* MAINTAINERS: add myself
[official-gcc.git] / libcpp / charset.c
blob5a1c929d83516f2c131a151a10ba7e1855b07fdd
1 /* CPP Library - charsets
2 Copyright (C) 1998-2015 Free Software Foundation, Inc.
4 Broken out of c-lex.c Apr 2003, adding valid C99 UCN ranges.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms of the GNU General Public License as published by the
8 Free Software Foundation; either version 3, or (at your option) any
9 later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 #include "config.h"
21 #include "system.h"
22 #include "cpplib.h"
23 #include "internal.h"
25 /* Character set handling for C-family languages.
27 Terminological note: In what follows, "charset" or "character set"
28 will be taken to mean both an abstract set of characters and an
29 encoding for that set.
31 The C99 standard discusses two character sets: source and execution.
32 The source character set is used for internal processing in translation
33 phases 1 through 4; the execution character set is used thereafter.
34 Both are required by 5.2.1.2p1 to be multibyte encodings, not wide
35 character encodings (see 3.7.2, 3.7.3 for the standardese meanings
36 of these terms). Furthermore, the "basic character set" (listed in
37 5.2.1p3) is to be encoded in each with values one byte wide, and is
38 to appear in the initial shift state.
40 It is not explicitly mentioned, but there is also a "wide execution
41 character set" used to encode wide character constants and wide
42 string literals; this is supposed to be the result of applying the
43 standard library function mbstowcs() to an equivalent narrow string
44 (6.4.5p5). However, the behavior of hexadecimal and octal
45 \-escapes is at odds with this; they are supposed to be translated
46 directly to wchar_t values (6.4.4.4p5,6).
48 The source character set is not necessarily the character set used
49 to encode physical source files on disk; translation phase 1 converts
50 from whatever that encoding is to the source character set.
52 The presence of universal character names in C99 (6.4.3 et seq.)
53 forces the source character set to be isomorphic to ISO 10646,
54 that is, Unicode. There is no such constraint on the execution
55 character set; note also that the conversion from source to
56 execution character set does not occur for identifiers (5.1.1.2p1#5).
58 For convenience of implementation, the source character set's
59 encoding of the basic character set should be identical to the
60 execution character set OF THE HOST SYSTEM's encoding of the basic
61 character set, and it should not be a state-dependent encoding.
63 cpplib uses UTF-8 or UTF-EBCDIC for the source character set,
64 depending on whether the host is based on ASCII or EBCDIC (see
65 respectively Unicode section 2.3/ISO10646 Amendment 2, and Unicode
66 Technical Report #16). With limited exceptions, it relies on the
67 system library's iconv() primitive to do charset conversion
68 (specified in SUSv2). */
70 #if !HAVE_ICONV
71 /* Make certain that the uses of iconv(), iconv_open(), iconv_close()
72 below, which are guarded only by if statements with compile-time
73 constant conditions, do not cause link errors. */
74 #define iconv_open(x, y) (errno = EINVAL, (iconv_t)-1)
75 #define iconv(a,b,c,d,e) (errno = EINVAL, (size_t)-1)
76 #define iconv_close(x) (void)0
77 #define ICONV_CONST
78 #endif
80 #if HOST_CHARSET == HOST_CHARSET_ASCII
81 #define SOURCE_CHARSET "UTF-8"
82 #define LAST_POSSIBLY_BASIC_SOURCE_CHAR 0x7e
83 #elif HOST_CHARSET == HOST_CHARSET_EBCDIC
84 #define SOURCE_CHARSET "UTF-EBCDIC"
85 #define LAST_POSSIBLY_BASIC_SOURCE_CHAR 0xFF
86 #else
87 #error "Unrecognized basic host character set"
88 #endif
90 #ifndef EILSEQ
91 #define EILSEQ EINVAL
92 #endif
94 /* This structure is used for a resizable string buffer throughout. */
95 /* Don't call it strbuf, as that conflicts with unistd.h on systems
96 such as DYNIX/ptx where unistd.h includes stropts.h. */
97 struct _cpp_strbuf
98 {
99 uchar *text;
100 size_t asize;
101 size_t len;
102 };
104 /* This is enough to hold any string that fits on a single 80-column
105 line, even if iconv quadruples its size (e.g. conversion from
106 ASCII to UTF-32) rounded up to a power of two. */
107 #define OUTBUF_BLOCK_SIZE 256
109 /* Conversions between UTF-8 and UTF-16/32 are implemented by custom
110 logic. This is because a depressing number of systems lack iconv,
111 or have have iconv libraries that do not do these conversions, so
112 we need a fallback implementation for them. To ensure the fallback
113 doesn't break due to neglect, it is used on all systems.
115 UTF-32 encoding is nice and simple: a four-byte binary number,
116 constrained to the range 00000000-7FFFFFFF to avoid questions of
117 signedness. We do have to cope with big- and little-endian
118 variants.
120 UTF-16 encoding uses two-byte binary numbers, again in big- and
121 little-endian variants, for all values in the 00000000-0000FFFF
122 range. Values in the 00010000-0010FFFF range are encoded as pairs
123 of two-byte numbers, called "surrogate pairs": given a number S in
124 this range, it is mapped to a pair (H, L) as follows:
126 H = (S - 0x10000) / 0x400 + 0xD800
127 L = (S - 0x10000) % 0x400 + 0xDC00
129 Two-byte values in the D800...DFFF range are ill-formed except as a
130 component of a surrogate pair. Even if the encoding within a
131 two-byte value is little-endian, the H member of the surrogate pair
132 comes first.
134 There is no way to encode values in the 00110000-7FFFFFFF range,
135 which is not currently a problem as there are no assigned code
136 points in that range; however, the author expects that it will
137 eventually become necessary to abandon UTF-16 due to this
138 limitation. Note also that, because of these pairs, UTF-16 does
139 not meet the requirements of the C standard for a wide character
140 encoding (see 3.7.3 and 6.4.4.4p11).
142 UTF-8 encoding looks like this:
144 value range encoded as
145 00000000-0000007F 0xxxxxxx
146 00000080-000007FF 110xxxxx 10xxxxxx
147 00000800-0000FFFF 1110xxxx 10xxxxxx 10xxxxxx
148 00010000-001FFFFF 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
149 00200000-03FFFFFF 111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
150 04000000-7FFFFFFF 1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
152 Values in the 0000D800 ... 0000DFFF range (surrogates) are invalid,
153 which means that three-byte sequences ED xx yy, with A0 <= xx <= BF,
154 never occur. Note also that any value that can be encoded by a
155 given row of the table can also be encoded by all successive rows,
156 but this is not done; only the shortest possible encoding for any
157 given value is valid. For instance, the character 07C0 could be
158 encoded as any of DF 80, E0 9F 80, F0 80 9F 80, F8 80 80 9F 80, or
159 FC 80 80 80 9F 80. Only the first is valid.
161 An implementation note: the transformation from UTF-16 to UTF-8, or
162 vice versa, is easiest done by using UTF-32 as an intermediary. */
164 /* Internal primitives which go from an UTF-8 byte stream to native-endian
165 UTF-32 in a cppchar_t, or vice versa; this avoids an extra marshal/unmarshal
166 operation in several places below. */
167 static inline int
168 one_utf8_to_cppchar (const uchar **inbufp, size_t *inbytesleftp,
169 cppchar_t *cp)
171 static const uchar masks[6] = { 0x7F, 0x1F, 0x0F, 0x07, 0x03, 0x01 };
172 static const uchar patns[6] = { 0x00, 0xC0, 0xE0, 0xF0, 0xF8, 0xFC };
174 cppchar_t c;
175 const uchar *inbuf = *inbufp;
176 size_t nbytes, i;
178 if (*inbytesleftp < 1)
179 return EINVAL;
181 c = *inbuf;
182 if (c < 0x80)
184 *cp = c;
185 *inbytesleftp -= 1;
186 *inbufp += 1;
187 return 0;
190 /* The number of leading 1-bits in the first byte indicates how many
191 bytes follow. */
192 for (nbytes = 2; nbytes < 7; nbytes++)
193 if ((c & ~masks[nbytes-1]) == patns[nbytes-1])
194 goto found;
195 return EILSEQ;
196 found:
198 if (*inbytesleftp < nbytes)
199 return EINVAL;
201 c = (c & masks[nbytes-1]);
202 inbuf++;
203 for (i = 1; i < nbytes; i++)
205 cppchar_t n = *inbuf++;
206 if ((n & 0xC0) != 0x80)
207 return EILSEQ;
208 c = ((c << 6) + (n & 0x3F));
211 /* Make sure the shortest possible encoding was used. */
212 if (c <= 0x7F && nbytes > 1) return EILSEQ;
213 if (c <= 0x7FF && nbytes > 2) return EILSEQ;
214 if (c <= 0xFFFF && nbytes > 3) return EILSEQ;
215 if (c <= 0x1FFFFF && nbytes > 4) return EILSEQ;
216 if (c <= 0x3FFFFFF && nbytes > 5) return EILSEQ;
218 /* Make sure the character is valid. */
219 if (c > 0x7FFFFFFF || (c >= 0xD800 && c <= 0xDFFF)) return EILSEQ;
221 *cp = c;
222 *inbufp = inbuf;
223 *inbytesleftp -= nbytes;
224 return 0;
227 static inline int
228 one_cppchar_to_utf8 (cppchar_t c, uchar **outbufp, size_t *outbytesleftp)
230 static const uchar masks[6] = { 0x00, 0xC0, 0xE0, 0xF0, 0xF8, 0xFC };
231 static const uchar limits[6] = { 0x80, 0xE0, 0xF0, 0xF8, 0xFC, 0xFE };
232 size_t nbytes;
233 uchar buf[6], *p = &buf[6];
234 uchar *outbuf = *outbufp;
236 nbytes = 1;
237 if (c < 0x80)
238 *--p = c;
239 else
241 do
243 *--p = ((c & 0x3F) | 0x80);
244 c >>= 6;
245 nbytes++;
247 while (c >= 0x3F || (c & limits[nbytes-1]));
248 *--p = (c | masks[nbytes-1]);
251 if (*outbytesleftp < nbytes)
252 return E2BIG;
254 while (p < &buf[6])
255 *outbuf++ = *p++;
256 *outbytesleftp -= nbytes;
257 *outbufp = outbuf;
258 return 0;
261 /* The following four functions transform one character between the two
262 encodings named in the function name. All have the signature
263 int (*)(iconv_t bigend, const uchar **inbufp, size_t *inbytesleftp,
264 uchar **outbufp, size_t *outbytesleftp)
266 BIGEND must have the value 0 or 1, coerced to (iconv_t); it is
267 interpreted as a boolean indicating whether big-endian or
268 little-endian encoding is to be used for the member of the pair
269 that is not UTF-8.
271 INBUFP, INBYTESLEFTP, OUTBUFP, OUTBYTESLEFTP work exactly as they
272 do for iconv.
274 The return value is either 0 for success, or an errno value for
275 failure, which may be E2BIG (need more space), EILSEQ (ill-formed
276 input sequence), ir EINVAL (incomplete input sequence). */
278 static inline int
279 one_utf8_to_utf32 (iconv_t bigend, const uchar **inbufp, size_t *inbytesleftp,
280 uchar **outbufp, size_t *outbytesleftp)
282 uchar *outbuf;
283 cppchar_t s = 0;
284 int rval;
286 /* Check for space first, since we know exactly how much we need. */
287 if (*outbytesleftp < 4)
288 return E2BIG;
290 rval = one_utf8_to_cppchar (inbufp, inbytesleftp, &s);
291 if (rval)
292 return rval;
294 outbuf = *outbufp;
295 outbuf[bigend ? 3 : 0] = (s & 0x000000FF);
296 outbuf[bigend ? 2 : 1] = (s & 0x0000FF00) >> 8;
297 outbuf[bigend ? 1 : 2] = (s & 0x00FF0000) >> 16;
298 outbuf[bigend ? 0 : 3] = (s & 0xFF000000) >> 24;
300 *outbufp += 4;
301 *outbytesleftp -= 4;
302 return 0;
305 static inline int
306 one_utf32_to_utf8 (iconv_t bigend, const uchar **inbufp, size_t *inbytesleftp,
307 uchar **outbufp, size_t *outbytesleftp)
309 cppchar_t s;
310 int rval;
311 const uchar *inbuf;
313 if (*inbytesleftp < 4)
314 return EINVAL;
316 inbuf = *inbufp;
318 s = inbuf[bigend ? 0 : 3] << 24;
319 s += inbuf[bigend ? 1 : 2] << 16;
320 s += inbuf[bigend ? 2 : 1] << 8;
321 s += inbuf[bigend ? 3 : 0];
323 if (s >= 0x7FFFFFFF || (s >= 0xD800 && s <= 0xDFFF))
324 return EILSEQ;
326 rval = one_cppchar_to_utf8 (s, outbufp, outbytesleftp);
327 if (rval)
328 return rval;
330 *inbufp += 4;
331 *inbytesleftp -= 4;
332 return 0;
335 static inline int
336 one_utf8_to_utf16 (iconv_t bigend, const uchar **inbufp, size_t *inbytesleftp,
337 uchar **outbufp, size_t *outbytesleftp)
339 int rval;
340 cppchar_t s = 0;
341 const uchar *save_inbuf = *inbufp;
342 size_t save_inbytesleft = *inbytesleftp;
343 uchar *outbuf = *outbufp;
345 rval = one_utf8_to_cppchar (inbufp, inbytesleftp, &s);
346 if (rval)
347 return rval;
349 if (s > 0x0010FFFF)
351 *inbufp = save_inbuf;
352 *inbytesleftp = save_inbytesleft;
353 return EILSEQ;
356 if (s <= 0xFFFF)
358 if (*outbytesleftp < 2)
360 *inbufp = save_inbuf;
361 *inbytesleftp = save_inbytesleft;
362 return E2BIG;
364 outbuf[bigend ? 1 : 0] = (s & 0x00FF);
365 outbuf[bigend ? 0 : 1] = (s & 0xFF00) >> 8;
367 *outbufp += 2;
368 *outbytesleftp -= 2;
369 return 0;
371 else
373 cppchar_t hi, lo;
375 if (*outbytesleftp < 4)
377 *inbufp = save_inbuf;
378 *inbytesleftp = save_inbytesleft;
379 return E2BIG;
382 hi = (s - 0x10000) / 0x400 + 0xD800;
383 lo = (s - 0x10000) % 0x400 + 0xDC00;
385 /* Even if we are little-endian, put the high surrogate first.
386 ??? Matches practice? */
387 outbuf[bigend ? 1 : 0] = (hi & 0x00FF);
388 outbuf[bigend ? 0 : 1] = (hi & 0xFF00) >> 8;
389 outbuf[bigend ? 3 : 2] = (lo & 0x00FF);
390 outbuf[bigend ? 2 : 3] = (lo & 0xFF00) >> 8;
392 *outbufp += 4;
393 *outbytesleftp -= 4;
394 return 0;
398 static inline int
399 one_utf16_to_utf8 (iconv_t bigend, const uchar **inbufp, size_t *inbytesleftp,
400 uchar **outbufp, size_t *outbytesleftp)
402 cppchar_t s;
403 const uchar *inbuf = *inbufp;
404 int rval;
406 if (*inbytesleftp < 2)
407 return EINVAL;
408 s = inbuf[bigend ? 0 : 1] << 8;
409 s += inbuf[bigend ? 1 : 0];
411 /* Low surrogate without immediately preceding high surrogate is invalid. */
412 if (s >= 0xDC00 && s <= 0xDFFF)
413 return EILSEQ;
414 /* High surrogate must have a following low surrogate. */
415 else if (s >= 0xD800 && s <= 0xDBFF)
417 cppchar_t hi = s, lo;
418 if (*inbytesleftp < 4)
419 return EINVAL;
421 lo = inbuf[bigend ? 2 : 3] << 8;
422 lo += inbuf[bigend ? 3 : 2];
424 if (lo < 0xDC00 || lo > 0xDFFF)
425 return EILSEQ;
427 s = (hi - 0xD800) * 0x400 + (lo - 0xDC00) + 0x10000;
430 rval = one_cppchar_to_utf8 (s, outbufp, outbytesleftp);
431 if (rval)
432 return rval;
434 /* Success - update the input pointers (one_cppchar_to_utf8 has done
435 the output pointers for us). */
436 if (s <= 0xFFFF)
438 *inbufp += 2;
439 *inbytesleftp -= 2;
441 else
443 *inbufp += 4;
444 *inbytesleftp -= 4;
446 return 0;
449 /* Helper routine for the next few functions. The 'const' on
450 one_conversion means that we promise not to modify what function is
451 pointed to, which lets the inliner see through it. */
453 static inline bool
454 conversion_loop (int (*const one_conversion)(iconv_t, const uchar **, size_t *,
455 uchar **, size_t *),
456 iconv_t cd, const uchar *from, size_t flen, struct _cpp_strbuf *to)
458 const uchar *inbuf;
459 uchar *outbuf;
460 size_t inbytesleft, outbytesleft;
461 int rval;
463 inbuf = from;
464 inbytesleft = flen;
465 outbuf = to->text + to->len;
466 outbytesleft = to->asize - to->len;
468 for (;;)
470 do
471 rval = one_conversion (cd, &inbuf, &inbytesleft,
472 &outbuf, &outbytesleft);
473 while (inbytesleft && !rval);
475 if (__builtin_expect (inbytesleft == 0, 1))
477 to->len = to->asize - outbytesleft;
478 return true;
480 if (rval != E2BIG)
482 errno = rval;
483 return false;
486 outbytesleft += OUTBUF_BLOCK_SIZE;
487 to->asize += OUTBUF_BLOCK_SIZE;
488 to->text = XRESIZEVEC (uchar, to->text, to->asize);
489 outbuf = to->text + to->asize - outbytesleft;
494 /* These functions convert entire strings between character sets.
495 They all have the signature
497 bool (*)(iconv_t cd, const uchar *from, size_t flen, struct _cpp_strbuf *to);
499 The input string FROM is converted as specified by the function
500 name plus the iconv descriptor CD (which may be fake), and the
501 result appended to TO. On any error, false is returned, otherwise true. */
503 /* These four use the custom conversion code above. */
504 static bool
505 convert_utf8_utf16 (iconv_t cd, const uchar *from, size_t flen,
506 struct _cpp_strbuf *to)
508 return conversion_loop (one_utf8_to_utf16, cd, from, flen, to);
511 static bool
512 convert_utf8_utf32 (iconv_t cd, const uchar *from, size_t flen,
513 struct _cpp_strbuf *to)
515 return conversion_loop (one_utf8_to_utf32, cd, from, flen, to);
518 static bool
519 convert_utf16_utf8 (iconv_t cd, const uchar *from, size_t flen,
520 struct _cpp_strbuf *to)
522 return conversion_loop (one_utf16_to_utf8, cd, from, flen, to);
525 static bool
526 convert_utf32_utf8 (iconv_t cd, const uchar *from, size_t flen,
527 struct _cpp_strbuf *to)
529 return conversion_loop (one_utf32_to_utf8, cd, from, flen, to);
532 /* Identity conversion, used when we have no alternative. */
533 static bool
534 convert_no_conversion (iconv_t cd ATTRIBUTE_UNUSED,
535 const uchar *from, size_t flen, struct _cpp_strbuf *to)
537 if (to->len + flen > to->asize)
539 to->asize = to->len + flen;
540 to->asize += to->asize / 4;
541 to->text = XRESIZEVEC (uchar, to->text, to->asize);
543 memcpy (to->text + to->len, from, flen);
544 to->len += flen;
545 return true;
548 /* And this one uses the system iconv primitive. It's a little
549 different, since iconv's interface is a little different. */
550 #if HAVE_ICONV
552 #define CONVERT_ICONV_GROW_BUFFER \
553 do { \
554 outbytesleft += OUTBUF_BLOCK_SIZE; \
555 to->asize += OUTBUF_BLOCK_SIZE; \
556 to->text = XRESIZEVEC (uchar, to->text, to->asize); \
557 outbuf = (char *)to->text + to->asize - outbytesleft; \
558 } while (0)
560 static bool
561 convert_using_iconv (iconv_t cd, const uchar *from, size_t flen,
562 struct _cpp_strbuf *to)
564 ICONV_CONST char *inbuf;
565 char *outbuf;
566 size_t inbytesleft, outbytesleft;
568 /* Reset conversion descriptor and check that it is valid. */
569 if (iconv (cd, 0, 0, 0, 0) == (size_t)-1)
570 return false;
572 inbuf = (ICONV_CONST char *)from;
573 inbytesleft = flen;
574 outbuf = (char *)to->text + to->len;
575 outbytesleft = to->asize - to->len;
577 for (;;)
579 iconv (cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft);
580 if (__builtin_expect (inbytesleft == 0, 1))
582 /* Close out any shift states, returning to the initial state. */
583 if (iconv (cd, 0, 0, &outbuf, &outbytesleft) == (size_t)-1)
585 if (errno != E2BIG)
586 return false;
588 CONVERT_ICONV_GROW_BUFFER;
589 if (iconv (cd, 0, 0, &outbuf, &outbytesleft) == (size_t)-1)
590 return false;
593 to->len = to->asize - outbytesleft;
594 return true;
596 if (errno != E2BIG)
597 return false;
599 CONVERT_ICONV_GROW_BUFFER;
602 #else
603 #define convert_using_iconv 0 /* prevent undefined symbol error below */
604 #endif
606 /* Arrange for the above custom conversion logic to be used automatically
607 when conversion between a suitable pair of character sets is requested. */
609 #define APPLY_CONVERSION(CONVERTER, FROM, FLEN, TO) \
610 CONVERTER.func (CONVERTER.cd, FROM, FLEN, TO)
612 struct cpp_conversion
614 const char *pair;
615 convert_f func;
616 iconv_t fake_cd;
617 };
618 static const struct cpp_conversion conversion_tab[] = {
619 { "UTF-8/UTF-32LE", convert_utf8_utf32, (iconv_t)0 },
620 { "UTF-8/UTF-32BE", convert_utf8_utf32, (iconv_t)1 },
621 { "UTF-8/UTF-16LE", convert_utf8_utf16, (iconv_t)0 },
622 { "UTF-8/UTF-16BE", convert_utf8_utf16, (iconv_t)1 },
623 { "UTF-32LE/UTF-8", convert_utf32_utf8, (iconv_t)0 },
624 { "UTF-32BE/UTF-8", convert_utf32_utf8, (iconv_t)1 },
625 { "UTF-16LE/UTF-8", convert_utf16_utf8, (iconv_t)0 },
626 { "UTF-16BE/UTF-8", convert_utf16_utf8, (iconv_t)1 },
627 };
629 /* Subroutine of cpp_init_iconv: initialize and return a
630 cset_converter structure for conversion from FROM to TO. If
631 iconv_open() fails, issue an error and return an identity
632 converter. Silently return an identity converter if FROM and TO
633 are identical. */
634 static struct cset_converter
635 init_iconv_desc (cpp_reader *pfile, const char *to, const char *from)
637 struct cset_converter ret;
638 char *pair;
639 size_t i;
641 if (!strcasecmp (to, from))
643 ret.func = convert_no_conversion;
644 ret.cd = (iconv_t) -1;
645 ret.width = -1;
646 return ret;
649 pair = (char *) alloca(strlen(to) + strlen(from) + 2);
651 strcpy(pair, from);
652 strcat(pair, "/");
653 strcat(pair, to);
654 for (i = 0; i < ARRAY_SIZE (conversion_tab); i++)
655 if (!strcasecmp (pair, conversion_tab[i].pair))
657 ret.func = conversion_tab[i].func;
658 ret.cd = conversion_tab[i].fake_cd;
659 ret.width = -1;
660 return ret;
663 /* No custom converter - try iconv. */
664 if (HAVE_ICONV)
666 ret.func = convert_using_iconv;
667 ret.cd = iconv_open (to, from);
668 ret.width = -1;
670 if (ret.cd == (iconv_t) -1)
672 if (errno == EINVAL)
673 cpp_error (pfile, CPP_DL_ERROR, /* FIXME should be DL_SORRY */
674 "conversion from %s to %s not supported by iconv",
675 from, to);
676 else
677 cpp_errno (pfile, CPP_DL_ERROR, "iconv_open");
679 ret.func = convert_no_conversion;
682 else
684 cpp_error (pfile, CPP_DL_ERROR, /* FIXME: should be DL_SORRY */
685 "no iconv implementation, cannot convert from %s to %s",
686 from, to);
687 ret.func = convert_no_conversion;
688 ret.cd = (iconv_t) -1;
689 ret.width = -1;
691 return ret;
694 /* If charset conversion is requested, initialize iconv(3) descriptors
695 for conversion from the source character set to the execution
696 character sets. If iconv is not present in the C library, and
697 conversion is requested, issue an error. */
699 void
700 cpp_init_iconv (cpp_reader *pfile)
702 const char *ncset = CPP_OPTION (pfile, narrow_charset);
703 const char *wcset = CPP_OPTION (pfile, wide_charset);
704 const char *default_wcset;
706 bool be = CPP_OPTION (pfile, bytes_big_endian);
708 if (CPP_OPTION (pfile, wchar_precision) >= 32)
709 default_wcset = be ? "UTF-32BE" : "UTF-32LE";
710 else if (CPP_OPTION (pfile, wchar_precision) >= 16)
711 default_wcset = be ? "UTF-16BE" : "UTF-16LE";
712 else
713 /* This effectively means that wide strings are not supported,
714 so don't do any conversion at all. */
715 default_wcset = SOURCE_CHARSET;
717 if (!ncset)
718 ncset = SOURCE_CHARSET;
719 if (!wcset)
720 wcset = default_wcset;
722 pfile->narrow_cset_desc = init_iconv_desc (pfile, ncset, SOURCE_CHARSET);
723 pfile->narrow_cset_desc.width = CPP_OPTION (pfile, char_precision);
724 pfile->utf8_cset_desc = init_iconv_desc (pfile, "UTF-8", SOURCE_CHARSET);
725 pfile->utf8_cset_desc.width = CPP_OPTION (pfile, char_precision);
726 pfile->char16_cset_desc = init_iconv_desc (pfile,
727 be ? "UTF-16BE" : "UTF-16LE",
728 SOURCE_CHARSET);
729 pfile->char16_cset_desc.width = 16;
730 pfile->char32_cset_desc = init_iconv_desc (pfile,
731 be ? "UTF-32BE" : "UTF-32LE",
732 SOURCE_CHARSET);
733 pfile->char32_cset_desc.width = 32;
734 pfile->wide_cset_desc = init_iconv_desc (pfile, wcset, SOURCE_CHARSET);
735 pfile->wide_cset_desc.width = CPP_OPTION (pfile, wchar_precision);
738 /* Destroy iconv(3) descriptors set up by cpp_init_iconv, if necessary. */
739 void
740 _cpp_destroy_iconv (cpp_reader *pfile)
742 if (HAVE_ICONV)
744 if (pfile->narrow_cset_desc.func == convert_using_iconv)
745 iconv_close (pfile->narrow_cset_desc.cd);
746 if (pfile->utf8_cset_desc.func == convert_using_iconv)
747 iconv_close (pfile->utf8_cset_desc.cd);
748 if (pfile->char16_cset_desc.func == convert_using_iconv)
749 iconv_close (pfile->char16_cset_desc.cd);
750 if (pfile->char32_cset_desc.func == convert_using_iconv)
751 iconv_close (pfile->char32_cset_desc.cd);
752 if (pfile->wide_cset_desc.func == convert_using_iconv)
753 iconv_close (pfile->wide_cset_desc.cd);
757 /* Utility routine for use by a full compiler. C is a character taken
758 from the *basic* source character set, encoded in the host's
759 execution encoding. Convert it to (the target's) execution
760 encoding, and return that value.
762 Issues an internal error if C's representation in the narrow
763 execution character set fails to be a single-byte value (C99
764 5.2.1p3: "The representation of each member of the source and
765 execution character sets shall fit in a byte.") May also issue an
766 internal error if C fails to be a member of the basic source
767 character set (testing this exactly is too hard, especially when
768 the host character set is EBCDIC). */
769 cppchar_t
770 cpp_host_to_exec_charset (cpp_reader *pfile, cppchar_t c)
772 uchar sbuf[1];
773 struct _cpp_strbuf tbuf;
775 /* This test is merely an approximation, but it suffices to catch
776 the most important thing, which is that we don't get handed a
777 character outside the unibyte range of the host character set. */
778 if (c > LAST_POSSIBLY_BASIC_SOURCE_CHAR)
780 cpp_error (pfile, CPP_DL_ICE,
781 "character 0x%lx is not in the basic source character set\n",
782 (unsigned long)c);
783 return 0;
786 /* Being a character in the unibyte range of the host character set,
787 we can safely splat it into a one-byte buffer and trust that that
788 is a well-formed string. */
789 sbuf[0] = c;
791 /* This should never need to reallocate, but just in case... */
792 tbuf.asize = 1;
793 tbuf.text = XNEWVEC (uchar, tbuf.asize);
794 tbuf.len = 0;
796 if (!APPLY_CONVERSION (pfile->narrow_cset_desc, sbuf, 1, &tbuf))
798 cpp_errno (pfile, CPP_DL_ICE, "converting to execution character set");
799 return 0;
801 if (tbuf.len != 1)
803 cpp_error (pfile, CPP_DL_ICE,
804 "character 0x%lx is not unibyte in execution character set",
805 (unsigned long)c);
806 return 0;
808 c = tbuf.text[0];
809 free(tbuf.text);
810 return c;
813 \f
815 /* Utility routine that computes a mask of the form 0000...111... with
816 WIDTH 1-bits. */
817 static inline size_t
818 width_to_mask (size_t width)
820 width = MIN (width, BITS_PER_CPPCHAR_T);
821 if (width >= CHAR_BIT * sizeof (size_t))
822 return ~(size_t) 0;
823 else
824 return ((size_t) 1 << width) - 1;
827 /* A large table of unicode character information. */
828 enum {
829 /* Valid in a C99 identifier? */
830 C99 = 1,
831 /* Valid in a C99 identifier, but not as the first character? */
832 N99 = 2,
833 /* Valid in a C++ identifier? */
834 CXX = 4,
835 /* Valid in a C11/C++11 identifier? */
836 C11 = 8,
837 /* Valid in a C11/C++11 identifier, but not as the first character? */
838 N11 = 16,
839 /* NFC representation is not valid in an identifier? */
840 CID = 32,
841 /* Might be valid NFC form? */
842 NFC = 64,
843 /* Might be valid NFKC form? */
844 NKC = 128,
845 /* Certain preceding characters might make it not valid NFC/NKFC form? */
846 CTX = 256
847 };
849 struct ucnrange {
850 /* Bitmap of flags above. */
851 unsigned short flags;
852 /* Combining class of the character. */
853 unsigned char combine;
854 /* Last character in the range described by this entry. */
855 unsigned int end;
856 };
857 #include "ucnid.h"
859 /* Returns 1 if C is valid in an identifier, 2 if C is valid except at
860 the start of an identifier, and 0 if C is not valid in an
861 identifier. We assume C has already gone through the checks of
862 _cpp_valid_ucn. Also update NST for C if returning nonzero. The
863 algorithm is a simple binary search on the table defined in
864 ucnid.h. */
866 static int
867 ucn_valid_in_identifier (cpp_reader *pfile, cppchar_t c,
868 struct normalize_state *nst)
870 int mn, mx, md;
871 unsigned short valid_flags, invalid_start_flags;
873 if (c > 0x10FFFF)
874 return 0;
876 mn = 0;
877 mx = ARRAY_SIZE (ucnranges) - 1;
878 while (mx != mn)
880 md = (mn + mx) / 2;
881 if (c <= ucnranges[md].end)
882 mx = md;
883 else
884 mn = md + 1;
887 /* When -pedantic, we require the character to have been listed by
888 the standard for the current language. Otherwise, we accept the
889 union of the acceptable sets for all supported language versions. */
890 valid_flags = C99 | CXX | C11;
891 if (CPP_PEDANTIC (pfile))
893 if (CPP_OPTION (pfile, c11_identifiers))
894 valid_flags = C11;
895 else if (CPP_OPTION (pfile, c99))
896 valid_flags = C99;
897 else if (CPP_OPTION (pfile, cplusplus))
898 valid_flags = CXX;
900 if (! (ucnranges[mn].flags & valid_flags))
901 return 0;
902 if (CPP_OPTION (pfile, c11_identifiers))
903 invalid_start_flags = N11;
904 else if (CPP_OPTION (pfile, c99))
905 invalid_start_flags = N99;
906 else
907 invalid_start_flags = 0;
909 /* Update NST. */
910 if (ucnranges[mn].combine != 0 && ucnranges[mn].combine < nst->prev_class)
911 nst->level = normalized_none;
912 else if (ucnranges[mn].flags & CTX)
914 bool safe;
915 cppchar_t p = nst->previous;
917 /* For Hangul, characters in the range AC00-D7A3 are NFC/NFKC,
918 and are combined algorithmically from a sequence of the form
919 1100-1112 1161-1175 11A8-11C2
920 (if the third is not present, it is treated as 11A7, which is not
921 really a valid character).
922 Unfortunately, C99 allows (only) the NFC form, but C++ allows
923 only the combining characters. */
924 if (c >= 0x1161 && c <= 0x1175)
925 safe = p < 0x1100 || p > 0x1112;
926 else if (c >= 0x11A8 && c <= 0x11C2)
927 safe = (p < 0xAC00 || p > 0xD7A3 || (p - 0xAC00) % 28 != 0);
928 else
929 safe = check_nfc (pfile, c, p);
930 if (!safe)
932 if ((c >= 0x1161 && c <= 0x1175) || (c >= 0x11A8 && c <= 0x11C2))
933 nst->level = MAX (nst->level, normalized_identifier_C);
934 else
935 nst->level = normalized_none;
938 else if (ucnranges[mn].flags & NKC)
940 else if (ucnranges[mn].flags & NFC)
941 nst->level = MAX (nst->level, normalized_C);
942 else if (ucnranges[mn].flags & CID)
943 nst->level = MAX (nst->level, normalized_identifier_C);
944 else
945 nst->level = normalized_none;
946 if (ucnranges[mn].combine == 0)
947 nst->previous = c;
948 nst->prev_class = ucnranges[mn].combine;
950 /* In C99, UCN digits may not begin identifiers. In C11 and C++11,
951 UCN combining characters may not begin identifiers. */
952 if (ucnranges[mn].flags & invalid_start_flags)
953 return 2;
955 return 1;
958 /* [lex.charset]: The character designated by the universal character
959 name \UNNNNNNNN is that character whose character short name in
960 ISO/IEC 10646 is NNNNNNNN; the character designated by the
961 universal character name \uNNNN is that character whose character
962 short name in ISO/IEC 10646 is 0000NNNN. If the hexadecimal value
963 for a universal character name corresponds to a surrogate code point
964 (in the range 0xD800-0xDFFF, inclusive), the program is ill-formed.
965 Additionally, if the hexadecimal value for a universal-character-name
966 outside a character or string literal corresponds to a control character
967 (in either of the ranges 0x00-0x1F or 0x7F-0x9F, both inclusive) or to a
968 character in the basic source character set, the program is ill-formed.
970 C99 6.4.3: A universal character name shall not specify a character
971 whose short identifier is less than 00A0 other than 0024 ($), 0040 (@),
972 or 0060 (`), nor one in the range D800 through DFFF inclusive.
974 *PSTR must be preceded by "\u" or "\U"; it is assumed that the
975 buffer end is delimited by a non-hex digit. Returns false if the
976 UCN has not been consumed, true otherwise.
978 The value of the UCN, whether valid or invalid, is returned in *CP.
979 Diagnostics are emitted for invalid values. PSTR is updated to point
980 one beyond the UCN, or to the syntactically invalid character.
982 IDENTIFIER_POS is 0 when not in an identifier, 1 for the start of
983 an identifier, or 2 otherwise. */
985 bool
986 _cpp_valid_ucn (cpp_reader *pfile, const uchar **pstr,
987 const uchar *limit, int identifier_pos,
988 struct normalize_state *nst, cppchar_t *cp)
990 cppchar_t result, c;
991 unsigned int length;
992 const uchar *str = *pstr;
993 const uchar *base = str - 2;
995 if (!CPP_OPTION (pfile, cplusplus) && !CPP_OPTION (pfile, c99))
996 cpp_error (pfile, CPP_DL_WARNING,
997 "universal character names are only valid in C++ and C99");
998 else if (CPP_OPTION (pfile, cpp_warn_c90_c99_compat) > 0
999 && !CPP_OPTION (pfile, cplusplus))
1000 cpp_error (pfile, CPP_DL_WARNING,
1001 "C99's universal character names are incompatible with C90");
1002 else if (CPP_WTRADITIONAL (pfile) && identifier_pos == 0)
1003 cpp_warning (pfile, CPP_W_TRADITIONAL,
1004 "the meaning of '\\%c' is different in traditional C",
1005 (int) str[-1]);
1007 if (str[-1] == 'u')
1008 length = 4;
1009 else if (str[-1] == 'U')
1010 length = 8;
1011 else
1013 cpp_error (pfile, CPP_DL_ICE, "In _cpp_valid_ucn but not a UCN");
1014 length = 4;
1017 result = 0;
1018 do
1020 c = *str;
1021 if (!ISXDIGIT (c))
1022 break;
1023 str++;
1024 result = (result << 4) + hex_value (c);
1026 while (--length && str < limit);
1028 /* Partial UCNs are not valid in strings, but decompose into
1029 multiple tokens in identifiers, so we can't give a helpful
1030 error message in that case. */
1031 if (length && identifier_pos)
1033 *cp = 0;
1034 return false;
1037 *pstr = str;
1038 if (length)
1040 cpp_error (pfile, CPP_DL_ERROR,
1041 "incomplete universal character name %.*s",
1042 (int) (str - base), base);
1043 result = 1;
1045 /* The C99 standard permits $, @ and ` to be specified as UCNs. We use
1046 hex escapes so that this also works with EBCDIC hosts.
1047 C++0x permits everything below 0xa0 within literals;
1048 ucn_valid_in_identifier will complain about identifiers. */
1049 else if ((result < 0xa0
1050 && !CPP_OPTION (pfile, cplusplus)
1051 && (result != 0x24 && result != 0x40 && result != 0x60))
1052 || (result & 0x80000000)
1053 || (result >= 0xD800 && result <= 0xDFFF))
1055 cpp_error (pfile, CPP_DL_ERROR,
1056 "%.*s is not a valid universal character",
1057 (int) (str - base), base);
1058 result = 1;
1060 else if (identifier_pos && result == 0x24
1061 && CPP_OPTION (pfile, dollars_in_ident))
1063 if (CPP_OPTION (pfile, warn_dollars) && !pfile->state.skipping)
1065 CPP_OPTION (pfile, warn_dollars) = 0;
1066 cpp_error (pfile, CPP_DL_PEDWARN, "'$' in identifier or number");
1068 NORMALIZE_STATE_UPDATE_IDNUM (nst, result);
1070 else if (identifier_pos)
1072 int validity = ucn_valid_in_identifier (pfile, result, nst);
1074 if (validity == 0)
1075 cpp_error (pfile, CPP_DL_ERROR,
1076 "universal character %.*s is not valid in an identifier",
1077 (int) (str - base), base);
1078 else if (validity == 2 && identifier_pos == 1)
1079 cpp_error (pfile, CPP_DL_ERROR,
1080 "universal character %.*s is not valid at the start of an identifier",
1081 (int) (str - base), base);
1084 *cp = result;
1085 return true;
1088 /* Convert an UCN, pointed to by FROM, to UTF-8 encoding, then translate
1089 it to the execution character set and write the result into TBUF.
1090 An advanced pointer is returned. Issues all relevant diagnostics. */
1091 static const uchar *
1092 convert_ucn (cpp_reader *pfile, const uchar *from, const uchar *limit,
1093 struct _cpp_strbuf *tbuf, struct cset_converter cvt)
1095 cppchar_t ucn;
1096 uchar buf[6];
1097 uchar *bufp = buf;
1098 size_t bytesleft = 6;
1099 int rval;
1100 struct normalize_state nst = INITIAL_NORMALIZE_STATE;
1102 from++; /* Skip u/U. */
1103 _cpp_valid_ucn (pfile, &from, limit, 0, &nst, &ucn);
1105 rval = one_cppchar_to_utf8 (ucn, &bufp, &bytesleft);
1106 if (rval)
1108 errno = rval;
1109 cpp_errno (pfile, CPP_DL_ERROR,
1110 "converting UCN to source character set");
1112 else if (!APPLY_CONVERSION (cvt, buf, 6 - bytesleft, tbuf))
1113 cpp_errno (pfile, CPP_DL_ERROR,
1114 "converting UCN to execution character set");
1116 return from;
1119 /* Subroutine of convert_hex and convert_oct. N is the representation
1120 in the execution character set of a numeric escape; write it into the
1121 string buffer TBUF and update the end-of-string pointer therein. WIDE
1122 is true if it's a wide string that's being assembled in TBUF. This
1123 function issues no diagnostics and never fails. */
1124 static void
1125 emit_numeric_escape (cpp_reader *pfile, cppchar_t n,
1126 struct _cpp_strbuf *tbuf, struct cset_converter cvt)
1128 size_t width = cvt.width;
1130 if (width != CPP_OPTION (pfile, char_precision))
1132 /* We have to render this into the target byte order, which may not
1133 be our byte order. */
1134 bool bigend = CPP_OPTION (pfile, bytes_big_endian);
1135 size_t cwidth = CPP_OPTION (pfile, char_precision);
1136 size_t cmask = width_to_mask (cwidth);
1137 size_t nbwc = width / cwidth;
1138 size_t i;
1139 size_t off = tbuf->len;
1140 cppchar_t c;
1142 if (tbuf->len + nbwc > tbuf->asize)
1144 tbuf->asize += OUTBUF_BLOCK_SIZE;
1145 tbuf->text = XRESIZEVEC (uchar, tbuf->text, tbuf->asize);
1148 for (i = 0; i < nbwc; i++)
1150 c = n & cmask;
1151 n >>= cwidth;
1152 tbuf->text[off + (bigend ? nbwc - i - 1 : i)] = c;
1154 tbuf->len += nbwc;
1156 else
1158 /* Note: this code does not handle the case where the target
1159 and host have a different number of bits in a byte. */
1160 if (tbuf->len + 1 > tbuf->asize)
1162 tbuf->asize += OUTBUF_BLOCK_SIZE;
1163 tbuf->text = XRESIZEVEC (uchar, tbuf->text, tbuf->asize);
1165 tbuf->text[tbuf->len++] = n;
1169 /* Convert a hexadecimal escape, pointed to by FROM, to the execution
1170 character set and write it into the string buffer TBUF. Returns an
1171 advanced pointer, and issues diagnostics as necessary.
1172 No character set translation occurs; this routine always produces the
1173 execution-set character with numeric value equal to the given hex
1174 number. You can, e.g. generate surrogate pairs this way. */
1175 static const uchar *
1176 convert_hex (cpp_reader *pfile, const uchar *from, const uchar *limit,
1177 struct _cpp_strbuf *tbuf, struct cset_converter cvt)
1179 cppchar_t c, n = 0, overflow = 0;
1180 int digits_found = 0;
1181 size_t width = cvt.width;
1182 size_t mask = width_to_mask (width);
1184 if (CPP_WTRADITIONAL (pfile))
1185 cpp_warning (pfile, CPP_W_TRADITIONAL,
1186 "the meaning of '\\x' is different in traditional C");
1188 from++; /* Skip 'x'. */
1189 while (from < limit)
1191 c = *from;
1192 if (! hex_p (c))
1193 break;
1194 from++;
1195 overflow |= n ^ (n << 4 >> 4);
1196 n = (n << 4) + hex_value (c);
1197 digits_found = 1;
1200 if (!digits_found)
1202 cpp_error (pfile, CPP_DL_ERROR,
1203 "\\x used with no following hex digits");
1204 return from;
1207 if (overflow | (n != (n & mask)))
1209 cpp_error (pfile, CPP_DL_PEDWARN,
1210 "hex escape sequence out of range");
1211 n &= mask;
1214 emit_numeric_escape (pfile, n, tbuf, cvt);
1216 return from;
1219 /* Convert an octal escape, pointed to by FROM, to the execution
1220 character set and write it into the string buffer TBUF. Returns an
1221 advanced pointer, and issues diagnostics as necessary.
1222 No character set translation occurs; this routine always produces the
1223 execution-set character with numeric value equal to the given octal
1224 number. */
1225 static const uchar *
1226 convert_oct (cpp_reader *pfile, const uchar *from, const uchar *limit,
1227 struct _cpp_strbuf *tbuf, struct cset_converter cvt)
1229 size_t count = 0;
1230 cppchar_t c, n = 0;
1231 size_t width = cvt.width;
1232 size_t mask = width_to_mask (width);
1233 bool overflow = false;
1235 while (from < limit && count++ < 3)
1237 c = *from;
1238 if (c < '0' || c > '7')
1239 break;
1240 from++;
1241 overflow |= n ^ (n << 3 >> 3);
1242 n = (n << 3) + c - '0';
1245 if (n != (n & mask))
1247 cpp_error (pfile, CPP_DL_PEDWARN,
1248 "octal escape sequence out of range");
1249 n &= mask;
1252 emit_numeric_escape (pfile, n, tbuf, cvt);
1254 return from;
1257 /* Convert an escape sequence (pointed to by FROM) to its value on
1258 the target, and to the execution character set. Do not scan past
1259 LIMIT. Write the converted value into TBUF. Returns an advanced
1260 pointer. Handles all relevant diagnostics. */
1261 static const uchar *
1262 convert_escape (cpp_reader *pfile, const uchar *from, const uchar *limit,
1263 struct _cpp_strbuf *tbuf, struct cset_converter cvt)
1265 /* Values of \a \b \e \f \n \r \t \v respectively. */
1266 #if HOST_CHARSET == HOST_CHARSET_ASCII
1267 static const uchar charconsts[] = { 7, 8, 27, 12, 10, 13, 9, 11 };
1268 #elif HOST_CHARSET == HOST_CHARSET_EBCDIC
1269 static const uchar charconsts[] = { 47, 22, 39, 12, 21, 13, 5, 11 };
1270 #else
1271 #error "unknown host character set"
1272 #endif
1274 uchar c;
1276 c = *from;
1277 switch (c)
1279 /* UCNs, hex escapes, and octal escapes are processed separately. */
1280 case 'u': case 'U':
1281 return convert_ucn (pfile, from, limit, tbuf, cvt);
1283 case 'x':
1284 return convert_hex (pfile, from, limit, tbuf, cvt);
1285 break;
1287 case '0': case '1': case '2': case '3':
1288 case '4': case '5': case '6': case '7':
1289 return convert_oct (pfile, from, limit, tbuf, cvt);
1291 /* Various letter escapes. Get the appropriate host-charset
1292 value into C. */
1293 case '\\': case '\'': case '"': case '?': break;
1295 case '(': case '{': case '[': case '%':
1296 /* '\(', etc, can be used at the beginning of a line in a long
1297 string split onto multiple lines with \-newline, to prevent
1298 Emacs or other text editors from getting confused. '\%' can
1299 be used to prevent SCCS from mangling printf format strings. */
1300 if (CPP_PEDANTIC (pfile))
1301 goto unknown;
1302 break;
1304 case 'b': c = charconsts[1]; break;
1305 case 'f': c = charconsts[3]; break;
1306 case 'n': c = charconsts[4]; break;
1307 case 'r': c = charconsts[5]; break;
1308 case 't': c = charconsts[6]; break;
1309 case 'v': c = charconsts[7]; break;
1311 case 'a':
1312 if (CPP_WTRADITIONAL (pfile))
1313 cpp_warning (pfile, CPP_W_TRADITIONAL,
1314 "the meaning of '\\a' is different in traditional C");
1315 c = charconsts[0];
1316 break;
1318 case 'e': case 'E':
1319 if (CPP_PEDANTIC (pfile))
1320 cpp_error (pfile, CPP_DL_PEDWARN,
1321 "non-ISO-standard escape sequence, '\\%c'", (int) c);
1322 c = charconsts[2];
1323 break;
1325 default:
1326 unknown:
1327 if (ISGRAPH (c))
1328 cpp_error (pfile, CPP_DL_PEDWARN,
1329 "unknown escape sequence: '\\%c'", (int) c);
1330 else
1332 /* diagnostic.c does not support "%03o". When it does, this
1333 code can use %03o directly in the diagnostic again. */
1334 char buf[32];
1335 sprintf(buf, "%03o", (int) c);
1336 cpp_error (pfile, CPP_DL_PEDWARN,
1337 "unknown escape sequence: '\\%s'", buf);
1341 /* Now convert what we have to the execution character set. */
1342 if (!APPLY_CONVERSION (cvt, &c, 1, tbuf))
1343 cpp_errno (pfile, CPP_DL_ERROR,
1344 "converting escape sequence to execution character set");
1346 return from + 1;
1348 \f
1349 /* TYPE is a token type. The return value is the conversion needed to
1350 convert from source to execution character set for the given type. */
1351 static struct cset_converter
1352 converter_for_type (cpp_reader *pfile, enum cpp_ttype type)
1354 switch (type)
1356 default:
1357 return pfile->narrow_cset_desc;
1358 case CPP_UTF8CHAR:
1359 case CPP_UTF8STRING:
1360 return pfile->utf8_cset_desc;
1361 case CPP_CHAR16:
1362 case CPP_STRING16:
1363 return pfile->char16_cset_desc;
1364 case CPP_CHAR32:
1365 case CPP_STRING32:
1366 return pfile->char32_cset_desc;
1367 case CPP_WCHAR:
1368 case CPP_WSTRING:
1369 return pfile->wide_cset_desc;
1373 /* FROM is an array of cpp_string structures of length COUNT. These
1374 are to be converted from the source to the execution character set,
1375 escape sequences translated, and finally all are to be
1376 concatenated. WIDE indicates whether or not to produce a wide
1377 string. The result is written into TO. Returns true for success,
1378 false for failure. */
1379 bool
1380 cpp_interpret_string (cpp_reader *pfile, const cpp_string *from, size_t count,
1381 cpp_string *to, enum cpp_ttype type)
1383 struct _cpp_strbuf tbuf;
1384 const uchar *p, *base, *limit;
1385 size_t i;
1386 struct cset_converter cvt = converter_for_type (pfile, type);
1388 tbuf.asize = MAX (OUTBUF_BLOCK_SIZE, from->len);
1389 tbuf.text = XNEWVEC (uchar, tbuf.asize);
1390 tbuf.len = 0;
1392 for (i = 0; i < count; i++)
1394 p = from[i].text;
1395 if (*p == 'u')
1397 if (*++p == '8')
1398 p++;
1400 else if (*p == 'L' || *p == 'U') p++;
1401 if (*p == 'R')
1403 const uchar *prefix;
1405 /* Skip over 'R"'. */
1406 p += 2;
1407 prefix = p;
1408 while (*p != '(')
1409 p++;
1410 p++;
1411 limit = from[i].text + from[i].len;
1412 if (limit >= p + (p - prefix) + 1)
1413 limit -= (p - prefix) + 1;
1415 /* Raw strings are all normal characters; these can be fed
1416 directly to convert_cset. */
1417 if (!APPLY_CONVERSION (cvt, p, limit - p, &tbuf))
1418 goto fail;
1420 continue;
1423 p++; /* Skip leading quote. */
1424 limit = from[i].text + from[i].len - 1; /* Skip trailing quote. */
1426 for (;;)
1428 base = p;
1429 while (p < limit && *p != '\\')
1430 p++;
1431 if (p > base)
1433 /* We have a run of normal characters; these can be fed
1434 directly to convert_cset. */
1435 if (!APPLY_CONVERSION (cvt, base, p - base, &tbuf))
1436 goto fail;
1438 if (p == limit)
1439 break;
1441 p = convert_escape (pfile, p + 1, limit, &tbuf, cvt);
1444 /* NUL-terminate the 'to' buffer and translate it to a cpp_string
1445 structure. */
1446 emit_numeric_escape (pfile, 0, &tbuf, cvt);
1447 tbuf.text = XRESIZEVEC (uchar, tbuf.text, tbuf.len);
1448 to->text = tbuf.text;
1449 to->len = tbuf.len;
1450 return true;
1452 fail:
1453 cpp_errno (pfile, CPP_DL_ERROR, "converting to execution character set");
1454 free (tbuf.text);
1455 return false;
1458 /* Subroutine of do_line and do_linemarker. Convert escape sequences
1459 in a string, but do not perform character set conversion. */
1460 bool
1461 cpp_interpret_string_notranslate (cpp_reader *pfile, const cpp_string *from,
1462 size_t count, cpp_string *to,
1463 enum cpp_ttype type ATTRIBUTE_UNUSED)
1465 struct cset_converter save_narrow_cset_desc = pfile->narrow_cset_desc;
1466 bool retval;
1468 pfile->narrow_cset_desc.func = convert_no_conversion;
1469 pfile->narrow_cset_desc.cd = (iconv_t) -1;
1470 pfile->narrow_cset_desc.width = CPP_OPTION (pfile, char_precision);
1472 retval = cpp_interpret_string (pfile, from, count, to, CPP_STRING);
1474 pfile->narrow_cset_desc = save_narrow_cset_desc;
1475 return retval;
1478 \f
1479 /* Subroutine of cpp_interpret_charconst which performs the conversion
1480 to a number, for narrow strings. STR is the string structure returned
1481 by cpp_interpret_string. PCHARS_SEEN and UNSIGNEDP are as for
1482 cpp_interpret_charconst. */
1483 static cppchar_t
1484 narrow_str_to_charconst (cpp_reader *pfile, cpp_string str,
1485 unsigned int *pchars_seen, int *unsignedp)
1487 size_t width = CPP_OPTION (pfile, char_precision);
1488 size_t max_chars = CPP_OPTION (pfile, int_precision) / width;
1489 size_t mask = width_to_mask (width);
1490 size_t i;
1491 cppchar_t result, c;
1492 bool unsigned_p;
1494 /* The value of a multi-character character constant, or a
1495 single-character character constant whose representation in the
1496 execution character set is more than one byte long, is
1497 implementation defined. This implementation defines it to be the
1498 number formed by interpreting the byte sequence in memory as a
1499 big-endian binary number. If overflow occurs, the high bytes are
1500 lost, and a warning is issued.
1502 We don't want to process the NUL terminator handed back by
1503 cpp_interpret_string. */
1504 result = 0;
1505 for (i = 0; i < str.len - 1; i++)
1507 c = str.text[i] & mask;
1508 if (width < BITS_PER_CPPCHAR_T)
1509 result = (result << width) | c;
1510 else
1511 result = c;
1514 if (i > max_chars)
1516 i = max_chars;
1517 cpp_error (pfile, CPP_DL_WARNING,
1518 "character constant too long for its type");
1520 else if (i > 1 && CPP_OPTION (pfile, warn_multichar))
1521 cpp_warning (pfile, CPP_W_MULTICHAR, "multi-character character constant");
1523 /* Multichar constants are of type int and therefore signed. */
1524 if (i > 1)
1525 unsigned_p = 0;
1526 else
1527 unsigned_p = CPP_OPTION (pfile, unsigned_char);
1529 /* Truncate the constant to its natural width, and simultaneously
1530 sign- or zero-extend to the full width of cppchar_t.
1531 For single-character constants, the value is WIDTH bits wide.
1532 For multi-character constants, the value is INT_PRECISION bits wide. */
1533 if (i > 1)
1534 width = CPP_OPTION (pfile, int_precision);
1535 if (width < BITS_PER_CPPCHAR_T)
1537 mask = ((cppchar_t) 1 << width) - 1;
1538 if (unsigned_p || !(result & (1 << (width - 1))))
1539 result &= mask;
1540 else
1541 result |= ~mask;
1543 *pchars_seen = i;
1544 *unsignedp = unsigned_p;
1545 return result;
1548 /* Subroutine of cpp_interpret_charconst which performs the conversion
1549 to a number, for wide strings. STR is the string structure returned
1550 by cpp_interpret_string. PCHARS_SEEN and UNSIGNEDP are as for
1551 cpp_interpret_charconst. TYPE is the token type. */
1552 static cppchar_t
1553 wide_str_to_charconst (cpp_reader *pfile, cpp_string str,
1554 unsigned int *pchars_seen, int *unsignedp,
1555 enum cpp_ttype type)
1557 bool bigend = CPP_OPTION (pfile, bytes_big_endian);
1558 size_t width = converter_for_type (pfile, type).width;
1559 size_t cwidth = CPP_OPTION (pfile, char_precision);
1560 size_t mask = width_to_mask (width);
1561 size_t cmask = width_to_mask (cwidth);
1562 size_t nbwc = width / cwidth;
1563 size_t off, i;
1564 cppchar_t result = 0, c;
1566 /* This is finicky because the string is in the target's byte order,
1567 which may not be our byte order. Only the last character, ignoring
1568 the NUL terminator, is relevant. */
1569 off = str.len - (nbwc * 2);
1570 result = 0;
1571 for (i = 0; i < nbwc; i++)
1573 c = bigend ? str.text[off + i] : str.text[off + nbwc - i - 1];
1574 result = (result << cwidth) | (c & cmask);
1577 /* Wide character constants have type wchar_t, and a single
1578 character exactly fills a wchar_t, so a multi-character wide
1579 character constant is guaranteed to overflow. */
1580 if (str.len > nbwc * 2)
1581 cpp_error (pfile, CPP_DL_WARNING,
1582 "character constant too long for its type");
1584 /* Truncate the constant to its natural width, and simultaneously
1585 sign- or zero-extend to the full width of cppchar_t. */
1586 if (width < BITS_PER_CPPCHAR_T)
1588 if (type == CPP_CHAR16 || type == CPP_CHAR32
1589 || CPP_OPTION (pfile, unsigned_wchar)
1590 || !(result & (1 << (width - 1))))
1591 result &= mask;
1592 else
1593 result |= ~mask;
1596 if (type == CPP_CHAR16 || type == CPP_CHAR32
1597 || CPP_OPTION (pfile, unsigned_wchar))
1598 *unsignedp = 1;
1599 else
1600 *unsignedp = 0;
1602 *pchars_seen = 1;
1603 return result;
1606 /* Interpret a (possibly wide) character constant in TOKEN.
1607 PCHARS_SEEN points to a variable that is filled in with the number
1608 of characters seen, and UNSIGNEDP to a variable that indicates
1609 whether the result has signed type. */
1610 cppchar_t
1611 cpp_interpret_charconst (cpp_reader *pfile, const cpp_token *token,
1612 unsigned int *pchars_seen, int *unsignedp)
1614 cpp_string str = { 0, 0 };
1615 bool wide = (token->type != CPP_CHAR && token->type != CPP_UTF8CHAR);
1616 int u8 = 2 * int(token->type == CPP_UTF8CHAR);
1617 cppchar_t result;
1619 /* An empty constant will appear as L'', u'', U'', u8'', or '' */
1620 if (token->val.str.len == (size_t) (2 + wide + u8))
1622 cpp_error (pfile, CPP_DL_ERROR, "empty character constant");
1623 return 0;
1625 else if (!cpp_interpret_string (pfile, &token->val.str, 1, &str, token->type))
1626 return 0;
1628 if (wide)
1629 result = wide_str_to_charconst (pfile, str, pchars_seen, unsignedp,
1630 token->type);
1631 else
1632 result = narrow_str_to_charconst (pfile, str, pchars_seen, unsignedp);
1634 if (str.text != token->val.str.text)
1635 free ((void *)str.text);
1637 return result;
1639 \f
1640 /* Convert an identifier denoted by ID and LEN, which might contain
1641 UCN escapes, to the source character set, either UTF-8 or
1642 UTF-EBCDIC. Assumes that the identifier is actually a valid identifier. */
1643 cpp_hashnode *
1644 _cpp_interpret_identifier (cpp_reader *pfile, const uchar *id, size_t len)
1646 /* It turns out that a UCN escape always turns into fewer characters
1647 than the escape itself, so we can allocate a temporary in advance. */
1648 uchar * buf = (uchar *) alloca (len + 1);
1649 uchar * bufp = buf;
1650 size_t idp;
1652 for (idp = 0; idp < len; idp++)
1653 if (id[idp] != '\\')
1654 *bufp++ = id[idp];
1655 else
1657 unsigned length = id[idp+1] == 'u' ? 4 : 8;
1658 cppchar_t value = 0;
1659 size_t bufleft = len - (bufp - buf);
1660 int rval;
1662 idp += 2;
1663 while (length && idp < len && ISXDIGIT (id[idp]))
1665 value = (value << 4) + hex_value (id[idp]);
1666 idp++;
1667 length--;
1669 idp--;
1671 /* Special case for EBCDIC: if the identifier contains
1672 a '$' specified using a UCN, translate it to EBCDIC. */
1673 if (value == 0x24)
1675 *bufp++ = '$';
1676 continue;
1679 rval = one_cppchar_to_utf8 (value, &bufp, &bufleft);
1680 if (rval)
1682 errno = rval;
1683 cpp_errno (pfile, CPP_DL_ERROR,
1684 "converting UCN to source character set");
1685 break;
1689 return CPP_HASHNODE (ht_lookup (pfile->hash_table,
1690 buf, bufp - buf, HT_ALLOC));
1692 \f
1693 /* Convert an input buffer (containing the complete contents of one
1694 source file) from INPUT_CHARSET to the source character set. INPUT
1695 points to the input buffer, SIZE is its allocated size, and LEN is
1696 the length of the meaningful data within the buffer. The
1697 translated buffer is returned, *ST_SIZE is set to the length of
1698 the meaningful data within the translated buffer, and *BUFFER_START
1699 is set to the start of the returned buffer. *BUFFER_START may
1700 differ from the return value in the case of a BOM or other ignored
1701 marker information.
1703 INPUT is expected to have been allocated with xmalloc. This
1704 function will either set *BUFFER_START to INPUT, or free it and set
1705 *BUFFER_START to a pointer to another xmalloc-allocated block of
1706 memory. */
1707 uchar *
1708 _cpp_convert_input (cpp_reader *pfile, const char *input_charset,
1709 uchar *input, size_t size, size_t len,
1710 const unsigned char **buffer_start, off_t *st_size)
1712 struct cset_converter input_cset;
1713 struct _cpp_strbuf to;
1714 unsigned char *buffer;
1716 input_cset = init_iconv_desc (pfile, SOURCE_CHARSET, input_charset);
1717 if (input_cset.func == convert_no_conversion)
1719 to.text = input;
1720 to.asize = size;
1721 to.len = len;
1723 else
1725 to.asize = MAX (65536, len);
1726 to.text = XNEWVEC (uchar, to.asize);
1727 to.len = 0;
1729 if (!APPLY_CONVERSION (input_cset, input, len, &to))
1730 cpp_error (pfile, CPP_DL_ERROR,
1731 "failure to convert %s to %s",
1732 CPP_OPTION (pfile, input_charset), SOURCE_CHARSET);
1734 free (input);
1737 /* Clean up the mess. */
1738 if (input_cset.func == convert_using_iconv)
1739 iconv_close (input_cset.cd);
1741 /* Resize buffer if we allocated substantially too much, or if we
1742 haven't enough space for the \n-terminator or following
1743 15 bytes of padding (used to quiet warnings from valgrind or
1744 Address Sanitizer, when the optimized lexer accesses aligned
1745 16-byte memory chunks, including the bytes after the malloced,
1746 area, and stops lexing on '\n'). */
1747 if (to.len + 4096 < to.asize || to.len + 16 > to.asize)
1748 to.text = XRESIZEVEC (uchar, to.text, to.len + 16);
1750 memset (to.text + to.len, '\0', 16);
1752 /* If the file is using old-school Mac line endings (\r only),
1753 terminate with another \r, not an \n, so that we do not mistake
1754 the \r\n sequence for a single DOS line ending and erroneously
1755 issue the "No newline at end of file" diagnostic. */
1756 if (to.len && to.text[to.len - 1] == '\r')
1757 to.text[to.len] = '\r';
1758 else
1759 to.text[to.len] = '\n';
1761 buffer = to.text;
1762 *st_size = to.len;
1763 #if HOST_CHARSET == HOST_CHARSET_ASCII
1764 /* The HOST_CHARSET test just above ensures that the source charset
1765 is UTF-8. So, ignore a UTF-8 BOM if we see one. Note that
1766 glib'c UTF-8 iconv() provider (as of glibc 2.7) does not ignore a
1767 BOM -- however, even if it did, we would still need this code due
1768 to the 'convert_no_conversion' case. */
1769 if (to.len >= 3 && to.text[0] == 0xef && to.text[1] == 0xbb
1770 && to.text[2] == 0xbf)
1772 *st_size -= 3;
1773 buffer += 3;
1775 #endif
1777 *buffer_start = to.text;
1778 return buffer;
1781 /* Decide on the default encoding to assume for input files. */
1782 const char *
1783 _cpp_default_encoding (void)
1785 const char *current_encoding = NULL;
1787 /* We disable this because the default codeset is 7-bit ASCII on
1788 most platforms, and this causes conversion failures on every
1789 file in GCC that happens to have one of the upper 128 characters
1790 in it -- most likely, as part of the name of a contributor.
1791 We should definitely recognize in-band markers of file encoding,
1792 like:
1793 - the appropriate Unicode byte-order mark (FE FF) to recognize
1794 UTF16 and UCS4 (in both big-endian and little-endian flavors)
1795 and UTF8
1796 - a "#i", "#d", "/ *", "//", " #p" or "#p" (for #pragma) to
1797 distinguish ASCII and EBCDIC.
1798 - now we can parse something like "#pragma GCC encoding <xyz>
1799 on the first line, or even Emacs/VIM's mode line tags (there's
1800 a problem here in that VIM uses the last line, and Emacs has
1801 its more elaborate "local variables" convention).
1802 - investigate whether Java has another common convention, which
1803 would be friendly to support.
1804 (Zack Weinberg and Paolo Bonzini, May 20th 2004) */
1805 #if defined (HAVE_LOCALE_H) && defined (HAVE_LANGINFO_CODESET) && 0
1806 setlocale (LC_CTYPE, "");
1807 current_encoding = nl_langinfo (CODESET);
1808 #endif
1809 if (current_encoding == NULL || *current_encoding == '\0')
1810 current_encoding = SOURCE_CHARSET;
1812 return current_encoding;