2 * Copyright 2010 Nexenta Systems, Inc. All rights reserved.
3 * Copyright (c) 1995 Alex Tatmanjants <alex@elvisti.kiev.ua>
4 * at Electronni Visti IA, Kiev, Ukraine.
7 * Copyright (c) 2011 The FreeBSD Foundation
9 * Portions of this software were developed by David Chisnall
10 * under sponsorship from the FreeBSD Foundation.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * Adapted to xlocale by John Marino <draco@marino.st>
36 #include "namespace.h"
45 #include <sys/types.h>
48 #include "un-namespace.h"
51 #include "setlocale.h"
54 struct xlocale_collate __xlocale_global_collate = {
55 {{0}, "C"}, 1, 0, 0, 0
58 struct xlocale_collate __xlocale_C_collate = {
59 {{0}, "C"}, 1, 0, 0, 0
62 #include "libc_private.h"
65 __collate_load_tables_l(const char *encoding, struct xlocale_collate *table);
68 destruct_collate(void *t)
70 struct xlocale_collate *table = t;
71 if (table->map && (table->maplen > 0)) {
72 (void) munmap(table->map, table->maplen);
78 __collate_load(const char *encoding, __unused locale_t unused)
80 if (strcmp(encoding, "C") == 0 || strcmp(encoding, "POSIX") == 0) {
81 return &__xlocale_C_collate;
83 struct xlocale_collate *table = calloc(sizeof(struct xlocale_collate), 1);
84 table->header.header.destructor = destruct_collate;
85 // FIXME: Make sure that _LDP_CACHE is never returned. We should be doing
86 // the caching outside of this section
87 if (__collate_load_tables_l(encoding, table) != _LDP_LOADED) {
88 xlocale_release(table);
95 * Load the collation tables for the specified encoding into the global table.
98 __collate_load_tables(const char *encoding)
100 int ret = __collate_load_tables_l(encoding, &__xlocale_global_collate);
105 __collate_load_tables_l(const char *encoding, struct xlocale_collate *table)
111 collate_info_t *info;
115 table->__collate_load_error = 1;
117 /* 'encoding' must be already checked. */
118 if (strcmp(encoding, "C") == 0 || strcmp(encoding, "POSIX") == 0) {
122 (void) snprintf(buf, sizeof (buf), "%s/%s/LC_COLLATE",
123 _PathLocale, encoding);
125 if ((fd = _open(buf, O_RDONLY)) < 0)
127 if (_fstat(fd, &sbuf) < 0) {
131 if (sbuf.st_size < (COLLATE_STR_LEN + sizeof (info))) {
136 map = mmap(NULL, sbuf.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
138 if ((TMP = map) == NULL) {
142 if (strncmp(TMP, COLLATE_VERSION, COLLATE_STR_LEN) != 0) {
143 (void) munmap(map, sbuf.st_size);
147 TMP += COLLATE_STR_LEN;
150 TMP += sizeof (*info);
152 if ((info->directive_count < 1) ||
153 (info->directive_count >= COLL_WEIGHTS_MAX) ||
154 ((chains = info->chain_count) < 0)) {
155 (void) munmap(map, sbuf.st_size);
160 i = (sizeof (collate_char_t) * (UCHAR_MAX + 1)) +
161 (sizeof (collate_chain_t) * chains) +
162 (sizeof (collate_large_t) * info->large_count);
163 for (z = 0; z < info->directive_count; z++) {
164 i += sizeof (collate_subst_t) * info->subst_count[z];
166 if (i != (sbuf.st_size - (TMP - map))) {
167 (void) munmap(map, sbuf.st_size);
173 table->char_pri_table = (void *)TMP;
174 TMP += sizeof (collate_char_t) * (UCHAR_MAX + 1);
176 for (z = 0; z < info->directive_count; z++) {
177 if (info->subst_count[z] > 0) {
178 table->subst_table[z] = (void *)TMP;
179 TMP += info->subst_count[z] * sizeof (collate_subst_t);
181 table->subst_table[z] = NULL;
186 table->chain_pri_table = (void *)TMP;
187 TMP += chains * sizeof (collate_chain_t);
189 table->chain_pri_table = NULL;
190 if (info->large_count > 0)
191 table->large_pri_table = (void *)TMP;
193 table->large_pri_table = NULL;
195 table->__collate_load_error = 0;
196 return (_LDP_LOADED);
199 static const int32_t *
200 substsearch(struct xlocale_collate *table, const wchar_t key, int pass)
202 const collate_subst_t *p;
203 int n = table->info->subst_count[pass];
208 if (pass >= table->info->directive_count)
211 if (!(key & COLLATE_SUBST_PRIORITY))
214 p = table->subst_table[pass] + (key & ~COLLATE_SUBST_PRIORITY);
215 assert(p->key == key);
220 * Note: for performance reasons, we have expanded bsearch here. This avoids
221 * function call overhead with each comparison.
224 static collate_chain_t *
225 chainsearch(struct xlocale_collate *table, const wchar_t *key, int *len)
231 collate_chain_t *tab;
233 if (table->info->chain_count == 0)
237 high = table->info->chain_count - 1;
238 tab = table->chain_pri_table;
240 while (low <= high) {
241 next = (low + high) / 2;
243 compar = *key - *p->str;
245 l = wcsnlen(p->str, COLLATE_STR_LEN);
246 compar = wcsncmp(key, p->str, l);
260 static collate_large_t *
261 largesearch(struct xlocale_collate *table, const wchar_t key)
264 int high = table->info->large_count - 1;
267 collate_large_t *tab = table->large_pri_table;
269 if (table->info->large_count == 0)
272 while (low <= high) {
273 next = (low + high) / 2;
275 compar = key - p->val;
287 _collate_lookup(struct xlocale_collate *table, const wchar_t *t, int *len,
288 int *pri, int which, const int **state)
291 collate_large_t *match;
296 * If this is the "last" pass for the UNDEFINED, then
297 * we just return the priority itself.
299 if (which >= table->info->directive_count) {
307 * If we have remaining substitution data from a previous
308 * call, consume it first.
310 if ((sptr = *state) != NULL) {
313 *state = *sptr ? sptr : NULL;
318 /* No active substitutions */
322 * Check for composites such as dipthongs that collate as a
323 * single element (aka chains or collating-elements).
325 if (((p2 = chainsearch(table, t, &l)) != NULL) &&
326 ((p = p2->pri[which]) >= 0)) {
331 } else if (*t <= UCHAR_MAX) {
334 * Character is a small (8-bit) character.
335 * We just look these up directly for speed.
337 *pri = table->char_pri_table[*t].pri[which];
339 } else if ((table->info->large_count > 0) &&
340 ((match = largesearch(table, *t)) != NULL)) {
343 * Character was found in the extended table.
345 *pri = match->pri.pri[which];
349 * Character lacks a specific definition.
351 if (table->info->directive[which] & DIRECTIVE_UNDEFINED) {
352 /* Mask off sign bit to prevent ordering confusion. */
353 *pri = (*t & COLLATE_MAX_PRIORITY);
355 *pri = table->info->undef_pri[which];
357 /* No substitutions for undefined characters! */
362 * Try substituting (expanding) the character. We are
363 * currently doing this *after* the chain compression. I
364 * think it should not matter, but this way might be slightly
367 * We do this after the priority search, as this will help us
368 * to identify a single key value. In order for this to work,
369 * its important that the priority assigned to a given element
370 * to be substituted be unique for that level. The localedef
371 * code ensures this for us.
373 if ((sptr = substsearch(table, *pri, which)) != NULL) {
374 if ((*pri = *sptr) != 0) {
376 *state = *sptr ? sptr : NULL;
383 * This is the meaty part of wcsxfrm & strxfrm. Note that it does
384 * NOT NULL terminate. That is left to the caller.
387 _collate_wxfrm(struct xlocale_collate *table, const wchar_t *src, wchar_t *xf,
396 const int32_t *state;
399 int ndir = table->info->directive_count;
403 for (pass = 0; pass <= ndir; pass++) {
408 /* insert level separator from the previous pass */
416 /* special pass for undefined */
418 direc = DIRECTIVE_FORWARD | DIRECTIVE_UNDEFINED;
420 direc = table->info->directive[pass];
425 if (direc & DIRECTIVE_BACKWARD) {
429 if ((tr = wcsdup(t)) == NULL) {
434 fp = tr + wcslen(tr) - 1;
440 t = (const wchar_t *)tr;
443 if (direc & DIRECTIVE_POSITION) {
444 while (*t || state) {
445 _collate_lookup(table, t, &len, &pri, pass, &state);
452 pri = COLLATE_MAX_PRIORITY;
462 while (*t || state) {
463 _collate_lookup(table, t, &len, &pri, pass, &state);
486 return ((size_t)(-1));
490 * In the non-POSIX case, we transform each character into a string of
491 * characters representing the character's priority. Since char is usually
492 * signed, we are limited by 7 bits per byte. To avoid zero, we need to add
493 * XFRM_OFFSET, so we can't use a full 7 bits. For simplicity, we choose 6
496 * It turns out that we sometimes have real priorities that are
497 * 31-bits wide. (But: be careful using priorities where the high
498 * order bit is set -- i.e. the priority is negative. The sort order
499 * may be surprising!)
501 * TODO: This would be a good area to optimize somewhat. It turns out
502 * that real prioririties *except for the last UNDEFINED pass* are generally
503 * very small. We need the localedef code to precalculate the max
504 * priority for us, and ideally also give us a mask, and then we could
505 * severely limit what we expand to.
508 #define XFRM_OFFSET ('0') /* make all printable characters */
510 #define XFRM_MASK ((1 << XFRM_SHIFT) - 1)
511 #define XFRM_SEP ('.') /* chosen to be less than XFRM_OFFSET */
514 xfrm(struct xlocale_collate *table, unsigned char *p, int pri, int pass)
516 /* we use unsigned to ensure zero fill on right shift */
517 uint32_t val = (uint32_t)table->info->pri_count[pass];
521 *p = (pri & XFRM_MASK) + XFRM_OFFSET;
531 _collate_sxfrm(struct xlocale_collate *table, const wchar_t *src, char *xf,
540 const int32_t *state;
544 uint8_t buf[XFRM_BYTES];
545 int ndir = table->info->directive_count;
549 for (pass = 0; pass <= ndir; pass++) {
554 /* insert level separator from the previous pass */
562 /* special pass for undefined */
564 direc = DIRECTIVE_FORWARD | DIRECTIVE_UNDEFINED;
566 direc = table->info->directive[pass];
571 if (direc & DIRECTIVE_BACKWARD) {
575 if ((tr = wcsdup(t)) == NULL) {
580 fp = tr + wcslen(tr) - 1;
586 t = (const wchar_t *)tr;
589 if (direc & DIRECTIVE_POSITION) {
590 while (*t || state) {
592 _collate_lookup(table, t, &len, &pri, pass, &state);
599 pri = COLLATE_MAX_PRIORITY;
602 b = xfrm(table, buf, pri, pass);
616 while (*t || state) {
617 _collate_lookup(table, t, &len, &pri, pass, &state);
627 b = xfrm(table, buf, pri, pass);
648 return ((size_t)(-1));
652 * __collate_equiv_value returns the primary collation value for the given
653 * collating symbol specified by str and len. Zero or negative is returned
654 * if the collating symbol was not found. This function is used by bracket
655 * code in the TRE regex library.
658 __collate_equiv_value(locale_t locale, const wchar_t *str, size_t len)
662 if (len < 1 || len >= COLLATE_STR_LEN)
666 struct xlocale_collate *table =
667 (struct xlocale_collate*)locale->components[XLC_COLLATE];
669 if (table->__collate_load_error)
670 return ((len == 1 && *str <= UCHAR_MAX) ? *str : -1);
674 if (*str <= UCHAR_MAX)
675 e = table->char_pri_table[*str].pri[0];
676 else if (table->info->large_count > 0) {
677 collate_large_t *match_large;
678 match_large = largesearch(table, *str);
680 e = match_large->pri.pri[0];
684 return (e > 0 ? e : 0);
686 if (table->info->chain_count > 0) {
687 wchar_t name[COLLATE_STR_LEN];
688 collate_chain_t *match_chain;
691 wcsncpy (name, str, len);
693 match_chain = chainsearch(table, name, &clen);
695 e = match_chain->pri[0];
698 return (e < 0 ? -e : e);