]>
Commit | Line | Data |
---|---|---|
6a5fa4e0 MG |
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 | * | |
21 | * Beware that some of this code is subtly aware of the way operator | |
22 | * precedence is structured in regular expressions. Serious changes in | |
23 | * regular-expression syntax might require a total rethink. | |
24 | * | |
25 | * *** NOTE: this code has been altered slightly for use in Tcl. *** | |
26 | * *** The only change is to use ckalloc and ckfree instead of *** | |
27 | * *** malloc and free. *** | |
28 | */ | |
29 | #include "tclint.h" | |
30 | ||
31 | /* | |
32 | * The "internal use only" fields in regexp.h are present to pass info from | |
33 | * compile to execute that permits the execute phase to run lots faster on | |
34 | * simple cases. They are: | |
35 | * | |
36 | * regstart char that must begin a match; '\0' if none obvious | |
37 | * reganch is the match anchored (at beginning-of-line only)? | |
38 | * regmust string (pointer into program) that match must include, or NULL | |
39 | * regmlen length of regmust string | |
40 | * | |
41 | * Regstart and reganch permit very fast decisions on suitable starting points | |
42 | * for a match, cutting down the work a lot. Regmust permits fast rejection | |
43 | * of lines that cannot possibly match. The regmust tests are costly enough | |
44 | * that regcomp() supplies a regmust only if the r.e. contains something | |
45 | * potentially expensive (at present, the only such thing detected is * or + | |
46 | * at the start of the r.e., which can involve a lot of backup). Regmlen is | |
47 | * supplied because the test in regexec() needs it and regcomp() is computing | |
48 | * it anyway. | |
49 | */ | |
50 | ||
51 | /* | |
52 | * Structure for regexp "program". This is essentially a linear encoding | |
53 | * of a nondeterministic finite-state machine (aka syntax charts or | |
54 | * "railroad normal form" in parsing technology). Each node is an opcode | |
55 | * plus a "next" pointer, possibly plus an operand. "Next" pointers of | |
56 | * all nodes except BRANCH implement concatenation; a "next" pointer with | |
57 | * a BRANCH on both ends of it is connecting two alternatives. (Here we | |
58 | * have one of the subtle syntax dependencies: an individual BRANCH (as | |
59 | * opposed to a collection of them) is never concatenated with anything | |
60 | * because of operator precedence.) The operand of some types of node is | |
61 | * a literal string; for others, it is a node leading into a sub-FSM. In | |
62 | * particular, the operand of a BRANCH node is the first node of the branch. | |
63 | * (NB this is *not* a tree structure: the tail of the branch connects | |
64 | * to the thing following the set of BRANCHes.) The opcodes are: | |
65 | */ | |
66 | ||
67 | /* definition number opnd? meaning */ | |
68 | #define END 0 /* no End of program. */ | |
69 | #define BOL 1 /* no Match "" at beginning of line. */ | |
70 | #define EOL 2 /* no Match "" at end of line. */ | |
71 | #define ANY 3 /* no Match any one character. */ | |
72 | #define ANYOF 4 /* str Match any character in this string. */ | |
73 | #define ANYBUT 5 /* str Match any character not in this string. */ | |
74 | #define BRANCH 6 /* node Match this alternative, or the next... */ | |
75 | #define BACK 7 /* no Match "", "next" ptr points backward. */ | |
76 | #define EXACTLY 8 /* str Match this string. */ | |
77 | #define NOTHING 9 /* no Match empty string. */ | |
78 | #define STAR 10 /* node Match this (simple) thing 0 or more times. */ | |
79 | #define PLUS 11 /* node Match this (simple) thing 1 or more times. */ | |
80 | #define OPEN 20 /* no Mark this point in input as start of #n. */ | |
81 | /* OPEN+1 is number 1, etc. */ | |
82 | #define CLOSE 30 /* no Analogous to OPEN. */ | |
83 | ||
84 | /* | |
85 | * Opcode notes: | |
86 | * | |
87 | * BRANCH The set of branches constituting a single choice are hooked | |
88 | * together with their "next" pointers, since precedence prevents | |
89 | * anything being concatenated to any individual branch. The | |
90 | * "next" pointer of the last BRANCH in a choice points to the | |
91 | * thing following the whole choice. This is also where the | |
92 | * final "next" pointer of each individual branch points; each | |
93 | * branch starts with the operand node of a BRANCH node. | |
94 | * | |
95 | * BACK Normal "next" pointers all implicitly point forward; BACK | |
96 | * exists to make loop structures possible. | |
97 | * | |
98 | * STAR,PLUS '?', and complex '*' and '+', are implemented as circular | |
99 | * BRANCH structures using BACK. Simple cases (one character | |
100 | * per match) are implemented with STAR and PLUS for speed | |
101 | * and to minimize recursive plunges. | |
102 | * | |
103 | * OPEN,CLOSE ...are numbered at compile time. | |
104 | */ | |
105 | ||
106 | /* | |
107 | * A node is one char of opcode followed by two chars of "next" pointer. | |
108 | * "Next" pointers are stored as two 8-bit pieces, high order first. The | |
109 | * value is a positive offset from the opcode of the node containing it. | |
110 | * An operand, if any, simply follows the node. (Note that much of the | |
111 | * code generation knows about this implicit relationship.) | |
112 | * | |
113 | * Using two bytes for the "next" pointer is vast overkill for most things, | |
114 | * but allows patterns to get big without disasters. | |
115 | */ | |
116 | #define OP(p) (*(p)) | |
117 | #define NEXT(p) (((*((p)+1)&0377)<<8) + (*((p)+2)&0377)) | |
118 | #define OPERAND(p) ((p) + 3) | |
119 | ||
120 | /* | |
121 | * See regmagic.h for one further detail of program structure. | |
122 | */ | |
123 | ||
124 | ||
125 | /* | |
126 | * Utility definitions. | |
127 | */ | |
128 | #ifndef CHARBITS | |
129 | #define UCHARAT(p) ((int)*(unsigned char *)(p)) | |
130 | #else | |
131 | #define UCHARAT(p) ((int)*(p)&CHARBITS) | |
132 | #endif | |
133 | ||
134 | #define FAIL(m) { regerror(m); return(NULL); } | |
135 | #define ISMULT(c) ((c) == '*' || (c) == '+' || (c) == '?') | |
136 | #define META "^$.[()|?+*\\" | |
137 | ||
138 | /* | |
139 | * Flags to be passed up and down. | |
140 | */ | |
141 | #define HASWIDTH 01 /* Known never to match null string. */ | |
142 | #define SIMPLE 02 /* Simple enough to be STAR/PLUS operand. */ | |
143 | #define SPSTART 04 /* Starts with * or +. */ | |
144 | #define WORST 0 /* Worst case. */ | |
145 | ||
146 | /* | |
147 | * Global work variables for regcomp(). | |
148 | */ | |
149 | static char *regparse; /* Input-scan pointer. */ | |
150 | static int regnpar; /* () count. */ | |
151 | static char regdummy; | |
152 | static char *regcode; /* Code-emit pointer; ®dummy = don't. */ | |
153 | static long regsize; /* Code size. */ | |
154 | ||
155 | /* | |
156 | * The first byte of the regexp internal "program" is actually this magic | |
157 | * number; the start node begins in the second byte. | |
158 | */ | |
159 | #define MAGIC 0234 | |
160 | ||
161 | ||
162 | /* | |
163 | * Forward declarations for regcomp()'s friends. | |
164 | */ | |
165 | #ifndef STATIC | |
166 | #define STATIC static | |
167 | #endif | |
168 | STATIC char *reg(); | |
169 | STATIC char *regbranch(); | |
170 | STATIC char *regpiece(); | |
171 | STATIC char *regatom(); | |
172 | STATIC char *regnode(); | |
173 | STATIC char *regnext(); | |
174 | STATIC void regc(); | |
175 | STATIC void reginsert(); | |
176 | STATIC void regtail(); | |
177 | STATIC void regoptail(); | |
178 | #ifdef STRCSPN | |
179 | STATIC int strcspn(); | |
180 | #endif | |
181 | ||
182 | /* | |
183 | - regcomp - compile a regular expression into internal code | |
184 | * | |
185 | * We can't allocate space until we know how big the compiled form will be, | |
186 | * but we can't compile it (and thus know how big it is) until we've got a | |
187 | * place to put the code. So we cheat: we compile it twice, once with code | |
188 | * generation turned off and size counting turned on, and once "for real". | |
189 | * This also means that we don't allocate space until we are sure that the | |
190 | * thing really will compile successfully, and we never have to move the | |
191 | * code and thus invalidate pointers into it. (Note that it has to be in | |
192 | * one piece because free() must be able to free it all.) | |
193 | * | |
194 | * Beware that the optimization-preparation code in here knows about some | |
195 | * of the structure of the compiled regexp. | |
196 | */ | |
197 | regexp * | |
198 | regcomp(exp) | |
199 | char *exp; | |
200 | { | |
201 | register regexp *r; | |
202 | register char *scan; | |
203 | register char *longest; | |
204 | register int len; | |
205 | int flags; | |
206 | ||
207 | if (exp == NULL) | |
208 | FAIL("NULL argument"); | |
209 | ||
210 | /* First pass: determine size, legality. */ | |
211 | regparse = exp; | |
212 | regnpar = 1; | |
213 | regsize = 0L; | |
214 | regcode = ®dummy; | |
215 | regc(MAGIC); | |
216 | if (reg(0, &flags) == NULL) | |
217 | return(NULL); | |
218 | ||
219 | /* Small enough for pointer-storage convention? */ | |
220 | if (regsize >= 32767L) /* Probably could be 65535L. */ | |
221 | FAIL("regexp too big"); | |
222 | ||
223 | /* Allocate space. */ | |
224 | r = (regexp *)ckalloc(sizeof(regexp) + (unsigned)regsize); | |
225 | if (r == NULL) | |
226 | FAIL("out of space"); | |
227 | ||
228 | /* Second pass: emit code. */ | |
229 | regparse = exp; | |
230 | regnpar = 1; | |
231 | regcode = r->program; | |
232 | regc(MAGIC); | |
233 | if (reg(0, &flags) == NULL) | |
234 | return(NULL); | |
235 | ||
236 | /* Dig out information for optimizations. */ | |
237 | r->regstart = '\0'; /* Worst-case defaults. */ | |
238 | r->reganch = 0; | |
239 | r->regmust = NULL; | |
240 | r->regmlen = 0; | |
241 | scan = r->program+1; /* First BRANCH. */ | |
242 | if (OP(regnext(scan)) == END) { /* Only one top-level choice. */ | |
243 | scan = OPERAND(scan); | |
244 | ||
245 | /* Starting-point info. */ | |
246 | if (OP(scan) == EXACTLY) | |
247 | r->regstart = *OPERAND(scan); | |
248 | else if (OP(scan) == BOL) | |
249 | r->reganch++; | |
250 | ||
251 | /* | |
252 | * If there's something expensive in the r.e., find the | |
253 | * longest literal string that must appear and make it the | |
254 | * regmust. Resolve ties in favor of later strings, since | |
255 | * the regstart check works with the beginning of the r.e. | |
256 | * and avoiding duplication strengthens checking. Not a | |
257 | * strong reason, but sufficient in the absence of others. | |
258 | */ | |
259 | if (flags&SPSTART) { | |
260 | longest = NULL; | |
261 | len = 0; | |
262 | for (; scan != NULL; scan = regnext(scan)) | |
263 | if (OP(scan) == EXACTLY && strlen(OPERAND(scan)) >= len) { | |
264 | longest = OPERAND(scan); | |
265 | len = strlen(OPERAND(scan)); | |
266 | } | |
267 | r->regmust = longest; | |
268 | r->regmlen = len; | |
269 | } | |
270 | } | |
271 | ||
272 | return(r); | |
273 | } | |
274 | ||
275 | /* | |
276 | - reg - regular expression, i.e. main body or parenthesized thing | |
277 | * | |
278 | * Caller must absorb opening parenthesis. | |
279 | * | |
280 | * Combining parenthesis handling with the base level of regular expression | |
281 | * is a trifle forced, but the need to tie the tails of the branches to what | |
282 | * follows makes it hard to avoid. | |
283 | */ | |
284 | static char * | |
285 | reg(paren, flagp) | |
286 | int paren; /* Parenthesized? */ | |
287 | int *flagp; | |
288 | { | |
289 | register char *ret; | |
290 | register char *br; | |
291 | register char *ender; | |
292 | register int parno = 0; | |
293 | int flags; | |
294 | ||
295 | *flagp = HASWIDTH; /* Tentatively. */ | |
296 | ||
297 | /* Make an OPEN node, if parenthesized. */ | |
298 | if (paren) { | |
299 | if (regnpar >= NSUBEXP) | |
300 | FAIL("too many ()"); | |
301 | parno = regnpar; | |
302 | regnpar++; | |
303 | ret = regnode(OPEN+parno); | |
304 | } else | |
305 | ret = NULL; | |
306 | ||
307 | /* Pick up the branches, linking them together. */ | |
308 | br = regbranch(&flags); | |
309 | if (br == NULL) | |
310 | return(NULL); | |
311 | if (ret != NULL) | |
312 | regtail(ret, br); /* OPEN -> first. */ | |
313 | else | |
314 | ret = br; | |
315 | if (!(flags&HASWIDTH)) | |
316 | *flagp &= ~HASWIDTH; | |
317 | *flagp |= flags&SPSTART; | |
318 | while (*regparse == '|') { | |
319 | regparse++; | |
320 | br = regbranch(&flags); | |
321 | if (br == NULL) | |
322 | return(NULL); | |
323 | regtail(ret, br); /* BRANCH -> BRANCH. */ | |
324 | if (!(flags&HASWIDTH)) | |
325 | *flagp &= ~HASWIDTH; | |
326 | *flagp |= flags&SPSTART; | |
327 | } | |
328 | ||
329 | /* Make a closing node, and hook it on the end. */ | |
330 | ender = regnode((paren) ? CLOSE+parno : END); | |
331 | regtail(ret, ender); | |
332 | ||
333 | /* Hook the tails of the branches to the closing node. */ | |
334 | for (br = ret; br != NULL; br = regnext(br)) | |
335 | regoptail(br, ender); | |
336 | ||
337 | /* Check for proper termination. */ | |
338 | if (paren && *regparse++ != ')') { | |
339 | FAIL("unmatched ()"); | |
340 | } else if (!paren && *regparse != '\0') { | |
341 | if (*regparse == ')') { | |
342 | FAIL("unmatched ()"); | |
343 | } else | |
344 | FAIL("junk on end"); /* "Can't happen". */ | |
345 | /* NOTREACHED */ | |
346 | } | |
347 | ||
348 | return(ret); | |
349 | } | |
350 | ||
351 | /* | |
352 | - regbranch - one alternative of an | operator | |
353 | * | |
354 | * Implements the concatenation operator. | |
355 | */ | |
356 | static char * | |
357 | regbranch(flagp) | |
358 | int *flagp; | |
359 | { | |
360 | register char *ret; | |
361 | register char *chain; | |
362 | register char *latest; | |
363 | int flags; | |
364 | ||
365 | *flagp = WORST; /* Tentatively. */ | |
366 | ||
367 | ret = regnode(BRANCH); | |
368 | chain = NULL; | |
369 | while (*regparse != '\0' && *regparse != '|' && *regparse != ')') { | |
370 | latest = regpiece(&flags); | |
371 | if (latest == NULL) | |
372 | return(NULL); | |
373 | *flagp |= flags&HASWIDTH; | |
374 | if (chain == NULL) /* First piece. */ | |
375 | *flagp |= flags&SPSTART; | |
376 | else | |
377 | regtail(chain, latest); | |
378 | chain = latest; | |
379 | } | |
380 | if (chain == NULL) /* Loop ran zero times. */ | |
381 | (void) regnode(NOTHING); | |
382 | ||
383 | return(ret); | |
384 | } | |
385 | ||
386 | /* | |
387 | - regpiece - something followed by possible [*+?] | |
388 | * | |
389 | * Note that the branching code sequences used for ? and the general cases | |
390 | * of * and + are somewhat optimized: they use the same NOTHING node as | |
391 | * both the endmarker for their branch list and the body of the last branch. | |
392 | * It might seem that this node could be dispensed with entirely, but the | |
393 | * endmarker role is not redundant. | |
394 | */ | |
395 | static char * | |
396 | regpiece(flagp) | |
397 | int *flagp; | |
398 | { | |
399 | register char *ret; | |
400 | register char op; | |
401 | register char *next; | |
402 | int flags; | |
403 | ||
404 | ret = regatom(&flags); | |
405 | if (ret == NULL) | |
406 | return(NULL); | |
407 | ||
408 | op = *regparse; | |
409 | if (!ISMULT(op)) { | |
410 | *flagp = flags; | |
411 | return(ret); | |
412 | } | |
413 | ||
414 | if (!(flags&HASWIDTH) && op != '?') | |
415 | FAIL("*+ operand could be empty"); | |
416 | *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH); | |
417 | ||
418 | if (op == '*' && (flags&SIMPLE)) | |
419 | reginsert(STAR, ret); | |
420 | else if (op == '*') { | |
421 | /* Emit x* as (x&|), where & means "self". */ | |
422 | reginsert(BRANCH, ret); /* Either x */ | |
423 | regoptail(ret, regnode(BACK)); /* and loop */ | |
424 | regoptail(ret, ret); /* back */ | |
425 | regtail(ret, regnode(BRANCH)); /* or */ | |
426 | regtail(ret, regnode(NOTHING)); /* null. */ | |
427 | } else if (op == '+' && (flags&SIMPLE)) | |
428 | reginsert(PLUS, ret); | |
429 | else if (op == '+') { | |
430 | /* Emit x+ as x(&|), where & means "self". */ | |
431 | next = regnode(BRANCH); /* Either */ | |
432 | regtail(ret, next); | |
433 | regtail(regnode(BACK), ret); /* loop back */ | |
434 | regtail(next, regnode(BRANCH)); /* or */ | |
435 | regtail(ret, regnode(NOTHING)); /* null. */ | |
436 | } else if (op == '?') { | |
437 | /* Emit x? as (x|) */ | |
438 | reginsert(BRANCH, ret); /* Either x */ | |
439 | regtail(ret, regnode(BRANCH)); /* or */ | |
440 | next = regnode(NOTHING); /* null. */ | |
441 | regtail(ret, next); | |
442 | regoptail(ret, next); | |
443 | } | |
444 | regparse++; | |
445 | if (ISMULT(*regparse)) | |
446 | FAIL("nested *?+"); | |
447 | ||
448 | return(ret); | |
449 | } | |
450 | ||
451 | /* | |
452 | - regatom - the lowest level | |
453 | * | |
454 | * Optimization: gobbles an entire sequence of ordinary characters so that | |
455 | * it can turn them into a single node, which is smaller to store and | |
456 | * faster to run. Backslashed characters are exceptions, each becoming a | |
457 | * separate node; the code is simpler that way and it's not worth fixing. | |
458 | */ | |
459 | static char * | |
460 | regatom(flagp) | |
461 | int *flagp; | |
462 | { | |
463 | register char *ret; | |
464 | int flags; | |
465 | ||
466 | *flagp = WORST; /* Tentatively. */ | |
467 | ||
468 | switch (*regparse++) { | |
469 | case '^': | |
470 | ret = regnode(BOL); | |
471 | break; | |
472 | case '$': | |
473 | ret = regnode(EOL); | |
474 | break; | |
475 | case '.': | |
476 | ret = regnode(ANY); | |
477 | *flagp |= HASWIDTH|SIMPLE; | |
478 | break; | |
479 | case '[': { | |
480 | register int clss; | |
481 | register int classend; | |
482 | ||
483 | if (*regparse == '^') { /* Complement of range. */ | |
484 | ret = regnode(ANYBUT); | |
485 | regparse++; | |
486 | } else | |
487 | ret = regnode(ANYOF); | |
488 | if (*regparse == ']' || *regparse == '-') | |
489 | regc(*regparse++); | |
490 | while (*regparse != '\0' && *regparse != ']') { | |
491 | if (*regparse == '-') { | |
492 | regparse++; | |
493 | if (*regparse == ']' || *regparse == '\0') | |
494 | regc('-'); | |
495 | else { | |
496 | clss = UCHARAT(regparse-2)+1; | |
497 | classend = UCHARAT(regparse); | |
498 | if (clss > classend+1) | |
499 | FAIL("invalid [] range"); | |
500 | for (; clss <= classend; clss++) | |
501 | regc(clss); | |
502 | regparse++; | |
503 | } | |
504 | } else | |
505 | regc(*regparse++); | |
506 | } | |
507 | regc('\0'); | |
508 | if (*regparse != ']') | |
509 | FAIL("unmatched []"); | |
510 | regparse++; | |
511 | *flagp |= HASWIDTH|SIMPLE; | |
512 | } | |
513 | break; | |
514 | case '(': | |
515 | ret = reg(1, &flags); | |
516 | if (ret == NULL) | |
517 | return(NULL); | |
518 | *flagp |= flags&(HASWIDTH|SPSTART); | |
519 | break; | |
520 | case '\0': | |
521 | case '|': | |
522 | case ')': | |
523 | FAIL("internal urp"); /* Supposed to be caught earlier. */ | |
524 | /* NOTREACHED */ | |
525 | break; | |
526 | case '?': | |
527 | case '+': | |
528 | case '*': | |
529 | FAIL("?+* follows nothing"); | |
530 | /* NOTREACHED */ | |
531 | break; | |
532 | case '\\': | |
533 | if (*regparse == '\0') | |
534 | FAIL("trailing \\"); | |
535 | ret = regnode(EXACTLY); | |
536 | regc(*regparse++); | |
537 | regc('\0'); | |
538 | *flagp |= HASWIDTH|SIMPLE; | |
539 | break; | |
540 | default: { | |
541 | register int len; | |
542 | register char ender; | |
543 | ||
544 | regparse--; | |
545 | len = strcspn(regparse, META); | |
546 | if (len <= 0) | |
547 | FAIL("internal disaster"); | |
548 | ender = *(regparse+len); | |
549 | if (len > 1 && ISMULT(ender)) | |
550 | len--; /* Back off clear of ?+* operand. */ | |
551 | *flagp |= HASWIDTH; | |
552 | if (len == 1) | |
553 | *flagp |= SIMPLE; | |
554 | ret = regnode(EXACTLY); | |
555 | while (len > 0) { | |
556 | regc(*regparse++); | |
557 | len--; | |
558 | } | |
559 | regc('\0'); | |
560 | } | |
561 | break; | |
562 | } | |
563 | ||
564 | return(ret); | |
565 | } | |
566 | ||
567 | /* | |
568 | - regnode - emit a node | |
569 | */ | |
570 | static char * /* Location. */ | |
571 | regnode(op) | |
572 | char op; | |
573 | { | |
574 | register char *ret; | |
575 | register char *ptr; | |
576 | ||
577 | ret = regcode; | |
578 | if (ret == ®dummy) { | |
579 | regsize += 3; | |
580 | return(ret); | |
581 | } | |
582 | ||
583 | ptr = ret; | |
584 | *ptr++ = op; | |
585 | *ptr++ = '\0'; /* Null "next" pointer. */ | |
586 | *ptr++ = '\0'; | |
587 | regcode = ptr; | |
588 | ||
589 | return(ret); | |
590 | } | |
591 | ||
592 | /* | |
593 | - regc - emit (if appropriate) a byte of code | |
594 | */ | |
595 | static void | |
596 | regc(b) | |
597 | char b; | |
598 | { | |
599 | if (regcode != ®dummy) | |
600 | *regcode++ = b; | |
601 | else | |
602 | regsize++; | |
603 | } | |
604 | ||
605 | /* | |
606 | - reginsert - insert an operator in front of already-emitted operand | |
607 | * | |
608 | * Means relocating the operand. | |
609 | */ | |
610 | static void | |
611 | reginsert(op, opnd) | |
612 | char op; | |
613 | char *opnd; | |
614 | { | |
615 | register char *src; | |
616 | register char *dst; | |
617 | register char *place; | |
618 | ||
619 | if (regcode == ®dummy) { | |
620 | regsize += 3; | |
621 | return; | |
622 | } | |
623 | ||
624 | src = regcode; | |
625 | regcode += 3; | |
626 | dst = regcode; | |
627 | while (src > opnd) | |
628 | *--dst = *--src; | |
629 | ||
630 | place = opnd; /* Op node, where operand used to be. */ | |
631 | *place++ = op; | |
632 | *place++ = '\0'; | |
633 | *place++ = '\0'; | |
634 | } | |
635 | ||
636 | /* | |
637 | - regtail - set the next-pointer at the end of a node chain | |
638 | */ | |
639 | static void | |
640 | regtail(p, val) | |
641 | char *p; | |
642 | char *val; | |
643 | { | |
644 | register char *scan; | |
645 | register char *temp; | |
646 | register int offset; | |
647 | ||
648 | if (p == ®dummy) | |
649 | return; | |
650 | ||
651 | /* Find last node. */ | |
652 | scan = p; | |
653 | for (;;) { | |
654 | temp = regnext(scan); | |
655 | if (temp == NULL) | |
656 | break; | |
657 | scan = temp; | |
658 | } | |
659 | ||
660 | if (OP(scan) == BACK) | |
661 | offset = scan - val; | |
662 | else | |
663 | offset = val - scan; | |
664 | *(scan+1) = (offset>>8)&0377; | |
665 | *(scan+2) = offset&0377; | |
666 | } | |
667 | ||
668 | /* | |
669 | - regoptail - regtail on operand of first argument; nop if operandless | |
670 | */ | |
671 | static void | |
672 | regoptail(p, val) | |
673 | char *p; | |
674 | char *val; | |
675 | { | |
676 | /* "Operandless" and "op != BRANCH" are synonymous in practice. */ | |
677 | if (p == NULL || p == ®dummy || OP(p) != BRANCH) | |
678 | return; | |
679 | regtail(OPERAND(p), val); | |
680 | } | |
681 | ||
682 | /* | |
683 | * regexec and friends | |
684 | */ | |
685 | ||
686 | /* | |
687 | * Global work variables for regexec(). | |
688 | */ | |
689 | static char *reginput; /* String-input pointer. */ | |
690 | static char *regbol; /* Beginning of input, for ^ check. */ | |
691 | static char **regstartp; /* Pointer to startp array. */ | |
692 | static char **regendp; /* Ditto for endp. */ | |
693 | ||
694 | /* | |
695 | * Forwards. | |
696 | */ | |
697 | STATIC int regtry(); | |
698 | STATIC int regmatch(); | |
699 | STATIC int regrepeat(); | |
700 | ||
701 | #ifdef DEBUG | |
702 | int regnarrate = 0; | |
703 | void regdump(); | |
704 | STATIC char *regprop(); | |
705 | #endif | |
706 | ||
707 | /* | |
708 | - regexec - match a regexp against a string | |
709 | */ | |
710 | int | |
711 | regexec(prog, string) | |
712 | register regexp *prog; | |
713 | register char *string; | |
714 | { | |
715 | register char *s; | |
716 | #ifndef IS_LINUX | |
717 | extern char *strchr(); | |
718 | #endif | |
719 | ||
720 | /* Be paranoid... */ | |
721 | if (prog == NULL || string == NULL) { | |
722 | regerror("NULL parameter"); | |
723 | return(0); | |
724 | } | |
725 | ||
726 | /* Check validity of program. */ | |
727 | if (UCHARAT(prog->program) != MAGIC) { | |
728 | regerror("corrupted program"); | |
729 | return(0); | |
730 | } | |
731 | ||
732 | /* If there is a "must appear" string, look for it. */ | |
733 | if (prog->regmust != NULL) { | |
734 | s = string; | |
735 | while ((s = strchr(s, prog->regmust[0])) != NULL) { | |
736 | if (strncmp(s, prog->regmust, prog->regmlen) == 0) | |
737 | break; /* Found it. */ | |
738 | s++; | |
739 | } | |
740 | if (s == NULL) /* Not present. */ | |
741 | return(0); | |
742 | } | |
743 | ||
744 | /* Mark beginning of line for ^ . */ | |
745 | regbol = string; | |
746 | ||
747 | /* Simplest case: anchored match need be tried only once. */ | |
748 | if (prog->reganch) | |
749 | return(regtry(prog, string)); | |
750 | ||
751 | /* Messy cases: unanchored match. */ | |
752 | s = string; | |
753 | if (prog->regstart != '\0') | |
754 | /* We know what char it must start with. */ | |
755 | while ((s = strchr(s, prog->regstart)) != NULL) { | |
756 | if (regtry(prog, s)) | |
757 | return(1); | |
758 | s++; | |
759 | } | |
760 | else | |
761 | /* We don't -- general case. */ | |
762 | do { | |
763 | if (regtry(prog, s)) | |
764 | return(1); | |
765 | } while (*s++ != '\0'); | |
766 | ||
767 | /* Failure. */ | |
768 | return(0); | |
769 | } | |
770 | ||
771 | /* | |
772 | - regtry - try match at specific point | |
773 | */ | |
774 | static int /* 0 failure, 1 success */ | |
775 | regtry(prog, string) | |
776 | regexp *prog; | |
777 | char *string; | |
778 | { | |
779 | register int i; | |
780 | register char **sp; | |
781 | register char **ep; | |
782 | ||
783 | reginput = string; | |
784 | regstartp = prog->startp; | |
785 | regendp = prog->endp; | |
786 | ||
787 | sp = prog->startp; | |
788 | ep = prog->endp; | |
789 | for (i = NSUBEXP; i > 0; i--) { | |
790 | *sp++ = NULL; | |
791 | *ep++ = NULL; | |
792 | } | |
793 | if (regmatch(prog->program + 1)) { | |
794 | prog->startp[0] = string; | |
795 | prog->endp[0] = reginput; | |
796 | return(1); | |
797 | } else | |
798 | return(0); | |
799 | } | |
800 | ||
801 | /* | |
802 | - regmatch - main matching routine | |
803 | * | |
804 | * Conceptually the strategy is simple: check to see whether the current | |
805 | * node matches, call self recursively to see whether the rest matches, | |
806 | * and then act accordingly. In practice we make some effort to avoid | |
807 | * recursion, in particular by going through "ordinary" nodes (that don't | |
808 | * need to know whether the rest of the match failed) by a loop instead of | |
809 | * by recursion. | |
810 | */ | |
811 | static int /* 0 failure, 1 success */ | |
812 | regmatch(prog) | |
813 | char *prog; | |
814 | { | |
815 | register char *scan; /* Current node. */ | |
816 | char *next; /* Next node. */ | |
817 | #ifndef IS_LINUX | |
818 | extern char *strchr(); | |
819 | #endif | |
820 | ||
821 | scan = prog; | |
822 | #ifdef DEBUG | |
823 | if (scan != NULL && regnarrate) | |
824 | fprintf(stderr, "%s(\n", regprop(scan)); | |
825 | #endif | |
826 | while (scan != NULL) { | |
827 | #ifdef DEBUG | |
828 | if (regnarrate) | |
829 | fprintf(stderr, "%s...\n", regprop(scan)); | |
830 | #endif | |
831 | next = regnext(scan); | |
832 | ||
833 | switch (OP(scan)) { | |
834 | case BOL: | |
835 | if (reginput != regbol) | |
836 | return(0); | |
837 | break; | |
838 | case EOL: | |
839 | if (*reginput != '\0') | |
840 | return(0); | |
841 | break; | |
842 | case ANY: | |
843 | if (*reginput == '\0') | |
844 | return(0); | |
845 | reginput++; | |
846 | break; | |
847 | case EXACTLY: { | |
848 | register int len; | |
849 | register char *opnd; | |
850 | ||
851 | opnd = OPERAND(scan); | |
852 | /* Inline the first character, for speed. */ | |
853 | if (*opnd != *reginput) | |
854 | return(0); | |
855 | len = strlen(opnd); | |
856 | if (len > 1 && strncmp(opnd, reginput, len) != 0) | |
857 | return(0); | |
858 | reginput += len; | |
859 | } | |
860 | break; | |
861 | case ANYOF: | |
862 | if (*reginput == '\0' || strchr(OPERAND(scan), *reginput) == NULL) | |
863 | return(0); | |
864 | reginput++; | |
865 | break; | |
866 | case ANYBUT: | |
867 | if (*reginput == '\0' || strchr(OPERAND(scan), *reginput) != NULL) | |
868 | return(0); | |
869 | reginput++; | |
870 | break; | |
871 | case NOTHING: | |
872 | break; | |
873 | case BACK: | |
874 | break; | |
875 | case OPEN+1: | |
876 | case OPEN+2: | |
877 | case OPEN+3: | |
878 | case OPEN+4: | |
879 | case OPEN+5: | |
880 | case OPEN+6: | |
881 | case OPEN+7: | |
882 | case OPEN+8: | |
883 | case OPEN+9: { | |
884 | register int no; | |
885 | register char *save; | |
886 | ||
887 | no = OP(scan) - OPEN; | |
888 | save = reginput; | |
889 | ||
890 | if (regmatch(next)) { | |
891 | /* | |
892 | * Don't set startp if some later | |
893 | * invocation of the same parentheses | |
894 | * already has. | |
895 | */ | |
896 | if (regstartp[no] == NULL) | |
897 | regstartp[no] = save; | |
898 | return(1); | |
899 | } else | |
900 | return(0); | |
901 | } | |
902 | /* NOTREACHED */ | |
903 | break; | |
904 | case CLOSE+1: | |
905 | case CLOSE+2: | |
906 | case CLOSE+3: | |
907 | case CLOSE+4: | |
908 | case CLOSE+5: | |
909 | case CLOSE+6: | |
910 | case CLOSE+7: | |
911 | case CLOSE+8: | |
912 | case CLOSE+9: { | |
913 | register int no; | |
914 | register char *save; | |
915 | ||
916 | no = OP(scan) - CLOSE; | |
917 | save = reginput; | |
918 | ||
919 | if (regmatch(next)) { | |
920 | /* | |
921 | * Don't set endp if some later | |
922 | * invocation of the same parentheses | |
923 | * already has. | |
924 | */ | |
925 | if (regendp[no] == NULL) | |
926 | regendp[no] = save; | |
927 | return(1); | |
928 | } else | |
929 | return(0); | |
930 | } | |
931 | /* NOTREACHED */ | |
932 | break; | |
933 | case BRANCH: { | |
934 | register char *save; | |
935 | ||
936 | if (OP(next) != BRANCH) /* No choice. */ | |
937 | next = OPERAND(scan); /* Avoid recursion. */ | |
938 | else { | |
939 | do { | |
940 | save = reginput; | |
941 | if (regmatch(OPERAND(scan))) | |
942 | return(1); | |
943 | reginput = save; | |
944 | scan = regnext(scan); | |
945 | } while (scan != NULL && OP(scan) == BRANCH); | |
946 | return(0); | |
947 | /* NOTREACHED */ | |
948 | } | |
949 | } | |
950 | /* NOTREACHED */ | |
951 | break; | |
952 | case STAR: | |
953 | case PLUS: { | |
954 | register char nextch; | |
955 | register int no; | |
956 | register char *save; | |
957 | register int min; | |
958 | ||
959 | /* | |
960 | * Lookahead to avoid useless match attempts | |
961 | * when we know what character comes next. | |
962 | */ | |
963 | nextch = '\0'; | |
964 | if (OP(next) == EXACTLY) | |
965 | nextch = *OPERAND(next); | |
966 | min = (OP(scan) == STAR) ? 0 : 1; | |
967 | save = reginput; | |
968 | no = regrepeat(OPERAND(scan)); | |
969 | while (no >= min) { | |
970 | /* If it could work, try it. */ | |
971 | if (nextch == '\0' || *reginput == nextch) | |
972 | if (regmatch(next)) | |
973 | return(1); | |
974 | /* Couldn't or didn't -- back up. */ | |
975 | no--; | |
976 | reginput = save + no; | |
977 | } | |
978 | return(0); | |
979 | } | |
980 | /* NOTREACHED */ | |
981 | break; | |
982 | case END: | |
983 | return(1); /* Success! */ | |
984 | /* NOTREACHED */ | |
985 | break; | |
986 | default: | |
987 | regerror("memory corruption"); | |
988 | return(0); | |
989 | /* NOTREACHED */ | |
990 | break; | |
991 | } | |
992 | ||
993 | scan = next; | |
994 | } | |
995 | ||
996 | /* | |
997 | * We get here only if there's trouble -- normally "case END" is | |
998 | * the terminating point. | |
999 | */ | |
1000 | regerror("corrupted pointers"); | |
1001 | return(0); | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | - regrepeat - repeatedly match something simple, report how many | |
1006 | */ | |
1007 | static int | |
1008 | regrepeat(p) | |
1009 | char *p; | |
1010 | { | |
1011 | register int count = 0; | |
1012 | register char *scan; | |
1013 | register char *opnd; | |
1014 | ||
1015 | scan = reginput; | |
1016 | opnd = OPERAND(p); | |
1017 | switch (OP(p)) { | |
1018 | case ANY: | |
1019 | count = strlen(scan); | |
1020 | scan += count; | |
1021 | break; | |
1022 | case EXACTLY: | |
1023 | while (*opnd == *scan) { | |
1024 | count++; | |
1025 | scan++; | |
1026 | } | |
1027 | break; | |
1028 | case ANYOF: | |
1029 | while (*scan != '\0' && strchr(opnd, *scan) != NULL) { | |
1030 | count++; | |
1031 | scan++; | |
1032 | } | |
1033 | break; | |
1034 | case ANYBUT: | |
1035 | while (*scan != '\0' && strchr(opnd, *scan) == NULL) { | |
1036 | count++; | |
1037 | scan++; | |
1038 | } | |
1039 | break; | |
1040 | default: /* Oh dear. Called inappropriately. */ | |
1041 | regerror("internal foulup"); | |
1042 | count = 0; /* Best compromise. */ | |
1043 | break; | |
1044 | } | |
1045 | reginput = scan; | |
1046 | ||
1047 | return(count); | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | - regnext - dig the "next" pointer out of a node | |
1052 | */ | |
1053 | static char * | |
1054 | regnext(p) | |
1055 | register char *p; | |
1056 | { | |
1057 | register int offset; | |
1058 | ||
1059 | if (p == ®dummy) | |
1060 | return(NULL); | |
1061 | ||
1062 | offset = NEXT(p); | |
1063 | if (offset == 0) | |
1064 | return(NULL); | |
1065 | ||
1066 | if (OP(p) == BACK) | |
1067 | return(p-offset); | |
1068 | else | |
1069 | return(p+offset); | |
1070 | } | |
1071 | ||
1072 | #ifdef DEBUG | |
1073 | ||
1074 | STATIC char *regprop(); | |
1075 | ||
1076 | /* | |
1077 | - regdump - dump a regexp onto stdout in vaguely comprehensible form | |
1078 | */ | |
1079 | void | |
1080 | regdump(r) | |
1081 | regexp *r; | |
1082 | { | |
1083 | register char *s; | |
1084 | register char op = EXACTLY; /* Arbitrary non-END op. */ | |
1085 | register char *next; | |
1086 | extern char *strchr(); | |
1087 | ||
1088 | ||
1089 | s = r->program + 1; | |
1090 | while (op != END) { /* While that wasn't END last time... */ | |
1091 | op = OP(s); | |
1092 | printf("%2d%s", s-r->program, regprop(s)); /* Where, what. */ | |
1093 | next = regnext(s); | |
1094 | if (next == NULL) /* Next ptr. */ | |
1095 | printf("(0)"); | |
1096 | else | |
1097 | printf("(%d)", (s-r->program)+(next-s)); | |
1098 | s += 3; | |
1099 | if (op == ANYOF || op == ANYBUT || op == EXACTLY) { | |
1100 | /* Literal string, where present. */ | |
1101 | while (*s != '\0') { | |
1102 | putchar(*s); | |
1103 | s++; | |
1104 | } | |
1105 | s++; | |
1106 | } | |
1107 | putchar('\n'); | |
1108 | } | |
1109 | ||
1110 | /* Header fields of interest. */ | |
1111 | if (r->regstart != '\0') | |
1112 | printf("start `%c' ", r->regstart); | |
1113 | if (r->reganch) | |
1114 | printf("anchored "); | |
1115 | if (r->regmust != NULL) | |
1116 | printf("must have \"%s\"", r->regmust); | |
1117 | printf("\n"); | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | - regprop - printable representation of opcode | |
1122 | */ | |
1123 | static char * | |
1124 | regprop(op) | |
1125 | char *op; | |
1126 | { | |
1127 | register char *p; | |
1128 | static char buf[50]; | |
1129 | ||
1130 | (void) strcpy(buf, ":"); | |
1131 | ||
1132 | switch (OP(op)) { | |
1133 | case BOL: | |
1134 | p = "BOL"; | |
1135 | break; | |
1136 | case EOL: | |
1137 | p = "EOL"; | |
1138 | break; | |
1139 | case ANY: | |
1140 | p = "ANY"; | |
1141 | break; | |
1142 | case ANYOF: | |
1143 | p = "ANYOF"; | |
1144 | break; | |
1145 | case ANYBUT: | |
1146 | p = "ANYBUT"; | |
1147 | break; | |
1148 | case BRANCH: | |
1149 | p = "BRANCH"; | |
1150 | break; | |
1151 | case EXACTLY: | |
1152 | p = "EXACTLY"; | |
1153 | break; | |
1154 | case NOTHING: | |
1155 | p = "NOTHING"; | |
1156 | break; | |
1157 | case BACK: | |
1158 | p = "BACK"; | |
1159 | break; | |
1160 | case END: | |
1161 | p = "END"; | |
1162 | break; | |
1163 | case OPEN+1: | |
1164 | case OPEN+2: | |
1165 | case OPEN+3: | |
1166 | case OPEN+4: | |
1167 | case OPEN+5: | |
1168 | case OPEN+6: | |
1169 | case OPEN+7: | |
1170 | case OPEN+8: | |
1171 | case OPEN+9: | |
1172 | sprintf(buf+strlen(buf), "OPEN%d", OP(op)-OPEN); | |
1173 | p = NULL; | |
1174 | break; | |
1175 | case CLOSE+1: | |
1176 | case CLOSE+2: | |
1177 | case CLOSE+3: | |
1178 | case CLOSE+4: | |
1179 | case CLOSE+5: | |
1180 | case CLOSE+6: | |
1181 | case CLOSE+7: | |
1182 | case CLOSE+8: | |
1183 | case CLOSE+9: | |
1184 | sprintf(buf+strlen(buf), "CLOSE%d", OP(op)-CLOSE); | |
1185 | p = NULL; | |
1186 | break; | |
1187 | case STAR: | |
1188 | p = "STAR"; | |
1189 | break; | |
1190 | case PLUS: | |
1191 | p = "PLUS"; | |
1192 | break; | |
1193 | default: | |
1194 | regerror("corrupted opcode"); | |
1195 | break; | |
1196 | } | |
1197 | if (p != NULL) | |
1198 | (void) strcat(buf, p); | |
1199 | return(buf); | |
1200 | } | |
1201 | #endif | |
1202 | ||
1203 | /* | |
1204 | * The following is provided for those people who do not have strcspn() in | |
1205 | * their C libraries. They should get off their butts and do something | |
1206 | * about it; at least one public-domain implementation of those (highly | |
1207 | * useful) string routines has been published on Usenet. | |
1208 | */ | |
1209 | #ifdef STRCSPN | |
1210 | /* | |
1211 | * strcspn - find length of initial segment of s1 consisting entirely | |
1212 | * of characters not from s2 | |
1213 | */ | |
1214 | ||
1215 | static int | |
1216 | strcspn(s1, s2) | |
1217 | char *s1; | |
1218 | char *s2; | |
1219 | { | |
1220 | register char *scan1; | |
1221 | register char *scan2; | |
1222 | register int count; | |
1223 | ||
1224 | count = 0; | |
1225 | for (scan1 = s1; *scan1 != '\0'; scan1++) { | |
1226 | for (scan2 = s2; *scan2 != '\0';) /* ++ moved down. */ | |
1227 | if (*scan1 == *scan2++) | |
1228 | return(count); | |
1229 | count++; | |
1230 | } | |
1231 | return(count); | |
1232 | } | |
1233 | #endif |