HCoop / bpt / emacs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Use functions, not macros, for up- and down-casing.
[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 #if defined STDC_HEADERS && !defined emacs
41 # include <stddef.h>
42 #else
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
45 #endif
46
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
49 #if defined _LIBC
50 #define WIDE_CHAR_SUPPORT 1
51 #else
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
54 #endif
55
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
58 #if WIDE_CHAR_SUPPORT
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60 # include <wchar.h>
61 # include <wctype.h>
62 #endif
63
64 #ifdef _LIBC
65 /* We have to keep the namespace clean. */
66 # define regfree(preg) __regfree (preg)
67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69 # define regerror(err_code, preg, errbuf, errbuf_size) \
70 __regerror(err_code, preg, errbuf, errbuf_size)
71 # define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75 # define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77 # define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79 # define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81 # define re_set_syntax(syntax) __re_set_syntax (syntax)
82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
85
86 /* Make sure we call libc's function even if the user overrides them. */
87 # define btowc __btowc
88 # define iswctype __iswctype
89 # define wctype __wctype
90
91 # define WEAK_ALIAS(a,b) weak_alias (a, b)
92
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
97 #else
98 # define WEAK_ALIAS(a,b)
99 #endif
100
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
104 #else
105 # define gettext(msgid) (msgid)
106 #endif
107
108 #ifndef gettext_noop
109 /* This define is so xgettext can find the internationalizable
110 strings. */
111 # define gettext_noop(String) String
112 #endif
113
114 /* The `emacs' switch turns on certain matching commands
115 that make sense only in Emacs. */
116 #ifdef emacs
117
118 # include <setjmp.h>
119 # include "lisp.h"
120 # include "buffer.h"
121
122 /* Make syntax table lookup grant data in gl_state. */
123 # define SYNTAX_ENTRY_VIA_PROPERTY
124
125 # include "syntax.h"
126 # include "character.h"
127 # include "category.h"
128
129 # ifdef malloc
130 # undef malloc
131 # endif
132 # define malloc xmalloc
133 # ifdef realloc
134 # undef realloc
135 # endif
136 # define realloc xrealloc
137 # ifdef free
138 # undef free
139 # endif
140 # define free xfree
141
142 /* Converts the pointer to the char to BEG-based offset from the start. */
143 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
144 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
145
146 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
147 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
148 # define RE_STRING_CHAR(p, multibyte) \
149 (multibyte ? (STRING_CHAR (p)) : (*(p)))
150 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
151 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
152
153 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
154
155 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
156
157 /* Set C a (possibly converted to multibyte) character before P. P
158 points into a string which is the virtual concatenation of STR1
159 (which ends at END1) or STR2 (which ends at END2). */
160 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
161 do { \
162 if (target_multibyte) \
163 { \
164 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
165 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
166 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
167 c = STRING_CHAR (dtemp); \
168 } \
169 else \
170 { \
171 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
172 (c) = RE_CHAR_TO_MULTIBYTE (c); \
173 } \
174 } while (0)
175
176 /* Set C a (possibly converted to multibyte) character at P, and set
177 LEN to the byte length of that character. */
178 # define GET_CHAR_AFTER(c, p, len) \
179 do { \
180 if (target_multibyte) \
181 (c) = STRING_CHAR_AND_LENGTH (p, len); \
182 else \
183 { \
184 (c) = *p; \
185 len = 1; \
186 (c) = RE_CHAR_TO_MULTIBYTE (c); \
187 } \
188 } while (0)
189
190 #else /* not emacs */
191
192 /* If we are not linking with Emacs proper,
193 we can't use the relocating allocator
194 even if config.h says that we can. */
195 # undef REL_ALLOC
196
197 # include <unistd.h>
198
199 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
200
201 void *
202 xmalloc (size_t size)
203 {
204 register void *val;
205 val = (void *) malloc (size);
206 if (!val && size)
207 {
208 write (2, "virtual memory exhausted\n", 25);
209 exit (1);
210 }
211 return val;
212 }
213
214 void *
215 xrealloc (void *block, size_t size)
216 {
217 register void *val;
218 /* We must call malloc explicitly when BLOCK is 0, since some
219 reallocs don't do this. */
220 if (! block)
221 val = (void *) malloc (size);
222 else
223 val = (void *) realloc (block, size);
224 if (!val && size)
225 {
226 write (2, "virtual memory exhausted\n", 25);
227 exit (1);
228 }
229 return val;
230 }
231
232 # ifdef malloc
233 # undef malloc
234 # endif
235 # define malloc xmalloc
236 # ifdef realloc
237 # undef realloc
238 # endif
239 # define realloc xrealloc
240
241 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
242 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
243 # include <string.h>
244 # else
245 # include <strings.h>
246 # ifndef memcmp
247 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
248 # endif
249 # ifndef memcpy
250 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
251 # endif
252 # endif
253
254 /* Define the syntax stuff for \<, \>, etc. */
255
256 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
257 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
258
259 # define SWITCH_ENUM_CAST(x) (x)
260
261 /* Dummy macros for non-Emacs environments. */
262 # define CHAR_CHARSET(c) 0
263 # define CHARSET_LEADING_CODE_BASE(c) 0
264 # define MAX_MULTIBYTE_LENGTH 1
265 # define RE_MULTIBYTE_P(x) 0
266 # define RE_TARGET_MULTIBYTE_P(x) 0
267 # define WORD_BOUNDARY_P(c1, c2) (0)
268 # define CHAR_HEAD_P(p) (1)
269 # define SINGLE_BYTE_CHAR_P(c) (1)
270 # define SAME_CHARSET_P(c1, c2) (1)
271 # define BYTES_BY_CHAR_HEAD(p) (1)
272 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
273 # define STRING_CHAR(p) (*(p))
274 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
275 # define CHAR_STRING(c, s) (*(s) = (c), 1)
276 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
277 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
278 # define RE_CHAR_TO_MULTIBYTE(c) (c)
279 # define RE_CHAR_TO_UNIBYTE(c) (c)
280 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
281 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
282 # define GET_CHAR_AFTER(c, p, len) \
283 (c = *p, len = 1)
284 # define MAKE_CHAR(charset, c1, c2) (c1)
285 # define BYTE8_TO_CHAR(c) (c)
286 # define CHAR_BYTE8_P(c) (0)
287 # define CHAR_LEADING_CODE(c) (c)
288
289 #endif /* not emacs */
290
291 #ifndef RE_TRANSLATE
292 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
293 # define RE_TRANSLATE_P(TBL) (TBL)
294 #endif
295 \f
296 /* Get the interface, including the syntax bits. */
297 #include "regex.h"
298
299 /* isalpha etc. are used for the character classes. */
300 #include <ctype.h>
301
302 #ifdef emacs
303
304 /* 1 if C is an ASCII character. */
305 # define IS_REAL_ASCII(c) ((c) < 0200)
306
307 /* 1 if C is a unibyte character. */
308 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
309
310 /* The Emacs definitions should not be directly affected by locales. */
311
312 /* In Emacs, these are only used for single-byte characters. */
313 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
314 # define ISCNTRL(c) ((c) < ' ')
315 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
316 || ((c) >= 'a' && (c) <= 'f') \
317 || ((c) >= 'A' && (c) <= 'F'))
318
319 /* This is only used for single-byte characters. */
320 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
321
322 /* The rest must handle multibyte characters. */
323
324 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
325 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
326 : 1)
327
328 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
329 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
330 : 1)
331
332 # define ISALNUM(c) (IS_REAL_ASCII (c) \
333 ? (((c) >= 'a' && (c) <= 'z') \
334 || ((c) >= 'A' && (c) <= 'Z') \
335 || ((c) >= '0' && (c) <= '9')) \
336 : SYNTAX (c) == Sword)
337
338 # define ISALPHA(c) (IS_REAL_ASCII (c) \
339 ? (((c) >= 'a' && (c) <= 'z') \
340 || ((c) >= 'A' && (c) <= 'Z')) \
341 : SYNTAX (c) == Sword)
342
343 # define ISLOWER(c) lowercasep (c)
344
345 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
346 ? ((c) > ' ' && (c) < 0177 \
347 && !(((c) >= 'a' && (c) <= 'z') \
348 || ((c) >= 'A' && (c) <= 'Z') \
349 || ((c) >= '0' && (c) <= '9'))) \
350 : SYNTAX (c) != Sword)
351
352 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
353
354 # define ISUPPER(c) uppercasep (c)
355
356 # define ISWORD(c) (SYNTAX (c) == Sword)
357
358 #else /* not emacs */
359
360 /* Jim Meyering writes:
361
362 "... Some ctype macros are valid only for character codes that
363 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
364 using /bin/cc or gcc but without giving an ansi option). So, all
365 ctype uses should be through macros like ISPRINT... If
366 STDC_HEADERS is defined, then autoconf has verified that the ctype
367 macros don't need to be guarded with references to isascii. ...
368 Defining isascii to 1 should let any compiler worth its salt
369 eliminate the && through constant folding."
370 Solaris defines some of these symbols so we must undefine them first. */
371
372 # undef ISASCII
373 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
374 # define ISASCII(c) 1
375 # else
376 # define ISASCII(c) isascii(c)
377 # endif
378
379 /* 1 if C is an ASCII character. */
380 # define IS_REAL_ASCII(c) ((c) < 0200)
381
382 /* This distinction is not meaningful, except in Emacs. */
383 # define ISUNIBYTE(c) 1
384
385 # ifdef isblank
386 # define ISBLANK(c) (ISASCII (c) && isblank (c))
387 # else
388 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
389 # endif
390 # ifdef isgraph
391 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
392 # else
393 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
394 # endif
395
396 # undef ISPRINT
397 # define ISPRINT(c) (ISASCII (c) && isprint (c))
398 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
399 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
400 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
401 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
402 # define ISLOWER(c) (ISASCII (c) && islower (c))
403 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
404 # define ISSPACE(c) (ISASCII (c) && isspace (c))
405 # define ISUPPER(c) (ISASCII (c) && isupper (c))
406 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
407
408 # define ISWORD(c) ISALPHA(c)
409
410 # ifdef _tolower
411 # define TOLOWER(c) _tolower(c)
412 # else
413 # define TOLOWER(c) tolower(c)
414 # endif
415
416 /* How many characters in the character set. */
417 # define CHAR_SET_SIZE 256
418
419 # ifdef SYNTAX_TABLE
420
421 extern char *re_syntax_table;
422
423 # else /* not SYNTAX_TABLE */
424
425 static char re_syntax_table[CHAR_SET_SIZE];
426
427 static void
428 init_syntax_once (void)
429 {
430 register int c;
431 static int done = 0;
432
433 if (done)
434 return;
435
436 memset (re_syntax_table, 0, sizeof re_syntax_table);
437
438 for (c = 0; c < CHAR_SET_SIZE; ++c)
439 if (ISALNUM (c))
440 re_syntax_table[c] = Sword;
441
442 re_syntax_table['_'] = Ssymbol;
443
444 done = 1;
445 }
446
447 # endif /* not SYNTAX_TABLE */
448
449 # define SYNTAX(c) re_syntax_table[(c)]
450
451 #endif /* not emacs */
452 \f
453 #ifndef NULL
454 # define NULL (void *)0
455 #endif
456
457 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
458 since ours (we hope) works properly with all combinations of
459 machines, compilers, `char' and `unsigned char' argument types.
460 (Per Bothner suggested the basic approach.) */
461 #undef SIGN_EXTEND_CHAR
462 #if __STDC__
463 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
464 #else /* not __STDC__ */
465 /* As in Harbison and Steele. */
466 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
467 #endif
468 \f
469 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
470 use `alloca' instead of `malloc'. This is because using malloc in
471 re_search* or re_match* could cause memory leaks when C-g is used in
472 Emacs; also, malloc is slower and causes storage fragmentation. On
473 the other hand, malloc is more portable, and easier to debug.
474
475 Because we sometimes use alloca, some routines have to be macros,
476 not functions -- `alloca'-allocated space disappears at the end of the
477 function it is called in. */
478
479 #ifdef REGEX_MALLOC
480
481 # define REGEX_ALLOCATE malloc
482 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
483 # define REGEX_FREE free
484
485 #else /* not REGEX_MALLOC */
486
487 /* Emacs already defines alloca, sometimes. */
488 # ifndef alloca
489
490 /* Make alloca work the best possible way. */
491 # ifdef __GNUC__
492 # define alloca __builtin_alloca
493 # else /* not __GNUC__ */
494 # ifdef HAVE_ALLOCA_H
495 # include <alloca.h>
496 # endif /* HAVE_ALLOCA_H */
497 # endif /* not __GNUC__ */
498
499 # endif /* not alloca */
500
501 # define REGEX_ALLOCATE alloca
502
503 /* Assumes a `char *destination' variable. */
504 # define REGEX_REALLOCATE(source, osize, nsize) \
505 (destination = (char *) alloca (nsize), \
506 memcpy (destination, source, osize))
507
508 /* No need to do anything to free, after alloca. */
509 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
510
511 #endif /* not REGEX_MALLOC */
512
513 /* Define how to allocate the failure stack. */
514
515 #if defined REL_ALLOC && defined REGEX_MALLOC
516
517 # define REGEX_ALLOCATE_STACK(size) \
518 r_alloc (&failure_stack_ptr, (size))
519 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
520 r_re_alloc (&failure_stack_ptr, (nsize))
521 # define REGEX_FREE_STACK(ptr) \
522 r_alloc_free (&failure_stack_ptr)
523
524 #else /* not using relocating allocator */
525
526 # ifdef REGEX_MALLOC
527
528 # define REGEX_ALLOCATE_STACK malloc
529 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
530 # define REGEX_FREE_STACK free
531
532 # else /* not REGEX_MALLOC */
533
534 # define REGEX_ALLOCATE_STACK alloca
535
536 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
537 REGEX_REALLOCATE (source, osize, nsize)
538 /* No need to explicitly free anything. */
539 # define REGEX_FREE_STACK(arg) ((void)0)
540
541 # endif /* not REGEX_MALLOC */
542 #endif /* not using relocating allocator */
543
544
545 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
546 `string1' or just past its end. This works if PTR is NULL, which is
547 a good thing. */
548 #define FIRST_STRING_P(ptr) \
549 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
550
551 /* (Re)Allocate N items of type T using malloc, or fail. */
552 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
553 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
554 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
555
556 #define BYTEWIDTH 8 /* In bits. */
557
558 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
559
560 #undef MAX
561 #undef MIN
562 #define MAX(a, b) ((a) > (b) ? (a) : (b))
563 #define MIN(a, b) ((a) < (b) ? (a) : (b))
564
565 /* Type of source-pattern and string chars. */
566 typedef const unsigned char re_char;
567
568 typedef char boolean;
569 #define false 0
570 #define true 1
571
572 static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
573 re_char *string1, int size1,
574 re_char *string2, int size2,
575 int pos,
576 struct re_registers *regs,
577 int stop));
578 \f
579 /* These are the command codes that appear in compiled regular
580 expressions. Some opcodes are followed by argument bytes. A
581 command code can specify any interpretation whatsoever for its
582 arguments. Zero bytes may appear in the compiled regular expression. */
583
584 typedef enum
585 {
586 no_op = 0,
587
588 /* Succeed right away--no more backtracking. */
589 succeed,
590
591 /* Followed by one byte giving n, then by n literal bytes. */
592 exactn,
593
594 /* Matches any (more or less) character. */
595 anychar,
596
597 /* Matches any one char belonging to specified set. First
598 following byte is number of bitmap bytes. Then come bytes
599 for a bitmap saying which chars are in. Bits in each byte
600 are ordered low-bit-first. A character is in the set if its
601 bit is 1. A character too large to have a bit in the map is
602 automatically not in the set.
603
604 If the length byte has the 0x80 bit set, then that stuff
605 is followed by a range table:
606 2 bytes of flags for character sets (low 8 bits, high 8 bits)
607 See RANGE_TABLE_WORK_BITS below.
608 2 bytes, the number of pairs that follow (upto 32767)
609 pairs, each 2 multibyte characters,
610 each multibyte character represented as 3 bytes. */
611 charset,
612
613 /* Same parameters as charset, but match any character that is
614 not one of those specified. */
615 charset_not,
616
617 /* Start remembering the text that is matched, for storing in a
618 register. Followed by one byte with the register number, in
619 the range 0 to one less than the pattern buffer's re_nsub
620 field. */
621 start_memory,
622
623 /* Stop remembering the text that is matched and store it in a
624 memory register. Followed by one byte with the register
625 number, in the range 0 to one less than `re_nsub' in the
626 pattern buffer. */
627 stop_memory,
628
629 /* Match a duplicate of something remembered. Followed by one
630 byte containing the register number. */
631 duplicate,
632
633 /* Fail unless at beginning of line. */
634 begline,
635
636 /* Fail unless at end of line. */
637 endline,
638
639 /* Succeeds if at beginning of buffer (if emacs) or at beginning
640 of string to be matched (if not). */
641 begbuf,
642
643 /* Analogously, for end of buffer/string. */
644 endbuf,
645
646 /* Followed by two byte relative address to which to jump. */
647 jump,
648
649 /* Followed by two-byte relative address of place to resume at
650 in case of failure. */
651 on_failure_jump,
652
653 /* Like on_failure_jump, but pushes a placeholder instead of the
654 current string position when executed. */
655 on_failure_keep_string_jump,
656
657 /* Just like `on_failure_jump', except that it checks that we
658 don't get stuck in an infinite loop (matching an empty string
659 indefinitely). */
660 on_failure_jump_loop,
661
662 /* Just like `on_failure_jump_loop', except that it checks for
663 a different kind of loop (the kind that shows up with non-greedy
664 operators). This operation has to be immediately preceded
665 by a `no_op'. */
666 on_failure_jump_nastyloop,
667
668 /* A smart `on_failure_jump' used for greedy * and + operators.
669 It analyses the loop before which it is put and if the
670 loop does not require backtracking, it changes itself to
671 `on_failure_keep_string_jump' and short-circuits the loop,
672 else it just defaults to changing itself into `on_failure_jump'.
673 It assumes that it is pointing to just past a `jump'. */
674 on_failure_jump_smart,
675
676 /* Followed by two-byte relative address and two-byte number n.
677 After matching N times, jump to the address upon failure.
678 Does not work if N starts at 0: use on_failure_jump_loop
679 instead. */
680 succeed_n,
681
682 /* Followed by two-byte relative address, and two-byte number n.
683 Jump to the address N times, then fail. */
684 jump_n,
685
686 /* Set the following two-byte relative address to the
687 subsequent two-byte number. The address *includes* the two
688 bytes of number. */
689 set_number_at,
690
691 wordbeg, /* Succeeds if at word beginning. */
692 wordend, /* Succeeds if at word end. */
693
694 wordbound, /* Succeeds if at a word boundary. */
695 notwordbound, /* Succeeds if not at a word boundary. */
696
697 symbeg, /* Succeeds if at symbol beginning. */
698 symend, /* Succeeds if at symbol end. */
699
700 /* Matches any character whose syntax is specified. Followed by
701 a byte which contains a syntax code, e.g., Sword. */
702 syntaxspec,
703
704 /* Matches any character whose syntax is not that specified. */
705 notsyntaxspec
706
707 #ifdef emacs
708 ,before_dot, /* Succeeds if before point. */
709 at_dot, /* Succeeds if at point. */
710 after_dot, /* Succeeds if after point. */
711
712 /* Matches any character whose category-set contains the specified
713 category. The operator is followed by a byte which contains a
714 category code (mnemonic ASCII character). */
715 categoryspec,
716
717 /* Matches any character whose category-set does not contain the
718 specified category. The operator is followed by a byte which
719 contains the category code (mnemonic ASCII character). */
720 notcategoryspec
721 #endif /* emacs */
722 } re_opcode_t;
723 \f
724 /* Common operations on the compiled pattern. */
725
726 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
727
728 #define STORE_NUMBER(destination, number) \
729 do { \
730 (destination)[0] = (number) & 0377; \
731 (destination)[1] = (number) >> 8; \
732 } while (0)
733
734 /* Same as STORE_NUMBER, except increment DESTINATION to
735 the byte after where the number is stored. Therefore, DESTINATION
736 must be an lvalue. */
737
738 #define STORE_NUMBER_AND_INCR(destination, number) \
739 do { \
740 STORE_NUMBER (destination, number); \
741 (destination) += 2; \
742 } while (0)
743
744 /* Put into DESTINATION a number stored in two contiguous bytes starting
745 at SOURCE. */
746
747 #define EXTRACT_NUMBER(destination, source) \
748 do { \
749 (destination) = *(source) & 0377; \
750 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
751 } while (0)
752
753 #ifdef DEBUG
754 static void extract_number _RE_ARGS ((int *dest, re_char *source));
755 static void
756 extract_number (dest, source)
757 int *dest;
758 re_char *source;
759 {
760 int temp = SIGN_EXTEND_CHAR (*(source + 1));
761 *dest = *source & 0377;
762 *dest += temp << 8;
763 }
764
765 # ifndef EXTRACT_MACROS /* To debug the macros. */
766 # undef EXTRACT_NUMBER
767 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
768 # endif /* not EXTRACT_MACROS */
769
770 #endif /* DEBUG */
771
772 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
773 SOURCE must be an lvalue. */
774
775 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
776 do { \
777 EXTRACT_NUMBER (destination, source); \
778 (source) += 2; \
779 } while (0)
780
781 #ifdef DEBUG
782 static void extract_number_and_incr _RE_ARGS ((int *destination,
783 re_char **source));
784 static void
785 extract_number_and_incr (destination, source)
786 int *destination;
787 re_char **source;
788 {
789 extract_number (destination, *source);
790 *source += 2;
791 }
792
793 # ifndef EXTRACT_MACROS
794 # undef EXTRACT_NUMBER_AND_INCR
795 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
796 extract_number_and_incr (&dest, &src)
797 # endif /* not EXTRACT_MACROS */
798
799 #endif /* DEBUG */
800 \f
801 /* Store a multibyte character in three contiguous bytes starting
802 DESTINATION, and increment DESTINATION to the byte after where the
803 character is stored. Therefore, DESTINATION must be an lvalue. */
804
805 #define STORE_CHARACTER_AND_INCR(destination, character) \
806 do { \
807 (destination)[0] = (character) & 0377; \
808 (destination)[1] = ((character) >> 8) & 0377; \
809 (destination)[2] = (character) >> 16; \
810 (destination) += 3; \
811 } while (0)
812
813 /* Put into DESTINATION a character stored in three contiguous bytes
814 starting at SOURCE. */
815
816 #define EXTRACT_CHARACTER(destination, source) \
817 do { \
818 (destination) = ((source)[0] \
819 | ((source)[1] << 8) \
820 | ((source)[2] << 16)); \
821 } while (0)
822
823
824 /* Macros for charset. */
825
826 /* Size of bitmap of charset P in bytes. P is a start of charset,
827 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
828 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
829
830 /* Nonzero if charset P has range table. */
831 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
832
833 /* Return the address of range table of charset P. But not the start
834 of table itself, but the before where the number of ranges is
835 stored. `2 +' means to skip re_opcode_t and size of bitmap,
836 and the 2 bytes of flags at the start of the range table. */
837 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
838
839 /* Extract the bit flags that start a range table. */
840 #define CHARSET_RANGE_TABLE_BITS(p) \
841 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
842 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
843
844 /* Return the address of end of RANGE_TABLE. COUNT is number of
845 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
846 is start of range and end of range. `* 3' is size of each start
847 and end. */
848 #define CHARSET_RANGE_TABLE_END(range_table, count) \
849 ((range_table) + (count) * 2 * 3)
850
851 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
852 COUNT is number of ranges in RANGE_TABLE. */
853 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
854 do \
855 { \
856 re_wchar_t range_start, range_end; \
857 re_char *rtp; \
858 re_char *range_table_end \
859 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
860 \
861 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
862 { \
863 EXTRACT_CHARACTER (range_start, rtp); \
864 EXTRACT_CHARACTER (range_end, rtp + 3); \
865 \
866 if (range_start <= (c) && (c) <= range_end) \
867 { \
868 (not) = !(not); \
869 break; \
870 } \
871 } \
872 } \
873 while (0)
874
875 /* Test if C is in range table of CHARSET. The flag NOT is negated if
876 C is listed in it. */
877 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
878 do \
879 { \
880 /* Number of ranges in range table. */ \
881 int count; \
882 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
883 \
884 EXTRACT_NUMBER_AND_INCR (count, range_table); \
885 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
886 } \
887 while (0)
888 \f
889 /* If DEBUG is defined, Regex prints many voluminous messages about what
890 it is doing (if the variable `debug' is nonzero). If linked with the
891 main program in `iregex.c', you can enter patterns and strings
892 interactively. And if linked with the main program in `main.c' and
893 the other test files, you can run the already-written tests. */
894
895 #ifdef DEBUG
896
897 /* We use standard I/O for debugging. */
898 # include <stdio.h>
899
900 /* It is useful to test things that ``must'' be true when debugging. */
901 # include <assert.h>
902
903 static int debug = -100000;
904
905 # define DEBUG_STATEMENT(e) e
906 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
907 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
908 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
909 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
910 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
911 if (debug > 0) print_partial_compiled_pattern (s, e)
912 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
913 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
914
915
916 /* Print the fastmap in human-readable form. */
917
918 void
919 print_fastmap (fastmap)
920 char *fastmap;
921 {
922 unsigned was_a_range = 0;
923 unsigned i = 0;
924
925 while (i < (1 << BYTEWIDTH))
926 {
927 if (fastmap[i++])
928 {
929 was_a_range = 0;
930 putchar (i - 1);
931 while (i < (1 << BYTEWIDTH) && fastmap[i])
932 {
933 was_a_range = 1;
934 i++;
935 }
936 if (was_a_range)
937 {
938 printf ("-");
939 putchar (i - 1);
940 }
941 }
942 }
943 putchar ('\n');
944 }
945
946
947 /* Print a compiled pattern string in human-readable form, starting at
948 the START pointer into it and ending just before the pointer END. */
949
950 void
951 print_partial_compiled_pattern (start, end)
952 re_char *start;
953 re_char *end;
954 {
955 int mcnt, mcnt2;
956 re_char *p = start;
957 re_char *pend = end;
958
959 if (start == NULL)
960 {
961 fprintf (stderr, "(null)\n");
962 return;
963 }
964
965 /* Loop over pattern commands. */
966 while (p < pend)
967 {
968 fprintf (stderr, "%d:\t", p - start);
969
970 switch ((re_opcode_t) *p++)
971 {
972 case no_op:
973 fprintf (stderr, "/no_op");
974 break;
975
976 case succeed:
977 fprintf (stderr, "/succeed");
978 break;
979
980 case exactn:
981 mcnt = *p++;
982 fprintf (stderr, "/exactn/%d", mcnt);
983 do
984 {
985 fprintf (stderr, "/%c", *p++);
986 }
987 while (--mcnt);
988 break;
989
990 case start_memory:
991 fprintf (stderr, "/start_memory/%d", *p++);
992 break;
993
994 case stop_memory:
995 fprintf (stderr, "/stop_memory/%d", *p++);
996 break;
997
998 case duplicate:
999 fprintf (stderr, "/duplicate/%d", *p++);
1000 break;
1001
1002 case anychar:
1003 fprintf (stderr, "/anychar");
1004 break;
1005
1006 case charset:
1007 case charset_not:
1008 {
1009 register int c, last = -100;
1010 register int in_range = 0;
1011 int length = CHARSET_BITMAP_SIZE (p - 1);
1012 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
1013
1014 fprintf (stderr, "/charset [%s",
1015 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
1016
1017 if (p + *p >= pend)
1018 fprintf (stderr, " !extends past end of pattern! ");
1019
1020 for (c = 0; c < 256; c++)
1021 if (c / 8 < length
1022 && (p[1 + (c/8)] & (1 << (c % 8))))
1023 {
1024 /* Are we starting a range? */
1025 if (last + 1 == c && ! in_range)
1026 {
1027 fprintf (stderr, "-");
1028 in_range = 1;
1029 }
1030 /* Have we broken a range? */
1031 else if (last + 1 != c && in_range)
1032 {
1033 fprintf (stderr, "%c", last);
1034 in_range = 0;
1035 }
1036
1037 if (! in_range)
1038 fprintf (stderr, "%c", c);
1039
1040 last = c;
1041 }
1042
1043 if (in_range)
1044 fprintf (stderr, "%c", last);
1045
1046 fprintf (stderr, "]");
1047
1048 p += 1 + length;
1049
1050 if (has_range_table)
1051 {
1052 int count;
1053 fprintf (stderr, "has-range-table");
1054
1055 /* ??? Should print the range table; for now, just skip it. */
1056 p += 2; /* skip range table bits */
1057 EXTRACT_NUMBER_AND_INCR (count, p);
1058 p = CHARSET_RANGE_TABLE_END (p, count);
1059 }
1060 }
1061 break;
1062
1063 case begline:
1064 fprintf (stderr, "/begline");
1065 break;
1066
1067 case endline:
1068 fprintf (stderr, "/endline");
1069 break;
1070
1071 case on_failure_jump:
1072 extract_number_and_incr (&mcnt, &p);
1073 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1074 break;
1075
1076 case on_failure_keep_string_jump:
1077 extract_number_and_incr (&mcnt, &p);
1078 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1079 break;
1080
1081 case on_failure_jump_nastyloop:
1082 extract_number_and_incr (&mcnt, &p);
1083 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1084 break;
1085
1086 case on_failure_jump_loop:
1087 extract_number_and_incr (&mcnt, &p);
1088 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1089 break;
1090
1091 case on_failure_jump_smart:
1092 extract_number_and_incr (&mcnt, &p);
1093 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1094 break;
1095
1096 case jump:
1097 extract_number_and_incr (&mcnt, &p);
1098 fprintf (stderr, "/jump to %d", p + mcnt - start);
1099 break;
1100
1101 case succeed_n:
1102 extract_number_and_incr (&mcnt, &p);
1103 extract_number_and_incr (&mcnt2, &p);
1104 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1105 break;
1106
1107 case jump_n:
1108 extract_number_and_incr (&mcnt, &p);
1109 extract_number_and_incr (&mcnt2, &p);
1110 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1111 break;
1112
1113 case set_number_at:
1114 extract_number_and_incr (&mcnt, &p);
1115 extract_number_and_incr (&mcnt2, &p);
1116 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1117 break;
1118
1119 case wordbound:
1120 fprintf (stderr, "/wordbound");
1121 break;
1122
1123 case notwordbound:
1124 fprintf (stderr, "/notwordbound");
1125 break;
1126
1127 case wordbeg:
1128 fprintf (stderr, "/wordbeg");
1129 break;
1130
1131 case wordend:
1132 fprintf (stderr, "/wordend");
1133 break;
1134
1135 case symbeg:
1136 fprintf (stderr, "/symbeg");
1137 break;
1138
1139 case symend:
1140 fprintf (stderr, "/symend");
1141 break;
1142
1143 case syntaxspec:
1144 fprintf (stderr, "/syntaxspec");
1145 mcnt = *p++;
1146 fprintf (stderr, "/%d", mcnt);
1147 break;
1148
1149 case notsyntaxspec:
1150 fprintf (stderr, "/notsyntaxspec");
1151 mcnt = *p++;
1152 fprintf (stderr, "/%d", mcnt);
1153 break;
1154
1155 # ifdef emacs
1156 case before_dot:
1157 fprintf (stderr, "/before_dot");
1158 break;
1159
1160 case at_dot:
1161 fprintf (stderr, "/at_dot");
1162 break;
1163
1164 case after_dot:
1165 fprintf (stderr, "/after_dot");
1166 break;
1167
1168 case categoryspec:
1169 fprintf (stderr, "/categoryspec");
1170 mcnt = *p++;
1171 fprintf (stderr, "/%d", mcnt);
1172 break;
1173
1174 case notcategoryspec:
1175 fprintf (stderr, "/notcategoryspec");
1176 mcnt = *p++;
1177 fprintf (stderr, "/%d", mcnt);
1178 break;
1179 # endif /* emacs */
1180
1181 case begbuf:
1182 fprintf (stderr, "/begbuf");
1183 break;
1184
1185 case endbuf:
1186 fprintf (stderr, "/endbuf");
1187 break;
1188
1189 default:
1190 fprintf (stderr, "?%d", *(p-1));
1191 }
1192
1193 fprintf (stderr, "\n");
1194 }
1195
1196 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1197 }
1198
1199
1200 void
1201 print_compiled_pattern (bufp)
1202 struct re_pattern_buffer *bufp;
1203 {
1204 re_char *buffer = bufp->buffer;
1205
1206 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1207 printf ("%ld bytes used/%ld bytes allocated.\n",
1208 bufp->used, bufp->allocated);
1209
1210 if (bufp->fastmap_accurate && bufp->fastmap)
1211 {
1212 printf ("fastmap: ");
1213 print_fastmap (bufp->fastmap);
1214 }
1215
1216 printf ("re_nsub: %d\t", bufp->re_nsub);
1217 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1218 printf ("can_be_null: %d\t", bufp->can_be_null);
1219 printf ("no_sub: %d\t", bufp->no_sub);
1220 printf ("not_bol: %d\t", bufp->not_bol);
1221 printf ("not_eol: %d\t", bufp->not_eol);
1222 printf ("syntax: %lx\n", bufp->syntax);
1223 fflush (stdout);
1224 /* Perhaps we should print the translate table? */
1225 }
1226
1227
1228 void
1229 print_double_string (where, string1, size1, string2, size2)
1230 re_char *where;
1231 re_char *string1;
1232 re_char *string2;
1233 int size1;
1234 int size2;
1235 {
1236 int this_char;
1237
1238 if (where == NULL)
1239 printf ("(null)");
1240 else
1241 {
1242 if (FIRST_STRING_P (where))
1243 {
1244 for (this_char = where - string1; this_char < size1; this_char++)
1245 putchar (string1[this_char]);
1246
1247 where = string2;
1248 }
1249
1250 for (this_char = where - string2; this_char < size2; this_char++)
1251 putchar (string2[this_char]);
1252 }
1253 }
1254
1255 #else /* not DEBUG */
1256
1257 # undef assert
1258 # define assert(e)
1259
1260 # define DEBUG_STATEMENT(e)
1261 # define DEBUG_PRINT1(x)
1262 # define DEBUG_PRINT2(x1, x2)
1263 # define DEBUG_PRINT3(x1, x2, x3)
1264 # define DEBUG_PRINT4(x1, x2, x3, x4)
1265 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1266 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1267
1268 #endif /* not DEBUG */
1269 \f
1270 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1271 also be assigned to arbitrarily: each pattern buffer stores its own
1272 syntax, so it can be changed between regex compilations. */
1273 /* This has no initializer because initialized variables in Emacs
1274 become read-only after dumping. */
1275 reg_syntax_t re_syntax_options;
1276
1277
1278 /* Specify the precise syntax of regexps for compilation. This provides
1279 for compatibility for various utilities which historically have
1280 different, incompatible syntaxes.
1281
1282 The argument SYNTAX is a bit mask comprised of the various bits
1283 defined in regex.h. We return the old syntax. */
1284
1285 reg_syntax_t
1286 re_set_syntax (reg_syntax_t syntax)
1287 {
1288 reg_syntax_t ret = re_syntax_options;
1289
1290 re_syntax_options = syntax;
1291 return ret;
1292 }
1293 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1294
1295 /* Regexp to use to replace spaces, or NULL meaning don't. */
1296 static re_char *whitespace_regexp;
1297
1298 void
1299 re_set_whitespace_regexp (const char *regexp)
1300 {
1301 whitespace_regexp = (re_char *) regexp;
1302 }
1303 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1304 \f
1305 /* This table gives an error message for each of the error codes listed
1306 in regex.h. Obviously the order here has to be same as there.
1307 POSIX doesn't require that we do anything for REG_NOERROR,
1308 but why not be nice? */
1309
1310 static const char *re_error_msgid[] =
1311 {
1312 gettext_noop ("Success"), /* REG_NOERROR */
1313 gettext_noop ("No match"), /* REG_NOMATCH */
1314 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1315 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1316 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1317 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1318 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1319 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1320 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1321 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1322 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1323 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1324 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1325 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1326 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1327 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1328 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1329 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1330 };
1331 \f
1332 /* Avoiding alloca during matching, to placate r_alloc. */
1333
1334 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1335 searching and matching functions should not call alloca. On some
1336 systems, alloca is implemented in terms of malloc, and if we're
1337 using the relocating allocator routines, then malloc could cause a
1338 relocation, which might (if the strings being searched are in the
1339 ralloc heap) shift the data out from underneath the regexp
1340 routines.
1341
1342 Here's another reason to avoid allocation: Emacs
1343 processes input from X in a signal handler; processing X input may
1344 call malloc; if input arrives while a matching routine is calling
1345 malloc, then we're scrod. But Emacs can't just block input while
1346 calling matching routines; then we don't notice interrupts when
1347 they come in. So, Emacs blocks input around all regexp calls
1348 except the matching calls, which it leaves unprotected, in the
1349 faith that they will not malloc. */
1350
1351 /* Normally, this is fine. */
1352 #define MATCH_MAY_ALLOCATE
1353
1354 /* The match routines may not allocate if (1) they would do it with malloc
1355 and (2) it's not safe for them to use malloc.
1356 Note that if REL_ALLOC is defined, matching would not use malloc for the
1357 failure stack, but we would still use it for the register vectors;
1358 so REL_ALLOC should not affect this. */
1359 #if defined REGEX_MALLOC && defined emacs
1360 # undef MATCH_MAY_ALLOCATE
1361 #endif
1362
1363 \f
1364 /* Failure stack declarations and macros; both re_compile_fastmap and
1365 re_match_2 use a failure stack. These have to be macros because of
1366 REGEX_ALLOCATE_STACK. */
1367
1368
1369 /* Approximate number of failure points for which to initially allocate space
1370 when matching. If this number is exceeded, we allocate more
1371 space, so it is not a hard limit. */
1372 #ifndef INIT_FAILURE_ALLOC
1373 # define INIT_FAILURE_ALLOC 20
1374 #endif
1375
1376 /* Roughly the maximum number of failure points on the stack. Would be
1377 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1378 This is a variable only so users of regex can assign to it; we never
1379 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1380 before using it, so it should probably be a byte-count instead. */
1381 # if defined MATCH_MAY_ALLOCATE
1382 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1383 whose default stack limit is 2mb. In order for a larger
1384 value to work reliably, you have to try to make it accord
1385 with the process stack limit. */
1386 size_t re_max_failures = 40000;
1387 # else
1388 size_t re_max_failures = 4000;
1389 # endif
1390
1391 union fail_stack_elt
1392 {
1393 re_char *pointer;
1394 /* This should be the biggest `int' that's no bigger than a pointer. */
1395 long integer;
1396 };
1397
1398 typedef union fail_stack_elt fail_stack_elt_t;
1399
1400 typedef struct
1401 {
1402 fail_stack_elt_t *stack;
1403 size_t size;
1404 size_t avail; /* Offset of next open position. */
1405 size_t frame; /* Offset of the cur constructed frame. */
1406 } fail_stack_type;
1407
1408 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1409
1410
1411 /* Define macros to initialize and free the failure stack.
1412 Do `return -2' if the alloc fails. */
1413
1414 #ifdef MATCH_MAY_ALLOCATE
1415 # define INIT_FAIL_STACK() \
1416 do { \
1417 fail_stack.stack = (fail_stack_elt_t *) \
1418 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1419 * sizeof (fail_stack_elt_t)); \
1420 \
1421 if (fail_stack.stack == NULL) \
1422 return -2; \
1423 \
1424 fail_stack.size = INIT_FAILURE_ALLOC; \
1425 fail_stack.avail = 0; \
1426 fail_stack.frame = 0; \
1427 } while (0)
1428 #else
1429 # define INIT_FAIL_STACK() \
1430 do { \
1431 fail_stack.avail = 0; \
1432 fail_stack.frame = 0; \
1433 } while (0)
1434
1435 # define RETALLOC_IF(addr, n, t) \
1436 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1437 #endif
1438
1439
1440 /* Double the size of FAIL_STACK, up to a limit
1441 which allows approximately `re_max_failures' items.
1442
1443 Return 1 if succeeds, and 0 if either ran out of memory
1444 allocating space for it or it was already too large.
1445
1446 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1447
1448 /* Factor to increase the failure stack size by
1449 when we increase it.
1450 This used to be 2, but 2 was too wasteful
1451 because the old discarded stacks added up to as much space
1452 were as ultimate, maximum-size stack. */
1453 #define FAIL_STACK_GROWTH_FACTOR 4
1454
1455 #define GROW_FAIL_STACK(fail_stack) \
1456 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1457 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1458 ? 0 \
1459 : ((fail_stack).stack \
1460 = (fail_stack_elt_t *) \
1461 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1462 (fail_stack).size * sizeof (fail_stack_elt_t), \
1463 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1464 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1465 * FAIL_STACK_GROWTH_FACTOR))), \
1466 \
1467 (fail_stack).stack == NULL \
1468 ? 0 \
1469 : ((fail_stack).size \
1470 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1471 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1472 * FAIL_STACK_GROWTH_FACTOR)) \
1473 / sizeof (fail_stack_elt_t)), \
1474 1)))
1475
1476
1477 /* Push a pointer value onto the failure stack.
1478 Assumes the variable `fail_stack'. Probably should only
1479 be called from within `PUSH_FAILURE_POINT'. */
1480 #define PUSH_FAILURE_POINTER(item) \
1481 fail_stack.stack[fail_stack.avail++].pointer = (item)
1482
1483 /* This pushes an integer-valued item onto the failure stack.
1484 Assumes the variable `fail_stack'. Probably should only
1485 be called from within `PUSH_FAILURE_POINT'. */
1486 #define PUSH_FAILURE_INT(item) \
1487 fail_stack.stack[fail_stack.avail++].integer = (item)
1488
1489 /* These POP... operations complement the PUSH... operations.
1490 All assume that `fail_stack' is nonempty. */
1491 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1492 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1493
1494 /* Individual items aside from the registers. */
1495 #define NUM_NONREG_ITEMS 3
1496
1497 /* Used to examine the stack (to detect infinite loops). */
1498 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1499 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1500 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1501 #define TOP_FAILURE_HANDLE() fail_stack.frame
1502
1503
1504 #define ENSURE_FAIL_STACK(space) \
1505 while (REMAINING_AVAIL_SLOTS <= space) { \
1506 if (!GROW_FAIL_STACK (fail_stack)) \
1507 return -2; \
1508 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1509 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1510 }
1511
1512 /* Push register NUM onto the stack. */
1513 #define PUSH_FAILURE_REG(num) \
1514 do { \
1515 char *destination; \
1516 ENSURE_FAIL_STACK(3); \
1517 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1518 num, regstart[num], regend[num]); \
1519 PUSH_FAILURE_POINTER (regstart[num]); \
1520 PUSH_FAILURE_POINTER (regend[num]); \
1521 PUSH_FAILURE_INT (num); \
1522 } while (0)
1523
1524 /* Change the counter's value to VAL, but make sure that it will
1525 be reset when backtracking. */
1526 #define PUSH_NUMBER(ptr,val) \
1527 do { \
1528 char *destination; \
1529 int c; \
1530 ENSURE_FAIL_STACK(3); \
1531 EXTRACT_NUMBER (c, ptr); \
1532 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1533 PUSH_FAILURE_INT (c); \
1534 PUSH_FAILURE_POINTER (ptr); \
1535 PUSH_FAILURE_INT (-1); \
1536 STORE_NUMBER (ptr, val); \
1537 } while (0)
1538
1539 /* Pop a saved register off the stack. */
1540 #define POP_FAILURE_REG_OR_COUNT() \
1541 do { \
1542 int pfreg = POP_FAILURE_INT (); \
1543 if (pfreg == -1) \
1544 { \
1545 /* It's a counter. */ \
1546 /* Here, we discard `const', making re_match non-reentrant. */ \
1547 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1548 pfreg = POP_FAILURE_INT (); \
1549 STORE_NUMBER (ptr, pfreg); \
1550 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, pfreg); \
1551 } \
1552 else \
1553 { \
1554 regend[pfreg] = POP_FAILURE_POINTER (); \
1555 regstart[pfreg] = POP_FAILURE_POINTER (); \
1556 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1557 pfreg, regstart[pfreg], regend[pfreg]); \
1558 } \
1559 } while (0)
1560
1561 /* Check that we are not stuck in an infinite loop. */
1562 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1563 do { \
1564 int failure = TOP_FAILURE_HANDLE (); \
1565 /* Check for infinite matching loops */ \
1566 while (failure > 0 \
1567 && (FAILURE_STR (failure) == string_place \
1568 || FAILURE_STR (failure) == NULL)) \
1569 { \
1570 assert (FAILURE_PAT (failure) >= bufp->buffer \
1571 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1572 if (FAILURE_PAT (failure) == pat_cur) \
1573 { \
1574 cycle = 1; \
1575 break; \
1576 } \
1577 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1578 failure = NEXT_FAILURE_HANDLE(failure); \
1579 } \
1580 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1581 } while (0)
1582
1583 /* Push the information about the state we will need
1584 if we ever fail back to it.
1585
1586 Requires variables fail_stack, regstart, regend and
1587 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1588 declared.
1589
1590 Does `return FAILURE_CODE' if runs out of memory. */
1591
1592 #define PUSH_FAILURE_POINT(pattern, string_place) \
1593 do { \
1594 char *destination; \
1595 /* Must be int, so when we don't save any registers, the arithmetic \
1596 of 0 + -1 isn't done as unsigned. */ \
1597 \
1598 DEBUG_STATEMENT (nfailure_points_pushed++); \
1599 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1600 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1601 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1602 \
1603 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1604 \
1605 DEBUG_PRINT1 ("\n"); \
1606 \
1607 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1608 PUSH_FAILURE_INT (fail_stack.frame); \
1609 \
1610 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1611 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1612 DEBUG_PRINT1 ("'\n"); \
1613 PUSH_FAILURE_POINTER (string_place); \
1614 \
1615 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1616 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1617 PUSH_FAILURE_POINTER (pattern); \
1618 \
1619 /* Close the frame by moving the frame pointer past it. */ \
1620 fail_stack.frame = fail_stack.avail; \
1621 } while (0)
1622
1623 /* Estimate the size of data pushed by a typical failure stack entry.
1624 An estimate is all we need, because all we use this for
1625 is to choose a limit for how big to make the failure stack. */
1626 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1627 #define TYPICAL_FAILURE_SIZE 20
1628
1629 /* How many items can still be added to the stack without overflowing it. */
1630 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1631
1632
1633 /* Pops what PUSH_FAIL_STACK pushes.
1634
1635 We restore into the parameters, all of which should be lvalues:
1636 STR -- the saved data position.
1637 PAT -- the saved pattern position.
1638 REGSTART, REGEND -- arrays of string positions.
1639
1640 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1641 `pend', `string1', `size1', `string2', and `size2'. */
1642
1643 #define POP_FAILURE_POINT(str, pat) \
1644 do { \
1645 assert (!FAIL_STACK_EMPTY ()); \
1646 \
1647 /* Remove failure points and point to how many regs pushed. */ \
1648 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1649 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1650 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1651 \
1652 /* Pop the saved registers. */ \
1653 while (fail_stack.frame < fail_stack.avail) \
1654 POP_FAILURE_REG_OR_COUNT (); \
1655 \
1656 pat = POP_FAILURE_POINTER (); \
1657 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1658 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1659 \
1660 /* If the saved string location is NULL, it came from an \
1661 on_failure_keep_string_jump opcode, and we want to throw away the \
1662 saved NULL, thus retaining our current position in the string. */ \
1663 str = POP_FAILURE_POINTER (); \
1664 DEBUG_PRINT2 (" Popping string %p: `", str); \
1665 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1666 DEBUG_PRINT1 ("'\n"); \
1667 \
1668 fail_stack.frame = POP_FAILURE_INT (); \
1669 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1670 \
1671 assert (fail_stack.avail >= 0); \
1672 assert (fail_stack.frame <= fail_stack.avail); \
1673 \
1674 DEBUG_STATEMENT (nfailure_points_popped++); \
1675 } while (0) /* POP_FAILURE_POINT */
1676
1677
1678 \f
1679 /* Registers are set to a sentinel when they haven't yet matched. */
1680 #define REG_UNSET(e) ((e) == NULL)
1681 \f
1682 /* Subroutine declarations and macros for regex_compile. */
1683
1684 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1685 reg_syntax_t syntax,
1686 struct re_pattern_buffer *bufp));
1687 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1688 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1689 int arg1, int arg2));
1690 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1691 int arg, unsigned char *end));
1692 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1693 int arg1, int arg2, unsigned char *end));
1694 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1695 re_char *p,
1696 reg_syntax_t syntax));
1697 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1698 re_char *pend,
1699 reg_syntax_t syntax));
1700 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1701 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1702 char *fastmap, const int multibyte));
1703
1704 /* Fetch the next character in the uncompiled pattern, with no
1705 translation. */
1706 #define PATFETCH(c) \
1707 do { \
1708 int len; \
1709 if (p == pend) return REG_EEND; \
1710 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1711 p += len; \
1712 } while (0)
1713
1714
1715 /* If `translate' is non-null, return translate[D], else just D. We
1716 cast the subscript to translate because some data is declared as
1717 `char *', to avoid warnings when a string constant is passed. But
1718 when we use a character as a subscript we must make it unsigned. */
1719 #ifndef TRANSLATE
1720 # define TRANSLATE(d) \
1721 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1722 #endif
1723
1724
1725 /* Macros for outputting the compiled pattern into `buffer'. */
1726
1727 /* If the buffer isn't allocated when it comes in, use this. */
1728 #define INIT_BUF_SIZE 32
1729
1730 /* Make sure we have at least N more bytes of space in buffer. */
1731 #define GET_BUFFER_SPACE(n) \
1732 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1733 EXTEND_BUFFER ()
1734
1735 /* Make sure we have one more byte of buffer space and then add C to it. */
1736 #define BUF_PUSH(c) \
1737 do { \
1738 GET_BUFFER_SPACE (1); \
1739 *b++ = (unsigned char) (c); \
1740 } while (0)
1741
1742
1743 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1744 #define BUF_PUSH_2(c1, c2) \
1745 do { \
1746 GET_BUFFER_SPACE (2); \
1747 *b++ = (unsigned char) (c1); \
1748 *b++ = (unsigned char) (c2); \
1749 } while (0)
1750
1751
1752 /* Store a jump with opcode OP at LOC to location TO. We store a
1753 relative address offset by the three bytes the jump itself occupies. */
1754 #define STORE_JUMP(op, loc, to) \
1755 store_op1 (op, loc, (to) - (loc) - 3)
1756
1757 /* Likewise, for a two-argument jump. */
1758 #define STORE_JUMP2(op, loc, to, arg) \
1759 store_op2 (op, loc, (to) - (loc) - 3, arg)
1760
1761 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1762 #define INSERT_JUMP(op, loc, to) \
1763 insert_op1 (op, loc, (to) - (loc) - 3, b)
1764
1765 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1766 #define INSERT_JUMP2(op, loc, to, arg) \
1767 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1768
1769
1770 /* This is not an arbitrary limit: the arguments which represent offsets
1771 into the pattern are two bytes long. So if 2^15 bytes turns out to
1772 be too small, many things would have to change. */
1773 # define MAX_BUF_SIZE (1L << 15)
1774
1775 #if 0 /* This is when we thought it could be 2^16 bytes. */
1776 /* Any other compiler which, like MSC, has allocation limit below 2^16
1777 bytes will have to use approach similar to what was done below for
1778 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1779 reallocating to 0 bytes. Such thing is not going to work too well.
1780 You have been warned!! */
1781 #if defined _MSC_VER && !defined WIN32
1782 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1783 # define MAX_BUF_SIZE 65500L
1784 #else
1785 # define MAX_BUF_SIZE (1L << 16)
1786 #endif
1787 #endif /* 0 */
1788
1789 /* Extend the buffer by twice its current size via realloc and
1790 reset the pointers that pointed into the old block to point to the
1791 correct places in the new one. If extending the buffer results in it
1792 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1793 #if __BOUNDED_POINTERS__
1794 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1795 # define MOVE_BUFFER_POINTER(P) \
1796 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1797 SET_HIGH_BOUND (P), \
1798 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1799 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1800 else \
1801 { \
1802 SET_HIGH_BOUND (b); \
1803 SET_HIGH_BOUND (begalt); \
1804 if (fixup_alt_jump) \
1805 SET_HIGH_BOUND (fixup_alt_jump); \
1806 if (laststart) \
1807 SET_HIGH_BOUND (laststart); \
1808 if (pending_exact) \
1809 SET_HIGH_BOUND (pending_exact); \
1810 }
1811 #else
1812 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1813 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1814 #endif
1815 #define EXTEND_BUFFER() \
1816 do { \
1817 unsigned char *old_buffer = bufp->buffer; \
1818 if (bufp->allocated == MAX_BUF_SIZE) \
1819 return REG_ESIZE; \
1820 bufp->allocated <<= 1; \
1821 if (bufp->allocated > MAX_BUF_SIZE) \
1822 bufp->allocated = MAX_BUF_SIZE; \
1823 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1824 if (bufp->buffer == NULL) \
1825 return REG_ESPACE; \
1826 /* If the buffer moved, move all the pointers into it. */ \
1827 if (old_buffer != bufp->buffer) \
1828 { \
1829 unsigned char *new_buffer = bufp->buffer; \
1830 MOVE_BUFFER_POINTER (b); \
1831 MOVE_BUFFER_POINTER (begalt); \
1832 if (fixup_alt_jump) \
1833 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1834 if (laststart) \
1835 MOVE_BUFFER_POINTER (laststart); \
1836 if (pending_exact) \
1837 MOVE_BUFFER_POINTER (pending_exact); \
1838 } \
1839 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1840 } while (0)
1841
1842
1843 /* Since we have one byte reserved for the register number argument to
1844 {start,stop}_memory, the maximum number of groups we can report
1845 things about is what fits in that byte. */
1846 #define MAX_REGNUM 255
1847
1848 /* But patterns can have more than `MAX_REGNUM' registers. We just
1849 ignore the excess. */
1850 typedef int regnum_t;
1851
1852
1853 /* Macros for the compile stack. */
1854
1855 /* Since offsets can go either forwards or backwards, this type needs to
1856 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1857 /* int may be not enough when sizeof(int) == 2. */
1858 typedef long pattern_offset_t;
1859
1860 typedef struct
1861 {
1862 pattern_offset_t begalt_offset;
1863 pattern_offset_t fixup_alt_jump;
1864 pattern_offset_t laststart_offset;
1865 regnum_t regnum;
1866 } compile_stack_elt_t;
1867
1868
1869 typedef struct
1870 {
1871 compile_stack_elt_t *stack;
1872 unsigned size;
1873 unsigned avail; /* Offset of next open position. */
1874 } compile_stack_type;
1875
1876
1877 #define INIT_COMPILE_STACK_SIZE 32
1878
1879 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1880 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1881
1882 /* The next available element. */
1883 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1884
1885 /* Explicit quit checking is only used on NTemacs and whenever we
1886 use polling to process input events. */
1887 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1888 extern int immediate_quit;
1889 # define IMMEDIATE_QUIT_CHECK \
1890 do { \
1891 if (immediate_quit) QUIT; \
1892 } while (0)
1893 #else
1894 # define IMMEDIATE_QUIT_CHECK ((void)0)
1895 #endif
1896 \f
1897 /* Structure to manage work area for range table. */
1898 struct range_table_work_area
1899 {
1900 int *table; /* actual work area. */
1901 int allocated; /* allocated size for work area in bytes. */
1902 int used; /* actually used size in words. */
1903 int bits; /* flag to record character classes */
1904 };
1905
1906 /* Make sure that WORK_AREA can hold more N multibyte characters.
1907 This is used only in set_image_of_range and set_image_of_range_1.
1908 It expects WORK_AREA to be a pointer.
1909 If it can't get the space, it returns from the surrounding function. */
1910
1911 #define EXTEND_RANGE_TABLE(work_area, n) \
1912 do { \
1913 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1914 { \
1915 extend_range_table_work_area (&work_area); \
1916 if ((work_area).table == 0) \
1917 return (REG_ESPACE); \
1918 } \
1919 } while (0)
1920
1921 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1922 (work_area).bits |= (bit)
1923
1924 /* Bits used to implement the multibyte-part of the various character classes
1925 such as [:alnum:] in a charset's range table. */
1926 #define BIT_WORD 0x1
1927 #define BIT_LOWER 0x2
1928 #define BIT_PUNCT 0x4
1929 #define BIT_SPACE 0x8
1930 #define BIT_UPPER 0x10
1931 #define BIT_MULTIBYTE 0x20
1932
1933 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1934 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1935 do { \
1936 EXTEND_RANGE_TABLE ((work_area), 2); \
1937 (work_area).table[(work_area).used++] = (range_start); \
1938 (work_area).table[(work_area).used++] = (range_end); \
1939 } while (0)
1940
1941 /* Free allocated memory for WORK_AREA. */
1942 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1943 do { \
1944 if ((work_area).table) \
1945 free ((work_area).table); \
1946 } while (0)
1947
1948 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1949 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1950 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1951 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1952 \f
1953
1954 /* Set the bit for character C in a list. */
1955 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1956
1957
1958 #ifdef emacs
1959
1960 /* Store characters in the range FROM to TO in the bitmap at B (for
1961 ASCII and unibyte characters) and WORK_AREA (for multibyte
1962 characters) while translating them and paying attention to the
1963 continuity of translated characters.
1964
1965 Implementation note: It is better to implement these fairly big
1966 macros by a function, but it's not that easy because macros called
1967 in this macro assume various local variables already declared. */
1968
1969 /* Both FROM and TO are ASCII characters. */
1970
1971 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1972 do { \
1973 int C0, C1; \
1974 \
1975 for (C0 = (FROM); C0 <= (TO); C0++) \
1976 { \
1977 C1 = TRANSLATE (C0); \
1978 if (! ASCII_CHAR_P (C1)) \
1979 { \
1980 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1981 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1982 C1 = C0; \
1983 } \
1984 SET_LIST_BIT (C1); \
1985 } \
1986 } while (0)
1987
1988
1989 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1990
1991 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1992 do { \
1993 int C0, C1, C2, I; \
1994 int USED = RANGE_TABLE_WORK_USED (work_area); \
1995 \
1996 for (C0 = (FROM); C0 <= (TO); C0++) \
1997 { \
1998 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1999 if (CHAR_BYTE8_P (C1)) \
2000 SET_LIST_BIT (C0); \
2001 else \
2002 { \
2003 C2 = TRANSLATE (C1); \
2004 if (C2 == C1 \
2005 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2006 C1 = C0; \
2007 SET_LIST_BIT (C1); \
2008 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2009 { \
2010 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2011 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2012 \
2013 if (C2 >= from - 1 && C2 <= to + 1) \
2014 { \
2015 if (C2 == from - 1) \
2016 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2017 else if (C2 == to + 1) \
2018 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2019 break; \
2020 } \
2021 } \
2022 if (I < USED) \
2023 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2024 } \
2025 } \
2026 } while (0)
2027
2028
2029 /* Both FROM and TO are multibyte characters. */
2030
2031 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2032 do { \
2033 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2034 \
2035 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2036 for (C0 = (FROM); C0 <= (TO); C0++) \
2037 { \
2038 C1 = TRANSLATE (C0); \
2039 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2040 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2041 SET_LIST_BIT (C2); \
2042 if (C1 >= (FROM) && C1 <= (TO)) \
2043 continue; \
2044 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2045 { \
2046 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2047 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2048 \
2049 if (C1 >= from - 1 && C1 <= to + 1) \
2050 { \
2051 if (C1 == from - 1) \
2052 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2053 else if (C1 == to + 1) \
2054 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2055 break; \
2056 } \
2057 } \
2058 if (I < USED) \
2059 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2060 } \
2061 } while (0)
2062
2063 #endif /* emacs */
2064
2065 /* Get the next unsigned number in the uncompiled pattern. */
2066 #define GET_UNSIGNED_NUMBER(num) \
2067 do { \
2068 if (p == pend) \
2069 FREE_STACK_RETURN (REG_EBRACE); \
2070 else \
2071 { \
2072 PATFETCH (c); \
2073 while ('0' <= c && c <= '9') \
2074 { \
2075 int prev; \
2076 if (num < 0) \
2077 num = 0; \
2078 prev = num; \
2079 num = num * 10 + c - '0'; \
2080 if (num / 10 != prev) \
2081 FREE_STACK_RETURN (REG_BADBR); \
2082 if (p == pend) \
2083 FREE_STACK_RETURN (REG_EBRACE); \
2084 PATFETCH (c); \
2085 } \
2086 } \
2087 } while (0)
2088 \f
2089 #if ! WIDE_CHAR_SUPPORT
2090
2091 /* Map a string to the char class it names (if any). */
2092 re_wctype_t
2093 re_wctype (const re_char *str)
2094 {
2095 const char *string = (const char *) str;
2096 if (STREQ (string, "alnum")) return RECC_ALNUM;
2097 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2098 else if (STREQ (string, "word")) return RECC_WORD;
2099 else if (STREQ (string, "ascii")) return RECC_ASCII;
2100 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2101 else if (STREQ (string, "graph")) return RECC_GRAPH;
2102 else if (STREQ (string, "lower")) return RECC_LOWER;
2103 else if (STREQ (string, "print")) return RECC_PRINT;
2104 else if (STREQ (string, "punct")) return RECC_PUNCT;
2105 else if (STREQ (string, "space")) return RECC_SPACE;
2106 else if (STREQ (string, "upper")) return RECC_UPPER;
2107 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2108 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2109 else if (STREQ (string, "digit")) return RECC_DIGIT;
2110 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2111 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2112 else if (STREQ (string, "blank")) return RECC_BLANK;
2113 else return 0;
2114 }
2115
2116 /* True if CH is in the char class CC. */
2117 boolean
2118 re_iswctype (int ch, re_wctype_t cc)
2119 {
2120 switch (cc)
2121 {
2122 case RECC_ALNUM: return ISALNUM (ch);
2123 case RECC_ALPHA: return ISALPHA (ch);
2124 case RECC_BLANK: return ISBLANK (ch);
2125 case RECC_CNTRL: return ISCNTRL (ch);
2126 case RECC_DIGIT: return ISDIGIT (ch);
2127 case RECC_GRAPH: return ISGRAPH (ch);
2128 case RECC_LOWER: return ISLOWER (ch);
2129 case RECC_PRINT: return ISPRINT (ch);
2130 case RECC_PUNCT: return ISPUNCT (ch);
2131 case RECC_SPACE: return ISSPACE (ch);
2132 case RECC_UPPER: return ISUPPER (ch);
2133 case RECC_XDIGIT: return ISXDIGIT (ch);
2134 case RECC_ASCII: return IS_REAL_ASCII (ch);
2135 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2136 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2137 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2138 case RECC_WORD: return ISWORD (ch);
2139 case RECC_ERROR: return false;
2140 default:
2141 abort();
2142 }
2143 }
2144
2145 /* Return a bit-pattern to use in the range-table bits to match multibyte
2146 chars of class CC. */
2147 static int
2148 re_wctype_to_bit (re_wctype_t cc)
2149 {
2150 switch (cc)
2151 {
2152 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2153 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2154 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2155 case RECC_LOWER: return BIT_LOWER;
2156 case RECC_UPPER: return BIT_UPPER;
2157 case RECC_PUNCT: return BIT_PUNCT;
2158 case RECC_SPACE: return BIT_SPACE;
2159 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2160 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2161 default:
2162 abort();
2163 }
2164 }
2165 #endif
2166 \f
2167 /* Filling in the work area of a range. */
2168
2169 /* Actually extend the space in WORK_AREA. */
2170
2171 static void
2172 extend_range_table_work_area (struct range_table_work_area *work_area)
2173 {
2174 work_area->allocated += 16 * sizeof (int);
2175 if (work_area->table)
2176 work_area->table
2177 = (int *) realloc (work_area->table, work_area->allocated);
2178 else
2179 work_area->table
2180 = (int *) malloc (work_area->allocated);
2181 }
2182
2183 #if 0
2184 #ifdef emacs
2185
2186 /* Carefully find the ranges of codes that are equivalent
2187 under case conversion to the range start..end when passed through
2188 TRANSLATE. Handle the case where non-letters can come in between
2189 two upper-case letters (which happens in Latin-1).
2190 Also handle the case of groups of more than 2 case-equivalent chars.
2191
2192 The basic method is to look at consecutive characters and see
2193 if they can form a run that can be handled as one.
2194
2195 Returns -1 if successful, REG_ESPACE if ran out of space. */
2196
2197 static int
2198 set_image_of_range_1 (work_area, start, end, translate)
2199 RE_TRANSLATE_TYPE translate;
2200 struct range_table_work_area *work_area;
2201 re_wchar_t start, end;
2202 {
2203 /* `one_case' indicates a character, or a run of characters,
2204 each of which is an isolate (no case-equivalents).
2205 This includes all ASCII non-letters.
2206
2207 `two_case' indicates a character, or a run of characters,
2208 each of which has two case-equivalent forms.
2209 This includes all ASCII letters.
2210
2211 `strange' indicates a character that has more than one
2212 case-equivalent. */
2213
2214 enum case_type {one_case, two_case, strange};
2215
2216 /* Describe the run that is in progress,
2217 which the next character can try to extend.
2218 If run_type is strange, that means there really is no run.
2219 If run_type is one_case, then run_start...run_end is the run.
2220 If run_type is two_case, then the run is run_start...run_end,
2221 and the case-equivalents end at run_eqv_end. */
2222
2223 enum case_type run_type = strange;
2224 int run_start, run_end, run_eqv_end;
2225
2226 Lisp_Object eqv_table;
2227
2228 if (!RE_TRANSLATE_P (translate))
2229 {
2230 EXTEND_RANGE_TABLE (work_area, 2);
2231 work_area->table[work_area->used++] = (start);
2232 work_area->table[work_area->used++] = (end);
2233 return -1;
2234 }
2235
2236 eqv_table = XCHAR_TABLE (translate)->extras[2];
2237
2238 for (; start <= end; start++)
2239 {
2240 enum case_type this_type;
2241 int eqv = RE_TRANSLATE (eqv_table, start);
2242 int minchar, maxchar;
2243
2244 /* Classify this character */
2245 if (eqv == start)
2246 this_type = one_case;
2247 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2248 this_type = two_case;
2249 else
2250 this_type = strange;
2251
2252 if (start < eqv)
2253 minchar = start, maxchar = eqv;
2254 else
2255 minchar = eqv, maxchar = start;
2256
2257 /* Can this character extend the run in progress? */
2258 if (this_type == strange || this_type != run_type
2259 || !(minchar == run_end + 1
2260 && (run_type == two_case
2261 ? maxchar == run_eqv_end + 1 : 1)))
2262 {
2263 /* No, end the run.
2264 Record each of its equivalent ranges. */
2265 if (run_type == one_case)
2266 {
2267 EXTEND_RANGE_TABLE (work_area, 2);
2268 work_area->table[work_area->used++] = run_start;
2269 work_area->table[work_area->used++] = run_end;
2270 }
2271 else if (run_type == two_case)
2272 {
2273 EXTEND_RANGE_TABLE (work_area, 4);
2274 work_area->table[work_area->used++] = run_start;
2275 work_area->table[work_area->used++] = run_end;
2276 work_area->table[work_area->used++]
2277 = RE_TRANSLATE (eqv_table, run_start);
2278 work_area->table[work_area->used++]
2279 = RE_TRANSLATE (eqv_table, run_end);
2280 }
2281 run_type = strange;
2282 }
2283
2284 if (this_type == strange)
2285 {
2286 /* For a strange character, add each of its equivalents, one
2287 by one. Don't start a range. */
2288 do
2289 {
2290 EXTEND_RANGE_TABLE (work_area, 2);
2291 work_area->table[work_area->used++] = eqv;
2292 work_area->table[work_area->used++] = eqv;
2293 eqv = RE_TRANSLATE (eqv_table, eqv);
2294 }
2295 while (eqv != start);
2296 }
2297
2298 /* Add this char to the run, or start a new run. */
2299 else if (run_type == strange)
2300 {
2301 /* Initialize a new range. */
2302 run_type = this_type;
2303 run_start = start;
2304 run_end = start;
2305 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2306 }
2307 else
2308 {
2309 /* Extend a running range. */
2310 run_end = minchar;
2311 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2312 }
2313 }
2314
2315 /* If a run is still in progress at the end, finish it now
2316 by recording its equivalent ranges. */
2317 if (run_type == one_case)
2318 {
2319 EXTEND_RANGE_TABLE (work_area, 2);
2320 work_area->table[work_area->used++] = run_start;
2321 work_area->table[work_area->used++] = run_end;
2322 }
2323 else if (run_type == two_case)
2324 {
2325 EXTEND_RANGE_TABLE (work_area, 4);
2326 work_area->table[work_area->used++] = run_start;
2327 work_area->table[work_area->used++] = run_end;
2328 work_area->table[work_area->used++]
2329 = RE_TRANSLATE (eqv_table, run_start);
2330 work_area->table[work_area->used++]
2331 = RE_TRANSLATE (eqv_table, run_end);
2332 }
2333
2334 return -1;
2335 }
2336
2337 #endif /* emacs */
2338
2339 /* Record the image of the range start..end when passed through
2340 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2341 and is not even necessarily contiguous.
2342 Normally we approximate it with the smallest contiguous range that contains
2343 all the chars we need. However, for Latin-1 we go to extra effort
2344 to do a better job.
2345
2346 This function is not called for ASCII ranges.
2347
2348 Returns -1 if successful, REG_ESPACE if ran out of space. */
2349
2350 static int
2351 set_image_of_range (work_area, start, end, translate)
2352 RE_TRANSLATE_TYPE translate;
2353 struct range_table_work_area *work_area;
2354 re_wchar_t start, end;
2355 {
2356 re_wchar_t cmin, cmax;
2357
2358 #ifdef emacs
2359 /* For Latin-1 ranges, use set_image_of_range_1
2360 to get proper handling of ranges that include letters and nonletters.
2361 For a range that includes the whole of Latin-1, this is not necessary.
2362 For other character sets, we don't bother to get this right. */
2363 if (RE_TRANSLATE_P (translate) && start < 04400
2364 && !(start < 04200 && end >= 04377))
2365 {
2366 int newend;
2367 int tem;
2368 newend = end;
2369 if (newend > 04377)
2370 newend = 04377;
2371 tem = set_image_of_range_1 (work_area, start, newend, translate);
2372 if (tem > 0)
2373 return tem;
2374
2375 start = 04400;
2376 if (end < 04400)
2377 return -1;
2378 }
2379 #endif
2380
2381 EXTEND_RANGE_TABLE (work_area, 2);
2382 work_area->table[work_area->used++] = (start);
2383 work_area->table[work_area->used++] = (end);
2384
2385 cmin = -1, cmax = -1;
2386
2387 if (RE_TRANSLATE_P (translate))
2388 {
2389 int ch;
2390
2391 for (ch = start; ch <= end; ch++)
2392 {
2393 re_wchar_t c = TRANSLATE (ch);
2394 if (! (start <= c && c <= end))
2395 {
2396 if (cmin == -1)
2397 cmin = c, cmax = c;
2398 else
2399 {
2400 cmin = MIN (cmin, c);
2401 cmax = MAX (cmax, c);
2402 }
2403 }
2404 }
2405
2406 if (cmin != -1)
2407 {
2408 EXTEND_RANGE_TABLE (work_area, 2);
2409 work_area->table[work_area->used++] = (cmin);
2410 work_area->table[work_area->used++] = (cmax);
2411 }
2412 }
2413
2414 return -1;
2415 }
2416 #endif /* 0 */
2417 \f
2418 #ifndef MATCH_MAY_ALLOCATE
2419
2420 /* If we cannot allocate large objects within re_match_2_internal,
2421 we make the fail stack and register vectors global.
2422 The fail stack, we grow to the maximum size when a regexp
2423 is compiled.
2424 The register vectors, we adjust in size each time we
2425 compile a regexp, according to the number of registers it needs. */
2426
2427 static fail_stack_type fail_stack;
2428
2429 /* Size with which the following vectors are currently allocated.
2430 That is so we can make them bigger as needed,
2431 but never make them smaller. */
2432 static int regs_allocated_size;
2433
2434 static re_char ** regstart, ** regend;
2435 static re_char **best_regstart, **best_regend;
2436
2437 /* Make the register vectors big enough for NUM_REGS registers,
2438 but don't make them smaller. */
2439
2440 static
2441 regex_grow_registers (num_regs)
2442 int num_regs;
2443 {
2444 if (num_regs > regs_allocated_size)
2445 {
2446 RETALLOC_IF (regstart, num_regs, re_char *);
2447 RETALLOC_IF (regend, num_regs, re_char *);
2448 RETALLOC_IF (best_regstart, num_regs, re_char *);
2449 RETALLOC_IF (best_regend, num_regs, re_char *);
2450
2451 regs_allocated_size = num_regs;
2452 }
2453 }
2454
2455 #endif /* not MATCH_MAY_ALLOCATE */
2456 \f
2457 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2458 compile_stack,
2459 regnum_t regnum));
2460
2461 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2462 Returns one of error codes defined in `regex.h', or zero for success.
2463
2464 Assumes the `allocated' (and perhaps `buffer') and `translate'
2465 fields are set in BUFP on entry.
2466
2467 If it succeeds, results are put in BUFP (if it returns an error, the
2468 contents of BUFP are undefined):
2469 `buffer' is the compiled pattern;
2470 `syntax' is set to SYNTAX;
2471 `used' is set to the length of the compiled pattern;
2472 `fastmap_accurate' is zero;
2473 `re_nsub' is the number of subexpressions in PATTERN;
2474 `not_bol' and `not_eol' are zero;
2475
2476 The `fastmap' field is neither examined nor set. */
2477
2478 /* Insert the `jump' from the end of last alternative to "here".
2479 The space for the jump has already been allocated. */
2480 #define FIXUP_ALT_JUMP() \
2481 do { \
2482 if (fixup_alt_jump) \
2483 STORE_JUMP (jump, fixup_alt_jump, b); \
2484 } while (0)
2485
2486
2487 /* Return, freeing storage we allocated. */
2488 #define FREE_STACK_RETURN(value) \
2489 do { \
2490 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2491 free (compile_stack.stack); \
2492 return value; \
2493 } while (0)
2494
2495 static reg_errcode_t
2496 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2497 {
2498 /* We fetch characters from PATTERN here. */
2499 register re_wchar_t c, c1;
2500
2501 /* Points to the end of the buffer, where we should append. */
2502 register unsigned char *b;
2503
2504 /* Keeps track of unclosed groups. */
2505 compile_stack_type compile_stack;
2506
2507 /* Points to the current (ending) position in the pattern. */
2508 #ifdef AIX
2509 /* `const' makes AIX compiler fail. */
2510 unsigned char *p = pattern;
2511 #else
2512 re_char *p = pattern;
2513 #endif
2514 re_char *pend = pattern + size;
2515
2516 /* How to translate the characters in the pattern. */
2517 RE_TRANSLATE_TYPE translate = bufp->translate;
2518
2519 /* Address of the count-byte of the most recently inserted `exactn'
2520 command. This makes it possible to tell if a new exact-match
2521 character can be added to that command or if the character requires
2522 a new `exactn' command. */
2523 unsigned char *pending_exact = 0;
2524
2525 /* Address of start of the most recently finished expression.
2526 This tells, e.g., postfix * where to find the start of its
2527 operand. Reset at the beginning of groups and alternatives. */
2528 unsigned char *laststart = 0;
2529
2530 /* Address of beginning of regexp, or inside of last group. */
2531 unsigned char *begalt;
2532
2533 /* Place in the uncompiled pattern (i.e., the {) to
2534 which to go back if the interval is invalid. */
2535 re_char *beg_interval;
2536
2537 /* Address of the place where a forward jump should go to the end of
2538 the containing expression. Each alternative of an `or' -- except the
2539 last -- ends with a forward jump of this sort. */
2540 unsigned char *fixup_alt_jump = 0;
2541
2542 /* Work area for range table of charset. */
2543 struct range_table_work_area range_table_work;
2544
2545 /* If the object matched can contain multibyte characters. */
2546 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2547
2548 /* Nonzero if we have pushed down into a subpattern. */
2549 int in_subpattern = 0;
2550
2551 /* These hold the values of p, pattern, and pend from the main
2552 pattern when we have pushed into a subpattern. */
2553 re_char *main_p IF_LINT (= NULL);
2554 re_char *main_pattern IF_LINT (= NULL);
2555 re_char *main_pend IF_LINT (= NULL);
2556
2557 #ifdef DEBUG
2558 debug++;
2559 DEBUG_PRINT1 ("\nCompiling pattern: ");
2560 if (debug > 0)
2561 {
2562 unsigned debug_count;
2563
2564 for (debug_count = 0; debug_count < size; debug_count++)
2565 putchar (pattern[debug_count]);
2566 putchar ('\n');
2567 }
2568 #endif /* DEBUG */
2569
2570 /* Initialize the compile stack. */
2571 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2572 if (compile_stack.stack == NULL)
2573 return REG_ESPACE;
2574
2575 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2576 compile_stack.avail = 0;
2577
2578 range_table_work.table = 0;
2579 range_table_work.allocated = 0;
2580
2581 /* Initialize the pattern buffer. */
2582 bufp->syntax = syntax;
2583 bufp->fastmap_accurate = 0;
2584 bufp->not_bol = bufp->not_eol = 0;
2585 bufp->used_syntax = 0;
2586
2587 /* Set `used' to zero, so that if we return an error, the pattern
2588 printer (for debugging) will think there's no pattern. We reset it
2589 at the end. */
2590 bufp->used = 0;
2591
2592 /* Always count groups, whether or not bufp->no_sub is set. */
2593 bufp->re_nsub = 0;
2594
2595 #if !defined emacs && !defined SYNTAX_TABLE
2596 /* Initialize the syntax table. */
2597 init_syntax_once ();
2598 #endif
2599
2600 if (bufp->allocated == 0)
2601 {
2602 if (bufp->buffer)
2603 { /* If zero allocated, but buffer is non-null, try to realloc
2604 enough space. This loses if buffer's address is bogus, but
2605 that is the user's responsibility. */
2606 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2607 }
2608 else
2609 { /* Caller did not allocate a buffer. Do it for them. */
2610 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2611 }
2612 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2613
2614 bufp->allocated = INIT_BUF_SIZE;
2615 }
2616
2617 begalt = b = bufp->buffer;
2618
2619 /* Loop through the uncompiled pattern until we're at the end. */
2620 while (1)
2621 {
2622 if (p == pend)
2623 {
2624 /* If this is the end of an included regexp,
2625 pop back to the main regexp and try again. */
2626 if (in_subpattern)
2627 {
2628 in_subpattern = 0;
2629 pattern = main_pattern;
2630 p = main_p;
2631 pend = main_pend;
2632 continue;
2633 }
2634 /* If this is the end of the main regexp, we are done. */
2635 break;
2636 }
2637
2638 PATFETCH (c);
2639
2640 switch (c)
2641 {
2642 case ' ':
2643 {
2644 re_char *p1 = p;
2645
2646 /* If there's no special whitespace regexp, treat
2647 spaces normally. And don't try to do this recursively. */
2648 if (!whitespace_regexp || in_subpattern)
2649 goto normal_char;
2650
2651 /* Peek past following spaces. */
2652 while (p1 != pend)
2653 {
2654 if (*p1 != ' ')
2655 break;
2656 p1++;
2657 }
2658 /* If the spaces are followed by a repetition op,
2659 treat them normally. */
2660 if (p1 != pend
2661 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2662 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2663 goto normal_char;
2664
2665 /* Replace the spaces with the whitespace regexp. */
2666 in_subpattern = 1;
2667 main_p = p1;
2668 main_pend = pend;
2669 main_pattern = pattern;
2670 p = pattern = whitespace_regexp;
2671 pend = p + strlen ((const char *) p);
2672 break;
2673 }
2674
2675 case '^':
2676 {
2677 if ( /* If at start of pattern, it's an operator. */
2678 p == pattern + 1
2679 /* If context independent, it's an operator. */
2680 || syntax & RE_CONTEXT_INDEP_ANCHORS
2681 /* Otherwise, depends on what's come before. */
2682 || at_begline_loc_p (pattern, p, syntax))
2683 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2684 else
2685 goto normal_char;
2686 }
2687 break;
2688
2689
2690 case '$':
2691 {
2692 if ( /* If at end of pattern, it's an operator. */
2693 p == pend
2694 /* If context independent, it's an operator. */
2695 || syntax & RE_CONTEXT_INDEP_ANCHORS
2696 /* Otherwise, depends on what's next. */
2697 || at_endline_loc_p (p, pend, syntax))
2698 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2699 else
2700 goto normal_char;
2701 }
2702 break;
2703
2704
2705 case '+':
2706 case '?':
2707 if ((syntax & RE_BK_PLUS_QM)
2708 || (syntax & RE_LIMITED_OPS))
2709 goto normal_char;
2710 handle_plus:
2711 case '*':
2712 /* If there is no previous pattern... */
2713 if (!laststart)
2714 {
2715 if (syntax & RE_CONTEXT_INVALID_OPS)
2716 FREE_STACK_RETURN (REG_BADRPT);
2717 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2718 goto normal_char;
2719 }
2720
2721 {
2722 /* 1 means zero (many) matches is allowed. */
2723 boolean zero_times_ok = 0, many_times_ok = 0;
2724 boolean greedy = 1;
2725
2726 /* If there is a sequence of repetition chars, collapse it
2727 down to just one (the right one). We can't combine
2728 interval operators with these because of, e.g., `a{2}*',
2729 which should only match an even number of `a's. */
2730
2731 for (;;)
2732 {
2733 if ((syntax & RE_FRUGAL)
2734 && c == '?' && (zero_times_ok || many_times_ok))
2735 greedy = 0;
2736 else
2737 {
2738 zero_times_ok |= c != '+';
2739 many_times_ok |= c != '?';
2740 }
2741
2742 if (p == pend)
2743 break;
2744 else if (*p == '*'
2745 || (!(syntax & RE_BK_PLUS_QM)
2746 && (*p == '+' || *p == '?')))
2747 ;
2748 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2749 {
2750 if (p+1 == pend)
2751 FREE_STACK_RETURN (REG_EESCAPE);
2752 if (p[1] == '+' || p[1] == '?')
2753 PATFETCH (c); /* Gobble up the backslash. */
2754 else
2755 break;
2756 }
2757 else
2758 break;
2759 /* If we get here, we found another repeat character. */
2760 PATFETCH (c);
2761 }
2762
2763 /* Star, etc. applied to an empty pattern is equivalent
2764 to an empty pattern. */
2765 if (!laststart || laststart == b)
2766 break;
2767
2768 /* Now we know whether or not zero matches is allowed
2769 and also whether or not two or more matches is allowed. */
2770 if (greedy)
2771 {
2772 if (many_times_ok)
2773 {
2774 boolean simple = skip_one_char (laststart) == b;
2775 unsigned int startoffset = 0;
2776 re_opcode_t ofj =
2777 /* Check if the loop can match the empty string. */
2778 (simple || !analyse_first (laststart, b, NULL, 0))
2779 ? on_failure_jump : on_failure_jump_loop;
2780 assert (skip_one_char (laststart) <= b);
2781
2782 if (!zero_times_ok && simple)
2783 { /* Since simple * loops can be made faster by using
2784 on_failure_keep_string_jump, we turn simple P+
2785 into PP* if P is simple. */
2786 unsigned char *p1, *p2;
2787 startoffset = b - laststart;
2788 GET_BUFFER_SPACE (startoffset);
2789 p1 = b; p2 = laststart;
2790 while (p2 < p1)
2791 *b++ = *p2++;
2792 zero_times_ok = 1;
2793 }
2794
2795 GET_BUFFER_SPACE (6);
2796 if (!zero_times_ok)
2797 /* A + loop. */
2798 STORE_JUMP (ofj, b, b + 6);
2799 else
2800 /* Simple * loops can use on_failure_keep_string_jump
2801 depending on what follows. But since we don't know
2802 that yet, we leave the decision up to
2803 on_failure_jump_smart. */
2804 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2805 laststart + startoffset, b + 6);
2806 b += 3;
2807 STORE_JUMP (jump, b, laststart + startoffset);
2808 b += 3;
2809 }
2810 else
2811 {
2812 /* A simple ? pattern. */
2813 assert (zero_times_ok);
2814 GET_BUFFER_SPACE (3);
2815 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2816 b += 3;
2817 }
2818 }
2819 else /* not greedy */
2820 { /* I wish the greedy and non-greedy cases could be merged. */
2821
2822 GET_BUFFER_SPACE (7); /* We might use less. */
2823 if (many_times_ok)
2824 {
2825 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2826
2827 /* The non-greedy multiple match looks like
2828 a repeat..until: we only need a conditional jump
2829 at the end of the loop. */
2830 if (emptyp) BUF_PUSH (no_op);
2831 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2832 : on_failure_jump, b, laststart);
2833 b += 3;
2834 if (zero_times_ok)
2835 {
2836 /* The repeat...until naturally matches one or more.
2837 To also match zero times, we need to first jump to
2838 the end of the loop (its conditional jump). */
2839 INSERT_JUMP (jump, laststart, b);
2840 b += 3;
2841 }
2842 }
2843 else
2844 {
2845 /* non-greedy a?? */
2846 INSERT_JUMP (jump, laststart, b + 3);
2847 b += 3;
2848 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2849 b += 3;
2850 }
2851 }
2852 }
2853 pending_exact = 0;
2854 break;
2855
2856
2857 case '.':
2858 laststart = b;
2859 BUF_PUSH (anychar);
2860 break;
2861
2862
2863 case '[':
2864 {
2865 re_char *p1;
2866
2867 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2868
2869 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2870
2871 /* Ensure that we have enough space to push a charset: the
2872 opcode, the length count, and the bitset; 34 bytes in all. */
2873 GET_BUFFER_SPACE (34);
2874
2875 laststart = b;
2876
2877 /* We test `*p == '^' twice, instead of using an if
2878 statement, so we only need one BUF_PUSH. */
2879 BUF_PUSH (*p == '^' ? charset_not : charset);
2880 if (*p == '^')
2881 p++;
2882
2883 /* Remember the first position in the bracket expression. */
2884 p1 = p;
2885
2886 /* Push the number of bytes in the bitmap. */
2887 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2888
2889 /* Clear the whole map. */
2890 memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH);
2891
2892 /* charset_not matches newline according to a syntax bit. */
2893 if ((re_opcode_t) b[-2] == charset_not
2894 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2895 SET_LIST_BIT ('\n');
2896
2897 /* Read in characters and ranges, setting map bits. */
2898 for (;;)
2899 {
2900 boolean escaped_char = false;
2901 const unsigned char *p2 = p;
2902 re_wchar_t ch;
2903
2904 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2905
2906 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2907 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2908 So the translation is done later in a loop. Example:
2909 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2910 PATFETCH (c);
2911
2912 /* \ might escape characters inside [...] and [^...]. */
2913 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2914 {
2915 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2916
2917 PATFETCH (c);
2918 escaped_char = true;
2919 }
2920 else
2921 {
2922 /* Could be the end of the bracket expression. If it's
2923 not (i.e., when the bracket expression is `[]' so
2924 far), the ']' character bit gets set way below. */
2925 if (c == ']' && p2 != p1)
2926 break;
2927 }
2928
2929 /* See if we're at the beginning of a possible character
2930 class. */
2931
2932 if (!escaped_char &&
2933 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2934 {
2935 /* Leave room for the null. */
2936 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2937 const unsigned char *class_beg;
2938
2939 PATFETCH (c);
2940 c1 = 0;
2941 class_beg = p;
2942
2943 /* If pattern is `[[:'. */
2944 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2945
2946 for (;;)
2947 {
2948 PATFETCH (c);
2949 if ((c == ':' && *p == ']') || p == pend)
2950 break;
2951 if (c1 < CHAR_CLASS_MAX_LENGTH)
2952 str[c1++] = c;
2953 else
2954 /* This is in any case an invalid class name. */
2955 str[0] = '\0';
2956 }
2957 str[c1] = '\0';
2958
2959 /* If isn't a word bracketed by `[:' and `:]':
2960 undo the ending character, the letters, and
2961 leave the leading `:' and `[' (but set bits for
2962 them). */
2963 if (c == ':' && *p == ']')
2964 {
2965 re_wctype_t cc = re_wctype (str);
2966
2967 if (cc == 0)
2968 FREE_STACK_RETURN (REG_ECTYPE);
2969
2970 /* Throw away the ] at the end of the character
2971 class. */
2972 PATFETCH (c);
2973
2974 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2975
2976 #ifndef emacs
2977 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
2978 if (re_iswctype (btowc (ch), cc))
2979 {
2980 c = TRANSLATE (ch);
2981 if (c < (1 << BYTEWIDTH))
2982 SET_LIST_BIT (c);
2983 }
2984 #else /* emacs */
2985 /* Most character classes in a multibyte match
2986 just set a flag. Exceptions are is_blank,
2987 is_digit, is_cntrl, and is_xdigit, since
2988 they can only match ASCII characters. We
2989 don't need to handle them for multibyte.
2990 They are distinguished by a negative wctype. */
2991
2992 /* Setup the gl_state object to its buffer-defined
2993 value. This hardcodes the buffer-global
2994 syntax-table for ASCII chars, while the other chars
2995 will obey syntax-table properties. It's not ideal,
2996 but it's the way it's been done until now. */
2997 SETUP_BUFFER_SYNTAX_TABLE ();
2998
2999 for (ch = 0; ch < 256; ++ch)
3000 {
3001 c = RE_CHAR_TO_MULTIBYTE (ch);
3002 if (! CHAR_BYTE8_P (c)
3003 && re_iswctype (c, cc))
3004 {
3005 SET_LIST_BIT (ch);
3006 c1 = TRANSLATE (c);
3007 if (c1 == c)
3008 continue;
3009 if (ASCII_CHAR_P (c1))
3010 SET_LIST_BIT (c1);
3011 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
3012 SET_LIST_BIT (c1);
3013 }
3014 }
3015 SET_RANGE_TABLE_WORK_AREA_BIT
3016 (range_table_work, re_wctype_to_bit (cc));
3017 #endif /* emacs */
3018 /* In most cases the matching rule for char classes
3019 only uses the syntax table for multibyte chars,
3020 so that the content of the syntax-table it is not
3021 hardcoded in the range_table. SPACE and WORD are
3022 the two exceptions. */
3023 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
3024 bufp->used_syntax = 1;
3025
3026 /* Repeat the loop. */
3027 continue;
3028 }
3029 else
3030 {
3031 /* Go back to right after the "[:". */
3032 p = class_beg;
3033 SET_LIST_BIT ('[');
3034
3035 /* Because the `:' may starts the range, we
3036 can't simply set bit and repeat the loop.
3037 Instead, just set it to C and handle below. */
3038 c = ':';
3039 }
3040 }
3041
3042 if (p < pend && p[0] == '-' && p[1] != ']')
3043 {
3044
3045 /* Discard the `-'. */
3046 PATFETCH (c1);
3047
3048 /* Fetch the character which ends the range. */
3049 PATFETCH (c1);
3050 #ifdef emacs
3051 if (CHAR_BYTE8_P (c1)
3052 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3053 /* Treat the range from a multibyte character to
3054 raw-byte character as empty. */
3055 c = c1 + 1;
3056 #endif /* emacs */
3057 }
3058 else
3059 /* Range from C to C. */
3060 c1 = c;
3061
3062 if (c > c1)
3063 {
3064 if (syntax & RE_NO_EMPTY_RANGES)
3065 FREE_STACK_RETURN (REG_ERANGEX);
3066 /* Else, repeat the loop. */
3067 }
3068 else
3069 {
3070 #ifndef emacs
3071 /* Set the range into bitmap */
3072 for (; c <= c1; c++)
3073 {
3074 ch = TRANSLATE (c);
3075 if (ch < (1 << BYTEWIDTH))
3076 SET_LIST_BIT (ch);
3077 }
3078 #else /* emacs */
3079 if (c < 128)
3080 {
3081 ch = MIN (127, c1);
3082 SETUP_ASCII_RANGE (range_table_work, c, ch);
3083 c = ch + 1;
3084 if (CHAR_BYTE8_P (c1))
3085 c = BYTE8_TO_CHAR (128);
3086 }
3087 if (c <= c1)
3088 {
3089 if (CHAR_BYTE8_P (c))
3090 {
3091 c = CHAR_TO_BYTE8 (c);
3092 c1 = CHAR_TO_BYTE8 (c1);
3093 for (; c <= c1; c++)
3094 SET_LIST_BIT (c);
3095 }
3096 else if (multibyte)
3097 {
3098 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3099 }
3100 else
3101 {
3102 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3103 }
3104 }
3105 #endif /* emacs */
3106 }
3107 }
3108
3109 /* Discard any (non)matching list bytes that are all 0 at the
3110 end of the map. Decrease the map-length byte too. */
3111 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3112 b[-1]--;
3113 b += b[-1];
3114
3115 /* Build real range table from work area. */
3116 if (RANGE_TABLE_WORK_USED (range_table_work)
3117 || RANGE_TABLE_WORK_BITS (range_table_work))
3118 {
3119 int i;
3120 int used = RANGE_TABLE_WORK_USED (range_table_work);
3121
3122 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3123 bytes for flags, two for COUNT, and three bytes for
3124 each character. */
3125 GET_BUFFER_SPACE (4 + used * 3);
3126
3127 /* Indicate the existence of range table. */
3128 laststart[1] |= 0x80;
3129
3130 /* Store the character class flag bits into the range table.
3131 If not in emacs, these flag bits are always 0. */
3132 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3133 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3134
3135 STORE_NUMBER_AND_INCR (b, used / 2);
3136 for (i = 0; i < used; i++)
3137 STORE_CHARACTER_AND_INCR
3138 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3139 }
3140 }
3141 break;
3142
3143
3144 case '(':
3145 if (syntax & RE_NO_BK_PARENS)
3146 goto handle_open;
3147 else
3148 goto normal_char;
3149
3150
3151 case ')':
3152 if (syntax & RE_NO_BK_PARENS)
3153 goto handle_close;
3154 else
3155 goto normal_char;
3156
3157
3158 case '\n':
3159 if (syntax & RE_NEWLINE_ALT)
3160 goto handle_alt;
3161 else
3162 goto normal_char;
3163
3164
3165 case '|':
3166 if (syntax & RE_NO_BK_VBAR)
3167 goto handle_alt;
3168 else
3169 goto normal_char;
3170
3171
3172 case '{':
3173 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3174 goto handle_interval;
3175 else
3176 goto normal_char;
3177
3178
3179 case '\\':
3180 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3181
3182 /* Do not translate the character after the \, so that we can
3183 distinguish, e.g., \B from \b, even if we normally would
3184 translate, e.g., B to b. */
3185 PATFETCH (c);
3186
3187 switch (c)
3188 {
3189 case '(':
3190 if (syntax & RE_NO_BK_PARENS)
3191 goto normal_backslash;
3192
3193 handle_open:
3194 {
3195 int shy = 0;
3196 regnum_t regnum = 0;
3197 if (p+1 < pend)
3198 {
3199 /* Look for a special (?...) construct */
3200 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3201 {
3202 PATFETCH (c); /* Gobble up the '?'. */
3203 while (!shy)
3204 {
3205 PATFETCH (c);
3206 switch (c)
3207 {
3208 case ':': shy = 1; break;
3209 case '0':
3210 /* An explicitly specified regnum must start
3211 with non-0. */
3212 if (regnum == 0)
3213 FREE_STACK_RETURN (REG_BADPAT);
3214 case '1': case '2': case '3': case '4':
3215 case '5': case '6': case '7': case '8': case '9':
3216 regnum = 10*regnum + (c - '0'); break;
3217 default:
3218 /* Only (?:...) is supported right now. */
3219 FREE_STACK_RETURN (REG_BADPAT);
3220 }
3221 }
3222 }
3223 }
3224
3225 if (!shy)
3226 regnum = ++bufp->re_nsub;
3227 else if (regnum)
3228 { /* It's actually not shy, but explicitly numbered. */
3229 shy = 0;
3230 if (regnum > bufp->re_nsub)
3231 bufp->re_nsub = regnum;
3232 else if (regnum > bufp->re_nsub
3233 /* Ideally, we'd want to check that the specified
3234 group can't have matched (i.e. all subgroups
3235 using the same regnum are in other branches of
3236 OR patterns), but we don't currently keep track
3237 of enough info to do that easily. */
3238 || group_in_compile_stack (compile_stack, regnum))
3239 FREE_STACK_RETURN (REG_BADPAT);
3240 }
3241 else
3242 /* It's really shy. */
3243 regnum = - bufp->re_nsub;
3244
3245 if (COMPILE_STACK_FULL)
3246 {
3247 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3248 compile_stack_elt_t);
3249 if (compile_stack.stack == NULL) return REG_ESPACE;
3250
3251 compile_stack.size <<= 1;
3252 }
3253
3254 /* These are the values to restore when we hit end of this
3255 group. They are all relative offsets, so that if the
3256 whole pattern moves because of realloc, they will still
3257 be valid. */
3258 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3259 COMPILE_STACK_TOP.fixup_alt_jump
3260 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3261 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3262 COMPILE_STACK_TOP.regnum = regnum;
3263
3264 /* Do not push a start_memory for groups beyond the last one
3265 we can represent in the compiled pattern. */
3266 if (regnum <= MAX_REGNUM && regnum > 0)
3267 BUF_PUSH_2 (start_memory, regnum);
3268
3269 compile_stack.avail++;
3270
3271 fixup_alt_jump = 0;
3272 laststart = 0;
3273 begalt = b;
3274 /* If we've reached MAX_REGNUM groups, then this open
3275 won't actually generate any code, so we'll have to
3276 clear pending_exact explicitly. */
3277 pending_exact = 0;
3278 break;
3279 }
3280
3281 case ')':
3282 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3283
3284 if (COMPILE_STACK_EMPTY)
3285 {
3286 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3287 goto normal_backslash;
3288 else
3289 FREE_STACK_RETURN (REG_ERPAREN);
3290 }
3291
3292 handle_close:
3293 FIXUP_ALT_JUMP ();
3294
3295 /* See similar code for backslashed left paren above. */
3296 if (COMPILE_STACK_EMPTY)
3297 {
3298 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3299 goto normal_char;
3300 else
3301 FREE_STACK_RETURN (REG_ERPAREN);
3302 }
3303
3304 /* Since we just checked for an empty stack above, this
3305 ``can't happen''. */
3306 assert (compile_stack.avail != 0);
3307 {
3308 /* We don't just want to restore into `regnum', because
3309 later groups should continue to be numbered higher,
3310 as in `(ab)c(de)' -- the second group is #2. */
3311 regnum_t regnum;
3312
3313 compile_stack.avail--;
3314 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3315 fixup_alt_jump
3316 = COMPILE_STACK_TOP.fixup_alt_jump
3317 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3318 : 0;
3319 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3320 regnum = COMPILE_STACK_TOP.regnum;
3321 /* If we've reached MAX_REGNUM groups, then this open
3322 won't actually generate any code, so we'll have to
3323 clear pending_exact explicitly. */
3324 pending_exact = 0;
3325
3326 /* We're at the end of the group, so now we know how many
3327 groups were inside this one. */
3328 if (regnum <= MAX_REGNUM && regnum > 0)
3329 BUF_PUSH_2 (stop_memory, regnum);
3330 }
3331 break;
3332
3333
3334 case '|': /* `\|'. */
3335 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3336 goto normal_backslash;
3337 handle_alt:
3338 if (syntax & RE_LIMITED_OPS)
3339 goto normal_char;
3340
3341 /* Insert before the previous alternative a jump which
3342 jumps to this alternative if the former fails. */
3343 GET_BUFFER_SPACE (3);
3344 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3345 pending_exact = 0;
3346 b += 3;
3347
3348 /* The alternative before this one has a jump after it
3349 which gets executed if it gets matched. Adjust that
3350 jump so it will jump to this alternative's analogous
3351 jump (put in below, which in turn will jump to the next
3352 (if any) alternative's such jump, etc.). The last such
3353 jump jumps to the correct final destination. A picture:
3354 _____ _____
3355 | | | |
3356 | v | v
3357 a | b | c
3358
3359 If we are at `b', then fixup_alt_jump right now points to a
3360 three-byte space after `a'. We'll put in the jump, set
3361 fixup_alt_jump to right after `b', and leave behind three
3362 bytes which we'll fill in when we get to after `c'. */
3363
3364 FIXUP_ALT_JUMP ();
3365
3366 /* Mark and leave space for a jump after this alternative,
3367 to be filled in later either by next alternative or
3368 when know we're at the end of a series of alternatives. */
3369 fixup_alt_jump = b;
3370 GET_BUFFER_SPACE (3);
3371 b += 3;
3372
3373 laststart = 0;
3374 begalt = b;
3375 break;
3376
3377
3378 case '{':
3379 /* If \{ is a literal. */
3380 if (!(syntax & RE_INTERVALS)
3381 /* If we're at `\{' and it's not the open-interval
3382 operator. */
3383 || (syntax & RE_NO_BK_BRACES))
3384 goto normal_backslash;
3385
3386 handle_interval:
3387 {
3388 /* If got here, then the syntax allows intervals. */
3389
3390 /* At least (most) this many matches must be made. */
3391 int lower_bound = 0, upper_bound = -1;
3392
3393 beg_interval = p;
3394
3395 GET_UNSIGNED_NUMBER (lower_bound);
3396
3397 if (c == ',')
3398 GET_UNSIGNED_NUMBER (upper_bound);
3399 else
3400 /* Interval such as `{1}' => match exactly once. */
3401 upper_bound = lower_bound;
3402
3403 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3404 || (upper_bound >= 0 && lower_bound > upper_bound))
3405 FREE_STACK_RETURN (REG_BADBR);
3406
3407 if (!(syntax & RE_NO_BK_BRACES))
3408 {
3409 if (c != '\\')
3410 FREE_STACK_RETURN (REG_BADBR);
3411 if (p == pend)
3412 FREE_STACK_RETURN (REG_EESCAPE);
3413 PATFETCH (c);
3414 }
3415
3416 if (c != '}')
3417 FREE_STACK_RETURN (REG_BADBR);
3418
3419 /* We just parsed a valid interval. */
3420
3421 /* If it's invalid to have no preceding re. */
3422 if (!laststart)
3423 {
3424 if (syntax & RE_CONTEXT_INVALID_OPS)
3425 FREE_STACK_RETURN (REG_BADRPT);
3426 else if (syntax & RE_CONTEXT_INDEP_OPS)
3427 laststart = b;
3428 else
3429 goto unfetch_interval;
3430 }
3431
3432 if (upper_bound == 0)
3433 /* If the upper bound is zero, just drop the sub pattern
3434 altogether. */
3435 b = laststart;
3436 else if (lower_bound == 1 && upper_bound == 1)
3437 /* Just match it once: nothing to do here. */
3438 ;
3439
3440 /* Otherwise, we have a nontrivial interval. When
3441 we're all done, the pattern will look like:
3442 set_number_at <jump count> <upper bound>
3443 set_number_at <succeed_n count> <lower bound>
3444 succeed_n <after jump addr> <succeed_n count>
3445 <body of loop>
3446 jump_n <succeed_n addr> <jump count>
3447 (The upper bound and `jump_n' are omitted if
3448 `upper_bound' is 1, though.) */
3449 else
3450 { /* If the upper bound is > 1, we need to insert
3451 more at the end of the loop. */
3452 unsigned int nbytes = (upper_bound < 0 ? 3
3453 : upper_bound > 1 ? 5 : 0);
3454 unsigned int startoffset = 0;
3455
3456 GET_BUFFER_SPACE (20); /* We might use less. */
3457
3458 if (lower_bound == 0)
3459 {
3460 /* A succeed_n that starts with 0 is really a
3461 a simple on_failure_jump_loop. */
3462 INSERT_JUMP (on_failure_jump_loop, laststart,
3463 b + 3 + nbytes);
3464 b += 3;
3465 }
3466 else
3467 {
3468 /* Initialize lower bound of the `succeed_n', even
3469 though it will be set during matching by its
3470 attendant `set_number_at' (inserted next),
3471 because `re_compile_fastmap' needs to know.
3472 Jump to the `jump_n' we might insert below. */
3473 INSERT_JUMP2 (succeed_n, laststart,
3474 b + 5 + nbytes,
3475 lower_bound);
3476 b += 5;
3477
3478 /* Code to initialize the lower bound. Insert
3479 before the `succeed_n'. The `5' is the last two
3480 bytes of this `set_number_at', plus 3 bytes of
3481 the following `succeed_n'. */
3482 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3483 b += 5;
3484 startoffset += 5;
3485 }
3486
3487 if (upper_bound < 0)
3488 {
3489 /* A negative upper bound stands for infinity,
3490 in which case it degenerates to a plain jump. */
3491 STORE_JUMP (jump, b, laststart + startoffset);
3492 b += 3;
3493 }
3494 else if (upper_bound > 1)
3495 { /* More than one repetition is allowed, so
3496 append a backward jump to the `succeed_n'
3497 that starts this interval.
3498
3499 When we've reached this during matching,
3500 we'll have matched the interval once, so
3501 jump back only `upper_bound - 1' times. */
3502 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3503 upper_bound - 1);
3504 b += 5;
3505
3506 /* The location we want to set is the second
3507 parameter of the `jump_n'; that is `b-2' as
3508 an absolute address. `laststart' will be
3509 the `set_number_at' we're about to insert;
3510 `laststart+3' the number to set, the source
3511 for the relative address. But we are
3512 inserting into the middle of the pattern --
3513 so everything is getting moved up by 5.
3514 Conclusion: (b - 2) - (laststart + 3) + 5,
3515 i.e., b - laststart.
3516
3517 We insert this at the beginning of the loop
3518 so that if we fail during matching, we'll
3519 reinitialize the bounds. */
3520 insert_op2 (set_number_at, laststart, b - laststart,
3521 upper_bound - 1, b);
3522 b += 5;
3523 }
3524 }
3525 pending_exact = 0;
3526 beg_interval = NULL;
3527 }
3528 break;
3529
3530 unfetch_interval:
3531 /* If an invalid interval, match the characters as literals. */
3532 assert (beg_interval);
3533 p = beg_interval;
3534 beg_interval = NULL;
3535
3536 /* normal_char and normal_backslash need `c'. */
3537 c = '{';
3538
3539 if (!(syntax & RE_NO_BK_BRACES))
3540 {
3541 assert (p > pattern && p[-1] == '\\');
3542 goto normal_backslash;
3543 }
3544 else
3545 goto normal_char;
3546
3547 #ifdef emacs
3548 /* There is no way to specify the before_dot and after_dot
3549 operators. rms says this is ok. --karl */
3550 case '=':
3551 BUF_PUSH (at_dot);
3552 break;
3553
3554 case 's':
3555 laststart = b;
3556 PATFETCH (c);
3557 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3558 break;
3559
3560 case 'S':
3561 laststart = b;
3562 PATFETCH (c);
3563 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3564 break;
3565
3566 case 'c':
3567 laststart = b;
3568 PATFETCH (c);
3569 BUF_PUSH_2 (categoryspec, c);
3570 break;
3571
3572 case 'C':
3573 laststart = b;
3574 PATFETCH (c);
3575 BUF_PUSH_2 (notcategoryspec, c);
3576 break;
3577 #endif /* emacs */
3578
3579
3580 case 'w':
3581 if (syntax & RE_NO_GNU_OPS)
3582 goto normal_char;
3583 laststart = b;
3584 BUF_PUSH_2 (syntaxspec, Sword);
3585 break;
3586
3587
3588 case 'W':
3589 if (syntax & RE_NO_GNU_OPS)
3590 goto normal_char;
3591 laststart = b;
3592 BUF_PUSH_2 (notsyntaxspec, Sword);
3593 break;
3594
3595
3596 case '<':
3597 if (syntax & RE_NO_GNU_OPS)
3598 goto normal_char;
3599 BUF_PUSH (wordbeg);
3600 break;
3601
3602 case '>':
3603 if (syntax & RE_NO_GNU_OPS)
3604 goto normal_char;
3605 BUF_PUSH (wordend);
3606 break;
3607
3608 case '_':
3609 if (syntax & RE_NO_GNU_OPS)
3610 goto normal_char;
3611 laststart = b;
3612 PATFETCH (c);
3613 if (c == '<')
3614 BUF_PUSH (symbeg);
3615 else if (c == '>')
3616 BUF_PUSH (symend);
3617 else
3618 FREE_STACK_RETURN (REG_BADPAT);
3619 break;
3620
3621 case 'b':
3622 if (syntax & RE_NO_GNU_OPS)
3623 goto normal_char;
3624 BUF_PUSH (wordbound);
3625 break;
3626
3627 case 'B':
3628 if (syntax & RE_NO_GNU_OPS)
3629 goto normal_char;
3630 BUF_PUSH (notwordbound);
3631 break;
3632
3633 case '`':
3634 if (syntax & RE_NO_GNU_OPS)
3635 goto normal_char;
3636 BUF_PUSH (begbuf);
3637 break;
3638
3639 case '\'':
3640 if (syntax & RE_NO_GNU_OPS)
3641 goto normal_char;
3642 BUF_PUSH (endbuf);
3643 break;
3644
3645 case '1': case '2': case '3': case '4': case '5':
3646 case '6': case '7': case '8': case '9':
3647 {
3648 regnum_t reg;
3649
3650 if (syntax & RE_NO_BK_REFS)
3651 goto normal_backslash;
3652
3653 reg = c - '0';
3654
3655 if (reg > bufp->re_nsub || reg < 1
3656 /* Can't back reference to a subexp before its end. */
3657 || group_in_compile_stack (compile_stack, reg))
3658 FREE_STACK_RETURN (REG_ESUBREG);
3659
3660 laststart = b;
3661 BUF_PUSH_2 (duplicate, reg);
3662 }
3663 break;
3664
3665
3666 case '+':
3667 case '?':
3668 if (syntax & RE_BK_PLUS_QM)
3669 goto handle_plus;
3670 else
3671 goto normal_backslash;
3672
3673 default:
3674 normal_backslash:
3675 /* You might think it would be useful for \ to mean
3676 not to translate; but if we don't translate it
3677 it will never match anything. */
3678 goto normal_char;
3679 }
3680 break;
3681
3682
3683 default:
3684 /* Expects the character in `c'. */
3685 normal_char:
3686 /* If no exactn currently being built. */
3687 if (!pending_exact
3688
3689 /* If last exactn not at current position. */
3690 || pending_exact + *pending_exact + 1 != b
3691
3692 /* We have only one byte following the exactn for the count. */
3693 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3694
3695 /* If followed by a repetition operator. */
3696 || (p != pend && (*p == '*' || *p == '^'))
3697 || ((syntax & RE_BK_PLUS_QM)
3698 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3699 : p != pend && (*p == '+' || *p == '?'))
3700 || ((syntax & RE_INTERVALS)
3701 && ((syntax & RE_NO_BK_BRACES)
3702 ? p != pend && *p == '{'
3703 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3704 {
3705 /* Start building a new exactn. */
3706
3707 laststart = b;
3708
3709 BUF_PUSH_2 (exactn, 0);
3710 pending_exact = b - 1;
3711 }
3712
3713 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3714 {
3715 int len;
3716
3717 if (multibyte)
3718 {
3719 c = TRANSLATE (c);
3720 len = CHAR_STRING (c, b);
3721 b += len;
3722 }
3723 else
3724 {
3725 c1 = RE_CHAR_TO_MULTIBYTE (c);
3726 if (! CHAR_BYTE8_P (c1))
3727 {
3728 re_wchar_t c2 = TRANSLATE (c1);
3729
3730 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3731 c = c1;
3732 }
3733 *b++ = c;
3734 len = 1;
3735 }
3736 (*pending_exact) += len;
3737 }
3738
3739 break;
3740 } /* switch (c) */
3741 } /* while p != pend */
3742
3743
3744 /* Through the pattern now. */
3745
3746 FIXUP_ALT_JUMP ();
3747
3748 if (!COMPILE_STACK_EMPTY)
3749 FREE_STACK_RETURN (REG_EPAREN);
3750
3751 /* If we don't want backtracking, force success
3752 the first time we reach the end of the compiled pattern. */
3753 if (syntax & RE_NO_POSIX_BACKTRACKING)
3754 BUF_PUSH (succeed);
3755
3756 /* We have succeeded; set the length of the buffer. */
3757 bufp->used = b - bufp->buffer;
3758
3759 #ifdef DEBUG
3760 if (debug > 0)
3761 {
3762 re_compile_fastmap (bufp);
3763 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3764 print_compiled_pattern (bufp);
3765 }
3766 debug--;
3767 #endif /* DEBUG */
3768
3769 #ifndef MATCH_MAY_ALLOCATE
3770 /* Initialize the failure stack to the largest possible stack. This
3771 isn't necessary unless we're trying to avoid calling alloca in
3772 the search and match routines. */
3773 {
3774 int num_regs = bufp->re_nsub + 1;
3775
3776 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3777 {
3778 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3779
3780 if (! fail_stack.stack)
3781 fail_stack.stack
3782 = (fail_stack_elt_t *) malloc (fail_stack.size
3783 * sizeof (fail_stack_elt_t));
3784 else
3785 fail_stack.stack
3786 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3787 (fail_stack.size
3788 * sizeof (fail_stack_elt_t)));
3789 }
3790
3791 regex_grow_registers (num_regs);
3792 }
3793 #endif /* not MATCH_MAY_ALLOCATE */
3794
3795 FREE_STACK_RETURN (REG_NOERROR);
3796 } /* regex_compile */
3797 \f
3798 /* Subroutines for `regex_compile'. */
3799
3800 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3801
3802 static void
3803 store_op1 (re_opcode_t op, unsigned char *loc, int arg)
3804 {
3805 *loc = (unsigned char) op;
3806 STORE_NUMBER (loc + 1, arg);
3807 }
3808
3809
3810 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3811
3812 static void
3813 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2)
3814 {
3815 *loc = (unsigned char) op;
3816 STORE_NUMBER (loc + 1, arg1);
3817 STORE_NUMBER (loc + 3, arg2);
3818 }
3819
3820
3821 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3822 for OP followed by two-byte integer parameter ARG. */
3823
3824 static void
3825 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end)
3826 {
3827 register unsigned char *pfrom = end;
3828 register unsigned char *pto = end + 3;
3829
3830 while (pfrom != loc)
3831 *--pto = *--pfrom;
3832
3833 store_op1 (op, loc, arg);
3834 }
3835
3836
3837 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3838
3839 static void
3840 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end)
3841 {
3842 register unsigned char *pfrom = end;
3843 register unsigned char *pto = end + 5;
3844
3845 while (pfrom != loc)
3846 *--pto = *--pfrom;
3847
3848 store_op2 (op, loc, arg1, arg2);
3849 }
3850
3851
3852 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3853 after an alternative or a begin-subexpression. We assume there is at
3854 least one character before the ^. */
3855
3856 static boolean
3857 at_begline_loc_p (const re_char *pattern, const re_char *p, reg_syntax_t syntax)
3858 {
3859 re_char *prev = p - 2;
3860 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3861
3862 return
3863 /* After a subexpression? */
3864 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3865 /* After an alternative? */
3866 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3867 /* After a shy subexpression? */
3868 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3869 && prev[-1] == '?' && prev[-2] == '('
3870 && (syntax & RE_NO_BK_PARENS
3871 || (prev - 3 >= pattern && prev[-3] == '\\')));
3872 }
3873
3874
3875 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3876 at least one character after the $, i.e., `P < PEND'. */
3877
3878 static boolean
3879 at_endline_loc_p (const re_char *p, const re_char *pend, reg_syntax_t syntax)
3880 {
3881 re_char *next = p;
3882 boolean next_backslash = *next == '\\';
3883 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3884
3885 return
3886 /* Before a subexpression? */
3887 (syntax & RE_NO_BK_PARENS ? *next == ')'
3888 : next_backslash && next_next && *next_next == ')')
3889 /* Before an alternative? */
3890 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3891 : next_backslash && next_next && *next_next == '|');
3892 }
3893
3894
3895 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3896 false if it's not. */
3897
3898 static boolean
3899 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
3900 {
3901 int this_element;
3902
3903 for (this_element = compile_stack.avail - 1;
3904 this_element >= 0;
3905 this_element--)
3906 if (compile_stack.stack[this_element].regnum == regnum)
3907 return true;
3908
3909 return false;
3910 }
3911 \f
3912 /* analyse_first.
3913 If fastmap is non-NULL, go through the pattern and fill fastmap
3914 with all the possible leading chars. If fastmap is NULL, don't
3915 bother filling it up (obviously) and only return whether the
3916 pattern could potentially match the empty string.
3917
3918 Return 1 if p..pend might match the empty string.
3919 Return 0 if p..pend matches at least one char.
3920 Return -1 if fastmap was not updated accurately. */
3921
3922 static int
3923 analyse_first (const re_char *p, const re_char *pend, char *fastmap, const int multibyte)
3924 {
3925 int j, k;
3926 boolean not;
3927
3928 /* If all elements for base leading-codes in fastmap is set, this
3929 flag is set true. */
3930 boolean match_any_multibyte_characters = false;
3931
3932 assert (p);
3933
3934 /* The loop below works as follows:
3935 - It has a working-list kept in the PATTERN_STACK and which basically
3936 starts by only containing a pointer to the first operation.
3937 - If the opcode we're looking at is a match against some set of
3938 chars, then we add those chars to the fastmap and go on to the
3939 next work element from the worklist (done via `break').
3940 - If the opcode is a control operator on the other hand, we either
3941 ignore it (if it's meaningless at this point, such as `start_memory')
3942 or execute it (if it's a jump). If the jump has several destinations
3943 (i.e. `on_failure_jump'), then we push the other destination onto the
3944 worklist.
3945 We guarantee termination by ignoring backward jumps (more or less),
3946 so that `p' is monotonically increasing. More to the point, we
3947 never set `p' (or push) anything `<= p1'. */
3948
3949 while (p < pend)
3950 {
3951 /* `p1' is used as a marker of how far back a `on_failure_jump'
3952 can go without being ignored. It is normally equal to `p'
3953 (which prevents any backward `on_failure_jump') except right
3954 after a plain `jump', to allow patterns such as:
3955 0: jump 10
3956 3..9: <body>
3957 10: on_failure_jump 3
3958 as used for the *? operator. */
3959 re_char *p1 = p;
3960
3961 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3962 {
3963 case succeed:
3964 return 1;
3965
3966 case duplicate:
3967 /* If the first character has to match a backreference, that means
3968 that the group was empty (since it already matched). Since this
3969 is the only case that interests us here, we can assume that the
3970 backreference must match the empty string. */
3971 p++;
3972 continue;
3973
3974
3975 /* Following are the cases which match a character. These end
3976 with `break'. */
3977
3978 case exactn:
3979 if (fastmap)
3980 {
3981 /* If multibyte is nonzero, the first byte of each
3982 character is an ASCII or a leading code. Otherwise,
3983 each byte is a character. Thus, this works in both
3984 cases. */
3985 fastmap[p[1]] = 1;
3986 if (! multibyte)
3987 {
3988 /* For the case of matching this unibyte regex
3989 against multibyte, we must set a leading code of
3990 the corresponding multibyte character. */
3991 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
3992
3993 fastmap[CHAR_LEADING_CODE (c)] = 1;
3994 }
3995 }
3996 break;
3997
3998
3999 case anychar:
4000 /* We could put all the chars except for \n (and maybe \0)
4001 but we don't bother since it is generally not worth it. */
4002 if (!fastmap) break;
4003 return -1;
4004
4005
4006 case charset_not:
4007 if (!fastmap) break;
4008 {
4009 /* Chars beyond end of bitmap are possible matches. */
4010 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
4011 j < (1 << BYTEWIDTH); j++)
4012 fastmap[j] = 1;
4013 }
4014
4015 /* Fallthrough */
4016 case charset:
4017 if (!fastmap) break;
4018 not = (re_opcode_t) *(p - 1) == charset_not;
4019 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
4020 j >= 0; j--)
4021 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
4022 fastmap[j] = 1;
4023
4024 #ifdef emacs
4025 if (/* Any leading code can possibly start a character
4026 which doesn't match the specified set of characters. */
4027 not
4028 ||
4029 /* If we can match a character class, we can match any
4030 multibyte characters. */
4031 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4032 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4033
4034 {
4035 if (match_any_multibyte_characters == false)
4036 {
4037 for (j = MIN_MULTIBYTE_LEADING_CODE;
4038 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4039 fastmap[j] = 1;
4040 match_any_multibyte_characters = true;
4041 }
4042 }
4043
4044 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4045 && match_any_multibyte_characters == false)
4046 {
4047 /* Set fastmap[I] to 1 where I is a leading code of each
4048 multibyte character in the range table. */
4049 int c, count;
4050 unsigned char lc1, lc2;
4051
4052 /* Make P points the range table. `+ 2' is to skip flag
4053 bits for a character class. */
4054 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4055
4056 /* Extract the number of ranges in range table into COUNT. */
4057 EXTRACT_NUMBER_AND_INCR (count, p);
4058 for (; count > 0; count--, p += 3)
4059 {
4060 /* Extract the start and end of each range. */
4061 EXTRACT_CHARACTER (c, p);
4062 lc1 = CHAR_LEADING_CODE (c);
4063 p += 3;
4064 EXTRACT_CHARACTER (c, p);
4065 lc2 = CHAR_LEADING_CODE (c);
4066 for (j = lc1; j <= lc2; j++)
4067 fastmap[j] = 1;
4068 }
4069 }
4070 #endif
4071 break;
4072
4073 case syntaxspec:
4074 case notsyntaxspec:
4075 if (!fastmap) break;
4076 #ifndef emacs
4077 not = (re_opcode_t)p[-1] == notsyntaxspec;
4078 k = *p++;
4079 for (j = 0; j < (1 << BYTEWIDTH); j++)
4080 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4081 fastmap[j] = 1;
4082 break;
4083 #else /* emacs */
4084 /* This match depends on text properties. These end with
4085 aborting optimizations. */
4086 return -1;
4087
4088 case categoryspec:
4089 case notcategoryspec:
4090 if (!fastmap) break;
4091 not = (re_opcode_t)p[-1] == notcategoryspec;
4092 k = *p++;
4093 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4094 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4095 fastmap[j] = 1;
4096
4097 /* Any leading code can possibly start a character which
4098 has or doesn't has the specified category. */
4099 if (match_any_multibyte_characters == false)
4100 {
4101 for (j = MIN_MULTIBYTE_LEADING_CODE;
4102 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4103 fastmap[j] = 1;
4104 match_any_multibyte_characters = true;
4105 }
4106 break;
4107
4108 /* All cases after this match the empty string. These end with
4109 `continue'. */
4110
4111 case before_dot:
4112 case at_dot:
4113 case after_dot:
4114 #endif /* !emacs */
4115 case no_op:
4116 case begline:
4117 case endline:
4118 case begbuf:
4119 case endbuf:
4120 case wordbound:
4121 case notwordbound:
4122 case wordbeg:
4123 case wordend:
4124 case symbeg:
4125 case symend:
4126 continue;
4127
4128
4129 case jump:
4130 EXTRACT_NUMBER_AND_INCR (j, p);
4131 if (j < 0)
4132 /* Backward jumps can only go back to code that we've already
4133 visited. `re_compile' should make sure this is true. */
4134 break;
4135 p += j;
4136 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4137 {
4138 case on_failure_jump:
4139 case on_failure_keep_string_jump:
4140 case on_failure_jump_loop:
4141 case on_failure_jump_nastyloop:
4142 case on_failure_jump_smart:
4143 p++;
4144 break;
4145 default:
4146 continue;
4147 };
4148 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4149 to jump back to "just after here". */
4150 /* Fallthrough */
4151
4152 case on_failure_jump:
4153 case on_failure_keep_string_jump:
4154 case on_failure_jump_nastyloop:
4155 case on_failure_jump_loop:
4156 case on_failure_jump_smart:
4157 EXTRACT_NUMBER_AND_INCR (j, p);
4158 if (p + j <= p1)
4159 ; /* Backward jump to be ignored. */
4160 else
4161 { /* We have to look down both arms.
4162 We first go down the "straight" path so as to minimize
4163 stack usage when going through alternatives. */
4164 int r = analyse_first (p, pend, fastmap, multibyte);
4165 if (r) return r;
4166 p += j;
4167 }
4168 continue;
4169
4170
4171 case jump_n:
4172 /* This code simply does not properly handle forward jump_n. */
4173 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4174 p += 4;
4175 /* jump_n can either jump or fall through. The (backward) jump
4176 case has already been handled, so we only need to look at the
4177 fallthrough case. */
4178 continue;
4179
4180 case succeed_n:
4181 /* If N == 0, it should be an on_failure_jump_loop instead. */
4182 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4183 p += 4;
4184 /* We only care about one iteration of the loop, so we don't
4185 need to consider the case where this behaves like an
4186 on_failure_jump. */
4187 continue;
4188
4189
4190 case set_number_at:
4191 p += 4;
4192 continue;
4193
4194
4195 case start_memory:
4196 case stop_memory:
4197 p += 1;
4198 continue;
4199
4200
4201 default:
4202 abort (); /* We have listed all the cases. */
4203 } /* switch *p++ */
4204
4205 /* Getting here means we have found the possible starting
4206 characters for one path of the pattern -- and that the empty
4207 string does not match. We need not follow this path further. */
4208 return 0;
4209 } /* while p */
4210
4211 /* We reached the end without matching anything. */
4212 return 1;
4213
4214 } /* analyse_first */
4215 \f
4216 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4217 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4218 characters can start a string that matches the pattern. This fastmap
4219 is used by re_search to skip quickly over impossible starting points.
4220
4221 Character codes above (1 << BYTEWIDTH) are not represented in the
4222 fastmap, but the leading codes are represented. Thus, the fastmap
4223 indicates which character sets could start a match.
4224
4225 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4226 area as BUFP->fastmap.
4227
4228 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4229 the pattern buffer.
4230
4231 Returns 0 if we succeed, -2 if an internal error. */
4232
4233 int
4234 re_compile_fastmap (struct re_pattern_buffer *bufp)
4235 {
4236 char *fastmap = bufp->fastmap;
4237 int analysis;
4238
4239 assert (fastmap && bufp->buffer);
4240
4241 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4242 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4243
4244 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4245 fastmap, RE_MULTIBYTE_P (bufp));
4246 bufp->can_be_null = (analysis != 0);
4247 return 0;
4248 } /* re_compile_fastmap */
4249 \f
4250 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4251 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4252 this memory for recording register information. STARTS and ENDS
4253 must be allocated using the malloc library routine, and must each
4254 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4255
4256 If NUM_REGS == 0, then subsequent matches should allocate their own
4257 register data.
4258
4259 Unless this function is called, the first search or match using
4260 PATTERN_BUFFER will allocate its own register data, without
4261 freeing the old data. */
4262
4263 void
4264 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4265 {
4266 if (num_regs)
4267 {
4268 bufp->regs_allocated = REGS_REALLOCATE;
4269 regs->num_regs = num_regs;
4270 regs->start = starts;
4271 regs->end = ends;
4272 }
4273 else
4274 {
4275 bufp->regs_allocated = REGS_UNALLOCATED;
4276 regs->num_regs = 0;
4277 regs->start = regs->end = (regoff_t *) 0;
4278 }
4279 }
4280 WEAK_ALIAS (__re_set_registers, re_set_registers)
4281 \f
4282 /* Searching routines. */
4283
4284 /* Like re_search_2, below, but only one string is specified, and
4285 doesn't let you say where to stop matching. */
4286
4287 int
4288 re_search (struct re_pattern_buffer *bufp, const char *string, int size, int startpos, int range, struct re_registers *regs)
4289 {
4290 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4291 regs, size);
4292 }
4293 WEAK_ALIAS (__re_search, re_search)
4294
4295 /* Head address of virtual concatenation of string. */
4296 #define HEAD_ADDR_VSTRING(P) \
4297 (((P) >= size1 ? string2 : string1))
4298
4299 /* Address of POS in the concatenation of virtual string. */
4300 #define POS_ADDR_VSTRING(POS) \
4301 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4302
4303 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4304 virtual concatenation of STRING1 and STRING2, starting first at index
4305 STARTPOS, then at STARTPOS + 1, and so on.
4306
4307 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4308
4309 RANGE is how far to scan while trying to match. RANGE = 0 means try
4310 only at STARTPOS; in general, the last start tried is STARTPOS +
4311 RANGE.
4312
4313 In REGS, return the indices of the virtual concatenation of STRING1
4314 and STRING2 that matched the entire BUFP->buffer and its contained
4315 subexpressions.
4316
4317 Do not consider matching one past the index STOP in the virtual
4318 concatenation of STRING1 and STRING2.
4319
4320 We return either the position in the strings at which the match was
4321 found, -1 if no match, or -2 if error (such as failure
4322 stack overflow). */
4323
4324 int
4325 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, int size1, const char *str2, int size2, int startpos, int range, struct re_registers *regs, int stop)
4326 {
4327 int val;
4328 re_char *string1 = (re_char*) str1;
4329 re_char *string2 = (re_char*) str2;
4330 register char *fastmap = bufp->fastmap;
4331 register RE_TRANSLATE_TYPE translate = bufp->translate;
4332 int total_size = size1 + size2;
4333 int endpos = startpos + range;
4334 boolean anchored_start;
4335 /* Nonzero if we are searching multibyte string. */
4336 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4337
4338 /* Check for out-of-range STARTPOS. */
4339 if (startpos < 0 || startpos > total_size)
4340 return -1;
4341
4342 /* Fix up RANGE if it might eventually take us outside
4343 the virtual concatenation of STRING1 and STRING2.
4344 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4345 if (endpos < 0)
4346 range = 0 - startpos;
4347 else if (endpos > total_size)
4348 range = total_size - startpos;
4349
4350 /* If the search isn't to be a backwards one, don't waste time in a
4351 search for a pattern anchored at beginning of buffer. */
4352 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4353 {
4354 if (startpos > 0)
4355 return -1;
4356 else
4357 range = 0;
4358 }
4359
4360 #ifdef emacs
4361 /* In a forward search for something that starts with \=.
4362 don't keep searching past point. */
4363 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4364 {
4365 range = PT_BYTE - BEGV_BYTE - startpos;
4366 if (range < 0)
4367 return -1;
4368 }
4369 #endif /* emacs */
4370
4371 /* Update the fastmap now if not correct already. */
4372 if (fastmap && !bufp->fastmap_accurate)
4373 re_compile_fastmap (bufp);
4374
4375 /* See whether the pattern is anchored. */
4376 anchored_start = (bufp->buffer[0] == begline);
4377
4378 #ifdef emacs
4379 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4380 {
4381 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4382
4383 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4384 }
4385 #endif
4386
4387 /* Loop through the string, looking for a place to start matching. */
4388 for (;;)
4389 {
4390 /* If the pattern is anchored,
4391 skip quickly past places we cannot match.
4392 We don't bother to treat startpos == 0 specially
4393 because that case doesn't repeat. */
4394 if (anchored_start && startpos > 0)
4395 {
4396 if (! ((startpos <= size1 ? string1[startpos - 1]
4397 : string2[startpos - size1 - 1])
4398 == '\n'))
4399 goto advance;
4400 }
4401
4402 /* If a fastmap is supplied, skip quickly over characters that
4403 cannot be the start of a match. If the pattern can match the
4404 null string, however, we don't need to skip characters; we want
4405 the first null string. */
4406 if (fastmap && startpos < total_size && !bufp->can_be_null)
4407 {
4408 register re_char *d;
4409 register re_wchar_t buf_ch;
4410
4411 d = POS_ADDR_VSTRING (startpos);
4412
4413 if (range > 0) /* Searching forwards. */
4414 {
4415 register int lim = 0;
4416 int irange = range;
4417
4418 if (startpos < size1 && startpos + range >= size1)
4419 lim = range - (size1 - startpos);
4420
4421 /* Written out as an if-else to avoid testing `translate'
4422 inside the loop. */
4423 if (RE_TRANSLATE_P (translate))
4424 {
4425 if (multibyte)
4426 while (range > lim)
4427 {
4428 int buf_charlen;
4429
4430 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4431 buf_ch = RE_TRANSLATE (translate, buf_ch);
4432 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4433 break;
4434
4435 range -= buf_charlen;
4436 d += buf_charlen;
4437 }
4438 else
4439 while (range > lim)
4440 {
4441 register re_wchar_t ch, translated;
4442
4443 buf_ch = *d;
4444 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4445 translated = RE_TRANSLATE (translate, ch);
4446 if (translated != ch
4447 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4448 buf_ch = ch;
4449 if (fastmap[buf_ch])
4450 break;
4451 d++;
4452 range--;
4453 }
4454 }
4455 else
4456 {
4457 if (multibyte)
4458 while (range > lim)
4459 {
4460 int buf_charlen;
4461
4462 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4463 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4464 break;
4465 range -= buf_charlen;
4466 d += buf_charlen;
4467 }
4468 else
4469 while (range > lim && !fastmap[*d])
4470 {
4471 d++;
4472 range--;
4473 }
4474 }
4475 startpos += irange - range;
4476 }
4477 else /* Searching backwards. */
4478 {
4479 if (multibyte)
4480 {
4481 buf_ch = STRING_CHAR (d);
4482 buf_ch = TRANSLATE (buf_ch);
4483 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4484 goto advance;
4485 }
4486 else
4487 {
4488 register re_wchar_t ch, translated;
4489
4490 buf_ch = *d;
4491 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4492 translated = TRANSLATE (ch);
4493 if (translated != ch
4494 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4495 buf_ch = ch;
4496 if (! fastmap[TRANSLATE (buf_ch)])
4497 goto advance;
4498 }
4499 }
4500 }
4501
4502 /* If can't match the null string, and that's all we have left, fail. */
4503 if (range >= 0 && startpos == total_size && fastmap
4504 && !bufp->can_be_null)
4505 return -1;
4506
4507 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4508 startpos, regs, stop);
4509
4510 if (val >= 0)
4511 return startpos;
4512
4513 if (val == -2)
4514 return -2;
4515
4516 advance:
4517 if (!range)
4518 break;
4519 else if (range > 0)
4520 {
4521 /* Update STARTPOS to the next character boundary. */
4522 if (multibyte)
4523 {
4524 re_char *p = POS_ADDR_VSTRING (startpos);
4525 int len = BYTES_BY_CHAR_HEAD (*p);
4526
4527 range -= len;
4528 if (range < 0)
4529 break;
4530 startpos += len;
4531 }
4532 else
4533 {
4534 range--;
4535 startpos++;
4536 }
4537 }
4538 else
4539 {
4540 range++;
4541 startpos--;
4542
4543 /* Update STARTPOS to the previous character boundary. */
4544 if (multibyte)
4545 {
4546 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4547 re_char *p0 = p;
4548 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4549
4550 /* Find the head of multibyte form. */
4551 PREV_CHAR_BOUNDARY (p, phead);
4552 range += p0 - 1 - p;
4553 if (range > 0)
4554 break;
4555
4556 startpos -= p0 - 1 - p;
4557 }
4558 }
4559 }
4560 return -1;
4561 } /* re_search_2 */
4562 WEAK_ALIAS (__re_search_2, re_search_2)
4563 \f
4564 /* Declarations and macros for re_match_2. */
4565
4566 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4567 register int len,
4568 RE_TRANSLATE_TYPE translate,
4569 const int multibyte));
4570
4571 /* This converts PTR, a pointer into one of the search strings `string1'
4572 and `string2' into an offset from the beginning of that string. */
4573 #define POINTER_TO_OFFSET(ptr) \
4574 (FIRST_STRING_P (ptr) \
4575 ? ((regoff_t) ((ptr) - string1)) \
4576 : ((regoff_t) ((ptr) - string2 + size1)))
4577
4578 /* Call before fetching a character with *d. This switches over to
4579 string2 if necessary.
4580 Check re_match_2_internal for a discussion of why end_match_2 might
4581 not be within string2 (but be equal to end_match_1 instead). */
4582 #define PREFETCH() \
4583 while (d == dend) \
4584 { \
4585 /* End of string2 => fail. */ \
4586 if (dend == end_match_2) \
4587 goto fail; \
4588 /* End of string1 => advance to string2. */ \
4589 d = string2; \
4590 dend = end_match_2; \
4591 }
4592
4593 /* Call before fetching a char with *d if you already checked other limits.
4594 This is meant for use in lookahead operations like wordend, etc..
4595 where we might need to look at parts of the string that might be
4596 outside of the LIMITs (i.e past `stop'). */
4597 #define PREFETCH_NOLIMIT() \
4598 if (d == end1) \
4599 { \
4600 d = string2; \
4601 dend = end_match_2; \
4602 } \
4603
4604 /* Test if at very beginning or at very end of the virtual concatenation
4605 of `string1' and `string2'. If only one string, it's `string2'. */
4606 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4607 #define AT_STRINGS_END(d) ((d) == end2)
4608
4609 /* Disabled due to a compiler bug -- see comment at case wordbound */
4610
4611 /* The comment at case wordbound is following one, but we don't use
4612 AT_WORD_BOUNDARY anymore to support multibyte form.
4613
4614 The DEC Alpha C compiler 3.x generates incorrect code for the
4615 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4616 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4617 macro and introducing temporary variables works around the bug. */
4618
4619 #if 0
4620 /* Test if D points to a character which is word-constituent. We have
4621 two special cases to check for: if past the end of string1, look at
4622 the first character in string2; and if before the beginning of
4623 string2, look at the last character in string1. */
4624 #define WORDCHAR_P(d) \
4625 (SYNTAX ((d) == end1 ? *string2 \
4626 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4627 == Sword)
4628
4629 /* Test if the character before D and the one at D differ with respect
4630 to being word-constituent. */
4631 #define AT_WORD_BOUNDARY(d) \
4632 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4633 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4634 #endif
4635
4636 /* Free everything we malloc. */
4637 #ifdef MATCH_MAY_ALLOCATE
4638 # define FREE_VAR(var) \
4639 do { \
4640 if (var) \
4641 { \
4642 REGEX_FREE (var); \
4643 var = NULL; \
4644 } \
4645 } while (0)
4646 # define FREE_VARIABLES() \
4647 do { \
4648 REGEX_FREE_STACK (fail_stack.stack); \
4649 FREE_VAR (regstart); \
4650 FREE_VAR (regend); \
4651 FREE_VAR (best_regstart); \
4652 FREE_VAR (best_regend); \
4653 } while (0)
4654 #else
4655 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4656 #endif /* not MATCH_MAY_ALLOCATE */
4657
4658 \f
4659 /* Optimization routines. */
4660
4661 /* If the operation is a match against one or more chars,
4662 return a pointer to the next operation, else return NULL. */
4663 static re_char *
4664 skip_one_char (const re_char *p)
4665 {
4666 switch (SWITCH_ENUM_CAST (*p++))
4667 {
4668 case anychar:
4669 break;
4670
4671 case exactn:
4672 p += *p + 1;
4673 break;
4674
4675 case charset_not:
4676 case charset:
4677 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4678 {
4679 int mcnt;
4680 p = CHARSET_RANGE_TABLE (p - 1);
4681 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4682 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4683 }
4684 else
4685 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4686 break;
4687
4688 case syntaxspec:
4689 case notsyntaxspec:
4690 #ifdef emacs
4691 case categoryspec:
4692 case notcategoryspec:
4693 #endif /* emacs */
4694 p++;
4695 break;
4696
4697 default:
4698 p = NULL;
4699 }
4700 return p;
4701 }
4702
4703
4704 /* Jump over non-matching operations. */
4705 static re_char *
4706 skip_noops (const re_char *p, const re_char *pend)
4707 {
4708 int mcnt;
4709 while (p < pend)
4710 {
4711 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4712 {
4713 case start_memory:
4714 case stop_memory:
4715 p += 2; break;
4716 case no_op:
4717 p += 1; break;
4718 case jump:
4719 p += 1;
4720 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4721 p += mcnt;
4722 break;
4723 default:
4724 return p;
4725 }
4726 }
4727 assert (p == pend);
4728 return p;
4729 }
4730
4731 /* Non-zero if "p1 matches something" implies "p2 fails". */
4732 static int
4733 mutually_exclusive_p (struct re_pattern_buffer *bufp, const re_char *p1, const re_char *p2)
4734 {
4735 re_opcode_t op2;
4736 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4737 unsigned char *pend = bufp->buffer + bufp->used;
4738
4739 assert (p1 >= bufp->buffer && p1 < pend
4740 && p2 >= bufp->buffer && p2 <= pend);
4741
4742 /* Skip over open/close-group commands.
4743 If what follows this loop is a ...+ construct,
4744 look at what begins its body, since we will have to
4745 match at least one of that. */
4746 p2 = skip_noops (p2, pend);
4747 /* The same skip can be done for p1, except that this function
4748 is only used in the case where p1 is a simple match operator. */
4749 /* p1 = skip_noops (p1, pend); */
4750
4751 assert (p1 >= bufp->buffer && p1 < pend
4752 && p2 >= bufp->buffer && p2 <= pend);
4753
4754 op2 = p2 == pend ? succeed : *p2;
4755
4756 switch (SWITCH_ENUM_CAST (op2))
4757 {
4758 case succeed:
4759 case endbuf:
4760 /* If we're at the end of the pattern, we can change. */
4761 if (skip_one_char (p1))
4762 {
4763 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4764 return 1;
4765 }
4766 break;
4767
4768 case endline:
4769 case exactn:
4770 {
4771 register re_wchar_t c
4772 = (re_opcode_t) *p2 == endline ? '\n'
4773 : RE_STRING_CHAR (p2 + 2, multibyte);
4774
4775 if ((re_opcode_t) *p1 == exactn)
4776 {
4777 if (c != RE_STRING_CHAR (p1 + 2, multibyte))
4778 {
4779 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4780 return 1;
4781 }
4782 }
4783
4784 else if ((re_opcode_t) *p1 == charset
4785 || (re_opcode_t) *p1 == charset_not)
4786 {
4787 int not = (re_opcode_t) *p1 == charset_not;
4788
4789 /* Test if C is listed in charset (or charset_not)
4790 at `p1'. */
4791 if (! multibyte || IS_REAL_ASCII (c))
4792 {
4793 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4794 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4795 not = !not;
4796 }
4797 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4798 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4799
4800 /* `not' is equal to 1 if c would match, which means
4801 that we can't change to pop_failure_jump. */
4802 if (!not)
4803 {
4804 DEBUG_PRINT1 (" No match => fast loop.\n");
4805 return 1;
4806 }
4807 }
4808 else if ((re_opcode_t) *p1 == anychar
4809 && c == '\n')
4810 {
4811 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4812 return 1;
4813 }
4814 }
4815 break;
4816
4817 case charset:
4818 {
4819 if ((re_opcode_t) *p1 == exactn)
4820 /* Reuse the code above. */
4821 return mutually_exclusive_p (bufp, p2, p1);
4822
4823 /* It is hard to list up all the character in charset
4824 P2 if it includes multibyte character. Give up in
4825 such case. */
4826 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4827 {
4828 /* Now, we are sure that P2 has no range table.
4829 So, for the size of bitmap in P2, `p2[1]' is
4830 enough. But P1 may have range table, so the
4831 size of bitmap table of P1 is extracted by
4832 using macro `CHARSET_BITMAP_SIZE'.
4833
4834 In a multibyte case, we know that all the character
4835 listed in P2 is ASCII. In a unibyte case, P1 has only a
4836 bitmap table. So, in both cases, it is enough to test
4837 only the bitmap table of P1. */
4838
4839 if ((re_opcode_t) *p1 == charset)
4840 {
4841 int idx;
4842 /* We win if the charset inside the loop
4843 has no overlap with the one after the loop. */
4844 for (idx = 0;
4845 (idx < (int) p2[1]
4846 && idx < CHARSET_BITMAP_SIZE (p1));
4847 idx++)
4848 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4849 break;
4850
4851 if (idx == p2[1]
4852 || idx == CHARSET_BITMAP_SIZE (p1))
4853 {
4854 DEBUG_PRINT1 (" No match => fast loop.\n");
4855 return 1;
4856 }
4857 }
4858 else if ((re_opcode_t) *p1 == charset_not)
4859 {
4860 int idx;
4861 /* We win if the charset_not inside the loop lists
4862 every character listed in the charset after. */
4863 for (idx = 0; idx < (int) p2[1]; idx++)
4864 if (! (p2[2 + idx] == 0
4865 || (idx < CHARSET_BITMAP_SIZE (p1)
4866 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4867 break;
4868
4869 if (idx == p2[1])
4870 {
4871 DEBUG_PRINT1 (" No match => fast loop.\n");
4872 return 1;
4873 }
4874 }
4875 }
4876 }
4877 break;
4878
4879 case charset_not:
4880 switch (SWITCH_ENUM_CAST (*p1))
4881 {
4882 case exactn:
4883 case charset:
4884 /* Reuse the code above. */
4885 return mutually_exclusive_p (bufp, p2, p1);
4886 case charset_not:
4887 /* When we have two charset_not, it's very unlikely that
4888 they don't overlap. The union of the two sets of excluded
4889 chars should cover all possible chars, which, as a matter of
4890 fact, is virtually impossible in multibyte buffers. */
4891 break;
4892 }
4893 break;
4894
4895 case wordend:
4896 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
4897 case symend:
4898 return ((re_opcode_t) *p1 == syntaxspec
4899 && (p1[1] == Ssymbol || p1[1] == Sword));
4900 case notsyntaxspec:
4901 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
4902
4903 case wordbeg:
4904 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
4905 case symbeg:
4906 return ((re_opcode_t) *p1 == notsyntaxspec
4907 && (p1[1] == Ssymbol || p1[1] == Sword));
4908 case syntaxspec:
4909 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
4910
4911 case wordbound:
4912 return (((re_opcode_t) *p1 == notsyntaxspec
4913 || (re_opcode_t) *p1 == syntaxspec)
4914 && p1[1] == Sword);
4915
4916 #ifdef emacs
4917 case categoryspec:
4918 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4919 case notcategoryspec:
4920 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4921 #endif /* emacs */
4922
4923 default:
4924 ;
4925 }
4926
4927 /* Safe default. */
4928 return 0;
4929 }
4930
4931 \f
4932 /* Matching routines. */
4933
4934 #ifndef emacs /* Emacs never uses this. */
4935 /* re_match is like re_match_2 except it takes only a single string. */
4936
4937 int
4938 re_match (struct re_pattern_buffer *bufp, const char *string,
4939 int size, int pos, struct re_registers *regs)
4940 {
4941 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
4942 pos, regs, size);
4943 return result;
4944 }
4945 WEAK_ALIAS (__re_match, re_match)
4946 #endif /* not emacs */
4947
4948 #ifdef emacs
4949 /* In Emacs, this is the string or buffer in which we
4950 are matching. It is used for looking up syntax properties. */
4951 Lisp_Object re_match_object;
4952 #endif
4953
4954 /* re_match_2 matches the compiled pattern in BUFP against the
4955 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4956 and SIZE2, respectively). We start matching at POS, and stop
4957 matching at STOP.
4958
4959 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4960 store offsets for the substring each group matched in REGS. See the
4961 documentation for exactly how many groups we fill.
4962
4963 We return -1 if no match, -2 if an internal error (such as the
4964 failure stack overflowing). Otherwise, we return the length of the
4965 matched substring. */
4966
4967 int
4968 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, const char *string2, int size2, int pos, struct re_registers *regs, int stop)
4969 {
4970 int result;
4971
4972 #ifdef emacs
4973 int charpos;
4974 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4975 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4976 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4977 #endif
4978
4979 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4980 (re_char*) string2, size2,
4981 pos, regs, stop);
4982 return result;
4983 }
4984 WEAK_ALIAS (__re_match_2, re_match_2)
4985
4986
4987 /* This is a separate function so that we can force an alloca cleanup
4988 afterwards. */
4989 static int
4990 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1, int size1, const re_char *string2, int size2, int pos, struct re_registers *regs, int stop)
4991 {
4992 /* General temporaries. */
4993 int mcnt;
4994 size_t reg;
4995
4996 /* Just past the end of the corresponding string. */
4997 re_char *end1, *end2;
4998
4999 /* Pointers into string1 and string2, just past the last characters in
5000 each to consider matching. */
5001 re_char *end_match_1, *end_match_2;
5002
5003 /* Where we are in the data, and the end of the current string. */
5004 re_char *d, *dend;
5005
5006 /* Used sometimes to remember where we were before starting matching
5007 an operator so that we can go back in case of failure. This "atomic"
5008 behavior of matching opcodes is indispensable to the correctness
5009 of the on_failure_keep_string_jump optimization. */
5010 re_char *dfail;
5011
5012 /* Where we are in the pattern, and the end of the pattern. */
5013 re_char *p = bufp->buffer;
5014 re_char *pend = p + bufp->used;
5015
5016 /* We use this to map every character in the string. */
5017 RE_TRANSLATE_TYPE translate = bufp->translate;
5018
5019 /* Nonzero if BUFP is setup from a multibyte regex. */
5020 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5021
5022 /* Nonzero if STRING1/STRING2 are multibyte. */
5023 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5024
5025 /* Failure point stack. Each place that can handle a failure further
5026 down the line pushes a failure point on this stack. It consists of
5027 regstart, and regend for all registers corresponding to
5028 the subexpressions we're currently inside, plus the number of such
5029 registers, and, finally, two char *'s. The first char * is where
5030 to resume scanning the pattern; the second one is where to resume
5031 scanning the strings. */
5032 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5033 fail_stack_type fail_stack;
5034 #endif
5035 #ifdef DEBUG
5036 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5037 #endif
5038
5039 #if defined REL_ALLOC && defined REGEX_MALLOC
5040 /* This holds the pointer to the failure stack, when
5041 it is allocated relocatably. */
5042 fail_stack_elt_t *failure_stack_ptr;
5043 #endif
5044
5045 /* We fill all the registers internally, independent of what we
5046 return, for use in backreferences. The number here includes
5047 an element for register zero. */
5048 size_t num_regs = bufp->re_nsub + 1;
5049
5050 /* Information on the contents of registers. These are pointers into
5051 the input strings; they record just what was matched (on this
5052 attempt) by a subexpression part of the pattern, that is, the
5053 regnum-th regstart pointer points to where in the pattern we began
5054 matching and the regnum-th regend points to right after where we
5055 stopped matching the regnum-th subexpression. (The zeroth register
5056 keeps track of what the whole pattern matches.) */
5057 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5058 re_char **regstart, **regend;
5059 #endif
5060
5061 /* The following record the register info as found in the above
5062 variables when we find a match better than any we've seen before.
5063 This happens as we backtrack through the failure points, which in
5064 turn happens only if we have not yet matched the entire string. */
5065 unsigned best_regs_set = false;
5066 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5067 re_char **best_regstart, **best_regend;
5068 #endif
5069
5070 /* Logically, this is `best_regend[0]'. But we don't want to have to
5071 allocate space for that if we're not allocating space for anything
5072 else (see below). Also, we never need info about register 0 for
5073 any of the other register vectors, and it seems rather a kludge to
5074 treat `best_regend' differently than the rest. So we keep track of
5075 the end of the best match so far in a separate variable. We
5076 initialize this to NULL so that when we backtrack the first time
5077 and need to test it, it's not garbage. */
5078 re_char *match_end = NULL;
5079
5080 #ifdef DEBUG
5081 /* Counts the total number of registers pushed. */
5082 unsigned num_regs_pushed = 0;
5083 #endif
5084
5085 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5086
5087 INIT_FAIL_STACK ();
5088
5089 #ifdef MATCH_MAY_ALLOCATE
5090 /* Do not bother to initialize all the register variables if there are
5091 no groups in the pattern, as it takes a fair amount of time. If
5092 there are groups, we include space for register 0 (the whole
5093 pattern), even though we never use it, since it simplifies the
5094 array indexing. We should fix this. */
5095 if (bufp->re_nsub)
5096 {
5097 regstart = REGEX_TALLOC (num_regs, re_char *);
5098 regend = REGEX_TALLOC (num_regs, re_char *);
5099 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5100 best_regend = REGEX_TALLOC (num_regs, re_char *);
5101
5102 if (!(regstart && regend && best_regstart && best_regend))
5103 {
5104 FREE_VARIABLES ();
5105 return -2;
5106 }
5107 }
5108 else
5109 {
5110 /* We must initialize all our variables to NULL, so that
5111 `FREE_VARIABLES' doesn't try to free them. */
5112 regstart = regend = best_regstart = best_regend = NULL;
5113 }
5114 #endif /* MATCH_MAY_ALLOCATE */
5115
5116 /* The starting position is bogus. */
5117 if (pos < 0 || pos > size1 + size2)
5118 {
5119 FREE_VARIABLES ();
5120 return -1;
5121 }
5122
5123 /* Initialize subexpression text positions to -1 to mark ones that no
5124 start_memory/stop_memory has been seen for. Also initialize the
5125 register information struct. */
5126 for (reg = 1; reg < num_regs; reg++)
5127 regstart[reg] = regend[reg] = NULL;
5128
5129 /* We move `string1' into `string2' if the latter's empty -- but not if
5130 `string1' is null. */
5131 if (size2 == 0 && string1 != NULL)
5132 {
5133 string2 = string1;
5134 size2 = size1;
5135 string1 = 0;
5136 size1 = 0;
5137 }
5138 end1 = string1 + size1;
5139 end2 = string2 + size2;
5140
5141 /* `p' scans through the pattern as `d' scans through the data.
5142 `dend' is the end of the input string that `d' points within. `d'
5143 is advanced into the following input string whenever necessary, but
5144 this happens before fetching; therefore, at the beginning of the
5145 loop, `d' can be pointing at the end of a string, but it cannot
5146 equal `string2'. */
5147 if (pos >= size1)
5148 {
5149 /* Only match within string2. */
5150 d = string2 + pos - size1;
5151 dend = end_match_2 = string2 + stop - size1;
5152 end_match_1 = end1; /* Just to give it a value. */
5153 }
5154 else
5155 {
5156 if (stop < size1)
5157 {
5158 /* Only match within string1. */
5159 end_match_1 = string1 + stop;
5160 /* BEWARE!
5161 When we reach end_match_1, PREFETCH normally switches to string2.
5162 But in the present case, this means that just doing a PREFETCH
5163 makes us jump from `stop' to `gap' within the string.
5164 What we really want here is for the search to stop as
5165 soon as we hit end_match_1. That's why we set end_match_2
5166 to end_match_1 (since PREFETCH fails as soon as we hit
5167 end_match_2). */
5168 end_match_2 = end_match_1;
5169 }
5170 else
5171 { /* It's important to use this code when stop == size so that
5172 moving `d' from end1 to string2 will not prevent the d == dend
5173 check from catching the end of string. */
5174 end_match_1 = end1;
5175 end_match_2 = string2 + stop - size1;
5176 }
5177 d = string1 + pos;
5178 dend = end_match_1;
5179 }
5180
5181 DEBUG_PRINT1 ("The compiled pattern is: ");
5182 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5183 DEBUG_PRINT1 ("The string to match is: `");
5184 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5185 DEBUG_PRINT1 ("'\n");
5186
5187 /* This loops over pattern commands. It exits by returning from the
5188 function if the match is complete, or it drops through if the match
5189 fails at this starting point in the input data. */
5190 for (;;)
5191 {
5192 DEBUG_PRINT2 ("\n%p: ", p);
5193
5194 if (p == pend)
5195 { /* End of pattern means we might have succeeded. */
5196 DEBUG_PRINT1 ("end of pattern ... ");
5197
5198 /* If we haven't matched the entire string, and we want the
5199 longest match, try backtracking. */
5200 if (d != end_match_2)
5201 {
5202 /* 1 if this match ends in the same string (string1 or string2)
5203 as the best previous match. */
5204 boolean same_str_p = (FIRST_STRING_P (match_end)
5205 == FIRST_STRING_P (d));
5206 /* 1 if this match is the best seen so far. */
5207 boolean best_match_p;
5208
5209 /* AIX compiler got confused when this was combined
5210 with the previous declaration. */
5211 if (same_str_p)
5212 best_match_p = d > match_end;
5213 else
5214 best_match_p = !FIRST_STRING_P (d);
5215
5216 DEBUG_PRINT1 ("backtracking.\n");
5217
5218 if (!FAIL_STACK_EMPTY ())
5219 { /* More failure points to try. */
5220
5221 /* If exceeds best match so far, save it. */
5222 if (!best_regs_set || best_match_p)
5223 {
5224 best_regs_set = true;
5225 match_end = d;
5226
5227 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5228
5229 for (reg = 1; reg < num_regs; reg++)
5230 {
5231 best_regstart[reg] = regstart[reg];
5232 best_regend[reg] = regend[reg];
5233 }
5234 }
5235 goto fail;
5236 }
5237
5238 /* If no failure points, don't restore garbage. And if
5239 last match is real best match, don't restore second
5240 best one. */
5241 else if (best_regs_set && !best_match_p)
5242 {
5243 restore_best_regs:
5244 /* Restore best match. It may happen that `dend ==
5245 end_match_1' while the restored d is in string2.
5246 For example, the pattern `x.*y.*z' against the
5247 strings `x-' and `y-z-', if the two strings are
5248 not consecutive in memory. */
5249 DEBUG_PRINT1 ("Restoring best registers.\n");
5250
5251 d = match_end;
5252 dend = ((d >= string1 && d <= end1)
5253 ? end_match_1 : end_match_2);
5254
5255 for (reg = 1; reg < num_regs; reg++)
5256 {
5257 regstart[reg] = best_regstart[reg];
5258 regend[reg] = best_regend[reg];
5259 }
5260 }
5261 } /* d != end_match_2 */
5262
5263 succeed_label:
5264 DEBUG_PRINT1 ("Accepting match.\n");
5265
5266 /* If caller wants register contents data back, do it. */
5267 if (regs && !bufp->no_sub)
5268 {
5269 /* Have the register data arrays been allocated? */
5270 if (bufp->regs_allocated == REGS_UNALLOCATED)
5271 { /* No. So allocate them with malloc. We need one
5272 extra element beyond `num_regs' for the `-1' marker
5273 GNU code uses. */
5274 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5275 regs->start = TALLOC (regs->num_regs, regoff_t);
5276 regs->end = TALLOC (regs->num_regs, regoff_t);
5277 if (regs->start == NULL || regs->end == NULL)
5278 {
5279 FREE_VARIABLES ();
5280 return -2;
5281 }
5282 bufp->regs_allocated = REGS_REALLOCATE;
5283 }
5284 else if (bufp->regs_allocated == REGS_REALLOCATE)
5285 { /* Yes. If we need more elements than were already
5286 allocated, reallocate them. If we need fewer, just
5287 leave it alone. */
5288 if (regs->num_regs < num_regs + 1)
5289 {
5290 regs->num_regs = num_regs + 1;
5291 RETALLOC (regs->start, regs->num_regs, regoff_t);
5292 RETALLOC (regs->end, regs->num_regs, regoff_t);
5293 if (regs->start == NULL || regs->end == NULL)
5294 {
5295 FREE_VARIABLES ();
5296 return -2;
5297 }
5298 }
5299 }
5300 else
5301 {
5302 /* These braces fend off a "empty body in an else-statement"
5303 warning under GCC when assert expands to nothing. */
5304 assert (bufp->regs_allocated == REGS_FIXED);
5305 }
5306
5307 /* Convert the pointer data in `regstart' and `regend' to
5308 indices. Register zero has to be set differently,
5309 since we haven't kept track of any info for it. */
5310 if (regs->num_regs > 0)
5311 {
5312 regs->start[0] = pos;
5313 regs->end[0] = POINTER_TO_OFFSET (d);
5314 }
5315
5316 /* Go through the first `min (num_regs, regs->num_regs)'
5317 registers, since that is all we initialized. */
5318 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5319 {
5320 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5321 regs->start[reg] = regs->end[reg] = -1;
5322 else
5323 {
5324 regs->start[reg]
5325 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5326 regs->end[reg]
5327 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5328 }
5329 }
5330
5331 /* If the regs structure we return has more elements than
5332 were in the pattern, set the extra elements to -1. If
5333 we (re)allocated the registers, this is the case,
5334 because we always allocate enough to have at least one
5335 -1 at the end. */
5336 for (reg = num_regs; reg < regs->num_regs; reg++)
5337 regs->start[reg] = regs->end[reg] = -1;
5338 } /* regs && !bufp->no_sub */
5339
5340 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5341 nfailure_points_pushed, nfailure_points_popped,
5342 nfailure_points_pushed - nfailure_points_popped);
5343 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5344
5345 mcnt = POINTER_TO_OFFSET (d) - pos;
5346
5347 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5348
5349 FREE_VARIABLES ();
5350 return mcnt;
5351 }
5352
5353 /* Otherwise match next pattern command. */
5354 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5355 {
5356 /* Ignore these. Used to ignore the n of succeed_n's which
5357 currently have n == 0. */
5358 case no_op:
5359 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5360 break;
5361
5362 case succeed:
5363 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5364 goto succeed_label;
5365
5366 /* Match the next n pattern characters exactly. The following
5367 byte in the pattern defines n, and the n bytes after that
5368 are the characters to match. */
5369 case exactn:
5370 mcnt = *p++;
5371 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5372
5373 /* Remember the start point to rollback upon failure. */
5374 dfail = d;
5375
5376 #ifndef emacs
5377 /* This is written out as an if-else so we don't waste time
5378 testing `translate' inside the loop. */
5379 if (RE_TRANSLATE_P (translate))
5380 do
5381 {
5382 PREFETCH ();
5383 if (RE_TRANSLATE (translate, *d) != *p++)
5384 {
5385 d = dfail;
5386 goto fail;
5387 }
5388 d++;
5389 }
5390 while (--mcnt);
5391 else
5392 do
5393 {
5394 PREFETCH ();
5395 if (*d++ != *p++)
5396 {
5397 d = dfail;
5398 goto fail;
5399 }
5400 }
5401 while (--mcnt);
5402 #else /* emacs */
5403 /* The cost of testing `translate' is comparatively small. */
5404 if (target_multibyte)
5405 do
5406 {
5407 int pat_charlen, buf_charlen;
5408 int pat_ch, buf_ch;
5409
5410 PREFETCH ();
5411 if (multibyte)
5412 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5413 else
5414 {
5415 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5416 pat_charlen = 1;
5417 }
5418 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5419
5420 if (TRANSLATE (buf_ch) != pat_ch)
5421 {
5422 d = dfail;
5423 goto fail;
5424 }
5425
5426 p += pat_charlen;
5427 d += buf_charlen;
5428 mcnt -= pat_charlen;
5429 }
5430 while (mcnt > 0);
5431 else
5432 do
5433 {
5434 int pat_charlen;
5435 int pat_ch, buf_ch;
5436
5437 PREFETCH ();
5438 if (multibyte)
5439 {
5440 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5441 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5442 }
5443 else
5444 {
5445 pat_ch = *p;
5446 pat_charlen = 1;
5447 }
5448 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5449 if (! CHAR_BYTE8_P (buf_ch))
5450 {
5451 buf_ch = TRANSLATE (buf_ch);
5452 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5453 if (buf_ch < 0)
5454 buf_ch = *d;
5455 }
5456 else
5457 buf_ch = *d;
5458 if (buf_ch != pat_ch)
5459 {
5460 d = dfail;
5461 goto fail;
5462 }
5463 p += pat_charlen;
5464 d++;
5465 }
5466 while (--mcnt);
5467 #endif
5468 break;
5469
5470
5471 /* Match any character except possibly a newline or a null. */
5472 case anychar:
5473 {
5474 int buf_charlen;
5475 re_wchar_t buf_ch;
5476
5477 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5478
5479 PREFETCH ();
5480 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5481 target_multibyte);
5482 buf_ch = TRANSLATE (buf_ch);
5483
5484 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5485 && buf_ch == '\n')
5486 || ((bufp->syntax & RE_DOT_NOT_NULL)
5487 && buf_ch == '\000'))
5488 goto fail;
5489
5490 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5491 d += buf_charlen;
5492 }
5493 break;
5494
5495
5496 case charset:
5497 case charset_not:
5498 {
5499 register unsigned int c;
5500 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5501 int len;
5502
5503 /* Start of actual range_table, or end of bitmap if there is no
5504 range table. */
5505 re_char *range_table IF_LINT (= NULL);
5506
5507 /* Nonzero if there is a range table. */
5508 int range_table_exists;
5509
5510 /* Number of ranges of range table. This is not included
5511 in the initial byte-length of the command. */
5512 int count = 0;
5513
5514 /* Whether matching against a unibyte character. */
5515 boolean unibyte_char = false;
5516
5517 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5518
5519 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5520
5521 if (range_table_exists)
5522 {
5523 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5524 EXTRACT_NUMBER_AND_INCR (count, range_table);
5525 }
5526
5527 PREFETCH ();
5528 c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5529 if (target_multibyte)
5530 {
5531 int c1;
5532
5533 c = TRANSLATE (c);
5534 c1 = RE_CHAR_TO_UNIBYTE (c);
5535 if (c1 >= 0)
5536 {
5537 unibyte_char = true;
5538 c = c1;
5539 }
5540 }
5541 else
5542 {
5543 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5544
5545 if (! CHAR_BYTE8_P (c1))
5546 {
5547 c1 = TRANSLATE (c1);
5548 c1 = RE_CHAR_TO_UNIBYTE (c1);
5549 if (c1 >= 0)
5550 {
5551 unibyte_char = true;
5552 c = c1;
5553 }
5554 }
5555 else
5556 unibyte_char = true;
5557 }
5558
5559 if (unibyte_char && c < (1 << BYTEWIDTH))
5560 { /* Lookup bitmap. */
5561 /* Cast to `unsigned' instead of `unsigned char' in
5562 case the bit list is a full 32 bytes long. */
5563 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5564 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5565 not = !not;
5566 }
5567 #ifdef emacs
5568 else if (range_table_exists)
5569 {
5570 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5571
5572 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5573 | (class_bits & BIT_MULTIBYTE)
5574 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5575 | (class_bits & BIT_SPACE && ISSPACE (c))
5576 | (class_bits & BIT_UPPER && ISUPPER (c))
5577 | (class_bits & BIT_WORD && ISWORD (c)))
5578 not = !not;
5579 else
5580 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5581 }
5582 #endif /* emacs */
5583
5584 if (range_table_exists)
5585 p = CHARSET_RANGE_TABLE_END (range_table, count);
5586 else
5587 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5588
5589 if (!not) goto fail;
5590
5591 d += len;
5592 }
5593 break;
5594
5595
5596 /* The beginning of a group is represented by start_memory.
5597 The argument is the register number. The text
5598 matched within the group is recorded (in the internal
5599 registers data structure) under the register number. */
5600 case start_memory:
5601 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5602
5603 /* In case we need to undo this operation (via backtracking). */
5604 PUSH_FAILURE_REG ((unsigned int)*p);
5605
5606 regstart[*p] = d;
5607 regend[*p] = NULL; /* probably unnecessary. -sm */
5608 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5609
5610 /* Move past the register number and inner group count. */
5611 p += 1;
5612 break;
5613
5614
5615 /* The stop_memory opcode represents the end of a group. Its
5616 argument is the same as start_memory's: the register number. */
5617 case stop_memory:
5618 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5619
5620 assert (!REG_UNSET (regstart[*p]));
5621 /* Strictly speaking, there should be code such as:
5622
5623 assert (REG_UNSET (regend[*p]));
5624 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5625
5626 But the only info to be pushed is regend[*p] and it is known to
5627 be UNSET, so there really isn't anything to push.
5628 Not pushing anything, on the other hand deprives us from the
5629 guarantee that regend[*p] is UNSET since undoing this operation
5630 will not reset its value properly. This is not important since
5631 the value will only be read on the next start_memory or at
5632 the very end and both events can only happen if this stop_memory
5633 is *not* undone. */
5634
5635 regend[*p] = d;
5636 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5637
5638 /* Move past the register number and the inner group count. */
5639 p += 1;
5640 break;
5641
5642
5643 /* \<digit> has been turned into a `duplicate' command which is
5644 followed by the numeric value of <digit> as the register number. */
5645 case duplicate:
5646 {
5647 register re_char *d2, *dend2;
5648 int regno = *p++; /* Get which register to match against. */
5649 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5650
5651 /* Can't back reference a group which we've never matched. */
5652 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5653 goto fail;
5654
5655 /* Where in input to try to start matching. */
5656 d2 = regstart[regno];
5657
5658 /* Remember the start point to rollback upon failure. */
5659 dfail = d;
5660
5661 /* Where to stop matching; if both the place to start and
5662 the place to stop matching are in the same string, then
5663 set to the place to stop, otherwise, for now have to use
5664 the end of the first string. */
5665
5666 dend2 = ((FIRST_STRING_P (regstart[regno])
5667 == FIRST_STRING_P (regend[regno]))
5668 ? regend[regno] : end_match_1);
5669 for (;;)
5670 {
5671 /* If necessary, advance to next segment in register
5672 contents. */
5673 while (d2 == dend2)
5674 {
5675 if (dend2 == end_match_2) break;
5676 if (dend2 == regend[regno]) break;
5677
5678 /* End of string1 => advance to string2. */
5679 d2 = string2;
5680 dend2 = regend[regno];
5681 }
5682 /* At end of register contents => success */
5683 if (d2 == dend2) break;
5684
5685 /* If necessary, advance to next segment in data. */
5686 PREFETCH ();
5687
5688 /* How many characters left in this segment to match. */
5689 mcnt = dend - d;
5690
5691 /* Want how many consecutive characters we can match in
5692 one shot, so, if necessary, adjust the count. */
5693 if (mcnt > dend2 - d2)
5694 mcnt = dend2 - d2;
5695
5696 /* Compare that many; failure if mismatch, else move
5697 past them. */
5698 if (RE_TRANSLATE_P (translate)
5699 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5700 : memcmp (d, d2, mcnt))
5701 {
5702 d = dfail;
5703 goto fail;
5704 }
5705 d += mcnt, d2 += mcnt;
5706 }
5707 }
5708 break;
5709
5710
5711 /* begline matches the empty string at the beginning of the string
5712 (unless `not_bol' is set in `bufp'), and after newlines. */
5713 case begline:
5714 DEBUG_PRINT1 ("EXECUTING begline.\n");
5715
5716 if (AT_STRINGS_BEG (d))
5717 {
5718 if (!bufp->not_bol) break;
5719 }
5720 else
5721 {
5722 unsigned c;
5723 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5724 if (c == '\n')
5725 break;
5726 }
5727 /* In all other cases, we fail. */
5728 goto fail;
5729
5730
5731 /* endline is the dual of begline. */
5732 case endline:
5733 DEBUG_PRINT1 ("EXECUTING endline.\n");
5734
5735 if (AT_STRINGS_END (d))
5736 {
5737 if (!bufp->not_eol) break;
5738 }
5739 else
5740 {
5741 PREFETCH_NOLIMIT ();
5742 if (*d == '\n')
5743 break;
5744 }
5745 goto fail;
5746
5747
5748 /* Match at the very beginning of the data. */
5749 case begbuf:
5750 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5751 if (AT_STRINGS_BEG (d))
5752 break;
5753 goto fail;
5754
5755
5756 /* Match at the very end of the data. */
5757 case endbuf:
5758 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5759 if (AT_STRINGS_END (d))
5760 break;
5761 goto fail;
5762
5763
5764 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5765 pushes NULL as the value for the string on the stack. Then
5766 `POP_FAILURE_POINT' will keep the current value for the
5767 string, instead of restoring it. To see why, consider
5768 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5769 then the . fails against the \n. But the next thing we want
5770 to do is match the \n against the \n; if we restored the
5771 string value, we would be back at the foo.
5772
5773 Because this is used only in specific cases, we don't need to
5774 check all the things that `on_failure_jump' does, to make
5775 sure the right things get saved on the stack. Hence we don't
5776 share its code. The only reason to push anything on the
5777 stack at all is that otherwise we would have to change
5778 `anychar's code to do something besides goto fail in this
5779 case; that seems worse than this. */
5780 case on_failure_keep_string_jump:
5781 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5782 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5783 mcnt, p + mcnt);
5784
5785 PUSH_FAILURE_POINT (p - 3, NULL);
5786 break;
5787
5788 /* A nasty loop is introduced by the non-greedy *? and +?.
5789 With such loops, the stack only ever contains one failure point
5790 at a time, so that a plain on_failure_jump_loop kind of
5791 cycle detection cannot work. Worse yet, such a detection
5792 can not only fail to detect a cycle, but it can also wrongly
5793 detect a cycle (between different instantiations of the same
5794 loop).
5795 So the method used for those nasty loops is a little different:
5796 We use a special cycle-detection-stack-frame which is pushed
5797 when the on_failure_jump_nastyloop failure-point is *popped*.
5798 This special frame thus marks the beginning of one iteration
5799 through the loop and we can hence easily check right here
5800 whether something matched between the beginning and the end of
5801 the loop. */
5802 case on_failure_jump_nastyloop:
5803 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5804 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5805 mcnt, p + mcnt);
5806
5807 assert ((re_opcode_t)p[-4] == no_op);
5808 {
5809 int cycle = 0;
5810 CHECK_INFINITE_LOOP (p - 4, d);
5811 if (!cycle)
5812 /* If there's a cycle, just continue without pushing
5813 this failure point. The failure point is the "try again"
5814 option, which shouldn't be tried.
5815 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5816 PUSH_FAILURE_POINT (p - 3, d);
5817 }
5818 break;
5819
5820 /* Simple loop detecting on_failure_jump: just check on the
5821 failure stack if the same spot was already hit earlier. */
5822 case on_failure_jump_loop:
5823 on_failure:
5824 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5825 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5826 mcnt, p + mcnt);
5827 {
5828 int cycle = 0;
5829 CHECK_INFINITE_LOOP (p - 3, d);
5830 if (cycle)
5831 /* If there's a cycle, get out of the loop, as if the matching
5832 had failed. We used to just `goto fail' here, but that was
5833 aborting the search a bit too early: we want to keep the
5834 empty-loop-match and keep matching after the loop.
5835 We want (x?)*y\1z to match both xxyz and xxyxz. */
5836 p += mcnt;
5837 else
5838 PUSH_FAILURE_POINT (p - 3, d);
5839 }
5840 break;
5841
5842
5843 /* Uses of on_failure_jump:
5844
5845 Each alternative starts with an on_failure_jump that points
5846 to the beginning of the next alternative. Each alternative
5847 except the last ends with a jump that in effect jumps past
5848 the rest of the alternatives. (They really jump to the
5849 ending jump of the following alternative, because tensioning
5850 these jumps is a hassle.)
5851
5852 Repeats start with an on_failure_jump that points past both
5853 the repetition text and either the following jump or
5854 pop_failure_jump back to this on_failure_jump. */
5855 case on_failure_jump:
5856 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5857 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5858 mcnt, p + mcnt);
5859
5860 PUSH_FAILURE_POINT (p -3, d);
5861 break;
5862
5863 /* This operation is used for greedy *.
5864 Compare the beginning of the repeat with what in the
5865 pattern follows its end. If we can establish that there
5866 is nothing that they would both match, i.e., that we
5867 would have to backtrack because of (as in, e.g., `a*a')
5868 then we can use a non-backtracking loop based on
5869 on_failure_keep_string_jump instead of on_failure_jump. */
5870 case on_failure_jump_smart:
5871 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5872 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5873 mcnt, p + mcnt);
5874 {
5875 re_char *p1 = p; /* Next operation. */
5876 /* Here, we discard `const', making re_match non-reentrant. */
5877 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5878 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5879
5880 p -= 3; /* Reset so that we will re-execute the
5881 instruction once it's been changed. */
5882
5883 EXTRACT_NUMBER (mcnt, p2 - 2);
5884
5885 /* Ensure this is a indeed the trivial kind of loop
5886 we are expecting. */
5887 assert (skip_one_char (p1) == p2 - 3);
5888 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5889 DEBUG_STATEMENT (debug += 2);
5890 if (mutually_exclusive_p (bufp, p1, p2))
5891 {
5892 /* Use a fast `on_failure_keep_string_jump' loop. */
5893 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5894 *p3 = (unsigned char) on_failure_keep_string_jump;
5895 STORE_NUMBER (p2 - 2, mcnt + 3);
5896 }
5897 else
5898 {
5899 /* Default to a safe `on_failure_jump' loop. */
5900 DEBUG_PRINT1 (" smart default => slow loop.\n");
5901 *p3 = (unsigned char) on_failure_jump;
5902 }
5903 DEBUG_STATEMENT (debug -= 2);
5904 }
5905 break;
5906
5907 /* Unconditionally jump (without popping any failure points). */
5908 case jump:
5909 unconditional_jump:
5910 IMMEDIATE_QUIT_CHECK;
5911 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5912 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5913 p += mcnt; /* Do the jump. */
5914 DEBUG_PRINT2 ("(to %p).\n", p);
5915 break;
5916
5917
5918 /* Have to succeed matching what follows at least n times.
5919 After that, handle like `on_failure_jump'. */
5920 case succeed_n:
5921 /* Signedness doesn't matter since we only compare MCNT to 0. */
5922 EXTRACT_NUMBER (mcnt, p + 2);
5923 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5924
5925 /* Originally, mcnt is how many times we HAVE to succeed. */
5926 if (mcnt != 0)
5927 {
5928 /* Here, we discard `const', making re_match non-reentrant. */
5929 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5930 mcnt--;
5931 p += 4;
5932 PUSH_NUMBER (p2, mcnt);
5933 }
5934 else
5935 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5936 goto on_failure;
5937 break;
5938
5939 case jump_n:
5940 /* Signedness doesn't matter since we only compare MCNT to 0. */
5941 EXTRACT_NUMBER (mcnt, p + 2);
5942 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5943
5944 /* Originally, this is how many times we CAN jump. */
5945 if (mcnt != 0)
5946 {
5947 /* Here, we discard `const', making re_match non-reentrant. */
5948 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5949 mcnt--;
5950 PUSH_NUMBER (p2, mcnt);
5951 goto unconditional_jump;
5952 }
5953 /* If don't have to jump any more, skip over the rest of command. */
5954 else
5955 p += 4;
5956 break;
5957
5958 case set_number_at:
5959 {
5960 unsigned char *p2; /* Location of the counter. */
5961 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5962
5963 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5964 /* Here, we discard `const', making re_match non-reentrant. */
5965 p2 = (unsigned char*) p + mcnt;
5966 /* Signedness doesn't matter since we only copy MCNT's bits . */
5967 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5968 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
5969 PUSH_NUMBER (p2, mcnt);
5970 break;
5971 }
5972
5973 case wordbound:
5974 case notwordbound:
5975 {
5976 boolean not = (re_opcode_t) *(p - 1) == notwordbound;
5977 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5978
5979 /* We SUCCEED (or FAIL) in one of the following cases: */
5980
5981 /* Case 1: D is at the beginning or the end of string. */
5982 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5983 not = !not;
5984 else
5985 {
5986 /* C1 is the character before D, S1 is the syntax of C1, C2
5987 is the character at D, and S2 is the syntax of C2. */
5988 re_wchar_t c1, c2;
5989 int s1, s2;
5990 int dummy;
5991 #ifdef emacs
5992 int offset = PTR_TO_OFFSET (d - 1);
5993 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5994 UPDATE_SYNTAX_TABLE (charpos);
5995 #endif
5996 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5997 s1 = SYNTAX (c1);
5998 #ifdef emacs
5999 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6000 #endif
6001 PREFETCH_NOLIMIT ();
6002 GET_CHAR_AFTER (c2, d, dummy);
6003 s2 = SYNTAX (c2);
6004
6005 if (/* Case 2: Only one of S1 and S2 is Sword. */
6006 ((s1 == Sword) != (s2 == Sword))
6007 /* Case 3: Both of S1 and S2 are Sword, and macro
6008 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6009 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
6010 not = !not;
6011 }
6012 if (not)
6013 break;
6014 else
6015 goto fail;
6016 }
6017
6018 case wordbeg:
6019 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6020
6021 /* We FAIL in one of the following cases: */
6022
6023 /* Case 1: D is at the end of string. */
6024 if (AT_STRINGS_END (d))
6025 goto fail;
6026 else
6027 {
6028 /* C1 is the character before D, S1 is the syntax of C1, C2
6029 is the character at D, and S2 is the syntax of C2. */
6030 re_wchar_t c1, c2;
6031 int s1, s2;
6032 int dummy;
6033 #ifdef emacs
6034 int offset = PTR_TO_OFFSET (d);
6035 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6036 UPDATE_SYNTAX_TABLE (charpos);
6037 #endif
6038 PREFETCH ();
6039 GET_CHAR_AFTER (c2, d, dummy);
6040 s2 = SYNTAX (c2);
6041
6042 /* Case 2: S2 is not Sword. */
6043 if (s2 != Sword)
6044 goto fail;
6045
6046 /* Case 3: D is not at the beginning of string ... */
6047 if (!AT_STRINGS_BEG (d))
6048 {
6049 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6050 #ifdef emacs
6051 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6052 #endif
6053 s1 = SYNTAX (c1);
6054
6055 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6056 returns 0. */
6057 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6058 goto fail;
6059 }
6060 }
6061 break;
6062
6063 case wordend:
6064 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6065
6066 /* We FAIL in one of the following cases: */
6067
6068 /* Case 1: D is at the beginning of string. */
6069 if (AT_STRINGS_BEG (d))
6070 goto fail;
6071 else
6072 {
6073 /* C1 is the character before D, S1 is the syntax of C1, C2
6074 is the character at D, and S2 is the syntax of C2. */
6075 re_wchar_t c1, c2;
6076 int s1, s2;
6077 int dummy;
6078 #ifdef emacs
6079 int offset = PTR_TO_OFFSET (d) - 1;
6080 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6081 UPDATE_SYNTAX_TABLE (charpos);
6082 #endif
6083 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6084 s1 = SYNTAX (c1);
6085
6086 /* Case 2: S1 is not Sword. */
6087 if (s1 != Sword)
6088 goto fail;
6089
6090 /* Case 3: D is not at the end of string ... */
6091 if (!AT_STRINGS_END (d))
6092 {
6093 PREFETCH_NOLIMIT ();
6094 GET_CHAR_AFTER (c2, d, dummy);
6095 #ifdef emacs
6096 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6097 #endif
6098 s2 = SYNTAX (c2);
6099
6100 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6101 returns 0. */
6102 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6103 goto fail;
6104 }
6105 }
6106 break;
6107
6108 case symbeg:
6109 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6110
6111 /* We FAIL in one of the following cases: */
6112
6113 /* Case 1: D is at the end of string. */
6114 if (AT_STRINGS_END (d))
6115 goto fail;
6116 else
6117 {
6118 /* C1 is the character before D, S1 is the syntax of C1, C2
6119 is the character at D, and S2 is the syntax of C2. */
6120 re_wchar_t c1, c2;
6121 int s1, s2;
6122 #ifdef emacs
6123 int offset = PTR_TO_OFFSET (d);
6124 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6125 UPDATE_SYNTAX_TABLE (charpos);
6126 #endif
6127 PREFETCH ();
6128 c2 = RE_STRING_CHAR (d, target_multibyte);
6129 s2 = SYNTAX (c2);
6130
6131 /* Case 2: S2 is neither Sword nor Ssymbol. */
6132 if (s2 != Sword && s2 != Ssymbol)
6133 goto fail;
6134
6135 /* Case 3: D is not at the beginning of string ... */
6136 if (!AT_STRINGS_BEG (d))
6137 {
6138 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6139 #ifdef emacs
6140 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6141 #endif
6142 s1 = SYNTAX (c1);
6143
6144 /* ... and S1 is Sword or Ssymbol. */
6145 if (s1 == Sword || s1 == Ssymbol)
6146 goto fail;
6147 }
6148 }
6149 break;
6150
6151 case symend:
6152 DEBUG_PRINT1 ("EXECUTING symend.\n");
6153
6154 /* We FAIL in one of the following cases: */
6155
6156 /* Case 1: D is at the beginning of string. */
6157 if (AT_STRINGS_BEG (d))
6158 goto fail;
6159 else
6160 {
6161 /* C1 is the character before D, S1 is the syntax of C1, C2
6162 is the character at D, and S2 is the syntax of C2. */
6163 re_wchar_t c1, c2;
6164 int s1, s2;
6165 #ifdef emacs
6166 int offset = PTR_TO_OFFSET (d) - 1;
6167 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6168 UPDATE_SYNTAX_TABLE (charpos);
6169 #endif
6170 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6171 s1 = SYNTAX (c1);
6172
6173 /* Case 2: S1 is neither Ssymbol nor Sword. */
6174 if (s1 != Sword && s1 != Ssymbol)
6175 goto fail;
6176
6177 /* Case 3: D is not at the end of string ... */
6178 if (!AT_STRINGS_END (d))
6179 {
6180 PREFETCH_NOLIMIT ();
6181 c2 = RE_STRING_CHAR (d, target_multibyte);
6182 #ifdef emacs
6183 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6184 #endif
6185 s2 = SYNTAX (c2);
6186
6187 /* ... and S2 is Sword or Ssymbol. */
6188 if (s2 == Sword || s2 == Ssymbol)
6189 goto fail;
6190 }
6191 }
6192 break;
6193
6194 case syntaxspec:
6195 case notsyntaxspec:
6196 {
6197 boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6198 mcnt = *p++;
6199 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6200 PREFETCH ();
6201 #ifdef emacs
6202 {
6203 int offset = PTR_TO_OFFSET (d);
6204 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6205 UPDATE_SYNTAX_TABLE (pos1);
6206 }
6207 #endif
6208 {
6209 int len;
6210 re_wchar_t c;
6211
6212 GET_CHAR_AFTER (c, d, len);
6213 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6214 goto fail;
6215 d += len;
6216 }
6217 }
6218 break;
6219
6220 #ifdef emacs
6221 case before_dot:
6222 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6223 if (PTR_BYTE_POS (d) >= PT_BYTE)
6224 goto fail;
6225 break;
6226
6227 case at_dot:
6228 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6229 if (PTR_BYTE_POS (d) != PT_BYTE)
6230 goto fail;
6231 break;
6232
6233 case after_dot:
6234 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6235 if (PTR_BYTE_POS (d) <= PT_BYTE)
6236 goto fail;
6237 break;
6238
6239 case categoryspec:
6240 case notcategoryspec:
6241 {
6242 boolean not = (re_opcode_t) *(p - 1) == notcategoryspec;
6243 mcnt = *p++;
6244 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n",
6245 not?"not":"", mcnt);
6246 PREFETCH ();
6247
6248 {
6249 int len;
6250 re_wchar_t c;
6251 GET_CHAR_AFTER (c, d, len);
6252 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6253 goto fail;
6254 d += len;
6255 }
6256 }
6257 break;
6258
6259 #endif /* emacs */
6260
6261 default:
6262 abort ();
6263 }
6264 continue; /* Successfully executed one pattern command; keep going. */
6265
6266
6267 /* We goto here if a matching operation fails. */
6268 fail:
6269 IMMEDIATE_QUIT_CHECK;
6270 if (!FAIL_STACK_EMPTY ())
6271 {
6272 re_char *str, *pat;
6273 /* A restart point is known. Restore to that state. */
6274 DEBUG_PRINT1 ("\nFAIL:\n");
6275 POP_FAILURE_POINT (str, pat);
6276 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6277 {
6278 case on_failure_keep_string_jump:
6279 assert (str == NULL);
6280 goto continue_failure_jump;
6281
6282 case on_failure_jump_nastyloop:
6283 assert ((re_opcode_t)pat[-2] == no_op);
6284 PUSH_FAILURE_POINT (pat - 2, str);
6285 /* Fallthrough */
6286
6287 case on_failure_jump_loop:
6288 case on_failure_jump:
6289 case succeed_n:
6290 d = str;
6291 continue_failure_jump:
6292 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6293 p = pat + mcnt;
6294 break;
6295
6296 case no_op:
6297 /* A special frame used for nastyloops. */
6298 goto fail;
6299
6300 default:
6301 abort();
6302 }
6303
6304 assert (p >= bufp->buffer && p <= pend);
6305
6306 if (d >= string1 && d <= end1)
6307 dend = end_match_1;
6308 }
6309 else
6310 break; /* Matching at this starting point really fails. */
6311 } /* for (;;) */
6312
6313 if (best_regs_set)
6314 goto restore_best_regs;
6315
6316 FREE_VARIABLES ();
6317
6318 return -1; /* Failure to match. */
6319 } /* re_match_2 */
6320 \f
6321 /* Subroutine definitions for re_match_2. */
6322
6323 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6324 bytes; nonzero otherwise. */
6325
6326 static int
6327 bcmp_translate (const re_char *s1, const re_char *s2, register int len,
6328 RE_TRANSLATE_TYPE translate, const int target_multibyte)
6329 {
6330 register re_char *p1 = s1, *p2 = s2;
6331 re_char *p1_end = s1 + len;
6332 re_char *p2_end = s2 + len;
6333
6334 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6335 different lengths, but relying on a single `len' would break this. -sm */
6336 while (p1 < p1_end && p2 < p2_end)
6337 {
6338 int p1_charlen, p2_charlen;
6339 re_wchar_t p1_ch, p2_ch;
6340
6341 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6342 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6343
6344 if (RE_TRANSLATE (translate, p1_ch)
6345 != RE_TRANSLATE (translate, p2_ch))
6346 return 1;
6347
6348 p1 += p1_charlen, p2 += p2_charlen;
6349 }
6350
6351 if (p1 != p1_end || p2 != p2_end)
6352 return 1;
6353
6354 return 0;
6355 }
6356 \f
6357 /* Entry points for GNU code. */
6358
6359 /* re_compile_pattern is the GNU regular expression compiler: it
6360 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6361 Returns 0 if the pattern was valid, otherwise an error string.
6362
6363 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6364 are set in BUFP on entry.
6365
6366 We call regex_compile to do the actual compilation. */
6367
6368 const char *
6369 re_compile_pattern (const char *pattern, size_t length, struct re_pattern_buffer *bufp)
6370 {
6371 reg_errcode_t ret;
6372
6373 /* GNU code is written to assume at least RE_NREGS registers will be set
6374 (and at least one extra will be -1). */
6375 bufp->regs_allocated = REGS_UNALLOCATED;
6376
6377 /* And GNU code determines whether or not to get register information
6378 by passing null for the REGS argument to re_match, etc., not by
6379 setting no_sub. */
6380 bufp->no_sub = 0;
6381
6382 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6383
6384 if (!ret)
6385 return NULL;
6386 return gettext (re_error_msgid[(int) ret]);
6387 }
6388 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6389 \f
6390 /* Entry points compatible with 4.2 BSD regex library. We don't define
6391 them unless specifically requested. */
6392
6393 #if defined _REGEX_RE_COMP || defined _LIBC
6394
6395 /* BSD has one and only one pattern buffer. */
6396 static struct re_pattern_buffer re_comp_buf;
6397
6398 char *
6399 # ifdef _LIBC
6400 /* Make these definitions weak in libc, so POSIX programs can redefine
6401 these names if they don't use our functions, and still use
6402 regcomp/regexec below without link errors. */
6403 weak_function
6404 # endif
6405 re_comp (s)
6406 const char *s;
6407 {
6408 reg_errcode_t ret;
6409
6410 if (!s)
6411 {
6412 if (!re_comp_buf.buffer)
6413 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6414 return (char *) gettext ("No previous regular expression");
6415 return 0;
6416 }
6417
6418 if (!re_comp_buf.buffer)
6419 {
6420 re_comp_buf.buffer = (unsigned char *) malloc (200);
6421 if (re_comp_buf.buffer == NULL)
6422 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6423 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6424 re_comp_buf.allocated = 200;
6425
6426 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6427 if (re_comp_buf.fastmap == NULL)
6428 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6429 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6430 }
6431
6432 /* Since `re_exec' always passes NULL for the `regs' argument, we
6433 don't need to initialize the pattern buffer fields which affect it. */
6434
6435 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6436
6437 if (!ret)
6438 return NULL;
6439
6440 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6441 return (char *) gettext (re_error_msgid[(int) ret]);
6442 }
6443
6444
6445 int
6446 # ifdef _LIBC
6447 weak_function
6448 # endif
6449 re_exec (s)
6450 const char *s;
6451 {
6452 const int len = strlen (s);
6453 return
6454 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6455 }
6456 #endif /* _REGEX_RE_COMP */
6457 \f
6458 /* POSIX.2 functions. Don't define these for Emacs. */
6459
6460 #ifndef emacs
6461
6462 /* regcomp takes a regular expression as a string and compiles it.
6463
6464 PREG is a regex_t *. We do not expect any fields to be initialized,
6465 since POSIX says we shouldn't. Thus, we set
6466
6467 `buffer' to the compiled pattern;
6468 `used' to the length of the compiled pattern;
6469 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6470 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6471 RE_SYNTAX_POSIX_BASIC;
6472 `fastmap' to an allocated space for the fastmap;
6473 `fastmap_accurate' to zero;
6474 `re_nsub' to the number of subexpressions in PATTERN.
6475
6476 PATTERN is the address of the pattern string.
6477
6478 CFLAGS is a series of bits which affect compilation.
6479
6480 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6481 use POSIX basic syntax.
6482
6483 If REG_NEWLINE is set, then . and [^...] don't match newline.
6484 Also, regexec will try a match beginning after every newline.
6485
6486 If REG_ICASE is set, then we considers upper- and lowercase
6487 versions of letters to be equivalent when matching.
6488
6489 If REG_NOSUB is set, then when PREG is passed to regexec, that
6490 routine will report only success or failure, and nothing about the
6491 registers.
6492
6493 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6494 the return codes and their meanings.) */
6495
6496 int
6497 regcomp (regex_t *__restrict preg, const char *__restrict pattern,
6498 int cflags)
6499 {
6500 reg_errcode_t ret;
6501 reg_syntax_t syntax
6502 = (cflags & REG_EXTENDED) ?
6503 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6504
6505 /* regex_compile will allocate the space for the compiled pattern. */
6506 preg->buffer = 0;
6507 preg->allocated = 0;
6508 preg->used = 0;
6509
6510 /* Try to allocate space for the fastmap. */
6511 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6512
6513 if (cflags & REG_ICASE)
6514 {
6515 unsigned i;
6516
6517 preg->translate
6518 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6519 * sizeof (*(RE_TRANSLATE_TYPE)0));
6520 if (preg->translate == NULL)
6521 return (int) REG_ESPACE;
6522
6523 /* Map uppercase characters to corresponding lowercase ones. */
6524 for (i = 0; i < CHAR_SET_SIZE; i++)
6525 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6526 }
6527 else
6528 preg->translate = NULL;
6529
6530 /* If REG_NEWLINE is set, newlines are treated differently. */
6531 if (cflags & REG_NEWLINE)
6532 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6533 syntax &= ~RE_DOT_NEWLINE;
6534 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6535 }
6536 else
6537 syntax |= RE_NO_NEWLINE_ANCHOR;
6538
6539 preg->no_sub = !!(cflags & REG_NOSUB);
6540
6541 /* POSIX says a null character in the pattern terminates it, so we
6542 can use strlen here in compiling the pattern. */
6543 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6544
6545 /* POSIX doesn't distinguish between an unmatched open-group and an
6546 unmatched close-group: both are REG_EPAREN. */
6547 if (ret == REG_ERPAREN)
6548 ret = REG_EPAREN;
6549
6550 if (ret == REG_NOERROR && preg->fastmap)
6551 { /* Compute the fastmap now, since regexec cannot modify the pattern
6552 buffer. */
6553 re_compile_fastmap (preg);
6554 if (preg->can_be_null)
6555 { /* The fastmap can't be used anyway. */
6556 free (preg->fastmap);
6557 preg->fastmap = NULL;
6558 }
6559 }
6560 return (int) ret;
6561 }
6562 WEAK_ALIAS (__regcomp, regcomp)
6563
6564
6565 /* regexec searches for a given pattern, specified by PREG, in the
6566 string STRING.
6567
6568 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6569 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6570 least NMATCH elements, and we set them to the offsets of the
6571 corresponding matched substrings.
6572
6573 EFLAGS specifies `execution flags' which affect matching: if
6574 REG_NOTBOL is set, then ^ does not match at the beginning of the
6575 string; if REG_NOTEOL is set, then $ does not match at the end.
6576
6577 We return 0 if we find a match and REG_NOMATCH if not. */
6578
6579 int
6580 regexec (const regex_t *__restrict preg, const char *__restrict string,
6581 size_t nmatch, regmatch_t pmatch[__restrict_arr], int eflags)
6582 {
6583 int ret;
6584 struct re_registers regs;
6585 regex_t private_preg;
6586 int len = strlen (string);
6587 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6588
6589 private_preg = *preg;
6590
6591 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6592 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6593
6594 /* The user has told us exactly how many registers to return
6595 information about, via `nmatch'. We have to pass that on to the
6596 matching routines. */
6597 private_preg.regs_allocated = REGS_FIXED;
6598
6599 if (want_reg_info)
6600 {
6601 regs.num_regs = nmatch;
6602 regs.start = TALLOC (nmatch * 2, regoff_t);
6603 if (regs.start == NULL)
6604 return (int) REG_NOMATCH;
6605 regs.end = regs.start + nmatch;
6606 }
6607
6608 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6609 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6610 was a little bit longer but still only matching the real part.
6611 This works because the `endline' will check for a '\n' and will find a
6612 '\0', correctly deciding that this is not the end of a line.
6613 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6614 a convenient '\0' there. For all we know, the string could be preceded
6615 by '\n' which would throw things off. */
6616
6617 /* Perform the searching operation. */
6618 ret = re_search (&private_preg, string, len,
6619 /* start: */ 0, /* range: */ len,
6620 want_reg_info ? &regs : (struct re_registers *) 0);
6621
6622 /* Copy the register information to the POSIX structure. */
6623 if (want_reg_info)
6624 {
6625 if (ret >= 0)
6626 {
6627 unsigned r;
6628
6629 for (r = 0; r < nmatch; r++)
6630 {
6631 pmatch[r].rm_so = regs.start[r];
6632 pmatch[r].rm_eo = regs.end[r];
6633 }
6634 }
6635
6636 /* If we needed the temporary register info, free the space now. */
6637 free (regs.start);
6638 }
6639
6640 /* We want zero return to mean success, unlike `re_search'. */
6641 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6642 }
6643 WEAK_ALIAS (__regexec, regexec)
6644
6645
6646 /* Returns a message corresponding to an error code, ERR_CODE, returned
6647 from either regcomp or regexec. We don't use PREG here.
6648
6649 ERR_CODE was previously called ERRCODE, but that name causes an
6650 error with msvc8 compiler. */
6651
6652 size_t
6653 regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size)
6654 {
6655 const char *msg;
6656 size_t msg_size;
6657
6658 if (err_code < 0
6659 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6660 /* Only error codes returned by the rest of the code should be passed
6661 to this routine. If we are given anything else, or if other regex
6662 code generates an invalid error code, then the program has a bug.
6663 Dump core so we can fix it. */
6664 abort ();
6665
6666 msg = gettext (re_error_msgid[err_code]);
6667
6668 msg_size = strlen (msg) + 1; /* Includes the null. */
6669
6670 if (errbuf_size != 0)
6671 {
6672 if (msg_size > errbuf_size)
6673 {
6674 strncpy (errbuf, msg, errbuf_size - 1);
6675 errbuf[errbuf_size - 1] = 0;
6676 }
6677 else
6678 strcpy (errbuf, msg);
6679 }
6680
6681 return msg_size;
6682 }
6683 WEAK_ALIAS (__regerror, regerror)
6684
6685
6686 /* Free dynamically allocated space used by PREG. */
6687
6688 void
6689 regfree (regex_t *preg)
6690 {
6691 free (preg->buffer);
6692 preg->buffer = NULL;
6693
6694 preg->allocated = 0;
6695 preg->used = 0;
6696
6697 free (preg->fastmap);
6698 preg->fastmap = NULL;
6699 preg->fastmap_accurate = 0;
6700
6701 free (preg->translate);
6702 preg->translate = NULL;
6703 }
6704 WEAK_ALIAS (__regfree, regfree)
6705
6706 #endif /* not emacs */
Unnamed repository; edit this file 'description' to name the repository.