1//===--- LiteralSupport.cpp - Code to parse and process literals ----------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file implements the NumericLiteralParser, CharLiteralParser, and
10// StringLiteralParser interfaces.
11//
12//===----------------------------------------------------------------------===//
13
14#include "clang/Lex/LiteralSupport.h"
15#include "clang/Basic/CharInfo.h"
16#include "clang/Basic/LangOptions.h"
17#include "clang/Basic/SourceLocation.h"
18#include "clang/Basic/TargetInfo.h"
19#include "clang/Lex/LexDiagnostic.h"
20#include "clang/Lex/Lexer.h"
21#include "clang/Lex/Preprocessor.h"
22#include "clang/Lex/Token.h"
23#include "llvm/ADT/APInt.h"
24#include "llvm/ADT/SmallVector.h"
25#include "llvm/ADT/StringExtras.h"
26#include "llvm/ADT/StringSwitch.h"
27#include "llvm/Support/ConvertUTF.h"
28#include "llvm/Support/Error.h"
29#include "llvm/Support/ErrorHandling.h"
30#include <algorithm>
31#include <cassert>
32#include <cstddef>
33#include <cstdint>
34#include <cstring>
35#include <string>
36
37using namespace clang;
38
39static unsigned getCharWidth(tok::TokenKind kind, const TargetInfo &Target) {
40  switch (kind) {
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();
56  }
57}
58
59static CharSourceRange MakeCharSourceRange(const LangOptions &Features,
60                                           FullSourceLoc TokLoc,
61                                           const char *TokBegin,
62                                           const char *TokRangeBegin,
63                                           const char *TokRangeEnd) {
64  SourceLocation Begin =
65    Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin,
66                                   TokLoc.getManager(), Features);
67  SourceLocation End =
68    Lexer::AdvanceToTokenCharacter(Begin, TokRangeEnd - TokRangeBegin,
69                                   TokLoc.getManager(), Features);
70  return CharSourceRange::getCharRange(Begin, End);
71}
72
73/// Produce a diagnostic highlighting some portion of a literal.
74///
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.
78static 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);
87}
88
89/// ProcessCharEscape - Parse a standard C escape sequence, which can occur in
90/// either a character or a string literal.
91static 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;
98
99  // Skip the '\' char.
100  ++ThisTokBuf;
101
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;
108
109    // These have fixed mappings.
110  case 'a':
111    // TODO: K&R: the meaning of '\\a' is different in traditional C
112    ResultChar = 7;
113    break;
114  case 'b':
115    ResultChar = 8;
116    break;
117  case 'e':
118    if (Diags)
119      Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
120           diag::ext_nonstandard_escape) << "e";
121    ResultChar = 27;
122    break;
123  case 'E':
124    if (Diags)
125      Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
126           diag::ext_nonstandard_escape) << "E";
127    ResultChar = 27;
128    break;
129  case 'f':
130    ResultChar = 12;
131    break;
132  case 'n':
133    ResultChar = 10;
134    break;
135  case 'r':
136    ResultChar = 13;
137    break;
138  case 't':
139    ResultChar = 9;
140    break;
141  case 'v':
142    ResultChar = 11;
143    break;
144  case 'x': { // Hex escape.
145    ResultChar = 0;
146    if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
147      if (Diags)
148        Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
149             diag::err_hex_escape_no_digits) << "x";
150      HadError = true;
151      break;
152    }
153
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)
161        Overflow = true;
162      ResultChar <<= 4;
163      ResultChar |= CharVal;
164    }
165
166    // See if any bits will be truncated when evaluated as a character.
167    if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
168      Overflow = true;
169      ResultChar &= ~0U >> (32-CharWidth);
170    }
171
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;
176    break;
177  }
178  case '0': case '1': case '2': case '3':
179  case '4': case '5': case '6': case '7': {
180    // Octal escapes.
181    --ThisTokBuf;
182    ResultChar = 0;
183
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;
187    do {
188      ResultChar <<= 3;
189      ResultChar |= *ThisTokBuf++ - '0';
190      ++NumDigits;
191    } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 &&
192             ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7');
193
194    // Check for overflow.  Reject '\777', but not L'\777'.
195    if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
196      if (Diags)
197        Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
198             diag::err_escape_too_large) << 1;
199      ResultChar &= ~0U >> (32-CharWidth);
200    }
201    break;
202  }
203
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.
207    if (Diags)
208      Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
209           diag::ext_nonstandard_escape)
210        << std::string(1, ResultChar);
211    break;
212  default:
213    if (!Diags)
214      break;
215
216    if (isPrintable(ResultChar))
217      Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
218           diag::ext_unknown_escape)
219        << std::string(1, ResultChar);
220    else
221      Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
222           diag::ext_unknown_escape)
223        << "x" + llvm::utohexstr(ResultChar);
224    break;
225  }
226
227  return ResultChar;
228}
229
230static void appendCodePoint(unsigned Codepoint,
231                            llvm::SmallVectorImpl<char> &Str) {
232  char ResultBuf[4];
233  char *ResultPtr = ResultBuf;
234  bool Res = llvm::ConvertCodePointToUTF8(Codepoint, ResultPtr);
235  (void)Res;
236  assert(Res && "Unexpected conversion failure");
237  Str.append(ResultBuf, ResultPtr);
238}
239
240void clang::expandUCNs(SmallVectorImpl<char> &Buf, StringRef Input) {
241  for (StringRef::iterator I = Input.begin(), E = Input.end(); I != E; ++I) {
242    if (*I != '\\') {
243      Buf.push_back(*I);
244      continue;
245    }
246
247    ++I;
248    assert(*I == 'u' || *I == 'U');
249
250    unsigned NumHexDigits;
251    if (*I == 'u')
252      NumHexDigits = 4;
253    else
254      NumHexDigits = 8;
255
256    assert(I + NumHexDigits <= E);
257
258    uint32_t CodePoint = 0;
259    for (++I; NumHexDigits != 0; ++I, --NumHexDigits) {
260      unsigned Value = llvm::hexDigitValue(*I);
261      assert(Value != -1U);
262
263      CodePoint <<= 4;
264      CodePoint += Value;
265    }
266
267    appendCodePoint(CodePoint, Buf);
268    --I;
269  }
270}
271
272/// ProcessUCNEscape - Read the Universal Character Name, check constraints and
273/// return the UTF32.
274static 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;
281
282  // Skip the '\u' char's.
283  ThisTokBuf += 2;
284
285  if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
286    if (Diags)
287      Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
288           diag::err_hex_escape_no_digits) << StringRef(&ThisTokBuf[-1], 1);
289    return false;
290  }
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;
296    UcnVal <<= 4;
297    UcnVal |= CharVal;
298  }
299  // If we didn't consume the proper number of digits, there is a problem.
300  if (UcnLenSave) {
301    if (Diags)
302      Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
303           diag::err_ucn_escape_incomplete);
304    return false;
305  }
306
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
310    if (Diags)
311      Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
312           diag::err_ucn_escape_invalid);
313    return false;
314  }
315
316  // C++11 allows UCNs that refer to control characters and basic source
317  // characters inside character and string literals
318  if (UcnVal < 0xa0 &&
319      (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60)) {  // $, @, `
320    bool IsError = (!Features.CPlusPlus11 || !in_char_string_literal);
321    if (Diags) {
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);
328      else
329        Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
330             IsError ? diag::err_ucn_control_character :
331                       diag::warn_cxx98_compat_literal_ucn_control_character);
332    }
333    if (IsError)
334      return false;
335  }
336
337  if (!Features.CPlusPlus && !Features.C99 && Diags)
338    Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
339         diag::warn_ucn_not_valid_in_c89_literal);
340
341  return true;
342}
343
344/// MeasureUCNEscape - Determine the number of bytes within the resulting string
345/// which this UCN will occupy.
346static 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)
351    return 4;
352
353  uint32_t UcnVal = 0;
354  unsigned short UcnLen = 0;
355  FullSourceLoc Loc;
356
357  if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal,
358                        UcnLen, Loc, nullptr, Features, true)) {
359    HadError = true;
360    return 0;
361  }
362
363  // UTF-16: 2 bytes for BMP, 4 bytes otherwise.
364  if (CharByteWidth == 2)
365    return UcnVal <= 0xFFFF ? 2 : 4;
366
367  // UTF-8.
368  if (UcnVal < 0x80)
369    return 1;
370  if (UcnVal < 0x800)
371    return 2;
372  if (UcnVal < 0x10000)
373    return 3;
374  return 4;
375}
376
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.
381static 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;
388  UTF32 UcnVal = 0;
389  unsigned short UcnLen = 0;
390  if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, UcnLen,
391                        Loc, Diags, Features, true)) {
392    HadError = true;
393    return;
394  }
395
396  assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth == 4) &&
397         "only character widths of 1, 2, or 4 bytes supported");
398
399  (void)UcnLen;
400  assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported");
401
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);
406    *ResultPtr = UcnVal;
407    ResultBuf += 4;
408    return;
409  }
410
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);
415
416    if (UcnVal <= (UTF32)0xFFFF) {
417      *ResultPtr = UcnVal;
418      ResultBuf += 2;
419      return;
420    }
421
422    // Convert to UTF16.
423    UcnVal -= 0x10000;
424    *ResultPtr     = 0xD800 + (UcnVal >> 10);
425    *(ResultPtr+1) = 0xDC00 + (UcnVal & 0x3FF);
426    ResultBuf += 4;
427    return;
428  }
429
430  assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters");
431
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;
437
438  unsigned short bytesToWrite = 0;
439  if (UcnVal < (UTF32)0x80)
440    bytesToWrite = 1;
441  else if (UcnVal < (UTF32)0x800)
442    bytesToWrite = 2;
443  else if (UcnVal < (UTF32)0x10000)
444    bytesToWrite = 3;
445  else
446    bytesToWrite = 4;
447
448  const unsigned byteMask = 0xBF;
449  const unsigned byteMark = 0x80;
450
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
455  };
456  // Finally, we write the bytes into ResultBuf.
457  ResultBuf += bytesToWrite;
458  switch (bytesToWrite) { // note: everything falls through.
459  case 4:
460    *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
461    LLVM_FALLTHROUGH;
462  case 3:
463    *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
464    LLVM_FALLTHROUGH;
465  case 2:
466    *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
467    LLVM_FALLTHROUGH;
468  case 1:
469    *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]);
470  }
471  // Update the buffer.
472  ResultBuf += bytesToWrite;
473}
474
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:
486///         nonzero-digit
487///         decimal-constant digit
488///       octal-constant:
489///         0
490///         octal-constant octal-digit
491///       hexadecimal-constant:
492///         hexadecimal-prefix hexadecimal-digit
493///         hexadecimal-constant hexadecimal-digit
494///       hexadecimal-prefix: one of
495///         0x 0X
496///       binary-literal:
497///         0b binary-digit
498///         0B binary-digit
499///         binary-literal binary-digit
500///       integer-suffix:
501///         unsigned-suffix [long-suffix]
502///         unsigned-suffix [long-long-suffix]
503///         long-suffix [unsigned-suffix]
504///         long-long-suffix [unsigned-sufix]
505///       nonzero-digit:
506///         1 2 3 4 5 6 7 8 9
507///       octal-digit:
508///         0 1 2 3 4 5 6 7
509///       hexadecimal-digit:
510///         0 1 2 3 4 5 6 7 8 9
511///         a b c d e f
512///         A B C D E F
513///       binary-digit:
514///         0
515///         1
516///       unsigned-suffix: one of
517///         u U
518///       long-suffix: one of
519///         l L
520///       long-long-suffix: one of
521///         ll LL
522///
523///       floating-constant: [C99 6.4.4.2]
524///         TODO: add rules...
525///
526NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
527                                           SourceLocation TokLoc,
528                                           const SourceManager &SM,
529                                           const LangOptions &LangOpts,
530                                           const TargetInfo &Target,
531                                           DiagnosticsEngine &Diags)
532    : SM(SM), LangOpts(LangOpts), Diags(Diags),
533      ThisTokBegin(TokSpelling.begin()), ThisTokEnd(TokSpelling.end()) {
534
535  // This routine assumes that the range begin/end matches the regex for integer
536  // and FP constants (specifically, the 'pp-number' regex), and assumes that
537  // the byte at "*end" is both valid and not part of the regex.  Because of
538  // this, it doesn't have to check for 'overscan' in various places.
539  assert(!isPreprocessingNumberBody(*ThisTokEnd) && "didn't maximally munch?");
540
541  s = DigitsBegin = ThisTokBegin;
542  saw_exponent = false;
543  saw_period = false;
544  saw_ud_suffix = false;
545  saw_fixed_point_suffix = false;
546  isLong = false;
547  isUnsigned = false;
548  isLongLong = false;
549  isSizeT = false;
550  isHalf = false;
551  isFloat = false;
552  isImaginary = false;
553  isFloat16 = false;
554  isFloat128 = false;
555  MicrosoftInteger = 0;
556  isFract = false;
557  isAccum = false;
558  hadError = false;
559
560  if (*s == '0') { // parse radix
561    ParseNumberStartingWithZero(TokLoc);
562    if (hadError)
563      return;
564  } else { // the first digit is non-zero
565    radix = 10;
566    s = SkipDigits(s);
567    if (s == ThisTokEnd) {
568      // Done.
569    } else {
570      ParseDecimalOrOctalCommon(TokLoc);
571      if (hadError)
572        return;
573    }
574  }
575
576  SuffixBegin = s;
577  checkSeparator(TokLoc, s, CSK_AfterDigits);
578
579  // Initial scan to lookahead for fixed point suffix.
580  if (LangOpts.FixedPoint) {
581    for (const char *c = s; c != ThisTokEnd; ++c) {
582      if (*c == 'r' || *c == 'k' || *c == 'R' || *c == 'K') {
583        saw_fixed_point_suffix = true;
584        break;
585      }
586    }
587  }
588
589  // Parse the suffix.  At this point we can classify whether we have an FP or
590  // integer constant.
591  bool isFixedPointConstant = isFixedPointLiteral();
592  bool isFPConstant = isFloatingLiteral();
593  bool HasSize = false;
594
595  // Loop over all of the characters of the suffix.  If we see something bad,
596  // we break out of the loop.
597  for (; s != ThisTokEnd; ++s) {
598    switch (*s) {
599    case 'R':
600    case 'r':
601      if (!LangOpts.FixedPoint)
602        break;
603      if (isFract || isAccum) break;
604      if (!(saw_period || saw_exponent)) break;
605      isFract = true;
606      continue;
607    case 'K':
608    case 'k':
609      if (!LangOpts.FixedPoint)
610        break;
611      if (isFract || isAccum) break;
612      if (!(saw_period || saw_exponent)) break;
613      isAccum = true;
614      continue;
615    case 'h':      // FP Suffix for "half".
616    case 'H':
617      // OpenCL Extension v1.2 s9.5 - h or H suffix for half type.
618      if (!(LangOpts.Half || LangOpts.FixedPoint))
619        break;
620      if (isIntegerLiteral()) break;  // Error for integer constant.
621      if (HasSize)
622        break;
623      HasSize = true;
624      isHalf = true;
625      continue;  // Success.
626    case 'f':      // FP Suffix for "float"
627    case 'F':
628      if (!isFPConstant) break;  // Error for integer constant.
629      if (HasSize)
630        break;
631      HasSize = true;
632
633      // CUDA host and device may have different _Float16 support, therefore
634      // allows f16 literals to avoid false alarm.
635      // ToDo: more precise check for CUDA.
636      if ((Target.hasFloat16Type() || LangOpts.CUDA) && s + 2 < ThisTokEnd &&
637          s[1] == '1' && s[2] == '6') {
638        s += 2; // success, eat up 2 characters.
639        isFloat16 = true;
640        continue;
641      }
642
643      isFloat = true;
644      continue;  // Success.
645    case 'q':    // FP Suffix for "__float128"
646    case 'Q':
647      if (!isFPConstant) break;  // Error for integer constant.
648      if (HasSize)
649        break;
650      HasSize = true;
651      isFloat128 = true;
652      continue;  // Success.
653    case 'u':
654    case 'U':
655      if (isFPConstant) break;  // Error for floating constant.
656      if (isUnsigned) break;    // Cannot be repeated.
657      isUnsigned = true;
658      continue;  // Success.
659    case 'l':
660    case 'L':
661      if (HasSize)
662        break;
663      HasSize = true;
664
665      // Check for long long.  The L's need to be adjacent and the same case.
666      if (s[1] == s[0]) {
667        assert(s + 1 < ThisTokEnd && "didn't maximally munch?");
668        if (isFPConstant) break;        // long long invalid for floats.
669        isLongLong = true;
670        ++s;  // Eat both of them.
671      } else {
672        isLong = true;
673      }
674      continue; // Success.
675    case 'z':
676    case 'Z':
677      if (isFPConstant)
678        break; // Invalid for floats.
679      if (HasSize)
680        break;
681      HasSize = true;
682      isSizeT = true;
683      continue;
684    case 'i':
685    case 'I':
686      if (LangOpts.MicrosoftExt && !isFPConstant) {
687        // Allow i8, i16, i32, and i64. First, look ahead and check if
688        // suffixes are Microsoft integers and not the imaginary unit.
689        uint8_t Bits = 0;
690        size_t ToSkip = 0;
691        switch (s[1]) {
692        case '8': // i8 suffix
693          Bits = 8;
694          ToSkip = 2;
695          break;
696        case '1':
697          if (s[2] == '6') { // i16 suffix
698            Bits = 16;
699            ToSkip = 3;
700          }
701          break;
702        case '3':
703          if (s[2] == '2') { // i32 suffix
704            Bits = 32;
705            ToSkip = 3;
706          }
707          break;
708        case '6':
709          if (s[2] == '4') { // i64 suffix
710            Bits = 64;
711            ToSkip = 3;
712          }
713          break;
714        default:
715          break;
716        }
717        if (Bits) {
718          if (HasSize)
719            break;
720          HasSize = true;
721          MicrosoftInteger = Bits;
722          s += ToSkip;
723          assert(s <= ThisTokEnd && "didn't maximally munch?");
724          break;
725        }
726      }
727      LLVM_FALLTHROUGH;
728    case 'j':
729    case 'J':
730      if (isImaginary) break;   // Cannot be repeated.
731      isImaginary = true;
732      continue;  // Success.
733    }
734    // If we reached here, there was an error or a ud-suffix.
735    break;
736  }
737
738  // "i", "if", and "il" are user-defined suffixes in C++1y.
739  if (s != ThisTokEnd || isImaginary) {
740    // FIXME: Don't bother expanding UCNs if !tok.hasUCN().
741    expandUCNs(UDSuffixBuf, StringRef(SuffixBegin, ThisTokEnd - SuffixBegin));
742    if (isValidUDSuffix(LangOpts, UDSuffixBuf)) {
743      if (!isImaginary) {
744        // Any suffix pieces we might have parsed are actually part of the
745        // ud-suffix.
746        isLong = false;
747        isUnsigned = false;
748        isLongLong = false;
749        isSizeT = false;
750        isFloat = false;
751        isFloat16 = false;
752        isHalf = false;
753        isImaginary = false;
754        MicrosoftInteger = 0;
755        saw_fixed_point_suffix = false;
756        isFract = false;
757        isAccum = false;
758      }
759
760      saw_ud_suffix = true;
761      return;
762    }
763
764    if (s != ThisTokEnd) {
765      // Report an error if there are any.
766      Diags.Report(Lexer::AdvanceToTokenCharacter(
767                       TokLoc, SuffixBegin - ThisTokBegin, SM, LangOpts),
768                   diag::err_invalid_suffix_constant)
769          << StringRef(SuffixBegin, ThisTokEnd - SuffixBegin)
770          << (isFixedPointConstant ? 2 : isFPConstant);
771      hadError = true;
772    }
773  }
774
775  if (!hadError && saw_fixed_point_suffix) {
776    assert(isFract || isAccum);
777  }
778}
779
780/// ParseDecimalOrOctalCommon - This method is called for decimal or octal
781/// numbers. It issues an error for illegal digits, and handles floating point
782/// parsing. If it detects a floating point number, the radix is set to 10.
783void NumericLiteralParser::ParseDecimalOrOctalCommon(SourceLocation TokLoc){
784  assert((radix == 8 || radix == 10) && "Unexpected radix");
785
786  // If we have a hex digit other than 'e' (which denotes a FP exponent) then
787  // the code is using an incorrect base.
788  if (isHexDigit(*s) && *s != 'e' && *s != 'E' &&
789      !isValidUDSuffix(LangOpts, StringRef(s, ThisTokEnd - s))) {
790    Diags.Report(
791        Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM, LangOpts),
792        diag::err_invalid_digit)
793        << StringRef(s, 1) << (radix == 8 ? 1 : 0);
794    hadError = true;
795    return;
796  }
797
798  if (*s == '.') {
799    checkSeparator(TokLoc, s, CSK_AfterDigits);
800    s++;
801    radix = 10;
802    saw_period = true;
803    checkSeparator(TokLoc, s, CSK_BeforeDigits);
804    s = SkipDigits(s); // Skip suffix.
805  }
806  if (*s == 'e' || *s == 'E') { // exponent
807    checkSeparator(TokLoc, s, CSK_AfterDigits);
808    const char *Exponent = s;
809    s++;
810    radix = 10;
811    saw_exponent = true;
812    if (s != ThisTokEnd && (*s == '+' || *s == '-'))  s++; // sign
813    const char *first_non_digit = SkipDigits(s);
814    if (containsDigits(s, first_non_digit)) {
815      checkSeparator(TokLoc, s, CSK_BeforeDigits);
816      s = first_non_digit;
817    } else {
818      if (!hadError) {
819        Diags.Report(Lexer::AdvanceToTokenCharacter(
820                         TokLoc, Exponent - ThisTokBegin, SM, LangOpts),
821                     diag::err_exponent_has_no_digits);
822        hadError = true;
823      }
824      return;
825    }
826  }
827}
828
829/// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
830/// suffixes as ud-suffixes, because the diagnostic experience is better if we
831/// treat it as an invalid suffix.
832bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
833                                           StringRef Suffix) {
834  if (!LangOpts.CPlusPlus11 || Suffix.empty())
835    return false;
836
837  // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid.
838  if (Suffix[0] == '_')
839    return true;
840
841  // In C++11, there are no library suffixes.
842  if (!LangOpts.CPlusPlus14)
843    return false;
844
845  // In C++14, "s", "h", "min", "ms", "us", and "ns" are used in the library.
846  // Per tweaked N3660, "il", "i", and "if" are also used in the library.
847  // In C++2a "d" and "y" are used in the library.
848  return llvm::StringSwitch<bool>(Suffix)
849      .Cases("h", "min", "s", true)
850      .Cases("ms", "us", "ns", true)
851      .Cases("il", "i", "if", true)
852      .Cases("d", "y", LangOpts.CPlusPlus20)
853      .Default(false);
854}
855
856void NumericLiteralParser::checkSeparator(SourceLocation TokLoc,
857                                          const char *Pos,
858                                          CheckSeparatorKind IsAfterDigits) {
859  if (IsAfterDigits == CSK_AfterDigits) {
860    if (Pos == ThisTokBegin)
861      return;
862    --Pos;
863  } else if (Pos == ThisTokEnd)
864    return;
865
866  if (isDigitSeparator(*Pos)) {
867    Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin, SM,
868                                                LangOpts),
869                 diag::err_digit_separator_not_between_digits)
870        << IsAfterDigits;
871    hadError = true;
872  }
873}
874
875/// ParseNumberStartingWithZero - This method is called when the first character
876/// of the number is found to be a zero.  This means it is either an octal
877/// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or
878/// a floating point number (01239.123e4).  Eat the prefix, determining the
879/// radix etc.
880void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
881  assert(s[0] == '0' && "Invalid method call");
882  s++;
883
884  int c1 = s[0];
885
886  // Handle a hex number like 0x1234.
887  if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) {
888    s++;
889    assert(s < ThisTokEnd && "didn't maximally munch?");
890    radix = 16;
891    DigitsBegin = s;
892    s = SkipHexDigits(s);
893    bool HasSignificandDigits = containsDigits(DigitsBegin, s);
894    if (s == ThisTokEnd) {
895      // Done.
896    } else if (*s == '.') {
897      s++;
898      saw_period = true;
899      const char *floatDigitsBegin = s;
900      s = SkipHexDigits(s);
901      if (containsDigits(floatDigitsBegin, s))
902        HasSignificandDigits = true;
903      if (HasSignificandDigits)
904        checkSeparator(TokLoc, floatDigitsBegin, CSK_BeforeDigits);
905    }
906
907    if (!HasSignificandDigits) {
908      Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM,
909                                                  LangOpts),
910                   diag::err_hex_constant_requires)
911          << LangOpts.CPlusPlus << 1;
912      hadError = true;
913      return;
914    }
915
916    // A binary exponent can appear with or with a '.'. If dotted, the
917    // binary exponent is required.
918    if (*s == 'p' || *s == 'P') {
919      checkSeparator(TokLoc, s, CSK_AfterDigits);
920      const char *Exponent = s;
921      s++;
922      saw_exponent = true;
923      if (s != ThisTokEnd && (*s == '+' || *s == '-'))  s++; // sign
924      const char *first_non_digit = SkipDigits(s);
925      if (!containsDigits(s, first_non_digit)) {
926        if (!hadError) {
927          Diags.Report(Lexer::AdvanceToTokenCharacter(
928                           TokLoc, Exponent - ThisTokBegin, SM, LangOpts),
929                       diag::err_exponent_has_no_digits);
930          hadError = true;
931        }
932        return;
933      }
934      checkSeparator(TokLoc, s, CSK_BeforeDigits);
935      s = first_non_digit;
936
937      if (!LangOpts.HexFloats)
938        Diags.Report(TokLoc, LangOpts.CPlusPlus
939                                 ? diag::ext_hex_literal_invalid
940                                 : diag::ext_hex_constant_invalid);
941      else if (LangOpts.CPlusPlus17)
942        Diags.Report(TokLoc, diag::warn_cxx17_hex_literal);
943    } else if (saw_period) {
944      Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM,
945                                                  LangOpts),
946                   diag::err_hex_constant_requires)
947          << LangOpts.CPlusPlus << 0;
948      hadError = true;
949    }
950    return;
951  }
952
953  // Handle simple binary numbers 0b01010
954  if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) {
955    // 0b101010 is a C++1y / GCC extension.
956    Diags.Report(TokLoc, LangOpts.CPlusPlus14
957                             ? diag::warn_cxx11_compat_binary_literal
958                         : LangOpts.CPlusPlus ? diag::ext_binary_literal_cxx14
959                                              : diag::ext_binary_literal);
960    ++s;
961    assert(s < ThisTokEnd && "didn't maximally munch?");
962    radix = 2;
963    DigitsBegin = s;
964    s = SkipBinaryDigits(s);
965    if (s == ThisTokEnd) {
966      // Done.
967    } else if (isHexDigit(*s) &&
968               !isValidUDSuffix(LangOpts, StringRef(s, ThisTokEnd - s))) {
969      Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM,
970                                                  LangOpts),
971                   diag::err_invalid_digit)
972          << StringRef(s, 1) << 2;
973      hadError = true;
974    }
975    // Other suffixes will be diagnosed by the caller.
976    return;
977  }
978
979  // For now, the radix is set to 8. If we discover that we have a
980  // floating point constant, the radix will change to 10. Octal floating
981  // point constants are not permitted (only decimal and hexadecimal).
982  radix = 8;
983  DigitsBegin = s;
984  s = SkipOctalDigits(s);
985  if (s == ThisTokEnd)
986    return; // Done, simple octal number like 01234
987
988  // If we have some other non-octal digit that *is* a decimal digit, see if
989  // this is part of a floating point number like 094.123 or 09e1.
990  if (isDigit(*s)) {
991    const char *EndDecimal = SkipDigits(s);
992    if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') {
993      s = EndDecimal;
994      radix = 10;
995    }
996  }
997
998  ParseDecimalOrOctalCommon(TokLoc);
999}
1000
1001static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) {
1002  switch (Radix) {
1003  case 2:
1004    return NumDigits <= 64;
1005  case 8:
1006    return NumDigits <= 64 / 3; // Digits are groups of 3 bits.
1007  case 10:
1008    return NumDigits <= 19; // floor(log10(2^64))
1009  case 16:
1010    return NumDigits <= 64 / 4; // Digits are groups of 4 bits.
1011  default:
1012    llvm_unreachable("impossible Radix");
1013  }
1014}
1015
1016/// GetIntegerValue - Convert this numeric literal value to an APInt that
1017/// matches Val's input width.  If there is an overflow, set Val to the low bits
1018/// of the result and return true.  Otherwise, return false.
1019bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) {
1020  // Fast path: Compute a conservative bound on the maximum number of
1021  // bits per digit in this radix. If we can't possibly overflow a
1022  // uint64 based on that bound then do the simple conversion to
1023  // integer. This avoids the expensive overflow checking below, and
1024  // handles the common cases that matter (small decimal integers and
1025  // hex/octal values which don't overflow).
1026  const unsigned NumDigits = SuffixBegin - DigitsBegin;
1027  if (alwaysFitsInto64Bits(radix, NumDigits)) {
1028    uint64_t N = 0;
1029    for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr)
1030      if (!isDigitSeparator(*Ptr))
1031        N = N * radix + llvm::hexDigitValue(*Ptr);
1032
1033    // This will truncate the value to Val's input width. Simply check
1034    // for overflow by comparing.
1035    Val = N;
1036    return Val.getZExtValue() != N;
1037  }
1038
1039  Val = 0;
1040  const char *Ptr = DigitsBegin;
1041
1042  llvm::APInt RadixVal(Val.getBitWidth(), radix);
1043  llvm::APInt CharVal(Val.getBitWidth(), 0);
1044  llvm::APInt OldVal = Val;
1045
1046  bool OverflowOccurred = false;
1047  while (Ptr < SuffixBegin) {
1048    if (isDigitSeparator(*Ptr)) {
1049      ++Ptr;
1050      continue;
1051    }
1052
1053    unsigned C = llvm::hexDigitValue(*Ptr++);
1054
1055    // If this letter is out of bound for this radix, reject it.
1056    assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1057
1058    CharVal = C;
1059
1060    // Add the digit to the value in the appropriate radix.  If adding in digits
1061    // made the value smaller, then this overflowed.
1062    OldVal = Val;
1063
1064    // Multiply by radix, did overflow occur on the multiply?
1065    Val *= RadixVal;
1066    OverflowOccurred |= Val.udiv(RadixVal) != OldVal;
1067
1068    // Add value, did overflow occur on the value?
1069    //   (a + b) ult b  <=> overflow
1070    Val += CharVal;
1071    OverflowOccurred |= Val.ult(CharVal);
1072  }
1073  return OverflowOccurred;
1074}
1075
1076llvm::APFloat::opStatus
1077NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) {
1078  using llvm::APFloat;
1079
1080  unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin);
1081
1082  llvm::SmallString<16> Buffer;
1083  StringRef Str(ThisTokBegin, n);
1084  if (Str.find('\'') != StringRef::npos) {
1085    Buffer.reserve(n);
1086    std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer),
1087                        &isDigitSeparator);
1088    Str = Buffer;
1089  }
1090
1091  auto StatusOrErr =
1092      Result.convertFromString(Str, APFloat::rmNearestTiesToEven);
1093  assert(StatusOrErr && "Invalid floating point representation");
1094  return !errorToBool(StatusOrErr.takeError()) ? *StatusOrErr
1095                                               : APFloat::opInvalidOp;
1096}
1097
1098static inline bool IsExponentPart(char c) {
1099  return c == 'p' || c == 'P' || c == 'e' || c == 'E';
1100}
1101
1102bool NumericLiteralParser::GetFixedPointValue(llvm::APInt &StoreVal, unsigned Scale) {
1103  assert(radix == 16 || radix == 10);
1104
1105  // Find how many digits are needed to store the whole literal.
1106  unsigned NumDigits = SuffixBegin - DigitsBegin;
1107  if (saw_period) --NumDigits;
1108
1109  // Initial scan of the exponent if it exists
1110  bool ExpOverflowOccurred = false;
1111  bool NegativeExponent = false;
1112  const char *ExponentBegin;
1113  uint64_t Exponent = 0;
1114  int64_t BaseShift = 0;
1115  if (saw_exponent) {
1116    const char *Ptr = DigitsBegin;
1117
1118    while (!IsExponentPart(*Ptr)) ++Ptr;
1119    ExponentBegin = Ptr;
1120    ++Ptr;
1121    NegativeExponent = *Ptr == '-';
1122    if (NegativeExponent) ++Ptr;
1123
1124    unsigned NumExpDigits = SuffixBegin - Ptr;
1125    if (alwaysFitsInto64Bits(radix, NumExpDigits)) {
1126      llvm::StringRef ExpStr(Ptr, NumExpDigits);
1127      llvm::APInt ExpInt(/*numBits=*/64, ExpStr, /*radix=*/10);
1128      Exponent = ExpInt.getZExtValue();
1129    } else {
1130      ExpOverflowOccurred = true;
1131    }
1132
1133    if (NegativeExponent) BaseShift -= Exponent;
1134    else BaseShift += Exponent;
1135  }
1136
1137  // Number of bits needed for decimal literal is
1138  //   ceil(NumDigits * log2(10))       Integral part
1139  // + Scale                            Fractional part
1140  // + ceil(Exponent * log2(10))        Exponent
1141  // --------------------------------------------------
1142  //   ceil((NumDigits + Exponent) * log2(10)) + Scale
1143  //
1144  // But for simplicity in handling integers, we can round up log2(10) to 4,
1145  // making:
1146  // 4 * (NumDigits + Exponent) + Scale
1147  //
1148  // Number of digits needed for hexadecimal literal is
1149  //   4 * NumDigits                    Integral part
1150  // + Scale                            Fractional part
1151  // + Exponent                         Exponent
1152  // --------------------------------------------------
1153  //   (4 * NumDigits) + Scale + Exponent
1154  uint64_t NumBitsNeeded;
1155  if (radix == 10)
1156    NumBitsNeeded = 4 * (NumDigits + Exponent) + Scale;
1157  else
1158    NumBitsNeeded = 4 * NumDigits + Exponent + Scale;
1159
1160  if (NumBitsNeeded > std::numeric_limits<unsigned>::max())
1161    ExpOverflowOccurred = true;
1162  llvm::APInt Val(static_cast<unsigned>(NumBitsNeeded), 0, /*isSigned=*/false);
1163
1164  bool FoundDecimal = false;
1165
1166  int64_t FractBaseShift = 0;
1167  const char *End = saw_exponent ? ExponentBegin : SuffixBegin;
1168  for (const char *Ptr = DigitsBegin; Ptr < End; ++Ptr) {
1169    if (*Ptr == '.') {
1170      FoundDecimal = true;
1171      continue;
1172    }
1173
1174    // Normal reading of an integer
1175    unsigned C = llvm::hexDigitValue(*Ptr);
1176    assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1177
1178    Val *= radix;
1179    Val += C;
1180
1181    if (FoundDecimal)
1182      // Keep track of how much we will need to adjust this value by from the
1183      // number of digits past the radix point.
1184      --FractBaseShift;
1185  }
1186
1187  // For a radix of 16, we will be multiplying by 2 instead of 16.
1188  if (radix == 16) FractBaseShift *= 4;
1189  BaseShift += FractBaseShift;
1190
1191  Val <<= Scale;
1192
1193  uint64_t Base = (radix == 16) ? 2 : 10;
1194  if (BaseShift > 0) {
1195    for (int64_t i = 0; i < BaseShift; ++i) {
1196      Val *= Base;
1197    }
1198  } else if (BaseShift < 0) {
1199    for (int64_t i = BaseShift; i < 0 && !Val.isNullValue(); ++i)
1200      Val = Val.udiv(Base);
1201  }
1202
1203  bool IntOverflowOccurred = false;
1204  auto MaxVal = llvm::APInt::getMaxValue(StoreVal.getBitWidth());
1205  if (Val.getBitWidth() > StoreVal.getBitWidth()) {
1206    IntOverflowOccurred |= Val.ugt(MaxVal.zext(Val.getBitWidth()));
1207    StoreVal = Val.trunc(StoreVal.getBitWidth());
1208  } else if (Val.getBitWidth() < StoreVal.getBitWidth()) {
1209    IntOverflowOccurred |= Val.zext(MaxVal.getBitWidth()).ugt(MaxVal);
1210    StoreVal = Val.zext(StoreVal.getBitWidth());
1211  } else {
1212    StoreVal = Val;
1213  }
1214
1215  return IntOverflowOccurred || ExpOverflowOccurred;
1216}
1217
1218/// \verbatim
1219///       user-defined-character-literal: [C++11 lex.ext]
1220///         character-literal ud-suffix
1221///       ud-suffix:
1222///         identifier
1223///       character-literal: [C++11 lex.ccon]
1224///         ' c-char-sequence '
1225///         u' c-char-sequence '
1226///         U' c-char-sequence '
1227///         L' c-char-sequence '
1228///         u8' c-char-sequence ' [C++1z lex.ccon]
1229///       c-char-sequence:
1230///         c-char
1231///         c-char-sequence c-char
1232///       c-char:
1233///         any member of the source character set except the single-quote ',
1234///           backslash \, or new-line character
1235///         escape-sequence
1236///         universal-character-name
1237///       escape-sequence:
1238///         simple-escape-sequence
1239///         octal-escape-sequence
1240///         hexadecimal-escape-sequence
1241///       simple-escape-sequence:
1242///         one of \' \" \? \\ \a \b \f \n \r \t \v
1243///       octal-escape-sequence:
1244///         \ octal-digit
1245///         \ octal-digit octal-digit
1246///         \ octal-digit octal-digit octal-digit
1247///       hexadecimal-escape-sequence:
1248///         \x hexadecimal-digit
1249///         hexadecimal-escape-sequence hexadecimal-digit
1250///       universal-character-name: [C++11 lex.charset]
1251///         \u hex-quad
1252///         \U hex-quad hex-quad
1253///       hex-quad:
1254///         hex-digit hex-digit hex-digit hex-digit
1255/// \endverbatim
1256///
1257CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
1258                                     SourceLocation Loc, Preprocessor &PP,
1259                                     tok::TokenKind kind) {
1260  // At this point we know that the character matches the regex "(L|u|U)?'.*'".
1261  HadError = false;
1262
1263  Kind = kind;
1264
1265  const char *TokBegin = begin;
1266
1267  // Skip over wide character determinant.
1268  if (Kind != tok::char_constant)
1269    ++begin;
1270  if (Kind == tok::utf8_char_constant)
1271    ++begin;
1272
1273  // Skip over the entry quote.
1274  assert(begin[0] == '\'' && "Invalid token lexed");
1275  ++begin;
1276
1277  // Remove an optional ud-suffix.
1278  if (end[-1] != '\'') {
1279    const char *UDSuffixEnd = end;
1280    do {
1281      --end;
1282    } while (end[-1] != '\'');
1283    // FIXME: Don't bother with this if !tok.hasUCN().
1284    expandUCNs(UDSuffixBuf, StringRef(end, UDSuffixEnd - end));
1285    UDSuffixOffset = end - TokBegin;
1286  }
1287
1288  // Trim the ending quote.
1289  assert(end != begin && "Invalid token lexed");
1290  --end;
1291
1292  // FIXME: The "Value" is an uint64_t so we can handle char literals of
1293  // up to 64-bits.
1294  // FIXME: This extensively assumes that 'char' is 8-bits.
1295  assert(PP.getTargetInfo().getCharWidth() == 8 &&
1296         "Assumes char is 8 bits");
1297  assert(PP.getTargetInfo().getIntWidth() <= 64 &&
1298         (PP.getTargetInfo().getIntWidth() & 7) == 0 &&
1299         "Assumes sizeof(int) on target is <= 64 and a multiple of char");
1300  assert(PP.getTargetInfo().getWCharWidth() <= 64 &&
1301         "Assumes sizeof(wchar) on target is <= 64");
1302
1303  SmallVector<uint32_t, 4> codepoint_buffer;
1304  codepoint_buffer.resize(end - begin);
1305  uint32_t *buffer_begin = &codepoint_buffer.front();
1306  uint32_t *buffer_end = buffer_begin + codepoint_buffer.size();
1307
1308  // Unicode escapes representing characters that cannot be correctly
1309  // represented in a single code unit are disallowed in character literals
1310  // by this implementation.
1311  uint32_t largest_character_for_kind;
1312  if (tok::wide_char_constant == Kind) {
1313    largest_character_for_kind =
1314        0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth());
1315  } else if (tok::utf8_char_constant == Kind) {
1316    largest_character_for_kind = 0x7F;
1317  } else if (tok::utf16_char_constant == Kind) {
1318    largest_character_for_kind = 0xFFFF;
1319  } else if (tok::utf32_char_constant == Kind) {
1320    largest_character_for_kind = 0x10FFFF;
1321  } else {
1322    largest_character_for_kind = 0x7Fu;
1323  }
1324
1325  while (begin != end) {
1326    // Is this a span of non-escape characters?
1327    if (begin[0] != '\\') {
1328      char const *start = begin;
1329      do {
1330        ++begin;
1331      } while (begin != end && *begin != '\\');
1332
1333      char const *tmp_in_start = start;
1334      uint32_t *tmp_out_start = buffer_begin;
1335      llvm::ConversionResult res =
1336          llvm::ConvertUTF8toUTF32(reinterpret_cast<llvm::UTF8 const **>(&start),
1337                             reinterpret_cast<llvm::UTF8 const *>(begin),
1338                             &buffer_begin, buffer_end, llvm::strictConversion);
1339      if (res != llvm::conversionOK) {
1340        // If we see bad encoding for unprefixed character literals, warn and
1341        // simply copy the byte values, for compatibility with gcc and
1342        // older versions of clang.
1343        bool NoErrorOnBadEncoding = isAscii();
1344        unsigned Msg = diag::err_bad_character_encoding;
1345        if (NoErrorOnBadEncoding)
1346          Msg = diag::warn_bad_character_encoding;
1347        PP.Diag(Loc, Msg);
1348        if (NoErrorOnBadEncoding) {
1349          start = tmp_in_start;
1350          buffer_begin = tmp_out_start;
1351          for (; start != begin; ++start, ++buffer_begin)
1352            *buffer_begin = static_cast<uint8_t>(*start);
1353        } else {
1354          HadError = true;
1355        }
1356      } else {
1357        for (; tmp_out_start < buffer_begin; ++tmp_out_start) {
1358          if (*tmp_out_start > largest_character_for_kind) {
1359            HadError = true;
1360            PP.Diag(Loc, diag::err_character_too_large);
1361          }
1362        }
1363      }
1364
1365      continue;
1366    }
1367    // Is this a Universal Character Name escape?
1368    if (begin[1] == 'u' || begin[1] == 'U') {
1369      unsigned short UcnLen = 0;
1370      if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen,
1371                            FullSourceLoc(Loc, PP.getSourceManager()),
1372                            &PP.getDiagnostics(), PP.getLangOpts(), true)) {
1373        HadError = true;
1374      } else if (*buffer_begin > largest_character_for_kind) {
1375        HadError = true;
1376        PP.Diag(Loc, diag::err_character_too_large);
1377      }
1378
1379      ++buffer_begin;
1380      continue;
1381    }
1382    unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo());
1383    uint64_t result =
1384      ProcessCharEscape(TokBegin, begin, end, HadError,
1385                        FullSourceLoc(Loc,PP.getSourceManager()),
1386                        CharWidth, &PP.getDiagnostics(), PP.getLangOpts());
1387    *buffer_begin++ = result;
1388  }
1389
1390  unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front();
1391
1392  if (NumCharsSoFar > 1) {
1393    if (isWide())
1394      PP.Diag(Loc, diag::warn_extraneous_char_constant);
1395    else if (isAscii() && NumCharsSoFar == 4)
1396      PP.Diag(Loc, diag::warn_four_char_character_literal);
1397    else if (isAscii())
1398      PP.Diag(Loc, diag::warn_multichar_character_literal);
1399    else
1400      PP.Diag(Loc, diag::err_multichar_utf_character_literal);
1401    IsMultiChar = true;
1402  } else {
1403    IsMultiChar = false;
1404  }
1405
1406  llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0);
1407
1408  // Narrow character literals act as though their value is concatenated
1409  // in this implementation, but warn on overflow.
1410  bool multi_char_too_long = false;
1411  if (isAscii() && isMultiChar()) {
1412    LitVal = 0;
1413    for (size_t i = 0; i < NumCharsSoFar; ++i) {
1414      // check for enough leading zeros to shift into
1415      multi_char_too_long |= (LitVal.countLeadingZeros() < 8);
1416      LitVal <<= 8;
1417      LitVal = LitVal + (codepoint_buffer[i] & 0xFF);
1418    }
1419  } else if (NumCharsSoFar > 0) {
1420    // otherwise just take the last character
1421    LitVal = buffer_begin[-1];
1422  }
1423
1424  if (!HadError && multi_char_too_long) {
1425    PP.Diag(Loc, diag::warn_char_constant_too_large);
1426  }
1427
1428  // Transfer the value from APInt to uint64_t
1429  Value = LitVal.getZExtValue();
1430
1431  // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1")
1432  // if 'char' is signed for this target (C99 6.4.4.4p10).  Note that multiple
1433  // character constants are not sign extended in the this implementation:
1434  // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC.
1435  if (isAscii() && NumCharsSoFar == 1 && (Value & 128) &&
1436      PP.getLangOpts().CharIsSigned)
1437    Value = (signed char)Value;
1438}
1439
1440/// \verbatim
1441///       string-literal: [C++0x lex.string]
1442///         encoding-prefix " [s-char-sequence] "
1443///         encoding-prefix R raw-string
1444///       encoding-prefix:
1445///         u8
1446///         u
1447///         U
1448///         L
1449///       s-char-sequence:
1450///         s-char
1451///         s-char-sequence s-char
1452///       s-char:
1453///         any member of the source character set except the double-quote ",
1454///           backslash \, or new-line character
1455///         escape-sequence
1456///         universal-character-name
1457///       raw-string:
1458///         " d-char-sequence ( r-char-sequence ) d-char-sequence "
1459///       r-char-sequence:
1460///         r-char
1461///         r-char-sequence r-char
1462///       r-char:
1463///         any member of the source character set, except a right parenthesis )
1464///           followed by the initial d-char-sequence (which may be empty)
1465///           followed by a double quote ".
1466///       d-char-sequence:
1467///         d-char
1468///         d-char-sequence d-char
1469///       d-char:
1470///         any member of the basic source character set except:
1471///           space, the left parenthesis (, the right parenthesis ),
1472///           the backslash \, and the control characters representing horizontal
1473///           tab, vertical tab, form feed, and newline.
1474///       escape-sequence: [C++0x lex.ccon]
1475///         simple-escape-sequence
1476///         octal-escape-sequence
1477///         hexadecimal-escape-sequence
1478///       simple-escape-sequence:
1479///         one of \' \" \? \\ \a \b \f \n \r \t \v
1480///       octal-escape-sequence:
1481///         \ octal-digit
1482///         \ octal-digit octal-digit
1483///         \ octal-digit octal-digit octal-digit
1484///       hexadecimal-escape-sequence:
1485///         \x hexadecimal-digit
1486///         hexadecimal-escape-sequence hexadecimal-digit
1487///       universal-character-name:
1488///         \u hex-quad
1489///         \U hex-quad hex-quad
1490///       hex-quad:
1491///         hex-digit hex-digit hex-digit hex-digit
1492/// \endverbatim
1493///
1494StringLiteralParser::
1495StringLiteralParser(ArrayRef<Token> StringToks,
1496                    Preprocessor &PP, bool Complain)
1497  : SM(PP.getSourceManager()), Features(PP.getLangOpts()),
1498    Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() :nullptr),
1499    MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown),
1500    ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) {
1501  init(StringToks);
1502}
1503
1504void StringLiteralParser::init(ArrayRef<Token> StringToks){
1505  // The literal token may have come from an invalid source location (e.g. due
1506  // to a PCH error), in which case the token length will be 0.
1507  if (StringToks.empty() || StringToks[0].getLength() < 2)
1508    return DiagnoseLexingError(SourceLocation());
1509
1510  // Scan all of the string portions, remember the max individual token length,
1511  // computing a bound on the concatenated string length, and see whether any
1512  // piece is a wide-string.  If any of the string portions is a wide-string
1513  // literal, the result is a wide-string literal [C99 6.4.5p4].
1514  assert(!StringToks.empty() && "expected at least one token");
1515  MaxTokenLength = StringToks[0].getLength();
1516  assert(StringToks[0].getLength() >= 2 && "literal token is invalid!");
1517  SizeBound = StringToks[0].getLength()-2;  // -2 for "".
1518  Kind = StringToks[0].getKind();
1519
1520  hadError = false;
1521
1522  // Implement Translation Phase #6: concatenation of string literals
1523  /// (C99 5.1.1.2p1).  The common case is only one string fragment.
1524  for (unsigned i = 1; i != StringToks.size(); ++i) {
1525    if (StringToks[i].getLength() < 2)
1526      return DiagnoseLexingError(StringToks[i].getLocation());
1527
1528    // The string could be shorter than this if it needs cleaning, but this is a
1529    // reasonable bound, which is all we need.
1530    assert(StringToks[i].getLength() >= 2 && "literal token is invalid!");
1531    SizeBound += StringToks[i].getLength()-2;  // -2 for "".
1532
1533    // Remember maximum string piece length.
1534    if (StringToks[i].getLength() > MaxTokenLength)
1535      MaxTokenLength = StringToks[i].getLength();
1536
1537    // Remember if we see any wide or utf-8/16/32 strings.
1538    // Also check for illegal concatenations.
1539    if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) {
1540      if (isAscii()) {
1541        Kind = StringToks[i].getKind();
1542      } else {
1543        if (Diags)
1544          Diags->Report(StringToks[i].getLocation(),
1545                        diag::err_unsupported_string_concat);
1546        hadError = true;
1547      }
1548    }
1549  }
1550
1551  // Include space for the null terminator.
1552  ++SizeBound;
1553
1554  // TODO: K&R warning: "traditional C rejects string constant concatenation"
1555
1556  // Get the width in bytes of char/wchar_t/char16_t/char32_t
1557  CharByteWidth = getCharWidth(Kind, Target);
1558  assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple");
1559  CharByteWidth /= 8;
1560
1561  // The output buffer size needs to be large enough to hold wide characters.
1562  // This is a worst-case assumption which basically corresponds to L"" "long".
1563  SizeBound *= CharByteWidth;
1564
1565  // Size the temporary buffer to hold the result string data.
1566  ResultBuf.resize(SizeBound);
1567
1568  // Likewise, but for each string piece.
1569  SmallString<512> TokenBuf;
1570  TokenBuf.resize(MaxTokenLength);
1571
1572  // Loop over all the strings, getting their spelling, and expanding them to
1573  // wide strings as appropriate.
1574  ResultPtr = &ResultBuf[0];   // Next byte to fill in.
1575
1576  Pascal = false;
1577
1578  SourceLocation UDSuffixTokLoc;
1579
1580  for (unsigned i = 0, e = StringToks.size(); i != e; ++i) {
1581    const char *ThisTokBuf = &TokenBuf[0];
1582    // Get the spelling of the token, which eliminates trigraphs, etc.  We know
1583    // that ThisTokBuf points to a buffer that is big enough for the whole token
1584    // and 'spelled' tokens can only shrink.
1585    bool StringInvalid = false;
1586    unsigned ThisTokLen =
1587      Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features,
1588                         &StringInvalid);
1589    if (StringInvalid)
1590      return DiagnoseLexingError(StringToks[i].getLocation());
1591
1592    const char *ThisTokBegin = ThisTokBuf;
1593    const char *ThisTokEnd = ThisTokBuf+ThisTokLen;
1594
1595    // Remove an optional ud-suffix.
1596    if (ThisTokEnd[-1] != '"') {
1597      const char *UDSuffixEnd = ThisTokEnd;
1598      do {
1599        --ThisTokEnd;
1600      } while (ThisTokEnd[-1] != '"');
1601
1602      StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd);
1603
1604      if (UDSuffixBuf.empty()) {
1605        if (StringToks[i].hasUCN())
1606          expandUCNs(UDSuffixBuf, UDSuffix);
1607        else
1608          UDSuffixBuf.assign(UDSuffix);
1609        UDSuffixToken = i;
1610        UDSuffixOffset = ThisTokEnd - ThisTokBuf;
1611        UDSuffixTokLoc = StringToks[i].getLocation();
1612      } else {
1613        SmallString<32> ExpandedUDSuffix;
1614        if (StringToks[i].hasUCN()) {
1615          expandUCNs(ExpandedUDSuffix, UDSuffix);
1616          UDSuffix = ExpandedUDSuffix;
1617        }
1618
1619        // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the
1620        // result of a concatenation involving at least one user-defined-string-
1621        // literal, all the participating user-defined-string-literals shall
1622        // have the same ud-suffix.
1623        if (UDSuffixBuf != UDSuffix) {
1624          if (Diags) {
1625            SourceLocation TokLoc = StringToks[i].getLocation();
1626            Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix)
1627              << UDSuffixBuf << UDSuffix
1628              << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc)
1629              << SourceRange(TokLoc, TokLoc);
1630          }
1631          hadError = true;
1632        }
1633      }
1634    }
1635
1636    // Strip the end quote.
1637    --ThisTokEnd;
1638
1639    // TODO: Input character set mapping support.
1640
1641    // Skip marker for wide or unicode strings.
1642    if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') {
1643      ++ThisTokBuf;
1644      // Skip 8 of u8 marker for utf8 strings.
1645      if (ThisTokBuf[0] == '8')
1646        ++ThisTokBuf;
1647    }
1648
1649    // Check for raw string
1650    if (ThisTokBuf[0] == 'R') {
1651      if (ThisTokBuf[1] != '"') {
1652        // The file may have come from PCH and then changed after loading the
1653        // PCH; Fail gracefully.
1654        return DiagnoseLexingError(StringToks[i].getLocation());
1655      }
1656      ThisTokBuf += 2; // skip R"
1657
1658      // C++11 [lex.string]p2: A `d-char-sequence` shall consist of at most 16
1659      // characters.
1660      constexpr unsigned MaxRawStrDelimLen = 16;
1661
1662      const char *Prefix = ThisTokBuf;
1663      while (ThisTokBuf - Prefix < MaxRawStrDelimLen && ThisTokBuf[0] != '(')
1664        ++ThisTokBuf;
1665      if (ThisTokBuf[0] != '(')
1666        return DiagnoseLexingError(StringToks[i].getLocation());
1667      ++ThisTokBuf; // skip '('
1668
1669      // Remove same number of characters from the end
1670      ThisTokEnd -= ThisTokBuf - Prefix;
1671      if (ThisTokEnd < ThisTokBuf)
1672        return DiagnoseLexingError(StringToks[i].getLocation());
1673
1674      // C++14 [lex.string]p4: A source-file new-line in a raw string literal
1675      // results in a new-line in the resulting execution string-literal.
1676      StringRef RemainingTokenSpan(ThisTokBuf, ThisTokEnd - ThisTokBuf);
1677      while (!RemainingTokenSpan.empty()) {
1678        // Split the string literal on \r\n boundaries.
1679        size_t CRLFPos = RemainingTokenSpan.find("\r\n");
1680        StringRef BeforeCRLF = RemainingTokenSpan.substr(0, CRLFPos);
1681        StringRef AfterCRLF = RemainingTokenSpan.substr(CRLFPos);
1682
1683        // Copy everything before the \r\n sequence into the string literal.
1684        if (CopyStringFragment(StringToks[i], ThisTokBegin, BeforeCRLF))
1685          hadError = true;
1686
1687        // Point into the \n inside the \r\n sequence and operate on the
1688        // remaining portion of the literal.
1689        RemainingTokenSpan = AfterCRLF.substr(1);
1690      }
1691    } else {
1692      if (ThisTokBuf[0] != '"') {
1693        // The file may have come from PCH and then changed after loading the
1694        // PCH; Fail gracefully.
1695        return DiagnoseLexingError(StringToks[i].getLocation());
1696      }
1697      ++ThisTokBuf; // skip "
1698
1699      // Check if this is a pascal string
1700      if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd &&
1701          ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') {
1702
1703        // If the \p sequence is found in the first token, we have a pascal string
1704        // Otherwise, if we already have a pascal string, ignore the first \p
1705        if (i == 0) {
1706          ++ThisTokBuf;
1707          Pascal = true;
1708        } else if (Pascal)
1709          ThisTokBuf += 2;
1710      }
1711
1712      while (ThisTokBuf != ThisTokEnd) {
1713        // Is this a span of non-escape characters?
1714        if (ThisTokBuf[0] != '\\') {
1715          const char *InStart = ThisTokBuf;
1716          do {
1717            ++ThisTokBuf;
1718          } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\');
1719
1720          // Copy the character span over.
1721          if (CopyStringFragment(StringToks[i], ThisTokBegin,
1722                                 StringRef(InStart, ThisTokBuf - InStart)))
1723            hadError = true;
1724          continue;
1725        }
1726        // Is this a Universal Character Name escape?
1727        if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') {
1728          EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd,
1729                          ResultPtr, hadError,
1730                          FullSourceLoc(StringToks[i].getLocation(), SM),
1731                          CharByteWidth, Diags, Features);
1732          continue;
1733        }
1734        // Otherwise, this is a non-UCN escape character.  Process it.
1735        unsigned ResultChar =
1736          ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError,
1737                            FullSourceLoc(StringToks[i].getLocation(), SM),
1738                            CharByteWidth*8, Diags, Features);
1739
1740        if (CharByteWidth == 4) {
1741          // FIXME: Make the type of the result buffer correct instead of
1742          // using reinterpret_cast.
1743          llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultPtr);
1744          *ResultWidePtr = ResultChar;
1745          ResultPtr += 4;
1746        } else if (CharByteWidth == 2) {
1747          // FIXME: Make the type of the result buffer correct instead of
1748          // using reinterpret_cast.
1749          llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultPtr);
1750          *ResultWidePtr = ResultChar & 0xFFFF;
1751          ResultPtr += 2;
1752        } else {
1753          assert(CharByteWidth == 1 && "Unexpected char width");
1754          *ResultPtr++ = ResultChar & 0xFF;
1755        }
1756      }
1757    }
1758  }
1759
1760  if (Pascal) {
1761    if (CharByteWidth == 4) {
1762      // FIXME: Make the type of the result buffer correct instead of
1763      // using reinterpret_cast.
1764      llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultBuf.data());
1765      ResultWidePtr[0] = GetNumStringChars() - 1;
1766    } else if (CharByteWidth == 2) {
1767      // FIXME: Make the type of the result buffer correct instead of
1768      // using reinterpret_cast.
1769      llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultBuf.data());
1770      ResultWidePtr[0] = GetNumStringChars() - 1;
1771    } else {
1772      assert(CharByteWidth == 1 && "Unexpected char width");
1773      ResultBuf[0] = GetNumStringChars() - 1;
1774    }
1775
1776    // Verify that pascal strings aren't too large.
1777    if (GetStringLength() > 256) {
1778      if (Diags)
1779        Diags->Report(StringToks.front().getLocation(),
1780                      diag::err_pascal_string_too_long)
1781          << SourceRange(StringToks.front().getLocation(),
1782                         StringToks.back().getLocation());
1783      hadError = true;
1784      return;
1785    }
1786  } else if (Diags) {
1787    // Complain if this string literal has too many characters.
1788    unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509;
1789
1790    if (GetNumStringChars() > MaxChars)
1791      Diags->Report(StringToks.front().getLocation(),
1792                    diag::ext_string_too_long)
1793        << GetNumStringChars() << MaxChars
1794        << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0)
1795        << SourceRange(StringToks.front().getLocation(),
1796                       StringToks.back().getLocation());
1797  }
1798}
1799
1800static const char *resyncUTF8(const char *Err, const char *End) {
1801  if (Err == End)
1802    return End;
1803  End = Err + std::min<unsigned>(llvm::getNumBytesForUTF8(*Err), End-Err);
1804  while (++Err != End && (*Err & 0xC0) == 0x80)
1805    ;
1806  return Err;
1807}
1808
1809/// This function copies from Fragment, which is a sequence of bytes
1810/// within Tok's contents (which begin at TokBegin) into ResultPtr.
1811/// Performs widening for multi-byte characters.
1812bool StringLiteralParser::CopyStringFragment(const Token &Tok,
1813                                             const char *TokBegin,
1814                                             StringRef Fragment) {
1815  const llvm::UTF8 *ErrorPtrTmp;
1816  if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp))
1817    return false;
1818
1819  // If we see bad encoding for unprefixed string literals, warn and
1820  // simply copy the byte values, for compatibility with gcc and older
1821  // versions of clang.
1822  bool NoErrorOnBadEncoding = isAscii();
1823  if (NoErrorOnBadEncoding) {
1824    memcpy(ResultPtr, Fragment.data(), Fragment.size());
1825    ResultPtr += Fragment.size();
1826  }
1827
1828  if (Diags) {
1829    const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1830
1831    FullSourceLoc SourceLoc(Tok.getLocation(), SM);
1832    const DiagnosticBuilder &Builder =
1833      Diag(Diags, Features, SourceLoc, TokBegin,
1834           ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()),
1835           NoErrorOnBadEncoding ? diag::warn_bad_string_encoding
1836                                : diag::err_bad_string_encoding);
1837
1838    const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1839    StringRef NextFragment(NextStart, Fragment.end()-NextStart);
1840
1841    // Decode into a dummy buffer.
1842    SmallString<512> Dummy;
1843    Dummy.reserve(Fragment.size() * CharByteWidth);
1844    char *Ptr = Dummy.data();
1845
1846    while (!ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) {
1847      const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1848      NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1849      Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin,
1850                                     ErrorPtr, NextStart);
1851      NextFragment = StringRef(NextStart, Fragment.end()-NextStart);
1852    }
1853  }
1854  return !NoErrorOnBadEncoding;
1855}
1856
1857void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) {
1858  hadError = true;
1859  if (Diags)
1860    Diags->Report(Loc, diag::err_lexing_string);
1861}
1862
1863/// getOffsetOfStringByte - This function returns the offset of the
1864/// specified byte of the string data represented by Token.  This handles
1865/// advancing over escape sequences in the string.
1866unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok,
1867                                                    unsigned ByteNo) const {
1868  // Get the spelling of the token.
1869  SmallString<32> SpellingBuffer;
1870  SpellingBuffer.resize(Tok.getLength());
1871
1872  bool StringInvalid = false;
1873  const char *SpellingPtr = &SpellingBuffer[0];
1874  unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features,
1875                                       &StringInvalid);
1876  if (StringInvalid)
1877    return 0;
1878
1879  const char *SpellingStart = SpellingPtr;
1880  const char *SpellingEnd = SpellingPtr+TokLen;
1881
1882  // Handle UTF-8 strings just like narrow strings.
1883  if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8')
1884    SpellingPtr += 2;
1885
1886  assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' &&
1887         SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet");
1888
1889  // For raw string literals, this is easy.
1890  if (SpellingPtr[0] == 'R') {
1891    assert(SpellingPtr[1] == '"' && "Should be a raw string literal!");
1892    // Skip 'R"'.
1893    SpellingPtr += 2;
1894    while (*SpellingPtr != '(') {
1895      ++SpellingPtr;
1896      assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal");
1897    }
1898    // Skip '('.
1899    ++SpellingPtr;
1900    return SpellingPtr - SpellingStart + ByteNo;
1901  }
1902
1903  // Skip over the leading quote
1904  assert(SpellingPtr[0] == '"' && "Should be a string literal!");
1905  ++SpellingPtr;
1906
1907  // Skip over bytes until we find the offset we're looking for.
1908  while (ByteNo) {
1909    assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!");
1910
1911    // Step over non-escapes simply.
1912    if (*SpellingPtr != '\\') {
1913      ++SpellingPtr;
1914      --ByteNo;
1915      continue;
1916    }
1917
1918    // Otherwise, this is an escape character.  Advance over it.
1919    bool HadError = false;
1920    if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') {
1921      const char *EscapePtr = SpellingPtr;
1922      unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd,
1923                                      1, Features, HadError);
1924      if (Len > ByteNo) {
1925        // ByteNo is somewhere within the escape sequence.
1926        SpellingPtr = EscapePtr;
1927        break;
1928      }
1929      ByteNo -= Len;
1930    } else {
1931      ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError,
1932                        FullSourceLoc(Tok.getLocation(), SM),
1933                        CharByteWidth*8, Diags, Features);
1934      --ByteNo;
1935    }
1936    assert(!HadError && "This method isn't valid on erroneous strings");
1937  }
1938
1939  return SpellingPtr-SpellingStart;
1940}
1941
1942/// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
1943/// suffixes as ud-suffixes, because the diagnostic experience is better if we
1944/// treat it as an invalid suffix.
1945bool StringLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
1946                                          StringRef Suffix) {
1947  return NumericLiteralParser::isValidUDSuffix(LangOpts, Suffix) ||
1948         Suffix == "sv";
1949}
1950