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 of the University of Toronto.
   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  * 3. All advertising materials mentioning features or use of this software
  18  *    must display the following acknowledgement:
  19  *      This product includes software developed by the University of
  20  *      California, Berkeley and its contributors.
  21  * 4. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      @(#)regcomp.c   8.4 (Berkeley) 3/19/94
  38  */
  39
  40 #if defined(LIBC_SCCS) && !defined(lint)
  41 static char sccsid[] = "@(#)regcomp.c   8.4 (Berkeley) 3/19/94";
  42 #endif /* LIBC_SCCS and not lint */
  43
  44 #include <sys/types.h>
  45 #include <stdio.h>
  46 #include <string.h>
  47 #include <ctype.h>
  48 #include <limits.h>
  49 #include <stdlib.h>
  50 #include <regex.h>
  51
  52 #include "utils.h"
  53 #include "regex2.h"
  54
  55 #include "cclass.h"
  56 #include "cname.h"
  57
  58 /*
  59  * parse structure, passed up and down to avoid global variables and
  60  * other clumsinesses
  61  */
  62 struct parse {
  63         RCHAR_T *next;          /* next character in RE */
  64         RCHAR_T *end;           /* end of string (-> NUL normally) */
  65         int error;              /* has an error been seen? */
  66         sop *strip;             /* malloced strip */
  67         sopno ssize;            /* malloced strip size (allocated) */
  68         sopno slen;             /* malloced strip length (used) */
  69         int ncsalloc;           /* number of csets allocated */
  70         struct re_guts *g;
  71 #       define  NPAREN  10      /* we need to remember () 1-9 for back refs */
  72         sopno pbegin[NPAREN];   /* -> ( ([0] unused) */
  73         sopno pend[NPAREN];     /* -> ) ([0] unused) */
  74 };
  75
  76 /* ========= begin header generated by ./mkh ========= */
  77 #ifdef __cplusplus
  78 extern "C" {
  79 #endif
  80
  81 /* === regcomp.c === */
  82 static void p_ere __P((struct parse *p, int stop));
  83 static void p_ere_exp __P((struct parse *p));
  84 static void p_str __P((struct parse *p));
  85 static void p_bre __P((struct parse *p, int end1, int end2));
  86 static int p_simp_re __P((struct parse *p, int starordinary));
  87 static int p_count __P((struct parse *p));
  88 static void p_bracket __P((struct parse *p));
  89 static void p_b_term __P((struct parse *p, cset *cs));
  90 static void p_b_cclass __P((struct parse *p, cset *cs));
  91 static void p_b_eclass __P((struct parse *p, cset *cs));
  92 static char p_b_symbol __P((struct parse *p));
  93 static char p_b_coll_elem __P((struct parse *p, int endc));
  94 static char othercase __P((int ch));
  95 static void bothcases __P((struct parse *p, int ch));
  96 static void ordinary __P((struct parse *p, int ch));
  97 static void nonnewline __P((struct parse *p));
  98 static void repeat __P((struct parse *p, sopno start, int from, int to));
  99 static int seterr __P((struct parse *p, int e));
 100 static cset *allocset __P((struct parse *p));
 101 static void freeset __P((struct parse *p, cset *cs));
 102 static int freezeset __P((struct parse *p, cset *cs));
 103 static int firstch __P((struct parse *p, cset *cs));
 104 static int nch __P((struct parse *p, cset *cs));
 105 static void mcadd __P((struct parse *p, cset *cs, char *cp));
 106 static void mcsub __P((cset *cs, char *cp));
 107 static int mcin __P((cset *cs, char *cp));
 108 static char *mcfind __P((cset *cs, char *cp));
 109 static void mcinvert __P((struct parse *p, cset *cs));
 110 static void mccase __P((struct parse *p, cset *cs));
 111 static int isinsets __P((struct re_guts *g, int c));
 112 static int samesets __P((struct re_guts *g, int c1, int c2));
 113 static void categorize __P((struct parse *p, struct re_guts *g));
 114 static sopno dupl __P((struct parse *p, sopno start, sopno finish));
 115 static void doemit __P((struct parse *p, sop op, size_t opnd));
 116 static void doinsert __P((struct parse *p, sop op, size_t opnd, sopno pos));
 117 static void dofwd __P((struct parse *p, sopno pos, sop value));
 118 static void enlarge __P((struct parse *p, sopno size));
 119 static void stripsnug __P((struct parse *p, struct re_guts *g));
 120 static void findmust __P((struct parse *p, struct re_guts *g));
 121 static sopno pluscount __P((struct parse *p, struct re_guts *g));
 122
 123 #ifdef __cplusplus
 124 }
 125 #endif
 126 /* ========= end header generated by ./mkh ========= */
 127
 128 static RCHAR_T nuls[10];                /* place to point scanner in event of error */
 129
 130 /*
 131  * macros for use with parse structure
 132  * BEWARE:  these know that the parse structure is named `p' !!!
 133  */
 134 #define PEEK()  (*p->next)
 135 #define PEEK2() (*(p->next+1))
 136 #define MORE()  (p->next < p->end)
 137 #define MORE2() (p->next+1 < p->end)
 138 #define SEE(c)  (MORE() && PEEK() == (c))
 139 #define SEETWO(a, b)    (MORE() && MORE2() && PEEK() == (a) && PEEK2() == (b))
 140 #define EAT(c)  ((SEE(c)) ? (NEXT(), 1) : 0)
 141 #define EATTWO(a, b)    ((SEETWO(a, b)) ? (NEXT2(), 1) : 0)
 142 #define NEXT()  (p->next++)
 143 #define NEXT2() (p->next += 2)
 144 #define NEXTn(n)        (p->next += (n))
 145 #define GETNEXT()       (*p->next++)
 146 #define SETERROR(e)     seterr(p, (e))
 147 #define REQUIRE(co, e)  ((co) || SETERROR(e))
 148 #define MUSTSEE(c, e)   (REQUIRE(MORE() && PEEK() == (c), e))
 149 #define MUSTEAT(c, e)   (REQUIRE(MORE() && GETNEXT() == (c), e))
 150 #define MUSTNOTSEE(c, e)        (REQUIRE(!MORE() || PEEK() != (c), e))
 151 #define EMIT(op, sopnd) doemit(p, (sop)(op), (size_t)(sopnd))
 152 #define INSERT(op, pos) doinsert(p, (sop)(op), HERE()-(pos)+1, pos)
 153 #define AHEAD(pos)              dofwd(p, pos, HERE()-(pos))
 154 #define ASTERN(sop, pos)        EMIT(sop, HERE()-pos)
 155 #define HERE()          (p->slen)
 156 #define THERE()         (p->slen - 1)
 157 #define THERETHERE()    (p->slen - 2)
 158 #define DROP(n) (p->slen -= (n))
 159
 160 #ifndef NDEBUG
 161 static int never = 0;           /* for use in asserts; shuts lint up */
 162 #else
 163 #define never   0               /* some <assert.h>s have bugs too */
 164 #endif
 165
 166 /*
 167  - regcomp - interface for parser and compilation
 168  = extern int regcomp(regex_t *, const RCHAR_T *, int);
 169  = #define      REG_BASIC       0000
 170  = #define      REG_EXTENDED    0001
 171  = #define      REG_ICASE       0002
 172  = #define      REG_NOSUB       0004
 173  = #define      REG_NEWLINE     0010
 174  = #define      REG_NOSPEC      0020
 175  = #define      REG_PEND        0040
 176  = #define      REG_DUMP        0200
 177  */
 178 int                             /* 0 success, otherwise REG_something */
 179 regcomp(regex_t *preg, const RCHAR_T *pattern, int cflags)
 180 {
 181         struct parse pa;
 182         register struct re_guts *g;
 183         register struct parse *p = &pa;
 184         register int i;
 185         register size_t len;
 186 #ifdef REDEBUG
 187 #       define  GOODFLAGS(f)    (f)
 188 #else
 189 #       define  GOODFLAGS(f)    ((f)&~REG_DUMP)
 190 #endif
 191
 192         cflags = GOODFLAGS(cflags);
 193         if ((cflags&REG_EXTENDED) && (cflags&REG_NOSPEC))
 194                 return(REG_INVARG);
 195
 196         if (cflags&REG_PEND) {
 197                 if (preg->re_endp < pattern)
 198                         return(REG_INVARG);
 199                 len = preg->re_endp - pattern;
 200         } else
 201                 len = STRLEN(pattern);
 202
 203         /* do the mallocs early so failure handling is easy */
 204         g = (struct re_guts *)malloc(sizeof(struct re_guts) +
 205                                                         (NC-1)*sizeof(cat_t));
 206         if (g == NULL)
 207                 return(REG_ESPACE);
 208         p->ssize = len/(size_t)2*(size_t)3 + (size_t)1; /* ugh */
 209         p->strip = (sop *)malloc(p->ssize * sizeof(sop));
 210         p->slen = 0;
 211         if (p->strip == NULL) {
 212                 free((char *)g);
 213                 return(REG_ESPACE);
 214         }
 215
 216         /* set things up */
 217         p->g = g;
 218         p->next = (RCHAR_T *)pattern;   /* convenience; we do not modify it */
 219         p->end = p->next + len;
 220         p->error = 0;
 221         p->ncsalloc = 0;
 222         for (i = 0; i < NPAREN; i++) {
 223                 p->pbegin[i] = 0;
 224                 p->pend[i] = 0;
 225         }
 226         g->csetsize = NC;
 227         g->sets = NULL;
 228         g->setbits = NULL;
 229         g->ncsets = 0;
 230         g->cflags = cflags;
 231         g->iflags = 0;
 232         g->nbol = 0;
 233         g->neol = 0;
 234         g->must = NULL;
 235         g->mlen = 0;
 236         g->nsub = 0;
 237 #if 0
 238         g->ncategories = 1;     /* category 0 is "everything else" */
 239         g->categories = &g->catspace[-(CHAR_MIN)];
 240         (void) memset((char *)g->catspace, 0, NC*sizeof(cat_t));
 241 #endif
 242         g->backrefs = 0;
 243
 244         /* do it */
 245         EMIT(OEND, 0);
 246         g->firststate = THERE();
 247         if (cflags&REG_EXTENDED)
 248                 p_ere(p, OUT);
 249         else if (cflags&REG_NOSPEC)
 250                 p_str(p);
 251         else
 252                 p_bre(p, OUT, OUT);
 253         EMIT(OEND, 0);
 254         g->laststate = THERE();
 255
 256         /* tidy up loose ends and fill things in */
 257         categorize(p, g);
 258         stripsnug(p, g);
 259         findmust(p, g);
 260         g->nplus = pluscount(p, g);
 261         g->magic = MAGIC2;
 262         preg->re_nsub = g->nsub;
 263         preg->re_g = g;
 264         preg->re_magic = MAGIC1;
 265 #ifndef REDEBUG
 266         /* not debugging, so can't rely on the assert() in regexec() */
 267         if (g->iflags&BAD)
 268                 SETERROR(REG_ASSERT);
 269 #endif
 270
 271         /* win or lose, we're done */
 272         if (p->error != 0)      /* lose */
 273                 regfree(preg);
 274         return(p->error);
 275 }
 276
 277 /*
 278  - p_ere - ERE parser top level, concatenation and alternation
 279  == static void p_ere(register struct parse *p, int stop);
 280  */
 281 static void
 282 p_ere(register struct parse *p, int stop)
 283
 284                                 /* character this ERE should end at */
 285 {
 286         register char c;
 287         register sopno prevback;
 288         register sopno prevfwd;
 289         register sopno conc;
 290         register int first = 1;         /* is this the first alternative? */
 291
 292         for (;;) {
 293                 /* do a bunch of concatenated expressions */
 294                 conc = HERE();
 295                 while (MORE() && (c = PEEK()) != '|' && c != stop)
 296                         p_ere_exp(p);
 297                 REQUIRE(HERE() != conc, REG_EMPTY);     /* require nonempty */
 298
 299                 if (!EAT('|'))
 300                         break;          /* NOTE BREAK OUT */
 301
 302                 if (first) {
 303                         INSERT(OCH_, conc);     /* offset is wrong */
 304                         prevfwd = conc;
 305                         prevback = conc;
 306                         first = 0;
 307                 }
 308                 ASTERN(OOR1, prevback);
 309                 prevback = THERE();
 310                 AHEAD(prevfwd);                 /* fix previous offset */
 311                 prevfwd = HERE();
 312                 EMIT(OOR2, 0);                  /* offset is very wrong */
 313         }
 314
 315         if (!first) {           /* tail-end fixups */
 316                 AHEAD(prevfwd);
 317                 ASTERN(O_CH, prevback);
 318         }
 319
 320         assert(!MORE() || SEE(stop));
 321 }
 322
 323 /*
 324  - p_ere_exp - parse one subERE, an atom possibly followed by a repetition op
 325  == static void p_ere_exp(register struct parse *p);
 326  */
 327 static void
 328 p_ere_exp(register struct parse *p)
 329 {
 330         register char c;
 331         register sopno pos;
 332         register int count;
 333         register int count2;
 334         register sopno subno;
 335         int wascaret = 0;
 336
 337         assert(MORE());         /* caller should have ensured this */
 338         c = GETNEXT();
 339
 340         pos = HERE();
 341         switch (c) {
 342         case '(':
 343                 REQUIRE(MORE(), REG_EPAREN);
 344                 p->g->nsub++;
 345                 subno = p->g->nsub;
 346                 if (subno < NPAREN)
 347                         p->pbegin[subno] = HERE();
 348                 EMIT(OLPAREN, subno);
 349                 if (!SEE(')'))
 350                         p_ere(p, ')');
 351                 if (subno < NPAREN) {
 352                         p->pend[subno] = HERE();
 353                         assert(p->pend[subno] != 0);
 354                 }
 355                 EMIT(ORPAREN, subno);
 356                 MUSTEAT(')', REG_EPAREN);
 357                 break;
 358 #ifndef POSIX_MISTAKE
 359         case ')':               /* happens only if no current unmatched ( */
 360                 /*
 361                  * You may ask, why the ifndef?  Because I didn't notice
 362                  * this until slightly too late for 1003.2, and none of the
 363                  * other 1003.2 regular-expression reviewers noticed it at
 364                  * all.  So an unmatched ) is legal POSIX, at least until
 365                  * we can get it fixed.
 366                  */
 367                 SETERROR(REG_EPAREN);
 368                 break;
 369 #endif
 370         case '^':
 371                 EMIT(OBOL, 0);
 372                 p->g->iflags |= USEBOL;
 373                 p->g->nbol++;
 374                 wascaret = 1;
 375                 break;
 376         case '$':
 377                 EMIT(OEOL, 0);
 378                 p->g->iflags |= USEEOL;
 379                 p->g->neol++;
 380                 break;
 381         case '|':
 382                 SETERROR(REG_EMPTY);
 383                 break;
 384         case '*':
 385         case '+':
 386         case '?':
 387                 SETERROR(REG_BADRPT);
 388                 break;
 389         case '.':
 390                 if (p->g->cflags&REG_NEWLINE)
 391                         nonnewline(p);
 392                 else
 393                         EMIT(OANY, 0);
 394                 break;
 395         case '[':
 396                 p_bracket(p);
 397                 break;
 398         case '\\':
 399                 REQUIRE(MORE(), REG_EESCAPE);
 400                 c = GETNEXT();
 401                 ordinary(p, c);
 402                 break;
 403         case '{':               /* okay as ordinary except if digit follows */
 404                 REQUIRE(!MORE() || !isdigit(PEEK()), REG_BADRPT);
 405                 /* FALLTHROUGH */
 406         default:
 407                 ordinary(p, c);
 408                 break;
 409         }
 410
 411         if (!MORE())
 412                 return;
 413         c = PEEK();
 414         /* we call { a repetition if followed by a digit */
 415         if (!( c == '*' || c == '+' || c == '?' ||
 416                                 (c == '{' && MORE2() && isdigit(PEEK2())) ))
 417                 return;         /* no repetition, we're done */
 418         NEXT();
 419
 420         REQUIRE(!wascaret, REG_BADRPT);
 421         switch (c) {
 422         case '*':       /* implemented as +? */
 423                 /* this case does not require the (y|) trick, noKLUDGE */
 424                 INSERT(OPLUS_, pos);
 425                 ASTERN(O_PLUS, pos);
 426                 INSERT(OQUEST_, pos);
 427                 ASTERN(O_QUEST, pos);
 428                 break;
 429         case '+':
 430                 INSERT(OPLUS_, pos);
 431                 ASTERN(O_PLUS, pos);
 432                 break;
 433         case '?':
 434                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
 435                 INSERT(OCH_, pos);              /* offset slightly wrong */
 436                 ASTERN(OOR1, pos);              /* this one's right */
 437                 AHEAD(pos);                     /* fix the OCH_ */
 438                 EMIT(OOR2, 0);                  /* offset very wrong... */
 439                 AHEAD(THERE());                 /* ...so fix it */
 440                 ASTERN(O_CH, THERETHERE());
 441                 break;
 442         case '{':
 443                 count = p_count(p);
 444                 if (EAT(',')) {
 445                         if (isdigit(PEEK())) {
 446                                 count2 = p_count(p);
 447                                 REQUIRE(count <= count2, REG_BADBR);
 448                         } else          /* single number with comma */
 449                                 count2 = INFINITY;
 450                 } else          /* just a single number */
 451                         count2 = count;
 452                 repeat(p, pos, count, count2);
 453                 if (!EAT('}')) {        /* error heuristics */
 454                         while (MORE() && PEEK() != '}')
 455                                 NEXT();
 456                         REQUIRE(MORE(), REG_EBRACE);
 457                         SETERROR(REG_BADBR);
 458                 }
 459                 break;
 460         }
 461
 462         if (!MORE())
 463                 return;
 464         c = PEEK();
 465         if (!( c == '*' || c == '+' || c == '?' ||
 466                                 (c == '{' && MORE2() && isdigit(PEEK2())) ) )
 467                 return;
 468         SETERROR(REG_BADRPT);
 469 }
 470
 471 /*
 472  - p_str - string (no metacharacters) "parser"
 473  == static void p_str(register struct parse *p);
 474  */
 475 static void
 476 p_str(register struct parse *p)
 477 {
 478         REQUIRE(MORE(), REG_EMPTY);
 479         while (MORE())
 480                 ordinary(p, GETNEXT());
 481 }
 482
 483 /*
 484  - p_bre - BRE parser top level, anchoring and concatenation
 485  == static void p_bre(register struct parse *p, register int end1, \
 486  ==     register int end2);
 487  * Giving end1 as OUT essentially eliminates the end1/end2 check.
 488  *
 489  * This implementation is a bit of a kludge, in that a trailing $ is first
 490  * taken as an ordinary character and then revised to be an anchor.  The
 491  * only undesirable side effect is that '$' gets included as a character
 492  * category in such cases.  This is fairly harmless; not worth fixing.
 493  * The amount of lookahead needed to avoid this kludge is excessive.
 494  */
 495 static void
 496 p_bre(register struct parse *p, register int end1, register int end2)
 497
 498                                 /* first terminating character */
 499                                 /* second terminating character */
 500 {
 501         register sopno start = HERE();
 502         register int first = 1;                 /* first subexpression? */
 503         register int wasdollar = 0;
 504
 505         if (EAT('^')) {
 506                 EMIT(OBOL, 0);
 507                 p->g->iflags |= USEBOL;
 508                 p->g->nbol++;
 509         }
 510         while (MORE() && !SEETWO(end1, end2)) {
 511                 wasdollar = p_simp_re(p, first);
 512                 first = 0;
 513         }
 514         if (wasdollar) {        /* oops, that was a trailing anchor */
 515                 DROP(1);
 516                 EMIT(OEOL, 0);
 517                 p->g->iflags |= USEEOL;
 518                 p->g->neol++;
 519         }
 520
 521         REQUIRE(HERE() != start, REG_EMPTY);    /* require nonempty */
 522 }
 523
 524 /*
 525  - p_simp_re - parse a simple RE, an atom possibly followed by a repetition
 526  == static int p_simp_re(register struct parse *p, int starordinary);
 527  */
 528 static int                      /* was the simple RE an unbackslashed $? */
 529 p_simp_re(register struct parse *p, int starordinary)
 530
 531                                 /* is a leading * an ordinary character? */
 532 {
 533         register int c;
 534         register int count;
 535         register int count2;
 536         register sopno pos;
 537         register int i;
 538         register sopno subno;
 539 #       define  BACKSL  (1<<RCHAR_BIT)
 540
 541         pos = HERE();           /* repetion op, if any, covers from here */
 542
 543         assert(MORE());         /* caller should have ensured this */
 544         c = GETNEXT();
 545         if (c == '\\') {
 546                 REQUIRE(MORE(), REG_EESCAPE);
 547                 c = BACKSL | (unsigned char)GETNEXT();
 548         }
 549         switch (c) {
 550         case '.':
 551                 if (p->g->cflags&REG_NEWLINE)
 552                         nonnewline(p);
 553                 else
 554                         EMIT(OANY, 0);
 555                 break;
 556         case '[':
 557                 p_bracket(p);
 558                 break;
 559         case BACKSL|'{':
 560                 SETERROR(REG_BADRPT);
 561                 break;
 562         case BACKSL|'(':
 563                 p->g->nsub++;
 564                 subno = p->g->nsub;
 565                 if (subno < NPAREN)
 566                         p->pbegin[subno] = HERE();
 567                 EMIT(OLPAREN, subno);
 568                 /* the MORE here is an error heuristic */
 569                 if (MORE() && !SEETWO('\\', ')'))
 570                         p_bre(p, '\\', ')');
 571                 if (subno < NPAREN) {
 572                         p->pend[subno] = HERE();
 573                         assert(p->pend[subno] != 0);
 574                 }
 575                 EMIT(ORPAREN, subno);
 576                 REQUIRE(EATTWO('\\', ')'), REG_EPAREN);
 577                 break;
 578         case BACKSL|')':        /* should not get here -- must be user */
 579         case BACKSL|'}':
 580                 SETERROR(REG_EPAREN);
 581                 break;
 582         case BACKSL|'1':
 583         case BACKSL|'2':
 584         case BACKSL|'3':
 585         case BACKSL|'4':
 586         case BACKSL|'5':
 587         case BACKSL|'6':
 588         case BACKSL|'7':
 589         case BACKSL|'8':
 590         case BACKSL|'9':
 591                 i = (c&~BACKSL) - '0';
 592                 assert(i < NPAREN);
 593                 if (p->pend[i] != 0) {
 594                         assert(i <= p->g->nsub);
 595                         EMIT(OBACK_, i);
 596                         assert(p->pbegin[i] != 0);
 597                         assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
 598                         assert(OP(p->strip[p->pend[i]]) == ORPAREN);
 599                         (void) dupl(p, p->pbegin[i]+1, p->pend[i]);
 600                         EMIT(O_BACK, i);
 601                 } else
 602                         SETERROR(REG_ESUBREG);
 603                 p->g->backrefs = 1;
 604                 break;
 605         case '*':
 606                 REQUIRE(starordinary, REG_BADRPT);
 607                 /* FALLTHROUGH */
 608         default:
 609                 ordinary(p, c &~ BACKSL);
 610                 break;
 611         }
 612
 613         if (EAT('*')) {         /* implemented as +? */
 614                 /* this case does not require the (y|) trick, noKLUDGE */
 615                 INSERT(OPLUS_, pos);
 616                 ASTERN(O_PLUS, pos);
 617                 INSERT(OQUEST_, pos);
 618                 ASTERN(O_QUEST, pos);
 619         } else if (EATTWO('\\', '{')) {
 620                 count = p_count(p);
 621                 if (EAT(',')) {
 622                         if (MORE() && isdigit(PEEK())) {
 623                                 count2 = p_count(p);
 624                                 REQUIRE(count <= count2, REG_BADBR);
 625                         } else          /* single number with comma */
 626                                 count2 = INFINITY;
 627                 } else          /* just a single number */
 628                         count2 = count;
 629                 repeat(p, pos, count, count2);
 630                 if (!EATTWO('\\', '}')) {       /* error heuristics */
 631                         while (MORE() && !SEETWO('\\', '}'))
 632                                 NEXT();
 633                         REQUIRE(MORE(), REG_EBRACE);
 634                         SETERROR(REG_BADBR);
 635                 }
 636         } else if (c == (unsigned char)'$')     /* $ (but not \$) ends it */
 637                 return(1);
 638
 639         return(0);
 640 }
 641
 642 /*
 643  - p_count - parse a repetition count
 644  == static int p_count(register struct parse *p);
 645  */
 646 static int                      /* the value */
 647 p_count(register struct parse *p)
 648 {
 649         register int count = 0;
 650         register int ndigits = 0;
 651
 652         while (MORE() && isdigit(PEEK()) && count <= DUPMAX) {
 653                 count = count*10 + (GETNEXT() - '0');
 654                 ndigits++;
 655         }
 656
 657         REQUIRE(ndigits > 0 && count <= DUPMAX, REG_BADBR);
 658         return(count);
 659 }
 660
 661 /*
 662  - p_bracket - parse a bracketed character list
 663  == static void p_bracket(register struct parse *p);
 664  *
 665  * Note a significant property of this code:  if the allocset() did SETERROR,
 666  * no set operations are done.
 667  */
 668 static void
 669 p_bracket(register struct parse *p)
 670 {
 671         register char c;
 672         register cset *cs = allocset(p);
 673         register int invert = 0;
 674         static RCHAR_T bow[] = { '[', ':', '<', ':', ']', ']' };
 675         static RCHAR_T eow[] = { '[', ':', '>', ':', ']', ']' };
 676
 677         /* Dept of Truly Sickening Special-Case Kludges */
 678         if (p->next + 5 < p->end && MEMCMP(p->next, bow, 6) == 0) {
 679                 EMIT(OBOW, 0);
 680                 NEXTn(6);
 681                 return;
 682         }
 683         if (p->next + 5 < p->end && MEMCMP(p->next, eow, 6) == 0) {
 684                 EMIT(OEOW, 0);
 685                 NEXTn(6);
 686                 return;
 687         }
 688
 689         if (EAT('^'))
 690                 invert++;       /* make note to invert set at end */
 691         if (EAT(']'))
 692                 CHadd(cs, ']');
 693         else if (EAT('-'))
 694                 CHadd(cs, '-');
 695         while (MORE() && PEEK() != ']' && !SEETWO('-', ']'))
 696                 p_b_term(p, cs);
 697         if (EAT('-'))
 698                 CHadd(cs, '-');
 699         MUSTEAT(']', REG_EBRACK);
 700
 701         if (p->error != 0)      /* don't mess things up further */
 702                 return;
 703
 704         if (p->g->cflags&REG_ICASE) {
 705                 register int i;
 706                 register int ci;
 707
 708                 for (i = p->g->csetsize - 1; i >= 0; i--)
 709                         if (CHIN(cs, i) && isalpha(i)) {
 710                                 ci = othercase(i);
 711                                 if (ci != i)
 712                                         CHadd(cs, ci);
 713                         }
 714                 if (cs->multis != NULL)
 715                         mccase(p, cs);
 716         }
 717         if (invert) {
 718                 register int i;
 719
 720                 for (i = p->g->csetsize - 1; i >= 0; i--)
 721                         if (CHIN(cs, i))
 722                                 CHsub(cs, i);
 723                         else
 724                                 CHadd(cs, i);
 725                 if (p->g->cflags&REG_NEWLINE)
 726                         CHsub(cs, '\n');
 727                 if (cs->multis != NULL)
 728                         mcinvert(p, cs);
 729         }
 730
 731         assert(cs->multis == NULL);             /* xxx */
 732
 733         if (nch(p, cs) == 1) {          /* optimize singleton sets */
 734                 ordinary(p, firstch(p, cs));
 735                 freeset(p, cs);
 736         } else
 737                 EMIT(OANYOF, freezeset(p, cs));
 738 }
 739
 740 /*
 741  - p_b_term - parse one term of a bracketed character list
 742  == static void p_b_term(register struct parse *p, register cset *cs);
 743  */
 744 static void
 745 p_b_term(register struct parse *p, register cset *cs)
 746 {
 747         register char c;
 748         register char start, finish;
 749         register int i;
 750
 751         /* classify what we've got */
 752         switch ((MORE()) ? PEEK() : '\0') {
 753         case '[':
 754                 c = (MORE2()) ? PEEK2() : '\0';
 755                 break;
 756         case '-':
 757                 SETERROR(REG_ERANGE);
 758                 return;                 /* NOTE RETURN */
 759                 break;
 760         default:
 761                 c = '\0';
 762                 break;
 763         }
 764
 765         switch (c) {
 766         case ':':               /* character class */
 767                 NEXT2();
 768                 REQUIRE(MORE(), REG_EBRACK);
 769                 c = PEEK();
 770                 REQUIRE(c != '-' && c != ']', REG_ECTYPE);
 771                 p_b_cclass(p, cs);
 772                 REQUIRE(MORE(), REG_EBRACK);
 773                 REQUIRE(EATTWO(':', ']'), REG_ECTYPE);
 774                 break;
 775         case '=':               /* equivalence class */
 776                 NEXT2();
 777                 REQUIRE(MORE(), REG_EBRACK);
 778                 c = PEEK();
 779                 REQUIRE(c != '-' && c != ']', REG_ECOLLATE);
 780                 p_b_eclass(p, cs);
 781                 REQUIRE(MORE(), REG_EBRACK);
 782                 REQUIRE(EATTWO('=', ']'), REG_ECOLLATE);
 783                 break;
 784         default:                /* symbol, ordinary character, or range */
 785 /* xxx revision needed for multichar stuff */
 786                 start = p_b_symbol(p);
 787                 if (SEE('-') && MORE2() && PEEK2() != ']') {
 788                         /* range */
 789                         NEXT();
 790                         if (EAT('-'))
 791                                 finish = '-';
 792                         else
 793                                 finish = p_b_symbol(p);
 794                 } else
 795                         finish = start;
 796 /* xxx what about signed chars here... */
 797                 REQUIRE(start <= finish, REG_ERANGE);
 798                 for (i = start; i <= finish; i++)
 799                         CHadd(cs, i);
 800                 break;
 801         }
 802 }
 803
 804 /*
 805  - p_b_cclass - parse a character-class name and deal with it
 806  == static void p_b_cclass(register struct parse *p, register cset *cs);
 807  */
 808 static void
 809 p_b_cclass(register struct parse *p, register cset *cs)
 810 {
 811         register RCHAR_T *sp = p->next;
 812         register struct cclass *cp;
 813         register size_t len;
 814         register char *u;
 815         register char c;
 816
 817         while (MORE() && isalpha(PEEK()))
 818                 NEXT();
 819         len = p->next - sp;
 820         for (cp = cclasses; cp->name != NULL; cp++)
 821                 if (STRLEN(cp->name) == len && !MEMCMP(cp->name, sp, len))
 822                         break;
 823         if (cp->name == NULL) {
 824                 /* oops, didn't find it */
 825                 SETERROR(REG_ECTYPE);
 826                 return;
 827         }
 828
 829         u = cp->chars;
 830         while ((c = *u++) != '\0')
 831                 CHadd(cs, c);
 832         for (u = cp->multis; *u != '\0'; u += strlen(u) + 1)
 833                 MCadd(p, cs, u);
 834 }
 835
 836 /*
 837  - p_b_eclass - parse an equivalence-class name and deal with it
 838  == static void p_b_eclass(register struct parse *p, register cset *cs);
 839  *
 840  * This implementation is incomplete. xxx
 841  */
 842 static void
 843 p_b_eclass(register struct parse *p, register cset *cs)
 844 {
 845         register char c;
 846
 847         c = p_b_coll_elem(p, '=');
 848         CHadd(cs, c);
 849 }
 850
 851 /*
 852  - p_b_symbol - parse a character or [..]ed multicharacter collating symbol
 853  == static char p_b_symbol(register struct parse *p);
 854  */
 855 static char                     /* value of symbol */
 856 p_b_symbol(register struct parse *p)
 857 {
 858         register char value;
 859
 860         REQUIRE(MORE(), REG_EBRACK);
 861         if (!EATTWO('[', '.'))
 862                 return(GETNEXT());
 863
 864         /* collating symbol */
 865         value = p_b_coll_elem(p, '.');
 866         REQUIRE(EATTWO('.', ']'), REG_ECOLLATE);
 867         return(value);
 868 }
 869
 870 /*
 871  - p_b_coll_elem - parse a collating-element name and look it up
 872  == static char p_b_coll_elem(register struct parse *p, int endc);
 873  */
 874 static char                     /* value of collating element */
 875 p_b_coll_elem(register struct parse *p, int endc)
 876
 877                                 /* name ended by endc,']' */
 878 {
 879         register RCHAR_T *sp = p->next;
 880         register struct cname *cp;
 881         register int len;
 882         register char c;
 883
 884         while (MORE() && !SEETWO(endc, ']'))
 885                 NEXT();
 886         if (!MORE()) {
 887                 SETERROR(REG_EBRACK);
 888                 return(0);
 889         }
 890         len = p->next - sp;
 891         for (cp = cnames; cp->name != NULL; cp++)
 892                 if (STRLEN(cp->name) == len && MEMCMP(cp->name, sp, len))
 893                         return(cp->code);       /* known name */
 894         if (len == 1)
 895                 return(*sp);    /* single character */
 896         SETERROR(REG_ECOLLATE);                 /* neither */
 897         return(0);
 898 }
 899
 900 /*
 901  - othercase - return the case counterpart of an alphabetic
 902  == static char othercase(int ch);
 903  */
 904 static char                     /* if no counterpart, return ch */
 905 othercase(int ch)
 906 {
 907         assert(isalpha(ch));
 908         if (isupper(ch))
 909                 return(tolower(ch));
 910         else if (islower(ch))
 911                 return(toupper(ch));
 912         else                    /* peculiar, but could happen */
 913                 return(ch);
 914 }
 915
 916 /*
 917  - bothcases - emit a dualcase version of a two-case character
 918  == static void bothcases(register struct parse *p, int ch);
 919  *
 920  * Boy, is this implementation ever a kludge...
 921  */
 922 static void
 923 bothcases(register struct parse *p, int ch)
 924 {
 925         register RCHAR_T *oldnext = p->next;
 926         register RCHAR_T *oldend = p->end;
 927         RCHAR_T bracket[3];
 928
 929         assert(othercase(ch) != ch);    /* p_bracket() would recurse */
 930         p->next = bracket;
 931         p->end = bracket+2;
 932         bracket[0] = ch;
 933         bracket[1] = ']';
 934         bracket[2] = '\0';
 935         p_bracket(p);
 936         assert(p->next == bracket+2);
 937         p->next = oldnext;
 938         p->end = oldend;
 939 }
 940
 941 /*
 942  - ordinary - emit an ordinary character
 943  == static void ordinary(register struct parse *p, register int ch);
 944  */
 945 static void
 946 ordinary(register struct parse *p, register int ch)
 947 {
 948 /*
 949         register cat_t *cap = p->g->categories;
 950 */
 951
 952         if ((p->g->cflags&REG_ICASE) && isalpha(ch) && othercase(ch) != ch)
 953                 bothcases(p, ch);
 954         else {
 955                 EMIT(OCHAR, (UCHAR_T)ch);
 956 /*
 957                 if (cap[ch] == 0)
 958                         cap[ch] = p->g->ncategories++;
 959 */
 960         }
 961 }
 962
 963 /*
 964  - nonnewline - emit REG_NEWLINE version of OANY
 965  == static void nonnewline(register struct parse *p);
 966  *
 967  * Boy, is this implementation ever a kludge...
 968  */
 969 static void
 970 nonnewline(register struct parse *p)
 971 {
 972         register RCHAR_T *oldnext = p->next;
 973         register RCHAR_T *oldend = p->end;
 974         RCHAR_T bracket[4];
 975
 976         p->next = bracket;
 977         p->end = bracket+3;
 978         bracket[0] = '^';
 979         bracket[1] = '\n';
 980         bracket[2] = ']';
 981         bracket[3] = '\0';
 982         p_bracket(p);
 983         assert(p->next == bracket+3);
 984         p->next = oldnext;
 985         p->end = oldend;
 986 }
 987
 988 /*
 989  - repeat - generate code for a bounded repetition, recursively if needed
 990  == static void repeat(register struct parse *p, sopno start, int from, int to);
 991  */
 992 static void
 993 repeat(register struct parse *p, sopno start, int from, int to)
 994
 995                                 /* operand from here to end of strip */
 996                                 /* repeated from this number */
 997                                 /* to this number of times (maybe INFINITY) */
 998 {
 999         register sopno finish = HERE();
1000 #       define  N       2
1001 #       define  INF     3
1002 #       define  REP(f, t)       ((f)*8 + (t))
1003 #       define  MAP(n)  (((n) <= 1) ? (n) : ((n) == INFINITY) ? INF : N)
1004         register sopno copy;
1005
1006         if (p->error != 0)      /* head off possible runaway recursion */
1007                 return;
1008
1009         assert(from <= to);
1010
1011         switch (REP(MAP(from), MAP(to))) {
1012         case REP(0, 0):                 /* must be user doing this */
1013                 DROP(finish-start);     /* drop the operand */
1014                 break;
1015         case REP(0, 1):                 /* as x{1,1}? */
1016         case REP(0, N):                 /* as x{1,n}? */
1017         case REP(0, INF):               /* as x{1,}? */
1018                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1019                 INSERT(OCH_, start);            /* offset is wrong... */
1020                 repeat(p, start+1, 1, to);
1021                 ASTERN(OOR1, start);
1022                 AHEAD(start);                   /* ... fix it */
1023                 EMIT(OOR2, 0);
1024                 AHEAD(THERE());
1025                 ASTERN(O_CH, THERETHERE());
1026                 break;
1027         case REP(1, 1):                 /* trivial case */
1028                 /* done */
1029                 break;
1030         case REP(1, N):                 /* as x?x{1,n-1} */
1031                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1032                 INSERT(OCH_, start);
1033                 ASTERN(OOR1, start);
1034                 AHEAD(start);
1035                 EMIT(OOR2, 0);                  /* offset very wrong... */
1036                 AHEAD(THERE());                 /* ...so fix it */
1037                 ASTERN(O_CH, THERETHERE());
1038                 copy = dupl(p, start+1, finish+1);
1039                 assert(copy == finish+4);
1040                 repeat(p, copy, 1, to-1);
1041                 break;
1042         case REP(1, INF):               /* as x+ */
1043                 INSERT(OPLUS_, start);
1044                 ASTERN(O_PLUS, start);
1045                 break;
1046         case REP(N, N):                 /* as xx{m-1,n-1} */
1047                 copy = dupl(p, start, finish);
1048                 repeat(p, copy, from-1, to-1);
1049                 break;
1050         case REP(N, INF):               /* as xx{n-1,INF} */
1051                 copy = dupl(p, start, finish);
1052                 repeat(p, copy, from-1, to);
1053                 break;
1054         default:                        /* "can't happen" */
1055                 SETERROR(REG_ASSERT);   /* just in case */
1056                 break;
1057         }
1058 }
1059
1060 /*
1061  - seterr - set an error condition
1062  == static int seterr(register struct parse *p, int e);
1063  */
1064 static int                      /* useless but makes type checking happy */
1065 seterr(register struct parse *p, int e)
1066 {
1067         if (p->error == 0)      /* keep earliest error condition */
1068                 p->error = e;
1069         p->next = nuls;         /* try to bring things to a halt */
1070         p->end = nuls;
1071         return(0);              /* make the return value well-defined */
1072 }
1073
1074 /*
1075  - allocset - allocate a set of characters for []
1076  == static cset *allocset(register struct parse *p);
1077  */
1078 static cset *
1079 allocset(register struct parse *p)
1080 {
1081         register int no = p->g->ncsets++;
1082         register size_t nc;
1083         register size_t nbytes;
1084         register cset *cs;
1085         register size_t css = (size_t)p->g->csetsize;
1086         register int i;
1087
1088         if (no >= p->ncsalloc) {        /* need another column of space */
1089                 p->ncsalloc += CHAR_BIT;
1090                 nc = p->ncsalloc;
1091                 assert(nc % CHAR_BIT == 0);
1092                 nbytes = nc / CHAR_BIT * css;
1093                 if (p->g->sets == NULL)
1094                         p->g->sets = (cset *)malloc(nc * sizeof(cset));
1095                 else
1096                         p->g->sets = (cset *)realloc((char *)p->g->sets,
1097                                                         nc * sizeof(cset));
1098                 if (p->g->setbits == NULL)
1099                         p->g->setbits = (uch *)malloc(nbytes);
1100                 else {
1101                         p->g->setbits = (uch *)realloc((char *)p->g->setbits,
1102                                                                 nbytes);
1103                         /* xxx this isn't right if setbits is now NULL */
1104                         for (i = 0; i < no; i++)
1105                                 p->g->sets[i].ptr = p->g->setbits + css*(i/CHAR_BIT);
1106                 }
1107                 if (p->g->sets != NULL && p->g->setbits != NULL)
1108                         (void) memset((char *)p->g->setbits + (nbytes - css),
1109                                                                 0, css);
1110                 else {
1111                         no = 0;
1112                         SETERROR(REG_ESPACE);
1113                         /* caller's responsibility not to do set ops */
1114                 }
1115         }
1116
1117         assert(p->g->sets != NULL);     /* xxx */
1118         cs = &p->g->sets[no];
1119         cs->ptr = p->g->setbits + css*((no)/CHAR_BIT);
1120         cs->mask = 1 << ((no) % CHAR_BIT);
1121         cs->hash = 0;
1122         cs->smultis = 0;
1123         cs->multis = NULL;
1124
1125         return(cs);
1126 }
1127
1128 /*
1129  - freeset - free a now-unused set
1130  == static void freeset(register struct parse *p, register cset *cs);
1131  */
1132 static void
1133 freeset(register struct parse *p, register cset *cs)
1134 {
1135         register int i;
1136         register cset *top = &p->g->sets[p->g->ncsets];
1137         register size_t css = (size_t)p->g->csetsize;
1138
1139         for (i = 0; i < css; i++)
1140                 CHsub(cs, i);
1141         if (cs == top-1)        /* recover only the easy case */
1142                 p->g->ncsets--;
1143 }
1144
1145 /*
1146  - freezeset - final processing on a set of characters
1147  == static int freezeset(register struct parse *p, register cset *cs);
1148  *
1149  * The main task here is merging identical sets.  This is usually a waste
1150  * of time (although the hash code minimizes the overhead), but can win
1151  * big if REG_ICASE is being used.  REG_ICASE, by the way, is why the hash
1152  * is done using addition rather than xor -- all ASCII [aA] sets xor to
1153  * the same value!
1154  */
1155 static int                      /* set number */
1156 freezeset(register struct parse *p, register cset *cs)
1157 {
1158         register uch h = cs->hash;
1159         register int i;
1160         register cset *top = &p->g->sets[p->g->ncsets];
1161         register cset *cs2;
1162         register size_t css = (size_t)p->g->csetsize;
1163
1164         /* look for an earlier one which is the same */
1165         for (cs2 = &p->g->sets[0]; cs2 < top; cs2++)
1166                 if (cs2->hash == h && cs2 != cs) {
1167                         /* maybe */
1168                         for (i = 0; i < css; i++)
1169                                 if (!!CHIN(cs2, i) != !!CHIN(cs, i))
1170                                         break;          /* no */
1171                         if (i == css)
1172                                 break;                  /* yes */
1173                 }
1174
1175         if (cs2 < top) {        /* found one */
1176                 freeset(p, cs);
1177                 cs = cs2;
1178         }
1179
1180         return((int)(cs - p->g->sets));
1181 }
1182
1183 /*
1184  - firstch - return first character in a set (which must have at least one)
1185  == static int firstch(register struct parse *p, register cset *cs);
1186  */
1187 static int                      /* character; there is no "none" value */
1188 firstch(register struct parse *p, register cset *cs)
1189 {
1190         register int i;
1191         register size_t css = (size_t)p->g->csetsize;
1192
1193         for (i = 0; i < css; i++)
1194                 if (CHIN(cs, i))
1195                         return((char)i);
1196         assert(never);
1197         return(0);              /* arbitrary */
1198 }
1199
1200 /*
1201  - nch - number of characters in a set
1202  == static int nch(register struct parse *p, register cset *cs);
1203  */
1204 static int
1205 nch(register struct parse *p, register cset *cs)
1206 {
1207         register int i;
1208         register size_t css = (size_t)p->g->csetsize;
1209         register int n = 0;
1210
1211         for (i = 0; i < css; i++)
1212                 if (CHIN(cs, i))
1213                         n++;
1214         return(n);
1215 }
1216
1217 /*
1218  - mcadd - add a collating element to a cset
1219  == static void mcadd(register struct parse *p, register cset *cs, \
1220  ==     register char *cp);
1221  */
1222 static void
1223 mcadd(register struct parse *p, register cset *cs, register char *cp)
1224 {
1225         register size_t oldend = cs->smultis;
1226
1227         cs->smultis += strlen(cp) + 1;
1228         if (cs->multis == NULL)
1229                 cs->multis = malloc(cs->smultis);
1230         else
1231                 cs->multis = realloc(cs->multis, cs->smultis);
1232         if (cs->multis == NULL) {
1233                 SETERROR(REG_ESPACE);
1234                 return;
1235         }
1236
1237         (void) strcpy(cs->multis + oldend - 1, cp);
1238         cs->multis[cs->smultis - 1] = '\0';
1239 }
1240
1241 /*
1242  - mcsub - subtract a collating element from a cset
1243  == static void mcsub(register cset *cs, register char *cp);
1244  */
1245 static void
1246 mcsub(register cset *cs, register char *cp)
1247 {
1248         register char *fp = mcfind(cs, cp);
1249         register size_t len = strlen(fp);
1250
1251         assert(fp != NULL);
1252         (void) memmove(fp, fp + len + 1,
1253                                 cs->smultis - (fp + len + 1 - cs->multis));
1254         cs->smultis -= len;
1255
1256         if (cs->smultis == 0) {
1257                 free(cs->multis);
1258                 cs->multis = NULL;
1259                 return;
1260         }
1261
1262         cs->multis = realloc(cs->multis, cs->smultis);
1263         assert(cs->multis != NULL);
1264 }
1265
1266 /*
1267  - mcin - is a collating element in a cset?
1268  == static int mcin(register cset *cs, register char *cp);
1269  */
1270 static int
1271 mcin(register cset *cs, register char *cp)
1272 {
1273         return(mcfind(cs, cp) != NULL);
1274 }
1275
1276 /*
1277  - mcfind - find a collating element in a cset
1278  == static char *mcfind(register cset *cs, register char *cp);
1279  */
1280 static char *
1281 mcfind(register cset *cs, register char *cp)
1282 {
1283         register char *p;
1284
1285         if (cs->multis == NULL)
1286                 return(NULL);
1287         for (p = cs->multis; *p != '\0'; p += strlen(p) + 1)
1288                 if (strcmp(cp, p) == 0)
1289                         return(p);
1290         return(NULL);
1291 }
1292
1293 /*
1294  - mcinvert - invert the list of collating elements in a cset
1295  == static void mcinvert(register struct parse *p, register cset *cs);
1296  *
1297  * This would have to know the set of possibilities.  Implementation
1298  * is deferred.
1299  */
1300 static void
1301 mcinvert(register struct parse *p, register cset *cs)
1302 {
1303         assert(cs->multis == NULL);     /* xxx */
1304 }
1305
1306 /*
1307  - mccase - add case counterparts of the list of collating elements in a cset
1308  == static void mccase(register struct parse *p, register cset *cs);
1309  *
1310  * This would have to know the set of possibilities.  Implementation
1311  * is deferred.
1312  */
1313 static void
1314 mccase(register struct parse *p, register cset *cs)
1315 {
1316         assert(cs->multis == NULL);     /* xxx */
1317 }
1318
1319 /*
1320  - isinsets - is this character in any sets?
1321  == static int isinsets(register struct re_guts *g, int c);
1322  */
1323 static int                      /* predicate */
1324 isinsets(register struct re_guts *g, int c)
1325 {
1326         register uch *col;
1327         register int i;
1328         register int ncols = (g->ncsets+(CHAR_BIT-1)) / CHAR_BIT;
1329         register unsigned uc = (unsigned char)c;
1330
1331         for (i = 0, col = g->setbits; i < ncols; i++, col += g->csetsize)
1332                 if (col[uc] != 0)
1333                         return(1);
1334         return(0);
1335 }
1336
1337 /*
1338  - samesets - are these two characters in exactly the same sets?
1339  == static int samesets(register struct re_guts *g, int c1, int c2);
1340  */
1341 static int                      /* predicate */
1342 samesets(register struct re_guts *g, int c1, int c2)
1343 {
1344         register uch *col;
1345         register int i;
1346         register int ncols = (g->ncsets+(CHAR_BIT-1)) / CHAR_BIT;
1347         register unsigned uc1 = (unsigned char)c1;
1348         register unsigned uc2 = (unsigned char)c2;
1349
1350         for (i = 0, col = g->setbits; i < ncols; i++, col += g->csetsize)
1351                 if (col[uc1] != col[uc2])
1352                         return(0);
1353         return(1);
1354 }
1355
1356 /*
1357  - categorize - sort out character categories
1358  == static void categorize(struct parse *p, register struct re_guts *g);
1359  */
1360 static void
1361 categorize(struct parse *p, register struct re_guts *g)
1362 {
1363 /*
1364         register cat_t *cats = g->categories;
1365 */
1366         register int c;
1367         register int c2;
1368         register cat_t cat;
1369
1370         /* avoid making error situations worse */
1371         if (p->error != 0)
1372                 return;
1373
1374 /*
1375         for (c = CHAR_MIN; c <= CHAR_MAX; c++)
1376                 if (cats[c] == 0 && isinsets(g, c)) {
1377                         cat = g->ncategories++;
1378                         cats[c] = cat;
1379                         for (c2 = c+1; c2 <= CHAR_MAX; c2++)
1380                                 if (cats[c2] == 0 && samesets(g, c, c2))
1381                                         cats[c2] = cat;
1382                 }
1383 */
1384 }
1385
1386 /*
1387  - dupl - emit a duplicate of a bunch of sops
1388  == static sopno dupl(register struct parse *p, sopno start, sopno finish);
1389  */
1390 static sopno                    /* start of duplicate */
1391 dupl(register struct parse *p, sopno start, sopno finish)
1392
1393                                 /* from here */
1394                                 /* to this less one */
1395 {
1396         register sopno ret = HERE();
1397         register sopno len = finish - start;
1398
1399         assert(finish >= start);
1400         if (len == 0)
1401                 return(ret);
1402         enlarge(p, p->ssize + len);     /* this many unexpected additions */
1403         assert(p->ssize >= p->slen + len);
1404         (void) memcpy((char *)(p->strip + p->slen),
1405                 (char *)(p->strip + start), (size_t)len*sizeof(sop));
1406         p->slen += len;
1407         return(ret);
1408 }
1409
1410 /*
1411  - doemit - emit a strip operator
1412  == static void doemit(register struct parse *p, sop op, size_t opnd);
1413  *
1414  * It might seem better to implement this as a macro with a function as
1415  * hard-case backup, but it's just too big and messy unless there are
1416  * some changes to the data structures.  Maybe later.
1417  */
1418 static void
1419 doemit(register struct parse *p, sop op, size_t opnd)
1420 {
1421         /* avoid making error situations worse */
1422         if (p->error != 0)
1423                 return;
1424
1425         /* deal with oversize operands ("can't happen", more or less) */
1426         assert(opnd < 1<<OPSHIFT);
1427
1428         /* deal with undersized strip */
1429         if (p->slen >= p->ssize)
1430                 enlarge(p, (p->ssize+1) / 2 * 3);       /* +50% */
1431         assert(p->slen < p->ssize);
1432
1433         /* finally, it's all reduced to the easy case */
1434         p->strip[p->slen++] = SOP(op, opnd);
1435 }
1436
1437 /*
1438  - doinsert - insert a sop into the strip
1439  == static void doinsert(register struct parse *p, sop op, size_t opnd, sopno pos);
1440  */
1441 static void
1442 doinsert(register struct parse *p, sop op, size_t opnd, sopno pos)
1443 {
1444         register sopno sn;
1445         register sop s;
1446         register int i;
1447
1448         /* avoid making error situations worse */
1449         if (p->error != 0)
1450                 return;
1451
1452         sn = HERE();
1453         EMIT(op, opnd);         /* do checks, ensure space */
1454         assert(HERE() == sn+1);
1455         s = p->strip[sn];
1456
1457         /* adjust paren pointers */
1458         assert(pos > 0);
1459         for (i = 1; i < NPAREN; i++) {
1460                 if (p->pbegin[i] >= pos) {
1461                         p->pbegin[i]++;
1462                 }
1463                 if (p->pend[i] >= pos) {
1464                         p->pend[i]++;
1465                 }
1466         }
1467
1468         memmove((char *)&p->strip[pos+1], (char *)&p->strip[pos],
1469                                                 (HERE()-pos-1)*sizeof(sop));
1470         p->strip[pos] = s;
1471 }
1472
1473 /*
1474  - dofwd - complete a forward reference
1475  == static void dofwd(register struct parse *p, sopno pos, sop value);
1476  */
1477 static void
1478 dofwd(register struct parse *p, register sopno pos, sop value)
1479 {
1480         /* avoid making error situations worse */
1481         if (p->error != 0)
1482                 return;
1483
1484         assert(value < 1<<OPSHIFT);
1485         p->strip[pos] = OP(p->strip[pos]) | value;
1486 }
1487
1488 /*
1489  - enlarge - enlarge the strip
1490  == static void enlarge(register struct parse *p, sopno size);
1491  */
1492 static void
1493 enlarge(register struct parse *p, register sopno size)
1494 {
1495         register sop *sp;
1496
1497         if (p->ssize >= size)
1498                 return;
1499
1500         sp = (sop *)realloc(p->strip, size*sizeof(sop));
1501         if (sp == NULL) {
1502                 SETERROR(REG_ESPACE);
1503                 return;
1504         }
1505         p->strip = sp;
1506         p->ssize = size;
1507 }
1508
1509 /*
1510  - stripsnug - compact the strip
1511  == static void stripsnug(register struct parse *p, register struct re_guts *g);
1512  */
1513 static void
1514 stripsnug(register struct parse *p, register struct re_guts *g)
1515 {
1516         g->nstates = p->slen;
1517         g->strip = (sop *)realloc((char *)p->strip, p->slen * sizeof(sop));
1518         if (g->strip == NULL) {
1519                 SETERROR(REG_ESPACE);
1520                 g->strip = p->strip;
1521         }
1522 }
1523
1524 /*
1525  - findmust - fill in must and mlen with longest mandatory literal string
1526  == static void findmust(register struct parse *p, register struct re_guts *g);
1527  *
1528  * This algorithm could do fancy things like analyzing the operands of |
1529  * for common subsequences.  Someday.  This code is simple and finds most
1530  * of the interesting cases.
1531  *
1532  * Note that must and mlen got initialized during setup.
1533  */
1534 static void
1535 findmust(struct parse *p, register struct re_guts *g)
1536 {
1537         register sop *scan;
1538         sop *start;
1539         register sop *newstart;
1540         register sopno newlen;
1541         register sop s;
1542         register RCHAR_T *cp;
1543         register sopno i;
1544
1545         /* avoid making error situations worse */
1546         if (p->error != 0)
1547                 return;
1548
1549         /* find the longest OCHAR sequence in strip */
1550         newlen = 0;
1551         scan = g->strip + 1;
1552         do {
1553                 s = *scan++;
1554                 switch (OP(s)) {
1555                 case OCHAR:             /* sequence member */
1556                         if (newlen == 0)                /* new sequence */
1557                                 newstart = scan - 1;
1558                         newlen++;
1559                         break;
1560                 case OPLUS_:            /* things that don't break one */
1561                 case OLPAREN:
1562                 case ORPAREN:
1563                         break;
1564                 case OQUEST_:           /* things that must be skipped */
1565                 case OCH_:
1566                         scan--;
1567                         do {
1568                                 scan += OPND(s);
1569                                 s = *scan;
1570                                 /* assert() interferes w debug printouts */
1571                                 if (OP(s) != O_QUEST && OP(s) != O_CH &&
1572                                                         OP(s) != OOR2) {
1573                                         g->iflags |= BAD;
1574                                         return;
1575                                 }
1576                         } while (OP(s) != O_QUEST && OP(s) != O_CH);
1577                         /* fallthrough */
1578                 default:                /* things that break a sequence */
1579                         if (newlen > g->mlen) {         /* ends one */
1580                                 start = newstart;
1581                                 g->mlen = newlen;
1582                         }
1583                         newlen = 0;
1584                         break;
1585                 }
1586         } while (OP(s) != OEND);
1587
1588         if (g->mlen == 0)               /* there isn't one */
1589                 return;
1590
1591         /* turn it into a character string */
1592         g->must = malloc(((size_t)g->mlen + 1) * sizeof(RCHAR_T));
1593         if (g->must == NULL) {          /* argh; just forget it */
1594                 g->mlen = 0;
1595                 return;
1596         }
1597         cp = g->must;
1598         scan = start;
1599         for (i = g->mlen; i > 0; i--) {
1600                 while (OP(s = *scan++) != OCHAR)
1601                         continue;
1602                 assert(cp < g->must + g->mlen);
1603                 *cp++ = (RCHAR_T)OPND(s);
1604         }
1605         assert(cp == g->must + g->mlen);
1606         *cp++ = '\0';           /* just on general principles */
1607 }
1608
1609 /*
1610  - pluscount - count + nesting
1611  == static sopno pluscount(register struct parse *p, register struct re_guts *g);
1612  */
1613 static sopno                    /* nesting depth */
1614 pluscount(struct parse *p, register struct re_guts *g)
1615 {
1616         register sop *scan;
1617         register sop s;
1618         register sopno plusnest = 0;
1619         register sopno maxnest = 0;
1620
1621         if (p->error != 0)
1622                 return(0);      /* there may not be an OEND */
1623
1624         scan = g->strip + 1;
1625         do {
1626                 s = *scan++;
1627                 switch (OP(s)) {
1628                 case OPLUS_:
1629                         plusnest++;
1630                         break;
1631                 case O_PLUS:
1632                         if (plusnest > maxnest)
1633                                 maxnest = plusnest;
1634                         plusnest--;
1635                         break;
1636                 }
1637         } while (OP(s) != OEND);
1638         if (plusnest != 0)
1639                 g->iflags |= BAD;
1640         return(maxnest);
1641 }