Home | History | Annotate | Line # | Download | only in regexp
      1 /*
      2  * regcomp and regexec -- regsub and regerror are elsewhere
      3  *
      4  *	Copyright (c) 1986 by University of Toronto.
      5  *	Written by Henry Spencer.  Not derived from licensed software.
      6  *
      7  *	Permission is granted to anyone to use this software for any
      8  *	purpose on any computer system, and to redistribute it freely,
      9  *	subject to the following restrictions:
     10  *
     11  *	1. The author is not responsible for the consequences of use of
     12  *		this software, no matter how awful, even if they arise
     13  *		from defects in it.
     14  *
     15  *	2. The origin of this software must not be misrepresented, either
     16  *		by explicit claim or by omission.
     17  *
     18  *	3. Altered versions must be plainly marked as such, and must not
     19  *		be misrepresented as being the original software.
     20  *** THIS IS AN ALTERED VERSION.  It was altered by John Gilmore,
     21  *** hoptoad!gnu, on 27 Dec 1986, to add \n as an alternative to |
     22  *** to assist in implementing egrep.
     23  *** THIS IS AN ALTERED VERSION.  It was altered by John Gilmore,
     24  *** hoptoad!gnu, on 27 Dec 1986, to add \< and \> for word-matching
     25  *** as in BSD grep and ex.
     26  *** THIS IS AN ALTERED VERSION.  It was altered by John Gilmore,
     27  *** hoptoad!gnu, on 28 Dec 1986, to optimize characters quoted with \.
     28  *** THIS IS AN ALTERED VERSION.  It was altered by James A. Woods,
     29  *** ames!jaw, on 19 June 1987, to quash a regcomp() redundancy.
     30  *
     31  * Beware that some of this code is subtly aware of the way operator
     32  * precedence is structured in regular expressions.  Serious changes in
     33  * regular-expression syntax might require a total rethink.
     34  */
     35 
     36 #include <sys/cdefs.h>
     37 #if defined(LIBC_SCCS) && !defined(lint)
     38 __RCSID("$NetBSD: regexp.c,v 1.19 2016/01/26 16:05:18 christos Exp $");
     39 #endif /* LIBC_SCCS and not lint */
     40 
     41 #include <ctype.h>
     42 #include <regexp.h>
     43 #include <stdio.h>
     44 #include <stdlib.h>
     45 #include <string.h>
     46 #include "regmagic.h"
     47 
     48 /*
     49  * The "internal use only" fields in regexp.h are present to pass info from
     50  * compile to execute that permits the execute phase to run lots faster on
     51  * simple cases.  They are:
     52  *
     53  * regstart	char that must begin a match; '\0' if none obvious
     54  * reganch	is the match anchored (at beginning-of-line only)?
     55  * regmust	string (pointer into program) that match must include, or NULL
     56  * regmlen	length of regmust string
     57  *
     58  * Regstart and reganch permit very fast decisions on suitable starting points
     59  * for a match, cutting down the work a lot.  Regmust permits fast rejection
     60  * of lines that cannot possibly match.  The regmust tests are costly enough
     61  * that regcomp() supplies a regmust only if the r.e. contains something
     62  * potentially expensive (at present, the only such thing detected is * or +
     63  * at the start of the r.e., which can involve a lot of backup).  Regmlen is
     64  * supplied because the test in regexec() needs it and regcomp() is computing
     65  * it anyway.
     66  */
     67 
     68 /*
     69  * Structure for regexp "program".  This is essentially a linear encoding
     70  * of a nondeterministic finite-state machine (aka syntax charts or
     71  * "railroad normal form" in parsing technology).  Each node is an opcode
     72  * plus a "next" pointer, possibly plus an operand.  "Next" pointers of
     73  * all nodes except BRANCH implement concatenation; a "next" pointer with
     74  * a BRANCH on both ends of it is connecting two alternatives.  (Here we
     75  * have one of the subtle syntax dependencies:  an individual BRANCH (as
     76  * opposed to a collection of them) is never concatenated with anything
     77  * because of operator precedence.)  The operand of some types of node is
     78  * a literal string; for others, it is a node leading into a sub-FSM.  In
     79  * particular, the operand of a BRANCH node is the first node of the branch.
     80  * (NB this is *not* a tree structure:  the tail of the branch connects
     81  * to the thing following the set of BRANCHes.)  The opcodes are:
     82  */
     83 
     84 /* definition	number	opnd?	meaning */
     85 #define	END	0	/* no	End of program. */
     86 #define	BOL	1	/* no	Match "" at beginning of line. */
     87 #define	EOL	2	/* no	Match "" at end of line. */
     88 #define	ANY	3	/* no	Match any one character. */
     89 #define	ANYOF	4	/* str	Match any character in this string. */
     90 #define	ANYBUT	5	/* str	Match any character not in this string. */
     91 #define	BRANCH	6	/* node	Match this alternative, or the next... */
     92 #define	BACK	7	/* no	Match "", "next" ptr points backward. */
     93 #define	EXACTLY	8	/* str	Match this string. */
     94 #define	NOTHING	9	/* no	Match empty string. */
     95 #define	STAR	10	/* node	Match this (simple) thing 0 or more times. */
     96 #define	PLUS	11	/* node	Match this (simple) thing 1 or more times. */
     97 #define	WORDA	12	/* no	Match "" at wordchar, where prev is nonword */
     98 #define	WORDZ	13	/* no	Match "" at nonwordchar, where prev is word */
     99 #define	OPEN	20	/* no	Mark this point in input as start of #n. */
    100 			/*	OPEN+1 is number 1, etc. */
    101 #define	CLOSE	30	/* no	Analogous to OPEN. */
    102 
    103 /*
    104  * Opcode notes:
    105  *
    106  * BRANCH	The set of branches constituting a single choice are hooked
    107  *		together with their "next" pointers, since precedence prevents
    108  *		anything being concatenated to any individual branch.  The
    109  *		"next" pointer of the last BRANCH in a choice points to the
    110  *		thing following the whole choice.  This is also where the
    111  *		final "next" pointer of each individual branch points; each
    112  *		branch starts with the operand node of a BRANCH node.
    113  *
    114  * BACK		Normal "next" pointers all implicitly point forward; BACK
    115  *		exists to make loop structures possible.
    116  *
    117  * STAR,PLUS	'?', and complex '*' and '+', are implemented as circular
    118  *		BRANCH structures using BACK.  Simple cases (one character
    119  *		per match) are implemented with STAR and PLUS for speed
    120  *		and to minimize recursive plunges.
    121  *
    122  * OPEN,CLOSE	...are numbered at compile time.
    123  */
    124 
    125 /*
    126  * A node is one char of opcode followed by two chars of "next" pointer.
    127  * "Next" pointers are stored as two 8-bit pieces, high order first.  The
    128  * value is a positive offset from the opcode of the node containing it.
    129  * An operand, if any, simply follows the node.  (Note that much of the
    130  * code generation knows about this implicit relationship.)
    131  *
    132  * Using two bytes for the "next" pointer is vast overkill for most things,
    133  * but allows patterns to get big without disasters.
    134  */
    135 #define	OP(p)	(*(p))
    136 #define	NEXT(p)	(((*((p)+1)&0377)<<8) + (*((p)+2)&0377))
    137 #define	OPERAND(p)	((p) + 3)
    138 
    139 /*
    140  * See regmagic.h for one further detail of program structure.
    141  */
    142 
    143 
    144 /*
    145  * Utility definitions.
    146  */
    147 #ifndef CHARBITS
    148 #define	UCHARAT(p)	((int)*(unsigned char *)(p))
    149 #else
    150 #define	UCHARAT(p)	((int)*(p)&CHARBITS)
    151 #endif
    152 
    153 #define	FAIL(m)	{ regerror(m); return(NULL); }
    154 #define	ISMULT(c)	((c) == '*' || (c) == '+' || (c) == '?')
    155 
    156 /*
    157  * Flags to be passed up and down.
    158  */
    159 #define	HASWIDTH	01	/* Known never to match null string. */
    160 #define	SIMPLE		02	/* Simple enough to be STAR/PLUS operand. */
    161 #define	SPSTART		04	/* Starts with * or +. */
    162 #define	WORST		0	/* Worst case. */
    163 
    164 /*
    165  * Global work variables for regcomp().
    166  */
    167 static char *regparse;		/* Input-scan pointer. */
    168 static int regnpar;		/* () count. */
    169 static char regdummy;
    170 static char *regcode;		/* Code-emit pointer; &regdummy = don't. */
    171 static long regsize;		/* Code size. */
    172 
    173 /*
    174  * Forward declarations for regcomp()'s friends.
    175  */
    176 #ifndef STATIC
    177 #define	STATIC	static
    178 #endif
    179 STATIC char *reg __P((int, int *));
    180 STATIC char *regbranch __P((int *));
    181 STATIC char *regpiece __P((int *));
    182 STATIC char *regatom __P((int *));
    183 STATIC char *regnode __P((int));
    184 STATIC char *regnext __P((char *));
    185 STATIC void regc __P((int));
    186 STATIC void reginsert __P((int, char *));
    187 STATIC void regtail __P((char *, char *));
    188 STATIC void regoptail __P((char *, char *));
    189 #ifdef STRCSPN
    190 STATIC int strcspn __P((char *, char *));
    191 #endif
    192 
    193 /*
    194  - regcomp - compile a regular expression into internal code
    195  *
    196  * We can't allocate space until we know how big the compiled form will be,
    197  * but we can't compile it (and thus know how big it is) until we've got a
    198  * place to put the code.  So we cheat:  we compile it twice, once with code
    199  * generation turned off and size counting turned on, and once "for real".
    200  * This also means that we don't allocate space until we are sure that the
    201  * thing really will compile successfully, and we never have to move the
    202  * code and thus invalidate pointers into it.  (Note that it has to be in
    203  * one piece because free() must be able to free it all.)
    204  *
    205  * Beware that the optimization-preparation code in here knows about some
    206  * of the structure of the compiled regexp.
    207  */
    208 regexp *
    209 __compat_regcomp(expn)
    210 const char *expn;
    211 {
    212 	regexp *r;
    213 	char *scan;
    214 	char *longest;
    215 	int len;
    216 	int flags;
    217 
    218 	if (expn == NULL)
    219 		FAIL("NULL argument");
    220 
    221 	/* First pass: determine size, legality. */
    222 #ifdef notdef
    223 	if (expn[0] == '.' && expn[1] == '*') expn += 2;  /* aid grep */
    224 #endif
    225 	/* LINTED const castaway */
    226 	regparse = (char *)expn;
    227 	regnpar = 1;
    228 	regsize = 0L;
    229 	regcode = &regdummy;
    230 	regc(MAGIC);
    231 	if (reg(0, &flags) == NULL)
    232 		return(NULL);
    233 
    234 	/* Small enough for pointer-storage convention? */
    235 	if (regsize >= 32767L)		/* Probably could be 65535L. */
    236 		FAIL("regexp too big");
    237 
    238 	/* Allocate space. */
    239 	r = malloc(sizeof(regexp) + (unsigned)regsize);
    240 	if (r == NULL)
    241 		FAIL("out of space");
    242 
    243 	/* Second pass: emit code. */
    244 	/* LINTED const castaway */
    245 	regparse = (char *)expn;
    246 	regnpar = 1;
    247 	regcode = r->program;
    248 	regc(MAGIC);
    249 	if (reg(0, &flags) == NULL) {
    250 		free(r);
    251 		return(NULL);
    252 	}
    253 
    254 	/* Dig out information for optimizations. */
    255 	r->regstart = '\0';	/* Worst-case defaults. */
    256 	r->reganch = 0;
    257 	r->regmust = NULL;
    258 	r->regmlen = 0;
    259 	scan = r->program+1;			/* First BRANCH. */
    260 	if (OP(regnext(scan)) == END) {		/* Only one top-level choice. */
    261 		scan = OPERAND(scan);
    262 
    263 		/* Starting-point info. */
    264 		if (OP(scan) == EXACTLY)
    265 			r->regstart = *OPERAND(scan);
    266 		else if (OP(scan) == BOL)
    267 			r->reganch++;
    268 
    269 		/*
    270 		 * If there's something expensive in the r.e., find the
    271 		 * longest literal string that must appear and make it the
    272 		 * regmust.  Resolve ties in favor of later strings, since
    273 		 * the regstart check works with the beginning of the r.e.
    274 		 * and avoiding duplication strengthens checking.  Not a
    275 		 * strong reason, but sufficient in the absence of others.
    276 		 */
    277 		if (flags&SPSTART) {
    278 			longest = NULL;
    279 			len = 0;
    280 			for (; scan != NULL; scan = regnext(scan))
    281 				if (OP(scan) == EXACTLY && (int) strlen(OPERAND(scan)) >= len) {
    282 					longest = OPERAND(scan);
    283 					len = strlen(OPERAND(scan));
    284 				}
    285 			r->regmust = longest;
    286 			r->regmlen = len;
    287 		}
    288 	}
    289 
    290 	return(r);
    291 }
    292 
    293 /*
    294  - reg - regular expression, i.e. main body or parenthesized thing
    295  *
    296  * Caller must absorb opening parenthesis.
    297  *
    298  * Combining parenthesis handling with the base level of regular expression
    299  * is a trifle forced, but the need to tie the tails of the branches to what
    300  * follows makes it hard to avoid.
    301  */
    302 static char *
    303 reg(paren, flagp)
    304 int paren;			/* Parenthesized? */
    305 int *flagp;
    306 {
    307 	char *ret;
    308 	char *br;
    309 	char *ender;
    310 	int parno = 0;
    311 	int flags;
    312 
    313 	*flagp = HASWIDTH;	/* Tentatively. */
    314 
    315 	/* Make an OPEN node, if parenthesized. */
    316 	if (paren) {
    317 		if (regnpar >= NSUBEXP)
    318 			FAIL("too many ()");
    319 		parno = regnpar;
    320 		regnpar++;
    321 		ret = regnode(OPEN+parno);
    322 	} else
    323 		ret = NULL;
    324 
    325 	/* Pick up the branches, linking them together. */
    326 	br = regbranch(&flags);
    327 	if (br == NULL)
    328 		return(NULL);
    329 	if (ret != NULL)
    330 		regtail(ret, br);	/* OPEN -> first. */
    331 	else
    332 		ret = br;
    333 	if (!(flags&HASWIDTH))
    334 		*flagp &= ~HASWIDTH;
    335 	*flagp |= flags&SPSTART;
    336 	while (*regparse == '|' || *regparse == '\n') {
    337 		regparse++;
    338 		br = regbranch(&flags);
    339 		if (br == NULL)
    340 			return(NULL);
    341 		regtail(ret, br);	/* BRANCH -> BRANCH. */
    342 		if (!(flags&HASWIDTH))
    343 			*flagp &= ~HASWIDTH;
    344 		*flagp |= flags&SPSTART;
    345 	}
    346 
    347 	/* Make a closing node, and hook it on the end. */
    348 	ender = regnode((paren) ? CLOSE+parno : END);
    349 	regtail(ret, ender);
    350 
    351 	/* Hook the tails of the branches to the closing node. */
    352 	for (br = ret; br != NULL; br = regnext(br))
    353 		regoptail(br, ender);
    354 
    355 	/* Check for proper termination. */
    356 	if (paren && *regparse++ != ')') {
    357 		FAIL("unmatched ()");
    358 	} else if (!paren && *regparse != '\0') {
    359 		if (*regparse == ')') {
    360 			FAIL("unmatched ()");
    361 		} else
    362 			FAIL("junk on end");	/* "Can't happen". */
    363 		/* NOTREACHED */
    364 	}
    365 
    366 	return(ret);
    367 }
    368 
    369 /*
    370  - regbranch - one alternative of an | operator
    371  *
    372  * Implements the concatenation operator.
    373  */
    374 static char *
    375 regbranch(flagp)
    376 int *flagp;
    377 {
    378 	char *ret;
    379 	char *chain;
    380 	char *latest;
    381 	int flags;
    382 
    383 	*flagp = WORST;		/* Tentatively. */
    384 
    385 	ret = regnode(BRANCH);
    386 	chain = NULL;
    387 	while (*regparse != '\0' && *regparse != ')' &&
    388 	       *regparse != '\n' && *regparse != '|') {
    389 		latest = regpiece(&flags);
    390 		if (latest == NULL)
    391 			return(NULL);
    392 		*flagp |= flags&HASWIDTH;
    393 		if (chain == NULL)	/* First piece. */
    394 			*flagp |= flags&SPSTART;
    395 		else
    396 			regtail(chain, latest);
    397 		chain = latest;
    398 	}
    399 	if (chain == NULL)	/* Loop ran zero times. */
    400 		(void) regnode(NOTHING);
    401 
    402 	return(ret);
    403 }
    404 
    405 /*
    406  - regpiece - something followed by possible [*+?]
    407  *
    408  * Note that the branching code sequences used for ? and the general cases
    409  * of * and + are somewhat optimized:  they use the same NOTHING node as
    410  * both the endmarker for their branch list and the body of the last branch.
    411  * It might seem that this node could be dispensed with entirely, but the
    412  * endmarker role is not redundant.
    413  */
    414 static char *
    415 regpiece(flagp)
    416 int *flagp;
    417 {
    418 	char *ret;
    419 	char op;
    420 	char *next;
    421 	int flags;
    422 
    423 	ret = regatom(&flags);
    424 	if (ret == NULL)
    425 		return(NULL);
    426 
    427 	op = *regparse;
    428 	if (!ISMULT(op)) {
    429 		*flagp = flags;
    430 		return(ret);
    431 	}
    432 
    433 	if (!(flags&HASWIDTH) && op != '?')
    434 		FAIL("*+ operand could be empty");
    435 	*flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH);
    436 
    437 	if (op == '*' && (flags&SIMPLE))
    438 		reginsert(STAR, ret);
    439 	else if (op == '*') {
    440 		/* Emit x* as (x&|), where & means "self". */
    441 		reginsert(BRANCH, ret);			/* Either x */
    442 		regoptail(ret, regnode(BACK));		/* and loop */
    443 		regoptail(ret, ret);			/* back */
    444 		regtail(ret, regnode(BRANCH));		/* or */
    445 		regtail(ret, regnode(NOTHING));		/* null. */
    446 	} else if (op == '+' && (flags&SIMPLE))
    447 		reginsert(PLUS, ret);
    448 	else if (op == '+') {
    449 		/* Emit x+ as x(&|), where & means "self". */
    450 		next = regnode(BRANCH);			/* Either */
    451 		regtail(ret, next);
    452 		regtail(regnode(BACK), ret);		/* loop back */
    453 		regtail(next, regnode(BRANCH));		/* or */
    454 		regtail(ret, regnode(NOTHING));		/* null. */
    455 	} else if (op == '?') {
    456 		/* Emit x? as (x|) */
    457 		reginsert(BRANCH, ret);			/* Either x */
    458 		regtail(ret, regnode(BRANCH));		/* or */
    459 		next = regnode(NOTHING);		/* null. */
    460 		regtail(ret, next);
    461 		regoptail(ret, next);
    462 	}
    463 	regparse++;
    464 	if (ISMULT(*regparse))
    465 		FAIL("nested *?+");
    466 
    467 	return(ret);
    468 }
    469 
    470 /*
    471  - regatom - the lowest level
    472  *
    473  * Optimization:  gobbles an entire sequence of ordinary characters so that
    474  * it can turn them into a single node, which is smaller to store and
    475  * faster to run.  Backslashed characters are exceptions, each becoming a
    476  * separate node; the code is simpler that way and it's not worth fixing.
    477  */
    478 static char *
    479 regatom(flagp)
    480 int *flagp;
    481 {
    482 	char *ret;
    483 	int flags;
    484 
    485 	*flagp = WORST;		/* Tentatively. */
    486 
    487 	switch (*regparse++) {
    488 	/* FIXME: these chars only have meaning at beg/end of pat? */
    489 	case '^':
    490 		ret = regnode(BOL);
    491 		break;
    492 	case '$':
    493 		ret = regnode(EOL);
    494 		break;
    495 	case '.':
    496 		ret = regnode(ANY);
    497 		*flagp |= HASWIDTH|SIMPLE;
    498 		break;
    499 	case '[': {
    500 			int class;
    501 			int classend;
    502 
    503 			if (*regparse == '^') {	/* Complement of range. */
    504 				ret = regnode(ANYBUT);
    505 				regparse++;
    506 			} else
    507 				ret = regnode(ANYOF);
    508 			if (*regparse == ']' || *regparse == '-')
    509 				regc(*regparse++);
    510 			while (*regparse != '\0' && *regparse != ']') {
    511 				if (*regparse == '-') {
    512 					regparse++;
    513 					if (*regparse == ']' || *regparse == '\0')
    514 						regc('-');
    515 					else {
    516 						class = UCHARAT(regparse-2)+1;
    517 						classend = UCHARAT(regparse);
    518 						if (class > classend+1)
    519 							FAIL("invalid [] range");
    520 						for (; class <= classend; class++)
    521 							regc(class);
    522 						regparse++;
    523 					}
    524 				} else
    525 					regc(*regparse++);
    526 			}
    527 			regc('\0');
    528 			if (*regparse != ']')
    529 				FAIL("unmatched []");
    530 			regparse++;
    531 			*flagp |= HASWIDTH|SIMPLE;
    532 		}
    533 		break;
    534 	case '(':
    535 		ret = reg(1, &flags);
    536 		if (ret == NULL)
    537 			return(NULL);
    538 		*flagp |= flags&(HASWIDTH|SPSTART);
    539 		break;
    540 	case '\0':
    541 	case '|':
    542 	case '\n':
    543 	case ')':
    544 		FAIL("internal urp");	/* Supposed to be caught earlier. */
    545 	case '?':
    546 	case '+':
    547 	case '*':
    548 		FAIL("?+* follows nothing");
    549 	case '\\':
    550 		switch (*regparse++) {
    551 		case '\0':
    552 			FAIL("trailing \\");
    553 		case '<':
    554 			ret = regnode(WORDA);
    555 			break;
    556 		case '>':
    557 			ret = regnode(WORDZ);
    558 			break;
    559 		/* FIXME: Someday handle \1, \2, ... */
    560 		default:
    561 			/* Handle general quoted chars in exact-match routine */
    562 			goto de_fault;
    563 		}
    564 		break;
    565 	de_fault:
    566 		/*FALLTHROUGH*/
    567 	default:
    568 		/*
    569 		 * Encode a string of characters to be matched exactly.
    570 		 *
    571 		 * This is a bit tricky due to quoted chars and due to
    572 		 * '*', '+', and '?' taking the SINGLE char previous
    573 		 * as their operand.
    574 		 *
    575 		 * On entry, the char at regparse[-1] is going to go
    576 		 * into the string, no matter what it is.  (It could be
    577 		 * following a \ if we are entered from the '\' case.)
    578 		 *
    579 		 * Basic idea is to pick up a good char in  ch  and
    580 		 * examine the next char.  If it's *+? then we twiddle.
    581 		 * If it's \ then we frozzle.  If it's other magic char
    582 		 * we push  ch  and terminate the string.  If none of the
    583 		 * above, we push  ch  on the string and go around again.
    584 		 *
    585 		 *  regprev  is used to remember where "the current char"
    586 		 * starts in the string, if due to a *+? we need to back
    587 		 * up and put the current char in a separate, 1-char, string.
    588 		 * When  regprev  is NULL,  ch  is the only char in the
    589 		 * string; this is used in *+? handling, and in setting
    590 		 * flags |= SIMPLE at the end.
    591 		 */
    592 		{
    593 			char *regprev;
    594 			char ch;
    595 
    596 			regparse--;			/* Look at cur char */
    597 			ret = regnode(EXACTLY);
    598 			for ( regprev = 0 ; ; ) {
    599 				ch = *regparse++;	/* Get current char */
    600 				switch (*regparse) {	/* look at next one */
    601 
    602 				default:
    603 					regc(ch);	/* Add cur to string */
    604 					break;
    605 
    606 				case '.': case '[': case '(':
    607 				case ')': case '|': case '\n':
    608 				case '$': case '^':
    609 				case '\0':
    610 				/* FIXME, $ and ^ should not always be magic */
    611 				magic:
    612 					regc(ch);	/* dump cur char */
    613 					goto done;	/* and we are done */
    614 
    615 				case '?': case '+': case '*':
    616 					if (!regprev) 	/* If just ch in str, */
    617 						goto magic;	/* use it */
    618 					/* End mult-char string one early */
    619 					regparse = regprev; /* Back up parse */
    620 					goto done;
    621 
    622 				case '\\':
    623 					regc(ch);	/* Cur char OK */
    624 					switch (regparse[1]){ /* Look after \ */
    625 					case '\0':
    626 					case '<':
    627 					case '>':
    628 					/* FIXME: Someday handle \1, \2, ... */
    629 						goto done; /* Not quoted */
    630 					default:
    631 						/* Backup point is \, scan							 * point is after it. */
    632 						regprev = regparse;
    633 						regparse++;
    634 						continue;	/* NOT break; */
    635 					}
    636 				}
    637 				regprev = regparse;	/* Set backup point */
    638 			}
    639 		done:
    640 			regc('\0');
    641 			*flagp |= HASWIDTH;
    642 			if (!regprev)		/* One char? */
    643 				*flagp |= SIMPLE;
    644 		}
    645 		break;
    646 	}
    647 
    648 	return(ret);
    649 }
    650 
    651 /*
    652  - regnode - emit a node
    653  */
    654 static char *			/* Location. */
    655 regnode(op)
    656 int op;
    657 {
    658 	char *ret;
    659 	char *ptr;
    660 
    661 	ret = regcode;
    662 	if (ret == &regdummy) {
    663 		regsize += 3;
    664 		return(ret);
    665 	}
    666 
    667 	ptr = ret;
    668 	*ptr++ = op;
    669 	*ptr++ = '\0';		/* Null "next" pointer. */
    670 	*ptr++ = '\0';
    671 	regcode = ptr;
    672 
    673 	return(ret);
    674 }
    675 
    676 /*
    677  - regc - emit (if appropriate) a byte of code
    678  */
    679 static void
    680 regc(b)
    681 int b;
    682 {
    683 	if (regcode != &regdummy)
    684 		*regcode++ = b;
    685 	else
    686 		regsize++;
    687 }
    688 
    689 /*
    690  - reginsert - insert an operator in front of already-emitted operand
    691  *
    692  * Means relocating the operand.
    693  */
    694 static void
    695 reginsert(op, opnd)
    696 int op;
    697 char *opnd;
    698 {
    699 	char *src;
    700 	char *dst;
    701 	char *place;
    702 
    703 	if (regcode == &regdummy) {
    704 		regsize += 3;
    705 		return;
    706 	}
    707 
    708 	src = regcode;
    709 	regcode += 3;
    710 	dst = regcode;
    711 	while (src > opnd)
    712 		*--dst = *--src;
    713 
    714 	place = opnd;		/* Op node, where operand used to be. */
    715 	*place++ = op;
    716 	*place++ = '\0';
    717 	*place++ = '\0';
    718 }
    719 
    720 /*
    721  - regtail - set the next-pointer at the end of a node chain
    722  */
    723 static void
    724 regtail(p, val)
    725 char *p;
    726 char *val;
    727 {
    728 	char *scan;
    729 	char *temp;
    730 	int offset;
    731 
    732 	if (p == &regdummy)
    733 		return;
    734 
    735 	/* Find last node. */
    736 	scan = p;
    737 	for (;;) {
    738 		temp = regnext(scan);
    739 		if (temp == NULL)
    740 			break;
    741 		scan = temp;
    742 	}
    743 
    744 	if (OP(scan) == BACK)
    745 		offset = scan - val;
    746 	else
    747 		offset = val - scan;
    748 	*(scan+1) = ((unsigned int)offset>>8)&0377;
    749 	*(scan+2) = offset&0377;
    750 }
    751 
    752 /*
    753  - regoptail - regtail on operand of first argument; nop if operandless
    754  */
    755 static void
    756 regoptail(p, val)
    757 char *p;
    758 char *val;
    759 {
    760 	/* "Operandless" and "op != BRANCH" are synonymous in practice. */
    761 	if (p == NULL || p == &regdummy || OP(p) != BRANCH)
    762 		return;
    763 	regtail(OPERAND(p), val);
    764 }
    765 
    766 /*
    767  * regexec and friends
    768  */
    769 
    770 /*
    771  * Global work variables for regexec().
    772  */
    773 static char *reginput;		/* String-input pointer. */
    774 static char *regbol;		/* Beginning of input, for ^ check. */
    775 static char **regstartp;	/* Pointer to startp array. */
    776 static char **regendp;		/* Ditto for endp. */
    777 
    778 /*
    779  * Forwards.
    780  */
    781 STATIC int regtry __P((const regexp *, const char *));
    782 STATIC int regmatch __P((char *));
    783 STATIC int regrepeat __P((char *));
    784 
    785 #ifdef DEBUG
    786 int regnarrate = 0;
    787 void regdump __P((regexp *));
    788 STATIC char *regprop __P((char *));
    789 #endif
    790 
    791 /*
    792  - regexec - match a regexp against a string
    793  */
    794 int
    795 __compat_regexec(prog, string)
    796 const regexp *prog;
    797 const char *string;
    798 {
    799 	char *s;
    800 
    801 	/* Be paranoid... */
    802 	if (prog == NULL || string == NULL) {
    803 		regerror("NULL parameter");
    804 		return(0);
    805 	}
    806 
    807 	/* Check validity of program. */
    808 	if (UCHARAT(prog->program) != MAGIC) {
    809 		regerror("corrupted program");
    810 		return(0);
    811 	}
    812 
    813 	/* If there is a "must appear" string, look for it. */
    814 	if (prog->regmust != NULL) {
    815 		/* LINTED const castaway */
    816 		s = (char *)string;
    817 		while ((s = strchr(s, prog->regmust[0])) != NULL) {
    818 			if (strncmp(s, prog->regmust,
    819 			    (size_t)prog->regmlen) == 0)
    820 				break;	/* Found it. */
    821 			s++;
    822 		}
    823 		if (s == NULL)	/* Not present. */
    824 			return(0);
    825 	}
    826 
    827 	/* Mark beginning of line for ^ . */
    828 	/* LINTED const castaway */
    829 	regbol = (char *)string;
    830 
    831 	/* Simplest case:  anchored match need be tried only once. */
    832 	if (prog->reganch)
    833 		return(regtry(prog, string));
    834 
    835 	/* Messy cases:  unanchored match. */
    836 	/* LINTED const castaway */
    837 	s = (char *)string;
    838 	if (prog->regstart != '\0')
    839 		/* We know what char it must start with. */
    840 		while ((s = strchr(s, prog->regstart)) != NULL) {
    841 			if (regtry(prog, s))
    842 				return(1);
    843 			s++;
    844 		}
    845 	else
    846 		/* We don't -- general case. */
    847 		do {
    848 			if (regtry(prog, s))
    849 				return(1);
    850 		} while (*s++ != '\0');
    851 
    852 	/* Failure. */
    853 	return(0);
    854 }
    855 
    856 /*
    857  - regtry - try match at specific point
    858  */
    859 static int			/* 0 failure, 1 success */
    860 regtry(prog, string)
    861 const regexp *prog;
    862 const char *string;
    863 {
    864 	int i;
    865 	char **sp;
    866 	char **ep;
    867 
    868 	/* LINTED const castaway */
    869 	reginput = (char *)string;				/* XXX */
    870 	regstartp = (char **)prog->startp;			/* XXX */
    871 	regendp = (char **)prog->endp;				/* XXX */
    872 
    873 	sp = (char **)prog->startp;				/* XXX */
    874 	ep = (char **)prog->endp;				/* XXX */
    875 	for (i = NSUBEXP; i > 0; i--) {
    876 		*sp++ = NULL;
    877 		*ep++ = NULL;
    878 	}
    879 	if (regmatch((char *)prog->program + 1)) {		/* XXX */
    880 		/* LINTED const castaway */
    881 		((regexp *)prog)->startp[0] = (char *)string;	/* XXX */
    882 		/* LINTED const castaway */
    883 		((regexp *)prog)->endp[0] = reginput;		/* XXX */
    884 		return(1);
    885 	} else
    886 		return(0);
    887 }
    888 
    889 /*
    890  - regmatch - main matching routine
    891  *
    892  * Conceptually the strategy is simple:  check to see whether the current
    893  * node matches, call self recursively to see whether the rest matches,
    894  * and then act accordingly.  In practice we make some effort to avoid
    895  * recursion, in particular by going through "ordinary" nodes (that don't
    896  * need to know whether the rest of the match failed) by a loop instead of
    897  * by recursion.
    898  */
    899 static int			/* 0 failure, 1 success */
    900 regmatch(prog)
    901 char *prog;
    902 {
    903 	char *scan;	/* Current node. */
    904 	char *next;		/* Next node. */
    905 
    906 	scan = prog;
    907 #ifdef DEBUG
    908 	if (scan != NULL && regnarrate)
    909 		fprintf(stderr, "%s(\n", regprop(scan));
    910 #endif
    911 	while (scan != NULL) {
    912 #ifdef DEBUG
    913 		if (regnarrate)
    914 			fprintf(stderr, "%s...\n", regprop(scan));
    915 #endif
    916 		next = regnext(scan);
    917 
    918 		switch (OP(scan)) {
    919 		case BOL:
    920 			if (reginput != regbol)
    921 				return(0);
    922 			break;
    923 		case EOL:
    924 			if (*reginput != '\0')
    925 				return(0);
    926 			break;
    927 		case WORDA:
    928 			/* Must be looking at a letter, digit, or _ */
    929 			if ((!isalnum(UCHARAT(reginput))) && *reginput != '_')
    930 				return(0);
    931 			/* Prev must be BOL or nonword */
    932 			if (reginput > regbol &&
    933 			    (isalnum(UCHARAT(reginput - 1))
    934 			     || reginput[-1] == '_'))
    935 				return(0);
    936 			break;
    937 		case WORDZ:
    938 			/* Must be looking at non letter, digit, or _ */
    939 			if (isalnum(UCHARAT(reginput)) || *reginput == '_')
    940 				return(0);
    941 			/* We don't care what the previous char was */
    942 			break;
    943 		case ANY:
    944 			if (*reginput == '\0')
    945 				return(0);
    946 			reginput++;
    947 			break;
    948 		case EXACTLY: {
    949 				int len;
    950 				char *opnd;
    951 
    952 				opnd = OPERAND(scan);
    953 				/* Inline the first character, for speed. */
    954 				if (*opnd != *reginput)
    955 					return(0);
    956 				len = strlen(opnd);
    957 				if (len > 1 && strncmp(opnd, reginput,
    958 				    (size_t)len) != 0)
    959 					return(0);
    960 				reginput += len;
    961 			}
    962 			break;
    963 		case ANYOF:
    964  			if (*reginput == '\0' || strchr(OPERAND(scan), *reginput) == NULL)
    965 				return(0);
    966 			reginput++;
    967 			break;
    968 		case ANYBUT:
    969  			if (*reginput == '\0' || strchr(OPERAND(scan), *reginput) != NULL)
    970 				return(0);
    971 			reginput++;
    972 			break;
    973 		case NOTHING:
    974 			break;
    975 		case BACK:
    976 			break;
    977 		case OPEN+1:
    978 		case OPEN+2:
    979 		case OPEN+3:
    980 		case OPEN+4:
    981 		case OPEN+5:
    982 		case OPEN+6:
    983 		case OPEN+7:
    984 		case OPEN+8:
    985 		case OPEN+9: {
    986 				int no;
    987 				char *save;
    988 
    989 				no = OP(scan) - OPEN;
    990 				save = reginput;
    991 
    992 				if (regmatch(next)) {
    993 					/*
    994 					 * Don't set startp if some later
    995 					 * invocation of the same parentheses
    996 					 * already has.
    997 					 */
    998 					if (regstartp[no] == NULL)
    999 						regstartp[no] = save;
   1000 					return(1);
   1001 				} else
   1002 					return(0);
   1003 			}
   1004 		case CLOSE+1:
   1005 		case CLOSE+2:
   1006 		case CLOSE+3:
   1007 		case CLOSE+4:
   1008 		case CLOSE+5:
   1009 		case CLOSE+6:
   1010 		case CLOSE+7:
   1011 		case CLOSE+8:
   1012 		case CLOSE+9: {
   1013 				int no;
   1014 				char *save;
   1015 
   1016 				no = OP(scan) - CLOSE;
   1017 				save = reginput;
   1018 
   1019 				if (regmatch(next)) {
   1020 					/*
   1021 					 * Don't set endp if some later
   1022 					 * invocation of the same parentheses
   1023 					 * already has.
   1024 					 */
   1025 					if (regendp[no] == NULL)
   1026 						regendp[no] = save;
   1027 					return(1);
   1028 				} else
   1029 					return(0);
   1030 			}
   1031 		case BRANCH: {
   1032 				char *save;
   1033 
   1034 				if (OP(next) != BRANCH)		/* No choice. */
   1035 					next = OPERAND(scan);	/* Avoid recursion. */
   1036 				else {
   1037 					do {
   1038 						save = reginput;
   1039 						if (regmatch(OPERAND(scan)))
   1040 							return(1);
   1041 						reginput = save;
   1042 						scan = regnext(scan);
   1043 					} while (scan != NULL && OP(scan) == BRANCH);
   1044 					return(0);
   1045 					/* NOTREACHED */
   1046 				}
   1047 			}
   1048 			break;
   1049 		case STAR:
   1050 		case PLUS: {
   1051 				char nextch;
   1052 				int no;
   1053 				char *save;
   1054 				int min;
   1055 
   1056 				/*
   1057 				 * Lookahead to avoid useless match attempts
   1058 				 * when we know what character comes next.
   1059 				 */
   1060 				nextch = '\0';
   1061 				if (OP(next) == EXACTLY)
   1062 					nextch = *OPERAND(next);
   1063 				min = (OP(scan) == STAR) ? 0 : 1;
   1064 				save = reginput;
   1065 				no = regrepeat(OPERAND(scan));
   1066 				while (no >= min) {
   1067 					/* If it could work, try it. */
   1068 					if (nextch == '\0' || *reginput == nextch)
   1069 						if (regmatch(next))
   1070 							return(1);
   1071 					/* Couldn't or didn't -- back up. */
   1072 					no--;
   1073 					reginput = save + no;
   1074 				}
   1075 				return(0);
   1076 			}
   1077 		case END:
   1078 			return(1);	/* Success! */
   1079 		default:
   1080 			regerror("memory corruption");
   1081 			return(0);
   1082 		}
   1083 
   1084 		scan = next;
   1085 	}
   1086 
   1087 	/*
   1088 	 * We get here only if there's trouble -- normally "case END" is
   1089 	 * the terminating point.
   1090 	 */
   1091 	regerror("corrupted pointers");
   1092 	return(0);
   1093 }
   1094 
   1095 /*
   1096  - regrepeat - repeatedly match something simple, report how many
   1097  */
   1098 static int
   1099 regrepeat(p)
   1100 char *p;
   1101 {
   1102 	int count = 0;
   1103 	char *scan;
   1104 	char *opnd;
   1105 
   1106 	scan = reginput;
   1107 	opnd = OPERAND(p);
   1108 	switch (OP(p)) {
   1109 	case ANY:
   1110 		count = strlen(scan);
   1111 		scan += count;
   1112 		break;
   1113 	case EXACTLY:
   1114 		while (*opnd == *scan) {
   1115 			count++;
   1116 			scan++;
   1117 		}
   1118 		break;
   1119 	case ANYOF:
   1120 		while (*scan != '\0' && strchr(opnd, *scan) != NULL) {
   1121 			count++;
   1122 			scan++;
   1123 		}
   1124 		break;
   1125 	case ANYBUT:
   1126 		while (*scan != '\0' && strchr(opnd, *scan) == NULL) {
   1127 			count++;
   1128 			scan++;
   1129 		}
   1130 		break;
   1131 	default:		/* Oh dear.  Called inappropriately. */
   1132 		regerror("internal foulup");
   1133 		count = 0;	/* Best compromise. */
   1134 		break;
   1135 	}
   1136 	reginput = scan;
   1137 
   1138 	return(count);
   1139 }
   1140 
   1141 /*
   1142  - regnext - dig the "next" pointer out of a node
   1143  */
   1144 static char *
   1145 regnext(p)
   1146 char *p;
   1147 {
   1148 	int offset;
   1149 
   1150 	if (p == &regdummy)
   1151 		return(NULL);
   1152 
   1153 	offset = NEXT(p);
   1154 	if (offset == 0)
   1155 		return(NULL);
   1156 
   1157 	if (OP(p) == BACK)
   1158 		return(p-offset);
   1159 	else
   1160 		return(p+offset);
   1161 }
   1162 
   1163 #ifdef DEBUG
   1164 
   1165 /*
   1166  - regdump - dump a regexp onto stdout in vaguely comprehensible form
   1167  */
   1168 void
   1169 regdump(r)
   1170 regexp *r;
   1171 {
   1172 	char *s;
   1173 	char op = EXACTLY;	/* Arbitrary non-END op. */
   1174 	char *next;
   1175 
   1176 
   1177 	s = r->program + 1;
   1178 	while (op != END) {	/* While that wasn't END last time... */
   1179 		op = OP(s);
   1180 		printf("%2td%s", s-r->program, regprop(s));	/* Where, what. */
   1181 		next = regnext(s);
   1182 		if (next == NULL)		/* Next ptr. */
   1183 			printf("(0)");
   1184 		else
   1185 			printf("(%td)", (s-r->program)+(next-s));
   1186 		s += 3;
   1187 		if (op == ANYOF || op == ANYBUT || op == EXACTLY) {
   1188 			/* Literal string, where present. */
   1189 			while (*s != '\0') {
   1190 				putchar(*s);
   1191 				s++;
   1192 			}
   1193 			s++;
   1194 		}
   1195 		putchar('\n');
   1196 	}
   1197 
   1198 	/* Header fields of interest. */
   1199 	if (r->regstart != '\0')
   1200 		printf("start `%c' ", r->regstart);
   1201 	if (r->reganch)
   1202 		printf("anchored ");
   1203 	if (r->regmust != NULL)
   1204 		printf("must have \"%s\"", r->regmust);
   1205 	printf("\n");
   1206 }
   1207 
   1208 /*
   1209  - regprop - printable representation of opcode
   1210  */
   1211 static char *
   1212 regprop(op)
   1213 char *op;
   1214 {
   1215 	char *p;
   1216 	static char buf[50];
   1217 
   1218 	(void)strncpy(buf, ":", sizeof(buf) - 1);
   1219 
   1220 	switch (OP(op)) {
   1221 	case BOL:
   1222 		p = "BOL";
   1223 		break;
   1224 	case EOL:
   1225 		p = "EOL";
   1226 		break;
   1227 	case ANY:
   1228 		p = "ANY";
   1229 		break;
   1230 	case ANYOF:
   1231 		p = "ANYOF";
   1232 		break;
   1233 	case ANYBUT:
   1234 		p = "ANYBUT";
   1235 		break;
   1236 	case BRANCH:
   1237 		p = "BRANCH";
   1238 		break;
   1239 	case EXACTLY:
   1240 		p = "EXACTLY";
   1241 		break;
   1242 	case NOTHING:
   1243 		p = "NOTHING";
   1244 		break;
   1245 	case BACK:
   1246 		p = "BACK";
   1247 		break;
   1248 	case END:
   1249 		p = "END";
   1250 		break;
   1251 	case OPEN+1:
   1252 	case OPEN+2:
   1253 	case OPEN+3:
   1254 	case OPEN+4:
   1255 	case OPEN+5:
   1256 	case OPEN+6:
   1257 	case OPEN+7:
   1258 	case OPEN+8:
   1259 	case OPEN+9:
   1260 		(void)snprintf(buf+strlen(buf), sizeof(buf) - strlen(buf),
   1261 		    "OPEN%d", OP(op)-OPEN);
   1262 		p = NULL;
   1263 		break;
   1264 	case CLOSE+1:
   1265 	case CLOSE+2:
   1266 	case CLOSE+3:
   1267 	case CLOSE+4:
   1268 	case CLOSE+5:
   1269 	case CLOSE+6:
   1270 	case CLOSE+7:
   1271 	case CLOSE+8:
   1272 	case CLOSE+9:
   1273 		(void)snprintf(buf+strlen(buf), sizeof(buf) - strlen(buf),
   1274 		    "CLOSE%d", OP(op)-CLOSE);
   1275 		p = NULL;
   1276 		break;
   1277 	case STAR:
   1278 		p = "STAR";
   1279 		break;
   1280 	case PLUS:
   1281 		p = "PLUS";
   1282 		break;
   1283 	case WORDA:
   1284 		p = "WORDA";
   1285 		break;
   1286 	case WORDZ:
   1287 		p = "WORDZ";
   1288 		break;
   1289 	default:
   1290 		p = NULL;
   1291 		regerror("corrupted opcode");
   1292 		break;
   1293 	}
   1294 	if (p != NULL)
   1295 		(void)strncat(buf, p, sizeof(buf) - strlen(buf) - 1);
   1296 	return(buf);
   1297 }
   1298 #endif
   1299 
   1300 /*
   1301  * The following is provided for those people who do not have strcspn() in
   1302  * their C libraries.  They should get off their butts and do something
   1303  * about it; at least one public-domain implementation of those (highly
   1304  * useful) string routines has been published on Usenet.
   1305  */
   1306 #ifdef STRCSPN
   1307 /*
   1308  * strcspn - find length of initial segment of s1 consisting entirely
   1309  * of characters not from s2
   1310  */
   1311 
   1312 static int
   1313 strcspn(s1, s2)
   1314 char *s1;
   1315 char *s2;
   1316 {
   1317 	char *scan1;
   1318 	char *scan2;
   1319 	int count;
   1320 
   1321 	count = 0;
   1322 	for (scan1 = s1; *scan1 != '\0'; scan1++) {
   1323 		for (scan2 = s2; *scan2 != '\0';)	/* ++ moved down. */
   1324 			if (*scan1 == *scan2++)
   1325 				return(count);
   1326 		count++;
   1327 	}
   1328 	return(count);
   1329 }
   1330 #endif
   1331