lib/Support/regcomp.c

   1 /*-
   2  * This code is derived from OpenBSD's libc/regex, original license follows:
   3  *
   4  * Copyright (c) 1992, 1993, 1994 Henry Spencer.
   5  * Copyright (c) 1992, 1993, 1994
   6  *      The Regents of the University of California.  All rights reserved.
   7  *
   8  * This code is derived from software contributed to Berkeley by
   9  * Henry Spencer.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. Neither the name of the University nor the names of its contributors
  20  *    may be used to endorse or promote products derived from this software
  21  *    without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  *      @(#)regcomp.c   8.5 (Berkeley) 3/20/94
  36  */
  37
  38 #include <sys/types.h>
  39 #include <stdint.h>
  40 #include <stdio.h>
  41 #include <string.h>
  42 #include <ctype.h>
  43 #include <limits.h>
  44 #include <stdlib.h>
  45 #include "regex_impl.h"
  46
  47 #include "regutils.h"
  48 #include "regex2.h"
  49
  50 #include "llvm/Config/config.h"
  51
  52 /* character-class table */
  53 static struct cclass {
  54         const char *name;
  55         const char *chars;
  56         const char *multis;
  57 } cclasses[] = {
  58         { "alnum",      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz\
  59 0123456789",                            ""} ,
  60         { "alpha",      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz",
  61                                         ""} ,
  62         { "blank",      " \t",          ""} ,
  63         { "cntrl",      "\007\b\t\n\v\f\r\1\2\3\4\5\6\16\17\20\21\22\23\24\
  64 \25\26\27\30\31\32\33\34\35\36\37\177", ""} ,
  65         { "digit",      "0123456789",   ""} ,
  66         { "graph",      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz\
  67 0123456789!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~",
  68                                         ""} ,
  69         { "lower",      "abcdefghijklmnopqrstuvwxyz",
  70                                         ""} ,
  71         { "print",      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz\
  72 0123456789!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~ ",
  73                                         ""} ,
  74         { "punct",      "!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~",
  75                                         ""} ,
  76         { "space",      "\t\n\v\f\r ",  ""} ,
  77         { "upper",      "ABCDEFGHIJKLMNOPQRSTUVWXYZ",
  78                                         ""} ,
  79         { "xdigit",     "0123456789ABCDEFabcdef",
  80                                         ""} ,
  81         { NULL,         0,              "" }
  82 };
  83
  84 /* character-name table */
  85 static struct cname {
  86         const char *name;
  87         char code;
  88 } cnames[] = {
  89         { "NUL",                        '\0' },
  90         { "SOH",                        '\001' },
  91         { "STX",                        '\002' },
  92         { "ETX",                        '\003' },
  93         { "EOT",                        '\004' },
  94         { "ENQ",                        '\005' },
  95         { "ACK",                        '\006' },
  96         { "BEL",                        '\007' },
  97         { "alert",                      '\007' },
  98         { "BS",                         '\010' },
  99         { "backspace",                  '\b' },
 100         { "HT",                         '\011' },
 101         { "tab",                        '\t' },
 102         { "LF",                         '\012' },
 103         { "newline",                    '\n' },
 104         { "VT",                         '\013' },
 105         { "vertical-tab",               '\v' },
 106         { "FF",                         '\014' },
 107         { "form-feed",                  '\f' },
 108         { "CR",                         '\015' },
 109         { "carriage-return",            '\r' },
 110         { "SO",                         '\016' },
 111         { "SI",                         '\017' },
 112         { "DLE",                        '\020' },
 113         { "DC1",                        '\021' },
 114         { "DC2",                        '\022' },
 115         { "DC3",                        '\023' },
 116         { "DC4",                        '\024' },
 117         { "NAK",                        '\025' },
 118         { "SYN",                        '\026' },
 119         { "ETB",                        '\027' },
 120         { "CAN",                        '\030' },
 121         { "EM",                         '\031' },
 122         { "SUB",                        '\032' },
 123         { "ESC",                        '\033' },
 124         { "IS4",                        '\034' },
 125         { "FS",                         '\034' },
 126         { "IS3",                        '\035' },
 127         { "GS",                         '\035' },
 128         { "IS2",                        '\036' },
 129         { "RS",                         '\036' },
 130         { "IS1",                        '\037' },
 131         { "US",                         '\037' },
 132         { "space",                      ' ' },
 133         { "exclamation-mark",           '!' },
 134         { "quotation-mark",             '"' },
 135         { "number-sign",                '#' },
 136         { "dollar-sign",                '$' },
 137         { "percent-sign",               '%' },
 138         { "ampersand",                  '&' },
 139         { "apostrophe",                 '\'' },
 140         { "left-parenthesis",           '(' },
 141         { "right-parenthesis",          ')' },
 142         { "asterisk",                   '*' },
 143         { "plus-sign",                  '+' },
 144         { "comma",                      ',' },
 145         { "hyphen",                     '-' },
 146         { "hyphen-minus",               '-' },
 147         { "period",                     '.' },
 148         { "full-stop",                  '.' },
 149         { "slash",                      '/' },
 150         { "solidus",                    '/' },
 151         { "zero",                       '0' },
 152         { "one",                        '1' },
 153         { "two",                        '2' },
 154         { "three",                      '3' },
 155         { "four",                       '4' },
 156         { "five",                       '5' },
 157         { "six",                        '6' },
 158         { "seven",                      '7' },
 159         { "eight",                      '8' },
 160         { "nine",                       '9' },
 161         { "colon",                      ':' },
 162         { "semicolon",                  ';' },
 163         { "less-than-sign",             '<' },
 164         { "equals-sign",                '=' },
 165         { "greater-than-sign",          '>' },
 166         { "question-mark",              '?' },
 167         { "commercial-at",              '@' },
 168         { "left-square-bracket",        '[' },
 169         { "backslash",                  '\\' },
 170         { "reverse-solidus",            '\\' },
 171         { "right-square-bracket",       ']' },
 172         { "circumflex",                 '^' },
 173         { "circumflex-accent",          '^' },
 174         { "underscore",                 '_' },
 175         { "low-line",                   '_' },
 176         { "grave-accent",               '`' },
 177         { "left-brace",                 '{' },
 178         { "left-curly-bracket",         '{' },
 179         { "vertical-line",              '|' },
 180         { "right-brace",                '}' },
 181         { "right-curly-bracket",        '}' },
 182         { "tilde",                      '~' },
 183         { "DEL",                        '\177' },
 184         { NULL,                         0 }
 185 };
 186
 187 /*
 188  * parse structure, passed up and down to avoid global variables and
 189  * other clumsinesses
 190  */
 191 struct parse {
 192         char *next;             /* next character in RE */
 193         char *end;              /* end of string (-> NUL normally) */
 194         int error;              /* has an error been seen? */
 195         sop *strip;             /* malloced strip */
 196         sopno ssize;            /* malloced strip size (allocated) */
 197         sopno slen;             /* malloced strip length (used) */
 198         int ncsalloc;           /* number of csets allocated */
 199         struct re_guts *g;
 200 #       define  NPAREN  10      /* we need to remember () 1-9 for back refs */
 201         sopno pbegin[NPAREN];   /* -> ( ([0] unused) */
 202         sopno pend[NPAREN];     /* -> ) ([0] unused) */
 203 };
 204
 205 static void p_ere(struct parse *, int);
 206 static void p_ere_exp(struct parse *);
 207 static void p_str(struct parse *);
 208 static void p_bre(struct parse *, int, int);
 209 static int p_simp_re(struct parse *, int);
 210 static int p_count(struct parse *);
 211 static void p_bracket(struct parse *);
 212 static void p_b_term(struct parse *, cset *);
 213 static void p_b_cclass(struct parse *, cset *);
 214 static void p_b_eclass(struct parse *, cset *);
 215 static char p_b_symbol(struct parse *);
 216 static char p_b_coll_elem(struct parse *, int);
 217 static char othercase(int);
 218 static void bothcases(struct parse *, int);
 219 static void ordinary(struct parse *, int);
 220 static void nonnewline(struct parse *);
 221 static void repeat(struct parse *, sopno, int, int);
 222 static int seterr(struct parse *, int);
 223 static cset *allocset(struct parse *);
 224 static void freeset(struct parse *, cset *);
 225 static int freezeset(struct parse *, cset *);
 226 static int firstch(struct parse *, cset *);
 227 static int nch(struct parse *, cset *);
 228 static void mcadd(struct parse *, cset *, const char *);
 229 static void mcinvert(struct parse *, cset *);
 230 static void mccase(struct parse *, cset *);
 231 static int isinsets(struct re_guts *, int);
 232 static int samesets(struct re_guts *, int, int);
 233 static void categorize(struct parse *, struct re_guts *);
 234 static sopno dupl(struct parse *, sopno, sopno);
 235 static void doemit(struct parse *, sop, size_t);
 236 static void doinsert(struct parse *, sop, size_t, sopno);
 237 static void dofwd(struct parse *, sopno, sop);
 238 static void enlarge(struct parse *, sopno);
 239 static void stripsnug(struct parse *, struct re_guts *);
 240 static void findmust(struct parse *, struct re_guts *);
 241 static sopno pluscount(struct parse *, struct re_guts *);
 242
 243 static char nuls[10];           /* place to point scanner in event of error */
 244
 245 /*
 246  * macros for use with parse structure
 247  * BEWARE:  these know that the parse structure is named `p' !!!
 248  */
 249 #define PEEK()  (*p->next)
 250 #define PEEK2() (*(p->next+1))
 251 #define MORE()  (p->next < p->end)
 252 #define MORE2() (p->next+1 < p->end)
 253 #define SEE(c)  (MORE() && PEEK() == (c))
 254 #define SEETWO(a, b)    (MORE() && MORE2() && PEEK() == (a) && PEEK2() == (b))
 255 #define EAT(c)  ((SEE(c)) ? (NEXT(), 1) : 0)
 256 #define EATTWO(a, b)    ((SEETWO(a, b)) ? (NEXT2(), 1) : 0)
 257 #define NEXT()  (p->next++)
 258 #define NEXT2() (p->next += 2)
 259 #define NEXTn(n)        (p->next += (n))
 260 #define GETNEXT()       (*p->next++)
 261 #define SETERROR(e)     seterr(p, (e))
 262 #define REQUIRE(co, e)  (void)((co) || SETERROR(e))
 263 #define MUSTSEE(c, e)   (REQUIRE(MORE() && PEEK() == (c), e))
 264 #define MUSTEAT(c, e)   (REQUIRE(MORE() && GETNEXT() == (c), e))
 265 #define MUSTNOTSEE(c, e)        (REQUIRE(!MORE() || PEEK() != (c), e))
 266 #define EMIT(op, sopnd) doemit(p, (sop)(op), (size_t)(sopnd))
 267 #define INSERT(op, pos) doinsert(p, (sop)(op), HERE()-(pos)+1, pos)
 268 #define AHEAD(pos)              dofwd(p, pos, HERE()-(pos))
 269 #define ASTERN(sop, pos)        EMIT(sop, HERE()-pos)
 270 #define HERE()          (p->slen)
 271 #define THERE()         (p->slen - 1)
 272 #define THERETHERE()    (p->slen - 2)
 273 #define DROP(n) (p->slen -= (n))
 274
 275 #ifdef  _POSIX2_RE_DUP_MAX
 276 #define DUPMAX  _POSIX2_RE_DUP_MAX
 277 #else
 278 #define DUPMAX  255
 279 #endif
 280 #define INFINITY        (DUPMAX + 1)
 281
 282 #ifndef NDEBUG
 283 static int never = 0;           /* for use in asserts; shuts lint up */
 284 #else
 285 #define never   0               /* some <assert.h>s have bugs too */
 286 #endif
 287
 288 /*
 289  - llvm_regcomp - interface for parser and compilation
 290  */
 291 int                             /* 0 success, otherwise REG_something */
 292 llvm_regcomp(llvm_regex_t *preg, const char *pattern, int cflags)
 293 {
 294         struct parse pa;
 295         struct re_guts *g;
 296         struct parse *p = &pa;
 297         int i;
 298         size_t len;
 299 #ifdef REDEBUG
 300 #       define  GOODFLAGS(f)    (f)
 301 #else
 302 #       define  GOODFLAGS(f)    ((f)&~REG_DUMP)
 303 #endif
 304
 305         cflags = GOODFLAGS(cflags);
 306         if ((cflags&REG_EXTENDED) && (cflags&REG_NOSPEC))
 307                 return(REG_INVARG);
 308
 309         if (cflags&REG_PEND) {
 310                 if (preg->re_endp < pattern)
 311                         return(REG_INVARG);
 312                 len = preg->re_endp - pattern;
 313         } else
 314                 len = strlen((const char *)pattern);
 315
 316         /* do the mallocs early so failure handling is easy */
 317         g = (struct re_guts *)malloc(sizeof(struct re_guts) +
 318                                                         (NC-1)*sizeof(cat_t));
 319         if (g == NULL)
 320                 return(REG_ESPACE);
 321         p->ssize = len/(size_t)2*(size_t)3 + (size_t)1; /* ugh */
 322         p->strip = (sop *)calloc(p->ssize, sizeof(sop));
 323         p->slen = 0;
 324         if (p->strip == NULL) {
 325                 free((char *)g);
 326                 return(REG_ESPACE);
 327         }
 328
 329         /* set things up */
 330         p->g = g;
 331         p->next = (char *)pattern;      /* convenience; we do not modify it */
 332         p->end = p->next + len;
 333         p->error = 0;
 334         p->ncsalloc = 0;
 335         for (i = 0; i < NPAREN; i++) {
 336                 p->pbegin[i] = 0;
 337                 p->pend[i] = 0;
 338         }
 339         g->csetsize = NC;
 340         g->sets = NULL;
 341         g->setbits = NULL;
 342         g->ncsets = 0;
 343         g->cflags = cflags;
 344         g->iflags = 0;
 345         g->nbol = 0;
 346         g->neol = 0;
 347         g->must = NULL;
 348         g->mlen = 0;
 349         g->nsub = 0;
 350         g->ncategories = 1;     /* category 0 is "everything else" */
 351         g->categories = &g->catspace[-(CHAR_MIN)];
 352         (void) memset((char *)g->catspace, 0, NC*sizeof(cat_t));
 353         g->backrefs = 0;
 354
 355         /* do it */
 356         EMIT(OEND, 0);
 357         g->firststate = THERE();
 358         if (cflags&REG_EXTENDED)
 359                 p_ere(p, OUT);
 360         else if (cflags&REG_NOSPEC)
 361                 p_str(p);
 362         else
 363                 p_bre(p, OUT, OUT);
 364         EMIT(OEND, 0);
 365         g->laststate = THERE();
 366
 367         /* tidy up loose ends and fill things in */
 368         categorize(p, g);
 369         stripsnug(p, g);
 370         findmust(p, g);
 371         g->nplus = pluscount(p, g);
 372         g->magic = MAGIC2;
 373         preg->re_nsub = g->nsub;
 374         preg->re_g = g;
 375         preg->re_magic = MAGIC1;
 376 #ifndef REDEBUG
 377         /* not debugging, so can't rely on the assert() in llvm_regexec() */
 378         if (g->iflags&REGEX_BAD)
 379                 SETERROR(REG_ASSERT);
 380 #endif
 381
 382         /* win or lose, we're done */
 383         if (p->error != 0)      /* lose */
 384                 llvm_regfree(preg);
 385         return(p->error);
 386 }
 387
 388 /*
 389  - p_ere - ERE parser top level, concatenation and alternation
 390  */
 391 static void
 392 p_ere(struct parse *p, int stop)        /* character this ERE should end at */
 393 {
 394         char c;
 395         sopno prevback = 0;
 396         sopno prevfwd = 0;
 397         sopno conc;
 398         int first = 1;          /* is this the first alternative? */
 399
 400         for (;;) {
 401                 /* do a bunch of concatenated expressions */
 402                 conc = HERE();
 403                 while (MORE() && (c = PEEK()) != '|' && c != stop)
 404                         p_ere_exp(p);
 405                 REQUIRE(HERE() != conc, REG_EMPTY);     /* require nonempty */
 406
 407                 if (!EAT('|'))
 408                         break;          /* NOTE BREAK OUT */
 409
 410                 if (first) {
 411                         INSERT(OCH_, conc);     /* offset is wrong */
 412                         prevfwd = conc;
 413                         prevback = conc;
 414                         first = 0;
 415                 }
 416                 ASTERN(OOR1, prevback);
 417                 prevback = THERE();
 418                 AHEAD(prevfwd);                 /* fix previous offset */
 419                 prevfwd = HERE();
 420                 EMIT(OOR2, 0);                  /* offset is very wrong */
 421         }
 422
 423         if (!first) {           /* tail-end fixups */
 424                 AHEAD(prevfwd);
 425                 ASTERN(O_CH, prevback);
 426         }
 427
 428         assert(!MORE() || SEE(stop));
 429 }
 430
 431 /*
 432  - p_ere_exp - parse one subERE, an atom possibly followed by a repetition op
 433  */
 434 static void
 435 p_ere_exp(struct parse *p)
 436 {
 437         char c;
 438         sopno pos;
 439         int count;
 440         int count2;
 441         int backrefnum;
 442         sopno subno;
 443         int wascaret = 0;
 444
 445         assert(MORE());         /* caller should have ensured this */
 446         c = GETNEXT();
 447
 448         pos = HERE();
 449         switch (c) {
 450         case '(':
 451                 REQUIRE(MORE(), REG_EPAREN);
 452                 p->g->nsub++;
 453                 subno = p->g->nsub;
 454                 if (subno < NPAREN)
 455                         p->pbegin[subno] = HERE();
 456                 EMIT(OLPAREN, subno);
 457                 if (!SEE(')'))
 458                         p_ere(p, ')');
 459                 if (subno < NPAREN) {
 460                         p->pend[subno] = HERE();
 461                         assert(p->pend[subno] != 0);
 462                 }
 463                 EMIT(ORPAREN, subno);
 464                 MUSTEAT(')', REG_EPAREN);
 465                 break;
 466 #ifndef POSIX_MISTAKE
 467         case ')':               /* happens only if no current unmatched ( */
 468                 /*
 469                  * You may ask, why the ifndef?  Because I didn't notice
 470                  * this until slightly too late for 1003.2, and none of the
 471                  * other 1003.2 regular-expression reviewers noticed it at
 472                  * all.  So an unmatched ) is legal POSIX, at least until
 473                  * we can get it fixed.
 474                  */
 475                 SETERROR(REG_EPAREN);
 476                 break;
 477 #endif
 478         case '^':
 479                 EMIT(OBOL, 0);
 480                 p->g->iflags |= USEBOL;
 481                 p->g->nbol++;
 482                 wascaret = 1;
 483                 break;
 484         case '$':
 485                 EMIT(OEOL, 0);
 486                 p->g->iflags |= USEEOL;
 487                 p->g->neol++;
 488                 break;
 489         case '|':
 490                 SETERROR(REG_EMPTY);
 491                 break;
 492         case '*':
 493         case '+':
 494         case '?':
 495                 SETERROR(REG_BADRPT);
 496                 break;
 497         case '.':
 498                 if (p->g->cflags&REG_NEWLINE)
 499                         nonnewline(p);
 500                 else
 501                         EMIT(OANY, 0);
 502                 break;
 503         case '[':
 504                 p_bracket(p);
 505                 break;
 506         case '\\':
 507                 REQUIRE(MORE(), REG_EESCAPE);
 508                 c = GETNEXT();
 509                 if (c >= '1' && c <= '9') {
 510                         /* \[0-9] is taken to be a back-reference to a previously specified
 511                          * matching group. backrefnum will hold the number. The matching
 512                          * group must exist (i.e. if \4 is found there must have been at
 513                          * least 4 matching groups specified in the pattern previously).
 514                          */
 515                         backrefnum = c - '0';
 516                         if (p->pend[backrefnum] == 0) {
 517                                 SETERROR(REG_ESUBREG);
 518                                 break;
 519                         }
 520
 521                         /* Make sure everything checks out and emit the sequence
 522                          * that marks a back-reference to the parse structure.
 523                          */
 524                         assert(backrefnum <= p->g->nsub);
 525                         EMIT(OBACK_, backrefnum);
 526                         assert(p->pbegin[backrefnum] != 0);
 527                         assert(OP(p->strip[p->pbegin[backrefnum]]) != OLPAREN);
 528                         assert(OP(p->strip[p->pend[backrefnum]]) != ORPAREN);
 529                         (void) dupl(p, p->pbegin[backrefnum]+1, p->pend[backrefnum]);
 530                         EMIT(O_BACK, backrefnum);
 531                         p->g->backrefs = 1;
 532                 } else {
 533                         /* Other chars are simply themselves when escaped with a backslash.
 534                          */
 535                         ordinary(p, c);
 536                 }
 537                 break;
 538         case '{':               /* okay as ordinary except if digit follows */
 539                 REQUIRE(!MORE() || !isdigit((uch)PEEK()), REG_BADRPT);
 540                 /* FALLTHROUGH */
 541         default:
 542                 ordinary(p, c);
 543                 break;
 544         }
 545
 546         if (!MORE())
 547                 return;
 548         c = PEEK();
 549         /* we call { a repetition if followed by a digit */
 550         if (!( c == '*' || c == '+' || c == '?' ||
 551                                 (c == '{' && MORE2() && isdigit((uch)PEEK2())) ))
 552                 return;         /* no repetition, we're done */
 553         NEXT();
 554
 555         REQUIRE(!wascaret, REG_BADRPT);
 556         switch (c) {
 557         case '*':       /* implemented as +? */
 558                 /* this case does not require the (y|) trick, noKLUDGE */
 559                 INSERT(OPLUS_, pos);
 560                 ASTERN(O_PLUS, pos);
 561                 INSERT(OQUEST_, pos);
 562                 ASTERN(O_QUEST, pos);
 563                 break;
 564         case '+':
 565                 INSERT(OPLUS_, pos);
 566                 ASTERN(O_PLUS, pos);
 567                 break;
 568         case '?':
 569                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
 570                 INSERT(OCH_, pos);              /* offset slightly wrong */
 571                 ASTERN(OOR1, pos);              /* this one's right */
 572                 AHEAD(pos);                     /* fix the OCH_ */
 573                 EMIT(OOR2, 0);                  /* offset very wrong... */
 574                 AHEAD(THERE());                 /* ...so fix it */
 575                 ASTERN(O_CH, THERETHERE());
 576                 break;
 577         case '{':
 578                 count = p_count(p);
 579                 if (EAT(',')) {
 580                         if (isdigit((uch)PEEK())) {
 581                                 count2 = p_count(p);
 582                                 REQUIRE(count <= count2, REG_BADBR);
 583                         } else          /* single number with comma */
 584                                 count2 = INFINITY;
 585                 } else          /* just a single number */
 586                         count2 = count;
 587                 repeat(p, pos, count, count2);
 588                 if (!EAT('}')) {        /* error heuristics */
 589                         while (MORE() && PEEK() != '}')
 590                                 NEXT();
 591                         REQUIRE(MORE(), REG_EBRACE);
 592                         SETERROR(REG_BADBR);
 593                 }
 594                 break;
 595         }
 596
 597         if (!MORE())
 598                 return;
 599         c = PEEK();
 600         if (!( c == '*' || c == '+' || c == '?' ||
 601                                 (c == '{' && MORE2() && isdigit((uch)PEEK2())) ) )
 602                 return;
 603         SETERROR(REG_BADRPT);
 604 }
 605
 606 /*
 607  - p_str - string (no metacharacters) "parser"
 608  */
 609 static void
 610 p_str(struct parse *p)
 611 {
 612         REQUIRE(MORE(), REG_EMPTY);
 613         while (MORE())
 614                 ordinary(p, GETNEXT());
 615 }
 616
 617 /*
 618  - p_bre - BRE parser top level, anchoring and concatenation
 619  * Giving end1 as OUT essentially eliminates the end1/end2 check.
 620  *
 621  * This implementation is a bit of a kludge, in that a trailing $ is first
 622  * taken as an ordinary character and then revised to be an anchor.  The
 623  * only undesirable side effect is that '$' gets included as a character
 624  * category in such cases.  This is fairly harmless; not worth fixing.
 625  * The amount of lookahead needed to avoid this kludge is excessive.
 626  */
 627 static void
 628 p_bre(struct parse *p,
 629     int end1,           /* first terminating character */
 630     int end2)           /* second terminating character */
 631 {
 632         sopno start = HERE();
 633         int first = 1;                  /* first subexpression? */
 634         int wasdollar = 0;
 635
 636         if (EAT('^')) {
 637                 EMIT(OBOL, 0);
 638                 p->g->iflags |= USEBOL;
 639                 p->g->nbol++;
 640         }
 641         while (MORE() && !SEETWO(end1, end2)) {
 642                 wasdollar = p_simp_re(p, first);
 643                 first = 0;
 644         }
 645         if (wasdollar) {        /* oops, that was a trailing anchor */
 646                 DROP(1);
 647                 EMIT(OEOL, 0);
 648                 p->g->iflags |= USEEOL;
 649                 p->g->neol++;
 650         }
 651
 652         REQUIRE(HERE() != start, REG_EMPTY);    /* require nonempty */
 653 }
 654
 655 /*
 656  - p_simp_re - parse a simple RE, an atom possibly followed by a repetition
 657  */
 658 static int                      /* was the simple RE an unbackslashed $? */
 659 p_simp_re(struct parse *p,
 660     int starordinary)           /* is a leading * an ordinary character? */
 661 {
 662         int c;
 663         int count;
 664         int count2;
 665         sopno pos;
 666         int i;
 667         sopno subno;
 668 #       define  BACKSL  (1<<CHAR_BIT)
 669
 670         pos = HERE(); /* repetition op, if any, covers from here */
 671
 672         assert(MORE()); /* caller should have ensured this */
 673         c = GETNEXT();
 674         if (c == '\\') {
 675                 REQUIRE(MORE(), REG_EESCAPE);
 676                 c = BACKSL | GETNEXT();
 677         }
 678         switch (c) {
 679         case '.':
 680                 if (p->g->cflags&REG_NEWLINE)
 681                         nonnewline(p);
 682                 else
 683                         EMIT(OANY, 0);
 684                 break;
 685         case '[':
 686                 p_bracket(p);
 687                 break;
 688         case BACKSL|'{':
 689                 SETERROR(REG_BADRPT);
 690                 break;
 691         case BACKSL|'(':
 692                 p->g->nsub++;
 693                 subno = p->g->nsub;
 694                 if (subno < NPAREN)
 695                         p->pbegin[subno] = HERE();
 696                 EMIT(OLPAREN, subno);
 697                 /* the MORE here is an error heuristic */
 698                 if (MORE() && !SEETWO('\\', ')'))
 699                         p_bre(p, '\\', ')');
 700                 if (subno < NPAREN) {
 701                         p->pend[subno] = HERE();
 702                         assert(p->pend[subno] != 0);
 703                 }
 704                 EMIT(ORPAREN, subno);
 705                 REQUIRE(EATTWO('\\', ')'), REG_EPAREN);
 706                 break;
 707         case BACKSL|')':        /* should not get here -- must be user */
 708         case BACKSL|'}':
 709                 SETERROR(REG_EPAREN);
 710                 break;
 711         case BACKSL|'1':
 712         case BACKSL|'2':
 713         case BACKSL|'3':
 714         case BACKSL|'4':
 715         case BACKSL|'5':
 716         case BACKSL|'6':
 717         case BACKSL|'7':
 718         case BACKSL|'8':
 719         case BACKSL|'9':
 720                 i = (c&~BACKSL) - '0';
 721                 assert(i < NPAREN);
 722                 if (p->pend[i] != 0) {
 723                         assert(i <= p->g->nsub);
 724                         EMIT(OBACK_, i);
 725                         assert(p->pbegin[i] != 0);
 726                         assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
 727                         assert(OP(p->strip[p->pend[i]]) == ORPAREN);
 728                         (void) dupl(p, p->pbegin[i]+1, p->pend[i]);
 729                         EMIT(O_BACK, i);
 730                 } else
 731                         SETERROR(REG_ESUBREG);
 732                 p->g->backrefs = 1;
 733                 break;
 734         case '*':
 735                 REQUIRE(starordinary, REG_BADRPT);
 736                 /* FALLTHROUGH */
 737         default:
 738                 ordinary(p, (char)c);
 739                 break;
 740         }
 741
 742         if (EAT('*')) {         /* implemented as +? */
 743                 /* this case does not require the (y|) trick, noKLUDGE */
 744                 INSERT(OPLUS_, pos);
 745                 ASTERN(O_PLUS, pos);
 746                 INSERT(OQUEST_, pos);
 747                 ASTERN(O_QUEST, pos);
 748         } else if (EATTWO('\\', '{')) {
 749                 count = p_count(p);
 750                 if (EAT(',')) {
 751                         if (MORE() && isdigit((uch)PEEK())) {
 752                                 count2 = p_count(p);
 753                                 REQUIRE(count <= count2, REG_BADBR);
 754                         } else          /* single number with comma */
 755                                 count2 = INFINITY;
 756                 } else          /* just a single number */
 757                         count2 = count;
 758                 repeat(p, pos, count, count2);
 759                 if (!EATTWO('\\', '}')) {       /* error heuristics */
 760                         while (MORE() && !SEETWO('\\', '}'))
 761                                 NEXT();
 762                         REQUIRE(MORE(), REG_EBRACE);
 763                         SETERROR(REG_BADBR);
 764                 }
 765         } else if (c == '$')    /* $ (but not \$) ends it */
 766                 return(1);
 767
 768         return(0);
 769 }
 770
 771 /*
 772  - p_count - parse a repetition count
 773  */
 774 static int                      /* the value */
 775 p_count(struct parse *p)
 776 {
 777         int count = 0;
 778         int ndigits = 0;
 779
 780         while (MORE() && isdigit((uch)PEEK()) && count <= DUPMAX) {
 781                 count = count*10 + (GETNEXT() - '0');
 782                 ndigits++;
 783         }
 784
 785         REQUIRE(ndigits > 0 && count <= DUPMAX, REG_BADBR);
 786         return(count);
 787 }
 788
 789 /*
 790  - p_bracket - parse a bracketed character list
 791  *
 792  * Note a significant property of this code:  if the allocset() did SETERROR,
 793  * no set operations are done.
 794  */
 795 static void
 796 p_bracket(struct parse *p)
 797 {
 798         cset *cs;
 799         int invert = 0;
 800
 801         /* Dept of Truly Sickening Special-Case Kludges */
 802         if (p->next + 5 < p->end && strncmp(p->next, "[:<:]]", 6) == 0) {
 803                 EMIT(OBOW, 0);
 804                 NEXTn(6);
 805                 return;
 806         }
 807         if (p->next + 5 < p->end && strncmp(p->next, "[:>:]]", 6) == 0) {
 808                 EMIT(OEOW, 0);
 809                 NEXTn(6);
 810                 return;
 811         }
 812
 813         if ((cs = allocset(p)) == NULL) {
 814                 /* allocset did set error status in p */
 815                 return;
 816         }
 817
 818         if (EAT('^'))
 819                 invert++;       /* make note to invert set at end */
 820         if (EAT(']'))
 821                 CHadd(cs, ']');
 822         else if (EAT('-'))
 823                 CHadd(cs, '-');
 824         while (MORE() && PEEK() != ']' && !SEETWO('-', ']'))
 825                 p_b_term(p, cs);
 826         if (EAT('-'))
 827                 CHadd(cs, '-');
 828         MUSTEAT(']', REG_EBRACK);
 829
 830         if (p->error != 0) {    /* don't mess things up further */
 831                 freeset(p, cs);
 832                 return;
 833         }
 834
 835         if (p->g->cflags&REG_ICASE) {
 836                 int i;
 837                 int ci;
 838
 839                 for (i = p->g->csetsize - 1; i >= 0; i--)
 840                         if (CHIN(cs, i) && isalpha(i)) {
 841                                 ci = othercase(i);
 842                                 if (ci != i)
 843                                         CHadd(cs, ci);
 844                         }
 845                 if (cs->multis != NULL)
 846                         mccase(p, cs);
 847         }
 848         if (invert) {
 849                 int i;
 850
 851                 for (i = p->g->csetsize - 1; i >= 0; i--)
 852                         if (CHIN(cs, i))
 853                                 CHsub(cs, i);
 854                         else
 855                                 CHadd(cs, i);
 856                 if (p->g->cflags&REG_NEWLINE)
 857                         CHsub(cs, '\n');
 858                 if (cs->multis != NULL)
 859                         mcinvert(p, cs);
 860         }
 861
 862         assert(cs->multis == NULL);             /* xxx */
 863
 864         if (nch(p, cs) == 1) {          /* optimize singleton sets */
 865                 ordinary(p, firstch(p, cs));
 866                 freeset(p, cs);
 867         } else
 868                 EMIT(OANYOF, freezeset(p, cs));
 869 }
 870
 871 /*
 872  - p_b_term - parse one term of a bracketed character list
 873  */
 874 static void
 875 p_b_term(struct parse *p, cset *cs)
 876 {
 877         char c;
 878         char start, finish;
 879         int i;
 880
 881         /* classify what we've got */
 882         switch ((MORE()) ? PEEK() : '\0') {
 883         case '[':
 884                 c = (MORE2()) ? PEEK2() : '\0';
 885                 break;
 886         case '-':
 887                 SETERROR(REG_ERANGE);
 888                 return;                 /* NOTE RETURN */
 889                 break;
 890         default:
 891                 c = '\0';
 892                 break;
 893         }
 894
 895         switch (c) {
 896         case ':':               /* character class */
 897                 NEXT2();
 898                 REQUIRE(MORE(), REG_EBRACK);
 899                 c = PEEK();
 900                 REQUIRE(c != '-' && c != ']', REG_ECTYPE);
 901                 p_b_cclass(p, cs);
 902                 REQUIRE(MORE(), REG_EBRACK);
 903                 REQUIRE(EATTWO(':', ']'), REG_ECTYPE);
 904                 break;
 905         case '=':               /* equivalence class */
 906                 NEXT2();
 907                 REQUIRE(MORE(), REG_EBRACK);
 908                 c = PEEK();
 909                 REQUIRE(c != '-' && c != ']', REG_ECOLLATE);
 910                 p_b_eclass(p, cs);
 911                 REQUIRE(MORE(), REG_EBRACK);
 912                 REQUIRE(EATTWO('=', ']'), REG_ECOLLATE);
 913                 break;
 914         default:                /* symbol, ordinary character, or range */
 915 /* xxx revision needed for multichar stuff */
 916                 start = p_b_symbol(p);
 917                 if (SEE('-') && MORE2() && PEEK2() != ']') {
 918                         /* range */
 919                         NEXT();
 920                         if (EAT('-'))
 921                                 finish = '-';
 922                         else
 923                                 finish = p_b_symbol(p);
 924                 } else
 925                         finish = start;
 926 /* xxx what about signed chars here... */
 927                 REQUIRE(start <= finish, REG_ERANGE);
 928                 for (i = start; i <= finish; i++)
 929                         CHadd(cs, i);
 930                 break;
 931         }
 932 }
 933
 934 /*
 935  - p_b_cclass - parse a character-class name and deal with it
 936  */
 937 static void
 938 p_b_cclass(struct parse *p, cset *cs)
 939 {
 940         char *sp = p->next;
 941         struct cclass *cp;
 942         size_t len;
 943         const char *u;
 944         char c;
 945
 946         while (MORE() && isalpha((uch)PEEK()))
 947                 NEXT();
 948         len = p->next - sp;
 949         for (cp = cclasses; cp->name != NULL; cp++)
 950                 if (strncmp(cp->name, sp, len) == 0 && cp->name[len] == '\0')
 951                         break;
 952         if (cp->name == NULL) {
 953                 /* oops, didn't find it */
 954                 SETERROR(REG_ECTYPE);
 955                 return;
 956         }
 957
 958         u = cp->chars;
 959         while ((c = *u++) != '\0')
 960                 CHadd(cs, c);
 961         for (u = cp->multis; *u != '\0'; u += strlen(u) + 1)
 962                 MCadd(p, cs, u);
 963 }
 964
 965 /*
 966  - p_b_eclass - parse an equivalence-class name and deal with it
 967  *
 968  * This implementation is incomplete. xxx
 969  */
 970 static void
 971 p_b_eclass(struct parse *p, cset *cs)
 972 {
 973         char c;
 974
 975         c = p_b_coll_elem(p, '=');
 976         CHadd(cs, c);
 977 }
 978
 979 /*
 980  - p_b_symbol - parse a character or [..]ed multicharacter collating symbol
 981  */
 982 static char                     /* value of symbol */
 983 p_b_symbol(struct parse *p)
 984 {
 985         char value;
 986
 987         REQUIRE(MORE(), REG_EBRACK);
 988         if (!EATTWO('[', '.'))
 989                 return(GETNEXT());
 990
 991         /* collating symbol */
 992         value = p_b_coll_elem(p, '.');
 993         REQUIRE(EATTWO('.', ']'), REG_ECOLLATE);
 994         return(value);
 995 }
 996
 997 /*
 998  - p_b_coll_elem - parse a collating-element name and look it up
 999  */
1000 static char                     /* value of collating element */
1001 p_b_coll_elem(struct parse *p,
1002     int endc)                   /* name ended by endc,']' */
1003 {
1004         char *sp = p->next;
1005         struct cname *cp;
1006         size_t len;
1007
1008         while (MORE() && !SEETWO(endc, ']'))
1009                 NEXT();
1010         if (!MORE()) {
1011                 SETERROR(REG_EBRACK);
1012                 return(0);
1013         }
1014         len = p->next - sp;
1015         for (cp = cnames; cp->name != NULL; cp++)
1016                 if (strncmp(cp->name, sp, len) == 0 && strlen(cp->name) == len)
1017                         return(cp->code);       /* known name */
1018         if (len == 1)
1019                 return(*sp);    /* single character */
1020         SETERROR(REG_ECOLLATE);                 /* neither */
1021         return(0);
1022 }
1023
1024 /*
1025  - othercase - return the case counterpart of an alphabetic
1026  */
1027 static char                     /* if no counterpart, return ch */
1028 othercase(int ch)
1029 {
1030         ch = (uch)ch;
1031         assert(isalpha(ch));
1032         if (isupper(ch))
1033                 return ((uch)tolower(ch));
1034         else if (islower(ch))
1035                 return ((uch)toupper(ch));
1036         else                    /* peculiar, but could happen */
1037                 return(ch);
1038 }
1039
1040 /*
1041  - bothcases - emit a dualcase version of a two-case character
1042  *
1043  * Boy, is this implementation ever a kludge...
1044  */
1045 static void
1046 bothcases(struct parse *p, int ch)
1047 {
1048         char *oldnext = p->next;
1049         char *oldend = p->end;
1050         char bracket[3];
1051
1052         ch = (uch)ch;
1053         assert(othercase(ch) != ch);    /* p_bracket() would recurse */
1054         p->next = bracket;
1055         p->end = bracket+2;
1056         bracket[0] = ch;
1057         bracket[1] = ']';
1058         bracket[2] = '\0';
1059         p_bracket(p);
1060         assert(p->next == bracket+2);
1061         p->next = oldnext;
1062         p->end = oldend;
1063 }
1064
1065 /*
1066  - ordinary - emit an ordinary character
1067  */
1068 static void
1069 ordinary(struct parse *p, int ch)
1070 {
1071         cat_t *cap = p->g->categories;
1072
1073         if ((p->g->cflags&REG_ICASE) && isalpha((uch)ch) && othercase(ch) != ch)
1074                 bothcases(p, ch);
1075         else {
1076                 EMIT(OCHAR, (uch)ch);
1077                 if (cap[ch] == 0)
1078                         cap[ch] = p->g->ncategories++;
1079         }
1080 }
1081
1082 /*
1083  - nonnewline - emit REG_NEWLINE version of OANY
1084  *
1085  * Boy, is this implementation ever a kludge...
1086  */
1087 static void
1088 nonnewline(struct parse *p)
1089 {
1090         char *oldnext = p->next;
1091         char *oldend = p->end;
1092         char bracket[4];
1093
1094         p->next = bracket;
1095         p->end = bracket+3;
1096         bracket[0] = '^';
1097         bracket[1] = '\n';
1098         bracket[2] = ']';
1099         bracket[3] = '\0';
1100         p_bracket(p);
1101         assert(p->next == bracket+3);
1102         p->next = oldnext;
1103         p->end = oldend;
1104 }
1105
1106 /*
1107  - repeat - generate code for a bounded repetition, recursively if needed
1108  */
1109 static void
1110 repeat(struct parse *p,
1111     sopno start,                /* operand from here to end of strip */
1112     int from,                   /* repeated from this number */
1113     int to)                     /* to this number of times (maybe INFINITY) */
1114 {
1115         sopno finish = HERE();
1116 #       define  N       2
1117 #       define  INF     3
1118 #       define  REP(f, t)       ((f)*8 + (t))
1119 #       define  MAP(n)  (((n) <= 1) ? (n) : ((n) == INFINITY) ? INF : N)
1120         sopno copy;
1121
1122         if (p->error != 0)      /* head off possible runaway recursion */
1123                 return;
1124
1125         assert(from <= to);
1126
1127         switch (REP(MAP(from), MAP(to))) {
1128         case REP(0, 0):                 /* must be user doing this */
1129                 DROP(finish-start);     /* drop the operand */
1130                 break;
1131         case REP(0, 1):                 /* as x{1,1}? */
1132         case REP(0, N):                 /* as x{1,n}? */
1133         case REP(0, INF):               /* as x{1,}? */
1134                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1135                 INSERT(OCH_, start);            /* offset is wrong... */
1136                 repeat(p, start+1, 1, to);
1137                 ASTERN(OOR1, start);
1138                 AHEAD(start);                   /* ... fix it */
1139                 EMIT(OOR2, 0);
1140                 AHEAD(THERE());
1141                 ASTERN(O_CH, THERETHERE());
1142                 break;
1143         case REP(1, 1):                 /* trivial case */
1144                 /* done */
1145                 break;
1146         case REP(1, N):                 /* as x?x{1,n-1} */
1147                 /* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1148                 INSERT(OCH_, start);
1149                 ASTERN(OOR1, start);
1150                 AHEAD(start);
1151                 EMIT(OOR2, 0);                  /* offset very wrong... */
1152                 AHEAD(THERE());                 /* ...so fix it */
1153                 ASTERN(O_CH, THERETHERE());
1154                 copy = dupl(p, start+1, finish+1);
1155                 assert(copy == finish+4);
1156                 repeat(p, copy, 1, to-1);
1157                 break;
1158         case REP(1, INF):               /* as x+ */
1159                 INSERT(OPLUS_, start);
1160                 ASTERN(O_PLUS, start);
1161                 break;
1162         case REP(N, N):                 /* as xx{m-1,n-1} */
1163                 copy = dupl(p, start, finish);
1164                 repeat(p, copy, from-1, to-1);
1165                 break;
1166         case REP(N, INF):               /* as xx{n-1,INF} */
1167                 copy = dupl(p, start, finish);
1168                 repeat(p, copy, from-1, to);
1169                 break;
1170         default:                        /* "can't happen" */
1171                 SETERROR(REG_ASSERT);   /* just in case */
1172                 break;
1173         }
1174 }
1175
1176 /*
1177  - seterr - set an error condition
1178  */
1179 static int                      /* useless but makes type checking happy */
1180 seterr(struct parse *p, int e)
1181 {
1182         if (p->error == 0)      /* keep earliest error condition */
1183                 p->error = e;
1184         p->next = nuls;         /* try to bring things to a halt */
1185         p->end = nuls;
1186         return(0);              /* make the return value well-defined */
1187 }
1188
1189 /*
1190  - allocset - allocate a set of characters for []
1191  */
1192 static cset *
1193 allocset(struct parse *p)
1194 {
1195         int no = p->g->ncsets++;
1196         size_t nc;
1197         size_t nbytes;
1198         cset *cs;
1199         size_t css = (size_t)p->g->csetsize;
1200         int i;
1201
1202         if (no >= p->ncsalloc) {        /* need another column of space */
1203                 void *ptr;
1204
1205                 p->ncsalloc += CHAR_BIT;
1206                 nc = p->ncsalloc;
1207                 if (nc > SIZE_MAX / sizeof(cset))
1208                         goto nomem;
1209                 assert(nc % CHAR_BIT == 0);
1210                 nbytes = nc / CHAR_BIT * css;
1211
1212                 ptr = (cset *)realloc((char *)p->g->sets, nc * sizeof(cset));
1213                 if (ptr == NULL)
1214                         goto nomem;
1215                 p->g->sets = ptr;
1216
1217                 ptr = (uch *)realloc((char *)p->g->setbits, nbytes);
1218                 if (ptr == NULL)
1219                         goto nomem;
1220                 p->g->setbits = ptr;
1221
1222                 for (i = 0; i < no; i++)
1223                         p->g->sets[i].ptr = p->g->setbits + css*(i/CHAR_BIT);
1224
1225                 (void) memset((char *)p->g->setbits + (nbytes - css), 0, css);
1226         }
1227         /* XXX should not happen */
1228         if (p->g->sets == NULL || p->g->setbits == NULL)
1229                 goto nomem;
1230
1231         cs = &p->g->sets[no];
1232         cs->ptr = p->g->setbits + css*((no)/CHAR_BIT);
1233         cs->mask = 1 << ((no) % CHAR_BIT);
1234         cs->hash = 0;
1235         cs->smultis = 0;
1236         cs->multis = NULL;
1237
1238         return(cs);
1239 nomem:
1240         free(p->g->sets);
1241         p->g->sets = NULL;
1242         free(p->g->setbits);
1243         p->g->setbits = NULL;
1244
1245         SETERROR(REG_ESPACE);
1246         /* caller's responsibility not to do set ops */
1247         return(NULL);
1248 }
1249
1250 /*
1251  - freeset - free a now-unused set
1252  */
1253 static void
1254 freeset(struct parse *p, cset *cs)
1255 {
1256         size_t i;
1257         cset *top = &p->g->sets[p->g->ncsets];
1258         size_t css = (size_t)p->g->csetsize;
1259
1260         for (i = 0; i < css; i++)
1261                 CHsub(cs, i);
1262         if (cs == top-1)        /* recover only the easy case */
1263                 p->g->ncsets--;
1264 }
1265
1266 /*
1267  - freezeset - final processing on a set of characters
1268  *
1269  * The main task here is merging identical sets.  This is usually a waste
1270  * of time (although the hash code minimizes the overhead), but can win
1271  * big if REG_ICASE is being used.  REG_ICASE, by the way, is why the hash
1272  * is done using addition rather than xor -- all ASCII [aA] sets xor to
1273  * the same value!
1274  */
1275 static int                      /* set number */
1276 freezeset(struct parse *p, cset *cs)
1277 {
1278         uch h = cs->hash;
1279         size_t i;
1280         cset *top = &p->g->sets[p->g->ncsets];
1281         cset *cs2;
1282         size_t css = (size_t)p->g->csetsize;
1283
1284         /* look for an earlier one which is the same */
1285         for (cs2 = &p->g->sets[0]; cs2 < top; cs2++)
1286                 if (cs2->hash == h && cs2 != cs) {
1287                         /* maybe */
1288                         for (i = 0; i < css; i++)
1289                                 if (!!CHIN(cs2, i) != !!CHIN(cs, i))
1290                                         break;          /* no */
1291                         if (i == css)
1292                                 break;                  /* yes */
1293                 }
1294
1295         if (cs2 < top) {        /* found one */
1296                 freeset(p, cs);
1297                 cs = cs2;
1298         }
1299
1300         return((int)(cs - p->g->sets));
1301 }
1302
1303 /*
1304  - firstch - return first character in a set (which must have at least one)
1305  */
1306 static int                      /* character; there is no "none" value */
1307 firstch(struct parse *p, cset *cs)
1308 {
1309         size_t i;
1310         size_t css = (size_t)p->g->csetsize;
1311
1312         for (i = 0; i < css; i++)
1313                 if (CHIN(cs, i))
1314                         return((char)i);
1315         assert(never);
1316         return(0);              /* arbitrary */
1317 }
1318
1319 /*
1320  - nch - number of characters in a set
1321  */
1322 static int
1323 nch(struct parse *p, cset *cs)
1324 {
1325         size_t i;
1326         size_t css = (size_t)p->g->csetsize;
1327         int n = 0;
1328
1329         for (i = 0; i < css; i++)
1330                 if (CHIN(cs, i))
1331                         n++;
1332         return(n);
1333 }
1334
1335 /*
1336  - mcadd - add a collating element to a cset
1337  */
1338 static void
1339 mcadd( struct parse *p, cset *cs, const char *cp)
1340 {
1341         size_t oldend = cs->smultis;
1342         void *np;
1343
1344         cs->smultis += strlen(cp) + 1;
1345         np = realloc(cs->multis, cs->smultis);
1346         if (np == NULL) {
1347                 if (cs->multis)
1348                         free(cs->multis);
1349                 cs->multis = NULL;
1350                 SETERROR(REG_ESPACE);
1351                 return;
1352         }
1353         cs->multis = np;
1354
1355         llvm_strlcpy(cs->multis + oldend - 1, cp, cs->smultis - oldend + 1);
1356 }
1357
1358 /*
1359  - mcinvert - invert the list of collating elements in a cset
1360  *
1361  * This would have to know the set of possibilities.  Implementation
1362  * is deferred.
1363  */
1364 /* ARGSUSED */
1365 static void
1366 mcinvert(struct parse *p, cset *cs)
1367 {
1368         assert(cs->multis == NULL);     /* xxx */
1369 }
1370
1371 /*
1372  - mccase - add case counterparts of the list of collating elements in a cset
1373  *
1374  * This would have to know the set of possibilities.  Implementation
1375  * is deferred.
1376  */
1377 /* ARGSUSED */
1378 static void
1379 mccase(struct parse *p, cset *cs)
1380 {
1381         assert(cs->multis == NULL);     /* xxx */
1382 }
1383
1384 /*
1385  - isinsets - is this character in any sets?
1386  */
1387 static int                      /* predicate */
1388 isinsets(struct re_guts *g, int c)
1389 {
1390         uch *col;
1391         int i;
1392         int ncols = (g->ncsets+(CHAR_BIT-1)) / CHAR_BIT;
1393         unsigned uc = (uch)c;
1394
1395         for (i = 0, col = g->setbits; i < ncols; i++, col += g->csetsize)
1396                 if (col[uc] != 0)
1397                         return(1);
1398         return(0);
1399 }
1400
1401 /*
1402  - samesets - are these two characters in exactly the same sets?
1403  */
1404 static int                      /* predicate */
1405 samesets(struct re_guts *g, int c1, int c2)
1406 {
1407         uch *col;
1408         int i;
1409         int ncols = (g->ncsets+(CHAR_BIT-1)) / CHAR_BIT;
1410         unsigned uc1 = (uch)c1;
1411         unsigned uc2 = (uch)c2;
1412
1413         for (i = 0, col = g->setbits; i < ncols; i++, col += g->csetsize)
1414                 if (col[uc1] != col[uc2])
1415                         return(0);
1416         return(1);
1417 }
1418
1419 /*
1420  - categorize - sort out character categories
1421  */
1422 static void
1423 categorize(struct parse *p, struct re_guts *g)
1424 {
1425         cat_t *cats = g->categories;
1426         int c;
1427         int c2;
1428         cat_t cat;
1429
1430         /* avoid making error situations worse */
1431         if (p->error != 0)
1432                 return;
1433
1434         for (c = CHAR_MIN; c <= CHAR_MAX; c++)
1435                 if (cats[c] == 0 && isinsets(g, c)) {
1436                         cat = g->ncategories++;
1437                         cats[c] = cat;
1438                         for (c2 = c+1; c2 <= CHAR_MAX; c2++)
1439                                 if (cats[c2] == 0 && samesets(g, c, c2))
1440                                         cats[c2] = cat;
1441                 }
1442 }
1443
1444 /*
1445  - dupl - emit a duplicate of a bunch of sops
1446  */
1447 static sopno                    /* start of duplicate */
1448 dupl(struct parse *p,
1449     sopno start,                /* from here */
1450     sopno finish)               /* to this less one */
1451 {
1452         sopno ret = HERE();
1453         sopno len = finish - start;
1454
1455         assert(finish >= start);
1456         if (len == 0)
1457                 return(ret);
1458         enlarge(p, p->ssize + len);     /* this many unexpected additions */
1459         assert(p->ssize >= p->slen + len);
1460         (void) memmove((char *)(p->strip + p->slen),
1461                 (char *)(p->strip + start), (size_t)len*sizeof(sop));
1462         p->slen += len;
1463         return(ret);
1464 }
1465
1466 /*
1467  - doemit - emit a strip operator
1468  *
1469  * It might seem better to implement this as a macro with a function as
1470  * hard-case backup, but it's just too big and messy unless there are
1471  * some changes to the data structures.  Maybe later.
1472  */
1473 static void
1474 doemit(struct parse *p, sop op, size_t opnd)
1475 {
1476         /* avoid making error situations worse */
1477         if (p->error != 0)
1478                 return;
1479
1480         /* deal with oversize operands ("can't happen", more or less) */
1481         assert(opnd < 1<<OPSHIFT);
1482
1483         /* deal with undersized strip */
1484         if (p->slen >= p->ssize)
1485                 enlarge(p, (p->ssize+1) / 2 * 3);       /* +50% */
1486         assert(p->slen < p->ssize);
1487
1488         /* finally, it's all reduced to the easy case */
1489         p->strip[p->slen++] = SOP(op, opnd);
1490 }
1491
1492 /*
1493  - doinsert - insert a sop into the strip
1494  */
1495 static void
1496 doinsert(struct parse *p, sop op, size_t opnd, sopno pos)
1497 {
1498         sopno sn;
1499         sop s;
1500         int i;
1501
1502         /* avoid making error situations worse */
1503         if (p->error != 0)
1504                 return;
1505
1506         sn = HERE();
1507         EMIT(op, opnd);         /* do checks, ensure space */
1508         assert(HERE() == sn+1);
1509         s = p->strip[sn];
1510
1511         /* adjust paren pointers */
1512         assert(pos > 0);
1513         for (i = 1; i < NPAREN; i++) {
1514                 if (p->pbegin[i] >= pos) {
1515                         p->pbegin[i]++;
1516                 }
1517                 if (p->pend[i] >= pos) {
1518                         p->pend[i]++;
1519                 }
1520         }
1521
1522         memmove((char *)&p->strip[pos+1], (char *)&p->strip[pos],
1523                                                 (HERE()-pos-1)*sizeof(sop));
1524         p->strip[pos] = s;
1525 }
1526
1527 /*
1528  - dofwd - complete a forward reference
1529  */
1530 static void
1531 dofwd(struct parse *p, sopno pos, sop value)
1532 {
1533         /* avoid making error situations worse */
1534         if (p->error != 0)
1535                 return;
1536
1537         assert(value < 1<<OPSHIFT);
1538         p->strip[pos] = OP(p->strip[pos]) | value;
1539 }
1540
1541 /*
1542  - enlarge - enlarge the strip
1543  */
1544 static void
1545 enlarge(struct parse *p, sopno size)
1546 {
1547         sop *sp;
1548
1549         if (p->ssize >= size)
1550                 return;
1551
1552         if ((uintptr_t)size > SIZE_MAX / sizeof(sop)) {
1553                 SETERROR(REG_ESPACE);
1554                 return;
1555         }
1556
1557         sp = (sop *)realloc(p->strip, size*sizeof(sop));
1558         if (sp == NULL) {
1559                 SETERROR(REG_ESPACE);
1560                 return;
1561         }
1562         p->strip = sp;
1563         p->ssize = size;
1564 }
1565
1566 /*
1567  - stripsnug - compact the strip
1568  */
1569 static void
1570 stripsnug(struct parse *p, struct re_guts *g)
1571 {
1572         g->nstates = p->slen;
1573         if ((uintptr_t)p->slen > SIZE_MAX / sizeof(sop)) {
1574                 g->strip = p->strip;
1575                 SETERROR(REG_ESPACE);
1576                 return;
1577         }
1578
1579         g->strip = (sop *)realloc((char *)p->strip, p->slen * sizeof(sop));
1580         if (g->strip == NULL) {
1581                 SETERROR(REG_ESPACE);
1582                 g->strip = p->strip;
1583         }
1584 }
1585
1586 /*
1587  - findmust - fill in must and mlen with longest mandatory literal string
1588  *
1589  * This algorithm could do fancy things like analyzing the operands of |
1590  * for common subsequences.  Someday.  This code is simple and finds most
1591  * of the interesting cases.
1592  *
1593  * Note that must and mlen got initialized during setup.
1594  */
1595 static void
1596 findmust(struct parse *p, struct re_guts *g)
1597 {
1598         sop *scan;
1599         sop *start = 0; /* start initialized in the default case, after that */
1600         sop *newstart = 0; /* newstart was initialized in the OCHAR case */
1601         sopno newlen;
1602         sop s;
1603         char *cp;
1604         sopno i;
1605
1606         /* avoid making error situations worse */
1607         if (p->error != 0)
1608                 return;
1609
1610         /* find the longest OCHAR sequence in strip */
1611         newlen = 0;
1612         scan = g->strip + 1;
1613         do {
1614                 s = *scan++;
1615                 switch (OP(s)) {
1616                 case OCHAR:             /* sequence member */
1617                         if (newlen == 0)                /* new sequence */
1618                                 newstart = scan - 1;
1619                         newlen++;
1620                         break;
1621                 case OPLUS_:            /* things that don't break one */
1622                 case OLPAREN:
1623                 case ORPAREN:
1624                         break;
1625                 case OQUEST_:           /* things that must be skipped */
1626                 case OCH_:
1627                         scan--;
1628                         do {
1629                                 scan += OPND(s);
1630                                 s = *scan;
1631                                 /* assert() interferes w debug printouts */
1632                                 if (OP(s) != O_QUEST && OP(s) != O_CH &&
1633                                                         OP(s) != OOR2) {
1634                                         g->iflags |= REGEX_BAD;
1635                                         return;
1636                                 }
1637                         } while (OP(s) != O_QUEST && OP(s) != O_CH);
1638                         /* fallthrough */
1639                 default:                /* things that break a sequence */
1640                         if (newlen > g->mlen) {         /* ends one */
1641                                 start = newstart;
1642                                 g->mlen = newlen;
1643                         }
1644                         newlen = 0;
1645                         break;
1646                 }
1647         } while (OP(s) != OEND);
1648
1649         if (g->mlen == 0)               /* there isn't one */
1650                 return;
1651
1652         /* turn it into a character string */
1653         g->must = malloc((size_t)g->mlen + 1);
1654         if (g->must == NULL) {          /* argh; just forget it */
1655                 g->mlen = 0;
1656                 return;
1657         }
1658         cp = g->must;
1659         scan = start;
1660         for (i = g->mlen; i > 0; i--) {
1661                 while (OP(s = *scan++) != OCHAR)
1662                         continue;
1663                 assert(cp < g->must + g->mlen);
1664                 *cp++ = (char)OPND(s);
1665         }
1666         assert(cp == g->must + g->mlen);
1667         *cp++ = '\0';           /* just on general principles */
1668 }
1669
1670 /*
1671  - pluscount - count + nesting
1672  */
1673 static sopno                    /* nesting depth */
1674 pluscount(struct parse *p, struct re_guts *g)
1675 {
1676         sop *scan;
1677         sop s;
1678         sopno plusnest = 0;
1679         sopno maxnest = 0;
1680
1681         if (p->error != 0)
1682                 return(0);      /* there may not be an OEND */
1683
1684         scan = g->strip + 1;
1685         do {
1686                 s = *scan++;
1687                 switch (OP(s)) {
1688                 case OPLUS_:
1689                         plusnest++;
1690                         break;
1691                 case O_PLUS:
1692                         if (plusnest > maxnest)
1693                                 maxnest = plusnest;
1694                         plusnest--;
1695                         break;
1696                 }
1697         } while (OP(s) != OEND);
1698         if (plusnest != 0)
1699                 g->iflags |= REGEX_BAD;
1700         return(maxnest);
1701 }