (HasSignificandDigits) 804 checkSeparator(TokLoc, floatDigitsBegin, CSK_BeforeDigits); 805 } 806 807 if (!HasSignificandDigits) { 808 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 809 diag::err_hex_constant_requires) 810 << PP.getLangOpts().CPlusPlus << 1; 811 hadError = true; 812 return; 813 } 814 815 // A binary exponent can appear with or with a '.'. If dotted, the 816 // binary exponent is required. 817 if (*s == 'p' || *s == 'P') { 818 checkSeparator(TokLoc, s, CSK_AfterDigits); 819 const char *Exponent = s; 820 s++; 821 saw_exponent = true; 822 if (*s == '+' || *s == '-') s++; // sign 823 const char *first_non_digit = SkipDigits(s); 824 if (!containsDigits(s, first_non_digit)) { 825 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 826 diag::err_exponent_has_no_digits); 827 hadError = true; 828 return; 829 } 830 checkSeparator(TokLoc, s, CSK_BeforeDigits); 831 s = first_non_digit; 832 833 if (!PP.getLangOpts().HexFloats) 834 PP.Diag(TokLoc, PP.getLangOpts().CPlusPlus 835 ? diag::ext_hex_literal_invalid 836 : diag::ext_hex_constant_invalid); 837 else if (PP.getLangOpts().CPlusPlus1z) 838 PP.Diag(TokLoc, diag::warn_cxx1z_hex_literal); 839 } else if (saw_period) { 840 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 841 diag::err_hex_constant_requires) 842 << PP.getLangOpts().CPlusPlus << 0; 843 hadError = true; 844 } 845 return; 846 } 847 848 // Handle simple binary numbers 0b01010 849 if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) { 850 // 0b101010 is a C++1y / GCC extension. 851 PP.Diag(TokLoc, 852 PP.getLangOpts().CPlusPlus14 853 ? diag::warn_cxx11_compat_binary_literal 854 : PP.getLangOpts().CPlusPlus 855 ? diag::ext_binary_literal_cxx14 856 : diag::ext_binary_literal); 857 ++s; 858 assert(s < ThisTokEnd && "didn't maximally munch?"); 859 radix = 2; 860 DigitsBegin = s; 861 s = SkipBinaryDigits(s); 862 if (s == ThisTokEnd) { 863 // Done. 864 } else if (isHexDigit(*s)) { 865 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 866 diag::err_invalid_digit) << StringRef(s, 1) << 2; 867 hadError = true; 868 } 869 // Other suffixes will be diagnosed by the caller. 870 return; 871 } 872 873 // For now, the radix is set to 8. If we discover that we have a 874 // floating point constant, the radix will change to 10. Octal floating 875 // point constants are not permitted (only decimal and hexadecimal). 876 radix = 8; 877 DigitsBegin = s; 878 s = SkipOctalDigits(s); 879 if (s == ThisTokEnd) 880 return; // Done, simple octal number like 01234 881 882 // If we have some other non-octal digit that *is* a decimal digit, see if 883 // this is part of a floating point number like 094.123 or 09e1. 884 if (isDigit(*s)) { 885 const char *EndDecimal = SkipDigits(s); 886 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') { 887 s = EndDecimal; 888 radix = 10; 889 } 890 } 891 892 ParseDecimalOrOctalCommon(TokLoc); 893 } 894 895 static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) { 896 switch (Radix) { 897 case 2: 898 return NumDigits <= 64; 899 case 8: 900 return NumDigits <= 64 / 3; // Digits are groups of 3 bits. 901 case 10: 902 return NumDigits <= 19; // floor(log10(2^64)) 903 case 16: 904 return NumDigits <= 64 / 4; // Digits are groups of 4 bits. 905 default: 906 llvm_unreachable("impossible Radix"); 907 } 908 } 909 910 /// GetIntegerValue - Convert this numeric literal value to an APInt that 911 /// matches Val's input width. If there is an overflow, set Val to the low bits 912 /// of the result and return true. Otherwise, return false. 913 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) { 914 // Fast path: Compute a conservative bound on the maximum number of 915 // bits per digit in this radix. If we can't possibly overflow a 916 // uint64 based on that bound then do the simple conversion to 917 // integer. This avoids the expensive overflow checking below, and 918 // handles the common cases that matter (small decimal integers and 919 // hex/octal values which don't overflow). 920 const unsigned NumDigits = SuffixBegin - DigitsBegin; 921 if (alwaysFitsInto64Bits(radix, NumDigits)) { 922 uint64_t N = 0; 923 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr) 924 if (!isDigitSeparator(*Ptr)) 925 N = N * radix + llvm::hexDigitValue(*Ptr); 926 927 // This will truncate the value to Val's input width. Simply check 928 // for overflow by comparing. 929 Val = N; 930 return Val.getZExtValue() != N; 931 } 932 933 Val = 0; 934 const char *Ptr = DigitsBegin; 935 936 llvm::APInt RadixVal(Val.getBitWidth(), radix); 937 llvm::APInt CharVal(Val.getBitWidth(), 0); 938 llvm::APInt OldVal = Val; 939 940 bool OverflowOccurred = false; 941 while (Ptr < SuffixBegin) { 942 if (isDigitSeparator(*Ptr)) { 943 ++Ptr; 944 continue; 945 } 946 947 unsigned C = llvm::hexDigitValue(*Ptr++); 948 949 // If this letter is out of bound for this radix, reject it. 950 assert(C < radix && "NumericLiteralParser ctor should have rejected this"); 951 952 CharVal = C; 953 954 // Add the digit to the value in the appropriate radix. If adding in digits 955 // made the value smaller, then this overflowed. 956 OldVal = Val; 957 958 // Multiply by radix, did overflow occur on the multiply? 959 Val *= RadixVal; 960 OverflowOccurred |= Val.udiv(RadixVal) != OldVal; 961 962 // Add value, did overflow occur on the value? 963 // (a + b) ult b <=> overflow 964 Val += CharVal; 965 OverflowOccurred |= Val.ult(CharVal); 966 } 967 return OverflowOccurred; 968 } 969 970 llvm::APFloat::opStatus 971 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) { 972 using llvm::APFloat; 973 974 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin); 975 976 llvm::SmallString<16> Buffer; 977 StringRef Str(ThisTokBegin, n); 978 if (Str.find('\'') != StringRef::npos) { 979 Buffer.reserve(n); 980 std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer), 981 &isDigitSeparator); 982 Str = Buffer; 983 } 984 985 return Result.convertFromString(Str, APFloat::rmNearestTiesToEven); 986 } 987 988 989 /// \verbatim 990 /// user-defined-character-literal: [C++11 lex.ext] 991 /// character-literal ud-suffix 992 /// ud-suffix: 993 /// identifier 994 /// character-literal: [C++11 lex.ccon] 995 /// ' c-char-sequence ' 996 /// u' c-char-sequence ' 997 /// U' c-char-sequence ' 998 /// L' c-char-sequence ' 999 /// u8' c-char-sequence ' [C++1z lex.ccon] 1000 /// c-char-sequence: 1001 /// c-char 1002 /// c-char-sequence c-char 1003 /// c-char: 1004 /// any member of the source character set except the single-quote ', 1005 /// backslash \, or new-line character 1006 /// escape-sequence 1007 /// universal-character-name 1008 /// escape-sequence: 1009 /// simple-escape-sequence 1010 /// octal-escape-sequence 1011 /// hexadecimal-escape-sequence 1012 /// simple-escape-sequence: 1013 /// one of \' \" \? \\ \a \b \f \n \r \t \v 1014 /// octal-escape-sequence: 1015 /// \ octal-digit 1016 /// \ octal-digit octal-digit 1017 /// \ octal-digit octal-digit octal-digit 1018 /// hexadecimal-escape-sequence: 1019 /// \x hexadecimal-digit 1020 /// hexadecimal-escape-sequence hexadecimal-digit 1021 /// universal-character-name: [C++11 lex.charset] 1022 /// \u hex-quad 1023 /// \U hex-quad hex-quad 1024 /// hex-quad: 1025 /// hex-digit hex-digit hex-digit hex-digit 1026 /// \endverbatim 1027 /// 1028 CharLiteralParser::CharLiteralParser(const char *begin, const char *end, 1029 SourceLocation Loc, Preprocessor &PP, 1030 tok::TokenKind kind) { 1031 // At this point we know that the character matches the regex "(L|u|U)?'.*'". 1032 HadError = false; 1033 1034 Kind = kind; 1035 1036 const char *TokBegin = begin; 1037 1038 // Skip over wide character determinant. 1039 if (Kind != tok::char_constant) 1040 ++begin; 1041 if (Kind == tok::utf8_char_constant) 1042 ++begin; 1043 1044 // Skip over the entry quote. 1045 assert(begin[0] == '\'' && "Invalid token lexed"); 1046 ++begin; 1047 1048 // Remove an optional ud-suffix. 1049 if (end[-1] != '\'') { 1050 const char *UDSuffixEnd = end; 1051 do { 1052 --end; 1053 } while (end[-1] != '\''); 1054 // FIXME: Don't bother with this if !tok.hasUCN(). 1055 expandUCNs(UDSuffixBuf, StringRef(end, UDSuffixEnd - end)); 1056 UDSuffixOffset = end - TokBegin; 1057 } 1058 1059 // Trim the ending quote. 1060 assert(end != begin && "Invalid token lexed"); 1061 --end; 1062 1063 // FIXME: The "Value" is an uint64_t so we can handle char literals of 1064 // up to 64-bits. 1065 // FIXME: This extensively assumes that 'char' is 8-bits. 1066 assert(PP.getTargetInfo().getCharWidth() == 8 && 1067 "Assumes char is 8 bits"); 1068 assert(PP.getTargetInfo().getIntWidth() <= 64 && 1069 (PP.getTargetInfo().getIntWidth() & 7) == 0 && 1070 "Assumes sizeof(int) on target is <= 64 and a multiple of char"); 1071 assert(PP.getTargetInfo().getWCharWidth() <= 64 && 1072 "Assumes sizeof(wchar) on target is <= 64"); 1073 1074 SmallVector<uint32_t, 4> codepoint_buffer; 1075 codepoint_buffer.resize(end - begin); 1076 uint32_t *buffer_begin = &codepoint_buffer.front(); 1077 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size(); 1078 1079 // Unicode escapes representing characters that cannot be correctly 1080 // represented in a single code unit are disallowed in character literals 1081 // by this implementation. 1082 uint32_t largest_character_for_kind; 1083 if (tok::wide_char_constant == Kind) { 1084 largest_character_for_kind = 1085 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth()); 1086 } else if (tok::utf8_char_constant == Kind) { 1087 largest_character_for_kind = 0x7F; 1088 } else if (tok::utf16_char_constant == Kind) { 1089 largest_character_for_kind = 0xFFFF; 1090 } else if (tok::utf32_char_constant == Kind) { 1091 largest_character_for_kind = 0x10FFFF; 1092 } else { 1093 largest_character_for_kind = 0x7Fu; 1094 } 1095 1096 while (begin != end) { 1097 // Is this a span of non-escape characters? 1098 if (begin[0] != '\\') { 1099 char const *start = begin; 1100 do { 1101 ++begin; 1102 } while (begin != end && *begin != '\\'); 1103 1104 char const *tmp_in_start = start; 1105 uint32_t *tmp_out_start = buffer_begin; 1106 ConversionResult res = 1107 ConvertUTF8toUTF32(reinterpret_cast<UTF8 const **>(&start), 1108 reinterpret_cast<UTF8 const *>(begin), 1109 &buffer_begin, buffer_end, strictConversion); 1110 if (res != conversionOK) { 1111 // If we see bad encoding for unprefixed character literals, warn and 1112 // simply copy the byte values, for compatibility with gcc and 1113 // older versions of clang. 1114 bool NoErrorOnBadEncoding = isAscii(); 1115 unsigned Msg = diag::err_bad_character_encoding; 1116 if (NoErrorOnBadEncoding) 1117 Msg = diag::warn_bad_character_encoding; 1118 PP.Diag(Loc, Msg); 1119 if (NoErrorOnBadEncoding) { 1120 start = tmp_in_start; 1121 buffer_begin = tmp_out_start; 1122 for (; start != begin; ++start, ++buffer_begin) 1123 *buffer_begin = static_cast<uint8_t>(*start); 1124 } else { 1125 HadError = true; 1126 } 1127 } else { 1128 for (; tmp_out_start < buffer_begin; ++tmp_out_start) { 1129 if (*tmp_out_start > largest_character_for_kind) { 1130 HadError = true; 1131 PP.Diag(Loc, diag::err_character_too_large); 1132 } 1133 } 1134 } 1135 1136 continue; 1137 } 1138 // Is this a Universal Character Name escape? 1139 if (begin[1] == 'u' || begin[1] == 'U') { 1140 unsigned short UcnLen = 0; 1141 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen, 1142 FullSourceLoc(Loc, PP.getSourceManager()), 1143 &PP.getDiagnostics(), PP.getLangOpts(), true)) { 1144 HadError = true; 1145 } else if (*buffer_begin > largest_character_for_kind) { 1146 HadError = true; 1147 PP.Diag(Loc, diag::err_character_too_large); 1148 } 1149 1150 ++buffer_begin; 1151 continue; 1152 } 1153 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo()); 1154 uint64_t result = 1155 ProcessCharEscape(TokBegin, begin, end, HadError, 1156 FullSourceLoc(Loc,PP.getSourceManager()), 1157 CharWidth, &PP.getDiagnostics(), PP.getLangOpts()); 1158 *buffer_begin++ = result; 1159 } 1160 1161 unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front(); 1162 1163 if (NumCharsSoFar > 1) { 1164 if (isWide()) 1165 PP.Diag(Loc, diag::warn_extraneous_char_constant); 1166 else if (isAscii() && NumCharsSoFar == 4) 1167 PP.Diag(Loc, diag::ext_four_char_character_literal); 1168 else if (isAscii()) 1169 PP.Diag(Loc, diag::ext_multichar_character_literal); 1170 else 1171 PP.Diag(Loc, diag::err_multichar_utf_character_literal); 1172 IsMultiChar = true; 1173 } else { 1174 IsMultiChar = false; 1175 } 1176 1177 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0); 1178 1179 // Narrow character literals act as though their value is concatenated 1180 // in this implementation, but warn on overflow. 1181 bool multi_char_too_long = false; 1182 if (isAscii() && isMultiChar()) { 1183 LitVal = 0; 1184 for (size_t i = 0; i < NumCharsSoFar; ++i) { 1185 // check for enough leading zeros to shift into 1186 multi_char_too_long |= (LitVal.countLeadingZeros() < 8); 1187 LitVal <<= 8; 1188 LitVal = LitVal + (codepoint_buffer[i] & 0xFF); 1189 } 1190 } else if (NumCharsSoFar > 0) { 1191 // otherwise just take the last character 1192 LitVal = buffer_begin[-1]; 1193 } 1194 1195 if (!HadError && multi_char_too_long) { 1196 PP.Diag(Loc, diag::warn_char_constant_too_large); 1197 } 1198 1199 // Transfer the value from APInt to uint64_t 1200 Value = LitVal.getZExtValue(); 1201 1202 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1") 1203 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple 1204 // character constants are not sign extended in the this implementation: 1205 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC. 1206 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) && 1207 PP.getLangOpts().CharIsSigned) 1208 Value = (signed char)Value; 1209 } 1210 1211 /// \verbatim 1212 /// string-literal: [C++0x lex.string] 1213 /// encoding-prefix " [s-char-sequence] " 1214 /// encoding-prefix R raw-string 1215 /// encoding-prefix: 1216 /// u8 1217 /// u 1218 /// U 1219 /// L 1220 /// s-char-sequence: 1221 /// s-char 1222 /// s-char-sequence s-char 1223 /// s-char: 1224 /// any member of the source character set except the double-quote ", 1225 /// backslash \, or new-line character 1226 /// escape-sequence 1227 /// universal-character-name 1228 /// raw-string: 1229 /// " d-char-sequence ( r-char-sequence ) d-char-sequence " 1230 /// r-char-sequence: 1231 /// r-char 1232 /// r-char-sequence r-char 1233 /// r-char: 1234 /// any member of the source character set, except a right parenthesis ) 1235 /// followed by the initial d-char-sequence (which may be empty) 1236 /// followed by a double quote ". 1237 /// d-char-sequence: 1238 /// d-char 1239 /// d-char-sequence d-char 1240 /// d-char: 1241 /// any member of the basic source character set except: 1242 /// space, the left parenthesis (, the right parenthesis ), 1243 /// the backslash \, and the control characters representing horizontal 1244 /// tab, vertical tab, form feed, and newline. 1245 /// escape-sequence: [C++0x lex.ccon] 1246 /// simple-escape-sequence 1247 /// octal-escape-sequence 1248 /// hexadecimal-escape-sequence 1249 /// simple-escape-sequence: 1250 /// one of \' \" \? \\ \a \b \f \n \r \t \v 1251 /// octal-escape-sequence: 1252 /// \ octal-digit 1253 /// \ octal-digit octal-digit 1254 /// \ octal-digit octal-digit octal-digit 1255 /// hexadecimal-escape-sequence: 1256 /// \x hexadecimal-digit 1257 /// hexadecimal-escape-sequence hexadecimal-digit 1258 /// universal-character-name: 1259 /// \u hex-quad 1260 /// \U hex-quad hex-quad 1261 /// hex-quad: 1262 /// hex-digit hex-digit hex-digit hex-digit 1263 /// \endverbatim 1264 /// 1265 StringLiteralParser:: 1266 StringLiteralParser(ArrayRef<Token> StringToks, 1267 Preprocessor &PP, bool Complain) 1268 : SM(PP.getSourceManager()), Features(PP.getLangOpts()), 1269 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() :nullptr), 1270 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown), 1271 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) { 1272 init(StringToks); 1273 } 1274 1275 void StringLiteralParser::init(ArrayRef<Token> StringToks){ 1276 // The literal token may have come from an invalid source location (e.g. due 1277 // to a PCH error), in which case the token length will be 0. 1278 if (StringToks.empty() || StringToks[0].getLength() < 2) 1279 return DiagnoseLexingError(SourceLocation()); 1280 1281 // Scan all of the string portions, remember the max individual token length, 1282 // computing a bound on the concatenated string length, and see whether any 1283 // piece is a wide-string. If any of the string portions is a wide-string 1284 // literal, the result is a wide-string literal [C99 6.4.5p4]. 1285 assert(!StringToks.empty() && "expected at least one token"); 1286 MaxTokenLength = StringToks[0].getLength(); 1287 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!"); 1288 SizeBound = StringToks[0].getLength()-2; // -2 for "". 1289 Kind = StringToks[0].getKind(); 1290 1291 hadError = false; 1292 1293 // Implement Translation Phase #6: concatenation of string literals 1294 /// (C99 5.1.1.2p1). The common case is only one string fragment. 1295 for (unsigned i = 1; i != StringToks.size(); ++i) { 1296 if (StringToks[i].getLength() < 2) 1297 return DiagnoseLexingError(StringToks[i].getLocation()); 1298 1299 // The string could be shorter than this if it needs cleaning, but this is a 1300 // reasonable bound, which is all we need. 1301 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!"); 1302 SizeBound += StringToks[i].getLength()-2; // -2 for "". 1303 1304 // Remember maximum string piece length. 1305 if (StringToks[i].getLength() > MaxTokenLength) 1306 MaxTokenLength = StringToks[i].getLength(); 1307 1308 // Remember if we see any wide or utf-8/16/32 strings. 1309 // Also check for illegal concatenations. 1310 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) { 1311 if (isAscii()) { 1312 Kind = StringToks[i].getKind(); 1313 } else { 1314 if (Diags) 1315 Diags->Report(StringToks[i].getLocation(), 1316 diag::err_unsupported_string_concat); 1317 hadError = true; 1318 } 1319 } 1320 } 1321 1322 // Include space for the null terminator. 1323 ++SizeBound; 1324 1325 // TODO: K&R warning: "traditional C rejects string constant concatenation" 1326 1327 // Get the width in bytes of char/wchar_t/char16_t/char32_t 1328 CharByteWidth = getCharWidth(Kind, Target); 1329 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple"); 1330 CharByteWidth /= 8; 1331 1332 // The output buffer size needs to be large enough to hold wide characters. 1333 // This is a worst-case assumption which basically corresponds to L"" "long". 1334 SizeBound *= CharByteWidth; 1335 1336 // Size the temporary buffer to hold the result string data. 1337 ResultBuf.resize(SizeBound); 1338 1339 // Likewise, but for each string piece. 1340 SmallString<512> TokenBuf; 1341 TokenBuf.resize(MaxTokenLength); 1342 1343 // Loop over all the strings, getting their spelling, and expanding them to 1344 // wide strings as appropriate. 1345 ResultPtr = &ResultBuf[0]; // Next byte to fill in. 1346 1347 Pascal = false; 1348 1349 SourceLocation UDSuffixTokLoc; 1350 1351 for (unsigned i = 0, e = StringToks.size(); i != e; ++i) { 1352 const char *ThisTokBuf = &TokenBuf[0]; 1353 // Get the spelling of the token, which eliminates trigraphs, etc. We know 1354 // that ThisTokBuf points to a buffer that is big enough for the whole token 1355 // and 'spelled' tokens can only shrink. 1356 bool StringInvalid = false; 1357 unsigned ThisTokLen = 1358 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features, 1359 &StringInvalid); 1360 if (StringInvalid) 1361 return DiagnoseLexingError(StringToks[i].getLocation()); 1362 1363 const char *ThisTokBegin = ThisTokBuf; 1364 const char *ThisTokEnd = ThisTokBuf+ThisTokLen; 1365 1366 // Remove an optional ud-suffix. 1367 if (ThisTokEnd[-1] != '"') { 1368 const char *UDSuffixEnd = ThisTokEnd; 1369 do { 1370 --ThisTokEnd; 1371 } while (ThisTokEnd[-1] != '"'); 1372 1373 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd); 1374 1375 if (UDSuffixBuf.empty()) { 1376 if (StringToks[i].hasUCN()) 1377 expandUCNs(UDSuffixBuf, UDSuffix); 1378 else 1379 UDSuffixBuf.assign(UDSuffix); 1380 UDSuffixToken = i; 1381 UDSuffixOffset = ThisTokEnd - ThisTokBuf; 1382 UDSuffixTokLoc = StringToks[i].getLocation(); 1383 } else { 1384 SmallString<32> ExpandedUDSuffix; 1385 if (StringToks[i].hasUCN()) { 1386 expandUCNs(ExpandedUDSuffix, UDSuffix); 1387 UDSuffix = ExpandedUDSuffix; 1388 } 1389 1390 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the 1391 // result of a concatenation involving at least one user-defined-string- 1392 // literal, all the participating user-defined-string-literals shall 1393 // have the same ud-suffix. 1394 if (UDSuffixBuf != UDSuffix) { 1395 if (Diags) { 1396 SourceLocation TokLoc = StringToks[i].getLocation(); 1397 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix) 1398 << UDSuffixBuf << UDSuffix 1399 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc) 1400 << SourceRange(TokLoc, TokLoc); 1401 } 1402 hadError = true; 1403 } 1404 } 1405 } 1406 1407 // Strip the end quote. 1408 --ThisTokEnd; 1409 1410 // TODO: Input character set mapping support. 1411 1412 // Skip marker for wide or unicode strings. 1413 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') { 1414 ++ThisTokBuf; 1415 // Skip 8 of u8 marker for utf8 strings. 1416 if (ThisTokBuf[0] == '8') 1417 ++ThisTokBuf; 1418 } 1419 1420 // Check for raw string 1421 if (ThisTokBuf[0] == 'R') { 1422 ThisTokBuf += 2; // skip R" 1423 1424 const char *Prefix = ThisTokBuf; 1425 while (ThisTokBuf[0] != '(') 1426 ++ThisTokBuf; 1427 ++ThisTokBuf; // skip '(' 1428 1429 // Remove same number of characters from the end 1430 ThisTokEnd -= ThisTokBuf - Prefix; 1431 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal"); 1432 1433 // C++14 [lex.string]p4: A source-file new-line in a raw string literal 1434 // results in a new-line in the resulting execution string-literal. 1435 StringRef RemainingTokenSpan(ThisTokBuf, ThisTokEnd - ThisTokBuf); 1436 while (!RemainingTokenSpan.empty()) { 1437 // Split the string literal on \r\n boundaries. 1438 size_t CRLFPos = RemainingTokenSpan.find("\r\n"); 1439 StringRef BeforeCRLF = RemainingTokenSpan.substr(0, CRLFPos); 1440 StringRef AfterCRLF = RemainingTokenSpan.substr(CRLFPos); 1441 1442 // Copy everything before the \r\n sequence into the string literal. 1443 if (CopyStringFragment(StringToks[i], ThisTokBegin, BeforeCRLF)) 1444 hadError = true; 1445 1446 // Point into the \n inside the \r\n sequence and operate on the 1447 // remaining portion of the literal. 1448 RemainingTokenSpan = AfterCRLF.substr(1); 1449 } 1450 } else { 1451 if (ThisTokBuf[0] != '"') { 1452 // The file may have come from PCH and then changed after loading the 1453 // PCH; Fail gracefully. 1454 return DiagnoseLexingError(StringToks[i].getLocation()); 1455 } 1456 ++ThisTokBuf; // skip " 1457 1458 // Check if this is a pascal string 1459 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd && 1460 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') { 1461 1462 // If the \p sequence is found in the first token, we have a pascal string 1463 // Otherwise, if we already have a pascal string, ignore the first \p 1464 if (i == 0) { 1465 ++ThisTokBuf; 1466 Pascal = true; 1467 } else if (Pascal) 1468 ThisTokBuf += 2; 1469 } 1470 1471 while (ThisTokBuf != ThisTokEnd) { 1472 // Is this a span of non-escape characters? 1473 if (ThisTokBuf[0] != '\\') { 1474 const char *InStart = ThisTokBuf; 1475 do { 1476 ++ThisTokBuf; 1477 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\'); 1478 1479 // Copy the character span over. 1480 if (CopyStringFragment(StringToks[i], ThisTokBegin, 1481 StringRef(InStart, ThisTokBuf - InStart))) 1482 hadError = true; 1483 continue; 1484 } 1485 // Is this a Universal Character Name escape? 1486 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') { 1487 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, 1488 ResultPtr, hadError, 1489 FullSourceLoc(StringToks[i].getLocation(), SM), 1490 CharByteWidth, Diags, Features); 1491 continue; 1492 } 1493 // Otherwise, this is a non-UCN escape character. Process it. 1494 unsigned ResultChar = 1495 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError, 1496 FullSourceLoc(StringToks[i].getLocation(), SM), 1497 CharByteWidth*8, Diags, Features); 1498 1499 if (CharByteWidth == 4) { 1500 // FIXME: Make the type of the result buffer correct instead of 1501 // using reinterpret_cast. 1502 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultPtr); 1503 *ResultWidePtr = ResultChar; 1504 ResultPtr += 4; 1505 } else if (CharByteWidth == 2) { 1506 // FIXME: Make the type of the result buffer correct instead of 1507 // using reinterpret_cast. 1508 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultPtr); 1509 *ResultWidePtr = ResultChar & 0xFFFF; 1510 ResultPtr += 2; 1511 } else { 1512 assert(CharByteWidth == 1 && "Unexpected char width"); 1513 *ResultPtr++ = ResultChar & 0xFF; 1514 } 1515 } 1516 } 1517 } 1518 1519 if (Pascal) { 1520 if (CharByteWidth == 4) { 1521 // FIXME: Make the type of the result buffer correct instead of 1522 // using reinterpret_cast. 1523 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultBuf.data()); 1524 ResultWidePtr[0] = GetNumStringChars() - 1; 1525 } else if (CharByteWidth == 2) { 1526 // FIXME: Make the type of the result buffer correct instead of 1527 // using reinterpret_cast. 1528 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultBuf.data()); 1529 ResultWidePtr[0] = GetNumStringChars() - 1; 1530 } else { 1531 assert(CharByteWidth == 1 && "Unexpected char width"); 1532 ResultBuf[0] = GetNumStringChars() - 1; 1533 } 1534 1535 // Verify that pascal strings aren't too large. 1536 if (GetStringLength() > 256) { 1537 if (Diags) 1538 Diags->Report(StringToks.front().getLocation(), 1539 diag::err_pascal_string_too_long) 1540 << SourceRange(StringToks.front().getLocation(), 1541 StringToks.back().getLocation()); 1542 hadError = true; 1543 return; 1544 } 1545 } else if (Diags) { 1546 // Complain if this string literal has too many characters. 1547 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509; 1548 1549 if (GetNumStringChars() > MaxChars) 1550 Diags->Report(StringToks.front().getLocation(), 1551 diag::ext_string_too_long) 1552 << GetNumStringChars() << MaxChars 1553 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0) 1554 << SourceRange(StringToks.front().getLocation(), 1555 StringToks.back().getLocation()); 1556 } 1557 } 1558 1559 static const char *resyncUTF8(const char *Err, const char *End) { 1560 if (Err == End) 1561 return End; 1562 End = Err + std::min<unsigned>(getNumBytesForUTF8(*Err), End-Err); 1563 while (++Err != End && (*Err & 0xC0) == 0x80) 1564 ; 1565 return Err; 1566 } 1567 1568 /// \brief This function copies from Fragment, which is a sequence of bytes 1569 /// within Tok's contents (which begin at TokBegin) into ResultPtr. 1570 /// Performs widening for multi-byte characters. 1571 bool StringLiteralParser::CopyStringFragment(const Token &Tok, 1572 const char *TokBegin, 1573 StringRef Fragment) { 1574 const UTF8 *ErrorPtrTmp; 1575 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp)) 1576 return false; 1577 1578 // If we see bad encoding for unprefixed string literals, warn and 1579 // simply copy the byte values, for compatibility with gcc and older 1580 // versions of clang. 1581 bool NoErrorOnBadEncoding = isAscii(); 1582 if (NoErrorOnBadEncoding) { 1583 memcpy(ResultPtr, Fragment.data(), Fragment.size()); 1584 ResultPtr += Fragment.size(); 1585 } 1586 1587 if (Diags) { 1588 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1589 1590 FullSourceLoc SourceLoc(Tok.getLocation(), SM); 1591 const DiagnosticBuilder &Builder = 1592 Diag(Diags, Features, SourceLoc, TokBegin, 1593 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()), 1594 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding 1595 : diag::err_bad_string_encoding); 1596 1597 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1598 StringRef NextFragment(NextStart, Fragment.end()-NextStart); 1599 1600 // Decode into a dummy buffer. 1601 SmallString<512> Dummy; 1602 Dummy.reserve(Fragment.size() * CharByteWidth); 1603 char *Ptr = Dummy.data(); 1604 1605 while (!ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) { 1606 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1607 NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1608 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin, 1609 ErrorPtr, NextStart); 1610 NextFragment = StringRef(NextStart, Fragment.end()-NextStart); 1611 } 1612 } 1613 return !NoErrorOnBadEncoding; 1614 } 1615 1616 void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) { 1617 hadError = true; 1618 if (Diags) 1619 Diags->Report(Loc, diag::err_lexing_string); 1620 } 1621 1622 /// getOffsetOfStringByte - This function returns the offset of the 1623 /// specified byte of the string data represented by Token. This handles 1624 /// advancing over escape sequences in the string. 1625 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok, 1626 unsigned ByteNo) const { 1627 // Get the spelling of the token. 1628 SmallString<32> SpellingBuffer; 1629 SpellingBuffer.resize(Tok.getLength()); 1630 1631 bool StringInvalid = false; 1632 const char *SpellingPtr = &SpellingBuffer[0]; 1633 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features, 1634 &StringInvalid); 1635 if (StringInvalid) 1636 return 0; 1637 1638 const char *SpellingStart = SpellingPtr; 1639 const char *SpellingEnd = SpellingPtr+TokLen; 1640 1641 // Handle UTF-8 strings just like narrow strings. 1642 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8') 1643 SpellingPtr += 2; 1644 1645 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' && 1646 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet"); 1647 1648 // For raw string literals, this is easy. 1649 if (SpellingPtr[0] == 'R') { 1650 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!"); 1651 // Skip 'R"'. 1652 SpellingPtr += 2; 1653 while (*SpellingPtr != '(') { 1654 ++SpellingPtr; 1655 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal"); 1656 } 1657 // Skip '('. 1658 ++SpellingPtr; 1659 return SpellingPtr - SpellingStart + ByteNo; 1660 } 1661 1662 // Skip over the leading quote 1663 assert(SpellingPtr[0] == '"' && "Should be a string literal!"); 1664 ++SpellingPtr; 1665 1666 // Skip over bytes until we find the offset we're looking for. 1667 while (ByteNo) { 1668 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!"); 1669 1670 // Step over non-escapes simply. 1671 if (*SpellingPtr != '\\') { 1672 ++SpellingPtr; 1673 --ByteNo; 1674 continue; 1675 } 1676 1677 // Otherwise, this is an escape character. Advance over it. 1678 bool HadError = false; 1679 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') { 1680 const char *EscapePtr = SpellingPtr; 1681 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd, 1682 1, Features, HadError); 1683 if (Len > ByteNo) { 1684 // ByteNo is somewhere within the escape sequence. 1685 SpellingPtr = EscapePtr; 1686 break; 1687 } 1688 ByteNo -= Len; 1689 } else { 1690 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError, 1691 FullSourceLoc(Tok.getLocation(), SM), 1692 CharByteWidth*8, Diags, Features); 1693 --ByteNo; 1694 } 1695 assert(!HadError && "This method isn't valid on erroneous strings"); 1696 } 1697 1698 return SpellingPtr-SpellingStart; 1699 } 1700