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
37 using namespace clang;
38
getCharWidth(tok::TokenKind kind,const TargetInfo & Target)39 static 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
MakeCharSourceRange(const LangOptions & Features,FullSourceLoc TokLoc,const char * TokBegin,const char * TokRangeBegin,const char * TokRangeEnd)59 static 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.
Diag(DiagnosticsEngine * Diags,const LangOptions & Features,FullSourceLoc TokLoc,const char * TokBegin,const char * TokRangeBegin,const char * TokRangeEnd,unsigned DiagID)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);
87 }
88
89 /// ProcessCharEscape - Parse a standard C escape sequence, which can occur in
90 /// either a character or a string literal.
ProcessCharEscape(const char * ThisTokBegin,const char * & ThisTokBuf,const char * ThisTokEnd,bool & HadError,FullSourceLoc Loc,unsigned CharWidth,DiagnosticsEngine * Diags,const LangOptions & Features)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;
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
appendCodePoint(unsigned Codepoint,llvm::SmallVectorImpl<char> & Str)230 static 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
expandUCNs(SmallVectorImpl<char> & Buf,StringRef Input)240 void 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.
ProcessUCNEscape(const char * ThisTokBegin,const char * & ThisTokBuf,const char * ThisTokEnd,uint32_t & UcnVal,unsigned short & UcnLen,FullSourceLoc Loc,DiagnosticsEngine * Diags,const LangOptions & Features,bool in_char_string_literal=false)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;
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.
MeasureUCNEscape(const char * ThisTokBegin,const char * & ThisTokBuf,const char * ThisTokEnd,unsigned CharByteWidth,const LangOptions & Features,bool & HadError)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)
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.
EncodeUCNEscape(const char * ThisTokBegin,const char * & ThisTokBuf,const char * ThisTokEnd,char * & ResultBuf,bool & HadError,FullSourceLoc Loc,unsigned CharByteWidth,DiagnosticsEngine * Diags,const LangOptions & Features)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;
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 ///
NumericLiteralParser(StringRef TokSpelling,SourceLocation TokLoc,const SourceManager & SM,const LangOptions & LangOpts,const TargetInfo & Target,DiagnosticsEngine & Diags)526 NumericLiteralParser::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 isHalf = false;
550 isFloat = false;
551 isImaginary = false;
552 isFloat16 = false;
553 isFloat128 = false;
554 MicrosoftInteger = 0;
555 isFract = false;
556 isAccum = false;
557 hadError = false;
558
559 if (*s == '0') { // parse radix
560 ParseNumberStartingWithZero(TokLoc);
561 if (hadError)
562 return;
563 } else { // the first digit is non-zero
564 radix = 10;
565 s = SkipDigits(s);
566 if (s == ThisTokEnd) {
567 // Done.
568 } else {
569 ParseDecimalOrOctalCommon(TokLoc);
570 if (hadError)
571 return;
572 }
573 }
574
575 SuffixBegin = s;
576 checkSeparator(TokLoc, s, CSK_AfterDigits);
577
578 // Initial scan to lookahead for fixed point suffix.
579 if (LangOpts.FixedPoint) {
580 for (const char *c = s; c != ThisTokEnd; ++c) {
581 if (*c == 'r' || *c == 'k' || *c == 'R' || *c == 'K') {
582 saw_fixed_point_suffix = true;
583 break;
584 }
585 }
586 }
587
588 // Parse the suffix. At this point we can classify whether we have an FP or
589 // integer constant.
590 bool isFixedPointConstant = isFixedPointLiteral();
591 bool isFPConstant = isFloatingLiteral();
592
593 // Loop over all of the characters of the suffix. If we see something bad,
594 // we break out of the loop.
595 for (; s != ThisTokEnd; ++s) {
596 switch (*s) {
597 case 'R':
598 case 'r':
599 if (!LangOpts.FixedPoint)
600 break;
601 if (isFract || isAccum) break;
602 if (!(saw_period || saw_exponent)) break;
603 isFract = true;
604 continue;
605 case 'K':
606 case 'k':
607 if (!LangOpts.FixedPoint)
608 break;
609 if (isFract || isAccum) break;
610 if (!(saw_period || saw_exponent)) break;
611 isAccum = true;
612 continue;
613 case 'h': // FP Suffix for "half".
614 case 'H':
615 // OpenCL Extension v1.2 s9.5 - h or H suffix for half type.
616 if (!(LangOpts.Half || LangOpts.FixedPoint))
617 break;
618 if (isIntegerLiteral()) break; // Error for integer constant.
619 if (isHalf || isFloat || isLong) break; // HH, FH, LH invalid.
620 isHalf = true;
621 continue; // Success.
622 case 'f': // FP Suffix for "float"
623 case 'F':
624 if (!isFPConstant) break; // Error for integer constant.
625 if (isHalf || isFloat || isLong || isFloat128)
626 break; // HF, FF, LF, QF invalid.
627
628 // CUDA host and device may have different _Float16 support, therefore
629 // allows f16 literals to avoid false alarm.
630 // ToDo: more precise check for CUDA.
631 if ((Target.hasFloat16Type() || LangOpts.CUDA) && s + 2 < ThisTokEnd &&
632 s[1] == '1' && s[2] == '6') {
633 s += 2; // success, eat up 2 characters.
634 isFloat16 = true;
635 continue;
636 }
637
638 isFloat = true;
639 continue; // Success.
640 case 'q': // FP Suffix for "__float128"
641 case 'Q':
642 if (!isFPConstant) break; // Error for integer constant.
643 if (isHalf || isFloat || isLong || isFloat128)
644 break; // HQ, FQ, LQ, QQ invalid.
645 isFloat128 = true;
646 continue; // Success.
647 case 'u':
648 case 'U':
649 if (isFPConstant) break; // Error for floating constant.
650 if (isUnsigned) break; // Cannot be repeated.
651 isUnsigned = true;
652 continue; // Success.
653 case 'l':
654 case 'L':
655 if (isLong || isLongLong) break; // Cannot be repeated.
656 if (isHalf || isFloat || isFloat128) break; // LH, LF, LQ invalid.
657
658 // Check for long long. The L's need to be adjacent and the same case.
659 if (s[1] == s[0]) {
660 assert(s + 1 < ThisTokEnd && "didn't maximally munch?");
661 if (isFPConstant) break; // long long invalid for floats.
662 isLongLong = true;
663 ++s; // Eat both of them.
664 } else {
665 isLong = true;
666 }
667 continue; // Success.
668 case 'i':
669 case 'I':
670 if (LangOpts.MicrosoftExt) {
671 if (isLong || isLongLong || MicrosoftInteger)
672 break;
673
674 if (!isFPConstant) {
675 // Allow i8, i16, i32, and i64.
676 switch (s[1]) {
677 case '8':
678 s += 2; // i8 suffix
679 MicrosoftInteger = 8;
680 break;
681 case '1':
682 if (s[2] == '6') {
683 s += 3; // i16 suffix
684 MicrosoftInteger = 16;
685 }
686 break;
687 case '3':
688 if (s[2] == '2') {
689 s += 3; // i32 suffix
690 MicrosoftInteger = 32;
691 }
692 break;
693 case '6':
694 if (s[2] == '4') {
695 s += 3; // i64 suffix
696 MicrosoftInteger = 64;
697 }
698 break;
699 default:
700 break;
701 }
702 }
703 if (MicrosoftInteger) {
704 assert(s <= ThisTokEnd && "didn't maximally munch?");
705 break;
706 }
707 }
708 LLVM_FALLTHROUGH;
709 case 'j':
710 case 'J':
711 if (isImaginary) break; // Cannot be repeated.
712 isImaginary = true;
713 continue; // Success.
714 }
715 // If we reached here, there was an error or a ud-suffix.
716 break;
717 }
718
719 // "i", "if", and "il" are user-defined suffixes in C++1y.
720 if (s != ThisTokEnd || isImaginary) {
721 // FIXME: Don't bother expanding UCNs if !tok.hasUCN().
722 expandUCNs(UDSuffixBuf, StringRef(SuffixBegin, ThisTokEnd - SuffixBegin));
723 if (isValidUDSuffix(LangOpts, UDSuffixBuf)) {
724 if (!isImaginary) {
725 // Any suffix pieces we might have parsed are actually part of the
726 // ud-suffix.
727 isLong = false;
728 isUnsigned = false;
729 isLongLong = false;
730 isFloat = false;
731 isFloat16 = false;
732 isHalf = false;
733 isImaginary = false;
734 MicrosoftInteger = 0;
735 saw_fixed_point_suffix = false;
736 isFract = false;
737 isAccum = false;
738 }
739
740 saw_ud_suffix = true;
741 return;
742 }
743
744 if (s != ThisTokEnd) {
745 // Report an error if there are any.
746 Diags.Report(Lexer::AdvanceToTokenCharacter(
747 TokLoc, SuffixBegin - ThisTokBegin, SM, LangOpts),
748 diag::err_invalid_suffix_constant)
749 << StringRef(SuffixBegin, ThisTokEnd - SuffixBegin)
750 << (isFixedPointConstant ? 2 : isFPConstant);
751 hadError = true;
752 }
753 }
754
755 if (!hadError && saw_fixed_point_suffix) {
756 assert(isFract || isAccum);
757 }
758 }
759
760 /// ParseDecimalOrOctalCommon - This method is called for decimal or octal
761 /// numbers. It issues an error for illegal digits, and handles floating point
762 /// parsing. If it detects a floating point number, the radix is set to 10.
ParseDecimalOrOctalCommon(SourceLocation TokLoc)763 void NumericLiteralParser::ParseDecimalOrOctalCommon(SourceLocation TokLoc){
764 assert((radix == 8 || radix == 10) && "Unexpected radix");
765
766 // If we have a hex digit other than 'e' (which denotes a FP exponent) then
767 // the code is using an incorrect base.
768 if (isHexDigit(*s) && *s != 'e' && *s != 'E' &&
769 !isValidUDSuffix(LangOpts, StringRef(s, ThisTokEnd - s))) {
770 Diags.Report(
771 Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM, LangOpts),
772 diag::err_invalid_digit)
773 << StringRef(s, 1) << (radix == 8 ? 1 : 0);
774 hadError = true;
775 return;
776 }
777
778 if (*s == '.') {
779 checkSeparator(TokLoc, s, CSK_AfterDigits);
780 s++;
781 radix = 10;
782 saw_period = true;
783 checkSeparator(TokLoc, s, CSK_BeforeDigits);
784 s = SkipDigits(s); // Skip suffix.
785 }
786 if (*s == 'e' || *s == 'E') { // exponent
787 checkSeparator(TokLoc, s, CSK_AfterDigits);
788 const char *Exponent = s;
789 s++;
790 radix = 10;
791 saw_exponent = true;
792 if (s != ThisTokEnd && (*s == '+' || *s == '-')) s++; // sign
793 const char *first_non_digit = SkipDigits(s);
794 if (containsDigits(s, first_non_digit)) {
795 checkSeparator(TokLoc, s, CSK_BeforeDigits);
796 s = first_non_digit;
797 } else {
798 if (!hadError) {
799 Diags.Report(Lexer::AdvanceToTokenCharacter(
800 TokLoc, Exponent - ThisTokBegin, SM, LangOpts),
801 diag::err_exponent_has_no_digits);
802 hadError = true;
803 }
804 return;
805 }
806 }
807 }
808
809 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
810 /// suffixes as ud-suffixes, because the diagnostic experience is better if we
811 /// treat it as an invalid suffix.
isValidUDSuffix(const LangOptions & LangOpts,StringRef Suffix)812 bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
813 StringRef Suffix) {
814 if (!LangOpts.CPlusPlus11 || Suffix.empty())
815 return false;
816
817 // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid.
818 if (Suffix[0] == '_')
819 return true;
820
821 // In C++11, there are no library suffixes.
822 if (!LangOpts.CPlusPlus14)
823 return false;
824
825 // In C++14, "s", "h", "min", "ms", "us", and "ns" are used in the library.
826 // Per tweaked N3660, "il", "i", and "if" are also used in the library.
827 // In C++2a "d" and "y" are used in the library.
828 return llvm::StringSwitch<bool>(Suffix)
829 .Cases("h", "min", "s", true)
830 .Cases("ms", "us", "ns", true)
831 .Cases("il", "i", "if", true)
832 .Cases("d", "y", LangOpts.CPlusPlus20)
833 .Default(false);
834 }
835
checkSeparator(SourceLocation TokLoc,const char * Pos,CheckSeparatorKind IsAfterDigits)836 void NumericLiteralParser::checkSeparator(SourceLocation TokLoc,
837 const char *Pos,
838 CheckSeparatorKind IsAfterDigits) {
839 if (IsAfterDigits == CSK_AfterDigits) {
840 if (Pos == ThisTokBegin)
841 return;
842 --Pos;
843 } else if (Pos == ThisTokEnd)
844 return;
845
846 if (isDigitSeparator(*Pos)) {
847 Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin, SM,
848 LangOpts),
849 diag::err_digit_separator_not_between_digits)
850 << IsAfterDigits;
851 hadError = true;
852 }
853 }
854
855 /// ParseNumberStartingWithZero - This method is called when the first character
856 /// of the number is found to be a zero. This means it is either an octal
857 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or
858 /// a floating point number (01239.123e4). Eat the prefix, determining the
859 /// radix etc.
ParseNumberStartingWithZero(SourceLocation TokLoc)860 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
861 assert(s[0] == '0' && "Invalid method call");
862 s++;
863
864 int c1 = s[0];
865
866 // Handle a hex number like 0x1234.
867 if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) {
868 s++;
869 assert(s < ThisTokEnd && "didn't maximally munch?");
870 radix = 16;
871 DigitsBegin = s;
872 s = SkipHexDigits(s);
873 bool HasSignificandDigits = containsDigits(DigitsBegin, s);
874 if (s == ThisTokEnd) {
875 // Done.
876 } else if (*s == '.') {
877 s++;
878 saw_period = true;
879 const char *floatDigitsBegin = s;
880 s = SkipHexDigits(s);
881 if (containsDigits(floatDigitsBegin, s))
882 HasSignificandDigits = true;
883 if (HasSignificandDigits)
884 checkSeparator(TokLoc, floatDigitsBegin, CSK_BeforeDigits);
885 }
886
887 if (!HasSignificandDigits) {
888 Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM,
889 LangOpts),
890 diag::err_hex_constant_requires)
891 << LangOpts.CPlusPlus << 1;
892 hadError = true;
893 return;
894 }
895
896 // A binary exponent can appear with or with a '.'. If dotted, the
897 // binary exponent is required.
898 if (*s == 'p' || *s == 'P') {
899 checkSeparator(TokLoc, s, CSK_AfterDigits);
900 const char *Exponent = s;
901 s++;
902 saw_exponent = true;
903 if (s != ThisTokEnd && (*s == '+' || *s == '-')) s++; // sign
904 const char *first_non_digit = SkipDigits(s);
905 if (!containsDigits(s, first_non_digit)) {
906 if (!hadError) {
907 Diags.Report(Lexer::AdvanceToTokenCharacter(
908 TokLoc, Exponent - ThisTokBegin, SM, LangOpts),
909 diag::err_exponent_has_no_digits);
910 hadError = true;
911 }
912 return;
913 }
914 checkSeparator(TokLoc, s, CSK_BeforeDigits);
915 s = first_non_digit;
916
917 if (!LangOpts.HexFloats)
918 Diags.Report(TokLoc, LangOpts.CPlusPlus
919 ? diag::ext_hex_literal_invalid
920 : diag::ext_hex_constant_invalid);
921 else if (LangOpts.CPlusPlus17)
922 Diags.Report(TokLoc, diag::warn_cxx17_hex_literal);
923 } else if (saw_period) {
924 Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM,
925 LangOpts),
926 diag::err_hex_constant_requires)
927 << LangOpts.CPlusPlus << 0;
928 hadError = true;
929 }
930 return;
931 }
932
933 // Handle simple binary numbers 0b01010
934 if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) {
935 // 0b101010 is a C++1y / GCC extension.
936 Diags.Report(TokLoc, LangOpts.CPlusPlus14
937 ? diag::warn_cxx11_compat_binary_literal
938 : LangOpts.CPlusPlus ? diag::ext_binary_literal_cxx14
939 : diag::ext_binary_literal);
940 ++s;
941 assert(s < ThisTokEnd && "didn't maximally munch?");
942 radix = 2;
943 DigitsBegin = s;
944 s = SkipBinaryDigits(s);
945 if (s == ThisTokEnd) {
946 // Done.
947 } else if (isHexDigit(*s) &&
948 !isValidUDSuffix(LangOpts, StringRef(s, ThisTokEnd - s))) {
949 Diags.Report(Lexer::AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin, SM,
950 LangOpts),
951 diag::err_invalid_digit)
952 << StringRef(s, 1) << 2;
953 hadError = true;
954 }
955 // Other suffixes will be diagnosed by the caller.
956 return;
957 }
958
959 // For now, the radix is set to 8. If we discover that we have a
960 // floating point constant, the radix will change to 10. Octal floating
961 // point constants are not permitted (only decimal and hexadecimal).
962 radix = 8;
963 DigitsBegin = s;
964 s = SkipOctalDigits(s);
965 if (s == ThisTokEnd)
966 return; // Done, simple octal number like 01234
967
968 // If we have some other non-octal digit that *is* a decimal digit, see if
969 // this is part of a floating point number like 094.123 or 09e1.
970 if (isDigit(*s)) {
971 const char *EndDecimal = SkipDigits(s);
972 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') {
973 s = EndDecimal;
974 radix = 10;
975 }
976 }
977
978 ParseDecimalOrOctalCommon(TokLoc);
979 }
980
alwaysFitsInto64Bits(unsigned Radix,unsigned NumDigits)981 static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) {
982 switch (Radix) {
983 case 2:
984 return NumDigits <= 64;
985 case 8:
986 return NumDigits <= 64 / 3; // Digits are groups of 3 bits.
987 case 10:
988 return NumDigits <= 19; // floor(log10(2^64))
989 case 16:
990 return NumDigits <= 64 / 4; // Digits are groups of 4 bits.
991 default:
992 llvm_unreachable("impossible Radix");
993 }
994 }
995
996 /// GetIntegerValue - Convert this numeric literal value to an APInt that
997 /// matches Val's input width. If there is an overflow, set Val to the low bits
998 /// of the result and return true. Otherwise, return false.
GetIntegerValue(llvm::APInt & Val)999 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) {
1000 // Fast path: Compute a conservative bound on the maximum number of
1001 // bits per digit in this radix. If we can't possibly overflow a
1002 // uint64 based on that bound then do the simple conversion to
1003 // integer. This avoids the expensive overflow checking below, and
1004 // handles the common cases that matter (small decimal integers and
1005 // hex/octal values which don't overflow).
1006 const unsigned NumDigits = SuffixBegin - DigitsBegin;
1007 if (alwaysFitsInto64Bits(radix, NumDigits)) {
1008 uint64_t N = 0;
1009 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr)
1010 if (!isDigitSeparator(*Ptr))
1011 N = N * radix + llvm::hexDigitValue(*Ptr);
1012
1013 // This will truncate the value to Val's input width. Simply check
1014 // for overflow by comparing.
1015 Val = N;
1016 return Val.getZExtValue() != N;
1017 }
1018
1019 Val = 0;
1020 const char *Ptr = DigitsBegin;
1021
1022 llvm::APInt RadixVal(Val.getBitWidth(), radix);
1023 llvm::APInt CharVal(Val.getBitWidth(), 0);
1024 llvm::APInt OldVal = Val;
1025
1026 bool OverflowOccurred = false;
1027 while (Ptr < SuffixBegin) {
1028 if (isDigitSeparator(*Ptr)) {
1029 ++Ptr;
1030 continue;
1031 }
1032
1033 unsigned C = llvm::hexDigitValue(*Ptr++);
1034
1035 // If this letter is out of bound for this radix, reject it.
1036 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1037
1038 CharVal = C;
1039
1040 // Add the digit to the value in the appropriate radix. If adding in digits
1041 // made the value smaller, then this overflowed.
1042 OldVal = Val;
1043
1044 // Multiply by radix, did overflow occur on the multiply?
1045 Val *= RadixVal;
1046 OverflowOccurred |= Val.udiv(RadixVal) != OldVal;
1047
1048 // Add value, did overflow occur on the value?
1049 // (a + b) ult b <=> overflow
1050 Val += CharVal;
1051 OverflowOccurred |= Val.ult(CharVal);
1052 }
1053 return OverflowOccurred;
1054 }
1055
1056 llvm::APFloat::opStatus
GetFloatValue(llvm::APFloat & Result)1057 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) {
1058 using llvm::APFloat;
1059
1060 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin);
1061
1062 llvm::SmallString<16> Buffer;
1063 StringRef Str(ThisTokBegin, n);
1064 if (Str.find('\'') != StringRef::npos) {
1065 Buffer.reserve(n);
1066 std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer),
1067 &isDigitSeparator);
1068 Str = Buffer;
1069 }
1070
1071 auto StatusOrErr =
1072 Result.convertFromString(Str, APFloat::rmNearestTiesToEven);
1073 assert(StatusOrErr && "Invalid floating point representation");
1074 return !errorToBool(StatusOrErr.takeError()) ? *StatusOrErr
1075 : APFloat::opInvalidOp;
1076 }
1077
IsExponentPart(char c)1078 static inline bool IsExponentPart(char c) {
1079 return c == 'p' || c == 'P' || c == 'e' || c == 'E';
1080 }
1081
GetFixedPointValue(llvm::APInt & StoreVal,unsigned Scale)1082 bool NumericLiteralParser::GetFixedPointValue(llvm::APInt &StoreVal, unsigned Scale) {
1083 assert(radix == 16 || radix == 10);
1084
1085 // Find how many digits are needed to store the whole literal.
1086 unsigned NumDigits = SuffixBegin - DigitsBegin;
1087 if (saw_period) --NumDigits;
1088
1089 // Initial scan of the exponent if it exists
1090 bool ExpOverflowOccurred = false;
1091 bool NegativeExponent = false;
1092 const char *ExponentBegin;
1093 uint64_t Exponent = 0;
1094 int64_t BaseShift = 0;
1095 if (saw_exponent) {
1096 const char *Ptr = DigitsBegin;
1097
1098 while (!IsExponentPart(*Ptr)) ++Ptr;
1099 ExponentBegin = Ptr;
1100 ++Ptr;
1101 NegativeExponent = *Ptr == '-';
1102 if (NegativeExponent) ++Ptr;
1103
1104 unsigned NumExpDigits = SuffixBegin - Ptr;
1105 if (alwaysFitsInto64Bits(radix, NumExpDigits)) {
1106 llvm::StringRef ExpStr(Ptr, NumExpDigits);
1107 llvm::APInt ExpInt(/*numBits=*/64, ExpStr, /*radix=*/10);
1108 Exponent = ExpInt.getZExtValue();
1109 } else {
1110 ExpOverflowOccurred = true;
1111 }
1112
1113 if (NegativeExponent) BaseShift -= Exponent;
1114 else BaseShift += Exponent;
1115 }
1116
1117 // Number of bits needed for decimal literal is
1118 // ceil(NumDigits * log2(10)) Integral part
1119 // + Scale Fractional part
1120 // + ceil(Exponent * log2(10)) Exponent
1121 // --------------------------------------------------
1122 // ceil((NumDigits + Exponent) * log2(10)) + Scale
1123 //
1124 // But for simplicity in handling integers, we can round up log2(10) to 4,
1125 // making:
1126 // 4 * (NumDigits + Exponent) + Scale
1127 //
1128 // Number of digits needed for hexadecimal literal is
1129 // 4 * NumDigits Integral part
1130 // + Scale Fractional part
1131 // + Exponent Exponent
1132 // --------------------------------------------------
1133 // (4 * NumDigits) + Scale + Exponent
1134 uint64_t NumBitsNeeded;
1135 if (radix == 10)
1136 NumBitsNeeded = 4 * (NumDigits + Exponent) + Scale;
1137 else
1138 NumBitsNeeded = 4 * NumDigits + Exponent + Scale;
1139
1140 if (NumBitsNeeded > std::numeric_limits<unsigned>::max())
1141 ExpOverflowOccurred = true;
1142 llvm::APInt Val(static_cast<unsigned>(NumBitsNeeded), 0, /*isSigned=*/false);
1143
1144 bool FoundDecimal = false;
1145
1146 int64_t FractBaseShift = 0;
1147 const char *End = saw_exponent ? ExponentBegin : SuffixBegin;
1148 for (const char *Ptr = DigitsBegin; Ptr < End; ++Ptr) {
1149 if (*Ptr == '.') {
1150 FoundDecimal = true;
1151 continue;
1152 }
1153
1154 // Normal reading of an integer
1155 unsigned C = llvm::hexDigitValue(*Ptr);
1156 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1157
1158 Val *= radix;
1159 Val += C;
1160
1161 if (FoundDecimal)
1162 // Keep track of how much we will need to adjust this value by from the
1163 // number of digits past the radix point.
1164 --FractBaseShift;
1165 }
1166
1167 // For a radix of 16, we will be multiplying by 2 instead of 16.
1168 if (radix == 16) FractBaseShift *= 4;
1169 BaseShift += FractBaseShift;
1170
1171 Val <<= Scale;
1172
1173 uint64_t Base = (radix == 16) ? 2 : 10;
1174 if (BaseShift > 0) {
1175 for (int64_t i = 0; i < BaseShift; ++i) {
1176 Val *= Base;
1177 }
1178 } else if (BaseShift < 0) {
1179 for (int64_t i = BaseShift; i < 0 && !Val.isNullValue(); ++i)
1180 Val = Val.udiv(Base);
1181 }
1182
1183 bool IntOverflowOccurred = false;
1184 auto MaxVal = llvm::APInt::getMaxValue(StoreVal.getBitWidth());
1185 if (Val.getBitWidth() > StoreVal.getBitWidth()) {
1186 IntOverflowOccurred |= Val.ugt(MaxVal.zext(Val.getBitWidth()));
1187 StoreVal = Val.trunc(StoreVal.getBitWidth());
1188 } else if (Val.getBitWidth() < StoreVal.getBitWidth()) {
1189 IntOverflowOccurred |= Val.zext(MaxVal.getBitWidth()).ugt(MaxVal);
1190 StoreVal = Val.zext(StoreVal.getBitWidth());
1191 } else {
1192 StoreVal = Val;
1193 }
1194
1195 return IntOverflowOccurred || ExpOverflowOccurred;
1196 }
1197
1198 /// \verbatim
1199 /// user-defined-character-literal: [C++11 lex.ext]
1200 /// character-literal ud-suffix
1201 /// ud-suffix:
1202 /// identifier
1203 /// character-literal: [C++11 lex.ccon]
1204 /// ' c-char-sequence '
1205 /// u' c-char-sequence '
1206 /// U' c-char-sequence '
1207 /// L' c-char-sequence '
1208 /// u8' c-char-sequence ' [C++1z lex.ccon]
1209 /// c-char-sequence:
1210 /// c-char
1211 /// c-char-sequence c-char
1212 /// c-char:
1213 /// any member of the source character set except the single-quote ',
1214 /// backslash \, or new-line character
1215 /// escape-sequence
1216 /// universal-character-name
1217 /// escape-sequence:
1218 /// simple-escape-sequence
1219 /// octal-escape-sequence
1220 /// hexadecimal-escape-sequence
1221 /// simple-escape-sequence:
1222 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1223 /// octal-escape-sequence:
1224 /// \ octal-digit
1225 /// \ octal-digit octal-digit
1226 /// \ octal-digit octal-digit octal-digit
1227 /// hexadecimal-escape-sequence:
1228 /// \x hexadecimal-digit
1229 /// hexadecimal-escape-sequence hexadecimal-digit
1230 /// universal-character-name: [C++11 lex.charset]
1231 /// \u hex-quad
1232 /// \U hex-quad hex-quad
1233 /// hex-quad:
1234 /// hex-digit hex-digit hex-digit hex-digit
1235 /// \endverbatim
1236 ///
CharLiteralParser(const char * begin,const char * end,SourceLocation Loc,Preprocessor & PP,tok::TokenKind kind)1237 CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
1238 SourceLocation Loc, Preprocessor &PP,
1239 tok::TokenKind kind) {
1240 // At this point we know that the character matches the regex "(L|u|U)?'.*'".
1241 HadError = false;
1242
1243 Kind = kind;
1244
1245 const char *TokBegin = begin;
1246
1247 // Skip over wide character determinant.
1248 if (Kind != tok::char_constant)
1249 ++begin;
1250 if (Kind == tok::utf8_char_constant)
1251 ++begin;
1252
1253 // Skip over the entry quote.
1254 assert(begin[0] == '\'' && "Invalid token lexed");
1255 ++begin;
1256
1257 // Remove an optional ud-suffix.
1258 if (end[-1] != '\'') {
1259 const char *UDSuffixEnd = end;
1260 do {
1261 --end;
1262 } while (end[-1] != '\'');
1263 // FIXME: Don't bother with this if !tok.hasUCN().
1264 expandUCNs(UDSuffixBuf, StringRef(end, UDSuffixEnd - end));
1265 UDSuffixOffset = end - TokBegin;
1266 }
1267
1268 // Trim the ending quote.
1269 assert(end != begin && "Invalid token lexed");
1270 --end;
1271
1272 // FIXME: The "Value" is an uint64_t so we can handle char literals of
1273 // up to 64-bits.
1274 // FIXME: This extensively assumes that 'char' is 8-bits.
1275 assert(PP.getTargetInfo().getCharWidth() == 8 &&
1276 "Assumes char is 8 bits");
1277 assert(PP.getTargetInfo().getIntWidth() <= 64 &&
1278 (PP.getTargetInfo().getIntWidth() & 7) == 0 &&
1279 "Assumes sizeof(int) on target is <= 64 and a multiple of char");
1280 assert(PP.getTargetInfo().getWCharWidth() <= 64 &&
1281 "Assumes sizeof(wchar) on target is <= 64");
1282
1283 SmallVector<uint32_t, 4> codepoint_buffer;
1284 codepoint_buffer.resize(end - begin);
1285 uint32_t *buffer_begin = &codepoint_buffer.front();
1286 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size();
1287
1288 // Unicode escapes representing characters that cannot be correctly
1289 // represented in a single code unit are disallowed in character literals
1290 // by this implementation.
1291 uint32_t largest_character_for_kind;
1292 if (tok::wide_char_constant == Kind) {
1293 largest_character_for_kind =
1294 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth());
1295 } else if (tok::utf8_char_constant == Kind) {
1296 largest_character_for_kind = 0x7F;
1297 } else if (tok::utf16_char_constant == Kind) {
1298 largest_character_for_kind = 0xFFFF;
1299 } else if (tok::utf32_char_constant == Kind) {
1300 largest_character_for_kind = 0x10FFFF;
1301 } else {
1302 largest_character_for_kind = 0x7Fu;
1303 }
1304
1305 while (begin != end) {
1306 // Is this a span of non-escape characters?
1307 if (begin[0] != '\\') {
1308 char const *start = begin;
1309 do {
1310 ++begin;
1311 } while (begin != end && *begin != '\\');
1312
1313 char const *tmp_in_start = start;
1314 uint32_t *tmp_out_start = buffer_begin;
1315 llvm::ConversionResult res =
1316 llvm::ConvertUTF8toUTF32(reinterpret_cast<llvm::UTF8 const **>(&start),
1317 reinterpret_cast<llvm::UTF8 const *>(begin),
1318 &buffer_begin, buffer_end, llvm::strictConversion);
1319 if (res != llvm::conversionOK) {
1320 // If we see bad encoding for unprefixed character literals, warn and
1321 // simply copy the byte values, for compatibility with gcc and
1322 // older versions of clang.
1323 bool NoErrorOnBadEncoding = isAscii();
1324 unsigned Msg = diag::err_bad_character_encoding;
1325 if (NoErrorOnBadEncoding)
1326 Msg = diag::warn_bad_character_encoding;
1327 PP.Diag(Loc, Msg);
1328 if (NoErrorOnBadEncoding) {
1329 start = tmp_in_start;
1330 buffer_begin = tmp_out_start;
1331 for (; start != begin; ++start, ++buffer_begin)
1332 *buffer_begin = static_cast<uint8_t>(*start);
1333 } else {
1334 HadError = true;
1335 }
1336 } else {
1337 for (; tmp_out_start < buffer_begin; ++tmp_out_start) {
1338 if (*tmp_out_start > largest_character_for_kind) {
1339 HadError = true;
1340 PP.Diag(Loc, diag::err_character_too_large);
1341 }
1342 }
1343 }
1344
1345 continue;
1346 }
1347 // Is this a Universal Character Name escape?
1348 if (begin[1] == 'u' || begin[1] == 'U') {
1349 unsigned short UcnLen = 0;
1350 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen,
1351 FullSourceLoc(Loc, PP.getSourceManager()),
1352 &PP.getDiagnostics(), PP.getLangOpts(), true)) {
1353 HadError = true;
1354 } else if (*buffer_begin > largest_character_for_kind) {
1355 HadError = true;
1356 PP.Diag(Loc, diag::err_character_too_large);
1357 }
1358
1359 ++buffer_begin;
1360 continue;
1361 }
1362 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo());
1363 uint64_t result =
1364 ProcessCharEscape(TokBegin, begin, end, HadError,
1365 FullSourceLoc(Loc,PP.getSourceManager()),
1366 CharWidth, &PP.getDiagnostics(), PP.getLangOpts());
1367 *buffer_begin++ = result;
1368 }
1369
1370 unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front();
1371
1372 if (NumCharsSoFar > 1) {
1373 if (isWide())
1374 PP.Diag(Loc, diag::warn_extraneous_char_constant);
1375 else if (isAscii() && NumCharsSoFar == 4)
1376 PP.Diag(Loc, diag::ext_four_char_character_literal);
1377 else if (isAscii())
1378 PP.Diag(Loc, diag::ext_multichar_character_literal);
1379 else
1380 PP.Diag(Loc, diag::err_multichar_utf_character_literal);
1381 IsMultiChar = true;
1382 } else {
1383 IsMultiChar = false;
1384 }
1385
1386 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0);
1387
1388 // Narrow character literals act as though their value is concatenated
1389 // in this implementation, but warn on overflow.
1390 bool multi_char_too_long = false;
1391 if (isAscii() && isMultiChar()) {
1392 LitVal = 0;
1393 for (size_t i = 0; i < NumCharsSoFar; ++i) {
1394 // check for enough leading zeros to shift into
1395 multi_char_too_long |= (LitVal.countLeadingZeros() < 8);
1396 LitVal <<= 8;
1397 LitVal = LitVal + (codepoint_buffer[i] & 0xFF);
1398 }
1399 } else if (NumCharsSoFar > 0) {
1400 // otherwise just take the last character
1401 LitVal = buffer_begin[-1];
1402 }
1403
1404 if (!HadError && multi_char_too_long) {
1405 PP.Diag(Loc, diag::warn_char_constant_too_large);
1406 }
1407
1408 // Transfer the value from APInt to uint64_t
1409 Value = LitVal.getZExtValue();
1410
1411 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1")
1412 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple
1413 // character constants are not sign extended in the this implementation:
1414 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC.
1415 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) &&
1416 PP.getLangOpts().CharIsSigned)
1417 Value = (signed char)Value;
1418 }
1419
1420 /// \verbatim
1421 /// string-literal: [C++0x lex.string]
1422 /// encoding-prefix " [s-char-sequence] "
1423 /// encoding-prefix R raw-string
1424 /// encoding-prefix:
1425 /// u8
1426 /// u
1427 /// U
1428 /// L
1429 /// s-char-sequence:
1430 /// s-char
1431 /// s-char-sequence s-char
1432 /// s-char:
1433 /// any member of the source character set except the double-quote ",
1434 /// backslash \, or new-line character
1435 /// escape-sequence
1436 /// universal-character-name
1437 /// raw-string:
1438 /// " d-char-sequence ( r-char-sequence ) d-char-sequence "
1439 /// r-char-sequence:
1440 /// r-char
1441 /// r-char-sequence r-char
1442 /// r-char:
1443 /// any member of the source character set, except a right parenthesis )
1444 /// followed by the initial d-char-sequence (which may be empty)
1445 /// followed by a double quote ".
1446 /// d-char-sequence:
1447 /// d-char
1448 /// d-char-sequence d-char
1449 /// d-char:
1450 /// any member of the basic source character set except:
1451 /// space, the left parenthesis (, the right parenthesis ),
1452 /// the backslash \, and the control characters representing horizontal
1453 /// tab, vertical tab, form feed, and newline.
1454 /// escape-sequence: [C++0x lex.ccon]
1455 /// simple-escape-sequence
1456 /// octal-escape-sequence
1457 /// hexadecimal-escape-sequence
1458 /// simple-escape-sequence:
1459 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1460 /// octal-escape-sequence:
1461 /// \ octal-digit
1462 /// \ octal-digit octal-digit
1463 /// \ octal-digit octal-digit octal-digit
1464 /// hexadecimal-escape-sequence:
1465 /// \x hexadecimal-digit
1466 /// hexadecimal-escape-sequence hexadecimal-digit
1467 /// universal-character-name:
1468 /// \u hex-quad
1469 /// \U hex-quad hex-quad
1470 /// hex-quad:
1471 /// hex-digit hex-digit hex-digit hex-digit
1472 /// \endverbatim
1473 ///
1474 StringLiteralParser::
StringLiteralParser(ArrayRef<Token> StringToks,Preprocessor & PP,bool Complain)1475 StringLiteralParser(ArrayRef<Token> StringToks,
1476 Preprocessor &PP, bool Complain)
1477 : SM(PP.getSourceManager()), Features(PP.getLangOpts()),
1478 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() :nullptr),
1479 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown),
1480 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) {
1481 init(StringToks);
1482 }
1483
init(ArrayRef<Token> StringToks)1484 void StringLiteralParser::init(ArrayRef<Token> StringToks){
1485 // The literal token may have come from an invalid source location (e.g. due
1486 // to a PCH error), in which case the token length will be 0.
1487 if (StringToks.empty() || StringToks[0].getLength() < 2)
1488 return DiagnoseLexingError(SourceLocation());
1489
1490 // Scan all of the string portions, remember the max individual token length,
1491 // computing a bound on the concatenated string length, and see whether any
1492 // piece is a wide-string. If any of the string portions is a wide-string
1493 // literal, the result is a wide-string literal [C99 6.4.5p4].
1494 assert(!StringToks.empty() && "expected at least one token");
1495 MaxTokenLength = StringToks[0].getLength();
1496 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!");
1497 SizeBound = StringToks[0].getLength()-2; // -2 for "".
1498 Kind = StringToks[0].getKind();
1499
1500 hadError = false;
1501
1502 // Implement Translation Phase #6: concatenation of string literals
1503 /// (C99 5.1.1.2p1). The common case is only one string fragment.
1504 for (unsigned i = 1; i != StringToks.size(); ++i) {
1505 if (StringToks[i].getLength() < 2)
1506 return DiagnoseLexingError(StringToks[i].getLocation());
1507
1508 // The string could be shorter than this if it needs cleaning, but this is a
1509 // reasonable bound, which is all we need.
1510 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!");
1511 SizeBound += StringToks[i].getLength()-2; // -2 for "".
1512
1513 // Remember maximum string piece length.
1514 if (StringToks[i].getLength() > MaxTokenLength)
1515 MaxTokenLength = StringToks[i].getLength();
1516
1517 // Remember if we see any wide or utf-8/16/32 strings.
1518 // Also check for illegal concatenations.
1519 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) {
1520 if (isAscii()) {
1521 Kind = StringToks[i].getKind();
1522 } else {
1523 if (Diags)
1524 Diags->Report(StringToks[i].getLocation(),
1525 diag::err_unsupported_string_concat);
1526 hadError = true;
1527 }
1528 }
1529 }
1530
1531 // Include space for the null terminator.
1532 ++SizeBound;
1533
1534 // TODO: K&R warning: "traditional C rejects string constant concatenation"
1535
1536 // Get the width in bytes of char/wchar_t/char16_t/char32_t
1537 CharByteWidth = getCharWidth(Kind, Target);
1538 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple");
1539 CharByteWidth /= 8;
1540
1541 // The output buffer size needs to be large enough to hold wide characters.
1542 // This is a worst-case assumption which basically corresponds to L"" "long".
1543 SizeBound *= CharByteWidth;
1544
1545 // Size the temporary buffer to hold the result string data.
1546 ResultBuf.resize(SizeBound);
1547
1548 // Likewise, but for each string piece.
1549 SmallString<512> TokenBuf;
1550 TokenBuf.resize(MaxTokenLength);
1551
1552 // Loop over all the strings, getting their spelling, and expanding them to
1553 // wide strings as appropriate.
1554 ResultPtr = &ResultBuf[0]; // Next byte to fill in.
1555
1556 Pascal = false;
1557
1558 SourceLocation UDSuffixTokLoc;
1559
1560 for (unsigned i = 0, e = StringToks.size(); i != e; ++i) {
1561 const char *ThisTokBuf = &TokenBuf[0];
1562 // Get the spelling of the token, which eliminates trigraphs, etc. We know
1563 // that ThisTokBuf points to a buffer that is big enough for the whole token
1564 // and 'spelled' tokens can only shrink.
1565 bool StringInvalid = false;
1566 unsigned ThisTokLen =
1567 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features,
1568 &StringInvalid);
1569 if (StringInvalid)
1570 return DiagnoseLexingError(StringToks[i].getLocation());
1571
1572 const char *ThisTokBegin = ThisTokBuf;
1573 const char *ThisTokEnd = ThisTokBuf+ThisTokLen;
1574
1575 // Remove an optional ud-suffix.
1576 if (ThisTokEnd[-1] != '"') {
1577 const char *UDSuffixEnd = ThisTokEnd;
1578 do {
1579 --ThisTokEnd;
1580 } while (ThisTokEnd[-1] != '"');
1581
1582 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd);
1583
1584 if (UDSuffixBuf.empty()) {
1585 if (StringToks[i].hasUCN())
1586 expandUCNs(UDSuffixBuf, UDSuffix);
1587 else
1588 UDSuffixBuf.assign(UDSuffix);
1589 UDSuffixToken = i;
1590 UDSuffixOffset = ThisTokEnd - ThisTokBuf;
1591 UDSuffixTokLoc = StringToks[i].getLocation();
1592 } else {
1593 SmallString<32> ExpandedUDSuffix;
1594 if (StringToks[i].hasUCN()) {
1595 expandUCNs(ExpandedUDSuffix, UDSuffix);
1596 UDSuffix = ExpandedUDSuffix;
1597 }
1598
1599 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the
1600 // result of a concatenation involving at least one user-defined-string-
1601 // literal, all the participating user-defined-string-literals shall
1602 // have the same ud-suffix.
1603 if (UDSuffixBuf != UDSuffix) {
1604 if (Diags) {
1605 SourceLocation TokLoc = StringToks[i].getLocation();
1606 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix)
1607 << UDSuffixBuf << UDSuffix
1608 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc)
1609 << SourceRange(TokLoc, TokLoc);
1610 }
1611 hadError = true;
1612 }
1613 }
1614 }
1615
1616 // Strip the end quote.
1617 --ThisTokEnd;
1618
1619 // TODO: Input character set mapping support.
1620
1621 // Skip marker for wide or unicode strings.
1622 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') {
1623 ++ThisTokBuf;
1624 // Skip 8 of u8 marker for utf8 strings.
1625 if (ThisTokBuf[0] == '8')
1626 ++ThisTokBuf;
1627 }
1628
1629 // Check for raw string
1630 if (ThisTokBuf[0] == 'R') {
1631 ThisTokBuf += 2; // skip R"
1632
1633 const char *Prefix = ThisTokBuf;
1634 while (ThisTokBuf[0] != '(')
1635 ++ThisTokBuf;
1636 ++ThisTokBuf; // skip '('
1637
1638 // Remove same number of characters from the end
1639 ThisTokEnd -= ThisTokBuf - Prefix;
1640 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal");
1641
1642 // C++14 [lex.string]p4: A source-file new-line in a raw string literal
1643 // results in a new-line in the resulting execution string-literal.
1644 StringRef RemainingTokenSpan(ThisTokBuf, ThisTokEnd - ThisTokBuf);
1645 while (!RemainingTokenSpan.empty()) {
1646 // Split the string literal on \r\n boundaries.
1647 size_t CRLFPos = RemainingTokenSpan.find("\r\n");
1648 StringRef BeforeCRLF = RemainingTokenSpan.substr(0, CRLFPos);
1649 StringRef AfterCRLF = RemainingTokenSpan.substr(CRLFPos);
1650
1651 // Copy everything before the \r\n sequence into the string literal.
1652 if (CopyStringFragment(StringToks[i], ThisTokBegin, BeforeCRLF))
1653 hadError = true;
1654
1655 // Point into the \n inside the \r\n sequence and operate on the
1656 // remaining portion of the literal.
1657 RemainingTokenSpan = AfterCRLF.substr(1);
1658 }
1659 } else {
1660 if (ThisTokBuf[0] != '"') {
1661 // The file may have come from PCH and then changed after loading the
1662 // PCH; Fail gracefully.
1663 return DiagnoseLexingError(StringToks[i].getLocation());
1664 }
1665 ++ThisTokBuf; // skip "
1666
1667 // Check if this is a pascal string
1668 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd &&
1669 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') {
1670
1671 // If the \p sequence is found in the first token, we have a pascal string
1672 // Otherwise, if we already have a pascal string, ignore the first \p
1673 if (i == 0) {
1674 ++ThisTokBuf;
1675 Pascal = true;
1676 } else if (Pascal)
1677 ThisTokBuf += 2;
1678 }
1679
1680 while (ThisTokBuf != ThisTokEnd) {
1681 // Is this a span of non-escape characters?
1682 if (ThisTokBuf[0] != '\\') {
1683 const char *InStart = ThisTokBuf;
1684 do {
1685 ++ThisTokBuf;
1686 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\');
1687
1688 // Copy the character span over.
1689 if (CopyStringFragment(StringToks[i], ThisTokBegin,
1690 StringRef(InStart, ThisTokBuf - InStart)))
1691 hadError = true;
1692 continue;
1693 }
1694 // Is this a Universal Character Name escape?
1695 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') {
1696 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd,
1697 ResultPtr, hadError,
1698 FullSourceLoc(StringToks[i].getLocation(), SM),
1699 CharByteWidth, Diags, Features);
1700 continue;
1701 }
1702 // Otherwise, this is a non-UCN escape character. Process it.
1703 unsigned ResultChar =
1704 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError,
1705 FullSourceLoc(StringToks[i].getLocation(), SM),
1706 CharByteWidth*8, Diags, Features);
1707
1708 if (CharByteWidth == 4) {
1709 // FIXME: Make the type of the result buffer correct instead of
1710 // using reinterpret_cast.
1711 llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultPtr);
1712 *ResultWidePtr = ResultChar;
1713 ResultPtr += 4;
1714 } else if (CharByteWidth == 2) {
1715 // FIXME: Make the type of the result buffer correct instead of
1716 // using reinterpret_cast.
1717 llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultPtr);
1718 *ResultWidePtr = ResultChar & 0xFFFF;
1719 ResultPtr += 2;
1720 } else {
1721 assert(CharByteWidth == 1 && "Unexpected char width");
1722 *ResultPtr++ = ResultChar & 0xFF;
1723 }
1724 }
1725 }
1726 }
1727
1728 if (Pascal) {
1729 if (CharByteWidth == 4) {
1730 // FIXME: Make the type of the result buffer correct instead of
1731 // using reinterpret_cast.
1732 llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultBuf.data());
1733 ResultWidePtr[0] = GetNumStringChars() - 1;
1734 } else if (CharByteWidth == 2) {
1735 // FIXME: Make the type of the result buffer correct instead of
1736 // using reinterpret_cast.
1737 llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultBuf.data());
1738 ResultWidePtr[0] = GetNumStringChars() - 1;
1739 } else {
1740 assert(CharByteWidth == 1 && "Unexpected char width");
1741 ResultBuf[0] = GetNumStringChars() - 1;
1742 }
1743
1744 // Verify that pascal strings aren't too large.
1745 if (GetStringLength() > 256) {
1746 if (Diags)
1747 Diags->Report(StringToks.front().getLocation(),
1748 diag::err_pascal_string_too_long)
1749 << SourceRange(StringToks.front().getLocation(),
1750 StringToks.back().getLocation());
1751 hadError = true;
1752 return;
1753 }
1754 } else if (Diags) {
1755 // Complain if this string literal has too many characters.
1756 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509;
1757
1758 if (GetNumStringChars() > MaxChars)
1759 Diags->Report(StringToks.front().getLocation(),
1760 diag::ext_string_too_long)
1761 << GetNumStringChars() << MaxChars
1762 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0)
1763 << SourceRange(StringToks.front().getLocation(),
1764 StringToks.back().getLocation());
1765 }
1766 }
1767
resyncUTF8(const char * Err,const char * End)1768 static const char *resyncUTF8(const char *Err, const char *End) {
1769 if (Err == End)
1770 return End;
1771 End = Err + std::min<unsigned>(llvm::getNumBytesForUTF8(*Err), End-Err);
1772 while (++Err != End && (*Err & 0xC0) == 0x80)
1773 ;
1774 return Err;
1775 }
1776
1777 /// This function copies from Fragment, which is a sequence of bytes
1778 /// within Tok's contents (which begin at TokBegin) into ResultPtr.
1779 /// Performs widening for multi-byte characters.
CopyStringFragment(const Token & Tok,const char * TokBegin,StringRef Fragment)1780 bool StringLiteralParser::CopyStringFragment(const Token &Tok,
1781 const char *TokBegin,
1782 StringRef Fragment) {
1783 const llvm::UTF8 *ErrorPtrTmp;
1784 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp))
1785 return false;
1786
1787 // If we see bad encoding for unprefixed string literals, warn and
1788 // simply copy the byte values, for compatibility with gcc and older
1789 // versions of clang.
1790 bool NoErrorOnBadEncoding = isAscii();
1791 if (NoErrorOnBadEncoding) {
1792 memcpy(ResultPtr, Fragment.data(), Fragment.size());
1793 ResultPtr += Fragment.size();
1794 }
1795
1796 if (Diags) {
1797 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1798
1799 FullSourceLoc SourceLoc(Tok.getLocation(), SM);
1800 const DiagnosticBuilder &Builder =
1801 Diag(Diags, Features, SourceLoc, TokBegin,
1802 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()),
1803 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding
1804 : diag::err_bad_string_encoding);
1805
1806 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1807 StringRef NextFragment(NextStart, Fragment.end()-NextStart);
1808
1809 // Decode into a dummy buffer.
1810 SmallString<512> Dummy;
1811 Dummy.reserve(Fragment.size() * CharByteWidth);
1812 char *Ptr = Dummy.data();
1813
1814 while (!ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) {
1815 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1816 NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1817 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin,
1818 ErrorPtr, NextStart);
1819 NextFragment = StringRef(NextStart, Fragment.end()-NextStart);
1820 }
1821 }
1822 return !NoErrorOnBadEncoding;
1823 }
1824
DiagnoseLexingError(SourceLocation Loc)1825 void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) {
1826 hadError = true;
1827 if (Diags)
1828 Diags->Report(Loc, diag::err_lexing_string);
1829 }
1830
1831 /// getOffsetOfStringByte - This function returns the offset of the
1832 /// specified byte of the string data represented by Token. This handles
1833 /// advancing over escape sequences in the string.
getOffsetOfStringByte(const Token & Tok,unsigned ByteNo) const1834 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok,
1835 unsigned ByteNo) const {
1836 // Get the spelling of the token.
1837 SmallString<32> SpellingBuffer;
1838 SpellingBuffer.resize(Tok.getLength());
1839
1840 bool StringInvalid = false;
1841 const char *SpellingPtr = &SpellingBuffer[0];
1842 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features,
1843 &StringInvalid);
1844 if (StringInvalid)
1845 return 0;
1846
1847 const char *SpellingStart = SpellingPtr;
1848 const char *SpellingEnd = SpellingPtr+TokLen;
1849
1850 // Handle UTF-8 strings just like narrow strings.
1851 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8')
1852 SpellingPtr += 2;
1853
1854 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' &&
1855 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet");
1856
1857 // For raw string literals, this is easy.
1858 if (SpellingPtr[0] == 'R') {
1859 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!");
1860 // Skip 'R"'.
1861 SpellingPtr += 2;
1862 while (*SpellingPtr != '(') {
1863 ++SpellingPtr;
1864 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal");
1865 }
1866 // Skip '('.
1867 ++SpellingPtr;
1868 return SpellingPtr - SpellingStart + ByteNo;
1869 }
1870
1871 // Skip over the leading quote
1872 assert(SpellingPtr[0] == '"' && "Should be a string literal!");
1873 ++SpellingPtr;
1874
1875 // Skip over bytes until we find the offset we're looking for.
1876 while (ByteNo) {
1877 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!");
1878
1879 // Step over non-escapes simply.
1880 if (*SpellingPtr != '\\') {
1881 ++SpellingPtr;
1882 --ByteNo;
1883 continue;
1884 }
1885
1886 // Otherwise, this is an escape character. Advance over it.
1887 bool HadError = false;
1888 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') {
1889 const char *EscapePtr = SpellingPtr;
1890 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd,
1891 1, Features, HadError);
1892 if (Len > ByteNo) {
1893 // ByteNo is somewhere within the escape sequence.
1894 SpellingPtr = EscapePtr;
1895 break;
1896 }
1897 ByteNo -= Len;
1898 } else {
1899 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError,
1900 FullSourceLoc(Tok.getLocation(), SM),
1901 CharByteWidth*8, Diags, Features);
1902 --ByteNo;
1903 }
1904 assert(!HadError && "This method isn't valid on erroneous strings");
1905 }
1906
1907 return SpellingPtr-SpellingStart;
1908 }
1909
1910 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
1911 /// suffixes as ud-suffixes, because the diagnostic experience is better if we
1912 /// treat it as an invalid suffix.
isValidUDSuffix(const LangOptions & LangOpts,StringRef Suffix)1913 bool StringLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
1914 StringRef Suffix) {
1915 return NumericLiteralParser::isValidUDSuffix(LangOpts, Suffix) ||
1916 Suffix == "sv";
1917 }
1918