xref: /freebsd/usr.bin/localedef/wide.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
1 /*-
2  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
3  * Copyright 2012 Garrett D'Amore <garrett@damore.org>  All rights reserved.
4  * Copyright 2015 John Marino <draco@marino.st>
5  *
6  * This source code is derived from the illumos localedef command, and
7  * provided under BSD-style license terms by Nexenta Systems, Inc.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
23  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * The functions in this file convert from the standard multibyte forms
34  * to the wide character forms used internally by libc.  Unfortunately,
35  * this approach means that we need a method for each and every encoding.
36  */
37 #include <sys/cdefs.h>
38 #include <ctype.h>
39 #include <stdlib.h>
40 #include <wchar.h>
41 #include <string.h>
42 #include <sys/types.h>
43 #include "localedef.h"
44 
45 static int towide_none(wchar_t *, const char *, unsigned);
46 static int towide_utf8(wchar_t *, const char *, unsigned);
47 static int towide_big5(wchar_t *, const char *, unsigned);
48 static int towide_gbk(wchar_t *, const char *, unsigned);
49 static int towide_gb2312(wchar_t *, const char *, unsigned);
50 static int towide_gb18030(wchar_t *, const char *, unsigned);
51 static int towide_mskanji(wchar_t *, const char *, unsigned);
52 static int towide_euccn(wchar_t *, const char *, unsigned);
53 static int towide_eucjp(wchar_t *, const char *, unsigned);
54 static int towide_euckr(wchar_t *, const char *, unsigned);
55 static int towide_euctw(wchar_t *, const char *, unsigned);
56 
57 static int tomb_none(char *, wchar_t);
58 static int tomb_utf8(char *, wchar_t);
59 static int tomb_mbs(char *, wchar_t);
60 
61 static int (*_towide)(wchar_t *, const char *, unsigned) = towide_none;
62 static int (*_tomb)(char *, wchar_t) = tomb_none;
63 static char _encoding_buffer[20] = {'N','O','N','E'};
64 static const char *_encoding = _encoding_buffer;
65 static int _nbits = 7;
66 
67 /*
68  * Table of supported encodings.  We only bother to list the multibyte
69  * encodings here, because single byte locales are handed by "NONE".
70  */
71 static struct {
72 	const char *name;
73 	/* the name that the underlying libc implemenation uses */
74 	const char *cname;
75 	/* the maximum number of bits required for priorities */
76 	int nbits;
77 	int (*towide)(wchar_t *, const char *, unsigned);
78 	int (*tomb)(char *, wchar_t);
79 } mb_encodings[] = {
80 	/*
81 	 * UTF8 values max out at 0x1fffff (although in theory there could
82 	 * be later extensions, but it won't happen.)  This means we only need
83 	 * 21 bits to be able to encode the entire range of priorities.
84 	 */
85 	{ "UTF-8",	"UTF-8",	21, towide_utf8, tomb_utf8 },
86 	{ "UTF8",	"UTF-8",	21, towide_utf8, tomb_utf8 },
87 	{ "utf8",	"UTF-8",	21, towide_utf8, tomb_utf8 },
88 	{ "utf-8",	"UTF-8",	21, towide_utf8, tomb_utf8 },
89 
90 	{ "EUC-CN",	"EUC-CN",	16, towide_euccn, tomb_mbs },
91 	{ "eucCN",	"EUC-CN",	16, towide_euccn, tomb_mbs },
92 	/*
93 	 * Because the 3-byte form of EUC-JP use the same leading byte,
94 	 * only 17 bits required to provide unique priorities.  (The low
95 	 * bit of that first byte is set.)  By setting this value low,
96 	 * we can get by with only 3 bytes in the strxfrm expansion.
97 	 */
98 	{ "EUC-JP",	"EUC-JP",	17, towide_eucjp, tomb_mbs },
99 	{ "eucJP",	"EUC-JP",	17, towide_eucjp, tomb_mbs },
100 
101 	{ "EUC-KR",	"EUC-KR",	16, towide_euckr, tomb_mbs },
102 	{ "eucKR",	"EUC-KR",	16, towide_euckr, tomb_mbs },
103 	/*
104 	 * EUC-TW uses 2 bytes most of the time, but 4 bytes if the
105 	 * high order byte is 0x8E.  However, with 4 byte encodings,
106 	 * the third byte will be A0-B0.  So we only need to consider
107 	 * the lower order 24 bits for collation.
108 	 */
109 	{ "EUC-TW",	"EUC-TW",	24, towide_euctw, tomb_mbs },
110 	{ "eucTW",	"EUC-TW",	24, towide_euctw, tomb_mbs },
111 
112 	{ "MS_Kanji",	"MSKanji",	16, towide_mskanji, tomb_mbs },
113 	{ "MSKanji",	"MSKanji",	16, towide_mskanji, tomb_mbs },
114 	{ "PCK",	"MSKanji",	16, towide_mskanji, tomb_mbs },
115 	{ "SJIS",	"MSKanji",	16, towide_mskanji, tomb_mbs },
116 	{ "Shift_JIS",	"MSKanji",	16, towide_mskanji, tomb_mbs },
117 
118 	{ "BIG5",	"BIG5",		16, towide_big5, tomb_mbs },
119 	{ "big5",	"BIG5",		16, towide_big5, tomb_mbs },
120 	{ "Big5",	"BIG5",		16, towide_big5, tomb_mbs },
121 
122 	{ "GBK",	"GBK",		16, towide_gbk,	tomb_mbs },
123 
124 	/*
125 	 * GB18030 can get away with just 31 bits.  This is because the
126 	 * high order bit is always set for 4 byte values, and the
127 	 * at least one of the other bits in that 4 byte value will
128 	 * be non-zero.
129 	 */
130 	{ "GB18030",	"GB18030",	31, towide_gb18030, tomb_mbs },
131 
132 	/*
133 	 * This should probably be an aliase for euc-cn, or vice versa.
134 	 */
135 	{ "GB2312",	"GB2312",	16, towide_gb2312, tomb_mbs },
136 
137 	{ NULL, NULL, 0, 0, 0 },
138 };
139 
140 static char *
141 show_mb(const char *mb)
142 {
143 	static char buf[64];
144 
145 	/* ASCII stuff we just print */
146 	if (isascii(*mb) && isgraph(*mb)) {
147 		buf[0] = *mb;
148 		buf[1] = 0;
149 		return (buf);
150 	}
151 	buf[0] = 0;
152 	while (*mb != 0) {
153 		char scr[8];
154 		(void) snprintf(scr, sizeof (scr), "\\x%02x", *mb);
155 		(void) strlcat(buf, scr, sizeof (buf));
156 		mb++;
157 	}
158 	return (buf);
159 }
160 
161 static char	*widemsg;
162 
163 void
164 werr(const char *fmt, ...)
165 {
166 	char	*msg;
167 
168 	va_list	va;
169 	va_start(va, fmt);
170 	(void) vasprintf(&msg, fmt, va);
171 	va_end(va);
172 
173 	free(widemsg);
174 	widemsg = msg;
175 }
176 
177 /*
178  * This is used for 8-bit encodings.
179  */
180 int
181 towide_none(wchar_t *c, const char *mb, unsigned n __unused)
182 {
183 	if (mb_cur_max != 1) {
184 		werr("invalid or unsupported multibyte locale");
185 		return (-1);
186 	}
187 	*c = (uint8_t)*mb;
188 	return (1);
189 }
190 
191 int
192 tomb_none(char *mb, wchar_t wc)
193 {
194 	if (mb_cur_max != 1) {
195 		werr("invalid or unsupported multibyte locale");
196 		return (-1);
197 	}
198 	*(uint8_t *)mb = (wc & 0xff);
199 	mb[1] = 0;
200 	return (1);
201 }
202 
203 /*
204  * UTF-8 stores wide characters in UTF-32 form.
205  */
206 int
207 towide_utf8(wchar_t *wc, const char *mb, unsigned n)
208 {
209 	wchar_t	c;
210 	int	nb;
211 	wchar_t	lv;	/* lowest legal value */
212 	int	i;
213 	const uint8_t *s = (const uint8_t *)mb;
214 
215 	c = *s;
216 
217 	if ((c & 0x80) == 0) {
218 		/* 7-bit ASCII */
219 		*wc = c;
220 		return (1);
221 	} else if ((c & 0xe0) == 0xc0) {
222 		/* u80-u7ff - two bytes encoded */
223 		nb = 2;
224 		lv = 0x80;
225 		c &= ~0xe0;
226 	} else if ((c & 0xf0) == 0xe0) {
227 		/* u800-uffff - three bytes encoded */
228 		nb = 3;
229 		lv = 0x800;
230 		c &= ~0xf0;
231 	} else if ((c & 0xf8) == 0xf0) {
232 		/* u1000-u1fffff - four bytes encoded */
233 		nb = 4;
234 		lv = 0x1000;
235 		c &= ~0xf8;
236 	} else {
237 		/* 5 and 6 byte encodings are not legal unicode */
238 		werr("utf8 encoding too large (%s)", show_mb(mb));
239 		return (-1);
240 	}
241 	if (nb > (int)n) {
242 		werr("incomplete utf8 sequence (%s)", show_mb(mb));
243 		return (-1);
244 	}
245 
246 	for (i = 1; i < nb; i++) {
247 		if (((s[i]) & 0xc0) != 0x80) {
248 			werr("illegal utf8 byte (%x)", s[i]);
249 			return (-1);
250 		}
251 		c <<= 6;
252 		c |= (s[i] & 0x3f);
253 	}
254 
255 	if (c < lv) {
256 		werr("illegal redundant utf8 encoding (%s)", show_mb(mb));
257 		return (-1);
258 	}
259 	*wc = c;
260 	return (nb);
261 }
262 
263 int
264 tomb_utf8(char *mb, wchar_t wc)
265 {
266 	uint8_t *s = (uint8_t *)mb;
267 	uint8_t msk;
268 	int cnt;
269 	int i;
270 
271 	if (wc <= 0x7f) {
272 		s[0] = wc & 0x7f;
273 		s[1] = 0;
274 		return (1);
275 	}
276 	if (wc <= 0x7ff) {
277 		cnt = 2;
278 		msk = 0xc0;
279 	} else if (wc <= 0xffff) {
280 		cnt = 3;
281 		msk = 0xe0;
282 	} else if (wc <= 0x1fffff) {
283 		cnt = 4;
284 		msk = 0xf0;
285 	} else {
286 		werr("illegal uf8 char (%x)", wc);
287 		return (-1);
288 	}
289 	for (i = cnt - 1; i; i--) {
290 		s[i] = (wc & 0x3f) | 0x80;
291 		wc >>= 6;
292 	}
293 	s[0] = (msk) | wc;
294 	s[cnt] = 0;
295 	return (cnt);
296 }
297 
298 /*
299  * Several encodings share a simplistic dual byte encoding.  In these
300  * forms, they all indicate that a two byte sequence is to be used if
301  * the first byte has its high bit set.  They all store this simple
302  * encoding as a 16-bit value, although a great many of the possible
303  * code points are not used in most character sets.  This gives a possible
304  * set of just over 32,000 valid code points.
305  *
306  * 0x00 - 0x7f		- 1 byte encoding
307  * 0x80 - 0x7fff	- illegal
308  * 0x8000 - 0xffff	- 2 byte encoding
309  */
310 
311 static int
312 towide_dbcs(wchar_t *wc, const char *mb, unsigned n)
313 {
314 	wchar_t	c;
315 
316 	c = *(const uint8_t *)mb;
317 
318 	if ((c & 0x80) == 0) {
319 		/* 7-bit */
320 		*wc = c;
321 		return (1);
322 	}
323 	if (n < 2) {
324 		werr("incomplete character sequence (%s)", show_mb(mb));
325 		return (-1);
326 	}
327 
328 	/* Store both bytes as a single 16-bit wide. */
329 	c <<= 8;
330 	c |= (uint8_t)(mb[1]);
331 	*wc = c;
332 	return (2);
333 }
334 
335 /*
336  * Most multibyte locales just convert the wide character to the multibyte
337  * form by stripping leading null bytes, and writing the 32-bit quantity
338  * in big-endian order.
339  */
340 int
341 tomb_mbs(char *mb, wchar_t wc)
342 {
343 	uint8_t *s = (uint8_t *)mb;
344 	int 	n = 0, c;
345 
346 	if ((wc & 0xff000000U) != 0) {
347 		n = 4;
348 	} else if ((wc & 0x00ff0000U) != 0) {
349 		n = 3;
350 	} else if ((wc & 0x0000ff00U) != 0) {
351 		n = 2;
352 	} else {
353 		n = 1;
354 	}
355 	c = n;
356 	while (n) {
357 		n--;
358 		s[n] = wc & 0xff;
359 		wc >>= 8;
360 	}
361 	/* ensure null termination */
362 	s[c] = 0;
363 	return (c);
364 }
365 
366 
367 /*
368  * big5 is a simple dual byte character set.
369  */
370 int
371 towide_big5(wchar_t *wc, const char *mb, unsigned n)
372 {
373 	return (towide_dbcs(wc, mb, n));
374 }
375 
376 /*
377  * GBK encodes wides in the same way that big5 does, the high order
378  * bit of the first byte indicates a double byte character.
379  */
380 int
381 towide_gbk(wchar_t *wc, const char *mb, unsigned n)
382 {
383 	return (towide_dbcs(wc, mb, n));
384 }
385 
386 /*
387  * GB2312 is another DBCS.  Its cleaner than others in that the second
388  * byte does not encode ASCII, but it supports characters.
389  */
390 int
391 towide_gb2312(wchar_t *wc, const char *mb, unsigned n)
392 {
393 	return (towide_dbcs(wc, mb, n));
394 }
395 
396 /*
397  * GB18030.  This encodes as 8, 16, or 32-bits.
398  * 7-bit values are in 1 byte,  4 byte sequences are used when
399  * the second byte encodes 0x30-39 and all other sequences are 2 bytes.
400  */
401 int
402 towide_gb18030(wchar_t *wc, const char *mb, unsigned n)
403 {
404 	wchar_t	c;
405 
406 	c = *(const uint8_t *)mb;
407 
408 	if ((c & 0x80) == 0) {
409 		/* 7-bit */
410 		*wc = c;
411 		return (1);
412 	}
413 	if (n < 2) {
414 		werr("incomplete character sequence (%s)", show_mb(mb));
415 		return (-1);
416 	}
417 
418 	/* pull in the second byte */
419 	c <<= 8;
420 	c |= (uint8_t)(mb[1]);
421 
422 	if (((c & 0xff) >= 0x30) && ((c & 0xff) <= 0x39)) {
423 		if (n < 4) {
424 			werr("incomplete 4-byte character sequence (%s)",
425 			    show_mb(mb));
426 			return (-1);
427 		}
428 		c <<= 8;
429 		c |= (uint8_t)(mb[2]);
430 		c <<= 8;
431 		c |= (uint8_t)(mb[3]);
432 		*wc = c;
433 		return (4);
434 	}
435 
436 	*wc = c;
437 	return (2);
438 }
439 
440 /*
441  * MS-Kanji (aka SJIS) is almost a clean DBCS like the others, but it
442  * also has a range of single byte characters above 0x80.  (0xa1-0xdf).
443  */
444 int
445 towide_mskanji(wchar_t *wc, const char *mb, unsigned n)
446 {
447 	wchar_t	c;
448 
449 	c = *(const uint8_t *)mb;
450 
451 	if ((c < 0x80) || ((c > 0xa0) && (c < 0xe0))) {
452 		/* 7-bit */
453 		*wc = c;
454 		return (1);
455 	}
456 
457 	if (n < 2) {
458 		werr("incomplete character sequence (%s)", show_mb(mb));
459 		return (-1);
460 	}
461 
462 	/* Store both bytes as a single 16-bit wide. */
463 	c <<= 8;
464 	c |= (uint8_t)(mb[1]);
465 	*wc = c;
466 	return (2);
467 }
468 
469 /*
470  * EUC forms.  EUC encodings are "variable".  FreeBSD carries some additional
471  * variable data to encode these, but we're going to treat each as independent
472  * instead.  Its the only way we can sensibly move forward.
473  *
474  * Note that the way in which the different EUC forms vary is how wide
475  * CS2 and CS3 are and what the first byte of them is.
476  */
477 static int
478 towide_euc_impl(wchar_t *wc, const char *mb, unsigned n,
479     uint8_t cs2, uint8_t cs2width, uint8_t cs3, uint8_t cs3width)
480 {
481 	int i;
482 	int width = 2;
483 	wchar_t	c;
484 
485 	c = *(const uint8_t *)mb;
486 
487 	/*
488 	 * All variations of EUC encode 7-bit ASCII as one byte, and use
489 	 * additional bytes for more than that.
490 	 */
491 	if ((c & 0x80) == 0) {
492 		/* 7-bit */
493 		*wc = c;
494 		return (1);
495 	}
496 
497 	/*
498 	 * All EUC variants reserve 0xa1-0xff to identify CS1, which
499 	 * is always two bytes wide.  Note that unused CS will be zero,
500 	 * and that cannot be true because we know that the high order
501 	 * bit must be set.
502 	 */
503 	if (c >= 0xa1) {
504 		width = 2;
505 	} else if (c == cs2) {
506 		width = cs2width;
507 	} else if (c == cs3) {
508 		width = cs3width;
509 	}
510 
511 	if ((int)n < width) {
512 		werr("incomplete character sequence (%s)", show_mb(mb));
513 		return (-1);
514 	}
515 
516 	for (i = 1; i < width; i++) {
517 		/* pull in the next byte */
518 		c <<= 8;
519 		c |= (uint8_t)(mb[i]);
520 	}
521 
522 	*wc = c;
523 	return (width);
524 }
525 
526 /*
527  * EUC-CN encodes as follows:
528  *
529  * Code set 0 (ASCII):				0x21-0x7E
530  * Code set 1 (CNS 11643-1992 Plane 1):		0xA1A1-0xFEFE
531  * Code set 2:					unused
532  * Code set 3:					unused
533  */
534 int
535 towide_euccn(wchar_t *wc, const char *mb, unsigned n)
536 {
537 	return (towide_euc_impl(wc, mb, n, 0x8e, 4, 0, 0));
538 }
539 
540 /*
541  * EUC-JP encodes as follows:
542  *
543  * Code set 0 (ASCII or JIS X 0201-1976 Roman):	0x21-0x7E
544  * Code set 1 (JIS X 0208):			0xA1A1-0xFEFE
545  * Code set 2 (half-width katakana):		0x8EA1-0x8EDF
546  * Code set 3 (JIS X 0212-1990):		0x8FA1A1-0x8FFEFE
547  */
548 int
549 towide_eucjp(wchar_t *wc, const char *mb, unsigned n)
550 {
551 	return (towide_euc_impl(wc, mb, n, 0x8e, 2, 0x8f, 3));
552 }
553 
554 /*
555  * EUC-KR encodes as follows:
556  *
557  * Code set 0 (ASCII or KS C 5636-1993):	0x21-0x7E
558  * Code set 1 (KS C 5601-1992):			0xA1A1-0xFEFE
559  * Code set 2:					unused
560  * Code set 3:					unused
561  */
562 int
563 towide_euckr(wchar_t *wc, const char *mb, unsigned n)
564 {
565 	return (towide_euc_impl(wc, mb, n, 0, 0, 0, 0));
566 }
567 
568 /*
569  * EUC-TW encodes as follows:
570  *
571  * Code set 0 (ASCII):				0x21-0x7E
572  * Code set 1 (CNS 11643-1992 Plane 1):		0xA1A1-0xFEFE
573  * Code set 2 (CNS 11643-1992 Planes 1-16):	0x8EA1A1A1-0x8EB0FEFE
574  * Code set 3:					unused
575  */
576 int
577 towide_euctw(wchar_t *wc, const char *mb, unsigned n)
578 {
579 	return (towide_euc_impl(wc, mb, n, 0x8e, 4, 0, 0));
580 }
581 
582 /*
583  * Public entry points.
584  */
585 
586 int
587 to_wide(wchar_t *wc, const char *mb)
588 {
589 	/* this won't fail hard */
590 	return (_towide(wc, mb, strlen(mb)));
591 }
592 
593 int
594 to_mb(char *mb, wchar_t wc)
595 {
596 	int	rv;
597 
598 	if ((rv = _tomb(mb, wc)) < 0) {
599 		warn("%s", widemsg);
600 		free(widemsg);
601 		widemsg = NULL;
602 	}
603 	return (rv);
604 }
605 
606 char *
607 to_mb_string(const wchar_t *wcs)
608 {
609 	char	*mbs;
610 	char	*ptr;
611 	int	len;
612 
613 	mbs = malloc((wcslen(wcs) * mb_cur_max) + 1);
614 	if (mbs == NULL) {
615 		warn("out of memory");
616 		return (NULL);
617 	}
618 	ptr = mbs;
619 	while (*wcs) {
620 		if ((len = to_mb(ptr, *wcs)) < 0) {
621 			INTERR;
622 			free(mbs);
623 			return (NULL);
624 		}
625 		wcs++;
626 		ptr += len;
627 	}
628 	*ptr = 0;
629 	return (mbs);
630 }
631 
632 void
633 set_wide_encoding(const char *encoding)
634 {
635 	int i;
636 
637 	_towide = towide_none;
638 	_tomb = tomb_none;
639 	_nbits = 8;
640 
641 	snprintf(_encoding_buffer, sizeof(_encoding_buffer), "NONE:%s",
642 	    encoding);
643 	for (i = 0; mb_encodings[i].name; i++) {
644 		if (strcasecmp(encoding, mb_encodings[i].name) == 0) {
645 			_towide = mb_encodings[i].towide;
646 			_tomb = mb_encodings[i].tomb;
647 			_encoding = mb_encodings[i].cname;
648 			_nbits = mb_encodings[i].nbits;
649 			break;
650 		}
651 	}
652 }
653 
654 const char *
655 get_wide_encoding(void)
656 {
657 	return (_encoding);
658 }
659 
660 int
661 max_wide(void)
662 {
663 	return ((int)((1U << _nbits) - 1));
664 }
665