xref: /netbsd-src/lib/libcompat/regexp/regexp.c (revision 3d64af45d4fba25af2578e868d29d5f2e4931611)
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 *
__compat_regcomp(expn)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 *
reg(paren,flagp)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 *
regbranch(flagp)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 *
regpiece(flagp)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 *
regatom(flagp)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. */
regnode(op)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
regc(b)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
reginsert(op,opnd)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
regtail(p,val)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
regoptail(p,val)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
__compat_regexec(prog,string)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 */
regtry(prog,string)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 */
regmatch(prog)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
regrepeat(p)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 *
regnext(p)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
regdump(r)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 *
regprop(op)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
strcspn(s1,s2)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