1 //===--- LiteralSupport.cpp - Code to parse and process literals ----------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the NumericLiteralParser, CharLiteralParser, and
11 // StringLiteralParser interfaces.
13 //===----------------------------------------------------------------------===//
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"
24 using namespace clang;
26 static unsigned getCharWidth(tok::TokenKind kind, const TargetInfo &Target) {
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();
46 static CharSourceRange MakeCharSourceRange(const LangOptions &Features,
49 const char *TokRangeBegin,
50 const char *TokRangeEnd) {
51 SourceLocation Begin =
52 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin,
53 TokLoc.getManager(), Features);
55 Lexer::AdvanceToTokenCharacter(Begin, TokRangeEnd - TokRangeBegin,
56 TokLoc.getManager(), Features);
57 return CharSourceRange::getCharRange(Begin, End);
60 /// \brief Produce a diagnostic highlighting some portion of a literal.
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);
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;
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++;
93 // These map to themselves.
94 case '\\': case '\'': case '"': case '?': break;
96 // These have fixed mappings.
98 // TODO: K&R: the meaning of '\\a' is different in traditional C
106 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
107 diag::ext_nonstandard_escape) << "e";
112 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
113 diag::ext_nonstandard_escape) << "E";
131 case 'x': { // Hex escape.
133 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
135 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
136 diag::err_hex_escape_no_digits) << "x";
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)
150 ResultChar |= CharVal;
153 // See if any bits will be truncated when evaluated as a character.
154 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
156 ResultChar &= ~0U >> (32-CharWidth);
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;
165 case '0': case '1': case '2': case '3':
166 case '4': case '5': case '6': case '7': {
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;
176 ResultChar |= *ThisTokBuf++ - '0';
178 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 &&
179 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7');
181 // Check for overflow. Reject '\777', but not L'\777'.
182 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
184 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
185 diag::err_escape_too_large) << 1;
186 ResultChar &= ~0U >> (32-CharWidth);
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.
195 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
196 diag::ext_nonstandard_escape)
197 << std::string(1, ResultChar);
203 if (isPrintable(ResultChar))
204 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
205 diag::ext_unknown_escape)
206 << std::string(1, ResultChar);
208 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
209 diag::ext_unknown_escape)
210 << "x" + llvm::utohexstr(ResultChar);
217 static void appendCodePoint(unsigned Codepoint,
218 llvm::SmallVectorImpl<char> &Str) {
220 char *ResultPtr = ResultBuf;
221 bool Res = llvm::ConvertCodePointToUTF8(Codepoint, ResultPtr);
223 assert(Res && "Unexpected conversion failure");
224 Str.append(ResultBuf, ResultPtr);
227 void clang::expandUCNs(SmallVectorImpl<char> &Buf, StringRef Input) {
228 for (StringRef::iterator I = Input.begin(), E = Input.end(); I != E; ++I) {
235 assert(*I == 'u' || *I == 'U');
237 unsigned NumHexDigits;
243 assert(I + NumHexDigits <= E);
245 uint32_t CodePoint = 0;
246 for (++I; NumHexDigits != 0; ++I, --NumHexDigits) {
247 unsigned Value = llvm::hexDigitValue(*I);
248 assert(Value != -1U);
254 appendCodePoint(CodePoint, Buf);
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;
269 // Skip the '\u' char's.
272 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
274 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
275 diag::err_hex_escape_no_digits) << StringRef(&ThisTokBuf[-1], 1);
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;
286 // If we didn't consume the proper number of digits, there is a problem.
289 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
290 diag::err_ucn_escape_incomplete);
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
298 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
299 diag::err_ucn_escape_invalid);
303 // C++11 allows UCNs that refer to control characters and basic source
304 // characters inside character and string literals
306 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60)) { // $, @, `
307 bool IsError = (!Features.CPlusPlus11 || !in_char_string_literal);
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);
316 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
317 IsError ? diag::err_ucn_control_character :
318 diag::warn_cxx98_compat_literal_ucn_control_character);
324 if (!Features.CPlusPlus && !Features.C99 && Diags)
325 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
326 diag::warn_ucn_not_valid_in_c89_literal);
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)
341 unsigned short UcnLen = 0;
344 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal,
345 UcnLen, Loc, nullptr, Features, true)) {
350 // UTF-16: 2 bytes for BMP, 4 bytes otherwise.
351 if (CharByteWidth == 2)
352 return UcnVal <= 0xFFFF ? 2 : 4;
359 if (UcnVal < 0x10000)
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;
376 unsigned short UcnLen = 0;
377 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, UcnLen,
378 Loc, Diags, Features, true)) {
383 assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth == 4) &&
384 "only character widths of 1, 2, or 4 bytes supported");
387 assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported");
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);
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);
403 if (UcnVal <= (UTF32)0xFFFF) {
411 *ResultPtr = 0xD800 + (UcnVal >> 10);
412 *(ResultPtr+1) = 0xDC00 + (UcnVal & 0x3FF);
417 assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters");
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;
425 unsigned short bytesToWrite = 0;
426 if (UcnVal < (UTF32)0x80)
428 else if (UcnVal < (UTF32)0x800)
430 else if (UcnVal < (UTF32)0x10000)
435 const unsigned byteMask = 0xBF;
436 const unsigned byteMark = 0x80;
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
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]);
451 // Update the buffer.
452 ResultBuf += bytesToWrite;
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:
468 /// decimal-constant digit
471 /// octal-constant octal-digit
472 /// hexadecimal-constant:
473 /// hexadecimal-prefix hexadecimal-digit
474 /// hexadecimal-constant hexadecimal-digit
475 /// hexadecimal-prefix: one of
480 /// binary-literal binary-digit
482 /// unsigned-suffix [long-suffix]
483 /// unsigned-suffix [long-long-suffix]
484 /// long-suffix [unsigned-suffix]
485 /// long-long-suffix [unsigned-sufix]
487 /// 1 2 3 4 5 6 7 8 9
490 /// hexadecimal-digit:
491 /// 0 1 2 3 4 5 6 7 8 9
497 /// unsigned-suffix: one of
499 /// long-suffix: one of
501 /// long-long-suffix: one of
504 /// floating-constant: [C99 6.4.4.2]
505 /// TODO: add rules...
507 NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
508 SourceLocation TokLoc,
510 : PP(PP), ThisTokBegin(TokSpelling.begin()), ThisTokEnd(TokSpelling.end()) {
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?");
518 s = DigitsBegin = ThisTokBegin;
519 saw_exponent = false;
521 saw_ud_suffix = false;
527 MicrosoftInteger = 0;
530 if (*s == '0') { // parse radix
531 ParseNumberStartingWithZero(TokLoc);
534 } else { // the first digit is non-zero
537 if (s == ThisTokEnd) {
539 } else if (isHexDigit(*s) && !(*s == 'e' || *s == 'E')) {
540 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin),
541 diag::err_invalid_digit) << StringRef(s, 1) << 0;
544 } else if (*s == '.') {
545 checkSeparator(TokLoc, s, CSK_AfterDigits);
548 checkSeparator(TokLoc, s, CSK_BeforeDigits);
551 if ((*s == 'e' || *s == 'E')) { // exponent
552 checkSeparator(TokLoc, s, CSK_AfterDigits);
553 const char *Exponent = s;
556 if (*s == '+' || *s == '-') s++; // sign
557 checkSeparator(TokLoc, s, CSK_BeforeDigits);
558 const char *first_non_digit = SkipDigits(s);
559 if (first_non_digit != s) {
562 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent - ThisTokBegin),
563 diag::err_exponent_has_no_digits);
571 checkSeparator(TokLoc, s, CSK_AfterDigits);
573 // Parse the suffix. At this point we can classify whether we have an FP or
575 bool isFPConstant = isFloatingLiteral();
576 const char *ImaginarySuffixLoc = nullptr;
578 // Loop over all of the characters of the suffix. If we see something bad,
579 // we break out of the loop.
580 for (; s != ThisTokEnd; ++s) {
582 case 'f': // FP Suffix for "float"
584 if (!isFPConstant) break; // Error for integer constant.
585 if (isFloat || isLong) break; // FF, LF invalid.
587 continue; // Success.
590 if (isFPConstant) break; // Error for floating constant.
591 if (isUnsigned) break; // Cannot be repeated.
593 continue; // Success.
596 if (isLong || isLongLong) break; // Cannot be repeated.
597 if (isFloat) break; // LF invalid.
599 // Check for long long. The L's need to be adjacent and the same case.
601 assert(s + 1 < ThisTokEnd && "didn't maximally munch?");
602 if (isFPConstant) break; // long long invalid for floats.
604 ++s; // Eat both of them.
608 continue; // Success.
611 if (PP.getLangOpts().MicrosoftExt) {
612 if (isLong || isLongLong || MicrosoftInteger)
616 // Allow i8, i16, i32, and i64.
620 MicrosoftInteger = 8;
624 s += 3; // i16 suffix
625 MicrosoftInteger = 16;
630 s += 3; // i32 suffix
631 MicrosoftInteger = 32;
636 s += 3; // i64 suffix
637 MicrosoftInteger = 64;
644 if (MicrosoftInteger) {
645 assert(s <= ThisTokEnd && "didn't maximally munch?");
649 // "i", "if", and "il" are user-defined suffixes in C++1y.
650 if (*s == 'i' && PP.getLangOpts().CPlusPlus14)
655 if (isImaginary) break; // Cannot be repeated.
657 ImaginarySuffixLoc = s;
658 continue; // Success.
660 // If we reached here, there was an error or a ud-suffix.
664 if (s != ThisTokEnd) {
665 // FIXME: Don't bother expanding UCNs if !tok.hasUCN().
666 expandUCNs(UDSuffixBuf, StringRef(SuffixBegin, ThisTokEnd - SuffixBegin));
667 if (isValidUDSuffix(PP.getLangOpts(), UDSuffixBuf)) {
668 // Any suffix pieces we might have parsed are actually part of the
675 MicrosoftInteger = 0;
677 saw_ud_suffix = true;
681 // Report an error if there are any.
682 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin),
683 diag::err_invalid_suffix_constant)
684 << StringRef(SuffixBegin, ThisTokEnd-SuffixBegin) << isFPConstant;
690 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc,
691 ImaginarySuffixLoc - ThisTokBegin),
692 diag::ext_imaginary_constant);
696 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
697 /// suffixes as ud-suffixes, because the diagnostic experience is better if we
698 /// treat it as an invalid suffix.
699 bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
701 if (!LangOpts.CPlusPlus11 || Suffix.empty())
704 // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid.
705 if (Suffix[0] == '_')
708 // In C++11, there are no library suffixes.
709 if (!LangOpts.CPlusPlus14)
712 // In C++1y, "s", "h", "min", "ms", "us", and "ns" are used in the library.
713 // Per tweaked N3660, "il", "i", and "if" are also used in the library.
714 return llvm::StringSwitch<bool>(Suffix)
715 .Cases("h", "min", "s", true)
716 .Cases("ms", "us", "ns", true)
717 .Cases("il", "i", "if", true)
721 void NumericLiteralParser::checkSeparator(SourceLocation TokLoc,
723 CheckSeparatorKind IsAfterDigits) {
724 if (IsAfterDigits == CSK_AfterDigits) {
725 if (Pos == ThisTokBegin)
728 } else if (Pos == ThisTokEnd)
731 if (isDigitSeparator(*Pos))
732 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin),
733 diag::err_digit_separator_not_between_digits)
737 /// ParseNumberStartingWithZero - This method is called when the first character
738 /// of the number is found to be a zero. This means it is either an octal
739 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or
740 /// a floating point number (01239.123e4). Eat the prefix, determining the
742 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
743 assert(s[0] == '0' && "Invalid method call");
748 // Handle a hex number like 0x1234.
749 if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) {
751 assert(s < ThisTokEnd && "didn't maximally munch?");
754 s = SkipHexDigits(s);
755 bool noSignificand = (s == DigitsBegin);
756 if (s == ThisTokEnd) {
758 } else if (*s == '.') {
761 const char *floatDigitsBegin = s;
762 checkSeparator(TokLoc, s, CSK_BeforeDigits);
763 s = SkipHexDigits(s);
764 noSignificand &= (floatDigitsBegin == s);
768 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin),
769 diag::err_hexconstant_requires) << 1;
774 // A binary exponent can appear with or with a '.'. If dotted, the
775 // binary exponent is required.
776 if (*s == 'p' || *s == 'P') {
777 checkSeparator(TokLoc, s, CSK_AfterDigits);
778 const char *Exponent = s;
781 if (*s == '+' || *s == '-') s++; // sign
782 const char *first_non_digit = SkipDigits(s);
783 if (first_non_digit == s) {
784 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
785 diag::err_exponent_has_no_digits);
789 checkSeparator(TokLoc, s, CSK_BeforeDigits);
792 if (!PP.getLangOpts().HexFloats)
793 PP.Diag(TokLoc, diag::ext_hexconstant_invalid);
794 } else if (saw_period) {
795 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
796 diag::err_hexconstant_requires) << 0;
802 // Handle simple binary numbers 0b01010
803 if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) {
804 // 0b101010 is a C++1y / GCC extension.
806 PP.getLangOpts().CPlusPlus14
807 ? diag::warn_cxx11_compat_binary_literal
808 : PP.getLangOpts().CPlusPlus
809 ? diag::ext_binary_literal_cxx14
810 : diag::ext_binary_literal);
812 assert(s < ThisTokEnd && "didn't maximally munch?");
815 s = SkipBinaryDigits(s);
816 if (s == ThisTokEnd) {
818 } else if (isHexDigit(*s)) {
819 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
820 diag::err_invalid_digit) << StringRef(s, 1) << 2;
823 // Other suffixes will be diagnosed by the caller.
827 // For now, the radix is set to 8. If we discover that we have a
828 // floating point constant, the radix will change to 10. Octal floating
829 // point constants are not permitted (only decimal and hexadecimal).
832 s = SkipOctalDigits(s);
834 return; // Done, simple octal number like 01234
836 // If we have some other non-octal digit that *is* a decimal digit, see if
837 // this is part of a floating point number like 094.123 or 09e1.
839 const char *EndDecimal = SkipDigits(s);
840 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') {
846 // If we have a hex digit other than 'e' (which denotes a FP exponent) then
847 // the code is using an incorrect base.
848 if (isHexDigit(*s) && *s != 'e' && *s != 'E') {
849 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
850 diag::err_invalid_digit) << StringRef(s, 1) << 1;
859 checkSeparator(TokLoc, s, CSK_BeforeDigits);
860 s = SkipDigits(s); // Skip suffix.
862 if (*s == 'e' || *s == 'E') { // exponent
863 checkSeparator(TokLoc, s, CSK_AfterDigits);
864 const char *Exponent = s;
868 if (*s == '+' || *s == '-') s++; // sign
869 const char *first_non_digit = SkipDigits(s);
870 if (first_non_digit != s) {
871 checkSeparator(TokLoc, s, CSK_BeforeDigits);
874 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
875 diag::err_exponent_has_no_digits);
882 static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) {
885 return NumDigits <= 64;
887 return NumDigits <= 64 / 3; // Digits are groups of 3 bits.
889 return NumDigits <= 19; // floor(log10(2^64))
891 return NumDigits <= 64 / 4; // Digits are groups of 4 bits.
893 llvm_unreachable("impossible Radix");
897 /// GetIntegerValue - Convert this numeric literal value to an APInt that
898 /// matches Val's input width. If there is an overflow, set Val to the low bits
899 /// of the result and return true. Otherwise, return false.
900 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) {
901 // Fast path: Compute a conservative bound on the maximum number of
902 // bits per digit in this radix. If we can't possibly overflow a
903 // uint64 based on that bound then do the simple conversion to
904 // integer. This avoids the expensive overflow checking below, and
905 // handles the common cases that matter (small decimal integers and
906 // hex/octal values which don't overflow).
907 const unsigned NumDigits = SuffixBegin - DigitsBegin;
908 if (alwaysFitsInto64Bits(radix, NumDigits)) {
910 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr)
911 if (!isDigitSeparator(*Ptr))
912 N = N * radix + llvm::hexDigitValue(*Ptr);
914 // This will truncate the value to Val's input width. Simply check
915 // for overflow by comparing.
917 return Val.getZExtValue() != N;
921 const char *Ptr = DigitsBegin;
923 llvm::APInt RadixVal(Val.getBitWidth(), radix);
924 llvm::APInt CharVal(Val.getBitWidth(), 0);
925 llvm::APInt OldVal = Val;
927 bool OverflowOccurred = false;
928 while (Ptr < SuffixBegin) {
929 if (isDigitSeparator(*Ptr)) {
934 unsigned C = llvm::hexDigitValue(*Ptr++);
936 // If this letter is out of bound for this radix, reject it.
937 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
941 // Add the digit to the value in the appropriate radix. If adding in digits
942 // made the value smaller, then this overflowed.
945 // Multiply by radix, did overflow occur on the multiply?
947 OverflowOccurred |= Val.udiv(RadixVal) != OldVal;
949 // Add value, did overflow occur on the value?
950 // (a + b) ult b <=> overflow
952 OverflowOccurred |= Val.ult(CharVal);
954 return OverflowOccurred;
957 llvm::APFloat::opStatus
958 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) {
961 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin);
963 llvm::SmallString<16> Buffer;
964 StringRef Str(ThisTokBegin, n);
965 if (Str.find('\'') != StringRef::npos) {
967 std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer),
972 return Result.convertFromString(Str, APFloat::rmNearestTiesToEven);
977 /// user-defined-character-literal: [C++11 lex.ext]
978 /// character-literal ud-suffix
981 /// character-literal: [C++11 lex.ccon]
982 /// ' c-char-sequence '
983 /// u' c-char-sequence '
984 /// U' c-char-sequence '
985 /// L' c-char-sequence '
988 /// c-char-sequence c-char
990 /// any member of the source character set except the single-quote ',
991 /// backslash \, or new-line character
993 /// universal-character-name
995 /// simple-escape-sequence
996 /// octal-escape-sequence
997 /// hexadecimal-escape-sequence
998 /// simple-escape-sequence:
999 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1000 /// octal-escape-sequence:
1002 /// \ octal-digit octal-digit
1003 /// \ octal-digit octal-digit octal-digit
1004 /// hexadecimal-escape-sequence:
1005 /// \x hexadecimal-digit
1006 /// hexadecimal-escape-sequence hexadecimal-digit
1007 /// universal-character-name: [C++11 lex.charset]
1009 /// \U hex-quad hex-quad
1011 /// hex-digit hex-digit hex-digit hex-digit
1014 CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
1015 SourceLocation Loc, Preprocessor &PP,
1016 tok::TokenKind kind) {
1017 // At this point we know that the character matches the regex "(L|u|U)?'.*'".
1022 const char *TokBegin = begin;
1024 // Skip over wide character determinant.
1025 if (Kind != tok::char_constant)
1027 if (Kind == tok::utf8_char_constant)
1030 // Skip over the entry quote.
1031 assert(begin[0] == '\'' && "Invalid token lexed");
1034 // Remove an optional ud-suffix.
1035 if (end[-1] != '\'') {
1036 const char *UDSuffixEnd = end;
1039 } while (end[-1] != '\'');
1040 // FIXME: Don't bother with this if !tok.hasUCN().
1041 expandUCNs(UDSuffixBuf, StringRef(end, UDSuffixEnd - end));
1042 UDSuffixOffset = end - TokBegin;
1045 // Trim the ending quote.
1046 assert(end != begin && "Invalid token lexed");
1049 // FIXME: The "Value" is an uint64_t so we can handle char literals of
1051 // FIXME: This extensively assumes that 'char' is 8-bits.
1052 assert(PP.getTargetInfo().getCharWidth() == 8 &&
1053 "Assumes char is 8 bits");
1054 assert(PP.getTargetInfo().getIntWidth() <= 64 &&
1055 (PP.getTargetInfo().getIntWidth() & 7) == 0 &&
1056 "Assumes sizeof(int) on target is <= 64 and a multiple of char");
1057 assert(PP.getTargetInfo().getWCharWidth() <= 64 &&
1058 "Assumes sizeof(wchar) on target is <= 64");
1060 SmallVector<uint32_t, 4> codepoint_buffer;
1061 codepoint_buffer.resize(end - begin);
1062 uint32_t *buffer_begin = &codepoint_buffer.front();
1063 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size();
1065 // Unicode escapes representing characters that cannot be correctly
1066 // represented in a single code unit are disallowed in character literals
1067 // by this implementation.
1068 uint32_t largest_character_for_kind;
1069 if (tok::wide_char_constant == Kind) {
1070 largest_character_for_kind =
1071 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth());
1072 } else if (tok::utf8_char_constant == Kind) {
1073 largest_character_for_kind = 0x7F;
1074 } else if (tok::utf16_char_constant == Kind) {
1075 largest_character_for_kind = 0xFFFF;
1076 } else if (tok::utf32_char_constant == Kind) {
1077 largest_character_for_kind = 0x10FFFF;
1079 largest_character_for_kind = 0x7Fu;
1082 while (begin != end) {
1083 // Is this a span of non-escape characters?
1084 if (begin[0] != '\\') {
1085 char const *start = begin;
1088 } while (begin != end && *begin != '\\');
1090 char const *tmp_in_start = start;
1091 uint32_t *tmp_out_start = buffer_begin;
1092 ConversionResult res =
1093 ConvertUTF8toUTF32(reinterpret_cast<UTF8 const **>(&start),
1094 reinterpret_cast<UTF8 const *>(begin),
1095 &buffer_begin, buffer_end, strictConversion);
1096 if (res != conversionOK) {
1097 // If we see bad encoding for unprefixed character literals, warn and
1098 // simply copy the byte values, for compatibility with gcc and
1099 // older versions of clang.
1100 bool NoErrorOnBadEncoding = isAscii();
1101 unsigned Msg = diag::err_bad_character_encoding;
1102 if (NoErrorOnBadEncoding)
1103 Msg = diag::warn_bad_character_encoding;
1105 if (NoErrorOnBadEncoding) {
1106 start = tmp_in_start;
1107 buffer_begin = tmp_out_start;
1108 for (; start != begin; ++start, ++buffer_begin)
1109 *buffer_begin = static_cast<uint8_t>(*start);
1114 for (; tmp_out_start < buffer_begin; ++tmp_out_start) {
1115 if (*tmp_out_start > largest_character_for_kind) {
1117 PP.Diag(Loc, diag::err_character_too_large);
1124 // Is this a Universal Character Name escape?
1125 if (begin[1] == 'u' || begin[1] == 'U') {
1126 unsigned short UcnLen = 0;
1127 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen,
1128 FullSourceLoc(Loc, PP.getSourceManager()),
1129 &PP.getDiagnostics(), PP.getLangOpts(), true)) {
1131 } else if (*buffer_begin > largest_character_for_kind) {
1133 PP.Diag(Loc, diag::err_character_too_large);
1139 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo());
1141 ProcessCharEscape(TokBegin, begin, end, HadError,
1142 FullSourceLoc(Loc,PP.getSourceManager()),
1143 CharWidth, &PP.getDiagnostics(), PP.getLangOpts());
1144 *buffer_begin++ = result;
1147 unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front();
1149 if (NumCharsSoFar > 1) {
1151 PP.Diag(Loc, diag::warn_extraneous_char_constant);
1152 else if (isAscii() && NumCharsSoFar == 4)
1153 PP.Diag(Loc, diag::ext_four_char_character_literal);
1155 PP.Diag(Loc, diag::ext_multichar_character_literal);
1157 PP.Diag(Loc, diag::err_multichar_utf_character_literal);
1160 IsMultiChar = false;
1163 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0);
1165 // Narrow character literals act as though their value is concatenated
1166 // in this implementation, but warn on overflow.
1167 bool multi_char_too_long = false;
1168 if (isAscii() && isMultiChar()) {
1170 for (size_t i = 0; i < NumCharsSoFar; ++i) {
1171 // check for enough leading zeros to shift into
1172 multi_char_too_long |= (LitVal.countLeadingZeros() < 8);
1174 LitVal = LitVal + (codepoint_buffer[i] & 0xFF);
1176 } else if (NumCharsSoFar > 0) {
1177 // otherwise just take the last character
1178 LitVal = buffer_begin[-1];
1181 if (!HadError && multi_char_too_long) {
1182 PP.Diag(Loc, diag::warn_char_constant_too_large);
1185 // Transfer the value from APInt to uint64_t
1186 Value = LitVal.getZExtValue();
1188 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1")
1189 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple
1190 // character constants are not sign extended in the this implementation:
1191 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC.
1192 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) &&
1193 PP.getLangOpts().CharIsSigned)
1194 Value = (signed char)Value;
1198 /// string-literal: [C++0x lex.string]
1199 /// encoding-prefix " [s-char-sequence] "
1200 /// encoding-prefix R raw-string
1201 /// encoding-prefix:
1206 /// s-char-sequence:
1208 /// s-char-sequence s-char
1210 /// any member of the source character set except the double-quote ",
1211 /// backslash \, or new-line character
1213 /// universal-character-name
1215 /// " d-char-sequence ( r-char-sequence ) d-char-sequence "
1216 /// r-char-sequence:
1218 /// r-char-sequence r-char
1220 /// any member of the source character set, except a right parenthesis )
1221 /// followed by the initial d-char-sequence (which may be empty)
1222 /// followed by a double quote ".
1223 /// d-char-sequence:
1225 /// d-char-sequence d-char
1227 /// any member of the basic source character set except:
1228 /// space, the left parenthesis (, the right parenthesis ),
1229 /// the backslash \, and the control characters representing horizontal
1230 /// tab, vertical tab, form feed, and newline.
1231 /// escape-sequence: [C++0x lex.ccon]
1232 /// simple-escape-sequence
1233 /// octal-escape-sequence
1234 /// hexadecimal-escape-sequence
1235 /// simple-escape-sequence:
1236 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1237 /// octal-escape-sequence:
1239 /// \ octal-digit octal-digit
1240 /// \ octal-digit octal-digit octal-digit
1241 /// hexadecimal-escape-sequence:
1242 /// \x hexadecimal-digit
1243 /// hexadecimal-escape-sequence hexadecimal-digit
1244 /// universal-character-name:
1246 /// \U hex-quad hex-quad
1248 /// hex-digit hex-digit hex-digit hex-digit
1251 StringLiteralParser::
1252 StringLiteralParser(ArrayRef<Token> StringToks,
1253 Preprocessor &PP, bool Complain)
1254 : SM(PP.getSourceManager()), Features(PP.getLangOpts()),
1255 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() :nullptr),
1256 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown),
1257 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) {
1261 void StringLiteralParser::init(ArrayRef<Token> StringToks){
1262 // The literal token may have come from an invalid source location (e.g. due
1263 // to a PCH error), in which case the token length will be 0.
1264 if (StringToks.empty() || StringToks[0].getLength() < 2)
1265 return DiagnoseLexingError(SourceLocation());
1267 // Scan all of the string portions, remember the max individual token length,
1268 // computing a bound on the concatenated string length, and see whether any
1269 // piece is a wide-string. If any of the string portions is a wide-string
1270 // literal, the result is a wide-string literal [C99 6.4.5p4].
1271 assert(!StringToks.empty() && "expected at least one token");
1272 MaxTokenLength = StringToks[0].getLength();
1273 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!");
1274 SizeBound = StringToks[0].getLength()-2; // -2 for "".
1275 Kind = StringToks[0].getKind();
1279 // Implement Translation Phase #6: concatenation of string literals
1280 /// (C99 5.1.1.2p1). The common case is only one string fragment.
1281 for (unsigned i = 1; i != StringToks.size(); ++i) {
1282 if (StringToks[i].getLength() < 2)
1283 return DiagnoseLexingError(StringToks[i].getLocation());
1285 // The string could be shorter than this if it needs cleaning, but this is a
1286 // reasonable bound, which is all we need.
1287 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!");
1288 SizeBound += StringToks[i].getLength()-2; // -2 for "".
1290 // Remember maximum string piece length.
1291 if (StringToks[i].getLength() > MaxTokenLength)
1292 MaxTokenLength = StringToks[i].getLength();
1294 // Remember if we see any wide or utf-8/16/32 strings.
1295 // Also check for illegal concatenations.
1296 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) {
1298 Kind = StringToks[i].getKind();
1301 Diags->Report(StringToks[i].getLocation(),
1302 diag::err_unsupported_string_concat);
1308 // Include space for the null terminator.
1311 // TODO: K&R warning: "traditional C rejects string constant concatenation"
1313 // Get the width in bytes of char/wchar_t/char16_t/char32_t
1314 CharByteWidth = getCharWidth(Kind, Target);
1315 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple");
1318 // The output buffer size needs to be large enough to hold wide characters.
1319 // This is a worst-case assumption which basically corresponds to L"" "long".
1320 SizeBound *= CharByteWidth;
1322 // Size the temporary buffer to hold the result string data.
1323 ResultBuf.resize(SizeBound);
1325 // Likewise, but for each string piece.
1326 SmallString<512> TokenBuf;
1327 TokenBuf.resize(MaxTokenLength);
1329 // Loop over all the strings, getting their spelling, and expanding them to
1330 // wide strings as appropriate.
1331 ResultPtr = &ResultBuf[0]; // Next byte to fill in.
1335 SourceLocation UDSuffixTokLoc;
1337 for (unsigned i = 0, e = StringToks.size(); i != e; ++i) {
1338 const char *ThisTokBuf = &TokenBuf[0];
1339 // Get the spelling of the token, which eliminates trigraphs, etc. We know
1340 // that ThisTokBuf points to a buffer that is big enough for the whole token
1341 // and 'spelled' tokens can only shrink.
1342 bool StringInvalid = false;
1343 unsigned ThisTokLen =
1344 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features,
1347 return DiagnoseLexingError(StringToks[i].getLocation());
1349 const char *ThisTokBegin = ThisTokBuf;
1350 const char *ThisTokEnd = ThisTokBuf+ThisTokLen;
1352 // Remove an optional ud-suffix.
1353 if (ThisTokEnd[-1] != '"') {
1354 const char *UDSuffixEnd = ThisTokEnd;
1357 } while (ThisTokEnd[-1] != '"');
1359 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd);
1361 if (UDSuffixBuf.empty()) {
1362 if (StringToks[i].hasUCN())
1363 expandUCNs(UDSuffixBuf, UDSuffix);
1365 UDSuffixBuf.assign(UDSuffix);
1367 UDSuffixOffset = ThisTokEnd - ThisTokBuf;
1368 UDSuffixTokLoc = StringToks[i].getLocation();
1370 SmallString<32> ExpandedUDSuffix;
1371 if (StringToks[i].hasUCN()) {
1372 expandUCNs(ExpandedUDSuffix, UDSuffix);
1373 UDSuffix = ExpandedUDSuffix;
1376 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the
1377 // result of a concatenation involving at least one user-defined-string-
1378 // literal, all the participating user-defined-string-literals shall
1379 // have the same ud-suffix.
1380 if (UDSuffixBuf != UDSuffix) {
1382 SourceLocation TokLoc = StringToks[i].getLocation();
1383 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix)
1384 << UDSuffixBuf << UDSuffix
1385 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc)
1386 << SourceRange(TokLoc, TokLoc);
1393 // Strip the end quote.
1396 // TODO: Input character set mapping support.
1398 // Skip marker for wide or unicode strings.
1399 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') {
1401 // Skip 8 of u8 marker for utf8 strings.
1402 if (ThisTokBuf[0] == '8')
1406 // Check for raw string
1407 if (ThisTokBuf[0] == 'R') {
1408 ThisTokBuf += 2; // skip R"
1410 const char *Prefix = ThisTokBuf;
1411 while (ThisTokBuf[0] != '(')
1413 ++ThisTokBuf; // skip '('
1415 // Remove same number of characters from the end
1416 ThisTokEnd -= ThisTokBuf - Prefix;
1417 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal");
1419 // C++14 [lex.string]p4: A source-file new-line in a raw string literal
1420 // results in a new-line in the resulting execution string-literal.
1421 StringRef RemainingTokenSpan(ThisTokBuf, ThisTokEnd - ThisTokBuf);
1422 while (!RemainingTokenSpan.empty()) {
1423 // Split the string literal on \r\n boundaries.
1424 size_t CRLFPos = RemainingTokenSpan.find("\r\n");
1425 StringRef BeforeCRLF = RemainingTokenSpan.substr(0, CRLFPos);
1426 StringRef AfterCRLF = RemainingTokenSpan.substr(CRLFPos);
1428 // Copy everything before the \r\n sequence into the string literal.
1429 if (CopyStringFragment(StringToks[i], ThisTokBegin, BeforeCRLF))
1432 // Point into the \n inside the \r\n sequence and operate on the
1433 // remaining portion of the literal.
1434 RemainingTokenSpan = AfterCRLF.substr(1);
1437 if (ThisTokBuf[0] != '"') {
1438 // The file may have come from PCH and then changed after loading the
1439 // PCH; Fail gracefully.
1440 return DiagnoseLexingError(StringToks[i].getLocation());
1442 ++ThisTokBuf; // skip "
1444 // Check if this is a pascal string
1445 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd &&
1446 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') {
1448 // If the \p sequence is found in the first token, we have a pascal string
1449 // Otherwise, if we already have a pascal string, ignore the first \p
1457 while (ThisTokBuf != ThisTokEnd) {
1458 // Is this a span of non-escape characters?
1459 if (ThisTokBuf[0] != '\\') {
1460 const char *InStart = ThisTokBuf;
1463 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\');
1465 // Copy the character span over.
1466 if (CopyStringFragment(StringToks[i], ThisTokBegin,
1467 StringRef(InStart, ThisTokBuf - InStart)))
1471 // Is this a Universal Character Name escape?
1472 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') {
1473 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd,
1474 ResultPtr, hadError,
1475 FullSourceLoc(StringToks[i].getLocation(), SM),
1476 CharByteWidth, Diags, Features);
1479 // Otherwise, this is a non-UCN escape character. Process it.
1480 unsigned ResultChar =
1481 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError,
1482 FullSourceLoc(StringToks[i].getLocation(), SM),
1483 CharByteWidth*8, Diags, Features);
1485 if (CharByteWidth == 4) {
1486 // FIXME: Make the type of the result buffer correct instead of
1487 // using reinterpret_cast.
1488 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultPtr);
1489 *ResultWidePtr = ResultChar;
1491 } else if (CharByteWidth == 2) {
1492 // FIXME: Make the type of the result buffer correct instead of
1493 // using reinterpret_cast.
1494 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultPtr);
1495 *ResultWidePtr = ResultChar & 0xFFFF;
1498 assert(CharByteWidth == 1 && "Unexpected char width");
1499 *ResultPtr++ = ResultChar & 0xFF;
1506 if (CharByteWidth == 4) {
1507 // FIXME: Make the type of the result buffer correct instead of
1508 // using reinterpret_cast.
1509 UTF32 *ResultWidePtr = reinterpret_cast<UTF32*>(ResultBuf.data());
1510 ResultWidePtr[0] = GetNumStringChars() - 1;
1511 } else if (CharByteWidth == 2) {
1512 // FIXME: Make the type of the result buffer correct instead of
1513 // using reinterpret_cast.
1514 UTF16 *ResultWidePtr = reinterpret_cast<UTF16*>(ResultBuf.data());
1515 ResultWidePtr[0] = GetNumStringChars() - 1;
1517 assert(CharByteWidth == 1 && "Unexpected char width");
1518 ResultBuf[0] = GetNumStringChars() - 1;
1521 // Verify that pascal strings aren't too large.
1522 if (GetStringLength() > 256) {
1524 Diags->Report(StringToks.front().getLocation(),
1525 diag::err_pascal_string_too_long)
1526 << SourceRange(StringToks.front().getLocation(),
1527 StringToks.back().getLocation());
1532 // Complain if this string literal has too many characters.
1533 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509;
1535 if (GetNumStringChars() > MaxChars)
1536 Diags->Report(StringToks.front().getLocation(),
1537 diag::ext_string_too_long)
1538 << GetNumStringChars() << MaxChars
1539 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0)
1540 << SourceRange(StringToks.front().getLocation(),
1541 StringToks.back().getLocation());
1545 static const char *resyncUTF8(const char *Err, const char *End) {
1548 End = Err + std::min<unsigned>(getNumBytesForUTF8(*Err), End-Err);
1549 while (++Err != End && (*Err & 0xC0) == 0x80)
1554 /// \brief This function copies from Fragment, which is a sequence of bytes
1555 /// within Tok's contents (which begin at TokBegin) into ResultPtr.
1556 /// Performs widening for multi-byte characters.
1557 bool StringLiteralParser::CopyStringFragment(const Token &Tok,
1558 const char *TokBegin,
1559 StringRef Fragment) {
1560 const UTF8 *ErrorPtrTmp;
1561 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp))
1564 // If we see bad encoding for unprefixed string literals, warn and
1565 // simply copy the byte values, for compatibility with gcc and older
1566 // versions of clang.
1567 bool NoErrorOnBadEncoding = isAscii();
1568 if (NoErrorOnBadEncoding) {
1569 memcpy(ResultPtr, Fragment.data(), Fragment.size());
1570 ResultPtr += Fragment.size();
1574 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1576 FullSourceLoc SourceLoc(Tok.getLocation(), SM);
1577 const DiagnosticBuilder &Builder =
1578 Diag(Diags, Features, SourceLoc, TokBegin,
1579 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()),
1580 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding
1581 : diag::err_bad_string_encoding);
1583 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1584 StringRef NextFragment(NextStart, Fragment.end()-NextStart);
1586 // Decode into a dummy buffer.
1587 SmallString<512> Dummy;
1588 Dummy.reserve(Fragment.size() * CharByteWidth);
1589 char *Ptr = Dummy.data();
1591 while (!ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) {
1592 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1593 NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1594 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin,
1595 ErrorPtr, NextStart);
1596 NextFragment = StringRef(NextStart, Fragment.end()-NextStart);
1599 return !NoErrorOnBadEncoding;
1602 void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) {
1605 Diags->Report(Loc, diag::err_lexing_string);
1608 /// getOffsetOfStringByte - This function returns the offset of the
1609 /// specified byte of the string data represented by Token. This handles
1610 /// advancing over escape sequences in the string.
1611 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok,
1612 unsigned ByteNo) const {
1613 // Get the spelling of the token.
1614 SmallString<32> SpellingBuffer;
1615 SpellingBuffer.resize(Tok.getLength());
1617 bool StringInvalid = false;
1618 const char *SpellingPtr = &SpellingBuffer[0];
1619 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features,
1624 const char *SpellingStart = SpellingPtr;
1625 const char *SpellingEnd = SpellingPtr+TokLen;
1627 // Handle UTF-8 strings just like narrow strings.
1628 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8')
1631 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' &&
1632 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet");
1634 // For raw string literals, this is easy.
1635 if (SpellingPtr[0] == 'R') {
1636 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!");
1639 while (*SpellingPtr != '(') {
1641 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal");
1645 return SpellingPtr - SpellingStart + ByteNo;
1648 // Skip over the leading quote
1649 assert(SpellingPtr[0] == '"' && "Should be a string literal!");
1652 // Skip over bytes until we find the offset we're looking for.
1654 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!");
1656 // Step over non-escapes simply.
1657 if (*SpellingPtr != '\\') {
1663 // Otherwise, this is an escape character. Advance over it.
1664 bool HadError = false;
1665 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') {
1666 const char *EscapePtr = SpellingPtr;
1667 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd,
1668 1, Features, HadError);
1670 // ByteNo is somewhere within the escape sequence.
1671 SpellingPtr = EscapePtr;
1676 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError,
1677 FullSourceLoc(Tok.getLocation(), SM),
1678 CharByteWidth*8, Diags, Features);
1681 assert(!HadError && "This method isn't valid on erroneous strings");
1684 return SpellingPtr-SpellingStart;