|
2
|
1 /* Extended regular expression matching and search library, |
|
47
|
2 version 0.12. |
|
2
|
3 (Implements POSIX draft P10003.2/D11.2, except for |
|
|
4 internationalization features.) |
|
|
5 |
|
34
|
6 Copyright (C) 1993 Free Software Foundation, Inc. |
|
2
|
7 |
|
|
8 This program is free software; you can redistribute it and/or modify |
|
|
9 it under the terms of the GNU General Public License as published by |
|
|
10 the Free Software Foundation; either version 2, or (at your option) |
|
|
11 any later version. |
|
|
12 |
|
|
13 This program is distributed in the hope that it will be useful, |
|
|
14 but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
|
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
|
|
16 GNU General Public License for more details. |
|
|
17 |
|
|
18 You should have received a copy of the GNU General Public License |
|
|
19 along with this program; if not, write to the Free Software |
|
|
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ |
|
|
21 |
|
|
22 /* AIX requires this to be the first thing in the file. */ |
|
|
23 #if defined (_AIX) && !defined (REGEX_MALLOC) |
|
|
24 #pragma alloca |
|
|
25 #endif |
|
|
26 |
|
|
27 #define _GNU_SOURCE |
|
|
28 |
|
124
|
29 #ifdef HAVE_CONFIG_H |
|
|
30 #if defined (CONFIG_BROKETS) |
|
|
31 /* We use <config.h> instead of "config.h" so that a compilation |
|
|
32 using -I. -I$srcdir will use ./config.h rather than $srcdir/config.h |
|
|
33 (which it would do because it found this file in $srcdir). */ |
|
|
34 #include <config.h> |
|
|
35 #else |
|
|
36 #include "config.h" |
|
|
37 #endif |
|
|
38 #endif |
|
|
39 |
|
2
|
40 /* We need this for `regex.h', and perhaps for the Emacs include files. */ |
|
|
41 #include <sys/types.h> |
|
|
42 |
|
29
|
43 #ifdef HAVE_CONFIG_H |
|
26
|
44 #include "config.h" |
|
|
45 #endif |
|
|
46 |
|
2
|
47 /* The `emacs' switch turns on certain matching commands |
|
|
48 that make sense only in Emacs. */ |
|
|
49 #ifdef emacs |
|
|
50 |
|
|
51 #include "lisp.h" |
|
|
52 #include "buffer.h" |
|
|
53 #include "syntax.h" |
|
|
54 |
|
|
55 /* Emacs uses `NULL' as a predicate. */ |
|
|
56 #undef NULL |
|
|
57 |
|
|
58 #else /* not emacs */ |
|
|
59 |
|
94
|
60 #ifdef STDC_HEADERS |
|
|
61 #include <stdlib.h> |
|
|
62 #else |
|
|
63 char *malloc (); |
|
|
64 char *realloc (); |
|
|
65 #endif |
|
|
66 |
|
|
67 |
|
2
|
68 /* We used to test for `BSTRING' here, but only GCC and Emacs define |
|
|
69 `BSTRING', as far as I know, and neither of them use this code. */ |
|
24
|
70 #if HAVE_STRING_H || STDC_HEADERS |
|
2
|
71 #include <string.h> |
|
23
|
72 #ifndef bcmp |
|
2
|
73 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n)) |
|
23
|
74 #endif |
|
|
75 #ifndef bcopy |
|
2
|
76 #define bcopy(s, d, n) memcpy ((d), (s), (n)) |
|
23
|
77 #endif |
|
|
78 #ifndef bzero |
|
2
|
79 #define bzero(s, n) memset ((s), 0, (n)) |
|
23
|
80 #endif |
|
2
|
81 #else |
|
|
82 #include <strings.h> |
|
|
83 #endif |
|
|
84 |
|
|
85 /* Define the syntax stuff for \<, \>, etc. */ |
|
|
86 |
|
|
87 /* This must be nonzero for the wordchar and notwordchar pattern |
|
|
88 commands in re_match_2. */ |
|
|
89 #ifndef Sword |
|
|
90 #define Sword 1 |
|
|
91 #endif |
|
|
92 |
|
|
93 #ifdef SYNTAX_TABLE |
|
|
94 |
|
|
95 extern char *re_syntax_table; |
|
|
96 |
|
|
97 #else /* not SYNTAX_TABLE */ |
|
|
98 |
|
|
99 /* How many characters in the character set. */ |
|
|
100 #define CHAR_SET_SIZE 256 |
|
|
101 |
|
|
102 static char re_syntax_table[CHAR_SET_SIZE]; |
|
|
103 |
|
|
104 static void |
|
|
105 init_syntax_once () |
|
|
106 { |
|
|
107 register int c; |
|
|
108 static int done = 0; |
|
|
109 |
|
|
110 if (done) |
|
|
111 return; |
|
|
112 |
|
|
113 bzero (re_syntax_table, sizeof re_syntax_table); |
|
|
114 |
|
|
115 for (c = 'a'; c <= 'z'; c++) |
|
|
116 re_syntax_table[c] = Sword; |
|
|
117 |
|
|
118 for (c = 'A'; c <= 'Z'; c++) |
|
|
119 re_syntax_table[c] = Sword; |
|
|
120 |
|
|
121 for (c = '0'; c <= '9'; c++) |
|
|
122 re_syntax_table[c] = Sword; |
|
|
123 |
|
|
124 re_syntax_table['_'] = Sword; |
|
|
125 |
|
|
126 done = 1; |
|
|
127 } |
|
|
128 |
|
|
129 #endif /* not SYNTAX_TABLE */ |
|
|
130 |
|
|
131 #define SYNTAX(c) re_syntax_table[c] |
|
|
132 |
|
|
133 #endif /* not emacs */ |
|
|
134 |
|
|
135 /* Get the interface, including the syntax bits. */ |
|
|
136 #include "regex.h" |
|
|
137 |
|
|
138 /* isalpha etc. are used for the character classes. */ |
|
|
139 #include <ctype.h> |
|
28
|
140 |
|
54
|
141 /* Jim Meyering writes: |
|
|
142 |
|
|
143 "... Some ctype macros are valid only for character codes that |
|
|
144 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when |
|
|
145 using /bin/cc or gcc but without giving an ansi option). So, all |
|
|
146 ctype uses should be through macros like ISPRINT... If |
|
|
147 STDC_HEADERS is defined, then autoconf has verified that the ctype |
|
|
148 macros don't need to be guarded with references to isascii. ... |
|
|
149 Defining isascii to 1 should let any compiler worth its salt |
|
|
150 eliminate the && through constant folding." */ |
|
|
151 #if ! defined (isascii) || defined (STDC_HEADERS) |
|
|
152 #undef isascii |
|
28
|
153 #define isascii(c) 1 |
|
|
154 #endif |
|
|
155 |
|
|
156 #ifdef isblank |
|
|
157 #define ISBLANK(c) (isascii (c) && isblank (c)) |
|
|
158 #else |
|
|
159 #define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
|
2
|
160 #endif |
|
28
|
161 #ifdef isgraph |
|
|
162 #define ISGRAPH(c) (isascii (c) && isgraph (c)) |
|
|
163 #else |
|
|
164 #define ISGRAPH(c) (isascii (c) && isprint (c) && !isspace (c)) |
|
2
|
165 #endif |
|
|
166 |
|
28
|
167 #define ISPRINT(c) (isascii (c) && isprint (c)) |
|
|
168 #define ISDIGIT(c) (isascii (c) && isdigit (c)) |
|
|
169 #define ISALNUM(c) (isascii (c) && isalnum (c)) |
|
|
170 #define ISALPHA(c) (isascii (c) && isalpha (c)) |
|
|
171 #define ISCNTRL(c) (isascii (c) && iscntrl (c)) |
|
|
172 #define ISLOWER(c) (isascii (c) && islower (c)) |
|
|
173 #define ISPUNCT(c) (isascii (c) && ispunct (c)) |
|
|
174 #define ISSPACE(c) (isascii (c) && isspace (c)) |
|
|
175 #define ISUPPER(c) (isascii (c) && isupper (c)) |
|
|
176 #define ISXDIGIT(c) (isascii (c) && isxdigit (c)) |
|
|
177 |
|
2
|
178 #ifndef NULL |
|
|
179 #define NULL 0 |
|
|
180 #endif |
|
|
181 |
|
|
182 /* We remove any previous definition of `SIGN_EXTEND_CHAR', |
|
|
183 since ours (we hope) works properly with all combinations of |
|
|
184 machines, compilers, `char' and `unsigned char' argument types. |
|
|
185 (Per Bothner suggested the basic approach.) */ |
|
|
186 #undef SIGN_EXTEND_CHAR |
|
|
187 #if __STDC__ |
|
|
188 #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
|
23
|
189 #else /* not __STDC__ */ |
|
2
|
190 /* As in Harbison and Steele. */ |
|
|
191 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
|
|
192 #endif |
|
|
193 |
|
|
194 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we |
|
|
195 use `alloca' instead of `malloc'. This is because using malloc in |
|
|
196 re_search* or re_match* could cause memory leaks when C-g is used in |
|
|
197 Emacs; also, malloc is slower and causes storage fragmentation. On |
|
|
198 the other hand, malloc is more portable, and easier to debug. |
|
|
199 |
|
|
200 Because we sometimes use alloca, some routines have to be macros, |
|
|
201 not functions -- `alloca'-allocated space disappears at the end of the |
|
|
202 function it is called in. */ |
|
|
203 |
|
|
204 #ifdef REGEX_MALLOC |
|
|
205 |
|
|
206 #define REGEX_ALLOCATE malloc |
|
|
207 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) |
|
|
208 |
|
|
209 #else /* not REGEX_MALLOC */ |
|
|
210 |
|
|
211 /* Emacs already defines alloca, sometimes. */ |
|
|
212 #ifndef alloca |
|
|
213 |
|
|
214 /* Make alloca work the best possible way. */ |
|
|
215 #ifdef __GNUC__ |
|
|
216 #define alloca __builtin_alloca |
|
|
217 #else /* not __GNUC__ */ |
|
|
218 #if HAVE_ALLOCA_H |
|
|
219 #include <alloca.h> |
|
|
220 #else /* not __GNUC__ or HAVE_ALLOCA_H */ |
|
|
221 #ifndef _AIX /* Already did AIX, up at the top. */ |
|
|
222 char *alloca (); |
|
|
223 #endif /* not _AIX */ |
|
|
224 #endif /* not HAVE_ALLOCA_H */ |
|
|
225 #endif /* not __GNUC__ */ |
|
|
226 |
|
|
227 #endif /* not alloca */ |
|
|
228 |
|
|
229 #define REGEX_ALLOCATE alloca |
|
|
230 |
|
|
231 /* Assumes a `char *destination' variable. */ |
|
|
232 #define REGEX_REALLOCATE(source, osize, nsize) \ |
|
|
233 (destination = (char *) alloca (nsize), \ |
|
|
234 bcopy (source, destination, osize), \ |
|
|
235 destination) |
|
|
236 |
|
|
237 #endif /* not REGEX_MALLOC */ |
|
|
238 |
|
|
239 |
|
|
240 /* True if `size1' is non-NULL and PTR is pointing anywhere inside |
|
|
241 `string1' or just past its end. This works if PTR is NULL, which is |
|
|
242 a good thing. */ |
|
|
243 #define FIRST_STRING_P(ptr) \ |
|
|
244 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
|
|
245 |
|
|
246 /* (Re)Allocate N items of type T using malloc, or fail. */ |
|
|
247 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) |
|
|
248 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) |
|
87
|
249 #define RETALLOC_IF(addr, n, t) \ |
|
|
250 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) |
|
2
|
251 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) |
|
|
252 |
|
|
253 #define BYTEWIDTH 8 /* In bits. */ |
|
|
254 |
|
|
255 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) |
|
|
256 |
|
|
257 #define MAX(a, b) ((a) > (b) ? (a) : (b)) |
|
|
258 #define MIN(a, b) ((a) < (b) ? (a) : (b)) |
|
|
259 |
|
|
260 typedef char boolean; |
|
|
261 #define false 0 |
|
|
262 #define true 1 |
|
|
263 |
|
|
264 /* These are the command codes that appear in compiled regular |
|
|
265 expressions. Some opcodes are followed by argument bytes. A |
|
|
266 command code can specify any interpretation whatsoever for its |
|
|
267 arguments. Zero bytes may appear in the compiled regular expression. |
|
|
268 |
|
|
269 The value of `exactn' is needed in search.c (search_buffer) in Emacs. |
|
|
270 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of |
|
|
271 `exactn' we use here must also be 1. */ |
|
|
272 |
|
|
273 typedef enum |
|
|
274 { |
|
|
275 no_op = 0, |
|
|
276 |
|
|
277 /* Followed by one byte giving n, then by n literal bytes. */ |
|
|
278 exactn = 1, |
|
|
279 |
|
|
280 /* Matches any (more or less) character. */ |
|
|
281 anychar, |
|
|
282 |
|
|
283 /* Matches any one char belonging to specified set. First |
|
|
284 following byte is number of bitmap bytes. Then come bytes |
|
|
285 for a bitmap saying which chars are in. Bits in each byte |
|
|
286 are ordered low-bit-first. A character is in the set if its |
|
|
287 bit is 1. A character too large to have a bit in the map is |
|
|
288 automatically not in the set. */ |
|
|
289 charset, |
|
|
290 |
|
|
291 /* Same parameters as charset, but match any character that is |
|
|
292 not one of those specified. */ |
|
|
293 charset_not, |
|
|
294 |
|
|
295 /* Start remembering the text that is matched, for storing in a |
|
|
296 register. Followed by one byte with the register number, in |
|
|
297 the range 0 to one less than the pattern buffer's re_nsub |
|
|
298 field. Then followed by one byte with the number of groups |
|
|
299 inner to this one. (This last has to be part of the |
|
|
300 start_memory only because we need it in the on_failure_jump |
|
|
301 of re_match_2.) */ |
|
|
302 start_memory, |
|
|
303 |
|
|
304 /* Stop remembering the text that is matched and store it in a |
|
|
305 memory register. Followed by one byte with the register |
|
|
306 number, in the range 0 to one less than `re_nsub' in the |
|
|
307 pattern buffer, and one byte with the number of inner groups, |
|
|
308 just like `start_memory'. (We need the number of inner |
|
|
309 groups here because we don't have any easy way of finding the |
|
|
310 corresponding start_memory when we're at a stop_memory.) */ |
|
|
311 stop_memory, |
|
|
312 |
|
|
313 /* Match a duplicate of something remembered. Followed by one |
|
|
314 byte containing the register number. */ |
|
|
315 duplicate, |
|
|
316 |
|
|
317 /* Fail unless at beginning of line. */ |
|
|
318 begline, |
|
|
319 |
|
|
320 /* Fail unless at end of line. */ |
|
|
321 endline, |
|
|
322 |
|
|
323 /* Succeeds if at beginning of buffer (if emacs) or at beginning |
|
|
324 of string to be matched (if not). */ |
|
|
325 begbuf, |
|
|
326 |
|
|
327 /* Analogously, for end of buffer/string. */ |
|
|
328 endbuf, |
|
|
329 |
|
|
330 /* Followed by two byte relative address to which to jump. */ |
|
|
331 jump, |
|
|
332 |
|
|
333 /* Same as jump, but marks the end of an alternative. */ |
|
|
334 jump_past_alt, |
|
|
335 |
|
|
336 /* Followed by two-byte relative address of place to resume at |
|
|
337 in case of failure. */ |
|
|
338 on_failure_jump, |
|
|
339 |
|
|
340 /* Like on_failure_jump, but pushes a placeholder instead of the |
|
|
341 current string position when executed. */ |
|
|
342 on_failure_keep_string_jump, |
|
|
343 |
|
|
344 /* Throw away latest failure point and then jump to following |
|
|
345 two-byte relative address. */ |
|
|
346 pop_failure_jump, |
|
|
347 |
|
|
348 /* Change to pop_failure_jump if know won't have to backtrack to |
|
|
349 match; otherwise change to jump. This is used to jump |
|
|
350 back to the beginning of a repeat. If what follows this jump |
|
|
351 clearly won't match what the repeat does, such that we can be |
|
|
352 sure that there is no use backtracking out of repetitions |
|
|
353 already matched, then we change it to a pop_failure_jump. |
|
|
354 Followed by two-byte address. */ |
|
|
355 maybe_pop_jump, |
|
|
356 |
|
|
357 /* Jump to following two-byte address, and push a dummy failure |
|
|
358 point. This failure point will be thrown away if an attempt |
|
|
359 is made to use it for a failure. A `+' construct makes this |
|
|
360 before the first repeat. Also used as an intermediary kind |
|
|
361 of jump when compiling an alternative. */ |
|
|
362 dummy_failure_jump, |
|
|
363 |
|
|
364 /* Push a dummy failure point and continue. Used at the end of |
|
|
365 alternatives. */ |
|
|
366 push_dummy_failure, |
|
|
367 |
|
|
368 /* Followed by two-byte relative address and two-byte number n. |
|
|
369 After matching N times, jump to the address upon failure. */ |
|
|
370 succeed_n, |
|
|
371 |
|
|
372 /* Followed by two-byte relative address, and two-byte number n. |
|
|
373 Jump to the address N times, then fail. */ |
|
|
374 jump_n, |
|
|
375 |
|
|
376 /* Set the following two-byte relative address to the |
|
|
377 subsequent two-byte number. The address *includes* the two |
|
|
378 bytes of number. */ |
|
|
379 set_number_at, |
|
|
380 |
|
|
381 wordchar, /* Matches any word-constituent character. */ |
|
|
382 notwordchar, /* Matches any char that is not a word-constituent. */ |
|
|
383 |
|
|
384 wordbeg, /* Succeeds if at word beginning. */ |
|
|
385 wordend, /* Succeeds if at word end. */ |
|
|
386 |
|
|
387 wordbound, /* Succeeds if at a word boundary. */ |
|
|
388 notwordbound /* Succeeds if not at a word boundary. */ |
|
|
389 |
|
|
390 #ifdef emacs |
|
|
391 ,before_dot, /* Succeeds if before point. */ |
|
|
392 at_dot, /* Succeeds if at point. */ |
|
|
393 after_dot, /* Succeeds if after point. */ |
|
|
394 |
|
|
395 /* Matches any character whose syntax is specified. Followed by |
|
|
396 a byte which contains a syntax code, e.g., Sword. */ |
|
|
397 syntaxspec, |
|
|
398 |
|
|
399 /* Matches any character whose syntax is not that specified. */ |
|
|
400 notsyntaxspec |
|
|
401 #endif /* emacs */ |
|
|
402 } re_opcode_t; |
|
|
403 |
|
|
404 /* Common operations on the compiled pattern. */ |
|
|
405 |
|
|
406 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ |
|
|
407 |
|
|
408 #define STORE_NUMBER(destination, number) \ |
|
|
409 do { \ |
|
|
410 (destination)[0] = (number) & 0377; \ |
|
|
411 (destination)[1] = (number) >> 8; \ |
|
|
412 } while (0) |
|
|
413 |
|
|
414 /* Same as STORE_NUMBER, except increment DESTINATION to |
|
|
415 the byte after where the number is stored. Therefore, DESTINATION |
|
|
416 must be an lvalue. */ |
|
|
417 |
|
|
418 #define STORE_NUMBER_AND_INCR(destination, number) \ |
|
|
419 do { \ |
|
|
420 STORE_NUMBER (destination, number); \ |
|
|
421 (destination) += 2; \ |
|
|
422 } while (0) |
|
|
423 |
|
|
424 /* Put into DESTINATION a number stored in two contiguous bytes starting |
|
|
425 at SOURCE. */ |
|
|
426 |
|
|
427 #define EXTRACT_NUMBER(destination, source) \ |
|
|
428 do { \ |
|
|
429 (destination) = *(source) & 0377; \ |
|
|
430 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ |
|
|
431 } while (0) |
|
|
432 |
|
|
433 #ifdef DEBUG |
|
|
434 static void |
|
|
435 extract_number (dest, source) |
|
|
436 int *dest; |
|
|
437 unsigned char *source; |
|
|
438 { |
|
|
439 int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
|
|
440 *dest = *source & 0377; |
|
|
441 *dest += temp << 8; |
|
|
442 } |
|
|
443 |
|
|
444 #ifndef EXTRACT_MACROS /* To debug the macros. */ |
|
|
445 #undef EXTRACT_NUMBER |
|
|
446 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) |
|
|
447 #endif /* not EXTRACT_MACROS */ |
|
|
448 |
|
|
449 #endif /* DEBUG */ |
|
|
450 |
|
|
451 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. |
|
|
452 SOURCE must be an lvalue. */ |
|
|
453 |
|
|
454 #define EXTRACT_NUMBER_AND_INCR(destination, source) \ |
|
|
455 do { \ |
|
|
456 EXTRACT_NUMBER (destination, source); \ |
|
|
457 (source) += 2; \ |
|
|
458 } while (0) |
|
|
459 |
|
|
460 #ifdef DEBUG |
|
|
461 static void |
|
|
462 extract_number_and_incr (destination, source) |
|
|
463 int *destination; |
|
|
464 unsigned char **source; |
|
|
465 { |
|
|
466 extract_number (destination, *source); |
|
|
467 *source += 2; |
|
|
468 } |
|
|
469 |
|
|
470 #ifndef EXTRACT_MACROS |
|
|
471 #undef EXTRACT_NUMBER_AND_INCR |
|
|
472 #define EXTRACT_NUMBER_AND_INCR(dest, src) \ |
|
|
473 extract_number_and_incr (&dest, &src) |
|
|
474 #endif /* not EXTRACT_MACROS */ |
|
|
475 |
|
|
476 #endif /* DEBUG */ |
|
|
477 |
|
|
478 /* If DEBUG is defined, Regex prints many voluminous messages about what |
|
|
479 it is doing (if the variable `debug' is nonzero). If linked with the |
|
|
480 main program in `iregex.c', you can enter patterns and strings |
|
|
481 interactively. And if linked with the main program in `main.c' and |
|
|
482 the other test files, you can run the already-written tests. */ |
|
|
483 |
|
|
484 #ifdef DEBUG |
|
|
485 |
|
|
486 /* We use standard I/O for debugging. */ |
|
|
487 #include <stdio.h> |
|
|
488 |
|
|
489 /* It is useful to test things that ``must'' be true when debugging. */ |
|
|
490 #include <assert.h> |
|
|
491 |
|
|
492 static int debug = 0; |
|
|
493 |
|
|
494 #define DEBUG_STATEMENT(e) e |
|
|
495 #define DEBUG_PRINT1(x) if (debug) printf (x) |
|
|
496 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) |
|
|
497 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) |
|
23
|
498 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) |
|
2
|
499 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
|
|
500 if (debug) print_partial_compiled_pattern (s, e) |
|
|
501 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
|
|
502 if (debug) print_double_string (w, s1, sz1, s2, sz2) |
|
|
503 |
|
|
504 |
|
|
505 extern void printchar (); |
|
|
506 |
|
|
507 /* Print the fastmap in human-readable form. */ |
|
|
508 |
|
|
509 void |
|
|
510 print_fastmap (fastmap) |
|
|
511 char *fastmap; |
|
|
512 { |
|
|
513 unsigned was_a_range = 0; |
|
|
514 unsigned i = 0; |
|
|
515 |
|
|
516 while (i < (1 << BYTEWIDTH)) |
|
|
517 { |
|
|
518 if (fastmap[i++]) |
|
|
519 { |
|
|
520 was_a_range = 0; |
|
|
521 printchar (i - 1); |
|
|
522 while (i < (1 << BYTEWIDTH) && fastmap[i]) |
|
|
523 { |
|
|
524 was_a_range = 1; |
|
|
525 i++; |
|
|
526 } |
|
|
527 if (was_a_range) |
|
|
528 { |
|
|
529 printf ("-"); |
|
|
530 printchar (i - 1); |
|
|
531 } |
|
|
532 } |
|
|
533 } |
|
|
534 putchar ('\n'); |
|
|
535 } |
|
|
536 |
|
|
537 |
|
|
538 /* Print a compiled pattern string in human-readable form, starting at |
|
|
539 the START pointer into it and ending just before the pointer END. */ |
|
|
540 |
|
|
541 void |
|
|
542 print_partial_compiled_pattern (start, end) |
|
|
543 unsigned char *start; |
|
|
544 unsigned char *end; |
|
|
545 { |
|
|
546 int mcnt, mcnt2; |
|
|
547 unsigned char *p = start; |
|
|
548 unsigned char *pend = end; |
|
|
549 |
|
|
550 if (start == NULL) |
|
|
551 { |
|
|
552 printf ("(null)\n"); |
|
|
553 return; |
|
|
554 } |
|
|
555 |
|
|
556 /* Loop over pattern commands. */ |
|
|
557 while (p < pend) |
|
|
558 { |
|
79
|
559 printf ("%d:\t", p - start); |
|
|
560 |
|
2
|
561 switch ((re_opcode_t) *p++) |
|
|
562 { |
|
|
563 case no_op: |
|
|
564 printf ("/no_op"); |
|
|
565 break; |
|
|
566 |
|
|
567 case exactn: |
|
|
568 mcnt = *p++; |
|
|
569 printf ("/exactn/%d", mcnt); |
|
|
570 do |
|
|
571 { |
|
|
572 putchar ('/'); |
|
|
573 printchar (*p++); |
|
|
574 } |
|
|
575 while (--mcnt); |
|
|
576 break; |
|
|
577 |
|
|
578 case start_memory: |
|
|
579 mcnt = *p++; |
|
|
580 printf ("/start_memory/%d/%d", mcnt, *p++); |
|
|
581 break; |
|
|
582 |
|
|
583 case stop_memory: |
|
|
584 mcnt = *p++; |
|
|
585 printf ("/stop_memory/%d/%d", mcnt, *p++); |
|
|
586 break; |
|
|
587 |
|
|
588 case duplicate: |
|
|
589 printf ("/duplicate/%d", *p++); |
|
|
590 break; |
|
|
591 |
|
|
592 case anychar: |
|
|
593 printf ("/anychar"); |
|
|
594 break; |
|
|
595 |
|
|
596 case charset: |
|
|
597 case charset_not: |
|
|
598 { |
|
79
|
599 register int c, last = -100; |
|
|
600 register int in_range = 0; |
|
|
601 |
|
|
602 printf ("/charset [%s", |
|
|
603 (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
|
2
|
604 |
|
|
605 assert (p + *p < pend); |
|
|
606 |
|
79
|
607 for (c = 0; c < 256; c++) |
|
|
608 if (c / 8 < *p |
|
|
609 && (p[1 + (c/8)] & (1 << (c % 8)))) |
|
|
610 { |
|
|
611 /* Are we starting a range? */ |
|
|
612 if (last + 1 == c && ! in_range) |
|
|
613 { |
|
|
614 putchar ('-'); |
|
|
615 in_range = 1; |
|
|
616 } |
|
|
617 /* Have we broken a range? */ |
|
|
618 else if (last + 1 != c && in_range) |
|
2
|
619 { |
|
79
|
620 printchar (last); |
|
|
621 in_range = 0; |
|
|
622 } |
|
2
|
623 |
|
79
|
624 if (! in_range) |
|
|
625 printchar (c); |
|
|
626 |
|
|
627 last = c; |
|
2
|
628 } |
|
79
|
629 |
|
|
630 if (in_range) |
|
|
631 printchar (last); |
|
|
632 |
|
|
633 putchar (']'); |
|
|
634 |
|
2
|
635 p += 1 + *p; |
|
|
636 } |
|
79
|
637 break; |
|
2
|
638 |
|
|
639 case begline: |
|
|
640 printf ("/begline"); |
|
|
641 break; |
|
|
642 |
|
|
643 case endline: |
|
|
644 printf ("/endline"); |
|
|
645 break; |
|
|
646 |
|
|
647 case on_failure_jump: |
|
|
648 extract_number_and_incr (&mcnt, &p); |
|
79
|
649 printf ("/on_failure_jump to %d", p + mcnt - start); |
|
2
|
650 break; |
|
|
651 |
|
|
652 case on_failure_keep_string_jump: |
|
|
653 extract_number_and_incr (&mcnt, &p); |
|
79
|
654 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); |
|
2
|
655 break; |
|
|
656 |
|
|
657 case dummy_failure_jump: |
|
|
658 extract_number_and_incr (&mcnt, &p); |
|
79
|
659 printf ("/dummy_failure_jump to %d", p + mcnt - start); |
|
2
|
660 break; |
|
|
661 |
|
|
662 case push_dummy_failure: |
|
|
663 printf ("/push_dummy_failure"); |
|
|
664 break; |
|
|
665 |
|
|
666 case maybe_pop_jump: |
|
|
667 extract_number_and_incr (&mcnt, &p); |
|
79
|
668 printf ("/maybe_pop_jump to %d", p + mcnt - start); |
|
2
|
669 break; |
|
|
670 |
|
|
671 case pop_failure_jump: |
|
|
672 extract_number_and_incr (&mcnt, &p); |
|
79
|
673 printf ("/pop_failure_jump to %d", p + mcnt - start); |
|
2
|
674 break; |
|
|
675 |
|
|
676 case jump_past_alt: |
|
|
677 extract_number_and_incr (&mcnt, &p); |
|
79
|
678 printf ("/jump_past_alt to %d", p + mcnt - start); |
|
2
|
679 break; |
|
|
680 |
|
|
681 case jump: |
|
|
682 extract_number_and_incr (&mcnt, &p); |
|
79
|
683 printf ("/jump to %d", p + mcnt - start); |
|
2
|
684 break; |
|
|
685 |
|
|
686 case succeed_n: |
|
|
687 extract_number_and_incr (&mcnt, &p); |
|
|
688 extract_number_and_incr (&mcnt2, &p); |
|
79
|
689 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2); |
|
2
|
690 break; |
|
|
691 |
|
|
692 case jump_n: |
|
|
693 extract_number_and_incr (&mcnt, &p); |
|
|
694 extract_number_and_incr (&mcnt2, &p); |
|
79
|
695 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2); |
|
2
|
696 break; |
|
|
697 |
|
|
698 case set_number_at: |
|
|
699 extract_number_and_incr (&mcnt, &p); |
|
|
700 extract_number_and_incr (&mcnt2, &p); |
|
79
|
701 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2); |
|
2
|
702 break; |
|
|
703 |
|
|
704 case wordbound: |
|
|
705 printf ("/wordbound"); |
|
|
706 break; |
|
|
707 |
|
|
708 case notwordbound: |
|
|
709 printf ("/notwordbound"); |
|
|
710 break; |
|
|
711 |
|
|
712 case wordbeg: |
|
|
713 printf ("/wordbeg"); |
|
|
714 break; |
|
|
715 |
|
|
716 case wordend: |
|
|
717 printf ("/wordend"); |
|
|
718 |
|
|
719 #ifdef emacs |
|
|
720 case before_dot: |
|
|
721 printf ("/before_dot"); |
|
|
722 break; |
|
|
723 |
|
|
724 case at_dot: |
|
|
725 printf ("/at_dot"); |
|
|
726 break; |
|
|
727 |
|
|
728 case after_dot: |
|
|
729 printf ("/after_dot"); |
|
|
730 break; |
|
|
731 |
|
|
732 case syntaxspec: |
|
|
733 printf ("/syntaxspec"); |
|
|
734 mcnt = *p++; |
|
|
735 printf ("/%d", mcnt); |
|
|
736 break; |
|
|
737 |
|
|
738 case notsyntaxspec: |
|
|
739 printf ("/notsyntaxspec"); |
|
|
740 mcnt = *p++; |
|
|
741 printf ("/%d", mcnt); |
|
|
742 break; |
|
|
743 #endif /* emacs */ |
|
|
744 |
|
|
745 case wordchar: |
|
|
746 printf ("/wordchar"); |
|
|
747 break; |
|
|
748 |
|
|
749 case notwordchar: |
|
|
750 printf ("/notwordchar"); |
|
|
751 break; |
|
|
752 |
|
|
753 case begbuf: |
|
|
754 printf ("/begbuf"); |
|
|
755 break; |
|
|
756 |
|
|
757 case endbuf: |
|
|
758 printf ("/endbuf"); |
|
|
759 break; |
|
|
760 |
|
|
761 default: |
|
|
762 printf ("?%d", *(p-1)); |
|
|
763 } |
|
79
|
764 |
|
|
765 putchar ('\n'); |
|
2
|
766 } |
|
79
|
767 |
|
|
768 printf ("%d:\tend of pattern.\n", p - start); |
|
2
|
769 } |
|
|
770 |
|
|
771 |
|
|
772 void |
|
|
773 print_compiled_pattern (bufp) |
|
|
774 struct re_pattern_buffer *bufp; |
|
|
775 { |
|
|
776 unsigned char *buffer = bufp->buffer; |
|
|
777 |
|
|
778 print_partial_compiled_pattern (buffer, buffer + bufp->used); |
|
|
779 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); |
|
|
780 |
|
|
781 if (bufp->fastmap_accurate && bufp->fastmap) |
|
|
782 { |
|
|
783 printf ("fastmap: "); |
|
|
784 print_fastmap (bufp->fastmap); |
|
|
785 } |
|
|
786 |
|
|
787 printf ("re_nsub: %d\t", bufp->re_nsub); |
|
|
788 printf ("regs_alloc: %d\t", bufp->regs_allocated); |
|
|
789 printf ("can_be_null: %d\t", bufp->can_be_null); |
|
|
790 printf ("newline_anchor: %d\n", bufp->newline_anchor); |
|
|
791 printf ("no_sub: %d\t", bufp->no_sub); |
|
|
792 printf ("not_bol: %d\t", bufp->not_bol); |
|
|
793 printf ("not_eol: %d\t", bufp->not_eol); |
|
|
794 printf ("syntax: %d\n", bufp->syntax); |
|
|
795 /* Perhaps we should print the translate table? */ |
|
|
796 } |
|
|
797 |
|
|
798 |
|
|
799 void |
|
|
800 print_double_string (where, string1, size1, string2, size2) |
|
|
801 const char *where; |
|
|
802 const char *string1; |
|
|
803 const char *string2; |
|
|
804 int size1; |
|
|
805 int size2; |
|
|
806 { |
|
|
807 unsigned this_char; |
|
|
808 |
|
|
809 if (where == NULL) |
|
|
810 printf ("(null)"); |
|
|
811 else |
|
|
812 { |
|
|
813 if (FIRST_STRING_P (where)) |
|
|
814 { |
|
|
815 for (this_char = where - string1; this_char < size1; this_char++) |
|
|
816 printchar (string1[this_char]); |
|
|
817 |
|
|
818 where = string2; |
|
|
819 } |
|
|
820 |
|
|
821 for (this_char = where - string2; this_char < size2; this_char++) |
|
|
822 printchar (string2[this_char]); |
|
|
823 } |
|
|
824 } |
|
|
825 |
|
|
826 #else /* not DEBUG */ |
|
|
827 |
|
|
828 #undef assert |
|
|
829 #define assert(e) |
|
|
830 |
|
|
831 #define DEBUG_STATEMENT(e) |
|
|
832 #define DEBUG_PRINT1(x) |
|
|
833 #define DEBUG_PRINT2(x1, x2) |
|
|
834 #define DEBUG_PRINT3(x1, x2, x3) |
|
23
|
835 #define DEBUG_PRINT4(x1, x2, x3, x4) |
|
2
|
836 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
|
|
837 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) |
|
|
838 |
|
|
839 #endif /* not DEBUG */ |
|
|
840 |
|
|
841 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can |
|
|
842 also be assigned to arbitrarily: each pattern buffer stores its own |
|
|
843 syntax, so it can be changed between regex compilations. */ |
|
|
844 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS; |
|
|
845 |
|
|
846 |
|
|
847 /* Specify the precise syntax of regexps for compilation. This provides |
|
|
848 for compatibility for various utilities which historically have |
|
|
849 different, incompatible syntaxes. |
|
|
850 |
|
|
851 The argument SYNTAX is a bit mask comprised of the various bits |
|
|
852 defined in regex.h. We return the old syntax. */ |
|
|
853 |
|
|
854 reg_syntax_t |
|
|
855 re_set_syntax (syntax) |
|
|
856 reg_syntax_t syntax; |
|
|
857 { |
|
|
858 reg_syntax_t ret = re_syntax_options; |
|
|
859 |
|
|
860 re_syntax_options = syntax; |
|
|
861 return ret; |
|
|
862 } |
|
|
863 |
|
|
864 /* This table gives an error message for each of the error codes listed |
|
|
865 in regex.h. Obviously the order here has to be same as there. */ |
|
|
866 |
|
|
867 static const char *re_error_msg[] = |
|
|
868 { NULL, /* REG_NOERROR */ |
|
|
869 "No match", /* REG_NOMATCH */ |
|
|
870 "Invalid regular expression", /* REG_BADPAT */ |
|
|
871 "Invalid collation character", /* REG_ECOLLATE */ |
|
|
872 "Invalid character class name", /* REG_ECTYPE */ |
|
|
873 "Trailing backslash", /* REG_EESCAPE */ |
|
|
874 "Invalid back reference", /* REG_ESUBREG */ |
|
|
875 "Unmatched [ or [^", /* REG_EBRACK */ |
|
|
876 "Unmatched ( or \\(", /* REG_EPAREN */ |
|
|
877 "Unmatched \\{", /* REG_EBRACE */ |
|
|
878 "Invalid content of \\{\\}", /* REG_BADBR */ |
|
|
879 "Invalid range end", /* REG_ERANGE */ |
|
|
880 "Memory exhausted", /* REG_ESPACE */ |
|
|
881 "Invalid preceding regular expression", /* REG_BADRPT */ |
|
|
882 "Premature end of regular expression", /* REG_EEND */ |
|
|
883 "Regular expression too big", /* REG_ESIZE */ |
|
|
884 "Unmatched ) or \\)", /* REG_ERPAREN */ |
|
|
885 }; |
|
|
886 |
|
87
|
887 /* Avoiding alloca during matching, to placate r_alloc. */ |
|
|
888 |
|
89
|
889 /* Define MATCH_MAY_ALLOCATE if we need to make sure that the |
|
87
|
890 searching and matching functions should not call alloca. On some |
|
|
891 systems, alloca is implemented in terms of malloc, and if we're |
|
|
892 using the relocating allocator routines, then malloc could cause a |
|
|
893 relocation, which might (if the strings being searched are in the |
|
|
894 ralloc heap) shift the data out from underneath the regexp |
|
93
|
895 routines. |
|
|
896 |
|
|
897 Here's another reason to avoid allocation: Emacs insists on |
|
|
898 processing input from X in a signal handler; processing X input may |
|
|
899 call malloc; if input arrives while a matching routine is calling |
|
|
900 malloc, then we're scrod. But Emacs can't just block input while |
|
|
901 calling matching routines; then we don't notice interrupts when |
|
|
902 they come in. So, Emacs blocks input around all regexp calls |
|
|
903 except the matching calls, which it leaves unprotected, in the |
|
|
904 faith that they will not malloc. */ |
|
89
|
905 |
|
|
906 /* Normally, this is fine. */ |
|
|
907 #define MATCH_MAY_ALLOCATE |
|
|
908 |
|
|
909 /* But under some circumstances, it's not. */ |
|
93
|
910 #if defined (emacs) || (defined (REL_ALLOC) && defined (C_ALLOCA)) |
|
89
|
911 #undef MATCH_MAY_ALLOCATE |
|
87
|
912 #endif |
|
|
913 |
|
|
914 |
|
|
915 /* Failure stack declarations and macros; both re_compile_fastmap and |
|
|
916 re_match_2 use a failure stack. These have to be macros because of |
|
|
917 REGEX_ALLOCATE. */ |
|
|
918 |
|
|
919 |
|
|
920 /* Number of failure points for which to initially allocate space |
|
|
921 when matching. If this number is exceeded, we allocate more |
|
|
922 space, so it is not a hard limit. */ |
|
|
923 #ifndef INIT_FAILURE_ALLOC |
|
|
924 #define INIT_FAILURE_ALLOC 5 |
|
|
925 #endif |
|
|
926 |
|
|
927 /* Roughly the maximum number of failure points on the stack. Would be |
|
|
928 exactly that if always used MAX_FAILURE_SPACE each time we failed. |
|
|
929 This is a variable only so users of regex can assign to it; we never |
|
|
930 change it ourselves. */ |
|
|
931 int re_max_failures = 2000; |
|
|
932 |
|
|
933 typedef const unsigned char *fail_stack_elt_t; |
|
|
934 |
|
|
935 typedef struct |
|
|
936 { |
|
|
937 fail_stack_elt_t *stack; |
|
|
938 unsigned size; |
|
|
939 unsigned avail; /* Offset of next open position. */ |
|
|
940 } fail_stack_type; |
|
|
941 |
|
|
942 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) |
|
|
943 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) |
|
|
944 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
|
|
945 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail]) |
|
|
946 |
|
|
947 |
|
|
948 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */ |
|
|
949 |
|
89
|
950 #ifdef MATCH_MAY_ALLOCATE |
|
87
|
951 #define INIT_FAIL_STACK() \ |
|
|
952 do { \ |
|
|
953 fail_stack.stack = (fail_stack_elt_t *) \ |
|
|
954 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ |
|
|
955 \ |
|
|
956 if (fail_stack.stack == NULL) \ |
|
|
957 return -2; \ |
|
|
958 \ |
|
|
959 fail_stack.size = INIT_FAILURE_ALLOC; \ |
|
|
960 fail_stack.avail = 0; \ |
|
|
961 } while (0) |
|
|
962 #else |
|
|
963 #define INIT_FAIL_STACK() \ |
|
|
964 do { \ |
|
|
965 fail_stack.avail = 0; \ |
|
|
966 } while (0) |
|
|
967 #endif |
|
|
968 |
|
|
969 |
|
|
970 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. |
|
|
971 |
|
|
972 Return 1 if succeeds, and 0 if either ran out of memory |
|
|
973 allocating space for it or it was already too large. |
|
|
974 |
|
|
975 REGEX_REALLOCATE requires `destination' be declared. */ |
|
|
976 |
|
|
977 #define DOUBLE_FAIL_STACK(fail_stack) \ |
|
|
978 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ |
|
|
979 ? 0 \ |
|
|
980 : ((fail_stack).stack = (fail_stack_elt_t *) \ |
|
|
981 REGEX_REALLOCATE ((fail_stack).stack, \ |
|
|
982 (fail_stack).size * sizeof (fail_stack_elt_t), \ |
|
|
983 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ |
|
|
984 \ |
|
|
985 (fail_stack).stack == NULL \ |
|
|
986 ? 0 \ |
|
|
987 : ((fail_stack).size <<= 1, \ |
|
|
988 1))) |
|
|
989 |
|
|
990 |
|
|
991 /* Push PATTERN_OP on FAIL_STACK. |
|
|
992 |
|
|
993 Return 1 if was able to do so and 0 if ran out of memory allocating |
|
|
994 space to do so. */ |
|
|
995 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \ |
|
|
996 ((FAIL_STACK_FULL () \ |
|
|
997 && !DOUBLE_FAIL_STACK (fail_stack)) \ |
|
|
998 ? 0 \ |
|
|
999 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \ |
|
|
1000 1)) |
|
|
1001 |
|
|
1002 /* This pushes an item onto the failure stack. Must be a four-byte |
|
|
1003 value. Assumes the variable `fail_stack'. Probably should only |
|
|
1004 be called from within `PUSH_FAILURE_POINT'. */ |
|
|
1005 #define PUSH_FAILURE_ITEM(item) \ |
|
|
1006 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item |
|
|
1007 |
|
|
1008 /* The complement operation. Assumes `fail_stack' is nonempty. */ |
|
|
1009 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail] |
|
|
1010 |
|
|
1011 /* Used to omit pushing failure point id's when we're not debugging. */ |
|
|
1012 #ifdef DEBUG |
|
|
1013 #define DEBUG_PUSH PUSH_FAILURE_ITEM |
|
|
1014 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM () |
|
|
1015 #else |
|
|
1016 #define DEBUG_PUSH(item) |
|
|
1017 #define DEBUG_POP(item_addr) |
|
|
1018 #endif |
|
|
1019 |
|
|
1020 |
|
|
1021 /* Push the information about the state we will need |
|
|
1022 if we ever fail back to it. |
|
|
1023 |
|
|
1024 Requires variables fail_stack, regstart, regend, reg_info, and |
|
|
1025 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be |
|
|
1026 declared. |
|
|
1027 |
|
|
1028 Does `return FAILURE_CODE' if runs out of memory. */ |
|
|
1029 |
|
|
1030 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ |
|
|
1031 do { \ |
|
|
1032 char *destination; \ |
|
|
1033 /* Must be int, so when we don't save any registers, the arithmetic \ |
|
|
1034 of 0 + -1 isn't done as unsigned. */ \ |
|
|
1035 int this_reg; \ |
|
|
1036 \ |
|
|
1037 DEBUG_STATEMENT (failure_id++); \ |
|
|
1038 DEBUG_STATEMENT (nfailure_points_pushed++); \ |
|
|
1039 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ |
|
|
1040 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ |
|
|
1041 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ |
|
|
1042 \ |
|
|
1043 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ |
|
|
1044 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ |
|
|
1045 \ |
|
|
1046 /* Ensure we have enough space allocated for what we will push. */ \ |
|
|
1047 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ |
|
|
1048 { \ |
|
|
1049 if (!DOUBLE_FAIL_STACK (fail_stack)) \ |
|
|
1050 return failure_code; \ |
|
|
1051 \ |
|
|
1052 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ |
|
|
1053 (fail_stack).size); \ |
|
|
1054 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ |
|
|
1055 } \ |
|
|
1056 \ |
|
|
1057 /* Push the info, starting with the registers. */ \ |
|
|
1058 DEBUG_PRINT1 ("\n"); \ |
|
|
1059 \ |
|
|
1060 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ |
|
|
1061 this_reg++) \ |
|
|
1062 { \ |
|
|
1063 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ |
|
|
1064 DEBUG_STATEMENT (num_regs_pushed++); \ |
|
|
1065 \ |
|
|
1066 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ |
|
|
1067 PUSH_FAILURE_ITEM (regstart[this_reg]); \ |
|
|
1068 \ |
|
|
1069 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ |
|
|
1070 PUSH_FAILURE_ITEM (regend[this_reg]); \ |
|
|
1071 \ |
|
|
1072 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ |
|
|
1073 DEBUG_PRINT2 (" match_null=%d", \ |
|
|
1074 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ |
|
|
1075 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ |
|
|
1076 DEBUG_PRINT2 (" matched_something=%d", \ |
|
|
1077 MATCHED_SOMETHING (reg_info[this_reg])); \ |
|
|
1078 DEBUG_PRINT2 (" ever_matched=%d", \ |
|
|
1079 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ |
|
|
1080 DEBUG_PRINT1 ("\n"); \ |
|
|
1081 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \ |
|
|
1082 } \ |
|
|
1083 \ |
|
|
1084 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ |
|
|
1085 PUSH_FAILURE_ITEM (lowest_active_reg); \ |
|
|
1086 \ |
|
|
1087 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ |
|
|
1088 PUSH_FAILURE_ITEM (highest_active_reg); \ |
|
|
1089 \ |
|
|
1090 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ |
|
|
1091 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ |
|
|
1092 PUSH_FAILURE_ITEM (pattern_place); \ |
|
|
1093 \ |
|
|
1094 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ |
|
|
1095 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ |
|
|
1096 size2); \ |
|
|
1097 DEBUG_PRINT1 ("'\n"); \ |
|
|
1098 PUSH_FAILURE_ITEM (string_place); \ |
|
|
1099 \ |
|
|
1100 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ |
|
|
1101 DEBUG_PUSH (failure_id); \ |
|
|
1102 } while (0) |
|
|
1103 |
|
|
1104 /* This is the number of items that are pushed and popped on the stack |
|
|
1105 for each register. */ |
|
|
1106 #define NUM_REG_ITEMS 3 |
|
|
1107 |
|
|
1108 /* Individual items aside from the registers. */ |
|
|
1109 #ifdef DEBUG |
|
|
1110 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ |
|
|
1111 #else |
|
|
1112 #define NUM_NONREG_ITEMS 4 |
|
|
1113 #endif |
|
|
1114 |
|
|
1115 /* We push at most this many items on the stack. */ |
|
|
1116 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS) |
|
|
1117 |
|
|
1118 /* We actually push this many items. */ |
|
|
1119 #define NUM_FAILURE_ITEMS \ |
|
|
1120 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \ |
|
|
1121 + NUM_NONREG_ITEMS) |
|
|
1122 |
|
|
1123 /* How many items can still be added to the stack without overflowing it. */ |
|
|
1124 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) |
|
|
1125 |
|
|
1126 |
|
|
1127 /* Pops what PUSH_FAIL_STACK pushes. |
|
|
1128 |
|
|
1129 We restore into the parameters, all of which should be lvalues: |
|
|
1130 STR -- the saved data position. |
|
|
1131 PAT -- the saved pattern position. |
|
|
1132 LOW_REG, HIGH_REG -- the highest and lowest active registers. |
|
|
1133 REGSTART, REGEND -- arrays of string positions. |
|
|
1134 REG_INFO -- array of information about each subexpression. |
|
|
1135 |
|
|
1136 Also assumes the variables `fail_stack' and (if debugging), `bufp', |
|
|
1137 `pend', `string1', `size1', `string2', and `size2'. */ |
|
|
1138 |
|
|
1139 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ |
|
|
1140 { \ |
|
|
1141 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ |
|
|
1142 int this_reg; \ |
|
|
1143 const unsigned char *string_temp; \ |
|
|
1144 \ |
|
|
1145 assert (!FAIL_STACK_EMPTY ()); \ |
|
|
1146 \ |
|
|
1147 /* Remove failure points and point to how many regs pushed. */ \ |
|
|
1148 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ |
|
|
1149 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ |
|
|
1150 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
|
|
1151 \ |
|
|
1152 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ |
|
|
1153 \ |
|
|
1154 DEBUG_POP (&failure_id); \ |
|
|
1155 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ |
|
|
1156 \ |
|
|
1157 /* If the saved string location is NULL, it came from an \ |
|
|
1158 on_failure_keep_string_jump opcode, and we want to throw away the \ |
|
|
1159 saved NULL, thus retaining our current position in the string. */ \ |
|
|
1160 string_temp = POP_FAILURE_ITEM (); \ |
|
|
1161 if (string_temp != NULL) \ |
|
|
1162 str = (const char *) string_temp; \ |
|
|
1163 \ |
|
|
1164 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ |
|
|
1165 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
|
|
1166 DEBUG_PRINT1 ("'\n"); \ |
|
|
1167 \ |
|
|
1168 pat = (unsigned char *) POP_FAILURE_ITEM (); \ |
|
|
1169 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ |
|
|
1170 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ |
|
|
1171 \ |
|
|
1172 /* Restore register info. */ \ |
|
|
1173 high_reg = (unsigned) POP_FAILURE_ITEM (); \ |
|
|
1174 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ |
|
|
1175 \ |
|
|
1176 low_reg = (unsigned) POP_FAILURE_ITEM (); \ |
|
|
1177 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ |
|
|
1178 \ |
|
|
1179 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ |
|
|
1180 { \ |
|
|
1181 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ |
|
|
1182 \ |
|
|
1183 reg_info[this_reg].word = POP_FAILURE_ITEM (); \ |
|
|
1184 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ |
|
|
1185 \ |
|
|
1186 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \ |
|
|
1187 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ |
|
|
1188 \ |
|
|
1189 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \ |
|
|
1190 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ |
|
|
1191 } \ |
|
|
1192 \ |
|
|
1193 DEBUG_STATEMENT (nfailure_points_popped++); \ |
|
|
1194 } /* POP_FAILURE_POINT */ |
|
|
1195 |
|
|
1196 |
|
|
1197 |
|
|
1198 /* Structure for per-register (a.k.a. per-group) information. |
|
|
1199 This must not be longer than one word, because we push this value |
|
|
1200 onto the failure stack. Other register information, such as the |
|
|
1201 starting and ending positions (which are addresses), and the list of |
|
|
1202 inner groups (which is a bits list) are maintained in separate |
|
|
1203 variables. |
|
|
1204 |
|
|
1205 We are making a (strictly speaking) nonportable assumption here: that |
|
|
1206 the compiler will pack our bit fields into something that fits into |
|
|
1207 the type of `word', i.e., is something that fits into one item on the |
|
|
1208 failure stack. */ |
|
|
1209 typedef union |
|
|
1210 { |
|
|
1211 fail_stack_elt_t word; |
|
|
1212 struct |
|
|
1213 { |
|
|
1214 /* This field is one if this group can match the empty string, |
|
|
1215 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ |
|
|
1216 #define MATCH_NULL_UNSET_VALUE 3 |
|
|
1217 unsigned match_null_string_p : 2; |
|
|
1218 unsigned is_active : 1; |
|
|
1219 unsigned matched_something : 1; |
|
|
1220 unsigned ever_matched_something : 1; |
|
|
1221 } bits; |
|
|
1222 } register_info_type; |
|
|
1223 |
|
|
1224 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) |
|
|
1225 #define IS_ACTIVE(R) ((R).bits.is_active) |
|
|
1226 #define MATCHED_SOMETHING(R) ((R).bits.matched_something) |
|
|
1227 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) |
|
|
1228 |
|
|
1229 |
|
|
1230 /* Call this when have matched a real character; it sets `matched' flags |
|
|
1231 for the subexpressions which we are currently inside. Also records |
|
|
1232 that those subexprs have matched. */ |
|
|
1233 #define SET_REGS_MATCHED() \ |
|
|
1234 do \ |
|
|
1235 { \ |
|
|
1236 unsigned r; \ |
|
|
1237 for (r = lowest_active_reg; r <= highest_active_reg; r++) \ |
|
|
1238 { \ |
|
|
1239 MATCHED_SOMETHING (reg_info[r]) \ |
|
|
1240 = EVER_MATCHED_SOMETHING (reg_info[r]) \ |
|
|
1241 = 1; \ |
|
|
1242 } \ |
|
|
1243 } \ |
|
|
1244 while (0) |
|
|
1245 |
|
|
1246 |
|
|
1247 /* Registers are set to a sentinel when they haven't yet matched. */ |
|
|
1248 #define REG_UNSET_VALUE ((char *) -1) |
|
|
1249 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) |
|
|
1250 |
|
|
1251 |
|
|
1252 |
|
89
|
1253 /* How do we implement a missing MATCH_MAY_ALLOCATE? |
|
87
|
1254 We make the fail stack a global thing, and then grow it to |
|
|
1255 re_max_failures when we compile. */ |
|
89
|
1256 #ifndef MATCH_MAY_ALLOCATE |
|
87
|
1257 static fail_stack_type fail_stack; |
|
|
1258 |
|
|
1259 static const char ** regstart, ** regend; |
|
|
1260 static const char ** old_regstart, ** old_regend; |
|
|
1261 static const char **best_regstart, **best_regend; |
|
|
1262 static register_info_type *reg_info; |
|
|
1263 static const char **reg_dummy; |
|
|
1264 static register_info_type *reg_info_dummy; |
|
|
1265 #endif |
|
|
1266 |
|
|
1267 |
|
2
|
1268 /* Subroutine declarations and macros for regex_compile. */ |
|
|
1269 |
|
|
1270 static void store_op1 (), store_op2 (); |
|
|
1271 static void insert_op1 (), insert_op2 (); |
|
|
1272 static boolean at_begline_loc_p (), at_endline_loc_p (); |
|
|
1273 static boolean group_in_compile_stack (); |
|
|
1274 static reg_errcode_t compile_range (); |
|
|
1275 |
|
|
1276 /* Fetch the next character in the uncompiled pattern---translating it |
|
|
1277 if necessary. Also cast from a signed character in the constant |
|
|
1278 string passed to us by the user to an unsigned char that we can use |
|
|
1279 as an array index (in, e.g., `translate'). */ |
|
|
1280 #define PATFETCH(c) \ |
|
|
1281 do {if (p == pend) return REG_EEND; \ |
|
|
1282 c = (unsigned char) *p++; \ |
|
|
1283 if (translate) c = translate[c]; \ |
|
|
1284 } while (0) |
|
|
1285 |
|
|
1286 /* Fetch the next character in the uncompiled pattern, with no |
|
|
1287 translation. */ |
|
|
1288 #define PATFETCH_RAW(c) \ |
|
|
1289 do {if (p == pend) return REG_EEND; \ |
|
|
1290 c = (unsigned char) *p++; \ |
|
|
1291 } while (0) |
|
|
1292 |
|
|
1293 /* Go backwards one character in the pattern. */ |
|
|
1294 #define PATUNFETCH p-- |
|
|
1295 |
|
|
1296 |
|
|
1297 /* If `translate' is non-null, return translate[D], else just D. We |
|
|
1298 cast the subscript to translate because some data is declared as |
|
|
1299 `char *', to avoid warnings when a string constant is passed. But |
|
|
1300 when we use a character as a subscript we must make it unsigned. */ |
|
|
1301 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d)) |
|
|
1302 |
|
|
1303 |
|
|
1304 /* Macros for outputting the compiled pattern into `buffer'. */ |
|
|
1305 |
|
|
1306 /* If the buffer isn't allocated when it comes in, use this. */ |
|
|
1307 #define INIT_BUF_SIZE 32 |
|
|
1308 |
|
|
1309 /* Make sure we have at least N more bytes of space in buffer. */ |
|
|
1310 #define GET_BUFFER_SPACE(n) \ |
|
|
1311 while (b - bufp->buffer + (n) > bufp->allocated) \ |
|
|
1312 EXTEND_BUFFER () |
|
|
1313 |
|
|
1314 /* Make sure we have one more byte of buffer space and then add C to it. */ |
|
|
1315 #define BUF_PUSH(c) \ |
|
|
1316 do { \ |
|
|
1317 GET_BUFFER_SPACE (1); \ |
|
|
1318 *b++ = (unsigned char) (c); \ |
|
|
1319 } while (0) |
|
|
1320 |
|
|
1321 |
|
|
1322 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ |
|
|
1323 #define BUF_PUSH_2(c1, c2) \ |
|
|
1324 do { \ |
|
|
1325 GET_BUFFER_SPACE (2); \ |
|
|
1326 *b++ = (unsigned char) (c1); \ |
|
|
1327 *b++ = (unsigned char) (c2); \ |
|
|
1328 } while (0) |
|
|
1329 |
|
|
1330 |
|
|
1331 /* As with BUF_PUSH_2, except for three bytes. */ |
|
|
1332 #define BUF_PUSH_3(c1, c2, c3) \ |
|
|
1333 do { \ |
|
|
1334 GET_BUFFER_SPACE (3); \ |
|
|
1335 *b++ = (unsigned char) (c1); \ |
|
|
1336 *b++ = (unsigned char) (c2); \ |
|
|
1337 *b++ = (unsigned char) (c3); \ |
|
|
1338 } while (0) |
|
|
1339 |
|
|
1340 |
|
|
1341 /* Store a jump with opcode OP at LOC to location TO. We store a |
|
|
1342 relative address offset by the three bytes the jump itself occupies. */ |
|
|
1343 #define STORE_JUMP(op, loc, to) \ |
|
|
1344 store_op1 (op, loc, (to) - (loc) - 3) |
|
|
1345 |
|
|
1346 /* Likewise, for a two-argument jump. */ |
|
|
1347 #define STORE_JUMP2(op, loc, to, arg) \ |
|
|
1348 store_op2 (op, loc, (to) - (loc) - 3, arg) |
|
|
1349 |
|
|
1350 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
|
|
1351 #define INSERT_JUMP(op, loc, to) \ |
|
|
1352 insert_op1 (op, loc, (to) - (loc) - 3, b) |
|
|
1353 |
|
|
1354 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ |
|
|
1355 #define INSERT_JUMP2(op, loc, to, arg) \ |
|
|
1356 insert_op2 (op, loc, (to) - (loc) - 3, arg, b) |
|
|
1357 |
|
|
1358 |
|
|
1359 /* This is not an arbitrary limit: the arguments which represent offsets |
|
|
1360 into the pattern are two bytes long. So if 2^16 bytes turns out to |
|
|
1361 be too small, many things would have to change. */ |
|
|
1362 #define MAX_BUF_SIZE (1L << 16) |
|
|
1363 |
|
|
1364 |
|
|
1365 /* Extend the buffer by twice its current size via realloc and |
|
|
1366 reset the pointers that pointed into the old block to point to the |
|
|
1367 correct places in the new one. If extending the buffer results in it |
|
|
1368 being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
|
|
1369 #define EXTEND_BUFFER() \ |
|
|
1370 do { \ |
|
|
1371 unsigned char *old_buffer = bufp->buffer; \ |
|
|
1372 if (bufp->allocated == MAX_BUF_SIZE) \ |
|
|
1373 return REG_ESIZE; \ |
|
|
1374 bufp->allocated <<= 1; \ |
|
|
1375 if (bufp->allocated > MAX_BUF_SIZE) \ |
|
|
1376 bufp->allocated = MAX_BUF_SIZE; \ |
|
|
1377 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ |
|
|
1378 if (bufp->buffer == NULL) \ |
|
|
1379 return REG_ESPACE; \ |
|
|
1380 /* If the buffer moved, move all the pointers into it. */ \ |
|
|
1381 if (old_buffer != bufp->buffer) \ |
|
|
1382 { \ |
|
|
1383 b = (b - old_buffer) + bufp->buffer; \ |
|
|
1384 begalt = (begalt - old_buffer) + bufp->buffer; \ |
|
|
1385 if (fixup_alt_jump) \ |
|
|
1386 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ |
|
|
1387 if (laststart) \ |
|
|
1388 laststart = (laststart - old_buffer) + bufp->buffer; \ |
|
|
1389 if (pending_exact) \ |
|
|
1390 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ |
|
|
1391 } \ |
|
|
1392 } while (0) |
|
|
1393 |
|
|
1394 |
|
|
1395 /* Since we have one byte reserved for the register number argument to |
|
|
1396 {start,stop}_memory, the maximum number of groups we can report |
|
|
1397 things about is what fits in that byte. */ |
|
|
1398 #define MAX_REGNUM 255 |
|
|
1399 |
|
|
1400 /* But patterns can have more than `MAX_REGNUM' registers. We just |
|
|
1401 ignore the excess. */ |
|
|
1402 typedef unsigned regnum_t; |
|
|
1403 |
|
|
1404 |
|
|
1405 /* Macros for the compile stack. */ |
|
|
1406 |
|
|
1407 /* Since offsets can go either forwards or backwards, this type needs to |
|
|
1408 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
|
|
1409 typedef int pattern_offset_t; |
|
|
1410 |
|
|
1411 typedef struct |
|
|
1412 { |
|
|
1413 pattern_offset_t begalt_offset; |
|
|
1414 pattern_offset_t fixup_alt_jump; |
|
|
1415 pattern_offset_t inner_group_offset; |
|
|
1416 pattern_offset_t laststart_offset; |
|
|
1417 regnum_t regnum; |
|
|
1418 } compile_stack_elt_t; |
|
|
1419 |
|
|
1420 |
|
|
1421 typedef struct |
|
|
1422 { |
|
|
1423 compile_stack_elt_t *stack; |
|
|
1424 unsigned size; |
|
|
1425 unsigned avail; /* Offset of next open position. */ |
|
|
1426 } compile_stack_type; |
|
|
1427 |
|
|
1428 |
|
|
1429 #define INIT_COMPILE_STACK_SIZE 32 |
|
|
1430 |
|
|
1431 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) |
|
|
1432 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) |
|
|
1433 |
|
|
1434 /* The next available element. */ |
|
|
1435 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
|
|
1436 |
|
|
1437 |
|
|
1438 /* Set the bit for character C in a list. */ |
|
|
1439 #define SET_LIST_BIT(c) \ |
|
|
1440 (b[((unsigned char) (c)) / BYTEWIDTH] \ |
|
|
1441 |= 1 << (((unsigned char) c) % BYTEWIDTH)) |
|
|
1442 |
|
|
1443 |
|
|
1444 /* Get the next unsigned number in the uncompiled pattern. */ |
|
|
1445 #define GET_UNSIGNED_NUMBER(num) \ |
|
|
1446 { if (p != pend) \ |
|
|
1447 { \ |
|
|
1448 PATFETCH (c); \ |
|
28
|
1449 while (ISDIGIT (c)) \ |
|
2
|
1450 { \ |
|
|
1451 if (num < 0) \ |
|
|
1452 num = 0; \ |
|
|
1453 num = num * 10 + c - '0'; \ |
|
|
1454 if (p == pend) \ |
|
|
1455 break; \ |
|
|
1456 PATFETCH (c); \ |
|
|
1457 } \ |
|
|
1458 } \ |
|
|
1459 } |
|
|
1460 |
|
|
1461 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ |
|
|
1462 |
|
|
1463 #define IS_CHAR_CLASS(string) \ |
|
|
1464 (STREQ (string, "alpha") || STREQ (string, "upper") \ |
|
|
1465 || STREQ (string, "lower") || STREQ (string, "digit") \ |
|
|
1466 || STREQ (string, "alnum") || STREQ (string, "xdigit") \ |
|
|
1467 || STREQ (string, "space") || STREQ (string, "print") \ |
|
|
1468 || STREQ (string, "punct") || STREQ (string, "graph") \ |
|
|
1469 || STREQ (string, "cntrl") || STREQ (string, "blank")) |
|
|
1470 |
|
|
1471 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
|
|
1472 Returns one of error codes defined in `regex.h', or zero for success. |
|
|
1473 |
|
|
1474 Assumes the `allocated' (and perhaps `buffer') and `translate' |
|
|
1475 fields are set in BUFP on entry. |
|
|
1476 |
|
|
1477 If it succeeds, results are put in BUFP (if it returns an error, the |
|
|
1478 contents of BUFP are undefined): |
|
|
1479 `buffer' is the compiled pattern; |
|
|
1480 `syntax' is set to SYNTAX; |
|
|
1481 `used' is set to the length of the compiled pattern; |
|
23
|
1482 `fastmap_accurate' is zero; |
|
|
1483 `re_nsub' is the number of subexpressions in PATTERN; |
|
|
1484 `not_bol' and `not_eol' are zero; |
|
2
|
1485 |
|
|
1486 The `fastmap' and `newline_anchor' fields are neither |
|
|
1487 examined nor set. */ |
|
|
1488 |
|
|
1489 static reg_errcode_t |
|
|
1490 regex_compile (pattern, size, syntax, bufp) |
|
|
1491 const char *pattern; |
|
|
1492 int size; |
|
|
1493 reg_syntax_t syntax; |
|
|
1494 struct re_pattern_buffer *bufp; |
|
|
1495 { |
|
|
1496 /* We fetch characters from PATTERN here. Even though PATTERN is |
|
|
1497 `char *' (i.e., signed), we declare these variables as unsigned, so |
|
|
1498 they can be reliably used as array indices. */ |
|
|
1499 register unsigned char c, c1; |
|
|
1500 |
|
|
1501 /* A random tempory spot in PATTERN. */ |
|
|
1502 const char *p1; |
|
|
1503 |
|
|
1504 /* Points to the end of the buffer, where we should append. */ |
|
|
1505 register unsigned char *b; |
|
|
1506 |
|
|
1507 /* Keeps track of unclosed groups. */ |
|
|
1508 compile_stack_type compile_stack; |
|
|
1509 |
|
|
1510 /* Points to the current (ending) position in the pattern. */ |
|
|
1511 const char *p = pattern; |
|
|
1512 const char *pend = pattern + size; |
|
|
1513 |
|
|
1514 /* How to translate the characters in the pattern. */ |
|
|
1515 char *translate = bufp->translate; |
|
|
1516 |
|
|
1517 /* Address of the count-byte of the most recently inserted `exactn' |
|
|
1518 command. This makes it possible to tell if a new exact-match |
|
|
1519 character can be added to that command or if the character requires |
|
|
1520 a new `exactn' command. */ |
|
|
1521 unsigned char *pending_exact = 0; |
|
|
1522 |
|
|
1523 /* Address of start of the most recently finished expression. |
|
|
1524 This tells, e.g., postfix * where to find the start of its |
|
|
1525 operand. Reset at the beginning of groups and alternatives. */ |
|
|
1526 unsigned char *laststart = 0; |
|
|
1527 |
|
|
1528 /* Address of beginning of regexp, or inside of last group. */ |
|
|
1529 unsigned char *begalt; |
|
|
1530 |
|
|
1531 /* Place in the uncompiled pattern (i.e., the {) to |
|
|
1532 which to go back if the interval is invalid. */ |
|
|
1533 const char *beg_interval; |
|
|
1534 |
|
|
1535 /* Address of the place where a forward jump should go to the end of |
|
|
1536 the containing expression. Each alternative of an `or' -- except the |
|
|
1537 last -- ends with a forward jump of this sort. */ |
|
|
1538 unsigned char *fixup_alt_jump = 0; |
|
|
1539 |
|
|
1540 /* Counts open-groups as they are encountered. Remembered for the |
|
|
1541 matching close-group on the compile stack, so the same register |
|
|
1542 number is put in the stop_memory as the start_memory. */ |
|
|
1543 regnum_t regnum = 0; |
|
|
1544 |
|
|
1545 #ifdef DEBUG |
|
|
1546 DEBUG_PRINT1 ("\nCompiling pattern: "); |
|
|
1547 if (debug) |
|
|
1548 { |
|
|
1549 unsigned debug_count; |
|
|
1550 |
|
|
1551 for (debug_count = 0; debug_count < size; debug_count++) |
|
|
1552 printchar (pattern[debug_count]); |
|
|
1553 putchar ('\n'); |
|
|
1554 } |
|
|
1555 #endif /* DEBUG */ |
|
|
1556 |
|
|
1557 /* Initialize the compile stack. */ |
|
|
1558 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); |
|
|
1559 if (compile_stack.stack == NULL) |
|
|
1560 return REG_ESPACE; |
|
|
1561 |
|
|
1562 compile_stack.size = INIT_COMPILE_STACK_SIZE; |
|
|
1563 compile_stack.avail = 0; |
|
|
1564 |
|
|
1565 /* Initialize the pattern buffer. */ |
|
|
1566 bufp->syntax = syntax; |
|
|
1567 bufp->fastmap_accurate = 0; |
|
|
1568 bufp->not_bol = bufp->not_eol = 0; |
|
|
1569 |
|
|
1570 /* Set `used' to zero, so that if we return an error, the pattern |
|
|
1571 printer (for debugging) will think there's no pattern. We reset it |
|
|
1572 at the end. */ |
|
|
1573 bufp->used = 0; |
|
|
1574 |
|
|
1575 /* Always count groups, whether or not bufp->no_sub is set. */ |
|
|
1576 bufp->re_nsub = 0; |
|
|
1577 |
|
|
1578 #if !defined (emacs) && !defined (SYNTAX_TABLE) |
|
|
1579 /* Initialize the syntax table. */ |
|
|
1580 init_syntax_once (); |
|
|
1581 #endif |
|
|
1582 |
|
|
1583 if (bufp->allocated == 0) |
|
|
1584 { |
|
|
1585 if (bufp->buffer) |
|
|
1586 { /* If zero allocated, but buffer is non-null, try to realloc |
|
|
1587 enough space. This loses if buffer's address is bogus, but |
|
|
1588 that is the user's responsibility. */ |
|
|
1589 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); |
|
|
1590 } |
|
|
1591 else |
|
|
1592 { /* Caller did not allocate a buffer. Do it for them. */ |
|
|
1593 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); |
|
|
1594 } |
|
|
1595 if (!bufp->buffer) return REG_ESPACE; |
|
|
1596 |
|
|
1597 bufp->allocated = INIT_BUF_SIZE; |
|
|
1598 } |
|
|
1599 |
|
|
1600 begalt = b = bufp->buffer; |
|
|
1601 |
|
|
1602 /* Loop through the uncompiled pattern until we're at the end. */ |
|
|
1603 while (p != pend) |
|
|
1604 { |
|
|
1605 PATFETCH (c); |
|
|
1606 |
|
|
1607 switch (c) |
|
|
1608 { |
|
|
1609 case '^': |
|
|
1610 { |
|
|
1611 if ( /* If at start of pattern, it's an operator. */ |
|
|
1612 p == pattern + 1 |
|
|
1613 /* If context independent, it's an operator. */ |
|
|
1614 || syntax & RE_CONTEXT_INDEP_ANCHORS |
|
|
1615 /* Otherwise, depends on what's come before. */ |
|
|
1616 || at_begline_loc_p (pattern, p, syntax)) |
|
|
1617 BUF_PUSH (begline); |
|
|
1618 else |
|
|
1619 goto normal_char; |
|
|
1620 } |
|
|
1621 break; |
|
|
1622 |
|
|
1623 |
|
|
1624 case '$': |
|
|
1625 { |
|
|
1626 if ( /* If at end of pattern, it's an operator. */ |
|
|
1627 p == pend |
|
|
1628 /* If context independent, it's an operator. */ |
|
|
1629 || syntax & RE_CONTEXT_INDEP_ANCHORS |
|
|
1630 /* Otherwise, depends on what's next. */ |
|
|
1631 || at_endline_loc_p (p, pend, syntax)) |
|
|
1632 BUF_PUSH (endline); |
|
|
1633 else |
|
|
1634 goto normal_char; |
|
|
1635 } |
|
|
1636 break; |
|
|
1637 |
|
|
1638 |
|
|
1639 case '+': |
|
|
1640 case '?': |
|
|
1641 if ((syntax & RE_BK_PLUS_QM) |
|
|
1642 || (syntax & RE_LIMITED_OPS)) |
|
|
1643 goto normal_char; |
|
|
1644 handle_plus: |
|
|
1645 case '*': |
|
|
1646 /* If there is no previous pattern... */ |
|
|
1647 if (!laststart) |
|
|
1648 { |
|
|
1649 if (syntax & RE_CONTEXT_INVALID_OPS) |
|
|
1650 return REG_BADRPT; |
|
|
1651 else if (!(syntax & RE_CONTEXT_INDEP_OPS)) |
|
|
1652 goto normal_char; |
|
|
1653 } |
|
|
1654 |
|
|
1655 { |
|
|
1656 /* Are we optimizing this jump? */ |
|
|
1657 boolean keep_string_p = false; |
|
|
1658 |
|
|
1659 /* 1 means zero (many) matches is allowed. */ |
|
|
1660 char zero_times_ok = 0, many_times_ok = 0; |
|
|
1661 |
|
|
1662 /* If there is a sequence of repetition chars, collapse it |
|
|
1663 down to just one (the right one). We can't combine |
|
|
1664 interval operators with these because of, e.g., `a{2}*', |
|
|
1665 which should only match an even number of `a's. */ |
|
|
1666 |
|
|
1667 for (;;) |
|
|
1668 { |
|
|
1669 zero_times_ok |= c != '+'; |
|
|
1670 many_times_ok |= c != '?'; |
|
|
1671 |
|
|
1672 if (p == pend) |
|
|
1673 break; |
|
|
1674 |
|
|
1675 PATFETCH (c); |
|
|
1676 |
|
|
1677 if (c == '*' |
|
|
1678 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) |
|
|
1679 ; |
|
|
1680 |
|
|
1681 else if (syntax & RE_BK_PLUS_QM && c == '\\') |
|
|
1682 { |
|
|
1683 if (p == pend) return REG_EESCAPE; |
|
|
1684 |
|
|
1685 PATFETCH (c1); |
|
|
1686 if (!(c1 == '+' || c1 == '?')) |
|
|
1687 { |
|
|
1688 PATUNFETCH; |
|
|
1689 PATUNFETCH; |
|
|
1690 break; |
|
|
1691 } |
|
|
1692 |
|
|
1693 c = c1; |
|
|
1694 } |
|
|
1695 else |
|
|
1696 { |
|
|
1697 PATUNFETCH; |
|
|
1698 break; |
|
|
1699 } |
|
|
1700 |
|
|
1701 /* If we get here, we found another repeat character. */ |
|
|
1702 } |
|
|
1703 |
|
|
1704 /* Star, etc. applied to an empty pattern is equivalent |
|
|
1705 to an empty pattern. */ |
|
|
1706 if (!laststart) |
|
|
1707 break; |
|
|
1708 |
|
|
1709 /* Now we know whether or not zero matches is allowed |
|
|
1710 and also whether or not two or more matches is allowed. */ |
|
|
1711 if (many_times_ok) |
|
|
1712 { /* More than one repetition is allowed, so put in at the |
|
|
1713 end a backward relative jump from `b' to before the next |
|
|
1714 jump we're going to put in below (which jumps from |
|
|
1715 laststart to after this jump). |
|
|
1716 |
|
|
1717 But if we are at the `*' in the exact sequence `.*\n', |
|
|
1718 insert an unconditional jump backwards to the ., |
|
|
1719 instead of the beginning of the loop. This way we only |
|
|
1720 push a failure point once, instead of every time |
|
|
1721 through the loop. */ |
|
|
1722 assert (p - 1 > pattern); |
|
|
1723 |
|
|
1724 /* Allocate the space for the jump. */ |
|
|
1725 GET_BUFFER_SPACE (3); |
|
|
1726 |
|
|
1727 /* We know we are not at the first character of the pattern, |
|
|
1728 because laststart was nonzero. And we've already |
|
|
1729 incremented `p', by the way, to be the character after |
|
|
1730 the `*'. Do we have to do something analogous here |
|
|
1731 for null bytes, because of RE_DOT_NOT_NULL? */ |
|
|
1732 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') |
|
45
|
1733 && zero_times_ok |
|
2
|
1734 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') |
|
|
1735 && !(syntax & RE_DOT_NEWLINE)) |
|
|
1736 { /* We have .*\n. */ |
|
|
1737 STORE_JUMP (jump, b, laststart); |
|
|
1738 keep_string_p = true; |
|
|
1739 } |
|
|
1740 else |
|
|
1741 /* Anything else. */ |
|
|
1742 STORE_JUMP (maybe_pop_jump, b, laststart - 3); |
|
|
1743 |
|
|
1744 /* We've added more stuff to the buffer. */ |
|
|
1745 b += 3; |
|
|
1746 } |
|
|
1747 |
|
|
1748 /* On failure, jump from laststart to b + 3, which will be the |
|
|
1749 end of the buffer after this jump is inserted. */ |
|
|
1750 GET_BUFFER_SPACE (3); |
|
|
1751 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump |
|
|
1752 : on_failure_jump, |
|
|
1753 laststart, b + 3); |
|
|
1754 pending_exact = 0; |
|
|
1755 b += 3; |
|
|
1756 |
|
|
1757 if (!zero_times_ok) |
|
|
1758 { |
|
|
1759 /* At least one repetition is required, so insert a |
|
|
1760 `dummy_failure_jump' before the initial |
|
|
1761 `on_failure_jump' instruction of the loop. This |
|
|
1762 effects a skip over that instruction the first time |
|
|
1763 we hit that loop. */ |
|
|
1764 GET_BUFFER_SPACE (3); |
|
|
1765 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); |
|
|
1766 b += 3; |
|
|
1767 } |
|
|
1768 } |
|
|
1769 break; |
|
|
1770 |
|
|
1771 |
|
|
1772 case '.': |
|
|
1773 laststart = b; |
|
|
1774 BUF_PUSH (anychar); |
|
|
1775 break; |
|
|
1776 |
|
|
1777 |
|
|
1778 case '[': |
|
|
1779 { |
|
|
1780 boolean had_char_class = false; |
|
|
1781 |
|
|
1782 if (p == pend) return REG_EBRACK; |
|
|
1783 |
|
|
1784 /* Ensure that we have enough space to push a charset: the |
|
|
1785 opcode, the length count, and the bitset; 34 bytes in all. */ |
|
|
1786 GET_BUFFER_SPACE (34); |
|
|
1787 |
|
|
1788 laststart = b; |
|
|
1789 |
|
|
1790 /* We test `*p == '^' twice, instead of using an if |
|
|
1791 statement, so we only need one BUF_PUSH. */ |
|
|
1792 BUF_PUSH (*p == '^' ? charset_not : charset); |
|
|
1793 if (*p == '^') |
|
|
1794 p++; |
|
|
1795 |
|
|
1796 /* Remember the first position in the bracket expression. */ |
|
|
1797 p1 = p; |
|
|
1798 |
|
|
1799 /* Push the number of bytes in the bitmap. */ |
|
|
1800 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); |
|
|
1801 |
|
|
1802 /* Clear the whole map. */ |
|
|
1803 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); |
|
|
1804 |
|
|
1805 /* charset_not matches newline according to a syntax bit. */ |
|
|
1806 if ((re_opcode_t) b[-2] == charset_not |
|
|
1807 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) |
|
|
1808 SET_LIST_BIT ('\n'); |
|
|
1809 |
|
|
1810 /* Read in characters and ranges, setting map bits. */ |
|
|
1811 for (;;) |
|
|
1812 { |
|
|
1813 if (p == pend) return REG_EBRACK; |
|
|
1814 |
|
|
1815 PATFETCH (c); |
|
|
1816 |
|
|
1817 /* \ might escape characters inside [...] and [^...]. */ |
|
|
1818 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') |
|
|
1819 { |
|
|
1820 if (p == pend) return REG_EESCAPE; |
|
|
1821 |
|
|
1822 PATFETCH (c1); |
|
|
1823 SET_LIST_BIT (c1); |
|
|
1824 continue; |
|
|
1825 } |
|
|
1826 |
|
|
1827 /* Could be the end of the bracket expression. If it's |
|
|
1828 not (i.e., when the bracket expression is `[]' so |
|
|
1829 far), the ']' character bit gets set way below. */ |
|
|
1830 if (c == ']' && p != p1 + 1) |
|
|
1831 break; |
|
|
1832 |
|
|
1833 /* Look ahead to see if it's a range when the last thing |
|
|
1834 was a character class. */ |
|
|
1835 if (had_char_class && c == '-' && *p != ']') |
|
|
1836 return REG_ERANGE; |
|
|
1837 |
|
|
1838 /* Look ahead to see if it's a range when the last thing |
|
|
1839 was a character: if this is a hyphen not at the |
|
|
1840 beginning or the end of a list, then it's the range |
|
|
1841 operator. */ |
|
|
1842 if (c == '-' |
|
|
1843 && !(p - 2 >= pattern && p[-2] == '[') |
|
|
1844 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') |
|
|
1845 && *p != ']') |
|
|
1846 { |
|
|
1847 reg_errcode_t ret |
|
|
1848 = compile_range (&p, pend, translate, syntax, b); |
|
|
1849 if (ret != REG_NOERROR) return ret; |
|
|
1850 } |
|
|
1851 |
|
|
1852 else if (p[0] == '-' && p[1] != ']') |
|
|
1853 { /* This handles ranges made up of characters only. */ |
|
|
1854 reg_errcode_t ret; |
|
|
1855 |
|
|
1856 /* Move past the `-'. */ |
|
|
1857 PATFETCH (c1); |
|
|
1858 |
|
|
1859 ret = compile_range (&p, pend, translate, syntax, b); |
|
|
1860 if (ret != REG_NOERROR) return ret; |
|
|
1861 } |
|
|
1862 |
|
|
1863 /* See if we're at the beginning of a possible character |
|
|
1864 class. */ |
|
|
1865 |
|
|
1866 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') |
|
|
1867 { /* Leave room for the null. */ |
|
|
1868 char str[CHAR_CLASS_MAX_LENGTH + 1]; |
|
|
1869 |
|
|
1870 PATFETCH (c); |
|
|
1871 c1 = 0; |
|
|
1872 |
|
|
1873 /* If pattern is `[[:'. */ |
|
|
1874 if (p == pend) return REG_EBRACK; |
|
|
1875 |
|
|
1876 for (;;) |
|
|
1877 { |
|
|
1878 PATFETCH (c); |
|
|
1879 if (c == ':' || c == ']' || p == pend |
|
|
1880 || c1 == CHAR_CLASS_MAX_LENGTH) |
|
|
1881 break; |
|
|
1882 str[c1++] = c; |
|
|
1883 } |
|
|
1884 str[c1] = '\0'; |
|
|
1885 |
|
|
1886 /* If isn't a word bracketed by `[:' and:`]': |
|
|
1887 undo the ending character, the letters, and leave |
|
|
1888 the leading `:' and `[' (but set bits for them). */ |
|
|
1889 if (c == ':' && *p == ']') |
|
|
1890 { |
|
|
1891 int ch; |
|
|
1892 boolean is_alnum = STREQ (str, "alnum"); |
|
|
1893 boolean is_alpha = STREQ (str, "alpha"); |
|
|
1894 boolean is_blank = STREQ (str, "blank"); |
|
|
1895 boolean is_cntrl = STREQ (str, "cntrl"); |
|
|
1896 boolean is_digit = STREQ (str, "digit"); |
|
|
1897 boolean is_graph = STREQ (str, "graph"); |
|
|
1898 boolean is_lower = STREQ (str, "lower"); |
|
|
1899 boolean is_print = STREQ (str, "print"); |
|
|
1900 boolean is_punct = STREQ (str, "punct"); |
|
|
1901 boolean is_space = STREQ (str, "space"); |
|
|
1902 boolean is_upper = STREQ (str, "upper"); |
|
|
1903 boolean is_xdigit = STREQ (str, "xdigit"); |
|
|
1904 |
|
|
1905 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE; |
|
|
1906 |
|
|
1907 /* Throw away the ] at the end of the character |
|
|
1908 class. */ |
|
|
1909 PATFETCH (c); |
|
|
1910 |
|
|
1911 if (p == pend) return REG_EBRACK; |
|
|
1912 |
|
|
1913 for (ch = 0; ch < 1 << BYTEWIDTH; ch++) |
|
|
1914 { |
|
28
|
1915 if ( (is_alnum && ISALNUM (ch)) |
|
|
1916 || (is_alpha && ISALPHA (ch)) |
|
|
1917 || (is_blank && ISBLANK (ch)) |
|
|
1918 || (is_cntrl && ISCNTRL (ch)) |
|
|
1919 || (is_digit && ISDIGIT (ch)) |
|
|
1920 || (is_graph && ISGRAPH (ch)) |
|
|
1921 || (is_lower && ISLOWER (ch)) |
|
|
1922 || (is_print && ISPRINT (ch)) |
|
|
1923 || (is_punct && ISPUNCT (ch)) |
|
|
1924 || (is_space && ISSPACE (ch)) |
|
|
1925 || (is_upper && ISUPPER (ch)) |
|
|
1926 || (is_xdigit && ISXDIGIT (ch))) |
|
2
|
1927 SET_LIST_BIT (ch); |
|
|
1928 } |
|
|
1929 had_char_class = true; |
|
|
1930 } |
|
|
1931 else |
|
|
1932 { |
|
|
1933 c1++; |
|
|
1934 while (c1--) |
|
|
1935 PATUNFETCH; |
|
|
1936 SET_LIST_BIT ('['); |
|
|
1937 SET_LIST_BIT (':'); |
|
|
1938 had_char_class = false; |
|
|
1939 } |
|
|
1940 } |
|
|
1941 else |
|
|
1942 { |
|
|
1943 had_char_class = false; |
|
|
1944 SET_LIST_BIT (c); |
|
|
1945 } |
|
|
1946 } |
|
|
1947 |
|
|
1948 /* Discard any (non)matching list bytes that are all 0 at the |
|
|
1949 end of the map. Decrease the map-length byte too. */ |
|
|
1950 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) |
|
|
1951 b[-1]--; |
|
|
1952 b += b[-1]; |
|
|
1953 } |
|
|
1954 break; |
|
|
1955 |
|
|
1956 |
|
|
1957 case '(': |
|
|
1958 if (syntax & RE_NO_BK_PARENS) |
|
|
1959 goto handle_open; |
|
|
1960 else |
|
|
1961 goto normal_char; |
|
|
1962 |
|
|
1963 |
|
|
1964 case ')': |
|
|
1965 if (syntax & RE_NO_BK_PARENS) |
|
|
1966 goto handle_close; |
|
|
1967 else |
|
|
1968 goto normal_char; |
|
|
1969 |
|
|
1970 |
|
|
1971 case '\n': |
|
|
1972 if (syntax & RE_NEWLINE_ALT) |
|
|
1973 goto handle_alt; |
|
|
1974 else |
|
|
1975 goto normal_char; |
|
|
1976 |
|
|
1977 |
|
|
1978 case '|': |
|
|
1979 if (syntax & RE_NO_BK_VBAR) |
|
|
1980 goto handle_alt; |
|
|
1981 else |
|
|
1982 goto normal_char; |
|
|
1983 |
|
|
1984 |
|
|
1985 case '{': |
|
|
1986 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) |
|
|
1987 goto handle_interval; |
|
|
1988 else |
|
|
1989 goto normal_char; |
|
|
1990 |
|
|
1991 |
|
|
1992 case '\\': |
|
|
1993 if (p == pend) return REG_EESCAPE; |
|
|
1994 |
|
|
1995 /* Do not translate the character after the \, so that we can |
|
|
1996 distinguish, e.g., \B from \b, even if we normally would |
|
|
1997 translate, e.g., B to b. */ |
|
|
1998 PATFETCH_RAW (c); |
|
|
1999 |
|
|
2000 switch (c) |
|
|
2001 { |
|
|
2002 case '(': |
|
|
2003 if (syntax & RE_NO_BK_PARENS) |
|
|
2004 goto normal_backslash; |
|
|
2005 |
|
|
2006 handle_open: |
|
|
2007 bufp->re_nsub++; |
|
|
2008 regnum++; |
|
|
2009 |
|
|
2010 if (COMPILE_STACK_FULL) |
|
|
2011 { |
|
|
2012 RETALLOC (compile_stack.stack, compile_stack.size << 1, |
|
|
2013 compile_stack_elt_t); |
|
|
2014 if (compile_stack.stack == NULL) return REG_ESPACE; |
|
|
2015 |
|
|
2016 compile_stack.size <<= 1; |
|
|
2017 } |
|
|
2018 |
|
|
2019 /* These are the values to restore when we hit end of this |
|
|
2020 group. They are all relative offsets, so that if the |
|
|
2021 whole pattern moves because of realloc, they will still |
|
|
2022 be valid. */ |
|
|
2023 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; |
|
|
2024 COMPILE_STACK_TOP.fixup_alt_jump |
|
|
2025 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; |
|
|
2026 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; |
|
|
2027 COMPILE_STACK_TOP.regnum = regnum; |
|
|
2028 |
|
|
2029 /* We will eventually replace the 0 with the number of |
|
|
2030 groups inner to this one. But do not push a |
|
|
2031 start_memory for groups beyond the last one we can |
|
|
2032 represent in the compiled pattern. */ |
|
|
2033 if (regnum <= MAX_REGNUM) |
|
|
2034 { |
|
|
2035 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; |
|
|
2036 BUF_PUSH_3 (start_memory, regnum, 0); |
|
|
2037 } |
|
|
2038 |
|
|
2039 compile_stack.avail++; |
|
|
2040 |
|
|
2041 fixup_alt_jump = 0; |
|
|
2042 laststart = 0; |
|
|
2043 begalt = b; |
|
45
|
2044 /* If we've reached MAX_REGNUM groups, then this open |
|
|
2045 won't actually generate any code, so we'll have to |
|
|
2046 clear pending_exact explicitly. */ |
|
|
2047 pending_exact = 0; |
|
2
|
2048 break; |
|
|
2049 |
|
|
2050 |
|
|
2051 case ')': |
|
|
2052 if (syntax & RE_NO_BK_PARENS) goto normal_backslash; |
|
|
2053 |
|
|
2054 if (COMPILE_STACK_EMPTY) |
|
|
2055 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
|
|
2056 goto normal_backslash; |
|
|
2057 else |
|
|
2058 return REG_ERPAREN; |
|
|
2059 |
|
|
2060 handle_close: |
|
|
2061 if (fixup_alt_jump) |
|
|
2062 { /* Push a dummy failure point at the end of the |
|
|
2063 alternative for a possible future |
|
|
2064 `pop_failure_jump' to pop. See comments at |
|
|
2065 `push_dummy_failure' in `re_match_2'. */ |
|
|
2066 BUF_PUSH (push_dummy_failure); |
|
|
2067 |
|
|
2068 /* We allocated space for this jump when we assigned |
|
|
2069 to `fixup_alt_jump', in the `handle_alt' case below. */ |
|
|
2070 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); |
|
|
2071 } |
|
|
2072 |
|
|
2073 /* See similar code for backslashed left paren above. */ |
|
|
2074 if (COMPILE_STACK_EMPTY) |
|
|
2075 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
|
|
2076 goto normal_char; |
|
|
2077 else |
|
|
2078 return REG_ERPAREN; |
|
|
2079 |
|
|
2080 /* Since we just checked for an empty stack above, this |
|
|
2081 ``can't happen''. */ |
|
|
2082 assert (compile_stack.avail != 0); |
|
|
2083 { |
|
|
2084 /* We don't just want to restore into `regnum', because |
|
|
2085 later groups should continue to be numbered higher, |
|
|
2086 as in `(ab)c(de)' -- the second group is #2. */ |
|
|
2087 regnum_t this_group_regnum; |
|
|
2088 |
|
|
2089 compile_stack.avail--; |
|
|
2090 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; |
|
|
2091 fixup_alt_jump |
|
|
2092 = COMPILE_STACK_TOP.fixup_alt_jump |
|
|
2093 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 |
|
|
2094 : 0; |
|
|
2095 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; |
|
|
2096 this_group_regnum = COMPILE_STACK_TOP.regnum; |
|
45
|
2097 /* If we've reached MAX_REGNUM groups, then this open |
|
|
2098 won't actually generate any code, so we'll have to |
|
|
2099 clear pending_exact explicitly. */ |
|
|
2100 pending_exact = 0; |
|
2
|
2101 |
|
|
2102 /* We're at the end of the group, so now we know how many |
|
|
2103 groups were inside this one. */ |
|
|
2104 if (this_group_regnum <= MAX_REGNUM) |
|
|
2105 { |
|
|
2106 unsigned char *inner_group_loc |
|
|
2107 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; |
|
|
2108 |
|
|
2109 *inner_group_loc = regnum - this_group_regnum; |
|
|
2110 BUF_PUSH_3 (stop_memory, this_group_regnum, |
|
|
2111 regnum - this_group_regnum); |
|
|
2112 } |
|
|
2113 } |
|
|
2114 break; |
|
|
2115 |
|
|
2116 |
|
|
2117 case '|': /* `\|'. */ |
|
|
2118 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) |
|
|
2119 goto normal_backslash; |
|
|
2120 handle_alt: |
|
|
2121 if (syntax & RE_LIMITED_OPS) |
|
|
2122 goto normal_char; |
|
|
2123 |
|
|
2124 /* Insert before the previous alternative a jump which |
|
|
2125 jumps to this alternative if the former fails. */ |
|
|
2126 GET_BUFFER_SPACE (3); |
|
|
2127 INSERT_JUMP (on_failure_jump, begalt, b + 6); |
|
|
2128 pending_exact = 0; |
|
|
2129 b += 3; |
|
|
2130 |
|
|
2131 /* The alternative before this one has a jump after it |
|
|
2132 which gets executed if it gets matched. Adjust that |
|
|
2133 jump so it will jump to this alternative's analogous |
|
|
2134 jump (put in below, which in turn will jump to the next |
|
|
2135 (if any) alternative's such jump, etc.). The last such |
|
|
2136 jump jumps to the correct final destination. A picture: |
|
|
2137 _____ _____ |
|
|
2138 | | | | |
|
|
2139 | v | v |
|
|
2140 a | b | c |
|
|
2141 |
|
23
|
2142 If we are at `b', then fixup_alt_jump right now points to a |
|
|
2143 three-byte space after `a'. We'll put in the jump, set |
|
|
2144 fixup_alt_jump to right after `b', and leave behind three |
|
|
2145 bytes which we'll fill in when we get to after `c'. */ |
|
2
|
2146 |
|
|
2147 if (fixup_alt_jump) |
|
|
2148 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); |
|
|
2149 |
|
|
2150 /* Mark and leave space for a jump after this alternative, |
|
|
2151 to be filled in later either by next alternative or |
|
|
2152 when know we're at the end of a series of alternatives. */ |
|
|
2153 fixup_alt_jump = b; |
|
|
2154 GET_BUFFER_SPACE (3); |
|
|
2155 b += 3; |
|
|
2156 |
|
|
2157 laststart = 0; |
|
|
2158 begalt = b; |
|
|
2159 break; |
|
|
2160 |
|
|
2161 |
|
|
2162 case '{': |
|
|
2163 /* If \{ is a literal. */ |
|
|
2164 if (!(syntax & RE_INTERVALS) |
|
|
2165 /* If we're at `\{' and it's not the open-interval |
|
|
2166 operator. */ |
|
|
2167 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) |
|
|
2168 || (p - 2 == pattern && p == pend)) |
|
|
2169 goto normal_backslash; |
|
|
2170 |
|
|
2171 handle_interval: |
|
|
2172 { |
|
|
2173 /* If got here, then the syntax allows intervals. */ |
|
|
2174 |
|
|
2175 /* At least (most) this many matches must be made. */ |
|
|
2176 int lower_bound = -1, upper_bound = -1; |
|
|
2177 |
|
|
2178 beg_interval = p - 1; |
|
|
2179 |
|
|
2180 if (p == pend) |
|
|
2181 { |
|
|
2182 if (syntax & RE_NO_BK_BRACES) |
|
|
2183 goto unfetch_interval; |
|
|
2184 else |
|
|
2185 return REG_EBRACE; |
|
|
2186 } |
|
|
2187 |
|
|
2188 GET_UNSIGNED_NUMBER (lower_bound); |
|
|
2189 |
|
|
2190 if (c == ',') |
|
|
2191 { |
|
|
2192 GET_UNSIGNED_NUMBER (upper_bound); |
|
|
2193 if (upper_bound < 0) upper_bound = RE_DUP_MAX; |
|
|
2194 } |
|
|
2195 else |
|
|
2196 /* Interval such as `{1}' => match exactly once. */ |
|
|
2197 upper_bound = lower_bound; |
|
|
2198 |
|
|
2199 if (lower_bound < 0 || upper_bound > RE_DUP_MAX |
|
|
2200 || lower_bound > upper_bound) |
|
|
2201 { |
|
|
2202 if (syntax & RE_NO_BK_BRACES) |
|
|
2203 goto unfetch_interval; |
|
|
2204 else |
|
|
2205 return REG_BADBR; |
|
|
2206 } |
|
|
2207 |
|
|
2208 if (!(syntax & RE_NO_BK_BRACES)) |
|
|
2209 { |
|
|
2210 if (c != '\\') return REG_EBRACE; |
|
|
2211 |
|
|
2212 PATFETCH (c); |
|
|
2213 } |
|
|
2214 |
|
|
2215 if (c != '}') |
|
|
2216 { |
|
|
2217 if (syntax & RE_NO_BK_BRACES) |
|
|
2218 goto unfetch_interval; |
|
|
2219 else |
|
|
2220 return REG_BADBR; |
|
|
2221 } |
|
|
2222 |
|
|
2223 /* We just parsed a valid interval. */ |
|
|
2224 |
|
|
2225 /* If it's invalid to have no preceding re. */ |
|
|
2226 if (!laststart) |
|
|
2227 { |
|
|
2228 if (syntax & RE_CONTEXT_INVALID_OPS) |
|
|
2229 return REG_BADRPT; |
|
|
2230 else if (syntax & RE_CONTEXT_INDEP_OPS) |
|
|
2231 laststart = b; |
|
|
2232 else |
|
|
2233 goto unfetch_interval; |
|
|
2234 } |
|
|
2235 |
|
|
2236 /* If the upper bound is zero, don't want to succeed at |
|
|
2237 all; jump from `laststart' to `b + 3', which will be |
|
|
2238 the end of the buffer after we insert the jump. */ |
|
|
2239 if (upper_bound == 0) |
|
|
2240 { |
|
|
2241 GET_BUFFER_SPACE (3); |
|
|
2242 INSERT_JUMP (jump, laststart, b + 3); |
|
|
2243 b += 3; |
|
|
2244 } |
|
|
2245 |
|
|
2246 /* Otherwise, we have a nontrivial interval. When |
|
|
2247 we're all done, the pattern will look like: |
|
|
2248 set_number_at <jump count> <upper bound> |
|
|
2249 set_number_at <succeed_n count> <lower bound> |
|
|
2250 succeed_n <after jump addr> <succed_n count> |
|
|
2251 <body of loop> |
|
|
2252 jump_n <succeed_n addr> <jump count> |
|
|
2253 (The upper bound and `jump_n' are omitted if |
|
|
2254 `upper_bound' is 1, though.) */ |
|
|
2255 else |
|
|
2256 { /* If the upper bound is > 1, we need to insert |
|
|
2257 more at the end of the loop. */ |
|
|
2258 unsigned nbytes = 10 + (upper_bound > 1) * 10; |
|
|
2259 |
|
|
2260 GET_BUFFER_SPACE (nbytes); |
|
|
2261 |
|
|
2262 /* Initialize lower bound of the `succeed_n', even |
|
|
2263 though it will be set during matching by its |
|
|
2264 attendant `set_number_at' (inserted next), |
|
|
2265 because `re_compile_fastmap' needs to know. |
|
|
2266 Jump to the `jump_n' we might insert below. */ |
|
|
2267 INSERT_JUMP2 (succeed_n, laststart, |
|
|
2268 b + 5 + (upper_bound > 1) * 5, |
|
|
2269 lower_bound); |
|
|
2270 b += 5; |
|
|
2271 |
|
|
2272 /* Code to initialize the lower bound. Insert |
|
|
2273 before the `succeed_n'. The `5' is the last two |
|
|
2274 bytes of this `set_number_at', plus 3 bytes of |
|
|
2275 the following `succeed_n'. */ |
|
|
2276 insert_op2 (set_number_at, laststart, 5, lower_bound, b); |
|
|
2277 b += 5; |
|
|
2278 |
|
|
2279 if (upper_bound > 1) |
|
|
2280 { /* More than one repetition is allowed, so |
|
|
2281 append a backward jump to the `succeed_n' |
|
|
2282 that starts this interval. |
|
|
2283 |
|
|
2284 When we've reached this during matching, |
|
|
2285 we'll have matched the interval once, so |
|
|
2286 jump back only `upper_bound - 1' times. */ |
|
|
2287 STORE_JUMP2 (jump_n, b, laststart + 5, |
|
|
2288 upper_bound - 1); |
|
|
2289 b += 5; |
|
|
2290 |
|
|
2291 /* The location we want to set is the second |
|
|
2292 parameter of the `jump_n'; that is `b-2' as |
|
|
2293 an absolute address. `laststart' will be |
|
|
2294 the `set_number_at' we're about to insert; |
|
|
2295 `laststart+3' the number to set, the source |
|
|
2296 for the relative address. But we are |
|
|
2297 inserting into the middle of the pattern -- |
|
|
2298 so everything is getting moved up by 5. |
|
|
2299 Conclusion: (b - 2) - (laststart + 3) + 5, |
|
|
2300 i.e., b - laststart. |
|
|
2301 |
|
|
2302 We insert this at the beginning of the loop |
|
|
2303 so that if we fail during matching, we'll |
|
|
2304 reinitialize the bounds. */ |
|
|
2305 insert_op2 (set_number_at, laststart, b - laststart, |
|
|
2306 upper_bound - 1, b); |
|
|
2307 b += 5; |
|
|
2308 } |
|
|
2309 } |
|
|
2310 pending_exact = 0; |
|
|
2311 beg_interval = NULL; |
|
|
2312 } |
|
|
2313 break; |
|
|
2314 |
|
|
2315 unfetch_interval: |
|
|
2316 /* If an invalid interval, match the characters as literals. */ |
|
|
2317 assert (beg_interval); |
|
|
2318 p = beg_interval; |
|
|
2319 beg_interval = NULL; |
|
|
2320 |
|
|
2321 /* normal_char and normal_backslash need `c'. */ |
|
|
2322 PATFETCH (c); |
|
|
2323 |
|
|
2324 if (!(syntax & RE_NO_BK_BRACES)) |
|
|
2325 { |
|
|
2326 if (p > pattern && p[-1] == '\\') |
|
|
2327 goto normal_backslash; |
|
|
2328 } |
|
|
2329 goto normal_char; |
|
|
2330 |
|
|
2331 #ifdef emacs |
|
|
2332 /* There is no way to specify the before_dot and after_dot |
|
|
2333 operators. rms says this is ok. --karl */ |
|
|
2334 case '=': |
|
|
2335 BUF_PUSH (at_dot); |
|
|
2336 break; |
|
|
2337 |
|
|
2338 case 's': |
|
|
2339 laststart = b; |
|
|
2340 PATFETCH (c); |
|
|
2341 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); |
|
|
2342 break; |
|
|
2343 |
|
|
2344 case 'S': |
|
|
2345 laststart = b; |
|
|
2346 PATFETCH (c); |
|
|
2347 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); |
|
|
2348 break; |
|
|
2349 #endif /* emacs */ |
|
|
2350 |
|
|
2351 |
|
|
2352 case 'w': |
|
|
2353 laststart = b; |
|
|
2354 BUF_PUSH (wordchar); |
|
|
2355 break; |
|
|
2356 |
|
|
2357 |
|
|
2358 case 'W': |
|
|
2359 laststart = b; |
|
|
2360 BUF_PUSH (notwordchar); |
|
|
2361 break; |
|
|
2362 |
|
|
2363 |
|
|
2364 case '<': |
|
|
2365 BUF_PUSH (wordbeg); |
|
|
2366 break; |
|
|
2367 |
|
|
2368 case '>': |
|
|
2369 BUF_PUSH (wordend); |
|
|
2370 break; |
|
|
2371 |
|
|
2372 case 'b': |
|
|
2373 BUF_PUSH (wordbound); |
|
|
2374 break; |
|
|
2375 |
|
|
2376 case 'B': |
|
|
2377 BUF_PUSH (notwordbound); |
|
|
2378 break; |
|
|
2379 |
|
|
2380 case '`': |
|
|
2381 BUF_PUSH (begbuf); |
|
|
2382 break; |
|
|
2383 |
|
|
2384 case '\'': |
|
|
2385 BUF_PUSH (endbuf); |
|
|
2386 break; |
|
|
2387 |
|
|
2388 case '1': case '2': case '3': case '4': case '5': |
|
|
2389 case '6': case '7': case '8': case '9': |
|
|
2390 if (syntax & RE_NO_BK_REFS) |
|
|
2391 goto normal_char; |
|
|
2392 |
|
|
2393 c1 = c - '0'; |
|
|
2394 |
|
|
2395 if (c1 > regnum) |
|
|
2396 return REG_ESUBREG; |
|
|
2397 |
|
|
2398 /* Can't back reference to a subexpression if inside of it. */ |
|
|
2399 if (group_in_compile_stack (compile_stack, c1)) |
|
|
2400 goto normal_char; |
|
|
2401 |
|
|
2402 laststart = b; |
|
|
2403 BUF_PUSH_2 (duplicate, c1); |
|
|
2404 break; |
|
|
2405 |
|
|
2406 |
|
|
2407 case '+': |
|
|
2408 case '?': |
|
|
2409 if (syntax & RE_BK_PLUS_QM) |
|
|
2410 goto handle_plus; |
|
|
2411 else |
|
|
2412 goto normal_backslash; |
|
|
2413 |
|
|
2414 default: |
|
|
2415 normal_backslash: |
|
|
2416 /* You might think it would be useful for \ to mean |
|
|
2417 not to translate; but if we don't translate it |
|
|
2418 it will never match anything. */ |
|
|
2419 c = TRANSLATE (c); |
|
|
2420 goto normal_char; |
|
|
2421 } |
|
|
2422 break; |
|
|
2423 |
|
|
2424 |
|
|
2425 default: |
|
|
2426 /* Expects the character in `c'. */ |
|
|
2427 normal_char: |
|
|
2428 /* If no exactn currently being built. */ |
|
|
2429 if (!pending_exact |
|
|
2430 |
|
|
2431 /* If last exactn not at current position. */ |
|
|
2432 || pending_exact + *pending_exact + 1 != b |
|
|
2433 |
|
|
2434 /* We have only one byte following the exactn for the count. */ |
|
|
2435 || *pending_exact == (1 << BYTEWIDTH) - 1 |
|
|
2436 |
|
|
2437 /* If followed by a repetition operator. */ |
|
|
2438 || *p == '*' || *p == '^' |
|
|
2439 || ((syntax & RE_BK_PLUS_QM) |
|
|
2440 ? *p == '\\' && (p[1] == '+' || p[1] == '?') |
|
|
2441 : (*p == '+' || *p == '?')) |
|
|
2442 || ((syntax & RE_INTERVALS) |
|
|
2443 && ((syntax & RE_NO_BK_BRACES) |
|
|
2444 ? *p == '{' |
|
|
2445 : (p[0] == '\\' && p[1] == '{')))) |
|
|
2446 { |
|
|
2447 /* Start building a new exactn. */ |
|
|
2448 |
|
|
2449 laststart = b; |
|
|
2450 |
|
|
2451 BUF_PUSH_2 (exactn, 0); |
|
|
2452 pending_exact = b - 1; |
|
|
2453 } |
|
|
2454 |
|
|
2455 BUF_PUSH (c); |
|
|
2456 (*pending_exact)++; |
|
|
2457 break; |
|
|
2458 } /* switch (c) */ |
|
|
2459 } /* while p != pend */ |
|
|
2460 |
|
|
2461 |
|
|
2462 /* Through the pattern now. */ |
|
|
2463 |
|
|
2464 if (fixup_alt_jump) |
|
|
2465 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); |
|
|
2466 |
|
|
2467 if (!COMPILE_STACK_EMPTY) |
|
|
2468 return REG_EPAREN; |
|
|
2469 |
|
|
2470 free (compile_stack.stack); |
|
|
2471 |
|
|
2472 /* We have succeeded; set the length of the buffer. */ |
|
|
2473 bufp->used = b - bufp->buffer; |
|
|
2474 |
|
|
2475 #ifdef DEBUG |
|
|
2476 if (debug) |
|
|
2477 { |
|
79
|
2478 DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
|
2
|
2479 print_compiled_pattern (bufp); |
|
|
2480 } |
|
|
2481 #endif /* DEBUG */ |
|
|
2482 |
|
89
|
2483 #ifndef MATCH_MAY_ALLOCATE |
|
87
|
2484 /* Initialize the failure stack to the largest possible stack. This |
|
|
2485 isn't necessary unless we're trying to avoid calling alloca in |
|
|
2486 the search and match routines. */ |
|
|
2487 { |
|
|
2488 int num_regs = bufp->re_nsub + 1; |
|
|
2489 |
|
|
2490 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size |
|
|
2491 is strictly greater than re_max_failures, the largest possible stack |
|
|
2492 is 2 * re_max_failures failure points. */ |
|
|
2493 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); |
|
|
2494 if (fail_stack.stack) |
|
|
2495 fail_stack.stack = |
|
|
2496 (fail_stack_elt_t *) realloc (fail_stack.stack, |
|
|
2497 (fail_stack.size |
|
|
2498 * sizeof (fail_stack_elt_t))); |
|
|
2499 else |
|
|
2500 fail_stack.stack = |
|
|
2501 (fail_stack_elt_t *) malloc (fail_stack.size |
|
|
2502 * sizeof (fail_stack_elt_t)); |
|
|
2503 |
|
|
2504 /* Initialize some other variables the matcher uses. */ |
|
|
2505 RETALLOC_IF (regstart, num_regs, const char *); |
|
|
2506 RETALLOC_IF (regend, num_regs, const char *); |
|
|
2507 RETALLOC_IF (old_regstart, num_regs, const char *); |
|
|
2508 RETALLOC_IF (old_regend, num_regs, const char *); |
|
|
2509 RETALLOC_IF (best_regstart, num_regs, const char *); |
|
|
2510 RETALLOC_IF (best_regend, num_regs, const char *); |
|
|
2511 RETALLOC_IF (reg_info, num_regs, register_info_type); |
|
|
2512 RETALLOC_IF (reg_dummy, num_regs, const char *); |
|
|
2513 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); |
|
|
2514 } |
|
|
2515 #endif |
|
|
2516 |
|
2
|
2517 return REG_NOERROR; |
|
|
2518 } /* regex_compile */ |
|
|
2519 |
|
|
2520 /* Subroutines for `regex_compile'. */ |
|
|
2521 |
|
|
2522 /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
|
|
2523 |
|
|
2524 static void |
|
|
2525 store_op1 (op, loc, arg) |
|
|
2526 re_opcode_t op; |
|
|
2527 unsigned char *loc; |
|
|
2528 int arg; |
|
|
2529 { |
|
|
2530 *loc = (unsigned char) op; |
|
|
2531 STORE_NUMBER (loc + 1, arg); |
|
|
2532 } |
|
|
2533 |
|
|
2534 |
|
|
2535 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ |
|
|
2536 |
|
|
2537 static void |
|
|
2538 store_op2 (op, loc, arg1, arg2) |
|
|
2539 re_opcode_t op; |
|
|
2540 unsigned char *loc; |
|
|
2541 int arg1, arg2; |
|
|
2542 { |
|
|
2543 *loc = (unsigned char) op; |
|
|
2544 STORE_NUMBER (loc + 1, arg1); |
|
|
2545 STORE_NUMBER (loc + 3, arg2); |
|
|
2546 } |
|
|
2547 |
|
|
2548 |
|
|
2549 /* Copy the bytes from LOC to END to open up three bytes of space at LOC |
|
|
2550 for OP followed by two-byte integer parameter ARG. */ |
|
|
2551 |
|
|
2552 static void |
|
|
2553 insert_op1 (op, loc, arg, end) |
|
|
2554 re_opcode_t op; |
|
|
2555 unsigned char *loc; |
|
|
2556 int arg; |
|
|
2557 unsigned char *end; |
|
|
2558 { |
|
|
2559 register unsigned char *pfrom = end; |
|
|
2560 register unsigned char *pto = end + 3; |
|
|
2561 |
|
|
2562 while (pfrom != loc) |
|
|
2563 *--pto = *--pfrom; |
|
|
2564 |
|
|
2565 store_op1 (op, loc, arg); |
|
|
2566 } |
|
|
2567 |
|
|
2568 |
|
|
2569 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ |
|
|
2570 |
|
|
2571 static void |
|
|
2572 insert_op2 (op, loc, arg1, arg2, end) |
|
|
2573 re_opcode_t op; |
|
|
2574 unsigned char *loc; |
|
|
2575 int arg1, arg2; |
|
|
2576 unsigned char *end; |
|
|
2577 { |
|
|
2578 register unsigned char *pfrom = end; |
|
|
2579 register unsigned char *pto = end + 5; |
|
|
2580 |
|
|
2581 while (pfrom != loc) |
|
|
2582 *--pto = *--pfrom; |
|
|
2583 |
|
|
2584 store_op2 (op, loc, arg1, arg2); |
|
|
2585 } |
|
|
2586 |
|
|
2587 |
|
|
2588 /* P points to just after a ^ in PATTERN. Return true if that ^ comes |
|
|
2589 after an alternative or a begin-subexpression. We assume there is at |
|
|
2590 least one character before the ^. */ |
|
|
2591 |
|
|
2592 static boolean |
|
|
2593 at_begline_loc_p (pattern, p, syntax) |
|
|
2594 const char *pattern, *p; |
|
|
2595 reg_syntax_t syntax; |
|
|
2596 { |
|
|
2597 const char *prev = p - 2; |
|
|
2598 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
|
|
2599 |
|
|
2600 return |
|
|
2601 /* After a subexpression? */ |
|
|
2602 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) |
|
|
2603 /* After an alternative? */ |
|
|
2604 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); |
|
|
2605 } |
|
|
2606 |
|
|
2607 |
|
|
2608 /* The dual of at_begline_loc_p. This one is for $. We assume there is |
|
|
2609 at least one character after the $, i.e., `P < PEND'. */ |
|
|
2610 |
|
|
2611 static boolean |
|
|
2612 at_endline_loc_p (p, pend, syntax) |
|
|
2613 const char *p, *pend; |
|
|
2614 int syntax; |
|
|
2615 { |
|
|
2616 const char *next = p; |
|
|
2617 boolean next_backslash = *next == '\\'; |
|
|
2618 const char *next_next = p + 1 < pend ? p + 1 : NULL; |
|
|
2619 |
|
|
2620 return |
|
|
2621 /* Before a subexpression? */ |
|
|
2622 (syntax & RE_NO_BK_PARENS ? *next == ')' |
|
|
2623 : next_backslash && next_next && *next_next == ')') |
|
|
2624 /* Before an alternative? */ |
|
|
2625 || (syntax & RE_NO_BK_VBAR ? *next == '|' |
|
|
2626 : next_backslash && next_next && *next_next == '|'); |
|
|
2627 } |
|
|
2628 |
|
|
2629 |
|
|
2630 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
|
|
2631 false if it's not. */ |
|
|
2632 |
|
|
2633 static boolean |
|
|
2634 group_in_compile_stack (compile_stack, regnum) |
|
|
2635 compile_stack_type compile_stack; |
|
|
2636 regnum_t regnum; |
|
|
2637 { |
|
|
2638 int this_element; |
|
|
2639 |
|
|
2640 for (this_element = compile_stack.avail - 1; |
|
|
2641 this_element >= 0; |
|
|
2642 this_element--) |
|
|
2643 if (compile_stack.stack[this_element].regnum == regnum) |
|
|
2644 return true; |
|
|
2645 |
|
|
2646 return false; |
|
|
2647 } |
|
|
2648 |
|
|
2649 |
|
|
2650 /* Read the ending character of a range (in a bracket expression) from the |
|
|
2651 uncompiled pattern *P_PTR (which ends at PEND). We assume the |
|
|
2652 starting character is in `P[-2]'. (`P[-1]' is the character `-'.) |
|
|
2653 Then we set the translation of all bits between the starting and |
|
|
2654 ending characters (inclusive) in the compiled pattern B. |
|
|
2655 |
|
|
2656 Return an error code. |
|
|
2657 |
|
|
2658 We use these short variable names so we can use the same macros as |
|
|
2659 `regex_compile' itself. */ |
|
|
2660 |
|
|
2661 static reg_errcode_t |
|
|
2662 compile_range (p_ptr, pend, translate, syntax, b) |
|
|
2663 const char **p_ptr, *pend; |
|
|
2664 char *translate; |
|
|
2665 reg_syntax_t syntax; |
|
|
2666 unsigned char *b; |
|
|
2667 { |
|
|
2668 unsigned this_char; |
|
|
2669 |
|
|
2670 const char *p = *p_ptr; |
|
31
|
2671 int range_start, range_end; |
|
2
|
2672 |
|
|
2673 if (p == pend) |
|
|
2674 return REG_ERANGE; |
|
|
2675 |
|
31
|
2676 /* Even though the pattern is a signed `char *', we need to fetch |
|
|
2677 with unsigned char *'s; if the high bit of the pattern character |
|
|
2678 is set, the range endpoints will be negative if we fetch using a |
|
|
2679 signed char *. |
|
|
2680 |
|
|
2681 We also want to fetch the endpoints without translating them; the |
|
|
2682 appropriate translation is done in the bit-setting loop below. */ |
|
|
2683 range_start = ((unsigned char *) p)[-2]; |
|
|
2684 range_end = ((unsigned char *) p)[0]; |
|
2
|
2685 |
|
|
2686 /* Have to increment the pointer into the pattern string, so the |
|
|
2687 caller isn't still at the ending character. */ |
|
|
2688 (*p_ptr)++; |
|
|
2689 |
|
|
2690 /* If the start is after the end, the range is empty. */ |
|
|
2691 if (range_start > range_end) |
|
|
2692 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; |
|
|
2693 |
|
|
2694 /* Here we see why `this_char' has to be larger than an `unsigned |
|
|
2695 char' -- the range is inclusive, so if `range_end' == 0xff |
|
|
2696 (assuming 8-bit characters), we would otherwise go into an infinite |
|
|
2697 loop, since all characters <= 0xff. */ |
|
|
2698 for (this_char = range_start; this_char <= range_end; this_char++) |
|
|
2699 { |
|
|
2700 SET_LIST_BIT (TRANSLATE (this_char)); |
|
|
2701 } |
|
|
2702 |
|
|
2703 return REG_NOERROR; |
|
|
2704 } |
|
|
2705 |
|
|
2706 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in |
|
|
2707 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible |
|
|
2708 characters can start a string that matches the pattern. This fastmap |
|
|
2709 is used by re_search to skip quickly over impossible starting points. |
|
|
2710 |
|
|
2711 The caller must supply the address of a (1 << BYTEWIDTH)-byte data |
|
|
2712 area as BUFP->fastmap. |
|
|
2713 |
|
|
2714 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in |
|
|
2715 the pattern buffer. |
|
|
2716 |
|
|
2717 Returns 0 if we succeed, -2 if an internal error. */ |
|
|
2718 |
|
|
2719 int |
|
|
2720 re_compile_fastmap (bufp) |
|
|
2721 struct re_pattern_buffer *bufp; |
|
|
2722 { |
|
|
2723 int j, k; |
|
89
|
2724 #ifdef MATCH_MAY_ALLOCATE |
|
2
|
2725 fail_stack_type fail_stack; |
|
87
|
2726 #endif |
|
2
|
2727 #ifndef REGEX_MALLOC |
|
|
2728 char *destination; |
|
|
2729 #endif |
|
|
2730 /* We don't push any register information onto the failure stack. */ |
|
|
2731 unsigned num_regs = 0; |
|
|
2732 |
|
|
2733 register char *fastmap = bufp->fastmap; |
|
|
2734 unsigned char *pattern = bufp->buffer; |
|
|
2735 unsigned long size = bufp->used; |
|
|
2736 const unsigned char *p = pattern; |
|
|
2737 register unsigned char *pend = pattern + size; |
|
|
2738 |
|
|
2739 /* Assume that each path through the pattern can be null until |
|
|
2740 proven otherwise. We set this false at the bottom of switch |
|
|
2741 statement, to which we get only if a particular path doesn't |
|
|
2742 match the empty string. */ |
|
|
2743 boolean path_can_be_null = true; |
|
|
2744 |
|
|
2745 /* We aren't doing a `succeed_n' to begin with. */ |
|
|
2746 boolean succeed_n_p = false; |
|
|
2747 |
|
|
2748 assert (fastmap != NULL && p != NULL); |
|
|
2749 |
|
|
2750 INIT_FAIL_STACK (); |
|
|
2751 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ |
|
|
2752 bufp->fastmap_accurate = 1; /* It will be when we're done. */ |
|
|
2753 bufp->can_be_null = 0; |
|
|
2754 |
|
|
2755 while (p != pend || !FAIL_STACK_EMPTY ()) |
|
|
2756 { |
|
|
2757 if (p == pend) |
|
|
2758 { |
|
|
2759 bufp->can_be_null |= path_can_be_null; |
|
|
2760 |
|
|
2761 /* Reset for next path. */ |
|
|
2762 path_can_be_null = true; |
|
|
2763 |
|
|
2764 p = fail_stack.stack[--fail_stack.avail]; |
|
|
2765 } |
|
|
2766 |
|
|
2767 /* We should never be about to go beyond the end of the pattern. */ |
|
|
2768 assert (p < pend); |
|
|
2769 |
|
|
2770 #ifdef SWITCH_ENUM_BUG |
|
|
2771 switch ((int) ((re_opcode_t) *p++)) |
|
|
2772 #else |
|
|
2773 switch ((re_opcode_t) *p++) |
|
|
2774 #endif |
|
|
2775 { |
|
|
2776 |
|
|
2777 /* I guess the idea here is to simply not bother with a fastmap |
|
|
2778 if a backreference is used, since it's too hard to figure out |
|
|
2779 the fastmap for the corresponding group. Setting |
|
|
2780 `can_be_null' stops `re_search_2' from using the fastmap, so |
|
|
2781 that is all we do. */ |
|
|
2782 case duplicate: |
|
|
2783 bufp->can_be_null = 1; |
|
|
2784 return 0; |
|
|
2785 |
|
|
2786 |
|
|
2787 /* Following are the cases which match a character. These end |
|
|
2788 with `break'. */ |
|
|
2789 |
|
|
2790 case exactn: |
|
|
2791 fastmap[p[1]] = 1; |
|
|
2792 break; |
|
|
2793 |
|
|
2794 |
|
|
2795 case charset: |
|
|
2796 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) |
|
|
2797 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) |
|
|
2798 fastmap[j] = 1; |
|
|
2799 break; |
|
|
2800 |
|
|
2801 |
|
|
2802 case charset_not: |
|
|
2803 /* Chars beyond end of map must be allowed. */ |
|
|
2804 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) |
|
|
2805 fastmap[j] = 1; |
|
|
2806 |
|
|
2807 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) |
|
|
2808 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) |
|
|
2809 fastmap[j] = 1; |
|
|
2810 break; |
|
|
2811 |
|
|
2812 |
|
|
2813 case wordchar: |
|
|
2814 for (j = 0; j < (1 << BYTEWIDTH); j++) |
|
|
2815 if (SYNTAX (j) == Sword) |
|
|
2816 fastmap[j] = 1; |
|
|
2817 break; |
|
|
2818 |
|
|
2819 |
|
|
2820 case notwordchar: |
|
|
2821 for (j = 0; j < (1 << BYTEWIDTH); j++) |
|
|
2822 if (SYNTAX (j) != Sword) |
|
|
2823 fastmap[j] = 1; |
|
|
2824 break; |
|
|
2825 |
|
|
2826 |
|
|
2827 case anychar: |
|
|
2828 /* `.' matches anything ... */ |
|
|
2829 for (j = 0; j < (1 << BYTEWIDTH); j++) |
|
|
2830 fastmap[j] = 1; |
|
|
2831 |
|
|
2832 /* ... except perhaps newline. */ |
|
|
2833 if (!(bufp->syntax & RE_DOT_NEWLINE)) |
|
|
2834 fastmap['\n'] = 0; |
|
|
2835 |
|
|
2836 /* Return if we have already set `can_be_null'; if we have, |
|
|
2837 then the fastmap is irrelevant. Something's wrong here. */ |
|
|
2838 else if (bufp->can_be_null) |
|
|
2839 return 0; |
|
|
2840 |
|
|
2841 /* Otherwise, have to check alternative paths. */ |
|
|
2842 break; |
|
|
2843 |
|
|
2844 |
|
|
2845 #ifdef emacs |
|
|
2846 case syntaxspec: |
|
|
2847 k = *p++; |
|
|
2848 for (j = 0; j < (1 << BYTEWIDTH); j++) |
|
|
2849 if (SYNTAX (j) == (enum syntaxcode) k) |
|
|
2850 fastmap[j] = 1; |
|
|
2851 break; |
|
|
2852 |
|
|
2853 |
|
|
2854 case notsyntaxspec: |
|
|
2855 k = *p++; |
|
|
2856 for (j = 0; j < (1 << BYTEWIDTH); j++) |
|
|
2857 if (SYNTAX (j) != (enum syntaxcode) k) |
|
|
2858 fastmap[j] = 1; |
|
|
2859 break; |
|
|
2860 |
|
|
2861 |
|
|
2862 /* All cases after this match the empty string. These end with |
|
|
2863 `continue'. */ |
|
|
2864 |
|
|
2865 |
|
|
2866 case before_dot: |
|
|
2867 case at_dot: |
|
|
2868 case after_dot: |
|
|
2869 continue; |
|
|
2870 #endif /* not emacs */ |
|
|
2871 |
|
|
2872 |
|
|
2873 case no_op: |
|
|
2874 case begline: |
|
|
2875 case endline: |
|
|
2876 case begbuf: |
|
|
2877 case endbuf: |
|
|
2878 case wordbound: |
|
|
2879 case notwordbound: |
|
|
2880 case wordbeg: |
|
|
2881 case wordend: |
|
|
2882 case push_dummy_failure: |
|
|
2883 continue; |
|
|
2884 |
|
|
2885 |
|
|
2886 case jump_n: |
|
|
2887 case pop_failure_jump: |
|
|
2888 case maybe_pop_jump: |
|
|
2889 case jump: |
|
|
2890 case jump_past_alt: |
|
|
2891 case dummy_failure_jump: |
|
|
2892 EXTRACT_NUMBER_AND_INCR (j, p); |
|
|
2893 p += j; |
|
|
2894 if (j > 0) |
|
|
2895 continue; |
|
|
2896 |
|
|
2897 /* Jump backward implies we just went through the body of a |
|
|
2898 loop and matched nothing. Opcode jumped to should be |
|
|
2899 `on_failure_jump' or `succeed_n'. Just treat it like an |
|
|
2900 ordinary jump. For a * loop, it has pushed its failure |
|
|
2901 point already; if so, discard that as redundant. */ |
|
|
2902 if ((re_opcode_t) *p != on_failure_jump |
|
|
2903 && (re_opcode_t) *p != succeed_n) |
|
|
2904 continue; |
|
|
2905 |
|
|
2906 p++; |
|
|
2907 EXTRACT_NUMBER_AND_INCR (j, p); |
|
|
2908 p += j; |
|
|
2909 |
|
|
2910 /* If what's on the stack is where we are now, pop it. */ |
|
|
2911 if (!FAIL_STACK_EMPTY () |
|
|
2912 && fail_stack.stack[fail_stack.avail - 1] == p) |
|
|
2913 fail_stack.avail--; |
|
|
2914 |
|
|
2915 continue; |
|
|
2916 |
|
|
2917 |
|
|
2918 case on_failure_jump: |
|
|
2919 case on_failure_keep_string_jump: |
|
|
2920 handle_on_failure_jump: |
|
|
2921 EXTRACT_NUMBER_AND_INCR (j, p); |
|
|
2922 |
|
|
2923 /* For some patterns, e.g., `(a?)?', `p+j' here points to the |
|
|
2924 end of the pattern. We don't want to push such a point, |
|
|
2925 since when we restore it above, entering the switch will |
|
|
2926 increment `p' past the end of the pattern. We don't need |
|
|
2927 to push such a point since we obviously won't find any more |
|
|
2928 fastmap entries beyond `pend'. Such a pattern can match |
|
|
2929 the null string, though. */ |
|
|
2930 if (p + j < pend) |
|
|
2931 { |
|
|
2932 if (!PUSH_PATTERN_OP (p + j, fail_stack)) |
|
|
2933 return -2; |
|
|
2934 } |
|
|
2935 else |
|
|
2936 bufp->can_be_null = 1; |
|
|
2937 |
|
|
2938 if (succeed_n_p) |
|
|
2939 { |
|
|
2940 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ |
|
|
2941 succeed_n_p = false; |
|
|
2942 } |
|
|
2943 |
|
|
2944 continue; |
|
|
2945 |
|
|
2946 |
|
|
2947 case succeed_n: |
|
|
2948 /* Get to the number of times to succeed. */ |
|
|
2949 p += 2; |
|
|
2950 |
|
|
2951 /* Increment p past the n for when k != 0. */ |
|
|
2952 EXTRACT_NUMBER_AND_INCR (k, p); |
|
|
2953 if (k == 0) |
|
|
2954 { |
|
|
2955 p -= 4; |
|
|
2956 succeed_n_p = true; /* Spaghetti code alert. */ |
|
|
2957 goto handle_on_failure_jump; |
|
|
2958 } |
|
|
2959 continue; |
|
|
2960 |
|
|
2961 |
|
|
2962 case set_number_at: |
|
|
2963 p += 4; |
|
|
2964 continue; |
|
|
2965 |
|
|
2966 |
|
|
2967 case start_memory: |
|
|
2968 case stop_memory: |
|
|
2969 p += 2; |
|
|
2970 continue; |
|
|
2971 |
|
|
2972 |
|
|
2973 default: |
|
|
2974 abort (); /* We have listed all the cases. */ |
|
|
2975 } /* switch *p++ */ |
|
|
2976 |
|
|
2977 /* Getting here means we have found the possible starting |
|
|
2978 characters for one path of the pattern -- and that the empty |
|
|
2979 string does not match. We need not follow this path further. |
|
|
2980 Instead, look at the next alternative (remembered on the |
|
|
2981 stack), or quit if no more. The test at the top of the loop |
|
|
2982 does these things. */ |
|
|
2983 path_can_be_null = false; |
|
|
2984 p = pend; |
|
|
2985 } /* while p */ |
|
|
2986 |
|
|
2987 /* Set `can_be_null' for the last path (also the first path, if the |
|
|
2988 pattern is empty). */ |
|
|
2989 bufp->can_be_null |= path_can_be_null; |
|
|
2990 return 0; |
|
|
2991 } /* re_compile_fastmap */ |
|
|
2992 |
|
|
2993 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and |
|
|
2994 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use |
|
|
2995 this memory for recording register information. STARTS and ENDS |
|
|
2996 must be allocated using the malloc library routine, and must each |
|
|
2997 be at least NUM_REGS * sizeof (regoff_t) bytes long. |
|
|
2998 |
|
|
2999 If NUM_REGS == 0, then subsequent matches should allocate their own |
|
|
3000 register data. |
|
|
3001 |
|
|
3002 Unless this function is called, the first search or match using |
|
|
3003 PATTERN_BUFFER will allocate its own register data, without |
|
|
3004 freeing the old data. */ |
|
|
3005 |
|
|
3006 void |
|
|
3007 re_set_registers (bufp, regs, num_regs, starts, ends) |
|
|
3008 struct re_pattern_buffer *bufp; |
|
|
3009 struct re_registers *regs; |
|
|
3010 unsigned num_regs; |
|
|
3011 regoff_t *starts, *ends; |
|
|
3012 { |
|
|
3013 if (num_regs) |
|
|
3014 { |
|
|
3015 bufp->regs_allocated = REGS_REALLOCATE; |
|
|
3016 regs->num_regs = num_regs; |
|
|
3017 regs->start = starts; |
|
|
3018 regs->end = ends; |
|
|
3019 } |
|
|
3020 else |
|
|
3021 { |
|
|
3022 bufp->regs_allocated = REGS_UNALLOCATED; |
|
|
3023 regs->num_regs = 0; |
|
|
3024 regs->start = regs->end = (regoff_t) 0; |
|
|
3025 } |
|
|
3026 } |
|
|
3027 |
|
|
3028 /* Searching routines. */ |
|
|
3029 |
|
|
3030 /* Like re_search_2, below, but only one string is specified, and |
|
|
3031 doesn't let you say where to stop matching. */ |
|
|
3032 |
|
|
3033 int |
|
|
3034 re_search (bufp, string, size, startpos, range, regs) |
|
|
3035 struct re_pattern_buffer *bufp; |
|
|
3036 const char *string; |
|
|
3037 int size, startpos, range; |
|
|
3038 struct re_registers *regs; |
|
|
3039 { |
|
|
3040 return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
|
|
3041 regs, size); |
|
|
3042 } |
|
|
3043 |
|
|
3044 |
|
|
3045 /* Using the compiled pattern in BUFP->buffer, first tries to match the |
|
|
3046 virtual concatenation of STRING1 and STRING2, starting first at index |
|
|
3047 STARTPOS, then at STARTPOS + 1, and so on. |
|
|
3048 |
|
|
3049 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
|
|
3050 |
|
|
3051 RANGE is how far to scan while trying to match. RANGE = 0 means try |
|
|
3052 only at STARTPOS; in general, the last start tried is STARTPOS + |
|
|
3053 RANGE. |
|
|
3054 |
|
|
3055 In REGS, return the indices of the virtual concatenation of STRING1 |
|
|
3056 and STRING2 that matched the entire BUFP->buffer and its contained |
|
|
3057 subexpressions. |
|
|
3058 |
|
|
3059 Do not consider matching one past the index STOP in the virtual |
|
|
3060 concatenation of STRING1 and STRING2. |
|
|
3061 |
|
|
3062 We return either the position in the strings at which the match was |
|
|
3063 found, -1 if no match, or -2 if error (such as failure |
|
|
3064 stack overflow). */ |
|
|
3065 |
|
|
3066 int |
|
|
3067 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) |
|
|
3068 struct re_pattern_buffer *bufp; |
|
|
3069 const char *string1, *string2; |
|
|
3070 int size1, size2; |
|
|
3071 int startpos; |
|
|
3072 int range; |
|
|
3073 struct re_registers *regs; |
|
|
3074 int stop; |
|
|
3075 { |
|
|
3076 int val; |
|
|
3077 register char *fastmap = bufp->fastmap; |
|
|
3078 register char *translate = bufp->translate; |
|
|
3079 int total_size = size1 + size2; |
|
|
3080 int endpos = startpos + range; |
|
|
3081 |
|
|
3082 /* Check for out-of-range STARTPOS. */ |
|
|
3083 if (startpos < 0 || startpos > total_size) |
|
|
3084 return -1; |
|
|
3085 |
|
|
3086 /* Fix up RANGE if it might eventually take us outside |
|
|
3087 the virtual concatenation of STRING1 and STRING2. */ |
|
|
3088 if (endpos < -1) |
|
|
3089 range = -1 - startpos; |
|
|
3090 else if (endpos > total_size) |
|
|
3091 range = total_size - startpos; |
|
|
3092 |
|
|
3093 /* If the search isn't to be a backwards one, don't waste time in a |
|
23
|
3094 search for a pattern that must be anchored. */ |
|
|
3095 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) |
|
2
|
3096 { |
|
|
3097 if (startpos > 0) |
|
|
3098 return -1; |
|
|
3099 else |
|
|
3100 range = 1; |
|
|
3101 } |
|
|
3102 |
|
23
|
3103 /* Update the fastmap now if not correct already. */ |
|
|
3104 if (fastmap && !bufp->fastmap_accurate) |
|
|
3105 if (re_compile_fastmap (bufp) == -2) |
|
|
3106 return -2; |
|
|
3107 |
|
|
3108 /* Loop through the string, looking for a place to start matching. */ |
|
2
|
3109 for (;;) |
|
|
3110 { |
|
|
3111 /* If a fastmap is supplied, skip quickly over characters that |
|
|
3112 cannot be the start of a match. If the pattern can match the |
|
|
3113 null string, however, we don't need to skip characters; we want |
|
|
3114 the first null string. */ |
|
|
3115 if (fastmap && startpos < total_size && !bufp->can_be_null) |
|
|
3116 { |
|
|
3117 if (range > 0) /* Searching forwards. */ |
|
|
3118 { |
|
|
3119 register const char *d; |
|
|
3120 register int lim = 0; |
|
|
3121 int irange = range; |
|
|
3122 |
|
|
3123 if (startpos < size1 && startpos + range >= size1) |
|
|
3124 lim = range - (size1 - startpos); |
|
|
3125 |
|
|
3126 d = (startpos >= size1 ? string2 - size1 : string1) + startpos; |
|
|
3127 |
|
|
3128 /* Written out as an if-else to avoid testing `translate' |
|
|
3129 inside the loop. */ |
|
|
3130 if (translate) |
|
|
3131 while (range > lim |
|
38
|
3132 && !fastmap[(unsigned char) |
|
|
3133 translate[(unsigned char) *d++]]) |
|
2
|
3134 range--; |
|
|
3135 else |
|
|
3136 while (range > lim && !fastmap[(unsigned char) *d++]) |
|
|
3137 range--; |
|
|
3138 |
|
|
3139 startpos += irange - range; |
|
|
3140 } |
|
|
3141 else /* Searching backwards. */ |
|
|
3142 { |
|
|
3143 register char c = (size1 == 0 || startpos >= size1 |
|
|
3144 ? string2[startpos - size1] |
|
|
3145 : string1[startpos]); |
|
|
3146 |
|
23
|
3147 if (!fastmap[(unsigned char) TRANSLATE (c)]) |
|
2
|
3148 goto advance; |
|
|
3149 } |
|
|
3150 } |
|
|
3151 |
|
|
3152 /* If can't match the null string, and that's all we have left, fail. */ |
|
|
3153 if (range >= 0 && startpos == total_size && fastmap |
|
|
3154 && !bufp->can_be_null) |
|
|
3155 return -1; |
|
|
3156 |
|
|
3157 val = re_match_2 (bufp, string1, size1, string2, size2, |
|
|
3158 startpos, regs, stop); |
|
|
3159 if (val >= 0) |
|
|
3160 return startpos; |
|
|
3161 |
|
|
3162 if (val == -2) |
|
|
3163 return -2; |
|
|
3164 |
|
|
3165 advance: |
|
|
3166 if (!range) |
|
|
3167 break; |
|
|
3168 else if (range > 0) |
|
|
3169 { |
|
|
3170 range--; |
|
|
3171 startpos++; |
|
|
3172 } |
|
|
3173 else |
|
|
3174 { |
|
|
3175 range++; |
|
|
3176 startpos--; |
|
|
3177 } |
|
|
3178 } |
|
|
3179 return -1; |
|
|
3180 } /* re_search_2 */ |
|
|
3181 |
|
|
3182 /* Declarations and macros for re_match_2. */ |
|
|
3183 |
|
|
3184 static int bcmp_translate (); |
|
|
3185 static boolean alt_match_null_string_p (), |
|
|
3186 common_op_match_null_string_p (), |
|
|
3187 group_match_null_string_p (); |
|
|
3188 |
|
|
3189 /* This converts PTR, a pointer into one of the search strings `string1' |
|
|
3190 and `string2' into an offset from the beginning of that string. */ |
|
|
3191 #define POINTER_TO_OFFSET(ptr) \ |
|
|
3192 (FIRST_STRING_P (ptr) ? (ptr) - string1 : (ptr) - string2 + size1) |
|
|
3193 |
|
|
3194 /* Macros for dealing with the split strings in re_match_2. */ |
|
|
3195 |
|
|
3196 #define MATCHING_IN_FIRST_STRING (dend == end_match_1) |
|
|
3197 |
|
|
3198 /* Call before fetching a character with *d. This switches over to |
|
|
3199 string2 if necessary. */ |
|
|
3200 #define PREFETCH() \ |
|
|
3201 while (d == dend) \ |
|
|
3202 { \ |
|
|
3203 /* End of string2 => fail. */ \ |
|
|
3204 if (dend == end_match_2) \ |
|
|
3205 goto fail; \ |
|
|
3206 /* End of string1 => advance to string2. */ \ |
|
|
3207 d = string2; \ |
|
|
3208 dend = end_match_2; \ |
|
|
3209 } |
|
|
3210 |
|
|
3211 |
|
|
3212 /* Test if at very beginning or at very end of the virtual concatenation |
|
|
3213 of `string1' and `string2'. If only one string, it's `string2'. */ |
|
23
|
3214 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
|
|
3215 #define AT_STRINGS_END(d) ((d) == end2) |
|
2
|
3216 |
|
|
3217 |
|
|
3218 /* Test if D points to a character which is word-constituent. We have |
|
|
3219 two special cases to check for: if past the end of string1, look at |
|
|
3220 the first character in string2; and if before the beginning of |
|
23
|
3221 string2, look at the last character in string1. */ |
|
|
3222 #define WORDCHAR_P(d) \ |
|
2
|
3223 (SYNTAX ((d) == end1 ? *string2 \ |
|
23
|
3224 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
|
|
3225 == Sword) |
|
2
|
3226 |
|
|
3227 /* Test if the character before D and the one at D differ with respect |
|
|
3228 to being word-constituent. */ |
|
|
3229 #define AT_WORD_BOUNDARY(d) \ |
|
23
|
3230 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
|
|
3231 || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) |
|
2
|
3232 |
|
|
3233 |
|
|
3234 /* Free everything we malloc. */ |
|
89
|
3235 #ifdef MATCH_MAY_ALLOCATE |
|
2
|
3236 #ifdef REGEX_MALLOC |
|
|
3237 #define FREE_VAR(var) if (var) free (var); var = NULL |
|
|
3238 #define FREE_VARIABLES() \ |
|
|
3239 do { \ |
|
|
3240 FREE_VAR (fail_stack.stack); \ |
|
|
3241 FREE_VAR (regstart); \ |
|
|
3242 FREE_VAR (regend); \ |
|
|
3243 FREE_VAR (old_regstart); \ |
|
|
3244 FREE_VAR (old_regend); \ |
|
|
3245 FREE_VAR (best_regstart); \ |
|
|
3246 FREE_VAR (best_regend); \ |
|
|
3247 FREE_VAR (reg_info); \ |
|
|
3248 FREE_VAR (reg_dummy); \ |
|
|
3249 FREE_VAR (reg_info_dummy); \ |
|
|
3250 } while (0) |
|
|
3251 #else /* not REGEX_MALLOC */ |
|
|
3252 /* Some MIPS systems (at least) want this to free alloca'd storage. */ |
|
|
3253 #define FREE_VARIABLES() alloca (0) |
|
|
3254 #endif /* not REGEX_MALLOC */ |
|
89
|
3255 #else |
|
|
3256 #define FREE_VARIABLES() /* Do nothing! */ |
|
|
3257 #endif /* not MATCH_MAY_ALLOCATE */ |
|
2
|
3258 |
|
|
3259 /* These values must meet several constraints. They must not be valid |
|
|
3260 register values; since we have a limit of 255 registers (because |
|
|
3261 we use only one byte in the pattern for the register number), we can |
|
|
3262 use numbers larger than 255. They must differ by 1, because of |
|
|
3263 NUM_FAILURE_ITEMS above. And the value for the lowest register must |
|
|
3264 be larger than the value for the highest register, so we do not try |
|
|
3265 to actually save any registers when none are active. */ |
|
|
3266 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) |
|
|
3267 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) |
|
|
3268 |
|
|
3269 /* Matching routines. */ |
|
|
3270 |
|
|
3271 #ifndef emacs /* Emacs never uses this. */ |
|
|
3272 /* re_match is like re_match_2 except it takes only a single string. */ |
|
|
3273 |
|
|
3274 int |
|
|
3275 re_match (bufp, string, size, pos, regs) |
|
|
3276 struct re_pattern_buffer *bufp; |
|
|
3277 const char *string; |
|
|
3278 int size, pos; |
|
|
3279 struct re_registers *regs; |
|
|
3280 { |
|
|
3281 return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size); |
|
|
3282 } |
|
|
3283 #endif /* not emacs */ |
|
|
3284 |
|
|
3285 |
|
|
3286 /* re_match_2 matches the compiled pattern in BUFP against the |
|
|
3287 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 |
|
|
3288 and SIZE2, respectively). We start matching at POS, and stop |
|
|
3289 matching at STOP. |
|
|
3290 |
|
|
3291 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
|
|
3292 store offsets for the substring each group matched in REGS. See the |
|
|
3293 documentation for exactly how many groups we fill. |
|
|
3294 |
|
|
3295 We return -1 if no match, -2 if an internal error (such as the |
|
|
3296 failure stack overflowing). Otherwise, we return the length of the |
|
|
3297 matched substring. */ |
|
|
3298 |
|
|
3299 int |
|
|
3300 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) |
|
|
3301 struct re_pattern_buffer *bufp; |
|
|
3302 const char *string1, *string2; |
|
|
3303 int size1, size2; |
|
|
3304 int pos; |
|
|
3305 struct re_registers *regs; |
|
|
3306 int stop; |
|
|
3307 { |
|
|
3308 /* General temporaries. */ |
|
|
3309 int mcnt; |
|
|
3310 unsigned char *p1; |
|
|
3311 |
|
|
3312 /* Just past the end of the corresponding string. */ |
|
|
3313 const char *end1, *end2; |
|
|
3314 |
|
|
3315 /* Pointers into string1 and string2, just past the last characters in |
|
|
3316 each to consider matching. */ |
|
|
3317 const char *end_match_1, *end_match_2; |
|
|
3318 |
|
|
3319 /* Where we are in the data, and the end of the current string. */ |
|
|
3320 const char *d, *dend; |
|
|
3321 |
|
|
3322 /* Where we are in the pattern, and the end of the pattern. */ |
|
|
3323 unsigned char *p = bufp->buffer; |
|
|
3324 register unsigned char *pend = p + bufp->used; |
|
|
3325 |
|
|
3326 /* We use this to map every character in the string. */ |
|
|
3327 char *translate = bufp->translate; |
|
|
3328 |
|
|
3329 /* Failure point stack. Each place that can handle a failure further |
|
|
3330 down the line pushes a failure point on this stack. It consists of |
|
|
3331 restart, regend, and reg_info for all registers corresponding to |
|
|
3332 the subexpressions we're currently inside, plus the number of such |
|
|
3333 registers, and, finally, two char *'s. The first char * is where |
|
|
3334 to resume scanning the pattern; the second one is where to resume |
|
|
3335 scanning the strings. If the latter is zero, the failure point is |
|
|
3336 a ``dummy''; if a failure happens and the failure point is a dummy, |
|
|
3337 it gets discarded and the next next one is tried. */ |
|
89
|
3338 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
|
2
|
3339 fail_stack_type fail_stack; |
|
87
|
3340 #endif |
|
2
|
3341 #ifdef DEBUG |
|
|
3342 static unsigned failure_id = 0; |
|
23
|
3343 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
|
2
|
3344 #endif |
|
|
3345 |
|
|
3346 /* We fill all the registers internally, independent of what we |
|
|
3347 return, for use in backreferences. The number here includes |
|
|
3348 an element for register zero. */ |
|
|
3349 unsigned num_regs = bufp->re_nsub + 1; |
|
|
3350 |
|
|
3351 /* The currently active registers. */ |
|
|
3352 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
|
|
3353 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
|
|
3354 |
|
|
3355 /* Information on the contents of registers. These are pointers into |
|
|
3356 the input strings; they record just what was matched (on this |
|
|
3357 attempt) by a subexpression part of the pattern, that is, the |
|
|
3358 regnum-th regstart pointer points to where in the pattern we began |
|
|
3359 matching and the regnum-th regend points to right after where we |
|
|
3360 stopped matching the regnum-th subexpression. (The zeroth register |
|
|
3361 keeps track of what the whole pattern matches.) */ |
|
89
|
3362 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
|
2
|
3363 const char **regstart, **regend; |
|
87
|
3364 #endif |
|
2
|
3365 |
|
|
3366 /* If a group that's operated upon by a repetition operator fails to |
|
|
3367 match anything, then the register for its start will need to be |
|
|
3368 restored because it will have been set to wherever in the string we |
|
|
3369 are when we last see its open-group operator. Similarly for a |
|
|
3370 register's end. */ |
|
89
|
3371 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
|
2
|
3372 const char **old_regstart, **old_regend; |
|
87
|
3373 #endif |
|
2
|
3374 |
|
|
3375 /* The is_active field of reg_info helps us keep track of which (possibly |
|
|
3376 nested) subexpressions we are currently in. The matched_something |
|
|
3377 field of reg_info[reg_num] helps us tell whether or not we have |
|
|
3378 matched any of the pattern so far this time through the reg_num-th |
|
|
3379 subexpression. These two fields get reset each time through any |
|
|
3380 loop their register is in. */ |
|
89
|
3381 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
|
2
|
3382 register_info_type *reg_info; |
|
87
|
3383 #endif |
|
2
|
3384 |
|
|
3385 /* The following record the register info as found in the above |
|
|
3386 variables when we find a match better than any we've seen before. |
|
|
3387 This happens as we backtrack through the failure points, which in |
|
|
3388 turn happens only if we have not yet matched the entire string. */ |
|
|
3389 unsigned best_regs_set = false; |
|
89
|
3390 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
|
2
|
3391 const char **best_regstart, **best_regend; |
|
87
|
3392 #endif |
|
2
|
3393 |
|
|
3394 /* Logically, this is `best_regend[0]'. But we don't want to have to |
|
|
3395 allocate space for that if we're not allocating space for anything |
|
|
3396 else (see below). Also, we never need info about register 0 for |
|
|
3397 any of the other register vectors, and it seems rather a kludge to |
|
|
3398 treat `best_regend' differently than the rest. So we keep track of |
|
|
3399 the end of the best match so far in a separate variable. We |
|
|
3400 initialize this to NULL so that when we backtrack the first time |
|
|
3401 and need to test it, it's not garbage. */ |
|
|
3402 const char *match_end = NULL; |
|
|
3403 |
|
|
3404 /* Used when we pop values we don't care about. */ |
|
89
|
3405 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
|
2
|
3406 const char **reg_dummy; |
|
|
3407 register_info_type *reg_info_dummy; |
|
87
|
3408 #endif |
|
2
|
3409 |
|
|
3410 #ifdef DEBUG |
|
|
3411 /* Counts the total number of registers pushed. */ |
|
|
3412 unsigned num_regs_pushed = 0; |
|
|
3413 #endif |
|
|
3414 |
|
|
3415 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); |
|
|
3416 |
|
|
3417 INIT_FAIL_STACK (); |
|
|
3418 |
|
89
|
3419 #ifdef MATCH_MAY_ALLOCATE |
|
2
|
3420 /* Do not bother to initialize all the register variables if there are |
|
|
3421 no groups in the pattern, as it takes a fair amount of time. If |
|
|
3422 there are groups, we include space for register 0 (the whole |
|
|
3423 pattern), even though we never use it, since it simplifies the |
|
|
3424 array indexing. We should fix this. */ |
|
|
3425 if (bufp->re_nsub) |
|
|
3426 { |
|
|
3427 regstart = REGEX_TALLOC (num_regs, const char *); |
|
|
3428 regend = REGEX_TALLOC (num_regs, const char *); |
|
|
3429 old_regstart = REGEX_TALLOC (num_regs, const char *); |
|
|
3430 old_regend = REGEX_TALLOC (num_regs, const char *); |
|
|
3431 best_regstart = REGEX_TALLOC (num_regs, const char *); |
|
|
3432 best_regend = REGEX_TALLOC (num_regs, const char *); |
|
|
3433 reg_info = REGEX_TALLOC (num_regs, register_info_type); |
|
|
3434 reg_dummy = REGEX_TALLOC (num_regs, const char *); |
|
|
3435 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); |
|
|
3436 |
|
|
3437 if (!(regstart && regend && old_regstart && old_regend && reg_info |
|
|
3438 && best_regstart && best_regend && reg_dummy && reg_info_dummy)) |
|
|
3439 { |
|
|
3440 FREE_VARIABLES (); |
|
|
3441 return -2; |
|
|
3442 } |
|
|
3443 } |
|
87
|
3444 #if defined (REGEX_MALLOC) |
|
2
|
3445 else |
|
|
3446 { |
|
|
3447 /* We must initialize all our variables to NULL, so that |
|
23
|
3448 `FREE_VARIABLES' doesn't try to free them. */ |
|
2
|
3449 regstart = regend = old_regstart = old_regend = best_regstart |
|
|
3450 = best_regend = reg_dummy = NULL; |
|
|
3451 reg_info = reg_info_dummy = (register_info_type *) NULL; |
|
|
3452 } |
|
|
3453 #endif /* REGEX_MALLOC */ |
|
89
|
3454 #endif /* MATCH_MAY_ALLOCATE */ |
|
2
|
3455 |
|
|
3456 /* The starting position is bogus. */ |
|
|
3457 if (pos < 0 || pos > size1 + size2) |
|
|
3458 { |
|
|
3459 FREE_VARIABLES (); |
|
|
3460 return -1; |
|
|
3461 } |
|
|
3462 |
|
|
3463 /* Initialize subexpression text positions to -1 to mark ones that no |
|
|
3464 start_memory/stop_memory has been seen for. Also initialize the |
|
|
3465 register information struct. */ |
|
|
3466 for (mcnt = 1; mcnt < num_regs; mcnt++) |
|
|
3467 { |
|
|
3468 regstart[mcnt] = regend[mcnt] |
|
|
3469 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; |
|
|
3470 |
|
|
3471 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; |
|
|
3472 IS_ACTIVE (reg_info[mcnt]) = 0; |
|
|
3473 MATCHED_SOMETHING (reg_info[mcnt]) = 0; |
|
|
3474 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; |
|
|
3475 } |
|
|
3476 |
|
|
3477 /* We move `string1' into `string2' if the latter's empty -- but not if |
|
|
3478 `string1' is null. */ |
|
|
3479 if (size2 == 0 && string1 != NULL) |
|
|
3480 { |
|
|
3481 string2 = string1; |
|
|
3482 size2 = size1; |
|
|
3483 string1 = 0; |
|
|
3484 size1 = 0; |
|
|
3485 } |
|
|
3486 end1 = string1 + size1; |
|
|
3487 end2 = string2 + size2; |
|
|
3488 |
|
|
3489 /* Compute where to stop matching, within the two strings. */ |
|
|
3490 if (stop <= size1) |
|
|
3491 { |
|
|
3492 end_match_1 = string1 + stop; |
|
|
3493 end_match_2 = string2; |
|
|
3494 } |
|
|
3495 else |
|
|
3496 { |
|
|
3497 end_match_1 = end1; |
|
|
3498 end_match_2 = string2 + stop - size1; |
|
|
3499 } |
|
|
3500 |
|
|
3501 /* `p' scans through the pattern as `d' scans through the data. |
|
|
3502 `dend' is the end of the input string that `d' points within. `d' |
|
|
3503 is advanced into the following input string whenever necessary, but |
|
|
3504 this happens before fetching; therefore, at the beginning of the |
|
|
3505 loop, `d' can be pointing at the end of a string, but it cannot |
|
|
3506 equal `string2'. */ |
|
|
3507 if (size1 > 0 && pos <= size1) |
|
|
3508 { |
|
|
3509 d = string1 + pos; |
|
|
3510 dend = end_match_1; |
|
|
3511 } |
|
|
3512 else |
|
|
3513 { |
|
|
3514 d = string2 + pos - size1; |
|
|
3515 dend = end_match_2; |
|
|
3516 } |
|
|
3517 |
|
|
3518 DEBUG_PRINT1 ("The compiled pattern is: "); |
|
|
3519 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); |
|
|
3520 DEBUG_PRINT1 ("The string to match is: `"); |
|
|
3521 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); |
|
|
3522 DEBUG_PRINT1 ("'\n"); |
|
|
3523 |
|
|
3524 /* This loops over pattern commands. It exits by returning from the |
|
|
3525 function if the match is complete, or it drops through if the match |
|
|
3526 fails at this starting point in the input data. */ |
|
|
3527 for (;;) |
|
|
3528 { |
|
|
3529 DEBUG_PRINT2 ("\n0x%x: ", p); |
|
|
3530 |
|
|
3531 if (p == pend) |
|
|
3532 { /* End of pattern means we might have succeeded. */ |
|
23
|
3533 DEBUG_PRINT1 ("end of pattern ... "); |
|
|
3534 |
|
|
3535 /* If we haven't matched the entire string, and we want the |
|
|
3536 longest match, try backtracking. */ |
|
2
|
3537 if (d != end_match_2) |
|
|
3538 { |
|
|
3539 DEBUG_PRINT1 ("backtracking.\n"); |
|
|
3540 |
|
|
3541 if (!FAIL_STACK_EMPTY ()) |
|
|
3542 { /* More failure points to try. */ |
|
|
3543 boolean same_str_p = (FIRST_STRING_P (match_end) |
|
|
3544 == MATCHING_IN_FIRST_STRING); |
|
|
3545 |
|
|
3546 /* If exceeds best match so far, save it. */ |
|
|
3547 if (!best_regs_set |
|
|
3548 || (same_str_p && d > match_end) |
|
|
3549 || (!same_str_p && !MATCHING_IN_FIRST_STRING)) |
|
|
3550 { |
|
|
3551 best_regs_set = true; |
|
|
3552 match_end = d; |
|
|
3553 |
|
|
3554 DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); |
|
|
3555 |
|
|
3556 for (mcnt = 1; mcnt < num_regs; mcnt++) |
|
|
3557 { |
|
|
3558 best_regstart[mcnt] = regstart[mcnt]; |
|
|
3559 best_regend[mcnt] = regend[mcnt]; |
|
|
3560 } |
|
|
3561 } |
|
|
3562 goto fail; |
|
|
3563 } |
|
|
3564 |
|
|
3565 /* If no failure points, don't restore garbage. */ |
|
|
3566 else if (best_regs_set) |
|
|
3567 { |
|
|
3568 restore_best_regs: |
|
|
3569 /* Restore best match. It may happen that `dend == |
|
|
3570 end_match_1' while the restored d is in string2. |
|
|
3571 For example, the pattern `x.*y.*z' against the |
|
|
3572 strings `x-' and `y-z-', if the two strings are |
|
|
3573 not consecutive in memory. */ |
|
23
|
3574 DEBUG_PRINT1 ("Restoring best registers.\n"); |
|
|
3575 |
|
2
|
3576 d = match_end; |
|
|
3577 dend = ((d >= string1 && d <= end1) |
|
|
3578 ? end_match_1 : end_match_2); |
|
|
3579 |
|
|
3580 for (mcnt = 1; mcnt < num_regs; mcnt++) |
|
|
3581 { |
|
|
3582 regstart[mcnt] = best_regstart[mcnt]; |
|
|
3583 regend[mcnt] = best_regend[mcnt]; |
|
|
3584 } |
|
|
3585 } |
|
|
3586 } /* d != end_match_2 */ |
|
|
3587 |
|
23
|
3588 DEBUG_PRINT1 ("Accepting match.\n"); |
|
2
|
3589 |
|
|
3590 /* If caller wants register contents data back, do it. */ |
|
|
3591 if (regs && !bufp->no_sub) |
|
|
3592 { |
|
|
3593 /* Have the register data arrays been allocated? */ |
|
|
3594 if (bufp->regs_allocated == REGS_UNALLOCATED) |
|
|
3595 { /* No. So allocate them with malloc. We need one |
|
|
3596 extra element beyond `num_regs' for the `-1' marker |
|
|
3597 GNU code uses. */ |
|
|
3598 regs->num_regs = MAX (RE_NREGS, num_regs + 1); |
|
|
3599 regs->start = TALLOC (regs->num_regs, regoff_t); |
|
|
3600 regs->end = TALLOC (regs->num_regs, regoff_t); |
|
|
3601 if (regs->start == NULL || regs->end == NULL) |
|
|
3602 return -2; |
|
|
3603 bufp->regs_allocated = REGS_REALLOCATE; |
|
|
3604 } |
|
|
3605 else if (bufp->regs_allocated == REGS_REALLOCATE) |
|
|
3606 { /* Yes. If we need more elements than were already |
|
|
3607 allocated, reallocate them. If we need fewer, just |
|
|
3608 leave it alone. */ |
|
|
3609 if (regs->num_regs < num_regs + 1) |
|
|
3610 { |
|
|
3611 regs->num_regs = num_regs + 1; |
|
|
3612 RETALLOC (regs->start, regs->num_regs, regoff_t); |
|
|
3613 RETALLOC (regs->end, regs->num_regs, regoff_t); |
|
|
3614 if (regs->start == NULL || regs->end == NULL) |
|
|
3615 return -2; |
|
|
3616 } |
|
|
3617 } |
|
|
3618 else |
|
54
|
3619 { |
|
|
3620 /* These braces fend off a "empty body in an else-statement" |
|
|
3621 warning under GCC when assert expands to nothing. */ |
|
|
3622 assert (bufp->regs_allocated == REGS_FIXED); |
|
|
3623 } |
|
2
|
3624 |
|
|
3625 /* Convert the pointer data in `regstart' and `regend' to |
|
|
3626 indices. Register zero has to be set differently, |
|
|
3627 since we haven't kept track of any info for it. */ |
|
|
3628 if (regs->num_regs > 0) |
|
|
3629 { |
|
|
3630 regs->start[0] = pos; |
|
|
3631 regs->end[0] = (MATCHING_IN_FIRST_STRING ? d - string1 |
|
|
3632 : d - string2 + size1); |
|
|
3633 } |
|
|
3634 |
|
|
3635 /* Go through the first `min (num_regs, regs->num_regs)' |
|
|
3636 registers, since that is all we initialized. */ |
|
|
3637 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) |
|
|
3638 { |
|
|
3639 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) |
|
|
3640 regs->start[mcnt] = regs->end[mcnt] = -1; |
|
|
3641 else |
|
|
3642 { |
|
|
3643 regs->start[mcnt] = POINTER_TO_OFFSET (regstart[mcnt]); |
|
|
3644 regs->end[mcnt] = POINTER_TO_OFFSET (regend[mcnt]); |
|
|
3645 } |
|
|
3646 } |
|
|
3647 |
|
|
3648 /* If the regs structure we return has more elements than |
|
|
3649 were in the pattern, set the extra elements to -1. If |
|
|
3650 we (re)allocated the registers, this is the case, |
|
|
3651 because we always allocate enough to have at least one |
|
|
3652 -1 at the end. */ |
|
|
3653 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) |
|
|
3654 regs->start[mcnt] = regs->end[mcnt] = -1; |
|
|
3655 } /* regs && !bufp->no_sub */ |
|
|
3656 |
|
|
3657 FREE_VARIABLES (); |
|
23
|
3658 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
|
|
3659 nfailure_points_pushed, nfailure_points_popped, |
|
|
3660 nfailure_points_pushed - nfailure_points_popped); |
|
|
3661 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); |
|
2
|
3662 |
|
|
3663 mcnt = d - pos - (MATCHING_IN_FIRST_STRING |
|
|
3664 ? string1 |
|
|
3665 : string2 - size1); |
|
|
3666 |
|
|
3667 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
|
|
3668 |
|
|
3669 return mcnt; |
|
|
3670 } |
|
|
3671 |
|
|
3672 /* Otherwise match next pattern command. */ |
|
|
3673 #ifdef SWITCH_ENUM_BUG |
|
|
3674 switch ((int) ((re_opcode_t) *p++)) |
|
|
3675 #else |
|
|
3676 switch ((re_opcode_t) *p++) |
|
|
3677 #endif |
|
|
3678 { |
|
|
3679 /* Ignore these. Used to ignore the n of succeed_n's which |
|
|
3680 currently have n == 0. */ |
|
|
3681 case no_op: |
|
|
3682 DEBUG_PRINT1 ("EXECUTING no_op.\n"); |
|
|
3683 break; |
|
|
3684 |
|
|
3685 |
|
|
3686 /* Match the next n pattern characters exactly. The following |
|
|
3687 byte in the pattern defines n, and the n bytes after that |
|
|
3688 are the characters to match. */ |
|
|
3689 case exactn: |
|
|
3690 mcnt = *p++; |
|
|
3691 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
|
|
3692 |
|
|
3693 /* This is written out as an if-else so we don't waste time |
|
|
3694 testing `translate' inside the loop. */ |
|
|
3695 if (translate) |
|
|
3696 { |
|
|
3697 do |
|
|
3698 { |
|
|
3699 PREFETCH (); |
|
|
3700 if (translate[(unsigned char) *d++] != (char) *p++) |
|
|
3701 goto fail; |
|
|
3702 } |
|
|
3703 while (--mcnt); |
|
|
3704 } |
|
|
3705 else |
|
|
3706 { |
|
|
3707 do |
|
|
3708 { |
|
|
3709 PREFETCH (); |
|
|
3710 if (*d++ != (char) *p++) goto fail; |
|
|
3711 } |
|
|
3712 while (--mcnt); |
|
|
3713 } |
|
|
3714 SET_REGS_MATCHED (); |
|
|
3715 break; |
|
|
3716 |
|
|
3717 |
|
|
3718 /* Match any character except possibly a newline or a null. */ |
|
|
3719 case anychar: |
|
|
3720 DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
|
|
3721 |
|
|
3722 PREFETCH (); |
|
|
3723 |
|
|
3724 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') |
|
|
3725 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) |
|
|
3726 goto fail; |
|
|
3727 |
|
|
3728 SET_REGS_MATCHED (); |
|
|
3729 DEBUG_PRINT2 (" Matched `%d'.\n", *d); |
|
|
3730 d++; |
|
|
3731 break; |
|
|
3732 |
|
|
3733 |
|
|
3734 case charset: |
|
|
3735 case charset_not: |
|
|
3736 { |
|
|
3737 register unsigned char c; |
|
|
3738 boolean not = (re_opcode_t) *(p - 1) == charset_not; |
|
|
3739 |
|
|
3740 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
|
|
3741 |
|
|
3742 PREFETCH (); |
|
|
3743 c = TRANSLATE (*d); /* The character to match. */ |
|
|
3744 |
|
|
3745 /* Cast to `unsigned' instead of `unsigned char' in case the |
|
|
3746 bit list is a full 32 bytes long. */ |
|
|
3747 if (c < (unsigned) (*p * BYTEWIDTH) |
|
|
3748 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
|
|
3749 not = !not; |
|
|
3750 |
|
|
3751 p += 1 + *p; |
|
|
3752 |
|
|
3753 if (!not) goto fail; |
|
|
3754 |
|
|
3755 SET_REGS_MATCHED (); |
|
|
3756 d++; |
|
|
3757 break; |
|
|
3758 } |
|
|
3759 |
|
|
3760 |
|
|
3761 /* The beginning of a group is represented by start_memory. |
|
|
3762 The arguments are the register number in the next byte, and the |
|
|
3763 number of groups inner to this one in the next. The text |
|
|
3764 matched within the group is recorded (in the internal |
|
|
3765 registers data structure) under the register number. */ |
|
|
3766 case start_memory: |
|
|
3767 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); |
|
|
3768 |
|
|
3769 /* Find out if this group can match the empty string. */ |
|
|
3770 p1 = p; /* To send to group_match_null_string_p. */ |
|
|
3771 |
|
|
3772 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) |
|
|
3773 REG_MATCH_NULL_STRING_P (reg_info[*p]) |
|
|
3774 = group_match_null_string_p (&p1, pend, reg_info); |
|
|
3775 |
|
|
3776 /* Save the position in the string where we were the last time |
|
|
3777 we were at this open-group operator in case the group is |
|
|
3778 operated upon by a repetition operator, e.g., with `(a*)*b' |
|
|
3779 against `ab'; then we want to ignore where we are now in |
|
|
3780 the string in case this attempt to match fails. */ |
|
|
3781 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) |
|
|
3782 ? REG_UNSET (regstart[*p]) ? d : regstart[*p] |
|
|
3783 : regstart[*p]; |
|
|
3784 DEBUG_PRINT2 (" old_regstart: %d\n", |
|
|
3785 POINTER_TO_OFFSET (old_regstart[*p])); |
|
|
3786 |
|
|
3787 regstart[*p] = d; |
|
|
3788 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
|
|
3789 |
|
|
3790 IS_ACTIVE (reg_info[*p]) = 1; |
|
|
3791 MATCHED_SOMETHING (reg_info[*p]) = 0; |
|
|
3792 |
|
|
3793 /* This is the new highest active register. */ |
|
|
3794 highest_active_reg = *p; |
|
|
3795 |
|
|
3796 /* If nothing was active before, this is the new lowest active |
|
|
3797 register. */ |
|
|
3798 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) |
|
|
3799 lowest_active_reg = *p; |
|
|
3800 |
|
|
3801 /* Move past the register number and inner group count. */ |
|
|
3802 p += 2; |
|
|
3803 break; |
|
|
3804 |
|
|
3805 |
|
|
3806 /* The stop_memory opcode represents the end of a group. Its |
|
|
3807 arguments are the same as start_memory's: the register |
|
|
3808 number, and the number of inner groups. */ |
|
|
3809 case stop_memory: |
|
|
3810 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); |
|
|
3811 |
|
|
3812 /* We need to save the string position the last time we were at |
|
|
3813 this close-group operator in case the group is operated |
|
|
3814 upon by a repetition operator, e.g., with `((a*)*(b*)*)*' |
|
|
3815 against `aba'; then we want to ignore where we are now in |
|
|
3816 the string in case this attempt to match fails. */ |
|
|
3817 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) |
|
|
3818 ? REG_UNSET (regend[*p]) ? d : regend[*p] |
|
|
3819 : regend[*p]; |
|
|
3820 DEBUG_PRINT2 (" old_regend: %d\n", |
|
|
3821 POINTER_TO_OFFSET (old_regend[*p])); |
|
|
3822 |
|
|
3823 regend[*p] = d; |
|
|
3824 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
|
|
3825 |
|
|
3826 /* This register isn't active anymore. */ |
|
|
3827 IS_ACTIVE (reg_info[*p]) = 0; |
|
|
3828 |
|
|
3829 /* If this was the only register active, nothing is active |
|
|
3830 anymore. */ |
|
|
3831 if (lowest_active_reg == highest_active_reg) |
|
|
3832 { |
|
|
3833 lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
|
|
3834 highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
|
|
3835 } |
|
|
3836 else |
|
|
3837 { /* We must scan for the new highest active register, since |
|
|
3838 it isn't necessarily one less than now: consider |
|
|
3839 (a(b)c(d(e)f)g). When group 3 ends, after the f), the |
|
|
3840 new highest active register is 1. */ |
|
|
3841 unsigned char r = *p - 1; |
|
|
3842 while (r > 0 && !IS_ACTIVE (reg_info[r])) |
|
|
3843 r--; |
|
|
3844 |
|
|
3845 /* If we end up at register zero, that means that we saved |
|
|
3846 the registers as the result of an `on_failure_jump', not |
|
|
3847 a `start_memory', and we jumped to past the innermost |
|
|
3848 `stop_memory'. For example, in ((.)*) we save |
|
|
3849 registers 1 and 2 as a result of the *, but when we pop |
|
|
3850 back to the second ), we are at the stop_memory 1. |
|
|
3851 Thus, nothing is active. */ |
|
|
3852 if (r == 0) |
|
|
3853 { |
|
|
3854 lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
|
|
3855 highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
|
|
3856 } |
|
|
3857 else |
|
|
3858 highest_active_reg = r; |
|
|
3859 } |
|
|
3860 |
|
|
3861 /* If just failed to match something this time around with a |
|
|
3862 group that's operated on by a repetition operator, try to |
|
23
|
3863 force exit from the ``loop'', and restore the register |
|
2
|
3864 information for this group that we had before trying this |
|
|
3865 last match. */ |
|
|
3866 if ((!MATCHED_SOMETHING (reg_info[*p]) |
|
|
3867 || (re_opcode_t) p[-3] == start_memory) |
|
|
3868 && (p + 2) < pend) |
|
|
3869 { |
|
|
3870 boolean is_a_jump_n = false; |
|
|
3871 |
|
|
3872 p1 = p + 2; |
|
|
3873 mcnt = 0; |
|
|
3874 switch ((re_opcode_t) *p1++) |
|
|
3875 { |
|
|
3876 case jump_n: |
|
|
3877 is_a_jump_n = true; |
|
|
3878 case pop_failure_jump: |
|
|
3879 case maybe_pop_jump: |
|
|
3880 case jump: |
|
|
3881 case dummy_failure_jump: |
|
|
3882 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
3883 if (is_a_jump_n) |
|
|
3884 p1 += 2; |
|
|
3885 break; |
|
|
3886 |
|
|
3887 default: |
|
|
3888 /* do nothing */ ; |
|
|
3889 } |
|
|
3890 p1 += mcnt; |
|
|
3891 |
|
|
3892 /* If the next operation is a jump backwards in the pattern |
|
|
3893 to an on_failure_jump right before the start_memory |
|
|
3894 corresponding to this stop_memory, exit from the loop |
|
|
3895 by forcing a failure after pushing on the stack the |
|
|
3896 on_failure_jump's jump in the pattern, and d. */ |
|
|
3897 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump |
|
|
3898 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) |
|
|
3899 { |
|
|
3900 /* If this group ever matched anything, then restore |
|
|
3901 what its registers were before trying this last |
|
|
3902 failed match, e.g., with `(a*)*b' against `ab' for |
|
|
3903 regstart[1], and, e.g., with `((a*)*(b*)*)*' |
|
|
3904 against `aba' for regend[3]. |
|
|
3905 |
|
|
3906 Also restore the registers for inner groups for, |
|
|
3907 e.g., `((a*)(b*))*' against `aba' (register 3 would |
|
|
3908 otherwise get trashed). */ |
|
|
3909 |
|
|
3910 if (EVER_MATCHED_SOMETHING (reg_info[*p])) |
|
|
3911 { |
|
|
3912 unsigned r; |
|
|
3913 |
|
|
3914 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; |
|
|
3915 |
|
|
3916 /* Restore this and inner groups' (if any) registers. */ |
|
|
3917 for (r = *p; r < *p + *(p + 1); r++) |
|
|
3918 { |
|
|
3919 regstart[r] = old_regstart[r]; |
|
|
3920 |
|
|
3921 /* xx why this test? */ |
|
|
3922 if ((int) old_regend[r] >= (int) regstart[r]) |
|
|
3923 regend[r] = old_regend[r]; |
|
|
3924 } |
|
|
3925 } |
|
|
3926 p1++; |
|
|
3927 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
3928 PUSH_FAILURE_POINT (p1 + mcnt, d, -2); |
|
|
3929 |
|
|
3930 goto fail; |
|
|
3931 } |
|
|
3932 } |
|
|
3933 |
|
|
3934 /* Move past the register number and the inner group count. */ |
|
|
3935 p += 2; |
|
|
3936 break; |
|
|
3937 |
|
|
3938 |
|
|
3939 /* \<digit> has been turned into a `duplicate' command which is |
|
|
3940 followed by the numeric value of <digit> as the register number. */ |
|
|
3941 case duplicate: |
|
|
3942 { |
|
|
3943 register const char *d2, *dend2; |
|
|
3944 int regno = *p++; /* Get which register to match against. */ |
|
|
3945 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
|
|
3946 |
|
|
3947 /* Can't back reference a group which we've never matched. */ |
|
|
3948 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) |
|
|
3949 goto fail; |
|
|
3950 |
|
|
3951 /* Where in input to try to start matching. */ |
|
|
3952 d2 = regstart[regno]; |
|
|
3953 |
|
|
3954 /* Where to stop matching; if both the place to start and |
|
|
3955 the place to stop matching are in the same string, then |
|
|
3956 set to the place to stop, otherwise, for now have to use |
|
|
3957 the end of the first string. */ |
|
|
3958 |
|
|
3959 dend2 = ((FIRST_STRING_P (regstart[regno]) |
|
|
3960 == FIRST_STRING_P (regend[regno])) |
|
|
3961 ? regend[regno] : end_match_1); |
|
|
3962 for (;;) |
|
|
3963 { |
|
|
3964 /* If necessary, advance to next segment in register |
|
|
3965 contents. */ |
|
|
3966 while (d2 == dend2) |
|
|
3967 { |
|
|
3968 if (dend2 == end_match_2) break; |
|
|
3969 if (dend2 == regend[regno]) break; |
|
|
3970 |
|
|
3971 /* End of string1 => advance to string2. */ |
|
|
3972 d2 = string2; |
|
|
3973 dend2 = regend[regno]; |
|
|
3974 } |
|
|
3975 /* At end of register contents => success */ |
|
|
3976 if (d2 == dend2) break; |
|
|
3977 |
|
|
3978 /* If necessary, advance to next segment in data. */ |
|
|
3979 PREFETCH (); |
|
|
3980 |
|
|
3981 /* How many characters left in this segment to match. */ |
|
|
3982 mcnt = dend - d; |
|
|
3983 |
|
|
3984 /* Want how many consecutive characters we can match in |
|
|
3985 one shot, so, if necessary, adjust the count. */ |
|
|
3986 if (mcnt > dend2 - d2) |
|
|
3987 mcnt = dend2 - d2; |
|
|
3988 |
|
|
3989 /* Compare that many; failure if mismatch, else move |
|
|
3990 past them. */ |
|
|
3991 if (translate |
|
|
3992 ? bcmp_translate (d, d2, mcnt, translate) |
|
|
3993 : bcmp (d, d2, mcnt)) |
|
|
3994 goto fail; |
|
|
3995 d += mcnt, d2 += mcnt; |
|
|
3996 } |
|
|
3997 } |
|
|
3998 break; |
|
|
3999 |
|
|
4000 |
|
|
4001 /* begline matches the empty string at the beginning of the string |
|
|
4002 (unless `not_bol' is set in `bufp'), and, if |
|
|
4003 `newline_anchor' is set, after newlines. */ |
|
|
4004 case begline: |
|
|
4005 DEBUG_PRINT1 ("EXECUTING begline.\n"); |
|
|
4006 |
|
23
|
4007 if (AT_STRINGS_BEG (d)) |
|
2
|
4008 { |
|
|
4009 if (!bufp->not_bol) break; |
|
|
4010 } |
|
|
4011 else if (d[-1] == '\n' && bufp->newline_anchor) |
|
|
4012 { |
|
|
4013 break; |
|
|
4014 } |
|
|
4015 /* In all other cases, we fail. */ |
|
|
4016 goto fail; |
|
|
4017 |
|
|
4018 |
|
|
4019 /* endline is the dual of begline. */ |
|
|
4020 case endline: |
|
|
4021 DEBUG_PRINT1 ("EXECUTING endline.\n"); |
|
|
4022 |
|
23
|
4023 if (AT_STRINGS_END (d)) |
|
2
|
4024 { |
|
|
4025 if (!bufp->not_eol) break; |
|
|
4026 } |
|
|
4027 |
|
|
4028 /* We have to ``prefetch'' the next character. */ |
|
|
4029 else if ((d == end1 ? *string2 : *d) == '\n' |
|
|
4030 && bufp->newline_anchor) |
|
|
4031 { |
|
|
4032 break; |
|
|
4033 } |
|
|
4034 goto fail; |
|
|
4035 |
|
|
4036 |
|
|
4037 /* Match at the very beginning of the data. */ |
|
|
4038 case begbuf: |
|
|
4039 DEBUG_PRINT1 ("EXECUTING begbuf.\n"); |
|
23
|
4040 if (AT_STRINGS_BEG (d)) |
|
2
|
4041 break; |
|
|
4042 goto fail; |
|
|
4043 |
|
|
4044 |
|
|
4045 /* Match at the very end of the data. */ |
|
|
4046 case endbuf: |
|
|
4047 DEBUG_PRINT1 ("EXECUTING endbuf.\n"); |
|
23
|
4048 if (AT_STRINGS_END (d)) |
|
2
|
4049 break; |
|
|
4050 goto fail; |
|
|
4051 |
|
|
4052 |
|
|
4053 /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
|
|
4054 pushes NULL as the value for the string on the stack. Then |
|
|
4055 `pop_failure_point' will keep the current value for the |
|
|
4056 string, instead of restoring it. To see why, consider |
|
|
4057 matching `foo\nbar' against `.*\n'. The .* matches the foo; |
|
|
4058 then the . fails against the \n. But the next thing we want |
|
|
4059 to do is match the \n against the \n; if we restored the |
|
|
4060 string value, we would be back at the foo. |
|
|
4061 |
|
|
4062 Because this is used only in specific cases, we don't need to |
|
|
4063 check all the things that `on_failure_jump' does, to make |
|
|
4064 sure the right things get saved on the stack. Hence we don't |
|
|
4065 share its code. The only reason to push anything on the |
|
|
4066 stack at all is that otherwise we would have to change |
|
|
4067 `anychar's code to do something besides goto fail in this |
|
|
4068 case; that seems worse than this. */ |
|
|
4069 case on_failure_keep_string_jump: |
|
|
4070 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); |
|
|
4071 |
|
|
4072 EXTRACT_NUMBER_AND_INCR (mcnt, p); |
|
|
4073 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); |
|
|
4074 |
|
|
4075 PUSH_FAILURE_POINT (p + mcnt, NULL, -2); |
|
|
4076 break; |
|
|
4077 |
|
|
4078 |
|
|
4079 /* Uses of on_failure_jump: |
|
|
4080 |
|
|
4081 Each alternative starts with an on_failure_jump that points |
|
|
4082 to the beginning of the next alternative. Each alternative |
|
|
4083 except the last ends with a jump that in effect jumps past |
|
|
4084 the rest of the alternatives. (They really jump to the |
|
|
4085 ending jump of the following alternative, because tensioning |
|
|
4086 these jumps is a hassle.) |
|
|
4087 |
|
|
4088 Repeats start with an on_failure_jump that points past both |
|
|
4089 the repetition text and either the following jump or |
|
|
4090 pop_failure_jump back to this on_failure_jump. */ |
|
|
4091 case on_failure_jump: |
|
|
4092 on_failure: |
|
|
4093 DEBUG_PRINT1 ("EXECUTING on_failure_jump"); |
|
|
4094 |
|
|
4095 EXTRACT_NUMBER_AND_INCR (mcnt, p); |
|
|
4096 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); |
|
|
4097 |
|
|
4098 /* If this on_failure_jump comes right before a group (i.e., |
|
|
4099 the original * applied to a group), save the information |
|
|
4100 for that group and all inner ones, so that if we fail back |
|
|
4101 to this point, the group's information will be correct. |
|
23
|
4102 For example, in \(a*\)*\1, we need the preceding group, |
|
2
|
4103 and in \(\(a*\)b*\)\2, we need the inner group. */ |
|
|
4104 |
|
|
4105 /* We can't use `p' to check ahead because we push |
|
|
4106 a failure point to `p + mcnt' after we do this. */ |
|
|
4107 p1 = p; |
|
|
4108 |
|
|
4109 /* We need to skip no_op's before we look for the |
|
|
4110 start_memory in case this on_failure_jump is happening as |
|
|
4111 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 |
|
|
4112 against aba. */ |
|
|
4113 while (p1 < pend && (re_opcode_t) *p1 == no_op) |
|
|
4114 p1++; |
|
|
4115 |
|
|
4116 if (p1 < pend && (re_opcode_t) *p1 == start_memory) |
|
|
4117 { |
|
|
4118 /* We have a new highest active register now. This will |
|
|
4119 get reset at the start_memory we are about to get to, |
|
|
4120 but we will have saved all the registers relevant to |
|
|
4121 this repetition op, as described above. */ |
|
|
4122 highest_active_reg = *(p1 + 1) + *(p1 + 2); |
|
|
4123 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) |
|
|
4124 lowest_active_reg = *(p1 + 1); |
|
|
4125 } |
|
|
4126 |
|
|
4127 DEBUG_PRINT1 (":\n"); |
|
|
4128 PUSH_FAILURE_POINT (p + mcnt, d, -2); |
|
|
4129 break; |
|
|
4130 |
|
|
4131 |
|
23
|
4132 /* A smart repeat ends with `maybe_pop_jump'. |
|
|
4133 We change it to either `pop_failure_jump' or `jump'. */ |
|
2
|
4134 case maybe_pop_jump: |
|
|
4135 EXTRACT_NUMBER_AND_INCR (mcnt, p); |
|
|
4136 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); |
|
|
4137 { |
|
|
4138 register unsigned char *p2 = p; |
|
|
4139 |
|
|
4140 /* Compare the beginning of the repeat with what in the |
|
|
4141 pattern follows its end. If we can establish that there |
|
|
4142 is nothing that they would both match, i.e., that we |
|
|
4143 would have to backtrack because of (as in, e.g., `a*a') |
|
|
4144 then we can change to pop_failure_jump, because we'll |
|
|
4145 never have to backtrack. |
|
|
4146 |
|
|
4147 This is not true in the case of alternatives: in |
|
|
4148 `(a|ab)*' we do need to backtrack to the `ab' alternative |
|
|
4149 (e.g., if the string was `ab'). But instead of trying to |
|
|
4150 detect that here, the alternative has put on a dummy |
|
|
4151 failure point which is what we will end up popping. */ |
|
|
4152 |
|
92
|
4153 /* Skip over open/close-group commands. |
|
|
4154 If what follows this loop is a ...+ construct, |
|
|
4155 look at what begins its body, since we will have to |
|
|
4156 match at least one of that. */ |
|
|
4157 while (1) |
|
|
4158 { |
|
|
4159 if (p2 + 2 < pend |
|
|
4160 && ((re_opcode_t) *p2 == stop_memory |
|
|
4161 || (re_opcode_t) *p2 == start_memory)) |
|
|
4162 p2 += 3; |
|
|
4163 else if (p2 + 6 < pend |
|
|
4164 && (re_opcode_t) *p2 == dummy_failure_jump) |
|
|
4165 p2 += 6; |
|
|
4166 else |
|
|
4167 break; |
|
|
4168 } |
|
|
4169 |
|
|
4170 p1 = p + mcnt; |
|
|
4171 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding |
|
|
4172 to the `maybe_finalize_jump' of this case. Examine what |
|
|
4173 follows. */ |
|
2
|
4174 |
|
|
4175 /* If we're at the end of the pattern, we can change. */ |
|
|
4176 if (p2 == pend) |
|
29
|
4177 { |
|
|
4178 /* Consider what happens when matching ":\(.*\)" |
|
|
4179 against ":/". I don't really understand this code |
|
|
4180 yet. */ |
|
2
|
4181 p[-3] = (unsigned char) pop_failure_jump; |
|
29
|
4182 DEBUG_PRINT1 |
|
|
4183 (" End of pattern: change to `pop_failure_jump'.\n"); |
|
2
|
4184 } |
|
|
4185 |
|
|
4186 else if ((re_opcode_t) *p2 == exactn |
|
|
4187 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) |
|
|
4188 { |
|
|
4189 register unsigned char c |
|
|
4190 = *p2 == (unsigned char) endline ? '\n' : p2[2]; |
|
92
|
4191 |
|
2
|
4192 if ((re_opcode_t) p1[3] == exactn && p1[5] != c) |
|
23
|
4193 { |
|
|
4194 p[-3] = (unsigned char) pop_failure_jump; |
|
|
4195 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", |
|
|
4196 c, p1[5]); |
|
|
4197 } |
|
|
4198 |
|
2
|
4199 else if ((re_opcode_t) p1[3] == charset |
|
|
4200 || (re_opcode_t) p1[3] == charset_not) |
|
|
4201 { |
|
|
4202 int not = (re_opcode_t) p1[3] == charset_not; |
|
|
4203 |
|
|
4204 if (c < (unsigned char) (p1[4] * BYTEWIDTH) |
|
|
4205 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
|
|
4206 not = !not; |
|
|
4207 |
|
|
4208 /* `not' is equal to 1 if c would match, which means |
|
|
4209 that we can't change to pop_failure_jump. */ |
|
|
4210 if (!not) |
|
|
4211 { |
|
|
4212 p[-3] = (unsigned char) pop_failure_jump; |
|
23
|
4213 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
|
2
|
4214 } |
|
|
4215 } |
|
|
4216 } |
|
92
|
4217 else if ((re_opcode_t) *p2 == charset) |
|
|
4218 { |
|
|
4219 register unsigned char c |
|
|
4220 = *p2 == (unsigned char) endline ? '\n' : p2[2]; |
|
|
4221 |
|
|
4222 if ((re_opcode_t) p1[3] == exactn |
|
|
4223 && ! (p2[1] * BYTEWIDTH > p1[4] |
|
|
4224 && (p2[1 + p1[4] / BYTEWIDTH] |
|
|
4225 & (1 << (p1[4] % BYTEWIDTH))))) |
|
|
4226 { |
|
|
4227 p[-3] = (unsigned char) pop_failure_jump; |
|
|
4228 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", |
|
|
4229 c, p1[5]); |
|
|
4230 } |
|
|
4231 |
|
|
4232 else if ((re_opcode_t) p1[3] == charset_not) |
|
|
4233 { |
|
|
4234 int idx; |
|
|
4235 /* We win if the charset_not inside the loop |
|
|
4236 lists every character listed in the charset after. */ |
|
|
4237 for (idx = 0; idx < p2[1]; idx++) |
|
|
4238 if (! (p2[2 + idx] == 0 |
|
|
4239 || (idx < p1[4] |
|
|
4240 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) |
|
|
4241 break; |
|
|
4242 |
|
|
4243 if (idx == p2[1]) |
|
|
4244 { |
|
|
4245 p[-3] = (unsigned char) pop_failure_jump; |
|
|
4246 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
|
|
4247 } |
|
|
4248 } |
|
|
4249 else if ((re_opcode_t) p1[3] == charset) |
|
|
4250 { |
|
|
4251 int idx; |
|
|
4252 /* We win if the charset inside the loop |
|
|
4253 has no overlap with the one after the loop. */ |
|
|
4254 for (idx = 0; idx < p2[1] && idx < p1[4]; idx++) |
|
|
4255 if ((p2[2 + idx] & p1[5 + idx]) != 0) |
|
|
4256 break; |
|
|
4257 |
|
|
4258 if (idx == p2[1] || idx == p1[4]) |
|
|
4259 { |
|
|
4260 p[-3] = (unsigned char) pop_failure_jump; |
|
|
4261 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
|
|
4262 } |
|
|
4263 } |
|
|
4264 } |
|
2
|
4265 } |
|
|
4266 p -= 2; /* Point at relative address again. */ |
|
|
4267 if ((re_opcode_t) p[-1] != pop_failure_jump) |
|
|
4268 { |
|
|
4269 p[-1] = (unsigned char) jump; |
|
23
|
4270 DEBUG_PRINT1 (" Match => jump.\n"); |
|
2
|
4271 goto unconditional_jump; |
|
|
4272 } |
|
|
4273 /* Note fall through. */ |
|
|
4274 |
|
|
4275 |
|
|
4276 /* The end of a simple repeat has a pop_failure_jump back to |
|
|
4277 its matching on_failure_jump, where the latter will push a |
|
|
4278 failure point. The pop_failure_jump takes off failure |
|
|
4279 points put on by this pop_failure_jump's matching |
|
|
4280 on_failure_jump; we got through the pattern to here from the |
|
|
4281 matching on_failure_jump, so didn't fail. */ |
|
|
4282 case pop_failure_jump: |
|
|
4283 { |
|
|
4284 /* We need to pass separate storage for the lowest and |
|
|
4285 highest registers, even though we don't care about the |
|
|
4286 actual values. Otherwise, we will restore only one |
|
|
4287 register from the stack, since lowest will == highest in |
|
|
4288 `pop_failure_point'. */ |
|
|
4289 unsigned dummy_low_reg, dummy_high_reg; |
|
|
4290 unsigned char *pdummy; |
|
|
4291 const char *sdummy; |
|
|
4292 |
|
|
4293 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); |
|
|
4294 POP_FAILURE_POINT (sdummy, pdummy, |
|
|
4295 dummy_low_reg, dummy_high_reg, |
|
|
4296 reg_dummy, reg_dummy, reg_info_dummy); |
|
|
4297 } |
|
|
4298 /* Note fall through. */ |
|
|
4299 |
|
|
4300 |
|
|
4301 /* Unconditionally jump (without popping any failure points). */ |
|
|
4302 case jump: |
|
|
4303 unconditional_jump: |
|
|
4304 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
|
|
4305 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
|
|
4306 p += mcnt; /* Do the jump. */ |
|
|
4307 DEBUG_PRINT2 ("(to 0x%x).\n", p); |
|
|
4308 break; |
|
|
4309 |
|
|
4310 |
|
|
4311 /* We need this opcode so we can detect where alternatives end |
|
|
4312 in `group_match_null_string_p' et al. */ |
|
|
4313 case jump_past_alt: |
|
|
4314 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); |
|
|
4315 goto unconditional_jump; |
|
|
4316 |
|
|
4317 |
|
|
4318 /* Normally, the on_failure_jump pushes a failure point, which |
|
|
4319 then gets popped at pop_failure_jump. We will end up at |
|
|
4320 pop_failure_jump, also, and with a pattern of, say, `a+', we |
|
|
4321 are skipping over the on_failure_jump, so we have to push |
|
|
4322 something meaningless for pop_failure_jump to pop. */ |
|
|
4323 case dummy_failure_jump: |
|
|
4324 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); |
|
|
4325 /* It doesn't matter what we push for the string here. What |
|
|
4326 the code at `fail' tests is the value for the pattern. */ |
|
|
4327 PUSH_FAILURE_POINT (0, 0, -2); |
|
|
4328 goto unconditional_jump; |
|
|
4329 |
|
|
4330 |
|
|
4331 /* At the end of an alternative, we need to push a dummy failure |
|
23
|
4332 point in case we are followed by a `pop_failure_jump', because |
|
2
|
4333 we don't want the failure point for the alternative to be |
|
|
4334 popped. For example, matching `(a|ab)*' against `aab' |
|
|
4335 requires that we match the `ab' alternative. */ |
|
|
4336 case push_dummy_failure: |
|
|
4337 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); |
|
|
4338 /* See comments just above at `dummy_failure_jump' about the |
|
|
4339 two zeroes. */ |
|
|
4340 PUSH_FAILURE_POINT (0, 0, -2); |
|
|
4341 break; |
|
|
4342 |
|
|
4343 /* Have to succeed matching what follows at least n times. |
|
|
4344 After that, handle like `on_failure_jump'. */ |
|
|
4345 case succeed_n: |
|
|
4346 EXTRACT_NUMBER (mcnt, p + 2); |
|
|
4347 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); |
|
|
4348 |
|
|
4349 assert (mcnt >= 0); |
|
|
4350 /* Originally, this is how many times we HAVE to succeed. */ |
|
|
4351 if (mcnt > 0) |
|
|
4352 { |
|
|
4353 mcnt--; |
|
|
4354 p += 2; |
|
|
4355 STORE_NUMBER_AND_INCR (p, mcnt); |
|
|
4356 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); |
|
|
4357 } |
|
|
4358 else if (mcnt == 0) |
|
|
4359 { |
|
|
4360 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); |
|
|
4361 p[2] = (unsigned char) no_op; |
|
|
4362 p[3] = (unsigned char) no_op; |
|
|
4363 goto on_failure; |
|
|
4364 } |
|
|
4365 break; |
|
|
4366 |
|
|
4367 case jump_n: |
|
|
4368 EXTRACT_NUMBER (mcnt, p + 2); |
|
|
4369 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); |
|
|
4370 |
|
|
4371 /* Originally, this is how many times we CAN jump. */ |
|
|
4372 if (mcnt) |
|
|
4373 { |
|
|
4374 mcnt--; |
|
|
4375 STORE_NUMBER (p + 2, mcnt); |
|
|
4376 goto unconditional_jump; |
|
|
4377 } |
|
|
4378 /* If don't have to jump any more, skip over the rest of command. */ |
|
|
4379 else |
|
|
4380 p += 4; |
|
|
4381 break; |
|
|
4382 |
|
|
4383 case set_number_at: |
|
|
4384 { |
|
|
4385 DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
|
|
4386 |
|
|
4387 EXTRACT_NUMBER_AND_INCR (mcnt, p); |
|
|
4388 p1 = p + mcnt; |
|
|
4389 EXTRACT_NUMBER_AND_INCR (mcnt, p); |
|
|
4390 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); |
|
|
4391 STORE_NUMBER (p1, mcnt); |
|
|
4392 break; |
|
|
4393 } |
|
|
4394 |
|
|
4395 case wordbound: |
|
|
4396 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); |
|
|
4397 if (AT_WORD_BOUNDARY (d)) |
|
|
4398 break; |
|
|
4399 goto fail; |
|
|
4400 |
|
|
4401 case notwordbound: |
|
|
4402 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); |
|
|
4403 if (AT_WORD_BOUNDARY (d)) |
|
|
4404 goto fail; |
|
|
4405 break; |
|
|
4406 |
|
|
4407 case wordbeg: |
|
|
4408 DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
|
23
|
4409 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) |
|
2
|
4410 break; |
|
|
4411 goto fail; |
|
|
4412 |
|
|
4413 case wordend: |
|
|
4414 DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
|
23
|
4415 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) |
|
|
4416 && (!WORDCHAR_P (d) || AT_STRINGS_END (d))) |
|
2
|
4417 break; |
|
|
4418 goto fail; |
|
|
4419 |
|
|
4420 #ifdef emacs |
|
|
4421 #ifdef emacs19 |
|
|
4422 case before_dot: |
|
|
4423 DEBUG_PRINT1 ("EXECUTING before_dot.\n"); |
|
|
4424 if (PTR_CHAR_POS ((unsigned char *) d) >= point) |
|
|
4425 goto fail; |
|
|
4426 break; |
|
|
4427 |
|
|
4428 case at_dot: |
|
|
4429 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); |
|
|
4430 if (PTR_CHAR_POS ((unsigned char *) d) != point) |
|
|
4431 goto fail; |
|
|
4432 break; |
|
|
4433 |
|
|
4434 case after_dot: |
|
|
4435 DEBUG_PRINT1 ("EXECUTING after_dot.\n"); |
|
|
4436 if (PTR_CHAR_POS ((unsigned char *) d) <= point) |
|
|
4437 goto fail; |
|
|
4438 break; |
|
|
4439 #else /* not emacs19 */ |
|
|
4440 case at_dot: |
|
|
4441 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); |
|
|
4442 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point) |
|
|
4443 goto fail; |
|
|
4444 break; |
|
|
4445 #endif /* not emacs19 */ |
|
|
4446 |
|
|
4447 case syntaxspec: |
|
|
4448 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); |
|
|
4449 mcnt = *p++; |
|
|
4450 goto matchsyntax; |
|
|
4451 |
|
|
4452 case wordchar: |
|
23
|
4453 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); |
|
2
|
4454 mcnt = (int) Sword; |
|
|
4455 matchsyntax: |
|
|
4456 PREFETCH (); |
|
23
|
4457 if (SYNTAX (*d++) != (enum syntaxcode) mcnt) |
|
|
4458 goto fail; |
|
2
|
4459 SET_REGS_MATCHED (); |
|
|
4460 break; |
|
|
4461 |
|
|
4462 case notsyntaxspec: |
|
|
4463 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); |
|
|
4464 mcnt = *p++; |
|
|
4465 goto matchnotsyntax; |
|
|
4466 |
|
|
4467 case notwordchar: |
|
23
|
4468 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); |
|
2
|
4469 mcnt = (int) Sword; |
|
23
|
4470 matchnotsyntax: |
|
2
|
4471 PREFETCH (); |
|
23
|
4472 if (SYNTAX (*d++) == (enum syntaxcode) mcnt) |
|
|
4473 goto fail; |
|
2
|
4474 SET_REGS_MATCHED (); |
|
|
4475 break; |
|
|
4476 |
|
|
4477 #else /* not emacs */ |
|
|
4478 case wordchar: |
|
|
4479 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); |
|
|
4480 PREFETCH (); |
|
23
|
4481 if (!WORDCHAR_P (d)) |
|
2
|
4482 goto fail; |
|
|
4483 SET_REGS_MATCHED (); |
|
23
|
4484 d++; |
|
2
|
4485 break; |
|
|
4486 |
|
|
4487 case notwordchar: |
|
|
4488 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); |
|
|
4489 PREFETCH (); |
|
23
|
4490 if (WORDCHAR_P (d)) |
|
2
|
4491 goto fail; |
|
|
4492 SET_REGS_MATCHED (); |
|
23
|
4493 d++; |
|
2
|
4494 break; |
|
|
4495 #endif /* not emacs */ |
|
|
4496 |
|
|
4497 default: |
|
|
4498 abort (); |
|
|
4499 } |
|
|
4500 continue; /* Successfully executed one pattern command; keep going. */ |
|
|
4501 |
|
|
4502 |
|
|
4503 /* We goto here if a matching operation fails. */ |
|
|
4504 fail: |
|
|
4505 if (!FAIL_STACK_EMPTY ()) |
|
|
4506 { /* A restart point is known. Restore to that state. */ |
|
|
4507 DEBUG_PRINT1 ("\nFAIL:\n"); |
|
|
4508 POP_FAILURE_POINT (d, p, |
|
|
4509 lowest_active_reg, highest_active_reg, |
|
|
4510 regstart, regend, reg_info); |
|
|
4511 |
|
|
4512 /* If this failure point is a dummy, try the next one. */ |
|
|
4513 if (!p) |
|
|
4514 goto fail; |
|
|
4515 |
|
|
4516 /* If we failed to the end of the pattern, don't examine *p. */ |
|
|
4517 assert (p <= pend); |
|
|
4518 if (p < pend) |
|
|
4519 { |
|
|
4520 boolean is_a_jump_n = false; |
|
|
4521 |
|
|
4522 /* If failed to a backwards jump that's part of a repetition |
|
|
4523 loop, need to pop this failure point and use the next one. */ |
|
|
4524 switch ((re_opcode_t) *p) |
|
|
4525 { |
|
|
4526 case jump_n: |
|
|
4527 is_a_jump_n = true; |
|
|
4528 case maybe_pop_jump: |
|
|
4529 case pop_failure_jump: |
|
|
4530 case jump: |
|
|
4531 p1 = p + 1; |
|
|
4532 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
4533 p1 += mcnt; |
|
|
4534 |
|
|
4535 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) |
|
|
4536 || (!is_a_jump_n |
|
|
4537 && (re_opcode_t) *p1 == on_failure_jump)) |
|
|
4538 goto fail; |
|
|
4539 break; |
|
|
4540 default: |
|
|
4541 /* do nothing */ ; |
|
|
4542 } |
|
|
4543 } |
|
|
4544 |
|
|
4545 if (d >= string1 && d <= end1) |
|
|
4546 dend = end_match_1; |
|
|
4547 } |
|
|
4548 else |
|
|
4549 break; /* Matching at this starting point really fails. */ |
|
|
4550 } /* for (;;) */ |
|
|
4551 |
|
|
4552 if (best_regs_set) |
|
|
4553 goto restore_best_regs; |
|
|
4554 |
|
|
4555 FREE_VARIABLES (); |
|
|
4556 |
|
|
4557 return -1; /* Failure to match. */ |
|
|
4558 } /* re_match_2 */ |
|
|
4559 |
|
|
4560 /* Subroutine definitions for re_match_2. */ |
|
|
4561 |
|
|
4562 |
|
|
4563 /* We are passed P pointing to a register number after a start_memory. |
|
|
4564 |
|
|
4565 Return true if the pattern up to the corresponding stop_memory can |
|
|
4566 match the empty string, and false otherwise. |
|
|
4567 |
|
|
4568 If we find the matching stop_memory, sets P to point to one past its number. |
|
|
4569 Otherwise, sets P to an undefined byte less than or equal to END. |
|
|
4570 |
|
|
4571 We don't handle duplicates properly (yet). */ |
|
|
4572 |
|
|
4573 static boolean |
|
|
4574 group_match_null_string_p (p, end, reg_info) |
|
|
4575 unsigned char **p, *end; |
|
|
4576 register_info_type *reg_info; |
|
|
4577 { |
|
|
4578 int mcnt; |
|
|
4579 /* Point to after the args to the start_memory. */ |
|
|
4580 unsigned char *p1 = *p + 2; |
|
|
4581 |
|
|
4582 while (p1 < end) |
|
|
4583 { |
|
|
4584 /* Skip over opcodes that can match nothing, and return true or |
|
|
4585 false, as appropriate, when we get to one that can't, or to the |
|
|
4586 matching stop_memory. */ |
|
|
4587 |
|
|
4588 switch ((re_opcode_t) *p1) |
|
|
4589 { |
|
|
4590 /* Could be either a loop or a series of alternatives. */ |
|
|
4591 case on_failure_jump: |
|
|
4592 p1++; |
|
|
4593 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
4594 |
|
|
4595 /* If the next operation is not a jump backwards in the |
|
|
4596 pattern. */ |
|
|
4597 |
|
|
4598 if (mcnt >= 0) |
|
|
4599 { |
|
|
4600 /* Go through the on_failure_jumps of the alternatives, |
|
|
4601 seeing if any of the alternatives cannot match nothing. |
|
|
4602 The last alternative starts with only a jump, |
|
|
4603 whereas the rest start with on_failure_jump and end |
|
|
4604 with a jump, e.g., here is the pattern for `a|b|c': |
|
|
4605 |
|
|
4606 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 |
|
|
4607 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 |
|
|
4608 /exactn/1/c |
|
|
4609 |
|
|
4610 So, we have to first go through the first (n-1) |
|
|
4611 alternatives and then deal with the last one separately. */ |
|
|
4612 |
|
|
4613 |
|
|
4614 /* Deal with the first (n-1) alternatives, which start |
|
|
4615 with an on_failure_jump (see above) that jumps to right |
|
|
4616 past a jump_past_alt. */ |
|
|
4617 |
|
|
4618 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) |
|
|
4619 { |
|
|
4620 /* `mcnt' holds how many bytes long the alternative |
|
|
4621 is, including the ending `jump_past_alt' and |
|
|
4622 its number. */ |
|
|
4623 |
|
|
4624 if (!alt_match_null_string_p (p1, p1 + mcnt - 3, |
|
|
4625 reg_info)) |
|
|
4626 return false; |
|
|
4627 |
|
|
4628 /* Move to right after this alternative, including the |
|
|
4629 jump_past_alt. */ |
|
|
4630 p1 += mcnt; |
|
|
4631 |
|
|
4632 /* Break if it's the beginning of an n-th alternative |
|
|
4633 that doesn't begin with an on_failure_jump. */ |
|
|
4634 if ((re_opcode_t) *p1 != on_failure_jump) |
|
|
4635 break; |
|
|
4636 |
|
|
4637 /* Still have to check that it's not an n-th |
|
|
4638 alternative that starts with an on_failure_jump. */ |
|
|
4639 p1++; |
|
|
4640 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
4641 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) |
|
|
4642 { |
|
|
4643 /* Get to the beginning of the n-th alternative. */ |
|
|
4644 p1 -= 3; |
|
|
4645 break; |
|
|
4646 } |
|
|
4647 } |
|
|
4648 |
|
|
4649 /* Deal with the last alternative: go back and get number |
|
|
4650 of the `jump_past_alt' just before it. `mcnt' contains |
|
|
4651 the length of the alternative. */ |
|
|
4652 EXTRACT_NUMBER (mcnt, p1 - 2); |
|
|
4653 |
|
|
4654 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) |
|
|
4655 return false; |
|
|
4656 |
|
|
4657 p1 += mcnt; /* Get past the n-th alternative. */ |
|
|
4658 } /* if mcnt > 0 */ |
|
|
4659 break; |
|
|
4660 |
|
|
4661 |
|
|
4662 case stop_memory: |
|
|
4663 assert (p1[1] == **p); |
|
|
4664 *p = p1 + 2; |
|
|
4665 return true; |
|
|
4666 |
|
|
4667 |
|
|
4668 default: |
|
|
4669 if (!common_op_match_null_string_p (&p1, end, reg_info)) |
|
|
4670 return false; |
|
|
4671 } |
|
|
4672 } /* while p1 < end */ |
|
|
4673 |
|
|
4674 return false; |
|
|
4675 } /* group_match_null_string_p */ |
|
|
4676 |
|
|
4677 |
|
|
4678 /* Similar to group_match_null_string_p, but doesn't deal with alternatives: |
|
|
4679 It expects P to be the first byte of a single alternative and END one |
|
|
4680 byte past the last. The alternative can contain groups. */ |
|
|
4681 |
|
|
4682 static boolean |
|
|
4683 alt_match_null_string_p (p, end, reg_info) |
|
|
4684 unsigned char *p, *end; |
|
|
4685 register_info_type *reg_info; |
|
|
4686 { |
|
|
4687 int mcnt; |
|
|
4688 unsigned char *p1 = p; |
|
|
4689 |
|
|
4690 while (p1 < end) |
|
|
4691 { |
|
|
4692 /* Skip over opcodes that can match nothing, and break when we get |
|
|
4693 to one that can't. */ |
|
|
4694 |
|
|
4695 switch ((re_opcode_t) *p1) |
|
|
4696 { |
|
|
4697 /* It's a loop. */ |
|
|
4698 case on_failure_jump: |
|
|
4699 p1++; |
|
|
4700 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
4701 p1 += mcnt; |
|
|
4702 break; |
|
|
4703 |
|
|
4704 default: |
|
|
4705 if (!common_op_match_null_string_p (&p1, end, reg_info)) |
|
|
4706 return false; |
|
|
4707 } |
|
|
4708 } /* while p1 < end */ |
|
|
4709 |
|
|
4710 return true; |
|
|
4711 } /* alt_match_null_string_p */ |
|
|
4712 |
|
|
4713 |
|
|
4714 /* Deals with the ops common to group_match_null_string_p and |
|
|
4715 alt_match_null_string_p. |
|
|
4716 |
|
|
4717 Sets P to one after the op and its arguments, if any. */ |
|
|
4718 |
|
|
4719 static boolean |
|
|
4720 common_op_match_null_string_p (p, end, reg_info) |
|
|
4721 unsigned char **p, *end; |
|
|
4722 register_info_type *reg_info; |
|
|
4723 { |
|
|
4724 int mcnt; |
|
|
4725 boolean ret; |
|
|
4726 int reg_no; |
|
|
4727 unsigned char *p1 = *p; |
|
|
4728 |
|
|
4729 switch ((re_opcode_t) *p1++) |
|
|
4730 { |
|
|
4731 case no_op: |
|
|
4732 case begline: |
|
|
4733 case endline: |
|
|
4734 case begbuf: |
|
|
4735 case endbuf: |
|
|
4736 case wordbeg: |
|
|
4737 case wordend: |
|
|
4738 case wordbound: |
|
|
4739 case notwordbound: |
|
|
4740 #ifdef emacs |
|
|
4741 case before_dot: |
|
|
4742 case at_dot: |
|
|
4743 case after_dot: |
|
|
4744 #endif |
|
|
4745 break; |
|
|
4746 |
|
|
4747 case start_memory: |
|
|
4748 reg_no = *p1; |
|
|
4749 assert (reg_no > 0 && reg_no <= MAX_REGNUM); |
|
|
4750 ret = group_match_null_string_p (&p1, end, reg_info); |
|
|
4751 |
|
|
4752 /* Have to set this here in case we're checking a group which |
|
|
4753 contains a group and a back reference to it. */ |
|
|
4754 |
|
|
4755 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) |
|
|
4756 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; |
|
|
4757 |
|
|
4758 if (!ret) |
|
|
4759 return false; |
|
|
4760 break; |
|
|
4761 |
|
|
4762 /* If this is an optimized succeed_n for zero times, make the jump. */ |
|
|
4763 case jump: |
|
|
4764 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
4765 if (mcnt >= 0) |
|
|
4766 p1 += mcnt; |
|
|
4767 else |
|
|
4768 return false; |
|
|
4769 break; |
|
|
4770 |
|
|
4771 case succeed_n: |
|
|
4772 /* Get to the number of times to succeed. */ |
|
|
4773 p1 += 2; |
|
|
4774 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
4775 |
|
|
4776 if (mcnt == 0) |
|
|
4777 { |
|
|
4778 p1 -= 4; |
|
|
4779 EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
|
|
4780 p1 += mcnt; |
|
|
4781 } |
|
|
4782 else |
|
|
4783 return false; |
|
|
4784 break; |
|
|
4785 |
|
|
4786 case duplicate: |
|
|
4787 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) |
|
|
4788 return false; |
|
|
4789 break; |
|
|
4790 |
|
|
4791 case set_number_at: |
|
|
4792 p1 += 4; |
|
|
4793 |
|
|
4794 default: |
|
|
4795 /* All other opcodes mean we cannot match the empty string. */ |
|
|
4796 return false; |
|
|
4797 } |
|
|
4798 |
|
|
4799 *p = p1; |
|
|
4800 return true; |
|
|
4801 } /* common_op_match_null_string_p */ |
|
|
4802 |
|
|
4803 |
|
|
4804 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
|
|
4805 bytes; nonzero otherwise. */ |
|
|
4806 |
|
|
4807 static int |
|
|
4808 bcmp_translate (s1, s2, len, translate) |
|
|
4809 unsigned char *s1, *s2; |
|
|
4810 register int len; |
|
|
4811 char *translate; |
|
|
4812 { |
|
|
4813 register unsigned char *p1 = s1, *p2 = s2; |
|
|
4814 while (len) |
|
|
4815 { |
|
|
4816 if (translate[*p1++] != translate[*p2++]) return 1; |
|
|
4817 len--; |
|
|
4818 } |
|
|
4819 return 0; |
|
|
4820 } |
|
|
4821 |
|
|
4822 /* Entry points for GNU code. */ |
|
|
4823 |
|
|
4824 /* re_compile_pattern is the GNU regular expression compiler: it |
|
|
4825 compiles PATTERN (of length SIZE) and puts the result in BUFP. |
|
|
4826 Returns 0 if the pattern was valid, otherwise an error string. |
|
|
4827 |
|
|
4828 Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
|
|
4829 are set in BUFP on entry. |
|
|
4830 |
|
|
4831 We call regex_compile to do the actual compilation. */ |
|
|
4832 |
|
|
4833 const char * |
|
|
4834 re_compile_pattern (pattern, length, bufp) |
|
|
4835 const char *pattern; |
|
|
4836 int length; |
|
|
4837 struct re_pattern_buffer *bufp; |
|
|
4838 { |
|
|
4839 reg_errcode_t ret; |
|
|
4840 |
|
|
4841 /* GNU code is written to assume at least RE_NREGS registers will be set |
|
|
4842 (and at least one extra will be -1). */ |
|
|
4843 bufp->regs_allocated = REGS_UNALLOCATED; |
|
|
4844 |
|
|
4845 /* And GNU code determines whether or not to get register information |
|
|
4846 by passing null for the REGS argument to re_match, etc., not by |
|
|
4847 setting no_sub. */ |
|
|
4848 bufp->no_sub = 0; |
|
|
4849 |
|
|
4850 /* Match anchors at newline. */ |
|
|
4851 bufp->newline_anchor = 1; |
|
|
4852 |
|
|
4853 ret = regex_compile (pattern, length, re_syntax_options, bufp); |
|
|
4854 |
|
|
4855 return re_error_msg[(int) ret]; |
|
|
4856 } |
|
|
4857 |
|
|
4858 /* Entry points compatible with 4.2 BSD regex library. We don't define |
|
|
4859 them if this is an Emacs or POSIX compilation. */ |
|
|
4860 |
|
|
4861 #if !defined (emacs) && !defined (_POSIX_SOURCE) |
|
|
4862 |
|
|
4863 /* BSD has one and only one pattern buffer. */ |
|
|
4864 static struct re_pattern_buffer re_comp_buf; |
|
|
4865 |
|
|
4866 char * |
|
|
4867 re_comp (s) |
|
|
4868 const char *s; |
|
|
4869 { |
|
|
4870 reg_errcode_t ret; |
|
|
4871 |
|
|
4872 if (!s) |
|
|
4873 { |
|
|
4874 if (!re_comp_buf.buffer) |
|
|
4875 return "No previous regular expression"; |
|
|
4876 return 0; |
|
|
4877 } |
|
|
4878 |
|
|
4879 if (!re_comp_buf.buffer) |
|
|
4880 { |
|
|
4881 re_comp_buf.buffer = (unsigned char *) malloc (200); |
|
|
4882 if (re_comp_buf.buffer == NULL) |
|
|
4883 return "Memory exhausted"; |
|
|
4884 re_comp_buf.allocated = 200; |
|
|
4885 |
|
|
4886 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); |
|
|
4887 if (re_comp_buf.fastmap == NULL) |
|
|
4888 return "Memory exhausted"; |
|
|
4889 } |
|
|
4890 |
|
|
4891 /* Since `re_exec' always passes NULL for the `regs' argument, we |
|
|
4892 don't need to initialize the pattern buffer fields which affect it. */ |
|
|
4893 |
|
|
4894 /* Match anchors at newlines. */ |
|
|
4895 re_comp_buf.newline_anchor = 1; |
|
|
4896 |
|
|
4897 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
|
|
4898 |
|
|
4899 /* Yes, we're discarding `const' here. */ |
|
|
4900 return (char *) re_error_msg[(int) ret]; |
|
|
4901 } |
|
|
4902 |
|
|
4903 |
|
|
4904 int |
|
|
4905 re_exec (s) |
|
|
4906 const char *s; |
|
|
4907 { |
|
|
4908 const int len = strlen (s); |
|
|
4909 return |
|
|
4910 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); |
|
|
4911 } |
|
|
4912 #endif /* not emacs and not _POSIX_SOURCE */ |
|
|
4913 |
|
|
4914 /* POSIX.2 functions. Don't define these for Emacs. */ |
|
|
4915 |
|
|
4916 #ifndef emacs |
|
|
4917 |
|
|
4918 /* regcomp takes a regular expression as a string and compiles it. |
|
|
4919 |
|
|
4920 PREG is a regex_t *. We do not expect any fields to be initialized, |
|
|
4921 since POSIX says we shouldn't. Thus, we set |
|
|
4922 |
|
|
4923 `buffer' to the compiled pattern; |
|
|
4924 `used' to the length of the compiled pattern; |
|
|
4925 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the |
|
|
4926 REG_EXTENDED bit in CFLAGS is set; otherwise, to |
|
|
4927 RE_SYNTAX_POSIX_BASIC; |
|
|
4928 `newline_anchor' to REG_NEWLINE being set in CFLAGS; |
|
|
4929 `fastmap' and `fastmap_accurate' to zero; |
|
|
4930 `re_nsub' to the number of subexpressions in PATTERN. |
|
|
4931 |
|
|
4932 PATTERN is the address of the pattern string. |
|
|
4933 |
|
|
4934 CFLAGS is a series of bits which affect compilation. |
|
|
4935 |
|
|
4936 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we |
|
|
4937 use POSIX basic syntax. |
|
|
4938 |
|
|
4939 If REG_NEWLINE is set, then . and [^...] don't match newline. |
|
|
4940 Also, regexec will try a match beginning after every newline. |
|
|
4941 |
|
|
4942 If REG_ICASE is set, then we considers upper- and lowercase |
|
|
4943 versions of letters to be equivalent when matching. |
|
|
4944 |
|
|
4945 If REG_NOSUB is set, then when PREG is passed to regexec, that |
|
|
4946 routine will report only success or failure, and nothing about the |
|
|
4947 registers. |
|
|
4948 |
|
|
4949 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
|
|
4950 the return codes and their meanings.) */ |
|
|
4951 |
|
|
4952 int |
|
|
4953 regcomp (preg, pattern, cflags) |
|
|
4954 regex_t *preg; |
|
|
4955 const char *pattern; |
|
|
4956 int cflags; |
|
|
4957 { |
|
|
4958 reg_errcode_t ret; |
|
|
4959 unsigned syntax |
|
25
|
4960 = (cflags & REG_EXTENDED) ? |
|
|
4961 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; |
|
2
|
4962 |
|
|
4963 /* regex_compile will allocate the space for the compiled pattern. */ |
|
|
4964 preg->buffer = 0; |
|
25
|
4965 preg->allocated = 0; |
|
84
|
4966 preg->used = 0; |
|
2
|
4967 |
|
|
4968 /* Don't bother to use a fastmap when searching. This simplifies the |
|
|
4969 REG_NEWLINE case: if we used a fastmap, we'd have to put all the |
|
|
4970 characters after newlines into the fastmap. This way, we just try |
|
|
4971 every character. */ |
|
|
4972 preg->fastmap = 0; |
|
|
4973 |
|
|
4974 if (cflags & REG_ICASE) |
|
|
4975 { |
|
|
4976 unsigned i; |
|
|
4977 |
|
|
4978 preg->translate = (char *) malloc (CHAR_SET_SIZE); |
|
|
4979 if (preg->translate == NULL) |
|
|
4980 return (int) REG_ESPACE; |
|
|
4981 |
|
|
4982 /* Map uppercase characters to corresponding lowercase ones. */ |
|
|
4983 for (i = 0; i < CHAR_SET_SIZE; i++) |
|
28
|
4984 preg->translate[i] = ISUPPER (i) ? tolower (i) : i; |
|
2
|
4985 } |
|
|
4986 else |
|
|
4987 preg->translate = NULL; |
|
|
4988 |
|
|
4989 /* If REG_NEWLINE is set, newlines are treated differently. */ |
|
|
4990 if (cflags & REG_NEWLINE) |
|
|
4991 { /* REG_NEWLINE implies neither . nor [^...] match newline. */ |
|
|
4992 syntax &= ~RE_DOT_NEWLINE; |
|
|
4993 syntax |= RE_HAT_LISTS_NOT_NEWLINE; |
|
|
4994 /* It also changes the matching behavior. */ |
|
|
4995 preg->newline_anchor = 1; |
|
|
4996 } |
|
|
4997 else |
|
|
4998 preg->newline_anchor = 0; |
|
|
4999 |
|
|
5000 preg->no_sub = !!(cflags & REG_NOSUB); |
|
|
5001 |
|
|
5002 /* POSIX says a null character in the pattern terminates it, so we |
|
|
5003 can use strlen here in compiling the pattern. */ |
|
|
5004 ret = regex_compile (pattern, strlen (pattern), syntax, preg); |
|
|
5005 |
|
|
5006 /* POSIX doesn't distinguish between an unmatched open-group and an |
|
|
5007 unmatched close-group: both are REG_EPAREN. */ |
|
|
5008 if (ret == REG_ERPAREN) ret = REG_EPAREN; |
|
|
5009 |
|
|
5010 return (int) ret; |
|
|
5011 } |
|
|
5012 |
|
|
5013 |
|
|
5014 /* regexec searches for a given pattern, specified by PREG, in the |
|
|
5015 string STRING. |
|
|
5016 |
|
|
5017 If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
|
|
5018 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
|
|
5019 least NMATCH elements, and we set them to the offsets of the |
|
|
5020 corresponding matched substrings. |
|
|
5021 |
|
|
5022 EFLAGS specifies `execution flags' which affect matching: if |
|
|
5023 REG_NOTBOL is set, then ^ does not match at the beginning of the |
|
|
5024 string; if REG_NOTEOL is set, then $ does not match at the end. |
|
|
5025 |
|
|
5026 We return 0 if we find a match and REG_NOMATCH if not. */ |
|
|
5027 |
|
|
5028 int |
|
|
5029 regexec (preg, string, nmatch, pmatch, eflags) |
|
|
5030 const regex_t *preg; |
|
|
5031 const char *string; |
|
|
5032 size_t nmatch; |
|
|
5033 regmatch_t pmatch[]; |
|
|
5034 int eflags; |
|
|
5035 { |
|
|
5036 int ret; |
|
|
5037 struct re_registers regs; |
|
|
5038 regex_t private_preg; |
|
|
5039 int len = strlen (string); |
|
|
5040 boolean want_reg_info = !preg->no_sub && nmatch > 0; |
|
|
5041 |
|
|
5042 private_preg = *preg; |
|
|
5043 |
|
|
5044 private_preg.not_bol = !!(eflags & REG_NOTBOL); |
|
|
5045 private_preg.not_eol = !!(eflags & REG_NOTEOL); |
|
|
5046 |
|
|
5047 /* The user has told us exactly how many registers to return |
|
|
5048 information about, via `nmatch'. We have to pass that on to the |
|
|
5049 matching routines. */ |
|
|
5050 private_preg.regs_allocated = REGS_FIXED; |
|
|
5051 |
|
|
5052 if (want_reg_info) |
|
|
5053 { |
|
|
5054 regs.num_regs = nmatch; |
|
|
5055 regs.start = TALLOC (nmatch, regoff_t); |
|
|
5056 regs.end = TALLOC (nmatch, regoff_t); |
|
|
5057 if (regs.start == NULL || regs.end == NULL) |
|
|
5058 return (int) REG_NOMATCH; |
|
|
5059 } |
|
|
5060 |
|
|
5061 /* Perform the searching operation. */ |
|
|
5062 ret = re_search (&private_preg, string, len, |
|
|
5063 /* start: */ 0, /* range: */ len, |
|
|
5064 want_reg_info ? ®s : (struct re_registers *) 0); |
|
|
5065 |
|
|
5066 /* Copy the register information to the POSIX structure. */ |
|
|
5067 if (want_reg_info) |
|
|
5068 { |
|
|
5069 if (ret >= 0) |
|
|
5070 { |
|
|
5071 unsigned r; |
|
|
5072 |
|
|
5073 for (r = 0; r < nmatch; r++) |
|
|
5074 { |
|
|
5075 pmatch[r].rm_so = regs.start[r]; |
|
|
5076 pmatch[r].rm_eo = regs.end[r]; |
|
|
5077 } |
|
|
5078 } |
|
|
5079 |
|
|
5080 /* If we needed the temporary register info, free the space now. */ |
|
|
5081 free (regs.start); |
|
|
5082 free (regs.end); |
|
|
5083 } |
|
|
5084 |
|
|
5085 /* We want zero return to mean success, unlike `re_search'. */ |
|
|
5086 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; |
|
|
5087 } |
|
|
5088 |
|
|
5089 |
|
|
5090 /* Returns a message corresponding to an error code, ERRCODE, returned |
|
23
|
5091 from either regcomp or regexec. We don't use PREG here. */ |
|
2
|
5092 |
|
|
5093 size_t |
|
|
5094 regerror (errcode, preg, errbuf, errbuf_size) |
|
|
5095 int errcode; |
|
|
5096 const regex_t *preg; |
|
|
5097 char *errbuf; |
|
|
5098 size_t errbuf_size; |
|
|
5099 { |
|
34
|
5100 const char *msg; |
|
|
5101 size_t msg_size; |
|
|
5102 |
|
|
5103 if (errcode < 0 |
|
|
5104 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0]))) |
|
|
5105 /* Only error codes returned by the rest of the code should be passed |
|
|
5106 to this routine. If we are given anything else, or if other regex |
|
|
5107 code generates an invalid error code, then the program has a bug. |
|
|
5108 Dump core so we can fix it. */ |
|
|
5109 abort (); |
|
|
5110 |
|
45
|
5111 msg = re_error_msg[errcode]; |
|
|
5112 |
|
|
5113 /* POSIX doesn't require that we do anything in this case, but why |
|
|
5114 not be nice. */ |
|
|
5115 if (! msg) |
|
|
5116 msg = "Success"; |
|
|
5117 |
|
34
|
5118 msg_size = strlen (msg) + 1; /* Includes the null. */ |
|
2
|
5119 |
|
|
5120 if (errbuf_size != 0) |
|
|
5121 { |
|
|
5122 if (msg_size > errbuf_size) |
|
|
5123 { |
|
|
5124 strncpy (errbuf, msg, errbuf_size - 1); |
|
|
5125 errbuf[errbuf_size - 1] = 0; |
|
|
5126 } |
|
|
5127 else |
|
|
5128 strcpy (errbuf, msg); |
|
|
5129 } |
|
|
5130 |
|
|
5131 return msg_size; |
|
|
5132 } |
|
|
5133 |
|
|
5134 |
|
|
5135 /* Free dynamically allocated space used by PREG. */ |
|
|
5136 |
|
|
5137 void |
|
|
5138 regfree (preg) |
|
|
5139 regex_t *preg; |
|
|
5140 { |
|
|
5141 if (preg->buffer != NULL) |
|
|
5142 free (preg->buffer); |
|
|
5143 preg->buffer = NULL; |
|
|
5144 |
|
|
5145 preg->allocated = 0; |
|
|
5146 preg->used = 0; |
|
|
5147 |
|
|
5148 if (preg->fastmap != NULL) |
|
|
5149 free (preg->fastmap); |
|
|
5150 preg->fastmap = NULL; |
|
|
5151 preg->fastmap_accurate = 0; |
|
|
5152 |
|
|
5153 if (preg->translate != NULL) |
|
|
5154 free (preg->translate); |
|
|
5155 preg->translate = NULL; |
|
|
5156 } |
|
|
5157 |
|
|
5158 #endif /* not emacs */ |
|
|
5159 |
|
|
5160 /* |
|
|
5161 Local variables: |
|
|
5162 make-backup-files: t |
|
|
5163 version-control: t |
|
|
5164 trim-versions-without-asking: nil |
|
|
5165 End: |
|
|
5166 */ |
