1/*-
2 * Copyright (c) 1992, 1993, 1994 Henry Spencer.
3 * Copyright (c) 1992, 1993, 1994
4 *	The Regents of the University of California.  All rights reserved.
5 *
6 * Copyright (c) 2011 The FreeBSD Foundation
7 * All rights reserved.
8 * Portions of this software were developed by David Chisnall
9 * under sponsorship from the FreeBSD Foundation.
10 *
11 * This code is derived from software contributed to Berkeley by
12 * Henry Spencer.
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 * 1. Redistributions of source code must retain the above copyright
18 *    notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 *    notice, this list of conditions and the following disclaimer in the
21 *    documentation and/or other materials provided with the distribution.
22 * 4. Neither the name of the University nor the names of its contributors
23 *    may be used to endorse or promote products derived from this software
24 *    without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 *	@(#)regcomp.c	8.5 (Berkeley) 3/20/94
39 */
40
41#if defined(LIBC_SCCS) && !defined(lint)
42static char sccsid[] = "@(#)regcomp.c	8.5 (Berkeley) 3/20/94";
43#endif /* LIBC_SCCS and not lint */
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD$");
46
47#include <sys/types.h>
48#include <stdio.h>
49#include <string.h>
50#include <ctype.h>
51#include <limits.h>
52#include <stdlib.h>
53#include <regex.h>
54#include <runetype.h>
55#include <wchar.h>
56#include <wctype.h>
57
58#include "collate.h"
59
60#include "utils.h"
61#include "regex2.h"
62
63#include "cname.h"
64
65/*
66 * parse structure, passed up and down to avoid global variables and
67 * other clumsinesses
68 */
69struct parse {
70	char *next;		/* next character in RE */
71	char *end;		/* end of string (-> NUL normally) */
72	int error;		/* has an error been seen? */
73	sop *strip;		/* malloced strip */
74	sopno ssize;		/* malloced strip size (allocated) */
75	sopno slen;		/* malloced strip length (used) */
76	int ncsalloc;		/* number of csets allocated */
77	struct re_guts *g;
78#	define	NPAREN	10	/* we need to remember () 1-9 for back refs */
79	sopno pbegin[NPAREN];	/* -> ( ([0] unused) */
80	sopno pend[NPAREN];	/* -> ) ([0] unused) */
81};
82
83/* ========= begin header generated by ./mkh ========= */
84#ifdef __cplusplus
85extern "C" {
86#endif
87
88/* === regcomp.c === */
89static void p_ere(struct parse *p, wint_t stop);
90static void p_ere_exp(struct parse *p);
91static void p_str(struct parse *p);
92static void p_bre(struct parse *p, wint_t end1, wint_t end2);
93static int p_simp_re(struct parse *p, int starordinary);
94static int p_count(struct parse *p);
95static void p_bracket(struct parse *p);
96static void p_b_term(struct parse *p, cset *cs);
97static void p_b_cclass(struct parse *p, cset *cs);
98static void p_b_eclass(struct parse *p, cset *cs);
99static wint_t p_b_symbol(struct parse *p);
100static wint_t p_b_coll_elem(struct parse *p, wint_t endc);
101static wint_t othercase(wint_t ch);
102static void bothcases(struct parse *p, wint_t ch);
103static void ordinary(struct parse *p, wint_t ch);
104static void nonnewline(struct parse *p);
105static void repeat(struct parse *p, sopno start, int from, int to);
106static int seterr(struct parse *p, int e);
107static cset *allocset(struct parse *p);
108static void freeset(struct parse *p, cset *cs);
109static void CHadd(struct parse *p, cset *cs, wint_t ch);
110static void CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max);
111static void CHaddtype(struct parse *p, cset *cs, wctype_t wct);
112static wint_t singleton(cset *cs);
113static sopno dupl(struct parse *p, sopno start, sopno finish);
114static void doemit(struct parse *p, sop op, size_t opnd);
115static void doinsert(struct parse *p, sop op, size_t opnd, sopno pos);
116static void dofwd(struct parse *p, sopno pos, sop value);
117static void enlarge(struct parse *p, sopno size);
118static void stripsnug(struct parse *p, struct re_guts *g);
119static void findmust(struct parse *p, struct re_guts *g);
120static int altoffset(sop *scan, int offset);
121static void computejumps(struct parse *p, struct re_guts *g);
122static void computematchjumps(struct parse *p, struct re_guts *g);
123static sopno pluscount(struct parse *p, struct re_guts *g);
124static wint_t wgetnext(struct parse *p);
125
126#ifdef __cplusplus
127}
128#endif
129/* ========= end header generated by ./mkh ========= */
130
131static char nuls[10];		/* place to point scanner in event of error */
132
133/*
134 * macros for use with parse structure
135 * BEWARE:  these know that the parse structure is named `p' !!!
136 */
137#define	PEEK()	(*p->next)
138#define	PEEK2()	(*(p->next+1))
139#define	MORE()	(p->next < p->end)
140#define	MORE2()	(p->next+1 < p->end)
141#define	SEE(c)	(MORE() && PEEK() == (c))
142#define	SEETWO(a, b)	(MORE() && MORE2() && PEEK() == (a) && PEEK2() == (b))
143#define	EAT(c)	((SEE(c)) ? (NEXT(), 1) : 0)
144#define	EATTWO(a, b)	((SEETWO(a, b)) ? (NEXT2(), 1) : 0)
145#define	NEXT()	(p->next++)
146#define	NEXT2()	(p->next += 2)
147#define	NEXTn(n)	(p->next += (n))
148#define	GETNEXT()	(*p->next++)
149#define	WGETNEXT()	wgetnext(p)
150#define	SETERROR(e)	seterr(p, (e))
151#define	REQUIRE(co, e)	((co) || SETERROR(e))
152#define	MUSTSEE(c, e)	(REQUIRE(MORE() && PEEK() == (c), e))
153#define	MUSTEAT(c, e)	(REQUIRE(MORE() && GETNEXT() == (c), e))
154#define	MUSTNOTSEE(c, e)	(REQUIRE(!MORE() || PEEK() != (c), e))
155#define	EMIT(op, sopnd)	doemit(p, (sop)(op), (size_t)(sopnd))
156#define	INSERT(op, pos)	doinsert(p, (sop)(op), HERE()-(pos)+1, pos)
157#define	AHEAD(pos)		dofwd(p, pos, HERE()-(pos))
158#define	ASTERN(sop, pos)	EMIT(sop, HERE()-pos)
159#define	HERE()		(p->slen)
160#define	THERE()		(p->slen - 1)
161#define	THERETHERE()	(p->slen - 2)
162#define	DROP(n)	(p->slen -= (n))
163
164#ifndef NDEBUG
165static int never = 0;		/* for use in asserts; shuts lint up */
166#else
167#define	never	0		/* some <assert.h>s have bugs too */
168#endif
169
170/* Macro used by computejump()/computematchjump() */
171#define MIN(a,b)	((a)<(b)?(a):(b))
172
173/*
174 - regcomp - interface for parser and compilation
175 = extern int regcomp(regex_t *, const char *, int);
176 = #define	REG_BASIC	0000
177 = #define	REG_EXTENDED	0001
178 = #define	REG_ICASE	0002
179 = #define	REG_NOSUB	0004
180 = #define	REG_NEWLINE	0010
181 = #define	REG_NOSPEC	0020
182 = #define	REG_PEND	0040
183 = #define	REG_DUMP	0200
184 */
185int				/* 0 success, otherwise REG_something */
186regcomp(regex_t * __restrict preg,
187	const char * __restrict pattern,
188	int cflags)
189{
190	struct parse pa;
191	struct re_guts *g;
192	struct parse *p = &pa;
193	int i;
194	size_t len;
195#ifdef REDEBUG
196#	define	GOODFLAGS(f)	(f)
197#else
198#	define	GOODFLAGS(f)	((f)&~REG_DUMP)
199#endif
200
201	cflags = GOODFLAGS(cflags);
202	if ((cflags&REG_EXTENDED) && (cflags&REG_NOSPEC))
203		return(REG_INVARG);
204
205	if (cflags&REG_PEND) {
206		if (preg->re_endp < pattern)
207			return(REG_INVARG);
208		len = preg->re_endp - pattern;
209	} else
210		len = strlen((char *)pattern);
211
212	/* do the mallocs early so failure handling is easy */
213	g = (struct re_guts *)malloc(sizeof(struct re_guts));
214	if (g == NULL)
215		return(REG_ESPACE);
216	p->ssize = len/(size_t)2*(size_t)3 + (size_t)1;	/* ugh */
217	p->strip = (sop *)malloc(p->ssize * sizeof(sop));
218	p->slen = 0;
219	if (p->strip == NULL) {
220		free((char *)g);
221		return(REG_ESPACE);
222	}
223
224	/* set things up */
225	p->g = g;
226	p->next = (char *)pattern;	/* convenience; we do not modify it */
227	p->end = p->next + len;
228	p->error = 0;
229	p->ncsalloc = 0;
230	for (i = 0; i < NPAREN; i++) {
231		p->pbegin[i] = 0;
232		p->pend[i] = 0;
233	}
234	g->sets = NULL;
235	g->ncsets = 0;
236	g->cflags = cflags;
237	g->iflags = 0;
238	g->nbol = 0;
239	g->neol = 0;
240	g->must = NULL;
241	g->moffset = -1;
242	g->charjump = NULL;
243	g->matchjump = NULL;
244	g->mlen = 0;
245	g->nsub = 0;
246	g->backrefs = 0;
247
248	/* do it */
249	EMIT(OEND, 0);
250	g->firststate = THERE();
251	if (cflags&REG_EXTENDED)
252		p_ere(p, OUT);
253	else if (cflags&REG_NOSPEC)
254		p_str(p);
255	else
256		p_bre(p, OUT, OUT);
257	EMIT(OEND, 0);
258	g->laststate = THERE();
259
260	/* tidy up loose ends and fill things in */
261	stripsnug(p, g);
262	findmust(p, g);
263	/* only use Boyer-Moore algorithm if the pattern is bigger
264	 * than three characters
265	 */
266	if(g->mlen > 3) {
267		computejumps(p, g);
268		computematchjumps(p, g);
269		if(g->matchjump == NULL && g->charjump != NULL) {
270			free(g->charjump);
271			g->charjump = NULL;
272		}
273	}
274	g->nplus = pluscount(p, g);
275	g->magic = MAGIC2;
276	preg->re_nsub = g->nsub;
277	preg->re_g = g;
278	preg->re_magic = MAGIC1;
279#ifndef REDEBUG
280	/* not debugging, so can't rely on the assert() in regexec() */
281	if (g->iflags&BAD)
282		SETERROR(REG_ASSERT);
283#endif
284
285	/* win or lose, we're done */
286	if (p->error != 0)	/* lose */
287		regfree(preg);
288	return(p->error);
289}
290
291/*
292 - p_ere - ERE parser top level, concatenation and alternation
293 == static void p_ere(struct parse *p, int stop);
294 */
295static void
296p_ere(struct parse *p,
297	int stop)		/* character this ERE should end at */
298{
299	char c;
300	sopno prevback;
301	sopno prevfwd;
302	sopno conc;
303	int first = 1;		/* is this the first alternative? */
304
305	for (;;) {
306		/* do a bunch of concatenated expressions */
307		conc = HERE();
308		while (MORE() && (c = PEEK()) != '|' && c != stop)
309			p_ere_exp(p);
310		(void)REQUIRE(HERE() != conc, REG_EMPTY);	/* require nonempty */
311
312		if (!EAT('|'))
313			break;		/* NOTE BREAK OUT */
314
315		if (first) {
316			INSERT(OCH_, conc);	/* offset is wrong */
317			prevfwd = conc;
318			prevback = conc;
319			first = 0;
320		}
321		ASTERN(OOR1, prevback);
322		prevback = THERE();
323		AHEAD(prevfwd);			/* fix previous offset */
324		prevfwd = HERE();
325		EMIT(OOR2, 0);			/* offset is very wrong */
326	}
327
328	if (!first) {		/* tail-end fixups */
329		AHEAD(prevfwd);
330		ASTERN(O_CH, prevback);
331	}
332
333	assert(!MORE() || SEE(stop));
334}
335
336/*
337 - p_ere_exp - parse one subERE, an atom possibly followed by a repetition op
338 == static void p_ere_exp(struct parse *p);
339 */
340static void
341p_ere_exp(struct parse *p)
342{
343	char c;
344	wint_t wc;
345	sopno pos;
346	int count;
347	int count2;
348	sopno subno;
349	int wascaret = 0;
350
351	assert(MORE());		/* caller should have ensured this */
352	c = GETNEXT();
353
354	pos = HERE();
355	switch (c) {
356	case '(':
357		(void)REQUIRE(MORE(), REG_EPAREN);
358		p->g->nsub++;
359		subno = p->g->nsub;
360		if (subno < NPAREN)
361			p->pbegin[subno] = HERE();
362		EMIT(OLPAREN, subno);
363		if (!SEE(')'))
364			p_ere(p, ')');
365		if (subno < NPAREN) {
366			p->pend[subno] = HERE();
367			assert(p->pend[subno] != 0);
368		}
369		EMIT(ORPAREN, subno);
370		(void)MUSTEAT(')', REG_EPAREN);
371		break;
372#ifndef POSIX_MISTAKE
373	case ')':		/* happens only if no current unmatched ( */
374		/*
375		 * You may ask, why the ifndef?  Because I didn't notice
376		 * this until slightly too late for 1003.2, and none of the
377		 * other 1003.2 regular-expression reviewers noticed it at
378		 * all.  So an unmatched ) is legal POSIX, at least until
379		 * we can get it fixed.
380		 */
381		SETERROR(REG_EPAREN);
382		break;
383#endif
384	case '^':
385		EMIT(OBOL, 0);
386		p->g->iflags |= USEBOL;
387		p->g->nbol++;
388		wascaret = 1;
389		break;
390	case '$':
391		EMIT(OEOL, 0);
392		p->g->iflags |= USEEOL;
393		p->g->neol++;
394		break;
395	case '|':
396		SETERROR(REG_EMPTY);
397		break;
398	case '*':
399	case '+':
400	case '?':
401		SETERROR(REG_BADRPT);
402		break;
403	case '.':
404		if (p->g->cflags&REG_NEWLINE)
405			nonnewline(p);
406		else
407			EMIT(OANY, 0);
408		break;
409	case '[':
410		p_bracket(p);
411		break;
412	case '\\':
413		(void)REQUIRE(MORE(), REG_EESCAPE);
414		wc = WGETNEXT();
415		ordinary(p, wc);
416		break;
417	case '{':		/* okay as ordinary except if digit follows */
418		(void)REQUIRE(!MORE() || !isdigit((uch)PEEK()), REG_BADRPT);
419		/* FALLTHROUGH */
420	default:
421		p->next--;
422		wc = WGETNEXT();
423		ordinary(p, wc);
424		break;
425	}
426
427	if (!MORE())
428		return;
429	c = PEEK();
430	/* we call { a repetition if followed by a digit */
431	if (!( c == '*' || c == '+' || c == '?' ||
432				(c == '{' && MORE2() && isdigit((uch)PEEK2())) ))
433		return;		/* no repetition, we're done */
434	NEXT();
435
436	(void)REQUIRE(!wascaret, REG_BADRPT);
437	switch (c) {
438	case '*':	/* implemented as +? */
439		/* this case does not require the (y|) trick, noKLUDGE */
440		INSERT(OPLUS_, pos);
441		ASTERN(O_PLUS, pos);
442		INSERT(OQUEST_, pos);
443		ASTERN(O_QUEST, pos);
444		break;
445	case '+':
446		INSERT(OPLUS_, pos);
447		ASTERN(O_PLUS, pos);
448		break;
449	case '?':
450		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
451		INSERT(OCH_, pos);		/* offset slightly wrong */
452		ASTERN(OOR1, pos);		/* this one's right */
453		AHEAD(pos);			/* fix the OCH_ */
454		EMIT(OOR2, 0);			/* offset very wrong... */
455		AHEAD(THERE());			/* ...so fix it */
456		ASTERN(O_CH, THERETHERE());
457		break;
458	case '{':
459		count = p_count(p);
460		if (EAT(',')) {
461			if (isdigit((uch)PEEK())) {
462				count2 = p_count(p);
463				(void)REQUIRE(count <= count2, REG_BADBR);
464			} else		/* single number with comma */
465				count2 = INFINITY;
466		} else		/* just a single number */
467			count2 = count;
468		repeat(p, pos, count, count2);
469		if (!EAT('}')) {	/* error heuristics */
470			while (MORE() && PEEK() != '}')
471				NEXT();
472			(void)REQUIRE(MORE(), REG_EBRACE);
473			SETERROR(REG_BADBR);
474		}
475		break;
476	}
477
478	if (!MORE())
479		return;
480	c = PEEK();
481	if (!( c == '*' || c == '+' || c == '?' ||
482				(c == '{' && MORE2() && isdigit((uch)PEEK2())) ) )
483		return;
484	SETERROR(REG_BADRPT);
485}
486
487/*
488 - p_str - string (no metacharacters) "parser"
489 == static void p_str(struct parse *p);
490 */
491static void
492p_str(struct parse *p)
493{
494	(void)REQUIRE(MORE(), REG_EMPTY);
495	while (MORE())
496		ordinary(p, WGETNEXT());
497}
498
499/*
500 - p_bre - BRE parser top level, anchoring and concatenation
501 == static void p_bre(struct parse *p, int end1, \
502 ==	int end2);
503 * Giving end1 as OUT essentially eliminates the end1/end2 check.
504 *
505 * This implementation is a bit of a kludge, in that a trailing $ is first
506 * taken as an ordinary character and then revised to be an anchor.
507 * The amount of lookahead needed to avoid this kludge is excessive.
508 */
509static void
510p_bre(struct parse *p,
511	int end1,		/* first terminating character */
512	int end2)		/* second terminating character */
513{
514	sopno start = HERE();
515	int first = 1;			/* first subexpression? */
516	int wasdollar = 0;
517
518	if (EAT('^')) {
519		EMIT(OBOL, 0);
520		p->g->iflags |= USEBOL;
521		p->g->nbol++;
522	}
523	while (MORE() && !SEETWO(end1, end2)) {
524		wasdollar = p_simp_re(p, first);
525		first = 0;
526	}
527	if (wasdollar) {	/* oops, that was a trailing anchor */
528		DROP(1);
529		EMIT(OEOL, 0);
530		p->g->iflags |= USEEOL;
531		p->g->neol++;
532	}
533
534	(void)REQUIRE(HERE() != start, REG_EMPTY);	/* require nonempty */
535}
536
537/*
538 - p_simp_re - parse a simple RE, an atom possibly followed by a repetition
539 == static int p_simp_re(struct parse *p, int starordinary);
540 */
541static int			/* was the simple RE an unbackslashed $? */
542p_simp_re(struct parse *p,
543	int starordinary)	/* is a leading * an ordinary character? */
544{
545	int c;
546	int count;
547	int count2;
548	sopno pos;
549	int i;
550	wint_t wc;
551	sopno subno;
552#	define	BACKSL	(1<<CHAR_BIT)
553
554	pos = HERE();		/* repetion op, if any, covers from here */
555
556	assert(MORE());		/* caller should have ensured this */
557	c = GETNEXT();
558	if (c == '\\') {
559		(void)REQUIRE(MORE(), REG_EESCAPE);
560		c = BACKSL | GETNEXT();
561	}
562	switch (c) {
563	case '.':
564		if (p->g->cflags&REG_NEWLINE)
565			nonnewline(p);
566		else
567			EMIT(OANY, 0);
568		break;
569	case '[':
570		p_bracket(p);
571		break;
572	case BACKSL|'{':
573		SETERROR(REG_BADRPT);
574		break;
575	case BACKSL|'(':
576		p->g->nsub++;
577		subno = p->g->nsub;
578		if (subno < NPAREN)
579			p->pbegin[subno] = HERE();
580		EMIT(OLPAREN, subno);
581		/* the MORE here is an error heuristic */
582		if (MORE() && !SEETWO('\\', ')'))
583			p_bre(p, '\\', ')');
584		if (subno < NPAREN) {
585			p->pend[subno] = HERE();
586			assert(p->pend[subno] != 0);
587		}
588		EMIT(ORPAREN, subno);
589		(void)REQUIRE(EATTWO('\\', ')'), REG_EPAREN);
590		break;
591	case BACKSL|')':	/* should not get here -- must be user */
592	case BACKSL|'}':
593		SETERROR(REG_EPAREN);
594		break;
595	case BACKSL|'1':
596	case BACKSL|'2':
597	case BACKSL|'3':
598	case BACKSL|'4':
599	case BACKSL|'5':
600	case BACKSL|'6':
601	case BACKSL|'7':
602	case BACKSL|'8':
603	case BACKSL|'9':
604		i = (c&~BACKSL) - '0';
605		assert(i < NPAREN);
606		if (p->pend[i] != 0) {
607			assert(i <= p->g->nsub);
608			EMIT(OBACK_, i);
609			assert(p->pbegin[i] != 0);
610			assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
611			assert(OP(p->strip[p->pend[i]]) == ORPAREN);
612			(void) dupl(p, p->pbegin[i]+1, p->pend[i]);
613			EMIT(O_BACK, i);
614		} else
615			SETERROR(REG_ESUBREG);
616		p->g->backrefs = 1;
617		break;
618	case '*':
619		(void)REQUIRE(starordinary, REG_BADRPT);
620		/* FALLTHROUGH */
621	default:
622		p->next--;
623		wc = WGETNEXT();
624		ordinary(p, wc);
625		break;
626	}
627
628	if (EAT('*')) {		/* implemented as +? */
629		/* this case does not require the (y|) trick, noKLUDGE */
630		INSERT(OPLUS_, pos);
631		ASTERN(O_PLUS, pos);
632		INSERT(OQUEST_, pos);
633		ASTERN(O_QUEST, pos);
634	} else if (EATTWO('\\', '{')) {
635		count = p_count(p);
636		if (EAT(',')) {
637			if (MORE() && isdigit((uch)PEEK())) {
638				count2 = p_count(p);
639				(void)REQUIRE(count <= count2, REG_BADBR);
640			} else		/* single number with comma */
641				count2 = INFINITY;
642		} else		/* just a single number */
643			count2 = count;
644		repeat(p, pos, count, count2);
645		if (!EATTWO('\\', '}')) {	/* error heuristics */
646			while (MORE() && !SEETWO('\\', '}'))
647				NEXT();
648			(void)REQUIRE(MORE(), REG_EBRACE);
649			SETERROR(REG_BADBR);
650		}
651	} else if (c == '$')     /* $ (but not \$) ends it */
652		return(1);
653
654	return(0);
655}
656
657/*
658 - p_count - parse a repetition count
659 == static int p_count(struct parse *p);
660 */
661static int			/* the value */
662p_count(struct parse *p)
663{
664	int count = 0;
665	int ndigits = 0;
666
667	while (MORE() && isdigit((uch)PEEK()) && count <= DUPMAX) {
668		count = count*10 + (GETNEXT() - '0');
669		ndigits++;
670	}
671
672	(void)REQUIRE(ndigits > 0 && count <= DUPMAX, REG_BADBR);
673	return(count);
674}
675
676/*
677 - p_bracket - parse a bracketed character list
678 == static void p_bracket(struct parse *p);
679 */
680static void
681p_bracket(struct parse *p)
682{
683	cset *cs;
684	wint_t ch;
685
686	/* Dept of Truly Sickening Special-Case Kludges */
687	if (p->next + 5 < p->end && strncmp(p->next, "[:<:]]", 6) == 0) {
688		EMIT(OBOW, 0);
689		NEXTn(6);
690		return;
691	}
692	if (p->next + 5 < p->end && strncmp(p->next, "[:>:]]", 6) == 0) {
693		EMIT(OEOW, 0);
694		NEXTn(6);
695		return;
696	}
697
698	if ((cs = allocset(p)) == NULL)
699		return;
700
701	if (p->g->cflags&REG_ICASE)
702		cs->icase = 1;
703	if (EAT('^'))
704		cs->invert = 1;
705	if (EAT(']'))
706		CHadd(p, cs, ']');
707	else if (EAT('-'))
708		CHadd(p, cs, '-');
709	while (MORE() && PEEK() != ']' && !SEETWO('-', ']'))
710		p_b_term(p, cs);
711	if (EAT('-'))
712		CHadd(p, cs, '-');
713	(void)MUSTEAT(']', REG_EBRACK);
714
715	if (p->error != 0)	/* don't mess things up further */
716		return;
717
718	if (cs->invert && p->g->cflags&REG_NEWLINE)
719		cs->bmp['\n' >> 3] |= 1 << ('\n' & 7);
720
721	if ((ch = singleton(cs)) != OUT) {	/* optimize singleton sets */
722		ordinary(p, ch);
723		freeset(p, cs);
724	} else
725		EMIT(OANYOF, (int)(cs - p->g->sets));
726}
727
728/*
729 - p_b_term - parse one term of a bracketed character list
730 == static void p_b_term(struct parse *p, cset *cs);
731 */
732static void
733p_b_term(struct parse *p, cset *cs)
734{
735	char c;
736	wint_t start, finish;
737	wint_t i;
738	struct xlocale_collate *table =
739		(struct xlocale_collate*)__get_locale()->components[XLC_COLLATE];
740
741	/* classify what we've got */
742	switch ((MORE()) ? PEEK() : '\0') {
743	case '[':
744		c = (MORE2()) ? PEEK2() : '\0';
745		break;
746	case '-':
747		SETERROR(REG_ERANGE);
748		return;			/* NOTE RETURN */
749	default:
750		c = '\0';
751		break;
752	}
753
754	switch (c) {
755	case ':':		/* character class */
756		NEXT2();
757		(void)REQUIRE(MORE(), REG_EBRACK);
758		c = PEEK();
759		(void)REQUIRE(c != '-' && c != ']', REG_ECTYPE);
760		p_b_cclass(p, cs);
761		(void)REQUIRE(MORE(), REG_EBRACK);
762		(void)REQUIRE(EATTWO(':', ']'), REG_ECTYPE);
763		break;
764	case '=':		/* equivalence class */
765		NEXT2();
766		(void)REQUIRE(MORE(), REG_EBRACK);
767		c = PEEK();
768		(void)REQUIRE(c != '-' && c != ']', REG_ECOLLATE);
769		p_b_eclass(p, cs);
770		(void)REQUIRE(MORE(), REG_EBRACK);
771		(void)REQUIRE(EATTWO('=', ']'), REG_ECOLLATE);
772		break;
773	default:		/* symbol, ordinary character, or range */
774		start = p_b_symbol(p);
775		if (SEE('-') && MORE2() && PEEK2() != ']') {
776			/* range */
777			NEXT();
778			if (EAT('-'))
779				finish = '-';
780			else
781				finish = p_b_symbol(p);
782		} else
783			finish = start;
784		if (start == finish)
785			CHadd(p, cs, start);
786		else {
787			if (table->__collate_load_error) {
788				(void)REQUIRE((uch)start <= (uch)finish, REG_ERANGE);
789				CHaddrange(p, cs, start, finish);
790			} else {
791				(void)REQUIRE(__collate_range_cmp(table, start, finish) <= 0, REG_ERANGE);
792				for (i = 0; i <= UCHAR_MAX; i++) {
793					if (   __collate_range_cmp(table, start, i) <= 0
794					    && __collate_range_cmp(table, i, finish) <= 0
795					   )
796						CHadd(p, cs, i);
797				}
798			}
799		}
800		break;
801	}
802}
803
804/*
805 - p_b_cclass - parse a character-class name and deal with it
806 == static void p_b_cclass(struct parse *p, cset *cs);
807 */
808static void
809p_b_cclass(struct parse *p, cset *cs)
810{
811	char *sp = p->next;
812	size_t len;
813	wctype_t wct;
814	char clname[16];
815
816	while (MORE() && isalpha((uch)PEEK()))
817		NEXT();
818	len = p->next - sp;
819	if (len >= sizeof(clname) - 1) {
820		SETERROR(REG_ECTYPE);
821		return;
822	}
823	memcpy(clname, sp, len);
824	clname[len] = '\0';
825	if ((wct = wctype(clname)) == 0) {
826		SETERROR(REG_ECTYPE);
827		return;
828	}
829	CHaddtype(p, cs, wct);
830}
831
832/*
833 - p_b_eclass - parse an equivalence-class name and deal with it
834 == static void p_b_eclass(struct parse *p, cset *cs);
835 *
836 * This implementation is incomplete. xxx
837 */
838static void
839p_b_eclass(struct parse *p, cset *cs)
840{
841	wint_t c;
842
843	c = p_b_coll_elem(p, '=');
844	CHadd(p, cs, c);
845}
846
847/*
848 - p_b_symbol - parse a character or [..]ed multicharacter collating symbol
849 == static char p_b_symbol(struct parse *p);
850 */
851static wint_t			/* value of symbol */
852p_b_symbol(struct parse *p)
853{
854	wint_t value;
855
856	(void)REQUIRE(MORE(), REG_EBRACK);
857	if (!EATTWO('[', '.'))
858		return(WGETNEXT());
859
860	/* collating symbol */
861	value = p_b_coll_elem(p, '.');
862	(void)REQUIRE(EATTWO('.', ']'), REG_ECOLLATE);
863	return(value);
864}
865
866/*
867 - p_b_coll_elem - parse a collating-element name and look it up
868 == static char p_b_coll_elem(struct parse *p, int endc);
869 */
870static wint_t			/* value of collating element */
871p_b_coll_elem(struct parse *p,
872	wint_t endc)		/* name ended by endc,']' */
873{
874	char *sp = p->next;
875	struct cname *cp;
876	int len;
877	mbstate_t mbs;
878	wchar_t wc;
879	size_t clen;
880
881	while (MORE() && !SEETWO(endc, ']'))
882		NEXT();
883	if (!MORE()) {
884		SETERROR(REG_EBRACK);
885		return(0);
886	}
887	len = p->next - sp;
888	for (cp = cnames; cp->name != NULL; cp++)
889		if (strncmp(cp->name, sp, len) == 0 && cp->name[len] == '\0')
890			return(cp->code);	/* known name */
891	memset(&mbs, 0, sizeof(mbs));
892	if ((clen = mbrtowc(&wc, sp, len, &mbs)) == len)
893		return (wc);			/* single character */
894	else if (clen == (size_t)-1 || clen == (size_t)-2)
895		SETERROR(REG_ILLSEQ);
896	else
897		SETERROR(REG_ECOLLATE);		/* neither */
898	return(0);
899}
900
901/*
902 - othercase - return the case counterpart of an alphabetic
903 == static char othercase(int ch);
904 */
905static wint_t			/* if no counterpart, return ch */
906othercase(wint_t ch)
907{
908	assert(iswalpha(ch));
909	if (iswupper(ch))
910		return(towlower(ch));
911	else if (iswlower(ch))
912		return(towupper(ch));
913	else			/* peculiar, but could happen */
914		return(ch);
915}
916
917/*
918 - bothcases - emit a dualcase version of a two-case character
919 == static void bothcases(struct parse *p, int ch);
920 *
921 * Boy, is this implementation ever a kludge...
922 */
923static void
924bothcases(struct parse *p, wint_t ch)
925{
926	char *oldnext = p->next;
927	char *oldend = p->end;
928	char bracket[3 + MB_LEN_MAX];
929	size_t n;
930	mbstate_t mbs;
931
932	assert(othercase(ch) != ch);	/* p_bracket() would recurse */
933	p->next = bracket;
934	memset(&mbs, 0, sizeof(mbs));
935	n = wcrtomb(bracket, ch, &mbs);
936	assert(n != (size_t)-1);
937	bracket[n] = ']';
938	bracket[n + 1] = '\0';
939	p->end = bracket+n+1;
940	p_bracket(p);
941	assert(p->next == p->end);
942	p->next = oldnext;
943	p->end = oldend;
944}
945
946/*
947 - ordinary - emit an ordinary character
948 == static void ordinary(struct parse *p, int ch);
949 */
950static void
951ordinary(struct parse *p, wint_t ch)
952{
953	cset *cs;
954
955	if ((p->g->cflags&REG_ICASE) && iswalpha(ch) && othercase(ch) != ch)
956		bothcases(p, ch);
957	else if ((ch & OPDMASK) == ch)
958		EMIT(OCHAR, ch);
959	else {
960		/*
961		 * Kludge: character is too big to fit into an OCHAR operand.
962		 * Emit a singleton set.
963		 */
964		if ((cs = allocset(p)) == NULL)
965			return;
966		CHadd(p, cs, ch);
967		EMIT(OANYOF, (int)(cs - p->g->sets));
968	}
969}
970
971/*
972 - nonnewline - emit REG_NEWLINE version of OANY
973 == static void nonnewline(struct parse *p);
974 *
975 * Boy, is this implementation ever a kludge...
976 */
977static void
978nonnewline(struct parse *p)
979{
980	char *oldnext = p->next;
981	char *oldend = p->end;
982	char bracket[4];
983
984	p->next = bracket;
985	p->end = bracket+3;
986	bracket[0] = '^';
987	bracket[1] = '\n';
988	bracket[2] = ']';
989	bracket[3] = '\0';
990	p_bracket(p);
991	assert(p->next == bracket+3);
992	p->next = oldnext;
993	p->end = oldend;
994}
995
996/*
997 - repeat - generate code for a bounded repetition, recursively if needed
998 == static void repeat(struct parse *p, sopno start, int from, int to);
999 */
1000static void
1001repeat(struct parse *p,
1002	sopno start,		/* operand from here to end of strip */
1003	int from,		/* repeated from this number */
1004	int to)			/* to this number of times (maybe INFINITY) */
1005{
1006	sopno finish = HERE();
1007#	define	N	2
1008#	define	INF	3
1009#	define	REP(f, t)	((f)*8 + (t))
1010#	define	MAP(n)	(((n) <= 1) ? (n) : ((n) == INFINITY) ? INF : N)
1011	sopno copy;
1012
1013	if (p->error != 0)	/* head off possible runaway recursion */
1014		return;
1015
1016	assert(from <= to);
1017
1018	switch (REP(MAP(from), MAP(to))) {
1019	case REP(0, 0):			/* must be user doing this */
1020		DROP(finish-start);	/* drop the operand */
1021		break;
1022	case REP(0, 1):			/* as x{1,1}? */
1023	case REP(0, N):			/* as x{1,n}? */
1024	case REP(0, INF):		/* as x{1,}? */
1025		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1026		INSERT(OCH_, start);		/* offset is wrong... */
1027		repeat(p, start+1, 1, to);
1028		ASTERN(OOR1, start);
1029		AHEAD(start);			/* ... fix it */
1030		EMIT(OOR2, 0);
1031		AHEAD(THERE());
1032		ASTERN(O_CH, THERETHERE());
1033		break;
1034	case REP(1, 1):			/* trivial case */
1035		/* done */
1036		break;
1037	case REP(1, N):			/* as x?x{1,n-1} */
1038		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
1039		INSERT(OCH_, start);
1040		ASTERN(OOR1, start);
1041		AHEAD(start);
1042		EMIT(OOR2, 0);			/* offset very wrong... */
1043		AHEAD(THERE());			/* ...so fix it */
1044		ASTERN(O_CH, THERETHERE());
1045		copy = dupl(p, start+1, finish+1);
1046		assert(copy == finish+4);
1047		repeat(p, copy, 1, to-1);
1048		break;
1049	case REP(1, INF):		/* as x+ */
1050		INSERT(OPLUS_, start);
1051		ASTERN(O_PLUS, start);
1052		break;
1053	case REP(N, N):			/* as xx{m-1,n-1} */
1054		copy = dupl(p, start, finish);
1055		repeat(p, copy, from-1, to-1);
1056		break;
1057	case REP(N, INF):		/* as xx{n-1,INF} */
1058		copy = dupl(p, start, finish);
1059		repeat(p, copy, from-1, to);
1060		break;
1061	default:			/* "can't happen" */
1062		SETERROR(REG_ASSERT);	/* just in case */
1063		break;
1064	}
1065}
1066
1067/*
1068 - wgetnext - helper function for WGETNEXT() macro. Gets the next wide
1069 - character from the parse struct, signals a REG_ILLSEQ error if the
1070 - character can't be converted. Returns the number of bytes consumed.
1071 */
1072static wint_t
1073wgetnext(struct parse *p)
1074{
1075	mbstate_t mbs;
1076	wchar_t wc;
1077	size_t n;
1078
1079	memset(&mbs, 0, sizeof(mbs));
1080	n = mbrtowc(&wc, p->next, p->end - p->next, &mbs);
1081	if (n == (size_t)-1 || n == (size_t)-2) {
1082		SETERROR(REG_ILLSEQ);
1083		return (0);
1084	}
1085	if (n == 0)
1086		n = 1;
1087	p->next += n;
1088	return (wc);
1089}
1090
1091/*
1092 - seterr - set an error condition
1093 == static int seterr(struct parse *p, int e);
1094 */
1095static int			/* useless but makes type checking happy */
1096seterr(struct parse *p, int e)
1097{
1098	if (p->error == 0)	/* keep earliest error condition */
1099		p->error = e;
1100	p->next = nuls;		/* try to bring things to a halt */
1101	p->end = nuls;
1102	return(0);		/* make the return value well-defined */
1103}
1104
1105/*
1106 - allocset - allocate a set of characters for []
1107 == static cset *allocset(struct parse *p);
1108 */
1109static cset *
1110allocset(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	memset(cs, 0, sizeof(*cs));
1122
1123	return(cs);
1124}
1125
1126/*
1127 - freeset - free a now-unused set
1128 == static void freeset(struct parse *p, cset *cs);
1129 */
1130static void
1131freeset(struct parse *p, cset *cs)
1132{
1133	cset *top = &p->g->sets[p->g->ncsets];
1134
1135	free(cs->wides);
1136	free(cs->ranges);
1137	free(cs->types);
1138	memset(cs, 0, sizeof(*cs));
1139	if (cs == top-1)	/* recover only the easy case */
1140		p->g->ncsets--;
1141}
1142
1143/*
1144 - singleton - Determine whether a set contains only one character,
1145 - returning it if so, otherwise returning OUT.
1146 */
1147static wint_t
1148singleton(cset *cs)
1149{
1150	wint_t i, s, n;
1151
1152	for (i = n = 0; i < NC; i++)
1153		if (CHIN(cs, i)) {
1154			n++;
1155			s = i;
1156		}
1157	if (n == 1)
1158		return (s);
1159	if (cs->nwides == 1 && cs->nranges == 0 && cs->ntypes == 0 &&
1160	    cs->icase == 0)
1161		return (cs->wides[0]);
1162	/* Don't bother handling the other cases. */
1163	return (OUT);
1164}
1165
1166/*
1167 - CHadd - add character to character set.
1168 */
1169static void
1170CHadd(struct parse *p, cset *cs, wint_t ch)
1171{
1172	wint_t nch, *newwides;
1173	assert(ch >= 0);
1174	if (ch < NC)
1175		cs->bmp[ch >> 3] |= 1 << (ch & 7);
1176	else {
1177		newwides = realloc(cs->wides, (cs->nwides + 1) *
1178		    sizeof(*cs->wides));
1179		if (newwides == NULL) {
1180			SETERROR(REG_ESPACE);
1181			return;
1182		}
1183		cs->wides = newwides;
1184		cs->wides[cs->nwides++] = ch;
1185	}
1186	if (cs->icase) {
1187		if ((nch = towlower(ch)) < NC)
1188			cs->bmp[nch >> 3] |= 1 << (nch & 7);
1189		if ((nch = towupper(ch)) < NC)
1190			cs->bmp[nch >> 3] |= 1 << (nch & 7);
1191	}
1192}
1193
1194/*
1195 - CHaddrange - add all characters in the range [min,max] to a character set.
1196 */
1197static void
1198CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max)
1199{
1200	crange *newranges;
1201
1202	for (; min < NC && min <= max; min++)
1203		CHadd(p, cs, min);
1204	if (min >= max)
1205		return;
1206	newranges = realloc(cs->ranges, (cs->nranges + 1) *
1207	    sizeof(*cs->ranges));
1208	if (newranges == NULL) {
1209		SETERROR(REG_ESPACE);
1210		return;
1211	}
1212	cs->ranges = newranges;
1213	cs->ranges[cs->nranges].min = min;
1214	cs->ranges[cs->nranges].max = max;
1215	cs->nranges++;
1216}
1217
1218/*
1219 - CHaddtype - add all characters of a certain type to a character set.
1220 */
1221static void
1222CHaddtype(struct parse *p, cset *cs, wctype_t wct)
1223{
1224	wint_t i;
1225	wctype_t *newtypes;
1226
1227	for (i = 0; i < NC; i++)
1228		if (iswctype(i, wct))
1229			CHadd(p, cs, i);
1230	newtypes = realloc(cs->types, (cs->ntypes + 1) *
1231	    sizeof(*cs->types));
1232	if (newtypes == NULL) {
1233		SETERROR(REG_ESPACE);
1234		return;
1235	}
1236	cs->types = newtypes;
1237	cs->types[cs->ntypes++] = wct;
1238}
1239
1240/*
1241 - dupl - emit a duplicate of a bunch of sops
1242 == static sopno dupl(struct parse *p, sopno start, sopno finish);
1243 */
1244static sopno			/* start of duplicate */
1245dupl(struct parse *p,
1246	sopno start,		/* from here */
1247	sopno finish)		/* to this less one */
1248{
1249	sopno ret = HERE();
1250	sopno len = finish - start;
1251
1252	assert(finish >= start);
1253	if (len == 0)
1254		return(ret);
1255	enlarge(p, p->ssize + len);	/* this many unexpected additions */
1256	assert(p->ssize >= p->slen + len);
1257	(void) memcpy((char *)(p->strip + p->slen),
1258		(char *)(p->strip + start), (size_t)len*sizeof(sop));
1259	p->slen += len;
1260	return(ret);
1261}
1262
1263/*
1264 - doemit - emit a strip operator
1265 == static void doemit(struct parse *p, sop op, size_t opnd);
1266 *
1267 * It might seem better to implement this as a macro with a function as
1268 * hard-case backup, but it's just too big and messy unless there are
1269 * some changes to the data structures.  Maybe later.
1270 */
1271static void
1272doemit(struct parse *p, sop op, size_t opnd)
1273{
1274	/* avoid making error situations worse */
1275	if (p->error != 0)
1276		return;
1277
1278	/* deal with oversize operands ("can't happen", more or less) */
1279	assert(opnd < 1<<OPSHIFT);
1280
1281	/* deal with undersized strip */
1282	if (p->slen >= p->ssize)
1283		enlarge(p, (p->ssize+1) / 2 * 3);	/* +50% */
1284	assert(p->slen < p->ssize);
1285
1286	/* finally, it's all reduced to the easy case */
1287	p->strip[p->slen++] = SOP(op, opnd);
1288}
1289
1290/*
1291 - doinsert - insert a sop into the strip
1292 == static void doinsert(struct parse *p, sop op, size_t opnd, sopno pos);
1293 */
1294static void
1295doinsert(struct parse *p, sop op, size_t opnd, sopno pos)
1296{
1297	sopno sn;
1298	sop s;
1299	int i;
1300
1301	/* avoid making error situations worse */
1302	if (p->error != 0)
1303		return;
1304
1305	sn = HERE();
1306	EMIT(op, opnd);		/* do checks, ensure space */
1307	assert(HERE() == sn+1);
1308	s = p->strip[sn];
1309
1310	/* adjust paren pointers */
1311	assert(pos > 0);
1312	for (i = 1; i < NPAREN; i++) {
1313		if (p->pbegin[i] >= pos) {
1314			p->pbegin[i]++;
1315		}
1316		if (p->pend[i] >= pos) {
1317			p->pend[i]++;
1318		}
1319	}
1320
1321	memmove((char *)&p->strip[pos+1], (char *)&p->strip[pos],
1322						(HERE()-pos-1)*sizeof(sop));
1323	p->strip[pos] = s;
1324}
1325
1326/*
1327 - dofwd - complete a forward reference
1328 == static void dofwd(struct parse *p, sopno pos, sop value);
1329 */
1330static void
1331dofwd(struct parse *p, sopno pos, sop value)
1332{
1333	/* avoid making error situations worse */
1334	if (p->error != 0)
1335		return;
1336
1337	assert(value < 1<<OPSHIFT);
1338	p->strip[pos] = OP(p->strip[pos]) | value;
1339}
1340
1341/*
1342 - enlarge - enlarge the strip
1343 == static void enlarge(struct parse *p, sopno size);
1344 */
1345static void
1346enlarge(struct parse *p, sopno size)
1347{
1348	sop *sp;
1349
1350	if (p->ssize >= size)
1351		return;
1352
1353	sp = (sop *)realloc(p->strip, size*sizeof(sop));
1354	if (sp == NULL) {
1355		SETERROR(REG_ESPACE);
1356		return;
1357	}
1358	p->strip = sp;
1359	p->ssize = size;
1360}
1361
1362/*
1363 - stripsnug - compact the strip
1364 == static void stripsnug(struct parse *p, struct re_guts *g);
1365 */
1366static void
1367stripsnug(struct parse *p, struct re_guts *g)
1368{
1369	g->nstates = p->slen;
1370	g->strip = (sop *)realloc((char *)p->strip, p->slen * sizeof(sop));
1371	if (g->strip == NULL) {
1372		SETERROR(REG_ESPACE);
1373		g->strip = p->strip;
1374	}
1375}
1376
1377/*
1378 - findmust - fill in must and mlen with longest mandatory literal string
1379 == static void findmust(struct parse *p, struct re_guts *g);
1380 *
1381 * This algorithm could do fancy things like analyzing the operands of |
1382 * for common subsequences.  Someday.  This code is simple and finds most
1383 * of the interesting cases.
1384 *
1385 * Note that must and mlen got initialized during setup.
1386 */
1387static void
1388findmust(struct parse *p, struct re_guts *g)
1389{
1390	sop *scan;
1391	sop *start;
1392	sop *newstart;
1393	sopno newlen;
1394	sop s;
1395	char *cp;
1396	int offset;
1397	char buf[MB_LEN_MAX];
1398	size_t clen;
1399	mbstate_t mbs;
1400
1401	/* avoid making error situations worse */
1402	if (p->error != 0)
1403		return;
1404
1405	/*
1406	 * It's not generally safe to do a ``char'' substring search on
1407	 * multibyte character strings, but it's safe for at least
1408	 * UTF-8 (see RFC 3629).
1409	 */
1410	if (MB_CUR_MAX > 1 &&
1411	    strcmp(_CurrentRuneLocale->__encoding, "UTF-8") != 0)
1412		return;
1413
1414	/* find the longest OCHAR sequence in strip */
1415	newlen = 0;
1416	offset = 0;
1417	g->moffset = 0;
1418	scan = g->strip + 1;
1419	do {
1420		s = *scan++;
1421		switch (OP(s)) {
1422		case OCHAR:		/* sequence member */
1423			if (newlen == 0) {		/* new sequence */
1424				memset(&mbs, 0, sizeof(mbs));
1425				newstart = scan - 1;
1426			}
1427			clen = wcrtomb(buf, OPND(s), &mbs);
1428			if (clen == (size_t)-1)
1429				goto toohard;
1430			newlen += clen;
1431			break;
1432		case OPLUS_:		/* things that don't break one */
1433		case OLPAREN:
1434		case ORPAREN:
1435			break;
1436		case OQUEST_:		/* things that must be skipped */
1437		case OCH_:
1438			offset = altoffset(scan, offset);
1439			scan--;
1440			do {
1441				scan += OPND(s);
1442				s = *scan;
1443				/* assert() interferes w debug printouts */
1444				if (OP(s) != O_QUEST && OP(s) != O_CH &&
1445							OP(s) != OOR2) {
1446					g->iflags |= BAD;
1447					return;
1448				}
1449			} while (OP(s) != O_QUEST && OP(s) != O_CH);
1450			/* FALLTHROUGH */
1451		case OBOW:		/* things that break a sequence */
1452		case OEOW:
1453		case OBOL:
1454		case OEOL:
1455		case O_QUEST:
1456		case O_CH:
1457		case OEND:
1458			if (newlen > g->mlen) {		/* ends one */
1459				start = newstart;
1460				g->mlen = newlen;
1461				if (offset > -1) {
1462					g->moffset += offset;
1463					offset = newlen;
1464				} else
1465					g->moffset = offset;
1466			} else {
1467				if (offset > -1)
1468					offset += newlen;
1469			}
1470			newlen = 0;
1471			break;
1472		case OANY:
1473			if (newlen > g->mlen) {		/* ends one */
1474				start = newstart;
1475				g->mlen = newlen;
1476				if (offset > -1) {
1477					g->moffset += offset;
1478					offset = newlen;
1479				} else
1480					g->moffset = offset;
1481			} else {
1482				if (offset > -1)
1483					offset += newlen;
1484			}
1485			if (offset > -1)
1486				offset++;
1487			newlen = 0;
1488			break;
1489		case OANYOF:		/* may or may not invalidate offset */
1490			/* First, everything as OANY */
1491			if (newlen > g->mlen) {		/* ends one */
1492				start = newstart;
1493				g->mlen = newlen;
1494				if (offset > -1) {
1495					g->moffset += offset;
1496					offset = newlen;
1497				} else
1498					g->moffset = offset;
1499			} else {
1500				if (offset > -1)
1501					offset += newlen;
1502			}
1503			if (offset > -1)
1504				offset++;
1505			newlen = 0;
1506			break;
1507		toohard:
1508		default:
1509			/* Anything here makes it impossible or too hard
1510			 * to calculate the offset -- so we give up;
1511			 * save the last known good offset, in case the
1512			 * must sequence doesn't occur later.
1513			 */
1514			if (newlen > g->mlen) {		/* ends one */
1515				start = newstart;
1516				g->mlen = newlen;
1517				if (offset > -1)
1518					g->moffset += offset;
1519				else
1520					g->moffset = offset;
1521			}
1522			offset = -1;
1523			newlen = 0;
1524			break;
1525		}
1526	} while (OP(s) != OEND);
1527
1528	if (g->mlen == 0) {		/* there isn't one */
1529		g->moffset = -1;
1530		return;
1531	}
1532
1533	/* turn it into a character string */
1534	g->must = malloc((size_t)g->mlen + 1);
1535	if (g->must == NULL) {		/* argh; just forget it */
1536		g->mlen = 0;
1537		g->moffset = -1;
1538		return;
1539	}
1540	cp = g->must;
1541	scan = start;
1542	memset(&mbs, 0, sizeof(mbs));
1543	while (cp < g->must + g->mlen) {
1544		while (OP(s = *scan++) != OCHAR)
1545			continue;
1546		clen = wcrtomb(cp, OPND(s), &mbs);
1547		assert(clen != (size_t)-1);
1548		cp += clen;
1549	}
1550	assert(cp == g->must + g->mlen);
1551	*cp++ = '\0';		/* just on general principles */
1552}
1553
1554/*
1555 - altoffset - choose biggest offset among multiple choices
1556 == static int altoffset(sop *scan, int offset);
1557 *
1558 * Compute, recursively if necessary, the largest offset among multiple
1559 * re paths.
1560 */
1561static int
1562altoffset(sop *scan, int offset)
1563{
1564	int largest;
1565	int try;
1566	sop s;
1567
1568	/* If we gave up already on offsets, return */
1569	if (offset == -1)
1570		return -1;
1571
1572	largest = 0;
1573	try = 0;
1574	s = *scan++;
1575	while (OP(s) != O_QUEST && OP(s) != O_CH) {
1576		switch (OP(s)) {
1577		case OOR1:
1578			if (try > largest)
1579				largest = try;
1580			try = 0;
1581			break;
1582		case OQUEST_:
1583		case OCH_:
1584			try = altoffset(scan, try);
1585			if (try == -1)
1586				return -1;
1587			scan--;
1588			do {
1589				scan += OPND(s);
1590				s = *scan;
1591				if (OP(s) != O_QUEST && OP(s) != O_CH &&
1592							OP(s) != OOR2)
1593					return -1;
1594			} while (OP(s) != O_QUEST && OP(s) != O_CH);
1595			/* We must skip to the next position, or we'll
1596			 * leave altoffset() too early.
1597			 */
1598			scan++;
1599			break;
1600		case OANYOF:
1601		case OCHAR:
1602		case OANY:
1603			try++;
1604		case OBOW:
1605		case OEOW:
1606		case OLPAREN:
1607		case ORPAREN:
1608		case OOR2:
1609			break;
1610		default:
1611			try = -1;
1612			break;
1613		}
1614		if (try == -1)
1615			return -1;
1616		s = *scan++;
1617	}
1618
1619	if (try > largest)
1620		largest = try;
1621
1622	return largest+offset;
1623}
1624
1625/*
1626 - computejumps - compute char jumps for BM scan
1627 == static void computejumps(struct parse *p, struct re_guts *g);
1628 *
1629 * This algorithm assumes g->must exists and is has size greater than
1630 * zero. It's based on the algorithm found on Computer Algorithms by
1631 * Sara Baase.
1632 *
1633 * A char jump is the number of characters one needs to jump based on
1634 * the value of the character from the text that was mismatched.
1635 */
1636static void
1637computejumps(struct parse *p, struct re_guts *g)
1638{
1639	int ch;
1640	int mindex;
1641
1642	/* Avoid making errors worse */
1643	if (p->error != 0)
1644		return;
1645
1646	g->charjump = (int*) malloc((NC + 1) * sizeof(int));
1647	if (g->charjump == NULL)	/* Not a fatal error */
1648		return;
1649	/* Adjust for signed chars, if necessary */
1650	g->charjump = &g->charjump[-(CHAR_MIN)];
1651
1652	/* If the character does not exist in the pattern, the jump
1653	 * is equal to the number of characters in the pattern.
1654	 */
1655	for (ch = CHAR_MIN; ch < (CHAR_MAX + 1); ch++)
1656		g->charjump[ch] = g->mlen;
1657
1658	/* If the character does exist, compute the jump that would
1659	 * take us to the last character in the pattern equal to it
1660	 * (notice that we match right to left, so that last character
1661	 * is the first one that would be matched).
1662	 */
1663	for (mindex = 0; mindex < g->mlen; mindex++)
1664		g->charjump[(int)g->must[mindex]] = g->mlen - mindex - 1;
1665}
1666
1667/*
1668 - computematchjumps - compute match jumps for BM scan
1669 == static void computematchjumps(struct parse *p, struct re_guts *g);
1670 *
1671 * This algorithm assumes g->must exists and is has size greater than
1672 * zero. It's based on the algorithm found on Computer Algorithms by
1673 * Sara Baase.
1674 *
1675 * A match jump is the number of characters one needs to advance based
1676 * on the already-matched suffix.
1677 * Notice that all values here are minus (g->mlen-1), because of the way
1678 * the search algorithm works.
1679 */
1680static void
1681computematchjumps(struct parse *p, struct re_guts *g)
1682{
1683	int mindex;		/* General "must" iterator */
1684	int suffix;		/* Keeps track of matching suffix */
1685	int ssuffix;		/* Keeps track of suffixes' suffix */
1686	int* pmatches;		/* pmatches[k] points to the next i
1687				 * such that i+1...mlen is a substring
1688				 * of k+1...k+mlen-i-1
1689				 */
1690
1691	/* Avoid making errors worse */
1692	if (p->error != 0)
1693		return;
1694
1695	pmatches = (int*) malloc(g->mlen * sizeof(unsigned int));
1696	if (pmatches == NULL) {
1697		g->matchjump = NULL;
1698		return;
1699	}
1700
1701	g->matchjump = (int*) malloc(g->mlen * sizeof(unsigned int));
1702	if (g->matchjump == NULL)	/* Not a fatal error */
1703		return;
1704
1705	/* Set maximum possible jump for each character in the pattern */
1706	for (mindex = 0; mindex < g->mlen; mindex++)
1707		g->matchjump[mindex] = 2*g->mlen - mindex - 1;
1708
1709	/* Compute pmatches[] */
1710	for (mindex = g->mlen - 1, suffix = g->mlen; mindex >= 0;
1711	    mindex--, suffix--) {
1712		pmatches[mindex] = suffix;
1713
1714		/* If a mismatch is found, interrupting the substring,
1715		 * compute the matchjump for that position. If no
1716		 * mismatch is found, then a text substring mismatched
1717		 * against the suffix will also mismatch against the
1718		 * substring.
1719		 */
1720		while (suffix < g->mlen
1721		    && g->must[mindex] != g->must[suffix]) {
1722			g->matchjump[suffix] = MIN(g->matchjump[suffix],
1723			    g->mlen - mindex - 1);
1724			suffix = pmatches[suffix];
1725		}
1726	}
1727
1728	/* Compute the matchjump up to the last substring found to jump
1729	 * to the beginning of the largest must pattern prefix matching
1730	 * it's own suffix.
1731	 */
1732	for (mindex = 0; mindex <= suffix; mindex++)
1733		g->matchjump[mindex] = MIN(g->matchjump[mindex],
1734		    g->mlen + suffix - mindex);
1735
1736        ssuffix = pmatches[suffix];
1737        while (suffix < g->mlen) {
1738                while (suffix <= ssuffix && suffix < g->mlen) {
1739                        g->matchjump[suffix] = MIN(g->matchjump[suffix],
1740			    g->mlen + ssuffix - suffix);
1741                        suffix++;
1742                }
1743		if (suffix < g->mlen)
1744                	ssuffix = pmatches[ssuffix];
1745        }
1746
1747	free(pmatches);
1748}
1749
1750/*
1751 - pluscount - count + nesting
1752 == static sopno pluscount(struct parse *p, struct re_guts *g);
1753 */
1754static sopno			/* nesting depth */
1755pluscount(struct parse *p, struct re_guts *g)
1756{
1757	sop *scan;
1758	sop s;
1759	sopno plusnest = 0;
1760	sopno maxnest = 0;
1761
1762	if (p->error != 0)
1763		return(0);	/* there may not be an OEND */
1764
1765	scan = g->strip + 1;
1766	do {
1767		s = *scan++;
1768		switch (OP(s)) {
1769		case OPLUS_:
1770			plusnest++;
1771			break;
1772		case O_PLUS:
1773			if (plusnest > maxnest)
1774				maxnest = plusnest;
1775			plusnest--;
1776			break;
1777		}
1778	} while (OP(s) != OEND);
1779	if (plusnest != 0)
1780		g->iflags |= BAD;
1781	return(maxnest);
1782}
1783