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/ErrorHandling.h"
29 #include <algorithm>
30 #include <cassert>
31 #include <cstddef>
32 #include <cstdint>
33 #include <cstring>
34 #include <string>
35
36 using namespace clang;
37
getCharWidth(tok::TokenKind kind,const TargetInfo & Target)38 static unsigned getCharWidth(tok::TokenKind kind, const TargetInfo &Target) {
39 switch (kind) {
40 default: llvm_unreachable("Unknown token type!");
41 case tok::char_constant:
42 case tok::string_literal:
43 case tok::utf8_char_constant:
44 case tok::utf8_string_literal:
45 return Target.getCharWidth();
46 case tok::wide_char_constant:
47 case tok::wide_string_literal:
48 return Target.getWCharWidth();
49 case tok::utf16_char_constant:
50 case tok::utf16_string_literal:
51 return Target.getChar16Width();
52 case tok::utf32_char_constant:
53 case tok::utf32_string_literal:
54 return Target.getChar32Width();
55 }
56 }
57
MakeCharSourceRange(const LangOptions & Features,FullSourceLoc TokLoc,const char * TokBegin,const char * TokRangeBegin,const char * TokRangeEnd)58 static CharSourceRange MakeCharSourceRange(const LangOptions &Features,
59 FullSourceLoc TokLoc,
60 const char *TokBegin,
61 const char *TokRangeBegin,
62 const char *TokRangeEnd) {
63 SourceLocation Begin =
64 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin,
65 TokLoc.getManager(), Features);
66 SourceLocation End =
67 Lexer::AdvanceToTokenCharacter(Begin, TokRangeEnd - TokRangeBegin,
68 TokLoc.getManager(), Features);
69 return CharSourceRange::getCharRange(Begin, End);
70 }
71
72 /// Produce a diagnostic highlighting some portion of a literal.
73 ///
74 /// Emits the diagnostic \p DiagID, highlighting the range of characters from
75 /// \p TokRangeBegin (inclusive) to \p TokRangeEnd (exclusive), which must be
76 /// 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)77 static DiagnosticBuilder Diag(DiagnosticsEngine *Diags,
78 const LangOptions &Features, FullSourceLoc TokLoc,
79 const char *TokBegin, const char *TokRangeBegin,
80 const char *TokRangeEnd, unsigned DiagID) {
81 SourceLocation Begin =
82 Lexer::AdvanceToTokenCharacter(TokLoc, TokRangeBegin - TokBegin,
83 TokLoc.getManager(), Features);
84 return Diags->Report(Begin, DiagID) <<
85 MakeCharSourceRange(Features, TokLoc, TokBegin, TokRangeBegin, TokRangeEnd);
86 }
87
88 /// ProcessCharEscape - Parse a standard C escape sequence, which can occur in
89 /// 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)90 static unsigned ProcessCharEscape(const char *ThisTokBegin,
91 const char *&ThisTokBuf,
92 const char *ThisTokEnd, bool &HadError,
93 FullSourceLoc Loc, unsigned CharWidth,
94 DiagnosticsEngine *Diags,
95 const LangOptions &Features) {
96 const char *EscapeBegin = ThisTokBuf;
97
98 // Skip the '\' char.
99 ++ThisTokBuf;
100
101 // We know that this character can't be off the end of the buffer, because
102 // that would have been \", which would not have been the end of string.
103 unsigned ResultChar = *ThisTokBuf++;
104 switch (ResultChar) {
105 // These map to themselves.
106 case '\\': case '\'': case '"': case '?': break;
107
108 // These have fixed mappings.
109 case 'a':
110 // TODO: K&R: the meaning of '\\a' is different in traditional C
111 ResultChar = 7;
112 break;
113 case 'b':
114 ResultChar = 8;
115 break;
116 case 'e':
117 if (Diags)
118 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
119 diag::ext_nonstandard_escape) << "e";
120 ResultChar = 27;
121 break;
122 case 'E':
123 if (Diags)
124 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
125 diag::ext_nonstandard_escape) << "E";
126 ResultChar = 27;
127 break;
128 case 'f':
129 ResultChar = 12;
130 break;
131 case 'n':
132 ResultChar = 10;
133 break;
134 case 'r':
135 ResultChar = 13;
136 break;
137 case 't':
138 ResultChar = 9;
139 break;
140 case 'v':
141 ResultChar = 11;
142 break;
143 case 'x': { // Hex escape.
144 ResultChar = 0;
145 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
146 if (Diags)
147 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
148 diag::err_hex_escape_no_digits) << "x";
149 HadError = true;
150 break;
151 }
152
153 // Hex escapes are a maximal series of hex digits.
154 bool Overflow = false;
155 for (; ThisTokBuf != ThisTokEnd; ++ThisTokBuf) {
156 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]);
157 if (CharVal == -1) break;
158 // About to shift out a digit?
159 if (ResultChar & 0xF0000000)
160 Overflow = true;
161 ResultChar <<= 4;
162 ResultChar |= CharVal;
163 }
164
165 // See if any bits will be truncated when evaluated as a character.
166 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
167 Overflow = true;
168 ResultChar &= ~0U >> (32-CharWidth);
169 }
170
171 // Check for overflow.
172 if (Overflow && Diags) // Too many digits to fit in
173 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
174 diag::err_escape_too_large) << 0;
175 break;
176 }
177 case '0': case '1': case '2': case '3':
178 case '4': case '5': case '6': case '7': {
179 // Octal escapes.
180 --ThisTokBuf;
181 ResultChar = 0;
182
183 // Octal escapes are a series of octal digits with maximum length 3.
184 // "\0123" is a two digit sequence equal to "\012" "3".
185 unsigned NumDigits = 0;
186 do {
187 ResultChar <<= 3;
188 ResultChar |= *ThisTokBuf++ - '0';
189 ++NumDigits;
190 } while (ThisTokBuf != ThisTokEnd && NumDigits < 3 &&
191 ThisTokBuf[0] >= '0' && ThisTokBuf[0] <= '7');
192
193 // Check for overflow. Reject '\777', but not L'\777'.
194 if (CharWidth != 32 && (ResultChar >> CharWidth) != 0) {
195 if (Diags)
196 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
197 diag::err_escape_too_large) << 1;
198 ResultChar &= ~0U >> (32-CharWidth);
199 }
200 break;
201 }
202
203 // Otherwise, these are not valid escapes.
204 case '(': case '{': case '[': case '%':
205 // GCC accepts these as extensions. We warn about them as such though.
206 if (Diags)
207 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
208 diag::ext_nonstandard_escape)
209 << std::string(1, ResultChar);
210 break;
211 default:
212 if (!Diags)
213 break;
214
215 if (isPrintable(ResultChar))
216 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
217 diag::ext_unknown_escape)
218 << std::string(1, ResultChar);
219 else
220 Diag(Diags, Features, Loc, ThisTokBegin, EscapeBegin, ThisTokBuf,
221 diag::ext_unknown_escape)
222 << "x" + llvm::utohexstr(ResultChar);
223 break;
224 }
225
226 return ResultChar;
227 }
228
appendCodePoint(unsigned Codepoint,llvm::SmallVectorImpl<char> & Str)229 static void appendCodePoint(unsigned Codepoint,
230 llvm::SmallVectorImpl<char> &Str) {
231 char ResultBuf[4];
232 char *ResultPtr = ResultBuf;
233 bool Res = llvm::ConvertCodePointToUTF8(Codepoint, ResultPtr);
234 (void)Res;
235 assert(Res && "Unexpected conversion failure");
236 Str.append(ResultBuf, ResultPtr);
237 }
238
expandUCNs(SmallVectorImpl<char> & Buf,StringRef Input)239 void clang::expandUCNs(SmallVectorImpl<char> &Buf, StringRef Input) {
240 for (StringRef::iterator I = Input.begin(), E = Input.end(); I != E; ++I) {
241 if (*I != '\\') {
242 Buf.push_back(*I);
243 continue;
244 }
245
246 ++I;
247 assert(*I == 'u' || *I == 'U');
248
249 unsigned NumHexDigits;
250 if (*I == 'u')
251 NumHexDigits = 4;
252 else
253 NumHexDigits = 8;
254
255 assert(I + NumHexDigits <= E);
256
257 uint32_t CodePoint = 0;
258 for (++I; NumHexDigits != 0; ++I, --NumHexDigits) {
259 unsigned Value = llvm::hexDigitValue(*I);
260 assert(Value != -1U);
261
262 CodePoint <<= 4;
263 CodePoint += Value;
264 }
265
266 appendCodePoint(CodePoint, Buf);
267 --I;
268 }
269 }
270
271 /// ProcessUCNEscape - Read the Universal Character Name, check constraints and
272 /// 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)273 static bool ProcessUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf,
274 const char *ThisTokEnd,
275 uint32_t &UcnVal, unsigned short &UcnLen,
276 FullSourceLoc Loc, DiagnosticsEngine *Diags,
277 const LangOptions &Features,
278 bool in_char_string_literal = false) {
279 const char *UcnBegin = ThisTokBuf;
280
281 // Skip the '\u' char's.
282 ThisTokBuf += 2;
283
284 if (ThisTokBuf == ThisTokEnd || !isHexDigit(*ThisTokBuf)) {
285 if (Diags)
286 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
287 diag::err_hex_escape_no_digits) << StringRef(&ThisTokBuf[-1], 1);
288 return false;
289 }
290 UcnLen = (ThisTokBuf[-1] == 'u' ? 4 : 8);
291 unsigned short UcnLenSave = UcnLen;
292 for (; ThisTokBuf != ThisTokEnd && UcnLenSave; ++ThisTokBuf, UcnLenSave--) {
293 int CharVal = llvm::hexDigitValue(ThisTokBuf[0]);
294 if (CharVal == -1) break;
295 UcnVal <<= 4;
296 UcnVal |= CharVal;
297 }
298 // If we didn't consume the proper number of digits, there is a problem.
299 if (UcnLenSave) {
300 if (Diags)
301 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
302 diag::err_ucn_escape_incomplete);
303 return false;
304 }
305
306 // Check UCN constraints (C99 6.4.3p2) [C++11 lex.charset p2]
307 if ((0xD800 <= UcnVal && UcnVal <= 0xDFFF) || // surrogate codepoints
308 UcnVal > 0x10FFFF) { // maximum legal UTF32 value
309 if (Diags)
310 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
311 diag::err_ucn_escape_invalid);
312 return false;
313 }
314
315 // C++11 allows UCNs that refer to control characters and basic source
316 // characters inside character and string literals
317 if (UcnVal < 0xa0 &&
318 (UcnVal != 0x24 && UcnVal != 0x40 && UcnVal != 0x60)) { // $, @, `
319 bool IsError = (!Features.CPlusPlus11 || !in_char_string_literal);
320 if (Diags) {
321 char BasicSCSChar = UcnVal;
322 if (UcnVal >= 0x20 && UcnVal < 0x7f)
323 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
324 IsError ? diag::err_ucn_escape_basic_scs :
325 diag::warn_cxx98_compat_literal_ucn_escape_basic_scs)
326 << StringRef(&BasicSCSChar, 1);
327 else
328 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
329 IsError ? diag::err_ucn_control_character :
330 diag::warn_cxx98_compat_literal_ucn_control_character);
331 }
332 if (IsError)
333 return false;
334 }
335
336 if (!Features.CPlusPlus && !Features.C99 && Diags)
337 Diag(Diags, Features, Loc, ThisTokBegin, UcnBegin, ThisTokBuf,
338 diag::warn_ucn_not_valid_in_c89_literal);
339
340 return true;
341 }
342
343 /// MeasureUCNEscape - Determine the number of bytes within the resulting string
344 /// which this UCN will occupy.
MeasureUCNEscape(const char * ThisTokBegin,const char * & ThisTokBuf,const char * ThisTokEnd,unsigned CharByteWidth,const LangOptions & Features,bool & HadError)345 static int MeasureUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf,
346 const char *ThisTokEnd, unsigned CharByteWidth,
347 const LangOptions &Features, bool &HadError) {
348 // UTF-32: 4 bytes per escape.
349 if (CharByteWidth == 4)
350 return 4;
351
352 uint32_t UcnVal = 0;
353 unsigned short UcnLen = 0;
354 FullSourceLoc Loc;
355
356 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal,
357 UcnLen, Loc, nullptr, Features, true)) {
358 HadError = true;
359 return 0;
360 }
361
362 // UTF-16: 2 bytes for BMP, 4 bytes otherwise.
363 if (CharByteWidth == 2)
364 return UcnVal <= 0xFFFF ? 2 : 4;
365
366 // UTF-8.
367 if (UcnVal < 0x80)
368 return 1;
369 if (UcnVal < 0x800)
370 return 2;
371 if (UcnVal < 0x10000)
372 return 3;
373 return 4;
374 }
375
376 /// EncodeUCNEscape - Read the Universal Character Name, check constraints and
377 /// convert the UTF32 to UTF8 or UTF16. This is a subroutine of
378 /// StringLiteralParser. When we decide to implement UCN's for identifiers,
379 /// 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)380 static void EncodeUCNEscape(const char *ThisTokBegin, const char *&ThisTokBuf,
381 const char *ThisTokEnd,
382 char *&ResultBuf, bool &HadError,
383 FullSourceLoc Loc, unsigned CharByteWidth,
384 DiagnosticsEngine *Diags,
385 const LangOptions &Features) {
386 typedef uint32_t UTF32;
387 UTF32 UcnVal = 0;
388 unsigned short UcnLen = 0;
389 if (!ProcessUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, UcnVal, UcnLen,
390 Loc, Diags, Features, true)) {
391 HadError = true;
392 return;
393 }
394
395 assert((CharByteWidth == 1 || CharByteWidth == 2 || CharByteWidth == 4) &&
396 "only character widths of 1, 2, or 4 bytes supported");
397
398 (void)UcnLen;
399 assert((UcnLen== 4 || UcnLen== 8) && "only ucn length of 4 or 8 supported");
400
401 if (CharByteWidth == 4) {
402 // FIXME: Make the type of the result buffer correct instead of
403 // using reinterpret_cast.
404 llvm::UTF32 *ResultPtr = reinterpret_cast<llvm::UTF32*>(ResultBuf);
405 *ResultPtr = UcnVal;
406 ResultBuf += 4;
407 return;
408 }
409
410 if (CharByteWidth == 2) {
411 // FIXME: Make the type of the result buffer correct instead of
412 // using reinterpret_cast.
413 llvm::UTF16 *ResultPtr = reinterpret_cast<llvm::UTF16*>(ResultBuf);
414
415 if (UcnVal <= (UTF32)0xFFFF) {
416 *ResultPtr = UcnVal;
417 ResultBuf += 2;
418 return;
419 }
420
421 // Convert to UTF16.
422 UcnVal -= 0x10000;
423 *ResultPtr = 0xD800 + (UcnVal >> 10);
424 *(ResultPtr+1) = 0xDC00 + (UcnVal & 0x3FF);
425 ResultBuf += 4;
426 return;
427 }
428
429 assert(CharByteWidth == 1 && "UTF-8 encoding is only for 1 byte characters");
430
431 // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8.
432 // The conversion below was inspired by:
433 // http://www.unicode.org/Public/PROGRAMS/CVTUTF/ConvertUTF.c
434 // First, we determine how many bytes the result will require.
435 typedef uint8_t UTF8;
436
437 unsigned short bytesToWrite = 0;
438 if (UcnVal < (UTF32)0x80)
439 bytesToWrite = 1;
440 else if (UcnVal < (UTF32)0x800)
441 bytesToWrite = 2;
442 else if (UcnVal < (UTF32)0x10000)
443 bytesToWrite = 3;
444 else
445 bytesToWrite = 4;
446
447 const unsigned byteMask = 0xBF;
448 const unsigned byteMark = 0x80;
449
450 // Once the bits are split out into bytes of UTF8, this is a mask OR-ed
451 // into the first byte, depending on how many bytes follow.
452 static const UTF8 firstByteMark[5] = {
453 0x00, 0x00, 0xC0, 0xE0, 0xF0
454 };
455 // Finally, we write the bytes into ResultBuf.
456 ResultBuf += bytesToWrite;
457 switch (bytesToWrite) { // note: everything falls through.
458 case 4:
459 *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
460 LLVM_FALLTHROUGH;
461 case 3:
462 *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
463 LLVM_FALLTHROUGH;
464 case 2:
465 *--ResultBuf = (UTF8)((UcnVal | byteMark) & byteMask); UcnVal >>= 6;
466 LLVM_FALLTHROUGH;
467 case 1:
468 *--ResultBuf = (UTF8) (UcnVal | firstByteMark[bytesToWrite]);
469 }
470 // Update the buffer.
471 ResultBuf += bytesToWrite;
472 }
473
474 /// integer-constant: [C99 6.4.4.1]
475 /// decimal-constant integer-suffix
476 /// octal-constant integer-suffix
477 /// hexadecimal-constant integer-suffix
478 /// binary-literal integer-suffix [GNU, C++1y]
479 /// user-defined-integer-literal: [C++11 lex.ext]
480 /// decimal-literal ud-suffix
481 /// octal-literal ud-suffix
482 /// hexadecimal-literal ud-suffix
483 /// binary-literal ud-suffix [GNU, C++1y]
484 /// decimal-constant:
485 /// nonzero-digit
486 /// decimal-constant digit
487 /// octal-constant:
488 /// 0
489 /// octal-constant octal-digit
490 /// hexadecimal-constant:
491 /// hexadecimal-prefix hexadecimal-digit
492 /// hexadecimal-constant hexadecimal-digit
493 /// hexadecimal-prefix: one of
494 /// 0x 0X
495 /// binary-literal:
496 /// 0b binary-digit
497 /// 0B binary-digit
498 /// binary-literal binary-digit
499 /// integer-suffix:
500 /// unsigned-suffix [long-suffix]
501 /// unsigned-suffix [long-long-suffix]
502 /// long-suffix [unsigned-suffix]
503 /// long-long-suffix [unsigned-sufix]
504 /// nonzero-digit:
505 /// 1 2 3 4 5 6 7 8 9
506 /// octal-digit:
507 /// 0 1 2 3 4 5 6 7
508 /// hexadecimal-digit:
509 /// 0 1 2 3 4 5 6 7 8 9
510 /// a b c d e f
511 /// A B C D E F
512 /// binary-digit:
513 /// 0
514 /// 1
515 /// unsigned-suffix: one of
516 /// u U
517 /// long-suffix: one of
518 /// l L
519 /// long-long-suffix: one of
520 /// ll LL
521 ///
522 /// floating-constant: [C99 6.4.4.2]
523 /// TODO: add rules...
524 ///
NumericLiteralParser(StringRef TokSpelling,SourceLocation TokLoc,Preprocessor & PP)525 NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
526 SourceLocation TokLoc,
527 Preprocessor &PP)
528 : PP(PP), ThisTokBegin(TokSpelling.begin()), ThisTokEnd(TokSpelling.end()) {
529
530 // This routine assumes that the range begin/end matches the regex for integer
531 // and FP constants (specifically, the 'pp-number' regex), and assumes that
532 // the byte at "*end" is both valid and not part of the regex. Because of
533 // this, it doesn't have to check for 'overscan' in various places.
534 assert(!isPreprocessingNumberBody(*ThisTokEnd) && "didn't maximally munch?");
535
536 s = DigitsBegin = ThisTokBegin;
537 saw_exponent = false;
538 saw_period = false;
539 saw_ud_suffix = false;
540 saw_fixed_point_suffix = false;
541 isLong = false;
542 isUnsigned = false;
543 isLongLong = false;
544 isHalf = false;
545 isFloat = false;
546 isImaginary = false;
547 isFloat16 = false;
548 isFloat128 = false;
549 MicrosoftInteger = 0;
550 isFract = false;
551 isAccum = false;
552 hadError = false;
553
554 if (*s == '0') { // parse radix
555 ParseNumberStartingWithZero(TokLoc);
556 if (hadError)
557 return;
558 } else { // the first digit is non-zero
559 radix = 10;
560 s = SkipDigits(s);
561 if (s == ThisTokEnd) {
562 // Done.
563 } else {
564 ParseDecimalOrOctalCommon(TokLoc);
565 if (hadError)
566 return;
567 }
568 }
569
570 SuffixBegin = s;
571 checkSeparator(TokLoc, s, CSK_AfterDigits);
572
573 // Initial scan to lookahead for fixed point suffix.
574 if (PP.getLangOpts().FixedPoint) {
575 for (const char *c = s; c != ThisTokEnd; ++c) {
576 if (*c == 'r' || *c == 'k' || *c == 'R' || *c == 'K') {
577 saw_fixed_point_suffix = true;
578 break;
579 }
580 }
581 }
582
583 // Parse the suffix. At this point we can classify whether we have an FP or
584 // integer constant.
585 bool isFPConstant = isFloatingLiteral();
586
587 // Loop over all of the characters of the suffix. If we see something bad,
588 // we break out of the loop.
589 for (; s != ThisTokEnd; ++s) {
590 switch (*s) {
591 case 'R':
592 case 'r':
593 if (!PP.getLangOpts().FixedPoint) break;
594 if (isFract || isAccum) break;
595 if (!(saw_period || saw_exponent)) break;
596 isFract = true;
597 continue;
598 case 'K':
599 case 'k':
600 if (!PP.getLangOpts().FixedPoint) break;
601 if (isFract || isAccum) break;
602 if (!(saw_period || saw_exponent)) break;
603 isAccum = true;
604 continue;
605 case 'h': // FP Suffix for "half".
606 case 'H':
607 // OpenCL Extension v1.2 s9.5 - h or H suffix for half type.
608 if (!(PP.getLangOpts().Half || PP.getLangOpts().FixedPoint)) break;
609 if (isIntegerLiteral()) break; // Error for integer constant.
610 if (isHalf || isFloat || isLong) break; // HH, FH, LH invalid.
611 isHalf = true;
612 continue; // Success.
613 case 'f': // FP Suffix for "float"
614 case 'F':
615 if (!isFPConstant) break; // Error for integer constant.
616 if (isHalf || isFloat || isLong || isFloat128)
617 break; // HF, FF, LF, QF invalid.
618
619 // CUDA host and device may have different _Float16 support, therefore
620 // allows f16 literals to avoid false alarm.
621 // ToDo: more precise check for CUDA.
622 if ((PP.getTargetInfo().hasFloat16Type() || PP.getLangOpts().CUDA) &&
623 s + 2 < ThisTokEnd && s[1] == '1' && s[2] == '6') {
624 s += 2; // success, eat up 2 characters.
625 isFloat16 = true;
626 continue;
627 }
628
629 isFloat = true;
630 continue; // Success.
631 case 'q': // FP Suffix for "__float128"
632 case 'Q':
633 if (!isFPConstant) break; // Error for integer constant.
634 if (isHalf || isFloat || isLong || isFloat128)
635 break; // HQ, FQ, LQ, QQ invalid.
636 isFloat128 = true;
637 continue; // Success.
638 case 'u':
639 case 'U':
640 if (isFPConstant) break; // Error for floating constant.
641 if (isUnsigned) break; // Cannot be repeated.
642 isUnsigned = true;
643 continue; // Success.
644 case 'l':
645 case 'L':
646 if (isLong || isLongLong) break; // Cannot be repeated.
647 if (isHalf || isFloat || isFloat128) break; // LH, LF, LQ invalid.
648
649 // Check for long long. The L's need to be adjacent and the same case.
650 if (s[1] == s[0]) {
651 assert(s + 1 < ThisTokEnd && "didn't maximally munch?");
652 if (isFPConstant) break; // long long invalid for floats.
653 isLongLong = true;
654 ++s; // Eat both of them.
655 } else {
656 isLong = true;
657 }
658 continue; // Success.
659 case 'i':
660 case 'I':
661 if (PP.getLangOpts().MicrosoftExt) {
662 if (isLong || isLongLong || MicrosoftInteger)
663 break;
664
665 if (!isFPConstant) {
666 // Allow i8, i16, i32, and i64.
667 switch (s[1]) {
668 case '8':
669 s += 2; // i8 suffix
670 MicrosoftInteger = 8;
671 break;
672 case '1':
673 if (s[2] == '6') {
674 s += 3; // i16 suffix
675 MicrosoftInteger = 16;
676 }
677 break;
678 case '3':
679 if (s[2] == '2') {
680 s += 3; // i32 suffix
681 MicrosoftInteger = 32;
682 }
683 break;
684 case '6':
685 if (s[2] == '4') {
686 s += 3; // i64 suffix
687 MicrosoftInteger = 64;
688 }
689 break;
690 default:
691 break;
692 }
693 }
694 if (MicrosoftInteger) {
695 assert(s <= ThisTokEnd && "didn't maximally munch?");
696 break;
697 }
698 }
699 LLVM_FALLTHROUGH;
700 case 'j':
701 case 'J':
702 if (isImaginary) break; // Cannot be repeated.
703 isImaginary = true;
704 continue; // Success.
705 }
706 // If we reached here, there was an error or a ud-suffix.
707 break;
708 }
709
710 // "i", "if", and "il" are user-defined suffixes in C++1y.
711 if (s != ThisTokEnd || isImaginary) {
712 // FIXME: Don't bother expanding UCNs if !tok.hasUCN().
713 expandUCNs(UDSuffixBuf, StringRef(SuffixBegin, ThisTokEnd - SuffixBegin));
714 if (isValidUDSuffix(PP.getLangOpts(), UDSuffixBuf)) {
715 if (!isImaginary) {
716 // Any suffix pieces we might have parsed are actually part of the
717 // ud-suffix.
718 isLong = false;
719 isUnsigned = false;
720 isLongLong = false;
721 isFloat = false;
722 isFloat16 = false;
723 isHalf = false;
724 isImaginary = false;
725 MicrosoftInteger = 0;
726 saw_fixed_point_suffix = false;
727 isFract = false;
728 isAccum = false;
729 }
730
731 saw_ud_suffix = true;
732 return;
733 }
734
735 if (s != ThisTokEnd) {
736 // Report an error if there are any.
737 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin),
738 diag::err_invalid_suffix_constant)
739 << StringRef(SuffixBegin, ThisTokEnd - SuffixBegin) << isFPConstant;
740 hadError = true;
741 }
742 }
743
744 if (!hadError && saw_fixed_point_suffix) {
745 assert(isFract || isAccum);
746 }
747 }
748
749 /// ParseDecimalOrOctalCommon - This method is called for decimal or octal
750 /// numbers. It issues an error for illegal digits, and handles floating point
751 /// parsing. If it detects a floating point number, the radix is set to 10.
ParseDecimalOrOctalCommon(SourceLocation TokLoc)752 void NumericLiteralParser::ParseDecimalOrOctalCommon(SourceLocation TokLoc){
753 assert((radix == 8 || radix == 10) && "Unexpected radix");
754
755 // If we have a hex digit other than 'e' (which denotes a FP exponent) then
756 // the code is using an incorrect base.
757 if (isHexDigit(*s) && *s != 'e' && *s != 'E' &&
758 !isValidUDSuffix(PP.getLangOpts(), StringRef(s, ThisTokEnd - s))) {
759 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
760 diag::err_invalid_digit) << StringRef(s, 1) << (radix == 8 ? 1 : 0);
761 hadError = true;
762 return;
763 }
764
765 if (*s == '.') {
766 checkSeparator(TokLoc, s, CSK_AfterDigits);
767 s++;
768 radix = 10;
769 saw_period = true;
770 checkSeparator(TokLoc, s, CSK_BeforeDigits);
771 s = SkipDigits(s); // Skip suffix.
772 }
773 if (*s == 'e' || *s == 'E') { // exponent
774 checkSeparator(TokLoc, s, CSK_AfterDigits);
775 const char *Exponent = s;
776 s++;
777 radix = 10;
778 saw_exponent = true;
779 if (s != ThisTokEnd && (*s == '+' || *s == '-')) s++; // sign
780 const char *first_non_digit = SkipDigits(s);
781 if (containsDigits(s, first_non_digit)) {
782 checkSeparator(TokLoc, s, CSK_BeforeDigits);
783 s = first_non_digit;
784 } else {
785 if (!hadError) {
786 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
787 diag::err_exponent_has_no_digits);
788 hadError = true;
789 }
790 return;
791 }
792 }
793 }
794
795 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
796 /// suffixes as ud-suffixes, because the diagnostic experience is better if we
797 /// treat it as an invalid suffix.
isValidUDSuffix(const LangOptions & LangOpts,StringRef Suffix)798 bool NumericLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
799 StringRef Suffix) {
800 if (!LangOpts.CPlusPlus11 || Suffix.empty())
801 return false;
802
803 // By C++11 [lex.ext]p10, ud-suffixes starting with an '_' are always valid.
804 if (Suffix[0] == '_')
805 return true;
806
807 // In C++11, there are no library suffixes.
808 if (!LangOpts.CPlusPlus14)
809 return false;
810
811 // In C++14, "s", "h", "min", "ms", "us", and "ns" are used in the library.
812 // Per tweaked N3660, "il", "i", and "if" are also used in the library.
813 // In C++2a "d" and "y" are used in the library.
814 return llvm::StringSwitch<bool>(Suffix)
815 .Cases("h", "min", "s", true)
816 .Cases("ms", "us", "ns", true)
817 .Cases("il", "i", "if", true)
818 .Cases("d", "y", LangOpts.CPlusPlus2a)
819 .Default(false);
820 }
821
checkSeparator(SourceLocation TokLoc,const char * Pos,CheckSeparatorKind IsAfterDigits)822 void NumericLiteralParser::checkSeparator(SourceLocation TokLoc,
823 const char *Pos,
824 CheckSeparatorKind IsAfterDigits) {
825 if (IsAfterDigits == CSK_AfterDigits) {
826 if (Pos == ThisTokBegin)
827 return;
828 --Pos;
829 } else if (Pos == ThisTokEnd)
830 return;
831
832 if (isDigitSeparator(*Pos)) {
833 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Pos - ThisTokBegin),
834 diag::err_digit_separator_not_between_digits)
835 << IsAfterDigits;
836 hadError = true;
837 }
838 }
839
840 /// ParseNumberStartingWithZero - This method is called when the first character
841 /// of the number is found to be a zero. This means it is either an octal
842 /// number (like '04') or a hex number ('0x123a') a binary number ('0b1010') or
843 /// a floating point number (01239.123e4). Eat the prefix, determining the
844 /// radix etc.
ParseNumberStartingWithZero(SourceLocation TokLoc)845 void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
846 assert(s[0] == '0' && "Invalid method call");
847 s++;
848
849 int c1 = s[0];
850
851 // Handle a hex number like 0x1234.
852 if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) {
853 s++;
854 assert(s < ThisTokEnd && "didn't maximally munch?");
855 radix = 16;
856 DigitsBegin = s;
857 s = SkipHexDigits(s);
858 bool HasSignificandDigits = containsDigits(DigitsBegin, s);
859 if (s == ThisTokEnd) {
860 // Done.
861 } else if (*s == '.') {
862 s++;
863 saw_period = true;
864 const char *floatDigitsBegin = s;
865 s = SkipHexDigits(s);
866 if (containsDigits(floatDigitsBegin, s))
867 HasSignificandDigits = true;
868 if (HasSignificandDigits)
869 checkSeparator(TokLoc, floatDigitsBegin, CSK_BeforeDigits);
870 }
871
872 if (!HasSignificandDigits) {
873 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin),
874 diag::err_hex_constant_requires)
875 << PP.getLangOpts().CPlusPlus << 1;
876 hadError = true;
877 return;
878 }
879
880 // A binary exponent can appear with or with a '.'. If dotted, the
881 // binary exponent is required.
882 if (*s == 'p' || *s == 'P') {
883 checkSeparator(TokLoc, s, CSK_AfterDigits);
884 const char *Exponent = s;
885 s++;
886 saw_exponent = true;
887 if (s != ThisTokEnd && (*s == '+' || *s == '-')) s++; // sign
888 const char *first_non_digit = SkipDigits(s);
889 if (!containsDigits(s, first_non_digit)) {
890 if (!hadError) {
891 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, Exponent-ThisTokBegin),
892 diag::err_exponent_has_no_digits);
893 hadError = true;
894 }
895 return;
896 }
897 checkSeparator(TokLoc, s, CSK_BeforeDigits);
898 s = first_non_digit;
899
900 if (!PP.getLangOpts().HexFloats)
901 PP.Diag(TokLoc, PP.getLangOpts().CPlusPlus
902 ? diag::ext_hex_literal_invalid
903 : diag::ext_hex_constant_invalid);
904 else if (PP.getLangOpts().CPlusPlus17)
905 PP.Diag(TokLoc, diag::warn_cxx17_hex_literal);
906 } else if (saw_period) {
907 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s - ThisTokBegin),
908 diag::err_hex_constant_requires)
909 << PP.getLangOpts().CPlusPlus << 0;
910 hadError = true;
911 }
912 return;
913 }
914
915 // Handle simple binary numbers 0b01010
916 if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) {
917 // 0b101010 is a C++1y / GCC extension.
918 PP.Diag(TokLoc,
919 PP.getLangOpts().CPlusPlus14
920 ? diag::warn_cxx11_compat_binary_literal
921 : PP.getLangOpts().CPlusPlus
922 ? diag::ext_binary_literal_cxx14
923 : diag::ext_binary_literal);
924 ++s;
925 assert(s < ThisTokEnd && "didn't maximally munch?");
926 radix = 2;
927 DigitsBegin = s;
928 s = SkipBinaryDigits(s);
929 if (s == ThisTokEnd) {
930 // Done.
931 } else if (isHexDigit(*s) &&
932 !isValidUDSuffix(PP.getLangOpts(),
933 StringRef(s, ThisTokEnd - s))) {
934 PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, s-ThisTokBegin),
935 diag::err_invalid_digit) << StringRef(s, 1) << 2;
936 hadError = true;
937 }
938 // Other suffixes will be diagnosed by the caller.
939 return;
940 }
941
942 // For now, the radix is set to 8. If we discover that we have a
943 // floating point constant, the radix will change to 10. Octal floating
944 // point constants are not permitted (only decimal and hexadecimal).
945 radix = 8;
946 DigitsBegin = s;
947 s = SkipOctalDigits(s);
948 if (s == ThisTokEnd)
949 return; // Done, simple octal number like 01234
950
951 // If we have some other non-octal digit that *is* a decimal digit, see if
952 // this is part of a floating point number like 094.123 or 09e1.
953 if (isDigit(*s)) {
954 const char *EndDecimal = SkipDigits(s);
955 if (EndDecimal[0] == '.' || EndDecimal[0] == 'e' || EndDecimal[0] == 'E') {
956 s = EndDecimal;
957 radix = 10;
958 }
959 }
960
961 ParseDecimalOrOctalCommon(TokLoc);
962 }
963
alwaysFitsInto64Bits(unsigned Radix,unsigned NumDigits)964 static bool alwaysFitsInto64Bits(unsigned Radix, unsigned NumDigits) {
965 switch (Radix) {
966 case 2:
967 return NumDigits <= 64;
968 case 8:
969 return NumDigits <= 64 / 3; // Digits are groups of 3 bits.
970 case 10:
971 return NumDigits <= 19; // floor(log10(2^64))
972 case 16:
973 return NumDigits <= 64 / 4; // Digits are groups of 4 bits.
974 default:
975 llvm_unreachable("impossible Radix");
976 }
977 }
978
979 /// GetIntegerValue - Convert this numeric literal value to an APInt that
980 /// matches Val's input width. If there is an overflow, set Val to the low bits
981 /// of the result and return true. Otherwise, return false.
GetIntegerValue(llvm::APInt & Val)982 bool NumericLiteralParser::GetIntegerValue(llvm::APInt &Val) {
983 // Fast path: Compute a conservative bound on the maximum number of
984 // bits per digit in this radix. If we can't possibly overflow a
985 // uint64 based on that bound then do the simple conversion to
986 // integer. This avoids the expensive overflow checking below, and
987 // handles the common cases that matter (small decimal integers and
988 // hex/octal values which don't overflow).
989 const unsigned NumDigits = SuffixBegin - DigitsBegin;
990 if (alwaysFitsInto64Bits(radix, NumDigits)) {
991 uint64_t N = 0;
992 for (const char *Ptr = DigitsBegin; Ptr != SuffixBegin; ++Ptr)
993 if (!isDigitSeparator(*Ptr))
994 N = N * radix + llvm::hexDigitValue(*Ptr);
995
996 // This will truncate the value to Val's input width. Simply check
997 // for overflow by comparing.
998 Val = N;
999 return Val.getZExtValue() != N;
1000 }
1001
1002 Val = 0;
1003 const char *Ptr = DigitsBegin;
1004
1005 llvm::APInt RadixVal(Val.getBitWidth(), radix);
1006 llvm::APInt CharVal(Val.getBitWidth(), 0);
1007 llvm::APInt OldVal = Val;
1008
1009 bool OverflowOccurred = false;
1010 while (Ptr < SuffixBegin) {
1011 if (isDigitSeparator(*Ptr)) {
1012 ++Ptr;
1013 continue;
1014 }
1015
1016 unsigned C = llvm::hexDigitValue(*Ptr++);
1017
1018 // If this letter is out of bound for this radix, reject it.
1019 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1020
1021 CharVal = C;
1022
1023 // Add the digit to the value in the appropriate radix. If adding in digits
1024 // made the value smaller, then this overflowed.
1025 OldVal = Val;
1026
1027 // Multiply by radix, did overflow occur on the multiply?
1028 Val *= RadixVal;
1029 OverflowOccurred |= Val.udiv(RadixVal) != OldVal;
1030
1031 // Add value, did overflow occur on the value?
1032 // (a + b) ult b <=> overflow
1033 Val += CharVal;
1034 OverflowOccurred |= Val.ult(CharVal);
1035 }
1036 return OverflowOccurred;
1037 }
1038
1039 llvm::APFloat::opStatus
GetFloatValue(llvm::APFloat & Result)1040 NumericLiteralParser::GetFloatValue(llvm::APFloat &Result) {
1041 using llvm::APFloat;
1042
1043 unsigned n = std::min(SuffixBegin - ThisTokBegin, ThisTokEnd - ThisTokBegin);
1044
1045 llvm::SmallString<16> Buffer;
1046 StringRef Str(ThisTokBegin, n);
1047 if (Str.find('\'') != StringRef::npos) {
1048 Buffer.reserve(n);
1049 std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer),
1050 &isDigitSeparator);
1051 Str = Buffer;
1052 }
1053
1054 auto StatusOrErr =
1055 Result.convertFromString(Str, APFloat::rmNearestTiesToEven);
1056 assert(StatusOrErr && "Invalid floating point representation");
1057 return !errorToBool(StatusOrErr.takeError()) ? *StatusOrErr
1058 : APFloat::opInvalidOp;
1059 }
1060
IsExponentPart(char c)1061 static inline bool IsExponentPart(char c) {
1062 return c == 'p' || c == 'P' || c == 'e' || c == 'E';
1063 }
1064
GetFixedPointValue(llvm::APInt & StoreVal,unsigned Scale)1065 bool NumericLiteralParser::GetFixedPointValue(llvm::APInt &StoreVal, unsigned Scale) {
1066 assert(radix == 16 || radix == 10);
1067
1068 // Find how many digits are needed to store the whole literal.
1069 unsigned NumDigits = SuffixBegin - DigitsBegin;
1070 if (saw_period) --NumDigits;
1071
1072 // Initial scan of the exponent if it exists
1073 bool ExpOverflowOccurred = false;
1074 bool NegativeExponent = false;
1075 const char *ExponentBegin;
1076 uint64_t Exponent = 0;
1077 int64_t BaseShift = 0;
1078 if (saw_exponent) {
1079 const char *Ptr = DigitsBegin;
1080
1081 while (!IsExponentPart(*Ptr)) ++Ptr;
1082 ExponentBegin = Ptr;
1083 ++Ptr;
1084 NegativeExponent = *Ptr == '-';
1085 if (NegativeExponent) ++Ptr;
1086
1087 unsigned NumExpDigits = SuffixBegin - Ptr;
1088 if (alwaysFitsInto64Bits(radix, NumExpDigits)) {
1089 llvm::StringRef ExpStr(Ptr, NumExpDigits);
1090 llvm::APInt ExpInt(/*numBits=*/64, ExpStr, /*radix=*/10);
1091 Exponent = ExpInt.getZExtValue();
1092 } else {
1093 ExpOverflowOccurred = true;
1094 }
1095
1096 if (NegativeExponent) BaseShift -= Exponent;
1097 else BaseShift += Exponent;
1098 }
1099
1100 // Number of bits needed for decimal literal is
1101 // ceil(NumDigits * log2(10)) Integral part
1102 // + Scale Fractional part
1103 // + ceil(Exponent * log2(10)) Exponent
1104 // --------------------------------------------------
1105 // ceil((NumDigits + Exponent) * log2(10)) + Scale
1106 //
1107 // But for simplicity in handling integers, we can round up log2(10) to 4,
1108 // making:
1109 // 4 * (NumDigits + Exponent) + Scale
1110 //
1111 // Number of digits needed for hexadecimal literal is
1112 // 4 * NumDigits Integral part
1113 // + Scale Fractional part
1114 // + Exponent Exponent
1115 // --------------------------------------------------
1116 // (4 * NumDigits) + Scale + Exponent
1117 uint64_t NumBitsNeeded;
1118 if (radix == 10)
1119 NumBitsNeeded = 4 * (NumDigits + Exponent) + Scale;
1120 else
1121 NumBitsNeeded = 4 * NumDigits + Exponent + Scale;
1122
1123 if (NumBitsNeeded > std::numeric_limits<unsigned>::max())
1124 ExpOverflowOccurred = true;
1125 llvm::APInt Val(static_cast<unsigned>(NumBitsNeeded), 0, /*isSigned=*/false);
1126
1127 bool FoundDecimal = false;
1128
1129 int64_t FractBaseShift = 0;
1130 const char *End = saw_exponent ? ExponentBegin : SuffixBegin;
1131 for (const char *Ptr = DigitsBegin; Ptr < End; ++Ptr) {
1132 if (*Ptr == '.') {
1133 FoundDecimal = true;
1134 continue;
1135 }
1136
1137 // Normal reading of an integer
1138 unsigned C = llvm::hexDigitValue(*Ptr);
1139 assert(C < radix && "NumericLiteralParser ctor should have rejected this");
1140
1141 Val *= radix;
1142 Val += C;
1143
1144 if (FoundDecimal)
1145 // Keep track of how much we will need to adjust this value by from the
1146 // number of digits past the radix point.
1147 --FractBaseShift;
1148 }
1149
1150 // For a radix of 16, we will be multiplying by 2 instead of 16.
1151 if (radix == 16) FractBaseShift *= 4;
1152 BaseShift += FractBaseShift;
1153
1154 Val <<= Scale;
1155
1156 uint64_t Base = (radix == 16) ? 2 : 10;
1157 if (BaseShift > 0) {
1158 for (int64_t i = 0; i < BaseShift; ++i) {
1159 Val *= Base;
1160 }
1161 } else if (BaseShift < 0) {
1162 for (int64_t i = BaseShift; i < 0 && !Val.isNullValue(); ++i)
1163 Val = Val.udiv(Base);
1164 }
1165
1166 bool IntOverflowOccurred = false;
1167 auto MaxVal = llvm::APInt::getMaxValue(StoreVal.getBitWidth());
1168 if (Val.getBitWidth() > StoreVal.getBitWidth()) {
1169 IntOverflowOccurred |= Val.ugt(MaxVal.zext(Val.getBitWidth()));
1170 StoreVal = Val.trunc(StoreVal.getBitWidth());
1171 } else if (Val.getBitWidth() < StoreVal.getBitWidth()) {
1172 IntOverflowOccurred |= Val.zext(MaxVal.getBitWidth()).ugt(MaxVal);
1173 StoreVal = Val.zext(StoreVal.getBitWidth());
1174 } else {
1175 StoreVal = Val;
1176 }
1177
1178 return IntOverflowOccurred || ExpOverflowOccurred;
1179 }
1180
1181 /// \verbatim
1182 /// user-defined-character-literal: [C++11 lex.ext]
1183 /// character-literal ud-suffix
1184 /// ud-suffix:
1185 /// identifier
1186 /// character-literal: [C++11 lex.ccon]
1187 /// ' c-char-sequence '
1188 /// u' c-char-sequence '
1189 /// U' c-char-sequence '
1190 /// L' c-char-sequence '
1191 /// u8' c-char-sequence ' [C++1z lex.ccon]
1192 /// c-char-sequence:
1193 /// c-char
1194 /// c-char-sequence c-char
1195 /// c-char:
1196 /// any member of the source character set except the single-quote ',
1197 /// backslash \, or new-line character
1198 /// escape-sequence
1199 /// universal-character-name
1200 /// escape-sequence:
1201 /// simple-escape-sequence
1202 /// octal-escape-sequence
1203 /// hexadecimal-escape-sequence
1204 /// simple-escape-sequence:
1205 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1206 /// octal-escape-sequence:
1207 /// \ octal-digit
1208 /// \ octal-digit octal-digit
1209 /// \ octal-digit octal-digit octal-digit
1210 /// hexadecimal-escape-sequence:
1211 /// \x hexadecimal-digit
1212 /// hexadecimal-escape-sequence hexadecimal-digit
1213 /// universal-character-name: [C++11 lex.charset]
1214 /// \u hex-quad
1215 /// \U hex-quad hex-quad
1216 /// hex-quad:
1217 /// hex-digit hex-digit hex-digit hex-digit
1218 /// \endverbatim
1219 ///
CharLiteralParser(const char * begin,const char * end,SourceLocation Loc,Preprocessor & PP,tok::TokenKind kind)1220 CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
1221 SourceLocation Loc, Preprocessor &PP,
1222 tok::TokenKind kind) {
1223 // At this point we know that the character matches the regex "(L|u|U)?'.*'".
1224 HadError = false;
1225
1226 Kind = kind;
1227
1228 const char *TokBegin = begin;
1229
1230 // Skip over wide character determinant.
1231 if (Kind != tok::char_constant)
1232 ++begin;
1233 if (Kind == tok::utf8_char_constant)
1234 ++begin;
1235
1236 // Skip over the entry quote.
1237 assert(begin[0] == '\'' && "Invalid token lexed");
1238 ++begin;
1239
1240 // Remove an optional ud-suffix.
1241 if (end[-1] != '\'') {
1242 const char *UDSuffixEnd = end;
1243 do {
1244 --end;
1245 } while (end[-1] != '\'');
1246 // FIXME: Don't bother with this if !tok.hasUCN().
1247 expandUCNs(UDSuffixBuf, StringRef(end, UDSuffixEnd - end));
1248 UDSuffixOffset = end - TokBegin;
1249 }
1250
1251 // Trim the ending quote.
1252 assert(end != begin && "Invalid token lexed");
1253 --end;
1254
1255 // FIXME: The "Value" is an uint64_t so we can handle char literals of
1256 // up to 64-bits.
1257 // FIXME: This extensively assumes that 'char' is 8-bits.
1258 assert(PP.getTargetInfo().getCharWidth() == 8 &&
1259 "Assumes char is 8 bits");
1260 assert(PP.getTargetInfo().getIntWidth() <= 64 &&
1261 (PP.getTargetInfo().getIntWidth() & 7) == 0 &&
1262 "Assumes sizeof(int) on target is <= 64 and a multiple of char");
1263 assert(PP.getTargetInfo().getWCharWidth() <= 64 &&
1264 "Assumes sizeof(wchar) on target is <= 64");
1265
1266 SmallVector<uint32_t, 4> codepoint_buffer;
1267 codepoint_buffer.resize(end - begin);
1268 uint32_t *buffer_begin = &codepoint_buffer.front();
1269 uint32_t *buffer_end = buffer_begin + codepoint_buffer.size();
1270
1271 // Unicode escapes representing characters that cannot be correctly
1272 // represented in a single code unit are disallowed in character literals
1273 // by this implementation.
1274 uint32_t largest_character_for_kind;
1275 if (tok::wide_char_constant == Kind) {
1276 largest_character_for_kind =
1277 0xFFFFFFFFu >> (32-PP.getTargetInfo().getWCharWidth());
1278 } else if (tok::utf8_char_constant == Kind) {
1279 largest_character_for_kind = 0x7F;
1280 } else if (tok::utf16_char_constant == Kind) {
1281 largest_character_for_kind = 0xFFFF;
1282 } else if (tok::utf32_char_constant == Kind) {
1283 largest_character_for_kind = 0x10FFFF;
1284 } else {
1285 largest_character_for_kind = 0x7Fu;
1286 }
1287
1288 while (begin != end) {
1289 // Is this a span of non-escape characters?
1290 if (begin[0] != '\\') {
1291 char const *start = begin;
1292 do {
1293 ++begin;
1294 } while (begin != end && *begin != '\\');
1295
1296 char const *tmp_in_start = start;
1297 uint32_t *tmp_out_start = buffer_begin;
1298 llvm::ConversionResult res =
1299 llvm::ConvertUTF8toUTF32(reinterpret_cast<llvm::UTF8 const **>(&start),
1300 reinterpret_cast<llvm::UTF8 const *>(begin),
1301 &buffer_begin, buffer_end, llvm::strictConversion);
1302 if (res != llvm::conversionOK) {
1303 // If we see bad encoding for unprefixed character literals, warn and
1304 // simply copy the byte values, for compatibility with gcc and
1305 // older versions of clang.
1306 bool NoErrorOnBadEncoding = isAscii();
1307 unsigned Msg = diag::err_bad_character_encoding;
1308 if (NoErrorOnBadEncoding)
1309 Msg = diag::warn_bad_character_encoding;
1310 PP.Diag(Loc, Msg);
1311 if (NoErrorOnBadEncoding) {
1312 start = tmp_in_start;
1313 buffer_begin = tmp_out_start;
1314 for (; start != begin; ++start, ++buffer_begin)
1315 *buffer_begin = static_cast<uint8_t>(*start);
1316 } else {
1317 HadError = true;
1318 }
1319 } else {
1320 for (; tmp_out_start < buffer_begin; ++tmp_out_start) {
1321 if (*tmp_out_start > largest_character_for_kind) {
1322 HadError = true;
1323 PP.Diag(Loc, diag::err_character_too_large);
1324 }
1325 }
1326 }
1327
1328 continue;
1329 }
1330 // Is this a Universal Character Name escape?
1331 if (begin[1] == 'u' || begin[1] == 'U') {
1332 unsigned short UcnLen = 0;
1333 if (!ProcessUCNEscape(TokBegin, begin, end, *buffer_begin, UcnLen,
1334 FullSourceLoc(Loc, PP.getSourceManager()),
1335 &PP.getDiagnostics(), PP.getLangOpts(), true)) {
1336 HadError = true;
1337 } else if (*buffer_begin > largest_character_for_kind) {
1338 HadError = true;
1339 PP.Diag(Loc, diag::err_character_too_large);
1340 }
1341
1342 ++buffer_begin;
1343 continue;
1344 }
1345 unsigned CharWidth = getCharWidth(Kind, PP.getTargetInfo());
1346 uint64_t result =
1347 ProcessCharEscape(TokBegin, begin, end, HadError,
1348 FullSourceLoc(Loc,PP.getSourceManager()),
1349 CharWidth, &PP.getDiagnostics(), PP.getLangOpts());
1350 *buffer_begin++ = result;
1351 }
1352
1353 unsigned NumCharsSoFar = buffer_begin - &codepoint_buffer.front();
1354
1355 if (NumCharsSoFar > 1) {
1356 if (isWide())
1357 PP.Diag(Loc, diag::warn_extraneous_char_constant);
1358 else if (isAscii() && NumCharsSoFar == 4)
1359 PP.Diag(Loc, diag::ext_four_char_character_literal);
1360 else if (isAscii())
1361 PP.Diag(Loc, diag::ext_multichar_character_literal);
1362 else
1363 PP.Diag(Loc, diag::err_multichar_utf_character_literal);
1364 IsMultiChar = true;
1365 } else {
1366 IsMultiChar = false;
1367 }
1368
1369 llvm::APInt LitVal(PP.getTargetInfo().getIntWidth(), 0);
1370
1371 // Narrow character literals act as though their value is concatenated
1372 // in this implementation, but warn on overflow.
1373 bool multi_char_too_long = false;
1374 if (isAscii() && isMultiChar()) {
1375 LitVal = 0;
1376 for (size_t i = 0; i < NumCharsSoFar; ++i) {
1377 // check for enough leading zeros to shift into
1378 multi_char_too_long |= (LitVal.countLeadingZeros() < 8);
1379 LitVal <<= 8;
1380 LitVal = LitVal + (codepoint_buffer[i] & 0xFF);
1381 }
1382 } else if (NumCharsSoFar > 0) {
1383 // otherwise just take the last character
1384 LitVal = buffer_begin[-1];
1385 }
1386
1387 if (!HadError && multi_char_too_long) {
1388 PP.Diag(Loc, diag::warn_char_constant_too_large);
1389 }
1390
1391 // Transfer the value from APInt to uint64_t
1392 Value = LitVal.getZExtValue();
1393
1394 // If this is a single narrow character, sign extend it (e.g. '\xFF' is "-1")
1395 // if 'char' is signed for this target (C99 6.4.4.4p10). Note that multiple
1396 // character constants are not sign extended in the this implementation:
1397 // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC.
1398 if (isAscii() && NumCharsSoFar == 1 && (Value & 128) &&
1399 PP.getLangOpts().CharIsSigned)
1400 Value = (signed char)Value;
1401 }
1402
1403 /// \verbatim
1404 /// string-literal: [C++0x lex.string]
1405 /// encoding-prefix " [s-char-sequence] "
1406 /// encoding-prefix R raw-string
1407 /// encoding-prefix:
1408 /// u8
1409 /// u
1410 /// U
1411 /// L
1412 /// s-char-sequence:
1413 /// s-char
1414 /// s-char-sequence s-char
1415 /// s-char:
1416 /// any member of the source character set except the double-quote ",
1417 /// backslash \, or new-line character
1418 /// escape-sequence
1419 /// universal-character-name
1420 /// raw-string:
1421 /// " d-char-sequence ( r-char-sequence ) d-char-sequence "
1422 /// r-char-sequence:
1423 /// r-char
1424 /// r-char-sequence r-char
1425 /// r-char:
1426 /// any member of the source character set, except a right parenthesis )
1427 /// followed by the initial d-char-sequence (which may be empty)
1428 /// followed by a double quote ".
1429 /// d-char-sequence:
1430 /// d-char
1431 /// d-char-sequence d-char
1432 /// d-char:
1433 /// any member of the basic source character set except:
1434 /// space, the left parenthesis (, the right parenthesis ),
1435 /// the backslash \, and the control characters representing horizontal
1436 /// tab, vertical tab, form feed, and newline.
1437 /// escape-sequence: [C++0x lex.ccon]
1438 /// simple-escape-sequence
1439 /// octal-escape-sequence
1440 /// hexadecimal-escape-sequence
1441 /// simple-escape-sequence:
1442 /// one of \' \" \? \\ \a \b \f \n \r \t \v
1443 /// octal-escape-sequence:
1444 /// \ octal-digit
1445 /// \ octal-digit octal-digit
1446 /// \ octal-digit octal-digit octal-digit
1447 /// hexadecimal-escape-sequence:
1448 /// \x hexadecimal-digit
1449 /// hexadecimal-escape-sequence hexadecimal-digit
1450 /// universal-character-name:
1451 /// \u hex-quad
1452 /// \U hex-quad hex-quad
1453 /// hex-quad:
1454 /// hex-digit hex-digit hex-digit hex-digit
1455 /// \endverbatim
1456 ///
1457 StringLiteralParser::
StringLiteralParser(ArrayRef<Token> StringToks,Preprocessor & PP,bool Complain)1458 StringLiteralParser(ArrayRef<Token> StringToks,
1459 Preprocessor &PP, bool Complain)
1460 : SM(PP.getSourceManager()), Features(PP.getLangOpts()),
1461 Target(PP.getTargetInfo()), Diags(Complain ? &PP.getDiagnostics() :nullptr),
1462 MaxTokenLength(0), SizeBound(0), CharByteWidth(0), Kind(tok::unknown),
1463 ResultPtr(ResultBuf.data()), hadError(false), Pascal(false) {
1464 init(StringToks);
1465 }
1466
init(ArrayRef<Token> StringToks)1467 void StringLiteralParser::init(ArrayRef<Token> StringToks){
1468 // The literal token may have come from an invalid source location (e.g. due
1469 // to a PCH error), in which case the token length will be 0.
1470 if (StringToks.empty() || StringToks[0].getLength() < 2)
1471 return DiagnoseLexingError(SourceLocation());
1472
1473 // Scan all of the string portions, remember the max individual token length,
1474 // computing a bound on the concatenated string length, and see whether any
1475 // piece is a wide-string. If any of the string portions is a wide-string
1476 // literal, the result is a wide-string literal [C99 6.4.5p4].
1477 assert(!StringToks.empty() && "expected at least one token");
1478 MaxTokenLength = StringToks[0].getLength();
1479 assert(StringToks[0].getLength() >= 2 && "literal token is invalid!");
1480 SizeBound = StringToks[0].getLength()-2; // -2 for "".
1481 Kind = StringToks[0].getKind();
1482
1483 hadError = false;
1484
1485 // Implement Translation Phase #6: concatenation of string literals
1486 /// (C99 5.1.1.2p1). The common case is only one string fragment.
1487 for (unsigned i = 1; i != StringToks.size(); ++i) {
1488 if (StringToks[i].getLength() < 2)
1489 return DiagnoseLexingError(StringToks[i].getLocation());
1490
1491 // The string could be shorter than this if it needs cleaning, but this is a
1492 // reasonable bound, which is all we need.
1493 assert(StringToks[i].getLength() >= 2 && "literal token is invalid!");
1494 SizeBound += StringToks[i].getLength()-2; // -2 for "".
1495
1496 // Remember maximum string piece length.
1497 if (StringToks[i].getLength() > MaxTokenLength)
1498 MaxTokenLength = StringToks[i].getLength();
1499
1500 // Remember if we see any wide or utf-8/16/32 strings.
1501 // Also check for illegal concatenations.
1502 if (StringToks[i].isNot(Kind) && StringToks[i].isNot(tok::string_literal)) {
1503 if (isAscii()) {
1504 Kind = StringToks[i].getKind();
1505 } else {
1506 if (Diags)
1507 Diags->Report(StringToks[i].getLocation(),
1508 diag::err_unsupported_string_concat);
1509 hadError = true;
1510 }
1511 }
1512 }
1513
1514 // Include space for the null terminator.
1515 ++SizeBound;
1516
1517 // TODO: K&R warning: "traditional C rejects string constant concatenation"
1518
1519 // Get the width in bytes of char/wchar_t/char16_t/char32_t
1520 CharByteWidth = getCharWidth(Kind, Target);
1521 assert((CharByteWidth & 7) == 0 && "Assumes character size is byte multiple");
1522 CharByteWidth /= 8;
1523
1524 // The output buffer size needs to be large enough to hold wide characters.
1525 // This is a worst-case assumption which basically corresponds to L"" "long".
1526 SizeBound *= CharByteWidth;
1527
1528 // Size the temporary buffer to hold the result string data.
1529 ResultBuf.resize(SizeBound);
1530
1531 // Likewise, but for each string piece.
1532 SmallString<512> TokenBuf;
1533 TokenBuf.resize(MaxTokenLength);
1534
1535 // Loop over all the strings, getting their spelling, and expanding them to
1536 // wide strings as appropriate.
1537 ResultPtr = &ResultBuf[0]; // Next byte to fill in.
1538
1539 Pascal = false;
1540
1541 SourceLocation UDSuffixTokLoc;
1542
1543 for (unsigned i = 0, e = StringToks.size(); i != e; ++i) {
1544 const char *ThisTokBuf = &TokenBuf[0];
1545 // Get the spelling of the token, which eliminates trigraphs, etc. We know
1546 // that ThisTokBuf points to a buffer that is big enough for the whole token
1547 // and 'spelled' tokens can only shrink.
1548 bool StringInvalid = false;
1549 unsigned ThisTokLen =
1550 Lexer::getSpelling(StringToks[i], ThisTokBuf, SM, Features,
1551 &StringInvalid);
1552 if (StringInvalid)
1553 return DiagnoseLexingError(StringToks[i].getLocation());
1554
1555 const char *ThisTokBegin = ThisTokBuf;
1556 const char *ThisTokEnd = ThisTokBuf+ThisTokLen;
1557
1558 // Remove an optional ud-suffix.
1559 if (ThisTokEnd[-1] != '"') {
1560 const char *UDSuffixEnd = ThisTokEnd;
1561 do {
1562 --ThisTokEnd;
1563 } while (ThisTokEnd[-1] != '"');
1564
1565 StringRef UDSuffix(ThisTokEnd, UDSuffixEnd - ThisTokEnd);
1566
1567 if (UDSuffixBuf.empty()) {
1568 if (StringToks[i].hasUCN())
1569 expandUCNs(UDSuffixBuf, UDSuffix);
1570 else
1571 UDSuffixBuf.assign(UDSuffix);
1572 UDSuffixToken = i;
1573 UDSuffixOffset = ThisTokEnd - ThisTokBuf;
1574 UDSuffixTokLoc = StringToks[i].getLocation();
1575 } else {
1576 SmallString<32> ExpandedUDSuffix;
1577 if (StringToks[i].hasUCN()) {
1578 expandUCNs(ExpandedUDSuffix, UDSuffix);
1579 UDSuffix = ExpandedUDSuffix;
1580 }
1581
1582 // C++11 [lex.ext]p8: At the end of phase 6, if a string literal is the
1583 // result of a concatenation involving at least one user-defined-string-
1584 // literal, all the participating user-defined-string-literals shall
1585 // have the same ud-suffix.
1586 if (UDSuffixBuf != UDSuffix) {
1587 if (Diags) {
1588 SourceLocation TokLoc = StringToks[i].getLocation();
1589 Diags->Report(TokLoc, diag::err_string_concat_mixed_suffix)
1590 << UDSuffixBuf << UDSuffix
1591 << SourceRange(UDSuffixTokLoc, UDSuffixTokLoc)
1592 << SourceRange(TokLoc, TokLoc);
1593 }
1594 hadError = true;
1595 }
1596 }
1597 }
1598
1599 // Strip the end quote.
1600 --ThisTokEnd;
1601
1602 // TODO: Input character set mapping support.
1603
1604 // Skip marker for wide or unicode strings.
1605 if (ThisTokBuf[0] == 'L' || ThisTokBuf[0] == 'u' || ThisTokBuf[0] == 'U') {
1606 ++ThisTokBuf;
1607 // Skip 8 of u8 marker for utf8 strings.
1608 if (ThisTokBuf[0] == '8')
1609 ++ThisTokBuf;
1610 }
1611
1612 // Check for raw string
1613 if (ThisTokBuf[0] == 'R') {
1614 ThisTokBuf += 2; // skip R"
1615
1616 const char *Prefix = ThisTokBuf;
1617 while (ThisTokBuf[0] != '(')
1618 ++ThisTokBuf;
1619 ++ThisTokBuf; // skip '('
1620
1621 // Remove same number of characters from the end
1622 ThisTokEnd -= ThisTokBuf - Prefix;
1623 assert(ThisTokEnd >= ThisTokBuf && "malformed raw string literal");
1624
1625 // C++14 [lex.string]p4: A source-file new-line in a raw string literal
1626 // results in a new-line in the resulting execution string-literal.
1627 StringRef RemainingTokenSpan(ThisTokBuf, ThisTokEnd - ThisTokBuf);
1628 while (!RemainingTokenSpan.empty()) {
1629 // Split the string literal on \r\n boundaries.
1630 size_t CRLFPos = RemainingTokenSpan.find("\r\n");
1631 StringRef BeforeCRLF = RemainingTokenSpan.substr(0, CRLFPos);
1632 StringRef AfterCRLF = RemainingTokenSpan.substr(CRLFPos);
1633
1634 // Copy everything before the \r\n sequence into the string literal.
1635 if (CopyStringFragment(StringToks[i], ThisTokBegin, BeforeCRLF))
1636 hadError = true;
1637
1638 // Point into the \n inside the \r\n sequence and operate on the
1639 // remaining portion of the literal.
1640 RemainingTokenSpan = AfterCRLF.substr(1);
1641 }
1642 } else {
1643 if (ThisTokBuf[0] != '"') {
1644 // The file may have come from PCH and then changed after loading the
1645 // PCH; Fail gracefully.
1646 return DiagnoseLexingError(StringToks[i].getLocation());
1647 }
1648 ++ThisTokBuf; // skip "
1649
1650 // Check if this is a pascal string
1651 if (Features.PascalStrings && ThisTokBuf + 1 != ThisTokEnd &&
1652 ThisTokBuf[0] == '\\' && ThisTokBuf[1] == 'p') {
1653
1654 // If the \p sequence is found in the first token, we have a pascal string
1655 // Otherwise, if we already have a pascal string, ignore the first \p
1656 if (i == 0) {
1657 ++ThisTokBuf;
1658 Pascal = true;
1659 } else if (Pascal)
1660 ThisTokBuf += 2;
1661 }
1662
1663 while (ThisTokBuf != ThisTokEnd) {
1664 // Is this a span of non-escape characters?
1665 if (ThisTokBuf[0] != '\\') {
1666 const char *InStart = ThisTokBuf;
1667 do {
1668 ++ThisTokBuf;
1669 } while (ThisTokBuf != ThisTokEnd && ThisTokBuf[0] != '\\');
1670
1671 // Copy the character span over.
1672 if (CopyStringFragment(StringToks[i], ThisTokBegin,
1673 StringRef(InStart, ThisTokBuf - InStart)))
1674 hadError = true;
1675 continue;
1676 }
1677 // Is this a Universal Character Name escape?
1678 if (ThisTokBuf[1] == 'u' || ThisTokBuf[1] == 'U') {
1679 EncodeUCNEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd,
1680 ResultPtr, hadError,
1681 FullSourceLoc(StringToks[i].getLocation(), SM),
1682 CharByteWidth, Diags, Features);
1683 continue;
1684 }
1685 // Otherwise, this is a non-UCN escape character. Process it.
1686 unsigned ResultChar =
1687 ProcessCharEscape(ThisTokBegin, ThisTokBuf, ThisTokEnd, hadError,
1688 FullSourceLoc(StringToks[i].getLocation(), SM),
1689 CharByteWidth*8, Diags, Features);
1690
1691 if (CharByteWidth == 4) {
1692 // FIXME: Make the type of the result buffer correct instead of
1693 // using reinterpret_cast.
1694 llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultPtr);
1695 *ResultWidePtr = ResultChar;
1696 ResultPtr += 4;
1697 } else if (CharByteWidth == 2) {
1698 // FIXME: Make the type of the result buffer correct instead of
1699 // using reinterpret_cast.
1700 llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultPtr);
1701 *ResultWidePtr = ResultChar & 0xFFFF;
1702 ResultPtr += 2;
1703 } else {
1704 assert(CharByteWidth == 1 && "Unexpected char width");
1705 *ResultPtr++ = ResultChar & 0xFF;
1706 }
1707 }
1708 }
1709 }
1710
1711 if (Pascal) {
1712 if (CharByteWidth == 4) {
1713 // FIXME: Make the type of the result buffer correct instead of
1714 // using reinterpret_cast.
1715 llvm::UTF32 *ResultWidePtr = reinterpret_cast<llvm::UTF32*>(ResultBuf.data());
1716 ResultWidePtr[0] = GetNumStringChars() - 1;
1717 } else if (CharByteWidth == 2) {
1718 // FIXME: Make the type of the result buffer correct instead of
1719 // using reinterpret_cast.
1720 llvm::UTF16 *ResultWidePtr = reinterpret_cast<llvm::UTF16*>(ResultBuf.data());
1721 ResultWidePtr[0] = GetNumStringChars() - 1;
1722 } else {
1723 assert(CharByteWidth == 1 && "Unexpected char width");
1724 ResultBuf[0] = GetNumStringChars() - 1;
1725 }
1726
1727 // Verify that pascal strings aren't too large.
1728 if (GetStringLength() > 256) {
1729 if (Diags)
1730 Diags->Report(StringToks.front().getLocation(),
1731 diag::err_pascal_string_too_long)
1732 << SourceRange(StringToks.front().getLocation(),
1733 StringToks.back().getLocation());
1734 hadError = true;
1735 return;
1736 }
1737 } else if (Diags) {
1738 // Complain if this string literal has too many characters.
1739 unsigned MaxChars = Features.CPlusPlus? 65536 : Features.C99 ? 4095 : 509;
1740
1741 if (GetNumStringChars() > MaxChars)
1742 Diags->Report(StringToks.front().getLocation(),
1743 diag::ext_string_too_long)
1744 << GetNumStringChars() << MaxChars
1745 << (Features.CPlusPlus ? 2 : Features.C99 ? 1 : 0)
1746 << SourceRange(StringToks.front().getLocation(),
1747 StringToks.back().getLocation());
1748 }
1749 }
1750
resyncUTF8(const char * Err,const char * End)1751 static const char *resyncUTF8(const char *Err, const char *End) {
1752 if (Err == End)
1753 return End;
1754 End = Err + std::min<unsigned>(llvm::getNumBytesForUTF8(*Err), End-Err);
1755 while (++Err != End && (*Err & 0xC0) == 0x80)
1756 ;
1757 return Err;
1758 }
1759
1760 /// This function copies from Fragment, which is a sequence of bytes
1761 /// within Tok's contents (which begin at TokBegin) into ResultPtr.
1762 /// Performs widening for multi-byte characters.
CopyStringFragment(const Token & Tok,const char * TokBegin,StringRef Fragment)1763 bool StringLiteralParser::CopyStringFragment(const Token &Tok,
1764 const char *TokBegin,
1765 StringRef Fragment) {
1766 const llvm::UTF8 *ErrorPtrTmp;
1767 if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp))
1768 return false;
1769
1770 // If we see bad encoding for unprefixed string literals, warn and
1771 // simply copy the byte values, for compatibility with gcc and older
1772 // versions of clang.
1773 bool NoErrorOnBadEncoding = isAscii();
1774 if (NoErrorOnBadEncoding) {
1775 memcpy(ResultPtr, Fragment.data(), Fragment.size());
1776 ResultPtr += Fragment.size();
1777 }
1778
1779 if (Diags) {
1780 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1781
1782 FullSourceLoc SourceLoc(Tok.getLocation(), SM);
1783 const DiagnosticBuilder &Builder =
1784 Diag(Diags, Features, SourceLoc, TokBegin,
1785 ErrorPtr, resyncUTF8(ErrorPtr, Fragment.end()),
1786 NoErrorOnBadEncoding ? diag::warn_bad_string_encoding
1787 : diag::err_bad_string_encoding);
1788
1789 const char *NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1790 StringRef NextFragment(NextStart, Fragment.end()-NextStart);
1791
1792 // Decode into a dummy buffer.
1793 SmallString<512> Dummy;
1794 Dummy.reserve(Fragment.size() * CharByteWidth);
1795 char *Ptr = Dummy.data();
1796
1797 while (!ConvertUTF8toWide(CharByteWidth, NextFragment, Ptr, ErrorPtrTmp)) {
1798 const char *ErrorPtr = reinterpret_cast<const char *>(ErrorPtrTmp);
1799 NextStart = resyncUTF8(ErrorPtr, Fragment.end());
1800 Builder << MakeCharSourceRange(Features, SourceLoc, TokBegin,
1801 ErrorPtr, NextStart);
1802 NextFragment = StringRef(NextStart, Fragment.end()-NextStart);
1803 }
1804 }
1805 return !NoErrorOnBadEncoding;
1806 }
1807
DiagnoseLexingError(SourceLocation Loc)1808 void StringLiteralParser::DiagnoseLexingError(SourceLocation Loc) {
1809 hadError = true;
1810 if (Diags)
1811 Diags->Report(Loc, diag::err_lexing_string);
1812 }
1813
1814 /// getOffsetOfStringByte - This function returns the offset of the
1815 /// specified byte of the string data represented by Token. This handles
1816 /// advancing over escape sequences in the string.
getOffsetOfStringByte(const Token & Tok,unsigned ByteNo) const1817 unsigned StringLiteralParser::getOffsetOfStringByte(const Token &Tok,
1818 unsigned ByteNo) const {
1819 // Get the spelling of the token.
1820 SmallString<32> SpellingBuffer;
1821 SpellingBuffer.resize(Tok.getLength());
1822
1823 bool StringInvalid = false;
1824 const char *SpellingPtr = &SpellingBuffer[0];
1825 unsigned TokLen = Lexer::getSpelling(Tok, SpellingPtr, SM, Features,
1826 &StringInvalid);
1827 if (StringInvalid)
1828 return 0;
1829
1830 const char *SpellingStart = SpellingPtr;
1831 const char *SpellingEnd = SpellingPtr+TokLen;
1832
1833 // Handle UTF-8 strings just like narrow strings.
1834 if (SpellingPtr[0] == 'u' && SpellingPtr[1] == '8')
1835 SpellingPtr += 2;
1836
1837 assert(SpellingPtr[0] != 'L' && SpellingPtr[0] != 'u' &&
1838 SpellingPtr[0] != 'U' && "Doesn't handle wide or utf strings yet");
1839
1840 // For raw string literals, this is easy.
1841 if (SpellingPtr[0] == 'R') {
1842 assert(SpellingPtr[1] == '"' && "Should be a raw string literal!");
1843 // Skip 'R"'.
1844 SpellingPtr += 2;
1845 while (*SpellingPtr != '(') {
1846 ++SpellingPtr;
1847 assert(SpellingPtr < SpellingEnd && "Missing ( for raw string literal");
1848 }
1849 // Skip '('.
1850 ++SpellingPtr;
1851 return SpellingPtr - SpellingStart + ByteNo;
1852 }
1853
1854 // Skip over the leading quote
1855 assert(SpellingPtr[0] == '"' && "Should be a string literal!");
1856 ++SpellingPtr;
1857
1858 // Skip over bytes until we find the offset we're looking for.
1859 while (ByteNo) {
1860 assert(SpellingPtr < SpellingEnd && "Didn't find byte offset!");
1861
1862 // Step over non-escapes simply.
1863 if (*SpellingPtr != '\\') {
1864 ++SpellingPtr;
1865 --ByteNo;
1866 continue;
1867 }
1868
1869 // Otherwise, this is an escape character. Advance over it.
1870 bool HadError = false;
1871 if (SpellingPtr[1] == 'u' || SpellingPtr[1] == 'U') {
1872 const char *EscapePtr = SpellingPtr;
1873 unsigned Len = MeasureUCNEscape(SpellingStart, SpellingPtr, SpellingEnd,
1874 1, Features, HadError);
1875 if (Len > ByteNo) {
1876 // ByteNo is somewhere within the escape sequence.
1877 SpellingPtr = EscapePtr;
1878 break;
1879 }
1880 ByteNo -= Len;
1881 } else {
1882 ProcessCharEscape(SpellingStart, SpellingPtr, SpellingEnd, HadError,
1883 FullSourceLoc(Tok.getLocation(), SM),
1884 CharByteWidth*8, Diags, Features);
1885 --ByteNo;
1886 }
1887 assert(!HadError && "This method isn't valid on erroneous strings");
1888 }
1889
1890 return SpellingPtr-SpellingStart;
1891 }
1892
1893 /// Determine whether a suffix is a valid ud-suffix. We avoid treating reserved
1894 /// suffixes as ud-suffixes, because the diagnostic experience is better if we
1895 /// treat it as an invalid suffix.
isValidUDSuffix(const LangOptions & LangOpts,StringRef Suffix)1896 bool StringLiteralParser::isValidUDSuffix(const LangOptions &LangOpts,
1897 StringRef Suffix) {
1898 return NumericLiteralParser::isValidUDSuffix(LangOpts, Suffix) ||
1899 Suffix == "sv";
1900 }
1901