regexp.c

Revision 247, 33.0 kB (checked in by athomas, 4 years ago)
Merged autoconf branch into trunk.

Line
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	*** THIS IS AN ALTERED VERSION. It was altered by Christopher Seiwald
31	*** seiwald@vix.com, on 28 August 1993, for use in jam. Regmagic.h
32	*** was moved into regexp.h, and the include of regexp.h now uses "'s
33	*** to avoid conflicting with the system regexp.h. Const, bless its
34	*** soul, was removed so it can compile everywhere. The declaration
35	*** of strchr() was in conflict on AIX, so it was removed (as it is
36	*** happily defined in string.h).
37	*** THIS IS AN ALTERED VERSION. It was altered by Christopher Seiwald
38	*** seiwald@perforce.com, on 20 January 2000, to use function prototypes.
39	*
40	* Beware that some of this code is subtly aware of the way operator
41	* precedence is structured in regular expressions. Serious changes in
42	* regular-expression syntax might require a total rethink.
43	*/
44	#include "regexp.h"
45	#include <stdio.h>
46	#include <ctype.h>
47	#ifndef ultrix
48	#include <stdlib.h>
49	#endif
50	#include <string.h>
51
52	/*
53	* The "internal use only" fields in regexp.h are present to pass info from
54	* compile to execute that permits the execute phase to run lots faster on
55	* simple cases. They are:
56	*
57	* regstart char that must begin a match; '\0' if none obvious
58	* reganch is the match anchored (at beginning-of-line only)?
59	* regmust string (pointer into program) that match must include, or NULL
60	* regmlen length of regmust string
61	*
62	* Regstart and reganch permit very fast decisions on suitable starting points
63	* for a match, cutting down the work a lot. Regmust permits fast rejection
64	* of lines that cannot possibly match. The regmust tests are costly enough
65	* that regcomp() supplies a regmust only if the r.e. contains something
66	* potentially expensive (at present, the only such thing detected is * or +
67	* at the start of the r.e., which can involve a lot of backup). Regmlen is
68	* supplied because the test in regexec() needs it and regcomp() is computing
69	* it anyway.
70	*/
71
72	/*
73	* Structure for regexp "program". This is essentially a linear encoding
74	* of a nondeterministic finite-state machine (aka syntax charts or
75	* "railroad normal form" in parsing technology). Each node is an opcode
76	* plus a "next" pointer, possibly plus an operand. "Next" pointers of
77	* all nodes except BRANCH implement concatenation; a "next" pointer with
78	* a BRANCH on both ends of it is connecting two alternatives. (Here we
79	* have one of the subtle syntax dependencies: an individual BRANCH (as
80	* opposed to a collection of them) is never concatenated with anything
81	* because of operator precedence.) The operand of some types of node is
82	* a literal string; for others, it is a node leading into a sub-FSM. In
83	* particular, the operand of a BRANCH node is the first node of the branch.
84	* (NB this is not a tree structure: the tail of the branch connects
85	* to the thing following the set of BRANCHes.) The opcodes are:
86	*/
87
88	/* definition number opnd? meaning */
89	#define END 0 /* no End of program. */
90	#define BOL 1 /* no Match "" at beginning of line. */
91	#define EOL 2 /* no Match "" at end of line. */
92	#define ANY 3 /* no Match any one character. */
93	#define ANYOF 4 /* str Match any character in this string. */
94	#define ANYBUT 5 /* str Match any character not in this string. */
95	#define BRANCH 6 /* node Match this alternative, or the next... */
96	#define BACK 7 /* no Match "", "next" ptr points backward. */
97	#define EXACTLY 8 /* str Match this string. */
98	#define NOTHING 9 /* no Match empty string. */
99	#define STAR 10 /* node Match this (simple) thing 0 or more times. */
100	#define PLUS 11 /* node Match this (simple) thing 1 or more times. */
101	#define WORDA 12 /* no Match "" at wordchar, where prev is nonword */
102	#define WORDZ 13 /* no Match "" at nonwordchar, where prev is word */
103	#define OPEN 20 /* no Mark this point in input as start of #n. */
104	/* OPEN+1 is number 1, etc. */
105	#define CLOSE 30 /* no Analogous to OPEN. */
106
107	/*
108	* Opcode notes:
109	*
110	* BRANCH The set of branches constituting a single choice are hooked
111	* together with their "next" pointers, since precedence prevents
112	* anything being concatenated to any individual branch. The
113	* "next" pointer of the last BRANCH in a choice points to the
114	* thing following the whole choice. This is also where the
115	* final "next" pointer of each individual branch points; each
116	* branch starts with the operand node of a BRANCH node.
117	*
118	* BACK Normal "next" pointers all implicitly point forward; BACK
119	* exists to make loop structures possible.
120	*
121	* STAR,PLUS '?', and complex '*' and '+', are implemented as circular
122	* BRANCH structures using BACK. Simple cases (one character
123	* per match) are implemented with STAR and PLUS for speed
124	* and to minimize recursive plunges.
125	*
126	* OPEN,CLOSE ...are numbered at compile time.
127	*/
128
129	/*
130	* A node is one char of opcode followed by two chars of "next" pointer.
131	* "Next" pointers are stored as two 8-bit pieces, high order first. The
132	* value is a positive offset from the opcode of the node containing it.
133	* An operand, if any, simply follows the node. (Note that much of the
134	* code generation knows about this implicit relationship.)
135	*
136	* Using two bytes for the "next" pointer is vast overkill for most things,
137	* but allows patterns to get big without disasters.
138	*/
139	#define OP(p) (*(p))
140	#define NEXT(p) (((((p)+1)&0377)<<8) + (((p)+2)&0377))
141	#define OPERAND(p) ((p) + 3)
142
143	/*
144	* See regmagic.h for one further detail of program structure.
145	*/
146
147
148	/*
149	* Utility definitions.
150	*/
151	#ifndef CHARBITS
152	#define UCHARAT(p) ((int)(unsigned char )(p))
153	#else
154	#define UCHARAT(p) ((int)*(p)&CHARBITS)
155	#endif
156
157	#define FAIL(m) { regerror(m); return(NULL); }
158	#define ISMULT(c) ((c) == '*' \|\| (c) == '+' \|\| (c) == '?')
159
160	/*
161	* Flags to be passed up and down.
162	*/
163	#define HASWIDTH 01 /* Known never to match null string. */
164	#define SIMPLE 02 /* Simple enough to be STAR/PLUS operand. */
165	#define SPSTART 04 /* Starts with * or +. */
166	#define WORST 0 /* Worst case. */
167
168	/*
169	* Global work variables for regcomp().
170	*/
171	static char regparse; / Input-scan pointer. */
172	static int regnpar; /* () count. */
173	static char regdummy;
174	static char regcode; / Code-emit pointer; &regdummy = don't. */
175	static long regsize; /* Code size. */
176
177	/*
178	* Forward declarations for regcomp()'s friends.
179	*/
180	#ifndef STATIC
181	#define STATIC static
182	#endif
183	STATIC char reg( int paren, int flagp );
184	STATIC char regbranch( int flagp );
185	STATIC char regpiece( int flagp );
186	STATIC char regatom( int flagp );
187	STATIC char *regnode( int op );
188	STATIC char regnext( register char p );
189	STATIC void regc( int b );
190	STATIC void reginsert( char op, char *opnd );
191	STATIC void regtail( char p, char val );
192	STATIC void regoptail( char p, char val );
193	#ifdef STRCSPN
194	STATIC int strcspn();
195	#endif
196
197	/*
198	- regcomp - compile a regular expression into internal code
199	*
200	* We can't allocate space until we know how big the compiled form will be,
201	* but we can't compile it (and thus know how big it is) until we've got a
202	* place to put the code. So we cheat: we compile it twice, once with code
203	* generation turned off and size counting turned on, and once "for real".
204	* This also means that we don't allocate space until we are sure that the
205	* thing really will compile successfully, and we never have to move the
206	* code and thus invalidate pointers into it. (Note that it has to be in
207	* one piece because free() must be able to free it all.)
208	*
209	* Beware that the optimization-preparation code in here knows about some
210	* of the structure of the compiled regexp.
211	*/
212	regexp *
213	regcomp( char *exp )
214	{
215	register regexp *r;
216	register char *scan;
217	register char *longest;
218	register unsigned len;
219	int flags;
220
221	if (exp == NULL)
222	FAIL("NULL argument");
223
224	/* First pass: determine size, legality. */
225	#ifdef notdef
226	if (exp[0] == '.' && exp[1] == '') exp += 2; / aid grep */
227	#endif
228	regparse = (char *)exp;
229	regnpar = 1;
230	regsize = 0L;
231	regcode = &regdummy;
232	regc(MAGIC);
233	if (reg(0, &flags) == NULL)
234	return(NULL);
235
236	/* Small enough for pointer-storage convention? */
237	if (regsize >= 32767L) /* Probably could be 65535L. */
238	FAIL("regexp too big");
239
240	/* Allocate space. */
241	r = (regexp *)malloc(sizeof(regexp) + (unsigned)regsize);
242	if (r == NULL)
243	FAIL("out of space");
244
245	/* Second pass: emit code. */
246	regparse = (char *)exp;
247	regnpar = 1;
248	regcode = r->program;
249	regc(MAGIC);
250	if (reg(0, &flags) == NULL)
251	return(NULL);
252
253	/* Dig out information for optimizations. */
254	r->regstart = '\0'; /* Worst-case defaults. */
255	r->reganch = 0;
256	r->regmust = NULL;
257	r->regmlen = 0;
258	scan = r->program+1; /* First BRANCH. */
259	if (OP(regnext(scan)) == END) { /* Only one top-level choice. */
260	scan = OPERAND(scan);
261
262	/* Starting-point info. */
263	if (OP(scan) == EXACTLY)
264	r->regstart = *OPERAND(scan);
265	else if (OP(scan) == BOL)
266	r->reganch++;
267
268	/*
269	* If there's something expensive in the r.e., find the
270	* longest literal string that must appear and make it the
271	* regmust. Resolve ties in favor of later strings, since
272	* the regstart check works with the beginning of the r.e.
273	* and avoiding duplication strengthens checking. Not a
274	* strong reason, but sufficient in the absence of others.
275	*/
276	if (flags&SPSTART) {
277	longest = NULL;
278	len = 0;
279	for (; scan != NULL; scan = regnext(scan))
280	if (OP(scan) == EXACTLY && strlen(OPERAND(scan)) >= len) {
281	longest = OPERAND(scan);
282	len = strlen(OPERAND(scan));
283	}
284	r->regmust = longest;
285	r->regmlen = len;
286	}
287	}
288
289	return(r);
290	}
291
292	/*
293	- reg - regular expression, i.e. main body or parenthesized thing
294	*
295	* Caller must absorb opening parenthesis.
296	*
297	* Combining parenthesis handling with the base level of regular expression
298	* is a trifle forced, but the need to tie the tails of the branches to what
299	* follows makes it hard to avoid.
300	*/
301	static char *
302	reg(
303	int paren, /* Parenthesized? */
304	int *flagp )
305	{
306	register char *ret;
307	register char *br;
308	register char *ender;
309	register int parno = 0;
310	int flags;
311
312	flagp = HASWIDTH; / Tentatively. */
313
314	/* Make an OPEN node, if parenthesized. */
315	if (paren) {
316	if (regnpar >= NSUBEXP)
317	FAIL("too many ()");
318	parno = regnpar;
319	regnpar++;
320	ret = regnode(OPEN+parno);
321	} else
322	ret = NULL;
323
324	/* Pick up the branches, linking them together. */
325	br = regbranch(&flags);
326	if (br == NULL)
327	return(NULL);
328	if (ret != NULL)
329	regtail(ret, br); /* OPEN -> first. */
330	else
331	ret = br;
332	if (!(flags&HASWIDTH))
333	*flagp &= ~HASWIDTH;
334	*flagp \|= flags&SPSTART;
335	while (regparse == '\|' \|\| regparse == '\n') {
336	regparse++;
337	br = regbranch(&flags);
338	if (br == NULL)
339	return(NULL);
340	regtail(ret, br); /* BRANCH -> BRANCH. */
341	if (!(flags&HASWIDTH))
342	*flagp &= ~HASWIDTH;
343	*flagp \|= flags&SPSTART;
344	}
345
346	/* Make a closing node, and hook it on the end. */
347	ender = regnode((paren) ? CLOSE+parno : END);
348	regtail(ret, ender);
349
350	/* Hook the tails of the branches to the closing node. */
351	for (br = ret; br != NULL; br = regnext(br))
352	regoptail(br, ender);
353
354	/* Check for proper termination. */
355	if (paren && *regparse++ != ')') {
356	FAIL("unmatched ()");
357	} else if (!paren && *regparse != '\0') {
358	if (*regparse == ')') {
359	FAIL("unmatched ()");
360	} else
361	FAIL("junk on end"); /* "Can't happen". */
362	/* NOTREACHED */
363	}
364
365	return(ret);
366	}
367
368	/*
369	- regbranch - one alternative of an \| operator
370	*
371	* Implements the concatenation operator.
372	*/
373	static char *
374	regbranch( int *flagp )
375	{
376	register char *ret;
377	register char *chain;
378	register char *latest;
379	int flags;
380
381	flagp = WORST; / Tentatively. */
382
383	ret = regnode(BRANCH);
384	chain = NULL;
385	while (regparse != '\0' && regparse != ')' &&
386	regparse != '\n' && regparse != '\|') {
387	latest = regpiece(&flags);
388	if (latest == NULL)
389	return(NULL);
390	*flagp \|= flags&HASWIDTH;
391	if (chain == NULL) /* First piece. */
392	*flagp \|= flags&SPSTART;
393	else
394	regtail(chain, latest);
395	chain = latest;
396	}
397	if (chain == NULL) /* Loop ran zero times. */
398	(void) regnode(NOTHING);
399
400	return(ret);
401	}
402
403	/*
404	- regpiece - something followed by possible [*+?]
405	*
406	* Note that the branching code sequences used for ? and the general cases
407	* of * and + are somewhat optimized: they use the same NOTHING node as
408	* both the endmarker for their branch list and the body of the last branch.
409	* It might seem that this node could be dispensed with entirely, but the
410	* endmarker role is not redundant.
411	*/
412	static char *
413	regpiece( int *flagp )
414	{
415	register char *ret;
416	register char op;
417	register char *next;
418	int flags;
419
420	ret = regatom(&flags);
421	if (ret == NULL)
422	return(NULL);
423
424	op = *regparse;
425	if (!ISMULT(op)) {
426	*flagp = flags;
427	return(ret);
428	}
429
430	if (!(flags&HASWIDTH) && op != '?')
431	FAIL("*+ operand could be empty");
432	*flagp = (op != '+') ? (WORST\|SPSTART) : (WORST\|HASWIDTH);
433
434	if (op == '*' && (flags&SIMPLE))
435	reginsert(STAR, ret);
436	else if (op == '*') {
437	/* Emit x* as (x&\|), where & means "self". */
438	reginsert(BRANCH, ret); /* Either x */
439	regoptail(ret, regnode(BACK)); /* and loop */
440	regoptail(ret, ret); /* back */
441	regtail(ret, regnode(BRANCH)); /* or */
442	regtail(ret, regnode(NOTHING)); /* null. */
443	} else if (op == '+' && (flags&SIMPLE))
444	reginsert(PLUS, ret);
445	else if (op == '+') {
446	/* Emit x+ as x(&\|), where & means "self". */
447	next = regnode(BRANCH); /* Either */
448	regtail(ret, next);
449	regtail(regnode(BACK), ret); /* loop back */
450	regtail(next, regnode(BRANCH)); /* or */
451	regtail(ret, regnode(NOTHING)); /* null. */
452	} else if (op == '?') {
453	/* Emit x? as (x\|) */
454	reginsert(BRANCH, ret); /* Either x */
455	regtail(ret, regnode(BRANCH)); /* or */
456	next = regnode(NOTHING); /* null. */
457	regtail(ret, next);
458	regoptail(ret, next);
459	}
460	regparse++;
461	if (ISMULT(*regparse))
462	FAIL("nested *?+");
463
464	return(ret);
465	}
466
467	/*
468	- regatom - the lowest level
469	*
470	* Optimization: gobbles an entire sequence of ordinary characters so that
471	* it can turn them into a single node, which is smaller to store and
472	* faster to run. Backslashed characters are exceptions, each becoming a
473	* separate node; the code is simpler that way and it's not worth fixing.
474	*/
475	static char *
476	regatom( int *flagp )
477	{
478	register char *ret;
479	int flags;
480
481	flagp = WORST; / Tentatively. */
482
483	switch (*regparse++) {
484	/* FIXME: these chars only have meaning at beg/end of pat? */
485	case '^':
486	ret = regnode(BOL);
487	break;
488	case '$':
489	ret = regnode(EOL);
490	break;
491	case '.':
492	ret = regnode(ANY);
493	*flagp \|= HASWIDTH\|SIMPLE;
494	break;
495	case '[': {
496	register int classr;
497	register int classend;
498
499	if (regparse == '^') { / Complement of range. */
500	ret = regnode(ANYBUT);
501	regparse++;
502	} else
503	ret = regnode(ANYOF);
504	if (regparse == ']' \|\| regparse == '-')
505	regc(*regparse++);
506	while (regparse != '\0' && regparse != ']') {
507	if (*regparse == '-') {
508	regparse++;
509	if (regparse == ']' \|\| regparse == '\0')
510	regc('-');
511	else {
512	classr = UCHARAT(regparse-2)+1;
513	classend = UCHARAT(regparse);
514	if (classr > classend+1)
515	FAIL("invalid [] range");
516	for (; classr <= classend; classr++)
517	regc(classr);
518	regparse++;
519	}
520	} else
521	regc(*regparse++);
522	}
523	regc('\0');
524	if (*regparse != ']')
525	FAIL("unmatched []");
526	regparse++;
527	*flagp \|= HASWIDTH\|SIMPLE;
528	}
529	break;
530	case '(':
531	ret = reg(1, &flags);
532	if (ret == NULL)
533	return(NULL);
534	*flagp \|= flags&(HASWIDTH\|SPSTART);
535	break;
536	case '\0':
537	case '\|':
538	case '\n':
539	case ')':
540	FAIL("internal urp"); /* Supposed to be caught earlier. */
541	break;
542	case '?':
543	case '+':
544	case '*':
545	FAIL("?+* follows nothing");
546	break;
547	case '\\':
548	switch (*regparse++) {
549	case '\0':
550	FAIL("trailing \\");
551	break;
552	case '<':
553	ret = regnode(WORDA);
554	break;
555	case '>':
556	ret = regnode(WORDZ);
557	break;
558	/* FIXME: Someday handle \1, \2, ... */
559	default:
560	/* Handle general quoted chars in exact-match routine */
561	goto de_fault;
562	}
563	break;
564	de_fault:
565	default:
566	/*
567	* Encode a string of characters to be matched exactly.
568	*
569	* This is a bit tricky due to quoted chars and due to
570	* '*', '+', and '?' taking the SINGLE char previous
571	* as their operand.
572	*
573	* On entry, the char at regparse[-1] is going to go
574	* into the string, no matter what it is. (It could be
575	* following a \ if we are entered from the '\' case.)
576	*
577	* Basic idea is to pick up a good char in ch and
578	* examine the next char. If it's *+? then we twiddle.
579	* If it's \ then we frozzle. If it's other magic char
580	* we push ch and terminate the string. If none of the
581	* above, we push ch on the string and go around again.
582	*
583	* regprev is used to remember where "the current char"
584	* starts in the string, if due to a *+? we need to back
585	* up and put the current char in a separate, 1-char, string.
586	* When regprev is NULL, ch is the only char in the
587	* string; this is used in *+? handling, and in setting
588	* flags \|= SIMPLE at the end.
589	*/
590	{
591	char *regprev;
592	register char ch;
593
594	regparse--; /* Look at cur char */
595	ret = regnode(EXACTLY);
596	for ( regprev = 0 ; ; ) {
597	ch = regparse++; / Get current char */
598	switch (regparse) { / look at next one */
599
600	default:
601	regc(ch); /* Add cur to string */
602	break;
603
604	case '.': case '[': case '(':
605	case ')': case '\|': case '\n':
606	case '$': case '^':
607	case '\0':
608	/* FIXME, $ and ^ should not always be magic */
609	magic:
610	regc(ch); /* dump cur char */
611	&n