Home | History | Annotate | Line # | Download | only in regex
regcomp.c revision 1.46
      1 /*	$NetBSD: regcomp.c,v 1.46 2021/03/11 15:00:29 christos Exp $	*/
      2 
      3 /*-
      4  * SPDX-License-Identifier: BSD-3-Clause
      5  *
      6  * Copyright (c) 1992, 1993, 1994 Henry Spencer.
      7  * Copyright (c) 1992, 1993, 1994
      8  *	The Regents of the University of California.  All rights reserved.
      9  *
     10  * Copyright (c) 2011 The FreeBSD Foundation
     11  * All rights reserved.
     12  * Portions of this software were developed by David Chisnall
     13  * under sponsorship from the FreeBSD Foundation.
     14  *
     15  * This code is derived from software contributed to Berkeley by
     16  * Henry Spencer.
     17  *
     18  * Redistribution and use in source and binary forms, with or without
     19  * modification, are permitted provided that the following conditions
     20  * are met:
     21  * 1. Redistributions of source code must retain the above copyright
     22  *    notice, this list of conditions and the following disclaimer.
     23  * 2. Redistributions in binary form must reproduce the above copyright
     24  *    notice, this list of conditions and the following disclaimer in the
     25  *    documentation and/or other materials provided with the distribution.
     26  * 3. Neither the name of the University nor the names of its contributors
     27  *    may be used to endorse or promote products derived from this software
     28  *    without specific prior written permission.
     29  *
     30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     40  * SUCH DAMAGE.
     41  *
     42  *	@(#)regcomp.c	8.5 (Berkeley) 3/20/94
     43  */
     44 
     45 #if HAVE_NBTOOL_CONFIG_H
     46 #include "nbtool_config.h"
     47 #endif
     48 
     49 #include <sys/cdefs.h>
     50 #if 0
     51 static char sccsid[] = "@(#)regcomp.c	8.5 (Berkeley) 3/20/94";
     52 __FBSDID("$FreeBSD: head/lib/libc/regex/regcomp.c 368359 2020-12-05 03:18:48Z kevans $");
     53 #endif
     54 __RCSID("$NetBSD: regcomp.c,v 1.46 2021/03/11 15:00:29 christos Exp $");
     55 
     56 #define _OPENBSD_SOURCE
     57 
     58 #ifndef LIBHACK
     59 #define REGEX_GNU_EXTENSIONS
     60 
     61 #include "namespace.h"
     62 #endif
     63 #include <sys/types.h>
     64 #include <stdio.h>
     65 #include <string.h>
     66 #include <ctype.h>
     67 #include <limits.h>
     68 #include <stdlib.h>
     69 #include <regex.h>
     70 #include <stdbool.h>
     71 
     72 #if defined(__weak_alias) && !defined(LIBHACK)
     73 __weak_alias(regcomp,_regcomp)
     74 #endif
     75 
     76 #ifdef REGEX_LIBC_COLLATE
     77 #include "collate.h"
     78 #endif
     79 
     80 #include "utils.h"
     81 #include "regex2.h"
     82 
     83 #include "cname.h"
     84 
     85 /*
     86  * Branching context, used to keep track of branch state for all of the branch-
     87  * aware functions. In addition to keeping track of branch positions for the
     88  * p_branch_* functions, we use this to simplify some clumsiness in BREs for
     89  * detection of whether ^ is acting as an anchor or being used erroneously and
     90  * also for whether we're in a sub-expression or not.
     91  */
     92 struct branchc {
     93 	sopno start;
     94 	sopno back;
     95 	sopno fwd;
     96 
     97 	int nbranch;
     98 	int nchain;
     99 	bool outer;
    100 	bool terminate;
    101 };
    102 
    103 /*
    104  * parse structure, passed up and down to avoid global variables and
    105  * other clumsinesses
    106  */
    107 struct parse {
    108 	const char *next;	/* next character in RE */
    109 	const char *end;	/* end of string (-> NUL normally) */
    110 	int error;		/* has an error been seen? */
    111 	int gnuext;
    112 	sop *strip;		/* malloced strip */
    113 	sopno ssize;		/* malloced strip size (allocated) */
    114 	sopno slen;		/* malloced strip length (used) */
    115 	size_t ncsalloc;	/* number of csets allocated */
    116 	struct re_guts *g;
    117 #	define	NPAREN	10	/* we need to remember () 1-9 for back refs */
    118 	sopno pbegin[NPAREN];	/* -> ( ([0] unused) */
    119 	sopno pend[NPAREN];	/* -> ) ([0] unused) */
    120 	bool allowbranch;	/* can this expression branch? */
    121 	bool bre;		/* convenience; is this a BRE? */
    122 	int pflags;		/* other parsing flags -- legacy escapes? */
    123 	bool (*parse_expr)(struct parse *, struct branchc *);
    124 	void (*pre_parse)(struct parse *, struct branchc *);
    125 	void (*post_parse)(struct parse *, struct branchc *);
    126 };
    127 
    128 #define PFLAG_LEGACY_ESC	0x00000001
    129 
    130 /* ========= begin header generated by ./mkh ========= */
    131 #ifdef __cplusplus
    132 extern "C" {
    133 #endif
    134 
    135 /* === regcomp.c === */
    136 static bool p_ere_exp(struct parse *p, struct branchc *bc);
    137 static void p_str(struct parse *p);
    138 static int p_branch_eat_delim(struct parse *p, struct branchc *bc);
    139 static void p_branch_ins_offset(struct parse *p, struct branchc *bc);
    140 static void p_branch_fix_tail(struct parse *p, struct branchc *bc);
    141 static bool p_branch_empty(struct parse *p, struct branchc *bc);
    142 static bool p_branch_do(struct parse *p, struct branchc *bc);
    143 static void p_bre_pre_parse(struct parse *p, struct branchc *bc);
    144 static void p_bre_post_parse(struct parse *p, struct branchc *bc);
    145 static void p_re(struct parse *p, int end1, int end2);
    146 static bool p_simp_re(struct parse *p, struct branchc *bc);
    147 static int p_count(struct parse *p);
    148 static void p_bracket(struct parse *p);
    149 static int p_range_cmp(wchar_t c1, wchar_t c2);
    150 static void p_b_term(struct parse *p, cset *cs);
    151 #ifdef REGEX_GNU_EXTENSIONS
    152 static int p_b_pseudoclass(struct parse *p, char c);
    153 #endif
    154 static void p_b_cclass(struct parse *p, cset *cs);
    155 static void p_b_cclass_named(struct parse *p, cset *cs, const char[]);
    156 static void p_b_eclass(struct parse *p, cset *cs);
    157 static wint_t p_b_symbol(struct parse *p);
    158 static wint_t p_b_coll_elem(struct parse *p, wint_t endc);
    159 static bool may_escape(struct parse *p, const wint_t ch);
    160 static wint_t othercase(wint_t ch);
    161 static void bothcases(struct parse *p, wint_t ch);
    162 static void ordinary(struct parse *p, wint_t ch);
    163 static void nonnewline(struct parse *p);
    164 static void repeat(struct parse *p, sopno start, int from, int to);
    165 static int seterr(struct parse *p, int e);
    166 static cset *allocset(struct parse *p);
    167 static void freeset(struct parse *p, cset *cs);
    168 static void CHadd(struct parse *p, cset *cs, wint_t ch);
    169 static void CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max);
    170 static void CHaddtype(struct parse *p, cset *cs, wctype_t wct);
    171 static wint_t singleton(cset *cs);
    172 static sopno dupl(struct parse *p, sopno start, sopno finish);
    173 static void doemit(struct parse *p, sop op, size_t opnd);
    174 static void doinsert(struct parse *p, sop op, size_t opnd, sopno pos);
    175 static void dofwd(struct parse *p, sopno pos, sop value);
    176 static int enlarge(struct parse *p, sopno size);
    177 static void stripsnug(struct parse *p, struct re_guts *g);
    178 static void findmust(struct parse *p, struct re_guts *g);
    179 static int altoffset(sop *scan, int offset);
    180 static void computejumps(struct parse *p, struct re_guts *g);
    181 static void computematchjumps(struct parse *p, struct re_guts *g);
    182 static sopno pluscount(struct parse *p, struct re_guts *g);
    183 static wint_t wgetnext(struct parse *p);
    184 
    185 #ifdef __cplusplus
    186 }
    187 #endif
    188 /* ========= end header generated by ./mkh ========= */
    189 
    190 static char nuls[10];		/* place to point scanner in event of error */
    191 
    192 /*
    193  * macros for use with parse structure
    194  * BEWARE:  these know that the parse structure is named `p' !!!
    195  */
    196 #define	PEEK()	(*p->next)
    197 #define	PEEK2()	(*(p->next+1))
    198 #define	MORE()	(p->next < p->end)
    199 #define	MORE2()	(p->next+1 < p->end)
    200 #define	SEE(c)	(MORE() && PEEK() == (c))
    201 #define	SEETWO(a, b)	(MORE() && MORE2() && PEEK() == (a) && PEEK2() == (b))
    202 #define	SEESPEC(a)	(p->bre ? SEETWO('\\', a) : SEE(a))
    203 #define	EAT(c)	((SEE(c)) ? (NEXT(), 1) : 0)
    204 #define	EATTWO(a, b)	((SEETWO(a, b)) ? (NEXT2(), 1) : 0)
    205 #define	EATSPEC(a)	(p->bre ? EATTWO('\\', a) : EAT(a))
    206 #define	NEXT()	(p->next++)
    207 #define	NEXT2()	(p->next += 2)
    208 #define	NEXTn(n)	(p->next += (n))
    209 #define	GETNEXT()	(*p->next++)
    210 #define	WGETNEXT()	wgetnext(p)
    211 #define	SETERROR(e)	seterr(p, (e))
    212 #define	REQUIRE(co, e)	((co) || SETERROR(e))
    213 #define	MUSTSEE(c, e)	(REQUIRE(MORE() && PEEK() == (c), e))
    214 #define	MUSTEAT(c, e)	(REQUIRE(MORE() && GETNEXT() == (c), e))
    215 #define	MUSTNOTSEE(c, e)	(REQUIRE(!MORE() || PEEK() != (c), e))
    216 #define	EMIT(op, sopnd)	doemit(p, (op), (sopnd))
    217 #define	INSERT(op, pos)	doinsert(p, (op), HERE()-(pos)+1, pos)
    218 #define	AHEAD(pos)		dofwd(p, pos, HERE()-(pos))
    219 #define	ASTERN(sop, pos)	EMIT(sop, HERE()-pos)
    220 #define	HERE()		(p->slen)
    221 #define	THERE()		(p->slen - 1)
    222 #define	THERETHERE()	(p->slen - 2)
    223 #define	DROP(n)	(p->slen -= (n))
    224 
    225 /* Macro used by computejump()/computematchjump() */
    226 #ifndef MIN
    227 #define MIN(a,b)	((a)<(b)?(a):(b))
    228 #endif
    229 
    230 #ifndef NLS
    231 static const struct {
    232 	const char *name;
    233 	int (*func)(int);
    234 } wctypes[] = {
    235 #define ADD(x) { .name = # x, .func = is ## x }
    236 	ADD(alnum),
    237 	ADD(alpha),
    238 	ADD(blank),
    239 	ADD(cntrl),
    240 	ADD(digit),
    241 	ADD(graph),
    242 	ADD(lower),
    243 	ADD(print),
    244 	ADD(punct),
    245 	ADD(space),
    246 	ADD(upper),
    247 	ADD(xdigit),
    248 #undef ADD
    249 };
    250 
    251 wctype_t
    252 __regex_wctype(const char *str)
    253 {
    254 	for (size_t i = 0; i < __arraycount(wctypes); i++) {
    255 		if (strcmp(wctypes[i].name, str) == 0)
    256 			return (wctype_t)(i + 1);
    257 	}
    258 	return (wctype_t)0;
    259 }
    260 
    261 int
    262 __regex_iswctype(wint_t c, wctype_t ct)
    263 {
    264 	if (ct == 0)
    265 		return 0;
    266 	return (*wctypes[ct - 1].func)(c);
    267 }
    268 #endif
    269 
    270 static int				/* 0 success, otherwise REG_something */
    271 regcomp_internal(regex_t * __restrict preg,
    272 	const char * __restrict pattern,
    273 	int cflags, int pflags)
    274 {
    275 	struct parse pa;
    276 	struct re_guts *g;
    277 	struct parse *p = &pa;
    278 	int i;
    279 	size_t len;
    280 	size_t maxlen;
    281 #ifdef REDEBUG
    282 #	define	GOODFLAGS(f)	(f)
    283 #else
    284 #	define	GOODFLAGS(f)	((f)&~REG_DUMP)
    285 #endif
    286 
    287 	_DIAGASSERT(preg != NULL);
    288 	_DIAGASSERT(pattern != NULL);
    289 
    290 	cflags = GOODFLAGS(cflags);
    291 	if ((cflags&REG_EXTENDED) && (cflags&REG_NOSPEC))
    292 		return(REG_INVARG);
    293 
    294 	if (cflags&REG_PEND) {
    295 		if (preg->re_endp < pattern)
    296 			return(REG_INVARG);
    297 		len = preg->re_endp - pattern;
    298 	} else
    299 		len = strlen(pattern);
    300 
    301 	/* do the mallocs early so failure handling is easy */
    302 	g = malloc(sizeof(*g));
    303 	if (g == NULL)
    304 		return(REG_ESPACE);
    305 	/*
    306 	 * Limit the pattern space to avoid a 32-bit overflow on buffer
    307 	 * extension.  Also avoid any signed overflow in case of conversion
    308 	 * so make the real limit based on a 31-bit overflow.
    309 	 *
    310 	 * Likely not applicable on 64-bit systems but handle the case
    311 	 * generically (who are we to stop people from using ~715MB+
    312 	 * patterns?).
    313 	 */
    314 	maxlen = ((size_t)-1 >> 1) / sizeof(*p->strip) * 2 / 3;
    315 	if (len >= maxlen) {
    316 		free(g);
    317 		return(REG_ESPACE);
    318 	}
    319 	p->ssize = (sopno)(len / 2 * 3 + 1);	/* ugh */
    320 	assert(p->ssize >= len);
    321 
    322 	p->strip = calloc(p->ssize, sizeof(*p->strip));
    323 	p->slen = 0;
    324 	if (p->strip == NULL) {
    325 		free(g);
    326 		return(REG_ESPACE);
    327 	}
    328 
    329 	/* set things up */
    330 	p->g = g;
    331 	p->next = pattern;	/* convenience; we do not modify it */
    332 	p->end = p->next + len;
    333 	p->error = 0;
    334 	p->ncsalloc = 0;
    335 	p->pflags = pflags;
    336 	for (i = 0; i < NPAREN; i++) {
    337 		p->pbegin[i] = 0;
    338 		p->pend[i] = 0;
    339 	}
    340 #ifdef REGEX_GNU_EXTENSIONS
    341 	if ((cflags & REG_GNU) == 0) {
    342 		p->gnuext = false;
    343 		p->allowbranch = (cflags & REG_EXTENDED) != 0;
    344 	} else
    345 		p->gnuext = p->allowbranch = true;
    346 #else
    347 	p->gnuext = false;
    348 	p->allowbranch = (cflags & REG_EXTENDED) != 0;
    349 #endif
    350 	if (cflags & REG_EXTENDED) {
    351 		p->bre = false;
    352 		p->parse_expr = p_ere_exp;
    353 		p->pre_parse = NULL;
    354 		p->post_parse = NULL;
    355 	} else {
    356 		p->bre = true;
    357 		p->parse_expr = p_simp_re;
    358 		p->pre_parse = p_bre_pre_parse;
    359 		p->post_parse = p_bre_post_parse;
    360 	}
    361 	g->sets = NULL;
    362 	g->ncsets = 0;
    363 	g->cflags = cflags;
    364 	g->iflags = 0;
    365 	g->nbol = 0;
    366 	g->neol = 0;
    367 	g->must = NULL;
    368 	g->moffset = -1;
    369 	g->charjump = NULL;
    370 	g->matchjump = NULL;
    371 	g->mlen = 0;
    372 	g->nsub = 0;
    373 	g->backrefs = 0;
    374 
    375 	/* do it */
    376 	EMIT(OEND, 0);
    377 	g->firststate = THERE();
    378 	if (cflags & REG_NOSPEC)
    379 		p_str(p);
    380 	else
    381 		p_re(p, OUT, OUT);
    382 	EMIT(OEND, 0);
    383 	g->laststate = THERE();
    384 
    385 	/* tidy up loose ends and fill things in */
    386 	stripsnug(p, g);
    387 	findmust(p, g);
    388 	/* only use Boyer-Moore algorithm if the pattern is bigger
    389 	 * than three characters
    390 	 */
    391 	if(g->mlen > 3) {
    392 		computejumps(p, g);
    393 		computematchjumps(p, g);
    394 		if(g->matchjump == NULL && g->charjump != NULL) {
    395 			free(g->charjump);
    396 			g->charjump = NULL;
    397 		}
    398 	}
    399 	g->nplus = pluscount(p, g);
    400 	g->magic = MAGIC2;
    401 	preg->re_nsub = g->nsub;
    402 	preg->re_g = g;
    403 	preg->re_magic = MAGIC1;
    404 #ifndef REDEBUG
    405 	/* not debugging, so can't rely on the assert() in regexec() */
    406 	if (g->iflags&BAD)
    407 		SETERROR(REG_ASSERT);
    408 #endif
    409 
    410 	/* win or lose, we're done */
    411 	if (p->error != 0)	/* lose */
    412 		regfree(preg);
    413 	return(p->error);
    414 }
    415 
    416 /*
    417  - regcomp - interface for parser and compilation
    418  = extern int regcomp(regex_t *, const char *, int);
    419  = #define	REG_BASIC	0000
    420  = #define	REG_EXTENDED	0001
    421  = #define	REG_ICASE	0002
    422  = #define	REG_NOSUB	0004
    423  = #define	REG_NEWLINE	0010
    424  = #define	REG_NOSPEC	0020
    425  = #define	REG_PEND	0040
    426  = #define	REG_DUMP	0200
    427  */
    428 int				/* 0 success, otherwise REG_something */
    429 regcomp(regex_t * __restrict preg,
    430 	const char * __restrict pattern,
    431 	int cflags)
    432 {
    433 
    434 	return (regcomp_internal(preg, pattern, cflags, 0));
    435 }
    436 
    437 /*
    438  - p_ere_exp - parse one subERE, an atom possibly followed by a repetition op,
    439  - return whether we should terminate or not
    440  == static bool p_ere_exp(struct parse *p);
    441  */
    442 static bool
    443 p_ere_exp(struct parse *p, struct branchc *bc)
    444 {
    445 	char c;
    446 	wint_t wc;
    447 	sopno pos;
    448 	int count;
    449 	int count2;
    450 #ifdef REGEX_GNU_EXTENSIONS
    451 	size_t i;
    452 	int handled;
    453 #endif
    454 	sopno subno;
    455 	int wascaret = 0;
    456 
    457 	_DIAGASSERT(p != NULL);
    458 
    459 	(void)bc;
    460 	assert(MORE());		/* caller should have ensured this */
    461 	c = GETNEXT();
    462 
    463 #ifdef REGEX_GNU_EXTENSIONS
    464 	handled = 0;
    465 #endif
    466 	pos = HERE();
    467 	switch (c) {
    468 	case '(':
    469 		(void)REQUIRE(MORE(), REG_EPAREN);
    470 		p->g->nsub++;
    471 		subno = (sopno)p->g->nsub;
    472 		if (subno < NPAREN)
    473 			p->pbegin[subno] = HERE();
    474 		EMIT(OLPAREN, subno);
    475 		if (!SEE(')'))
    476 			p_re(p, ')', IGN);
    477 		if (subno < NPAREN) {
    478 			p->pend[subno] = HERE();
    479 			assert(p->pend[subno] != 0);
    480 		}
    481 		EMIT(ORPAREN, subno);
    482 		(void)MUSTEAT(')', REG_EPAREN);
    483 		break;
    484 #ifndef POSIX_MISTAKE
    485 	case ')':		/* happens only if no current unmatched ( */
    486 		/*
    487 		 * You may ask, why the ifndef?  Because I didn't notice
    488 		 * this until slightly too late for 1003.2, and none of the
    489 		 * other 1003.2 regular-expression reviewers noticed it at
    490 		 * all.  So an unmatched ) is legal POSIX, at least until
    491 		 * we can get it fixed.
    492 		 */
    493 		SETERROR(REG_EPAREN);
    494 		break;
    495 #endif
    496 	case '^':
    497 		EMIT(OBOL, 0);
    498 		p->g->iflags |= USEBOL;
    499 		p->g->nbol++;
    500 		wascaret = 1;
    501 		break;
    502 	case '$':
    503 		EMIT(OEOL, 0);
    504 		p->g->iflags |= USEEOL;
    505 		p->g->neol++;
    506 		break;
    507 	case '|':
    508 		SETERROR(REG_EMPTY);
    509 		break;
    510 	case '*':
    511 	case '+':
    512 	case '?':
    513 	case '{':
    514 		SETERROR(REG_BADRPT);
    515 		break;
    516 	case '.':
    517 		if (p->g->cflags&REG_NEWLINE)
    518 			nonnewline(p);
    519 		else
    520 			EMIT(OANY, 0);
    521 		break;
    522 	case '[':
    523 		p_bracket(p);
    524 		break;
    525 	case '\\':
    526 		(void)REQUIRE(MORE(), REG_EESCAPE);
    527 		wc = WGETNEXT();
    528 #ifdef REGEX_GNU_EXTENSIONS
    529 		if (p->gnuext) {
    530 			handled = 1;
    531 			switch (wc) {
    532 			case '`':
    533 				EMIT(OBOS, 0);
    534 				break;
    535 			case '\'':
    536 				EMIT(OEOS, 0);
    537 				break;
    538 			case 'B':
    539 				EMIT(ONWBND, 0);
    540 				break;
    541 			case 'b':
    542 				EMIT(OWBND, 0);
    543 				break;
    544 			case 'W':
    545 			case 'w':
    546 			case 'S':
    547 			case 's':
    548 				p_b_pseudoclass(p, wc);
    549 				break;
    550 			case 'a':
    551 				ordinary(p, '\a');
    552 				break;
    553 			case 'e':
    554 				ordinary(p, '\e');
    555 				break;
    556 			case 'f':
    557 				ordinary(p, '\f');
    558 				break;
    559 			case 'n':
    560 				ordinary(p, '\n');
    561 				break;
    562 			case 'r':
    563 				ordinary(p, '\r');
    564 				break;
    565 			case 't':
    566 				ordinary(p, '\t');
    567 				break;
    568 			case 'v':
    569 				ordinary(p, '\v');
    570 				break;
    571 			case '1':
    572 			case '2':
    573 			case '3':
    574 			case '4':
    575 			case '5':
    576 			case '6':
    577 			case '7':
    578 			case '8':
    579 			case '9':
    580 				i = wc - '0';
    581 				assert(i < NPAREN);
    582 				if (p->pend[i] != 0) {
    583 					assert(i <= p->g->nsub);
    584 					EMIT(OBACK_, i);
    585 					assert(p->pbegin[i] != 0);
    586 					assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
    587 					assert(OP(p->strip[p->pend[i]]) == ORPAREN);
    588 					(void) dupl(p, p->pbegin[i]+1, p->pend[i]);
    589 					EMIT(O_BACK, i);
    590 				} else
    591 					SETERROR(REG_ESUBREG);
    592 				p->g->backrefs = 1;
    593 				break;
    594 			default:
    595 				handled = 0;
    596 			}
    597 			/* Don't proceed to the POSIX bits if we've already handled it */
    598 			if (handled)
    599 				break;
    600 		}
    601 #endif
    602 		switch (wc) {
    603 		case '<':
    604 			EMIT(OBOW, 0);
    605 			break;
    606 		case '>':
    607 			EMIT(OEOW, 0);
    608 			break;
    609 		default:
    610 			if (may_escape(p, wc))
    611 				ordinary(p, wc);
    612 			else
    613 				SETERROR(REG_EESCAPE);
    614 			break;
    615 		}
    616 		break;
    617 	default:
    618 		if (p->error != 0)
    619 			return (false);
    620 		p->next--;
    621 		wc = WGETNEXT();
    622 		ordinary(p, wc);
    623 		break;
    624 	}
    625 
    626 	if (!MORE())
    627 		return (false);
    628 	c = PEEK();
    629 	/* we call { a repetition if followed by a digit */
    630 	if (!( c == '*' || c == '+' || c == '?' || c == '{'))
    631 		return (false);		/* no repetition, we're done */
    632 	else if (c == '{')
    633 		(void)REQUIRE(MORE2() && \
    634 		    (isdigit((uch)PEEK2()) || PEEK2() == ','), REG_BADRPT);
    635 	NEXT();
    636 
    637 	(void)REQUIRE(!wascaret, REG_BADRPT);
    638 	switch (c) {
    639 	case '*':	/* implemented as +? */
    640 		/* this case does not require the (y|) trick, noKLUDGE */
    641 		INSERT(OPLUS_, pos);
    642 		ASTERN(O_PLUS, pos);
    643 		INSERT(OQUEST_, pos);
    644 		ASTERN(O_QUEST, pos);
    645 		break;
    646 	case '+':
    647 		INSERT(OPLUS_, pos);
    648 		ASTERN(O_PLUS, pos);
    649 		break;
    650 	case '?':
    651 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
    652 		INSERT(OCH_, pos);		/* offset slightly wrong */
    653 		ASTERN(OOR1, pos);		/* this one's right */
    654 		AHEAD(pos);			/* fix the OCH_ */
    655 		EMIT(OOR2, 0);			/* offset very wrong... */
    656 		AHEAD(THERE());			/* ...so fix it */
    657 		ASTERN(O_CH, THERETHERE());
    658 		break;
    659 	case '{':
    660 		count = p_count(p);
    661 		if (EAT(',')) {
    662 			if (isdigit((uch)PEEK())) {
    663 				count2 = p_count(p);
    664 				(void)REQUIRE(count <= count2, REG_BADBR);
    665 			} else		/* single number with comma */
    666 				count2 = INFINITY;
    667 		} else		/* just a single number */
    668 			count2 = count;
    669 		repeat(p, pos, count, count2);
    670 		if (!EAT('}')) {	/* error heuristics */
    671 			while (MORE() && PEEK() != '}')
    672 				NEXT();
    673 			(void)REQUIRE(MORE(), REG_EBRACE);
    674 			SETERROR(REG_BADBR);
    675 		}
    676 		break;
    677 	}
    678 
    679 	if (!MORE())
    680 		return (false);
    681 	c = PEEK();
    682 	if (!( c == '*' || c == '+' || c == '?' ||
    683 				(c == '{' && MORE2() && isdigit((uch)PEEK2())) ) )
    684 		return (false);
    685 	SETERROR(REG_BADRPT);
    686 	return (false);
    687 }
    688 
    689 /*
    690  - p_str - string (no metacharacters) "parser"
    691  == static void p_str(struct parse *p);
    692  */
    693 static void
    694 p_str(struct parse *p)
    695 {
    696 	(void)REQUIRE(MORE(), REG_EMPTY);
    697 	while (MORE())
    698 		ordinary(p, WGETNEXT());
    699 }
    700 
    701 /*
    702  * Eat consecutive branch delimiters for the kind of expression that we are
    703  * parsing, return the number of delimiters that we ate.
    704  */
    705 static int
    706 p_branch_eat_delim(struct parse *p, struct branchc *bc)
    707 {
    708 	int nskip;
    709 
    710 	(void)bc;
    711 	nskip = 0;
    712 	while (EATSPEC('|'))
    713 		++nskip;
    714 	return (nskip);
    715 }
    716 
    717 /*
    718  * Insert necessary branch book-keeping operations. This emits a
    719  * bogus 'next' offset, since we still have more to parse
    720  */
    721 static void
    722 p_branch_ins_offset(struct parse *p, struct branchc *bc)
    723 {
    724 
    725 	if (bc->nbranch == 0) {
    726 		INSERT(OCH_, bc->start);	/* offset is wrong */
    727 		bc->fwd = bc->start;
    728 		bc->back = bc->start;
    729 	}
    730 
    731 	ASTERN(OOR1, bc->back);
    732 	bc->back = THERE();
    733 	AHEAD(bc->fwd);			/* fix previous offset */
    734 	bc->fwd = HERE();
    735 	EMIT(OOR2, 0);			/* offset is very wrong */
    736 	++bc->nbranch;
    737 }
    738 
    739 /*
    740  * Fix the offset of the tail branch, if we actually had any branches.
    741  * This is to correct the bogus placeholder offset that we use.
    742  */
    743 static void
    744 p_branch_fix_tail(struct parse *p, struct branchc *bc)
    745 {
    746 
    747 	/* Fix bogus offset at the tail if we actually have branches */
    748 	if (bc->nbranch > 0) {
    749 		AHEAD(bc->fwd);
    750 		ASTERN(O_CH, bc->back);
    751 	}
    752 }
    753 
    754 /*
    755  * Signal to the parser that an empty branch has been encountered; this will,
    756  * in the future, be used to allow for more permissive behavior with empty
    757  * branches. The return value should indicate whether parsing may continue
    758  * or not.
    759  */
    760 static bool
    761 p_branch_empty(struct parse *p, struct branchc *bc)
    762 {
    763 
    764 	(void)bc;
    765 	SETERROR(REG_EMPTY);
    766 	return (false);
    767 }
    768 
    769 /*
    770  * Take care of any branching requirements. This includes inserting the
    771  * appropriate branching instructions as well as eating all of the branch
    772  * delimiters until we either run out of pattern or need to parse more pattern.
    773  */
    774 static bool
    775 p_branch_do(struct parse *p, struct branchc *bc)
    776 {
    777 	int ate = 0;
    778 
    779 	ate = p_branch_eat_delim(p, bc);
    780 	if (ate == 0)
    781 		return (false);
    782 	else if ((ate > 1 || (bc->outer && !MORE())) && !p_branch_empty(p, bc))
    783 		/*
    784 		 * Halt parsing only if we have an empty branch and p_branch_empty
    785 		 * indicates that we must not continue. In the future, this will not
    786 		 * necessarily be an error.
    787 		 */
    788 		return (false);
    789 	p_branch_ins_offset(p, bc);
    790 
    791 	return (true);
    792 }
    793 
    794 static void
    795 p_bre_pre_parse(struct parse *p, struct branchc *bc)
    796 {
    797 
    798 	(void)bc;
    799 	/*
    800 	 * Does not move cleanly into expression parser because of
    801 	 * ordinary interpration of * at the beginning position of
    802 	 * an expression.
    803 	 */
    804 	if (EAT('^')) {
    805 		EMIT(OBOL, 0);
    806 		p->g->iflags |= USEBOL;
    807 		p->g->nbol++;
    808 	}
    809 }
    810 
    811 static void
    812 p_bre_post_parse(struct parse *p, struct branchc *bc)
    813 {
    814 
    815 	/* Expression is terminating due to EOL token */
    816 	if (bc->terminate) {
    817 		DROP(1);
    818 		EMIT(OEOL, 0);
    819 		p->g->iflags |= USEEOL;
    820 		p->g->neol++;
    821 	}
    822 }
    823 
    824 /*
    825  - p_re - Top level parser, concatenation and BRE anchoring
    826  == static void p_re(struct parse *p, int end1, int end2);
    827  * Giving end1 as OUT essentially eliminates the end1/end2 check.
    828  *
    829  * This implementation is a bit of a kludge, in that a trailing $ is first
    830  * taken as an ordinary character and then revised to be an anchor.
    831  * The amount of lookahead needed to avoid this kludge is excessive.
    832  */
    833 static void
    834 p_re(struct parse *p,
    835 	int end1,	/* first terminating character */
    836 	int end2)	/* second terminating character; ignored for EREs */
    837 {
    838 	struct branchc bc;
    839 
    840 	bc.nbranch = 0;
    841 	if (end1 == OUT && end2 == OUT)
    842 		bc.outer = true;
    843 	else
    844 		bc.outer = false;
    845 #define	SEEEND()	(!p->bre ? SEE(end1) : SEETWO(end1, end2))
    846 	for (;;) {
    847 		bc.start = HERE();
    848 		bc.nchain = 0;
    849 		bc.terminate = false;
    850 		if (p->pre_parse != NULL)
    851 			p->pre_parse(p, &bc);
    852 		while (MORE() && (!p->allowbranch || !SEESPEC('|')) && !SEEEND()) {
    853 			bc.terminate = p->parse_expr(p, &bc);
    854 			++bc.nchain;
    855 		}
    856 		if (p->post_parse != NULL)
    857 			p->post_parse(p, &bc);
    858 		(void) REQUIRE(p->gnuext || HERE() != bc.start, REG_EMPTY);
    859 #ifdef REGEX_GNU_EXTENSIONS
    860 		if (p->gnuext && HERE() == bc.start && !p_branch_empty(p, &bc))
    861 			break;
    862 #endif
    863 		if (!p->allowbranch)
    864 			break;
    865 		/*
    866 		 * p_branch_do's return value indicates whether we should
    867 		 * continue parsing or not. This is both for correctness and
    868 		 * a slight optimization, because it will check if we've
    869 		 * encountered an empty branch or the end of the string
    870 		 * immediately following a branch delimiter.
    871 		 */
    872 		if (!p_branch_do(p, &bc))
    873 			break;
    874 	}
    875 #undef SEE_END
    876 	if (p->allowbranch)
    877 		p_branch_fix_tail(p, &bc);
    878 	assert(!MORE() || SEE(end1));
    879 }
    880 
    881 /*
    882  - p_simp_re - parse a simple RE, an atom possibly followed by a repetition
    883  == static bool p_simp_re(struct parse *p, struct branchc *bc);
    884  */
    885 static bool			/* was the simple RE an unbackslashed $? */
    886 p_simp_re(struct parse *p, struct branchc *bc)
    887 {
    888 	int c;
    889 	int cc;			/* convenient/control character */
    890 	int count;
    891 	int count2;
    892 	sopno pos;
    893 	bool handled;
    894 	size_t i;
    895 	wint_t wc;
    896 	sopno subno;
    897 #	define	BACKSL	(1<<CHAR_BIT)
    898 
    899 	pos = HERE();		/* repetition op, if any, covers from here */
    900 	handled = false;
    901 
    902 	assert(MORE());		/* caller should have ensured this */
    903 	c = GETNEXT();
    904 	if (c == '\\') {
    905 		(void)REQUIRE(MORE(), REG_EESCAPE);
    906 		cc = GETNEXT();
    907 		c = BACKSL | cc;
    908 #ifdef REGEX_GNU_EXTENSIONS
    909 		if (p->gnuext) {
    910 			handled = true;
    911 			switch (c) {
    912 			case BACKSL|'`':
    913 				EMIT(OBOS, 0);
    914 				break;
    915 			case BACKSL|'\'':
    916 				EMIT(OEOS, 0);
    917 				break;
    918 			case BACKSL|'B':
    919 				EMIT(ONWBND, 0);
    920 				break;
    921 			case BACKSL|'b':
    922 				EMIT(OWBND, 0);
    923 				break;
    924 			case BACKSL|'W':
    925 			case BACKSL|'w':
    926 			case BACKSL|'S':
    927 			case BACKSL|'s':
    928 				p_b_pseudoclass(p, cc);
    929 				break;
    930 			case BACKSL|'a':
    931 				ordinary(p, '\a');
    932 				break;
    933 			case BACKSL|'e':
    934 				ordinary(p, '\e');
    935 				break;
    936 			case BACKSL|'f':
    937 				ordinary(p, '\f');
    938 				break;
    939 			case BACKSL|'n':
    940 				ordinary(p, '\n');
    941 				break;
    942 			case BACKSL|'r':
    943 				ordinary(p, '\r');
    944 				break;
    945 			case BACKSL|'t':
    946 				ordinary(p, '\t');
    947 				break;
    948 			case BACKSL|'v':
    949 				ordinary(p, '\v');
    950 				break;
    951 			default:
    952 				handled = false;
    953 			}
    954 		}
    955 #endif
    956 	}
    957 	if (!handled) {
    958 		switch (c) {
    959 		case '.':
    960 			if (p->g->cflags&REG_NEWLINE)
    961 				nonnewline(p);
    962 			else
    963 				EMIT(OANY, 0);
    964 			break;
    965 		case '[':
    966 			p_bracket(p);
    967 			break;
    968 		case BACKSL|'<':
    969 			EMIT(OBOW, 0);
    970 			break;
    971 		case BACKSL|'>':
    972 			EMIT(OEOW, 0);
    973 			break;
    974 		case BACKSL|'{':
    975 			SETERROR(REG_BADRPT);
    976 			break;
    977 		case BACKSL|'(':
    978 			p->g->nsub++;
    979 			subno = (sopno)p->g->nsub;
    980 			if (subno < NPAREN)
    981 				p->pbegin[subno] = HERE();
    982 			EMIT(OLPAREN, subno);
    983 			/* the MORE here is an error heuristic */
    984 			if (MORE() && !SEETWO('\\', ')'))
    985 				p_re(p, '\\', ')');
    986 			if (subno < NPAREN) {
    987 				p->pend[subno] = HERE();
    988 				assert(p->pend[subno] != 0);
    989 			}
    990 			EMIT(ORPAREN, subno);
    991 			(void)REQUIRE(EATTWO('\\', ')'), REG_EPAREN);
    992 			break;
    993 		case BACKSL|')':	/* should not get here -- must be user */
    994 			SETERROR(REG_EPAREN);
    995 			break;
    996 		case BACKSL|'1':
    997 		case BACKSL|'2':
    998 		case BACKSL|'3':
    999 		case BACKSL|'4':
   1000 		case BACKSL|'5':
   1001 		case BACKSL|'6':
   1002 		case BACKSL|'7':
   1003 		case BACKSL|'8':
   1004 		case BACKSL|'9':
   1005 			i = (c&~BACKSL) - '0';
   1006 			assert(i < NPAREN);
   1007 			if (p->pend[i] != 0) {
   1008 				assert(i <= p->g->nsub);
   1009 				EMIT(OBACK_, i);
   1010 				assert(p->pbegin[i] != 0);
   1011 				assert(OP(p->strip[p->pbegin[i]]) == OLPAREN);
   1012 				assert(OP(p->strip[p->pend[i]]) == ORPAREN);
   1013 				(void) dupl(p, p->pbegin[i]+1, p->pend[i]);
   1014 				EMIT(O_BACK, i);
   1015 			} else
   1016 				SETERROR(REG_ESUBREG);
   1017 			p->g->backrefs = 1;
   1018 			break;
   1019 		case '*':
   1020 			/*
   1021 			 * Ordinary if used as the first character beyond BOL anchor of
   1022 			 * a (sub-)expression, counts as a bad repetition operator if it
   1023 			 * appears otherwise.
   1024 			 */
   1025 			(void)REQUIRE(bc->nchain == 0, REG_BADRPT);
   1026 			/* FALLTHROUGH */
   1027 		default:
   1028 			if (p->error != 0)
   1029 				return (false);	/* Definitely not $... */
   1030 			p->next--;
   1031 			wc = WGETNEXT();
   1032 			if ((c & BACKSL) == 0 || may_escape(p, wc))
   1033 				ordinary(p, wc);
   1034 			else
   1035 				SETERROR(REG_EESCAPE);
   1036 			break;
   1037 		}
   1038 	}
   1039 
   1040 	if (EAT('*')) {		/* implemented as +? */
   1041 		/* this case does not require the (y|) trick, noKLUDGE */
   1042 		INSERT(OPLUS_, pos);
   1043 		ASTERN(O_PLUS, pos);
   1044 		INSERT(OQUEST_, pos);
   1045 		ASTERN(O_QUEST, pos);
   1046 #ifdef REGEX_GNU_EXTENSIONS
   1047 	} else if (p->gnuext && EATTWO('\\', '?')) {
   1048 		INSERT(OQUEST_, pos);
   1049 		ASTERN(O_QUEST, pos);
   1050 	} else if (p->gnuext && EATTWO('\\', '+')) {
   1051 		INSERT(OPLUS_, pos);
   1052 		ASTERN(O_PLUS, pos);
   1053 #endif
   1054 	} else if (EATTWO('\\', '{')) {
   1055 		count = p_count(p);
   1056 		if (EAT(',')) {
   1057 			if (MORE() && isdigit((uch)PEEK())) {
   1058 				count2 = p_count(p);
   1059 				(void)REQUIRE(count <= count2, REG_BADBR);
   1060 			} else		/* single number with comma */
   1061 				count2 = INFINITY;
   1062 		} else		/* just a single number */
   1063 			count2 = count;
   1064 		repeat(p, pos, count, count2);
   1065 		if (!EATTWO('\\', '}')) {	/* error heuristics */
   1066 			while (MORE() && !SEETWO('\\', '}'))
   1067 				NEXT();
   1068 			(void)REQUIRE(MORE(), REG_EBRACE);
   1069 			SETERROR(REG_BADBR);
   1070 		}
   1071 	} else if (c == '$')     /* $ (but not \$) ends it */
   1072 		return (true);
   1073 
   1074 	return (false);
   1075 }
   1076 
   1077 /*
   1078  - p_count - parse a repetition count
   1079  == static int p_count(struct parse *p);
   1080  */
   1081 static int			/* the value */
   1082 p_count(struct parse *p)
   1083 {
   1084 	int count = 0;
   1085 	int ndigits = 0;
   1086 
   1087 	while (MORE() && isdigit((uch)PEEK()) && count <= DUPMAX) {
   1088 		count = count*10 + (GETNEXT() - '0');
   1089 		ndigits++;
   1090 	}
   1091 
   1092 	(void)REQUIRE(ndigits > 0 && count <= DUPMAX, REG_BADBR);
   1093 	return(count);
   1094 }
   1095 
   1096 /*
   1097  - p_bracket - parse a bracketed character list
   1098  == static void p_bracket(struct parse *p);
   1099  */
   1100 static void
   1101 p_bracket(struct parse *p)
   1102 {
   1103 	cset *cs;
   1104 	wint_t ch;
   1105 
   1106 	/* Dept of Truly Sickening Special-Case Kludges */
   1107 	if (p->next + 5 < p->end && strncmp(p->next, "[:<:]]", 6) == 0) {
   1108 		EMIT(OBOW, 0);
   1109 		NEXTn(6);
   1110 		return;
   1111 	}
   1112 	if (p->next + 5 < p->end && strncmp(p->next, "[:>:]]", 6) == 0) {
   1113 		EMIT(OEOW, 0);
   1114 		NEXTn(6);
   1115 		return;
   1116 	}
   1117 
   1118 	if ((cs = allocset(p)) == NULL)
   1119 		return;
   1120 
   1121 	if (p->g->cflags&REG_ICASE)
   1122 		cs->icase = 1;
   1123 	if (EAT('^'))
   1124 		cs->invert = 1;
   1125 	if (EAT(']'))
   1126 		CHadd(p, cs, ']');
   1127 	else if (EAT('-'))
   1128 		CHadd(p, cs, '-');
   1129 	while (MORE() && PEEK() != ']' && !SEETWO('-', ']'))
   1130 		p_b_term(p, cs);
   1131 	if (EAT('-'))
   1132 		CHadd(p, cs, '-');
   1133 	(void)MUSTEAT(']', REG_EBRACK);
   1134 
   1135 	if (p->error != 0)	/* don't mess things up further */
   1136 		return;
   1137 
   1138 	if (cs->invert && p->g->cflags&REG_NEWLINE)
   1139 		cs->bmp['\n' >> 3] |= 1 << ('\n' & 7);
   1140 
   1141 	if ((ch = singleton(cs)) != OUT) {	/* optimize singleton sets */
   1142 		ordinary(p, ch);
   1143 		freeset(p, cs);
   1144 	} else
   1145 		EMIT(OANYOF, (size_t)(cs - p->g->sets));
   1146 }
   1147 
   1148 static int
   1149 p_range_cmp(wchar_t c1, wchar_t c2)
   1150 {
   1151 #ifdef REGEX_LIBC_COLLATE
   1152 	return __wcollate_range_cmp(c1, c2);
   1153 #elif defined(NLS)
   1154 	/* Copied from libc/collate __wcollate_range_cmp */
   1155 	wchar_t s1[2], s2[2];
   1156 
   1157 	s1[0] = c1;
   1158 	s1[1] = L'\0';
   1159 	s2[0] = c2;
   1160 	s2[1] = L'\0';
   1161 	return wcscoll(s1, s2);
   1162 #else
   1163 	char s1[2], s2[2];
   1164 
   1165 	s1[0] = (char)c1;
   1166 	s1[1] = '\0';
   1167 	s2[0] = (char)c2;
   1168 	s2[1] = '\0';
   1169 	return strcoll(s1, s2);
   1170 #endif
   1171 }
   1172 
   1173 /*
   1174  - p_b_term - parse one term of a bracketed character list
   1175  == static void p_b_term(struct parse *p, cset *cs);
   1176  */
   1177 static void
   1178 p_b_term(struct parse *p, cset *cs)
   1179 {
   1180 	char c;
   1181 	wint_t start, finish;
   1182 	wint_t i;
   1183 #ifdef REGEX_LIBC_COLLATE
   1184 	struct xlocale_collate *table =
   1185 		(struct xlocale_collate*)__get_locale()->components[XLC_COLLATE];
   1186 #endif
   1187 
   1188 	_DIAGASSERT(p != NULL);
   1189 	_DIAGASSERT(cs != NULL);
   1190 
   1191 	/* classify what we've got */
   1192 	switch ((MORE()) ? PEEK() : '\0') {
   1193 	case '[':
   1194 		c = (MORE2()) ? PEEK2() : '\0';
   1195 		break;
   1196 	case '-':
   1197 		SETERROR(REG_ERANGE);
   1198 		return;			/* NOTE RETURN */
   1199 	default:
   1200 		c = '\0';
   1201 		break;
   1202 	}
   1203 
   1204 	switch (c) {
   1205 	case ':':		/* character class */
   1206 		NEXT2();
   1207 		(void)REQUIRE(MORE(), REG_EBRACK);
   1208 		c = PEEK();
   1209 		(void)REQUIRE(c != '-' && c != ']', REG_ECTYPE);
   1210 		p_b_cclass(p, cs);
   1211 		(void)REQUIRE(MORE(), REG_EBRACK);
   1212 		(void)REQUIRE(EATTWO(':', ']'), REG_ECTYPE);
   1213 		break;
   1214 	case '=':		/* equivalence class */
   1215 		NEXT2();
   1216 		(void)REQUIRE(MORE(), REG_EBRACK);
   1217 		c = PEEK();
   1218 		(void)REQUIRE(c != '-' && c != ']', REG_ECOLLATE);
   1219 		p_b_eclass(p, cs);
   1220 		(void)REQUIRE(MORE(), REG_EBRACK);
   1221 		(void)REQUIRE(EATTWO('=', ']'), REG_ECOLLATE);
   1222 		break;
   1223 	default:		/* symbol, ordinary character, or range */
   1224 		start = p_b_symbol(p);
   1225 		if (SEE('-') && MORE2() && PEEK2() != ']') {
   1226 			/* range */
   1227 			NEXT();
   1228 			if (EAT('-'))
   1229 				finish = '-';
   1230 			else
   1231 				finish = p_b_symbol(p);
   1232 		} else
   1233 			finish = start;
   1234 		if (start == finish)
   1235 			CHadd(p, cs, start);
   1236 		else {
   1237 #ifdef REGEX_LIBC_COLLATE
   1238 			if (table->__collate_load_error || MB_CUR_MAX > 1) {
   1239 #else
   1240 			if (MB_CUR_MAX > 1) {
   1241 #endif
   1242 				(void)REQUIRE(start <= finish, REG_ERANGE);
   1243 				CHaddrange(p, cs, start, finish);
   1244 			} else {
   1245 				(void)REQUIRE(p_range_cmp(start, finish) <= 0, REG_ERANGE);
   1246 				for (i = 0; i <= UCHAR_MAX; i++) {
   1247 					if (p_range_cmp(start, i) <= 0 &&
   1248 					    p_range_cmp(i, finish) <= 0 )
   1249 						CHadd(p, cs, i);
   1250 				}
   1251 			}
   1252 		}
   1253 		break;
   1254 	}
   1255 }
   1256 
   1257 #ifdef REGEX_GNU_EXTENSIONS
   1258 /*
   1259  - p_b_pseudoclass - parse a pseudo-class (\w, \W, \s, \S)
   1260  == static int p_b_pseudoclass(struct parse *p, char c)
   1261  */
   1262 static int
   1263 p_b_pseudoclass(struct parse *p, char c) {
   1264 	cset *cs;
   1265 
   1266 	if ((cs = allocset(p)) == NULL)
   1267 		return(0);
   1268 
   1269 	if (p->g->cflags&REG_ICASE)
   1270 		cs->icase = 1;
   1271 
   1272 	switch (c) {
   1273 	case 'W':
   1274 		cs->invert = 1;
   1275 		/* FALLTHROUGH */
   1276 	case 'w':
   1277 		p_b_cclass_named(p, cs, "alnum");
   1278 		break;
   1279 	case 'S':
   1280 		cs->invert = 1;
   1281 		/* FALLTHROUGH */
   1282 	case 's':
   1283 		p_b_cclass_named(p, cs, "space");
   1284 		break;
   1285 	default:
   1286 		return(0);
   1287 	}
   1288 
   1289 	EMIT(OANYOF, (size_t)(cs - p->g->sets));
   1290 	return(1);
   1291 }
   1292 #endif
   1293 
   1294 /*
   1295  - p_b_cclass - parse a character-class name and deal with it
   1296  == static void p_b_cclass(struct parse *p, cset *cs);
   1297  */
   1298 static void
   1299 p_b_cclass(struct parse *p, cset *cs)
   1300 {
   1301 	const char *sp = p->next;
   1302 	size_t len;
   1303 	char clname[16];
   1304 
   1305 	while (MORE() && isalpha((uch)PEEK()))
   1306 		NEXT();
   1307 	len = p->next - sp;
   1308 	if (len >= sizeof(clname) - 1) {
   1309 		SETERROR(REG_ECTYPE);
   1310 		return;
   1311 	}
   1312 	memcpy(clname, sp, len);
   1313 	clname[len] = '\0';
   1314 
   1315 	p_b_cclass_named(p, cs, clname);
   1316 }
   1317 
   1318 /*
   1319  - p_b_cclass_named - deal with a named character class
   1320  == static void p_b_cclass_named(struct parse *p, cset *cs, const char []);
   1321  */
   1322 static void
   1323 p_b_cclass_named(struct parse *p, cset *cs, const char clname[]) {
   1324 	wctype_t wct;
   1325 
   1326 	if ((wct = wctype(clname)) == 0) {
   1327 		SETERROR(REG_ECTYPE);
   1328 		return;
   1329 	}
   1330 	CHaddtype(p, cs, wct);
   1331 }
   1332 
   1333 /*
   1334  - p_b_eclass - parse an equivalence-class name and deal with it
   1335  == static void p_b_eclass(struct parse *p, cset *cs);
   1336  *
   1337  * This implementation is incomplete. xxx
   1338  */
   1339 static void
   1340 p_b_eclass(struct parse *p, cset *cs)
   1341 {
   1342 	wint_t c;
   1343 
   1344 	_DIAGASSERT(p != NULL);
   1345 	_DIAGASSERT(cs != NULL);
   1346 
   1347 	c = p_b_coll_elem(p, '=');
   1348 	CHadd(p, cs, c);
   1349 }
   1350 
   1351 /*
   1352  - p_b_symbol - parse a character or [..]ed multicharacter collating symbol
   1353  == static wint_t p_b_symbol(struct parse *p);
   1354  */
   1355 static wint_t			/* value of symbol */
   1356 p_b_symbol(struct parse *p)
   1357 {
   1358 	wint_t value;
   1359 
   1360 	_DIAGASSERT(p != NULL);
   1361 
   1362 	(void)REQUIRE(MORE(), REG_EBRACK);
   1363 	if (!EATTWO('[', '.'))
   1364 		return(WGETNEXT());
   1365 
   1366 	/* collating symbol */
   1367 	value = p_b_coll_elem(p, '.');
   1368 	(void)REQUIRE(EATTWO('.', ']'), REG_ECOLLATE);
   1369 	return(value);
   1370 }
   1371 
   1372 /*
   1373  - p_b_coll_elem - parse a collating-element name and look it up
   1374  == static wint_t p_b_coll_elem(struct parse *p, wint_t endc);
   1375  */
   1376 static wint_t			/* value of collating element */
   1377 p_b_coll_elem(struct parse *p,
   1378 	wint_t endc)		/* name ended by endc,']' */
   1379 {
   1380 	const char *sp = p->next;
   1381 	struct cname *cp;
   1382 	size_t len;
   1383 
   1384 	_DIAGASSERT(p != NULL);
   1385 
   1386 	while (MORE() && !SEETWO(endc, ']'))
   1387 		NEXT();
   1388 	if (!MORE()) {
   1389 		SETERROR(REG_EBRACK);
   1390 		return(0);
   1391 	}
   1392 	len = p->next - sp;
   1393 	for (cp = cnames; cp->name != NULL; cp++)
   1394 		if (strncmp(cp->name, sp, len) == 0 && strlen(cp->name) == len)
   1395 			return(cp->code);	/* known name */
   1396 #ifdef NLS
   1397 	mbstate_t mbs;
   1398 	wchar_t wc;
   1399 	size_t clen;
   1400 
   1401 	memset(&mbs, 0, sizeof(mbs));
   1402 	if ((clen = mbrtowc(&wc, sp, len, &mbs)) == len)
   1403 		return (wc);			/* single character */
   1404 	else if (clen == (size_t)-1 || clen == (size_t)-2)
   1405 		SETERROR(REG_ILLSEQ);
   1406 	else
   1407 		SETERROR(REG_ECOLLATE);		/* neither */
   1408 	return(0);
   1409 #else
   1410 	if (len == 1)
   1411 		return *sp;    /* single character */
   1412 	SETERROR(REG_ECOLLATE);                 /* neither */
   1413 	return 0;
   1414 #endif
   1415 }
   1416 
   1417 /*
   1418  - may_escape - determine whether 'ch' is escape-able in the current context
   1419  == static int may_escape(struct parse *p, const wint_t ch)
   1420  */
   1421 static bool
   1422 may_escape(struct parse *p, const wint_t ch)
   1423 {
   1424 
   1425 	if ((p->pflags & PFLAG_LEGACY_ESC) != 0)
   1426 		return (true);
   1427 	if (isalpha(ch) || ch == '\'' || ch == '`')
   1428 		return (false);
   1429 	return (true);
   1430 #ifdef NOTYET
   1431 	/*
   1432 	 * Build a whitelist of characters that may be escaped to produce an
   1433 	 * ordinary in the current context. This assumes that these have not
   1434 	 * been otherwise interpreted as a special character. Escaping an
   1435 	 * ordinary character yields undefined results according to
   1436 	 * IEEE 1003.1-2008. Some extensions (notably, some GNU extensions) take
   1437 	 * advantage of this and use escaped ordinary characters to provide
   1438 	 * special meaning, e.g. \b, \B, \w, \W, \s, \S.
   1439 	 */
   1440 	switch(ch) {
   1441 	case '|':
   1442 	case '+':
   1443 	case '?':
   1444 		/* The above characters may not be escaped in BREs */
   1445 		if (!(p->g->cflags&REG_EXTENDED))
   1446 			return (false);
   1447 		/* Fallthrough */
   1448 	case '(':
   1449 	case ')':
   1450 	case '{':
   1451 	case '}':
   1452 	case '.':
   1453 	case '[':
   1454 	case ']':
   1455 	case '\\':
   1456 	case '*':
   1457 	case '^':
   1458 	case '$':
   1459 		return (true);
   1460 	default:
   1461 		return (false);
   1462 	}
   1463 #endif
   1464 }
   1465 
   1466 /*
   1467  - othercase - return the case counterpart of an alphabetic
   1468  == static wint_t othercase(wint_t ch);
   1469  */
   1470 static wint_t			/* if no counterpart, return ch */
   1471 othercase(wint_t ch)
   1472 {
   1473 	assert(iswalpha(ch));
   1474 	if (iswupper(ch))
   1475 		return(towlower(ch));
   1476 	else if (iswlower(ch))
   1477 		return(towupper(ch));
   1478 	else			/* peculiar, but could happen */
   1479 		return(ch);
   1480 }
   1481 
   1482 /*
   1483  - bothcases - emit a dualcase version of a two-case character
   1484  == static void bothcases(struct parse *p, wint_t ch);
   1485  *
   1486  * Boy, is this implementation ever a kludge...
   1487  */
   1488 static void
   1489 bothcases(struct parse *p, wint_t ch)
   1490 {
   1491 	const char *oldnext = p->next;
   1492 	const char *oldend = p->end;
   1493 	char bracket[3 + MB_LEN_MAX];
   1494 	size_t n;
   1495 
   1496 	_DIAGASSERT(p != NULL);
   1497 
   1498 	assert(othercase(ch) != ch);	/* p_bracket() would recurse */
   1499 	p->next = bracket;
   1500 #ifdef NLS
   1501 	mbstate_t mbs;
   1502 	memset(&mbs, 0, sizeof(mbs));
   1503 	n = wcrtomb(bracket, ch, &mbs);
   1504 	assert(n != (size_t)-1);
   1505 #else
   1506 	n = 0;
   1507 	bracket[n++] = ch;
   1508 #endif
   1509 	bracket[n] = ']';
   1510 	bracket[n + 1] = '\0';
   1511 	p->end = bracket+n+1;
   1512 	p_bracket(p);
   1513 	assert(p->next == p->end);
   1514 	p->next = oldnext;
   1515 	p->end = oldend;
   1516 }
   1517 
   1518 /*
   1519  - ordinary - emit an ordinary character
   1520  == static void ordinary(struct parse *p, wint_t ch);
   1521  */
   1522 static void
   1523 ordinary(struct parse *p, wint_t ch)
   1524 {
   1525 	cset *cs;
   1526 
   1527 	_DIAGASSERT(p != NULL);
   1528 
   1529 	if ((p->g->cflags&REG_ICASE) && iswalpha(ch) && othercase(ch) != ch)
   1530 		bothcases(p, ch);
   1531 	else if ((wint_t)(ch & OPDMASK) == ch)
   1532 		EMIT(OCHAR, (size_t)ch);
   1533 	else {
   1534 		/*
   1535 		 * Kludge: character is too big to fit into an OCHAR operand.
   1536 		 * Emit a singleton set.
   1537 		 */
   1538 		if ((cs = allocset(p)) == NULL)
   1539 			return;
   1540 		CHadd(p, cs, ch);
   1541 		EMIT(OANYOF, (size_t)(cs - p->g->sets));
   1542 	}
   1543 }
   1544 
   1545 /*
   1546  - nonnewline - emit REG_NEWLINE version of OANY
   1547  == static void nonnewline(struct parse *p);
   1548  *
   1549  * Boy, is this implementation ever a kludge...
   1550  */
   1551 static void
   1552 nonnewline(struct parse *p)
   1553 {
   1554 	const char *oldnext = p->next;
   1555 	const char *oldend = p->end;
   1556 	char bracket[4];
   1557 
   1558 	_DIAGASSERT(p != NULL);
   1559 
   1560 	p->next = bracket;
   1561 	p->end = bracket+3;
   1562 	bracket[0] = '^';
   1563 	bracket[1] = '\n';
   1564 	bracket[2] = ']';
   1565 	bracket[3] = '\0';
   1566 	p_bracket(p);
   1567 	assert(p->next == bracket+3);
   1568 	p->next = oldnext;
   1569 	p->end = oldend;
   1570 }
   1571 
   1572 /*
   1573  - repeat - generate code for a bounded repetition, recursively if needed
   1574  == static void repeat(struct parse *p, sopno start, int from, int to);
   1575  */
   1576 static void
   1577 repeat(struct parse *p,
   1578 	sopno start,		/* operand from here to end of strip */
   1579 	int from,		/* repeated from this number */
   1580 	int to)			/* to this number of times (maybe INFINITY) */
   1581 {
   1582 	sopno finish = HERE();
   1583 #	define	N	2
   1584 #	define	INF	3
   1585 #	define	REP(f, t)	((f)*8 + (t))
   1586 #	define	MAP(n)	(((n) <= 1) ? (n) : ((n) == INFINITY) ? INF : N)
   1587 	sopno copy;
   1588 
   1589 	_DIAGASSERT(p != NULL);
   1590 
   1591 	if (p->error != 0)	/* head off possible runaway recursion */
   1592 		return;
   1593 
   1594 	assert(from <= to);
   1595 
   1596 	switch (REP(MAP(from), MAP(to))) {
   1597 	case REP(0, 0):			/* must be user doing this */
   1598 		DROP(finish-start);	/* drop the operand */
   1599 		break;
   1600 	case REP(0, 1):			/* as x{1,1}? */
   1601 	case REP(0, N):			/* as x{1,n}? */
   1602 	case REP(0, INF):		/* as x{1,}? */
   1603 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
   1604 		INSERT(OCH_, start);		/* offset is wrong... */
   1605 		repeat(p, start+1, 1, to);
   1606 		ASTERN(OOR1, start);
   1607 		AHEAD(start);			/* ... fix it */
   1608 		EMIT(OOR2, 0);
   1609 		AHEAD(THERE());
   1610 		ASTERN(O_CH, THERETHERE());
   1611 		break;
   1612 	case REP(1, 1):			/* trivial case */
   1613 		/* done */
   1614 		break;
   1615 	case REP(1, N):			/* as x?x{1,n-1} */
   1616 		/* KLUDGE: emit y? as (y|) until subtle bug gets fixed */
   1617 		INSERT(OCH_, start);
   1618 		ASTERN(OOR1, start);
   1619 		AHEAD(start);
   1620 		EMIT(OOR2, 0);			/* offset very wrong... */
   1621 		AHEAD(THERE());			/* ...so fix it */
   1622 		ASTERN(O_CH, THERETHERE());
   1623 		copy = dupl(p, start+1, finish+1);
   1624 		assert(copy == finish+4);
   1625 		repeat(p, copy, 1, to-1);
   1626 		break;
   1627 	case REP(1, INF):		/* as x+ */
   1628 		INSERT(OPLUS_, start);
   1629 		ASTERN(O_PLUS, start);
   1630 		break;
   1631 	case REP(N, N):			/* as xx{m-1,n-1} */
   1632 		copy = dupl(p, start, finish);
   1633 		repeat(p, copy, from-1, to-1);
   1634 		break;
   1635 	case REP(N, INF):		/* as xx{n-1,INF} */
   1636 		copy = dupl(p, start, finish);
   1637 		repeat(p, copy, from-1, to);
   1638 		break;
   1639 	default:			/* "can't happen" */
   1640 		SETERROR(REG_ASSERT);	/* just in case */
   1641 		break;
   1642 	}
   1643 }
   1644 
   1645 /*
   1646  - wgetnext - helper function for WGETNEXT() macro. Gets the next wide
   1647  - character from the parse struct, signals a REG_ILLSEQ error if the
   1648  - character can't be converted. Returns the number of bytes consumed.
   1649  */
   1650 static wint_t
   1651 wgetnext(struct parse *p)
   1652 {
   1653 #ifdef NLS
   1654 	mbstate_t mbs;
   1655 	wchar_t wc;
   1656 	size_t n;
   1657 
   1658 	memset(&mbs, 0, sizeof(mbs));
   1659 	n = mbrtowc(&wc, p->next, (size_t)(p->end - p->next), &mbs);
   1660 	if (n == (size_t)-1 || n == (size_t)-2) {
   1661 		SETERROR(REG_ILLSEQ);
   1662 		return (0);
   1663 	}
   1664 	if (n == 0)
   1665 		n = 1;
   1666 	p->next += n;
   1667 	return wc;
   1668 #else
   1669 	return *p->next++;
   1670 #endif
   1671 }
   1672 
   1673 /*
   1674  - seterr - set an error condition
   1675  == static int seterr(struct parse *p, int e);
   1676  */
   1677 static int			/* useless but makes type checking happy */
   1678 seterr(struct parse *p, int e)
   1679 {
   1680 
   1681 	_DIAGASSERT(p != NULL);
   1682 
   1683 	if (p->error == 0)	/* keep earliest error condition */
   1684 		p->error = e;
   1685 	p->next = nuls;		/* try to bring things to a halt */
   1686 	p->end = nuls;
   1687 	return(0);		/* make the return value well-defined */
   1688 }
   1689 
   1690 /*
   1691  - allocset - allocate a set of characters for []
   1692  == static cset *allocset(struct parse *p);
   1693  */
   1694 static cset *
   1695 allocset(struct parse *p)
   1696 {
   1697 	cset *cs, *ncs;
   1698 
   1699 	_DIAGASSERT(p != NULL);
   1700 
   1701 	ncs = reallocarray(p->g->sets, p->g->ncsets + 1, sizeof(*ncs));
   1702 	if (ncs == NULL) {
   1703 		SETERROR(REG_ESPACE);
   1704 		return (NULL);
   1705 	}
   1706 	p->g->sets = ncs;
   1707 	cs = &p->g->sets[p->g->ncsets++];
   1708 	memset(cs, 0, sizeof(*cs));
   1709 
   1710 	return(cs);
   1711 }
   1712 
   1713 /*
   1714  - freeset - free a now-unused set
   1715  == static void freeset(struct parse *p, cset *cs);
   1716  */
   1717 static void
   1718 freeset(struct parse *p, cset *cs)
   1719 {
   1720 	cset *top;
   1721 
   1722 	_DIAGASSERT(p != NULL);
   1723 	_DIAGASSERT(cs != NULL);
   1724 
   1725 	top = &p->g->sets[p->g->ncsets];
   1726 
   1727 	free(cs->wides);
   1728 	free(cs->ranges);
   1729 	free(cs->types);
   1730 	memset(cs, 0, sizeof(*cs));
   1731 	if (cs == top-1)	/* recover only the easy case */
   1732 		p->g->ncsets--;
   1733 }
   1734 
   1735 /*
   1736  - singleton - Determine whether a set contains only one character,
   1737  - returning it if so, otherwise returning OUT.
   1738  */
   1739 static wint_t
   1740 singleton(cset *cs)
   1741 {
   1742 	wint_t i, s, n;
   1743 
   1744 	for (i = n = 0; i < NC; i++)
   1745 		if (CHIN(cs, i)) {
   1746 			n++;
   1747 			s = i;
   1748 		}
   1749 	if (n == 1)
   1750 		return (s);
   1751 	if (cs->nwides == 1 && cs->nranges == 0 && cs->ntypes == 0 &&
   1752 	    cs->icase == 0)
   1753 		return (cs->wides[0]);
   1754 	/* Don't bother handling the other cases. */
   1755 	return (OUT);
   1756 }
   1757 
   1758 /*
   1759  - CHadd - add character to character set.
   1760  */
   1761 static void
   1762 CHadd(struct parse *p, cset *cs, wint_t ch)
   1763 {
   1764 	wint_t nch, *newwides;
   1765 
   1766 	_DIAGASSERT(p != NULL);
   1767 	_DIAGASSERT(cs != NULL);
   1768 
   1769 	assert(ch >= 0);
   1770 	if (ch < NC)
   1771 		cs->bmp[(unsigned)ch >> 3] |= 1 << (ch & 7);
   1772 	else {
   1773 		newwides = reallocarray(cs->wides, cs->nwides + 1,
   1774 		    sizeof(*cs->wides));
   1775 		if (newwides == NULL) {
   1776 			SETERROR(REG_ESPACE);
   1777 			return;
   1778 		}
   1779 		cs->wides = newwides;
   1780 		cs->wides[cs->nwides++] = ch;
   1781 	}
   1782 	if (cs->icase) {
   1783 		if ((nch = towlower(ch)) < NC)
   1784 			cs->bmp[(unsigned)nch >> 3] |= 1 << (nch & 7);
   1785 		if ((nch = towupper(ch)) < NC)
   1786 			cs->bmp[(unsigned)nch >> 3] |= 1 << (nch & 7);
   1787 	}
   1788 }
   1789 
   1790 /*
   1791  - CHaddrange - add all characters in the range [min,max] to a character set.
   1792  */
   1793 static void
   1794 CHaddrange(struct parse *p, cset *cs, wint_t min, wint_t max)
   1795 {
   1796 	crange *newranges;
   1797 
   1798 	_DIAGASSERT(p != NULL);
   1799 	_DIAGASSERT(cs != NULL);
   1800 
   1801 	for (; min < NC && min <= max; min++)
   1802 		CHadd(p, cs, min);
   1803 	if (min >= max)
   1804 		return;
   1805 	newranges = reallocarray(cs->ranges, cs->nranges + 1,
   1806 	    sizeof(*cs->ranges));
   1807 	if (newranges == NULL) {
   1808 		SETERROR(REG_ESPACE);
   1809 		return;
   1810 	}
   1811 	cs->ranges = newranges;
   1812 	cs->ranges[cs->nranges].min = min;
   1813 	cs->ranges[cs->nranges].max = max;
   1814 	cs->nranges++;
   1815 }
   1816 
   1817 /*
   1818  - CHaddtype - add all characters of a certain type to a character set.
   1819  */
   1820 static void
   1821 CHaddtype(struct parse *p, cset *cs, wctype_t wct)
   1822 {
   1823 	wint_t i;
   1824 	wctype_t *newtypes;
   1825 
   1826 	_DIAGASSERT(p != NULL);
   1827 	_DIAGASSERT(cs != NULL);
   1828 
   1829 	for (i = 0; i < NC; i++)
   1830 		if (iswctype(i, wct))
   1831 			CHadd(p, cs, i);
   1832 	newtypes = reallocarray(cs->types, cs->ntypes + 1,
   1833 	    sizeof(*cs->types));
   1834 	if (newtypes == NULL) {
   1835 		SETERROR(REG_ESPACE);
   1836 		return;
   1837 	}
   1838 	cs->types = newtypes;
   1839 	cs->types[cs->ntypes++] = wct;
   1840 }
   1841 
   1842 /*
   1843  - dupl - emit a duplicate of a bunch of sops
   1844  == static sopno dupl(struct parse *p, sopno start, sopno finish);
   1845  */
   1846 static sopno			/* start of duplicate */
   1847 dupl(struct parse *p,
   1848 	sopno start,		/* from here */
   1849 	sopno finish)		/* to this less one */
   1850 {
   1851 	sopno ret = HERE();
   1852 	sopno len = finish - start;
   1853 
   1854 	_DIAGASSERT(p != NULL);
   1855 
   1856 	assert(finish >= start);
   1857 	if (len == 0)
   1858 		return(ret);
   1859 	if (!enlarge(p, p->ssize + len)) /* this many unexpected additions */
   1860 		return(ret);
   1861 	(void) memcpy(p->strip + p->slen,
   1862 	    p->strip + start, len * sizeof(*p->strip));
   1863 	p->slen += len;
   1864 	return(ret);
   1865 }
   1866 
   1867 /*
   1868  - doemit - emit a strip operator
   1869  == static void doemit(struct parse *p, sop op, size_t opnd);
   1870  *
   1871  * It might seem better to implement this as a macro with a function as
   1872  * hard-case backup, but it's just too big and messy unless there are
   1873  * some changes to the data structures.  Maybe later.
   1874  */
   1875 static void
   1876 doemit(struct parse *p, sop op, size_t opnd)
   1877 {
   1878 	/* avoid making error situations worse */
   1879 	if (p->error != 0)
   1880 		return;
   1881 
   1882 	_DIAGASSERT(p != NULL);
   1883 
   1884 	/* deal with oversize operands ("can't happen", more or less) */
   1885 	assert(opnd < 1<<OPSHIFT);
   1886 
   1887 	/* deal with undersized strip */
   1888 	if (p->slen >= p->ssize)
   1889 		if (!enlarge(p, (p->ssize+1) / 2 * 3))	/* +50% */
   1890 			return;
   1891 
   1892 	/* finally, it's all reduced to the easy case */
   1893 	p->strip[p->slen++] = (sopno)SOP(op, opnd);
   1894 }
   1895 
   1896 /*
   1897  - doinsert - insert a sop into the strip
   1898  == static void doinsert(struct parse *p, sop op, size_t opnd, sopno pos);
   1899  */
   1900 static void
   1901 doinsert(struct parse *p, sop op, size_t opnd, sopno pos)
   1902 {
   1903 	sopno sn;
   1904 	sop s;
   1905 	int i;
   1906 
   1907 	_DIAGASSERT(p != NULL);
   1908 
   1909 	/* avoid making error situations worse */
   1910 	if (p->error != 0)
   1911 		return;
   1912 
   1913 	sn = HERE();
   1914 	EMIT(op, opnd);		/* do checks, ensure space */
   1915 	assert(HERE() == sn+1);
   1916 	s = p->strip[sn];
   1917 
   1918 	/* adjust paren pointers */
   1919 	assert(pos > 0);
   1920 	for (i = 1; i < NPAREN; i++) {
   1921 		if (p->pbegin[i] >= pos) {
   1922 			p->pbegin[i]++;
   1923 		}
   1924 		if (p->pend[i] >= pos) {
   1925 			p->pend[i]++;
   1926 		}
   1927 	}
   1928 
   1929 	memmove(&p->strip[pos+1], &p->strip[pos],
   1930 	    (HERE()-pos-1)*sizeof(*p->strip));
   1931 	p->strip[pos] = s;
   1932 }
   1933 
   1934 /*
   1935  - dofwd - complete a forward reference
   1936  == static void dofwd(struct parse *p, sopno pos, sop value);
   1937  */
   1938 static void
   1939 dofwd(struct parse *p, sopno pos, sop value)
   1940 {
   1941 
   1942 	_DIAGASSERT(p != NULL);
   1943 
   1944 	/* avoid making error situations worse */
   1945 	if (p->error != 0)
   1946 		return;
   1947 
   1948 	assert(value < 1<<OPSHIFT);
   1949 	p->strip[pos] = OP(p->strip[pos]) | value;
   1950 }
   1951 
   1952 /*
   1953  - enlarge - enlarge the strip
   1954  == static int enlarge(struct parse *p, sopno size);
   1955  */
   1956 static int
   1957 enlarge(struct parse *p, sopno size)
   1958 {
   1959 	sop *sp;
   1960 
   1961 	_DIAGASSERT(p != NULL);
   1962 
   1963 	if (p->ssize >= size)
   1964 		return 1;
   1965 
   1966 	sp = reallocarray(p->strip, size, sizeof(*p->strip));
   1967 	if (sp == NULL) {
   1968 		SETERROR(REG_ESPACE);
   1969 		return 0;
   1970 	}
   1971 	p->strip = sp;
   1972 	p->ssize = size;
   1973 	return 1;
   1974 }
   1975 
   1976 /*
   1977  - stripsnug - compact the strip
   1978  == static void stripsnug(struct parse *p, struct re_guts *g);
   1979  */
   1980 static void
   1981 stripsnug(struct parse *p, struct re_guts *g)
   1982 {
   1983 
   1984 	_DIAGASSERT(p != NULL);
   1985 	_DIAGASSERT(g != NULL);
   1986 
   1987 	g->nstates = p->slen;
   1988 	g->strip = reallocarray(p->strip, p->slen, sizeof(*p->strip));
   1989 	if (g->strip == NULL) {
   1990 		SETERROR(REG_ESPACE);
   1991 		g->strip = p->strip;
   1992 	}
   1993 }
   1994 
   1995 /*
   1996  - findmust - fill in must and mlen with longest mandatory literal string
   1997  == static void findmust(struct parse *p, struct re_guts *g);
   1998  *
   1999  * This algorithm could do fancy things like analyzing the operands of |
   2000  * for common subsequences.  Someday.  This code is simple and finds most
   2001  * of the interesting cases.
   2002  *
   2003  * Note that must and mlen got initialized during setup.
   2004  */
   2005 static void
   2006 findmust(struct parse *p, struct re_guts *g)
   2007 {
   2008 	sop *scan;
   2009 	sop *start = NULL;
   2010 	sop *newstart = NULL;
   2011 	sopno newlen;
   2012 	sop s;
   2013 	char *cp;
   2014 	int offset;
   2015 	mbstate_t mbs;
   2016 
   2017 	_DIAGASSERT(p != NULL);
   2018 	_DIAGASSERT(g != NULL);
   2019 
   2020 	/* avoid making error situations worse */
   2021 	if (p->error != 0)
   2022 		return;
   2023 
   2024 #ifdef notyet
   2025 	/*
   2026 	 * It's not generally safe to do a ``char'' substring search on
   2027 	 * multibyte character strings, but it's safe for at least
   2028 	 * UTF-8 (see RFC 3629).
   2029 	 */
   2030 	if (MB_CUR_MAX > 1 &&
   2031 	    strcmp(_CurrentRuneLocale->__encoding, "UTF-8") != 0)
   2032 		return;
   2033 #endif
   2034 
   2035 	/* find the longest OCHAR sequence in strip */
   2036 	newlen = 0;
   2037 	offset = 0;
   2038 	g->moffset = 0;
   2039 	scan = g->strip + 1;
   2040 	do {
   2041 		s = *scan++;
   2042 		switch (OP(s)) {
   2043 		case OCHAR:		/* sequence member */
   2044 			if (newlen == 0) {		/* new sequence */
   2045 				memset(&mbs, 0, sizeof(mbs));
   2046 				newstart = scan - 1;
   2047 			}
   2048 #ifdef NLS
   2049 			char buf[MB_LEN_MAX];
   2050 			size_t clen = wcrtomb(buf, (int)OPND(s), &mbs);
   2051 			if (clen == (size_t)-1)
   2052 				goto toohard;
   2053 			newlen += (sopno)clen;
   2054 #else
   2055 			newlen++;
   2056 #endif
   2057 			break;
   2058 		case OPLUS_:		/* things that don't break one */
   2059 		case OLPAREN:
   2060 		case ORPAREN:
   2061 			break;
   2062 		case OQUEST_:		/* things that must be skipped */
   2063 		case OCH_:
   2064 			offset = altoffset(scan, offset);
   2065 			scan--;
   2066 			do {
   2067 				scan += OPND(s);
   2068 				s = *scan;
   2069 				/* assert() interferes w debug printouts */
   2070 				if (OP(s) != O_QUEST &&
   2071 				    OP(s) != O_CH && OP(s) != OOR2) {
   2072 					g->iflags |= BAD;
   2073 					return;
   2074 				}
   2075 			} while (OP(s) != O_QUEST && OP(s) != O_CH);
   2076 			/* FALLTHROUGH */
   2077 		case OBOW:		/* things that break a sequence */
   2078 		case OEOW:
   2079 		case OBOL:
   2080 		case OEOL:
   2081 		case OBOS:
   2082 		case OEOS:
   2083 		case OWBND:
   2084 		case ONWBND:
   2085 		case O_QUEST:
   2086 		case O_CH:
   2087 		case OEND:
   2088 			if (newlen > (sopno)g->mlen) {		/* ends one */
   2089 				start = newstart;
   2090 				g->mlen = newlen;
   2091 				if (offset > -1) {
   2092 					g->moffset += offset;
   2093 					offset = newlen;
   2094 				} else
   2095 					g->moffset = offset;
   2096 			} else {
   2097 				if (offset > -1)
   2098 					offset += newlen;
   2099 			}
   2100 			newlen = 0;
   2101 			break;
   2102 		case OANY:
   2103 			if (newlen > (sopno)g->mlen) {		/* ends one */
   2104 				start = newstart;
   2105 				g->mlen = newlen;
   2106 				if (offset > -1) {
   2107 					g->moffset += offset;
   2108 					offset = newlen;
   2109 				} else
   2110 					g->moffset = offset;
   2111 			} else {
   2112 				if (offset > -1)
   2113 					offset += newlen;
   2114 			}
   2115 			if (offset > -1)
   2116 				offset++;
   2117 			newlen = 0;
   2118 			break;
   2119 		case OANYOF:		/* may or may not invalidate offset */
   2120 			/* First, everything as OANY */
   2121 			if (newlen > (sopno)g->mlen) {		/* ends one */
   2122 				start = newstart;
   2123 				g->mlen = newlen;
   2124 				if (offset > -1) {
   2125 					g->moffset += offset;
   2126 					offset = newlen;
   2127 				} else
   2128 					g->moffset = offset;
   2129 			} else {
   2130 				if (offset > -1)
   2131 					offset += newlen;
   2132 			}
   2133 			if (offset > -1)
   2134 				offset++;
   2135 			newlen = 0;
   2136 			break;
   2137 #ifdef NLS
   2138 		toohard:/*FALLTHROUGH*/
   2139 #endif
   2140 		default:
   2141 			/* Anything here makes it impossible or too hard
   2142 			 * to calculate the offset -- so we give up;
   2143 			 * save the last known good offset, in case the
   2144 			 * must sequence doesn't occur later.
   2145 			 */
   2146 			if (newlen > (sopno)g->mlen) {		/* ends one */
   2147 				start = newstart;
   2148 				g->mlen = newlen;
   2149 				if (offset > -1)
   2150 					g->moffset += offset;
   2151 				else
   2152 					g->moffset = offset;
   2153 			}
   2154 			offset = -1;
   2155 			newlen = 0;
   2156 			break;
   2157 		}
   2158 	} while (OP(s) != OEND);
   2159 
   2160 	if (g->mlen == 0) {		/* there isn't one */
   2161 		g->moffset = -1;
   2162 		return;
   2163 	}
   2164 
   2165 	/* turn it into a character string */
   2166 	g->must = malloc((size_t)g->mlen + 1);
   2167 	if (g->must == NULL) {		/* argh; just forget it */
   2168 		g->mlen = 0;
   2169 		g->moffset = -1;
   2170 		return;
   2171 	}
   2172 	cp = g->must;
   2173 	scan = start;
   2174 	memset(&mbs, 0, sizeof(mbs));
   2175 	while (cp < g->must + g->mlen) {
   2176 		while (OP(s = *scan++) != OCHAR)
   2177 			continue;
   2178 #ifdef NLS
   2179 		size_t clen = wcrtomb(cp, (int)OPND(s), &mbs);
   2180 		assert(clen != (size_t)-1);
   2181 		cp += clen;
   2182 #else
   2183 		*cp++ = OPND(s);
   2184 #endif
   2185 	}
   2186 	assert(cp == g->must + g->mlen);
   2187 	*cp++ = '\0';		/* just on general principles */
   2188 }
   2189 
   2190 /*
   2191  - altoffset - choose biggest offset among multiple choices
   2192  == static int altoffset(sop *scan, int offset);
   2193  *
   2194  * Compute, recursively if necessary, the largest offset among multiple
   2195  * re paths.
   2196  */
   2197 static int
   2198 altoffset(sop *scan, int offset)
   2199 {
   2200 	int largest;
   2201 	int try;
   2202 	sop s;
   2203 
   2204 	_DIAGASSERT(scan != NULL);
   2205 
   2206 	/* If we gave up already on offsets, return */
   2207 	if (offset == -1)
   2208 		return -1;
   2209 
   2210 	largest = 0;
   2211 	try = 0;
   2212 	s = *scan++;
   2213 	while (OP(s) != O_QUEST && OP(s) != O_CH) {
   2214 		switch (OP(s)) {
   2215 		case OOR1:
   2216 			if (try > largest)
   2217 				largest = try;
   2218 			try = 0;
   2219 			break;
   2220 		case OQUEST_:
   2221 		case OCH_:
   2222 			try = altoffset(scan, try);
   2223 			if (try == -1)
   2224 				return -1;
   2225 			scan--;
   2226 			do {
   2227 				scan += OPND(s);
   2228 				s = *scan;
   2229 				if (OP(s) != O_QUEST &&
   2230 				    OP(s) != O_CH && OP(s) != OOR2)
   2231 					return -1;
   2232 			} while (OP(s) != O_QUEST && OP(s) != O_CH);
   2233 			/* We must skip to the next position, or we'll
   2234 			 * leave altoffset() too early.
   2235 			 */
   2236 			scan++;
   2237 			break;
   2238 		case OANYOF:
   2239 		case OCHAR:
   2240 		case OANY:
   2241 			try++;
   2242 			/*FALLTHROUGH*/
   2243 		case OBOW:
   2244 		case OEOW:
   2245 		case OWBND:
   2246 		case ONWBND:
   2247 		case OLPAREN:
   2248 		case ORPAREN:
   2249 		case OOR2:
   2250 			break;
   2251 		default:
   2252 			try = -1;
   2253 			break;
   2254 		}
   2255 		if (try == -1)
   2256 			return -1;
   2257 		s = *scan++;
   2258 	}
   2259 
   2260 	if (try > largest)
   2261 		largest = try;
   2262 
   2263 	return largest+offset;
   2264 }
   2265 
   2266 /*
   2267  - computejumps - compute char jumps for BM scan
   2268  == static void computejumps(struct parse *p, struct re_guts *g);
   2269  *
   2270  * This algorithm assumes g->must exists and is has size greater than
   2271  * zero. It's based on the algorithm found on Computer Algorithms by
   2272  * Sara Baase.
   2273  *
   2274  * A char jump is the number of characters one needs to jump based on
   2275  * the value of the character from the text that was mismatched.
   2276  */
   2277 static void
   2278 computejumps(struct parse *p, struct re_guts *g)
   2279 {
   2280 	int ch;
   2281 	size_t mindex;
   2282 
   2283 	_DIAGASSERT(p != NULL);
   2284 	_DIAGASSERT(g != NULL);
   2285 
   2286 	/* Avoid making errors worse */
   2287 	if (p->error != 0)
   2288 		return;
   2289 
   2290 	g->charjump = calloc((NC_MAX + 1), sizeof(*g->charjump));
   2291 	if (g->charjump == NULL)	/* Not a fatal error */
   2292 		return;
   2293 	/* Adjust for signed chars, if necessary */
   2294 	g->charjump = &g->charjump[-(CHAR_MIN)];
   2295 
   2296 	/* If the character does not exist in the pattern, the jump
   2297 	 * is equal to the number of characters in the pattern.
   2298 	 */
   2299 	for (ch = CHAR_MIN; ch < (CHAR_MAX + 1); ch++)
   2300 		g->charjump[ch] = g->mlen;
   2301 
   2302 	/* If the character does exist, compute the jump that would
   2303 	 * take us to the last character in the pattern equal to it
   2304 	 * (notice that we match right to left, so that last character
   2305 	 * is the first one that would be matched).
   2306 	 */
   2307 	for (mindex = 0; mindex < g->mlen; mindex++)
   2308 		g->charjump[(int)g->must[mindex]] = g->mlen - mindex - 1;
   2309 }
   2310 
   2311 /*
   2312  - computematchjumps - compute match jumps for BM scan
   2313  == static void computematchjumps(struct parse *p, struct re_guts *g);
   2314  *
   2315  * This algorithm assumes g->must exists and is has size greater than
   2316  * zero. It's based on the algorithm found on Computer Algorithms by
   2317  * Sara Baase.
   2318  *
   2319  * A match jump is the number of characters one needs to advance based
   2320  * on the already-matched suffix.
   2321  * Notice that all values here are minus (g->mlen-1), because of the way
   2322  * the search algorithm works.
   2323  */
   2324 static void
   2325 computematchjumps(struct parse *p, struct re_guts *g)
   2326 {
   2327 	size_t mindex;		/* General "must" iterator */
   2328 	size_t suffix;		/* Keeps track of matching suffix */
   2329 	size_t ssuffix;		/* Keeps track of suffixes' suffix */
   2330 	size_t* pmatches;	/* pmatches[k] points to the next i
   2331 				 * such that i+1...mlen is a substring
   2332 				 * of k+1...k+mlen-i-1
   2333 				 */
   2334 
   2335 	_DIAGASSERT(p != NULL);
   2336 	_DIAGASSERT(g != NULL);
   2337 
   2338 	/* Avoid making errors worse */
   2339 	if (p->error != 0)
   2340 		return;
   2341 
   2342 	pmatches = calloc(g->mlen, sizeof(*pmatches));
   2343 	if (pmatches == NULL) {
   2344 		g->matchjump = NULL;
   2345 		return;
   2346 	}
   2347 
   2348 	g->matchjump = calloc(g->mlen, sizeof(*g->matchjump));
   2349 	if (g->matchjump == NULL) {	/* Not a fatal error */
   2350 		free(pmatches);
   2351 		return;
   2352 	}
   2353 
   2354 	/* Set maximum possible jump for each character in the pattern */
   2355 	for (mindex = 0; mindex < g->mlen; mindex++)
   2356 		g->matchjump[mindex] = 2 * g->mlen - mindex - 1;
   2357 
   2358 	/* Compute pmatches[] */
   2359 	for (suffix = mindex = g->mlen; mindex-- > 0; suffix--) {
   2360 		pmatches[mindex] = suffix;
   2361 
   2362 		/* If a mismatch is found, interrupting the substring,
   2363 		 * compute the matchjump for that position. If no
   2364 		 * mismatch is found, then a text substring mismatched
   2365 		 * against the suffix will also mismatch against the
   2366 		 * substring.
   2367 		 */
   2368 		while (suffix < g->mlen
   2369 		    && g->must[mindex] != g->must[suffix]) {
   2370 			g->matchjump[suffix] = MIN(g->matchjump[suffix],
   2371 			    g->mlen - mindex - 1);
   2372 			suffix = pmatches[suffix];
   2373 		}
   2374 	}
   2375 
   2376 	/* Compute the matchjump up to the last substring found to jump
   2377 	 * to the beginning of the largest must pattern prefix matching
   2378 	 * it's own suffix.
   2379 	 */
   2380 	for (mindex = 0; mindex <= suffix; mindex++)
   2381 		g->matchjump[mindex] = MIN(g->matchjump[mindex],
   2382 		    g->mlen + suffix - mindex);
   2383 
   2384         ssuffix = pmatches[suffix];
   2385         while (suffix < g->mlen) {
   2386                 while (suffix <= ssuffix && suffix < g->mlen) {
   2387                         g->matchjump[suffix] = MIN(g->matchjump[suffix],
   2388 			    g->mlen + ssuffix - suffix);
   2389                         suffix++;
   2390                 }
   2391 		if (suffix < g->mlen)
   2392                 	ssuffix = pmatches[ssuffix];
   2393         }
   2394 
   2395 	free(pmatches);
   2396 }
   2397 
   2398 /*
   2399  - pluscount - count + nesting
   2400  == static sopno pluscount(struct parse *p, struct re_guts *g);
   2401  */
   2402 static sopno			/* nesting depth */
   2403 pluscount(struct parse *p, struct re_guts *g)
   2404 {
   2405 	sop *scan;
   2406 	sop s;
   2407 	sopno plusnest = 0;
   2408 	sopno maxnest = 0;
   2409 
   2410 	_DIAGASSERT(p != NULL);
   2411 	_DIAGASSERT(g != NULL);
   2412 
   2413 	if (p->error != 0)
   2414 		return(0);	/* there may not be an OEND */
   2415 
   2416 	scan = g->strip + 1;
   2417 	do {
   2418 		s = *scan++;
   2419 		switch (OP(s)) {
   2420 		case OPLUS_:
   2421 			plusnest++;
   2422 			break;
   2423 		case O_PLUS:
   2424 			if (plusnest > maxnest)
   2425 				maxnest = plusnest;
   2426 			plusnest--;
   2427 			break;
   2428 		}
   2429 	} while (OP(s) != OEND);
   2430 	if (plusnest != 0)
   2431 		g->iflags |= BAD;
   2432 	return(maxnest);
   2433 }
   2434