Commit | Line | Data |
---|---|---|
2a6ef469 DD |
1 | /* Extended regular expression matching and search library, |
2 | version 0.12. | |
3 | (Implements POSIX draft P1003.2/D11.2, except for some of the | |
4 | internationalization features.) | |
5 | Copyright (C) 1993-1999, 2000, 2001 Free Software Foundation, Inc. | |
6ad8a379 | 6 | This file is part of the GNU C Library. |
2a6ef469 DD |
7 | |
8 | The GNU C Library is free software; you can redistribute it and/or | |
6ad8a379 DD |
9 | modify it under the terms of the GNU Lesser General Public |
10 | License as published by the Free Software Foundation; either | |
11 | version 2.1 of the License, or (at your option) any later version. | |
2a6ef469 DD |
12 | |
13 | The GNU C Library 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 GNU | |
6ad8a379 | 16 | Lesser General Public License for more details. |
2a6ef469 | 17 | |
6ad8a379 DD |
18 | You should have received a copy of the GNU Lesser General Public |
19 | License along with the GNU C Library; if not, write to the Free | |
20 | Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA | |
21 | 02111-1307 USA. */ | |
2a6ef469 DD |
22 | |
23 | /* This file has been modified for usage in libiberty. It includes "xregex.h" | |
24 | instead of <regex.h>. The "xregex.h" header file renames all external | |
25 | routines with an "x" prefix so they do not collide with the native regex | |
26 | routines or with other components regex routines. */ | |
6ad8a379 | 27 | /* AIX requires this to be the first thing in the file. */ |
2a6ef469 DD |
28 | #if defined _AIX && !defined REGEX_MALLOC |
29 | #pragma alloca | |
30 | #endif | |
31 | ||
32 | #undef _GNU_SOURCE | |
33 | #define _GNU_SOURCE | |
34 | ||
35 | #ifdef HAVE_CONFIG_H | |
36 | # include <config.h> | |
37 | #endif | |
38 | ||
39 | #ifndef PARAMS | |
40 | # if defined __GNUC__ || (defined __STDC__ && __STDC__) | |
41 | # define PARAMS(args) args | |
42 | # else | |
43 | # define PARAMS(args) () | |
44 | # endif /* GCC. */ | |
45 | #endif /* Not PARAMS. */ | |
46 | ||
47 | #ifndef INSIDE_RECURSION | |
48 | ||
49 | # if defined STDC_HEADERS && !defined emacs | |
50 | # include <stddef.h> | |
51 | # else | |
52 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
53 | # include <sys/types.h> | |
54 | # endif | |
55 | ||
56 | # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC) | |
57 | ||
58 | /* For platform which support the ISO C amendement 1 functionality we | |
59 | support user defined character classes. */ | |
60 | # if defined _LIBC || WIDE_CHAR_SUPPORT | |
61 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ | |
62 | # include <wchar.h> | |
63 | # include <wctype.h> | |
64 | # endif | |
65 | ||
66 | # ifdef _LIBC | |
67 | /* We have to keep the namespace clean. */ | |
68 | # define regfree(preg) __regfree (preg) | |
69 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) | |
70 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) | |
71 | # define regerror(errcode, preg, errbuf, errbuf_size) \ | |
72 | __regerror(errcode, preg, errbuf, errbuf_size) | |
73 | # define re_set_registers(bu, re, nu, st, en) \ | |
74 | __re_set_registers (bu, re, nu, st, en) | |
75 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ | |
76 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
77 | # define re_match(bufp, string, size, pos, regs) \ | |
78 | __re_match (bufp, string, size, pos, regs) | |
79 | # define re_search(bufp, string, size, startpos, range, regs) \ | |
80 | __re_search (bufp, string, size, startpos, range, regs) | |
81 | # define re_compile_pattern(pattern, length, bufp) \ | |
82 | __re_compile_pattern (pattern, length, bufp) | |
83 | # define re_set_syntax(syntax) __re_set_syntax (syntax) | |
84 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ | |
85 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) | |
86 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) | |
87 | ||
88 | # define btowc __btowc | |
89 | ||
90 | /* We are also using some library internals. */ | |
91 | # include <locale/localeinfo.h> | |
92 | # include <locale/elem-hash.h> | |
93 | # include <langinfo.h> | |
94 | # include <locale/coll-lookup.h> | |
95 | # endif | |
96 | ||
97 | /* This is for other GNU distributions with internationalized messages. */ | |
98 | # if HAVE_LIBINTL_H || defined _LIBC | |
99 | # include <libintl.h> | |
100 | # ifdef _LIBC | |
101 | # undef gettext | |
102 | # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES) | |
103 | # endif | |
104 | # else | |
105 | # define gettext(msgid) (msgid) | |
106 | # endif | |
107 | ||
108 | # ifndef gettext_noop | |
109 | /* This define is so xgettext can find the internationalizable | |
110 | strings. */ | |
111 | # define gettext_noop(String) String | |
112 | # endif | |
113 | ||
114 | /* The `emacs' switch turns on certain matching commands | |
115 | that make sense only in Emacs. */ | |
116 | # ifdef emacs | |
117 | ||
118 | # include "lisp.h" | |
119 | # include "buffer.h" | |
120 | # include "syntax.h" | |
121 | ||
122 | # else /* not emacs */ | |
123 | ||
124 | /* If we are not linking with Emacs proper, | |
125 | we can't use the relocating allocator | |
126 | even if config.h says that we can. */ | |
127 | # undef REL_ALLOC | |
128 | ||
129 | # if defined STDC_HEADERS || defined _LIBC | |
130 | # include <stdlib.h> | |
131 | # else | |
132 | char *malloc (); | |
133 | char *realloc (); | |
134 | # endif | |
135 | ||
136 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. | |
137 | If nothing else has been done, use the method below. */ | |
138 | # ifdef INHIBIT_STRING_HEADER | |
139 | # if !(defined HAVE_BZERO && defined HAVE_BCOPY) | |
140 | # if !defined bzero && !defined bcopy | |
141 | # undef INHIBIT_STRING_HEADER | |
142 | # endif | |
143 | # endif | |
144 | # endif | |
145 | ||
146 | /* This is the normal way of making sure we have a bcopy and a bzero. | |
147 | This is used in most programs--a few other programs avoid this | |
148 | by defining INHIBIT_STRING_HEADER. */ | |
149 | # ifndef INHIBIT_STRING_HEADER | |
150 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC | |
151 | # include <string.h> | |
152 | # ifndef bzero | |
153 | # ifndef _LIBC | |
154 | # define bzero(s, n) (memset (s, '\0', n), (s)) | |
155 | # else | |
156 | # define bzero(s, n) __bzero (s, n) | |
157 | # endif | |
158 | # endif | |
159 | # else | |
160 | # include <strings.h> | |
161 | # ifndef memcmp | |
162 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) | |
163 | # endif | |
164 | # ifndef memcpy | |
165 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) | |
166 | # endif | |
167 | # endif | |
168 | # endif | |
169 | ||
170 | /* Define the syntax stuff for \<, \>, etc. */ | |
171 | ||
172 | /* This must be nonzero for the wordchar and notwordchar pattern | |
173 | commands in re_match_2. */ | |
174 | # ifndef Sword | |
175 | # define Sword 1 | |
176 | # endif | |
177 | ||
178 | # ifdef SWITCH_ENUM_BUG | |
179 | # define SWITCH_ENUM_CAST(x) ((int)(x)) | |
180 | # else | |
181 | # define SWITCH_ENUM_CAST(x) (x) | |
182 | # endif | |
183 | ||
184 | # endif /* not emacs */ | |
185 | ||
186 | # if defined _LIBC || HAVE_LIMITS_H | |
187 | # include <limits.h> | |
188 | # endif | |
189 | ||
190 | # ifndef MB_LEN_MAX | |
191 | # define MB_LEN_MAX 1 | |
192 | # endif | |
193 | \f | |
194 | /* Get the interface, including the syntax bits. */ | |
195 | # include "xregex.h" /* change for libiberty */ | |
196 | ||
197 | /* isalpha etc. are used for the character classes. */ | |
198 | # include <ctype.h> | |
199 | ||
200 | /* Jim Meyering writes: | |
201 | ||
202 | "... Some ctype macros are valid only for character codes that | |
203 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
204 | using /bin/cc or gcc but without giving an ansi option). So, all | |
205 | ctype uses should be through macros like ISPRINT... If | |
206 | STDC_HEADERS is defined, then autoconf has verified that the ctype | |
207 | macros don't need to be guarded with references to isascii. ... | |
208 | Defining isascii to 1 should let any compiler worth its salt | |
209 | eliminate the && through constant folding." | |
210 | Solaris defines some of these symbols so we must undefine them first. */ | |
211 | ||
212 | # undef ISASCII | |
213 | # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) | |
214 | # define ISASCII(c) 1 | |
215 | # else | |
216 | # define ISASCII(c) isascii(c) | |
217 | # endif | |
218 | ||
219 | # ifdef isblank | |
220 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
221 | # else | |
222 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
223 | # endif | |
224 | # ifdef isgraph | |
225 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
226 | # else | |
227 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
228 | # endif | |
229 | ||
230 | # undef ISPRINT | |
231 | # define ISPRINT(c) (ISASCII (c) && isprint (c)) | |
232 | # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
233 | # define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
234 | # define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
235 | # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
236 | # define ISLOWER(c) (ISASCII (c) && islower (c)) | |
237 | # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
238 | # define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
239 | # define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
240 | # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
241 | ||
242 | # ifdef _tolower | |
243 | # define TOLOWER(c) _tolower(c) | |
244 | # else | |
245 | # define TOLOWER(c) tolower(c) | |
246 | # endif | |
247 | ||
248 | # ifndef NULL | |
249 | # define NULL (void *)0 | |
250 | # endif | |
251 | ||
252 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
253 | since ours (we hope) works properly with all combinations of | |
254 | machines, compilers, `char' and `unsigned char' argument types. | |
255 | (Per Bothner suggested the basic approach.) */ | |
256 | # undef SIGN_EXTEND_CHAR | |
257 | # if __STDC__ | |
258 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
259 | # else /* not __STDC__ */ | |
260 | /* As in Harbison and Steele. */ | |
261 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
262 | # endif | |
263 | \f | |
264 | # ifndef emacs | |
265 | /* How many characters in the character set. */ | |
266 | # define CHAR_SET_SIZE 256 | |
267 | ||
268 | # ifdef SYNTAX_TABLE | |
269 | ||
270 | extern char *re_syntax_table; | |
271 | ||
272 | # else /* not SYNTAX_TABLE */ | |
273 | ||
274 | static char re_syntax_table[CHAR_SET_SIZE]; | |
275 | ||
276 | static void init_syntax_once PARAMS ((void)); | |
277 | ||
278 | static void | |
279 | init_syntax_once () | |
280 | { | |
281 | register int c; | |
282 | static int done = 0; | |
283 | ||
284 | if (done) | |
285 | return; | |
286 | bzero (re_syntax_table, sizeof re_syntax_table); | |
287 | ||
288 | for (c = 0; c < CHAR_SET_SIZE; ++c) | |
289 | if (ISALNUM (c)) | |
290 | re_syntax_table[c] = Sword; | |
291 | ||
292 | re_syntax_table['_'] = Sword; | |
293 | ||
294 | done = 1; | |
295 | } | |
296 | ||
297 | # endif /* not SYNTAX_TABLE */ | |
298 | ||
299 | # define SYNTAX(c) re_syntax_table[(unsigned char) (c)] | |
300 | ||
301 | # endif /* emacs */ | |
302 | \f | |
303 | /* Integer type for pointers. */ | |
304 | # if !defined _LIBC | |
305 | typedef unsigned long int uintptr_t; | |
306 | # endif | |
307 | ||
308 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
309 | use `alloca' instead of `malloc'. This is because using malloc in | |
310 | re_search* or re_match* could cause memory leaks when C-g is used in | |
311 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
312 | the other hand, malloc is more portable, and easier to debug. | |
313 | ||
314 | Because we sometimes use alloca, some routines have to be macros, | |
315 | not functions -- `alloca'-allocated space disappears at the end of the | |
316 | function it is called in. */ | |
317 | ||
318 | # ifdef REGEX_MALLOC | |
319 | ||
320 | # define REGEX_ALLOCATE malloc | |
321 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
322 | # define REGEX_FREE free | |
323 | ||
324 | # else /* not REGEX_MALLOC */ | |
325 | ||
326 | /* Emacs already defines alloca, sometimes. */ | |
327 | # ifndef alloca | |
328 | ||
329 | /* Make alloca work the best possible way. */ | |
330 | # ifdef __GNUC__ | |
331 | # define alloca __builtin_alloca | |
332 | # else /* not __GNUC__ */ | |
333 | # if HAVE_ALLOCA_H | |
334 | # include <alloca.h> | |
335 | # endif /* HAVE_ALLOCA_H */ | |
336 | # endif /* not __GNUC__ */ | |
337 | ||
338 | # endif /* not alloca */ | |
339 | ||
340 | # define REGEX_ALLOCATE alloca | |
341 | ||
342 | /* Assumes a `char *destination' variable. */ | |
343 | # define REGEX_REALLOCATE(source, osize, nsize) \ | |
344 | (destination = (char *) alloca (nsize), \ | |
345 | memcpy (destination, source, osize)) | |
346 | ||
347 | /* No need to do anything to free, after alloca. */ | |
348 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ | |
349 | ||
350 | # endif /* not REGEX_MALLOC */ | |
351 | ||
352 | /* Define how to allocate the failure stack. */ | |
353 | ||
354 | # if defined REL_ALLOC && defined REGEX_MALLOC | |
355 | ||
356 | # define REGEX_ALLOCATE_STACK(size) \ | |
357 | r_alloc (&failure_stack_ptr, (size)) | |
358 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ | |
359 | r_re_alloc (&failure_stack_ptr, (nsize)) | |
360 | # define REGEX_FREE_STACK(ptr) \ | |
361 | r_alloc_free (&failure_stack_ptr) | |
362 | ||
363 | # else /* not using relocating allocator */ | |
364 | ||
365 | # ifdef REGEX_MALLOC | |
366 | ||
367 | # define REGEX_ALLOCATE_STACK malloc | |
368 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
369 | # define REGEX_FREE_STACK free | |
370 | ||
371 | # else /* not REGEX_MALLOC */ | |
372 | ||
373 | # define REGEX_ALLOCATE_STACK alloca | |
374 | ||
375 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ | |
376 | REGEX_REALLOCATE (source, osize, nsize) | |
377 | /* No need to explicitly free anything. */ | |
378 | # define REGEX_FREE_STACK(arg) | |
379 | ||
380 | # endif /* not REGEX_MALLOC */ | |
381 | # endif /* not using relocating allocator */ | |
382 | ||
383 | ||
384 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
385 | `string1' or just past its end. This works if PTR is NULL, which is | |
386 | a good thing. */ | |
387 | # define FIRST_STRING_P(ptr) \ | |
388 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
389 | ||
390 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
391 | # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
392 | # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
393 | # define RETALLOC_IF(addr, n, t) \ | |
394 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
395 | # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
396 | ||
397 | # define BYTEWIDTH 8 /* In bits. */ | |
398 | ||
399 | # define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
400 | ||
401 | # undef MAX | |
402 | # undef MIN | |
403 | # define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
404 | # define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
405 | ||
406 | typedef char boolean; | |
407 | # define false 0 | |
408 | # define true 1 | |
409 | ||
410 | static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size, | |
411 | reg_syntax_t syntax, | |
412 | struct re_pattern_buffer *bufp)); | |
2a6ef469 DD |
413 | |
414 | static int byte_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp, | |
415 | const char *string1, int size1, | |
416 | const char *string2, int size2, | |
417 | int pos, | |
418 | struct re_registers *regs, | |
419 | int stop)); | |
86710ce2 DD |
420 | static int byte_re_search_2 PARAMS ((struct re_pattern_buffer *bufp, |
421 | const char *string1, int size1, | |
422 | const char *string2, int size2, | |
423 | int startpos, int range, | |
424 | struct re_registers *regs, int stop)); | |
425 | static int byte_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp)); | |
426 | ||
427 | #ifdef MBS_SUPPORT | |
428 | static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size, | |
429 | reg_syntax_t syntax, | |
430 | struct re_pattern_buffer *bufp)); | |
431 | ||
432 | ||
2a6ef469 DD |
433 | static int wcs_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp, |
434 | const char *cstring1, int csize1, | |
435 | const char *cstring2, int csize2, | |
436 | int pos, | |
437 | struct re_registers *regs, | |
438 | int stop, | |
439 | wchar_t *string1, int size1, | |
440 | wchar_t *string2, int size2, | |
441 | int *mbs_offset1, int *mbs_offset2)); | |
2a6ef469 DD |
442 | static int wcs_re_search_2 PARAMS ((struct re_pattern_buffer *bufp, |
443 | const char *string1, int size1, | |
444 | const char *string2, int size2, | |
445 | int startpos, int range, | |
446 | struct re_registers *regs, int stop)); | |
2a6ef469 | 447 | static int wcs_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp)); |
86710ce2 | 448 | #endif |
2a6ef469 DD |
449 | \f |
450 | /* These are the command codes that appear in compiled regular | |
451 | expressions. Some opcodes are followed by argument bytes. A | |
452 | command code can specify any interpretation whatsoever for its | |
453 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
454 | ||
455 | typedef enum | |
456 | { | |
457 | no_op = 0, | |
458 | ||
459 | /* Succeed right away--no more backtracking. */ | |
460 | succeed, | |
461 | ||
462 | /* Followed by one byte giving n, then by n literal bytes. */ | |
463 | exactn, | |
464 | ||
465 | # ifdef MBS_SUPPORT | |
466 | /* Same as exactn, but contains binary data. */ | |
467 | exactn_bin, | |
468 | # endif | |
469 | ||
470 | /* Matches any (more or less) character. */ | |
471 | anychar, | |
472 | ||
473 | /* Matches any one char belonging to specified set. First | |
474 | following byte is number of bitmap bytes. Then come bytes | |
475 | for a bitmap saying which chars are in. Bits in each byte | |
476 | are ordered low-bit-first. A character is in the set if its | |
477 | bit is 1. A character too large to have a bit in the map is | |
478 | automatically not in the set. */ | |
479 | /* ifdef MBS_SUPPORT, following element is length of character | |
480 | classes, length of collating symbols, length of equivalence | |
481 | classes, length of character ranges, and length of characters. | |
482 | Next, character class element, collating symbols elements, | |
483 | equivalence class elements, range elements, and character | |
484 | elements follow. | |
485 | See regex_compile function. */ | |
486 | charset, | |
487 | ||
488 | /* Same parameters as charset, but match any character that is | |
489 | not one of those specified. */ | |
490 | charset_not, | |
491 | ||
492 | /* Start remembering the text that is matched, for storing in a | |
493 | register. Followed by one byte with the register number, in | |
494 | the range 0 to one less than the pattern buffer's re_nsub | |
495 | field. Then followed by one byte with the number of groups | |
496 | inner to this one. (This last has to be part of the | |
497 | start_memory only because we need it in the on_failure_jump | |
498 | of re_match_2.) */ | |
499 | start_memory, | |
500 | ||
501 | /* Stop remembering the text that is matched and store it in a | |
502 | memory register. Followed by one byte with the register | |
503 | number, in the range 0 to one less than `re_nsub' in the | |
504 | pattern buffer, and one byte with the number of inner groups, | |
505 | just like `start_memory'. (We need the number of inner | |
506 | groups here because we don't have any easy way of finding the | |
507 | corresponding start_memory when we're at a stop_memory.) */ | |
508 | stop_memory, | |
509 | ||
510 | /* Match a duplicate of something remembered. Followed by one | |
511 | byte containing the register number. */ | |
512 | duplicate, | |
513 | ||
514 | /* Fail unless at beginning of line. */ | |
515 | begline, | |
516 | ||
517 | /* Fail unless at end of line. */ | |
518 | endline, | |
519 | ||
520 | /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
521 | of string to be matched (if not). */ | |
522 | begbuf, | |
523 | ||
524 | /* Analogously, for end of buffer/string. */ | |
525 | endbuf, | |
526 | ||
527 | /* Followed by two byte relative address to which to jump. */ | |
528 | jump, | |
529 | ||
530 | /* Same as jump, but marks the end of an alternative. */ | |
531 | jump_past_alt, | |
532 | ||
533 | /* Followed by two-byte relative address of place to resume at | |
534 | in case of failure. */ | |
535 | /* ifdef MBS_SUPPORT, the size of address is 1. */ | |
536 | on_failure_jump, | |
537 | ||
538 | /* Like on_failure_jump, but pushes a placeholder instead of the | |
539 | current string position when executed. */ | |
540 | on_failure_keep_string_jump, | |
541 | ||
542 | /* Throw away latest failure point and then jump to following | |
543 | two-byte relative address. */ | |
544 | /* ifdef MBS_SUPPORT, the size of address is 1. */ | |
545 | pop_failure_jump, | |
546 | ||
547 | /* Change to pop_failure_jump if know won't have to backtrack to | |
548 | match; otherwise change to jump. This is used to jump | |
549 | back to the beginning of a repeat. If what follows this jump | |
550 | clearly won't match what the repeat does, such that we can be | |
551 | sure that there is no use backtracking out of repetitions | |
552 | already matched, then we change it to a pop_failure_jump. | |
553 | Followed by two-byte address. */ | |
554 | /* ifdef MBS_SUPPORT, the size of address is 1. */ | |
555 | maybe_pop_jump, | |
556 | ||
557 | /* Jump to following two-byte address, and push a dummy failure | |
558 | point. This failure point will be thrown away if an attempt | |
559 | is made to use it for a failure. A `+' construct makes this | |
560 | before the first repeat. Also used as an intermediary kind | |
561 | of jump when compiling an alternative. */ | |
562 | /* ifdef MBS_SUPPORT, the size of address is 1. */ | |
563 | dummy_failure_jump, | |
564 | ||
565 | /* Push a dummy failure point and continue. Used at the end of | |
566 | alternatives. */ | |
567 | push_dummy_failure, | |
568 | ||
569 | /* Followed by two-byte relative address and two-byte number n. | |
570 | After matching N times, jump to the address upon failure. */ | |
571 | /* ifdef MBS_SUPPORT, the size of address is 1. */ | |
572 | succeed_n, | |
573 | ||
574 | /* Followed by two-byte relative address, and two-byte number n. | |
575 | Jump to the address N times, then fail. */ | |
576 | /* ifdef MBS_SUPPORT, the size of address is 1. */ | |
577 | jump_n, | |
578 | ||
579 | /* Set the following two-byte relative address to the | |
580 | subsequent two-byte number. The address *includes* the two | |
581 | bytes of number. */ | |
582 | /* ifdef MBS_SUPPORT, the size of address is 1. */ | |
583 | set_number_at, | |
584 | ||
585 | wordchar, /* Matches any word-constituent character. */ | |
586 | notwordchar, /* Matches any char that is not a word-constituent. */ | |
587 | ||
588 | wordbeg, /* Succeeds if at word beginning. */ | |
589 | wordend, /* Succeeds if at word end. */ | |
590 | ||
591 | wordbound, /* Succeeds if at a word boundary. */ | |
592 | notwordbound /* Succeeds if not at a word boundary. */ | |
593 | ||
594 | # ifdef emacs | |
595 | ,before_dot, /* Succeeds if before point. */ | |
596 | at_dot, /* Succeeds if at point. */ | |
597 | after_dot, /* Succeeds if after point. */ | |
598 | ||
599 | /* Matches any character whose syntax is specified. Followed by | |
600 | a byte which contains a syntax code, e.g., Sword. */ | |
601 | syntaxspec, | |
602 | ||
603 | /* Matches any character whose syntax is not that specified. */ | |
604 | notsyntaxspec | |
605 | # endif /* emacs */ | |
606 | } re_opcode_t; | |
607 | #endif /* not INSIDE_RECURSION */ | |
608 | \f | |
609 | ||
610 | #ifdef BYTE | |
611 | # define CHAR_T char | |
612 | # define UCHAR_T unsigned char | |
613 | # define COMPILED_BUFFER_VAR bufp->buffer | |
614 | # define OFFSET_ADDRESS_SIZE 2 | |
615 | # define PREFIX(name) byte_##name | |
616 | # define ARG_PREFIX(name) name | |
617 | # define PUT_CHAR(c) putchar (c) | |
2a6ef469 | 618 | #else |
86710ce2 DD |
619 | # ifdef WCHAR |
620 | # define CHAR_T wchar_t | |
621 | # define UCHAR_T wchar_t | |
622 | # define COMPILED_BUFFER_VAR wc_buffer | |
623 | # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */ | |
624 | # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1) | |
625 | # define PREFIX(name) wcs_##name | |
626 | # define ARG_PREFIX(name) c##name | |
627 | /* Should we use wide stream?? */ | |
628 | # define PUT_CHAR(c) printf ("%C", c); | |
629 | # define TRUE 1 | |
630 | # define FALSE 0 | |
631 | # else | |
632 | # ifdef MBS_SUPPORT | |
633 | # define WCHAR | |
634 | # define INSIDE_RECURSION | |
635 | # include "regex.c" | |
636 | # undef INSIDE_RECURSION | |
637 | # endif | |
638 | # define BYTE | |
2a6ef469 DD |
639 | # define INSIDE_RECURSION |
640 | # include "regex.c" | |
641 | # undef INSIDE_RECURSION | |
642 | # endif | |
2a6ef469 DD |
643 | #endif |
644 | ||
645 | #ifdef INSIDE_RECURSION | |
646 | /* Common operations on the compiled pattern. */ | |
647 | ||
648 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
649 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ | |
650 | ||
651 | # ifdef WCHAR | |
652 | # define STORE_NUMBER(destination, number) \ | |
653 | do { \ | |
654 | *(destination) = (UCHAR_T)(number); \ | |
655 | } while (0) | |
656 | # else /* BYTE */ | |
657 | # define STORE_NUMBER(destination, number) \ | |
658 | do { \ | |
659 | (destination)[0] = (number) & 0377; \ | |
660 | (destination)[1] = (number) >> 8; \ | |
661 | } while (0) | |
662 | # endif /* WCHAR */ | |
663 | ||
664 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
665 | the byte after where the number is stored. Therefore, DESTINATION | |
666 | must be an lvalue. */ | |
667 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ | |
668 | ||
669 | # define STORE_NUMBER_AND_INCR(destination, number) \ | |
670 | do { \ | |
671 | STORE_NUMBER (destination, number); \ | |
672 | (destination) += OFFSET_ADDRESS_SIZE; \ | |
673 | } while (0) | |
674 | ||
675 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
676 | at SOURCE. */ | |
677 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ | |
678 | ||
679 | # ifdef WCHAR | |
680 | # define EXTRACT_NUMBER(destination, source) \ | |
681 | do { \ | |
682 | (destination) = *(source); \ | |
683 | } while (0) | |
684 | # else /* BYTE */ | |
685 | # define EXTRACT_NUMBER(destination, source) \ | |
686 | do { \ | |
687 | (destination) = *(source) & 0377; \ | |
688 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
689 | } while (0) | |
690 | # endif | |
691 | ||
692 | # ifdef DEBUG | |
693 | static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source)); | |
694 | static void | |
695 | PREFIX(extract_number) (dest, source) | |
696 | int *dest; | |
697 | UCHAR_T *source; | |
698 | { | |
699 | # ifdef WCHAR | |
700 | *dest = *source; | |
701 | # else /* BYTE */ | |
702 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
703 | *dest = *source & 0377; | |
704 | *dest += temp << 8; | |
705 | # endif | |
706 | } | |
707 | ||
708 | # ifndef EXTRACT_MACROS /* To debug the macros. */ | |
709 | # undef EXTRACT_NUMBER | |
710 | # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src) | |
711 | # endif /* not EXTRACT_MACROS */ | |
712 | ||
713 | # endif /* DEBUG */ | |
714 | ||
715 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
716 | SOURCE must be an lvalue. */ | |
717 | ||
718 | # define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
719 | do { \ | |
720 | EXTRACT_NUMBER (destination, source); \ | |
721 | (source) += OFFSET_ADDRESS_SIZE; \ | |
722 | } while (0) | |
723 | ||
724 | # ifdef DEBUG | |
725 | static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination, | |
726 | UCHAR_T **source)); | |
727 | static void | |
728 | PREFIX(extract_number_and_incr) (destination, source) | |
729 | int *destination; | |
730 | UCHAR_T **source; | |
731 | { | |
732 | PREFIX(extract_number) (destination, *source); | |
733 | *source += OFFSET_ADDRESS_SIZE; | |
734 | } | |
735 | ||
736 | # ifndef EXTRACT_MACROS | |
737 | # undef EXTRACT_NUMBER_AND_INCR | |
738 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
739 | PREFIX(extract_number_and_incr) (&dest, &src) | |
740 | # endif /* not EXTRACT_MACROS */ | |
741 | ||
742 | # endif /* DEBUG */ | |
743 | ||
744 | \f | |
745 | ||
746 | /* If DEBUG is defined, Regex prints many voluminous messages about what | |
747 | it is doing (if the variable `debug' is nonzero). If linked with the | |
748 | main program in `iregex.c', you can enter patterns and strings | |
749 | interactively. And if linked with the main program in `main.c' and | |
750 | the other test files, you can run the already-written tests. */ | |
751 | ||
752 | # ifdef DEBUG | |
753 | ||
754 | # ifndef DEFINED_ONCE | |
755 | ||
756 | /* We use standard I/O for debugging. */ | |
757 | # include <stdio.h> | |
758 | ||
759 | /* It is useful to test things that ``must'' be true when debugging. */ | |
760 | # include <assert.h> | |
761 | ||
762 | static int debug; | |
763 | ||
764 | # define DEBUG_STATEMENT(e) e | |
765 | # define DEBUG_PRINT1(x) if (debug) printf (x) | |
766 | # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
767 | # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
768 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) | |
769 | # endif /* not DEFINED_ONCE */ | |
770 | ||
771 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
772 | if (debug) PREFIX(print_partial_compiled_pattern) (s, e) | |
773 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
774 | if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2) | |
775 | ||
776 | ||
777 | /* Print the fastmap in human-readable form. */ | |
778 | ||
779 | # ifndef DEFINED_ONCE | |
780 | void | |
781 | print_fastmap (fastmap) | |
782 | char *fastmap; | |
783 | { | |
784 | unsigned was_a_range = 0; | |
785 | unsigned i = 0; | |
786 | ||
787 | while (i < (1 << BYTEWIDTH)) | |
788 | { | |
789 | if (fastmap[i++]) | |
790 | { | |
791 | was_a_range = 0; | |
792 | putchar (i - 1); | |
793 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
794 | { | |
795 | was_a_range = 1; | |
796 | i++; | |
797 | } | |
798 | if (was_a_range) | |
799 | { | |
800 | printf ("-"); | |
801 | putchar (i - 1); | |
802 | } | |
803 | } | |
804 | } | |
805 | putchar ('\n'); | |
806 | } | |
807 | # endif /* not DEFINED_ONCE */ | |
808 | ||
809 | ||
810 | /* Print a compiled pattern string in human-readable form, starting at | |
811 | the START pointer into it and ending just before the pointer END. */ | |
812 | ||
813 | void | |
814 | PREFIX(print_partial_compiled_pattern) (start, end) | |
815 | UCHAR_T *start; | |
816 | UCHAR_T *end; | |
817 | { | |
818 | int mcnt, mcnt2; | |
819 | UCHAR_T *p1; | |
820 | UCHAR_T *p = start; | |
821 | UCHAR_T *pend = end; | |
822 | ||
823 | if (start == NULL) | |
824 | { | |
825 | printf ("(null)\n"); | |
826 | return; | |
827 | } | |
828 | ||
829 | /* Loop over pattern commands. */ | |
830 | while (p < pend) | |
831 | { | |
832 | # ifdef _LIBC | |
833 | printf ("%td:\t", p - start); | |
834 | # else | |
835 | printf ("%ld:\t", (long int) (p - start)); | |
836 | # endif | |
837 | ||
838 | switch ((re_opcode_t) *p++) | |
839 | { | |
840 | case no_op: | |
841 | printf ("/no_op"); | |
842 | break; | |
843 | ||
844 | case exactn: | |
845 | mcnt = *p++; | |
846 | printf ("/exactn/%d", mcnt); | |
847 | do | |
848 | { | |
849 | putchar ('/'); | |
850 | PUT_CHAR (*p++); | |
851 | } | |
852 | while (--mcnt); | |
853 | break; | |
854 | ||
855 | # ifdef MBS_SUPPORT | |
856 | case exactn_bin: | |
857 | mcnt = *p++; | |
858 | printf ("/exactn_bin/%d", mcnt); | |
859 | do | |
860 | { | |
861 | printf("/%lx", (long int) *p++); | |
862 | } | |
863 | while (--mcnt); | |
864 | break; | |
865 | # endif /* MBS_SUPPORT */ | |
866 | ||
867 | case start_memory: | |
868 | mcnt = *p++; | |
869 | printf ("/start_memory/%d/%ld", mcnt, (long int) *p++); | |
870 | break; | |
871 | ||
872 | case stop_memory: | |
873 | mcnt = *p++; | |
874 | printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++); | |
875 | break; | |
876 | ||
877 | case duplicate: | |
878 | printf ("/duplicate/%ld", (long int) *p++); | |
879 | break; | |
880 | ||
881 | case anychar: | |
882 | printf ("/anychar"); | |
883 | break; | |
884 | ||
885 | case charset: | |
886 | case charset_not: | |
887 | { | |
888 | # ifdef WCHAR | |
889 | int i, length; | |
890 | wchar_t *workp = p; | |
891 | printf ("/charset [%s", | |
892 | (re_opcode_t) *(workp - 1) == charset_not ? "^" : ""); | |
893 | p += 5; | |
894 | length = *workp++; /* the length of char_classes */ | |
895 | for (i=0 ; i<length ; i++) | |
896 | printf("[:%lx:]", (long int) *p++); | |
897 | length = *workp++; /* the length of collating_symbol */ | |
898 | for (i=0 ; i<length ;) | |
899 | { | |
900 | printf("[."); | |
901 | while(*p != 0) | |
902 | PUT_CHAR((i++,*p++)); | |
903 | i++,p++; | |
904 | printf(".]"); | |
905 | } | |
906 | length = *workp++; /* the length of equivalence_class */ | |
907 | for (i=0 ; i<length ;) | |
908 | { | |
909 | printf("[="); | |
910 | while(*p != 0) | |
911 | PUT_CHAR((i++,*p++)); | |
912 | i++,p++; | |
913 | printf("=]"); | |
914 | } | |
915 | length = *workp++; /* the length of char_range */ | |
916 | for (i=0 ; i<length ; i++) | |
917 | { | |
918 | wchar_t range_start = *p++; | |
919 | wchar_t range_end = *p++; | |
920 | printf("%C-%C", range_start, range_end); | |
921 | } | |
922 | length = *workp++; /* the length of char */ | |
923 | for (i=0 ; i<length ; i++) | |
924 | printf("%C", *p++); | |
925 | putchar (']'); | |
926 | # else | |
927 | register int c, last = -100; | |
928 | register int in_range = 0; | |
929 | ||
930 | printf ("/charset [%s", | |
931 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); | |
932 | ||
933 | assert (p + *p < pend); | |
934 | ||
935 | for (c = 0; c < 256; c++) | |
936 | if (c / 8 < *p | |
937 | && (p[1 + (c/8)] & (1 << (c % 8)))) | |
938 | { | |
939 | /* Are we starting a range? */ | |
940 | if (last + 1 == c && ! in_range) | |
941 | { | |
942 | putchar ('-'); | |
943 | in_range = 1; | |
944 | } | |
945 | /* Have we broken a range? */ | |
946 | else if (last + 1 != c && in_range) | |
947 | { | |
948 | putchar (last); | |
949 | in_range = 0; | |
950 | } | |
951 | ||
952 | if (! in_range) | |
953 | putchar (c); | |
954 | ||
955 | last = c; | |
956 | } | |
957 | ||
958 | if (in_range) | |
959 | putchar (last); | |
960 | ||
961 | putchar (']'); | |
962 | ||
963 | p += 1 + *p; | |
964 | # endif /* WCHAR */ | |
965 | } | |
966 | break; | |
967 | ||
968 | case begline: | |
969 | printf ("/begline"); | |
970 | break; | |
971 | ||
972 | case endline: | |
973 | printf ("/endline"); | |
974 | break; | |
975 | ||
976 | case on_failure_jump: | |
977 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
978 | # ifdef _LIBC | |
979 | printf ("/on_failure_jump to %td", p + mcnt - start); | |
980 | # else | |
981 | printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start)); | |
982 | # endif | |
983 | break; | |
984 | ||
985 | case on_failure_keep_string_jump: | |
986 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
987 | # ifdef _LIBC | |
988 | printf ("/on_failure_keep_string_jump to %td", p + mcnt - start); | |
989 | # else | |
990 | printf ("/on_failure_keep_string_jump to %ld", | |
991 | (long int) (p + mcnt - start)); | |
992 | # endif | |
993 | break; | |
994 | ||
995 | case dummy_failure_jump: | |
996 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
997 | # ifdef _LIBC | |
998 | printf ("/dummy_failure_jump to %td", p + mcnt - start); | |
999 | # else | |
1000 | printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start)); | |
1001 | # endif | |
1002 | break; | |
1003 | ||
1004 | case push_dummy_failure: | |
1005 | printf ("/push_dummy_failure"); | |
1006 | break; | |
1007 | ||
1008 | case maybe_pop_jump: | |
1009 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
1010 | # ifdef _LIBC | |
1011 | printf ("/maybe_pop_jump to %td", p + mcnt - start); | |
1012 | # else | |
1013 | printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start)); | |
1014 | # endif | |
1015 | break; | |
1016 | ||
1017 | case pop_failure_jump: | |
1018 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
1019 | # ifdef _LIBC | |
1020 | printf ("/pop_failure_jump to %td", p + mcnt - start); | |
1021 | # else | |
1022 | printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start)); | |
1023 | # endif | |
1024 | break; | |
1025 | ||
1026 | case jump_past_alt: | |
1027 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
1028 | # ifdef _LIBC | |
1029 | printf ("/jump_past_alt to %td", p + mcnt - start); | |
1030 | # else | |
1031 | printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start)); | |
1032 | # endif | |
1033 | break; | |
1034 | ||
1035 | case jump: | |
1036 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
1037 | # ifdef _LIBC | |
1038 | printf ("/jump to %td", p + mcnt - start); | |
1039 | # else | |
1040 | printf ("/jump to %ld", (long int) (p + mcnt - start)); | |
1041 | # endif | |
1042 | break; | |
1043 | ||
1044 | case succeed_n: | |
1045 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
1046 | p1 = p + mcnt; | |
1047 | PREFIX(extract_number_and_incr) (&mcnt2, &p); | |
1048 | # ifdef _LIBC | |
1049 | printf ("/succeed_n to %td, %d times", p1 - start, mcnt2); | |
1050 | # else | |
1051 | printf ("/succeed_n to %ld, %d times", | |
1052 | (long int) (p1 - start), mcnt2); | |
1053 | # endif | |
1054 | break; | |
1055 | ||
1056 | case jump_n: | |
1057 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
1058 | p1 = p + mcnt; | |
1059 | PREFIX(extract_number_and_incr) (&mcnt2, &p); | |
1060 | printf ("/jump_n to %d, %d times", p1 - start, mcnt2); | |
1061 | break; | |
1062 | ||
1063 | case set_number_at: | |
1064 | PREFIX(extract_number_and_incr) (&mcnt, &p); | |
1065 | p1 = p + mcnt; | |
1066 | PREFIX(extract_number_and_incr) (&mcnt2, &p); | |
1067 | # ifdef _LIBC | |
1068 | printf ("/set_number_at location %td to %d", p1 - start, mcnt2); | |
1069 | # else | |
1070 | printf ("/set_number_at location %ld to %d", | |
1071 | (long int) (p1 - start), mcnt2); | |
1072 | # endif | |
1073 | break; | |
1074 | ||
1075 | case wordbound: | |
1076 | printf ("/wordbound"); | |
1077 | break; | |
1078 | ||
1079 | case notwordbound: | |
1080 | printf ("/notwordbound"); | |
1081 | break; | |
1082 | ||
1083 | case wordbeg: | |
1084 | printf ("/wordbeg"); | |
1085 | break; | |
1086 | ||
1087 | case wordend: | |
1088 | printf ("/wordend"); | |
1089 | break; | |
1090 | ||
1091 | # ifdef emacs | |
1092 | case before_dot: | |
1093 | printf ("/before_dot"); | |
1094 | break; | |
1095 | ||
1096 | case at_dot: | |
1097 | printf ("/at_dot"); | |
1098 | break; | |
1099 | ||
1100 | case after_dot: | |
1101 | printf ("/after_dot"); | |
1102 | break; | |
1103 | ||
1104 | case syntaxspec: | |
1105 | printf ("/syntaxspec"); | |
1106 | mcnt = *p++; | |
1107 | printf ("/%d", mcnt); | |
1108 | break; | |
1109 | ||
1110 | case notsyntaxspec: | |
1111 | printf ("/notsyntaxspec"); | |
1112 | mcnt = *p++; | |
1113 | printf ("/%d", mcnt); | |
1114 | break; | |
1115 | # endif /* emacs */ | |
1116 | ||
1117 | case wordchar: | |
1118 | printf ("/wordchar"); | |
1119 | break; | |
1120 | ||
1121 | case notwordchar: | |
1122 | printf ("/notwordchar"); | |
1123 | break; | |
1124 | ||
1125 | case begbuf: | |
1126 | printf ("/begbuf"); | |
1127 | break; | |
1128 | ||
1129 | case endbuf: | |
1130 | printf ("/endbuf"); | |
1131 | break; | |
1132 | ||
1133 | default: | |
1134 | printf ("?%ld", (long int) *(p-1)); | |
1135 | } | |
1136 | ||
1137 | putchar ('\n'); | |
1138 | } | |
1139 | ||
1140 | # ifdef _LIBC | |
1141 | printf ("%td:\tend of pattern.\n", p - start); | |
1142 | # else | |
1143 | printf ("%ld:\tend of pattern.\n", (long int) (p - start)); | |
1144 | # endif | |
1145 | } | |
1146 | ||
1147 | ||
1148 | void | |
1149 | PREFIX(print_compiled_pattern) (bufp) | |
1150 | struct re_pattern_buffer *bufp; | |
1151 | { | |
1152 | UCHAR_T *buffer = (UCHAR_T*) bufp->buffer; | |
1153 | ||
1154 | PREFIX(print_partial_compiled_pattern) (buffer, buffer | |
1155 | + bufp->used / sizeof(UCHAR_T)); | |
1156 | printf ("%ld bytes used/%ld bytes allocated.\n", | |
1157 | bufp->used, bufp->allocated); | |
1158 | ||
1159 | if (bufp->fastmap_accurate && bufp->fastmap) | |
1160 | { | |
1161 | printf ("fastmap: "); | |
1162 | print_fastmap (bufp->fastmap); | |
1163 | } | |
1164 | ||
1165 | # ifdef _LIBC | |
1166 | printf ("re_nsub: %Zd\t", bufp->re_nsub); | |
1167 | # else | |
1168 | printf ("re_nsub: %ld\t", (long int) bufp->re_nsub); | |
1169 | # endif | |
1170 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
1171 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
1172 | printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
1173 | printf ("no_sub: %d\t", bufp->no_sub); | |
1174 | printf ("not_bol: %d\t", bufp->not_bol); | |
1175 | printf ("not_eol: %d\t", bufp->not_eol); | |
1176 | printf ("syntax: %lx\n", bufp->syntax); | |
1177 | /* Perhaps we should print the translate table? */ | |
1178 | } | |
1179 | ||
1180 | ||
1181 | void | |
1182 | PREFIX(print_double_string) (where, string1, size1, string2, size2) | |
1183 | const CHAR_T *where; | |
1184 | const CHAR_T *string1; | |
1185 | const CHAR_T *string2; | |
1186 | int size1; | |
1187 | int size2; | |
1188 | { | |
1189 | int this_char; | |
1190 | ||
1191 | if (where == NULL) | |
1192 | printf ("(null)"); | |
1193 | else | |
1194 | { | |
6ad8a379 DD |
1195 | int cnt; |
1196 | ||
2a6ef469 DD |
1197 | if (FIRST_STRING_P (where)) |
1198 | { | |
1199 | for (this_char = where - string1; this_char < size1; this_char++) | |
1200 | PUT_CHAR (string1[this_char]); | |
1201 | ||
1202 | where = string2; | |
1203 | } | |
1204 | ||
6ad8a379 | 1205 | cnt = 0; |
2a6ef469 | 1206 | for (this_char = where - string2; this_char < size2; this_char++) |
6ad8a379 DD |
1207 | { |
1208 | PUT_CHAR (string2[this_char]); | |
1209 | if (++cnt > 100) | |
1210 | { | |
1211 | fputs ("...", stdout); | |
1212 | break; | |
1213 | } | |
1214 | } | |
2a6ef469 DD |
1215 | } |
1216 | } | |
1217 | ||
1218 | # ifndef DEFINED_ONCE | |
1219 | void | |
1220 | printchar (c) | |
1221 | int c; | |
1222 | { | |
1223 | putc (c, stderr); | |
1224 | } | |
1225 | # endif | |
1226 | ||
1227 | # else /* not DEBUG */ | |
1228 | ||
1229 | # ifndef DEFINED_ONCE | |
1230 | # undef assert | |
1231 | # define assert(e) | |
1232 | ||
1233 | # define DEBUG_STATEMENT(e) | |
1234 | # define DEBUG_PRINT1(x) | |
1235 | # define DEBUG_PRINT2(x1, x2) | |
1236 | # define DEBUG_PRINT3(x1, x2, x3) | |
1237 | # define DEBUG_PRINT4(x1, x2, x3, x4) | |
1238 | # endif /* not DEFINED_ONCE */ | |
1239 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1240 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
1241 | ||
1242 | # endif /* not DEBUG */ | |
1243 | ||
1244 | \f | |
1245 | ||
1246 | # ifdef WCHAR | |
1247 | /* This convert a multibyte string to a wide character string. | |
1248 | And write their correspondances to offset_buffer(see below) | |
1249 | and write whether each wchar_t is binary data to is_binary. | |
1250 | This assume invalid multibyte sequences as binary data. | |
1251 | We assume offset_buffer and is_binary is already allocated | |
1252 | enough space. */ | |
1253 | ||
1254 | static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src, | |
1255 | size_t len, int *offset_buffer, | |
1256 | char *is_binary); | |
1257 | static size_t | |
1258 | convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary) | |
1259 | CHAR_T *dest; | |
1260 | const unsigned char* src; | |
1261 | size_t len; /* the length of multibyte string. */ | |
1262 | ||
1263 | /* It hold correspondances between src(char string) and | |
1264 | dest(wchar_t string) for optimization. | |
1265 | e.g. src = "xxxyzz" | |
1266 | dest = {'X', 'Y', 'Z'} | |
1267 | (each "xxx", "y" and "zz" represent one multibyte character | |
1268 | corresponding to 'X', 'Y' and 'Z'.) | |
1269 | offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")} | |
1270 | = {0, 3, 4, 6} | |
1271 | */ | |
1272 | int *offset_buffer; | |
1273 | char *is_binary; | |
1274 | { | |
1275 | wchar_t *pdest = dest; | |
1276 | const unsigned char *psrc = src; | |
1277 | size_t wc_count = 0; | |
1278 | ||
1279 | mbstate_t mbs; | |
1280 | int i, consumed; | |
1281 | size_t mb_remain = len; | |
1282 | size_t mb_count = 0; | |
1283 | ||
1284 | /* Initialize the conversion state. */ | |
1285 | memset (&mbs, 0, sizeof (mbstate_t)); | |
1286 | ||
1287 | offset_buffer[0] = 0; | |
1288 | for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed, | |
1289 | psrc += consumed) | |
1290 | { | |
dc676635 DD |
1291 | #ifdef _LIBC |
1292 | consumed = __mbrtowc (pdest, psrc, mb_remain, &mbs); | |
1293 | #else | |
2a6ef469 | 1294 | consumed = mbrtowc (pdest, psrc, mb_remain, &mbs); |
dc676635 | 1295 | #endif |
2a6ef469 DD |
1296 | |
1297 | if (consumed <= 0) | |
1298 | /* failed to convert. maybe src contains binary data. | |
1299 | So we consume 1 byte manualy. */ | |
1300 | { | |
1301 | *pdest = *psrc; | |
1302 | consumed = 1; | |
1303 | is_binary[wc_count] = TRUE; | |
1304 | } | |
1305 | else | |
1306 | is_binary[wc_count] = FALSE; | |
1307 | /* In sjis encoding, we use yen sign as escape character in | |
1308 | place of reverse solidus. So we convert 0x5c(yen sign in | |
1309 | sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse | |
1310 | solidus in UCS2). */ | |
1311 | if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5) | |
1312 | *pdest = (wchar_t) *psrc; | |
1313 | ||
1314 | offset_buffer[wc_count + 1] = mb_count += consumed; | |
1315 | } | |
1316 | ||
1317 | /* Fill remain of the buffer with sentinel. */ | |
1318 | for (i = wc_count + 1 ; i <= len ; i++) | |
1319 | offset_buffer[i] = mb_count + 1; | |
1320 | ||
1321 | return wc_count; | |
1322 | } | |
1323 | ||
1324 | # endif /* WCHAR */ | |
1325 | ||
1326 | #else /* not INSIDE_RECURSION */ | |
1327 | ||
1328 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
1329 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1330 | syntax, so it can be changed between regex compilations. */ | |
1331 | /* This has no initializer because initialized variables in Emacs | |
1332 | become read-only after dumping. */ | |
1333 | reg_syntax_t re_syntax_options; | |
1334 | ||
1335 | ||
1336 | /* Specify the precise syntax of regexps for compilation. This provides | |
1337 | for compatibility for various utilities which historically have | |
1338 | different, incompatible syntaxes. | |
1339 | ||
1340 | The argument SYNTAX is a bit mask comprised of the various bits | |
1341 | defined in regex.h. We return the old syntax. */ | |
1342 | ||
1343 | reg_syntax_t | |
1344 | re_set_syntax (syntax) | |
1345 | reg_syntax_t syntax; | |
1346 | { | |
1347 | reg_syntax_t ret = re_syntax_options; | |
1348 | ||
1349 | re_syntax_options = syntax; | |
1350 | # ifdef DEBUG | |
1351 | if (syntax & RE_DEBUG) | |
1352 | debug = 1; | |
1353 | else if (debug) /* was on but now is not */ | |
1354 | debug = 0; | |
1355 | # endif /* DEBUG */ | |
1356 | return ret; | |
1357 | } | |
1358 | # ifdef _LIBC | |
1359 | weak_alias (__re_set_syntax, re_set_syntax) | |
1360 | # endif | |
1361 | \f | |
1362 | /* This table gives an error message for each of the error codes listed | |
1363 | in regex.h. Obviously the order here has to be same as there. | |
1364 | POSIX doesn't require that we do anything for REG_NOERROR, | |
1365 | but why not be nice? */ | |
1366 | ||
1367 | static const char re_error_msgid[] = | |
1368 | { | |
1369 | # define REG_NOERROR_IDX 0 | |
1370 | gettext_noop ("Success") /* REG_NOERROR */ | |
1371 | "\0" | |
1372 | # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success") | |
1373 | gettext_noop ("No match") /* REG_NOMATCH */ | |
1374 | "\0" | |
1375 | # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match") | |
1376 | gettext_noop ("Invalid regular expression") /* REG_BADPAT */ | |
1377 | "\0" | |
1378 | # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression") | |
1379 | gettext_noop ("Invalid collation character") /* REG_ECOLLATE */ | |
1380 | "\0" | |
1381 | # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character") | |
1382 | gettext_noop ("Invalid character class name") /* REG_ECTYPE */ | |
1383 | "\0" | |
1384 | # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name") | |
1385 | gettext_noop ("Trailing backslash") /* REG_EESCAPE */ | |
1386 | "\0" | |
1387 | # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash") | |
1388 | gettext_noop ("Invalid back reference") /* REG_ESUBREG */ | |
1389 | "\0" | |
1390 | # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference") | |
1391 | gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */ | |
1392 | "\0" | |
1393 | # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^") | |
1394 | gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */ | |
1395 | "\0" | |
1396 | # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(") | |
1397 | gettext_noop ("Unmatched \\{") /* REG_EBRACE */ | |
1398 | "\0" | |
1399 | # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{") | |
1400 | gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */ | |
1401 | "\0" | |
1402 | # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}") | |
1403 | gettext_noop ("Invalid range end") /* REG_ERANGE */ | |
1404 | "\0" | |
1405 | # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end") | |
1406 | gettext_noop ("Memory exhausted") /* REG_ESPACE */ | |
1407 | "\0" | |
1408 | # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted") | |
1409 | gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */ | |
1410 | "\0" | |
1411 | # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression") | |
1412 | gettext_noop ("Premature end of regular expression") /* REG_EEND */ | |
1413 | "\0" | |
1414 | # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression") | |
1415 | gettext_noop ("Regular expression too big") /* REG_ESIZE */ | |
1416 | "\0" | |
1417 | # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big") | |
1418 | gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */ | |
1419 | }; | |
1420 | ||
1421 | static const size_t re_error_msgid_idx[] = | |
1422 | { | |
1423 | REG_NOERROR_IDX, | |
1424 | REG_NOMATCH_IDX, | |
1425 | REG_BADPAT_IDX, | |
1426 | REG_ECOLLATE_IDX, | |
1427 | REG_ECTYPE_IDX, | |
1428 | REG_EESCAPE_IDX, | |
1429 | REG_ESUBREG_IDX, | |
1430 | REG_EBRACK_IDX, | |
1431 | REG_EPAREN_IDX, | |
1432 | REG_EBRACE_IDX, | |
1433 | REG_BADBR_IDX, | |
1434 | REG_ERANGE_IDX, | |
1435 | REG_ESPACE_IDX, | |
1436 | REG_BADRPT_IDX, | |
1437 | REG_EEND_IDX, | |
1438 | REG_ESIZE_IDX, | |
1439 | REG_ERPAREN_IDX | |
1440 | }; | |
1441 | \f | |
1442 | #endif /* INSIDE_RECURSION */ | |
1443 | ||
1444 | #ifndef DEFINED_ONCE | |
1445 | /* Avoiding alloca during matching, to placate r_alloc. */ | |
1446 | ||
1447 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1448 | searching and matching functions should not call alloca. On some | |
1449 | systems, alloca is implemented in terms of malloc, and if we're | |
1450 | using the relocating allocator routines, then malloc could cause a | |
1451 | relocation, which might (if the strings being searched are in the | |
1452 | ralloc heap) shift the data out from underneath the regexp | |
1453 | routines. | |
1454 | ||
1455 | Here's another reason to avoid allocation: Emacs | |
1456 | processes input from X in a signal handler; processing X input may | |
1457 | call malloc; if input arrives while a matching routine is calling | |
1458 | malloc, then we're scrod. But Emacs can't just block input while | |
1459 | calling matching routines; then we don't notice interrupts when | |
1460 | they come in. So, Emacs blocks input around all regexp calls | |
1461 | except the matching calls, which it leaves unprotected, in the | |
1462 | faith that they will not malloc. */ | |
1463 | ||
1464 | /* Normally, this is fine. */ | |
1465 | # define MATCH_MAY_ALLOCATE | |
1466 | ||
1467 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
1468 | what config.h may say. So don't take precautions for it. */ | |
1469 | # ifdef __GNUC__ | |
1470 | # undef C_ALLOCA | |
1471 | # endif | |
1472 | ||
1473 | /* The match routines may not allocate if (1) they would do it with malloc | |
1474 | and (2) it's not safe for them to use malloc. | |
1475 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1476 | failure stack, but we would still use it for the register vectors; | |
1477 | so REL_ALLOC should not affect this. */ | |
1478 | # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs | |
1479 | # undef MATCH_MAY_ALLOCATE | |
1480 | # endif | |
1481 | #endif /* not DEFINED_ONCE */ | |
1482 | \f | |
1483 | #ifdef INSIDE_RECURSION | |
1484 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1485 | re_match_2 use a failure stack. These have to be macros because of | |
1486 | REGEX_ALLOCATE_STACK. */ | |
1487 | ||
1488 | ||
1489 | /* Number of failure points for which to initially allocate space | |
1490 | when matching. If this number is exceeded, we allocate more | |
1491 | space, so it is not a hard limit. */ | |
1492 | # ifndef INIT_FAILURE_ALLOC | |
1493 | # define INIT_FAILURE_ALLOC 5 | |
1494 | # endif | |
1495 | ||
1496 | /* Roughly the maximum number of failure points on the stack. Would be | |
1497 | exactly that if always used MAX_FAILURE_ITEMS items each time we failed. | |
1498 | This is a variable only so users of regex can assign to it; we never | |
1499 | change it ourselves. */ | |
1500 | ||
1501 | # ifdef INT_IS_16BIT | |
1502 | ||
1503 | # ifndef DEFINED_ONCE | |
1504 | # if defined MATCH_MAY_ALLOCATE | |
1505 | /* 4400 was enough to cause a crash on Alpha OSF/1, | |
1506 | whose default stack limit is 2mb. */ | |
1507 | long int re_max_failures = 4000; | |
1508 | # else | |
1509 | long int re_max_failures = 2000; | |
1510 | # endif | |
1511 | # endif | |
1512 | ||
1513 | union PREFIX(fail_stack_elt) | |
1514 | { | |
1515 | UCHAR_T *pointer; | |
1516 | long int integer; | |
1517 | }; | |
1518 | ||
1519 | typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); | |
1520 | ||
1521 | typedef struct | |
1522 | { | |
1523 | PREFIX(fail_stack_elt_t) *stack; | |
1524 | unsigned long int size; | |
1525 | unsigned long int avail; /* Offset of next open position. */ | |
1526 | } PREFIX(fail_stack_type); | |
1527 | ||
1528 | # else /* not INT_IS_16BIT */ | |
1529 | ||
1530 | # ifndef DEFINED_ONCE | |
1531 | # if defined MATCH_MAY_ALLOCATE | |
1532 | /* 4400 was enough to cause a crash on Alpha OSF/1, | |
1533 | whose default stack limit is 2mb. */ | |
1534 | int re_max_failures = 4000; | |
1535 | # else | |
1536 | int re_max_failures = 2000; | |
1537 | # endif | |
1538 | # endif | |
1539 | ||
1540 | union PREFIX(fail_stack_elt) | |
1541 | { | |
1542 | UCHAR_T *pointer; | |
1543 | int integer; | |
1544 | }; | |
1545 | ||
1546 | typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); | |
1547 | ||
1548 | typedef struct | |
1549 | { | |
1550 | PREFIX(fail_stack_elt_t) *stack; | |
1551 | unsigned size; | |
1552 | unsigned avail; /* Offset of next open position. */ | |
1553 | } PREFIX(fail_stack_type); | |
1554 | ||
1555 | # endif /* INT_IS_16BIT */ | |
1556 | ||
1557 | # ifndef DEFINED_ONCE | |
1558 | # define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
1559 | # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
1560 | # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
1561 | # endif | |
1562 | ||
1563 | ||
1564 | /* Define macros to initialize and free the failure stack. | |
1565 | Do `return -2' if the alloc fails. */ | |
1566 | ||
1567 | # ifdef MATCH_MAY_ALLOCATE | |
1568 | # define INIT_FAIL_STACK() \ | |
1569 | do { \ | |
1570 | fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \ | |
1571 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \ | |
1572 | \ | |
1573 | if (fail_stack.stack == NULL) \ | |
1574 | return -2; \ | |
1575 | \ | |
1576 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1577 | fail_stack.avail = 0; \ | |
1578 | } while (0) | |
1579 | ||
1580 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) | |
1581 | # else | |
1582 | # define INIT_FAIL_STACK() \ | |
1583 | do { \ | |
1584 | fail_stack.avail = 0; \ | |
1585 | } while (0) | |
1586 | ||
1587 | # define RESET_FAIL_STACK() | |
1588 | # endif | |
1589 | ||
1590 | ||
1591 | /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
1592 | ||
1593 | Return 1 if succeeds, and 0 if either ran out of memory | |
1594 | allocating space for it or it was already too large. | |
1595 | ||
1596 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ | |
1597 | ||
1598 | # define DOUBLE_FAIL_STACK(fail_stack) \ | |
1599 | ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \ | |
1600 | ? 0 \ | |
1601 | : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \ | |
1602 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ | |
1603 | (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \ | |
1604 | ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\ | |
1605 | \ | |
1606 | (fail_stack).stack == NULL \ | |
1607 | ? 0 \ | |
1608 | : ((fail_stack).size <<= 1, \ | |
1609 | 1))) | |
1610 | ||
1611 | ||
1612 | /* Push pointer POINTER on FAIL_STACK. | |
1613 | Return 1 if was able to do so and 0 if ran out of memory allocating | |
1614 | space to do so. */ | |
1615 | # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ | |
1616 | ((FAIL_STACK_FULL () \ | |
1617 | && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ | |
1618 | ? 0 \ | |
1619 | : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ | |
1620 | 1)) | |
1621 | ||
1622 | /* Push a pointer value onto the failure stack. | |
1623 | Assumes the variable `fail_stack'. Probably should only | |
1624 | be called from within `PUSH_FAILURE_POINT'. */ | |
1625 | # define PUSH_FAILURE_POINTER(item) \ | |
1626 | fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item) | |
1627 | ||
1628 | /* This pushes an integer-valued item onto the failure stack. | |
1629 | Assumes the variable `fail_stack'. Probably should only | |
1630 | be called from within `PUSH_FAILURE_POINT'. */ | |
1631 | # define PUSH_FAILURE_INT(item) \ | |
1632 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1633 | ||
1634 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1635 | Assumes the variable `fail_stack'. Probably should only | |
1636 | be called from within `PUSH_FAILURE_POINT'. */ | |
1637 | # define PUSH_FAILURE_ELT(item) \ | |
1638 | fail_stack.stack[fail_stack.avail++] = (item) | |
1639 | ||
1640 | /* These three POP... operations complement the three PUSH... operations. | |
1641 | All assume that `fail_stack' is nonempty. */ | |
1642 | # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1643 | # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1644 | # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1645 | ||
1646 | /* Used to omit pushing failure point id's when we're not debugging. */ | |
1647 | # ifdef DEBUG | |
1648 | # define DEBUG_PUSH PUSH_FAILURE_INT | |
1649 | # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () | |
1650 | # else | |
1651 | # define DEBUG_PUSH(item) | |
1652 | # define DEBUG_POP(item_addr) | |
1653 | # endif | |
1654 | ||
1655 | ||
1656 | /* Push the information about the state we will need | |
1657 | if we ever fail back to it. | |
1658 | ||
1659 | Requires variables fail_stack, regstart, regend, reg_info, and | |
1660 | num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination' | |
1661 | be declared. | |
1662 | ||
1663 | Does `return FAILURE_CODE' if runs out of memory. */ | |
1664 | ||
1665 | # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
1666 | do { \ | |
1667 | char *destination; \ | |
1668 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1669 | of 0 + -1 isn't done as unsigned. */ \ | |
1670 | /* Can't be int, since there is not a shred of a guarantee that int \ | |
1671 | is wide enough to hold a value of something to which pointer can \ | |
1672 | be assigned */ \ | |
1673 | active_reg_t this_reg; \ | |
1674 | \ | |
1675 | DEBUG_STATEMENT (failure_id++); \ | |
1676 | DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
1677 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
1678 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
1679 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1680 | \ | |
1681 | DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \ | |
1682 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ | |
1683 | \ | |
1684 | /* Ensure we have enough space allocated for what we will push. */ \ | |
1685 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
1686 | { \ | |
1687 | if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
1688 | return failure_code; \ | |
1689 | \ | |
1690 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
1691 | (fail_stack).size); \ | |
1692 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1693 | } \ | |
1694 | \ | |
1695 | /* Push the info, starting with the registers. */ \ | |
1696 | DEBUG_PRINT1 ("\n"); \ | |
1697 | \ | |
1698 | if (1) \ | |
1699 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ | |
1700 | this_reg++) \ | |
1701 | { \ | |
1702 | DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \ | |
1703 | DEBUG_STATEMENT (num_regs_pushed++); \ | |
1704 | \ | |
1705 | DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ | |
1706 | PUSH_FAILURE_POINTER (regstart[this_reg]); \ | |
1707 | \ | |
1708 | DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ | |
1709 | PUSH_FAILURE_POINTER (regend[this_reg]); \ | |
1710 | \ | |
1711 | DEBUG_PRINT2 (" info: %p\n ", \ | |
1712 | reg_info[this_reg].word.pointer); \ | |
1713 | DEBUG_PRINT2 (" match_null=%d", \ | |
1714 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
1715 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
1716 | DEBUG_PRINT2 (" matched_something=%d", \ | |
1717 | MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1718 | DEBUG_PRINT2 (" ever_matched=%d", \ | |
1719 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1720 | DEBUG_PRINT1 ("\n"); \ | |
1721 | PUSH_FAILURE_ELT (reg_info[this_reg].word); \ | |
1722 | } \ | |
1723 | \ | |
1724 | DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\ | |
1725 | PUSH_FAILURE_INT (lowest_active_reg); \ | |
1726 | \ | |
1727 | DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\ | |
1728 | PUSH_FAILURE_INT (highest_active_reg); \ | |
1729 | \ | |
1730 | DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \ | |
1731 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
1732 | PUSH_FAILURE_POINTER (pattern_place); \ | |
1733 | \ | |
1734 | DEBUG_PRINT2 (" Pushing string %p: `", string_place); \ | |
1735 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
1736 | size2); \ | |
1737 | DEBUG_PRINT1 ("'\n"); \ | |
1738 | PUSH_FAILURE_POINTER (string_place); \ | |
1739 | \ | |
1740 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
1741 | DEBUG_PUSH (failure_id); \ | |
1742 | } while (0) | |
1743 | ||
1744 | # ifndef DEFINED_ONCE | |
1745 | /* This is the number of items that are pushed and popped on the stack | |
1746 | for each register. */ | |
1747 | # define NUM_REG_ITEMS 3 | |
1748 | ||
1749 | /* Individual items aside from the registers. */ | |
1750 | # ifdef DEBUG | |
1751 | # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
1752 | # else | |
1753 | # define NUM_NONREG_ITEMS 4 | |
1754 | # endif | |
1755 | ||
1756 | /* We push at most this many items on the stack. */ | |
1757 | /* We used to use (num_regs - 1), which is the number of registers | |
1758 | this regexp will save; but that was changed to 5 | |
1759 | to avoid stack overflow for a regexp with lots of parens. */ | |
1760 | # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
1761 | ||
1762 | /* We actually push this many items. */ | |
1763 | # define NUM_FAILURE_ITEMS \ | |
1764 | (((0 \ | |
1765 | ? 0 : highest_active_reg - lowest_active_reg + 1) \ | |
1766 | * NUM_REG_ITEMS) \ | |
1767 | + NUM_NONREG_ITEMS) | |
1768 | ||
1769 | /* How many items can still be added to the stack without overflowing it. */ | |
1770 | # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1771 | # endif /* not DEFINED_ONCE */ | |
1772 | ||
1773 | ||
1774 | /* Pops what PUSH_FAIL_STACK pushes. | |
1775 | ||
1776 | We restore into the parameters, all of which should be lvalues: | |
1777 | STR -- the saved data position. | |
1778 | PAT -- the saved pattern position. | |
1779 | LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
1780 | REGSTART, REGEND -- arrays of string positions. | |
1781 | REG_INFO -- array of information about each subexpression. | |
1782 | ||
1783 | Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
1784 | `pend', `string1', `size1', `string2', and `size2'. */ | |
1785 | # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
1786 | { \ | |
1787 | DEBUG_STATEMENT (unsigned failure_id;) \ | |
1788 | active_reg_t this_reg; \ | |
1789 | const UCHAR_T *string_temp; \ | |
1790 | \ | |
1791 | assert (!FAIL_STACK_EMPTY ()); \ | |
1792 | \ | |
1793 | /* Remove failure points and point to how many regs pushed. */ \ | |
1794 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1795 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
1796 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
1797 | \ | |
1798 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
1799 | \ | |
1800 | DEBUG_POP (&failure_id); \ | |
1801 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
1802 | \ | |
1803 | /* If the saved string location is NULL, it came from an \ | |
1804 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1805 | saved NULL, thus retaining our current position in the string. */ \ | |
1806 | string_temp = POP_FAILURE_POINTER (); \ | |
1807 | if (string_temp != NULL) \ | |
1808 | str = (const CHAR_T *) string_temp; \ | |
1809 | \ | |
1810 | DEBUG_PRINT2 (" Popping string %p: `", str); \ | |
1811 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
1812 | DEBUG_PRINT1 ("'\n"); \ | |
1813 | \ | |
1814 | pat = (UCHAR_T *) POP_FAILURE_POINTER (); \ | |
1815 | DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \ | |
1816 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
1817 | \ | |
1818 | /* Restore register info. */ \ | |
1819 | high_reg = (active_reg_t) POP_FAILURE_INT (); \ | |
1820 | DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \ | |
1821 | \ | |
1822 | low_reg = (active_reg_t) POP_FAILURE_INT (); \ | |
1823 | DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \ | |
1824 | \ | |
1825 | if (1) \ | |
1826 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
1827 | { \ | |
1828 | DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \ | |
1829 | \ | |
1830 | reg_info[this_reg].word = POP_FAILURE_ELT (); \ | |
1831 | DEBUG_PRINT2 (" info: %p\n", \ | |
1832 | reg_info[this_reg].word.pointer); \ | |
1833 | \ | |
1834 | regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ | |
1835 | DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ | |
1836 | \ | |
1837 | regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ | |
1838 | DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ | |
1839 | } \ | |
1840 | else \ | |
1841 | { \ | |
1842 | for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ | |
1843 | { \ | |
1844 | reg_info[this_reg].word.integer = 0; \ | |
1845 | regend[this_reg] = 0; \ | |
1846 | regstart[this_reg] = 0; \ | |
1847 | } \ | |
1848 | highest_active_reg = high_reg; \ | |
1849 | } \ | |
1850 | \ | |
1851 | set_regs_matched_done = 0; \ | |
1852 | DEBUG_STATEMENT (nfailure_points_popped++); \ | |
1853 | } /* POP_FAILURE_POINT */ | |
1854 | \f | |
1855 | /* Structure for per-register (a.k.a. per-group) information. | |
1856 | Other register information, such as the | |
1857 | starting and ending positions (which are addresses), and the list of | |
1858 | inner groups (which is a bits list) are maintained in separate | |
1859 | variables. | |
1860 | ||
1861 | We are making a (strictly speaking) nonportable assumption here: that | |
1862 | the compiler will pack our bit fields into something that fits into | |
1863 | the type of `word', i.e., is something that fits into one item on the | |
1864 | failure stack. */ | |
1865 | ||
1866 | ||
1867 | /* Declarations and macros for re_match_2. */ | |
1868 | ||
1869 | typedef union | |
1870 | { | |
1871 | PREFIX(fail_stack_elt_t) word; | |
1872 | struct | |
1873 | { | |
1874 | /* This field is one if this group can match the empty string, | |
1875 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
1876 | # define MATCH_NULL_UNSET_VALUE 3 | |
1877 | unsigned match_null_string_p : 2; | |
1878 | unsigned is_active : 1; | |
1879 | unsigned matched_something : 1; | |
1880 | unsigned ever_matched_something : 1; | |
1881 | } bits; | |
1882 | } PREFIX(register_info_type); | |
1883 | ||
1884 | # ifndef DEFINED_ONCE | |
1885 | # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
1886 | # define IS_ACTIVE(R) ((R).bits.is_active) | |
1887 | # define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
1888 | # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
1889 | ||
1890 | ||
1891 | /* Call this when have matched a real character; it sets `matched' flags | |
1892 | for the subexpressions which we are currently inside. Also records | |
1893 | that those subexprs have matched. */ | |
1894 | # define SET_REGS_MATCHED() \ | |
1895 | do \ | |
1896 | { \ | |
1897 | if (!set_regs_matched_done) \ | |
1898 | { \ | |
1899 | active_reg_t r; \ | |
1900 | set_regs_matched_done = 1; \ | |
1901 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
1902 | { \ | |
1903 | MATCHED_SOMETHING (reg_info[r]) \ | |
1904 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
1905 | = 1; \ | |
1906 | } \ | |
1907 | } \ | |
1908 | } \ | |
1909 | while (0) | |
1910 | # endif /* not DEFINED_ONCE */ | |
1911 | ||
1912 | /* Registers are set to a sentinel when they haven't yet matched. */ | |
1913 | static CHAR_T PREFIX(reg_unset_dummy); | |
1914 | # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy)) | |
1915 | # define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
1916 | ||
1917 | /* Subroutine declarations and macros for regex_compile. */ | |
1918 | static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg)); | |
1919 | static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, | |
1920 | int arg1, int arg2)); | |
1921 | static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, | |
1922 | int arg, UCHAR_T *end)); | |
1923 | static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, | |
1924 | int arg1, int arg2, UCHAR_T *end)); | |
1925 | static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern, | |
1926 | const CHAR_T *p, | |
1927 | reg_syntax_t syntax)); | |
1928 | static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p, | |
1929 | const CHAR_T *pend, | |
1930 | reg_syntax_t syntax)); | |
1931 | # ifdef WCHAR | |
1932 | static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start, | |
1933 | const CHAR_T **p_ptr, | |
1934 | const CHAR_T *pend, | |
1935 | char *translate, | |
1936 | reg_syntax_t syntax, | |
1937 | UCHAR_T *b, | |
1938 | CHAR_T *char_set)); | |
1939 | static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end)); | |
1940 | # else /* BYTE */ | |
1941 | static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start, | |
1942 | const char **p_ptr, | |
1943 | const char *pend, | |
1944 | char *translate, | |
1945 | reg_syntax_t syntax, | |
1946 | unsigned char *b)); | |
1947 | # endif /* WCHAR */ | |
1948 | ||
1949 | /* Fetch the next character in the uncompiled pattern---translating it | |
1950 | if necessary. Also cast from a signed character in the constant | |
1951 | string passed to us by the user to an unsigned char that we can use | |
1952 | as an array index (in, e.g., `translate'). */ | |
1953 | /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, | |
1954 | because it is impossible to allocate 4GB array for some encodings | |
1955 | which have 4 byte character_set like UCS4. */ | |
1956 | # ifndef PATFETCH | |
1957 | # ifdef WCHAR | |
1958 | # define PATFETCH(c) \ | |
1959 | do {if (p == pend) return REG_EEND; \ | |
1960 | c = (UCHAR_T) *p++; \ | |
1961 | if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \ | |
1962 | } while (0) | |
1963 | # else /* BYTE */ | |
1964 | # define PATFETCH(c) \ | |
1965 | do {if (p == pend) return REG_EEND; \ | |
1966 | c = (unsigned char) *p++; \ | |
1967 | if (translate) c = (unsigned char) translate[c]; \ | |
1968 | } while (0) | |
1969 | # endif /* WCHAR */ | |
1970 | # endif | |
1971 | ||
1972 | /* Fetch the next character in the uncompiled pattern, with no | |
1973 | translation. */ | |
1974 | # define PATFETCH_RAW(c) \ | |
1975 | do {if (p == pend) return REG_EEND; \ | |
1976 | c = (UCHAR_T) *p++; \ | |
1977 | } while (0) | |
1978 | ||
1979 | /* Go backwards one character in the pattern. */ | |
1980 | # define PATUNFETCH p-- | |
1981 | ||
1982 | ||
1983 | /* If `translate' is non-null, return translate[D], else just D. We | |
1984 | cast the subscript to translate because some data is declared as | |
1985 | `char *', to avoid warnings when a string constant is passed. But | |
1986 | when we use a character as a subscript we must make it unsigned. */ | |
1987 | /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, | |
1988 | because it is impossible to allocate 4GB array for some encodings | |
1989 | which have 4 byte character_set like UCS4. */ | |
1990 | ||
1991 | # ifndef TRANSLATE | |
1992 | # ifdef WCHAR | |
1993 | # define TRANSLATE(d) \ | |
1994 | ((translate && ((UCHAR_T) (d)) <= 0xff) \ | |
1995 | ? (char) translate[(unsigned char) (d)] : (d)) | |
1996 | # else /* BYTE */ | |
1997 | # define TRANSLATE(d) \ | |
1998 | (translate ? (char) translate[(unsigned char) (d)] : (d)) | |
1999 | # endif /* WCHAR */ | |
2000 | # endif | |
2001 | ||
2002 | ||
2003 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
2004 | ||
2005 | /* If the buffer isn't allocated when it comes in, use this. */ | |
2006 | # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T)) | |
2007 | ||
2008 | /* Make sure we have at least N more bytes of space in buffer. */ | |
2009 | # ifdef WCHAR | |
2010 | # define GET_BUFFER_SPACE(n) \ | |
2011 | while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \ | |
2012 | + (n)*sizeof(CHAR_T)) > bufp->allocated) \ | |
2013 | EXTEND_BUFFER () | |
2014 | # else /* BYTE */ | |
2015 | # define GET_BUFFER_SPACE(n) \ | |
2016 | while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \ | |
2017 | EXTEND_BUFFER () | |
2018 | # endif /* WCHAR */ | |
2019 | ||
2020 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
2021 | # define BUF_PUSH(c) \ | |
2022 | do { \ | |
2023 | GET_BUFFER_SPACE (1); \ | |
2024 | *b++ = (UCHAR_T) (c); \ | |
2025 | } while (0) | |
2026 | ||
2027 | ||
2028 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
2029 | # define BUF_PUSH_2(c1, c2) \ | |
2030 | do { \ | |
2031 | GET_BUFFER_SPACE (2); \ | |
2032 | *b++ = (UCHAR_T) (c1); \ | |
2033 | *b++ = (UCHAR_T) (c2); \ | |
2034 | } while (0) | |
2035 | ||
2036 | ||
2037 | /* As with BUF_PUSH_2, except for three bytes. */ | |
2038 | # define BUF_PUSH_3(c1, c2, c3) \ | |
2039 | do { \ | |
2040 | GET_BUFFER_SPACE (3); \ | |
2041 | *b++ = (UCHAR_T) (c1); \ | |
2042 | *b++ = (UCHAR_T) (c2); \ | |
2043 | *b++ = (UCHAR_T) (c3); \ | |
2044 | } while (0) | |
2045 | ||
2046 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
2047 | relative address offset by the three bytes the jump itself occupies. */ | |
2048 | # define STORE_JUMP(op, loc, to) \ | |
2049 | PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE))) | |
2050 | ||
2051 | /* Likewise, for a two-argument jump. */ | |
2052 | # define STORE_JUMP2(op, loc, to, arg) \ | |
2053 | PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg) | |
2054 | ||
2055 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
2056 | # define INSERT_JUMP(op, loc, to) \ | |
2057 | PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b) | |
2058 | ||
2059 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
2060 | # define INSERT_JUMP2(op, loc, to, arg) \ | |
2061 | PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\ | |
2062 | arg, b) | |
2063 | ||
2064 | /* This is not an arbitrary limit: the arguments which represent offsets | |
2065 | into the pattern are two bytes long. So if 2^16 bytes turns out to | |
2066 | be too small, many things would have to change. */ | |
2067 | /* Any other compiler which, like MSC, has allocation limit below 2^16 | |
2068 | bytes will have to use approach similar to what was done below for | |
2069 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up | |
2070 | reallocating to 0 bytes. Such thing is not going to work too well. | |
2071 | You have been warned!! */ | |
2072 | # ifndef DEFINED_ONCE | |
2073 | # if defined _MSC_VER && !defined WIN32 | |
2074 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. | |
2075 | The REALLOC define eliminates a flurry of conversion warnings, | |
2076 | but is not required. */ | |
2077 | # define MAX_BUF_SIZE 65500L | |
2078 | # define REALLOC(p,s) realloc ((p), (size_t) (s)) | |
2079 | # else | |
2080 | # define MAX_BUF_SIZE (1L << 16) | |
2081 | # define REALLOC(p,s) realloc ((p), (s)) | |
2082 | # endif | |
2083 | ||
2084 | /* Extend the buffer by twice its current size via realloc and | |
2085 | reset the pointers that pointed into the old block to point to the | |
2086 | correct places in the new one. If extending the buffer results in it | |
2087 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
2088 | # if __BOUNDED_POINTERS__ | |
2089 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) | |
2090 | # define MOVE_BUFFER_POINTER(P) \ | |
2091 | (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) | |
2092 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
2093 | else \ | |
2094 | { \ | |
2095 | SET_HIGH_BOUND (b); \ | |
2096 | SET_HIGH_BOUND (begalt); \ | |
2097 | if (fixup_alt_jump) \ | |
2098 | SET_HIGH_BOUND (fixup_alt_jump); \ | |
2099 | if (laststart) \ | |
2100 | SET_HIGH_BOUND (laststart); \ | |
2101 | if (pending_exact) \ | |
2102 | SET_HIGH_BOUND (pending_exact); \ | |
2103 | } | |
2104 | # else | |
2105 | # define MOVE_BUFFER_POINTER(P) (P) += incr | |
2106 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND | |
2107 | # endif | |
2108 | # endif /* not DEFINED_ONCE */ | |
2109 | ||
2110 | # ifdef WCHAR | |
2111 | # define EXTEND_BUFFER() \ | |
2112 | do { \ | |
2113 | UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ | |
2114 | int wchar_count; \ | |
2115 | if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \ | |
2116 | return REG_ESIZE; \ | |
2117 | bufp->allocated <<= 1; \ | |
2118 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
2119 | bufp->allocated = MAX_BUF_SIZE; \ | |
2120 | /* How many characters the new buffer can have? */ \ | |
2121 | wchar_count = bufp->allocated / sizeof(UCHAR_T); \ | |
2122 | if (wchar_count == 0) wchar_count = 1; \ | |
2123 | /* Truncate the buffer to CHAR_T align. */ \ | |
2124 | bufp->allocated = wchar_count * sizeof(UCHAR_T); \ | |
2125 | RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \ | |
2126 | bufp->buffer = (char*)COMPILED_BUFFER_VAR; \ | |
2127 | if (COMPILED_BUFFER_VAR == NULL) \ | |
2128 | return REG_ESPACE; \ | |
2129 | /* If the buffer moved, move all the pointers into it. */ \ | |
2130 | if (old_buffer != COMPILED_BUFFER_VAR) \ | |
2131 | { \ | |
2132 | int incr = COMPILED_BUFFER_VAR - old_buffer; \ | |
2133 | MOVE_BUFFER_POINTER (b); \ | |
2134 | MOVE_BUFFER_POINTER (begalt); \ | |
2135 | if (fixup_alt_jump) \ | |
2136 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ | |
2137 | if (laststart) \ | |
2138 | MOVE_BUFFER_POINTER (laststart); \ | |
2139 | if (pending_exact) \ | |
2140 | MOVE_BUFFER_POINTER (pending_exact); \ | |
2141 | } \ | |
2142 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
2143 | } while (0) | |
2144 | # else /* BYTE */ | |
2145 | # define EXTEND_BUFFER() \ | |
2146 | do { \ | |
2147 | UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ | |
2148 | if (bufp->allocated == MAX_BUF_SIZE) \ | |
2149 | return REG_ESIZE; \ | |
2150 | bufp->allocated <<= 1; \ | |
2151 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
2152 | bufp->allocated = MAX_BUF_SIZE; \ | |
2153 | bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \ | |
2154 | bufp->allocated); \ | |
2155 | if (COMPILED_BUFFER_VAR == NULL) \ | |
2156 | return REG_ESPACE; \ | |
2157 | /* If the buffer moved, move all the pointers into it. */ \ | |
2158 | if (old_buffer != COMPILED_BUFFER_VAR) \ | |
2159 | { \ | |
2160 | int incr = COMPILED_BUFFER_VAR - old_buffer; \ | |
2161 | MOVE_BUFFER_POINTER (b); \ | |
2162 | MOVE_BUFFER_POINTER (begalt); \ | |
2163 | if (fixup_alt_jump) \ | |
2164 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ | |
2165 | if (laststart) \ | |
2166 | MOVE_BUFFER_POINTER (laststart); \ | |
2167 | if (pending_exact) \ | |
2168 | MOVE_BUFFER_POINTER (pending_exact); \ | |
2169 | } \ | |
2170 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
2171 | } while (0) | |
2172 | # endif /* WCHAR */ | |
2173 | ||
2174 | # ifndef DEFINED_ONCE | |
2175 | /* Since we have one byte reserved for the register number argument to | |
2176 | {start,stop}_memory, the maximum number of groups we can report | |
2177 | things about is what fits in that byte. */ | |
2178 | # define MAX_REGNUM 255 | |
2179 | ||
2180 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
2181 | ignore the excess. */ | |
2182 | typedef unsigned regnum_t; | |
2183 | ||
2184 | ||
2185 | /* Macros for the compile stack. */ | |
2186 | ||
2187 | /* Since offsets can go either forwards or backwards, this type needs to | |
2188 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
2189 | /* int may be not enough when sizeof(int) == 2. */ | |
2190 | typedef long pattern_offset_t; | |
2191 | ||
2192 | typedef struct | |
2193 | { | |
2194 | pattern_offset_t begalt_offset; | |
2195 | pattern_offset_t fixup_alt_jump; | |
2196 | pattern_offset_t inner_group_offset; | |
2197 | pattern_offset_t laststart_offset; | |
2198 | regnum_t regnum; | |
2199 | } compile_stack_elt_t; | |
2200 | ||
2201 | ||
2202 | typedef struct | |
2203 | { | |
2204 | compile_stack_elt_t *stack; | |
2205 | unsigned size; | |
2206 | unsigned avail; /* Offset of next open position. */ | |
2207 | } compile_stack_type; | |
2208 | ||
2209 | ||
2210 | # define INIT_COMPILE_STACK_SIZE 32 | |
2211 | ||
2212 | # define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
2213 | # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
2214 | ||
2215 | /* The next available element. */ | |
2216 | # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
2217 | ||
2218 | # endif /* not DEFINED_ONCE */ | |
2219 | ||
2220 | /* Set the bit for character C in a list. */ | |
2221 | # ifndef DEFINED_ONCE | |
2222 | # define SET_LIST_BIT(c) \ | |
2223 | (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
2224 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
2225 | # endif /* DEFINED_ONCE */ | |
2226 | ||
2227 | /* Get the next unsigned number in the uncompiled pattern. */ | |
2228 | # define GET_UNSIGNED_NUMBER(num) \ | |
2229 | { \ | |
2230 | while (p != pend) \ | |
2231 | { \ | |
2232 | PATFETCH (c); \ | |
2233 | if (c < '0' || c > '9') \ | |
2234 | break; \ | |
2235 | if (num <= RE_DUP_MAX) \ | |
2236 | { \ | |
2237 | if (num < 0) \ | |
2238 | num = 0; \ | |
2239 | num = num * 10 + c - '0'; \ | |
2240 | } \ | |
2241 | } \ | |
2242 | } | |
2243 | ||
2244 | # ifndef DEFINED_ONCE | |
2245 | # if defined _LIBC || WIDE_CHAR_SUPPORT | |
2246 | /* The GNU C library provides support for user-defined character classes | |
2247 | and the functions from ISO C amendement 1. */ | |
2248 | # ifdef CHARCLASS_NAME_MAX | |
2249 | # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX | |
2250 | # else | |
2251 | /* This shouldn't happen but some implementation might still have this | |
2252 | problem. Use a reasonable default value. */ | |
2253 | # define CHAR_CLASS_MAX_LENGTH 256 | |
2254 | # endif | |
2255 | ||
2256 | # ifdef _LIBC | |
2257 | # define IS_CHAR_CLASS(string) __wctype (string) | |
2258 | # else | |
2259 | # define IS_CHAR_CLASS(string) wctype (string) | |
2260 | # endif | |
2261 | # else | |
2262 | # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
2263 | ||
2264 | # define IS_CHAR_CLASS(string) \ | |
2265 | (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
2266 | || STREQ (string, "lower") || STREQ (string, "digit") \ | |
2267 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
2268 | || STREQ (string, "space") || STREQ (string, "print") \ | |
2269 | || STREQ (string, "punct") || STREQ (string, "graph") \ | |
2270 | || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
2271 | # endif | |
2272 | # endif /* DEFINED_ONCE */ | |
2273 | \f | |
2274 | # ifndef MATCH_MAY_ALLOCATE | |
2275 | ||
2276 | /* If we cannot allocate large objects within re_match_2_internal, | |
2277 | we make the fail stack and register vectors global. | |
2278 | The fail stack, we grow to the maximum size when a regexp | |
2279 | is compiled. | |
2280 | The register vectors, we adjust in size each time we | |
2281 | compile a regexp, according to the number of registers it needs. */ | |
2282 | ||
2283 | static PREFIX(fail_stack_type) fail_stack; | |
2284 | ||
2285 | /* Size with which the following vectors are currently allocated. | |
2286 | That is so we can make them bigger as needed, | |
2287 | but never make them smaller. */ | |
2288 | # ifdef DEFINED_ONCE | |
2289 | static int regs_allocated_size; | |
2290 | ||
2291 | static const char ** regstart, ** regend; | |
2292 | static const char ** old_regstart, ** old_regend; | |
2293 | static const char **best_regstart, **best_regend; | |
2294 | static const char **reg_dummy; | |
2295 | # endif /* DEFINED_ONCE */ | |
2296 | ||
2297 | static PREFIX(register_info_type) *PREFIX(reg_info); | |
2298 | static PREFIX(register_info_type) *PREFIX(reg_info_dummy); | |
2299 | ||
2300 | /* Make the register vectors big enough for NUM_REGS registers, | |
2301 | but don't make them smaller. */ | |
2302 | ||
2303 | static void | |
2304 | PREFIX(regex_grow_registers) (num_regs) | |
2305 | int num_regs; | |
2306 | { | |
2307 | if (num_regs > regs_allocated_size) | |
2308 | { | |
2309 | RETALLOC_IF (regstart, num_regs, const char *); | |
2310 | RETALLOC_IF (regend, num_regs, const char *); | |
2311 | RETALLOC_IF (old_regstart, num_regs, const char *); | |
2312 | RETALLOC_IF (old_regend, num_regs, const char *); | |
2313 | RETALLOC_IF (best_regstart, num_regs, const char *); | |
2314 | RETALLOC_IF (best_regend, num_regs, const char *); | |
2315 | RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type)); | |
2316 | RETALLOC_IF (reg_dummy, num_regs, const char *); | |
2317 | RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type)); | |
2318 | ||
2319 | regs_allocated_size = num_regs; | |
2320 | } | |
2321 | } | |
2322 | ||
2323 | # endif /* not MATCH_MAY_ALLOCATE */ | |
2324 | \f | |
2325 | # ifndef DEFINED_ONCE | |
2326 | static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type | |
2327 | compile_stack, | |
2328 | regnum_t regnum)); | |
2329 | # endif /* not DEFINED_ONCE */ | |
2330 | ||
2331 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
2332 | Returns one of error codes defined in `regex.h', or zero for success. | |
2333 | ||
2334 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
2335 | fields are set in BUFP on entry. | |
2336 | ||
2337 | If it succeeds, results are put in BUFP (if it returns an error, the | |
2338 | contents of BUFP are undefined): | |
2339 | `buffer' is the compiled pattern; | |
2340 | `syntax' is set to SYNTAX; | |
2341 | `used' is set to the length of the compiled pattern; | |
2342 | `fastmap_accurate' is zero; | |
2343 | `re_nsub' is the number of subexpressions in PATTERN; | |
2344 | `not_bol' and `not_eol' are zero; | |
2345 | ||
2346 | The `fastmap' and `newline_anchor' fields are neither | |
2347 | examined nor set. */ | |
2348 | ||
2349 | /* Return, freeing storage we allocated. */ | |
2350 | # ifdef WCHAR | |
2351 | # define FREE_STACK_RETURN(value) \ | |
2352 | return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value) | |
2353 | # else | |
2354 | # define FREE_STACK_RETURN(value) \ | |
2355 | return (free (compile_stack.stack), value) | |
2356 | # endif /* WCHAR */ | |
2357 | ||
2358 | static reg_errcode_t | |
2359 | PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp) | |
2360 | const char *ARG_PREFIX(pattern); | |
2361 | size_t ARG_PREFIX(size); | |
2362 | reg_syntax_t syntax; | |
2363 | struct re_pattern_buffer *bufp; | |
2364 | { | |
2365 | /* We fetch characters from PATTERN here. Even though PATTERN is | |
2366 | `char *' (i.e., signed), we declare these variables as unsigned, so | |
2367 | they can be reliably used as array indices. */ | |
2368 | register UCHAR_T c, c1; | |
2369 | ||
2370 | #ifdef WCHAR | |
2371 | /* A temporary space to keep wchar_t pattern and compiled pattern. */ | |
2372 | CHAR_T *pattern, *COMPILED_BUFFER_VAR; | |
2373 | size_t size; | |
2374 | /* offset buffer for optimization. See convert_mbs_to_wc. */ | |
2375 | int *mbs_offset = NULL; | |
2376 | /* It hold whether each wchar_t is binary data or not. */ | |
2377 | char *is_binary = NULL; | |
2378 | /* A flag whether exactn is handling binary data or not. */ | |
2379 | char is_exactn_bin = FALSE; | |
2380 | #endif /* WCHAR */ | |
2381 | ||
2382 | /* A random temporary spot in PATTERN. */ | |
2383 | const CHAR_T *p1; | |
2384 | ||
2385 | /* Points to the end of the buffer, where we should append. */ | |
2386 | register UCHAR_T *b; | |
2387 | ||
2388 | /* Keeps track of unclosed groups. */ | |
2389 | compile_stack_type compile_stack; | |
2390 | ||
2391 | /* Points to the current (ending) position in the pattern. */ | |
2392 | #ifdef WCHAR | |
2393 | const CHAR_T *p; | |
2394 | const CHAR_T *pend; | |
2395 | #else /* BYTE */ | |
2396 | const CHAR_T *p = pattern; | |
2397 | const CHAR_T *pend = pattern + size; | |
2398 | #endif /* WCHAR */ | |
2399 | ||
2400 | /* How to translate the characters in the pattern. */ | |
2401 | RE_TRANSLATE_TYPE translate = bufp->translate; | |
2402 | ||
2403 | /* Address of the count-byte of the most recently inserted `exactn' | |
2404 | command. This makes it possible to tell if a new exact-match | |
2405 | character can be added to that command or if the character requires | |
2406 | a new `exactn' command. */ | |
2407 | UCHAR_T *pending_exact = 0; | |
2408 | ||
2409 | /* Address of start of the most recently finished expression. | |
2410 | This tells, e.g., postfix * where to find the start of its | |
2411 | operand. Reset at the beginning of groups and alternatives. */ | |
2412 | UCHAR_T *laststart = 0; | |
2413 | ||
2414 | /* Address of beginning of regexp, or inside of last group. */ | |
2415 | UCHAR_T *begalt; | |
2416 | ||
2417 | /* Address of the place where a forward jump should go to the end of | |
2418 | the containing expression. Each alternative of an `or' -- except the | |
2419 | last -- ends with a forward jump of this sort. */ | |
2420 | UCHAR_T *fixup_alt_jump = 0; | |
2421 | ||
2422 | /* Counts open-groups as they are encountered. Remembered for the | |
2423 | matching close-group on the compile stack, so the same register | |
2424 | number is put in the stop_memory as the start_memory. */ | |
2425 | regnum_t regnum = 0; | |
2426 | ||
2427 | #ifdef WCHAR | |
2428 | /* Initialize the wchar_t PATTERN and offset_buffer. */ | |
2429 | p = pend = pattern = TALLOC(csize + 1, CHAR_T); | |
2430 | mbs_offset = TALLOC(csize + 1, int); | |
2431 | is_binary = TALLOC(csize + 1, char); | |
2432 | if (pattern == NULL || mbs_offset == NULL || is_binary == NULL) | |
2433 | { | |
2434 | free(pattern); | |
2435 | free(mbs_offset); | |
2436 | free(is_binary); | |
2437 | return REG_ESPACE; | |
2438 | } | |
2439 | pattern[csize] = L'\0'; /* sentinel */ | |
2440 | size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary); | |
2441 | pend = p + size; | |
2442 | if (size < 0) | |
2443 | { | |
2444 | free(pattern); | |
2445 | free(mbs_offset); | |
2446 | free(is_binary); | |
2447 | return REG_BADPAT; | |
2448 | } | |
2449 | #endif | |
2450 | ||
2451 | #ifdef DEBUG | |
2452 | DEBUG_PRINT1 ("\nCompiling pattern: "); | |
2453 | if (debug) | |
2454 | { | |
2455 | unsigned debug_count; | |
2456 | ||
2457 | for (debug_count = 0; debug_count < size; debug_count++) | |
2458 | PUT_CHAR (pattern[debug_count]); | |
2459 | putchar ('\n'); | |
2460 | } | |
2461 | #endif /* DEBUG */ | |
2462 | ||
2463 | /* Initialize the compile stack. */ | |
2464 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
2465 | if (compile_stack.stack == NULL) | |
2466 | { | |
2467 | #ifdef WCHAR | |
2468 | free(pattern); | |
2469 | free(mbs_offset); | |
2470 | free(is_binary); | |
2471 | #endif | |
2472 | return REG_ESPACE; | |
2473 | } | |
2474 | ||
2475 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
2476 | compile_stack.avail = 0; | |
2477 | ||
2478 | /* Initialize the pattern buffer. */ | |
2479 | bufp->syntax = syntax; | |
2480 | bufp->fastmap_accurate = 0; | |
2481 | bufp->not_bol = bufp->not_eol = 0; | |
2482 | ||
2483 | /* Set `used' to zero, so that if we return an error, the pattern | |
2484 | printer (for debugging) will think there's no pattern. We reset it | |
2485 | at the end. */ | |
2486 | bufp->used = 0; | |
2487 | ||
2488 | /* Always count groups, whether or not bufp->no_sub is set. */ | |
2489 | bufp->re_nsub = 0; | |
2490 | ||
2491 | #if !defined emacs && !defined SYNTAX_TABLE | |
2492 | /* Initialize the syntax table. */ | |
2493 | init_syntax_once (); | |
2494 | #endif | |
2495 | ||
2496 | if (bufp->allocated == 0) | |
2497 | { | |
2498 | if (bufp->buffer) | |
2499 | { /* If zero allocated, but buffer is non-null, try to realloc | |
2500 | enough space. This loses if buffer's address is bogus, but | |
2501 | that is the user's responsibility. */ | |
2502 | #ifdef WCHAR | |
2503 | /* Free bufp->buffer and allocate an array for wchar_t pattern | |
2504 | buffer. */ | |
2505 | free(bufp->buffer); | |
2506 | COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T), | |
2507 | UCHAR_T); | |
2508 | #else | |
2509 | RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T); | |
2510 | #endif /* WCHAR */ | |
2511 | } | |
2512 | else | |
2513 | { /* Caller did not allocate a buffer. Do it for them. */ | |
2514 | COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T), | |
2515 | UCHAR_T); | |
2516 | } | |
2517 | ||
2518 | if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE); | |
2519 | #ifdef WCHAR | |
2520 | bufp->buffer = (char*)COMPILED_BUFFER_VAR; | |
2521 | #endif /* WCHAR */ | |
2522 | bufp->allocated = INIT_BUF_SIZE; | |
2523 | } | |
2524 | #ifdef WCHAR | |
2525 | else | |
2526 | COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer; | |
2527 | #endif | |
2528 | ||
2529 | begalt = b = COMPILED_BUFFER_VAR; | |
2530 | ||
2531 | /* Loop through the uncompiled pattern until we're at the end. */ | |
2532 | while (p != pend) | |
2533 | { | |
2534 | PATFETCH (c); | |
2535 | ||
2536 | switch (c) | |
2537 | { | |
2538 | case '^': | |
2539 | { | |
2540 | if ( /* If at start of pattern, it's an operator. */ | |
2541 | p == pattern + 1 | |
2542 | /* If context independent, it's an operator. */ | |
2543 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2544 | /* Otherwise, depends on what's come before. */ | |
2545 | || PREFIX(at_begline_loc_p) (pattern, p, syntax)) | |
2546 | BUF_PUSH (begline); | |
2547 | else | |
2548 | goto normal_char; | |
2549 | } | |
2550 | break; | |
2551 | ||
2552 | ||
2553 | case '$': | |
2554 | { | |
2555 | if ( /* If at end of pattern, it's an operator. */ | |
2556 | p == pend | |
2557 | /* If context independent, it's an operator. */ | |
2558 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2559 | /* Otherwise, depends on what's next. */ | |
2560 | || PREFIX(at_endline_loc_p) (p, pend, syntax)) | |
2561 | BUF_PUSH (endline); | |
2562 | else | |
2563 | goto normal_char; | |
2564 | } | |
2565 | break; | |
2566 | ||
2567 | ||
2568 | case '+': | |
2569 | case '?': | |
2570 | if ((syntax & RE_BK_PLUS_QM) | |
2571 | || (syntax & RE_LIMITED_OPS)) | |
2572 | goto normal_char; | |
2573 | handle_plus: | |
2574 | case '*': | |
2575 | /* If there is no previous pattern... */ | |
2576 | if (!laststart) | |
2577 | { | |
2578 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2579 | FREE_STACK_RETURN (REG_BADRPT); | |
2580 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2581 | goto normal_char; | |
2582 | } | |
2583 | ||
2584 | { | |
2585 | /* Are we optimizing this jump? */ | |
2586 | boolean keep_string_p = false; | |
2587 | ||
2588 | /* 1 means zero (many) matches is allowed. */ | |
2589 | char zero_times_ok = 0, many_times_ok = 0; | |
2590 | ||
2591 | /* If there is a sequence of repetition chars, collapse it | |
2592 | down to just one (the right one). We can't combine | |
2593 | interval operators with these because of, e.g., `a{2}*', | |
2594 | which should only match an even number of `a's. */ | |
2595 | ||
2596 | for (;;) | |
2597 | { | |
2598 | zero_times_ok |= c != '+'; | |
2599 | many_times_ok |= c != '?'; | |
2600 | ||
2601 | if (p == pend) | |
2602 | break; | |
2603 | ||
2604 | PATFETCH (c); | |
2605 | ||
2606 | if (c == '*' | |
2607 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
2608 | ; | |
2609 | ||
2610 | else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
2611 | { | |
2612 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2613 | ||
2614 | PATFETCH (c1); | |
2615 | if (!(c1 == '+' || c1 == '?')) | |
2616 | { | |
2617 | PATUNFETCH; | |
2618 | PATUNFETCH; | |
2619 | break; | |
2620 | } | |
2621 | ||
2622 | c = c1; | |
2623 | } | |
2624 | else | |
2625 | { | |
2626 | PATUNFETCH; | |
2627 | break; | |
2628 | } | |
2629 | ||
2630 | /* If we get here, we found another repeat character. */ | |
2631 | } | |
2632 | ||
2633 | /* Star, etc. applied to an empty pattern is equivalent | |
2634 | to an empty pattern. */ | |
2635 | if (!laststart) | |
2636 | break; | |
2637 | ||
2638 | /* Now we know whether or not zero matches is allowed | |
2639 | and also whether or not two or more matches is allowed. */ | |
2640 | if (many_times_ok) | |
2641 | { /* More than one repetition is allowed, so put in at the | |
2642 | end a backward relative jump from `b' to before the next | |
2643 | jump we're going to put in below (which jumps from | |
2644 | laststart to after this jump). | |
2645 | ||
2646 | But if we are at the `*' in the exact sequence `.*\n', | |
2647 | insert an unconditional jump backwards to the ., | |
2648 | instead of the beginning of the loop. This way we only | |
2649 | push a failure point once, instead of every time | |
2650 | through the loop. */ | |
2651 | assert (p - 1 > pattern); | |
2652 | ||
2653 | /* Allocate the space for the jump. */ | |
2654 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); | |
2655 | ||
2656 | /* We know we are not at the first character of the pattern, | |
2657 | because laststart was nonzero. And we've already | |
2658 | incremented `p', by the way, to be the character after | |
2659 | the `*'. Do we have to do something analogous here | |
2660 | for null bytes, because of RE_DOT_NOT_NULL? */ | |
2661 | if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
2662 | && zero_times_ok | |
2663 | && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') | |
2664 | && !(syntax & RE_DOT_NEWLINE)) | |
2665 | { /* We have .*\n. */ | |
2666 | STORE_JUMP (jump, b, laststart); | |
2667 | keep_string_p = true; | |
2668 | } | |
2669 | else | |
2670 | /* Anything else. */ | |
2671 | STORE_JUMP (maybe_pop_jump, b, laststart - | |
2672 | (1 + OFFSET_ADDRESS_SIZE)); | |
2673 | ||
2674 | /* We've added more stuff to the buffer. */ | |
2675 | b += 1 + OFFSET_ADDRESS_SIZE; | |
2676 | } | |
2677 | ||
2678 | /* On failure, jump from laststart to b + 3, which will be the | |
2679 | end of the buffer after this jump is inserted. */ | |
2680 | /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of | |
2681 | 'b + 3'. */ | |
2682 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); | |
2683 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
2684 | : on_failure_jump, | |
2685 | laststart, b + 1 + OFFSET_ADDRESS_SIZE); | |
2686 | pending_exact = 0; | |
2687 | b += 1 + OFFSET_ADDRESS_SIZE; | |
2688 | ||
2689 | if (!zero_times_ok) | |
2690 | { | |
2691 | /* At least one repetition is required, so insert a | |
2692 | `dummy_failure_jump' before the initial | |
2693 | `on_failure_jump' instruction of the loop. This | |
2694 | effects a skip over that instruction the first time | |
2695 | we hit that loop. */ | |
2696 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); | |
2697 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + | |
2698 | 2 + 2 * OFFSET_ADDRESS_SIZE); | |
2699 | b += 1 + OFFSET_ADDRESS_SIZE; | |
2700 | } | |
2701 | } | |
2702 | break; | |
2703 | ||
2704 | ||
2705 | case '.': | |
2706 | laststart = b; | |
2707 | BUF_PUSH (anychar); | |
2708 | break; | |
2709 | ||
2710 | ||
2711 | case '[': | |
2712 | { | |
2713 | boolean had_char_class = false; | |
2714 | #ifdef WCHAR | |
2715 | CHAR_T range_start = 0xffffffff; | |
2716 | #else | |
2717 | unsigned int range_start = 0xffffffff; | |
2718 | #endif | |
2719 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2720 | ||
2721 | #ifdef WCHAR | |
2722 | /* We assume a charset(_not) structure as a wchar_t array. | |
2723 | charset[0] = (re_opcode_t) charset(_not) | |
2724 | charset[1] = l (= length of char_classes) | |
2725 | charset[2] = m (= length of collating_symbols) | |
2726 | charset[3] = n (= length of equivalence_classes) | |
2727 | charset[4] = o (= length of char_ranges) | |
2728 | charset[5] = p (= length of chars) | |
2729 | ||
2730 | charset[6] = char_class (wctype_t) | |
2731 | charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t) | |
2732 | ... | |
2733 | charset[l+5] = char_class (wctype_t) | |
2734 | ||
2735 | charset[l+6] = collating_symbol (wchar_t) | |
2736 | ... | |
2737 | charset[l+m+5] = collating_symbol (wchar_t) | |
2738 | ifdef _LIBC we use the index if | |
2739 | _NL_COLLATE_SYMB_EXTRAMB instead of | |
2740 | wchar_t string. | |
2741 | ||
2742 | charset[l+m+6] = equivalence_classes (wchar_t) | |
2743 | ... | |
2744 | charset[l+m+n+5] = equivalence_classes (wchar_t) | |
2745 | ifdef _LIBC we use the index in | |
2746 | _NL_COLLATE_WEIGHT instead of | |
2747 | wchar_t string. | |
2748 | ||
2749 | charset[l+m+n+6] = range_start | |
2750 | charset[l+m+n+7] = range_end | |
2751 | ... | |
2752 | charset[l+m+n+2o+4] = range_start | |
2753 | charset[l+m+n+2o+5] = range_end | |
2754 | ifdef _LIBC we use the value looked up | |
2755 | in _NL_COLLATE_COLLSEQ instead of | |
2756 | wchar_t character. | |
2757 | ||
2758 | charset[l+m+n+2o+6] = char | |
2759 | ... | |
2760 | charset[l+m+n+2o+p+5] = char | |
2761 | ||
2762 | */ | |
2763 | ||
2764 | /* We need at least 6 spaces: the opcode, the length of | |
2765 | char_classes, the length of collating_symbols, the length of | |
2766 | equivalence_classes, the length of char_ranges, the length of | |
2767 | chars. */ | |
2768 | GET_BUFFER_SPACE (6); | |
2769 | ||
2770 | /* Save b as laststart. And We use laststart as the pointer | |
2771 | to the first element of the charset here. | |
2772 | In other words, laststart[i] indicates charset[i]. */ | |
2773 | laststart = b; | |
2774 | ||
2775 | /* We test `*p == '^' twice, instead of using an if | |
2776 | statement, so we only need one BUF_PUSH. */ | |
2777 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
2778 | if (*p == '^') | |
2779 | p++; | |
2780 | ||
2781 | /* Push the length of char_classes, the length of | |
2782 | collating_symbols, the length of equivalence_classes, the | |
2783 | length of char_ranges and the length of chars. */ | |
2784 | BUF_PUSH_3 (0, 0, 0); | |
2785 | BUF_PUSH_2 (0, 0); | |
2786 | ||
2787 | /* Remember the first position in the bracket expression. */ | |
2788 | p1 = p; | |
2789 | ||
2790 | /* charset_not matches newline according to a syntax bit. */ | |
2791 | if ((re_opcode_t) b[-6] == charset_not | |
2792 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2793 | { | |
2794 | BUF_PUSH('\n'); | |
2795 | laststart[5]++; /* Update the length of characters */ | |
2796 | } | |
2797 | ||
2798 | /* Read in characters and ranges, setting map bits. */ | |
2799 | for (;;) | |
2800 | { | |
2801 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2802 | ||
2803 | PATFETCH (c); | |
2804 | ||
2805 | /* \ might escape characters inside [...] and [^...]. */ | |
2806 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2807 | { | |
2808 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2809 | ||
2810 | PATFETCH (c1); | |
2811 | BUF_PUSH(c1); | |
2812 | laststart[5]++; /* Update the length of chars */ | |
2813 | range_start = c1; | |
2814 | continue; | |
2815 | } | |
2816 | ||
2817 | /* Could be the end of the bracket expression. If it's | |
2818 | not (i.e., when the bracket expression is `[]' so | |
2819 | far), the ']' character bit gets set way below. */ | |
2820 | if (c == ']' && p != p1 + 1) | |
2821 | break; | |
2822 | ||
2823 | /* Look ahead to see if it's a range when the last thing | |
2824 | was a character class. */ | |
2825 | if (had_char_class && c == '-' && *p != ']') | |
2826 | FREE_STACK_RETURN (REG_ERANGE); | |
2827 | ||
2828 | /* Look ahead to see if it's a range when the last thing | |
2829 | was a character: if this is a hyphen not at the | |
2830 | beginning or the end of a list, then it's the range | |
2831 | operator. */ | |
2832 | if (c == '-' | |
2833 | && !(p - 2 >= pattern && p[-2] == '[') | |
2834 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
2835 | && *p != ']') | |
2836 | { | |
2837 | reg_errcode_t ret; | |
2838 | /* Allocate the space for range_start and range_end. */ | |
2839 | GET_BUFFER_SPACE (2); | |
2840 | /* Update the pointer to indicate end of buffer. */ | |
2841 | b += 2; | |
2842 | ret = wcs_compile_range (range_start, &p, pend, translate, | |
2843 | syntax, b, laststart); | |
2844 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
2845 | range_start = 0xffffffff; | |
2846 | } | |
2847 | else if (p[0] == '-' && p[1] != ']') | |
2848 | { /* This handles ranges made up of characters only. */ | |
2849 | reg_errcode_t ret; | |
2850 | ||
2851 | /* Move past the `-'. */ | |
2852 | PATFETCH (c1); | |
2853 | /* Allocate the space for range_start and range_end. */ | |
2854 | GET_BUFFER_SPACE (2); | |
2855 | /* Update the pointer to indicate end of buffer. */ | |
2856 | b += 2; | |
2857 | ret = wcs_compile_range (c, &p, pend, translate, syntax, b, | |
2858 | laststart); | |
2859 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
2860 | range_start = 0xffffffff; | |
2861 | } | |
2862 | ||
2863 | /* See if we're at the beginning of a possible character | |
2864 | class. */ | |
2865 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
2866 | { /* Leave room for the null. */ | |
2867 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
2868 | ||
2869 | PATFETCH (c); | |
2870 | c1 = 0; | |
2871 | ||
2872 | /* If pattern is `[[:'. */ | |
2873 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2874 | ||
2875 | for (;;) | |
2876 | { | |
2877 | PATFETCH (c); | |
2878 | if ((c == ':' && *p == ']') || p == pend) | |
2879 | break; | |
2880 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
2881 | str[c1++] = c; | |
2882 | else | |
2883 | /* This is in any case an invalid class name. */ | |
2884 | str[0] = '\0'; | |
2885 | } | |
2886 | str[c1] = '\0'; | |
2887 | ||
2888 | /* If isn't a word bracketed by `[:' and `:]': | |
2889 | undo the ending character, the letters, and leave | |
2890 | the leading `:' and `[' (but store them as character). */ | |
2891 | if (c == ':' && *p == ']') | |
2892 | { | |
2893 | wctype_t wt; | |
2894 | uintptr_t alignedp; | |
2895 | ||
2896 | /* Query the character class as wctype_t. */ | |
2897 | wt = IS_CHAR_CLASS (str); | |
2898 | if (wt == 0) | |
2899 | FREE_STACK_RETURN (REG_ECTYPE); | |
2900 | ||
2901 | /* Throw away the ] at the end of the character | |
2902 | class. */ | |
2903 | PATFETCH (c); | |
2904 | ||
2905 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2906 | ||
2907 | /* Allocate the space for character class. */ | |
2908 | GET_BUFFER_SPACE(CHAR_CLASS_SIZE); | |
2909 | /* Update the pointer to indicate end of buffer. */ | |
2910 | b += CHAR_CLASS_SIZE; | |
2911 | /* Move data which follow character classes | |
2912 | not to violate the data. */ | |
2913 | insert_space(CHAR_CLASS_SIZE, | |
2914 | laststart + 6 + laststart[1], | |
2915 | b - 1); | |
2916 | alignedp = ((uintptr_t)(laststart + 6 + laststart[1]) | |
2917 | + __alignof__(wctype_t) - 1) | |
2918 | & ~(uintptr_t)(__alignof__(wctype_t) - 1); | |
2919 | /* Store the character class. */ | |
2920 | *((wctype_t*)alignedp) = wt; | |
2921 | /* Update length of char_classes */ | |
2922 | laststart[1] += CHAR_CLASS_SIZE; | |
2923 | ||
2924 | had_char_class = true; | |
2925 | } | |
2926 | else | |
2927 | { | |
2928 | c1++; | |
2929 | while (c1--) | |
2930 | PATUNFETCH; | |
2931 | BUF_PUSH ('['); | |
2932 | BUF_PUSH (':'); | |
2933 | laststart[5] += 2; /* Update the length of characters */ | |
2934 | range_start = ':'; | |
2935 | had_char_class = false; | |
2936 | } | |
2937 | } | |
2938 | else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '=' | |
2939 | || *p == '.')) | |
2940 | { | |
2941 | CHAR_T str[128]; /* Should be large enough. */ | |
2942 | CHAR_T delim = *p; /* '=' or '.' */ | |
2943 | # ifdef _LIBC | |
2944 | uint32_t nrules = | |
2945 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
2946 | # endif | |
2947 | PATFETCH (c); | |
2948 | c1 = 0; | |
2949 | ||
2950 | /* If pattern is `[[=' or '[[.'. */ | |
2951 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2952 | ||
2953 | for (;;) | |
2954 | { | |
2955 | PATFETCH (c); | |
2956 | if ((c == delim && *p == ']') || p == pend) | |
2957 | break; | |
2958 | if (c1 < sizeof (str) - 1) | |
2959 | str[c1++] = c; | |
2960 | else | |
2961 | /* This is in any case an invalid class name. */ | |
2962 | str[0] = '\0'; | |
2963 | } | |
2964 | str[c1] = '\0'; | |
2965 | ||
2966 | if (c == delim && *p == ']' && str[0] != '\0') | |
2967 | { | |
2968 | unsigned int i, offset; | |
2969 | /* If we have no collation data we use the default | |
2970 | collation in which each character is in a class | |
2971 | by itself. It also means that ASCII is the | |
2972 | character set and therefore we cannot have character | |
2973 | with more than one byte in the multibyte | |
2974 | representation. */ | |
2975 | ||
2976 | /* If not defined _LIBC, we push the name and | |
2977 | `\0' for the sake of matching performance. */ | |
2978 | int datasize = c1 + 1; | |
2979 | ||
2980 | # ifdef _LIBC | |
2981 | int32_t idx = 0; | |
2982 | if (nrules == 0) | |
2983 | # endif | |
2984 | { | |
2985 | if (c1 != 1) | |
2986 | FREE_STACK_RETURN (REG_ECOLLATE); | |
2987 | } | |
2988 | # ifdef _LIBC | |
2989 | else | |
2990 | { | |
2991 | const int32_t *table; | |
2992 | const int32_t *weights; | |
2993 | const int32_t *extra; | |
2994 | const int32_t *indirect; | |
2995 | wint_t *cp; | |
2996 | ||
2997 | /* This #include defines a local function! */ | |
2998 | # include <locale/weightwc.h> | |
2999 | ||
3000 | if(delim == '=') | |
3001 | { | |
3002 | /* We push the index for equivalence class. */ | |
3003 | cp = (wint_t*)str; | |
3004 | ||
3005 | table = (const int32_t *) | |
3006 | _NL_CURRENT (LC_COLLATE, | |
3007 | _NL_COLLATE_TABLEWC); | |
3008 | weights = (const int32_t *) | |
3009 | _NL_CURRENT (LC_COLLATE, | |
3010 | _NL_COLLATE_WEIGHTWC); | |
3011 | extra = (const int32_t *) | |
3012 | _NL_CURRENT (LC_COLLATE, | |
3013 | _NL_COLLATE_EXTRAWC); | |
3014 | indirect = (const int32_t *) | |
3015 | _NL_CURRENT (LC_COLLATE, | |
3016 | _NL_COLLATE_INDIRECTWC); | |
3017 | ||
3018 | idx = findidx ((const wint_t**)&cp); | |
3019 | if (idx == 0 || cp < (wint_t*) str + c1) | |
3020 | /* This is no valid character. */ | |
3021 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3022 | ||
3023 | str[0] = (wchar_t)idx; | |
3024 | } | |
3025 | else /* delim == '.' */ | |
3026 | { | |
3027 | /* We push collation sequence value | |
3028 | for collating symbol. */ | |
3029 | int32_t table_size; | |
3030 | const int32_t *symb_table; | |
3031 | const unsigned char *extra; | |
3032 | int32_t idx; | |
3033 | int32_t elem; | |
3034 | int32_t second; | |
3035 | int32_t hash; | |
3036 | char char_str[c1]; | |
3037 | ||
3038 | /* We have to convert the name to a single-byte | |
3039 | string. This is possible since the names | |
3040 | consist of ASCII characters and the internal | |
3041 | representation is UCS4. */ | |
3042 | for (i = 0; i < c1; ++i) | |
3043 | char_str[i] = str[i]; | |
3044 | ||
3045 | table_size = | |
3046 | _NL_CURRENT_WORD (LC_COLLATE, | |
3047 | _NL_COLLATE_SYMB_HASH_SIZEMB); | |
3048 | symb_table = (const int32_t *) | |
3049 | _NL_CURRENT (LC_COLLATE, | |
3050 | _NL_COLLATE_SYMB_TABLEMB); | |
3051 | extra = (const unsigned char *) | |
3052 | _NL_CURRENT (LC_COLLATE, | |
3053 | _NL_COLLATE_SYMB_EXTRAMB); | |
3054 | ||
3055 | /* Locate the character in the hashing table. */ | |
3056 | hash = elem_hash (char_str, c1); | |
3057 | ||
3058 | idx = 0; | |
3059 | elem = hash % table_size; | |
3060 | second = hash % (table_size - 2); | |
3061 | while (symb_table[2 * elem] != 0) | |
3062 | { | |
3063 | /* First compare the hashing value. */ | |
3064 | if (symb_table[2 * elem] == hash | |
3065 | && c1 == extra[symb_table[2 * elem + 1]] | |
a9022147 | 3066 | && memcmp (char_str, |
2a6ef469 DD |
3067 | &extra[symb_table[2 * elem + 1] |
3068 | + 1], c1) == 0) | |
3069 | { | |
3070 | /* Yep, this is the entry. */ | |
3071 | idx = symb_table[2 * elem + 1]; | |
3072 | idx += 1 + extra[idx]; | |
3073 | break; | |
3074 | } | |
3075 | ||
3076 | /* Next entry. */ | |
3077 | elem += second; | |
3078 | } | |
3079 | ||
3080 | if (symb_table[2 * elem] != 0) | |
3081 | { | |
3082 | /* Compute the index of the byte sequence | |
3083 | in the table. */ | |
3084 | idx += 1 + extra[idx]; | |
3085 | /* Adjust for the alignment. */ | |
a9022147 | 3086 | idx = (idx + 3) & ~3; |
2a6ef469 DD |
3087 | |
3088 | str[0] = (wchar_t) idx + 4; | |
3089 | } | |
3090 | else if (symb_table[2 * elem] == 0 && c1 == 1) | |
3091 | { | |
3092 | /* No valid character. Match it as a | |
3093 | single byte character. */ | |
3094 | had_char_class = false; | |
3095 | BUF_PUSH(str[0]); | |
3096 | /* Update the length of characters */ | |
3097 | laststart[5]++; | |
3098 | range_start = str[0]; | |
3099 | ||
3100 | /* Throw away the ] at the end of the | |
3101 | collating symbol. */ | |
3102 | PATFETCH (c); | |
3103 | /* exit from the switch block. */ | |
3104 | continue; | |
3105 | } | |
3106 | else | |
3107 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3108 | } | |
3109 | datasize = 1; | |
3110 | } | |
3111 | # endif | |
3112 | /* Throw away the ] at the end of the equivalence | |
3113 | class (or collating symbol). */ | |
3114 | PATFETCH (c); | |
3115 | ||
3116 | /* Allocate the space for the equivalence class | |
3117 | (or collating symbol) (and '\0' if needed). */ | |
3118 | GET_BUFFER_SPACE(datasize); | |
3119 | /* Update the pointer to indicate end of buffer. */ | |
3120 | b += datasize; | |
3121 | ||
3122 | if (delim == '=') | |
3123 | { /* equivalence class */ | |
3124 | /* Calculate the offset of char_ranges, | |
3125 | which is next to equivalence_classes. */ | |
3126 | offset = laststart[1] + laststart[2] | |
3127 | + laststart[3] +6; | |
3128 | /* Insert space. */ | |
3129 | insert_space(datasize, laststart + offset, b - 1); | |
3130 | ||
3131 | /* Write the equivalence_class and \0. */ | |
3132 | for (i = 0 ; i < datasize ; i++) | |
3133 | laststart[offset + i] = str[i]; | |
3134 | ||
3135 | /* Update the length of equivalence_classes. */ | |
3136 | laststart[3] += datasize; | |
3137 | had_char_class = true; | |
3138 | } | |
3139 | else /* delim == '.' */ | |
3140 | { /* collating symbol */ | |
3141 | /* Calculate the offset of the equivalence_classes, | |
3142 | which is next to collating_symbols. */ | |
3143 | offset = laststart[1] + laststart[2] + 6; | |
3144 | /* Insert space and write the collationg_symbol | |
3145 | and \0. */ | |
3146 | insert_space(datasize, laststart + offset, b-1); | |
3147 | for (i = 0 ; i < datasize ; i++) | |
3148 | laststart[offset + i] = str[i]; | |
3149 | ||
3150 | /* In re_match_2_internal if range_start < -1, we | |
3151 | assume -range_start is the offset of the | |
3152 | collating symbol which is specified as | |
3153 | the character of the range start. So we assign | |
3154 | -(laststart[1] + laststart[2] + 6) to | |
3155 | range_start. */ | |
3156 | range_start = -(laststart[1] + laststart[2] + 6); | |
3157 | /* Update the length of collating_symbol. */ | |
3158 | laststart[2] += datasize; | |
3159 | had_char_class = false; | |
3160 | } | |
3161 | } | |
3162 | else | |
3163 | { | |
3164 | c1++; | |
3165 | while (c1--) | |
3166 | PATUNFETCH; | |
3167 | BUF_PUSH ('['); | |
3168 | BUF_PUSH (delim); | |
3169 | laststart[5] += 2; /* Update the length of characters */ | |
3170 | range_start = delim; | |
3171 | had_char_class = false; | |
3172 | } | |
3173 | } | |
3174 | else | |
3175 | { | |
3176 | had_char_class = false; | |
3177 | BUF_PUSH(c); | |
3178 | laststart[5]++; /* Update the length of characters */ | |
3179 | range_start = c; | |
3180 | } | |
3181 | } | |
3182 | ||
3183 | #else /* BYTE */ | |
3184 | /* Ensure that we have enough space to push a charset: the | |
3185 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
3186 | GET_BUFFER_SPACE (34); | |
3187 | ||
3188 | laststart = b; | |
3189 | ||
3190 | /* We test `*p == '^' twice, instead of using an if | |
3191 | statement, so we only need one BUF_PUSH. */ | |
3192 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
3193 | if (*p == '^') | |
3194 | p++; | |
3195 | ||
3196 | /* Remember the first position in the bracket expression. */ | |
3197 | p1 = p; | |
3198 | ||
3199 | /* Push the number of bytes in the bitmap. */ | |
3200 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
3201 | ||
3202 | /* Clear the whole map. */ | |
3203 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
3204 | ||
3205 | /* charset_not matches newline according to a syntax bit. */ | |
3206 | if ((re_opcode_t) b[-2] == charset_not | |
3207 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
3208 | SET_LIST_BIT ('\n'); | |
3209 | ||
3210 | /* Read in characters and ranges, setting map bits. */ | |
3211 | for (;;) | |
3212 | { | |
3213 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3214 | ||
3215 | PATFETCH (c); | |
3216 | ||
3217 | /* \ might escape characters inside [...] and [^...]. */ | |
3218 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
3219 | { | |
3220 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
3221 | ||
3222 | PATFETCH (c1); | |
3223 | SET_LIST_BIT (c1); | |
3224 | range_start = c1; | |
3225 | continue; | |
3226 | } | |
3227 | ||
3228 | /* Could be the end of the bracket expression. If it's | |
3229 | not (i.e., when the bracket expression is `[]' so | |
3230 | far), the ']' character bit gets set way below. */ | |
3231 | if (c == ']' && p != p1 + 1) | |
3232 | break; | |
3233 | ||
3234 | /* Look ahead to see if it's a range when the last thing | |
3235 | was a character class. */ | |
3236 | if (had_char_class && c == '-' && *p != ']') | |
3237 | FREE_STACK_RETURN (REG_ERANGE); | |
3238 | ||
3239 | /* Look ahead to see if it's a range when the last thing | |
3240 | was a character: if this is a hyphen not at the | |
3241 | beginning or the end of a list, then it's the range | |
3242 | operator. */ | |
3243 | if (c == '-' | |
3244 | && !(p - 2 >= pattern && p[-2] == '[') | |
3245 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
3246 | && *p != ']') | |
3247 | { | |
3248 | reg_errcode_t ret | |
3249 | = byte_compile_range (range_start, &p, pend, translate, | |
3250 | syntax, b); | |
3251 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
3252 | range_start = 0xffffffff; | |
3253 | } | |
3254 | ||
3255 | else if (p[0] == '-' && p[1] != ']') | |
3256 | { /* This handles ranges made up of characters only. */ | |
3257 | reg_errcode_t ret; | |
3258 | ||
3259 | /* Move past the `-'. */ | |
3260 | PATFETCH (c1); | |
3261 | ||
3262 | ret = byte_compile_range (c, &p, pend, translate, syntax, b); | |
3263 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
3264 | range_start = 0xffffffff; | |
3265 | } | |
3266 | ||
3267 | /* See if we're at the beginning of a possible character | |
3268 | class. */ | |
3269 | ||
3270 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
3271 | { /* Leave room for the null. */ | |
3272 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
3273 | ||
3274 | PATFETCH (c); | |
3275 | c1 = 0; | |
3276 | ||
3277 | /* If pattern is `[[:'. */ | |
3278 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3279 | ||
3280 | for (;;) | |
3281 | { | |
3282 | PATFETCH (c); | |
3283 | if ((c == ':' && *p == ']') || p == pend) | |
3284 | break; | |
3285 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
3286 | str[c1++] = c; | |
3287 | else | |
3288 | /* This is in any case an invalid class name. */ | |
3289 | str[0] = '\0'; | |
3290 | } | |
3291 | str[c1] = '\0'; | |
3292 | ||
3293 | /* If isn't a word bracketed by `[:' and `:]': | |
3294 | undo the ending character, the letters, and leave | |
3295 | the leading `:' and `[' (but set bits for them). */ | |
3296 | if (c == ':' && *p == ']') | |
3297 | { | |
3298 | # if defined _LIBC || WIDE_CHAR_SUPPORT | |
3299 | boolean is_lower = STREQ (str, "lower"); | |
3300 | boolean is_upper = STREQ (str, "upper"); | |
3301 | wctype_t wt; | |
3302 | int ch; | |
3303 | ||
3304 | wt = IS_CHAR_CLASS (str); | |
3305 | if (wt == 0) | |
3306 | FREE_STACK_RETURN (REG_ECTYPE); | |
3307 | ||
3308 | /* Throw away the ] at the end of the character | |
3309 | class. */ | |
3310 | PATFETCH (c); | |
3311 | ||
3312 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3313 | ||
3314 | for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) | |
3315 | { | |
3316 | # ifdef _LIBC | |
3317 | if (__iswctype (__btowc (ch), wt)) | |
3318 | SET_LIST_BIT (ch); | |
3319 | # else | |
3320 | if (iswctype (btowc (ch), wt)) | |
3321 | SET_LIST_BIT (ch); | |
3322 | # endif | |
3323 | ||
3324 | if (translate && (is_upper || is_lower) | |
3325 | && (ISUPPER (ch) || ISLOWER (ch))) | |
3326 | SET_LIST_BIT (ch); | |
3327 | } | |
3328 | ||
3329 | had_char_class = true; | |
3330 | # else | |
3331 | int ch; | |
3332 | boolean is_alnum = STREQ (str, "alnum"); | |
3333 | boolean is_alpha = STREQ (str, "alpha"); | |
3334 | boolean is_blank = STREQ (str, "blank"); | |
3335 | boolean is_cntrl = STREQ (str, "cntrl"); | |
3336 | boolean is_digit = STREQ (str, "digit"); | |
3337 | boolean is_graph = STREQ (str, "graph"); | |
3338 | boolean is_lower = STREQ (str, "lower"); | |
3339 | boolean is_print = STREQ (str, "print"); | |
3340 | boolean is_punct = STREQ (str, "punct"); | |
3341 | boolean is_space = STREQ (str, "space"); | |
3342 | boolean is_upper = STREQ (str, "upper"); | |
3343 | boolean is_xdigit = STREQ (str, "xdigit"); | |
3344 | ||
3345 | if (!IS_CHAR_CLASS (str)) | |
3346 | FREE_STACK_RETURN (REG_ECTYPE); | |
3347 | ||
3348 | /* Throw away the ] at the end of the character | |
3349 | class. */ | |
3350 | PATFETCH (c); | |
3351 | ||
3352 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3353 | ||
3354 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
3355 | { | |
3356 | /* This was split into 3 if's to | |
3357 | avoid an arbitrary limit in some compiler. */ | |
3358 | if ( (is_alnum && ISALNUM (ch)) | |
3359 | || (is_alpha && ISALPHA (ch)) | |
3360 | || (is_blank && ISBLANK (ch)) | |
3361 | || (is_cntrl && ISCNTRL (ch))) | |
3362 | SET_LIST_BIT (ch); | |
3363 | if ( (is_digit && ISDIGIT (ch)) | |
3364 | || (is_graph && ISGRAPH (ch)) | |
3365 | || (is_lower && ISLOWER (ch)) | |
3366 | || (is_print && ISPRINT (ch))) | |
3367 | SET_LIST_BIT (ch); | |
3368 | if ( (is_punct && ISPUNCT (ch)) | |
3369 | || (is_space && ISSPACE (ch)) | |
3370 | || (is_upper && ISUPPER (ch)) | |
3371 | || (is_xdigit && ISXDIGIT (ch))) | |
3372 | SET_LIST_BIT (ch); | |
3373 | if ( translate && (is_upper || is_lower) | |
3374 | && (ISUPPER (ch) || ISLOWER (ch))) | |
3375 | SET_LIST_BIT (ch); | |
3376 | } | |
3377 | had_char_class = true; | |
3378 | # endif /* libc || wctype.h */ | |
3379 | } | |
3380 | else | |
3381 | { | |
3382 | c1++; | |
3383 | while (c1--) | |
3384 | PATUNFETCH; | |
3385 | SET_LIST_BIT ('['); | |
3386 | SET_LIST_BIT (':'); | |
3387 | range_start = ':'; | |
3388 | had_char_class = false; | |
3389 | } | |
3390 | } | |
3391 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=') | |
3392 | { | |
3393 | unsigned char str[MB_LEN_MAX + 1]; | |
3394 | # ifdef _LIBC | |
3395 | uint32_t nrules = | |
3396 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
3397 | # endif | |
3398 | ||
3399 | PATFETCH (c); | |
3400 | c1 = 0; | |
3401 | ||
3402 | /* If pattern is `[[='. */ | |
3403 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3404 | ||
3405 | for (;;) | |
3406 | { | |
3407 | PATFETCH (c); | |
3408 | if ((c == '=' && *p == ']') || p == pend) | |
3409 | break; | |
3410 | if (c1 < MB_LEN_MAX) | |
3411 | str[c1++] = c; | |
3412 | else | |
3413 | /* This is in any case an invalid class name. */ | |
3414 | str[0] = '\0'; | |
3415 | } | |
3416 | str[c1] = '\0'; | |
3417 | ||
3418 | if (c == '=' && *p == ']' && str[0] != '\0') | |
3419 | { | |
3420 | /* If we have no collation data we use the default | |
3421 | collation in which each character is in a class | |
3422 | by itself. It also means that ASCII is the | |
3423 | character set and therefore we cannot have character | |
3424 | with more than one byte in the multibyte | |
3425 | representation. */ | |
3426 | # ifdef _LIBC | |
3427 | if (nrules == 0) | |
3428 | # endif | |
3429 | { | |
3430 | if (c1 != 1) | |
3431 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3432 | ||
3433 | /* Throw away the ] at the end of the equivalence | |
3434 | class. */ | |
3435 | PATFETCH (c); | |
3436 | ||
3437 | /* Set the bit for the character. */ | |
3438 | SET_LIST_BIT (str[0]); | |
3439 | } | |
3440 | # ifdef _LIBC | |
3441 | else | |
3442 | { | |
3443 | /* Try to match the byte sequence in `str' against | |
3444 | those known to the collate implementation. | |
3445 | First find out whether the bytes in `str' are | |
3446 | actually from exactly one character. */ | |
3447 | const int32_t *table; | |
3448 | const unsigned char *weights; | |
3449 | const unsigned char *extra; | |
3450 | const int32_t *indirect; | |
3451 | int32_t idx; | |
3452 | const unsigned char *cp = str; | |
3453 | int ch; | |
3454 | ||
3455 | /* This #include defines a local function! */ | |
3456 | # include <locale/weight.h> | |
3457 | ||
3458 | table = (const int32_t *) | |
3459 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB); | |
3460 | weights = (const unsigned char *) | |
3461 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB); | |
3462 | extra = (const unsigned char *) | |
3463 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB); | |
3464 | indirect = (const int32_t *) | |
3465 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB); | |
3466 | ||
3467 | idx = findidx (&cp); | |
3468 | if (idx == 0 || cp < str + c1) | |
3469 | /* This is no valid character. */ | |
3470 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3471 | ||
3472 | /* Throw away the ] at the end of the equivalence | |
3473 | class. */ | |
3474 | PATFETCH (c); | |
3475 | ||
3476 | /* Now we have to go throught the whole table | |
3477 | and find all characters which have the same | |
3478 | first level weight. | |
3479 | ||
3480 | XXX Note that this is not entirely correct. | |
3481 | we would have to match multibyte sequences | |
3482 | but this is not possible with the current | |
3483 | implementation. */ | |
3484 | for (ch = 1; ch < 256; ++ch) | |
3485 | /* XXX This test would have to be changed if we | |
3486 | would allow matching multibyte sequences. */ | |
3487 | if (table[ch] > 0) | |
3488 | { | |
3489 | int32_t idx2 = table[ch]; | |
3490 | size_t len = weights[idx2]; | |
3491 | ||
3492 | /* Test whether the lenghts match. */ | |
3493 | if (weights[idx] == len) | |
3494 | { | |
3495 | /* They do. New compare the bytes of | |
3496 | the weight. */ | |
3497 | size_t cnt = 0; | |
3498 | ||
3499 | while (cnt < len | |
3500 | && (weights[idx + 1 + cnt] | |
3501 | == weights[idx2 + 1 + cnt])) | |
3502 | ++cnt; | |
3503 | ||
3504 | if (cnt == len) | |
3505 | /* They match. Mark the character as | |
3506 | acceptable. */ | |
3507 | SET_LIST_BIT (ch); | |
3508 | } | |
3509 | } | |
3510 | } | |
3511 | # endif | |
3512 | had_char_class = true; | |
3513 | } | |
3514 | else | |
3515 | { | |
3516 | c1++; | |
3517 | while (c1--) | |
3518 | PATUNFETCH; | |
3519 | SET_LIST_BIT ('['); | |
3520 | SET_LIST_BIT ('='); | |
3521 | range_start = '='; | |
3522 | had_char_class = false; | |
3523 | } | |
3524 | } | |
3525 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.') | |
3526 | { | |
3527 | unsigned char str[128]; /* Should be large enough. */ | |
3528 | # ifdef _LIBC | |
3529 | uint32_t nrules = | |
3530 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
3531 | # endif | |
3532 | ||
3533 | PATFETCH (c); | |
3534 | c1 = 0; | |
3535 | ||
3536 | /* If pattern is `[[.'. */ | |
3537 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3538 | ||
3539 | for (;;) | |
3540 | { | |
3541 | PATFETCH (c); | |
3542 | if ((c == '.' && *p == ']') || p == pend) | |
3543 | break; | |
3544 | if (c1 < sizeof (str)) | |
3545 | str[c1++] = c; | |
3546 | else | |
3547 | /* This is in any case an invalid class name. */ | |
3548 | str[0] = '\0'; | |
3549 | } | |
3550 | str[c1] = '\0'; | |
3551 | ||
3552 | if (c == '.' && *p == ']' && str[0] != '\0') | |
3553 | { | |
3554 | /* If we have no collation data we use the default | |
3555 | collation in which each character is the name | |
3556 | for its own class which contains only the one | |
3557 | character. It also means that ASCII is the | |
3558 | character set and therefore we cannot have character | |
3559 | with more than one byte in the multibyte | |
3560 | representation. */ | |
3561 | # ifdef _LIBC | |
3562 | if (nrules == 0) | |
3563 | # endif | |
3564 | { | |
3565 | if (c1 != 1) | |
3566 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3567 | ||
3568 | /* Throw away the ] at the end of the equivalence | |
3569 | class. */ | |
3570 | PATFETCH (c); | |
3571 | ||
3572 | /* Set the bit for the character. */ | |
3573 | SET_LIST_BIT (str[0]); | |
3574 | range_start = ((const unsigned char *) str)[0]; | |
3575 | } | |
3576 | # ifdef _LIBC | |
3577 | else | |
3578 | { | |
3579 | /* Try to match the byte sequence in `str' against | |
3580 | those known to the collate implementation. | |
3581 | First find out whether the bytes in `str' are | |
3582 | actually from exactly one character. */ | |
3583 | int32_t table_size; | |
3584 | const int32_t *symb_table; | |
3585 | const unsigned char *extra; | |
3586 | int32_t idx; | |
3587 | int32_t elem; | |
3588 | int32_t second; | |
3589 | int32_t hash; | |
3590 | ||
3591 | table_size = | |
3592 | _NL_CURRENT_WORD (LC_COLLATE, | |
3593 | _NL_COLLATE_SYMB_HASH_SIZEMB); | |
3594 | symb_table = (const int32_t *) | |
3595 | _NL_CURRENT (LC_COLLATE, | |
3596 | _NL_COLLATE_SYMB_TABLEMB); | |
3597 | extra = (const unsigned char *) | |
3598 | _NL_CURRENT (LC_COLLATE, | |
3599 | _NL_COLLATE_SYMB_EXTRAMB); | |
3600 | ||
3601 | /* Locate the character in the hashing table. */ | |
3602 | hash = elem_hash (str, c1); | |
3603 | ||
3604 | idx = 0; | |
3605 | elem = hash % table_size; | |
3606 | second = hash % (table_size - 2); | |
3607 | while (symb_table[2 * elem] != 0) | |
3608 | { | |
3609 | /* First compare the hashing value. */ | |
3610 | if (symb_table[2 * elem] == hash | |
3611 | && c1 == extra[symb_table[2 * elem + 1]] | |
3612 | && memcmp (str, | |
3613 | &extra[symb_table[2 * elem + 1] | |
3614 | + 1], | |
3615 | c1) == 0) | |
3616 | { | |
3617 | /* Yep, this is the entry. */ | |
3618 | idx = symb_table[2 * elem + 1]; | |
3619 | idx += 1 + extra[idx]; | |
3620 | break; | |
3621 | } | |
3622 | ||
3623 | /* Next entry. */ | |
3624 | elem += second; | |
3625 | } | |
3626 | ||
3627 | if (symb_table[2 * elem] == 0) | |
3628 | /* This is no valid character. */ | |
3629 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3630 | ||
3631 | /* Throw away the ] at the end of the equivalence | |
3632 | class. */ | |
3633 | PATFETCH (c); | |
3634 | ||
3635 | /* Now add the multibyte character(s) we found | |
3636 | to the accept list. | |
3637 | ||
3638 | XXX Note that this is not entirely correct. | |
3639 | we would have to match multibyte sequences | |
3640 | but this is not possible with the current | |
3641 | implementation. Also, we have to match | |
3642 | collating symbols, which expand to more than | |
3643 | one file, as a whole and not allow the | |
3644 | individual bytes. */ | |
3645 | c1 = extra[idx++]; | |
3646 | if (c1 == 1) | |
3647 | range_start = extra[idx]; | |
3648 | while (c1-- > 0) | |
3649 | { | |
3650 | SET_LIST_BIT (extra[idx]); | |
3651 | ++idx; | |
3652 | } | |
3653 | } | |
3654 | # endif | |
3655 | had_char_class = false; | |
3656 | } | |
3657 | else | |
3658 | { | |
3659 | c1++; | |
3660 | while (c1--) | |
3661 | PATUNFETCH; | |
3662 | SET_LIST_BIT ('['); | |
3663 | SET_LIST_BIT ('.'); | |
3664 | range_start = '.'; | |
3665 | had_char_class = false; | |
3666 | } | |
3667 | } | |
3668 | else | |
3669 | { | |
3670 | had_char_class = false; | |
3671 | SET_LIST_BIT (c); | |
3672 | range_start = c; | |
3673 | } | |
3674 | } | |
3675 | ||
3676 | /* Discard any (non)matching list bytes that are all 0 at the | |
3677 | end of the map. Decrease the map-length byte too. */ | |
3678 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
3679 | b[-1]--; | |
3680 | b += b[-1]; | |
3681 | #endif /* WCHAR */ | |
3682 | } | |
3683 | break; | |
3684 | ||
3685 | ||
3686 | case '(': | |
3687 | if (syntax & RE_NO_BK_PARENS) | |
3688 | goto handle_open; | |
3689 | else | |
3690 | goto normal_char; | |
3691 | ||
3692 | ||
3693 | case ')': | |
3694 | if (syntax & RE_NO_BK_PARENS) | |
3695 | goto handle_close; | |
3696 | else | |
3697 | goto normal_char; | |
3698 | ||
3699 | ||
3700 | case '\n': | |
3701 | if (syntax & RE_NEWLINE_ALT) | |
3702 | goto handle_alt; | |
3703 | else | |
3704 | goto normal_char; | |
3705 | ||
3706 | ||
3707 | case '|': | |
3708 | if (syntax & RE_NO_BK_VBAR) | |
3709 | goto handle_alt; | |
3710 | else | |
3711 | goto normal_char; | |
3712 | ||
3713 | ||
3714 | case '{': | |
3715 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
3716 | goto handle_interval; | |
3717 | else | |
3718 | goto normal_char; | |
3719 | ||
3720 | ||
3721 | case '\\': | |
3722 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
3723 | ||
3724 | /* Do not translate the character after the \, so that we can | |
3725 | distinguish, e.g., \B from \b, even if we normally would | |
3726 | translate, e.g., B to b. */ | |
3727 | PATFETCH_RAW (c); | |
3728 | ||
3729 | switch (c) | |
3730 | { | |
3731 | case '(': | |
3732 | if (syntax & RE_NO_BK_PARENS) | |
3733 | goto normal_backslash; | |
3734 | ||
3735 | handle_open: | |
3736 | bufp->re_nsub++; | |
3737 | regnum++; | |
3738 | ||
3739 | if (COMPILE_STACK_FULL) | |
3740 | { | |
3741 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
3742 | compile_stack_elt_t); | |
3743 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
3744 | ||
3745 | compile_stack.size <<= 1; | |
3746 | } | |
3747 | ||
3748 | /* These are the values to restore when we hit end of this | |
3749 | group. They are all relative offsets, so that if the | |
3750 | whole pattern moves because of realloc, they will still | |
3751 | be valid. */ | |
3752 | COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR; | |
3753 | COMPILE_STACK_TOP.fixup_alt_jump | |
3754 | = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0; | |
3755 | COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR; | |
3756 | COMPILE_STACK_TOP.regnum = regnum; | |
3757 | ||
3758 | /* We will eventually replace the 0 with the number of | |
3759 | groups inner to this one. But do not push a | |
3760 | start_memory for groups beyond the last one we can | |
3761 | represent in the compiled pattern. */ | |
3762 | if (regnum <= MAX_REGNUM) | |
3763 | { | |
3764 | COMPILE_STACK_TOP.inner_group_offset = b | |
3765 | - COMPILED_BUFFER_VAR + 2; | |
3766 | BUF_PUSH_3 (start_memory, regnum, 0); | |
3767 | } | |
3768 | ||
3769 | compile_stack.avail++; | |
3770 | ||
3771 | fixup_alt_jump = 0; | |
3772 | laststart = 0; | |
3773 | begalt = b; | |
3774 | /* If we've reached MAX_REGNUM groups, then this open | |
3775 | won't actually generate any code, so we'll have to | |
3776 | clear pending_exact explicitly. */ | |
3777 | pending_exact = 0; | |
3778 | break; | |
3779 | ||
3780 | ||
3781 | case ')': | |
3782 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
3783 | ||
3784 | if (COMPILE_STACK_EMPTY) | |
3785 | { | |
3786 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3787 | goto normal_backslash; | |
3788 | else | |
3789 | FREE_STACK_RETURN (REG_ERPAREN); | |
3790 | } | |
3791 | ||
3792 | handle_close: | |
3793 | if (fixup_alt_jump) | |
3794 | { /* Push a dummy failure point at the end of the | |
3795 | alternative for a possible future | |
3796 | `pop_failure_jump' to pop. See comments at | |
3797 | `push_dummy_failure' in `re_match_2'. */ | |
3798 | BUF_PUSH (push_dummy_failure); | |
3799 | ||
3800 | /* We allocated space for this jump when we assigned | |
3801 | to `fixup_alt_jump', in the `handle_alt' case below. */ | |
3802 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
3803 | } | |
3804 | ||
3805 | /* See similar code for backslashed left paren above. */ | |
3806 | if (COMPILE_STACK_EMPTY) | |
3807 | { | |
3808 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3809 | goto normal_char; | |
3810 | else | |
3811 | FREE_STACK_RETURN (REG_ERPAREN); | |
3812 | } | |
3813 | ||
3814 | /* Since we just checked for an empty stack above, this | |
3815 | ``can't happen''. */ | |
3816 | assert (compile_stack.avail != 0); | |
3817 | { | |
3818 | /* We don't just want to restore into `regnum', because | |
3819 | later groups should continue to be numbered higher, | |
3820 | as in `(ab)c(de)' -- the second group is #2. */ | |
3821 | regnum_t this_group_regnum; | |
3822 | ||
3823 | compile_stack.avail--; | |
3824 | begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset; | |
3825 | fixup_alt_jump | |
3826 | = COMPILE_STACK_TOP.fixup_alt_jump | |
3827 | ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
3828 | : 0; | |
3829 | laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset; | |
3830 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
3831 | /* If we've reached MAX_REGNUM groups, then this open | |
3832 | won't actually generate any code, so we'll have to | |
3833 | clear pending_exact explicitly. */ | |
3834 | pending_exact = 0; | |
3835 | ||
3836 | /* We're at the end of the group, so now we know how many | |
3837 | groups were inside this one. */ | |
3838 | if (this_group_regnum <= MAX_REGNUM) | |
3839 | { | |
3840 | UCHAR_T *inner_group_loc | |
3841 | = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset; | |
3842 | ||
3843 | *inner_group_loc = regnum - this_group_regnum; | |
3844 | BUF_PUSH_3 (stop_memory, this_group_regnum, | |
3845 | regnum - this_group_regnum); | |
3846 | } | |
3847 | } | |
3848 | break; | |
3849 | ||
3850 | ||
3851 | case '|': /* `\|'. */ | |
3852 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
3853 | goto normal_backslash; | |
3854 | handle_alt: | |
3855 | if (syntax & RE_LIMITED_OPS) | |
3856 | goto normal_char; | |
3857 | ||
3858 | /* Insert before the previous alternative a jump which | |
3859 | jumps to this alternative if the former fails. */ | |
3860 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); | |
3861 | INSERT_JUMP (on_failure_jump, begalt, | |
3862 | b + 2 + 2 * OFFSET_ADDRESS_SIZE); | |
3863 | pending_exact = 0; | |
3864 | b += 1 + OFFSET_ADDRESS_SIZE; | |
3865 | ||
3866 | /* The alternative before this one has a jump after it | |
3867 | which gets executed if it gets matched. Adjust that | |
3868 | jump so it will jump to this alternative's analogous | |
3869 | jump (put in below, which in turn will jump to the next | |
3870 | (if any) alternative's such jump, etc.). The last such | |
3871 | jump jumps to the correct final destination. A picture: | |
3872 | _____ _____ | |
3873 | | | | | | |
3874 | | v | v | |
3875 | a | b | c | |
3876 | ||
3877 | If we are at `b', then fixup_alt_jump right now points to a | |
3878 | three-byte space after `a'. We'll put in the jump, set | |
3879 | fixup_alt_jump to right after `b', and leave behind three | |
3880 | bytes which we'll fill in when we get to after `c'. */ | |
3881 | ||
3882 | if (fixup_alt_jump) | |
3883 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
3884 | ||
3885 | /* Mark and leave space for a jump after this alternative, | |
3886 | to be filled in later either by next alternative or | |
3887 | when know we're at the end of a series of alternatives. */ | |
3888 | fixup_alt_jump = b; | |
3889 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); | |
3890 | b += 1 + OFFSET_ADDRESS_SIZE; | |
3891 | ||
3892 | laststart = 0; | |
3893 | begalt = b; | |
3894 | break; | |
3895 | ||
3896 | ||
3897 | case '{': | |
3898 | /* If \{ is a literal. */ | |
3899 | if (!(syntax & RE_INTERVALS) | |
3900 | /* If we're at `\{' and it's not the open-interval | |
3901 | operator. */ | |
3902 | || (syntax & RE_NO_BK_BRACES)) | |
3903 | goto normal_backslash; | |
3904 | ||
3905 | handle_interval: | |
3906 | { | |
3907 | /* If got here, then the syntax allows intervals. */ | |
3908 | ||
3909 | /* At least (most) this many matches must be made. */ | |
3910 | int lower_bound = -1, upper_bound = -1; | |
3911 | ||
3912 | /* Place in the uncompiled pattern (i.e., just after | |
3913 | the '{') to go back to if the interval is invalid. */ | |
3914 | const CHAR_T *beg_interval = p; | |
3915 | ||
3916 | if (p == pend) | |
3917 | goto invalid_interval; | |
3918 | ||
3919 | GET_UNSIGNED_NUMBER (lower_bound); | |
3920 | ||
3921 | if (c == ',') | |
3922 | { | |
3923 | GET_UNSIGNED_NUMBER (upper_bound); | |
3924 | if (upper_bound < 0) | |
3925 | upper_bound = RE_DUP_MAX; | |
3926 | } | |
3927 | else | |
3928 | /* Interval such as `{1}' => match exactly once. */ | |
3929 | upper_bound = lower_bound; | |
3930 | ||
3931 | if (! (0 <= lower_bound && lower_bound <= upper_bound)) | |
3932 | goto invalid_interval; | |
3933 | ||
3934 | if (!(syntax & RE_NO_BK_BRACES)) | |
3935 | { | |
3936 | if (c != '\\' || p == pend) | |
3937 | goto invalid_interval; | |
3938 | PATFETCH (c); | |
3939 | } | |
3940 | ||
3941 | if (c != '}') | |
3942 | goto invalid_interval; | |
3943 | ||
3944 | /* If it's invalid to have no preceding re. */ | |
3945 | if (!laststart) | |
3946 | { | |
3947 | if (syntax & RE_CONTEXT_INVALID_OPS | |
3948 | && !(syntax & RE_INVALID_INTERVAL_ORD)) | |
3949 | FREE_STACK_RETURN (REG_BADRPT); | |
3950 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
3951 | laststart = b; | |
3952 | else | |
3953 | goto unfetch_interval; | |
3954 | } | |
3955 | ||
3956 | /* We just parsed a valid interval. */ | |
3957 | ||
3958 | if (RE_DUP_MAX < upper_bound) | |
3959 | FREE_STACK_RETURN (REG_BADBR); | |
3960 | ||
3961 | /* If the upper bound is zero, don't want to succeed at | |
3962 | all; jump from `laststart' to `b + 3', which will be | |
3963 | the end of the buffer after we insert the jump. */ | |
3964 | /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' | |
3965 | instead of 'b + 3'. */ | |
3966 | if (upper_bound == 0) | |
3967 | { | |
3968 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); | |
3969 | INSERT_JUMP (jump, laststart, b + 1 | |
3970 | + OFFSET_ADDRESS_SIZE); | |
3971 | b += 1 + OFFSET_ADDRESS_SIZE; | |
3972 | } | |
3973 | ||
3974 | /* Otherwise, we have a nontrivial interval. When | |
3975 | we're all done, the pattern will look like: | |
3976 | set_number_at <jump count> <upper bound> | |
3977 | set_number_at <succeed_n count> <lower bound> | |
3978 | succeed_n <after jump addr> <succeed_n count> | |
3979 | <body of loop> | |
3980 | jump_n <succeed_n addr> <jump count> | |
3981 | (The upper bound and `jump_n' are omitted if | |
3982 | `upper_bound' is 1, though.) */ | |
3983 | else | |
3984 | { /* If the upper bound is > 1, we need to insert | |
3985 | more at the end of the loop. */ | |
3986 | unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE + | |
3987 | (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE); | |
3988 | ||
3989 | GET_BUFFER_SPACE (nbytes); | |
3990 | ||
3991 | /* Initialize lower bound of the `succeed_n', even | |
3992 | though it will be set during matching by its | |
3993 | attendant `set_number_at' (inserted next), | |
3994 | because `re_compile_fastmap' needs to know. | |
3995 | Jump to the `jump_n' we might insert below. */ | |
3996 | INSERT_JUMP2 (succeed_n, laststart, | |
3997 | b + 1 + 2 * OFFSET_ADDRESS_SIZE | |
3998 | + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE) | |
3999 | , lower_bound); | |
4000 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; | |
4001 | ||
4002 | /* Code to initialize the lower bound. Insert | |
4003 | before the `succeed_n'. The `5' is the last two | |
4004 | bytes of this `set_number_at', plus 3 bytes of | |
4005 | the following `succeed_n'. */ | |
4006 | /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE' | |
4007 | is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE' | |
4008 | of the following `succeed_n'. */ | |
4009 | PREFIX(insert_op2) (set_number_at, laststart, 1 | |
4010 | + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b); | |
4011 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; | |
4012 | ||
4013 | if (upper_bound > 1) | |
4014 | { /* More than one repetition is allowed, so | |
4015 | append a backward jump to the `succeed_n' | |
4016 | that starts this interval. | |
4017 | ||
4018 | When we've reached this during matching, | |
4019 | we'll have matched the interval once, so | |
4020 | jump back only `upper_bound - 1' times. */ | |
4021 | STORE_JUMP2 (jump_n, b, laststart | |
4022 | + 2 * OFFSET_ADDRESS_SIZE + 1, | |
4023 | upper_bound - 1); | |
4024 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; | |
4025 | ||
4026 | /* The location we want to set is the second | |
4027 | parameter of the `jump_n'; that is `b-2' as | |
4028 | an absolute address. `laststart' will be | |
4029 | the `set_number_at' we're about to insert; | |
4030 | `laststart+3' the number to set, the source | |
4031 | for the relative address. But we are | |
4032 | inserting into the middle of the pattern -- | |
4033 | so everything is getting moved up by 5. | |
4034 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
4035 | i.e., b - laststart. | |
4036 | ||
4037 | We insert this at the beginning of the loop | |
4038 | so that if we fail during matching, we'll | |
4039 | reinitialize the bounds. */ | |
4040 | PREFIX(insert_op2) (set_number_at, laststart, | |
4041 | b - laststart, | |
4042 | upper_bound - 1, b); | |
4043 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; | |
4044 | } | |
4045 | } | |
4046 | pending_exact = 0; | |
4047 | break; | |
4048 | ||
4049 | invalid_interval: | |
4050 | if (!(syntax & RE_INVALID_INTERVAL_ORD)) | |
4051 | FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR); | |
4052 | unfetch_interval: | |
4053 | /* Match the characters as literals. */ | |
4054 | p = beg_interval; | |
4055 | c = '{'; | |
4056 | if (syntax & RE_NO_BK_BRACES) | |
4057 | goto normal_char; | |
4058 | else | |
4059 | goto normal_backslash; | |
4060 | } | |
4061 | ||
4062 | #ifdef emacs | |
4063 | /* There is no way to specify the before_dot and after_dot | |
4064 | operators. rms says this is ok. --karl */ | |
4065 | case '=': | |
4066 | BUF_PUSH (at_dot); | |
4067 | break; | |
4068 | ||
4069 | case 's': | |
4070 | laststart = b; | |
4071 | PATFETCH (c); | |
4072 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
4073 | break; | |
4074 | ||
4075 | case 'S': | |
4076 | laststart = b; | |
4077 | PATFETCH (c); | |
4078 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
4079 | break; | |
4080 | #endif /* emacs */ | |
4081 | ||
4082 | ||
4083 | case 'w': | |
4084 | if (syntax & RE_NO_GNU_OPS) | |
4085 | goto normal_char; | |
4086 | laststart = b; | |
4087 | BUF_PUSH (wordchar); | |
4088 | break; | |
4089 | ||
4090 | ||
4091 | case 'W': | |
4092 | if (syntax & RE_NO_GNU_OPS) | |
4093 | goto normal_char; | |
4094 | laststart = b; | |
4095 | BUF_PUSH (notwordchar); | |
4096 | break; | |
4097 | ||
4098 | ||
4099 | case '<': | |
4100 | if (syntax & RE_NO_GNU_OPS) | |
4101 | goto normal_char; | |
4102 | BUF_PUSH (wordbeg); | |
4103 | break; | |
4104 | ||
4105 | case '>': | |
4106 | if (syntax & RE_NO_GNU_OPS) | |
4107 | goto normal_char; | |
4108 | BUF_PUSH (wordend); | |
4109 | break; | |
4110 | ||
4111 | case 'b': | |
4112 | if (syntax & RE_NO_GNU_OPS) | |
4113 | goto normal_char; | |
4114 | BUF_PUSH (wordbound); | |
4115 | break; | |
4116 | ||
4117 | case 'B': | |
4118 | if (syntax & RE_NO_GNU_OPS) | |
4119 | goto normal_char; | |
4120 | BUF_PUSH (notwordbound); | |
4121 | break; | |
4122 | ||
4123 | case '`': | |
4124 | if (syntax & RE_NO_GNU_OPS) | |
4125 | goto normal_char; | |
4126 | BUF_PUSH (begbuf); | |
4127 | break; | |
4128 | ||
4129 | case '\'': | |
4130 | if (syntax & RE_NO_GNU_OPS) | |
4131 | goto normal_char; | |
4132 | BUF_PUSH (endbuf); | |
4133 | break; | |
4134 | ||
4135 | case '1': case '2': case '3': case '4': case '5': | |
4136 | case '6': case '7': case '8': case '9': | |
4137 | if (syntax & RE_NO_BK_REFS) | |
4138 | goto normal_char; | |
4139 | ||
4140 | c1 = c - '0'; | |
4141 | ||
4142 | if (c1 > regnum) | |
4143 | FREE_STACK_RETURN (REG_ESUBREG); | |
4144 | ||
4145 | /* Can't back reference to a subexpression if inside of it. */ | |
4146 | if (group_in_compile_stack (compile_stack, (regnum_t) c1)) | |
4147 | goto normal_char; | |
4148 | ||
4149 | laststart = b; | |
4150 | BUF_PUSH_2 (duplicate, c1); | |
4151 | break; | |
4152 | ||
4153 | ||
4154 | case '+': | |
4155 | case '?': | |
4156 | if (syntax & RE_BK_PLUS_QM) | |
4157 | goto handle_plus; | |
4158 | else | |
4159 | goto normal_backslash; | |
4160 | ||
4161 | default: | |
4162 | normal_backslash: | |
4163 | /* You might think it would be useful for \ to mean | |
4164 | not to translate; but if we don't translate it | |
4165 | it will never match anything. */ | |
4166 | c = TRANSLATE (c); | |
4167 | goto normal_char; | |
4168 | } | |
4169 | break; | |
4170 | ||
4171 | ||
4172 | default: | |
4173 | /* Expects the character in `c'. */ | |
4174 | normal_char: | |
4175 | /* If no exactn currently being built. */ | |
4176 | if (!pending_exact | |
4177 | #ifdef WCHAR | |
4178 | /* If last exactn handle binary(or character) and | |
4179 | new exactn handle character(or binary). */ | |
4180 | || is_exactn_bin != is_binary[p - 1 - pattern] | |
4181 | #endif /* WCHAR */ | |
4182 | ||
4183 | /* If last exactn not at current position. */ | |
4184 | || pending_exact + *pending_exact + 1 != b | |
4185 | ||
4186 | /* We have only one byte following the exactn for the count. */ | |
4187 | || *pending_exact == (1 << BYTEWIDTH) - 1 | |
4188 | ||
4189 | /* If followed by a repetition operator. */ | |
4190 | || *p == '*' || *p == '^' | |
4191 | || ((syntax & RE_BK_PLUS_QM) | |
4192 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
4193 | : (*p == '+' || *p == '?')) | |
4194 | || ((syntax & RE_INTERVALS) | |
4195 | && ((syntax & RE_NO_BK_BRACES) | |
4196 | ? *p == '{' | |
4197 | : (p[0] == '\\' && p[1] == '{')))) | |
4198 | { | |
4199 | /* Start building a new exactn. */ | |
4200 | ||
4201 | laststart = b; | |
4202 | ||
4203 | #ifdef WCHAR | |
4204 | /* Is this exactn binary data or character? */ | |
4205 | is_exactn_bin = is_binary[p - 1 - pattern]; | |
4206 | if (is_exactn_bin) | |
4207 | BUF_PUSH_2 (exactn_bin, 0); | |
4208 | else | |
4209 | BUF_PUSH_2 (exactn, 0); | |
4210 | #else | |
4211 | BUF_PUSH_2 (exactn, 0); | |
4212 | #endif /* WCHAR */ | |
4213 | pending_exact = b - 1; | |
4214 | } | |
4215 | ||
4216 | BUF_PUSH (c); | |
4217 | (*pending_exact)++; | |
4218 | break; | |
4219 | } /* switch (c) */ | |
4220 | } /* while p != pend */ | |
4221 | ||
4222 | ||
4223 | /* Through the pattern now. */ | |
4224 | ||
4225 | if (fixup_alt_jump) | |
4226 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
4227 | ||
4228 | if (!COMPILE_STACK_EMPTY) | |
4229 | FREE_STACK_RETURN (REG_EPAREN); | |
4230 | ||
4231 | /* If we don't want backtracking, force success | |
4232 | the first time we reach the end of the compiled pattern. */ | |
4233 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
4234 | BUF_PUSH (succeed); | |
4235 | ||
4236 | #ifdef WCHAR | |
4237 | free (pattern); | |
4238 | free (mbs_offset); | |
4239 | free (is_binary); | |
4240 | #endif | |
4241 | free (compile_stack.stack); | |
4242 | ||
4243 | /* We have succeeded; set the length of the buffer. */ | |
4244 | #ifdef WCHAR | |
4245 | bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR; | |
4246 | #else | |
4247 | bufp->used = b - bufp->buffer; | |
4248 | #endif | |
4249 | ||
4250 | #ifdef DEBUG | |
4251 | if (debug) | |
4252 | { | |
4253 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); | |
4254 | PREFIX(print_compiled_pattern) (bufp); | |
4255 | } | |
4256 | #endif /* DEBUG */ | |
4257 | ||
4258 | #ifndef MATCH_MAY_ALLOCATE | |
4259 | /* Initialize the failure stack to the largest possible stack. This | |
4260 | isn't necessary unless we're trying to avoid calling alloca in | |
4261 | the search and match routines. */ | |
4262 | { | |
4263 | int num_regs = bufp->re_nsub + 1; | |
4264 | ||
4265 | /* Since DOUBLE_FAIL_STACK refuses to double only if the current size | |
4266 | is strictly greater than re_max_failures, the largest possible stack | |
4267 | is 2 * re_max_failures failure points. */ | |
4268 | if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) | |
4269 | { | |
4270 | fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); | |
4271 | ||
4272 | # ifdef emacs | |
4273 | if (! fail_stack.stack) | |
4274 | fail_stack.stack | |
4275 | = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size | |
4276 | * sizeof (PREFIX(fail_stack_elt_t))); | |
4277 | else | |
4278 | fail_stack.stack | |
4279 | = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack, | |
4280 | (fail_stack.size | |
4281 | * sizeof (PREFIX(fail_stack_elt_t)))); | |
4282 | # else /* not emacs */ | |
4283 | if (! fail_stack.stack) | |
4284 | fail_stack.stack | |
4285 | = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size | |
4286 | * sizeof (PREFIX(fail_stack_elt_t))); | |
4287 | else | |
4288 | fail_stack.stack | |
4289 | = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack, | |
4290 | (fail_stack.size | |
4291 | * sizeof (PREFIX(fail_stack_elt_t)))); | |
4292 | # endif /* not emacs */ | |
4293 | } | |
4294 | ||
4295 | PREFIX(regex_grow_registers) (num_regs); | |
4296 | } | |
4297 | #endif /* not MATCH_MAY_ALLOCATE */ | |
4298 | ||
4299 | return REG_NOERROR; | |
4300 | } /* regex_compile */ | |
4301 | ||
4302 | /* Subroutines for `regex_compile'. */ | |
4303 | ||
4304 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
4305 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ | |
4306 | ||
4307 | static void | |
4308 | PREFIX(store_op1) (op, loc, arg) | |
4309 | re_opcode_t op; | |
4310 | UCHAR_T *loc; | |
4311 | int arg; | |
4312 | { | |
4313 | *loc = (UCHAR_T) op; | |
4314 | STORE_NUMBER (loc + 1, arg); | |
4315 | } | |
4316 | ||
4317 | ||
4318 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
4319 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ | |
4320 | ||
4321 | static void | |
4322 | PREFIX(store_op2) (op, loc, arg1, arg2) | |
4323 | re_opcode_t op; | |
4324 | UCHAR_T *loc; | |
4325 | int arg1, arg2; | |
4326 | { | |
4327 | *loc = (UCHAR_T) op; | |
4328 | STORE_NUMBER (loc + 1, arg1); | |
4329 | STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2); | |
4330 | } | |
4331 | ||
4332 | ||
4333 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
4334 | for OP followed by two-byte integer parameter ARG. */ | |
4335 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ | |
4336 | ||
4337 | static void | |
4338 | PREFIX(insert_op1) (op, loc, arg, end) | |
4339 | re_opcode_t op; | |
4340 | UCHAR_T *loc; | |
4341 | int arg; | |
4342 | UCHAR_T *end; | |
4343 | { | |
4344 | register UCHAR_T *pfrom = end; | |
4345 | register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE; | |
4346 | ||
4347 | while (pfrom != loc) | |
4348 | *--pto = *--pfrom; | |
4349 | ||
4350 | PREFIX(store_op1) (op, loc, arg); | |
4351 | } | |
4352 | ||
4353 | ||
4354 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
4355 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ | |
4356 | ||
4357 | static void | |
4358 | PREFIX(insert_op2) (op, loc, arg1, arg2, end) | |
4359 | re_opcode_t op; | |
4360 | UCHAR_T *loc; | |
4361 | int arg1, arg2; | |
4362 | UCHAR_T *end; | |
4363 | { | |
4364 | register UCHAR_T *pfrom = end; | |
4365 | register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE; | |
4366 | ||
4367 | while (pfrom != loc) | |
4368 | *--pto = *--pfrom; | |
4369 | ||
4370 | PREFIX(store_op2) (op, loc, arg1, arg2); | |
4371 | } | |
4372 | ||
4373 | ||
4374 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
4375 | after an alternative or a begin-subexpression. We assume there is at | |
4376 | least one character before the ^. */ | |
4377 | ||
4378 | static boolean | |
4379 | PREFIX(at_begline_loc_p) (pattern, p, syntax) | |
4380 | const CHAR_T *pattern, *p; | |
4381 | reg_syntax_t syntax; | |
4382 | { | |
4383 | const CHAR_T *prev = p - 2; | |
4384 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
4385 | ||
4386 | return | |
4387 | /* After a subexpression? */ | |
4388 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
4389 | /* After an alternative? */ | |
4390 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
4391 | } | |
4392 | ||
4393 | ||
4394 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
4395 | at least one character after the $, i.e., `P < PEND'. */ | |
4396 | ||
4397 | static boolean | |
4398 | PREFIX(at_endline_loc_p) (p, pend, syntax) | |
4399 | const CHAR_T *p, *pend; | |
4400 | reg_syntax_t syntax; | |
4401 | { | |
4402 | const CHAR_T *next = p; | |
4403 | boolean next_backslash = *next == '\\'; | |
4404 | const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0; | |
4405 | ||
4406 | return | |
4407 | /* Before a subexpression? */ | |
4408 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
4409 | : next_backslash && next_next && *next_next == ')') | |
4410 | /* Before an alternative? */ | |
4411 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
4412 | : next_backslash && next_next && *next_next == '|'); | |
4413 | } | |
4414 | ||
4415 | #else /* not INSIDE_RECURSION */ | |
4416 | ||
4417 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and | |
4418 | false if it's not. */ | |
4419 | ||
4420 | static boolean | |
4421 | group_in_compile_stack (compile_stack, regnum) | |
4422 | compile_stack_type compile_stack; | |
4423 | regnum_t regnum; | |
4424 | { | |
4425 | int this_element; | |
4426 | ||
4427 | for (this_element = compile_stack.avail - 1; | |
4428 | this_element >= 0; | |
4429 | this_element--) | |
4430 | if (compile_stack.stack[this_element].regnum == regnum) | |
4431 | return true; | |
4432 | ||
4433 | return false; | |
4434 | } | |
4435 | #endif /* not INSIDE_RECURSION */ | |
4436 | ||
4437 | #ifdef INSIDE_RECURSION | |
4438 | ||
4439 | #ifdef WCHAR | |
4440 | /* This insert space, which size is "num", into the pattern at "loc". | |
4441 | "end" must point the end of the allocated buffer. */ | |
4442 | static void | |
4443 | insert_space (num, loc, end) | |
4444 | int num; | |
4445 | CHAR_T *loc; | |
4446 | CHAR_T *end; | |
4447 | { | |
4448 | register CHAR_T *pto = end; | |
4449 | register CHAR_T *pfrom = end - num; | |
4450 | ||
4451 | while (pfrom >= loc) | |
4452 | *pto-- = *pfrom--; | |
4453 | } | |
4454 | #endif /* WCHAR */ | |
4455 | ||
4456 | #ifdef WCHAR | |
4457 | static reg_errcode_t | |
4458 | wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b, | |
4459 | char_set) | |
4460 | CHAR_T range_start_char; | |
4461 | const CHAR_T **p_ptr, *pend; | |
4462 | CHAR_T *char_set, *b; | |
4463 | RE_TRANSLATE_TYPE translate; | |
4464 | reg_syntax_t syntax; | |
4465 | { | |
4466 | const CHAR_T *p = *p_ptr; | |
4467 | CHAR_T range_start, range_end; | |
4468 | reg_errcode_t ret; | |
4469 | # ifdef _LIBC | |
4470 | uint32_t nrules; | |
4471 | uint32_t start_val, end_val; | |
4472 | # endif | |
4473 | if (p == pend) | |
4474 | return REG_ERANGE; | |
4475 | ||
4476 | # ifdef _LIBC | |
4477 | nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
4478 | if (nrules != 0) | |
4479 | { | |
4480 | const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE, | |
4481 | _NL_COLLATE_COLLSEQWC); | |
4482 | const unsigned char *extra = (const unsigned char *) | |
4483 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); | |
4484 | ||
4485 | if (range_start_char < -1) | |
4486 | { | |
4487 | /* range_start is a collating symbol. */ | |
4488 | int32_t *wextra; | |
4489 | /* Retreive the index and get collation sequence value. */ | |
4490 | wextra = (int32_t*)(extra + char_set[-range_start_char]); | |
4491 | start_val = wextra[1 + *wextra]; | |
4492 | } | |
4493 | else | |
4494 | start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char)); | |
4495 | ||
4496 | end_val = collseq_table_lookup (collseq, TRANSLATE (p[0])); | |
4497 | ||
4498 | /* Report an error if the range is empty and the syntax prohibits | |
4499 | this. */ | |
4500 | ret = ((syntax & RE_NO_EMPTY_RANGES) | |
4501 | && (start_val > end_val))? REG_ERANGE : REG_NOERROR; | |
4502 | ||
4503 | /* Insert space to the end of the char_ranges. */ | |
4504 | insert_space(2, b - char_set[5] - 2, b - 1); | |
4505 | *(b - char_set[5] - 2) = (wchar_t)start_val; | |
4506 | *(b - char_set[5] - 1) = (wchar_t)end_val; | |
4507 | char_set[4]++; /* ranges_index */ | |
4508 | } | |
4509 | else | |
4510 | # endif | |
4511 | { | |
4512 | range_start = (range_start_char >= 0)? TRANSLATE (range_start_char): | |
4513 | range_start_char; | |
4514 | range_end = TRANSLATE (p[0]); | |
4515 | /* Report an error if the range is empty and the syntax prohibits | |
4516 | this. */ | |
4517 | ret = ((syntax & RE_NO_EMPTY_RANGES) | |
4518 | && (range_start > range_end))? REG_ERANGE : REG_NOERROR; | |
4519 | ||
4520 | /* Insert space to the end of the char_ranges. */ | |
4521 | insert_space(2, b - char_set[5] - 2, b - 1); | |
4522 | *(b - char_set[5] - 2) = range_start; | |
4523 | *(b - char_set[5] - 1) = range_end; | |
4524 | char_set[4]++; /* ranges_index */ | |
4525 | } | |
4526 | /* Have to increment the pointer into the pattern string, so the | |
4527 | caller isn't still at the ending character. */ | |
4528 | (*p_ptr)++; | |
4529 | ||
4530 | return ret; | |
4531 | } | |
4532 | #else /* BYTE */ | |
4533 | /* Read the ending character of a range (in a bracket expression) from the | |
4534 | uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
4535 | starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
4536 | Then we set the translation of all bits between the starting and | |
4537 | ending characters (inclusive) in the compiled pattern B. | |
4538 | ||
4539 | Return an error code. | |
4540 | ||
4541 | We use these short variable names so we can use the same macros as | |
4542 | `regex_compile' itself. */ | |
4543 | ||
4544 | static reg_errcode_t | |
4545 | byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b) | |
4546 | unsigned int range_start_char; | |
4547 | const char **p_ptr, *pend; | |
4548 | RE_TRANSLATE_TYPE translate; | |
4549 | reg_syntax_t syntax; | |
4550 | unsigned char *b; | |
4551 | { | |
4552 | unsigned this_char; | |
4553 | const char *p = *p_ptr; | |
4554 | reg_errcode_t ret; | |
4555 | # if _LIBC | |
4556 | const unsigned char *collseq; | |
4557 | unsigned int start_colseq; | |
4558 | unsigned int end_colseq; | |
4559 | # else | |
4560 | unsigned end_char; | |
4561 | # endif | |
4562 | ||
4563 | if (p == pend) | |
4564 | return REG_ERANGE; | |
4565 | ||
4566 | /* Have to increment the pointer into the pattern string, so the | |
4567 | caller isn't still at the ending character. */ | |
4568 | (*p_ptr)++; | |
4569 | ||
4570 | /* Report an error if the range is empty and the syntax prohibits this. */ | |
4571 | ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
4572 | ||
4573 | # if _LIBC | |
4574 | collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE, | |
4575 | _NL_COLLATE_COLLSEQMB); | |
4576 | ||
4577 | start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)]; | |
4578 | end_colseq = collseq[(unsigned char) TRANSLATE (p[0])]; | |
4579 | for (this_char = 0; this_char <= (unsigned char) -1; ++this_char) | |
4580 | { | |
4581 | unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)]; | |
4582 | ||
4583 | if (start_colseq <= this_colseq && this_colseq <= end_colseq) | |
4584 | { | |
4585 | SET_LIST_BIT (TRANSLATE (this_char)); | |
4586 | ret = REG_NOERROR; | |
4587 | } | |
4588 | } | |
4589 | # else | |
4590 | /* Here we see why `this_char' has to be larger than an `unsigned | |
4591 | char' -- we would otherwise go into an infinite loop, since all | |
4592 | characters <= 0xff. */ | |
4593 | range_start_char = TRANSLATE (range_start_char); | |
4594 | /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE, | |
4595 | and some compilers cast it to int implicitly, so following for_loop | |
4596 | may fall to (almost) infinite loop. | |
4597 | e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff. | |
4598 | To avoid this, we cast p[0] to unsigned int and truncate it. */ | |
4599 | end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1)); | |
4600 | ||
4601 | for (this_char = range_start_char; this_char <= end_char; ++this_char) | |
4602 | { | |
4603 | SET_LIST_BIT (TRANSLATE (this_char)); | |
4604 | ret = REG_NOERROR; | |
4605 | } | |
4606 | # endif | |
4607 | ||
4608 | return ret; | |
4609 | } | |
4610 | #endif /* WCHAR */ | |
4611 | \f | |
4612 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
4613 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
4614 | characters can start a string that matches the pattern. This fastmap | |
4615 | is used by re_search to skip quickly over impossible starting points. | |
4616 | ||
4617 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
4618 | area as BUFP->fastmap. | |
4619 | ||
4620 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
4621 | the pattern buffer. | |
4622 | ||
4623 | Returns 0 if we succeed, -2 if an internal error. */ | |
4624 | ||
4625 | #ifdef WCHAR | |
4626 | /* local function for re_compile_fastmap. | |
4627 | truncate wchar_t character to char. */ | |
4628 | static unsigned char truncate_wchar (CHAR_T c); | |
4629 | ||
4630 | static unsigned char | |
4631 | truncate_wchar (c) | |
4632 | CHAR_T c; | |
4633 | { | |
dc676635 DD |
4634 | unsigned char buf[MB_CUR_MAX]; |
4635 | mbstate_t state; | |
4636 | int retval; | |
4637 | memset (&state, '\0', sizeof (state)); | |
4638 | # ifdef _LIBC | |
4639 | retval = __wcrtomb (buf, c, &state); | |
4640 | # else | |
4641 | retval = wcrtomb (buf, c, &state); | |
4642 | # endif | |
4643 | return retval > 0 ? buf[0] : (unsigned char) c; | |
2a6ef469 DD |
4644 | } |
4645 | #endif /* WCHAR */ | |
4646 | ||
4647 | static int | |
4648 | PREFIX(re_compile_fastmap) (bufp) | |
4649 | struct re_pattern_buffer *bufp; | |
4650 | { | |
4651 | int j, k; | |
4652 | #ifdef MATCH_MAY_ALLOCATE | |
4653 | PREFIX(fail_stack_type) fail_stack; | |
4654 | #endif | |
4655 | #ifndef REGEX_MALLOC | |
4656 | char *destination; | |
4657 | #endif | |
4658 | ||
4659 | register char *fastmap = bufp->fastmap; | |
4660 | ||
4661 | #ifdef WCHAR | |
4662 | /* We need to cast pattern to (wchar_t*), because we casted this compiled | |
4663 | pattern to (char*) in regex_compile. */ | |
4664 | UCHAR_T *pattern = (UCHAR_T*)bufp->buffer; | |
4665 | register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used); | |
4666 | #else /* BYTE */ | |
4667 | UCHAR_T *pattern = bufp->buffer; | |
4668 | register UCHAR_T *pend = pattern + bufp->used; | |
4669 | #endif /* WCHAR */ | |
4670 | UCHAR_T *p = pattern; | |
4671 | ||
4672 | #ifdef REL_ALLOC | |
4673 | /* This holds the pointer to the failure stack, when | |
4674 | it is allocated relocatably. */ | |
4675 | fail_stack_elt_t *failure_stack_ptr; | |
4676 | #endif | |
4677 | ||
4678 | /* Assume that each path through the pattern can be null until | |
4679 | proven otherwise. We set this false at the bottom of switch | |
4680 | statement, to which we get only if a particular path doesn't | |
4681 | match the empty string. */ | |
4682 | boolean path_can_be_null = true; | |
4683 | ||
4684 | /* We aren't doing a `succeed_n' to begin with. */ | |
4685 | boolean succeed_n_p = false; | |
4686 | ||
4687 | assert (fastmap != NULL && p != NULL); | |
4688 | ||
4689 | INIT_FAIL_STACK (); | |
4690 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
4691 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
4692 | bufp->can_be_null = 0; | |
4693 | ||
4694 | while (1) | |
4695 | { | |
4696 | if (p == pend || *p == succeed) | |
4697 | { | |
4698 | /* We have reached the (effective) end of pattern. */ | |
4699 | if (!FAIL_STACK_EMPTY ()) | |
4700 | { | |
4701 | bufp->can_be_null |= path_can_be_null; | |
4702 | ||
4703 | /* Reset for next path. */ | |
4704 | path_can_be_null = true; | |
4705 | ||
4706 | p = fail_stack.stack[--fail_stack.avail].pointer; | |
4707 | ||
4708 | continue; | |
4709 | } | |
4710 | else | |
4711 | break; | |
4712 | } | |
4713 | ||
4714 | /* We should never be about to go beyond the end of the pattern. */ | |
4715 | assert (p < pend); | |
4716 | ||
4717 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) | |
4718 | { | |
4719 | ||
4720 | /* I guess the idea here is to simply not bother with a fastmap | |
4721 | if a backreference is used, since it's too hard to figure out | |
4722 | the fastmap for the corresponding group. Setting | |
4723 | `can_be_null' stops `re_search_2' from using the fastmap, so | |
4724 | that is all we do. */ | |
4725 | case duplicate: | |
4726 | bufp->can_be_null = 1; | |
4727 | goto done; | |
4728 | ||
4729 | ||
4730 | /* Following are the cases which match a character. These end | |
4731 | with `break'. */ | |
4732 | ||
4733 | #ifdef WCHAR | |
4734 | case exactn: | |
4735 | fastmap[truncate_wchar(p[1])] = 1; | |
4736 | break; | |
4737 | #else /* BYTE */ | |
4738 | case exactn: | |
4739 | fastmap[p[1]] = 1; | |
4740 | break; | |
4741 | #endif /* WCHAR */ | |
4742 | #ifdef MBS_SUPPORT | |
4743 | case exactn_bin: | |
4744 | fastmap[p[1]] = 1; | |
4745 | break; | |
4746 | #endif | |
4747 | ||
4748 | #ifdef WCHAR | |
4749 | /* It is hard to distinguish fastmap from (multi byte) characters | |
4750 | which depends on current locale. */ | |
4751 | case charset: | |
4752 | case charset_not: | |
4753 | case wordchar: | |
4754 | case notwordchar: | |
4755 | bufp->can_be_null = 1; | |
4756 | goto done; | |
4757 | #else /* BYTE */ | |
4758 | case charset: | |
4759 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
4760 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
4761 | fastmap[j] = 1; | |
4762 | break; | |
4763 | ||
4764 | ||
4765 | case charset_not: | |
4766 | /* Chars beyond end of map must be allowed. */ | |
4767 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
4768 | fastmap[j] = 1; | |
4769 | ||
4770 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
4771 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
4772 | fastmap[j] = 1; | |
4773 | break; | |
4774 | ||
4775 | ||
4776 | case wordchar: | |
4777 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4778 | if (SYNTAX (j) == Sword) | |
4779 | fastmap[j] = 1; | |
4780 | break; | |
4781 | ||
4782 | ||
4783 | case notwordchar: | |
4784 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4785 | if (SYNTAX (j) != Sword) | |
4786 | fastmap[j] = 1; | |
4787 | break; | |
4788 | #endif /* WCHAR */ | |
4789 | ||
4790 | case anychar: | |
4791 | { | |
4792 | int fastmap_newline = fastmap['\n']; | |
4793 | ||
4794 | /* `.' matches anything ... */ | |
4795 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4796 | fastmap[j] = 1; | |
4797 | ||
4798 | /* ... except perhaps newline. */ | |
4799 | if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
4800 | fastmap['\n'] = fastmap_newline; | |
4801 | ||
4802 | /* Return if we have already set `can_be_null'; if we have, | |
4803 | then the fastmap is irrelevant. Something's wrong here. */ | |
4804 | else if (bufp->can_be_null) | |
4805 | goto done; | |
4806 | ||
4807 | /* Otherwise, have to check alternative paths. */ | |
4808 | break; | |
4809 | } | |
4810 | ||
4811 | #ifdef emacs | |
4812 | case syntaxspec: | |
4813 | k = *p++; | |
4814 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4815 | if (SYNTAX (j) == (enum syntaxcode) k) | |
4816 | fastmap[j] = 1; | |
4817 | break; | |
4818 | ||
4819 | ||
4820 | case notsyntaxspec: | |
4821 | k = *p++; | |
4822 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4823 | if (SYNTAX (j) != (enum syntaxcode) k) | |
4824 | fastmap[j] = 1; | |
4825 | break; | |
4826 | ||
4827 | ||
4828 | /* All cases after this match the empty string. These end with | |
4829 | `continue'. */ | |
4830 | ||
4831 | ||
4832 | case before_dot: | |
4833 | case at_dot: | |
4834 | case after_dot: | |
4835 | continue; | |
4836 | #endif /* emacs */ | |
4837 | ||
4838 | ||
4839 | case no_op: | |
4840 | case begline: | |
4841 | case endline: | |
4842 | case begbuf: | |
4843 | case endbuf: | |
4844 | case wordbound: | |
4845 | case notwordbound: | |
4846 | case wordbeg: | |
4847 | case wordend: | |
4848 | case push_dummy_failure: | |
4849 | continue; | |
4850 | ||
4851 | ||
4852 | case jump_n: | |
4853 | case pop_failure_jump: | |
4854 | case maybe_pop_jump: | |
4855 | case jump: | |
4856 | case jump_past_alt: | |
4857 | case dummy_failure_jump: | |
4858 | EXTRACT_NUMBER_AND_INCR (j, p); | |
4859 | p += j; | |
4860 | if (j > 0) | |
4861 | continue; | |
4862 | ||
4863 | /* Jump backward implies we just went through the body of a | |
4864 | loop and matched nothing. Opcode jumped to should be | |
4865 | `on_failure_jump' or `succeed_n'. Just treat it like an | |
4866 | ordinary jump. For a * loop, it has pushed its failure | |
4867 | point already; if so, discard that as redundant. */ | |
4868 | if ((re_opcode_t) *p != on_failure_jump | |
4869 | && (re_opcode_t) *p != succeed_n) | |
4870 | continue; | |
4871 | ||
4872 | p++; | |
4873 | EXTRACT_NUMBER_AND_INCR (j, p); | |
4874 | p += j; | |
4875 | ||
4876 | /* If what's on the stack is where we are now, pop it. */ | |
4877 | if (!FAIL_STACK_EMPTY () | |
4878 | && fail_stack.stack[fail_stack.avail - 1].pointer == p) | |
4879 | fail_stack.avail--; | |
4880 | ||
4881 | continue; | |
4882 | ||
4883 | ||
4884 | case on_failure_jump: | |
4885 | case on_failure_keep_string_jump: | |
4886 | handle_on_failure_jump: | |
4887 | EXTRACT_NUMBER_AND_INCR (j, p); | |
4888 | ||
4889 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
4890 | end of the pattern. We don't want to push such a point, | |
4891 | since when we restore it above, entering the switch will | |
4892 | increment `p' past the end of the pattern. We don't need | |
4893 | to push such a point since we obviously won't find any more | |
4894 | fastmap entries beyond `pend'. Such a pattern can match | |
4895 | the null string, though. */ | |
4896 | if (p + j < pend) | |
4897 | { | |
4898 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
4899 | { | |
4900 | RESET_FAIL_STACK (); | |
4901 | return -2; | |
4902 | } | |
4903 | } | |
4904 | else | |
4905 | bufp->can_be_null = 1; | |
4906 | ||
4907 | if (succeed_n_p) | |
4908 | { | |
4909 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
4910 | succeed_n_p = false; | |
4911 | } | |
4912 | ||
4913 | continue; | |
4914 | ||
4915 | ||
4916 | case succeed_n: | |
4917 | /* Get to the number of times to succeed. */ | |
4918 | p += OFFSET_ADDRESS_SIZE; | |
4919 | ||
4920 | /* Increment p past the n for when k != 0. */ | |
4921 | EXTRACT_NUMBER_AND_INCR (k, p); | |
4922 | if (k == 0) | |
4923 | { | |
4924 | p -= 2 * OFFSET_ADDRESS_SIZE; | |
4925 | succeed_n_p = true; /* Spaghetti code alert. */ | |
4926 | goto handle_on_failure_jump; | |
4927 | } | |
4928 | continue; | |
4929 | ||
4930 | ||
4931 | case set_number_at: | |
4932 | p += 2 * OFFSET_ADDRESS_SIZE; | |
4933 | continue; | |
4934 | ||
4935 | ||
4936 | case start_memory: | |
4937 | case stop_memory: | |
4938 | p += 2; | |
4939 | continue; | |
4940 | ||
4941 | ||
4942 | default: | |
4943 | abort (); /* We have listed all the cases. */ | |
4944 | } /* switch *p++ */ | |
4945 | ||
4946 | /* Getting here means we have found the possible starting | |
4947 | characters for one path of the pattern -- and that the empty | |
4948 | string does not match. We need not follow this path further. | |
4949 | Instead, look at the next alternative (remembered on the | |
4950 | stack), or quit if no more. The test at the top of the loop | |
4951 | does these things. */ | |
4952 | path_can_be_null = false; | |
4953 | p = pend; | |
4954 | } /* while p */ | |
4955 | ||
4956 | /* Set `can_be_null' for the last path (also the first path, if the | |
4957 | pattern is empty). */ | |
4958 | bufp->can_be_null |= path_can_be_null; | |
4959 | ||
4960 | done: | |
4961 | RESET_FAIL_STACK (); | |
4962 | return 0; | |
4963 | } | |
4964 | ||
4965 | #else /* not INSIDE_RECURSION */ | |
4966 | ||
4967 | int | |
4968 | re_compile_fastmap (bufp) | |
4969 | struct re_pattern_buffer *bufp; | |
4970 | { | |
4971 | # ifdef MBS_SUPPORT | |
4972 | if (MB_CUR_MAX != 1) | |
4973 | return wcs_re_compile_fastmap(bufp); | |
4974 | else | |
4975 | # endif | |
4976 | return byte_re_compile_fastmap(bufp); | |
4977 | } /* re_compile_fastmap */ | |
4978 | #ifdef _LIBC | |
4979 | weak_alias (__re_compile_fastmap, re_compile_fastmap) | |
4980 | #endif | |
4981 | \f | |
4982 | ||
4983 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
4984 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
4985 | this memory for recording register information. STARTS and ENDS | |
4986 | must be allocated using the malloc library routine, and must each | |
4987 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
4988 | ||
4989 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
4990 | register data. | |
4991 | ||
4992 | Unless this function is called, the first search or match using | |
4993 | PATTERN_BUFFER will allocate its own register data, without | |
4994 | freeing the old data. */ | |
4995 | ||
4996 | void | |
4997 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
4998 | struct re_pattern_buffer *bufp; | |
4999 | struct re_registers *regs; | |
5000 | unsigned num_regs; | |
5001 | regoff_t *starts, *ends; | |
5002 | { | |
5003 | if (num_regs) | |
5004 | { | |
5005 | bufp->regs_allocated = REGS_REALLOCATE; | |
5006 | regs->num_regs = num_regs; | |
5007 | regs->start = starts; | |
5008 | regs->end = ends; | |
5009 | } | |
5010 | else | |
5011 | { | |
5012 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5013 | regs->num_regs = 0; | |
5014 | regs->start = regs->end = (regoff_t *) 0; | |
5015 | } | |
5016 | } | |
5017 | #ifdef _LIBC | |
5018 | weak_alias (__re_set_registers, re_set_registers) | |
5019 | #endif | |
5020 | \f | |
5021 | /* Searching routines. */ | |
5022 | ||
5023 | /* Like re_search_2, below, but only one string is specified, and | |
5024 | doesn't let you say where to stop matching. */ | |
5025 | ||
5026 | int | |
5027 | re_search (bufp, string, size, startpos, range, regs) | |
5028 | struct re_pattern_buffer *bufp; | |
5029 | const char *string; | |
5030 | int size, startpos, range; | |
5031 | struct re_registers *regs; | |
5032 | { | |
5033 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
5034 | regs, size); | |
5035 | } | |
5036 | #ifdef _LIBC | |
5037 | weak_alias (__re_search, re_search) | |
5038 | #endif | |
5039 | ||
5040 | ||
5041 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
5042 | virtual concatenation of STRING1 and STRING2, starting first at index | |
5043 | STARTPOS, then at STARTPOS + 1, and so on. | |
5044 | ||
5045 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
5046 | ||
5047 | RANGE is how far to scan while trying to match. RANGE = 0 means try | |
5048 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
5049 | RANGE. | |
5050 | ||
5051 | In REGS, return the indices of the virtual concatenation of STRING1 | |
5052 | and STRING2 that matched the entire BUFP->buffer and its contained | |
5053 | subexpressions. | |
5054 | ||
5055 | Do not consider matching one past the index STOP in the virtual | |
5056 | concatenation of STRING1 and STRING2. | |
5057 | ||
5058 | We return either the position in the strings at which the match was | |
5059 | found, -1 if no match, or -2 if error (such as failure | |
5060 | stack overflow). */ | |
5061 | ||
5062 | int | |
5063 | re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
5064 | struct re_pattern_buffer *bufp; | |
5065 | const char *string1, *string2; | |
5066 | int size1, size2; | |
5067 | int startpos; | |
5068 | int range; | |
5069 | struct re_registers *regs; | |
5070 | int stop; | |
5071 | { | |
5072 | # ifdef MBS_SUPPORT | |
5073 | if (MB_CUR_MAX != 1) | |
5074 | return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos, | |
5075 | range, regs, stop); | |
5076 | else | |
5077 | # endif | |
5078 | return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos, | |
5079 | range, regs, stop); | |
5080 | } /* re_search_2 */ | |
5081 | #ifdef _LIBC | |
5082 | weak_alias (__re_search_2, re_search_2) | |
5083 | #endif | |
5084 | ||
5085 | #endif /* not INSIDE_RECURSION */ | |
5086 | ||
5087 | #ifdef INSIDE_RECURSION | |
5088 | ||
5089 | #ifdef MATCH_MAY_ALLOCATE | |
5090 | # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL | |
5091 | #else | |
5092 | # define FREE_VAR(var) if (var) free (var); var = NULL | |
5093 | #endif | |
5094 | ||
5095 | #ifdef WCHAR | |
86710ce2 DD |
5096 | # define MAX_ALLOCA_SIZE 2000 |
5097 | ||
5098 | # define FREE_WCS_BUFFERS() \ | |
5099 | do { \ | |
5100 | if (size1 > MAX_ALLOCA_SIZE) \ | |
5101 | { \ | |
5102 | free (wcs_string1); \ | |
5103 | free (mbs_offset1); \ | |
5104 | } \ | |
5105 | else \ | |
5106 | { \ | |
5107 | FREE_VAR (wcs_string1); \ | |
5108 | FREE_VAR (mbs_offset1); \ | |
5109 | } \ | |
5110 | if (size2 > MAX_ALLOCA_SIZE) \ | |
5111 | { \ | |
5112 | free (wcs_string2); \ | |
5113 | free (mbs_offset2); \ | |
5114 | } \ | |
5115 | else \ | |
5116 | { \ | |
5117 | FREE_VAR (wcs_string2); \ | |
5118 | FREE_VAR (mbs_offset2); \ | |
5119 | } \ | |
2a6ef469 DD |
5120 | } while (0) |
5121 | ||
5122 | #endif | |
5123 | ||
86710ce2 | 5124 | |
2a6ef469 DD |
5125 | static int |
5126 | PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range, | |
5127 | regs, stop) | |
5128 | struct re_pattern_buffer *bufp; | |
5129 | const char *string1, *string2; | |
5130 | int size1, size2; | |
5131 | int startpos; | |
5132 | int range; | |
5133 | struct re_registers *regs; | |
5134 | int stop; | |
5135 | { | |
5136 | int val; | |
5137 | register char *fastmap = bufp->fastmap; | |
5138 | register RE_TRANSLATE_TYPE translate = bufp->translate; | |
5139 | int total_size = size1 + size2; | |
5140 | int endpos = startpos + range; | |
5141 | #ifdef WCHAR | |
5142 | /* We need wchar_t* buffers correspond to cstring1, cstring2. */ | |
5143 | wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL; | |
5144 | /* We need the size of wchar_t buffers correspond to csize1, csize2. */ | |
5145 | int wcs_size1 = 0, wcs_size2 = 0; | |
5146 | /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ | |
5147 | int *mbs_offset1 = NULL, *mbs_offset2 = NULL; | |
5148 | /* They hold whether each wchar_t is binary data or not. */ | |
5149 | char *is_binary = NULL; | |
5150 | #endif /* WCHAR */ | |
5151 | ||
5152 | /* Check for out-of-range STARTPOS. */ | |
5153 | if (startpos < 0 || startpos > total_size) | |
5154 | return -1; | |
5155 | ||
5156 | /* Fix up RANGE if it might eventually take us outside | |
5157 | the virtual concatenation of STRING1 and STRING2. | |
5158 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ | |
5159 | if (endpos < 0) | |
5160 | range = 0 - startpos; | |
5161 | else if (endpos > total_size) | |
5162 | range = total_size - startpos; | |
5163 | ||
5164 | /* If the search isn't to be a backwards one, don't waste time in a | |
5165 | search for a pattern that must be anchored. */ | |
5166 | if (bufp->used > 0 && range > 0 | |
5167 | && ((re_opcode_t) bufp->buffer[0] == begbuf | |
5168 | /* `begline' is like `begbuf' if it cannot match at newlines. */ | |
5169 | || ((re_opcode_t) bufp->buffer[0] == begline | |
5170 | && !bufp->newline_anchor))) | |
5171 | { | |
5172 | if (startpos > 0) | |
5173 | return -1; | |
5174 | else | |
5175 | range = 1; | |
5176 | } | |
5177 | ||
5178 | #ifdef emacs | |
5179 | /* In a forward search for something that starts with \=. | |
5180 | don't keep searching past point. */ | |
5181 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
5182 | { | |
5183 | range = PT - startpos; | |
5184 | if (range <= 0) | |
5185 | return -1; | |
5186 | } | |
5187 | #endif /* emacs */ | |
5188 | ||
5189 | /* Update the fastmap now if not correct already. */ | |
5190 | if (fastmap && !bufp->fastmap_accurate) | |
5191 | if (re_compile_fastmap (bufp) == -2) | |
5192 | return -2; | |
5193 | ||
5194 | #ifdef WCHAR | |
5195 | /* Allocate wchar_t array for wcs_string1 and wcs_string2 and | |
5196 | fill them with converted string. */ | |
5197 | if (size1 != 0) | |
5198 | { | |
86710ce2 DD |
5199 | if (size1 > MAX_ALLOCA_SIZE) |
5200 | { | |
5201 | wcs_string1 = TALLOC (size1 + 1, CHAR_T); | |
5202 | mbs_offset1 = TALLOC (size1 + 1, int); | |
5203 | is_binary = TALLOC (size1 + 1, char); | |
5204 | } | |
5205 | else | |
5206 | { | |
5207 | wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T); | |
5208 | mbs_offset1 = REGEX_TALLOC (size1 + 1, int); | |
5209 | is_binary = REGEX_TALLOC (size1 + 1, char); | |
5210 | } | |
2a6ef469 DD |
5211 | if (!wcs_string1 || !mbs_offset1 || !is_binary) |
5212 | { | |
86710ce2 DD |
5213 | if (size1 > MAX_ALLOCA_SIZE) |
5214 | { | |
5215 | free (wcs_string1); | |
5216 | free (mbs_offset1); | |
5217 | free (is_binary); | |
5218 | } | |
5219 | else | |
5220 | { | |
5221 | FREE_VAR (wcs_string1); | |
5222 | FREE_VAR (mbs_offset1); | |
5223 | FREE_VAR (is_binary); | |
5224 | } | |
2a6ef469 DD |
5225 | return -2; |
5226 | } | |
5227 | wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1, | |
5228 | mbs_offset1, is_binary); | |
5229 | wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */ | |
86710ce2 DD |
5230 | if (size1 > MAX_ALLOCA_SIZE) |
5231 | free (is_binary); | |
5232 | else | |
5233 | FREE_VAR (is_binary); | |
2a6ef469 DD |
5234 | } |
5235 | if (size2 != 0) | |
5236 | { | |
86710ce2 DD |
5237 | if (size2 > MAX_ALLOCA_SIZE) |
5238 | { | |
5239 | wcs_string2 = TALLOC (size2 + 1, CHAR_T); | |
5240 | mbs_offset2 = TALLOC (size2 + 1, int); | |
5241 | is_binary = TALLOC (size2 + 1, char); | |
5242 | } | |
5243 | else | |
5244 | { | |
5245 | wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T); | |
5246 | mbs_offset2 = REGEX_TALLOC (size2 + 1, int); | |
5247 | is_binary = REGEX_TALLOC (size2 + 1, char); | |
5248 | } | |
2a6ef469 DD |
5249 | if (!wcs_string2 || !mbs_offset2 || !is_binary) |
5250 | { | |
5251 | FREE_WCS_BUFFERS (); | |
86710ce2 DD |
5252 | if (size2 > MAX_ALLOCA_SIZE) |
5253 | free (is_binary); | |
5254 | else | |
5255 | FREE_VAR (is_binary); | |
2a6ef469 DD |
5256 | return -2; |
5257 | } | |
5258 | wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2, | |
5259 | mbs_offset2, is_binary); | |
5260 | wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */ | |
86710ce2 DD |
5261 | if (size2 > MAX_ALLOCA_SIZE) |
5262 | free (is_binary); | |
5263 | else | |
5264 | FREE_VAR (is_binary); | |
2a6ef469 DD |
5265 | } |
5266 | #endif /* WCHAR */ | |
5267 | ||
5268 | ||
5269 | /* Loop through the string, looking for a place to start matching. */ | |
5270 | for (;;) | |
5271 | { | |
5272 | /* If a fastmap is supplied, skip quickly over characters that | |
5273 | cannot be the start of a match. If the pattern can match the | |
5274 | null string, however, we don't need to skip characters; we want | |
5275 | the first null string. */ | |
5276 | if (fastmap && startpos < total_size && !bufp->can_be_null) | |
5277 | { | |
5278 | if (range > 0) /* Searching forwards. */ | |
5279 | { | |
5280 | register const char *d; | |
5281 | register int lim = 0; | |
5282 | int irange = range; | |
5283 | ||
5284 | if (startpos < size1 && startpos + range >= size1) | |
5285 | lim = range - (size1 - startpos); | |
5286 | ||
5287 | d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
5288 | ||
5289 | /* Written out as an if-else to avoid testing `translate' | |
5290 | inside the loop. */ | |
5291 | if (translate) | |
5292 | while (range > lim | |
5293 | && !fastmap[(unsigned char) | |
5294 | translate[(unsigned char) *d++]]) | |
5295 | range--; | |
5296 | else | |
5297 | while (range > lim && !fastmap[(unsigned char) *d++]) | |
5298 | range--; | |
5299 | ||
5300 | startpos += irange - range; | |
5301 | } | |
5302 | else /* Searching backwards. */ | |
5303 | { | |
5304 | register CHAR_T c = (size1 == 0 || startpos >= size1 | |
5305 | ? string2[startpos - size1] | |
5306 | : string1[startpos]); | |
5307 | ||
5308 | if (!fastmap[(unsigned char) TRANSLATE (c)]) | |
5309 | goto advance; | |
5310 | } | |
5311 | } | |
5312 | ||
5313 | /* If can't match the null string, and that's all we have left, fail. */ | |
5314 | if (range >= 0 && startpos == total_size && fastmap | |
5315 | && !bufp->can_be_null) | |
5316 | { | |
5317 | #ifdef WCHAR | |
5318 | FREE_WCS_BUFFERS (); | |
5319 | #endif | |
5320 | return -1; | |
5321 | } | |
5322 | ||
5323 | #ifdef WCHAR | |
5324 | val = wcs_re_match_2_internal (bufp, string1, size1, string2, | |
5325 | size2, startpos, regs, stop, | |
5326 | wcs_string1, wcs_size1, | |
5327 | wcs_string2, wcs_size2, | |
5328 | mbs_offset1, mbs_offset2); | |
5329 | #else /* BYTE */ | |
5330 | val = byte_re_match_2_internal (bufp, string1, size1, string2, | |
5331 | size2, startpos, regs, stop); | |
5332 | #endif /* BYTE */ | |
5333 | ||
5334 | #ifndef REGEX_MALLOC | |
5335 | # ifdef C_ALLOCA | |
5336 | alloca (0); | |
5337 | # endif | |
5338 | #endif | |
5339 | ||
5340 | if (val >= 0) | |
5341 | { | |
5342 | #ifdef WCHAR | |
5343 | FREE_WCS_BUFFERS (); | |
5344 | #endif | |
5345 | return startpos; | |
5346 | } | |
5347 | ||
5348 | if (val == -2) | |
5349 | { | |
5350 | #ifdef WCHAR | |
5351 | FREE_WCS_BUFFERS (); | |
5352 | #endif | |
5353 | return -2; | |
5354 | } | |
5355 | ||
5356 | advance: | |
5357 | if (!range) | |
5358 | break; | |
5359 | else if (range > 0) | |
5360 | { | |
5361 | range--; | |
5362 | startpos++; | |
5363 | } | |
5364 | else | |
5365 | { | |
5366 | range++; | |
5367 | startpos--; | |
5368 | } | |
5369 | } | |
5370 | #ifdef WCHAR | |
5371 | FREE_WCS_BUFFERS (); | |
5372 | #endif | |
5373 | return -1; | |
5374 | } | |
5375 | ||
5376 | #ifdef WCHAR | |
5377 | /* This converts PTR, a pointer into one of the search wchar_t strings | |
5378 | `string1' and `string2' into an multibyte string offset from the | |
5379 | beginning of that string. We use mbs_offset to optimize. | |
5380 | See convert_mbs_to_wcs. */ | |
5381 | # define POINTER_TO_OFFSET(ptr) \ | |
5382 | (FIRST_STRING_P (ptr) \ | |
5383 | ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \ | |
5384 | : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \ | |
5385 | + csize1))) | |
5386 | #else /* BYTE */ | |
5387 | /* This converts PTR, a pointer into one of the search strings `string1' | |
5388 | and `string2' into an offset from the beginning of that string. */ | |
5389 | # define POINTER_TO_OFFSET(ptr) \ | |
5390 | (FIRST_STRING_P (ptr) \ | |
5391 | ? ((regoff_t) ((ptr) - string1)) \ | |
5392 | : ((regoff_t) ((ptr) - string2 + size1))) | |
5393 | #endif /* WCHAR */ | |
5394 | ||
5395 | /* Macros for dealing with the split strings in re_match_2. */ | |
5396 | ||
5397 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
5398 | ||
5399 | /* Call before fetching a character with *d. This switches over to | |
5400 | string2 if necessary. */ | |
5401 | #define PREFETCH() \ | |
5402 | while (d == dend) \ | |
5403 | { \ | |
5404 | /* End of string2 => fail. */ \ | |
5405 | if (dend == end_match_2) \ | |
5406 | goto fail; \ | |
5407 | /* End of string1 => advance to string2. */ \ | |
5408 | d = string2; \ | |
5409 | dend = end_match_2; \ | |
5410 | } | |
5411 | ||
5412 | /* Test if at very beginning or at very end of the virtual concatenation | |
5413 | of `string1' and `string2'. If only one string, it's `string2'. */ | |
5414 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) | |
5415 | #define AT_STRINGS_END(d) ((d) == end2) | |
5416 | ||
5417 | ||
5418 | /* Test if D points to a character which is word-constituent. We have | |
5419 | two special cases to check for: if past the end of string1, look at | |
5420 | the first character in string2; and if before the beginning of | |
5421 | string2, look at the last character in string1. */ | |
5422 | #ifdef WCHAR | |
5423 | /* Use internationalized API instead of SYNTAX. */ | |
5424 | # define WORDCHAR_P(d) \ | |
5425 | (iswalnum ((wint_t)((d) == end1 ? *string2 \ | |
6ad8a379 DD |
5426 | : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \ |
5427 | || ((d) == end1 ? *string2 \ | |
5428 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_') | |
2a6ef469 DD |
5429 | #else /* BYTE */ |
5430 | # define WORDCHAR_P(d) \ | |
5431 | (SYNTAX ((d) == end1 ? *string2 \ | |
5432 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ | |
5433 | == Sword) | |
5434 | #endif /* WCHAR */ | |
5435 | ||
5436 | /* Disabled due to a compiler bug -- see comment at case wordbound */ | |
5437 | #if 0 | |
5438 | /* Test if the character before D and the one at D differ with respect | |
5439 | to being word-constituent. */ | |
5440 | #define AT_WORD_BOUNDARY(d) \ | |
5441 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
5442 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
5443 | #endif | |
5444 | ||
5445 | /* Free everything we malloc. */ | |
5446 | #ifdef MATCH_MAY_ALLOCATE | |
5447 | # ifdef WCHAR | |
5448 | # define FREE_VARIABLES() \ | |
5449 | do { \ | |
5450 | REGEX_FREE_STACK (fail_stack.stack); \ | |
5451 | FREE_VAR (regstart); \ | |
5452 | FREE_VAR (regend); \ | |
5453 | FREE_VAR (old_regstart); \ | |
5454 | FREE_VAR (old_regend); \ | |
5455 | FREE_VAR (best_regstart); \ | |
5456 | FREE_VAR (best_regend); \ | |
5457 | FREE_VAR (reg_info); \ | |
5458 | FREE_VAR (reg_dummy); \ | |
5459 | FREE_VAR (reg_info_dummy); \ | |
5460 | if (!cant_free_wcs_buf) \ | |
5461 | { \ | |
5462 | FREE_VAR (string1); \ | |
5463 | FREE_VAR (string2); \ | |
5464 | FREE_VAR (mbs_offset1); \ | |
5465 | FREE_VAR (mbs_offset2); \ | |
5466 | } \ | |
5467 | } while (0) | |
5468 | # else /* BYTE */ | |
5469 | # define FREE_VARIABLES() \ | |
5470 | do { \ | |
5471 | REGEX_FREE_STACK (fail_stack.stack); \ | |
5472 | FREE_VAR (regstart); \ | |
5473 | FREE_VAR (regend); \ | |
5474 | FREE_VAR (old_regstart); \ | |
5475 | FREE_VAR (old_regend); \ | |
5476 | FREE_VAR (best_regstart); \ | |
5477 | FREE_VAR (best_regend); \ | |
5478 | FREE_VAR (reg_info); \ | |
5479 | FREE_VAR (reg_dummy); \ | |
5480 | FREE_VAR (reg_info_dummy); \ | |
5481 | } while (0) | |
5482 | # endif /* WCHAR */ | |
5483 | #else | |
5484 | # ifdef WCHAR | |
5485 | # define FREE_VARIABLES() \ | |
5486 | do { \ | |
5487 | if (!cant_free_wcs_buf) \ | |
5488 | { \ | |
5489 | FREE_VAR (string1); \ | |
5490 | FREE_VAR (string2); \ | |
5491 | FREE_VAR (mbs_offset1); \ | |
5492 | FREE_VAR (mbs_offset2); \ | |
5493 | } \ | |
5494 | } while (0) | |
5495 | # else /* BYTE */ | |
5496 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ | |
5497 | # endif /* WCHAR */ | |
5498 | #endif /* not MATCH_MAY_ALLOCATE */ | |
5499 | ||
5500 | /* These values must meet several constraints. They must not be valid | |
5501 | register values; since we have a limit of 255 registers (because | |
5502 | we use only one byte in the pattern for the register number), we can | |
5503 | use numbers larger than 255. They must differ by 1, because of | |
5504 | NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
5505 | be larger than the value for the highest register, so we do not try | |
5506 | to actually save any registers when none are active. */ | |
5507 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
5508 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
5509 | \f | |
5510 | #else /* not INSIDE_RECURSION */ | |
5511 | /* Matching routines. */ | |
5512 | ||
5513 | #ifndef emacs /* Emacs never uses this. */ | |
5514 | /* re_match is like re_match_2 except it takes only a single string. */ | |
5515 | ||
5516 | int | |
5517 | re_match (bufp, string, size, pos, regs) | |
5518 | struct re_pattern_buffer *bufp; | |
5519 | const char *string; | |
5520 | int size, pos; | |
5521 | struct re_registers *regs; | |
5522 | { | |
5523 | int result; | |
5524 | # ifdef MBS_SUPPORT | |
5525 | if (MB_CUR_MAX != 1) | |
5526 | result = wcs_re_match_2_internal (bufp, NULL, 0, string, size, | |
5527 | pos, regs, size, | |
5528 | NULL, 0, NULL, 0, NULL, NULL); | |
5529 | else | |
5530 | # endif | |
5531 | result = byte_re_match_2_internal (bufp, NULL, 0, string, size, | |
5532 | pos, regs, size); | |
5533 | # ifndef REGEX_MALLOC | |
5534 | # ifdef C_ALLOCA | |
5535 | alloca (0); | |
5536 | # endif | |
5537 | # endif | |
5538 | return result; | |
5539 | } | |
5540 | # ifdef _LIBC | |
5541 | weak_alias (__re_match, re_match) | |
5542 | # endif | |
5543 | #endif /* not emacs */ | |
5544 | ||
5545 | #endif /* not INSIDE_RECURSION */ | |
5546 | ||
5547 | #ifdef INSIDE_RECURSION | |
5548 | static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p, | |
5549 | UCHAR_T *end, | |
5550 | PREFIX(register_info_type) *reg_info)); | |
5551 | static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p, | |
5552 | UCHAR_T *end, | |
5553 | PREFIX(register_info_type) *reg_info)); | |
5554 | static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p, | |
5555 | UCHAR_T *end, | |
5556 | PREFIX(register_info_type) *reg_info)); | |
5557 | static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2, | |
5558 | int len, char *translate)); | |
5559 | #else /* not INSIDE_RECURSION */ | |
5560 | ||
5561 | /* re_match_2 matches the compiled pattern in BUFP against the | |
5562 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
5563 | and SIZE2, respectively). We start matching at POS, and stop | |
5564 | matching at STOP. | |
5565 | ||
5566 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
5567 | store offsets for the substring each group matched in REGS. See the | |
5568 | documentation for exactly how many groups we fill. | |
5569 | ||
5570 | We return -1 if no match, -2 if an internal error (such as the | |
5571 | failure stack overflowing). Otherwise, we return the length of the | |
5572 | matched substring. */ | |
5573 | ||
5574 | int | |
5575 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
5576 | struct re_pattern_buffer *bufp; | |
5577 | const char *string1, *string2; | |
5578 | int size1, size2; | |
5579 | int pos; | |
5580 | struct re_registers *regs; | |
5581 | int stop; | |
5582 | { | |
5583 | int result; | |
5584 | # ifdef MBS_SUPPORT | |
5585 | if (MB_CUR_MAX != 1) | |
5586 | result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2, | |
5587 | pos, regs, stop, | |
5588 | NULL, 0, NULL, 0, NULL, NULL); | |
5589 | else | |
5590 | # endif | |
5591 | result = byte_re_match_2_internal (bufp, string1, size1, string2, size2, | |
5592 | pos, regs, stop); | |
5593 | ||
5594 | #ifndef REGEX_MALLOC | |
5595 | # ifdef C_ALLOCA | |
5596 | alloca (0); | |
5597 | # endif | |
5598 | #endif | |
5599 | return result; | |
5600 | } | |
5601 | #ifdef _LIBC | |
5602 | weak_alias (__re_match_2, re_match_2) | |
5603 | #endif | |
5604 | ||
5605 | #endif /* not INSIDE_RECURSION */ | |
5606 | ||
5607 | #ifdef INSIDE_RECURSION | |
5608 | ||
5609 | #ifdef WCHAR | |
5610 | static int count_mbs_length PARAMS ((int *, int)); | |
5611 | ||
5612 | /* This check the substring (from 0, to length) of the multibyte string, | |
5613 | to which offset_buffer correspond. And count how many wchar_t_characters | |
5614 | the substring occupy. We use offset_buffer to optimization. | |
5615 | See convert_mbs_to_wcs. */ | |
5616 | ||
5617 | static int | |
5618 | count_mbs_length(offset_buffer, length) | |
5619 | int *offset_buffer; | |
5620 | int length; | |
5621 | { | |
5622 | int upper, lower; | |
5623 | ||
5624 | /* Check whether the size is valid. */ | |
5625 | if (length < 0) | |
5626 | return -1; | |
5627 | ||
5628 | if (offset_buffer == NULL) | |
5629 | return 0; | |
5630 | ||
5631 | /* If there are no multibyte character, offset_buffer[i] == i. | |
5632 | Optmize for this case. */ | |
5633 | if (offset_buffer[length] == length) | |
5634 | return length; | |
5635 | ||
5636 | /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */ | |
5637 | upper = length; | |
5638 | lower = 0; | |
5639 | ||
5640 | while (true) | |
5641 | { | |
5642 | int middle = (lower + upper) / 2; | |
5643 | if (middle == lower || middle == upper) | |
5644 | break; | |
5645 | if (offset_buffer[middle] > length) | |
5646 | upper = middle; | |
5647 | else if (offset_buffer[middle] < length) | |
5648 | lower = middle; | |
5649 | else | |
5650 | return middle; | |
5651 | } | |
5652 | ||
5653 | return -1; | |
5654 | } | |
5655 | #endif /* WCHAR */ | |
5656 | ||
5657 | /* This is a separate function so that we can force an alloca cleanup | |
5658 | afterwards. */ | |
5659 | #ifdef WCHAR | |
5660 | static int | |
5661 | wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos, | |
5662 | regs, stop, string1, size1, string2, size2, | |
5663 | mbs_offset1, mbs_offset2) | |
5664 | struct re_pattern_buffer *bufp; | |
5665 | const char *cstring1, *cstring2; | |
5666 | int csize1, csize2; | |
5667 | int pos; | |
5668 | struct re_registers *regs; | |
5669 | int stop; | |
5670 | /* string1 == string2 == NULL means string1/2, size1/2 and | |
5671 | mbs_offset1/2 need seting up in this function. */ | |
5672 | /* We need wchar_t* buffers correspond to cstring1, cstring2. */ | |
5673 | wchar_t *string1, *string2; | |
5674 | /* We need the size of wchar_t buffers correspond to csize1, csize2. */ | |
5675 | int size1, size2; | |
5676 | /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ | |
5677 | int *mbs_offset1, *mbs_offset2; | |
5678 | #else /* BYTE */ | |
5679 | static int | |
5680 | byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos, | |
5681 | regs, stop) | |
5682 | struct re_pattern_buffer *bufp; | |
5683 | const char *string1, *string2; | |
5684 | int size1, size2; | |
5685 | int pos; | |
5686 | struct re_registers *regs; | |
5687 | int stop; | |
5688 | #endif /* BYTE */ | |
5689 | { | |
5690 | /* General temporaries. */ | |
5691 | int mcnt; | |
5692 | UCHAR_T *p1; | |
5693 | #ifdef WCHAR | |
5694 | /* They hold whether each wchar_t is binary data or not. */ | |
5695 | char *is_binary = NULL; | |
5696 | /* If true, we can't free string1/2, mbs_offset1/2. */ | |
5697 | int cant_free_wcs_buf = 1; | |
5698 | #endif /* WCHAR */ | |
5699 | ||
5700 | /* Just past the end of the corresponding string. */ | |
5701 | const CHAR_T *end1, *end2; | |
5702 | ||
5703 | /* Pointers into string1 and string2, just past the last characters in | |
5704 | each to consider matching. */ | |
5705 | const CHAR_T *end_match_1, *end_match_2; | |
5706 | ||
5707 | /* Where we are in the data, and the end of the current string. */ | |
5708 | const CHAR_T *d, *dend; | |
5709 | ||
5710 | /* Where we are in the pattern, and the end of the pattern. */ | |
5711 | #ifdef WCHAR | |
5712 | UCHAR_T *pattern, *p; | |
5713 | register UCHAR_T *pend; | |
5714 | #else /* BYTE */ | |
5715 | UCHAR_T *p = bufp->buffer; | |
5716 | register UCHAR_T *pend = p + bufp->used; | |
5717 | #endif /* WCHAR */ | |
5718 | ||
5719 | /* Mark the opcode just after a start_memory, so we can test for an | |
5720 | empty subpattern when we get to the stop_memory. */ | |
5721 | UCHAR_T *just_past_start_mem = 0; | |
5722 | ||
5723 | /* We use this to map every character in the string. */ | |
5724 | RE_TRANSLATE_TYPE translate = bufp->translate; | |
5725 | ||
5726 | /* Failure point stack. Each place that can handle a failure further | |
5727 | down the line pushes a failure point on this stack. It consists of | |
5728 | restart, regend, and reg_info for all registers corresponding to | |
5729 | the subexpressions we're currently inside, plus the number of such | |
5730 | registers, and, finally, two char *'s. The first char * is where | |
5731 | to resume scanning the pattern; the second one is where to resume | |
5732 | scanning the strings. If the latter is zero, the failure point is | |
5733 | a ``dummy''; if a failure happens and the failure point is a dummy, | |
5734 | it gets discarded and the next next one is tried. */ | |
5735 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
5736 | PREFIX(fail_stack_type) fail_stack; | |
5737 | #endif | |
5738 | #ifdef DEBUG | |
5739 | static unsigned failure_id; | |
5740 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; | |
5741 | #endif | |
5742 | ||
5743 | #ifdef REL_ALLOC | |
5744 | /* This holds the pointer to the failure stack, when | |
5745 | it is allocated relocatably. */ | |
5746 | fail_stack_elt_t *failure_stack_ptr; | |
5747 | #endif | |
5748 | ||
5749 | /* We fill all the registers internally, independent of what we | |
5750 | return, for use in backreferences. The number here includes | |
5751 | an element for register zero. */ | |
5752 | size_t num_regs = bufp->re_nsub + 1; | |
5753 | ||
5754 | /* The currently active registers. */ | |
5755 | active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
5756 | active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
5757 | ||
5758 | /* Information on the contents of registers. These are pointers into | |
5759 | the input strings; they record just what was matched (on this | |
5760 | attempt) by a subexpression part of the pattern, that is, the | |
5761 | regnum-th regstart pointer points to where in the pattern we began | |
5762 | matching and the regnum-th regend points to right after where we | |
5763 | stopped matching the regnum-th subexpression. (The zeroth register | |
5764 | keeps track of what the whole pattern matches.) */ | |
5765 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
5766 | const CHAR_T **regstart, **regend; | |
5767 | #endif | |
5768 | ||
5769 | /* If a group that's operated upon by a repetition operator fails to | |
5770 | match anything, then the register for its start will need to be | |
5771 | restored because it will have been set to wherever in the string we | |
5772 | are when we last see its open-group operator. Similarly for a | |
5773 | register's end. */ | |
5774 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
5775 | const CHAR_T **old_regstart, **old_regend; | |
5776 | #endif | |
5777 | ||
5778 | /* The is_active field of reg_info helps us keep track of which (possibly | |
5779 | nested) subexpressions we are currently in. The matched_something | |
5780 | field of reg_info[reg_num] helps us tell whether or not we have | |
5781 | matched any of the pattern so far this time through the reg_num-th | |
5782 | subexpression. These two fields get reset each time through any | |
5783 | loop their register is in. */ | |
5784 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
5785 | PREFIX(register_info_type) *reg_info; | |
5786 | #endif | |
5787 | ||
5788 | /* The following record the register info as found in the above | |
5789 | variables when we find a match better than any we've seen before. | |
5790 | This happens as we backtrack through the failure points, which in | |
5791 | turn happens only if we have not yet matched the entire string. */ | |
5792 | unsigned best_regs_set = false; | |
5793 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
5794 | const CHAR_T **best_regstart, **best_regend; | |
5795 | #endif | |
5796 | ||
5797 | /* Logically, this is `best_regend[0]'. But we don't want to have to | |
5798 | allocate space for that if we're not allocating space for anything | |
5799 | else (see below). Also, we never need info about register 0 for | |
5800 | any of the other register vectors, and it seems rather a kludge to | |
5801 | treat `best_regend' differently than the rest. So we keep track of | |
5802 | the end of the best match so far in a separate variable. We | |
5803 | initialize this to NULL so that when we backtrack the first time | |
5804 | and need to test it, it's not garbage. */ | |
5805 | const CHAR_T *match_end = NULL; | |
5806 | ||
5807 | /* This helps SET_REGS_MATCHED avoid doing redundant work. */ | |
5808 | int set_regs_matched_done = 0; | |
5809 | ||
5810 | /* Used when we pop values we don't care about. */ | |
5811 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
5812 | const CHAR_T **reg_dummy; | |
5813 | PREFIX(register_info_type) *reg_info_dummy; | |
5814 | #endif | |
5815 | ||
5816 | #ifdef DEBUG | |
5817 | /* Counts the total number of registers pushed. */ | |
5818 | unsigned num_regs_pushed = 0; | |
5819 | #endif | |
5820 | ||
5821 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
5822 | ||
5823 | INIT_FAIL_STACK (); | |
5824 | ||
5825 | #ifdef MATCH_MAY_ALLOCATE | |
5826 | /* Do not bother to initialize all the register variables if there are | |
5827 | no groups in the pattern, as it takes a fair amount of time. If | |
5828 | there are groups, we include space for register 0 (the whole | |
5829 | pattern), even though we never use it, since it simplifies the | |
5830 | array indexing. We should fix this. */ | |
5831 | if (bufp->re_nsub) | |
5832 | { | |
5833 | regstart = REGEX_TALLOC (num_regs, const CHAR_T *); | |
5834 | regend = REGEX_TALLOC (num_regs, const CHAR_T *); | |
5835 | old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); | |
5836 | old_regend = REGEX_TALLOC (num_regs, const CHAR_T *); | |
5837 | best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); | |
5838 | best_regend = REGEX_TALLOC (num_regs, const CHAR_T *); | |
5839 | reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); | |
5840 | reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *); | |
5841 | reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); | |
5842 | ||
5843 | if (!(regstart && regend && old_regstart && old_regend && reg_info | |
5844 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
5845 | { | |
5846 | FREE_VARIABLES (); | |
5847 | return -2; | |
5848 | } | |
5849 | } | |
5850 | else | |
5851 | { | |
5852 | /* We must initialize all our variables to NULL, so that | |
5853 | `FREE_VARIABLES' doesn't try to free them. */ | |
5854 | regstart = regend = old_regstart = old_regend = best_regstart | |
5855 | = best_regend = reg_dummy = NULL; | |
5856 | reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL; | |
5857 | } | |
5858 | #endif /* MATCH_MAY_ALLOCATE */ | |
5859 | ||
5860 | /* The starting position is bogus. */ | |
5861 | #ifdef WCHAR | |
5862 | if (pos < 0 || pos > csize1 + csize2) | |
5863 | #else /* BYTE */ | |
5864 | if (pos < 0 || pos > size1 + size2) | |
5865 | #endif | |
5866 | { | |
5867 | FREE_VARIABLES (); | |
5868 | return -1; | |
5869 | } | |
5870 | ||
5871 | #ifdef WCHAR | |
5872 | /* Allocate wchar_t array for string1 and string2 and | |
5873 | fill them with converted string. */ | |
5874 | if (string1 == NULL && string2 == NULL) | |
5875 | { | |
5876 | /* We need seting up buffers here. */ | |
5877 | ||
5878 | /* We must free wcs buffers in this function. */ | |
5879 | cant_free_wcs_buf = 0; | |
5880 | ||
5881 | if (csize1 != 0) | |
5882 | { | |
5883 | string1 = REGEX_TALLOC (csize1 + 1, CHAR_T); | |
5884 | mbs_offset1 = REGEX_TALLOC (csize1 + 1, int); | |
5885 | is_binary = REGEX_TALLOC (csize1 + 1, char); | |
5886 | if (!string1 || !mbs_offset1 || !is_binary) | |
5887 | { | |
5888 | FREE_VAR (string1); | |
5889 | FREE_VAR (mbs_offset1); | |
5890 | FREE_VAR (is_binary); | |
5891 | return -2; | |
5892 | } | |
5893 | } | |
5894 | if (csize2 != 0) | |
5895 | { | |
5896 | string2 = REGEX_TALLOC (csize2 + 1, CHAR_T); | |
5897 | mbs_offset2 = REGEX_TALLOC (csize2 + 1, int); | |
5898 | is_binary = REGEX_TALLOC (csize2 + 1, char); | |
5899 | if (!string2 || !mbs_offset2 || !is_binary) | |
5900 | { | |
5901 | FREE_VAR (string1); | |
5902 | FREE_VAR (mbs_offset1); | |
5903 | FREE_VAR (string2); | |
5904 | FREE_VAR (mbs_offset2); | |
5905 | FREE_VAR (is_binary); | |
5906 | return -2; | |
5907 | } | |
5908 | size2 = convert_mbs_to_wcs(string2, cstring2, csize2, | |
5909 | mbs_offset2, is_binary); | |
5910 | string2[size2] = L'\0'; /* for a sentinel */ | |
5911 | FREE_VAR (is_binary); | |
5912 | } | |
5913 | } | |
5914 | ||
5915 | /* We need to cast pattern to (wchar_t*), because we casted this compiled | |
5916 | pattern to (char*) in regex_compile. */ | |
5917 | p = pattern = (CHAR_T*)bufp->buffer; | |
5918 | pend = (CHAR_T*)(bufp->buffer + bufp->used); | |
5919 | ||
5920 | #endif /* WCHAR */ | |
5921 | ||
5922 | /* Initialize subexpression text positions to -1 to mark ones that no | |
5923 | start_memory/stop_memory has been seen for. Also initialize the | |
5924 | register information struct. */ | |
5925 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) | |
5926 | { | |
5927 | regstart[mcnt] = regend[mcnt] | |
5928 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
5929 | ||
5930 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
5931 | IS_ACTIVE (reg_info[mcnt]) = 0; | |
5932 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
5933 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
5934 | } | |
5935 | ||
5936 | /* We move `string1' into `string2' if the latter's empty -- but not if | |
5937 | `string1' is null. */ | |
5938 | if (size2 == 0 && string1 != NULL) | |
5939 | { | |
5940 | string2 = string1; | |
5941 | size2 = size1; | |
5942 | string1 = 0; | |
5943 | size1 = 0; | |
5944 | #ifdef WCHAR | |
5945 | mbs_offset2 = mbs_offset1; | |
5946 | csize2 = csize1; | |
5947 | mbs_offset1 = NULL; | |
5948 | csize1 = 0; | |
5949 | #endif | |
5950 | } | |
5951 | end1 = string1 + size1; | |
5952 | end2 = string2 + size2; | |
5953 | ||
5954 | /* Compute where to stop matching, within the two strings. */ | |
5955 | #ifdef WCHAR | |
5956 | if (stop <= csize1) | |
5957 | { | |
5958 | mcnt = count_mbs_length(mbs_offset1, stop); | |
5959 | end_match_1 = string1 + mcnt; | |
5960 | end_match_2 = string2; | |
5961 | } | |
5962 | else | |
5963 | { | |
5964 | if (stop > csize1 + csize2) | |
5965 | stop = csize1 + csize2; | |
5966 | end_match_1 = end1; | |
5967 | mcnt = count_mbs_length(mbs_offset2, stop-csize1); | |
5968 | end_match_2 = string2 + mcnt; | |
5969 | } | |
5970 | if (mcnt < 0) | |
5971 | { /* count_mbs_length return error. */ | |
5972 | FREE_VARIABLES (); | |
5973 | return -1; | |
5974 | } | |
5975 | #else | |
5976 | if (stop <= size1) | |
5977 | { | |
5978 | end_match_1 = string1 + stop; | |
5979 | end_match_2 = string2; | |
5980 | } | |
5981 | else | |
5982 | { | |
5983 | end_match_1 = end1; | |
5984 | end_match_2 = string2 + stop - size1; | |
5985 | } | |
5986 | #endif /* WCHAR */ | |
5987 | ||
5988 | /* `p' scans through the pattern as `d' scans through the data. | |
5989 | `dend' is the end of the input string that `d' points within. `d' | |
5990 | is advanced into the following input string whenever necessary, but | |
5991 | this happens before fetching; therefore, at the beginning of the | |
5992 | loop, `d' can be pointing at the end of a string, but it cannot | |
5993 | equal `string2'. */ | |
5994 | #ifdef WCHAR | |
5995 | if (size1 > 0 && pos <= csize1) | |
5996 | { | |
5997 | mcnt = count_mbs_length(mbs_offset1, pos); | |
5998 | d = string1 + mcnt; | |
5999 | dend = end_match_1; | |
6000 | } | |
6001 | else | |
6002 | { | |
6003 | mcnt = count_mbs_length(mbs_offset2, pos-csize1); | |
6004 | d = string2 + mcnt; | |
6005 | dend = end_match_2; | |
6006 | } | |
6007 | ||
6008 | if (mcnt < 0) | |
6009 | { /* count_mbs_length return error. */ | |
6010 | FREE_VARIABLES (); | |
6011 | return -1; | |
6012 | } | |
6013 | #else | |
6014 | if (size1 > 0 && pos <= size1) | |
6015 | { | |
6016 | d = string1 + pos; | |
6017 | dend = end_match_1; | |
6018 | } | |
6019 | else | |
6020 | { | |
6021 | d = string2 + pos - size1; | |
6022 | dend = end_match_2; | |
6023 | } | |
6024 | #endif /* WCHAR */ | |
6025 | ||
6026 | DEBUG_PRINT1 ("The compiled pattern is:\n"); | |
6027 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
6028 | DEBUG_PRINT1 ("The string to match is: `"); | |
6029 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
6030 | DEBUG_PRINT1 ("'\n"); | |
6031 | ||
6032 | /* This loops over pattern commands. It exits by returning from the | |
6033 | function if the match is complete, or it drops through if the match | |
6034 | fails at this starting point in the input data. */ | |
6035 | for (;;) | |
6036 | { | |
6037 | #ifdef _LIBC | |
6038 | DEBUG_PRINT2 ("\n%p: ", p); | |
6039 | #else | |
6040 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
6041 | #endif | |
6042 | ||
6043 | if (p == pend) | |
6044 | { /* End of pattern means we might have succeeded. */ | |
6045 | DEBUG_PRINT1 ("end of pattern ... "); | |
6046 | ||
6047 | /* If we haven't matched the entire string, and we want the | |
6048 | longest match, try backtracking. */ | |
6049 | if (d != end_match_2) | |
6050 | { | |
6051 | /* 1 if this match ends in the same string (string1 or string2) | |
6052 | as the best previous match. */ | |
6053 | boolean same_str_p = (FIRST_STRING_P (match_end) | |
6054 | == MATCHING_IN_FIRST_STRING); | |
6055 | /* 1 if this match is the best seen so far. */ | |
6056 | boolean best_match_p; | |
6057 | ||
6058 | /* AIX compiler got confused when this was combined | |
6059 | with the previous declaration. */ | |
6060 | if (same_str_p) | |
6061 | best_match_p = d > match_end; | |
6062 | else | |
6063 | best_match_p = !MATCHING_IN_FIRST_STRING; | |
6064 | ||
6065 | DEBUG_PRINT1 ("backtracking.\n"); | |
6066 | ||
6067 | if (!FAIL_STACK_EMPTY ()) | |
6068 | { /* More failure points to try. */ | |
6069 | ||
6070 | /* If exceeds best match so far, save it. */ | |
6071 | if (!best_regs_set || best_match_p) | |
6072 | { | |
6073 | best_regs_set = true; | |
6074 | match_end = d; | |
6075 | ||
6076 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
6077 | ||
6078 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) | |
6079 | { | |
6080 | best_regstart[mcnt] = regstart[mcnt]; | |
6081 | best_regend[mcnt] = regend[mcnt]; | |
6082 | } | |
6083 | } | |
6084 | goto fail; | |
6085 | } | |
6086 | ||
6087 | /* If no failure points, don't restore garbage. And if | |
6088 | last match is real best match, don't restore second | |
6089 | best one. */ | |
6090 | else if (best_regs_set && !best_match_p) | |
6091 | { | |
6092 | restore_best_regs: | |
6093 | /* Restore best match. It may happen that `dend == | |
6094 | end_match_1' while the restored d is in string2. | |
6095 | For example, the pattern `x.*y.*z' against the | |
6096 | strings `x-' and `y-z-', if the two strings are | |
6097 | not consecutive in memory. */ | |
6098 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
6099 | ||
6100 | d = match_end; | |
6101 | dend = ((d >= string1 && d <= end1) | |
6102 | ? end_match_1 : end_match_2); | |
6103 | ||
6104 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) | |
6105 | { | |
6106 | regstart[mcnt] = best_regstart[mcnt]; | |
6107 | regend[mcnt] = best_regend[mcnt]; | |
6108 | } | |
6109 | } | |
6110 | } /* d != end_match_2 */ | |
6111 | ||
6112 | succeed_label: | |
6113 | DEBUG_PRINT1 ("Accepting match.\n"); | |
6114 | /* If caller wants register contents data back, do it. */ | |
6115 | if (regs && !bufp->no_sub) | |
6116 | { | |
6117 | /* Have the register data arrays been allocated? */ | |
6118 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
6119 | { /* No. So allocate them with malloc. We need one | |
6120 | extra element beyond `num_regs' for the `-1' marker | |
6121 | GNU code uses. */ | |
6122 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
6123 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
6124 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
6125 | if (regs->start == NULL || regs->end == NULL) | |
6126 | { | |
6127 | FREE_VARIABLES (); | |
6128 | return -2; | |
6129 | } | |
6130 | bufp->regs_allocated = REGS_REALLOCATE; | |
6131 | } | |
6132 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
6133 | { /* Yes. If we need more elements than were already | |
6134 | allocated, reallocate them. If we need fewer, just | |
6135 | leave it alone. */ | |
6136 | if (regs->num_regs < num_regs + 1) | |
6137 | { | |
6138 | regs->num_regs = num_regs + 1; | |
6139 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
6140 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
6141 | if (regs->start == NULL || regs->end == NULL) | |
6142 | { | |
6143 | FREE_VARIABLES (); | |
6144 | return -2; | |
6145 | } | |
6146 | } | |
6147 | } | |
6148 | else | |
6149 | { | |
6150 | /* These braces fend off a "empty body in an else-statement" | |
6151 | warning under GCC when assert expands to nothing. */ | |
6152 | assert (bufp->regs_allocated == REGS_FIXED); | |
6153 | } | |
6154 | ||
6155 | /* Convert the pointer data in `regstart' and `regend' to | |
6156 | indices. Register zero has to be set differently, | |
6157 | since we haven't kept track of any info for it. */ | |
6158 | if (regs->num_regs > 0) | |
6159 | { | |
6160 | regs->start[0] = pos; | |
6161 | #ifdef WCHAR | |
6162 | if (MATCHING_IN_FIRST_STRING) | |
6163 | regs->end[0] = mbs_offset1 != NULL ? | |
6164 | mbs_offset1[d-string1] : 0; | |
6165 | else | |
6166 | regs->end[0] = csize1 + (mbs_offset2 != NULL ? | |
6167 | mbs_offset2[d-string2] : 0); | |
6168 | #else | |
6169 | regs->end[0] = (MATCHING_IN_FIRST_STRING | |
6170 | ? ((regoff_t) (d - string1)) | |
6171 | : ((regoff_t) (d - string2 + size1))); | |
6172 | #endif /* WCHAR */ | |
6173 | } | |
6174 | ||
6175 | /* Go through the first `min (num_regs, regs->num_regs)' | |
6176 | registers, since that is all we initialized. */ | |
6177 | for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs); | |
6178 | mcnt++) | |
6179 | { | |
6180 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
6181 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
6182 | else | |
6183 | { | |
6184 | regs->start[mcnt] | |
6185 | = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); | |
6186 | regs->end[mcnt] | |
6187 | = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); | |
6188 | } | |
6189 | } | |
6190 | ||
6191 | /* If the regs structure we return has more elements than | |
6192 | were in the pattern, set the extra elements to -1. If | |
6193 | we (re)allocated the registers, this is the case, | |
6194 | because we always allocate enough to have at least one | |
6195 | -1 at the end. */ | |
6196 | for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++) | |
6197 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
6198 | } /* regs && !bufp->no_sub */ | |
6199 | ||
6200 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", | |
6201 | nfailure_points_pushed, nfailure_points_popped, | |
6202 | nfailure_points_pushed - nfailure_points_popped); | |
6203 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
6204 | ||
6205 | #ifdef WCHAR | |
6206 | if (MATCHING_IN_FIRST_STRING) | |
6207 | mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0; | |
6208 | else | |
6209 | mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) + | |
6210 | csize1; | |
6211 | mcnt -= pos; | |
6212 | #else | |
6213 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
6214 | ? string1 | |
6215 | : string2 - size1); | |
6216 | #endif /* WCHAR */ | |
6217 | ||
6218 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
6219 | ||
6220 | FREE_VARIABLES (); | |
6221 | return mcnt; | |
6222 | } | |
6223 | ||
6224 | /* Otherwise match next pattern command. */ | |
6225 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) | |
6226 | { | |
6227 | /* Ignore these. Used to ignore the n of succeed_n's which | |
6228 | currently have n == 0. */ | |
6229 | case no_op: | |
6230 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
6231 | break; | |
6232 | ||
6233 | case succeed: | |
6234 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); | |
6235 | goto succeed_label; | |
6236 | ||
6237 | /* Match the next n pattern characters exactly. The following | |
6238 | byte in the pattern defines n, and the n bytes after that | |
6239 | are the characters to match. */ | |
6240 | case exactn: | |
6241 | #ifdef MBS_SUPPORT | |
6242 | case exactn_bin: | |
6243 | #endif | |
6244 | mcnt = *p++; | |
6245 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
6246 | ||
6247 | /* This is written out as an if-else so we don't waste time | |
6248 | testing `translate' inside the loop. */ | |
6249 | if (translate) | |
6250 | { | |
6251 | do | |
6252 | { | |
6253 | PREFETCH (); | |
6254 | #ifdef WCHAR | |
6255 | if (*d <= 0xff) | |
6256 | { | |
6257 | if ((UCHAR_T) translate[(unsigned char) *d++] | |
6258 | != (UCHAR_T) *p++) | |
6259 | goto fail; | |
6260 | } | |
6261 | else | |
6262 | { | |
6263 | if (*d++ != (CHAR_T) *p++) | |
6264 | goto fail; | |
6265 | } | |
6266 | #else | |
6267 | if ((UCHAR_T) translate[(unsigned char) *d++] | |
6268 | != (UCHAR_T) *p++) | |
6269 | goto fail; | |
6270 | #endif /* WCHAR */ | |
6271 | } | |
6272 | while (--mcnt); | |
6273 | } | |
6274 | else | |
6275 | { | |
6276 | do | |
6277 | { | |
6278 | PREFETCH (); | |
6279 | if (*d++ != (CHAR_T) *p++) goto fail; | |
6280 | } | |
6281 | while (--mcnt); | |
6282 | } | |
6283 | SET_REGS_MATCHED (); | |
6284 | break; | |
6285 | ||
6286 | ||
6287 | /* Match any character except possibly a newline or a null. */ | |
6288 | case anychar: | |
6289 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
6290 | ||
6291 | PREFETCH (); | |
6292 | ||
6293 | if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
6294 | || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
6295 | goto fail; | |
6296 | ||
6297 | SET_REGS_MATCHED (); | |
6298 | DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d); | |
6299 | d++; | |
6300 | break; | |
6301 | ||
6302 | ||
6303 | case charset: | |
6304 | case charset_not: | |
6305 | { | |
6306 | register UCHAR_T c; | |
6307 | #ifdef WCHAR | |
6308 | unsigned int i, char_class_length, coll_symbol_length, | |
6309 | equiv_class_length, ranges_length, chars_length, length; | |
6310 | CHAR_T *workp, *workp2, *charset_top; | |
6311 | #define WORK_BUFFER_SIZE 128 | |
6312 | CHAR_T str_buf[WORK_BUFFER_SIZE]; | |
6313 | # ifdef _LIBC | |
6314 | uint32_t nrules; | |
6315 | # endif /* _LIBC */ | |
6316 | #endif /* WCHAR */ | |
6317 | boolean not = (re_opcode_t) *(p - 1) == charset_not; | |
6318 | ||
6319 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
6320 | PREFETCH (); | |
6321 | c = TRANSLATE (*d); /* The character to match. */ | |
6322 | #ifdef WCHAR | |
6323 | # ifdef _LIBC | |
6324 | nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
6325 | # endif /* _LIBC */ | |
6326 | charset_top = p - 1; | |
6327 | char_class_length = *p++; | |
6328 | coll_symbol_length = *p++; | |
6329 | equiv_class_length = *p++; | |
6330 | ranges_length = *p++; | |
6331 | chars_length = *p++; | |
6332 | /* p points charset[6], so the address of the next instruction | |
6333 | (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'], | |
6334 | where l=length of char_classes, m=length of collating_symbol, | |
6335 | n=equivalence_class, o=length of char_range, | |
6336 | p'=length of character. */ | |
6337 | workp = p; | |
6338 | /* Update p to indicate the next instruction. */ | |
6339 | p += char_class_length + coll_symbol_length+ equiv_class_length + | |
6340 | 2*ranges_length + chars_length; | |
6341 | ||
6342 | /* match with char_class? */ | |
6343 | for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE) | |
6344 | { | |
6345 | wctype_t wctype; | |
6346 | uintptr_t alignedp = ((uintptr_t)workp | |
6347 | + __alignof__(wctype_t) - 1) | |
6348 | & ~(uintptr_t)(__alignof__(wctype_t) - 1); | |
6349 | wctype = *((wctype_t*)alignedp); | |
6350 | workp += CHAR_CLASS_SIZE; | |
dc676635 DD |
6351 | # ifdef _LIBC |
6352 | if (__iswctype((wint_t)c, wctype)) | |
6353 | goto char_set_matched; | |
6354 | # else | |
2a6ef469 DD |
6355 | if (iswctype((wint_t)c, wctype)) |
6356 | goto char_set_matched; | |
dc676635 | 6357 | # endif |
2a6ef469 DD |
6358 | } |
6359 | ||
6360 | /* match with collating_symbol? */ | |
6361 | # ifdef _LIBC | |
6362 | if (nrules != 0) | |
6363 | { | |
6364 | const unsigned char *extra = (const unsigned char *) | |
6365 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); | |
6366 | ||
6367 | for (workp2 = workp + coll_symbol_length ; workp < workp2 ; | |
6368 | workp++) | |
6369 | { | |
6370 | int32_t *wextra; | |
6371 | wextra = (int32_t*)(extra + *workp++); | |
6372 | for (i = 0; i < *wextra; ++i) | |
6373 | if (TRANSLATE(d[i]) != wextra[1 + i]) | |
6374 | break; | |
6375 | ||
6376 | if (i == *wextra) | |
6377 | { | |
6378 | /* Update d, however d will be incremented at | |
6379 | char_set_matched:, we decrement d here. */ | |
6380 | d += i - 1; | |
6381 | goto char_set_matched; | |
6382 | } | |
6383 | } | |
6384 | } | |
6385 | else /* (nrules == 0) */ | |
6386 | # endif | |
6387 | /* If we can't look up collation data, we use wcscoll | |
6388 | instead. */ | |
6389 | { | |
6390 | for (workp2 = workp + coll_symbol_length ; workp < workp2 ;) | |
6391 | { | |
6392 | const CHAR_T *backup_d = d, *backup_dend = dend; | |
dc676635 DD |
6393 | # ifdef _LIBC |
6394 | length = __wcslen (workp); | |
6395 | # else | |
6396 | length = wcslen (workp); | |
6397 | # endif | |
2a6ef469 DD |
6398 | |
6399 | /* If wcscoll(the collating symbol, whole string) > 0, | |
6400 | any substring of the string never match with the | |
6401 | collating symbol. */ | |
dc676635 DD |
6402 | # ifdef _LIBC |
6403 | if (__wcscoll (workp, d) > 0) | |
6404 | # else | |
6405 | if (wcscoll (workp, d) > 0) | |
6406 | # endif | |
2a6ef469 DD |
6407 | { |
6408 | workp += length + 1; | |
6409 | continue; | |
6410 | } | |
6411 | ||
6412 | /* First, we compare the collating symbol with | |
6413 | the first character of the string. | |
6414 | If it don't match, we add the next character to | |
6415 | the compare buffer in turn. */ | |
6416 | for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++) | |
6417 | { | |
6418 | int match; | |
6419 | if (d == dend) | |
6420 | { | |
6421 | if (dend == end_match_2) | |
6422 | break; | |
6423 | d = string2; | |
6424 | dend = end_match_2; | |
6425 | } | |
6426 | ||
6427 | /* add next character to the compare buffer. */ | |
6428 | str_buf[i] = TRANSLATE(*d); | |
6429 | str_buf[i+1] = '\0'; | |
6430 | ||
dc676635 DD |
6431 | # ifdef _LIBC |
6432 | match = __wcscoll (workp, str_buf); | |
6433 | # else | |
6434 | match = wcscoll (workp, str_buf); | |
6435 | # endif | |
2a6ef469 DD |
6436 | if (match == 0) |
6437 | goto char_set_matched; | |
6438 | ||
6439 | if (match < 0) | |
6440 | /* (str_buf > workp) indicate (str_buf + X > workp), | |
6441 | because for all X (str_buf + X > str_buf). | |
6442 | So we don't need continue this loop. */ | |
6443 | break; | |
6444 | ||
6445 | /* Otherwise(str_buf < workp), | |
6446 | (str_buf+next_character) may equals (workp). | |
6447 | So we continue this loop. */ | |
6448 | } | |
6449 | /* not matched */ | |
6450 | d = backup_d; | |
6451 | dend = backup_dend; | |
6452 | workp += length + 1; | |
6453 | } | |
6454 | } | |
6455 | /* match with equivalence_class? */ | |
6456 | # ifdef _LIBC | |
6457 | if (nrules != 0) | |
6458 | { | |
6459 | const CHAR_T *backup_d = d, *backup_dend = dend; | |
6460 | /* Try to match the equivalence class against | |
6461 | those known to the collate implementation. */ | |
6462 | const int32_t *table; | |
6463 | const int32_t *weights; | |
6464 | const int32_t *extra; | |
6465 | const int32_t *indirect; | |
6466 | int32_t idx, idx2; | |
6467 | wint_t *cp; | |
6468 | size_t len; | |
6469 | ||
6470 | /* This #include defines a local function! */ | |
6471 | # include <locale/weightwc.h> | |
6472 | ||
6473 | table = (const int32_t *) | |
6474 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC); | |
6475 | weights = (const wint_t *) | |
6476 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC); | |
6477 | extra = (const wint_t *) | |
6478 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC); | |
6479 | indirect = (const int32_t *) | |
6480 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC); | |
6481 | ||
6482 | /* Write 1 collating element to str_buf, and | |
6483 | get its index. */ | |
6484 | idx2 = 0; | |
6485 | ||
6486 | for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++) | |
6487 | { | |
6488 | cp = (wint_t*)str_buf; | |
6489 | if (d == dend) | |
6490 | { | |
6491 | if (dend == end_match_2) | |
6492 | break; | |
6493 | d = string2; | |
6494 | dend = end_match_2; | |
6495 | } | |
6496 | str_buf[i] = TRANSLATE(*(d+i)); | |
6497 | str_buf[i+1] = '\0'; /* sentinel */ | |
6498 | idx2 = findidx ((const wint_t**)&cp); | |
6499 | } | |
6500 | ||
6501 | /* Update d, however d will be incremented at | |
6502 | char_set_matched:, we decrement d here. */ | |
6503 | d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1); | |
6504 | if (d >= dend) | |
6505 | { | |
6506 | if (dend == end_match_2) | |
6507 | d = dend; | |
6508 | else | |
6509 | { | |
6510 | d = string2; | |
6511 | dend = end_match_2; | |
6512 | } | |
6513 | } | |
6514 | ||
6515 | len = weights[idx2]; | |
6516 | ||
6517 | for (workp2 = workp + equiv_class_length ; workp < workp2 ; | |
6518 | workp++) | |
6519 | { | |
6520 | idx = (int32_t)*workp; | |
6521 | /* We already checked idx != 0 in regex_compile. */ | |
6522 | ||
6523 | if (idx2 != 0 && len == weights[idx]) | |
6524 | { | |
6525 | int cnt = 0; | |
6526 | while (cnt < len && (weights[idx + 1 + cnt] | |
6527 | == weights[idx2 + 1 + cnt])) | |
6528 | ++cnt; | |
6529 | ||
6530 | if (cnt == len) | |
6531 | goto char_set_matched; | |
6532 | } | |
6533 | } | |
6534 | /* not matched */ | |
6535 | d = backup_d; | |
6536 | dend = backup_dend; | |
6537 | } | |
6538 | else /* (nrules == 0) */ | |
6539 | # endif | |
6540 | /* If we can't look up collation data, we use wcscoll | |
6541 | instead. */ | |
6542 | { | |
6543 | for (workp2 = workp + equiv_class_length ; workp < workp2 ;) | |
6544 | { | |
6545 | const CHAR_T *backup_d = d, *backup_dend = dend; | |
dc676635 DD |
6546 | # ifdef _LIBC |
6547 | length = __wcslen (workp); | |
6548 | # else | |
6549 | length = wcslen (workp); | |
6550 | # endif | |
2a6ef469 DD |
6551 | |
6552 | /* If wcscoll(the collating symbol, whole string) > 0, | |
6553 | any substring of the string never match with the | |
6554 | collating symbol. */ | |
dc676635 DD |
6555 | # ifdef _LIBC |
6556 | if (__wcscoll (workp, d) > 0) | |
6557 | # else | |
6558 | if (wcscoll (workp, d) > 0) | |
6559 | # endif | |
2a6ef469 DD |
6560 | { |
6561 | workp += length + 1; | |
6562 | break; | |
6563 | } | |
6564 | ||
6565 | /* First, we compare the equivalence class with | |
6566 | the first character of the string. | |
6567 | If it don't match, we add the next character to | |
6568 | the compare buffer in turn. */ | |
6569 | for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++) | |
6570 | { | |
6571 | int match; | |
6572 | if (d == dend) | |
6573 | { | |
6574 | if (dend == end_match_2) | |
6575 | break; | |
6576 | d = string2; | |
6577 | dend = end_match_2; | |
6578 | } | |
6579 | ||
6580 | /* add next character to the compare buffer. */ | |
6581 | str_buf[i] = TRANSLATE(*d); | |
6582 | str_buf[i+1] = '\0'; | |
6583 | ||
dc676635 DD |
6584 | # ifdef _LIBC |
6585 | match = __wcscoll (workp, str_buf); | |
6586 | # else | |
6587 | match = wcscoll (workp, str_buf); | |
6588 | # endif | |
2a6ef469 DD |
6589 | |
6590 | if (match == 0) | |
6591 | goto char_set_matched; | |
6592 | ||
6593 | if (match < 0) | |
6594 | /* (str_buf > workp) indicate (str_buf + X > workp), | |
6595 | because for all X (str_buf + X > str_buf). | |
6596 | So we don't need continue this loop. */ | |
6597 | break; | |
6598 | ||
6599 | /* Otherwise(str_buf < workp), | |
6600 | (str_buf+next_character) may equals (workp). | |
6601 | So we continue this loop. */ | |
6602 | } | |
6603 | /* not matched */ | |
6604 | d = backup_d; | |
6605 | dend = backup_dend; | |
6606 | workp += length + 1; | |
6607 | } | |
6608 | } | |
6609 | ||
6610 | /* match with char_range? */ | |
dc676635 | 6611 | # ifdef _LIBC |
2a6ef469 DD |
6612 | if (nrules != 0) |
6613 | { | |
6614 | uint32_t collseqval; | |
6615 | const char *collseq = (const char *) | |
6616 | _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC); | |
6617 | ||
6618 | collseqval = collseq_table_lookup (collseq, c); | |
6619 | ||
6620 | for (; workp < p - chars_length ;) | |
6621 | { | |
6622 | uint32_t start_val, end_val; | |
6623 | ||
6624 | /* We already compute the collation sequence value | |
6625 | of the characters (or collating symbols). */ | |
6626 | start_val = (uint32_t) *workp++; /* range_start */ | |
6627 | end_val = (uint32_t) *workp++; /* range_end */ | |
6628 | ||
6629 | if (start_val <= collseqval && collseqval <= end_val) | |
6630 | goto char_set_matched; | |
6631 | } | |
6632 | } | |
6633 | else | |
dc676635 | 6634 | # endif |
2a6ef469 DD |
6635 | { |
6636 | /* We set range_start_char at str_buf[0], range_end_char | |
6637 | at str_buf[4], and compared char at str_buf[2]. */ | |
6638 | str_buf[1] = 0; | |
6639 | str_buf[2] = c; | |
6640 | str_buf[3] = 0; | |
6641 | str_buf[5] = 0; | |
6642 | for (; workp < p - chars_length ;) | |
6643 | { | |
6644 | wchar_t *range_start_char, *range_end_char; | |
6645 | ||
6646 | /* match if (range_start_char <= c <= range_end_char). */ | |
6647 | ||
6648 | /* If range_start(or end) < 0, we assume -range_start(end) | |
6649 | is the offset of the collating symbol which is specified | |
6650 | as the character of the range start(end). */ | |
6651 | ||
6652 | /* range_start */ | |
6653 | if (*workp < 0) | |
6654 | range_start_char = charset_top - (*workp++); | |
6655 | else | |
6656 | { | |
6657 | str_buf[0] = *workp++; | |
6658 | range_start_char = str_buf; | |
6659 | } | |
6660 | ||
6661 | /* range_end */ | |
6662 | if (*workp < 0) | |
6663 | range_end_char = charset_top - (*workp++); | |
6664 | else | |
6665 | { | |
6666 | str_buf[4] = *workp++; | |
6667 | range_end_char = str_buf + 4; | |
6668 | } | |
6669 | ||
dc676635 DD |
6670 | # ifdef _LIBC |
6671 | if (__wcscoll (range_start_char, str_buf+2) <= 0 | |
6672 | && __wcscoll (str_buf+2, range_end_char) <= 0) | |
6673 | # else | |
6674 | if (wcscoll (range_start_char, str_buf+2) <= 0 | |
6675 | && wcscoll (str_buf+2, range_end_char) <= 0) | |
6676 | # endif | |
2a6ef469 DD |
6677 | goto char_set_matched; |
6678 | } | |
6679 | } | |
6680 | ||
6681 | /* match with char? */ | |
6682 | for (; workp < p ; workp++) | |
6683 | if (c == *workp) | |
6684 | goto char_set_matched; | |
6685 | ||
6686 | not = !not; | |
6687 | ||
6688 | char_set_matched: | |
6689 | if (not) goto fail; | |
6690 | #else | |
6691 | /* Cast to `unsigned' instead of `unsigned char' in case the | |
6692 | bit list is a full 32 bytes long. */ | |
6693 | if (c < (unsigned) (*p * BYTEWIDTH) | |
6694 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
6695 | not = !not; | |
6696 | ||
6697 | p += 1 + *p; | |
6698 | ||
6699 | if (!not) goto fail; | |
6700 | #undef WORK_BUFFER_SIZE | |
6701 | #endif /* WCHAR */ | |
6702 | SET_REGS_MATCHED (); | |
6703 | d++; | |
6704 | break; | |
6705 | } | |
6706 | ||
6707 | ||
6708 | /* The beginning of a group is represented by start_memory. | |
6709 | The arguments are the register number in the next byte, and the | |
6710 | number of groups inner to this one in the next. The text | |
6711 | matched within the group is recorded (in the internal | |
6712 | registers data structure) under the register number. */ | |
6713 | case start_memory: | |
6714 | DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n", | |
6715 | (long int) *p, (long int) p[1]); | |
6716 | ||
6717 | /* Find out if this group can match the empty string. */ | |
6718 | p1 = p; /* To send to group_match_null_string_p. */ | |
6719 | ||
6720 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
6721 | REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
6722 | = PREFIX(group_match_null_string_p) (&p1, pend, reg_info); | |
6723 | ||
6724 | /* Save the position in the string where we were the last time | |
6725 | we were at this open-group operator in case the group is | |
6726 | operated upon by a repetition operator, e.g., with `(a*)*b' | |
6727 | against `ab'; then we want to ignore where we are now in | |
6728 | the string in case this attempt to match fails. */ | |
6729 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
6730 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
6731 | : regstart[*p]; | |
6732 | DEBUG_PRINT2 (" old_regstart: %d\n", | |
6733 | POINTER_TO_OFFSET (old_regstart[*p])); | |
6734 | ||
6735 | regstart[*p] = d; | |
6736 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
6737 | ||
6738 | IS_ACTIVE (reg_info[*p]) = 1; | |
6739 | MATCHED_SOMETHING (reg_info[*p]) = 0; | |
6740 | ||
6741 | /* Clear this whenever we change the register activity status. */ | |
6742 | set_regs_matched_done = 0; | |
6743 | ||
6744 | /* This is the new highest active register. */ | |
6745 | highest_active_reg = *p; | |
6746 | ||
6747 | /* If nothing was active before, this is the new lowest active | |
6748 | register. */ | |
6749 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
6750 | lowest_active_reg = *p; | |
6751 | ||
6752 | /* Move past the register number and inner group count. */ | |
6753 | p += 2; | |
6754 | just_past_start_mem = p; | |
6755 | ||
6756 | break; | |
6757 | ||
6758 | ||
6759 | /* The stop_memory opcode represents the end of a group. Its | |
6760 | arguments are the same as start_memory's: the register | |
6761 | number, and the number of inner groups. */ | |
6762 | case stop_memory: | |
6763 | DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n", | |
6764 | (long int) *p, (long int) p[1]); | |
6765 | ||
6766 | /* We need to save the string position the last time we were at | |
6767 | this close-group operator in case the group is operated | |
6768 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
6769 | against `aba'; then we want to ignore where we are now in | |
6770 | the string in case this attempt to match fails. */ | |
6771 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
6772 | ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
6773 | : regend[*p]; | |
6774 | DEBUG_PRINT2 (" old_regend: %d\n", | |
6775 | POINTER_TO_OFFSET (old_regend[*p])); | |
6776 | ||
6777 | regend[*p] = d; | |
6778 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
6779 | ||
6780 | /* This register isn't active anymore. */ | |
6781 | IS_ACTIVE (reg_info[*p]) = 0; | |
6782 | ||
6783 | /* Clear this whenever we change the register activity status. */ | |
6784 | set_regs_matched_done = 0; | |
6785 | ||
6786 | /* If this was the only register active, nothing is active | |
6787 | anymore. */ | |
6788 | if (lowest_active_reg == highest_active_reg) | |
6789 | { | |
6790 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
6791 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
6792 | } | |
6793 | else | |
6794 | { /* We must scan for the new highest active register, since | |
6795 | it isn't necessarily one less than now: consider | |
6796 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
6797 | new highest active register is 1. */ | |
6798 | UCHAR_T r = *p - 1; | |
6799 | while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
6800 | r--; | |
6801 | ||
6802 | /* If we end up at register zero, that means that we saved | |
6803 | the registers as the result of an `on_failure_jump', not | |
6804 | a `start_memory', and we jumped to past the innermost | |
6805 | `stop_memory'. For example, in ((.)*) we save | |
6806 | registers 1 and 2 as a result of the *, but when we pop | |
6807 | back to the second ), we are at the stop_memory 1. | |
6808 | Thus, nothing is active. */ | |
6809 | if (r == 0) | |
6810 | { | |
6811 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
6812 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
6813 | } | |
6814 | else | |
6815 | highest_active_reg = r; | |
6816 | } | |
6817 | ||
6818 | /* If just failed to match something this time around with a | |
6819 | group that's operated on by a repetition operator, try to | |
6820 | force exit from the ``loop'', and restore the register | |
6821 | information for this group that we had before trying this | |
6822 | last match. */ | |
6823 | if ((!MATCHED_SOMETHING (reg_info[*p]) | |
6824 | || just_past_start_mem == p - 1) | |
6825 | && (p + 2) < pend) | |
6826 | { | |
6827 | boolean is_a_jump_n = false; | |
6828 | ||
6829 | p1 = p + 2; | |
6830 | mcnt = 0; | |
6831 | switch ((re_opcode_t) *p1++) | |
6832 | { | |
6833 | case jump_n: | |
6834 | is_a_jump_n = true; | |
6835 | case pop_failure_jump: | |
6836 | case maybe_pop_jump: | |
6837 | case jump: | |
6838 | case dummy_failure_jump: | |
6839 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
6840 | if (is_a_jump_n) | |
6841 | p1 += OFFSET_ADDRESS_SIZE; | |
6842 | break; | |
6843 | ||
6844 | default: | |
6845 | /* do nothing */ ; | |
6846 | } | |
6847 | p1 += mcnt; | |
6848 | ||
6849 | /* If the next operation is a jump backwards in the pattern | |
6850 | to an on_failure_jump right before the start_memory | |
6851 | corresponding to this stop_memory, exit from the loop | |
6852 | by forcing a failure after pushing on the stack the | |
6853 | on_failure_jump's jump in the pattern, and d. */ | |
6854 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
6855 | && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory | |
6856 | && p1[2+OFFSET_ADDRESS_SIZE] == *p) | |
6857 | { | |
6858 | /* If this group ever matched anything, then restore | |
6859 | what its registers were before trying this last | |
6860 | failed match, e.g., with `(a*)*b' against `ab' for | |
6861 | regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
6862 | against `aba' for regend[3]. | |
6863 | ||
6864 | Also restore the registers for inner groups for, | |
6865 | e.g., `((a*)(b*))*' against `aba' (register 3 would | |
6866 | otherwise get trashed). */ | |
6867 | ||
6868 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
6869 | { | |
6870 | unsigned r; | |
6871 | ||
6872 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
6873 | ||
6874 | /* Restore this and inner groups' (if any) registers. */ | |
6875 | for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1); | |
6876 | r++) | |
6877 | { | |
6878 | regstart[r] = old_regstart[r]; | |
6879 | ||
6880 | /* xx why this test? */ | |
6881 | if (old_regend[r] >= regstart[r]) | |
6882 | regend[r] = old_regend[r]; | |
6883 | } | |
6884 | } | |
6885 | p1++; | |
6886 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
6887 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
6888 | ||
6889 | goto fail; | |
6890 | } | |
6891 | } | |
6892 | ||
6893 | /* Move past the register number and the inner group count. */ | |
6894 | p += 2; | |
6895 | break; | |
6896 | ||
6897 | ||
6898 | /* \<digit> has been turned into a `duplicate' command which is | |
6899 | followed by the numeric value of <digit> as the register number. */ | |
6900 | case duplicate: | |
6901 | { | |
6902 | register const CHAR_T *d2, *dend2; | |
6903 | int regno = *p++; /* Get which register to match against. */ | |
6904 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
6905 | ||
6906 | /* Can't back reference a group which we've never matched. */ | |
6907 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
6908 | goto fail; | |
6909 | ||
6910 | /* Where in input to try to start matching. */ | |
6911 | d2 = regstart[regno]; | |
6912 | ||
6913 | /* Where to stop matching; if both the place to start and | |
6914 | the place to stop matching are in the same string, then | |
6915 | set to the place to stop, otherwise, for now have to use | |
6916 | the end of the first string. */ | |
6917 | ||
6918 | dend2 = ((FIRST_STRING_P (regstart[regno]) | |
6919 | == FIRST_STRING_P (regend[regno])) | |
6920 | ? regend[regno] : end_match_1); | |
6921 | for (;;) | |
6922 | { | |
6923 | /* If necessary, advance to next segment in register | |
6924 | contents. */ | |
6925 | while (d2 == dend2) | |
6926 | { | |
6927 | if (dend2 == end_match_2) break; | |
6928 | if (dend2 == regend[regno]) break; | |
6929 | ||
6930 | /* End of string1 => advance to string2. */ | |
6931 | d2 = string2; | |
6932 | dend2 = regend[regno]; | |
6933 | } | |
6934 | /* At end of register contents => success */ | |
6935 | if (d2 == dend2) break; | |
6936 | ||
6937 | /* If necessary, advance to next segment in data. */ | |
6938 | PREFETCH (); | |
6939 | ||
6940 | /* How many characters left in this segment to match. */ | |
6941 | mcnt = dend - d; | |
6942 | ||
6943 | /* Want how many consecutive characters we can match in | |
6944 | one shot, so, if necessary, adjust the count. */ | |
6945 | if (mcnt > dend2 - d2) | |
6946 | mcnt = dend2 - d2; | |
6947 | ||
6948 | /* Compare that many; failure if mismatch, else move | |
6949 | past them. */ | |
6950 | if (translate | |
6951 | ? PREFIX(bcmp_translate) (d, d2, mcnt, translate) | |
6952 | : memcmp (d, d2, mcnt*sizeof(UCHAR_T))) | |
6953 | goto fail; | |
6954 | d += mcnt, d2 += mcnt; | |
6955 | ||
6956 | /* Do this because we've match some characters. */ | |
6957 | SET_REGS_MATCHED (); | |
6958 | } | |
6959 | } | |
6960 | break; | |
6961 | ||
6962 | ||
6963 | /* begline matches the empty string at the beginning of the string | |
6964 | (unless `not_bol' is set in `bufp'), and, if | |
6965 | `newline_anchor' is set, after newlines. */ | |
6966 | case begline: | |
6967 | DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
6968 | ||
6969 | if (AT_STRINGS_BEG (d)) | |
6970 | { | |
6971 | if (!bufp->not_bol) break; | |
6972 | } | |
6973 | else if (d[-1] == '\n' && bufp->newline_anchor) | |
6974 | { | |
6975 | break; | |
6976 | } | |
6977 | /* In all other cases, we fail. */ | |
6978 | goto fail; | |
6979 | ||
6980 | ||
6981 | /* endline is the dual of begline. */ | |
6982 | case endline: | |
6983 | DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
6984 | ||
6985 | if (AT_STRINGS_END (d)) | |
6986 | { | |
6987 | if (!bufp->not_eol) break; | |
6988 | } | |
6989 | ||
6990 | /* We have to ``prefetch'' the next character. */ | |
6991 | else if ((d == end1 ? *string2 : *d) == '\n' | |
6992 | && bufp->newline_anchor) | |
6993 | { | |
6994 | break; | |
6995 | } | |
6996 | goto fail; | |
6997 | ||
6998 | ||
6999 | /* Match at the very beginning of the data. */ | |
7000 | case begbuf: | |
7001 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
7002 | if (AT_STRINGS_BEG (d)) | |
7003 | break; | |
7004 | goto fail; | |
7005 | ||
7006 | ||
7007 | /* Match at the very end of the data. */ | |
7008 | case endbuf: | |
7009 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
7010 | if (AT_STRINGS_END (d)) | |
7011 | break; | |
7012 | goto fail; | |
7013 | ||
7014 | ||
7015 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
7016 | pushes NULL as the value for the string on the stack. Then | |
7017 | `pop_failure_point' will keep the current value for the | |
7018 | string, instead of restoring it. To see why, consider | |
7019 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
7020 | then the . fails against the \n. But the next thing we want | |
7021 | to do is match the \n against the \n; if we restored the | |
7022 | string value, we would be back at the foo. | |
7023 | ||
7024 | Because this is used only in specific cases, we don't need to | |
7025 | check all the things that `on_failure_jump' does, to make | |
7026 | sure the right things get saved on the stack. Hence we don't | |
7027 | share its code. The only reason to push anything on the | |
7028 | stack at all is that otherwise we would have to change | |
7029 | `anychar's code to do something besides goto fail in this | |
7030 | case; that seems worse than this. */ | |
7031 | case on_failure_keep_string_jump: | |
7032 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
7033 | ||
7034 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
7035 | #ifdef _LIBC | |
7036 | DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt); | |
7037 | #else | |
7038 | DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
7039 | #endif | |
7040 | ||
7041 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
7042 | break; | |
7043 | ||
7044 | ||
7045 | /* Uses of on_failure_jump: | |
7046 | ||
7047 | Each alternative starts with an on_failure_jump that points | |
7048 | to the beginning of the next alternative. Each alternative | |
7049 | except the last ends with a jump that in effect jumps past | |
7050 | the rest of the alternatives. (They really jump to the | |
7051 | ending jump of the following alternative, because tensioning | |
7052 | these jumps is a hassle.) | |
7053 | ||
7054 | Repeats start with an on_failure_jump that points past both | |
7055 | the repetition text and either the following jump or | |
7056 | pop_failure_jump back to this on_failure_jump. */ | |
7057 | case on_failure_jump: | |
7058 | on_failure: | |
7059 | DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
7060 | ||
7061 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
7062 | #ifdef _LIBC | |
7063 | DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt); | |
7064 | #else | |
7065 | DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
7066 | #endif | |
7067 | ||
7068 | /* If this on_failure_jump comes right before a group (i.e., | |
7069 | the original * applied to a group), save the information | |
7070 | for that group and all inner ones, so that if we fail back | |
7071 | to this point, the group's information will be correct. | |
7072 | For example, in \(a*\)*\1, we need the preceding group, | |
7073 | and in \(zz\(a*\)b*\)\2, we need the inner group. */ | |
7074 | ||
7075 | /* We can't use `p' to check ahead because we push | |
7076 | a failure point to `p + mcnt' after we do this. */ | |
7077 | p1 = p; | |
7078 | ||
7079 | /* We need to skip no_op's before we look for the | |
7080 | start_memory in case this on_failure_jump is happening as | |
7081 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
7082 | against aba. */ | |
7083 | while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
7084 | p1++; | |
7085 | ||
7086 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
7087 | { | |
7088 | /* We have a new highest active register now. This will | |
7089 | get reset at the start_memory we are about to get to, | |
7090 | but we will have saved all the registers relevant to | |
7091 | this repetition op, as described above. */ | |
7092 | highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
7093 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
7094 | lowest_active_reg = *(p1 + 1); | |
7095 | } | |
7096 | ||
7097 | DEBUG_PRINT1 (":\n"); | |
7098 | PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
7099 | break; | |
7100 | ||
7101 | ||
7102 | /* A smart repeat ends with `maybe_pop_jump'. | |
7103 | We change it to either `pop_failure_jump' or `jump'. */ | |
7104 | case maybe_pop_jump: | |
7105 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
7106 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
7107 | { | |
7108 | register UCHAR_T *p2 = p; | |
7109 | ||
7110 | /* Compare the beginning of the repeat with what in the | |
7111 | pattern follows its end. If we can establish that there | |
7112 | is nothing that they would both match, i.e., that we | |
7113 | would have to backtrack because of (as in, e.g., `a*a') | |
7114 | then we can change to pop_failure_jump, because we'll | |
7115 | never have to backtrack. | |
7116 | ||
7117 | This is not true in the case of alternatives: in | |
7118 | `(a|ab)*' we do need to backtrack to the `ab' alternative | |
7119 | (e.g., if the string was `ab'). But instead of trying to | |
7120 | detect that here, the alternative has put on a dummy | |
7121 | failure point which is what we will end up popping. */ | |
7122 | ||
7123 | /* Skip over open/close-group commands. | |
7124 | If what follows this loop is a ...+ construct, | |
7125 | look at what begins its body, since we will have to | |
7126 | match at least one of that. */ | |
7127 | while (1) | |
7128 | { | |
7129 | if (p2 + 2 < pend | |
7130 | && ((re_opcode_t) *p2 == stop_memory | |
7131 | || (re_opcode_t) *p2 == start_memory)) | |
7132 | p2 += 3; | |
7133 | else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend | |
7134 | && (re_opcode_t) *p2 == dummy_failure_jump) | |
7135 | p2 += 2 + 2 * OFFSET_ADDRESS_SIZE; | |
7136 | else | |
7137 | break; | |
7138 | } | |
7139 | ||
7140 | p1 = p + mcnt; | |
7141 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
7142 | to the `maybe_finalize_jump' of this case. Examine what | |
7143 | follows. */ | |
7144 | ||
7145 | /* If we're at the end of the pattern, we can change. */ | |
7146 | if (p2 == pend) | |
7147 | { | |
7148 | /* Consider what happens when matching ":\(.*\)" | |
7149 | against ":/". I don't really understand this code | |
7150 | yet. */ | |
7151 | p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) | |
7152 | pop_failure_jump; | |
7153 | DEBUG_PRINT1 | |
7154 | (" End of pattern: change to `pop_failure_jump'.\n"); | |
7155 | } | |
7156 | ||
7157 | else if ((re_opcode_t) *p2 == exactn | |
7158 | #ifdef MBS_SUPPORT | |
7159 | || (re_opcode_t) *p2 == exactn_bin | |
7160 | #endif | |
7161 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
7162 | { | |
7163 | register UCHAR_T c | |
7164 | = *p2 == (UCHAR_T) endline ? '\n' : p2[2]; | |
7165 | ||
7166 | if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn | |
7167 | #ifdef MBS_SUPPORT | |
7168 | || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin | |
7169 | #endif | |
7170 | ) && p1[3+OFFSET_ADDRESS_SIZE] != c) | |
7171 | { | |
7172 | p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) | |
7173 | pop_failure_jump; | |
7174 | #ifdef WCHAR | |
7175 | DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n", | |
7176 | (wint_t) c, | |
7177 | (wint_t) p1[3+OFFSET_ADDRESS_SIZE]); | |
7178 | #else | |
7179 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
7180 | (char) c, | |
7181 | (char) p1[3+OFFSET_ADDRESS_SIZE]); | |
7182 | #endif | |
7183 | } | |
7184 | ||
7185 | #ifndef WCHAR | |
7186 | else if ((re_opcode_t) p1[3] == charset | |
7187 | || (re_opcode_t) p1[3] == charset_not) | |
7188 | { | |
7189 | int not = (re_opcode_t) p1[3] == charset_not; | |
7190 | ||
7191 | if (c < (unsigned) (p1[4] * BYTEWIDTH) | |
7192 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
7193 | not = !not; | |
7194 | ||
7195 | /* `not' is equal to 1 if c would match, which means | |
7196 | that we can't change to pop_failure_jump. */ | |
7197 | if (!not) | |
7198 | { | |
7199 | p[-3] = (unsigned char) pop_failure_jump; | |
7200 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
7201 | } | |
7202 | } | |
7203 | #endif /* not WCHAR */ | |
7204 | } | |
7205 | #ifndef WCHAR | |
7206 | else if ((re_opcode_t) *p2 == charset) | |
7207 | { | |
7208 | /* We win if the first character of the loop is not part | |
7209 | of the charset. */ | |
7210 | if ((re_opcode_t) p1[3] == exactn | |
7211 | && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] | |
7212 | && (p2[2 + p1[5] / BYTEWIDTH] | |
7213 | & (1 << (p1[5] % BYTEWIDTH))))) | |
7214 | { | |
7215 | p[-3] = (unsigned char) pop_failure_jump; | |
7216 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
7217 | } | |
7218 | ||
7219 | else if ((re_opcode_t) p1[3] == charset_not) | |
7220 | { | |
7221 | int idx; | |
7222 | /* We win if the charset_not inside the loop | |
7223 | lists every character listed in the charset after. */ | |
7224 | for (idx = 0; idx < (int) p2[1]; idx++) | |
7225 | if (! (p2[2 + idx] == 0 | |
7226 | || (idx < (int) p1[4] | |
7227 | && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) | |
7228 | break; | |
7229 | ||
7230 | if (idx == p2[1]) | |
7231 | { | |
7232 | p[-3] = (unsigned char) pop_failure_jump; | |
7233 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
7234 | } | |
7235 | } | |
7236 | else if ((re_opcode_t) p1[3] == charset) | |
7237 | { | |
7238 | int idx; | |
7239 | /* We win if the charset inside the loop | |
7240 | has no overlap with the one after the loop. */ | |
7241 | for (idx = 0; | |
7242 | idx < (int) p2[1] && idx < (int) p1[4]; | |
7243 | idx++) | |
7244 | if ((p2[2 + idx] & p1[5 + idx]) != 0) | |
7245 | break; | |
7246 | ||
7247 | if (idx == p2[1] || idx == p1[4]) | |
7248 | { | |
7249 | p[-3] = (unsigned char) pop_failure_jump; | |
7250 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
7251 | } | |
7252 | } | |
7253 | } | |
7254 | #endif /* not WCHAR */ | |
7255 | } | |
7256 | p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */ | |
7257 | if ((re_opcode_t) p[-1] != pop_failure_jump) | |
7258 | { | |
7259 | p[-1] = (UCHAR_T) jump; | |
7260 | DEBUG_PRINT1 (" Match => jump.\n"); | |
7261 | goto unconditional_jump; | |
7262 | } | |
7263 | /* Note fall through. */ | |
7264 | ||
7265 | ||
7266 | /* The end of a simple repeat has a pop_failure_jump back to | |
7267 | its matching on_failure_jump, where the latter will push a | |
7268 | failure point. The pop_failure_jump takes off failure | |
7269 | points put on by this pop_failure_jump's matching | |
7270 | on_failure_jump; we got through the pattern to here from the | |
7271 | matching on_failure_jump, so didn't fail. */ | |
7272 | case pop_failure_jump: | |
7273 | { | |
7274 | /* We need to pass separate storage for the lowest and | |
7275 | highest registers, even though we don't care about the | |
7276 | actual values. Otherwise, we will restore only one | |
7277 | register from the stack, since lowest will == highest in | |
7278 | `pop_failure_point'. */ | |
7279 | active_reg_t dummy_low_reg, dummy_high_reg; | |
7280 | UCHAR_T *pdummy = NULL; | |
7281 | const CHAR_T *sdummy = NULL; | |
7282 | ||
7283 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
7284 | POP_FAILURE_POINT (sdummy, pdummy, | |
7285 | dummy_low_reg, dummy_high_reg, | |
7286 | reg_dummy, reg_dummy, reg_info_dummy); | |
7287 | } | |
7288 | /* Note fall through. */ | |
7289 | ||
7290 | unconditional_jump: | |
7291 | #ifdef _LIBC | |
7292 | DEBUG_PRINT2 ("\n%p: ", p); | |
7293 | #else | |
7294 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
7295 | #endif | |
7296 | /* Note fall through. */ | |
7297 | ||
7298 | /* Unconditionally jump (without popping any failure points). */ | |
7299 | case jump: | |
7300 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
7301 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
7302 | p += mcnt; /* Do the jump. */ | |
7303 | #ifdef _LIBC | |
7304 | DEBUG_PRINT2 ("(to %p).\n", p); | |
7305 | #else | |
7306 | DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
7307 | #endif | |
7308 | break; | |
7309 | ||
7310 | ||
7311 | /* We need this opcode so we can detect where alternatives end | |
7312 | in `group_match_null_string_p' et al. */ | |
7313 | case jump_past_alt: | |
7314 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
7315 | goto unconditional_jump; | |
7316 | ||
7317 | ||
7318 | /* Normally, the on_failure_jump pushes a failure point, which | |
7319 | then gets popped at pop_failure_jump. We will end up at | |
7320 | pop_failure_jump, also, and with a pattern of, say, `a+', we | |
7321 | are skipping over the on_failure_jump, so we have to push | |
7322 | something meaningless for pop_failure_jump to pop. */ | |
7323 | case dummy_failure_jump: | |
7324 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
7325 | /* It doesn't matter what we push for the string here. What | |
7326 | the code at `fail' tests is the value for the pattern. */ | |
7327 | PUSH_FAILURE_POINT (NULL, NULL, -2); | |
7328 | goto unconditional_jump; | |
7329 | ||
7330 | ||
7331 | /* At the end of an alternative, we need to push a dummy failure | |
7332 | point in case we are followed by a `pop_failure_jump', because | |
7333 | we don't want the failure point for the alternative to be | |
7334 | popped. For example, matching `(a|ab)*' against `aab' | |
7335 | requires that we match the `ab' alternative. */ | |
7336 | case push_dummy_failure: | |
7337 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
7338 | /* See comments just above at `dummy_failure_jump' about the | |
7339 | two zeroes. */ | |
7340 | PUSH_FAILURE_POINT (NULL, NULL, -2); | |
7341 | break; | |
7342 | ||
7343 | /* Have to succeed matching what follows at least n times. | |
7344 | After that, handle like `on_failure_jump'. */ | |
7345 | case succeed_n: | |
7346 | EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); | |
7347 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
7348 | ||
7349 | assert (mcnt >= 0); | |
7350 | /* Originally, this is how many times we HAVE to succeed. */ | |
7351 | if (mcnt > 0) | |
7352 | { | |
7353 | mcnt--; | |
7354 | p += OFFSET_ADDRESS_SIZE; | |
7355 | STORE_NUMBER_AND_INCR (p, mcnt); | |
7356 | #ifdef _LIBC | |
7357 | DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE | |
7358 | , mcnt); | |
7359 | #else | |
7360 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE | |
7361 | , mcnt); | |
7362 | #endif | |
7363 | } | |
7364 | else if (mcnt == 0) | |
7365 | { | |
7366 | #ifdef _LIBC | |
7367 | DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", | |
7368 | p + OFFSET_ADDRESS_SIZE); | |
7369 | #else | |
7370 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", | |
7371 | p + OFFSET_ADDRESS_SIZE); | |
7372 | #endif /* _LIBC */ | |
7373 | ||
7374 | #ifdef WCHAR | |
7375 | p[1] = (UCHAR_T) no_op; | |
7376 | #else | |
7377 | p[2] = (UCHAR_T) no_op; | |
7378 | p[3] = (UCHAR_T) no_op; | |
7379 | #endif /* WCHAR */ | |
7380 | goto on_failure; | |
7381 | } | |
7382 | break; | |
7383 | ||
7384 | case jump_n: | |
7385 | EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); | |
7386 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
7387 | ||
7388 | /* Originally, this is how many times we CAN jump. */ | |
7389 | if (mcnt) | |
7390 | { | |
7391 | mcnt--; | |
7392 | STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt); | |
7393 | ||
7394 | #ifdef _LIBC | |
7395 | DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE, | |
7396 | mcnt); | |
7397 | #else | |
7398 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE, | |
7399 | mcnt); | |
7400 | #endif /* _LIBC */ | |
7401 | goto unconditional_jump; | |
7402 | } | |
7403 | /* If don't have to jump any more, skip over the rest of command. */ | |
7404 | else | |
7405 | p += 2 * OFFSET_ADDRESS_SIZE; | |
7406 | break; | |
7407 | ||
7408 | case set_number_at: | |
7409 | { | |
7410 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
7411 | ||
7412 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
7413 | p1 = p + mcnt; | |
7414 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
7415 | #ifdef _LIBC | |
7416 | DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt); | |
7417 | #else | |
7418 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
7419 | #endif | |
7420 | STORE_NUMBER (p1, mcnt); | |
7421 | break; | |
7422 | } | |
7423 | ||
7424 | #if 0 | |
7425 | /* The DEC Alpha C compiler 3.x generates incorrect code for the | |
7426 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of | |
7427 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the | |
7428 | macro and introducing temporary variables works around the bug. */ | |
7429 | ||
7430 | case wordbound: | |
7431 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
7432 | if (AT_WORD_BOUNDARY (d)) | |
7433 | break; | |
7434 | goto fail; | |
7435 | ||
7436 | case notwordbound: | |
7437 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
7438 | if (AT_WORD_BOUNDARY (d)) | |
7439 | goto fail; | |
7440 | break; | |
7441 | #else | |
7442 | case wordbound: | |
7443 | { | |
7444 | boolean prevchar, thischar; | |
7445 | ||
7446 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
7447 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) | |
7448 | break; | |
7449 | ||
7450 | prevchar = WORDCHAR_P (d - 1); | |
7451 | thischar = WORDCHAR_P (d); | |
7452 | if (prevchar != thischar) | |
7453 | break; | |
7454 | goto fail; | |
7455 | } | |
7456 | ||
7457 | case notwordbound: | |
7458 | { | |
7459 | boolean prevchar, thischar; | |
7460 | ||
7461 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
7462 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) | |
7463 | goto fail; | |
7464 | ||
7465 | prevchar = WORDCHAR_P (d - 1); | |
7466 | thischar = WORDCHAR_P (d); | |
7467 | if (prevchar != thischar) | |
7468 | goto fail; | |
7469 | break; | |
7470 | } | |
7471 | #endif | |
7472 | ||
7473 | case wordbeg: | |
7474 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
7475 | if (!AT_STRINGS_END (d) && WORDCHAR_P (d) | |
7476 | && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) | |
7477 | break; | |
7478 | goto fail; | |
7479 | ||
7480 | case wordend: | |
7481 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
7482 | if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) | |
7483 | && (AT_STRINGS_END (d) || !WORDCHAR_P (d))) | |
7484 | break; | |
7485 | goto fail; | |
7486 | ||
7487 | #ifdef emacs | |
7488 | case before_dot: | |
7489 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
7490 | if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
7491 | goto fail; | |
7492 | break; | |
7493 | ||
7494 | case at_dot: | |
7495 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
7496 | if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
7497 | goto fail; | |
7498 | break; | |
7499 | ||
7500 | case after_dot: | |
7501 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
7502 | if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
7503 | goto fail; | |
7504 | break; | |
7505 | ||
7506 | case syntaxspec: | |
7507 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
7508 | mcnt = *p++; | |
7509 | goto matchsyntax; | |
7510 | ||
7511 | case wordchar: | |
7512 | DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); | |
7513 | mcnt = (int) Sword; | |
7514 | matchsyntax: | |
7515 | PREFETCH (); | |
7516 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ | |
7517 | d++; | |
7518 | if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) | |
7519 | goto fail; | |
7520 | SET_REGS_MATCHED (); | |
7521 | break; | |
7522 | ||
7523 | case notsyntaxspec: | |
7524 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
7525 | mcnt = *p++; | |
7526 | goto matchnotsyntax; | |
7527 | ||
7528 | case notwordchar: | |
7529 | DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); | |
7530 | mcnt = (int) Sword; | |
7531 | matchnotsyntax: | |
7532 | PREFETCH (); | |
7533 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ | |
7534 | d++; | |
7535 | if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) | |
7536 | goto fail; | |
7537 | SET_REGS_MATCHED (); | |
7538 | break; | |
7539 | ||
7540 | #else /* not emacs */ | |
7541 | case wordchar: | |
7542 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
7543 | PREFETCH (); | |
7544 | if (!WORDCHAR_P (d)) | |
7545 | goto fail; | |
7546 | SET_REGS_MATCHED (); | |
7547 | d++; | |
7548 | break; | |
7549 | ||
7550 | case notwordchar: | |
7551 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
7552 | PREFETCH (); | |
7553 | if (WORDCHAR_P (d)) | |
7554 | goto fail; | |
7555 | SET_REGS_MATCHED (); | |
7556 | d++; | |
7557 | break; | |
7558 | #endif /* not emacs */ | |
7559 | ||
7560 | default: | |
7561 | abort (); | |
7562 | } | |
7563 | continue; /* Successfully executed one pattern command; keep going. */ | |
7564 | ||
7565 | ||
7566 | /* We goto here if a matching operation fails. */ | |
7567 | fail: | |
7568 | if (!FAIL_STACK_EMPTY ()) | |
7569 | { /* A restart point is known. Restore to that state. */ | |
7570 | DEBUG_PRINT1 ("\nFAIL:\n"); | |
7571 | POP_FAILURE_POINT (d, p, | |
7572 | lowest_active_reg, highest_active_reg, | |
7573 | regstart, regend, reg_info); | |
7574 | ||
7575 | /* If this failure point is a dummy, try the next one. */ | |
7576 | if (!p) | |
7577 | goto fail; | |
7578 | ||
7579 | /* If we failed to the end of the pattern, don't examine *p. */ | |
7580 | assert (p <= pend); | |
7581 | if (p < pend) | |
7582 | { | |
7583 | boolean is_a_jump_n = false; | |
7584 | ||
7585 | /* If failed to a backwards jump that's part of a repetition | |
7586 | loop, need to pop this failure point and use the next one. */ | |
7587 | switch ((re_opcode_t) *p) | |
7588 | { | |
7589 | case jump_n: | |
7590 | is_a_jump_n = true; | |
7591 | case maybe_pop_jump: | |
7592 | case pop_failure_jump: | |
7593 | case jump: | |
7594 | p1 = p + 1; | |
7595 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7596 | p1 += mcnt; | |
7597 | ||
7598 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
7599 | || (!is_a_jump_n | |
7600 | && (re_opcode_t) *p1 == on_failure_jump)) | |
7601 | goto fail; | |
7602 | break; | |
7603 | default: | |
7604 | /* do nothing */ ; | |
7605 | } | |
7606 | } | |
7607 | ||
7608 | if (d >= string1 && d <= end1) | |
7609 | dend = end_match_1; | |
7610 | } | |
7611 | else | |
7612 | break; /* Matching at this starting point really fails. */ | |
7613 | } /* for (;;) */ | |
7614 | ||
7615 | if (best_regs_set) | |
7616 | goto restore_best_regs; | |
7617 | ||
7618 | FREE_VARIABLES (); | |
7619 | ||
7620 | return -1; /* Failure to match. */ | |
7621 | } /* re_match_2 */ | |
7622 | \f | |
7623 | /* Subroutine definitions for re_match_2. */ | |
7624 | ||
7625 | ||
7626 | /* We are passed P pointing to a register number after a start_memory. | |
7627 | ||
7628 | Return true if the pattern up to the corresponding stop_memory can | |
7629 | match the empty string, and false otherwise. | |
7630 | ||
7631 | If we find the matching stop_memory, sets P to point to one past its number. | |
7632 | Otherwise, sets P to an undefined byte less than or equal to END. | |
7633 | ||
7634 | We don't handle duplicates properly (yet). */ | |
7635 | ||
7636 | static boolean | |
7637 | PREFIX(group_match_null_string_p) (p, end, reg_info) | |
7638 | UCHAR_T **p, *end; | |
7639 | PREFIX(register_info_type) *reg_info; | |
7640 | { | |
7641 | int mcnt; | |
7642 | /* Point to after the args to the start_memory. */ | |
7643 | UCHAR_T *p1 = *p + 2; | |
7644 | ||
7645 | while (p1 < end) | |
7646 | { | |
7647 | /* Skip over opcodes that can match nothing, and return true or | |
7648 | false, as appropriate, when we get to one that can't, or to the | |
7649 | matching stop_memory. */ | |
7650 | ||
7651 | switch ((re_opcode_t) *p1) | |
7652 | { | |
7653 | /* Could be either a loop or a series of alternatives. */ | |
7654 | case on_failure_jump: | |
7655 | p1++; | |
7656 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7657 | ||
7658 | /* If the next operation is not a jump backwards in the | |
7659 | pattern. */ | |
7660 | ||
7661 | if (mcnt >= 0) | |
7662 | { | |
7663 | /* Go through the on_failure_jumps of the alternatives, | |
7664 | seeing if any of the alternatives cannot match nothing. | |
7665 | The last alternative starts with only a jump, | |
7666 | whereas the rest start with on_failure_jump and end | |
7667 | with a jump, e.g., here is the pattern for `a|b|c': | |
7668 | ||
7669 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
7670 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
7671 | /exactn/1/c | |
7672 | ||
7673 | So, we have to first go through the first (n-1) | |
7674 | alternatives and then deal with the last one separately. */ | |
7675 | ||
7676 | ||
7677 | /* Deal with the first (n-1) alternatives, which start | |
7678 | with an on_failure_jump (see above) that jumps to right | |
7679 | past a jump_past_alt. */ | |
7680 | ||
7681 | while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] == | |
7682 | jump_past_alt) | |
7683 | { | |
7684 | /* `mcnt' holds how many bytes long the alternative | |
7685 | is, including the ending `jump_past_alt' and | |
7686 | its number. */ | |
7687 | ||
7688 | if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt - | |
7689 | (1 + OFFSET_ADDRESS_SIZE), | |
7690 | reg_info)) | |
7691 | return false; | |
7692 | ||
7693 | /* Move to right after this alternative, including the | |
7694 | jump_past_alt. */ | |
7695 | p1 += mcnt; | |
7696 | ||
7697 | /* Break if it's the beginning of an n-th alternative | |
7698 | that doesn't begin with an on_failure_jump. */ | |
7699 | if ((re_opcode_t) *p1 != on_failure_jump) | |
7700 | break; | |
7701 | ||
7702 | /* Still have to check that it's not an n-th | |
7703 | alternative that starts with an on_failure_jump. */ | |
7704 | p1++; | |
7705 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7706 | if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] != | |
7707 | jump_past_alt) | |
7708 | { | |
7709 | /* Get to the beginning of the n-th alternative. */ | |
7710 | p1 -= 1 + OFFSET_ADDRESS_SIZE; | |
7711 | break; | |
7712 | } | |
7713 | } | |
7714 | ||
7715 | /* Deal with the last alternative: go back and get number | |
7716 | of the `jump_past_alt' just before it. `mcnt' contains | |
7717 | the length of the alternative. */ | |
7718 | EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE); | |
7719 | ||
7720 | if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info)) | |
7721 | return false; | |
7722 | ||
7723 | p1 += mcnt; /* Get past the n-th alternative. */ | |
7724 | } /* if mcnt > 0 */ | |
7725 | break; | |
7726 | ||
7727 | ||
7728 | case stop_memory: | |
7729 | assert (p1[1] == **p); | |
7730 | *p = p1 + 2; | |
7731 | return true; | |
7732 | ||
7733 | ||
7734 | default: | |
7735 | if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) | |
7736 | return false; | |
7737 | } | |
7738 | } /* while p1 < end */ | |
7739 | ||
7740 | return false; | |
7741 | } /* group_match_null_string_p */ | |
7742 | ||
7743 | ||
7744 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
7745 | It expects P to be the first byte of a single alternative and END one | |
7746 | byte past the last. The alternative can contain groups. */ | |
7747 | ||
7748 | static boolean | |
7749 | PREFIX(alt_match_null_string_p) (p, end, reg_info) | |
7750 | UCHAR_T *p, *end; | |
7751 | PREFIX(register_info_type) *reg_info; | |
7752 | { | |
7753 | int mcnt; | |
7754 | UCHAR_T *p1 = p; | |
7755 | ||
7756 | while (p1 < end) | |
7757 | { | |
7758 | /* Skip over opcodes that can match nothing, and break when we get | |
7759 | to one that can't. */ | |
7760 | ||
7761 | switch ((re_opcode_t) *p1) | |
7762 | { | |
7763 | /* It's a loop. */ | |
7764 | case on_failure_jump: | |
7765 | p1++; | |
7766 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7767 | p1 += mcnt; | |
7768 | break; | |
7769 | ||
7770 | default: | |
7771 | if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) | |
7772 | return false; | |
7773 | } | |
7774 | } /* while p1 < end */ | |
7775 | ||
7776 | return true; | |
7777 | } /* alt_match_null_string_p */ | |
7778 | ||
7779 | ||
7780 | /* Deals with the ops common to group_match_null_string_p and | |
7781 | alt_match_null_string_p. | |
7782 | ||
7783 | Sets P to one after the op and its arguments, if any. */ | |
7784 | ||
7785 | static boolean | |
7786 | PREFIX(common_op_match_null_string_p) (p, end, reg_info) | |
7787 | UCHAR_T **p, *end; | |
7788 | PREFIX(register_info_type) *reg_info; | |
7789 | { | |
7790 | int mcnt; | |
7791 | boolean ret; | |
7792 | int reg_no; | |
7793 | UCHAR_T *p1 = *p; | |
7794 | ||
7795 | switch ((re_opcode_t) *p1++) | |
7796 | { | |
7797 | case no_op: | |
7798 | case begline: | |
7799 | case endline: | |
7800 | case begbuf: | |
7801 | case endbuf: | |
7802 | case wordbeg: | |
7803 | case wordend: | |
7804 | case wordbound: | |
7805 | case notwordbound: | |
7806 | #ifdef emacs | |
7807 | case before_dot: | |
7808 | case at_dot: | |
7809 | case after_dot: | |
7810 | #endif | |
7811 | break; | |
7812 | ||
7813 | case start_memory: | |
7814 | reg_no = *p1; | |
7815 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
7816 | ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info); | |
7817 | ||
7818 | /* Have to set this here in case we're checking a group which | |
7819 | contains a group and a back reference to it. */ | |
7820 | ||
7821 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
7822 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
7823 | ||
7824 | if (!ret) | |
7825 | return false; | |
7826 | break; | |
7827 | ||
7828 | /* If this is an optimized succeed_n for zero times, make the jump. */ | |
7829 | case jump: | |
7830 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7831 | if (mcnt >= 0) | |
7832 | p1 += mcnt; | |
7833 | else | |
7834 | return false; | |
7835 | break; | |
7836 | ||
7837 | case succeed_n: | |
7838 | /* Get to the number of times to succeed. */ | |
7839 | p1 += OFFSET_ADDRESS_SIZE; | |
7840 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7841 | ||
7842 | if (mcnt == 0) | |
7843 | { | |
7844 | p1 -= 2 * OFFSET_ADDRESS_SIZE; | |
7845 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7846 | p1 += mcnt; | |
7847 | } | |
7848 | else | |
7849 | return false; | |
7850 | break; | |
7851 | ||
7852 | case duplicate: | |
7853 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
7854 | return false; | |
7855 | break; | |
7856 | ||
7857 | case set_number_at: | |
7858 | p1 += 2 * OFFSET_ADDRESS_SIZE; | |
7859 | ||
7860 | default: | |
7861 | /* All other opcodes mean we cannot match the empty string. */ | |
7862 | return false; | |
7863 | } | |
7864 | ||
7865 | *p = p1; | |
7866 | return true; | |
7867 | } /* common_op_match_null_string_p */ | |
7868 | ||
7869 | ||
7870 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
7871 | bytes; nonzero otherwise. */ | |
7872 | ||
7873 | static int | |
7874 | PREFIX(bcmp_translate) (s1, s2, len, translate) | |
7875 | const CHAR_T *s1, *s2; | |
7876 | register int len; | |
7877 | RE_TRANSLATE_TYPE translate; | |
7878 | { | |
7879 | register const UCHAR_T *p1 = (const UCHAR_T *) s1; | |
7880 | register const UCHAR_T *p2 = (const UCHAR_T *) s2; | |
7881 | while (len) | |
7882 | { | |
7883 | #ifdef WCHAR | |
7884 | if (((*p1<=0xff)?translate[*p1++]:*p1++) | |
7885 | != ((*p2<=0xff)?translate[*p2++]:*p2++)) | |
7886 | return 1; | |
7887 | #else /* BYTE */ | |
7888 | if (translate[*p1++] != translate[*p2++]) return 1; | |
7889 | #endif /* WCHAR */ | |
7890 | len--; | |
7891 | } | |
7892 | return 0; | |
7893 | } | |
7894 | \f | |
7895 | ||
7896 | #else /* not INSIDE_RECURSION */ | |
7897 | ||
7898 | /* Entry points for GNU code. */ | |
7899 | ||
7900 | /* re_compile_pattern is the GNU regular expression compiler: it | |
7901 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
7902 | Returns 0 if the pattern was valid, otherwise an error string. | |
7903 | ||
7904 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
7905 | are set in BUFP on entry. | |
7906 | ||
7907 | We call regex_compile to do the actual compilation. */ | |
7908 | ||
7909 | const char * | |
7910 | re_compile_pattern (pattern, length, bufp) | |
7911 | const char *pattern; | |
7912 | size_t length; | |
7913 | struct re_pattern_buffer *bufp; | |
7914 | { | |
7915 | reg_errcode_t ret; | |
7916 | ||
7917 | /* GNU code is written to assume at least RE_NREGS registers will be set | |
7918 | (and at least one extra will be -1). */ | |
7919 | bufp->regs_allocated = REGS_UNALLOCATED; | |
7920 | ||
7921 | /* And GNU code determines whether or not to get register information | |
7922 | by passing null for the REGS argument to re_match, etc., not by | |
7923 | setting no_sub. */ | |
7924 | bufp->no_sub = 0; | |
7925 | ||
7926 | /* Match anchors at newline. */ | |
7927 | bufp->newline_anchor = 1; | |
7928 | ||
7929 | # ifdef MBS_SUPPORT | |
7930 | if (MB_CUR_MAX != 1) | |
7931 | ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp); | |
7932 | else | |
7933 | # endif | |
7934 | ret = byte_regex_compile (pattern, length, re_syntax_options, bufp); | |
7935 | ||
7936 | if (!ret) | |
7937 | return NULL; | |
7938 | return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]); | |
7939 | } | |
7940 | #ifdef _LIBC | |
7941 | weak_alias (__re_compile_pattern, re_compile_pattern) | |
7942 | #endif | |
7943 | \f | |
7944 | /* Entry points compatible with 4.2 BSD regex library. We don't define | |
7945 | them unless specifically requested. */ | |
7946 | ||
7947 | #if defined _REGEX_RE_COMP || defined _LIBC | |
7948 | ||
7949 | /* BSD has one and only one pattern buffer. */ | |
7950 | static struct re_pattern_buffer re_comp_buf; | |
7951 | ||
7952 | char * | |
7953 | #ifdef _LIBC | |
7954 | /* Make these definitions weak in libc, so POSIX programs can redefine | |
7955 | these names if they don't use our functions, and still use | |
7956 | regcomp/regexec below without link errors. */ | |
7957 | weak_function | |
7958 | #endif | |
7959 | re_comp (s) | |
7960 | const char *s; | |
7961 | { | |
7962 | reg_errcode_t ret; | |
7963 | ||
7964 | if (!s) | |
7965 | { | |
7966 | if (!re_comp_buf.buffer) | |
7967 | return gettext ("No previous regular expression"); | |
7968 | return 0; | |
7969 | } | |
7970 | ||
7971 | if (!re_comp_buf.buffer) | |
7972 | { | |
7973 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
7974 | if (re_comp_buf.buffer == NULL) | |
7975 | return (char *) gettext (re_error_msgid | |
7976 | + re_error_msgid_idx[(int) REG_ESPACE]); | |
7977 | re_comp_buf.allocated = 200; | |
7978 | ||
7979 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
7980 | if (re_comp_buf.fastmap == NULL) | |
7981 | return (char *) gettext (re_error_msgid | |
7982 | + re_error_msgid_idx[(int) REG_ESPACE]); | |
7983 | } | |
7984 | ||
7985 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
7986 | don't need to initialize the pattern buffer fields which affect it. */ | |
7987 | ||
7988 | /* Match anchors at newlines. */ | |
7989 | re_comp_buf.newline_anchor = 1; | |
7990 | ||
7991 | # ifdef MBS_SUPPORT | |
7992 | if (MB_CUR_MAX != 1) | |
7993 | ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
7994 | else | |
7995 | # endif | |
7996 | ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
7997 | ||
7998 | if (!ret) | |
7999 | return NULL; | |
8000 | ||
8001 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
8002 | return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]); | |
8003 | } | |
8004 | ||
8005 | ||
8006 | int | |
8007 | #ifdef _LIBC | |
8008 | weak_function | |
8009 | #endif | |
8010 | re_exec (s) | |
8011 | const char *s; | |
8012 | { | |
8013 | const int len = strlen (s); | |
8014 | return | |
8015 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
8016 | } | |
8017 | ||
8018 | #endif /* _REGEX_RE_COMP */ | |
8019 | \f | |
8020 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
8021 | ||
8022 | #ifndef emacs | |
8023 | ||
8024 | /* regcomp takes a regular expression as a string and compiles it. | |
8025 | ||
8026 | PREG is a regex_t *. We do not expect any fields to be initialized, | |
8027 | since POSIX says we shouldn't. Thus, we set | |
8028 | ||
8029 | `buffer' to the compiled pattern; | |
8030 | `used' to the length of the compiled pattern; | |
8031 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
8032 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
8033 | RE_SYNTAX_POSIX_BASIC; | |
8034 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
8035 | `fastmap' to an allocated space for the fastmap; | |
8036 | `fastmap_accurate' to zero; | |
8037 | `re_nsub' to the number of subexpressions in PATTERN. | |
8038 | ||
8039 | PATTERN is the address of the pattern string. | |
8040 | ||
8041 | CFLAGS is a series of bits which affect compilation. | |
8042 | ||
8043 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
8044 | use POSIX basic syntax. | |
8045 | ||
8046 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
8047 | Also, regexec will try a match beginning after every newline. | |
8048 | ||
8049 | If REG_ICASE is set, then we considers upper- and lowercase | |
8050 | versions of letters to be equivalent when matching. | |
8051 | ||
8052 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
8053 | routine will report only success or failure, and nothing about the | |
8054 | registers. | |
8055 | ||
8056 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
8057 | the return codes and their meanings.) */ | |
8058 | ||
8059 | int | |
8060 | regcomp (preg, pattern, cflags) | |
8061 | regex_t *preg; | |
8062 | const char *pattern; | |
8063 | int cflags; | |
8064 | { | |
8065 | reg_errcode_t ret; | |
8066 | reg_syntax_t syntax | |
8067 | = (cflags & REG_EXTENDED) ? | |
8068 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
8069 | ||
8070 | /* regex_compile will allocate the space for the compiled pattern. */ | |
8071 | preg->buffer = 0; | |
8072 | preg->allocated = 0; | |
8073 | preg->used = 0; | |
8074 | ||
8075 | /* Try to allocate space for the fastmap. */ | |
8076 | preg->fastmap = (char *) malloc (1 << BYTEWIDTH); | |
8077 | ||
8078 | if (cflags & REG_ICASE) | |
8079 | { | |
8080 | unsigned i; | |
8081 | ||
8082 | preg->translate | |
8083 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
8084 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
8085 | if (preg->translate == NULL) | |
8086 | return (int) REG_ESPACE; | |
8087 | ||
8088 | /* Map uppercase characters to corresponding lowercase ones. */ | |
8089 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
8090 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; | |
8091 | } | |
8092 | else | |
8093 | preg->translate = NULL; | |
8094 | ||
8095 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
8096 | if (cflags & REG_NEWLINE) | |
8097 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
8098 | syntax &= ~RE_DOT_NEWLINE; | |
8099 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
8100 | /* It also changes the matching behavior. */ | |
8101 | preg->newline_anchor = 1; | |
8102 | } | |
8103 | else | |
8104 | preg->newline_anchor = 0; | |
8105 | ||
8106 | preg->no_sub = !!(cflags & REG_NOSUB); | |
8107 | ||
8108 | /* POSIX says a null character in the pattern terminates it, so we | |
8109 | can use strlen here in compiling the pattern. */ | |
8110 | # ifdef MBS_SUPPORT | |
8111 | if (MB_CUR_MAX != 1) | |
8112 | ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg); | |
8113 | else | |
8114 | # endif | |
8115 | ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg); | |
8116 | ||
8117 | /* POSIX doesn't distinguish between an unmatched open-group and an | |
8118 | unmatched close-group: both are REG_EPAREN. */ | |
8119 | if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
8120 | ||
8121 | if (ret == REG_NOERROR && preg->fastmap) | |
8122 | { | |
8123 | /* Compute the fastmap now, since regexec cannot modify the pattern | |
8124 | buffer. */ | |
8125 | if (re_compile_fastmap (preg) == -2) | |
8126 | { | |
8127 | /* Some error occurred while computing the fastmap, just forget | |
8128 | about it. */ | |
8129 | free (preg->fastmap); | |
8130 | preg->fastmap = NULL; | |
8131 | } | |
8132 | } | |
8133 | ||
8134 | return (int) ret; | |
8135 | } | |
8136 | #ifdef _LIBC | |
8137 | weak_alias (__regcomp, regcomp) | |
8138 | #endif | |
8139 | ||
8140 | ||
8141 | /* regexec searches for a given pattern, specified by PREG, in the | |
8142 | string STRING. | |
8143 | ||
8144 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
8145 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
8146 | least NMATCH elements, and we set them to the offsets of the | |
8147 | corresponding matched substrings. | |
8148 | ||
8149 | EFLAGS specifies `execution flags' which affect matching: if | |
8150 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
8151 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
8152 | ||
8153 | We return 0 if we find a match and REG_NOMATCH if not. */ | |
8154 | ||
8155 | int | |
8156 | regexec (preg, string, nmatch, pmatch, eflags) | |
8157 | const regex_t *preg; | |
8158 | const char *string; | |
8159 | size_t nmatch; | |
8160 | regmatch_t pmatch[]; | |
8161 | int eflags; | |
8162 | { | |
8163 | int ret; | |
8164 | struct re_registers regs; | |
8165 | regex_t private_preg; | |
8166 | int len = strlen (string); | |
8167 | boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
8168 | ||
8169 | private_preg = *preg; | |
8170 | ||
8171 | private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
8172 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
8173 | ||
8174 | /* The user has told us exactly how many registers to return | |
8175 | information about, via `nmatch'. We have to pass that on to the | |
8176 | matching routines. */ | |
8177 | private_preg.regs_allocated = REGS_FIXED; | |
8178 | ||
8179 | if (want_reg_info) | |
8180 | { | |
8181 | regs.num_regs = nmatch; | |
8182 | regs.start = TALLOC (nmatch * 2, regoff_t); | |
8183 | if (regs.start == NULL) | |
8184 | return (int) REG_NOMATCH; | |
8185 | regs.end = regs.start + nmatch; | |
8186 | } | |
8187 | ||
8188 | /* Perform the searching operation. */ | |
8189 | ret = re_search (&private_preg, string, len, | |
8190 | /* start: */ 0, /* range: */ len, | |
8191 | want_reg_info ? ®s : (struct re_registers *) 0); | |
8192 | ||
8193 | /* Copy the register information to the POSIX structure. */ | |
8194 | if (want_reg_info) | |
8195 | { | |
8196 | if (ret >= 0) | |
8197 | { | |
8198 | unsigned r; | |
8199 | ||
8200 | for (r = 0; r < nmatch; r++) | |
8201 | { | |
8202 | pmatch[r].rm_so = regs.start[r]; | |
8203 | pmatch[r].rm_eo = regs.end[r]; | |
8204 | } | |
8205 | } | |
8206 | ||
8207 | /* If we needed the temporary register info, free the space now. */ | |
8208 | free (regs.start); | |
8209 | } | |
8210 | ||
8211 | /* We want zero return to mean success, unlike `re_search'. */ | |
8212 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
8213 | } | |
8214 | #ifdef _LIBC | |
8215 | weak_alias (__regexec, regexec) | |
8216 | #endif | |
8217 | ||
8218 | ||
8219 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
8220 | from either regcomp or regexec. We don't use PREG here. */ | |
8221 | ||
8222 | size_t | |
8223 | regerror (errcode, preg, errbuf, errbuf_size) | |
8224 | int errcode; | |
8225 | const regex_t *preg; | |
8226 | char *errbuf; | |
8227 | size_t errbuf_size; | |
8228 | { | |
8229 | const char *msg; | |
8230 | size_t msg_size; | |
8231 | ||
8232 | if (errcode < 0 | |
8233 | || errcode >= (int) (sizeof (re_error_msgid_idx) | |
8234 | / sizeof (re_error_msgid_idx[0]))) | |
8235 | /* Only error codes returned by the rest of the code should be passed | |
8236 | to this routine. If we are given anything else, or if other regex | |
8237 | code generates an invalid error code, then the program has a bug. | |
8238 | Dump core so we can fix it. */ | |
8239 | abort (); | |
8240 | ||
8241 | msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]); | |
8242 | ||
8243 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
8244 | ||
8245 | if (errbuf_size != 0) | |
8246 | { | |
8247 | if (msg_size > errbuf_size) | |
8248 | { | |
8249 | #if defined HAVE_MEMPCPY || defined _LIBC | |
8250 | *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0'; | |
8251 | #else | |
8252 | memcpy (errbuf, msg, errbuf_size - 1); | |
8253 | errbuf[errbuf_size - 1] = 0; | |
8254 | #endif | |
8255 | } | |
8256 | else | |
8257 | memcpy (errbuf, msg, msg_size); | |
8258 | } | |
8259 | ||
8260 | return msg_size; | |
8261 | } | |
8262 | #ifdef _LIBC | |
8263 | weak_alias (__regerror, regerror) | |
8264 | #endif | |
8265 | ||
8266 | ||
8267 | /* Free dynamically allocated space used by PREG. */ | |
8268 | ||
8269 | void | |
8270 | regfree (preg) | |
8271 | regex_t *preg; | |
8272 | { | |
8273 | if (preg->buffer != NULL) | |
8274 | free (preg->buffer); | |
8275 | preg->buffer = NULL; | |
8276 | ||
8277 | preg->allocated = 0; | |
8278 | preg->used = 0; | |
8279 | ||
8280 | if (preg->fastmap != NULL) | |
8281 | free (preg->fastmap); | |
8282 | preg->fastmap = NULL; | |
8283 | preg->fastmap_accurate = 0; | |
8284 | ||
8285 | if (preg->translate != NULL) | |
8286 | free (preg->translate); | |
8287 | preg->translate = NULL; | |
8288 | } | |
8289 | #ifdef _LIBC | |
8290 | weak_alias (__regfree, regfree) | |
8291 | #endif | |
8292 | ||
8293 | #endif /* not emacs */ | |
8294 | ||
8295 | #endif /* not INSIDE_RECURSION */ | |
8296 | ||
8297 | \f | |
8298 | #undef STORE_NUMBER | |
8299 | #undef STORE_NUMBER_AND_INCR | |
8300 | #undef EXTRACT_NUMBER | |
8301 | #undef EXTRACT_NUMBER_AND_INCR | |
8302 | ||
8303 | #undef DEBUG_PRINT_COMPILED_PATTERN | |
8304 | #undef DEBUG_PRINT_DOUBLE_STRING | |
8305 | ||
8306 | #undef INIT_FAIL_STACK | |
8307 | #undef RESET_FAIL_STACK | |
8308 | #undef DOUBLE_FAIL_STACK | |
8309 | #undef PUSH_PATTERN_OP | |
8310 | #undef PUSH_FAILURE_POINTER | |
8311 | #undef PUSH_FAILURE_INT | |
8312 | #undef PUSH_FAILURE_ELT | |
8313 | #undef POP_FAILURE_POINTER | |
8314 | #undef POP_FAILURE_INT | |
8315 | #undef POP_FAILURE_ELT | |
8316 | #undef DEBUG_PUSH | |
8317 | #undef DEBUG_POP | |
8318 | #undef PUSH_FAILURE_POINT | |
8319 | #undef POP_FAILURE_POINT | |
8320 | ||
8321 | #undef REG_UNSET_VALUE | |
8322 | #undef REG_UNSET | |
8323 | ||
8324 | #undef PATFETCH | |
8325 | #undef PATFETCH_RAW | |
8326 | #undef PATUNFETCH | |
8327 | #undef TRANSLATE | |
8328 | ||
8329 | #undef INIT_BUF_SIZE | |
8330 | #undef GET_BUFFER_SPACE | |
8331 | #undef BUF_PUSH | |
8332 | #undef BUF_PUSH_2 | |
8333 | #undef BUF_PUSH_3 | |
8334 | #undef STORE_JUMP | |
8335 | #undef STORE_JUMP2 | |
8336 | #undef INSERT_JUMP | |
8337 | #undef INSERT_JUMP2 | |
8338 | #undef EXTEND_BUFFER | |
8339 | #undef GET_UNSIGNED_NUMBER | |
8340 | #undef FREE_STACK_RETURN | |
8341 | ||
8342 | # undef POINTER_TO_OFFSET | |
8343 | # undef MATCHING_IN_FRST_STRING | |
8344 | # undef PREFETCH | |
8345 | # undef AT_STRINGS_BEG | |
8346 | # undef AT_STRINGS_END | |
8347 | # undef WORDCHAR_P | |
8348 | # undef FREE_VAR | |
8349 | # undef FREE_VARIABLES | |
8350 | # undef NO_HIGHEST_ACTIVE_REG | |
8351 | # undef NO_LOWEST_ACTIVE_REG | |
8352 | ||
8353 | # undef CHAR_T | |
8354 | # undef UCHAR_T | |
8355 | # undef COMPILED_BUFFER_VAR | |
8356 | # undef OFFSET_ADDRESS_SIZE | |
8357 | # undef CHAR_CLASS_SIZE | |
8358 | # undef PREFIX | |
8359 | # undef ARG_PREFIX | |
8360 | # undef PUT_CHAR | |
8361 | # undef BYTE | |
8362 | # undef WCHAR | |
8363 | ||
8364 | # define DEFINED_ONCE |