HCoop / bpt / emacs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge from trunk.
[bpt/emacs.git] / src / regex.c
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 #include <stddef.h>
41
42 #ifdef emacs
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 # include <string.h>
242
243 /* Define the syntax stuff for \<, \>, etc. */
244
245 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
246 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
247
248 # define SWITCH_ENUM_CAST(x) (x)
249
250 /* Dummy macros for non-Emacs environments. */
251 # define CHAR_CHARSET(c) 0
252 # define CHARSET_LEADING_CODE_BASE(c) 0
253 # define MAX_MULTIBYTE_LENGTH 1
254 # define RE_MULTIBYTE_P(x) 0
255 # define RE_TARGET_MULTIBYTE_P(x) 0
256 # define WORD_BOUNDARY_P(c1, c2) (0)
257 # define CHAR_HEAD_P(p) (1)
258 # define SINGLE_BYTE_CHAR_P(c) (1)
259 # define SAME_CHARSET_P(c1, c2) (1)
260 # define BYTES_BY_CHAR_HEAD(p) (1)
261 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
262 # define STRING_CHAR(p) (*(p))
263 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
264 # define CHAR_STRING(c, s) (*(s) = (c), 1)
265 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
266 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
267 # define RE_CHAR_TO_MULTIBYTE(c) (c)
268 # define RE_CHAR_TO_UNIBYTE(c) (c)
269 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
270 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
271 # define GET_CHAR_AFTER(c, p, len) \
272 (c = *p, len = 1)
273 # define MAKE_CHAR(charset, c1, c2) (c1)
274 # define BYTE8_TO_CHAR(c) (c)
275 # define CHAR_BYTE8_P(c) (0)
276 # define CHAR_LEADING_CODE(c) (c)
277
278 #endif /* not emacs */
279
280 #ifndef RE_TRANSLATE
281 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
282 # define RE_TRANSLATE_P(TBL) (TBL)
283 #endif
284 \f
285 /* Get the interface, including the syntax bits. */
286 #include "regex.h"
287
288 /* isalpha etc. are used for the character classes. */
289 #include <ctype.h>
290
291 #ifdef emacs
292
293 /* 1 if C is an ASCII character. */
294 # define IS_REAL_ASCII(c) ((c) < 0200)
295
296 /* 1 if C is a unibyte character. */
297 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
298
299 /* The Emacs definitions should not be directly affected by locales. */
300
301 /* In Emacs, these are only used for single-byte characters. */
302 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
303 # define ISCNTRL(c) ((c) < ' ')
304 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
305 || ((c) >= 'a' && (c) <= 'f') \
306 || ((c) >= 'A' && (c) <= 'F'))
307
308 /* This is only used for single-byte characters. */
309 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
310
311 /* The rest must handle multibyte characters. */
312
313 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
314 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
315 : 1)
316
317 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
318 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
319 : 1)
320
321 # define ISALNUM(c) (IS_REAL_ASCII (c) \
322 ? (((c) >= 'a' && (c) <= 'z') \
323 || ((c) >= 'A' && (c) <= 'Z') \
324 || ((c) >= '0' && (c) <= '9')) \
325 : SYNTAX (c) == Sword)
326
327 # define ISALPHA(c) (IS_REAL_ASCII (c) \
328 ? (((c) >= 'a' && (c) <= 'z') \
329 || ((c) >= 'A' && (c) <= 'Z')) \
330 : SYNTAX (c) == Sword)
331
332 # define ISLOWER(c) lowercasep (c)
333
334 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
335 ? ((c) > ' ' && (c) < 0177 \
336 && !(((c) >= 'a' && (c) <= 'z') \
337 || ((c) >= 'A' && (c) <= 'Z') \
338 || ((c) >= '0' && (c) <= '9'))) \
339 : SYNTAX (c) != Sword)
340
341 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
342
343 # define ISUPPER(c) uppercasep (c)
344
345 # define ISWORD(c) (SYNTAX (c) == Sword)
346
347 #else /* not emacs */
348
349 /* 1 if C is an ASCII character. */
350 # define IS_REAL_ASCII(c) ((c) < 0200)
351
352 /* This distinction is not meaningful, except in Emacs. */
353 # define ISUNIBYTE(c) 1
354
355 # ifdef isblank
356 # define ISBLANK(c) isblank (c)
357 # else
358 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
359 # endif
360 # ifdef isgraph
361 # define ISGRAPH(c) isgraph (c)
362 # else
363 # define ISGRAPH(c) (isprint (c) && !isspace (c))
364 # endif
365
366 /* Solaris defines ISPRINT so we must undefine it first. */
367 # undef ISPRINT
368 # define ISPRINT(c) isprint (c)
369 # define ISDIGIT(c) isdigit (c)
370 # define ISALNUM(c) isalnum (c)
371 # define ISALPHA(c) isalpha (c)
372 # define ISCNTRL(c) iscntrl (c)
373 # define ISLOWER(c) islower (c)
374 # define ISPUNCT(c) ispunct (c)
375 # define ISSPACE(c) isspace (c)
376 # define ISUPPER(c) isupper (c)
377 # define ISXDIGIT(c) isxdigit (c)
378
379 # define ISWORD(c) ISALPHA(c)
380
381 # ifdef _tolower
382 # define TOLOWER(c) _tolower(c)
383 # else
384 # define TOLOWER(c) tolower(c)
385 # endif
386
387 /* How many characters in the character set. */
388 # define CHAR_SET_SIZE 256
389
390 # ifdef SYNTAX_TABLE
391
392 extern char *re_syntax_table;
393
394 # else /* not SYNTAX_TABLE */
395
396 static char re_syntax_table[CHAR_SET_SIZE];
397
398 static void
399 init_syntax_once (void)
400 {
401 register int c;
402 static int done = 0;
403
404 if (done)
405 return;
406
407 memset (re_syntax_table, 0, sizeof re_syntax_table);
408
409 for (c = 0; c < CHAR_SET_SIZE; ++c)
410 if (ISALNUM (c))
411 re_syntax_table[c] = Sword;
412
413 re_syntax_table['_'] = Ssymbol;
414
415 done = 1;
416 }
417
418 # endif /* not SYNTAX_TABLE */
419
420 # define SYNTAX(c) re_syntax_table[(c)]
421
422 #endif /* not emacs */
423 \f
424 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
425 since ours (we hope) works properly with all combinations of
426 machines, compilers, `char' and `unsigned char' argument types.
427 (Per Bothner suggested the basic approach.) */
428 #undef SIGN_EXTEND_CHAR
429 #if __STDC__
430 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
431 #else /* not __STDC__ */
432 /* As in Harbison and Steele. */
433 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
434 #endif
435 \f
436 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
437 use `alloca' instead of `malloc'. This is because using malloc in
438 re_search* or re_match* could cause memory leaks when C-g is used in
439 Emacs; also, malloc is slower and causes storage fragmentation. On
440 the other hand, malloc is more portable, and easier to debug.
441
442 Because we sometimes use alloca, some routines have to be macros,
443 not functions -- `alloca'-allocated space disappears at the end of the
444 function it is called in. */
445
446 #ifdef REGEX_MALLOC
447
448 # define REGEX_ALLOCATE malloc
449 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
450 # define REGEX_FREE free
451
452 #else /* not REGEX_MALLOC */
453
454 /* Emacs already defines alloca, sometimes. */
455 # ifndef alloca
456
457 /* Make alloca work the best possible way. */
458 # ifdef __GNUC__
459 # define alloca __builtin_alloca
460 # else /* not __GNUC__ */
461 # ifdef HAVE_ALLOCA_H
462 # include <alloca.h>
463 # endif /* HAVE_ALLOCA_H */
464 # endif /* not __GNUC__ */
465
466 # endif /* not alloca */
467
468 # define REGEX_ALLOCATE alloca
469
470 /* Assumes a `char *destination' variable. */
471 # define REGEX_REALLOCATE(source, osize, nsize) \
472 (destination = (char *) alloca (nsize), \
473 memcpy (destination, source, osize))
474
475 /* No need to do anything to free, after alloca. */
476 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
477
478 #endif /* not REGEX_MALLOC */
479
480 /* Define how to allocate the failure stack. */
481
482 #if defined REL_ALLOC && defined REGEX_MALLOC
483
484 # define REGEX_ALLOCATE_STACK(size) \
485 r_alloc (&failure_stack_ptr, (size))
486 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
487 r_re_alloc (&failure_stack_ptr, (nsize))
488 # define REGEX_FREE_STACK(ptr) \
489 r_alloc_free (&failure_stack_ptr)
490
491 #else /* not using relocating allocator */
492
493 # ifdef REGEX_MALLOC
494
495 # define REGEX_ALLOCATE_STACK malloc
496 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
497 # define REGEX_FREE_STACK free
498
499 # else /* not REGEX_MALLOC */
500
501 # define REGEX_ALLOCATE_STACK alloca
502
503 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
504 REGEX_REALLOCATE (source, osize, nsize)
505 /* No need to explicitly free anything. */
506 # define REGEX_FREE_STACK(arg) ((void)0)
507
508 # endif /* not REGEX_MALLOC */
509 #endif /* not using relocating allocator */
510
511
512 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
513 `string1' or just past its end. This works if PTR is NULL, which is
514 a good thing. */
515 #define FIRST_STRING_P(ptr) \
516 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
517
518 /* (Re)Allocate N items of type T using malloc, or fail. */
519 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
520 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
521 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
522
523 #define BYTEWIDTH 8 /* In bits. */
524
525 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
526
527 #undef MAX
528 #undef MIN
529 #define MAX(a, b) ((a) > (b) ? (a) : (b))
530 #define MIN(a, b) ((a) < (b) ? (a) : (b))
531
532 /* Type of source-pattern and string chars. */
533 typedef const unsigned char re_char;
534
535 typedef char boolean;
536 #define false 0
537 #define true 1
538
539 static regoff_t re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
540 re_char *string1, size_t size1,
541 re_char *string2, size_t size2,
542 ssize_t pos,
543 struct re_registers *regs,
544 ssize_t stop));
545 \f
546 /* These are the command codes that appear in compiled regular
547 expressions. Some opcodes are followed by argument bytes. A
548 command code can specify any interpretation whatsoever for its
549 arguments. Zero bytes may appear in the compiled regular expression. */
550
551 typedef enum
552 {
553 no_op = 0,
554
555 /* Succeed right away--no more backtracking. */
556 succeed,
557
558 /* Followed by one byte giving n, then by n literal bytes. */
559 exactn,
560
561 /* Matches any (more or less) character. */
562 anychar,
563
564 /* Matches any one char belonging to specified set. First
565 following byte is number of bitmap bytes. Then come bytes
566 for a bitmap saying which chars are in. Bits in each byte
567 are ordered low-bit-first. A character is in the set if its
568 bit is 1. A character too large to have a bit in the map is
569 automatically not in the set.
570
571 If the length byte has the 0x80 bit set, then that stuff
572 is followed by a range table:
573 2 bytes of flags for character sets (low 8 bits, high 8 bits)
574 See RANGE_TABLE_WORK_BITS below.
575 2 bytes, the number of pairs that follow (upto 32767)
576 pairs, each 2 multibyte characters,
577 each multibyte character represented as 3 bytes. */
578 charset,
579
580 /* Same parameters as charset, but match any character that is
581 not one of those specified. */
582 charset_not,
583
584 /* Start remembering the text that is matched, for storing in a
585 register. Followed by one byte with the register number, in
586 the range 0 to one less than the pattern buffer's re_nsub
587 field. */
588 start_memory,
589
590 /* Stop remembering the text that is matched and store it in a
591 memory register. Followed by one byte with the register
592 number, in the range 0 to one less than `re_nsub' in the
593 pattern buffer. */
594 stop_memory,
595
596 /* Match a duplicate of something remembered. Followed by one
597 byte containing the register number. */
598 duplicate,
599
600 /* Fail unless at beginning of line. */
601 begline,
602
603 /* Fail unless at end of line. */
604 endline,
605
606 /* Succeeds if at beginning of buffer (if emacs) or at beginning
607 of string to be matched (if not). */
608 begbuf,
609
610 /* Analogously, for end of buffer/string. */
611 endbuf,
612
613 /* Followed by two byte relative address to which to jump. */
614 jump,
615
616 /* Followed by two-byte relative address of place to resume at
617 in case of failure. */
618 on_failure_jump,
619
620 /* Like on_failure_jump, but pushes a placeholder instead of the
621 current string position when executed. */
622 on_failure_keep_string_jump,
623
624 /* Just like `on_failure_jump', except that it checks that we
625 don't get stuck in an infinite loop (matching an empty string
626 indefinitely). */
627 on_failure_jump_loop,
628
629 /* Just like `on_failure_jump_loop', except that it checks for
630 a different kind of loop (the kind that shows up with non-greedy
631 operators). This operation has to be immediately preceded
632 by a `no_op'. */
633 on_failure_jump_nastyloop,
634
635 /* A smart `on_failure_jump' used for greedy * and + operators.
636 It analyses the loop before which it is put and if the
637 loop does not require backtracking, it changes itself to
638 `on_failure_keep_string_jump' and short-circuits the loop,
639 else it just defaults to changing itself into `on_failure_jump'.
640 It assumes that it is pointing to just past a `jump'. */
641 on_failure_jump_smart,
642
643 /* Followed by two-byte relative address and two-byte number n.
644 After matching N times, jump to the address upon failure.
645 Does not work if N starts at 0: use on_failure_jump_loop
646 instead. */
647 succeed_n,
648
649 /* Followed by two-byte relative address, and two-byte number n.
650 Jump to the address N times, then fail. */
651 jump_n,
652
653 /* Set the following two-byte relative address to the
654 subsequent two-byte number. The address *includes* the two
655 bytes of number. */
656 set_number_at,
657
658 wordbeg, /* Succeeds if at word beginning. */
659 wordend, /* Succeeds if at word end. */
660
661 wordbound, /* Succeeds if at a word boundary. */
662 notwordbound, /* Succeeds if not at a word boundary. */
663
664 symbeg, /* Succeeds if at symbol beginning. */
665 symend, /* Succeeds if at symbol end. */
666
667 /* Matches any character whose syntax is specified. Followed by
668 a byte which contains a syntax code, e.g., Sword. */
669 syntaxspec,
670
671 /* Matches any character whose syntax is not that specified. */
672 notsyntaxspec
673
674 #ifdef emacs
675 ,before_dot, /* Succeeds if before point. */
676 at_dot, /* Succeeds if at point. */
677 after_dot, /* Succeeds if after point. */
678
679 /* Matches any character whose category-set contains the specified
680 category. The operator is followed by a byte which contains a
681 category code (mnemonic ASCII character). */
682 categoryspec,
683
684 /* Matches any character whose category-set does not contain the
685 specified category. The operator is followed by a byte which
686 contains the category code (mnemonic ASCII character). */
687 notcategoryspec
688 #endif /* emacs */
689 } re_opcode_t;
690 \f
691 /* Common operations on the compiled pattern. */
692
693 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
694
695 #define STORE_NUMBER(destination, number) \
696 do { \
697 (destination)[0] = (number) & 0377; \
698 (destination)[1] = (number) >> 8; \
699 } while (0)
700
701 /* Same as STORE_NUMBER, except increment DESTINATION to
702 the byte after where the number is stored. Therefore, DESTINATION
703 must be an lvalue. */
704
705 #define STORE_NUMBER_AND_INCR(destination, number) \
706 do { \
707 STORE_NUMBER (destination, number); \
708 (destination) += 2; \
709 } while (0)
710
711 /* Put into DESTINATION a number stored in two contiguous bytes starting
712 at SOURCE. */
713
714 #define EXTRACT_NUMBER(destination, source) \
715 do { \
716 (destination) = *(source) & 0377; \
717 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
718 } while (0)
719
720 #ifdef DEBUG
721 static void extract_number _RE_ARGS ((int *dest, re_char *source));
722 static void
723 extract_number (dest, source)
724 int *dest;
725 re_char *source;
726 {
727 int temp = SIGN_EXTEND_CHAR (*(source + 1));
728 *dest = *source & 0377;
729 *dest += temp << 8;
730 }
731
732 # ifndef EXTRACT_MACROS /* To debug the macros. */
733 # undef EXTRACT_NUMBER
734 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
735 # endif /* not EXTRACT_MACROS */
736
737 #endif /* DEBUG */
738
739 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
740 SOURCE must be an lvalue. */
741
742 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
743 do { \
744 EXTRACT_NUMBER (destination, source); \
745 (source) += 2; \
746 } while (0)
747
748 #ifdef DEBUG
749 static void extract_number_and_incr _RE_ARGS ((int *destination,
750 re_char **source));
751 static void
752 extract_number_and_incr (destination, source)
753 int *destination;
754 re_char **source;
755 {
756 extract_number (destination, *source);
757 *source += 2;
758 }
759
760 # ifndef EXTRACT_MACROS
761 # undef EXTRACT_NUMBER_AND_INCR
762 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
763 extract_number_and_incr (&dest, &src)
764 # endif /* not EXTRACT_MACROS */
765
766 #endif /* DEBUG */
767 \f
768 /* Store a multibyte character in three contiguous bytes starting
769 DESTINATION, and increment DESTINATION to the byte after where the
770 character is stored. Therefore, DESTINATION must be an lvalue. */
771
772 #define STORE_CHARACTER_AND_INCR(destination, character) \
773 do { \
774 (destination)[0] = (character) & 0377; \
775 (destination)[1] = ((character) >> 8) & 0377; \
776 (destination)[2] = (character) >> 16; \
777 (destination) += 3; \
778 } while (0)
779
780 /* Put into DESTINATION a character stored in three contiguous bytes
781 starting at SOURCE. */
782
783 #define EXTRACT_CHARACTER(destination, source) \
784 do { \
785 (destination) = ((source)[0] \
786 | ((source)[1] << 8) \
787 | ((source)[2] << 16)); \
788 } while (0)
789
790
791 /* Macros for charset. */
792
793 /* Size of bitmap of charset P in bytes. P is a start of charset,
794 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
795 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
796
797 /* Nonzero if charset P has range table. */
798 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
799
800 /* Return the address of range table of charset P. But not the start
801 of table itself, but the before where the number of ranges is
802 stored. `2 +' means to skip re_opcode_t and size of bitmap,
803 and the 2 bytes of flags at the start of the range table. */
804 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
805
806 /* Extract the bit flags that start a range table. */
807 #define CHARSET_RANGE_TABLE_BITS(p) \
808 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
809 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
810
811 /* Return the address of end of RANGE_TABLE. COUNT is number of
812 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
813 is start of range and end of range. `* 3' is size of each start
814 and end. */
815 #define CHARSET_RANGE_TABLE_END(range_table, count) \
816 ((range_table) + (count) * 2 * 3)
817
818 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
819 COUNT is number of ranges in RANGE_TABLE. */
820 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
821 do \
822 { \
823 re_wchar_t range_start, range_end; \
824 re_char *rtp; \
825 re_char *range_table_end \
826 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
827 \
828 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
829 { \
830 EXTRACT_CHARACTER (range_start, rtp); \
831 EXTRACT_CHARACTER (range_end, rtp + 3); \
832 \
833 if (range_start <= (c) && (c) <= range_end) \
834 { \
835 (not) = !(not); \
836 break; \
837 } \
838 } \
839 } \
840 while (0)
841
842 /* Test if C is in range table of CHARSET. The flag NOT is negated if
843 C is listed in it. */
844 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
845 do \
846 { \
847 /* Number of ranges in range table. */ \
848 int count; \
849 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
850 \
851 EXTRACT_NUMBER_AND_INCR (count, range_table); \
852 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
853 } \
854 while (0)
855 \f
856 /* If DEBUG is defined, Regex prints many voluminous messages about what
857 it is doing (if the variable `debug' is nonzero). If linked with the
858 main program in `iregex.c', you can enter patterns and strings
859 interactively. And if linked with the main program in `main.c' and
860 the other test files, you can run the already-written tests. */
861
862 #ifdef DEBUG
863
864 /* We use standard I/O for debugging. */
865 # include <stdio.h>
866
867 /* It is useful to test things that ``must'' be true when debugging. */
868 # include <assert.h>
869
870 static int debug = -100000;
871
872 # define DEBUG_STATEMENT(e) e
873 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
874 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
875 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
876 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
877 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
878 if (debug > 0) print_partial_compiled_pattern (s, e)
879 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
880 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
881
882
883 /* Print the fastmap in human-readable form. */
884
885 void
886 print_fastmap (fastmap)
887 char *fastmap;
888 {
889 unsigned was_a_range = 0;
890 unsigned i = 0;
891
892 while (i < (1 << BYTEWIDTH))
893 {
894 if (fastmap[i++])
895 {
896 was_a_range = 0;
897 putchar (i - 1);
898 while (i < (1 << BYTEWIDTH) && fastmap[i])
899 {
900 was_a_range = 1;
901 i++;
902 }
903 if (was_a_range)
904 {
905 printf ("-");
906 putchar (i - 1);
907 }
908 }
909 }
910 putchar ('\n');
911 }
912
913
914 /* Print a compiled pattern string in human-readable form, starting at
915 the START pointer into it and ending just before the pointer END. */
916
917 void
918 print_partial_compiled_pattern (start, end)
919 re_char *start;
920 re_char *end;
921 {
922 int mcnt, mcnt2;
923 re_char *p = start;
924 re_char *pend = end;
925
926 if (start == NULL)
927 {
928 fprintf (stderr, "(null)\n");
929 return;
930 }
931
932 /* Loop over pattern commands. */
933 while (p < pend)
934 {
935 fprintf (stderr, "%d:\t", p - start);
936
937 switch ((re_opcode_t) *p++)
938 {
939 case no_op:
940 fprintf (stderr, "/no_op");
941 break;
942
943 case succeed:
944 fprintf (stderr, "/succeed");
945 break;
946
947 case exactn:
948 mcnt = *p++;
949 fprintf (stderr, "/exactn/%d", mcnt);
950 do
951 {
952 fprintf (stderr, "/%c", *p++);
953 }
954 while (--mcnt);
955 break;
956
957 case start_memory:
958 fprintf (stderr, "/start_memory/%d", *p++);
959 break;
960
961 case stop_memory:
962 fprintf (stderr, "/stop_memory/%d", *p++);
963 break;
964
965 case duplicate:
966 fprintf (stderr, "/duplicate/%d", *p++);
967 break;
968
969 case anychar:
970 fprintf (stderr, "/anychar");
971 break;
972
973 case charset:
974 case charset_not:
975 {
976 register int c, last = -100;
977 register int in_range = 0;
978 int length = CHARSET_BITMAP_SIZE (p - 1);
979 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
980
981 fprintf (stderr, "/charset [%s",
982 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
983
984 if (p + *p >= pend)
985 fprintf (stderr, " !extends past end of pattern! ");
986
987 for (c = 0; c < 256; c++)
988 if (c / 8 < length
989 && (p[1 + (c/8)] & (1 << (c % 8))))
990 {
991 /* Are we starting a range? */
992 if (last + 1 == c && ! in_range)
993 {
994 fprintf (stderr, "-");
995 in_range = 1;
996 }
997 /* Have we broken a range? */
998 else if (last + 1 != c && in_range)
999 {
1000 fprintf (stderr, "%c", last);
1001 in_range = 0;
1002 }
1003
1004 if (! in_range)
1005 fprintf (stderr, "%c", c);
1006
1007 last = c;
1008 }
1009
1010 if (in_range)
1011 fprintf (stderr, "%c", last);
1012
1013 fprintf (stderr, "]");
1014
1015 p += 1 + length;
1016
1017 if (has_range_table)
1018 {
1019 int count;
1020 fprintf (stderr, "has-range-table");
1021
1022 /* ??? Should print the range table; for now, just skip it. */
1023 p += 2; /* skip range table bits */
1024 EXTRACT_NUMBER_AND_INCR (count, p);
1025 p = CHARSET_RANGE_TABLE_END (p, count);
1026 }
1027 }
1028 break;
1029
1030 case begline:
1031 fprintf (stderr, "/begline");
1032 break;
1033
1034 case endline:
1035 fprintf (stderr, "/endline");
1036 break;
1037
1038 case on_failure_jump:
1039 extract_number_and_incr (&mcnt, &p);
1040 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1041 break;
1042
1043 case on_failure_keep_string_jump:
1044 extract_number_and_incr (&mcnt, &p);
1045 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1046 break;
1047
1048 case on_failure_jump_nastyloop:
1049 extract_number_and_incr (&mcnt, &p);
1050 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1051 break;
1052
1053 case on_failure_jump_loop:
1054 extract_number_and_incr (&mcnt, &p);
1055 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1056 break;
1057
1058 case on_failure_jump_smart:
1059 extract_number_and_incr (&mcnt, &p);
1060 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1061 break;
1062
1063 case jump:
1064 extract_number_and_incr (&mcnt, &p);
1065 fprintf (stderr, "/jump to %d", p + mcnt - start);
1066 break;
1067
1068 case succeed_n:
1069 extract_number_and_incr (&mcnt, &p);
1070 extract_number_and_incr (&mcnt2, &p);
1071 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1072 break;
1073
1074 case jump_n:
1075 extract_number_and_incr (&mcnt, &p);
1076 extract_number_and_incr (&mcnt2, &p);
1077 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1078 break;
1079
1080 case set_number_at:
1081 extract_number_and_incr (&mcnt, &p);
1082 extract_number_and_incr (&mcnt2, &p);
1083 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1084 break;
1085
1086 case wordbound:
1087 fprintf (stderr, "/wordbound");
1088 break;
1089
1090 case notwordbound:
1091 fprintf (stderr, "/notwordbound");
1092 break;
1093
1094 case wordbeg:
1095 fprintf (stderr, "/wordbeg");
1096 break;
1097
1098 case wordend:
1099 fprintf (stderr, "/wordend");
1100 break;
1101
1102 case symbeg:
1103 fprintf (stderr, "/symbeg");
1104 break;
1105
1106 case symend:
1107 fprintf (stderr, "/symend");
1108 break;
1109
1110 case syntaxspec:
1111 fprintf (stderr, "/syntaxspec");
1112 mcnt = *p++;
1113 fprintf (stderr, "/%d", mcnt);
1114 break;
1115
1116 case notsyntaxspec:
1117 fprintf (stderr, "/notsyntaxspec");
1118 mcnt = *p++;
1119 fprintf (stderr, "/%d", mcnt);
1120 break;
1121
1122 # ifdef emacs
1123 case before_dot:
1124 fprintf (stderr, "/before_dot");
1125 break;
1126
1127 case at_dot:
1128 fprintf (stderr, "/at_dot");
1129 break;
1130
1131 case after_dot:
1132 fprintf (stderr, "/after_dot");
1133 break;
1134
1135 case categoryspec:
1136 fprintf (stderr, "/categoryspec");
1137 mcnt = *p++;
1138 fprintf (stderr, "/%d", mcnt);
1139 break;
1140
1141 case notcategoryspec:
1142 fprintf (stderr, "/notcategoryspec");
1143 mcnt = *p++;
1144 fprintf (stderr, "/%d", mcnt);
1145 break;
1146 # endif /* emacs */
1147
1148 case begbuf:
1149 fprintf (stderr, "/begbuf");
1150 break;
1151
1152 case endbuf:
1153 fprintf (stderr, "/endbuf");
1154 break;
1155
1156 default:
1157 fprintf (stderr, "?%d", *(p-1));
1158 }
1159
1160 fprintf (stderr, "\n");
1161 }
1162
1163 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1164 }
1165
1166
1167 void
1168 print_compiled_pattern (bufp)
1169 struct re_pattern_buffer *bufp;
1170 {
1171 re_char *buffer = bufp->buffer;
1172
1173 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1174 printf ("%ld bytes used/%ld bytes allocated.\n",
1175 bufp->used, bufp->allocated);
1176
1177 if (bufp->fastmap_accurate && bufp->fastmap)
1178 {
1179 printf ("fastmap: ");
1180 print_fastmap (bufp->fastmap);
1181 }
1182
1183 printf ("re_nsub: %d\t", bufp->re_nsub);
1184 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1185 printf ("can_be_null: %d\t", bufp->can_be_null);
1186 printf ("no_sub: %d\t", bufp->no_sub);
1187 printf ("not_bol: %d\t", bufp->not_bol);
1188 printf ("not_eol: %d\t", bufp->not_eol);
1189 printf ("syntax: %lx\n", bufp->syntax);
1190 fflush (stdout);
1191 /* Perhaps we should print the translate table? */
1192 }
1193
1194
1195 void
1196 print_double_string (where, string1, size1, string2, size2)
1197 re_char *where;
1198 re_char *string1;
1199 re_char *string2;
1200 ssize_t size1;
1201 ssize_t size2;
1202 {
1203 ssize_t this_char;
1204
1205 if (where == NULL)
1206 printf ("(null)");
1207 else
1208 {
1209 if (FIRST_STRING_P (where))
1210 {
1211 for (this_char = where - string1; this_char < size1; this_char++)
1212 putchar (string1[this_char]);
1213
1214 where = string2;
1215 }
1216
1217 for (this_char = where - string2; this_char < size2; this_char++)
1218 putchar (string2[this_char]);
1219 }
1220 }
1221
1222 #else /* not DEBUG */
1223
1224 # undef assert
1225 # define assert(e)
1226
1227 # define DEBUG_STATEMENT(e)
1228 # define DEBUG_PRINT1(x)
1229 # define DEBUG_PRINT2(x1, x2)
1230 # define DEBUG_PRINT3(x1, x2, x3)
1231 # define DEBUG_PRINT4(x1, x2, x3, x4)
1232 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1233 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1234
1235 #endif /* not DEBUG */
1236 \f
1237 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1238 #ifdef lint
1239 # define IF_LINT(Code) Code
1240 #else
1241 # define IF_LINT(Code) /* empty */
1242 #endif
1243 \f
1244 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1245 also be assigned to arbitrarily: each pattern buffer stores its own
1246 syntax, so it can be changed between regex compilations. */
1247 /* This has no initializer because initialized variables in Emacs
1248 become read-only after dumping. */
1249 reg_syntax_t re_syntax_options;
1250
1251
1252 /* Specify the precise syntax of regexps for compilation. This provides
1253 for compatibility for various utilities which historically have
1254 different, incompatible syntaxes.
1255
1256 The argument SYNTAX is a bit mask comprised of the various bits
1257 defined in regex.h. We return the old syntax. */
1258
1259 reg_syntax_t
1260 re_set_syntax (reg_syntax_t syntax)
1261 {
1262 reg_syntax_t ret = re_syntax_options;
1263
1264 re_syntax_options = syntax;
1265 return ret;
1266 }
1267 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1268
1269 /* Regexp to use to replace spaces, or NULL meaning don't. */
1270 static re_char *whitespace_regexp;
1271
1272 void
1273 re_set_whitespace_regexp (const char *regexp)
1274 {
1275 whitespace_regexp = (re_char *) regexp;
1276 }
1277 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1278 \f
1279 /* This table gives an error message for each of the error codes listed
1280 in regex.h. Obviously the order here has to be same as there.
1281 POSIX doesn't require that we do anything for REG_NOERROR,
1282 but why not be nice? */
1283
1284 static const char *re_error_msgid[] =
1285 {
1286 gettext_noop ("Success"), /* REG_NOERROR */
1287 gettext_noop ("No match"), /* REG_NOMATCH */
1288 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1289 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1290 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1291 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1292 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1293 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1294 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1295 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1296 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1297 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1298 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1299 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1300 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1301 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1302 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1303 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1304 };
1305 \f
1306 /* Avoiding alloca during matching, to placate r_alloc. */
1307
1308 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1309 searching and matching functions should not call alloca. On some
1310 systems, alloca is implemented in terms of malloc, and if we're
1311 using the relocating allocator routines, then malloc could cause a
1312 relocation, which might (if the strings being searched are in the
1313 ralloc heap) shift the data out from underneath the regexp
1314 routines.
1315
1316 Here's another reason to avoid allocation: Emacs
1317 processes input from X in a signal handler; processing X input may
1318 call malloc; if input arrives while a matching routine is calling
1319 malloc, then we're scrod. But Emacs can't just block input while
1320 calling matching routines; then we don't notice interrupts when
1321 they come in. So, Emacs blocks input around all regexp calls
1322 except the matching calls, which it leaves unprotected, in the
1323 faith that they will not malloc. */
1324
1325 /* Normally, this is fine. */
1326 #define MATCH_MAY_ALLOCATE
1327
1328 /* The match routines may not allocate if (1) they would do it with malloc
1329 and (2) it's not safe for them to use malloc.
1330 Note that if REL_ALLOC is defined, matching would not use malloc for the
1331 failure stack, but we would still use it for the register vectors;
1332 so REL_ALLOC should not affect this. */
1333 #if defined REGEX_MALLOC && defined emacs
1334 # undef MATCH_MAY_ALLOCATE
1335 #endif
1336
1337 \f
1338 /* Failure stack declarations and macros; both re_compile_fastmap and
1339 re_match_2 use a failure stack. These have to be macros because of
1340 REGEX_ALLOCATE_STACK. */
1341
1342
1343 /* Approximate number of failure points for which to initially allocate space
1344 when matching. If this number is exceeded, we allocate more
1345 space, so it is not a hard limit. */
1346 #ifndef INIT_FAILURE_ALLOC
1347 # define INIT_FAILURE_ALLOC 20
1348 #endif
1349
1350 /* Roughly the maximum number of failure points on the stack. Would be
1351 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1352 This is a variable only so users of regex can assign to it; we never
1353 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1354 before using it, so it should probably be a byte-count instead. */
1355 # if defined MATCH_MAY_ALLOCATE
1356 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1357 whose default stack limit is 2mb. In order for a larger
1358 value to work reliably, you have to try to make it accord
1359 with the process stack limit. */
1360 size_t re_max_failures = 40000;
1361 # else
1362 size_t re_max_failures = 4000;
1363 # endif
1364
1365 union fail_stack_elt
1366 {
1367 re_char *pointer;
1368 /* This should be the biggest `int' that's no bigger than a pointer. */
1369 long integer;
1370 };
1371
1372 typedef union fail_stack_elt fail_stack_elt_t;
1373
1374 typedef struct
1375 {
1376 fail_stack_elt_t *stack;
1377 size_t size;
1378 size_t avail; /* Offset of next open position. */
1379 size_t frame; /* Offset of the cur constructed frame. */
1380 } fail_stack_type;
1381
1382 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1383
1384
1385 /* Define macros to initialize and free the failure stack.
1386 Do `return -2' if the alloc fails. */
1387
1388 #ifdef MATCH_MAY_ALLOCATE
1389 # define INIT_FAIL_STACK() \
1390 do { \
1391 fail_stack.stack = (fail_stack_elt_t *) \
1392 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1393 * sizeof (fail_stack_elt_t)); \
1394 \
1395 if (fail_stack.stack == NULL) \
1396 return -2; \
1397 \
1398 fail_stack.size = INIT_FAILURE_ALLOC; \
1399 fail_stack.avail = 0; \
1400 fail_stack.frame = 0; \
1401 } while (0)
1402 #else
1403 # define INIT_FAIL_STACK() \
1404 do { \
1405 fail_stack.avail = 0; \
1406 fail_stack.frame = 0; \
1407 } while (0)
1408
1409 # define RETALLOC_IF(addr, n, t) \
1410 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1411 #endif
1412
1413
1414 /* Double the size of FAIL_STACK, up to a limit
1415 which allows approximately `re_max_failures' items.
1416
1417 Return 1 if succeeds, and 0 if either ran out of memory
1418 allocating space for it or it was already too large.
1419
1420 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1421
1422 /* Factor to increase the failure stack size by
1423 when we increase it.
1424 This used to be 2, but 2 was too wasteful
1425 because the old discarded stacks added up to as much space
1426 were as ultimate, maximum-size stack. */
1427 #define FAIL_STACK_GROWTH_FACTOR 4
1428
1429 #define GROW_FAIL_STACK(fail_stack) \
1430 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1431 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1432 ? 0 \
1433 : ((fail_stack).stack \
1434 = (fail_stack_elt_t *) \
1435 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1436 (fail_stack).size * sizeof (fail_stack_elt_t), \
1437 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1438 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1439 * FAIL_STACK_GROWTH_FACTOR))), \
1440 \
1441 (fail_stack).stack == NULL \
1442 ? 0 \
1443 : ((fail_stack).size \
1444 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1445 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1446 * FAIL_STACK_GROWTH_FACTOR)) \
1447 / sizeof (fail_stack_elt_t)), \
1448 1)))
1449
1450
1451 /* Push a pointer value onto the failure stack.
1452 Assumes the variable `fail_stack'. Probably should only
1453 be called from within `PUSH_FAILURE_POINT'. */
1454 #define PUSH_FAILURE_POINTER(item) \
1455 fail_stack.stack[fail_stack.avail++].pointer = (item)
1456
1457 /* This pushes an integer-valued item onto the failure stack.
1458 Assumes the variable `fail_stack'. Probably should only
1459 be called from within `PUSH_FAILURE_POINT'. */
1460 #define PUSH_FAILURE_INT(item) \
1461 fail_stack.stack[fail_stack.avail++].integer = (item)
1462
1463 /* These POP... operations complement the PUSH... operations.
1464 All assume that `fail_stack' is nonempty. */
1465 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1466 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1467
1468 /* Individual items aside from the registers. */
1469 #define NUM_NONREG_ITEMS 3
1470
1471 /* Used to examine the stack (to detect infinite loops). */
1472 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1473 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1474 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1475 #define TOP_FAILURE_HANDLE() fail_stack.frame
1476
1477
1478 #define ENSURE_FAIL_STACK(space) \
1479 while (REMAINING_AVAIL_SLOTS <= space) { \
1480 if (!GROW_FAIL_STACK (fail_stack)) \
1481 return -2; \
1482 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1483 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1484 }
1485
1486 /* Push register NUM onto the stack. */
1487 #define PUSH_FAILURE_REG(num) \
1488 do { \
1489 char *destination; \
1490 ENSURE_FAIL_STACK(3); \
1491 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1492 num, regstart[num], regend[num]); \
1493 PUSH_FAILURE_POINTER (regstart[num]); \
1494 PUSH_FAILURE_POINTER (regend[num]); \
1495 PUSH_FAILURE_INT (num); \
1496 } while (0)
1497
1498 /* Change the counter's value to VAL, but make sure that it will
1499 be reset when backtracking. */
1500 #define PUSH_NUMBER(ptr,val) \
1501 do { \
1502 char *destination; \
1503 int c; \
1504 ENSURE_FAIL_STACK(3); \
1505 EXTRACT_NUMBER (c, ptr); \
1506 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1507 PUSH_FAILURE_INT (c); \
1508 PUSH_FAILURE_POINTER (ptr); \
1509 PUSH_FAILURE_INT (-1); \
1510 STORE_NUMBER (ptr, val); \
1511 } while (0)
1512
1513 /* Pop a saved register off the stack. */
1514 #define POP_FAILURE_REG_OR_COUNT() \
1515 do { \
1516 long pfreg = POP_FAILURE_INT (); \
1517 if (pfreg == -1) \
1518 { \
1519 /* It's a counter. */ \
1520 /* Here, we discard `const', making re_match non-reentrant. */ \
1521 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1522 pfreg = POP_FAILURE_INT (); \
1523 STORE_NUMBER (ptr, pfreg); \
1524 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, pfreg); \
1525 } \
1526 else \
1527 { \
1528 regend[pfreg] = POP_FAILURE_POINTER (); \
1529 regstart[pfreg] = POP_FAILURE_POINTER (); \
1530 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1531 pfreg, regstart[pfreg], regend[pfreg]); \
1532 } \
1533 } while (0)
1534
1535 /* Check that we are not stuck in an infinite loop. */
1536 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1537 do { \
1538 ssize_t failure = TOP_FAILURE_HANDLE (); \
1539 /* Check for infinite matching loops */ \
1540 while (failure > 0 \
1541 && (FAILURE_STR (failure) == string_place \
1542 || FAILURE_STR (failure) == NULL)) \
1543 { \
1544 assert (FAILURE_PAT (failure) >= bufp->buffer \
1545 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1546 if (FAILURE_PAT (failure) == pat_cur) \
1547 { \
1548 cycle = 1; \
1549 break; \
1550 } \
1551 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1552 failure = NEXT_FAILURE_HANDLE(failure); \
1553 } \
1554 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1555 } while (0)
1556
1557 /* Push the information about the state we will need
1558 if we ever fail back to it.
1559
1560 Requires variables fail_stack, regstart, regend and
1561 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1562 declared.
1563
1564 Does `return FAILURE_CODE' if runs out of memory. */
1565
1566 #define PUSH_FAILURE_POINT(pattern, string_place) \
1567 do { \
1568 char *destination; \
1569 /* Must be int, so when we don't save any registers, the arithmetic \
1570 of 0 + -1 isn't done as unsigned. */ \
1571 \
1572 DEBUG_STATEMENT (nfailure_points_pushed++); \
1573 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1574 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1575 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1576 \
1577 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1578 \
1579 DEBUG_PRINT1 ("\n"); \
1580 \
1581 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1582 PUSH_FAILURE_INT (fail_stack.frame); \
1583 \
1584 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1585 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1586 DEBUG_PRINT1 ("'\n"); \
1587 PUSH_FAILURE_POINTER (string_place); \
1588 \
1589 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1590 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1591 PUSH_FAILURE_POINTER (pattern); \
1592 \
1593 /* Close the frame by moving the frame pointer past it. */ \
1594 fail_stack.frame = fail_stack.avail; \
1595 } while (0)
1596
1597 /* Estimate the size of data pushed by a typical failure stack entry.
1598 An estimate is all we need, because all we use this for
1599 is to choose a limit for how big to make the failure stack. */
1600 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1601 #define TYPICAL_FAILURE_SIZE 20
1602
1603 /* How many items can still be added to the stack without overflowing it. */
1604 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1605
1606
1607 /* Pops what PUSH_FAIL_STACK pushes.
1608
1609 We restore into the parameters, all of which should be lvalues:
1610 STR -- the saved data position.
1611 PAT -- the saved pattern position.
1612 REGSTART, REGEND -- arrays of string positions.
1613
1614 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1615 `pend', `string1', `size1', `string2', and `size2'. */
1616
1617 #define POP_FAILURE_POINT(str, pat) \
1618 do { \
1619 assert (!FAIL_STACK_EMPTY ()); \
1620 \
1621 /* Remove failure points and point to how many regs pushed. */ \
1622 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1623 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1624 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1625 \
1626 /* Pop the saved registers. */ \
1627 while (fail_stack.frame < fail_stack.avail) \
1628 POP_FAILURE_REG_OR_COUNT (); \
1629 \
1630 pat = POP_FAILURE_POINTER (); \
1631 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1632 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1633 \
1634 /* If the saved string location is NULL, it came from an \
1635 on_failure_keep_string_jump opcode, and we want to throw away the \
1636 saved NULL, thus retaining our current position in the string. */ \
1637 str = POP_FAILURE_POINTER (); \
1638 DEBUG_PRINT2 (" Popping string %p: `", str); \
1639 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1640 DEBUG_PRINT1 ("'\n"); \
1641 \
1642 fail_stack.frame = POP_FAILURE_INT (); \
1643 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1644 \
1645 assert (fail_stack.avail >= 0); \
1646 assert (fail_stack.frame <= fail_stack.avail); \
1647 \
1648 DEBUG_STATEMENT (nfailure_points_popped++); \
1649 } while (0) /* POP_FAILURE_POINT */
1650
1651
1652 \f
1653 /* Registers are set to a sentinel when they haven't yet matched. */
1654 #define REG_UNSET(e) ((e) == NULL)
1655 \f
1656 /* Subroutine declarations and macros for regex_compile. */
1657
1658 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1659 reg_syntax_t syntax,
1660 struct re_pattern_buffer *bufp));
1661 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1662 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1663 int arg1, int arg2));
1664 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1665 int arg, unsigned char *end));
1666 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1667 int arg1, int arg2, unsigned char *end));
1668 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1669 re_char *p,
1670 reg_syntax_t syntax));
1671 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1672 re_char *pend,
1673 reg_syntax_t syntax));
1674 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1675 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1676 char *fastmap, const int multibyte));
1677
1678 /* Fetch the next character in the uncompiled pattern, with no
1679 translation. */
1680 #define PATFETCH(c) \
1681 do { \
1682 int len; \
1683 if (p == pend) return REG_EEND; \
1684 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1685 p += len; \
1686 } while (0)
1687
1688
1689 /* If `translate' is non-null, return translate[D], else just D. We
1690 cast the subscript to translate because some data is declared as
1691 `char *', to avoid warnings when a string constant is passed. But
1692 when we use a character as a subscript we must make it unsigned. */
1693 #ifndef TRANSLATE
1694 # define TRANSLATE(d) \
1695 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1696 #endif
1697
1698
1699 /* Macros for outputting the compiled pattern into `buffer'. */
1700
1701 /* If the buffer isn't allocated when it comes in, use this. */
1702 #define INIT_BUF_SIZE 32
1703
1704 /* Make sure we have at least N more bytes of space in buffer. */
1705 #define GET_BUFFER_SPACE(n) \
1706 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1707 EXTEND_BUFFER ()
1708
1709 /* Make sure we have one more byte of buffer space and then add C to it. */
1710 #define BUF_PUSH(c) \
1711 do { \
1712 GET_BUFFER_SPACE (1); \
1713 *b++ = (unsigned char) (c); \
1714 } while (0)
1715
1716
1717 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1718 #define BUF_PUSH_2(c1, c2) \
1719 do { \
1720 GET_BUFFER_SPACE (2); \
1721 *b++ = (unsigned char) (c1); \
1722 *b++ = (unsigned char) (c2); \
1723 } while (0)
1724
1725
1726 /* Store a jump with opcode OP at LOC to location TO. We store a
1727 relative address offset by the three bytes the jump itself occupies. */
1728 #define STORE_JUMP(op, loc, to) \
1729 store_op1 (op, loc, (to) - (loc) - 3)
1730
1731 /* Likewise, for a two-argument jump. */
1732 #define STORE_JUMP2(op, loc, to, arg) \
1733 store_op2 (op, loc, (to) - (loc) - 3, arg)
1734
1735 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1736 #define INSERT_JUMP(op, loc, to) \
1737 insert_op1 (op, loc, (to) - (loc) - 3, b)
1738
1739 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1740 #define INSERT_JUMP2(op, loc, to, arg) \
1741 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1742
1743
1744 /* This is not an arbitrary limit: the arguments which represent offsets
1745 into the pattern are two bytes long. So if 2^15 bytes turns out to
1746 be too small, many things would have to change. */
1747 # define MAX_BUF_SIZE (1L << 15)
1748
1749 #if 0 /* This is when we thought it could be 2^16 bytes. */
1750 /* Any other compiler which, like MSC, has allocation limit below 2^16
1751 bytes will have to use approach similar to what was done below for
1752 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1753 reallocating to 0 bytes. Such thing is not going to work too well.
1754 You have been warned!! */
1755 #if defined _MSC_VER && !defined WIN32
1756 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1757 # define MAX_BUF_SIZE 65500L
1758 #else
1759 # define MAX_BUF_SIZE (1L << 16)
1760 #endif
1761 #endif /* 0 */
1762
1763 /* Extend the buffer by twice its current size via realloc and
1764 reset the pointers that pointed into the old block to point to the
1765 correct places in the new one. If extending the buffer results in it
1766 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1767 #if __BOUNDED_POINTERS__
1768 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1769 # define MOVE_BUFFER_POINTER(P) \
1770 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1771 SET_HIGH_BOUND (P), \
1772 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1773 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1774 else \
1775 { \
1776 SET_HIGH_BOUND (b); \
1777 SET_HIGH_BOUND (begalt); \
1778 if (fixup_alt_jump) \
1779 SET_HIGH_BOUND (fixup_alt_jump); \
1780 if (laststart) \
1781 SET_HIGH_BOUND (laststart); \
1782 if (pending_exact) \
1783 SET_HIGH_BOUND (pending_exact); \
1784 }
1785 #else
1786 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1787 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1788 #endif
1789 #define EXTEND_BUFFER() \
1790 do { \
1791 unsigned char *old_buffer = bufp->buffer; \
1792 if (bufp->allocated == MAX_BUF_SIZE) \
1793 return REG_ESIZE; \
1794 bufp->allocated <<= 1; \
1795 if (bufp->allocated > MAX_BUF_SIZE) \
1796 bufp->allocated = MAX_BUF_SIZE; \
1797 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1798 if (bufp->buffer == NULL) \
1799 return REG_ESPACE; \
1800 /* If the buffer moved, move all the pointers into it. */ \
1801 if (old_buffer != bufp->buffer) \
1802 { \
1803 unsigned char *new_buffer = bufp->buffer; \
1804 MOVE_BUFFER_POINTER (b); \
1805 MOVE_BUFFER_POINTER (begalt); \
1806 if (fixup_alt_jump) \
1807 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1808 if (laststart) \
1809 MOVE_BUFFER_POINTER (laststart); \
1810 if (pending_exact) \
1811 MOVE_BUFFER_POINTER (pending_exact); \
1812 } \
1813 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1814 } while (0)
1815
1816
1817 /* Since we have one byte reserved for the register number argument to
1818 {start,stop}_memory, the maximum number of groups we can report
1819 things about is what fits in that byte. */
1820 #define MAX_REGNUM 255
1821
1822 /* But patterns can have more than `MAX_REGNUM' registers. We just
1823 ignore the excess. */
1824 typedef int regnum_t;
1825
1826
1827 /* Macros for the compile stack. */
1828
1829 /* Since offsets can go either forwards or backwards, this type needs to
1830 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1831 /* int may be not enough when sizeof(int) == 2. */
1832 typedef long pattern_offset_t;
1833
1834 typedef struct
1835 {
1836 pattern_offset_t begalt_offset;
1837 pattern_offset_t fixup_alt_jump;
1838 pattern_offset_t laststart_offset;
1839 regnum_t regnum;
1840 } compile_stack_elt_t;
1841
1842
1843 typedef struct
1844 {
1845 compile_stack_elt_t *stack;
1846 size_t size;
1847 size_t avail; /* Offset of next open position. */
1848 } compile_stack_type;
1849
1850
1851 #define INIT_COMPILE_STACK_SIZE 32
1852
1853 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1854 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1855
1856 /* The next available element. */
1857 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1858
1859 /* Explicit quit checking is only used on NTemacs and whenever we
1860 use polling to process input events. */
1861 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1862 extern int immediate_quit;
1863 # define IMMEDIATE_QUIT_CHECK \
1864 do { \
1865 if (immediate_quit) QUIT; \
1866 } while (0)
1867 #else
1868 # define IMMEDIATE_QUIT_CHECK ((void)0)
1869 #endif
1870 \f
1871 /* Structure to manage work area for range table. */
1872 struct range_table_work_area
1873 {
1874 int *table; /* actual work area. */
1875 int allocated; /* allocated size for work area in bytes. */
1876 int used; /* actually used size in words. */
1877 int bits; /* flag to record character classes */
1878 };
1879
1880 /* Make sure that WORK_AREA can hold more N multibyte characters.
1881 This is used only in set_image_of_range and set_image_of_range_1.
1882 It expects WORK_AREA to be a pointer.
1883 If it can't get the space, it returns from the surrounding function. */
1884
1885 #define EXTEND_RANGE_TABLE(work_area, n) \
1886 do { \
1887 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1888 { \
1889 extend_range_table_work_area (&work_area); \
1890 if ((work_area).table == 0) \
1891 return (REG_ESPACE); \
1892 } \
1893 } while (0)
1894
1895 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1896 (work_area).bits |= (bit)
1897
1898 /* Bits used to implement the multibyte-part of the various character classes
1899 such as [:alnum:] in a charset's range table. */
1900 #define BIT_WORD 0x1
1901 #define BIT_LOWER 0x2
1902 #define BIT_PUNCT 0x4
1903 #define BIT_SPACE 0x8
1904 #define BIT_UPPER 0x10
1905 #define BIT_MULTIBYTE 0x20
1906
1907 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1908 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1909 do { \
1910 EXTEND_RANGE_TABLE ((work_area), 2); \
1911 (work_area).table[(work_area).used++] = (range_start); \
1912 (work_area).table[(work_area).used++] = (range_end); \
1913 } while (0)
1914
1915 /* Free allocated memory for WORK_AREA. */
1916 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1917 do { \
1918 if ((work_area).table) \
1919 free ((work_area).table); \
1920 } while (0)
1921
1922 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1923 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1924 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1925 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1926 \f
1927
1928 /* Set the bit for character C in a list. */
1929 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1930
1931
1932 #ifdef emacs
1933
1934 /* Store characters in the range FROM to TO in the bitmap at B (for
1935 ASCII and unibyte characters) and WORK_AREA (for multibyte
1936 characters) while translating them and paying attention to the
1937 continuity of translated characters.
1938
1939 Implementation note: It is better to implement these fairly big
1940 macros by a function, but it's not that easy because macros called
1941 in this macro assume various local variables already declared. */
1942
1943 /* Both FROM and TO are ASCII characters. */
1944
1945 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1946 do { \
1947 int C0, C1; \
1948 \
1949 for (C0 = (FROM); C0 <= (TO); C0++) \
1950 { \
1951 C1 = TRANSLATE (C0); \
1952 if (! ASCII_CHAR_P (C1)) \
1953 { \
1954 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1955 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1956 C1 = C0; \
1957 } \
1958 SET_LIST_BIT (C1); \
1959 } \
1960 } while (0)
1961
1962
1963 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1964
1965 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1966 do { \
1967 int C0, C1, C2, I; \
1968 int USED = RANGE_TABLE_WORK_USED (work_area); \
1969 \
1970 for (C0 = (FROM); C0 <= (TO); C0++) \
1971 { \
1972 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1973 if (CHAR_BYTE8_P (C1)) \
1974 SET_LIST_BIT (C0); \
1975 else \
1976 { \
1977 C2 = TRANSLATE (C1); \
1978 if (C2 == C1 \
1979 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1980 C1 = C0; \
1981 SET_LIST_BIT (C1); \
1982 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1983 { \
1984 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1985 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1986 \
1987 if (C2 >= from - 1 && C2 <= to + 1) \
1988 { \
1989 if (C2 == from - 1) \
1990 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1991 else if (C2 == to + 1) \
1992 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1993 break; \
1994 } \
1995 } \
1996 if (I < USED) \
1997 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1998 } \
1999 } \
2000 } while (0)
2001
2002
2003 /* Both FROM and TO are multibyte characters. */
2004
2005 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2006 do { \
2007 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2008 \
2009 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2010 for (C0 = (FROM); C0 <= (TO); C0++) \
2011 { \
2012 C1 = TRANSLATE (C0); \
2013 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2014 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2015 SET_LIST_BIT (C2); \
2016 if (C1 >= (FROM) && C1 <= (TO)) \
2017 continue; \
2018 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2019 { \
2020 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2021 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2022 \
2023 if (C1 >= from - 1 && C1 <= to + 1) \
2024 { \
2025 if (C1 == from - 1) \
2026 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2027 else if (C1 == to + 1) \
2028 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2029 break; \
2030 } \
2031 } \
2032 if (I < USED) \
2033 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2034 } \
2035 } while (0)
2036
2037 #endif /* emacs */
2038
2039 /* Get the next unsigned number in the uncompiled pattern. */
2040 #define GET_UNSIGNED_NUMBER(num) \
2041 do { \
2042 if (p == pend) \
2043 FREE_STACK_RETURN (REG_EBRACE); \
2044 else \
2045 { \
2046 PATFETCH (c); \
2047 while ('0' <= c && c <= '9') \
2048 { \
2049 int prev; \
2050 if (num < 0) \
2051 num = 0; \
2052 prev = num; \
2053 num = num * 10 + c - '0'; \
2054 if (num / 10 != prev) \
2055 FREE_STACK_RETURN (REG_BADBR); \
2056 if (p == pend) \
2057 FREE_STACK_RETURN (REG_EBRACE); \
2058 PATFETCH (c); \
2059 } \
2060 } \
2061 } while (0)
2062 \f
2063 #if ! WIDE_CHAR_SUPPORT
2064
2065 /* Map a string to the char class it names (if any). */
2066 re_wctype_t
2067 re_wctype (const re_char *str)
2068 {
2069 const char *string = (const char *) str;
2070 if (STREQ (string, "alnum")) return RECC_ALNUM;
2071 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2072 else if (STREQ (string, "word")) return RECC_WORD;
2073 else if (STREQ (string, "ascii")) return RECC_ASCII;
2074 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2075 else if (STREQ (string, "graph")) return RECC_GRAPH;
2076 else if (STREQ (string, "lower")) return RECC_LOWER;
2077 else if (STREQ (string, "print")) return RECC_PRINT;
2078 else if (STREQ (string, "punct")) return RECC_PUNCT;
2079 else if (STREQ (string, "space")) return RECC_SPACE;
2080 else if (STREQ (string, "upper")) return RECC_UPPER;
2081 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2082 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2083 else if (STREQ (string, "digit")) return RECC_DIGIT;
2084 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2085 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2086 else if (STREQ (string, "blank")) return RECC_BLANK;
2087 else return 0;
2088 }
2089
2090 /* True if CH is in the char class CC. */
2091 boolean
2092 re_iswctype (int ch, re_wctype_t cc)
2093 {
2094 switch (cc)
2095 {
2096 case RECC_ALNUM: return ISALNUM (ch);
2097 case RECC_ALPHA: return ISALPHA (ch);
2098 case RECC_BLANK: return ISBLANK (ch);
2099 case RECC_CNTRL: return ISCNTRL (ch);
2100 case RECC_DIGIT: return ISDIGIT (ch);
2101 case RECC_GRAPH: return ISGRAPH (ch);
2102 case RECC_LOWER: return ISLOWER (ch);
2103 case RECC_PRINT: return ISPRINT (ch);
2104 case RECC_PUNCT: return ISPUNCT (ch);
2105 case RECC_SPACE: return ISSPACE (ch);
2106 case RECC_UPPER: return ISUPPER (ch);
2107 case RECC_XDIGIT: return ISXDIGIT (ch);
2108 case RECC_ASCII: return IS_REAL_ASCII (ch);
2109 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2110 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2111 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2112 case RECC_WORD: return ISWORD (ch);
2113 case RECC_ERROR: return false;
2114 default:
2115 abort();
2116 }
2117 }
2118
2119 /* Return a bit-pattern to use in the range-table bits to match multibyte
2120 chars of class CC. */
2121 static int
2122 re_wctype_to_bit (re_wctype_t cc)
2123 {
2124 switch (cc)
2125 {
2126 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2127 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2128 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2129 case RECC_LOWER: return BIT_LOWER;
2130 case RECC_UPPER: return BIT_UPPER;
2131 case RECC_PUNCT: return BIT_PUNCT;
2132 case RECC_SPACE: return BIT_SPACE;
2133 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2134 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2135 default:
2136 abort();
2137 }
2138 }
2139 #endif
2140 \f
2141 /* Filling in the work area of a range. */
2142
2143 /* Actually extend the space in WORK_AREA. */
2144
2145 static void
2146 extend_range_table_work_area (struct range_table_work_area *work_area)
2147 {
2148 work_area->allocated += 16 * sizeof (int);
2149 if (work_area->table)
2150 work_area->table
2151 = (int *) realloc (work_area->table, work_area->allocated);
2152 else
2153 work_area->table
2154 = (int *) malloc (work_area->allocated);
2155 }
2156
2157 #if 0
2158 #ifdef emacs
2159
2160 /* Carefully find the ranges of codes that are equivalent
2161 under case conversion to the range start..end when passed through
2162 TRANSLATE. Handle the case where non-letters can come in between
2163 two upper-case letters (which happens in Latin-1).
2164 Also handle the case of groups of more than 2 case-equivalent chars.
2165
2166 The basic method is to look at consecutive characters and see
2167 if they can form a run that can be handled as one.
2168
2169 Returns -1 if successful, REG_ESPACE if ran out of space. */
2170
2171 static int
2172 set_image_of_range_1 (struct range_table_work_area *work_area,
2173 re_wchar_t start, re_wchar_t end,
2174 RE_TRANSLATE_TYPE translate)
2175 {
2176 /* `one_case' indicates a character, or a run of characters,
2177 each of which is an isolate (no case-equivalents).
2178 This includes all ASCII non-letters.
2179
2180 `two_case' indicates a character, or a run of characters,
2181 each of which has two case-equivalent forms.
2182 This includes all ASCII letters.
2183
2184 `strange' indicates a character that has more than one
2185 case-equivalent. */
2186
2187 enum case_type {one_case, two_case, strange};
2188
2189 /* Describe the run that is in progress,
2190 which the next character can try to extend.
2191 If run_type is strange, that means there really is no run.
2192 If run_type is one_case, then run_start...run_end is the run.
2193 If run_type is two_case, then the run is run_start...run_end,
2194 and the case-equivalents end at run_eqv_end. */
2195
2196 enum case_type run_type = strange;
2197 int run_start, run_end, run_eqv_end;
2198
2199 Lisp_Object eqv_table;
2200
2201 if (!RE_TRANSLATE_P (translate))
2202 {
2203 EXTEND_RANGE_TABLE (work_area, 2);
2204 work_area->table[work_area->used++] = (start);
2205 work_area->table[work_area->used++] = (end);
2206 return -1;
2207 }
2208
2209 eqv_table = XCHAR_TABLE (translate)->extras[2];
2210
2211 for (; start <= end; start++)
2212 {
2213 enum case_type this_type;
2214 int eqv = RE_TRANSLATE (eqv_table, start);
2215 int minchar, maxchar;
2216
2217 /* Classify this character */
2218 if (eqv == start)
2219 this_type = one_case;
2220 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2221 this_type = two_case;
2222 else
2223 this_type = strange;
2224
2225 if (start < eqv)
2226 minchar = start, maxchar = eqv;
2227 else
2228 minchar = eqv, maxchar = start;
2229
2230 /* Can this character extend the run in progress? */
2231 if (this_type == strange || this_type != run_type
2232 || !(minchar == run_end + 1
2233 && (run_type == two_case
2234 ? maxchar == run_eqv_end + 1 : 1)))
2235 {
2236 /* No, end the run.
2237 Record each of its equivalent ranges. */
2238 if (run_type == one_case)
2239 {
2240 EXTEND_RANGE_TABLE (work_area, 2);
2241 work_area->table[work_area->used++] = run_start;
2242 work_area->table[work_area->used++] = run_end;
2243 }
2244 else if (run_type == two_case)
2245 {
2246 EXTEND_RANGE_TABLE (work_area, 4);
2247 work_area->table[work_area->used++] = run_start;
2248 work_area->table[work_area->used++] = run_end;
2249 work_area->table[work_area->used++]
2250 = RE_TRANSLATE (eqv_table, run_start);
2251 work_area->table[work_area->used++]
2252 = RE_TRANSLATE (eqv_table, run_end);
2253 }
2254 run_type = strange;
2255 }
2256
2257 if (this_type == strange)
2258 {
2259 /* For a strange character, add each of its equivalents, one
2260 by one. Don't start a range. */
2261 do
2262 {
2263 EXTEND_RANGE_TABLE (work_area, 2);
2264 work_area->table[work_area->used++] = eqv;
2265 work_area->table[work_area->used++] = eqv;
2266 eqv = RE_TRANSLATE (eqv_table, eqv);
2267 }
2268 while (eqv != start);
2269 }
2270
2271 /* Add this char to the run, or start a new run. */
2272 else if (run_type == strange)
2273 {
2274 /* Initialize a new range. */
2275 run_type = this_type;
2276 run_start = start;
2277 run_end = start;
2278 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2279 }
2280 else
2281 {
2282 /* Extend a running range. */
2283 run_end = minchar;
2284 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2285 }
2286 }
2287
2288 /* If a run is still in progress at the end, finish it now
2289 by recording its equivalent ranges. */
2290 if (run_type == one_case)
2291 {
2292 EXTEND_RANGE_TABLE (work_area, 2);
2293 work_area->table[work_area->used++] = run_start;
2294 work_area->table[work_area->used++] = run_end;
2295 }
2296 else if (run_type == two_case)
2297 {
2298 EXTEND_RANGE_TABLE (work_area, 4);
2299 work_area->table[work_area->used++] = run_start;
2300 work_area->table[work_area->used++] = run_end;
2301 work_area->table[work_area->used++]
2302 = RE_TRANSLATE (eqv_table, run_start);
2303 work_area->table[work_area->used++]
2304 = RE_TRANSLATE (eqv_table, run_end);
2305 }
2306
2307 return -1;
2308 }
2309
2310 #endif /* emacs */
2311
2312 /* Record the image of the range start..end when passed through
2313 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2314 and is not even necessarily contiguous.
2315 Normally we approximate it with the smallest contiguous range that contains
2316 all the chars we need. However, for Latin-1 we go to extra effort
2317 to do a better job.
2318
2319 This function is not called for ASCII ranges.
2320
2321 Returns -1 if successful, REG_ESPACE if ran out of space. */
2322
2323 static int
2324 set_image_of_range (struct range_table_work_area *work_area,
2325 re_wchar_t start, re_wchar_t end,
2326 RE_TRANSLATE_TYPE translate)
2327 {
2328 re_wchar_t cmin, cmax;
2329
2330 #ifdef emacs
2331 /* For Latin-1 ranges, use set_image_of_range_1
2332 to get proper handling of ranges that include letters and nonletters.
2333 For a range that includes the whole of Latin-1, this is not necessary.
2334 For other character sets, we don't bother to get this right. */
2335 if (RE_TRANSLATE_P (translate) && start < 04400
2336 && !(start < 04200 && end >= 04377))
2337 {
2338 int newend;
2339 int tem;
2340 newend = end;
2341 if (newend > 04377)
2342 newend = 04377;
2343 tem = set_image_of_range_1 (work_area, start, newend, translate);
2344 if (tem > 0)
2345 return tem;
2346
2347 start = 04400;
2348 if (end < 04400)
2349 return -1;
2350 }
2351 #endif
2352
2353 EXTEND_RANGE_TABLE (work_area, 2);
2354 work_area->table[work_area->used++] = (start);
2355 work_area->table[work_area->used++] = (end);
2356
2357 cmin = -1, cmax = -1;
2358
2359 if (RE_TRANSLATE_P (translate))
2360 {
2361 int ch;
2362
2363 for (ch = start; ch <= end; ch++)
2364 {
2365 re_wchar_t c = TRANSLATE (ch);
2366 if (! (start <= c && c <= end))
2367 {
2368 if (cmin == -1)
2369 cmin = c, cmax = c;
2370 else
2371 {
2372 cmin = MIN (cmin, c);
2373 cmax = MAX (cmax, c);
2374 }
2375 }
2376 }
2377
2378 if (cmin != -1)
2379 {
2380 EXTEND_RANGE_TABLE (work_area, 2);
2381 work_area->table[work_area->used++] = (cmin);
2382 work_area->table[work_area->used++] = (cmax);
2383 }
2384 }
2385
2386 return -1;
2387 }
2388 #endif /* 0 */
2389 \f
2390 #ifndef MATCH_MAY_ALLOCATE
2391
2392 /* If we cannot allocate large objects within re_match_2_internal,
2393 we make the fail stack and register vectors global.
2394 The fail stack, we grow to the maximum size when a regexp
2395 is compiled.
2396 The register vectors, we adjust in size each time we
2397 compile a regexp, according to the number of registers it needs. */
2398
2399 static fail_stack_type fail_stack;
2400
2401 /* Size with which the following vectors are currently allocated.
2402 That is so we can make them bigger as needed,
2403 but never make them smaller. */
2404 static int regs_allocated_size;
2405
2406 static re_char ** regstart, ** regend;
2407 static re_char **best_regstart, **best_regend;
2408
2409 /* Make the register vectors big enough for NUM_REGS registers,
2410 but don't make them smaller. */
2411
2412 static
2413 regex_grow_registers (int num_regs)
2414 {
2415 if (num_regs > regs_allocated_size)
2416 {
2417 RETALLOC_IF (regstart, num_regs, re_char *);
2418 RETALLOC_IF (regend, num_regs, re_char *);
2419 RETALLOC_IF (best_regstart, num_regs, re_char *);
2420 RETALLOC_IF (best_regend, num_regs, re_char *);
2421
2422 regs_allocated_size = num_regs;
2423 }
2424 }
2425
2426 #endif /* not MATCH_MAY_ALLOCATE */
2427 \f
2428 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2429 compile_stack,
2430 regnum_t regnum));
2431
2432 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2433 Returns one of error codes defined in `regex.h', or zero for success.
2434
2435 Assumes the `allocated' (and perhaps `buffer') and `translate'
2436 fields are set in BUFP on entry.
2437
2438 If it succeeds, results are put in BUFP (if it returns an error, the
2439 contents of BUFP are undefined):
2440 `buffer' is the compiled pattern;
2441 `syntax' is set to SYNTAX;
2442 `used' is set to the length of the compiled pattern;
2443 `fastmap_accurate' is zero;
2444 `re_nsub' is the number of subexpressions in PATTERN;
2445 `not_bol' and `not_eol' are zero;
2446
2447 The `fastmap' field is neither examined nor set. */
2448
2449 /* Insert the `jump' from the end of last alternative to "here".
2450 The space for the jump has already been allocated. */
2451 #define FIXUP_ALT_JUMP() \
2452 do { \
2453 if (fixup_alt_jump) \
2454 STORE_JUMP (jump, fixup_alt_jump, b); \
2455 } while (0)
2456
2457
2458 /* Return, freeing storage we allocated. */
2459 #define FREE_STACK_RETURN(value) \
2460 do { \
2461 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2462 free (compile_stack.stack); \
2463 return value; \
2464 } while (0)
2465
2466 static reg_errcode_t
2467 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2468 {
2469 /* We fetch characters from PATTERN here. */
2470 register re_wchar_t c, c1;
2471
2472 /* Points to the end of the buffer, where we should append. */
2473 register unsigned char *b;
2474
2475 /* Keeps track of unclosed groups. */
2476 compile_stack_type compile_stack;
2477
2478 /* Points to the current (ending) position in the pattern. */
2479 #ifdef AIX
2480 /* `const' makes AIX compiler fail. */
2481 unsigned char *p = pattern;
2482 #else
2483 re_char *p = pattern;
2484 #endif
2485 re_char *pend = pattern + size;
2486
2487 /* How to translate the characters in the pattern. */
2488 RE_TRANSLATE_TYPE translate = bufp->translate;
2489
2490 /* Address of the count-byte of the most recently inserted `exactn'
2491 command. This makes it possible to tell if a new exact-match
2492 character can be added to that command or if the character requires
2493 a new `exactn' command. */
2494 unsigned char *pending_exact = 0;
2495
2496 /* Address of start of the most recently finished expression.
2497 This tells, e.g., postfix * where to find the start of its
2498 operand. Reset at the beginning of groups and alternatives. */
2499 unsigned char *laststart = 0;
2500
2501 /* Address of beginning of regexp, or inside of last group. */
2502 unsigned char *begalt;
2503
2504 /* Place in the uncompiled pattern (i.e., the {) to
2505 which to go back if the interval is invalid. */
2506 re_char *beg_interval;
2507
2508 /* Address of the place where a forward jump should go to the end of
2509 the containing expression. Each alternative of an `or' -- except the
2510 last -- ends with a forward jump of this sort. */
2511 unsigned char *fixup_alt_jump = 0;
2512
2513 /* Work area for range table of charset. */
2514 struct range_table_work_area range_table_work;
2515
2516 /* If the object matched can contain multibyte characters. */
2517 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2518
2519 /* Nonzero if we have pushed down into a subpattern. */
2520 int in_subpattern = 0;
2521
2522 /* These hold the values of p, pattern, and pend from the main
2523 pattern when we have pushed into a subpattern. */
2524 re_char *main_p IF_LINT (= NULL);
2525 re_char *main_pattern IF_LINT (= NULL);
2526 re_char *main_pend IF_LINT (= NULL);
2527
2528 #ifdef DEBUG
2529 debug++;
2530 DEBUG_PRINT1 ("\nCompiling pattern: ");
2531 if (debug > 0)
2532 {
2533 unsigned debug_count;
2534
2535 for (debug_count = 0; debug_count < size; debug_count++)
2536 putchar (pattern[debug_count]);
2537 putchar ('\n');
2538 }
2539 #endif /* DEBUG */
2540
2541 /* Initialize the compile stack. */
2542 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2543 if (compile_stack.stack == NULL)
2544 return REG_ESPACE;
2545
2546 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2547 compile_stack.avail = 0;
2548
2549 range_table_work.table = 0;
2550 range_table_work.allocated = 0;
2551
2552 /* Initialize the pattern buffer. */
2553 bufp->syntax = syntax;
2554 bufp->fastmap_accurate = 0;
2555 bufp->not_bol = bufp->not_eol = 0;
2556 bufp->used_syntax = 0;
2557
2558 /* Set `used' to zero, so that if we return an error, the pattern
2559 printer (for debugging) will think there's no pattern. We reset it
2560 at the end. */
2561 bufp->used = 0;
2562
2563 /* Always count groups, whether or not bufp->no_sub is set. */
2564 bufp->re_nsub = 0;
2565
2566 #if !defined emacs && !defined SYNTAX_TABLE
2567 /* Initialize the syntax table. */
2568 init_syntax_once ();
2569 #endif
2570
2571 if (bufp->allocated == 0)
2572 {
2573 if (bufp->buffer)
2574 { /* If zero allocated, but buffer is non-null, try to realloc
2575 enough space. This loses if buffer's address is bogus, but
2576 that is the user's responsibility. */
2577 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2578 }
2579 else
2580 { /* Caller did not allocate a buffer. Do it for them. */
2581 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2582 }
2583 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2584
2585 bufp->allocated = INIT_BUF_SIZE;
2586 }
2587
2588 begalt = b = bufp->buffer;
2589
2590 /* Loop through the uncompiled pattern until we're at the end. */
2591 while (1)
2592 {
2593 if (p == pend)
2594 {
2595 /* If this is the end of an included regexp,
2596 pop back to the main regexp and try again. */
2597 if (in_subpattern)
2598 {
2599 in_subpattern = 0;
2600 pattern = main_pattern;
2601 p = main_p;
2602 pend = main_pend;
2603 continue;
2604 }
2605 /* If this is the end of the main regexp, we are done. */
2606 break;
2607 }
2608
2609 PATFETCH (c);
2610
2611 switch (c)
2612 {
2613 case ' ':
2614 {
2615 re_char *p1 = p;
2616
2617 /* If there's no special whitespace regexp, treat
2618 spaces normally. And don't try to do this recursively. */
2619 if (!whitespace_regexp || in_subpattern)
2620 goto normal_char;
2621
2622 /* Peek past following spaces. */
2623 while (p1 != pend)
2624 {
2625 if (*p1 != ' ')
2626 break;
2627 p1++;
2628 }
2629 /* If the spaces are followed by a repetition op,
2630 treat them normally. */
2631 if (p1 != pend
2632 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2633 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2634 goto normal_char;
2635
2636 /* Replace the spaces with the whitespace regexp. */
2637 in_subpattern = 1;
2638 main_p = p1;
2639 main_pend = pend;
2640 main_pattern = pattern;
2641 p = pattern = whitespace_regexp;
2642 pend = p + strlen ((const char *) p);
2643 break;
2644 }
2645
2646 case '^':
2647 {
2648 if ( /* If at start of pattern, it's an operator. */
2649 p == pattern + 1
2650 /* If context independent, it's an operator. */
2651 || syntax & RE_CONTEXT_INDEP_ANCHORS
2652 /* Otherwise, depends on what's come before. */
2653 || at_begline_loc_p (pattern, p, syntax))
2654 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2655 else
2656 goto normal_char;
2657 }
2658 break;
2659
2660
2661 case '$':
2662 {
2663 if ( /* If at end of pattern, it's an operator. */
2664 p == pend
2665 /* If context independent, it's an operator. */
2666 || syntax & RE_CONTEXT_INDEP_ANCHORS
2667 /* Otherwise, depends on what's next. */
2668 || at_endline_loc_p (p, pend, syntax))
2669 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2670 else
2671 goto normal_char;
2672 }
2673 break;
2674
2675
2676 case '+':
2677 case '?':
2678 if ((syntax & RE_BK_PLUS_QM)
2679 || (syntax & RE_LIMITED_OPS))
2680 goto normal_char;
2681 handle_plus:
2682 case '*':
2683 /* If there is no previous pattern... */
2684 if (!laststart)
2685 {
2686 if (syntax & RE_CONTEXT_INVALID_OPS)
2687 FREE_STACK_RETURN (REG_BADRPT);
2688 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2689 goto normal_char;
2690 }
2691
2692 {
2693 /* 1 means zero (many) matches is allowed. */
2694 boolean zero_times_ok = 0, many_times_ok = 0;
2695 boolean greedy = 1;
2696
2697 /* If there is a sequence of repetition chars, collapse it
2698 down to just one (the right one). We can't combine
2699 interval operators with these because of, e.g., `a{2}*',
2700 which should only match an even number of `a's. */
2701
2702 for (;;)
2703 {
2704 if ((syntax & RE_FRUGAL)
2705 && c == '?' && (zero_times_ok || many_times_ok))
2706 greedy = 0;
2707 else
2708 {
2709 zero_times_ok |= c != '+';
2710 many_times_ok |= c != '?';
2711 }
2712
2713 if (p == pend)
2714 break;
2715 else if (*p == '*'
2716 || (!(syntax & RE_BK_PLUS_QM)
2717 && (*p == '+' || *p == '?')))
2718 ;
2719 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2720 {
2721 if (p+1 == pend)
2722 FREE_STACK_RETURN (REG_EESCAPE);
2723 if (p[1] == '+' || p[1] == '?')
2724 PATFETCH (c); /* Gobble up the backslash. */
2725 else
2726 break;
2727 }
2728 else
2729 break;
2730 /* If we get here, we found another repeat character. */
2731 PATFETCH (c);
2732 }
2733
2734 /* Star, etc. applied to an empty pattern is equivalent
2735 to an empty pattern. */
2736 if (!laststart || laststart == b)
2737 break;
2738
2739 /* Now we know whether or not zero matches is allowed
2740 and also whether or not two or more matches is allowed. */
2741 if (greedy)
2742 {
2743 if (many_times_ok)
2744 {
2745 boolean simple = skip_one_char (laststart) == b;
2746 size_t startoffset = 0;
2747 re_opcode_t ofj =
2748 /* Check if the loop can match the empty string. */
2749 (simple || !analyse_first (laststart, b, NULL, 0))
2750 ? on_failure_jump : on_failure_jump_loop;
2751 assert (skip_one_char (laststart) <= b);
2752
2753 if (!zero_times_ok && simple)
2754 { /* Since simple * loops can be made faster by using
2755 on_failure_keep_string_jump, we turn simple P+
2756 into PP* if P is simple. */
2757 unsigned char *p1, *p2;
2758 startoffset = b - laststart;
2759 GET_BUFFER_SPACE (startoffset);
2760 p1 = b; p2 = laststart;
2761 while (p2 < p1)
2762 *b++ = *p2++;
2763 zero_times_ok = 1;
2764 }
2765
2766 GET_BUFFER_SPACE (6);
2767 if (!zero_times_ok)
2768 /* A + loop. */
2769 STORE_JUMP (ofj, b, b + 6);
2770 else
2771 /* Simple * loops can use on_failure_keep_string_jump
2772 depending on what follows. But since we don't know
2773 that yet, we leave the decision up to
2774 on_failure_jump_smart. */
2775 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2776 laststart + startoffset, b + 6);
2777 b += 3;
2778 STORE_JUMP (jump, b, laststart + startoffset);
2779 b += 3;
2780 }
2781 else
2782 {
2783 /* A simple ? pattern. */
2784 assert (zero_times_ok);
2785 GET_BUFFER_SPACE (3);
2786 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2787 b += 3;
2788 }
2789 }
2790 else /* not greedy */
2791 { /* I wish the greedy and non-greedy cases could be merged. */
2792
2793 GET_BUFFER_SPACE (7); /* We might use less. */
2794 if (many_times_ok)
2795 {
2796 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2797
2798 /* The non-greedy multiple match looks like
2799 a repeat..until: we only need a conditional jump
2800 at the end of the loop. */
2801 if (emptyp) BUF_PUSH (no_op);
2802 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2803 : on_failure_jump, b, laststart);
2804 b += 3;
2805 if (zero_times_ok)
2806 {
2807 /* The repeat...until naturally matches one or more.
2808 To also match zero times, we need to first jump to
2809 the end of the loop (its conditional jump). */
2810 INSERT_JUMP (jump, laststart, b);
2811 b += 3;
2812 }
2813 }
2814 else
2815 {
2816 /* non-greedy a?? */
2817 INSERT_JUMP (jump, laststart, b + 3);
2818 b += 3;
2819 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2820 b += 3;
2821 }
2822 }
2823 }
2824 pending_exact = 0;
2825 break;
2826
2827
2828 case '.':
2829 laststart = b;
2830 BUF_PUSH (anychar);
2831 break;
2832
2833
2834 case '[':
2835 {
2836 re_char *p1;
2837
2838 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2839
2840 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2841
2842 /* Ensure that we have enough space to push a charset: the
2843 opcode, the length count, and the bitset; 34 bytes in all. */
2844 GET_BUFFER_SPACE (34);
2845
2846 laststart = b;
2847
2848 /* We test `*p == '^' twice, instead of using an if
2849 statement, so we only need one BUF_PUSH. */
2850 BUF_PUSH (*p == '^' ? charset_not : charset);
2851 if (*p == '^')
2852 p++;
2853
2854 /* Remember the first position in the bracket expression. */
2855 p1 = p;
2856
2857 /* Push the number of bytes in the bitmap. */
2858 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2859
2860 /* Clear the whole map. */
2861 memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH);
2862
2863 /* charset_not matches newline according to a syntax bit. */
2864 if ((re_opcode_t) b[-2] == charset_not
2865 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2866 SET_LIST_BIT ('\n');
2867
2868 /* Read in characters and ranges, setting map bits. */
2869 for (;;)
2870 {
2871 boolean escaped_char = false;
2872 const unsigned char *p2 = p;
2873 re_wchar_t ch;
2874
2875 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2876
2877 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2878 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2879 So the translation is done later in a loop. Example:
2880 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2881 PATFETCH (c);
2882
2883 /* \ might escape characters inside [...] and [^...]. */
2884 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2885 {
2886 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2887
2888 PATFETCH (c);
2889 escaped_char = true;
2890 }
2891 else
2892 {
2893 /* Could be the end of the bracket expression. If it's
2894 not (i.e., when the bracket expression is `[]' so
2895 far), the ']' character bit gets set way below. */
2896 if (c == ']' && p2 != p1)
2897 break;
2898 }
2899
2900 /* See if we're at the beginning of a possible character
2901 class. */
2902
2903 if (!escaped_char &&
2904 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2905 {
2906 /* Leave room for the null. */
2907 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2908 const unsigned char *class_beg;
2909
2910 PATFETCH (c);
2911 c1 = 0;
2912 class_beg = p;
2913
2914 /* If pattern is `[[:'. */
2915 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2916
2917 for (;;)
2918 {
2919 PATFETCH (c);
2920 if ((c == ':' && *p == ']') || p == pend)
2921 break;
2922 if (c1 < CHAR_CLASS_MAX_LENGTH)
2923 str[c1++] = c;
2924 else
2925 /* This is in any case an invalid class name. */
2926 str[0] = '\0';
2927 }
2928 str[c1] = '\0';
2929
2930 /* If isn't a word bracketed by `[:' and `:]':
2931 undo the ending character, the letters, and
2932 leave the leading `:' and `[' (but set bits for
2933 them). */
2934 if (c == ':' && *p == ']')
2935 {
2936 re_wctype_t cc = re_wctype (str);
2937
2938 if (cc == 0)
2939 FREE_STACK_RETURN (REG_ECTYPE);
2940
2941 /* Throw away the ] at the end of the character
2942 class. */
2943 PATFETCH (c);
2944
2945 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2946
2947 #ifndef emacs
2948 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
2949 if (re_iswctype (btowc (ch), cc))
2950 {
2951 c = TRANSLATE (ch);
2952 if (c < (1 << BYTEWIDTH))
2953 SET_LIST_BIT (c);
2954 }
2955 #else /* emacs */
2956 /* Most character classes in a multibyte match
2957 just set a flag. Exceptions are is_blank,
2958 is_digit, is_cntrl, and is_xdigit, since
2959 they can only match ASCII characters. We
2960 don't need to handle them for multibyte.
2961 They are distinguished by a negative wctype. */
2962
2963 /* Setup the gl_state object to its buffer-defined
2964 value. This hardcodes the buffer-global
2965 syntax-table for ASCII chars, while the other chars
2966 will obey syntax-table properties. It's not ideal,
2967 but it's the way it's been done until now. */
2968 SETUP_BUFFER_SYNTAX_TABLE ();
2969
2970 for (ch = 0; ch < 256; ++ch)
2971 {
2972 c = RE_CHAR_TO_MULTIBYTE (ch);
2973 if (! CHAR_BYTE8_P (c)
2974 && re_iswctype (c, cc))
2975 {
2976 SET_LIST_BIT (ch);
2977 c1 = TRANSLATE (c);
2978 if (c1 == c)
2979 continue;
2980 if (ASCII_CHAR_P (c1))
2981 SET_LIST_BIT (c1);
2982 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
2983 SET_LIST_BIT (c1);
2984 }
2985 }
2986 SET_RANGE_TABLE_WORK_AREA_BIT
2987 (range_table_work, re_wctype_to_bit (cc));
2988 #endif /* emacs */
2989 /* In most cases the matching rule for char classes
2990 only uses the syntax table for multibyte chars,
2991 so that the content of the syntax-table it is not
2992 hardcoded in the range_table. SPACE and WORD are
2993 the two exceptions. */
2994 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
2995 bufp->used_syntax = 1;
2996
2997 /* Repeat the loop. */
2998 continue;
2999 }
3000 else
3001 {
3002 /* Go back to right after the "[:". */
3003 p = class_beg;
3004 SET_LIST_BIT ('[');
3005
3006 /* Because the `:' may starts the range, we
3007 can't simply set bit and repeat the loop.
3008 Instead, just set it to C and handle below. */
3009 c = ':';
3010 }
3011 }
3012
3013 if (p < pend && p[0] == '-' && p[1] != ']')
3014 {
3015
3016 /* Discard the `-'. */
3017 PATFETCH (c1);
3018
3019 /* Fetch the character which ends the range. */
3020 PATFETCH (c1);
3021 #ifdef emacs
3022 if (CHAR_BYTE8_P (c1)
3023 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3024 /* Treat the range from a multibyte character to
3025 raw-byte character as empty. */
3026 c = c1 + 1;
3027 #endif /* emacs */
3028 }
3029 else
3030 /* Range from C to C. */
3031 c1 = c;
3032
3033 if (c > c1)
3034 {
3035 if (syntax & RE_NO_EMPTY_RANGES)
3036 FREE_STACK_RETURN (REG_ERANGEX);
3037 /* Else, repeat the loop. */
3038 }
3039 else
3040 {
3041 #ifndef emacs
3042 /* Set the range into bitmap */
3043 for (; c <= c1; c++)
3044 {
3045 ch = TRANSLATE (c);
3046 if (ch < (1 << BYTEWIDTH))
3047 SET_LIST_BIT (ch);
3048 }
3049 #else /* emacs */
3050 if (c < 128)
3051 {
3052 ch = MIN (127, c1);
3053 SETUP_ASCII_RANGE (range_table_work, c, ch);
3054 c = ch + 1;
3055 if (CHAR_BYTE8_P (c1))
3056 c = BYTE8_TO_CHAR (128);
3057 }
3058 if (c <= c1)
3059 {
3060 if (CHAR_BYTE8_P (c))
3061 {
3062 c = CHAR_TO_BYTE8 (c);
3063 c1 = CHAR_TO_BYTE8 (c1);
3064 for (; c <= c1; c++)
3065 SET_LIST_BIT (c);
3066 }
3067 else if (multibyte)
3068 {
3069 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3070 }
3071 else
3072 {
3073 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3074 }
3075 }
3076 #endif /* emacs */
3077 }
3078 }
3079
3080 /* Discard any (non)matching list bytes that are all 0 at the
3081 end of the map. Decrease the map-length byte too. */
3082 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3083 b[-1]--;
3084 b += b[-1];
3085
3086 /* Build real range table from work area. */
3087 if (RANGE_TABLE_WORK_USED (range_table_work)
3088 || RANGE_TABLE_WORK_BITS (range_table_work))
3089 {
3090 int i;
3091 int used = RANGE_TABLE_WORK_USED (range_table_work);
3092
3093 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3094 bytes for flags, two for COUNT, and three bytes for
3095 each character. */
3096 GET_BUFFER_SPACE (4 + used * 3);
3097
3098 /* Indicate the existence of range table. */
3099 laststart[1] |= 0x80;
3100
3101 /* Store the character class flag bits into the range table.
3102 If not in emacs, these flag bits are always 0. */
3103 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3104 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3105
3106 STORE_NUMBER_AND_INCR (b, used / 2);
3107 for (i = 0; i < used; i++)
3108 STORE_CHARACTER_AND_INCR
3109 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3110 }
3111 }
3112 break;
3113
3114
3115 case '(':
3116 if (syntax & RE_NO_BK_PARENS)
3117 goto handle_open;
3118 else
3119 goto normal_char;
3120
3121
3122 case ')':
3123 if (syntax & RE_NO_BK_PARENS)
3124 goto handle_close;
3125 else
3126 goto normal_char;
3127
3128
3129 case '\n':
3130 if (syntax & RE_NEWLINE_ALT)
3131 goto handle_alt;
3132 else
3133 goto normal_char;
3134
3135
3136 case '|':
3137 if (syntax & RE_NO_BK_VBAR)
3138 goto handle_alt;
3139 else
3140 goto normal_char;
3141
3142
3143 case '{':
3144 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3145 goto handle_interval;
3146 else
3147 goto normal_char;
3148
3149
3150 case '\\':
3151 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3152
3153 /* Do not translate the character after the \, so that we can
3154 distinguish, e.g., \B from \b, even if we normally would
3155 translate, e.g., B to b. */
3156 PATFETCH (c);
3157
3158 switch (c)
3159 {
3160 case '(':
3161 if (syntax & RE_NO_BK_PARENS)
3162 goto normal_backslash;
3163
3164 handle_open:
3165 {
3166 int shy = 0;
3167 regnum_t regnum = 0;
3168 if (p+1 < pend)
3169 {
3170 /* Look for a special (?...) construct */
3171 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3172 {
3173 PATFETCH (c); /* Gobble up the '?'. */
3174 while (!shy)
3175 {
3176 PATFETCH (c);
3177 switch (c)
3178 {
3179 case ':': shy = 1; break;
3180 case '0':
3181 /* An explicitly specified regnum must start
3182 with non-0. */
3183 if (regnum == 0)
3184 FREE_STACK_RETURN (REG_BADPAT);
3185 case '1': case '2': case '3': case '4':
3186 case '5': case '6': case '7': case '8': case '9':
3187 regnum = 10*regnum + (c - '0'); break;
3188 default:
3189 /* Only (?:...) is supported right now. */
3190 FREE_STACK_RETURN (REG_BADPAT);
3191 }
3192 }
3193 }
3194 }
3195
3196 if (!shy)
3197 regnum = ++bufp->re_nsub;
3198 else if (regnum)
3199 { /* It's actually not shy, but explicitly numbered. */
3200 shy = 0;
3201 if (regnum > bufp->re_nsub)
3202 bufp->re_nsub = regnum;
3203 else if (regnum > bufp->re_nsub
3204 /* Ideally, we'd want to check that the specified
3205 group can't have matched (i.e. all subgroups
3206 using the same regnum are in other branches of
3207 OR patterns), but we don't currently keep track
3208 of enough info to do that easily. */
3209 || group_in_compile_stack (compile_stack, regnum))
3210 FREE_STACK_RETURN (REG_BADPAT);
3211 }
3212 else
3213 /* It's really shy. */
3214 regnum = - bufp->re_nsub;
3215
3216 if (COMPILE_STACK_FULL)
3217 {
3218 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3219 compile_stack_elt_t);
3220 if (compile_stack.stack == NULL) return REG_ESPACE;
3221
3222 compile_stack.size <<= 1;
3223 }
3224
3225 /* These are the values to restore when we hit end of this
3226 group. They are all relative offsets, so that if the
3227 whole pattern moves because of realloc, they will still
3228 be valid. */
3229 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3230 COMPILE_STACK_TOP.fixup_alt_jump
3231 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3232 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3233 COMPILE_STACK_TOP.regnum = regnum;
3234
3235 /* Do not push a start_memory for groups beyond the last one
3236 we can represent in the compiled pattern. */
3237 if (regnum <= MAX_REGNUM && regnum > 0)
3238 BUF_PUSH_2 (start_memory, regnum);
3239
3240 compile_stack.avail++;
3241
3242 fixup_alt_jump = 0;
3243 laststart = 0;
3244 begalt = b;
3245 /* If we've reached MAX_REGNUM groups, then this open
3246 won't actually generate any code, so we'll have to
3247 clear pending_exact explicitly. */
3248 pending_exact = 0;
3249 break;
3250 }
3251
3252 case ')':
3253 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3254
3255 if (COMPILE_STACK_EMPTY)
3256 {
3257 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3258 goto normal_backslash;
3259 else
3260 FREE_STACK_RETURN (REG_ERPAREN);
3261 }
3262
3263 handle_close:
3264 FIXUP_ALT_JUMP ();
3265
3266 /* See similar code for backslashed left paren above. */
3267 if (COMPILE_STACK_EMPTY)
3268 {
3269 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3270 goto normal_char;
3271 else
3272 FREE_STACK_RETURN (REG_ERPAREN);
3273 }
3274
3275 /* Since we just checked for an empty stack above, this
3276 ``can't happen''. */
3277 assert (compile_stack.avail != 0);
3278 {
3279 /* We don't just want to restore into `regnum', because
3280 later groups should continue to be numbered higher,
3281 as in `(ab)c(de)' -- the second group is #2. */
3282 regnum_t regnum;
3283
3284 compile_stack.avail--;
3285 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3286 fixup_alt_jump
3287 = COMPILE_STACK_TOP.fixup_alt_jump
3288 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3289 : 0;
3290 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3291 regnum = COMPILE_STACK_TOP.regnum;
3292 /* If we've reached MAX_REGNUM groups, then this open
3293 won't actually generate any code, so we'll have to
3294 clear pending_exact explicitly. */
3295 pending_exact = 0;
3296
3297 /* We're at the end of the group, so now we know how many
3298 groups were inside this one. */
3299 if (regnum <= MAX_REGNUM && regnum > 0)
3300 BUF_PUSH_2 (stop_memory, regnum);
3301 }
3302 break;
3303
3304
3305 case '|': /* `\|'. */
3306 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3307 goto normal_backslash;
3308 handle_alt:
3309 if (syntax & RE_LIMITED_OPS)
3310 goto normal_char;
3311
3312 /* Insert before the previous alternative a jump which
3313 jumps to this alternative if the former fails. */
3314 GET_BUFFER_SPACE (3);
3315 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3316 pending_exact = 0;
3317 b += 3;
3318
3319 /* The alternative before this one has a jump after it
3320 which gets executed if it gets matched. Adjust that
3321 jump so it will jump to this alternative's analogous
3322 jump (put in below, which in turn will jump to the next
3323 (if any) alternative's such jump, etc.). The last such
3324 jump jumps to the correct final destination. A picture:
3325 _____ _____
3326 | | | |
3327 | v | v
3328 a | b | c
3329
3330 If we are at `b', then fixup_alt_jump right now points to a
3331 three-byte space after `a'. We'll put in the jump, set
3332 fixup_alt_jump to right after `b', and leave behind three
3333 bytes which we'll fill in when we get to after `c'. */
3334
3335 FIXUP_ALT_JUMP ();
3336
3337 /* Mark and leave space for a jump after this alternative,
3338 to be filled in later either by next alternative or
3339 when know we're at the end of a series of alternatives. */
3340 fixup_alt_jump = b;
3341 GET_BUFFER_SPACE (3);
3342 b += 3;
3343
3344 laststart = 0;
3345 begalt = b;
3346 break;
3347
3348
3349 case '{':
3350 /* If \{ is a literal. */
3351 if (!(syntax & RE_INTERVALS)
3352 /* If we're at `\{' and it's not the open-interval
3353 operator. */
3354 || (syntax & RE_NO_BK_BRACES))
3355 goto normal_backslash;
3356
3357 handle_interval:
3358 {
3359 /* If got here, then the syntax allows intervals. */
3360
3361 /* At least (most) this many matches must be made. */
3362 int lower_bound = 0, upper_bound = -1;
3363
3364 beg_interval = p;
3365
3366 GET_UNSIGNED_NUMBER (lower_bound);
3367
3368 if (c == ',')
3369 GET_UNSIGNED_NUMBER (upper_bound);
3370 else
3371 /* Interval such as `{1}' => match exactly once. */
3372 upper_bound = lower_bound;
3373
3374 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3375 || (upper_bound >= 0 && lower_bound > upper_bound))
3376 FREE_STACK_RETURN (REG_BADBR);
3377
3378 if (!(syntax & RE_NO_BK_BRACES))
3379 {
3380 if (c != '\\')
3381 FREE_STACK_RETURN (REG_BADBR);
3382 if (p == pend)
3383 FREE_STACK_RETURN (REG_EESCAPE);
3384 PATFETCH (c);
3385 }
3386
3387 if (c != '}')
3388 FREE_STACK_RETURN (REG_BADBR);
3389
3390 /* We just parsed a valid interval. */
3391
3392 /* If it's invalid to have no preceding re. */
3393 if (!laststart)
3394 {
3395 if (syntax & RE_CONTEXT_INVALID_OPS)
3396 FREE_STACK_RETURN (REG_BADRPT);
3397 else if (syntax & RE_CONTEXT_INDEP_OPS)
3398 laststart = b;
3399 else
3400 goto unfetch_interval;
3401 }
3402
3403 if (upper_bound == 0)
3404 /* If the upper bound is zero, just drop the sub pattern
3405 altogether. */
3406 b = laststart;
3407 else if (lower_bound == 1 && upper_bound == 1)
3408 /* Just match it once: nothing to do here. */
3409 ;
3410
3411 /* Otherwise, we have a nontrivial interval. When
3412 we're all done, the pattern will look like:
3413 set_number_at <jump count> <upper bound>
3414 set_number_at <succeed_n count> <lower bound>
3415 succeed_n <after jump addr> <succeed_n count>
3416 <body of loop>
3417 jump_n <succeed_n addr> <jump count>
3418 (The upper bound and `jump_n' are omitted if
3419 `upper_bound' is 1, though.) */
3420 else
3421 { /* If the upper bound is > 1, we need to insert
3422 more at the end of the loop. */
3423 unsigned int nbytes = (upper_bound < 0 ? 3
3424 : upper_bound > 1 ? 5 : 0);
3425 unsigned int startoffset = 0;
3426
3427 GET_BUFFER_SPACE (20); /* We might use less. */
3428
3429 if (lower_bound == 0)
3430 {
3431 /* A succeed_n that starts with 0 is really a
3432 a simple on_failure_jump_loop. */
3433 INSERT_JUMP (on_failure_jump_loop, laststart,
3434 b + 3 + nbytes);
3435 b += 3;
3436 }
3437 else
3438 {
3439 /* Initialize lower bound of the `succeed_n', even
3440 though it will be set during matching by its
3441 attendant `set_number_at' (inserted next),
3442 because `re_compile_fastmap' needs to know.
3443 Jump to the `jump_n' we might insert below. */
3444 INSERT_JUMP2 (succeed_n, laststart,
3445 b + 5 + nbytes,
3446 lower_bound);
3447 b += 5;
3448
3449 /* Code to initialize the lower bound. Insert
3450 before the `succeed_n'. The `5' is the last two
3451 bytes of this `set_number_at', plus 3 bytes of
3452 the following `succeed_n'. */
3453 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3454 b += 5;
3455 startoffset += 5;
3456 }
3457
3458 if (upper_bound < 0)
3459 {
3460 /* A negative upper bound stands for infinity,
3461 in which case it degenerates to a plain jump. */
3462 STORE_JUMP (jump, b, laststart + startoffset);
3463 b += 3;
3464 }
3465 else if (upper_bound > 1)
3466 { /* More than one repetition is allowed, so
3467 append a backward jump to the `succeed_n'
3468 that starts this interval.
3469
3470 When we've reached this during matching,
3471 we'll have matched the interval once, so
3472 jump back only `upper_bound - 1' times. */
3473 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3474 upper_bound - 1);
3475 b += 5;
3476
3477 /* The location we want to set is the second
3478 parameter of the `jump_n'; that is `b-2' as
3479 an absolute address. `laststart' will be
3480 the `set_number_at' we're about to insert;
3481 `laststart+3' the number to set, the source
3482 for the relative address. But we are
3483 inserting into the middle of the pattern --
3484 so everything is getting moved up by 5.
3485 Conclusion: (b - 2) - (laststart + 3) + 5,
3486 i.e., b - laststart.
3487
3488 We insert this at the beginning of the loop
3489 so that if we fail during matching, we'll
3490 reinitialize the bounds. */
3491 insert_op2 (set_number_at, laststart, b - laststart,
3492 upper_bound - 1, b);
3493 b += 5;
3494 }
3495 }
3496 pending_exact = 0;
3497 beg_interval = NULL;
3498 }
3499 break;
3500
3501 unfetch_interval:
3502 /* If an invalid interval, match the characters as literals. */
3503 assert (beg_interval);
3504 p = beg_interval;
3505 beg_interval = NULL;
3506
3507 /* normal_char and normal_backslash need `c'. */
3508 c = '{';
3509
3510 if (!(syntax & RE_NO_BK_BRACES))
3511 {
3512 assert (p > pattern && p[-1] == '\\');
3513 goto normal_backslash;
3514 }
3515 else
3516 goto normal_char;
3517
3518 #ifdef emacs
3519 /* There is no way to specify the before_dot and after_dot
3520 operators. rms says this is ok. --karl */
3521 case '=':
3522 BUF_PUSH (at_dot);
3523 break;
3524
3525 case 's':
3526 laststart = b;
3527 PATFETCH (c);
3528 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3529 break;
3530
3531 case 'S':
3532 laststart = b;
3533 PATFETCH (c);
3534 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3535 break;
3536
3537 case 'c':
3538 laststart = b;
3539 PATFETCH (c);
3540 BUF_PUSH_2 (categoryspec, c);
3541 break;
3542
3543 case 'C':
3544 laststart = b;
3545 PATFETCH (c);
3546 BUF_PUSH_2 (notcategoryspec, c);
3547 break;
3548 #endif /* emacs */
3549
3550
3551 case 'w':
3552 if (syntax & RE_NO_GNU_OPS)
3553 goto normal_char;
3554 laststart = b;
3555 BUF_PUSH_2 (syntaxspec, Sword);
3556 break;
3557
3558
3559 case 'W':
3560 if (syntax & RE_NO_GNU_OPS)
3561 goto normal_char;
3562 laststart = b;
3563 BUF_PUSH_2 (notsyntaxspec, Sword);
3564 break;
3565
3566
3567 case '<':
3568 if (syntax & RE_NO_GNU_OPS)
3569 goto normal_char;
3570 BUF_PUSH (wordbeg);
3571 break;
3572
3573 case '>':
3574 if (syntax & RE_NO_GNU_OPS)
3575 goto normal_char;
3576 BUF_PUSH (wordend);
3577 break;
3578
3579 case '_':
3580 if (syntax & RE_NO_GNU_OPS)
3581 goto normal_char;
3582 laststart = b;
3583 PATFETCH (c);
3584 if (c == '<')
3585 BUF_PUSH (symbeg);
3586 else if (c == '>')
3587 BUF_PUSH (symend);
3588 else
3589 FREE_STACK_RETURN (REG_BADPAT);
3590 break;
3591
3592 case 'b':
3593 if (syntax & RE_NO_GNU_OPS)
3594 goto normal_char;
3595 BUF_PUSH (wordbound);
3596 break;
3597
3598 case 'B':
3599 if (syntax & RE_NO_GNU_OPS)
3600 goto normal_char;
3601 BUF_PUSH (notwordbound);
3602 break;
3603
3604 case '`':
3605 if (syntax & RE_NO_GNU_OPS)
3606 goto normal_char;
3607 BUF_PUSH (begbuf);
3608 break;
3609
3610 case '\'':
3611 if (syntax & RE_NO_GNU_OPS)
3612 goto normal_char;
3613 BUF_PUSH (endbuf);
3614 break;
3615
3616 case '1': case '2': case '3': case '4': case '5':
3617 case '6': case '7': case '8': case '9':
3618 {
3619 regnum_t reg;
3620
3621 if (syntax & RE_NO_BK_REFS)
3622 goto normal_backslash;
3623
3624 reg = c - '0';
3625
3626 if (reg > bufp->re_nsub || reg < 1
3627 /* Can't back reference to a subexp before its end. */
3628 || group_in_compile_stack (compile_stack, reg))
3629 FREE_STACK_RETURN (REG_ESUBREG);
3630
3631 laststart = b;
3632 BUF_PUSH_2 (duplicate, reg);
3633 }
3634 break;
3635
3636
3637 case '+':
3638 case '?':
3639 if (syntax & RE_BK_PLUS_QM)
3640 goto handle_plus;
3641 else
3642 goto normal_backslash;
3643
3644 default:
3645 normal_backslash:
3646 /* You might think it would be useful for \ to mean
3647 not to translate; but if we don't translate it
3648 it will never match anything. */
3649 goto normal_char;
3650 }
3651 break;
3652
3653
3654 default:
3655 /* Expects the character in `c'. */
3656 normal_char:
3657 /* If no exactn currently being built. */
3658 if (!pending_exact
3659
3660 /* If last exactn not at current position. */
3661 || pending_exact + *pending_exact + 1 != b
3662
3663 /* We have only one byte following the exactn for the count. */
3664 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3665
3666 /* If followed by a repetition operator. */
3667 || (p != pend && (*p == '*' || *p == '^'))
3668 || ((syntax & RE_BK_PLUS_QM)
3669 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3670 : p != pend && (*p == '+' || *p == '?'))
3671 || ((syntax & RE_INTERVALS)
3672 && ((syntax & RE_NO_BK_BRACES)
3673 ? p != pend && *p == '{'
3674 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3675 {
3676 /* Start building a new exactn. */
3677
3678 laststart = b;
3679
3680 BUF_PUSH_2 (exactn, 0);
3681 pending_exact = b - 1;
3682 }
3683
3684 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3685 {
3686 int len;
3687
3688 if (multibyte)
3689 {
3690 c = TRANSLATE (c);
3691 len = CHAR_STRING (c, b);
3692 b += len;
3693 }
3694 else
3695 {
3696 c1 = RE_CHAR_TO_MULTIBYTE (c);
3697 if (! CHAR_BYTE8_P (c1))
3698 {
3699 re_wchar_t c2 = TRANSLATE (c1);
3700
3701 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3702 c = c1;
3703 }
3704 *b++ = c;
3705 len = 1;
3706 }
3707 (*pending_exact) += len;
3708 }
3709
3710 break;
3711 } /* switch (c) */
3712 } /* while p != pend */
3713
3714
3715 /* Through the pattern now. */
3716
3717 FIXUP_ALT_JUMP ();
3718
3719 if (!COMPILE_STACK_EMPTY)
3720 FREE_STACK_RETURN (REG_EPAREN);
3721
3722 /* If we don't want backtracking, force success
3723 the first time we reach the end of the compiled pattern. */
3724 if (syntax & RE_NO_POSIX_BACKTRACKING)
3725 BUF_PUSH (succeed);
3726
3727 /* We have succeeded; set the length of the buffer. */
3728 bufp->used = b - bufp->buffer;
3729
3730 #ifdef DEBUG
3731 if (debug > 0)
3732 {
3733 re_compile_fastmap (bufp);
3734 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3735 print_compiled_pattern (bufp);
3736 }
3737 debug--;
3738 #endif /* DEBUG */
3739
3740 #ifndef MATCH_MAY_ALLOCATE
3741 /* Initialize the failure stack to the largest possible stack. This
3742 isn't necessary unless we're trying to avoid calling alloca in
3743 the search and match routines. */
3744 {
3745 int num_regs = bufp->re_nsub + 1;
3746
3747 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3748 {
3749 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3750
3751 if (! fail_stack.stack)
3752 fail_stack.stack
3753 = (fail_stack_elt_t *) malloc (fail_stack.size
3754 * sizeof (fail_stack_elt_t));
3755 else
3756 fail_stack.stack
3757 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3758 (fail_stack.size
3759 * sizeof (fail_stack_elt_t)));
3760 }
3761
3762 regex_grow_registers (num_regs);
3763 }
3764 #endif /* not MATCH_MAY_ALLOCATE */
3765
3766 FREE_STACK_RETURN (REG_NOERROR);
3767 } /* regex_compile */
3768 \f
3769 /* Subroutines for `regex_compile'. */
3770
3771 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3772
3773 static void
3774 store_op1 (re_opcode_t op, unsigned char *loc, int arg)
3775 {
3776 *loc = (unsigned char) op;
3777 STORE_NUMBER (loc + 1, arg);
3778 }
3779
3780
3781 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3782
3783 static void
3784 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2)
3785 {
3786 *loc = (unsigned char) op;
3787 STORE_NUMBER (loc + 1, arg1);
3788 STORE_NUMBER (loc + 3, arg2);
3789 }
3790
3791
3792 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3793 for OP followed by two-byte integer parameter ARG. */
3794
3795 static void
3796 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end)
3797 {
3798 register unsigned char *pfrom = end;
3799 register unsigned char *pto = end + 3;
3800
3801 while (pfrom != loc)
3802 *--pto = *--pfrom;
3803
3804 store_op1 (op, loc, arg);
3805 }
3806
3807
3808 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3809
3810 static void
3811 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end)
3812 {
3813 register unsigned char *pfrom = end;
3814 register unsigned char *pto = end + 5;
3815
3816 while (pfrom != loc)
3817 *--pto = *--pfrom;
3818
3819 store_op2 (op, loc, arg1, arg2);
3820 }
3821
3822
3823 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3824 after an alternative or a begin-subexpression. We assume there is at
3825 least one character before the ^. */
3826
3827 static boolean
3828 at_begline_loc_p (const re_char *pattern, const re_char *p, reg_syntax_t syntax)
3829 {
3830 re_char *prev = p - 2;
3831 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3832
3833 return
3834 /* After a subexpression? */
3835 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3836 /* After an alternative? */
3837 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3838 /* After a shy subexpression? */
3839 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3840 && prev[-1] == '?' && prev[-2] == '('
3841 && (syntax & RE_NO_BK_PARENS
3842 || (prev - 3 >= pattern && prev[-3] == '\\')));
3843 }
3844
3845
3846 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3847 at least one character after the $, i.e., `P < PEND'. */
3848
3849 static boolean
3850 at_endline_loc_p (const re_char *p, const re_char *pend, reg_syntax_t syntax)
3851 {
3852 re_char *next = p;
3853 boolean next_backslash = *next == '\\';
3854 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3855
3856 return
3857 /* Before a subexpression? */
3858 (syntax & RE_NO_BK_PARENS ? *next == ')'
3859 : next_backslash && next_next && *next_next == ')')
3860 /* Before an alternative? */
3861 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3862 : next_backslash && next_next && *next_next == '|');
3863 }
3864
3865
3866 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3867 false if it's not. */
3868
3869 static boolean
3870 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
3871 {
3872 ssize_t this_element;
3873
3874 for (this_element = compile_stack.avail - 1;
3875 this_element >= 0;
3876 this_element--)
3877 if (compile_stack.stack[this_element].regnum == regnum)
3878 return true;
3879
3880 return false;
3881 }
3882 \f
3883 /* analyse_first.
3884 If fastmap is non-NULL, go through the pattern and fill fastmap
3885 with all the possible leading chars. If fastmap is NULL, don't
3886 bother filling it up (obviously) and only return whether the
3887 pattern could potentially match the empty string.
3888
3889 Return 1 if p..pend might match the empty string.
3890 Return 0 if p..pend matches at least one char.
3891 Return -1 if fastmap was not updated accurately. */
3892
3893 static int
3894 analyse_first (const re_char *p, const re_char *pend, char *fastmap, const int multibyte)
3895 {
3896 int j, k;
3897 boolean not;
3898
3899 /* If all elements for base leading-codes in fastmap is set, this
3900 flag is set true. */
3901 boolean match_any_multibyte_characters = false;
3902
3903 assert (p);
3904
3905 /* The loop below works as follows:
3906 - It has a working-list kept in the PATTERN_STACK and which basically
3907 starts by only containing a pointer to the first operation.
3908 - If the opcode we're looking at is a match against some set of
3909 chars, then we add those chars to the fastmap and go on to the
3910 next work element from the worklist (done via `break').
3911 - If the opcode is a control operator on the other hand, we either
3912 ignore it (if it's meaningless at this point, such as `start_memory')
3913 or execute it (if it's a jump). If the jump has several destinations
3914 (i.e. `on_failure_jump'), then we push the other destination onto the
3915 worklist.
3916 We guarantee termination by ignoring backward jumps (more or less),
3917 so that `p' is monotonically increasing. More to the point, we
3918 never set `p' (or push) anything `<= p1'. */
3919
3920 while (p < pend)
3921 {
3922 /* `p1' is used as a marker of how far back a `on_failure_jump'
3923 can go without being ignored. It is normally equal to `p'
3924 (which prevents any backward `on_failure_jump') except right
3925 after a plain `jump', to allow patterns such as:
3926 0: jump 10
3927 3..9: <body>
3928 10: on_failure_jump 3
3929 as used for the *? operator. */
3930 re_char *p1 = p;
3931
3932 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3933 {
3934 case succeed:
3935 return 1;
3936
3937 case duplicate:
3938 /* If the first character has to match a backreference, that means
3939 that the group was empty (since it already matched). Since this
3940 is the only case that interests us here, we can assume that the
3941 backreference must match the empty string. */
3942 p++;
3943 continue;
3944
3945
3946 /* Following are the cases which match a character. These end
3947 with `break'. */
3948
3949 case exactn:
3950 if (fastmap)
3951 {
3952 /* If multibyte is nonzero, the first byte of each
3953 character is an ASCII or a leading code. Otherwise,
3954 each byte is a character. Thus, this works in both
3955 cases. */
3956 fastmap[p[1]] = 1;
3957 if (! multibyte)
3958 {
3959 /* For the case of matching this unibyte regex
3960 against multibyte, we must set a leading code of
3961 the corresponding multibyte character. */
3962 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
3963
3964 fastmap[CHAR_LEADING_CODE (c)] = 1;
3965 }
3966 }
3967 break;
3968
3969
3970 case anychar:
3971 /* We could put all the chars except for \n (and maybe \0)
3972 but we don't bother since it is generally not worth it. */
3973 if (!fastmap) break;
3974 return -1;
3975
3976
3977 case charset_not:
3978 if (!fastmap) break;
3979 {
3980 /* Chars beyond end of bitmap are possible matches. */
3981 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3982 j < (1 << BYTEWIDTH); j++)
3983 fastmap[j] = 1;
3984 }
3985
3986 /* Fallthrough */
3987 case charset:
3988 if (!fastmap) break;
3989 not = (re_opcode_t) *(p - 1) == charset_not;
3990 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3991 j >= 0; j--)
3992 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3993 fastmap[j] = 1;
3994
3995 #ifdef emacs
3996 if (/* Any leading code can possibly start a character
3997 which doesn't match the specified set of characters. */
3998 not
3999 ||
4000 /* If we can match a character class, we can match any
4001 multibyte characters. */
4002 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4003 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4004
4005 {
4006 if (match_any_multibyte_characters == false)
4007 {
4008 for (j = MIN_MULTIBYTE_LEADING_CODE;
4009 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4010 fastmap[j] = 1;
4011 match_any_multibyte_characters = true;
4012 }
4013 }
4014
4015 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4016 && match_any_multibyte_characters == false)
4017 {
4018 /* Set fastmap[I] to 1 where I is a leading code of each
4019 multibyte character in the range table. */
4020 int c, count;
4021 unsigned char lc1, lc2;
4022
4023 /* Make P points the range table. `+ 2' is to skip flag
4024 bits for a character class. */
4025 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4026
4027 /* Extract the number of ranges in range table into COUNT. */
4028 EXTRACT_NUMBER_AND_INCR (count, p);
4029 for (; count > 0; count--, p += 3)
4030 {
4031 /* Extract the start and end of each range. */
4032 EXTRACT_CHARACTER (c, p);
4033 lc1 = CHAR_LEADING_CODE (c);
4034 p += 3;
4035 EXTRACT_CHARACTER (c, p);
4036 lc2 = CHAR_LEADING_CODE (c);
4037 for (j = lc1; j <= lc2; j++)
4038 fastmap[j] = 1;
4039 }
4040 }
4041 #endif
4042 break;
4043
4044 case syntaxspec:
4045 case notsyntaxspec:
4046 if (!fastmap) break;
4047 #ifndef emacs
4048 not = (re_opcode_t)p[-1] == notsyntaxspec;
4049 k = *p++;
4050 for (j = 0; j < (1 << BYTEWIDTH); j++)
4051 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4052 fastmap[j] = 1;
4053 break;
4054 #else /* emacs */
4055 /* This match depends on text properties. These end with
4056 aborting optimizations. */
4057 return -1;
4058
4059 case categoryspec:
4060 case notcategoryspec:
4061 if (!fastmap) break;
4062 not = (re_opcode_t)p[-1] == notcategoryspec;
4063 k = *p++;
4064 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4065 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4066 fastmap[j] = 1;
4067
4068 /* Any leading code can possibly start a character which
4069 has or doesn't has the specified category. */
4070 if (match_any_multibyte_characters == false)
4071 {
4072 for (j = MIN_MULTIBYTE_LEADING_CODE;
4073 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4074 fastmap[j] = 1;
4075 match_any_multibyte_characters = true;
4076 }
4077 break;
4078
4079 /* All cases after this match the empty string. These end with
4080 `continue'. */
4081
4082 case before_dot:
4083 case at_dot:
4084 case after_dot:
4085 #endif /* !emacs */
4086 case no_op:
4087 case begline:
4088 case endline:
4089 case begbuf:
4090 case endbuf:
4091 case wordbound:
4092 case notwordbound:
4093 case wordbeg:
4094 case wordend:
4095 case symbeg:
4096 case symend:
4097 continue;
4098
4099
4100 case jump:
4101 EXTRACT_NUMBER_AND_INCR (j, p);
4102 if (j < 0)
4103 /* Backward jumps can only go back to code that we've already
4104 visited. `re_compile' should make sure this is true. */
4105 break;
4106 p += j;
4107 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4108 {
4109 case on_failure_jump:
4110 case on_failure_keep_string_jump:
4111 case on_failure_jump_loop:
4112 case on_failure_jump_nastyloop:
4113 case on_failure_jump_smart:
4114 p++;
4115 break;
4116 default:
4117 continue;
4118 };
4119 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4120 to jump back to "just after here". */
4121 /* Fallthrough */
4122
4123 case on_failure_jump:
4124 case on_failure_keep_string_jump:
4125 case on_failure_jump_nastyloop:
4126 case on_failure_jump_loop:
4127 case on_failure_jump_smart:
4128 EXTRACT_NUMBER_AND_INCR (j, p);
4129 if (p + j <= p1)
4130 ; /* Backward jump to be ignored. */
4131 else
4132 { /* We have to look down both arms.
4133 We first go down the "straight" path so as to minimize
4134 stack usage when going through alternatives. */
4135 int r = analyse_first (p, pend, fastmap, multibyte);
4136 if (r) return r;
4137 p += j;
4138 }
4139 continue;
4140
4141
4142 case jump_n:
4143 /* This code simply does not properly handle forward jump_n. */
4144 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4145 p += 4;
4146 /* jump_n can either jump or fall through. The (backward) jump
4147 case has already been handled, so we only need to look at the
4148 fallthrough case. */
4149 continue;
4150
4151 case succeed_n:
4152 /* If N == 0, it should be an on_failure_jump_loop instead. */
4153 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4154 p += 4;
4155 /* We only care about one iteration of the loop, so we don't
4156 need to consider the case where this behaves like an
4157 on_failure_jump. */
4158 continue;
4159
4160
4161 case set_number_at:
4162 p += 4;
4163 continue;
4164
4165
4166 case start_memory:
4167 case stop_memory:
4168 p += 1;
4169 continue;
4170
4171
4172 default:
4173 abort (); /* We have listed all the cases. */
4174 } /* switch *p++ */
4175
4176 /* Getting here means we have found the possible starting
4177 characters for one path of the pattern -- and that the empty
4178 string does not match. We need not follow this path further. */
4179 return 0;
4180 } /* while p */
4181
4182 /* We reached the end without matching anything. */
4183 return 1;
4184
4185 } /* analyse_first */
4186 \f
4187 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4188 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4189 characters can start a string that matches the pattern. This fastmap
4190 is used by re_search to skip quickly over impossible starting points.
4191
4192 Character codes above (1 << BYTEWIDTH) are not represented in the
4193 fastmap, but the leading codes are represented. Thus, the fastmap
4194 indicates which character sets could start a match.
4195
4196 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4197 area as BUFP->fastmap.
4198
4199 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4200 the pattern buffer.
4201
4202 Returns 0 if we succeed, -2 if an internal error. */
4203
4204 int
4205 re_compile_fastmap (struct re_pattern_buffer *bufp)
4206 {
4207 char *fastmap = bufp->fastmap;
4208 int analysis;
4209
4210 assert (fastmap && bufp->buffer);
4211
4212 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4213 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4214
4215 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4216 fastmap, RE_MULTIBYTE_P (bufp));
4217 bufp->can_be_null = (analysis != 0);
4218 return 0;
4219 } /* re_compile_fastmap */
4220 \f
4221 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4222 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4223 this memory for recording register information. STARTS and ENDS
4224 must be allocated using the malloc library routine, and must each
4225 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4226
4227 If NUM_REGS == 0, then subsequent matches should allocate their own
4228 register data.
4229
4230 Unless this function is called, the first search or match using
4231 PATTERN_BUFFER will allocate its own register data, without
4232 freeing the old data. */
4233
4234 void
4235 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4236 {
4237 if (num_regs)
4238 {
4239 bufp->regs_allocated = REGS_REALLOCATE;
4240 regs->num_regs = num_regs;
4241 regs->start = starts;
4242 regs->end = ends;
4243 }
4244 else
4245 {
4246 bufp->regs_allocated = REGS_UNALLOCATED;
4247 regs->num_regs = 0;
4248 regs->start = regs->end = (regoff_t *) 0;
4249 }
4250 }
4251 WEAK_ALIAS (__re_set_registers, re_set_registers)
4252 \f
4253 /* Searching routines. */
4254
4255 /* Like re_search_2, below, but only one string is specified, and
4256 doesn't let you say where to stop matching. */
4257
4258 regoff_t
4259 re_search (struct re_pattern_buffer *bufp, const char *string, size_t size,
4260 ssize_t startpos, ssize_t range, struct re_registers *regs)
4261 {
4262 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4263 regs, size);
4264 }
4265 WEAK_ALIAS (__re_search, re_search)
4266
4267 /* Head address of virtual concatenation of string. */
4268 #define HEAD_ADDR_VSTRING(P) \
4269 (((P) >= size1 ? string2 : string1))
4270
4271 /* Address of POS in the concatenation of virtual string. */
4272 #define POS_ADDR_VSTRING(POS) \
4273 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4274
4275 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4276 virtual concatenation of STRING1 and STRING2, starting first at index
4277 STARTPOS, then at STARTPOS + 1, and so on.
4278
4279 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4280
4281 RANGE is how far to scan while trying to match. RANGE = 0 means try
4282 only at STARTPOS; in general, the last start tried is STARTPOS +
4283 RANGE.
4284
4285 In REGS, return the indices of the virtual concatenation of STRING1
4286 and STRING2 that matched the entire BUFP->buffer and its contained
4287 subexpressions.
4288
4289 Do not consider matching one past the index STOP in the virtual
4290 concatenation of STRING1 and STRING2.
4291
4292 We return either the position in the strings at which the match was
4293 found, -1 if no match, or -2 if error (such as failure
4294 stack overflow). */
4295
4296 regoff_t
4297 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, size_t size1,
4298 const char *str2, size_t size2, ssize_t startpos, ssize_t range,
4299 struct re_registers *regs, ssize_t stop)
4300 {
4301 regoff_t val;
4302 re_char *string1 = (re_char*) str1;
4303 re_char *string2 = (re_char*) str2;
4304 register char *fastmap = bufp->fastmap;
4305 register RE_TRANSLATE_TYPE translate = bufp->translate;
4306 size_t total_size = size1 + size2;
4307 ssize_t endpos = startpos + range;
4308 boolean anchored_start;
4309 /* Nonzero if we are searching multibyte string. */
4310 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4311
4312 /* Check for out-of-range STARTPOS. */
4313 if (startpos < 0 || startpos > total_size)
4314 return -1;
4315
4316 /* Fix up RANGE if it might eventually take us outside
4317 the virtual concatenation of STRING1 and STRING2.
4318 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4319 if (endpos < 0)
4320 range = 0 - startpos;
4321 else if (endpos > total_size)
4322 range = total_size - startpos;
4323
4324 /* If the search isn't to be a backwards one, don't waste time in a
4325 search for a pattern anchored at beginning of buffer. */
4326 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4327 {
4328 if (startpos > 0)
4329 return -1;
4330 else
4331 range = 0;
4332 }
4333
4334 #ifdef emacs
4335 /* In a forward search for something that starts with \=.
4336 don't keep searching past point. */
4337 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4338 {
4339 range = PT_BYTE - BEGV_BYTE - startpos;
4340 if (range < 0)
4341 return -1;
4342 }
4343 #endif /* emacs */
4344
4345 /* Update the fastmap now if not correct already. */
4346 if (fastmap && !bufp->fastmap_accurate)
4347 re_compile_fastmap (bufp);
4348
4349 /* See whether the pattern is anchored. */
4350 anchored_start = (bufp->buffer[0] == begline);
4351
4352 #ifdef emacs
4353 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4354 {
4355 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4356
4357 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4358 }
4359 #endif
4360
4361 /* Loop through the string, looking for a place to start matching. */
4362 for (;;)
4363 {
4364 /* If the pattern is anchored,
4365 skip quickly past places we cannot match.
4366 We don't bother to treat startpos == 0 specially
4367 because that case doesn't repeat. */
4368 if (anchored_start && startpos > 0)
4369 {
4370 if (! ((startpos <= size1 ? string1[startpos - 1]
4371 : string2[startpos - size1 - 1])
4372 == '\n'))
4373 goto advance;
4374 }
4375
4376 /* If a fastmap is supplied, skip quickly over characters that
4377 cannot be the start of a match. If the pattern can match the
4378 null string, however, we don't need to skip characters; we want
4379 the first null string. */
4380 if (fastmap && startpos < total_size && !bufp->can_be_null)
4381 {
4382 register re_char *d;
4383 register re_wchar_t buf_ch;
4384
4385 d = POS_ADDR_VSTRING (startpos);
4386
4387 if (range > 0) /* Searching forwards. */
4388 {
4389 register int lim = 0;
4390 ssize_t irange = range;
4391
4392 if (startpos < size1 && startpos + range >= size1)
4393 lim = range - (size1 - startpos);
4394
4395 /* Written out as an if-else to avoid testing `translate'
4396 inside the loop. */
4397 if (RE_TRANSLATE_P (translate))
4398 {
4399 if (multibyte)
4400 while (range > lim)
4401 {
4402 int buf_charlen;
4403
4404 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4405 buf_ch = RE_TRANSLATE (translate, buf_ch);
4406 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4407 break;
4408
4409 range -= buf_charlen;
4410 d += buf_charlen;
4411 }
4412 else
4413 while (range > lim)
4414 {
4415 register re_wchar_t ch, translated;
4416
4417 buf_ch = *d;
4418 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4419 translated = RE_TRANSLATE (translate, ch);
4420 if (translated != ch
4421 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4422 buf_ch = ch;
4423 if (fastmap[buf_ch])
4424 break;
4425 d++;
4426 range--;
4427 }
4428 }
4429 else
4430 {
4431 if (multibyte)
4432 while (range > lim)
4433 {
4434 int buf_charlen;
4435
4436 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4437 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4438 break;
4439 range -= buf_charlen;
4440 d += buf_charlen;
4441 }
4442 else
4443 while (range > lim && !fastmap[*d])
4444 {
4445 d++;
4446 range--;
4447 }
4448 }
4449 startpos += irange - range;
4450 }
4451 else /* Searching backwards. */
4452 {
4453 if (multibyte)
4454 {
4455 buf_ch = STRING_CHAR (d);
4456 buf_ch = TRANSLATE (buf_ch);
4457 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4458 goto advance;
4459 }
4460 else
4461 {
4462 register re_wchar_t ch, translated;
4463
4464 buf_ch = *d;
4465 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4466 translated = TRANSLATE (ch);
4467 if (translated != ch
4468 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4469 buf_ch = ch;
4470 if (! fastmap[TRANSLATE (buf_ch)])
4471 goto advance;
4472 }
4473 }
4474 }
4475
4476 /* If can't match the null string, and that's all we have left, fail. */
4477 if (range >= 0 && startpos == total_size && fastmap
4478 && !bufp->can_be_null)
4479 return -1;
4480
4481 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4482 startpos, regs, stop);
4483
4484 if (val >= 0)
4485 return startpos;
4486
4487 if (val == -2)
4488 return -2;
4489
4490 advance:
4491 if (!range)
4492 break;
4493 else if (range > 0)
4494 {
4495 /* Update STARTPOS to the next character boundary. */
4496 if (multibyte)
4497 {
4498 re_char *p = POS_ADDR_VSTRING (startpos);
4499 int len = BYTES_BY_CHAR_HEAD (*p);
4500
4501 range -= len;
4502 if (range < 0)
4503 break;
4504 startpos += len;
4505 }
4506 else
4507 {
4508 range--;
4509 startpos++;
4510 }
4511 }
4512 else
4513 {
4514 range++;
4515 startpos--;
4516
4517 /* Update STARTPOS to the previous character boundary. */
4518 if (multibyte)
4519 {
4520 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4521 re_char *p0 = p;
4522 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4523
4524 /* Find the head of multibyte form. */
4525 PREV_CHAR_BOUNDARY (p, phead);
4526 range += p0 - 1 - p;
4527 if (range > 0)
4528 break;
4529
4530 startpos -= p0 - 1 - p;
4531 }
4532 }
4533 }
4534 return -1;
4535 } /* re_search_2 */
4536 WEAK_ALIAS (__re_search_2, re_search_2)
4537 \f
4538 /* Declarations and macros for re_match_2. */
4539
4540 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4541 register ssize_t len,
4542 RE_TRANSLATE_TYPE translate,
4543 const int multibyte));
4544
4545 /* This converts PTR, a pointer into one of the search strings `string1'
4546 and `string2' into an offset from the beginning of that string. */
4547 #define POINTER_TO_OFFSET(ptr) \
4548 (FIRST_STRING_P (ptr) \
4549 ? ((regoff_t) ((ptr) - string1)) \
4550 : ((regoff_t) ((ptr) - string2 + size1)))
4551
4552 /* Call before fetching a character with *d. This switches over to
4553 string2 if necessary.
4554 Check re_match_2_internal for a discussion of why end_match_2 might
4555 not be within string2 (but be equal to end_match_1 instead). */
4556 #define PREFETCH() \
4557 while (d == dend) \
4558 { \
4559 /* End of string2 => fail. */ \
4560 if (dend == end_match_2) \
4561 goto fail; \
4562 /* End of string1 => advance to string2. */ \
4563 d = string2; \
4564 dend = end_match_2; \
4565 }
4566
4567 /* Call before fetching a char with *d if you already checked other limits.
4568 This is meant for use in lookahead operations like wordend, etc..
4569 where we might need to look at parts of the string that might be
4570 outside of the LIMITs (i.e past `stop'). */
4571 #define PREFETCH_NOLIMIT() \
4572 if (d == end1) \
4573 { \
4574 d = string2; \
4575 dend = end_match_2; \
4576 } \
4577
4578 /* Test if at very beginning or at very end of the virtual concatenation
4579 of `string1' and `string2'. If only one string, it's `string2'. */
4580 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4581 #define AT_STRINGS_END(d) ((d) == end2)
4582
4583 /* Disabled due to a compiler bug -- see comment at case wordbound */
4584
4585 /* The comment at case wordbound is following one, but we don't use
4586 AT_WORD_BOUNDARY anymore to support multibyte form.
4587
4588 The DEC Alpha C compiler 3.x generates incorrect code for the
4589 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4590 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4591 macro and introducing temporary variables works around the bug. */
4592
4593 #if 0
4594 /* Test if D points to a character which is word-constituent. We have
4595 two special cases to check for: if past the end of string1, look at
4596 the first character in string2; and if before the beginning of
4597 string2, look at the last character in string1. */
4598 #define WORDCHAR_P(d) \
4599 (SYNTAX ((d) == end1 ? *string2 \
4600 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4601 == Sword)
4602
4603 /* Test if the character before D and the one at D differ with respect
4604 to being word-constituent. */
4605 #define AT_WORD_BOUNDARY(d) \
4606 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4607 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4608 #endif
4609
4610 /* Free everything we malloc. */
4611 #ifdef MATCH_MAY_ALLOCATE
4612 # define FREE_VAR(var) \
4613 do { \
4614 if (var) \
4615 { \
4616 REGEX_FREE (var); \
4617 var = NULL; \
4618 } \
4619 } while (0)
4620 # define FREE_VARIABLES() \
4621 do { \
4622 REGEX_FREE_STACK (fail_stack.stack); \
4623 FREE_VAR (regstart); \
4624 FREE_VAR (regend); \
4625 FREE_VAR (best_regstart); \
4626 FREE_VAR (best_regend); \
4627 } while (0)
4628 #else
4629 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4630 #endif /* not MATCH_MAY_ALLOCATE */
4631
4632 \f
4633 /* Optimization routines. */
4634
4635 /* If the operation is a match against one or more chars,
4636 return a pointer to the next operation, else return NULL. */
4637 static re_char *
4638 skip_one_char (const re_char *p)
4639 {
4640 switch (SWITCH_ENUM_CAST (*p++))
4641 {
4642 case anychar:
4643 break;
4644
4645 case exactn:
4646 p += *p + 1;
4647 break;
4648
4649 case charset_not:
4650 case charset:
4651 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4652 {
4653 int mcnt;
4654 p = CHARSET_RANGE_TABLE (p - 1);
4655 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4656 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4657 }
4658 else
4659 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4660 break;
4661
4662 case syntaxspec:
4663 case notsyntaxspec:
4664 #ifdef emacs
4665 case categoryspec:
4666 case notcategoryspec:
4667 #endif /* emacs */
4668 p++;
4669 break;
4670
4671 default:
4672 p = NULL;
4673 }
4674 return p;
4675 }
4676
4677
4678 /* Jump over non-matching operations. */
4679 static re_char *
4680 skip_noops (const re_char *p, const re_char *pend)
4681 {
4682 int mcnt;
4683 while (p < pend)
4684 {
4685 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4686 {
4687 case start_memory:
4688 case stop_memory:
4689 p += 2; break;
4690 case no_op:
4691 p += 1; break;
4692 case jump:
4693 p += 1;
4694 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4695 p += mcnt;
4696 break;
4697 default:
4698 return p;
4699 }
4700 }
4701 assert (p == pend);
4702 return p;
4703 }
4704
4705 /* Non-zero if "p1 matches something" implies "p2 fails". */
4706 static int
4707 mutually_exclusive_p (struct re_pattern_buffer *bufp, const re_char *p1, const re_char *p2)
4708 {
4709 re_opcode_t op2;
4710 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4711 unsigned char *pend = bufp->buffer + bufp->used;
4712
4713 assert (p1 >= bufp->buffer && p1 < pend
4714 && p2 >= bufp->buffer && p2 <= pend);
4715
4716 /* Skip over open/close-group commands.
4717 If what follows this loop is a ...+ construct,
4718 look at what begins its body, since we will have to
4719 match at least one of that. */
4720 p2 = skip_noops (p2, pend);
4721 /* The same skip can be done for p1, except that this function
4722 is only used in the case where p1 is a simple match operator. */
4723 /* p1 = skip_noops (p1, pend); */
4724
4725 assert (p1 >= bufp->buffer && p1 < pend
4726 && p2 >= bufp->buffer && p2 <= pend);
4727
4728 op2 = p2 == pend ? succeed : *p2;
4729
4730 switch (SWITCH_ENUM_CAST (op2))
4731 {
4732 case succeed:
4733 case endbuf:
4734 /* If we're at the end of the pattern, we can change. */
4735 if (skip_one_char (p1))
4736 {
4737 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4738 return 1;
4739 }
4740 break;
4741
4742 case endline:
4743 case exactn:
4744 {
4745 register re_wchar_t c
4746 = (re_opcode_t) *p2 == endline ? '\n'
4747 : RE_STRING_CHAR (p2 + 2, multibyte);
4748
4749 if ((re_opcode_t) *p1 == exactn)
4750 {
4751 if (c != RE_STRING_CHAR (p1 + 2, multibyte))
4752 {
4753 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4754 return 1;
4755 }
4756 }
4757
4758 else if ((re_opcode_t) *p1 == charset
4759 || (re_opcode_t) *p1 == charset_not)
4760 {
4761 int not = (re_opcode_t) *p1 == charset_not;
4762
4763 /* Test if C is listed in charset (or charset_not)
4764 at `p1'. */
4765 if (! multibyte || IS_REAL_ASCII (c))
4766 {
4767 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4768 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4769 not = !not;
4770 }
4771 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4772 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4773
4774 /* `not' is equal to 1 if c would match, which means
4775 that we can't change to pop_failure_jump. */
4776 if (!not)
4777 {
4778 DEBUG_PRINT1 (" No match => fast loop.\n");
4779 return 1;
4780 }
4781 }
4782 else if ((re_opcode_t) *p1 == anychar
4783 && c == '\n')
4784 {
4785 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4786 return 1;
4787 }
4788 }
4789 break;
4790
4791 case charset:
4792 {
4793 if ((re_opcode_t) *p1 == exactn)
4794 /* Reuse the code above. */
4795 return mutually_exclusive_p (bufp, p2, p1);
4796
4797 /* It is hard to list up all the character in charset
4798 P2 if it includes multibyte character. Give up in
4799 such case. */
4800 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4801 {
4802 /* Now, we are sure that P2 has no range table.
4803 So, for the size of bitmap in P2, `p2[1]' is
4804 enough. But P1 may have range table, so the
4805 size of bitmap table of P1 is extracted by
4806 using macro `CHARSET_BITMAP_SIZE'.
4807
4808 In a multibyte case, we know that all the character
4809 listed in P2 is ASCII. In a unibyte case, P1 has only a
4810 bitmap table. So, in both cases, it is enough to test
4811 only the bitmap table of P1. */
4812
4813 if ((re_opcode_t) *p1 == charset)
4814 {
4815 int idx;
4816 /* We win if the charset inside the loop
4817 has no overlap with the one after the loop. */
4818 for (idx = 0;
4819 (idx < (int) p2[1]
4820 && idx < CHARSET_BITMAP_SIZE (p1));
4821 idx++)
4822 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4823 break;
4824
4825 if (idx == p2[1]
4826 || idx == CHARSET_BITMAP_SIZE (p1))
4827 {
4828 DEBUG_PRINT1 (" No match => fast loop.\n");
4829 return 1;
4830 }
4831 }
4832 else if ((re_opcode_t) *p1 == charset_not)
4833 {
4834 int idx;
4835 /* We win if the charset_not inside the loop lists
4836 every character listed in the charset after. */
4837 for (idx = 0; idx < (int) p2[1]; idx++)
4838 if (! (p2[2 + idx] == 0
4839 || (idx < CHARSET_BITMAP_SIZE (p1)
4840 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4841 break;
4842
4843 if (idx == p2[1])
4844 {
4845 DEBUG_PRINT1 (" No match => fast loop.\n");
4846 return 1;
4847 }
4848 }
4849 }
4850 }
4851 break;
4852
4853 case charset_not:
4854 switch (SWITCH_ENUM_CAST (*p1))
4855 {
4856 case exactn:
4857 case charset:
4858 /* Reuse the code above. */
4859 return mutually_exclusive_p (bufp, p2, p1);
4860 case charset_not:
4861 /* When we have two charset_not, it's very unlikely that
4862 they don't overlap. The union of the two sets of excluded
4863 chars should cover all possible chars, which, as a matter of
4864 fact, is virtually impossible in multibyte buffers. */
4865 break;
4866 }
4867 break;
4868
4869 case wordend:
4870 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
4871 case symend:
4872 return ((re_opcode_t) *p1 == syntaxspec
4873 && (p1[1] == Ssymbol || p1[1] == Sword));
4874 case notsyntaxspec:
4875 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
4876
4877 case wordbeg:
4878 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
4879 case symbeg:
4880 return ((re_opcode_t) *p1 == notsyntaxspec
4881 && (p1[1] == Ssymbol || p1[1] == Sword));
4882 case syntaxspec:
4883 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
4884
4885 case wordbound:
4886 return (((re_opcode_t) *p1 == notsyntaxspec
4887 || (re_opcode_t) *p1 == syntaxspec)
4888 && p1[1] == Sword);
4889
4890 #ifdef emacs
4891 case categoryspec:
4892 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4893 case notcategoryspec:
4894 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4895 #endif /* emacs */
4896
4897 default:
4898 ;
4899 }
4900
4901 /* Safe default. */
4902 return 0;
4903 }
4904
4905 \f
4906 /* Matching routines. */
4907
4908 #ifndef emacs /* Emacs never uses this. */
4909 /* re_match is like re_match_2 except it takes only a single string. */
4910
4911 regoff_t
4912 re_match (struct re_pattern_buffer *bufp, const char *string,
4913 size_t size, ssize_t pos, struct re_registers *regs)
4914 {
4915 regoff_t result = re_match_2_internal (bufp, NULL, 0, (re_char*) string,
4916 size, pos, regs, size);
4917 return result;
4918 }
4919 WEAK_ALIAS (__re_match, re_match)
4920 #endif /* not emacs */
4921
4922 #ifdef emacs
4923 /* In Emacs, this is the string or buffer in which we
4924 are matching. It is used for looking up syntax properties. */
4925 Lisp_Object re_match_object;
4926 #endif
4927
4928 /* re_match_2 matches the compiled pattern in BUFP against the
4929 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4930 and SIZE2, respectively). We start matching at POS, and stop
4931 matching at STOP.
4932
4933 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4934 store offsets for the substring each group matched in REGS. See the
4935 documentation for exactly how many groups we fill.
4936
4937 We return -1 if no match, -2 if an internal error (such as the
4938 failure stack overflowing). Otherwise, we return the length of the
4939 matched substring. */
4940
4941 regoff_t
4942 re_match_2 (struct re_pattern_buffer *bufp, const char *string1,
4943 size_t size1, const char *string2, size_t size2, ssize_t pos,
4944 struct re_registers *regs, ssize_t stop)
4945 {
4946 regoff_t result;
4947
4948 #ifdef emacs
4949 ssize_t charpos;
4950 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4951 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4952 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4953 #endif
4954
4955 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4956 (re_char*) string2, size2,
4957 pos, regs, stop);
4958 return result;
4959 }
4960 WEAK_ALIAS (__re_match_2, re_match_2)
4961
4962
4963 /* This is a separate function so that we can force an alloca cleanup
4964 afterwards. */
4965 static regoff_t
4966 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1,
4967 size_t size1, const re_char *string2, size_t size2,
4968 ssize_t pos, struct re_registers *regs, ssize_t stop)
4969 {
4970 /* General temporaries. */
4971 ssize_t mcnt;
4972 size_t reg;
4973
4974 /* Just past the end of the corresponding string. */
4975 re_char *end1, *end2;
4976
4977 /* Pointers into string1 and string2, just past the last characters in
4978 each to consider matching. */
4979 re_char *end_match_1, *end_match_2;
4980
4981 /* Where we are in the data, and the end of the current string. */
4982 re_char *d, *dend;
4983
4984 /* Used sometimes to remember where we were before starting matching
4985 an operator so that we can go back in case of failure. This "atomic"
4986 behavior of matching opcodes is indispensable to the correctness
4987 of the on_failure_keep_string_jump optimization. */
4988 re_char *dfail;
4989
4990 /* Where we are in the pattern, and the end of the pattern. */
4991 re_char *p = bufp->buffer;
4992 re_char *pend = p + bufp->used;
4993
4994 /* We use this to map every character in the string. */
4995 RE_TRANSLATE_TYPE translate = bufp->translate;
4996
4997 /* Nonzero if BUFP is setup from a multibyte regex. */
4998 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4999
5000 /* Nonzero if STRING1/STRING2 are multibyte. */
5001 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5002
5003 /* Failure point stack. Each place that can handle a failure further
5004 down the line pushes a failure point on this stack. It consists of
5005 regstart, and regend for all registers corresponding to
5006 the subexpressions we're currently inside, plus the number of such
5007 registers, and, finally, two char *'s. The first char * is where
5008 to resume scanning the pattern; the second one is where to resume
5009 scanning the strings. */
5010 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5011 fail_stack_type fail_stack;
5012 #endif
5013 #ifdef DEBUG
5014 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5015 #endif
5016
5017 #if defined REL_ALLOC && defined REGEX_MALLOC
5018 /* This holds the pointer to the failure stack, when
5019 it is allocated relocatably. */
5020 fail_stack_elt_t *failure_stack_ptr;
5021 #endif
5022
5023 /* We fill all the registers internally, independent of what we
5024 return, for use in backreferences. The number here includes
5025 an element for register zero. */
5026 size_t num_regs = bufp->re_nsub + 1;
5027
5028 /* Information on the contents of registers. These are pointers into
5029 the input strings; they record just what was matched (on this
5030 attempt) by a subexpression part of the pattern, that is, the
5031 regnum-th regstart pointer points to where in the pattern we began
5032 matching and the regnum-th regend points to right after where we
5033 stopped matching the regnum-th subexpression. (The zeroth register
5034 keeps track of what the whole pattern matches.) */
5035 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5036 re_char **regstart, **regend;
5037 #endif
5038
5039 /* The following record the register info as found in the above
5040 variables when we find a match better than any we've seen before.
5041 This happens as we backtrack through the failure points, which in
5042 turn happens only if we have not yet matched the entire string. */
5043 unsigned best_regs_set = false;
5044 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5045 re_char **best_regstart, **best_regend;
5046 #endif
5047
5048 /* Logically, this is `best_regend[0]'. But we don't want to have to
5049 allocate space for that if we're not allocating space for anything
5050 else (see below). Also, we never need info about register 0 for
5051 any of the other register vectors, and it seems rather a kludge to
5052 treat `best_regend' differently than the rest. So we keep track of
5053 the end of the best match so far in a separate variable. We
5054 initialize this to NULL so that when we backtrack the first time
5055 and need to test it, it's not garbage. */
5056 re_char *match_end = NULL;
5057
5058 #ifdef DEBUG
5059 /* Counts the total number of registers pushed. */
5060 unsigned num_regs_pushed = 0;
5061 #endif
5062
5063 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5064
5065 INIT_FAIL_STACK ();
5066
5067 #ifdef MATCH_MAY_ALLOCATE
5068 /* Do not bother to initialize all the register variables if there are
5069 no groups in the pattern, as it takes a fair amount of time. If
5070 there are groups, we include space for register 0 (the whole
5071 pattern), even though we never use it, since it simplifies the
5072 array indexing. We should fix this. */
5073 if (bufp->re_nsub)
5074 {
5075 regstart = REGEX_TALLOC (num_regs, re_char *);
5076 regend = REGEX_TALLOC (num_regs, re_char *);
5077 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5078 best_regend = REGEX_TALLOC (num_regs, re_char *);
5079
5080 if (!(regstart && regend && best_regstart && best_regend))
5081 {
5082 FREE_VARIABLES ();
5083 return -2;
5084 }
5085 }
5086 else
5087 {
5088 /* We must initialize all our variables to NULL, so that
5089 `FREE_VARIABLES' doesn't try to free them. */
5090 regstart = regend = best_regstart = best_regend = NULL;
5091 }
5092 #endif /* MATCH_MAY_ALLOCATE */
5093
5094 /* The starting position is bogus. */
5095 if (pos < 0 || pos > size1 + size2)
5096 {
5097 FREE_VARIABLES ();
5098 return -1;
5099 }
5100
5101 /* Initialize subexpression text positions to -1 to mark ones that no
5102 start_memory/stop_memory has been seen for. Also initialize the
5103 register information struct. */
5104 for (reg = 1; reg < num_regs; reg++)
5105 regstart[reg] = regend[reg] = NULL;
5106
5107 /* We move `string1' into `string2' if the latter's empty -- but not if
5108 `string1' is null. */
5109 if (size2 == 0 && string1 != NULL)
5110 {
5111 string2 = string1;
5112 size2 = size1;
5113 string1 = 0;
5114 size1 = 0;
5115 }
5116 end1 = string1 + size1;
5117 end2 = string2 + size2;
5118
5119 /* `p' scans through the pattern as `d' scans through the data.
5120 `dend' is the end of the input string that `d' points within. `d'
5121 is advanced into the following input string whenever necessary, but
5122 this happens before fetching; therefore, at the beginning of the
5123 loop, `d' can be pointing at the end of a string, but it cannot
5124 equal `string2'. */
5125 if (pos >= size1)
5126 {
5127 /* Only match within string2. */
5128 d = string2 + pos - size1;
5129 dend = end_match_2 = string2 + stop - size1;
5130 end_match_1 = end1; /* Just to give it a value. */
5131 }
5132 else
5133 {
5134 if (stop < size1)
5135 {
5136 /* Only match within string1. */
5137 end_match_1 = string1 + stop;
5138 /* BEWARE!
5139 When we reach end_match_1, PREFETCH normally switches to string2.
5140 But in the present case, this means that just doing a PREFETCH
5141 makes us jump from `stop' to `gap' within the string.
5142 What we really want here is for the search to stop as
5143 soon as we hit end_match_1. That's why we set end_match_2
5144 to end_match_1 (since PREFETCH fails as soon as we hit
5145 end_match_2). */
5146 end_match_2 = end_match_1;
5147 }
5148 else
5149 { /* It's important to use this code when stop == size so that
5150 moving `d' from end1 to string2 will not prevent the d == dend
5151 check from catching the end of string. */
5152 end_match_1 = end1;
5153 end_match_2 = string2 + stop - size1;
5154 }
5155 d = string1 + pos;
5156 dend = end_match_1;
5157 }
5158
5159 DEBUG_PRINT1 ("The compiled pattern is: ");
5160 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5161 DEBUG_PRINT1 ("The string to match is: `");
5162 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5163 DEBUG_PRINT1 ("'\n");
5164
5165 /* This loops over pattern commands. It exits by returning from the
5166 function if the match is complete, or it drops through if the match
5167 fails at this starting point in the input data. */
5168 for (;;)
5169 {
5170 DEBUG_PRINT2 ("\n%p: ", p);
5171
5172 if (p == pend)
5173 { /* End of pattern means we might have succeeded. */
5174 DEBUG_PRINT1 ("end of pattern ... ");
5175
5176 /* If we haven't matched the entire string, and we want the
5177 longest match, try backtracking. */
5178 if (d != end_match_2)
5179 {
5180 /* 1 if this match ends in the same string (string1 or string2)
5181 as the best previous match. */
5182 boolean same_str_p = (FIRST_STRING_P (match_end)
5183 == FIRST_STRING_P (d));
5184 /* 1 if this match is the best seen so far. */
5185 boolean best_match_p;
5186
5187 /* AIX compiler got confused when this was combined
5188 with the previous declaration. */
5189 if (same_str_p)
5190 best_match_p = d > match_end;
5191 else
5192 best_match_p = !FIRST_STRING_P (d);
5193
5194 DEBUG_PRINT1 ("backtracking.\n");
5195
5196 if (!FAIL_STACK_EMPTY ())
5197 { /* More failure points to try. */
5198
5199 /* If exceeds best match so far, save it. */
5200 if (!best_regs_set || best_match_p)
5201 {
5202 best_regs_set = true;
5203 match_end = d;
5204
5205 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5206
5207 for (reg = 1; reg < num_regs; reg++)
5208 {
5209 best_regstart[reg] = regstart[reg];
5210 best_regend[reg] = regend[reg];
5211 }
5212 }
5213 goto fail;
5214 }
5215
5216 /* If no failure points, don't restore garbage. And if
5217 last match is real best match, don't restore second
5218 best one. */
5219 else if (best_regs_set && !best_match_p)
5220 {
5221 restore_best_regs:
5222 /* Restore best match. It may happen that `dend ==
5223 end_match_1' while the restored d is in string2.
5224 For example, the pattern `x.*y.*z' against the
5225 strings `x-' and `y-z-', if the two strings are
5226 not consecutive in memory. */
5227 DEBUG_PRINT1 ("Restoring best registers.\n");
5228
5229 d = match_end;
5230 dend = ((d >= string1 && d <= end1)
5231 ? end_match_1 : end_match_2);
5232
5233 for (reg = 1; reg < num_regs; reg++)
5234 {
5235 regstart[reg] = best_regstart[reg];
5236 regend[reg] = best_regend[reg];
5237 }
5238 }
5239 } /* d != end_match_2 */
5240
5241 succeed_label:
5242 DEBUG_PRINT1 ("Accepting match.\n");
5243
5244 /* If caller wants register contents data back, do it. */
5245 if (regs && !bufp->no_sub)
5246 {
5247 /* Have the register data arrays been allocated? */
5248 if (bufp->regs_allocated == REGS_UNALLOCATED)
5249 { /* No. So allocate them with malloc. We need one
5250 extra element beyond `num_regs' for the `-1' marker
5251 GNU code uses. */
5252 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5253 regs->start = TALLOC (regs->num_regs, regoff_t);
5254 regs->end = TALLOC (regs->num_regs, regoff_t);
5255 if (regs->start == NULL || regs->end == NULL)
5256 {
5257 FREE_VARIABLES ();
5258 return -2;
5259 }
5260 bufp->regs_allocated = REGS_REALLOCATE;
5261 }
5262 else if (bufp->regs_allocated == REGS_REALLOCATE)
5263 { /* Yes. If we need more elements than were already
5264 allocated, reallocate them. If we need fewer, just
5265 leave it alone. */
5266 if (regs->num_regs < num_regs + 1)
5267 {
5268 regs->num_regs = num_regs + 1;
5269 RETALLOC (regs->start, regs->num_regs, regoff_t);
5270 RETALLOC (regs->end, regs->num_regs, regoff_t);
5271 if (regs->start == NULL || regs->end == NULL)
5272 {
5273 FREE_VARIABLES ();
5274 return -2;
5275 }
5276 }
5277 }
5278 else
5279 {
5280 /* These braces fend off a "empty body in an else-statement"
5281 warning under GCC when assert expands to nothing. */
5282 assert (bufp->regs_allocated == REGS_FIXED);
5283 }
5284
5285 /* Convert the pointer data in `regstart' and `regend' to
5286 indices. Register zero has to be set differently,
5287 since we haven't kept track of any info for it. */
5288 if (regs->num_regs > 0)
5289 {
5290 regs->start[0] = pos;
5291 regs->end[0] = POINTER_TO_OFFSET (d);
5292 }
5293
5294 /* Go through the first `min (num_regs, regs->num_regs)'
5295 registers, since that is all we initialized. */
5296 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5297 {
5298 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5299 regs->start[reg] = regs->end[reg] = -1;
5300 else
5301 {
5302 regs->start[reg]
5303 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5304 regs->end[reg]
5305 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5306 }
5307 }
5308
5309 /* If the regs structure we return has more elements than
5310 were in the pattern, set the extra elements to -1. If
5311 we (re)allocated the registers, this is the case,
5312 because we always allocate enough to have at least one
5313 -1 at the end. */
5314 for (reg = num_regs; reg < regs->num_regs; reg++)
5315 regs->start[reg] = regs->end[reg] = -1;
5316 } /* regs && !bufp->no_sub */
5317
5318 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5319 nfailure_points_pushed, nfailure_points_popped,
5320 nfailure_points_pushed - nfailure_points_popped);
5321 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5322
5323 mcnt = POINTER_TO_OFFSET (d) - pos;
5324
5325 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5326
5327 FREE_VARIABLES ();
5328 return mcnt;
5329 }
5330
5331 /* Otherwise match next pattern command. */
5332 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5333 {
5334 /* Ignore these. Used to ignore the n of succeed_n's which
5335 currently have n == 0. */
5336 case no_op:
5337 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5338 break;
5339
5340 case succeed:
5341 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5342 goto succeed_label;
5343
5344 /* Match the next n pattern characters exactly. The following
5345 byte in the pattern defines n, and the n bytes after that
5346 are the characters to match. */
5347 case exactn:
5348 mcnt = *p++;
5349 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5350
5351 /* Remember the start point to rollback upon failure. */
5352 dfail = d;
5353
5354 #ifndef emacs
5355 /* This is written out as an if-else so we don't waste time
5356 testing `translate' inside the loop. */
5357 if (RE_TRANSLATE_P (translate))
5358 do
5359 {
5360 PREFETCH ();
5361 if (RE_TRANSLATE (translate, *d) != *p++)
5362 {
5363 d = dfail;
5364 goto fail;
5365 }
5366 d++;
5367 }
5368 while (--mcnt);
5369 else
5370 do
5371 {
5372 PREFETCH ();
5373 if (*d++ != *p++)
5374 {
5375 d = dfail;
5376 goto fail;
5377 }
5378 }
5379 while (--mcnt);
5380 #else /* emacs */
5381 /* The cost of testing `translate' is comparatively small. */
5382 if (target_multibyte)
5383 do
5384 {
5385 int pat_charlen, buf_charlen;
5386 int pat_ch, buf_ch;
5387
5388 PREFETCH ();
5389 if (multibyte)
5390 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5391 else
5392 {
5393 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5394 pat_charlen = 1;
5395 }
5396 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5397
5398 if (TRANSLATE (buf_ch) != pat_ch)
5399 {
5400 d = dfail;
5401 goto fail;
5402 }
5403
5404 p += pat_charlen;
5405 d += buf_charlen;
5406 mcnt -= pat_charlen;
5407 }
5408 while (mcnt > 0);
5409 else
5410 do
5411 {
5412 int pat_charlen;
5413 int pat_ch, buf_ch;
5414
5415 PREFETCH ();
5416 if (multibyte)
5417 {
5418 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5419 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5420 }
5421 else
5422 {
5423 pat_ch = *p;
5424 pat_charlen = 1;
5425 }
5426 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5427 if (! CHAR_BYTE8_P (buf_ch))
5428 {
5429 buf_ch = TRANSLATE (buf_ch);
5430 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5431 if (buf_ch < 0)
5432 buf_ch = *d;
5433 }
5434 else
5435 buf_ch = *d;
5436 if (buf_ch != pat_ch)
5437 {
5438 d = dfail;
5439 goto fail;
5440 }
5441 p += pat_charlen;
5442 d++;
5443 }
5444 while (--mcnt);
5445 #endif
5446 break;
5447
5448
5449 /* Match any character except possibly a newline or a null. */
5450 case anychar:
5451 {
5452 int buf_charlen;
5453 re_wchar_t buf_ch;
5454
5455 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5456
5457 PREFETCH ();
5458 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5459 target_multibyte);
5460 buf_ch = TRANSLATE (buf_ch);
5461
5462 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5463 && buf_ch == '\n')
5464 || ((bufp->syntax & RE_DOT_NOT_NULL)
5465 && buf_ch == '\000'))
5466 goto fail;
5467
5468 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5469 d += buf_charlen;
5470 }
5471 break;
5472
5473
5474 case charset:
5475 case charset_not:
5476 {
5477 register unsigned int c;
5478 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5479 int len;
5480
5481 /* Start of actual range_table, or end of bitmap if there is no
5482 range table. */
5483 re_char *range_table IF_LINT (= NULL);
5484
5485 /* Nonzero if there is a range table. */
5486 int range_table_exists;
5487
5488 /* Number of ranges of range table. This is not included
5489 in the initial byte-length of the command. */
5490 int count = 0;
5491
5492 /* Whether matching against a unibyte character. */
5493 boolean unibyte_char = false;
5494
5495 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5496
5497 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5498
5499 if (range_table_exists)
5500 {
5501 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5502 EXTRACT_NUMBER_AND_INCR (count, range_table);
5503 }
5504
5505 PREFETCH ();
5506 c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5507 if (target_multibyte)
5508 {
5509 int c1;
5510
5511 c = TRANSLATE (c);
5512 c1 = RE_CHAR_TO_UNIBYTE (c);
5513 if (c1 >= 0)
5514 {
5515 unibyte_char = true;
5516 c = c1;
5517 }
5518 }
5519 else
5520 {
5521 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5522
5523 if (! CHAR_BYTE8_P (c1))
5524 {
5525 c1 = TRANSLATE (c1);
5526 c1 = RE_CHAR_TO_UNIBYTE (c1);
5527 if (c1 >= 0)
5528 {
5529 unibyte_char = true;
5530 c = c1;
5531 }
5532 }
5533 else
5534 unibyte_char = true;
5535 }
5536
5537 if (unibyte_char && c < (1 << BYTEWIDTH))
5538 { /* Lookup bitmap. */
5539 /* Cast to `unsigned' instead of `unsigned char' in
5540 case the bit list is a full 32 bytes long. */
5541 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5542 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5543 not = !not;
5544 }
5545 #ifdef emacs
5546 else if (range_table_exists)
5547 {
5548 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5549
5550 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5551 | (class_bits & BIT_MULTIBYTE)
5552 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5553 | (class_bits & BIT_SPACE && ISSPACE (c))
5554 | (class_bits & BIT_UPPER && ISUPPER (c))
5555 | (class_bits & BIT_WORD && ISWORD (c)))
5556 not = !not;
5557 else
5558 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5559 }
5560 #endif /* emacs */
5561
5562 if (range_table_exists)
5563 p = CHARSET_RANGE_TABLE_END (range_table, count);
5564 else
5565 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5566
5567 if (!not) goto fail;
5568
5569 d += len;
5570 }
5571 break;
5572
5573
5574 /* The beginning of a group is represented by start_memory.
5575 The argument is the register number. The text
5576 matched within the group is recorded (in the internal
5577 registers data structure) under the register number. */
5578 case start_memory:
5579 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5580
5581 /* In case we need to undo this operation (via backtracking). */
5582 PUSH_FAILURE_REG ((unsigned int)*p);
5583
5584 regstart[*p] = d;
5585 regend[*p] = NULL; /* probably unnecessary. -sm */
5586 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5587
5588 /* Move past the register number and inner group count. */
5589 p += 1;
5590 break;
5591
5592
5593 /* The stop_memory opcode represents the end of a group. Its
5594 argument is the same as start_memory's: the register number. */
5595 case stop_memory:
5596 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5597
5598 assert (!REG_UNSET (regstart[*p]));
5599 /* Strictly speaking, there should be code such as:
5600
5601 assert (REG_UNSET (regend[*p]));
5602 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5603
5604 But the only info to be pushed is regend[*p] and it is known to
5605 be UNSET, so there really isn't anything to push.
5606 Not pushing anything, on the other hand deprives us from the
5607 guarantee that regend[*p] is UNSET since undoing this operation
5608 will not reset its value properly. This is not important since
5609 the value will only be read on the next start_memory or at
5610 the very end and both events can only happen if this stop_memory
5611 is *not* undone. */
5612
5613 regend[*p] = d;
5614 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5615
5616 /* Move past the register number and the inner group count. */
5617 p += 1;
5618 break;
5619
5620
5621 /* \<digit> has been turned into a `duplicate' command which is
5622 followed by the numeric value of <digit> as the register number. */
5623 case duplicate:
5624 {
5625 register re_char *d2, *dend2;
5626 int regno = *p++; /* Get which register to match against. */
5627 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5628
5629 /* Can't back reference a group which we've never matched. */
5630 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5631 goto fail;
5632
5633 /* Where in input to try to start matching. */
5634 d2 = regstart[regno];
5635
5636 /* Remember the start point to rollback upon failure. */
5637 dfail = d;
5638
5639 /* Where to stop matching; if both the place to start and
5640 the place to stop matching are in the same string, then
5641 set to the place to stop, otherwise, for now have to use
5642 the end of the first string. */
5643
5644 dend2 = ((FIRST_STRING_P (regstart[regno])
5645 == FIRST_STRING_P (regend[regno]))
5646 ? regend[regno] : end_match_1);
5647 for (;;)
5648 {
5649 /* If necessary, advance to next segment in register
5650 contents. */
5651 while (d2 == dend2)
5652 {
5653 if (dend2 == end_match_2) break;
5654 if (dend2 == regend[regno]) break;
5655
5656 /* End of string1 => advance to string2. */
5657 d2 = string2;
5658 dend2 = regend[regno];
5659 }
5660 /* At end of register contents => success */
5661 if (d2 == dend2) break;
5662
5663 /* If necessary, advance to next segment in data. */
5664 PREFETCH ();
5665
5666 /* How many characters left in this segment to match. */
5667 mcnt = dend - d;
5668
5669 /* Want how many consecutive characters we can match in
5670 one shot, so, if necessary, adjust the count. */
5671 if (mcnt > dend2 - d2)
5672 mcnt = dend2 - d2;
5673
5674 /* Compare that many; failure if mismatch, else move
5675 past them. */
5676 if (RE_TRANSLATE_P (translate)
5677 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5678 : memcmp (d, d2, mcnt))
5679 {
5680 d = dfail;
5681 goto fail;
5682 }
5683 d += mcnt, d2 += mcnt;
5684 }
5685 }
5686 break;
5687
5688
5689 /* begline matches the empty string at the beginning of the string
5690 (unless `not_bol' is set in `bufp'), and after newlines. */
5691 case begline:
5692 DEBUG_PRINT1 ("EXECUTING begline.\n");
5693
5694 if (AT_STRINGS_BEG (d))
5695 {
5696 if (!bufp->not_bol) break;
5697 }
5698 else
5699 {
5700 unsigned c;
5701 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5702 if (c == '\n')
5703 break;
5704 }
5705 /* In all other cases, we fail. */
5706 goto fail;
5707
5708
5709 /* endline is the dual of begline. */
5710 case endline:
5711 DEBUG_PRINT1 ("EXECUTING endline.\n");
5712
5713 if (AT_STRINGS_END (d))
5714 {
5715 if (!bufp->not_eol) break;
5716 }
5717 else
5718 {
5719 PREFETCH_NOLIMIT ();
5720 if (*d == '\n')
5721 break;
5722 }
5723 goto fail;
5724
5725
5726 /* Match at the very beginning of the data. */
5727 case begbuf:
5728 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5729 if (AT_STRINGS_BEG (d))
5730 break;
5731 goto fail;
5732
5733
5734 /* Match at the very end of the data. */
5735 case endbuf:
5736 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5737 if (AT_STRINGS_END (d))
5738 break;
5739 goto fail;
5740
5741
5742 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5743 pushes NULL as the value for the string on the stack. Then
5744 `POP_FAILURE_POINT' will keep the current value for the
5745 string, instead of restoring it. To see why, consider
5746 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5747 then the . fails against the \n. But the next thing we want
5748 to do is match the \n against the \n; if we restored the
5749 string value, we would be back at the foo.
5750
5751 Because this is used only in specific cases, we don't need to
5752 check all the things that `on_failure_jump' does, to make
5753 sure the right things get saved on the stack. Hence we don't
5754 share its code. The only reason to push anything on the
5755 stack at all is that otherwise we would have to change
5756 `anychar's code to do something besides goto fail in this
5757 case; that seems worse than this. */
5758 case on_failure_keep_string_jump:
5759 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5760 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5761 mcnt, p + mcnt);
5762
5763 PUSH_FAILURE_POINT (p - 3, NULL);
5764 break;
5765
5766 /* A nasty loop is introduced by the non-greedy *? and +?.
5767 With such loops, the stack only ever contains one failure point
5768 at a time, so that a plain on_failure_jump_loop kind of
5769 cycle detection cannot work. Worse yet, such a detection
5770 can not only fail to detect a cycle, but it can also wrongly
5771 detect a cycle (between different instantiations of the same
5772 loop).
5773 So the method used for those nasty loops is a little different:
5774 We use a special cycle-detection-stack-frame which is pushed
5775 when the on_failure_jump_nastyloop failure-point is *popped*.
5776 This special frame thus marks the beginning of one iteration
5777 through the loop and we can hence easily check right here
5778 whether something matched between the beginning and the end of
5779 the loop. */
5780 case on_failure_jump_nastyloop:
5781 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5782 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5783 mcnt, p + mcnt);
5784
5785 assert ((re_opcode_t)p[-4] == no_op);
5786 {
5787 int cycle = 0;
5788 CHECK_INFINITE_LOOP (p - 4, d);
5789 if (!cycle)
5790 /* If there's a cycle, just continue without pushing
5791 this failure point. The failure point is the "try again"
5792 option, which shouldn't be tried.
5793 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5794 PUSH_FAILURE_POINT (p - 3, d);
5795 }
5796 break;
5797
5798 /* Simple loop detecting on_failure_jump: just check on the
5799 failure stack if the same spot was already hit earlier. */
5800 case on_failure_jump_loop:
5801 on_failure:
5802 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5803 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5804 mcnt, p + mcnt);
5805 {
5806 int cycle = 0;
5807 CHECK_INFINITE_LOOP (p - 3, d);
5808 if (cycle)
5809 /* If there's a cycle, get out of the loop, as if the matching
5810 had failed. We used to just `goto fail' here, but that was
5811 aborting the search a bit too early: we want to keep the
5812 empty-loop-match and keep matching after the loop.
5813 We want (x?)*y\1z to match both xxyz and xxyxz. */
5814 p += mcnt;
5815 else
5816 PUSH_FAILURE_POINT (p - 3, d);
5817 }
5818 break;
5819
5820
5821 /* Uses of on_failure_jump:
5822
5823 Each alternative starts with an on_failure_jump that points
5824 to the beginning of the next alternative. Each alternative
5825 except the last ends with a jump that in effect jumps past
5826 the rest of the alternatives. (They really jump to the
5827 ending jump of the following alternative, because tensioning
5828 these jumps is a hassle.)
5829
5830 Repeats start with an on_failure_jump that points past both
5831 the repetition text and either the following jump or
5832 pop_failure_jump back to this on_failure_jump. */
5833 case on_failure_jump:
5834 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5835 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5836 mcnt, p + mcnt);
5837
5838 PUSH_FAILURE_POINT (p -3, d);
5839 break;
5840
5841 /* This operation is used for greedy *.
5842 Compare the beginning of the repeat with what in the
5843 pattern follows its end. If we can establish that there
5844 is nothing that they would both match, i.e., that we
5845 would have to backtrack because of (as in, e.g., `a*a')
5846 then we can use a non-backtracking loop based on
5847 on_failure_keep_string_jump instead of on_failure_jump. */
5848 case on_failure_jump_smart:
5849 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5850 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5851 mcnt, p + mcnt);
5852 {
5853 re_char *p1 = p; /* Next operation. */
5854 /* Here, we discard `const', making re_match non-reentrant. */
5855 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5856 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5857
5858 p -= 3; /* Reset so that we will re-execute the
5859 instruction once it's been changed. */
5860
5861 EXTRACT_NUMBER (mcnt, p2 - 2);
5862
5863 /* Ensure this is a indeed the trivial kind of loop
5864 we are expecting. */
5865 assert (skip_one_char (p1) == p2 - 3);
5866 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5867 DEBUG_STATEMENT (debug += 2);
5868 if (mutually_exclusive_p (bufp, p1, p2))
5869 {
5870 /* Use a fast `on_failure_keep_string_jump' loop. */
5871 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5872 *p3 = (unsigned char) on_failure_keep_string_jump;
5873 STORE_NUMBER (p2 - 2, mcnt + 3);
5874 }
5875 else
5876 {
5877 /* Default to a safe `on_failure_jump' loop. */
5878 DEBUG_PRINT1 (" smart default => slow loop.\n");
5879 *p3 = (unsigned char) on_failure_jump;
5880 }
5881 DEBUG_STATEMENT (debug -= 2);
5882 }
5883 break;
5884
5885 /* Unconditionally jump (without popping any failure points). */
5886 case jump:
5887 unconditional_jump:
5888 IMMEDIATE_QUIT_CHECK;
5889 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5890 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5891 p += mcnt; /* Do the jump. */
5892 DEBUG_PRINT2 ("(to %p).\n", p);
5893 break;
5894
5895
5896 /* Have to succeed matching what follows at least n times.
5897 After that, handle like `on_failure_jump'. */
5898 case succeed_n:
5899 /* Signedness doesn't matter since we only compare MCNT to 0. */
5900 EXTRACT_NUMBER (mcnt, p + 2);
5901 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5902
5903 /* Originally, mcnt is how many times we HAVE to succeed. */
5904 if (mcnt != 0)
5905 {
5906 /* Here, we discard `const', making re_match non-reentrant. */
5907 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5908 mcnt--;
5909 p += 4;
5910 PUSH_NUMBER (p2, mcnt);
5911 }
5912 else
5913 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5914 goto on_failure;
5915 break;
5916
5917 case jump_n:
5918 /* Signedness doesn't matter since we only compare MCNT to 0. */
5919 EXTRACT_NUMBER (mcnt, p + 2);
5920 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5921
5922 /* Originally, this is how many times we CAN jump. */
5923 if (mcnt != 0)
5924 {
5925 /* Here, we discard `const', making re_match non-reentrant. */
5926 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5927 mcnt--;
5928 PUSH_NUMBER (p2, mcnt);
5929 goto unconditional_jump;
5930 }
5931 /* If don't have to jump any more, skip over the rest of command. */
5932 else
5933 p += 4;
5934 break;
5935
5936 case set_number_at:
5937 {
5938 unsigned char *p2; /* Location of the counter. */
5939 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5940
5941 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5942 /* Here, we discard `const', making re_match non-reentrant. */
5943 p2 = (unsigned char*) p + mcnt;
5944 /* Signedness doesn't matter since we only copy MCNT's bits . */
5945 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5946 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
5947 PUSH_NUMBER (p2, mcnt);
5948 break;
5949 }
5950
5951 case wordbound:
5952 case notwordbound:
5953 {
5954 boolean not = (re_opcode_t) *(p - 1) == notwordbound;
5955 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5956
5957 /* We SUCCEED (or FAIL) in one of the following cases: */
5958
5959 /* Case 1: D is at the beginning or the end of string. */
5960 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5961 not = !not;
5962 else
5963 {
5964 /* C1 is the character before D, S1 is the syntax of C1, C2
5965 is the character at D, and S2 is the syntax of C2. */
5966 re_wchar_t c1, c2;
5967 int s1, s2;
5968 int dummy;
5969 #ifdef emacs
5970 ssize_t offset = PTR_TO_OFFSET (d - 1);
5971 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5972 UPDATE_SYNTAX_TABLE (charpos);
5973 #endif
5974 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5975 s1 = SYNTAX (c1);
5976 #ifdef emacs
5977 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
5978 #endif
5979 PREFETCH_NOLIMIT ();
5980 GET_CHAR_AFTER (c2, d, dummy);
5981 s2 = SYNTAX (c2);
5982
5983 if (/* Case 2: Only one of S1 and S2 is Sword. */
5984 ((s1 == Sword) != (s2 == Sword))
5985 /* Case 3: Both of S1 and S2 are Sword, and macro
5986 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5987 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5988 not = !not;
5989 }
5990 if (not)
5991 break;
5992 else
5993 goto fail;
5994 }
5995
5996 case wordbeg:
5997 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5998
5999 /* We FAIL in one of the following cases: */
6000
6001 /* Case 1: D is at the end of string. */
6002 if (AT_STRINGS_END (d))
6003 goto fail;
6004 else
6005 {
6006 /* C1 is the character before D, S1 is the syntax of C1, C2
6007 is the character at D, and S2 is the syntax of C2. */
6008 re_wchar_t c1, c2;
6009 int s1, s2;
6010 int dummy;
6011 #ifdef emacs
6012 ssize_t offset = PTR_TO_OFFSET (d);
6013 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6014 UPDATE_SYNTAX_TABLE (charpos);
6015 #endif
6016 PREFETCH ();
6017 GET_CHAR_AFTER (c2, d, dummy);
6018 s2 = SYNTAX (c2);
6019
6020 /* Case 2: S2 is not Sword. */
6021 if (s2 != Sword)
6022 goto fail;
6023
6024 /* Case 3: D is not at the beginning of string ... */
6025 if (!AT_STRINGS_BEG (d))
6026 {
6027 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6028 #ifdef emacs
6029 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6030 #endif
6031 s1 = SYNTAX (c1);
6032
6033 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6034 returns 0. */
6035 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6036 goto fail;
6037 }
6038 }
6039 break;
6040
6041 case wordend:
6042 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6043
6044 /* We FAIL in one of the following cases: */
6045
6046 /* Case 1: D is at the beginning of string. */
6047 if (AT_STRINGS_BEG (d))
6048 goto fail;
6049 else
6050 {
6051 /* C1 is the character before D, S1 is the syntax of C1, C2
6052 is the character at D, and S2 is the syntax of C2. */
6053 re_wchar_t c1, c2;
6054 int s1, s2;
6055 int dummy;
6056 #ifdef emacs
6057 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6058 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6059 UPDATE_SYNTAX_TABLE (charpos);
6060 #endif
6061 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6062 s1 = SYNTAX (c1);
6063
6064 /* Case 2: S1 is not Sword. */
6065 if (s1 != Sword)
6066 goto fail;
6067
6068 /* Case 3: D is not at the end of string ... */
6069 if (!AT_STRINGS_END (d))
6070 {
6071 PREFETCH_NOLIMIT ();
6072 GET_CHAR_AFTER (c2, d, dummy);
6073 #ifdef emacs
6074 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6075 #endif
6076 s2 = SYNTAX (c2);
6077
6078 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6079 returns 0. */
6080 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6081 goto fail;
6082 }
6083 }
6084 break;
6085
6086 case symbeg:
6087 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6088
6089 /* We FAIL in one of the following cases: */
6090
6091 /* Case 1: D is at the end of string. */
6092 if (AT_STRINGS_END (d))
6093 goto fail;
6094 else
6095 {
6096 /* C1 is the character before D, S1 is the syntax of C1, C2
6097 is the character at D, and S2 is the syntax of C2. */
6098 re_wchar_t c1, c2;
6099 int s1, s2;
6100 #ifdef emacs
6101 ssize_t offset = PTR_TO_OFFSET (d);
6102 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6103 UPDATE_SYNTAX_TABLE (charpos);
6104 #endif
6105 PREFETCH ();
6106 c2 = RE_STRING_CHAR (d, target_multibyte);
6107 s2 = SYNTAX (c2);
6108
6109 /* Case 2: S2 is neither Sword nor Ssymbol. */
6110 if (s2 != Sword && s2 != Ssymbol)
6111 goto fail;
6112
6113 /* Case 3: D is not at the beginning of string ... */
6114 if (!AT_STRINGS_BEG (d))
6115 {
6116 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6117 #ifdef emacs
6118 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6119 #endif
6120 s1 = SYNTAX (c1);
6121
6122 /* ... and S1 is Sword or Ssymbol. */
6123 if (s1 == Sword || s1 == Ssymbol)
6124 goto fail;
6125 }
6126 }
6127 break;
6128
6129 case symend:
6130 DEBUG_PRINT1 ("EXECUTING symend.\n");
6131
6132 /* We FAIL in one of the following cases: */
6133
6134 /* Case 1: D is at the beginning of string. */
6135 if (AT_STRINGS_BEG (d))
6136 goto fail;
6137 else
6138 {
6139 /* C1 is the character before D, S1 is the syntax of C1, C2
6140 is the character at D, and S2 is the syntax of C2. */
6141 re_wchar_t c1, c2;
6142 int s1, s2;
6143 #ifdef emacs
6144 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6145 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6146 UPDATE_SYNTAX_TABLE (charpos);
6147 #endif
6148 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6149 s1 = SYNTAX (c1);
6150
6151 /* Case 2: S1 is neither Ssymbol nor Sword. */
6152 if (s1 != Sword && s1 != Ssymbol)
6153 goto fail;
6154
6155 /* Case 3: D is not at the end of string ... */
6156 if (!AT_STRINGS_END (d))
6157 {
6158 PREFETCH_NOLIMIT ();
6159 c2 = RE_STRING_CHAR (d, target_multibyte);
6160 #ifdef emacs
6161 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6162 #endif
6163 s2 = SYNTAX (c2);
6164
6165 /* ... and S2 is Sword or Ssymbol. */
6166 if (s2 == Sword || s2 == Ssymbol)
6167 goto fail;
6168 }
6169 }
6170 break;
6171
6172 case syntaxspec:
6173 case notsyntaxspec:
6174 {
6175 boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6176 mcnt = *p++;
6177 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6178 PREFETCH ();
6179 #ifdef emacs
6180 {
6181 ssize_t offset = PTR_TO_OFFSET (d);
6182 ssize_t pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6183 UPDATE_SYNTAX_TABLE (pos1);
6184 }
6185 #endif
6186 {
6187 int len;
6188 re_wchar_t c;
6189
6190 GET_CHAR_AFTER (c, d, len);
6191 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6192 goto fail;
6193 d += len;
6194 }
6195 }
6196 break;
6197
6198 #ifdef emacs
6199 case before_dot:
6200 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6201 if (PTR_BYTE_POS (d) >= PT_BYTE)
6202 goto fail;
6203 break;
6204
6205 case at_dot:
6206 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6207 if (PTR_BYTE_POS (d) != PT_BYTE)
6208 goto fail;
6209 break;
6210
6211 case after_dot:
6212 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6213 if (PTR_BYTE_POS (d) <= PT_BYTE)
6214 goto fail;
6215 break;
6216
6217 case categoryspec:
6218 case notcategoryspec:
6219 {
6220 boolean not = (re_opcode_t) *(p - 1) == notcategoryspec;
6221 mcnt = *p++;
6222 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n",
6223 not?"not":"", mcnt);
6224 PREFETCH ();
6225
6226 {
6227 int len;
6228 re_wchar_t c;
6229 GET_CHAR_AFTER (c, d, len);
6230 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6231 goto fail;
6232 d += len;
6233 }
6234 }
6235 break;
6236
6237 #endif /* emacs */
6238
6239 default:
6240 abort ();
6241 }
6242 continue; /* Successfully executed one pattern command; keep going. */
6243
6244
6245 /* We goto here if a matching operation fails. */
6246 fail:
6247 IMMEDIATE_QUIT_CHECK;
6248 if (!FAIL_STACK_EMPTY ())
6249 {
6250 re_char *str, *pat;
6251 /* A restart point is known. Restore to that state. */
6252 DEBUG_PRINT1 ("\nFAIL:\n");
6253 POP_FAILURE_POINT (str, pat);
6254 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6255 {
6256 case on_failure_keep_string_jump:
6257 assert (str == NULL);
6258 goto continue_failure_jump;
6259
6260 case on_failure_jump_nastyloop:
6261 assert ((re_opcode_t)pat[-2] == no_op);
6262 PUSH_FAILURE_POINT (pat - 2, str);
6263 /* Fallthrough */
6264
6265 case on_failure_jump_loop:
6266 case on_failure_jump:
6267 case succeed_n:
6268 d = str;
6269 continue_failure_jump:
6270 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6271 p = pat + mcnt;
6272 break;
6273
6274 case no_op:
6275 /* A special frame used for nastyloops. */
6276 goto fail;
6277
6278 default:
6279 abort();
6280 }
6281
6282 assert (p >= bufp->buffer && p <= pend);
6283
6284 if (d >= string1 && d <= end1)
6285 dend = end_match_1;
6286 }
6287 else
6288 break; /* Matching at this starting point really fails. */
6289 } /* for (;;) */
6290
6291 if (best_regs_set)
6292 goto restore_best_regs;
6293
6294 FREE_VARIABLES ();
6295
6296 return -1; /* Failure to match. */
6297 } /* re_match_2 */
6298 \f
6299 /* Subroutine definitions for re_match_2. */
6300
6301 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6302 bytes; nonzero otherwise. */
6303
6304 static int
6305 bcmp_translate (const re_char *s1, const re_char *s2, register ssize_t len,
6306 RE_TRANSLATE_TYPE translate, const int target_multibyte)
6307 {
6308 register re_char *p1 = s1, *p2 = s2;
6309 re_char *p1_end = s1 + len;
6310 re_char *p2_end = s2 + len;
6311
6312 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6313 different lengths, but relying on a single `len' would break this. -sm */
6314 while (p1 < p1_end && p2 < p2_end)
6315 {
6316 int p1_charlen, p2_charlen;
6317 re_wchar_t p1_ch, p2_ch;
6318
6319 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6320 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6321
6322 if (RE_TRANSLATE (translate, p1_ch)
6323 != RE_TRANSLATE (translate, p2_ch))
6324 return 1;
6325
6326 p1 += p1_charlen, p2 += p2_charlen;
6327 }
6328
6329 if (p1 != p1_end || p2 != p2_end)
6330 return 1;
6331
6332 return 0;
6333 }
6334 \f
6335 /* Entry points for GNU code. */
6336
6337 /* re_compile_pattern is the GNU regular expression compiler: it
6338 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6339 Returns 0 if the pattern was valid, otherwise an error string.
6340
6341 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6342 are set in BUFP on entry.
6343
6344 We call regex_compile to do the actual compilation. */
6345
6346 const char *
6347 re_compile_pattern (const char *pattern, size_t length,
6348 struct re_pattern_buffer *bufp)
6349 {
6350 reg_errcode_t ret;
6351
6352 /* GNU code is written to assume at least RE_NREGS registers will be set
6353 (and at least one extra will be -1). */
6354 bufp->regs_allocated = REGS_UNALLOCATED;
6355
6356 /* And GNU code determines whether or not to get register information
6357 by passing null for the REGS argument to re_match, etc., not by
6358 setting no_sub. */
6359 bufp->no_sub = 0;
6360
6361 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6362
6363 if (!ret)
6364 return NULL;
6365 return gettext (re_error_msgid[(int) ret]);
6366 }
6367 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6368 \f
6369 /* Entry points compatible with 4.2 BSD regex library. We don't define
6370 them unless specifically requested. */
6371
6372 #if defined _REGEX_RE_COMP || defined _LIBC
6373
6374 /* BSD has one and only one pattern buffer. */
6375 static struct re_pattern_buffer re_comp_buf;
6376
6377 char *
6378 # ifdef _LIBC
6379 /* Make these definitions weak in libc, so POSIX programs can redefine
6380 these names if they don't use our functions, and still use
6381 regcomp/regexec below without link errors. */
6382 weak_function
6383 # endif
6384 re_comp (s)
6385 const char *s;
6386 {
6387 reg_errcode_t ret;
6388
6389 if (!s)
6390 {
6391 if (!re_comp_buf.buffer)
6392 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6393 return (char *) gettext ("No previous regular expression");
6394 return 0;
6395 }
6396
6397 if (!re_comp_buf.buffer)
6398 {
6399 re_comp_buf.buffer = (unsigned char *) malloc (200);
6400 if (re_comp_buf.buffer == NULL)
6401 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6402 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6403 re_comp_buf.allocated = 200;
6404
6405 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6406 if (re_comp_buf.fastmap == NULL)
6407 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6408 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6409 }
6410
6411 /* Since `re_exec' always passes NULL for the `regs' argument, we
6412 don't need to initialize the pattern buffer fields which affect it. */
6413
6414 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6415
6416 if (!ret)
6417 return NULL;
6418
6419 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6420 return (char *) gettext (re_error_msgid[(int) ret]);
6421 }
6422
6423
6424 regoff_t
6425 # ifdef _LIBC
6426 weak_function
6427 # endif
6428 re_exec (const char *s)
6429 {
6430 const size_t len = strlen (s);
6431 return
6432 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6433 }
6434 #endif /* _REGEX_RE_COMP */
6435 \f
6436 /* POSIX.2 functions. Don't define these for Emacs. */
6437
6438 #ifndef emacs
6439
6440 /* regcomp takes a regular expression as a string and compiles it.
6441
6442 PREG is a regex_t *. We do not expect any fields to be initialized,
6443 since POSIX says we shouldn't. Thus, we set
6444
6445 `buffer' to the compiled pattern;
6446 `used' to the length of the compiled pattern;
6447 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6448 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6449 RE_SYNTAX_POSIX_BASIC;
6450 `fastmap' to an allocated space for the fastmap;
6451 `fastmap_accurate' to zero;
6452 `re_nsub' to the number of subexpressions in PATTERN.
6453
6454 PATTERN is the address of the pattern string.
6455
6456 CFLAGS is a series of bits which affect compilation.
6457
6458 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6459 use POSIX basic syntax.
6460
6461 If REG_NEWLINE is set, then . and [^...] don't match newline.
6462 Also, regexec will try a match beginning after every newline.
6463
6464 If REG_ICASE is set, then we considers upper- and lowercase
6465 versions of letters to be equivalent when matching.
6466
6467 If REG_NOSUB is set, then when PREG is passed to regexec, that
6468 routine will report only success or failure, and nothing about the
6469 registers.
6470
6471 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6472 the return codes and their meanings.) */
6473
6474 reg_errcode_t
6475 regcomp (regex_t *__restrict preg, const char *__restrict pattern,
6476 int cflags)
6477 {
6478 reg_errcode_t ret;
6479 reg_syntax_t syntax
6480 = (cflags & REG_EXTENDED) ?
6481 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6482
6483 /* regex_compile will allocate the space for the compiled pattern. */
6484 preg->buffer = 0;
6485 preg->allocated = 0;
6486 preg->used = 0;
6487
6488 /* Try to allocate space for the fastmap. */
6489 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6490
6491 if (cflags & REG_ICASE)
6492 {
6493 unsigned i;
6494
6495 preg->translate
6496 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6497 * sizeof (*(RE_TRANSLATE_TYPE)0));
6498 if (preg->translate == NULL)
6499 return (int) REG_ESPACE;
6500
6501 /* Map uppercase characters to corresponding lowercase ones. */
6502 for (i = 0; i < CHAR_SET_SIZE; i++)
6503 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6504 }
6505 else
6506 preg->translate = NULL;
6507
6508 /* If REG_NEWLINE is set, newlines are treated differently. */
6509 if (cflags & REG_NEWLINE)
6510 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6511 syntax &= ~RE_DOT_NEWLINE;
6512 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6513 }
6514 else
6515 syntax |= RE_NO_NEWLINE_ANCHOR;
6516
6517 preg->no_sub = !!(cflags & REG_NOSUB);
6518
6519 /* POSIX says a null character in the pattern terminates it, so we
6520 can use strlen here in compiling the pattern. */
6521 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6522
6523 /* POSIX doesn't distinguish between an unmatched open-group and an
6524 unmatched close-group: both are REG_EPAREN. */
6525 if (ret == REG_ERPAREN)
6526 ret = REG_EPAREN;
6527
6528 if (ret == REG_NOERROR && preg->fastmap)
6529 { /* Compute the fastmap now, since regexec cannot modify the pattern
6530 buffer. */
6531 re_compile_fastmap (preg);
6532 if (preg->can_be_null)
6533 { /* The fastmap can't be used anyway. */
6534 free (preg->fastmap);
6535 preg->fastmap = NULL;
6536 }
6537 }
6538 return ret;
6539 }
6540 WEAK_ALIAS (__regcomp, regcomp)
6541
6542
6543 /* regexec searches for a given pattern, specified by PREG, in the
6544 string STRING.
6545
6546 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6547 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6548 least NMATCH elements, and we set them to the offsets of the
6549 corresponding matched substrings.
6550
6551 EFLAGS specifies `execution flags' which affect matching: if
6552 REG_NOTBOL is set, then ^ does not match at the beginning of the
6553 string; if REG_NOTEOL is set, then $ does not match at the end.
6554
6555 We return 0 if we find a match and REG_NOMATCH if not. */
6556
6557 reg_errcode_t
6558 regexec (const regex_t *__restrict preg, const char *__restrict string,
6559 size_t nmatch, regmatch_t pmatch[__restrict_arr], int eflags)
6560 {
6561 reg_errcode_t ret;
6562 struct re_registers regs;
6563 regex_t private_preg;
6564 size_t len = strlen (string);
6565 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6566
6567 private_preg = *preg;
6568
6569 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6570 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6571
6572 /* The user has told us exactly how many registers to return
6573 information about, via `nmatch'. We have to pass that on to the
6574 matching routines. */
6575 private_preg.regs_allocated = REGS_FIXED;
6576
6577 if (want_reg_info)
6578 {
6579 regs.num_regs = nmatch;
6580 regs.start = TALLOC (nmatch * 2, regoff_t);
6581 if (regs.start == NULL)
6582 return REG_NOMATCH;
6583 regs.end = regs.start + nmatch;
6584 }
6585
6586 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6587 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6588 was a little bit longer but still only matching the real part.
6589 This works because the `endline' will check for a '\n' and will find a
6590 '\0', correctly deciding that this is not the end of a line.
6591 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6592 a convenient '\0' there. For all we know, the string could be preceded
6593 by '\n' which would throw things off. */
6594
6595 /* Perform the searching operation. */
6596 ret = re_search (&private_preg, string, len,
6597 /* start: */ 0, /* range: */ len,
6598 want_reg_info ? &regs : (struct re_registers *) 0);
6599
6600 /* Copy the register information to the POSIX structure. */
6601 if (want_reg_info)
6602 {
6603 if (ret >= 0)
6604 {
6605 unsigned r;
6606
6607 for (r = 0; r < nmatch; r++)
6608 {
6609 pmatch[r].rm_so = regs.start[r];
6610 pmatch[r].rm_eo = regs.end[r];
6611 }
6612 }
6613
6614 /* If we needed the temporary register info, free the space now. */
6615 free (regs.start);
6616 }
6617
6618 /* We want zero return to mean success, unlike `re_search'. */
6619 return ret >= 0 ? REG_NOERROR : REG_NOMATCH;
6620 }
6621 WEAK_ALIAS (__regexec, regexec)
6622
6623
6624 /* Returns a message corresponding to an error code, ERR_CODE, returned
6625 from either regcomp or regexec. We don't use PREG here.
6626
6627 ERR_CODE was previously called ERRCODE, but that name causes an
6628 error with msvc8 compiler. */
6629
6630 size_t
6631 regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size)
6632 {
6633 const char *msg;
6634 size_t msg_size;
6635
6636 if (err_code < 0
6637 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6638 /* Only error codes returned by the rest of the code should be passed
6639 to this routine. If we are given anything else, or if other regex
6640 code generates an invalid error code, then the program has a bug.
6641 Dump core so we can fix it. */
6642 abort ();
6643
6644 msg = gettext (re_error_msgid[err_code]);
6645
6646 msg_size = strlen (msg) + 1; /* Includes the null. */
6647
6648 if (errbuf_size != 0)
6649 {
6650 if (msg_size > errbuf_size)
6651 {
6652 strncpy (errbuf, msg, errbuf_size - 1);
6653 errbuf[errbuf_size - 1] = 0;
6654 }
6655 else
6656 strcpy (errbuf, msg);
6657 }
6658
6659 return msg_size;
6660 }
6661 WEAK_ALIAS (__regerror, regerror)
6662
6663
6664 /* Free dynamically allocated space used by PREG. */
6665
6666 void
6667 regfree (regex_t *preg)
6668 {
6669 free (preg->buffer);
6670 preg->buffer = NULL;
6671
6672 preg->allocated = 0;
6673 preg->used = 0;
6674
6675 free (preg->fastmap);
6676 preg->fastmap = NULL;
6677 preg->fastmap_accurate = 0;
6678
6679 free (preg->translate);
6680 preg->translate = NULL;
6681 }
6682 WEAK_ALIAS (__regfree, regfree)
6683
6684 #endif /* not emacs */
Unnamed repository; edit this file 'description' to name the repository.