Commit | Line | Data |
---|---|---|
bc78d348 KB |
1 | /* Extended regular expression matching and search library, |
2 | version 0.11. | |
3 | (Implements POSIX draft P10003.2/D11.2, except for | |
4 | internationalization features.) | |
5 | ||
85586484 | 6 | Copyright (C) 1993 Free Software Foundation, Inc. |
bc78d348 KB |
7 | |
8 | This program is free software; you can redistribute it and/or modify | |
9 | it under the terms of the GNU General Public License as published by | |
10 | the Free Software Foundation; either version 2, or (at your option) | |
11 | any later version. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | GNU General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; if not, write to the Free Software | |
20 | Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ | |
21 | ||
22 | /* AIX requires this to be the first thing in the file. */ | |
23 | #if defined (_AIX) && !defined (REGEX_MALLOC) | |
24 | #pragma alloca | |
25 | #endif | |
26 | ||
27 | #define _GNU_SOURCE | |
28 | ||
29 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
30 | #include <sys/types.h> | |
31 | ||
71715da9 | 32 | #ifdef HAVE_CONFIG_H |
a8cfc9f7 DM |
33 | #include "config.h" |
34 | #endif | |
35 | ||
bc78d348 KB |
36 | /* The `emacs' switch turns on certain matching commands |
37 | that make sense only in Emacs. */ | |
38 | #ifdef emacs | |
39 | ||
bc78d348 KB |
40 | #include "lisp.h" |
41 | #include "buffer.h" | |
42 | #include "syntax.h" | |
43 | ||
44 | /* Emacs uses `NULL' as a predicate. */ | |
45 | #undef NULL | |
46 | ||
47 | #else /* not emacs */ | |
48 | ||
49 | /* We used to test for `BSTRING' here, but only GCC and Emacs define | |
50 | `BSTRING', as far as I know, and neither of them use this code. */ | |
18ec5b05 | 51 | #if HAVE_STRING_H || STDC_HEADERS |
bc78d348 | 52 | #include <string.h> |
9114e279 | 53 | #ifndef bcmp |
bc78d348 | 54 | #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n)) |
9114e279 KB |
55 | #endif |
56 | #ifndef bcopy | |
bc78d348 | 57 | #define bcopy(s, d, n) memcpy ((d), (s), (n)) |
9114e279 KB |
58 | #endif |
59 | #ifndef bzero | |
bc78d348 | 60 | #define bzero(s, n) memset ((s), 0, (n)) |
9114e279 | 61 | #endif |
bc78d348 KB |
62 | #else |
63 | #include <strings.h> | |
64 | #endif | |
65 | ||
66 | #ifdef STDC_HEADERS | |
67 | #include <stdlib.h> | |
68 | #else | |
69 | char *malloc (); | |
70 | char *realloc (); | |
71 | #endif | |
72 | ||
73 | ||
74 | /* Define the syntax stuff for \<, \>, etc. */ | |
75 | ||
76 | /* This must be nonzero for the wordchar and notwordchar pattern | |
77 | commands in re_match_2. */ | |
78 | #ifndef Sword | |
79 | #define Sword 1 | |
80 | #endif | |
81 | ||
82 | #ifdef SYNTAX_TABLE | |
83 | ||
84 | extern char *re_syntax_table; | |
85 | ||
86 | #else /* not SYNTAX_TABLE */ | |
87 | ||
88 | /* How many characters in the character set. */ | |
89 | #define CHAR_SET_SIZE 256 | |
90 | ||
91 | static char re_syntax_table[CHAR_SET_SIZE]; | |
92 | ||
93 | static void | |
94 | init_syntax_once () | |
95 | { | |
96 | register int c; | |
97 | static int done = 0; | |
98 | ||
99 | if (done) | |
100 | return; | |
101 | ||
102 | bzero (re_syntax_table, sizeof re_syntax_table); | |
103 | ||
104 | for (c = 'a'; c <= 'z'; c++) | |
105 | re_syntax_table[c] = Sword; | |
106 | ||
107 | for (c = 'A'; c <= 'Z'; c++) | |
108 | re_syntax_table[c] = Sword; | |
109 | ||
110 | for (c = '0'; c <= '9'; c++) | |
111 | re_syntax_table[c] = Sword; | |
112 | ||
113 | re_syntax_table['_'] = Sword; | |
114 | ||
115 | done = 1; | |
116 | } | |
117 | ||
118 | #endif /* not SYNTAX_TABLE */ | |
119 | ||
120 | #define SYNTAX(c) re_syntax_table[c] | |
121 | ||
122 | #endif /* not emacs */ | |
123 | \f | |
124 | /* Get the interface, including the syntax bits. */ | |
125 | #include "regex.h" | |
126 | ||
bc78d348 KB |
127 | /* isalpha etc. are used for the character classes. */ |
128 | #include <ctype.h> | |
c6b40788 JM |
129 | |
130 | #ifndef isascii | |
131 | #define isascii(c) 1 | |
bc78d348 | 132 | #endif |
c6b40788 JM |
133 | |
134 | #ifdef isblank | |
135 | #define ISBLANK(c) (isascii (c) && isblank (c)) | |
136 | #else | |
137 | #define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
bc78d348 | 138 | #endif |
c6b40788 JM |
139 | #ifdef isgraph |
140 | #define ISGRAPH(c) (isascii (c) && isgraph (c)) | |
141 | #else | |
142 | #define ISGRAPH(c) (isascii (c) && isprint (c) && !isspace (c)) | |
143 | #endif | |
144 | ||
145 | #define ISPRINT(c) (isascii (c) && isprint (c)) | |
146 | #define ISDIGIT(c) (isascii (c) && isdigit (c)) | |
147 | #define ISALNUM(c) (isascii (c) && isalnum (c)) | |
148 | #define ISALPHA(c) (isascii (c) && isalpha (c)) | |
149 | #define ISCNTRL(c) (isascii (c) && iscntrl (c)) | |
150 | #define ISLOWER(c) (isascii (c) && islower (c)) | |
151 | #define ISPUNCT(c) (isascii (c) && ispunct (c)) | |
152 | #define ISSPACE(c) (isascii (c) && isspace (c)) | |
153 | #define ISUPPER(c) (isascii (c) && isupper (c)) | |
154 | #define ISXDIGIT(c) (isascii (c) && isxdigit (c)) | |
bc78d348 KB |
155 | |
156 | #ifndef NULL | |
157 | #define NULL 0 | |
158 | #endif | |
159 | ||
160 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
161 | since ours (we hope) works properly with all combinations of | |
162 | machines, compilers, `char' and `unsigned char' argument types. | |
163 | (Per Bothner suggested the basic approach.) */ | |
164 | #undef SIGN_EXTEND_CHAR | |
165 | #if __STDC__ | |
166 | #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
9114e279 | 167 | #else /* not __STDC__ */ |
bc78d348 KB |
168 | /* As in Harbison and Steele. */ |
169 | #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
170 | #endif | |
171 | \f | |
172 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
173 | use `alloca' instead of `malloc'. This is because using malloc in | |
174 | re_search* or re_match* could cause memory leaks when C-g is used in | |
175 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
176 | the other hand, malloc is more portable, and easier to debug. | |
177 | ||
178 | Because we sometimes use alloca, some routines have to be macros, | |
179 | not functions -- `alloca'-allocated space disappears at the end of the | |
180 | function it is called in. */ | |
181 | ||
182 | #ifdef REGEX_MALLOC | |
183 | ||
184 | #define REGEX_ALLOCATE malloc | |
185 | #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
186 | ||
187 | #else /* not REGEX_MALLOC */ | |
188 | ||
189 | /* Emacs already defines alloca, sometimes. */ | |
190 | #ifndef alloca | |
191 | ||
192 | /* Make alloca work the best possible way. */ | |
193 | #ifdef __GNUC__ | |
194 | #define alloca __builtin_alloca | |
195 | #else /* not __GNUC__ */ | |
196 | #if HAVE_ALLOCA_H | |
197 | #include <alloca.h> | |
198 | #else /* not __GNUC__ or HAVE_ALLOCA_H */ | |
199 | #ifndef _AIX /* Already did AIX, up at the top. */ | |
200 | char *alloca (); | |
201 | #endif /* not _AIX */ | |
202 | #endif /* not HAVE_ALLOCA_H */ | |
203 | #endif /* not __GNUC__ */ | |
204 | ||
205 | #endif /* not alloca */ | |
206 | ||
207 | #define REGEX_ALLOCATE alloca | |
208 | ||
209 | /* Assumes a `char *destination' variable. */ | |
210 | #define REGEX_REALLOCATE(source, osize, nsize) \ | |
211 | (destination = (char *) alloca (nsize), \ | |
212 | bcopy (source, destination, osize), \ | |
213 | destination) | |
214 | ||
215 | #endif /* not REGEX_MALLOC */ | |
216 | ||
217 | ||
218 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
219 | `string1' or just past its end. This works if PTR is NULL, which is | |
220 | a good thing. */ | |
221 | #define FIRST_STRING_P(ptr) \ | |
222 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
223 | ||
224 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
225 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
226 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
227 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
228 | ||
229 | #define BYTEWIDTH 8 /* In bits. */ | |
230 | ||
231 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
232 | ||
233 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
234 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
235 | ||
236 | typedef char boolean; | |
237 | #define false 0 | |
238 | #define true 1 | |
239 | \f | |
240 | /* These are the command codes that appear in compiled regular | |
241 | expressions. Some opcodes are followed by argument bytes. A | |
242 | command code can specify any interpretation whatsoever for its | |
243 | arguments. Zero bytes may appear in the compiled regular expression. | |
244 | ||
245 | The value of `exactn' is needed in search.c (search_buffer) in Emacs. | |
246 | So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of | |
247 | `exactn' we use here must also be 1. */ | |
248 | ||
249 | typedef enum | |
250 | { | |
251 | no_op = 0, | |
252 | ||
253 | /* Followed by one byte giving n, then by n literal bytes. */ | |
254 | exactn = 1, | |
255 | ||
256 | /* Matches any (more or less) character. */ | |
257 | anychar, | |
258 | ||
259 | /* Matches any one char belonging to specified set. First | |
260 | following byte is number of bitmap bytes. Then come bytes | |
261 | for a bitmap saying which chars are in. Bits in each byte | |
262 | are ordered low-bit-first. A character is in the set if its | |
263 | bit is 1. A character too large to have a bit in the map is | |
264 | automatically not in the set. */ | |
265 | charset, | |
266 | ||
267 | /* Same parameters as charset, but match any character that is | |
268 | not one of those specified. */ | |
269 | charset_not, | |
270 | ||
271 | /* Start remembering the text that is matched, for storing in a | |
272 | register. Followed by one byte with the register number, in | |
273 | the range 0 to one less than the pattern buffer's re_nsub | |
274 | field. Then followed by one byte with the number of groups | |
275 | inner to this one. (This last has to be part of the | |
276 | start_memory only because we need it in the on_failure_jump | |
277 | of re_match_2.) */ | |
278 | start_memory, | |
279 | ||
280 | /* Stop remembering the text that is matched and store it in a | |
281 | memory register. Followed by one byte with the register | |
282 | number, in the range 0 to one less than `re_nsub' in the | |
283 | pattern buffer, and one byte with the number of inner groups, | |
284 | just like `start_memory'. (We need the number of inner | |
285 | groups here because we don't have any easy way of finding the | |
286 | corresponding start_memory when we're at a stop_memory.) */ | |
287 | stop_memory, | |
288 | ||
289 | /* Match a duplicate of something remembered. Followed by one | |
290 | byte containing the register number. */ | |
291 | duplicate, | |
292 | ||
293 | /* Fail unless at beginning of line. */ | |
294 | begline, | |
295 | ||
296 | /* Fail unless at end of line. */ | |
297 | endline, | |
298 | ||
299 | /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
300 | of string to be matched (if not). */ | |
301 | begbuf, | |
302 | ||
303 | /* Analogously, for end of buffer/string. */ | |
304 | endbuf, | |
305 | ||
306 | /* Followed by two byte relative address to which to jump. */ | |
307 | jump, | |
308 | ||
309 | /* Same as jump, but marks the end of an alternative. */ | |
310 | jump_past_alt, | |
311 | ||
312 | /* Followed by two-byte relative address of place to resume at | |
313 | in case of failure. */ | |
314 | on_failure_jump, | |
315 | ||
316 | /* Like on_failure_jump, but pushes a placeholder instead of the | |
317 | current string position when executed. */ | |
318 | on_failure_keep_string_jump, | |
319 | ||
320 | /* Throw away latest failure point and then jump to following | |
321 | two-byte relative address. */ | |
322 | pop_failure_jump, | |
323 | ||
324 | /* Change to pop_failure_jump if know won't have to backtrack to | |
325 | match; otherwise change to jump. This is used to jump | |
326 | back to the beginning of a repeat. If what follows this jump | |
327 | clearly won't match what the repeat does, such that we can be | |
328 | sure that there is no use backtracking out of repetitions | |
329 | already matched, then we change it to a pop_failure_jump. | |
330 | Followed by two-byte address. */ | |
331 | maybe_pop_jump, | |
332 | ||
333 | /* Jump to following two-byte address, and push a dummy failure | |
334 | point. This failure point will be thrown away if an attempt | |
335 | is made to use it for a failure. A `+' construct makes this | |
336 | before the first repeat. Also used as an intermediary kind | |
337 | of jump when compiling an alternative. */ | |
338 | dummy_failure_jump, | |
339 | ||
340 | /* Push a dummy failure point and continue. Used at the end of | |
341 | alternatives. */ | |
342 | push_dummy_failure, | |
343 | ||
344 | /* Followed by two-byte relative address and two-byte number n. | |
345 | After matching N times, jump to the address upon failure. */ | |
346 | succeed_n, | |
347 | ||
348 | /* Followed by two-byte relative address, and two-byte number n. | |
349 | Jump to the address N times, then fail. */ | |
350 | jump_n, | |
351 | ||
352 | /* Set the following two-byte relative address to the | |
353 | subsequent two-byte number. The address *includes* the two | |
354 | bytes of number. */ | |
355 | set_number_at, | |
356 | ||
357 | wordchar, /* Matches any word-constituent character. */ | |
358 | notwordchar, /* Matches any char that is not a word-constituent. */ | |
359 | ||
360 | wordbeg, /* Succeeds if at word beginning. */ | |
361 | wordend, /* Succeeds if at word end. */ | |
362 | ||
363 | wordbound, /* Succeeds if at a word boundary. */ | |
364 | notwordbound /* Succeeds if not at a word boundary. */ | |
365 | ||
366 | #ifdef emacs | |
367 | ,before_dot, /* Succeeds if before point. */ | |
368 | at_dot, /* Succeeds if at point. */ | |
369 | after_dot, /* Succeeds if after point. */ | |
370 | ||
371 | /* Matches any character whose syntax is specified. Followed by | |
372 | a byte which contains a syntax code, e.g., Sword. */ | |
373 | syntaxspec, | |
374 | ||
375 | /* Matches any character whose syntax is not that specified. */ | |
376 | notsyntaxspec | |
377 | #endif /* emacs */ | |
378 | } re_opcode_t; | |
379 | \f | |
380 | /* Common operations on the compiled pattern. */ | |
381 | ||
382 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
383 | ||
384 | #define STORE_NUMBER(destination, number) \ | |
385 | do { \ | |
386 | (destination)[0] = (number) & 0377; \ | |
387 | (destination)[1] = (number) >> 8; \ | |
388 | } while (0) | |
389 | ||
390 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
391 | the byte after where the number is stored. Therefore, DESTINATION | |
392 | must be an lvalue. */ | |
393 | ||
394 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
395 | do { \ | |
396 | STORE_NUMBER (destination, number); \ | |
397 | (destination) += 2; \ | |
398 | } while (0) | |
399 | ||
400 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
401 | at SOURCE. */ | |
402 | ||
403 | #define EXTRACT_NUMBER(destination, source) \ | |
404 | do { \ | |
405 | (destination) = *(source) & 0377; \ | |
406 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
407 | } while (0) | |
408 | ||
409 | #ifdef DEBUG | |
410 | static void | |
411 | extract_number (dest, source) | |
412 | int *dest; | |
413 | unsigned char *source; | |
414 | { | |
415 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
416 | *dest = *source & 0377; | |
417 | *dest += temp << 8; | |
418 | } | |
419 | ||
420 | #ifndef EXTRACT_MACROS /* To debug the macros. */ | |
421 | #undef EXTRACT_NUMBER | |
422 | #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
423 | #endif /* not EXTRACT_MACROS */ | |
424 | ||
425 | #endif /* DEBUG */ | |
426 | ||
427 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
428 | SOURCE must be an lvalue. */ | |
429 | ||
430 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
431 | do { \ | |
432 | EXTRACT_NUMBER (destination, source); \ | |
433 | (source) += 2; \ | |
434 | } while (0) | |
435 | ||
436 | #ifdef DEBUG | |
437 | static void | |
438 | extract_number_and_incr (destination, source) | |
439 | int *destination; | |
440 | unsigned char **source; | |
441 | { | |
442 | extract_number (destination, *source); | |
443 | *source += 2; | |
444 | } | |
445 | ||
446 | #ifndef EXTRACT_MACROS | |
447 | #undef EXTRACT_NUMBER_AND_INCR | |
448 | #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
449 | extract_number_and_incr (&dest, &src) | |
450 | #endif /* not EXTRACT_MACROS */ | |
451 | ||
452 | #endif /* DEBUG */ | |
453 | \f | |
454 | /* If DEBUG is defined, Regex prints many voluminous messages about what | |
455 | it is doing (if the variable `debug' is nonzero). If linked with the | |
456 | main program in `iregex.c', you can enter patterns and strings | |
457 | interactively. And if linked with the main program in `main.c' and | |
458 | the other test files, you can run the already-written tests. */ | |
459 | ||
460 | #ifdef DEBUG | |
461 | ||
462 | /* We use standard I/O for debugging. */ | |
463 | #include <stdio.h> | |
464 | ||
465 | /* It is useful to test things that ``must'' be true when debugging. */ | |
466 | #include <assert.h> | |
467 | ||
468 | static int debug = 0; | |
469 | ||
470 | #define DEBUG_STATEMENT(e) e | |
471 | #define DEBUG_PRINT1(x) if (debug) printf (x) | |
472 | #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
473 | #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
9114e279 | 474 | #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) |
bc78d348 KB |
475 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
476 | if (debug) print_partial_compiled_pattern (s, e) | |
477 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
478 | if (debug) print_double_string (w, s1, sz1, s2, sz2) | |
479 | ||
480 | ||
481 | extern void printchar (); | |
482 | ||
483 | /* Print the fastmap in human-readable form. */ | |
484 | ||
485 | void | |
486 | print_fastmap (fastmap) | |
487 | char *fastmap; | |
488 | { | |
489 | unsigned was_a_range = 0; | |
490 | unsigned i = 0; | |
491 | ||
492 | while (i < (1 << BYTEWIDTH)) | |
493 | { | |
494 | if (fastmap[i++]) | |
495 | { | |
496 | was_a_range = 0; | |
497 | printchar (i - 1); | |
498 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
499 | { | |
500 | was_a_range = 1; | |
501 | i++; | |
502 | } | |
503 | if (was_a_range) | |
504 | { | |
505 | printf ("-"); | |
506 | printchar (i - 1); | |
507 | } | |
508 | } | |
509 | } | |
510 | putchar ('\n'); | |
511 | } | |
512 | ||
513 | ||
514 | /* Print a compiled pattern string in human-readable form, starting at | |
515 | the START pointer into it and ending just before the pointer END. */ | |
516 | ||
517 | void | |
518 | print_partial_compiled_pattern (start, end) | |
519 | unsigned char *start; | |
520 | unsigned char *end; | |
521 | { | |
522 | int mcnt, mcnt2; | |
523 | unsigned char *p = start; | |
524 | unsigned char *pend = end; | |
525 | ||
526 | if (start == NULL) | |
527 | { | |
528 | printf ("(null)\n"); | |
529 | return; | |
530 | } | |
531 | ||
532 | /* Loop over pattern commands. */ | |
533 | while (p < pend) | |
534 | { | |
535 | switch ((re_opcode_t) *p++) | |
536 | { | |
537 | case no_op: | |
538 | printf ("/no_op"); | |
539 | break; | |
540 | ||
541 | case exactn: | |
542 | mcnt = *p++; | |
543 | printf ("/exactn/%d", mcnt); | |
544 | do | |
545 | { | |
546 | putchar ('/'); | |
547 | printchar (*p++); | |
548 | } | |
549 | while (--mcnt); | |
550 | break; | |
551 | ||
552 | case start_memory: | |
553 | mcnt = *p++; | |
554 | printf ("/start_memory/%d/%d", mcnt, *p++); | |
555 | break; | |
556 | ||
557 | case stop_memory: | |
558 | mcnt = *p++; | |
559 | printf ("/stop_memory/%d/%d", mcnt, *p++); | |
560 | break; | |
561 | ||
562 | case duplicate: | |
563 | printf ("/duplicate/%d", *p++); | |
564 | break; | |
565 | ||
566 | case anychar: | |
567 | printf ("/anychar"); | |
568 | break; | |
569 | ||
570 | case charset: | |
571 | case charset_not: | |
572 | { | |
573 | register int c; | |
574 | ||
575 | printf ("/charset%s", | |
576 | (re_opcode_t) *(p - 1) == charset_not ? "_not" : ""); | |
577 | ||
578 | assert (p + *p < pend); | |
579 | ||
580 | for (c = 0; c < *p; c++) | |
581 | { | |
582 | unsigned bit; | |
583 | unsigned char map_byte = p[1 + c]; | |
584 | ||
585 | putchar ('/'); | |
586 | ||
587 | for (bit = 0; bit < BYTEWIDTH; bit++) | |
588 | if (map_byte & (1 << bit)) | |
589 | printchar (c * BYTEWIDTH + bit); | |
590 | } | |
591 | p += 1 + *p; | |
592 | break; | |
593 | } | |
594 | ||
595 | case begline: | |
596 | printf ("/begline"); | |
597 | break; | |
598 | ||
599 | case endline: | |
600 | printf ("/endline"); | |
601 | break; | |
602 | ||
603 | case on_failure_jump: | |
604 | extract_number_and_incr (&mcnt, &p); | |
605 | printf ("/on_failure_jump/0/%d", mcnt); | |
606 | break; | |
607 | ||
608 | case on_failure_keep_string_jump: | |
609 | extract_number_and_incr (&mcnt, &p); | |
610 | printf ("/on_failure_keep_string_jump/0/%d", mcnt); | |
611 | break; | |
612 | ||
613 | case dummy_failure_jump: | |
614 | extract_number_and_incr (&mcnt, &p); | |
615 | printf ("/dummy_failure_jump/0/%d", mcnt); | |
616 | break; | |
617 | ||
618 | case push_dummy_failure: | |
619 | printf ("/push_dummy_failure"); | |
620 | break; | |
621 | ||
622 | case maybe_pop_jump: | |
623 | extract_number_and_incr (&mcnt, &p); | |
624 | printf ("/maybe_pop_jump/0/%d", mcnt); | |
625 | break; | |
626 | ||
627 | case pop_failure_jump: | |
628 | extract_number_and_incr (&mcnt, &p); | |
629 | printf ("/pop_failure_jump/0/%d", mcnt); | |
630 | break; | |
631 | ||
632 | case jump_past_alt: | |
633 | extract_number_and_incr (&mcnt, &p); | |
634 | printf ("/jump_past_alt/0/%d", mcnt); | |
635 | break; | |
636 | ||
637 | case jump: | |
638 | extract_number_and_incr (&mcnt, &p); | |
639 | printf ("/jump/0/%d", mcnt); | |
640 | break; | |
641 | ||
642 | case succeed_n: | |
643 | extract_number_and_incr (&mcnt, &p); | |
644 | extract_number_and_incr (&mcnt2, &p); | |
645 | printf ("/succeed_n/0/%d/0/%d", mcnt, mcnt2); | |
646 | break; | |
647 | ||
648 | case jump_n: | |
649 | extract_number_and_incr (&mcnt, &p); | |
650 | extract_number_and_incr (&mcnt2, &p); | |
651 | printf ("/jump_n/0/%d/0/%d", mcnt, mcnt2); | |
652 | break; | |
653 | ||
654 | case set_number_at: | |
655 | extract_number_and_incr (&mcnt, &p); | |
656 | extract_number_and_incr (&mcnt2, &p); | |
657 | printf ("/set_number_at/0/%d/0/%d", mcnt, mcnt2); | |
658 | break; | |
659 | ||
660 | case wordbound: | |
661 | printf ("/wordbound"); | |
662 | break; | |
663 | ||
664 | case notwordbound: | |
665 | printf ("/notwordbound"); | |
666 | break; | |
667 | ||
668 | case wordbeg: | |
669 | printf ("/wordbeg"); | |
670 | break; | |
671 | ||
672 | case wordend: | |
673 | printf ("/wordend"); | |
674 | ||
675 | #ifdef emacs | |
676 | case before_dot: | |
677 | printf ("/before_dot"); | |
678 | break; | |
679 | ||
680 | case at_dot: | |
681 | printf ("/at_dot"); | |
682 | break; | |
683 | ||
684 | case after_dot: | |
685 | printf ("/after_dot"); | |
686 | break; | |
687 | ||
688 | case syntaxspec: | |
689 | printf ("/syntaxspec"); | |
690 | mcnt = *p++; | |
691 | printf ("/%d", mcnt); | |
692 | break; | |
693 | ||
694 | case notsyntaxspec: | |
695 | printf ("/notsyntaxspec"); | |
696 | mcnt = *p++; | |
697 | printf ("/%d", mcnt); | |
698 | break; | |
699 | #endif /* emacs */ | |
700 | ||
701 | case wordchar: | |
702 | printf ("/wordchar"); | |
703 | break; | |
704 | ||
705 | case notwordchar: | |
706 | printf ("/notwordchar"); | |
707 | break; | |
708 | ||
709 | case begbuf: | |
710 | printf ("/begbuf"); | |
711 | break; | |
712 | ||
713 | case endbuf: | |
714 | printf ("/endbuf"); | |
715 | break; | |
716 | ||
717 | default: | |
718 | printf ("?%d", *(p-1)); | |
719 | } | |
720 | } | |
721 | printf ("/\n"); | |
722 | } | |
723 | ||
724 | ||
725 | void | |
726 | print_compiled_pattern (bufp) | |
727 | struct re_pattern_buffer *bufp; | |
728 | { | |
729 | unsigned char *buffer = bufp->buffer; | |
730 | ||
731 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
732 | printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); | |
733 | ||
734 | if (bufp->fastmap_accurate && bufp->fastmap) | |
735 | { | |
736 | printf ("fastmap: "); | |
737 | print_fastmap (bufp->fastmap); | |
738 | } | |
739 | ||
740 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
741 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
742 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
743 | printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
744 | printf ("no_sub: %d\t", bufp->no_sub); | |
745 | printf ("not_bol: %d\t", bufp->not_bol); | |
746 | printf ("not_eol: %d\t", bufp->not_eol); | |
747 | printf ("syntax: %d\n", bufp->syntax); | |
748 | /* Perhaps we should print the translate table? */ | |
749 | } | |
750 | ||
751 | ||
752 | void | |
753 | print_double_string (where, string1, size1, string2, size2) | |
754 | const char *where; | |
755 | const char *string1; | |
756 | const char *string2; | |
757 | int size1; | |
758 | int size2; | |
759 | { | |
760 | unsigned this_char; | |
761 | ||
762 | if (where == NULL) | |
763 | printf ("(null)"); | |
764 | else | |
765 | { | |
766 | if (FIRST_STRING_P (where)) | |
767 | { | |
768 | for (this_char = where - string1; this_char < size1; this_char++) | |
769 | printchar (string1[this_char]); | |
770 | ||
771 | where = string2; | |
772 | } | |
773 | ||
774 | for (this_char = where - string2; this_char < size2; this_char++) | |
775 | printchar (string2[this_char]); | |
776 | } | |
777 | } | |
778 | ||
779 | #else /* not DEBUG */ | |
780 | ||
781 | #undef assert | |
782 | #define assert(e) | |
783 | ||
784 | #define DEBUG_STATEMENT(e) | |
785 | #define DEBUG_PRINT1(x) | |
786 | #define DEBUG_PRINT2(x1, x2) | |
787 | #define DEBUG_PRINT3(x1, x2, x3) | |
9114e279 | 788 | #define DEBUG_PRINT4(x1, x2, x3, x4) |
bc78d348 KB |
789 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
790 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
791 | ||
792 | #endif /* not DEBUG */ | |
793 | \f | |
794 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
795 | also be assigned to arbitrarily: each pattern buffer stores its own | |
796 | syntax, so it can be changed between regex compilations. */ | |
797 | reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS; | |
798 | ||
799 | ||
800 | /* Specify the precise syntax of regexps for compilation. This provides | |
801 | for compatibility for various utilities which historically have | |
802 | different, incompatible syntaxes. | |
803 | ||
804 | The argument SYNTAX is a bit mask comprised of the various bits | |
805 | defined in regex.h. We return the old syntax. */ | |
806 | ||
807 | reg_syntax_t | |
808 | re_set_syntax (syntax) | |
809 | reg_syntax_t syntax; | |
810 | { | |
811 | reg_syntax_t ret = re_syntax_options; | |
812 | ||
813 | re_syntax_options = syntax; | |
814 | return ret; | |
815 | } | |
816 | \f | |
817 | /* This table gives an error message for each of the error codes listed | |
818 | in regex.h. Obviously the order here has to be same as there. */ | |
819 | ||
820 | static const char *re_error_msg[] = | |
821 | { NULL, /* REG_NOERROR */ | |
822 | "No match", /* REG_NOMATCH */ | |
823 | "Invalid regular expression", /* REG_BADPAT */ | |
824 | "Invalid collation character", /* REG_ECOLLATE */ | |
825 | "Invalid character class name", /* REG_ECTYPE */ | |
826 | "Trailing backslash", /* REG_EESCAPE */ | |
827 | "Invalid back reference", /* REG_ESUBREG */ | |
828 | "Unmatched [ or [^", /* REG_EBRACK */ | |
829 | "Unmatched ( or \\(", /* REG_EPAREN */ | |
830 | "Unmatched \\{", /* REG_EBRACE */ | |
831 | "Invalid content of \\{\\}", /* REG_BADBR */ | |
832 | "Invalid range end", /* REG_ERANGE */ | |
833 | "Memory exhausted", /* REG_ESPACE */ | |
834 | "Invalid preceding regular expression", /* REG_BADRPT */ | |
835 | "Premature end of regular expression", /* REG_EEND */ | |
836 | "Regular expression too big", /* REG_ESIZE */ | |
837 | "Unmatched ) or \\)", /* REG_ERPAREN */ | |
838 | }; | |
839 | \f | |
840 | /* Subroutine declarations and macros for regex_compile. */ | |
841 | ||
842 | static void store_op1 (), store_op2 (); | |
843 | static void insert_op1 (), insert_op2 (); | |
844 | static boolean at_begline_loc_p (), at_endline_loc_p (); | |
845 | static boolean group_in_compile_stack (); | |
846 | static reg_errcode_t compile_range (); | |
847 | ||
848 | /* Fetch the next character in the uncompiled pattern---translating it | |
849 | if necessary. Also cast from a signed character in the constant | |
850 | string passed to us by the user to an unsigned char that we can use | |
851 | as an array index (in, e.g., `translate'). */ | |
852 | #define PATFETCH(c) \ | |
853 | do {if (p == pend) return REG_EEND; \ | |
854 | c = (unsigned char) *p++; \ | |
855 | if (translate) c = translate[c]; \ | |
856 | } while (0) | |
857 | ||
858 | /* Fetch the next character in the uncompiled pattern, with no | |
859 | translation. */ | |
860 | #define PATFETCH_RAW(c) \ | |
861 | do {if (p == pend) return REG_EEND; \ | |
862 | c = (unsigned char) *p++; \ | |
863 | } while (0) | |
864 | ||
865 | /* Go backwards one character in the pattern. */ | |
866 | #define PATUNFETCH p-- | |
867 | ||
868 | ||
869 | /* If `translate' is non-null, return translate[D], else just D. We | |
870 | cast the subscript to translate because some data is declared as | |
871 | `char *', to avoid warnings when a string constant is passed. But | |
872 | when we use a character as a subscript we must make it unsigned. */ | |
873 | #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d)) | |
874 | ||
875 | ||
876 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
877 | ||
878 | /* If the buffer isn't allocated when it comes in, use this. */ | |
879 | #define INIT_BUF_SIZE 32 | |
880 | ||
881 | /* Make sure we have at least N more bytes of space in buffer. */ | |
882 | #define GET_BUFFER_SPACE(n) \ | |
883 | while (b - bufp->buffer + (n) > bufp->allocated) \ | |
884 | EXTEND_BUFFER () | |
885 | ||
886 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
887 | #define BUF_PUSH(c) \ | |
888 | do { \ | |
889 | GET_BUFFER_SPACE (1); \ | |
890 | *b++ = (unsigned char) (c); \ | |
891 | } while (0) | |
892 | ||
893 | ||
894 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
895 | #define BUF_PUSH_2(c1, c2) \ | |
896 | do { \ | |
897 | GET_BUFFER_SPACE (2); \ | |
898 | *b++ = (unsigned char) (c1); \ | |
899 | *b++ = (unsigned char) (c2); \ | |
900 | } while (0) | |
901 | ||
902 | ||
903 | /* As with BUF_PUSH_2, except for three bytes. */ | |
904 | #define BUF_PUSH_3(c1, c2, c3) \ | |
905 | do { \ | |
906 | GET_BUFFER_SPACE (3); \ | |
907 | *b++ = (unsigned char) (c1); \ | |
908 | *b++ = (unsigned char) (c2); \ | |
909 | *b++ = (unsigned char) (c3); \ | |
910 | } while (0) | |
911 | ||
912 | ||
913 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
914 | relative address offset by the three bytes the jump itself occupies. */ | |
915 | #define STORE_JUMP(op, loc, to) \ | |
916 | store_op1 (op, loc, (to) - (loc) - 3) | |
917 | ||
918 | /* Likewise, for a two-argument jump. */ | |
919 | #define STORE_JUMP2(op, loc, to, arg) \ | |
920 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
921 | ||
922 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
923 | #define INSERT_JUMP(op, loc, to) \ | |
924 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
925 | ||
926 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
927 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
928 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
929 | ||
930 | ||
931 | /* This is not an arbitrary limit: the arguments which represent offsets | |
932 | into the pattern are two bytes long. So if 2^16 bytes turns out to | |
933 | be too small, many things would have to change. */ | |
934 | #define MAX_BUF_SIZE (1L << 16) | |
935 | ||
936 | ||
937 | /* Extend the buffer by twice its current size via realloc and | |
938 | reset the pointers that pointed into the old block to point to the | |
939 | correct places in the new one. If extending the buffer results in it | |
940 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
941 | #define EXTEND_BUFFER() \ | |
942 | do { \ | |
943 | unsigned char *old_buffer = bufp->buffer; \ | |
944 | if (bufp->allocated == MAX_BUF_SIZE) \ | |
945 | return REG_ESIZE; \ | |
946 | bufp->allocated <<= 1; \ | |
947 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
948 | bufp->allocated = MAX_BUF_SIZE; \ | |
949 | bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ | |
950 | if (bufp->buffer == NULL) \ | |
951 | return REG_ESPACE; \ | |
952 | /* If the buffer moved, move all the pointers into it. */ \ | |
953 | if (old_buffer != bufp->buffer) \ | |
954 | { \ | |
955 | b = (b - old_buffer) + bufp->buffer; \ | |
956 | begalt = (begalt - old_buffer) + bufp->buffer; \ | |
957 | if (fixup_alt_jump) \ | |
958 | fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ | |
959 | if (laststart) \ | |
960 | laststart = (laststart - old_buffer) + bufp->buffer; \ | |
961 | if (pending_exact) \ | |
962 | pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
963 | } \ | |
964 | } while (0) | |
965 | ||
966 | ||
967 | /* Since we have one byte reserved for the register number argument to | |
968 | {start,stop}_memory, the maximum number of groups we can report | |
969 | things about is what fits in that byte. */ | |
970 | #define MAX_REGNUM 255 | |
971 | ||
972 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
973 | ignore the excess. */ | |
974 | typedef unsigned regnum_t; | |
975 | ||
976 | ||
977 | /* Macros for the compile stack. */ | |
978 | ||
979 | /* Since offsets can go either forwards or backwards, this type needs to | |
980 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
981 | typedef int pattern_offset_t; | |
982 | ||
983 | typedef struct | |
984 | { | |
985 | pattern_offset_t begalt_offset; | |
986 | pattern_offset_t fixup_alt_jump; | |
987 | pattern_offset_t inner_group_offset; | |
988 | pattern_offset_t laststart_offset; | |
989 | regnum_t regnum; | |
990 | } compile_stack_elt_t; | |
991 | ||
992 | ||
993 | typedef struct | |
994 | { | |
995 | compile_stack_elt_t *stack; | |
996 | unsigned size; | |
997 | unsigned avail; /* Offset of next open position. */ | |
998 | } compile_stack_type; | |
999 | ||
1000 | ||
1001 | #define INIT_COMPILE_STACK_SIZE 32 | |
1002 | ||
1003 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1004 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1005 | ||
1006 | /* The next available element. */ | |
1007 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
1008 | ||
1009 | ||
1010 | /* Set the bit for character C in a list. */ | |
1011 | #define SET_LIST_BIT(c) \ | |
1012 | (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
1013 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
1014 | ||
1015 | ||
1016 | /* Get the next unsigned number in the uncompiled pattern. */ | |
1017 | #define GET_UNSIGNED_NUMBER(num) \ | |
1018 | { if (p != pend) \ | |
1019 | { \ | |
1020 | PATFETCH (c); \ | |
c6b40788 | 1021 | while (ISDIGIT (c)) \ |
bc78d348 KB |
1022 | { \ |
1023 | if (num < 0) \ | |
1024 | num = 0; \ | |
1025 | num = num * 10 + c - '0'; \ | |
1026 | if (p == pend) \ | |
1027 | break; \ | |
1028 | PATFETCH (c); \ | |
1029 | } \ | |
1030 | } \ | |
1031 | } | |
1032 | ||
1033 | #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
1034 | ||
1035 | #define IS_CHAR_CLASS(string) \ | |
1036 | (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
1037 | || STREQ (string, "lower") || STREQ (string, "digit") \ | |
1038 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
1039 | || STREQ (string, "space") || STREQ (string, "print") \ | |
1040 | || STREQ (string, "punct") || STREQ (string, "graph") \ | |
1041 | || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
1042 | \f | |
1043 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
1044 | Returns one of error codes defined in `regex.h', or zero for success. | |
1045 | ||
1046 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
1047 | fields are set in BUFP on entry. | |
1048 | ||
1049 | If it succeeds, results are put in BUFP (if it returns an error, the | |
1050 | contents of BUFP are undefined): | |
1051 | `buffer' is the compiled pattern; | |
1052 | `syntax' is set to SYNTAX; | |
1053 | `used' is set to the length of the compiled pattern; | |
9114e279 KB |
1054 | `fastmap_accurate' is zero; |
1055 | `re_nsub' is the number of subexpressions in PATTERN; | |
1056 | `not_bol' and `not_eol' are zero; | |
bc78d348 KB |
1057 | |
1058 | The `fastmap' and `newline_anchor' fields are neither | |
1059 | examined nor set. */ | |
1060 | ||
1061 | static reg_errcode_t | |
1062 | regex_compile (pattern, size, syntax, bufp) | |
1063 | const char *pattern; | |
1064 | int size; | |
1065 | reg_syntax_t syntax; | |
1066 | struct re_pattern_buffer *bufp; | |
1067 | { | |
1068 | /* We fetch characters from PATTERN here. Even though PATTERN is | |
1069 | `char *' (i.e., signed), we declare these variables as unsigned, so | |
1070 | they can be reliably used as array indices. */ | |
1071 | register unsigned char c, c1; | |
1072 | ||
1073 | /* A random tempory spot in PATTERN. */ | |
1074 | const char *p1; | |
1075 | ||
1076 | /* Points to the end of the buffer, where we should append. */ | |
1077 | register unsigned char *b; | |
1078 | ||
1079 | /* Keeps track of unclosed groups. */ | |
1080 | compile_stack_type compile_stack; | |
1081 | ||
1082 | /* Points to the current (ending) position in the pattern. */ | |
1083 | const char *p = pattern; | |
1084 | const char *pend = pattern + size; | |
1085 | ||
1086 | /* How to translate the characters in the pattern. */ | |
1087 | char *translate = bufp->translate; | |
1088 | ||
1089 | /* Address of the count-byte of the most recently inserted `exactn' | |
1090 | command. This makes it possible to tell if a new exact-match | |
1091 | character can be added to that command or if the character requires | |
1092 | a new `exactn' command. */ | |
1093 | unsigned char *pending_exact = 0; | |
1094 | ||
1095 | /* Address of start of the most recently finished expression. | |
1096 | This tells, e.g., postfix * where to find the start of its | |
1097 | operand. Reset at the beginning of groups and alternatives. */ | |
1098 | unsigned char *laststart = 0; | |
1099 | ||
1100 | /* Address of beginning of regexp, or inside of last group. */ | |
1101 | unsigned char *begalt; | |
1102 | ||
1103 | /* Place in the uncompiled pattern (i.e., the {) to | |
1104 | which to go back if the interval is invalid. */ | |
1105 | const char *beg_interval; | |
1106 | ||
1107 | /* Address of the place where a forward jump should go to the end of | |
1108 | the containing expression. Each alternative of an `or' -- except the | |
1109 | last -- ends with a forward jump of this sort. */ | |
1110 | unsigned char *fixup_alt_jump = 0; | |
1111 | ||
1112 | /* Counts open-groups as they are encountered. Remembered for the | |
1113 | matching close-group on the compile stack, so the same register | |
1114 | number is put in the stop_memory as the start_memory. */ | |
1115 | regnum_t regnum = 0; | |
1116 | ||
1117 | #ifdef DEBUG | |
1118 | DEBUG_PRINT1 ("\nCompiling pattern: "); | |
1119 | if (debug) | |
1120 | { | |
1121 | unsigned debug_count; | |
1122 | ||
1123 | for (debug_count = 0; debug_count < size; debug_count++) | |
1124 | printchar (pattern[debug_count]); | |
1125 | putchar ('\n'); | |
1126 | } | |
1127 | #endif /* DEBUG */ | |
1128 | ||
1129 | /* Initialize the compile stack. */ | |
1130 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
1131 | if (compile_stack.stack == NULL) | |
1132 | return REG_ESPACE; | |
1133 | ||
1134 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
1135 | compile_stack.avail = 0; | |
1136 | ||
1137 | /* Initialize the pattern buffer. */ | |
1138 | bufp->syntax = syntax; | |
1139 | bufp->fastmap_accurate = 0; | |
1140 | bufp->not_bol = bufp->not_eol = 0; | |
1141 | ||
1142 | /* Set `used' to zero, so that if we return an error, the pattern | |
1143 | printer (for debugging) will think there's no pattern. We reset it | |
1144 | at the end. */ | |
1145 | bufp->used = 0; | |
1146 | ||
1147 | /* Always count groups, whether or not bufp->no_sub is set. */ | |
1148 | bufp->re_nsub = 0; | |
1149 | ||
1150 | #if !defined (emacs) && !defined (SYNTAX_TABLE) | |
1151 | /* Initialize the syntax table. */ | |
1152 | init_syntax_once (); | |
1153 | #endif | |
1154 | ||
1155 | if (bufp->allocated == 0) | |
1156 | { | |
1157 | if (bufp->buffer) | |
1158 | { /* If zero allocated, but buffer is non-null, try to realloc | |
1159 | enough space. This loses if buffer's address is bogus, but | |
1160 | that is the user's responsibility. */ | |
1161 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
1162 | } | |
1163 | else | |
1164 | { /* Caller did not allocate a buffer. Do it for them. */ | |
1165 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
1166 | } | |
1167 | if (!bufp->buffer) return REG_ESPACE; | |
1168 | ||
1169 | bufp->allocated = INIT_BUF_SIZE; | |
1170 | } | |
1171 | ||
1172 | begalt = b = bufp->buffer; | |
1173 | ||
1174 | /* Loop through the uncompiled pattern until we're at the end. */ | |
1175 | while (p != pend) | |
1176 | { | |
1177 | PATFETCH (c); | |
1178 | ||
1179 | switch (c) | |
1180 | { | |
1181 | case '^': | |
1182 | { | |
1183 | if ( /* If at start of pattern, it's an operator. */ | |
1184 | p == pattern + 1 | |
1185 | /* If context independent, it's an operator. */ | |
1186 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1187 | /* Otherwise, depends on what's come before. */ | |
1188 | || at_begline_loc_p (pattern, p, syntax)) | |
1189 | BUF_PUSH (begline); | |
1190 | else | |
1191 | goto normal_char; | |
1192 | } | |
1193 | break; | |
1194 | ||
1195 | ||
1196 | case '$': | |
1197 | { | |
1198 | if ( /* If at end of pattern, it's an operator. */ | |
1199 | p == pend | |
1200 | /* If context independent, it's an operator. */ | |
1201 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1202 | /* Otherwise, depends on what's next. */ | |
1203 | || at_endline_loc_p (p, pend, syntax)) | |
1204 | BUF_PUSH (endline); | |
1205 | else | |
1206 | goto normal_char; | |
1207 | } | |
1208 | break; | |
1209 | ||
1210 | ||
1211 | case '+': | |
1212 | case '?': | |
1213 | if ((syntax & RE_BK_PLUS_QM) | |
1214 | || (syntax & RE_LIMITED_OPS)) | |
1215 | goto normal_char; | |
1216 | handle_plus: | |
1217 | case '*': | |
1218 | /* If there is no previous pattern... */ | |
1219 | if (!laststart) | |
1220 | { | |
1221 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
1222 | return REG_BADRPT; | |
1223 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
1224 | goto normal_char; | |
1225 | } | |
1226 | ||
1227 | { | |
1228 | /* Are we optimizing this jump? */ | |
1229 | boolean keep_string_p = false; | |
1230 | ||
1231 | /* 1 means zero (many) matches is allowed. */ | |
1232 | char zero_times_ok = 0, many_times_ok = 0; | |
1233 | ||
1234 | /* If there is a sequence of repetition chars, collapse it | |
1235 | down to just one (the right one). We can't combine | |
1236 | interval operators with these because of, e.g., `a{2}*', | |
1237 | which should only match an even number of `a's. */ | |
1238 | ||
1239 | for (;;) | |
1240 | { | |
1241 | zero_times_ok |= c != '+'; | |
1242 | many_times_ok |= c != '?'; | |
1243 | ||
1244 | if (p == pend) | |
1245 | break; | |
1246 | ||
1247 | PATFETCH (c); | |
1248 | ||
1249 | if (c == '*' | |
1250 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
1251 | ; | |
1252 | ||
1253 | else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
1254 | { | |
1255 | if (p == pend) return REG_EESCAPE; | |
1256 | ||
1257 | PATFETCH (c1); | |
1258 | if (!(c1 == '+' || c1 == '?')) | |
1259 | { | |
1260 | PATUNFETCH; | |
1261 | PATUNFETCH; | |
1262 | break; | |
1263 | } | |
1264 | ||
1265 | c = c1; | |
1266 | } | |
1267 | else | |
1268 | { | |
1269 | PATUNFETCH; | |
1270 | break; | |
1271 | } | |
1272 | ||
1273 | /* If we get here, we found another repeat character. */ | |
1274 | } | |
1275 | ||
1276 | /* Star, etc. applied to an empty pattern is equivalent | |
1277 | to an empty pattern. */ | |
1278 | if (!laststart) | |
1279 | break; | |
1280 | ||
1281 | /* Now we know whether or not zero matches is allowed | |
1282 | and also whether or not two or more matches is allowed. */ | |
1283 | if (many_times_ok) | |
1284 | { /* More than one repetition is allowed, so put in at the | |
1285 | end a backward relative jump from `b' to before the next | |
1286 | jump we're going to put in below (which jumps from | |
1287 | laststart to after this jump). | |
1288 | ||
1289 | But if we are at the `*' in the exact sequence `.*\n', | |
1290 | insert an unconditional jump backwards to the ., | |
1291 | instead of the beginning of the loop. This way we only | |
1292 | push a failure point once, instead of every time | |
1293 | through the loop. */ | |
1294 | assert (p - 1 > pattern); | |
1295 | ||
1296 | /* Allocate the space for the jump. */ | |
1297 | GET_BUFFER_SPACE (3); | |
1298 | ||
1299 | /* We know we are not at the first character of the pattern, | |
1300 | because laststart was nonzero. And we've already | |
1301 | incremented `p', by the way, to be the character after | |
1302 | the `*'. Do we have to do something analogous here | |
1303 | for null bytes, because of RE_DOT_NOT_NULL? */ | |
1304 | if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
1305 | && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') | |
1306 | && !(syntax & RE_DOT_NEWLINE)) | |
1307 | { /* We have .*\n. */ | |
1308 | STORE_JUMP (jump, b, laststart); | |
1309 | keep_string_p = true; | |
1310 | } | |
1311 | else | |
1312 | /* Anything else. */ | |
1313 | STORE_JUMP (maybe_pop_jump, b, laststart - 3); | |
1314 | ||
1315 | /* We've added more stuff to the buffer. */ | |
1316 | b += 3; | |
1317 | } | |
1318 | ||
1319 | /* On failure, jump from laststart to b + 3, which will be the | |
1320 | end of the buffer after this jump is inserted. */ | |
1321 | GET_BUFFER_SPACE (3); | |
1322 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
1323 | : on_failure_jump, | |
1324 | laststart, b + 3); | |
1325 | pending_exact = 0; | |
1326 | b += 3; | |
1327 | ||
1328 | if (!zero_times_ok) | |
1329 | { | |
1330 | /* At least one repetition is required, so insert a | |
1331 | `dummy_failure_jump' before the initial | |
1332 | `on_failure_jump' instruction of the loop. This | |
1333 | effects a skip over that instruction the first time | |
1334 | we hit that loop. */ | |
1335 | GET_BUFFER_SPACE (3); | |
1336 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
1337 | b += 3; | |
1338 | } | |
1339 | } | |
1340 | break; | |
1341 | ||
1342 | ||
1343 | case '.': | |
1344 | laststart = b; | |
1345 | BUF_PUSH (anychar); | |
1346 | break; | |
1347 | ||
1348 | ||
1349 | case '[': | |
1350 | { | |
1351 | boolean had_char_class = false; | |
1352 | ||
1353 | if (p == pend) return REG_EBRACK; | |
1354 | ||
1355 | /* Ensure that we have enough space to push a charset: the | |
1356 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
1357 | GET_BUFFER_SPACE (34); | |
1358 | ||
1359 | laststart = b; | |
1360 | ||
1361 | /* We test `*p == '^' twice, instead of using an if | |
1362 | statement, so we only need one BUF_PUSH. */ | |
1363 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
1364 | if (*p == '^') | |
1365 | p++; | |
1366 | ||
1367 | /* Remember the first position in the bracket expression. */ | |
1368 | p1 = p; | |
1369 | ||
1370 | /* Push the number of bytes in the bitmap. */ | |
1371 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
1372 | ||
1373 | /* Clear the whole map. */ | |
1374 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
1375 | ||
1376 | /* charset_not matches newline according to a syntax bit. */ | |
1377 | if ((re_opcode_t) b[-2] == charset_not | |
1378 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
1379 | SET_LIST_BIT ('\n'); | |
1380 | ||
1381 | /* Read in characters and ranges, setting map bits. */ | |
1382 | for (;;) | |
1383 | { | |
1384 | if (p == pend) return REG_EBRACK; | |
1385 | ||
1386 | PATFETCH (c); | |
1387 | ||
1388 | /* \ might escape characters inside [...] and [^...]. */ | |
1389 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
1390 | { | |
1391 | if (p == pend) return REG_EESCAPE; | |
1392 | ||
1393 | PATFETCH (c1); | |
1394 | SET_LIST_BIT (c1); | |
1395 | continue; | |
1396 | } | |
1397 | ||
1398 | /* Could be the end of the bracket expression. If it's | |
1399 | not (i.e., when the bracket expression is `[]' so | |
1400 | far), the ']' character bit gets set way below. */ | |
1401 | if (c == ']' && p != p1 + 1) | |
1402 | break; | |
1403 | ||
1404 | /* Look ahead to see if it's a range when the last thing | |
1405 | was a character class. */ | |
1406 | if (had_char_class && c == '-' && *p != ']') | |
1407 | return REG_ERANGE; | |
1408 | ||
1409 | /* Look ahead to see if it's a range when the last thing | |
1410 | was a character: if this is a hyphen not at the | |
1411 | beginning or the end of a list, then it's the range | |
1412 | operator. */ | |
1413 | if (c == '-' | |
1414 | && !(p - 2 >= pattern && p[-2] == '[') | |
1415 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
1416 | && *p != ']') | |
1417 | { | |
1418 | reg_errcode_t ret | |
1419 | = compile_range (&p, pend, translate, syntax, b); | |
1420 | if (ret != REG_NOERROR) return ret; | |
1421 | } | |
1422 | ||
1423 | else if (p[0] == '-' && p[1] != ']') | |
1424 | { /* This handles ranges made up of characters only. */ | |
1425 | reg_errcode_t ret; | |
1426 | ||
1427 | /* Move past the `-'. */ | |
1428 | PATFETCH (c1); | |
1429 | ||
1430 | ret = compile_range (&p, pend, translate, syntax, b); | |
1431 | if (ret != REG_NOERROR) return ret; | |
1432 | } | |
1433 | ||
1434 | /* See if we're at the beginning of a possible character | |
1435 | class. */ | |
1436 | ||
1437 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
1438 | { /* Leave room for the null. */ | |
1439 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
1440 | ||
1441 | PATFETCH (c); | |
1442 | c1 = 0; | |
1443 | ||
1444 | /* If pattern is `[[:'. */ | |
1445 | if (p == pend) return REG_EBRACK; | |
1446 | ||
1447 | for (;;) | |
1448 | { | |
1449 | PATFETCH (c); | |
1450 | if (c == ':' || c == ']' || p == pend | |
1451 | || c1 == CHAR_CLASS_MAX_LENGTH) | |
1452 | break; | |
1453 | str[c1++] = c; | |
1454 | } | |
1455 | str[c1] = '\0'; | |
1456 | ||
1457 | /* If isn't a word bracketed by `[:' and:`]': | |
1458 | undo the ending character, the letters, and leave | |
1459 | the leading `:' and `[' (but set bits for them). */ | |
1460 | if (c == ':' && *p == ']') | |
1461 | { | |
1462 | int ch; | |
1463 | boolean is_alnum = STREQ (str, "alnum"); | |
1464 | boolean is_alpha = STREQ (str, "alpha"); | |
1465 | boolean is_blank = STREQ (str, "blank"); | |
1466 | boolean is_cntrl = STREQ (str, "cntrl"); | |
1467 | boolean is_digit = STREQ (str, "digit"); | |
1468 | boolean is_graph = STREQ (str, "graph"); | |
1469 | boolean is_lower = STREQ (str, "lower"); | |
1470 | boolean is_print = STREQ (str, "print"); | |
1471 | boolean is_punct = STREQ (str, "punct"); | |
1472 | boolean is_space = STREQ (str, "space"); | |
1473 | boolean is_upper = STREQ (str, "upper"); | |
1474 | boolean is_xdigit = STREQ (str, "xdigit"); | |
1475 | ||
1476 | if (!IS_CHAR_CLASS (str)) return REG_ECTYPE; | |
1477 | ||
1478 | /* Throw away the ] at the end of the character | |
1479 | class. */ | |
1480 | PATFETCH (c); | |
1481 | ||
1482 | if (p == pend) return REG_EBRACK; | |
1483 | ||
1484 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
1485 | { | |
c6b40788 JM |
1486 | if ( (is_alnum && ISALNUM (ch)) |
1487 | || (is_alpha && ISALPHA (ch)) | |
1488 | || (is_blank && ISBLANK (ch)) | |
1489 | || (is_cntrl && ISCNTRL (ch)) | |
1490 | || (is_digit && ISDIGIT (ch)) | |
1491 | || (is_graph && ISGRAPH (ch)) | |
1492 | || (is_lower && ISLOWER (ch)) | |
1493 | || (is_print && ISPRINT (ch)) | |
1494 | || (is_punct && ISPUNCT (ch)) | |
1495 | || (is_space && ISSPACE (ch)) | |
1496 | || (is_upper && ISUPPER (ch)) | |
1497 | || (is_xdigit && ISXDIGIT (ch))) | |
bc78d348 KB |
1498 | SET_LIST_BIT (ch); |
1499 | } | |
1500 | had_char_class = true; | |
1501 | } | |
1502 | else | |
1503 | { | |
1504 | c1++; | |
1505 | while (c1--) | |
1506 | PATUNFETCH; | |
1507 | SET_LIST_BIT ('['); | |
1508 | SET_LIST_BIT (':'); | |
1509 | had_char_class = false; | |
1510 | } | |
1511 | } | |
1512 | else | |
1513 | { | |
1514 | had_char_class = false; | |
1515 | SET_LIST_BIT (c); | |
1516 | } | |
1517 | } | |
1518 | ||
1519 | /* Discard any (non)matching list bytes that are all 0 at the | |
1520 | end of the map. Decrease the map-length byte too. */ | |
1521 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
1522 | b[-1]--; | |
1523 | b += b[-1]; | |
1524 | } | |
1525 | break; | |
1526 | ||
1527 | ||
1528 | case '(': | |
1529 | if (syntax & RE_NO_BK_PARENS) | |
1530 | goto handle_open; | |
1531 | else | |
1532 | goto normal_char; | |
1533 | ||
1534 | ||
1535 | case ')': | |
1536 | if (syntax & RE_NO_BK_PARENS) | |
1537 | goto handle_close; | |
1538 | else | |
1539 | goto normal_char; | |
1540 | ||
1541 | ||
1542 | case '\n': | |
1543 | if (syntax & RE_NEWLINE_ALT) | |
1544 | goto handle_alt; | |
1545 | else | |
1546 | goto normal_char; | |
1547 | ||
1548 | ||
1549 | case '|': | |
1550 | if (syntax & RE_NO_BK_VBAR) | |
1551 | goto handle_alt; | |
1552 | else | |
1553 | goto normal_char; | |
1554 | ||
1555 | ||
1556 | case '{': | |
1557 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
1558 | goto handle_interval; | |
1559 | else | |
1560 | goto normal_char; | |
1561 | ||
1562 | ||
1563 | case '\\': | |
1564 | if (p == pend) return REG_EESCAPE; | |
1565 | ||
1566 | /* Do not translate the character after the \, so that we can | |
1567 | distinguish, e.g., \B from \b, even if we normally would | |
1568 | translate, e.g., B to b. */ | |
1569 | PATFETCH_RAW (c); | |
1570 | ||
1571 | switch (c) | |
1572 | { | |
1573 | case '(': | |
1574 | if (syntax & RE_NO_BK_PARENS) | |
1575 | goto normal_backslash; | |
1576 | ||
1577 | handle_open: | |
1578 | bufp->re_nsub++; | |
1579 | regnum++; | |
1580 | ||
1581 | if (COMPILE_STACK_FULL) | |
1582 | { | |
1583 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
1584 | compile_stack_elt_t); | |
1585 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
1586 | ||
1587 | compile_stack.size <<= 1; | |
1588 | } | |
1589 | ||
1590 | /* These are the values to restore when we hit end of this | |
1591 | group. They are all relative offsets, so that if the | |
1592 | whole pattern moves because of realloc, they will still | |
1593 | be valid. */ | |
1594 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
1595 | COMPILE_STACK_TOP.fixup_alt_jump | |
1596 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
1597 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
1598 | COMPILE_STACK_TOP.regnum = regnum; | |
1599 | ||
1600 | /* We will eventually replace the 0 with the number of | |
1601 | groups inner to this one. But do not push a | |
1602 | start_memory for groups beyond the last one we can | |
1603 | represent in the compiled pattern. */ | |
1604 | if (regnum <= MAX_REGNUM) | |
1605 | { | |
1606 | COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; | |
1607 | BUF_PUSH_3 (start_memory, regnum, 0); | |
1608 | } | |
1609 | ||
1610 | compile_stack.avail++; | |
1611 | ||
1612 | fixup_alt_jump = 0; | |
1613 | laststart = 0; | |
1614 | begalt = b; | |
1615 | break; | |
1616 | ||
1617 | ||
1618 | case ')': | |
1619 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
1620 | ||
1621 | if (COMPILE_STACK_EMPTY) | |
1622 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
1623 | goto normal_backslash; | |
1624 | else | |
1625 | return REG_ERPAREN; | |
1626 | ||
1627 | handle_close: | |
1628 | if (fixup_alt_jump) | |
1629 | { /* Push a dummy failure point at the end of the | |
1630 | alternative for a possible future | |
1631 | `pop_failure_jump' to pop. See comments at | |
1632 | `push_dummy_failure' in `re_match_2'. */ | |
1633 | BUF_PUSH (push_dummy_failure); | |
1634 | ||
1635 | /* We allocated space for this jump when we assigned | |
1636 | to `fixup_alt_jump', in the `handle_alt' case below. */ | |
1637 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
1638 | } | |
1639 | ||
1640 | /* See similar code for backslashed left paren above. */ | |
1641 | if (COMPILE_STACK_EMPTY) | |
1642 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
1643 | goto normal_char; | |
1644 | else | |
1645 | return REG_ERPAREN; | |
1646 | ||
1647 | /* Since we just checked for an empty stack above, this | |
1648 | ``can't happen''. */ | |
1649 | assert (compile_stack.avail != 0); | |
1650 | { | |
1651 | /* We don't just want to restore into `regnum', because | |
1652 | later groups should continue to be numbered higher, | |
1653 | as in `(ab)c(de)' -- the second group is #2. */ | |
1654 | regnum_t this_group_regnum; | |
1655 | ||
1656 | compile_stack.avail--; | |
1657 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
1658 | fixup_alt_jump | |
1659 | = COMPILE_STACK_TOP.fixup_alt_jump | |
1660 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
1661 | : 0; | |
1662 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
1663 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
1664 | ||
1665 | /* We're at the end of the group, so now we know how many | |
1666 | groups were inside this one. */ | |
1667 | if (this_group_regnum <= MAX_REGNUM) | |
1668 | { | |
1669 | unsigned char *inner_group_loc | |
1670 | = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
1671 | ||
1672 | *inner_group_loc = regnum - this_group_regnum; | |
1673 | BUF_PUSH_3 (stop_memory, this_group_regnum, | |
1674 | regnum - this_group_regnum); | |
1675 | } | |
1676 | } | |
1677 | break; | |
1678 | ||
1679 | ||
1680 | case '|': /* `\|'. */ | |
1681 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
1682 | goto normal_backslash; | |
1683 | handle_alt: | |
1684 | if (syntax & RE_LIMITED_OPS) | |
1685 | goto normal_char; | |
1686 | ||
1687 | /* Insert before the previous alternative a jump which | |
1688 | jumps to this alternative if the former fails. */ | |
1689 | GET_BUFFER_SPACE (3); | |
1690 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
1691 | pending_exact = 0; | |
1692 | b += 3; | |
1693 | ||
1694 | /* The alternative before this one has a jump after it | |
1695 | which gets executed if it gets matched. Adjust that | |
1696 | jump so it will jump to this alternative's analogous | |
1697 | jump (put in below, which in turn will jump to the next | |
1698 | (if any) alternative's such jump, etc.). The last such | |
1699 | jump jumps to the correct final destination. A picture: | |
1700 | _____ _____ | |
1701 | | | | | | |
1702 | | v | v | |
1703 | a | b | c | |
1704 | ||
9114e279 KB |
1705 | If we are at `b', then fixup_alt_jump right now points to a |
1706 | three-byte space after `a'. We'll put in the jump, set | |
1707 | fixup_alt_jump to right after `b', and leave behind three | |
1708 | bytes which we'll fill in when we get to after `c'. */ | |
bc78d348 KB |
1709 | |
1710 | if (fixup_alt_jump) | |
1711 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
1712 | ||
1713 | /* Mark and leave space for a jump after this alternative, | |
1714 | to be filled in later either by next alternative or | |
1715 | when know we're at the end of a series of alternatives. */ | |
1716 | fixup_alt_jump = b; | |
1717 | GET_BUFFER_SPACE (3); | |
1718 | b += 3; | |
1719 | ||
1720 | laststart = 0; | |
1721 | begalt = b; | |
1722 | break; | |
1723 | ||
1724 | ||
1725 | case '{': | |
1726 | /* If \{ is a literal. */ | |
1727 | if (!(syntax & RE_INTERVALS) | |
1728 | /* If we're at `\{' and it's not the open-interval | |
1729 | operator. */ | |
1730 | || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
1731 | || (p - 2 == pattern && p == pend)) | |
1732 | goto normal_backslash; | |
1733 | ||
1734 | handle_interval: | |
1735 | { | |
1736 | /* If got here, then the syntax allows intervals. */ | |
1737 | ||
1738 | /* At least (most) this many matches must be made. */ | |
1739 | int lower_bound = -1, upper_bound = -1; | |
1740 | ||
1741 | beg_interval = p - 1; | |
1742 | ||
1743 | if (p == pend) | |
1744 | { | |
1745 | if (syntax & RE_NO_BK_BRACES) | |
1746 | goto unfetch_interval; | |
1747 | else | |
1748 | return REG_EBRACE; | |
1749 | } | |
1750 | ||
1751 | GET_UNSIGNED_NUMBER (lower_bound); | |
1752 | ||
1753 | if (c == ',') | |
1754 | { | |
1755 | GET_UNSIGNED_NUMBER (upper_bound); | |
1756 | if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
1757 | } | |
1758 | else | |
1759 | /* Interval such as `{1}' => match exactly once. */ | |
1760 | upper_bound = lower_bound; | |
1761 | ||
1762 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
1763 | || lower_bound > upper_bound) | |
1764 | { | |
1765 | if (syntax & RE_NO_BK_BRACES) | |
1766 | goto unfetch_interval; | |
1767 | else | |
1768 | return REG_BADBR; | |
1769 | } | |
1770 | ||
1771 | if (!(syntax & RE_NO_BK_BRACES)) | |
1772 | { | |
1773 | if (c != '\\') return REG_EBRACE; | |
1774 | ||
1775 | PATFETCH (c); | |
1776 | } | |
1777 | ||
1778 | if (c != '}') | |
1779 | { | |
1780 | if (syntax & RE_NO_BK_BRACES) | |
1781 | goto unfetch_interval; | |
1782 | else | |
1783 | return REG_BADBR; | |
1784 | } | |
1785 | ||
1786 | /* We just parsed a valid interval. */ | |
1787 | ||
1788 | /* If it's invalid to have no preceding re. */ | |
1789 | if (!laststart) | |
1790 | { | |
1791 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
1792 | return REG_BADRPT; | |
1793 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
1794 | laststart = b; | |
1795 | else | |
1796 | goto unfetch_interval; | |
1797 | } | |
1798 | ||
1799 | /* If the upper bound is zero, don't want to succeed at | |
1800 | all; jump from `laststart' to `b + 3', which will be | |
1801 | the end of the buffer after we insert the jump. */ | |
1802 | if (upper_bound == 0) | |
1803 | { | |
1804 | GET_BUFFER_SPACE (3); | |
1805 | INSERT_JUMP (jump, laststart, b + 3); | |
1806 | b += 3; | |
1807 | } | |
1808 | ||
1809 | /* Otherwise, we have a nontrivial interval. When | |
1810 | we're all done, the pattern will look like: | |
1811 | set_number_at <jump count> <upper bound> | |
1812 | set_number_at <succeed_n count> <lower bound> | |
1813 | succeed_n <after jump addr> <succed_n count> | |
1814 | <body of loop> | |
1815 | jump_n <succeed_n addr> <jump count> | |
1816 | (The upper bound and `jump_n' are omitted if | |
1817 | `upper_bound' is 1, though.) */ | |
1818 | else | |
1819 | { /* If the upper bound is > 1, we need to insert | |
1820 | more at the end of the loop. */ | |
1821 | unsigned nbytes = 10 + (upper_bound > 1) * 10; | |
1822 | ||
1823 | GET_BUFFER_SPACE (nbytes); | |
1824 | ||
1825 | /* Initialize lower bound of the `succeed_n', even | |
1826 | though it will be set during matching by its | |
1827 | attendant `set_number_at' (inserted next), | |
1828 | because `re_compile_fastmap' needs to know. | |
1829 | Jump to the `jump_n' we might insert below. */ | |
1830 | INSERT_JUMP2 (succeed_n, laststart, | |
1831 | b + 5 + (upper_bound > 1) * 5, | |
1832 | lower_bound); | |
1833 | b += 5; | |
1834 | ||
1835 | /* Code to initialize the lower bound. Insert | |
1836 | before the `succeed_n'. The `5' is the last two | |
1837 | bytes of this `set_number_at', plus 3 bytes of | |
1838 | the following `succeed_n'. */ | |
1839 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
1840 | b += 5; | |
1841 | ||
1842 | if (upper_bound > 1) | |
1843 | { /* More than one repetition is allowed, so | |
1844 | append a backward jump to the `succeed_n' | |
1845 | that starts this interval. | |
1846 | ||
1847 | When we've reached this during matching, | |
1848 | we'll have matched the interval once, so | |
1849 | jump back only `upper_bound - 1' times. */ | |
1850 | STORE_JUMP2 (jump_n, b, laststart + 5, | |
1851 | upper_bound - 1); | |
1852 | b += 5; | |
1853 | ||
1854 | /* The location we want to set is the second | |
1855 | parameter of the `jump_n'; that is `b-2' as | |
1856 | an absolute address. `laststart' will be | |
1857 | the `set_number_at' we're about to insert; | |
1858 | `laststart+3' the number to set, the source | |
1859 | for the relative address. But we are | |
1860 | inserting into the middle of the pattern -- | |
1861 | so everything is getting moved up by 5. | |
1862 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
1863 | i.e., b - laststart. | |
1864 | ||
1865 | We insert this at the beginning of the loop | |
1866 | so that if we fail during matching, we'll | |
1867 | reinitialize the bounds. */ | |
1868 | insert_op2 (set_number_at, laststart, b - laststart, | |
1869 | upper_bound - 1, b); | |
1870 | b += 5; | |
1871 | } | |
1872 | } | |
1873 | pending_exact = 0; | |
1874 | beg_interval = NULL; | |
1875 | } | |
1876 | break; | |
1877 | ||
1878 | unfetch_interval: | |
1879 | /* If an invalid interval, match the characters as literals. */ | |
1880 | assert (beg_interval); | |
1881 | p = beg_interval; | |
1882 | beg_interval = NULL; | |
1883 | ||
1884 | /* normal_char and normal_backslash need `c'. */ | |
1885 | PATFETCH (c); | |
1886 | ||
1887 | if (!(syntax & RE_NO_BK_BRACES)) | |
1888 | { | |
1889 | if (p > pattern && p[-1] == '\\') | |
1890 | goto normal_backslash; | |
1891 | } | |
1892 | goto normal_char; | |
1893 | ||
1894 | #ifdef emacs | |
1895 | /* There is no way to specify the before_dot and after_dot | |
1896 | operators. rms says this is ok. --karl */ | |
1897 | case '=': | |
1898 | BUF_PUSH (at_dot); | |
1899 | break; | |
1900 | ||
1901 | case 's': | |
1902 | laststart = b; | |
1903 | PATFETCH (c); | |
1904 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
1905 | break; | |
1906 | ||
1907 | case 'S': | |
1908 | laststart = b; | |
1909 | PATFETCH (c); | |
1910 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
1911 | break; | |
1912 | #endif /* emacs */ | |
1913 | ||
1914 | ||
1915 | case 'w': | |
1916 | laststart = b; | |
1917 | BUF_PUSH (wordchar); | |
1918 | break; | |
1919 | ||
1920 | ||
1921 | case 'W': | |
1922 | laststart = b; | |
1923 | BUF_PUSH (notwordchar); | |
1924 | break; | |
1925 | ||
1926 | ||
1927 | case '<': | |
1928 | BUF_PUSH (wordbeg); | |
1929 | break; | |
1930 | ||
1931 | case '>': | |
1932 | BUF_PUSH (wordend); | |
1933 | break; | |
1934 | ||
1935 | case 'b': | |
1936 | BUF_PUSH (wordbound); | |
1937 | break; | |
1938 | ||
1939 | case 'B': | |
1940 | BUF_PUSH (notwordbound); | |
1941 | break; | |
1942 | ||
1943 | case '`': | |
1944 | BUF_PUSH (begbuf); | |
1945 | break; | |
1946 | ||
1947 | case '\'': | |
1948 | BUF_PUSH (endbuf); | |
1949 | break; | |
1950 | ||
1951 | case '1': case '2': case '3': case '4': case '5': | |
1952 | case '6': case '7': case '8': case '9': | |
1953 | if (syntax & RE_NO_BK_REFS) | |
1954 | goto normal_char; | |
1955 | ||
1956 | c1 = c - '0'; | |
1957 | ||
1958 | if (c1 > regnum) | |
1959 | return REG_ESUBREG; | |
1960 | ||
1961 | /* Can't back reference to a subexpression if inside of it. */ | |
1962 | if (group_in_compile_stack (compile_stack, c1)) | |
1963 | goto normal_char; | |
1964 | ||
1965 | laststart = b; | |
1966 | BUF_PUSH_2 (duplicate, c1); | |
1967 | break; | |
1968 | ||
1969 | ||
1970 | case '+': | |
1971 | case '?': | |
1972 | if (syntax & RE_BK_PLUS_QM) | |
1973 | goto handle_plus; | |
1974 | else | |
1975 | goto normal_backslash; | |
1976 | ||
1977 | default: | |
1978 | normal_backslash: | |
1979 | /* You might think it would be useful for \ to mean | |
1980 | not to translate; but if we don't translate it | |
1981 | it will never match anything. */ | |
1982 | c = TRANSLATE (c); | |
1983 | goto normal_char; | |
1984 | } | |
1985 | break; | |
1986 | ||
1987 | ||
1988 | default: | |
1989 | /* Expects the character in `c'. */ | |
1990 | normal_char: | |
1991 | /* If no exactn currently being built. */ | |
1992 | if (!pending_exact | |
1993 | ||
1994 | /* If last exactn not at current position. */ | |
1995 | || pending_exact + *pending_exact + 1 != b | |
1996 | ||
1997 | /* We have only one byte following the exactn for the count. */ | |
1998 | || *pending_exact == (1 << BYTEWIDTH) - 1 | |
1999 | ||
2000 | /* If followed by a repetition operator. */ | |
2001 | || *p == '*' || *p == '^' | |
2002 | || ((syntax & RE_BK_PLUS_QM) | |
2003 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
2004 | : (*p == '+' || *p == '?')) | |
2005 | || ((syntax & RE_INTERVALS) | |
2006 | && ((syntax & RE_NO_BK_BRACES) | |
2007 | ? *p == '{' | |
2008 | : (p[0] == '\\' && p[1] == '{')))) | |
2009 | { | |
2010 | /* Start building a new exactn. */ | |
2011 | ||
2012 | laststart = b; | |
2013 | ||
2014 | BUF_PUSH_2 (exactn, 0); | |
2015 | pending_exact = b - 1; | |
2016 | } | |
2017 | ||
2018 | BUF_PUSH (c); | |
2019 | (*pending_exact)++; | |
2020 | break; | |
2021 | } /* switch (c) */ | |
2022 | } /* while p != pend */ | |
2023 | ||
2024 | ||
2025 | /* Through the pattern now. */ | |
2026 | ||
2027 | if (fixup_alt_jump) | |
2028 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2029 | ||
2030 | if (!COMPILE_STACK_EMPTY) | |
2031 | return REG_EPAREN; | |
2032 | ||
2033 | free (compile_stack.stack); | |
2034 | ||
2035 | /* We have succeeded; set the length of the buffer. */ | |
2036 | bufp->used = b - bufp->buffer; | |
2037 | ||
2038 | #ifdef DEBUG | |
2039 | if (debug) | |
2040 | { | |
2041 | DEBUG_PRINT1 ("\nCompiled pattern: "); | |
2042 | print_compiled_pattern (bufp); | |
2043 | } | |
2044 | #endif /* DEBUG */ | |
2045 | ||
2046 | return REG_NOERROR; | |
2047 | } /* regex_compile */ | |
2048 | \f | |
2049 | /* Subroutines for `regex_compile'. */ | |
2050 | ||
2051 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
2052 | ||
2053 | static void | |
2054 | store_op1 (op, loc, arg) | |
2055 | re_opcode_t op; | |
2056 | unsigned char *loc; | |
2057 | int arg; | |
2058 | { | |
2059 | *loc = (unsigned char) op; | |
2060 | STORE_NUMBER (loc + 1, arg); | |
2061 | } | |
2062 | ||
2063 | ||
2064 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2065 | ||
2066 | static void | |
2067 | store_op2 (op, loc, arg1, arg2) | |
2068 | re_opcode_t op; | |
2069 | unsigned char *loc; | |
2070 | int arg1, arg2; | |
2071 | { | |
2072 | *loc = (unsigned char) op; | |
2073 | STORE_NUMBER (loc + 1, arg1); | |
2074 | STORE_NUMBER (loc + 3, arg2); | |
2075 | } | |
2076 | ||
2077 | ||
2078 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
2079 | for OP followed by two-byte integer parameter ARG. */ | |
2080 | ||
2081 | static void | |
2082 | insert_op1 (op, loc, arg, end) | |
2083 | re_opcode_t op; | |
2084 | unsigned char *loc; | |
2085 | int arg; | |
2086 | unsigned char *end; | |
2087 | { | |
2088 | register unsigned char *pfrom = end; | |
2089 | register unsigned char *pto = end + 3; | |
2090 | ||
2091 | while (pfrom != loc) | |
2092 | *--pto = *--pfrom; | |
2093 | ||
2094 | store_op1 (op, loc, arg); | |
2095 | } | |
2096 | ||
2097 | ||
2098 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2099 | ||
2100 | static void | |
2101 | insert_op2 (op, loc, arg1, arg2, end) | |
2102 | re_opcode_t op; | |
2103 | unsigned char *loc; | |
2104 | int arg1, arg2; | |
2105 | unsigned char *end; | |
2106 | { | |
2107 | register unsigned char *pfrom = end; | |
2108 | register unsigned char *pto = end + 5; | |
2109 | ||
2110 | while (pfrom != loc) | |
2111 | *--pto = *--pfrom; | |
2112 | ||
2113 | store_op2 (op, loc, arg1, arg2); | |
2114 | } | |
2115 | ||
2116 | ||
2117 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
2118 | after an alternative or a begin-subexpression. We assume there is at | |
2119 | least one character before the ^. */ | |
2120 | ||
2121 | static boolean | |
2122 | at_begline_loc_p (pattern, p, syntax) | |
2123 | const char *pattern, *p; | |
2124 | reg_syntax_t syntax; | |
2125 | { | |
2126 | const char *prev = p - 2; | |
2127 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
2128 | ||
2129 | return | |
2130 | /* After a subexpression? */ | |
2131 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
2132 | /* After an alternative? */ | |
2133 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
2134 | } | |
2135 | ||
2136 | ||
2137 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
2138 | at least one character after the $, i.e., `P < PEND'. */ | |
2139 | ||
2140 | static boolean | |
2141 | at_endline_loc_p (p, pend, syntax) | |
2142 | const char *p, *pend; | |
2143 | int syntax; | |
2144 | { | |
2145 | const char *next = p; | |
2146 | boolean next_backslash = *next == '\\'; | |
2147 | const char *next_next = p + 1 < pend ? p + 1 : NULL; | |
2148 | ||
2149 | return | |
2150 | /* Before a subexpression? */ | |
2151 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
2152 | : next_backslash && next_next && *next_next == ')') | |
2153 | /* Before an alternative? */ | |
2154 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
2155 | : next_backslash && next_next && *next_next == '|'); | |
2156 | } | |
2157 | ||
2158 | ||
2159 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and | |
2160 | false if it's not. */ | |
2161 | ||
2162 | static boolean | |
2163 | group_in_compile_stack (compile_stack, regnum) | |
2164 | compile_stack_type compile_stack; | |
2165 | regnum_t regnum; | |
2166 | { | |
2167 | int this_element; | |
2168 | ||
2169 | for (this_element = compile_stack.avail - 1; | |
2170 | this_element >= 0; | |
2171 | this_element--) | |
2172 | if (compile_stack.stack[this_element].regnum == regnum) | |
2173 | return true; | |
2174 | ||
2175 | return false; | |
2176 | } | |
2177 | ||
2178 | ||
2179 | /* Read the ending character of a range (in a bracket expression) from the | |
2180 | uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
2181 | starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
2182 | Then we set the translation of all bits between the starting and | |
2183 | ending characters (inclusive) in the compiled pattern B. | |
2184 | ||
2185 | Return an error code. | |
2186 | ||
2187 | We use these short variable names so we can use the same macros as | |
2188 | `regex_compile' itself. */ | |
2189 | ||
2190 | static reg_errcode_t | |
2191 | compile_range (p_ptr, pend, translate, syntax, b) | |
2192 | const char **p_ptr, *pend; | |
2193 | char *translate; | |
2194 | reg_syntax_t syntax; | |
2195 | unsigned char *b; | |
2196 | { | |
2197 | unsigned this_char; | |
2198 | ||
2199 | const char *p = *p_ptr; | |
4af46de6 | 2200 | int range_start, range_end; |
bc78d348 | 2201 | |
bc78d348 KB |
2202 | if (p == pend) |
2203 | return REG_ERANGE; | |
2204 | ||
4af46de6 JB |
2205 | /* Even though the pattern is a signed `char *', we need to fetch |
2206 | with unsigned char *'s; if the high bit of the pattern character | |
2207 | is set, the range endpoints will be negative if we fetch using a | |
2208 | signed char *. | |
2209 | ||
2210 | We also want to fetch the endpoints without translating them; the | |
2211 | appropriate translation is done in the bit-setting loop below. */ | |
2212 | range_start = ((unsigned char *) p)[-2]; | |
2213 | range_end = ((unsigned char *) p)[0]; | |
bc78d348 KB |
2214 | |
2215 | /* Have to increment the pointer into the pattern string, so the | |
2216 | caller isn't still at the ending character. */ | |
2217 | (*p_ptr)++; | |
2218 | ||
2219 | /* If the start is after the end, the range is empty. */ | |
2220 | if (range_start > range_end) | |
2221 | return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
2222 | ||
2223 | /* Here we see why `this_char' has to be larger than an `unsigned | |
2224 | char' -- the range is inclusive, so if `range_end' == 0xff | |
2225 | (assuming 8-bit characters), we would otherwise go into an infinite | |
2226 | loop, since all characters <= 0xff. */ | |
2227 | for (this_char = range_start; this_char <= range_end; this_char++) | |
2228 | { | |
2229 | SET_LIST_BIT (TRANSLATE (this_char)); | |
2230 | } | |
2231 | ||
2232 | return REG_NOERROR; | |
2233 | } | |
2234 | \f | |
2235 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
2236 | re_match_2 use a failure stack. These have to be macros because of | |
2237 | REGEX_ALLOCATE. */ | |
2238 | ||
2239 | ||
2240 | /* Number of failure points for which to initially allocate space | |
2241 | when matching. If this number is exceeded, we allocate more | |
2242 | space, so it is not a hard limit. */ | |
2243 | #ifndef INIT_FAILURE_ALLOC | |
2244 | #define INIT_FAILURE_ALLOC 5 | |
2245 | #endif | |
2246 | ||
2247 | /* Roughly the maximum number of failure points on the stack. Would be | |
2248 | exactly that if always used MAX_FAILURE_SPACE each time we failed. | |
2249 | This is a variable only so users of regex can assign to it; we never | |
2250 | change it ourselves. */ | |
2251 | int re_max_failures = 2000; | |
2252 | ||
2253 | typedef const unsigned char *fail_stack_elt_t; | |
2254 | ||
2255 | typedef struct | |
2256 | { | |
2257 | fail_stack_elt_t *stack; | |
2258 | unsigned size; | |
2259 | unsigned avail; /* Offset of next open position. */ | |
2260 | } fail_stack_type; | |
2261 | ||
2262 | #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
2263 | #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
2264 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
2265 | #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail]) | |
2266 | ||
2267 | ||
2268 | /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */ | |
2269 | ||
2270 | #define INIT_FAIL_STACK() \ | |
2271 | do { \ | |
2272 | fail_stack.stack = (fail_stack_elt_t *) \ | |
2273 | REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ | |
2274 | \ | |
2275 | if (fail_stack.stack == NULL) \ | |
2276 | return -2; \ | |
2277 | \ | |
2278 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
2279 | fail_stack.avail = 0; \ | |
2280 | } while (0) | |
2281 | ||
2282 | ||
2283 | /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
2284 | ||
2285 | Return 1 if succeeds, and 0 if either ran out of memory | |
2286 | allocating space for it or it was already too large. | |
2287 | ||
2288 | REGEX_REALLOCATE requires `destination' be declared. */ | |
2289 | ||
2290 | #define DOUBLE_FAIL_STACK(fail_stack) \ | |
2291 | ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ | |
2292 | ? 0 \ | |
2293 | : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
2294 | REGEX_REALLOCATE ((fail_stack).stack, \ | |
2295 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
2296 | ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
2297 | \ | |
2298 | (fail_stack).stack == NULL \ | |
2299 | ? 0 \ | |
2300 | : ((fail_stack).size <<= 1, \ | |
2301 | 1))) | |
2302 | ||
2303 | ||
2304 | /* Push PATTERN_OP on FAIL_STACK. | |
2305 | ||
2306 | Return 1 if was able to do so and 0 if ran out of memory allocating | |
2307 | space to do so. */ | |
2308 | #define PUSH_PATTERN_OP(pattern_op, fail_stack) \ | |
2309 | ((FAIL_STACK_FULL () \ | |
2310 | && !DOUBLE_FAIL_STACK (fail_stack)) \ | |
2311 | ? 0 \ | |
2312 | : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \ | |
2313 | 1)) | |
2314 | ||
2315 | /* This pushes an item onto the failure stack. Must be a four-byte | |
2316 | value. Assumes the variable `fail_stack'. Probably should only | |
2317 | be called from within `PUSH_FAILURE_POINT'. */ | |
2318 | #define PUSH_FAILURE_ITEM(item) \ | |
2319 | fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item | |
2320 | ||
2321 | /* The complement operation. Assumes `fail_stack' is nonempty. */ | |
2322 | #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail] | |
2323 | ||
2324 | /* Used to omit pushing failure point id's when we're not debugging. */ | |
2325 | #ifdef DEBUG | |
2326 | #define DEBUG_PUSH PUSH_FAILURE_ITEM | |
2327 | #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM () | |
2328 | #else | |
2329 | #define DEBUG_PUSH(item) | |
2330 | #define DEBUG_POP(item_addr) | |
2331 | #endif | |
2332 | ||
2333 | ||
2334 | /* Push the information about the state we will need | |
2335 | if we ever fail back to it. | |
2336 | ||
2337 | Requires variables fail_stack, regstart, regend, reg_info, and | |
2338 | num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be | |
2339 | declared. | |
2340 | ||
2341 | Does `return FAILURE_CODE' if runs out of memory. */ | |
2342 | ||
2343 | #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
2344 | do { \ | |
2345 | char *destination; \ | |
2346 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
2347 | of 0 + -1 isn't done as unsigned. */ \ | |
2348 | int this_reg; \ | |
2349 | \ | |
2350 | DEBUG_STATEMENT (failure_id++); \ | |
9114e279 | 2351 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
bc78d348 KB |
2352 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ |
2353 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
2354 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
2355 | \ | |
2356 | DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ | |
2357 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ | |
2358 | \ | |
2359 | /* Ensure we have enough space allocated for what we will push. */ \ | |
2360 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
2361 | { \ | |
2362 | if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
2363 | return failure_code; \ | |
2364 | \ | |
2365 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
2366 | (fail_stack).size); \ | |
2367 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
2368 | } \ | |
2369 | \ | |
2370 | /* Push the info, starting with the registers. */ \ | |
2371 | DEBUG_PRINT1 ("\n"); \ | |
2372 | \ | |
2373 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ | |
2374 | this_reg++) \ | |
2375 | { \ | |
2376 | DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ | |
2377 | DEBUG_STATEMENT (num_regs_pushed++); \ | |
2378 | \ | |
2379 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
2380 | PUSH_FAILURE_ITEM (regstart[this_reg]); \ | |
2381 | \ | |
2382 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
2383 | PUSH_FAILURE_ITEM (regend[this_reg]); \ | |
2384 | \ | |
2385 | DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ | |
2386 | DEBUG_PRINT2 (" match_null=%d", \ | |
2387 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
2388 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
2389 | DEBUG_PRINT2 (" matched_something=%d", \ | |
2390 | MATCHED_SOMETHING (reg_info[this_reg])); \ | |
2391 | DEBUG_PRINT2 (" ever_matched=%d", \ | |
2392 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
2393 | DEBUG_PRINT1 ("\n"); \ | |
2394 | PUSH_FAILURE_ITEM (reg_info[this_reg].word); \ | |
2395 | } \ | |
2396 | \ | |
2397 | DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ | |
2398 | PUSH_FAILURE_ITEM (lowest_active_reg); \ | |
2399 | \ | |
2400 | DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ | |
2401 | PUSH_FAILURE_ITEM (highest_active_reg); \ | |
2402 | \ | |
2403 | DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ | |
2404 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
2405 | PUSH_FAILURE_ITEM (pattern_place); \ | |
2406 | \ | |
2407 | DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ | |
2408 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
2409 | size2); \ | |
2410 | DEBUG_PRINT1 ("'\n"); \ | |
2411 | PUSH_FAILURE_ITEM (string_place); \ | |
2412 | \ | |
2413 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
2414 | DEBUG_PUSH (failure_id); \ | |
2415 | } while (0) | |
2416 | ||
2417 | /* This is the number of items that are pushed and popped on the stack | |
2418 | for each register. */ | |
2419 | #define NUM_REG_ITEMS 3 | |
2420 | ||
2421 | /* Individual items aside from the registers. */ | |
2422 | #ifdef DEBUG | |
2423 | #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
2424 | #else | |
2425 | #define NUM_NONREG_ITEMS 4 | |
2426 | #endif | |
2427 | ||
2428 | /* We push at most this many items on the stack. */ | |
2429 | #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
2430 | ||
2431 | /* We actually push this many items. */ | |
2432 | #define NUM_FAILURE_ITEMS \ | |
2433 | ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \ | |
2434 | + NUM_NONREG_ITEMS) | |
2435 | ||
2436 | /* How many items can still be added to the stack without overflowing it. */ | |
2437 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
2438 | ||
2439 | ||
2440 | /* Pops what PUSH_FAIL_STACK pushes. | |
2441 | ||
2442 | We restore into the parameters, all of which should be lvalues: | |
2443 | STR -- the saved data position. | |
2444 | PAT -- the saved pattern position. | |
2445 | LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
2446 | REGSTART, REGEND -- arrays of string positions. | |
2447 | REG_INFO -- array of information about each subexpression. | |
2448 | ||
2449 | Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
2450 | `pend', `string1', `size1', `string2', and `size2'. */ | |
2451 | ||
2452 | #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
2453 | { \ | |
2454 | DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ | |
2455 | int this_reg; \ | |
2456 | const unsigned char *string_temp; \ | |
2457 | \ | |
2458 | assert (!FAIL_STACK_EMPTY ()); \ | |
2459 | \ | |
2460 | /* Remove failure points and point to how many regs pushed. */ \ | |
2461 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
2462 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
2463 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
2464 | \ | |
2465 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
2466 | \ | |
2467 | DEBUG_POP (&failure_id); \ | |
2468 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
2469 | \ | |
2470 | /* If the saved string location is NULL, it came from an \ | |
2471 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
2472 | saved NULL, thus retaining our current position in the string. */ \ | |
2473 | string_temp = POP_FAILURE_ITEM (); \ | |
2474 | if (string_temp != NULL) \ | |
2475 | str = (const char *) string_temp; \ | |
2476 | \ | |
2477 | DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ | |
2478 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
2479 | DEBUG_PRINT1 ("'\n"); \ | |
2480 | \ | |
2481 | pat = (unsigned char *) POP_FAILURE_ITEM (); \ | |
2482 | DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ | |
2483 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
2484 | \ | |
2485 | /* Restore register info. */ \ | |
2486 | high_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
2487 | DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ | |
2488 | \ | |
2489 | low_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
2490 | DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ | |
2491 | \ | |
2492 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
2493 | { \ | |
2494 | DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ | |
2495 | \ | |
2496 | reg_info[this_reg].word = POP_FAILURE_ITEM (); \ | |
2497 | DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ | |
2498 | \ | |
2499 | regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
2500 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
2501 | \ | |
2502 | regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
2503 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
2504 | } \ | |
9114e279 KB |
2505 | \ |
2506 | DEBUG_STATEMENT (nfailure_points_popped++); \ | |
bc78d348 KB |
2507 | } /* POP_FAILURE_POINT */ |
2508 | \f | |
2509 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
2510 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
2511 | characters can start a string that matches the pattern. This fastmap | |
2512 | is used by re_search to skip quickly over impossible starting points. | |
2513 | ||
2514 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
2515 | area as BUFP->fastmap. | |
2516 | ||
2517 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
2518 | the pattern buffer. | |
2519 | ||
2520 | Returns 0 if we succeed, -2 if an internal error. */ | |
2521 | ||
2522 | int | |
2523 | re_compile_fastmap (bufp) | |
2524 | struct re_pattern_buffer *bufp; | |
2525 | { | |
2526 | int j, k; | |
2527 | fail_stack_type fail_stack; | |
2528 | #ifndef REGEX_MALLOC | |
2529 | char *destination; | |
2530 | #endif | |
2531 | /* We don't push any register information onto the failure stack. */ | |
2532 | unsigned num_regs = 0; | |
2533 | ||
2534 | register char *fastmap = bufp->fastmap; | |
2535 | unsigned char *pattern = bufp->buffer; | |
2536 | unsigned long size = bufp->used; | |
2537 | const unsigned char *p = pattern; | |
2538 | register unsigned char *pend = pattern + size; | |
2539 | ||
2540 | /* Assume that each path through the pattern can be null until | |
2541 | proven otherwise. We set this false at the bottom of switch | |
2542 | statement, to which we get only if a particular path doesn't | |
2543 | match the empty string. */ | |
2544 | boolean path_can_be_null = true; | |
2545 | ||
2546 | /* We aren't doing a `succeed_n' to begin with. */ | |
2547 | boolean succeed_n_p = false; | |
2548 | ||
2549 | assert (fastmap != NULL && p != NULL); | |
2550 | ||
2551 | INIT_FAIL_STACK (); | |
2552 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
2553 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
2554 | bufp->can_be_null = 0; | |
2555 | ||
2556 | while (p != pend || !FAIL_STACK_EMPTY ()) | |
2557 | { | |
2558 | if (p == pend) | |
2559 | { | |
2560 | bufp->can_be_null |= path_can_be_null; | |
2561 | ||
2562 | /* Reset for next path. */ | |
2563 | path_can_be_null = true; | |
2564 | ||
2565 | p = fail_stack.stack[--fail_stack.avail]; | |
2566 | } | |
2567 | ||
2568 | /* We should never be about to go beyond the end of the pattern. */ | |
2569 | assert (p < pend); | |
2570 | ||
2571 | #ifdef SWITCH_ENUM_BUG | |
2572 | switch ((int) ((re_opcode_t) *p++)) | |
2573 | #else | |
2574 | switch ((re_opcode_t) *p++) | |
2575 | #endif | |
2576 | { | |
2577 | ||
2578 | /* I guess the idea here is to simply not bother with a fastmap | |
2579 | if a backreference is used, since it's too hard to figure out | |
2580 | the fastmap for the corresponding group. Setting | |
2581 | `can_be_null' stops `re_search_2' from using the fastmap, so | |
2582 | that is all we do. */ | |
2583 | case duplicate: | |
2584 | bufp->can_be_null = 1; | |
2585 | return 0; | |
2586 | ||
2587 | ||
2588 | /* Following are the cases which match a character. These end | |
2589 | with `break'. */ | |
2590 | ||
2591 | case exactn: | |
2592 | fastmap[p[1]] = 1; | |
2593 | break; | |
2594 | ||
2595 | ||
2596 | case charset: | |
2597 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2598 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
2599 | fastmap[j] = 1; | |
2600 | break; | |
2601 | ||
2602 | ||
2603 | case charset_not: | |
2604 | /* Chars beyond end of map must be allowed. */ | |
2605 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
2606 | fastmap[j] = 1; | |
2607 | ||
2608 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2609 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
2610 | fastmap[j] = 1; | |
2611 | break; | |
2612 | ||
2613 | ||
2614 | case wordchar: | |
2615 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2616 | if (SYNTAX (j) == Sword) | |
2617 | fastmap[j] = 1; | |
2618 | break; | |
2619 | ||
2620 | ||
2621 | case notwordchar: | |
2622 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2623 | if (SYNTAX (j) != Sword) | |
2624 | fastmap[j] = 1; | |
2625 | break; | |
2626 | ||
2627 | ||
2628 | case anychar: | |
2629 | /* `.' matches anything ... */ | |
2630 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2631 | fastmap[j] = 1; | |
2632 | ||
2633 | /* ... except perhaps newline. */ | |
2634 | if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
2635 | fastmap['\n'] = 0; | |
2636 | ||
2637 | /* Return if we have already set `can_be_null'; if we have, | |
2638 | then the fastmap is irrelevant. Something's wrong here. */ | |
2639 | else if (bufp->can_be_null) | |
2640 | return 0; | |
2641 | ||
2642 | /* Otherwise, have to check alternative paths. */ | |
2643 | break; | |
2644 | ||
2645 | ||
2646 | #ifdef emacs | |
2647 | case syntaxspec: | |
2648 | k = *p++; | |
2649 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2650 | if (SYNTAX (j) == (enum syntaxcode) k) | |
2651 | fastmap[j] = 1; | |
2652 | break; | |
2653 | ||
2654 | ||
2655 | case notsyntaxspec: | |
2656 | k = *p++; | |
2657 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2658 | if (SYNTAX (j) != (enum syntaxcode) k) | |
2659 | fastmap[j] = 1; | |
2660 | break; | |
2661 | ||
2662 | ||
2663 | /* All cases after this match the empty string. These end with | |
2664 | `continue'. */ | |
2665 | ||
2666 | ||
2667 | case before_dot: | |
2668 | case at_dot: | |
2669 | case after_dot: | |
2670 | continue; | |
2671 | #endif /* not emacs */ | |
2672 | ||
2673 | ||
2674 | case no_op: | |
2675 | case begline: | |
2676 | case endline: | |
2677 | case begbuf: | |
2678 | case endbuf: | |
2679 | case wordbound: | |
2680 | case notwordbound: | |
2681 | case wordbeg: | |
2682 | case wordend: | |
2683 | case push_dummy_failure: | |
2684 | continue; | |
2685 | ||
2686 | ||
2687 | case jump_n: | |
2688 | case pop_failure_jump: | |
2689 | case maybe_pop_jump: | |
2690 | case jump: | |
2691 | case jump_past_alt: | |
2692 | case dummy_failure_jump: | |
2693 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2694 | p += j; | |
2695 | if (j > 0) | |
2696 | continue; | |
2697 | ||
2698 | /* Jump backward implies we just went through the body of a | |
2699 | loop and matched nothing. Opcode jumped to should be | |
2700 | `on_failure_jump' or `succeed_n'. Just treat it like an | |
2701 | ordinary jump. For a * loop, it has pushed its failure | |
2702 | point already; if so, discard that as redundant. */ | |
2703 | if ((re_opcode_t) *p != on_failure_jump | |
2704 | && (re_opcode_t) *p != succeed_n) | |
2705 | continue; | |
2706 | ||
2707 | p++; | |
2708 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2709 | p += j; | |
2710 | ||
2711 | /* If what's on the stack is where we are now, pop it. */ | |
2712 | if (!FAIL_STACK_EMPTY () | |
2713 | && fail_stack.stack[fail_stack.avail - 1] == p) | |
2714 | fail_stack.avail--; | |
2715 | ||
2716 | continue; | |
2717 | ||
2718 | ||
2719 | case on_failure_jump: | |
2720 | case on_failure_keep_string_jump: | |
2721 | handle_on_failure_jump: | |
2722 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2723 | ||
2724 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
2725 | end of the pattern. We don't want to push such a point, | |
2726 | since when we restore it above, entering the switch will | |
2727 | increment `p' past the end of the pattern. We don't need | |
2728 | to push such a point since we obviously won't find any more | |
2729 | fastmap entries beyond `pend'. Such a pattern can match | |
2730 | the null string, though. */ | |
2731 | if (p + j < pend) | |
2732 | { | |
2733 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
2734 | return -2; | |
2735 | } | |
2736 | else | |
2737 | bufp->can_be_null = 1; | |
2738 | ||
2739 | if (succeed_n_p) | |
2740 | { | |
2741 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
2742 | succeed_n_p = false; | |
2743 | } | |
2744 | ||
2745 | continue; | |
2746 | ||
2747 | ||
2748 | case succeed_n: | |
2749 | /* Get to the number of times to succeed. */ | |
2750 | p += 2; | |
2751 | ||
2752 | /* Increment p past the n for when k != 0. */ | |
2753 | EXTRACT_NUMBER_AND_INCR (k, p); | |
2754 | if (k == 0) | |
2755 | { | |
2756 | p -= 4; | |
2757 | succeed_n_p = true; /* Spaghetti code alert. */ | |
2758 | goto handle_on_failure_jump; | |
2759 | } | |
2760 | continue; | |
2761 | ||
2762 | ||
2763 | case set_number_at: | |
2764 | p += 4; | |
2765 | continue; | |
2766 | ||
2767 | ||
2768 | case start_memory: | |
2769 | case stop_memory: | |
2770 | p += 2; | |
2771 | continue; | |
2772 | ||
2773 | ||
2774 | default: | |
2775 | abort (); /* We have listed all the cases. */ | |
2776 | } /* switch *p++ */ | |
2777 | ||
2778 | /* Getting here means we have found the possible starting | |
2779 | characters for one path of the pattern -- and that the empty | |
2780 | string does not match. We need not follow this path further. | |
2781 | Instead, look at the next alternative (remembered on the | |
2782 | stack), or quit if no more. The test at the top of the loop | |
2783 | does these things. */ | |
2784 | path_can_be_null = false; | |
2785 | p = pend; | |
2786 | } /* while p */ | |
2787 | ||
2788 | /* Set `can_be_null' for the last path (also the first path, if the | |
2789 | pattern is empty). */ | |
2790 | bufp->can_be_null |= path_can_be_null; | |
2791 | return 0; | |
2792 | } /* re_compile_fastmap */ | |
2793 | \f | |
2794 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
2795 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
2796 | this memory for recording register information. STARTS and ENDS | |
2797 | must be allocated using the malloc library routine, and must each | |
2798 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
2799 | ||
2800 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
2801 | register data. | |
2802 | ||
2803 | Unless this function is called, the first search or match using | |
2804 | PATTERN_BUFFER will allocate its own register data, without | |
2805 | freeing the old data. */ | |
2806 | ||
2807 | void | |
2808 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
2809 | struct re_pattern_buffer *bufp; | |
2810 | struct re_registers *regs; | |
2811 | unsigned num_regs; | |
2812 | regoff_t *starts, *ends; | |
2813 | { | |
2814 | if (num_regs) | |
2815 | { | |
2816 | bufp->regs_allocated = REGS_REALLOCATE; | |
2817 | regs->num_regs = num_regs; | |
2818 | regs->start = starts; | |
2819 | regs->end = ends; | |
2820 | } | |
2821 | else | |
2822 | { | |
2823 | bufp->regs_allocated = REGS_UNALLOCATED; | |
2824 | regs->num_regs = 0; | |
2825 | regs->start = regs->end = (regoff_t) 0; | |
2826 | } | |
2827 | } | |
2828 | \f | |
2829 | /* Searching routines. */ | |
2830 | ||
2831 | /* Like re_search_2, below, but only one string is specified, and | |
2832 | doesn't let you say where to stop matching. */ | |
2833 | ||
2834 | int | |
2835 | re_search (bufp, string, size, startpos, range, regs) | |
2836 | struct re_pattern_buffer *bufp; | |
2837 | const char *string; | |
2838 | int size, startpos, range; | |
2839 | struct re_registers *regs; | |
2840 | { | |
2841 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
2842 | regs, size); | |
2843 | } | |
2844 | ||
2845 | ||
2846 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
2847 | virtual concatenation of STRING1 and STRING2, starting first at index | |
2848 | STARTPOS, then at STARTPOS + 1, and so on. | |
2849 | ||
2850 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
2851 | ||
2852 | RANGE is how far to scan while trying to match. RANGE = 0 means try | |
2853 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
2854 | RANGE. | |
2855 | ||
2856 | In REGS, return the indices of the virtual concatenation of STRING1 | |
2857 | and STRING2 that matched the entire BUFP->buffer and its contained | |
2858 | subexpressions. | |
2859 | ||
2860 | Do not consider matching one past the index STOP in the virtual | |
2861 | concatenation of STRING1 and STRING2. | |
2862 | ||
2863 | We return either the position in the strings at which the match was | |
2864 | found, -1 if no match, or -2 if error (such as failure | |
2865 | stack overflow). */ | |
2866 | ||
2867 | int | |
2868 | re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
2869 | struct re_pattern_buffer *bufp; | |
2870 | const char *string1, *string2; | |
2871 | int size1, size2; | |
2872 | int startpos; | |
2873 | int range; | |
2874 | struct re_registers *regs; | |
2875 | int stop; | |
2876 | { | |
2877 | int val; | |
2878 | register char *fastmap = bufp->fastmap; | |
2879 | register char *translate = bufp->translate; | |
2880 | int total_size = size1 + size2; | |
2881 | int endpos = startpos + range; | |
2882 | ||
2883 | /* Check for out-of-range STARTPOS. */ | |
2884 | if (startpos < 0 || startpos > total_size) | |
2885 | return -1; | |
2886 | ||
2887 | /* Fix up RANGE if it might eventually take us outside | |
2888 | the virtual concatenation of STRING1 and STRING2. */ | |
2889 | if (endpos < -1) | |
2890 | range = -1 - startpos; | |
2891 | else if (endpos > total_size) | |
2892 | range = total_size - startpos; | |
2893 | ||
bc78d348 | 2894 | /* If the search isn't to be a backwards one, don't waste time in a |
9114e279 KB |
2895 | search for a pattern that must be anchored. */ |
2896 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) | |
bc78d348 KB |
2897 | { |
2898 | if (startpos > 0) | |
2899 | return -1; | |
2900 | else | |
2901 | range = 1; | |
2902 | } | |
2903 | ||
9114e279 KB |
2904 | /* Update the fastmap now if not correct already. */ |
2905 | if (fastmap && !bufp->fastmap_accurate) | |
2906 | if (re_compile_fastmap (bufp) == -2) | |
2907 | return -2; | |
2908 | ||
2909 | /* Loop through the string, looking for a place to start matching. */ | |
bc78d348 KB |
2910 | for (;;) |
2911 | { | |
2912 | /* If a fastmap is supplied, skip quickly over characters that | |
2913 | cannot be the start of a match. If the pattern can match the | |
2914 | null string, however, we don't need to skip characters; we want | |
2915 | the first null string. */ | |
2916 | if (fastmap && startpos < total_size && !bufp->can_be_null) | |
2917 | { | |
2918 | if (range > 0) /* Searching forwards. */ | |
2919 | { | |
2920 | register const char *d; | |
2921 | register int lim = 0; | |
2922 | int irange = range; | |
2923 | ||
2924 | if (startpos < size1 && startpos + range >= size1) | |
2925 | lim = range - (size1 - startpos); | |
2926 | ||
2927 | d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
2928 | ||
2929 | /* Written out as an if-else to avoid testing `translate' | |
2930 | inside the loop. */ | |
2931 | if (translate) | |
2932 | while (range > lim | |
2933 | && !fastmap[(unsigned char) translate[*d++]]) | |
2934 | range--; | |
2935 | else | |
2936 | while (range > lim && !fastmap[(unsigned char) *d++]) | |
2937 | range--; | |
2938 | ||
2939 | startpos += irange - range; | |
2940 | } | |
2941 | else /* Searching backwards. */ | |
2942 | { | |
2943 | register char c = (size1 == 0 || startpos >= size1 | |
2944 | ? string2[startpos - size1] | |
2945 | : string1[startpos]); | |
2946 | ||
9114e279 | 2947 | if (!fastmap[(unsigned char) TRANSLATE (c)]) |
bc78d348 KB |
2948 | goto advance; |
2949 | } | |
2950 | } | |
2951 | ||
2952 | /* If can't match the null string, and that's all we have left, fail. */ | |
2953 | if (range >= 0 && startpos == total_size && fastmap | |
2954 | && !bufp->can_be_null) | |
2955 | return -1; | |
2956 | ||
2957 | val = re_match_2 (bufp, string1, size1, string2, size2, | |
2958 | startpos, regs, stop); | |
2959 | if (val >= 0) | |
2960 | return startpos; | |
2961 | ||
2962 | if (val == -2) | |
2963 | return -2; | |
2964 | ||
2965 | advance: | |
2966 | if (!range) | |
2967 | break; | |
2968 | else if (range > 0) | |
2969 | { | |
2970 | range--; | |
2971 | startpos++; | |
2972 | } | |
2973 | else | |
2974 | { | |
2975 | range++; | |
2976 | startpos--; | |
2977 | } | |
2978 | } | |
2979 | return -1; | |
2980 | } /* re_search_2 */ | |
2981 | \f | |
2982 | /* Declarations and macros for re_match_2. */ | |
2983 | ||
2984 | static int bcmp_translate (); | |
2985 | static boolean alt_match_null_string_p (), | |
2986 | common_op_match_null_string_p (), | |
2987 | group_match_null_string_p (); | |
2988 | ||
2989 | /* Structure for per-register (a.k.a. per-group) information. | |
2990 | This must not be longer than one word, because we push this value | |
2991 | onto the failure stack. Other register information, such as the | |
2992 | starting and ending positions (which are addresses), and the list of | |
2993 | inner groups (which is a bits list) are maintained in separate | |
2994 | variables. | |
2995 | ||
2996 | We are making a (strictly speaking) nonportable assumption here: that | |
2997 | the compiler will pack our bit fields into something that fits into | |
2998 | the type of `word', i.e., is something that fits into one item on the | |
2999 | failure stack. */ | |
3000 | typedef union | |
3001 | { | |
3002 | fail_stack_elt_t word; | |
3003 | struct | |
3004 | { | |
3005 | /* This field is one if this group can match the empty string, | |
3006 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
3007 | #define MATCH_NULL_UNSET_VALUE 3 | |
3008 | unsigned match_null_string_p : 2; | |
3009 | unsigned is_active : 1; | |
3010 | unsigned matched_something : 1; | |
3011 | unsigned ever_matched_something : 1; | |
3012 | } bits; | |
3013 | } register_info_type; | |
3014 | ||
3015 | #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
3016 | #define IS_ACTIVE(R) ((R).bits.is_active) | |
3017 | #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
3018 | #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
3019 | ||
3020 | ||
9114e279 KB |
3021 | /* Call this when have matched a real character; it sets `matched' flags |
3022 | for the subexpressions which we are currently inside. Also records | |
3023 | that those subexprs have matched. */ | |
bc78d348 KB |
3024 | #define SET_REGS_MATCHED() \ |
3025 | do \ | |
3026 | { \ | |
3027 | unsigned r; \ | |
3028 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
3029 | { \ | |
3030 | MATCHED_SOMETHING (reg_info[r]) \ | |
3031 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
3032 | = 1; \ | |
3033 | } \ | |
3034 | } \ | |
3035 | while (0) | |
3036 | ||
3037 | ||
3038 | /* This converts PTR, a pointer into one of the search strings `string1' | |
3039 | and `string2' into an offset from the beginning of that string. */ | |
3040 | #define POINTER_TO_OFFSET(ptr) \ | |
3041 | (FIRST_STRING_P (ptr) ? (ptr) - string1 : (ptr) - string2 + size1) | |
3042 | ||
3043 | /* Registers are set to a sentinel when they haven't yet matched. */ | |
3044 | #define REG_UNSET_VALUE ((char *) -1) | |
3045 | #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
3046 | ||
3047 | ||
3048 | /* Macros for dealing with the split strings in re_match_2. */ | |
3049 | ||
3050 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
3051 | ||
3052 | /* Call before fetching a character with *d. This switches over to | |
3053 | string2 if necessary. */ | |
3054 | #define PREFETCH() \ | |
3055 | while (d == dend) \ | |
3056 | { \ | |
3057 | /* End of string2 => fail. */ \ | |
3058 | if (dend == end_match_2) \ | |
3059 | goto fail; \ | |
3060 | /* End of string1 => advance to string2. */ \ | |
3061 | d = string2; \ | |
3062 | dend = end_match_2; \ | |
3063 | } | |
3064 | ||
3065 | ||
3066 | /* Test if at very beginning or at very end of the virtual concatenation | |
3067 | of `string1' and `string2'. If only one string, it's `string2'. */ | |
9114e279 KB |
3068 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
3069 | #define AT_STRINGS_END(d) ((d) == end2) | |
bc78d348 KB |
3070 | |
3071 | ||
3072 | /* Test if D points to a character which is word-constituent. We have | |
3073 | two special cases to check for: if past the end of string1, look at | |
3074 | the first character in string2; and if before the beginning of | |
9114e279 KB |
3075 | string2, look at the last character in string1. */ |
3076 | #define WORDCHAR_P(d) \ | |
bc78d348 | 3077 | (SYNTAX ((d) == end1 ? *string2 \ |
9114e279 KB |
3078 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
3079 | == Sword) | |
bc78d348 KB |
3080 | |
3081 | /* Test if the character before D and the one at D differ with respect | |
3082 | to being word-constituent. */ | |
3083 | #define AT_WORD_BOUNDARY(d) \ | |
9114e279 KB |
3084 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
3085 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
bc78d348 KB |
3086 | |
3087 | ||
3088 | /* Free everything we malloc. */ | |
3089 | #ifdef REGEX_MALLOC | |
3090 | #define FREE_VAR(var) if (var) free (var); var = NULL | |
3091 | #define FREE_VARIABLES() \ | |
3092 | do { \ | |
3093 | FREE_VAR (fail_stack.stack); \ | |
3094 | FREE_VAR (regstart); \ | |
3095 | FREE_VAR (regend); \ | |
3096 | FREE_VAR (old_regstart); \ | |
3097 | FREE_VAR (old_regend); \ | |
3098 | FREE_VAR (best_regstart); \ | |
3099 | FREE_VAR (best_regend); \ | |
3100 | FREE_VAR (reg_info); \ | |
3101 | FREE_VAR (reg_dummy); \ | |
3102 | FREE_VAR (reg_info_dummy); \ | |
3103 | } while (0) | |
3104 | #else /* not REGEX_MALLOC */ | |
3105 | /* Some MIPS systems (at least) want this to free alloca'd storage. */ | |
3106 | #define FREE_VARIABLES() alloca (0) | |
3107 | #endif /* not REGEX_MALLOC */ | |
3108 | ||
3109 | ||
3110 | /* These values must meet several constraints. They must not be valid | |
3111 | register values; since we have a limit of 255 registers (because | |
3112 | we use only one byte in the pattern for the register number), we can | |
3113 | use numbers larger than 255. They must differ by 1, because of | |
3114 | NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
3115 | be larger than the value for the highest register, so we do not try | |
3116 | to actually save any registers when none are active. */ | |
3117 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
3118 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
3119 | \f | |
3120 | /* Matching routines. */ | |
3121 | ||
3122 | #ifndef emacs /* Emacs never uses this. */ | |
3123 | /* re_match is like re_match_2 except it takes only a single string. */ | |
3124 | ||
3125 | int | |
3126 | re_match (bufp, string, size, pos, regs) | |
3127 | struct re_pattern_buffer *bufp; | |
3128 | const char *string; | |
3129 | int size, pos; | |
3130 | struct re_registers *regs; | |
3131 | { | |
3132 | return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size); | |
3133 | } | |
3134 | #endif /* not emacs */ | |
3135 | ||
3136 | ||
3137 | /* re_match_2 matches the compiled pattern in BUFP against the | |
3138 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
3139 | and SIZE2, respectively). We start matching at POS, and stop | |
3140 | matching at STOP. | |
3141 | ||
3142 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
3143 | store offsets for the substring each group matched in REGS. See the | |
3144 | documentation for exactly how many groups we fill. | |
3145 | ||
3146 | We return -1 if no match, -2 if an internal error (such as the | |
3147 | failure stack overflowing). Otherwise, we return the length of the | |
3148 | matched substring. */ | |
3149 | ||
3150 | int | |
3151 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3152 | struct re_pattern_buffer *bufp; | |
3153 | const char *string1, *string2; | |
3154 | int size1, size2; | |
3155 | int pos; | |
3156 | struct re_registers *regs; | |
3157 | int stop; | |
3158 | { | |
3159 | /* General temporaries. */ | |
3160 | int mcnt; | |
3161 | unsigned char *p1; | |
3162 | ||
3163 | /* Just past the end of the corresponding string. */ | |
3164 | const char *end1, *end2; | |
3165 | ||
3166 | /* Pointers into string1 and string2, just past the last characters in | |
3167 | each to consider matching. */ | |
3168 | const char *end_match_1, *end_match_2; | |
3169 | ||
3170 | /* Where we are in the data, and the end of the current string. */ | |
3171 | const char *d, *dend; | |
3172 | ||
3173 | /* Where we are in the pattern, and the end of the pattern. */ | |
3174 | unsigned char *p = bufp->buffer; | |
3175 | register unsigned char *pend = p + bufp->used; | |
3176 | ||
3177 | /* We use this to map every character in the string. */ | |
3178 | char *translate = bufp->translate; | |
3179 | ||
3180 | /* Failure point stack. Each place that can handle a failure further | |
3181 | down the line pushes a failure point on this stack. It consists of | |
3182 | restart, regend, and reg_info for all registers corresponding to | |
3183 | the subexpressions we're currently inside, plus the number of such | |
3184 | registers, and, finally, two char *'s. The first char * is where | |
3185 | to resume scanning the pattern; the second one is where to resume | |
3186 | scanning the strings. If the latter is zero, the failure point is | |
3187 | a ``dummy''; if a failure happens and the failure point is a dummy, | |
3188 | it gets discarded and the next next one is tried. */ | |
3189 | fail_stack_type fail_stack; | |
3190 | #ifdef DEBUG | |
3191 | static unsigned failure_id = 0; | |
9114e279 | 3192 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
bc78d348 KB |
3193 | #endif |
3194 | ||
3195 | /* We fill all the registers internally, independent of what we | |
3196 | return, for use in backreferences. The number here includes | |
3197 | an element for register zero. */ | |
3198 | unsigned num_regs = bufp->re_nsub + 1; | |
3199 | ||
3200 | /* The currently active registers. */ | |
3201 | unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3202 | unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3203 | ||
3204 | /* Information on the contents of registers. These are pointers into | |
3205 | the input strings; they record just what was matched (on this | |
3206 | attempt) by a subexpression part of the pattern, that is, the | |
3207 | regnum-th regstart pointer points to where in the pattern we began | |
3208 | matching and the regnum-th regend points to right after where we | |
3209 | stopped matching the regnum-th subexpression. (The zeroth register | |
3210 | keeps track of what the whole pattern matches.) */ | |
3211 | const char **regstart, **regend; | |
3212 | ||
3213 | /* If a group that's operated upon by a repetition operator fails to | |
3214 | match anything, then the register for its start will need to be | |
3215 | restored because it will have been set to wherever in the string we | |
3216 | are when we last see its open-group operator. Similarly for a | |
3217 | register's end. */ | |
3218 | const char **old_regstart, **old_regend; | |
3219 | ||
3220 | /* The is_active field of reg_info helps us keep track of which (possibly | |
3221 | nested) subexpressions we are currently in. The matched_something | |
3222 | field of reg_info[reg_num] helps us tell whether or not we have | |
3223 | matched any of the pattern so far this time through the reg_num-th | |
3224 | subexpression. These two fields get reset each time through any | |
3225 | loop their register is in. */ | |
3226 | register_info_type *reg_info; | |
3227 | ||
3228 | /* The following record the register info as found in the above | |
3229 | variables when we find a match better than any we've seen before. | |
3230 | This happens as we backtrack through the failure points, which in | |
3231 | turn happens only if we have not yet matched the entire string. */ | |
3232 | unsigned best_regs_set = false; | |
3233 | const char **best_regstart, **best_regend; | |
3234 | ||
3235 | /* Logically, this is `best_regend[0]'. But we don't want to have to | |
3236 | allocate space for that if we're not allocating space for anything | |
3237 | else (see below). Also, we never need info about register 0 for | |
3238 | any of the other register vectors, and it seems rather a kludge to | |
3239 | treat `best_regend' differently than the rest. So we keep track of | |
3240 | the end of the best match so far in a separate variable. We | |
3241 | initialize this to NULL so that when we backtrack the first time | |
3242 | and need to test it, it's not garbage. */ | |
3243 | const char *match_end = NULL; | |
3244 | ||
3245 | /* Used when we pop values we don't care about. */ | |
3246 | const char **reg_dummy; | |
3247 | register_info_type *reg_info_dummy; | |
3248 | ||
3249 | #ifdef DEBUG | |
3250 | /* Counts the total number of registers pushed. */ | |
3251 | unsigned num_regs_pushed = 0; | |
3252 | #endif | |
3253 | ||
3254 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
3255 | ||
3256 | INIT_FAIL_STACK (); | |
3257 | ||
3258 | /* Do not bother to initialize all the register variables if there are | |
3259 | no groups in the pattern, as it takes a fair amount of time. If | |
3260 | there are groups, we include space for register 0 (the whole | |
3261 | pattern), even though we never use it, since it simplifies the | |
3262 | array indexing. We should fix this. */ | |
3263 | if (bufp->re_nsub) | |
3264 | { | |
3265 | regstart = REGEX_TALLOC (num_regs, const char *); | |
3266 | regend = REGEX_TALLOC (num_regs, const char *); | |
3267 | old_regstart = REGEX_TALLOC (num_regs, const char *); | |
3268 | old_regend = REGEX_TALLOC (num_regs, const char *); | |
3269 | best_regstart = REGEX_TALLOC (num_regs, const char *); | |
3270 | best_regend = REGEX_TALLOC (num_regs, const char *); | |
3271 | reg_info = REGEX_TALLOC (num_regs, register_info_type); | |
3272 | reg_dummy = REGEX_TALLOC (num_regs, const char *); | |
3273 | reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); | |
3274 | ||
3275 | if (!(regstart && regend && old_regstart && old_regend && reg_info | |
3276 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
3277 | { | |
3278 | FREE_VARIABLES (); | |
3279 | return -2; | |
3280 | } | |
3281 | } | |
3282 | #ifdef REGEX_MALLOC | |
3283 | else | |
3284 | { | |
3285 | /* We must initialize all our variables to NULL, so that | |
9114e279 | 3286 | `FREE_VARIABLES' doesn't try to free them. */ |
bc78d348 KB |
3287 | regstart = regend = old_regstart = old_regend = best_regstart |
3288 | = best_regend = reg_dummy = NULL; | |
3289 | reg_info = reg_info_dummy = (register_info_type *) NULL; | |
3290 | } | |
3291 | #endif /* REGEX_MALLOC */ | |
3292 | ||
3293 | /* The starting position is bogus. */ | |
3294 | if (pos < 0 || pos > size1 + size2) | |
3295 | { | |
3296 | FREE_VARIABLES (); | |
3297 | return -1; | |
3298 | } | |
3299 | ||
3300 | /* Initialize subexpression text positions to -1 to mark ones that no | |
3301 | start_memory/stop_memory has been seen for. Also initialize the | |
3302 | register information struct. */ | |
3303 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3304 | { | |
3305 | regstart[mcnt] = regend[mcnt] | |
3306 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
3307 | ||
3308 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
3309 | IS_ACTIVE (reg_info[mcnt]) = 0; | |
3310 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3311 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3312 | } | |
3313 | ||
3314 | /* We move `string1' into `string2' if the latter's empty -- but not if | |
3315 | `string1' is null. */ | |
3316 | if (size2 == 0 && string1 != NULL) | |
3317 | { | |
3318 | string2 = string1; | |
3319 | size2 = size1; | |
3320 | string1 = 0; | |
3321 | size1 = 0; | |
3322 | } | |
3323 | end1 = string1 + size1; | |
3324 | end2 = string2 + size2; | |
3325 | ||
3326 | /* Compute where to stop matching, within the two strings. */ | |
3327 | if (stop <= size1) | |
3328 | { | |
3329 | end_match_1 = string1 + stop; | |
3330 | end_match_2 = string2; | |
3331 | } | |
3332 | else | |
3333 | { | |
3334 | end_match_1 = end1; | |
3335 | end_match_2 = string2 + stop - size1; | |
3336 | } | |
3337 | ||
3338 | /* `p' scans through the pattern as `d' scans through the data. | |
3339 | `dend' is the end of the input string that `d' points within. `d' | |
3340 | is advanced into the following input string whenever necessary, but | |
3341 | this happens before fetching; therefore, at the beginning of the | |
3342 | loop, `d' can be pointing at the end of a string, but it cannot | |
3343 | equal `string2'. */ | |
3344 | if (size1 > 0 && pos <= size1) | |
3345 | { | |
3346 | d = string1 + pos; | |
3347 | dend = end_match_1; | |
3348 | } | |
3349 | else | |
3350 | { | |
3351 | d = string2 + pos - size1; | |
3352 | dend = end_match_2; | |
3353 | } | |
3354 | ||
3355 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
3356 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
3357 | DEBUG_PRINT1 ("The string to match is: `"); | |
3358 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
3359 | DEBUG_PRINT1 ("'\n"); | |
3360 | ||
3361 | /* This loops over pattern commands. It exits by returning from the | |
3362 | function if the match is complete, or it drops through if the match | |
3363 | fails at this starting point in the input data. */ | |
3364 | for (;;) | |
3365 | { | |
3366 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
3367 | ||
3368 | if (p == pend) | |
3369 | { /* End of pattern means we might have succeeded. */ | |
9114e279 KB |
3370 | DEBUG_PRINT1 ("end of pattern ... "); |
3371 | ||
3372 | /* If we haven't matched the entire string, and we want the | |
3373 | longest match, try backtracking. */ | |
bc78d348 KB |
3374 | if (d != end_match_2) |
3375 | { | |
3376 | DEBUG_PRINT1 ("backtracking.\n"); | |
3377 | ||
3378 | if (!FAIL_STACK_EMPTY ()) | |
3379 | { /* More failure points to try. */ | |
3380 | boolean same_str_p = (FIRST_STRING_P (match_end) | |
3381 | == MATCHING_IN_FIRST_STRING); | |
3382 | ||
3383 | /* If exceeds best match so far, save it. */ | |
3384 | if (!best_regs_set | |
3385 | || (same_str_p && d > match_end) | |
3386 | || (!same_str_p && !MATCHING_IN_FIRST_STRING)) | |
3387 | { | |
3388 | best_regs_set = true; | |
3389 | match_end = d; | |
3390 | ||
3391 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
3392 | ||
3393 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3394 | { | |
3395 | best_regstart[mcnt] = regstart[mcnt]; | |
3396 | best_regend[mcnt] = regend[mcnt]; | |
3397 | } | |
3398 | } | |
3399 | goto fail; | |
3400 | } | |
3401 | ||
3402 | /* If no failure points, don't restore garbage. */ | |
3403 | else if (best_regs_set) | |
3404 | { | |
3405 | restore_best_regs: | |
3406 | /* Restore best match. It may happen that `dend == | |
3407 | end_match_1' while the restored d is in string2. | |
3408 | For example, the pattern `x.*y.*z' against the | |
3409 | strings `x-' and `y-z-', if the two strings are | |
3410 | not consecutive in memory. */ | |
9114e279 KB |
3411 | DEBUG_PRINT1 ("Restoring best registers.\n"); |
3412 | ||
bc78d348 KB |
3413 | d = match_end; |
3414 | dend = ((d >= string1 && d <= end1) | |
3415 | ? end_match_1 : end_match_2); | |
3416 | ||
3417 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3418 | { | |
3419 | regstart[mcnt] = best_regstart[mcnt]; | |
3420 | regend[mcnt] = best_regend[mcnt]; | |
3421 | } | |
3422 | } | |
3423 | } /* d != end_match_2 */ | |
3424 | ||
9114e279 | 3425 | DEBUG_PRINT1 ("Accepting match.\n"); |
bc78d348 KB |
3426 | |
3427 | /* If caller wants register contents data back, do it. */ | |
3428 | if (regs && !bufp->no_sub) | |
3429 | { | |
3430 | /* Have the register data arrays been allocated? */ | |
3431 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
3432 | { /* No. So allocate them with malloc. We need one | |
3433 | extra element beyond `num_regs' for the `-1' marker | |
3434 | GNU code uses. */ | |
3435 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
3436 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
3437 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
3438 | if (regs->start == NULL || regs->end == NULL) | |
3439 | return -2; | |
3440 | bufp->regs_allocated = REGS_REALLOCATE; | |
3441 | } | |
3442 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
3443 | { /* Yes. If we need more elements than were already | |
3444 | allocated, reallocate them. If we need fewer, just | |
3445 | leave it alone. */ | |
3446 | if (regs->num_regs < num_regs + 1) | |
3447 | { | |
3448 | regs->num_regs = num_regs + 1; | |
3449 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
3450 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
3451 | if (regs->start == NULL || regs->end == NULL) | |
3452 | return -2; | |
3453 | } | |
3454 | } | |
3455 | else | |
3456 | assert (bufp->regs_allocated == REGS_FIXED); | |
3457 | ||
3458 | /* Convert the pointer data in `regstart' and `regend' to | |
3459 | indices. Register zero has to be set differently, | |
3460 | since we haven't kept track of any info for it. */ | |
3461 | if (regs->num_regs > 0) | |
3462 | { | |
3463 | regs->start[0] = pos; | |
3464 | regs->end[0] = (MATCHING_IN_FIRST_STRING ? d - string1 | |
3465 | : d - string2 + size1); | |
3466 | } | |
3467 | ||
3468 | /* Go through the first `min (num_regs, regs->num_regs)' | |
3469 | registers, since that is all we initialized. */ | |
3470 | for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) | |
3471 | { | |
3472 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
3473 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3474 | else | |
3475 | { | |
3476 | regs->start[mcnt] = POINTER_TO_OFFSET (regstart[mcnt]); | |
3477 | regs->end[mcnt] = POINTER_TO_OFFSET (regend[mcnt]); | |
3478 | } | |
3479 | } | |
3480 | ||
3481 | /* If the regs structure we return has more elements than | |
3482 | were in the pattern, set the extra elements to -1. If | |
3483 | we (re)allocated the registers, this is the case, | |
3484 | because we always allocate enough to have at least one | |
3485 | -1 at the end. */ | |
3486 | for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) | |
3487 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3488 | } /* regs && !bufp->no_sub */ | |
3489 | ||
3490 | FREE_VARIABLES (); | |
9114e279 KB |
3491 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
3492 | nfailure_points_pushed, nfailure_points_popped, | |
3493 | nfailure_points_pushed - nfailure_points_popped); | |
3494 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
bc78d348 KB |
3495 | |
3496 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
3497 | ? string1 | |
3498 | : string2 - size1); | |
3499 | ||
3500 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
3501 | ||
3502 | return mcnt; | |
3503 | } | |
3504 | ||
3505 | /* Otherwise match next pattern command. */ | |
3506 | #ifdef SWITCH_ENUM_BUG | |
3507 | switch ((int) ((re_opcode_t) *p++)) | |
3508 | #else | |
3509 | switch ((re_opcode_t) *p++) | |
3510 | #endif | |
3511 | { | |
3512 | /* Ignore these. Used to ignore the n of succeed_n's which | |
3513 | currently have n == 0. */ | |
3514 | case no_op: | |
3515 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
3516 | break; | |
3517 | ||
3518 | ||
3519 | /* Match the next n pattern characters exactly. The following | |
3520 | byte in the pattern defines n, and the n bytes after that | |
3521 | are the characters to match. */ | |
3522 | case exactn: | |
3523 | mcnt = *p++; | |
3524 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
3525 | ||
3526 | /* This is written out as an if-else so we don't waste time | |
3527 | testing `translate' inside the loop. */ | |
3528 | if (translate) | |
3529 | { | |
3530 | do | |
3531 | { | |
3532 | PREFETCH (); | |
3533 | if (translate[(unsigned char) *d++] != (char) *p++) | |
3534 | goto fail; | |
3535 | } | |
3536 | while (--mcnt); | |
3537 | } | |
3538 | else | |
3539 | { | |
3540 | do | |
3541 | { | |
3542 | PREFETCH (); | |
3543 | if (*d++ != (char) *p++) goto fail; | |
3544 | } | |
3545 | while (--mcnt); | |
3546 | } | |
3547 | SET_REGS_MATCHED (); | |
3548 | break; | |
3549 | ||
3550 | ||
3551 | /* Match any character except possibly a newline or a null. */ | |
3552 | case anychar: | |
3553 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
3554 | ||
3555 | PREFETCH (); | |
3556 | ||
3557 | if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
3558 | || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
3559 | goto fail; | |
3560 | ||
3561 | SET_REGS_MATCHED (); | |
3562 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); | |
3563 | d++; | |
3564 | break; | |
3565 | ||
3566 | ||
3567 | case charset: | |
3568 | case charset_not: | |
3569 | { | |
3570 | register unsigned char c; | |
3571 | boolean not = (re_opcode_t) *(p - 1) == charset_not; | |
3572 | ||
3573 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
3574 | ||
3575 | PREFETCH (); | |
3576 | c = TRANSLATE (*d); /* The character to match. */ | |
3577 | ||
3578 | /* Cast to `unsigned' instead of `unsigned char' in case the | |
3579 | bit list is a full 32 bytes long. */ | |
3580 | if (c < (unsigned) (*p * BYTEWIDTH) | |
3581 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
3582 | not = !not; | |
3583 | ||
3584 | p += 1 + *p; | |
3585 | ||
3586 | if (!not) goto fail; | |
3587 | ||
3588 | SET_REGS_MATCHED (); | |
3589 | d++; | |
3590 | break; | |
3591 | } | |
3592 | ||
3593 | ||
3594 | /* The beginning of a group is represented by start_memory. | |
3595 | The arguments are the register number in the next byte, and the | |
3596 | number of groups inner to this one in the next. The text | |
3597 | matched within the group is recorded (in the internal | |
3598 | registers data structure) under the register number. */ | |
3599 | case start_memory: | |
3600 | DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); | |
3601 | ||
3602 | /* Find out if this group can match the empty string. */ | |
3603 | p1 = p; /* To send to group_match_null_string_p. */ | |
3604 | ||
3605 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
3606 | REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3607 | = group_match_null_string_p (&p1, pend, reg_info); | |
3608 | ||
3609 | /* Save the position in the string where we were the last time | |
3610 | we were at this open-group operator in case the group is | |
3611 | operated upon by a repetition operator, e.g., with `(a*)*b' | |
3612 | against `ab'; then we want to ignore where we are now in | |
3613 | the string in case this attempt to match fails. */ | |
3614 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3615 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
3616 | : regstart[*p]; | |
3617 | DEBUG_PRINT2 (" old_regstart: %d\n", | |
3618 | POINTER_TO_OFFSET (old_regstart[*p])); | |
3619 | ||
3620 | regstart[*p] = d; | |
3621 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
3622 | ||
3623 | IS_ACTIVE (reg_info[*p]) = 1; | |
3624 | MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3625 | ||
3626 | /* This is the new highest active register. */ | |
3627 | highest_active_reg = *p; | |
3628 | ||
3629 | /* If nothing was active before, this is the new lowest active | |
3630 | register. */ | |
3631 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
3632 | lowest_active_reg = *p; | |
3633 | ||
3634 | /* Move past the register number and inner group count. */ | |
3635 | p += 2; | |
3636 | break; | |
3637 | ||
3638 | ||
3639 | /* The stop_memory opcode represents the end of a group. Its | |
3640 | arguments are the same as start_memory's: the register | |
3641 | number, and the number of inner groups. */ | |
3642 | case stop_memory: | |
3643 | DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); | |
3644 | ||
3645 | /* We need to save the string position the last time we were at | |
3646 | this close-group operator in case the group is operated | |
3647 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
3648 | against `aba'; then we want to ignore where we are now in | |
3649 | the string in case this attempt to match fails. */ | |
3650 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3651 | ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
3652 | : regend[*p]; | |
3653 | DEBUG_PRINT2 (" old_regend: %d\n", | |
3654 | POINTER_TO_OFFSET (old_regend[*p])); | |
3655 | ||
3656 | regend[*p] = d; | |
3657 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
3658 | ||
3659 | /* This register isn't active anymore. */ | |
3660 | IS_ACTIVE (reg_info[*p]) = 0; | |
3661 | ||
3662 | /* If this was the only register active, nothing is active | |
3663 | anymore. */ | |
3664 | if (lowest_active_reg == highest_active_reg) | |
3665 | { | |
3666 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3667 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3668 | } | |
3669 | else | |
3670 | { /* We must scan for the new highest active register, since | |
3671 | it isn't necessarily one less than now: consider | |
3672 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
3673 | new highest active register is 1. */ | |
3674 | unsigned char r = *p - 1; | |
3675 | while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
3676 | r--; | |
3677 | ||
3678 | /* If we end up at register zero, that means that we saved | |
3679 | the registers as the result of an `on_failure_jump', not | |
3680 | a `start_memory', and we jumped to past the innermost | |
3681 | `stop_memory'. For example, in ((.)*) we save | |
3682 | registers 1 and 2 as a result of the *, but when we pop | |
3683 | back to the second ), we are at the stop_memory 1. | |
3684 | Thus, nothing is active. */ | |
3685 | if (r == 0) | |
3686 | { | |
3687 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3688 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3689 | } | |
3690 | else | |
3691 | highest_active_reg = r; | |
3692 | } | |
3693 | ||
3694 | /* If just failed to match something this time around with a | |
3695 | group that's operated on by a repetition operator, try to | |
9114e279 | 3696 | force exit from the ``loop'', and restore the register |
bc78d348 KB |
3697 | information for this group that we had before trying this |
3698 | last match. */ | |
3699 | if ((!MATCHED_SOMETHING (reg_info[*p]) | |
3700 | || (re_opcode_t) p[-3] == start_memory) | |
3701 | && (p + 2) < pend) | |
3702 | { | |
3703 | boolean is_a_jump_n = false; | |
3704 | ||
3705 | p1 = p + 2; | |
3706 | mcnt = 0; | |
3707 | switch ((re_opcode_t) *p1++) | |
3708 | { | |
3709 | case jump_n: | |
3710 | is_a_jump_n = true; | |
3711 | case pop_failure_jump: | |
3712 | case maybe_pop_jump: | |
3713 | case jump: | |
3714 | case dummy_failure_jump: | |
3715 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3716 | if (is_a_jump_n) | |
3717 | p1 += 2; | |
3718 | break; | |
3719 | ||
3720 | default: | |
3721 | /* do nothing */ ; | |
3722 | } | |
3723 | p1 += mcnt; | |
3724 | ||
3725 | /* If the next operation is a jump backwards in the pattern | |
3726 | to an on_failure_jump right before the start_memory | |
3727 | corresponding to this stop_memory, exit from the loop | |
3728 | by forcing a failure after pushing on the stack the | |
3729 | on_failure_jump's jump in the pattern, and d. */ | |
3730 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
3731 | && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
3732 | { | |
3733 | /* If this group ever matched anything, then restore | |
3734 | what its registers were before trying this last | |
3735 | failed match, e.g., with `(a*)*b' against `ab' for | |
3736 | regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
3737 | against `aba' for regend[3]. | |
3738 | ||
3739 | Also restore the registers for inner groups for, | |
3740 | e.g., `((a*)(b*))*' against `aba' (register 3 would | |
3741 | otherwise get trashed). */ | |
3742 | ||
3743 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
3744 | { | |
3745 | unsigned r; | |
3746 | ||
3747 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3748 | ||
3749 | /* Restore this and inner groups' (if any) registers. */ | |
3750 | for (r = *p; r < *p + *(p + 1); r++) | |
3751 | { | |
3752 | regstart[r] = old_regstart[r]; | |
3753 | ||
3754 | /* xx why this test? */ | |
3755 | if ((int) old_regend[r] >= (int) regstart[r]) | |
3756 | regend[r] = old_regend[r]; | |
3757 | } | |
3758 | } | |
3759 | p1++; | |
3760 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3761 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
3762 | ||
3763 | goto fail; | |
3764 | } | |
3765 | } | |
3766 | ||
3767 | /* Move past the register number and the inner group count. */ | |
3768 | p += 2; | |
3769 | break; | |
3770 | ||
3771 | ||
3772 | /* \<digit> has been turned into a `duplicate' command which is | |
3773 | followed by the numeric value of <digit> as the register number. */ | |
3774 | case duplicate: | |
3775 | { | |
3776 | register const char *d2, *dend2; | |
3777 | int regno = *p++; /* Get which register to match against. */ | |
3778 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
3779 | ||
3780 | /* Can't back reference a group which we've never matched. */ | |
3781 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
3782 | goto fail; | |
3783 | ||
3784 | /* Where in input to try to start matching. */ | |
3785 | d2 = regstart[regno]; | |
3786 | ||
3787 | /* Where to stop matching; if both the place to start and | |
3788 | the place to stop matching are in the same string, then | |
3789 | set to the place to stop, otherwise, for now have to use | |
3790 | the end of the first string. */ | |
3791 | ||
3792 | dend2 = ((FIRST_STRING_P (regstart[regno]) | |
3793 | == FIRST_STRING_P (regend[regno])) | |
3794 | ? regend[regno] : end_match_1); | |
3795 | for (;;) | |
3796 | { | |
3797 | /* If necessary, advance to next segment in register | |
3798 | contents. */ | |
3799 | while (d2 == dend2) | |
3800 | { | |
3801 | if (dend2 == end_match_2) break; | |
3802 | if (dend2 == regend[regno]) break; | |
3803 | ||
3804 | /* End of string1 => advance to string2. */ | |
3805 | d2 = string2; | |
3806 | dend2 = regend[regno]; | |
3807 | } | |
3808 | /* At end of register contents => success */ | |
3809 | if (d2 == dend2) break; | |
3810 | ||
3811 | /* If necessary, advance to next segment in data. */ | |
3812 | PREFETCH (); | |
3813 | ||
3814 | /* How many characters left in this segment to match. */ | |
3815 | mcnt = dend - d; | |
3816 | ||
3817 | /* Want how many consecutive characters we can match in | |
3818 | one shot, so, if necessary, adjust the count. */ | |
3819 | if (mcnt > dend2 - d2) | |
3820 | mcnt = dend2 - d2; | |
3821 | ||
3822 | /* Compare that many; failure if mismatch, else move | |
3823 | past them. */ | |
3824 | if (translate | |
3825 | ? bcmp_translate (d, d2, mcnt, translate) | |
3826 | : bcmp (d, d2, mcnt)) | |
3827 | goto fail; | |
3828 | d += mcnt, d2 += mcnt; | |
3829 | } | |
3830 | } | |
3831 | break; | |
3832 | ||
3833 | ||
3834 | /* begline matches the empty string at the beginning of the string | |
3835 | (unless `not_bol' is set in `bufp'), and, if | |
3836 | `newline_anchor' is set, after newlines. */ | |
3837 | case begline: | |
3838 | DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
3839 | ||
9114e279 | 3840 | if (AT_STRINGS_BEG (d)) |
bc78d348 KB |
3841 | { |
3842 | if (!bufp->not_bol) break; | |
3843 | } | |
3844 | else if (d[-1] == '\n' && bufp->newline_anchor) | |
3845 | { | |
3846 | break; | |
3847 | } | |
3848 | /* In all other cases, we fail. */ | |
3849 | goto fail; | |
3850 | ||
3851 | ||
3852 | /* endline is the dual of begline. */ | |
3853 | case endline: | |
3854 | DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
3855 | ||
9114e279 | 3856 | if (AT_STRINGS_END (d)) |
bc78d348 KB |
3857 | { |
3858 | if (!bufp->not_eol) break; | |
3859 | } | |
3860 | ||
3861 | /* We have to ``prefetch'' the next character. */ | |
3862 | else if ((d == end1 ? *string2 : *d) == '\n' | |
3863 | && bufp->newline_anchor) | |
3864 | { | |
3865 | break; | |
3866 | } | |
3867 | goto fail; | |
3868 | ||
3869 | ||
3870 | /* Match at the very beginning of the data. */ | |
3871 | case begbuf: | |
3872 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
9114e279 | 3873 | if (AT_STRINGS_BEG (d)) |
bc78d348 KB |
3874 | break; |
3875 | goto fail; | |
3876 | ||
3877 | ||
3878 | /* Match at the very end of the data. */ | |
3879 | case endbuf: | |
3880 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
9114e279 | 3881 | if (AT_STRINGS_END (d)) |
bc78d348 KB |
3882 | break; |
3883 | goto fail; | |
3884 | ||
3885 | ||
3886 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
3887 | pushes NULL as the value for the string on the stack. Then | |
3888 | `pop_failure_point' will keep the current value for the | |
3889 | string, instead of restoring it. To see why, consider | |
3890 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
3891 | then the . fails against the \n. But the next thing we want | |
3892 | to do is match the \n against the \n; if we restored the | |
3893 | string value, we would be back at the foo. | |
3894 | ||
3895 | Because this is used only in specific cases, we don't need to | |
3896 | check all the things that `on_failure_jump' does, to make | |
3897 | sure the right things get saved on the stack. Hence we don't | |
3898 | share its code. The only reason to push anything on the | |
3899 | stack at all is that otherwise we would have to change | |
3900 | `anychar's code to do something besides goto fail in this | |
3901 | case; that seems worse than this. */ | |
3902 | case on_failure_keep_string_jump: | |
3903 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
3904 | ||
3905 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
3906 | DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
3907 | ||
3908 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
3909 | break; | |
3910 | ||
3911 | ||
3912 | /* Uses of on_failure_jump: | |
3913 | ||
3914 | Each alternative starts with an on_failure_jump that points | |
3915 | to the beginning of the next alternative. Each alternative | |
3916 | except the last ends with a jump that in effect jumps past | |
3917 | the rest of the alternatives. (They really jump to the | |
3918 | ending jump of the following alternative, because tensioning | |
3919 | these jumps is a hassle.) | |
3920 | ||
3921 | Repeats start with an on_failure_jump that points past both | |
3922 | the repetition text and either the following jump or | |
3923 | pop_failure_jump back to this on_failure_jump. */ | |
3924 | case on_failure_jump: | |
3925 | on_failure: | |
3926 | DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
3927 | ||
3928 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
3929 | DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
3930 | ||
3931 | /* If this on_failure_jump comes right before a group (i.e., | |
3932 | the original * applied to a group), save the information | |
3933 | for that group and all inner ones, so that if we fail back | |
3934 | to this point, the group's information will be correct. | |
9114e279 | 3935 | For example, in \(a*\)*\1, we need the preceding group, |
bc78d348 KB |
3936 | and in \(\(a*\)b*\)\2, we need the inner group. */ |
3937 | ||
3938 | /* We can't use `p' to check ahead because we push | |
3939 | a failure point to `p + mcnt' after we do this. */ | |
3940 | p1 = p; | |
3941 | ||
3942 | /* We need to skip no_op's before we look for the | |
3943 | start_memory in case this on_failure_jump is happening as | |
3944 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
3945 | against aba. */ | |
3946 | while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
3947 | p1++; | |
3948 | ||
3949 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
3950 | { | |
3951 | /* We have a new highest active register now. This will | |
3952 | get reset at the start_memory we are about to get to, | |
3953 | but we will have saved all the registers relevant to | |
3954 | this repetition op, as described above. */ | |
3955 | highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
3956 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
3957 | lowest_active_reg = *(p1 + 1); | |
3958 | } | |
3959 | ||
3960 | DEBUG_PRINT1 (":\n"); | |
3961 | PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
3962 | break; | |
3963 | ||
3964 | ||
9114e279 KB |
3965 | /* A smart repeat ends with `maybe_pop_jump'. |
3966 | We change it to either `pop_failure_jump' or `jump'. */ | |
bc78d348 KB |
3967 | case maybe_pop_jump: |
3968 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
3969 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
3970 | { | |
3971 | register unsigned char *p2 = p; | |
3972 | ||
3973 | /* Compare the beginning of the repeat with what in the | |
3974 | pattern follows its end. If we can establish that there | |
3975 | is nothing that they would both match, i.e., that we | |
3976 | would have to backtrack because of (as in, e.g., `a*a') | |
3977 | then we can change to pop_failure_jump, because we'll | |
3978 | never have to backtrack. | |
3979 | ||
3980 | This is not true in the case of alternatives: in | |
3981 | `(a|ab)*' we do need to backtrack to the `ab' alternative | |
3982 | (e.g., if the string was `ab'). But instead of trying to | |
3983 | detect that here, the alternative has put on a dummy | |
3984 | failure point which is what we will end up popping. */ | |
3985 | ||
3986 | /* Skip over open/close-group commands. */ | |
3987 | while (p2 + 2 < pend | |
3988 | && ((re_opcode_t) *p2 == stop_memory | |
3989 | || (re_opcode_t) *p2 == start_memory)) | |
3990 | p2 += 3; /* Skip over args, too. */ | |
3991 | ||
3992 | /* If we're at the end of the pattern, we can change. */ | |
3993 | if (p2 == pend) | |
71715da9 JB |
3994 | { |
3995 | /* Consider what happens when matching ":\(.*\)" | |
3996 | against ":/". I don't really understand this code | |
3997 | yet. */ | |
bc78d348 | 3998 | p[-3] = (unsigned char) pop_failure_jump; |
71715da9 JB |
3999 | DEBUG_PRINT1 |
4000 | (" End of pattern: change to `pop_failure_jump'.\n"); | |
bc78d348 KB |
4001 | } |
4002 | ||
4003 | else if ((re_opcode_t) *p2 == exactn | |
4004 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
4005 | { | |
4006 | register unsigned char c | |
4007 | = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
4008 | p1 = p + mcnt; | |
4009 | ||
4010 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
4011 | to the `maybe_finalize_jump' of this case. Examine what | |
4012 | follows. */ | |
4013 | if ((re_opcode_t) p1[3] == exactn && p1[5] != c) | |
9114e279 KB |
4014 | { |
4015 | p[-3] = (unsigned char) pop_failure_jump; | |
4016 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
4017 | c, p1[5]); | |
4018 | } | |
4019 | ||
bc78d348 KB |
4020 | else if ((re_opcode_t) p1[3] == charset |
4021 | || (re_opcode_t) p1[3] == charset_not) | |
4022 | { | |
4023 | int not = (re_opcode_t) p1[3] == charset_not; | |
4024 | ||
4025 | if (c < (unsigned char) (p1[4] * BYTEWIDTH) | |
4026 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4027 | not = !not; | |
4028 | ||
4029 | /* `not' is equal to 1 if c would match, which means | |
4030 | that we can't change to pop_failure_jump. */ | |
4031 | if (!not) | |
4032 | { | |
4033 | p[-3] = (unsigned char) pop_failure_jump; | |
9114e279 | 4034 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
bc78d348 KB |
4035 | } |
4036 | } | |
4037 | } | |
4038 | } | |
4039 | p -= 2; /* Point at relative address again. */ | |
4040 | if ((re_opcode_t) p[-1] != pop_failure_jump) | |
4041 | { | |
4042 | p[-1] = (unsigned char) jump; | |
9114e279 | 4043 | DEBUG_PRINT1 (" Match => jump.\n"); |
bc78d348 KB |
4044 | goto unconditional_jump; |
4045 | } | |
4046 | /* Note fall through. */ | |
4047 | ||
4048 | ||
4049 | /* The end of a simple repeat has a pop_failure_jump back to | |
4050 | its matching on_failure_jump, where the latter will push a | |
4051 | failure point. The pop_failure_jump takes off failure | |
4052 | points put on by this pop_failure_jump's matching | |
4053 | on_failure_jump; we got through the pattern to here from the | |
4054 | matching on_failure_jump, so didn't fail. */ | |
4055 | case pop_failure_jump: | |
4056 | { | |
4057 | /* We need to pass separate storage for the lowest and | |
4058 | highest registers, even though we don't care about the | |
4059 | actual values. Otherwise, we will restore only one | |
4060 | register from the stack, since lowest will == highest in | |
4061 | `pop_failure_point'. */ | |
4062 | unsigned dummy_low_reg, dummy_high_reg; | |
4063 | unsigned char *pdummy; | |
4064 | const char *sdummy; | |
4065 | ||
4066 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
4067 | POP_FAILURE_POINT (sdummy, pdummy, | |
4068 | dummy_low_reg, dummy_high_reg, | |
4069 | reg_dummy, reg_dummy, reg_info_dummy); | |
4070 | } | |
4071 | /* Note fall through. */ | |
4072 | ||
4073 | ||
4074 | /* Unconditionally jump (without popping any failure points). */ | |
4075 | case jump: | |
4076 | unconditional_jump: | |
4077 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
4078 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
4079 | p += mcnt; /* Do the jump. */ | |
4080 | DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
4081 | break; | |
4082 | ||
4083 | ||
4084 | /* We need this opcode so we can detect where alternatives end | |
4085 | in `group_match_null_string_p' et al. */ | |
4086 | case jump_past_alt: | |
4087 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
4088 | goto unconditional_jump; | |
4089 | ||
4090 | ||
4091 | /* Normally, the on_failure_jump pushes a failure point, which | |
4092 | then gets popped at pop_failure_jump. We will end up at | |
4093 | pop_failure_jump, also, and with a pattern of, say, `a+', we | |
4094 | are skipping over the on_failure_jump, so we have to push | |
4095 | something meaningless for pop_failure_jump to pop. */ | |
4096 | case dummy_failure_jump: | |
4097 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
4098 | /* It doesn't matter what we push for the string here. What | |
4099 | the code at `fail' tests is the value for the pattern. */ | |
4100 | PUSH_FAILURE_POINT (0, 0, -2); | |
4101 | goto unconditional_jump; | |
4102 | ||
4103 | ||
4104 | /* At the end of an alternative, we need to push a dummy failure | |
9114e279 | 4105 | point in case we are followed by a `pop_failure_jump', because |
bc78d348 KB |
4106 | we don't want the failure point for the alternative to be |
4107 | popped. For example, matching `(a|ab)*' against `aab' | |
4108 | requires that we match the `ab' alternative. */ | |
4109 | case push_dummy_failure: | |
4110 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
4111 | /* See comments just above at `dummy_failure_jump' about the | |
4112 | two zeroes. */ | |
4113 | PUSH_FAILURE_POINT (0, 0, -2); | |
4114 | break; | |
4115 | ||
4116 | /* Have to succeed matching what follows at least n times. | |
4117 | After that, handle like `on_failure_jump'. */ | |
4118 | case succeed_n: | |
4119 | EXTRACT_NUMBER (mcnt, p + 2); | |
4120 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
4121 | ||
4122 | assert (mcnt >= 0); | |
4123 | /* Originally, this is how many times we HAVE to succeed. */ | |
4124 | if (mcnt > 0) | |
4125 | { | |
4126 | mcnt--; | |
4127 | p += 2; | |
4128 | STORE_NUMBER_AND_INCR (p, mcnt); | |
4129 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); | |
4130 | } | |
4131 | else if (mcnt == 0) | |
4132 | { | |
4133 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); | |
4134 | p[2] = (unsigned char) no_op; | |
4135 | p[3] = (unsigned char) no_op; | |
4136 | goto on_failure; | |
4137 | } | |
4138 | break; | |
4139 | ||
4140 | case jump_n: | |
4141 | EXTRACT_NUMBER (mcnt, p + 2); | |
4142 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
4143 | ||
4144 | /* Originally, this is how many times we CAN jump. */ | |
4145 | if (mcnt) | |
4146 | { | |
4147 | mcnt--; | |
4148 | STORE_NUMBER (p + 2, mcnt); | |
4149 | goto unconditional_jump; | |
4150 | } | |
4151 | /* If don't have to jump any more, skip over the rest of command. */ | |
4152 | else | |
4153 | p += 4; | |
4154 | break; | |
4155 | ||
4156 | case set_number_at: | |
4157 | { | |
4158 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
4159 | ||
4160 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4161 | p1 = p + mcnt; | |
4162 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4163 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
4164 | STORE_NUMBER (p1, mcnt); | |
4165 | break; | |
4166 | } | |
4167 | ||
4168 | case wordbound: | |
4169 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
4170 | if (AT_WORD_BOUNDARY (d)) | |
4171 | break; | |
4172 | goto fail; | |
4173 | ||
4174 | case notwordbound: | |
4175 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
4176 | if (AT_WORD_BOUNDARY (d)) | |
4177 | goto fail; | |
4178 | break; | |
4179 | ||
4180 | case wordbeg: | |
4181 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
9114e279 | 4182 | if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) |
bc78d348 KB |
4183 | break; |
4184 | goto fail; | |
4185 | ||
4186 | case wordend: | |
4187 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
9114e279 KB |
4188 | if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) |
4189 | && (!WORDCHAR_P (d) || AT_STRINGS_END (d))) | |
bc78d348 KB |
4190 | break; |
4191 | goto fail; | |
4192 | ||
4193 | #ifdef emacs | |
4194 | #ifdef emacs19 | |
4195 | case before_dot: | |
4196 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
4197 | if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
4198 | goto fail; | |
4199 | break; | |
4200 | ||
4201 | case at_dot: | |
4202 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4203 | if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
4204 | goto fail; | |
4205 | break; | |
4206 | ||
4207 | case after_dot: | |
4208 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
4209 | if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
4210 | goto fail; | |
4211 | break; | |
4212 | #else /* not emacs19 */ | |
4213 | case at_dot: | |
4214 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4215 | if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point) | |
4216 | goto fail; | |
4217 | break; | |
4218 | #endif /* not emacs19 */ | |
4219 | ||
4220 | case syntaxspec: | |
4221 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
4222 | mcnt = *p++; | |
4223 | goto matchsyntax; | |
4224 | ||
4225 | case wordchar: | |
9114e279 | 4226 | DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); |
bc78d348 KB |
4227 | mcnt = (int) Sword; |
4228 | matchsyntax: | |
4229 | PREFETCH (); | |
9114e279 KB |
4230 | if (SYNTAX (*d++) != (enum syntaxcode) mcnt) |
4231 | goto fail; | |
bc78d348 KB |
4232 | SET_REGS_MATCHED (); |
4233 | break; | |
4234 | ||
4235 | case notsyntaxspec: | |
4236 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
4237 | mcnt = *p++; | |
4238 | goto matchnotsyntax; | |
4239 | ||
4240 | case notwordchar: | |
9114e279 | 4241 | DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); |
bc78d348 | 4242 | mcnt = (int) Sword; |
9114e279 | 4243 | matchnotsyntax: |
bc78d348 | 4244 | PREFETCH (); |
9114e279 KB |
4245 | if (SYNTAX (*d++) == (enum syntaxcode) mcnt) |
4246 | goto fail; | |
bc78d348 KB |
4247 | SET_REGS_MATCHED (); |
4248 | break; | |
4249 | ||
4250 | #else /* not emacs */ | |
4251 | case wordchar: | |
4252 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
4253 | PREFETCH (); | |
9114e279 | 4254 | if (!WORDCHAR_P (d)) |
bc78d348 KB |
4255 | goto fail; |
4256 | SET_REGS_MATCHED (); | |
9114e279 | 4257 | d++; |
bc78d348 KB |
4258 | break; |
4259 | ||
4260 | case notwordchar: | |
4261 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
4262 | PREFETCH (); | |
9114e279 | 4263 | if (WORDCHAR_P (d)) |
bc78d348 KB |
4264 | goto fail; |
4265 | SET_REGS_MATCHED (); | |
9114e279 | 4266 | d++; |
bc78d348 KB |
4267 | break; |
4268 | #endif /* not emacs */ | |
4269 | ||
4270 | default: | |
4271 | abort (); | |
4272 | } | |
4273 | continue; /* Successfully executed one pattern command; keep going. */ | |
4274 | ||
4275 | ||
4276 | /* We goto here if a matching operation fails. */ | |
4277 | fail: | |
4278 | if (!FAIL_STACK_EMPTY ()) | |
4279 | { /* A restart point is known. Restore to that state. */ | |
4280 | DEBUG_PRINT1 ("\nFAIL:\n"); | |
4281 | POP_FAILURE_POINT (d, p, | |
4282 | lowest_active_reg, highest_active_reg, | |
4283 | regstart, regend, reg_info); | |
4284 | ||
4285 | /* If this failure point is a dummy, try the next one. */ | |
4286 | if (!p) | |
4287 | goto fail; | |
4288 | ||
4289 | /* If we failed to the end of the pattern, don't examine *p. */ | |
4290 | assert (p <= pend); | |
4291 | if (p < pend) | |
4292 | { | |
4293 | boolean is_a_jump_n = false; | |
4294 | ||
4295 | /* If failed to a backwards jump that's part of a repetition | |
4296 | loop, need to pop this failure point and use the next one. */ | |
4297 | switch ((re_opcode_t) *p) | |
4298 | { | |
4299 | case jump_n: | |
4300 | is_a_jump_n = true; | |
4301 | case maybe_pop_jump: | |
4302 | case pop_failure_jump: | |
4303 | case jump: | |
4304 | p1 = p + 1; | |
4305 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4306 | p1 += mcnt; | |
4307 | ||
4308 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
4309 | || (!is_a_jump_n | |
4310 | && (re_opcode_t) *p1 == on_failure_jump)) | |
4311 | goto fail; | |
4312 | break; | |
4313 | default: | |
4314 | /* do nothing */ ; | |
4315 | } | |
4316 | } | |
4317 | ||
4318 | if (d >= string1 && d <= end1) | |
4319 | dend = end_match_1; | |
4320 | } | |
4321 | else | |
4322 | break; /* Matching at this starting point really fails. */ | |
4323 | } /* for (;;) */ | |
4324 | ||
4325 | if (best_regs_set) | |
4326 | goto restore_best_regs; | |
4327 | ||
4328 | FREE_VARIABLES (); | |
4329 | ||
4330 | return -1; /* Failure to match. */ | |
4331 | } /* re_match_2 */ | |
4332 | \f | |
4333 | /* Subroutine definitions for re_match_2. */ | |
4334 | ||
4335 | ||
4336 | /* We are passed P pointing to a register number after a start_memory. | |
4337 | ||
4338 | Return true if the pattern up to the corresponding stop_memory can | |
4339 | match the empty string, and false otherwise. | |
4340 | ||
4341 | If we find the matching stop_memory, sets P to point to one past its number. | |
4342 | Otherwise, sets P to an undefined byte less than or equal to END. | |
4343 | ||
4344 | We don't handle duplicates properly (yet). */ | |
4345 | ||
4346 | static boolean | |
4347 | group_match_null_string_p (p, end, reg_info) | |
4348 | unsigned char **p, *end; | |
4349 | register_info_type *reg_info; | |
4350 | { | |
4351 | int mcnt; | |
4352 | /* Point to after the args to the start_memory. */ | |
4353 | unsigned char *p1 = *p + 2; | |
4354 | ||
4355 | while (p1 < end) | |
4356 | { | |
4357 | /* Skip over opcodes that can match nothing, and return true or | |
4358 | false, as appropriate, when we get to one that can't, or to the | |
4359 | matching stop_memory. */ | |
4360 | ||
4361 | switch ((re_opcode_t) *p1) | |
4362 | { | |
4363 | /* Could be either a loop or a series of alternatives. */ | |
4364 | case on_failure_jump: | |
4365 | p1++; | |
4366 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4367 | ||
4368 | /* If the next operation is not a jump backwards in the | |
4369 | pattern. */ | |
4370 | ||
4371 | if (mcnt >= 0) | |
4372 | { | |
4373 | /* Go through the on_failure_jumps of the alternatives, | |
4374 | seeing if any of the alternatives cannot match nothing. | |
4375 | The last alternative starts with only a jump, | |
4376 | whereas the rest start with on_failure_jump and end | |
4377 | with a jump, e.g., here is the pattern for `a|b|c': | |
4378 | ||
4379 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
4380 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
4381 | /exactn/1/c | |
4382 | ||
4383 | So, we have to first go through the first (n-1) | |
4384 | alternatives and then deal with the last one separately. */ | |
4385 | ||
4386 | ||
4387 | /* Deal with the first (n-1) alternatives, which start | |
4388 | with an on_failure_jump (see above) that jumps to right | |
4389 | past a jump_past_alt. */ | |
4390 | ||
4391 | while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) | |
4392 | { | |
4393 | /* `mcnt' holds how many bytes long the alternative | |
4394 | is, including the ending `jump_past_alt' and | |
4395 | its number. */ | |
4396 | ||
4397 | if (!alt_match_null_string_p (p1, p1 + mcnt - 3, | |
4398 | reg_info)) | |
4399 | return false; | |
4400 | ||
4401 | /* Move to right after this alternative, including the | |
4402 | jump_past_alt. */ | |
4403 | p1 += mcnt; | |
4404 | ||
4405 | /* Break if it's the beginning of an n-th alternative | |
4406 | that doesn't begin with an on_failure_jump. */ | |
4407 | if ((re_opcode_t) *p1 != on_failure_jump) | |
4408 | break; | |
4409 | ||
4410 | /* Still have to check that it's not an n-th | |
4411 | alternative that starts with an on_failure_jump. */ | |
4412 | p1++; | |
4413 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4414 | if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
4415 | { | |
4416 | /* Get to the beginning of the n-th alternative. */ | |
4417 | p1 -= 3; | |
4418 | break; | |
4419 | } | |
4420 | } | |
4421 | ||
4422 | /* Deal with the last alternative: go back and get number | |
4423 | of the `jump_past_alt' just before it. `mcnt' contains | |
4424 | the length of the alternative. */ | |
4425 | EXTRACT_NUMBER (mcnt, p1 - 2); | |
4426 | ||
4427 | if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
4428 | return false; | |
4429 | ||
4430 | p1 += mcnt; /* Get past the n-th alternative. */ | |
4431 | } /* if mcnt > 0 */ | |
4432 | break; | |
4433 | ||
4434 | ||
4435 | case stop_memory: | |
4436 | assert (p1[1] == **p); | |
4437 | *p = p1 + 2; | |
4438 | return true; | |
4439 | ||
4440 | ||
4441 | default: | |
4442 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4443 | return false; | |
4444 | } | |
4445 | } /* while p1 < end */ | |
4446 | ||
4447 | return false; | |
4448 | } /* group_match_null_string_p */ | |
4449 | ||
4450 | ||
4451 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
4452 | It expects P to be the first byte of a single alternative and END one | |
4453 | byte past the last. The alternative can contain groups. */ | |
4454 | ||
4455 | static boolean | |
4456 | alt_match_null_string_p (p, end, reg_info) | |
4457 | unsigned char *p, *end; | |
4458 | register_info_type *reg_info; | |
4459 | { | |
4460 | int mcnt; | |
4461 | unsigned char *p1 = p; | |
4462 | ||
4463 | while (p1 < end) | |
4464 | { | |
4465 | /* Skip over opcodes that can match nothing, and break when we get | |
4466 | to one that can't. */ | |
4467 | ||
4468 | switch ((re_opcode_t) *p1) | |
4469 | { | |
4470 | /* It's a loop. */ | |
4471 | case on_failure_jump: | |
4472 | p1++; | |
4473 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4474 | p1 += mcnt; | |
4475 | break; | |
4476 | ||
4477 | default: | |
4478 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4479 | return false; | |
4480 | } | |
4481 | } /* while p1 < end */ | |
4482 | ||
4483 | return true; | |
4484 | } /* alt_match_null_string_p */ | |
4485 | ||
4486 | ||
4487 | /* Deals with the ops common to group_match_null_string_p and | |
4488 | alt_match_null_string_p. | |
4489 | ||
4490 | Sets P to one after the op and its arguments, if any. */ | |
4491 | ||
4492 | static boolean | |
4493 | common_op_match_null_string_p (p, end, reg_info) | |
4494 | unsigned char **p, *end; | |
4495 | register_info_type *reg_info; | |
4496 | { | |
4497 | int mcnt; | |
4498 | boolean ret; | |
4499 | int reg_no; | |
4500 | unsigned char *p1 = *p; | |
4501 | ||
4502 | switch ((re_opcode_t) *p1++) | |
4503 | { | |
4504 | case no_op: | |
4505 | case begline: | |
4506 | case endline: | |
4507 | case begbuf: | |
4508 | case endbuf: | |
4509 | case wordbeg: | |
4510 | case wordend: | |
4511 | case wordbound: | |
4512 | case notwordbound: | |
4513 | #ifdef emacs | |
4514 | case before_dot: | |
4515 | case at_dot: | |
4516 | case after_dot: | |
4517 | #endif | |
4518 | break; | |
4519 | ||
4520 | case start_memory: | |
4521 | reg_no = *p1; | |
4522 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
4523 | ret = group_match_null_string_p (&p1, end, reg_info); | |
4524 | ||
4525 | /* Have to set this here in case we're checking a group which | |
4526 | contains a group and a back reference to it. */ | |
4527 | ||
4528 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
4529 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
4530 | ||
4531 | if (!ret) | |
4532 | return false; | |
4533 | break; | |
4534 | ||
4535 | /* If this is an optimized succeed_n for zero times, make the jump. */ | |
4536 | case jump: | |
4537 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4538 | if (mcnt >= 0) | |
4539 | p1 += mcnt; | |
4540 | else | |
4541 | return false; | |
4542 | break; | |
4543 | ||
4544 | case succeed_n: | |
4545 | /* Get to the number of times to succeed. */ | |
4546 | p1 += 2; | |
4547 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4548 | ||
4549 | if (mcnt == 0) | |
4550 | { | |
4551 | p1 -= 4; | |
4552 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4553 | p1 += mcnt; | |
4554 | } | |
4555 | else | |
4556 | return false; | |
4557 | break; | |
4558 | ||
4559 | case duplicate: | |
4560 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
4561 | return false; | |
4562 | break; | |
4563 | ||
4564 | case set_number_at: | |
4565 | p1 += 4; | |
4566 | ||
4567 | default: | |
4568 | /* All other opcodes mean we cannot match the empty string. */ | |
4569 | return false; | |
4570 | } | |
4571 | ||
4572 | *p = p1; | |
4573 | return true; | |
4574 | } /* common_op_match_null_string_p */ | |
4575 | ||
4576 | ||
4577 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
4578 | bytes; nonzero otherwise. */ | |
4579 | ||
4580 | static int | |
4581 | bcmp_translate (s1, s2, len, translate) | |
4582 | unsigned char *s1, *s2; | |
4583 | register int len; | |
4584 | char *translate; | |
4585 | { | |
4586 | register unsigned char *p1 = s1, *p2 = s2; | |
4587 | while (len) | |
4588 | { | |
4589 | if (translate[*p1++] != translate[*p2++]) return 1; | |
4590 | len--; | |
4591 | } | |
4592 | return 0; | |
4593 | } | |
4594 | \f | |
4595 | /* Entry points for GNU code. */ | |
4596 | ||
4597 | /* re_compile_pattern is the GNU regular expression compiler: it | |
4598 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
4599 | Returns 0 if the pattern was valid, otherwise an error string. | |
4600 | ||
4601 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
4602 | are set in BUFP on entry. | |
4603 | ||
4604 | We call regex_compile to do the actual compilation. */ | |
4605 | ||
4606 | const char * | |
4607 | re_compile_pattern (pattern, length, bufp) | |
4608 | const char *pattern; | |
4609 | int length; | |
4610 | struct re_pattern_buffer *bufp; | |
4611 | { | |
4612 | reg_errcode_t ret; | |
4613 | ||
4614 | /* GNU code is written to assume at least RE_NREGS registers will be set | |
4615 | (and at least one extra will be -1). */ | |
4616 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4617 | ||
4618 | /* And GNU code determines whether or not to get register information | |
4619 | by passing null for the REGS argument to re_match, etc., not by | |
4620 | setting no_sub. */ | |
4621 | bufp->no_sub = 0; | |
4622 | ||
4623 | /* Match anchors at newline. */ | |
4624 | bufp->newline_anchor = 1; | |
4625 | ||
4626 | ret = regex_compile (pattern, length, re_syntax_options, bufp); | |
4627 | ||
4628 | return re_error_msg[(int) ret]; | |
4629 | } | |
4630 | \f | |
4631 | /* Entry points compatible with 4.2 BSD regex library. We don't define | |
4632 | them if this is an Emacs or POSIX compilation. */ | |
4633 | ||
4634 | #if !defined (emacs) && !defined (_POSIX_SOURCE) | |
4635 | ||
4636 | /* BSD has one and only one pattern buffer. */ | |
4637 | static struct re_pattern_buffer re_comp_buf; | |
4638 | ||
4639 | char * | |
4640 | re_comp (s) | |
4641 | const char *s; | |
4642 | { | |
4643 | reg_errcode_t ret; | |
4644 | ||
4645 | if (!s) | |
4646 | { | |
4647 | if (!re_comp_buf.buffer) | |
4648 | return "No previous regular expression"; | |
4649 | return 0; | |
4650 | } | |
4651 | ||
4652 | if (!re_comp_buf.buffer) | |
4653 | { | |
4654 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
4655 | if (re_comp_buf.buffer == NULL) | |
4656 | return "Memory exhausted"; | |
4657 | re_comp_buf.allocated = 200; | |
4658 | ||
4659 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
4660 | if (re_comp_buf.fastmap == NULL) | |
4661 | return "Memory exhausted"; | |
4662 | } | |
4663 | ||
4664 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
4665 | don't need to initialize the pattern buffer fields which affect it. */ | |
4666 | ||
4667 | /* Match anchors at newlines. */ | |
4668 | re_comp_buf.newline_anchor = 1; | |
4669 | ||
4670 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
4671 | ||
4672 | /* Yes, we're discarding `const' here. */ | |
4673 | return (char *) re_error_msg[(int) ret]; | |
4674 | } | |
4675 | ||
4676 | ||
4677 | int | |
4678 | re_exec (s) | |
4679 | const char *s; | |
4680 | { | |
4681 | const int len = strlen (s); | |
4682 | return | |
4683 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
4684 | } | |
4685 | #endif /* not emacs and not _POSIX_SOURCE */ | |
4686 | \f | |
4687 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
4688 | ||
4689 | #ifndef emacs | |
4690 | ||
4691 | /* regcomp takes a regular expression as a string and compiles it. | |
4692 | ||
4693 | PREG is a regex_t *. We do not expect any fields to be initialized, | |
4694 | since POSIX says we shouldn't. Thus, we set | |
4695 | ||
4696 | `buffer' to the compiled pattern; | |
4697 | `used' to the length of the compiled pattern; | |
4698 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
4699 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
4700 | RE_SYNTAX_POSIX_BASIC; | |
4701 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
4702 | `fastmap' and `fastmap_accurate' to zero; | |
4703 | `re_nsub' to the number of subexpressions in PATTERN. | |
4704 | ||
4705 | PATTERN is the address of the pattern string. | |
4706 | ||
4707 | CFLAGS is a series of bits which affect compilation. | |
4708 | ||
4709 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
4710 | use POSIX basic syntax. | |
4711 | ||
4712 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
4713 | Also, regexec will try a match beginning after every newline. | |
4714 | ||
4715 | If REG_ICASE is set, then we considers upper- and lowercase | |
4716 | versions of letters to be equivalent when matching. | |
4717 | ||
4718 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
4719 | routine will report only success or failure, and nothing about the | |
4720 | registers. | |
4721 | ||
4722 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
4723 | the return codes and their meanings.) */ | |
4724 | ||
4725 | int | |
4726 | regcomp (preg, pattern, cflags) | |
4727 | regex_t *preg; | |
4728 | const char *pattern; | |
4729 | int cflags; | |
4730 | { | |
4731 | reg_errcode_t ret; | |
4732 | unsigned syntax | |
d9088577 DM |
4733 | = (cflags & REG_EXTENDED) ? |
4734 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
bc78d348 KB |
4735 | |
4736 | /* regex_compile will allocate the space for the compiled pattern. */ | |
4737 | preg->buffer = 0; | |
d9088577 | 4738 | preg->allocated = 0; |
bc78d348 KB |
4739 | |
4740 | /* Don't bother to use a fastmap when searching. This simplifies the | |
4741 | REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
4742 | characters after newlines into the fastmap. This way, we just try | |
4743 | every character. */ | |
4744 | preg->fastmap = 0; | |
4745 | ||
4746 | if (cflags & REG_ICASE) | |
4747 | { | |
4748 | unsigned i; | |
4749 | ||
4750 | preg->translate = (char *) malloc (CHAR_SET_SIZE); | |
4751 | if (preg->translate == NULL) | |
4752 | return (int) REG_ESPACE; | |
4753 | ||
4754 | /* Map uppercase characters to corresponding lowercase ones. */ | |
4755 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
c6b40788 | 4756 | preg->translate[i] = ISUPPER (i) ? tolower (i) : i; |
bc78d348 KB |
4757 | } |
4758 | else | |
4759 | preg->translate = NULL; | |
4760 | ||
4761 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
4762 | if (cflags & REG_NEWLINE) | |
4763 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
4764 | syntax &= ~RE_DOT_NEWLINE; | |
4765 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
4766 | /* It also changes the matching behavior. */ | |
4767 | preg->newline_anchor = 1; | |
4768 | } | |
4769 | else | |
4770 | preg->newline_anchor = 0; | |
4771 | ||
4772 | preg->no_sub = !!(cflags & REG_NOSUB); | |
4773 | ||
4774 | /* POSIX says a null character in the pattern terminates it, so we | |
4775 | can use strlen here in compiling the pattern. */ | |
4776 | ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
4777 | ||
4778 | /* POSIX doesn't distinguish between an unmatched open-group and an | |
4779 | unmatched close-group: both are REG_EPAREN. */ | |
4780 | if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
4781 | ||
4782 | return (int) ret; | |
4783 | } | |
4784 | ||
4785 | ||
4786 | /* regexec searches for a given pattern, specified by PREG, in the | |
4787 | string STRING. | |
4788 | ||
4789 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
4790 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
4791 | least NMATCH elements, and we set them to the offsets of the | |
4792 | corresponding matched substrings. | |
4793 | ||
4794 | EFLAGS specifies `execution flags' which affect matching: if | |
4795 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
4796 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
4797 | ||
4798 | We return 0 if we find a match and REG_NOMATCH if not. */ | |
4799 | ||
4800 | int | |
4801 | regexec (preg, string, nmatch, pmatch, eflags) | |
4802 | const regex_t *preg; | |
4803 | const char *string; | |
4804 | size_t nmatch; | |
4805 | regmatch_t pmatch[]; | |
4806 | int eflags; | |
4807 | { | |
4808 | int ret; | |
4809 | struct re_registers regs; | |
4810 | regex_t private_preg; | |
4811 | int len = strlen (string); | |
4812 | boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
4813 | ||
4814 | private_preg = *preg; | |
4815 | ||
4816 | private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
4817 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
4818 | ||
4819 | /* The user has told us exactly how many registers to return | |
4820 | information about, via `nmatch'. We have to pass that on to the | |
4821 | matching routines. */ | |
4822 | private_preg.regs_allocated = REGS_FIXED; | |
4823 | ||
4824 | if (want_reg_info) | |
4825 | { | |
4826 | regs.num_regs = nmatch; | |
4827 | regs.start = TALLOC (nmatch, regoff_t); | |
4828 | regs.end = TALLOC (nmatch, regoff_t); | |
4829 | if (regs.start == NULL || regs.end == NULL) | |
4830 | return (int) REG_NOMATCH; | |
4831 | } | |
4832 | ||
4833 | /* Perform the searching operation. */ | |
4834 | ret = re_search (&private_preg, string, len, | |
4835 | /* start: */ 0, /* range: */ len, | |
4836 | want_reg_info ? ®s : (struct re_registers *) 0); | |
4837 | ||
4838 | /* Copy the register information to the POSIX structure. */ | |
4839 | if (want_reg_info) | |
4840 | { | |
4841 | if (ret >= 0) | |
4842 | { | |
4843 | unsigned r; | |
4844 | ||
4845 | for (r = 0; r < nmatch; r++) | |
4846 | { | |
4847 | pmatch[r].rm_so = regs.start[r]; | |
4848 | pmatch[r].rm_eo = regs.end[r]; | |
4849 | } | |
4850 | } | |
4851 | ||
4852 | /* If we needed the temporary register info, free the space now. */ | |
4853 | free (regs.start); | |
4854 | free (regs.end); | |
4855 | } | |
4856 | ||
4857 | /* We want zero return to mean success, unlike `re_search'. */ | |
4858 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
4859 | } | |
4860 | ||
4861 | ||
4862 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
9114e279 | 4863 | from either regcomp or regexec. We don't use PREG here. */ |
bc78d348 KB |
4864 | |
4865 | size_t | |
4866 | regerror (errcode, preg, errbuf, errbuf_size) | |
4867 | int errcode; | |
4868 | const regex_t *preg; | |
4869 | char *errbuf; | |
4870 | size_t errbuf_size; | |
4871 | { | |
85586484 JB |
4872 | const char *msg; |
4873 | size_t msg_size; | |
4874 | ||
4875 | if (errcode < 0 | |
4876 | || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0]))) | |
4877 | /* Only error codes returned by the rest of the code should be passed | |
4878 | to this routine. If we are given anything else, or if other regex | |
4879 | code generates an invalid error code, then the program has a bug. | |
4880 | Dump core so we can fix it. */ | |
4881 | abort (); | |
4882 | ||
4883 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
bc78d348 KB |
4884 | |
4885 | if (errbuf_size != 0) | |
4886 | { | |
4887 | if (msg_size > errbuf_size) | |
4888 | { | |
4889 | strncpy (errbuf, msg, errbuf_size - 1); | |
4890 | errbuf[errbuf_size - 1] = 0; | |
4891 | } | |
4892 | else | |
4893 | strcpy (errbuf, msg); | |
4894 | } | |
4895 | ||
4896 | return msg_size; | |
4897 | } | |
4898 | ||
4899 | ||
4900 | /* Free dynamically allocated space used by PREG. */ | |
4901 | ||
4902 | void | |
4903 | regfree (preg) | |
4904 | regex_t *preg; | |
4905 | { | |
4906 | if (preg->buffer != NULL) | |
4907 | free (preg->buffer); | |
4908 | preg->buffer = NULL; | |
4909 | ||
4910 | preg->allocated = 0; | |
4911 | preg->used = 0; | |
4912 | ||
4913 | if (preg->fastmap != NULL) | |
4914 | free (preg->fastmap); | |
4915 | preg->fastmap = NULL; | |
4916 | preg->fastmap_accurate = 0; | |
4917 | ||
4918 | if (preg->translate != NULL) | |
4919 | free (preg->translate); | |
4920 | preg->translate = NULL; | |
4921 | } | |
4922 | ||
4923 | #endif /* not emacs */ | |
4924 | \f | |
4925 | /* | |
4926 | Local variables: | |
4927 | make-backup-files: t | |
4928 | version-control: t | |
4929 | trim-versions-without-asking: nil | |
4930 | End: | |
4931 | */ |