xref: /openbsd-src/lib/libc/regex/regcomp.c (revision daf88648c0e349d5c02e1504293082072c981640)
1 /*	$OpenBSD: regcomp.c,v 1.16 2006/03/31 05:36:36 deraadt Exp $ */
2 /*-
3  * Copyright (c) 1992, 1993, 1994 Henry Spencer.
4  * Copyright (c) 1992, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Henry Spencer.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)regcomp.c	8.5 (Berkeley) 3/20/94
35  */
36 
37 #include <sys/types.h>
38 #include <stdio.h>
39 #include <string.h>
40 #include <ctype.h>
41 #include <limits.h>
42 #include <stdlib.h>
43 #include <regex.h>
44 
45 #include "utils.h"
46 #include "regex2.h"
47 
48 #include "cclass.h"
49 #include "cname.h"
50 
51 /*
52  * parse structure, passed up and down to avoid global variables and
53  * other clumsinesses
54  */
55 struct parse {
56 	char *next;		/* next character in RE */
57 	char *end;		/* end of string (-> NUL normally) */
58 	int error;		/* has an error been seen? */
59 	sop *strip;		/* malloced strip */
60 	sopno ssize;		/* malloced strip size (allocated) */
61 	sopno slen;		/* malloced strip length (used) */
62 	int ncsalloc;		/* number of csets allocated */
63 	struct re_guts *g;
64 #	define	NPAREN	10	/* we need to remember () 1-9 for back refs */
65 	sopno pbegin[NPAREN];	/* -> ( ([0] unused) */
66 	sopno pend[NPAREN];	/* -> ) ([0] unused) */
67 };
68 
69 static void p_ere(struct parse *, int);
70 static void p_ere_exp(struct parse *);
71 static void p_str(struct parse *);
72 static void p_bre(struct parse *, int, int);
73 static int p_simp_re(struct parse *, int);
74 static int p_count(struct parse *);
75 static void p_bracket(struct parse *);
76 static void p_b_term(struct parse *, cset *);
77 static void p_b_cclass(struct parse *, cset *);
78 static void p_b_eclass(struct parse *, cset *);
79 static char p_b_symbol(struct parse *);
80 static char p_b_coll_elem(struct parse *, int);
81 static char othercase(int);
82 static void bothcases(struct parse *, int);
83 static void ordinary(struct parse *, int);
84 static void nonnewline(struct parse *);
85 static void repeat(struct parse *, sopno, int, int);
86 static int seterr(struct parse *, int);
87 static cset *allocset(struct parse *);
88 static void freeset(struct parse *, cset *);
89 static int freezeset(struct parse *, cset *);
90 static int firstch(struct parse *, cset *);
91 static int nch(struct parse *, cset *);
92 static void mcadd(struct parse *, cset *, char *);
93 static void mcinvert(struct parse *, cset *);
94 static void mccase(struct parse *, cset *);
95 static int isinsets(struct re_guts *, int);
96 static int samesets(struct re_guts *, int, int);
97 static void categorize(struct parse *, struct re_guts *);
98 static sopno dupl(struct parse *, sopno, sopno);
99 static void doemit(struct parse *, sop, size_t);
100 static void doinsert(struct parse *, sop, size_t, sopno);
101 static void dofwd(struct parse *, sopno, sop);
102 static void enlarge(struct parse *, sopno);
103 static void stripsnug(struct parse *, struct re_guts *);
104 static void findmust(struct parse *, struct re_guts *);
105 static sopno pluscount(struct parse *, struct re_guts *);
106 
107 static char nuls[10];		/* place to point scanner in event of error */
108 
109 /*
110  * macros for use with parse structure
111  * BEWARE:  these know that the parse structure is named `p' !!!
112  */
113 #define	PEEK()	(*p->next)
114 #define	PEEK2()	(*(p->next+1))
115 #define	MORE()	(p->next < p->end)
116 #define	MORE2()	(p->next+1 < p->end)
117 #define	SEE(c)	(MORE() && PEEK() == (c))
118 #define	SEETWO(a, b)	(MORE() && MORE2() && PEEK() == (a) && PEEK2() == (b))
119 #define	EAT(c)	((SEE(c)) ? (NEXT(), 1) : 0)
120 #define	EATTWO(a, b)	((SEETWO(a, b)) ? (NEXT2(), 1) : 0)
121 #define	NEXT()	(p->next++)
122 #define	NEXT2()	(p->next += 2)
123 #define	NEXTn(n)	(p->next += (n))
124 #define	GETNEXT()	(*p->next++)
125 #define	SETERROR(e)	seterr(p, (e))
126 #define	REQUIRE(co, e)	((co) || SETERROR(e))
127 #define	MUSTSEE(c, e)	(REQUIRE(MORE() && PEEK() == (c), e))
128 #define	MUSTEAT(c, e)	(REQUIRE(MORE() && GETNEXT() == (c), e))
129 #define	MUSTNOTSEE(c, e)	(REQUIRE(!MORE() || PEEK() != (c), e))
130 #define	EMIT(op, sopnd)	doemit(p, (sop)(op), (size_t)(sopnd))
131 #define	INSERT(op, pos)	doinsert(p, (sop)(op), HERE()-(pos)+1, pos)
132 #define	AHEAD(pos)		dofwd(p, pos, HERE()-(pos))
133 #define	ASTERN(sop, pos)	EMIT(sop, HERE()-pos)
134 #define	HERE()		(p->slen)
135 #define	THERE()		(p->slen - 1)
136 #define	THERETHERE()	(p->slen - 2)
137 #define	DROP(n)	(p->slen -= (n))
138 
139 #ifndef NDEBUG
140 static int never = 0;		/* for use in asserts; shuts lint up */
141 #else
142 #define	never	0		/* some <assert.h>s have bugs too */
143 #endif
144 
145 /*
146  - regcomp - interface for parser and compilation
147  */
148 int				/* 0 success, otherwise REG_something */
149 regcomp(regex_t *preg, const char *pattern, int cflags)
150 {
151 	struct parse pa;
152 	struct re_guts *g;
153 	struct parse *p = &pa;
154 	int i;
155 	size_t len;
156 #ifdef REDEBUG
157 #	define	GOODFLAGS(f)	(f)
158 #else
159 #	define	GOODFLAGS(f)	((f)&~REG_DUMP)
160 #endif
161 
162 	cflags = GOODFLAGS(cflags);
163 	if ((cflags&REG_EXTENDED) && (cflags&REG_NOSPEC))
164 		return(REG_INVARG);
165 
166 	if (cflags&REG_PEND) {
167 		if (preg->re_endp < pattern)
168 			return(REG_INVARG);
169 		len = preg->re_endp - pattern;
170 	} else
171 		len = strlen((char *)pattern);
172 
173 	/* do the mallocs early so failure handling is easy */
174 	g = (struct re_guts *)malloc(sizeof(struct re_guts) +
175 							(NC-1)*sizeof(cat_t));
176 	if (g == NULL)
177 		return(REG_ESPACE);
178 	p->ssize = len/(size_t)2*(size_t)3 + (size_t)1;	/* ugh */
179 	p->strip = (sop *)malloc(p->ssize * sizeof(sop));
180 	p->slen = 0;
181 	if (p->strip == NULL) {
182 		free((char *)g);
183 		return(REG_ESPACE);
184 	}
185 
186 	/* set things up */
187 	p->g = g;
188 	p->next = (char *)pattern;	/* convenience; we do not modify it */
189 	p->end = p->next + len;
190 	p->error = 0;
191 	p->ncsalloc = 0;
192 	for (i = 0; i < NPAREN; i++) {
193 		p->pbegin[i] = 0;
194 		p->pend[i] = 0;
195 	}
196 	g->csetsize = NC;
197 	g->sets = NULL;
198 	g->setbits = NULL;
199 	g->ncsets = 0;
200 	g->cflags = cflags;
201 	g->iflags = 0;
202 	g->nbol = 0;
203 	g->neol = 0;
204 	g->must = NULL;
205 	g->mlen = 0;
206 	g->nsub = 0;
207 	g->ncategories = 1;	/* category 0 is "everything else" */
208 	g->categories = &g->catspace[-(CHAR_MIN)];
209 	(void) memset((char *)g->catspace, 0, NC*sizeof(cat_t));
210 	g->backrefs = 0;
211 
212 	/* do it */
213 	EMIT(OEND, 0);
214 	g->firststate = THERE();
215 	if (cflags&REG_EXTENDED)
216 		p_ere(p, OUT);
217 	else if (cflags&REG_NOSPEC)
218 		p_str(p);
219 	else
220 		p_bre(p, OUT, OUT);
221 	EMIT(OEND, 0);
222 	g->laststate = THERE();
223 
224 	/* tidy up loose ends and fill things in */
225 	categorize(p, g);
226 	stripsnug(p, g);
227 	findmust(p, g);
228 	g->nplus = pluscount(p, g);
229 	g->magic = MAGIC2;
230 	preg->re_nsub = g->nsub;
231 	preg->re_g = g;
232 	preg->re_magic = MAGIC1;
233 #ifndef REDEBUG
234 	/* not debugging, so can't rely on the assert() in regexec() */
235 	if (g->iflags&BAD)
236 		SETERROR(REG_ASSERT);
237 #endif
238 
239 	/* win or lose, we're done */
240 	if (p->error != 0)	/* lose */
241 		regfree(preg);
242 	return(p->error);
243 }
244 
245 /*
246  - p_ere - ERE parser top level, concatenation and alternation
247  */
248 static void
249 p_ere(struct parse *p, int stop)	/* character this ERE should end at */
250 {
251 	char c;
252 	sopno prevback;
253 	sopno prevfwd;
254 	sopno conc;
255 	int first = 1;		/* is this the first alternative? */
256 
257 	for (;;) {
258 		/* do a bunch of concatenated expressions */
259 		conc = HERE();
260 		while (MORE() && (c = PEEK()) != '|' && c != stop)
261 			p_ere_exp(p);
262 		REQUIRE(HERE() != conc, REG_EMPTY);	/* require nonempty */
263 
264 		if (!EAT('|'))
265 			break;		/* NOTE BREAK OUT */
266 
267 		if (first) {
268 			INSERT(OCH_, conc);	/* offset is wrong */
269 			prevfwd = conc;
270 			prevback = conc;
271 			first = 0;
272 		}
273 		ASTERN(OOR1, prevback);
274 		prevback = THERE();
275 		AHEAD(prevfwd);			/* fix previous offset */
276 		prevfwd = HERE();
277 		EMIT(OOR2, 0);			/* offset is very wrong */
278 	}
279 
280 	if (!first) {		/* tail-end fixups */
281 		AHEAD(prevfwd);
282 		ASTERN(O_CH, prevback);
283 	}
284 
285 	assert(!MORE() || SEE(stop));
286 }
287 
288 /*
289  - p_ere_exp - parse one subERE, an atom possibly followed by a repetition op
290  */
291 static void
292 p_ere_exp(struct parse *p)
293 {
294 	char c;
295 	sopno pos;
296 	int count;
297 	int count2;
298 	sopno subno;
299 	int wascaret = 0;
300 
301 	assert(MORE());		/* caller should have ensured this */
302 	c = GETNEXT();
303 
304 	pos = HERE();
305 	switch (c) {
306 	case '(':
307 		REQUIRE(MORE(), REG_EPAREN);
308 		p->g->nsub++;
309 		subno = p->g->nsub;
310 		if (subno < NPAREN)
311 			p->pbegin[subno] = HERE();
312 		EMIT(OLPAREN, subno);
313 		if (!SEE(')'))
314 			p_ere(p, ')');
315 		if (subno < NPAREN) {
316 			p->pend[subno] = HERE();
317 			assert(p->pend[subno] != 0);
318 		}
319 		EMIT(ORPAREN, subno);
320 		MUSTEAT(')', REG_EPAREN);
321 		break;
322 #ifndef POSIX_MISTAKE
323 	case ')':		/* happens only if no current unmatched ( */
324 		/*
325 		 * You may ask, why the ifndef?  Because I didn't notice
326 		 * this until slightly too late for 1003.2, and none of the
327 		 * other 1003.2 regular-expression reviewers noticed it at
328 		 * all.  So an unmatched ) is legal POSIX, at least until
329 		 * we can get it fixed.
330 		 */
331 		SETERROR(REG_EPAREN);
332 		break;
333 #endif
334 	case '^':
335 		EMIT(OBOL, 0);
336 		p->g->iflags |= USEBOL;
337 		p->g->nbol++;
338 		wascaret = 1;
339 		break;
340 	case '$':
341 		EMIT(OEOL, 0);
342 		p->g->iflags |= USEEOL;
343 		p->g->neol++;
344 		break;
345 	case '|':
346 		SETERROR(REG_EMPTY);
347 		break;
348 	case '*':
349 	case '+':
350 	case '?':
351 		SETERROR(REG_BADRPT);
352 		break;
353 	case '.':
354 		if (p->g->cflags&REG_NEWLINE)
355 			nonnewline(p);
356 		else
357 			EMIT(OANY, 0);
358 		break;
359 	case '[':
360 		p_bracket(p);
361 		break;
362 	case '\\':
363 		REQUIRE(MORE(), REG_EESCAPE);
364 		c = GETNEXT();
365 		ordinary(p, c);
366 		break;
367 	case '{':		/* okay as ordinary except if digit follows */
368 		REQUIRE(!MORE() || !isdigit((uch)PEEK()), REG_BADRPT);
369 		/* FALLTHROUGH */
370 	default:
371 		ordinary(p, c);
372 		break;
373 	}
374 
375 	if (!MORE())
376 		return;
377 	c = PEEK();
378 	/* we call { a repetition if followed by a digit */
379 	if (!( c == '*' || c == '+' || c == '?' ||
380 				(c == '{' && MORE2() && isdigit((uch)PEEK2())) ))
381 		return;		/* no repetition, we're done */
382 	NEXT();
383 
384 	REQUIRE(!wascaret, REG_BADRPT);
385 	switch (c) {
386 	case '*':	/* implemented as +? */
387 		/* this case does not require the (y|) trick, noKLUDGE */
388 		INSERT(OPLUS_, pos);
389 		ASTERN(O_PLUS, pos);
390 		INSERT(OQUEST_, pos);
391 		ASTERN(O_QUEST, pos);
392 		break;
393 	case '+':
394 		INSERT(OPLUS_, pos);
395 		ASTERN(O_PLUS, pos);
396 		break;
397 	case '?':
398 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
399 		INSERT(OCH_, pos);		/* offset slightly wrong */
400 		ASTERN(OOR1, pos);		/* this one's right */
401 		AHEAD(pos);			/* fix the OCH_ */
402 		EMIT(OOR2, 0);			/* offset very wrong... */
403 		AHEAD(THERE());			/* ...so fix it */
404 		ASTERN(O_CH, THERETHERE());
405 		break;
406 	case '{':
407 		count = p_count(p);
408 		if (EAT(',')) {
409 			if (isdigit((uch)PEEK())) {
410 				count2 = p_count(p);
411 				REQUIRE(count <= count2, REG_BADBR);
412 			} else		/* single number with comma */
413 				count2 = INFINITY;
414 		} else		/* just a single number */
415 			count2 = count;
416 		repeat(p, pos, count, count2);
417 		if (!EAT('}')) {	/* error heuristics */
418 			while (MORE() && PEEK() != '}')
419 				NEXT();
420 			REQUIRE(MORE(), REG_EBRACE);
421 			SETERROR(REG_BADBR);
422 		}
423 		break;
424 	}
425 
426 	if (!MORE())
427 		return;
428 	c = PEEK();
429 	if (!( c == '*' || c == '+' || c == '?' ||
430 				(c == '{' && MORE2() && isdigit((uch)PEEK2())) ) )
431 		return;
432 	SETERROR(REG_BADRPT);
433 }
434 
435 /*
436  - p_str - string (no metacharacters) "parser"
437  */
438 static void
439 p_str(struct parse *p)
440 {
441 	REQUIRE(MORE(), REG_EMPTY);
442 	while (MORE())
443 		ordinary(p, GETNEXT());
444 }
445 
446 /*
447  - p_bre - BRE parser top level, anchoring and concatenation
448  * Giving end1 as OUT essentially eliminates the end1/end2 check.
449  *
450  * This implementation is a bit of a kludge, in that a trailing $ is first
451  * taken as an ordinary character and then revised to be an anchor.  The
452  * only undesirable side effect is that '$' gets included as a character
453  * category in such cases.  This is fairly harmless; not worth fixing.
454  * The amount of lookahead needed to avoid this kludge is excessive.
455  */
456 static void
457 p_bre(struct parse *p,
458     int end1,		/* first terminating character */
459     int end2)		/* second terminating character */
460 {
461 	sopno start = HERE();
462 	int first = 1;			/* first subexpression? */
463 	int wasdollar = 0;
464 
465 	if (EAT('^')) {
466 		EMIT(OBOL, 0);
467 		p->g->iflags |= USEBOL;
468 		p->g->nbol++;
469 	}
470 	while (MORE() && !SEETWO(end1, end2)) {
471 		wasdollar = p_simp_re(p, first);
472 		first = 0;
473 	}
474 	if (wasdollar) {	/* oops, that was a trailing anchor */
475 		DROP(1);
476 		EMIT(OEOL, 0);
477 		p->g->iflags |= USEEOL;
478 		p->g->neol++;
479 	}
480 
481 	REQUIRE(HERE() != start, REG_EMPTY);	/* require nonempty */
482 }
483 
484 /*
485  - p_simp_re - parse a simple RE, an atom possibly followed by a repetition
486  */
487 static int			/* was the simple RE an unbackslashed $? */
488 p_simp_re(struct parse *p,
489     int starordinary)		/* is a leading * an ordinary character? */
490 {
491 	int c;
492 	int count;
493 	int count2;
494 	sopno pos;
495 	int i;
496 	sopno subno;
497 #	define	BACKSL	(1<<CHAR_BIT)
498 
499 	pos = HERE();		/* repetion op, if any, covers from here */
500 
501 	assert(MORE());		/* caller should have ensured this */
502 	c = GETNEXT();
503 	if (c == '\\') {
504 		REQUIRE(MORE(), REG_EESCAPE);
505 		c = BACKSL | GETNEXT();
506 	}
507 	switch (c) {
508 	case '.':
509 		if (p->g->cflags&REG_NEWLINE)
510 			nonnewline(p);
511 		else
512 			EMIT(OANY, 0);
513 		break;
514 	case '[':
515 		p_bracket(p);
516 		break;
517 	case BACKSL|'{':
518 		SETERROR(REG_BADRPT);
519 		break;
520 	case BACKSL|'(':
521 		p->g->nsub++;
522 		subno = p->g->nsub;
523 		if (subno < NPAREN)
524 			p->pbegin[subno] = HERE();
525 		EMIT(OLPAREN, subno);
526 		/* the MORE here is an error heuristic */
527 		if (MORE() && !SEETWO('\\', ')'))
528 			p_bre(p, '\\', ')');
529 		if (subno < NPAREN) {
530 			p->pend[subno] = HERE();
531 			assert(p->pend[subno] != 0);
532 		}
533 		EMIT(ORPAREN, subno);
534 		REQUIRE(EATTWO('\\', ')'), REG_EPAREN);
535 		break;
536 	case BACKSL|')':	/* should not get here -- must be user */
537 	case BACKSL|'}':
538 		SETERROR(REG_EPAREN);
539 		break;
540 	case BACKSL|'1':
541 	case BACKSL|'2':
542 	case BACKSL|'3':
543 	case BACKSL|'4':
544 	case BACKSL|'5':
545 	case BACKSL|'6':
546 	case BACKSL|'7':
547 	case BACKSL|'8':
548 	case BACKSL|'9':
549 		i = (c&~BACKSL) - '0';
550 		assert(i < NPAREN);
551 		if (p->pend[i] != 0) {
552 			assert(i <= p->g->nsub);
553 			EMIT(OBACK_, i);
554 			assert(p->pbegin[i] != 0);
555 			assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
556 			assert(OP(p->strip[p->pend[i]]) == ORPAREN);
557 			(void) dupl(p, p->pbegin[i]+1, p->pend[i]);
558 			EMIT(O_BACK, i);
559 		} else
560 			SETERROR(REG_ESUBREG);
561 		p->g->backrefs = 1;
562 		break;
563 	case '*':
564 		REQUIRE(starordinary, REG_BADRPT);
565 		/* FALLTHROUGH */
566 	default:
567 		ordinary(p, (char)c);
568 		break;
569 	}
570 
571 	if (EAT('*')) {		/* implemented as +? */
572 		/* this case does not require the (y|) trick, noKLUDGE */
573 		INSERT(OPLUS_, pos);
574 		ASTERN(O_PLUS, pos);
575 		INSERT(OQUEST_, pos);
576 		ASTERN(O_QUEST, pos);
577 	} else if (EATTWO('\\', '{')) {
578 		count = p_count(p);
579 		if (EAT(',')) {
580 			if (MORE() && 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 (!EATTWO('\\', '}')) {	/* error heuristics */
589 			while (MORE() && !SEETWO('\\', '}'))
590 				NEXT();
591 			REQUIRE(MORE(), REG_EBRACE);
592 			SETERROR(REG_BADBR);
593 		}
594 	} else if (c == '$')	/* $ (but not \$) ends it */
595 		return(1);
596 
597 	return(0);
598 }
599 
600 /*
601  - p_count - parse a repetition count
602  */
603 static int			/* the value */
604 p_count(struct parse *p)
605 {
606 	int count = 0;
607 	int ndigits = 0;
608 
609 	while (MORE() && isdigit((uch)PEEK()) && count <= DUPMAX) {
610 		count = count*10 + (GETNEXT() - '0');
611 		ndigits++;
612 	}
613 
614 	REQUIRE(ndigits > 0 && count <= DUPMAX, REG_BADBR);
615 	return(count);
616 }
617 
618 /*
619  - p_bracket - parse a bracketed character list
620  *
621  * Note a significant property of this code:  if the allocset() did SETERROR,
622  * no set operations are done.
623  */
624 static void
625 p_bracket(struct parse *p)
626 {
627 	cset *cs;
628 	int invert = 0;
629 
630 	/* Dept of Truly Sickening Special-Case Kludges */
631 	if (p->next + 5 < p->end && strncmp(p->next, "[:<:]]", 6) == 0) {
632 		EMIT(OBOW, 0);
633 		NEXTn(6);
634 		return;
635 	}
636 	if (p->next + 5 < p->end && strncmp(p->next, "[:>:]]", 6) == 0) {
637 		EMIT(OEOW, 0);
638 		NEXTn(6);
639 		return;
640 	}
641 
642 	cs = allocset(p);
643 
644 	if (EAT('^'))
645 		invert++;	/* make note to invert set at end */
646 	if (EAT(']'))
647 		CHadd(cs, ']');
648 	else if (EAT('-'))
649 		CHadd(cs, '-');
650 	while (MORE() && PEEK() != ']' && !SEETWO('-', ']'))
651 		p_b_term(p, cs);
652 	if (EAT('-'))
653 		CHadd(cs, '-');
654 	MUSTEAT(']', REG_EBRACK);
655 
656 	if (p->error != 0) {	/* don't mess things up further */
657 		freeset(p, cs);
658 		return;
659 	}
660 
661 	if (p->g->cflags&REG_ICASE) {
662 		int i;
663 		int ci;
664 
665 		for (i = p->g->csetsize - 1; i >= 0; i--)
666 			if (CHIN(cs, i) && isalpha(i)) {
667 				ci = othercase(i);
668 				if (ci != i)
669 					CHadd(cs, ci);
670 			}
671 		if (cs->multis != NULL)
672 			mccase(p, cs);
673 	}
674 	if (invert) {
675 		int i;
676 
677 		for (i = p->g->csetsize - 1; i >= 0; i--)
678 			if (CHIN(cs, i))
679 				CHsub(cs, i);
680 			else
681 				CHadd(cs, i);
682 		if (p->g->cflags&REG_NEWLINE)
683 			CHsub(cs, '\n');
684 		if (cs->multis != NULL)
685 			mcinvert(p, cs);
686 	}
687 
688 	assert(cs->multis == NULL);		/* xxx */
689 
690 	if (nch(p, cs) == 1) {		/* optimize singleton sets */
691 		ordinary(p, firstch(p, cs));
692 		freeset(p, cs);
693 	} else
694 		EMIT(OANYOF, freezeset(p, cs));
695 }
696 
697 /*
698  - p_b_term - parse one term of a bracketed character list
699  */
700 static void
701 p_b_term(struct parse *p, cset *cs)
702 {
703 	char c;
704 	char start, finish;
705 	int i;
706 
707 	/* classify what we've got */
708 	switch ((MORE()) ? PEEK() : '\0') {
709 	case '[':
710 		c = (MORE2()) ? PEEK2() : '\0';
711 		break;
712 	case '-':
713 		SETERROR(REG_ERANGE);
714 		return;			/* NOTE RETURN */
715 		break;
716 	default:
717 		c = '\0';
718 		break;
719 	}
720 
721 	switch (c) {
722 	case ':':		/* character class */
723 		NEXT2();
724 		REQUIRE(MORE(), REG_EBRACK);
725 		c = PEEK();
726 		REQUIRE(c != '-' && c != ']', REG_ECTYPE);
727 		p_b_cclass(p, cs);
728 		REQUIRE(MORE(), REG_EBRACK);
729 		REQUIRE(EATTWO(':', ']'), REG_ECTYPE);
730 		break;
731 	case '=':		/* equivalence class */
732 		NEXT2();
733 		REQUIRE(MORE(), REG_EBRACK);
734 		c = PEEK();
735 		REQUIRE(c != '-' && c != ']', REG_ECOLLATE);
736 		p_b_eclass(p, cs);
737 		REQUIRE(MORE(), REG_EBRACK);
738 		REQUIRE(EATTWO('=', ']'), REG_ECOLLATE);
739 		break;
740 	default:		/* symbol, ordinary character, or range */
741 /* xxx revision needed for multichar stuff */
742 		start = p_b_symbol(p);
743 		if (SEE('-') && MORE2() && PEEK2() != ']') {
744 			/* range */
745 			NEXT();
746 			if (EAT('-'))
747 				finish = '-';
748 			else
749 				finish = p_b_symbol(p);
750 		} else
751 			finish = start;
752 /* xxx what about signed chars here... */
753 		REQUIRE(start <= finish, REG_ERANGE);
754 		for (i = start; i <= finish; i++)
755 			CHadd(cs, i);
756 		break;
757 	}
758 }
759 
760 /*
761  - p_b_cclass - parse a character-class name and deal with it
762  */
763 static void
764 p_b_cclass(struct parse *p, cset *cs)
765 {
766 	char *sp = p->next;
767 	struct cclass *cp;
768 	size_t len;
769 	char *u;
770 	char c;
771 
772 	while (MORE() && isalpha(PEEK()))
773 		NEXT();
774 	len = p->next - sp;
775 	for (cp = cclasses; cp->name != NULL; cp++)
776 		if (strncmp(cp->name, sp, len) == 0 && cp->name[len] == '\0')
777 			break;
778 	if (cp->name == NULL) {
779 		/* oops, didn't find it */
780 		SETERROR(REG_ECTYPE);
781 		return;
782 	}
783 
784 	u = cp->chars;
785 	while ((c = *u++) != '\0')
786 		CHadd(cs, c);
787 	for (u = cp->multis; *u != '\0'; u += strlen(u) + 1)
788 		MCadd(p, cs, u);
789 }
790 
791 /*
792  - p_b_eclass - parse an equivalence-class name and deal with it
793  *
794  * This implementation is incomplete. xxx
795  */
796 static void
797 p_b_eclass(struct parse *p, cset *cs)
798 {
799 	char c;
800 
801 	c = p_b_coll_elem(p, '=');
802 	CHadd(cs, c);
803 }
804 
805 /*
806  - p_b_symbol - parse a character or [..]ed multicharacter collating symbol
807  */
808 static char			/* value of symbol */
809 p_b_symbol(struct parse *p)
810 {
811 	char value;
812 
813 	REQUIRE(MORE(), REG_EBRACK);
814 	if (!EATTWO('[', '.'))
815 		return(GETNEXT());
816 
817 	/* collating symbol */
818 	value = p_b_coll_elem(p, '.');
819 	REQUIRE(EATTWO('.', ']'), REG_ECOLLATE);
820 	return(value);
821 }
822 
823 /*
824  - p_b_coll_elem - parse a collating-element name and look it up
825  */
826 static char			/* value of collating element */
827 p_b_coll_elem(struct parse *p,
828     int endc)			/* name ended by endc,']' */
829 {
830 	char *sp = p->next;
831 	struct cname *cp;
832 	int len;
833 
834 	while (MORE() && !SEETWO(endc, ']'))
835 		NEXT();
836 	if (!MORE()) {
837 		SETERROR(REG_EBRACK);
838 		return(0);
839 	}
840 	len = p->next - sp;
841 	for (cp = cnames; cp->name != NULL; cp++)
842 		if (strncmp(cp->name, sp, len) == 0 && cp->name[len] == '\0')
843 			return(cp->code);	/* known name */
844 	if (len == 1)
845 		return(*sp);	/* single character */
846 	SETERROR(REG_ECOLLATE);			/* neither */
847 	return(0);
848 }
849 
850 /*
851  - othercase - return the case counterpart of an alphabetic
852  */
853 static char			/* if no counterpart, return ch */
854 othercase(int ch)
855 {
856 	ch = (uch)ch;
857 	assert(isalpha(ch));
858 	if (isupper(ch))
859 		return ((uch)tolower(ch));
860 	else if (islower(ch))
861 		return ((uch)toupper(ch));
862 	else			/* peculiar, but could happen */
863 		return(ch);
864 }
865 
866 /*
867  - bothcases - emit a dualcase version of a two-case character
868  *
869  * Boy, is this implementation ever a kludge...
870  */
871 static void
872 bothcases(struct parse *p, int ch)
873 {
874 	char *oldnext = p->next;
875 	char *oldend = p->end;
876 	char bracket[3];
877 
878 	ch = (uch)ch;
879 	assert(othercase(ch) != ch);	/* p_bracket() would recurse */
880 	p->next = bracket;
881 	p->end = bracket+2;
882 	bracket[0] = ch;
883 	bracket[1] = ']';
884 	bracket[2] = '\0';
885 	p_bracket(p);
886 	assert(p->next == bracket+2);
887 	p->next = oldnext;
888 	p->end = oldend;
889 }
890 
891 /*
892  - ordinary - emit an ordinary character
893  */
894 static void
895 ordinary(struct parse *p, int ch)
896 {
897 	cat_t *cap = p->g->categories;
898 
899 	if ((p->g->cflags&REG_ICASE) && isalpha((uch)ch) && othercase(ch) != ch)
900 		bothcases(p, ch);
901 	else {
902 		EMIT(OCHAR, (uch)ch);
903 		if (cap[ch] == 0)
904 			cap[ch] = p->g->ncategories++;
905 	}
906 }
907 
908 /*
909  - nonnewline - emit REG_NEWLINE version of OANY
910  *
911  * Boy, is this implementation ever a kludge...
912  */
913 static void
914 nonnewline(struct parse *p)
915 {
916 	char *oldnext = p->next;
917 	char *oldend = p->end;
918 	char bracket[4];
919 
920 	p->next = bracket;
921 	p->end = bracket+3;
922 	bracket[0] = '^';
923 	bracket[1] = '\n';
924 	bracket[2] = ']';
925 	bracket[3] = '\0';
926 	p_bracket(p);
927 	assert(p->next == bracket+3);
928 	p->next = oldnext;
929 	p->end = oldend;
930 }
931 
932 /*
933  - repeat - generate code for a bounded repetition, recursively if needed
934  */
935 static void
936 repeat(struct parse *p,
937     sopno start,		/* operand from here to end of strip */
938     int from,			/* repeated from this number */
939     int to)			/* to this number of times (maybe INFINITY) */
940 {
941 	sopno finish = HERE();
942 #	define	N	2
943 #	define	INF	3
944 #	define	REP(f, t)	((f)*8 + (t))
945 #	define	MAP(n)	(((n) <= 1) ? (n) : ((n) == INFINITY) ? INF : N)
946 	sopno copy;
947 
948 	if (p->error != 0)	/* head off possible runaway recursion */
949 		return;
950 
951 	assert(from <= to);
952 
953 	switch (REP(MAP(from), MAP(to))) {
954 	case REP(0, 0):			/* must be user doing this */
955 		DROP(finish-start);	/* drop the operand */
956 		break;
957 	case REP(0, 1):			/* as x{1,1}? */
958 	case REP(0, N):			/* as x{1,n}? */
959 	case REP(0, INF):		/* as x{1,}? */
960 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
961 		INSERT(OCH_, start);		/* offset is wrong... */
962 		repeat(p, start+1, 1, to);
963 		ASTERN(OOR1, start);
964 		AHEAD(start);			/* ... fix it */
965 		EMIT(OOR2, 0);
966 		AHEAD(THERE());
967 		ASTERN(O_CH, THERETHERE());
968 		break;
969 	case REP(1, 1):			/* trivial case */
970 		/* done */
971 		break;
972 	case REP(1, N):			/* as x?x{1,n-1} */
973 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
974 		INSERT(OCH_, start);
975 		ASTERN(OOR1, start);
976 		AHEAD(start);
977 		EMIT(OOR2, 0);			/* offset very wrong... */
978 		AHEAD(THERE());			/* ...so fix it */
979 		ASTERN(O_CH, THERETHERE());
980 		copy = dupl(p, start+1, finish+1);
981 		assert(copy == finish+4);
982 		repeat(p, copy, 1, to-1);
983 		break;
984 	case REP(1, INF):		/* as x+ */
985 		INSERT(OPLUS_, start);
986 		ASTERN(O_PLUS, start);
987 		break;
988 	case REP(N, N):			/* as xx{m-1,n-1} */
989 		copy = dupl(p, start, finish);
990 		repeat(p, copy, from-1, to-1);
991 		break;
992 	case REP(N, INF):		/* as xx{n-1,INF} */
993 		copy = dupl(p, start, finish);
994 		repeat(p, copy, from-1, to);
995 		break;
996 	default:			/* "can't happen" */
997 		SETERROR(REG_ASSERT);	/* just in case */
998 		break;
999 	}
1000 }
1001 
1002 /*
1003  - seterr - set an error condition
1004  */
1005 static int			/* useless but makes type checking happy */
1006 seterr(struct parse *p, int e)
1007 {
1008 	if (p->error == 0)	/* keep earliest error condition */
1009 		p->error = e;
1010 	p->next = nuls;		/* try to bring things to a halt */
1011 	p->end = nuls;
1012 	return(0);		/* make the return value well-defined */
1013 }
1014 
1015 /*
1016  - allocset - allocate a set of characters for []
1017  */
1018 static cset *
1019 allocset(struct parse *p)
1020 {
1021 	int no = p->g->ncsets++;
1022 	size_t nc;
1023 	size_t nbytes;
1024 	cset *cs;
1025 	size_t css = (size_t)p->g->csetsize;
1026 	int i;
1027 
1028 	if (no >= p->ncsalloc) {	/* need another column of space */
1029 		p->ncsalloc += CHAR_BIT;
1030 		nc = p->ncsalloc;
1031 		assert(nc % CHAR_BIT == 0);
1032 		nbytes = nc / CHAR_BIT * css;
1033 		if (p->g->sets == NULL)
1034 			p->g->sets = (cset *)malloc(nc * sizeof(cset));
1035 		else {
1036 			cset *ptr;
1037 			ptr = (cset *)realloc((char *)p->g->sets,
1038 			    nc * sizeof(cset));
1039 			if (ptr == NULL) {
1040 				free(p->g->sets);
1041 				p->g->sets = NULL;
1042 			} else
1043 				p->g->sets = ptr;
1044 		}
1045 		if (p->g->sets == NULL)
1046 			goto nomem;
1047 
1048 		if (p->g->setbits == NULL)
1049 			p->g->setbits = (uch *)malloc(nbytes);
1050 		else {
1051 			uch *ptr;
1052 
1053 			ptr = (uch *)realloc((char *)p->g->setbits, nbytes);
1054 			if (ptr == NULL) {
1055 				free(p->g->setbits);
1056 				p->g->setbits = NULL;
1057 			} else {
1058 				p->g->setbits = ptr;
1059 
1060 				for (i = 0; i < no; i++)
1061 					p->g->sets[i].ptr = p->g->setbits +
1062 					    css*(i/CHAR_BIT);
1063 			}
1064 		}
1065 
1066 		if (p->g->sets == NULL || p->g->setbits == NULL) {
1067 nomem:
1068 			no = 0;
1069 			SETERROR(REG_ESPACE);
1070 			/* caller's responsibility not to do set ops */
1071 		} else
1072 			(void) memset((char *)p->g->setbits + (nbytes - css),
1073 			    0, css);
1074 	}
1075 
1076 	assert(p->g->sets != NULL);	/* xxx */
1077 	if (p->g->sets != NULL && p->g->setbits != NULL) {
1078 		cs = &p->g->sets[no];
1079 		cs->ptr = p->g->setbits + css*((no)/CHAR_BIT);
1080 	}
1081 	cs->mask = 1 << ((no) % CHAR_BIT);
1082 	cs->hash = 0;
1083 	cs->smultis = 0;
1084 	cs->multis = NULL;
1085 
1086 	return(cs);
1087 }
1088 
1089 /*
1090  - freeset - free a now-unused set
1091  */
1092 static void
1093 freeset(struct parse *p, cset *cs)
1094 {
1095 	int i;
1096 	cset *top = &p->g->sets[p->g->ncsets];
1097 	size_t css = (size_t)p->g->csetsize;
1098 
1099 	for (i = 0; i < css; i++)
1100 		CHsub(cs, i);
1101 	if (cs == top-1)	/* recover only the easy case */
1102 		p->g->ncsets--;
1103 }
1104 
1105 /*
1106  - freezeset - final processing on a set of characters
1107  *
1108  * The main task here is merging identical sets.  This is usually a waste
1109  * of time (although the hash code minimizes the overhead), but can win
1110  * big if REG_ICASE is being used.  REG_ICASE, by the way, is why the hash
1111  * is done using addition rather than xor -- all ASCII [aA] sets xor to
1112  * the same value!
1113  */
1114 static int			/* set number */
1115 freezeset(struct parse *p, cset *cs)
1116 {
1117 	uch h = cs->hash;
1118 	int i;
1119 	cset *top = &p->g->sets[p->g->ncsets];
1120 	cset *cs2;
1121 	size_t css = (size_t)p->g->csetsize;
1122 
1123 	/* look for an earlier one which is the same */
1124 	for (cs2 = &p->g->sets[0]; cs2 < top; cs2++)
1125 		if (cs2->hash == h && cs2 != cs) {
1126 			/* maybe */
1127 			for (i = 0; i < css; i++)
1128 				if (!!CHIN(cs2, i) != !!CHIN(cs, i))
1129 					break;		/* no */
1130 			if (i == css)
1131 				break;			/* yes */
1132 		}
1133 
1134 	if (cs2 < top) {	/* found one */
1135 		freeset(p, cs);
1136 		cs = cs2;
1137 	}
1138 
1139 	return((int)(cs - p->g->sets));
1140 }
1141 
1142 /*
1143  - firstch - return first character in a set (which must have at least one)
1144  */
1145 static int			/* character; there is no "none" value */
1146 firstch(struct parse *p, cset *cs)
1147 {
1148 	int i;
1149 	size_t css = (size_t)p->g->csetsize;
1150 
1151 	for (i = 0; i < css; i++)
1152 		if (CHIN(cs, i))
1153 			return((char)i);
1154 	assert(never);
1155 	return(0);		/* arbitrary */
1156 }
1157 
1158 /*
1159  - nch - number of characters in a set
1160  */
1161 static int
1162 nch(struct parse *p, cset *cs)
1163 {
1164 	int i;
1165 	size_t css = (size_t)p->g->csetsize;
1166 	int n = 0;
1167 
1168 	for (i = 0; i < css; i++)
1169 		if (CHIN(cs, i))
1170 			n++;
1171 	return(n);
1172 }
1173 
1174 /*
1175  - mcadd - add a collating element to a cset
1176  */
1177 static void
1178 mcadd( struct parse *p, cset *cs, char *cp)
1179 {
1180 	size_t oldend = cs->smultis;
1181 	void *np;
1182 
1183 	cs->smultis += strlen(cp) + 1;
1184 	if (cs->multis == NULL)
1185 		np = malloc(cs->smultis);
1186 	else
1187 		np = realloc(cs->multis, cs->smultis);
1188 	if (np == NULL) {
1189 		if (cs->multis)
1190 			free(cs->multis);
1191 		cs->multis = NULL;
1192 		SETERROR(REG_ESPACE);
1193 		return;
1194 	}
1195 	cs->multis = np;
1196 
1197 	strlcpy(cs->multis + oldend - 1, cp, cs->smultis - oldend + 1);
1198 }
1199 
1200 /*
1201  - mcinvert - invert the list of collating elements in a cset
1202  *
1203  * This would have to know the set of possibilities.  Implementation
1204  * is deferred.
1205  */
1206 /* ARGSUSED */
1207 static void
1208 mcinvert(struct parse *p, cset *cs)
1209 {
1210 	assert(cs->multis == NULL);	/* xxx */
1211 }
1212 
1213 /*
1214  - mccase - add case counterparts of the list of collating elements in a cset
1215  *
1216  * This would have to know the set of possibilities.  Implementation
1217  * is deferred.
1218  */
1219 /* ARGSUSED */
1220 static void
1221 mccase(struct parse *p, cset *cs)
1222 {
1223 	assert(cs->multis == NULL);	/* xxx */
1224 }
1225 
1226 /*
1227  - isinsets - is this character in any sets?
1228  */
1229 static int			/* predicate */
1230 isinsets(struct re_guts *g, int c)
1231 {
1232 	uch *col;
1233 	int i;
1234 	int ncols = (g->ncsets+(CHAR_BIT-1)) / CHAR_BIT;
1235 	unsigned uc = (uch)c;
1236 
1237 	for (i = 0, col = g->setbits; i < ncols; i++, col += g->csetsize)
1238 		if (col[uc] != 0)
1239 			return(1);
1240 	return(0);
1241 }
1242 
1243 /*
1244  - samesets - are these two characters in exactly the same sets?
1245  */
1246 static int			/* predicate */
1247 samesets(struct re_guts *g, int c1, int c2)
1248 {
1249 	uch *col;
1250 	int i;
1251 	int ncols = (g->ncsets+(CHAR_BIT-1)) / CHAR_BIT;
1252 	unsigned uc1 = (uch)c1;
1253 	unsigned uc2 = (uch)c2;
1254 
1255 	for (i = 0, col = g->setbits; i < ncols; i++, col += g->csetsize)
1256 		if (col[uc1] != col[uc2])
1257 			return(0);
1258 	return(1);
1259 }
1260 
1261 /*
1262  - categorize - sort out character categories
1263  */
1264 static void
1265 categorize(struct parse *p, struct re_guts *g)
1266 {
1267 	cat_t *cats = g->categories;
1268 	int c;
1269 	int c2;
1270 	cat_t cat;
1271 
1272 	/* avoid making error situations worse */
1273 	if (p->error != 0)
1274 		return;
1275 
1276 	for (c = CHAR_MIN; c <= CHAR_MAX; c++)
1277 		if (cats[c] == 0 && isinsets(g, c)) {
1278 			cat = g->ncategories++;
1279 			cats[c] = cat;
1280 			for (c2 = c+1; c2 <= CHAR_MAX; c2++)
1281 				if (cats[c2] == 0 && samesets(g, c, c2))
1282 					cats[c2] = cat;
1283 		}
1284 }
1285 
1286 /*
1287  - dupl - emit a duplicate of a bunch of sops
1288  */
1289 static sopno			/* start of duplicate */
1290 dupl(struct parse *p,
1291     sopno start,		/* from here */
1292     sopno finish)		/* to this less one */
1293 {
1294 	sopno ret = HERE();
1295 	sopno len = finish - start;
1296 
1297 	assert(finish >= start);
1298 	if (len == 0)
1299 		return(ret);
1300 	enlarge(p, p->ssize + len);	/* this many unexpected additions */
1301 	assert(p->ssize >= p->slen + len);
1302 	(void) memcpy((char *)(p->strip + p->slen),
1303 		(char *)(p->strip + start), (size_t)len*sizeof(sop));
1304 	p->slen += len;
1305 	return(ret);
1306 }
1307 
1308 /*
1309  - doemit - emit a strip operator
1310  *
1311  * It might seem better to implement this as a macro with a function as
1312  * hard-case backup, but it's just too big and messy unless there are
1313  * some changes to the data structures.  Maybe later.
1314  */
1315 static void
1316 doemit(struct parse *p, sop op, size_t opnd)
1317 {
1318 	/* avoid making error situations worse */
1319 	if (p->error != 0)
1320 		return;
1321 
1322 	/* deal with oversize operands ("can't happen", more or less) */
1323 	assert(opnd < 1<<OPSHIFT);
1324 
1325 	/* deal with undersized strip */
1326 	if (p->slen >= p->ssize)
1327 		enlarge(p, (p->ssize+1) / 2 * 3);	/* +50% */
1328 	assert(p->slen < p->ssize);
1329 
1330 	/* finally, it's all reduced to the easy case */
1331 	p->strip[p->slen++] = SOP(op, opnd);
1332 }
1333 
1334 /*
1335  - doinsert - insert a sop into the strip
1336  */
1337 static void
1338 doinsert(struct parse *p, sop op, size_t opnd, sopno pos)
1339 {
1340 	sopno sn;
1341 	sop s;
1342 	int i;
1343 
1344 	/* avoid making error situations worse */
1345 	if (p->error != 0)
1346 		return;
1347 
1348 	sn = HERE();
1349 	EMIT(op, opnd);		/* do checks, ensure space */
1350 	assert(HERE() == sn+1);
1351 	s = p->strip[sn];
1352 
1353 	/* adjust paren pointers */
1354 	assert(pos > 0);
1355 	for (i = 1; i < NPAREN; i++) {
1356 		if (p->pbegin[i] >= pos) {
1357 			p->pbegin[i]++;
1358 		}
1359 		if (p->pend[i] >= pos) {
1360 			p->pend[i]++;
1361 		}
1362 	}
1363 
1364 	memmove((char *)&p->strip[pos+1], (char *)&p->strip[pos],
1365 						(HERE()-pos-1)*sizeof(sop));
1366 	p->strip[pos] = s;
1367 }
1368 
1369 /*
1370  - dofwd - complete a forward reference
1371  */
1372 static void
1373 dofwd(struct parse *p, sopno pos, sop value)
1374 {
1375 	/* avoid making error situations worse */
1376 	if (p->error != 0)
1377 		return;
1378 
1379 	assert(value < 1<<OPSHIFT);
1380 	p->strip[pos] = OP(p->strip[pos]) | value;
1381 }
1382 
1383 /*
1384  - enlarge - enlarge the strip
1385  */
1386 static void
1387 enlarge(struct parse *p, sopno size)
1388 {
1389 	sop *sp;
1390 
1391 	if (p->ssize >= size)
1392 		return;
1393 
1394 	sp = (sop *)realloc(p->strip, size*sizeof(sop));
1395 	if (sp == NULL) {
1396 		SETERROR(REG_ESPACE);
1397 		return;
1398 	}
1399 	p->strip = sp;
1400 	p->ssize = size;
1401 }
1402 
1403 /*
1404  - stripsnug - compact the strip
1405  */
1406 static void
1407 stripsnug(struct parse *p, struct re_guts *g)
1408 {
1409 	g->nstates = p->slen;
1410 	g->strip = (sop *)realloc((char *)p->strip, p->slen * sizeof(sop));
1411 	if (g->strip == NULL) {
1412 		SETERROR(REG_ESPACE);
1413 		g->strip = p->strip;
1414 	}
1415 }
1416 
1417 /*
1418  - findmust - fill in must and mlen with longest mandatory literal string
1419  *
1420  * This algorithm could do fancy things like analyzing the operands of |
1421  * for common subsequences.  Someday.  This code is simple and finds most
1422  * of the interesting cases.
1423  *
1424  * Note that must and mlen got initialized during setup.
1425  */
1426 static void
1427 findmust(struct parse *p, struct re_guts *g)
1428 {
1429 	sop *scan;
1430 	sop *start;
1431 	sop *newstart;
1432 	sopno newlen;
1433 	sop s;
1434 	char *cp;
1435 	sopno i;
1436 
1437 	/* avoid making error situations worse */
1438 	if (p->error != 0)
1439 		return;
1440 
1441 	/* find the longest OCHAR sequence in strip */
1442 	newlen = 0;
1443 	scan = g->strip + 1;
1444 	do {
1445 		s = *scan++;
1446 		switch (OP(s)) {
1447 		case OCHAR:		/* sequence member */
1448 			if (newlen == 0)		/* new sequence */
1449 				newstart = scan - 1;
1450 			newlen++;
1451 			break;
1452 		case OPLUS_:		/* things that don't break one */
1453 		case OLPAREN:
1454 		case ORPAREN:
1455 			break;
1456 		case OQUEST_:		/* things that must be skipped */
1457 		case OCH_:
1458 			scan--;
1459 			do {
1460 				scan += OPND(s);
1461 				s = *scan;
1462 				/* assert() interferes w debug printouts */
1463 				if (OP(s) != O_QUEST && OP(s) != O_CH &&
1464 							OP(s) != OOR2) {
1465 					g->iflags |= BAD;
1466 					return;
1467 				}
1468 			} while (OP(s) != O_QUEST && OP(s) != O_CH);
1469 			/* fallthrough */
1470 		default:		/* things that break a sequence */
1471 			if (newlen > g->mlen) {		/* ends one */
1472 				start = newstart;
1473 				g->mlen = newlen;
1474 			}
1475 			newlen = 0;
1476 			break;
1477 		}
1478 	} while (OP(s) != OEND);
1479 
1480 	if (g->mlen == 0)		/* there isn't one */
1481 		return;
1482 
1483 	/* turn it into a character string */
1484 	g->must = malloc((size_t)g->mlen + 1);
1485 	if (g->must == NULL) {		/* argh; just forget it */
1486 		g->mlen = 0;
1487 		return;
1488 	}
1489 	cp = g->must;
1490 	scan = start;
1491 	for (i = g->mlen; i > 0; i--) {
1492 		while (OP(s = *scan++) != OCHAR)
1493 			continue;
1494 		assert(cp < g->must + g->mlen);
1495 		*cp++ = (char)OPND(s);
1496 	}
1497 	assert(cp == g->must + g->mlen);
1498 	*cp++ = '\0';		/* just on general principles */
1499 }
1500 
1501 /*
1502  - pluscount - count + nesting
1503  */
1504 static sopno			/* nesting depth */
1505 pluscount(struct parse *p, struct re_guts *g)
1506 {
1507 	sop *scan;
1508 	sop s;
1509 	sopno plusnest = 0;
1510 	sopno maxnest = 0;
1511 
1512 	if (p->error != 0)
1513 		return(0);	/* there may not be an OEND */
1514 
1515 	scan = g->strip + 1;
1516 	do {
1517 		s = *scan++;
1518 		switch (OP(s)) {
1519 		case OPLUS_:
1520 			plusnest++;
1521 			break;
1522 		case O_PLUS:
1523 			if (plusnest > maxnest)
1524 				maxnest = plusnest;
1525 			plusnest--;
1526 			break;
1527 		}
1528 	} while (OP(s) != OEND);
1529 	if (plusnest != 0)
1530 		g->iflags |= BAD;
1531 	return(maxnest);
1532 }
1533