compiler/clib/regex/regcomp.c

   1 /*-
   2  * Copyright (c) 1992, 1993, 1994 Henry Spencer.
   3  * Copyright (c) 1992, 1993, 1994
   4  *      The Regents of the University of California.  All rights reserved.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * Henry Spencer.
   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  * 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  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``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 REGENTS OR CONTRIBUTORS 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
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)regcomp.c   8.5 (Berkeley) 3/20/94
  34  */
  35
  36 #if defined(LIBC_SCCS) && !defined(lint)
  37 static char sccsid[] = "@(#)regcomp.c   8.5 (Berkeley) 3/20/94";
  38 #endif /* LIBC_SCCS and not lint */
  39 #include <sys/cdefs.h>
  40 __FBSDID("$FreeBSD: src/lib/libc/regex/regcomp.c,v 1.36 2007/06/11 03:05:54 delphij Exp $");
  41
  42 #include <sys/types.h>
  43 #include <stdio.h>
  44 #include <string.h>
  45 #include <ctype.h>
  46 #include <limits.h>
  47 #include <stdlib.h>
  48 #include <regex.h>
  49 #include <runetype.h>
  50 #include <wchar.h>
  51 #include <wctype.h>
  52
  53 #include "locale/collate.h"
  54
  55 #include "utils.h"
  56 #include "regex2.h"
  57
  58 #include "cname.h"
  59
  60 /*
  61  * parse structure, passed up and down to avoid global variables and
  62  * other clumsinesses
  63  */
  64 struct parse {
  65         char *next;             /* next character in RE */
  66         char *end;              /* end of string (-> NUL normally) */
  67         int error;              /* has an error been seen? */
  68         sop *strip;             /* malloced strip */
  69         sopno ssize;            /* malloced strip size (allocated) */
  70         sopno slen;             /* malloced strip length (used) */
  71         int ncsalloc;           /* number of csets allocated */
  72         struct re_guts *g;
  73 #       define  NPAREN  10      /* we need to remember () 1-9 for back refs */
  74         sopno pbegin[NPAREN];   /* -> ( ([0] unused) */
  75         sopno pend[NPAREN];     /* -> ) ([0] unused) */
  76 };
  77
  78 /* ========= begin header generated by ./mkh ========= */
  79 #ifdef __cplusplus
  80 extern "C" {
  81 #endif
  82
  83 /* === regcomp.c === */
  84 static void p_ere(struct parse *p, wint_t stop);
  85 static void p_ere_exp(struct parse *p);
  86 static void p_str(struct parse *p);
  87 static void p_bre(struct parse *p, wint_t end1, wint_t end2);
  88 static int p_simp_re(struct parse *p, int starordinary);
  89 static int p_count(struct parse *p);
  90 static void p_bracket(struct parse *p);
  91 static void p_b_term(struct parse *p, cset *cs);
  92 static void p_b_cclass(struct parse *p, cset *cs);
  93 static void p_b_eclass(struct parse *p, cset *cs);
  94 static wint_t p_b_symbol(struct parse *p);
  95 static wint_t p_b_coll_elem(struct parse *p, wint_t endc);
  96 static wint_t othercase(wint_t ch);
  97 static void bothcases(struct parse *p, wint_t ch);
  98 static void ordinary(struct parse *p, wint_t ch);
  99 static void nonnewline(struct parse *p);
 100 static void repeat(struct parse *p, sopno start, int from, int to);
 101 static int seterr(struct parse *p, int e);
 102 static cset *allocset(struct parse *p);
 103 static void freeset(struct parse *p, cset *cs);
 104 static void CHadd(struct parse *p, cset *cs, wint_t ch);
 105 static void CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max);
 106 static void CHaddtype(struct parse *p, cset *cs, wctype_t wct);
 107 static wint_t singleton(cset *cs);
 108 static sopno dupl(struct parse *p, sopno start, sopno finish);
 109 static void doemit(struct parse *p, sop op, size_t opnd);
 110 static void doinsert(struct parse *p, sop op, size_t opnd, sopno pos);
 111 static void dofwd(struct parse *p, sopno pos, sop value);
 112 static void enlarge(struct parse *p, sopno size);
 113 static void stripsnug(struct parse *p, struct re_guts *g);
 114 static void findmust(struct parse *p, struct re_guts *g);
 115 static int altoffset(sop *scan, int offset);
 116 static void computejumps(struct parse *p, struct re_guts *g);
 117 static void computematchjumps(struct parse *p, struct re_guts *g);
 118 static sopno pluscount(struct parse *p, struct re_guts *g);
 119 static wint_t wgetnext(struct parse *p);
 120
 121 #ifdef __cplusplus
 122 }
 123 #endif
 124 /* ========= end header generated by ./mkh ========= */
 125
 126 static char nuls[10];           /* place to point scanner in event of error */
 127
 128 /*
 129  * macros for use with parse structure
 130  * BEWARE:  these know that the parse structure is named `p' !!!
 131  */
 132 #define PEEK()  (*p->next)
 133 #define PEEK2() (*(p->next+1))
 134 #define MORE()  (p->next < p->end)
 135 #define MORE2() (p->next+1 < p->end)
 136 #define SEE(c)  (MORE() && PEEK() == (c))
 137 #define SEETWO(a, b)    (MORE() && MORE2() && PEEK() == (a) && PEEK2() == (b))
 138 #define EAT(c)  ((SEE(c)) ? (NEXT(), 1) : 0)
 139 #define EATTWO(a, b)    ((SEETWO(a, b)) ? (NEXT2(), 1) : 0)
 140 #define NEXT()  (p->next++)
 141 #define NEXT2() (p->next += 2)
 142 #define NEXTn(n)        (p->next += (n))
 143 #define GETNEXT()       (*p->next++)
 144 #define WGETNEXT()      wgetnext(p)
 145 #define SETERROR(e)     seterr(p, (e))
 146 #define REQUIRE(co, e)  ((co) || SETERROR(e))
 147 #define MUSTSEE(c, e)   (REQUIRE(MORE() && PEEK() == (c), e))
 148 #define MUSTEAT(c, e)   (REQUIRE(MORE() && GETNEXT() == (c), e))
 149 #define MUSTNOTSEE(c, e)        (REQUIRE(!MORE() || PEEK() != (c), e))
 150 #define EMIT(op, sopnd) doemit(p, (sop)(op), (size_t)(sopnd))
 151 #define INSERT(op, pos) doinsert(p, (sop)(op), HERE()-(pos)+1, pos)
 152 #define AHEAD(pos)              dofwd(p, pos, HERE()-(pos))
 153 #define ASTERN(sop, pos)        EMIT(sop, HERE()-pos)
 154 #define HERE()          (p->slen)
 155 #define THERE()         (p->slen - 1)
 156 #define THERETHERE()    (p->slen - 2)
 157 #define DROP(n) (p->slen -= (n))
 158
 159 #ifndef NDEBUG
 160 static int never = 0;           /* for use in asserts; shuts lint up */
 161 #else
 162 #define never   0               /* some <assert.h>s have bugs too */
 163 #endif
 164
 165 /* Macro used by computejump()/computematchjump() */
 166 #define MIN(a,b)        ((a)<(b)?(a):(b))
 167
 168 /*
 169  - regcomp - interface for parser and compilation
 170  = extern int regcomp(regex_t *, const char *, int);
 171  = #define      REG_BASIC       0000
 172  = #define      REG_EXTENDED    0001
 173  = #define      REG_ICASE       0002
 174  = #define      REG_NOSUB       0004
 175  = #define      REG_NEWLINE     0010
 176  = #define      REG_NOSPEC      0020
 177  = #define      REG_PEND        0040
 178  = #define      REG_DUMP        0200
 179  */
 180 int                             /* 0 success, otherwise REG_something */
 181 regcomp(regex_t * __restrict preg,
 182         const char * __restrict pattern,
 183         int cflags)
 184 {
 185         struct parse pa;
 186         struct re_guts *g;
 187         struct parse *p = &pa;
 188         int i;
 189         size_t len;
 190 #ifdef REDEBUG
 191 #       define  GOODFLAGS(f)    (f)
 192 #else
 193 #       define  GOODFLAGS(f)    ((f)&~REG_DUMP)
 194 #endif
 195
 196         cflags = GOODFLAGS(cflags);
 197         if ((cflags&REG_EXTENDED) && (cflags&REG_NOSPEC))
 198                 return(REG_INVARG);
 199
 200         if (cflags&REG_PEND) {
 201                 if (preg->re_endp < pattern)
 202                         return(REG_INVARG);
 203                 len = preg->re_endp - pattern;
 204         } else
 205                 len = strlen((char *)pattern);
 206
 207         /* do the mallocs early so failure handling is easy */
 208         g = (struct re_guts *)malloc(sizeof(struct re_guts));
 209         if (g == NULL)
 210                 return(REG_ESPACE);
 211         p->ssize = len/(size_t)2*(size_t)3 + (size_t)1; /* ugh */
 212         p->strip = (sop *)malloc(p->ssize * sizeof(sop));
 213         p->slen = 0;
 214         if (p->strip == NULL) {
 215                 free((char *)g);
 216                 return(REG_ESPACE);
 217         }
 218
 219         /* set things up */
 220         p->g = g;
 221         p->next = (char *)pattern;      /* convenience; we do not modify it */
 222         p->end = p->next + len;
 223         p->error = 0;
 224         p->ncsalloc = 0;
 225         for (i = 0; i < NPAREN; i++) {
 226                 p->pbegin[i] = 0;
 227                 p->pend[i] = 0;
 228         }
 229         g->sets = NULL;
 230         g->ncsets = 0;
 231         g->cflags = cflags;
 232         g->iflags = 0;
 233         g->nbol = 0;
 234         g->neol = 0;
 235         g->must = NULL;
 236         g->moffset = -1;
 237         g->charjump = NULL;
 238         g->matchjump = NULL;
 239         g->mlen = 0;
 240         g->nsub = 0;
 241         g->backrefs = 0;
 242
 243         /* do it */
 244         EMIT(OEND, 0);
 245         g->firststate = THERE();
 246         if (cflags&REG_EXTENDED)
 247                 p_ere(p, OUT);
 248         else if (cflags&REG_NOSPEC)
 249                 p_str(p);
 250         else
 251                 p_bre(p, OUT, OUT);
 252         EMIT(OEND, 0);
 253         g->laststate = THERE();
 254
 255         /* tidy up loose ends and fill things in */
 256         stripsnug(p, g);
 257         findmust(p, g);
 258         /* only use Boyer-Moore algorithm if the pattern is bigger
 259          * than three characters
 260          */
 261         if(g->mlen > 3) {
 262                 computejumps(p, g);
 263                 computematchjumps(p, g);
 264                 if(g->matchjump == NULL && g->charjump != NULL) {
 265                         free(g->charjump);
 266                         g->charjump = NULL;
 267                 }
 268         }
 269         g->nplus = pluscount(p, g);
 270         g->magic = MAGIC2;
 271         preg->re_nsub = g->nsub;
 272         preg->re_g = g;
 273         preg->re_magic = MAGIC1;
 274 #ifndef REDEBUG
 275         /* not debugging, so can't rely on the assert() in regexec() */
 276         if (g->iflags&BAD)
 277                 SETERROR(REG_ASSERT);
 278 #endif
 279
 280         /* win or lose, we're done */
 281         if (p->error != 0)      /* lose */
 282                 regfree(preg);
 283         return(p->error);
 284 }
 285
 286 /*
 287  - p_ere - ERE parser top level, concatenation and alternation
 288  == static void p_ere(struct parse *p, int stop);
 289  */
 290 static void
 291 p_ere(struct parse *p,
 292         int stop)               /* character this ERE should end at */
 293 {
 294         char c;
 295         sopno prevback;
 296         sopno prevfwd;
 297         sopno conc;
 298         int first = 1;          /* is this the first alternative? */
 299
 300         for (;;) {
 301                 /* do a bunch of concatenated expressions */
 302                 conc = HERE();
 303                 while (MORE() && (c = PEEK()) != '|' && c != stop)
 304                         p_ere_exp(p);
 305                 (void)REQUIRE(HERE() != conc, REG_EMPTY);       /* require nonempty */
 306
 307                 if (!EAT('|'))
 308                         break;          /* NOTE BREAK OUT */
 309
 310                 if (first) {
 311                         INSERT(OCH_, conc);     /* offset is wrong */
 312                         prevfwd = conc;
 313                         prevback = conc;
 314                         first = 0;
 315                 }
 316                 ASTERN(OOR1, prevback);
 317                 prevback = THERE();
 318                 AHEAD(prevfwd);                 /* fix previous offset */
 319                 prevfwd = HERE();
 320                 EMIT(OOR2, 0);                  /* offset is very wrong */
 321         }
 322
 323         if (!first) {           /* tail-end fixups */
 324                 AHEAD(prevfwd);
 325                 ASTERN(O_CH, prevback);
 326         }
 327
 328         assert(!MORE() || SEE(stop));
 329 }
 330
 331 /*
 332  - p_ere_exp - parse one subERE, an atom possibly followed by a repetition op
 333  == static void p_ere_exp(struct parse *p);
 334  */
 335 static void
 336 p_ere_exp(struct parse *p)
 337 {
 338         char c;
 339         wint_t wc;
 340         sopno pos;
 341         int count;
 342         int count2;
 343         sopno subno;
 344         int wascaret = 0;
 345
 346         assert(MORE());         /* caller should have ensured this */
 347         c = GETNEXT();
 348
 349         pos = HERE();
 350         switch (c) {
 351         case '(':
 352                 (void)REQUIRE(MORE(), REG_EPAREN);
 353                 p->g->nsub++;
 354                 subno = p->g->nsub;
 355                 if (subno < NPAREN)
 356                         p->pbegin[subno] = HERE();
 357                 EMIT(OLPAREN, subno);
 358                 if (!SEE(')'))
 359                         p_ere(p, ')');
 360                 if (subno < NPAREN) {
 361                         p->pend[subno] = HERE();
 362                         assert(p->pend[subno] != 0);
 363                 }
 364                 EMIT(ORPAREN, subno);
 365                 (void)MUSTEAT(')', REG_EPAREN);
 366                 break;
 367 #ifndef POSIX_MISTAKE
 368         case ')':               /* happens only if no current unmatched ( */
 369                 /*
 370                  * You may ask, why the ifndef?  Because I didn't notice
 371                  * this until slightly too late for 1003.2, and none of the
 372                  * other 1003.2 regular-expression reviewers noticed it at
 373                  * all.  So an unmatched ) is legal POSIX, at least until
 374                  * we can get it fixed.
 375                  */
 376                 SETERROR(REG_EPAREN);
 377                 break;
 378 #endif
 379         case '^':
 380                 EMIT(OBOL, 0);
 381                 p->g->iflags |= USEBOL;
 382                 p->g->nbol++;
 383                 wascaret = 1;
 384                 break;
 385         case '$':
 386                 EMIT(OEOL, 0);
 387                 p->g->iflags |= USEEOL;
 388                 p->g->neol++;
 389                 break;
 390         case '|':
 391                 SETERROR(REG_EMPTY);
 392                 break;
 393         case '*':
 394         case '+':
 395         case '?':
 396                 SETERROR(REG_BADRPT);
 397                 break;
 398         case '.':
 399                 if (p->g->cflags&REG_NEWLINE)
 400                         nonnewline(p);
 401                 else
 402                         EMIT(OANY, 0);
 403                 break;
 404         case '[':
 405                 p_bracket(p);
 406                 break;
 407         case '\\':
 408                 (void)REQUIRE(MORE(), REG_EESCAPE);
 409                 wc = WGETNEXT();
 410                 ordinary(p, wc);
 411                 break;
 412         case '{':               /* okay as ordinary except if digit follows */
 413                 (void)REQUIRE(!MORE() || !isdigit((uch)PEEK()), REG_BADRPT);
 414                 /* FALLTHROUGH */
 415         default:
 416                 p->next--;
 417                 wc = WGETNEXT();
 418                 ordinary(p, wc);
 419                 break;
 420         }
 421
 422         if (!MORE())
 423                 return;
 424         c = PEEK();
 425         /* we call { a repetition if followed by a digit */
 426         if (!( c == '*' || c == '+' || c == '?' ||
 427                                 (c == '{' && MORE2() && isdigit((uch)PEEK2())) ))
 428                 return;         /* no repetition, we're done */
 429         NEXT();
 430
 431         (void)REQUIRE(!wascaret, REG_BADRPT);
 432         switch (c) {
 433         case '*':       /* implemented as +? */
 434                 /* this case does not require the (y|) trick, noKLUDGE */
 435                 INSERT(OPLUS_, pos);
 436                 ASTERN(O_PLUS, pos);
 437                 INSERT(OQUEST_, pos);
 438                 ASTERN(O_QUEST, pos);
 439                 break;
 440         case '+':
 441                 INSERT(OPLUS_, pos);
 442                 ASTERN(O_PLUS, pos);
 443                 break;
 444         case '?':
 445                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
 446                 INSERT(OCH_, pos);              /* offset slightly wrong */
 447                 ASTERN(OOR1, pos);              /* this one's right */
 448                 AHEAD(pos);                     /* fix the OCH_ */
 449                 EMIT(OOR2, 0);                  /* offset very wrong... */
 450                 AHEAD(THERE());                 /* ...so fix it */
 451                 ASTERN(O_CH, THERETHERE());
 452                 break;
 453         case '{':
 454                 count = p_count(p);
 455                 if (EAT(',')) {
 456                         if (isdigit((uch)PEEK())) {
 457                                 count2 = p_count(p);
 458                                 (void)REQUIRE(count <= count2, REG_BADBR);
 459                         } else          /* single number with comma */
 460                                 count2 = INFINITY;
 461                 } else          /* just a single number */
 462                         count2 = count;
 463                 repeat(p, pos, count, count2);
 464                 if (!EAT('}')) {        /* error heuristics */
 465                         while (MORE() && PEEK() != '}')
 466                                 NEXT();
 467                         (void)REQUIRE(MORE(), REG_EBRACE);
 468                         SETERROR(REG_BADBR);
 469                 }
 470                 break;
 471         }
 472
 473         if (!MORE())
 474                 return;
 475         c = PEEK();
 476         if (!( c == '*' || c == '+' || c == '?' ||
 477                                 (c == '{' && MORE2() && isdigit((uch)PEEK2())) ) )
 478                 return;
 479         SETERROR(REG_BADRPT);
 480 }
 481
 482 /*
 483  - p_str - string (no metacharacters) "parser"
 484  == static void p_str(struct parse *p);
 485  */
 486 static void
 487 p_str(struct parse *p)
 488 {
 489         (void)REQUIRE(MORE(), REG_EMPTY);
 490         while (MORE())
 491                 ordinary(p, WGETNEXT());
 492 }
 493
 494 /*
 495  - p_bre - BRE parser top level, anchoring and concatenation
 496  == static void p_bre(struct parse *p, int end1, \
 497  ==     int end2);
 498  * Giving end1 as OUT essentially eliminates the end1/end2 check.
 499  *
 500  * This implementation is a bit of a kludge, in that a trailing $ is first
 501  * taken as an ordinary character and then revised to be an anchor.
 502  * The amount of lookahead needed to avoid this kludge is excessive.
 503  */
 504 static void
 505 p_bre(struct parse *p,
 506         int end1,               /* first terminating character */
 507         int end2)               /* second terminating character */
 508 {
 509         sopno start = HERE();
 510         int first = 1;                  /* first subexpression? */
 511         int wasdollar = 0;
 512
 513         if (EAT('^')) {
 514                 EMIT(OBOL, 0);
 515                 p->g->iflags |= USEBOL;
 516                 p->g->nbol++;
 517         }
 518         while (MORE() && !SEETWO(end1, end2)) {
 519                 wasdollar = p_simp_re(p, first);
 520                 first = 0;
 521         }
 522         if (wasdollar) {        /* oops, that was a trailing anchor */
 523                 DROP(1);
 524                 EMIT(OEOL, 0);
 525                 p->g->iflags |= USEEOL;
 526                 p->g->neol++;
 527         }
 528
 529         (void)REQUIRE(HERE() != start, REG_EMPTY);      /* require nonempty */
 530 }
 531
 532 /*
 533  - p_simp_re - parse a simple RE, an atom possibly followed by a repetition
 534  == static int p_simp_re(struct parse *p, int starordinary);
 535  */
 536 static int                      /* was the simple RE an unbackslashed $? */
 537 p_simp_re(struct parse *p,
 538         int starordinary)       /* is a leading * an ordinary character? */
 539 {
 540         int c;
 541         int count;
 542         int count2;
 543         sopno pos;
 544         int i;
 545         wint_t wc;
 546         sopno subno;
 547 #       define  BACKSL  (1<<CHAR_BIT)
 548
 549         pos = HERE();           /* repetion op, if any, covers from here */
 550
 551         assert(MORE());         /* caller should have ensured this */
 552         c = GETNEXT();
 553         if (c == '\\') {
 554                 (void)REQUIRE(MORE(), REG_EESCAPE);
 555                 c = BACKSL | GETNEXT();
 556         }
 557         switch (c) {
 558         case '.':
 559                 if (p->g->cflags&REG_NEWLINE)
 560                         nonnewline(p);
 561                 else
 562                         EMIT(OANY, 0);
 563                 break;
 564         case '[':
 565                 p_bracket(p);
 566                 break;
 567         case BACKSL|'{':
 568                 SETERROR(REG_BADRPT);
 569                 break;
 570         case BACKSL|'(':
 571                 p->g->nsub++;
 572                 subno = p->g->nsub;
 573                 if (subno < NPAREN)
 574                         p->pbegin[subno] = HERE();
 575                 EMIT(OLPAREN, subno);
 576                 /* the MORE here is an error heuristic */
 577                 if (MORE() && !SEETWO('\\', ')'))
 578                         p_bre(p, '\\', ')');
 579                 if (subno < NPAREN) {
 580                         p->pend[subno] = HERE();
 581                         assert(p->pend[subno] != 0);
 582                 }
 583                 EMIT(ORPAREN, subno);
 584                 (void)REQUIRE(EATTWO('\\', ')'), REG_EPAREN);
 585                 break;
 586         case BACKSL|')':        /* should not get here -- must be user */
 587         case BACKSL|'}':
 588                 SETERROR(REG_EPAREN);
 589                 break;
 590         case BACKSL|'1':
 591         case BACKSL|'2':
 592         case BACKSL|'3':
 593         case BACKSL|'4':
 594         case BACKSL|'5':
 595         case BACKSL|'6':
 596         case BACKSL|'7':
 597         case BACKSL|'8':
 598         case BACKSL|'9':
 599                 i = (c&~BACKSL) - '0';
 600                 assert(i < NPAREN);
 601                 if (p->pend[i] != 0) {
 602                         assert(i <= p->g->nsub);
 603                         EMIT(OBACK_, i);
 604                         assert(p->pbegin[i] != 0);
 605                         assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
 606                         assert(OP(p->strip[p->pend[i]]) == ORPAREN);
 607                         (void) dupl(p, p->pbegin[i]+1, p->pend[i]);
 608                         EMIT(O_BACK, i);
 609                 } else
 610                         SETERROR(REG_ESUBREG);
 611                 p->g->backrefs = 1;
 612                 break;
 613         case '*':
 614                 (void)REQUIRE(starordinary, REG_BADRPT);
 615                 /* FALLTHROUGH */
 616         default:
 617                 p->next--;
 618                 wc = WGETNEXT();
 619                 ordinary(p, wc);
 620                 break;
 621         }
 622
 623         if (EAT('*')) {         /* implemented as +? */
 624                 /* this case does not require the (y|) trick, noKLUDGE */
 625                 INSERT(OPLUS_, pos);
 626                 ASTERN(O_PLUS, pos);
 627                 INSERT(OQUEST_, pos);
 628                 ASTERN(O_QUEST, pos);
 629         } else if (EATTWO('\\', '{')) {
 630                 count = p_count(p);
 631                 if (EAT(',')) {
 632                         if (MORE() && isdigit((uch)PEEK())) {
 633                                 count2 = p_count(p);
 634                                 (void)REQUIRE(count <= count2, REG_BADBR);
 635                         } else          /* single number with comma */
 636                                 count2 = INFINITY;
 637                 } else          /* just a single number */
 638                         count2 = count;
 639                 repeat(p, pos, count, count2);
 640                 if (!EATTWO('\\', '}')) {       /* error heuristics */
 641                         while (MORE() && !SEETWO('\\', '}'))
 642                                 NEXT();
 643                         (void)REQUIRE(MORE(), REG_EBRACE);
 644                         SETERROR(REG_BADBR);
 645                 }
 646         } else if (c == '$')     /* $ (but not \$) ends it */
 647                 return(1);
 648
 649         return(0);
 650 }
 651
 652 /*
 653  - p_count - parse a repetition count
 654  == static int p_count(struct parse *p);
 655  */
 656 static int                      /* the value */
 657 p_count(struct parse *p)
 658 {
 659         int count = 0;
 660         int ndigits = 0;
 661
 662         while (MORE() && isdigit((uch)PEEK()) && count <= DUPMAX) {
 663                 count = count*10 + (GETNEXT() - '0');
 664                 ndigits++;
 665         }
 666
 667         (void)REQUIRE(ndigits > 0 && count <= DUPMAX, REG_BADBR);
 668         return(count);
 669 }
 670
 671 /*
 672  - p_bracket - parse a bracketed character list
 673  == static void p_bracket(struct parse *p);
 674  */
 675 static void
 676 p_bracket(struct parse *p)
 677 {
 678         cset *cs;
 679         wint_t ch;
 680
 681         /* Dept of Truly Sickening Special-Case Kludges */
 682         if (p->next + 5 < p->end && strncmp(p->next, "[:<:]]", 6) == 0) {
 683                 EMIT(OBOW, 0);
 684                 NEXTn(6);
 685                 return;
 686         }
 687         if (p->next + 5 < p->end && strncmp(p->next, "[:>:]]", 6) == 0) {
 688                 EMIT(OEOW, 0);
 689                 NEXTn(6);
 690                 return;
 691         }
 692
 693         if ((cs = allocset(p)) == NULL)
 694                 return;
 695
 696         if (p->g->cflags&REG_ICASE)
 697                 cs->icase = 1;
 698         if (EAT('^'))
 699                 cs->invert = 1;
 700         if (EAT(']'))
 701                 CHadd(p, cs, ']');
 702         else if (EAT('-'))
 703                 CHadd(p, cs, '-');
 704         while (MORE() && PEEK() != ']' && !SEETWO('-', ']'))
 705                 p_b_term(p, cs);
 706         if (EAT('-'))
 707                 CHadd(p, cs, '-');
 708         (void)MUSTEAT(']', REG_EBRACK);
 709
 710         if (p->error != 0)      /* don't mess things up further */
 711                 return;
 712
 713         if (cs->invert && p->g->cflags&REG_NEWLINE)
 714                 cs->bmp['\n' >> 3] |= 1 << ('\n' & 7);
 715
 716         if ((ch = singleton(cs)) != OUT) {      /* optimize singleton sets */
 717                 ordinary(p, ch);
 718                 freeset(p, cs);
 719         } else
 720                 EMIT(OANYOF, (int)(cs - p->g->sets));
 721 }
 722
 723 /*
 724  - p_b_term - parse one term of a bracketed character list
 725  == static void p_b_term(struct parse *p, cset *cs);
 726  */
 727 static void
 728 p_b_term(struct parse *p, cset *cs)
 729 {
 730         char c;
 731         wint_t start, finish;
 732         wint_t i;
 733
 734         /* classify what we've got */
 735         switch ((MORE()) ? PEEK() : '\0') {
 736         case '[':
 737                 c = (MORE2()) ? PEEK2() : '\0';
 738                 break;
 739         case '-':
 740                 SETERROR(REG_ERANGE);
 741                 return;                 /* NOTE RETURN */
 742                 break;
 743         default:
 744                 c = '\0';
 745                 break;
 746         }
 747
 748         switch (c) {
 749         case ':':               /* character class */
 750                 NEXT2();
 751                 (void)REQUIRE(MORE(), REG_EBRACK);
 752                 c = PEEK();
 753                 (void)REQUIRE(c != '-' && c != ']', REG_ECTYPE);
 754                 p_b_cclass(p, cs);
 755                 (void)REQUIRE(MORE(), REG_EBRACK);
 756                 (void)REQUIRE(EATTWO(':', ']'), REG_ECTYPE);
 757                 break;
 758         case '=':               /* equivalence class */
 759                 NEXT2();
 760                 (void)REQUIRE(MORE(), REG_EBRACK);
 761                 c = PEEK();
 762                 (void)REQUIRE(c != '-' && c != ']', REG_ECOLLATE);
 763                 p_b_eclass(p, cs);
 764                 (void)REQUIRE(MORE(), REG_EBRACK);
 765                 (void)REQUIRE(EATTWO('=', ']'), REG_ECOLLATE);
 766                 break;
 767         default:                /* symbol, ordinary character, or range */
 768                 start = p_b_symbol(p);
 769                 if (SEE('-') && MORE2() && PEEK2() != ']') {
 770                         /* range */
 771                         NEXT();
 772                         if (EAT('-'))
 773                                 finish = '-';
 774                         else
 775                                 finish = p_b_symbol(p);
 776                 } else
 777                         finish = start;
 778                 if (start == finish)
 779                         CHadd(p, cs, start);
 780                 else {
 781                         if (__collate_load_error) {
 782                                 (void)REQUIRE((uch)start <= (uch)finish, REG_ERANGE);
 783                                 CHaddrange(p, cs, start, finish);
 784                         } else {
 785                                 (void)REQUIRE(__collate_range_cmp(start, finish) <= 0, REG_ERANGE);
 786                                 for (i = 0; i <= UCHAR_MAX; i++) {
 787                                         if (   __collate_range_cmp(start, i) <= 0
 788                                             && __collate_range_cmp(i, finish) <= 0
 789                                            )
 790                                                 CHadd(p, cs, i);
 791                                 }
 792                         }
 793                 }
 794                 break;
 795         }
 796 }
 797
 798 /*
 799  - p_b_cclass - parse a character-class name and deal with it
 800  == static void p_b_cclass(struct parse *p, cset *cs);
 801  */
 802 static void
 803 p_b_cclass(struct parse *p, cset *cs)
 804 {
 805         char *sp = p->next;
 806         size_t len;
 807         wctype_t wct;
 808         char clname[16];
 809
 810         while (MORE() && isalpha((uch)PEEK()))
 811                 NEXT();
 812         len = p->next - sp;
 813         if (len >= sizeof(clname) - 1) {
 814                 SETERROR(REG_ECTYPE);
 815                 return;
 816         }
 817         memcpy(clname, sp, len);
 818         clname[len] = '\0';
 819         if ((wct = wctype(clname)) == 0) {
 820                 SETERROR(REG_ECTYPE);
 821                 return;
 822         }
 823         CHaddtype(p, cs, wct);
 824 }
 825
 826 /*
 827  - p_b_eclass - parse an equivalence-class name and deal with it
 828  == static void p_b_eclass(struct parse *p, cset *cs);
 829  *
 830  * This implementation is incomplete. xxx
 831  */
 832 static void
 833 p_b_eclass(struct parse *p, cset *cs)
 834 {
 835         wint_t c;
 836
 837         c = p_b_coll_elem(p, '=');
 838         CHadd(p, cs, c);
 839 }
 840
 841 /*
 842  - p_b_symbol - parse a character or [..]ed multicharacter collating symbol
 843  == static char p_b_symbol(struct parse *p);
 844  */
 845 static wint_t                   /* value of symbol */
 846 p_b_symbol(struct parse *p)
 847 {
 848         wint_t value;
 849
 850         (void)REQUIRE(MORE(), REG_EBRACK);
 851         if (!EATTWO('[', '.'))
 852                 return(WGETNEXT());
 853
 854         /* collating symbol */
 855         value = p_b_coll_elem(p, '.');
 856         (void)REQUIRE(EATTWO('.', ']'), REG_ECOLLATE);
 857         return(value);
 858 }
 859
 860 /*
 861  - p_b_coll_elem - parse a collating-element name and look it up
 862  == static char p_b_coll_elem(struct parse *p, int endc);
 863  */
 864 static wint_t                   /* value of collating element */
 865 p_b_coll_elem(struct parse *p,
 866         wint_t endc)            /* name ended by endc,']' */
 867 {
 868         char *sp = p->next;
 869         struct cname *cp;
 870         int len;
 871         mbstate_t mbs;
 872         wchar_t wc;
 873         size_t clen;
 874
 875         while (MORE() && !SEETWO(endc, ']'))
 876                 NEXT();
 877         if (!MORE()) {
 878                 SETERROR(REG_EBRACK);
 879                 return(0);
 880         }
 881         len = p->next - sp;
 882         for (cp = cnames; cp->name != NULL; cp++)
 883                 if (strncmp(cp->name, sp, len) == 0 && cp->name[len] == '\0')
 884                         return(cp->code);       /* known name */
 885         memset(&mbs, 0, sizeof(mbs));
 886         if ((clen = mbrtowc(&wc, sp, len, &mbs)) == len)
 887                 return (wc);                    /* single character */
 888         else if (clen == (size_t)-1 || clen == (size_t)-2)
 889                 SETERROR(REG_ILLSEQ);
 890         else
 891                 SETERROR(REG_ECOLLATE);         /* neither */
 892         return(0);
 893 }
 894
 895 /*
 896  - othercase - return the case counterpart of an alphabetic
 897  == static char othercase(int ch);
 898  */
 899 static wint_t                   /* if no counterpart, return ch */
 900 othercase(wint_t ch)
 901 {
 902         assert(iswalpha(ch));
 903         if (iswupper(ch))
 904                 return(towlower(ch));
 905         else if (iswlower(ch))
 906                 return(towupper(ch));
 907         else                    /* peculiar, but could happen */
 908                 return(ch);
 909 }
 910
 911 /*
 912  - bothcases - emit a dualcase version of a two-case character
 913  == static void bothcases(struct parse *p, int ch);
 914  *
 915  * Boy, is this implementation ever a kludge...
 916  */
 917 static void
 918 bothcases(struct parse *p, wint_t ch)
 919 {
 920         char *oldnext = p->next;
 921         char *oldend = p->end;
 922         char bracket[3 + MB_LEN_MAX];
 923         size_t n;
 924         mbstate_t mbs;
 925
 926         assert(othercase(ch) != ch);    /* p_bracket() would recurse */
 927         p->next = bracket;
 928         memset(&mbs, 0, sizeof(mbs));
 929         n = wcrtomb(bracket, ch, &mbs);
 930         assert(n != (size_t)-1);
 931         bracket[n] = ']';
 932         bracket[n + 1] = '\0';
 933         p->end = bracket+n+1;
 934         p_bracket(p);
 935         assert(p->next == p->end);
 936         p->next = oldnext;
 937         p->end = oldend;
 938 }
 939
 940 /*
 941  - ordinary - emit an ordinary character
 942  == static void ordinary(struct parse *p, int ch);
 943  */
 944 static void
 945 ordinary(struct parse *p, wint_t ch)
 946 {
 947         cset *cs;
 948
 949         if ((p->g->cflags&REG_ICASE) && iswalpha(ch) && othercase(ch) != ch)
 950                 bothcases(p, ch);
 951         else if ((ch & OPDMASK) == ch)
 952                 EMIT(OCHAR, ch);
 953         else {
 954                 /*
 955                  * Kludge: character is too big to fit into an OCHAR operand.
 956                  * Emit a singleton set.
 957                  */
 958                 if ((cs = allocset(p)) == NULL)
 959                         return;
 960                 CHadd(p, cs, ch);
 961                 EMIT(OANYOF, (int)(cs - p->g->sets));
 962         }
 963 }
 964
 965 /*
 966  - nonnewline - emit REG_NEWLINE version of OANY
 967  == static void nonnewline(struct parse *p);
 968  *
 969  * Boy, is this implementation ever a kludge...
 970  */
 971 static void
 972 nonnewline(struct parse *p)
 973 {
 974         char *oldnext = p->next;
 975         char *oldend = p->end;
 976         char bracket[4];
 977
 978         p->next = bracket;
 979         p->end = bracket+3;
 980         bracket[0] = '^';
 981         bracket[1] = '\n';
 982         bracket[2] = ']';
 983         bracket[3] = '\0';
 984         p_bracket(p);
 985         assert(p->next == bracket+3);
 986         p->next = oldnext;
 987         p->end = oldend;
 988 }
 989
 990 /*
 991  - repeat - generate code for a bounded repetition, recursively if needed
 992  == static void repeat(struct parse *p, sopno start, int from, int to);
 993  */
 994 static void
 995 repeat(struct parse *p,
 996         sopno start,            /* operand from here to end of strip */
 997         int from,               /* repeated from this number */
 998         int to)                 /* to this number of times (maybe INFINITY) */
 999 {
1000         sopno finish = HERE();
1001 #       define  N       2
1002 #       define  INF     3
1003 #       define  REP(f, t)       ((f)*8 + (t))
1004 #       define  MAP(n)  (((n) <= 1) ? (n) : ((n) == INFINITY) ? INF : N)
1005         sopno copy;
1006
1007         if (p->error != 0)      /* head off possible runaway recursion */
1008                 return;
1009
1010         assert(from <= to);
1011
1012         switch (REP(MAP(from), MAP(to))) {
1013         case REP(0, 0):                 /* must be user doing this */
1014                 DROP(finish-start);     /* drop the operand */
1015                 break;
1016         case REP(0, 1):                 /* as x{1,1}? */
1017         case REP(0, N):                 /* as x{1,n}? */
1018         case REP(0, INF):               /* as x{1,}? */
1019                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1020                 INSERT(OCH_, start);            /* offset is wrong... */
1021                 repeat(p, start+1, 1, to);
1022                 ASTERN(OOR1, start);
1023                 AHEAD(start);                   /* ... fix it */
1024                 EMIT(OOR2, 0);
1025                 AHEAD(THERE());
1026                 ASTERN(O_CH, THERETHERE());
1027                 break;
1028         case REP(1, 1):                 /* trivial case */
1029                 /* done */
1030                 break;
1031         case REP(1, N):                 /* as x?x{1,n-1} */
1032                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1033                 INSERT(OCH_, start);
1034                 ASTERN(OOR1, start);
1035                 AHEAD(start);
1036                 EMIT(OOR2, 0);                  /* offset very wrong... */
1037                 AHEAD(THERE());                 /* ...so fix it */
1038                 ASTERN(O_CH, THERETHERE());
1039                 copy = dupl(p, start+1, finish+1);
1040                 assert(copy == finish+4);
1041                 repeat(p, copy, 1, to-1);
1042                 break;
1043         case REP(1, INF):               /* as x+ */
1044                 INSERT(OPLUS_, start);
1045                 ASTERN(O_PLUS, start);
1046                 break;
1047         case REP(N, N):                 /* as xx{m-1,n-1} */
1048                 copy = dupl(p, start, finish);
1049                 repeat(p, copy, from-1, to-1);
1050                 break;
1051         case REP(N, INF):               /* as xx{n-1,INF} */
1052                 copy = dupl(p, start, finish);
1053                 repeat(p, copy, from-1, to);
1054                 break;
1055         default:                        /* "can't happen" */
1056                 SETERROR(REG_ASSERT);   /* just in case */
1057                 break;
1058         }
1059 }
1060
1061 /*
1062  - wgetnext - helper function for WGETNEXT() macro. Gets the next wide
1063  - character from the parse struct, signals a REG_ILLSEQ error if the
1064  - character can't be converted. Returns the number of bytes consumed.
1065  */
1066 static wint_t
1067 wgetnext(struct parse *p)
1068 {
1069         mbstate_t mbs;
1070         wchar_t wc;
1071         size_t n;
1072
1073         memset(&mbs, 0, sizeof(mbs));
1074         n = mbrtowc(&wc, p->next, p->end - p->next, &mbs);
1075         if (n == (size_t)-1 || n == (size_t)-2) {
1076                 SETERROR(REG_ILLSEQ);
1077                 return (0);
1078         }
1079         if (n == 0)
1080                 n = 1;
1081         p->next += n;
1082         return (wc);
1083 }
1084
1085 /*
1086  - seterr - set an error condition
1087  == static int seterr(struct parse *p, int e);
1088  */
1089 static int                      /* useless but makes type checking happy */
1090 seterr(struct parse *p, int e)
1091 {
1092         if (p->error == 0)      /* keep earliest error condition */
1093                 p->error = e;
1094         p->next = nuls;         /* try to bring things to a halt */
1095         p->end = nuls;
1096         return(0);              /* make the return value well-defined */
1097 }
1098
1099 /*
1100  - allocset - allocate a set of characters for []
1101  == static cset *allocset(struct parse *p);
1102  */
1103 static cset *
1104 allocset(struct parse *p)
1105 {
1106         cset *cs, *ncs;
1107
1108         ncs = realloc(p->g->sets, (p->g->ncsets + 1) * sizeof(*ncs));
1109         if (ncs == NULL) {
1110                 SETERROR(REG_ESPACE);
1111                 return (NULL);
1112         }
1113         p->g->sets = ncs;
1114         cs = &p->g->sets[p->g->ncsets++];
1115         memset(cs, 0, sizeof(*cs));
1116
1117         return(cs);
1118 }
1119
1120 /*
1121  - freeset - free a now-unused set
1122  == static void freeset(struct parse *p, cset *cs);
1123  */
1124 static void
1125 freeset(struct parse *p, cset *cs)
1126 {
1127         cset *top = &p->g->sets[p->g->ncsets];
1128
1129         free(cs->wides);
1130         free(cs->ranges);
1131         free(cs->types);
1132         memset(cs, 0, sizeof(*cs));
1133         if (cs == top-1)        /* recover only the easy case */
1134                 p->g->ncsets--;
1135 }
1136
1137 /*
1138  - singleton - Determine whether a set contains only one character,
1139  - returning it if so, otherwise returning OUT.
1140  */
1141 static wint_t
1142 singleton(cset *cs)
1143 {
1144         wint_t i, s, n;
1145
1146         for (i = n = 0; i < NC; i++)
1147                 if (CHIN(cs, i)) {
1148                         n++;
1149                         s = i;
1150                 }
1151         if (n == 1)
1152                 return (s);
1153         if (cs->nwides == 1 && cs->nranges == 0 && cs->ntypes == 0 &&
1154             cs->icase == 0)
1155                 return (cs->wides[0]);
1156         /* Don't bother handling the other cases. */
1157         return (OUT);
1158 }
1159
1160 /*
1161  - CHadd - add character to character set.
1162  */
1163 static void
1164 CHadd(struct parse *p, cset *cs, wint_t ch)
1165 {
1166         wint_t nch, *newwides;
1167         assert(ch >= 0);
1168         if (ch < NC)
1169                 cs->bmp[ch >> 3] |= 1 << (ch & 7);
1170         else {
1171                 newwides = realloc(cs->wides, (cs->nwides + 1) *
1172                     sizeof(*cs->wides));
1173                 if (newwides == NULL) {
1174                         SETERROR(REG_ESPACE);
1175                         return;
1176                 }
1177                 cs->wides = newwides;
1178                 cs->wides[cs->nwides++] = ch;
1179         }
1180         if (cs->icase) {
1181                 if ((nch = towlower(ch)) < NC)
1182                         cs->bmp[nch >> 3] |= 1 << (nch & 7);
1183                 if ((nch = towupper(ch)) < NC)
1184                         cs->bmp[nch >> 3] |= 1 << (nch & 7);
1185         }
1186 }
1187
1188 /*
1189  - CHaddrange - add all characters in the range [min,max] to a character set.
1190  */
1191 static void
1192 CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max)
1193 {
1194         crange *newranges;
1195
1196         for (; min < NC && min <= max; min++)
1197                 CHadd(p, cs, min);
1198         if (min >= max)
1199                 return;
1200         newranges = realloc(cs->ranges, (cs->nranges + 1) *
1201             sizeof(*cs->ranges));
1202         if (newranges == NULL) {
1203                 SETERROR(REG_ESPACE);
1204                 return;
1205         }
1206         cs->ranges = newranges;
1207         cs->ranges[cs->nranges].min = min;
1208         cs->ranges[cs->nranges].min = max;
1209         cs->nranges++;
1210 }
1211
1212 /*
1213  - CHaddtype - add all characters of a certain type to a character set.
1214  */
1215 static void
1216 CHaddtype(struct parse *p, cset *cs, wctype_t wct)
1217 {
1218         wint_t i;
1219         wctype_t *newtypes;
1220
1221         for (i = 0; i < NC; i++)
1222                 if (iswctype(i, wct))
1223                         CHadd(p, cs, i);
1224         newtypes = realloc(cs->types, (cs->ntypes + 1) *
1225             sizeof(*cs->types));
1226         if (newtypes == NULL) {
1227                 SETERROR(REG_ESPACE);
1228                 return;
1229         }
1230         cs->types = newtypes;
1231         cs->types[cs->ntypes++] = wct;
1232 }
1233
1234 /*
1235  - dupl - emit a duplicate of a bunch of sops
1236  == static sopno dupl(struct parse *p, sopno start, sopno finish);
1237  */
1238 static sopno                    /* start of duplicate */
1239 dupl(struct parse *p,
1240         sopno start,            /* from here */
1241         sopno finish)           /* to this less one */
1242 {
1243         sopno ret = HERE();
1244         sopno len = finish - start;
1245
1246         assert(finish >= start);
1247         if (len == 0)
1248                 return(ret);
1249         enlarge(p, p->ssize + len);     /* this many unexpected additions */
1250         assert(p->ssize >= p->slen + len);
1251         (void) memcpy((char *)(p->strip + p->slen),
1252                 (char *)(p->strip + start), (size_t)len*sizeof(sop));
1253         p->slen += len;
1254         return(ret);
1255 }
1256
1257 /*
1258  - doemit - emit a strip operator
1259  == static void doemit(struct parse *p, sop op, size_t opnd);
1260  *
1261  * It might seem better to implement this as a macro with a function as
1262  * hard-case backup, but it's just too big and messy unless there are
1263  * some changes to the data structures.  Maybe later.
1264  */
1265 static void
1266 doemit(struct parse *p, sop op, size_t opnd)
1267 {
1268         /* avoid making error situations worse */
1269         if (p->error != 0)
1270                 return;
1271
1272         /* deal with oversize operands ("can't happen", more or less) */
1273         assert(opnd < 1<<OPSHIFT);
1274
1275         /* deal with undersized strip */
1276         if (p->slen >= p->ssize)
1277                 enlarge(p, (p->ssize+1) / 2 * 3);       /* +50% */
1278         assert(p->slen < p->ssize);
1279
1280         /* finally, it's all reduced to the easy case */
1281         p->strip[p->slen++] = SOP(op, opnd);
1282 }
1283
1284 /*
1285  - doinsert - insert a sop into the strip
1286  == static void doinsert(struct parse *p, sop op, size_t opnd, sopno pos);
1287  */
1288 static void
1289 doinsert(struct parse *p, sop op, size_t opnd, sopno pos)
1290 {
1291         sopno sn;
1292         sop s;
1293         int i;
1294
1295         /* avoid making error situations worse */
1296         if (p->error != 0)
1297                 return;
1298
1299         sn = HERE();
1300         EMIT(op, opnd);         /* do checks, ensure space */
1301         assert(HERE() == sn+1);
1302         s = p->strip[sn];
1303
1304         /* adjust paren pointers */
1305         assert(pos > 0);
1306         for (i = 1; i < NPAREN; i++) {
1307                 if (p->pbegin[i] >= pos) {
1308                         p->pbegin[i]++;
1309                 }
1310                 if (p->pend[i] >= pos) {
1311                         p->pend[i]++;
1312                 }
1313         }
1314
1315         memmove((char *)&p->strip[pos+1], (char *)&p->strip[pos],
1316                                                 (HERE()-pos-1)*sizeof(sop));
1317         p->strip[pos] = s;
1318 }
1319
1320 /*
1321  - dofwd - complete a forward reference
1322  == static void dofwd(struct parse *p, sopno pos, sop value);
1323  */
1324 static void
1325 dofwd(struct parse *p, sopno pos, sop value)
1326 {
1327         /* avoid making error situations worse */
1328         if (p->error != 0)
1329                 return;
1330
1331         assert(value < 1<<OPSHIFT);
1332         p->strip[pos] = OP(p->strip[pos]) | value;
1333 }
1334
1335 /*
1336  - enlarge - enlarge the strip
1337  == static void enlarge(struct parse *p, sopno size);
1338  */
1339 static void
1340 enlarge(struct parse *p, sopno size)
1341 {
1342         sop *sp;
1343
1344         if (p->ssize >= size)
1345                 return;
1346
1347         sp = (sop *)realloc(p->strip, size*sizeof(sop));
1348         if (sp == NULL) {
1349                 SETERROR(REG_ESPACE);
1350                 return;
1351         }
1352         p->strip = sp;
1353         p->ssize = size;
1354 }
1355
1356 /*
1357  - stripsnug - compact the strip
1358  == static void stripsnug(struct parse *p, struct re_guts *g);
1359  */
1360 static void
1361 stripsnug(struct parse *p, struct re_guts *g)
1362 {
1363         g->nstates = p->slen;
1364         g->strip = (sop *)realloc((char *)p->strip, p->slen * sizeof(sop));
1365         if (g->strip == NULL) {
1366                 SETERROR(REG_ESPACE);
1367                 g->strip = p->strip;
1368         }
1369 }
1370
1371 /*
1372  - findmust - fill in must and mlen with longest mandatory literal string
1373  == static void findmust(struct parse *p, struct re_guts *g);
1374  *
1375  * This algorithm could do fancy things like analyzing the operands of |
1376  * for common subsequences.  Someday.  This code is simple and finds most
1377  * of the interesting cases.
1378  *
1379  * Note that must and mlen got initialized during setup.
1380  */
1381 static void
1382 findmust(struct parse *p, struct re_guts *g)
1383 {
1384         sop *scan;
1385         sop *start;
1386         sop *newstart;
1387         sopno newlen;
1388         sop s;
1389         char *cp;
1390         int offset;
1391         char buf[MB_LEN_MAX];
1392         size_t clen;
1393         mbstate_t mbs;
1394
1395         /* avoid making error situations worse */
1396         if (p->error != 0)
1397                 return;
1398
1399         /*
1400          * It's not generally safe to do a ``char'' substring search on
1401          * multibyte character strings, but it's safe for at least
1402          * UTF-8 (see RFC 3629).
1403          */
1404         if (MB_CUR_MAX > 1 &&
1405             strcmp(_CurrentRuneLocale->__encoding, "UTF-8") != 0)
1406                 return;
1407
1408         /* find the longest OCHAR sequence in strip */
1409         newlen = 0;
1410         offset = 0;
1411         g->moffset = 0;
1412         scan = g->strip + 1;
1413         do {
1414                 s = *scan++;
1415                 switch (OP(s)) {
1416                 case OCHAR:             /* sequence member */
1417                         if (newlen == 0) {              /* new sequence */
1418                                 memset(&mbs, 0, sizeof(mbs));
1419                                 newstart = scan - 1;
1420                         }
1421                         clen = wcrtomb(buf, OPND(s), &mbs);
1422                         if (clen == (size_t)-1)
1423                                 goto toohard;
1424                         newlen += clen;
1425                         break;
1426                 case OPLUS_:            /* things that don't break one */
1427                 case OLPAREN:
1428                 case ORPAREN:
1429                         break;
1430                 case OQUEST_:           /* things that must be skipped */
1431                 case OCH_:
1432                         offset = altoffset(scan, offset);
1433                         scan--;
1434                         do {
1435                                 scan += OPND(s);
1436                                 s = *scan;
1437                                 /* assert() interferes w debug printouts */
1438                                 if (OP(s) != O_QUEST && OP(s) != O_CH &&
1439                                                         OP(s) != OOR2) {
1440                                         g->iflags |= BAD;
1441                                         return;
1442                                 }
1443                         } while (OP(s) != O_QUEST && OP(s) != O_CH);
1444                         /* FALLTHROUGH */
1445                 case OBOW:              /* things that break a sequence */
1446                 case OEOW:
1447                 case OBOL:
1448                 case OEOL:
1449                 case O_QUEST:
1450                 case O_CH:
1451                 case OEND:
1452                         if (newlen > g->mlen) {         /* ends one */
1453                                 start = newstart;
1454                                 g->mlen = newlen;
1455                                 if (offset > -1) {
1456                                         g->moffset += offset;
1457                                         offset = newlen;
1458                                 } else
1459                                         g->moffset = offset;
1460                         } else {
1461                                 if (offset > -1)
1462                                         offset += newlen;
1463                         }
1464                         newlen = 0;
1465                         break;
1466                 case OANY:
1467                         if (newlen > g->mlen) {         /* ends one */
1468                                 start = newstart;
1469                                 g->mlen = newlen;
1470                                 if (offset > -1) {
1471                                         g->moffset += offset;
1472                                         offset = newlen;
1473                                 } else
1474                                         g->moffset = offset;
1475                         } else {
1476                                 if (offset > -1)
1477                                         offset += newlen;
1478                         }
1479                         if (offset > -1)
1480                                 offset++;
1481                         newlen = 0;
1482                         break;
1483                 case OANYOF:            /* may or may not invalidate offset */
1484                         /* First, everything as OANY */
1485                         if (newlen > g->mlen) {         /* ends one */
1486                                 start = newstart;
1487                                 g->mlen = newlen;
1488                                 if (offset > -1) {
1489                                         g->moffset += offset;
1490                                         offset = newlen;
1491                                 } else
1492                                         g->moffset = offset;
1493                         } else {
1494                                 if (offset > -1)
1495                                         offset += newlen;
1496                         }
1497                         if (offset > -1)
1498                                 offset++;
1499                         newlen = 0;
1500                         break;
1501                 toohard:
1502                 default:
1503                         /* Anything here makes it impossible or too hard
1504                          * to calculate the offset -- so we give up;
1505                          * save the last known good offset, in case the
1506                          * must sequence doesn't occur later.
1507                          */
1508                         if (newlen > g->mlen) {         /* ends one */
1509                                 start = newstart;
1510                                 g->mlen = newlen;
1511                                 if (offset > -1)
1512                                         g->moffset += offset;
1513                                 else
1514                                         g->moffset = offset;
1515                         }
1516                         offset = -1;
1517                         newlen = 0;
1518                         break;
1519                 }
1520         } while (OP(s) != OEND);
1521
1522         if (g->mlen == 0) {             /* there isn't one */
1523                 g->moffset = -1;
1524                 return;
1525         }
1526
1527         /* turn it into a character string */
1528         g->must = malloc((size_t)g->mlen + 1);
1529         if (g->must == NULL) {          /* argh; just forget it */
1530                 g->mlen = 0;
1531                 g->moffset = -1;
1532                 return;
1533         }
1534         cp = g->must;
1535         scan = start;
1536         memset(&mbs, 0, sizeof(mbs));
1537         while (cp < g->must + g->mlen) {
1538                 while (OP(s = *scan++) != OCHAR)
1539                         continue;
1540                 clen = wcrtomb(cp, OPND(s), &mbs);
1541                 assert(clen != (size_t)-1);
1542                 cp += clen;
1543         }
1544         assert(cp == g->must + g->mlen);
1545         *cp++ = '\0';           /* just on general principles */
1546 }
1547
1548 /*
1549  - altoffset - choose biggest offset among multiple choices
1550  == static int altoffset(sop *scan, int offset);
1551  *
1552  * Compute, recursively if necessary, the largest offset among multiple
1553  * re paths.
1554  */
1555 static int
1556 altoffset(sop *scan, int offset)
1557 {
1558         int largest;
1559         int try;
1560         sop s;
1561
1562         /* If we gave up already on offsets, return */
1563         if (offset == -1)
1564                 return -1;
1565
1566         largest = 0;
1567         try = 0;
1568         s = *scan++;
1569         while (OP(s) != O_QUEST && OP(s) != O_CH) {
1570                 switch (OP(s)) {
1571                 case OOR1:
1572                         if (try > largest)
1573                                 largest = try;
1574                         try = 0;
1575                         break;
1576                 case OQUEST_:
1577                 case OCH_:
1578                         try = altoffset(scan, try);
1579                         if (try == -1)
1580                                 return -1;
1581                         scan--;
1582                         do {
1583                                 scan += OPND(s);
1584                                 s = *scan;
1585                                 if (OP(s) != O_QUEST && OP(s) != O_CH &&
1586                                                         OP(s) != OOR2)
1587                                         return -1;
1588                         } while (OP(s) != O_QUEST && OP(s) != O_CH);
1589                         /* We must skip to the next position, or we'll
1590                          * leave altoffset() too early.
1591                          */
1592                         scan++;
1593                         break;
1594                 case OANYOF:
1595                 case OCHAR:
1596                 case OANY:
1597                         try++;
1598                 case OBOW:
1599                 case OEOW:
1600                 case OLPAREN:
1601                 case ORPAREN:
1602                 case OOR2:
1603                         break;
1604                 default:
1605                         try = -1;
1606                         break;
1607                 }
1608                 if (try == -1)
1609                         return -1;
1610                 s = *scan++;
1611         }
1612
1613         if (try > largest)
1614                 largest = try;
1615
1616         return largest+offset;
1617 }
1618
1619 /*
1620  - computejumps - compute char jumps for BM scan
1621  == static void computejumps(struct parse *p, struct re_guts *g);
1622  *
1623  * This algorithm assumes g->must exists and is has size greater than
1624  * zero. It's based on the algorithm found on Computer Algorithms by
1625  * Sara Baase.
1626  *
1627  * A char jump is the number of characters one needs to jump based on
1628  * the value of the character from the text that was mismatched.
1629  */
1630 static void
1631 computejumps(struct parse *p, struct re_guts *g)
1632 {
1633         int ch;
1634         int mindex;
1635
1636         /* Avoid making errors worse */
1637         if (p->error != 0)
1638                 return;
1639
1640         g->charjump = (int*) malloc((NC + 1) * sizeof(int));
1641         if (g->charjump == NULL)        /* Not a fatal error */
1642                 return;
1643         /* Adjust for signed chars, if necessary */
1644         g->charjump = &g->charjump[-(CHAR_MIN)];
1645
1646         /* If the character does not exist in the pattern, the jump
1647          * is equal to the number of characters in the pattern.
1648          */
1649         for (ch = CHAR_MIN; ch < (CHAR_MAX + 1); ch++)
1650                 g->charjump[ch] = g->mlen;
1651
1652         /* If the character does exist, compute the jump that would
1653          * take us to the last character in the pattern equal to it
1654          * (notice that we match right to left, so that last character
1655          * is the first one that would be matched).
1656          */
1657         for (mindex = 0; mindex < g->mlen; mindex++)
1658                 g->charjump[(int)g->must[mindex]] = g->mlen - mindex - 1;
1659 }
1660
1661 /*
1662  - computematchjumps - compute match jumps for BM scan
1663  == static void computematchjumps(struct parse *p, struct re_guts *g);
1664  *
1665  * This algorithm assumes g->must exists and is has size greater than
1666  * zero. It's based on the algorithm found on Computer Algorithms by
1667  * Sara Baase.
1668  *
1669  * A match jump is the number of characters one needs to advance based
1670  * on the already-matched suffix.
1671  * Notice that all values here are minus (g->mlen-1), because of the way
1672  * the search algorithm works.
1673  */
1674 static void
1675 computematchjumps(struct parse *p, struct re_guts *g)
1676 {
1677         int mindex;             /* General "must" iterator */
1678         int suffix;             /* Keeps track of matching suffix */
1679         int ssuffix;            /* Keeps track of suffixes' suffix */
1680         int* pmatches;          /* pmatches[k] points to the next i
1681                                  * such that i+1...mlen is a substring
1682                                  * of k+1...k+mlen-i-1
1683                                  */
1684
1685         /* Avoid making errors worse */
1686         if (p->error != 0)
1687                 return;
1688
1689         pmatches = (int*) malloc(g->mlen * sizeof(unsigned int));
1690         if (pmatches == NULL) {
1691                 g->matchjump = NULL;
1692                 return;
1693         }
1694
1695         g->matchjump = (int*) malloc(g->mlen * sizeof(unsigned int));
1696         if (g->matchjump == NULL)       /* Not a fatal error */
1697                 return;
1698
1699         /* Set maximum possible jump for each character in the pattern */
1700         for (mindex = 0; mindex < g->mlen; mindex++)
1701                 g->matchjump[mindex] = 2*g->mlen - mindex - 1;
1702
1703         /* Compute pmatches[] */
1704         for (mindex = g->mlen - 1, suffix = g->mlen; mindex >= 0;
1705             mindex--, suffix--) {
1706                 pmatches[mindex] = suffix;
1707
1708                 /* If a mismatch is found, interrupting the substring,
1709                  * compute the matchjump for that position. If no
1710                  * mismatch is found, then a text substring mismatched
1711                  * against the suffix will also mismatch against the
1712                  * substring.
1713                  */
1714                 while (suffix < g->mlen
1715                     && g->must[mindex] != g->must[suffix]) {
1716                         g->matchjump[suffix] = MIN(g->matchjump[suffix],
1717                             g->mlen - mindex - 1);
1718                         suffix = pmatches[suffix];
1719                 }
1720         }
1721
1722         /* Compute the matchjump up to the last substring found to jump
1723          * to the beginning of the largest must pattern prefix matching
1724          * it's own suffix.
1725          */
1726         for (mindex = 0; mindex <= suffix; mindex++)
1727                 g->matchjump[mindex] = MIN(g->matchjump[mindex],
1728                     g->mlen + suffix - mindex);
1729
1730         ssuffix = pmatches[suffix];
1731         while (suffix < g->mlen) {
1732                 while (suffix <= ssuffix && suffix < g->mlen) {
1733                         g->matchjump[suffix] = MIN(g->matchjump[suffix],
1734                             g->mlen + ssuffix - suffix);
1735                         suffix++;
1736                 }
1737                 if (suffix < g->mlen)
1738                         ssuffix = pmatches[ssuffix];
1739         }
1740
1741         free(pmatches);
1742 }
1743
1744 /*
1745  - pluscount - count + nesting
1746  == static sopno pluscount(struct parse *p, struct re_guts *g);
1747  */
1748 static sopno                    /* nesting depth */
1749 pluscount(struct parse *p, struct re_guts *g)
1750 {
1751         sop *scan;
1752         sop s;
1753         sopno plusnest = 0;
1754         sopno maxnest = 0;
1755
1756         if (p->error != 0)
1757                 return(0);      /* there may not be an OEND */
1758
1759         scan = g->strip + 1;
1760         do {
1761                 s = *scan++;
1762                 switch (OP(s)) {
1763                 case OPLUS_:
1764                         plusnest++;
1765                         break;
1766                 case O_PLUS:
1767                         if (plusnest > maxnest)
1768                                 maxnest = plusnest;
1769                         plusnest--;
1770                         break;
1771                 }
1772         } while (OP(s) != OEND);
1773         if (plusnest != 0)
1774                 g->iflags |= BAD;
1775         return(maxnest);
1776 }