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/Lex/Preprocessor.h"
17 #include "clang/Lex/LexDiagnostic.h"
18 #include "clang/Basic/TargetInfo.h"
19 #include "llvm/ADT/StringExtras.h"
20 #include "llvm/Support/ErrorHandling.h"
21 using namespace clang;
23 /// HexDigitValue - Return the value of the specified hex digit, or -1 if it's
25 static int HexDigitValue(char C) {
26 if (C >= '0' && C <= '9') return C-'0';
27 if (C >= 'a' && C <= 'f') return C-'a'+10;
28 if (C >= 'A' && C <= 'F') return C-'A'+10;
32 static unsigned getCharWidth(tok::TokenKind kind, const TargetInfo &Target) {
34 default: llvm_unreachable("Unknown token type!");
35 case tok::char_constant:
36 case tok::string_literal:
37 case tok::utf8_string_literal:
38 return Target.getCharWidth();
39 case tok::wide_char_constant:
40 case tok::wide_string_literal:
41 return Target.getWCharWidth();
42 case tok::utf16_char_constant:
43 case tok::utf16_string_literal:
44 return Target.getChar16Width();
45 case tok::utf32_char_constant:
46 case tok::utf32_string_literal:
47 return Target.getChar32Width();
51 /// ProcessCharEscape - Parse a standard C escape sequence, which can occur in
52 /// either a character or a string literal.
53 static unsigned ProcessCharEscape(const char *&ThisTokBuf,
54 const char *ThisTokEnd, bool &HadError,
55 FullSourceLoc Loc, unsigned CharWidth,
56 DiagnosticsEngine *Diags) {
60 // We know that this character can't be off the end of the buffer, because
61 // that would have been \", which would not have been the end of string.
62 unsigned ResultChar = *ThisTokBuf++;
64 // These map to themselves.
65 case '\\': case '\'': case '"': case '?': break;
67 // These have fixed mappings.
69 // TODO: K&R: the meaning of '\\a' is different in traditional C
77 Diags->Report(Loc, diag::ext_nonstandard_escape) << "e";
82 Diags->Report(Loc, diag::ext_nonstandard_escape) << "E";
100 case 'x': { // Hex escape.
102 if (ThisTokBuf == ThisTokEnd || !isxdigit(*ThisTokBuf)) {
104 Diags->Report(Loc, diag::err_hex_escape_no_digits);
109 // Hex escapes are a maximal series of hex digits.
110 bool Overflow = false;
111 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) {
112 int CharVal = HexDigitValue(ThisTokBuf[0]);
113 if (CharVal == -1) break;
114 // About to shift out a digit?
115 Overflow |= (ResultChar & 0xF0000000) ? true : false;
117 ResultChar |= CharVal;
120 // See if any bits will be truncated when evaluated as a character.
121 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
123 ResultChar &= ~0U >> (32-CharWidth);
126 // Check for overflow.
127 if (Overflow && Diags) // Too many digits to fit in
128 Diags->Report(Loc, diag::warn_hex_escape_too_large);
131 case '0': case '1': case '2': case '3':
132 case '4': case '5': case '6': case '7': {
137 // Octal escapes are a series of octal digits with maximum length 3.
138 // "\0123" is a two digit sequence equal to "\012" "3".
139 unsigned NumDigits = 0;
142 ResultChar |= *ThisTokBuf++ - '0';
144 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 &&
145 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7');
147 // Check for overflow. Reject '\777', but not L'\777'.
148 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
150 Diags->Report(Loc, diag::warn_octal_escape_too_large);
151 ResultChar &= ~0U >> (32-CharWidth);
156 // Otherwise, these are not valid escapes.
157 case '(': case '{': case '[': case '%':
158 // GCC accepts these as extensions. We warn about them as such though.
160 Diags->Report(Loc, diag::ext_nonstandard_escape)
161 << std::string()+(char)ResultChar;
167 if (isgraph(ResultChar))
168 Diags->Report(Loc, diag::ext_unknown_escape)
169 << std::string()+(char)ResultChar;
171 Diags->Report(Loc, diag::ext_unknown_escape)
172 << "x"+llvm::utohexstr(ResultChar);
179 /// ProcessUCNEscape - Read the Universal Character Name, check constraints and
180 /// return the UTF32.
181 static bool ProcessUCNEscape(const char *&ThisTokBuf, const char *ThisTokEnd,
182 uint32_t &UcnVal, unsigned short &UcnLen,
183 FullSourceLoc Loc, DiagnosticsEngine *Diags,
184 const LangOptions &Features) {
185 if (!Features.CPlusPlus && !Features.C99 && Diags)
186 Diags->Report(Loc, diag::warn_ucn_not_valid_in_c89);
188 // Save the beginning of the string (for error diagnostics).
189 const char *ThisTokBegin = ThisTokBuf;
191 // Skip the '\u' char's.
194 if (ThisTokBuf == ThisTokEnd || !isxdigit(*ThisTokBuf)) {
196 Diags->Report(Loc, diag::err_ucn_escape_no_digits);
199 UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8);
200 unsigned short UcnLenSave = UcnLen;
201 for (; ThisTokBuf != ThisTokEnd && UcnLenSave; ++ThisTokBuf, UcnLenSave--) {
202 int CharVal = HexDigitValue(ThisTokBuf[0]);
203 if (CharVal == -1) break;
207 // If we didn't consume the proper number of digits, there is a problem.
211 Lexer::AdvanceToTokenCharacter(Loc, ThisTokBuf-ThisTokBegin,
212 Loc.getManager(), Features);
213 Diags->Report(FullSourceLoc(L, Loc.getManager()),
214 diag::err_ucn_escape_incomplete);
218 // Check UCN constraints (C99 6.4.3p2).
219 if ((UcnVal < 0xa0 &&
220 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60 )) // $, @, `
221 || (UcnVal >= 0xD800 && UcnVal <= 0xDFFF)
222 || (UcnVal > 0x10FFFF)) /* the maximum legal UTF32 value */ {
224 Diags->Report(Loc, diag::err_ucn_escape_invalid);
230 /// EncodeUCNEscape - Read the Universal Character Name, check constraints and
231 /// convert the UTF32 to UTF8 or UTF16. This is a subroutine of
232 /// StringLiteralParser. When we decide to implement UCN's for identifiers,
233 /// we will likely rework our support for UCN's.
234 static void EncodeUCNEscape(const char *&ThisTokBuf, const char *ThisTokEnd,
235 char *&ResultBuf, bool &HadError,
236 FullSourceLoc Loc, unsigned CharByteWidth,
237 DiagnosticsEngine *Diags,
238 const LangOptions &Features) {
239 typedef uint32_t UTF32;
241 unsigned short UcnLen = 0;
242 if (!ProcessUCNEscape(ThisTokBuf, ThisTokEnd, UcnVal, UcnLen, Loc, Diags,
248 assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth) &&
249 "only character widths of 1, 2, or 4 bytes supported");
252 assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported");
254 if (CharByteWidth == 4) {
255 // Note: our internal rep of wide char tokens is always little-endian.
256 *ResultBuf++ = (UcnVal & 0x000000FF);
257 *ResultBuf++ = (UcnVal & 0x0000FF00) >> 8;
258 *ResultBuf++ = (UcnVal & 0x00FF0000) >> 16;
259 *ResultBuf++ = (UcnVal & 0xFF000000) >> 24;
263 if (CharByteWidth == 2) {
265 if (UcnVal < (UTF32)0xFFFF) {
266 *ResultBuf++ = (UcnVal & 0x000000FF);
267 *ResultBuf++ = (UcnVal & 0x0000FF00) >> 8;
270 if (Diags) Diags->Report(Loc, diag::warn_ucn_escape_too_large);
272 typedef uint16_t UTF16;
274 UTF16 surrogate1 = 0xD800 + (UcnVal >> 10);
275 UTF16 surrogate2 = 0xDC00 + (UcnVal & 0x3FF);
276 *ResultBuf++ = (surrogate1 & 0x000000FF);
277 *ResultBuf++ = (surrogate1 & 0x0000FF00) >> 8;
278 *ResultBuf++ = (surrogate2 & 0x000000FF);
279 *ResultBuf++ = (surrogate2 & 0x0000FF00) >> 8;
283 assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters");
285 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8.
286 // The conversion below was inspired by:
287 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c
288 // First, we determine how many bytes the result will require.
289 typedef uint8_t UTF8;
291 unsigned short bytesToWrite = 0;
292 if (UcnVal < (UTF32)0x80)
294 else if (UcnVal < (UTF32)0x800)
296 else if (UcnVal < (UTF32)0x10000)
301 const unsigned byteMask = 0xBF;
302 const unsigned byteMark = 0x80;
304 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed
305 // into the first byte, depending on how many bytes follow.
306 static const UTF8 firstByteMark[5] = {
307 0x00, 0x00, 0xC0, 0xE0, 0xF0
309 // Finally, we write the bytes into ResultBuf.
310 ResultBuf += bytesToWrite;
311 switch (bytesToWrite) { // note: everything falls through.
312 case 4: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
313 case 3: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
314 case 2: *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
315 case 1: *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]);
317 // Update the buffer.
318 ResultBuf += bytesToWrite;
322 /// integer-constant: [C99 6.4.4.1]
323 /// decimal-constant integer-suffix
324 /// octal-constant integer-suffix
325 /// hexadecimal-constant integer-suffix
326 /// decimal-constant:
328 /// decimal-constant digit
331 /// octal-constant octal-digit
332 /// hexadecimal-constant:
333 /// hexadecimal-prefix hexadecimal-digit
334 /// hexadecimal-constant hexadecimal-digit
335 /// hexadecimal-prefix: one of
338 /// unsigned-suffix [long-suffix]
339 /// unsigned-suffix [long-long-suffix]
340 /// long-suffix [unsigned-suffix]
341 /// long-long-suffix [unsigned-sufix]
343 /// 1 2 3 4 5 6 7 8 9
346 /// hexadecimal-digit:
347 /// 0 1 2 3 4 5 6 7 8 9
350 /// unsigned-suffix: one of
352 /// long-suffix: one of
354 /// long-long-suffix: one of
357 /// floating-constant: [C99 6.4.4.2]
358 /// TODO: add rules...
360 NumericLiteralParser::
361 NumericLiteralParser(const char *begin, const char *end,
362 SourceLocation TokLoc, Preprocessor &pp)
363 : PP(pp), ThisTokBegin(begin), ThisTokEnd(end) {
365 // This routine assumes that the range begin/end matches the regex for integer
366 // and FP constants (specifically, the 'pp-number' regex), and assumes that
367 // the byte at "*end" is both valid and not part of the regex. Because of
368 // this, it doesn't have to check for 'overscan' in various places.
369 assert(!isalnum(*end) && *end != '.' && *end != '_' &&
370 "Lexer didn't maximally munch?");
372 s = DigitsBegin = begin;
373 saw_exponent = false;
380 isMicrosoftInteger = false;
383 if (*s == '0') { // parse radix
384 ParseNumberStartingWithZero(TokLoc);
387 } else { // the first digit is non-zero
390 if (s == ThisTokEnd) {
392 } else if (isxdigit(*s) && !(*s == 'e' || *s == 'E')) {
393 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin),
394 diag::err_invalid_decimal_digit) << StringRef(s, 1);
397 } else if (*s == '.') {
402 if ((*s == 'e' || *s == 'E')) { // exponent
403 const char *Exponent = s;
406 if (*s == '+' || *s == '-') s++; // sign
407 const char *first_non_digit = SkipDigits(s);
408 if (first_non_digit != s) {
411 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-begin),
412 diag::err_exponent_has_no_digits);
421 // Parse the suffix. At this point we can classify whether we have an FP or
423 bool isFPConstant = isFloatingLiteral();
425 // Loop over all of the characters of the suffix. If we see something bad,
426 // we break out of the loop.
427 for (; s != ThisTokEnd; ++s) {
429 case 'f': // FP Suffix for "float"
431 if (!isFPConstant) break; // Error for integer constant.
432 if (isFloat || isLong) break; // FF, LF invalid.
434 continue; // Success.
437 if (isFPConstant) break; // Error for floating constant.
438 if (isUnsigned) break; // Cannot be repeated.
440 continue; // Success.
443 if (isLong || isLongLong) break; // Cannot be repeated.
444 if (isFloat) break; // LF invalid.
446 // Check for long long. The L's need to be adjacent and the same case.
447 if (s+1 != ThisTokEnd && s[1] == s[0]) {
448 if (isFPConstant) break; // long long invalid for floats.
450 ++s; // Eat both of them.
454 continue; // Success.
457 if (PP.getLangOptions().MicrosoftExt) {
458 if (isFPConstant || isLong || isLongLong) break;
460 // Allow i8, i16, i32, i64, and i128.
461 if (s + 1 != ThisTokEnd) {
465 isMicrosoftInteger = true;
468 if (s + 2 == ThisTokEnd) break;
470 s += 3; // i16 suffix
471 isMicrosoftInteger = true;
473 else if (s[2] == '2') {
474 if (s + 3 == ThisTokEnd) break;
476 s += 4; // i128 suffix
477 isMicrosoftInteger = true;
482 if (s + 2 == ThisTokEnd) break;
484 s += 3; // i32 suffix
486 isMicrosoftInteger = true;
490 if (s + 2 == ThisTokEnd) break;
492 s += 3; // i64 suffix
494 isMicrosoftInteger = true;
506 if (isImaginary) break; // Cannot be repeated.
507 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin),
508 diag::ext_imaginary_constant);
510 continue; // Success.
512 // If we reached here, there was an error.
516 // Report an error if there are any.
517 if (s != ThisTokEnd) {
518 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-begin),
519 isFPConstant ? diag::err_invalid_suffix_float_constant :
520 diag::err_invalid_suffix_integer_constant)
521 << StringRef(SuffixBegin, ThisTokEnd-SuffixBegin);
527 /// ParseNumberStartingWithZero - This method is called when the first character
528 /// of the number is found to be a zero. This means it is either an octal
529 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or
530 /// a floating point number (01239.123e4). Eat the prefix, determining the
532 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
533 assert(s[0] == '0' && "Invalid method call");
536 // Handle a hex number like 0x1234.
537 if ((*s == 'x' || *s == 'X') && (isxdigit(s[1]) || s[1] == '.')) {
541 s = SkipHexDigits(s);
542 if (s == ThisTokEnd) {
544 } else if (*s == '.') {
547 s = SkipHexDigits(s);
549 // A binary exponent can appear with or with a '.'. If dotted, the
550 // binary exponent is required.
551 if (*s == 'p' || *s == 'P') {
552 const char *Exponent = s;
555 if (*s == '+' || *s == '-') s++; // sign
556 const char *first_non_digit = SkipDigits(s);
557 if (first_non_digit == s) {
558 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
559 diag::err_exponent_has_no_digits);
565 if (!PP.getLangOptions().HexFloats)
566 PP.Diag(TokLoc, diag::ext_hexconstant_invalid);
567 } else if (saw_period) {
568 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
569 diag::err_hexconstant_requires_exponent);
575 // Handle simple binary numbers 0b01010
576 if (*s == 'b' || *s == 'B') {
577 // 0b101010 is a GCC extension.
578 PP.Diag(TokLoc, diag::ext_binary_literal);
582 s = SkipBinaryDigits(s);
583 if (s == ThisTokEnd) {
585 } else if (isxdigit(*s)) {
586 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
587 diag::err_invalid_binary_digit) << StringRef(s, 1);
590 // Other suffixes will be diagnosed by the caller.
594 // For now, the radix is set to 8. If we discover that we have a
595 // floating point constant, the radix will change to 10. Octal floating
596 // point constants are not permitted (only decimal and hexadecimal).
599 s = SkipOctalDigits(s);
601 return; // Done, simple octal number like 01234
603 // If we have some other non-octal digit that *is* a decimal digit, see if
604 // this is part of a floating point number like 094.123 or 09e1.
606 const char *EndDecimal = SkipDigits(s);
607 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') {
613 // If we have a hex digit other than 'e' (which denotes a FP exponent) then
614 // the code is using an incorrect base.
615 if (isxdigit(*s) && *s != 'e' && *s != 'E') {
616 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
617 diag::err_invalid_octal_digit) << StringRef(s, 1);
626 s = SkipDigits(s); // Skip suffix.
628 if (*s == 'e' || *s == 'E') { // exponent
629 const char *Exponent = s;
633 if (*s == '+' || *s == '-') s++; // sign
634 const char *first_non_digit = SkipDigits(s);
635 if (first_non_digit != s) {
638 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
639 diag::err_exponent_has_no_digits);
647 /// GetIntegerValue - Convert this numeric literal value to an APInt that
648 /// matches Val's input width. If there is an overflow, set Val to the low bits
649 /// of the result and return true. Otherwise, return false.
650 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) {
651 // Fast path: Compute a conservative bound on the maximum number of
652 // bits per digit in this radix. If we can't possibly overflow a
653 // uint64 based on that bound then do the simple conversion to
654 // integer. This avoids the expensive overflow checking below, and
655 // handles the common cases that matter (small decimal integers and
656 // hex/octal values which don't overflow).
657 unsigned MaxBitsPerDigit = 1;
658 while ((1U << MaxBitsPerDigit) < radix)
659 MaxBitsPerDigit += 1;
660 if ((SuffixBegin - DigitsBegin) * MaxBitsPerDigit <= 64) {
662 for (s = DigitsBegin; s != SuffixBegin; ++s)
663 N = N*radix + HexDigitValue(*s);
665 // This will truncate the value to Val's input width. Simply check
666 // for overflow by comparing.
668 return Val.getZExtValue() != N;
674 llvm::APInt RadixVal(Val.getBitWidth(), radix);
675 llvm::APInt CharVal(Val.getBitWidth(), 0);
676 llvm::APInt OldVal = Val;
678 bool OverflowOccurred = false;
679 while (s < SuffixBegin) {
680 unsigned C = HexDigitValue(*s++);
682 // If this letter is out of bound for this radix, reject it.
683 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
687 // Add the digit to the value in the appropriate radix. If adding in digits
688 // made the value smaller, then this overflowed.
691 // Multiply by radix, did overflow occur on the multiply?
693 OverflowOccurred |= Val.udiv(RadixVal) != OldVal;
695 // Add value, did overflow occur on the value?
696 // (a + b) ult b <=> overflow
698 OverflowOccurred |= Val.ult(CharVal);
700 return OverflowOccurred;
703 llvm::APFloat::opStatus
704 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) {
707 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin);
708 return Result.convertFromString(StringRef(ThisTokBegin, n),
709 APFloat::rmNearestTiesToEven);
713 /// character-literal: [C++0x lex.ccon]
714 /// ' c-char-sequence '
715 /// u' c-char-sequence '
716 /// U' c-char-sequence '
717 /// L' c-char-sequence '
720 /// c-char-sequence c-char
722 /// any member of the source character set except the single-quote ',
723 /// backslash \, or new-line character
725 /// universal-character-name
726 /// escape-sequence: [C++0x lex.ccon]
727 /// simple-escape-sequence
728 /// octal-escape-sequence
729 /// hexadecimal-escape-sequence
730 /// simple-escape-sequence:
731 /// one of \' \" \? \\ \a \b \f \n \r \t \v
732 /// octal-escape-sequence:
734 /// \ octal-digit octal-digit
735 /// \ octal-digit octal-digit octal-digit
736 /// hexadecimal-escape-sequence:
737 /// \x hexadecimal-digit
738 /// hexadecimal-escape-sequence hexadecimal-digit
739 /// universal-character-name:
741 /// \U hex-quad hex-quad
743 /// hex-digit hex-digit hex-digit hex-digit
745 CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
746 SourceLocation Loc, Preprocessor &PP,
747 tok::TokenKind kind) {
748 // At this point we know that the character matches the regex "L?'.*'".
753 // Determine if this is a wide or UTF character.
754 if (Kind == tok::wide_char_constant || Kind == tok::utf16_char_constant ||
755 Kind == tok::utf32_char_constant) {
759 // Skip over the entry quote.
760 assert(begin[0] == '\'' && "Invalid token lexed");
763 // FIXME: The "Value" is an uint64_t so we can handle char literals of
765 // FIXME: This extensively assumes that 'char' is 8-bits.
766 assert(PP.getTargetInfo().getCharWidth() == 8 &&
767 "Assumes char is 8 bits");
768 assert(PP.getTargetInfo().getIntWidth() <= 64 &&
769 (PP.getTargetInfo().getIntWidth() & 7) == 0 &&
770 "Assumes sizeof(int) on target is <= 64 and a multiple of char");
771 assert(PP.getTargetInfo().getWCharWidth() <= 64 &&
772 "Assumes sizeof(wchar) on target is <= 64");
774 // This is what we will use for overflow detection
775 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0);
777 unsigned NumCharsSoFar = 0;
779 while (begin[0] != '\'') {
782 // Is this a Universal Character Name escape?
783 if (begin[0] != '\\') // If this is a normal character, consume it.
784 ResultChar = (unsigned char)*begin++;
785 else { // Otherwise, this is an escape character.
786 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo());
788 if (begin[1] == 'u' || begin[1] == 'U') {
790 unsigned short UcnLen = 0;
791 if (!ProcessUCNEscape(begin, end, utf32, UcnLen,
792 FullSourceLoc(Loc, PP.getSourceManager()),
793 &PP.getDiagnostics(), PP.getLangOptions())) {
797 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
798 PP.Diag(Loc, diag::warn_ucn_escape_too_large);
799 ResultChar &= ~0U >> (32-CharWidth);
802 // Otherwise, this is a non-UCN escape character. Process it.
803 ResultChar = ProcessCharEscape(begin, end, HadError,
804 FullSourceLoc(Loc,PP.getSourceManager()),
805 CharWidth, &PP.getDiagnostics());
809 // If this is a multi-character constant (e.g. 'abc'), handle it. These are
810 // implementation defined (C99 6.4.4.4p10).
813 // Emulate GCC's (unintentional?) behavior: L'ab' -> L'b'.
816 // Narrow character literals act as though their value is concatenated
817 // in this implementation, but warn on overflow.
818 if (LitVal.countLeadingZeros() < 8 && !Warned) {
819 PP.Diag(Loc, diag::warn_char_constant_too_large);
826 LitVal = LitVal + ResultChar;
830 // If this is the second character being processed, do special handling.
831 if (NumCharsSoFar > 1) {
832 // Warn about discarding the top bits for multi-char wide-character
833 // constants (L'abcd').
835 PP.Diag(Loc, diag::warn_extraneous_char_constant);
836 else if (NumCharsSoFar != 4)
837 PP.Diag(Loc, diag::ext_multichar_character_literal);
839 PP.Diag(Loc, diag::ext_four_char_character_literal);
844 // Transfer the value from APInt to uint64_t
845 Value = LitVal.getZExtValue();
847 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1")
848 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple
849 // character constants are not sign extended in the this implementation:
850 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC.
851 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) &&
852 PP.getLangOptions().CharIsSigned)
853 Value = (signed char)Value;
857 /// string-literal: [C++0x lex.string]
858 /// encoding-prefix " [s-char-sequence] "
859 /// encoding-prefix R raw-string
867 /// s-char-sequence s-char
869 /// any member of the source character set except the double-quote ",
870 /// backslash \, or new-line character
872 /// universal-character-name
874 /// " d-char-sequence ( r-char-sequence ) d-char-sequence "
877 /// r-char-sequence r-char
879 /// any member of the source character set, except a right parenthesis )
880 /// followed by the initial d-char-sequence (which may be empty)
881 /// followed by a double quote ".
884 /// d-char-sequence d-char
886 /// any member of the basic source character set except:
887 /// space, the left parenthesis (, the right parenthesis ),
888 /// the backslash \, and the control characters representing horizontal
889 /// tab, vertical tab, form feed, and newline.
890 /// escape-sequence: [C++0x lex.ccon]
891 /// simple-escape-sequence
892 /// octal-escape-sequence
893 /// hexadecimal-escape-sequence
894 /// simple-escape-sequence:
895 /// one of \' \" \? \\ \a \b \f \n \r \t \v
896 /// octal-escape-sequence:
898 /// \ octal-digit octal-digit
899 /// \ octal-digit octal-digit octal-digit
900 /// hexadecimal-escape-sequence:
901 /// \x hexadecimal-digit
902 /// hexadecimal-escape-sequence hexadecimal-digit
903 /// universal-character-name:
905 /// \U hex-quad hex-quad
907 /// hex-digit hex-digit hex-digit hex-digit
909 StringLiteralParser::
910 StringLiteralParser(const Token *StringToks, unsigned NumStringToks,
911 Preprocessor &PP, bool Complain)
912 : SM(PP.getSourceManager()), Features(PP.getLangOptions()),
913 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() : 0),
914 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown),
915 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) {
916 init(StringToks, NumStringToks);
919 void StringLiteralParser::init(const Token *StringToks, unsigned NumStringToks){
920 // The literal token may have come from an invalid source location (e.g. due
921 // to a PCH error), in which case the token length will be 0.
922 if (NumStringToks == 0 || StringToks[0].getLength() < 2) {
927 // Scan all of the string portions, remember the max individual token length,
928 // computing a bound on the concatenated string length, and see whether any
929 // piece is a wide-string. If any of the string portions is a wide-string
930 // literal, the result is a wide-string literal [C99 6.4.5p4].
931 assert(NumStringToks && "expected at least one token");
932 MaxTokenLength = StringToks[0].getLength();
933 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!");
934 SizeBound = StringToks[0].getLength()-2; // -2 for "".
935 Kind = StringToks[0].getKind();
939 // Implement Translation Phase #6: concatenation of string literals
940 /// (C99 5.1.1.2p1). The common case is only one string fragment.
941 for (unsigned i = 1; i != NumStringToks; ++i) {
942 if (StringToks[i].getLength() < 2) {
947 // The string could be shorter than this if it needs cleaning, but this is a
948 // reasonable bound, which is all we need.
949 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!");
950 SizeBound += StringToks[i].getLength()-2; // -2 for "".
952 // Remember maximum string piece length.
953 if (StringToks[i].getLength() > MaxTokenLength)
954 MaxTokenLength = StringToks[i].getLength();
956 // Remember if we see any wide or utf-8/16/32 strings.
957 // Also check for illegal concatenations.
958 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) {
960 Kind = StringToks[i].getKind();
963 Diags->Report(FullSourceLoc(StringToks[i].getLocation(), SM),
964 diag::err_unsupported_string_concat);
970 // Include space for the null terminator.
973 // TODO: K&R warning: "traditional C rejects string constant concatenation"
975 // Get the width in bytes of char/wchar_t/char16_t/char32_t
976 CharByteWidth = getCharWidth(Kind, Target);
977 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple");
980 // The output buffer size needs to be large enough to hold wide characters.
981 // This is a worst-case assumption which basically corresponds to L"" "long".
982 SizeBound *= CharByteWidth;
984 // Size the temporary buffer to hold the result string data.
985 ResultBuf.resize(SizeBound);
987 // Likewise, but for each string piece.
988 llvm::SmallString<512> TokenBuf;
989 TokenBuf.resize(MaxTokenLength);
991 // Loop over all the strings, getting their spelling, and expanding them to
992 // wide strings as appropriate.
993 ResultPtr = &ResultBuf[0]; // Next byte to fill in.
997 for (unsigned i = 0, e = NumStringToks; i != e; ++i) {
998 const char *ThisTokBuf = &TokenBuf[0];
999 // Get the spelling of the token, which eliminates trigraphs, etc. We know
1000 // that ThisTokBuf points to a buffer that is big enough for the whole token
1001 // and 'spelled' tokens can only shrink.
1002 bool StringInvalid = false;
1003 unsigned ThisTokLen =
1004 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features,
1006 if (StringInvalid) {
1011 const char *ThisTokEnd = ThisTokBuf+ThisTokLen-1; // Skip end quote.
1012 // TODO: Input character set mapping support.
1014 // Skip marker for wide or unicode strings.
1015 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') {
1017 // Skip 8 of u8 marker for utf8 strings.
1018 if (ThisTokBuf[0] == '8')
1022 // Check for raw string
1023 if (ThisTokBuf[0] == 'R') {
1024 ThisTokBuf += 2; // skip R"
1026 const char *Prefix = ThisTokBuf;
1027 while (ThisTokBuf[0] != '(')
1029 ++ThisTokBuf; // skip '('
1031 // remove same number of characters from the end
1032 if (ThisTokEnd >= ThisTokBuf + (ThisTokBuf - Prefix))
1033 ThisTokEnd -= (ThisTokBuf - Prefix);
1035 // Copy the string over
1036 CopyStringFragment(StringRef(ThisTokBuf, ThisTokEnd - ThisTokBuf));
1038 assert(ThisTokBuf[0] == '"' && "Expected quote, lexer broken?");
1039 ++ThisTokBuf; // skip "
1041 // Check if this is a pascal string
1042 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd &&
1043 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') {
1045 // If the \p sequence is found in the first token, we have a pascal string
1046 // Otherwise, if we already have a pascal string, ignore the first \p
1054 while (ThisTokBuf != ThisTokEnd) {
1055 // Is this a span of non-escape characters?
1056 if (ThisTokBuf[0] != '\\') {
1057 const char *InStart = ThisTokBuf;
1060 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\');
1062 // Copy the character span over.
1063 CopyStringFragment(StringRef(InStart, ThisTokBuf - InStart));
1066 // Is this a Universal Character Name escape?
1067 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') {
1068 EncodeUCNEscape(ThisTokBuf, ThisTokEnd, ResultPtr,
1069 hadError, FullSourceLoc(StringToks[i].getLocation(),SM),
1070 CharByteWidth, Diags, Features);
1073 // Otherwise, this is a non-UCN escape character. Process it.
1074 unsigned ResultChar =
1075 ProcessCharEscape(ThisTokBuf, ThisTokEnd, hadError,
1076 FullSourceLoc(StringToks[i].getLocation(), SM),
1077 CharByteWidth*8, Diags);
1079 // Note: our internal rep of wide char tokens is always little-endian.
1080 *ResultPtr++ = ResultChar & 0xFF;
1082 for (unsigned i = 1, e = CharByteWidth; i != e; ++i)
1083 *ResultPtr++ = ResultChar >> i*8;
1089 ResultBuf[0] = ResultPtr-&ResultBuf[0]-1;
1090 ResultBuf[0] /= CharByteWidth;
1092 // Verify that pascal strings aren't too large.
1093 if (GetStringLength() > 256) {
1095 Diags->Report(FullSourceLoc(StringToks[0].getLocation(), SM),
1096 diag::err_pascal_string_too_long)
1097 << SourceRange(StringToks[0].getLocation(),
1098 StringToks[NumStringToks-1].getLocation());
1103 // Complain if this string literal has too many characters.
1104 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509;
1106 if (GetNumStringChars() > MaxChars)
1107 Diags->Report(FullSourceLoc(StringToks[0].getLocation(), SM),
1108 diag::ext_string_too_long)
1109 << GetNumStringChars() << MaxChars
1110 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0)
1111 << SourceRange(StringToks[0].getLocation(),
1112 StringToks[NumStringToks-1].getLocation());
1117 /// copyStringFragment - This function copies from Start to End into ResultPtr.
1118 /// Performs widening for multi-byte characters.
1119 void StringLiteralParser::CopyStringFragment(StringRef Fragment) {
1120 // Copy the character span over.
1121 if (CharByteWidth == 1) {
1122 memcpy(ResultPtr, Fragment.data(), Fragment.size());
1123 ResultPtr += Fragment.size();
1125 // Note: our internal rep of wide char tokens is always little-endian.
1126 for (StringRef::iterator I=Fragment.begin(), E=Fragment.end(); I!=E; ++I) {
1128 // Add zeros at the end.
1129 for (unsigned i = 1, e = CharByteWidth; i != e; ++i)
1136 /// getOffsetOfStringByte - This function returns the offset of the
1137 /// specified byte of the string data represented by Token. This handles
1138 /// advancing over escape sequences in the string.
1139 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok,
1140 unsigned ByteNo) const {
1141 // Get the spelling of the token.
1142 llvm::SmallString<32> SpellingBuffer;
1143 SpellingBuffer.resize(Tok.getLength());
1145 bool StringInvalid = false;
1146 const char *SpellingPtr = &SpellingBuffer[0];
1147 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features,
1152 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' &&
1153 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet");
1156 const char *SpellingStart = SpellingPtr;
1157 const char *SpellingEnd = SpellingPtr+TokLen;
1159 // Skip over the leading quote.
1160 assert(SpellingPtr[0] == '"' && "Should be a string literal!");
1163 // Skip over bytes until we find the offset we're looking for.
1165 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!");
1167 // Step over non-escapes simply.
1168 if (*SpellingPtr != '\\') {
1174 // Otherwise, this is an escape character. Advance over it.
1175 bool HadError = false;
1176 ProcessCharEscape(SpellingPtr, SpellingEnd, HadError,
1177 FullSourceLoc(Tok.getLocation(), SM),
1178 CharByteWidth*8, Diags);
1179 assert(!HadError && "This method isn't valid on erroneous strings");
1183 return SpellingPtr-SpellingStart;