xref: /illumos-gate/usr/src/lib/libc/port/regex/regcomp.c (revision fb3b6bbf2c922827ddf076fbf2172d96c58a71de)
1 /*
2  * Copyright 2013 Garrett D'Amore <garrett@damore.org>
3  * Copyright 2010 Nexenta Systems, Inc.  All rights reserved.
4  * Copyright 2012 Milan Jurik. All rights reserved.
5  * Copyright (c) 1992, 1993, 1994 Henry Spencer.
6  * Copyright (c) 1992, 1993, 1994
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Henry Spencer.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include "lint.h"
38 #include "file64.h"
39 #include <sys/types.h>
40 #include <stdio.h>
41 #include <string.h>
42 #include <ctype.h>
43 #include <limits.h>
44 #include <stdlib.h>
45 #include <regex.h>
46 #include <wchar.h>
47 #include <wctype.h>
48 
49 #include "../locale/runetype.h"
50 #include "../locale/collate.h"
51 
52 #include "utils.h"
53 #include "regex2.h"
54 
55 #include "cname.h"
56 #include "../locale/mblocal.h"
57 
58 /*
59  * parse structure, passed up and down to avoid global variables and
60  * other clumsinesses
61  */
62 struct parse {
63 	const char *next;	/* next character in RE */
64 	const char *end;	/* end of string (-> NUL normally) */
65 	int error;		/* has an error been seen? */
66 	sop *strip;		/* malloced strip */
67 	sopno ssize;		/* malloced strip size (allocated) */
68 	sopno slen;		/* malloced strip length (used) */
69 	int ncsalloc;		/* number of csets allocated */
70 	struct re_guts *g;
71 #define	NPAREN	10		/* we need to remember () 1-9 for back refs */
72 	sopno pbegin[NPAREN];	/* -> ( ([0] unused) */
73 	sopno pend[NPAREN];	/* -> ) ([0] unused) */
74 };
75 
76 /* ========= begin header generated by ./mkh ========= */
77 #ifdef __cplusplus
78 extern "C" {
79 #endif
80 
81 /* === regcomp.c === */
82 static void p_ere(struct parse *p, int stop);
83 static void p_ere_exp(struct parse *p);
84 static void p_str(struct parse *p);
85 static void p_bre(struct parse *p, int end1, int end2);
86 static int p_simp_re(struct parse *p, int starordinary);
87 static int p_count(struct parse *p);
88 static void p_bracket(struct parse *p);
89 static void p_b_term(struct parse *p, cset *cs);
90 static void p_b_cclass(struct parse *p, cset *cs);
91 static void p_b_eclass(struct parse *p, cset *cs);
92 static wint_t p_b_symbol(struct parse *p);
93 static wint_t p_b_coll_elem(struct parse *p, wint_t endc);
94 static wint_t othercase(wint_t ch);
95 static void bothcases(struct parse *p, wint_t ch);
96 static void ordinary(struct parse *p, wint_t ch);
97 static void nonnewline(struct parse *p);
98 static void repeat(struct parse *p, sopno start, int from, int to);
99 static int seterr(struct parse *p, int e);
100 static cset *allocset(struct parse *p);
101 static void freeset(struct parse *p, cset *cs);
102 static void CHadd(struct parse *p, cset *cs, wint_t ch);
103 static void CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max);
104 static void CHaddtype(struct parse *p, cset *cs, wctype_t wct);
105 static wint_t singleton(cset *cs);
106 static sopno dupl(struct parse *p, sopno start, sopno finish);
107 static void doemit(struct parse *p, sop op, size_t opnd);
108 static void doinsert(struct parse *p, sop op, size_t opnd, sopno pos);
109 static void dofwd(struct parse *p, sopno pos, sop value);
110 static int enlarge(struct parse *p, sopno size);
111 static void stripsnug(struct parse *p, struct re_guts *g);
112 static void findmust(struct parse *p, struct re_guts *g);
113 static int altoffset(sop *scan, int offset);
114 static void computejumps(struct parse *p, struct re_guts *g);
115 static void computematchjumps(struct parse *p, struct re_guts *g);
116 static sopno pluscount(struct parse *p, struct re_guts *g);
117 static wint_t wgetnext(struct parse *p);
118 
119 #ifdef __cplusplus
120 }
121 #endif
122 /* ========= end header generated by ./mkh ========= */
123 
124 static char nuls[10];		/* place to point scanner in event of error */
125 
126 /*
127  * macros for use with parse structure
128  * BEWARE:  these know that the parse structure is named `p' !!!
129  */
130 #define	PEEK()	(*p->next)
131 #define	PEEK2()	(*(p->next+1))
132 #define	MORE()	(p->next < p->end)
133 #define	MORE2()	(p->next+1 < p->end)
134 #define	SEE(c)	(MORE() && PEEK() == (c))
135 #define	SEETWO(a, b)	(MORE() && MORE2() && PEEK() == (a) && PEEK2() == (b))
136 #define	EAT(c)	((SEE(c)) ? (NEXT(), 1) : 0)
137 #define	EATTWO(a, b)	((SEETWO(a, b)) ? (NEXT2(), 1) : 0)
138 #define	NEXT()	(p->next++)
139 #define	NEXT2()	(p->next += 2)
140 #define	NEXTn(n)	(p->next += (n))
141 #define	GETNEXT()	(*p->next++)
142 #define	WGETNEXT()	wgetnext(p)
143 #define	SETERROR(e)	((void)seterr(p, (e)))
144 #define	REQUIRE(co, e)	((co) || seterr(p, e))
145 #define	MUSTSEE(c, e)	(REQUIRE(MORE() && PEEK() == (c), e))
146 #define	MUSTEAT(c, e)	(REQUIRE(MORE() && GETNEXT() == (c), e))
147 #define	MUSTNOTSEE(c, e)	(REQUIRE(!MORE() || PEEK() != (c), e))
148 #define	EMIT(op, sopnd)	doemit(p, (sop)(op), (size_t)(sopnd))
149 #define	INSERT(op, pos)	doinsert(p, (sop)(op), HERE()-(pos)+1, pos)
150 #define	AHEAD(pos)		dofwd(p, pos, HERE()-(pos))
151 #define	ASTERN(sop, pos)	EMIT(sop, HERE()-pos)
152 #define	HERE()		(p->slen)
153 #define	THERE()		(p->slen - 1)
154 #define	THERETHERE()	(p->slen - 2)
155 #define	DROP(n)	(p->slen -= (n))
156 
157 #ifndef NDEBUG
158 static int never = 0;		/* for use in asserts; shuts lint up */
159 #else
160 #define	never	0		/* some <assert.h>s have bugs too */
161 #endif
162 
163 /*
164  * regcomp - interface for parser and compilation
165  */
166 int				/* 0 success, otherwise REG_something */
167 regcomp(regex_t *_RESTRICT_KYWD preg, const char *_RESTRICT_KYWD pattern,
168     int cflags)
169 {
170 	struct parse pa;
171 	struct re_guts *g;
172 	struct parse *p = &pa;
173 	int i;
174 	size_t len;
175 	size_t maxlen;
176 #ifdef REDEBUG
177 #define	GOODFLAGS(f)	(f)
178 #else
179 #define	GOODFLAGS(f)	((f)&~REG_DUMP)
180 #endif
181 
182 	/* We had REG_INVARG, but we don't have that on Solaris. */
183 	cflags = GOODFLAGS(cflags);
184 	if ((cflags&REG_EXTENDED) && (cflags&REG_NOSPEC))
185 		return (REG_EFATAL);
186 
187 	if (cflags&REG_PEND) {
188 		if (preg->re_endp < pattern)
189 			return (REG_EFATAL);
190 		len = preg->re_endp - pattern;
191 	} else
192 		len = strlen(pattern);
193 
194 	/* do the mallocs early so failure handling is easy */
195 	g = (struct re_guts *)malloc(sizeof (struct re_guts));
196 	if (g == NULL)
197 		return (REG_ESPACE);
198 	/*
199 	 * Limit the pattern space to avoid a 32-bit overflow on buffer
200 	 * extension.  Also avoid any signed overflow in case of conversion
201 	 * so make the real limit based on a 31-bit overflow.
202 	 *
203 	 * Likely not applicable on 64-bit systems but handle the case
204 	 * generically (who are we to stop people from using ~715MB+
205 	 * patterns?).
206 	 */
207 	maxlen = ((size_t)-1 >> 1) / sizeof (sop) * 2 / 3;
208 	if (len >= maxlen) {
209 		free((char *)g);
210 		return (REG_ESPACE);
211 	}
212 	p->ssize = len/(size_t)2*(size_t)3 + (size_t)1;	/* ugh */
213 	assert(p->ssize >= len);
214 
215 	p->strip = (sop *)malloc(p->ssize * sizeof (sop));
216 	p->slen = 0;
217 	if (p->strip == NULL) {
218 		free((char *)g);
219 		return (REG_ESPACE);
220 	}
221 
222 	/* set things up */
223 	p->g = g;
224 	p->next = pattern;	/* convenience; we do not modify it */
225 	p->end = p->next + len;
226 	p->error = 0;
227 	p->ncsalloc = 0;
228 	for (i = 0; i < NPAREN; i++) {
229 		p->pbegin[i] = 0;
230 		p->pend[i] = 0;
231 	}
232 	g->sets = NULL;
233 	g->ncsets = 0;
234 	g->cflags = cflags;
235 	g->iflags = 0;
236 	g->nbol = 0;
237 	g->neol = 0;
238 	g->must = NULL;
239 	g->moffset = -1;
240 	g->charjump = NULL;
241 	g->matchjump = NULL;
242 	g->mlen = 0;
243 	g->nsub = 0;
244 	g->backrefs = 0;
245 
246 	/* do it */
247 	EMIT(OEND, 0);
248 	g->firststate = THERE();
249 	if (cflags&REG_EXTENDED)
250 		p_ere(p, OUT);
251 	else if (cflags&REG_NOSPEC)
252 		p_str(p);
253 	else
254 		p_bre(p, OUT, OUT);
255 	EMIT(OEND, 0);
256 	g->laststate = THERE();
257 
258 	/* tidy up loose ends and fill things in */
259 	stripsnug(p, g);
260 	findmust(p, g);
261 	/*
262 	 * only use Boyer-Moore algorithm if the pattern is bigger
263 	 * than three characters
264 	 */
265 	if (g->mlen > 3) {
266 		computejumps(p, g);
267 		computematchjumps(p, g);
268 		if (g->matchjump == NULL && g->charjump != NULL) {
269 			free(g->charjump);
270 			g->charjump = NULL;
271 		}
272 	}
273 	g->nplus = pluscount(p, g);
274 	g->magic = MAGIC2;
275 	preg->re_nsub = g->nsub;
276 	preg->re_g = g;
277 	preg->re_magic = MAGIC1;
278 #ifndef REDEBUG
279 	/* not debugging, so can't rely on the assert() in regexec() */
280 	if (g->iflags&BAD)
281 		SETERROR(REG_EFATAL);
282 #endif
283 
284 	/* win or lose, we're done */
285 	if (p->error != 0)	/* lose */
286 		regfree(preg);
287 	return (p->error);
288 }
289 
290 /*
291  * p_ere - ERE parser top level, concatenation and alternation
292  */
293 static void
294 p_ere(struct parse *p,
295     int stop)		/* character this ERE should end at */
296 {
297 	char c;
298 	sopno prevback;
299 	sopno prevfwd;
300 	sopno conc;
301 	int first = 1;		/* is this the first alternative? */
302 
303 	for (;;) {
304 		/* do a bunch of concatenated expressions */
305 		conc = HERE();
306 		while (MORE() && (c = PEEK()) != '|' && c != stop)
307 			p_ere_exp(p);
308 		/* require nonempty */
309 		(void) REQUIRE(HERE() != conc, REG_BADPAT);
310 
311 		if (!EAT('|'))
312 			break;		/* NOTE BREAK OUT */
313 
314 		if (first) {
315 			INSERT(OCH_, conc);	/* offset is wrong */
316 			prevfwd = conc;
317 			prevback = conc;
318 			first = 0;
319 		}
320 		ASTERN(OOR1, prevback);
321 		prevback = THERE();
322 		AHEAD(prevfwd);			/* fix previous offset */
323 		prevfwd = HERE();
324 		EMIT(OOR2, 0);			/* offset is very wrong */
325 	}
326 
327 	if (!first) {		/* tail-end fixups */
328 		AHEAD(prevfwd);
329 		ASTERN(O_CH, prevback);
330 	}
331 
332 	assert(!MORE() || SEE(stop));
333 }
334 
335 /*
336  * p_ere_exp - parse one subERE, an atom possibly followed by a repetition op
337  */
338 static void
339 p_ere_exp(struct parse *p)
340 {
341 	char c;
342 	wint_t wc;
343 	sopno pos;
344 	int count;
345 	int count2;
346 	sopno subno;
347 	int wascaret = 0;
348 
349 	assert(MORE());		/* caller should have ensured this */
350 	c = GETNEXT();
351 
352 	pos = HERE();
353 	switch (c) {
354 	case '(':
355 		(void) REQUIRE(MORE(), REG_EPAREN);
356 		p->g->nsub++;
357 		subno = p->g->nsub;
358 		if (subno < NPAREN)
359 			p->pbegin[subno] = HERE();
360 		EMIT(OLPAREN, subno);
361 		if (!SEE(')'))
362 			p_ere(p, ')');
363 		if (subno < NPAREN) {
364 			p->pend[subno] = HERE();
365 			assert(p->pend[subno] != 0);
366 		}
367 		EMIT(ORPAREN, subno);
368 		(void) MUSTEAT(')', REG_EPAREN);
369 		break;
370 #ifndef POSIX_MISTAKE
371 	case ')':		/* happens only if no current unmatched ( */
372 		/*
373 		 * You may ask, why the ifndef?  Because I didn't notice
374 		 * this until slightly too late for 1003.2, and none of the
375 		 * other 1003.2 regular-expression reviewers noticed it at
376 		 * all.  So an unmatched ) is legal POSIX, at least until
377 		 * we can get it fixed.
378 		 */
379 		SETERROR(REG_EPAREN);
380 		break;
381 #endif
382 	case '^':
383 		EMIT(OBOL, 0);
384 		p->g->iflags |= USEBOL;
385 		p->g->nbol++;
386 		wascaret = 1;
387 		break;
388 	case '$':
389 		EMIT(OEOL, 0);
390 		p->g->iflags |= USEEOL;
391 		p->g->neol++;
392 		break;
393 	case '|':
394 		SETERROR(REG_BADPAT);
395 		break;
396 	case '*':
397 	case '+':
398 	case '?':
399 	case '{':
400 		SETERROR(REG_BADRPT);
401 		break;
402 	case '.':
403 		if (p->g->cflags&REG_NEWLINE)
404 			nonnewline(p);
405 		else
406 			EMIT(OANY, 0);
407 		break;
408 	case '[':
409 		p_bracket(p);
410 		break;
411 	case '\\':
412 		(void) REQUIRE(MORE(), REG_EESCAPE);
413 		wc = WGETNEXT();
414 		switch (wc) {
415 		case '<':
416 			EMIT(OBOW, 0);
417 			break;
418 		case '>':
419 			EMIT(OEOW, 0);
420 			break;
421 		default:
422 			ordinary(p, wc);
423 			break;
424 		}
425 		break;
426 	default:
427 		if (p->error != 0)
428 			return;
429 		p->next--;
430 		wc = WGETNEXT();
431 		ordinary(p, wc);
432 		break;
433 	}
434 
435 	if (!MORE())
436 		return;
437 	c = PEEK();
438 	/* we call { a repetition if followed by a digit */
439 	if (!(c == '*' || c == '+' || c == '?' || c == '{'))
440 		return;		/* no repetition, we're done */
441 	else if (c == '{')
442 		(void) REQUIRE(MORE2() && \
443 		    (isdigit((uch)PEEK2()) || PEEK2() == ','), REG_BADRPT);
444 	NEXT();
445 
446 	(void) REQUIRE(!wascaret, REG_BADRPT);
447 	switch (c) {
448 	case '*':	/* implemented as +? */
449 		/* this case does not require the (y|) trick, noKLUDGE */
450 		INSERT(OPLUS_, pos);
451 		ASTERN(O_PLUS, pos);
452 		INSERT(OQUEST_, pos);
453 		ASTERN(O_QUEST, pos);
454 		break;
455 	case '+':
456 		INSERT(OPLUS_, pos);
457 		ASTERN(O_PLUS, pos);
458 		break;
459 	case '?':
460 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
461 		INSERT(OCH_, pos);		/* offset slightly wrong */
462 		ASTERN(OOR1, pos);		/* this one's right */
463 		AHEAD(pos);			/* fix the OCH_ */
464 		EMIT(OOR2, 0);			/* offset very wrong... */
465 		AHEAD(THERE());			/* ...so fix it */
466 		ASTERN(O_CH, THERETHERE());
467 		break;
468 	case '{':
469 		count = p_count(p);
470 		if (EAT(',')) {
471 			if (isdigit((uch)PEEK())) {
472 				count2 = p_count(p);
473 				(void) REQUIRE(count <= count2, REG_BADBR);
474 			} else		/* single number with comma */
475 				count2 = INFINITY;
476 		} else		/* just a single number */
477 			count2 = count;
478 		repeat(p, pos, count, count2);
479 		if (!EAT('}')) {	/* error heuristics */
480 			while (MORE() && PEEK() != '}')
481 				NEXT();
482 			(void) REQUIRE(MORE(), REG_EBRACE);
483 			SETERROR(REG_BADBR);
484 		}
485 		break;
486 	}
487 
488 	if (!MORE())
489 		return;
490 	c = PEEK();
491 	if (!(c == '*' || c == '+' || c == '?' ||
492 	    (c == '{' && MORE2() && isdigit((uch)PEEK2()))))
493 		return;
494 	SETERROR(REG_BADRPT);
495 }
496 
497 /*
498  * p_str - string (no metacharacters) "parser"
499  */
500 static void
501 p_str(struct parse *p)
502 {
503 	(void) REQUIRE(MORE(), REG_BADPAT);
504 	while (MORE())
505 		ordinary(p, WGETNEXT());
506 }
507 
508 /*
509  * p_bre - BRE parser top level, anchoring and concatenation
510  * Giving end1 as OUT essentially eliminates the end1/end2 check.
511  *
512  * This implementation is a bit of a kludge, in that a trailing $ is first
513  * taken as an ordinary character and then revised to be an anchor.
514  * The amount of lookahead needed to avoid this kludge is excessive.
515  */
516 static void
517 p_bre(struct parse *p,
518     int end1,		/* first terminating character */
519     int end2)		/* second terminating character */
520 {
521 	sopno start = HERE();
522 	int first = 1;			/* first subexpression? */
523 	int wasdollar = 0;
524 
525 	if (EAT('^')) {
526 		EMIT(OBOL, 0);
527 		p->g->iflags |= USEBOL;
528 		p->g->nbol++;
529 	}
530 	while (MORE() && !SEETWO(end1, end2)) {
531 		wasdollar = p_simp_re(p, first);
532 		first = 0;
533 	}
534 	if (wasdollar) {	/* oops, that was a trailing anchor */
535 		DROP(1);
536 		EMIT(OEOL, 0);
537 		p->g->iflags |= USEEOL;
538 		p->g->neol++;
539 	}
540 
541 	(void) REQUIRE(HERE() != start, REG_BADPAT);	/* require nonempty */
542 }
543 
544 /*
545  * p_simp_re - parse a simple RE, an atom possibly followed by a repetition
546  */
547 static int			/* was the simple RE an unbackslashed $? */
548 p_simp_re(struct parse *p,
549     int starordinary)	/* is a leading * an ordinary character? */
550 {
551 	int c;
552 	int count;
553 	int count2;
554 	sopno pos;
555 	int i;
556 	wint_t wc;
557 	sopno subno;
558 #define	BACKSL	(1<<CHAR_BIT)
559 
560 	pos = HERE();		/* repetition op, if any, covers from here */
561 
562 	assert(MORE());		/* caller should have ensured this */
563 	c = GETNEXT();
564 	if (c == '\\') {
565 		(void) REQUIRE(MORE(), REG_EESCAPE);
566 		c = BACKSL | GETNEXT();
567 	}
568 	switch (c) {
569 	case '.':
570 		if (p->g->cflags&REG_NEWLINE)
571 			nonnewline(p);
572 		else
573 			EMIT(OANY, 0);
574 		break;
575 	case '[':
576 		p_bracket(p);
577 		break;
578 	case BACKSL|'<':
579 		EMIT(OBOW, 0);
580 		break;
581 	case BACKSL|'>':
582 		EMIT(OEOW, 0);
583 		break;
584 	case BACKSL|'{':
585 		SETERROR(REG_BADRPT);
586 		break;
587 	case BACKSL|'(':
588 		p->g->nsub++;
589 		subno = p->g->nsub;
590 		if (subno < NPAREN)
591 			p->pbegin[subno] = HERE();
592 		EMIT(OLPAREN, subno);
593 		/* the MORE here is an error heuristic */
594 		if (MORE() && !SEETWO('\\', ')'))
595 			p_bre(p, '\\', ')');
596 		if (subno < NPAREN) {
597 			p->pend[subno] = HERE();
598 			assert(p->pend[subno] != 0);
599 		}
600 		EMIT(ORPAREN, subno);
601 		(void) REQUIRE(EATTWO('\\', ')'), REG_EPAREN);
602 		break;
603 	case BACKSL|')':	/* should not get here -- must be user */
604 		SETERROR(REG_EPAREN);
605 		break;
606 	case BACKSL|'1':
607 	case BACKSL|'2':
608 	case BACKSL|'3':
609 	case BACKSL|'4':
610 	case BACKSL|'5':
611 	case BACKSL|'6':
612 	case BACKSL|'7':
613 	case BACKSL|'8':
614 	case BACKSL|'9':
615 		i = (c&~BACKSL) - '0';
616 		assert(i < NPAREN);
617 		if (p->pend[i] != 0) {
618 			assert(i <= p->g->nsub);
619 			EMIT(OBACK_, i);
620 			assert(p->pbegin[i] != 0);
621 			assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
622 			assert(OP(p->strip[p->pend[i]]) == ORPAREN);
623 			(void) dupl(p, p->pbegin[i]+1, p->pend[i]);
624 			EMIT(O_BACK, i);
625 		} else
626 			SETERROR(REG_ESUBREG);
627 		p->g->backrefs = 1;
628 		break;
629 	case '*':
630 		(void) REQUIRE(starordinary, REG_BADRPT);
631 		/* FALLTHROUGH */
632 	default:
633 		if (p->error != 0)
634 			return (0);	/* Definitely not $... */
635 		p->next--;
636 		wc = WGETNEXT();
637 		ordinary(p, wc);
638 		break;
639 	}
640 
641 	if (EAT('*')) {		/* implemented as +? */
642 		/* this case does not require the (y|) trick, noKLUDGE */
643 		INSERT(OPLUS_, pos);
644 		ASTERN(O_PLUS, pos);
645 		INSERT(OQUEST_, pos);
646 		ASTERN(O_QUEST, pos);
647 	} else if (EATTWO('\\', '{')) {
648 		count = p_count(p);
649 		if (EAT(',')) {
650 			if (MORE() && isdigit((uch)PEEK())) {
651 				count2 = p_count(p);
652 				(void) REQUIRE(count <= count2, REG_BADBR);
653 			} else		/* single number with comma */
654 				count2 = INFINITY;
655 		} else		/* just a single number */
656 			count2 = count;
657 		repeat(p, pos, count, count2);
658 		if (!EATTWO('\\', '}')) {	/* error heuristics */
659 			while (MORE() && !SEETWO('\\', '}'))
660 				NEXT();
661 			(void) REQUIRE(MORE(), REG_EBRACE);
662 			SETERROR(REG_BADBR);
663 		}
664 	} else if (c == '$')	/* $ (but not \$) ends it */
665 		return (1);
666 
667 	return (0);
668 }
669 
670 /*
671  * p_count - parse a repetition count
672  */
673 static int			/* the value */
674 p_count(struct parse *p)
675 {
676 	int count = 0;
677 	int ndigits = 0;
678 
679 	while (MORE() && isdigit((uch)PEEK()) && count <= DUPMAX) {
680 		count = count*10 + (GETNEXT() - '0');
681 		ndigits++;
682 	}
683 
684 	(void) REQUIRE(ndigits > 0 && count <= DUPMAX, REG_BADBR);
685 	return (count);
686 }
687 
688 /*
689  * p_bracket - parse a bracketed character list
690  */
691 static void
692 p_bracket(struct parse *p)
693 {
694 	cset *cs;
695 	wint_t ch;
696 
697 	/* Dept of Truly Sickening Special-Case Kludges */
698 	if (p->next + 5 < p->end && strncmp(p->next, "[:<:]]", 6) == 0) {
699 		EMIT(OBOW, 0);
700 		NEXTn(6);
701 		return;
702 	}
703 	if (p->next + 5 < p->end && strncmp(p->next, "[:>:]]", 6) == 0) {
704 		EMIT(OEOW, 0);
705 		NEXTn(6);
706 		return;
707 	}
708 
709 	if ((cs = allocset(p)) == NULL)
710 		return;
711 
712 	if (p->g->cflags&REG_ICASE)
713 		cs->icase = 1;
714 	if (EAT('^'))
715 		cs->invert = 1;
716 	if (EAT(']'))
717 		CHadd(p, cs, ']');
718 	else if (EAT('-'))
719 		CHadd(p, cs, '-');
720 	while (MORE() && PEEK() != ']' && !SEETWO('-', ']'))
721 		p_b_term(p, cs);
722 	if (EAT('-'))
723 		CHadd(p, cs, '-');
724 	(void) MUSTEAT(']', REG_EBRACK);
725 
726 	if (p->error != 0)	/* don't mess things up further */
727 		return;
728 
729 	if (cs->invert && p->g->cflags&REG_NEWLINE)
730 		cs->bmp['\n' >> 3] |= 1 << ('\n' & 7);
731 
732 	if ((ch = singleton(cs)) != OUT) {	/* optimize singleton sets */
733 		ordinary(p, ch);
734 		freeset(p, cs);
735 	} else
736 		EMIT(OANYOF, (int)(cs - p->g->sets));
737 }
738 
739 /*
740  * p_b_term - parse one term of a bracketed character list
741  */
742 static void
743 p_b_term(struct parse *p, cset *cs)
744 {
745 	char c;
746 	wint_t start, finish;
747 	wint_t i;
748 	locale_t loc = uselocale(NULL);
749 
750 	/* classify what we've got */
751 	switch ((MORE()) ? PEEK() : '\0') {
752 	case '[':
753 		c = (MORE2()) ? PEEK2() : '\0';
754 		break;
755 	case '-':
756 		SETERROR(REG_ERANGE);
757 		return;			/* NOTE RETURN */
758 	default:
759 		c = '\0';
760 		break;
761 	}
762 
763 	switch (c) {
764 	case ':':		/* character class */
765 		NEXT2();
766 		(void) REQUIRE(MORE(), REG_EBRACK);
767 		c = PEEK();
768 		(void) REQUIRE(c != '-' && c != ']', REG_ECTYPE);
769 		p_b_cclass(p, cs);
770 		(void) REQUIRE(MORE(), REG_EBRACK);
771 		(void) REQUIRE(EATTWO(':', ']'), REG_ECTYPE);
772 		break;
773 	case '=':		/* equivalence class */
774 		NEXT2();
775 		(void) REQUIRE(MORE(), REG_EBRACK);
776 		c = PEEK();
777 		(void) REQUIRE(c != '-' && c != ']', REG_ECOLLATE);
778 		p_b_eclass(p, cs);
779 		(void) REQUIRE(MORE(), REG_EBRACK);
780 		(void) REQUIRE(EATTWO('=', ']'), REG_ECOLLATE);
781 		break;
782 	default:		/* symbol, ordinary character, or range */
783 		start = p_b_symbol(p);
784 		if (SEE('-') && MORE2() && PEEK2() != ']') {
785 			/* range */
786 			NEXT();
787 			if (EAT('-'))
788 				finish = '-';
789 			else
790 				finish = p_b_symbol(p);
791 		} else
792 			finish = start;
793 		if (start == finish)
794 			CHadd(p, cs, start);
795 		else {
796 			if (loc->collate->lc_is_posix) {
797 				(void) REQUIRE((uch)start <= (uch)finish,
798 				    REG_ERANGE);
799 				CHaddrange(p, cs, start, finish);
800 			} else {
801 				(void) REQUIRE(_collate_range_cmp(start,
802 				    finish, loc) <= 0, REG_ERANGE);
803 				for (i = 0; i <= UCHAR_MAX; i++) {
804 					if (_collate_range_cmp(start, i, loc)
805 					    <= 0 &&
806 					    _collate_range_cmp(i, finish, loc)
807 					    <= 0)
808 						CHadd(p, cs, i);
809 				}
810 			}
811 		}
812 		break;
813 	}
814 }
815 
816 /*
817  * p_b_cclass - parse a character-class name and deal with it
818  */
819 static void
820 p_b_cclass(struct parse *p, cset *cs)
821 {
822 	const char *sp = p->next;
823 	size_t len;
824 	wctype_t wct;
825 	char clname[16];
826 
827 	while (MORE() && isalpha((uch)PEEK()))
828 		NEXT();
829 	len = p->next - sp;
830 	if (len >= sizeof (clname) - 1) {
831 		SETERROR(REG_ECTYPE);
832 		return;
833 	}
834 	(void) memcpy(clname, sp, len);
835 	clname[len] = '\0';
836 	if ((wct = wctype(clname)) == 0) {
837 		SETERROR(REG_ECTYPE);
838 		return;
839 	}
840 	CHaddtype(p, cs, wct);
841 }
842 
843 /*
844  * p_b_eclass - parse an equivalence-class name and deal with it
845  *
846  * This implementation is incomplete. xxx
847  */
848 static void
849 p_b_eclass(struct parse *p, cset *cs)
850 {
851 	wint_t c;
852 
853 	c = p_b_coll_elem(p, '=');
854 	CHadd(p, cs, c);
855 }
856 
857 /*
858  * p_b_symbol - parse a character or [..]ed multicharacter collating symbol
859  */
860 static wint_t			/* value of symbol */
861 p_b_symbol(struct parse *p)
862 {
863 	wint_t value;
864 
865 	(void) REQUIRE(MORE(), REG_EBRACK);
866 	if (!EATTWO('[', '.'))
867 		return (WGETNEXT());
868 
869 	/* collating symbol */
870 	value = p_b_coll_elem(p, '.');
871 	(void) REQUIRE(EATTWO('.', ']'), REG_ECOLLATE);
872 	return (value);
873 }
874 
875 /*
876  * p_b_coll_elem - parse a collating-element name and look it up
877  */
878 static wint_t			/* value of collating element */
879 p_b_coll_elem(struct parse *p,
880     wint_t endc)		/* name ended by endc,']' */
881 {
882 	const char *sp = p->next;
883 	struct cname *cp;
884 	mbstate_t mbs;
885 	wchar_t wc;
886 	size_t clen, len;
887 
888 	while (MORE() && !SEETWO(endc, ']'))
889 		NEXT();
890 	if (!MORE()) {
891 		SETERROR(REG_EBRACK);
892 		return (0);
893 	}
894 	len = p->next - sp;
895 	for (cp = cnames; cp->name != NULL; cp++)
896 		if (strncmp(cp->name, sp, len) == 0 && cp->name[len] == '\0')
897 			return (cp->code);	/* known name */
898 	(void) memset(&mbs, 0, sizeof (mbs));
899 	if ((clen = mbrtowc(&wc, sp, len, &mbs)) == len)
900 		return (wc);			/* single character */
901 	else if (clen == (size_t)-1 || clen == (size_t)-2)
902 		SETERROR(REG_ECHAR);
903 	else
904 		SETERROR(REG_ECOLLATE);		/* neither */
905 	return (0);
906 }
907 
908 /*
909  * othercase - return the case counterpart of an alphabetic
910  */
911 static wint_t			/* if no counterpart, return ch */
912 othercase(wint_t ch)
913 {
914 	assert(iswalpha(ch));
915 	if (iswupper(ch))
916 		return (towlower(ch));
917 	else if (iswlower(ch))
918 		return (towupper(ch));
919 	else			/* peculiar, but could happen */
920 		return (ch);
921 }
922 
923 /*
924  * bothcases - emit a dualcase version of a two-case character
925  *
926  * Boy, is this implementation ever a kludge...
927  */
928 static void
929 bothcases(struct parse *p, wint_t ch)
930 {
931 	const char *oldnext = p->next;
932 	const char *oldend = p->end;
933 	char bracket[3 + MB_LEN_MAX];
934 	size_t n;
935 	mbstate_t mbs;
936 
937 	assert(othercase(ch) != ch);	/* p_bracket() would recurse */
938 	p->next = bracket;
939 	(void) memset(&mbs, 0, sizeof (mbs));
940 	n = wcrtomb(bracket, ch, &mbs);
941 	assert(n != (size_t)-1);
942 	bracket[n] = ']';
943 	bracket[n + 1] = '\0';
944 	p->end = bracket+n+1;
945 	p_bracket(p);
946 	assert(p->next == p->end);
947 	p->next = oldnext;
948 	p->end = oldend;
949 }
950 
951 /*
952  * ordinary - emit an ordinary character
953  */
954 static void
955 ordinary(struct parse *p, wint_t ch)
956 {
957 	cset *cs;
958 
959 	if ((p->g->cflags&REG_ICASE) && iswalpha(ch) && othercase(ch) != ch)
960 		bothcases(p, ch);
961 	else if ((ch & OPDMASK) == ch)
962 		EMIT(OCHAR, ch);
963 	else {
964 		/*
965 		 * Kludge: character is too big to fit into an OCHAR operand.
966 		 * Emit a singleton set.
967 		 */
968 		if ((cs = allocset(p)) == NULL)
969 			return;
970 		CHadd(p, cs, ch);
971 		EMIT(OANYOF, (int)(cs - p->g->sets));
972 	}
973 }
974 
975 /*
976  * nonnewline - emit REG_NEWLINE version of OANY
977  *
978  * Boy, is this implementation ever a kludge...
979  */
980 static void
981 nonnewline(struct parse *p)
982 {
983 	const char *oldnext = p->next;
984 	const char *oldend = p->end;
985 	char bracket[4];
986 
987 	p->next = bracket;
988 	p->end = bracket+3;
989 	bracket[0] = '^';
990 	bracket[1] = '\n';
991 	bracket[2] = ']';
992 	bracket[3] = '\0';
993 	p_bracket(p);
994 	assert(p->next == bracket+3);
995 	p->next = oldnext;
996 	p->end = oldend;
997 }
998 
999 /*
1000  * repeat - generate code for a bounded repetition, recursively if needed
1001  */
1002 static void
1003 repeat(struct parse *p,
1004     sopno start,		/* operand from here to end of strip */
1005     int from,			/* repeated from this number */
1006     int to)			/* to this number of times (maybe INFINITY) */
1007 {
1008 	sopno finish = HERE();
1009 #define	N	2
1010 #define	INF	3
1011 #define	REP(f, t)	((f)*8 + (t))
1012 #define	MAP(n)	(((n) <= 1) ? (n) : ((n) == INFINITY) ? INF : N)
1013 	sopno copy;
1014 
1015 	if (p->error != 0)	/* head off possible runaway recursion */
1016 		return;
1017 
1018 	assert(from <= to);
1019 
1020 	switch (REP(MAP(from), MAP(to))) {
1021 	case REP(0, 0):			/* must be user doing this */
1022 		DROP(finish-start);	/* drop the operand */
1023 		break;
1024 	case REP(0, 1):			/* as x{1,1}? */
1025 	case REP(0, N):			/* as x{1,n}? */
1026 	case REP(0, INF):		/* as x{1,}? */
1027 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1028 		INSERT(OCH_, start);		/* offset is wrong... */
1029 		repeat(p, start+1, 1, to);
1030 		ASTERN(OOR1, start);
1031 		AHEAD(start);			/* ... fix it */
1032 		EMIT(OOR2, 0);
1033 		AHEAD(THERE());
1034 		ASTERN(O_CH, THERETHERE());
1035 		break;
1036 	case REP(1, 1):			/* trivial case */
1037 		/* done */
1038 		break;
1039 	case REP(1, N):			/* as x?x{1,n-1} */
1040 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1041 		INSERT(OCH_, start);
1042 		ASTERN(OOR1, start);
1043 		AHEAD(start);
1044 		EMIT(OOR2, 0);			/* offset very wrong... */
1045 		AHEAD(THERE());			/* ...so fix it */
1046 		ASTERN(O_CH, THERETHERE());
1047 		copy = dupl(p, start+1, finish+1);
1048 		assert(copy == finish+4);
1049 		repeat(p, copy, 1, to-1);
1050 		break;
1051 	case REP(1, INF):		/* as x+ */
1052 		INSERT(OPLUS_, start);
1053 		ASTERN(O_PLUS, start);
1054 		break;
1055 	case REP(N, N):			/* as xx{m-1,n-1} */
1056 		copy = dupl(p, start, finish);
1057 		repeat(p, copy, from-1, to-1);
1058 		break;
1059 	case REP(N, INF):		/* as xx{n-1,INF} */
1060 		copy = dupl(p, start, finish);
1061 		repeat(p, copy, from-1, to);
1062 		break;
1063 	default:			/* "can't happen" */
1064 		SETERROR(REG_EFATAL);	/* just in case */
1065 		break;
1066 	}
1067 }
1068 
1069 /*
1070  * wgetnext - helper function for WGETNEXT() macro. Gets the next wide
1071  * character from the parse struct, signals a REG_ILLSEQ error if the
1072  * character can't be converted. Returns the number of bytes consumed.
1073  */
1074 static wint_t
1075 wgetnext(struct parse *p)
1076 {
1077 	mbstate_t mbs;
1078 	wchar_t wc;
1079 	size_t n;
1080 
1081 	(void) memset(&mbs, 0, sizeof (mbs));
1082 	n = mbrtowc(&wc, p->next, p->end - p->next, &mbs);
1083 	if (n == (size_t)-1 || n == (size_t)-2) {
1084 		SETERROR(REG_ECHAR);
1085 		return (0);
1086 	}
1087 	if (n == 0)
1088 		n = 1;
1089 	p->next += n;
1090 	return (wc);
1091 }
1092 
1093 /*
1094  * seterr - set an error condition
1095  */
1096 static int			/* useless but makes type checking happy */
1097 seterr(struct parse *p, int e)
1098 {
1099 	if (p->error == 0)	/* keep earliest error condition */
1100 		p->error = e;
1101 	p->next = nuls;		/* try to bring things to a halt */
1102 	p->end = nuls;
1103 	return (0);		/* make the return value well-defined */
1104 }
1105 
1106 /*
1107  * allocset - allocate a set of characters for []
1108  */
1109 static cset *
1110 allocset(struct parse *p)
1111 {
1112 	cset *cs, *ncs;
1113 
1114 	ncs = realloc(p->g->sets, (p->g->ncsets + 1) * sizeof (*ncs));
1115 	if (ncs == NULL) {
1116 		SETERROR(REG_ESPACE);
1117 		return (NULL);
1118 	}
1119 	p->g->sets = ncs;
1120 	cs = &p->g->sets[p->g->ncsets++];
1121 	(void) memset(cs, 0, sizeof (*cs));
1122 
1123 	return (cs);
1124 }
1125 
1126 /*
1127  * freeset - free a now-unused set
1128  */
1129 static void
1130 freeset(struct parse *p, cset *cs)
1131 {
1132 	cset *top = &p->g->sets[p->g->ncsets];
1133 
1134 	free(cs->wides);
1135 	free(cs->ranges);
1136 	free(cs->types);
1137 	(void) memset(cs, 0, sizeof (*cs));
1138 	if (cs == top-1)	/* recover only the easy case */
1139 		p->g->ncsets--;
1140 }
1141 
1142 /*
1143  * singleton - Determine whether a set contains only one character,
1144  * returning it if so, otherwise returning OUT.
1145  */
1146 static wint_t
1147 singleton(cset *cs)
1148 {
1149 	wint_t i, s, n;
1150 
1151 	for (i = n = 0; i < NC; i++)
1152 		if (CHIN(cs, i)) {
1153 			n++;
1154 			s = i;
1155 		}
1156 	if (n == 1)
1157 		return (s);
1158 	if (cs->nwides == 1 && cs->nranges == 0 && cs->ntypes == 0 &&
1159 	    cs->icase == 0)
1160 		return (cs->wides[0]);
1161 	/* Don't bother handling the other cases. */
1162 	return (OUT);
1163 }
1164 
1165 /*
1166  * CHadd - add character to character set.
1167  */
1168 static void
1169 CHadd(struct parse *p, cset *cs, wint_t ch)
1170 {
1171 	wint_t nch, *newwides;
1172 	assert(ch >= 0);
1173 	if (ch < NC)
1174 		cs->bmp[ch >> 3] |= 1 << (ch & 7);
1175 	else {
1176 		newwides = realloc(cs->wides, (cs->nwides + 1) *
1177 		    sizeof (*cs->wides));
1178 		if (newwides == NULL) {
1179 			SETERROR(REG_ESPACE);
1180 			return;
1181 		}
1182 		cs->wides = newwides;
1183 		cs->wides[cs->nwides++] = ch;
1184 	}
1185 	if (cs->icase) {
1186 		if ((nch = towlower(ch)) < NC)
1187 			cs->bmp[nch >> 3] |= 1 << (nch & 7);
1188 		if ((nch = towupper(ch)) < NC)
1189 			cs->bmp[nch >> 3] |= 1 << (nch & 7);
1190 	}
1191 }
1192 
1193 /*
1194  * CHaddrange - add all characters in the range [min,max] to a character set.
1195  */
1196 static void
1197 CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max)
1198 {
1199 	crange *newranges;
1200 
1201 	for (; min < NC && min <= max; min++)
1202 		CHadd(p, cs, min);
1203 	if (min >= max)
1204 		return;
1205 	newranges = realloc(cs->ranges, (cs->nranges + 1) *
1206 	    sizeof (*cs->ranges));
1207 	if (newranges == NULL) {
1208 		SETERROR(REG_ESPACE);
1209 		return;
1210 	}
1211 	cs->ranges = newranges;
1212 	cs->ranges[cs->nranges].min = min;
1213 	cs->ranges[cs->nranges].max = max;
1214 	cs->nranges++;
1215 }
1216 
1217 /*
1218  * CHaddtype - add all characters of a certain type to a character set.
1219  */
1220 static void
1221 CHaddtype(struct parse *p, cset *cs, wctype_t wct)
1222 {
1223 	wint_t i;
1224 	wctype_t *newtypes;
1225 
1226 	for (i = 0; i < NC; i++)
1227 		if (iswctype(i, wct))
1228 			CHadd(p, cs, i);
1229 	newtypes = realloc(cs->types, (cs->ntypes + 1) *
1230 	    sizeof (*cs->types));
1231 	if (newtypes == NULL) {
1232 		SETERROR(REG_ESPACE);
1233 		return;
1234 	}
1235 	cs->types = newtypes;
1236 	cs->types[cs->ntypes++] = wct;
1237 }
1238 
1239 /*
1240  * dupl - emit a duplicate of a bunch of sops
1241  */
1242 static sopno			/* start of duplicate */
1243 dupl(struct parse *p,
1244     sopno start,		/* from here */
1245     sopno finish)		/* to this less one */
1246 {
1247 	sopno ret = HERE();
1248 	sopno len = finish - start;
1249 
1250 	assert(finish >= start);
1251 	if (len == 0)
1252 		return (ret);
1253 	if (!enlarge(p, p->ssize + len)) /* this many unexpected additions */
1254 		return (ret);
1255 	assert(p->ssize >= p->slen + len);
1256 	(void) memcpy((char *)(p->strip + p->slen),
1257 	    (char *)(p->strip + start), (size_t)len*sizeof (sop));
1258 	p->slen += len;
1259 	return (ret);
1260 }
1261 
1262 /*
1263  * doemit - emit a strip operator
1264  *
1265  * It might seem better to implement this as a macro with a function as
1266  * hard-case backup, but it's just too big and messy unless there are
1267  * some changes to the data structures.  Maybe later.
1268  */
1269 static void
1270 doemit(struct parse *p, sop op, size_t opnd)
1271 {
1272 	/* avoid making error situations worse */
1273 	if (p->error != 0)
1274 		return;
1275 
1276 	/* deal with oversize operands ("can't happen", more or less) */
1277 	assert(opnd < 1<<OPSHIFT);
1278 
1279 	/* deal with undersized strip */
1280 	if (p->slen >= p->ssize)
1281 		if (!enlarge(p, (p->ssize+1) / 2 * 3))	/* +50% */
1282 			return;
1283 
1284 	/* finally, it's all reduced to the easy case */
1285 	p->strip[p->slen++] = SOP(op, opnd);
1286 }
1287 
1288 /*
1289  * doinsert - insert a sop into the strip
1290  */
1291 static void
1292 doinsert(struct parse *p, sop op, size_t opnd, sopno pos)
1293 {
1294 	sopno sn;
1295 	sop s;
1296 	int i;
1297 
1298 	/* avoid making error situations worse */
1299 	if (p->error != 0)
1300 		return;
1301 
1302 	sn = HERE();
1303 	EMIT(op, opnd);		/* do checks, ensure space */
1304 	assert(HERE() == sn+1);
1305 	s = p->strip[sn];
1306 
1307 	/* adjust paren pointers */
1308 	assert(pos > 0);
1309 	for (i = 1; i < NPAREN; i++) {
1310 		if (p->pbegin[i] >= pos) {
1311 			p->pbegin[i]++;
1312 		}
1313 		if (p->pend[i] >= pos) {
1314 			p->pend[i]++;
1315 		}
1316 	}
1317 
1318 	(void) memmove((char *)&p->strip[pos+1], (char *)&p->strip[pos],
1319 	    (HERE()-pos-1)*sizeof (sop));
1320 	p->strip[pos] = s;
1321 }
1322 
1323 /*
1324  * dofwd - complete a forward reference
1325  */
1326 static void
1327 dofwd(struct parse *p, sopno pos, sop value)
1328 {
1329 	/* avoid making error situations worse */
1330 	if (p->error != 0)
1331 		return;
1332 
1333 	assert(value < 1<<OPSHIFT);
1334 	p->strip[pos] = OP(p->strip[pos]) | value;
1335 }
1336 
1337 /*
1338  * enlarge - enlarge the strip
1339  */
1340 static int
1341 enlarge(struct parse *p, sopno size)
1342 {
1343 	sop *sp;
1344 
1345 	if (p->ssize >= size)
1346 		return (1);
1347 
1348 	sp = (sop *)realloc(p->strip, size*sizeof (sop));
1349 	if (sp == NULL) {
1350 		SETERROR(REG_ESPACE);
1351 		return (0);
1352 	}
1353 	p->strip = sp;
1354 	p->ssize = size;
1355 	return (1);
1356 }
1357 
1358 /*
1359  * stripsnug - compact the strip
1360  */
1361 static void
1362 stripsnug(struct parse *p, struct re_guts *g)
1363 {
1364 	g->nstates = p->slen;
1365 	g->strip = (sop *)realloc((char *)p->strip, p->slen * sizeof (sop));
1366 	if (g->strip == NULL) {
1367 		SETERROR(REG_ESPACE);
1368 		g->strip = p->strip;
1369 	}
1370 }
1371 
1372 /*
1373  * findmust - fill in must and mlen with longest mandatory literal string
1374  *
1375  * This algorithm could do fancy things like analyzing the operands of |
1376  * for common subsequences.  Someday.  This code is simple and finds most
1377  * of the interesting cases.
1378  *
1379  * Note that must and mlen got initialized during setup.
1380  */
1381 static void
1382 findmust(struct parse *p, struct re_guts *g)
1383 {
1384 	sop *scan;
1385 	sop *start = NULL;
1386 	sop *newstart = NULL;
1387 	sopno newlen;
1388 	sop s;
1389 	char *cp;
1390 	int offset;
1391 	char buf[MB_LEN_MAX];
1392 	size_t clen;
1393 	mbstate_t mbs;
1394 	locale_t loc = uselocale(NULL);
1395 
1396 	/* avoid making error situations worse */
1397 	if (p->error != 0)
1398 		return;
1399 
1400 	/*
1401 	 * It's not generally safe to do a ``char'' substring search on
1402 	 * multibyte character strings, but it's safe for at least
1403 	 * UTF-8 (see RFC 3629).
1404 	 */
1405 	if (MB_CUR_MAX > 1 &&
1406 	    strcmp(loc->runelocale->__encoding, "UTF-8") != 0)
1407 		return;
1408 
1409 	/* find the longest OCHAR sequence in strip */
1410 	newlen = 0;
1411 	offset = 0;
1412 	g->moffset = 0;
1413 	scan = g->strip + 1;
1414 	do {
1415 		s = *scan++;
1416 		switch (OP(s)) {
1417 		case OCHAR:		/* sequence member */
1418 			if (newlen == 0) {		/* new sequence */
1419 				(void) memset(&mbs, 0, sizeof (mbs));
1420 				newstart = scan - 1;
1421 			}
1422 			clen = wcrtomb(buf, OPND(s), &mbs);
1423 			if (clen == (size_t)-1)
1424 				goto toohard;
1425 			newlen += clen;
1426 			break;
1427 		case OPLUS_:		/* things that don't break one */
1428 		case OLPAREN:
1429 		case ORPAREN:
1430 			break;
1431 		case OQUEST_:		/* things that must be skipped */
1432 		case OCH_:
1433 			offset = altoffset(scan, offset);
1434 			scan--;
1435 			do {
1436 				scan += OPND(s);
1437 				s = *scan;
1438 				/* assert() interferes w debug printouts */
1439 				if (OP(s) != O_QUEST && OP(s) != O_CH &&
1440 				    OP(s) != OOR2) {
1441 					g->iflags |= BAD;
1442 					return;
1443 				}
1444 			} while (OP(s) != O_QUEST && OP(s) != O_CH);
1445 			/* FALLTHROUGH */
1446 		case OBOW:		/* things that break a sequence */
1447 		case OEOW:
1448 		case OBOL:
1449 		case OEOL:
1450 		case O_QUEST:
1451 		case O_CH:
1452 		case OEND:
1453 			if (newlen > g->mlen) {		/* ends one */
1454 				start = newstart;
1455 				g->mlen = newlen;
1456 				if (offset > -1) {
1457 					g->moffset += offset;
1458 					offset = newlen;
1459 				} else
1460 					g->moffset = offset;
1461 			} else {
1462 				if (offset > -1)
1463 					offset += newlen;
1464 			}
1465 			newlen = 0;
1466 			break;
1467 		case OANY:
1468 			if (newlen > g->mlen) {		/* ends one */
1469 				start = newstart;
1470 				g->mlen = newlen;
1471 				if (offset > -1) {
1472 					g->moffset += offset;
1473 					offset = newlen;
1474 				} else
1475 					g->moffset = offset;
1476 			} else {
1477 				if (offset > -1)
1478 					offset += newlen;
1479 			}
1480 			if (offset > -1)
1481 				offset++;
1482 			newlen = 0;
1483 			break;
1484 		case OANYOF:		/* may or may not invalidate offset */
1485 			/* First, everything as OANY */
1486 			if (newlen > g->mlen) {		/* ends one */
1487 				start = newstart;
1488 				g->mlen = newlen;
1489 				if (offset > -1) {
1490 					g->moffset += offset;
1491 					offset = newlen;
1492 				} else
1493 					g->moffset = offset;
1494 			} else {
1495 				if (offset > -1)
1496 					offset += newlen;
1497 			}
1498 			if (offset > -1)
1499 				offset++;
1500 			newlen = 0;
1501 			break;
1502 		toohard:
1503 		default:
1504 			/*
1505 			 * Anything here makes it impossible or too hard
1506 			 * to calculate the offset -- so we give up;
1507 			 * save the last known good offset, in case the
1508 			 * must sequence doesn't occur later.
1509 			 */
1510 			if (newlen > g->mlen) {		/* ends one */
1511 				start = newstart;
1512 				g->mlen = newlen;
1513 				if (offset > -1)
1514 					g->moffset += offset;
1515 				else
1516 					g->moffset = offset;
1517 			}
1518 			offset = -1;
1519 			newlen = 0;
1520 			break;
1521 		}
1522 	} while (OP(s) != OEND);
1523 
1524 	if (g->mlen == 0) {		/* there isn't one */
1525 		g->moffset = -1;
1526 		return;
1527 	}
1528 
1529 	/* turn it into a character string */
1530 	g->must = malloc((size_t)g->mlen + 1);
1531 	if (g->must == NULL) {		/* argh; just forget it */
1532 		g->mlen = 0;
1533 		g->moffset = -1;
1534 		return;
1535 	}
1536 	cp = g->must;
1537 	scan = start;
1538 	(void) memset(&mbs, 0, sizeof (mbs));
1539 	while (cp < g->must + g->mlen) {
1540 		while (OP(s = *scan++) != OCHAR)
1541 			continue;
1542 		clen = wcrtomb(cp, OPND(s), &mbs);
1543 		assert(clen != (size_t)-1);
1544 		cp += clen;
1545 	}
1546 	assert(cp == g->must + g->mlen);
1547 	*cp++ = '\0';		/* just on general principles */
1548 }
1549 
1550 /*
1551  * altoffset - choose biggest offset among multiple choices
1552  *
1553  * Compute, recursively if necessary, the largest offset among multiple
1554  * re paths.
1555  */
1556 static int
1557 altoffset(sop *scan, int offset)
1558 {
1559 	int largest;
1560 	int try;
1561 	sop s;
1562 
1563 	/* If we gave up already on offsets, return */
1564 	if (offset == -1)
1565 		return (-1);
1566 
1567 	largest = 0;
1568 	try = 0;
1569 	s = *scan++;
1570 	while (OP(s) != O_QUEST && OP(s) != O_CH) {
1571 		switch (OP(s)) {
1572 		case OOR1:
1573 			if (try > largest)
1574 				largest = try;
1575 			try = 0;
1576 			break;
1577 		case OQUEST_:
1578 		case OCH_:
1579 			try = altoffset(scan, try);
1580 			if (try == -1)
1581 				return (-1);
1582 			scan--;
1583 			do {
1584 				scan += OPND(s);
1585 				s = *scan;
1586 				if (OP(s) != O_QUEST && OP(s) != O_CH &&
1587 				    OP(s) != OOR2)
1588 					return (-1);
1589 			} while (OP(s) != O_QUEST && OP(s) != O_CH);
1590 			/*
1591 			 * We must skip to the next position, or we'll
1592 			 * leave altoffset() too early.
1593 			 */
1594 			scan++;
1595 			break;
1596 		case OANYOF:
1597 		case OCHAR:
1598 		case OANY:
1599 			try++;
1600 			/*FALLTHRU*/
1601 		case OBOW:
1602 		case OEOW:
1603 		case OLPAREN:
1604 		case ORPAREN:
1605 		case OOR2:
1606 			break;
1607 		default:
1608 			try = -1;
1609 			break;
1610 		}
1611 		if (try == -1)
1612 			return (-1);
1613 		s = *scan++;
1614 	}
1615 
1616 	if (try > largest)
1617 		largest = try;
1618 
1619 	return (largest+offset);
1620 }
1621 
1622 /*
1623  * computejumps - compute char jumps for BM scan
1624  *
1625  * This algorithm assumes g->must exists and is has size greater than
1626  * zero. It's based on the algorithm found on Computer Algorithms by
1627  * Sara Baase.
1628  *
1629  * A char jump is the number of characters one needs to jump based on
1630  * the value of the character from the text that was mismatched.
1631  */
1632 static void
1633 computejumps(struct parse *p, struct re_guts *g)
1634 {
1635 	int ch;
1636 	int mindex;
1637 
1638 	/* Avoid making errors worse */
1639 	if (p->error != 0)
1640 		return;
1641 
1642 	g->charjump = (int *)malloc((NC + 1) * sizeof (int));
1643 	if (g->charjump == NULL)	/* Not a fatal error */
1644 		return;
1645 	/* Adjust for signed chars, if necessary */
1646 	g->charjump = &g->charjump[-(CHAR_MIN)];
1647 
1648 	/*
1649 	 * If the character does not exist in the pattern, the jump
1650 	 * is equal to the number of characters in the pattern.
1651 	 */
1652 	for (ch = CHAR_MIN; ch < (CHAR_MAX + 1); ch++)
1653 		g->charjump[ch] = g->mlen;
1654 
1655 	/*
1656 	 * If the character does exist, compute the jump that would
1657 	 * take us to the last character in the pattern equal to it
1658 	 * (notice that we match right to left, so that last character
1659 	 * is the first one that would be matched).
1660 	 */
1661 	for (mindex = 0; mindex < g->mlen; mindex++)
1662 		g->charjump[(int)g->must[mindex]] = g->mlen - mindex - 1;
1663 }
1664 
1665 /*
1666  * computematchjumps - compute match jumps for BM scan
1667  *
1668  * This algorithm assumes g->must exists and is has size greater than
1669  * zero. It's based on the algorithm found on Computer Algorithms by
1670  * Sara Baase.
1671  *
1672  * A match jump is the number of characters one needs to advance based
1673  * on the already-matched suffix.
1674  * Notice that all values here are minus (g->mlen-1), because of the way
1675  * the search algorithm works.
1676  */
1677 static void
1678 computematchjumps(struct parse *p, struct re_guts *g)
1679 {
1680 	int mindex;		/* General "must" iterator */
1681 	int suffix;		/* Keeps track of matching suffix */
1682 	int ssuffix;		/* Keeps track of suffixes' suffix */
1683 	int *pmatches;
1684 				/*
1685 				 * pmatches[k] points to the next i
1686 				 * such that i+1...mlen is a substring
1687 				 * of k+1...k+mlen-i-1
1688 				 */
1689 
1690 	/* Avoid making errors worse */
1691 	if (p->error != 0)
1692 		return;
1693 
1694 	pmatches = (int *)malloc(g->mlen * sizeof (unsigned int));
1695 	if (pmatches == NULL) {
1696 		g->matchjump = NULL;
1697 		return;
1698 	}
1699 
1700 	g->matchjump = (int *)malloc(g->mlen * sizeof (unsigned int));
1701 	if (g->matchjump == NULL) {	/* Not a fatal error */
1702 		free(pmatches);
1703 		return;
1704 	}
1705 
1706 	/* Set maximum possible jump for each character in the pattern */
1707 	for (mindex = 0; mindex < g->mlen; mindex++)
1708 		g->matchjump[mindex] = 2*g->mlen - mindex - 1;
1709 
1710 	/* Compute pmatches[] */
1711 	for (mindex = g->mlen - 1, suffix = g->mlen; mindex >= 0;
1712 	    mindex--, suffix--) {
1713 		pmatches[mindex] = suffix;
1714 
1715 		/*
1716 		 * If a mismatch is found, interrupting the substring,
1717 		 * compute the matchjump for that position. If no
1718 		 * mismatch is found, then a text substring mismatched
1719 		 * against the suffix will also mismatch against the
1720 		 * substring.
1721 		 */
1722 		while (suffix < g->mlen && g->must[mindex] != g->must[suffix]) {
1723 			g->matchjump[suffix] = MIN(g->matchjump[suffix],
1724 			    g->mlen - mindex - 1);
1725 			suffix = pmatches[suffix];
1726 		}
1727 	}
1728 
1729 	/*
1730 	 * Compute the matchjump up to the last substring found to jump
1731 	 * to the beginning of the largest must pattern prefix matching
1732 	 * it's own suffix.
1733 	 */
1734 	for (mindex = 0; mindex <= suffix; mindex++)
1735 		g->matchjump[mindex] = MIN(g->matchjump[mindex],
1736 		    g->mlen + suffix - mindex);
1737 
1738 	ssuffix = pmatches[suffix];
1739 	while (suffix < g->mlen) {
1740 		while (suffix <= ssuffix && suffix < g->mlen) {
1741 			g->matchjump[suffix] = MIN(g->matchjump[suffix],
1742 			    g->mlen + ssuffix - suffix);
1743 			suffix++;
1744 		}
1745 		if (suffix < g->mlen)
1746 			ssuffix = pmatches[ssuffix];
1747 	}
1748 
1749 	free(pmatches);
1750 }
1751 
1752 /*
1753  * pluscount - count + nesting
1754  */
1755 static sopno			/* nesting depth */
1756 pluscount(struct parse *p, struct re_guts *g)
1757 {
1758 	sop *scan;
1759 	sop s;
1760 	sopno plusnest = 0;
1761 	sopno maxnest = 0;
1762 
1763 	if (p->error != 0)
1764 		return (0);	/* there may not be an OEND */
1765 
1766 	scan = g->strip + 1;
1767 	do {
1768 		s = *scan++;
1769 		switch (OP(s)) {
1770 		case OPLUS_:
1771 			plusnest++;
1772 			break;
1773 		case O_PLUS:
1774 			if (plusnest > maxnest)
1775 				maxnest = plusnest;
1776 			plusnest--;
1777 			break;
1778 		}
1779 	} while (OP(s) != OEND);
1780 	if (plusnest != 0)
1781 		g->iflags |= BAD;
1782 	return (maxnest);
1783 }
1784