collate.c revision 286484
1/* 2 * Copyright 2010 Nexenta Systems, Inc. All rights reserved. 3 * Copyright 2015 John Marino <draco@marino.st> 4 * 5 * This source code is derived from the illumos localedef command, and 6 * provided under BSD-style license terms by Nexenta Systems, Inc. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 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. 17 * 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. 29 */ 30 31/* 32 * LC_COLLATE database generation routines for localedef. 33 */ 34#include <sys/cdefs.h> 35__FBSDID("$FreeBSD: projects/collation/usr.bin/localedef/collate.c 286484 2015-08-08 22:57:17Z bapt $"); 36 37#include <sys/types.h> 38#include <sys/tree.h> 39 40#include <stdio.h> 41#include <stddef.h> 42#include <stdlib.h> 43#include <errno.h> 44#include <string.h> 45#include <unistd.h> 46#include <wchar.h> 47#include <limits.h> 48#include "localedef.h" 49#include "parser.h" 50#include "collate.h" 51 52/* 53 * Design notes. 54 * 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. 58 * 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: 62 * 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 67 * table. 68 * 69 * b) We eliminate duplicate entries into the substitution table. 70 * This saves a lot of space. 71 * 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 77 * smaller. 78 * 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. 81 * 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. 85 * 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. 89 * 90 * To process the entire data to make the database, we actually run 91 * multiple passes over the data. 92 * 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 98 * reference. 99 * 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". 104 * 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. 109 * 110 * The fourth pass just writes the data out. 111 */ 112 113/* 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. 116 * 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. 121 * 122 * RESOLVED is used for an entry where the priority indicates the final 123 * numeric weight. 124 * 125 * REFER is used for entries that reference other entries. Typically 126 * this is used for forward references. A collating-symbol can never 127 * have this value. 128 * 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>.) 132 */ 133typedef enum { 134 UNKNOWN, /* priority is totally unknown */ 135 RESOLVED, /* priority value fully resolved */ 136 REFER /* priority is a reference (index) */ 137} res_t; 138 139typedef struct weight { 140 int32_t pri; 141 int opt; 142 RB_ENTRY(weight) entry; 143} weight_t; 144 145typedef struct priority { 146 res_t res; 147 int32_t pri; 148 int pass; 149 int lineno; 150} collpri_t; 151 152#define NUM_WT collinfo.directive_count 153 154/* 155 * These are the abstract collating symbols, which are just a symbolic 156 * way to reference a priority. 157 */ 158struct collsym { 159 char *name; 160 int32_t ref; 161 RB_ENTRY(collsym) entry; 162}; 163 164/* 165 * These are also abstract collating symbols, but we allow them to have 166 * different priorities at different levels. 167 */ 168typedef struct collundef { 169 char *name; 170 int32_t ref[COLL_WEIGHTS_MAX]; 171 RB_ENTRY(collundef) entry; 172} collundef_t; 173 174/* 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>. 179 */ 180struct collelem { 181 char *symbol; 182 wchar_t *expand; 183 int32_t ref[COLL_WEIGHTS_MAX]; 184 RB_ENTRY(collelem) rb_bysymbol; 185 RB_ENTRY(collelem) rb_byexpand; 186}; 187 188/* 189 * Individual characters have a sequence of weights as well. 190 */ 191typedef struct collchar { 192 wchar_t wc; 193 int32_t ref[COLL_WEIGHTS_MAX]; 194 RB_ENTRY(collchar) entry; 195} collchar_t; 196 197/* 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. 202 */ 203typedef struct subst{ 204 int32_t key; 205 int32_t ref[COLLATE_STR_LEN]; 206 RB_ENTRY(subst) entry; 207 RB_ENTRY(subst) entry_ref; 208} subst_t; 209 210static RB_HEAD(collsyms, collsym) collsyms; 211static RB_HEAD(collundefs, collundef) collundefs; 212static RB_HEAD(elem_by_symbol, collelem) elem_by_symbol; 213static RB_HEAD(elem_by_expand, collelem) elem_by_expand; 214static RB_HEAD(collchars, collchar) collchars; 215static RB_HEAD(substs, subst) substs[COLL_WEIGHTS_MAX]; 216static RB_HEAD(substs_ref, subst) substs_ref[COLL_WEIGHTS_MAX]; 217static RB_HEAD(weights, weight) weights[COLL_WEIGHTS_MAX]; 218static int32_t nweight[COLL_WEIGHTS_MAX]; 219 220/* 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. 226 */ 227static int currorder = EOF; 228static int lastorder = EOF; 229static collelem_t *currelem; 230static collchar_t *currchar; 231static collundef_t *currundef; 232static wchar_t ellipsis_start = 0; 233static int32_t ellipsis_weights[COLL_WEIGHTS_MAX]; 234 235/* 236 * We keep a running tally of weights. 237 */ 238static int nextpri = 1; 239static int nextsubst[COLL_WEIGHTS_MAX] = { 0 }; 240 241/* 242 * This array collects up the weights for each level. 243 */ 244static int32_t order_weights[COLL_WEIGHTS_MAX]; 245static int curr_weight = 0; 246static int32_t subst_weights[COLLATE_STR_LEN]; 247static int curr_subst = 0; 248 249/* 250 * Some initial priority values. 251 */ 252static int32_t pri_undefined[COLL_WEIGHTS_MAX]; 253static int32_t pri_ignore; 254 255static collate_info_t collinfo; 256 257static collpri_t *prilist = NULL; 258static int numpri = 0; 259static int maxpri = 0; 260 261static void start_order(int); 262 263static int32_t 264new_pri(void) 265{ 266 int i; 267 268 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"); 273 return (-1); 274 } 275 for (i = numpri; i < maxpri; i++) { 276 prilist[i].res = UNKNOWN; 277 prilist[i].pri = 0; 278 prilist[i].pass = 0; 279 } 280 } 281 return (numpri++); 282} 283 284static collpri_t * 285get_pri(int32_t ref) 286{ 287 if ((ref < 0) || (ref > numpri)) { 288 INTERR; 289 return (NULL); 290 } 291 return (&prilist[ref]); 292} 293 294static void 295set_pri(int32_t ref, int32_t v, res_t res) 296{ 297 collpri_t *pri; 298 299 pri = get_pri(ref); 300 301 if ((res == REFER) && ((v < 0) || (v >= numpri))) { 302 INTERR; 303 } 304 305 /* Resolve self references */ 306 if ((res == REFER) && (ref == v)) { 307 v = nextpri; 308 res = RESOLVED; 309 } 310 311 if (pri->res != UNKNOWN) { 312 warn("repeated item in order list (first on %d)", 313 pri->lineno); 314 return; 315 } 316 pri->lineno = lineno; 317 pri->pri = v; 318 pri->res = res; 319} 320 321static int32_t 322resolve_pri(int32_t ref) 323{ 324 collpri_t *pri; 325 static int32_t pass = 0; 326 327 pri = get_pri(ref); 328 pass++; 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"); 334 return (-1); 335 } 336 if ((pri->pri < 0) || (pri->pri >= numpri)) { 337 INTERR; 338 return (-1); 339 } 340 pri->pass = pass; 341 pri = &prilist[pri->pri]; 342 } 343 344 if (pri->res == UNKNOWN) { 345 return (-1); 346 } 347 if (pri->res != RESOLVED) 348 INTERR; 349 350 return (pri->pri); 351} 352 353static int 354weight_compare(const void *n1, const void *n2) 355{ 356 int32_t k1 = ((const weight_t *)n1)->pri; 357 int32_t k2 = ((const weight_t *)n2)->pri; 358 359 return (k1 < k2 ? -1 : k1 > k2 ? 1 : 0); 360} 361 362RB_GENERATE_STATIC(weights, weight, entry, weight_compare); 363 364static int 365collsym_compare(const void *n1, const void *n2) 366{ 367 const collsym_t *c1 = n1; 368 const collsym_t *c2 = n2; 369 int rv; 370 371 rv = strcmp(c1->name, c2->name); 372 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0); 373} 374 375RB_GENERATE_STATIC(collsyms, collsym, entry, collsym_compare); 376 377static int 378collundef_compare(const void *n1, const void *n2) 379{ 380 const collundef_t *c1 = n1; 381 const collundef_t *c2 = n2; 382 int rv; 383 384 rv = strcmp(c1->name, c2->name); 385 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0); 386} 387 388RB_GENERATE_STATIC(collundefs, collundef, entry, collundef_compare); 389 390static int 391element_compare_symbol(const void *n1, const void *n2) 392{ 393 const collelem_t *c1 = n1; 394 const collelem_t *c2 = n2; 395 int rv; 396 397 rv = strcmp(c1->symbol, c2->symbol); 398 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0); 399} 400 401RB_GENERATE_STATIC(elem_by_symbol, collelem, rb_bysymbol, element_compare_symbol); 402 403static int 404element_compare_expand(const void *n1, const void *n2) 405{ 406 const collelem_t *c1 = n1; 407 const collelem_t *c2 = n2; 408 int rv; 409 410 rv = wcscmp(c1->expand, c2->expand); 411 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0); 412} 413 414RB_GENERATE_STATIC(elem_by_expand, collelem, rb_byexpand, element_compare_expand); 415 416static int 417collchar_compare(const void *n1, const void *n2) 418{ 419 wchar_t k1 = ((const collchar_t *)n1)->wc; 420 wchar_t k2 = ((const collchar_t *)n2)->wc; 421 422 return (k1 < k2 ? -1 : k1 > k2 ? 1 : 0); 423} 424 425RB_GENERATE_STATIC(collchars, collchar, entry, collchar_compare); 426 427static int 428subst_compare(const void *n1, const void *n2) 429{ 430 int32_t k1 = ((const subst_t *)n1)->key; 431 int32_t k2 = ((const subst_t *)n2)->key; 432 433 return (k1 < k2 ? -1 : k1 > k2 ? 1 : 0); 434} 435 436RB_GENERATE_STATIC(substs, subst, entry, subst_compare); 437 438#pragma GCC diagnostic push 439#pragma GCC diagnostic ignored "-Wcast-qual" 440 441static int 442subst_compare_ref(const void *n1, const void *n2) 443{ 444 int32_t *c1 = ((subst_t *)n1)->ref; 445 int32_t *c2 = ((subst_t *)n2)->ref; 446 int rv; 447 448 rv = wcscmp((wchar_t *)c1, (wchar_t *)c2); 449 return ((rv < 0) ? -1 : (rv > 0) ? 1 : 0); 450} 451 452RB_GENERATE_STATIC(substs_ref, subst, entry_ref, subst_compare_ref); 453 454#pragma GCC diagnostic pop 455 456void 457init_collate(void) 458{ 459 int i; 460 461 RB_INIT(&collsyms); 462 463 RB_INIT(&collundefs); 464 465 RB_INIT(&elem_by_symbol); 466 467 RB_INIT(&elem_by_expand); 468 469 RB_INIT(&collchars); 470 471 for (i = 0; i < COLL_WEIGHTS_MAX; i++) { 472 RB_INIT(&substs[i]); 473 RB_INIT(&substs_ref[i]); 474 RB_INIT(&weights[i]); 475 nweight[i] = 1; 476 } 477 478 (void) memset(&collinfo, 0, sizeof (collinfo)); 479 480 /* allocate some initial priorities */ 481 pri_ignore = new_pri(); 482 483 set_pri(pri_ignore, 0, RESOLVED); 484 485 for (i = 0; i < COLL_WEIGHTS_MAX; i++) { 486 pri_undefined[i] = new_pri(); 487 488 /* we will override this later */ 489 set_pri(pri_undefined[i], COLLATE_MAX_PRIORITY, UNKNOWN); 490 } 491} 492 493void 494define_collsym(char *name) 495{ 496 collsym_t *sym; 497 498 if ((sym = calloc(sizeof (*sym), 1)) == NULL) { 499 fprintf(stderr,"out of memory"); 500 return; 501 } 502 sym->name = name; 503 sym->ref = new_pri(); 504 505 if (RB_FIND(collsyms, &collsyms, sym) != NULL) { 506 /* 507 * This should never happen because we are only called 508 * for undefined symbols. 509 */ 510 INTERR; 511 return; 512 } 513 RB_INSERT(collsyms, &collsyms, sym); 514} 515 516collsym_t * 517lookup_collsym(char *name) 518{ 519 collsym_t srch; 520 521 srch.name = name; 522 return (RB_FIND(collsyms, &collsyms, &srch)); 523} 524 525collelem_t * 526lookup_collelem(char *symbol) 527{ 528 collelem_t srch; 529 530 srch.symbol = symbol; 531 return (RB_FIND(elem_by_symbol, &elem_by_symbol, &srch)); 532} 533 534static collundef_t * 535get_collundef(char *name) 536{ 537 collundef_t srch; 538 collundef_t *ud; 539 int i; 540 541 srch.name = 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"); 546 return (NULL); 547 } 548 for (i = 0; i < NUM_WT; i++) { 549 ud->ref[i] = new_pri(); 550 } 551 RB_INSERT(collundefs, &collundefs, ud); 552 } 553 add_charmap_undefined(name); 554 return (ud); 555} 556 557static collchar_t * 558get_collchar(wchar_t wc, int create) 559{ 560 collchar_t srch; 561 collchar_t *cc; 562 int i; 563 564 srch.wc = wc; 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"); 569 return (NULL); 570 } 571 for (i = 0; i < NUM_WT; i++) { 572 cc->ref[i] = new_pri(); 573 } 574 cc->wc = wc; 575 RB_INSERT(collchars, &collchars, cc); 576 } 577 return (cc); 578} 579 580void 581end_order_collsym(collsym_t *sym) 582{ 583 start_order(T_COLLSYM); 584 /* update the weight */ 585 586 set_pri(sym->ref, nextpri, RESOLVED); 587 nextpri++; 588} 589 590void 591end_order(void) 592{ 593 int i; 594 int32_t pri; 595 int32_t ref; 596 collpri_t *p; 597 598 /* advance the priority/weight */ 599 pri = nextpri; 600 601 switch (currorder) { 602 case T_CHAR: 603 for (i = 0; i < NUM_WT; i++) { 604 if (((ref = order_weights[i]) < 0) || 605 ((p = get_pri(ref)) == NULL) || 606 (p->pri == -1)) { 607 /* unspecified weight is a self reference */ 608 set_pri(currchar->ref[i], pri, RESOLVED); 609 } else { 610 set_pri(currchar->ref[i], ref, REFER); 611 } 612 order_weights[i] = -1; 613 } 614 615 /* leave a cookie trail in case next symbol is ellipsis */ 616 ellipsis_start = currchar->wc + 1; 617 currchar = NULL; 618 break; 619 620 case T_ELLIPSIS: 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; 625 } 626 break; 627 628 case T_COLLELEM: 629 if (currelem == NULL) { 630 INTERR; 631 } else { 632 for (i = 0; i < NUM_WT; i++) { 633 634 if (((ref = order_weights[i]) < 0) || 635 ((p = get_pri(ref)) == NULL) || 636 (p->pri == -1)) { 637 set_pri(currelem->ref[i], pri, 638 RESOLVED); 639 } else { 640 set_pri(currelem->ref[i], ref, REFER); 641 } 642 order_weights[i] = -1; 643 } 644 } 645 break; 646 647 case T_UNDEFINED: 648 for (i = 0; i < NUM_WT; i++) { 649 if (((ref = order_weights[i]) < 0) || 650 ((p = get_pri(ref)) == NULL) || 651 (p->pri == -1)) { 652 set_pri(pri_undefined[i], -1, RESOLVED); 653 } else { 654 set_pri(pri_undefined[i], ref, REFER); 655 } 656 order_weights[i] = -1; 657 } 658 break; 659 660 case T_SYMBOL: 661 for (i = 0; i < NUM_WT; i++) { 662 if (((ref = order_weights[i]) < 0) || 663 ((p = get_pri(ref)) == NULL) || 664 (p->pri == -1)) { 665 set_pri(currundef->ref[i], pri, RESOLVED); 666 } else { 667 set_pri(currundef->ref[i], ref, REFER); 668 } 669 order_weights[i] = -1; 670 } 671 break; 672 673 default: 674 INTERR; 675 } 676 677 nextpri++; 678} 679 680static void 681start_order(int type) 682{ 683 int i; 684 685 lastorder = currorder; 686 currorder = type; 687 688 /* this is used to protect ELLIPSIS processing */ 689 if ((lastorder == T_ELLIPSIS) && (type != T_CHAR)) { 690 fprintf(stderr, "character value expected"); 691 } 692 693 for (i = 0; i < COLL_WEIGHTS_MAX; i++) { 694 order_weights[i] = -1; 695 } 696 curr_weight = 0; 697} 698 699void 700start_order_undefined(void) 701{ 702 start_order(T_UNDEFINED); 703} 704 705void 706start_order_symbol(char *name) 707{ 708 currundef = get_collundef(name); 709 start_order(T_SYMBOL); 710} 711 712void 713start_order_char(wchar_t wc) 714{ 715 collchar_t *cc; 716 int32_t ref; 717 718 start_order(T_CHAR); 719 720 /* 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. 726 */ 727 if (lastorder == T_ELLIPSIS) { 728 int i; 729 730 if (wc < ellipsis_start) { 731 fprintf(stderr, "malformed range!"); 732 return; 733 } 734 while (ellipsis_start < wc) { 735 /* 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. 740 */ 741 if ((cc = get_collchar(ellipsis_start, 1)) == NULL) { 742 INTERR; 743 return; 744 } 745 for (i = 0; i < NUM_WT; i++) { 746 collpri_t *p; 747 if (((ref = ellipsis_weights[i]) == -1) || 748 ((p = get_pri(ref)) == NULL) || 749 (p->pri == -1)) { 750 set_pri(cc->ref[i], nextpri, RESOLVED); 751 } else { 752 set_pri(cc->ref[i], ref, REFER); 753 } 754 ellipsis_weights[i] = 0; 755 } 756 ellipsis_start++; 757 nextpri++; 758 } 759 } 760 761 currchar = get_collchar(wc, 1); 762} 763 764void 765start_order_collelem(collelem_t *e) 766{ 767 start_order(T_COLLELEM); 768 currelem = e; 769} 770 771void 772start_order_ellipsis(void) 773{ 774 int i; 775 776 start_order(T_ELLIPSIS); 777 778 if (lastorder != T_CHAR) { 779 fprintf(stderr, "illegal starting point for range"); 780 return; 781 } 782 783 for (i = 0; i < NUM_WT; i++) { 784 ellipsis_weights[i] = order_weights[i]; 785 } 786} 787 788void 789define_collelem(char *name, wchar_t *wcs) 790{ 791 collelem_t *e; 792 int i; 793 794 if (wcslen(wcs) >= COLLATE_STR_LEN) { 795 fprintf(stderr,"expanded collation element too long"); 796 return; 797 } 798 799 if ((e = calloc(sizeof (*e), 1)) == NULL) { 800 fprintf(stderr, "out of memory"); 801 return; 802 } 803 e->expand = wcs; 804 e->symbol = name; 805 806 /* 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. 811 */ 812 for (i = 0; i < COLL_WEIGHTS_MAX; i++) { 813 e->ref[i] = new_pri(); 814 } 815 816 /* 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"); 820 return; 821 } 822 RB_INSERT(elem_by_symbol, &elem_by_symbol, e); 823 RB_INSERT(elem_by_expand, &elem_by_expand, e); 824} 825 826void 827add_order_bit(int kw) 828{ 829 uint8_t bit = DIRECTIVE_UNDEF; 830 831 switch (kw) { 832 case T_FORWARD: 833 bit = DIRECTIVE_FORWARD; 834 break; 835 case T_BACKWARD: 836 bit = DIRECTIVE_BACKWARD; 837 break; 838 case T_POSITION: 839 bit = DIRECTIVE_POSITION; 840 break; 841 default: 842 INTERR; 843 break; 844 } 845 collinfo.directive[collinfo.directive_count] |= bit; 846} 847 848void 849add_order_directive(void) 850{ 851 if (collinfo.directive_count >= COLL_WEIGHTS_MAX) { 852 fprintf(stderr,"too many directives (max %d)", COLL_WEIGHTS_MAX); 853 } 854 collinfo.directive_count++; 855} 856 857static void 858add_order_pri(int32_t ref) 859{ 860 if (curr_weight >= NUM_WT) { 861 fprintf(stderr,"too many weights (max %d)", NUM_WT); 862 return; 863 } 864 order_weights[curr_weight] = ref; 865 curr_weight++; 866} 867 868void 869add_order_collsym(collsym_t *s) 870{ 871 add_order_pri(s->ref); 872} 873 874void 875add_order_char(wchar_t wc) 876{ 877 collchar_t *cc; 878 879 if ((cc = get_collchar(wc, 1)) == NULL) { 880 INTERR; 881 return; 882 } 883 884 add_order_pri(cc->ref[curr_weight]); 885} 886 887void 888add_order_collelem(collelem_t *e) 889{ 890 add_order_pri(e->ref[curr_weight]); 891} 892 893void 894add_order_ignore(void) 895{ 896 add_order_pri(pri_ignore); 897} 898 899void 900add_order_symbol(char *sym) 901{ 902 collundef_t *c; 903 if ((c = get_collundef(sym)) == NULL) { 904 INTERR; 905 return; 906 } 907 add_order_pri(c->ref[curr_weight]); 908} 909 910void 911add_order_ellipsis(void) 912{ 913 /* special NULL value indicates self reference */ 914 add_order_pri(0); 915} 916 917void 918add_order_subst(void) 919{ 920 subst_t srch; 921 subst_t *s; 922 int i; 923 924 (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; 928 } 929 s = RB_FIND(substs_ref, &substs_ref[curr_weight], &srch); 930 931 if (s == NULL) { 932 if ((s = calloc(sizeof (*s), 1)) == NULL) { 933 fprintf(stderr,"out of memory"); 934 return; 935 } 936 s->key = new_pri(); 937 938 /* 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.) 945 */ 946 set_pri(s->key, 947 (nextsubst[curr_weight] | COLLATE_SUBST_PRIORITY), 948 RESOLVED); 949 nextsubst[curr_weight] += 1; 950 951 for (i = 0; i < curr_subst; i++) { 952 s->ref[i] = srch.ref[i]; 953 } 954 955 RB_INSERT(substs_ref, &substs_ref[curr_weight], s); 956 957 if (RB_FIND(substs, &substs[curr_weight], s) != NULL) { 958 INTERR; 959 return; 960 } 961 RB_INSERT(substs, &substs[curr_weight], s); 962 } 963 curr_subst = 0; 964 965 966 /* 967 * We are using the current (unique) priority as a search key 968 * in the substitution table. 969 */ 970 add_order_pri(s->key); 971} 972 973static void 974add_subst_pri(int32_t ref) 975{ 976 if (curr_subst >= COLLATE_STR_LEN) { 977 fprintf(stderr,"substitution string is too long"); 978 return; 979 } 980 subst_weights[curr_subst] = ref; 981 curr_subst++; 982} 983 984void 985add_subst_char(wchar_t wc) 986{ 987 collchar_t *cc; 988 989 990 if (((cc = get_collchar(wc, 1)) == NULL) || 991 (cc->wc != wc)) { 992 INTERR; 993 return; 994 } 995 /* we take the weight for the character at that position */ 996 add_subst_pri(cc->ref[curr_weight]); 997} 998 999void 1000add_subst_collelem(collelem_t *e) 1001{ 1002 add_subst_pri(e->ref[curr_weight]); 1003} 1004 1005void 1006add_subst_collsym(collsym_t *s) 1007{ 1008 add_subst_pri(s->ref); 1009} 1010 1011void 1012add_subst_symbol(char *ptr) 1013{ 1014 collundef_t *cu; 1015 1016 if ((cu = get_collundef(ptr)) != NULL) { 1017 add_subst_pri(cu->ref[curr_weight]); 1018 } 1019} 1020 1021void 1022add_weight(int32_t ref, int pass) 1023{ 1024 weight_t srch; 1025 weight_t *w; 1026 1027 srch.pri = resolve_pri(ref); 1028 1029 /* No translation of ignores */ 1030 if (srch.pri == 0) 1031 return; 1032 1033 /* Substitution priorities are not weights */ 1034 if (srch.pri & COLLATE_SUBST_PRIORITY) 1035 return; 1036 1037 if (RB_FIND(weights, &weights[pass], &srch) != NULL) 1038 return; 1039 1040 if ((w = calloc(sizeof (*w), 1)) == NULL) { 1041 fprintf(stderr, "out of memory"); 1042 return; 1043 } 1044 w->pri = srch.pri; 1045 RB_INSERT(weights, &weights[pass], w); 1046} 1047 1048void 1049add_weights(int32_t *refs) 1050{ 1051 int i; 1052 for (i = 0; i < NUM_WT; i++) { 1053 add_weight(refs[i], i); 1054 } 1055} 1056 1057int32_t 1058get_weight(int32_t ref, int pass) 1059{ 1060 weight_t srch; 1061 weight_t *w; 1062 int32_t pri; 1063 1064 pri = resolve_pri(ref); 1065 if (pri & COLLATE_SUBST_PRIORITY) { 1066 return (pri); 1067 } 1068 if (pri <= 0) { 1069 return (pri); 1070 } 1071 srch.pri = pri; 1072 if ((w = RB_FIND(weights, &weights[pass], &srch)) == NULL) { 1073 INTERR; 1074 return (-1); 1075 } 1076 return (w->opt); 1077} 1078 1079wchar_t * 1080wsncpy(wchar_t *s1, const wchar_t *s2, size_t n) 1081{ 1082 wchar_t *os1 = s1; 1083 1084 n++; 1085 while (--n > 0 && (*s1++ = *s2++) != 0) 1086 continue; 1087 if (n > 0) 1088 while (--n > 0) 1089 *s1++ = 0; 1090 return (os1); 1091} 1092 1093#define RB_NUMNODES(type, name, head, cnt) do { \ 1094 type *t; \ 1095 cnt = 0; \ 1096 RB_FOREACH(t, name, head) { \ 1097 cnt++; \ 1098 } \ 1099} while (0); 1100 1101void 1102dump_collate(void) 1103{ 1104 FILE *f; 1105 int i, j, n; 1106 size_t sz; 1107 int32_t pri; 1108 collelem_t *ce; 1109 collchar_t *cc; 1110 subst_t *sb; 1111 char vers[COLLATE_STR_LEN]; 1112 collate_char_t chars[UCHAR_MAX + 1]; 1113 collate_large_t *large; 1114 collate_subst_t *subst[COLL_WEIGHTS_MAX]; 1115 collate_chain_t *chain; 1116 1117 /* 1118 * We have to run throught a preliminary pass to identify all the 1119 * weights that we use for each sorting level. 1120 */ 1121 for (i = 0; i < NUM_WT; i++) { 1122 add_weight(pri_ignore, i); 1123 } 1124 for (i = 0; i < NUM_WT; i++) { 1125 RB_FOREACH(sb, substs, &substs[i]) { 1126 for (j = 0; sb->ref[j]; j++) { 1127 add_weight(sb->ref[j], i); 1128 } 1129 } 1130 } 1131 RB_FOREACH(ce, elem_by_expand, &elem_by_expand) { 1132 add_weights(ce->ref); 1133 } 1134 RB_FOREACH(cc, collchars, &collchars) { 1135 add_weights(cc->ref); 1136 } 1137 1138 /* 1139 * Now we walk the entire set of weights, removing the gaps 1140 * in the weights. This gives us optimum usage. The walk 1141 * occurs in priority. 1142 */ 1143 for (i = 0; i < NUM_WT; i++) { 1144 weight_t *w; 1145 RB_FOREACH(w, weights, &weights[i]) { 1146 w->opt = nweight[i]; 1147 nweight[i] += 1; 1148 } 1149 } 1150 1151 (void) memset(&chars, 0, sizeof (chars)); 1152 (void) memset(vers, 0, COLLATE_STR_LEN); 1153 (void) strlcpy(vers, COLLATE_VERSION, sizeof (vers)); 1154 1155 /* 1156 * We need to make sure we arrange for the UNDEFINED field 1157 * to show up. Also, set the total weight counts. 1158 */ 1159 for (i = 0; i < NUM_WT; i++) { 1160 if (resolve_pri(pri_undefined[i]) == -1) { 1161 set_pri(pri_undefined[i], -1, RESOLVED); 1162 /* they collate at the end of everything else */ 1163 collinfo.undef_pri[i] = COLLATE_MAX_PRIORITY; 1164 } 1165 collinfo.pri_count[i] = nweight[i]; 1166 } 1167 1168 collinfo.pri_count[NUM_WT] = max_wide(); 1169 collinfo.undef_pri[NUM_WT] = COLLATE_MAX_PRIORITY; 1170 collinfo.directive[NUM_WT] = DIRECTIVE_UNDEFINED; 1171 1172 /* 1173 * Ordinary character priorities 1174 */ 1175 for (i = 0; i <= UCHAR_MAX; i++) { 1176 if ((cc = get_collchar(i, 0)) != NULL) { 1177 for (j = 0; j < NUM_WT; j++) { 1178 chars[i].pri[j] = get_weight(cc->ref[j], j); 1179 } 1180 } else { 1181 for (j = 0; j < NUM_WT; j++) { 1182 chars[i].pri[j] = 1183 get_weight(pri_undefined[j], j); 1184 } 1185 /* 1186 * Per POSIX, for undefined characters, we 1187 * also have to add a last item, which is the 1188 * character code. 1189 */ 1190 chars[i].pri[NUM_WT] = i; 1191 } 1192 } 1193 1194 /* 1195 * Substitution tables 1196 */ 1197 for (i = 0; i < NUM_WT; i++) { 1198 collate_subst_t *st = NULL; 1199 RB_NUMNODES(subst_t, substs, &substs[i], n); 1200 collinfo.subst_count[i] = n; 1201 if ((st = calloc(sizeof (collate_subst_t) * n, 1)) == NULL) { 1202 fprintf(stderr, "out of memory"); 1203 return; 1204 } 1205 n = 0; 1206 RB_FOREACH(sb, substs, &substs[i]) { 1207 if ((st[n].key = resolve_pri(sb->key)) < 0) { 1208 /* by definition these resolve! */ 1209 INTERR; 1210 } 1211 if (st[n].key != (n | COLLATE_SUBST_PRIORITY)) { 1212 INTERR; 1213 } 1214 for (j = 0; sb->ref[j]; j++) { 1215 st[n].pri[j] = get_weight(sb->ref[j], i); 1216 } 1217 n++; 1218 } 1219 if (n != collinfo.subst_count[i]) 1220 INTERR; 1221 subst[i] = st; 1222 } 1223 1224 1225 /* 1226 * Chains, i.e. collating elements 1227 */ 1228 RB_NUMNODES(collelem_t, elem_by_expand, &elem_by_expand, 1229 collinfo.chain_count); 1230 chain = calloc(sizeof (collate_chain_t), collinfo.chain_count); 1231 if (chain == NULL) { 1232 fprintf(stderr, "out of memory"); 1233 return; 1234 } 1235 n = 0; 1236 RB_FOREACH(ce, elem_by_expand, &elem_by_expand) { 1237 n++; 1238 (void) wsncpy(chain[n].str, ce->expand, COLLATE_STR_LEN); 1239 for (i = 0; i < NUM_WT; i++) { 1240 chain[n].pri[i] = get_weight(ce->ref[i], i); 1241 } 1242 } 1243 if (n != collinfo.chain_count) 1244 INTERR; 1245 1246 /* 1247 * Large (> UCHAR_MAX) character priorities 1248 */ 1249 RB_NUMNODES(collchar_t, collchars, &collchars, n); 1250 large = calloc(sizeof (collate_large_t) * n, 1); 1251 if (large == NULL) { 1252 fprintf(stderr, "out of memory"); 1253 return; 1254 } 1255 1256 i = 0; 1257 RB_FOREACH(cc, collchars, &collchars) { 1258 int undef = 0; 1259 /* we already gathered those */ 1260 if (cc->wc <= UCHAR_MAX) 1261 continue; 1262 for (j = 0; j < NUM_WT; j++) { 1263 if ((pri = get_weight(cc->ref[j], j)) < 0) { 1264 undef = 1; 1265 } 1266 if (undef && (pri >= 0)) { 1267 /* if undefined, then all priorities are */ 1268 INTERR; 1269 } else { 1270 large[i].pri.pri[j] = pri; 1271 } 1272 } 1273 if (!undef) { 1274 large[i].val = cc->wc; 1275 collinfo.large_count = i++; 1276 } 1277 } 1278 1279 if ((f = open_category()) == NULL) { 1280 return; 1281 } 1282 1283 /* Time to write the entire data set out */ 1284 1285 if ((wr_category(vers, COLLATE_STR_LEN, f) < 0) || 1286 (wr_category(&collinfo, sizeof (collinfo), f) < 0) || 1287 (wr_category(&chars, sizeof (chars), f) < 0)) { 1288 return; 1289 } 1290 1291 for (i = 0; i < NUM_WT; i++) { 1292 sz = sizeof (collate_subst_t) * collinfo.subst_count[i]; 1293 if (wr_category(subst[i], sz, f) < 0) { 1294 return; 1295 } 1296 } 1297 sz = sizeof (collate_chain_t) * collinfo.chain_count; 1298 if (wr_category(chain, sz, f) < 0) { 1299 return; 1300 } 1301 sz = sizeof (collate_large_t) * collinfo.large_count; 1302 if (wr_category(large, sz, f) < 0) { 1303 return; 1304 } 1305 1306 close_category(f); 1307} 1308