Commit | Line | Data |
---|---|---|
bc78d348 | 1 | /* Extended regular expression matching and search library, |
8b20806d | 2 | version 0.12. |
bc78d348 KB |
3 | (Implements POSIX draft P10003.2/D11.2, except for |
4 | internationalization features.) | |
5 | ||
6cd14acb | 6 | Copyright (C) 1993, 1994 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 | ||
2f4e547f | 29 | #ifdef HAVE_CONFIG_H |
972a693a | 30 | #if defined (emacs) || defined (CONFIG_BROKETS) |
2f4e547f JM |
31 | /* We use <config.h> instead of "config.h" so that a compilation |
32 | using -I. -I$srcdir will use ./config.h rather than $srcdir/config.h | |
33 | (which it would do because it found this file in $srcdir). */ | |
34 | #include <config.h> | |
35 | #else | |
36 | #include "config.h" | |
37 | #endif | |
38 | #endif | |
39 | ||
bc78d348 KB |
40 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ |
41 | #include <sys/types.h> | |
42 | ||
43 | /* The `emacs' switch turns on certain matching commands | |
44 | that make sense only in Emacs. */ | |
45 | #ifdef emacs | |
46 | ||
bc78d348 KB |
47 | #include "lisp.h" |
48 | #include "buffer.h" | |
49 | #include "syntax.h" | |
50 | ||
51 | /* Emacs uses `NULL' as a predicate. */ | |
52 | #undef NULL | |
53 | ||
54 | #else /* not emacs */ | |
55 | ||
74da5007 JB |
56 | #ifdef STDC_HEADERS |
57 | #include <stdlib.h> | |
58 | #else | |
59 | char *malloc (); | |
60 | char *realloc (); | |
61 | #endif | |
62 | ||
63 | ||
bc78d348 KB |
64 | /* We used to test for `BSTRING' here, but only GCC and Emacs define |
65 | `BSTRING', as far as I know, and neither of them use this code. */ | |
3fa89e8c | 66 | #ifndef INHIBIT_STRING_HEADER |
18ec5b05 | 67 | #if HAVE_STRING_H || STDC_HEADERS |
bc78d348 | 68 | #include <string.h> |
9114e279 | 69 | #ifndef bcmp |
bc78d348 | 70 | #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n)) |
9114e279 KB |
71 | #endif |
72 | #ifndef bcopy | |
bc78d348 | 73 | #define bcopy(s, d, n) memcpy ((d), (s), (n)) |
9114e279 KB |
74 | #endif |
75 | #ifndef bzero | |
bc78d348 | 76 | #define bzero(s, n) memset ((s), 0, (n)) |
9114e279 | 77 | #endif |
bc78d348 KB |
78 | #else |
79 | #include <strings.h> | |
80 | #endif | |
3fa89e8c | 81 | #endif |
bc78d348 | 82 | |
bc78d348 KB |
83 | /* Define the syntax stuff for \<, \>, etc. */ |
84 | ||
85 | /* This must be nonzero for the wordchar and notwordchar pattern | |
86 | commands in re_match_2. */ | |
87 | #ifndef Sword | |
88 | #define Sword 1 | |
89 | #endif | |
90 | ||
91 | #ifdef SYNTAX_TABLE | |
92 | ||
93 | extern char *re_syntax_table; | |
94 | ||
95 | #else /* not SYNTAX_TABLE */ | |
96 | ||
97 | /* How many characters in the character set. */ | |
98 | #define CHAR_SET_SIZE 256 | |
99 | ||
100 | static char re_syntax_table[CHAR_SET_SIZE]; | |
101 | ||
102 | static void | |
103 | init_syntax_once () | |
104 | { | |
105 | register int c; | |
106 | static int done = 0; | |
107 | ||
108 | if (done) | |
109 | return; | |
110 | ||
111 | bzero (re_syntax_table, sizeof re_syntax_table); | |
112 | ||
113 | for (c = 'a'; c <= 'z'; c++) | |
114 | re_syntax_table[c] = Sword; | |
115 | ||
116 | for (c = 'A'; c <= 'Z'; c++) | |
117 | re_syntax_table[c] = Sword; | |
118 | ||
119 | for (c = '0'; c <= '9'; c++) | |
120 | re_syntax_table[c] = Sword; | |
121 | ||
122 | re_syntax_table['_'] = Sword; | |
123 | ||
124 | done = 1; | |
125 | } | |
126 | ||
127 | #endif /* not SYNTAX_TABLE */ | |
128 | ||
129 | #define SYNTAX(c) re_syntax_table[c] | |
130 | ||
131 | #endif /* not emacs */ | |
132 | \f | |
133 | /* Get the interface, including the syntax bits. */ | |
134 | #include "regex.h" | |
135 | ||
bc78d348 KB |
136 | /* isalpha etc. are used for the character classes. */ |
137 | #include <ctype.h> | |
c6b40788 | 138 | |
a1dd004b JB |
139 | /* Jim Meyering writes: |
140 | ||
141 | "... Some ctype macros are valid only for character codes that | |
142 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
143 | using /bin/cc or gcc but without giving an ansi option). So, all | |
144 | ctype uses should be through macros like ISPRINT... If | |
145 | STDC_HEADERS is defined, then autoconf has verified that the ctype | |
146 | macros don't need to be guarded with references to isascii. ... | |
147 | Defining isascii to 1 should let any compiler worth its salt | |
148 | eliminate the && through constant folding." */ | |
6b528f2f KH |
149 | |
150 | #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII)) | |
151 | #define ISASCII(c) 1 | |
152 | #else | |
153 | #define ISASCII(c) isascii(c) | |
bc78d348 | 154 | #endif |
c6b40788 JM |
155 | |
156 | #ifdef isblank | |
6b528f2f | 157 | #define ISBLANK(c) (ISASCII (c) && isblank (c)) |
c6b40788 JM |
158 | #else |
159 | #define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
bc78d348 | 160 | #endif |
c6b40788 | 161 | #ifdef isgraph |
6b528f2f | 162 | #define ISGRAPH(c) (ISASCII (c) && isgraph (c)) |
c6b40788 | 163 | #else |
6b528f2f | 164 | #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) |
c6b40788 JM |
165 | #endif |
166 | ||
6b528f2f KH |
167 | #define ISPRINT(c) (ISASCII (c) && isprint (c)) |
168 | #define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
169 | #define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
170 | #define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
171 | #define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
172 | #define ISLOWER(c) (ISASCII (c) && islower (c)) | |
173 | #define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
174 | #define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
175 | #define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
176 | #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
bc78d348 KB |
177 | |
178 | #ifndef NULL | |
179 | #define NULL 0 | |
180 | #endif | |
181 | ||
182 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
183 | since ours (we hope) works properly with all combinations of | |
184 | machines, compilers, `char' and `unsigned char' argument types. | |
185 | (Per Bothner suggested the basic approach.) */ | |
186 | #undef SIGN_EXTEND_CHAR | |
187 | #if __STDC__ | |
188 | #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
9114e279 | 189 | #else /* not __STDC__ */ |
bc78d348 KB |
190 | /* As in Harbison and Steele. */ |
191 | #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
192 | #endif | |
193 | \f | |
194 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
195 | use `alloca' instead of `malloc'. This is because using malloc in | |
196 | re_search* or re_match* could cause memory leaks when C-g is used in | |
197 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
198 | the other hand, malloc is more portable, and easier to debug. | |
199 | ||
200 | Because we sometimes use alloca, some routines have to be macros, | |
201 | not functions -- `alloca'-allocated space disappears at the end of the | |
202 | function it is called in. */ | |
203 | ||
204 | #ifdef REGEX_MALLOC | |
205 | ||
206 | #define REGEX_ALLOCATE malloc | |
207 | #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
208 | ||
209 | #else /* not REGEX_MALLOC */ | |
210 | ||
211 | /* Emacs already defines alloca, sometimes. */ | |
212 | #ifndef alloca | |
213 | ||
214 | /* Make alloca work the best possible way. */ | |
215 | #ifdef __GNUC__ | |
216 | #define alloca __builtin_alloca | |
217 | #else /* not __GNUC__ */ | |
218 | #if HAVE_ALLOCA_H | |
219 | #include <alloca.h> | |
220 | #else /* not __GNUC__ or HAVE_ALLOCA_H */ | |
221 | #ifndef _AIX /* Already did AIX, up at the top. */ | |
222 | char *alloca (); | |
223 | #endif /* not _AIX */ | |
224 | #endif /* not HAVE_ALLOCA_H */ | |
225 | #endif /* not __GNUC__ */ | |
226 | ||
227 | #endif /* not alloca */ | |
228 | ||
229 | #define REGEX_ALLOCATE alloca | |
230 | ||
231 | /* Assumes a `char *destination' variable. */ | |
232 | #define REGEX_REALLOCATE(source, osize, nsize) \ | |
233 | (destination = (char *) alloca (nsize), \ | |
234 | bcopy (source, destination, osize), \ | |
235 | destination) | |
236 | ||
237 | #endif /* not REGEX_MALLOC */ | |
238 | ||
239 | ||
240 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
241 | `string1' or just past its end. This works if PTR is NULL, which is | |
242 | a good thing. */ | |
243 | #define FIRST_STRING_P(ptr) \ | |
244 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
245 | ||
246 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
247 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
248 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
e959bada JB |
249 | #define RETALLOC_IF(addr, n, t) \ |
250 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
bc78d348 KB |
251 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) |
252 | ||
253 | #define BYTEWIDTH 8 /* In bits. */ | |
254 | ||
255 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
256 | ||
63268f76 KH |
257 | #undef MAX |
258 | #undef MIN | |
bc78d348 KB |
259 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) |
260 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
261 | ||
262 | typedef char boolean; | |
263 | #define false 0 | |
264 | #define true 1 | |
f765858e KH |
265 | |
266 | static int re_match_2_internal (); | |
bc78d348 KB |
267 | \f |
268 | /* These are the command codes that appear in compiled regular | |
269 | expressions. Some opcodes are followed by argument bytes. A | |
270 | command code can specify any interpretation whatsoever for its | |
271 | arguments. Zero bytes may appear in the compiled regular expression. | |
272 | ||
273 | The value of `exactn' is needed in search.c (search_buffer) in Emacs. | |
274 | So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of | |
275 | `exactn' we use here must also be 1. */ | |
276 | ||
277 | typedef enum | |
278 | { | |
279 | no_op = 0, | |
280 | ||
281 | /* Followed by one byte giving n, then by n literal bytes. */ | |
282 | exactn = 1, | |
283 | ||
284 | /* Matches any (more or less) character. */ | |
285 | anychar, | |
286 | ||
287 | /* Matches any one char belonging to specified set. First | |
288 | following byte is number of bitmap bytes. Then come bytes | |
289 | for a bitmap saying which chars are in. Bits in each byte | |
290 | are ordered low-bit-first. A character is in the set if its | |
291 | bit is 1. A character too large to have a bit in the map is | |
292 | automatically not in the set. */ | |
293 | charset, | |
294 | ||
295 | /* Same parameters as charset, but match any character that is | |
296 | not one of those specified. */ | |
297 | charset_not, | |
298 | ||
299 | /* Start remembering the text that is matched, for storing in a | |
300 | register. Followed by one byte with the register number, in | |
301 | the range 0 to one less than the pattern buffer's re_nsub | |
302 | field. Then followed by one byte with the number of groups | |
303 | inner to this one. (This last has to be part of the | |
304 | start_memory only because we need it in the on_failure_jump | |
305 | of re_match_2.) */ | |
306 | start_memory, | |
307 | ||
308 | /* Stop remembering the text that is matched and store it in a | |
309 | memory register. Followed by one byte with the register | |
310 | number, in the range 0 to one less than `re_nsub' in the | |
311 | pattern buffer, and one byte with the number of inner groups, | |
312 | just like `start_memory'. (We need the number of inner | |
313 | groups here because we don't have any easy way of finding the | |
314 | corresponding start_memory when we're at a stop_memory.) */ | |
315 | stop_memory, | |
316 | ||
317 | /* Match a duplicate of something remembered. Followed by one | |
318 | byte containing the register number. */ | |
319 | duplicate, | |
320 | ||
321 | /* Fail unless at beginning of line. */ | |
322 | begline, | |
323 | ||
324 | /* Fail unless at end of line. */ | |
325 | endline, | |
326 | ||
327 | /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
328 | of string to be matched (if not). */ | |
329 | begbuf, | |
330 | ||
331 | /* Analogously, for end of buffer/string. */ | |
332 | endbuf, | |
333 | ||
334 | /* Followed by two byte relative address to which to jump. */ | |
335 | jump, | |
336 | ||
337 | /* Same as jump, but marks the end of an alternative. */ | |
338 | jump_past_alt, | |
339 | ||
340 | /* Followed by two-byte relative address of place to resume at | |
341 | in case of failure. */ | |
342 | on_failure_jump, | |
343 | ||
344 | /* Like on_failure_jump, but pushes a placeholder instead of the | |
345 | current string position when executed. */ | |
346 | on_failure_keep_string_jump, | |
347 | ||
348 | /* Throw away latest failure point and then jump to following | |
349 | two-byte relative address. */ | |
350 | pop_failure_jump, | |
351 | ||
352 | /* Change to pop_failure_jump if know won't have to backtrack to | |
353 | match; otherwise change to jump. This is used to jump | |
354 | back to the beginning of a repeat. If what follows this jump | |
355 | clearly won't match what the repeat does, such that we can be | |
356 | sure that there is no use backtracking out of repetitions | |
357 | already matched, then we change it to a pop_failure_jump. | |
358 | Followed by two-byte address. */ | |
359 | maybe_pop_jump, | |
360 | ||
361 | /* Jump to following two-byte address, and push a dummy failure | |
362 | point. This failure point will be thrown away if an attempt | |
363 | is made to use it for a failure. A `+' construct makes this | |
364 | before the first repeat. Also used as an intermediary kind | |
365 | of jump when compiling an alternative. */ | |
366 | dummy_failure_jump, | |
367 | ||
368 | /* Push a dummy failure point and continue. Used at the end of | |
369 | alternatives. */ | |
370 | push_dummy_failure, | |
371 | ||
372 | /* Followed by two-byte relative address and two-byte number n. | |
373 | After matching N times, jump to the address upon failure. */ | |
374 | succeed_n, | |
375 | ||
376 | /* Followed by two-byte relative address, and two-byte number n. | |
377 | Jump to the address N times, then fail. */ | |
378 | jump_n, | |
379 | ||
380 | /* Set the following two-byte relative address to the | |
381 | subsequent two-byte number. The address *includes* the two | |
382 | bytes of number. */ | |
383 | set_number_at, | |
384 | ||
385 | wordchar, /* Matches any word-constituent character. */ | |
386 | notwordchar, /* Matches any char that is not a word-constituent. */ | |
387 | ||
388 | wordbeg, /* Succeeds if at word beginning. */ | |
389 | wordend, /* Succeeds if at word end. */ | |
390 | ||
391 | wordbound, /* Succeeds if at a word boundary. */ | |
392 | notwordbound /* Succeeds if not at a word boundary. */ | |
393 | ||
394 | #ifdef emacs | |
395 | ,before_dot, /* Succeeds if before point. */ | |
396 | at_dot, /* Succeeds if at point. */ | |
397 | after_dot, /* Succeeds if after point. */ | |
398 | ||
399 | /* Matches any character whose syntax is specified. Followed by | |
400 | a byte which contains a syntax code, e.g., Sword. */ | |
401 | syntaxspec, | |
402 | ||
403 | /* Matches any character whose syntax is not that specified. */ | |
404 | notsyntaxspec | |
405 | #endif /* emacs */ | |
406 | } re_opcode_t; | |
407 | \f | |
408 | /* Common operations on the compiled pattern. */ | |
409 | ||
410 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
411 | ||
412 | #define STORE_NUMBER(destination, number) \ | |
413 | do { \ | |
414 | (destination)[0] = (number) & 0377; \ | |
415 | (destination)[1] = (number) >> 8; \ | |
416 | } while (0) | |
417 | ||
418 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
419 | the byte after where the number is stored. Therefore, DESTINATION | |
420 | must be an lvalue. */ | |
421 | ||
422 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
423 | do { \ | |
424 | STORE_NUMBER (destination, number); \ | |
425 | (destination) += 2; \ | |
426 | } while (0) | |
427 | ||
428 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
429 | at SOURCE. */ | |
430 | ||
431 | #define EXTRACT_NUMBER(destination, source) \ | |
432 | do { \ | |
433 | (destination) = *(source) & 0377; \ | |
434 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
435 | } while (0) | |
436 | ||
437 | #ifdef DEBUG | |
438 | static void | |
439 | extract_number (dest, source) | |
440 | int *dest; | |
441 | unsigned char *source; | |
442 | { | |
443 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
444 | *dest = *source & 0377; | |
445 | *dest += temp << 8; | |
446 | } | |
447 | ||
448 | #ifndef EXTRACT_MACROS /* To debug the macros. */ | |
449 | #undef EXTRACT_NUMBER | |
450 | #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
451 | #endif /* not EXTRACT_MACROS */ | |
452 | ||
453 | #endif /* DEBUG */ | |
454 | ||
455 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
456 | SOURCE must be an lvalue. */ | |
457 | ||
458 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
459 | do { \ | |
460 | EXTRACT_NUMBER (destination, source); \ | |
461 | (source) += 2; \ | |
462 | } while (0) | |
463 | ||
464 | #ifdef DEBUG | |
465 | static void | |
466 | extract_number_and_incr (destination, source) | |
467 | int *destination; | |
468 | unsigned char **source; | |
469 | { | |
470 | extract_number (destination, *source); | |
471 | *source += 2; | |
472 | } | |
473 | ||
474 | #ifndef EXTRACT_MACROS | |
475 | #undef EXTRACT_NUMBER_AND_INCR | |
476 | #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
477 | extract_number_and_incr (&dest, &src) | |
478 | #endif /* not EXTRACT_MACROS */ | |
479 | ||
480 | #endif /* DEBUG */ | |
481 | \f | |
482 | /* If DEBUG is defined, Regex prints many voluminous messages about what | |
483 | it is doing (if the variable `debug' is nonzero). If linked with the | |
484 | main program in `iregex.c', you can enter patterns and strings | |
485 | interactively. And if linked with the main program in `main.c' and | |
486 | the other test files, you can run the already-written tests. */ | |
487 | ||
488 | #ifdef DEBUG | |
489 | ||
490 | /* We use standard I/O for debugging. */ | |
491 | #include <stdio.h> | |
492 | ||
493 | /* It is useful to test things that ``must'' be true when debugging. */ | |
494 | #include <assert.h> | |
495 | ||
496 | static int debug = 0; | |
497 | ||
498 | #define DEBUG_STATEMENT(e) e | |
499 | #define DEBUG_PRINT1(x) if (debug) printf (x) | |
500 | #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
501 | #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
9114e279 | 502 | #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) |
bc78d348 KB |
503 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
504 | if (debug) print_partial_compiled_pattern (s, e) | |
505 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
506 | if (debug) print_double_string (w, s1, sz1, s2, sz2) | |
507 | ||
508 | ||
509 | extern void printchar (); | |
510 | ||
511 | /* Print the fastmap in human-readable form. */ | |
512 | ||
513 | void | |
514 | print_fastmap (fastmap) | |
515 | char *fastmap; | |
516 | { | |
517 | unsigned was_a_range = 0; | |
518 | unsigned i = 0; | |
519 | ||
520 | while (i < (1 << BYTEWIDTH)) | |
521 | { | |
522 | if (fastmap[i++]) | |
523 | { | |
524 | was_a_range = 0; | |
525 | printchar (i - 1); | |
526 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
527 | { | |
528 | was_a_range = 1; | |
529 | i++; | |
530 | } | |
531 | if (was_a_range) | |
532 | { | |
533 | printf ("-"); | |
534 | printchar (i - 1); | |
535 | } | |
536 | } | |
537 | } | |
538 | putchar ('\n'); | |
539 | } | |
540 | ||
541 | ||
542 | /* Print a compiled pattern string in human-readable form, starting at | |
543 | the START pointer into it and ending just before the pointer END. */ | |
544 | ||
545 | void | |
546 | print_partial_compiled_pattern (start, end) | |
547 | unsigned char *start; | |
548 | unsigned char *end; | |
549 | { | |
550 | int mcnt, mcnt2; | |
551 | unsigned char *p = start; | |
552 | unsigned char *pend = end; | |
553 | ||
554 | if (start == NULL) | |
555 | { | |
556 | printf ("(null)\n"); | |
557 | return; | |
558 | } | |
559 | ||
560 | /* Loop over pattern commands. */ | |
561 | while (p < pend) | |
562 | { | |
b0bef8c5 JB |
563 | printf ("%d:\t", p - start); |
564 | ||
bc78d348 KB |
565 | switch ((re_opcode_t) *p++) |
566 | { | |
567 | case no_op: | |
568 | printf ("/no_op"); | |
569 | break; | |
570 | ||
571 | case exactn: | |
572 | mcnt = *p++; | |
573 | printf ("/exactn/%d", mcnt); | |
574 | do | |
575 | { | |
576 | putchar ('/'); | |
577 | printchar (*p++); | |
578 | } | |
579 | while (--mcnt); | |
580 | break; | |
581 | ||
582 | case start_memory: | |
583 | mcnt = *p++; | |
584 | printf ("/start_memory/%d/%d", mcnt, *p++); | |
585 | break; | |
586 | ||
587 | case stop_memory: | |
588 | mcnt = *p++; | |
589 | printf ("/stop_memory/%d/%d", mcnt, *p++); | |
590 | break; | |
591 | ||
592 | case duplicate: | |
593 | printf ("/duplicate/%d", *p++); | |
594 | break; | |
595 | ||
596 | case anychar: | |
597 | printf ("/anychar"); | |
598 | break; | |
599 | ||
600 | case charset: | |
601 | case charset_not: | |
602 | { | |
b0bef8c5 JB |
603 | register int c, last = -100; |
604 | register int in_range = 0; | |
bc78d348 | 605 | |
b0bef8c5 JB |
606 | printf ("/charset [%s", |
607 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); | |
bc78d348 KB |
608 | |
609 | assert (p + *p < pend); | |
610 | ||
b0bef8c5 JB |
611 | for (c = 0; c < 256; c++) |
612 | if (c / 8 < *p | |
613 | && (p[1 + (c/8)] & (1 << (c % 8)))) | |
614 | { | |
615 | /* Are we starting a range? */ | |
616 | if (last + 1 == c && ! in_range) | |
617 | { | |
618 | putchar ('-'); | |
619 | in_range = 1; | |
620 | } | |
621 | /* Have we broken a range? */ | |
622 | else if (last + 1 != c && in_range) | |
bc78d348 | 623 | { |
b0bef8c5 JB |
624 | printchar (last); |
625 | in_range = 0; | |
626 | } | |
bc78d348 | 627 | |
b0bef8c5 JB |
628 | if (! in_range) |
629 | printchar (c); | |
bc78d348 | 630 | |
b0bef8c5 | 631 | last = c; |
bc78d348 | 632 | } |
b0bef8c5 JB |
633 | |
634 | if (in_range) | |
635 | printchar (last); | |
636 | ||
637 | putchar (']'); | |
638 | ||
bc78d348 | 639 | p += 1 + *p; |
bc78d348 | 640 | } |
b0bef8c5 | 641 | break; |
bc78d348 KB |
642 | |
643 | case begline: | |
644 | printf ("/begline"); | |
645 | break; | |
646 | ||
647 | case endline: | |
648 | printf ("/endline"); | |
649 | break; | |
650 | ||
651 | case on_failure_jump: | |
652 | extract_number_and_incr (&mcnt, &p); | |
b0bef8c5 | 653 | printf ("/on_failure_jump to %d", p + mcnt - start); |
bc78d348 KB |
654 | break; |
655 | ||
656 | case on_failure_keep_string_jump: | |
657 | extract_number_and_incr (&mcnt, &p); | |
b0bef8c5 | 658 | printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); |
bc78d348 KB |
659 | break; |
660 | ||
661 | case dummy_failure_jump: | |
662 | extract_number_and_incr (&mcnt, &p); | |
b0bef8c5 | 663 | printf ("/dummy_failure_jump to %d", p + mcnt - start); |
bc78d348 KB |
664 | break; |
665 | ||
666 | case push_dummy_failure: | |
667 | printf ("/push_dummy_failure"); | |
668 | break; | |
669 | ||
670 | case maybe_pop_jump: | |
671 | extract_number_and_incr (&mcnt, &p); | |
b0bef8c5 | 672 | printf ("/maybe_pop_jump to %d", p + mcnt - start); |
bc78d348 KB |
673 | break; |
674 | ||
675 | case pop_failure_jump: | |
676 | extract_number_and_incr (&mcnt, &p); | |
b0bef8c5 | 677 | printf ("/pop_failure_jump to %d", p + mcnt - start); |
bc78d348 KB |
678 | break; |
679 | ||
680 | case jump_past_alt: | |
681 | extract_number_and_incr (&mcnt, &p); | |
b0bef8c5 | 682 | printf ("/jump_past_alt to %d", p + mcnt - start); |
bc78d348 KB |
683 | break; |
684 | ||
685 | case jump: | |
686 | extract_number_and_incr (&mcnt, &p); | |
b0bef8c5 | 687 | printf ("/jump to %d", p + mcnt - start); |
bc78d348 KB |
688 | break; |
689 | ||
690 | case succeed_n: | |
691 | extract_number_and_incr (&mcnt, &p); | |
692 | extract_number_and_incr (&mcnt2, &p); | |
b0bef8c5 | 693 | printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2); |
bc78d348 KB |
694 | break; |
695 | ||
696 | case jump_n: | |
697 | extract_number_and_incr (&mcnt, &p); | |
698 | extract_number_and_incr (&mcnt2, &p); | |
b0bef8c5 | 699 | printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2); |
bc78d348 KB |
700 | break; |
701 | ||
702 | case set_number_at: | |
703 | extract_number_and_incr (&mcnt, &p); | |
704 | extract_number_and_incr (&mcnt2, &p); | |
b0bef8c5 | 705 | printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2); |
bc78d348 KB |
706 | break; |
707 | ||
708 | case wordbound: | |
709 | printf ("/wordbound"); | |
710 | break; | |
711 | ||
712 | case notwordbound: | |
713 | printf ("/notwordbound"); | |
714 | break; | |
715 | ||
716 | case wordbeg: | |
717 | printf ("/wordbeg"); | |
718 | break; | |
719 | ||
720 | case wordend: | |
721 | printf ("/wordend"); | |
722 | ||
723 | #ifdef emacs | |
724 | case before_dot: | |
725 | printf ("/before_dot"); | |
726 | break; | |
727 | ||
728 | case at_dot: | |
729 | printf ("/at_dot"); | |
730 | break; | |
731 | ||
732 | case after_dot: | |
733 | printf ("/after_dot"); | |
734 | break; | |
735 | ||
736 | case syntaxspec: | |
737 | printf ("/syntaxspec"); | |
738 | mcnt = *p++; | |
739 | printf ("/%d", mcnt); | |
740 | break; | |
741 | ||
742 | case notsyntaxspec: | |
743 | printf ("/notsyntaxspec"); | |
744 | mcnt = *p++; | |
745 | printf ("/%d", mcnt); | |
746 | break; | |
747 | #endif /* emacs */ | |
748 | ||
749 | case wordchar: | |
750 | printf ("/wordchar"); | |
751 | break; | |
752 | ||
753 | case notwordchar: | |
754 | printf ("/notwordchar"); | |
755 | break; | |
756 | ||
757 | case begbuf: | |
758 | printf ("/begbuf"); | |
759 | break; | |
760 | ||
761 | case endbuf: | |
762 | printf ("/endbuf"); | |
763 | break; | |
764 | ||
765 | default: | |
766 | printf ("?%d", *(p-1)); | |
767 | } | |
b0bef8c5 JB |
768 | |
769 | putchar ('\n'); | |
bc78d348 | 770 | } |
b0bef8c5 JB |
771 | |
772 | printf ("%d:\tend of pattern.\n", p - start); | |
bc78d348 KB |
773 | } |
774 | ||
775 | ||
776 | void | |
777 | print_compiled_pattern (bufp) | |
778 | struct re_pattern_buffer *bufp; | |
779 | { | |
780 | unsigned char *buffer = bufp->buffer; | |
781 | ||
782 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
783 | printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); | |
784 | ||
785 | if (bufp->fastmap_accurate && bufp->fastmap) | |
786 | { | |
787 | printf ("fastmap: "); | |
788 | print_fastmap (bufp->fastmap); | |
789 | } | |
790 | ||
791 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
792 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
793 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
794 | printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
795 | printf ("no_sub: %d\t", bufp->no_sub); | |
796 | printf ("not_bol: %d\t", bufp->not_bol); | |
797 | printf ("not_eol: %d\t", bufp->not_eol); | |
798 | printf ("syntax: %d\n", bufp->syntax); | |
799 | /* Perhaps we should print the translate table? */ | |
800 | } | |
801 | ||
802 | ||
803 | void | |
804 | print_double_string (where, string1, size1, string2, size2) | |
805 | const char *where; | |
806 | const char *string1; | |
807 | const char *string2; | |
808 | int size1; | |
809 | int size2; | |
810 | { | |
811 | unsigned this_char; | |
812 | ||
813 | if (where == NULL) | |
814 | printf ("(null)"); | |
815 | else | |
816 | { | |
817 | if (FIRST_STRING_P (where)) | |
818 | { | |
819 | for (this_char = where - string1; this_char < size1; this_char++) | |
820 | printchar (string1[this_char]); | |
821 | ||
822 | where = string2; | |
823 | } | |
824 | ||
825 | for (this_char = where - string2; this_char < size2; this_char++) | |
826 | printchar (string2[this_char]); | |
827 | } | |
828 | } | |
829 | ||
830 | #else /* not DEBUG */ | |
831 | ||
832 | #undef assert | |
833 | #define assert(e) | |
834 | ||
835 | #define DEBUG_STATEMENT(e) | |
836 | #define DEBUG_PRINT1(x) | |
837 | #define DEBUG_PRINT2(x1, x2) | |
838 | #define DEBUG_PRINT3(x1, x2, x3) | |
9114e279 | 839 | #define DEBUG_PRINT4(x1, x2, x3, x4) |
bc78d348 KB |
840 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
841 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
842 | ||
843 | #endif /* not DEBUG */ | |
844 | \f | |
845 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
846 | also be assigned to arbitrarily: each pattern buffer stores its own | |
847 | syntax, so it can be changed between regex compilations. */ | |
848 | reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS; | |
849 | ||
850 | ||
851 | /* Specify the precise syntax of regexps for compilation. This provides | |
852 | for compatibility for various utilities which historically have | |
853 | different, incompatible syntaxes. | |
854 | ||
855 | The argument SYNTAX is a bit mask comprised of the various bits | |
856 | defined in regex.h. We return the old syntax. */ | |
857 | ||
858 | reg_syntax_t | |
859 | re_set_syntax (syntax) | |
860 | reg_syntax_t syntax; | |
861 | { | |
862 | reg_syntax_t ret = re_syntax_options; | |
863 | ||
864 | re_syntax_options = syntax; | |
865 | return ret; | |
866 | } | |
867 | \f | |
868 | /* This table gives an error message for each of the error codes listed | |
869 | in regex.h. Obviously the order here has to be same as there. */ | |
870 | ||
871 | static const char *re_error_msg[] = | |
872 | { NULL, /* REG_NOERROR */ | |
873 | "No match", /* REG_NOMATCH */ | |
874 | "Invalid regular expression", /* REG_BADPAT */ | |
875 | "Invalid collation character", /* REG_ECOLLATE */ | |
876 | "Invalid character class name", /* REG_ECTYPE */ | |
877 | "Trailing backslash", /* REG_EESCAPE */ | |
878 | "Invalid back reference", /* REG_ESUBREG */ | |
879 | "Unmatched [ or [^", /* REG_EBRACK */ | |
880 | "Unmatched ( or \\(", /* REG_EPAREN */ | |
881 | "Unmatched \\{", /* REG_EBRACE */ | |
882 | "Invalid content of \\{\\}", /* REG_BADBR */ | |
883 | "Invalid range end", /* REG_ERANGE */ | |
884 | "Memory exhausted", /* REG_ESPACE */ | |
885 | "Invalid preceding regular expression", /* REG_BADRPT */ | |
886 | "Premature end of regular expression", /* REG_EEND */ | |
887 | "Regular expression too big", /* REG_ESIZE */ | |
888 | "Unmatched ) or \\)", /* REG_ERPAREN */ | |
889 | }; | |
890 | \f | |
e959bada JB |
891 | /* Avoiding alloca during matching, to placate r_alloc. */ |
892 | ||
501b568d | 893 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the |
e959bada JB |
894 | searching and matching functions should not call alloca. On some |
895 | systems, alloca is implemented in terms of malloc, and if we're | |
896 | using the relocating allocator routines, then malloc could cause a | |
897 | relocation, which might (if the strings being searched are in the | |
898 | ralloc heap) shift the data out from underneath the regexp | |
85be3101 JB |
899 | routines. |
900 | ||
d0a11f95 RS |
901 | Here's another reason to avoid allocation: Emacs |
902 | processes input from X in a signal handler; processing X input may | |
85be3101 JB |
903 | call malloc; if input arrives while a matching routine is calling |
904 | malloc, then we're scrod. But Emacs can't just block input while | |
905 | calling matching routines; then we don't notice interrupts when | |
906 | they come in. So, Emacs blocks input around all regexp calls | |
907 | except the matching calls, which it leaves unprotected, in the | |
908 | faith that they will not malloc. */ | |
a822d92f JB |
909 | |
910 | /* Normally, this is fine. */ | |
911 | #define MATCH_MAY_ALLOCATE | |
912 | ||
d0a11f95 RS |
913 | /* The match routines may not allocate if (1) they would do it with malloc |
914 | and (2) it's not safe for htem to use malloc. */ | |
915 | #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC)) | |
a822d92f | 916 | #undef MATCH_MAY_ALLOCATE |
e959bada | 917 | #endif |
bc78d348 | 918 | |
e959bada JB |
919 | \f |
920 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
921 | re_match_2 use a failure stack. These have to be macros because of | |
922 | REGEX_ALLOCATE. */ | |
923 | ||
bc78d348 | 924 | |
e959bada JB |
925 | /* Number of failure points for which to initially allocate space |
926 | when matching. If this number is exceeded, we allocate more | |
927 | space, so it is not a hard limit. */ | |
928 | #ifndef INIT_FAILURE_ALLOC | |
929 | #define INIT_FAILURE_ALLOC 5 | |
930 | #endif | |
bc78d348 | 931 | |
e959bada JB |
932 | /* Roughly the maximum number of failure points on the stack. Would be |
933 | exactly that if always used MAX_FAILURE_SPACE each time we failed. | |
934 | This is a variable only so users of regex can assign to it; we never | |
935 | change it ourselves. */ | |
936 | int re_max_failures = 2000; | |
bc78d348 | 937 | |
794ce2c1 | 938 | typedef unsigned char *fail_stack_elt_t; |
bc78d348 | 939 | |
e959bada JB |
940 | typedef struct |
941 | { | |
942 | fail_stack_elt_t *stack; | |
943 | unsigned size; | |
944 | unsigned avail; /* Offset of next open position. */ | |
945 | } fail_stack_type; | |
bc78d348 | 946 | |
e959bada JB |
947 | #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) |
948 | #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
949 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
950 | #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail]) | |
bc78d348 | 951 | |
bc78d348 | 952 | |
e959bada | 953 | /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */ |
bc78d348 | 954 | |
a822d92f | 955 | #ifdef MATCH_MAY_ALLOCATE |
e959bada | 956 | #define INIT_FAIL_STACK() \ |
bc78d348 | 957 | do { \ |
e959bada JB |
958 | fail_stack.stack = (fail_stack_elt_t *) \ |
959 | REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ | |
960 | \ | |
961 | if (fail_stack.stack == NULL) \ | |
962 | return -2; \ | |
963 | \ | |
964 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
965 | fail_stack.avail = 0; \ | |
bc78d348 | 966 | } while (0) |
e959bada JB |
967 | #else |
968 | #define INIT_FAIL_STACK() \ | |
bc78d348 | 969 | do { \ |
e959bada | 970 | fail_stack.avail = 0; \ |
bc78d348 | 971 | } while (0) |
e959bada | 972 | #endif |
bc78d348 KB |
973 | |
974 | ||
e959bada | 975 | /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. |
bc78d348 | 976 | |
e959bada JB |
977 | Return 1 if succeeds, and 0 if either ran out of memory |
978 | allocating space for it or it was already too large. | |
979 | ||
980 | REGEX_REALLOCATE requires `destination' be declared. */ | |
bc78d348 | 981 | |
e959bada JB |
982 | #define DOUBLE_FAIL_STACK(fail_stack) \ |
983 | ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ | |
984 | ? 0 \ | |
985 | : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
986 | REGEX_REALLOCATE ((fail_stack).stack, \ | |
987 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
988 | ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
989 | \ | |
990 | (fail_stack).stack == NULL \ | |
991 | ? 0 \ | |
992 | : ((fail_stack).size <<= 1, \ | |
993 | 1))) | |
bc78d348 | 994 | |
bc78d348 | 995 | |
e959bada | 996 | /* Push PATTERN_OP on FAIL_STACK. |
bc78d348 | 997 | |
e959bada JB |
998 | Return 1 if was able to do so and 0 if ran out of memory allocating |
999 | space to do so. */ | |
1000 | #define PUSH_PATTERN_OP(pattern_op, fail_stack) \ | |
1001 | ((FAIL_STACK_FULL () \ | |
1002 | && !DOUBLE_FAIL_STACK (fail_stack)) \ | |
1003 | ? 0 \ | |
1004 | : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \ | |
1005 | 1)) | |
bc78d348 | 1006 | |
e959bada JB |
1007 | /* This pushes an item onto the failure stack. Must be a four-byte |
1008 | value. Assumes the variable `fail_stack'. Probably should only | |
1009 | be called from within `PUSH_FAILURE_POINT'. */ | |
1010 | #define PUSH_FAILURE_ITEM(item) \ | |
1011 | fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item | |
bc78d348 | 1012 | |
e959bada JB |
1013 | /* The complement operation. Assumes `fail_stack' is nonempty. */ |
1014 | #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail] | |
bc78d348 | 1015 | |
e959bada JB |
1016 | /* Used to omit pushing failure point id's when we're not debugging. */ |
1017 | #ifdef DEBUG | |
1018 | #define DEBUG_PUSH PUSH_FAILURE_ITEM | |
1019 | #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM () | |
1020 | #else | |
1021 | #define DEBUG_PUSH(item) | |
1022 | #define DEBUG_POP(item_addr) | |
1023 | #endif | |
bc78d348 | 1024 | |
bc78d348 | 1025 | |
e959bada JB |
1026 | /* Push the information about the state we will need |
1027 | if we ever fail back to it. | |
1028 | ||
1029 | Requires variables fail_stack, regstart, regend, reg_info, and | |
1030 | num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be | |
1031 | declared. | |
1032 | ||
1033 | Does `return FAILURE_CODE' if runs out of memory. */ | |
bc78d348 | 1034 | |
e959bada JB |
1035 | #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ |
1036 | do { \ | |
1037 | char *destination; \ | |
1038 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1039 | of 0 + -1 isn't done as unsigned. */ \ | |
1040 | int this_reg; \ | |
1041 | \ | |
1042 | DEBUG_STATEMENT (failure_id++); \ | |
1043 | DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
1044 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
1045 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
1046 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1047 | \ | |
1048 | DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ | |
1049 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ | |
1050 | \ | |
1051 | /* Ensure we have enough space allocated for what we will push. */ \ | |
1052 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
1053 | { \ | |
1054 | if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
1055 | return failure_code; \ | |
1056 | \ | |
1057 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
1058 | (fail_stack).size); \ | |
1059 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1060 | } \ | |
1061 | \ | |
1062 | /* Push the info, starting with the registers. */ \ | |
1063 | DEBUG_PRINT1 ("\n"); \ | |
1064 | \ | |
1065 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ | |
1066 | this_reg++) \ | |
1067 | { \ | |
1068 | DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ | |
1069 | DEBUG_STATEMENT (num_regs_pushed++); \ | |
1070 | \ | |
1071 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
1072 | PUSH_FAILURE_ITEM (regstart[this_reg]); \ | |
1073 | \ | |
1074 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
1075 | PUSH_FAILURE_ITEM (regend[this_reg]); \ | |
1076 | \ | |
1077 | DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ | |
1078 | DEBUG_PRINT2 (" match_null=%d", \ | |
1079 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
1080 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
1081 | DEBUG_PRINT2 (" matched_something=%d", \ | |
1082 | MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1083 | DEBUG_PRINT2 (" ever_matched=%d", \ | |
1084 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1085 | DEBUG_PRINT1 ("\n"); \ | |
1086 | PUSH_FAILURE_ITEM (reg_info[this_reg].word); \ | |
1087 | } \ | |
1088 | \ | |
1089 | DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ | |
1090 | PUSH_FAILURE_ITEM (lowest_active_reg); \ | |
1091 | \ | |
1092 | DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ | |
1093 | PUSH_FAILURE_ITEM (highest_active_reg); \ | |
1094 | \ | |
1095 | DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ | |
1096 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
1097 | PUSH_FAILURE_ITEM (pattern_place); \ | |
1098 | \ | |
1099 | DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ | |
1100 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
1101 | size2); \ | |
1102 | DEBUG_PRINT1 ("'\n"); \ | |
1103 | PUSH_FAILURE_ITEM (string_place); \ | |
1104 | \ | |
1105 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
1106 | DEBUG_PUSH (failure_id); \ | |
1107 | } while (0) | |
1108 | ||
1109 | /* This is the number of items that are pushed and popped on the stack | |
1110 | for each register. */ | |
1111 | #define NUM_REG_ITEMS 3 | |
1112 | ||
1113 | /* Individual items aside from the registers. */ | |
1114 | #ifdef DEBUG | |
1115 | #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
1116 | #else | |
1117 | #define NUM_NONREG_ITEMS 4 | |
1118 | #endif | |
1119 | ||
1120 | /* We push at most this many items on the stack. */ | |
1121 | #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
1122 | ||
1123 | /* We actually push this many items. */ | |
1124 | #define NUM_FAILURE_ITEMS \ | |
1125 | ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \ | |
1126 | + NUM_NONREG_ITEMS) | |
1127 | ||
1128 | /* How many items can still be added to the stack without overflowing it. */ | |
1129 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1130 | ||
1131 | ||
1132 | /* Pops what PUSH_FAIL_STACK pushes. | |
1133 | ||
1134 | We restore into the parameters, all of which should be lvalues: | |
1135 | STR -- the saved data position. | |
1136 | PAT -- the saved pattern position. | |
1137 | LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
1138 | REGSTART, REGEND -- arrays of string positions. | |
1139 | REG_INFO -- array of information about each subexpression. | |
1140 | ||
1141 | Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
1142 | `pend', `string1', `size1', `string2', and `size2'. */ | |
1143 | ||
1144 | #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
1145 | { \ | |
1146 | DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ | |
1147 | int this_reg; \ | |
1148 | const unsigned char *string_temp; \ | |
1149 | \ | |
1150 | assert (!FAIL_STACK_EMPTY ()); \ | |
1151 | \ | |
1152 | /* Remove failure points and point to how many regs pushed. */ \ | |
1153 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1154 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
1155 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
1156 | \ | |
1157 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
1158 | \ | |
1159 | DEBUG_POP (&failure_id); \ | |
1160 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
1161 | \ | |
1162 | /* If the saved string location is NULL, it came from an \ | |
1163 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1164 | saved NULL, thus retaining our current position in the string. */ \ | |
1165 | string_temp = POP_FAILURE_ITEM (); \ | |
1166 | if (string_temp != NULL) \ | |
1167 | str = (const char *) string_temp; \ | |
1168 | \ | |
1169 | DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ | |
1170 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
1171 | DEBUG_PRINT1 ("'\n"); \ | |
1172 | \ | |
1173 | pat = (unsigned char *) POP_FAILURE_ITEM (); \ | |
1174 | DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ | |
1175 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
1176 | \ | |
1177 | /* Restore register info. */ \ | |
1178 | high_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
1179 | DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ | |
1180 | \ | |
1181 | low_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
1182 | DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ | |
1183 | \ | |
1184 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
1185 | { \ | |
1186 | DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ | |
1187 | \ | |
1188 | reg_info[this_reg].word = POP_FAILURE_ITEM (); \ | |
1189 | DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ | |
1190 | \ | |
1191 | regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
1192 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
1193 | \ | |
1194 | regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
1195 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
1196 | } \ | |
1197 | \ | |
1198 | DEBUG_STATEMENT (nfailure_points_popped++); \ | |
1199 | } /* POP_FAILURE_POINT */ | |
1200 | ||
1201 | ||
1202 | \f | |
1203 | /* Structure for per-register (a.k.a. per-group) information. | |
1204 | This must not be longer than one word, because we push this value | |
1205 | onto the failure stack. Other register information, such as the | |
1206 | starting and ending positions (which are addresses), and the list of | |
1207 | inner groups (which is a bits list) are maintained in separate | |
1208 | variables. | |
1209 | ||
1210 | We are making a (strictly speaking) nonportable assumption here: that | |
1211 | the compiler will pack our bit fields into something that fits into | |
1212 | the type of `word', i.e., is something that fits into one item on the | |
1213 | failure stack. */ | |
1214 | typedef union | |
1215 | { | |
1216 | fail_stack_elt_t word; | |
1217 | struct | |
1218 | { | |
1219 | /* This field is one if this group can match the empty string, | |
1220 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
1221 | #define MATCH_NULL_UNSET_VALUE 3 | |
1222 | unsigned match_null_string_p : 2; | |
1223 | unsigned is_active : 1; | |
1224 | unsigned matched_something : 1; | |
1225 | unsigned ever_matched_something : 1; | |
1226 | } bits; | |
1227 | } register_info_type; | |
1228 | ||
1229 | #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
1230 | #define IS_ACTIVE(R) ((R).bits.is_active) | |
1231 | #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
1232 | #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
1233 | ||
1234 | ||
1235 | /* Call this when have matched a real character; it sets `matched' flags | |
1236 | for the subexpressions which we are currently inside. Also records | |
1237 | that those subexprs have matched. */ | |
1238 | #define SET_REGS_MATCHED() \ | |
1239 | do \ | |
1240 | { \ | |
1241 | unsigned r; \ | |
1242 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
1243 | { \ | |
1244 | MATCHED_SOMETHING (reg_info[r]) \ | |
1245 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
1246 | = 1; \ | |
1247 | } \ | |
1248 | } \ | |
1249 | while (0) | |
1250 | ||
1251 | ||
1252 | /* Registers are set to a sentinel when they haven't yet matched. */ | |
1253 | #define REG_UNSET_VALUE ((char *) -1) | |
1254 | #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
1255 | ||
1256 | ||
1257 | \f | |
a822d92f | 1258 | /* How do we implement a missing MATCH_MAY_ALLOCATE? |
e959bada JB |
1259 | We make the fail stack a global thing, and then grow it to |
1260 | re_max_failures when we compile. */ | |
a822d92f | 1261 | #ifndef MATCH_MAY_ALLOCATE |
e959bada JB |
1262 | static fail_stack_type fail_stack; |
1263 | ||
1264 | static const char ** regstart, ** regend; | |
1265 | static const char ** old_regstart, ** old_regend; | |
1266 | static const char **best_regstart, **best_regend; | |
1267 | static register_info_type *reg_info; | |
1268 | static const char **reg_dummy; | |
1269 | static register_info_type *reg_info_dummy; | |
1270 | #endif | |
1271 | ||
1272 | \f | |
1273 | /* Subroutine declarations and macros for regex_compile. */ | |
1274 | ||
1275 | static void store_op1 (), store_op2 (); | |
1276 | static void insert_op1 (), insert_op2 (); | |
1277 | static boolean at_begline_loc_p (), at_endline_loc_p (); | |
1278 | static boolean group_in_compile_stack (); | |
1279 | static reg_errcode_t compile_range (); | |
1280 | ||
1281 | /* Fetch the next character in the uncompiled pattern---translating it | |
1282 | if necessary. Also cast from a signed character in the constant | |
1283 | string passed to us by the user to an unsigned char that we can use | |
1284 | as an array index (in, e.g., `translate'). */ | |
1285 | #define PATFETCH(c) \ | |
1286 | do {if (p == pend) return REG_EEND; \ | |
1287 | c = (unsigned char) *p++; \ | |
1288 | if (translate) c = translate[c]; \ | |
1289 | } while (0) | |
1290 | ||
1291 | /* Fetch the next character in the uncompiled pattern, with no | |
1292 | translation. */ | |
1293 | #define PATFETCH_RAW(c) \ | |
1294 | do {if (p == pend) return REG_EEND; \ | |
1295 | c = (unsigned char) *p++; \ | |
1296 | } while (0) | |
1297 | ||
1298 | /* Go backwards one character in the pattern. */ | |
1299 | #define PATUNFETCH p-- | |
1300 | ||
1301 | ||
1302 | /* If `translate' is non-null, return translate[D], else just D. We | |
1303 | cast the subscript to translate because some data is declared as | |
1304 | `char *', to avoid warnings when a string constant is passed. But | |
1305 | when we use a character as a subscript we must make it unsigned. */ | |
1306 | #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d)) | |
1307 | ||
1308 | ||
1309 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1310 | ||
1311 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1312 | #define INIT_BUF_SIZE 32 | |
1313 | ||
1314 | /* Make sure we have at least N more bytes of space in buffer. */ | |
1315 | #define GET_BUFFER_SPACE(n) \ | |
1316 | while (b - bufp->buffer + (n) > bufp->allocated) \ | |
1317 | EXTEND_BUFFER () | |
1318 | ||
1319 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1320 | #define BUF_PUSH(c) \ | |
1321 | do { \ | |
1322 | GET_BUFFER_SPACE (1); \ | |
1323 | *b++ = (unsigned char) (c); \ | |
1324 | } while (0) | |
1325 | ||
1326 | ||
1327 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1328 | #define BUF_PUSH_2(c1, c2) \ | |
1329 | do { \ | |
1330 | GET_BUFFER_SPACE (2); \ | |
1331 | *b++ = (unsigned char) (c1); \ | |
1332 | *b++ = (unsigned char) (c2); \ | |
1333 | } while (0) | |
1334 | ||
1335 | ||
1336 | /* As with BUF_PUSH_2, except for three bytes. */ | |
1337 | #define BUF_PUSH_3(c1, c2, c3) \ | |
1338 | do { \ | |
1339 | GET_BUFFER_SPACE (3); \ | |
1340 | *b++ = (unsigned char) (c1); \ | |
1341 | *b++ = (unsigned char) (c2); \ | |
1342 | *b++ = (unsigned char) (c3); \ | |
1343 | } while (0) | |
1344 | ||
1345 | ||
1346 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
1347 | relative address offset by the three bytes the jump itself occupies. */ | |
1348 | #define STORE_JUMP(op, loc, to) \ | |
1349 | store_op1 (op, loc, (to) - (loc) - 3) | |
1350 | ||
1351 | /* Likewise, for a two-argument jump. */ | |
1352 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1353 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1354 | ||
1355 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
1356 | #define INSERT_JUMP(op, loc, to) \ | |
1357 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1358 | ||
1359 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1360 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1361 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1362 | ||
1363 | ||
1364 | /* This is not an arbitrary limit: the arguments which represent offsets | |
1365 | into the pattern are two bytes long. So if 2^16 bytes turns out to | |
1366 | be too small, many things would have to change. */ | |
1367 | #define MAX_BUF_SIZE (1L << 16) | |
1368 | ||
1369 | ||
1370 | /* Extend the buffer by twice its current size via realloc and | |
1371 | reset the pointers that pointed into the old block to point to the | |
1372 | correct places in the new one. If extending the buffer results in it | |
1373 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
1374 | #define EXTEND_BUFFER() \ | |
1375 | do { \ | |
1376 | unsigned char *old_buffer = bufp->buffer; \ | |
1377 | if (bufp->allocated == MAX_BUF_SIZE) \ | |
1378 | return REG_ESIZE; \ | |
1379 | bufp->allocated <<= 1; \ | |
1380 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
1381 | bufp->allocated = MAX_BUF_SIZE; \ | |
1382 | bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ | |
1383 | if (bufp->buffer == NULL) \ | |
1384 | return REG_ESPACE; \ | |
1385 | /* If the buffer moved, move all the pointers into it. */ \ | |
1386 | if (old_buffer != bufp->buffer) \ | |
1387 | { \ | |
1388 | b = (b - old_buffer) + bufp->buffer; \ | |
1389 | begalt = (begalt - old_buffer) + bufp->buffer; \ | |
1390 | if (fixup_alt_jump) \ | |
1391 | fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ | |
1392 | if (laststart) \ | |
1393 | laststart = (laststart - old_buffer) + bufp->buffer; \ | |
1394 | if (pending_exact) \ | |
1395 | pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
1396 | } \ | |
1397 | } while (0) | |
1398 | ||
1399 | ||
1400 | /* Since we have one byte reserved for the register number argument to | |
1401 | {start,stop}_memory, the maximum number of groups we can report | |
1402 | things about is what fits in that byte. */ | |
1403 | #define MAX_REGNUM 255 | |
1404 | ||
1405 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1406 | ignore the excess. */ | |
1407 | typedef unsigned regnum_t; | |
1408 | ||
1409 | ||
1410 | /* Macros for the compile stack. */ | |
1411 | ||
1412 | /* Since offsets can go either forwards or backwards, this type needs to | |
1413 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
1414 | typedef int pattern_offset_t; | |
1415 | ||
1416 | typedef struct | |
1417 | { | |
1418 | pattern_offset_t begalt_offset; | |
1419 | pattern_offset_t fixup_alt_jump; | |
1420 | pattern_offset_t inner_group_offset; | |
1421 | pattern_offset_t laststart_offset; | |
1422 | regnum_t regnum; | |
1423 | } compile_stack_elt_t; | |
1424 | ||
1425 | ||
1426 | typedef struct | |
1427 | { | |
1428 | compile_stack_elt_t *stack; | |
1429 | unsigned size; | |
1430 | unsigned avail; /* Offset of next open position. */ | |
1431 | } compile_stack_type; | |
1432 | ||
1433 | ||
1434 | #define INIT_COMPILE_STACK_SIZE 32 | |
1435 | ||
1436 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1437 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1438 | ||
1439 | /* The next available element. */ | |
1440 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
1441 | ||
1442 | ||
1443 | /* Set the bit for character C in a list. */ | |
1444 | #define SET_LIST_BIT(c) \ | |
1445 | (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
1446 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
1447 | ||
1448 | ||
1449 | /* Get the next unsigned number in the uncompiled pattern. */ | |
1450 | #define GET_UNSIGNED_NUMBER(num) \ | |
1451 | { if (p != pend) \ | |
1452 | { \ | |
1453 | PATFETCH (c); \ | |
1454 | while (ISDIGIT (c)) \ | |
1455 | { \ | |
1456 | if (num < 0) \ | |
1457 | num = 0; \ | |
1458 | num = num * 10 + c - '0'; \ | |
1459 | if (p == pend) \ | |
1460 | break; \ | |
1461 | PATFETCH (c); \ | |
1462 | } \ | |
bc78d348 KB |
1463 | } \ |
1464 | } | |
1465 | ||
1466 | #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
1467 | ||
1468 | #define IS_CHAR_CLASS(string) \ | |
1469 | (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
1470 | || STREQ (string, "lower") || STREQ (string, "digit") \ | |
1471 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
1472 | || STREQ (string, "space") || STREQ (string, "print") \ | |
1473 | || STREQ (string, "punct") || STREQ (string, "graph") \ | |
1474 | || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
1475 | \f | |
1476 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
1477 | Returns one of error codes defined in `regex.h', or zero for success. | |
1478 | ||
1479 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
1480 | fields are set in BUFP on entry. | |
1481 | ||
1482 | If it succeeds, results are put in BUFP (if it returns an error, the | |
1483 | contents of BUFP are undefined): | |
1484 | `buffer' is the compiled pattern; | |
1485 | `syntax' is set to SYNTAX; | |
1486 | `used' is set to the length of the compiled pattern; | |
9114e279 KB |
1487 | `fastmap_accurate' is zero; |
1488 | `re_nsub' is the number of subexpressions in PATTERN; | |
1489 | `not_bol' and `not_eol' are zero; | |
bc78d348 KB |
1490 | |
1491 | The `fastmap' and `newline_anchor' fields are neither | |
1492 | examined nor set. */ | |
1493 | ||
69c38b46 RS |
1494 | /* Return, freeing storage we allocated. */ |
1495 | #define FREE_STACK_RETURN(value) \ | |
5330e99c | 1496 | return (free (compile_stack.stack), value) |
69c38b46 | 1497 | |
bc78d348 KB |
1498 | static reg_errcode_t |
1499 | regex_compile (pattern, size, syntax, bufp) | |
1500 | const char *pattern; | |
1501 | int size; | |
1502 | reg_syntax_t syntax; | |
1503 | struct re_pattern_buffer *bufp; | |
1504 | { | |
1505 | /* We fetch characters from PATTERN here. Even though PATTERN is | |
1506 | `char *' (i.e., signed), we declare these variables as unsigned, so | |
1507 | they can be reliably used as array indices. */ | |
1508 | register unsigned char c, c1; | |
1509 | ||
63268f76 | 1510 | /* A random temporary spot in PATTERN. */ |
bc78d348 KB |
1511 | const char *p1; |
1512 | ||
1513 | /* Points to the end of the buffer, where we should append. */ | |
1514 | register unsigned char *b; | |
1515 | ||
1516 | /* Keeps track of unclosed groups. */ | |
1517 | compile_stack_type compile_stack; | |
1518 | ||
1519 | /* Points to the current (ending) position in the pattern. */ | |
1520 | const char *p = pattern; | |
1521 | const char *pend = pattern + size; | |
1522 | ||
1523 | /* How to translate the characters in the pattern. */ | |
1524 | char *translate = bufp->translate; | |
1525 | ||
1526 | /* Address of the count-byte of the most recently inserted `exactn' | |
1527 | command. This makes it possible to tell if a new exact-match | |
1528 | character can be added to that command or if the character requires | |
1529 | a new `exactn' command. */ | |
1530 | unsigned char *pending_exact = 0; | |
1531 | ||
1532 | /* Address of start of the most recently finished expression. | |
1533 | This tells, e.g., postfix * where to find the start of its | |
1534 | operand. Reset at the beginning of groups and alternatives. */ | |
1535 | unsigned char *laststart = 0; | |
1536 | ||
1537 | /* Address of beginning of regexp, or inside of last group. */ | |
1538 | unsigned char *begalt; | |
1539 | ||
1540 | /* Place in the uncompiled pattern (i.e., the {) to | |
1541 | which to go back if the interval is invalid. */ | |
1542 | const char *beg_interval; | |
1543 | ||
1544 | /* Address of the place where a forward jump should go to the end of | |
1545 | the containing expression. Each alternative of an `or' -- except the | |
1546 | last -- ends with a forward jump of this sort. */ | |
1547 | unsigned char *fixup_alt_jump = 0; | |
1548 | ||
1549 | /* Counts open-groups as they are encountered. Remembered for the | |
1550 | matching close-group on the compile stack, so the same register | |
1551 | number is put in the stop_memory as the start_memory. */ | |
1552 | regnum_t regnum = 0; | |
1553 | ||
1554 | #ifdef DEBUG | |
1555 | DEBUG_PRINT1 ("\nCompiling pattern: "); | |
1556 | if (debug) | |
1557 | { | |
1558 | unsigned debug_count; | |
1559 | ||
1560 | for (debug_count = 0; debug_count < size; debug_count++) | |
1561 | printchar (pattern[debug_count]); | |
1562 | putchar ('\n'); | |
1563 | } | |
1564 | #endif /* DEBUG */ | |
1565 | ||
1566 | /* Initialize the compile stack. */ | |
1567 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
1568 | if (compile_stack.stack == NULL) | |
1569 | return REG_ESPACE; | |
1570 | ||
1571 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
1572 | compile_stack.avail = 0; | |
1573 | ||
1574 | /* Initialize the pattern buffer. */ | |
1575 | bufp->syntax = syntax; | |
1576 | bufp->fastmap_accurate = 0; | |
1577 | bufp->not_bol = bufp->not_eol = 0; | |
1578 | ||
1579 | /* Set `used' to zero, so that if we return an error, the pattern | |
1580 | printer (for debugging) will think there's no pattern. We reset it | |
1581 | at the end. */ | |
1582 | bufp->used = 0; | |
1583 | ||
1584 | /* Always count groups, whether or not bufp->no_sub is set. */ | |
1585 | bufp->re_nsub = 0; | |
1586 | ||
1587 | #if !defined (emacs) && !defined (SYNTAX_TABLE) | |
1588 | /* Initialize the syntax table. */ | |
1589 | init_syntax_once (); | |
1590 | #endif | |
1591 | ||
1592 | if (bufp->allocated == 0) | |
1593 | { | |
1594 | if (bufp->buffer) | |
1595 | { /* If zero allocated, but buffer is non-null, try to realloc | |
1596 | enough space. This loses if buffer's address is bogus, but | |
1597 | that is the user's responsibility. */ | |
1598 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
1599 | } | |
1600 | else | |
1601 | { /* Caller did not allocate a buffer. Do it for them. */ | |
1602 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
1603 | } | |
69c38b46 | 1604 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
bc78d348 KB |
1605 | |
1606 | bufp->allocated = INIT_BUF_SIZE; | |
1607 | } | |
1608 | ||
1609 | begalt = b = bufp->buffer; | |
1610 | ||
1611 | /* Loop through the uncompiled pattern until we're at the end. */ | |
1612 | while (p != pend) | |
1613 | { | |
1614 | PATFETCH (c); | |
1615 | ||
1616 | switch (c) | |
1617 | { | |
1618 | case '^': | |
1619 | { | |
1620 | if ( /* If at start of pattern, it's an operator. */ | |
1621 | p == pattern + 1 | |
1622 | /* If context independent, it's an operator. */ | |
1623 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1624 | /* Otherwise, depends on what's come before. */ | |
1625 | || at_begline_loc_p (pattern, p, syntax)) | |
1626 | BUF_PUSH (begline); | |
1627 | else | |
1628 | goto normal_char; | |
1629 | } | |
1630 | break; | |
1631 | ||
1632 | ||
1633 | case '$': | |
1634 | { | |
1635 | if ( /* If at end of pattern, it's an operator. */ | |
1636 | p == pend | |
1637 | /* If context independent, it's an operator. */ | |
1638 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1639 | /* Otherwise, depends on what's next. */ | |
1640 | || at_endline_loc_p (p, pend, syntax)) | |
1641 | BUF_PUSH (endline); | |
1642 | else | |
1643 | goto normal_char; | |
1644 | } | |
1645 | break; | |
1646 | ||
1647 | ||
1648 | case '+': | |
1649 | case '?': | |
1650 | if ((syntax & RE_BK_PLUS_QM) | |
1651 | || (syntax & RE_LIMITED_OPS)) | |
1652 | goto normal_char; | |
1653 | handle_plus: | |
1654 | case '*': | |
1655 | /* If there is no previous pattern... */ | |
1656 | if (!laststart) | |
1657 | { | |
1658 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
69c38b46 | 1659 | FREE_STACK_RETURN (REG_BADRPT); |
bc78d348 KB |
1660 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) |
1661 | goto normal_char; | |
1662 | } | |
1663 | ||
1664 | { | |
1665 | /* Are we optimizing this jump? */ | |
1666 | boolean keep_string_p = false; | |
1667 | ||
1668 | /* 1 means zero (many) matches is allowed. */ | |
1669 | char zero_times_ok = 0, many_times_ok = 0; | |
1670 | ||
1671 | /* If there is a sequence of repetition chars, collapse it | |
1672 | down to just one (the right one). We can't combine | |
1673 | interval operators with these because of, e.g., `a{2}*', | |
1674 | which should only match an even number of `a's. */ | |
1675 | ||
1676 | for (;;) | |
1677 | { | |
1678 | zero_times_ok |= c != '+'; | |
1679 | many_times_ok |= c != '?'; | |
1680 | ||
1681 | if (p == pend) | |
1682 | break; | |
1683 | ||
1684 | PATFETCH (c); | |
1685 | ||
1686 | if (c == '*' | |
1687 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
1688 | ; | |
1689 | ||
1690 | else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
1691 | { | |
69c38b46 | 1692 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
bc78d348 KB |
1693 | |
1694 | PATFETCH (c1); | |
1695 | if (!(c1 == '+' || c1 == '?')) | |
1696 | { | |
1697 | PATUNFETCH; | |
1698 | PATUNFETCH; | |
1699 | break; | |
1700 | } | |
1701 | ||
1702 | c = c1; | |
1703 | } | |
1704 | else | |
1705 | { | |
1706 | PATUNFETCH; | |
1707 | break; | |
1708 | } | |
1709 | ||
1710 | /* If we get here, we found another repeat character. */ | |
1711 | } | |
1712 | ||
1713 | /* Star, etc. applied to an empty pattern is equivalent | |
1714 | to an empty pattern. */ | |
1715 | if (!laststart) | |
1716 | break; | |
1717 | ||
1718 | /* Now we know whether or not zero matches is allowed | |
1719 | and also whether or not two or more matches is allowed. */ | |
1720 | if (many_times_ok) | |
1721 | { /* More than one repetition is allowed, so put in at the | |
1722 | end a backward relative jump from `b' to before the next | |
1723 | jump we're going to put in below (which jumps from | |
1724 | laststart to after this jump). | |
1725 | ||
1726 | But if we are at the `*' in the exact sequence `.*\n', | |
1727 | insert an unconditional jump backwards to the ., | |
1728 | instead of the beginning of the loop. This way we only | |
1729 | push a failure point once, instead of every time | |
1730 | through the loop. */ | |
1731 | assert (p - 1 > pattern); | |
1732 | ||
1733 | /* Allocate the space for the jump. */ | |
1734 | GET_BUFFER_SPACE (3); | |
1735 | ||
1736 | /* We know we are not at the first character of the pattern, | |
1737 | because laststart was nonzero. And we've already | |
1738 | incremented `p', by the way, to be the character after | |
1739 | the `*'. Do we have to do something analogous here | |
1740 | for null bytes, because of RE_DOT_NOT_NULL? */ | |
1741 | if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
5745b086 | 1742 | && zero_times_ok |
bc78d348 KB |
1743 | && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') |
1744 | && !(syntax & RE_DOT_NEWLINE)) | |
1745 | { /* We have .*\n. */ | |
1746 | STORE_JUMP (jump, b, laststart); | |
1747 | keep_string_p = true; | |
1748 | } | |
1749 | else | |
1750 | /* Anything else. */ | |
1751 | STORE_JUMP (maybe_pop_jump, b, laststart - 3); | |
1752 | ||
1753 | /* We've added more stuff to the buffer. */ | |
1754 | b += 3; | |
1755 | } | |
1756 | ||
1757 | /* On failure, jump from laststart to b + 3, which will be the | |
1758 | end of the buffer after this jump is inserted. */ | |
1759 | GET_BUFFER_SPACE (3); | |
1760 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
1761 | : on_failure_jump, | |
1762 | laststart, b + 3); | |
1763 | pending_exact = 0; | |
1764 | b += 3; | |
1765 | ||
1766 | if (!zero_times_ok) | |
1767 | { | |
1768 | /* At least one repetition is required, so insert a | |
1769 | `dummy_failure_jump' before the initial | |
1770 | `on_failure_jump' instruction of the loop. This | |
1771 | effects a skip over that instruction the first time | |
1772 | we hit that loop. */ | |
1773 | GET_BUFFER_SPACE (3); | |
1774 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
1775 | b += 3; | |
1776 | } | |
1777 | } | |
1778 | break; | |
1779 | ||
1780 | ||
1781 | case '.': | |
1782 | laststart = b; | |
1783 | BUF_PUSH (anychar); | |
1784 | break; | |
1785 | ||
1786 | ||
1787 | case '[': | |
1788 | { | |
1789 | boolean had_char_class = false; | |
1790 | ||
69c38b46 | 1791 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
bc78d348 KB |
1792 | |
1793 | /* Ensure that we have enough space to push a charset: the | |
1794 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
1795 | GET_BUFFER_SPACE (34); | |
1796 | ||
1797 | laststart = b; | |
1798 | ||
1799 | /* We test `*p == '^' twice, instead of using an if | |
1800 | statement, so we only need one BUF_PUSH. */ | |
1801 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
1802 | if (*p == '^') | |
1803 | p++; | |
1804 | ||
1805 | /* Remember the first position in the bracket expression. */ | |
1806 | p1 = p; | |
1807 | ||
1808 | /* Push the number of bytes in the bitmap. */ | |
1809 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
1810 | ||
1811 | /* Clear the whole map. */ | |
1812 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
1813 | ||
1814 | /* charset_not matches newline according to a syntax bit. */ | |
1815 | if ((re_opcode_t) b[-2] == charset_not | |
1816 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
1817 | SET_LIST_BIT ('\n'); | |
1818 | ||
1819 | /* Read in characters and ranges, setting map bits. */ | |
1820 | for (;;) | |
1821 | { | |
69c38b46 | 1822 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
bc78d348 KB |
1823 | |
1824 | PATFETCH (c); | |
1825 | ||
1826 | /* \ might escape characters inside [...] and [^...]. */ | |
1827 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
1828 | { | |
69c38b46 | 1829 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
bc78d348 KB |
1830 | |
1831 | PATFETCH (c1); | |
1832 | SET_LIST_BIT (c1); | |
1833 | continue; | |
1834 | } | |
1835 | ||
1836 | /* Could be the end of the bracket expression. If it's | |
1837 | not (i.e., when the bracket expression is `[]' so | |
1838 | far), the ']' character bit gets set way below. */ | |
1839 | if (c == ']' && p != p1 + 1) | |
1840 | break; | |
1841 | ||
1842 | /* Look ahead to see if it's a range when the last thing | |
1843 | was a character class. */ | |
1844 | if (had_char_class && c == '-' && *p != ']') | |
69c38b46 | 1845 | FREE_STACK_RETURN (REG_ERANGE); |
bc78d348 KB |
1846 | |
1847 | /* Look ahead to see if it's a range when the last thing | |
1848 | was a character: if this is a hyphen not at the | |
1849 | beginning or the end of a list, then it's the range | |
1850 | operator. */ | |
1851 | if (c == '-' | |
1852 | && !(p - 2 >= pattern && p[-2] == '[') | |
1853 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
1854 | && *p != ']') | |
1855 | { | |
1856 | reg_errcode_t ret | |
1857 | = compile_range (&p, pend, translate, syntax, b); | |
69c38b46 | 1858 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
bc78d348 KB |
1859 | } |
1860 | ||
1861 | else if (p[0] == '-' && p[1] != ']') | |
1862 | { /* This handles ranges made up of characters only. */ | |
1863 | reg_errcode_t ret; | |
1864 | ||
1865 | /* Move past the `-'. */ | |
1866 | PATFETCH (c1); | |
1867 | ||
1868 | ret = compile_range (&p, pend, translate, syntax, b); | |
69c38b46 | 1869 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
bc78d348 KB |
1870 | } |
1871 | ||
1872 | /* See if we're at the beginning of a possible character | |
1873 | class. */ | |
1874 | ||
1875 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
1876 | { /* Leave room for the null. */ | |
1877 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
1878 | ||
1879 | PATFETCH (c); | |
1880 | c1 = 0; | |
1881 | ||
1882 | /* If pattern is `[[:'. */ | |
69c38b46 | 1883 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
bc78d348 KB |
1884 | |
1885 | for (;;) | |
1886 | { | |
1887 | PATFETCH (c); | |
1888 | if (c == ':' || c == ']' || p == pend | |
1889 | || c1 == CHAR_CLASS_MAX_LENGTH) | |
1890 | break; | |
1891 | str[c1++] = c; | |
1892 | } | |
1893 | str[c1] = '\0'; | |
1894 | ||
1895 | /* If isn't a word bracketed by `[:' and:`]': | |
1896 | undo the ending character, the letters, and leave | |
1897 | the leading `:' and `[' (but set bits for them). */ | |
1898 | if (c == ':' && *p == ']') | |
1899 | { | |
1900 | int ch; | |
1901 | boolean is_alnum = STREQ (str, "alnum"); | |
1902 | boolean is_alpha = STREQ (str, "alpha"); | |
1903 | boolean is_blank = STREQ (str, "blank"); | |
1904 | boolean is_cntrl = STREQ (str, "cntrl"); | |
1905 | boolean is_digit = STREQ (str, "digit"); | |
1906 | boolean is_graph = STREQ (str, "graph"); | |
1907 | boolean is_lower = STREQ (str, "lower"); | |
1908 | boolean is_print = STREQ (str, "print"); | |
1909 | boolean is_punct = STREQ (str, "punct"); | |
1910 | boolean is_space = STREQ (str, "space"); | |
1911 | boolean is_upper = STREQ (str, "upper"); | |
1912 | boolean is_xdigit = STREQ (str, "xdigit"); | |
1913 | ||
69c38b46 RS |
1914 | if (!IS_CHAR_CLASS (str)) |
1915 | FREE_STACK_RETURN (REG_ECTYPE); | |
bc78d348 KB |
1916 | |
1917 | /* Throw away the ] at the end of the character | |
1918 | class. */ | |
1919 | PATFETCH (c); | |
1920 | ||
69c38b46 | 1921 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
bc78d348 KB |
1922 | |
1923 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
1924 | { | |
c6b40788 JM |
1925 | if ( (is_alnum && ISALNUM (ch)) |
1926 | || (is_alpha && ISALPHA (ch)) | |
1927 | || (is_blank && ISBLANK (ch)) | |
1928 | || (is_cntrl && ISCNTRL (ch)) | |
1929 | || (is_digit && ISDIGIT (ch)) | |
1930 | || (is_graph && ISGRAPH (ch)) | |
1931 | || (is_lower && ISLOWER (ch)) | |
1932 | || (is_print && ISPRINT (ch)) | |
1933 | || (is_punct && ISPUNCT (ch)) | |
1934 | || (is_space && ISSPACE (ch)) | |
1935 | || (is_upper && ISUPPER (ch)) | |
1936 | || (is_xdigit && ISXDIGIT (ch))) | |
bc78d348 KB |
1937 | SET_LIST_BIT (ch); |
1938 | } | |
1939 | had_char_class = true; | |
1940 | } | |
1941 | else | |
1942 | { | |
1943 | c1++; | |
1944 | while (c1--) | |
1945 | PATUNFETCH; | |
1946 | SET_LIST_BIT ('['); | |
1947 | SET_LIST_BIT (':'); | |
1948 | had_char_class = false; | |
1949 | } | |
1950 | } | |
1951 | else | |
1952 | { | |
1953 | had_char_class = false; | |
1954 | SET_LIST_BIT (c); | |
1955 | } | |
1956 | } | |
1957 | ||
1958 | /* Discard any (non)matching list bytes that are all 0 at the | |
1959 | end of the map. Decrease the map-length byte too. */ | |
1960 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
1961 | b[-1]--; | |
1962 | b += b[-1]; | |
1963 | } | |
1964 | break; | |
1965 | ||
1966 | ||
1967 | case '(': | |
1968 | if (syntax & RE_NO_BK_PARENS) | |
1969 | goto handle_open; | |
1970 | else | |
1971 | goto normal_char; | |
1972 | ||
1973 | ||
1974 | case ')': | |
1975 | if (syntax & RE_NO_BK_PARENS) | |
1976 | goto handle_close; | |
1977 | else | |
1978 | goto normal_char; | |
1979 | ||
1980 | ||
1981 | case '\n': | |
1982 | if (syntax & RE_NEWLINE_ALT) | |
1983 | goto handle_alt; | |
1984 | else | |
1985 | goto normal_char; | |
1986 | ||
1987 | ||
1988 | case '|': | |
1989 | if (syntax & RE_NO_BK_VBAR) | |
1990 | goto handle_alt; | |
1991 | else | |
1992 | goto normal_char; | |
1993 | ||
1994 | ||
1995 | case '{': | |
1996 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
1997 | goto handle_interval; | |
1998 | else | |
1999 | goto normal_char; | |
2000 | ||
2001 | ||
2002 | case '\\': | |
69c38b46 | 2003 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
bc78d348 KB |
2004 | |
2005 | /* Do not translate the character after the \, so that we can | |
2006 | distinguish, e.g., \B from \b, even if we normally would | |
2007 | translate, e.g., B to b. */ | |
2008 | PATFETCH_RAW (c); | |
2009 | ||
2010 | switch (c) | |
2011 | { | |
2012 | case '(': | |
2013 | if (syntax & RE_NO_BK_PARENS) | |
2014 | goto normal_backslash; | |
2015 | ||
2016 | handle_open: | |
2017 | bufp->re_nsub++; | |
2018 | regnum++; | |
2019 | ||
2020 | if (COMPILE_STACK_FULL) | |
2021 | { | |
2022 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
2023 | compile_stack_elt_t); | |
2024 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
2025 | ||
2026 | compile_stack.size <<= 1; | |
2027 | } | |
2028 | ||
2029 | /* These are the values to restore when we hit end of this | |
2030 | group. They are all relative offsets, so that if the | |
2031 | whole pattern moves because of realloc, they will still | |
2032 | be valid. */ | |
2033 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
2034 | COMPILE_STACK_TOP.fixup_alt_jump | |
2035 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
2036 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
2037 | COMPILE_STACK_TOP.regnum = regnum; | |
2038 | ||
2039 | /* We will eventually replace the 0 with the number of | |
2040 | groups inner to this one. But do not push a | |
2041 | start_memory for groups beyond the last one we can | |
2042 | represent in the compiled pattern. */ | |
2043 | if (regnum <= MAX_REGNUM) | |
2044 | { | |
2045 | COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; | |
2046 | BUF_PUSH_3 (start_memory, regnum, 0); | |
2047 | } | |
2048 | ||
2049 | compile_stack.avail++; | |
2050 | ||
2051 | fixup_alt_jump = 0; | |
2052 | laststart = 0; | |
2053 | begalt = b; | |
5745b086 JB |
2054 | /* If we've reached MAX_REGNUM groups, then this open |
2055 | won't actually generate any code, so we'll have to | |
2056 | clear pending_exact explicitly. */ | |
2057 | pending_exact = 0; | |
bc78d348 KB |
2058 | break; |
2059 | ||
2060 | ||
2061 | case ')': | |
2062 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
2063 | ||
2064 | if (COMPILE_STACK_EMPTY) | |
2065 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2066 | goto normal_backslash; | |
2067 | else | |
69c38b46 | 2068 | FREE_STACK_RETURN (REG_ERPAREN); |
bc78d348 KB |
2069 | |
2070 | handle_close: | |
2071 | if (fixup_alt_jump) | |
2072 | { /* Push a dummy failure point at the end of the | |
2073 | alternative for a possible future | |
2074 | `pop_failure_jump' to pop. See comments at | |
2075 | `push_dummy_failure' in `re_match_2'. */ | |
2076 | BUF_PUSH (push_dummy_failure); | |
2077 | ||
2078 | /* We allocated space for this jump when we assigned | |
2079 | to `fixup_alt_jump', in the `handle_alt' case below. */ | |
2080 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
2081 | } | |
2082 | ||
2083 | /* See similar code for backslashed left paren above. */ | |
2084 | if (COMPILE_STACK_EMPTY) | |
2085 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2086 | goto normal_char; | |
2087 | else | |
69c38b46 | 2088 | FREE_STACK_RETURN (REG_ERPAREN); |
bc78d348 KB |
2089 | |
2090 | /* Since we just checked for an empty stack above, this | |
2091 | ``can't happen''. */ | |
2092 | assert (compile_stack.avail != 0); | |
2093 | { | |
2094 | /* We don't just want to restore into `regnum', because | |
2095 | later groups should continue to be numbered higher, | |
2096 | as in `(ab)c(de)' -- the second group is #2. */ | |
2097 | regnum_t this_group_regnum; | |
2098 | ||
2099 | compile_stack.avail--; | |
2100 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
2101 | fixup_alt_jump | |
2102 | = COMPILE_STACK_TOP.fixup_alt_jump | |
2103 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
2104 | : 0; | |
2105 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
2106 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
5745b086 JB |
2107 | /* If we've reached MAX_REGNUM groups, then this open |
2108 | won't actually generate any code, so we'll have to | |
2109 | clear pending_exact explicitly. */ | |
2110 | pending_exact = 0; | |
bc78d348 KB |
2111 | |
2112 | /* We're at the end of the group, so now we know how many | |
2113 | groups were inside this one. */ | |
2114 | if (this_group_regnum <= MAX_REGNUM) | |
2115 | { | |
2116 | unsigned char *inner_group_loc | |
2117 | = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
2118 | ||
2119 | *inner_group_loc = regnum - this_group_regnum; | |
2120 | BUF_PUSH_3 (stop_memory, this_group_regnum, | |
2121 | regnum - this_group_regnum); | |
2122 | } | |
2123 | } | |
2124 | break; | |
2125 | ||
2126 | ||
2127 | case '|': /* `\|'. */ | |
2128 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
2129 | goto normal_backslash; | |
2130 | handle_alt: | |
2131 | if (syntax & RE_LIMITED_OPS) | |
2132 | goto normal_char; | |
2133 | ||
2134 | /* Insert before the previous alternative a jump which | |
2135 | jumps to this alternative if the former fails. */ | |
2136 | GET_BUFFER_SPACE (3); | |
2137 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
2138 | pending_exact = 0; | |
2139 | b += 3; | |
2140 | ||
2141 | /* The alternative before this one has a jump after it | |
2142 | which gets executed if it gets matched. Adjust that | |
2143 | jump so it will jump to this alternative's analogous | |
2144 | jump (put in below, which in turn will jump to the next | |
2145 | (if any) alternative's such jump, etc.). The last such | |
2146 | jump jumps to the correct final destination. A picture: | |
2147 | _____ _____ | |
2148 | | | | | | |
2149 | | v | v | |
2150 | a | b | c | |
2151 | ||
9114e279 KB |
2152 | If we are at `b', then fixup_alt_jump right now points to a |
2153 | three-byte space after `a'. We'll put in the jump, set | |
2154 | fixup_alt_jump to right after `b', and leave behind three | |
2155 | bytes which we'll fill in when we get to after `c'. */ | |
bc78d348 KB |
2156 | |
2157 | if (fixup_alt_jump) | |
2158 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2159 | ||
2160 | /* Mark and leave space for a jump after this alternative, | |
2161 | to be filled in later either by next alternative or | |
2162 | when know we're at the end of a series of alternatives. */ | |
2163 | fixup_alt_jump = b; | |
2164 | GET_BUFFER_SPACE (3); | |
2165 | b += 3; | |
2166 | ||
2167 | laststart = 0; | |
2168 | begalt = b; | |
2169 | break; | |
2170 | ||
2171 | ||
2172 | case '{': | |
2173 | /* If \{ is a literal. */ | |
2174 | if (!(syntax & RE_INTERVALS) | |
2175 | /* If we're at `\{' and it's not the open-interval | |
2176 | operator. */ | |
2177 | || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
2178 | || (p - 2 == pattern && p == pend)) | |
2179 | goto normal_backslash; | |
2180 | ||
2181 | handle_interval: | |
2182 | { | |
2183 | /* If got here, then the syntax allows intervals. */ | |
2184 | ||
2185 | /* At least (most) this many matches must be made. */ | |
2186 | int lower_bound = -1, upper_bound = -1; | |
2187 | ||
2188 | beg_interval = p - 1; | |
2189 | ||
2190 | if (p == pend) | |
2191 | { | |
2192 | if (syntax & RE_NO_BK_BRACES) | |
2193 | goto unfetch_interval; | |
2194 | else | |
69c38b46 | 2195 | FREE_STACK_RETURN (REG_EBRACE); |
bc78d348 KB |
2196 | } |
2197 | ||
2198 | GET_UNSIGNED_NUMBER (lower_bound); | |
2199 | ||
2200 | if (c == ',') | |
2201 | { | |
2202 | GET_UNSIGNED_NUMBER (upper_bound); | |
2203 | if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
2204 | } | |
2205 | else | |
2206 | /* Interval such as `{1}' => match exactly once. */ | |
2207 | upper_bound = lower_bound; | |
2208 | ||
2209 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
2210 | || lower_bound > upper_bound) | |
2211 | { | |
2212 | if (syntax & RE_NO_BK_BRACES) | |
2213 | goto unfetch_interval; | |
2214 | else | |
69c38b46 | 2215 | FREE_STACK_RETURN (REG_BADBR); |
bc78d348 KB |
2216 | } |
2217 | ||
2218 | if (!(syntax & RE_NO_BK_BRACES)) | |
2219 | { | |
69c38b46 | 2220 | if (c != '\\') FREE_STACK_RETURN (REG_EBRACE); |
bc78d348 KB |
2221 | |
2222 | PATFETCH (c); | |
2223 | } | |
2224 | ||
2225 | if (c != '}') | |
2226 | { | |
2227 | if (syntax & RE_NO_BK_BRACES) | |
2228 | goto unfetch_interval; | |
2229 | else | |
69c38b46 | 2230 | FREE_STACK_RETURN (REG_BADBR); |
bc78d348 KB |
2231 | } |
2232 | ||
2233 | /* We just parsed a valid interval. */ | |
2234 | ||
2235 | /* If it's invalid to have no preceding re. */ | |
2236 | if (!laststart) | |
2237 | { | |
2238 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
69c38b46 | 2239 | FREE_STACK_RETURN (REG_BADRPT); |
bc78d348 KB |
2240 | else if (syntax & RE_CONTEXT_INDEP_OPS) |
2241 | laststart = b; | |
2242 | else | |
2243 | goto unfetch_interval; | |
2244 | } | |
2245 | ||
2246 | /* If the upper bound is zero, don't want to succeed at | |
2247 | all; jump from `laststart' to `b + 3', which will be | |
2248 | the end of the buffer after we insert the jump. */ | |
2249 | if (upper_bound == 0) | |
2250 | { | |
2251 | GET_BUFFER_SPACE (3); | |
2252 | INSERT_JUMP (jump, laststart, b + 3); | |
2253 | b += 3; | |
2254 | } | |
2255 | ||
2256 | /* Otherwise, we have a nontrivial interval. When | |
2257 | we're all done, the pattern will look like: | |
2258 | set_number_at <jump count> <upper bound> | |
2259 | set_number_at <succeed_n count> <lower bound> | |
63268f76 | 2260 | succeed_n <after jump addr> <succeed_n count> |
bc78d348 KB |
2261 | <body of loop> |
2262 | jump_n <succeed_n addr> <jump count> | |
2263 | (The upper bound and `jump_n' are omitted if | |
2264 | `upper_bound' is 1, though.) */ | |
2265 | else | |
2266 | { /* If the upper bound is > 1, we need to insert | |
2267 | more at the end of the loop. */ | |
2268 | unsigned nbytes = 10 + (upper_bound > 1) * 10; | |
2269 | ||
2270 | GET_BUFFER_SPACE (nbytes); | |
2271 | ||
2272 | /* Initialize lower bound of the `succeed_n', even | |
2273 | though it will be set during matching by its | |
2274 | attendant `set_number_at' (inserted next), | |
2275 | because `re_compile_fastmap' needs to know. | |
2276 | Jump to the `jump_n' we might insert below. */ | |
2277 | INSERT_JUMP2 (succeed_n, laststart, | |
2278 | b + 5 + (upper_bound > 1) * 5, | |
2279 | lower_bound); | |
2280 | b += 5; | |
2281 | ||
2282 | /* Code to initialize the lower bound. Insert | |
2283 | before the `succeed_n'. The `5' is the last two | |
2284 | bytes of this `set_number_at', plus 3 bytes of | |
2285 | the following `succeed_n'. */ | |
2286 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
2287 | b += 5; | |
2288 | ||
2289 | if (upper_bound > 1) | |
2290 | { /* More than one repetition is allowed, so | |
2291 | append a backward jump to the `succeed_n' | |
2292 | that starts this interval. | |
2293 | ||
2294 | When we've reached this during matching, | |
2295 | we'll have matched the interval once, so | |
2296 | jump back only `upper_bound - 1' times. */ | |
2297 | STORE_JUMP2 (jump_n, b, laststart + 5, | |
2298 | upper_bound - 1); | |
2299 | b += 5; | |
2300 | ||
2301 | /* The location we want to set is the second | |
2302 | parameter of the `jump_n'; that is `b-2' as | |
2303 | an absolute address. `laststart' will be | |
2304 | the `set_number_at' we're about to insert; | |
2305 | `laststart+3' the number to set, the source | |
2306 | for the relative address. But we are | |
2307 | inserting into the middle of the pattern -- | |
2308 | so everything is getting moved up by 5. | |
2309 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
2310 | i.e., b - laststart. | |
2311 | ||
2312 | We insert this at the beginning of the loop | |
2313 | so that if we fail during matching, we'll | |
2314 | reinitialize the bounds. */ | |
2315 | insert_op2 (set_number_at, laststart, b - laststart, | |
2316 | upper_bound - 1, b); | |
2317 | b += 5; | |
2318 | } | |
2319 | } | |
2320 | pending_exact = 0; | |
2321 | beg_interval = NULL; | |
2322 | } | |
2323 | break; | |
2324 | ||
2325 | unfetch_interval: | |
2326 | /* If an invalid interval, match the characters as literals. */ | |
2327 | assert (beg_interval); | |
2328 | p = beg_interval; | |
2329 | beg_interval = NULL; | |
2330 | ||
2331 | /* normal_char and normal_backslash need `c'. */ | |
2332 | PATFETCH (c); | |
2333 | ||
2334 | if (!(syntax & RE_NO_BK_BRACES)) | |
2335 | { | |
2336 | if (p > pattern && p[-1] == '\\') | |
2337 | goto normal_backslash; | |
2338 | } | |
2339 | goto normal_char; | |
2340 | ||
2341 | #ifdef emacs | |
2342 | /* There is no way to specify the before_dot and after_dot | |
2343 | operators. rms says this is ok. --karl */ | |
2344 | case '=': | |
2345 | BUF_PUSH (at_dot); | |
2346 | break; | |
2347 | ||
2348 | case 's': | |
2349 | laststart = b; | |
2350 | PATFETCH (c); | |
2351 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
2352 | break; | |
2353 | ||
2354 | case 'S': | |
2355 | laststart = b; | |
2356 | PATFETCH (c); | |
2357 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
2358 | break; | |
2359 | #endif /* emacs */ | |
2360 | ||
2361 | ||
2362 | case 'w': | |
2363 | laststart = b; | |
2364 | BUF_PUSH (wordchar); | |
2365 | break; | |
2366 | ||
2367 | ||
2368 | case 'W': | |
2369 | laststart = b; | |
2370 | BUF_PUSH (notwordchar); | |
2371 | break; | |
2372 | ||
2373 | ||
2374 | case '<': | |
2375 | BUF_PUSH (wordbeg); | |
2376 | break; | |
2377 | ||
2378 | case '>': | |
2379 | BUF_PUSH (wordend); | |
2380 | break; | |
2381 | ||
2382 | case 'b': | |
2383 | BUF_PUSH (wordbound); | |
2384 | break; | |
2385 | ||
2386 | case 'B': | |
2387 | BUF_PUSH (notwordbound); | |
2388 | break; | |
2389 | ||
2390 | case '`': | |
2391 | BUF_PUSH (begbuf); | |
2392 | break; | |
2393 | ||
2394 | case '\'': | |
2395 | BUF_PUSH (endbuf); | |
2396 | break; | |
2397 | ||
2398 | case '1': case '2': case '3': case '4': case '5': | |
2399 | case '6': case '7': case '8': case '9': | |
2400 | if (syntax & RE_NO_BK_REFS) | |
2401 | goto normal_char; | |
2402 | ||
2403 | c1 = c - '0'; | |
2404 | ||
2405 | if (c1 > regnum) | |
69c38b46 | 2406 | FREE_STACK_RETURN (REG_ESUBREG); |
bc78d348 KB |
2407 | |
2408 | /* Can't back reference to a subexpression if inside of it. */ | |
2409 | if (group_in_compile_stack (compile_stack, c1)) | |
2410 | goto normal_char; | |
2411 | ||
2412 | laststart = b; | |
2413 | BUF_PUSH_2 (duplicate, c1); | |
2414 | break; | |
2415 | ||
2416 | ||
2417 | case '+': | |
2418 | case '?': | |
2419 | if (syntax & RE_BK_PLUS_QM) | |
2420 | goto handle_plus; | |
2421 | else | |
2422 | goto normal_backslash; | |
2423 | ||
2424 | default: | |
2425 | normal_backslash: | |
2426 | /* You might think it would be useful for \ to mean | |
2427 | not to translate; but if we don't translate it | |
2428 | it will never match anything. */ | |
2429 | c = TRANSLATE (c); | |
2430 | goto normal_char; | |
2431 | } | |
2432 | break; | |
2433 | ||
2434 | ||
2435 | default: | |
2436 | /* Expects the character in `c'. */ | |
2437 | normal_char: | |
2438 | /* If no exactn currently being built. */ | |
2439 | if (!pending_exact | |
2440 | ||
2441 | /* If last exactn not at current position. */ | |
2442 | || pending_exact + *pending_exact + 1 != b | |
2443 | ||
2444 | /* We have only one byte following the exactn for the count. */ | |
2445 | || *pending_exact == (1 << BYTEWIDTH) - 1 | |
2446 | ||
2447 | /* If followed by a repetition operator. */ | |
2448 | || *p == '*' || *p == '^' | |
2449 | || ((syntax & RE_BK_PLUS_QM) | |
2450 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
2451 | : (*p == '+' || *p == '?')) | |
2452 | || ((syntax & RE_INTERVALS) | |
2453 | && ((syntax & RE_NO_BK_BRACES) | |
2454 | ? *p == '{' | |
2455 | : (p[0] == '\\' && p[1] == '{')))) | |
2456 | { | |
2457 | /* Start building a new exactn. */ | |
2458 | ||
2459 | laststart = b; | |
2460 | ||
2461 | BUF_PUSH_2 (exactn, 0); | |
2462 | pending_exact = b - 1; | |
2463 | } | |
2464 | ||
2465 | BUF_PUSH (c); | |
2466 | (*pending_exact)++; | |
2467 | break; | |
2468 | } /* switch (c) */ | |
e959bada | 2469 | } /* while p != pend */ |
4af46de6 | 2470 | |
e959bada JB |
2471 | |
2472 | /* Through the pattern now. */ | |
2473 | ||
2474 | if (fixup_alt_jump) | |
2475 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
bc78d348 | 2476 | |
e959bada | 2477 | if (!COMPILE_STACK_EMPTY) |
69c38b46 | 2478 | FREE_STACK_RETURN (REG_EPAREN); |
bc78d348 | 2479 | |
e959bada | 2480 | free (compile_stack.stack); |
bc78d348 | 2481 | |
e959bada JB |
2482 | /* We have succeeded; set the length of the buffer. */ |
2483 | bufp->used = b - bufp->buffer; | |
2484 | ||
2485 | #ifdef DEBUG | |
2486 | if (debug) | |
bc78d348 | 2487 | { |
e959bada JB |
2488 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
2489 | print_compiled_pattern (bufp); | |
bc78d348 | 2490 | } |
e959bada | 2491 | #endif /* DEBUG */ |
bc78d348 | 2492 | |
a822d92f | 2493 | #ifndef MATCH_MAY_ALLOCATE |
e959bada JB |
2494 | /* Initialize the failure stack to the largest possible stack. This |
2495 | isn't necessary unless we're trying to avoid calling alloca in | |
2496 | the search and match routines. */ | |
2497 | { | |
2498 | int num_regs = bufp->re_nsub + 1; | |
2499 | ||
2500 | /* Since DOUBLE_FAIL_STACK refuses to double only if the current size | |
2501 | is strictly greater than re_max_failures, the largest possible stack | |
2502 | is 2 * re_max_failures failure points. */ | |
69fdec9e | 2503 | if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) |
501b568d | 2504 | { |
69fdec9e | 2505 | fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); |
cbbe9053 RS |
2506 | |
2507 | #ifdef emacs | |
501b568d | 2508 | if (! fail_stack.stack) |
cbbe9053 RS |
2509 | fail_stack.stack |
2510 | = (fail_stack_elt_t *) xmalloc (fail_stack.size | |
2511 | * sizeof (fail_stack_elt_t)); | |
501b568d | 2512 | else |
cbbe9053 RS |
2513 | fail_stack.stack |
2514 | = (fail_stack_elt_t *) xrealloc (fail_stack.stack, | |
2515 | (fail_stack.size | |
2516 | * sizeof (fail_stack_elt_t))); | |
2517 | #else /* not emacs */ | |
2518 | if (! fail_stack.stack) | |
2519 | fail_stack.stack | |
2520 | = (fail_stack_elt_t *) malloc (fail_stack.size | |
2521 | * sizeof (fail_stack_elt_t)); | |
2522 | else | |
2523 | fail_stack.stack | |
2524 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
2525 | (fail_stack.size | |
2526 | * sizeof (fail_stack_elt_t))); | |
2527 | #endif /* not emacs */ | |
501b568d | 2528 | } |
e959bada JB |
2529 | |
2530 | /* Initialize some other variables the matcher uses. */ | |
2531 | RETALLOC_IF (regstart, num_regs, const char *); | |
2532 | RETALLOC_IF (regend, num_regs, const char *); | |
2533 | RETALLOC_IF (old_regstart, num_regs, const char *); | |
2534 | RETALLOC_IF (old_regend, num_regs, const char *); | |
2535 | RETALLOC_IF (best_regstart, num_regs, const char *); | |
2536 | RETALLOC_IF (best_regend, num_regs, const char *); | |
2537 | RETALLOC_IF (reg_info, num_regs, register_info_type); | |
2538 | RETALLOC_IF (reg_dummy, num_regs, const char *); | |
2539 | RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); | |
2540 | } | |
bc78d348 KB |
2541 | #endif |
2542 | ||
e959bada JB |
2543 | return REG_NOERROR; |
2544 | } /* regex_compile */ | |
2545 | \f | |
2546 | /* Subroutines for `regex_compile'. */ | |
bc78d348 | 2547 | |
e959bada | 2548 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
bc78d348 | 2549 | |
e959bada JB |
2550 | static void |
2551 | store_op1 (op, loc, arg) | |
2552 | re_opcode_t op; | |
2553 | unsigned char *loc; | |
2554 | int arg; | |
bc78d348 | 2555 | { |
e959bada JB |
2556 | *loc = (unsigned char) op; |
2557 | STORE_NUMBER (loc + 1, arg); | |
2558 | } | |
bc78d348 | 2559 | |
bc78d348 | 2560 | |
e959bada | 2561 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ |
bc78d348 | 2562 | |
e959bada JB |
2563 | static void |
2564 | store_op2 (op, loc, arg1, arg2) | |
2565 | re_opcode_t op; | |
2566 | unsigned char *loc; | |
2567 | int arg1, arg2; | |
2568 | { | |
2569 | *loc = (unsigned char) op; | |
2570 | STORE_NUMBER (loc + 1, arg1); | |
2571 | STORE_NUMBER (loc + 3, arg2); | |
2572 | } | |
bc78d348 | 2573 | |
bc78d348 | 2574 | |
e959bada JB |
2575 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC |
2576 | for OP followed by two-byte integer parameter ARG. */ | |
bc78d348 | 2577 | |
e959bada JB |
2578 | static void |
2579 | insert_op1 (op, loc, arg, end) | |
2580 | re_opcode_t op; | |
2581 | unsigned char *loc; | |
2582 | int arg; | |
2583 | unsigned char *end; | |
2584 | { | |
2585 | register unsigned char *pfrom = end; | |
2586 | register unsigned char *pto = end + 3; | |
bc78d348 | 2587 | |
e959bada JB |
2588 | while (pfrom != loc) |
2589 | *--pto = *--pfrom; | |
2590 | ||
2591 | store_op1 (op, loc, arg); | |
2592 | } | |
bc78d348 | 2593 | |
bc78d348 | 2594 | |
e959bada | 2595 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ |
bc78d348 | 2596 | |
e959bada JB |
2597 | static void |
2598 | insert_op2 (op, loc, arg1, arg2, end) | |
2599 | re_opcode_t op; | |
2600 | unsigned char *loc; | |
2601 | int arg1, arg2; | |
2602 | unsigned char *end; | |
2603 | { | |
2604 | register unsigned char *pfrom = end; | |
2605 | register unsigned char *pto = end + 5; | |
bc78d348 | 2606 | |
e959bada JB |
2607 | while (pfrom != loc) |
2608 | *--pto = *--pfrom; | |
2609 | ||
2610 | store_op2 (op, loc, arg1, arg2); | |
2611 | } | |
bc78d348 | 2612 | |
bc78d348 | 2613 | |
e959bada JB |
2614 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes |
2615 | after an alternative or a begin-subexpression. We assume there is at | |
2616 | least one character before the ^. */ | |
bc78d348 | 2617 | |
e959bada JB |
2618 | static boolean |
2619 | at_begline_loc_p (pattern, p, syntax) | |
2620 | const char *pattern, *p; | |
2621 | reg_syntax_t syntax; | |
2622 | { | |
2623 | const char *prev = p - 2; | |
2624 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
2625 | ||
2626 | return | |
2627 | /* After a subexpression? */ | |
2628 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
2629 | /* After an alternative? */ | |
2630 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
2631 | } | |
bc78d348 | 2632 | |
bc78d348 | 2633 | |
e959bada JB |
2634 | /* The dual of at_begline_loc_p. This one is for $. We assume there is |
2635 | at least one character after the $, i.e., `P < PEND'. */ | |
2636 | ||
2637 | static boolean | |
2638 | at_endline_loc_p (p, pend, syntax) | |
2639 | const char *p, *pend; | |
2640 | int syntax; | |
2641 | { | |
2642 | const char *next = p; | |
2643 | boolean next_backslash = *next == '\\'; | |
2644 | const char *next_next = p + 1 < pend ? p + 1 : NULL; | |
2645 | ||
2646 | return | |
2647 | /* Before a subexpression? */ | |
2648 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
2649 | : next_backslash && next_next && *next_next == ')') | |
2650 | /* Before an alternative? */ | |
2651 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
2652 | : next_backslash && next_next && *next_next == '|'); | |
2653 | } | |
bc78d348 KB |
2654 | |
2655 | ||
e959bada JB |
2656 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
2657 | false if it's not. */ | |
bc78d348 | 2658 | |
e959bada JB |
2659 | static boolean |
2660 | group_in_compile_stack (compile_stack, regnum) | |
2661 | compile_stack_type compile_stack; | |
2662 | regnum_t regnum; | |
2663 | { | |
2664 | int this_element; | |
2665 | ||
2666 | for (this_element = compile_stack.avail - 1; | |
2667 | this_element >= 0; | |
2668 | this_element--) | |
2669 | if (compile_stack.stack[this_element].regnum == regnum) | |
2670 | return true; | |
2671 | ||
2672 | return false; | |
2673 | } | |
2674 | ||
2675 | ||
2676 | /* Read the ending character of a range (in a bracket expression) from the | |
2677 | uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
2678 | starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
2679 | Then we set the translation of all bits between the starting and | |
2680 | ending characters (inclusive) in the compiled pattern B. | |
bc78d348 | 2681 | |
e959bada JB |
2682 | Return an error code. |
2683 | ||
2684 | We use these short variable names so we can use the same macros as | |
2685 | `regex_compile' itself. */ | |
bc78d348 | 2686 | |
e959bada JB |
2687 | static reg_errcode_t |
2688 | compile_range (p_ptr, pend, translate, syntax, b) | |
2689 | const char **p_ptr, *pend; | |
2690 | char *translate; | |
2691 | reg_syntax_t syntax; | |
2692 | unsigned char *b; | |
2693 | { | |
2694 | unsigned this_char; | |
2695 | ||
2696 | const char *p = *p_ptr; | |
2697 | int range_start, range_end; | |
2698 | ||
2699 | if (p == pend) | |
2700 | return REG_ERANGE; | |
2701 | ||
2702 | /* Even though the pattern is a signed `char *', we need to fetch | |
2703 | with unsigned char *'s; if the high bit of the pattern character | |
2704 | is set, the range endpoints will be negative if we fetch using a | |
2705 | signed char *. | |
2706 | ||
2707 | We also want to fetch the endpoints without translating them; the | |
2708 | appropriate translation is done in the bit-setting loop below. */ | |
2709 | range_start = ((unsigned char *) p)[-2]; | |
2710 | range_end = ((unsigned char *) p)[0]; | |
2711 | ||
2712 | /* Have to increment the pointer into the pattern string, so the | |
2713 | caller isn't still at the ending character. */ | |
2714 | (*p_ptr)++; | |
2715 | ||
2716 | /* If the start is after the end, the range is empty. */ | |
2717 | if (range_start > range_end) | |
2718 | return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
2719 | ||
2720 | /* Here we see why `this_char' has to be larger than an `unsigned | |
2721 | char' -- the range is inclusive, so if `range_end' == 0xff | |
2722 | (assuming 8-bit characters), we would otherwise go into an infinite | |
2723 | loop, since all characters <= 0xff. */ | |
2724 | for (this_char = range_start; this_char <= range_end; this_char++) | |
2725 | { | |
2726 | SET_LIST_BIT (TRANSLATE (this_char)); | |
2727 | } | |
2728 | ||
2729 | return REG_NOERROR; | |
2730 | } | |
bc78d348 KB |
2731 | \f |
2732 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
2733 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
2734 | characters can start a string that matches the pattern. This fastmap | |
2735 | is used by re_search to skip quickly over impossible starting points. | |
2736 | ||
2737 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
2738 | area as BUFP->fastmap. | |
2739 | ||
2740 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
2741 | the pattern buffer. | |
2742 | ||
2743 | Returns 0 if we succeed, -2 if an internal error. */ | |
2744 | ||
2745 | int | |
2746 | re_compile_fastmap (bufp) | |
2747 | struct re_pattern_buffer *bufp; | |
2748 | { | |
2749 | int j, k; | |
a822d92f | 2750 | #ifdef MATCH_MAY_ALLOCATE |
bc78d348 | 2751 | fail_stack_type fail_stack; |
e959bada | 2752 | #endif |
bc78d348 KB |
2753 | #ifndef REGEX_MALLOC |
2754 | char *destination; | |
2755 | #endif | |
2756 | /* We don't push any register information onto the failure stack. */ | |
2757 | unsigned num_regs = 0; | |
2758 | ||
2759 | register char *fastmap = bufp->fastmap; | |
2760 | unsigned char *pattern = bufp->buffer; | |
2761 | unsigned long size = bufp->used; | |
794ce2c1 | 2762 | unsigned char *p = pattern; |
bc78d348 KB |
2763 | register unsigned char *pend = pattern + size; |
2764 | ||
2765 | /* Assume that each path through the pattern can be null until | |
2766 | proven otherwise. We set this false at the bottom of switch | |
2767 | statement, to which we get only if a particular path doesn't | |
2768 | match the empty string. */ | |
2769 | boolean path_can_be_null = true; | |
2770 | ||
2771 | /* We aren't doing a `succeed_n' to begin with. */ | |
2772 | boolean succeed_n_p = false; | |
2773 | ||
2774 | assert (fastmap != NULL && p != NULL); | |
2775 | ||
2776 | INIT_FAIL_STACK (); | |
2777 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
2778 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
2779 | bufp->can_be_null = 0; | |
2780 | ||
2781 | while (p != pend || !FAIL_STACK_EMPTY ()) | |
2782 | { | |
2783 | if (p == pend) | |
2784 | { | |
2785 | bufp->can_be_null |= path_can_be_null; | |
2786 | ||
2787 | /* Reset for next path. */ | |
2788 | path_can_be_null = true; | |
2789 | ||
2790 | p = fail_stack.stack[--fail_stack.avail]; | |
2791 | } | |
2792 | ||
2793 | /* We should never be about to go beyond the end of the pattern. */ | |
2794 | assert (p < pend); | |
2795 | ||
2796 | #ifdef SWITCH_ENUM_BUG | |
2797 | switch ((int) ((re_opcode_t) *p++)) | |
2798 | #else | |
2799 | switch ((re_opcode_t) *p++) | |
2800 | #endif | |
2801 | { | |
2802 | ||
2803 | /* I guess the idea here is to simply not bother with a fastmap | |
2804 | if a backreference is used, since it's too hard to figure out | |
2805 | the fastmap for the corresponding group. Setting | |
2806 | `can_be_null' stops `re_search_2' from using the fastmap, so | |
2807 | that is all we do. */ | |
2808 | case duplicate: | |
2809 | bufp->can_be_null = 1; | |
2810 | return 0; | |
2811 | ||
2812 | ||
2813 | /* Following are the cases which match a character. These end | |
2814 | with `break'. */ | |
2815 | ||
2816 | case exactn: | |
2817 | fastmap[p[1]] = 1; | |
2818 | break; | |
2819 | ||
2820 | ||
2821 | case charset: | |
2822 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2823 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
2824 | fastmap[j] = 1; | |
2825 | break; | |
2826 | ||
2827 | ||
2828 | case charset_not: | |
2829 | /* Chars beyond end of map must be allowed. */ | |
2830 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
2831 | fastmap[j] = 1; | |
2832 | ||
2833 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2834 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
2835 | fastmap[j] = 1; | |
2836 | break; | |
2837 | ||
2838 | ||
2839 | case wordchar: | |
2840 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2841 | if (SYNTAX (j) == Sword) | |
2842 | fastmap[j] = 1; | |
2843 | break; | |
2844 | ||
2845 | ||
2846 | case notwordchar: | |
2847 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2848 | if (SYNTAX (j) != Sword) | |
2849 | fastmap[j] = 1; | |
2850 | break; | |
2851 | ||
2852 | ||
2853 | case anychar: | |
73e84c96 RS |
2854 | { |
2855 | int fastmap_newline = fastmap['\n']; | |
bc78d348 | 2856 | |
73e84c96 RS |
2857 | /* `.' matches anything ... */ |
2858 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2859 | fastmap[j] = 1; | |
bc78d348 | 2860 | |
73e84c96 RS |
2861 | /* ... except perhaps newline. */ |
2862 | if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
2863 | fastmap['\n'] = fastmap_newline; | |
bc78d348 | 2864 | |
73e84c96 RS |
2865 | /* Return if we have already set `can_be_null'; if we have, |
2866 | then the fastmap is irrelevant. Something's wrong here. */ | |
2867 | else if (bufp->can_be_null) | |
2868 | return 0; | |
bc78d348 | 2869 | |
73e84c96 RS |
2870 | /* Otherwise, have to check alternative paths. */ |
2871 | break; | |
2872 | } | |
bc78d348 KB |
2873 | |
2874 | #ifdef emacs | |
2875 | case syntaxspec: | |
2876 | k = *p++; | |
2877 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2878 | if (SYNTAX (j) == (enum syntaxcode) k) | |
2879 | fastmap[j] = 1; | |
2880 | break; | |
2881 | ||
2882 | ||
2883 | case notsyntaxspec: | |
2884 | k = *p++; | |
2885 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2886 | if (SYNTAX (j) != (enum syntaxcode) k) | |
2887 | fastmap[j] = 1; | |
2888 | break; | |
2889 | ||
2890 | ||
2891 | /* All cases after this match the empty string. These end with | |
2892 | `continue'. */ | |
2893 | ||
2894 | ||
2895 | case before_dot: | |
2896 | case at_dot: | |
2897 | case after_dot: | |
2898 | continue; | |
2899 | #endif /* not emacs */ | |
2900 | ||
2901 | ||
2902 | case no_op: | |
2903 | case begline: | |
2904 | case endline: | |
2905 | case begbuf: | |
2906 | case endbuf: | |
2907 | case wordbound: | |
2908 | case notwordbound: | |
2909 | case wordbeg: | |
2910 | case wordend: | |
2911 | case push_dummy_failure: | |
2912 | continue; | |
2913 | ||
2914 | ||
2915 | case jump_n: | |
2916 | case pop_failure_jump: | |
2917 | case maybe_pop_jump: | |
2918 | case jump: | |
2919 | case jump_past_alt: | |
2920 | case dummy_failure_jump: | |
2921 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2922 | p += j; | |
2923 | if (j > 0) | |
2924 | continue; | |
2925 | ||
2926 | /* Jump backward implies we just went through the body of a | |
2927 | loop and matched nothing. Opcode jumped to should be | |
2928 | `on_failure_jump' or `succeed_n'. Just treat it like an | |
2929 | ordinary jump. For a * loop, it has pushed its failure | |
2930 | point already; if so, discard that as redundant. */ | |
2931 | if ((re_opcode_t) *p != on_failure_jump | |
2932 | && (re_opcode_t) *p != succeed_n) | |
2933 | continue; | |
2934 | ||
2935 | p++; | |
2936 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2937 | p += j; | |
2938 | ||
2939 | /* If what's on the stack is where we are now, pop it. */ | |
2940 | if (!FAIL_STACK_EMPTY () | |
2941 | && fail_stack.stack[fail_stack.avail - 1] == p) | |
2942 | fail_stack.avail--; | |
2943 | ||
2944 | continue; | |
2945 | ||
2946 | ||
2947 | case on_failure_jump: | |
2948 | case on_failure_keep_string_jump: | |
2949 | handle_on_failure_jump: | |
2950 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2951 | ||
2952 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
2953 | end of the pattern. We don't want to push such a point, | |
2954 | since when we restore it above, entering the switch will | |
2955 | increment `p' past the end of the pattern. We don't need | |
2956 | to push such a point since we obviously won't find any more | |
2957 | fastmap entries beyond `pend'. Such a pattern can match | |
2958 | the null string, though. */ | |
2959 | if (p + j < pend) | |
2960 | { | |
2961 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
2962 | return -2; | |
2963 | } | |
2964 | else | |
2965 | bufp->can_be_null = 1; | |
2966 | ||
2967 | if (succeed_n_p) | |
2968 | { | |
2969 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
2970 | succeed_n_p = false; | |
2971 | } | |
2972 | ||
2973 | continue; | |
2974 | ||
2975 | ||
2976 | case succeed_n: | |
2977 | /* Get to the number of times to succeed. */ | |
2978 | p += 2; | |
2979 | ||
2980 | /* Increment p past the n for when k != 0. */ | |
2981 | EXTRACT_NUMBER_AND_INCR (k, p); | |
2982 | if (k == 0) | |
2983 | { | |
2984 | p -= 4; | |
2985 | succeed_n_p = true; /* Spaghetti code alert. */ | |
2986 | goto handle_on_failure_jump; | |
2987 | } | |
2988 | continue; | |
2989 | ||
2990 | ||
2991 | case set_number_at: | |
2992 | p += 4; | |
2993 | continue; | |
2994 | ||
2995 | ||
2996 | case start_memory: | |
2997 | case stop_memory: | |
2998 | p += 2; | |
2999 | continue; | |
3000 | ||
3001 | ||
3002 | default: | |
3003 | abort (); /* We have listed all the cases. */ | |
3004 | } /* switch *p++ */ | |
3005 | ||
3006 | /* Getting here means we have found the possible starting | |
3007 | characters for one path of the pattern -- and that the empty | |
3008 | string does not match. We need not follow this path further. | |
3009 | Instead, look at the next alternative (remembered on the | |
3010 | stack), or quit if no more. The test at the top of the loop | |
3011 | does these things. */ | |
3012 | path_can_be_null = false; | |
3013 | p = pend; | |
3014 | } /* while p */ | |
3015 | ||
3016 | /* Set `can_be_null' for the last path (also the first path, if the | |
3017 | pattern is empty). */ | |
3018 | bufp->can_be_null |= path_can_be_null; | |
3019 | return 0; | |
3020 | } /* re_compile_fastmap */ | |
3021 | \f | |
3022 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
3023 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
3024 | this memory for recording register information. STARTS and ENDS | |
3025 | must be allocated using the malloc library routine, and must each | |
3026 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
3027 | ||
3028 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
3029 | register data. | |
3030 | ||
3031 | Unless this function is called, the first search or match using | |
3032 | PATTERN_BUFFER will allocate its own register data, without | |
3033 | freeing the old data. */ | |
3034 | ||
3035 | void | |
3036 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
3037 | struct re_pattern_buffer *bufp; | |
3038 | struct re_registers *regs; | |
3039 | unsigned num_regs; | |
3040 | regoff_t *starts, *ends; | |
3041 | { | |
3042 | if (num_regs) | |
3043 | { | |
3044 | bufp->regs_allocated = REGS_REALLOCATE; | |
3045 | regs->num_regs = num_regs; | |
3046 | regs->start = starts; | |
3047 | regs->end = ends; | |
3048 | } | |
3049 | else | |
3050 | { | |
3051 | bufp->regs_allocated = REGS_UNALLOCATED; | |
3052 | regs->num_regs = 0; | |
368481a5 | 3053 | regs->start = regs->end = (regoff_t *) 0; |
bc78d348 KB |
3054 | } |
3055 | } | |
3056 | \f | |
3057 | /* Searching routines. */ | |
3058 | ||
3059 | /* Like re_search_2, below, but only one string is specified, and | |
3060 | doesn't let you say where to stop matching. */ | |
3061 | ||
3062 | int | |
3063 | re_search (bufp, string, size, startpos, range, regs) | |
3064 | struct re_pattern_buffer *bufp; | |
3065 | const char *string; | |
3066 | int size, startpos, range; | |
3067 | struct re_registers *regs; | |
3068 | { | |
3069 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
3070 | regs, size); | |
3071 | } | |
3072 | ||
3073 | ||
3074 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
3075 | virtual concatenation of STRING1 and STRING2, starting first at index | |
3076 | STARTPOS, then at STARTPOS + 1, and so on. | |
3077 | ||
3078 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
3079 | ||
3080 | RANGE is how far to scan while trying to match. RANGE = 0 means try | |
3081 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
3082 | RANGE. | |
3083 | ||
3084 | In REGS, return the indices of the virtual concatenation of STRING1 | |
3085 | and STRING2 that matched the entire BUFP->buffer and its contained | |
3086 | subexpressions. | |
3087 | ||
3088 | Do not consider matching one past the index STOP in the virtual | |
3089 | concatenation of STRING1 and STRING2. | |
3090 | ||
3091 | We return either the position in the strings at which the match was | |
3092 | found, -1 if no match, or -2 if error (such as failure | |
3093 | stack overflow). */ | |
3094 | ||
3095 | int | |
3096 | re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
3097 | struct re_pattern_buffer *bufp; | |
3098 | const char *string1, *string2; | |
3099 | int size1, size2; | |
3100 | int startpos; | |
3101 | int range; | |
3102 | struct re_registers *regs; | |
3103 | int stop; | |
3104 | { | |
3105 | int val; | |
3106 | register char *fastmap = bufp->fastmap; | |
3107 | register char *translate = bufp->translate; | |
3108 | int total_size = size1 + size2; | |
3109 | int endpos = startpos + range; | |
3110 | ||
3111 | /* Check for out-of-range STARTPOS. */ | |
3112 | if (startpos < 0 || startpos > total_size) | |
3113 | return -1; | |
3114 | ||
3115 | /* Fix up RANGE if it might eventually take us outside | |
3116 | the virtual concatenation of STRING1 and STRING2. */ | |
3117 | if (endpos < -1) | |
3118 | range = -1 - startpos; | |
3119 | else if (endpos > total_size) | |
3120 | range = total_size - startpos; | |
3121 | ||
bc78d348 | 3122 | /* If the search isn't to be a backwards one, don't waste time in a |
9114e279 KB |
3123 | search for a pattern that must be anchored. */ |
3124 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) | |
bc78d348 KB |
3125 | { |
3126 | if (startpos > 0) | |
3127 | return -1; | |
3128 | else | |
3129 | range = 1; | |
3130 | } | |
3131 | ||
9114e279 KB |
3132 | /* Update the fastmap now if not correct already. */ |
3133 | if (fastmap && !bufp->fastmap_accurate) | |
3134 | if (re_compile_fastmap (bufp) == -2) | |
3135 | return -2; | |
3136 | ||
3137 | /* Loop through the string, looking for a place to start matching. */ | |
bc78d348 KB |
3138 | for (;;) |
3139 | { | |
3140 | /* If a fastmap is supplied, skip quickly over characters that | |
3141 | cannot be the start of a match. If the pattern can match the | |
3142 | null string, however, we don't need to skip characters; we want | |
3143 | the first null string. */ | |
3144 | if (fastmap && startpos < total_size && !bufp->can_be_null) | |
3145 | { | |
3146 | if (range > 0) /* Searching forwards. */ | |
3147 | { | |
3148 | register const char *d; | |
3149 | register int lim = 0; | |
3150 | int irange = range; | |
3151 | ||
3152 | if (startpos < size1 && startpos + range >= size1) | |
3153 | lim = range - (size1 - startpos); | |
3154 | ||
3155 | d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
3156 | ||
3157 | /* Written out as an if-else to avoid testing `translate' | |
3158 | inside the loop. */ | |
3159 | if (translate) | |
3160 | while (range > lim | |
11a364d9 JB |
3161 | && !fastmap[(unsigned char) |
3162 | translate[(unsigned char) *d++]]) | |
bc78d348 KB |
3163 | range--; |
3164 | else | |
3165 | while (range > lim && !fastmap[(unsigned char) *d++]) | |
3166 | range--; | |
3167 | ||
3168 | startpos += irange - range; | |
3169 | } | |
3170 | else /* Searching backwards. */ | |
3171 | { | |
3172 | register char c = (size1 == 0 || startpos >= size1 | |
3173 | ? string2[startpos - size1] | |
3174 | : string1[startpos]); | |
3175 | ||
9114e279 | 3176 | if (!fastmap[(unsigned char) TRANSLATE (c)]) |
bc78d348 KB |
3177 | goto advance; |
3178 | } | |
3179 | } | |
3180 | ||
3181 | /* If can't match the null string, and that's all we have left, fail. */ | |
3182 | if (range >= 0 && startpos == total_size && fastmap | |
3183 | && !bufp->can_be_null) | |
3184 | return -1; | |
3185 | ||
f765858e KH |
3186 | val = re_match_2_internal (bufp, string1, size1, string2, size2, |
3187 | startpos, regs, stop); | |
f46831b0 RS |
3188 | #ifndef REGEX_MALLOC |
3189 | #ifdef C_ALLOCA | |
f765858e | 3190 | alloca (0); |
f46831b0 RS |
3191 | #endif |
3192 | #endif | |
f765858e | 3193 | |
bc78d348 KB |
3194 | if (val >= 0) |
3195 | return startpos; | |
3196 | ||
3197 | if (val == -2) | |
3198 | return -2; | |
3199 | ||
3200 | advance: | |
3201 | if (!range) | |
3202 | break; | |
3203 | else if (range > 0) | |
3204 | { | |
3205 | range--; | |
3206 | startpos++; | |
3207 | } | |
3208 | else | |
3209 | { | |
3210 | range++; | |
3211 | startpos--; | |
3212 | } | |
3213 | } | |
3214 | return -1; | |
3215 | } /* re_search_2 */ | |
3216 | \f | |
3217 | /* Declarations and macros for re_match_2. */ | |
3218 | ||
3219 | static int bcmp_translate (); | |
3220 | static boolean alt_match_null_string_p (), | |
3221 | common_op_match_null_string_p (), | |
3222 | group_match_null_string_p (); | |
3223 | ||
bc78d348 KB |
3224 | /* This converts PTR, a pointer into one of the search strings `string1' |
3225 | and `string2' into an offset from the beginning of that string. */ | |
794ce2c1 RS |
3226 | #define POINTER_TO_OFFSET(ptr) \ |
3227 | (FIRST_STRING_P (ptr) \ | |
3228 | ? ((regoff_t) ((ptr) - string1)) \ | |
3229 | : ((regoff_t) ((ptr) - string2 + size1))) | |
bc78d348 | 3230 | |
bc78d348 KB |
3231 | /* Macros for dealing with the split strings in re_match_2. */ |
3232 | ||
3233 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
3234 | ||
3235 | /* Call before fetching a character with *d. This switches over to | |
3236 | string2 if necessary. */ | |
3237 | #define PREFETCH() \ | |
3238 | while (d == dend) \ | |
3239 | { \ | |
3240 | /* End of string2 => fail. */ \ | |
3241 | if (dend == end_match_2) \ | |
3242 | goto fail; \ | |
3243 | /* End of string1 => advance to string2. */ \ | |
3244 | d = string2; \ | |
3245 | dend = end_match_2; \ | |
3246 | } | |
3247 | ||
3248 | ||
3249 | /* Test if at very beginning or at very end of the virtual concatenation | |
3250 | of `string1' and `string2'. If only one string, it's `string2'. */ | |
9114e279 KB |
3251 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
3252 | #define AT_STRINGS_END(d) ((d) == end2) | |
bc78d348 KB |
3253 | |
3254 | ||
3255 | /* Test if D points to a character which is word-constituent. We have | |
3256 | two special cases to check for: if past the end of string1, look at | |
3257 | the first character in string2; and if before the beginning of | |
9114e279 KB |
3258 | string2, look at the last character in string1. */ |
3259 | #define WORDCHAR_P(d) \ | |
bc78d348 | 3260 | (SYNTAX ((d) == end1 ? *string2 \ |
9114e279 KB |
3261 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
3262 | == Sword) | |
bc78d348 KB |
3263 | |
3264 | /* Test if the character before D and the one at D differ with respect | |
3265 | to being word-constituent. */ | |
3266 | #define AT_WORD_BOUNDARY(d) \ | |
9114e279 KB |
3267 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
3268 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
bc78d348 KB |
3269 | |
3270 | ||
3271 | /* Free everything we malloc. */ | |
a822d92f | 3272 | #ifdef MATCH_MAY_ALLOCATE |
bc78d348 KB |
3273 | #ifdef REGEX_MALLOC |
3274 | #define FREE_VAR(var) if (var) free (var); var = NULL | |
3275 | #define FREE_VARIABLES() \ | |
3276 | do { \ | |
3277 | FREE_VAR (fail_stack.stack); \ | |
3278 | FREE_VAR (regstart); \ | |
3279 | FREE_VAR (regend); \ | |
3280 | FREE_VAR (old_regstart); \ | |
3281 | FREE_VAR (old_regend); \ | |
3282 | FREE_VAR (best_regstart); \ | |
3283 | FREE_VAR (best_regend); \ | |
3284 | FREE_VAR (reg_info); \ | |
3285 | FREE_VAR (reg_dummy); \ | |
3286 | FREE_VAR (reg_info_dummy); \ | |
3287 | } while (0) | |
3288 | #else /* not REGEX_MALLOC */ | |
f765858e KH |
3289 | /* This used to do alloca (0), but now we do that in the caller. */ |
3290 | #define FREE_VARIABLES() /* Nothing */ | |
bc78d348 | 3291 | #endif /* not REGEX_MALLOC */ |
a822d92f JB |
3292 | #else |
3293 | #define FREE_VARIABLES() /* Do nothing! */ | |
3294 | #endif /* not MATCH_MAY_ALLOCATE */ | |
bc78d348 KB |
3295 | |
3296 | /* These values must meet several constraints. They must not be valid | |
3297 | register values; since we have a limit of 255 registers (because | |
3298 | we use only one byte in the pattern for the register number), we can | |
3299 | use numbers larger than 255. They must differ by 1, because of | |
3300 | NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
3301 | be larger than the value for the highest register, so we do not try | |
3302 | to actually save any registers when none are active. */ | |
3303 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
3304 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
3305 | \f | |
3306 | /* Matching routines. */ | |
3307 | ||
3308 | #ifndef emacs /* Emacs never uses this. */ | |
3309 | /* re_match is like re_match_2 except it takes only a single string. */ | |
3310 | ||
3311 | int | |
3312 | re_match (bufp, string, size, pos, regs) | |
3313 | struct re_pattern_buffer *bufp; | |
3314 | const char *string; | |
3315 | int size, pos; | |
3316 | struct re_registers *regs; | |
f765858e KH |
3317 | { |
3318 | int result = re_match_2_internal (bufp, NULL, 0, string, size, | |
3319 | pos, regs, size); | |
3320 | alloca (0); | |
3321 | return result; | |
bc78d348 KB |
3322 | } |
3323 | #endif /* not emacs */ | |
3324 | ||
3325 | ||
3326 | /* re_match_2 matches the compiled pattern in BUFP against the | |
3327 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
3328 | and SIZE2, respectively). We start matching at POS, and stop | |
3329 | matching at STOP. | |
3330 | ||
3331 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
3332 | store offsets for the substring each group matched in REGS. See the | |
3333 | documentation for exactly how many groups we fill. | |
3334 | ||
3335 | We return -1 if no match, -2 if an internal error (such as the | |
3336 | failure stack overflowing). Otherwise, we return the length of the | |
3337 | matched substring. */ | |
3338 | ||
3339 | int | |
3340 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3341 | struct re_pattern_buffer *bufp; | |
3342 | const char *string1, *string2; | |
3343 | int size1, size2; | |
3344 | int pos; | |
3345 | struct re_registers *regs; | |
3346 | int stop; | |
f765858e KH |
3347 | { |
3348 | int result = re_match_2_internal (bufp, string1, size1, string2, size2, | |
3349 | pos, regs, stop); | |
3350 | alloca (0); | |
3351 | return result; | |
3352 | } | |
3353 | ||
3354 | /* This is a separate function so that we can force an alloca cleanup | |
3355 | afterwards. */ | |
3356 | static int | |
3357 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3358 | struct re_pattern_buffer *bufp; | |
3359 | const char *string1, *string2; | |
3360 | int size1, size2; | |
3361 | int pos; | |
3362 | struct re_registers *regs; | |
3363 | int stop; | |
bc78d348 KB |
3364 | { |
3365 | /* General temporaries. */ | |
3366 | int mcnt; | |
3367 | unsigned char *p1; | |
3368 | ||
3369 | /* Just past the end of the corresponding string. */ | |
3370 | const char *end1, *end2; | |
3371 | ||
3372 | /* Pointers into string1 and string2, just past the last characters in | |
3373 | each to consider matching. */ | |
3374 | const char *end_match_1, *end_match_2; | |
3375 | ||
3376 | /* Where we are in the data, and the end of the current string. */ | |
3377 | const char *d, *dend; | |
3378 | ||
3379 | /* Where we are in the pattern, and the end of the pattern. */ | |
3380 | unsigned char *p = bufp->buffer; | |
3381 | register unsigned char *pend = p + bufp->used; | |
3382 | ||
b391d2bf JM |
3383 | /* Mark the opcode just after a start_memory, so we can test for an |
3384 | empty subpattern when we get to the stop_memory. */ | |
3385 | unsigned char *just_past_start_mem = 0; | |
3386 | ||
bc78d348 KB |
3387 | /* We use this to map every character in the string. */ |
3388 | char *translate = bufp->translate; | |
3389 | ||
3390 | /* Failure point stack. Each place that can handle a failure further | |
3391 | down the line pushes a failure point on this stack. It consists of | |
3392 | restart, regend, and reg_info for all registers corresponding to | |
3393 | the subexpressions we're currently inside, plus the number of such | |
3394 | registers, and, finally, two char *'s. The first char * is where | |
3395 | to resume scanning the pattern; the second one is where to resume | |
3396 | scanning the strings. If the latter is zero, the failure point is | |
3397 | a ``dummy''; if a failure happens and the failure point is a dummy, | |
3398 | it gets discarded and the next next one is tried. */ | |
a822d92f | 3399 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
bc78d348 | 3400 | fail_stack_type fail_stack; |
e959bada | 3401 | #endif |
bc78d348 KB |
3402 | #ifdef DEBUG |
3403 | static unsigned failure_id = 0; | |
9114e279 | 3404 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
bc78d348 KB |
3405 | #endif |
3406 | ||
3407 | /* We fill all the registers internally, independent of what we | |
3408 | return, for use in backreferences. The number here includes | |
3409 | an element for register zero. */ | |
3410 | unsigned num_regs = bufp->re_nsub + 1; | |
3411 | ||
3412 | /* The currently active registers. */ | |
3413 | unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3414 | unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3415 | ||
3416 | /* Information on the contents of registers. These are pointers into | |
3417 | the input strings; they record just what was matched (on this | |
3418 | attempt) by a subexpression part of the pattern, that is, the | |
3419 | regnum-th regstart pointer points to where in the pattern we began | |
3420 | matching and the regnum-th regend points to right after where we | |
3421 | stopped matching the regnum-th subexpression. (The zeroth register | |
3422 | keeps track of what the whole pattern matches.) */ | |
a822d92f | 3423 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
bc78d348 | 3424 | const char **regstart, **regend; |
e959bada | 3425 | #endif |
bc78d348 KB |
3426 | |
3427 | /* If a group that's operated upon by a repetition operator fails to | |
3428 | match anything, then the register for its start will need to be | |
3429 | restored because it will have been set to wherever in the string we | |
3430 | are when we last see its open-group operator. Similarly for a | |
3431 | register's end. */ | |
a822d92f | 3432 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
bc78d348 | 3433 | const char **old_regstart, **old_regend; |
e959bada | 3434 | #endif |
bc78d348 KB |
3435 | |
3436 | /* The is_active field of reg_info helps us keep track of which (possibly | |
3437 | nested) subexpressions we are currently in. The matched_something | |
3438 | field of reg_info[reg_num] helps us tell whether or not we have | |
3439 | matched any of the pattern so far this time through the reg_num-th | |
3440 | subexpression. These two fields get reset each time through any | |
3441 | loop their register is in. */ | |
a822d92f | 3442 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
bc78d348 | 3443 | register_info_type *reg_info; |
e959bada | 3444 | #endif |
bc78d348 KB |
3445 | |
3446 | /* The following record the register info as found in the above | |
3447 | variables when we find a match better than any we've seen before. | |
3448 | This happens as we backtrack through the failure points, which in | |
3449 | turn happens only if we have not yet matched the entire string. */ | |
3450 | unsigned best_regs_set = false; | |
a822d92f | 3451 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
bc78d348 | 3452 | const char **best_regstart, **best_regend; |
e959bada | 3453 | #endif |
bc78d348 KB |
3454 | |
3455 | /* Logically, this is `best_regend[0]'. But we don't want to have to | |
3456 | allocate space for that if we're not allocating space for anything | |
3457 | else (see below). Also, we never need info about register 0 for | |
3458 | any of the other register vectors, and it seems rather a kludge to | |
3459 | treat `best_regend' differently than the rest. So we keep track of | |
3460 | the end of the best match so far in a separate variable. We | |
3461 | initialize this to NULL so that when we backtrack the first time | |
3462 | and need to test it, it's not garbage. */ | |
3463 | const char *match_end = NULL; | |
3464 | ||
3465 | /* Used when we pop values we don't care about. */ | |
a822d92f | 3466 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
bc78d348 KB |
3467 | const char **reg_dummy; |
3468 | register_info_type *reg_info_dummy; | |
e959bada | 3469 | #endif |
bc78d348 KB |
3470 | |
3471 | #ifdef DEBUG | |
3472 | /* Counts the total number of registers pushed. */ | |
3473 | unsigned num_regs_pushed = 0; | |
3474 | #endif | |
3475 | ||
3476 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
3477 | ||
3478 | INIT_FAIL_STACK (); | |
3479 | ||
a822d92f | 3480 | #ifdef MATCH_MAY_ALLOCATE |
bc78d348 KB |
3481 | /* Do not bother to initialize all the register variables if there are |
3482 | no groups in the pattern, as it takes a fair amount of time. If | |
3483 | there are groups, we include space for register 0 (the whole | |
3484 | pattern), even though we never use it, since it simplifies the | |
3485 | array indexing. We should fix this. */ | |
3486 | if (bufp->re_nsub) | |
3487 | { | |
3488 | regstart = REGEX_TALLOC (num_regs, const char *); | |
3489 | regend = REGEX_TALLOC (num_regs, const char *); | |
3490 | old_regstart = REGEX_TALLOC (num_regs, const char *); | |
3491 | old_regend = REGEX_TALLOC (num_regs, const char *); | |
3492 | best_regstart = REGEX_TALLOC (num_regs, const char *); | |
3493 | best_regend = REGEX_TALLOC (num_regs, const char *); | |
3494 | reg_info = REGEX_TALLOC (num_regs, register_info_type); | |
3495 | reg_dummy = REGEX_TALLOC (num_regs, const char *); | |
3496 | reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); | |
3497 | ||
3498 | if (!(regstart && regend && old_regstart && old_regend && reg_info | |
3499 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
3500 | { | |
3501 | FREE_VARIABLES (); | |
3502 | return -2; | |
3503 | } | |
3504 | } | |
e959bada | 3505 | #if defined (REGEX_MALLOC) |
bc78d348 KB |
3506 | else |
3507 | { | |
3508 | /* We must initialize all our variables to NULL, so that | |
9114e279 | 3509 | `FREE_VARIABLES' doesn't try to free them. */ |
bc78d348 KB |
3510 | regstart = regend = old_regstart = old_regend = best_regstart |
3511 | = best_regend = reg_dummy = NULL; | |
3512 | reg_info = reg_info_dummy = (register_info_type *) NULL; | |
3513 | } | |
3514 | #endif /* REGEX_MALLOC */ | |
a822d92f | 3515 | #endif /* MATCH_MAY_ALLOCATE */ |
bc78d348 KB |
3516 | |
3517 | /* The starting position is bogus. */ | |
3518 | if (pos < 0 || pos > size1 + size2) | |
3519 | { | |
3520 | FREE_VARIABLES (); | |
3521 | return -1; | |
3522 | } | |
3523 | ||
3524 | /* Initialize subexpression text positions to -1 to mark ones that no | |
3525 | start_memory/stop_memory has been seen for. Also initialize the | |
3526 | register information struct. */ | |
3527 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3528 | { | |
3529 | regstart[mcnt] = regend[mcnt] | |
3530 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
3531 | ||
3532 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
3533 | IS_ACTIVE (reg_info[mcnt]) = 0; | |
3534 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3535 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3536 | } | |
3537 | ||
3538 | /* We move `string1' into `string2' if the latter's empty -- but not if | |
3539 | `string1' is null. */ | |
3540 | if (size2 == 0 && string1 != NULL) | |
3541 | { | |
3542 | string2 = string1; | |
3543 | size2 = size1; | |
3544 | string1 = 0; | |
3545 | size1 = 0; | |
3546 | } | |
3547 | end1 = string1 + size1; | |
3548 | end2 = string2 + size2; | |
3549 | ||
3550 | /* Compute where to stop matching, within the two strings. */ | |
3551 | if (stop <= size1) | |
3552 | { | |
3553 | end_match_1 = string1 + stop; | |
3554 | end_match_2 = string2; | |
3555 | } | |
3556 | else | |
3557 | { | |
3558 | end_match_1 = end1; | |
3559 | end_match_2 = string2 + stop - size1; | |
3560 | } | |
3561 | ||
3562 | /* `p' scans through the pattern as `d' scans through the data. | |
3563 | `dend' is the end of the input string that `d' points within. `d' | |
3564 | is advanced into the following input string whenever necessary, but | |
3565 | this happens before fetching; therefore, at the beginning of the | |
3566 | loop, `d' can be pointing at the end of a string, but it cannot | |
3567 | equal `string2'. */ | |
3568 | if (size1 > 0 && pos <= size1) | |
3569 | { | |
3570 | d = string1 + pos; | |
3571 | dend = end_match_1; | |
3572 | } | |
3573 | else | |
3574 | { | |
3575 | d = string2 + pos - size1; | |
3576 | dend = end_match_2; | |
3577 | } | |
3578 | ||
3579 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
3580 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
3581 | DEBUG_PRINT1 ("The string to match is: `"); | |
3582 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
3583 | DEBUG_PRINT1 ("'\n"); | |
3584 | ||
3585 | /* This loops over pattern commands. It exits by returning from the | |
3586 | function if the match is complete, or it drops through if the match | |
3587 | fails at this starting point in the input data. */ | |
3588 | for (;;) | |
3589 | { | |
3590 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
3591 | ||
3592 | if (p == pend) | |
3593 | { /* End of pattern means we might have succeeded. */ | |
9114e279 KB |
3594 | DEBUG_PRINT1 ("end of pattern ... "); |
3595 | ||
3596 | /* If we haven't matched the entire string, and we want the | |
3597 | longest match, try backtracking. */ | |
bc78d348 KB |
3598 | if (d != end_match_2) |
3599 | { | |
73e84c96 RS |
3600 | /* 1 if this match ends in the same string (string1 or string2) |
3601 | as the best previous match. */ | |
3602 | boolean same_str_p = (FIRST_STRING_P (match_end) | |
3603 | == MATCHING_IN_FIRST_STRING); | |
3604 | /* 1 if this match is the best seen so far. */ | |
3605 | boolean best_match_p = (same_str_p ? d > match_end | |
3606 | : !MATCHING_IN_FIRST_STRING); | |
3607 | ||
bc78d348 KB |
3608 | DEBUG_PRINT1 ("backtracking.\n"); |
3609 | ||
3610 | if (!FAIL_STACK_EMPTY ()) | |
3611 | { /* More failure points to try. */ | |
bc78d348 KB |
3612 | |
3613 | /* If exceeds best match so far, save it. */ | |
73e84c96 | 3614 | if (!best_regs_set || best_match_p) |
bc78d348 KB |
3615 | { |
3616 | best_regs_set = true; | |
3617 | match_end = d; | |
3618 | ||
3619 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
3620 | ||
3621 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3622 | { | |
3623 | best_regstart[mcnt] = regstart[mcnt]; | |
3624 | best_regend[mcnt] = regend[mcnt]; | |
3625 | } | |
3626 | } | |
3627 | goto fail; | |
3628 | } | |
3629 | ||
73e84c96 RS |
3630 | /* If no failure points, don't restore garbage. And if |
3631 | last match is real best match, don't restore second | |
3632 | best one. */ | |
3633 | else if (best_regs_set && !best_match_p) | |
bc78d348 KB |
3634 | { |
3635 | restore_best_regs: | |
3636 | /* Restore best match. It may happen that `dend == | |
3637 | end_match_1' while the restored d is in string2. | |
3638 | For example, the pattern `x.*y.*z' against the | |
3639 | strings `x-' and `y-z-', if the two strings are | |
3640 | not consecutive in memory. */ | |
9114e279 KB |
3641 | DEBUG_PRINT1 ("Restoring best registers.\n"); |
3642 | ||
bc78d348 KB |
3643 | d = match_end; |
3644 | dend = ((d >= string1 && d <= end1) | |
3645 | ? end_match_1 : end_match_2); | |
3646 | ||
3647 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3648 | { | |
3649 | regstart[mcnt] = best_regstart[mcnt]; | |
3650 | regend[mcnt] = best_regend[mcnt]; | |
3651 | } | |
3652 | } | |
3653 | } /* d != end_match_2 */ | |
3654 | ||
9114e279 | 3655 | DEBUG_PRINT1 ("Accepting match.\n"); |
bc78d348 KB |
3656 | |
3657 | /* If caller wants register contents data back, do it. */ | |
3658 | if (regs && !bufp->no_sub) | |
3659 | { | |
3660 | /* Have the register data arrays been allocated? */ | |
3661 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
3662 | { /* No. So allocate them with malloc. We need one | |
3663 | extra element beyond `num_regs' for the `-1' marker | |
3664 | GNU code uses. */ | |
3665 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
3666 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
3667 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
3668 | if (regs->start == NULL || regs->end == NULL) | |
3669 | return -2; | |
3670 | bufp->regs_allocated = REGS_REALLOCATE; | |
3671 | } | |
3672 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
3673 | { /* Yes. If we need more elements than were already | |
3674 | allocated, reallocate them. If we need fewer, just | |
3675 | leave it alone. */ | |
3676 | if (regs->num_regs < num_regs + 1) | |
3677 | { | |
3678 | regs->num_regs = num_regs + 1; | |
3679 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
3680 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
3681 | if (regs->start == NULL || regs->end == NULL) | |
3682 | return -2; | |
3683 | } | |
3684 | } | |
3685 | else | |
a1dd004b JB |
3686 | { |
3687 | /* These braces fend off a "empty body in an else-statement" | |
3688 | warning under GCC when assert expands to nothing. */ | |
3689 | assert (bufp->regs_allocated == REGS_FIXED); | |
3690 | } | |
bc78d348 KB |
3691 | |
3692 | /* Convert the pointer data in `regstart' and `regend' to | |
3693 | indices. Register zero has to be set differently, | |
3694 | since we haven't kept track of any info for it. */ | |
3695 | if (regs->num_regs > 0) | |
3696 | { | |
3697 | regs->start[0] = pos; | |
794ce2c1 RS |
3698 | regs->end[0] = (MATCHING_IN_FIRST_STRING |
3699 | ? ((regoff_t) (d - string1)) | |
3700 | : ((regoff_t) (d - string2 + size1))); | |
bc78d348 KB |
3701 | } |
3702 | ||
3703 | /* Go through the first `min (num_regs, regs->num_regs)' | |
3704 | registers, since that is all we initialized. */ | |
3705 | for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) | |
3706 | { | |
3707 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
3708 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3709 | else | |
3710 | { | |
794ce2c1 RS |
3711 | regs->start[mcnt] |
3712 | = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); | |
3713 | regs->end[mcnt] | |
3714 | = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); | |
bc78d348 KB |
3715 | } |
3716 | } | |
3717 | ||
3718 | /* If the regs structure we return has more elements than | |
3719 | were in the pattern, set the extra elements to -1. If | |
3720 | we (re)allocated the registers, this is the case, | |
3721 | because we always allocate enough to have at least one | |
3722 | -1 at the end. */ | |
3723 | for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) | |
3724 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3725 | } /* regs && !bufp->no_sub */ | |
3726 | ||
3727 | FREE_VARIABLES (); | |
9114e279 KB |
3728 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
3729 | nfailure_points_pushed, nfailure_points_popped, | |
3730 | nfailure_points_pushed - nfailure_points_popped); | |
3731 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
bc78d348 KB |
3732 | |
3733 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
3734 | ? string1 | |
3735 | : string2 - size1); | |
3736 | ||
3737 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
3738 | ||
3739 | return mcnt; | |
3740 | } | |
3741 | ||
3742 | /* Otherwise match next pattern command. */ | |
3743 | #ifdef SWITCH_ENUM_BUG | |
3744 | switch ((int) ((re_opcode_t) *p++)) | |
3745 | #else | |
3746 | switch ((re_opcode_t) *p++) | |
3747 | #endif | |
3748 | { | |
3749 | /* Ignore these. Used to ignore the n of succeed_n's which | |
3750 | currently have n == 0. */ | |
3751 | case no_op: | |
3752 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
3753 | break; | |
3754 | ||
3755 | ||
3756 | /* Match the next n pattern characters exactly. The following | |
3757 | byte in the pattern defines n, and the n bytes after that | |
3758 | are the characters to match. */ | |
3759 | case exactn: | |
3760 | mcnt = *p++; | |
3761 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
3762 | ||
3763 | /* This is written out as an if-else so we don't waste time | |
3764 | testing `translate' inside the loop. */ | |
3765 | if (translate) | |
3766 | { | |
3767 | do | |
3768 | { | |
3769 | PREFETCH (); | |
3770 | if (translate[(unsigned char) *d++] != (char) *p++) | |
3771 | goto fail; | |
3772 | } | |
3773 | while (--mcnt); | |
3774 | } | |
3775 | else | |
3776 | { | |
3777 | do | |
3778 | { | |
3779 | PREFETCH (); | |
3780 | if (*d++ != (char) *p++) goto fail; | |
3781 | } | |
3782 | while (--mcnt); | |
3783 | } | |
3784 | SET_REGS_MATCHED (); | |
3785 | break; | |
3786 | ||
3787 | ||
3788 | /* Match any character except possibly a newline or a null. */ | |
3789 | case anychar: | |
3790 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
3791 | ||
3792 | PREFETCH (); | |
3793 | ||
3794 | if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
3795 | || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
3796 | goto fail; | |
3797 | ||
3798 | SET_REGS_MATCHED (); | |
3799 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); | |
3800 | d++; | |
3801 | break; | |
3802 | ||
3803 | ||
3804 | case charset: | |
3805 | case charset_not: | |
3806 | { | |
3807 | register unsigned char c; | |
3808 | boolean not = (re_opcode_t) *(p - 1) == charset_not; | |
3809 | ||
3810 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
3811 | ||
3812 | PREFETCH (); | |
3813 | c = TRANSLATE (*d); /* The character to match. */ | |
3814 | ||
3815 | /* Cast to `unsigned' instead of `unsigned char' in case the | |
3816 | bit list is a full 32 bytes long. */ | |
3817 | if (c < (unsigned) (*p * BYTEWIDTH) | |
3818 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
3819 | not = !not; | |
3820 | ||
3821 | p += 1 + *p; | |
3822 | ||
3823 | if (!not) goto fail; | |
3824 | ||
3825 | SET_REGS_MATCHED (); | |
3826 | d++; | |
3827 | break; | |
3828 | } | |
3829 | ||
3830 | ||
3831 | /* The beginning of a group is represented by start_memory. | |
3832 | The arguments are the register number in the next byte, and the | |
3833 | number of groups inner to this one in the next. The text | |
3834 | matched within the group is recorded (in the internal | |
3835 | registers data structure) under the register number. */ | |
3836 | case start_memory: | |
3837 | DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); | |
3838 | ||
3839 | /* Find out if this group can match the empty string. */ | |
3840 | p1 = p; /* To send to group_match_null_string_p. */ | |
3841 | ||
3842 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
3843 | REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3844 | = group_match_null_string_p (&p1, pend, reg_info); | |
3845 | ||
3846 | /* Save the position in the string where we were the last time | |
3847 | we were at this open-group operator in case the group is | |
3848 | operated upon by a repetition operator, e.g., with `(a*)*b' | |
3849 | against `ab'; then we want to ignore where we are now in | |
3850 | the string in case this attempt to match fails. */ | |
3851 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3852 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
3853 | : regstart[*p]; | |
3854 | DEBUG_PRINT2 (" old_regstart: %d\n", | |
3855 | POINTER_TO_OFFSET (old_regstart[*p])); | |
3856 | ||
3857 | regstart[*p] = d; | |
3858 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
3859 | ||
3860 | IS_ACTIVE (reg_info[*p]) = 1; | |
3861 | MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3862 | ||
3863 | /* This is the new highest active register. */ | |
3864 | highest_active_reg = *p; | |
3865 | ||
3866 | /* If nothing was active before, this is the new lowest active | |
3867 | register. */ | |
3868 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
3869 | lowest_active_reg = *p; | |
3870 | ||
3871 | /* Move past the register number and inner group count. */ | |
3872 | p += 2; | |
b391d2bf | 3873 | just_past_start_mem = p; |
bc78d348 KB |
3874 | break; |
3875 | ||
3876 | ||
3877 | /* The stop_memory opcode represents the end of a group. Its | |
3878 | arguments are the same as start_memory's: the register | |
3879 | number, and the number of inner groups. */ | |
3880 | case stop_memory: | |
3881 | DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); | |
3882 | ||
3883 | /* We need to save the string position the last time we were at | |
3884 | this close-group operator in case the group is operated | |
3885 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
3886 | against `aba'; then we want to ignore where we are now in | |
3887 | the string in case this attempt to match fails. */ | |
3888 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3889 | ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
3890 | : regend[*p]; | |
3891 | DEBUG_PRINT2 (" old_regend: %d\n", | |
3892 | POINTER_TO_OFFSET (old_regend[*p])); | |
3893 | ||
3894 | regend[*p] = d; | |
3895 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
3896 | ||
3897 | /* This register isn't active anymore. */ | |
3898 | IS_ACTIVE (reg_info[*p]) = 0; | |
3899 | ||
3900 | /* If this was the only register active, nothing is active | |
3901 | anymore. */ | |
3902 | if (lowest_active_reg == highest_active_reg) | |
3903 | { | |
3904 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3905 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3906 | } | |
3907 | else | |
3908 | { /* We must scan for the new highest active register, since | |
3909 | it isn't necessarily one less than now: consider | |
3910 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
3911 | new highest active register is 1. */ | |
3912 | unsigned char r = *p - 1; | |
3913 | while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
3914 | r--; | |
3915 | ||
3916 | /* If we end up at register zero, that means that we saved | |
3917 | the registers as the result of an `on_failure_jump', not | |
3918 | a `start_memory', and we jumped to past the innermost | |
3919 | `stop_memory'. For example, in ((.)*) we save | |
3920 | registers 1 and 2 as a result of the *, but when we pop | |
3921 | back to the second ), we are at the stop_memory 1. | |
3922 | Thus, nothing is active. */ | |
3923 | if (r == 0) | |
3924 | { | |
3925 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3926 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3927 | } | |
3928 | else | |
3929 | highest_active_reg = r; | |
3930 | } | |
3931 | ||
3932 | /* If just failed to match something this time around with a | |
3933 | group that's operated on by a repetition operator, try to | |
9114e279 | 3934 | force exit from the ``loop'', and restore the register |
bc78d348 KB |
3935 | information for this group that we had before trying this |
3936 | last match. */ | |
3937 | if ((!MATCHED_SOMETHING (reg_info[*p]) | |
b391d2bf | 3938 | || just_past_start_mem == p - 1) |
bc78d348 KB |
3939 | && (p + 2) < pend) |
3940 | { | |
3941 | boolean is_a_jump_n = false; | |
3942 | ||
3943 | p1 = p + 2; | |
3944 | mcnt = 0; | |
3945 | switch ((re_opcode_t) *p1++) | |
3946 | { | |
3947 | case jump_n: | |
3948 | is_a_jump_n = true; | |
3949 | case pop_failure_jump: | |
3950 | case maybe_pop_jump: | |
3951 | case jump: | |
3952 | case dummy_failure_jump: | |
3953 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3954 | if (is_a_jump_n) | |
3955 | p1 += 2; | |
3956 | break; | |
3957 | ||
3958 | default: | |
3959 | /* do nothing */ ; | |
3960 | } | |
3961 | p1 += mcnt; | |
3962 | ||
3963 | /* If the next operation is a jump backwards in the pattern | |
3964 | to an on_failure_jump right before the start_memory | |
3965 | corresponding to this stop_memory, exit from the loop | |
3966 | by forcing a failure after pushing on the stack the | |
3967 | on_failure_jump's jump in the pattern, and d. */ | |
3968 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
3969 | && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
3970 | { | |
3971 | /* If this group ever matched anything, then restore | |
3972 | what its registers were before trying this last | |
3973 | failed match, e.g., with `(a*)*b' against `ab' for | |
3974 | regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
3975 | against `aba' for regend[3]. | |
3976 | ||
3977 | Also restore the registers for inner groups for, | |
3978 | e.g., `((a*)(b*))*' against `aba' (register 3 would | |
3979 | otherwise get trashed). */ | |
3980 | ||
3981 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
3982 | { | |
3983 | unsigned r; | |
3984 | ||
3985 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3986 | ||
3987 | /* Restore this and inner groups' (if any) registers. */ | |
3988 | for (r = *p; r < *p + *(p + 1); r++) | |
3989 | { | |
3990 | regstart[r] = old_regstart[r]; | |
3991 | ||
3992 | /* xx why this test? */ | |
3993 | if ((int) old_regend[r] >= (int) regstart[r]) | |
3994 | regend[r] = old_regend[r]; | |
3995 | } | |
3996 | } | |
3997 | p1++; | |
3998 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3999 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
4000 | ||
4001 | goto fail; | |
4002 | } | |
4003 | } | |
4004 | ||
4005 | /* Move past the register number and the inner group count. */ | |
4006 | p += 2; | |
4007 | break; | |
4008 | ||
4009 | ||
4010 | /* \<digit> has been turned into a `duplicate' command which is | |
4011 | followed by the numeric value of <digit> as the register number. */ | |
4012 | case duplicate: | |
4013 | { | |
4014 | register const char *d2, *dend2; | |
4015 | int regno = *p++; /* Get which register to match against. */ | |
4016 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
4017 | ||
4018 | /* Can't back reference a group which we've never matched. */ | |
4019 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
4020 | goto fail; | |
4021 | ||
4022 | /* Where in input to try to start matching. */ | |
4023 | d2 = regstart[regno]; | |
4024 | ||
4025 | /* Where to stop matching; if both the place to start and | |
4026 | the place to stop matching are in the same string, then | |
4027 | set to the place to stop, otherwise, for now have to use | |
4028 | the end of the first string. */ | |
4029 | ||
4030 | dend2 = ((FIRST_STRING_P (regstart[regno]) | |
4031 | == FIRST_STRING_P (regend[regno])) | |
4032 | ? regend[regno] : end_match_1); | |
4033 | for (;;) | |
4034 | { | |
4035 | /* If necessary, advance to next segment in register | |
4036 | contents. */ | |
4037 | while (d2 == dend2) | |
4038 | { | |
4039 | if (dend2 == end_match_2) break; | |
4040 | if (dend2 == regend[regno]) break; | |
4041 | ||
4042 | /* End of string1 => advance to string2. */ | |
4043 | d2 = string2; | |
4044 | dend2 = regend[regno]; | |
4045 | } | |
4046 | /* At end of register contents => success */ | |
4047 | if (d2 == dend2) break; | |
4048 | ||
4049 | /* If necessary, advance to next segment in data. */ | |
4050 | PREFETCH (); | |
4051 | ||
4052 | /* How many characters left in this segment to match. */ | |
4053 | mcnt = dend - d; | |
4054 | ||
4055 | /* Want how many consecutive characters we can match in | |
4056 | one shot, so, if necessary, adjust the count. */ | |
4057 | if (mcnt > dend2 - d2) | |
4058 | mcnt = dend2 - d2; | |
4059 | ||
4060 | /* Compare that many; failure if mismatch, else move | |
4061 | past them. */ | |
4062 | if (translate | |
4063 | ? bcmp_translate (d, d2, mcnt, translate) | |
4064 | : bcmp (d, d2, mcnt)) | |
4065 | goto fail; | |
4066 | d += mcnt, d2 += mcnt; | |
4067 | } | |
4068 | } | |
4069 | break; | |
4070 | ||
4071 | ||
4072 | /* begline matches the empty string at the beginning of the string | |
4073 | (unless `not_bol' is set in `bufp'), and, if | |
4074 | `newline_anchor' is set, after newlines. */ | |
4075 | case begline: | |
4076 | DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
4077 | ||
9114e279 | 4078 | if (AT_STRINGS_BEG (d)) |
bc78d348 KB |
4079 | { |
4080 | if (!bufp->not_bol) break; | |
4081 | } | |
4082 | else if (d[-1] == '\n' && bufp->newline_anchor) | |
4083 | { | |
4084 | break; | |
4085 | } | |
4086 | /* In all other cases, we fail. */ | |
4087 | goto fail; | |
4088 | ||
4089 | ||
4090 | /* endline is the dual of begline. */ | |
4091 | case endline: | |
4092 | DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
4093 | ||
9114e279 | 4094 | if (AT_STRINGS_END (d)) |
bc78d348 KB |
4095 | { |
4096 | if (!bufp->not_eol) break; | |
4097 | } | |
4098 | ||
4099 | /* We have to ``prefetch'' the next character. */ | |
4100 | else if ((d == end1 ? *string2 : *d) == '\n' | |
4101 | && bufp->newline_anchor) | |
4102 | { | |
4103 | break; | |
4104 | } | |
4105 | goto fail; | |
4106 | ||
4107 | ||
4108 | /* Match at the very beginning of the data. */ | |
4109 | case begbuf: | |
4110 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
9114e279 | 4111 | if (AT_STRINGS_BEG (d)) |
bc78d348 KB |
4112 | break; |
4113 | goto fail; | |
4114 | ||
4115 | ||
4116 | /* Match at the very end of the data. */ | |
4117 | case endbuf: | |
4118 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
9114e279 | 4119 | if (AT_STRINGS_END (d)) |
bc78d348 KB |
4120 | break; |
4121 | goto fail; | |
4122 | ||
4123 | ||
4124 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
4125 | pushes NULL as the value for the string on the stack. Then | |
4126 | `pop_failure_point' will keep the current value for the | |
4127 | string, instead of restoring it. To see why, consider | |
4128 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
4129 | then the . fails against the \n. But the next thing we want | |
4130 | to do is match the \n against the \n; if we restored the | |
4131 | string value, we would be back at the foo. | |
4132 | ||
4133 | Because this is used only in specific cases, we don't need to | |
4134 | check all the things that `on_failure_jump' does, to make | |
4135 | sure the right things get saved on the stack. Hence we don't | |
4136 | share its code. The only reason to push anything on the | |
4137 | stack at all is that otherwise we would have to change | |
4138 | `anychar's code to do something besides goto fail in this | |
4139 | case; that seems worse than this. */ | |
4140 | case on_failure_keep_string_jump: | |
4141 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
4142 | ||
4143 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4144 | DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
4145 | ||
4146 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
4147 | break; | |
4148 | ||
4149 | ||
4150 | /* Uses of on_failure_jump: | |
4151 | ||
4152 | Each alternative starts with an on_failure_jump that points | |
4153 | to the beginning of the next alternative. Each alternative | |
4154 | except the last ends with a jump that in effect jumps past | |
4155 | the rest of the alternatives. (They really jump to the | |
4156 | ending jump of the following alternative, because tensioning | |
4157 | these jumps is a hassle.) | |
4158 | ||
4159 | Repeats start with an on_failure_jump that points past both | |
4160 | the repetition text and either the following jump or | |
4161 | pop_failure_jump back to this on_failure_jump. */ | |
4162 | case on_failure_jump: | |
4163 | on_failure: | |
4164 | DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
4165 | ||
4166 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4167 | DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
4168 | ||
4169 | /* If this on_failure_jump comes right before a group (i.e., | |
4170 | the original * applied to a group), save the information | |
4171 | for that group and all inner ones, so that if we fail back | |
4172 | to this point, the group's information will be correct. | |
9114e279 | 4173 | For example, in \(a*\)*\1, we need the preceding group, |
bc78d348 KB |
4174 | and in \(\(a*\)b*\)\2, we need the inner group. */ |
4175 | ||
4176 | /* We can't use `p' to check ahead because we push | |
4177 | a failure point to `p + mcnt' after we do this. */ | |
4178 | p1 = p; | |
4179 | ||
4180 | /* We need to skip no_op's before we look for the | |
4181 | start_memory in case this on_failure_jump is happening as | |
4182 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
4183 | against aba. */ | |
4184 | while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
4185 | p1++; | |
4186 | ||
4187 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
4188 | { | |
4189 | /* We have a new highest active register now. This will | |
4190 | get reset at the start_memory we are about to get to, | |
4191 | but we will have saved all the registers relevant to | |
4192 | this repetition op, as described above. */ | |
4193 | highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
4194 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
4195 | lowest_active_reg = *(p1 + 1); | |
4196 | } | |
4197 | ||
4198 | DEBUG_PRINT1 (":\n"); | |
4199 | PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
4200 | break; | |
4201 | ||
4202 | ||
9114e279 KB |
4203 | /* A smart repeat ends with `maybe_pop_jump'. |
4204 | We change it to either `pop_failure_jump' or `jump'. */ | |
bc78d348 KB |
4205 | case maybe_pop_jump: |
4206 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4207 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
4208 | { | |
4209 | register unsigned char *p2 = p; | |
4210 | ||
4211 | /* Compare the beginning of the repeat with what in the | |
4212 | pattern follows its end. If we can establish that there | |
4213 | is nothing that they would both match, i.e., that we | |
4214 | would have to backtrack because of (as in, e.g., `a*a') | |
4215 | then we can change to pop_failure_jump, because we'll | |
4216 | never have to backtrack. | |
4217 | ||
4218 | This is not true in the case of alternatives: in | |
4219 | `(a|ab)*' we do need to backtrack to the `ab' alternative | |
4220 | (e.g., if the string was `ab'). But instead of trying to | |
4221 | detect that here, the alternative has put on a dummy | |
4222 | failure point which is what we will end up popping. */ | |
4223 | ||
4f9d3ad8 RS |
4224 | /* Skip over open/close-group commands. |
4225 | If what follows this loop is a ...+ construct, | |
4226 | look at what begins its body, since we will have to | |
4227 | match at least one of that. */ | |
4228 | while (1) | |
4229 | { | |
4230 | if (p2 + 2 < pend | |
4231 | && ((re_opcode_t) *p2 == stop_memory | |
4232 | || (re_opcode_t) *p2 == start_memory)) | |
4233 | p2 += 3; | |
4234 | else if (p2 + 6 < pend | |
4235 | && (re_opcode_t) *p2 == dummy_failure_jump) | |
4236 | p2 += 6; | |
4237 | else | |
4238 | break; | |
4239 | } | |
4240 | ||
4241 | p1 = p + mcnt; | |
4242 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
4243 | to the `maybe_finalize_jump' of this case. Examine what | |
4244 | follows. */ | |
bc78d348 KB |
4245 | |
4246 | /* If we're at the end of the pattern, we can change. */ | |
4247 | if (p2 == pend) | |
71715da9 JB |
4248 | { |
4249 | /* Consider what happens when matching ":\(.*\)" | |
4250 | against ":/". I don't really understand this code | |
4251 | yet. */ | |
bc78d348 | 4252 | p[-3] = (unsigned char) pop_failure_jump; |
71715da9 JB |
4253 | DEBUG_PRINT1 |
4254 | (" End of pattern: change to `pop_failure_jump'.\n"); | |
bc78d348 KB |
4255 | } |
4256 | ||
4257 | else if ((re_opcode_t) *p2 == exactn | |
4258 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
4259 | { | |
4260 | register unsigned char c | |
4261 | = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
bc78d348 | 4262 | |
bc78d348 | 4263 | if ((re_opcode_t) p1[3] == exactn && p1[5] != c) |
9114e279 KB |
4264 | { |
4265 | p[-3] = (unsigned char) pop_failure_jump; | |
4266 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
4267 | c, p1[5]); | |
4268 | } | |
4269 | ||
bc78d348 KB |
4270 | else if ((re_opcode_t) p1[3] == charset |
4271 | || (re_opcode_t) p1[3] == charset_not) | |
4272 | { | |
4273 | int not = (re_opcode_t) p1[3] == charset_not; | |
4274 | ||
4275 | if (c < (unsigned char) (p1[4] * BYTEWIDTH) | |
4276 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4277 | not = !not; | |
4278 | ||
4279 | /* `not' is equal to 1 if c would match, which means | |
4280 | that we can't change to pop_failure_jump. */ | |
4281 | if (!not) | |
4282 | { | |
4283 | p[-3] = (unsigned char) pop_failure_jump; | |
9114e279 | 4284 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
bc78d348 KB |
4285 | } |
4286 | } | |
4287 | } | |
4f9d3ad8 RS |
4288 | else if ((re_opcode_t) *p2 == charset) |
4289 | { | |
7babf5f6 | 4290 | #ifdef DEBUG |
4f9d3ad8 RS |
4291 | register unsigned char c |
4292 | = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
7babf5f6 | 4293 | #endif |
4f9d3ad8 RS |
4294 | |
4295 | if ((re_opcode_t) p1[3] == exactn | |
5330e99c | 4296 | && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4] |
4f9d3ad8 RS |
4297 | && (p2[1 + p1[4] / BYTEWIDTH] |
4298 | & (1 << (p1[4] % BYTEWIDTH))))) | |
4299 | { | |
4300 | p[-3] = (unsigned char) pop_failure_jump; | |
4301 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
4302 | c, p1[5]); | |
4303 | } | |
4304 | ||
4305 | else if ((re_opcode_t) p1[3] == charset_not) | |
4306 | { | |
4307 | int idx; | |
4308 | /* We win if the charset_not inside the loop | |
4309 | lists every character listed in the charset after. */ | |
5330e99c | 4310 | for (idx = 0; idx < (int) p2[1]; idx++) |
4f9d3ad8 | 4311 | if (! (p2[2 + idx] == 0 |
5330e99c | 4312 | || (idx < (int) p1[4] |
4f9d3ad8 RS |
4313 | && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) |
4314 | break; | |
4315 | ||
4316 | if (idx == p2[1]) | |
4317 | { | |
4318 | p[-3] = (unsigned char) pop_failure_jump; | |
4319 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
4320 | } | |
4321 | } | |
4322 | else if ((re_opcode_t) p1[3] == charset) | |
4323 | { | |
4324 | int idx; | |
4325 | /* We win if the charset inside the loop | |
4326 | has no overlap with the one after the loop. */ | |
5330e99c RS |
4327 | for (idx = 0; |
4328 | idx < (int) p2[1] && idx < (int) p1[4]; | |
4329 | idx++) | |
4f9d3ad8 RS |
4330 | if ((p2[2 + idx] & p1[5 + idx]) != 0) |
4331 | break; | |
4332 | ||
4333 | if (idx == p2[1] || idx == p1[4]) | |
4334 | { | |
4335 | p[-3] = (unsigned char) pop_failure_jump; | |
4336 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
4337 | } | |
4338 | } | |
4339 | } | |
bc78d348 KB |
4340 | } |
4341 | p -= 2; /* Point at relative address again. */ | |
4342 | if ((re_opcode_t) p[-1] != pop_failure_jump) | |
4343 | { | |
4344 | p[-1] = (unsigned char) jump; | |
9114e279 | 4345 | DEBUG_PRINT1 (" Match => jump.\n"); |
bc78d348 KB |
4346 | goto unconditional_jump; |
4347 | } | |
4348 | /* Note fall through. */ | |
4349 | ||
4350 | ||
4351 | /* The end of a simple repeat has a pop_failure_jump back to | |
4352 | its matching on_failure_jump, where the latter will push a | |
4353 | failure point. The pop_failure_jump takes off failure | |
4354 | points put on by this pop_failure_jump's matching | |
4355 | on_failure_jump; we got through the pattern to here from the | |
4356 | matching on_failure_jump, so didn't fail. */ | |
4357 | case pop_failure_jump: | |
4358 | { | |
4359 | /* We need to pass separate storage for the lowest and | |
4360 | highest registers, even though we don't care about the | |
4361 | actual values. Otherwise, we will restore only one | |
4362 | register from the stack, since lowest will == highest in | |
4363 | `pop_failure_point'. */ | |
4364 | unsigned dummy_low_reg, dummy_high_reg; | |
4365 | unsigned char *pdummy; | |
4366 | const char *sdummy; | |
4367 | ||
4368 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
4369 | POP_FAILURE_POINT (sdummy, pdummy, | |
4370 | dummy_low_reg, dummy_high_reg, | |
4371 | reg_dummy, reg_dummy, reg_info_dummy); | |
4372 | } | |
4373 | /* Note fall through. */ | |
4374 | ||
4375 | ||
4376 | /* Unconditionally jump (without popping any failure points). */ | |
4377 | case jump: | |
4378 | unconditional_jump: | |
4379 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
4380 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
4381 | p += mcnt; /* Do the jump. */ | |
4382 | DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
4383 | break; | |
4384 | ||
4385 | ||
4386 | /* We need this opcode so we can detect where alternatives end | |
4387 | in `group_match_null_string_p' et al. */ | |
4388 | case jump_past_alt: | |
4389 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
4390 | goto unconditional_jump; | |
4391 | ||
4392 | ||
4393 | /* Normally, the on_failure_jump pushes a failure point, which | |
4394 | then gets popped at pop_failure_jump. We will end up at | |
4395 | pop_failure_jump, also, and with a pattern of, say, `a+', we | |
4396 | are skipping over the on_failure_jump, so we have to push | |
4397 | something meaningless for pop_failure_jump to pop. */ | |
4398 | case dummy_failure_jump: | |
4399 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
4400 | /* It doesn't matter what we push for the string here. What | |
4401 | the code at `fail' tests is the value for the pattern. */ | |
4402 | PUSH_FAILURE_POINT (0, 0, -2); | |
4403 | goto unconditional_jump; | |
4404 | ||
4405 | ||
4406 | /* At the end of an alternative, we need to push a dummy failure | |
9114e279 | 4407 | point in case we are followed by a `pop_failure_jump', because |
bc78d348 KB |
4408 | we don't want the failure point for the alternative to be |
4409 | popped. For example, matching `(a|ab)*' against `aab' | |
4410 | requires that we match the `ab' alternative. */ | |
4411 | case push_dummy_failure: | |
4412 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
4413 | /* See comments just above at `dummy_failure_jump' about the | |
4414 | two zeroes. */ | |
4415 | PUSH_FAILURE_POINT (0, 0, -2); | |
4416 | break; | |
4417 | ||
4418 | /* Have to succeed matching what follows at least n times. | |
4419 | After that, handle like `on_failure_jump'. */ | |
4420 | case succeed_n: | |
4421 | EXTRACT_NUMBER (mcnt, p + 2); | |
4422 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
4423 | ||
4424 | assert (mcnt >= 0); | |
4425 | /* Originally, this is how many times we HAVE to succeed. */ | |
4426 | if (mcnt > 0) | |
4427 | { | |
4428 | mcnt--; | |
4429 | p += 2; | |
4430 | STORE_NUMBER_AND_INCR (p, mcnt); | |
4431 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); | |
4432 | } | |
4433 | else if (mcnt == 0) | |
4434 | { | |
4435 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); | |
4436 | p[2] = (unsigned char) no_op; | |
4437 | p[3] = (unsigned char) no_op; | |
4438 | goto on_failure; | |
4439 | } | |
4440 | break; | |
4441 | ||
4442 | case jump_n: | |
4443 | EXTRACT_NUMBER (mcnt, p + 2); | |
4444 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
4445 | ||
4446 | /* Originally, this is how many times we CAN jump. */ | |
4447 | if (mcnt) | |
4448 | { | |
4449 | mcnt--; | |
4450 | STORE_NUMBER (p + 2, mcnt); | |
4451 | goto unconditional_jump; | |
4452 | } | |
4453 | /* If don't have to jump any more, skip over the rest of command. */ | |
4454 | else | |
4455 | p += 4; | |
4456 | break; | |
4457 | ||
4458 | case set_number_at: | |
4459 | { | |
4460 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
4461 | ||
4462 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4463 | p1 = p + mcnt; | |
4464 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4465 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
4466 | STORE_NUMBER (p1, mcnt); | |
4467 | break; | |
4468 | } | |
4469 | ||
4470 | case wordbound: | |
4471 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
4472 | if (AT_WORD_BOUNDARY (d)) | |
4473 | break; | |
4474 | goto fail; | |
4475 | ||
4476 | case notwordbound: | |
4477 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
4478 | if (AT_WORD_BOUNDARY (d)) | |
4479 | goto fail; | |
4480 | break; | |
4481 | ||
4482 | case wordbeg: | |
4483 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
9114e279 | 4484 | if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) |
bc78d348 KB |
4485 | break; |
4486 | goto fail; | |
4487 | ||
4488 | case wordend: | |
4489 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
9114e279 KB |
4490 | if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) |
4491 | && (!WORDCHAR_P (d) || AT_STRINGS_END (d))) | |
bc78d348 KB |
4492 | break; |
4493 | goto fail; | |
4494 | ||
4495 | #ifdef emacs | |
bc78d348 KB |
4496 | case before_dot: |
4497 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
4498 | if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
4499 | goto fail; | |
4500 | break; | |
4501 | ||
4502 | case at_dot: | |
4503 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4504 | if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
4505 | goto fail; | |
4506 | break; | |
4507 | ||
4508 | case after_dot: | |
4509 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
4510 | if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
4511 | goto fail; | |
4512 | break; | |
5b5bb5e4 | 4513 | #if 0 /* not emacs19 */ |
bc78d348 KB |
4514 | case at_dot: |
4515 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4516 | if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point) | |
4517 | goto fail; | |
4518 | break; | |
4519 | #endif /* not emacs19 */ | |
4520 | ||
4521 | case syntaxspec: | |
4522 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
4523 | mcnt = *p++; | |
4524 | goto matchsyntax; | |
4525 | ||
4526 | case wordchar: | |
9114e279 | 4527 | DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); |
bc78d348 KB |
4528 | mcnt = (int) Sword; |
4529 | matchsyntax: | |
4530 | PREFETCH (); | |
165c6714 KH |
4531 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ |
4532 | d++; | |
4533 | if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) | |
4534 | goto fail; | |
bc78d348 KB |
4535 | SET_REGS_MATCHED (); |
4536 | break; | |
4537 | ||
4538 | case notsyntaxspec: | |
4539 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
4540 | mcnt = *p++; | |
4541 | goto matchnotsyntax; | |
4542 | ||
4543 | case notwordchar: | |
9114e279 | 4544 | DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); |
bc78d348 | 4545 | mcnt = (int) Sword; |
9114e279 | 4546 | matchnotsyntax: |
bc78d348 | 4547 | PREFETCH (); |
165c6714 KH |
4548 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ |
4549 | d++; | |
4550 | if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) | |
4551 | goto fail; | |
bc78d348 KB |
4552 | SET_REGS_MATCHED (); |
4553 | break; | |
4554 | ||
4555 | #else /* not emacs */ | |
4556 | case wordchar: | |
4557 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
4558 | PREFETCH (); | |
9114e279 | 4559 | if (!WORDCHAR_P (d)) |
bc78d348 KB |
4560 | goto fail; |
4561 | SET_REGS_MATCHED (); | |
9114e279 | 4562 | d++; |
bc78d348 KB |
4563 | break; |
4564 | ||
4565 | case notwordchar: | |
4566 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
4567 | PREFETCH (); | |
9114e279 | 4568 | if (WORDCHAR_P (d)) |
bc78d348 KB |
4569 | goto fail; |
4570 | SET_REGS_MATCHED (); | |
9114e279 | 4571 | d++; |
bc78d348 KB |
4572 | break; |
4573 | #endif /* not emacs */ | |
4574 | ||
4575 | default: | |
4576 | abort (); | |
4577 | } | |
4578 | continue; /* Successfully executed one pattern command; keep going. */ | |
4579 | ||
4580 | ||
4581 | /* We goto here if a matching operation fails. */ | |
4582 | fail: | |
4583 | if (!FAIL_STACK_EMPTY ()) | |
4584 | { /* A restart point is known. Restore to that state. */ | |
4585 | DEBUG_PRINT1 ("\nFAIL:\n"); | |
4586 | POP_FAILURE_POINT (d, p, | |
4587 | lowest_active_reg, highest_active_reg, | |
4588 | regstart, regend, reg_info); | |
4589 | ||
4590 | /* If this failure point is a dummy, try the next one. */ | |
4591 | if (!p) | |
4592 | goto fail; | |
4593 | ||
4594 | /* If we failed to the end of the pattern, don't examine *p. */ | |
4595 | assert (p <= pend); | |
4596 | if (p < pend) | |
4597 | { | |
4598 | boolean is_a_jump_n = false; | |
4599 | ||
4600 | /* If failed to a backwards jump that's part of a repetition | |
4601 | loop, need to pop this failure point and use the next one. */ | |
4602 | switch ((re_opcode_t) *p) | |
4603 | { | |
4604 | case jump_n: | |
4605 | is_a_jump_n = true; | |
4606 | case maybe_pop_jump: | |
4607 | case pop_failure_jump: | |
4608 | case jump: | |
4609 | p1 = p + 1; | |
4610 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4611 | p1 += mcnt; | |
4612 | ||
4613 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
4614 | || (!is_a_jump_n | |
4615 | && (re_opcode_t) *p1 == on_failure_jump)) | |
4616 | goto fail; | |
4617 | break; | |
4618 | default: | |
4619 | /* do nothing */ ; | |
4620 | } | |
4621 | } | |
4622 | ||
4623 | if (d >= string1 && d <= end1) | |
4624 | dend = end_match_1; | |
4625 | } | |
4626 | else | |
4627 | break; /* Matching at this starting point really fails. */ | |
4628 | } /* for (;;) */ | |
4629 | ||
4630 | if (best_regs_set) | |
4631 | goto restore_best_regs; | |
4632 | ||
4633 | FREE_VARIABLES (); | |
4634 | ||
4635 | return -1; /* Failure to match. */ | |
4636 | } /* re_match_2 */ | |
4637 | \f | |
4638 | /* Subroutine definitions for re_match_2. */ | |
4639 | ||
4640 | ||
4641 | /* We are passed P pointing to a register number after a start_memory. | |
4642 | ||
4643 | Return true if the pattern up to the corresponding stop_memory can | |
4644 | match the empty string, and false otherwise. | |
4645 | ||
4646 | If we find the matching stop_memory, sets P to point to one past its number. | |
4647 | Otherwise, sets P to an undefined byte less than or equal to END. | |
4648 | ||
4649 | We don't handle duplicates properly (yet). */ | |
4650 | ||
4651 | static boolean | |
4652 | group_match_null_string_p (p, end, reg_info) | |
4653 | unsigned char **p, *end; | |
4654 | register_info_type *reg_info; | |
4655 | { | |
4656 | int mcnt; | |
4657 | /* Point to after the args to the start_memory. */ | |
4658 | unsigned char *p1 = *p + 2; | |
4659 | ||
4660 | while (p1 < end) | |
4661 | { | |
4662 | /* Skip over opcodes that can match nothing, and return true or | |
4663 | false, as appropriate, when we get to one that can't, or to the | |
4664 | matching stop_memory. */ | |
4665 | ||
4666 | switch ((re_opcode_t) *p1) | |
4667 | { | |
4668 | /* Could be either a loop or a series of alternatives. */ | |
4669 | case on_failure_jump: | |
4670 | p1++; | |
4671 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4672 | ||
4673 | /* If the next operation is not a jump backwards in the | |
4674 | pattern. */ | |
4675 | ||
4676 | if (mcnt >= 0) | |
4677 | { | |
4678 | /* Go through the on_failure_jumps of the alternatives, | |
4679 | seeing if any of the alternatives cannot match nothing. | |
4680 | The last alternative starts with only a jump, | |
4681 | whereas the rest start with on_failure_jump and end | |
4682 | with a jump, e.g., here is the pattern for `a|b|c': | |
4683 | ||
4684 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
4685 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
4686 | /exactn/1/c | |
4687 | ||
4688 | So, we have to first go through the first (n-1) | |
4689 | alternatives and then deal with the last one separately. */ | |
4690 | ||
4691 | ||
4692 | /* Deal with the first (n-1) alternatives, which start | |
4693 | with an on_failure_jump (see above) that jumps to right | |
4694 | past a jump_past_alt. */ | |
4695 | ||
4696 | while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) | |
4697 | { | |
4698 | /* `mcnt' holds how many bytes long the alternative | |
4699 | is, including the ending `jump_past_alt' and | |
4700 | its number. */ | |
4701 | ||
4702 | if (!alt_match_null_string_p (p1, p1 + mcnt - 3, | |
4703 | reg_info)) | |
4704 | return false; | |
4705 | ||
4706 | /* Move to right after this alternative, including the | |
4707 | jump_past_alt. */ | |
4708 | p1 += mcnt; | |
4709 | ||
4710 | /* Break if it's the beginning of an n-th alternative | |
4711 | that doesn't begin with an on_failure_jump. */ | |
4712 | if ((re_opcode_t) *p1 != on_failure_jump) | |
4713 | break; | |
4714 | ||
4715 | /* Still have to check that it's not an n-th | |
4716 | alternative that starts with an on_failure_jump. */ | |
4717 | p1++; | |
4718 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4719 | if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
4720 | { | |
4721 | /* Get to the beginning of the n-th alternative. */ | |
4722 | p1 -= 3; | |
4723 | break; | |
4724 | } | |
4725 | } | |
4726 | ||
4727 | /* Deal with the last alternative: go back and get number | |
4728 | of the `jump_past_alt' just before it. `mcnt' contains | |
4729 | the length of the alternative. */ | |
4730 | EXTRACT_NUMBER (mcnt, p1 - 2); | |
4731 | ||
4732 | if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
4733 | return false; | |
4734 | ||
4735 | p1 += mcnt; /* Get past the n-th alternative. */ | |
4736 | } /* if mcnt > 0 */ | |
4737 | break; | |
4738 | ||
4739 | ||
4740 | case stop_memory: | |
4741 | assert (p1[1] == **p); | |
4742 | *p = p1 + 2; | |
4743 | return true; | |
4744 | ||
4745 | ||
4746 | default: | |
4747 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4748 | return false; | |
4749 | } | |
4750 | } /* while p1 < end */ | |
4751 | ||
4752 | return false; | |
4753 | } /* group_match_null_string_p */ | |
4754 | ||
4755 | ||
4756 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
4757 | It expects P to be the first byte of a single alternative and END one | |
4758 | byte past the last. The alternative can contain groups. */ | |
4759 | ||
4760 | static boolean | |
4761 | alt_match_null_string_p (p, end, reg_info) | |
4762 | unsigned char *p, *end; | |
4763 | register_info_type *reg_info; | |
4764 | { | |
4765 | int mcnt; | |
4766 | unsigned char *p1 = p; | |
4767 | ||
4768 | while (p1 < end) | |
4769 | { | |
4770 | /* Skip over opcodes that can match nothing, and break when we get | |
4771 | to one that can't. */ | |
4772 | ||
4773 | switch ((re_opcode_t) *p1) | |
4774 | { | |
4775 | /* It's a loop. */ | |
4776 | case on_failure_jump: | |
4777 | p1++; | |
4778 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4779 | p1 += mcnt; | |
4780 | break; | |
4781 | ||
4782 | default: | |
4783 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4784 | return false; | |
4785 | } | |
4786 | } /* while p1 < end */ | |
4787 | ||
4788 | return true; | |
4789 | } /* alt_match_null_string_p */ | |
4790 | ||
4791 | ||
4792 | /* Deals with the ops common to group_match_null_string_p and | |
4793 | alt_match_null_string_p. | |
4794 | ||
4795 | Sets P to one after the op and its arguments, if any. */ | |
4796 | ||
4797 | static boolean | |
4798 | common_op_match_null_string_p (p, end, reg_info) | |
4799 | unsigned char **p, *end; | |
4800 | register_info_type *reg_info; | |
4801 | { | |
4802 | int mcnt; | |
4803 | boolean ret; | |
4804 | int reg_no; | |
4805 | unsigned char *p1 = *p; | |
4806 | ||
4807 | switch ((re_opcode_t) *p1++) | |
4808 | { | |
4809 | case no_op: | |
4810 | case begline: | |
4811 | case endline: | |
4812 | case begbuf: | |
4813 | case endbuf: | |
4814 | case wordbeg: | |
4815 | case wordend: | |
4816 | case wordbound: | |
4817 | case notwordbound: | |
4818 | #ifdef emacs | |
4819 | case before_dot: | |
4820 | case at_dot: | |
4821 | case after_dot: | |
4822 | #endif | |
4823 | break; | |
4824 | ||
4825 | case start_memory: | |
4826 | reg_no = *p1; | |
4827 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
4828 | ret = group_match_null_string_p (&p1, end, reg_info); | |
4829 | ||
4830 | /* Have to set this here in case we're checking a group which | |
4831 | contains a group and a back reference to it. */ | |
4832 | ||
4833 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
4834 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
4835 | ||
4836 | if (!ret) | |
4837 | return false; | |
4838 | break; | |
4839 | ||
4840 | /* If this is an optimized succeed_n for zero times, make the jump. */ | |
4841 | case jump: | |
4842 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4843 | if (mcnt >= 0) | |
4844 | p1 += mcnt; | |
4845 | else | |
4846 | return false; | |
4847 | break; | |
4848 | ||
4849 | case succeed_n: | |
4850 | /* Get to the number of times to succeed. */ | |
4851 | p1 += 2; | |
4852 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4853 | ||
4854 | if (mcnt == 0) | |
4855 | { | |
4856 | p1 -= 4; | |
4857 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4858 | p1 += mcnt; | |
4859 | } | |
4860 | else | |
4861 | return false; | |
4862 | break; | |
4863 | ||
4864 | case duplicate: | |
4865 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
4866 | return false; | |
4867 | break; | |
4868 | ||
4869 | case set_number_at: | |
4870 | p1 += 4; | |
4871 | ||
4872 | default: | |
4873 | /* All other opcodes mean we cannot match the empty string. */ | |
4874 | return false; | |
4875 | } | |
4876 | ||
4877 | *p = p1; | |
4878 | return true; | |
4879 | } /* common_op_match_null_string_p */ | |
4880 | ||
4881 | ||
4882 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
4883 | bytes; nonzero otherwise. */ | |
4884 | ||
4885 | static int | |
4886 | bcmp_translate (s1, s2, len, translate) | |
4887 | unsigned char *s1, *s2; | |
4888 | register int len; | |
4889 | char *translate; | |
4890 | { | |
4891 | register unsigned char *p1 = s1, *p2 = s2; | |
4892 | while (len) | |
4893 | { | |
4894 | if (translate[*p1++] != translate[*p2++]) return 1; | |
4895 | len--; | |
4896 | } | |
4897 | return 0; | |
4898 | } | |
4899 | \f | |
4900 | /* Entry points for GNU code. */ | |
4901 | ||
4902 | /* re_compile_pattern is the GNU regular expression compiler: it | |
4903 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
4904 | Returns 0 if the pattern was valid, otherwise an error string. | |
4905 | ||
4906 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
4907 | are set in BUFP on entry. | |
4908 | ||
4909 | We call regex_compile to do the actual compilation. */ | |
4910 | ||
4911 | const char * | |
4912 | re_compile_pattern (pattern, length, bufp) | |
4913 | const char *pattern; | |
4914 | int length; | |
4915 | struct re_pattern_buffer *bufp; | |
4916 | { | |
4917 | reg_errcode_t ret; | |
4918 | ||
4919 | /* GNU code is written to assume at least RE_NREGS registers will be set | |
4920 | (and at least one extra will be -1). */ | |
4921 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4922 | ||
4923 | /* And GNU code determines whether or not to get register information | |
4924 | by passing null for the REGS argument to re_match, etc., not by | |
4925 | setting no_sub. */ | |
4926 | bufp->no_sub = 0; | |
4927 | ||
4928 | /* Match anchors at newline. */ | |
4929 | bufp->newline_anchor = 1; | |
4930 | ||
4931 | ret = regex_compile (pattern, length, re_syntax_options, bufp); | |
4932 | ||
4933 | return re_error_msg[(int) ret]; | |
4934 | } | |
4935 | \f | |
4936 | /* Entry points compatible with 4.2 BSD regex library. We don't define | |
4937 | them if this is an Emacs or POSIX compilation. */ | |
4938 | ||
4939 | #if !defined (emacs) && !defined (_POSIX_SOURCE) | |
4940 | ||
4941 | /* BSD has one and only one pattern buffer. */ | |
4942 | static struct re_pattern_buffer re_comp_buf; | |
4943 | ||
4944 | char * | |
4945 | re_comp (s) | |
4946 | const char *s; | |
4947 | { | |
4948 | reg_errcode_t ret; | |
4949 | ||
4950 | if (!s) | |
4951 | { | |
4952 | if (!re_comp_buf.buffer) | |
4953 | return "No previous regular expression"; | |
4954 | return 0; | |
4955 | } | |
4956 | ||
4957 | if (!re_comp_buf.buffer) | |
4958 | { | |
4959 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
4960 | if (re_comp_buf.buffer == NULL) | |
4961 | return "Memory exhausted"; | |
4962 | re_comp_buf.allocated = 200; | |
4963 | ||
4964 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
4965 | if (re_comp_buf.fastmap == NULL) | |
4966 | return "Memory exhausted"; | |
4967 | } | |
4968 | ||
4969 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
4970 | don't need to initialize the pattern buffer fields which affect it. */ | |
4971 | ||
4972 | /* Match anchors at newlines. */ | |
4973 | re_comp_buf.newline_anchor = 1; | |
4974 | ||
4975 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
4976 | ||
4977 | /* Yes, we're discarding `const' here. */ | |
4978 | return (char *) re_error_msg[(int) ret]; | |
4979 | } | |
4980 | ||
4981 | ||
4982 | int | |
4983 | re_exec (s) | |
4984 | const char *s; | |
4985 | { | |
4986 | const int len = strlen (s); | |
4987 | return | |
4988 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
4989 | } | |
4990 | #endif /* not emacs and not _POSIX_SOURCE */ | |
4991 | \f | |
4992 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
4993 | ||
4994 | #ifndef emacs | |
4995 | ||
4996 | /* regcomp takes a regular expression as a string and compiles it. | |
4997 | ||
4998 | PREG is a regex_t *. We do not expect any fields to be initialized, | |
4999 | since POSIX says we shouldn't. Thus, we set | |
5000 | ||
5001 | `buffer' to the compiled pattern; | |
5002 | `used' to the length of the compiled pattern; | |
5003 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
5004 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
5005 | RE_SYNTAX_POSIX_BASIC; | |
5006 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
5007 | `fastmap' and `fastmap_accurate' to zero; | |
5008 | `re_nsub' to the number of subexpressions in PATTERN. | |
5009 | ||
5010 | PATTERN is the address of the pattern string. | |
5011 | ||
5012 | CFLAGS is a series of bits which affect compilation. | |
5013 | ||
5014 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
5015 | use POSIX basic syntax. | |
5016 | ||
5017 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
5018 | Also, regexec will try a match beginning after every newline. | |
5019 | ||
5020 | If REG_ICASE is set, then we considers upper- and lowercase | |
5021 | versions of letters to be equivalent when matching. | |
5022 | ||
5023 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
5024 | routine will report only success or failure, and nothing about the | |
5025 | registers. | |
5026 | ||
5027 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
5028 | the return codes and their meanings.) */ | |
5029 | ||
5030 | int | |
5031 | regcomp (preg, pattern, cflags) | |
5032 | regex_t *preg; | |
5033 | const char *pattern; | |
5034 | int cflags; | |
5035 | { | |
5036 | reg_errcode_t ret; | |
5037 | unsigned syntax | |
d9088577 DM |
5038 | = (cflags & REG_EXTENDED) ? |
5039 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
bc78d348 KB |
5040 | |
5041 | /* regex_compile will allocate the space for the compiled pattern. */ | |
5042 | preg->buffer = 0; | |
d9088577 | 5043 | preg->allocated = 0; |
8b4cc189 | 5044 | preg->used = 0; |
bc78d348 KB |
5045 | |
5046 | /* Don't bother to use a fastmap when searching. This simplifies the | |
5047 | REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
5048 | characters after newlines into the fastmap. This way, we just try | |
5049 | every character. */ | |
5050 | preg->fastmap = 0; | |
5051 | ||
5052 | if (cflags & REG_ICASE) | |
5053 | { | |
5054 | unsigned i; | |
5055 | ||
5056 | preg->translate = (char *) malloc (CHAR_SET_SIZE); | |
5057 | if (preg->translate == NULL) | |
5058 | return (int) REG_ESPACE; | |
5059 | ||
5060 | /* Map uppercase characters to corresponding lowercase ones. */ | |
5061 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
c6b40788 | 5062 | preg->translate[i] = ISUPPER (i) ? tolower (i) : i; |
bc78d348 KB |
5063 | } |
5064 | else | |
5065 | preg->translate = NULL; | |
5066 | ||
5067 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
5068 | if (cflags & REG_NEWLINE) | |
5069 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
5070 | syntax &= ~RE_DOT_NEWLINE; | |
5071 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
5072 | /* It also changes the matching behavior. */ | |
5073 | preg->newline_anchor = 1; | |
5074 | } | |
5075 | else | |
5076 | preg->newline_anchor = 0; | |
5077 | ||
5078 | preg->no_sub = !!(cflags & REG_NOSUB); | |
5079 | ||
5080 | /* POSIX says a null character in the pattern terminates it, so we | |
5081 | can use strlen here in compiling the pattern. */ | |
5082 | ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
5083 | ||
5084 | /* POSIX doesn't distinguish between an unmatched open-group and an | |
5085 | unmatched close-group: both are REG_EPAREN. */ | |
5086 | if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
5087 | ||
5088 | return (int) ret; | |
5089 | } | |
5090 | ||
5091 | ||
5092 | /* regexec searches for a given pattern, specified by PREG, in the | |
5093 | string STRING. | |
5094 | ||
5095 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
5096 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
5097 | least NMATCH elements, and we set them to the offsets of the | |
5098 | corresponding matched substrings. | |
5099 | ||
5100 | EFLAGS specifies `execution flags' which affect matching: if | |
5101 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
5102 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5103 | ||
5104 | We return 0 if we find a match and REG_NOMATCH if not. */ | |
5105 | ||
5106 | int | |
5107 | regexec (preg, string, nmatch, pmatch, eflags) | |
5108 | const regex_t *preg; | |
5109 | const char *string; | |
5110 | size_t nmatch; | |
5111 | regmatch_t pmatch[]; | |
5112 | int eflags; | |
5113 | { | |
5114 | int ret; | |
5115 | struct re_registers regs; | |
5116 | regex_t private_preg; | |
5117 | int len = strlen (string); | |
5118 | boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
5119 | ||
5120 | private_preg = *preg; | |
5121 | ||
5122 | private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
5123 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5124 | ||
5125 | /* The user has told us exactly how many registers to return | |
5126 | information about, via `nmatch'. We have to pass that on to the | |
5127 | matching routines. */ | |
5128 | private_preg.regs_allocated = REGS_FIXED; | |
5129 | ||
5130 | if (want_reg_info) | |
5131 | { | |
5132 | regs.num_regs = nmatch; | |
5133 | regs.start = TALLOC (nmatch, regoff_t); | |
5134 | regs.end = TALLOC (nmatch, regoff_t); | |
5135 | if (regs.start == NULL || regs.end == NULL) | |
5136 | return (int) REG_NOMATCH; | |
5137 | } | |
5138 | ||
5139 | /* Perform the searching operation. */ | |
5140 | ret = re_search (&private_preg, string, len, | |
5141 | /* start: */ 0, /* range: */ len, | |
5142 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5143 | ||
5144 | /* Copy the register information to the POSIX structure. */ | |
5145 | if (want_reg_info) | |
5146 | { | |
5147 | if (ret >= 0) | |
5148 | { | |
5149 | unsigned r; | |
5150 | ||
5151 | for (r = 0; r < nmatch; r++) | |
5152 | { | |
5153 | pmatch[r].rm_so = regs.start[r]; | |
5154 | pmatch[r].rm_eo = regs.end[r]; | |
5155 | } | |
5156 | } | |
5157 | ||
5158 | /* If we needed the temporary register info, free the space now. */ | |
5159 | free (regs.start); | |
5160 | free (regs.end); | |
5161 | } | |
5162 | ||
5163 | /* We want zero return to mean success, unlike `re_search'. */ | |
5164 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
5165 | } | |
5166 | ||
5167 | ||
5168 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
9114e279 | 5169 | from either regcomp or regexec. We don't use PREG here. */ |
bc78d348 KB |
5170 | |
5171 | size_t | |
5172 | regerror (errcode, preg, errbuf, errbuf_size) | |
5173 | int errcode; | |
5174 | const regex_t *preg; | |
5175 | char *errbuf; | |
5176 | size_t errbuf_size; | |
5177 | { | |
85586484 JB |
5178 | const char *msg; |
5179 | size_t msg_size; | |
5180 | ||
5181 | if (errcode < 0 | |
5182 | || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0]))) | |
5183 | /* Only error codes returned by the rest of the code should be passed | |
5184 | to this routine. If we are given anything else, or if other regex | |
5185 | code generates an invalid error code, then the program has a bug. | |
5186 | Dump core so we can fix it. */ | |
5187 | abort (); | |
5188 | ||
5745b086 JB |
5189 | msg = re_error_msg[errcode]; |
5190 | ||
5191 | /* POSIX doesn't require that we do anything in this case, but why | |
5192 | not be nice. */ | |
5193 | if (! msg) | |
5194 | msg = "Success"; | |
5195 | ||
85586484 | 5196 | msg_size = strlen (msg) + 1; /* Includes the null. */ |
bc78d348 KB |
5197 | |
5198 | if (errbuf_size != 0) | |
5199 | { | |
5200 | if (msg_size > errbuf_size) | |
5201 | { | |
5202 | strncpy (errbuf, msg, errbuf_size - 1); | |
5203 | errbuf[errbuf_size - 1] = 0; | |
5204 | } | |
5205 | else | |
5206 | strcpy (errbuf, msg); | |
5207 | } | |
5208 | ||
5209 | return msg_size; | |
5210 | } | |
5211 | ||
5212 | ||
5213 | /* Free dynamically allocated space used by PREG. */ | |
5214 | ||
5215 | void | |
5216 | regfree (preg) | |
5217 | regex_t *preg; | |
5218 | { | |
5219 | if (preg->buffer != NULL) | |
5220 | free (preg->buffer); | |
5221 | preg->buffer = NULL; | |
5222 | ||
5223 | preg->allocated = 0; | |
5224 | preg->used = 0; | |
5225 | ||
5226 | if (preg->fastmap != NULL) | |
5227 | free (preg->fastmap); | |
5228 | preg->fastmap = NULL; | |
5229 | preg->fastmap_accurate = 0; | |
5230 | ||
5231 | if (preg->translate != NULL) | |
5232 | free (preg->translate); | |
5233 | preg->translate = NULL; | |
5234 | } | |
5235 | ||
5236 | #endif /* not emacs */ | |
5237 | \f | |
5238 | /* | |
5239 | Local variables: | |
5240 | make-backup-files: t | |
5241 | version-control: t | |
5242 | trim-versions-without-asking: nil | |
5243 | End: | |
5244 | */ |