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 24using namespace clang; 25 26static 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_string_literal: 32 return Target.getCharWidth(); 33 case tok::wide_char_constant: 34 case tok::wide_string_literal: 35 return Target.getWCharWidth(); 36 case tok::utf16_char_constant: 37 case tok::utf16_string_literal: 38 return Target.getChar16Width(); 39 case tok::utf32_char_constant: 40 case tok::utf32_string_literal: 41 return Target.getChar32Width(); 42 } 43} 44 45static CharSourceRange MakeCharSourceRange(const LangOptions &Features, 46 FullSourceLoc TokLoc, 47 const char *TokBegin, 48 const char *TokRangeBegin, 49 const char *TokRangeEnd) { 50 SourceLocation Begin = 51 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin, 52 TokLoc.getManager(), Features); 53 SourceLocation End = 54 Lexer::AdvanceToTokenCharacter(Begin, TokRangeEnd - TokRangeBegin, 55 TokLoc.getManager(), Features); 56 return CharSourceRange::getCharRange(Begin, End); 57} 58 59/// \brief Produce a diagnostic highlighting some portion of a literal. 60/// 61/// Emits the diagnostic \p DiagID, highlighting the range of characters from 62/// \p TokRangeBegin (inclusive) to \p TokRangeEnd (exclusive), which must be 63/// a substring of a spelling buffer for the token beginning at \p TokBegin. 64static DiagnosticBuilder Diag(DiagnosticsEngine *Diags, 65 const LangOptions &Features, FullSourceLoc TokLoc, 66 const char *TokBegin, const char *TokRangeBegin, 67 const char *TokRangeEnd, unsigned DiagID) { 68 SourceLocation Begin = 69 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin, 70 TokLoc.getManager(), Features); 71 return Diags->Report(Begin, DiagID) << 72 MakeCharSourceRange(Features, TokLoc, TokBegin, TokRangeBegin, TokRangeEnd); 73} 74 75/// ProcessCharEscape - Parse a standard C escape sequence, which can occur in 76/// either a character or a string literal. 77static unsigned ProcessCharEscape(const char *ThisTokBegin, 78 const char *&ThisTokBuf, 79 const char *ThisTokEnd, bool &HadError, 80 FullSourceLoc Loc, unsigned CharWidth, 81 DiagnosticsEngine *Diags, 82 const LangOptions &Features) { 83 const char *EscapeBegin = ThisTokBuf; 84 85 // Skip the '\' char. 86 ++ThisTokBuf; 87 88 // We know that this character can't be off the end of the buffer, because 89 // that would have been \", which would not have been the end of string. 90 unsigned ResultChar = *ThisTokBuf++; 91 switch (ResultChar) { 92 // These map to themselves. 93 case '\\': case '\'': case '"': case '?': break; 94 95 // These have fixed mappings. 96 case 'a': 97 // TODO: K&R: the meaning of '\\a' is different in traditional C 98 ResultChar = 7; 99 break; 100 case 'b': 101 ResultChar = 8; 102 break; 103 case 'e': 104 if (Diags) 105 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 106 diag::ext_nonstandard_escape) << "e"; 107 ResultChar = 27; 108 break; 109 case 'E': 110 if (Diags) 111 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 112 diag::ext_nonstandard_escape) << "E"; 113 ResultChar = 27; 114 break; 115 case 'f': 116 ResultChar = 12; 117 break; 118 case 'n': 119 ResultChar = 10; 120 break; 121 case 'r': 122 ResultChar = 13; 123 break; 124 case 't': 125 ResultChar = 9; 126 break; 127 case 'v': 128 ResultChar = 11; 129 break; 130 case 'x': { // Hex escape. 131 ResultChar = 0; 132 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) { 133 if (Diags) 134 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 135 diag::err_hex_escape_no_digits) << "x"; 136 HadError = 1; 137 break; 138 } 139 140 // Hex escapes are a maximal series of hex digits. 141 bool Overflow = false; 142 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) { 143 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]); 144 if (CharVal == -1) break; 145 // About to shift out a digit? 146 Overflow |= (ResultChar & 0xF0000000) ? true : false; 147 ResultChar <<= 4; 148 ResultChar |= CharVal; 149 } 150 151 // See if any bits will be truncated when evaluated as a character. 152 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 153 Overflow = true; 154 ResultChar &= ~0U >> (32-CharWidth); 155 } 156 157 // Check for overflow. 158 if (Overflow && Diags) // Too many digits to fit in 159 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 160 diag::err_hex_escape_too_large); 161 break; 162 } 163 case '0': case '1': case '2': case '3': 164 case '4': case '5': case '6': case '7': { 165 // Octal escapes. 166 --ThisTokBuf; 167 ResultChar = 0; 168 169 // Octal escapes are a series of octal digits with maximum length 3. 170 // "\0123" is a two digit sequence equal to "\012" "3". 171 unsigned NumDigits = 0; 172 do { 173 ResultChar <<= 3; 174 ResultChar |= *ThisTokBuf++ - '0'; 175 ++NumDigits; 176 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 && 177 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7'); 178 179 // Check for overflow. Reject '\777', but not L'\777'. 180 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) { 181 if (Diags) 182 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 183 diag::err_octal_escape_too_large); 184 ResultChar &= ~0U >> (32-CharWidth); 185 } 186 break; 187 } 188 189 // Otherwise, these are not valid escapes. 190 case '(': case '{': case '[': case '%': 191 // GCC accepts these as extensions. We warn about them as such though. 192 if (Diags) 193 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 194 diag::ext_nonstandard_escape) 195 << std::string(1, ResultChar); 196 break; 197 default: 198 if (Diags == 0) 199 break; 200 201 if (isPrintable(ResultChar)) 202 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 203 diag::ext_unknown_escape) 204 << std::string(1, ResultChar); 205 else 206 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf, 207 diag::ext_unknown_escape) 208 << "x" + llvm::utohexstr(ResultChar); 209 break; 210 } 211 212 return ResultChar; 213} 214 215/// ProcessUCNEscape - Read the Universal Character Name, check constraints and 216/// return the UTF32. 217static bool ProcessUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 218 const char *ThisTokEnd, 219 uint32_t &UcnVal, unsigned short &UcnLen, 220 FullSourceLoc Loc, DiagnosticsEngine *Diags, 221 const LangOptions &Features, 222 bool in_char_string_literal = false) { 223 const char *UcnBegin = ThisTokBuf; 224 225 // Skip the '\u' char's. 226 ThisTokBuf += 2; 227 228 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) { 229 if (Diags) 230 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 231 diag::err_hex_escape_no_digits) << StringRef(&ThisTokBuf[-1], 1); 232 return false; 233 } 234 UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8); 235 unsigned short UcnLenSave = UcnLen; 236 for (; ThisTokBuf != ThisTokEnd && UcnLenSave; ++ThisTokBuf, UcnLenSave--) { 237 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]); 238 if (CharVal == -1) break; 239 UcnVal <<= 4; 240 UcnVal |= CharVal; 241 } 242 // If we didn't consume the proper number of digits, there is a problem. 243 if (UcnLenSave) { 244 if (Diags) 245 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 246 diag::err_ucn_escape_incomplete); 247 return false; 248 } 249 250 // Check UCN constraints (C99 6.4.3p2) [C++11 lex.charset p2] 251 if ((0xD800 <= UcnVal && UcnVal <= 0xDFFF) || // surrogate codepoints 252 UcnVal > 0x10FFFF) { // maximum legal UTF32 value 253 if (Diags) 254 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 255 diag::err_ucn_escape_invalid); 256 return false; 257 } 258 259 // C++11 allows UCNs that refer to control characters and basic source 260 // characters inside character and string literals 261 if (UcnVal < 0xa0 && 262 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60)) { // $, @, ` 263 bool IsError = (!Features.CPlusPlus11 || !in_char_string_literal); 264 if (Diags) { 265 char BasicSCSChar = UcnVal; 266 if (UcnVal >= 0x20 && UcnVal < 0x7f) 267 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 268 IsError ? diag::err_ucn_escape_basic_scs : 269 diag::warn_cxx98_compat_literal_ucn_escape_basic_scs) 270 << StringRef(&BasicSCSChar, 1); 271 else 272 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 273 IsError ? diag::err_ucn_control_character : 274 diag::warn_cxx98_compat_literal_ucn_control_character); 275 } 276 if (IsError) 277 return false; 278 } 279 280 if (!Features.CPlusPlus && !Features.C99 && Diags) 281 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf, 282 diag::warn_ucn_not_valid_in_c89_literal); 283 284 return true; 285} 286 287/// MeasureUCNEscape - Determine the number of bytes within the resulting string 288/// which this UCN will occupy. 289static int MeasureUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 290 const char *ThisTokEnd, unsigned CharByteWidth, 291 const LangOptions &Features, bool &HadError) { 292 // UTF-32: 4 bytes per escape. 293 if (CharByteWidth == 4) 294 return 4; 295 296 uint32_t UcnVal = 0; 297 unsigned short UcnLen = 0; 298 FullSourceLoc Loc; 299 300 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, 301 UcnLen, Loc, 0, Features, true)) { 302 HadError = true; 303 return 0; 304 } 305 306 // UTF-16: 2 bytes for BMP, 4 bytes otherwise. 307 if (CharByteWidth == 2) 308 return UcnVal <= 0xFFFF ? 2 : 4; 309 310 // UTF-8. 311 if (UcnVal < 0x80) 312 return 1; 313 if (UcnVal < 0x800) 314 return 2; 315 if (UcnVal < 0x10000) 316 return 3; 317 return 4; 318} 319 320/// EncodeUCNEscape - Read the Universal Character Name, check constraints and 321/// convert the UTF32 to UTF8 or UTF16. This is a subroutine of 322/// StringLiteralParser. When we decide to implement UCN's for identifiers, 323/// we will likely rework our support for UCN's. 324static void EncodeUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf, 325 const char *ThisTokEnd, 326 char *&ResultBuf, bool &HadError, 327 FullSourceLoc Loc, unsigned CharByteWidth, 328 DiagnosticsEngine *Diags, 329 const LangOptions &Features) { 330 typedef uint32_t UTF32; 331 UTF32 UcnVal = 0; 332 unsigned short UcnLen = 0; 333 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, UcnLen, 334 Loc, Diags, Features, true)) { 335 HadError = true; 336 return; 337 } 338 339 assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth == 4) && 340 "only character widths of 1, 2, or 4 bytes supported"); 341 342 (void)UcnLen; 343 assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported"); 344 345 if (CharByteWidth == 4) { 346 // FIXME: Make the type of the result buffer correct instead of 347 // using reinterpret_cast. 348 UTF32 *ResultPtr = reinterpret_cast<UTF32*>(ResultBuf); 349 *ResultPtr = UcnVal; 350 ResultBuf += 4; 351 return; 352 } 353 354 if (CharByteWidth == 2) { 355 // FIXME: Make the type of the result buffer correct instead of 356 // using reinterpret_cast. 357 UTF16 *ResultPtr = reinterpret_cast<UTF16*>(ResultBuf); 358 359 if (UcnVal <= (UTF32)0xFFFF) { 360 *ResultPtr = UcnVal; 361 ResultBuf += 2; 362 return; 363 } 364 365 // Convert to UTF16. 366 UcnVal -= 0x10000; 367 *ResultPtr = 0xD800 + (UcnVal >> 10); 368 *(ResultPtr+1) = 0xDC00 + (UcnVal & 0x3FF); 369 ResultBuf += 4; 370 return; 371 } 372 373 assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters"); 374 375 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8. 376 // The conversion below was inspired by: 377 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c 378 // First, we determine how many bytes the result will require. 379 typedef uint8_t UTF8; 380 381 unsigned short bytesToWrite = 0; 382 if (UcnVal < (UTF32)0x80) 383 bytesToWrite = 1; 384 else if (UcnVal < (UTF32)0x800) 385 bytesToWrite = 2; 386 else if (UcnVal < (UTF32)0x10000) 387 bytesToWrite = 3; 388 else 389 bytesToWrite = 4; 390 391 const unsigned byteMask = 0xBF; 392 const unsigned byteMark = 0x80; 393 394 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed 395 // into the first byte, depending on how many bytes follow. 396 static const UTF8 firstByteMark[5] = { 397 0x00, 0x00, 0xC0, 0xE0, 0xF0 398 }; 399 // Finally, we write the bytes into ResultBuf. 400 ResultBuf += bytesToWrite; 401 switch (bytesToWrite) { // note: everything falls through. 402 case 4: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 403 case 3: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 404 case 2: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6; 405 case 1: *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]); 406 } 407 // Update the buffer. 408 ResultBuf += bytesToWrite; 409} 410 411 412/// integer-constant: [C99 6.4.4.1] 413/// decimal-constant integer-suffix 414/// octal-constant integer-suffix 415/// hexadecimal-constant integer-suffix 416/// binary-literal integer-suffix [GNU, C++1y] 417/// user-defined-integer-literal: [C++11 lex.ext] 418/// decimal-literal ud-suffix 419/// octal-literal ud-suffix 420/// hexadecimal-literal ud-suffix 421/// binary-literal ud-suffix [GNU, C++1y] 422/// decimal-constant: 423/// nonzero-digit 424/// decimal-constant digit 425/// octal-constant: 426/// 0 427/// octal-constant octal-digit 428/// hexadecimal-constant: 429/// hexadecimal-prefix hexadecimal-digit 430/// hexadecimal-constant hexadecimal-digit 431/// hexadecimal-prefix: one of 432/// 0x 0X 433/// binary-literal: 434/// 0b binary-digit 435/// 0B binary-digit 436/// binary-literal binary-digit 437/// integer-suffix: 438/// unsigned-suffix [long-suffix] 439/// unsigned-suffix [long-long-suffix] 440/// long-suffix [unsigned-suffix] 441/// long-long-suffix [unsigned-sufix] 442/// nonzero-digit: 443/// 1 2 3 4 5 6 7 8 9 444/// octal-digit: 445/// 0 1 2 3 4 5 6 7 446/// hexadecimal-digit: 447/// 0 1 2 3 4 5 6 7 8 9 448/// a b c d e f 449/// A B C D E F 450/// binary-digit: 451/// 0 452/// 1 453/// unsigned-suffix: one of 454/// u U 455/// long-suffix: one of 456/// l L 457/// long-long-suffix: one of 458/// ll LL 459/// 460/// floating-constant: [C99 6.4.4.2] 461/// TODO: add rules... 462/// 463NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling, 464 SourceLocation TokLoc, 465 Preprocessor &PP) 466 : PP(PP), ThisTokBegin(TokSpelling.begin()), ThisTokEnd(TokSpelling.end()) { 467 468 // This routine assumes that the range begin/end matches the regex for integer 469 // and FP constants (specifically, the 'pp-number' regex), and assumes that 470 // the byte at "*end" is both valid and not part of the regex. Because of 471 // this, it doesn't have to check for 'overscan' in various places. 472 assert(!isPreprocessingNumberBody(*ThisTokEnd) && "didn't maximally munch?"); 473 474 s = DigitsBegin = ThisTokBegin; 475 saw_exponent = false; 476 saw_period = false; 477 saw_ud_suffix = false; 478 isLong = false; 479 isUnsigned = false; 480 isLongLong = false; 481 isFloat = false; 482 isImaginary = false; 483 isMicrosoftInteger = false; 484 hadError = false; 485 486 if (*s == '0') { // parse radix 487 ParseNumberStartingWithZero(TokLoc); 488 if (hadError) 489 return; 490 } else { // the first digit is non-zero 491 radix = 10; 492 s = SkipDigits(s); 493 if (s == ThisTokEnd) { 494 // Done. 495 } else if (isHexDigit(*s) && !(*s == 'e' || *s == 'E')) { 496 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 497 diag::err_invalid_decimal_digit) << StringRef(s, 1); 498 hadError = true; 499 return; 500 } else if (*s == '.') { 501 checkSeparator(TokLoc, s, CSK_AfterDigits); 502 s++; 503 saw_period = true; 504 checkSeparator(TokLoc, s, CSK_BeforeDigits); 505 s = SkipDigits(s); 506 } 507 if ((*s == 'e' || *s == 'E')) { // exponent 508 checkSeparator(TokLoc, s, CSK_AfterDigits); 509 const char *Exponent = s; 510 s++; 511 saw_exponent = true; 512 if (*s == '+' || *s == '-') s++; // sign 513 checkSeparator(TokLoc, s, CSK_BeforeDigits); 514 const char *first_non_digit = SkipDigits(s); 515 if (first_non_digit != s) { 516 s = first_non_digit; 517 } else { 518 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent - ThisTokBegin), 519 diag::err_exponent_has_no_digits); 520 hadError = true; 521 return; 522 } 523 } 524 } 525 526 SuffixBegin = s; 527 checkSeparator(TokLoc, s, CSK_AfterDigits); 528 529 // Parse the suffix. At this point we can classify whether we have an FP or 530 // integer constant. 531 bool isFPConstant = isFloatingLiteral(); 532 const char *ImaginarySuffixLoc = 0; 533 534 // Loop over all of the characters of the suffix. If we see something bad, 535 // we break out of the loop. 536 for (; s != ThisTokEnd; ++s) { 537 switch (*s) { 538 case 'f': // FP Suffix for "float" 539 case 'F': 540 if (!isFPConstant) break; // Error for integer constant. 541 if (isFloat || isLong) break; // FF, LF invalid. 542 isFloat = true; 543 continue; // Success. 544 case 'u': 545 case 'U': 546 if (isFPConstant) break; // Error for floating constant. 547 if (isUnsigned) break; // Cannot be repeated. 548 isUnsigned = true; 549 continue; // Success. 550 case 'l': 551 case 'L': 552 if (isLong || isLongLong) break; // Cannot be repeated. 553 if (isFloat) break; // LF invalid. 554 555 // Check for long long. The L's need to be adjacent and the same case. 556 if (s+1 != ThisTokEnd && s[1] == s[0]) { 557 if (isFPConstant) break; // long long invalid for floats. 558 isLongLong = true; 559 ++s; // Eat both of them. 560 } else { 561 isLong = true; 562 } 563 continue; // Success. 564 case 'i': 565 case 'I': 566 if (PP.getLangOpts().MicrosoftExt) { 567 if (isFPConstant || isLong || isLongLong) break; 568 569 // Allow i8, i16, i32, i64, and i128. 570 if (s + 1 != ThisTokEnd) { 571 switch (s[1]) { 572 case '8': 573 s += 2; // i8 suffix 574 isMicrosoftInteger = true; 575 break; 576 case '1': 577 if (s + 2 == ThisTokEnd) break; 578 if (s[2] == '6') { 579 s += 3; // i16 suffix 580 isMicrosoftInteger = true; 581 } 582 else if (s[2] == '2') { 583 if (s + 3 == ThisTokEnd) break; 584 if (s[3] == '8') { 585 s += 4; // i128 suffix 586 isMicrosoftInteger = true; 587 } 588 } 589 break; 590 case '3': 591 if (s + 2 == ThisTokEnd) break; 592 if (s[2] == '2') { 593 s += 3; // i32 suffix 594 isLong = true; 595 isMicrosoftInteger = true; 596 } 597 break; 598 case '6': 599 if (s + 2 == ThisTokEnd) break; 600 if (s[2] == '4') { 601 s += 3; // i64 suffix 602 isLongLong = true; 603 isMicrosoftInteger = true; 604 } 605 break; 606 default: 607 break; 608 } 609 break; 610 } 611 } 612 // "i", "if", and "il" are user-defined suffixes in C++1y. 613 if (PP.getLangOpts().CPlusPlus1y && *s == 'i') 614 break; 615 // fall through. 616 case 'j': 617 case 'J': 618 if (isImaginary) break; // Cannot be repeated. 619 isImaginary = true; 620 ImaginarySuffixLoc = s; 621 continue; // Success. 622 } 623 // If we reached here, there was an error or a ud-suffix. 624 break; 625 } 626 627 if (s != ThisTokEnd) { 628 if (isValidUDSuffix(PP.getLangOpts(), 629 StringRef(SuffixBegin, ThisTokEnd - SuffixBegin))) { 630 // Any suffix pieces we might have parsed are actually part of the 631 // ud-suffix. 632 isLong = false; 633 isUnsigned = false; 634 isLongLong = false; 635 isFloat = false; 636 isImaginary = false; 637 isMicrosoftInteger = false; 638 639 saw_ud_suffix = true; 640 return; 641 } 642 643 // Report an error if there are any. 644 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin), 645 isFPConstant ? diag::err_invalid_suffix_float_constant : 646 diag::err_invalid_suffix_integer_constant) 647 << StringRef(SuffixBegin, ThisTokEnd-SuffixBegin); 648 hadError = true; 649 return; 650 } 651 652 if (isImaginary) { 653 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, 654 ImaginarySuffixLoc - ThisTokBegin), 655 diag::ext_imaginary_constant); 656 } 657} 658 659/// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved 660/// suffixes as ud-suffixes, because the diagnostic experience is better if we 661/// treat it as an invalid suffix. 662bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts, 663 StringRef Suffix) { 664 if (!LangOpts.CPlusPlus11 || Suffix.empty()) 665 return false; 666 667 // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid. 668 if (Suffix[0] == '_') 669 return true; 670 671 // In C++11, there are no library suffixes. 672 if (!LangOpts.CPlusPlus1y) 673 return false; 674 675 // In C++1y, "s", "h", "min", "ms", "us", and "ns" are used in the library. 676 // Per tweaked N3660, "il", "i", and "if" are also used in the library. 677 return llvm::StringSwitch<bool>(Suffix) 678 .Cases("h", "min", "s", true) 679 .Cases("ms", "us", "ns", true) 680 .Cases("il", "i", "if", true) 681 .Default(false); 682} 683 684void NumericLiteralParser::checkSeparator(SourceLocation TokLoc, 685 const char *Pos, 686 CheckSeparatorKind IsAfterDigits) { 687 if (IsAfterDigits == CSK_AfterDigits) { 688 if (Pos == ThisTokBegin) 689 return; 690 --Pos; 691 } else if (Pos == ThisTokEnd) 692 return; 693 694 if (isDigitSeparator(*Pos)) 695 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin), 696 diag::err_digit_separator_not_between_digits) 697 << IsAfterDigits; 698} 699 700/// ParseNumberStartingWithZero - This method is called when the first character 701/// of the number is found to be a zero. This means it is either an octal 702/// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or 703/// a floating point number (01239.123e4). Eat the prefix, determining the 704/// radix etc. 705void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) { 706 assert(s[0] == '0' && "Invalid method call"); 707 s++; 708 709 int c1 = s[0]; 710 int c2 = s[1]; 711 712 // Handle a hex number like 0x1234. 713 if ((c1 == 'x' || c1 == 'X') && (isHexDigit(c2) || c2 == '.')) { 714 s++; 715 radix = 16; 716 DigitsBegin = s; 717 s = SkipHexDigits(s); 718 bool noSignificand = (s == DigitsBegin); 719 if (s == ThisTokEnd) { 720 // Done. 721 } else if (*s == '.') { 722 s++; 723 saw_period = true; 724 const char *floatDigitsBegin = s; 725 s = SkipHexDigits(s); 726 noSignificand &= (floatDigitsBegin == s); 727 } 728 729 if (noSignificand) { 730 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin), 731 diag::err_hexconstant_requires_digits); 732 hadError = true; 733 return; 734 } 735 736 // A binary exponent can appear with or with a '.'. If dotted, the 737 // binary exponent is required. 738 if (*s == 'p' || *s == 'P') { 739 const char *Exponent = s; 740 s++; 741 saw_exponent = true; 742 if (*s == '+' || *s == '-') s++; // sign 743 const char *first_non_digit = SkipDigits(s); 744 if (first_non_digit == s) { 745 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 746 diag::err_exponent_has_no_digits); 747 hadError = true; 748 return; 749 } 750 s = first_non_digit; 751 752 if (!PP.getLangOpts().HexFloats) 753 PP.Diag(TokLoc, diag::ext_hexconstant_invalid); 754 } else if (saw_period) { 755 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 756 diag::err_hexconstant_requires_exponent); 757 hadError = true; 758 } 759 return; 760 } 761 762 // Handle simple binary numbers 0b01010 763 if ((c1 == 'b' || c1 == 'B') && (c2 == '0' || c2 == '1')) { 764 // 0b101010 is a C++1y / GCC extension. 765 PP.Diag(TokLoc, 766 PP.getLangOpts().CPlusPlus1y 767 ? diag::warn_cxx11_compat_binary_literal 768 : PP.getLangOpts().CPlusPlus 769 ? diag::ext_binary_literal_cxx1y 770 : diag::ext_binary_literal); 771 ++s; 772 radix = 2; 773 DigitsBegin = s; 774 s = SkipBinaryDigits(s); 775 if (s == ThisTokEnd) { 776 // Done. 777 } else if (isHexDigit(*s)) { 778 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 779 diag::err_invalid_binary_digit) << StringRef(s, 1); 780 hadError = true; 781 } 782 // Other suffixes will be diagnosed by the caller. 783 return; 784 } 785 786 // For now, the radix is set to 8. If we discover that we have a 787 // floating point constant, the radix will change to 10. Octal floating 788 // point constants are not permitted (only decimal and hexadecimal). 789 radix = 8; 790 DigitsBegin = s; 791 s = SkipOctalDigits(s); 792 if (s == ThisTokEnd) 793 return; // Done, simple octal number like 01234 794 795 // If we have some other non-octal digit that *is* a decimal digit, see if 796 // this is part of a floating point number like 094.123 or 09e1. 797 if (isDigit(*s)) { 798 const char *EndDecimal = SkipDigits(s); 799 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') { 800 s = EndDecimal; 801 radix = 10; 802 } 803 } 804 805 // If we have a hex digit other than 'e' (which denotes a FP exponent) then 806 // the code is using an incorrect base. 807 if (isHexDigit(*s) && *s != 'e' && *s != 'E') { 808 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin), 809 diag::err_invalid_octal_digit) << StringRef(s, 1); 810 hadError = true; 811 return; 812 } 813 814 if (*s == '.') { 815 s++; 816 radix = 10; 817 saw_period = true; 818 s = SkipDigits(s); // Skip suffix. 819 } 820 if (*s == 'e' || *s == 'E') { // exponent 821 const char *Exponent = s; 822 s++; 823 radix = 10; 824 saw_exponent = true; 825 if (*s == '+' || *s == '-') s++; // sign 826 const char *first_non_digit = SkipDigits(s); 827 if (first_non_digit != s) { 828 s = first_non_digit; 829 } else { 830 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin), 831 diag::err_exponent_has_no_digits); 832 hadError = true; 833 return; 834 } 835 } 836} 837 838static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) { 839 switch (Radix) { 840 case 2: 841 return NumDigits <= 64; 842 case 8: 843 return NumDigits <= 64 / 3; // Digits are groups of 3 bits. 844 case 10: 845 return NumDigits <= 19; // floor(log10(2^64)) 846 case 16: 847 return NumDigits <= 64 / 4; // Digits are groups of 4 bits. 848 default: 849 llvm_unreachable("impossible Radix"); 850 } 851} 852 853/// GetIntegerValue - Convert this numeric literal value to an APInt that 854/// matches Val's input width. If there is an overflow, set Val to the low bits 855/// of the result and return true. Otherwise, return false. 856bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) { 857 // Fast path: Compute a conservative bound on the maximum number of 858 // bits per digit in this radix. If we can't possibly overflow a 859 // uint64 based on that bound then do the simple conversion to 860 // integer. This avoids the expensive overflow checking below, and 861 // handles the common cases that matter (small decimal integers and 862 // hex/octal values which don't overflow). 863 const unsigned NumDigits = SuffixBegin - DigitsBegin; 864 if (alwaysFitsInto64Bits(radix, NumDigits)) { 865 uint64_t N = 0; 866 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr) 867 if (!isDigitSeparator(*Ptr)) 868 N = N * radix + llvm::hexDigitValue(*Ptr); 869 870 // This will truncate the value to Val's input width. Simply check 871 // for overflow by comparing. 872 Val = N; 873 return Val.getZExtValue() != N; 874 } 875 876 Val = 0; 877 const char *Ptr = DigitsBegin; 878 879 llvm::APInt RadixVal(Val.getBitWidth(), radix); 880 llvm::APInt CharVal(Val.getBitWidth(), 0); 881 llvm::APInt OldVal = Val; 882 883 bool OverflowOccurred = false; 884 while (Ptr < SuffixBegin) { 885 if (isDigitSeparator(*Ptr)) { 886 ++Ptr; 887 continue; 888 } 889 890 unsigned C = llvm::hexDigitValue(*Ptr++); 891 892 // If this letter is out of bound for this radix, reject it. 893 assert(C < radix && "NumericLiteralParser ctor should have rejected this"); 894 895 CharVal = C; 896 897 // Add the digit to the value in the appropriate radix. If adding in digits 898 // made the value smaller, then this overflowed. 899 OldVal = Val; 900 901 // Multiply by radix, did overflow occur on the multiply? 902 Val *= RadixVal; 903 OverflowOccurred |= Val.udiv(RadixVal) != OldVal; 904 905 // Add value, did overflow occur on the value? 906 // (a + b) ult b <=> overflow 907 Val += CharVal; 908 OverflowOccurred |= Val.ult(CharVal); 909 } 910 return OverflowOccurred; 911} 912 913llvm::APFloat::opStatus 914NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) { 915 using llvm::APFloat; 916 917 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin); 918 919 llvm::SmallString<16> Buffer; 920 StringRef Str(ThisTokBegin, n); 921 if (Str.find('\'') != StringRef::npos) { 922 Buffer.reserve(n); 923 std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer), 924 &isDigitSeparator); 925 Str = Buffer; 926 } 927 928 return Result.convertFromString(Str, APFloat::rmNearestTiesToEven); 929} 930 931 932/// \verbatim 933/// user-defined-character-literal: [C++11 lex.ext] 934/// character-literal ud-suffix 935/// ud-suffix: 936/// identifier 937/// character-literal: [C++11 lex.ccon] 938/// ' c-char-sequence ' 939/// u' c-char-sequence ' 940/// U' c-char-sequence ' 941/// L' c-char-sequence ' 942/// c-char-sequence: 943/// c-char 944/// c-char-sequence c-char 945/// c-char: 946/// any member of the source character set except the single-quote ', 947/// backslash \, or new-line character 948/// escape-sequence 949/// universal-character-name 950/// escape-sequence: 951/// simple-escape-sequence 952/// octal-escape-sequence 953/// hexadecimal-escape-sequence 954/// simple-escape-sequence: 955/// one of \' \" \? \\ \a \b \f \n \r \t \v 956/// octal-escape-sequence: 957/// \ octal-digit 958/// \ octal-digit octal-digit 959/// \ octal-digit octal-digit octal-digit 960/// hexadecimal-escape-sequence: 961/// \x hexadecimal-digit 962/// hexadecimal-escape-sequence hexadecimal-digit 963/// universal-character-name: [C++11 lex.charset] 964/// \u hex-quad 965/// \U hex-quad hex-quad 966/// hex-quad: 967/// hex-digit hex-digit hex-digit hex-digit 968/// \endverbatim 969/// 970CharLiteralParser::CharLiteralParser(const char *begin, const char *end, 971 SourceLocation Loc, Preprocessor &PP, 972 tok::TokenKind kind) { 973 // At this point we know that the character matches the regex "(L|u|U)?'.*'". 974 HadError = false; 975 976 Kind = kind; 977 978 const char *TokBegin = begin; 979 980 // Skip over wide character determinant. 981 if (Kind != tok::char_constant) { 982 ++begin; 983 } 984 985 // Skip over the entry quote. 986 assert(begin[0] == '\'' && "Invalid token lexed"); 987 ++begin; 988 989 // Remove an optional ud-suffix. 990 if (end[-1] != '\'') { 991 const char *UDSuffixEnd = end; 992 do { 993 --end; 994 } while (end[-1] != '\''); 995 UDSuffixBuf.assign(end, UDSuffixEnd); 996 UDSuffixOffset = end - TokBegin; 997 } 998 999 // Trim the ending quote. 1000 assert(end != begin && "Invalid token lexed"); 1001 --end; 1002 1003 // FIXME: The "Value" is an uint64_t so we can handle char literals of 1004 // up to 64-bits. 1005 // FIXME: This extensively assumes that 'char' is 8-bits. 1006 assert(PP.getTargetInfo().getCharWidth() == 8 && 1007 "Assumes char is 8 bits"); 1008 assert(PP.getTargetInfo().getIntWidth() <= 64 && 1009 (PP.getTargetInfo().getIntWidth() & 7) == 0 && 1010 "Assumes sizeof(int) on target is <= 64 and a multiple of char"); 1011 assert(PP.getTargetInfo().getWCharWidth() <= 64 && 1012 "Assumes sizeof(wchar) on target is <= 64"); 1013 1014 SmallVector<uint32_t, 4> codepoint_buffer; 1015 codepoint_buffer.resize(end - begin); 1016 uint32_t *buffer_begin = &codepoint_buffer.front(); 1017 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size(); 1018 1019 // Unicode escapes representing characters that cannot be correctly 1020 // represented in a single code unit are disallowed in character literals 1021 // by this implementation. 1022 uint32_t largest_character_for_kind; 1023 if (tok::wide_char_constant == Kind) { 1024 largest_character_for_kind = 1025 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth()); 1026 } else if (tok::utf16_char_constant == Kind) { 1027 largest_character_for_kind = 0xFFFF; 1028 } else if (tok::utf32_char_constant == Kind) { 1029 largest_character_for_kind = 0x10FFFF; 1030 } else { 1031 largest_character_for_kind = 0x7Fu; 1032 } 1033 1034 while (begin != end) { 1035 // Is this a span of non-escape characters? 1036 if (begin[0] != '\\') { 1037 char const *start = begin; 1038 do { 1039 ++begin; 1040 } while (begin != end && *begin != '\\'); 1041 1042 char const *tmp_in_start = start; 1043 uint32_t *tmp_out_start = buffer_begin; 1044 ConversionResult res = 1045 ConvertUTF8toUTF32(reinterpret_cast<UTF8 const **>(&start), 1046 reinterpret_cast<UTF8 const *>(begin), 1047 &buffer_begin, buffer_end, strictConversion); 1048 if (res != conversionOK) { 1049 // If we see bad encoding for unprefixed character literals, warn and 1050 // simply copy the byte values, for compatibility with gcc and 1051 // older versions of clang. 1052 bool NoErrorOnBadEncoding = isAscii(); 1053 unsigned Msg = diag::err_bad_character_encoding; 1054 if (NoErrorOnBadEncoding) 1055 Msg = diag::warn_bad_character_encoding; 1056 PP.Diag(Loc, Msg); 1057 if (NoErrorOnBadEncoding) { 1058 start = tmp_in_start; 1059 buffer_begin = tmp_out_start; 1060 for (; start != begin; ++start, ++buffer_begin) 1061 *buffer_begin = static_cast<uint8_t>(*start); 1062 } else { 1063 HadError = true; 1064 } 1065 } else { 1066 for (; tmp_out_start < buffer_begin; ++tmp_out_start) { 1067 if (*tmp_out_start > largest_character_for_kind) { 1068 HadError = true; 1069 PP.Diag(Loc, diag::err_character_too_large); 1070 } 1071 } 1072 } 1073 1074 continue; 1075 } 1076 // Is this a Universal Character Name escape? 1077 if (begin[1] == 'u' || begin[1] == 'U') { 1078 unsigned short UcnLen = 0; 1079 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen, 1080 FullSourceLoc(Loc, PP.getSourceManager()), 1081 &PP.getDiagnostics(), PP.getLangOpts(), true)) { 1082 HadError = true; 1083 } else if (*buffer_begin > largest_character_for_kind) { 1084 HadError = true; 1085 PP.Diag(Loc, diag::err_character_too_large); 1086 } 1087 1088 ++buffer_begin; 1089 continue; 1090 } 1091 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo()); 1092 uint64_t result = 1093 ProcessCharEscape(TokBegin, begin, end, HadError, 1094 FullSourceLoc(Loc,PP.getSourceManager()), 1095 CharWidth, &PP.getDiagnostics(), PP.getLangOpts()); 1096 *buffer_begin++ = result; 1097 } 1098 1099 unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front(); 1100 1101 if (NumCharsSoFar > 1) { 1102 if (isWide()) 1103 PP.Diag(Loc, diag::warn_extraneous_char_constant); 1104 else if (isAscii() && NumCharsSoFar == 4) 1105 PP.Diag(Loc, diag::ext_four_char_character_literal); 1106 else if (isAscii()) 1107 PP.Diag(Loc, diag::ext_multichar_character_literal); 1108 else 1109 PP.Diag(Loc, diag::err_multichar_utf_character_literal); 1110 IsMultiChar = true; 1111 } else { 1112 IsMultiChar = false; 1113 } 1114 1115 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0); 1116 1117 // Narrow character literals act as though their value is concatenated 1118 // in this implementation, but warn on overflow. 1119 bool multi_char_too_long = false; 1120 if (isAscii() && isMultiChar()) { 1121 LitVal = 0; 1122 for (size_t i = 0; i < NumCharsSoFar; ++i) { 1123 // check for enough leading zeros to shift into 1124 multi_char_too_long |= (LitVal.countLeadingZeros() < 8); 1125 LitVal <<= 8; 1126 LitVal = LitVal + (codepoint_buffer[i] & 0xFF); 1127 } 1128 } else if (NumCharsSoFar > 0) { 1129 // otherwise just take the last character 1130 LitVal = buffer_begin[-1]; 1131 } 1132 1133 if (!HadError && multi_char_too_long) { 1134 PP.Diag(Loc, diag::warn_char_constant_too_large); 1135 } 1136 1137 // Transfer the value from APInt to uint64_t 1138 Value = LitVal.getZExtValue(); 1139 1140 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1") 1141 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple 1142 // character constants are not sign extended in the this implementation: 1143 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC. 1144 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) && 1145 PP.getLangOpts().CharIsSigned) 1146 Value = (signed char)Value; 1147} 1148 1149/// \verbatim 1150/// string-literal: [C++0x lex.string] 1151/// encoding-prefix " [s-char-sequence] " 1152/// encoding-prefix R raw-string 1153/// encoding-prefix: 1154/// u8 1155/// u 1156/// U 1157/// L 1158/// s-char-sequence: 1159/// s-char 1160/// s-char-sequence s-char 1161/// s-char: 1162/// any member of the source character set except the double-quote ", 1163/// backslash \, or new-line character 1164/// escape-sequence 1165/// universal-character-name 1166/// raw-string: 1167/// " d-char-sequence ( r-char-sequence ) d-char-sequence " 1168/// r-char-sequence: 1169/// r-char 1170/// r-char-sequence r-char 1171/// r-char: 1172/// any member of the source character set, except a right parenthesis ) 1173/// followed by the initial d-char-sequence (which may be empty) 1174/// followed by a double quote ". 1175/// d-char-sequence: 1176/// d-char 1177/// d-char-sequence d-char 1178/// d-char: 1179/// any member of the basic source character set except: 1180/// space, the left parenthesis (, the right parenthesis ), 1181/// the backslash \, and the control characters representing horizontal 1182/// tab, vertical tab, form feed, and newline. 1183/// escape-sequence: [C++0x lex.ccon] 1184/// simple-escape-sequence 1185/// octal-escape-sequence 1186/// hexadecimal-escape-sequence 1187/// simple-escape-sequence: 1188/// one of \' \" \? \\ \a \b \f \n \r \t \v 1189/// octal-escape-sequence: 1190/// \ octal-digit 1191/// \ octal-digit octal-digit 1192/// \ octal-digit octal-digit octal-digit 1193/// hexadecimal-escape-sequence: 1194/// \x hexadecimal-digit 1195/// hexadecimal-escape-sequence hexadecimal-digit 1196/// universal-character-name: 1197/// \u hex-quad 1198/// \U hex-quad hex-quad 1199/// hex-quad: 1200/// hex-digit hex-digit hex-digit hex-digit 1201/// \endverbatim 1202/// 1203StringLiteralParser:: 1204StringLiteralParser(const Token *StringToks, unsigned NumStringToks, 1205 Preprocessor &PP, bool Complain) 1206 : SM(PP.getSourceManager()), Features(PP.getLangOpts()), 1207 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() : 0), 1208 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown), 1209 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) { 1210 init(StringToks, NumStringToks); 1211} 1212 1213void StringLiteralParser::init(const Token *StringToks, unsigned NumStringToks){ 1214 // The literal token may have come from an invalid source location (e.g. due 1215 // to a PCH error), in which case the token length will be 0. 1216 if (NumStringToks == 0 || StringToks[0].getLength() < 2) 1217 return DiagnoseLexingError(SourceLocation()); 1218 1219 // Scan all of the string portions, remember the max individual token length, 1220 // computing a bound on the concatenated string length, and see whether any 1221 // piece is a wide-string. If any of the string portions is a wide-string 1222 // literal, the result is a wide-string literal [C99 6.4.5p4]. 1223 assert(NumStringToks && "expected at least one token"); 1224 MaxTokenLength = StringToks[0].getLength(); 1225 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!"); 1226 SizeBound = StringToks[0].getLength()-2; // -2 for "". 1227 Kind = StringToks[0].getKind(); 1228 1229 hadError = false; 1230 1231 // Implement Translation Phase #6: concatenation of string literals 1232 /// (C99 5.1.1.2p1). The common case is only one string fragment. 1233 for (unsigned i = 1; i != NumStringToks; ++i) { 1234 if (StringToks[i].getLength() < 2) 1235 return DiagnoseLexingError(StringToks[i].getLocation()); 1236 1237 // The string could be shorter than this if it needs cleaning, but this is a 1238 // reasonable bound, which is all we need. 1239 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!"); 1240 SizeBound += StringToks[i].getLength()-2; // -2 for "". 1241 1242 // Remember maximum string piece length. 1243 if (StringToks[i].getLength() > MaxTokenLength) 1244 MaxTokenLength = StringToks[i].getLength(); 1245 1246 // Remember if we see any wide or utf-8/16/32 strings. 1247 // Also check for illegal concatenations. 1248 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) { 1249 if (isAscii()) { 1250 Kind = StringToks[i].getKind(); 1251 } else { 1252 if (Diags) 1253 Diags->Report(StringToks[i].getLocation(), 1254 diag::err_unsupported_string_concat); 1255 hadError = true; 1256 } 1257 } 1258 } 1259 1260 // Include space for the null terminator. 1261 ++SizeBound; 1262 1263 // TODO: K&R warning: "traditional C rejects string constant concatenation" 1264 1265 // Get the width in bytes of char/wchar_t/char16_t/char32_t 1266 CharByteWidth = getCharWidth(Kind, Target); 1267 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple"); 1268 CharByteWidth /= 8; 1269 1270 // The output buffer size needs to be large enough to hold wide characters. 1271 // This is a worst-case assumption which basically corresponds to L"" "long". 1272 SizeBound *= CharByteWidth; 1273 1274 // Size the temporary buffer to hold the result string data. 1275 ResultBuf.resize(SizeBound); 1276 1277 // Likewise, but for each string piece. 1278 SmallString<512> TokenBuf; 1279 TokenBuf.resize(MaxTokenLength); 1280 1281 // Loop over all the strings, getting their spelling, and expanding them to 1282 // wide strings as appropriate. 1283 ResultPtr = &ResultBuf[0]; // Next byte to fill in. 1284 1285 Pascal = false; 1286 1287 SourceLocation UDSuffixTokLoc; 1288 1289 for (unsigned i = 0, e = NumStringToks; i != e; ++i) { 1290 const char *ThisTokBuf = &TokenBuf[0]; 1291 // Get the spelling of the token, which eliminates trigraphs, etc. We know 1292 // that ThisTokBuf points to a buffer that is big enough for the whole token 1293 // and 'spelled' tokens can only shrink. 1294 bool StringInvalid = false; 1295 unsigned ThisTokLen = 1296 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features, 1297 &StringInvalid); 1298 if (StringInvalid) 1299 return DiagnoseLexingError(StringToks[i].getLocation()); 1300 1301 const char *ThisTokBegin = ThisTokBuf; 1302 const char *ThisTokEnd = ThisTokBuf+ThisTokLen; 1303 1304 // Remove an optional ud-suffix. 1305 if (ThisTokEnd[-1] != '"') { 1306 const char *UDSuffixEnd = ThisTokEnd; 1307 do { 1308 --ThisTokEnd; 1309 } while (ThisTokEnd[-1] != '"'); 1310 1311 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd); 1312 1313 if (UDSuffixBuf.empty()) { 1314 UDSuffixBuf.assign(UDSuffix); 1315 UDSuffixToken = i; 1316 UDSuffixOffset = ThisTokEnd - ThisTokBuf; 1317 UDSuffixTokLoc = StringToks[i].getLocation(); 1318 } else if (!UDSuffixBuf.equals(UDSuffix)) { 1319 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the 1320 // result of a concatenation involving at least one user-defined-string- 1321 // literal, all the participating user-defined-string-literals shall 1322 // have the same ud-suffix. 1323 if (Diags) { 1324 SourceLocation TokLoc = StringToks[i].getLocation(); 1325 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix) 1326 << UDSuffixBuf << UDSuffix 1327 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc) 1328 << SourceRange(TokLoc, TokLoc); 1329 } 1330 hadError = true; 1331 } 1332 } 1333 1334 // Strip the end quote. 1335 --ThisTokEnd; 1336 1337 // TODO: Input character set mapping support. 1338 1339 // Skip marker for wide or unicode strings. 1340 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') { 1341 ++ThisTokBuf; 1342 // Skip 8 of u8 marker for utf8 strings. 1343 if (ThisTokBuf[0] == '8') 1344 ++ThisTokBuf; 1345 } 1346 1347 // Check for raw string 1348 if (ThisTokBuf[0] == 'R') { 1349 ThisTokBuf += 2; // skip R" 1350 1351 const char *Prefix = ThisTokBuf; 1352 while (ThisTokBuf[0] != '(') 1353 ++ThisTokBuf; 1354 ++ThisTokBuf; // skip '(' 1355 1356 // Remove same number of characters from the end 1357 ThisTokEnd -= ThisTokBuf - Prefix; 1358 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal"); 1359 1360 // Copy the string over 1361 if (CopyStringFragment(StringToks[i], ThisTokBegin, 1362 StringRef(ThisTokBuf, ThisTokEnd - ThisTokBuf))) 1363 hadError = true; 1364 } else { 1365 if (ThisTokBuf[0] != '"') { 1366 // The file may have come from PCH and then changed after loading the 1367 // PCH; Fail gracefully. 1368 return DiagnoseLexingError(StringToks[i].getLocation()); 1369 } 1370 ++ThisTokBuf; // skip " 1371 1372 // Check if this is a pascal string 1373 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd && 1374 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') { 1375 1376 // If the \p sequence is found in the first token, we have a pascal string 1377 // Otherwise, if we already have a pascal string, ignore the first \p 1378 if (i == 0) { 1379 ++ThisTokBuf; 1380 Pascal = true; 1381 } else if (Pascal) 1382 ThisTokBuf += 2; 1383 } 1384 1385 while (ThisTokBuf != ThisTokEnd) { 1386 // Is this a span of non-escape characters? 1387 if (ThisTokBuf[0] != '\\') { 1388 const char *InStart = ThisTokBuf; 1389 do { 1390 ++ThisTokBuf; 1391 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\'); 1392 1393 // Copy the character span over. 1394 if (CopyStringFragment(StringToks[i], ThisTokBegin, 1395 StringRef(InStart, ThisTokBuf - InStart))) 1396 hadError = true; 1397 continue; 1398 } 1399 // Is this a Universal Character Name escape? 1400 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') { 1401 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, 1402 ResultPtr, hadError, 1403 FullSourceLoc(StringToks[i].getLocation(), SM), 1404 CharByteWidth, Diags, Features); 1405 continue; 1406 } 1407 // Otherwise, this is a non-UCN escape character. Process it. 1408 unsigned ResultChar = 1409 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError, 1410 FullSourceLoc(StringToks[i].getLocation(), SM), 1411 CharByteWidth*8, Diags, Features); 1412 1413 if (CharByteWidth == 4) { 1414 // FIXME: Make the type of the result buffer correct instead of 1415 // using reinterpret_cast. 1416 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultPtr); 1417 *ResultWidePtr = ResultChar; 1418 ResultPtr += 4; 1419 } else if (CharByteWidth == 2) { 1420 // FIXME: Make the type of the result buffer correct instead of 1421 // using reinterpret_cast. 1422 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultPtr); 1423 *ResultWidePtr = ResultChar & 0xFFFF; 1424 ResultPtr += 2; 1425 } else { 1426 assert(CharByteWidth == 1 && "Unexpected char width"); 1427 *ResultPtr++ = ResultChar & 0xFF; 1428 } 1429 } 1430 } 1431 } 1432 1433 if (Pascal) { 1434 if (CharByteWidth == 4) { 1435 // FIXME: Make the type of the result buffer correct instead of 1436 // using reinterpret_cast. 1437 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultBuf.data()); 1438 ResultWidePtr[0] = GetNumStringChars() - 1; 1439 } else if (CharByteWidth == 2) { 1440 // FIXME: Make the type of the result buffer correct instead of 1441 // using reinterpret_cast. 1442 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultBuf.data()); 1443 ResultWidePtr[0] = GetNumStringChars() - 1; 1444 } else { 1445 assert(CharByteWidth == 1 && "Unexpected char width"); 1446 ResultBuf[0] = GetNumStringChars() - 1; 1447 } 1448 1449 // Verify that pascal strings aren't too large. 1450 if (GetStringLength() > 256) { 1451 if (Diags) 1452 Diags->Report(StringToks[0].getLocation(), 1453 diag::err_pascal_string_too_long) 1454 << SourceRange(StringToks[0].getLocation(), 1455 StringToks[NumStringToks-1].getLocation()); 1456 hadError = true; 1457 return; 1458 } 1459 } else if (Diags) { 1460 // Complain if this string literal has too many characters. 1461 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509; 1462 1463 if (GetNumStringChars() > MaxChars) 1464 Diags->Report(StringToks[0].getLocation(), 1465 diag::ext_string_too_long) 1466 << GetNumStringChars() << MaxChars 1467 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0) 1468 << SourceRange(StringToks[0].getLocation(), 1469 StringToks[NumStringToks-1].getLocation()); 1470 } 1471} 1472 1473static const char *resyncUTF8(const char *Err, const char *End) { 1474 if (Err == End) 1475 return End; 1476 End = Err + std::min<unsigned>(getNumBytesForUTF8(*Err), End-Err); 1477 while (++Err != End && (*Err & 0xC0) == 0x80) 1478 ; 1479 return Err; 1480} 1481 1482/// \brief This function copies from Fragment, which is a sequence of bytes 1483/// within Tok's contents (which begin at TokBegin) into ResultPtr. 1484/// Performs widening for multi-byte characters. 1485bool StringLiteralParser::CopyStringFragment(const Token &Tok, 1486 const char *TokBegin, 1487 StringRef Fragment) { 1488 const UTF8 *ErrorPtrTmp; 1489 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp)) 1490 return false; 1491 1492 // If we see bad encoding for unprefixed string literals, warn and 1493 // simply copy the byte values, for compatibility with gcc and older 1494 // versions of clang. 1495 bool NoErrorOnBadEncoding = isAscii(); 1496 if (NoErrorOnBadEncoding) { 1497 memcpy(ResultPtr, Fragment.data(), Fragment.size()); 1498 ResultPtr += Fragment.size(); 1499 } 1500 1501 if (Diags) { 1502 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1503 1504 FullSourceLoc SourceLoc(Tok.getLocation(), SM); 1505 const DiagnosticBuilder &Builder = 1506 Diag(Diags, Features, SourceLoc, TokBegin, 1507 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()), 1508 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding 1509 : diag::err_bad_string_encoding); 1510 1511 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1512 StringRef NextFragment(NextStart, Fragment.end()-NextStart); 1513 1514 // Decode into a dummy buffer. 1515 SmallString<512> Dummy; 1516 Dummy.reserve(Fragment.size() * CharByteWidth); 1517 char *Ptr = Dummy.data(); 1518 1519 while (!Builder.hasMaxRanges() && 1520 !ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) { 1521 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp); 1522 NextStart = resyncUTF8(ErrorPtr, Fragment.end()); 1523 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin, 1524 ErrorPtr, NextStart); 1525 NextFragment = StringRef(NextStart, Fragment.end()-NextStart); 1526 } 1527 } 1528 return !NoErrorOnBadEncoding; 1529} 1530 1531void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) { 1532 hadError = true; 1533 if (Diags) 1534 Diags->Report(Loc, diag::err_lexing_string); 1535} 1536 1537/// getOffsetOfStringByte - This function returns the offset of the 1538/// specified byte of the string data represented by Token. This handles 1539/// advancing over escape sequences in the string. 1540unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok, 1541 unsigned ByteNo) const { 1542 // Get the spelling of the token. 1543 SmallString<32> SpellingBuffer; 1544 SpellingBuffer.resize(Tok.getLength()); 1545 1546 bool StringInvalid = false; 1547 const char *SpellingPtr = &SpellingBuffer[0]; 1548 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features, 1549 &StringInvalid); 1550 if (StringInvalid) 1551 return 0; 1552 1553 const char *SpellingStart = SpellingPtr; 1554 const char *SpellingEnd = SpellingPtr+TokLen; 1555 1556 // Handle UTF-8 strings just like narrow strings. 1557 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8') 1558 SpellingPtr += 2; 1559 1560 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' && 1561 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet"); 1562 1563 // For raw string literals, this is easy. 1564 if (SpellingPtr[0] == 'R') { 1565 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!"); 1566 // Skip 'R"'. 1567 SpellingPtr += 2; 1568 while (*SpellingPtr != '(') { 1569 ++SpellingPtr; 1570 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal"); 1571 } 1572 // Skip '('. 1573 ++SpellingPtr; 1574 return SpellingPtr - SpellingStart + ByteNo; 1575 } 1576 1577 // Skip over the leading quote 1578 assert(SpellingPtr[0] == '"' && "Should be a string literal!"); 1579 ++SpellingPtr; 1580 1581 // Skip over bytes until we find the offset we're looking for. 1582 while (ByteNo) { 1583 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!"); 1584 1585 // Step over non-escapes simply. 1586 if (*SpellingPtr != '\\') { 1587 ++SpellingPtr; 1588 --ByteNo; 1589 continue; 1590 } 1591 1592 // Otherwise, this is an escape character. Advance over it. 1593 bool HadError = false; 1594 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') { 1595 const char *EscapePtr = SpellingPtr; 1596 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd, 1597 1, Features, HadError); 1598 if (Len > ByteNo) { 1599 // ByteNo is somewhere within the escape sequence. 1600 SpellingPtr = EscapePtr; 1601 break; 1602 } 1603 ByteNo -= Len; 1604 } else { 1605 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError, 1606 FullSourceLoc(Tok.getLocation(), SM), 1607 CharByteWidth*8, Diags, Features); 1608 --ByteNo; 1609 } 1610 assert(!HadError && "This method isn't valid on erroneous strings"); 1611 } 1612 1613 return SpellingPtr-SpellingStart; 1614} 1615