2 * Copyright 2010 Nexenta Systems, Inc. All rights reserved.
3 * Copyright 2015 John Marino <draco@marino.st>
5 * This source code is derived from the illumos localedef command, and
6 * provided under BSD-style license terms by Nexenta Systems, Inc.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
19 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
22 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
32 * LC_COLLATE database generation routines for localedef.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include <sys/types.h>
48 #include "localedef.h"
55 * It will be extremely helpful to the reader if they have access to
56 * the localedef and locale file format specifications available.
57 * Latest versions of these are available from www.opengroup.org.
59 * The design for the collation code is a bit complex. The goal is a
60 * single collation database as described in collate.h (in
61 * libc/port/locale). However, there are some other tidbits:
63 * a) The substitution entries are now a directly indexable array. A
64 * priority elsewhere in the table is taken as an index into the
65 * substitution table if it has a high bit (COLLATE_SUBST_PRIORITY)
66 * set. (The bit is cleared and the result is the index into the
69 * b) We eliminate duplicate entries into the substitution table.
70 * This saves a lot of space.
72 * c) The priorities for each level are "compressed", so that each
73 * sorting level has consecutively numbered priorities starting at 1.
74 * (O is reserved for the ignore priority.) This means sort levels
75 * which only have a few distinct priorities can represent the
76 * priority level in fewer bits, which makes the strxfrm output
79 * d) We record the total number of priorities so that strxfrm can
80 * figure out how many bytes to expand a numeric priority into.
82 * e) For the UNDEFINED pass (the last pass), we record the maximum
83 * number of bits needed to uniquely prioritize these entries, so that
84 * the last pass can also use smaller strxfrm output when possible.
86 * f) Priorities with the sign bit set are verboten. This works out
87 * because no active character set needs that bit to carry significant
88 * information once the character is in wide form.
90 * To process the entire data to make the database, we actually run
91 * multiple passes over the data.
93 * The first pass, which is done at parse time, identifies elements,
94 * substitutions, and such, and records them in priority order. As
95 * some priorities can refer to other priorities, using forward
96 * references, we use a table of references indicating whether the
97 * priority's value has been resolved, or whether it is still a
100 * The second pass walks over all the items in priority order, noting
101 * that they are used directly, and not just an indirect reference.
102 * This is done by creating a "weight" structure for the item. The
103 * weights are stashed in an RB tree sorted by relative "priority".
105 * The third pass walks over all the weight structures, in priority
106 * order, and assigns a new monotonically increasing (per sort level)
107 * weight value to them. These are the values that will actually be
108 * written to the file.
110 * The fourth pass just writes the data out.
114 * In order to resolve the priorities, we create a table of priorities.
115 * Entries in the table can be in one of three states.
117 * UNKNOWN is for newly allocated entries, and indicates that nothing
118 * is known about the priority. (For example, when new entries are created
119 * for collating-symbols, this is the value assigned for them until the
120 * collating symbol's order has been determined.
122 * RESOLVED is used for an entry where the priority indicates the final
125 * REFER is used for entries that reference other entries. Typically
126 * this is used for forward references. A collating-symbol can never
129 * The "pass" field is used during final resolution to aid in detection
130 * of referencing loops. (For example <A> depends on <B>, but <B> has its
131 * priority dependent on <A>.)
134 UNKNOWN, /* priority is totally unknown */
135 RESOLVED, /* priority value fully resolved */
136 REFER /* priority is a reference (index) */
139 typedef struct weight {
142 RB_ENTRY(weight) entry;
145 typedef struct priority {
152 #define NUM_WT collinfo.directive_count
155 * These are the abstract collating symbols, which are just a symbolic
156 * way to reference a priority.
161 RB_ENTRY(collsym) entry;
165 * These are also abstract collating symbols, but we allow them to have
166 * different priorities at different levels.
168 typedef struct collundef {
170 int32_t ref[COLL_WEIGHTS_MAX];
171 RB_ENTRY(collundef) entry;
175 * These are called "chains" in libc. This records the fact that two
176 * more characters should be treated as a single collating entity when
177 * they appear together. For example, in Spanish <C><h> gets collated
178 * as a character between <C> and <D>.
183 int32_t ref[COLL_WEIGHTS_MAX];
184 RB_ENTRY(collelem) rb_bysymbol;
185 RB_ENTRY(collelem) rb_byexpand;
189 * Individual characters have a sequence of weights as well.
191 typedef struct collchar {
193 int32_t ref[COLL_WEIGHTS_MAX];
194 RB_ENTRY(collchar) entry;
198 * Substitution entries. The key is itself a priority. Note that
199 * when we create one of these, we *automatically* wind up with a
200 * fully resolved priority for the key, because creation of
201 * substitutions creates a resolved priority at the same time.
203 typedef struct subst{
205 int32_t ref[COLLATE_STR_LEN];
206 RB_ENTRY(subst) entry;
207 RB_ENTRY(subst) entry_ref;
210 static RB_HEAD(collsyms, collsym) collsyms;
211 static RB_HEAD(collundefs, collundef) collundefs;
212 static RB_HEAD(elem_by_symbol, collelem) elem_by_symbol;
213 static RB_HEAD(elem_by_expand, collelem) elem_by_expand;
214 static RB_HEAD(collchars, collchar) collchars;
215 static RB_HEAD(substs, subst) substs[COLL_WEIGHTS_MAX];
216 static RB_HEAD(substs_ref, subst) substs_ref[COLL_WEIGHTS_MAX];
217 static RB_HEAD(weights, weight) weights[COLL_WEIGHTS_MAX];
218 static int32_t nweight[COLL_WEIGHTS_MAX];
221 * This is state tracking for the ellipsis token. Note that we start
222 * the initial values so that the ellipsis logic will think we got a
223 * magic starting value of NUL. It starts at minus one because the
224 * starting point is exclusive -- i.e. the starting point is not
225 * itself handled by the ellipsis code.
227 static int currorder = EOF;
228 static int lastorder = EOF;
229 static collelem_t *currelem;
230 static collchar_t *currchar;
231 static collundef_t *currundef;
232 static wchar_t ellipsis_start = 0;
233 static int32_t ellipsis_weights[COLL_WEIGHTS_MAX];
236 * We keep a running tally of weights.
238 static int nextpri = 1;
239 static int nextsubst[COLL_WEIGHTS_MAX] = { 0 };
242 * This array collects up the weights for each level.
244 static int32_t order_weights[COLL_WEIGHTS_MAX];
245 static int curr_weight = 0;
246 static int32_t subst_weights[COLLATE_STR_LEN];
247 static int curr_subst = 0;
250 * Some initial priority values.
252 static int32_t pri_undefined[COLL_WEIGHTS_MAX];
253 static int32_t pri_ignore;
255 static collate_info_t collinfo;
257 static collpri_t *prilist = NULL;
258 static int numpri = 0;
259 static int maxpri = 0;
261 static void start_order(int);
268 if (numpri >= maxpri) {
269 maxpri = maxpri ? maxpri * 2 : 1024;
270 prilist = realloc(prilist, sizeof (collpri_t) * maxpri);
271 if (prilist == NULL) {
272 fprintf(stderr,"out of memory");
275 for (i = numpri; i < maxpri; i++) {
276 prilist[i].res = UNKNOWN;
287 if ((ref < 0) || (ref > numpri)) {
291 return (&prilist[ref]);
295 set_pri(int32_t ref, int32_t v, res_t res)
301 if ((res == REFER) && ((v < 0) || (v >= numpri))) {
305 /* Resolve self references */
306 if ((res == REFER) && (ref == v)) {
311 if (pri->res != UNKNOWN) {
312 warn("repeated item in order list (first on %d)",
316 pri->lineno = lineno;
322 resolve_pri(int32_t ref)
325 static int32_t pass = 0;
329 while (pri->res == REFER) {
330 if (pri->pass == pass) {
331 /* report a line with the circular symbol */
332 lineno = pri->lineno;
333 fprintf(stderr,"circular reference in order list");
336 if ((pri->pri < 0) || (pri->pri >= numpri)) {
341 pri = &prilist[pri->pri];
344 if (pri->res == UNKNOWN) {
347 if (pri->res != RESOLVED)
354 weight_compare(const void *n1, const void *n2)
356 int32_t k1 = ((const weight_t *)n1)->pri;
357 int32_t k2 = ((const weight_t *)n2)->pri;
359 return (k1 < k2 ? -1 : k1 > k2 ? 1 : 0);
362 RB_GENERATE_STATIC(weights, weight, entry, weight_compare);
365 collsym_compare(const void *n1, const void *n2)
367 const collsym_t *c1 = n1;
368 const collsym_t *c2 = n2;
371 rv = strcmp(c1->name, c2->name);
372 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0);
375 RB_GENERATE_STATIC(collsyms, collsym, entry, collsym_compare);
378 collundef_compare(const void *n1, const void *n2)
380 const collundef_t *c1 = n1;
381 const collundef_t *c2 = n2;
384 rv = strcmp(c1->name, c2->name);
385 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0);
388 RB_GENERATE_STATIC(collundefs, collundef, entry, collundef_compare);
391 element_compare_symbol(const void *n1, const void *n2)
393 const collelem_t *c1 = n1;
394 const collelem_t *c2 = n2;
397 rv = strcmp(c1->symbol, c2->symbol);
398 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0);
401 RB_GENERATE_STATIC(elem_by_symbol, collelem, rb_bysymbol, element_compare_symbol);
404 element_compare_expand(const void *n1, const void *n2)
406 const collelem_t *c1 = n1;
407 const collelem_t *c2 = n2;
410 rv = wcscmp(c1->expand, c2->expand);
411 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0);
414 RB_GENERATE_STATIC(elem_by_expand, collelem, rb_byexpand, element_compare_expand);
417 collchar_compare(const void *n1, const void *n2)
419 wchar_t k1 = ((const collchar_t *)n1)->wc;
420 wchar_t k2 = ((const collchar_t *)n2)->wc;
422 return (k1 < k2 ? -1 : k1 > k2 ? 1 : 0);
425 RB_GENERATE_STATIC(collchars, collchar, entry, collchar_compare);
428 subst_compare(const void *n1, const void *n2)
430 int32_t k1 = ((const subst_t *)n1)->key;
431 int32_t k2 = ((const subst_t *)n2)->key;
433 return (k1 < k2 ? -1 : k1 > k2 ? 1 : 0);
436 RB_GENERATE_STATIC(substs, subst, entry, subst_compare);
438 #pragma GCC diagnostic push
439 #pragma GCC diagnostic ignored "-Wcast-qual"
442 subst_compare_ref(const void *n1, const void *n2)
444 int32_t *c1 = ((subst_t *)n1)->ref;
445 int32_t *c2 = ((subst_t *)n2)->ref;
448 rv = wcscmp((wchar_t *)c1, (wchar_t *)c2);
449 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0);
452 RB_GENERATE_STATIC(substs_ref, subst, entry_ref, subst_compare_ref);
454 #pragma GCC diagnostic pop
463 RB_INIT(&collundefs);
465 RB_INIT(&elem_by_symbol);
467 RB_INIT(&elem_by_expand);
471 for (i = 0; i < COLL_WEIGHTS_MAX; i++) {
473 RB_INIT(&substs_ref[i]);
474 RB_INIT(&weights[i]);
478 (void) memset(&collinfo, 0, sizeof (collinfo));
480 /* allocate some initial priorities */
481 pri_ignore = new_pri();
483 set_pri(pri_ignore, 0, RESOLVED);
485 for (i = 0; i < COLL_WEIGHTS_MAX; i++) {
486 pri_undefined[i] = new_pri();
488 /* we will override this later */
489 set_pri(pri_undefined[i], COLLATE_MAX_PRIORITY, UNKNOWN);
494 define_collsym(char *name)
498 if ((sym = calloc(sizeof (*sym), 1)) == NULL) {
499 fprintf(stderr,"out of memory");
503 sym->ref = new_pri();
505 if (RB_FIND(collsyms, &collsyms, sym) != NULL) {
507 * This should never happen because we are only called
508 * for undefined symbols.
513 RB_INSERT(collsyms, &collsyms, sym);
517 lookup_collsym(char *name)
522 return (RB_FIND(collsyms, &collsyms, &srch));
526 lookup_collelem(char *symbol)
530 srch.symbol = symbol;
531 return (RB_FIND(elem_by_symbol, &elem_by_symbol, &srch));
535 get_collundef(char *name)
542 if ((ud = RB_FIND(collundefs, &collundefs, &srch)) == NULL) {
543 if (((ud = calloc(sizeof (*ud), 1)) == NULL) ||
544 ((ud->name = strdup(name)) == NULL)) {
545 fprintf(stderr,"out of memory");
548 for (i = 0; i < NUM_WT; i++) {
549 ud->ref[i] = new_pri();
551 RB_INSERT(collundefs, &collundefs, ud);
553 add_charmap_undefined(name);
558 get_collchar(wchar_t wc, int create)
565 cc = RB_FIND(collchars, &collchars, &srch);
566 if ((cc == NULL) && create) {
567 if ((cc = calloc(sizeof (*cc), 1)) == NULL) {
568 fprintf(stderr, "out of memory");
571 for (i = 0; i < NUM_WT; i++) {
572 cc->ref[i] = new_pri();
575 RB_INSERT(collchars, &collchars, cc);
581 end_order_collsym(collsym_t *sym)
583 start_order(T_COLLSYM);
584 /* update the weight */
586 set_pri(sym->ref, nextpri, RESOLVED);
598 /* advance the priority/weight */
603 for (i = 0; i < NUM_WT; i++) {
604 if (((ref = order_weights[i]) < 0) ||
605 ((p = get_pri(ref)) == NULL) ||
607 /* unspecified weight is a self reference */
608 set_pri(currchar->ref[i], pri, RESOLVED);
610 set_pri(currchar->ref[i], ref, REFER);
612 order_weights[i] = -1;
615 /* leave a cookie trail in case next symbol is ellipsis */
616 ellipsis_start = currchar->wc + 1;
621 /* save off the weights were we can find them */
622 for (i = 0; i < NUM_WT; i++) {
623 ellipsis_weights[i] = order_weights[i];
624 order_weights[i] = -1;
629 if (currelem == NULL) {
632 for (i = 0; i < NUM_WT; i++) {
634 if (((ref = order_weights[i]) < 0) ||
635 ((p = get_pri(ref)) == NULL) ||
637 set_pri(currelem->ref[i], pri,
640 set_pri(currelem->ref[i], ref, REFER);
642 order_weights[i] = -1;
648 for (i = 0; i < NUM_WT; i++) {
649 if (((ref = order_weights[i]) < 0) ||
650 ((p = get_pri(ref)) == NULL) ||
652 set_pri(pri_undefined[i], -1, RESOLVED);
654 set_pri(pri_undefined[i], ref, REFER);
656 order_weights[i] = -1;
661 for (i = 0; i < NUM_WT; i++) {
662 if (((ref = order_weights[i]) < 0) ||
663 ((p = get_pri(ref)) == NULL) ||
665 set_pri(currundef->ref[i], pri, RESOLVED);
667 set_pri(currundef->ref[i], ref, REFER);
669 order_weights[i] = -1;
681 start_order(int type)
685 lastorder = currorder;
688 /* this is used to protect ELLIPSIS processing */
689 if ((lastorder == T_ELLIPSIS) && (type != T_CHAR)) {
690 fprintf(stderr, "character value expected");
693 for (i = 0; i < COLL_WEIGHTS_MAX; i++) {
694 order_weights[i] = -1;
700 start_order_undefined(void)
702 start_order(T_UNDEFINED);
706 start_order_symbol(char *name)
708 currundef = get_collundef(name);
709 start_order(T_SYMBOL);
713 start_order_char(wchar_t wc)
721 * If we last saw an ellipsis, then we need to close the range.
722 * Handle that here. Note that we have to be careful because the
723 * items *inside* the range are treated exclusiveley to the items
724 * outside of the range. The ends of the range can have quite
725 * different weights than the range members.
727 if (lastorder == T_ELLIPSIS) {
730 if (wc < ellipsis_start) {
731 fprintf(stderr, "malformed range!");
734 while (ellipsis_start < wc) {
736 * pick all of the saved weights for the
737 * ellipsis. note that -1 encodes for the
738 * ellipsis itself, which means to take the
739 * current relative priority.
741 if ((cc = get_collchar(ellipsis_start, 1)) == NULL) {
745 for (i = 0; i < NUM_WT; i++) {
747 if (((ref = ellipsis_weights[i]) == -1) ||
748 ((p = get_pri(ref)) == NULL) ||
750 set_pri(cc->ref[i], nextpri, RESOLVED);
752 set_pri(cc->ref[i], ref, REFER);
754 ellipsis_weights[i] = 0;
761 currchar = get_collchar(wc, 1);
765 start_order_collelem(collelem_t *e)
767 start_order(T_COLLELEM);
772 start_order_ellipsis(void)
776 start_order(T_ELLIPSIS);
778 if (lastorder != T_CHAR) {
779 fprintf(stderr, "illegal starting point for range");
783 for (i = 0; i < NUM_WT; i++) {
784 ellipsis_weights[i] = order_weights[i];
789 define_collelem(char *name, wchar_t *wcs)
794 if (wcslen(wcs) >= COLLATE_STR_LEN) {
795 fprintf(stderr,"expanded collation element too long");
799 if ((e = calloc(sizeof (*e), 1)) == NULL) {
800 fprintf(stderr, "out of memory");
807 * This is executed before the order statement, so we don't
808 * know how many priorities we *really* need. We allocate one
809 * for each possible weight. Not a big deal, as collating-elements
810 * prove to be quite rare.
812 for (i = 0; i < COLL_WEIGHTS_MAX; i++) {
813 e->ref[i] = new_pri();
816 /* A character sequence can only reduce to one element. */
817 if ((RB_FIND(elem_by_symbol, &elem_by_symbol, e) != NULL) ||
818 (RB_FIND(elem_by_expand, &elem_by_expand, e) != NULL)) {
819 fprintf(stderr, "duplicate collating element definition");
822 RB_INSERT(elem_by_symbol, &elem_by_symbol, e);
823 RB_INSERT(elem_by_expand, &elem_by_expand, e);
827 add_order_bit(int kw)
829 uint8_t bit = DIRECTIVE_UNDEF;
833 bit = DIRECTIVE_FORWARD;
836 bit = DIRECTIVE_BACKWARD;
839 bit = DIRECTIVE_POSITION;
845 collinfo.directive[collinfo.directive_count] |= bit;
849 add_order_directive(void)
851 if (collinfo.directive_count >= COLL_WEIGHTS_MAX) {
852 fprintf(stderr,"too many directives (max %d)", COLL_WEIGHTS_MAX);
854 collinfo.directive_count++;
858 add_order_pri(int32_t ref)
860 if (curr_weight >= NUM_WT) {
861 fprintf(stderr,"too many weights (max %d)", NUM_WT);
864 order_weights[curr_weight] = ref;
869 add_order_collsym(collsym_t *s)
871 add_order_pri(s->ref);
875 add_order_char(wchar_t wc)
879 if ((cc = get_collchar(wc, 1)) == NULL) {
884 add_order_pri(cc->ref[curr_weight]);
888 add_order_collelem(collelem_t *e)
890 add_order_pri(e->ref[curr_weight]);
894 add_order_ignore(void)
896 add_order_pri(pri_ignore);
900 add_order_symbol(char *sym)
903 if ((c = get_collundef(sym)) == NULL) {
907 add_order_pri(c->ref[curr_weight]);
911 add_order_ellipsis(void)
913 /* special NULL value indicates self reference */
918 add_order_subst(void)
924 (void) memset(&srch, 0, sizeof (srch));
925 for (i = 0; i < curr_subst; i++) {
926 srch.ref[i] = subst_weights[i];
927 subst_weights[i] = 0;
929 s = RB_FIND(substs_ref, &substs_ref[curr_weight], &srch);
932 if ((s = calloc(sizeof (*s), 1)) == NULL) {
933 fprintf(stderr,"out of memory");
939 * We use a self reference for our key, but we set a
940 * high bit to indicate that this is a substitution
941 * reference. This will expedite table lookups later,
942 * and prevent table lookups for situations that don't
943 * require it. (In short, its a big win, because we
944 * can skip a lot of binary searching.)
947 (nextsubst[curr_weight] | COLLATE_SUBST_PRIORITY),
949 nextsubst[curr_weight] += 1;
951 for (i = 0; i < curr_subst; i++) {
952 s->ref[i] = srch.ref[i];
955 RB_INSERT(substs_ref, &substs_ref[curr_weight], s);
957 if (RB_FIND(substs, &substs[curr_weight], s) != NULL) {
961 RB_INSERT(substs, &substs[curr_weight], s);
967 * We are using the current (unique) priority as a search key
968 * in the substitution table.
970 add_order_pri(s->key);
974 add_subst_pri(int32_t ref)
976 if (curr_subst >= COLLATE_STR_LEN) {
977 fprintf(stderr,"substitution string is too long");
980 subst_weights[curr_subst] = ref;
985 add_subst_char(wchar_t wc)
990 if (((cc = get_collchar(wc, 1)) == NULL) ||
995 /* we take the weight for the character at that position */
996 add_subst_pri(cc->ref[curr_weight]);
1000 add_subst_collelem(collelem_t *e)
1002 add_subst_pri(e->ref[curr_weight]);
1006 add_subst_collsym(collsym_t *s)
1008 add_subst_pri(s->ref);
1012 add_subst_symbol(char *ptr)
1016 if ((cu = get_collundef(ptr)) != NULL) {
1017 add_subst_pri(cu->ref[curr_weight]);
1022 add_weight(int32_t ref, int pass)
1027 srch.pri = resolve_pri(ref);
1029 /* No translation of ignores */
1033 /* Substitution priorities are not weights */
1034 if (srch.pri & COLLATE_SUBST_PRIORITY)
1037 if (RB_FIND(weights, &weights[pass], &srch) != NULL)
1040 if ((w = calloc(sizeof (*w), 1)) == NULL) {
1041 fprintf(stderr, "out of memory");
1045 RB_INSERT(weights, &weights[pass], w);
1049 add_weights(int32_t *refs)
1052 for (i = 0; i < NUM_WT; i++) {
1053 add_weight(refs[i], i);
1058 get_weight(int32_t ref, int pass)
1064 pri = resolve_pri(ref);
1065 if (pri & COLLATE_SUBST_PRIORITY) {
1072 if ((w = RB_FIND(weights, &weights[pass], &srch)) == NULL) {
1080 wsncpy(wchar_t *s1, const wchar_t *s2, size_t n)
1085 while (--n > 0 && (*s1++ = *s2++) != 0)
1093 #define RB_COUNT(x, name, head, cnt) do { \
1095 RB_FOREACH(x, name, (head)) { \
1100 #define RB_NUMNODES(type, name, head, cnt) do { \
1103 RB_FOREACH(t, name, head) { \
1118 char vers[COLLATE_STR_LEN];
1119 collate_char_t chars[UCHAR_MAX + 1];
1120 collate_large_t *large;
1121 collate_subst_t *subst[COLL_WEIGHTS_MAX];
1122 collate_chain_t *chain;
1125 * We have to run throught a preliminary pass to identify all the
1126 * weights that we use for each sorting level.
1128 for (i = 0; i < NUM_WT; i++) {
1129 add_weight(pri_ignore, i);
1131 for (i = 0; i < NUM_WT; i++) {
1132 RB_FOREACH(sb, substs, &substs[i]) {
1133 for (j = 0; sb->ref[j]; j++) {
1134 add_weight(sb->ref[j], i);
1138 RB_FOREACH(ce, elem_by_expand, &elem_by_expand) {
1139 add_weights(ce->ref);
1141 RB_FOREACH(cc, collchars, &collchars) {
1142 add_weights(cc->ref);
1146 * Now we walk the entire set of weights, removing the gaps
1147 * in the weights. This gives us optimum usage. The walk
1148 * occurs in priority.
1150 for (i = 0; i < NUM_WT; i++) {
1152 RB_FOREACH(w, weights, &weights[i]) {
1153 w->opt = nweight[i];
1158 (void) memset(&chars, 0, sizeof (chars));
1159 (void) memset(vers, 0, COLLATE_STR_LEN);
1160 (void) strlcpy(vers, COLLATE_VERSION, sizeof (vers));
1163 * We need to make sure we arrange for the UNDEFINED field
1164 * to show up. Also, set the total weight counts.
1166 for (i = 0; i < NUM_WT; i++) {
1167 if (resolve_pri(pri_undefined[i]) == -1) {
1168 set_pri(pri_undefined[i], -1, RESOLVED);
1169 /* they collate at the end of everything else */
1170 collinfo.undef_pri[i] = COLLATE_MAX_PRIORITY;
1172 collinfo.pri_count[i] = nweight[i];
1175 collinfo.pri_count[NUM_WT] = max_wide();
1176 collinfo.undef_pri[NUM_WT] = COLLATE_MAX_PRIORITY;
1177 collinfo.directive[NUM_WT] = DIRECTIVE_UNDEFINED;
1180 * Ordinary character priorities
1182 for (i = 0; i <= UCHAR_MAX; i++) {
1183 if ((cc = get_collchar(i, 0)) != NULL) {
1184 for (j = 0; j < NUM_WT; j++) {
1185 chars[i].pri[j] = get_weight(cc->ref[j], j);
1188 for (j = 0; j < NUM_WT; j++) {
1190 get_weight(pri_undefined[j], j);
1193 * Per POSIX, for undefined characters, we
1194 * also have to add a last item, which is the
1197 chars[i].pri[NUM_WT] = i;
1202 * Substitution tables
1204 for (i = 0; i < NUM_WT; i++) {
1205 collate_subst_t *st = NULL;
1207 RB_COUNT(temp, substs, &substs[i], n);
1208 collinfo.subst_count[i] = n;
1209 if ((st = calloc(sizeof (collate_subst_t) * n, 1)) == NULL) {
1210 fprintf(stderr, "out of memory");
1214 RB_FOREACH(sb, substs, &substs[i]) {
1215 if ((st[n].key = resolve_pri(sb->key)) < 0) {
1216 /* by definition these resolve! */
1219 if (st[n].key != (n | COLLATE_SUBST_PRIORITY)) {
1222 for (j = 0; sb->ref[j]; j++) {
1223 st[n].pri[j] = get_weight(sb->ref[j], i);
1227 if (n != collinfo.subst_count[i])
1234 * Chains, i.e. collating elements
1236 RB_NUMNODES(collelem_t, elem_by_expand, &elem_by_expand,
1237 collinfo.chain_count);
1238 chain = calloc(sizeof (collate_chain_t), collinfo.chain_count);
1239 if (chain == NULL) {
1240 fprintf(stderr, "out of memory");
1244 RB_FOREACH(ce, elem_by_expand, &elem_by_expand) {
1246 (void) wsncpy(chain[n].str, ce->expand, COLLATE_STR_LEN);
1247 for (i = 0; i < NUM_WT; i++) {
1248 chain[n].pri[i] = get_weight(ce->ref[i], i);
1251 if (n != collinfo.chain_count)
1255 * Large (> UCHAR_MAX) character priorities
1257 RB_NUMNODES(collchar_t, collchars, &collchars, n);
1258 large = malloc(sizeof (collate_large_t) * n);
1259 if (large == NULL) {
1260 fprintf(stderr, "out of memory");
1263 memset(large, 0, sizeof (collate_large_t) * n);
1266 RB_FOREACH(cc, collchars, &collchars) {
1268 /* we already gathered those */
1269 if (cc->wc <= UCHAR_MAX)
1271 for (j = 0; j < NUM_WT; j++) {
1272 if ((pri = get_weight(cc->ref[j], j)) < 0) {
1275 if (undef && (pri >= 0)) {
1276 /* if undefined, then all priorities are */
1279 large[i].pri.pri[j] = pri;
1283 large[i].val = cc->wc;
1284 collinfo.large_count = i++;
1288 if ((f = open_category()) == NULL) {
1292 /* Time to write the entire data set out */
1294 if ((wr_category(vers, COLLATE_STR_LEN, f) < 0) ||
1295 (wr_category(&collinfo, sizeof (collinfo), f) < 0) ||
1296 (wr_category(&chars, sizeof (chars), f) < 0)) {
1300 for (i = 0; i < NUM_WT; i++) {
1301 sz = sizeof (collate_subst_t) * collinfo.subst_count[i];
1302 if (wr_category(subst[i], sz, f) < 0) {
1306 sz = sizeof (collate_chain_t) * collinfo.chain_count;
1307 if (wr_category(chain, sz, f) < 0) {
1310 sz = sizeof (collate_large_t) * collinfo.large_count;
1311 if (wr_category(large, sz, f) < 0) {