?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge from emacs-23; up to 2010-06-10T12:56:11Z!michael.albinus@gmx.de.
[emacs.git] / src / regex.c
blob479239897bc89dd53b075befdb329cf4086beda4
1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
4
5 Copyright (C) 1993-2011 Free Software Foundation, Inc.
6
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
10 any later version.
11
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
20 USA. */
21
22 /* TODO:
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
29 */
30
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
33 #pragma alloca
34 #endif
35
36 #ifdef HAVE_CONFIG_H
37 # include <config.h>
38 #endif
39
40 #if defined STDC_HEADERS && !defined emacs
41 # include <stddef.h>
42 #else
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
45 #endif
46
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
49 #if defined _LIBC
50 #define WIDE_CHAR_SUPPORT 1
51 #else
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
54 #endif
55
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
58 #if WIDE_CHAR_SUPPORT
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60 # include <wchar.h>
61 # include <wctype.h>
62 #endif
63
64 #ifdef _LIBC
65 /* We have to keep the namespace clean. */
66 # define regfree(preg) __regfree (preg)
67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69 # define regerror(err_code, preg, errbuf, errbuf_size) \
70 __regerror(err_code, preg, errbuf, errbuf_size)
71 # define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75 # define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77 # define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79 # define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81 # define re_set_syntax(syntax) __re_set_syntax (syntax)
82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
85
86 /* Make sure we call libc's function even if the user overrides them. */
87 # define btowc __btowc
88 # define iswctype __iswctype
89 # define wctype __wctype
90
91 # define WEAK_ALIAS(a,b) weak_alias (a, b)
92
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
97 #else
98 # define WEAK_ALIAS(a,b)
99 #endif
100
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
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 <setjmp.h>
119 # include "lisp.h"
120 # include "buffer.h"
121
122 /* Make syntax table lookup grant data in gl_state. */
123 # define SYNTAX_ENTRY_VIA_PROPERTY
124
125 # include "syntax.h"
126 # include "character.h"
127 # include "category.h"
128
129 # ifdef malloc
130 # undef malloc
131 # endif
132 # define malloc xmalloc
133 # ifdef realloc
134 # undef realloc
135 # endif
136 # define realloc xrealloc
137 # ifdef free
138 # undef free
139 # endif
140 # define free xfree
141
142 /* Converts the pointer to the char to BEG-based offset from the start. */
143 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
144 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
145
146 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
147 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
148 # define RE_STRING_CHAR(p, multibyte) \
149 (multibyte ? (STRING_CHAR (p)) : (*(p)))
150 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
151 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
152
153 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
154
155 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
156
157 /* Set C a (possibly converted to multibyte) character before P. P
158 points into a string which is the virtual concatenation of STR1
159 (which ends at END1) or STR2 (which ends at END2). */
160 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
161 do { \
162 if (target_multibyte) \
163 { \
164 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
165 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
166 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
167 c = STRING_CHAR (dtemp); \
168 } \
169 else \
170 { \
171 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
172 (c) = RE_CHAR_TO_MULTIBYTE (c); \
173 } \
174 } while (0)
175
176 /* Set C a (possibly converted to multibyte) character at P, and set
177 LEN to the byte length of that character. */
178 # define GET_CHAR_AFTER(c, p, len) \
179 do { \
180 if (target_multibyte) \
181 (c) = STRING_CHAR_AND_LENGTH (p, len); \
182 else \
183 { \
184 (c) = *p; \
185 len = 1; \
186 (c) = RE_CHAR_TO_MULTIBYTE (c); \
187 } \
188 } while (0)
189
190 #else /* not emacs */
191
192 /* If we are not linking with Emacs proper,
193 we can't use the relocating allocator
194 even if config.h says that we can. */
195 # undef REL_ALLOC
196
197 # include <unistd.h>
198
199 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
200
201 void *
202 xmalloc (size_t size)
203 {
204 register void *val;
205 val = (void *) malloc (size);
206 if (!val && size)
207 {
208 write (2, "virtual memory exhausted\n", 25);
209 exit (1);
210 }
211 return val;
212 }
213
214 void *
215 xrealloc (void *block, size_t size)
216 {
217 register void *val;
218 /* We must call malloc explicitly when BLOCK is 0, since some
219 reallocs don't do this. */
220 if (! block)
221 val = (void *) malloc (size);
222 else
223 val = (void *) realloc (block, size);
224 if (!val && size)
225 {
226 write (2, "virtual memory exhausted\n", 25);
227 exit (1);
228 }
229 return val;
230 }
231
232 # ifdef malloc
233 # undef malloc
234 # endif
235 # define malloc xmalloc
236 # ifdef realloc
237 # undef realloc
238 # endif
239 # define realloc xrealloc
240
241 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
242 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
243 # include <string.h>
244 # else
245 # include <strings.h>
246 # ifndef memcmp
247 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
248 # endif
249 # ifndef memcpy
250 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
251 # endif
252 # endif
253
254 /* Define the syntax stuff for \<, \>, etc. */
255
256 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
257 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
258
259 # define SWITCH_ENUM_CAST(x) (x)
260
261 /* Dummy macros for non-Emacs environments. */
262 # define CHAR_CHARSET(c) 0
263 # define CHARSET_LEADING_CODE_BASE(c) 0
264 # define MAX_MULTIBYTE_LENGTH 1
265 # define RE_MULTIBYTE_P(x) 0
266 # define RE_TARGET_MULTIBYTE_P(x) 0
267 # define WORD_BOUNDARY_P(c1, c2) (0)
268 # define CHAR_HEAD_P(p) (1)
269 # define SINGLE_BYTE_CHAR_P(c) (1)
270 # define SAME_CHARSET_P(c1, c2) (1)
271 # define BYTES_BY_CHAR_HEAD(p) (1)
272 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
273 # define STRING_CHAR(p) (*(p))
274 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
275 # define CHAR_STRING(c, s) (*(s) = (c), 1)
276 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
277 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
278 # define RE_CHAR_TO_MULTIBYTE(c) (c)
279 # define RE_CHAR_TO_UNIBYTE(c) (c)
280 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
281 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
282 # define GET_CHAR_AFTER(c, p, len) \
283 (c = *p, len = 1)
284 # define MAKE_CHAR(charset, c1, c2) (c1)
285 # define BYTE8_TO_CHAR(c) (c)
286 # define CHAR_BYTE8_P(c) (0)
287 # define CHAR_LEADING_CODE(c) (c)
288
289 #endif /* not emacs */
290
291 #ifndef RE_TRANSLATE
292 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
293 # define RE_TRANSLATE_P(TBL) (TBL)
294 #endif
295 \f
296 /* Get the interface, including the syntax bits. */
297 #include "regex.h"
298
299 /* isalpha etc. are used for the character classes. */
300 #include <ctype.h>
301
302 #ifdef emacs
303
304 /* 1 if C is an ASCII character. */
305 # define IS_REAL_ASCII(c) ((c) < 0200)
306
307 /* 1 if C is a unibyte character. */
308 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
309
310 /* The Emacs definitions should not be directly affected by locales. */
311
312 /* In Emacs, these are only used for single-byte characters. */
313 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
314 # define ISCNTRL(c) ((c) < ' ')
315 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
316 || ((c) >= 'a' && (c) <= 'f') \
317 || ((c) >= 'A' && (c) <= 'F'))
318
319 /* This is only used for single-byte characters. */
320 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
321
322 /* The rest must handle multibyte characters. */
323
324 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
325 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
326 : 1)
327
328 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
329 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
330 : 1)
331
332 # define ISALNUM(c) (IS_REAL_ASCII (c) \
333 ? (((c) >= 'a' && (c) <= 'z') \
334 || ((c) >= 'A' && (c) <= 'Z') \
335 || ((c) >= '0' && (c) <= '9')) \
336 : SYNTAX (c) == Sword)
337
338 # define ISALPHA(c) (IS_REAL_ASCII (c) \
339 ? (((c) >= 'a' && (c) <= 'z') \
340 || ((c) >= 'A' && (c) <= 'Z')) \
341 : SYNTAX (c) == Sword)
342
343 # define ISLOWER(c) lowercasep (c)
344
345 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
346 ? ((c) > ' ' && (c) < 0177 \
347 && !(((c) >= 'a' && (c) <= 'z') \
348 || ((c) >= 'A' && (c) <= 'Z') \
349 || ((c) >= '0' && (c) <= '9'))) \
350 : SYNTAX (c) != Sword)
351
352 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
353
354 # define ISUPPER(c) uppercasep (c)
355
356 # define ISWORD(c) (SYNTAX (c) == Sword)
357
358 #else /* not emacs */
359
360 /* Jim Meyering writes:
361
362 "... Some ctype macros are valid only for character codes that
363 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
364 using /bin/cc or gcc but without giving an ansi option). So, all
365 ctype uses should be through macros like ISPRINT... If
366 STDC_HEADERS is defined, then autoconf has verified that the ctype
367 macros don't need to be guarded with references to isascii. ...
368 Defining isascii to 1 should let any compiler worth its salt
369 eliminate the && through constant folding."
370 Solaris defines some of these symbols so we must undefine them first. */
371
372 # undef ISASCII
373 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
374 # define ISASCII(c) 1
375 # else
376 # define ISASCII(c) isascii(c)
377 # endif
378
379 /* 1 if C is an ASCII character. */
380 # define IS_REAL_ASCII(c) ((c) < 0200)
381
382 /* This distinction is not meaningful, except in Emacs. */
383 # define ISUNIBYTE(c) 1
384
385 # ifdef isblank
386 # define ISBLANK(c) (ISASCII (c) && isblank (c))
387 # else
388 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
389 # endif
390 # ifdef isgraph
391 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
392 # else
393 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
394 # endif
395
396 # undef ISPRINT
397 # define ISPRINT(c) (ISASCII (c) && isprint (c))
398 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
399 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
400 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
401 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
402 # define ISLOWER(c) (ISASCII (c) && islower (c))
403 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
404 # define ISSPACE(c) (ISASCII (c) && isspace (c))
405 # define ISUPPER(c) (ISASCII (c) && isupper (c))
406 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
407
408 # define ISWORD(c) ISALPHA(c)
409
410 # ifdef _tolower
411 # define TOLOWER(c) _tolower(c)
412 # else
413 # define TOLOWER(c) tolower(c)
414 # endif
415
416 /* How many characters in the character set. */
417 # define CHAR_SET_SIZE 256
418
419 # ifdef SYNTAX_TABLE
420
421 extern char *re_syntax_table;
422
423 # else /* not SYNTAX_TABLE */
424
425 static char re_syntax_table[CHAR_SET_SIZE];
426
427 static void
428 init_syntax_once (void)
429 {
430 register int c;
431 static int done = 0;
432
433 if (done)
434 return;
435
436 memset (re_syntax_table, 0, sizeof re_syntax_table);
437
438 for (c = 0; c < CHAR_SET_SIZE; ++c)
439 if (ISALNUM (c))
440 re_syntax_table[c] = Sword;
441
442 re_syntax_table['_'] = Ssymbol;
443
444 done = 1;
445 }
446
447 # endif /* not SYNTAX_TABLE */
448
449 # define SYNTAX(c) re_syntax_table[(c)]
450
451 #endif /* not emacs */
452 \f
453 #ifndef NULL
454 # define NULL (void *)0
455 #endif
456
457 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
458 since ours (we hope) works properly with all combinations of
459 machines, compilers, `char' and `unsigned char' argument types.
460 (Per Bothner suggested the basic approach.) */
461 #undef SIGN_EXTEND_CHAR
462 #if __STDC__
463 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
464 #else /* not __STDC__ */
465 /* As in Harbison and Steele. */
466 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
467 #endif
468 \f
469 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
470 use `alloca' instead of `malloc'. This is because using malloc in
471 re_search* or re_match* could cause memory leaks when C-g is used in
472 Emacs; also, malloc is slower and causes storage fragmentation. On
473 the other hand, malloc is more portable, and easier to debug.
474
475 Because we sometimes use alloca, some routines have to be macros,
476 not functions -- `alloca'-allocated space disappears at the end of the
477 function it is called in. */
478
479 #ifdef REGEX_MALLOC
480
481 # define REGEX_ALLOCATE malloc
482 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
483 # define REGEX_FREE free
484
485 #else /* not REGEX_MALLOC */
486
487 /* Emacs already defines alloca, sometimes. */
488 # ifndef alloca
489
490 /* Make alloca work the best possible way. */
491 # ifdef __GNUC__
492 # define alloca __builtin_alloca
493 # else /* not __GNUC__ */
494 # ifdef HAVE_ALLOCA_H
495 # include <alloca.h>
496 # endif /* HAVE_ALLOCA_H */
497 # endif /* not __GNUC__ */
498
499 # endif /* not alloca */
500
501 # define REGEX_ALLOCATE alloca
502
503 /* Assumes a `char *destination' variable. */
504 # define REGEX_REALLOCATE(source, osize, nsize) \
505 (destination = (char *) alloca (nsize), \
506 memcpy (destination, source, osize))
507
508 /* No need to do anything to free, after alloca. */
509 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
510
511 #endif /* not REGEX_MALLOC */
512
513 /* Define how to allocate the failure stack. */
514
515 #if defined REL_ALLOC && defined REGEX_MALLOC
516
517 # define REGEX_ALLOCATE_STACK(size) \
518 r_alloc (&failure_stack_ptr, (size))
519 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
520 r_re_alloc (&failure_stack_ptr, (nsize))
521 # define REGEX_FREE_STACK(ptr) \
522 r_alloc_free (&failure_stack_ptr)
523
524 #else /* not using relocating allocator */
525
526 # ifdef REGEX_MALLOC
527
528 # define REGEX_ALLOCATE_STACK malloc
529 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
530 # define REGEX_FREE_STACK free
531
532 # else /* not REGEX_MALLOC */
533
534 # define REGEX_ALLOCATE_STACK alloca
535
536 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
537 REGEX_REALLOCATE (source, osize, nsize)
538 /* No need to explicitly free anything. */
539 # define REGEX_FREE_STACK(arg) ((void)0)
540
541 # endif /* not REGEX_MALLOC */
542 #endif /* not using relocating allocator */
543
544
545 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
546 `string1' or just past its end. This works if PTR is NULL, which is
547 a good thing. */
548 #define FIRST_STRING_P(ptr) \
549 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
550
551 /* (Re)Allocate N items of type T using malloc, or fail. */
552 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
553 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
554 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
555
556 #define BYTEWIDTH 8 /* In bits. */
557
558 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
559
560 #undef MAX
561 #undef MIN
562 #define MAX(a, b) ((a) > (b) ? (a) : (b))
563 #define MIN(a, b) ((a) < (b) ? (a) : (b))
564
565 /* Type of source-pattern and string chars. */
566 typedef const unsigned char re_char;
567
568 typedef char boolean;
569 #define false 0
570 #define true 1
571
572 static regoff_t re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
573 re_char *string1, size_t size1,
574 re_char *string2, size_t size2,
575 ssize_t pos,
576 struct re_registers *regs,
577 ssize_t stop));
578 \f
579 /* These are the command codes that appear in compiled regular
580 expressions. Some opcodes are followed by argument bytes. A
581 command code can specify any interpretation whatsoever for its
582 arguments. Zero bytes may appear in the compiled regular expression. */
583
584 typedef enum
585 {
586 no_op = 0,
587
588 /* Succeed right away--no more backtracking. */
589 succeed,
590
591 /* Followed by one byte giving n, then by n literal bytes. */
592 exactn,
593
594 /* Matches any (more or less) character. */
595 anychar,
596
597 /* Matches any one char belonging to specified set. First
598 following byte is number of bitmap bytes. Then come bytes
599 for a bitmap saying which chars are in. Bits in each byte
600 are ordered low-bit-first. A character is in the set if its
601 bit is 1. A character too large to have a bit in the map is
602 automatically not in the set.
603
604 If the length byte has the 0x80 bit set, then that stuff
605 is followed by a range table:
606 2 bytes of flags for character sets (low 8 bits, high 8 bits)
607 See RANGE_TABLE_WORK_BITS below.
608 2 bytes, the number of pairs that follow (upto 32767)
609 pairs, each 2 multibyte characters,
610 each multibyte character represented as 3 bytes. */
611 charset,
612
613 /* Same parameters as charset, but match any character that is
614 not one of those specified. */
615 charset_not,
616
617 /* Start remembering the text that is matched, for storing in a
618 register. Followed by one byte with the register number, in
619 the range 0 to one less than the pattern buffer's re_nsub
620 field. */
621 start_memory,
622
623 /* Stop remembering the text that is matched and store it in a
624 memory register. Followed by one byte with the register
625 number, in the range 0 to one less than `re_nsub' in the
626 pattern buffer. */
627 stop_memory,
628
629 /* Match a duplicate of something remembered. Followed by one
630 byte containing the register number. */
631 duplicate,
632
633 /* Fail unless at beginning of line. */
634 begline,
635
636 /* Fail unless at end of line. */
637 endline,
638
639 /* Succeeds if at beginning of buffer (if emacs) or at beginning
640 of string to be matched (if not). */
641 begbuf,
642
643 /* Analogously, for end of buffer/string. */
644 endbuf,
645
646 /* Followed by two byte relative address to which to jump. */
647 jump,
648
649 /* Followed by two-byte relative address of place to resume at
650 in case of failure. */
651 on_failure_jump,
652
653 /* Like on_failure_jump, but pushes a placeholder instead of the
654 current string position when executed. */
655 on_failure_keep_string_jump,
656
657 /* Just like `on_failure_jump', except that it checks that we
658 don't get stuck in an infinite loop (matching an empty string
659 indefinitely). */
660 on_failure_jump_loop,
661
662 /* Just like `on_failure_jump_loop', except that it checks for
663 a different kind of loop (the kind that shows up with non-greedy
664 operators). This operation has to be immediately preceded
665 by a `no_op'. */
666 on_failure_jump_nastyloop,
667
668 /* A smart `on_failure_jump' used for greedy * and + operators.
669 It analyses the loop before which it is put and if the
670 loop does not require backtracking, it changes itself to
671 `on_failure_keep_string_jump' and short-circuits the loop,
672 else it just defaults to changing itself into `on_failure_jump'.
673 It assumes that it is pointing to just past a `jump'. */
674 on_failure_jump_smart,
675
676 /* Followed by two-byte relative address and two-byte number n.
677 After matching N times, jump to the address upon failure.
678 Does not work if N starts at 0: use on_failure_jump_loop
679 instead. */
680 succeed_n,
681
682 /* Followed by two-byte relative address, and two-byte number n.
683 Jump to the address N times, then fail. */
684 jump_n,
685
686 /* Set the following two-byte relative address to the
687 subsequent two-byte number. The address *includes* the two
688 bytes of number. */
689 set_number_at,
690
691 wordbeg, /* Succeeds if at word beginning. */
692 wordend, /* Succeeds if at word end. */
693
694 wordbound, /* Succeeds if at a word boundary. */
695 notwordbound, /* Succeeds if not at a word boundary. */
696
697 symbeg, /* Succeeds if at symbol beginning. */
698 symend, /* Succeeds if at symbol end. */
699
700 /* Matches any character whose syntax is specified. Followed by
701 a byte which contains a syntax code, e.g., Sword. */
702 syntaxspec,
703
704 /* Matches any character whose syntax is not that specified. */
705 notsyntaxspec
706
707 #ifdef emacs
708 ,before_dot, /* Succeeds if before point. */
709 at_dot, /* Succeeds if at point. */
710 after_dot, /* Succeeds if after point. */
711
712 /* Matches any character whose category-set contains the specified
713 category. The operator is followed by a byte which contains a
714 category code (mnemonic ASCII character). */
715 categoryspec,
716
717 /* Matches any character whose category-set does not contain the
718 specified category. The operator is followed by a byte which
719 contains the category code (mnemonic ASCII character). */
720 notcategoryspec
721 #endif /* emacs */
722 } re_opcode_t;
723 \f
724 /* Common operations on the compiled pattern. */
725
726 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
727
728 #define STORE_NUMBER(destination, number) \
729 do { \
730 (destination)[0] = (number) & 0377; \
731 (destination)[1] = (number) >> 8; \
732 } while (0)
733
734 /* Same as STORE_NUMBER, except increment DESTINATION to
735 the byte after where the number is stored. Therefore, DESTINATION
736 must be an lvalue. */
737
738 #define STORE_NUMBER_AND_INCR(destination, number) \
739 do { \
740 STORE_NUMBER (destination, number); \
741 (destination) += 2; \
742 } while (0)
743
744 /* Put into DESTINATION a number stored in two contiguous bytes starting
745 at SOURCE. */
746
747 #define EXTRACT_NUMBER(destination, source) \
748 do { \
749 (destination) = *(source) & 0377; \
750 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
751 } while (0)
752
753 #ifdef DEBUG
754 static void extract_number _RE_ARGS ((int *dest, re_char *source));
755 static void
756 extract_number (dest, source)
757 int *dest;
758 re_char *source;
759 {
760 int temp = SIGN_EXTEND_CHAR (*(source + 1));
761 *dest = *source & 0377;
762 *dest += temp << 8;
763 }
764
765 # ifndef EXTRACT_MACROS /* To debug the macros. */
766 # undef EXTRACT_NUMBER
767 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
768 # endif /* not EXTRACT_MACROS */
769
770 #endif /* DEBUG */
771
772 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
773 SOURCE must be an lvalue. */
774
775 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
776 do { \
777 EXTRACT_NUMBER (destination, source); \
778 (source) += 2; \
779 } while (0)
780
781 #ifdef DEBUG
782 static void extract_number_and_incr _RE_ARGS ((int *destination,
783 re_char **source));
784 static void
785 extract_number_and_incr (destination, source)
786 int *destination;
787 re_char **source;
788 {
789 extract_number (destination, *source);
790 *source += 2;
791 }
792
793 # ifndef EXTRACT_MACROS
794 # undef EXTRACT_NUMBER_AND_INCR
795 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
796 extract_number_and_incr (&dest, &src)
797 # endif /* not EXTRACT_MACROS */
798
799 #endif /* DEBUG */
800 \f
801 /* Store a multibyte character in three contiguous bytes starting
802 DESTINATION, and increment DESTINATION to the byte after where the
803 character is stored. Therefore, DESTINATION must be an lvalue. */
804
805 #define STORE_CHARACTER_AND_INCR(destination, character) \
806 do { \
807 (destination)[0] = (character) & 0377; \
808 (destination)[1] = ((character) >> 8) & 0377; \
809 (destination)[2] = (character) >> 16; \
810 (destination) += 3; \
811 } while (0)
812
813 /* Put into DESTINATION a character stored in three contiguous bytes
814 starting at SOURCE. */
815
816 #define EXTRACT_CHARACTER(destination, source) \
817 do { \
818 (destination) = ((source)[0] \
819 | ((source)[1] << 8) \
820 | ((source)[2] << 16)); \
821 } while (0)
822
823
824 /* Macros for charset. */
825
826 /* Size of bitmap of charset P in bytes. P is a start of charset,
827 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
828 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
829
830 /* Nonzero if charset P has range table. */
831 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
832
833 /* Return the address of range table of charset P. But not the start
834 of table itself, but the before where the number of ranges is
835 stored. `2 +' means to skip re_opcode_t and size of bitmap,
836 and the 2 bytes of flags at the start of the range table. */
837 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
838
839 /* Extract the bit flags that start a range table. */
840 #define CHARSET_RANGE_TABLE_BITS(p) \
841 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
842 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
843
844 /* Return the address of end of RANGE_TABLE. COUNT is number of
845 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
846 is start of range and end of range. `* 3' is size of each start
847 and end. */
848 #define CHARSET_RANGE_TABLE_END(range_table, count) \
849 ((range_table) + (count) * 2 * 3)
850
851 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
852 COUNT is number of ranges in RANGE_TABLE. */
853 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
854 do \
855 { \
856 re_wchar_t range_start, range_end; \
857 re_char *rtp; \
858 re_char *range_table_end \
859 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
860 \
861 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
862 { \
863 EXTRACT_CHARACTER (range_start, rtp); \
864 EXTRACT_CHARACTER (range_end, rtp + 3); \
865 \
866 if (range_start <= (c) && (c) <= range_end) \
867 { \
868 (not) = !(not); \
869 break; \
870 } \
871 } \
872 } \
873 while (0)
874
875 /* Test if C is in range table of CHARSET. The flag NOT is negated if
876 C is listed in it. */
877 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
878 do \
879 { \
880 /* Number of ranges in range table. */ \
881 int count; \
882 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
883 \
884 EXTRACT_NUMBER_AND_INCR (count, range_table); \
885 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
886 } \
887 while (0)
888 \f
889 /* If DEBUG is defined, Regex prints many voluminous messages about what
890 it is doing (if the variable `debug' is nonzero). If linked with the
891 main program in `iregex.c', you can enter patterns and strings
892 interactively. And if linked with the main program in `main.c' and
893 the other test files, you can run the already-written tests. */
894
895 #ifdef DEBUG
896
897 /* We use standard I/O for debugging. */
898 # include <stdio.h>
899
900 /* It is useful to test things that ``must'' be true when debugging. */
901 # include <assert.h>
902
903 static int debug = -100000;
904
905 # define DEBUG_STATEMENT(e) e
906 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
907 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
908 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
909 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
910 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
911 if (debug > 0) print_partial_compiled_pattern (s, e)
912 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
913 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
914
915
916 /* Print the fastmap in human-readable form. */
917
918 void
919 print_fastmap (fastmap)
920 char *fastmap;
921 {
922 unsigned was_a_range = 0;
923 unsigned i = 0;
924
925 while (i < (1 << BYTEWIDTH))
926 {
927 if (fastmap[i++])
928 {
929 was_a_range = 0;
930 putchar (i - 1);
931 while (i < (1 << BYTEWIDTH) && fastmap[i])
932 {
933 was_a_range = 1;
934 i++;
935 }
936 if (was_a_range)
937 {
938 printf ("-");
939 putchar (i - 1);
940 }
941 }
942 }
943 putchar ('\n');
944 }
945
946
947 /* Print a compiled pattern string in human-readable form, starting at
948 the START pointer into it and ending just before the pointer END. */
949
950 void
951 print_partial_compiled_pattern (start, end)
952 re_char *start;
953 re_char *end;
954 {
955 int mcnt, mcnt2;
956 re_char *p = start;
957 re_char *pend = end;
958
959 if (start == NULL)
960 {
961 fprintf (stderr, "(null)\n");
962 return;
963 }
964
965 /* Loop over pattern commands. */
966 while (p < pend)
967 {
968 fprintf (stderr, "%d:\t", p - start);
969
970 switch ((re_opcode_t) *p++)
971 {
972 case no_op:
973 fprintf (stderr, "/no_op");
974 break;
975
976 case succeed:
977 fprintf (stderr, "/succeed");
978 break;
979
980 case exactn:
981 mcnt = *p++;
982 fprintf (stderr, "/exactn/%d", mcnt);
983 do
984 {
985 fprintf (stderr, "/%c", *p++);
986 }
987 while (--mcnt);
988 break;
989
990 case start_memory:
991 fprintf (stderr, "/start_memory/%d", *p++);
992 break;
993
994 case stop_memory:
995 fprintf (stderr, "/stop_memory/%d", *p++);
996 break;
997
998 case duplicate:
999 fprintf (stderr, "/duplicate/%d", *p++);
1000 break;
1001
1002 case anychar:
1003 fprintf (stderr, "/anychar");
1004 break;
1005
1006 case charset:
1007 case charset_not:
1008 {
1009 register int c, last = -100;
1010 register int in_range = 0;
1011 int length = CHARSET_BITMAP_SIZE (p - 1);
1012 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
1013
1014 fprintf (stderr, "/charset [%s",
1015 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
1016
1017 if (p + *p >= pend)
1018 fprintf (stderr, " !extends past end of pattern! ");
1019
1020 for (c = 0; c < 256; c++)
1021 if (c / 8 < length
1022 && (p[1 + (c/8)] & (1 << (c % 8))))
1023 {
1024 /* Are we starting a range? */
1025 if (last + 1 == c && ! in_range)
1026 {
1027 fprintf (stderr, "-");
1028 in_range = 1;
1029 }
1030 /* Have we broken a range? */
1031 else if (last + 1 != c && in_range)
1032 {
1033 fprintf (stderr, "%c", last);
1034 in_range = 0;
1035 }
1036
1037 if (! in_range)
1038 fprintf (stderr, "%c", c);
1039
1040 last = c;
1041 }
1042
1043 if (in_range)
1044 fprintf (stderr, "%c", last);
1045
1046 fprintf (stderr, "]");
1047
1048 p += 1 + length;
1049
1050 if (has_range_table)
1051 {
1052 int count;
1053 fprintf (stderr, "has-range-table");
1054
1055 /* ??? Should print the range table; for now, just skip it. */
1056 p += 2; /* skip range table bits */
1057 EXTRACT_NUMBER_AND_INCR (count, p);
1058 p = CHARSET_RANGE_TABLE_END (p, count);
1059 }
1060 }
1061 break;
1062
1063 case begline:
1064 fprintf (stderr, "/begline");
1065 break;
1066
1067 case endline:
1068 fprintf (stderr, "/endline");
1069 break;
1070
1071 case on_failure_jump:
1072 extract_number_and_incr (&mcnt, &p);
1073 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1074 break;
1075
1076 case on_failure_keep_string_jump:
1077 extract_number_and_incr (&mcnt, &p);
1078 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1079 break;
1080
1081 case on_failure_jump_nastyloop:
1082 extract_number_and_incr (&mcnt, &p);
1083 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1084 break;
1085
1086 case on_failure_jump_loop:
1087 extract_number_and_incr (&mcnt, &p);
1088 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1089 break;
1090
1091 case on_failure_jump_smart:
1092 extract_number_and_incr (&mcnt, &p);
1093 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1094 break;
1095
1096 case jump:
1097 extract_number_and_incr (&mcnt, &p);
1098 fprintf (stderr, "/jump to %d", p + mcnt - start);
1099 break;
1100
1101 case succeed_n:
1102 extract_number_and_incr (&mcnt, &p);
1103 extract_number_and_incr (&mcnt2, &p);
1104 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1105 break;
1106
1107 case jump_n:
1108 extract_number_and_incr (&mcnt, &p);
1109 extract_number_and_incr (&mcnt2, &p);
1110 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1111 break;
1112
1113 case set_number_at:
1114 extract_number_and_incr (&mcnt, &p);
1115 extract_number_and_incr (&mcnt2, &p);
1116 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1117 break;
1118
1119 case wordbound:
1120 fprintf (stderr, "/wordbound");
1121 break;
1122
1123 case notwordbound:
1124 fprintf (stderr, "/notwordbound");
1125 break;
1126
1127 case wordbeg:
1128 fprintf (stderr, "/wordbeg");
1129 break;
1130
1131 case wordend:
1132 fprintf (stderr, "/wordend");
1133 break;
1134
1135 case symbeg:
1136 fprintf (stderr, "/symbeg");
1137 break;
1138
1139 case symend:
1140 fprintf (stderr, "/symend");
1141 break;
1142
1143 case syntaxspec:
1144 fprintf (stderr, "/syntaxspec");
1145 mcnt = *p++;
1146 fprintf (stderr, "/%d", mcnt);
1147 break;
1148
1149 case notsyntaxspec:
1150 fprintf (stderr, "/notsyntaxspec");
1151 mcnt = *p++;
1152 fprintf (stderr, "/%d", mcnt);
1153 break;
1154
1155 # ifdef emacs
1156 case before_dot:
1157 fprintf (stderr, "/before_dot");
1158 break;
1159
1160 case at_dot:
1161 fprintf (stderr, "/at_dot");
1162 break;
1163
1164 case after_dot:
1165 fprintf (stderr, "/after_dot");
1166 break;
1167
1168 case categoryspec:
1169 fprintf (stderr, "/categoryspec");
1170 mcnt = *p++;
1171 fprintf (stderr, "/%d", mcnt);
1172 break;
1173
1174 case notcategoryspec:
1175 fprintf (stderr, "/notcategoryspec");
1176 mcnt = *p++;
1177 fprintf (stderr, "/%d", mcnt);
1178 break;
1179 # endif /* emacs */
1180
1181 case begbuf:
1182 fprintf (stderr, "/begbuf");
1183 break;
1184
1185 case endbuf:
1186 fprintf (stderr, "/endbuf");
1187 break;
1188
1189 default:
1190 fprintf (stderr, "?%d", *(p-1));
1191 }
1192
1193 fprintf (stderr, "\n");
1194 }
1195
1196 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1197 }
1198
1199
1200 void
1201 print_compiled_pattern (bufp)
1202 struct re_pattern_buffer *bufp;
1203 {
1204 re_char *buffer = bufp->buffer;
1205
1206 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1207 printf ("%ld bytes used/%ld bytes allocated.\n",
1208 bufp->used, bufp->allocated);
1209
1210 if (bufp->fastmap_accurate && bufp->fastmap)
1211 {
1212 printf ("fastmap: ");
1213 print_fastmap (bufp->fastmap);
1214 }
1215
1216 printf ("re_nsub: %d\t", bufp->re_nsub);
1217 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1218 printf ("can_be_null: %d\t", bufp->can_be_null);
1219 printf ("no_sub: %d\t", bufp->no_sub);
1220 printf ("not_bol: %d\t", bufp->not_bol);
1221 printf ("not_eol: %d\t", bufp->not_eol);
1222 printf ("syntax: %lx\n", bufp->syntax);
1223 fflush (stdout);
1224 /* Perhaps we should print the translate table? */
1225 }
1226
1227
1228 void
1229 print_double_string (where, string1, size1, string2, size2)
1230 re_char *where;
1231 re_char *string1;
1232 re_char *string2;
1233 ssize_t size1;
1234 ssize_t size2;
1235 {
1236 ssize_t this_char;
1237
1238 if (where == NULL)
1239 printf ("(null)");
1240 else
1241 {
1242 if (FIRST_STRING_P (where))
1243 {
1244 for (this_char = where - string1; this_char < size1; this_char++)
1245 putchar (string1[this_char]);
1246
1247 where = string2;
1248 }
1249
1250 for (this_char = where - string2; this_char < size2; this_char++)
1251 putchar (string2[this_char]);
1252 }
1253 }
1254
1255 #else /* not DEBUG */
1256
1257 # undef assert
1258 # define assert(e)
1259
1260 # define DEBUG_STATEMENT(e)
1261 # define DEBUG_PRINT1(x)
1262 # define DEBUG_PRINT2(x1, x2)
1263 # define DEBUG_PRINT3(x1, x2, x3)
1264 # define DEBUG_PRINT4(x1, x2, x3, x4)
1265 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1266 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1267
1268 #endif /* not DEBUG */
1269 \f
1270 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1271 #ifdef lint
1272 # define IF_LINT(Code) Code
1273 #else
1274 # define IF_LINT(Code) /* empty */
1275 #endif
1276 \f
1277 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1278 also be assigned to arbitrarily: each pattern buffer stores its own
1279 syntax, so it can be changed between regex compilations. */
1280 /* This has no initializer because initialized variables in Emacs
1281 become read-only after dumping. */
1282 reg_syntax_t re_syntax_options;
1283
1284
1285 /* Specify the precise syntax of regexps for compilation. This provides
1286 for compatibility for various utilities which historically have
1287 different, incompatible syntaxes.
1288
1289 The argument SYNTAX is a bit mask comprised of the various bits
1290 defined in regex.h. We return the old syntax. */
1291
1292 reg_syntax_t
1293 re_set_syntax (reg_syntax_t syntax)
1294 {
1295 reg_syntax_t ret = re_syntax_options;
1296
1297 re_syntax_options = syntax;
1298 return ret;
1299 }
1300 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1301
1302 /* Regexp to use to replace spaces, or NULL meaning don't. */
1303 static re_char *whitespace_regexp;
1304
1305 void
1306 re_set_whitespace_regexp (const char *regexp)
1307 {
1308 whitespace_regexp = (re_char *) regexp;
1309 }
1310 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1311 \f
1312 /* This table gives an error message for each of the error codes listed
1313 in regex.h. Obviously the order here has to be same as there.
1314 POSIX doesn't require that we do anything for REG_NOERROR,
1315 but why not be nice? */
1316
1317 static const char *re_error_msgid[] =
1318 {
1319 gettext_noop ("Success"), /* REG_NOERROR */
1320 gettext_noop ("No match"), /* REG_NOMATCH */
1321 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1322 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1323 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1324 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1325 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1326 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1327 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1328 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1329 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1330 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1331 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1332 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1333 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1334 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1335 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1336 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1337 };
1338 \f
1339 /* Avoiding alloca during matching, to placate r_alloc. */
1340
1341 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1342 searching and matching functions should not call alloca. On some
1343 systems, alloca is implemented in terms of malloc, and if we're
1344 using the relocating allocator routines, then malloc could cause a
1345 relocation, which might (if the strings being searched are in the
1346 ralloc heap) shift the data out from underneath the regexp
1347 routines.
1348
1349 Here's another reason to avoid allocation: Emacs
1350 processes input from X in a signal handler; processing X input may
1351 call malloc; if input arrives while a matching routine is calling
1352 malloc, then we're scrod. But Emacs can't just block input while
1353 calling matching routines; then we don't notice interrupts when
1354 they come in. So, Emacs blocks input around all regexp calls
1355 except the matching calls, which it leaves unprotected, in the
1356 faith that they will not malloc. */
1357
1358 /* Normally, this is fine. */
1359 #define MATCH_MAY_ALLOCATE
1360
1361 /* The match routines may not allocate if (1) they would do it with malloc
1362 and (2) it's not safe for them to use malloc.
1363 Note that if REL_ALLOC is defined, matching would not use malloc for the
1364 failure stack, but we would still use it for the register vectors;
1365 so REL_ALLOC should not affect this. */
1366 #if defined REGEX_MALLOC && defined emacs
1367 # undef MATCH_MAY_ALLOCATE
1368 #endif
1369
1370 \f
1371 /* Failure stack declarations and macros; both re_compile_fastmap and
1372 re_match_2 use a failure stack. These have to be macros because of
1373 REGEX_ALLOCATE_STACK. */
1374
1375
1376 /* Approximate number of failure points for which to initially allocate space
1377 when matching. If this number is exceeded, we allocate more
1378 space, so it is not a hard limit. */
1379 #ifndef INIT_FAILURE_ALLOC
1380 # define INIT_FAILURE_ALLOC 20
1381 #endif
1382
1383 /* Roughly the maximum number of failure points on the stack. Would be
1384 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1385 This is a variable only so users of regex can assign to it; we never
1386 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1387 before using it, so it should probably be a byte-count instead. */
1388 # if defined MATCH_MAY_ALLOCATE
1389 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1390 whose default stack limit is 2mb. In order for a larger
1391 value to work reliably, you have to try to make it accord
1392 with the process stack limit. */
1393 size_t re_max_failures = 40000;
1394 # else
1395 size_t re_max_failures = 4000;
1396 # endif
1397
1398 union fail_stack_elt
1399 {
1400 re_char *pointer;
1401 /* This should be the biggest `int' that's no bigger than a pointer. */
1402 long integer;
1403 };
1404
1405 typedef union fail_stack_elt fail_stack_elt_t;
1406
1407 typedef struct
1408 {
1409 fail_stack_elt_t *stack;
1410 size_t size;
1411 size_t avail; /* Offset of next open position. */
1412 size_t frame; /* Offset of the cur constructed frame. */
1413 } fail_stack_type;
1414
1415 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1416
1417
1418 /* Define macros to initialize and free the failure stack.
1419 Do `return -2' if the alloc fails. */
1420
1421 #ifdef MATCH_MAY_ALLOCATE
1422 # define INIT_FAIL_STACK() \
1423 do { \
1424 fail_stack.stack = (fail_stack_elt_t *) \
1425 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1426 * sizeof (fail_stack_elt_t)); \
1427 \
1428 if (fail_stack.stack == NULL) \
1429 return -2; \
1430 \
1431 fail_stack.size = INIT_FAILURE_ALLOC; \
1432 fail_stack.avail = 0; \
1433 fail_stack.frame = 0; \
1434 } while (0)
1435 #else
1436 # define INIT_FAIL_STACK() \
1437 do { \
1438 fail_stack.avail = 0; \
1439 fail_stack.frame = 0; \
1440 } while (0)
1441
1442 # define RETALLOC_IF(addr, n, t) \
1443 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1444 #endif
1445
1446
1447 /* Double the size of FAIL_STACK, up to a limit
1448 which allows approximately `re_max_failures' items.
1449
1450 Return 1 if succeeds, and 0 if either ran out of memory
1451 allocating space for it or it was already too large.
1452
1453 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1454
1455 /* Factor to increase the failure stack size by
1456 when we increase it.
1457 This used to be 2, but 2 was too wasteful
1458 because the old discarded stacks added up to as much space
1459 were as ultimate, maximum-size stack. */
1460 #define FAIL_STACK_GROWTH_FACTOR 4
1461
1462 #define GROW_FAIL_STACK(fail_stack) \
1463 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1464 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1465 ? 0 \
1466 : ((fail_stack).stack \
1467 = (fail_stack_elt_t *) \
1468 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1469 (fail_stack).size * sizeof (fail_stack_elt_t), \
1470 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1471 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1472 * FAIL_STACK_GROWTH_FACTOR))), \
1473 \
1474 (fail_stack).stack == NULL \
1475 ? 0 \
1476 : ((fail_stack).size \
1477 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1478 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1479 * FAIL_STACK_GROWTH_FACTOR)) \
1480 / sizeof (fail_stack_elt_t)), \
1481 1)))
1482
1483
1484 /* Push a pointer value onto the failure stack.
1485 Assumes the variable `fail_stack'. Probably should only
1486 be called from within `PUSH_FAILURE_POINT'. */
1487 #define PUSH_FAILURE_POINTER(item) \
1488 fail_stack.stack[fail_stack.avail++].pointer = (item)
1489
1490 /* This pushes an integer-valued item onto the failure stack.
1491 Assumes the variable `fail_stack'. Probably should only
1492 be called from within `PUSH_FAILURE_POINT'. */
1493 #define PUSH_FAILURE_INT(item) \
1494 fail_stack.stack[fail_stack.avail++].integer = (item)
1495
1496 /* These POP... operations complement the PUSH... operations.
1497 All assume that `fail_stack' is nonempty. */
1498 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1499 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1500
1501 /* Individual items aside from the registers. */
1502 #define NUM_NONREG_ITEMS 3
1503
1504 /* Used to examine the stack (to detect infinite loops). */
1505 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1506 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1507 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1508 #define TOP_FAILURE_HANDLE() fail_stack.frame
1509
1510
1511 #define ENSURE_FAIL_STACK(space) \
1512 while (REMAINING_AVAIL_SLOTS <= space) { \
1513 if (!GROW_FAIL_STACK (fail_stack)) \
1514 return -2; \
1515 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1516 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1517 }
1518
1519 /* Push register NUM onto the stack. */
1520 #define PUSH_FAILURE_REG(num) \
1521 do { \
1522 char *destination; \
1523 ENSURE_FAIL_STACK(3); \
1524 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1525 num, regstart[num], regend[num]); \
1526 PUSH_FAILURE_POINTER (regstart[num]); \
1527 PUSH_FAILURE_POINTER (regend[num]); \
1528 PUSH_FAILURE_INT (num); \
1529 } while (0)
1530
1531 /* Change the counter's value to VAL, but make sure that it will
1532 be reset when backtracking. */
1533 #define PUSH_NUMBER(ptr,val) \
1534 do { \
1535 char *destination; \
1536 int c; \
1537 ENSURE_FAIL_STACK(3); \
1538 EXTRACT_NUMBER (c, ptr); \
1539 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1540 PUSH_FAILURE_INT (c); \
1541 PUSH_FAILURE_POINTER (ptr); \
1542 PUSH_FAILURE_INT (-1); \
1543 STORE_NUMBER (ptr, val); \
1544 } while (0)
1545
1546 /* Pop a saved register off the stack. */
1547 #define POP_FAILURE_REG_OR_COUNT() \
1548 do { \
1549 long pfreg = POP_FAILURE_INT (); \
1550 if (pfreg == -1) \
1551 { \
1552 /* It's a counter. */ \
1553 /* Here, we discard `const', making re_match non-reentrant. */ \
1554 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1555 pfreg = POP_FAILURE_INT (); \
1556 STORE_NUMBER (ptr, pfreg); \
1557 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, pfreg); \
1558 } \
1559 else \
1560 { \
1561 regend[pfreg] = POP_FAILURE_POINTER (); \
1562 regstart[pfreg] = POP_FAILURE_POINTER (); \
1563 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1564 pfreg, regstart[pfreg], regend[pfreg]); \
1565 } \
1566 } while (0)
1567
1568 /* Check that we are not stuck in an infinite loop. */
1569 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1570 do { \
1571 ssize_t failure = TOP_FAILURE_HANDLE (); \
1572 /* Check for infinite matching loops */ \
1573 while (failure > 0 \
1574 && (FAILURE_STR (failure) == string_place \
1575 || FAILURE_STR (failure) == NULL)) \
1576 { \
1577 assert (FAILURE_PAT (failure) >= bufp->buffer \
1578 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1579 if (FAILURE_PAT (failure) == pat_cur) \
1580 { \
1581 cycle = 1; \
1582 break; \
1583 } \
1584 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1585 failure = NEXT_FAILURE_HANDLE(failure); \
1586 } \
1587 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1588 } while (0)
1589
1590 /* Push the information about the state we will need
1591 if we ever fail back to it.
1592
1593 Requires variables fail_stack, regstart, regend and
1594 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1595 declared.
1596
1597 Does `return FAILURE_CODE' if runs out of memory. */
1598
1599 #define PUSH_FAILURE_POINT(pattern, string_place) \
1600 do { \
1601 char *destination; \
1602 /* Must be int, so when we don't save any registers, the arithmetic \
1603 of 0 + -1 isn't done as unsigned. */ \
1604 \
1605 DEBUG_STATEMENT (nfailure_points_pushed++); \
1606 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1607 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1608 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1609 \
1610 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1611 \
1612 DEBUG_PRINT1 ("\n"); \
1613 \
1614 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1615 PUSH_FAILURE_INT (fail_stack.frame); \
1616 \
1617 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1618 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1619 DEBUG_PRINT1 ("'\n"); \
1620 PUSH_FAILURE_POINTER (string_place); \
1621 \
1622 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1623 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1624 PUSH_FAILURE_POINTER (pattern); \
1625 \
1626 /* Close the frame by moving the frame pointer past it. */ \
1627 fail_stack.frame = fail_stack.avail; \
1628 } while (0)
1629
1630 /* Estimate the size of data pushed by a typical failure stack entry.
1631 An estimate is all we need, because all we use this for
1632 is to choose a limit for how big to make the failure stack. */
1633 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1634 #define TYPICAL_FAILURE_SIZE 20
1635
1636 /* How many items can still be added to the stack without overflowing it. */
1637 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1638
1639
1640 /* Pops what PUSH_FAIL_STACK pushes.
1641
1642 We restore into the parameters, all of which should be lvalues:
1643 STR -- the saved data position.
1644 PAT -- the saved pattern position.
1645 REGSTART, REGEND -- arrays of string positions.
1646
1647 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1648 `pend', `string1', `size1', `string2', and `size2'. */
1649
1650 #define POP_FAILURE_POINT(str, pat) \
1651 do { \
1652 assert (!FAIL_STACK_EMPTY ()); \
1653 \
1654 /* Remove failure points and point to how many regs pushed. */ \
1655 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1656 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1657 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1658 \
1659 /* Pop the saved registers. */ \
1660 while (fail_stack.frame < fail_stack.avail) \
1661 POP_FAILURE_REG_OR_COUNT (); \
1662 \
1663 pat = POP_FAILURE_POINTER (); \
1664 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1665 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1666 \
1667 /* If the saved string location is NULL, it came from an \
1668 on_failure_keep_string_jump opcode, and we want to throw away the \
1669 saved NULL, thus retaining our current position in the string. */ \
1670 str = POP_FAILURE_POINTER (); \
1671 DEBUG_PRINT2 (" Popping string %p: `", str); \
1672 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1673 DEBUG_PRINT1 ("'\n"); \
1674 \
1675 fail_stack.frame = POP_FAILURE_INT (); \
1676 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1677 \
1678 assert (fail_stack.avail >= 0); \
1679 assert (fail_stack.frame <= fail_stack.avail); \
1680 \
1681 DEBUG_STATEMENT (nfailure_points_popped++); \
1682 } while (0) /* POP_FAILURE_POINT */
1683
1684
1685 \f
1686 /* Registers are set to a sentinel when they haven't yet matched. */
1687 #define REG_UNSET(e) ((e) == NULL)
1688 \f
1689 /* Subroutine declarations and macros for regex_compile. */
1690
1691 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1692 reg_syntax_t syntax,
1693 struct re_pattern_buffer *bufp));
1694 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1695 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1696 int arg1, int arg2));
1697 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1698 int arg, unsigned char *end));
1699 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1700 int arg1, int arg2, unsigned char *end));
1701 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1702 re_char *p,
1703 reg_syntax_t syntax));
1704 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1705 re_char *pend,
1706 reg_syntax_t syntax));
1707 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1708 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1709 char *fastmap, const int multibyte));
1710
1711 /* Fetch the next character in the uncompiled pattern, with no
1712 translation. */
1713 #define PATFETCH(c) \
1714 do { \
1715 int len; \
1716 if (p == pend) return REG_EEND; \
1717 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1718 p += len; \
1719 } while (0)
1720
1721
1722 /* If `translate' is non-null, return translate[D], else just D. We
1723 cast the subscript to translate because some data is declared as
1724 `char *', to avoid warnings when a string constant is passed. But
1725 when we use a character as a subscript we must make it unsigned. */
1726 #ifndef TRANSLATE
1727 # define TRANSLATE(d) \
1728 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1729 #endif
1730
1731
1732 /* Macros for outputting the compiled pattern into `buffer'. */
1733
1734 /* If the buffer isn't allocated when it comes in, use this. */
1735 #define INIT_BUF_SIZE 32
1736
1737 /* Make sure we have at least N more bytes of space in buffer. */
1738 #define GET_BUFFER_SPACE(n) \
1739 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1740 EXTEND_BUFFER ()
1741
1742 /* Make sure we have one more byte of buffer space and then add C to it. */
1743 #define BUF_PUSH(c) \
1744 do { \
1745 GET_BUFFER_SPACE (1); \
1746 *b++ = (unsigned char) (c); \
1747 } while (0)
1748
1749
1750 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1751 #define BUF_PUSH_2(c1, c2) \
1752 do { \
1753 GET_BUFFER_SPACE (2); \
1754 *b++ = (unsigned char) (c1); \
1755 *b++ = (unsigned char) (c2); \
1756 } while (0)
1757
1758
1759 /* Store a jump with opcode OP at LOC to location TO. We store a
1760 relative address offset by the three bytes the jump itself occupies. */
1761 #define STORE_JUMP(op, loc, to) \
1762 store_op1 (op, loc, (to) - (loc) - 3)
1763
1764 /* Likewise, for a two-argument jump. */
1765 #define STORE_JUMP2(op, loc, to, arg) \
1766 store_op2 (op, loc, (to) - (loc) - 3, arg)
1767
1768 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1769 #define INSERT_JUMP(op, loc, to) \
1770 insert_op1 (op, loc, (to) - (loc) - 3, b)
1771
1772 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1773 #define INSERT_JUMP2(op, loc, to, arg) \
1774 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1775
1776
1777 /* This is not an arbitrary limit: the arguments which represent offsets
1778 into the pattern are two bytes long. So if 2^15 bytes turns out to
1779 be too small, many things would have to change. */
1780 # define MAX_BUF_SIZE (1L << 15)
1781
1782 #if 0 /* This is when we thought it could be 2^16 bytes. */
1783 /* Any other compiler which, like MSC, has allocation limit below 2^16
1784 bytes will have to use approach similar to what was done below for
1785 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1786 reallocating to 0 bytes. Such thing is not going to work too well.
1787 You have been warned!! */
1788 #if defined _MSC_VER && !defined WIN32
1789 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1790 # define MAX_BUF_SIZE 65500L
1791 #else
1792 # define MAX_BUF_SIZE (1L << 16)
1793 #endif
1794 #endif /* 0 */
1795
1796 /* Extend the buffer by twice its current size via realloc and
1797 reset the pointers that pointed into the old block to point to the
1798 correct places in the new one. If extending the buffer results in it
1799 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1800 #if __BOUNDED_POINTERS__
1801 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1802 # define MOVE_BUFFER_POINTER(P) \
1803 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1804 SET_HIGH_BOUND (P), \
1805 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1806 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1807 else \
1808 { \
1809 SET_HIGH_BOUND (b); \
1810 SET_HIGH_BOUND (begalt); \
1811 if (fixup_alt_jump) \
1812 SET_HIGH_BOUND (fixup_alt_jump); \
1813 if (laststart) \
1814 SET_HIGH_BOUND (laststart); \
1815 if (pending_exact) \
1816 SET_HIGH_BOUND (pending_exact); \
1817 }
1818 #else
1819 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1820 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1821 #endif
1822 #define EXTEND_BUFFER() \
1823 do { \
1824 unsigned char *old_buffer = bufp->buffer; \
1825 if (bufp->allocated == MAX_BUF_SIZE) \
1826 return REG_ESIZE; \
1827 bufp->allocated <<= 1; \
1828 if (bufp->allocated > MAX_BUF_SIZE) \
1829 bufp->allocated = MAX_BUF_SIZE; \
1830 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1831 if (bufp->buffer == NULL) \
1832 return REG_ESPACE; \
1833 /* If the buffer moved, move all the pointers into it. */ \
1834 if (old_buffer != bufp->buffer) \
1835 { \
1836 unsigned char *new_buffer = bufp->buffer; \
1837 MOVE_BUFFER_POINTER (b); \
1838 MOVE_BUFFER_POINTER (begalt); \
1839 if (fixup_alt_jump) \
1840 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1841 if (laststart) \
1842 MOVE_BUFFER_POINTER (laststart); \
1843 if (pending_exact) \
1844 MOVE_BUFFER_POINTER (pending_exact); \
1845 } \
1846 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1847 } while (0)
1848
1849
1850 /* Since we have one byte reserved for the register number argument to
1851 {start,stop}_memory, the maximum number of groups we can report
1852 things about is what fits in that byte. */
1853 #define MAX_REGNUM 255
1854
1855 /* But patterns can have more than `MAX_REGNUM' registers. We just
1856 ignore the excess. */
1857 typedef int regnum_t;
1858
1859
1860 /* Macros for the compile stack. */
1861
1862 /* Since offsets can go either forwards or backwards, this type needs to
1863 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1864 /* int may be not enough when sizeof(int) == 2. */
1865 typedef long pattern_offset_t;
1866
1867 typedef struct
1868 {
1869 pattern_offset_t begalt_offset;
1870 pattern_offset_t fixup_alt_jump;
1871 pattern_offset_t laststart_offset;
1872 regnum_t regnum;
1873 } compile_stack_elt_t;
1874
1875
1876 typedef struct
1877 {
1878 compile_stack_elt_t *stack;
1879 size_t size;
1880 size_t avail; /* Offset of next open position. */
1881 } compile_stack_type;
1882
1883
1884 #define INIT_COMPILE_STACK_SIZE 32
1885
1886 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1887 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1888
1889 /* The next available element. */
1890 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1891
1892 /* Explicit quit checking is only used on NTemacs and whenever we
1893 use polling to process input events. */
1894 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1895 extern int immediate_quit;
1896 # define IMMEDIATE_QUIT_CHECK \
1897 do { \
1898 if (immediate_quit) QUIT; \
1899 } while (0)
1900 #else
1901 # define IMMEDIATE_QUIT_CHECK ((void)0)
1902 #endif
1903 \f
1904 /* Structure to manage work area for range table. */
1905 struct range_table_work_area
1906 {
1907 int *table; /* actual work area. */
1908 int allocated; /* allocated size for work area in bytes. */
1909 int used; /* actually used size in words. */
1910 int bits; /* flag to record character classes */
1911 };
1912
1913 /* Make sure that WORK_AREA can hold more N multibyte characters.
1914 This is used only in set_image_of_range and set_image_of_range_1.
1915 It expects WORK_AREA to be a pointer.
1916 If it can't get the space, it returns from the surrounding function. */
1917
1918 #define EXTEND_RANGE_TABLE(work_area, n) \
1919 do { \
1920 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1921 { \
1922 extend_range_table_work_area (&work_area); \
1923 if ((work_area).table == 0) \
1924 return (REG_ESPACE); \
1925 } \
1926 } while (0)
1927
1928 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1929 (work_area).bits |= (bit)
1930
1931 /* Bits used to implement the multibyte-part of the various character classes
1932 such as [:alnum:] in a charset's range table. */
1933 #define BIT_WORD 0x1
1934 #define BIT_LOWER 0x2
1935 #define BIT_PUNCT 0x4
1936 #define BIT_SPACE 0x8
1937 #define BIT_UPPER 0x10
1938 #define BIT_MULTIBYTE 0x20
1939
1940 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1941 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1942 do { \
1943 EXTEND_RANGE_TABLE ((work_area), 2); \
1944 (work_area).table[(work_area).used++] = (range_start); \
1945 (work_area).table[(work_area).used++] = (range_end); \
1946 } while (0)
1947
1948 /* Free allocated memory for WORK_AREA. */
1949 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1950 do { \
1951 if ((work_area).table) \
1952 free ((work_area).table); \
1953 } while (0)
1954
1955 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1956 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1957 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1958 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1959 \f
1960
1961 /* Set the bit for character C in a list. */
1962 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1963
1964
1965 #ifdef emacs
1966
1967 /* Store characters in the range FROM to TO in the bitmap at B (for
1968 ASCII and unibyte characters) and WORK_AREA (for multibyte
1969 characters) while translating them and paying attention to the
1970 continuity of translated characters.
1971
1972 Implementation note: It is better to implement these fairly big
1973 macros by a function, but it's not that easy because macros called
1974 in this macro assume various local variables already declared. */
1975
1976 /* Both FROM and TO are ASCII characters. */
1977
1978 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1979 do { \
1980 int C0, C1; \
1981 \
1982 for (C0 = (FROM); C0 <= (TO); C0++) \
1983 { \
1984 C1 = TRANSLATE (C0); \
1985 if (! ASCII_CHAR_P (C1)) \
1986 { \
1987 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1988 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1989 C1 = C0; \
1990 } \
1991 SET_LIST_BIT (C1); \
1992 } \
1993 } while (0)
1994
1995
1996 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1997
1998 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1999 do { \
2000 int C0, C1, C2, I; \
2001 int USED = RANGE_TABLE_WORK_USED (work_area); \
2002 \
2003 for (C0 = (FROM); C0 <= (TO); C0++) \
2004 { \
2005 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2006 if (CHAR_BYTE8_P (C1)) \
2007 SET_LIST_BIT (C0); \
2008 else \
2009 { \
2010 C2 = TRANSLATE (C1); \
2011 if (C2 == C1 \
2012 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2013 C1 = C0; \
2014 SET_LIST_BIT (C1); \
2015 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2016 { \
2017 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2018 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2019 \
2020 if (C2 >= from - 1 && C2 <= to + 1) \
2021 { \
2022 if (C2 == from - 1) \
2023 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2024 else if (C2 == to + 1) \
2025 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2026 break; \
2027 } \
2028 } \
2029 if (I < USED) \
2030 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2031 } \
2032 } \
2033 } while (0)
2034
2035
2036 /* Both FROM and TO are multibyte characters. */
2037
2038 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2039 do { \
2040 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2041 \
2042 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2043 for (C0 = (FROM); C0 <= (TO); C0++) \
2044 { \
2045 C1 = TRANSLATE (C0); \
2046 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2047 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2048 SET_LIST_BIT (C2); \
2049 if (C1 >= (FROM) && C1 <= (TO)) \
2050 continue; \
2051 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2052 { \
2053 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2054 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2055 \
2056 if (C1 >= from - 1 && C1 <= to + 1) \
2057 { \
2058 if (C1 == from - 1) \
2059 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2060 else if (C1 == to + 1) \
2061 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2062 break; \
2063 } \
2064 } \
2065 if (I < USED) \
2066 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2067 } \
2068 } while (0)
2069
2070 #endif /* emacs */
2071
2072 /* Get the next unsigned number in the uncompiled pattern. */
2073 #define GET_UNSIGNED_NUMBER(num) \
2074 do { \
2075 if (p == pend) \
2076 FREE_STACK_RETURN (REG_EBRACE); \
2077 else \
2078 { \
2079 PATFETCH (c); \
2080 while ('0' <= c && c <= '9') \
2081 { \
2082 int prev; \
2083 if (num < 0) \
2084 num = 0; \
2085 prev = num; \
2086 num = num * 10 + c - '0'; \
2087 if (num / 10 != prev) \
2088 FREE_STACK_RETURN (REG_BADBR); \
2089 if (p == pend) \
2090 FREE_STACK_RETURN (REG_EBRACE); \
2091 PATFETCH (c); \
2092 } \
2093 } \
2094 } while (0)
2095 \f
2096 #if ! WIDE_CHAR_SUPPORT
2097
2098 /* Map a string to the char class it names (if any). */
2099 re_wctype_t
2100 re_wctype (const re_char *str)
2101 {
2102 const char *string = (const char *) str;
2103 if (STREQ (string, "alnum")) return RECC_ALNUM;
2104 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2105 else if (STREQ (string, "word")) return RECC_WORD;
2106 else if (STREQ (string, "ascii")) return RECC_ASCII;
2107 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2108 else if (STREQ (string, "graph")) return RECC_GRAPH;
2109 else if (STREQ (string, "lower")) return RECC_LOWER;
2110 else if (STREQ (string, "print")) return RECC_PRINT;
2111 else if (STREQ (string, "punct")) return RECC_PUNCT;
2112 else if (STREQ (string, "space")) return RECC_SPACE;
2113 else if (STREQ (string, "upper")) return RECC_UPPER;
2114 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2115 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2116 else if (STREQ (string, "digit")) return RECC_DIGIT;
2117 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2118 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2119 else if (STREQ (string, "blank")) return RECC_BLANK;
2120 else return 0;
2121 }
2122
2123 /* True if CH is in the char class CC. */
2124 boolean
2125 re_iswctype (int ch, re_wctype_t cc)
2126 {
2127 switch (cc)
2128 {
2129 case RECC_ALNUM: return ISALNUM (ch);
2130 case RECC_ALPHA: return ISALPHA (ch);
2131 case RECC_BLANK: return ISBLANK (ch);
2132 case RECC_CNTRL: return ISCNTRL (ch);
2133 case RECC_DIGIT: return ISDIGIT (ch);
2134 case RECC_GRAPH: return ISGRAPH (ch);
2135 case RECC_LOWER: return ISLOWER (ch);
2136 case RECC_PRINT: return ISPRINT (ch);
2137 case RECC_PUNCT: return ISPUNCT (ch);
2138 case RECC_SPACE: return ISSPACE (ch);
2139 case RECC_UPPER: return ISUPPER (ch);
2140 case RECC_XDIGIT: return ISXDIGIT (ch);
2141 case RECC_ASCII: return IS_REAL_ASCII (ch);
2142 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2143 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2144 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2145 case RECC_WORD: return ISWORD (ch);
2146 case RECC_ERROR: return false;
2147 default:
2148 abort();
2149 }
2150 }
2151
2152 /* Return a bit-pattern to use in the range-table bits to match multibyte
2153 chars of class CC. */
2154 static int
2155 re_wctype_to_bit (re_wctype_t cc)
2156 {
2157 switch (cc)
2158 {
2159 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2160 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2161 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2162 case RECC_LOWER: return BIT_LOWER;
2163 case RECC_UPPER: return BIT_UPPER;
2164 case RECC_PUNCT: return BIT_PUNCT;
2165 case RECC_SPACE: return BIT_SPACE;
2166 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2167 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2168 default:
2169 abort();
2170 }
2171 }
2172 #endif
2173 \f
2174 /* Filling in the work area of a range. */
2175
2176 /* Actually extend the space in WORK_AREA. */
2177
2178 static void
2179 extend_range_table_work_area (struct range_table_work_area *work_area)
2180 {
2181 work_area->allocated += 16 * sizeof (int);
2182 if (work_area->table)
2183 work_area->table
2184 = (int *) realloc (work_area->table, work_area->allocated);
2185 else
2186 work_area->table
2187 = (int *) malloc (work_area->allocated);
2188 }
2189
2190 #if 0
2191 #ifdef emacs
2192
2193 /* Carefully find the ranges of codes that are equivalent
2194 under case conversion to the range start..end when passed through
2195 TRANSLATE. Handle the case where non-letters can come in between
2196 two upper-case letters (which happens in Latin-1).
2197 Also handle the case of groups of more than 2 case-equivalent chars.
2198
2199 The basic method is to look at consecutive characters and see
2200 if they can form a run that can be handled as one.
2201
2202 Returns -1 if successful, REG_ESPACE if ran out of space. */
2203
2204 static int
2205 set_image_of_range_1 (work_area, start, end, translate)
2206 RE_TRANSLATE_TYPE translate;
2207 struct range_table_work_area *work_area;
2208 re_wchar_t start, end;
2209 {
2210 /* `one_case' indicates a character, or a run of characters,
2211 each of which is an isolate (no case-equivalents).
2212 This includes all ASCII non-letters.
2213
2214 `two_case' indicates a character, or a run of characters,
2215 each of which has two case-equivalent forms.
2216 This includes all ASCII letters.
2217
2218 `strange' indicates a character that has more than one
2219 case-equivalent. */
2220
2221 enum case_type {one_case, two_case, strange};
2222
2223 /* Describe the run that is in progress,
2224 which the next character can try to extend.
2225 If run_type is strange, that means there really is no run.
2226 If run_type is one_case, then run_start...run_end is the run.
2227 If run_type is two_case, then the run is run_start...run_end,
2228 and the case-equivalents end at run_eqv_end. */
2229
2230 enum case_type run_type = strange;
2231 int run_start, run_end, run_eqv_end;
2232
2233 Lisp_Object eqv_table;
2234
2235 if (!RE_TRANSLATE_P (translate))
2236 {
2237 EXTEND_RANGE_TABLE (work_area, 2);
2238 work_area->table[work_area->used++] = (start);
2239 work_area->table[work_area->used++] = (end);
2240 return -1;
2241 }
2242
2243 eqv_table = XCHAR_TABLE (translate)->extras[2];
2244
2245 for (; start <= end; start++)
2246 {
2247 enum case_type this_type;
2248 int eqv = RE_TRANSLATE (eqv_table, start);
2249 int minchar, maxchar;
2250
2251 /* Classify this character */
2252 if (eqv == start)
2253 this_type = one_case;
2254 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2255 this_type = two_case;
2256 else
2257 this_type = strange;
2258
2259 if (start < eqv)
2260 minchar = start, maxchar = eqv;
2261 else
2262 minchar = eqv, maxchar = start;
2263
2264 /* Can this character extend the run in progress? */
2265 if (this_type == strange || this_type != run_type
2266 || !(minchar == run_end + 1
2267 && (run_type == two_case
2268 ? maxchar == run_eqv_end + 1 : 1)))
2269 {
2270 /* No, end the run.
2271 Record each of its equivalent ranges. */
2272 if (run_type == one_case)
2273 {
2274 EXTEND_RANGE_TABLE (work_area, 2);
2275 work_area->table[work_area->used++] = run_start;
2276 work_area->table[work_area->used++] = run_end;
2277 }
2278 else if (run_type == two_case)
2279 {
2280 EXTEND_RANGE_TABLE (work_area, 4);
2281 work_area->table[work_area->used++] = run_start;
2282 work_area->table[work_area->used++] = run_end;
2283 work_area->table[work_area->used++]
2284 = RE_TRANSLATE (eqv_table, run_start);
2285 work_area->table[work_area->used++]
2286 = RE_TRANSLATE (eqv_table, run_end);
2287 }
2288 run_type = strange;
2289 }
2290
2291 if (this_type == strange)
2292 {
2293 /* For a strange character, add each of its equivalents, one
2294 by one. Don't start a range. */
2295 do
2296 {
2297 EXTEND_RANGE_TABLE (work_area, 2);
2298 work_area->table[work_area->used++] = eqv;
2299 work_area->table[work_area->used++] = eqv;
2300 eqv = RE_TRANSLATE (eqv_table, eqv);
2301 }
2302 while (eqv != start);
2303 }
2304
2305 /* Add this char to the run, or start a new run. */
2306 else if (run_type == strange)
2307 {
2308 /* Initialize a new range. */
2309 run_type = this_type;
2310 run_start = start;
2311 run_end = start;
2312 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2313 }
2314 else
2315 {
2316 /* Extend a running range. */
2317 run_end = minchar;
2318 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2319 }
2320 }
2321
2322 /* If a run is still in progress at the end, finish it now
2323 by recording its equivalent ranges. */
2324 if (run_type == one_case)
2325 {
2326 EXTEND_RANGE_TABLE (work_area, 2);
2327 work_area->table[work_area->used++] = run_start;
2328 work_area->table[work_area->used++] = run_end;
2329 }
2330 else if (run_type == two_case)
2331 {
2332 EXTEND_RANGE_TABLE (work_area, 4);
2333 work_area->table[work_area->used++] = run_start;
2334 work_area->table[work_area->used++] = run_end;
2335 work_area->table[work_area->used++]
2336 = RE_TRANSLATE (eqv_table, run_start);
2337 work_area->table[work_area->used++]
2338 = RE_TRANSLATE (eqv_table, run_end);
2339 }
2340
2341 return -1;
2342 }
2343
2344 #endif /* emacs */
2345
2346 /* Record the image of the range start..end when passed through
2347 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2348 and is not even necessarily contiguous.
2349 Normally we approximate it with the smallest contiguous range that contains
2350 all the chars we need. However, for Latin-1 we go to extra effort
2351 to do a better job.
2352
2353 This function is not called for ASCII ranges.
2354
2355 Returns -1 if successful, REG_ESPACE if ran out of space. */
2356
2357 static int
2358 set_image_of_range (work_area, start, end, translate)
2359 RE_TRANSLATE_TYPE translate;
2360 struct range_table_work_area *work_area;
2361 re_wchar_t start, end;
2362 {
2363 re_wchar_t cmin, cmax;
2364
2365 #ifdef emacs
2366 /* For Latin-1 ranges, use set_image_of_range_1
2367 to get proper handling of ranges that include letters and nonletters.
2368 For a range that includes the whole of Latin-1, this is not necessary.
2369 For other character sets, we don't bother to get this right. */
2370 if (RE_TRANSLATE_P (translate) && start < 04400
2371 && !(start < 04200 && end >= 04377))
2372 {
2373 int newend;
2374 int tem;
2375 newend = end;
2376 if (newend > 04377)
2377 newend = 04377;
2378 tem = set_image_of_range_1 (work_area, start, newend, translate);
2379 if (tem > 0)
2380 return tem;
2381
2382 start = 04400;
2383 if (end < 04400)
2384 return -1;
2385 }
2386 #endif
2387
2388 EXTEND_RANGE_TABLE (work_area, 2);
2389 work_area->table[work_area->used++] = (start);
2390 work_area->table[work_area->used++] = (end);
2391
2392 cmin = -1, cmax = -1;
2393
2394 if (RE_TRANSLATE_P (translate))
2395 {
2396 int ch;
2397
2398 for (ch = start; ch <= end; ch++)
2399 {
2400 re_wchar_t c = TRANSLATE (ch);
2401 if (! (start <= c && c <= end))
2402 {
2403 if (cmin == -1)
2404 cmin = c, cmax = c;
2405 else
2406 {
2407 cmin = MIN (cmin, c);
2408 cmax = MAX (cmax, c);
2409 }
2410 }
2411 }
2412
2413 if (cmin != -1)
2414 {
2415 EXTEND_RANGE_TABLE (work_area, 2);
2416 work_area->table[work_area->used++] = (cmin);
2417 work_area->table[work_area->used++] = (cmax);
2418 }
2419 }
2420
2421 return -1;
2422 }
2423 #endif /* 0 */
2424 \f
2425 #ifndef MATCH_MAY_ALLOCATE
2426
2427 /* If we cannot allocate large objects within re_match_2_internal,
2428 we make the fail stack and register vectors global.
2429 The fail stack, we grow to the maximum size when a regexp
2430 is compiled.
2431 The register vectors, we adjust in size each time we
2432 compile a regexp, according to the number of registers it needs. */
2433
2434 static fail_stack_type fail_stack;
2435
2436 /* Size with which the following vectors are currently allocated.
2437 That is so we can make them bigger as needed,
2438 but never make them smaller. */
2439 static int regs_allocated_size;
2440
2441 static re_char ** regstart, ** regend;
2442 static re_char **best_regstart, **best_regend;
2443
2444 /* Make the register vectors big enough for NUM_REGS registers,
2445 but don't make them smaller. */
2446
2447 static
2448 regex_grow_registers (num_regs)
2449 int num_regs;
2450 {
2451 if (num_regs > regs_allocated_size)
2452 {
2453 RETALLOC_IF (regstart, num_regs, re_char *);
2454 RETALLOC_IF (regend, num_regs, re_char *);
2455 RETALLOC_IF (best_regstart, num_regs, re_char *);
2456 RETALLOC_IF (best_regend, num_regs, re_char *);
2457
2458 regs_allocated_size = num_regs;
2459 }
2460 }
2461
2462 #endif /* not MATCH_MAY_ALLOCATE */
2463 \f
2464 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2465 compile_stack,
2466 regnum_t regnum));
2467
2468 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2469 Returns one of error codes defined in `regex.h', or zero for success.
2470
2471 Assumes the `allocated' (and perhaps `buffer') and `translate'
2472 fields are set in BUFP on entry.
2473
2474 If it succeeds, results are put in BUFP (if it returns an error, the
2475 contents of BUFP are undefined):
2476 `buffer' is the compiled pattern;
2477 `syntax' is set to SYNTAX;
2478 `used' is set to the length of the compiled pattern;
2479 `fastmap_accurate' is zero;
2480 `re_nsub' is the number of subexpressions in PATTERN;
2481 `not_bol' and `not_eol' are zero;
2482
2483 The `fastmap' field is neither examined nor set. */
2484
2485 /* Insert the `jump' from the end of last alternative to "here".
2486 The space for the jump has already been allocated. */
2487 #define FIXUP_ALT_JUMP() \
2488 do { \
2489 if (fixup_alt_jump) \
2490 STORE_JUMP (jump, fixup_alt_jump, b); \
2491 } while (0)
2492
2493
2494 /* Return, freeing storage we allocated. */
2495 #define FREE_STACK_RETURN(value) \
2496 do { \
2497 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2498 free (compile_stack.stack); \
2499 return value; \
2500 } while (0)
2501
2502 static reg_errcode_t
2503 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2504 {
2505 /* We fetch characters from PATTERN here. */
2506 register re_wchar_t c, c1;
2507
2508 /* Points to the end of the buffer, where we should append. */
2509 register unsigned char *b;
2510
2511 /* Keeps track of unclosed groups. */
2512 compile_stack_type compile_stack;
2513
2514 /* Points to the current (ending) position in the pattern. */
2515 #ifdef AIX
2516 /* `const' makes AIX compiler fail. */
2517 unsigned char *p = pattern;
2518 #else
2519 re_char *p = pattern;
2520 #endif
2521 re_char *pend = pattern + size;
2522
2523 /* How to translate the characters in the pattern. */
2524 RE_TRANSLATE_TYPE translate = bufp->translate;
2525
2526 /* Address of the count-byte of the most recently inserted `exactn'
2527 command. This makes it possible to tell if a new exact-match
2528 character can be added to that command or if the character requires
2529 a new `exactn' command. */
2530 unsigned char *pending_exact = 0;
2531
2532 /* Address of start of the most recently finished expression.
2533 This tells, e.g., postfix * where to find the start of its
2534 operand. Reset at the beginning of groups and alternatives. */
2535 unsigned char *laststart = 0;
2536
2537 /* Address of beginning of regexp, or inside of last group. */
2538 unsigned char *begalt;
2539
2540 /* Place in the uncompiled pattern (i.e., the {) to
2541 which to go back if the interval is invalid. */
2542 re_char *beg_interval;
2543
2544 /* Address of the place where a forward jump should go to the end of
2545 the containing expression. Each alternative of an `or' -- except the
2546 last -- ends with a forward jump of this sort. */
2547 unsigned char *fixup_alt_jump = 0;
2548
2549 /* Work area for range table of charset. */
2550 struct range_table_work_area range_table_work;
2551
2552 /* If the object matched can contain multibyte characters. */
2553 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2554
2555 /* Nonzero if we have pushed down into a subpattern. */
2556 int in_subpattern = 0;
2557
2558 /* These hold the values of p, pattern, and pend from the main
2559 pattern when we have pushed into a subpattern. */
2560 re_char *main_p IF_LINT (= NULL);
2561 re_char *main_pattern IF_LINT (= NULL);
2562 re_char *main_pend IF_LINT (= NULL);
2563
2564 #ifdef DEBUG
2565 debug++;
2566 DEBUG_PRINT1 ("\nCompiling pattern: ");
2567 if (debug > 0)
2568 {
2569 unsigned debug_count;
2570
2571 for (debug_count = 0; debug_count < size; debug_count++)
2572 putchar (pattern[debug_count]);
2573 putchar ('\n');
2574 }
2575 #endif /* DEBUG */
2576
2577 /* Initialize the compile stack. */
2578 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2579 if (compile_stack.stack == NULL)
2580 return REG_ESPACE;
2581
2582 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2583 compile_stack.avail = 0;
2584
2585 range_table_work.table = 0;
2586 range_table_work.allocated = 0;
2587
2588 /* Initialize the pattern buffer. */
2589 bufp->syntax = syntax;
2590 bufp->fastmap_accurate = 0;
2591 bufp->not_bol = bufp->not_eol = 0;
2592 bufp->used_syntax = 0;
2593
2594 /* Set `used' to zero, so that if we return an error, the pattern
2595 printer (for debugging) will think there's no pattern. We reset it
2596 at the end. */
2597 bufp->used = 0;
2598
2599 /* Always count groups, whether or not bufp->no_sub is set. */
2600 bufp->re_nsub = 0;
2601
2602 #if !defined emacs && !defined SYNTAX_TABLE
2603 /* Initialize the syntax table. */
2604 init_syntax_once ();
2605 #endif
2606
2607 if (bufp->allocated == 0)
2608 {
2609 if (bufp->buffer)
2610 { /* If zero allocated, but buffer is non-null, try to realloc
2611 enough space. This loses if buffer's address is bogus, but
2612 that is the user's responsibility. */
2613 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2614 }
2615 else
2616 { /* Caller did not allocate a buffer. Do it for them. */
2617 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2618 }
2619 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2620
2621 bufp->allocated = INIT_BUF_SIZE;
2622 }
2623
2624 begalt = b = bufp->buffer;
2625
2626 /* Loop through the uncompiled pattern until we're at the end. */
2627 while (1)
2628 {
2629 if (p == pend)
2630 {
2631 /* If this is the end of an included regexp,
2632 pop back to the main regexp and try again. */
2633 if (in_subpattern)
2634 {
2635 in_subpattern = 0;
2636 pattern = main_pattern;
2637 p = main_p;
2638 pend = main_pend;
2639 continue;
2640 }
2641 /* If this is the end of the main regexp, we are done. */
2642 break;
2643 }
2644
2645 PATFETCH (c);
2646
2647 switch (c)
2648 {
2649 case ' ':
2650 {
2651 re_char *p1 = p;
2652
2653 /* If there's no special whitespace regexp, treat
2654 spaces normally. And don't try to do this recursively. */
2655 if (!whitespace_regexp || in_subpattern)
2656 goto normal_char;
2657
2658 /* Peek past following spaces. */
2659 while (p1 != pend)
2660 {
2661 if (*p1 != ' ')
2662 break;
2663 p1++;
2664 }
2665 /* If the spaces are followed by a repetition op,
2666 treat them normally. */
2667 if (p1 != pend
2668 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2669 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2670 goto normal_char;
2671
2672 /* Replace the spaces with the whitespace regexp. */
2673 in_subpattern = 1;
2674 main_p = p1;
2675 main_pend = pend;
2676 main_pattern = pattern;
2677 p = pattern = whitespace_regexp;
2678 pend = p + strlen ((const char *) p);
2679 break;
2680 }
2681
2682 case '^':
2683 {
2684 if ( /* If at start of pattern, it's an operator. */
2685 p == pattern + 1
2686 /* If context independent, it's an operator. */
2687 || syntax & RE_CONTEXT_INDEP_ANCHORS
2688 /* Otherwise, depends on what's come before. */
2689 || at_begline_loc_p (pattern, p, syntax))
2690 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2691 else
2692 goto normal_char;
2693 }
2694 break;
2695
2696
2697 case '$':
2698 {
2699 if ( /* If at end of pattern, it's an operator. */
2700 p == pend
2701 /* If context independent, it's an operator. */
2702 || syntax & RE_CONTEXT_INDEP_ANCHORS
2703 /* Otherwise, depends on what's next. */
2704 || at_endline_loc_p (p, pend, syntax))
2705 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2706 else
2707 goto normal_char;
2708 }
2709 break;
2710
2711
2712 case '+':
2713 case '?':
2714 if ((syntax & RE_BK_PLUS_QM)
2715 || (syntax & RE_LIMITED_OPS))
2716 goto normal_char;
2717 handle_plus:
2718 case '*':
2719 /* If there is no previous pattern... */
2720 if (!laststart)
2721 {
2722 if (syntax & RE_CONTEXT_INVALID_OPS)
2723 FREE_STACK_RETURN (REG_BADRPT);
2724 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2725 goto normal_char;
2726 }
2727
2728 {
2729 /* 1 means zero (many) matches is allowed. */
2730 boolean zero_times_ok = 0, many_times_ok = 0;
2731 boolean greedy = 1;
2732
2733 /* If there is a sequence of repetition chars, collapse it
2734 down to just one (the right one). We can't combine
2735 interval operators with these because of, e.g., `a{2}*',
2736 which should only match an even number of `a's. */
2737
2738 for (;;)
2739 {
2740 if ((syntax & RE_FRUGAL)
2741 && c == '?' && (zero_times_ok || many_times_ok))
2742 greedy = 0;
2743 else
2744 {
2745 zero_times_ok |= c != '+';
2746 many_times_ok |= c != '?';
2747 }
2748
2749 if (p == pend)
2750 break;
2751 else if (*p == '*'
2752 || (!(syntax & RE_BK_PLUS_QM)
2753 && (*p == '+' || *p == '?')))
2754 ;
2755 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2756 {
2757 if (p+1 == pend)
2758 FREE_STACK_RETURN (REG_EESCAPE);
2759 if (p[1] == '+' || p[1] == '?')
2760 PATFETCH (c); /* Gobble up the backslash. */
2761 else
2762 break;
2763 }
2764 else
2765 break;
2766 /* If we get here, we found another repeat character. */
2767 PATFETCH (c);
2768 }
2769
2770 /* Star, etc. applied to an empty pattern is equivalent
2771 to an empty pattern. */
2772 if (!laststart || laststart == b)
2773 break;
2774
2775 /* Now we know whether or not zero matches is allowed
2776 and also whether or not two or more matches is allowed. */
2777 if (greedy)
2778 {
2779 if (many_times_ok)
2780 {
2781 boolean simple = skip_one_char (laststart) == b;
2782 size_t startoffset = 0;
2783 re_opcode_t ofj =
2784 /* Check if the loop can match the empty string. */
2785 (simple || !analyse_first (laststart, b, NULL, 0))
2786 ? on_failure_jump : on_failure_jump_loop;
2787 assert (skip_one_char (laststart) <= b);
2788
2789 if (!zero_times_ok && simple)
2790 { /* Since simple * loops can be made faster by using
2791 on_failure_keep_string_jump, we turn simple P+
2792 into PP* if P is simple. */
2793 unsigned char *p1, *p2;
2794 startoffset = b - laststart;
2795 GET_BUFFER_SPACE (startoffset);
2796 p1 = b; p2 = laststart;
2797 while (p2 < p1)
2798 *b++ = *p2++;
2799 zero_times_ok = 1;
2800 }
2801
2802 GET_BUFFER_SPACE (6);
2803 if (!zero_times_ok)
2804 /* A + loop. */
2805 STORE_JUMP (ofj, b, b + 6);
2806 else
2807 /* Simple * loops can use on_failure_keep_string_jump
2808 depending on what follows. But since we don't know
2809 that yet, we leave the decision up to
2810 on_failure_jump_smart. */
2811 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2812 laststart + startoffset, b + 6);
2813 b += 3;
2814 STORE_JUMP (jump, b, laststart + startoffset);
2815 b += 3;
2816 }
2817 else
2818 {
2819 /* A simple ? pattern. */
2820 assert (zero_times_ok);
2821 GET_BUFFER_SPACE (3);
2822 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2823 b += 3;
2824 }
2825 }
2826 else /* not greedy */
2827 { /* I wish the greedy and non-greedy cases could be merged. */
2828
2829 GET_BUFFER_SPACE (7); /* We might use less. */
2830 if (many_times_ok)
2831 {
2832 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2833
2834 /* The non-greedy multiple match looks like
2835 a repeat..until: we only need a conditional jump
2836 at the end of the loop. */
2837 if (emptyp) BUF_PUSH (no_op);
2838 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2839 : on_failure_jump, b, laststart);
2840 b += 3;
2841 if (zero_times_ok)
2842 {
2843 /* The repeat...until naturally matches one or more.
2844 To also match zero times, we need to first jump to
2845 the end of the loop (its conditional jump). */
2846 INSERT_JUMP (jump, laststart, b);
2847 b += 3;
2848 }
2849 }
2850 else
2851 {
2852 /* non-greedy a?? */
2853 INSERT_JUMP (jump, laststart, b + 3);
2854 b += 3;
2855 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2856 b += 3;
2857 }
2858 }
2859 }
2860 pending_exact = 0;
2861 break;
2862
2863
2864 case '.':
2865 laststart = b;
2866 BUF_PUSH (anychar);
2867 break;
2868
2869
2870 case '[':
2871 {
2872 re_char *p1;
2873
2874 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2875
2876 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2877
2878 /* Ensure that we have enough space to push a charset: the
2879 opcode, the length count, and the bitset; 34 bytes in all. */
2880 GET_BUFFER_SPACE (34);
2881
2882 laststart = b;
2883
2884 /* We test `*p == '^' twice, instead of using an if
2885 statement, so we only need one BUF_PUSH. */
2886 BUF_PUSH (*p == '^' ? charset_not : charset);
2887 if (*p == '^')
2888 p++;
2889
2890 /* Remember the first position in the bracket expression. */
2891 p1 = p;
2892
2893 /* Push the number of bytes in the bitmap. */
2894 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2895
2896 /* Clear the whole map. */
2897 memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH);
2898
2899 /* charset_not matches newline according to a syntax bit. */
2900 if ((re_opcode_t) b[-2] == charset_not
2901 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2902 SET_LIST_BIT ('\n');
2903
2904 /* Read in characters and ranges, setting map bits. */
2905 for (;;)
2906 {
2907 boolean escaped_char = false;
2908 const unsigned char *p2 = p;
2909 re_wchar_t ch;
2910
2911 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2912
2913 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2914 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2915 So the translation is done later in a loop. Example:
2916 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2917 PATFETCH (c);
2918
2919 /* \ might escape characters inside [...] and [^...]. */
2920 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2921 {
2922 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2923
2924 PATFETCH (c);
2925 escaped_char = true;
2926 }
2927 else
2928 {
2929 /* Could be the end of the bracket expression. If it's
2930 not (i.e., when the bracket expression is `[]' so
2931 far), the ']' character bit gets set way below. */
2932 if (c == ']' && p2 != p1)
2933 break;
2934 }
2935
2936 /* See if we're at the beginning of a possible character
2937 class. */
2938
2939 if (!escaped_char &&
2940 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2941 {
2942 /* Leave room for the null. */
2943 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2944 const unsigned char *class_beg;
2945
2946 PATFETCH (c);
2947 c1 = 0;
2948 class_beg = p;
2949
2950 /* If pattern is `[[:'. */
2951 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2952
2953 for (;;)
2954 {
2955 PATFETCH (c);
2956 if ((c == ':' && *p == ']') || p == pend)
2957 break;
2958 if (c1 < CHAR_CLASS_MAX_LENGTH)
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 isn't a word bracketed by `[:' and `:]':
2967 undo the ending character, the letters, and
2968 leave the leading `:' and `[' (but set bits for
2969 them). */
2970 if (c == ':' && *p == ']')
2971 {
2972 re_wctype_t cc = re_wctype (str);
2973
2974 if (cc == 0)
2975 FREE_STACK_RETURN (REG_ECTYPE);
2976
2977 /* Throw away the ] at the end of the character
2978 class. */
2979 PATFETCH (c);
2980
2981 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2982
2983 #ifndef emacs
2984 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
2985 if (re_iswctype (btowc (ch), cc))
2986 {
2987 c = TRANSLATE (ch);
2988 if (c < (1 << BYTEWIDTH))
2989 SET_LIST_BIT (c);
2990 }
2991 #else /* emacs */
2992 /* Most character classes in a multibyte match
2993 just set a flag. Exceptions are is_blank,
2994 is_digit, is_cntrl, and is_xdigit, since
2995 they can only match ASCII characters. We
2996 don't need to handle them for multibyte.
2997 They are distinguished by a negative wctype. */
2998
2999 /* Setup the gl_state object to its buffer-defined
3000 value. This hardcodes the buffer-global
3001 syntax-table for ASCII chars, while the other chars
3002 will obey syntax-table properties. It's not ideal,
3003 but it's the way it's been done until now. */
3004 SETUP_BUFFER_SYNTAX_TABLE ();
3005
3006 for (ch = 0; ch < 256; ++ch)
3007 {
3008 c = RE_CHAR_TO_MULTIBYTE (ch);
3009 if (! CHAR_BYTE8_P (c)
3010 && re_iswctype (c, cc))
3011 {
3012 SET_LIST_BIT (ch);
3013 c1 = TRANSLATE (c);
3014 if (c1 == c)
3015 continue;
3016 if (ASCII_CHAR_P (c1))
3017 SET_LIST_BIT (c1);
3018 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
3019 SET_LIST_BIT (c1);
3020 }
3021 }
3022 SET_RANGE_TABLE_WORK_AREA_BIT
3023 (range_table_work, re_wctype_to_bit (cc));
3024 #endif /* emacs */
3025 /* In most cases the matching rule for char classes
3026 only uses the syntax table for multibyte chars,
3027 so that the content of the syntax-table it is not
3028 hardcoded in the range_table. SPACE and WORD are
3029 the two exceptions. */
3030 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
3031 bufp->used_syntax = 1;
3032
3033 /* Repeat the loop. */
3034 continue;
3035 }
3036 else
3037 {
3038 /* Go back to right after the "[:". */
3039 p = class_beg;
3040 SET_LIST_BIT ('[');
3041
3042 /* Because the `:' may starts the range, we
3043 can't simply set bit and repeat the loop.
3044 Instead, just set it to C and handle below. */
3045 c = ':';
3046 }
3047 }
3048
3049 if (p < pend && p[0] == '-' && p[1] != ']')
3050 {
3051
3052 /* Discard the `-'. */
3053 PATFETCH (c1);
3054
3055 /* Fetch the character which ends the range. */
3056 PATFETCH (c1);
3057 #ifdef emacs
3058 if (CHAR_BYTE8_P (c1)
3059 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3060 /* Treat the range from a multibyte character to
3061 raw-byte character as empty. */
3062 c = c1 + 1;
3063 #endif /* emacs */
3064 }
3065 else
3066 /* Range from C to C. */
3067 c1 = c;
3068
3069 if (c > c1)
3070 {
3071 if (syntax & RE_NO_EMPTY_RANGES)
3072 FREE_STACK_RETURN (REG_ERANGEX);
3073 /* Else, repeat the loop. */
3074 }
3075 else
3076 {
3077 #ifndef emacs
3078 /* Set the range into bitmap */
3079 for (; c <= c1; c++)
3080 {
3081 ch = TRANSLATE (c);
3082 if (ch < (1 << BYTEWIDTH))
3083 SET_LIST_BIT (ch);
3084 }
3085 #else /* emacs */
3086 if (c < 128)
3087 {
3088 ch = MIN (127, c1);
3089 SETUP_ASCII_RANGE (range_table_work, c, ch);
3090 c = ch + 1;
3091 if (CHAR_BYTE8_P (c1))
3092 c = BYTE8_TO_CHAR (128);
3093 }
3094 if (c <= c1)
3095 {
3096 if (CHAR_BYTE8_P (c))
3097 {
3098 c = CHAR_TO_BYTE8 (c);
3099 c1 = CHAR_TO_BYTE8 (c1);
3100 for (; c <= c1; c++)
3101 SET_LIST_BIT (c);
3102 }
3103 else if (multibyte)
3104 {
3105 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3106 }
3107 else
3108 {
3109 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3110 }
3111 }
3112 #endif /* emacs */
3113 }
3114 }
3115
3116 /* Discard any (non)matching list bytes that are all 0 at the
3117 end of the map. Decrease the map-length byte too. */
3118 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3119 b[-1]--;
3120 b += b[-1];
3121
3122 /* Build real range table from work area. */
3123 if (RANGE_TABLE_WORK_USED (range_table_work)
3124 || RANGE_TABLE_WORK_BITS (range_table_work))
3125 {
3126 int i;
3127 int used = RANGE_TABLE_WORK_USED (range_table_work);
3128
3129 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3130 bytes for flags, two for COUNT, and three bytes for
3131 each character. */
3132 GET_BUFFER_SPACE (4 + used * 3);
3133
3134 /* Indicate the existence of range table. */
3135 laststart[1] |= 0x80;
3136
3137 /* Store the character class flag bits into the range table.
3138 If not in emacs, these flag bits are always 0. */
3139 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3140 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3141
3142 STORE_NUMBER_AND_INCR (b, used / 2);
3143 for (i = 0; i < used; i++)
3144 STORE_CHARACTER_AND_INCR
3145 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3146 }
3147 }
3148 break;
3149
3150
3151 case '(':
3152 if (syntax & RE_NO_BK_PARENS)
3153 goto handle_open;
3154 else
3155 goto normal_char;
3156
3157
3158 case ')':
3159 if (syntax & RE_NO_BK_PARENS)
3160 goto handle_close;
3161 else
3162 goto normal_char;
3163
3164
3165 case '\n':
3166 if (syntax & RE_NEWLINE_ALT)
3167 goto handle_alt;
3168 else
3169 goto normal_char;
3170
3171
3172 case '|':
3173 if (syntax & RE_NO_BK_VBAR)
3174 goto handle_alt;
3175 else
3176 goto normal_char;
3177
3178
3179 case '{':
3180 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3181 goto handle_interval;
3182 else
3183 goto normal_char;
3184
3185
3186 case '\\':
3187 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3188
3189 /* Do not translate the character after the \, so that we can
3190 distinguish, e.g., \B from \b, even if we normally would
3191 translate, e.g., B to b. */
3192 PATFETCH (c);
3193
3194 switch (c)
3195 {
3196 case '(':
3197 if (syntax & RE_NO_BK_PARENS)
3198 goto normal_backslash;
3199
3200 handle_open:
3201 {
3202 int shy = 0;
3203 regnum_t regnum = 0;
3204 if (p+1 < pend)
3205 {
3206 /* Look for a special (?...) construct */
3207 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3208 {
3209 PATFETCH (c); /* Gobble up the '?'. */
3210 while (!shy)
3211 {
3212 PATFETCH (c);
3213 switch (c)
3214 {
3215 case ':': shy = 1; break;
3216 case '0':
3217 /* An explicitly specified regnum must start
3218 with non-0. */
3219 if (regnum == 0)
3220 FREE_STACK_RETURN (REG_BADPAT);
3221 case '1': case '2': case '3': case '4':
3222 case '5': case '6': case '7': case '8': case '9':
3223 regnum = 10*regnum + (c - '0'); break;
3224 default:
3225 /* Only (?:...) is supported right now. */
3226 FREE_STACK_RETURN (REG_BADPAT);
3227 }
3228 }
3229 }
3230 }
3231
3232 if (!shy)
3233 regnum = ++bufp->re_nsub;
3234 else if (regnum)
3235 { /* It's actually not shy, but explicitly numbered. */
3236 shy = 0;
3237 if (regnum > bufp->re_nsub)
3238 bufp->re_nsub = regnum;
3239 else if (regnum > bufp->re_nsub
3240 /* Ideally, we'd want to check that the specified
3241 group can't have matched (i.e. all subgroups
3242 using the same regnum are in other branches of
3243 OR patterns), but we don't currently keep track
3244 of enough info to do that easily. */
3245 || group_in_compile_stack (compile_stack, regnum))
3246 FREE_STACK_RETURN (REG_BADPAT);
3247 }
3248 else
3249 /* It's really shy. */
3250 regnum = - bufp->re_nsub;
3251
3252 if (COMPILE_STACK_FULL)
3253 {
3254 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3255 compile_stack_elt_t);
3256 if (compile_stack.stack == NULL) return REG_ESPACE;
3257
3258 compile_stack.size <<= 1;
3259 }
3260
3261 /* These are the values to restore when we hit end of this
3262 group. They are all relative offsets, so that if the
3263 whole pattern moves because of realloc, they will still
3264 be valid. */
3265 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3266 COMPILE_STACK_TOP.fixup_alt_jump
3267 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3268 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3269 COMPILE_STACK_TOP.regnum = regnum;
3270
3271 /* Do not push a start_memory for groups beyond the last one
3272 we can represent in the compiled pattern. */
3273 if (regnum <= MAX_REGNUM && regnum > 0)
3274 BUF_PUSH_2 (start_memory, regnum);
3275
3276 compile_stack.avail++;
3277
3278 fixup_alt_jump = 0;
3279 laststart = 0;
3280 begalt = b;
3281 /* If we've reached MAX_REGNUM groups, then this open
3282 won't actually generate any code, so we'll have to
3283 clear pending_exact explicitly. */
3284 pending_exact = 0;
3285 break;
3286 }
3287
3288 case ')':
3289 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3290
3291 if (COMPILE_STACK_EMPTY)
3292 {
3293 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3294 goto normal_backslash;
3295 else
3296 FREE_STACK_RETURN (REG_ERPAREN);
3297 }
3298
3299 handle_close:
3300 FIXUP_ALT_JUMP ();
3301
3302 /* See similar code for backslashed left paren above. */
3303 if (COMPILE_STACK_EMPTY)
3304 {
3305 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3306 goto normal_char;
3307 else
3308 FREE_STACK_RETURN (REG_ERPAREN);
3309 }
3310
3311 /* Since we just checked for an empty stack above, this
3312 ``can't happen''. */
3313 assert (compile_stack.avail != 0);
3314 {
3315 /* We don't just want to restore into `regnum', because
3316 later groups should continue to be numbered higher,
3317 as in `(ab)c(de)' -- the second group is #2. */
3318 regnum_t regnum;
3319
3320 compile_stack.avail--;
3321 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3322 fixup_alt_jump
3323 = COMPILE_STACK_TOP.fixup_alt_jump
3324 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3325 : 0;
3326 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3327 regnum = COMPILE_STACK_TOP.regnum;
3328 /* If we've reached MAX_REGNUM groups, then this open
3329 won't actually generate any code, so we'll have to
3330 clear pending_exact explicitly. */
3331 pending_exact = 0;
3332
3333 /* We're at the end of the group, so now we know how many
3334 groups were inside this one. */
3335 if (regnum <= MAX_REGNUM && regnum > 0)
3336 BUF_PUSH_2 (stop_memory, regnum);
3337 }
3338 break;
3339
3340
3341 case '|': /* `\|'. */
3342 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3343 goto normal_backslash;
3344 handle_alt:
3345 if (syntax & RE_LIMITED_OPS)
3346 goto normal_char;
3347
3348 /* Insert before the previous alternative a jump which
3349 jumps to this alternative if the former fails. */
3350 GET_BUFFER_SPACE (3);
3351 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3352 pending_exact = 0;
3353 b += 3;
3354
3355 /* The alternative before this one has a jump after it
3356 which gets executed if it gets matched. Adjust that
3357 jump so it will jump to this alternative's analogous
3358 jump (put in below, which in turn will jump to the next
3359 (if any) alternative's such jump, etc.). The last such
3360 jump jumps to the correct final destination. A picture:
3361 _____ _____
3362 | | | |
3363 | v | v
3364 a | b | c
3365
3366 If we are at `b', then fixup_alt_jump right now points to a
3367 three-byte space after `a'. We'll put in the jump, set
3368 fixup_alt_jump to right after `b', and leave behind three
3369 bytes which we'll fill in when we get to after `c'. */
3370
3371 FIXUP_ALT_JUMP ();
3372
3373 /* Mark and leave space for a jump after this alternative,
3374 to be filled in later either by next alternative or
3375 when know we're at the end of a series of alternatives. */
3376 fixup_alt_jump = b;
3377 GET_BUFFER_SPACE (3);
3378 b += 3;
3379
3380 laststart = 0;
3381 begalt = b;
3382 break;
3383
3384
3385 case '{':
3386 /* If \{ is a literal. */
3387 if (!(syntax & RE_INTERVALS)
3388 /* If we're at `\{' and it's not the open-interval
3389 operator. */
3390 || (syntax & RE_NO_BK_BRACES))
3391 goto normal_backslash;
3392
3393 handle_interval:
3394 {
3395 /* If got here, then the syntax allows intervals. */
3396
3397 /* At least (most) this many matches must be made. */
3398 int lower_bound = 0, upper_bound = -1;
3399
3400 beg_interval = p;
3401
3402 GET_UNSIGNED_NUMBER (lower_bound);
3403
3404 if (c == ',')
3405 GET_UNSIGNED_NUMBER (upper_bound);
3406 else
3407 /* Interval such as `{1}' => match exactly once. */
3408 upper_bound = lower_bound;
3409
3410 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3411 || (upper_bound >= 0 && lower_bound > upper_bound))
3412 FREE_STACK_RETURN (REG_BADBR);
3413
3414 if (!(syntax & RE_NO_BK_BRACES))
3415 {
3416 if (c != '\\')
3417 FREE_STACK_RETURN (REG_BADBR);
3418 if (p == pend)
3419 FREE_STACK_RETURN (REG_EESCAPE);
3420 PATFETCH (c);
3421 }
3422
3423 if (c != '}')
3424 FREE_STACK_RETURN (REG_BADBR);
3425
3426 /* We just parsed a valid interval. */
3427
3428 /* If it's invalid to have no preceding re. */
3429 if (!laststart)
3430 {
3431 if (syntax & RE_CONTEXT_INVALID_OPS)
3432 FREE_STACK_RETURN (REG_BADRPT);
3433 else if (syntax & RE_CONTEXT_INDEP_OPS)
3434 laststart = b;
3435 else
3436 goto unfetch_interval;
3437 }
3438
3439 if (upper_bound == 0)
3440 /* If the upper bound is zero, just drop the sub pattern
3441 altogether. */
3442 b = laststart;
3443 else if (lower_bound == 1 && upper_bound == 1)
3444 /* Just match it once: nothing to do here. */
3445 ;
3446
3447 /* Otherwise, we have a nontrivial interval. When
3448 we're all done, the pattern will look like:
3449 set_number_at <jump count> <upper bound>
3450 set_number_at <succeed_n count> <lower bound>
3451 succeed_n <after jump addr> <succeed_n count>
3452 <body of loop>
3453 jump_n <succeed_n addr> <jump count>
3454 (The upper bound and `jump_n' are omitted if
3455 `upper_bound' is 1, though.) */
3456 else
3457 { /* If the upper bound is > 1, we need to insert
3458 more at the end of the loop. */
3459 unsigned int nbytes = (upper_bound < 0 ? 3
3460 : upper_bound > 1 ? 5 : 0);
3461 unsigned int startoffset = 0;
3462
3463 GET_BUFFER_SPACE (20); /* We might use less. */
3464
3465 if (lower_bound == 0)
3466 {
3467 /* A succeed_n that starts with 0 is really a
3468 a simple on_failure_jump_loop. */
3469 INSERT_JUMP (on_failure_jump_loop, laststart,
3470 b + 3 + nbytes);
3471 b += 3;
3472 }
3473 else
3474 {
3475 /* Initialize lower bound of the `succeed_n', even
3476 though it will be set during matching by its
3477 attendant `set_number_at' (inserted next),
3478 because `re_compile_fastmap' needs to know.
3479 Jump to the `jump_n' we might insert below. */
3480 INSERT_JUMP2 (succeed_n, laststart,
3481 b + 5 + nbytes,
3482 lower_bound);
3483 b += 5;
3484
3485 /* Code to initialize the lower bound. Insert
3486 before the `succeed_n'. The `5' is the last two
3487 bytes of this `set_number_at', plus 3 bytes of
3488 the following `succeed_n'. */
3489 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3490 b += 5;
3491 startoffset += 5;
3492 }
3493
3494 if (upper_bound < 0)
3495 {
3496 /* A negative upper bound stands for infinity,
3497 in which case it degenerates to a plain jump. */
3498 STORE_JUMP (jump, b, laststart + startoffset);
3499 b += 3;
3500 }
3501 else if (upper_bound > 1)
3502 { /* More than one repetition is allowed, so
3503 append a backward jump to the `succeed_n'
3504 that starts this interval.
3505
3506 When we've reached this during matching,
3507 we'll have matched the interval once, so
3508 jump back only `upper_bound - 1' times. */
3509 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3510 upper_bound - 1);
3511 b += 5;
3512
3513 /* The location we want to set is the second
3514 parameter of the `jump_n'; that is `b-2' as
3515 an absolute address. `laststart' will be
3516 the `set_number_at' we're about to insert;
3517 `laststart+3' the number to set, the source
3518 for the relative address. But we are
3519 inserting into the middle of the pattern --
3520 so everything is getting moved up by 5.
3521 Conclusion: (b - 2) - (laststart + 3) + 5,
3522 i.e., b - laststart.
3523
3524 We insert this at the beginning of the loop
3525 so that if we fail during matching, we'll
3526 reinitialize the bounds. */
3527 insert_op2 (set_number_at, laststart, b - laststart,
3528 upper_bound - 1, b);
3529 b += 5;
3530 }
3531 }
3532 pending_exact = 0;
3533 beg_interval = NULL;
3534 }
3535 break;
3536
3537 unfetch_interval:
3538 /* If an invalid interval, match the characters as literals. */
3539 assert (beg_interval);
3540 p = beg_interval;
3541 beg_interval = NULL;
3542
3543 /* normal_char and normal_backslash need `c'. */
3544 c = '{';
3545
3546 if (!(syntax & RE_NO_BK_BRACES))
3547 {
3548 assert (p > pattern && p[-1] == '\\');
3549 goto normal_backslash;
3550 }
3551 else
3552 goto normal_char;
3553
3554 #ifdef emacs
3555 /* There is no way to specify the before_dot and after_dot
3556 operators. rms says this is ok. --karl */
3557 case '=':
3558 BUF_PUSH (at_dot);
3559 break;
3560
3561 case 's':
3562 laststart = b;
3563 PATFETCH (c);
3564 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3565 break;
3566
3567 case 'S':
3568 laststart = b;
3569 PATFETCH (c);
3570 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3571 break;
3572
3573 case 'c':
3574 laststart = b;
3575 PATFETCH (c);
3576 BUF_PUSH_2 (categoryspec, c);
3577 break;
3578
3579 case 'C':
3580 laststart = b;
3581 PATFETCH (c);
3582 BUF_PUSH_2 (notcategoryspec, c);
3583 break;
3584 #endif /* emacs */
3585
3586
3587 case 'w':
3588 if (syntax & RE_NO_GNU_OPS)
3589 goto normal_char;
3590 laststart = b;
3591 BUF_PUSH_2 (syntaxspec, Sword);
3592 break;
3593
3594
3595 case 'W':
3596 if (syntax & RE_NO_GNU_OPS)
3597 goto normal_char;
3598 laststart = b;
3599 BUF_PUSH_2 (notsyntaxspec, Sword);
3600 break;
3601
3602
3603 case '<':
3604 if (syntax & RE_NO_GNU_OPS)
3605 goto normal_char;
3606 BUF_PUSH (wordbeg);
3607 break;
3608
3609 case '>':
3610 if (syntax & RE_NO_GNU_OPS)
3611 goto normal_char;
3612 BUF_PUSH (wordend);
3613 break;
3614
3615 case '_':
3616 if (syntax & RE_NO_GNU_OPS)
3617 goto normal_char;
3618 laststart = b;
3619 PATFETCH (c);
3620 if (c == '<')
3621 BUF_PUSH (symbeg);
3622 else if (c == '>')
3623 BUF_PUSH (symend);
3624 else
3625 FREE_STACK_RETURN (REG_BADPAT);
3626 break;
3627
3628 case 'b':
3629 if (syntax & RE_NO_GNU_OPS)
3630 goto normal_char;
3631 BUF_PUSH (wordbound);
3632 break;
3633
3634 case 'B':
3635 if (syntax & RE_NO_GNU_OPS)
3636 goto normal_char;
3637 BUF_PUSH (notwordbound);
3638 break;
3639
3640 case '`':
3641 if (syntax & RE_NO_GNU_OPS)
3642 goto normal_char;
3643 BUF_PUSH (begbuf);
3644 break;
3645
3646 case '\'':
3647 if (syntax & RE_NO_GNU_OPS)
3648 goto normal_char;
3649 BUF_PUSH (endbuf);
3650 break;
3651
3652 case '1': case '2': case '3': case '4': case '5':
3653 case '6': case '7': case '8': case '9':
3654 {
3655 regnum_t reg;
3656
3657 if (syntax & RE_NO_BK_REFS)
3658 goto normal_backslash;
3659
3660 reg = c - '0';
3661
3662 if (reg > bufp->re_nsub || reg < 1
3663 /* Can't back reference to a subexp before its end. */
3664 || group_in_compile_stack (compile_stack, reg))
3665 FREE_STACK_RETURN (REG_ESUBREG);
3666
3667 laststart = b;
3668 BUF_PUSH_2 (duplicate, reg);
3669 }
3670 break;
3671
3672
3673 case '+':
3674 case '?':
3675 if (syntax & RE_BK_PLUS_QM)
3676 goto handle_plus;
3677 else
3678 goto normal_backslash;
3679
3680 default:
3681 normal_backslash:
3682 /* You might think it would be useful for \ to mean
3683 not to translate; but if we don't translate it
3684 it will never match anything. */
3685 goto normal_char;
3686 }
3687 break;
3688
3689
3690 default:
3691 /* Expects the character in `c'. */
3692 normal_char:
3693 /* If no exactn currently being built. */
3694 if (!pending_exact
3695
3696 /* If last exactn not at current position. */
3697 || pending_exact + *pending_exact + 1 != b
3698
3699 /* We have only one byte following the exactn for the count. */
3700 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3701
3702 /* If followed by a repetition operator. */
3703 || (p != pend && (*p == '*' || *p == '^'))
3704 || ((syntax & RE_BK_PLUS_QM)
3705 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3706 : p != pend && (*p == '+' || *p == '?'))
3707 || ((syntax & RE_INTERVALS)
3708 && ((syntax & RE_NO_BK_BRACES)
3709 ? p != pend && *p == '{'
3710 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3711 {
3712 /* Start building a new exactn. */
3713
3714 laststart = b;
3715
3716 BUF_PUSH_2 (exactn, 0);
3717 pending_exact = b - 1;
3718 }
3719
3720 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3721 {
3722 int len;
3723
3724 if (multibyte)
3725 {
3726 c = TRANSLATE (c);
3727 len = CHAR_STRING (c, b);
3728 b += len;
3729 }
3730 else
3731 {
3732 c1 = RE_CHAR_TO_MULTIBYTE (c);
3733 if (! CHAR_BYTE8_P (c1))
3734 {
3735 re_wchar_t c2 = TRANSLATE (c1);
3736
3737 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3738 c = c1;
3739 }
3740 *b++ = c;
3741 len = 1;
3742 }
3743 (*pending_exact) += len;
3744 }
3745
3746 break;
3747 } /* switch (c) */
3748 } /* while p != pend */
3749
3750
3751 /* Through the pattern now. */
3752
3753 FIXUP_ALT_JUMP ();
3754
3755 if (!COMPILE_STACK_EMPTY)
3756 FREE_STACK_RETURN (REG_EPAREN);
3757
3758 /* If we don't want backtracking, force success
3759 the first time we reach the end of the compiled pattern. */
3760 if (syntax & RE_NO_POSIX_BACKTRACKING)
3761 BUF_PUSH (succeed);
3762
3763 /* We have succeeded; set the length of the buffer. */
3764 bufp->used = b - bufp->buffer;
3765
3766 #ifdef DEBUG
3767 if (debug > 0)
3768 {
3769 re_compile_fastmap (bufp);
3770 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3771 print_compiled_pattern (bufp);
3772 }
3773 debug--;
3774 #endif /* DEBUG */
3775
3776 #ifndef MATCH_MAY_ALLOCATE
3777 /* Initialize the failure stack to the largest possible stack. This
3778 isn't necessary unless we're trying to avoid calling alloca in
3779 the search and match routines. */
3780 {
3781 int num_regs = bufp->re_nsub + 1;
3782
3783 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3784 {
3785 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3786
3787 if (! fail_stack.stack)
3788 fail_stack.stack
3789 = (fail_stack_elt_t *) malloc (fail_stack.size
3790 * sizeof (fail_stack_elt_t));
3791 else
3792 fail_stack.stack
3793 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3794 (fail_stack.size
3795 * sizeof (fail_stack_elt_t)));
3796 }
3797
3798 regex_grow_registers (num_regs);
3799 }
3800 #endif /* not MATCH_MAY_ALLOCATE */
3801
3802 FREE_STACK_RETURN (REG_NOERROR);
3803 } /* regex_compile */
3804 \f
3805 /* Subroutines for `regex_compile'. */
3806
3807 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3808
3809 static void
3810 store_op1 (re_opcode_t op, unsigned char *loc, int arg)
3811 {
3812 *loc = (unsigned char) op;
3813 STORE_NUMBER (loc + 1, arg);
3814 }
3815
3816
3817 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3818
3819 static void
3820 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2)
3821 {
3822 *loc = (unsigned char) op;
3823 STORE_NUMBER (loc + 1, arg1);
3824 STORE_NUMBER (loc + 3, arg2);
3825 }
3826
3827
3828 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3829 for OP followed by two-byte integer parameter ARG. */
3830
3831 static void
3832 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end)
3833 {
3834 register unsigned char *pfrom = end;
3835 register unsigned char *pto = end + 3;
3836
3837 while (pfrom != loc)
3838 *--pto = *--pfrom;
3839
3840 store_op1 (op, loc, arg);
3841 }
3842
3843
3844 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3845
3846 static void
3847 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end)
3848 {
3849 register unsigned char *pfrom = end;
3850 register unsigned char *pto = end + 5;
3851
3852 while (pfrom != loc)
3853 *--pto = *--pfrom;
3854
3855 store_op2 (op, loc, arg1, arg2);
3856 }
3857
3858
3859 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3860 after an alternative or a begin-subexpression. We assume there is at
3861 least one character before the ^. */
3862
3863 static boolean
3864 at_begline_loc_p (const re_char *pattern, const re_char *p, reg_syntax_t syntax)
3865 {
3866 re_char *prev = p - 2;
3867 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3868
3869 return
3870 /* After a subexpression? */
3871 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3872 /* After an alternative? */
3873 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3874 /* After a shy subexpression? */
3875 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3876 && prev[-1] == '?' && prev[-2] == '('
3877 && (syntax & RE_NO_BK_PARENS
3878 || (prev - 3 >= pattern && prev[-3] == '\\')));
3879 }
3880
3881
3882 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3883 at least one character after the $, i.e., `P < PEND'. */
3884
3885 static boolean
3886 at_endline_loc_p (const re_char *p, const re_char *pend, reg_syntax_t syntax)
3887 {
3888 re_char *next = p;
3889 boolean next_backslash = *next == '\\';
3890 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3891
3892 return
3893 /* Before a subexpression? */
3894 (syntax & RE_NO_BK_PARENS ? *next == ')'
3895 : next_backslash && next_next && *next_next == ')')
3896 /* Before an alternative? */
3897 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3898 : next_backslash && next_next && *next_next == '|');
3899 }
3900
3901
3902 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3903 false if it's not. */
3904
3905 static boolean
3906 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
3907 {
3908 ssize_t this_element;
3909
3910 for (this_element = compile_stack.avail - 1;
3911 this_element >= 0;
3912 this_element--)
3913 if (compile_stack.stack[this_element].regnum == regnum)
3914 return true;
3915
3916 return false;
3917 }
3918 \f
3919 /* analyse_first.
3920 If fastmap is non-NULL, go through the pattern and fill fastmap
3921 with all the possible leading chars. If fastmap is NULL, don't
3922 bother filling it up (obviously) and only return whether the
3923 pattern could potentially match the empty string.
3924
3925 Return 1 if p..pend might match the empty string.
3926 Return 0 if p..pend matches at least one char.
3927 Return -1 if fastmap was not updated accurately. */
3928
3929 static int
3930 analyse_first (const re_char *p, const re_char *pend, char *fastmap, const int multibyte)
3931 {
3932 int j, k;
3933 boolean not;
3934
3935 /* If all elements for base leading-codes in fastmap is set, this
3936 flag is set true. */
3937 boolean match_any_multibyte_characters = false;
3938
3939 assert (p);
3940
3941 /* The loop below works as follows:
3942 - It has a working-list kept in the PATTERN_STACK and which basically
3943 starts by only containing a pointer to the first operation.
3944 - If the opcode we're looking at is a match against some set of
3945 chars, then we add those chars to the fastmap and go on to the
3946 next work element from the worklist (done via `break').
3947 - If the opcode is a control operator on the other hand, we either
3948 ignore it (if it's meaningless at this point, such as `start_memory')
3949 or execute it (if it's a jump). If the jump has several destinations
3950 (i.e. `on_failure_jump'), then we push the other destination onto the
3951 worklist.
3952 We guarantee termination by ignoring backward jumps (more or less),
3953 so that `p' is monotonically increasing. More to the point, we
3954 never set `p' (or push) anything `<= p1'. */
3955
3956 while (p < pend)
3957 {
3958 /* `p1' is used as a marker of how far back a `on_failure_jump'
3959 can go without being ignored. It is normally equal to `p'
3960 (which prevents any backward `on_failure_jump') except right
3961 after a plain `jump', to allow patterns such as:
3962 0: jump 10
3963 3..9: <body>
3964 10: on_failure_jump 3
3965 as used for the *? operator. */
3966 re_char *p1 = p;
3967
3968 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3969 {
3970 case succeed:
3971 return 1;
3972
3973 case duplicate:
3974 /* If the first character has to match a backreference, that means
3975 that the group was empty (since it already matched). Since this
3976 is the only case that interests us here, we can assume that the
3977 backreference must match the empty string. */
3978 p++;
3979 continue;
3980
3981
3982 /* Following are the cases which match a character. These end
3983 with `break'. */
3984
3985 case exactn:
3986 if (fastmap)
3987 {
3988 /* If multibyte is nonzero, the first byte of each
3989 character is an ASCII or a leading code. Otherwise,
3990 each byte is a character. Thus, this works in both
3991 cases. */
3992 fastmap[p[1]] = 1;
3993 if (! multibyte)
3994 {
3995 /* For the case of matching this unibyte regex
3996 against multibyte, we must set a leading code of
3997 the corresponding multibyte character. */
3998 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
3999
4000 fastmap[CHAR_LEADING_CODE (c)] = 1;
4001 }
4002 }
4003 break;
4004
4005
4006 case anychar:
4007 /* We could put all the chars except for \n (and maybe \0)
4008 but we don't bother since it is generally not worth it. */
4009 if (!fastmap) break;
4010 return -1;
4011
4012
4013 case charset_not:
4014 if (!fastmap) break;
4015 {
4016 /* Chars beyond end of bitmap are possible matches. */
4017 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
4018 j < (1 << BYTEWIDTH); j++)
4019 fastmap[j] = 1;
4020 }
4021
4022 /* Fallthrough */
4023 case charset:
4024 if (!fastmap) break;
4025 not = (re_opcode_t) *(p - 1) == charset_not;
4026 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
4027 j >= 0; j--)
4028 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
4029 fastmap[j] = 1;
4030
4031 #ifdef emacs
4032 if (/* Any leading code can possibly start a character
4033 which doesn't match the specified set of characters. */
4034 not
4035 ||
4036 /* If we can match a character class, we can match any
4037 multibyte characters. */
4038 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4039 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4040
4041 {
4042 if (match_any_multibyte_characters == false)
4043 {
4044 for (j = MIN_MULTIBYTE_LEADING_CODE;
4045 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4046 fastmap[j] = 1;
4047 match_any_multibyte_characters = true;
4048 }
4049 }
4050
4051 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4052 && match_any_multibyte_characters == false)
4053 {
4054 /* Set fastmap[I] to 1 where I is a leading code of each
4055 multibyte character in the range table. */
4056 int c, count;
4057 unsigned char lc1, lc2;
4058
4059 /* Make P points the range table. `+ 2' is to skip flag
4060 bits for a character class. */
4061 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4062
4063 /* Extract the number of ranges in range table into COUNT. */
4064 EXTRACT_NUMBER_AND_INCR (count, p);
4065 for (; count > 0; count--, p += 3)
4066 {
4067 /* Extract the start and end of each range. */
4068 EXTRACT_CHARACTER (c, p);
4069 lc1 = CHAR_LEADING_CODE (c);
4070 p += 3;
4071 EXTRACT_CHARACTER (c, p);
4072 lc2 = CHAR_LEADING_CODE (c);
4073 for (j = lc1; j <= lc2; j++)
4074 fastmap[j] = 1;
4075 }
4076 }
4077 #endif
4078 break;
4079
4080 case syntaxspec:
4081 case notsyntaxspec:
4082 if (!fastmap) break;
4083 #ifndef emacs
4084 not = (re_opcode_t)p[-1] == notsyntaxspec;
4085 k = *p++;
4086 for (j = 0; j < (1 << BYTEWIDTH); j++)
4087 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4088 fastmap[j] = 1;
4089 break;
4090 #else /* emacs */
4091 /* This match depends on text properties. These end with
4092 aborting optimizations. */
4093 return -1;
4094
4095 case categoryspec:
4096 case notcategoryspec:
4097 if (!fastmap) break;
4098 not = (re_opcode_t)p[-1] == notcategoryspec;
4099 k = *p++;
4100 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4101 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4102 fastmap[j] = 1;
4103
4104 /* Any leading code can possibly start a character which
4105 has or doesn't has the specified category. */
4106 if (match_any_multibyte_characters == false)
4107 {
4108 for (j = MIN_MULTIBYTE_LEADING_CODE;
4109 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4110 fastmap[j] = 1;
4111 match_any_multibyte_characters = true;
4112 }
4113 break;
4114
4115 /* All cases after this match the empty string. These end with
4116 `continue'. */
4117
4118 case before_dot:
4119 case at_dot:
4120 case after_dot:
4121 #endif /* !emacs */
4122 case no_op:
4123 case begline:
4124 case endline:
4125 case begbuf:
4126 case endbuf:
4127 case wordbound:
4128 case notwordbound:
4129 case wordbeg:
4130 case wordend:
4131 case symbeg:
4132 case symend:
4133 continue;
4134
4135
4136 case jump:
4137 EXTRACT_NUMBER_AND_INCR (j, p);
4138 if (j < 0)
4139 /* Backward jumps can only go back to code that we've already
4140 visited. `re_compile' should make sure this is true. */
4141 break;
4142 p += j;
4143 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4144 {
4145 case on_failure_jump:
4146 case on_failure_keep_string_jump:
4147 case on_failure_jump_loop:
4148 case on_failure_jump_nastyloop:
4149 case on_failure_jump_smart:
4150 p++;
4151 break;
4152 default:
4153 continue;
4154 };
4155 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4156 to jump back to "just after here". */
4157 /* Fallthrough */
4158
4159 case on_failure_jump:
4160 case on_failure_keep_string_jump:
4161 case on_failure_jump_nastyloop:
4162 case on_failure_jump_loop:
4163 case on_failure_jump_smart:
4164 EXTRACT_NUMBER_AND_INCR (j, p);
4165 if (p + j <= p1)
4166 ; /* Backward jump to be ignored. */
4167 else
4168 { /* We have to look down both arms.
4169 We first go down the "straight" path so as to minimize
4170 stack usage when going through alternatives. */
4171 int r = analyse_first (p, pend, fastmap, multibyte);
4172 if (r) return r;
4173 p += j;
4174 }
4175 continue;
4176
4177
4178 case jump_n:
4179 /* This code simply does not properly handle forward jump_n. */
4180 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4181 p += 4;
4182 /* jump_n can either jump or fall through. The (backward) jump
4183 case has already been handled, so we only need to look at the
4184 fallthrough case. */
4185 continue;
4186
4187 case succeed_n:
4188 /* If N == 0, it should be an on_failure_jump_loop instead. */
4189 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4190 p += 4;
4191 /* We only care about one iteration of the loop, so we don't
4192 need to consider the case where this behaves like an
4193 on_failure_jump. */
4194 continue;
4195
4196
4197 case set_number_at:
4198 p += 4;
4199 continue;
4200
4201
4202 case start_memory:
4203 case stop_memory:
4204 p += 1;
4205 continue;
4206
4207
4208 default:
4209 abort (); /* We have listed all the cases. */
4210 } /* switch *p++ */
4211
4212 /* Getting here means we have found the possible starting
4213 characters for one path of the pattern -- and that the empty
4214 string does not match. We need not follow this path further. */
4215 return 0;
4216 } /* while p */
4217
4218 /* We reached the end without matching anything. */
4219 return 1;
4220
4221 } /* analyse_first */
4222 \f
4223 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4224 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4225 characters can start a string that matches the pattern. This fastmap
4226 is used by re_search to skip quickly over impossible starting points.
4227
4228 Character codes above (1 << BYTEWIDTH) are not represented in the
4229 fastmap, but the leading codes are represented. Thus, the fastmap
4230 indicates which character sets could start a match.
4231
4232 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4233 area as BUFP->fastmap.
4234
4235 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4236 the pattern buffer.
4237
4238 Returns 0 if we succeed, -2 if an internal error. */
4239
4240 int
4241 re_compile_fastmap (struct re_pattern_buffer *bufp)
4242 {
4243 char *fastmap = bufp->fastmap;
4244 int analysis;
4245
4246 assert (fastmap && bufp->buffer);
4247
4248 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4249 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4250
4251 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4252 fastmap, RE_MULTIBYTE_P (bufp));
4253 bufp->can_be_null = (analysis != 0);
4254 return 0;
4255 } /* re_compile_fastmap */
4256 \f
4257 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4258 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4259 this memory for recording register information. STARTS and ENDS
4260 must be allocated using the malloc library routine, and must each
4261 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4262
4263 If NUM_REGS == 0, then subsequent matches should allocate their own
4264 register data.
4265
4266 Unless this function is called, the first search or match using
4267 PATTERN_BUFFER will allocate its own register data, without
4268 freeing the old data. */
4269
4270 void
4271 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4272 {
4273 if (num_regs)
4274 {
4275 bufp->regs_allocated = REGS_REALLOCATE;
4276 regs->num_regs = num_regs;
4277 regs->start = starts;
4278 regs->end = ends;
4279 }
4280 else
4281 {
4282 bufp->regs_allocated = REGS_UNALLOCATED;
4283 regs->num_regs = 0;
4284 regs->start = regs->end = (regoff_t *) 0;
4285 }
4286 }
4287 WEAK_ALIAS (__re_set_registers, re_set_registers)
4288 \f
4289 /* Searching routines. */
4290
4291 /* Like re_search_2, below, but only one string is specified, and
4292 doesn't let you say where to stop matching. */
4293
4294 regoff_t
4295 re_search (struct re_pattern_buffer *bufp, const char *string, size_t size,
4296 ssize_t startpos, ssize_t range, struct re_registers *regs)
4297 {
4298 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4299 regs, size);
4300 }
4301 WEAK_ALIAS (__re_search, re_search)
4302
4303 /* Head address of virtual concatenation of string. */
4304 #define HEAD_ADDR_VSTRING(P) \
4305 (((P) >= size1 ? string2 : string1))
4306
4307 /* Address of POS in the concatenation of virtual string. */
4308 #define POS_ADDR_VSTRING(POS) \
4309 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4310
4311 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4312 virtual concatenation of STRING1 and STRING2, starting first at index
4313 STARTPOS, then at STARTPOS + 1, and so on.
4314
4315 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4316
4317 RANGE is how far to scan while trying to match. RANGE = 0 means try
4318 only at STARTPOS; in general, the last start tried is STARTPOS +
4319 RANGE.
4320
4321 In REGS, return the indices of the virtual concatenation of STRING1
4322 and STRING2 that matched the entire BUFP->buffer and its contained
4323 subexpressions.
4324
4325 Do not consider matching one past the index STOP in the virtual
4326 concatenation of STRING1 and STRING2.
4327
4328 We return either the position in the strings at which the match was
4329 found, -1 if no match, or -2 if error (such as failure
4330 stack overflow). */
4331
4332 regoff_t
4333 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, size_t size1,
4334 const char *str2, size_t size2, ssize_t startpos, ssize_t range,
4335 struct re_registers *regs, ssize_t stop)
4336 {
4337 regoff_t val;
4338 re_char *string1 = (re_char*) str1;
4339 re_char *string2 = (re_char*) str2;
4340 register char *fastmap = bufp->fastmap;
4341 register RE_TRANSLATE_TYPE translate = bufp->translate;
4342 size_t total_size = size1 + size2;
4343 ssize_t endpos = startpos + range;
4344 boolean anchored_start;
4345 /* Nonzero if we are searching multibyte string. */
4346 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4347
4348 /* Check for out-of-range STARTPOS. */
4349 if (startpos < 0 || startpos > total_size)
4350 return -1;
4351
4352 /* Fix up RANGE if it might eventually take us outside
4353 the virtual concatenation of STRING1 and STRING2.
4354 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4355 if (endpos < 0)
4356 range = 0 - startpos;
4357 else if (endpos > total_size)
4358 range = total_size - startpos;
4359
4360 /* If the search isn't to be a backwards one, don't waste time in a
4361 search for a pattern anchored at beginning of buffer. */
4362 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4363 {
4364 if (startpos > 0)
4365 return -1;
4366 else
4367 range = 0;
4368 }
4369
4370 #ifdef emacs
4371 /* In a forward search for something that starts with \=.
4372 don't keep searching past point. */
4373 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4374 {
4375 range = PT_BYTE - BEGV_BYTE - startpos;
4376 if (range < 0)
4377 return -1;
4378 }
4379 #endif /* emacs */
4380
4381 /* Update the fastmap now if not correct already. */
4382 if (fastmap && !bufp->fastmap_accurate)
4383 re_compile_fastmap (bufp);
4384
4385 /* See whether the pattern is anchored. */
4386 anchored_start = (bufp->buffer[0] == begline);
4387
4388 #ifdef emacs
4389 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4390 {
4391 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4392
4393 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4394 }
4395 #endif
4396
4397 /* Loop through the string, looking for a place to start matching. */
4398 for (;;)
4399 {
4400 /* If the pattern is anchored,
4401 skip quickly past places we cannot match.
4402 We don't bother to treat startpos == 0 specially
4403 because that case doesn't repeat. */
4404 if (anchored_start && startpos > 0)
4405 {
4406 if (! ((startpos <= size1 ? string1[startpos - 1]
4407 : string2[startpos - size1 - 1])
4408 == '\n'))
4409 goto advance;
4410 }
4411
4412 /* If a fastmap is supplied, skip quickly over characters that
4413 cannot be the start of a match. If the pattern can match the
4414 null string, however, we don't need to skip characters; we want
4415 the first null string. */
4416 if (fastmap && startpos < total_size && !bufp->can_be_null)
4417 {
4418 register re_char *d;
4419 register re_wchar_t buf_ch;
4420
4421 d = POS_ADDR_VSTRING (startpos);
4422
4423 if (range > 0) /* Searching forwards. */
4424 {
4425 register int lim = 0;
4426 ssize_t irange = range;
4427
4428 if (startpos < size1 && startpos + range >= size1)
4429 lim = range - (size1 - startpos);
4430
4431 /* Written out as an if-else to avoid testing `translate'
4432 inside the loop. */
4433 if (RE_TRANSLATE_P (translate))
4434 {
4435 if (multibyte)
4436 while (range > lim)
4437 {
4438 int buf_charlen;
4439
4440 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4441 buf_ch = RE_TRANSLATE (translate, buf_ch);
4442 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4443 break;
4444
4445 range -= buf_charlen;
4446 d += buf_charlen;
4447 }
4448 else
4449 while (range > lim)
4450 {
4451 register re_wchar_t ch, translated;
4452
4453 buf_ch = *d;
4454 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4455 translated = RE_TRANSLATE (translate, ch);
4456 if (translated != ch
4457 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4458 buf_ch = ch;
4459 if (fastmap[buf_ch])
4460 break;
4461 d++;
4462 range--;
4463 }
4464 }
4465 else
4466 {
4467 if (multibyte)
4468 while (range > lim)
4469 {
4470 int buf_charlen;
4471
4472 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4473 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4474 break;
4475 range -= buf_charlen;
4476 d += buf_charlen;
4477 }
4478 else
4479 while (range > lim && !fastmap[*d])
4480 {
4481 d++;
4482 range--;
4483 }
4484 }
4485 startpos += irange - range;
4486 }
4487 else /* Searching backwards. */
4488 {
4489 if (multibyte)
4490 {
4491 buf_ch = STRING_CHAR (d);
4492 buf_ch = TRANSLATE (buf_ch);
4493 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4494 goto advance;
4495 }
4496 else
4497 {
4498 register re_wchar_t ch, translated;
4499
4500 buf_ch = *d;
4501 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4502 translated = TRANSLATE (ch);
4503 if (translated != ch
4504 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4505 buf_ch = ch;
4506 if (! fastmap[TRANSLATE (buf_ch)])
4507 goto advance;
4508 }
4509 }
4510 }
4511
4512 /* If can't match the null string, and that's all we have left, fail. */
4513 if (range >= 0 && startpos == total_size && fastmap
4514 && !bufp->can_be_null)
4515 return -1;
4516
4517 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4518 startpos, regs, stop);
4519
4520 if (val >= 0)
4521 return startpos;
4522
4523 if (val == -2)
4524 return -2;
4525
4526 advance:
4527 if (!range)
4528 break;
4529 else if (range > 0)
4530 {
4531 /* Update STARTPOS to the next character boundary. */
4532 if (multibyte)
4533 {
4534 re_char *p = POS_ADDR_VSTRING (startpos);
4535 int len = BYTES_BY_CHAR_HEAD (*p);
4536
4537 range -= len;
4538 if (range < 0)
4539 break;
4540 startpos += len;
4541 }
4542 else
4543 {
4544 range--;
4545 startpos++;
4546 }
4547 }
4548 else
4549 {
4550 range++;
4551 startpos--;
4552
4553 /* Update STARTPOS to the previous character boundary. */
4554 if (multibyte)
4555 {
4556 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4557 re_char *p0 = p;
4558 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4559
4560 /* Find the head of multibyte form. */
4561 PREV_CHAR_BOUNDARY (p, phead);
4562 range += p0 - 1 - p;
4563 if (range > 0)
4564 break;
4565
4566 startpos -= p0 - 1 - p;
4567 }
4568 }
4569 }
4570 return -1;
4571 } /* re_search_2 */
4572 WEAK_ALIAS (__re_search_2, re_search_2)
4573 \f
4574 /* Declarations and macros for re_match_2. */
4575
4576 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4577 register ssize_t len,
4578 RE_TRANSLATE_TYPE translate,
4579 const int multibyte));
4580
4581 /* This converts PTR, a pointer into one of the search strings `string1'
4582 and `string2' into an offset from the beginning of that string. */
4583 #define POINTER_TO_OFFSET(ptr) \
4584 (FIRST_STRING_P (ptr) \
4585 ? ((regoff_t) ((ptr) - string1)) \
4586 : ((regoff_t) ((ptr) - string2 + size1)))
4587
4588 /* Call before fetching a character with *d. This switches over to
4589 string2 if necessary.
4590 Check re_match_2_internal for a discussion of why end_match_2 might
4591 not be within string2 (but be equal to end_match_1 instead). */
4592 #define PREFETCH() \
4593 while (d == dend) \
4594 { \
4595 /* End of string2 => fail. */ \
4596 if (dend == end_match_2) \
4597 goto fail; \
4598 /* End of string1 => advance to string2. */ \
4599 d = string2; \
4600 dend = end_match_2; \
4601 }
4602
4603 /* Call before fetching a char with *d if you already checked other limits.
4604 This is meant for use in lookahead operations like wordend, etc..
4605 where we might need to look at parts of the string that might be
4606 outside of the LIMITs (i.e past `stop'). */
4607 #define PREFETCH_NOLIMIT() \
4608 if (d == end1) \
4609 { \
4610 d = string2; \
4611 dend = end_match_2; \
4612 } \
4613
4614 /* Test if at very beginning or at very end of the virtual concatenation
4615 of `string1' and `string2'. If only one string, it's `string2'. */
4616 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4617 #define AT_STRINGS_END(d) ((d) == end2)
4618
4619 /* Disabled due to a compiler bug -- see comment at case wordbound */
4620
4621 /* The comment at case wordbound is following one, but we don't use
4622 AT_WORD_BOUNDARY anymore to support multibyte form.
4623
4624 The DEC Alpha C compiler 3.x generates incorrect code for the
4625 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4626 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4627 macro and introducing temporary variables works around the bug. */
4628
4629 #if 0
4630 /* Test if D points to a character which is word-constituent. We have
4631 two special cases to check for: if past the end of string1, look at
4632 the first character in string2; and if before the beginning of
4633 string2, look at the last character in string1. */
4634 #define WORDCHAR_P(d) \
4635 (SYNTAX ((d) == end1 ? *string2 \
4636 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4637 == Sword)
4638
4639 /* Test if the character before D and the one at D differ with respect
4640 to being word-constituent. */
4641 #define AT_WORD_BOUNDARY(d) \
4642 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4643 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4644 #endif
4645
4646 /* Free everything we malloc. */
4647 #ifdef MATCH_MAY_ALLOCATE
4648 # define FREE_VAR(var) \
4649 do { \
4650 if (var) \
4651 { \
4652 REGEX_FREE (var); \
4653 var = NULL; \
4654 } \
4655 } while (0)
4656 # define FREE_VARIABLES() \
4657 do { \
4658 REGEX_FREE_STACK (fail_stack.stack); \
4659 FREE_VAR (regstart); \
4660 FREE_VAR (regend); \
4661 FREE_VAR (best_regstart); \
4662 FREE_VAR (best_regend); \
4663 } while (0)
4664 #else
4665 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4666 #endif /* not MATCH_MAY_ALLOCATE */
4667
4668 \f
4669 /* Optimization routines. */
4670
4671 /* If the operation is a match against one or more chars,
4672 return a pointer to the next operation, else return NULL. */
4673 static re_char *
4674 skip_one_char (const re_char *p)
4675 {
4676 switch (SWITCH_ENUM_CAST (*p++))
4677 {
4678 case anychar:
4679 break;
4680
4681 case exactn:
4682 p += *p + 1;
4683 break;
4684
4685 case charset_not:
4686 case charset:
4687 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4688 {
4689 int mcnt;
4690 p = CHARSET_RANGE_TABLE (p - 1);
4691 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4692 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4693 }
4694 else
4695 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4696 break;
4697
4698 case syntaxspec:
4699 case notsyntaxspec:
4700 #ifdef emacs
4701 case categoryspec:
4702 case notcategoryspec:
4703 #endif /* emacs */
4704 p++;
4705 break;
4706
4707 default:
4708 p = NULL;
4709 }
4710 return p;
4711 }
4712
4713
4714 /* Jump over non-matching operations. */
4715 static re_char *
4716 skip_noops (const re_char *p, const re_char *pend)
4717 {
4718 int mcnt;
4719 while (p < pend)
4720 {
4721 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4722 {
4723 case start_memory:
4724 case stop_memory:
4725 p += 2; break;
4726 case no_op:
4727 p += 1; break;
4728 case jump:
4729 p += 1;
4730 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4731 p += mcnt;
4732 break;
4733 default:
4734 return p;
4735 }
4736 }
4737 assert (p == pend);
4738 return p;
4739 }
4740
4741 /* Non-zero if "p1 matches something" implies "p2 fails". */
4742 static int
4743 mutually_exclusive_p (struct re_pattern_buffer *bufp, const re_char *p1, const re_char *p2)
4744 {
4745 re_opcode_t op2;
4746 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4747 unsigned char *pend = bufp->buffer + bufp->used;
4748
4749 assert (p1 >= bufp->buffer && p1 < pend
4750 && p2 >= bufp->buffer && p2 <= pend);
4751
4752 /* Skip over open/close-group commands.
4753 If what follows this loop is a ...+ construct,
4754 look at what begins its body, since we will have to
4755 match at least one of that. */
4756 p2 = skip_noops (p2, pend);
4757 /* The same skip can be done for p1, except that this function
4758 is only used in the case where p1 is a simple match operator. */
4759 /* p1 = skip_noops (p1, pend); */
4760
4761 assert (p1 >= bufp->buffer && p1 < pend
4762 && p2 >= bufp->buffer && p2 <= pend);
4763
4764 op2 = p2 == pend ? succeed : *p2;
4765
4766 switch (SWITCH_ENUM_CAST (op2))
4767 {
4768 case succeed:
4769 case endbuf:
4770 /* If we're at the end of the pattern, we can change. */
4771 if (skip_one_char (p1))
4772 {
4773 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4774 return 1;
4775 }
4776 break;
4777
4778 case endline:
4779 case exactn:
4780 {
4781 register re_wchar_t c
4782 = (re_opcode_t) *p2 == endline ? '\n'
4783 : RE_STRING_CHAR (p2 + 2, multibyte);
4784
4785 if ((re_opcode_t) *p1 == exactn)
4786 {
4787 if (c != RE_STRING_CHAR (p1 + 2, multibyte))
4788 {
4789 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4790 return 1;
4791 }
4792 }
4793
4794 else if ((re_opcode_t) *p1 == charset
4795 || (re_opcode_t) *p1 == charset_not)
4796 {
4797 int not = (re_opcode_t) *p1 == charset_not;
4798
4799 /* Test if C is listed in charset (or charset_not)
4800 at `p1'. */
4801 if (! multibyte || IS_REAL_ASCII (c))
4802 {
4803 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4804 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4805 not = !not;
4806 }
4807 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4808 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4809
4810 /* `not' is equal to 1 if c would match, which means
4811 that we can't change to pop_failure_jump. */
4812 if (!not)
4813 {
4814 DEBUG_PRINT1 (" No match => fast loop.\n");
4815 return 1;
4816 }
4817 }
4818 else if ((re_opcode_t) *p1 == anychar
4819 && c == '\n')
4820 {
4821 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4822 return 1;
4823 }
4824 }
4825 break;
4826
4827 case charset:
4828 {
4829 if ((re_opcode_t) *p1 == exactn)
4830 /* Reuse the code above. */
4831 return mutually_exclusive_p (bufp, p2, p1);
4832
4833 /* It is hard to list up all the character in charset
4834 P2 if it includes multibyte character. Give up in
4835 such case. */
4836 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4837 {
4838 /* Now, we are sure that P2 has no range table.
4839 So, for the size of bitmap in P2, `p2[1]' is
4840 enough. But P1 may have range table, so the
4841 size of bitmap table of P1 is extracted by
4842 using macro `CHARSET_BITMAP_SIZE'.
4843
4844 In a multibyte case, we know that all the character
4845 listed in P2 is ASCII. In a unibyte case, P1 has only a
4846 bitmap table. So, in both cases, it is enough to test
4847 only the bitmap table of P1. */
4848
4849 if ((re_opcode_t) *p1 == charset)
4850 {
4851 int idx;
4852 /* We win if the charset inside the loop
4853 has no overlap with the one after the loop. */
4854 for (idx = 0;
4855 (idx < (int) p2[1]
4856 && idx < CHARSET_BITMAP_SIZE (p1));
4857 idx++)
4858 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4859 break;
4860
4861 if (idx == p2[1]
4862 || idx == CHARSET_BITMAP_SIZE (p1))
4863 {
4864 DEBUG_PRINT1 (" No match => fast loop.\n");
4865 return 1;
4866 }
4867 }
4868 else if ((re_opcode_t) *p1 == charset_not)
4869 {
4870 int idx;
4871 /* We win if the charset_not inside the loop lists
4872 every character listed in the charset after. */
4873 for (idx = 0; idx < (int) p2[1]; idx++)
4874 if (! (p2[2 + idx] == 0
4875 || (idx < CHARSET_BITMAP_SIZE (p1)
4876 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4877 break;
4878
4879 if (idx == p2[1])
4880 {
4881 DEBUG_PRINT1 (" No match => fast loop.\n");
4882 return 1;
4883 }
4884 }
4885 }
4886 }
4887 break;
4888
4889 case charset_not:
4890 switch (SWITCH_ENUM_CAST (*p1))
4891 {
4892 case exactn:
4893 case charset:
4894 /* Reuse the code above. */
4895 return mutually_exclusive_p (bufp, p2, p1);
4896 case charset_not:
4897 /* When we have two charset_not, it's very unlikely that
4898 they don't overlap. The union of the two sets of excluded
4899 chars should cover all possible chars, which, as a matter of
4900 fact, is virtually impossible in multibyte buffers. */
4901 break;
4902 }
4903 break;
4904
4905 case wordend:
4906 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
4907 case symend:
4908 return ((re_opcode_t) *p1 == syntaxspec
4909 && (p1[1] == Ssymbol || p1[1] == Sword));
4910 case notsyntaxspec:
4911 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
4912
4913 case wordbeg:
4914 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
4915 case symbeg:
4916 return ((re_opcode_t) *p1 == notsyntaxspec
4917 && (p1[1] == Ssymbol || p1[1] == Sword));
4918 case syntaxspec:
4919 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
4920
4921 case wordbound:
4922 return (((re_opcode_t) *p1 == notsyntaxspec
4923 || (re_opcode_t) *p1 == syntaxspec)
4924 && p1[1] == Sword);
4925
4926 #ifdef emacs
4927 case categoryspec:
4928 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4929 case notcategoryspec:
4930 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4931 #endif /* emacs */
4932
4933 default:
4934 ;
4935 }
4936
4937 /* Safe default. */
4938 return 0;
4939 }
4940
4941 \f
4942 /* Matching routines. */
4943
4944 #ifndef emacs /* Emacs never uses this. */
4945 /* re_match is like re_match_2 except it takes only a single string. */
4946
4947 regoff_t
4948 re_match (struct re_pattern_buffer *bufp, const char *string,
4949 size_t size, ssize_t pos, struct re_registers *regs)
4950 {
4951 regoff_t result = re_match_2_internal (bufp, NULL, 0, (re_char*) string,
4952 size, pos, regs, size);
4953 return result;
4954 }
4955 WEAK_ALIAS (__re_match, re_match)
4956 #endif /* not emacs */
4957
4958 #ifdef emacs
4959 /* In Emacs, this is the string or buffer in which we
4960 are matching. It is used for looking up syntax properties. */
4961 Lisp_Object re_match_object;
4962 #endif
4963
4964 /* re_match_2 matches the compiled pattern in BUFP against the
4965 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4966 and SIZE2, respectively). We start matching at POS, and stop
4967 matching at STOP.
4968
4969 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4970 store offsets for the substring each group matched in REGS. See the
4971 documentation for exactly how many groups we fill.
4972
4973 We return -1 if no match, -2 if an internal error (such as the
4974 failure stack overflowing). Otherwise, we return the length of the
4975 matched substring. */
4976
4977 regoff_t
4978 re_match_2 (struct re_pattern_buffer *bufp, const char *string1,
4979 size_t size1, const char *string2, size_t size2, ssize_t pos,
4980 struct re_registers *regs, ssize_t stop)
4981 {
4982 regoff_t result;
4983
4984 #ifdef emacs
4985 ssize_t charpos;
4986 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4987 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4988 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4989 #endif
4990
4991 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4992 (re_char*) string2, size2,
4993 pos, regs, stop);
4994 return result;
4995 }
4996 WEAK_ALIAS (__re_match_2, re_match_2)
4997
4998
4999 /* This is a separate function so that we can force an alloca cleanup
5000 afterwards. */
5001 static regoff_t
5002 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1,
5003 size_t size1, const re_char *string2, size_t size2,
5004 ssize_t pos, struct re_registers *regs, ssize_t stop)
5005 {
5006 /* General temporaries. */
5007 ssize_t mcnt;
5008 size_t reg;
5009
5010 /* Just past the end of the corresponding string. */
5011 re_char *end1, *end2;
5012
5013 /* Pointers into string1 and string2, just past the last characters in
5014 each to consider matching. */
5015 re_char *end_match_1, *end_match_2;
5016
5017 /* Where we are in the data, and the end of the current string. */
5018 re_char *d, *dend;
5019
5020 /* Used sometimes to remember where we were before starting matching
5021 an operator so that we can go back in case of failure. This "atomic"
5022 behavior of matching opcodes is indispensable to the correctness
5023 of the on_failure_keep_string_jump optimization. */
5024 re_char *dfail;
5025
5026 /* Where we are in the pattern, and the end of the pattern. */
5027 re_char *p = bufp->buffer;
5028 re_char *pend = p + bufp->used;
5029
5030 /* We use this to map every character in the string. */
5031 RE_TRANSLATE_TYPE translate = bufp->translate;
5032
5033 /* Nonzero if BUFP is setup from a multibyte regex. */
5034 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5035
5036 /* Nonzero if STRING1/STRING2 are multibyte. */
5037 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5038
5039 /* Failure point stack. Each place that can handle a failure further
5040 down the line pushes a failure point on this stack. It consists of
5041 regstart, and regend for all registers corresponding to
5042 the subexpressions we're currently inside, plus the number of such
5043 registers, and, finally, two char *'s. The first char * is where
5044 to resume scanning the pattern; the second one is where to resume
5045 scanning the strings. */
5046 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5047 fail_stack_type fail_stack;
5048 #endif
5049 #ifdef DEBUG
5050 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5051 #endif
5052
5053 #if defined REL_ALLOC && defined REGEX_MALLOC
5054 /* This holds the pointer to the failure stack, when
5055 it is allocated relocatably. */
5056 fail_stack_elt_t *failure_stack_ptr;
5057 #endif
5058
5059 /* We fill all the registers internally, independent of what we
5060 return, for use in backreferences. The number here includes
5061 an element for register zero. */
5062 size_t num_regs = bufp->re_nsub + 1;
5063
5064 /* Information on the contents of registers. These are pointers into
5065 the input strings; they record just what was matched (on this
5066 attempt) by a subexpression part of the pattern, that is, the
5067 regnum-th regstart pointer points to where in the pattern we began
5068 matching and the regnum-th regend points to right after where we
5069 stopped matching the regnum-th subexpression. (The zeroth register
5070 keeps track of what the whole pattern matches.) */
5071 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5072 re_char **regstart, **regend;
5073 #endif
5074
5075 /* The following record the register info as found in the above
5076 variables when we find a match better than any we've seen before.
5077 This happens as we backtrack through the failure points, which in
5078 turn happens only if we have not yet matched the entire string. */
5079 unsigned best_regs_set = false;
5080 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5081 re_char **best_regstart, **best_regend;
5082 #endif
5083
5084 /* Logically, this is `best_regend[0]'. But we don't want to have to
5085 allocate space for that if we're not allocating space for anything
5086 else (see below). Also, we never need info about register 0 for
5087 any of the other register vectors, and it seems rather a kludge to
5088 treat `best_regend' differently than the rest. So we keep track of
5089 the end of the best match so far in a separate variable. We
5090 initialize this to NULL so that when we backtrack the first time
5091 and need to test it, it's not garbage. */
5092 re_char *match_end = NULL;
5093
5094 #ifdef DEBUG
5095 /* Counts the total number of registers pushed. */
5096 unsigned num_regs_pushed = 0;
5097 #endif
5098
5099 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5100
5101 INIT_FAIL_STACK ();
5102
5103 #ifdef MATCH_MAY_ALLOCATE
5104 /* Do not bother to initialize all the register variables if there are
5105 no groups in the pattern, as it takes a fair amount of time. If
5106 there are groups, we include space for register 0 (the whole
5107 pattern), even though we never use it, since it simplifies the
5108 array indexing. We should fix this. */
5109 if (bufp->re_nsub)
5110 {
5111 regstart = REGEX_TALLOC (num_regs, re_char *);
5112 regend = REGEX_TALLOC (num_regs, re_char *);
5113 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5114 best_regend = REGEX_TALLOC (num_regs, re_char *);
5115
5116 if (!(regstart && regend && best_regstart && best_regend))
5117 {
5118 FREE_VARIABLES ();
5119 return -2;
5120 }
5121 }
5122 else
5123 {
5124 /* We must initialize all our variables to NULL, so that
5125 `FREE_VARIABLES' doesn't try to free them. */
5126 regstart = regend = best_regstart = best_regend = NULL;
5127 }
5128 #endif /* MATCH_MAY_ALLOCATE */
5129
5130 /* The starting position is bogus. */
5131 if (pos < 0 || pos > size1 + size2)
5132 {
5133 FREE_VARIABLES ();
5134 return -1;
5135 }
5136
5137 /* Initialize subexpression text positions to -1 to mark ones that no
5138 start_memory/stop_memory has been seen for. Also initialize the
5139 register information struct. */
5140 for (reg = 1; reg < num_regs; reg++)
5141 regstart[reg] = regend[reg] = NULL;
5142
5143 /* We move `string1' into `string2' if the latter's empty -- but not if
5144 `string1' is null. */
5145 if (size2 == 0 && string1 != NULL)
5146 {
5147 string2 = string1;
5148 size2 = size1;
5149 string1 = 0;
5150 size1 = 0;
5151 }
5152 end1 = string1 + size1;
5153 end2 = string2 + size2;
5154
5155 /* `p' scans through the pattern as `d' scans through the data.
5156 `dend' is the end of the input string that `d' points within. `d'
5157 is advanced into the following input string whenever necessary, but
5158 this happens before fetching; therefore, at the beginning of the
5159 loop, `d' can be pointing at the end of a string, but it cannot
5160 equal `string2'. */
5161 if (pos >= size1)
5162 {
5163 /* Only match within string2. */
5164 d = string2 + pos - size1;
5165 dend = end_match_2 = string2 + stop - size1;
5166 end_match_1 = end1; /* Just to give it a value. */
5167 }
5168 else
5169 {
5170 if (stop < size1)
5171 {
5172 /* Only match within string1. */
5173 end_match_1 = string1 + stop;
5174 /* BEWARE!
5175 When we reach end_match_1, PREFETCH normally switches to string2.
5176 But in the present case, this means that just doing a PREFETCH
5177 makes us jump from `stop' to `gap' within the string.
5178 What we really want here is for the search to stop as
5179 soon as we hit end_match_1. That's why we set end_match_2
5180 to end_match_1 (since PREFETCH fails as soon as we hit
5181 end_match_2). */
5182 end_match_2 = end_match_1;
5183 }
5184 else
5185 { /* It's important to use this code when stop == size so that
5186 moving `d' from end1 to string2 will not prevent the d == dend
5187 check from catching the end of string. */
5188 end_match_1 = end1;
5189 end_match_2 = string2 + stop - size1;
5190 }
5191 d = string1 + pos;
5192 dend = end_match_1;
5193 }
5194
5195 DEBUG_PRINT1 ("The compiled pattern is: ");
5196 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5197 DEBUG_PRINT1 ("The string to match is: `");
5198 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5199 DEBUG_PRINT1 ("'\n");
5200
5201 /* This loops over pattern commands. It exits by returning from the
5202 function if the match is complete, or it drops through if the match
5203 fails at this starting point in the input data. */
5204 for (;;)
5205 {
5206 DEBUG_PRINT2 ("\n%p: ", p);
5207
5208 if (p == pend)
5209 { /* End of pattern means we might have succeeded. */
5210 DEBUG_PRINT1 ("end of pattern ... ");
5211
5212 /* If we haven't matched the entire string, and we want the
5213 longest match, try backtracking. */
5214 if (d != end_match_2)
5215 {
5216 /* 1 if this match ends in the same string (string1 or string2)
5217 as the best previous match. */
5218 boolean same_str_p = (FIRST_STRING_P (match_end)
5219 == FIRST_STRING_P (d));
5220 /* 1 if this match is the best seen so far. */
5221 boolean best_match_p;
5222
5223 /* AIX compiler got confused when this was combined
5224 with the previous declaration. */
5225 if (same_str_p)
5226 best_match_p = d > match_end;
5227 else
5228 best_match_p = !FIRST_STRING_P (d);
5229
5230 DEBUG_PRINT1 ("backtracking.\n");
5231
5232 if (!FAIL_STACK_EMPTY ())
5233 { /* More failure points to try. */
5234
5235 /* If exceeds best match so far, save it. */
5236 if (!best_regs_set || best_match_p)
5237 {
5238 best_regs_set = true;
5239 match_end = d;
5240
5241 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5242
5243 for (reg = 1; reg < num_regs; reg++)
5244 {
5245 best_regstart[reg] = regstart[reg];
5246 best_regend[reg] = regend[reg];
5247 }
5248 }
5249 goto fail;
5250 }
5251
5252 /* If no failure points, don't restore garbage. And if
5253 last match is real best match, don't restore second
5254 best one. */
5255 else if (best_regs_set && !best_match_p)
5256 {
5257 restore_best_regs:
5258 /* Restore best match. It may happen that `dend ==
5259 end_match_1' while the restored d is in string2.
5260 For example, the pattern `x.*y.*z' against the
5261 strings `x-' and `y-z-', if the two strings are
5262 not consecutive in memory. */
5263 DEBUG_PRINT1 ("Restoring best registers.\n");
5264
5265 d = match_end;
5266 dend = ((d >= string1 && d <= end1)
5267 ? end_match_1 : end_match_2);
5268
5269 for (reg = 1; reg < num_regs; reg++)
5270 {
5271 regstart[reg] = best_regstart[reg];
5272 regend[reg] = best_regend[reg];
5273 }
5274 }
5275 } /* d != end_match_2 */
5276
5277 succeed_label:
5278 DEBUG_PRINT1 ("Accepting match.\n");
5279
5280 /* If caller wants register contents data back, do it. */
5281 if (regs && !bufp->no_sub)
5282 {
5283 /* Have the register data arrays been allocated? */
5284 if (bufp->regs_allocated == REGS_UNALLOCATED)
5285 { /* No. So allocate them with malloc. We need one
5286 extra element beyond `num_regs' for the `-1' marker
5287 GNU code uses. */
5288 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5289 regs->start = TALLOC (regs->num_regs, regoff_t);
5290 regs->end = TALLOC (regs->num_regs, regoff_t);
5291 if (regs->start == NULL || regs->end == NULL)
5292 {
5293 FREE_VARIABLES ();
5294 return -2;
5295 }
5296 bufp->regs_allocated = REGS_REALLOCATE;
5297 }
5298 else if (bufp->regs_allocated == REGS_REALLOCATE)
5299 { /* Yes. If we need more elements than were already
5300 allocated, reallocate them. If we need fewer, just
5301 leave it alone. */
5302 if (regs->num_regs < num_regs + 1)
5303 {
5304 regs->num_regs = num_regs + 1;
5305 RETALLOC (regs->start, regs->num_regs, regoff_t);
5306 RETALLOC (regs->end, regs->num_regs, regoff_t);
5307 if (regs->start == NULL || regs->end == NULL)
5308 {
5309 FREE_VARIABLES ();
5310 return -2;
5311 }
5312 }
5313 }
5314 else
5315 {
5316 /* These braces fend off a "empty body in an else-statement"
5317 warning under GCC when assert expands to nothing. */
5318 assert (bufp->regs_allocated == REGS_FIXED);
5319 }
5320
5321 /* Convert the pointer data in `regstart' and `regend' to
5322 indices. Register zero has to be set differently,
5323 since we haven't kept track of any info for it. */
5324 if (regs->num_regs > 0)
5325 {
5326 regs->start[0] = pos;
5327 regs->end[0] = POINTER_TO_OFFSET (d);
5328 }
5329
5330 /* Go through the first `min (num_regs, regs->num_regs)'
5331 registers, since that is all we initialized. */
5332 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5333 {
5334 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5335 regs->start[reg] = regs->end[reg] = -1;
5336 else
5337 {
5338 regs->start[reg]
5339 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5340 regs->end[reg]
5341 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5342 }
5343 }
5344
5345 /* If the regs structure we return has more elements than
5346 were in the pattern, set the extra elements to -1. If
5347 we (re)allocated the registers, this is the case,
5348 because we always allocate enough to have at least one
5349 -1 at the end. */
5350 for (reg = num_regs; reg < regs->num_regs; reg++)
5351 regs->start[reg] = regs->end[reg] = -1;
5352 } /* regs && !bufp->no_sub */
5353
5354 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5355 nfailure_points_pushed, nfailure_points_popped,
5356 nfailure_points_pushed - nfailure_points_popped);
5357 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5358
5359 mcnt = POINTER_TO_OFFSET (d) - pos;
5360
5361 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5362
5363 FREE_VARIABLES ();
5364 return mcnt;
5365 }
5366
5367 /* Otherwise match next pattern command. */
5368 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5369 {
5370 /* Ignore these. Used to ignore the n of succeed_n's which
5371 currently have n == 0. */
5372 case no_op:
5373 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5374 break;
5375
5376 case succeed:
5377 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5378 goto succeed_label;
5379
5380 /* Match the next n pattern characters exactly. The following
5381 byte in the pattern defines n, and the n bytes after that
5382 are the characters to match. */
5383 case exactn:
5384 mcnt = *p++;
5385 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5386
5387 /* Remember the start point to rollback upon failure. */
5388 dfail = d;
5389
5390 #ifndef emacs
5391 /* This is written out as an if-else so we don't waste time
5392 testing `translate' inside the loop. */
5393 if (RE_TRANSLATE_P (translate))
5394 do
5395 {
5396 PREFETCH ();
5397 if (RE_TRANSLATE (translate, *d) != *p++)
5398 {
5399 d = dfail;
5400 goto fail;
5401 }
5402 d++;
5403 }
5404 while (--mcnt);
5405 else
5406 do
5407 {
5408 PREFETCH ();
5409 if (*d++ != *p++)
5410 {
5411 d = dfail;
5412 goto fail;
5413 }
5414 }
5415 while (--mcnt);
5416 #else /* emacs */
5417 /* The cost of testing `translate' is comparatively small. */
5418 if (target_multibyte)
5419 do
5420 {
5421 int pat_charlen, buf_charlen;
5422 int pat_ch, buf_ch;
5423
5424 PREFETCH ();
5425 if (multibyte)
5426 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5427 else
5428 {
5429 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5430 pat_charlen = 1;
5431 }
5432 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5433
5434 if (TRANSLATE (buf_ch) != pat_ch)
5435 {
5436 d = dfail;
5437 goto fail;
5438 }
5439
5440 p += pat_charlen;
5441 d += buf_charlen;
5442 mcnt -= pat_charlen;
5443 }
5444 while (mcnt > 0);
5445 else
5446 do
5447 {
5448 int pat_charlen;
5449 int pat_ch, buf_ch;
5450
5451 PREFETCH ();
5452 if (multibyte)
5453 {
5454 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5455 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5456 }
5457 else
5458 {
5459 pat_ch = *p;
5460 pat_charlen = 1;
5461 }
5462 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5463 if (! CHAR_BYTE8_P (buf_ch))
5464 {
5465 buf_ch = TRANSLATE (buf_ch);
5466 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5467 if (buf_ch < 0)
5468 buf_ch = *d;
5469 }
5470 else
5471 buf_ch = *d;
5472 if (buf_ch != pat_ch)
5473 {
5474 d = dfail;
5475 goto fail;
5476 }
5477 p += pat_charlen;
5478 d++;
5479 }
5480 while (--mcnt);
5481 #endif
5482 break;
5483
5484
5485 /* Match any character except possibly a newline or a null. */
5486 case anychar:
5487 {
5488 int buf_charlen;
5489 re_wchar_t buf_ch;
5490
5491 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5492
5493 PREFETCH ();
5494 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5495 target_multibyte);
5496 buf_ch = TRANSLATE (buf_ch);
5497
5498 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5499 && buf_ch == '\n')
5500 || ((bufp->syntax & RE_DOT_NOT_NULL)
5501 && buf_ch == '\000'))
5502 goto fail;
5503
5504 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5505 d += buf_charlen;
5506 }
5507 break;
5508
5509
5510 case charset:
5511 case charset_not:
5512 {
5513 register unsigned int c;
5514 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5515 int len;
5516
5517 /* Start of actual range_table, or end of bitmap if there is no
5518 range table. */
5519 re_char *range_table IF_LINT (= NULL);
5520
5521 /* Nonzero if there is a range table. */
5522 int range_table_exists;
5523
5524 /* Number of ranges of range table. This is not included
5525 in the initial byte-length of the command. */
5526 int count = 0;
5527
5528 /* Whether matching against a unibyte character. */
5529 boolean unibyte_char = false;
5530
5531 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5532
5533 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5534
5535 if (range_table_exists)
5536 {
5537 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5538 EXTRACT_NUMBER_AND_INCR (count, range_table);
5539 }
5540
5541 PREFETCH ();
5542 c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5543 if (target_multibyte)
5544 {
5545 int c1;
5546
5547 c = TRANSLATE (c);
5548 c1 = RE_CHAR_TO_UNIBYTE (c);
5549 if (c1 >= 0)
5550 {
5551 unibyte_char = true;
5552 c = c1;
5553 }
5554 }
5555 else
5556 {
5557 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5558
5559 if (! CHAR_BYTE8_P (c1))
5560 {
5561 c1 = TRANSLATE (c1);
5562 c1 = RE_CHAR_TO_UNIBYTE (c1);
5563 if (c1 >= 0)
5564 {
5565 unibyte_char = true;
5566 c = c1;
5567 }
5568 }
5569 else
5570 unibyte_char = true;
5571 }
5572
5573 if (unibyte_char && c < (1 << BYTEWIDTH))
5574 { /* Lookup bitmap. */
5575 /* Cast to `unsigned' instead of `unsigned char' in
5576 case the bit list is a full 32 bytes long. */
5577 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5578 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5579 not = !not;
5580 }
5581 #ifdef emacs
5582 else if (range_table_exists)
5583 {
5584 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5585
5586 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5587 | (class_bits & BIT_MULTIBYTE)
5588 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5589 | (class_bits & BIT_SPACE && ISSPACE (c))
5590 | (class_bits & BIT_UPPER && ISUPPER (c))
5591 | (class_bits & BIT_WORD && ISWORD (c)))
5592 not = !not;
5593 else
5594 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5595 }
5596 #endif /* emacs */
5597
5598 if (range_table_exists)
5599 p = CHARSET_RANGE_TABLE_END (range_table, count);
5600 else
5601 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5602
5603 if (!not) goto fail;
5604
5605 d += len;
5606 }
5607 break;
5608
5609
5610 /* The beginning of a group is represented by start_memory.
5611 The argument is the register number. The text
5612 matched within the group is recorded (in the internal
5613 registers data structure) under the register number. */
5614 case start_memory:
5615 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5616
5617 /* In case we need to undo this operation (via backtracking). */
5618 PUSH_FAILURE_REG ((unsigned int)*p);
5619
5620 regstart[*p] = d;
5621 regend[*p] = NULL; /* probably unnecessary. -sm */
5622 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5623
5624 /* Move past the register number and inner group count. */
5625 p += 1;
5626 break;
5627
5628
5629 /* The stop_memory opcode represents the end of a group. Its
5630 argument is the same as start_memory's: the register number. */
5631 case stop_memory:
5632 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5633
5634 assert (!REG_UNSET (regstart[*p]));
5635 /* Strictly speaking, there should be code such as:
5636
5637 assert (REG_UNSET (regend[*p]));
5638 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5639
5640 But the only info to be pushed is regend[*p] and it is known to
5641 be UNSET, so there really isn't anything to push.
5642 Not pushing anything, on the other hand deprives us from the
5643 guarantee that regend[*p] is UNSET since undoing this operation
5644 will not reset its value properly. This is not important since
5645 the value will only be read on the next start_memory or at
5646 the very end and both events can only happen if this stop_memory
5647 is *not* undone. */
5648
5649 regend[*p] = d;
5650 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5651
5652 /* Move past the register number and the inner group count. */
5653 p += 1;
5654 break;
5655
5656
5657 /* \<digit> has been turned into a `duplicate' command which is
5658 followed by the numeric value of <digit> as the register number. */
5659 case duplicate:
5660 {
5661 register re_char *d2, *dend2;
5662 int regno = *p++; /* Get which register to match against. */
5663 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5664
5665 /* Can't back reference a group which we've never matched. */
5666 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5667 goto fail;
5668
5669 /* Where in input to try to start matching. */
5670 d2 = regstart[regno];
5671
5672 /* Remember the start point to rollback upon failure. */
5673 dfail = d;
5674
5675 /* Where to stop matching; if both the place to start and
5676 the place to stop matching are in the same string, then
5677 set to the place to stop, otherwise, for now have to use
5678 the end of the first string. */
5679
5680 dend2 = ((FIRST_STRING_P (regstart[regno])
5681 == FIRST_STRING_P (regend[regno]))
5682 ? regend[regno] : end_match_1);
5683 for (;;)
5684 {
5685 /* If necessary, advance to next segment in register
5686 contents. */
5687 while (d2 == dend2)
5688 {
5689 if (dend2 == end_match_2) break;
5690 if (dend2 == regend[regno]) break;
5691
5692 /* End of string1 => advance to string2. */
5693 d2 = string2;
5694 dend2 = regend[regno];
5695 }
5696 /* At end of register contents => success */
5697 if (d2 == dend2) break;
5698
5699 /* If necessary, advance to next segment in data. */
5700 PREFETCH ();
5701
5702 /* How many characters left in this segment to match. */
5703 mcnt = dend - d;
5704
5705 /* Want how many consecutive characters we can match in
5706 one shot, so, if necessary, adjust the count. */
5707 if (mcnt > dend2 - d2)
5708 mcnt = dend2 - d2;
5709
5710 /* Compare that many; failure if mismatch, else move
5711 past them. */
5712 if (RE_TRANSLATE_P (translate)
5713 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5714 : memcmp (d, d2, mcnt))
5715 {
5716 d = dfail;
5717 goto fail;
5718 }
5719 d += mcnt, d2 += mcnt;
5720 }
5721 }
5722 break;
5723
5724
5725 /* begline matches the empty string at the beginning of the string
5726 (unless `not_bol' is set in `bufp'), and after newlines. */
5727 case begline:
5728 DEBUG_PRINT1 ("EXECUTING begline.\n");
5729
5730 if (AT_STRINGS_BEG (d))
5731 {
5732 if (!bufp->not_bol) break;
5733 }
5734 else
5735 {
5736 unsigned c;
5737 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5738 if (c == '\n')
5739 break;
5740 }
5741 /* In all other cases, we fail. */
5742 goto fail;
5743
5744
5745 /* endline is the dual of begline. */
5746 case endline:
5747 DEBUG_PRINT1 ("EXECUTING endline.\n");
5748
5749 if (AT_STRINGS_END (d))
5750 {
5751 if (!bufp->not_eol) break;
5752 }
5753 else
5754 {
5755 PREFETCH_NOLIMIT ();
5756 if (*d == '\n')
5757 break;
5758 }
5759 goto fail;
5760
5761
5762 /* Match at the very beginning of the data. */
5763 case begbuf:
5764 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5765 if (AT_STRINGS_BEG (d))
5766 break;
5767 goto fail;
5768
5769
5770 /* Match at the very end of the data. */
5771 case endbuf:
5772 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5773 if (AT_STRINGS_END (d))
5774 break;
5775 goto fail;
5776
5777
5778 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5779 pushes NULL as the value for the string on the stack. Then
5780 `POP_FAILURE_POINT' will keep the current value for the
5781 string, instead of restoring it. To see why, consider
5782 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5783 then the . fails against the \n. But the next thing we want
5784 to do is match the \n against the \n; if we restored the
5785 string value, we would be back at the foo.
5786
5787 Because this is used only in specific cases, we don't need to
5788 check all the things that `on_failure_jump' does, to make
5789 sure the right things get saved on the stack. Hence we don't
5790 share its code. The only reason to push anything on the
5791 stack at all is that otherwise we would have to change
5792 `anychar's code to do something besides goto fail in this
5793 case; that seems worse than this. */
5794 case on_failure_keep_string_jump:
5795 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5796 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5797 mcnt, p + mcnt);
5798
5799 PUSH_FAILURE_POINT (p - 3, NULL);
5800 break;
5801
5802 /* A nasty loop is introduced by the non-greedy *? and +?.
5803 With such loops, the stack only ever contains one failure point
5804 at a time, so that a plain on_failure_jump_loop kind of
5805 cycle detection cannot work. Worse yet, such a detection
5806 can not only fail to detect a cycle, but it can also wrongly
5807 detect a cycle (between different instantiations of the same
5808 loop).
5809 So the method used for those nasty loops is a little different:
5810 We use a special cycle-detection-stack-frame which is pushed
5811 when the on_failure_jump_nastyloop failure-point is *popped*.
5812 This special frame thus marks the beginning of one iteration
5813 through the loop and we can hence easily check right here
5814 whether something matched between the beginning and the end of
5815 the loop. */
5816 case on_failure_jump_nastyloop:
5817 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5818 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5819 mcnt, p + mcnt);
5820
5821 assert ((re_opcode_t)p[-4] == no_op);
5822 {
5823 int cycle = 0;
5824 CHECK_INFINITE_LOOP (p - 4, d);
5825 if (!cycle)
5826 /* If there's a cycle, just continue without pushing
5827 this failure point. The failure point is the "try again"
5828 option, which shouldn't be tried.
5829 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5830 PUSH_FAILURE_POINT (p - 3, d);
5831 }
5832 break;
5833
5834 /* Simple loop detecting on_failure_jump: just check on the
5835 failure stack if the same spot was already hit earlier. */
5836 case on_failure_jump_loop:
5837 on_failure:
5838 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5839 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5840 mcnt, p + mcnt);
5841 {
5842 int cycle = 0;
5843 CHECK_INFINITE_LOOP (p - 3, d);
5844 if (cycle)
5845 /* If there's a cycle, get out of the loop, as if the matching
5846 had failed. We used to just `goto fail' here, but that was
5847 aborting the search a bit too early: we want to keep the
5848 empty-loop-match and keep matching after the loop.
5849 We want (x?)*y\1z to match both xxyz and xxyxz. */
5850 p += mcnt;
5851 else
5852 PUSH_FAILURE_POINT (p - 3, d);
5853 }
5854 break;
5855
5856
5857 /* Uses of on_failure_jump:
5858
5859 Each alternative starts with an on_failure_jump that points
5860 to the beginning of the next alternative. Each alternative
5861 except the last ends with a jump that in effect jumps past
5862 the rest of the alternatives. (They really jump to the
5863 ending jump of the following alternative, because tensioning
5864 these jumps is a hassle.)
5865
5866 Repeats start with an on_failure_jump that points past both
5867 the repetition text and either the following jump or
5868 pop_failure_jump back to this on_failure_jump. */
5869 case on_failure_jump:
5870 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5871 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5872 mcnt, p + mcnt);
5873
5874 PUSH_FAILURE_POINT (p -3, d);
5875 break;
5876
5877 /* This operation is used for greedy *.
5878 Compare the beginning of the repeat with what in the
5879 pattern follows its end. If we can establish that there
5880 is nothing that they would both match, i.e., that we
5881 would have to backtrack because of (as in, e.g., `a*a')
5882 then we can use a non-backtracking loop based on
5883 on_failure_keep_string_jump instead of on_failure_jump. */
5884 case on_failure_jump_smart:
5885 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5886 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5887 mcnt, p + mcnt);
5888 {
5889 re_char *p1 = p; /* Next operation. */
5890 /* Here, we discard `const', making re_match non-reentrant. */
5891 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5892 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5893
5894 p -= 3; /* Reset so that we will re-execute the
5895 instruction once it's been changed. */
5896
5897 EXTRACT_NUMBER (mcnt, p2 - 2);
5898
5899 /* Ensure this is a indeed the trivial kind of loop
5900 we are expecting. */
5901 assert (skip_one_char (p1) == p2 - 3);
5902 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5903 DEBUG_STATEMENT (debug += 2);
5904 if (mutually_exclusive_p (bufp, p1, p2))
5905 {
5906 /* Use a fast `on_failure_keep_string_jump' loop. */
5907 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5908 *p3 = (unsigned char) on_failure_keep_string_jump;
5909 STORE_NUMBER (p2 - 2, mcnt + 3);
5910 }
5911 else
5912 {
5913 /* Default to a safe `on_failure_jump' loop. */
5914 DEBUG_PRINT1 (" smart default => slow loop.\n");
5915 *p3 = (unsigned char) on_failure_jump;
5916 }
5917 DEBUG_STATEMENT (debug -= 2);
5918 }
5919 break;
5920
5921 /* Unconditionally jump (without popping any failure points). */
5922 case jump:
5923 unconditional_jump:
5924 IMMEDIATE_QUIT_CHECK;
5925 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5926 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5927 p += mcnt; /* Do the jump. */
5928 DEBUG_PRINT2 ("(to %p).\n", p);
5929 break;
5930
5931
5932 /* Have to succeed matching what follows at least n times.
5933 After that, handle like `on_failure_jump'. */
5934 case succeed_n:
5935 /* Signedness doesn't matter since we only compare MCNT to 0. */
5936 EXTRACT_NUMBER (mcnt, p + 2);
5937 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5938
5939 /* Originally, mcnt is how many times we HAVE to succeed. */
5940 if (mcnt != 0)
5941 {
5942 /* Here, we discard `const', making re_match non-reentrant. */
5943 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5944 mcnt--;
5945 p += 4;
5946 PUSH_NUMBER (p2, mcnt);
5947 }
5948 else
5949 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5950 goto on_failure;
5951 break;
5952
5953 case jump_n:
5954 /* Signedness doesn't matter since we only compare MCNT to 0. */
5955 EXTRACT_NUMBER (mcnt, p + 2);
5956 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5957
5958 /* Originally, this is how many times we CAN jump. */
5959 if (mcnt != 0)
5960 {
5961 /* Here, we discard `const', making re_match non-reentrant. */
5962 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5963 mcnt--;
5964 PUSH_NUMBER (p2, mcnt);
5965 goto unconditional_jump;
5966 }
5967 /* If don't have to jump any more, skip over the rest of command. */
5968 else
5969 p += 4;
5970 break;
5971
5972 case set_number_at:
5973 {
5974 unsigned char *p2; /* Location of the counter. */
5975 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5976
5977 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5978 /* Here, we discard `const', making re_match non-reentrant. */
5979 p2 = (unsigned char*) p + mcnt;
5980 /* Signedness doesn't matter since we only copy MCNT's bits . */
5981 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5982 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
5983 PUSH_NUMBER (p2, mcnt);
5984 break;
5985 }
5986
5987 case wordbound:
5988 case notwordbound:
5989 {
5990 boolean not = (re_opcode_t) *(p - 1) == notwordbound;
5991 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5992
5993 /* We SUCCEED (or FAIL) in one of the following cases: */
5994
5995 /* Case 1: D is at the beginning or the end of string. */
5996 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5997 not = !not;
5998 else
5999 {
6000 /* C1 is the character before D, S1 is the syntax of C1, C2
6001 is the character at D, and S2 is the syntax of C2. */
6002 re_wchar_t c1, c2;
6003 int s1, s2;
6004 int dummy;
6005 #ifdef emacs
6006 ssize_t offset = PTR_TO_OFFSET (d - 1);
6007 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6008 UPDATE_SYNTAX_TABLE (charpos);
6009 #endif
6010 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6011 s1 = SYNTAX (c1);
6012 #ifdef emacs
6013 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6014 #endif
6015 PREFETCH_NOLIMIT ();
6016 GET_CHAR_AFTER (c2, d, dummy);
6017 s2 = SYNTAX (c2);
6018
6019 if (/* Case 2: Only one of S1 and S2 is Sword. */
6020 ((s1 == Sword) != (s2 == Sword))
6021 /* Case 3: Both of S1 and S2 are Sword, and macro
6022 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6023 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
6024 not = !not;
6025 }
6026 if (not)
6027 break;
6028 else
6029 goto fail;
6030 }
6031
6032 case wordbeg:
6033 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6034
6035 /* We FAIL in one of the following cases: */
6036
6037 /* Case 1: D is at the end of string. */
6038 if (AT_STRINGS_END (d))
6039 goto fail;
6040 else
6041 {
6042 /* C1 is the character before D, S1 is the syntax of C1, C2
6043 is the character at D, and S2 is the syntax of C2. */
6044 re_wchar_t c1, c2;
6045 int s1, s2;
6046 int dummy;
6047 #ifdef emacs
6048 ssize_t offset = PTR_TO_OFFSET (d);
6049 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6050 UPDATE_SYNTAX_TABLE (charpos);
6051 #endif
6052 PREFETCH ();
6053 GET_CHAR_AFTER (c2, d, dummy);
6054 s2 = SYNTAX (c2);
6055
6056 /* Case 2: S2 is not Sword. */
6057 if (s2 != Sword)
6058 goto fail;
6059
6060 /* Case 3: D is not at the beginning of string ... */
6061 if (!AT_STRINGS_BEG (d))
6062 {
6063 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6064 #ifdef emacs
6065 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6066 #endif
6067 s1 = SYNTAX (c1);
6068
6069 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6070 returns 0. */
6071 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6072 goto fail;
6073 }
6074 }
6075 break;
6076
6077 case wordend:
6078 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6079
6080 /* We FAIL in one of the following cases: */
6081
6082 /* Case 1: D is at the beginning of string. */
6083 if (AT_STRINGS_BEG (d))
6084 goto fail;
6085 else
6086 {
6087 /* C1 is the character before D, S1 is the syntax of C1, C2
6088 is the character at D, and S2 is the syntax of C2. */
6089 re_wchar_t c1, c2;
6090 int s1, s2;
6091 int dummy;
6092 #ifdef emacs
6093 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6094 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6095 UPDATE_SYNTAX_TABLE (charpos);
6096 #endif
6097 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6098 s1 = SYNTAX (c1);
6099
6100 /* Case 2: S1 is not Sword. */
6101 if (s1 != Sword)
6102 goto fail;
6103
6104 /* Case 3: D is not at the end of string ... */
6105 if (!AT_STRINGS_END (d))
6106 {
6107 PREFETCH_NOLIMIT ();
6108 GET_CHAR_AFTER (c2, d, dummy);
6109 #ifdef emacs
6110 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6111 #endif
6112 s2 = SYNTAX (c2);
6113
6114 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6115 returns 0. */
6116 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6117 goto fail;
6118 }
6119 }
6120 break;
6121
6122 case symbeg:
6123 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6124
6125 /* We FAIL in one of the following cases: */
6126
6127 /* Case 1: D is at the end of string. */
6128 if (AT_STRINGS_END (d))
6129 goto fail;
6130 else
6131 {
6132 /* C1 is the character before D, S1 is the syntax of C1, C2
6133 is the character at D, and S2 is the syntax of C2. */
6134 re_wchar_t c1, c2;
6135 int s1, s2;
6136 #ifdef emacs
6137 ssize_t offset = PTR_TO_OFFSET (d);
6138 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6139 UPDATE_SYNTAX_TABLE (charpos);
6140 #endif
6141 PREFETCH ();
6142 c2 = RE_STRING_CHAR (d, target_multibyte);
6143 s2 = SYNTAX (c2);
6144
6145 /* Case 2: S2 is neither Sword nor Ssymbol. */
6146 if (s2 != Sword && s2 != Ssymbol)
6147 goto fail;
6148
6149 /* Case 3: D is not at the beginning of string ... */
6150 if (!AT_STRINGS_BEG (d))
6151 {
6152 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6153 #ifdef emacs
6154 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6155 #endif
6156 s1 = SYNTAX (c1);
6157
6158 /* ... and S1 is Sword or Ssymbol. */
6159 if (s1 == Sword || s1 == Ssymbol)
6160 goto fail;
6161 }
6162 }
6163 break;
6164
6165 case symend:
6166 DEBUG_PRINT1 ("EXECUTING symend.\n");
6167
6168 /* We FAIL in one of the following cases: */
6169
6170 /* Case 1: D is at the beginning of string. */
6171 if (AT_STRINGS_BEG (d))
6172 goto fail;
6173 else
6174 {
6175 /* C1 is the character before D, S1 is the syntax of C1, C2
6176 is the character at D, and S2 is the syntax of C2. */
6177 re_wchar_t c1, c2;
6178 int s1, s2;
6179 #ifdef emacs
6180 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6181 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6182 UPDATE_SYNTAX_TABLE (charpos);
6183 #endif
6184 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6185 s1 = SYNTAX (c1);
6186
6187 /* Case 2: S1 is neither Ssymbol nor Sword. */
6188 if (s1 != Sword && s1 != Ssymbol)
6189 goto fail;
6190
6191 /* Case 3: D is not at the end of string ... */
6192 if (!AT_STRINGS_END (d))
6193 {
6194 PREFETCH_NOLIMIT ();
6195 c2 = RE_STRING_CHAR (d, target_multibyte);
6196 #ifdef emacs
6197 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6198 #endif
6199 s2 = SYNTAX (c2);
6200
6201 /* ... and S2 is Sword or Ssymbol. */
6202 if (s2 == Sword || s2 == Ssymbol)
6203 goto fail;
6204 }
6205 }
6206 break;
6207
6208 case syntaxspec:
6209 case notsyntaxspec:
6210 {
6211 boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6212 mcnt = *p++;
6213 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6214 PREFETCH ();
6215 #ifdef emacs
6216 {
6217 ssize_t offset = PTR_TO_OFFSET (d);
6218 ssize_t pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6219 UPDATE_SYNTAX_TABLE (pos1);
6220 }
6221 #endif
6222 {
6223 int len;
6224 re_wchar_t c;
6225
6226 GET_CHAR_AFTER (c, d, len);
6227 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6228 goto fail;
6229 d += len;
6230 }
6231 }
6232 break;
6233
6234 #ifdef emacs
6235 case before_dot:
6236 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6237 if (PTR_BYTE_POS (d) >= PT_BYTE)
6238 goto fail;
6239 break;
6240
6241 case at_dot:
6242 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6243 if (PTR_BYTE_POS (d) != PT_BYTE)
6244 goto fail;
6245 break;
6246
6247 case after_dot:
6248 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6249 if (PTR_BYTE_POS (d) <= PT_BYTE)
6250 goto fail;
6251 break;
6252
6253 case categoryspec:
6254 case notcategoryspec:
6255 {
6256 boolean not = (re_opcode_t) *(p - 1) == notcategoryspec;
6257 mcnt = *p++;
6258 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n",
6259 not?"not":"", mcnt);
6260 PREFETCH ();
6261
6262 {
6263 int len;
6264 re_wchar_t c;
6265 GET_CHAR_AFTER (c, d, len);
6266 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6267 goto fail;
6268 d += len;
6269 }
6270 }
6271 break;
6272
6273 #endif /* emacs */
6274
6275 default:
6276 abort ();
6277 }
6278 continue; /* Successfully executed one pattern command; keep going. */
6279
6280
6281 /* We goto here if a matching operation fails. */
6282 fail:
6283 IMMEDIATE_QUIT_CHECK;
6284 if (!FAIL_STACK_EMPTY ())
6285 {
6286 re_char *str, *pat;
6287 /* A restart point is known. Restore to that state. */
6288 DEBUG_PRINT1 ("\nFAIL:\n");
6289 POP_FAILURE_POINT (str, pat);
6290 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6291 {
6292 case on_failure_keep_string_jump:
6293 assert (str == NULL);
6294 goto continue_failure_jump;
6295
6296 case on_failure_jump_nastyloop:
6297 assert ((re_opcode_t)pat[-2] == no_op);
6298 PUSH_FAILURE_POINT (pat - 2, str);
6299 /* Fallthrough */
6300
6301 case on_failure_jump_loop:
6302 case on_failure_jump:
6303 case succeed_n:
6304 d = str;
6305 continue_failure_jump:
6306 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6307 p = pat + mcnt;
6308 break;
6309
6310 case no_op:
6311 /* A special frame used for nastyloops. */
6312 goto fail;
6313
6314 default:
6315 abort();
6316 }
6317
6318 assert (p >= bufp->buffer && p <= pend);
6319
6320 if (d >= string1 && d <= end1)
6321 dend = end_match_1;
6322 }
6323 else
6324 break; /* Matching at this starting point really fails. */
6325 } /* for (;;) */
6326
6327 if (best_regs_set)
6328 goto restore_best_regs;
6329
6330 FREE_VARIABLES ();
6331
6332 return -1; /* Failure to match. */
6333 } /* re_match_2 */
6334 \f
6335 /* Subroutine definitions for re_match_2. */
6336
6337 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6338 bytes; nonzero otherwise. */
6339
6340 static int
6341 bcmp_translate (const re_char *s1, const re_char *s2, register ssize_t len,
6342 RE_TRANSLATE_TYPE translate, const int target_multibyte)
6343 {
6344 register re_char *p1 = s1, *p2 = s2;
6345 re_char *p1_end = s1 + len;
6346 re_char *p2_end = s2 + len;
6347
6348 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6349 different lengths, but relying on a single `len' would break this. -sm */
6350 while (p1 < p1_end && p2 < p2_end)
6351 {
6352 int p1_charlen, p2_charlen;
6353 re_wchar_t p1_ch, p2_ch;
6354
6355 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6356 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6357
6358 if (RE_TRANSLATE (translate, p1_ch)
6359 != RE_TRANSLATE (translate, p2_ch))
6360 return 1;
6361
6362 p1 += p1_charlen, p2 += p2_charlen;
6363 }
6364
6365 if (p1 != p1_end || p2 != p2_end)
6366 return 1;
6367
6368 return 0;
6369 }
6370 \f
6371 /* Entry points for GNU code. */
6372
6373 /* re_compile_pattern is the GNU regular expression compiler: it
6374 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6375 Returns 0 if the pattern was valid, otherwise an error string.
6376
6377 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6378 are set in BUFP on entry.
6379
6380 We call regex_compile to do the actual compilation. */
6381
6382 const char *
6383 re_compile_pattern (const char *pattern, size_t length,
6384 struct re_pattern_buffer *bufp)
6385 {
6386 reg_errcode_t ret;
6387
6388 /* GNU code is written to assume at least RE_NREGS registers will be set
6389 (and at least one extra will be -1). */
6390 bufp->regs_allocated = REGS_UNALLOCATED;
6391
6392 /* And GNU code determines whether or not to get register information
6393 by passing null for the REGS argument to re_match, etc., not by
6394 setting no_sub. */
6395 bufp->no_sub = 0;
6396
6397 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6398
6399 if (!ret)
6400 return NULL;
6401 return gettext (re_error_msgid[(int) ret]);
6402 }
6403 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6404 \f
6405 /* Entry points compatible with 4.2 BSD regex library. We don't define
6406 them unless specifically requested. */
6407
6408 #if defined _REGEX_RE_COMP || defined _LIBC
6409
6410 /* BSD has one and only one pattern buffer. */
6411 static struct re_pattern_buffer re_comp_buf;
6412
6413 char *
6414 # ifdef _LIBC
6415 /* Make these definitions weak in libc, so POSIX programs can redefine
6416 these names if they don't use our functions, and still use
6417 regcomp/regexec below without link errors. */
6418 weak_function
6419 # endif
6420 re_comp (s)
6421 const char *s;
6422 {
6423 reg_errcode_t ret;
6424
6425 if (!s)
6426 {
6427 if (!re_comp_buf.buffer)
6428 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6429 return (char *) gettext ("No previous regular expression");
6430 return 0;
6431 }
6432
6433 if (!re_comp_buf.buffer)
6434 {
6435 re_comp_buf.buffer = (unsigned char *) malloc (200);
6436 if (re_comp_buf.buffer == NULL)
6437 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6438 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6439 re_comp_buf.allocated = 200;
6440
6441 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6442 if (re_comp_buf.fastmap == NULL)
6443 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6444 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6445 }
6446
6447 /* Since `re_exec' always passes NULL for the `regs' argument, we
6448 don't need to initialize the pattern buffer fields which affect it. */
6449
6450 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6451
6452 if (!ret)
6453 return NULL;
6454
6455 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6456 return (char *) gettext (re_error_msgid[(int) ret]);
6457 }
6458
6459
6460 regoff_t
6461 # ifdef _LIBC
6462 weak_function
6463 # endif
6464 re_exec (const char *s)
6465 {
6466 const size_t len = strlen (s);
6467 return
6468 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6469 }
6470 #endif /* _REGEX_RE_COMP */
6471 \f
6472 /* POSIX.2 functions. Don't define these for Emacs. */
6473
6474 #ifndef emacs
6475
6476 /* regcomp takes a regular expression as a string and compiles it.
6477
6478 PREG is a regex_t *. We do not expect any fields to be initialized,
6479 since POSIX says we shouldn't. Thus, we set
6480
6481 `buffer' to the compiled pattern;
6482 `used' to the length of the compiled pattern;
6483 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6484 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6485 RE_SYNTAX_POSIX_BASIC;
6486 `fastmap' to an allocated space for the fastmap;
6487 `fastmap_accurate' to zero;
6488 `re_nsub' to the number of subexpressions in PATTERN.
6489
6490 PATTERN is the address of the pattern string.
6491
6492 CFLAGS is a series of bits which affect compilation.
6493
6494 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6495 use POSIX basic syntax.
6496
6497 If REG_NEWLINE is set, then . and [^...] don't match newline.
6498 Also, regexec will try a match beginning after every newline.
6499
6500 If REG_ICASE is set, then we considers upper- and lowercase
6501 versions of letters to be equivalent when matching.
6502
6503 If REG_NOSUB is set, then when PREG is passed to regexec, that
6504 routine will report only success or failure, and nothing about the
6505 registers.
6506
6507 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6508 the return codes and their meanings.) */
6509
6510 reg_errcode_t
6511 regcomp (regex_t *__restrict preg, const char *__restrict pattern,
6512 int cflags)
6513 {
6514 reg_errcode_t ret;
6515 reg_syntax_t syntax
6516 = (cflags & REG_EXTENDED) ?
6517 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6518
6519 /* regex_compile will allocate the space for the compiled pattern. */
6520 preg->buffer = 0;
6521 preg->allocated = 0;
6522 preg->used = 0;
6523
6524 /* Try to allocate space for the fastmap. */
6525 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6526
6527 if (cflags & REG_ICASE)
6528 {
6529 unsigned i;
6530
6531 preg->translate
6532 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6533 * sizeof (*(RE_TRANSLATE_TYPE)0));
6534 if (preg->translate == NULL)
6535 return (int) REG_ESPACE;
6536
6537 /* Map uppercase characters to corresponding lowercase ones. */
6538 for (i = 0; i < CHAR_SET_SIZE; i++)
6539 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6540 }
6541 else
6542 preg->translate = NULL;
6543
6544 /* If REG_NEWLINE is set, newlines are treated differently. */
6545 if (cflags & REG_NEWLINE)
6546 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6547 syntax &= ~RE_DOT_NEWLINE;
6548 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6549 }
6550 else
6551 syntax |= RE_NO_NEWLINE_ANCHOR;
6552
6553 preg->no_sub = !!(cflags & REG_NOSUB);
6554
6555 /* POSIX says a null character in the pattern terminates it, so we
6556 can use strlen here in compiling the pattern. */
6557 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6558
6559 /* POSIX doesn't distinguish between an unmatched open-group and an
6560 unmatched close-group: both are REG_EPAREN. */
6561 if (ret == REG_ERPAREN)
6562 ret = REG_EPAREN;
6563
6564 if (ret == REG_NOERROR && preg->fastmap)
6565 { /* Compute the fastmap now, since regexec cannot modify the pattern
6566 buffer. */
6567 re_compile_fastmap (preg);
6568 if (preg->can_be_null)
6569 { /* The fastmap can't be used anyway. */
6570 free (preg->fastmap);
6571 preg->fastmap = NULL;
6572 }
6573 }
6574 return ret;
6575 }
6576 WEAK_ALIAS (__regcomp, regcomp)
6577
6578
6579 /* regexec searches for a given pattern, specified by PREG, in the
6580 string STRING.
6581
6582 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6583 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6584 least NMATCH elements, and we set them to the offsets of the
6585 corresponding matched substrings.
6586
6587 EFLAGS specifies `execution flags' which affect matching: if
6588 REG_NOTBOL is set, then ^ does not match at the beginning of the
6589 string; if REG_NOTEOL is set, then $ does not match at the end.
6590
6591 We return 0 if we find a match and REG_NOMATCH if not. */
6592
6593 reg_errcode_t
6594 regexec (const regex_t *__restrict preg, const char *__restrict string,
6595 size_t nmatch, regmatch_t pmatch[__restrict_arr], int eflags)
6596 {
6597 reg_errcode_t ret;
6598 struct re_registers regs;
6599 regex_t private_preg;
6600 size_t len = strlen (string);
6601 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6602
6603 private_preg = *preg;
6604
6605 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6606 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6607
6608 /* The user has told us exactly how many registers to return
6609 information about, via `nmatch'. We have to pass that on to the
6610 matching routines. */
6611 private_preg.regs_allocated = REGS_FIXED;
6612
6613 if (want_reg_info)
6614 {
6615 regs.num_regs = nmatch;
6616 regs.start = TALLOC (nmatch * 2, regoff_t);
6617 if (regs.start == NULL)
6618 return REG_NOMATCH;
6619 regs.end = regs.start + nmatch;
6620 }
6621
6622 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6623 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6624 was a little bit longer but still only matching the real part.
6625 This works because the `endline' will check for a '\n' and will find a
6626 '\0', correctly deciding that this is not the end of a line.
6627 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6628 a convenient '\0' there. For all we know, the string could be preceded
6629 by '\n' which would throw things off. */
6630
6631 /* Perform the searching operation. */
6632 ret = re_search (&private_preg, string, len,
6633 /* start: */ 0, /* range: */ len,
6634 want_reg_info ? &regs : (struct re_registers *) 0);
6635
6636 /* Copy the register information to the POSIX structure. */
6637 if (want_reg_info)
6638 {
6639 if (ret >= 0)
6640 {
6641 unsigned r;
6642
6643 for (r = 0; r < nmatch; r++)
6644 {
6645 pmatch[r].rm_so = regs.start[r];
6646 pmatch[r].rm_eo = regs.end[r];
6647 }
6648 }
6649
6650 /* If we needed the temporary register info, free the space now. */
6651 free (regs.start);
6652 }
6653
6654 /* We want zero return to mean success, unlike `re_search'. */
6655 return ret >= 0 ? REG_NOERROR : REG_NOMATCH;
6656 }
6657 WEAK_ALIAS (__regexec, regexec)
6658
6659
6660 /* Returns a message corresponding to an error code, ERR_CODE, returned
6661 from either regcomp or regexec. We don't use PREG here.
6662
6663 ERR_CODE was previously called ERRCODE, but that name causes an
6664 error with msvc8 compiler. */
6665
6666 size_t
6667 regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size)
6668 {
6669 const char *msg;
6670 size_t msg_size;
6671
6672 if (err_code < 0
6673 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6674 /* Only error codes returned by the rest of the code should be passed
6675 to this routine. If we are given anything else, or if other regex
6676 code generates an invalid error code, then the program has a bug.
6677 Dump core so we can fix it. */
6678 abort ();
6679
6680 msg = gettext (re_error_msgid[err_code]);
6681
6682 msg_size = strlen (msg) + 1; /* Includes the null. */
6683
6684 if (errbuf_size != 0)
6685 {
6686 if (msg_size > errbuf_size)
6687 {
6688 strncpy (errbuf, msg, errbuf_size - 1);
6689 errbuf[errbuf_size - 1] = 0;
6690 }
6691 else
6692 strcpy (errbuf, msg);
6693 }
6694
6695 return msg_size;
6696 }
6697 WEAK_ALIAS (__regerror, regerror)
6698
6699
6700 /* Free dynamically allocated space used by PREG. */
6701
6702 void
6703 regfree (regex_t *preg)
6704 {
6705 free (preg->buffer);
6706 preg->buffer = NULL;
6707
6708 preg->allocated = 0;
6709 preg->used = 0;
6710
6711 free (preg->fastmap);
6712 preg->fastmap = NULL;
6713 preg->fastmap_accurate = 0;
6714
6715 free (preg->translate);
6716 preg->translate = NULL;
6717 }
6718 WEAK_ALIAS (__regfree, regfree)
6719
6720 #endif /* not emacs */
Emacs repository mirror