Remove advertising header from man pages.
[dragonfly.git] / lib / libcompat / regexp / regexp.c
blob16b9af83b497a47fc192c31f28b8389e7fb7d13f
1 /*
2 * regcomp and regexec -- regsub and regerror are elsewhere
4 * Copyright (c) 1986 by University of Toronto.
5 * Written by Henry Spencer. Not derived from licensed software.
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:
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.
15 * 2. The origin of this software must not be misrepresented, either
16 * by explicit claim or by omission.
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.
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.
35 #include <limits.h>
36 #include <regexp.h>
37 #include <stdio.h>
38 #include <ctype.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include "collate.h"
42 #include "regmagic.h"
45 * The "internal use only" fields in regexp.h are present to pass info from
46 * compile to execute that permits the execute phase to run lots faster on
47 * simple cases. They are:
49 * regstart char that must begin a match; '\0' if none obvious
50 * reganch is the match anchored (at beginning-of-line only)?
51 * regmust string (pointer into program) that match must include, or NULL
52 * regmlen length of regmust string
54 * Regstart and reganch permit very fast decisions on suitable starting points
55 * for a match, cutting down the work a lot. Regmust permits fast rejection
56 * of lines that cannot possibly match. The regmust tests are costly enough
57 * that regcomp() supplies a regmust only if the r.e. contains something
58 * potentially expensive (at present, the only such thing detected is * or +
59 * at the start of the r.e., which can involve a lot of backup). Regmlen is
60 * supplied because the test in regexec() needs it and regcomp() is computing
61 * it anyway.
65 * Structure for regexp "program". This is essentially a linear encoding
66 * of a nondeterministic finite-state machine (aka syntax charts or
67 * "railroad normal form" in parsing technology). Each node is an opcode
68 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
69 * all nodes except BRANCH implement concatenation; a "next" pointer with
70 * a BRANCH on both ends of it is connecting two alternatives. (Here we
71 * have one of the subtle syntax dependencies: an individual BRANCH (as
72 * opposed to a collection of them) is never concatenated with anything
73 * because of operator precedence.) The operand of some types of node is
74 * a literal string; for others, it is a node leading into a sub-FSM. In
75 * particular, the operand of a BRANCH node is the first node of the branch.
76 * (NB this is *not* a tree structure: the tail of the branch connects
77 * to the thing following the set of BRANCHes.) The opcodes are:
80 /* definition number opnd? meaning */
81 #define END 0 /* no End of program. */
82 #define BOL 1 /* no Match "" at beginning of line. */
83 #define EOL 2 /* no Match "" at end of line. */
84 #define ANY 3 /* no Match any one character. */
85 #define ANYOF 4 /* str Match any character in this string. */
86 #define ANYBUT 5 /* str Match any character not in this string. */
87 #define BRANCH 6 /* node Match this alternative, or the next... */
88 #define BACK 7 /* no Match "", "next" ptr points backward. */
89 #define EXACTLY 8 /* str Match this string. */
90 #define NOTHING 9 /* no Match empty string. */
91 #define STAR 10 /* node Match this (simple) thing 0 or more times. */
92 #define PLUS 11 /* node Match this (simple) thing 1 or more times. */
93 #define WORDA 12 /* no Match "" at wordchar, where prev is nonword */
94 #define WORDZ 13 /* no Match "" at nonwordchar, where prev is word */
95 #define OPEN 20 /* no Mark this point in input as start of #n. */
96 /* OPEN+1 is number 1, etc. */
97 #define CLOSE 30 /* no Analogous to OPEN. */
100 * Opcode notes:
102 * BRANCH The set of branches constituting a single choice are hooked
103 * together with their "next" pointers, since precedence prevents
104 * anything being concatenated to any individual branch. The
105 * "next" pointer of the last BRANCH in a choice points to the
106 * thing following the whole choice. This is also where the
107 * final "next" pointer of each individual branch points; each
108 * branch starts with the operand node of a BRANCH node.
110 * BACK Normal "next" pointers all implicitly point forward; BACK
111 * exists to make loop structures possible.
113 * STAR,PLUS '?', and complex '*' and '+', are implemented as circular
114 * BRANCH structures using BACK. Simple cases (one character
115 * per match) are implemented with STAR and PLUS for speed
116 * and to minimize recursive plunges.
118 * OPEN,CLOSE ...are numbered at compile time.
122 * A node is one char of opcode followed by two chars of "next" pointer.
123 * "Next" pointers are stored as two 8-bit pieces, high order first. The
124 * value is a positive offset from the opcode of the node containing it.
125 * An operand, if any, simply follows the node. (Note that much of the
126 * code generation knows about this implicit relationship.)
128 * Using two bytes for the "next" pointer is vast overkill for most things,
129 * but allows patterns to get big without disasters.
131 #define OP(p) (*(p))
132 #define NEXT(p) (((*((p)+1)&0377)<<8) + (*((p)+2)&0377))
133 #define OPERAND(p) ((p) + 3)
136 * See regmagic.h for one further detail of program structure.
141 * Utility definitions.
143 #ifndef CHARBITS
144 #define UCHARAT(p) ((int)*(unsigned char *)(p))
145 #else
146 #define UCHARAT(p) ((int)*(p)&CHARBITS)
147 #endif
149 #define FAIL(m) { regerror(m); return(NULL); }
150 #define ISMULT(c) ((c) == '*' || (c) == '+' || (c) == '?')
153 * Flags to be passed up and down.
155 #define HASWIDTH 01 /* Known never to match null string. */
156 #define SIMPLE 02 /* Simple enough to be STAR/PLUS operand. */
157 #define SPSTART 04 /* Starts with * or +. */
158 #define WORST 0 /* Worst case. */
161 * Global work variables for regcomp().
163 static char *regparse; /* Input-scan pointer. */
164 static int regnpar; /* () count. */
165 static char regdummy;
166 static char *regcode; /* Code-emit pointer; &regdummy = don't. */
167 static long regsize; /* Code size. */
170 * Forward declarations for regcomp()'s friends.
172 #ifndef STATIC
173 #define STATIC static
174 #endif
175 STATIC char *reg(int, int *);
176 STATIC char *regbranch(int *);
177 STATIC char *regpiece(int *);
178 STATIC char *regatom(int *);
179 STATIC char *regnode(char);
180 STATIC char *regnext(char *);
181 STATIC void regc(char);
182 STATIC void reginsert(char, char *);
183 STATIC void regtail(char *, char *);
184 STATIC void regoptail(char *, char *);
185 #ifdef STRCSPN
186 STATIC int strcspn(char *, char *);
187 #endif
190 - regcomp - compile a regular expression into internal code
192 * We can't allocate space until we know how big the compiled form will be,
193 * but we can't compile it (and thus know how big it is) until we've got a
194 * place to put the code. So we cheat: we compile it twice, once with code
195 * generation turned off and size counting turned on, and once "for real".
196 * This also means that we don't allocate space until we are sure that the
197 * thing really will compile successfully, and we never have to move the
198 * code and thus invalidate pointers into it. (Note that it has to be in
199 * one piece because free() must be able to free it all.)
201 * Beware that the optimization-preparation code in here knows about some
202 * of the structure of the compiled regexp.
204 regexp *
205 regcomp(const char *exp)
207 regexp *r;
208 char *scan;
209 char *longest;
210 int len;
211 int flags;
213 if (exp == NULL)
214 FAIL("NULL argument");
216 /* First pass: determine size, legality. */
217 #ifdef notdef
218 if (exp[0] == '.' && exp[1] == '*') exp += 2; /* aid grep */
219 #endif
220 regparse = (char *)exp;
221 regnpar = 1;
222 regsize = 0L;
223 regcode = &regdummy;
224 regc(MAGIC);
225 if (reg(0, &flags) == NULL)
226 return(NULL);
228 /* Small enough for pointer-storage convention? */
229 if (regsize >= 32767L) /* Probably could be 65535L. */
230 FAIL("regexp too big");
232 /* Allocate space. */
233 r = (regexp *)malloc(sizeof(regexp) + (unsigned)regsize);
234 if (r == NULL)
235 FAIL("out of space");
237 /* Second pass: emit code. */
238 regparse = (char *)exp;
239 regnpar = 1;
240 regcode = r->program;
241 regc(MAGIC);
242 if (reg(0, &flags) == NULL)
243 return(NULL);
245 /* Dig out information for optimizations. */
246 r->regstart = '\0'; /* Worst-case defaults. */
247 r->reganch = 0;
248 r->regmust = NULL;
249 r->regmlen = 0;
250 scan = r->program+1; /* First BRANCH. */
251 if (OP(regnext(scan)) == END) { /* Only one top-level choice. */
252 scan = OPERAND(scan);
254 /* Starting-point info. */
255 if (OP(scan) == EXACTLY)
256 r->regstart = *OPERAND(scan);
257 else if (OP(scan) == BOL)
258 r->reganch++;
261 * If there's something expensive in the r.e., find the
262 * longest literal string that must appear and make it the
263 * regmust. Resolve ties in favor of later strings, since
264 * the regstart check works with the beginning of the r.e.
265 * and avoiding duplication strengthens checking. Not a
266 * strong reason, but sufficient in the absence of others.
268 if (flags&SPSTART) {
269 longest = NULL;
270 len = 0;
271 for (; scan != NULL; scan = regnext(scan))
272 if (OP(scan) == EXACTLY && strlen(OPERAND(scan)) >= len) {
273 longest = OPERAND(scan);
274 len = strlen(OPERAND(scan));
276 r->regmust = longest;
277 r->regmlen = len;
281 return(r);
285 - reg - regular expression, i.e. main body or parenthesized thing
287 * Caller must absorb opening parenthesis.
289 * Combining parenthesis handling with the base level of regular expression
290 * is a trifle forced, but the need to tie the tails of the branches to what
291 * follows makes it hard to avoid.
293 static char *
294 reg(int paren, int *flagp)
296 char *ret;
297 char *br;
298 char *ender;
299 int parno;
300 int flags;
302 *flagp = HASWIDTH; /* Tentatively. */
304 /* Make an OPEN node, if parenthesized. */
305 if (paren) {
306 if (regnpar >= NSUBEXP)
307 FAIL("too many ()");
308 parno = regnpar;
309 regnpar++;
310 ret = regnode(OPEN+parno);
311 } else
312 ret = NULL;
314 /* Pick up the branches, linking them together. */
315 br = regbranch(&flags);
316 if (br == NULL)
317 return(NULL);
318 if (ret != NULL)
319 regtail(ret, br); /* OPEN -> first. */
320 else
321 ret = br;
322 if (!(flags&HASWIDTH))
323 *flagp &= ~HASWIDTH;
324 *flagp |= flags&SPSTART;
325 while (*regparse == '|' || *regparse == '\n') {
326 regparse++;
327 br = regbranch(&flags);
328 if (br == NULL)
329 return(NULL);
330 regtail(ret, br); /* BRANCH -> BRANCH. */
331 if (!(flags&HASWIDTH))
332 *flagp &= ~HASWIDTH;
333 *flagp |= flags&SPSTART;
336 /* Make a closing node, and hook it on the end. */
337 ender = regnode((paren) ? CLOSE+parno : END);
338 regtail(ret, ender);
340 /* Hook the tails of the branches to the closing node. */
341 for (br = ret; br != NULL; br = regnext(br))
342 regoptail(br, ender);
344 /* Check for proper termination. */
345 if (paren && *regparse++ != ')') {
346 FAIL("unmatched ()");
347 } else if (!paren && *regparse != '\0') {
348 if (*regparse == ')') {
349 FAIL("unmatched ()");
350 } else
351 FAIL("junk on end"); /* "Can't happen". */
352 /* NOTREACHED */
355 return(ret);
359 - regbranch - one alternative of an | operator
361 * Implements the concatenation operator.
363 static char *
364 regbranch(int *flagp)
366 char *ret;
367 char *chain;
368 char *latest;
369 int flags;
371 *flagp = WORST; /* Tentatively. */
373 ret = regnode(BRANCH);
374 chain = NULL;
375 while (*regparse != '\0' && *regparse != ')' &&
376 *regparse != '\n' && *regparse != '|') {
377 latest = regpiece(&flags);
378 if (latest == NULL)
379 return(NULL);
380 *flagp |= flags&HASWIDTH;
381 if (chain == NULL) /* First piece. */
382 *flagp |= flags&SPSTART;
383 else
384 regtail(chain, latest);
385 chain = latest;
387 if (chain == NULL) /* Loop ran zero times. */
388 (void) regnode(NOTHING);
390 return(ret);
394 - regpiece - something followed by possible [*+?]
396 * Note that the branching code sequences used for ? and the general cases
397 * of * and + are somewhat optimized: they use the same NOTHING node as
398 * both the endmarker for their branch list and the body of the last branch.
399 * It might seem that this node could be dispensed with entirely, but the
400 * endmarker role is not redundant.
402 static char *
403 regpiece(int *flagp)
405 char *ret;
406 char op;
407 char *next;
408 int flags;
410 ret = regatom(&flags);
411 if (ret == NULL)
412 return(NULL);
414 op = *regparse;
415 if (!ISMULT(op)) {
416 *flagp = flags;
417 return(ret);
420 if (!(flags&HASWIDTH) && op != '?')
421 FAIL("*+ operand could be empty");
422 *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH);
424 if (op == '*' && (flags&SIMPLE))
425 reginsert(STAR, ret);
426 else if (op == '*') {
427 /* Emit x* as (x&|), where & means "self". */
428 reginsert(BRANCH, ret); /* Either x */
429 regoptail(ret, regnode(BACK)); /* and loop */
430 regoptail(ret, ret); /* back */
431 regtail(ret, regnode(BRANCH)); /* or */
432 regtail(ret, regnode(NOTHING)); /* null. */
433 } else if (op == '+' && (flags&SIMPLE))
434 reginsert(PLUS, ret);
435 else if (op == '+') {
436 /* Emit x+ as x(&|), where & means "self". */
437 next = regnode(BRANCH); /* Either */
438 regtail(ret, next);
439 regtail(regnode(BACK), ret); /* loop back */
440 regtail(next, regnode(BRANCH)); /* or */
441 regtail(ret, regnode(NOTHING)); /* null. */
442 } else if (op == '?') {
443 /* Emit x? as (x|) */
444 reginsert(BRANCH, ret); /* Either x */
445 regtail(ret, regnode(BRANCH)); /* or */
446 next = regnode(NOTHING); /* null. */
447 regtail(ret, next);
448 regoptail(ret, next);
450 regparse++;
451 if (ISMULT(*regparse))
452 FAIL("nested *?+");
454 return(ret);
458 - regatom - the lowest level
460 * Optimization: gobbles an entire sequence of ordinary characters so that
461 * it can turn them into a single node, which is smaller to store and
462 * faster to run. Backslashed characters are exceptions, each becoming a
463 * separate node; the code is simpler that way and it's not worth fixing.
465 static char *
466 regatom(int *flagp)
468 char *ret;
469 int flags;
471 *flagp = WORST; /* Tentatively. */
473 switch (*regparse++) {
474 /* FIXME: these chars only have meaning at beg/end of pat? */
475 case '^':
476 ret = regnode(BOL);
477 break;
478 case '$':
479 ret = regnode(EOL);
480 break;
481 case '.':
482 ret = regnode(ANY);
483 *flagp |= HASWIDTH|SIMPLE;
484 break;
485 case '[': {
486 int class;
487 int classend;
488 int i;
490 if (*regparse == '^') { /* Complement of range. */
491 ret = regnode(ANYBUT);
492 regparse++;
493 } else
494 ret = regnode(ANYOF);
495 if (*regparse == ']' || *regparse == '-')
496 regc(*regparse++);
497 while (*regparse != '\0' && *regparse != ']') {
498 if (*regparse == '-') {
499 regparse++;
500 if (*regparse == ']' || *regparse == '\0')
501 regc('-');
502 else {
503 class = UCHARAT(regparse-2);
504 classend = UCHARAT(regparse);
505 if (__collate_load_error) {
506 if (class > classend)
507 FAIL("invalid [] range");
508 for (class++; class <= classend; class++)
509 regc(class);
510 } else {
511 if (__collate_range_cmp(class, classend) > 0)
512 FAIL("invalid [] range");
513 for (i = 0; i <= UCHAR_MAX; i++)
514 if ( i != class
515 && __collate_range_cmp(class, i) <= 0
516 && __collate_range_cmp(i, classend) <= 0
518 regc(i);
520 regparse++;
522 } else
523 regc(*regparse++);
525 regc('\0');
526 if (*regparse != ']')
527 FAIL("unmatched []");
528 regparse++;
529 *flagp |= HASWIDTH|SIMPLE;
531 break;
532 case '(':
533 ret = reg(1, &flags);
534 if (ret == NULL)
535 return(NULL);
536 *flagp |= flags&(HASWIDTH|SPSTART);
537 break;
538 case '\0':
539 case '|':
540 case '\n':
541 case ')':
542 FAIL("internal urp"); /* Supposed to be caught earlier. */
543 break;
544 case '?':
545 case '+':
546 case '*':
547 FAIL("?+* follows nothing");
548 break;
549 case '\\':
550 switch (*regparse++) {
551 case '\0':
552 FAIL("trailing \\");
553 break;
554 case '<':
555 ret = regnode(WORDA);
556 break;
557 case '>':
558 ret = regnode(WORDZ);
559 break;
560 /* FIXME: Someday handle \1, \2, ... */
561 default:
562 /* Handle general quoted chars in exact-match routine */
563 goto de_fault;
565 break;
566 de_fault:
567 default:
569 * Encode a string of characters to be matched exactly.
571 * This is a bit tricky due to quoted chars and due to
572 * '*', '+', and '?' taking the SINGLE char previous
573 * as their operand.
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.)
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.
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.
593 char *regprev;
594 char ch;
596 regparse--; /* Look at cur char */
597 ret = regnode(EXACTLY);
598 for ( regprev = NULL ; ; ) {
599 ch = *regparse++; /* Get current char */
600 switch (*regparse) { /* look at next one */
602 default:
603 regc(ch); /* Add cur to string */
604 break;
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 */
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;
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; */
637 regprev = regparse; /* Set backup point */
639 done:
640 regc('\0');
641 *flagp |= HASWIDTH;
642 if (!regprev) /* One char? */
643 *flagp |= SIMPLE;
645 break;
648 return(ret);
652 - regnode - emit a node
654 static char * /* Location. */
655 regnode(char op)
657 char *ret;
658 char *ptr;
660 ret = regcode;
661 if (ret == &regdummy) {
662 regsize += 3;
663 return(ret);
666 ptr = ret;
667 *ptr++ = op;
668 *ptr++ = '\0'; /* Null "next" pointer. */
669 *ptr++ = '\0';
670 regcode = ptr;
672 return(ret);
676 - regc - emit (if appropriate) a byte of code
678 static void
679 regc(char b)
681 if (regcode != &regdummy)
682 *regcode++ = b;
683 else
684 regsize++;
688 - reginsert - insert an operator in front of already-emitted operand
690 * Means relocating the operand.
692 static void
693 reginsert(char op, char *opnd)
695 char *src;
696 char *dst;
697 char *place;
699 if (regcode == &regdummy) {
700 regsize += 3;
701 return;
704 src = regcode;
705 regcode += 3;
706 dst = regcode;
707 while (src > opnd)
708 *--dst = *--src;
710 place = opnd; /* Op node, where operand used to be. */
711 *place++ = op;
712 *place++ = '\0';
713 *place++ = '\0';
717 - regtail - set the next-pointer at the end of a node chain
719 static void
720 regtail(char *p, char *val)
722 char *scan;
723 char *temp;
724 int offset;
726 if (p == &regdummy)
727 return;
729 /* Find last node. */
730 scan = p;
731 for (;;) {
732 temp = regnext(scan);
733 if (temp == NULL)
734 break;
735 scan = temp;
738 if (OP(scan) == BACK)
739 offset = scan - val;
740 else
741 offset = val - scan;
742 *(scan+1) = (offset>>8)&0377;
743 *(scan+2) = offset&0377;
747 - regoptail - regtail on operand of first argument; nop if operandless
749 static void
750 regoptail(char *p, char *val)
752 /* "Operandless" and "op != BRANCH" are synonymous in practice. */
753 if (p == NULL || p == &regdummy || OP(p) != BRANCH)
754 return;
755 regtail(OPERAND(p), val);
759 * regexec and friends
763 * Global work variables for regexec().
765 static char *reginput; /* String-input pointer. */
766 static char *regbol; /* Beginning of input, for ^ check. */
767 static char **regstartp; /* Pointer to startp array. */
768 static char **regendp; /* Ditto for endp. */
771 * Forwards.
773 STATIC int regtry();
774 STATIC int regmatch();
775 STATIC int regrepeat();
777 #ifdef DEBUG
778 int regnarrate = 0;
779 void regdump();
780 STATIC char *regprop();
781 #endif
784 - regexec - match a regexp against a string
787 regexec(const regexp *prog, const char *string)
789 char *s;
790 extern char *strchr();
792 /* Be paranoid... */
793 if (prog == NULL || string == NULL) {
794 regerror("NULL parameter");
795 return(0);
798 /* Check validity of program. */
799 if (UCHARAT(prog->program) != MAGIC) {
800 regerror("corrupted program");
801 return(0);
804 /* If there is a "must appear" string, look for it. */
805 if (prog->regmust != NULL) {
806 s = (char *)string;
807 while ((s = strchr(s, prog->regmust[0])) != NULL) {
808 if (strncmp(s, prog->regmust, prog->regmlen) == 0)
809 break; /* Found it. */
810 s++;
812 if (s == NULL) /* Not present. */
813 return(0);
816 /* Mark beginning of line for ^ . */
817 regbol = (char *)string;
819 /* Simplest case: anchored match need be tried only once. */
820 if (prog->reganch)
821 return(regtry(prog, string));
823 /* Messy cases: unanchored match. */
824 s = (char *)string;
825 if (prog->regstart != '\0')
826 /* We know what char it must start with. */
827 while ((s = strchr(s, prog->regstart)) != NULL) {
828 if (regtry(prog, s))
829 return(1);
830 s++;
832 else
833 /* We don't -- general case. */
834 do {
835 if (regtry(prog, s))
836 return(1);
837 } while (*s++ != '\0');
839 /* Failure. */
840 return(0);
844 - regtry - try match at specific point
846 static int /* 0 failure, 1 success */
847 regtry(regexp *prog, char *string)
849 int i;
850 char **sp;
851 char **ep;
853 reginput = string;
854 regstartp = prog->startp;
855 regendp = prog->endp;
857 sp = prog->startp;
858 ep = prog->endp;
859 for (i = NSUBEXP; i > 0; i--) {
860 *sp++ = NULL;
861 *ep++ = NULL;
863 if (regmatch(prog->program + 1)) {
864 prog->startp[0] = string;
865 prog->endp[0] = reginput;
866 return(1);
867 } else
868 return(0);
872 - regmatch - main matching routine
874 * Conceptually the strategy is simple: check to see whether the current
875 * node matches, call self recursively to see whether the rest matches,
876 * and then act accordingly. In practice we make some effort to avoid
877 * recursion, in particular by going through "ordinary" nodes (that don't
878 * need to know whether the rest of the match failed) by a loop instead of
879 * by recursion.
881 static int /* 0 failure, 1 success */
882 regmatch(char *prog)
884 char *scan; /* Current node. */
885 char *next; /* Next node. */
887 scan = prog;
888 #ifdef DEBUG
889 if (scan != NULL && regnarrate)
890 fprintf(stderr, "%s(\n", regprop(scan));
891 #endif
892 while (scan != NULL) {
893 #ifdef DEBUG
894 if (regnarrate)
895 fprintf(stderr, "%s...\n", regprop(scan));
896 #endif
897 next = regnext(scan);
899 switch (OP(scan)) {
900 case BOL:
901 if (reginput != regbol)
902 return(0);
903 break;
904 case EOL:
905 if (*reginput != '\0')
906 return(0);
907 break;
908 case WORDA:
909 /* Must be looking at a letter, digit, or _ */
910 if ((!isalnum((unsigned char)*reginput)) && *reginput != '_')
911 return(0);
912 /* Prev must be BOL or nonword */
913 if (reginput > regbol &&
914 (isalnum((unsigned char)reginput[-1]) || reginput[-1] == '_'))
915 return(0);
916 break;
917 case WORDZ:
918 /* Must be looking at non letter, digit, or _ */
919 if (isalnum((unsigned char)*reginput) || *reginput == '_')
920 return(0);
921 /* We don't care what the previous char was */
922 break;
923 case ANY:
924 if (*reginput == '\0')
925 return(0);
926 reginput++;
927 break;
928 case EXACTLY: {
929 int len;
930 char *opnd;
932 opnd = OPERAND(scan);
933 /* Inline the first character, for speed. */
934 if (*opnd != *reginput)
935 return(0);
936 len = strlen(opnd);
937 if (len > 1 && strncmp(opnd, reginput, len) != 0)
938 return(0);
939 reginput += len;
941 break;
942 case ANYOF:
943 if (*reginput == '\0' || strchr(OPERAND(scan), *reginput) == NULL)
944 return(0);
945 reginput++;
946 break;
947 case ANYBUT:
948 if (*reginput == '\0' || strchr(OPERAND(scan), *reginput) != NULL)
949 return(0);
950 reginput++;
951 break;
952 case NOTHING:
953 break;
954 case BACK:
955 break;
956 case OPEN+1:
957 case OPEN+2:
958 case OPEN+3:
959 case OPEN+4:
960 case OPEN+5:
961 case OPEN+6:
962 case OPEN+7:
963 case OPEN+8:
964 case OPEN+9: {
965 int no;
966 char *save;
968 no = OP(scan) - OPEN;
969 save = reginput;
971 if (regmatch(next)) {
973 * Don't set startp if some later
974 * invocation of the same parentheses
975 * already has.
977 if (regstartp[no] == NULL)
978 regstartp[no] = save;
979 return(1);
980 } else
981 return(0);
983 break;
984 case CLOSE+1:
985 case CLOSE+2:
986 case CLOSE+3:
987 case CLOSE+4:
988 case CLOSE+5:
989 case CLOSE+6:
990 case CLOSE+7:
991 case CLOSE+8:
992 case CLOSE+9: {
993 int no;
994 char *save;
996 no = OP(scan) - CLOSE;
997 save = reginput;
999 if (regmatch(next)) {
1001 * Don't set endp if some later
1002 * invocation of the same parentheses
1003 * already has.
1005 if (regendp[no] == NULL)
1006 regendp[no] = save;
1007 return(1);
1008 } else
1009 return(0);
1011 break;
1012 case BRANCH: {
1013 char *save;
1015 if (OP(next) != BRANCH) /* No choice. */
1016 next = OPERAND(scan); /* Avoid recursion. */
1017 else {
1018 do {
1019 save = reginput;
1020 if (regmatch(OPERAND(scan)))
1021 return(1);
1022 reginput = save;
1023 scan = regnext(scan);
1024 } while (scan != NULL && OP(scan) == BRANCH);
1025 return(0);
1026 /* NOTREACHED */
1029 break;
1030 case STAR:
1031 case PLUS: {
1032 char nextch;
1033 int no;
1034 char *save;
1035 int min;
1038 * Lookahead to avoid useless match attempts
1039 * when we know what character comes next.
1041 nextch = '\0';
1042 if (OP(next) == EXACTLY)
1043 nextch = *OPERAND(next);
1044 min = (OP(scan) == STAR) ? 0 : 1;
1045 save = reginput;
1046 no = regrepeat(OPERAND(scan));
1047 while (no >= min) {
1048 /* If it could work, try it. */
1049 if (nextch == '\0' || *reginput == nextch)
1050 if (regmatch(next))
1051 return(1);
1052 /* Couldn't or didn't -- back up. */
1053 no--;
1054 reginput = save + no;
1056 return(0);
1058 break;
1059 case END:
1060 return(1); /* Success! */
1061 break;
1062 default:
1063 regerror("memory corruption");
1064 return(0);
1065 break;
1068 scan = next;
1072 * We get here only if there's trouble -- normally "case END" is
1073 * the terminating point.
1075 regerror("corrupted pointers");
1076 return(0);
1080 - regrepeat - repeatedly match something simple, report how many
1082 static int
1083 regrepeat(char *p)
1085 int count = 0;
1086 char *scan;
1087 char *opnd;
1089 scan = reginput;
1090 opnd = OPERAND(p);
1091 switch (OP(p)) {
1092 case ANY:
1093 count = strlen(scan);
1094 scan += count;
1095 break;
1096 case EXACTLY:
1097 while (*opnd == *scan) {
1098 count++;
1099 scan++;
1101 break;
1102 case ANYOF:
1103 while (*scan != '\0' && strchr(opnd, *scan) != NULL) {
1104 count++;
1105 scan++;
1107 break;
1108 case ANYBUT:
1109 while (*scan != '\0' && strchr(opnd, *scan) == NULL) {
1110 count++;
1111 scan++;
1113 break;
1114 default: /* Oh dear. Called inappropriately. */
1115 regerror("internal foulup");
1116 count = 0; /* Best compromise. */
1117 break;
1119 reginput = scan;
1121 return(count);
1125 - regnext - dig the "next" pointer out of a node
1127 static char *
1128 regnext(char *p)
1130 int offset;
1132 if (p == &regdummy)
1133 return(NULL);
1135 offset = NEXT(p);
1136 if (offset == 0)
1137 return(NULL);
1139 if (OP(p) == BACK)
1140 return(p-offset);
1141 else
1142 return(p+offset);
1145 #ifdef DEBUG
1147 STATIC char *regprop();
1150 - regdump - dump a regexp onto stdout in vaguely comprehensible form
1152 void
1153 regdump(regexp *r)
1155 char *s;
1156 char op = EXACTLY; /* Arbitrary non-END op. */
1157 char *next;
1158 extern char *strchr();
1161 s = r->program + 1;
1162 while (op != END) { /* While that wasn't END last time... */
1163 op = OP(s);
1164 printf("%2d%s", s-r->program, regprop(s)); /* Where, what. */
1165 next = regnext(s);
1166 if (next == NULL) /* Next ptr. */
1167 printf("(0)");
1168 else
1169 printf("(%d)", (s-r->program)+(next-s));
1170 s += 3;
1171 if (op == ANYOF || op == ANYBUT || op == EXACTLY) {
1172 /* Literal string, where present. */
1173 while (*s != '\0') {
1174 putchar(*s);
1175 s++;
1177 s++;
1179 putchar('\n');
1182 /* Header fields of interest. */
1183 if (r->regstart != '\0')
1184 printf("start `%c' ", r->regstart);
1185 if (r->reganch)
1186 printf("anchored ");
1187 if (r->regmust != NULL)
1188 printf("must have \"%s\"", r->regmust);
1189 printf("\n");
1193 - regprop - printable representation of opcode
1195 static char *
1196 regprop(char *op)
1198 char *p;
1199 static char buf[50];
1201 (void) strcpy(buf, ":");
1203 switch (OP(op)) {
1204 case BOL:
1205 p = "BOL";
1206 break;
1207 case EOL:
1208 p = "EOL";
1209 break;
1210 case ANY:
1211 p = "ANY";
1212 break;
1213 case ANYOF:
1214 p = "ANYOF";
1215 break;
1216 case ANYBUT:
1217 p = "ANYBUT";
1218 break;
1219 case BRANCH:
1220 p = "BRANCH";
1221 break;
1222 case EXACTLY:
1223 p = "EXACTLY";
1224 break;
1225 case NOTHING:
1226 p = "NOTHING";
1227 break;
1228 case BACK:
1229 p = "BACK";
1230 break;
1231 case END:
1232 p = "END";
1233 break;
1234 case OPEN+1:
1235 case OPEN+2:
1236 case OPEN+3:
1237 case OPEN+4:
1238 case OPEN+5:
1239 case OPEN+6:
1240 case OPEN+7:
1241 case OPEN+8:
1242 case OPEN+9:
1243 sprintf(buf+strlen(buf), "OPEN%d", OP(op)-OPEN);
1244 p = NULL;
1245 break;
1246 case CLOSE+1:
1247 case CLOSE+2:
1248 case CLOSE+3:
1249 case CLOSE+4:
1250 case CLOSE+5:
1251 case CLOSE+6:
1252 case CLOSE+7:
1253 case CLOSE+8:
1254 case CLOSE+9:
1255 sprintf(buf+strlen(buf), "CLOSE%d", OP(op)-CLOSE);
1256 p = NULL;
1257 break;
1258 case STAR:
1259 p = "STAR";
1260 break;
1261 case PLUS:
1262 p = "PLUS";
1263 break;
1264 case WORDA:
1265 p = "WORDA";
1266 break;
1267 case WORDZ:
1268 p = "WORDZ";
1269 break;
1270 default:
1271 regerror("corrupted opcode");
1272 break;
1274 if (p != NULL)
1275 (void) strcat(buf, p);
1276 return(buf);
1278 #endif
1281 * The following is provided for those people who do not have strcspn() in
1282 * their C libraries. They should get off their butts and do something
1283 * about it; at least one public-domain implementation of those (highly
1284 * useful) string routines has been published on Usenet.
1286 #ifdef STRCSPN
1288 * strcspn - find length of initial segment of s1 consisting entirely
1289 * of characters not from s2
1292 static int
1293 strcspn(char *s1, char *s2)
1295 char *scan1;
1296 char *scan2;
1297 int count;
1299 count = 0;
1300 for (scan1 = s1; *scan1 != '\0'; scan1++) {
1301 for (scan2 = s2; *scan2 != '\0';) /* ++ moved down. */
1302 if (*scan1 == *scan2++)
1303 return(count);
1304 count++;
1306 return(count);
1308 #endif