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/LangOptions.h"
18 #include "clang/Basic/SourceLocation.h"
19 #include "clang/Basic/TargetInfo.h"
20 #include "clang/Lex/LexDiagnostic.h"
21 #include "clang/Lex/Lexer.h"
22 #include "clang/Lex/Preprocessor.h"
23 #include "clang/Lex/Token.h"
24 #include "llvm/ADT/APInt.h"
25 #include "llvm/ADT/SmallVector.h"
26 #include "llvm/ADT/StringExtras.h"
27 #include "llvm/ADT/StringSwitch.h"
28 #include "llvm/Support/ConvertUTF.h"
29 #include "llvm/Support/ErrorHandling.h"
37 using namespace clang;
39 static unsigned getCharWidth(tok::TokenKind kind, const TargetInfo &Target) {
41 default: llvm_unreachable("Unknown token type!");
42 case tok::char_constant:
43 case tok::string_literal:
44 case tok::utf8_char_constant:
45 case tok::utf8_string_literal:
46 return Target.getCharWidth();
47 case tok::wide_char_constant:
48 case tok::wide_string_literal:
49 return Target.getWCharWidth();
50 case tok::utf16_char_constant:
51 case tok::utf16_string_literal:
52 return Target.getChar16Width();
53 case tok::utf32_char_constant:
54 case tok::utf32_string_literal:
55 return Target.getChar32Width();
59 static CharSourceRange MakeCharSourceRange(const LangOptions &Features,
62 const char *TokRangeBegin,
63 const char *TokRangeEnd) {
64 SourceLocation Begin =
65 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin,
66 TokLoc.getManager(), Features);
68 Lexer::AdvanceToTokenCharacter(Begin, TokRangeEnd - TokRangeBegin,
69 TokLoc.getManager(), Features);
70 return CharSourceRange::getCharRange(Begin, End);
73 /// Produce a diagnostic highlighting some portion of a literal.
75 /// Emits the diagnostic \p DiagID, highlighting the range of characters from
76 /// \p TokRangeBegin (inclusive) to \p TokRangeEnd (exclusive), which must be
77 /// a substring of a spelling buffer for the token beginning at \p TokBegin.
78 static DiagnosticBuilder Diag(DiagnosticsEngine *Diags,
79 const LangOptions &Features, FullSourceLoc TokLoc,
80 const char *TokBegin, const char *TokRangeBegin,
81 const char *TokRangeEnd, unsigned DiagID) {
82 SourceLocation Begin =
83 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin,
84 TokLoc.getManager(), Features);
85 return Diags->Report(Begin, DiagID) <<
86 MakeCharSourceRange(Features, TokLoc, TokBegin, TokRangeBegin, TokRangeEnd);
89 /// ProcessCharEscape - Parse a standard C escape sequence, which can occur in
90 /// either a character or a string literal.
91 static unsigned ProcessCharEscape(const char *ThisTokBegin,
92 const char *&ThisTokBuf,
93 const char *ThisTokEnd, bool &HadError,
94 FullSourceLoc Loc, unsigned CharWidth,
95 DiagnosticsEngine *Diags,
96 const LangOptions &Features) {
97 const char *EscapeBegin = ThisTokBuf;
102 // We know that this character can't be off the end of the buffer, because
103 // that would have been \", which would not have been the end of string.
104 unsigned ResultChar = *ThisTokBuf++;
105 switch (ResultChar) {
106 // These map to themselves.
107 case '\\': case '\'': case '"': case '?': break;
109 // These have fixed mappings.
111 // TODO: K&R: the meaning of '\\a' is different in traditional C
119 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
120 diag::ext_nonstandard_escape) << "e";
125 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
126 diag::ext_nonstandard_escape) << "E";
144 case 'x': { // Hex escape.
146 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
148 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
149 diag::err_hex_escape_no_digits) << "x";
154 // Hex escapes are a maximal series of hex digits.
155 bool Overflow = false;
156 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) {
157 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]);
158 if (CharVal == -1) break;
159 // About to shift out a digit?
160 if (ResultChar & 0xF0000000)
163 ResultChar |= CharVal;
166 // See if any bits will be truncated when evaluated as a character.
167 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
169 ResultChar &= ~0U >> (32-CharWidth);
172 // Check for overflow.
173 if (Overflow && Diags) // Too many digits to fit in
174 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
175 diag::err_escape_too_large) << 0;
178 case '0': case '1': case '2': case '3':
179 case '4': case '5': case '6': case '7': {
184 // Octal escapes are a series of octal digits with maximum length 3.
185 // "\0123" is a two digit sequence equal to "\012" "3".
186 unsigned NumDigits = 0;
189 ResultChar |= *ThisTokBuf++ - '0';
191 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 &&
192 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7');
194 // Check for overflow. Reject '\777', but not L'\777'.
195 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
197 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
198 diag::err_escape_too_large) << 1;
199 ResultChar &= ~0U >> (32-CharWidth);
204 // Otherwise, these are not valid escapes.
205 case '(': case '{': case '[': case '%':
206 // GCC accepts these as extensions. We warn about them as such though.
208 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
209 diag::ext_nonstandard_escape)
210 << std::string(1, ResultChar);
216 if (isPrintable(ResultChar))
217 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
218 diag::ext_unknown_escape)
219 << std::string(1, ResultChar);
221 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
222 diag::ext_unknown_escape)
223 << "x" + llvm::utohexstr(ResultChar);
230 static void appendCodePoint(unsigned Codepoint,
231 llvm::SmallVectorImpl<char> &Str) {
233 char *ResultPtr = ResultBuf;
234 bool Res = llvm::ConvertCodePointToUTF8(Codepoint, ResultPtr);
236 assert(Res && "Unexpected conversion failure");
237 Str.append(ResultBuf, ResultPtr);
240 void clang::expandUCNs(SmallVectorImpl<char> &Buf, StringRef Input) {
241 for (StringRef::iterator I = Input.begin(), E = Input.end(); I != E; ++I) {
248 assert(*I == 'u' || *I == 'U');
250 unsigned NumHexDigits;
256 assert(I + NumHexDigits <= E);
258 uint32_t CodePoint = 0;
259 for (++I; NumHexDigits != 0; ++I, --NumHexDigits) {
260 unsigned Value = llvm::hexDigitValue(*I);
261 assert(Value != -1U);
267 appendCodePoint(CodePoint, Buf);
272 /// ProcessUCNEscape - Read the Universal Character Name, check constraints and
273 /// return the UTF32.
274 static bool ProcessUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf,
275 const char *ThisTokEnd,
276 uint32_t &UcnVal, unsigned short &UcnLen,
277 FullSourceLoc Loc, DiagnosticsEngine *Diags,
278 const LangOptions &Features,
279 bool in_char_string_literal = false) {
280 const char *UcnBegin = ThisTokBuf;
282 // Skip the '\u' char's.
285 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
287 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
288 diag::err_hex_escape_no_digits) << StringRef(&ThisTokBuf[-1], 1);
291 UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8);
292 unsigned short UcnLenSave = UcnLen;
293 for (; ThisTokBuf != ThisTokEnd && UcnLenSave; ++ThisTokBuf, UcnLenSave--) {
294 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]);
295 if (CharVal == -1) break;
299 // If we didn't consume the proper number of digits, there is a problem.
302 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
303 diag::err_ucn_escape_incomplete);
307 // Check UCN constraints (C99 6.4.3p2) [C++11 lex.charset p2]
308 if ((0xD800 <= UcnVal && UcnVal <= 0xDFFF) || // surrogate codepoints
309 UcnVal > 0x10FFFF) { // maximum legal UTF32 value
311 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
312 diag::err_ucn_escape_invalid);
316 // C++11 allows UCNs that refer to control characters and basic source
317 // characters inside character and string literals
319 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60)) { // $, @, `
320 bool IsError = (!Features.CPlusPlus11 || !in_char_string_literal);
322 char BasicSCSChar = UcnVal;
323 if (UcnVal >= 0x20 && UcnVal < 0x7f)
324 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
325 IsError ? diag::err_ucn_escape_basic_scs :
326 diag::warn_cxx98_compat_literal_ucn_escape_basic_scs)
327 << StringRef(&BasicSCSChar, 1);
329 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
330 IsError ? diag::err_ucn_control_character :
331 diag::warn_cxx98_compat_literal_ucn_control_character);
337 if (!Features.CPlusPlus && !Features.C99 && Diags)
338 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
339 diag::warn_ucn_not_valid_in_c89_literal);
344 /// MeasureUCNEscape - Determine the number of bytes within the resulting string
345 /// which this UCN will occupy.
346 static int MeasureUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf,
347 const char *ThisTokEnd, unsigned CharByteWidth,
348 const LangOptions &Features, bool &HadError) {
349 // UTF-32: 4 bytes per escape.
350 if (CharByteWidth == 4)
354 unsigned short UcnLen = 0;
357 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal,
358 UcnLen, Loc, nullptr, Features, true)) {
363 // UTF-16: 2 bytes for BMP, 4 bytes otherwise.
364 if (CharByteWidth == 2)
365 return UcnVal <= 0xFFFF ? 2 : 4;
372 if (UcnVal < 0x10000)
377 /// EncodeUCNEscape - Read the Universal Character Name, check constraints and
378 /// convert the UTF32 to UTF8 or UTF16. This is a subroutine of
379 /// StringLiteralParser. When we decide to implement UCN's for identifiers,
380 /// we will likely rework our support for UCN's.
381 static void EncodeUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf,
382 const char *ThisTokEnd,
383 char *&ResultBuf, bool &HadError,
384 FullSourceLoc Loc, unsigned CharByteWidth,
385 DiagnosticsEngine *Diags,
386 const LangOptions &Features) {
387 typedef uint32_t UTF32;
389 unsigned short UcnLen = 0;
390 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, UcnLen,
391 Loc, Diags, Features, true)) {
396 assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth == 4) &&
397 "only character widths of 1, 2, or 4 bytes supported");
400 assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported");
402 if (CharByteWidth == 4) {
403 // FIXME: Make the type of the result buffer correct instead of
404 // using reinterpret_cast.
405 llvm::UTF32 *ResultPtr = reinterpret_cast<llvm::UTF32*>(ResultBuf);
411 if (CharByteWidth == 2) {
412 // FIXME: Make the type of the result buffer correct instead of
413 // using reinterpret_cast.
414 llvm::UTF16 *ResultPtr = reinterpret_cast<llvm::UTF16*>(ResultBuf);
416 if (UcnVal <= (UTF32)0xFFFF) {
424 *ResultPtr = 0xD800 + (UcnVal >> 10);
425 *(ResultPtr+1) = 0xDC00 + (UcnVal & 0x3FF);
430 assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters");
432 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8.
433 // The conversion below was inspired by:
434 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c
435 // First, we determine how many bytes the result will require.
436 typedef uint8_t UTF8;
438 unsigned short bytesToWrite = 0;
439 if (UcnVal < (UTF32)0x80)
441 else if (UcnVal < (UTF32)0x800)
443 else if (UcnVal < (UTF32)0x10000)
448 const unsigned byteMask = 0xBF;
449 const unsigned byteMark = 0x80;
451 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed
452 // into the first byte, depending on how many bytes follow.
453 static const UTF8 firstByteMark[5] = {
454 0x00, 0x00, 0xC0, 0xE0, 0xF0
456 // Finally, we write the bytes into ResultBuf.
457 ResultBuf += bytesToWrite;
458 switch (bytesToWrite) { // note: everything falls through.
460 *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
463 *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
466 *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
469 *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]);
471 // Update the buffer.
472 ResultBuf += bytesToWrite;
475 /// integer-constant: [C99 6.4.4.1]
476 /// decimal-constant integer-suffix
477 /// octal-constant integer-suffix
478 /// hexadecimal-constant integer-suffix
479 /// binary-literal integer-suffix [GNU, C++1y]
480 /// user-defined-integer-literal: [C++11 lex.ext]
481 /// decimal-literal ud-suffix
482 /// octal-literal ud-suffix
483 /// hexadecimal-literal ud-suffix
484 /// binary-literal ud-suffix [GNU, C++1y]
485 /// decimal-constant:
487 /// decimal-constant digit
490 /// octal-constant octal-digit
491 /// hexadecimal-constant:
492 /// hexadecimal-prefix hexadecimal-digit
493 /// hexadecimal-constant hexadecimal-digit
494 /// hexadecimal-prefix: one of
499 /// binary-literal binary-digit
501 /// unsigned-suffix [long-suffix]
502 /// unsigned-suffix [long-long-suffix]
503 /// long-suffix [unsigned-suffix]
504 /// long-long-suffix [unsigned-sufix]
506 /// 1 2 3 4 5 6 7 8 9
509 /// hexadecimal-digit:
510 /// 0 1 2 3 4 5 6 7 8 9
516 /// unsigned-suffix: one of
518 /// long-suffix: one of
520 /// long-long-suffix: one of
523 /// floating-constant: [C99 6.4.4.2]
524 /// TODO: add rules...
526 NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
527 SourceLocation TokLoc,
529 : PP(PP), ThisTokBegin(TokSpelling.begin()), ThisTokEnd(TokSpelling.end()) {
531 // This routine assumes that the range begin/end matches the regex for integer
532 // and FP constants (specifically, the 'pp-number' regex), and assumes that
533 // the byte at "*end" is both valid and not part of the regex. Because of
534 // this, it doesn't have to check for 'overscan' in various places.
535 assert(!isPreprocessingNumberBody(*ThisTokEnd) && "didn't maximally munch?");
537 s = DigitsBegin = ThisTokBegin;
538 saw_exponent = false;
540 saw_ud_suffix = false;
541 saw_fixed_point_suffix = false;
550 MicrosoftInteger = 0;
555 if (*s == '0') { // parse radix
556 ParseNumberStartingWithZero(TokLoc);
559 } else { // the first digit is non-zero
562 if (s == ThisTokEnd) {
565 ParseDecimalOrOctalCommon(TokLoc);
572 checkSeparator(TokLoc, s, CSK_AfterDigits);
574 // Initial scan to lookahead for fixed point suffix.
575 if (PP.getLangOpts().FixedPoint) {
576 for (const char *c = s; c != ThisTokEnd; ++c) {
577 if (*c == 'r' || *c == 'k' || *c == 'R' || *c == 'K') {
578 saw_fixed_point_suffix = true;
584 // Parse the suffix. At this point we can classify whether we have an FP or
586 bool isFPConstant = isFloatingLiteral();
588 // Loop over all of the characters of the suffix. If we see something bad,
589 // we break out of the loop.
590 for (; s != ThisTokEnd; ++s) {
594 if (!PP.getLangOpts().FixedPoint) break;
595 if (isFract || isAccum) break;
596 if (!(saw_period || saw_exponent)) break;
601 if (!PP.getLangOpts().FixedPoint) break;
602 if (isFract || isAccum) break;
603 if (!(saw_period || saw_exponent)) break;
606 case 'h': // FP Suffix for "half".
608 // OpenCL Extension v1.2 s9.5 - h or H suffix for half type.
609 if (!(PP.getLangOpts().Half || PP.getLangOpts().FixedPoint)) break;
610 if (isIntegerLiteral()) break; // Error for integer constant.
611 if (isHalf || isFloat || isLong) break; // HH, FH, LH invalid.
613 continue; // Success.
614 case 'f': // FP Suffix for "float"
616 if (!isFPConstant) break; // Error for integer constant.
617 if (isHalf || isFloat || isLong || isFloat128)
618 break; // HF, FF, LF, QF invalid.
620 if (PP.getTargetInfo().hasFloat16Type() && s + 2 < ThisTokEnd &&
621 s[1] == '1' && s[2] == '6') {
622 s += 2; // success, eat up 2 characters.
628 continue; // Success.
629 case 'q': // FP Suffix for "__float128"
631 if (!isFPConstant) break; // Error for integer constant.
632 if (isHalf || isFloat || isLong || isFloat128)
633 break; // HQ, FQ, LQ, QQ invalid.
635 continue; // Success.
638 if (isFPConstant) break; // Error for floating constant.
639 if (isUnsigned) break; // Cannot be repeated.
641 continue; // Success.
644 if (isLong || isLongLong) break; // Cannot be repeated.
645 if (isHalf || isFloat || isFloat128) break; // LH, LF, LQ invalid.
647 // Check for long long. The L's need to be adjacent and the same case.
649 assert(s + 1 < ThisTokEnd && "didn't maximally munch?");
650 if (isFPConstant) break; // long long invalid for floats.
652 ++s; // Eat both of them.
656 continue; // Success.
659 if (PP.getLangOpts().MicrosoftExt) {
660 if (isLong || isLongLong || MicrosoftInteger)
664 // Allow i8, i16, i32, and i64.
668 MicrosoftInteger = 8;
672 s += 3; // i16 suffix
673 MicrosoftInteger = 16;
678 s += 3; // i32 suffix
679 MicrosoftInteger = 32;
684 s += 3; // i64 suffix
685 MicrosoftInteger = 64;
692 if (MicrosoftInteger) {
693 assert(s <= ThisTokEnd && "didn't maximally munch?");
700 if (isImaginary) break; // Cannot be repeated.
702 continue; // Success.
704 // If we reached here, there was an error or a ud-suffix.
708 // "i", "if", and "il" are user-defined suffixes in C++1y.
709 if (s != ThisTokEnd || isImaginary) {
710 // FIXME: Don't bother expanding UCNs if !tok.hasUCN().
711 expandUCNs(UDSuffixBuf, StringRef(SuffixBegin, ThisTokEnd - SuffixBegin));
712 if (isValidUDSuffix(PP.getLangOpts(), UDSuffixBuf)) {
714 // Any suffix pieces we might have parsed are actually part of the
723 MicrosoftInteger = 0;
724 saw_fixed_point_suffix = false;
729 saw_ud_suffix = true;
733 if (s != ThisTokEnd) {
734 // Report an error if there are any.
735 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin),
736 diag::err_invalid_suffix_constant)
737 << StringRef(SuffixBegin, ThisTokEnd - SuffixBegin) << isFPConstant;
742 if (!hadError && saw_fixed_point_suffix) {
743 assert(isFract || isAccum);
747 /// ParseDecimalOrOctalCommon - This method is called for decimal or octal
748 /// numbers. It issues an error for illegal digits, and handles floating point
749 /// parsing. If it detects a floating point number, the radix is set to 10.
750 void NumericLiteralParser::ParseDecimalOrOctalCommon(SourceLocation TokLoc){
751 assert((radix == 8 || radix == 10) && "Unexpected radix");
753 // If we have a hex digit other than 'e' (which denotes a FP exponent) then
754 // the code is using an incorrect base.
755 if (isHexDigit(*s) && *s != 'e' && *s != 'E' &&
756 !isValidUDSuffix(PP.getLangOpts(), StringRef(s, ThisTokEnd - s))) {
757 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
758 diag::err_invalid_digit) << StringRef(s, 1) << (radix == 8 ? 1 : 0);
764 checkSeparator(TokLoc, s, CSK_AfterDigits);
768 checkSeparator(TokLoc, s, CSK_BeforeDigits);
769 s = SkipDigits(s); // Skip suffix.
771 if (*s == 'e' || *s == 'E') { // exponent
772 checkSeparator(TokLoc, s, CSK_AfterDigits);
773 const char *Exponent = s;
777 if (s != ThisTokEnd && (*s == '+' || *s == '-')) s++; // sign
778 const char *first_non_digit = SkipDigits(s);
779 if (containsDigits(s, first_non_digit)) {
780 checkSeparator(TokLoc, s, CSK_BeforeDigits);
784 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
785 diag::err_exponent_has_no_digits);
793 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
794 /// suffixes as ud-suffixes, because the diagnostic experience is better if we
795 /// treat it as an invalid suffix.
796 bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
798 if (!LangOpts.CPlusPlus11 || Suffix.empty())
801 // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid.
802 if (Suffix[0] == '_')
805 // In C++11, there are no library suffixes.
806 if (!LangOpts.CPlusPlus14)
809 // In C++14, "s", "h", "min", "ms", "us", and "ns" are used in the library.
810 // Per tweaked N3660, "il", "i", and "if" are also used in the library.
811 // In C++2a "d" and "y" are used in the library.
812 return llvm::StringSwitch<bool>(Suffix)
813 .Cases("h", "min", "s", true)
814 .Cases("ms", "us", "ns", true)
815 .Cases("il", "i", "if", true)
816 .Cases("d", "y", LangOpts.CPlusPlus2a)
820 void NumericLiteralParser::checkSeparator(SourceLocation TokLoc,
822 CheckSeparatorKind IsAfterDigits) {
823 if (IsAfterDigits == CSK_AfterDigits) {
824 if (Pos == ThisTokBegin)
827 } else if (Pos == ThisTokEnd)
830 if (isDigitSeparator(*Pos)) {
831 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin),
832 diag::err_digit_separator_not_between_digits)
838 /// ParseNumberStartingWithZero - This method is called when the first character
839 /// of the number is found to be a zero. This means it is either an octal
840 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or
841 /// a floating point number (01239.123e4). Eat the prefix, determining the
843 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
844 assert(s[0] == '0' && "Invalid method call");
849 // Handle a hex number like 0x1234.
850 if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) {
852 assert(s < ThisTokEnd && "didn't maximally munch?");
855 s = SkipHexDigits(s);
856 bool HasSignificandDigits = containsDigits(DigitsBegin, s);
857 if (s == ThisTokEnd) {
859 } else if (*s == '.') {
862 const char *floatDigitsBegin = s;
863 s = SkipHexDigits(s);
864 if (containsDigits(floatDigitsBegin, s))
865 HasSignificandDigits = true;
866 if (HasSignificandDigits)
867 checkSeparator(TokLoc, floatDigitsBegin, CSK_BeforeDigits);
870 if (!HasSignificandDigits) {
871 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin),
872 diag::err_hex_constant_requires)
873 << PP.getLangOpts().CPlusPlus << 1;
878 // A binary exponent can appear with or with a '.'. If dotted, the
879 // binary exponent is required.
880 if (*s == 'p' || *s == 'P') {
881 checkSeparator(TokLoc, s, CSK_AfterDigits);
882 const char *Exponent = s;
885 if (s != ThisTokEnd && (*s == '+' || *s == '-')) s++; // sign
886 const char *first_non_digit = SkipDigits(s);
887 if (!containsDigits(s, first_non_digit)) {
889 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
890 diag::err_exponent_has_no_digits);
895 checkSeparator(TokLoc, s, CSK_BeforeDigits);
898 if (!PP.getLangOpts().HexFloats)
899 PP.Diag(TokLoc, PP.getLangOpts().CPlusPlus
900 ? diag::ext_hex_literal_invalid
901 : diag::ext_hex_constant_invalid);
902 else if (PP.getLangOpts().CPlusPlus17)
903 PP.Diag(TokLoc, diag::warn_cxx17_hex_literal);
904 } else if (saw_period) {
905 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin),
906 diag::err_hex_constant_requires)
907 << PP.getLangOpts().CPlusPlus << 0;
913 // Handle simple binary numbers 0b01010
914 if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) {
915 // 0b101010 is a C++1y / GCC extension.
917 PP.getLangOpts().CPlusPlus14
918 ? diag::warn_cxx11_compat_binary_literal
919 : PP.getLangOpts().CPlusPlus
920 ? diag::ext_binary_literal_cxx14
921 : diag::ext_binary_literal);
923 assert(s < ThisTokEnd && "didn't maximally munch?");
926 s = SkipBinaryDigits(s);
927 if (s == ThisTokEnd) {
929 } else if (isHexDigit(*s) &&
930 !isValidUDSuffix(PP.getLangOpts(),
931 StringRef(s, ThisTokEnd - s))) {
932 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
933 diag::err_invalid_digit) << StringRef(s, 1) << 2;
936 // Other suffixes will be diagnosed by the caller.
940 // For now, the radix is set to 8. If we discover that we have a
941 // floating point constant, the radix will change to 10. Octal floating
942 // point constants are not permitted (only decimal and hexadecimal).
945 s = SkipOctalDigits(s);
947 return; // Done, simple octal number like 01234
949 // If we have some other non-octal digit that *is* a decimal digit, see if
950 // this is part of a floating point number like 094.123 or 09e1.
952 const char *EndDecimal = SkipDigits(s);
953 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') {
959 ParseDecimalOrOctalCommon(TokLoc);
962 static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) {
965 return NumDigits <= 64;
967 return NumDigits <= 64 / 3; // Digits are groups of 3 bits.
969 return NumDigits <= 19; // floor(log10(2^64))
971 return NumDigits <= 64 / 4; // Digits are groups of 4 bits.
973 llvm_unreachable("impossible Radix");
977 /// GetIntegerValue - Convert this numeric literal value to an APInt that
978 /// matches Val's input width. If there is an overflow, set Val to the low bits
979 /// of the result and return true. Otherwise, return false.
980 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) {
981 // Fast path: Compute a conservative bound on the maximum number of
982 // bits per digit in this radix. If we can't possibly overflow a
983 // uint64 based on that bound then do the simple conversion to
984 // integer. This avoids the expensive overflow checking below, and
985 // handles the common cases that matter (small decimal integers and
986 // hex/octal values which don't overflow).
987 const unsigned NumDigits = SuffixBegin - DigitsBegin;
988 if (alwaysFitsInto64Bits(radix, NumDigits)) {
990 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr)
991 if (!isDigitSeparator(*Ptr))
992 N = N * radix + llvm::hexDigitValue(*Ptr);
994 // This will truncate the value to Val's input width. Simply check
995 // for overflow by comparing.
997 return Val.getZExtValue() != N;
1001 const char *Ptr = DigitsBegin;
1003 llvm::APInt RadixVal(Val.getBitWidth(), radix);
1004 llvm::APInt CharVal(Val.getBitWidth(), 0);
1005 llvm::APInt OldVal = Val;
1007 bool OverflowOccurred = false;
1008 while (Ptr < SuffixBegin) {
1009 if (isDigitSeparator(*Ptr)) {
1014 unsigned C = llvm::hexDigitValue(*Ptr++);
1016 // If this letter is out of bound for this radix, reject it.
1017 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1021 // Add the digit to the value in the appropriate radix. If adding in digits
1022 // made the value smaller, then this overflowed.
1025 // Multiply by radix, did overflow occur on the multiply?
1027 OverflowOccurred |= Val.udiv(RadixVal) != OldVal;
1029 // Add value, did overflow occur on the value?
1030 // (a + b) ult b <=> overflow
1032 OverflowOccurred |= Val.ult(CharVal);
1034 return OverflowOccurred;
1037 llvm::APFloat::opStatus
1038 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) {
1039 using llvm::APFloat;
1041 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin);
1043 llvm::SmallString<16> Buffer;
1044 StringRef Str(ThisTokBegin, n);
1045 if (Str.find('\'') != StringRef::npos) {
1047 std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer),
1052 return Result.convertFromString(Str, APFloat::rmNearestTiesToEven);
1055 static inline bool IsExponentPart(char c) {
1056 return c == 'p' || c == 'P' || c == 'e' || c == 'E';
1059 bool NumericLiteralParser::GetFixedPointValue(llvm::APInt &StoreVal, unsigned Scale) {
1060 assert(radix == 16 || radix == 10);
1062 // Find how many digits are needed to store the whole literal.
1063 unsigned NumDigits = SuffixBegin - DigitsBegin;
1064 if (saw_period) --NumDigits;
1066 // Initial scan of the exponent if it exists
1067 bool ExpOverflowOccurred = false;
1068 bool NegativeExponent = false;
1069 const char *ExponentBegin;
1070 uint64_t Exponent = 0;
1071 int64_t BaseShift = 0;
1073 const char *Ptr = DigitsBegin;
1075 while (!IsExponentPart(*Ptr)) ++Ptr;
1076 ExponentBegin = Ptr;
1078 NegativeExponent = *Ptr == '-';
1079 if (NegativeExponent) ++Ptr;
1081 unsigned NumExpDigits = SuffixBegin - Ptr;
1082 if (alwaysFitsInto64Bits(radix, NumExpDigits)) {
1083 llvm::StringRef ExpStr(Ptr, NumExpDigits);
1084 llvm::APInt ExpInt(/*numBits=*/64, ExpStr, /*radix=*/10);
1085 Exponent = ExpInt.getZExtValue();
1087 ExpOverflowOccurred = true;
1090 if (NegativeExponent) BaseShift -= Exponent;
1091 else BaseShift += Exponent;
1094 // Number of bits needed for decimal literal is
1095 // ceil(NumDigits * log2(10)) Integral part
1096 // + Scale Fractional part
1097 // + ceil(Exponent * log2(10)) Exponent
1098 // --------------------------------------------------
1099 // ceil((NumDigits + Exponent) * log2(10)) + Scale
1101 // But for simplicity in handling integers, we can round up log2(10) to 4,
1103 // 4 * (NumDigits + Exponent) + Scale
1105 // Number of digits needed for hexadecimal literal is
1106 // 4 * NumDigits Integral part
1107 // + Scale Fractional part
1108 // + Exponent Exponent
1109 // --------------------------------------------------
1110 // (4 * NumDigits) + Scale + Exponent
1111 uint64_t NumBitsNeeded;
1113 NumBitsNeeded = 4 * (NumDigits + Exponent) + Scale;
1115 NumBitsNeeded = 4 * NumDigits + Exponent + Scale;
1117 if (NumBitsNeeded > std::numeric_limits<unsigned>::max())
1118 ExpOverflowOccurred = true;
1119 llvm::APInt Val(static_cast<unsigned>(NumBitsNeeded), 0, /*isSigned=*/false);
1121 bool FoundDecimal = false;
1123 int64_t FractBaseShift = 0;
1124 const char *End = saw_exponent ? ExponentBegin : SuffixBegin;
1125 for (const char *Ptr = DigitsBegin; Ptr < End; ++Ptr) {
1127 FoundDecimal = true;
1131 // Normal reading of an integer
1132 unsigned C = llvm::hexDigitValue(*Ptr);
1133 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1139 // Keep track of how much we will need to adjust this value by from the
1140 // number of digits past the radix point.
1144 // For a radix of 16, we will be multiplying by 2 instead of 16.
1145 if (radix == 16) FractBaseShift *= 4;
1146 BaseShift += FractBaseShift;
1150 uint64_t Base = (radix == 16) ? 2 : 10;
1151 if (BaseShift > 0) {
1152 for (int64_t i = 0; i < BaseShift; ++i) {
1155 } else if (BaseShift < 0) {
1156 for (int64_t i = BaseShift; i < 0 && !Val.isNullValue(); ++i)
1157 Val = Val.udiv(Base);
1160 bool IntOverflowOccurred = false;
1161 auto MaxVal = llvm::APInt::getMaxValue(StoreVal.getBitWidth());
1162 if (Val.getBitWidth() > StoreVal.getBitWidth()) {
1163 IntOverflowOccurred |= Val.ugt(MaxVal.zext(Val.getBitWidth()));
1164 StoreVal = Val.trunc(StoreVal.getBitWidth());
1165 } else if (Val.getBitWidth() < StoreVal.getBitWidth()) {
1166 IntOverflowOccurred |= Val.zext(MaxVal.getBitWidth()).ugt(MaxVal);
1167 StoreVal = Val.zext(StoreVal.getBitWidth());
1172 return IntOverflowOccurred || ExpOverflowOccurred;
1176 /// user-defined-character-literal: [C++11 lex.ext]
1177 /// character-literal ud-suffix
1180 /// character-literal: [C++11 lex.ccon]
1181 /// ' c-char-sequence '
1182 /// u' c-char-sequence '
1183 /// U' c-char-sequence '
1184 /// L' c-char-sequence '
1185 /// u8' c-char-sequence ' [C++1z lex.ccon]
1186 /// c-char-sequence:
1188 /// c-char-sequence c-char
1190 /// any member of the source character set except the single-quote ',
1191 /// backslash \, or new-line character
1193 /// universal-character-name
1194 /// escape-sequence:
1195 /// simple-escape-sequence
1196 /// octal-escape-sequence
1197 /// hexadecimal-escape-sequence
1198 /// simple-escape-sequence:
1199 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1200 /// octal-escape-sequence:
1202 /// \ octal-digit octal-digit
1203 /// \ octal-digit octal-digit octal-digit
1204 /// hexadecimal-escape-sequence:
1205 /// \x hexadecimal-digit
1206 /// hexadecimal-escape-sequence hexadecimal-digit
1207 /// universal-character-name: [C++11 lex.charset]
1209 /// \U hex-quad hex-quad
1211 /// hex-digit hex-digit hex-digit hex-digit
1214 CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
1215 SourceLocation Loc, Preprocessor &PP,
1216 tok::TokenKind kind) {
1217 // At this point we know that the character matches the regex "(L|u|U)?'.*'".
1222 const char *TokBegin = begin;
1224 // Skip over wide character determinant.
1225 if (Kind != tok::char_constant)
1227 if (Kind == tok::utf8_char_constant)
1230 // Skip over the entry quote.
1231 assert(begin[0] == '\'' && "Invalid token lexed");
1234 // Remove an optional ud-suffix.
1235 if (end[-1] != '\'') {
1236 const char *UDSuffixEnd = end;
1239 } while (end[-1] != '\'');
1240 // FIXME: Don't bother with this if !tok.hasUCN().
1241 expandUCNs(UDSuffixBuf, StringRef(end, UDSuffixEnd - end));
1242 UDSuffixOffset = end - TokBegin;
1245 // Trim the ending quote.
1246 assert(end != begin && "Invalid token lexed");
1249 // FIXME: The "Value" is an uint64_t so we can handle char literals of
1251 // FIXME: This extensively assumes that 'char' is 8-bits.
1252 assert(PP.getTargetInfo().getCharWidth() == 8 &&
1253 "Assumes char is 8 bits");
1254 assert(PP.getTargetInfo().getIntWidth() <= 64 &&
1255 (PP.getTargetInfo().getIntWidth() & 7) == 0 &&
1256 "Assumes sizeof(int) on target is <= 64 and a multiple of char");
1257 assert(PP.getTargetInfo().getWCharWidth() <= 64 &&
1258 "Assumes sizeof(wchar) on target is <= 64");
1260 SmallVector<uint32_t, 4> codepoint_buffer;
1261 codepoint_buffer.resize(end - begin);
1262 uint32_t *buffer_begin = &codepoint_buffer.front();
1263 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size();
1265 // Unicode escapes representing characters that cannot be correctly
1266 // represented in a single code unit are disallowed in character literals
1267 // by this implementation.
1268 uint32_t largest_character_for_kind;
1269 if (tok::wide_char_constant == Kind) {
1270 largest_character_for_kind =
1271 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth());
1272 } else if (tok::utf8_char_constant == Kind) {
1273 largest_character_for_kind = 0x7F;
1274 } else if (tok::utf16_char_constant == Kind) {
1275 largest_character_for_kind = 0xFFFF;
1276 } else if (tok::utf32_char_constant == Kind) {
1277 largest_character_for_kind = 0x10FFFF;
1279 largest_character_for_kind = 0x7Fu;
1282 while (begin != end) {
1283 // Is this a span of non-escape characters?
1284 if (begin[0] != '\\') {
1285 char const *start = begin;
1288 } while (begin != end && *begin != '\\');
1290 char const *tmp_in_start = start;
1291 uint32_t *tmp_out_start = buffer_begin;
1292 llvm::ConversionResult res =
1293 llvm::ConvertUTF8toUTF32(reinterpret_cast<llvm::UTF8 const **>(&start),
1294 reinterpret_cast<llvm::UTF8 const *>(begin),
1295 &buffer_begin, buffer_end, llvm::strictConversion);
1296 if (res != llvm::conversionOK) {
1297 // If we see bad encoding for unprefixed character literals, warn and
1298 // simply copy the byte values, for compatibility with gcc and
1299 // older versions of clang.
1300 bool NoErrorOnBadEncoding = isAscii();
1301 unsigned Msg = diag::err_bad_character_encoding;
1302 if (NoErrorOnBadEncoding)
1303 Msg = diag::warn_bad_character_encoding;
1305 if (NoErrorOnBadEncoding) {
1306 start = tmp_in_start;
1307 buffer_begin = tmp_out_start;
1308 for (; start != begin; ++start, ++buffer_begin)
1309 *buffer_begin = static_cast<uint8_t>(*start);
1314 for (; tmp_out_start < buffer_begin; ++tmp_out_start) {
1315 if (*tmp_out_start > largest_character_for_kind) {
1317 PP.Diag(Loc, diag::err_character_too_large);
1324 // Is this a Universal Character Name escape?
1325 if (begin[1] == 'u' || begin[1] == 'U') {
1326 unsigned short UcnLen = 0;
1327 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen,
1328 FullSourceLoc(Loc, PP.getSourceManager()),
1329 &PP.getDiagnostics(), PP.getLangOpts(), true)) {
1331 } else if (*buffer_begin > largest_character_for_kind) {
1333 PP.Diag(Loc, diag::err_character_too_large);
1339 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo());
1341 ProcessCharEscape(TokBegin, begin, end, HadError,
1342 FullSourceLoc(Loc,PP.getSourceManager()),
1343 CharWidth, &PP.getDiagnostics(), PP.getLangOpts());
1344 *buffer_begin++ = result;
1347 unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front();
1349 if (NumCharsSoFar > 1) {
1351 PP.Diag(Loc, diag::warn_extraneous_char_constant);
1352 else if (isAscii() && NumCharsSoFar == 4)
1353 PP.Diag(Loc, diag::ext_four_char_character_literal);
1355 PP.Diag(Loc, diag::ext_multichar_character_literal);
1357 PP.Diag(Loc, diag::err_multichar_utf_character_literal);
1360 IsMultiChar = false;
1363 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0);
1365 // Narrow character literals act as though their value is concatenated
1366 // in this implementation, but warn on overflow.
1367 bool multi_char_too_long = false;
1368 if (isAscii() && isMultiChar()) {
1370 for (size_t i = 0; i < NumCharsSoFar; ++i) {
1371 // check for enough leading zeros to shift into
1372 multi_char_too_long |= (LitVal.countLeadingZeros() < 8);
1374 LitVal = LitVal + (codepoint_buffer[i] & 0xFF);
1376 } else if (NumCharsSoFar > 0) {
1377 // otherwise just take the last character
1378 LitVal = buffer_begin[-1];
1381 if (!HadError && multi_char_too_long) {
1382 PP.Diag(Loc, diag::warn_char_constant_too_large);
1385 // Transfer the value from APInt to uint64_t
1386 Value = LitVal.getZExtValue();
1388 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1")
1389 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple
1390 // character constants are not sign extended in the this implementation:
1391 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC.
1392 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) &&
1393 PP.getLangOpts().CharIsSigned)
1394 Value = (signed char)Value;
1398 /// string-literal: [C++0x lex.string]
1399 /// encoding-prefix " [s-char-sequence] "
1400 /// encoding-prefix R raw-string
1401 /// encoding-prefix:
1406 /// s-char-sequence:
1408 /// s-char-sequence s-char
1410 /// any member of the source character set except the double-quote ",
1411 /// backslash \, or new-line character
1413 /// universal-character-name
1415 /// " d-char-sequence ( r-char-sequence ) d-char-sequence "
1416 /// r-char-sequence:
1418 /// r-char-sequence r-char
1420 /// any member of the source character set, except a right parenthesis )
1421 /// followed by the initial d-char-sequence (which may be empty)
1422 /// followed by a double quote ".
1423 /// d-char-sequence:
1425 /// d-char-sequence d-char
1427 /// any member of the basic source character set except:
1428 /// space, the left parenthesis (, the right parenthesis ),
1429 /// the backslash \, and the control characters representing horizontal
1430 /// tab, vertical tab, form feed, and newline.
1431 /// escape-sequence: [C++0x lex.ccon]
1432 /// simple-escape-sequence
1433 /// octal-escape-sequence
1434 /// hexadecimal-escape-sequence
1435 /// simple-escape-sequence:
1436 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1437 /// octal-escape-sequence:
1439 /// \ octal-digit octal-digit
1440 /// \ octal-digit octal-digit octal-digit
1441 /// hexadecimal-escape-sequence:
1442 /// \x hexadecimal-digit
1443 /// hexadecimal-escape-sequence hexadecimal-digit
1444 /// universal-character-name:
1446 /// \U hex-quad hex-quad
1448 /// hex-digit hex-digit hex-digit hex-digit
1451 StringLiteralParser::
1452 StringLiteralParser(ArrayRef<Token> StringToks,
1453 Preprocessor &PP, bool Complain)
1454 : SM(PP.getSourceManager()), Features(PP.getLangOpts()),
1455 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() :nullptr),
1456 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown),
1457 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) {
1461 void StringLiteralParser::init(ArrayRef<Token> StringToks){
1462 // The literal token may have come from an invalid source location (e.g. due
1463 // to a PCH error), in which case the token length will be 0.
1464 if (StringToks.empty() || StringToks[0].getLength() < 2)
1465 return DiagnoseLexingError(SourceLocation());
1467 // Scan all of the string portions, remember the max individual token length,
1468 // computing a bound on the concatenated string length, and see whether any
1469 // piece is a wide-string. If any of the string portions is a wide-string
1470 // literal, the result is a wide-string literal [C99 6.4.5p4].
1471 assert(!StringToks.empty() && "expected at least one token");
1472 MaxTokenLength = StringToks[0].getLength();
1473 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!");
1474 SizeBound = StringToks[0].getLength()-2; // -2 for "".
1475 Kind = StringToks[0].getKind();
1479 // Implement Translation Phase #6: concatenation of string literals
1480 /// (C99 5.1.1.2p1). The common case is only one string fragment.
1481 for (unsigned i = 1; i != StringToks.size(); ++i) {
1482 if (StringToks[i].getLength() < 2)
1483 return DiagnoseLexingError(StringToks[i].getLocation());
1485 // The string could be shorter than this if it needs cleaning, but this is a
1486 // reasonable bound, which is all we need.
1487 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!");
1488 SizeBound += StringToks[i].getLength()-2; // -2 for "".
1490 // Remember maximum string piece length.
1491 if (StringToks[i].getLength() > MaxTokenLength)
1492 MaxTokenLength = StringToks[i].getLength();
1494 // Remember if we see any wide or utf-8/16/32 strings.
1495 // Also check for illegal concatenations.
1496 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) {
1498 Kind = StringToks[i].getKind();
1501 Diags->Report(StringToks[i].getLocation(),
1502 diag::err_unsupported_string_concat);
1508 // Include space for the null terminator.
1511 // TODO: K&R warning: "traditional C rejects string constant concatenation"
1513 // Get the width in bytes of char/wchar_t/char16_t/char32_t
1514 CharByteWidth = getCharWidth(Kind, Target);
1515 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple");
1518 // The output buffer size needs to be large enough to hold wide characters.
1519 // This is a worst-case assumption which basically corresponds to L"" "long".
1520 SizeBound *= CharByteWidth;
1522 // Size the temporary buffer to hold the result string data.
1523 ResultBuf.resize(SizeBound);
1525 // Likewise, but for each string piece.
1526 SmallString<512> TokenBuf;
1527 TokenBuf.resize(MaxTokenLength);
1529 // Loop over all the strings, getting their spelling, and expanding them to
1530 // wide strings as appropriate.
1531 ResultPtr = &ResultBuf[0]; // Next byte to fill in.
1535 SourceLocation UDSuffixTokLoc;
1537 for (unsigned i = 0, e = StringToks.size(); i != e; ++i) {
1538 const char *ThisTokBuf = &TokenBuf[0];
1539 // Get the spelling of the token, which eliminates trigraphs, etc. We know
1540 // that ThisTokBuf points to a buffer that is big enough for the whole token
1541 // and 'spelled' tokens can only shrink.
1542 bool StringInvalid = false;
1543 unsigned ThisTokLen =
1544 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features,
1547 return DiagnoseLexingError(StringToks[i].getLocation());
1549 const char *ThisTokBegin = ThisTokBuf;
1550 const char *ThisTokEnd = ThisTokBuf+ThisTokLen;
1552 // Remove an optional ud-suffix.
1553 if (ThisTokEnd[-1] != '"') {
1554 const char *UDSuffixEnd = ThisTokEnd;
1557 } while (ThisTokEnd[-1] != '"');
1559 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd);
1561 if (UDSuffixBuf.empty()) {
1562 if (StringToks[i].hasUCN())
1563 expandUCNs(UDSuffixBuf, UDSuffix);
1565 UDSuffixBuf.assign(UDSuffix);
1567 UDSuffixOffset = ThisTokEnd - ThisTokBuf;
1568 UDSuffixTokLoc = StringToks[i].getLocation();
1570 SmallString<32> ExpandedUDSuffix;
1571 if (StringToks[i].hasUCN()) {
1572 expandUCNs(ExpandedUDSuffix, UDSuffix);
1573 UDSuffix = ExpandedUDSuffix;
1576 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the
1577 // result of a concatenation involving at least one user-defined-string-
1578 // literal, all the participating user-defined-string-literals shall
1579 // have the same ud-suffix.
1580 if (UDSuffixBuf != UDSuffix) {
1582 SourceLocation TokLoc = StringToks[i].getLocation();
1583 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix)
1584 << UDSuffixBuf << UDSuffix
1585 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc)
1586 << SourceRange(TokLoc, TokLoc);
1593 // Strip the end quote.
1596 // TODO: Input character set mapping support.
1598 // Skip marker for wide or unicode strings.
1599 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') {
1601 // Skip 8 of u8 marker for utf8 strings.
1602 if (ThisTokBuf[0] == '8')
1606 // Check for raw string
1607 if (ThisTokBuf[0] == 'R') {
1608 ThisTokBuf += 2; // skip R"
1610 const char *Prefix = ThisTokBuf;
1611 while (ThisTokBuf[0] != '(')
1613 ++ThisTokBuf; // skip '('
1615 // Remove same number of characters from the end
1616 ThisTokEnd -= ThisTokBuf - Prefix;
1617 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal");
1619 // C++14 [lex.string]p4: A source-file new-line in a raw string literal
1620 // results in a new-line in the resulting execution string-literal.
1621 StringRef RemainingTokenSpan(ThisTokBuf, ThisTokEnd - ThisTokBuf);
1622 while (!RemainingTokenSpan.empty()) {
1623 // Split the string literal on \r\n boundaries.
1624 size_t CRLFPos = RemainingTokenSpan.find("\r\n");
1625 StringRef BeforeCRLF = RemainingTokenSpan.substr(0, CRLFPos);
1626 StringRef AfterCRLF = RemainingTokenSpan.substr(CRLFPos);
1628 // Copy everything before the \r\n sequence into the string literal.
1629 if (CopyStringFragment(StringToks[i], ThisTokBegin, BeforeCRLF))
1632 // Point into the \n inside the \r\n sequence and operate on the
1633 // remaining portion of the literal.
1634 RemainingTokenSpan = AfterCRLF.substr(1);
1637 if (ThisTokBuf[0] != '"') {
1638 // The file may have come from PCH and then changed after loading the
1639 // PCH; Fail gracefully.
1640 return DiagnoseLexingError(StringToks[i].getLocation());
1642 ++ThisTokBuf; // skip "
1644 // Check if this is a pascal string
1645 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd &&
1646 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') {
1648 // If the \p sequence is found in the first token, we have a pascal string
1649 // Otherwise, if we already have a pascal string, ignore the first \p
1657 while (ThisTokBuf != ThisTokEnd) {
1658 // Is this a span of non-escape characters?
1659 if (ThisTokBuf[0] != '\\') {
1660 const char *InStart = ThisTokBuf;
1663 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\');
1665 // Copy the character span over.
1666 if (CopyStringFragment(StringToks[i], ThisTokBegin,
1667 StringRef(InStart, ThisTokBuf - InStart)))
1671 // Is this a Universal Character Name escape?
1672 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') {
1673 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd,
1674 ResultPtr, hadError,
1675 FullSourceLoc(StringToks[i].getLocation(), SM),
1676 CharByteWidth, Diags, Features);
1679 // Otherwise, this is a non-UCN escape character. Process it.
1680 unsigned ResultChar =
1681 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError,
1682 FullSourceLoc(StringToks[i].getLocation(), SM),
1683 CharByteWidth*8, Diags, Features);
1685 if (CharByteWidth == 4) {
1686 // FIXME: Make the type of the result buffer correct instead of
1687 // using reinterpret_cast.
1688 llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultPtr);
1689 *ResultWidePtr = ResultChar;
1691 } else if (CharByteWidth == 2) {
1692 // FIXME: Make the type of the result buffer correct instead of
1693 // using reinterpret_cast.
1694 llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultPtr);
1695 *ResultWidePtr = ResultChar & 0xFFFF;
1698 assert(CharByteWidth == 1 && "Unexpected char width");
1699 *ResultPtr++ = ResultChar & 0xFF;
1706 if (CharByteWidth == 4) {
1707 // FIXME: Make the type of the result buffer correct instead of
1708 // using reinterpret_cast.
1709 llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultBuf.data());
1710 ResultWidePtr[0] = GetNumStringChars() - 1;
1711 } else if (CharByteWidth == 2) {
1712 // FIXME: Make the type of the result buffer correct instead of
1713 // using reinterpret_cast.
1714 llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultBuf.data());
1715 ResultWidePtr[0] = GetNumStringChars() - 1;
1717 assert(CharByteWidth == 1 && "Unexpected char width");
1718 ResultBuf[0] = GetNumStringChars() - 1;
1721 // Verify that pascal strings aren't too large.
1722 if (GetStringLength() > 256) {
1724 Diags->Report(StringToks.front().getLocation(),
1725 diag::err_pascal_string_too_long)
1726 << SourceRange(StringToks.front().getLocation(),
1727 StringToks.back().getLocation());
1732 // Complain if this string literal has too many characters.
1733 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509;
1735 if (GetNumStringChars() > MaxChars)
1736 Diags->Report(StringToks.front().getLocation(),
1737 diag::ext_string_too_long)
1738 << GetNumStringChars() << MaxChars
1739 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0)
1740 << SourceRange(StringToks.front().getLocation(),
1741 StringToks.back().getLocation());
1745 static const char *resyncUTF8(const char *Err, const char *End) {
1748 End = Err + std::min<unsigned>(llvm::getNumBytesForUTF8(*Err), End-Err);
1749 while (++Err != End && (*Err & 0xC0) == 0x80)
1754 /// This function copies from Fragment, which is a sequence of bytes
1755 /// within Tok's contents (which begin at TokBegin) into ResultPtr.
1756 /// Performs widening for multi-byte characters.
1757 bool StringLiteralParser::CopyStringFragment(const Token &Tok,
1758 const char *TokBegin,
1759 StringRef Fragment) {
1760 const llvm::UTF8 *ErrorPtrTmp;
1761 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp))
1764 // If we see bad encoding for unprefixed string literals, warn and
1765 // simply copy the byte values, for compatibility with gcc and older
1766 // versions of clang.
1767 bool NoErrorOnBadEncoding = isAscii();
1768 if (NoErrorOnBadEncoding) {
1769 memcpy(ResultPtr, Fragment.data(), Fragment.size());
1770 ResultPtr += Fragment.size();
1774 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1776 FullSourceLoc SourceLoc(Tok.getLocation(), SM);
1777 const DiagnosticBuilder &Builder =
1778 Diag(Diags, Features, SourceLoc, TokBegin,
1779 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()),
1780 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding
1781 : diag::err_bad_string_encoding);
1783 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1784 StringRef NextFragment(NextStart, Fragment.end()-NextStart);
1786 // Decode into a dummy buffer.
1787 SmallString<512> Dummy;
1788 Dummy.reserve(Fragment.size() * CharByteWidth);
1789 char *Ptr = Dummy.data();
1791 while (!ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) {
1792 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1793 NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1794 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin,
1795 ErrorPtr, NextStart);
1796 NextFragment = StringRef(NextStart, Fragment.end()-NextStart);
1799 return !NoErrorOnBadEncoding;
1802 void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) {
1805 Diags->Report(Loc, diag::err_lexing_string);
1808 /// getOffsetOfStringByte - This function returns the offset of the
1809 /// specified byte of the string data represented by Token. This handles
1810 /// advancing over escape sequences in the string.
1811 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok,
1812 unsigned ByteNo) const {
1813 // Get the spelling of the token.
1814 SmallString<32> SpellingBuffer;
1815 SpellingBuffer.resize(Tok.getLength());
1817 bool StringInvalid = false;
1818 const char *SpellingPtr = &SpellingBuffer[0];
1819 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features,
1824 const char *SpellingStart = SpellingPtr;
1825 const char *SpellingEnd = SpellingPtr+TokLen;
1827 // Handle UTF-8 strings just like narrow strings.
1828 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8')
1831 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' &&
1832 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet");
1834 // For raw string literals, this is easy.
1835 if (SpellingPtr[0] == 'R') {
1836 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!");
1839 while (*SpellingPtr != '(') {
1841 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal");
1845 return SpellingPtr - SpellingStart + ByteNo;
1848 // Skip over the leading quote
1849 assert(SpellingPtr[0] == '"' && "Should be a string literal!");
1852 // Skip over bytes until we find the offset we're looking for.
1854 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!");
1856 // Step over non-escapes simply.
1857 if (*SpellingPtr != '\\') {
1863 // Otherwise, this is an escape character. Advance over it.
1864 bool HadError = false;
1865 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') {
1866 const char *EscapePtr = SpellingPtr;
1867 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd,
1868 1, Features, HadError);
1870 // ByteNo is somewhere within the escape sequence.
1871 SpellingPtr = EscapePtr;
1876 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError,
1877 FullSourceLoc(Tok.getLocation(), SM),
1878 CharByteWidth*8, Diags, Features);
1881 assert(!HadError && "This method isn't valid on erroneous strings");
1884 return SpellingPtr-SpellingStart;
1887 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
1888 /// suffixes as ud-suffixes, because the diagnostic experience is better if we
1889 /// treat it as an invalid suffix.
1890 bool StringLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
1892 return NumericLiteralParser::isValidUDSuffix(LangOpts, Suffix) ||