Commit | Line | Data |
---|---|---|
e318085a RS |
1 | /* Extended regular expression matching and search library, version |
2 | 0.12. (Implements POSIX draft P10003.2/D11.2, except for | |
bc78d348 KB |
3 | internationalization features.) |
4 | ||
b18215fc | 5 | Copyright (C) 1993, 1994, 1995, 1996, 1997 Free Software Foundation, Inc. |
bc78d348 | 6 | |
fa9a63c5 RM |
7 | This program is free software; you can redistribute it and/or modify |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 2, or (at your option) | |
10 | any later version. | |
11 | ||
12 | This program is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
25fe55af | 14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
fa9a63c5 RM |
15 | GNU General Public License for more details. |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with this program; if not, write to the Free Software | |
ba4a8e51 | 19 | Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, |
25fe55af | 20 | USA. */ |
fa9a63c5 RM |
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 | ||
68d96f02 | 27 | #undef _GNU_SOURCE |
fa9a63c5 RM |
28 | #define _GNU_SOURCE |
29 | ||
25fe55af RS |
30 | /* Converts the pointer to the char to BEG-based offset from the start. */ |
31 | #define PTR_TO_OFFSET(d) \ | |
b18215fc RS |
32 | POS_AS_IN_BUFFER (MATCHING_IN_FIRST_STRING \ |
33 | ? (d) - string1 : (d) - (string2 - size1)) | |
34 | #define POS_AS_IN_BUFFER(p) ((p) + 1) | |
35 | ||
fa9a63c5 RM |
36 | #ifdef HAVE_CONFIG_H |
37 | #include <config.h> | |
38 | #endif | |
39 | ||
25fe55af | 40 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ |
fa9a63c5 RM |
41 | #include <sys/types.h> |
42 | ||
25fe55af | 43 | /* This is for other GNU distributions with internationalized messages. */ |
fa9a63c5 RM |
44 | #if HAVE_LIBINTL_H || defined (_LIBC) |
45 | # include <libintl.h> | |
46 | #else | |
47 | # define gettext(msgid) (msgid) | |
48 | #endif | |
49 | ||
5e69f11e RM |
50 | #ifndef gettext_noop |
51 | /* This define is so xgettext can find the internationalizable | |
52 | strings. */ | |
53 | #define gettext_noop(String) String | |
54 | #endif | |
55 | ||
fa9a63c5 RM |
56 | /* The `emacs' switch turns on certain matching commands |
57 | that make sense only in Emacs. */ | |
58 | #ifdef emacs | |
59 | ||
60 | #include "lisp.h" | |
61 | #include "buffer.h" | |
b18215fc RS |
62 | |
63 | /* Make syntax table lookup grant data in gl_state. */ | |
64 | #define SYNTAX_ENTRY_VIA_PROPERTY | |
65 | ||
fa9a63c5 | 66 | #include "syntax.h" |
b18215fc RS |
67 | #include "charset.h" |
68 | #include "category.h" | |
fa9a63c5 | 69 | |
9abbd165 | 70 | #define malloc xmalloc |
64e3c718 | 71 | #define realloc xrealloc |
9abbd165 RS |
72 | #define free xfree |
73 | ||
fa9a63c5 RM |
74 | #else /* not emacs */ |
75 | ||
76 | /* If we are not linking with Emacs proper, | |
77 | we can't use the relocating allocator | |
78 | even if config.h says that we can. */ | |
79 | #undef REL_ALLOC | |
80 | ||
81 | #if defined (STDC_HEADERS) || defined (_LIBC) | |
82 | #include <stdlib.h> | |
83 | #else | |
84 | char *malloc (); | |
85 | char *realloc (); | |
86 | #endif | |
87 | ||
9e4ecb26 | 88 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
25fe55af | 89 | If nothing else has been done, use the method below. */ |
9e4ecb26 KH |
90 | #ifdef INHIBIT_STRING_HEADER |
91 | #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY)) | |
92 | #if !defined (bzero) && !defined (bcopy) | |
93 | #undef INHIBIT_STRING_HEADER | |
94 | #endif | |
95 | #endif | |
96 | #endif | |
97 | ||
98 | /* This is the normal way of making sure we have a bcopy and a bzero. | |
99 | This is used in most programs--a few other programs avoid this | |
100 | by defining INHIBIT_STRING_HEADER. */ | |
fa9a63c5 | 101 | #ifndef INHIBIT_STRING_HEADER |
7f998252 | 102 | #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC) |
fa9a63c5 RM |
103 | #include <string.h> |
104 | #ifndef bcmp | |
105 | #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n)) | |
106 | #endif | |
107 | #ifndef bcopy | |
108 | #define bcopy(s, d, n) memcpy ((d), (s), (n)) | |
109 | #endif | |
110 | #ifndef bzero | |
111 | #define bzero(s, n) memset ((s), 0, (n)) | |
112 | #endif | |
113 | #else | |
114 | #include <strings.h> | |
115 | #endif | |
116 | #endif | |
117 | ||
118 | /* Define the syntax stuff for \<, \>, etc. */ | |
119 | ||
120 | /* This must be nonzero for the wordchar and notwordchar pattern | |
121 | commands in re_match_2. */ | |
5e69f11e | 122 | #ifndef Sword |
fa9a63c5 RM |
123 | #define Sword 1 |
124 | #endif | |
125 | ||
126 | #ifdef SWITCH_ENUM_BUG | |
127 | #define SWITCH_ENUM_CAST(x) ((int)(x)) | |
128 | #else | |
129 | #define SWITCH_ENUM_CAST(x) (x) | |
130 | #endif | |
131 | ||
132 | #ifdef SYNTAX_TABLE | |
133 | ||
134 | extern char *re_syntax_table; | |
135 | ||
136 | #else /* not SYNTAX_TABLE */ | |
137 | ||
138 | /* How many characters in the character set. */ | |
139 | #define CHAR_SET_SIZE 256 | |
140 | ||
141 | static char re_syntax_table[CHAR_SET_SIZE]; | |
142 | ||
143 | static void | |
144 | init_syntax_once () | |
145 | { | |
146 | register int c; | |
147 | static int done = 0; | |
148 | ||
149 | if (done) | |
150 | return; | |
151 | ||
152 | bzero (re_syntax_table, sizeof re_syntax_table); | |
153 | ||
154 | for (c = 'a'; c <= 'z'; c++) | |
155 | re_syntax_table[c] = Sword; | |
156 | ||
157 | for (c = 'A'; c <= 'Z'; c++) | |
158 | re_syntax_table[c] = Sword; | |
159 | ||
160 | for (c = '0'; c <= '9'; c++) | |
161 | re_syntax_table[c] = Sword; | |
162 | ||
163 | re_syntax_table['_'] = Sword; | |
164 | ||
165 | done = 1; | |
166 | } | |
167 | ||
168 | #endif /* not SYNTAX_TABLE */ | |
169 | ||
170 | #define SYNTAX(c) re_syntax_table[c] | |
171 | ||
e934739e | 172 | /* Dummy macros for non-Emacs environments. */ |
b18215fc RS |
173 | #define BASE_LEADING_CODE_P(c) (0) |
174 | #define WORD_BOUNDARY_P(c1, c2) (0) | |
175 | #define CHAR_HEAD_P(p) (1) | |
176 | #define SINGLE_BYTE_CHAR_P(c) (1) | |
177 | #define SAME_CHARSET_P(c1, c2) (1) | |
178 | #define MULTIBYTE_FORM_LENGTH(p, s) (1) | |
179 | #define STRING_CHAR(p, s) (*(p)) | |
180 | #define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p)) | |
181 | #define GET_CHAR_AFTER_2(c, p, str1, end1, str2, end2) \ | |
182 | (c = ((p) == (end1) ? *(str2) : *(p))) | |
183 | #define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ | |
184 | (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1))) | |
fa9a63c5 RM |
185 | #endif /* not emacs */ |
186 | \f | |
187 | /* Get the interface, including the syntax bits. */ | |
188 | #include "regex.h" | |
189 | ||
190 | /* isalpha etc. are used for the character classes. */ | |
191 | #include <ctype.h> | |
192 | ||
193 | /* Jim Meyering writes: | |
194 | ||
195 | "... Some ctype macros are valid only for character codes that | |
196 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
197 | using /bin/cc or gcc but without giving an ansi option). So, all | |
25fe55af | 198 | ctype uses should be through macros like ISPRINT... If |
fa9a63c5 RM |
199 | STDC_HEADERS is defined, then autoconf has verified that the ctype |
200 | macros don't need to be guarded with references to isascii. ... | |
201 | Defining isascii to 1 should let any compiler worth its salt | |
25fe55af | 202 | eliminate the && through constant folding." */ |
fa9a63c5 RM |
203 | |
204 | #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII)) | |
205 | #define ISASCII(c) 1 | |
206 | #else | |
207 | #define ISASCII(c) isascii(c) | |
208 | #endif | |
209 | ||
210 | #ifdef isblank | |
211 | #define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
212 | #else | |
213 | #define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
214 | #endif | |
215 | #ifdef isgraph | |
216 | #define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
217 | #else | |
218 | #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
219 | #endif | |
220 | ||
221 | #define ISPRINT(c) (ISASCII (c) && isprint (c)) | |
222 | #define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
223 | #define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
224 | #define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
225 | #define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
226 | #define ISLOWER(c) (ISASCII (c) && islower (c)) | |
227 | #define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
228 | #define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
229 | #define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
230 | #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
231 | ||
232 | #ifndef NULL | |
075f06ec | 233 | #define NULL (void *)0 |
fa9a63c5 RM |
234 | #endif |
235 | ||
236 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
237 | since ours (we hope) works properly with all combinations of | |
238 | machines, compilers, `char' and `unsigned char' argument types. | |
25fe55af | 239 | (Per Bothner suggested the basic approach.) */ |
fa9a63c5 RM |
240 | #undef SIGN_EXTEND_CHAR |
241 | #if __STDC__ | |
242 | #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
243 | #else /* not __STDC__ */ | |
244 | /* As in Harbison and Steele. */ | |
245 | #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
246 | #endif | |
247 | \f | |
248 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
249 | use `alloca' instead of `malloc'. This is because using malloc in | |
250 | re_search* or re_match* could cause memory leaks when C-g is used in | |
251 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
5e69f11e RM |
252 | the other hand, malloc is more portable, and easier to debug. |
253 | ||
fa9a63c5 RM |
254 | Because we sometimes use alloca, some routines have to be macros, |
255 | not functions -- `alloca'-allocated space disappears at the end of the | |
256 | function it is called in. */ | |
257 | ||
258 | #ifdef REGEX_MALLOC | |
259 | ||
260 | #define REGEX_ALLOCATE malloc | |
261 | #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
262 | #define REGEX_FREE free | |
263 | ||
264 | #else /* not REGEX_MALLOC */ | |
265 | ||
266 | /* Emacs already defines alloca, sometimes. */ | |
267 | #ifndef alloca | |
268 | ||
269 | /* Make alloca work the best possible way. */ | |
270 | #ifdef __GNUC__ | |
271 | #define alloca __builtin_alloca | |
272 | #else /* not __GNUC__ */ | |
273 | #if HAVE_ALLOCA_H | |
274 | #include <alloca.h> | |
275 | #else /* not __GNUC__ or HAVE_ALLOCA_H */ | |
f3c4387f | 276 | #if 0 /* It is a bad idea to declare alloca. We always cast the result. */ |
25fe55af | 277 | #ifndef _AIX /* Already did AIX, up at the top. */ |
fa9a63c5 RM |
278 | char *alloca (); |
279 | #endif /* not _AIX */ | |
f3c4387f | 280 | #endif |
5e69f11e | 281 | #endif /* not HAVE_ALLOCA_H */ |
fa9a63c5 RM |
282 | #endif /* not __GNUC__ */ |
283 | ||
284 | #endif /* not alloca */ | |
285 | ||
286 | #define REGEX_ALLOCATE alloca | |
287 | ||
288 | /* Assumes a `char *destination' variable. */ | |
289 | #define REGEX_REALLOCATE(source, osize, nsize) \ | |
290 | (destination = (char *) alloca (nsize), \ | |
291 | bcopy (source, destination, osize), \ | |
292 | destination) | |
293 | ||
294 | /* No need to do anything to free, after alloca. */ | |
c2e1680a | 295 | #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
296 | |
297 | #endif /* not REGEX_MALLOC */ | |
298 | ||
299 | /* Define how to allocate the failure stack. */ | |
300 | ||
33487cc8 | 301 | #if defined (REL_ALLOC) && defined (REGEX_MALLOC) |
4297555e | 302 | |
fa9a63c5 RM |
303 | #define REGEX_ALLOCATE_STACK(size) \ |
304 | r_alloc (&failure_stack_ptr, (size)) | |
305 | #define REGEX_REALLOCATE_STACK(source, osize, nsize) \ | |
306 | r_re_alloc (&failure_stack_ptr, (nsize)) | |
307 | #define REGEX_FREE_STACK(ptr) \ | |
308 | r_alloc_free (&failure_stack_ptr) | |
309 | ||
4297555e | 310 | #else /* not using relocating allocator */ |
fa9a63c5 RM |
311 | |
312 | #ifdef REGEX_MALLOC | |
313 | ||
314 | #define REGEX_ALLOCATE_STACK malloc | |
315 | #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
316 | #define REGEX_FREE_STACK free | |
317 | ||
318 | #else /* not REGEX_MALLOC */ | |
319 | ||
320 | #define REGEX_ALLOCATE_STACK alloca | |
321 | ||
322 | #define REGEX_REALLOCATE_STACK(source, osize, nsize) \ | |
323 | REGEX_REALLOCATE (source, osize, nsize) | |
25fe55af | 324 | /* No need to explicitly free anything. */ |
fa9a63c5 RM |
325 | #define REGEX_FREE_STACK(arg) |
326 | ||
327 | #endif /* not REGEX_MALLOC */ | |
4297555e | 328 | #endif /* not using relocating allocator */ |
fa9a63c5 RM |
329 | |
330 | ||
331 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
332 | `string1' or just past its end. This works if PTR is NULL, which is | |
333 | a good thing. */ | |
25fe55af | 334 | #define FIRST_STRING_P(ptr) \ |
fa9a63c5 RM |
335 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
336 | ||
337 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
338 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
339 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
340 | #define RETALLOC_IF(addr, n, t) \ | |
341 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
342 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
343 | ||
25fe55af | 344 | #define BYTEWIDTH 8 /* In bits. */ |
fa9a63c5 RM |
345 | |
346 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
347 | ||
348 | #undef MAX | |
349 | #undef MIN | |
350 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
351 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
352 | ||
353 | typedef char boolean; | |
354 | #define false 0 | |
355 | #define true 1 | |
356 | ||
357 | static int re_match_2_internal (); | |
358 | \f | |
359 | /* These are the command codes that appear in compiled regular | |
25fe55af | 360 | expressions. Some opcodes are followed by argument bytes. A |
fa9a63c5 RM |
361 | command code can specify any interpretation whatsoever for its |
362 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
363 | ||
364 | typedef enum | |
365 | { | |
366 | no_op = 0, | |
367 | ||
25fe55af | 368 | /* Succeed right away--no more backtracking. */ |
fa9a63c5 RM |
369 | succeed, |
370 | ||
25fe55af | 371 | /* Followed by one byte giving n, then by n literal bytes. */ |
fa9a63c5 RM |
372 | exactn, |
373 | ||
25fe55af | 374 | /* Matches any (more or less) character. */ |
fa9a63c5 RM |
375 | anychar, |
376 | ||
25fe55af RS |
377 | /* Matches any one char belonging to specified set. First |
378 | following byte is number of bitmap bytes. Then come bytes | |
379 | for a bitmap saying which chars are in. Bits in each byte | |
380 | are ordered low-bit-first. A character is in the set if its | |
381 | bit is 1. A character too large to have a bit in the map is | |
382 | automatically not in the set. */ | |
fa9a63c5 RM |
383 | charset, |
384 | ||
25fe55af RS |
385 | /* Same parameters as charset, but match any character that is |
386 | not one of those specified. */ | |
fa9a63c5 RM |
387 | charset_not, |
388 | ||
25fe55af RS |
389 | /* Start remembering the text that is matched, for storing in a |
390 | register. Followed by one byte with the register number, in | |
391 | the range 0 to one less than the pattern buffer's re_nsub | |
392 | field. Then followed by one byte with the number of groups | |
393 | inner to this one. (This last has to be part of the | |
394 | start_memory only because we need it in the on_failure_jump | |
395 | of re_match_2.) */ | |
fa9a63c5 RM |
396 | start_memory, |
397 | ||
25fe55af RS |
398 | /* Stop remembering the text that is matched and store it in a |
399 | memory register. Followed by one byte with the register | |
400 | number, in the range 0 to one less than `re_nsub' in the | |
401 | pattern buffer, and one byte with the number of inner groups, | |
402 | just like `start_memory'. (We need the number of inner | |
403 | groups here because we don't have any easy way of finding the | |
404 | corresponding start_memory when we're at a stop_memory.) */ | |
fa9a63c5 RM |
405 | stop_memory, |
406 | ||
25fe55af RS |
407 | /* Match a duplicate of something remembered. Followed by one |
408 | byte containing the register number. */ | |
fa9a63c5 RM |
409 | duplicate, |
410 | ||
25fe55af | 411 | /* Fail unless at beginning of line. */ |
fa9a63c5 RM |
412 | begline, |
413 | ||
25fe55af | 414 | /* Fail unless at end of line. */ |
fa9a63c5 RM |
415 | endline, |
416 | ||
25fe55af RS |
417 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
418 | of string to be matched (if not). */ | |
fa9a63c5 RM |
419 | begbuf, |
420 | ||
25fe55af | 421 | /* Analogously, for end of buffer/string. */ |
fa9a63c5 | 422 | endbuf, |
5e69f11e | 423 | |
25fe55af | 424 | /* Followed by two byte relative address to which to jump. */ |
5e69f11e | 425 | jump, |
fa9a63c5 RM |
426 | |
427 | /* Same as jump, but marks the end of an alternative. */ | |
428 | jump_past_alt, | |
429 | ||
25fe55af RS |
430 | /* Followed by two-byte relative address of place to resume at |
431 | in case of failure. */ | |
fa9a63c5 | 432 | on_failure_jump, |
5e69f11e | 433 | |
25fe55af RS |
434 | /* Like on_failure_jump, but pushes a placeholder instead of the |
435 | current string position when executed. */ | |
fa9a63c5 | 436 | on_failure_keep_string_jump, |
5e69f11e | 437 | |
25fe55af RS |
438 | /* Throw away latest failure point and then jump to following |
439 | two-byte relative address. */ | |
fa9a63c5 RM |
440 | pop_failure_jump, |
441 | ||
25fe55af RS |
442 | /* Change to pop_failure_jump if know won't have to backtrack to |
443 | match; otherwise change to jump. This is used to jump | |
444 | back to the beginning of a repeat. If what follows this jump | |
445 | clearly won't match what the repeat does, such that we can be | |
446 | sure that there is no use backtracking out of repetitions | |
447 | already matched, then we change it to a pop_failure_jump. | |
448 | Followed by two-byte address. */ | |
fa9a63c5 RM |
449 | maybe_pop_jump, |
450 | ||
25fe55af RS |
451 | /* Jump to following two-byte address, and push a dummy failure |
452 | point. This failure point will be thrown away if an attempt | |
453 | is made to use it for a failure. A `+' construct makes this | |
454 | before the first repeat. Also used as an intermediary kind | |
455 | of jump when compiling an alternative. */ | |
fa9a63c5 RM |
456 | dummy_failure_jump, |
457 | ||
458 | /* Push a dummy failure point and continue. Used at the end of | |
459 | alternatives. */ | |
460 | push_dummy_failure, | |
461 | ||
25fe55af RS |
462 | /* Followed by two-byte relative address and two-byte number n. |
463 | After matching N times, jump to the address upon failure. */ | |
fa9a63c5 RM |
464 | succeed_n, |
465 | ||
25fe55af RS |
466 | /* Followed by two-byte relative address, and two-byte number n. |
467 | Jump to the address N times, then fail. */ | |
fa9a63c5 RM |
468 | jump_n, |
469 | ||
25fe55af RS |
470 | /* Set the following two-byte relative address to the |
471 | subsequent two-byte number. The address *includes* the two | |
472 | bytes of number. */ | |
fa9a63c5 RM |
473 | set_number_at, |
474 | ||
475 | wordchar, /* Matches any word-constituent character. */ | |
476 | notwordchar, /* Matches any char that is not a word-constituent. */ | |
477 | ||
478 | wordbeg, /* Succeeds if at word beginning. */ | |
479 | wordend, /* Succeeds if at word end. */ | |
480 | ||
481 | wordbound, /* Succeeds if at a word boundary. */ | |
25fe55af | 482 | notwordbound /* Succeeds if not at a word boundary. */ |
fa9a63c5 RM |
483 | |
484 | #ifdef emacs | |
485 | ,before_dot, /* Succeeds if before point. */ | |
486 | at_dot, /* Succeeds if at point. */ | |
487 | after_dot, /* Succeeds if after point. */ | |
488 | ||
489 | /* Matches any character whose syntax is specified. Followed by | |
25fe55af | 490 | a byte which contains a syntax code, e.g., Sword. */ |
fa9a63c5 RM |
491 | syntaxspec, |
492 | ||
493 | /* Matches any character whose syntax is not that specified. */ | |
b18215fc RS |
494 | notsyntaxspec, |
495 | ||
496 | /* Matches any character whose category-set contains the specified | |
25fe55af RS |
497 | category. The operator is followed by a byte which contains a |
498 | category code (mnemonic ASCII character). */ | |
b18215fc RS |
499 | categoryspec, |
500 | ||
501 | /* Matches any character whose category-set does not contain the | |
502 | specified category. The operator is followed by a byte which | |
503 | contains the category code (mnemonic ASCII character). */ | |
504 | notcategoryspec | |
fa9a63c5 RM |
505 | #endif /* emacs */ |
506 | } re_opcode_t; | |
507 | \f | |
508 | /* Common operations on the compiled pattern. */ | |
509 | ||
510 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
511 | ||
512 | #define STORE_NUMBER(destination, number) \ | |
513 | do { \ | |
514 | (destination)[0] = (number) & 0377; \ | |
515 | (destination)[1] = (number) >> 8; \ | |
516 | } while (0) | |
517 | ||
518 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
519 | the byte after where the number is stored. Therefore, DESTINATION | |
520 | must be an lvalue. */ | |
521 | ||
522 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
523 | do { \ | |
524 | STORE_NUMBER (destination, number); \ | |
525 | (destination) += 2; \ | |
526 | } while (0) | |
527 | ||
528 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
529 | at SOURCE. */ | |
530 | ||
531 | #define EXTRACT_NUMBER(destination, source) \ | |
532 | do { \ | |
533 | (destination) = *(source) & 0377; \ | |
534 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
535 | } while (0) | |
536 | ||
537 | #ifdef DEBUG | |
538 | static void | |
539 | extract_number (dest, source) | |
540 | int *dest; | |
541 | unsigned char *source; | |
542 | { | |
5e69f11e | 543 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
fa9a63c5 RM |
544 | *dest = *source & 0377; |
545 | *dest += temp << 8; | |
546 | } | |
547 | ||
25fe55af | 548 | #ifndef EXTRACT_MACROS /* To debug the macros. */ |
fa9a63c5 RM |
549 | #undef EXTRACT_NUMBER |
550 | #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
551 | #endif /* not EXTRACT_MACROS */ | |
552 | ||
553 | #endif /* DEBUG */ | |
554 | ||
555 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
556 | SOURCE must be an lvalue. */ | |
557 | ||
558 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
559 | do { \ | |
560 | EXTRACT_NUMBER (destination, source); \ | |
25fe55af | 561 | (source) += 2; \ |
fa9a63c5 RM |
562 | } while (0) |
563 | ||
564 | #ifdef DEBUG | |
565 | static void | |
566 | extract_number_and_incr (destination, source) | |
567 | int *destination; | |
568 | unsigned char **source; | |
5e69f11e | 569 | { |
fa9a63c5 RM |
570 | extract_number (destination, *source); |
571 | *source += 2; | |
572 | } | |
573 | ||
574 | #ifndef EXTRACT_MACROS | |
575 | #undef EXTRACT_NUMBER_AND_INCR | |
576 | #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
577 | extract_number_and_incr (&dest, &src) | |
578 | #endif /* not EXTRACT_MACROS */ | |
579 | ||
580 | #endif /* DEBUG */ | |
581 | \f | |
b18215fc RS |
582 | /* Store a multibyte character in three contiguous bytes starting |
583 | DESTINATION, and increment DESTINATION to the byte after where the | |
25fe55af | 584 | character is stored. Therefore, DESTINATION must be an lvalue. */ |
b18215fc RS |
585 | |
586 | #define STORE_CHARACTER_AND_INCR(destination, character) \ | |
587 | do { \ | |
588 | (destination)[0] = (character) & 0377; \ | |
589 | (destination)[1] = ((character) >> 8) & 0377; \ | |
590 | (destination)[2] = (character) >> 16; \ | |
591 | (destination) += 3; \ | |
592 | } while (0) | |
593 | ||
594 | /* Put into DESTINATION a character stored in three contiguous bytes | |
25fe55af | 595 | starting at SOURCE. */ |
b18215fc RS |
596 | |
597 | #define EXTRACT_CHARACTER(destination, source) \ | |
598 | do { \ | |
599 | (destination) = ((source)[0] \ | |
600 | | ((source)[1] << 8) \ | |
601 | | ((source)[2] << 16)); \ | |
602 | } while (0) | |
603 | ||
604 | ||
605 | /* Macros for charset. */ | |
606 | ||
607 | /* Size of bitmap of charset P in bytes. P is a start of charset, | |
608 | i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */ | |
609 | #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F) | |
610 | ||
611 | /* Nonzero if charset P has range table. */ | |
25fe55af | 612 | #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80) |
b18215fc RS |
613 | |
614 | /* Return the address of range table of charset P. But not the start | |
615 | of table itself, but the before where the number of ranges is | |
25fe55af | 616 | stored. `2 +' means to skip re_opcode_t and size of bitmap. */ |
b18215fc RS |
617 | #define CHARSET_RANGE_TABLE(p) (&(p)[2 + CHARSET_BITMAP_SIZE (p)]) |
618 | ||
619 | /* Test if C is listed in the bitmap of charset P. */ | |
620 | #define CHARSET_LOOKUP_BITMAP(p, c) \ | |
621 | ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \ | |
622 | && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH))) | |
623 | ||
624 | /* Return the address of end of RANGE_TABLE. COUNT is number of | |
25fe55af RS |
625 | ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2' |
626 | is start of range and end of range. `* 3' is size of each start | |
b18215fc RS |
627 | and end. */ |
628 | #define CHARSET_RANGE_TABLE_END(range_table, count) \ | |
629 | ((range_table) + (count) * 2 * 3) | |
630 | ||
25fe55af | 631 | /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in. |
b18215fc RS |
632 | COUNT is number of ranges in RANGE_TABLE. */ |
633 | #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \ | |
634 | do \ | |
635 | { \ | |
636 | int range_start, range_end; \ | |
637 | unsigned char *p; \ | |
638 | unsigned char *range_table_end \ | |
639 | = CHARSET_RANGE_TABLE_END ((range_table), (count)); \ | |
640 | \ | |
641 | for (p = (range_table); p < range_table_end; p += 2 * 3) \ | |
642 | { \ | |
643 | EXTRACT_CHARACTER (range_start, p); \ | |
644 | EXTRACT_CHARACTER (range_end, p + 3); \ | |
645 | \ | |
646 | if (range_start <= (c) && (c) <= range_end) \ | |
647 | { \ | |
648 | (not) = !(not); \ | |
649 | break; \ | |
650 | } \ | |
651 | } \ | |
652 | } \ | |
653 | while (0) | |
654 | ||
655 | /* Test if C is in range table of CHARSET. The flag NOT is negated if | |
656 | C is listed in it. */ | |
657 | #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \ | |
658 | do \ | |
659 | { \ | |
660 | /* Number of ranges in range table. */ \ | |
661 | int count; \ | |
662 | unsigned char *range_table = CHARSET_RANGE_TABLE (charset); \ | |
663 | \ | |
664 | EXTRACT_NUMBER_AND_INCR (count, range_table); \ | |
665 | CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \ | |
666 | } \ | |
667 | while (0) | |
668 | \f | |
fa9a63c5 RM |
669 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
670 | it is doing (if the variable `debug' is nonzero). If linked with the | |
671 | main program in `iregex.c', you can enter patterns and strings | |
672 | interactively. And if linked with the main program in `main.c' and | |
25fe55af | 673 | the other test files, you can run the already-written tests. */ |
fa9a63c5 RM |
674 | |
675 | #ifdef DEBUG | |
676 | ||
677 | /* We use standard I/O for debugging. */ | |
678 | #include <stdio.h> | |
679 | ||
680 | /* It is useful to test things that ``must'' be true when debugging. */ | |
681 | #include <assert.h> | |
682 | ||
683 | static int debug = 0; | |
684 | ||
685 | #define DEBUG_STATEMENT(e) e | |
686 | #define DEBUG_PRINT1(x) if (debug) printf (x) | |
687 | #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
688 | #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
689 | #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) | |
25fe55af | 690 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
fa9a63c5 RM |
691 | if (debug) print_partial_compiled_pattern (s, e) |
692 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
693 | if (debug) print_double_string (w, s1, sz1, s2, sz2) | |
694 | ||
695 | ||
696 | /* Print the fastmap in human-readable form. */ | |
697 | ||
698 | void | |
699 | print_fastmap (fastmap) | |
700 | char *fastmap; | |
701 | { | |
702 | unsigned was_a_range = 0; | |
5e69f11e RM |
703 | unsigned i = 0; |
704 | ||
fa9a63c5 RM |
705 | while (i < (1 << BYTEWIDTH)) |
706 | { | |
707 | if (fastmap[i++]) | |
708 | { | |
709 | was_a_range = 0; | |
25fe55af RS |
710 | putchar (i - 1); |
711 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
712 | { | |
713 | was_a_range = 1; | |
714 | i++; | |
715 | } | |
fa9a63c5 | 716 | if (was_a_range) |
25fe55af RS |
717 | { |
718 | printf ("-"); | |
719 | putchar (i - 1); | |
720 | } | |
721 | } | |
fa9a63c5 | 722 | } |
5e69f11e | 723 | putchar ('\n'); |
fa9a63c5 RM |
724 | } |
725 | ||
726 | ||
727 | /* Print a compiled pattern string in human-readable form, starting at | |
728 | the START pointer into it and ending just before the pointer END. */ | |
729 | ||
730 | void | |
731 | print_partial_compiled_pattern (start, end) | |
732 | unsigned char *start; | |
733 | unsigned char *end; | |
734 | { | |
735 | int mcnt, mcnt2; | |
736 | unsigned char *p = start; | |
737 | unsigned char *pend = end; | |
738 | ||
739 | if (start == NULL) | |
740 | { | |
741 | printf ("(null)\n"); | |
742 | return; | |
743 | } | |
5e69f11e | 744 | |
fa9a63c5 RM |
745 | /* Loop over pattern commands. */ |
746 | while (p < pend) | |
747 | { | |
748 | printf ("%d:\t", p - start); | |
749 | ||
750 | switch ((re_opcode_t) *p++) | |
751 | { | |
25fe55af RS |
752 | case no_op: |
753 | printf ("/no_op"); | |
754 | break; | |
fa9a63c5 RM |
755 | |
756 | case exactn: | |
757 | mcnt = *p++; | |
25fe55af RS |
758 | printf ("/exactn/%d", mcnt); |
759 | do | |
fa9a63c5 | 760 | { |
25fe55af | 761 | putchar ('/'); |
fa9a63c5 | 762 | putchar (*p++); |
25fe55af RS |
763 | } |
764 | while (--mcnt); | |
765 | break; | |
fa9a63c5 RM |
766 | |
767 | case start_memory: | |
25fe55af RS |
768 | mcnt = *p++; |
769 | printf ("/start_memory/%d/%d", mcnt, *p++); | |
770 | break; | |
fa9a63c5 RM |
771 | |
772 | case stop_memory: | |
25fe55af | 773 | mcnt = *p++; |
fa9a63c5 | 774 | printf ("/stop_memory/%d/%d", mcnt, *p++); |
25fe55af | 775 | break; |
fa9a63c5 RM |
776 | |
777 | case duplicate: | |
778 | printf ("/duplicate/%d", *p++); | |
779 | break; | |
780 | ||
781 | case anychar: | |
782 | printf ("/anychar"); | |
783 | break; | |
784 | ||
785 | case charset: | |
25fe55af RS |
786 | case charset_not: |
787 | { | |
788 | register int c, last = -100; | |
fa9a63c5 RM |
789 | register int in_range = 0; |
790 | ||
791 | printf ("/charset [%s", | |
25fe55af | 792 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
5e69f11e | 793 | |
25fe55af | 794 | assert (p + *p < pend); |
fa9a63c5 | 795 | |
25fe55af | 796 | for (c = 0; c < 256; c++) |
fa9a63c5 RM |
797 | if (c / 8 < *p |
798 | && (p[1 + (c/8)] & (1 << (c % 8)))) | |
799 | { | |
800 | /* Are we starting a range? */ | |
801 | if (last + 1 == c && ! in_range) | |
802 | { | |
803 | putchar ('-'); | |
804 | in_range = 1; | |
805 | } | |
806 | /* Have we broken a range? */ | |
807 | else if (last + 1 != c && in_range) | |
25fe55af | 808 | { |
fa9a63c5 RM |
809 | putchar (last); |
810 | in_range = 0; | |
811 | } | |
5e69f11e | 812 | |
fa9a63c5 RM |
813 | if (! in_range) |
814 | putchar (c); | |
815 | ||
816 | last = c; | |
25fe55af | 817 | } |
fa9a63c5 RM |
818 | |
819 | if (in_range) | |
820 | putchar (last); | |
821 | ||
822 | putchar (']'); | |
823 | ||
824 | p += 1 + *p; | |
825 | } | |
826 | break; | |
827 | ||
828 | case begline: | |
829 | printf ("/begline"); | |
25fe55af | 830 | break; |
fa9a63c5 RM |
831 | |
832 | case endline: | |
25fe55af RS |
833 | printf ("/endline"); |
834 | break; | |
fa9a63c5 RM |
835 | |
836 | case on_failure_jump: | |
25fe55af RS |
837 | extract_number_and_incr (&mcnt, &p); |
838 | printf ("/on_failure_jump to %d", p + mcnt - start); | |
839 | break; | |
fa9a63c5 RM |
840 | |
841 | case on_failure_keep_string_jump: | |
25fe55af RS |
842 | extract_number_and_incr (&mcnt, &p); |
843 | printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); | |
844 | break; | |
fa9a63c5 RM |
845 | |
846 | case dummy_failure_jump: | |
25fe55af RS |
847 | extract_number_and_incr (&mcnt, &p); |
848 | printf ("/dummy_failure_jump to %d", p + mcnt - start); | |
849 | break; | |
fa9a63c5 RM |
850 | |
851 | case push_dummy_failure: | |
25fe55af RS |
852 | printf ("/push_dummy_failure"); |
853 | break; | |
5e69f11e | 854 | |
25fe55af RS |
855 | case maybe_pop_jump: |
856 | extract_number_and_incr (&mcnt, &p); | |
857 | printf ("/maybe_pop_jump to %d", p + mcnt - start); | |
fa9a63c5 RM |
858 | break; |
859 | ||
25fe55af | 860 | case pop_failure_jump: |
fa9a63c5 | 861 | extract_number_and_incr (&mcnt, &p); |
25fe55af | 862 | printf ("/pop_failure_jump to %d", p + mcnt - start); |
5e69f11e RM |
863 | break; |
864 | ||
25fe55af | 865 | case jump_past_alt: |
fa9a63c5 | 866 | extract_number_and_incr (&mcnt, &p); |
25fe55af | 867 | printf ("/jump_past_alt to %d", p + mcnt - start); |
5e69f11e RM |
868 | break; |
869 | ||
25fe55af | 870 | case jump: |
fa9a63c5 | 871 | extract_number_and_incr (&mcnt, &p); |
25fe55af | 872 | printf ("/jump to %d", p + mcnt - start); |
fa9a63c5 RM |
873 | break; |
874 | ||
25fe55af RS |
875 | case succeed_n: |
876 | extract_number_and_incr (&mcnt, &p); | |
877 | extract_number_and_incr (&mcnt2, &p); | |
fa9a63c5 | 878 | printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2); |
25fe55af | 879 | break; |
5e69f11e | 880 | |
25fe55af RS |
881 | case jump_n: |
882 | extract_number_and_incr (&mcnt, &p); | |
883 | extract_number_and_incr (&mcnt2, &p); | |
fa9a63c5 | 884 | printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2); |
25fe55af | 885 | break; |
5e69f11e | 886 | |
25fe55af RS |
887 | case set_number_at: |
888 | extract_number_and_incr (&mcnt, &p); | |
889 | extract_number_and_incr (&mcnt2, &p); | |
fa9a63c5 | 890 | printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2); |
25fe55af | 891 | break; |
5e69f11e | 892 | |
25fe55af | 893 | case wordbound: |
fa9a63c5 RM |
894 | printf ("/wordbound"); |
895 | break; | |
896 | ||
897 | case notwordbound: | |
898 | printf ("/notwordbound"); | |
25fe55af | 899 | break; |
fa9a63c5 RM |
900 | |
901 | case wordbeg: | |
902 | printf ("/wordbeg"); | |
903 | break; | |
5e69f11e | 904 | |
fa9a63c5 RM |
905 | case wordend: |
906 | printf ("/wordend"); | |
5e69f11e | 907 | |
fa9a63c5 RM |
908 | #ifdef emacs |
909 | case before_dot: | |
910 | printf ("/before_dot"); | |
25fe55af | 911 | break; |
fa9a63c5 RM |
912 | |
913 | case at_dot: | |
914 | printf ("/at_dot"); | |
25fe55af | 915 | break; |
fa9a63c5 RM |
916 | |
917 | case after_dot: | |
918 | printf ("/after_dot"); | |
25fe55af | 919 | break; |
fa9a63c5 RM |
920 | |
921 | case syntaxspec: | |
25fe55af | 922 | printf ("/syntaxspec"); |
fa9a63c5 RM |
923 | mcnt = *p++; |
924 | printf ("/%d", mcnt); | |
25fe55af | 925 | break; |
5e69f11e | 926 | |
fa9a63c5 | 927 | case notsyntaxspec: |
25fe55af | 928 | printf ("/notsyntaxspec"); |
fa9a63c5 RM |
929 | mcnt = *p++; |
930 | printf ("/%d", mcnt); | |
931 | break; | |
932 | #endif /* emacs */ | |
933 | ||
934 | case wordchar: | |
935 | printf ("/wordchar"); | |
25fe55af | 936 | break; |
5e69f11e | 937 | |
fa9a63c5 RM |
938 | case notwordchar: |
939 | printf ("/notwordchar"); | |
25fe55af | 940 | break; |
fa9a63c5 RM |
941 | |
942 | case begbuf: | |
943 | printf ("/begbuf"); | |
25fe55af | 944 | break; |
fa9a63c5 RM |
945 | |
946 | case endbuf: | |
947 | printf ("/endbuf"); | |
25fe55af | 948 | break; |
fa9a63c5 | 949 | |
25fe55af RS |
950 | default: |
951 | printf ("?%d", *(p-1)); | |
fa9a63c5 RM |
952 | } |
953 | ||
954 | putchar ('\n'); | |
955 | } | |
956 | ||
957 | printf ("%d:\tend of pattern.\n", p - start); | |
958 | } | |
959 | ||
960 | ||
961 | void | |
962 | print_compiled_pattern (bufp) | |
963 | struct re_pattern_buffer *bufp; | |
964 | { | |
965 | unsigned char *buffer = bufp->buffer; | |
966 | ||
967 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
968 | printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); | |
969 | ||
970 | if (bufp->fastmap_accurate && bufp->fastmap) | |
971 | { | |
972 | printf ("fastmap: "); | |
973 | print_fastmap (bufp->fastmap); | |
974 | } | |
975 | ||
976 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
977 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
978 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
979 | printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
980 | printf ("no_sub: %d\t", bufp->no_sub); | |
981 | printf ("not_bol: %d\t", bufp->not_bol); | |
982 | printf ("not_eol: %d\t", bufp->not_eol); | |
983 | printf ("syntax: %d\n", bufp->syntax); | |
984 | /* Perhaps we should print the translate table? */ | |
985 | } | |
986 | ||
987 | ||
988 | void | |
989 | print_double_string (where, string1, size1, string2, size2) | |
990 | const char *where; | |
991 | const char *string1; | |
992 | const char *string2; | |
993 | int size1; | |
994 | int size2; | |
995 | { | |
996 | unsigned this_char; | |
5e69f11e | 997 | |
fa9a63c5 RM |
998 | if (where == NULL) |
999 | printf ("(null)"); | |
1000 | else | |
1001 | { | |
1002 | if (FIRST_STRING_P (where)) | |
25fe55af RS |
1003 | { |
1004 | for (this_char = where - string1; this_char < size1; this_char++) | |
1005 | putchar (string1[this_char]); | |
fa9a63c5 | 1006 | |
25fe55af RS |
1007 | where = string2; |
1008 | } | |
fa9a63c5 RM |
1009 | |
1010 | for (this_char = where - string2; this_char < size2; this_char++) | |
25fe55af | 1011 | putchar (string2[this_char]); |
fa9a63c5 RM |
1012 | } |
1013 | } | |
1014 | ||
1015 | #else /* not DEBUG */ | |
1016 | ||
1017 | #undef assert | |
1018 | #define assert(e) | |
1019 | ||
1020 | #define DEBUG_STATEMENT(e) | |
1021 | #define DEBUG_PRINT1(x) | |
1022 | #define DEBUG_PRINT2(x1, x2) | |
1023 | #define DEBUG_PRINT3(x1, x2, x3) | |
1024 | #define DEBUG_PRINT4(x1, x2, x3, x4) | |
1025 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1026 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
1027 | ||
1028 | #endif /* not DEBUG */ | |
1029 | \f | |
1030 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
1031 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1032 | syntax, so it can be changed between regex compilations. */ | |
1033 | /* This has no initializer because initialized variables in Emacs | |
1034 | become read-only after dumping. */ | |
1035 | reg_syntax_t re_syntax_options; | |
1036 | ||
1037 | ||
1038 | /* Specify the precise syntax of regexps for compilation. This provides | |
1039 | for compatibility for various utilities which historically have | |
1040 | different, incompatible syntaxes. | |
1041 | ||
1042 | The argument SYNTAX is a bit mask comprised of the various bits | |
25fe55af | 1043 | defined in regex.h. We return the old syntax. */ |
fa9a63c5 RM |
1044 | |
1045 | reg_syntax_t | |
1046 | re_set_syntax (syntax) | |
1047 | reg_syntax_t syntax; | |
1048 | { | |
1049 | reg_syntax_t ret = re_syntax_options; | |
5e69f11e | 1050 | |
fa9a63c5 RM |
1051 | re_syntax_options = syntax; |
1052 | return ret; | |
1053 | } | |
1054 | \f | |
1055 | /* This table gives an error message for each of the error codes listed | |
25fe55af | 1056 | in regex.h. Obviously the order here has to be same as there. |
fa9a63c5 | 1057 | POSIX doesn't require that we do anything for REG_NOERROR, |
25fe55af | 1058 | but why not be nice? */ |
fa9a63c5 RM |
1059 | |
1060 | static const char *re_error_msgid[] = | |
5e69f11e RM |
1061 | { |
1062 | gettext_noop ("Success"), /* REG_NOERROR */ | |
1063 | gettext_noop ("No match"), /* REG_NOMATCH */ | |
1064 | gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ | |
1065 | gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ | |
1066 | gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ | |
1067 | gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ | |
1068 | gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ | |
1069 | gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ | |
1070 | gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ | |
1071 | gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ | |
1072 | gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ | |
1073 | gettext_noop ("Invalid range end"), /* REG_ERANGE */ | |
1074 | gettext_noop ("Memory exhausted"), /* REG_ESPACE */ | |
1075 | gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ | |
1076 | gettext_noop ("Premature end of regular expression"), /* REG_EEND */ | |
1077 | gettext_noop ("Regular expression too big"), /* REG_ESIZE */ | |
1078 | gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ | |
fa9a63c5 RM |
1079 | }; |
1080 | \f | |
25fe55af | 1081 | /* Avoiding alloca during matching, to placate r_alloc. */ |
fa9a63c5 RM |
1082 | |
1083 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1084 | searching and matching functions should not call alloca. On some | |
1085 | systems, alloca is implemented in terms of malloc, and if we're | |
1086 | using the relocating allocator routines, then malloc could cause a | |
1087 | relocation, which might (if the strings being searched are in the | |
1088 | ralloc heap) shift the data out from underneath the regexp | |
1089 | routines. | |
1090 | ||
5e69f11e | 1091 | Here's another reason to avoid allocation: Emacs |
fa9a63c5 RM |
1092 | processes input from X in a signal handler; processing X input may |
1093 | call malloc; if input arrives while a matching routine is calling | |
1094 | malloc, then we're scrod. But Emacs can't just block input while | |
1095 | calling matching routines; then we don't notice interrupts when | |
1096 | they come in. So, Emacs blocks input around all regexp calls | |
1097 | except the matching calls, which it leaves unprotected, in the | |
1098 | faith that they will not malloc. */ | |
1099 | ||
1100 | /* Normally, this is fine. */ | |
1101 | #define MATCH_MAY_ALLOCATE | |
1102 | ||
1103 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
1104 | what config.h may say. So don't take precautions for it. */ | |
1105 | #ifdef __GNUC__ | |
1106 | #undef C_ALLOCA | |
1107 | #endif | |
1108 | ||
1109 | /* The match routines may not allocate if (1) they would do it with malloc | |
1110 | and (2) it's not safe for them to use malloc. | |
1111 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1112 | failure stack, but we would still use it for the register vectors; | |
25fe55af | 1113 | so REL_ALLOC should not affect this. */ |
fa9a63c5 RM |
1114 | #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs) |
1115 | #undef MATCH_MAY_ALLOCATE | |
1116 | #endif | |
1117 | ||
1118 | \f | |
1119 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1120 | re_match_2 use a failure stack. These have to be macros because of | |
1121 | REGEX_ALLOCATE_STACK. */ | |
5e69f11e | 1122 | |
fa9a63c5 | 1123 | |
320a2a73 | 1124 | /* Approximate number of failure points for which to initially allocate space |
fa9a63c5 RM |
1125 | when matching. If this number is exceeded, we allocate more |
1126 | space, so it is not a hard limit. */ | |
1127 | #ifndef INIT_FAILURE_ALLOC | |
320a2a73 | 1128 | #define INIT_FAILURE_ALLOC 20 |
fa9a63c5 RM |
1129 | #endif |
1130 | ||
1131 | /* Roughly the maximum number of failure points on the stack. Would be | |
320a2a73 | 1132 | exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed. |
fa9a63c5 | 1133 | This is a variable only so users of regex can assign to it; we never |
25fe55af | 1134 | change it ourselves. */ |
fa9a63c5 | 1135 | #if defined (MATCH_MAY_ALLOCATE) |
320a2a73 KH |
1136 | /* Note that 4400 is enough to cause a crash on Alpha OSF/1, |
1137 | whose default stack limit is 2mb. In order for a larger | |
1138 | value to work reliably, you have to try to make it accord | |
1139 | with the process stack limit. */ | |
1140 | int re_max_failures = 40000; | |
fa9a63c5 | 1141 | #else |
320a2a73 | 1142 | int re_max_failures = 4000; |
fa9a63c5 RM |
1143 | #endif |
1144 | ||
1145 | union fail_stack_elt | |
1146 | { | |
1147 | unsigned char *pointer; | |
1148 | int integer; | |
1149 | }; | |
1150 | ||
1151 | typedef union fail_stack_elt fail_stack_elt_t; | |
1152 | ||
1153 | typedef struct | |
1154 | { | |
1155 | fail_stack_elt_t *stack; | |
1156 | unsigned size; | |
1157 | unsigned avail; /* Offset of next open position. */ | |
1158 | } fail_stack_type; | |
1159 | ||
1160 | #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
1161 | #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
1162 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
1163 | ||
1164 | ||
1165 | /* Define macros to initialize and free the failure stack. | |
1166 | Do `return -2' if the alloc fails. */ | |
1167 | ||
1168 | #ifdef MATCH_MAY_ALLOCATE | |
1169 | #define INIT_FAIL_STACK() \ | |
1170 | do { \ | |
1171 | fail_stack.stack = (fail_stack_elt_t *) \ | |
320a2a73 KH |
1172 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \ |
1173 | * sizeof (fail_stack_elt_t)); \ | |
fa9a63c5 RM |
1174 | \ |
1175 | if (fail_stack.stack == NULL) \ | |
1176 | return -2; \ | |
1177 | \ | |
1178 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1179 | fail_stack.avail = 0; \ | |
1180 | } while (0) | |
1181 | ||
1182 | #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) | |
1183 | #else | |
1184 | #define INIT_FAIL_STACK() \ | |
1185 | do { \ | |
1186 | fail_stack.avail = 0; \ | |
1187 | } while (0) | |
1188 | ||
1189 | #define RESET_FAIL_STACK() | |
1190 | #endif | |
1191 | ||
1192 | ||
320a2a73 KH |
1193 | /* Double the size of FAIL_STACK, up to a limit |
1194 | which allows approximately `re_max_failures' items. | |
fa9a63c5 RM |
1195 | |
1196 | Return 1 if succeeds, and 0 if either ran out of memory | |
5e69f11e RM |
1197 | allocating space for it or it was already too large. |
1198 | ||
25fe55af | 1199 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
fa9a63c5 | 1200 | |
320a2a73 KH |
1201 | /* Factor to increase the failure stack size by |
1202 | when we increase it. | |
1203 | This used to be 2, but 2 was too wasteful | |
1204 | because the old discarded stacks added up to as much space | |
1205 | were as ultimate, maximum-size stack. */ | |
1206 | #define FAIL_STACK_GROWTH_FACTOR 4 | |
1207 | ||
1208 | #define GROW_FAIL_STACK(fail_stack) \ | |
eead07d6 KH |
1209 | (((fail_stack).size * sizeof (fail_stack_elt_t) \ |
1210 | >= re_max_failures * TYPICAL_FAILURE_SIZE) \ | |
fa9a63c5 | 1211 | ? 0 \ |
320a2a73 KH |
1212 | : ((fail_stack).stack \ |
1213 | = (fail_stack_elt_t *) \ | |
25fe55af RS |
1214 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
1215 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
320a2a73 KH |
1216 | MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
1217 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1218 | * FAIL_STACK_GROWTH_FACTOR))), \ | |
fa9a63c5 RM |
1219 | \ |
1220 | (fail_stack).stack == NULL \ | |
1221 | ? 0 \ | |
6453db45 KH |
1222 | : ((fail_stack).size \ |
1223 | = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ | |
1224 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1225 | * FAIL_STACK_GROWTH_FACTOR)) \ | |
1226 | / sizeof (fail_stack_elt_t)), \ | |
25fe55af | 1227 | 1))) |
fa9a63c5 RM |
1228 | |
1229 | ||
5e69f11e | 1230 | /* Push pointer POINTER on FAIL_STACK. |
fa9a63c5 RM |
1231 | Return 1 if was able to do so and 0 if ran out of memory allocating |
1232 | space to do so. */ | |
1233 | #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ | |
1234 | ((FAIL_STACK_FULL () \ | |
320a2a73 | 1235 | && !GROW_FAIL_STACK (FAIL_STACK)) \ |
fa9a63c5 RM |
1236 | ? 0 \ |
1237 | : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ | |
1238 | 1)) | |
1239 | ||
1240 | /* Push a pointer value onto the failure stack. | |
1241 | Assumes the variable `fail_stack'. Probably should only | |
25fe55af | 1242 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1243 | #define PUSH_FAILURE_POINTER(item) \ |
1244 | fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item) | |
1245 | ||
1246 | /* This pushes an integer-valued item onto the failure stack. | |
1247 | Assumes the variable `fail_stack'. Probably should only | |
25fe55af | 1248 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1249 | #define PUSH_FAILURE_INT(item) \ |
1250 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1251 | ||
1252 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1253 | Assumes the variable `fail_stack'. Probably should only | |
25fe55af | 1254 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1255 | #define PUSH_FAILURE_ELT(item) \ |
1256 | fail_stack.stack[fail_stack.avail++] = (item) | |
1257 | ||
1258 | /* These three POP... operations complement the three PUSH... operations. | |
1259 | All assume that `fail_stack' is nonempty. */ | |
1260 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1261 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1262 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1263 | ||
1264 | /* Used to omit pushing failure point id's when we're not debugging. */ | |
1265 | #ifdef DEBUG | |
1266 | #define DEBUG_PUSH PUSH_FAILURE_INT | |
1267 | #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () | |
1268 | #else | |
1269 | #define DEBUG_PUSH(item) | |
1270 | #define DEBUG_POP(item_addr) | |
1271 | #endif | |
1272 | ||
1273 | ||
1274 | /* Push the information about the state we will need | |
5e69f11e RM |
1275 | if we ever fail back to it. |
1276 | ||
fa9a63c5 | 1277 | Requires variables fail_stack, regstart, regend, reg_info, and |
320a2a73 | 1278 | num_regs be declared. GROW_FAIL_STACK requires `destination' be |
fa9a63c5 | 1279 | declared. |
5e69f11e | 1280 | |
fa9a63c5 RM |
1281 | Does `return FAILURE_CODE' if runs out of memory. */ |
1282 | ||
1283 | #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
1284 | do { \ | |
1285 | char *destination; \ | |
1286 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1287 | of 0 + -1 isn't done as unsigned. */ \ | |
1288 | int this_reg; \ | |
25fe55af | 1289 | \ |
fa9a63c5 RM |
1290 | DEBUG_STATEMENT (failure_id++); \ |
1291 | DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
1292 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
1293 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
25fe55af | 1294 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ |
fa9a63c5 RM |
1295 | \ |
1296 | DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ | |
25fe55af | 1297 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ |
fa9a63c5 RM |
1298 | \ |
1299 | /* Ensure we have enough space allocated for what we will push. */ \ | |
1300 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
1301 | { \ | |
320a2a73 | 1302 | if (!GROW_FAIL_STACK (fail_stack)) \ |
25fe55af | 1303 | return failure_code; \ |
fa9a63c5 | 1304 | \ |
25fe55af | 1305 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ |
fa9a63c5 | 1306 | (fail_stack).size); \ |
25fe55af | 1307 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ |
fa9a63c5 RM |
1308 | } \ |
1309 | \ | |
1310 | /* Push the info, starting with the registers. */ \ | |
1311 | DEBUG_PRINT1 ("\n"); \ | |
1312 | \ | |
68d96f02 | 1313 | if (1) \ |
faec11db RS |
1314 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ |
1315 | this_reg++) \ | |
1316 | { \ | |
1317 | DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ | |
1318 | DEBUG_STATEMENT (num_regs_pushed++); \ | |
fa9a63c5 | 1319 | \ |
faec11db RS |
1320 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ |
1321 | PUSH_FAILURE_POINTER (regstart[this_reg]); \ | |
fa9a63c5 | 1322 | \ |
faec11db RS |
1323 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ |
1324 | PUSH_FAILURE_POINTER (regend[this_reg]); \ | |
1325 | \ | |
1326 | DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ | |
1327 | DEBUG_PRINT2 (" match_null=%d", \ | |
1328 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
1329 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
1330 | DEBUG_PRINT2 (" matched_something=%d", \ | |
1331 | MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1332 | DEBUG_PRINT2 (" ever_matched=%d", \ | |
1333 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1334 | DEBUG_PRINT1 ("\n"); \ | |
1335 | PUSH_FAILURE_ELT (reg_info[this_reg].word); \ | |
1336 | } \ | |
fa9a63c5 RM |
1337 | \ |
1338 | DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ | |
1339 | PUSH_FAILURE_INT (lowest_active_reg); \ | |
1340 | \ | |
1341 | DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ | |
1342 | PUSH_FAILURE_INT (highest_active_reg); \ | |
1343 | \ | |
1344 | DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ | |
1345 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
1346 | PUSH_FAILURE_POINTER (pattern_place); \ | |
1347 | \ | |
1348 | DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ | |
25fe55af | 1349 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ |
fa9a63c5 RM |
1350 | size2); \ |
1351 | DEBUG_PRINT1 ("'\n"); \ | |
1352 | PUSH_FAILURE_POINTER (string_place); \ | |
1353 | \ | |
1354 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
1355 | DEBUG_PUSH (failure_id); \ | |
1356 | } while (0) | |
1357 | ||
1358 | /* This is the number of items that are pushed and popped on the stack | |
1359 | for each register. */ | |
1360 | #define NUM_REG_ITEMS 3 | |
1361 | ||
1362 | /* Individual items aside from the registers. */ | |
1363 | #ifdef DEBUG | |
1364 | #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
1365 | #else | |
1366 | #define NUM_NONREG_ITEMS 4 | |
1367 | #endif | |
1368 | ||
320a2a73 KH |
1369 | /* Estimate the size of data pushed by a typical failure stack entry. |
1370 | An estimate is all we need, because all we use this for | |
1371 | is to choose a limit for how big to make the failure stack. */ | |
1372 | ||
1373 | #define TYPICAL_FAILURE_SIZE 20 | |
fa9a63c5 | 1374 | |
320a2a73 KH |
1375 | /* This is how many items we actually use for a failure point. |
1376 | It depends on the regexp. */ | |
faec11db | 1377 | #define NUM_FAILURE_ITEMS \ |
68d96f02 | 1378 | (((0 \ |
faec11db RS |
1379 | ? 0 : highest_active_reg - lowest_active_reg + 1) \ |
1380 | * NUM_REG_ITEMS) \ | |
1381 | + NUM_NONREG_ITEMS) | |
fa9a63c5 RM |
1382 | |
1383 | /* How many items can still be added to the stack without overflowing it. */ | |
1384 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1385 | ||
1386 | ||
1387 | /* Pops what PUSH_FAIL_STACK pushes. | |
1388 | ||
1389 | We restore into the parameters, all of which should be lvalues: | |
1390 | STR -- the saved data position. | |
1391 | PAT -- the saved pattern position. | |
1392 | LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
1393 | REGSTART, REGEND -- arrays of string positions. | |
1394 | REG_INFO -- array of information about each subexpression. | |
5e69f11e | 1395 | |
fa9a63c5 | 1396 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
25fe55af | 1397 | `pend', `string1', `size1', `string2', and `size2'. */ |
fa9a63c5 RM |
1398 | |
1399 | #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
1400 | { \ | |
1401 | DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ | |
1402 | int this_reg; \ | |
1403 | const unsigned char *string_temp; \ | |
1404 | \ | |
1405 | assert (!FAIL_STACK_EMPTY ()); \ | |
1406 | \ | |
1407 | /* Remove failure points and point to how many regs pushed. */ \ | |
1408 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1409 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
25fe55af | 1410 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
fa9a63c5 RM |
1411 | \ |
1412 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
1413 | \ | |
1414 | DEBUG_POP (&failure_id); \ | |
1415 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
1416 | \ | |
1417 | /* If the saved string location is NULL, it came from an \ | |
1418 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1419 | saved NULL, thus retaining our current position in the string. */ \ | |
1420 | string_temp = POP_FAILURE_POINTER (); \ | |
1421 | if (string_temp != NULL) \ | |
1422 | str = (const char *) string_temp; \ | |
1423 | \ | |
1424 | DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ | |
1425 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
1426 | DEBUG_PRINT1 ("'\n"); \ | |
1427 | \ | |
1428 | pat = (unsigned char *) POP_FAILURE_POINTER (); \ | |
1429 | DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ | |
1430 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
1431 | \ | |
1432 | /* Restore register info. */ \ | |
1433 | high_reg = (unsigned) POP_FAILURE_INT (); \ | |
1434 | DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ | |
1435 | \ | |
1436 | low_reg = (unsigned) POP_FAILURE_INT (); \ | |
1437 | DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ | |
1438 | \ | |
68d96f02 | 1439 | if (1) \ |
faec11db RS |
1440 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ |
1441 | { \ | |
25fe55af | 1442 | DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ |
fa9a63c5 | 1443 | \ |
faec11db | 1444 | reg_info[this_reg].word = POP_FAILURE_ELT (); \ |
25fe55af | 1445 | DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ |
fa9a63c5 | 1446 | \ |
faec11db | 1447 | regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \ |
25fe55af | 1448 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ |
fa9a63c5 | 1449 | \ |
faec11db | 1450 | regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \ |
25fe55af | 1451 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ |
faec11db | 1452 | } \ |
6676cb1c RS |
1453 | else \ |
1454 | { \ | |
1455 | for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ | |
1456 | { \ | |
4fcecab1 | 1457 | reg_info[this_reg].word.integer = 0; \ |
6676cb1c RS |
1458 | regend[this_reg] = 0; \ |
1459 | regstart[this_reg] = 0; \ | |
1460 | } \ | |
1461 | highest_active_reg = high_reg; \ | |
1462 | } \ | |
fa9a63c5 RM |
1463 | \ |
1464 | set_regs_matched_done = 0; \ | |
1465 | DEBUG_STATEMENT (nfailure_points_popped++); \ | |
1466 | } /* POP_FAILURE_POINT */ | |
1467 | ||
1468 | ||
1469 | \f | |
1470 | /* Structure for per-register (a.k.a. per-group) information. | |
1471 | Other register information, such as the | |
1472 | starting and ending positions (which are addresses), and the list of | |
1473 | inner groups (which is a bits list) are maintained in separate | |
5e69f11e RM |
1474 | variables. |
1475 | ||
fa9a63c5 RM |
1476 | We are making a (strictly speaking) nonportable assumption here: that |
1477 | the compiler will pack our bit fields into something that fits into | |
1478 | the type of `word', i.e., is something that fits into one item on the | |
1479 | failure stack. */ | |
1480 | ||
1481 | typedef union | |
1482 | { | |
1483 | fail_stack_elt_t word; | |
1484 | struct | |
1485 | { | |
1486 | /* This field is one if this group can match the empty string, | |
25fe55af | 1487 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ |
fa9a63c5 RM |
1488 | #define MATCH_NULL_UNSET_VALUE 3 |
1489 | unsigned match_null_string_p : 2; | |
1490 | unsigned is_active : 1; | |
1491 | unsigned matched_something : 1; | |
1492 | unsigned ever_matched_something : 1; | |
1493 | } bits; | |
1494 | } register_info_type; | |
1495 | ||
1496 | #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
1497 | #define IS_ACTIVE(R) ((R).bits.is_active) | |
1498 | #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
1499 | #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
1500 | ||
1501 | ||
1502 | /* Call this when have matched a real character; it sets `matched' flags | |
1503 | for the subexpressions which we are currently inside. Also records | |
1504 | that those subexprs have matched. */ | |
1505 | #define SET_REGS_MATCHED() \ | |
1506 | do \ | |
1507 | { \ | |
1508 | if (!set_regs_matched_done) \ | |
1509 | { \ | |
1510 | unsigned r; \ | |
1511 | set_regs_matched_done = 1; \ | |
1512 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
1513 | { \ | |
1514 | MATCHED_SOMETHING (reg_info[r]) \ | |
1515 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
1516 | = 1; \ | |
1517 | } \ | |
1518 | } \ | |
1519 | } \ | |
1520 | while (0) | |
1521 | ||
1522 | /* Registers are set to a sentinel when they haven't yet matched. */ | |
1523 | static char reg_unset_dummy; | |
1524 | #define REG_UNSET_VALUE (®_unset_dummy) | |
1525 | #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
1526 | \f | |
1527 | /* Subroutine declarations and macros for regex_compile. */ | |
1528 | ||
1529 | static void store_op1 (), store_op2 (); | |
1530 | static void insert_op1 (), insert_op2 (); | |
1531 | static boolean at_begline_loc_p (), at_endline_loc_p (); | |
1532 | static boolean group_in_compile_stack (); | |
1533 | static reg_errcode_t compile_range (); | |
1534 | ||
5e69f11e | 1535 | /* Fetch the next character in the uncompiled pattern---translating it |
fa9a63c5 RM |
1536 | if necessary. Also cast from a signed character in the constant |
1537 | string passed to us by the user to an unsigned char that we can use | |
1538 | as an array index (in, e.g., `translate'). */ | |
6676cb1c | 1539 | #ifndef PATFETCH |
fa9a63c5 RM |
1540 | #define PATFETCH(c) \ |
1541 | do {if (p == pend) return REG_EEND; \ | |
1542 | c = (unsigned char) *p++; \ | |
e934739e | 1543 | if (translate) c = RE_TRANSLATE (translate, c); \ |
fa9a63c5 | 1544 | } while (0) |
6676cb1c | 1545 | #endif |
fa9a63c5 RM |
1546 | |
1547 | /* Fetch the next character in the uncompiled pattern, with no | |
25fe55af | 1548 | translation. */ |
fa9a63c5 RM |
1549 | #define PATFETCH_RAW(c) \ |
1550 | do {if (p == pend) return REG_EEND; \ | |
25fe55af | 1551 | c = (unsigned char) *p++; \ |
fa9a63c5 RM |
1552 | } while (0) |
1553 | ||
1554 | /* Go backwards one character in the pattern. */ | |
1555 | #define PATUNFETCH p-- | |
1556 | ||
1557 | ||
1558 | /* If `translate' is non-null, return translate[D], else just D. We | |
1559 | cast the subscript to translate because some data is declared as | |
1560 | `char *', to avoid warnings when a string constant is passed. But | |
1561 | when we use a character as a subscript we must make it unsigned. */ | |
6676cb1c RS |
1562 | #ifndef TRANSLATE |
1563 | #define TRANSLATE(d) \ | |
e934739e | 1564 | (translate ? (unsigned) RE_TRANSLATE (translate, (unsigned) (d)) : (d)) |
6676cb1c | 1565 | #endif |
fa9a63c5 RM |
1566 | |
1567 | ||
1568 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1569 | ||
1570 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1571 | #define INIT_BUF_SIZE 32 | |
1572 | ||
25fe55af | 1573 | /* Make sure we have at least N more bytes of space in buffer. */ |
fa9a63c5 RM |
1574 | #define GET_BUFFER_SPACE(n) \ |
1575 | while (b - bufp->buffer + (n) > bufp->allocated) \ | |
1576 | EXTEND_BUFFER () | |
1577 | ||
1578 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1579 | #define BUF_PUSH(c) \ | |
1580 | do { \ | |
1581 | GET_BUFFER_SPACE (1); \ | |
1582 | *b++ = (unsigned char) (c); \ | |
1583 | } while (0) | |
1584 | ||
1585 | ||
1586 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1587 | #define BUF_PUSH_2(c1, c2) \ | |
1588 | do { \ | |
1589 | GET_BUFFER_SPACE (2); \ | |
1590 | *b++ = (unsigned char) (c1); \ | |
1591 | *b++ = (unsigned char) (c2); \ | |
1592 | } while (0) | |
1593 | ||
1594 | ||
25fe55af | 1595 | /* As with BUF_PUSH_2, except for three bytes. */ |
fa9a63c5 RM |
1596 | #define BUF_PUSH_3(c1, c2, c3) \ |
1597 | do { \ | |
1598 | GET_BUFFER_SPACE (3); \ | |
1599 | *b++ = (unsigned char) (c1); \ | |
1600 | *b++ = (unsigned char) (c2); \ | |
1601 | *b++ = (unsigned char) (c3); \ | |
1602 | } while (0) | |
1603 | ||
1604 | ||
1605 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
25fe55af | 1606 | relative address offset by the three bytes the jump itself occupies. */ |
fa9a63c5 RM |
1607 | #define STORE_JUMP(op, loc, to) \ |
1608 | store_op1 (op, loc, (to) - (loc) - 3) | |
1609 | ||
1610 | /* Likewise, for a two-argument jump. */ | |
1611 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1612 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1613 | ||
25fe55af | 1614 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
fa9a63c5 RM |
1615 | #define INSERT_JUMP(op, loc, to) \ |
1616 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1617 | ||
1618 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1619 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1620 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1621 | ||
1622 | ||
1623 | /* This is not an arbitrary limit: the arguments which represent offsets | |
25fe55af | 1624 | into the pattern are two bytes long. So if 2^16 bytes turns out to |
fa9a63c5 RM |
1625 | be too small, many things would have to change. */ |
1626 | #define MAX_BUF_SIZE (1L << 16) | |
1627 | ||
1628 | ||
1629 | /* Extend the buffer by twice its current size via realloc and | |
1630 | reset the pointers that pointed into the old block to point to the | |
1631 | correct places in the new one. If extending the buffer results in it | |
25fe55af | 1632 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
fa9a63c5 | 1633 | #define EXTEND_BUFFER() \ |
25fe55af | 1634 | do { \ |
fa9a63c5 | 1635 | unsigned char *old_buffer = bufp->buffer; \ |
25fe55af | 1636 | if (bufp->allocated == MAX_BUF_SIZE) \ |
fa9a63c5 RM |
1637 | return REG_ESIZE; \ |
1638 | bufp->allocated <<= 1; \ | |
1639 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
25fe55af | 1640 | bufp->allocated = MAX_BUF_SIZE; \ |
fa9a63c5 RM |
1641 | bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ |
1642 | if (bufp->buffer == NULL) \ | |
1643 | return REG_ESPACE; \ | |
1644 | /* If the buffer moved, move all the pointers into it. */ \ | |
1645 | if (old_buffer != bufp->buffer) \ | |
1646 | { \ | |
25fe55af RS |
1647 | b = (b - old_buffer) + bufp->buffer; \ |
1648 | begalt = (begalt - old_buffer) + bufp->buffer; \ | |
1649 | if (fixup_alt_jump) \ | |
1650 | fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ | |
1651 | if (laststart) \ | |
1652 | laststart = (laststart - old_buffer) + bufp->buffer; \ | |
1653 | if (pending_exact) \ | |
1654 | pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
fa9a63c5 RM |
1655 | } \ |
1656 | } while (0) | |
1657 | ||
1658 | ||
1659 | /* Since we have one byte reserved for the register number argument to | |
1660 | {start,stop}_memory, the maximum number of groups we can report | |
1661 | things about is what fits in that byte. */ | |
1662 | #define MAX_REGNUM 255 | |
1663 | ||
1664 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1665 | ignore the excess. */ | |
1666 | typedef unsigned regnum_t; | |
1667 | ||
1668 | ||
1669 | /* Macros for the compile stack. */ | |
1670 | ||
1671 | /* Since offsets can go either forwards or backwards, this type needs to | |
25fe55af | 1672 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
fa9a63c5 RM |
1673 | typedef int pattern_offset_t; |
1674 | ||
1675 | typedef struct | |
1676 | { | |
1677 | pattern_offset_t begalt_offset; | |
1678 | pattern_offset_t fixup_alt_jump; | |
1679 | pattern_offset_t inner_group_offset; | |
5e69f11e | 1680 | pattern_offset_t laststart_offset; |
fa9a63c5 RM |
1681 | regnum_t regnum; |
1682 | } compile_stack_elt_t; | |
1683 | ||
1684 | ||
1685 | typedef struct | |
1686 | { | |
1687 | compile_stack_elt_t *stack; | |
1688 | unsigned size; | |
1689 | unsigned avail; /* Offset of next open position. */ | |
1690 | } compile_stack_type; | |
1691 | ||
1692 | ||
1693 | #define INIT_COMPILE_STACK_SIZE 32 | |
1694 | ||
1695 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1696 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1697 | ||
25fe55af | 1698 | /* The next available element. */ |
fa9a63c5 RM |
1699 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
1700 | ||
1701 | ||
b18215fc RS |
1702 | /* Structure to manage work area for range table. */ |
1703 | struct range_table_work_area | |
1704 | { | |
1705 | int *table; /* actual work area. */ | |
1706 | int allocated; /* allocated size for work area in bytes. */ | |
25fe55af | 1707 | int used; /* actually used size in words. */ |
b18215fc RS |
1708 | }; |
1709 | ||
1710 | /* Make sure that WORK_AREA can hold more N multibyte characters. */ | |
1711 | #define EXTEND_RANGE_TABLE_WORK_AREA(work_area, n) \ | |
1712 | do { \ | |
1713 | if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \ | |
1714 | { \ | |
1715 | (work_area).allocated += 16 * sizeof (int); \ | |
1716 | if ((work_area).table) \ | |
1717 | (work_area).table \ | |
1718 | = (int *) realloc ((work_area).table, (work_area).allocated); \ | |
1719 | else \ | |
1720 | (work_area).table \ | |
1721 | = (int *) malloc ((work_area).allocated); \ | |
1722 | if ((work_area).table == 0) \ | |
1723 | FREE_STACK_RETURN (REG_ESPACE); \ | |
1724 | } \ | |
1725 | } while (0) | |
1726 | ||
1727 | /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */ | |
1728 | #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \ | |
1729 | do { \ | |
1730 | EXTEND_RANGE_TABLE_WORK_AREA ((work_area), 2); \ | |
1731 | (work_area).table[(work_area).used++] = (range_start); \ | |
1732 | (work_area).table[(work_area).used++] = (range_end); \ | |
1733 | } while (0) | |
1734 | ||
25fe55af | 1735 | /* Free allocated memory for WORK_AREA. */ |
b18215fc RS |
1736 | #define FREE_RANGE_TABLE_WORK_AREA(work_area) \ |
1737 | do { \ | |
1738 | if ((work_area).table) \ | |
1739 | free ((work_area).table); \ | |
1740 | } while (0) | |
1741 | ||
1742 | #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0) | |
1743 | #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used) | |
1744 | #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i]) | |
1745 | ||
1746 | ||
fa9a63c5 | 1747 | /* Set the bit for character C in a list. */ |
25fe55af RS |
1748 | #define SET_LIST_BIT(c) \ |
1749 | (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
fa9a63c5 RM |
1750 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) |
1751 | ||
1752 | ||
1753 | /* Get the next unsigned number in the uncompiled pattern. */ | |
25fe55af | 1754 | #define GET_UNSIGNED_NUMBER(num) \ |
fa9a63c5 RM |
1755 | { if (p != pend) \ |
1756 | { \ | |
25fe55af RS |
1757 | PATFETCH (c); \ |
1758 | while (ISDIGIT (c)) \ | |
1759 | { \ | |
1760 | if (num < 0) \ | |
1761 | num = 0; \ | |
1762 | num = num * 10 + c - '0'; \ | |
1763 | if (p == pend) \ | |
1764 | break; \ | |
1765 | PATFETCH (c); \ | |
1766 | } \ | |
1767 | } \ | |
5e69f11e | 1768 | } |
fa9a63c5 RM |
1769 | |
1770 | #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
1771 | ||
1772 | #define IS_CHAR_CLASS(string) \ | |
1773 | (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
1774 | || STREQ (string, "lower") || STREQ (string, "digit") \ | |
1775 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
1776 | || STREQ (string, "space") || STREQ (string, "print") \ | |
1777 | || STREQ (string, "punct") || STREQ (string, "graph") \ | |
1778 | || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
1779 | \f | |
1780 | #ifndef MATCH_MAY_ALLOCATE | |
1781 | ||
1782 | /* If we cannot allocate large objects within re_match_2_internal, | |
1783 | we make the fail stack and register vectors global. | |
1784 | The fail stack, we grow to the maximum size when a regexp | |
1785 | is compiled. | |
1786 | The register vectors, we adjust in size each time we | |
1787 | compile a regexp, according to the number of registers it needs. */ | |
1788 | ||
1789 | static fail_stack_type fail_stack; | |
1790 | ||
1791 | /* Size with which the following vectors are currently allocated. | |
1792 | That is so we can make them bigger as needed, | |
25fe55af | 1793 | but never make them smaller. */ |
fa9a63c5 RM |
1794 | static int regs_allocated_size; |
1795 | ||
25fe55af | 1796 | static const char ** regstart, ** regend; |
fa9a63c5 RM |
1797 | static const char ** old_regstart, ** old_regend; |
1798 | static const char **best_regstart, **best_regend; | |
5e69f11e | 1799 | static register_info_type *reg_info; |
fa9a63c5 RM |
1800 | static const char **reg_dummy; |
1801 | static register_info_type *reg_info_dummy; | |
1802 | ||
1803 | /* Make the register vectors big enough for NUM_REGS registers, | |
25fe55af | 1804 | but don't make them smaller. */ |
fa9a63c5 RM |
1805 | |
1806 | static | |
1807 | regex_grow_registers (num_regs) | |
1808 | int num_regs; | |
1809 | { | |
1810 | if (num_regs > regs_allocated_size) | |
1811 | { | |
1812 | RETALLOC_IF (regstart, num_regs, const char *); | |
1813 | RETALLOC_IF (regend, num_regs, const char *); | |
1814 | RETALLOC_IF (old_regstart, num_regs, const char *); | |
1815 | RETALLOC_IF (old_regend, num_regs, const char *); | |
1816 | RETALLOC_IF (best_regstart, num_regs, const char *); | |
1817 | RETALLOC_IF (best_regend, num_regs, const char *); | |
1818 | RETALLOC_IF (reg_info, num_regs, register_info_type); | |
1819 | RETALLOC_IF (reg_dummy, num_regs, const char *); | |
1820 | RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); | |
1821 | ||
1822 | regs_allocated_size = num_regs; | |
1823 | } | |
1824 | } | |
1825 | ||
1826 | #endif /* not MATCH_MAY_ALLOCATE */ | |
1827 | \f | |
1828 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
1829 | Returns one of error codes defined in `regex.h', or zero for success. | |
1830 | ||
1831 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
1832 | fields are set in BUFP on entry. | |
1833 | ||
1834 | If it succeeds, results are put in BUFP (if it returns an error, the | |
1835 | contents of BUFP are undefined): | |
1836 | `buffer' is the compiled pattern; | |
1837 | `syntax' is set to SYNTAX; | |
1838 | `used' is set to the length of the compiled pattern; | |
1839 | `fastmap_accurate' is zero; | |
1840 | `re_nsub' is the number of subexpressions in PATTERN; | |
1841 | `not_bol' and `not_eol' are zero; | |
5e69f11e | 1842 | |
fa9a63c5 RM |
1843 | The `fastmap' and `newline_anchor' fields are neither |
1844 | examined nor set. */ | |
1845 | ||
1846 | /* Return, freeing storage we allocated. */ | |
1847 | #define FREE_STACK_RETURN(value) \ | |
b18215fc RS |
1848 | do { \ |
1849 | FREE_RANGE_TABLE_WORK_AREA (range_table_work); \ | |
1850 | free (compile_stack.stack); \ | |
1851 | return value; \ | |
1852 | } while (0) | |
fa9a63c5 RM |
1853 | |
1854 | static reg_errcode_t | |
1855 | regex_compile (pattern, size, syntax, bufp) | |
1856 | const char *pattern; | |
1857 | int size; | |
1858 | reg_syntax_t syntax; | |
1859 | struct re_pattern_buffer *bufp; | |
1860 | { | |
1861 | /* We fetch characters from PATTERN here. Even though PATTERN is | |
1862 | `char *' (i.e., signed), we declare these variables as unsigned, so | |
1863 | they can be reliably used as array indices. */ | |
b18215fc | 1864 | register unsigned int c, c1; |
5e69f11e | 1865 | |
fa9a63c5 RM |
1866 | /* A random temporary spot in PATTERN. */ |
1867 | const char *p1; | |
1868 | ||
1869 | /* Points to the end of the buffer, where we should append. */ | |
1870 | register unsigned char *b; | |
5e69f11e | 1871 | |
fa9a63c5 RM |
1872 | /* Keeps track of unclosed groups. */ |
1873 | compile_stack_type compile_stack; | |
1874 | ||
1875 | /* Points to the current (ending) position in the pattern. */ | |
1876 | const char *p = pattern; | |
1877 | const char *pend = pattern + size; | |
5e69f11e | 1878 | |
fa9a63c5 | 1879 | /* How to translate the characters in the pattern. */ |
6676cb1c | 1880 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
1881 | |
1882 | /* Address of the count-byte of the most recently inserted `exactn' | |
1883 | command. This makes it possible to tell if a new exact-match | |
1884 | character can be added to that command or if the character requires | |
1885 | a new `exactn' command. */ | |
1886 | unsigned char *pending_exact = 0; | |
1887 | ||
1888 | /* Address of start of the most recently finished expression. | |
1889 | This tells, e.g., postfix * where to find the start of its | |
1890 | operand. Reset at the beginning of groups and alternatives. */ | |
1891 | unsigned char *laststart = 0; | |
1892 | ||
1893 | /* Address of beginning of regexp, or inside of last group. */ | |
1894 | unsigned char *begalt; | |
1895 | ||
1896 | /* Place in the uncompiled pattern (i.e., the {) to | |
1897 | which to go back if the interval is invalid. */ | |
1898 | const char *beg_interval; | |
5e69f11e | 1899 | |
fa9a63c5 | 1900 | /* Address of the place where a forward jump should go to the end of |
25fe55af | 1901 | the containing expression. Each alternative of an `or' -- except the |
fa9a63c5 RM |
1902 | last -- ends with a forward jump of this sort. */ |
1903 | unsigned char *fixup_alt_jump = 0; | |
1904 | ||
1905 | /* Counts open-groups as they are encountered. Remembered for the | |
1906 | matching close-group on the compile stack, so the same register | |
1907 | number is put in the stop_memory as the start_memory. */ | |
1908 | regnum_t regnum = 0; | |
1909 | ||
b18215fc RS |
1910 | /* Work area for range table of charset. */ |
1911 | struct range_table_work_area range_table_work; | |
1912 | ||
fa9a63c5 RM |
1913 | #ifdef DEBUG |
1914 | DEBUG_PRINT1 ("\nCompiling pattern: "); | |
1915 | if (debug) | |
1916 | { | |
1917 | unsigned debug_count; | |
5e69f11e | 1918 | |
fa9a63c5 | 1919 | for (debug_count = 0; debug_count < size; debug_count++) |
25fe55af | 1920 | putchar (pattern[debug_count]); |
fa9a63c5 RM |
1921 | putchar ('\n'); |
1922 | } | |
1923 | #endif /* DEBUG */ | |
1924 | ||
1925 | /* Initialize the compile stack. */ | |
1926 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
1927 | if (compile_stack.stack == NULL) | |
1928 | return REG_ESPACE; | |
1929 | ||
1930 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
1931 | compile_stack.avail = 0; | |
1932 | ||
b18215fc RS |
1933 | range_table_work.table = 0; |
1934 | range_table_work.allocated = 0; | |
1935 | ||
fa9a63c5 RM |
1936 | /* Initialize the pattern buffer. */ |
1937 | bufp->syntax = syntax; | |
1938 | bufp->fastmap_accurate = 0; | |
1939 | bufp->not_bol = bufp->not_eol = 0; | |
1940 | ||
1941 | /* Set `used' to zero, so that if we return an error, the pattern | |
1942 | printer (for debugging) will think there's no pattern. We reset it | |
1943 | at the end. */ | |
1944 | bufp->used = 0; | |
5e69f11e | 1945 | |
fa9a63c5 | 1946 | /* Always count groups, whether or not bufp->no_sub is set. */ |
5e69f11e | 1947 | bufp->re_nsub = 0; |
fa9a63c5 | 1948 | |
b18215fc RS |
1949 | #ifdef emacs |
1950 | /* bufp->multibyte is set before regex_compile is called, so don't alter | |
1951 | it. */ | |
1952 | #else /* not emacs */ | |
1953 | /* Nothing is recognized as a multibyte character. */ | |
1954 | bufp->multibyte = 0; | |
1955 | #endif | |
1956 | ||
fa9a63c5 RM |
1957 | #if !defined (emacs) && !defined (SYNTAX_TABLE) |
1958 | /* Initialize the syntax table. */ | |
1959 | init_syntax_once (); | |
1960 | #endif | |
1961 | ||
1962 | if (bufp->allocated == 0) | |
1963 | { | |
1964 | if (bufp->buffer) | |
1965 | { /* If zero allocated, but buffer is non-null, try to realloc | |
25fe55af RS |
1966 | enough space. This loses if buffer's address is bogus, but |
1967 | that is the user's responsibility. */ | |
1968 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
1969 | } | |
fa9a63c5 | 1970 | else |
25fe55af RS |
1971 | { /* Caller did not allocate a buffer. Do it for them. */ |
1972 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
1973 | } | |
fa9a63c5 RM |
1974 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
1975 | ||
1976 | bufp->allocated = INIT_BUF_SIZE; | |
1977 | } | |
1978 | ||
1979 | begalt = b = bufp->buffer; | |
1980 | ||
1981 | /* Loop through the uncompiled pattern until we're at the end. */ | |
1982 | while (p != pend) | |
1983 | { | |
1984 | PATFETCH (c); | |
1985 | ||
1986 | switch (c) | |
25fe55af RS |
1987 | { |
1988 | case '^': | |
1989 | { | |
1990 | if ( /* If at start of pattern, it's an operator. */ | |
1991 | p == pattern + 1 | |
1992 | /* If context independent, it's an operator. */ | |
1993 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1994 | /* Otherwise, depends on what's come before. */ | |
1995 | || at_begline_loc_p (pattern, p, syntax)) | |
1996 | BUF_PUSH (begline); | |
1997 | else | |
1998 | goto normal_char; | |
1999 | } | |
2000 | break; | |
2001 | ||
2002 | ||
2003 | case '$': | |
2004 | { | |
2005 | if ( /* If at end of pattern, it's an operator. */ | |
2006 | p == pend | |
2007 | /* If context independent, it's an operator. */ | |
2008 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2009 | /* Otherwise, depends on what's next. */ | |
2010 | || at_endline_loc_p (p, pend, syntax)) | |
2011 | BUF_PUSH (endline); | |
2012 | else | |
2013 | goto normal_char; | |
2014 | } | |
2015 | break; | |
fa9a63c5 RM |
2016 | |
2017 | ||
2018 | case '+': | |
25fe55af RS |
2019 | case '?': |
2020 | if ((syntax & RE_BK_PLUS_QM) | |
2021 | || (syntax & RE_LIMITED_OPS)) | |
2022 | goto normal_char; | |
2023 | handle_plus: | |
2024 | case '*': | |
2025 | /* If there is no previous pattern... */ | |
2026 | if (!laststart) | |
2027 | { | |
2028 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2029 | FREE_STACK_RETURN (REG_BADRPT); | |
2030 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2031 | goto normal_char; | |
2032 | } | |
2033 | ||
2034 | { | |
2035 | /* Are we optimizing this jump? */ | |
2036 | boolean keep_string_p = false; | |
2037 | ||
2038 | /* 1 means zero (many) matches is allowed. */ | |
2039 | char zero_times_ok = 0, many_times_ok = 0; | |
2040 | ||
2041 | /* If there is a sequence of repetition chars, collapse it | |
2042 | down to just one (the right one). We can't combine | |
2043 | interval operators with these because of, e.g., `a{2}*', | |
2044 | which should only match an even number of `a's. */ | |
2045 | ||
2046 | for (;;) | |
2047 | { | |
2048 | zero_times_ok |= c != '+'; | |
2049 | many_times_ok |= c != '?'; | |
2050 | ||
2051 | if (p == pend) | |
2052 | break; | |
2053 | ||
2054 | PATFETCH (c); | |
2055 | ||
2056 | if (c == '*' | |
2057 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
2058 | ; | |
2059 | ||
2060 | else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
2061 | { | |
2062 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2063 | ||
2064 | PATFETCH (c1); | |
2065 | if (!(c1 == '+' || c1 == '?')) | |
2066 | { | |
2067 | PATUNFETCH; | |
2068 | PATUNFETCH; | |
2069 | break; | |
2070 | } | |
2071 | ||
2072 | c = c1; | |
2073 | } | |
2074 | else | |
2075 | { | |
2076 | PATUNFETCH; | |
2077 | break; | |
2078 | } | |
2079 | ||
2080 | /* If we get here, we found another repeat character. */ | |
2081 | } | |
2082 | ||
2083 | /* Star, etc. applied to an empty pattern is equivalent | |
2084 | to an empty pattern. */ | |
2085 | if (!laststart) | |
2086 | break; | |
2087 | ||
2088 | /* Now we know whether or not zero matches is allowed | |
2089 | and also whether or not two or more matches is allowed. */ | |
2090 | if (many_times_ok) | |
2091 | { /* More than one repetition is allowed, so put in at the | |
2092 | end a backward relative jump from `b' to before the next | |
2093 | jump we're going to put in below (which jumps from | |
2094 | laststart to after this jump). | |
2095 | ||
2096 | But if we are at the `*' in the exact sequence `.*\n', | |
2097 | insert an unconditional jump backwards to the ., | |
2098 | instead of the beginning of the loop. This way we only | |
2099 | push a failure point once, instead of every time | |
2100 | through the loop. */ | |
2101 | assert (p - 1 > pattern); | |
2102 | ||
2103 | /* Allocate the space for the jump. */ | |
2104 | GET_BUFFER_SPACE (3); | |
2105 | ||
2106 | /* We know we are not at the first character of the pattern, | |
2107 | because laststart was nonzero. And we've already | |
2108 | incremented `p', by the way, to be the character after | |
2109 | the `*'. Do we have to do something analogous here | |
2110 | for null bytes, because of RE_DOT_NOT_NULL? */ | |
e934739e | 2111 | if (TRANSLATE ((unsigned char)*(p - 2)) == TRANSLATE ('.') |
fa9a63c5 | 2112 | && zero_times_ok |
e934739e RS |
2113 | && p < pend |
2114 | && TRANSLATE ((unsigned char)*p) == TRANSLATE ('\n') | |
25fe55af RS |
2115 | && !(syntax & RE_DOT_NEWLINE)) |
2116 | { /* We have .*\n. */ | |
2117 | STORE_JUMP (jump, b, laststart); | |
2118 | keep_string_p = true; | |
2119 | } | |
2120 | else | |
2121 | /* Anything else. */ | |
2122 | STORE_JUMP (maybe_pop_jump, b, laststart - 3); | |
2123 | ||
2124 | /* We've added more stuff to the buffer. */ | |
2125 | b += 3; | |
2126 | } | |
2127 | ||
2128 | /* On failure, jump from laststart to b + 3, which will be the | |
2129 | end of the buffer after this jump is inserted. */ | |
2130 | GET_BUFFER_SPACE (3); | |
2131 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
2132 | : on_failure_jump, | |
2133 | laststart, b + 3); | |
2134 | pending_exact = 0; | |
2135 | b += 3; | |
2136 | ||
2137 | if (!zero_times_ok) | |
2138 | { | |
2139 | /* At least one repetition is required, so insert a | |
2140 | `dummy_failure_jump' before the initial | |
2141 | `on_failure_jump' instruction of the loop. This | |
2142 | effects a skip over that instruction the first time | |
2143 | we hit that loop. */ | |
2144 | GET_BUFFER_SPACE (3); | |
2145 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
2146 | b += 3; | |
2147 | } | |
2148 | } | |
fa9a63c5 RM |
2149 | break; |
2150 | ||
2151 | ||
2152 | case '.': | |
25fe55af RS |
2153 | laststart = b; |
2154 | BUF_PUSH (anychar); | |
2155 | break; | |
fa9a63c5 RM |
2156 | |
2157 | ||
25fe55af RS |
2158 | case '[': |
2159 | { | |
b18215fc | 2160 | CLEAR_RANGE_TABLE_WORK_USED (range_table_work); |
fa9a63c5 | 2161 | |
25fe55af | 2162 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2163 | |
25fe55af RS |
2164 | /* Ensure that we have enough space to push a charset: the |
2165 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
fa9a63c5 RM |
2166 | GET_BUFFER_SPACE (34); |
2167 | ||
25fe55af | 2168 | laststart = b; |
e318085a | 2169 | |
25fe55af RS |
2170 | /* We test `*p == '^' twice, instead of using an if |
2171 | statement, so we only need one BUF_PUSH. */ | |
2172 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
2173 | if (*p == '^') | |
2174 | p++; | |
e318085a | 2175 | |
25fe55af RS |
2176 | /* Remember the first position in the bracket expression. */ |
2177 | p1 = p; | |
e318085a | 2178 | |
25fe55af RS |
2179 | /* Push the number of bytes in the bitmap. */ |
2180 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2181 | |
25fe55af RS |
2182 | /* Clear the whole map. */ |
2183 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2184 | |
25fe55af RS |
2185 | /* charset_not matches newline according to a syntax bit. */ |
2186 | if ((re_opcode_t) b[-2] == charset_not | |
2187 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2188 | SET_LIST_BIT ('\n'); | |
fa9a63c5 | 2189 | |
25fe55af RS |
2190 | /* Read in characters and ranges, setting map bits. */ |
2191 | for (;;) | |
2192 | { | |
b18215fc RS |
2193 | int len; |
2194 | boolean escaped_char = false; | |
e318085a | 2195 | |
25fe55af | 2196 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
e318085a | 2197 | |
25fe55af | 2198 | PATFETCH (c); |
e318085a | 2199 | |
25fe55af RS |
2200 | /* \ might escape characters inside [...] and [^...]. */ |
2201 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2202 | { | |
2203 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
e318085a RS |
2204 | |
2205 | PATFETCH (c); | |
b18215fc | 2206 | escaped_char = true; |
25fe55af | 2207 | } |
b18215fc RS |
2208 | else |
2209 | { | |
657fcfbd RS |
2210 | /* Could be the end of the bracket expression. If it's |
2211 | not (i.e., when the bracket expression is `[]' so | |
2212 | far), the ']' character bit gets set way below. */ | |
2213 | if (c == ']' && p != p1 + 1) | |
2214 | break; | |
25fe55af | 2215 | } |
b18215fc RS |
2216 | |
2217 | /* If C indicates start of multibyte char, get the | |
2218 | actual character code in C, and set the pattern | |
2219 | pointer P to the next character boundary. */ | |
2220 | if (bufp->multibyte && BASE_LEADING_CODE_P (c)) | |
2221 | { | |
2222 | PATUNFETCH; | |
2223 | c = STRING_CHAR_AND_LENGTH (p, pend - p, len); | |
2224 | p += len; | |
25fe55af | 2225 | } |
b18215fc RS |
2226 | /* What should we do for the character which is |
2227 | greater than 0x7F, but not BASE_LEADING_CODE_P? | |
2228 | XXX */ | |
2229 | ||
25fe55af RS |
2230 | /* See if we're at the beginning of a possible character |
2231 | class. */ | |
b18215fc RS |
2232 | |
2233 | else if (!escaped_char && | |
2234 | syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
657fcfbd RS |
2235 | { |
2236 | /* Leave room for the null. */ | |
25fe55af | 2237 | char str[CHAR_CLASS_MAX_LENGTH + 1]; |
b18215fc | 2238 | |
25fe55af RS |
2239 | PATFETCH (c); |
2240 | c1 = 0; | |
b18215fc | 2241 | |
25fe55af RS |
2242 | /* If pattern is `[[:'. */ |
2243 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
b18215fc | 2244 | |
25fe55af RS |
2245 | for (;;) |
2246 | { | |
2247 | PATFETCH (c); | |
2248 | if (c == ':' || c == ']' || p == pend | |
2249 | || c1 == CHAR_CLASS_MAX_LENGTH) | |
2250 | break; | |
2251 | str[c1++] = c; | |
2252 | } | |
2253 | str[c1] = '\0'; | |
b18215fc RS |
2254 | |
2255 | /* If isn't a word bracketed by `[:' and `:]': | |
2256 | undo the ending character, the letters, and | |
2257 | leave the leading `:' and `[' (but set bits for | |
2258 | them). */ | |
25fe55af RS |
2259 | if (c == ':' && *p == ']') |
2260 | { | |
2261 | int ch; | |
2262 | boolean is_alnum = STREQ (str, "alnum"); | |
2263 | boolean is_alpha = STREQ (str, "alpha"); | |
2264 | boolean is_blank = STREQ (str, "blank"); | |
2265 | boolean is_cntrl = STREQ (str, "cntrl"); | |
2266 | boolean is_digit = STREQ (str, "digit"); | |
2267 | boolean is_graph = STREQ (str, "graph"); | |
2268 | boolean is_lower = STREQ (str, "lower"); | |
2269 | boolean is_print = STREQ (str, "print"); | |
2270 | boolean is_punct = STREQ (str, "punct"); | |
2271 | boolean is_space = STREQ (str, "space"); | |
2272 | boolean is_upper = STREQ (str, "upper"); | |
2273 | boolean is_xdigit = STREQ (str, "xdigit"); | |
2274 | ||
2275 | if (!IS_CHAR_CLASS (str)) | |
fa9a63c5 RM |
2276 | FREE_STACK_RETURN (REG_ECTYPE); |
2277 | ||
25fe55af RS |
2278 | /* Throw away the ] at the end of the character |
2279 | class. */ | |
2280 | PATFETCH (c); | |
fa9a63c5 | 2281 | |
25fe55af | 2282 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2283 | |
25fe55af RS |
2284 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) |
2285 | { | |
7ae68633 | 2286 | int translated = TRANSLATE (ch); |
fa9a63c5 RM |
2287 | /* This was split into 3 if's to |
2288 | avoid an arbitrary limit in some compiler. */ | |
25fe55af RS |
2289 | if ( (is_alnum && ISALNUM (ch)) |
2290 | || (is_alpha && ISALPHA (ch)) | |
2291 | || (is_blank && ISBLANK (ch)) | |
2292 | || (is_cntrl && ISCNTRL (ch))) | |
7ae68633 | 2293 | SET_LIST_BIT (translated); |
fa9a63c5 | 2294 | if ( (is_digit && ISDIGIT (ch)) |
25fe55af RS |
2295 | || (is_graph && ISGRAPH (ch)) |
2296 | || (is_lower && ISLOWER (ch)) | |
2297 | || (is_print && ISPRINT (ch))) | |
7ae68633 | 2298 | SET_LIST_BIT (translated); |
fa9a63c5 | 2299 | if ( (is_punct && ISPUNCT (ch)) |
25fe55af RS |
2300 | || (is_space && ISSPACE (ch)) |
2301 | || (is_upper && ISUPPER (ch)) | |
2302 | || (is_xdigit && ISXDIGIT (ch))) | |
7ae68633 | 2303 | SET_LIST_BIT (translated); |
25fe55af | 2304 | } |
b18215fc RS |
2305 | |
2306 | /* Repeat the loop. */ | |
2307 | continue; | |
25fe55af RS |
2308 | } |
2309 | else | |
2310 | { | |
2311 | c1++; | |
2312 | while (c1--) | |
2313 | PATUNFETCH; | |
2314 | SET_LIST_BIT ('['); | |
b18215fc RS |
2315 | |
2316 | /* Because the `:' may starts the range, we | |
2317 | can't simply set bit and repeat the loop. | |
25fe55af | 2318 | Instead, just set it to C and handle below. */ |
b18215fc | 2319 | c = ':'; |
25fe55af RS |
2320 | } |
2321 | } | |
b18215fc RS |
2322 | |
2323 | if (p < pend && p[0] == '-' && p[1] != ']') | |
2324 | { | |
2325 | ||
2326 | /* Discard the `-'. */ | |
2327 | PATFETCH (c1); | |
2328 | ||
2329 | /* Fetch the character which ends the range. */ | |
2330 | PATFETCH (c1); | |
2331 | if (bufp->multibyte && BASE_LEADING_CODE_P (c1)) | |
e318085a | 2332 | { |
b18215fc RS |
2333 | PATUNFETCH; |
2334 | c1 = STRING_CHAR_AND_LENGTH (p, pend - p, len); | |
2335 | p += len; | |
e318085a | 2336 | } |
b18215fc | 2337 | |
e934739e RS |
2338 | if (SINGLE_BYTE_CHAR_P (c) |
2339 | && ! SINGLE_BYTE_CHAR_P (c1)) | |
2340 | { | |
2341 | /* Handle a range such as \177-\377 in multibyte mode. | |
2342 | Split that into two ranges,, | |
2343 | the low one ending at 0237, and the high one | |
2344 | starting at ...040. */ | |
2345 | int c1_base = (c1 & ~0177) | 040; | |
2346 | SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1); | |
2347 | c1 = 0237; | |
2348 | } | |
2349 | else if (!SAME_CHARSET_P (c, c1)) | |
b18215fc | 2350 | FREE_STACK_RETURN (REG_ERANGE); |
e318085a | 2351 | } |
25fe55af | 2352 | else |
b18215fc RS |
2353 | /* Range from C to C. */ |
2354 | c1 = c; | |
2355 | ||
2356 | /* Set the range ... */ | |
2357 | if (SINGLE_BYTE_CHAR_P (c)) | |
2358 | /* ... into bitmap. */ | |
25fe55af | 2359 | { |
b18215fc RS |
2360 | unsigned this_char; |
2361 | int range_start = c, range_end = c1; | |
2362 | ||
2363 | /* If the start is after the end, the range is empty. */ | |
2364 | if (range_start > range_end) | |
2365 | { | |
2366 | if (syntax & RE_NO_EMPTY_RANGES) | |
2367 | FREE_STACK_RETURN (REG_ERANGE); | |
2368 | /* Else, repeat the loop. */ | |
2369 | } | |
2370 | else | |
2371 | { | |
2372 | for (this_char = range_start; this_char <= range_end; | |
2373 | this_char++) | |
2374 | SET_LIST_BIT (TRANSLATE (this_char)); | |
e934739e | 2375 | } |
25fe55af | 2376 | } |
e318085a | 2377 | else |
b18215fc RS |
2378 | /* ... into range table. */ |
2379 | SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1); | |
e318085a RS |
2380 | } |
2381 | ||
25fe55af RS |
2382 | /* Discard any (non)matching list bytes that are all 0 at the |
2383 | end of the map. Decrease the map-length byte too. */ | |
2384 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
2385 | b[-1]--; | |
2386 | b += b[-1]; | |
fa9a63c5 | 2387 | |
b18215fc RS |
2388 | /* Build real range table from work area. */ |
2389 | if (RANGE_TABLE_WORK_USED (range_table_work)) | |
2390 | { | |
2391 | int i; | |
2392 | int used = RANGE_TABLE_WORK_USED (range_table_work); | |
fa9a63c5 | 2393 | |
b18215fc | 2394 | /* Allocate space for COUNT + RANGE_TABLE. Needs two |
25fe55af | 2395 | bytes for COUNT and three bytes for each character. */ |
b18215fc | 2396 | GET_BUFFER_SPACE (2 + used * 3); |
fa9a63c5 | 2397 | |
b18215fc RS |
2398 | /* Indicate the existence of range table. */ |
2399 | laststart[1] |= 0x80; | |
fa9a63c5 | 2400 | |
b18215fc RS |
2401 | STORE_NUMBER_AND_INCR (b, used / 2); |
2402 | for (i = 0; i < used; i++) | |
2403 | STORE_CHARACTER_AND_INCR | |
2404 | (b, RANGE_TABLE_WORK_ELT (range_table_work, i)); | |
2405 | } | |
25fe55af RS |
2406 | } |
2407 | break; | |
fa9a63c5 RM |
2408 | |
2409 | ||
b18215fc | 2410 | case '(': |
25fe55af RS |
2411 | if (syntax & RE_NO_BK_PARENS) |
2412 | goto handle_open; | |
2413 | else | |
2414 | goto normal_char; | |
fa9a63c5 RM |
2415 | |
2416 | ||
25fe55af RS |
2417 | case ')': |
2418 | if (syntax & RE_NO_BK_PARENS) | |
2419 | goto handle_close; | |
2420 | else | |
2421 | goto normal_char; | |
e318085a RS |
2422 | |
2423 | ||
25fe55af RS |
2424 | case '\n': |
2425 | if (syntax & RE_NEWLINE_ALT) | |
2426 | goto handle_alt; | |
2427 | else | |
2428 | goto normal_char; | |
e318085a RS |
2429 | |
2430 | ||
b18215fc | 2431 | case '|': |
25fe55af RS |
2432 | if (syntax & RE_NO_BK_VBAR) |
2433 | goto handle_alt; | |
2434 | else | |
2435 | goto normal_char; | |
2436 | ||
2437 | ||
2438 | case '{': | |
2439 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
2440 | goto handle_interval; | |
2441 | else | |
2442 | goto normal_char; | |
2443 | ||
2444 | ||
2445 | case '\\': | |
2446 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2447 | ||
2448 | /* Do not translate the character after the \, so that we can | |
2449 | distinguish, e.g., \B from \b, even if we normally would | |
2450 | translate, e.g., B to b. */ | |
2451 | PATFETCH_RAW (c); | |
2452 | ||
2453 | switch (c) | |
2454 | { | |
2455 | case '(': | |
2456 | if (syntax & RE_NO_BK_PARENS) | |
2457 | goto normal_backslash; | |
2458 | ||
2459 | handle_open: | |
2460 | bufp->re_nsub++; | |
2461 | regnum++; | |
2462 | ||
2463 | if (COMPILE_STACK_FULL) | |
2464 | { | |
2465 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
2466 | compile_stack_elt_t); | |
2467 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
2468 | ||
2469 | compile_stack.size <<= 1; | |
2470 | } | |
2471 | ||
2472 | /* These are the values to restore when we hit end of this | |
2473 | group. They are all relative offsets, so that if the | |
2474 | whole pattern moves because of realloc, they will still | |
2475 | be valid. */ | |
2476 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
2477 | COMPILE_STACK_TOP.fixup_alt_jump | |
2478 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
2479 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
2480 | COMPILE_STACK_TOP.regnum = regnum; | |
2481 | ||
2482 | /* We will eventually replace the 0 with the number of | |
2483 | groups inner to this one. But do not push a | |
2484 | start_memory for groups beyond the last one we can | |
2485 | represent in the compiled pattern. */ | |
2486 | if (regnum <= MAX_REGNUM) | |
2487 | { | |
2488 | COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; | |
2489 | BUF_PUSH_3 (start_memory, regnum, 0); | |
2490 | } | |
2491 | ||
2492 | compile_stack.avail++; | |
2493 | ||
2494 | fixup_alt_jump = 0; | |
2495 | laststart = 0; | |
2496 | begalt = b; | |
b18215fc RS |
2497 | /* If we've reached MAX_REGNUM groups, then this open |
2498 | won't actually generate any code, so we'll have to | |
2499 | clear pending_exact explicitly. */ | |
2500 | pending_exact = 0; | |
25fe55af RS |
2501 | break; |
2502 | ||
2503 | ||
2504 | case ')': | |
2505 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
2506 | ||
2507 | if (COMPILE_STACK_EMPTY) | |
2508 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2509 | goto normal_backslash; | |
2510 | else | |
2511 | FREE_STACK_RETURN (REG_ERPAREN); | |
2512 | ||
2513 | handle_close: | |
2514 | if (fixup_alt_jump) | |
2515 | { /* Push a dummy failure point at the end of the | |
2516 | alternative for a possible future | |
2517 | `pop_failure_jump' to pop. See comments at | |
2518 | `push_dummy_failure' in `re_match_2'. */ | |
2519 | BUF_PUSH (push_dummy_failure); | |
2520 | ||
2521 | /* We allocated space for this jump when we assigned | |
2522 | to `fixup_alt_jump', in the `handle_alt' case below. */ | |
2523 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
2524 | } | |
2525 | ||
2526 | /* See similar code for backslashed left paren above. */ | |
2527 | if (COMPILE_STACK_EMPTY) | |
2528 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2529 | goto normal_char; | |
2530 | else | |
2531 | FREE_STACK_RETURN (REG_ERPAREN); | |
2532 | ||
2533 | /* Since we just checked for an empty stack above, this | |
2534 | ``can't happen''. */ | |
2535 | assert (compile_stack.avail != 0); | |
2536 | { | |
2537 | /* We don't just want to restore into `regnum', because | |
2538 | later groups should continue to be numbered higher, | |
2539 | as in `(ab)c(de)' -- the second group is #2. */ | |
2540 | regnum_t this_group_regnum; | |
2541 | ||
2542 | compile_stack.avail--; | |
2543 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
2544 | fixup_alt_jump | |
2545 | = COMPILE_STACK_TOP.fixup_alt_jump | |
2546 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
2547 | : 0; | |
2548 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
2549 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
b18215fc RS |
2550 | /* If we've reached MAX_REGNUM groups, then this open |
2551 | won't actually generate any code, so we'll have to | |
2552 | clear pending_exact explicitly. */ | |
2553 | pending_exact = 0; | |
e318085a | 2554 | |
25fe55af RS |
2555 | /* We're at the end of the group, so now we know how many |
2556 | groups were inside this one. */ | |
2557 | if (this_group_regnum <= MAX_REGNUM) | |
2558 | { | |
2559 | unsigned char *inner_group_loc | |
2560 | = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
2561 | ||
2562 | *inner_group_loc = regnum - this_group_regnum; | |
2563 | BUF_PUSH_3 (stop_memory, this_group_regnum, | |
2564 | regnum - this_group_regnum); | |
2565 | } | |
2566 | } | |
2567 | break; | |
2568 | ||
2569 | ||
2570 | case '|': /* `\|'. */ | |
2571 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
2572 | goto normal_backslash; | |
2573 | handle_alt: | |
2574 | if (syntax & RE_LIMITED_OPS) | |
2575 | goto normal_char; | |
2576 | ||
2577 | /* Insert before the previous alternative a jump which | |
2578 | jumps to this alternative if the former fails. */ | |
2579 | GET_BUFFER_SPACE (3); | |
2580 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
2581 | pending_exact = 0; | |
2582 | b += 3; | |
2583 | ||
2584 | /* The alternative before this one has a jump after it | |
2585 | which gets executed if it gets matched. Adjust that | |
2586 | jump so it will jump to this alternative's analogous | |
2587 | jump (put in below, which in turn will jump to the next | |
2588 | (if any) alternative's such jump, etc.). The last such | |
2589 | jump jumps to the correct final destination. A picture: | |
2590 | _____ _____ | |
2591 | | | | | | |
2592 | | v | v | |
2593 | a | b | c | |
2594 | ||
2595 | If we are at `b', then fixup_alt_jump right now points to a | |
2596 | three-byte space after `a'. We'll put in the jump, set | |
2597 | fixup_alt_jump to right after `b', and leave behind three | |
2598 | bytes which we'll fill in when we get to after `c'. */ | |
2599 | ||
2600 | if (fixup_alt_jump) | |
2601 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2602 | ||
2603 | /* Mark and leave space for a jump after this alternative, | |
2604 | to be filled in later either by next alternative or | |
2605 | when know we're at the end of a series of alternatives. */ | |
2606 | fixup_alt_jump = b; | |
2607 | GET_BUFFER_SPACE (3); | |
2608 | b += 3; | |
2609 | ||
2610 | laststart = 0; | |
2611 | begalt = b; | |
2612 | break; | |
2613 | ||
2614 | ||
2615 | case '{': | |
2616 | /* If \{ is a literal. */ | |
2617 | if (!(syntax & RE_INTERVALS) | |
2618 | /* If we're at `\{' and it's not the open-interval | |
2619 | operator. */ | |
2620 | || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
2621 | || (p - 2 == pattern && p == pend)) | |
2622 | goto normal_backslash; | |
2623 | ||
2624 | handle_interval: | |
2625 | { | |
2626 | /* If got here, then the syntax allows intervals. */ | |
2627 | ||
2628 | /* At least (most) this many matches must be made. */ | |
2629 | int lower_bound = -1, upper_bound = -1; | |
2630 | ||
2631 | beg_interval = p - 1; | |
2632 | ||
2633 | if (p == pend) | |
2634 | { | |
2635 | if (syntax & RE_NO_BK_BRACES) | |
2636 | goto unfetch_interval; | |
2637 | else | |
2638 | FREE_STACK_RETURN (REG_EBRACE); | |
2639 | } | |
2640 | ||
2641 | GET_UNSIGNED_NUMBER (lower_bound); | |
2642 | ||
2643 | if (c == ',') | |
2644 | { | |
2645 | GET_UNSIGNED_NUMBER (upper_bound); | |
2646 | if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
2647 | } | |
2648 | else | |
2649 | /* Interval such as `{1}' => match exactly once. */ | |
2650 | upper_bound = lower_bound; | |
2651 | ||
2652 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
2653 | || lower_bound > upper_bound) | |
2654 | { | |
2655 | if (syntax & RE_NO_BK_BRACES) | |
2656 | goto unfetch_interval; | |
2657 | else | |
2658 | FREE_STACK_RETURN (REG_BADBR); | |
2659 | } | |
2660 | ||
2661 | if (!(syntax & RE_NO_BK_BRACES)) | |
2662 | { | |
2663 | if (c != '\\') FREE_STACK_RETURN (REG_EBRACE); | |
2664 | ||
2665 | PATFETCH (c); | |
2666 | } | |
2667 | ||
2668 | if (c != '}') | |
2669 | { | |
2670 | if (syntax & RE_NO_BK_BRACES) | |
2671 | goto unfetch_interval; | |
2672 | else | |
2673 | FREE_STACK_RETURN (REG_BADBR); | |
2674 | } | |
2675 | ||
2676 | /* We just parsed a valid interval. */ | |
2677 | ||
2678 | /* If it's invalid to have no preceding re. */ | |
2679 | if (!laststart) | |
2680 | { | |
2681 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2682 | FREE_STACK_RETURN (REG_BADRPT); | |
2683 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
2684 | laststart = b; | |
2685 | else | |
2686 | goto unfetch_interval; | |
2687 | } | |
2688 | ||
2689 | /* If the upper bound is zero, don't want to succeed at | |
2690 | all; jump from `laststart' to `b + 3', which will be | |
2691 | the end of the buffer after we insert the jump. */ | |
2692 | if (upper_bound == 0) | |
2693 | { | |
2694 | GET_BUFFER_SPACE (3); | |
2695 | INSERT_JUMP (jump, laststart, b + 3); | |
2696 | b += 3; | |
2697 | } | |
2698 | ||
2699 | /* Otherwise, we have a nontrivial interval. When | |
2700 | we're all done, the pattern will look like: | |
2701 | set_number_at <jump count> <upper bound> | |
2702 | set_number_at <succeed_n count> <lower bound> | |
2703 | succeed_n <after jump addr> <succeed_n count> | |
2704 | <body of loop> | |
2705 | jump_n <succeed_n addr> <jump count> | |
2706 | (The upper bound and `jump_n' are omitted if | |
2707 | `upper_bound' is 1, though.) */ | |
2708 | else | |
2709 | { /* If the upper bound is > 1, we need to insert | |
2710 | more at the end of the loop. */ | |
2711 | unsigned nbytes = 10 + (upper_bound > 1) * 10; | |
2712 | ||
2713 | GET_BUFFER_SPACE (nbytes); | |
2714 | ||
2715 | /* Initialize lower bound of the `succeed_n', even | |
2716 | though it will be set during matching by its | |
2717 | attendant `set_number_at' (inserted next), | |
2718 | because `re_compile_fastmap' needs to know. | |
2719 | Jump to the `jump_n' we might insert below. */ | |
2720 | INSERT_JUMP2 (succeed_n, laststart, | |
2721 | b + 5 + (upper_bound > 1) * 5, | |
2722 | lower_bound); | |
2723 | b += 5; | |
2724 | ||
2725 | /* Code to initialize the lower bound. Insert | |
2726 | before the `succeed_n'. The `5' is the last two | |
2727 | bytes of this `set_number_at', plus 3 bytes of | |
2728 | the following `succeed_n'. */ | |
2729 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
2730 | b += 5; | |
2731 | ||
2732 | if (upper_bound > 1) | |
2733 | { /* More than one repetition is allowed, so | |
2734 | append a backward jump to the `succeed_n' | |
2735 | that starts this interval. | |
2736 | ||
2737 | When we've reached this during matching, | |
2738 | we'll have matched the interval once, so | |
2739 | jump back only `upper_bound - 1' times. */ | |
2740 | STORE_JUMP2 (jump_n, b, laststart + 5, | |
2741 | upper_bound - 1); | |
2742 | b += 5; | |
2743 | ||
2744 | /* The location we want to set is the second | |
2745 | parameter of the `jump_n'; that is `b-2' as | |
2746 | an absolute address. `laststart' will be | |
2747 | the `set_number_at' we're about to insert; | |
2748 | `laststart+3' the number to set, the source | |
2749 | for the relative address. But we are | |
2750 | inserting into the middle of the pattern -- | |
2751 | so everything is getting moved up by 5. | |
2752 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
2753 | i.e., b - laststart. | |
2754 | ||
2755 | We insert this at the beginning of the loop | |
2756 | so that if we fail during matching, we'll | |
2757 | reinitialize the bounds. */ | |
2758 | insert_op2 (set_number_at, laststart, b - laststart, | |
2759 | upper_bound - 1, b); | |
2760 | b += 5; | |
2761 | } | |
2762 | } | |
2763 | pending_exact = 0; | |
2764 | beg_interval = NULL; | |
2765 | } | |
2766 | break; | |
2767 | ||
2768 | unfetch_interval: | |
2769 | /* If an invalid interval, match the characters as literals. */ | |
2770 | assert (beg_interval); | |
2771 | p = beg_interval; | |
2772 | beg_interval = NULL; | |
2773 | ||
2774 | /* normal_char and normal_backslash need `c'. */ | |
2775 | PATFETCH (c); | |
2776 | ||
2777 | if (!(syntax & RE_NO_BK_BRACES)) | |
2778 | { | |
2779 | if (p > pattern && p[-1] == '\\') | |
2780 | goto normal_backslash; | |
2781 | } | |
2782 | goto normal_char; | |
e318085a | 2783 | |
b18215fc | 2784 | #ifdef emacs |
25fe55af RS |
2785 | /* There is no way to specify the before_dot and after_dot |
2786 | operators. rms says this is ok. --karl */ | |
2787 | case '=': | |
2788 | BUF_PUSH (at_dot); | |
2789 | break; | |
2790 | ||
2791 | case 's': | |
2792 | laststart = b; | |
2793 | PATFETCH (c); | |
2794 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
2795 | break; | |
2796 | ||
2797 | case 'S': | |
2798 | laststart = b; | |
2799 | PATFETCH (c); | |
2800 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
2801 | break; | |
b18215fc RS |
2802 | |
2803 | case 'c': | |
2804 | laststart = b; | |
2805 | PATFETCH_RAW (c); | |
2806 | BUF_PUSH_2 (categoryspec, c); | |
2807 | break; | |
e318085a | 2808 | |
b18215fc RS |
2809 | case 'C': |
2810 | laststart = b; | |
2811 | PATFETCH_RAW (c); | |
2812 | BUF_PUSH_2 (notcategoryspec, c); | |
2813 | break; | |
2814 | #endif /* emacs */ | |
e318085a | 2815 | |
e318085a | 2816 | |
25fe55af RS |
2817 | case 'w': |
2818 | laststart = b; | |
2819 | BUF_PUSH (wordchar); | |
2820 | break; | |
e318085a | 2821 | |
e318085a | 2822 | |
25fe55af RS |
2823 | case 'W': |
2824 | laststart = b; | |
2825 | BUF_PUSH (notwordchar); | |
2826 | break; | |
e318085a RS |
2827 | |
2828 | ||
25fe55af RS |
2829 | case '<': |
2830 | BUF_PUSH (wordbeg); | |
2831 | break; | |
e318085a | 2832 | |
25fe55af RS |
2833 | case '>': |
2834 | BUF_PUSH (wordend); | |
2835 | break; | |
e318085a | 2836 | |
25fe55af RS |
2837 | case 'b': |
2838 | BUF_PUSH (wordbound); | |
2839 | break; | |
e318085a | 2840 | |
25fe55af RS |
2841 | case 'B': |
2842 | BUF_PUSH (notwordbound); | |
2843 | break; | |
fa9a63c5 | 2844 | |
25fe55af RS |
2845 | case '`': |
2846 | BUF_PUSH (begbuf); | |
2847 | break; | |
e318085a | 2848 | |
25fe55af RS |
2849 | case '\'': |
2850 | BUF_PUSH (endbuf); | |
2851 | break; | |
e318085a | 2852 | |
25fe55af RS |
2853 | case '1': case '2': case '3': case '4': case '5': |
2854 | case '6': case '7': case '8': case '9': | |
2855 | if (syntax & RE_NO_BK_REFS) | |
2856 | goto normal_char; | |
e318085a | 2857 | |
25fe55af | 2858 | c1 = c - '0'; |
e318085a | 2859 | |
25fe55af RS |
2860 | if (c1 > regnum) |
2861 | FREE_STACK_RETURN (REG_ESUBREG); | |
e318085a | 2862 | |
25fe55af RS |
2863 | /* Can't back reference to a subexpression if inside of it. */ |
2864 | if (group_in_compile_stack (compile_stack, c1)) | |
2865 | goto normal_char; | |
e318085a | 2866 | |
25fe55af RS |
2867 | laststart = b; |
2868 | BUF_PUSH_2 (duplicate, c1); | |
2869 | break; | |
e318085a | 2870 | |
e318085a | 2871 | |
25fe55af RS |
2872 | case '+': |
2873 | case '?': | |
2874 | if (syntax & RE_BK_PLUS_QM) | |
2875 | goto handle_plus; | |
2876 | else | |
2877 | goto normal_backslash; | |
2878 | ||
2879 | default: | |
2880 | normal_backslash: | |
2881 | /* You might think it would be useful for \ to mean | |
2882 | not to translate; but if we don't translate it | |
2883 | it will never match anything. */ | |
2884 | c = TRANSLATE (c); | |
2885 | goto normal_char; | |
2886 | } | |
2887 | break; | |
fa9a63c5 RM |
2888 | |
2889 | ||
2890 | default: | |
25fe55af | 2891 | /* Expects the character in `c'. */ |
fa9a63c5 | 2892 | normal_char: |
b18215fc RS |
2893 | p1 = p - 1; /* P1 points the head of C. */ |
2894 | #ifdef emacs | |
2895 | if (bufp->multibyte) | |
2896 | /* Set P to the next character boundary. */ | |
2897 | p += MULTIBYTE_FORM_LENGTH (p1, pend - p1) - 1; | |
2898 | #endif | |
fa9a63c5 | 2899 | /* If no exactn currently being built. */ |
25fe55af | 2900 | if (!pending_exact |
fa9a63c5 | 2901 | |
25fe55af RS |
2902 | /* If last exactn not at current position. */ |
2903 | || pending_exact + *pending_exact + 1 != b | |
5e69f11e | 2904 | |
25fe55af | 2905 | /* We have only one byte following the exactn for the count. */ |
b18215fc | 2906 | || *pending_exact >= (1 << BYTEWIDTH) - (p - p1) |
fa9a63c5 | 2907 | |
25fe55af RS |
2908 | /* If followed by a repetition operator. */ |
2909 | || *p == '*' || *p == '^' | |
fa9a63c5 RM |
2910 | || ((syntax & RE_BK_PLUS_QM) |
2911 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
2912 | : (*p == '+' || *p == '?')) | |
2913 | || ((syntax & RE_INTERVALS) | |
25fe55af | 2914 | && ((syntax & RE_NO_BK_BRACES) |
fa9a63c5 | 2915 | ? *p == '{' |
25fe55af | 2916 | : (p[0] == '\\' && p[1] == '{')))) |
fa9a63c5 RM |
2917 | { |
2918 | /* Start building a new exactn. */ | |
5e69f11e | 2919 | |
25fe55af | 2920 | laststart = b; |
fa9a63c5 RM |
2921 | |
2922 | BUF_PUSH_2 (exactn, 0); | |
2923 | pending_exact = b - 1; | |
25fe55af | 2924 | } |
5e69f11e | 2925 | |
b18215fc RS |
2926 | /* Here, C may translated, therefore C may not equal to *P1. */ |
2927 | while (1) | |
2928 | { | |
e934739e RS |
2929 | BUF_PUSH (c); |
2930 | (*pending_exact)++; | |
b18215fc RS |
2931 | if (++p1 == p) |
2932 | break; | |
2933 | ||
2934 | /* Rest of multibyte form should be copied literally. */ | |
2935 | c = *(unsigned char *)p1; | |
2936 | } | |
fa9a63c5 | 2937 | break; |
25fe55af | 2938 | } /* switch (c) */ |
fa9a63c5 RM |
2939 | } /* while p != pend */ |
2940 | ||
5e69f11e | 2941 | |
fa9a63c5 | 2942 | /* Through the pattern now. */ |
5e69f11e | 2943 | |
fa9a63c5 RM |
2944 | if (fixup_alt_jump) |
2945 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2946 | ||
5e69f11e | 2947 | if (!COMPILE_STACK_EMPTY) |
fa9a63c5 RM |
2948 | FREE_STACK_RETURN (REG_EPAREN); |
2949 | ||
2950 | /* If we don't want backtracking, force success | |
2951 | the first time we reach the end of the compiled pattern. */ | |
2952 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
2953 | BUF_PUSH (succeed); | |
2954 | ||
2955 | free (compile_stack.stack); | |
2956 | ||
2957 | /* We have succeeded; set the length of the buffer. */ | |
2958 | bufp->used = b - bufp->buffer; | |
2959 | ||
2960 | #ifdef DEBUG | |
2961 | if (debug) | |
2962 | { | |
2963 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); | |
2964 | print_compiled_pattern (bufp); | |
2965 | } | |
2966 | #endif /* DEBUG */ | |
2967 | ||
2968 | #ifndef MATCH_MAY_ALLOCATE | |
2969 | /* Initialize the failure stack to the largest possible stack. This | |
2970 | isn't necessary unless we're trying to avoid calling alloca in | |
2971 | the search and match routines. */ | |
2972 | { | |
2973 | int num_regs = bufp->re_nsub + 1; | |
2974 | ||
320a2a73 | 2975 | if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE) |
fa9a63c5 | 2976 | { |
a26f4ccd | 2977 | fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE; |
fa9a63c5 RM |
2978 | |
2979 | #ifdef emacs | |
2980 | if (! fail_stack.stack) | |
2981 | fail_stack.stack | |
5e69f11e | 2982 | = (fail_stack_elt_t *) xmalloc (fail_stack.size |
fa9a63c5 RM |
2983 | * sizeof (fail_stack_elt_t)); |
2984 | else | |
2985 | fail_stack.stack | |
2986 | = (fail_stack_elt_t *) xrealloc (fail_stack.stack, | |
2987 | (fail_stack.size | |
2988 | * sizeof (fail_stack_elt_t))); | |
2989 | #else /* not emacs */ | |
2990 | if (! fail_stack.stack) | |
2991 | fail_stack.stack | |
5e69f11e | 2992 | = (fail_stack_elt_t *) malloc (fail_stack.size |
fa9a63c5 RM |
2993 | * sizeof (fail_stack_elt_t)); |
2994 | else | |
2995 | fail_stack.stack | |
2996 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
2997 | (fail_stack.size | |
2998 | * sizeof (fail_stack_elt_t))); | |
2999 | #endif /* not emacs */ | |
3000 | } | |
3001 | ||
3002 | regex_grow_registers (num_regs); | |
3003 | } | |
3004 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3005 | ||
3006 | return REG_NOERROR; | |
3007 | } /* regex_compile */ | |
3008 | \f | |
3009 | /* Subroutines for `regex_compile'. */ | |
3010 | ||
25fe55af | 3011 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
fa9a63c5 RM |
3012 | |
3013 | static void | |
3014 | store_op1 (op, loc, arg) | |
3015 | re_opcode_t op; | |
3016 | unsigned char *loc; | |
3017 | int arg; | |
3018 | { | |
3019 | *loc = (unsigned char) op; | |
3020 | STORE_NUMBER (loc + 1, arg); | |
3021 | } | |
3022 | ||
3023 | ||
3024 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3025 | ||
3026 | static void | |
3027 | store_op2 (op, loc, arg1, arg2) | |
3028 | re_opcode_t op; | |
3029 | unsigned char *loc; | |
3030 | int arg1, arg2; | |
3031 | { | |
3032 | *loc = (unsigned char) op; | |
3033 | STORE_NUMBER (loc + 1, arg1); | |
3034 | STORE_NUMBER (loc + 3, arg2); | |
3035 | } | |
3036 | ||
3037 | ||
3038 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
3039 | for OP followed by two-byte integer parameter ARG. */ | |
3040 | ||
3041 | static void | |
3042 | insert_op1 (op, loc, arg, end) | |
3043 | re_opcode_t op; | |
3044 | unsigned char *loc; | |
3045 | int arg; | |
5e69f11e | 3046 | unsigned char *end; |
fa9a63c5 RM |
3047 | { |
3048 | register unsigned char *pfrom = end; | |
3049 | register unsigned char *pto = end + 3; | |
3050 | ||
3051 | while (pfrom != loc) | |
3052 | *--pto = *--pfrom; | |
5e69f11e | 3053 | |
fa9a63c5 RM |
3054 | store_op1 (op, loc, arg); |
3055 | } | |
3056 | ||
3057 | ||
3058 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3059 | ||
3060 | static void | |
3061 | insert_op2 (op, loc, arg1, arg2, end) | |
3062 | re_opcode_t op; | |
3063 | unsigned char *loc; | |
3064 | int arg1, arg2; | |
5e69f11e | 3065 | unsigned char *end; |
fa9a63c5 RM |
3066 | { |
3067 | register unsigned char *pfrom = end; | |
3068 | register unsigned char *pto = end + 5; | |
3069 | ||
3070 | while (pfrom != loc) | |
3071 | *--pto = *--pfrom; | |
5e69f11e | 3072 | |
fa9a63c5 RM |
3073 | store_op2 (op, loc, arg1, arg2); |
3074 | } | |
3075 | ||
3076 | ||
3077 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
3078 | after an alternative or a begin-subexpression. We assume there is at | |
3079 | least one character before the ^. */ | |
3080 | ||
3081 | static boolean | |
3082 | at_begline_loc_p (pattern, p, syntax) | |
3083 | const char *pattern, *p; | |
3084 | reg_syntax_t syntax; | |
3085 | { | |
3086 | const char *prev = p - 2; | |
3087 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
5e69f11e | 3088 | |
fa9a63c5 RM |
3089 | return |
3090 | /* After a subexpression? */ | |
3091 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
25fe55af | 3092 | /* After an alternative? */ |
fa9a63c5 RM |
3093 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); |
3094 | } | |
3095 | ||
3096 | ||
3097 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
3098 | at least one character after the $, i.e., `P < PEND'. */ | |
3099 | ||
3100 | static boolean | |
3101 | at_endline_loc_p (p, pend, syntax) | |
3102 | const char *p, *pend; | |
3103 | int syntax; | |
3104 | { | |
3105 | const char *next = p; | |
3106 | boolean next_backslash = *next == '\\'; | |
5bb52971 | 3107 | const char *next_next = p + 1 < pend ? p + 1 : 0; |
5e69f11e | 3108 | |
fa9a63c5 RM |
3109 | return |
3110 | /* Before a subexpression? */ | |
3111 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
25fe55af | 3112 | : next_backslash && next_next && *next_next == ')') |
fa9a63c5 RM |
3113 | /* Before an alternative? */ |
3114 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
25fe55af | 3115 | : next_backslash && next_next && *next_next == '|'); |
fa9a63c5 RM |
3116 | } |
3117 | ||
3118 | ||
5e69f11e | 3119 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
fa9a63c5 RM |
3120 | false if it's not. */ |
3121 | ||
3122 | static boolean | |
3123 | group_in_compile_stack (compile_stack, regnum) | |
3124 | compile_stack_type compile_stack; | |
3125 | regnum_t regnum; | |
3126 | { | |
3127 | int this_element; | |
3128 | ||
5e69f11e RM |
3129 | for (this_element = compile_stack.avail - 1; |
3130 | this_element >= 0; | |
fa9a63c5 RM |
3131 | this_element--) |
3132 | if (compile_stack.stack[this_element].regnum == regnum) | |
3133 | return true; | |
3134 | ||
3135 | return false; | |
3136 | } | |
fa9a63c5 RM |
3137 | \f |
3138 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
3139 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
3140 | characters can start a string that matches the pattern. This fastmap | |
3141 | is used by re_search to skip quickly over impossible starting points. | |
3142 | ||
3143 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
3144 | area as BUFP->fastmap. | |
5e69f11e | 3145 | |
fa9a63c5 RM |
3146 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in |
3147 | the pattern buffer. | |
3148 | ||
3149 | Returns 0 if we succeed, -2 if an internal error. */ | |
3150 | ||
3151 | int | |
3152 | re_compile_fastmap (bufp) | |
3153 | struct re_pattern_buffer *bufp; | |
3154 | { | |
b18215fc | 3155 | int i, j, k; |
fa9a63c5 RM |
3156 | #ifdef MATCH_MAY_ALLOCATE |
3157 | fail_stack_type fail_stack; | |
3158 | #endif | |
3159 | #ifndef REGEX_MALLOC | |
3160 | char *destination; | |
3161 | #endif | |
3162 | /* We don't push any register information onto the failure stack. */ | |
3163 | unsigned num_regs = 0; | |
5e69f11e | 3164 | |
fa9a63c5 RM |
3165 | register char *fastmap = bufp->fastmap; |
3166 | unsigned char *pattern = bufp->buffer; | |
3167 | unsigned long size = bufp->used; | |
3168 | unsigned char *p = pattern; | |
3169 | register unsigned char *pend = pattern + size; | |
3170 | ||
3171 | /* This holds the pointer to the failure stack, when | |
3172 | it is allocated relocatably. */ | |
3173 | fail_stack_elt_t *failure_stack_ptr; | |
3174 | ||
3175 | /* Assume that each path through the pattern can be null until | |
25fe55af | 3176 | proven otherwise. We set this false at the bottom of switch |
fa9a63c5 RM |
3177 | statement, to which we get only if a particular path doesn't |
3178 | match the empty string. */ | |
3179 | boolean path_can_be_null = true; | |
3180 | ||
3181 | /* We aren't doing a `succeed_n' to begin with. */ | |
3182 | boolean succeed_n_p = false; | |
3183 | ||
b18215fc | 3184 | /* If all elements for base leading-codes in fastmap is set, this |
25fe55af | 3185 | flag is set true. */ |
b18215fc RS |
3186 | boolean match_any_multibyte_characters = false; |
3187 | ||
3188 | /* Maximum code of simple (single byte) character. */ | |
3189 | int simple_char_max; | |
3190 | ||
fa9a63c5 | 3191 | assert (fastmap != NULL && p != NULL); |
5e69f11e | 3192 | |
fa9a63c5 | 3193 | INIT_FAIL_STACK (); |
25fe55af | 3194 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ |
fa9a63c5 RM |
3195 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ |
3196 | bufp->can_be_null = 0; | |
5e69f11e | 3197 | |
fa9a63c5 RM |
3198 | while (1) |
3199 | { | |
3200 | if (p == pend || *p == succeed) | |
3201 | { | |
3202 | /* We have reached the (effective) end of pattern. */ | |
3203 | if (!FAIL_STACK_EMPTY ()) | |
3204 | { | |
3205 | bufp->can_be_null |= path_can_be_null; | |
3206 | ||
3207 | /* Reset for next path. */ | |
3208 | path_can_be_null = true; | |
3209 | ||
3210 | p = fail_stack.stack[--fail_stack.avail].pointer; | |
3211 | ||
3212 | continue; | |
3213 | } | |
3214 | else | |
3215 | break; | |
3216 | } | |
3217 | ||
25fe55af | 3218 | /* We should never be about to go beyond the end of the pattern. */ |
fa9a63c5 | 3219 | assert (p < pend); |
5e69f11e | 3220 | |
fa9a63c5 RM |
3221 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
3222 | { | |
3223 | ||
25fe55af RS |
3224 | /* I guess the idea here is to simply not bother with a fastmap |
3225 | if a backreference is used, since it's too hard to figure out | |
3226 | the fastmap for the corresponding group. Setting | |
3227 | `can_be_null' stops `re_search_2' from using the fastmap, so | |
3228 | that is all we do. */ | |
fa9a63c5 RM |
3229 | case duplicate: |
3230 | bufp->can_be_null = 1; | |
25fe55af | 3231 | goto done; |
fa9a63c5 RM |
3232 | |
3233 | ||
3234 | /* Following are the cases which match a character. These end | |
25fe55af | 3235 | with `break'. */ |
fa9a63c5 RM |
3236 | |
3237 | case exactn: | |
25fe55af | 3238 | fastmap[p[1]] = 1; |
fa9a63c5 RM |
3239 | break; |
3240 | ||
3241 | ||
b18215fc | 3242 | #ifndef emacs |
25fe55af RS |
3243 | case charset: |
3244 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
fa9a63c5 | 3245 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) |
25fe55af | 3246 | fastmap[j] = 1; |
fa9a63c5 RM |
3247 | break; |
3248 | ||
3249 | ||
3250 | case charset_not: | |
3251 | /* Chars beyond end of map must be allowed. */ | |
3252 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
25fe55af | 3253 | fastmap[j] = 1; |
fa9a63c5 RM |
3254 | |
3255 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
3256 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
25fe55af RS |
3257 | fastmap[j] = 1; |
3258 | break; | |
fa9a63c5 RM |
3259 | |
3260 | ||
3261 | case wordchar: | |
3262 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
3263 | if (SYNTAX (j) == Sword) | |
3264 | fastmap[j] = 1; | |
3265 | break; | |
3266 | ||
3267 | ||
3268 | case notwordchar: | |
3269 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
3270 | if (SYNTAX (j) != Sword) | |
3271 | fastmap[j] = 1; | |
3272 | break; | |
b18215fc RS |
3273 | #else /* emacs */ |
3274 | case charset: | |
3275 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; | |
3276 | j >= 0; j--) | |
3277 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
3278 | fastmap[j] = 1; | |
3279 | ||
3280 | if (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) | |
3281 | && match_any_multibyte_characters == false) | |
3282 | { | |
3283 | /* Set fastmap[I] 1 where I is a base leading code of each | |
3284 | multibyte character in the range table. */ | |
3285 | int c, count; | |
3286 | ||
3287 | /* Make P points the range table. */ | |
3288 | p += CHARSET_BITMAP_SIZE (&p[-2]); | |
3289 | ||
3290 | /* Extract the number of ranges in range table into | |
25fe55af | 3291 | COUNT. */ |
b18215fc RS |
3292 | EXTRACT_NUMBER_AND_INCR (count, p); |
3293 | for (; count > 0; count--, p += 2 * 3) /* XXX */ | |
3294 | { | |
3295 | /* Extract the start of each range. */ | |
3296 | EXTRACT_CHARACTER (c, p); | |
3297 | j = CHAR_CHARSET (c); | |
3298 | fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1; | |
3299 | } | |
3300 | } | |
3301 | break; | |
fa9a63c5 RM |
3302 | |
3303 | ||
b18215fc RS |
3304 | case charset_not: |
3305 | /* Chars beyond end of map must be allowed. End of map is | |
3306 | `127' if bufp->multibyte is nonzero. */ | |
3307 | simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH); | |
3308 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH; | |
3309 | j < simple_char_max; j++) | |
3310 | fastmap[j] = 1; | |
3311 | ||
3312 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; | |
3313 | j >= 0; j--) | |
3314 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
3315 | fastmap[j] = 1; | |
3316 | ||
3317 | if (bufp->multibyte) | |
3318 | /* Any character set can possibly contain a character | |
3319 | which doesn't match the specified set of characters. */ | |
3320 | { | |
3321 | set_fastmap_for_multibyte_characters: | |
3322 | if (match_any_multibyte_characters == false) | |
3323 | { | |
3324 | for (j = 0x80; j < 0xA0; j++) /* XXX */ | |
3325 | if (BASE_LEADING_CODE_P (j)) | |
3326 | fastmap[j] = 1; | |
3327 | match_any_multibyte_characters = true; | |
3328 | } | |
3329 | } | |
3330 | break; | |
3331 | ||
3332 | ||
3333 | case wordchar: | |
3334 | simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH); | |
3335 | for (j = 0; j < simple_char_max; j++) | |
3336 | if (SYNTAX (j) == Sword) | |
3337 | fastmap[j] = 1; | |
3338 | ||
3339 | if (bufp->multibyte) | |
3340 | /* Any character set can possibly contain a character | |
25fe55af | 3341 | whose syntax is `Sword'. */ |
b18215fc RS |
3342 | goto set_fastmap_for_multibyte_characters; |
3343 | break; | |
3344 | ||
3345 | ||
3346 | case notwordchar: | |
3347 | simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH); | |
3348 | for (j = 0; j < simple_char_max; j++) | |
3349 | if (SYNTAX (j) != Sword) | |
3350 | fastmap[j] = 1; | |
3351 | ||
3352 | if (bufp->multibyte) | |
3353 | /* Any character set can possibly contain a character | |
3354 | whose syntax is not `Sword'. */ | |
3355 | goto set_fastmap_for_multibyte_characters; | |
3356 | break; | |
3357 | #endif | |
3358 | ||
25fe55af | 3359 | case anychar: |
fa9a63c5 RM |
3360 | { |
3361 | int fastmap_newline = fastmap['\n']; | |
3362 | ||
b18215fc RS |
3363 | /* `.' matches anything (but if bufp->multibyte is |
3364 | nonzero, matches `\000' .. `\127' and possible multibyte | |
3365 | character) ... */ | |
3366 | if (bufp->multibyte) | |
3367 | { | |
3368 | simple_char_max = 0x80; | |
3369 | ||
3370 | for (j = 0x80; j < 0xA0; j++) | |
3371 | if (BASE_LEADING_CODE_P (j)) | |
3372 | fastmap[j] = 1; | |
3373 | match_any_multibyte_characters = true; | |
3374 | } | |
3375 | else | |
3376 | simple_char_max = (1 << BYTEWIDTH); | |
3377 | ||
3378 | for (j = 0; j < simple_char_max; j++) | |
fa9a63c5 RM |
3379 | fastmap[j] = 1; |
3380 | ||
3381 | /* ... except perhaps newline. */ | |
3382 | if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
3383 | fastmap['\n'] = fastmap_newline; | |
3384 | ||
3385 | /* Return if we have already set `can_be_null'; if we have, | |
25fe55af | 3386 | then the fastmap is irrelevant. Something's wrong here. */ |
fa9a63c5 RM |
3387 | else if (bufp->can_be_null) |
3388 | goto done; | |
3389 | ||
3390 | /* Otherwise, have to check alternative paths. */ | |
3391 | break; | |
3392 | } | |
3393 | ||
3394 | #ifdef emacs | |
b18215fc RS |
3395 | case wordbound: |
3396 | case notwordbound: | |
3397 | case wordbeg: | |
3398 | case wordend: | |
3399 | case notsyntaxspec: | |
25fe55af | 3400 | case syntaxspec: |
b18215fc RS |
3401 | /* This match depends on text properties. These end with |
3402 | aborting optimizations. */ | |
3403 | bufp->can_be_null = 1; | |
25fe55af | 3404 | goto done; |
b18215fc | 3405 | #if 0 |
fa9a63c5 | 3406 | k = *p++; |
b18215fc RS |
3407 | simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH); |
3408 | for (j = 0; j < simple_char_max; j++) | |
fa9a63c5 RM |
3409 | if (SYNTAX (j) == (enum syntaxcode) k) |
3410 | fastmap[j] = 1; | |
fa9a63c5 | 3411 | |
b18215fc RS |
3412 | if (bufp->multibyte) |
3413 | /* Any character set can possibly contain a character | |
3414 | whose syntax is K. */ | |
3415 | goto set_fastmap_for_multibyte_characters; | |
3416 | break; | |
fa9a63c5 RM |
3417 | |
3418 | case notsyntaxspec: | |
3419 | k = *p++; | |
b18215fc RS |
3420 | simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH); |
3421 | for (j = 0; j < simple_char_max; j++) | |
fa9a63c5 RM |
3422 | if (SYNTAX (j) != (enum syntaxcode) k) |
3423 | fastmap[j] = 1; | |
b18215fc RS |
3424 | |
3425 | if (bufp->multibyte) | |
3426 | /* Any character set can possibly contain a character | |
3427 | whose syntax is not K. */ | |
3428 | goto set_fastmap_for_multibyte_characters; | |
3429 | break; | |
3430 | #endif | |
3431 | ||
3432 | ||
3433 | case categoryspec: | |
3434 | k = *p++; | |
3435 | simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH); | |
3436 | for (j = 0; j < simple_char_max; j++) | |
3437 | if (CHAR_HAS_CATEGORY (j, k)) | |
3438 | fastmap[j] = 1; | |
3439 | ||
3440 | if (bufp->multibyte) | |
3441 | /* Any character set can possibly contain a character | |
3442 | whose category is K. */ | |
3443 | goto set_fastmap_for_multibyte_characters; | |
fa9a63c5 RM |
3444 | break; |
3445 | ||
3446 | ||
b18215fc RS |
3447 | case notcategoryspec: |
3448 | k = *p++; | |
3449 | simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH); | |
3450 | for (j = 0; j < simple_char_max; j++) | |
3451 | if (!CHAR_HAS_CATEGORY (j, k)) | |
3452 | fastmap[j] = 1; | |
3453 | ||
3454 | if (bufp->multibyte) | |
3455 | /* Any character set can possibly contain a character | |
25fe55af | 3456 | whose category is not K. */ |
b18215fc RS |
3457 | goto set_fastmap_for_multibyte_characters; |
3458 | break; | |
3459 | ||
fa9a63c5 | 3460 | /* All cases after this match the empty string. These end with |
25fe55af | 3461 | `continue'. */ |
fa9a63c5 RM |
3462 | |
3463 | ||
3464 | case before_dot: | |
3465 | case at_dot: | |
3466 | case after_dot: | |
25fe55af | 3467 | continue; |
ae4788a8 | 3468 | #endif /* emacs */ |
fa9a63c5 RM |
3469 | |
3470 | ||
25fe55af RS |
3471 | case no_op: |
3472 | case begline: | |
3473 | case endline: | |
fa9a63c5 RM |
3474 | case begbuf: |
3475 | case endbuf: | |
b18215fc | 3476 | #ifndef emacs |
fa9a63c5 RM |
3477 | case wordbound: |
3478 | case notwordbound: | |
3479 | case wordbeg: | |
3480 | case wordend: | |
25fe55af RS |
3481 | #endif |
3482 | case push_dummy_failure: | |
3483 | continue; | |
fa9a63c5 RM |
3484 | |
3485 | ||
3486 | case jump_n: | |
25fe55af | 3487 | case pop_failure_jump: |
fa9a63c5 RM |
3488 | case maybe_pop_jump: |
3489 | case jump: | |
25fe55af | 3490 | case jump_past_alt: |
fa9a63c5 | 3491 | case dummy_failure_jump: |
25fe55af | 3492 | EXTRACT_NUMBER_AND_INCR (j, p); |
5e69f11e | 3493 | p += j; |
fa9a63c5 RM |
3494 | if (j > 0) |
3495 | continue; | |
5e69f11e | 3496 | |
25fe55af RS |
3497 | /* Jump backward implies we just went through the body of a |
3498 | loop and matched nothing. Opcode jumped to should be | |
3499 | `on_failure_jump' or `succeed_n'. Just treat it like an | |
3500 | ordinary jump. For a * loop, it has pushed its failure | |
3501 | point already; if so, discard that as redundant. */ | |
3502 | if ((re_opcode_t) *p != on_failure_jump | |
fa9a63c5 RM |
3503 | && (re_opcode_t) *p != succeed_n) |
3504 | continue; | |
3505 | ||
25fe55af RS |
3506 | p++; |
3507 | EXTRACT_NUMBER_AND_INCR (j, p); | |
3508 | p += j; | |
5e69f11e | 3509 | |
25fe55af RS |
3510 | /* If what's on the stack is where we are now, pop it. */ |
3511 | if (!FAIL_STACK_EMPTY () | |
fa9a63c5 | 3512 | && fail_stack.stack[fail_stack.avail - 1].pointer == p) |
25fe55af | 3513 | fail_stack.avail--; |
fa9a63c5 | 3514 | |
25fe55af | 3515 | continue; |
fa9a63c5 RM |
3516 | |
3517 | ||
25fe55af RS |
3518 | case on_failure_jump: |
3519 | case on_failure_keep_string_jump: | |
fa9a63c5 | 3520 | handle_on_failure_jump: |
25fe55af RS |
3521 | EXTRACT_NUMBER_AND_INCR (j, p); |
3522 | ||
3523 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
3524 | end of the pattern. We don't want to push such a point, | |
3525 | since when we restore it above, entering the switch will | |
3526 | increment `p' past the end of the pattern. We don't need | |
3527 | to push such a point since we obviously won't find any more | |
3528 | fastmap entries beyond `pend'. Such a pattern can match | |
3529 | the null string, though. */ | |
3530 | if (p + j < pend) | |
3531 | { | |
3532 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
fa9a63c5 RM |
3533 | { |
3534 | RESET_FAIL_STACK (); | |
3535 | return -2; | |
3536 | } | |
fa9a63c5 | 3537 | } |
25fe55af RS |
3538 | else |
3539 | bufp->can_be_null = 1; | |
fa9a63c5 | 3540 | |
25fe55af RS |
3541 | if (succeed_n_p) |
3542 | { | |
3543 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
3544 | succeed_n_p = false; | |
3545 | } | |
3546 | ||
3547 | continue; | |
fa9a63c5 RM |
3548 | |
3549 | ||
3550 | case succeed_n: | |
25fe55af RS |
3551 | /* Get to the number of times to succeed. */ |
3552 | p += 2; | |
fa9a63c5 | 3553 | |
25fe55af RS |
3554 | /* Increment p past the n for when k != 0. */ |
3555 | EXTRACT_NUMBER_AND_INCR (k, p); | |
3556 | if (k == 0) | |
fa9a63c5 | 3557 | { |
25fe55af RS |
3558 | p -= 4; |
3559 | succeed_n_p = true; /* Spaghetti code alert. */ | |
3560 | goto handle_on_failure_jump; | |
3561 | } | |
3562 | continue; | |
fa9a63c5 RM |
3563 | |
3564 | ||
3565 | case set_number_at: | |
25fe55af RS |
3566 | p += 4; |
3567 | continue; | |
fa9a63c5 RM |
3568 | |
3569 | ||
3570 | case start_memory: | |
25fe55af | 3571 | case stop_memory: |
fa9a63c5 RM |
3572 | p += 2; |
3573 | continue; | |
3574 | ||
3575 | ||
3576 | default: | |
25fe55af RS |
3577 | abort (); /* We have listed all the cases. */ |
3578 | } /* switch *p++ */ | |
fa9a63c5 RM |
3579 | |
3580 | /* Getting here means we have found the possible starting | |
25fe55af RS |
3581 | characters for one path of the pattern -- and that the empty |
3582 | string does not match. We need not follow this path further. | |
3583 | Instead, look at the next alternative (remembered on the | |
3584 | stack), or quit if no more. The test at the top of the loop | |
3585 | does these things. */ | |
fa9a63c5 RM |
3586 | path_can_be_null = false; |
3587 | p = pend; | |
3588 | } /* while p */ | |
3589 | ||
3590 | /* Set `can_be_null' for the last path (also the first path, if the | |
25fe55af | 3591 | pattern is empty). */ |
fa9a63c5 RM |
3592 | bufp->can_be_null |= path_can_be_null; |
3593 | ||
3594 | done: | |
3595 | RESET_FAIL_STACK (); | |
3596 | return 0; | |
3597 | } /* re_compile_fastmap */ | |
3598 | \f | |
3599 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
3600 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
3601 | this memory for recording register information. STARTS and ENDS | |
3602 | must be allocated using the malloc library routine, and must each | |
3603 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
3604 | ||
3605 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
3606 | register data. | |
3607 | ||
3608 | Unless this function is called, the first search or match using | |
3609 | PATTERN_BUFFER will allocate its own register data, without | |
3610 | freeing the old data. */ | |
3611 | ||
3612 | void | |
3613 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
3614 | struct re_pattern_buffer *bufp; | |
3615 | struct re_registers *regs; | |
3616 | unsigned num_regs; | |
3617 | regoff_t *starts, *ends; | |
3618 | { | |
3619 | if (num_regs) | |
3620 | { | |
3621 | bufp->regs_allocated = REGS_REALLOCATE; | |
3622 | regs->num_regs = num_regs; | |
3623 | regs->start = starts; | |
3624 | regs->end = ends; | |
3625 | } | |
3626 | else | |
3627 | { | |
3628 | bufp->regs_allocated = REGS_UNALLOCATED; | |
3629 | regs->num_regs = 0; | |
3630 | regs->start = regs->end = (regoff_t *) 0; | |
3631 | } | |
3632 | } | |
3633 | \f | |
25fe55af | 3634 | /* Searching routines. */ |
fa9a63c5 RM |
3635 | |
3636 | /* Like re_search_2, below, but only one string is specified, and | |
3637 | doesn't let you say where to stop matching. */ | |
3638 | ||
3639 | int | |
3640 | re_search (bufp, string, size, startpos, range, regs) | |
3641 | struct re_pattern_buffer *bufp; | |
3642 | const char *string; | |
3643 | int size, startpos, range; | |
3644 | struct re_registers *regs; | |
3645 | { | |
5e69f11e | 3646 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
fa9a63c5 RM |
3647 | regs, size); |
3648 | } | |
3649 | ||
b18215fc RS |
3650 | /* End address of virtual concatenation of string. */ |
3651 | #define STOP_ADDR_VSTRING(P) \ | |
3652 | (((P) >= size1 ? string2 + size2 : string1 + size1)) | |
3653 | ||
3654 | /* Address of POS in the concatenation of virtual string. */ | |
3655 | #define POS_ADDR_VSTRING(POS) \ | |
3656 | (((POS) >= size1 ? string2 - size1 : string1) + (POS)) | |
fa9a63c5 RM |
3657 | |
3658 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
3659 | virtual concatenation of STRING1 and STRING2, starting first at index | |
3660 | STARTPOS, then at STARTPOS + 1, and so on. | |
5e69f11e | 3661 | |
fa9a63c5 | 3662 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
5e69f11e | 3663 | |
fa9a63c5 RM |
3664 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
3665 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
3666 | RANGE. | |
5e69f11e | 3667 | |
fa9a63c5 RM |
3668 | In REGS, return the indices of the virtual concatenation of STRING1 |
3669 | and STRING2 that matched the entire BUFP->buffer and its contained | |
3670 | subexpressions. | |
5e69f11e | 3671 | |
fa9a63c5 RM |
3672 | Do not consider matching one past the index STOP in the virtual |
3673 | concatenation of STRING1 and STRING2. | |
3674 | ||
3675 | We return either the position in the strings at which the match was | |
3676 | found, -1 if no match, or -2 if error (such as failure | |
3677 | stack overflow). */ | |
3678 | ||
3679 | int | |
3680 | re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
3681 | struct re_pattern_buffer *bufp; | |
3682 | const char *string1, *string2; | |
3683 | int size1, size2; | |
3684 | int startpos; | |
3685 | int range; | |
3686 | struct re_registers *regs; | |
3687 | int stop; | |
3688 | { | |
3689 | int val; | |
3690 | register char *fastmap = bufp->fastmap; | |
6676cb1c | 3691 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
3692 | int total_size = size1 + size2; |
3693 | int endpos = startpos + range; | |
c8499ba5 | 3694 | int anchored_start = 0; |
fa9a63c5 | 3695 | |
25fe55af | 3696 | /* Nonzero if we have to concern multibyte character. */ |
b18215fc RS |
3697 | int multibyte = bufp->multibyte; |
3698 | ||
fa9a63c5 RM |
3699 | /* Check for out-of-range STARTPOS. */ |
3700 | if (startpos < 0 || startpos > total_size) | |
3701 | return -1; | |
5e69f11e | 3702 | |
fa9a63c5 | 3703 | /* Fix up RANGE if it might eventually take us outside |
34597fa9 | 3704 | the virtual concatenation of STRING1 and STRING2. |
5e69f11e | 3705 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
34597fa9 RS |
3706 | if (endpos < 0) |
3707 | range = 0 - startpos; | |
fa9a63c5 RM |
3708 | else if (endpos > total_size) |
3709 | range = total_size - startpos; | |
3710 | ||
3711 | /* If the search isn't to be a backwards one, don't waste time in a | |
3712 | search for a pattern that must be anchored. */ | |
3713 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) | |
3714 | { | |
3715 | if (startpos > 0) | |
3716 | return -1; | |
3717 | else | |
3718 | range = 1; | |
3719 | } | |
3720 | ||
ae4788a8 RS |
3721 | #ifdef emacs |
3722 | /* In a forward search for something that starts with \=. | |
3723 | don't keep searching past point. */ | |
3724 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
3725 | { | |
3726 | range = PT - startpos; | |
3727 | if (range <= 0) | |
3728 | return -1; | |
3729 | } | |
3730 | #endif /* emacs */ | |
3731 | ||
fa9a63c5 RM |
3732 | /* Update the fastmap now if not correct already. */ |
3733 | if (fastmap && !bufp->fastmap_accurate) | |
3734 | if (re_compile_fastmap (bufp) == -2) | |
3735 | return -2; | |
5e69f11e | 3736 | |
c8499ba5 RS |
3737 | /* See whether the pattern is anchored. */ |
3738 | if (bufp->buffer[0] == begline) | |
3739 | anchored_start = 1; | |
3740 | ||
b18215fc | 3741 | #ifdef emacs |
cc9b4df2 KH |
3742 | gl_state.object = re_match_object; |
3743 | { | |
3744 | int charpos | |
3745 | = SYNTAX_TABLE_BYTE_TO_CHAR (startpos > 0 ? startpos : startpos + 1); | |
3746 | ||
3747 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); | |
3748 | } | |
b18215fc RS |
3749 | #endif |
3750 | ||
fa9a63c5 RM |
3751 | /* Loop through the string, looking for a place to start matching. */ |
3752 | for (;;) | |
5e69f11e | 3753 | { |
c8499ba5 RS |
3754 | /* If the pattern is anchored, |
3755 | skip quickly past places we cannot match. | |
3756 | We don't bother to treat startpos == 0 specially | |
3757 | because that case doesn't repeat. */ | |
3758 | if (anchored_start && startpos > 0) | |
3759 | { | |
3760 | if (! (bufp->newline_anchor | |
3761 | && ((startpos <= size1 ? string1[startpos - 1] | |
3762 | : string2[startpos - size1 - 1]) | |
3763 | == '\n'))) | |
3764 | goto advance; | |
3765 | } | |
3766 | ||
fa9a63c5 | 3767 | /* If a fastmap is supplied, skip quickly over characters that |
25fe55af RS |
3768 | cannot be the start of a match. If the pattern can match the |
3769 | null string, however, we don't need to skip characters; we want | |
3770 | the first null string. */ | |
fa9a63c5 RM |
3771 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
3772 | { | |
e934739e RS |
3773 | register const char *d; |
3774 | register unsigned int buf_ch; | |
3775 | ||
3776 | d = POS_ADDR_VSTRING (startpos); | |
3777 | ||
25fe55af | 3778 | if (range > 0) /* Searching forwards. */ |
fa9a63c5 | 3779 | { |
fa9a63c5 RM |
3780 | register int lim = 0; |
3781 | int irange = range; | |
3782 | ||
25fe55af RS |
3783 | if (startpos < size1 && startpos + range >= size1) |
3784 | lim = range - (size1 - startpos); | |
fa9a63c5 | 3785 | |
25fe55af RS |
3786 | /* Written out as an if-else to avoid testing `translate' |
3787 | inside the loop. */ | |
fa9a63c5 | 3788 | if (translate) |
e934739e RS |
3789 | { |
3790 | if (multibyte) | |
3791 | while (range > lim) | |
3792 | { | |
3793 | int buf_charlen; | |
3794 | ||
3795 | buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim, | |
3796 | buf_charlen); | |
3797 | ||
3798 | buf_ch = RE_TRANSLATE (translate, buf_ch); | |
3799 | if (buf_ch >= 0400 | |
3800 | || fastmap[buf_ch]) | |
3801 | break; | |
3802 | ||
3803 | range -= buf_charlen; | |
3804 | d += buf_charlen; | |
3805 | } | |
3806 | else | |
3807 | while (range > lim | |
3808 | && !fastmap[(unsigned char) | |
3809 | RE_TRANSLATE (translate, (unsigned char) *d++)]) | |
3810 | range--; | |
3811 | } | |
fa9a63c5 | 3812 | else |
25fe55af RS |
3813 | while (range > lim && !fastmap[(unsigned char) *d++]) |
3814 | range--; | |
fa9a63c5 RM |
3815 | |
3816 | startpos += irange - range; | |
3817 | } | |
25fe55af | 3818 | else /* Searching backwards. */ |
fa9a63c5 | 3819 | { |
e934739e RS |
3820 | int room = (size1 == 0 || startpos >= size1 |
3821 | ? size2 + size1 - startpos | |
3822 | : size1 - startpos); | |
3823 | ||
3824 | buf_ch = STRING_CHAR (d, room); | |
3825 | if (translate) | |
3826 | buf_ch = RE_TRANSLATE (translate, buf_ch); | |
fa9a63c5 | 3827 | |
e934739e RS |
3828 | if (! (buf_ch >= 0400 |
3829 | || fastmap[buf_ch])) | |
fa9a63c5 RM |
3830 | goto advance; |
3831 | } | |
3832 | } | |
3833 | ||
3834 | /* If can't match the null string, and that's all we have left, fail. */ | |
3835 | if (range >= 0 && startpos == total_size && fastmap | |
25fe55af | 3836 | && !bufp->can_be_null) |
fa9a63c5 RM |
3837 | return -1; |
3838 | ||
3839 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
3840 | startpos, regs, stop); | |
3841 | #ifndef REGEX_MALLOC | |
3842 | #ifdef C_ALLOCA | |
3843 | alloca (0); | |
3844 | #endif | |
3845 | #endif | |
3846 | ||
3847 | if (val >= 0) | |
3848 | return startpos; | |
5e69f11e | 3849 | |
fa9a63c5 RM |
3850 | if (val == -2) |
3851 | return -2; | |
3852 | ||
3853 | advance: | |
5e69f11e | 3854 | if (!range) |
25fe55af | 3855 | break; |
5e69f11e | 3856 | else if (range > 0) |
25fe55af | 3857 | { |
b18215fc RS |
3858 | /* Update STARTPOS to the next character boundary. */ |
3859 | if (multibyte) | |
3860 | { | |
b560c397 RS |
3861 | const unsigned char *p |
3862 | = (const unsigned char *) POS_ADDR_VSTRING (startpos); | |
3863 | const unsigned char *pend | |
3864 | = (const unsigned char *) STOP_ADDR_VSTRING (startpos); | |
b18215fc RS |
3865 | int len = MULTIBYTE_FORM_LENGTH (p, pend - p); |
3866 | ||
3867 | range -= len; | |
3868 | if (range < 0) | |
3869 | break; | |
3870 | startpos += len; | |
3871 | } | |
3872 | else | |
3873 | { | |
b560c397 RS |
3874 | range--; |
3875 | startpos++; | |
3876 | } | |
e318085a | 3877 | } |
fa9a63c5 | 3878 | else |
25fe55af RS |
3879 | { |
3880 | range++; | |
3881 | startpos--; | |
b18215fc RS |
3882 | |
3883 | /* Update STARTPOS to the previous character boundary. */ | |
3884 | if (multibyte) | |
3885 | { | |
b560c397 RS |
3886 | const unsigned char *p |
3887 | = (const unsigned char *) POS_ADDR_VSTRING (startpos); | |
b18215fc RS |
3888 | int len = 0; |
3889 | ||
3890 | /* Find the head of multibyte form. */ | |
5d967c7a | 3891 | while (!CHAR_HEAD_P (*p)) |
b18215fc RS |
3892 | p--, len++; |
3893 | ||
3894 | /* Adjust it. */ | |
3895 | #if 0 /* XXX */ | |
3896 | if (MULTIBYTE_FORM_LENGTH (p, len + 1) != (len + 1)) | |
3897 | ; | |
3898 | else | |
3899 | #endif | |
3900 | { | |
3901 | range += len; | |
3902 | if (range > 0) | |
3903 | break; | |
3904 | ||
3905 | startpos -= len; | |
3906 | } | |
3907 | } | |
25fe55af | 3908 | } |
fa9a63c5 RM |
3909 | } |
3910 | return -1; | |
3911 | } /* re_search_2 */ | |
3912 | \f | |
3913 | /* Declarations and macros for re_match_2. */ | |
3914 | ||
3915 | static int bcmp_translate (); | |
3916 | static boolean alt_match_null_string_p (), | |
25fe55af RS |
3917 | common_op_match_null_string_p (), |
3918 | group_match_null_string_p (); | |
fa9a63c5 RM |
3919 | |
3920 | /* This converts PTR, a pointer into one of the search strings `string1' | |
3921 | and `string2' into an offset from the beginning of that string. */ | |
3922 | #define POINTER_TO_OFFSET(ptr) \ | |
3923 | (FIRST_STRING_P (ptr) \ | |
3924 | ? ((regoff_t) ((ptr) - string1)) \ | |
3925 | : ((regoff_t) ((ptr) - string2 + size1))) | |
3926 | ||
3927 | /* Macros for dealing with the split strings in re_match_2. */ | |
3928 | ||
3929 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
3930 | ||
3931 | /* Call before fetching a character with *d. This switches over to | |
3932 | string2 if necessary. */ | |
3933 | #define PREFETCH() \ | |
25fe55af | 3934 | while (d == dend) \ |
fa9a63c5 RM |
3935 | { \ |
3936 | /* End of string2 => fail. */ \ | |
25fe55af RS |
3937 | if (dend == end_match_2) \ |
3938 | goto fail; \ | |
3939 | /* End of string1 => advance to string2. */ \ | |
3940 | d = string2; \ | |
fa9a63c5 RM |
3941 | dend = end_match_2; \ |
3942 | } | |
3943 | ||
3944 | ||
3945 | /* Test if at very beginning or at very end of the virtual concatenation | |
25fe55af | 3946 | of `string1' and `string2'. If only one string, it's `string2'. */ |
fa9a63c5 | 3947 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
5e69f11e | 3948 | #define AT_STRINGS_END(d) ((d) == end2) |
fa9a63c5 RM |
3949 | |
3950 | ||
3951 | /* Test if D points to a character which is word-constituent. We have | |
3952 | two special cases to check for: if past the end of string1, look at | |
3953 | the first character in string2; and if before the beginning of | |
3954 | string2, look at the last character in string1. */ | |
3955 | #define WORDCHAR_P(d) \ | |
3956 | (SYNTAX ((d) == end1 ? *string2 \ | |
25fe55af | 3957 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
fa9a63c5 RM |
3958 | == Sword) |
3959 | ||
9121ca40 | 3960 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
b18215fc RS |
3961 | |
3962 | /* The comment at case wordbound is following one, but we don't use | |
3963 | AT_WORD_BOUNDARY anymore to support multibyte form. | |
3964 | ||
3965 | The DEC Alpha C compiler 3.x generates incorrect code for the | |
25fe55af RS |
3966 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
3967 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the | |
b18215fc RS |
3968 | macro and introducing temporary variables works around the bug. */ |
3969 | ||
9121ca40 | 3970 | #if 0 |
fa9a63c5 RM |
3971 | /* Test if the character before D and the one at D differ with respect |
3972 | to being word-constituent. */ | |
3973 | #define AT_WORD_BOUNDARY(d) \ | |
3974 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
3975 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
9121ca40 | 3976 | #endif |
fa9a63c5 RM |
3977 | |
3978 | /* Free everything we malloc. */ | |
3979 | #ifdef MATCH_MAY_ALLOCATE | |
00049484 | 3980 | #define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else |
fa9a63c5 RM |
3981 | #define FREE_VARIABLES() \ |
3982 | do { \ | |
3983 | REGEX_FREE_STACK (fail_stack.stack); \ | |
3984 | FREE_VAR (regstart); \ | |
3985 | FREE_VAR (regend); \ | |
3986 | FREE_VAR (old_regstart); \ | |
3987 | FREE_VAR (old_regend); \ | |
3988 | FREE_VAR (best_regstart); \ | |
3989 | FREE_VAR (best_regend); \ | |
3990 | FREE_VAR (reg_info); \ | |
3991 | FREE_VAR (reg_dummy); \ | |
3992 | FREE_VAR (reg_info_dummy); \ | |
3993 | } while (0) | |
3994 | #else | |
3995 | #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ | |
3996 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3997 | ||
25fe55af | 3998 | /* These values must meet several constraints. They must not be valid |
fa9a63c5 RM |
3999 | register values; since we have a limit of 255 registers (because |
4000 | we use only one byte in the pattern for the register number), we can | |
25fe55af | 4001 | use numbers larger than 255. They must differ by 1, because of |
fa9a63c5 RM |
4002 | NUM_FAILURE_ITEMS above. And the value for the lowest register must |
4003 | be larger than the value for the highest register, so we do not try | |
25fe55af | 4004 | to actually save any registers when none are active. */ |
fa9a63c5 RM |
4005 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) |
4006 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
4007 | \f | |
4008 | /* Matching routines. */ | |
4009 | ||
25fe55af | 4010 | #ifndef emacs /* Emacs never uses this. */ |
fa9a63c5 RM |
4011 | /* re_match is like re_match_2 except it takes only a single string. */ |
4012 | ||
4013 | int | |
4014 | re_match (bufp, string, size, pos, regs) | |
4015 | struct re_pattern_buffer *bufp; | |
4016 | const char *string; | |
4017 | int size, pos; | |
4018 | struct re_registers *regs; | |
4019 | { | |
4020 | int result = re_match_2_internal (bufp, NULL, 0, string, size, | |
4021 | pos, regs, size); | |
4022 | alloca (0); | |
4023 | return result; | |
4024 | } | |
4025 | #endif /* not emacs */ | |
4026 | ||
b18215fc RS |
4027 | #ifdef emacs |
4028 | /* In Emacs, this is the string or buffer in which we | |
25fe55af | 4029 | are matching. It is used for looking up syntax properties. */ |
b18215fc RS |
4030 | Lisp_Object re_match_object; |
4031 | #endif | |
fa9a63c5 RM |
4032 | |
4033 | /* re_match_2 matches the compiled pattern in BUFP against the | |
4034 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
4035 | and SIZE2, respectively). We start matching at POS, and stop | |
4036 | matching at STOP. | |
5e69f11e | 4037 | |
fa9a63c5 | 4038 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
25fe55af | 4039 | store offsets for the substring each group matched in REGS. See the |
fa9a63c5 RM |
4040 | documentation for exactly how many groups we fill. |
4041 | ||
4042 | We return -1 if no match, -2 if an internal error (such as the | |
25fe55af | 4043 | failure stack overflowing). Otherwise, we return the length of the |
fa9a63c5 RM |
4044 | matched substring. */ |
4045 | ||
4046 | int | |
4047 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4048 | struct re_pattern_buffer *bufp; | |
4049 | const char *string1, *string2; | |
4050 | int size1, size2; | |
4051 | int pos; | |
4052 | struct re_registers *regs; | |
4053 | int stop; | |
4054 | { | |
b18215fc | 4055 | int result; |
25fe55af | 4056 | |
b18215fc | 4057 | #ifdef emacs |
cc9b4df2 KH |
4058 | int charpos; |
4059 | gl_state.object = re_match_object; | |
4060 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos)); | |
4061 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); | |
b18215fc RS |
4062 | #endif |
4063 | ||
4064 | result = re_match_2_internal (bufp, string1, size1, string2, size2, | |
cc9b4df2 | 4065 | pos, regs, stop); |
fa9a63c5 RM |
4066 | alloca (0); |
4067 | return result; | |
4068 | } | |
4069 | ||
4070 | /* This is a separate function so that we can force an alloca cleanup | |
25fe55af | 4071 | afterwards. */ |
fa9a63c5 RM |
4072 | static int |
4073 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4074 | struct re_pattern_buffer *bufp; | |
4075 | const char *string1, *string2; | |
4076 | int size1, size2; | |
4077 | int pos; | |
4078 | struct re_registers *regs; | |
4079 | int stop; | |
4080 | { | |
4081 | /* General temporaries. */ | |
4082 | int mcnt; | |
4083 | unsigned char *p1; | |
4084 | ||
4085 | /* Just past the end of the corresponding string. */ | |
4086 | const char *end1, *end2; | |
4087 | ||
4088 | /* Pointers into string1 and string2, just past the last characters in | |
25fe55af | 4089 | each to consider matching. */ |
fa9a63c5 RM |
4090 | const char *end_match_1, *end_match_2; |
4091 | ||
4092 | /* Where we are in the data, and the end of the current string. */ | |
4093 | const char *d, *dend; | |
5e69f11e | 4094 | |
fa9a63c5 RM |
4095 | /* Where we are in the pattern, and the end of the pattern. */ |
4096 | unsigned char *p = bufp->buffer; | |
4097 | register unsigned char *pend = p + bufp->used; | |
4098 | ||
4099 | /* Mark the opcode just after a start_memory, so we can test for an | |
4100 | empty subpattern when we get to the stop_memory. */ | |
4101 | unsigned char *just_past_start_mem = 0; | |
4102 | ||
25fe55af | 4103 | /* We use this to map every character in the string. */ |
6676cb1c | 4104 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 | 4105 | |
25fe55af | 4106 | /* Nonzero if we have to concern multibyte character. */ |
b18215fc RS |
4107 | int multibyte = bufp->multibyte; |
4108 | ||
fa9a63c5 RM |
4109 | /* Failure point stack. Each place that can handle a failure further |
4110 | down the line pushes a failure point on this stack. It consists of | |
4111 | restart, regend, and reg_info for all registers corresponding to | |
4112 | the subexpressions we're currently inside, plus the number of such | |
4113 | registers, and, finally, two char *'s. The first char * is where | |
4114 | to resume scanning the pattern; the second one is where to resume | |
4115 | scanning the strings. If the latter is zero, the failure point is | |
4116 | a ``dummy''; if a failure happens and the failure point is a dummy, | |
25fe55af RS |
4117 | it gets discarded and the next next one is tried. */ |
4118 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
fa9a63c5 RM |
4119 | fail_stack_type fail_stack; |
4120 | #endif | |
4121 | #ifdef DEBUG | |
4122 | static unsigned failure_id = 0; | |
4123 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; | |
4124 | #endif | |
4125 | ||
4126 | /* This holds the pointer to the failure stack, when | |
4127 | it is allocated relocatably. */ | |
4128 | fail_stack_elt_t *failure_stack_ptr; | |
4129 | ||
4130 | /* We fill all the registers internally, independent of what we | |
25fe55af | 4131 | return, for use in backreferences. The number here includes |
fa9a63c5 RM |
4132 | an element for register zero. */ |
4133 | unsigned num_regs = bufp->re_nsub + 1; | |
5e69f11e | 4134 | |
fa9a63c5 RM |
4135 | /* The currently active registers. */ |
4136 | unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
4137 | unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
4138 | ||
4139 | /* Information on the contents of registers. These are pointers into | |
4140 | the input strings; they record just what was matched (on this | |
4141 | attempt) by a subexpression part of the pattern, that is, the | |
4142 | regnum-th regstart pointer points to where in the pattern we began | |
4143 | matching and the regnum-th regend points to right after where we | |
4144 | stopped matching the regnum-th subexpression. (The zeroth register | |
4145 | keeps track of what the whole pattern matches.) */ | |
4146 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
4147 | const char **regstart, **regend; | |
4148 | #endif | |
4149 | ||
4150 | /* If a group that's operated upon by a repetition operator fails to | |
4151 | match anything, then the register for its start will need to be | |
4152 | restored because it will have been set to wherever in the string we | |
4153 | are when we last see its open-group operator. Similarly for a | |
4154 | register's end. */ | |
4155 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
4156 | const char **old_regstart, **old_regend; | |
4157 | #endif | |
4158 | ||
4159 | /* The is_active field of reg_info helps us keep track of which (possibly | |
4160 | nested) subexpressions we are currently in. The matched_something | |
4161 | field of reg_info[reg_num] helps us tell whether or not we have | |
4162 | matched any of the pattern so far this time through the reg_num-th | |
4163 | subexpression. These two fields get reset each time through any | |
25fe55af RS |
4164 | loop their register is in. */ |
4165 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
5e69f11e | 4166 | register_info_type *reg_info; |
fa9a63c5 RM |
4167 | #endif |
4168 | ||
4169 | /* The following record the register info as found in the above | |
5e69f11e | 4170 | variables when we find a match better than any we've seen before. |
fa9a63c5 RM |
4171 | This happens as we backtrack through the failure points, which in |
4172 | turn happens only if we have not yet matched the entire string. */ | |
4173 | unsigned best_regs_set = false; | |
4174 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
4175 | const char **best_regstart, **best_regend; | |
4176 | #endif | |
5e69f11e | 4177 | |
fa9a63c5 RM |
4178 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
4179 | allocate space for that if we're not allocating space for anything | |
25fe55af | 4180 | else (see below). Also, we never need info about register 0 for |
fa9a63c5 RM |
4181 | any of the other register vectors, and it seems rather a kludge to |
4182 | treat `best_regend' differently than the rest. So we keep track of | |
4183 | the end of the best match so far in a separate variable. We | |
4184 | initialize this to NULL so that when we backtrack the first time | |
4185 | and need to test it, it's not garbage. */ | |
4186 | const char *match_end = NULL; | |
4187 | ||
4188 | /* This helps SET_REGS_MATCHED avoid doing redundant work. */ | |
4189 | int set_regs_matched_done = 0; | |
4190 | ||
4191 | /* Used when we pop values we don't care about. */ | |
4192 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
4193 | const char **reg_dummy; | |
4194 | register_info_type *reg_info_dummy; | |
4195 | #endif | |
4196 | ||
4197 | #ifdef DEBUG | |
4198 | /* Counts the total number of registers pushed. */ | |
5e69f11e | 4199 | unsigned num_regs_pushed = 0; |
fa9a63c5 RM |
4200 | #endif |
4201 | ||
4202 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
5e69f11e | 4203 | |
fa9a63c5 | 4204 | INIT_FAIL_STACK (); |
5e69f11e | 4205 | |
fa9a63c5 RM |
4206 | #ifdef MATCH_MAY_ALLOCATE |
4207 | /* Do not bother to initialize all the register variables if there are | |
4208 | no groups in the pattern, as it takes a fair amount of time. If | |
4209 | there are groups, we include space for register 0 (the whole | |
4210 | pattern), even though we never use it, since it simplifies the | |
4211 | array indexing. We should fix this. */ | |
4212 | if (bufp->re_nsub) | |
4213 | { | |
4214 | regstart = REGEX_TALLOC (num_regs, const char *); | |
4215 | regend = REGEX_TALLOC (num_regs, const char *); | |
4216 | old_regstart = REGEX_TALLOC (num_regs, const char *); | |
4217 | old_regend = REGEX_TALLOC (num_regs, const char *); | |
4218 | best_regstart = REGEX_TALLOC (num_regs, const char *); | |
4219 | best_regend = REGEX_TALLOC (num_regs, const char *); | |
4220 | reg_info = REGEX_TALLOC (num_regs, register_info_type); | |
4221 | reg_dummy = REGEX_TALLOC (num_regs, const char *); | |
4222 | reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); | |
4223 | ||
5e69f11e | 4224 | if (!(regstart && regend && old_regstart && old_regend && reg_info |
25fe55af RS |
4225 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) |
4226 | { | |
4227 | FREE_VARIABLES (); | |
4228 | return -2; | |
4229 | } | |
fa9a63c5 RM |
4230 | } |
4231 | else | |
4232 | { | |
4233 | /* We must initialize all our variables to NULL, so that | |
25fe55af | 4234 | `FREE_VARIABLES' doesn't try to free them. */ |
fa9a63c5 | 4235 | regstart = regend = old_regstart = old_regend = best_regstart |
25fe55af | 4236 | = best_regend = reg_dummy = NULL; |
fa9a63c5 RM |
4237 | reg_info = reg_info_dummy = (register_info_type *) NULL; |
4238 | } | |
4239 | #endif /* MATCH_MAY_ALLOCATE */ | |
4240 | ||
4241 | /* The starting position is bogus. */ | |
4242 | if (pos < 0 || pos > size1 + size2) | |
4243 | { | |
4244 | FREE_VARIABLES (); | |
4245 | return -1; | |
4246 | } | |
5e69f11e | 4247 | |
fa9a63c5 RM |
4248 | /* Initialize subexpression text positions to -1 to mark ones that no |
4249 | start_memory/stop_memory has been seen for. Also initialize the | |
4250 | register information struct. */ | |
4251 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
4252 | { | |
5e69f11e | 4253 | regstart[mcnt] = regend[mcnt] |
25fe55af | 4254 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; |
5e69f11e | 4255 | |
fa9a63c5 RM |
4256 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; |
4257 | IS_ACTIVE (reg_info[mcnt]) = 0; | |
4258 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
4259 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
4260 | } | |
5e69f11e | 4261 | |
fa9a63c5 | 4262 | /* We move `string1' into `string2' if the latter's empty -- but not if |
25fe55af | 4263 | `string1' is null. */ |
fa9a63c5 RM |
4264 | if (size2 == 0 && string1 != NULL) |
4265 | { | |
4266 | string2 = string1; | |
4267 | size2 = size1; | |
4268 | string1 = 0; | |
4269 | size1 = 0; | |
4270 | } | |
4271 | end1 = string1 + size1; | |
4272 | end2 = string2 + size2; | |
4273 | ||
4274 | /* Compute where to stop matching, within the two strings. */ | |
4275 | if (stop <= size1) | |
4276 | { | |
4277 | end_match_1 = string1 + stop; | |
4278 | end_match_2 = string2; | |
4279 | } | |
4280 | else | |
4281 | { | |
4282 | end_match_1 = end1; | |
4283 | end_match_2 = string2 + stop - size1; | |
4284 | } | |
4285 | ||
5e69f11e | 4286 | /* `p' scans through the pattern as `d' scans through the data. |
fa9a63c5 RM |
4287 | `dend' is the end of the input string that `d' points within. `d' |
4288 | is advanced into the following input string whenever necessary, but | |
4289 | this happens before fetching; therefore, at the beginning of the | |
4290 | loop, `d' can be pointing at the end of a string, but it cannot | |
4291 | equal `string2'. */ | |
4292 | if (size1 > 0 && pos <= size1) | |
4293 | { | |
4294 | d = string1 + pos; | |
4295 | dend = end_match_1; | |
4296 | } | |
4297 | else | |
4298 | { | |
4299 | d = string2 + pos - size1; | |
4300 | dend = end_match_2; | |
4301 | } | |
4302 | ||
4303 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
4304 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
4305 | DEBUG_PRINT1 ("The string to match is: `"); | |
4306 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
4307 | DEBUG_PRINT1 ("'\n"); | |
5e69f11e | 4308 | |
25fe55af | 4309 | /* This loops over pattern commands. It exits by returning from the |
fa9a63c5 RM |
4310 | function if the match is complete, or it drops through if the match |
4311 | fails at this starting point in the input data. */ | |
4312 | for (;;) | |
4313 | { | |
4314 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
4315 | ||
4316 | if (p == pend) | |
4317 | { /* End of pattern means we might have succeeded. */ | |
25fe55af | 4318 | DEBUG_PRINT1 ("end of pattern ... "); |
5e69f11e | 4319 | |
fa9a63c5 | 4320 | /* If we haven't matched the entire string, and we want the |
25fe55af RS |
4321 | longest match, try backtracking. */ |
4322 | if (d != end_match_2) | |
fa9a63c5 RM |
4323 | { |
4324 | /* 1 if this match ends in the same string (string1 or string2) | |
4325 | as the best previous match. */ | |
5e69f11e | 4326 | boolean same_str_p = (FIRST_STRING_P (match_end) |
fa9a63c5 RM |
4327 | == MATCHING_IN_FIRST_STRING); |
4328 | /* 1 if this match is the best seen so far. */ | |
4329 | boolean best_match_p; | |
4330 | ||
4331 | /* AIX compiler got confused when this was combined | |
25fe55af | 4332 | with the previous declaration. */ |
fa9a63c5 RM |
4333 | if (same_str_p) |
4334 | best_match_p = d > match_end; | |
4335 | else | |
4336 | best_match_p = !MATCHING_IN_FIRST_STRING; | |
4337 | ||
25fe55af RS |
4338 | DEBUG_PRINT1 ("backtracking.\n"); |
4339 | ||
4340 | if (!FAIL_STACK_EMPTY ()) | |
4341 | { /* More failure points to try. */ | |
4342 | ||
4343 | /* If exceeds best match so far, save it. */ | |
4344 | if (!best_regs_set || best_match_p) | |
4345 | { | |
4346 | best_regs_set = true; | |
4347 | match_end = d; | |
4348 | ||
4349 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
4350 | ||
4351 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
4352 | { | |
4353 | best_regstart[mcnt] = regstart[mcnt]; | |
4354 | best_regend[mcnt] = regend[mcnt]; | |
4355 | } | |
4356 | } | |
4357 | goto fail; | |
4358 | } | |
4359 | ||
4360 | /* If no failure points, don't restore garbage. And if | |
4361 | last match is real best match, don't restore second | |
4362 | best one. */ | |
4363 | else if (best_regs_set && !best_match_p) | |
4364 | { | |
4365 | restore_best_regs: | |
4366 | /* Restore best match. It may happen that `dend == | |
4367 | end_match_1' while the restored d is in string2. | |
4368 | For example, the pattern `x.*y.*z' against the | |
4369 | strings `x-' and `y-z-', if the two strings are | |
4370 | not consecutive in memory. */ | |
4371 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
4372 | ||
4373 | d = match_end; | |
4374 | dend = ((d >= string1 && d <= end1) | |
4375 | ? end_match_1 : end_match_2); | |
fa9a63c5 RM |
4376 | |
4377 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
4378 | { | |
4379 | regstart[mcnt] = best_regstart[mcnt]; | |
4380 | regend[mcnt] = best_regend[mcnt]; | |
4381 | } | |
25fe55af RS |
4382 | } |
4383 | } /* d != end_match_2 */ | |
fa9a63c5 RM |
4384 | |
4385 | succeed_label: | |
25fe55af | 4386 | DEBUG_PRINT1 ("Accepting match.\n"); |
fa9a63c5 | 4387 | |
25fe55af RS |
4388 | /* If caller wants register contents data back, do it. */ |
4389 | if (regs && !bufp->no_sub) | |
fa9a63c5 | 4390 | { |
25fe55af RS |
4391 | /* Have the register data arrays been allocated? */ |
4392 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
4393 | { /* No. So allocate them with malloc. We need one | |
4394 | extra element beyond `num_regs' for the `-1' marker | |
4395 | GNU code uses. */ | |
4396 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
4397 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
4398 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
4399 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
4400 | { |
4401 | FREE_VARIABLES (); | |
4402 | return -2; | |
4403 | } | |
25fe55af RS |
4404 | bufp->regs_allocated = REGS_REALLOCATE; |
4405 | } | |
4406 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
4407 | { /* Yes. If we need more elements than were already | |
4408 | allocated, reallocate them. If we need fewer, just | |
4409 | leave it alone. */ | |
4410 | if (regs->num_regs < num_regs + 1) | |
4411 | { | |
4412 | regs->num_regs = num_regs + 1; | |
4413 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
4414 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
4415 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
4416 | { |
4417 | FREE_VARIABLES (); | |
4418 | return -2; | |
4419 | } | |
25fe55af RS |
4420 | } |
4421 | } | |
4422 | else | |
fa9a63c5 RM |
4423 | { |
4424 | /* These braces fend off a "empty body in an else-statement" | |
25fe55af | 4425 | warning under GCC when assert expands to nothing. */ |
fa9a63c5 RM |
4426 | assert (bufp->regs_allocated == REGS_FIXED); |
4427 | } | |
4428 | ||
25fe55af RS |
4429 | /* Convert the pointer data in `regstart' and `regend' to |
4430 | indices. Register zero has to be set differently, | |
4431 | since we haven't kept track of any info for it. */ | |
4432 | if (regs->num_regs > 0) | |
4433 | { | |
4434 | regs->start[0] = pos; | |
4435 | regs->end[0] = (MATCHING_IN_FIRST_STRING | |
fa9a63c5 | 4436 | ? ((regoff_t) (d - string1)) |
25fe55af RS |
4437 | : ((regoff_t) (d - string2 + size1))); |
4438 | } | |
5e69f11e | 4439 | |
25fe55af RS |
4440 | /* Go through the first `min (num_regs, regs->num_regs)' |
4441 | registers, since that is all we initialized. */ | |
fa9a63c5 RM |
4442 | for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) |
4443 | { | |
25fe55af RS |
4444 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) |
4445 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
4446 | else | |
4447 | { | |
fa9a63c5 RM |
4448 | regs->start[mcnt] |
4449 | = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); | |
25fe55af | 4450 | regs->end[mcnt] |
fa9a63c5 | 4451 | = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); |
25fe55af | 4452 | } |
fa9a63c5 | 4453 | } |
5e69f11e | 4454 | |
25fe55af RS |
4455 | /* If the regs structure we return has more elements than |
4456 | were in the pattern, set the extra elements to -1. If | |
4457 | we (re)allocated the registers, this is the case, | |
4458 | because we always allocate enough to have at least one | |
4459 | -1 at the end. */ | |
4460 | for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) | |
4461 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
fa9a63c5 RM |
4462 | } /* regs && !bufp->no_sub */ |
4463 | ||
25fe55af RS |
4464 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
4465 | nfailure_points_pushed, nfailure_points_popped, | |
4466 | nfailure_points_pushed - nfailure_points_popped); | |
4467 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
fa9a63c5 | 4468 | |
25fe55af | 4469 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING |
5e69f11e | 4470 | ? string1 |
fa9a63c5 RM |
4471 | : string2 - size1); |
4472 | ||
25fe55af | 4473 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
fa9a63c5 | 4474 | |
25fe55af RS |
4475 | FREE_VARIABLES (); |
4476 | return mcnt; | |
4477 | } | |
fa9a63c5 | 4478 | |
25fe55af | 4479 | /* Otherwise match next pattern command. */ |
fa9a63c5 RM |
4480 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
4481 | { | |
25fe55af RS |
4482 | /* Ignore these. Used to ignore the n of succeed_n's which |
4483 | currently have n == 0. */ | |
4484 | case no_op: | |
4485 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
4486 | break; | |
fa9a63c5 RM |
4487 | |
4488 | case succeed: | |
25fe55af | 4489 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); |
fa9a63c5 RM |
4490 | goto succeed_label; |
4491 | ||
25fe55af RS |
4492 | /* Match the next n pattern characters exactly. The following |
4493 | byte in the pattern defines n, and the n bytes after that | |
4494 | are the characters to match. */ | |
fa9a63c5 RM |
4495 | case exactn: |
4496 | mcnt = *p++; | |
25fe55af | 4497 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
fa9a63c5 | 4498 | |
25fe55af RS |
4499 | /* This is written out as an if-else so we don't waste time |
4500 | testing `translate' inside the loop. */ | |
4501 | if (translate) | |
fa9a63c5 | 4502 | { |
e934739e RS |
4503 | #ifdef emacs |
4504 | if (multibyte) | |
4505 | do | |
4506 | { | |
4507 | int pat_charlen, buf_charlen; | |
e71b1971 | 4508 | unsigned int pat_ch, buf_ch; |
e934739e RS |
4509 | |
4510 | PREFETCH (); | |
4511 | pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen); | |
4512 | buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); | |
4513 | ||
4514 | if (RE_TRANSLATE (translate, buf_ch) | |
4515 | != pat_ch) | |
4516 | goto fail; | |
4517 | ||
4518 | p += pat_charlen; | |
4519 | d += buf_charlen; | |
4520 | mcnt -= pat_charlen; | |
4521 | } | |
4522 | while (mcnt > 0); | |
4523 | else | |
4524 | #endif /* not emacs */ | |
4525 | do | |
4526 | { | |
4527 | PREFETCH (); | |
4528 | if ((unsigned char) RE_TRANSLATE (translate, (unsigned char) *d++) | |
4529 | != (unsigned char) *p++) | |
4530 | goto fail; | |
4531 | } | |
4532 | while (--mcnt); | |
fa9a63c5 RM |
4533 | } |
4534 | else | |
4535 | { | |
4536 | do | |
4537 | { | |
4538 | PREFETCH (); | |
4539 | if (*d++ != (char) *p++) goto fail; | |
4540 | } | |
4541 | while (--mcnt); | |
4542 | } | |
4543 | SET_REGS_MATCHED (); | |
25fe55af | 4544 | break; |
fa9a63c5 RM |
4545 | |
4546 | ||
25fe55af | 4547 | /* Match any character except possibly a newline or a null. */ |
fa9a63c5 | 4548 | case anychar: |
e934739e RS |
4549 | { |
4550 | int buf_charlen; | |
e71b1971 | 4551 | unsigned int buf_ch; |
fa9a63c5 | 4552 | |
e934739e | 4553 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
fa9a63c5 | 4554 | |
e934739e | 4555 | PREFETCH (); |
fa9a63c5 | 4556 | |
e934739e RS |
4557 | #ifdef emacs |
4558 | if (multibyte) | |
4559 | buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); | |
4560 | else | |
4561 | #endif /* not emacs */ | |
4562 | { | |
85bc1d4e | 4563 | buf_ch = (unsigned char) *d; |
e934739e RS |
4564 | buf_charlen = 1; |
4565 | } | |
4566 | ||
4567 | buf_ch = TRANSLATE (buf_ch); | |
4568 | ||
4569 | if ((!(bufp->syntax & RE_DOT_NEWLINE) | |
4570 | && buf_ch == '\n') | |
4571 | || ((bufp->syntax & RE_DOT_NOT_NULL) | |
4572 | && buf_ch == '\000')) | |
4573 | goto fail; | |
4574 | ||
4575 | SET_REGS_MATCHED (); | |
4576 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); | |
4577 | d += buf_charlen; | |
4578 | } | |
fa9a63c5 RM |
4579 | break; |
4580 | ||
4581 | ||
4582 | case charset: | |
4583 | case charset_not: | |
4584 | { | |
b18215fc | 4585 | register unsigned int c; |
fa9a63c5 | 4586 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
b18215fc RS |
4587 | int len; |
4588 | ||
4589 | /* Start of actual range_table, or end of bitmap if there is no | |
4590 | range table. */ | |
4591 | unsigned char *range_table; | |
4592 | ||
25fe55af | 4593 | /* Nonzero if there is range table. */ |
b18215fc RS |
4594 | int range_table_exists; |
4595 | ||
25fe55af | 4596 | /* Number of ranges of range table. Not in bytes. */ |
b18215fc | 4597 | int count; |
fa9a63c5 | 4598 | |
25fe55af | 4599 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
fa9a63c5 RM |
4600 | |
4601 | PREFETCH (); | |
b18215fc | 4602 | c = (unsigned char) *d; |
fa9a63c5 | 4603 | |
b18215fc RS |
4604 | range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */ |
4605 | range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]); | |
4606 | if (range_table_exists) | |
4607 | EXTRACT_NUMBER_AND_INCR (count, range_table); | |
4608 | else | |
4609 | count = 0; | |
4610 | ||
4611 | if (multibyte && BASE_LEADING_CODE_P (c)) | |
4612 | c = STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
4613 | ||
4614 | if (SINGLE_BYTE_CHAR_P (c)) | |
4615 | { /* Lookup bitmap. */ | |
4616 | c = TRANSLATE (c); /* The character to match. */ | |
4617 | len = 1; | |
4618 | ||
4619 | /* Cast to `unsigned' instead of `unsigned char' in | |
4620 | case the bit list is a full 32 bytes long. */ | |
4621 | if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH) | |
fa9a63c5 RM |
4622 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
4623 | not = !not; | |
b18215fc RS |
4624 | } |
4625 | else if (range_table_exists) | |
4626 | CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count); | |
fa9a63c5 | 4627 | |
b18215fc | 4628 | p = CHARSET_RANGE_TABLE_END (range_table, count); |
fa9a63c5 RM |
4629 | |
4630 | if (!not) goto fail; | |
5e69f11e | 4631 | |
fa9a63c5 | 4632 | SET_REGS_MATCHED (); |
b18215fc | 4633 | d += len; |
fa9a63c5 RM |
4634 | break; |
4635 | } | |
4636 | ||
4637 | ||
25fe55af RS |
4638 | /* The beginning of a group is represented by start_memory. |
4639 | The arguments are the register number in the next byte, and the | |
4640 | number of groups inner to this one in the next. The text | |
4641 | matched within the group is recorded (in the internal | |
4642 | registers data structure) under the register number. */ | |
4643 | case start_memory: | |
fa9a63c5 RM |
4644 | DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); |
4645 | ||
25fe55af | 4646 | /* Find out if this group can match the empty string. */ |
fa9a63c5 | 4647 | p1 = p; /* To send to group_match_null_string_p. */ |
5e69f11e | 4648 | |
25fe55af RS |
4649 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) |
4650 | REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
4651 | = group_match_null_string_p (&p1, pend, reg_info); | |
4652 | ||
4653 | /* Save the position in the string where we were the last time | |
4654 | we were at this open-group operator in case the group is | |
4655 | operated upon by a repetition operator, e.g., with `(a*)*b' | |
4656 | against `ab'; then we want to ignore where we are now in | |
4657 | the string in case this attempt to match fails. */ | |
4658 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
4659 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
4660 | : regstart[*p]; | |
5e69f11e | 4661 | DEBUG_PRINT2 (" old_regstart: %d\n", |
fa9a63c5 RM |
4662 | POINTER_TO_OFFSET (old_regstart[*p])); |
4663 | ||
25fe55af | 4664 | regstart[*p] = d; |
fa9a63c5 RM |
4665 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
4666 | ||
25fe55af RS |
4667 | IS_ACTIVE (reg_info[*p]) = 1; |
4668 | MATCHED_SOMETHING (reg_info[*p]) = 0; | |
fa9a63c5 RM |
4669 | |
4670 | /* Clear this whenever we change the register activity status. */ | |
4671 | set_regs_matched_done = 0; | |
5e69f11e | 4672 | |
25fe55af RS |
4673 | /* This is the new highest active register. */ |
4674 | highest_active_reg = *p; | |
5e69f11e | 4675 | |
25fe55af RS |
4676 | /* If nothing was active before, this is the new lowest active |
4677 | register. */ | |
4678 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
4679 | lowest_active_reg = *p; | |
fa9a63c5 | 4680 | |
25fe55af RS |
4681 | /* Move past the register number and inner group count. */ |
4682 | p += 2; | |
fa9a63c5 RM |
4683 | just_past_start_mem = p; |
4684 | ||
25fe55af | 4685 | break; |
fa9a63c5 RM |
4686 | |
4687 | ||
25fe55af RS |
4688 | /* The stop_memory opcode represents the end of a group. Its |
4689 | arguments are the same as start_memory's: the register | |
4690 | number, and the number of inner groups. */ | |
fa9a63c5 RM |
4691 | case stop_memory: |
4692 | DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); | |
5e69f11e | 4693 | |
25fe55af RS |
4694 | /* We need to save the string position the last time we were at |
4695 | this close-group operator in case the group is operated | |
4696 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
4697 | against `aba'; then we want to ignore where we are now in | |
4698 | the string in case this attempt to match fails. */ | |
4699 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
4700 | ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
fa9a63c5 | 4701 | : regend[*p]; |
5e69f11e | 4702 | DEBUG_PRINT2 (" old_regend: %d\n", |
fa9a63c5 RM |
4703 | POINTER_TO_OFFSET (old_regend[*p])); |
4704 | ||
25fe55af | 4705 | regend[*p] = d; |
fa9a63c5 RM |
4706 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
4707 | ||
25fe55af RS |
4708 | /* This register isn't active anymore. */ |
4709 | IS_ACTIVE (reg_info[*p]) = 0; | |
fa9a63c5 RM |
4710 | |
4711 | /* Clear this whenever we change the register activity status. */ | |
4712 | set_regs_matched_done = 0; | |
4713 | ||
25fe55af RS |
4714 | /* If this was the only register active, nothing is active |
4715 | anymore. */ | |
4716 | if (lowest_active_reg == highest_active_reg) | |
4717 | { | |
4718 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
4719 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
4720 | } | |
4721 | else | |
4722 | { /* We must scan for the new highest active register, since | |
4723 | it isn't necessarily one less than now: consider | |
4724 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
4725 | new highest active register is 1. */ | |
4726 | unsigned char r = *p - 1; | |
4727 | while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
4728 | r--; | |
4729 | ||
4730 | /* If we end up at register zero, that means that we saved | |
4731 | the registers as the result of an `on_failure_jump', not | |
4732 | a `start_memory', and we jumped to past the innermost | |
4733 | `stop_memory'. For example, in ((.)*) we save | |
4734 | registers 1 and 2 as a result of the *, but when we pop | |
4735 | back to the second ), we are at the stop_memory 1. | |
4736 | Thus, nothing is active. */ | |
fa9a63c5 | 4737 | if (r == 0) |
25fe55af RS |
4738 | { |
4739 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
4740 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
4741 | } | |
4742 | else | |
4743 | highest_active_reg = r; | |
4744 | } | |
4745 | ||
4746 | /* If just failed to match something this time around with a | |
4747 | group that's operated on by a repetition operator, try to | |
4748 | force exit from the ``loop'', and restore the register | |
4749 | information for this group that we had before trying this | |
4750 | last match. */ | |
4751 | if ((!MATCHED_SOMETHING (reg_info[*p]) | |
4752 | || just_past_start_mem == p - 1) | |
5e69f11e | 4753 | && (p + 2) < pend) |
25fe55af RS |
4754 | { |
4755 | boolean is_a_jump_n = false; | |
4756 | ||
4757 | p1 = p + 2; | |
4758 | mcnt = 0; | |
4759 | switch ((re_opcode_t) *p1++) | |
4760 | { | |
4761 | case jump_n: | |
fa9a63c5 | 4762 | is_a_jump_n = true; |
25fe55af | 4763 | case pop_failure_jump: |
fa9a63c5 RM |
4764 | case maybe_pop_jump: |
4765 | case jump: | |
4766 | case dummy_failure_jump: | |
25fe55af | 4767 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
fa9a63c5 RM |
4768 | if (is_a_jump_n) |
4769 | p1 += 2; | |
25fe55af | 4770 | break; |
5e69f11e | 4771 | |
25fe55af RS |
4772 | default: |
4773 | /* do nothing */ ; | |
4774 | } | |
fa9a63c5 | 4775 | p1 += mcnt; |
5e69f11e | 4776 | |
25fe55af RS |
4777 | /* If the next operation is a jump backwards in the pattern |
4778 | to an on_failure_jump right before the start_memory | |
4779 | corresponding to this stop_memory, exit from the loop | |
4780 | by forcing a failure after pushing on the stack the | |
4781 | on_failure_jump's jump in the pattern, and d. */ | |
4782 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
4783 | && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
fa9a63c5 | 4784 | { |
25fe55af RS |
4785 | /* If this group ever matched anything, then restore |
4786 | what its registers were before trying this last | |
4787 | failed match, e.g., with `(a*)*b' against `ab' for | |
4788 | regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
4789 | against `aba' for regend[3]. | |
5e69f11e | 4790 | |
25fe55af RS |
4791 | Also restore the registers for inner groups for, |
4792 | e.g., `((a*)(b*))*' against `aba' (register 3 would | |
4793 | otherwise get trashed). */ | |
5e69f11e | 4794 | |
25fe55af | 4795 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) |
fa9a63c5 | 4796 | { |
5e69f11e RM |
4797 | unsigned r; |
4798 | ||
25fe55af | 4799 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; |
5e69f11e | 4800 | |
fa9a63c5 | 4801 | /* Restore this and inner groups' (if any) registers. */ |
25fe55af RS |
4802 | for (r = *p; r < *p + *(p + 1); r++) |
4803 | { | |
4804 | regstart[r] = old_regstart[r]; | |
4805 | ||
4806 | /* xx why this test? */ | |
4807 | if (old_regend[r] >= regstart[r]) | |
4808 | regend[r] = old_regend[r]; | |
4809 | } | |
4810 | } | |
fa9a63c5 | 4811 | p1++; |
25fe55af RS |
4812 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
4813 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
fa9a63c5 | 4814 | |
25fe55af RS |
4815 | goto fail; |
4816 | } | |
4817 | } | |
5e69f11e | 4818 | |
25fe55af RS |
4819 | /* Move past the register number and the inner group count. */ |
4820 | p += 2; | |
4821 | break; | |
fa9a63c5 RM |
4822 | |
4823 | ||
4824 | /* \<digit> has been turned into a `duplicate' command which is | |
25fe55af RS |
4825 | followed by the numeric value of <digit> as the register number. */ |
4826 | case duplicate: | |
fa9a63c5 RM |
4827 | { |
4828 | register const char *d2, *dend2; | |
25fe55af | 4829 | int regno = *p++; /* Get which register to match against. */ |
fa9a63c5 RM |
4830 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
4831 | ||
25fe55af RS |
4832 | /* Can't back reference a group which we've never matched. */ |
4833 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
4834 | goto fail; | |
5e69f11e | 4835 | |
25fe55af RS |
4836 | /* Where in input to try to start matching. */ |
4837 | d2 = regstart[regno]; | |
5e69f11e | 4838 | |
25fe55af RS |
4839 | /* Where to stop matching; if both the place to start and |
4840 | the place to stop matching are in the same string, then | |
4841 | set to the place to stop, otherwise, for now have to use | |
4842 | the end of the first string. */ | |
fa9a63c5 | 4843 | |
25fe55af | 4844 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
fa9a63c5 RM |
4845 | == FIRST_STRING_P (regend[regno])) |
4846 | ? regend[regno] : end_match_1); | |
4847 | for (;;) | |
4848 | { | |
4849 | /* If necessary, advance to next segment in register | |
25fe55af | 4850 | contents. */ |
fa9a63c5 RM |
4851 | while (d2 == dend2) |
4852 | { | |
4853 | if (dend2 == end_match_2) break; | |
4854 | if (dend2 == regend[regno]) break; | |
4855 | ||
25fe55af RS |
4856 | /* End of string1 => advance to string2. */ |
4857 | d2 = string2; | |
4858 | dend2 = regend[regno]; | |
fa9a63c5 RM |
4859 | } |
4860 | /* At end of register contents => success */ | |
4861 | if (d2 == dend2) break; | |
4862 | ||
4863 | /* If necessary, advance to next segment in data. */ | |
4864 | PREFETCH (); | |
4865 | ||
4866 | /* How many characters left in this segment to match. */ | |
4867 | mcnt = dend - d; | |
5e69f11e | 4868 | |
fa9a63c5 | 4869 | /* Want how many consecutive characters we can match in |
25fe55af RS |
4870 | one shot, so, if necessary, adjust the count. */ |
4871 | if (mcnt > dend2 - d2) | |
fa9a63c5 | 4872 | mcnt = dend2 - d2; |
5e69f11e | 4873 | |
fa9a63c5 | 4874 | /* Compare that many; failure if mismatch, else move |
25fe55af | 4875 | past them. */ |
5e69f11e | 4876 | if (translate |
25fe55af RS |
4877 | ? bcmp_translate (d, d2, mcnt, translate) |
4878 | : bcmp (d, d2, mcnt)) | |
fa9a63c5 RM |
4879 | goto fail; |
4880 | d += mcnt, d2 += mcnt; | |
4881 | ||
25fe55af | 4882 | /* Do this because we've match some characters. */ |
fa9a63c5 RM |
4883 | SET_REGS_MATCHED (); |
4884 | } | |
4885 | } | |
4886 | break; | |
4887 | ||
4888 | ||
25fe55af RS |
4889 | /* begline matches the empty string at the beginning of the string |
4890 | (unless `not_bol' is set in `bufp'), and, if | |
4891 | `newline_anchor' is set, after newlines. */ | |
fa9a63c5 | 4892 | case begline: |
25fe55af | 4893 | DEBUG_PRINT1 ("EXECUTING begline.\n"); |
5e69f11e | 4894 | |
25fe55af RS |
4895 | if (AT_STRINGS_BEG (d)) |
4896 | { | |
4897 | if (!bufp->not_bol) break; | |
4898 | } | |
4899 | else if (d[-1] == '\n' && bufp->newline_anchor) | |
4900 | { | |
4901 | break; | |
4902 | } | |
4903 | /* In all other cases, we fail. */ | |
4904 | goto fail; | |
fa9a63c5 RM |
4905 | |
4906 | ||
25fe55af | 4907 | /* endline is the dual of begline. */ |
fa9a63c5 | 4908 | case endline: |
25fe55af | 4909 | DEBUG_PRINT1 ("EXECUTING endline.\n"); |
fa9a63c5 | 4910 | |
25fe55af RS |
4911 | if (AT_STRINGS_END (d)) |
4912 | { | |
4913 | if (!bufp->not_eol) break; | |
4914 | } | |
5e69f11e | 4915 | |
25fe55af RS |
4916 | /* We have to ``prefetch'' the next character. */ |
4917 | else if ((d == end1 ? *string2 : *d) == '\n' | |
4918 | && bufp->newline_anchor) | |
4919 | { | |
4920 | break; | |
4921 | } | |
4922 | goto fail; | |
fa9a63c5 RM |
4923 | |
4924 | ||
4925 | /* Match at the very beginning of the data. */ | |
25fe55af RS |
4926 | case begbuf: |
4927 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
4928 | if (AT_STRINGS_BEG (d)) | |
4929 | break; | |
4930 | goto fail; | |
fa9a63c5 RM |
4931 | |
4932 | ||
4933 | /* Match at the very end of the data. */ | |
25fe55af RS |
4934 | case endbuf: |
4935 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
fa9a63c5 RM |
4936 | if (AT_STRINGS_END (d)) |
4937 | break; | |
25fe55af | 4938 | goto fail; |
5e69f11e | 4939 | |
5e69f11e | 4940 | |
25fe55af RS |
4941 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
4942 | pushes NULL as the value for the string on the stack. Then | |
4943 | `pop_failure_point' will keep the current value for the | |
4944 | string, instead of restoring it. To see why, consider | |
4945 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
4946 | then the . fails against the \n. But the next thing we want | |
4947 | to do is match the \n against the \n; if we restored the | |
4948 | string value, we would be back at the foo. | |
4949 | ||
4950 | Because this is used only in specific cases, we don't need to | |
4951 | check all the things that `on_failure_jump' does, to make | |
4952 | sure the right things get saved on the stack. Hence we don't | |
4953 | share its code. The only reason to push anything on the | |
4954 | stack at all is that otherwise we would have to change | |
4955 | `anychar's code to do something besides goto fail in this | |
4956 | case; that seems worse than this. */ | |
4957 | case on_failure_keep_string_jump: | |
4958 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
fa9a63c5 | 4959 | |
25fe55af RS |
4960 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
4961 | DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
b18215fc | 4962 | |
25fe55af RS |
4963 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); |
4964 | break; | |
fa9a63c5 RM |
4965 | |
4966 | ||
4967 | /* Uses of on_failure_jump: | |
5e69f11e | 4968 | |
25fe55af RS |
4969 | Each alternative starts with an on_failure_jump that points |
4970 | to the beginning of the next alternative. Each alternative | |
4971 | except the last ends with a jump that in effect jumps past | |
4972 | the rest of the alternatives. (They really jump to the | |
4973 | ending jump of the following alternative, because tensioning | |
4974 | these jumps is a hassle.) | |
fa9a63c5 | 4975 | |
25fe55af RS |
4976 | Repeats start with an on_failure_jump that points past both |
4977 | the repetition text and either the following jump or | |
4978 | pop_failure_jump back to this on_failure_jump. */ | |
fa9a63c5 | 4979 | case on_failure_jump: |
25fe55af RS |
4980 | on_failure: |
4981 | DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
4982 | ||
4983 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4984 | DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
4985 | ||
4986 | /* If this on_failure_jump comes right before a group (i.e., | |
4987 | the original * applied to a group), save the information | |
4988 | for that group and all inner ones, so that if we fail back | |
4989 | to this point, the group's information will be correct. | |
4990 | For example, in \(a*\)*\1, we need the preceding group, | |
4991 | and in \(zz\(a*\)b*\)\2, we need the inner group. */ | |
4992 | ||
4993 | /* We can't use `p' to check ahead because we push | |
4994 | a failure point to `p + mcnt' after we do this. */ | |
4995 | p1 = p; | |
4996 | ||
4997 | /* We need to skip no_op's before we look for the | |
4998 | start_memory in case this on_failure_jump is happening as | |
4999 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
5000 | against aba. */ | |
5001 | while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
5002 | p1++; | |
5003 | ||
5004 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
5005 | { | |
5006 | /* We have a new highest active register now. This will | |
5007 | get reset at the start_memory we are about to get to, | |
5008 | but we will have saved all the registers relevant to | |
5009 | this repetition op, as described above. */ | |
5010 | highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
5011 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
5012 | lowest_active_reg = *(p1 + 1); | |
5013 | } | |
5014 | ||
5015 | DEBUG_PRINT1 (":\n"); | |
5016 | PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
5017 | break; | |
5018 | ||
5019 | ||
5020 | /* A smart repeat ends with `maybe_pop_jump'. | |
5021 | We change it to either `pop_failure_jump' or `jump'. */ | |
5022 | case maybe_pop_jump: | |
5023 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5024 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
5025 | { | |
fa9a63c5 RM |
5026 | register unsigned char *p2 = p; |
5027 | ||
25fe55af RS |
5028 | /* Compare the beginning of the repeat with what in the |
5029 | pattern follows its end. If we can establish that there | |
5030 | is nothing that they would both match, i.e., that we | |
5031 | would have to backtrack because of (as in, e.g., `a*a') | |
5032 | then we can change to pop_failure_jump, because we'll | |
5033 | never have to backtrack. | |
5e69f11e | 5034 | |
25fe55af RS |
5035 | This is not true in the case of alternatives: in |
5036 | `(a|ab)*' we do need to backtrack to the `ab' alternative | |
5037 | (e.g., if the string was `ab'). But instead of trying to | |
5038 | detect that here, the alternative has put on a dummy | |
5039 | failure point which is what we will end up popping. */ | |
fa9a63c5 RM |
5040 | |
5041 | /* Skip over open/close-group commands. | |
5042 | If what follows this loop is a ...+ construct, | |
5043 | look at what begins its body, since we will have to | |
5044 | match at least one of that. */ | |
5045 | while (1) | |
5046 | { | |
5047 | if (p2 + 2 < pend | |
5048 | && ((re_opcode_t) *p2 == stop_memory | |
5049 | || (re_opcode_t) *p2 == start_memory)) | |
5050 | p2 += 3; | |
5051 | else if (p2 + 6 < pend | |
5052 | && (re_opcode_t) *p2 == dummy_failure_jump) | |
5053 | p2 += 6; | |
5054 | else | |
5055 | break; | |
5056 | } | |
5057 | ||
5058 | p1 = p + mcnt; | |
5059 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
5e69f11e | 5060 | to the `maybe_finalize_jump' of this case. Examine what |
25fe55af | 5061 | follows. */ |
fa9a63c5 | 5062 | |
25fe55af RS |
5063 | /* If we're at the end of the pattern, we can change. */ |
5064 | if (p2 == pend) | |
fa9a63c5 RM |
5065 | { |
5066 | /* Consider what happens when matching ":\(.*\)" | |
5067 | against ":/". I don't really understand this code | |
25fe55af RS |
5068 | yet. */ |
5069 | p[-3] = (unsigned char) pop_failure_jump; | |
5070 | DEBUG_PRINT1 | |
5071 | (" End of pattern: change to `pop_failure_jump'.\n"); | |
5072 | } | |
fa9a63c5 | 5073 | |
25fe55af | 5074 | else if ((re_opcode_t) *p2 == exactn |
fa9a63c5 RM |
5075 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) |
5076 | { | |
b18215fc | 5077 | register unsigned int c |
25fe55af | 5078 | = *p2 == (unsigned char) endline ? '\n' : p2[2]; |
fa9a63c5 | 5079 | |
b18215fc | 5080 | if ((re_opcode_t) p1[3] == exactn) |
e318085a | 5081 | { |
b18215fc RS |
5082 | if (!(multibyte /* && (c != '\n') */ |
5083 | && BASE_LEADING_CODE_P (c)) | |
5084 | ? c != p1[5] | |
5085 | : (STRING_CHAR (&p2[2], pend - &p2[2]) | |
5086 | != STRING_CHAR (&p1[5], pend - &p1[5]))) | |
25fe55af RS |
5087 | { |
5088 | p[-3] = (unsigned char) pop_failure_jump; | |
5089 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
5090 | c, p1[5]); | |
5091 | } | |
e318085a | 5092 | } |
5e69f11e | 5093 | |
fa9a63c5 RM |
5094 | else if ((re_opcode_t) p1[3] == charset |
5095 | || (re_opcode_t) p1[3] == charset_not) | |
5096 | { | |
5097 | int not = (re_opcode_t) p1[3] == charset_not; | |
5e69f11e | 5098 | |
b18215fc RS |
5099 | if (multibyte /* && (c != '\n') */ |
5100 | && BASE_LEADING_CODE_P (c)) | |
5101 | c = STRING_CHAR (&p2[2], pend - &p2[2]); | |
5102 | ||
5103 | /* Test if C is listed in charset (or charset_not) | |
5104 | at `&p1[3]'. */ | |
5105 | if (SINGLE_BYTE_CHAR_P (c)) | |
5106 | { | |
5107 | if (c < CHARSET_BITMAP_SIZE (&p1[3]) * BYTEWIDTH | |
fa9a63c5 RM |
5108 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
5109 | not = !not; | |
b18215fc RS |
5110 | } |
5111 | else if (CHARSET_RANGE_TABLE_EXISTS_P (&p1[3])) | |
5112 | CHARSET_LOOKUP_RANGE_TABLE (not, c, &p1[3]); | |
fa9a63c5 | 5113 | |
25fe55af RS |
5114 | /* `not' is equal to 1 if c would match, which means |
5115 | that we can't change to pop_failure_jump. */ | |
fa9a63c5 | 5116 | if (!not) |
25fe55af RS |
5117 | { |
5118 | p[-3] = (unsigned char) pop_failure_jump; | |
5119 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
5120 | } | |
fa9a63c5 RM |
5121 | } |
5122 | } | |
25fe55af | 5123 | else if ((re_opcode_t) *p2 == charset) |
fa9a63c5 | 5124 | { |
b18215fc | 5125 | if ((re_opcode_t) p1[3] == exactn) |
e318085a | 5126 | { |
b18215fc RS |
5127 | register unsigned int c = p1[5]; |
5128 | int not = 0; | |
5129 | ||
5130 | if (multibyte && BASE_LEADING_CODE_P (c)) | |
5131 | c = STRING_CHAR (&p1[5], pend - &p1[5]); | |
5132 | ||
25fe55af | 5133 | /* Test if C is listed in charset at `p2'. */ |
b18215fc RS |
5134 | if (SINGLE_BYTE_CHAR_P (c)) |
5135 | { | |
5136 | if (c < CHARSET_BITMAP_SIZE (p2) * BYTEWIDTH | |
5137 | && (p2[2 + c / BYTEWIDTH] | |
5138 | & (1 << (c % BYTEWIDTH)))) | |
5139 | not = !not; | |
5140 | } | |
5141 | else if (CHARSET_RANGE_TABLE_EXISTS_P (p2)) | |
5142 | CHARSET_LOOKUP_RANGE_TABLE (not, c, p2); | |
5143 | ||
5144 | if (!not) | |
25fe55af RS |
5145 | { |
5146 | p[-3] = (unsigned char) pop_failure_jump; | |
5147 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
b18215fc | 5148 | } |
25fe55af | 5149 | } |
5e69f11e | 5150 | |
b18215fc RS |
5151 | /* It is hard to list up all the character in charset |
5152 | P2 if it includes multibyte character. Give up in | |
5153 | such case. */ | |
5154 | else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2)) | |
5155 | { | |
5156 | /* Now, we are sure that P2 has no range table. | |
5157 | So, for the size of bitmap in P2, `p2[1]' is | |
25fe55af | 5158 | enough. But P1 may have range table, so the |
b18215fc RS |
5159 | size of bitmap table of P1 is extracted by |
5160 | using macro `CHARSET_BITMAP_SIZE'. | |
5161 | ||
5162 | Since we know that all the character listed in | |
5163 | P2 is ASCII, it is enough to test only bitmap | |
5164 | table of P1. */ | |
25fe55af | 5165 | |
b18215fc | 5166 | if ((re_opcode_t) p1[3] == charset_not) |
fa9a63c5 RM |
5167 | { |
5168 | int idx; | |
b18215fc | 5169 | /* We win if the charset_not inside the loop lists |
25fe55af | 5170 | every character listed in the charset after. */ |
fa9a63c5 RM |
5171 | for (idx = 0; idx < (int) p2[1]; idx++) |
5172 | if (! (p2[2 + idx] == 0 | |
b18215fc | 5173 | || (idx < CHARSET_BITMAP_SIZE (&p1[3]) |
fa9a63c5 RM |
5174 | && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) |
5175 | break; | |
5176 | ||
5177 | if (idx == p2[1]) | |
25fe55af RS |
5178 | { |
5179 | p[-3] = (unsigned char) pop_failure_jump; | |
5180 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
5181 | } | |
fa9a63c5 RM |
5182 | } |
5183 | else if ((re_opcode_t) p1[3] == charset) | |
5184 | { | |
5185 | int idx; | |
5186 | /* We win if the charset inside the loop | |
5187 | has no overlap with the one after the loop. */ | |
5188 | for (idx = 0; | |
b18215fc RS |
5189 | (idx < (int) p2[1] |
5190 | && idx < CHARSET_BITMAP_SIZE (&p1[3])); | |
fa9a63c5 RM |
5191 | idx++) |
5192 | if ((p2[2 + idx] & p1[5 + idx]) != 0) | |
5193 | break; | |
5194 | ||
b18215fc RS |
5195 | if (idx == p2[1] |
5196 | || idx == CHARSET_BITMAP_SIZE (&p1[3])) | |
25fe55af RS |
5197 | { |
5198 | p[-3] = (unsigned char) pop_failure_jump; | |
5199 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
5200 | } | |
fa9a63c5 RM |
5201 | } |
5202 | } | |
5203 | } | |
b18215fc | 5204 | } |
fa9a63c5 RM |
5205 | p -= 2; /* Point at relative address again. */ |
5206 | if ((re_opcode_t) p[-1] != pop_failure_jump) | |
5207 | { | |
5208 | p[-1] = (unsigned char) jump; | |
25fe55af | 5209 | DEBUG_PRINT1 (" Match => jump.\n"); |
fa9a63c5 RM |
5210 | goto unconditional_jump; |
5211 | } | |
25fe55af | 5212 | /* Note fall through. */ |
fa9a63c5 RM |
5213 | |
5214 | ||
5215 | /* The end of a simple repeat has a pop_failure_jump back to | |
25fe55af RS |
5216 | its matching on_failure_jump, where the latter will push a |
5217 | failure point. The pop_failure_jump takes off failure | |
5218 | points put on by this pop_failure_jump's matching | |
5219 | on_failure_jump; we got through the pattern to here from the | |
5220 | matching on_failure_jump, so didn't fail. */ | |
5221 | case pop_failure_jump: | |
5222 | { | |
5223 | /* We need to pass separate storage for the lowest and | |
5224 | highest registers, even though we don't care about the | |
5225 | actual values. Otherwise, we will restore only one | |
5226 | register from the stack, since lowest will == highest in | |
5227 | `pop_failure_point'. */ | |
5228 | unsigned dummy_low_reg, dummy_high_reg; | |
5229 | unsigned char *pdummy; | |
5230 | const char *sdummy; | |
5231 | ||
5232 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
5233 | POP_FAILURE_POINT (sdummy, pdummy, | |
5234 | dummy_low_reg, dummy_high_reg, | |
5235 | reg_dummy, reg_dummy, reg_info_dummy); | |
5236 | } | |
5237 | /* Note fall through. */ | |
5238 | ||
5239 | ||
5240 | /* Unconditionally jump (without popping any failure points). */ | |
5241 | case jump: | |
fa9a63c5 RM |
5242 | unconditional_jump: |
5243 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
25fe55af RS |
5244 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
5245 | p += mcnt; /* Do the jump. */ | |
5246 | DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
5247 | break; | |
5248 | ||
5249 | ||
5250 | /* We need this opcode so we can detect where alternatives end | |
5251 | in `group_match_null_string_p' et al. */ | |
5252 | case jump_past_alt: | |
5253 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
5254 | goto unconditional_jump; | |
5255 | ||
5256 | ||
5257 | /* Normally, the on_failure_jump pushes a failure point, which | |
5258 | then gets popped at pop_failure_jump. We will end up at | |
5259 | pop_failure_jump, also, and with a pattern of, say, `a+', we | |
5260 | are skipping over the on_failure_jump, so we have to push | |
5261 | something meaningless for pop_failure_jump to pop. */ | |
5262 | case dummy_failure_jump: | |
5263 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
5264 | /* It doesn't matter what we push for the string here. What | |
5265 | the code at `fail' tests is the value for the pattern. */ | |
5266 | PUSH_FAILURE_POINT (0, 0, -2); | |
5267 | goto unconditional_jump; | |
5268 | ||
5269 | ||
5270 | /* At the end of an alternative, we need to push a dummy failure | |
5271 | point in case we are followed by a `pop_failure_jump', because | |
5272 | we don't want the failure point for the alternative to be | |
5273 | popped. For example, matching `(a|ab)*' against `aab' | |
5274 | requires that we match the `ab' alternative. */ | |
5275 | case push_dummy_failure: | |
5276 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
5277 | /* See comments just above at `dummy_failure_jump' about the | |
5278 | two zeroes. */ | |
5279 | PUSH_FAILURE_POINT (0, 0, -2); | |
fa9a63c5 RM |
5280 | break; |
5281 | ||
25fe55af RS |
5282 | /* Have to succeed matching what follows at least n times. |
5283 | After that, handle like `on_failure_jump'. */ | |
5284 | case succeed_n: | |
5285 | EXTRACT_NUMBER (mcnt, p + 2); | |
5286 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
5e69f11e | 5287 | |
25fe55af RS |
5288 | assert (mcnt >= 0); |
5289 | /* Originally, this is how many times we HAVE to succeed. */ | |
5290 | if (mcnt > 0) | |
5291 | { | |
5292 | mcnt--; | |
fa9a63c5 | 5293 | p += 2; |
25fe55af RS |
5294 | STORE_NUMBER_AND_INCR (p, mcnt); |
5295 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); | |
5296 | } | |
fa9a63c5 | 5297 | else if (mcnt == 0) |
25fe55af RS |
5298 | { |
5299 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); | |
fa9a63c5 | 5300 | p[2] = (unsigned char) no_op; |
25fe55af RS |
5301 | p[3] = (unsigned char) no_op; |
5302 | goto on_failure; | |
5303 | } | |
5304 | break; | |
5305 | ||
5306 | case jump_n: | |
5307 | EXTRACT_NUMBER (mcnt, p + 2); | |
5308 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
5309 | ||
5310 | /* Originally, this is how many times we CAN jump. */ | |
5311 | if (mcnt) | |
5312 | { | |
5313 | mcnt--; | |
5314 | STORE_NUMBER (p + 2, mcnt); | |
5e69f11e | 5315 | goto unconditional_jump; |
25fe55af RS |
5316 | } |
5317 | /* If don't have to jump any more, skip over the rest of command. */ | |
5e69f11e RM |
5318 | else |
5319 | p += 4; | |
25fe55af | 5320 | break; |
5e69f11e | 5321 | |
fa9a63c5 RM |
5322 | case set_number_at: |
5323 | { | |
25fe55af | 5324 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
fa9a63c5 | 5325 | |
25fe55af RS |
5326 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
5327 | p1 = p + mcnt; | |
5328 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5329 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
fa9a63c5 | 5330 | STORE_NUMBER (p1, mcnt); |
25fe55af RS |
5331 | break; |
5332 | } | |
9121ca40 KH |
5333 | |
5334 | case wordbound: | |
5335 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
fa9a63c5 | 5336 | |
b18215fc | 5337 | /* We SUCCEED in one of the following cases: */ |
9121ca40 | 5338 | |
b18215fc | 5339 | /* Case 1: D is at the beginning or the end of string. */ |
9121ca40 KH |
5340 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
5341 | break; | |
b18215fc RS |
5342 | else |
5343 | { | |
5344 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5345 | is the character at D, and S2 is the syntax of C2. */ | |
5346 | int c1, c2, s1, s2; | |
5347 | int pos1 = PTR_TO_OFFSET (d - 1); | |
5d967c7a | 5348 | int charpos; |
9121ca40 | 5349 | |
b18215fc RS |
5350 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
5351 | GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2); | |
5352 | #ifdef emacs | |
92432794 | 5353 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (pos1 ? pos1 : 1); |
5d967c7a | 5354 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5355 | #endif |
b18215fc RS |
5356 | s1 = SYNTAX (c1); |
5357 | #ifdef emacs | |
5d967c7a | 5358 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
25fe55af | 5359 | #endif |
b18215fc RS |
5360 | s2 = SYNTAX (c2); |
5361 | ||
5362 | if (/* Case 2: Only one of S1 and S2 is Sword. */ | |
5363 | ((s1 == Sword) != (s2 == Sword)) | |
5364 | /* Case 3: Both of S1 and S2 are Sword, and macro | |
25fe55af | 5365 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
b18215fc | 5366 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
9121ca40 | 5367 | break; |
9121ca40 | 5368 | } |
b18215fc | 5369 | goto fail; |
9121ca40 KH |
5370 | |
5371 | case notwordbound: | |
9121ca40 | 5372 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); |
b18215fc RS |
5373 | |
5374 | /* We FAIL in one of the following cases: */ | |
5375 | ||
5376 | /* Case 1: D is at the beginning or the end of string. */ | |
9121ca40 KH |
5377 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
5378 | goto fail; | |
b18215fc RS |
5379 | else |
5380 | { | |
5381 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5382 | is the character at D, and S2 is the syntax of C2. */ | |
5383 | int c1, c2, s1, s2; | |
5384 | int pos1 = PTR_TO_OFFSET (d - 1); | |
5d967c7a | 5385 | int charpos; |
9121ca40 | 5386 | |
b18215fc RS |
5387 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
5388 | GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2); | |
5389 | #ifdef emacs | |
92432794 | 5390 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (pos1); |
5d967c7a | 5391 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5392 | #endif |
b18215fc RS |
5393 | s1 = SYNTAX (c1); |
5394 | #ifdef emacs | |
5d967c7a | 5395 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
25fe55af | 5396 | #endif |
b18215fc RS |
5397 | s2 = SYNTAX (c2); |
5398 | ||
5399 | if (/* Case 2: Only one of S1 and S2 is Sword. */ | |
5400 | ((s1 == Sword) != (s2 == Sword)) | |
5401 | /* Case 3: Both of S1 and S2 are Sword, and macro | |
25fe55af | 5402 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
b18215fc | 5403 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
9121ca40 | 5404 | goto fail; |
9121ca40 | 5405 | } |
b18215fc | 5406 | break; |
fa9a63c5 RM |
5407 | |
5408 | case wordbeg: | |
25fe55af | 5409 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
fa9a63c5 | 5410 | |
b18215fc RS |
5411 | /* We FAIL in one of the following cases: */ |
5412 | ||
25fe55af | 5413 | /* Case 1: D is at the end of string. */ |
b18215fc | 5414 | if (AT_STRINGS_END (d)) |
25fe55af | 5415 | goto fail; |
b18215fc RS |
5416 | else |
5417 | { | |
5418 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5419 | is the character at D, and S2 is the syntax of C2. */ | |
5420 | int c1, c2, s1, s2; | |
5421 | int pos1 = PTR_TO_OFFSET (d); | |
5d967c7a | 5422 | int charpos; |
fa9a63c5 | 5423 | |
b18215fc | 5424 | GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2); |
fa9a63c5 | 5425 | #ifdef emacs |
92432794 RS |
5426 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (pos1); |
5427 | UPDATE_SYNTAX_TABLE (charpos); | |
25fe55af | 5428 | #endif |
b18215fc | 5429 | s2 = SYNTAX (c2); |
25fe55af | 5430 | |
b18215fc RS |
5431 | /* Case 2: S2 is not Sword. */ |
5432 | if (s2 != Sword) | |
5433 | goto fail; | |
5434 | ||
5435 | /* Case 3: D is not at the beginning of string ... */ | |
5436 | if (!AT_STRINGS_BEG (d)) | |
5437 | { | |
5438 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5439 | #ifdef emacs | |
5d967c7a | 5440 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
25fe55af | 5441 | #endif |
b18215fc RS |
5442 | s1 = SYNTAX (c1); |
5443 | ||
5444 | /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5445 | returns 0. */ |
b18215fc RS |
5446 | if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
5447 | goto fail; | |
5448 | } | |
5449 | } | |
e318085a RS |
5450 | break; |
5451 | ||
b18215fc | 5452 | case wordend: |
25fe55af | 5453 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
b18215fc RS |
5454 | |
5455 | /* We FAIL in one of the following cases: */ | |
5456 | ||
5457 | /* Case 1: D is at the beginning of string. */ | |
5458 | if (AT_STRINGS_BEG (d)) | |
e318085a | 5459 | goto fail; |
b18215fc RS |
5460 | else |
5461 | { | |
5462 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5463 | is the character at D, and S2 is the syntax of C2. */ | |
5464 | int c1, c2, s1, s2; | |
5d967c7a RS |
5465 | int pos1 = PTR_TO_OFFSET (d); |
5466 | int charpos; | |
b18215fc RS |
5467 | |
5468 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5d967c7a | 5469 | #ifdef emacs |
92432794 RS |
5470 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (pos1 - 1); |
5471 | UPDATE_SYNTAX_TABLE (charpos); | |
5d967c7a | 5472 | #endif |
b18215fc RS |
5473 | s1 = SYNTAX (c1); |
5474 | ||
5475 | /* Case 2: S1 is not Sword. */ | |
5476 | if (s1 != Sword) | |
5477 | goto fail; | |
5478 | ||
5479 | /* Case 3: D is not at the end of string ... */ | |
5480 | if (!AT_STRINGS_END (d)) | |
5481 | { | |
5482 | GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2); | |
5d967c7a RS |
5483 | #ifdef emacs |
5484 | UPDATE_SYNTAX_TABLE_FORWARD (charpos); | |
5485 | #endif | |
b18215fc RS |
5486 | s2 = SYNTAX (c2); |
5487 | ||
5488 | /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5489 | returns 0. */ |
b18215fc | 5490 | if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
25fe55af | 5491 | goto fail; |
b18215fc RS |
5492 | } |
5493 | } | |
e318085a RS |
5494 | break; |
5495 | ||
b18215fc | 5496 | #ifdef emacs |
25fe55af RS |
5497 | case before_dot: |
5498 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
5d967c7a | 5499 | if (PTR_BYTE_POS ((unsigned char *) d) >= PT_BYTE) |
25fe55af RS |
5500 | goto fail; |
5501 | break; | |
b18215fc | 5502 | |
25fe55af RS |
5503 | case at_dot: |
5504 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
5d967c7a | 5505 | if (PTR_BYTE_POS ((unsigned char *) d) != PT_BYTE) |
25fe55af RS |
5506 | goto fail; |
5507 | break; | |
b18215fc | 5508 | |
25fe55af RS |
5509 | case after_dot: |
5510 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
5d967c7a | 5511 | if (PTR_BYTE_POS ((unsigned char *) d) <= PT_BYTE) |
25fe55af RS |
5512 | goto fail; |
5513 | break; | |
fa9a63c5 RM |
5514 | |
5515 | case syntaxspec: | |
25fe55af | 5516 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); |
fa9a63c5 RM |
5517 | mcnt = *p++; |
5518 | goto matchsyntax; | |
5519 | ||
25fe55af RS |
5520 | case wordchar: |
5521 | DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); | |
fa9a63c5 | 5522 | mcnt = (int) Sword; |
25fe55af | 5523 | matchsyntax: |
fa9a63c5 | 5524 | PREFETCH (); |
b18215fc RS |
5525 | #ifdef emacs |
5526 | { | |
92432794 | 5527 | int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (PTR_TO_OFFSET (d)); |
b18215fc RS |
5528 | UPDATE_SYNTAX_TABLE (pos1); |
5529 | } | |
25fe55af | 5530 | #endif |
b18215fc RS |
5531 | { |
5532 | int c, len; | |
5533 | ||
5534 | if (multibyte) | |
5535 | /* we must concern about multibyte form, ... */ | |
5536 | c = STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
5537 | else | |
5538 | /* everything should be handled as ASCII, even though it | |
5539 | looks like multibyte form. */ | |
5540 | c = *d, len = 1; | |
5541 | ||
5542 | if (SYNTAX (c) != (enum syntaxcode) mcnt) | |
fa9a63c5 | 5543 | goto fail; |
b18215fc RS |
5544 | d += len; |
5545 | } | |
25fe55af | 5546 | SET_REGS_MATCHED (); |
fa9a63c5 RM |
5547 | break; |
5548 | ||
5549 | case notsyntaxspec: | |
25fe55af | 5550 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); |
fa9a63c5 RM |
5551 | mcnt = *p++; |
5552 | goto matchnotsyntax; | |
5553 | ||
25fe55af RS |
5554 | case notwordchar: |
5555 | DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); | |
fa9a63c5 | 5556 | mcnt = (int) Sword; |
25fe55af | 5557 | matchnotsyntax: |
fa9a63c5 | 5558 | PREFETCH (); |
b18215fc RS |
5559 | #ifdef emacs |
5560 | { | |
92432794 | 5561 | int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (PTR_TO_OFFSET (d)); |
b18215fc RS |
5562 | UPDATE_SYNTAX_TABLE (pos1); |
5563 | } | |
25fe55af | 5564 | #endif |
b18215fc RS |
5565 | { |
5566 | int c, len; | |
5567 | ||
5568 | if (multibyte) | |
5569 | c = STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
5570 | else | |
5571 | c = *d, len = 1; | |
5572 | ||
5573 | if (SYNTAX (c) == (enum syntaxcode) mcnt) | |
fa9a63c5 | 5574 | goto fail; |
b18215fc RS |
5575 | d += len; |
5576 | } | |
5577 | SET_REGS_MATCHED (); | |
5578 | break; | |
5579 | ||
5580 | case categoryspec: | |
5581 | DEBUG_PRINT2 ("EXECUTING categoryspec %d.\n", *p); | |
5582 | mcnt = *p++; | |
5583 | PREFETCH (); | |
5584 | { | |
5585 | int c, len; | |
5586 | ||
5587 | if (multibyte) | |
5588 | c = STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
5589 | else | |
5590 | c = *d, len = 1; | |
5591 | ||
5592 | if (!CHAR_HAS_CATEGORY (c, mcnt)) | |
5593 | goto fail; | |
5594 | d += len; | |
5595 | } | |
fa9a63c5 | 5596 | SET_REGS_MATCHED (); |
e318085a | 5597 | break; |
fa9a63c5 | 5598 | |
b18215fc RS |
5599 | case notcategoryspec: |
5600 | DEBUG_PRINT2 ("EXECUTING notcategoryspec %d.\n", *p); | |
5601 | mcnt = *p++; | |
5602 | PREFETCH (); | |
5603 | { | |
5604 | int c, len; | |
5605 | ||
5606 | if (multibyte) | |
5607 | c = STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
5608 | else | |
5609 | c = *d, len = 1; | |
5610 | ||
5611 | if (CHAR_HAS_CATEGORY (c, mcnt)) | |
5612 | goto fail; | |
5613 | d += len; | |
5614 | } | |
5615 | SET_REGS_MATCHED (); | |
5616 | break; | |
5617 | ||
fa9a63c5 RM |
5618 | #else /* not emacs */ |
5619 | case wordchar: | |
b18215fc | 5620 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); |
fa9a63c5 | 5621 | PREFETCH (); |
b18215fc RS |
5622 | if (!WORDCHAR_P (d)) |
5623 | goto fail; | |
fa9a63c5 | 5624 | SET_REGS_MATCHED (); |
b18215fc | 5625 | d++; |
fa9a63c5 | 5626 | break; |
5e69f11e | 5627 | |
fa9a63c5 | 5628 | case notwordchar: |
b18215fc | 5629 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); |
fa9a63c5 RM |
5630 | PREFETCH (); |
5631 | if (WORDCHAR_P (d)) | |
b18215fc RS |
5632 | goto fail; |
5633 | SET_REGS_MATCHED (); | |
5634 | d++; | |
fa9a63c5 RM |
5635 | break; |
5636 | #endif /* not emacs */ | |
5e69f11e | 5637 | |
b18215fc RS |
5638 | default: |
5639 | abort (); | |
fa9a63c5 | 5640 | } |
b18215fc | 5641 | continue; /* Successfully executed one pattern command; keep going. */ |
fa9a63c5 RM |
5642 | |
5643 | ||
5644 | /* We goto here if a matching operation fails. */ | |
5645 | fail: | |
5646 | if (!FAIL_STACK_EMPTY ()) | |
b18215fc RS |
5647 | { /* A restart point is known. Restore to that state. */ |
5648 | DEBUG_PRINT1 ("\nFAIL:\n"); | |
5649 | POP_FAILURE_POINT (d, p, | |
5650 | lowest_active_reg, highest_active_reg, | |
5651 | regstart, regend, reg_info); | |
5652 | ||
5653 | /* If this failure point is a dummy, try the next one. */ | |
5654 | if (!p) | |
fa9a63c5 RM |
5655 | goto fail; |
5656 | ||
b18215fc | 5657 | /* If we failed to the end of the pattern, don't examine *p. */ |
fa9a63c5 | 5658 | assert (p <= pend); |
b18215fc RS |
5659 | if (p < pend) |
5660 | { | |
5661 | boolean is_a_jump_n = false; | |
5662 | ||
5663 | /* If failed to a backwards jump that's part of a repetition | |
5664 | loop, need to pop this failure point and use the next one. */ | |
5665 | switch ((re_opcode_t) *p) | |
5666 | { | |
5667 | case jump_n: | |
5668 | is_a_jump_n = true; | |
5669 | case maybe_pop_jump: | |
5670 | case pop_failure_jump: | |
5671 | case jump: | |
5672 | p1 = p + 1; | |
5673 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5674 | p1 += mcnt; | |
5675 | ||
5676 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
5677 | || (!is_a_jump_n | |
5678 | && (re_opcode_t) *p1 == on_failure_jump)) | |
5679 | goto fail; | |
5680 | break; | |
5681 | default: | |
5682 | /* do nothing */ ; | |
5683 | } | |
5684 | } | |
5685 | ||
5686 | if (d >= string1 && d <= end1) | |
fa9a63c5 | 5687 | dend = end_match_1; |
b18215fc | 5688 | } |
fa9a63c5 | 5689 | else |
b18215fc | 5690 | break; /* Matching at this starting point really fails. */ |
fa9a63c5 RM |
5691 | } /* for (;;) */ |
5692 | ||
5693 | if (best_regs_set) | |
5694 | goto restore_best_regs; | |
5695 | ||
5696 | FREE_VARIABLES (); | |
5697 | ||
b18215fc | 5698 | return -1; /* Failure to match. */ |
fa9a63c5 RM |
5699 | } /* re_match_2 */ |
5700 | \f | |
5701 | /* Subroutine definitions for re_match_2. */ | |
5702 | ||
5703 | ||
5704 | /* We are passed P pointing to a register number after a start_memory. | |
5e69f11e | 5705 | |
fa9a63c5 RM |
5706 | Return true if the pattern up to the corresponding stop_memory can |
5707 | match the empty string, and false otherwise. | |
5e69f11e | 5708 | |
fa9a63c5 RM |
5709 | If we find the matching stop_memory, sets P to point to one past its number. |
5710 | Otherwise, sets P to an undefined byte less than or equal to END. | |
5711 | ||
5712 | We don't handle duplicates properly (yet). */ | |
5713 | ||
5714 | static boolean | |
5715 | group_match_null_string_p (p, end, reg_info) | |
5716 | unsigned char **p, *end; | |
5717 | register_info_type *reg_info; | |
5718 | { | |
5719 | int mcnt; | |
5720 | /* Point to after the args to the start_memory. */ | |
5721 | unsigned char *p1 = *p + 2; | |
5e69f11e | 5722 | |
fa9a63c5 RM |
5723 | while (p1 < end) |
5724 | { | |
5725 | /* Skip over opcodes that can match nothing, and return true or | |
5726 | false, as appropriate, when we get to one that can't, or to the | |
b18215fc | 5727 | matching stop_memory. */ |
5e69f11e | 5728 | |
fa9a63c5 | 5729 | switch ((re_opcode_t) *p1) |
b18215fc RS |
5730 | { |
5731 | /* Could be either a loop or a series of alternatives. */ | |
5732 | case on_failure_jump: | |
5733 | p1++; | |
5734 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5e69f11e | 5735 | |
b18215fc | 5736 | /* If the next operation is not a jump backwards in the |
fa9a63c5 RM |
5737 | pattern. */ |
5738 | ||
5739 | if (mcnt >= 0) | |
5740 | { | |
b18215fc RS |
5741 | /* Go through the on_failure_jumps of the alternatives, |
5742 | seeing if any of the alternatives cannot match nothing. | |
5743 | The last alternative starts with only a jump, | |
5744 | whereas the rest start with on_failure_jump and end | |
5745 | with a jump, e.g., here is the pattern for `a|b|c': | |
fa9a63c5 | 5746 | |
b18215fc RS |
5747 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 |
5748 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
5749 | /exactn/1/c | |
fa9a63c5 | 5750 | |
b18215fc RS |
5751 | So, we have to first go through the first (n-1) |
5752 | alternatives and then deal with the last one separately. */ | |
fa9a63c5 RM |
5753 | |
5754 | ||
b18215fc RS |
5755 | /* Deal with the first (n-1) alternatives, which start |
5756 | with an on_failure_jump (see above) that jumps to right | |
5757 | past a jump_past_alt. */ | |
fa9a63c5 | 5758 | |
b18215fc RS |
5759 | while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) |
5760 | { | |
5761 | /* `mcnt' holds how many bytes long the alternative | |
5762 | is, including the ending `jump_past_alt' and | |
5763 | its number. */ | |
fa9a63c5 | 5764 | |
b18215fc RS |
5765 | if (!alt_match_null_string_p (p1, p1 + mcnt - 3, |
5766 | reg_info)) | |
5767 | return false; | |
fa9a63c5 | 5768 | |
b18215fc | 5769 | /* Move to right after this alternative, including the |
fa9a63c5 | 5770 | jump_past_alt. */ |
b18215fc | 5771 | p1 += mcnt; |
fa9a63c5 | 5772 | |
b18215fc RS |
5773 | /* Break if it's the beginning of an n-th alternative |
5774 | that doesn't begin with an on_failure_jump. */ | |
5775 | if ((re_opcode_t) *p1 != on_failure_jump) | |
5776 | break; | |
5e69f11e | 5777 | |
fa9a63c5 RM |
5778 | /* Still have to check that it's not an n-th |
5779 | alternative that starts with an on_failure_jump. */ | |
5780 | p1++; | |
b18215fc RS |
5781 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
5782 | if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
5783 | { | |
5784 | /* Get to the beginning of the n-th alternative. */ | |
5785 | p1 -= 3; | |
5786 | break; | |
5787 | } | |
5788 | } | |
fa9a63c5 | 5789 | |
b18215fc RS |
5790 | /* Deal with the last alternative: go back and get number |
5791 | of the `jump_past_alt' just before it. `mcnt' contains | |
5792 | the length of the alternative. */ | |
5793 | EXTRACT_NUMBER (mcnt, p1 - 2); | |
fa9a63c5 | 5794 | |
b18215fc RS |
5795 | if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) |
5796 | return false; | |
fa9a63c5 | 5797 | |
b18215fc RS |
5798 | p1 += mcnt; /* Get past the n-th alternative. */ |
5799 | } /* if mcnt > 0 */ | |
5800 | break; | |
fa9a63c5 | 5801 | |
5e69f11e | 5802 | |
b18215fc | 5803 | case stop_memory: |
fa9a63c5 | 5804 | assert (p1[1] == **p); |
b18215fc RS |
5805 | *p = p1 + 2; |
5806 | return true; | |
fa9a63c5 | 5807 | |
5e69f11e | 5808 | |
b18215fc RS |
5809 | default: |
5810 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
5811 | return false; | |
5812 | } | |
fa9a63c5 RM |
5813 | } /* while p1 < end */ |
5814 | ||
5815 | return false; | |
5816 | } /* group_match_null_string_p */ | |
5817 | ||
5818 | ||
5819 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
5820 | It expects P to be the first byte of a single alternative and END one | |
5821 | byte past the last. The alternative can contain groups. */ | |
5e69f11e | 5822 | |
fa9a63c5 RM |
5823 | static boolean |
5824 | alt_match_null_string_p (p, end, reg_info) | |
5825 | unsigned char *p, *end; | |
5826 | register_info_type *reg_info; | |
5827 | { | |
5828 | int mcnt; | |
5829 | unsigned char *p1 = p; | |
5e69f11e | 5830 | |
fa9a63c5 RM |
5831 | while (p1 < end) |
5832 | { | |
5e69f11e | 5833 | /* Skip over opcodes that can match nothing, and break when we get |
b18215fc | 5834 | to one that can't. */ |
5e69f11e | 5835 | |
fa9a63c5 | 5836 | switch ((re_opcode_t) *p1) |
b18215fc RS |
5837 | { |
5838 | /* It's a loop. */ | |
5839 | case on_failure_jump: | |
5840 | p1++; | |
5841 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5842 | p1 += mcnt; | |
5843 | break; | |
5e69f11e RM |
5844 | |
5845 | default: | |
b18215fc RS |
5846 | if (!common_op_match_null_string_p (&p1, end, reg_info)) |
5847 | return false; | |
5848 | } | |
fa9a63c5 RM |
5849 | } /* while p1 < end */ |
5850 | ||
5851 | return true; | |
5852 | } /* alt_match_null_string_p */ | |
5853 | ||
5854 | ||
5855 | /* Deals with the ops common to group_match_null_string_p and | |
5e69f11e RM |
5856 | alt_match_null_string_p. |
5857 | ||
fa9a63c5 RM |
5858 | Sets P to one after the op and its arguments, if any. */ |
5859 | ||
5860 | static boolean | |
5861 | common_op_match_null_string_p (p, end, reg_info) | |
5862 | unsigned char **p, *end; | |
5863 | register_info_type *reg_info; | |
5864 | { | |
5865 | int mcnt; | |
5866 | boolean ret; | |
5867 | int reg_no; | |
5868 | unsigned char *p1 = *p; | |
5869 | ||
5870 | switch ((re_opcode_t) *p1++) | |
5871 | { | |
5872 | case no_op: | |
5873 | case begline: | |
5874 | case endline: | |
5875 | case begbuf: | |
5876 | case endbuf: | |
5877 | case wordbeg: | |
5878 | case wordend: | |
5879 | case wordbound: | |
5880 | case notwordbound: | |
5881 | #ifdef emacs | |
5882 | case before_dot: | |
5883 | case at_dot: | |
5884 | case after_dot: | |
5885 | #endif | |
5886 | break; | |
5887 | ||
5888 | case start_memory: | |
5889 | reg_no = *p1; | |
5890 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
5891 | ret = group_match_null_string_p (&p1, end, reg_info); | |
5e69f11e | 5892 | |
fa9a63c5 | 5893 | /* Have to set this here in case we're checking a group which |
b18215fc | 5894 | contains a group and a back reference to it. */ |
fa9a63c5 RM |
5895 | |
5896 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
b18215fc | 5897 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; |
fa9a63c5 RM |
5898 | |
5899 | if (!ret) | |
b18215fc | 5900 | return false; |
fa9a63c5 | 5901 | break; |
5e69f11e | 5902 | |
b18215fc | 5903 | /* If this is an optimized succeed_n for zero times, make the jump. */ |
fa9a63c5 RM |
5904 | case jump: |
5905 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5906 | if (mcnt >= 0) | |
b18215fc | 5907 | p1 += mcnt; |
fa9a63c5 | 5908 | else |
b18215fc | 5909 | return false; |
fa9a63c5 RM |
5910 | break; |
5911 | ||
5912 | case succeed_n: | |
b18215fc | 5913 | /* Get to the number of times to succeed. */ |
5e69f11e | 5914 | p1 += 2; |
fa9a63c5 RM |
5915 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
5916 | ||
5917 | if (mcnt == 0) | |
b18215fc RS |
5918 | { |
5919 | p1 -= 4; | |
5920 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5921 | p1 += mcnt; | |
5922 | } | |
fa9a63c5 | 5923 | else |
b18215fc | 5924 | return false; |
fa9a63c5 RM |
5925 | break; |
5926 | ||
5e69f11e | 5927 | case duplicate: |
fa9a63c5 | 5928 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) |
b18215fc | 5929 | return false; |
fa9a63c5 RM |
5930 | break; |
5931 | ||
5932 | case set_number_at: | |
5933 | p1 += 4; | |
5934 | ||
5935 | default: | |
5936 | /* All other opcodes mean we cannot match the empty string. */ | |
5937 | return false; | |
5938 | } | |
5939 | ||
5940 | *p = p1; | |
5941 | return true; | |
5942 | } /* common_op_match_null_string_p */ | |
5943 | ||
5944 | ||
5945 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
5946 | bytes; nonzero otherwise. */ | |
5e69f11e | 5947 | |
fa9a63c5 RM |
5948 | static int |
5949 | bcmp_translate (s1, s2, len, translate) | |
5950 | unsigned char *s1, *s2; | |
5951 | register int len; | |
6676cb1c | 5952 | RE_TRANSLATE_TYPE translate; |
fa9a63c5 RM |
5953 | { |
5954 | register unsigned char *p1 = s1, *p2 = s2; | |
e934739e RS |
5955 | unsigned char *p1_end = s1 + len; |
5956 | unsigned char *p2_end = s2 + len; | |
5957 | ||
5958 | while (p1 != p1_end && p2 != p2_end) | |
fa9a63c5 | 5959 | { |
e934739e RS |
5960 | int p1_charlen, p2_charlen; |
5961 | int p1_ch, p2_ch; | |
5962 | ||
5963 | p1_ch = STRING_CHAR_AND_LENGTH (p1, p1_end - p1, p1_charlen); | |
5964 | p2_ch = STRING_CHAR_AND_LENGTH (p2, p2_end - p2, p2_charlen); | |
5965 | ||
5966 | if (RE_TRANSLATE (translate, p1_ch) | |
5967 | != RE_TRANSLATE (translate, p2_ch)) | |
bc192b5b | 5968 | return 1; |
e934739e RS |
5969 | |
5970 | p1 += p1_charlen, p2 += p2_charlen; | |
fa9a63c5 | 5971 | } |
e934739e RS |
5972 | |
5973 | if (p1 != p1_end || p2 != p2_end) | |
5974 | return 1; | |
5975 | ||
fa9a63c5 RM |
5976 | return 0; |
5977 | } | |
5978 | \f | |
5979 | /* Entry points for GNU code. */ | |
5980 | ||
5981 | /* re_compile_pattern is the GNU regular expression compiler: it | |
5982 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
5983 | Returns 0 if the pattern was valid, otherwise an error string. | |
5e69f11e | 5984 | |
fa9a63c5 RM |
5985 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
5986 | are set in BUFP on entry. | |
5e69f11e | 5987 | |
b18215fc | 5988 | We call regex_compile to do the actual compilation. */ |
fa9a63c5 RM |
5989 | |
5990 | const char * | |
5991 | re_compile_pattern (pattern, length, bufp) | |
5992 | const char *pattern; | |
5993 | int length; | |
5994 | struct re_pattern_buffer *bufp; | |
5995 | { | |
5996 | reg_errcode_t ret; | |
5e69f11e | 5997 | |
fa9a63c5 RM |
5998 | /* GNU code is written to assume at least RE_NREGS registers will be set |
5999 | (and at least one extra will be -1). */ | |
6000 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5e69f11e | 6001 | |
fa9a63c5 RM |
6002 | /* And GNU code determines whether or not to get register information |
6003 | by passing null for the REGS argument to re_match, etc., not by | |
6004 | setting no_sub. */ | |
6005 | bufp->no_sub = 0; | |
5e69f11e | 6006 | |
b18215fc | 6007 | /* Match anchors at newline. */ |
fa9a63c5 | 6008 | bufp->newline_anchor = 1; |
5e69f11e | 6009 | |
fa9a63c5 RM |
6010 | ret = regex_compile (pattern, length, re_syntax_options, bufp); |
6011 | ||
6012 | if (!ret) | |
6013 | return NULL; | |
6014 | return gettext (re_error_msgid[(int) ret]); | |
5e69f11e | 6015 | } |
fa9a63c5 | 6016 | \f |
b18215fc RS |
6017 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
6018 | them unless specifically requested. */ | |
fa9a63c5 | 6019 | |
0c085854 | 6020 | #if defined (_REGEX_RE_COMP) || defined (_LIBC) |
fa9a63c5 RM |
6021 | |
6022 | /* BSD has one and only one pattern buffer. */ | |
6023 | static struct re_pattern_buffer re_comp_buf; | |
6024 | ||
6025 | char * | |
48afdd44 RM |
6026 | #ifdef _LIBC |
6027 | /* Make these definitions weak in libc, so POSIX programs can redefine | |
6028 | these names if they don't use our functions, and still use | |
6029 | regcomp/regexec below without link errors. */ | |
6030 | weak_function | |
6031 | #endif | |
fa9a63c5 RM |
6032 | re_comp (s) |
6033 | const char *s; | |
6034 | { | |
6035 | reg_errcode_t ret; | |
5e69f11e | 6036 | |
fa9a63c5 RM |
6037 | if (!s) |
6038 | { | |
6039 | if (!re_comp_buf.buffer) | |
6040 | return gettext ("No previous regular expression"); | |
6041 | return 0; | |
6042 | } | |
6043 | ||
6044 | if (!re_comp_buf.buffer) | |
6045 | { | |
6046 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
6047 | if (re_comp_buf.buffer == NULL) | |
b18215fc | 6048 | return gettext (re_error_msgid[(int) REG_ESPACE]); |
fa9a63c5 RM |
6049 | re_comp_buf.allocated = 200; |
6050 | ||
6051 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
6052 | if (re_comp_buf.fastmap == NULL) | |
6053 | return gettext (re_error_msgid[(int) REG_ESPACE]); | |
6054 | } | |
6055 | ||
6056 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
6057 | don't need to initialize the pattern buffer fields which affect it. */ | |
6058 | ||
b18215fc | 6059 | /* Match anchors at newlines. */ |
fa9a63c5 RM |
6060 | re_comp_buf.newline_anchor = 1; |
6061 | ||
6062 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
5e69f11e | 6063 | |
fa9a63c5 RM |
6064 | if (!ret) |
6065 | return NULL; | |
6066 | ||
6067 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
6068 | return (char *) gettext (re_error_msgid[(int) ret]); | |
6069 | } | |
6070 | ||
6071 | ||
6072 | int | |
48afdd44 RM |
6073 | #ifdef _LIBC |
6074 | weak_function | |
6075 | #endif | |
fa9a63c5 RM |
6076 | re_exec (s) |
6077 | const char *s; | |
6078 | { | |
6079 | const int len = strlen (s); | |
6080 | return | |
6081 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
6082 | } | |
6083 | #endif /* _REGEX_RE_COMP */ | |
6084 | \f | |
6085 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
6086 | ||
6087 | #ifndef emacs | |
6088 | ||
6089 | /* regcomp takes a regular expression as a string and compiles it. | |
6090 | ||
b18215fc | 6091 | PREG is a regex_t *. We do not expect any fields to be initialized, |
fa9a63c5 RM |
6092 | since POSIX says we shouldn't. Thus, we set |
6093 | ||
6094 | `buffer' to the compiled pattern; | |
6095 | `used' to the length of the compiled pattern; | |
6096 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
6097 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
6098 | RE_SYNTAX_POSIX_BASIC; | |
6099 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
6100 | `fastmap' and `fastmap_accurate' to zero; | |
6101 | `re_nsub' to the number of subexpressions in PATTERN. | |
6102 | ||
6103 | PATTERN is the address of the pattern string. | |
6104 | ||
6105 | CFLAGS is a series of bits which affect compilation. | |
6106 | ||
6107 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
6108 | use POSIX basic syntax. | |
6109 | ||
6110 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
6111 | Also, regexec will try a match beginning after every newline. | |
6112 | ||
6113 | If REG_ICASE is set, then we considers upper- and lowercase | |
6114 | versions of letters to be equivalent when matching. | |
6115 | ||
6116 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
6117 | routine will report only success or failure, and nothing about the | |
6118 | registers. | |
6119 | ||
b18215fc | 6120 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
fa9a63c5 RM |
6121 | the return codes and their meanings.) */ |
6122 | ||
6123 | int | |
6124 | regcomp (preg, pattern, cflags) | |
6125 | regex_t *preg; | |
5e69f11e | 6126 | const char *pattern; |
fa9a63c5 RM |
6127 | int cflags; |
6128 | { | |
6129 | reg_errcode_t ret; | |
6130 | unsigned syntax | |
6131 | = (cflags & REG_EXTENDED) ? | |
6132 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
6133 | ||
6134 | /* regex_compile will allocate the space for the compiled pattern. */ | |
6135 | preg->buffer = 0; | |
6136 | preg->allocated = 0; | |
6137 | preg->used = 0; | |
5e69f11e | 6138 | |
fa9a63c5 RM |
6139 | /* Don't bother to use a fastmap when searching. This simplifies the |
6140 | REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
6141 | characters after newlines into the fastmap. This way, we just try | |
6142 | every character. */ | |
6143 | preg->fastmap = 0; | |
5e69f11e | 6144 | |
fa9a63c5 RM |
6145 | if (cflags & REG_ICASE) |
6146 | { | |
6147 | unsigned i; | |
5e69f11e | 6148 | |
6676cb1c RS |
6149 | preg->translate |
6150 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
6151 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
fa9a63c5 | 6152 | if (preg->translate == NULL) |
b18215fc | 6153 | return (int) REG_ESPACE; |
fa9a63c5 RM |
6154 | |
6155 | /* Map uppercase characters to corresponding lowercase ones. */ | |
6156 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
b18215fc | 6157 | preg->translate[i] = ISUPPER (i) ? tolower (i) : i; |
fa9a63c5 RM |
6158 | } |
6159 | else | |
6160 | preg->translate = NULL; | |
6161 | ||
6162 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
6163 | if (cflags & REG_NEWLINE) | |
6164 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
6165 | syntax &= ~RE_DOT_NEWLINE; | |
6166 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
b18215fc | 6167 | /* It also changes the matching behavior. */ |
fa9a63c5 RM |
6168 | preg->newline_anchor = 1; |
6169 | } | |
6170 | else | |
6171 | preg->newline_anchor = 0; | |
6172 | ||
6173 | preg->no_sub = !!(cflags & REG_NOSUB); | |
6174 | ||
5e69f11e | 6175 | /* POSIX says a null character in the pattern terminates it, so we |
fa9a63c5 RM |
6176 | can use strlen here in compiling the pattern. */ |
6177 | ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
5e69f11e | 6178 | |
fa9a63c5 RM |
6179 | /* POSIX doesn't distinguish between an unmatched open-group and an |
6180 | unmatched close-group: both are REG_EPAREN. */ | |
6181 | if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
5e69f11e | 6182 | |
fa9a63c5 RM |
6183 | return (int) ret; |
6184 | } | |
6185 | ||
6186 | ||
6187 | /* regexec searches for a given pattern, specified by PREG, in the | |
6188 | string STRING. | |
5e69f11e | 6189 | |
fa9a63c5 | 6190 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
b18215fc | 6191 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
fa9a63c5 RM |
6192 | least NMATCH elements, and we set them to the offsets of the |
6193 | corresponding matched substrings. | |
5e69f11e | 6194 | |
fa9a63c5 RM |
6195 | EFLAGS specifies `execution flags' which affect matching: if |
6196 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
6197 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5e69f11e | 6198 | |
fa9a63c5 RM |
6199 | We return 0 if we find a match and REG_NOMATCH if not. */ |
6200 | ||
6201 | int | |
6202 | regexec (preg, string, nmatch, pmatch, eflags) | |
6203 | const regex_t *preg; | |
5e69f11e RM |
6204 | const char *string; |
6205 | size_t nmatch; | |
6206 | regmatch_t pmatch[]; | |
fa9a63c5 RM |
6207 | int eflags; |
6208 | { | |
6209 | int ret; | |
6210 | struct re_registers regs; | |
6211 | regex_t private_preg; | |
6212 | int len = strlen (string); | |
6213 | boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
6214 | ||
6215 | private_preg = *preg; | |
5e69f11e | 6216 | |
fa9a63c5 RM |
6217 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
6218 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5e69f11e | 6219 | |
fa9a63c5 RM |
6220 | /* The user has told us exactly how many registers to return |
6221 | information about, via `nmatch'. We have to pass that on to the | |
b18215fc | 6222 | matching routines. */ |
fa9a63c5 | 6223 | private_preg.regs_allocated = REGS_FIXED; |
5e69f11e | 6224 | |
fa9a63c5 RM |
6225 | if (want_reg_info) |
6226 | { | |
6227 | regs.num_regs = nmatch; | |
6228 | regs.start = TALLOC (nmatch, regoff_t); | |
6229 | regs.end = TALLOC (nmatch, regoff_t); | |
6230 | if (regs.start == NULL || regs.end == NULL) | |
b18215fc | 6231 | return (int) REG_NOMATCH; |
fa9a63c5 RM |
6232 | } |
6233 | ||
6234 | /* Perform the searching operation. */ | |
6235 | ret = re_search (&private_preg, string, len, | |
b18215fc RS |
6236 | /* start: */ 0, /* range: */ len, |
6237 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5e69f11e | 6238 | |
fa9a63c5 RM |
6239 | /* Copy the register information to the POSIX structure. */ |
6240 | if (want_reg_info) | |
6241 | { | |
6242 | if (ret >= 0) | |
b18215fc RS |
6243 | { |
6244 | unsigned r; | |
fa9a63c5 | 6245 | |
b18215fc RS |
6246 | for (r = 0; r < nmatch; r++) |
6247 | { | |
6248 | pmatch[r].rm_so = regs.start[r]; | |
6249 | pmatch[r].rm_eo = regs.end[r]; | |
6250 | } | |
6251 | } | |
fa9a63c5 | 6252 | |
b18215fc | 6253 | /* If we needed the temporary register info, free the space now. */ |
fa9a63c5 RM |
6254 | free (regs.start); |
6255 | free (regs.end); | |
6256 | } | |
6257 | ||
6258 | /* We want zero return to mean success, unlike `re_search'. */ | |
6259 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
6260 | } | |
6261 | ||
6262 | ||
6263 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
6264 | from either regcomp or regexec. We don't use PREG here. */ | |
6265 | ||
6266 | size_t | |
6267 | regerror (errcode, preg, errbuf, errbuf_size) | |
6268 | int errcode; | |
6269 | const regex_t *preg; | |
6270 | char *errbuf; | |
6271 | size_t errbuf_size; | |
6272 | { | |
6273 | const char *msg; | |
6274 | size_t msg_size; | |
6275 | ||
6276 | if (errcode < 0 | |
6277 | || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) | |
5e69f11e | 6278 | /* Only error codes returned by the rest of the code should be passed |
b18215fc | 6279 | to this routine. If we are given anything else, or if other regex |
fa9a63c5 RM |
6280 | code generates an invalid error code, then the program has a bug. |
6281 | Dump core so we can fix it. */ | |
6282 | abort (); | |
6283 | ||
6284 | msg = gettext (re_error_msgid[errcode]); | |
6285 | ||
6286 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
5e69f11e | 6287 | |
fa9a63c5 RM |
6288 | if (errbuf_size != 0) |
6289 | { | |
6290 | if (msg_size > errbuf_size) | |
b18215fc RS |
6291 | { |
6292 | strncpy (errbuf, msg, errbuf_size - 1); | |
6293 | errbuf[errbuf_size - 1] = 0; | |
6294 | } | |
fa9a63c5 | 6295 | else |
b18215fc | 6296 | strcpy (errbuf, msg); |
fa9a63c5 RM |
6297 | } |
6298 | ||
6299 | return msg_size; | |
6300 | } | |
6301 | ||
6302 | ||
6303 | /* Free dynamically allocated space used by PREG. */ | |
6304 | ||
6305 | void | |
6306 | regfree (preg) | |
6307 | regex_t *preg; | |
6308 | { | |
6309 | if (preg->buffer != NULL) | |
6310 | free (preg->buffer); | |
6311 | preg->buffer = NULL; | |
5e69f11e | 6312 | |
fa9a63c5 RM |
6313 | preg->allocated = 0; |
6314 | preg->used = 0; | |
6315 | ||
6316 | if (preg->fastmap != NULL) | |
6317 | free (preg->fastmap); | |
6318 | preg->fastmap = NULL; | |
6319 | preg->fastmap_accurate = 0; | |
6320 | ||
6321 | if (preg->translate != NULL) | |
6322 | free (preg->translate); | |
6323 | preg->translate = NULL; | |
6324 | } | |
6325 | ||
6326 | #endif /* not emacs */ |