]> code.delx.au - gnu-emacs/blob - src/regex.c
delx.au / code / gnu-emacs / blob
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Avoid errors in 'newline'
[gnu-emacs] / 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) 1993-2016 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, see <http://www.gnu.org/licenses/>. */
19
20 /* TODO:
21 - structure the opcode space into opcode+flag.
22 - merge with glibc's regex.[ch].
23 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
24 need to modify the compiled regexp so that re_match can be reentrant.
25 - get rid of on_failure_jump_smart by doing the optimization in re_comp
26 rather than at run-time, so that re_match can be reentrant.
27 */
28
29 /* AIX requires this to be the first thing in the file. */
30 #if defined _AIX && !defined REGEX_MALLOC
31 #pragma alloca
32 #endif
33
34 /* Ignore some GCC warnings for now. This section should go away
35 once the Emacs and Gnulib regex code is merged. */
36 #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) || defined __clang__
37 # pragma GCC diagnostic ignored "-Wstrict-overflow"
38 # ifndef emacs
39 # pragma GCC diagnostic ignored "-Wunused-function"
40 # pragma GCC diagnostic ignored "-Wunused-macros"
41 # pragma GCC diagnostic ignored "-Wunused-result"
42 # pragma GCC diagnostic ignored "-Wunused-variable"
43 # endif
44 #endif
45
46 #if 4 < __GNUC__ + (6 <= __GNUC_MINOR__) && ! defined __clang__
47 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
48 #endif
49
50 #include <config.h>
51
52 #include <stddef.h>
53
54 #ifdef emacs
55 /* We need this for `regex.h', and perhaps for the Emacs include files. */
56 # include <sys/types.h>
57 #endif
58
59 /* Whether to use ISO C Amendment 1 wide char functions.
60 Those should not be used for Emacs since it uses its own. */
61 #if defined _LIBC
62 #define WIDE_CHAR_SUPPORT 1
63 #else
64 #define WIDE_CHAR_SUPPORT \
65 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
66 #endif
67
68 /* For platform which support the ISO C amendment 1 functionality we
69 support user defined character classes. */
70 #if WIDE_CHAR_SUPPORT
71 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
72 # include <wchar.h>
73 # include <wctype.h>
74 #endif
75
76 #ifdef _LIBC
77 /* We have to keep the namespace clean. */
78 # define regfree(preg) __regfree (preg)
79 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
80 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
81 # define regerror(err_code, preg, errbuf, errbuf_size) \
82 __regerror (err_code, preg, errbuf, errbuf_size)
83 # define re_set_registers(bu, re, nu, st, en) \
84 __re_set_registers (bu, re, nu, st, en)
85 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
86 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
87 # define re_match(bufp, string, size, pos, regs) \
88 __re_match (bufp, string, size, pos, regs)
89 # define re_search(bufp, string, size, startpos, range, regs) \
90 __re_search (bufp, string, size, startpos, range, regs)
91 # define re_compile_pattern(pattern, length, bufp) \
92 __re_compile_pattern (pattern, length, bufp)
93 # define re_set_syntax(syntax) __re_set_syntax (syntax)
94 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
95 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
96 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
97
98 /* Make sure we call libc's function even if the user overrides them. */
99 # define btowc __btowc
100 # define iswctype __iswctype
101 # define wctype __wctype
102
103 # define WEAK_ALIAS(a,b) weak_alias (a, b)
104
105 /* We are also using some library internals. */
106 # include <locale/localeinfo.h>
107 # include <locale/elem-hash.h>
108 # include <langinfo.h>
109 #else
110 # define WEAK_ALIAS(a,b)
111 #endif
112
113 /* This is for other GNU distributions with internationalized messages. */
114 #if HAVE_LIBINTL_H || defined _LIBC
115 # include <libintl.h>
116 #else
117 # define gettext(msgid) (msgid)
118 #endif
119
120 #ifndef gettext_noop
121 /* This define is so xgettext can find the internationalizable
122 strings. */
123 # define gettext_noop(String) String
124 #endif
125
126 /* The `emacs' switch turns on certain matching commands
127 that make sense only in Emacs. */
128 #ifdef emacs
129
130 # include "lisp.h"
131 # include "character.h"
132 # include "buffer.h"
133
134 # include "syntax.h"
135 # include "category.h"
136
137 /* Make syntax table lookup grant data in gl_state. */
138 # define SYNTAX(c) syntax_property (c, 1)
139
140 # ifdef malloc
141 # undef malloc
142 # endif
143 # define malloc xmalloc
144 # ifdef realloc
145 # undef realloc
146 # endif
147 # define realloc xrealloc
148 # ifdef free
149 # undef free
150 # endif
151 # define free xfree
152
153 /* Converts the pointer to the char to BEG-based offset from the start. */
154 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
155 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
156
157 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
158 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
159 # define RE_STRING_CHAR(p, multibyte) \
160 (multibyte ? (STRING_CHAR (p)) : (*(p)))
161 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
162 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
163
164 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
165
166 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
167
168 /* Set C a (possibly converted to multibyte) character before P. P
169 points into a string which is the virtual concatenation of STR1
170 (which ends at END1) or STR2 (which ends at END2). */
171 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
172 do { \
173 if (target_multibyte) \
174 { \
175 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
176 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
177 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
178 c = STRING_CHAR (dtemp); \
179 } \
180 else \
181 { \
182 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
183 (c) = RE_CHAR_TO_MULTIBYTE (c); \
184 } \
185 } while (0)
186
187 /* Set C a (possibly converted to multibyte) character at P, and set
188 LEN to the byte length of that character. */
189 # define GET_CHAR_AFTER(c, p, len) \
190 do { \
191 if (target_multibyte) \
192 (c) = STRING_CHAR_AND_LENGTH (p, len); \
193 else \
194 { \
195 (c) = *p; \
196 len = 1; \
197 (c) = RE_CHAR_TO_MULTIBYTE (c); \
198 } \
199 } while (0)
200
201 #else /* not emacs */
202
203 /* If we are not linking with Emacs proper,
204 we can't use the relocating allocator
205 even if config.h says that we can. */
206 # undef REL_ALLOC
207
208 # include <unistd.h>
209
210 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
211
212 static void *
213 xmalloc (size_t size)
214 {
215 void *val = malloc (size);
216 if (!val && size)
217 {
218 write (2, "virtual memory exhausted\n", 25);
219 exit (1);
220 }
221 return val;
222 }
223
224 static void *
225 xrealloc (void *block, size_t size)
226 {
227 void *val;
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
230 if (! block)
231 val = malloc (size);
232 else
233 val = realloc (block, size);
234 if (!val && size)
235 {
236 write (2, "virtual memory exhausted\n", 25);
237 exit (1);
238 }
239 return val;
240 }
241
242 # ifdef malloc
243 # undef malloc
244 # endif
245 # define malloc xmalloc
246 # ifdef realloc
247 # undef realloc
248 # endif
249 # define realloc xrealloc
250
251 # include <stdbool.h>
252 # include <string.h>
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 /* Dummy macros for non-Emacs environments. */
260 # define MAX_MULTIBYTE_LENGTH 1
261 # define RE_MULTIBYTE_P(x) 0
262 # define RE_TARGET_MULTIBYTE_P(x) 0
263 # define WORD_BOUNDARY_P(c1, c2) (0)
264 # define BYTES_BY_CHAR_HEAD(p) (1)
265 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
266 # define STRING_CHAR(p) (*(p))
267 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
268 # define CHAR_STRING(c, s) (*(s) = (c), 1)
269 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
270 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
271 # define RE_CHAR_TO_MULTIBYTE(c) (c)
272 # define RE_CHAR_TO_UNIBYTE(c) (c)
273 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
274 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
275 # define GET_CHAR_AFTER(c, p, len) \
276 (c = *p, len = 1)
277 # define CHAR_BYTE8_P(c) (0)
278 # define CHAR_LEADING_CODE(c) (c)
279
280 #endif /* not emacs */
281
282 #ifndef RE_TRANSLATE
283 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
284 # define RE_TRANSLATE_P(TBL) (TBL)
285 #endif
286 \f
287 /* Get the interface, including the syntax bits. */
288 #include "regex.h"
289
290 /* isalpha etc. are used for the character classes. */
291 #include <ctype.h>
292
293 #ifdef emacs
294
295 /* 1 if C is an ASCII character. */
296 # define IS_REAL_ASCII(c) ((c) < 0200)
297
298 /* 1 if C is a unibyte character. */
299 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
300
301 /* The Emacs definitions should not be directly affected by locales. */
302
303 /* In Emacs, these are only used for single-byte characters. */
304 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
305 # define ISCNTRL(c) ((c) < ' ')
306 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
307 || ((c) >= 'a' && (c) <= 'f') \
308 || ((c) >= 'A' && (c) <= 'F'))
309
310 /* This is only used for single-byte characters. */
311 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
312
313 /* The rest must handle multibyte characters. */
314
315 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
316 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0240) \
317 : graphicp (c))
318
319 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
320 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
321 : printablep (c))
322
323 # define ISALNUM(c) (IS_REAL_ASCII (c) \
324 ? (((c) >= 'a' && (c) <= 'z') \
325 || ((c) >= 'A' && (c) <= 'Z') \
326 || ((c) >= '0' && (c) <= '9')) \
327 : (alphabeticp (c) || decimalnump (c)))
328
329 # define ISALPHA(c) (IS_REAL_ASCII (c) \
330 ? (((c) >= 'a' && (c) <= 'z') \
331 || ((c) >= 'A' && (c) <= 'Z')) \
332 : alphabeticp (c))
333
334 # define ISLOWER(c) lowercasep (c)
335
336 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
337 ? ((c) > ' ' && (c) < 0177 \
338 && !(((c) >= 'a' && (c) <= 'z') \
339 || ((c) >= 'A' && (c) <= 'Z') \
340 || ((c) >= '0' && (c) <= '9'))) \
341 : SYNTAX (c) != Sword)
342
343 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
344
345 # define ISUPPER(c) uppercasep (c)
346
347 # define ISWORD(c) (SYNTAX (c) == Sword)
348
349 #else /* not emacs */
350
351 /* 1 if C is an ASCII character. */
352 # define IS_REAL_ASCII(c) ((c) < 0200)
353
354 /* This distinction is not meaningful, except in Emacs. */
355 # define ISUNIBYTE(c) 1
356
357 # ifdef isblank
358 # define ISBLANK(c) isblank (c)
359 # else
360 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
361 # endif
362 # ifdef isgraph
363 # define ISGRAPH(c) isgraph (c)
364 # else
365 # define ISGRAPH(c) (isprint (c) && !isspace (c))
366 # endif
367
368 /* Solaris defines ISPRINT so we must undefine it first. */
369 # undef ISPRINT
370 # define ISPRINT(c) isprint (c)
371 # define ISDIGIT(c) isdigit (c)
372 # define ISALNUM(c) isalnum (c)
373 # define ISALPHA(c) isalpha (c)
374 # define ISCNTRL(c) iscntrl (c)
375 # define ISLOWER(c) islower (c)
376 # define ISPUNCT(c) ispunct (c)
377 # define ISSPACE(c) isspace (c)
378 # define ISUPPER(c) isupper (c)
379 # define ISXDIGIT(c) isxdigit (c)
380
381 # define ISWORD(c) ISALPHA (c)
382
383 # ifdef _tolower
384 # define TOLOWER(c) _tolower (c)
385 # else
386 # define TOLOWER(c) tolower (c)
387 # endif
388
389 /* How many characters in the character set. */
390 # define CHAR_SET_SIZE 256
391
392 # ifdef SYNTAX_TABLE
393
394 extern char *re_syntax_table;
395
396 # else /* not SYNTAX_TABLE */
397
398 static char re_syntax_table[CHAR_SET_SIZE];
399
400 static void
401 init_syntax_once (void)
402 {
403 register int c;
404 static int done = 0;
405
406 if (done)
407 return;
408
409 memset (re_syntax_table, 0, sizeof re_syntax_table);
410
411 for (c = 0; c < CHAR_SET_SIZE; ++c)
412 if (ISALNUM (c))
413 re_syntax_table[c] = Sword;
414
415 re_syntax_table['_'] = Ssymbol;
416
417 done = 1;
418 }
419
420 # endif /* not SYNTAX_TABLE */
421
422 # define SYNTAX(c) re_syntax_table[(c)]
423
424 #endif /* not emacs */
425 \f
426 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
427 \f
428 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
429 use `alloca' instead of `malloc'. This is because using malloc in
430 re_search* or re_match* could cause memory leaks when C-g is used in
431 Emacs; also, malloc is slower and causes storage fragmentation. On
432 the other hand, malloc is more portable, and easier to debug.
433
434 Because we sometimes use alloca, some routines have to be macros,
435 not functions -- `alloca'-allocated space disappears at the end of the
436 function it is called in. */
437
438 #ifdef REGEX_MALLOC
439
440 # define REGEX_ALLOCATE malloc
441 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
442 # define REGEX_FREE free
443
444 #else /* not REGEX_MALLOC */
445
446 /* Emacs already defines alloca, sometimes. */
447 # ifndef alloca
448
449 /* Make alloca work the best possible way. */
450 # ifdef __GNUC__
451 # define alloca __builtin_alloca
452 # else /* not __GNUC__ */
453 # ifdef HAVE_ALLOCA_H
454 # include <alloca.h>
455 # endif /* HAVE_ALLOCA_H */
456 # endif /* not __GNUC__ */
457
458 # endif /* not alloca */
459
460 # ifdef emacs
461 # define REGEX_USE_SAFE_ALLOCA USE_SAFE_ALLOCA
462 # define REGEX_SAFE_FREE() SAFE_FREE ()
463 # define REGEX_ALLOCATE SAFE_ALLOCA
464 # else
465 # define REGEX_ALLOCATE alloca
466 # endif
467
468 /* Assumes a `char *destination' variable. */
469 # define REGEX_REALLOCATE(source, osize, nsize) \
470 (destination = REGEX_ALLOCATE (nsize), \
471 memcpy (destination, source, osize))
472
473 /* No need to do anything to free, after alloca. */
474 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
475
476 #endif /* not REGEX_MALLOC */
477
478 #ifndef REGEX_USE_SAFE_ALLOCA
479 # define REGEX_USE_SAFE_ALLOCA ((void) 0)
480 # define REGEX_SAFE_FREE() ((void) 0)
481 #endif
482
483 /* Define how to allocate the failure stack. */
484
485 #if defined REL_ALLOC && defined REGEX_MALLOC
486
487 # define REGEX_ALLOCATE_STACK(size) \
488 r_alloc (&failure_stack_ptr, (size))
489 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
490 r_re_alloc (&failure_stack_ptr, (nsize))
491 # define REGEX_FREE_STACK(ptr) \
492 r_alloc_free (&failure_stack_ptr)
493
494 #else /* not using relocating allocator */
495
496 # define REGEX_ALLOCATE_STACK(size) REGEX_ALLOCATE (size)
497 # define REGEX_REALLOCATE_STACK(source, o, n) REGEX_REALLOCATE (source, o, n)
498 # define REGEX_FREE_STACK(ptr) REGEX_FREE (ptr)
499
500 #endif /* not using relocating allocator */
501
502
503 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
504 `string1' or just past its end. This works if PTR is NULL, which is
505 a good thing. */
506 #define FIRST_STRING_P(ptr) \
507 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
508
509 /* (Re)Allocate N items of type T using malloc, or fail. */
510 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
511 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
512 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
513
514 #define BYTEWIDTH 8 /* In bits. */
515
516 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
517
518 #ifndef emacs
519 # undef max
520 # undef min
521 # define max(a, b) ((a) > (b) ? (a) : (b))
522 # define min(a, b) ((a) < (b) ? (a) : (b))
523 #endif
524
525 /* Type of source-pattern and string chars. */
526 #ifdef _MSC_VER
527 typedef unsigned char re_char;
528 typedef const re_char const_re_char;
529 #else
530 typedef const unsigned char re_char;
531 typedef re_char const_re_char;
532 #endif
533
534 typedef char boolean;
535
536 static regoff_t re_match_2_internal (struct re_pattern_buffer *bufp,
537 re_char *string1, size_t size1,
538 re_char *string2, size_t size2,
539 ssize_t pos,
540 struct re_registers *regs,
541 ssize_t stop);
542 \f
543 /* These are the command codes that appear in compiled regular
544 expressions. Some opcodes are followed by argument bytes. A
545 command code can specify any interpretation whatsoever for its
546 arguments. Zero bytes may appear in the compiled regular expression. */
547
548 typedef enum
549 {
550 no_op = 0,
551
552 /* Succeed right away--no more backtracking. */
553 succeed,
554
555 /* Followed by one byte giving n, then by n literal bytes. */
556 exactn,
557
558 /* Matches any (more or less) character. */
559 anychar,
560
561 /* Matches any one char belonging to specified set. First
562 following byte is number of bitmap bytes. Then come bytes
563 for a bitmap saying which chars are in. Bits in each byte
564 are ordered low-bit-first. A character is in the set if its
565 bit is 1. A character too large to have a bit in the map is
566 automatically not in the set.
567
568 If the length byte has the 0x80 bit set, then that stuff
569 is followed by a range table:
570 2 bytes of flags for character sets (low 8 bits, high 8 bits)
571 See RANGE_TABLE_WORK_BITS below.
572 2 bytes, the number of pairs that follow (upto 32767)
573 pairs, each 2 multibyte characters,
574 each multibyte character represented as 3 bytes. */
575 charset,
576
577 /* Same parameters as charset, but match any character that is
578 not one of those specified. */
579 charset_not,
580
581 /* Start remembering the text that is matched, for storing in a
582 register. Followed by one byte with the register number, in
583 the range 0 to one less than the pattern buffer's re_nsub
584 field. */
585 start_memory,
586
587 /* Stop remembering the text that is matched and store it in a
588 memory register. Followed by one byte with the register
589 number, in the range 0 to one less than `re_nsub' in the
590 pattern buffer. */
591 stop_memory,
592
593 /* Match a duplicate of something remembered. Followed by one
594 byte containing the register number. */
595 duplicate,
596
597 /* Fail unless at beginning of line. */
598 begline,
599
600 /* Fail unless at end of line. */
601 endline,
602
603 /* Succeeds if at beginning of buffer (if emacs) or at beginning
604 of string to be matched (if not). */
605 begbuf,
606
607 /* Analogously, for end of buffer/string. */
608 endbuf,
609
610 /* Followed by two byte relative address to which to jump. */
611 jump,
612
613 /* Followed by two-byte relative address of place to resume at
614 in case of failure. */
615 on_failure_jump,
616
617 /* Like on_failure_jump, but pushes a placeholder instead of the
618 current string position when executed. */
619 on_failure_keep_string_jump,
620
621 /* Just like `on_failure_jump', except that it checks that we
622 don't get stuck in an infinite loop (matching an empty string
623 indefinitely). */
624 on_failure_jump_loop,
625
626 /* Just like `on_failure_jump_loop', except that it checks for
627 a different kind of loop (the kind that shows up with non-greedy
628 operators). This operation has to be immediately preceded
629 by a `no_op'. */
630 on_failure_jump_nastyloop,
631
632 /* A smart `on_failure_jump' used for greedy * and + operators.
633 It analyzes the loop before which it is put and if the
634 loop does not require backtracking, it changes itself to
635 `on_failure_keep_string_jump' and short-circuits the loop,
636 else it just defaults to changing itself into `on_failure_jump'.
637 It assumes that it is pointing to just past a `jump'. */
638 on_failure_jump_smart,
639
640 /* Followed by two-byte relative address and two-byte number n.
641 After matching N times, jump to the address upon failure.
642 Does not work if N starts at 0: use on_failure_jump_loop
643 instead. */
644 succeed_n,
645
646 /* Followed by two-byte relative address, and two-byte number n.
647 Jump to the address N times, then fail. */
648 jump_n,
649
650 /* Set the following two-byte relative address to the
651 subsequent two-byte number. The address *includes* the two
652 bytes of number. */
653 set_number_at,
654
655 wordbeg, /* Succeeds if at word beginning. */
656 wordend, /* Succeeds if at word end. */
657
658 wordbound, /* Succeeds if at a word boundary. */
659 notwordbound, /* Succeeds if not at a word boundary. */
660
661 symbeg, /* Succeeds if at symbol beginning. */
662 symend, /* Succeeds if at symbol end. */
663
664 /* Matches any character whose syntax is specified. Followed by
665 a byte which contains a syntax code, e.g., Sword. */
666 syntaxspec,
667
668 /* Matches any character whose syntax is not that specified. */
669 notsyntaxspec
670
671 #ifdef emacs
672 ,before_dot, /* Succeeds if before point. */
673 at_dot, /* Succeeds if at point. */
674 after_dot, /* Succeeds if after point. */
675
676 /* Matches any character whose category-set contains the specified
677 category. The operator is followed by a byte which contains a
678 category code (mnemonic ASCII character). */
679 categoryspec,
680
681 /* Matches any character whose category-set does not contain the
682 specified category. The operator is followed by a byte which
683 contains the category code (mnemonic ASCII character). */
684 notcategoryspec
685 #endif /* emacs */
686 } re_opcode_t;
687 \f
688 /* Common operations on the compiled pattern. */
689
690 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
691
692 #define STORE_NUMBER(destination, number) \
693 do { \
694 (destination)[0] = (number) & 0377; \
695 (destination)[1] = (number) >> 8; \
696 } while (0)
697
698 /* Same as STORE_NUMBER, except increment DESTINATION to
699 the byte after where the number is stored. Therefore, DESTINATION
700 must be an lvalue. */
701
702 #define STORE_NUMBER_AND_INCR(destination, number) \
703 do { \
704 STORE_NUMBER (destination, number); \
705 (destination) += 2; \
706 } while (0)
707
708 /* Put into DESTINATION a number stored in two contiguous bytes starting
709 at SOURCE. */
710
711 #define EXTRACT_NUMBER(destination, source) \
712 ((destination) = extract_number (source))
713
714 static int
715 extract_number (re_char *source)
716 {
717 unsigned leading_byte = SIGN_EXTEND_CHAR (source[1]);
718 return (leading_byte << 8) + source[0];
719 }
720
721 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
722 SOURCE must be an lvalue. */
723
724 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
725 ((destination) = extract_number_and_incr (&source))
726
727 static int
728 extract_number_and_incr (re_char **source)
729 {
730 int num = extract_number (*source);
731 *source += 2;
732 return num;
733 }
734 \f
735 /* Store a multibyte character in three contiguous bytes starting
736 DESTINATION, and increment DESTINATION to the byte after where the
737 character is stored. Therefore, DESTINATION must be an lvalue. */
738
739 #define STORE_CHARACTER_AND_INCR(destination, character) \
740 do { \
741 (destination)[0] = (character) & 0377; \
742 (destination)[1] = ((character) >> 8) & 0377; \
743 (destination)[2] = (character) >> 16; \
744 (destination) += 3; \
745 } while (0)
746
747 /* Put into DESTINATION a character stored in three contiguous bytes
748 starting at SOURCE. */
749
750 #define EXTRACT_CHARACTER(destination, source) \
751 do { \
752 (destination) = ((source)[0] \
753 | ((source)[1] << 8) \
754 | ((source)[2] << 16)); \
755 } while (0)
756
757
758 /* Macros for charset. */
759
760 /* Size of bitmap of charset P in bytes. P is a start of charset,
761 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
762 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
763
764 /* Nonzero if charset P has range table. */
765 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
766
767 /* Return the address of range table of charset P. But not the start
768 of table itself, but the before where the number of ranges is
769 stored. `2 +' means to skip re_opcode_t and size of bitmap,
770 and the 2 bytes of flags at the start of the range table. */
771 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
772
773 #ifdef emacs
774 /* Extract the bit flags that start a range table. */
775 #define CHARSET_RANGE_TABLE_BITS(p) \
776 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
777 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
778 #endif
779
780 /* Return the address of end of RANGE_TABLE. COUNT is number of
781 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
782 is start of range and end of range. `* 3' is size of each start
783 and end. */
784 #define CHARSET_RANGE_TABLE_END(range_table, count) \
785 ((range_table) + (count) * 2 * 3)
786
787 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
788 COUNT is number of ranges in RANGE_TABLE. */
789 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
790 do \
791 { \
792 re_wchar_t range_start, range_end; \
793 re_char *rtp; \
794 re_char *range_table_end \
795 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
796 \
797 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
798 { \
799 EXTRACT_CHARACTER (range_start, rtp); \
800 EXTRACT_CHARACTER (range_end, rtp + 3); \
801 \
802 if (range_start <= (c) && (c) <= range_end) \
803 { \
804 (not) = !(not); \
805 break; \
806 } \
807 } \
808 } \
809 while (0)
810
811 /* Test if C is in range table of CHARSET. The flag NOT is negated if
812 C is listed in it. */
813 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
814 do \
815 { \
816 /* Number of ranges in range table. */ \
817 int count; \
818 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
819 \
820 EXTRACT_NUMBER_AND_INCR (count, range_table); \
821 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
822 } \
823 while (0)
824 \f
825 /* If DEBUG is defined, Regex prints many voluminous messages about what
826 it is doing (if the variable `debug' is nonzero). If linked with the
827 main program in `iregex.c', you can enter patterns and strings
828 interactively. And if linked with the main program in `main.c' and
829 the other test files, you can run the already-written tests. */
830
831 #ifdef DEBUG
832
833 /* We use standard I/O for debugging. */
834 # include <stdio.h>
835
836 /* It is useful to test things that ``must'' be true when debugging. */
837 # include <assert.h>
838
839 static int debug = -100000;
840
841 # define DEBUG_STATEMENT(e) e
842 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
843 # define DEBUG_COMPILES_ARGUMENTS
844 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
845 if (debug > 0) print_partial_compiled_pattern (s, e)
846 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
847 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
848
849
850 /* Print the fastmap in human-readable form. */
851
852 static void
853 print_fastmap (char *fastmap)
854 {
855 unsigned was_a_range = 0;
856 unsigned i = 0;
857
858 while (i < (1 << BYTEWIDTH))
859 {
860 if (fastmap[i++])
861 {
862 was_a_range = 0;
863 putchar (i - 1);
864 while (i < (1 << BYTEWIDTH) && fastmap[i])
865 {
866 was_a_range = 1;
867 i++;
868 }
869 if (was_a_range)
870 {
871 printf ("-");
872 putchar (i - 1);
873 }
874 }
875 }
876 putchar ('\n');
877 }
878
879
880 /* Print a compiled pattern string in human-readable form, starting at
881 the START pointer into it and ending just before the pointer END. */
882
883 static void
884 print_partial_compiled_pattern (re_char *start, re_char *end)
885 {
886 int mcnt, mcnt2;
887 re_char *p = start;
888 re_char *pend = end;
889
890 if (start == NULL)
891 {
892 fprintf (stderr, "(null)\n");
893 return;
894 }
895
896 /* Loop over pattern commands. */
897 while (p < pend)
898 {
899 fprintf (stderr, "%td:\t", p - start);
900
901 switch ((re_opcode_t) *p++)
902 {
903 case no_op:
904 fprintf (stderr, "/no_op");
905 break;
906
907 case succeed:
908 fprintf (stderr, "/succeed");
909 break;
910
911 case exactn:
912 mcnt = *p++;
913 fprintf (stderr, "/exactn/%d", mcnt);
914 do
915 {
916 fprintf (stderr, "/%c", *p++);
917 }
918 while (--mcnt);
919 break;
920
921 case start_memory:
922 fprintf (stderr, "/start_memory/%d", *p++);
923 break;
924
925 case stop_memory:
926 fprintf (stderr, "/stop_memory/%d", *p++);
927 break;
928
929 case duplicate:
930 fprintf (stderr, "/duplicate/%d", *p++);
931 break;
932
933 case anychar:
934 fprintf (stderr, "/anychar");
935 break;
936
937 case charset:
938 case charset_not:
939 {
940 register int c, last = -100;
941 register int in_range = 0;
942 int length = CHARSET_BITMAP_SIZE (p - 1);
943 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
944
945 fprintf (stderr, "/charset [%s",
946 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
947
948 if (p + *p >= pend)
949 fprintf (stderr, " !extends past end of pattern! ");
950
951 for (c = 0; c < 256; c++)
952 if (c / 8 < length
953 && (p[1 + (c/8)] & (1 << (c % 8))))
954 {
955 /* Are we starting a range? */
956 if (last + 1 == c && ! in_range)
957 {
958 fprintf (stderr, "-");
959 in_range = 1;
960 }
961 /* Have we broken a range? */
962 else if (last + 1 != c && in_range)
963 {
964 fprintf (stderr, "%c", last);
965 in_range = 0;
966 }
967
968 if (! in_range)
969 fprintf (stderr, "%c", c);
970
971 last = c;
972 }
973
974 if (in_range)
975 fprintf (stderr, "%c", last);
976
977 fprintf (stderr, "]");
978
979 p += 1 + length;
980
981 if (has_range_table)
982 {
983 int count;
984 fprintf (stderr, "has-range-table");
985
986 /* ??? Should print the range table; for now, just skip it. */
987 p += 2; /* skip range table bits */
988 EXTRACT_NUMBER_AND_INCR (count, p);
989 p = CHARSET_RANGE_TABLE_END (p, count);
990 }
991 }
992 break;
993
994 case begline:
995 fprintf (stderr, "/begline");
996 break;
997
998 case endline:
999 fprintf (stderr, "/endline");
1000 break;
1001
1002 case on_failure_jump:
1003 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1004 fprintf (stderr, "/on_failure_jump to %td", p + mcnt - start);
1005 break;
1006
1007 case on_failure_keep_string_jump:
1008 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1009 fprintf (stderr, "/on_failure_keep_string_jump to %td",
1010 p + mcnt - start);
1011 break;
1012
1013 case on_failure_jump_nastyloop:
1014 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1015 fprintf (stderr, "/on_failure_jump_nastyloop to %td",
1016 p + mcnt - start);
1017 break;
1018
1019 case on_failure_jump_loop:
1020 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1021 fprintf (stderr, "/on_failure_jump_loop to %td",
1022 p + mcnt - start);
1023 break;
1024
1025 case on_failure_jump_smart:
1026 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1027 fprintf (stderr, "/on_failure_jump_smart to %td",
1028 p + mcnt - start);
1029 break;
1030
1031 case jump:
1032 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1033 fprintf (stderr, "/jump to %td", p + mcnt - start);
1034 break;
1035
1036 case succeed_n:
1037 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1038 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1039 fprintf (stderr, "/succeed_n to %td, %d times",
1040 p - 2 + mcnt - start, mcnt2);
1041 break;
1042
1043 case jump_n:
1044 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1045 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1046 fprintf (stderr, "/jump_n to %td, %d times",
1047 p - 2 + mcnt - start, mcnt2);
1048 break;
1049
1050 case set_number_at:
1051 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1052 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1053 fprintf (stderr, "/set_number_at location %td to %d",
1054 p - 2 + mcnt - start, mcnt2);
1055 break;
1056
1057 case wordbound:
1058 fprintf (stderr, "/wordbound");
1059 break;
1060
1061 case notwordbound:
1062 fprintf (stderr, "/notwordbound");
1063 break;
1064
1065 case wordbeg:
1066 fprintf (stderr, "/wordbeg");
1067 break;
1068
1069 case wordend:
1070 fprintf (stderr, "/wordend");
1071 break;
1072
1073 case symbeg:
1074 fprintf (stderr, "/symbeg");
1075 break;
1076
1077 case symend:
1078 fprintf (stderr, "/symend");
1079 break;
1080
1081 case syntaxspec:
1082 fprintf (stderr, "/syntaxspec");
1083 mcnt = *p++;
1084 fprintf (stderr, "/%d", mcnt);
1085 break;
1086
1087 case notsyntaxspec:
1088 fprintf (stderr, "/notsyntaxspec");
1089 mcnt = *p++;
1090 fprintf (stderr, "/%d", mcnt);
1091 break;
1092
1093 # ifdef emacs
1094 case before_dot:
1095 fprintf (stderr, "/before_dot");
1096 break;
1097
1098 case at_dot:
1099 fprintf (stderr, "/at_dot");
1100 break;
1101
1102 case after_dot:
1103 fprintf (stderr, "/after_dot");
1104 break;
1105
1106 case categoryspec:
1107 fprintf (stderr, "/categoryspec");
1108 mcnt = *p++;
1109 fprintf (stderr, "/%d", mcnt);
1110 break;
1111
1112 case notcategoryspec:
1113 fprintf (stderr, "/notcategoryspec");
1114 mcnt = *p++;
1115 fprintf (stderr, "/%d", mcnt);
1116 break;
1117 # endif /* emacs */
1118
1119 case begbuf:
1120 fprintf (stderr, "/begbuf");
1121 break;
1122
1123 case endbuf:
1124 fprintf (stderr, "/endbuf");
1125 break;
1126
1127 default:
1128 fprintf (stderr, "?%d", *(p-1));
1129 }
1130
1131 fprintf (stderr, "\n");
1132 }
1133
1134 fprintf (stderr, "%td:\tend of pattern.\n", p - start);
1135 }
1136
1137
1138 static void
1139 print_compiled_pattern (struct re_pattern_buffer *bufp)
1140 {
1141 re_char *buffer = bufp->buffer;
1142
1143 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1144 printf ("%ld bytes used/%ld bytes allocated.\n",
1145 bufp->used, bufp->allocated);
1146
1147 if (bufp->fastmap_accurate && bufp->fastmap)
1148 {
1149 printf ("fastmap: ");
1150 print_fastmap (bufp->fastmap);
1151 }
1152
1153 printf ("re_nsub: %zu\t", bufp->re_nsub);
1154 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1155 printf ("can_be_null: %d\t", bufp->can_be_null);
1156 printf ("no_sub: %d\t", bufp->no_sub);
1157 printf ("not_bol: %d\t", bufp->not_bol);
1158 printf ("not_eol: %d\t", bufp->not_eol);
1159 printf ("syntax: %lx\n", bufp->syntax);
1160 fflush (stdout);
1161 /* Perhaps we should print the translate table? */
1162 }
1163
1164
1165 static void
1166 print_double_string (re_char *where, re_char *string1, ssize_t size1,
1167 re_char *string2, ssize_t size2)
1168 {
1169 ssize_t this_char;
1170
1171 if (where == NULL)
1172 printf ("(null)");
1173 else
1174 {
1175 if (FIRST_STRING_P (where))
1176 {
1177 for (this_char = where - string1; this_char < size1; this_char++)
1178 putchar (string1[this_char]);
1179
1180 where = string2;
1181 }
1182
1183 for (this_char = where - string2; this_char < size2; this_char++)
1184 putchar (string2[this_char]);
1185 }
1186 }
1187
1188 #else /* not DEBUG */
1189
1190 # undef assert
1191 # define assert(e)
1192
1193 # define DEBUG_STATEMENT(e)
1194 # define DEBUG_PRINT(...)
1195 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1196 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1197
1198 #endif /* not DEBUG */
1199 \f
1200 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1201 #ifdef lint
1202 # define IF_LINT(Code) Code
1203 #else
1204 # define IF_LINT(Code) /* empty */
1205 #endif
1206 \f
1207 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1208 also be assigned to arbitrarily: each pattern buffer stores its own
1209 syntax, so it can be changed between regex compilations. */
1210 /* This has no initializer because initialized variables in Emacs
1211 become read-only after dumping. */
1212 reg_syntax_t re_syntax_options;
1213
1214
1215 /* Specify the precise syntax of regexps for compilation. This provides
1216 for compatibility for various utilities which historically have
1217 different, incompatible syntaxes.
1218
1219 The argument SYNTAX is a bit mask comprised of the various bits
1220 defined in regex.h. We return the old syntax. */
1221
1222 reg_syntax_t
1223 re_set_syntax (reg_syntax_t syntax)
1224 {
1225 reg_syntax_t ret = re_syntax_options;
1226
1227 re_syntax_options = syntax;
1228 return ret;
1229 }
1230 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1231
1232 /* Regexp to use to replace spaces, or NULL meaning don't. */
1233 static const_re_char *whitespace_regexp;
1234
1235 void
1236 re_set_whitespace_regexp (const char *regexp)
1237 {
1238 whitespace_regexp = (const_re_char *) regexp;
1239 }
1240 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1241 \f
1242 /* This table gives an error message for each of the error codes listed
1243 in regex.h. Obviously the order here has to be same as there.
1244 POSIX doesn't require that we do anything for REG_NOERROR,
1245 but why not be nice? */
1246
1247 static const char *re_error_msgid[] =
1248 {
1249 gettext_noop ("Success"), /* REG_NOERROR */
1250 gettext_noop ("No match"), /* REG_NOMATCH */
1251 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1252 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1253 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1254 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1255 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1256 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1257 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1258 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1259 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1260 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1261 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1262 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1263 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1264 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1265 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1266 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1267 };
1268 \f
1269 /* Avoiding alloca during matching, to placate r_alloc. */
1270
1271 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1272 searching and matching functions should not call alloca. On some
1273 systems, alloca is implemented in terms of malloc, and if we're
1274 using the relocating allocator routines, then malloc could cause a
1275 relocation, which might (if the strings being searched are in the
1276 ralloc heap) shift the data out from underneath the regexp
1277 routines.
1278
1279 Here's another reason to avoid allocation: Emacs
1280 processes input from X in a signal handler; processing X input may
1281 call malloc; if input arrives while a matching routine is calling
1282 malloc, then we're scrod. But Emacs can't just block input while
1283 calling matching routines; then we don't notice interrupts when
1284 they come in. So, Emacs blocks input around all regexp calls
1285 except the matching calls, which it leaves unprotected, in the
1286 faith that they will not malloc. */
1287
1288 /* Normally, this is fine. */
1289 #define MATCH_MAY_ALLOCATE
1290
1291 /* The match routines may not allocate if (1) they would do it with malloc
1292 and (2) it's not safe for them to use malloc.
1293 Note that if REL_ALLOC is defined, matching would not use malloc for the
1294 failure stack, but we would still use it for the register vectors;
1295 so REL_ALLOC should not affect this. */
1296 #if defined REGEX_MALLOC && defined emacs
1297 # undef MATCH_MAY_ALLOCATE
1298 #endif
1299
1300 \f
1301 /* Failure stack declarations and macros; both re_compile_fastmap and
1302 re_match_2 use a failure stack. These have to be macros because of
1303 REGEX_ALLOCATE_STACK. */
1304
1305
1306 /* Approximate number of failure points for which to initially allocate space
1307 when matching. If this number is exceeded, we allocate more
1308 space, so it is not a hard limit. */
1309 #ifndef INIT_FAILURE_ALLOC
1310 # define INIT_FAILURE_ALLOC 20
1311 #endif
1312
1313 /* Roughly the maximum number of failure points on the stack. Would be
1314 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1315 This is a variable only so users of regex can assign to it; we never
1316 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1317 before using it, so it should probably be a byte-count instead. */
1318 # if defined MATCH_MAY_ALLOCATE
1319 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1320 whose default stack limit is 2mb. In order for a larger
1321 value to work reliably, you have to try to make it accord
1322 with the process stack limit. */
1323 size_t re_max_failures = 40000;
1324 # else
1325 size_t re_max_failures = 4000;
1326 # endif
1327
1328 union fail_stack_elt
1329 {
1330 re_char *pointer;
1331 /* This should be the biggest `int' that's no bigger than a pointer. */
1332 long integer;
1333 };
1334
1335 typedef union fail_stack_elt fail_stack_elt_t;
1336
1337 typedef struct
1338 {
1339 fail_stack_elt_t *stack;
1340 size_t size;
1341 size_t avail; /* Offset of next open position. */
1342 size_t frame; /* Offset of the cur constructed frame. */
1343 } fail_stack_type;
1344
1345 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1346
1347
1348 /* Define macros to initialize and free the failure stack.
1349 Do `return -2' if the alloc fails. */
1350
1351 #ifdef MATCH_MAY_ALLOCATE
1352 # define INIT_FAIL_STACK() \
1353 do { \
1354 fail_stack.stack = \
1355 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1356 * sizeof (fail_stack_elt_t)); \
1357 \
1358 if (fail_stack.stack == NULL) \
1359 return -2; \
1360 \
1361 fail_stack.size = INIT_FAILURE_ALLOC; \
1362 fail_stack.avail = 0; \
1363 fail_stack.frame = 0; \
1364 } while (0)
1365 #else
1366 # define INIT_FAIL_STACK() \
1367 do { \
1368 fail_stack.avail = 0; \
1369 fail_stack.frame = 0; \
1370 } while (0)
1371
1372 # define RETALLOC_IF(addr, n, t) \
1373 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1374 #endif
1375
1376
1377 /* Double the size of FAIL_STACK, up to a limit
1378 which allows approximately `re_max_failures' items.
1379
1380 Return 1 if succeeds, and 0 if either ran out of memory
1381 allocating space for it or it was already too large.
1382
1383 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1384
1385 /* Factor to increase the failure stack size by
1386 when we increase it.
1387 This used to be 2, but 2 was too wasteful
1388 because the old discarded stacks added up to as much space
1389 were as ultimate, maximum-size stack. */
1390 #define FAIL_STACK_GROWTH_FACTOR 4
1391
1392 #define GROW_FAIL_STACK(fail_stack) \
1393 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1394 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1395 ? 0 \
1396 : ((fail_stack).stack \
1397 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1398 (fail_stack).size * sizeof (fail_stack_elt_t), \
1399 min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1400 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1401 * FAIL_STACK_GROWTH_FACTOR))), \
1402 \
1403 (fail_stack).stack == NULL \
1404 ? 0 \
1405 : ((fail_stack).size \
1406 = (min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1407 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1408 * FAIL_STACK_GROWTH_FACTOR)) \
1409 / sizeof (fail_stack_elt_t)), \
1410 1)))
1411
1412
1413 /* Push a pointer value onto the failure stack.
1414 Assumes the variable `fail_stack'. Probably should only
1415 be called from within `PUSH_FAILURE_POINT'. */
1416 #define PUSH_FAILURE_POINTER(item) \
1417 fail_stack.stack[fail_stack.avail++].pointer = (item)
1418
1419 /* This pushes an integer-valued item onto the failure stack.
1420 Assumes the variable `fail_stack'. Probably should only
1421 be called from within `PUSH_FAILURE_POINT'. */
1422 #define PUSH_FAILURE_INT(item) \
1423 fail_stack.stack[fail_stack.avail++].integer = (item)
1424
1425 /* These POP... operations complement the PUSH... operations.
1426 All assume that `fail_stack' is nonempty. */
1427 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1428 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1429
1430 /* Individual items aside from the registers. */
1431 #define NUM_NONREG_ITEMS 3
1432
1433 /* Used to examine the stack (to detect infinite loops). */
1434 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1435 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1436 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1437 #define TOP_FAILURE_HANDLE() fail_stack.frame
1438
1439
1440 #define ENSURE_FAIL_STACK(space) \
1441 while (REMAINING_AVAIL_SLOTS <= space) { \
1442 if (!GROW_FAIL_STACK (fail_stack)) \
1443 return -2; \
1444 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1445 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1446 }
1447
1448 /* Push register NUM onto the stack. */
1449 #define PUSH_FAILURE_REG(num) \
1450 do { \
1451 char *destination; \
1452 long n = num; \
1453 ENSURE_FAIL_STACK(3); \
1454 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1455 n, regstart[n], regend[n]); \
1456 PUSH_FAILURE_POINTER (regstart[n]); \
1457 PUSH_FAILURE_POINTER (regend[n]); \
1458 PUSH_FAILURE_INT (n); \
1459 } while (0)
1460
1461 /* Change the counter's value to VAL, but make sure that it will
1462 be reset when backtracking. */
1463 #define PUSH_NUMBER(ptr,val) \
1464 do { \
1465 char *destination; \
1466 int c; \
1467 ENSURE_FAIL_STACK(3); \
1468 EXTRACT_NUMBER (c, ptr); \
1469 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1470 PUSH_FAILURE_INT (c); \
1471 PUSH_FAILURE_POINTER (ptr); \
1472 PUSH_FAILURE_INT (-1); \
1473 STORE_NUMBER (ptr, val); \
1474 } while (0)
1475
1476 /* Pop a saved register off the stack. */
1477 #define POP_FAILURE_REG_OR_COUNT() \
1478 do { \
1479 long pfreg = POP_FAILURE_INT (); \
1480 if (pfreg == -1) \
1481 { \
1482 /* It's a counter. */ \
1483 /* Here, we discard `const', making re_match non-reentrant. */ \
1484 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1485 pfreg = POP_FAILURE_INT (); \
1486 STORE_NUMBER (ptr, pfreg); \
1487 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1488 } \
1489 else \
1490 { \
1491 regend[pfreg] = POP_FAILURE_POINTER (); \
1492 regstart[pfreg] = POP_FAILURE_POINTER (); \
1493 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1494 pfreg, regstart[pfreg], regend[pfreg]); \
1495 } \
1496 } while (0)
1497
1498 /* Check that we are not stuck in an infinite loop. */
1499 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1500 do { \
1501 ssize_t failure = TOP_FAILURE_HANDLE (); \
1502 /* Check for infinite matching loops */ \
1503 while (failure > 0 \
1504 && (FAILURE_STR (failure) == string_place \
1505 || FAILURE_STR (failure) == NULL)) \
1506 { \
1507 assert (FAILURE_PAT (failure) >= bufp->buffer \
1508 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1509 if (FAILURE_PAT (failure) == pat_cur) \
1510 { \
1511 cycle = 1; \
1512 break; \
1513 } \
1514 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1515 failure = NEXT_FAILURE_HANDLE(failure); \
1516 } \
1517 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1518 } while (0)
1519
1520 /* Push the information about the state we will need
1521 if we ever fail back to it.
1522
1523 Requires variables fail_stack, regstart, regend and
1524 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1525 declared.
1526
1527 Does `return FAILURE_CODE' if runs out of memory. */
1528
1529 #define PUSH_FAILURE_POINT(pattern, string_place) \
1530 do { \
1531 char *destination; \
1532 /* Must be int, so when we don't save any registers, the arithmetic \
1533 of 0 + -1 isn't done as unsigned. */ \
1534 \
1535 DEBUG_STATEMENT (nfailure_points_pushed++); \
1536 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1537 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1538 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1539 \
1540 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1541 \
1542 DEBUG_PRINT ("\n"); \
1543 \
1544 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1545 PUSH_FAILURE_INT (fail_stack.frame); \
1546 \
1547 DEBUG_PRINT (" Push string %p: \"", string_place); \
1548 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1549 DEBUG_PRINT ("\"\n"); \
1550 PUSH_FAILURE_POINTER (string_place); \
1551 \
1552 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1553 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1554 PUSH_FAILURE_POINTER (pattern); \
1555 \
1556 /* Close the frame by moving the frame pointer past it. */ \
1557 fail_stack.frame = fail_stack.avail; \
1558 } while (0)
1559
1560 /* Estimate the size of data pushed by a typical failure stack entry.
1561 An estimate is all we need, because all we use this for
1562 is to choose a limit for how big to make the failure stack. */
1563 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1564 #define TYPICAL_FAILURE_SIZE 20
1565
1566 /* How many items can still be added to the stack without overflowing it. */
1567 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1568
1569
1570 /* Pops what PUSH_FAIL_STACK pushes.
1571
1572 We restore into the parameters, all of which should be lvalues:
1573 STR -- the saved data position.
1574 PAT -- the saved pattern position.
1575 REGSTART, REGEND -- arrays of string positions.
1576
1577 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1578 `pend', `string1', `size1', `string2', and `size2'. */
1579
1580 #define POP_FAILURE_POINT(str, pat) \
1581 do { \
1582 assert (!FAIL_STACK_EMPTY ()); \
1583 \
1584 /* Remove failure points and point to how many regs pushed. */ \
1585 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1586 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1587 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1588 \
1589 /* Pop the saved registers. */ \
1590 while (fail_stack.frame < fail_stack.avail) \
1591 POP_FAILURE_REG_OR_COUNT (); \
1592 \
1593 pat = POP_FAILURE_POINTER (); \
1594 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1595 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1596 \
1597 /* If the saved string location is NULL, it came from an \
1598 on_failure_keep_string_jump opcode, and we want to throw away the \
1599 saved NULL, thus retaining our current position in the string. */ \
1600 str = POP_FAILURE_POINTER (); \
1601 DEBUG_PRINT (" Popping string %p: \"", str); \
1602 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1603 DEBUG_PRINT ("\"\n"); \
1604 \
1605 fail_stack.frame = POP_FAILURE_INT (); \
1606 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1607 \
1608 assert (fail_stack.avail >= 0); \
1609 assert (fail_stack.frame <= fail_stack.avail); \
1610 \
1611 DEBUG_STATEMENT (nfailure_points_popped++); \
1612 } while (0) /* POP_FAILURE_POINT */
1613
1614
1615 \f
1616 /* Registers are set to a sentinel when they haven't yet matched. */
1617 #define REG_UNSET(e) ((e) == NULL)
1618 \f
1619 /* Subroutine declarations and macros for regex_compile. */
1620
1621 static reg_errcode_t regex_compile (re_char *pattern, size_t size,
1622 reg_syntax_t syntax,
1623 struct re_pattern_buffer *bufp);
1624 static void store_op1 (re_opcode_t op, unsigned char *loc, int arg);
1625 static void store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2);
1626 static void insert_op1 (re_opcode_t op, unsigned char *loc,
1627 int arg, unsigned char *end);
1628 static void insert_op2 (re_opcode_t op, unsigned char *loc,
1629 int arg1, int arg2, unsigned char *end);
1630 static boolean at_begline_loc_p (re_char *pattern, re_char *p,
1631 reg_syntax_t syntax);
1632 static boolean at_endline_loc_p (re_char *p, re_char *pend,
1633 reg_syntax_t syntax);
1634 static re_char *skip_one_char (re_char *p);
1635 static int analyze_first (re_char *p, re_char *pend,
1636 char *fastmap, const int multibyte);
1637
1638 /* Fetch the next character in the uncompiled pattern, with no
1639 translation. */
1640 #define PATFETCH(c) \
1641 do { \
1642 int len; \
1643 if (p == pend) return REG_EEND; \
1644 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1645 p += len; \
1646 } while (0)
1647
1648
1649 /* If `translate' is non-null, return translate[D], else just D. We
1650 cast the subscript to translate because some data is declared as
1651 `char *', to avoid warnings when a string constant is passed. But
1652 when we use a character as a subscript we must make it unsigned. */
1653 #ifndef TRANSLATE
1654 # define TRANSLATE(d) \
1655 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1656 #endif
1657
1658
1659 /* Macros for outputting the compiled pattern into `buffer'. */
1660
1661 /* If the buffer isn't allocated when it comes in, use this. */
1662 #define INIT_BUF_SIZE 32
1663
1664 /* Make sure we have at least N more bytes of space in buffer. */
1665 #define GET_BUFFER_SPACE(n) \
1666 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1667 EXTEND_BUFFER ()
1668
1669 /* Make sure we have one more byte of buffer space and then add C to it. */
1670 #define BUF_PUSH(c) \
1671 do { \
1672 GET_BUFFER_SPACE (1); \
1673 *b++ = (unsigned char) (c); \
1674 } while (0)
1675
1676
1677 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1678 #define BUF_PUSH_2(c1, c2) \
1679 do { \
1680 GET_BUFFER_SPACE (2); \
1681 *b++ = (unsigned char) (c1); \
1682 *b++ = (unsigned char) (c2); \
1683 } while (0)
1684
1685
1686 /* Store a jump with opcode OP at LOC to location TO. We store a
1687 relative address offset by the three bytes the jump itself occupies. */
1688 #define STORE_JUMP(op, loc, to) \
1689 store_op1 (op, loc, (to) - (loc) - 3)
1690
1691 /* Likewise, for a two-argument jump. */
1692 #define STORE_JUMP2(op, loc, to, arg) \
1693 store_op2 (op, loc, (to) - (loc) - 3, arg)
1694
1695 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1696 #define INSERT_JUMP(op, loc, to) \
1697 insert_op1 (op, loc, (to) - (loc) - 3, b)
1698
1699 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1700 #define INSERT_JUMP2(op, loc, to, arg) \
1701 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1702
1703
1704 /* This is not an arbitrary limit: the arguments which represent offsets
1705 into the pattern are two bytes long. So if 2^15 bytes turns out to
1706 be too small, many things would have to change. */
1707 # define MAX_BUF_SIZE (1L << 15)
1708
1709 /* Extend the buffer by twice its current size via realloc and
1710 reset the pointers that pointed into the old block to point to the
1711 correct places in the new one. If extending the buffer results in it
1712 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1713 #if __BOUNDED_POINTERS__
1714 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1715 # define MOVE_BUFFER_POINTER(P) \
1716 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1717 SET_HIGH_BOUND (P), \
1718 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1719 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1720 else \
1721 { \
1722 SET_HIGH_BOUND (b); \
1723 SET_HIGH_BOUND (begalt); \
1724 if (fixup_alt_jump) \
1725 SET_HIGH_BOUND (fixup_alt_jump); \
1726 if (laststart) \
1727 SET_HIGH_BOUND (laststart); \
1728 if (pending_exact) \
1729 SET_HIGH_BOUND (pending_exact); \
1730 }
1731 #else
1732 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1733 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1734 #endif
1735 #define EXTEND_BUFFER() \
1736 do { \
1737 unsigned char *old_buffer = bufp->buffer; \
1738 if (bufp->allocated == MAX_BUF_SIZE) \
1739 return REG_ESIZE; \
1740 bufp->allocated <<= 1; \
1741 if (bufp->allocated > MAX_BUF_SIZE) \
1742 bufp->allocated = MAX_BUF_SIZE; \
1743 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1744 if (bufp->buffer == NULL) \
1745 return REG_ESPACE; \
1746 /* If the buffer moved, move all the pointers into it. */ \
1747 if (old_buffer != bufp->buffer) \
1748 { \
1749 unsigned char *new_buffer = bufp->buffer; \
1750 MOVE_BUFFER_POINTER (b); \
1751 MOVE_BUFFER_POINTER (begalt); \
1752 if (fixup_alt_jump) \
1753 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1754 if (laststart) \
1755 MOVE_BUFFER_POINTER (laststart); \
1756 if (pending_exact) \
1757 MOVE_BUFFER_POINTER (pending_exact); \
1758 } \
1759 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1760 } while (0)
1761
1762
1763 /* Since we have one byte reserved for the register number argument to
1764 {start,stop}_memory, the maximum number of groups we can report
1765 things about is what fits in that byte. */
1766 #define MAX_REGNUM 255
1767
1768 /* But patterns can have more than `MAX_REGNUM' registers. We just
1769 ignore the excess. */
1770 typedef int regnum_t;
1771
1772
1773 /* Macros for the compile stack. */
1774
1775 /* Since offsets can go either forwards or backwards, this type needs to
1776 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1777 /* int may be not enough when sizeof(int) == 2. */
1778 typedef long pattern_offset_t;
1779
1780 typedef struct
1781 {
1782 pattern_offset_t begalt_offset;
1783 pattern_offset_t fixup_alt_jump;
1784 pattern_offset_t laststart_offset;
1785 regnum_t regnum;
1786 } compile_stack_elt_t;
1787
1788
1789 typedef struct
1790 {
1791 compile_stack_elt_t *stack;
1792 size_t size;
1793 size_t avail; /* Offset of next open position. */
1794 } compile_stack_type;
1795
1796
1797 #define INIT_COMPILE_STACK_SIZE 32
1798
1799 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1800 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1801
1802 /* The next available element. */
1803 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1804
1805 /* Explicit quit checking is needed for Emacs, which uses polling to
1806 process input events. */
1807 #ifdef emacs
1808 # define IMMEDIATE_QUIT_CHECK \
1809 do { \
1810 if (immediate_quit) QUIT; \
1811 } while (0)
1812 #else
1813 # define IMMEDIATE_QUIT_CHECK ((void)0)
1814 #endif
1815 \f
1816 /* Structure to manage work area for range table. */
1817 struct range_table_work_area
1818 {
1819 int *table; /* actual work area. */
1820 int allocated; /* allocated size for work area in bytes. */
1821 int used; /* actually used size in words. */
1822 int bits; /* flag to record character classes */
1823 };
1824
1825 #ifdef emacs
1826
1827 /* Make sure that WORK_AREA can hold more N multibyte characters.
1828 This is used only in set_image_of_range and set_image_of_range_1.
1829 It expects WORK_AREA to be a pointer.
1830 If it can't get the space, it returns from the surrounding function. */
1831
1832 #define EXTEND_RANGE_TABLE(work_area, n) \
1833 do { \
1834 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1835 { \
1836 extend_range_table_work_area (&work_area); \
1837 if ((work_area).table == 0) \
1838 return (REG_ESPACE); \
1839 } \
1840 } while (0)
1841
1842 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1843 (work_area).bits |= (bit)
1844
1845 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1846 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1847 do { \
1848 EXTEND_RANGE_TABLE ((work_area), 2); \
1849 (work_area).table[(work_area).used++] = (range_start); \
1850 (work_area).table[(work_area).used++] = (range_end); \
1851 } while (0)
1852
1853 #endif /* emacs */
1854
1855 /* Free allocated memory for WORK_AREA. */
1856 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1857 do { \
1858 if ((work_area).table) \
1859 free ((work_area).table); \
1860 } while (0)
1861
1862 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1863 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1864 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1865 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1866
1867 /* Bits used to implement the multibyte-part of the various character classes
1868 such as [:alnum:] in a charset's range table. The code currently assumes
1869 that only the low 16 bits are used. */
1870 #define BIT_WORD 0x1
1871 #define BIT_LOWER 0x2
1872 #define BIT_PUNCT 0x4
1873 #define BIT_SPACE 0x8
1874 #define BIT_UPPER 0x10
1875 #define BIT_MULTIBYTE 0x20
1876 #define BIT_ALPHA 0x40
1877 #define BIT_ALNUM 0x80
1878 #define BIT_GRAPH 0x100
1879 #define BIT_PRINT 0x200
1880 \f
1881
1882 /* Set the bit for character C in a list. */
1883 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1884
1885
1886 #ifdef emacs
1887
1888 /* Store characters in the range FROM to TO in the bitmap at B (for
1889 ASCII and unibyte characters) and WORK_AREA (for multibyte
1890 characters) while translating them and paying attention to the
1891 continuity of translated characters.
1892
1893 Implementation note: It is better to implement these fairly big
1894 macros by a function, but it's not that easy because macros called
1895 in this macro assume various local variables already declared. */
1896
1897 /* Both FROM and TO are ASCII characters. */
1898
1899 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1900 do { \
1901 int C0, C1; \
1902 \
1903 for (C0 = (FROM); C0 <= (TO); C0++) \
1904 { \
1905 C1 = TRANSLATE (C0); \
1906 if (! ASCII_CHAR_P (C1)) \
1907 { \
1908 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1909 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1910 C1 = C0; \
1911 } \
1912 SET_LIST_BIT (C1); \
1913 } \
1914 } while (0)
1915
1916
1917 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1918
1919 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1920 do { \
1921 int C0, C1, C2, I; \
1922 int USED = RANGE_TABLE_WORK_USED (work_area); \
1923 \
1924 for (C0 = (FROM); C0 <= (TO); C0++) \
1925 { \
1926 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1927 if (CHAR_BYTE8_P (C1)) \
1928 SET_LIST_BIT (C0); \
1929 else \
1930 { \
1931 C2 = TRANSLATE (C1); \
1932 if (C2 == C1 \
1933 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1934 C1 = C0; \
1935 SET_LIST_BIT (C1); \
1936 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1937 { \
1938 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1939 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1940 \
1941 if (C2 >= from - 1 && C2 <= to + 1) \
1942 { \
1943 if (C2 == from - 1) \
1944 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1945 else if (C2 == to + 1) \
1946 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1947 break; \
1948 } \
1949 } \
1950 if (I < USED) \
1951 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1952 } \
1953 } \
1954 } while (0)
1955
1956
1957 /* Both FROM and TO are multibyte characters. */
1958
1959 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1960 do { \
1961 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1962 \
1963 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1964 for (C0 = (FROM); C0 <= (TO); C0++) \
1965 { \
1966 C1 = TRANSLATE (C0); \
1967 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1968 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1969 SET_LIST_BIT (C2); \
1970 if (C1 >= (FROM) && C1 <= (TO)) \
1971 continue; \
1972 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1973 { \
1974 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1975 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1976 \
1977 if (C1 >= from - 1 && C1 <= to + 1) \
1978 { \
1979 if (C1 == from - 1) \
1980 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1981 else if (C1 == to + 1) \
1982 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1983 break; \
1984 } \
1985 } \
1986 if (I < USED) \
1987 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1988 } \
1989 } while (0)
1990
1991 #endif /* emacs */
1992
1993 /* Get the next unsigned number in the uncompiled pattern. */
1994 #define GET_INTERVAL_COUNT(num) \
1995 do { \
1996 if (p == pend) \
1997 FREE_STACK_RETURN (REG_EBRACE); \
1998 else \
1999 { \
2000 PATFETCH (c); \
2001 while ('0' <= c && c <= '9') \
2002 { \
2003 if (num < 0) \
2004 num = 0; \
2005 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
2006 FREE_STACK_RETURN (REG_BADBR); \
2007 num = num * 10 + c - '0'; \
2008 if (p == pend) \
2009 FREE_STACK_RETURN (REG_EBRACE); \
2010 PATFETCH (c); \
2011 } \
2012 } \
2013 } while (0)
2014 \f
2015 #if ! WIDE_CHAR_SUPPORT
2016
2017 /* Map a string to the char class it names (if any). */
2018 re_wctype_t
2019 re_wctype (const_re_char *str)
2020 {
2021 const char *string = (const char *) str;
2022 if (STREQ (string, "alnum")) return RECC_ALNUM;
2023 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2024 else if (STREQ (string, "word")) return RECC_WORD;
2025 else if (STREQ (string, "ascii")) return RECC_ASCII;
2026 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2027 else if (STREQ (string, "graph")) return RECC_GRAPH;
2028 else if (STREQ (string, "lower")) return RECC_LOWER;
2029 else if (STREQ (string, "print")) return RECC_PRINT;
2030 else if (STREQ (string, "punct")) return RECC_PUNCT;
2031 else if (STREQ (string, "space")) return RECC_SPACE;
2032 else if (STREQ (string, "upper")) return RECC_UPPER;
2033 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2034 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2035 else if (STREQ (string, "digit")) return RECC_DIGIT;
2036 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2037 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2038 else if (STREQ (string, "blank")) return RECC_BLANK;
2039 else return 0;
2040 }
2041
2042 /* True if CH is in the char class CC. */
2043 boolean
2044 re_iswctype (int ch, re_wctype_t cc)
2045 {
2046 switch (cc)
2047 {
2048 case RECC_ALNUM: return ISALNUM (ch) != 0;
2049 case RECC_ALPHA: return ISALPHA (ch) != 0;
2050 case RECC_BLANK: return ISBLANK (ch) != 0;
2051 case RECC_CNTRL: return ISCNTRL (ch) != 0;
2052 case RECC_DIGIT: return ISDIGIT (ch) != 0;
2053 case RECC_GRAPH: return ISGRAPH (ch) != 0;
2054 case RECC_LOWER: return ISLOWER (ch) != 0;
2055 case RECC_PRINT: return ISPRINT (ch) != 0;
2056 case RECC_PUNCT: return ISPUNCT (ch) != 0;
2057 case RECC_SPACE: return ISSPACE (ch) != 0;
2058 case RECC_UPPER: return ISUPPER (ch) != 0;
2059 case RECC_XDIGIT: return ISXDIGIT (ch) != 0;
2060 case RECC_ASCII: return IS_REAL_ASCII (ch) != 0;
2061 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2062 case RECC_UNIBYTE: return ISUNIBYTE (ch) != 0;
2063 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2064 case RECC_WORD: return ISWORD (ch) != 0;
2065 case RECC_ERROR: return false;
2066 default:
2067 abort ();
2068 }
2069 }
2070
2071 /* Return a bit-pattern to use in the range-table bits to match multibyte
2072 chars of class CC. */
2073 static int
2074 re_wctype_to_bit (re_wctype_t cc)
2075 {
2076 switch (cc)
2077 {
2078 case RECC_NONASCII:
2079 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2080 case RECC_ALPHA: return BIT_ALPHA;
2081 case RECC_ALNUM: return BIT_ALNUM;
2082 case RECC_WORD: return BIT_WORD;
2083 case RECC_LOWER: return BIT_LOWER;
2084 case RECC_UPPER: return BIT_UPPER;
2085 case RECC_PUNCT: return BIT_PUNCT;
2086 case RECC_SPACE: return BIT_SPACE;
2087 case RECC_GRAPH: return BIT_GRAPH;
2088 case RECC_PRINT: return BIT_PRINT;
2089 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2090 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2091 default:
2092 abort ();
2093 }
2094 }
2095 #endif
2096 \f
2097 /* Filling in the work area of a range. */
2098
2099 /* Actually extend the space in WORK_AREA. */
2100
2101 static void
2102 extend_range_table_work_area (struct range_table_work_area *work_area)
2103 {
2104 work_area->allocated += 16 * sizeof (int);
2105 work_area->table = realloc (work_area->table, work_area->allocated);
2106 }
2107
2108 #if 0
2109 #ifdef emacs
2110
2111 /* Carefully find the ranges of codes that are equivalent
2112 under case conversion to the range start..end when passed through
2113 TRANSLATE. Handle the case where non-letters can come in between
2114 two upper-case letters (which happens in Latin-1).
2115 Also handle the case of groups of more than 2 case-equivalent chars.
2116
2117 The basic method is to look at consecutive characters and see
2118 if they can form a run that can be handled as one.
2119
2120 Returns -1 if successful, REG_ESPACE if ran out of space. */
2121
2122 static int
2123 set_image_of_range_1 (struct range_table_work_area *work_area,
2124 re_wchar_t start, re_wchar_t end,
2125 RE_TRANSLATE_TYPE translate)
2126 {
2127 /* `one_case' indicates a character, or a run of characters,
2128 each of which is an isolate (no case-equivalents).
2129 This includes all ASCII non-letters.
2130
2131 `two_case' indicates a character, or a run of characters,
2132 each of which has two case-equivalent forms.
2133 This includes all ASCII letters.
2134
2135 `strange' indicates a character that has more than one
2136 case-equivalent. */
2137
2138 enum case_type {one_case, two_case, strange};
2139
2140 /* Describe the run that is in progress,
2141 which the next character can try to extend.
2142 If run_type is strange, that means there really is no run.
2143 If run_type is one_case, then run_start...run_end is the run.
2144 If run_type is two_case, then the run is run_start...run_end,
2145 and the case-equivalents end at run_eqv_end. */
2146
2147 enum case_type run_type = strange;
2148 int run_start, run_end, run_eqv_end;
2149
2150 Lisp_Object eqv_table;
2151
2152 if (!RE_TRANSLATE_P (translate))
2153 {
2154 EXTEND_RANGE_TABLE (work_area, 2);
2155 work_area->table[work_area->used++] = (start);
2156 work_area->table[work_area->used++] = (end);
2157 return -1;
2158 }
2159
2160 eqv_table = XCHAR_TABLE (translate)->extras[2];
2161
2162 for (; start <= end; start++)
2163 {
2164 enum case_type this_type;
2165 int eqv = RE_TRANSLATE (eqv_table, start);
2166 int minchar, maxchar;
2167
2168 /* Classify this character */
2169 if (eqv == start)
2170 this_type = one_case;
2171 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2172 this_type = two_case;
2173 else
2174 this_type = strange;
2175
2176 if (start < eqv)
2177 minchar = start, maxchar = eqv;
2178 else
2179 minchar = eqv, maxchar = start;
2180
2181 /* Can this character extend the run in progress? */
2182 if (this_type == strange || this_type != run_type
2183 || !(minchar == run_end + 1
2184 && (run_type == two_case
2185 ? maxchar == run_eqv_end + 1 : 1)))
2186 {
2187 /* No, end the run.
2188 Record each of its equivalent ranges. */
2189 if (run_type == one_case)
2190 {
2191 EXTEND_RANGE_TABLE (work_area, 2);
2192 work_area->table[work_area->used++] = run_start;
2193 work_area->table[work_area->used++] = run_end;
2194 }
2195 else if (run_type == two_case)
2196 {
2197 EXTEND_RANGE_TABLE (work_area, 4);
2198 work_area->table[work_area->used++] = run_start;
2199 work_area->table[work_area->used++] = run_end;
2200 work_area->table[work_area->used++]
2201 = RE_TRANSLATE (eqv_table, run_start);
2202 work_area->table[work_area->used++]
2203 = RE_TRANSLATE (eqv_table, run_end);
2204 }
2205 run_type = strange;
2206 }
2207
2208 if (this_type == strange)
2209 {
2210 /* For a strange character, add each of its equivalents, one
2211 by one. Don't start a range. */
2212 do
2213 {
2214 EXTEND_RANGE_TABLE (work_area, 2);
2215 work_area->table[work_area->used++] = eqv;
2216 work_area->table[work_area->used++] = eqv;
2217 eqv = RE_TRANSLATE (eqv_table, eqv);
2218 }
2219 while (eqv != start);
2220 }
2221
2222 /* Add this char to the run, or start a new run. */
2223 else if (run_type == strange)
2224 {
2225 /* Initialize a new range. */
2226 run_type = this_type;
2227 run_start = start;
2228 run_end = start;
2229 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2230 }
2231 else
2232 {
2233 /* Extend a running range. */
2234 run_end = minchar;
2235 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2236 }
2237 }
2238
2239 /* If a run is still in progress at the end, finish it now
2240 by recording its equivalent ranges. */
2241 if (run_type == one_case)
2242 {
2243 EXTEND_RANGE_TABLE (work_area, 2);
2244 work_area->table[work_area->used++] = run_start;
2245 work_area->table[work_area->used++] = run_end;
2246 }
2247 else if (run_type == two_case)
2248 {
2249 EXTEND_RANGE_TABLE (work_area, 4);
2250 work_area->table[work_area->used++] = run_start;
2251 work_area->table[work_area->used++] = run_end;
2252 work_area->table[work_area->used++]
2253 = RE_TRANSLATE (eqv_table, run_start);
2254 work_area->table[work_area->used++]
2255 = RE_TRANSLATE (eqv_table, run_end);
2256 }
2257
2258 return -1;
2259 }
2260
2261 #endif /* emacs */
2262
2263 /* Record the image of the range start..end when passed through
2264 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2265 and is not even necessarily contiguous.
2266 Normally we approximate it with the smallest contiguous range that contains
2267 all the chars we need. However, for Latin-1 we go to extra effort
2268 to do a better job.
2269
2270 This function is not called for ASCII ranges.
2271
2272 Returns -1 if successful, REG_ESPACE if ran out of space. */
2273
2274 static int
2275 set_image_of_range (struct range_table_work_area *work_area,
2276 re_wchar_t start, re_wchar_t end,
2277 RE_TRANSLATE_TYPE translate)
2278 {
2279 re_wchar_t cmin, cmax;
2280
2281 #ifdef emacs
2282 /* For Latin-1 ranges, use set_image_of_range_1
2283 to get proper handling of ranges that include letters and nonletters.
2284 For a range that includes the whole of Latin-1, this is not necessary.
2285 For other character sets, we don't bother to get this right. */
2286 if (RE_TRANSLATE_P (translate) && start < 04400
2287 && !(start < 04200 && end >= 04377))
2288 {
2289 int newend;
2290 int tem;
2291 newend = end;
2292 if (newend > 04377)
2293 newend = 04377;
2294 tem = set_image_of_range_1 (work_area, start, newend, translate);
2295 if (tem > 0)
2296 return tem;
2297
2298 start = 04400;
2299 if (end < 04400)
2300 return -1;
2301 }
2302 #endif
2303
2304 EXTEND_RANGE_TABLE (work_area, 2);
2305 work_area->table[work_area->used++] = (start);
2306 work_area->table[work_area->used++] = (end);
2307
2308 cmin = -1, cmax = -1;
2309
2310 if (RE_TRANSLATE_P (translate))
2311 {
2312 int ch;
2313
2314 for (ch = start; ch <= end; ch++)
2315 {
2316 re_wchar_t c = TRANSLATE (ch);
2317 if (! (start <= c && c <= end))
2318 {
2319 if (cmin == -1)
2320 cmin = c, cmax = c;
2321 else
2322 {
2323 cmin = min (cmin, c);
2324 cmax = max (cmax, c);
2325 }
2326 }
2327 }
2328
2329 if (cmin != -1)
2330 {
2331 EXTEND_RANGE_TABLE (work_area, 2);
2332 work_area->table[work_area->used++] = (cmin);
2333 work_area->table[work_area->used++] = (cmax);
2334 }
2335 }
2336
2337 return -1;
2338 }
2339 #endif /* 0 */
2340 \f
2341 #ifndef MATCH_MAY_ALLOCATE
2342
2343 /* If we cannot allocate large objects within re_match_2_internal,
2344 we make the fail stack and register vectors global.
2345 The fail stack, we grow to the maximum size when a regexp
2346 is compiled.
2347 The register vectors, we adjust in size each time we
2348 compile a regexp, according to the number of registers it needs. */
2349
2350 static fail_stack_type fail_stack;
2351
2352 /* Size with which the following vectors are currently allocated.
2353 That is so we can make them bigger as needed,
2354 but never make them smaller. */
2355 static int regs_allocated_size;
2356
2357 static re_char ** regstart, ** regend;
2358 static re_char **best_regstart, **best_regend;
2359
2360 /* Make the register vectors big enough for NUM_REGS registers,
2361 but don't make them smaller. */
2362
2363 static
2364 regex_grow_registers (int num_regs)
2365 {
2366 if (num_regs > regs_allocated_size)
2367 {
2368 RETALLOC_IF (regstart, num_regs, re_char *);
2369 RETALLOC_IF (regend, num_regs, re_char *);
2370 RETALLOC_IF (best_regstart, num_regs, re_char *);
2371 RETALLOC_IF (best_regend, num_regs, re_char *);
2372
2373 regs_allocated_size = num_regs;
2374 }
2375 }
2376
2377 #endif /* not MATCH_MAY_ALLOCATE */
2378 \f
2379 static boolean group_in_compile_stack (compile_stack_type compile_stack,
2380 regnum_t regnum);
2381
2382 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2383 Returns one of error codes defined in `regex.h', or zero for success.
2384
2385 Assumes the `allocated' (and perhaps `buffer') and `translate'
2386 fields are set in BUFP on entry.
2387
2388 If it succeeds, results are put in BUFP (if it returns an error, the
2389 contents of BUFP are undefined):
2390 `buffer' is the compiled pattern;
2391 `syntax' is set to SYNTAX;
2392 `used' is set to the length of the compiled pattern;
2393 `fastmap_accurate' is zero;
2394 `re_nsub' is the number of subexpressions in PATTERN;
2395 `not_bol' and `not_eol' are zero;
2396
2397 The `fastmap' field is neither examined nor set. */
2398
2399 /* Insert the `jump' from the end of last alternative to "here".
2400 The space for the jump has already been allocated. */
2401 #define FIXUP_ALT_JUMP() \
2402 do { \
2403 if (fixup_alt_jump) \
2404 STORE_JUMP (jump, fixup_alt_jump, b); \
2405 } while (0)
2406
2407
2408 /* Return, freeing storage we allocated. */
2409 #define FREE_STACK_RETURN(value) \
2410 do { \
2411 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2412 free (compile_stack.stack); \
2413 return value; \
2414 } while (0)
2415
2416 static reg_errcode_t
2417 regex_compile (const_re_char *pattern, size_t size, reg_syntax_t syntax,
2418 struct re_pattern_buffer *bufp)
2419 {
2420 /* We fetch characters from PATTERN here. */
2421 register re_wchar_t c, c1;
2422
2423 /* Points to the end of the buffer, where we should append. */
2424 register unsigned char *b;
2425
2426 /* Keeps track of unclosed groups. */
2427 compile_stack_type compile_stack;
2428
2429 /* Points to the current (ending) position in the pattern. */
2430 #ifdef AIX
2431 /* `const' makes AIX compiler fail. */
2432 unsigned char *p = pattern;
2433 #else
2434 re_char *p = pattern;
2435 #endif
2436 re_char *pend = pattern + size;
2437
2438 /* How to translate the characters in the pattern. */
2439 RE_TRANSLATE_TYPE translate = bufp->translate;
2440
2441 /* Address of the count-byte of the most recently inserted `exactn'
2442 command. This makes it possible to tell if a new exact-match
2443 character can be added to that command or if the character requires
2444 a new `exactn' command. */
2445 unsigned char *pending_exact = 0;
2446
2447 /* Address of start of the most recently finished expression.
2448 This tells, e.g., postfix * where to find the start of its
2449 operand. Reset at the beginning of groups and alternatives. */
2450 unsigned char *laststart = 0;
2451
2452 /* Address of beginning of regexp, or inside of last group. */
2453 unsigned char *begalt;
2454
2455 /* Place in the uncompiled pattern (i.e., the {) to
2456 which to go back if the interval is invalid. */
2457 re_char *beg_interval;
2458
2459 /* Address of the place where a forward jump should go to the end of
2460 the containing expression. Each alternative of an `or' -- except the
2461 last -- ends with a forward jump of this sort. */
2462 unsigned char *fixup_alt_jump = 0;
2463
2464 /* Work area for range table of charset. */
2465 struct range_table_work_area range_table_work;
2466
2467 /* If the object matched can contain multibyte characters. */
2468 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2469
2470 /* Nonzero if we have pushed down into a subpattern. */
2471 int in_subpattern = 0;
2472
2473 /* These hold the values of p, pattern, and pend from the main
2474 pattern when we have pushed into a subpattern. */
2475 re_char *main_p IF_LINT (= NULL);
2476 re_char *main_pattern IF_LINT (= NULL);
2477 re_char *main_pend IF_LINT (= NULL);
2478
2479 #ifdef DEBUG
2480 debug++;
2481 DEBUG_PRINT ("\nCompiling pattern: ");
2482 if (debug > 0)
2483 {
2484 unsigned debug_count;
2485
2486 for (debug_count = 0; debug_count < size; debug_count++)
2487 putchar (pattern[debug_count]);
2488 putchar ('\n');
2489 }
2490 #endif /* DEBUG */
2491
2492 /* Initialize the compile stack. */
2493 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2494 if (compile_stack.stack == NULL)
2495 return REG_ESPACE;
2496
2497 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2498 compile_stack.avail = 0;
2499
2500 range_table_work.table = 0;
2501 range_table_work.allocated = 0;
2502
2503 /* Initialize the pattern buffer. */
2504 bufp->syntax = syntax;
2505 bufp->fastmap_accurate = 0;
2506 bufp->not_bol = bufp->not_eol = 0;
2507 bufp->used_syntax = 0;
2508
2509 /* Set `used' to zero, so that if we return an error, the pattern
2510 printer (for debugging) will think there's no pattern. We reset it
2511 at the end. */
2512 bufp->used = 0;
2513
2514 /* Always count groups, whether or not bufp->no_sub is set. */
2515 bufp->re_nsub = 0;
2516
2517 #if !defined emacs && !defined SYNTAX_TABLE
2518 /* Initialize the syntax table. */
2519 init_syntax_once ();
2520 #endif
2521
2522 if (bufp->allocated == 0)
2523 {
2524 if (bufp->buffer)
2525 { /* If zero allocated, but buffer is non-null, try to realloc
2526 enough space. This loses if buffer's address is bogus, but
2527 that is the user's responsibility. */
2528 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2529 }
2530 else
2531 { /* Caller did not allocate a buffer. Do it for them. */
2532 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2533 }
2534 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2535
2536 bufp->allocated = INIT_BUF_SIZE;
2537 }
2538
2539 begalt = b = bufp->buffer;
2540
2541 /* Loop through the uncompiled pattern until we're at the end. */
2542 while (1)
2543 {
2544 if (p == pend)
2545 {
2546 /* If this is the end of an included regexp,
2547 pop back to the main regexp and try again. */
2548 if (in_subpattern)
2549 {
2550 in_subpattern = 0;
2551 pattern = main_pattern;
2552 p = main_p;
2553 pend = main_pend;
2554 continue;
2555 }
2556 /* If this is the end of the main regexp, we are done. */
2557 break;
2558 }
2559
2560 PATFETCH (c);
2561
2562 switch (c)
2563 {
2564 case ' ':
2565 {
2566 re_char *p1 = p;
2567
2568 /* If there's no special whitespace regexp, treat
2569 spaces normally. And don't try to do this recursively. */
2570 if (!whitespace_regexp || in_subpattern)
2571 goto normal_char;
2572
2573 /* Peek past following spaces. */
2574 while (p1 != pend)
2575 {
2576 if (*p1 != ' ')
2577 break;
2578 p1++;
2579 }
2580 /* If the spaces are followed by a repetition op,
2581 treat them normally. */
2582 if (p1 != pend
2583 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2584 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2585 goto normal_char;
2586
2587 /* Replace the spaces with the whitespace regexp. */
2588 in_subpattern = 1;
2589 main_p = p1;
2590 main_pend = pend;
2591 main_pattern = pattern;
2592 p = pattern = whitespace_regexp;
2593 pend = p + strlen ((const char *) p);
2594 break;
2595 }
2596
2597 case '^':
2598 {
2599 if ( /* If at start of pattern, it's an operator. */
2600 p == pattern + 1
2601 /* If context independent, it's an operator. */
2602 || syntax & RE_CONTEXT_INDEP_ANCHORS
2603 /* Otherwise, depends on what's come before. */
2604 || at_begline_loc_p (pattern, p, syntax))
2605 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2606 else
2607 goto normal_char;
2608 }
2609 break;
2610
2611
2612 case '$':
2613 {
2614 if ( /* If at end of pattern, it's an operator. */
2615 p == pend
2616 /* If context independent, it's an operator. */
2617 || syntax & RE_CONTEXT_INDEP_ANCHORS
2618 /* Otherwise, depends on what's next. */
2619 || at_endline_loc_p (p, pend, syntax))
2620 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2621 else
2622 goto normal_char;
2623 }
2624 break;
2625
2626
2627 case '+':
2628 case '?':
2629 if ((syntax & RE_BK_PLUS_QM)
2630 || (syntax & RE_LIMITED_OPS))
2631 goto normal_char;
2632 handle_plus:
2633 case '*':
2634 /* If there is no previous pattern... */
2635 if (!laststart)
2636 {
2637 if (syntax & RE_CONTEXT_INVALID_OPS)
2638 FREE_STACK_RETURN (REG_BADRPT);
2639 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2640 goto normal_char;
2641 }
2642
2643 {
2644 /* 1 means zero (many) matches is allowed. */
2645 boolean zero_times_ok = 0, many_times_ok = 0;
2646 boolean greedy = 1;
2647
2648 /* If there is a sequence of repetition chars, collapse it
2649 down to just one (the right one). We can't combine
2650 interval operators with these because of, e.g., `a{2}*',
2651 which should only match an even number of `a's. */
2652
2653 for (;;)
2654 {
2655 if ((syntax & RE_FRUGAL)
2656 && c == '?' && (zero_times_ok || many_times_ok))
2657 greedy = 0;
2658 else
2659 {
2660 zero_times_ok |= c != '+';
2661 many_times_ok |= c != '?';
2662 }
2663
2664 if (p == pend)
2665 break;
2666 else if (*p == '*'
2667 || (!(syntax & RE_BK_PLUS_QM)
2668 && (*p == '+' || *p == '?')))
2669 ;
2670 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2671 {
2672 if (p+1 == pend)
2673 FREE_STACK_RETURN (REG_EESCAPE);
2674 if (p[1] == '+' || p[1] == '?')
2675 PATFETCH (c); /* Gobble up the backslash. */
2676 else
2677 break;
2678 }
2679 else
2680 break;
2681 /* If we get here, we found another repeat character. */
2682 PATFETCH (c);
2683 }
2684
2685 /* Star, etc. applied to an empty pattern is equivalent
2686 to an empty pattern. */
2687 if (!laststart || laststart == b)
2688 break;
2689
2690 /* Now we know whether or not zero matches is allowed
2691 and also whether or not two or more matches is allowed. */
2692 if (greedy)
2693 {
2694 if (many_times_ok)
2695 {
2696 boolean simple = skip_one_char (laststart) == b;
2697 size_t startoffset = 0;
2698 re_opcode_t ofj =
2699 /* Check if the loop can match the empty string. */
2700 (simple || !analyze_first (laststart, b, NULL, 0))
2701 ? on_failure_jump : on_failure_jump_loop;
2702 assert (skip_one_char (laststart) <= b);
2703
2704 if (!zero_times_ok && simple)
2705 { /* Since simple * loops can be made faster by using
2706 on_failure_keep_string_jump, we turn simple P+
2707 into PP* if P is simple. */
2708 unsigned char *p1, *p2;
2709 startoffset = b - laststart;
2710 GET_BUFFER_SPACE (startoffset);
2711 p1 = b; p2 = laststart;
2712 while (p2 < p1)
2713 *b++ = *p2++;
2714 zero_times_ok = 1;
2715 }
2716
2717 GET_BUFFER_SPACE (6);
2718 if (!zero_times_ok)
2719 /* A + loop. */
2720 STORE_JUMP (ofj, b, b + 6);
2721 else
2722 /* Simple * loops can use on_failure_keep_string_jump
2723 depending on what follows. But since we don't know
2724 that yet, we leave the decision up to
2725 on_failure_jump_smart. */
2726 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2727 laststart + startoffset, b + 6);
2728 b += 3;
2729 STORE_JUMP (jump, b, laststart + startoffset);
2730 b += 3;
2731 }
2732 else
2733 {
2734 /* A simple ? pattern. */
2735 assert (zero_times_ok);
2736 GET_BUFFER_SPACE (3);
2737 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2738 b += 3;
2739 }
2740 }
2741 else /* not greedy */
2742 { /* I wish the greedy and non-greedy cases could be merged. */
2743
2744 GET_BUFFER_SPACE (7); /* We might use less. */
2745 if (many_times_ok)
2746 {
2747 boolean emptyp = analyze_first (laststart, b, NULL, 0);
2748
2749 /* The non-greedy multiple match looks like
2750 a repeat..until: we only need a conditional jump
2751 at the end of the loop. */
2752 if (emptyp) BUF_PUSH (no_op);
2753 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2754 : on_failure_jump, b, laststart);
2755 b += 3;
2756 if (zero_times_ok)
2757 {
2758 /* The repeat...until naturally matches one or more.
2759 To also match zero times, we need to first jump to
2760 the end of the loop (its conditional jump). */
2761 INSERT_JUMP (jump, laststart, b);
2762 b += 3;
2763 }
2764 }
2765 else
2766 {
2767 /* non-greedy a?? */
2768 INSERT_JUMP (jump, laststart, b + 3);
2769 b += 3;
2770 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2771 b += 3;
2772 }
2773 }
2774 }
2775 pending_exact = 0;
2776 break;
2777
2778
2779 case '.':
2780 laststart = b;
2781 BUF_PUSH (anychar);
2782 break;
2783
2784
2785 case '[':
2786 {
2787 re_char *p1;
2788
2789 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2790
2791 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2792
2793 /* Ensure that we have enough space to push a charset: the
2794 opcode, the length count, and the bitset; 34 bytes in all. */
2795 GET_BUFFER_SPACE (34);
2796
2797 laststart = b;
2798
2799 /* We test `*p == '^' twice, instead of using an if
2800 statement, so we only need one BUF_PUSH. */
2801 BUF_PUSH (*p == '^' ? charset_not : charset);
2802 if (*p == '^')
2803 p++;
2804
2805 /* Remember the first position in the bracket expression. */
2806 p1 = p;
2807
2808 /* Push the number of bytes in the bitmap. */
2809 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2810
2811 /* Clear the whole map. */
2812 memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH);
2813
2814 /* charset_not matches newline according to a syntax bit. */
2815 if ((re_opcode_t) b[-2] == charset_not
2816 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2817 SET_LIST_BIT ('\n');
2818
2819 /* Read in characters and ranges, setting map bits. */
2820 for (;;)
2821 {
2822 boolean escaped_char = false;
2823 const unsigned char *p2 = p;
2824 re_wchar_t ch;
2825
2826 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2827
2828 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2829 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2830 So the translation is done later in a loop. Example:
2831 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2832 PATFETCH (c);
2833
2834 /* \ might escape characters inside [...] and [^...]. */
2835 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2836 {
2837 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2838
2839 PATFETCH (c);
2840 escaped_char = true;
2841 }
2842 else
2843 {
2844 /* Could be the end of the bracket expression. If it's
2845 not (i.e., when the bracket expression is `[]' so
2846 far), the ']' character bit gets set way below. */
2847 if (c == ']' && p2 != p1)
2848 break;
2849 }
2850
2851 /* See if we're at the beginning of a possible character
2852 class. */
2853
2854 if (!escaped_char &&
2855 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2856 {
2857 /* Leave room for the null. */
2858 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2859 const unsigned char *class_beg;
2860
2861 PATFETCH (c);
2862 c1 = 0;
2863 class_beg = p;
2864
2865 /* If pattern is `[[:'. */
2866 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2867
2868 for (;;)
2869 {
2870 PATFETCH (c);
2871 if ((c == ':' && *p == ']') || p == pend)
2872 break;
2873 if (c1 < CHAR_CLASS_MAX_LENGTH)
2874 str[c1++] = c;
2875 else
2876 /* This is in any case an invalid class name. */
2877 str[0] = '\0';
2878 }
2879 str[c1] = '\0';
2880
2881 /* If isn't a word bracketed by `[:' and `:]':
2882 undo the ending character, the letters, and
2883 leave the leading `:' and `[' (but set bits for
2884 them). */
2885 if (c == ':' && *p == ']')
2886 {
2887 re_wctype_t cc = re_wctype (str);
2888
2889 if (cc == 0)
2890 FREE_STACK_RETURN (REG_ECTYPE);
2891
2892 /* Throw away the ] at the end of the character
2893 class. */
2894 PATFETCH (c);
2895
2896 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2897
2898 #ifndef emacs
2899 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
2900 if (re_iswctype (btowc (ch), cc))
2901 {
2902 c = TRANSLATE (ch);
2903 if (c < (1 << BYTEWIDTH))
2904 SET_LIST_BIT (c);
2905 }
2906 #else /* emacs */
2907 /* Most character classes in a multibyte match
2908 just set a flag. Exceptions are is_blank,
2909 is_digit, is_cntrl, and is_xdigit, since
2910 they can only match ASCII characters. We
2911 don't need to handle them for multibyte.
2912 They are distinguished by a negative wctype. */
2913
2914 /* Setup the gl_state object to its buffer-defined
2915 value. This hardcodes the buffer-global
2916 syntax-table for ASCII chars, while the other chars
2917 will obey syntax-table properties. It's not ideal,
2918 but it's the way it's been done until now. */
2919 SETUP_BUFFER_SYNTAX_TABLE ();
2920
2921 for (ch = 0; ch < 256; ++ch)
2922 {
2923 c = RE_CHAR_TO_MULTIBYTE (ch);
2924 if (! CHAR_BYTE8_P (c)
2925 && re_iswctype (c, cc))
2926 {
2927 SET_LIST_BIT (ch);
2928 c1 = TRANSLATE (c);
2929 if (c1 == c)
2930 continue;
2931 if (ASCII_CHAR_P (c1))
2932 SET_LIST_BIT (c1);
2933 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
2934 SET_LIST_BIT (c1);
2935 }
2936 }
2937 SET_RANGE_TABLE_WORK_AREA_BIT
2938 (range_table_work, re_wctype_to_bit (cc));
2939 #endif /* emacs */
2940 /* In most cases the matching rule for char classes
2941 only uses the syntax table for multibyte chars,
2942 so that the content of the syntax-table is not
2943 hardcoded in the range_table. SPACE and WORD are
2944 the two exceptions. */
2945 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
2946 bufp->used_syntax = 1;
2947
2948 /* Repeat the loop. */
2949 continue;
2950 }
2951 else
2952 {
2953 /* Go back to right after the "[:". */
2954 p = class_beg;
2955 SET_LIST_BIT ('[');
2956
2957 /* Because the `:' may start the range, we
2958 can't simply set bit and repeat the loop.
2959 Instead, just set it to C and handle below. */
2960 c = ':';
2961 }
2962 }
2963
2964 if (p < pend && p[0] == '-' && p[1] != ']')
2965 {
2966
2967 /* Discard the `-'. */
2968 PATFETCH (c1);
2969
2970 /* Fetch the character which ends the range. */
2971 PATFETCH (c1);
2972 #ifdef emacs
2973 if (CHAR_BYTE8_P (c1)
2974 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
2975 /* Treat the range from a multibyte character to
2976 raw-byte character as empty. */
2977 c = c1 + 1;
2978 #endif /* emacs */
2979 }
2980 else
2981 /* Range from C to C. */
2982 c1 = c;
2983
2984 if (c > c1)
2985 {
2986 if (syntax & RE_NO_EMPTY_RANGES)
2987 FREE_STACK_RETURN (REG_ERANGEX);
2988 /* Else, repeat the loop. */
2989 }
2990 else
2991 {
2992 #ifndef emacs
2993 /* Set the range into bitmap */
2994 for (; c <= c1; c++)
2995 {
2996 ch = TRANSLATE (c);
2997 if (ch < (1 << BYTEWIDTH))
2998 SET_LIST_BIT (ch);
2999 }
3000 #else /* emacs */
3001 if (c < 128)
3002 {
3003 ch = min (127, c1);
3004 SETUP_ASCII_RANGE (range_table_work, c, ch);
3005 c = ch + 1;
3006 if (CHAR_BYTE8_P (c1))
3007 c = BYTE8_TO_CHAR (128);
3008 }
3009 if (c <= c1)
3010 {
3011 if (CHAR_BYTE8_P (c))
3012 {
3013 c = CHAR_TO_BYTE8 (c);
3014 c1 = CHAR_TO_BYTE8 (c1);
3015 for (; c <= c1; c++)
3016 SET_LIST_BIT (c);
3017 }
3018 else if (multibyte)
3019 {
3020 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3021 }
3022 else
3023 {
3024 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3025 }
3026 }
3027 #endif /* emacs */
3028 }
3029 }
3030
3031 /* Discard any (non)matching list bytes that are all 0 at the
3032 end of the map. Decrease the map-length byte too. */
3033 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3034 b[-1]--;
3035 b += b[-1];
3036
3037 /* Build real range table from work area. */
3038 if (RANGE_TABLE_WORK_USED (range_table_work)
3039 || RANGE_TABLE_WORK_BITS (range_table_work))
3040 {
3041 int i;
3042 int used = RANGE_TABLE_WORK_USED (range_table_work);
3043
3044 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3045 bytes for flags, two for COUNT, and three bytes for
3046 each character. */
3047 GET_BUFFER_SPACE (4 + used * 3);
3048
3049 /* Indicate the existence of range table. */
3050 laststart[1] |= 0x80;
3051
3052 /* Store the character class flag bits into the range table.
3053 If not in emacs, these flag bits are always 0. */
3054 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3055 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3056
3057 STORE_NUMBER_AND_INCR (b, used / 2);
3058 for (i = 0; i < used; i++)
3059 STORE_CHARACTER_AND_INCR
3060 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3061 }
3062 }
3063 break;
3064
3065
3066 case '(':
3067 if (syntax & RE_NO_BK_PARENS)
3068 goto handle_open;
3069 else
3070 goto normal_char;
3071
3072
3073 case ')':
3074 if (syntax & RE_NO_BK_PARENS)
3075 goto handle_close;
3076 else
3077 goto normal_char;
3078
3079
3080 case '\n':
3081 if (syntax & RE_NEWLINE_ALT)
3082 goto handle_alt;
3083 else
3084 goto normal_char;
3085
3086
3087 case '|':
3088 if (syntax & RE_NO_BK_VBAR)
3089 goto handle_alt;
3090 else
3091 goto normal_char;
3092
3093
3094 case '{':
3095 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3096 goto handle_interval;
3097 else
3098 goto normal_char;
3099
3100
3101 case '\\':
3102 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3103
3104 /* Do not translate the character after the \, so that we can
3105 distinguish, e.g., \B from \b, even if we normally would
3106 translate, e.g., B to b. */
3107 PATFETCH (c);
3108
3109 switch (c)
3110 {
3111 case '(':
3112 if (syntax & RE_NO_BK_PARENS)
3113 goto normal_backslash;
3114
3115 handle_open:
3116 {
3117 int shy = 0;
3118 regnum_t regnum = 0;
3119 if (p+1 < pend)
3120 {
3121 /* Look for a special (?...) construct */
3122 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3123 {
3124 PATFETCH (c); /* Gobble up the '?'. */
3125 while (!shy)
3126 {
3127 PATFETCH (c);
3128 switch (c)
3129 {
3130 case ':': shy = 1; break;
3131 case '0':
3132 /* An explicitly specified regnum must start
3133 with non-0. */
3134 if (regnum == 0)
3135 FREE_STACK_RETURN (REG_BADPAT);
3136 case '1': case '2': case '3': case '4':
3137 case '5': case '6': case '7': case '8': case '9':
3138 regnum = 10*regnum + (c - '0'); break;
3139 default:
3140 /* Only (?:...) is supported right now. */
3141 FREE_STACK_RETURN (REG_BADPAT);
3142 }
3143 }
3144 }
3145 }
3146
3147 if (!shy)
3148 regnum = ++bufp->re_nsub;
3149 else if (regnum)
3150 { /* It's actually not shy, but explicitly numbered. */
3151 shy = 0;
3152 if (regnum > bufp->re_nsub)
3153 bufp->re_nsub = regnum;
3154 else if (regnum > bufp->re_nsub
3155 /* Ideally, we'd want to check that the specified
3156 group can't have matched (i.e. all subgroups
3157 using the same regnum are in other branches of
3158 OR patterns), but we don't currently keep track
3159 of enough info to do that easily. */
3160 || group_in_compile_stack (compile_stack, regnum))
3161 FREE_STACK_RETURN (REG_BADPAT);
3162 }
3163 else
3164 /* It's really shy. */
3165 regnum = - bufp->re_nsub;
3166
3167 if (COMPILE_STACK_FULL)
3168 {
3169 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3170 compile_stack_elt_t);
3171 if (compile_stack.stack == NULL) return REG_ESPACE;
3172
3173 compile_stack.size <<= 1;
3174 }
3175
3176 /* These are the values to restore when we hit end of this
3177 group. They are all relative offsets, so that if the
3178 whole pattern moves because of realloc, they will still
3179 be valid. */
3180 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3181 COMPILE_STACK_TOP.fixup_alt_jump
3182 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3183 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3184 COMPILE_STACK_TOP.regnum = regnum;
3185
3186 /* Do not push a start_memory for groups beyond the last one
3187 we can represent in the compiled pattern. */
3188 if (regnum <= MAX_REGNUM && regnum > 0)
3189 BUF_PUSH_2 (start_memory, regnum);
3190
3191 compile_stack.avail++;
3192
3193 fixup_alt_jump = 0;
3194 laststart = 0;
3195 begalt = b;
3196 /* If we've reached MAX_REGNUM groups, then this open
3197 won't actually generate any code, so we'll have to
3198 clear pending_exact explicitly. */
3199 pending_exact = 0;
3200 break;
3201 }
3202
3203 case ')':
3204 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3205
3206 if (COMPILE_STACK_EMPTY)
3207 {
3208 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3209 goto normal_backslash;
3210 else
3211 FREE_STACK_RETURN (REG_ERPAREN);
3212 }
3213
3214 handle_close:
3215 FIXUP_ALT_JUMP ();
3216
3217 /* See similar code for backslashed left paren above. */
3218 if (COMPILE_STACK_EMPTY)
3219 {
3220 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3221 goto normal_char;
3222 else
3223 FREE_STACK_RETURN (REG_ERPAREN);
3224 }
3225
3226 /* Since we just checked for an empty stack above, this
3227 ``can't happen''. */
3228 assert (compile_stack.avail != 0);
3229 {
3230 /* We don't just want to restore into `regnum', because
3231 later groups should continue to be numbered higher,
3232 as in `(ab)c(de)' -- the second group is #2. */
3233 regnum_t regnum;
3234
3235 compile_stack.avail--;
3236 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3237 fixup_alt_jump
3238 = COMPILE_STACK_TOP.fixup_alt_jump
3239 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3240 : 0;
3241 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3242 regnum = COMPILE_STACK_TOP.regnum;
3243 /* If we've reached MAX_REGNUM groups, then this open
3244 won't actually generate any code, so we'll have to
3245 clear pending_exact explicitly. */
3246 pending_exact = 0;
3247
3248 /* We're at the end of the group, so now we know how many
3249 groups were inside this one. */
3250 if (regnum <= MAX_REGNUM && regnum > 0)
3251 BUF_PUSH_2 (stop_memory, regnum);
3252 }
3253 break;
3254
3255
3256 case '|': /* `\|'. */
3257 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3258 goto normal_backslash;
3259 handle_alt:
3260 if (syntax & RE_LIMITED_OPS)
3261 goto normal_char;
3262
3263 /* Insert before the previous alternative a jump which
3264 jumps to this alternative if the former fails. */
3265 GET_BUFFER_SPACE (3);
3266 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3267 pending_exact = 0;
3268 b += 3;
3269
3270 /* The alternative before this one has a jump after it
3271 which gets executed if it gets matched. Adjust that
3272 jump so it will jump to this alternative's analogous
3273 jump (put in below, which in turn will jump to the next
3274 (if any) alternative's such jump, etc.). The last such
3275 jump jumps to the correct final destination. A picture:
3276 _____ _____
3277 | | | |
3278 | v | v
3279 a | b | c
3280
3281 If we are at `b', then fixup_alt_jump right now points to a
3282 three-byte space after `a'. We'll put in the jump, set
3283 fixup_alt_jump to right after `b', and leave behind three
3284 bytes which we'll fill in when we get to after `c'. */
3285
3286 FIXUP_ALT_JUMP ();
3287
3288 /* Mark and leave space for a jump after this alternative,
3289 to be filled in later either by next alternative or
3290 when know we're at the end of a series of alternatives. */
3291 fixup_alt_jump = b;
3292 GET_BUFFER_SPACE (3);
3293 b += 3;
3294
3295 laststart = 0;
3296 begalt = b;
3297 break;
3298
3299
3300 case '{':
3301 /* If \{ is a literal. */
3302 if (!(syntax & RE_INTERVALS)
3303 /* If we're at `\{' and it's not the open-interval
3304 operator. */
3305 || (syntax & RE_NO_BK_BRACES))
3306 goto normal_backslash;
3307
3308 handle_interval:
3309 {
3310 /* If got here, then the syntax allows intervals. */
3311
3312 /* At least (most) this many matches must be made. */
3313 int lower_bound = 0, upper_bound = -1;
3314
3315 beg_interval = p;
3316
3317 GET_INTERVAL_COUNT (lower_bound);
3318
3319 if (c == ',')
3320 GET_INTERVAL_COUNT (upper_bound);
3321 else
3322 /* Interval such as `{1}' => match exactly once. */
3323 upper_bound = lower_bound;
3324
3325 if (lower_bound < 0
3326 || (0 <= upper_bound && upper_bound < lower_bound))
3327 FREE_STACK_RETURN (REG_BADBR);
3328
3329 if (!(syntax & RE_NO_BK_BRACES))
3330 {
3331 if (c != '\\')
3332 FREE_STACK_RETURN (REG_BADBR);
3333 if (p == pend)
3334 FREE_STACK_RETURN (REG_EESCAPE);
3335 PATFETCH (c);
3336 }
3337
3338 if (c != '}')
3339 FREE_STACK_RETURN (REG_BADBR);
3340
3341 /* We just parsed a valid interval. */
3342
3343 /* If it's invalid to have no preceding re. */
3344 if (!laststart)
3345 {
3346 if (syntax & RE_CONTEXT_INVALID_OPS)
3347 FREE_STACK_RETURN (REG_BADRPT);
3348 else if (syntax & RE_CONTEXT_INDEP_OPS)
3349 laststart = b;
3350 else
3351 goto unfetch_interval;
3352 }
3353
3354 if (upper_bound == 0)
3355 /* If the upper bound is zero, just drop the sub pattern
3356 altogether. */
3357 b = laststart;
3358 else if (lower_bound == 1 && upper_bound == 1)
3359 /* Just match it once: nothing to do here. */
3360 ;
3361
3362 /* Otherwise, we have a nontrivial interval. When
3363 we're all done, the pattern will look like:
3364 set_number_at <jump count> <upper bound>
3365 set_number_at <succeed_n count> <lower bound>
3366 succeed_n <after jump addr> <succeed_n count>
3367 <body of loop>
3368 jump_n <succeed_n addr> <jump count>
3369 (The upper bound and `jump_n' are omitted if
3370 `upper_bound' is 1, though.) */
3371 else
3372 { /* If the upper bound is > 1, we need to insert
3373 more at the end of the loop. */
3374 unsigned int nbytes = (upper_bound < 0 ? 3
3375 : upper_bound > 1 ? 5 : 0);
3376 unsigned int startoffset = 0;
3377
3378 GET_BUFFER_SPACE (20); /* We might use less. */
3379
3380 if (lower_bound == 0)
3381 {
3382 /* A succeed_n that starts with 0 is really a
3383 a simple on_failure_jump_loop. */
3384 INSERT_JUMP (on_failure_jump_loop, laststart,
3385 b + 3 + nbytes);
3386 b += 3;
3387 }
3388 else
3389 {
3390 /* Initialize lower bound of the `succeed_n', even
3391 though it will be set during matching by its
3392 attendant `set_number_at' (inserted next),
3393 because `re_compile_fastmap' needs to know.
3394 Jump to the `jump_n' we might insert below. */
3395 INSERT_JUMP2 (succeed_n, laststart,
3396 b + 5 + nbytes,
3397 lower_bound);
3398 b += 5;
3399
3400 /* Code to initialize the lower bound. Insert
3401 before the `succeed_n'. The `5' is the last two
3402 bytes of this `set_number_at', plus 3 bytes of
3403 the following `succeed_n'. */
3404 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3405 b += 5;
3406 startoffset += 5;
3407 }
3408
3409 if (upper_bound < 0)
3410 {
3411 /* A negative upper bound stands for infinity,
3412 in which case it degenerates to a plain jump. */
3413 STORE_JUMP (jump, b, laststart + startoffset);
3414 b += 3;
3415 }
3416 else if (upper_bound > 1)
3417 { /* More than one repetition is allowed, so
3418 append a backward jump to the `succeed_n'
3419 that starts this interval.
3420
3421 When we've reached this during matching,
3422 we'll have matched the interval once, so
3423 jump back only `upper_bound - 1' times. */
3424 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3425 upper_bound - 1);
3426 b += 5;
3427
3428 /* The location we want to set is the second
3429 parameter of the `jump_n'; that is `b-2' as
3430 an absolute address. `laststart' will be
3431 the `set_number_at' we're about to insert;
3432 `laststart+3' the number to set, the source
3433 for the relative address. But we are
3434 inserting into the middle of the pattern --
3435 so everything is getting moved up by 5.
3436 Conclusion: (b - 2) - (laststart + 3) + 5,
3437 i.e., b - laststart.
3438
3439 We insert this at the beginning of the loop
3440 so that if we fail during matching, we'll
3441 reinitialize the bounds. */
3442 insert_op2 (set_number_at, laststart, b - laststart,
3443 upper_bound - 1, b);
3444 b += 5;
3445 }
3446 }
3447 pending_exact = 0;
3448 beg_interval = NULL;
3449 }
3450 break;
3451
3452 unfetch_interval:
3453 /* If an invalid interval, match the characters as literals. */
3454 assert (beg_interval);
3455 p = beg_interval;
3456 beg_interval = NULL;
3457
3458 /* normal_char and normal_backslash need `c'. */
3459 c = '{';
3460
3461 if (!(syntax & RE_NO_BK_BRACES))
3462 {
3463 assert (p > pattern && p[-1] == '\\');
3464 goto normal_backslash;
3465 }
3466 else
3467 goto normal_char;
3468
3469 #ifdef emacs
3470 /* There is no way to specify the before_dot and after_dot
3471 operators. rms says this is ok. --karl */
3472 case '=':
3473 laststart = b;
3474 BUF_PUSH (at_dot);
3475 break;
3476
3477 case 's':
3478 laststart = b;
3479 PATFETCH (c);
3480 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3481 break;
3482
3483 case 'S':
3484 laststart = b;
3485 PATFETCH (c);
3486 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3487 break;
3488
3489 case 'c':
3490 laststart = b;
3491 PATFETCH (c);
3492 BUF_PUSH_2 (categoryspec, c);
3493 break;
3494
3495 case 'C':
3496 laststart = b;
3497 PATFETCH (c);
3498 BUF_PUSH_2 (notcategoryspec, c);
3499 break;
3500 #endif /* emacs */
3501
3502
3503 case 'w':
3504 if (syntax & RE_NO_GNU_OPS)
3505 goto normal_char;
3506 laststart = b;
3507 BUF_PUSH_2 (syntaxspec, Sword);
3508 break;
3509
3510
3511 case 'W':
3512 if (syntax & RE_NO_GNU_OPS)
3513 goto normal_char;
3514 laststart = b;
3515 BUF_PUSH_2 (notsyntaxspec, Sword);
3516 break;
3517
3518
3519 case '<':
3520 if (syntax & RE_NO_GNU_OPS)
3521 goto normal_char;
3522 laststart = b;
3523 BUF_PUSH (wordbeg);
3524 break;
3525
3526 case '>':
3527 if (syntax & RE_NO_GNU_OPS)
3528 goto normal_char;
3529 laststart = b;
3530 BUF_PUSH (wordend);
3531 break;
3532
3533 case '_':
3534 if (syntax & RE_NO_GNU_OPS)
3535 goto normal_char;
3536 laststart = b;
3537 PATFETCH (c);
3538 if (c == '<')
3539 BUF_PUSH (symbeg);
3540 else if (c == '>')
3541 BUF_PUSH (symend);
3542 else
3543 FREE_STACK_RETURN (REG_BADPAT);
3544 break;
3545
3546 case 'b':
3547 if (syntax & RE_NO_GNU_OPS)
3548 goto normal_char;
3549 BUF_PUSH (wordbound);
3550 break;
3551
3552 case 'B':
3553 if (syntax & RE_NO_GNU_OPS)
3554 goto normal_char;
3555 BUF_PUSH (notwordbound);
3556 break;
3557
3558 case '`':
3559 if (syntax & RE_NO_GNU_OPS)
3560 goto normal_char;
3561 BUF_PUSH (begbuf);
3562 break;
3563
3564 case '\'':
3565 if (syntax & RE_NO_GNU_OPS)
3566 goto normal_char;
3567 BUF_PUSH (endbuf);
3568 break;
3569
3570 case '1': case '2': case '3': case '4': case '5':
3571 case '6': case '7': case '8': case '9':
3572 {
3573 regnum_t reg;
3574
3575 if (syntax & RE_NO_BK_REFS)
3576 goto normal_backslash;
3577
3578 reg = c - '0';
3579
3580 if (reg > bufp->re_nsub || reg < 1
3581 /* Can't back reference to a subexp before its end. */
3582 || group_in_compile_stack (compile_stack, reg))
3583 FREE_STACK_RETURN (REG_ESUBREG);
3584
3585 laststart = b;
3586 BUF_PUSH_2 (duplicate, reg);
3587 }
3588 break;
3589
3590
3591 case '+':
3592 case '?':
3593 if (syntax & RE_BK_PLUS_QM)
3594 goto handle_plus;
3595 else
3596 goto normal_backslash;
3597
3598 default:
3599 normal_backslash:
3600 /* You might think it would be useful for \ to mean
3601 not to translate; but if we don't translate it
3602 it will never match anything. */
3603 goto normal_char;
3604 }
3605 break;
3606
3607
3608 default:
3609 /* Expects the character in `c'. */
3610 normal_char:
3611 /* If no exactn currently being built. */
3612 if (!pending_exact
3613
3614 /* If last exactn not at current position. */
3615 || pending_exact + *pending_exact + 1 != b
3616
3617 /* We have only one byte following the exactn for the count. */
3618 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3619
3620 /* If followed by a repetition operator. */
3621 || (p != pend && (*p == '*' || *p == '^'))
3622 || ((syntax & RE_BK_PLUS_QM)
3623 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3624 : p != pend && (*p == '+' || *p == '?'))
3625 || ((syntax & RE_INTERVALS)
3626 && ((syntax & RE_NO_BK_BRACES)
3627 ? p != pend && *p == '{'
3628 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3629 {
3630 /* Start building a new exactn. */
3631
3632 laststart = b;
3633
3634 BUF_PUSH_2 (exactn, 0);
3635 pending_exact = b - 1;
3636 }
3637
3638 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3639 {
3640 int len;
3641
3642 if (multibyte)
3643 {
3644 c = TRANSLATE (c);
3645 len = CHAR_STRING (c, b);
3646 b += len;
3647 }
3648 else
3649 {
3650 c1 = RE_CHAR_TO_MULTIBYTE (c);
3651 if (! CHAR_BYTE8_P (c1))
3652 {
3653 re_wchar_t c2 = TRANSLATE (c1);
3654
3655 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3656 c = c1;
3657 }
3658 *b++ = c;
3659 len = 1;
3660 }
3661 (*pending_exact) += len;
3662 }
3663
3664 break;
3665 } /* switch (c) */
3666 } /* while p != pend */
3667
3668
3669 /* Through the pattern now. */
3670
3671 FIXUP_ALT_JUMP ();
3672
3673 if (!COMPILE_STACK_EMPTY)
3674 FREE_STACK_RETURN (REG_EPAREN);
3675
3676 /* If we don't want backtracking, force success
3677 the first time we reach the end of the compiled pattern. */
3678 if (syntax & RE_NO_POSIX_BACKTRACKING)
3679 BUF_PUSH (succeed);
3680
3681 /* We have succeeded; set the length of the buffer. */
3682 bufp->used = b - bufp->buffer;
3683
3684 #ifdef DEBUG
3685 if (debug > 0)
3686 {
3687 re_compile_fastmap (bufp);
3688 DEBUG_PRINT ("\nCompiled pattern: \n");
3689 print_compiled_pattern (bufp);
3690 }
3691 debug--;
3692 #endif /* DEBUG */
3693
3694 #ifndef MATCH_MAY_ALLOCATE
3695 /* Initialize the failure stack to the largest possible stack. This
3696 isn't necessary unless we're trying to avoid calling alloca in
3697 the search and match routines. */
3698 {
3699 int num_regs = bufp->re_nsub + 1;
3700
3701 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3702 {
3703 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3704 falk_stack.stack = realloc (fail_stack.stack,
3705 fail_stack.size * sizeof *falk_stack.stack);
3706 }
3707
3708 regex_grow_registers (num_regs);
3709 }
3710 #endif /* not MATCH_MAY_ALLOCATE */
3711
3712 FREE_STACK_RETURN (REG_NOERROR);
3713 } /* regex_compile */
3714 \f
3715 /* Subroutines for `regex_compile'. */
3716
3717 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3718
3719 static void
3720 store_op1 (re_opcode_t op, unsigned char *loc, int arg)
3721 {
3722 *loc = (unsigned char) op;
3723 STORE_NUMBER (loc + 1, arg);
3724 }
3725
3726
3727 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3728
3729 static void
3730 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2)
3731 {
3732 *loc = (unsigned char) op;
3733 STORE_NUMBER (loc + 1, arg1);
3734 STORE_NUMBER (loc + 3, arg2);
3735 }
3736
3737
3738 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3739 for OP followed by two-byte integer parameter ARG. */
3740
3741 static void
3742 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end)
3743 {
3744 register unsigned char *pfrom = end;
3745 register unsigned char *pto = end + 3;
3746
3747 while (pfrom != loc)
3748 *--pto = *--pfrom;
3749
3750 store_op1 (op, loc, arg);
3751 }
3752
3753
3754 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3755
3756 static void
3757 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end)
3758 {
3759 register unsigned char *pfrom = end;
3760 register unsigned char *pto = end + 5;
3761
3762 while (pfrom != loc)
3763 *--pto = *--pfrom;
3764
3765 store_op2 (op, loc, arg1, arg2);
3766 }
3767
3768
3769 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3770 after an alternative or a begin-subexpression. We assume there is at
3771 least one character before the ^. */
3772
3773 static boolean
3774 at_begline_loc_p (const_re_char *pattern, const_re_char *p, reg_syntax_t syntax)
3775 {
3776 re_char *prev = p - 2;
3777 boolean odd_backslashes;
3778
3779 /* After a subexpression? */
3780 if (*prev == '(')
3781 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3782
3783 /* After an alternative? */
3784 else if (*prev == '|')
3785 odd_backslashes = (syntax & RE_NO_BK_VBAR) == 0;
3786
3787 /* After a shy subexpression? */
3788 else if (*prev == ':' && (syntax & RE_SHY_GROUPS))
3789 {
3790 /* Skip over optional regnum. */
3791 while (prev - 1 >= pattern && prev[-1] >= '0' && prev[-1] <= '9')
3792 --prev;
3793
3794 if (!(prev - 2 >= pattern
3795 && prev[-1] == '?' && prev[-2] == '('))
3796 return false;
3797 prev -= 2;
3798 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3799 }
3800 else
3801 return false;
3802
3803 /* Count the number of preceding backslashes. */
3804 p = prev;
3805 while (prev - 1 >= pattern && prev[-1] == '\\')
3806 --prev;
3807 return (p - prev) & odd_backslashes;
3808 }
3809
3810
3811 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3812 at least one character after the $, i.e., `P < PEND'. */
3813
3814 static boolean
3815 at_endline_loc_p (const_re_char *p, const_re_char *pend, reg_syntax_t syntax)
3816 {
3817 re_char *next = p;
3818 boolean next_backslash = *next == '\\';
3819 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3820
3821 return
3822 /* Before a subexpression? */
3823 (syntax & RE_NO_BK_PARENS ? *next == ')'
3824 : next_backslash && next_next && *next_next == ')')
3825 /* Before an alternative? */
3826 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3827 : next_backslash && next_next && *next_next == '|');
3828 }
3829
3830
3831 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3832 false if it's not. */
3833
3834 static boolean
3835 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
3836 {
3837 ssize_t this_element;
3838
3839 for (this_element = compile_stack.avail - 1;
3840 this_element >= 0;
3841 this_element--)
3842 if (compile_stack.stack[this_element].regnum == regnum)
3843 return true;
3844
3845 return false;
3846 }
3847 \f
3848 /* analyze_first.
3849 If fastmap is non-NULL, go through the pattern and fill fastmap
3850 with all the possible leading chars. If fastmap is NULL, don't
3851 bother filling it up (obviously) and only return whether the
3852 pattern could potentially match the empty string.
3853
3854 Return 1 if p..pend might match the empty string.
3855 Return 0 if p..pend matches at least one char.
3856 Return -1 if fastmap was not updated accurately. */
3857
3858 static int
3859 analyze_first (const_re_char *p, const_re_char *pend, char *fastmap,
3860 const int multibyte)
3861 {
3862 int j, k;
3863 boolean not;
3864
3865 /* If all elements for base leading-codes in fastmap is set, this
3866 flag is set true. */
3867 boolean match_any_multibyte_characters = false;
3868
3869 assert (p);
3870
3871 /* The loop below works as follows:
3872 - It has a working-list kept in the PATTERN_STACK and which basically
3873 starts by only containing a pointer to the first operation.
3874 - If the opcode we're looking at is a match against some set of
3875 chars, then we add those chars to the fastmap and go on to the
3876 next work element from the worklist (done via `break').
3877 - If the opcode is a control operator on the other hand, we either
3878 ignore it (if it's meaningless at this point, such as `start_memory')
3879 or execute it (if it's a jump). If the jump has several destinations
3880 (i.e. `on_failure_jump'), then we push the other destination onto the
3881 worklist.
3882 We guarantee termination by ignoring backward jumps (more or less),
3883 so that `p' is monotonically increasing. More to the point, we
3884 never set `p' (or push) anything `<= p1'. */
3885
3886 while (p < pend)
3887 {
3888 /* `p1' is used as a marker of how far back a `on_failure_jump'
3889 can go without being ignored. It is normally equal to `p'
3890 (which prevents any backward `on_failure_jump') except right
3891 after a plain `jump', to allow patterns such as:
3892 0: jump 10
3893 3..9: <body>
3894 10: on_failure_jump 3
3895 as used for the *? operator. */
3896 re_char *p1 = p;
3897
3898 switch (*p++)
3899 {
3900 case succeed:
3901 return 1;
3902
3903 case duplicate:
3904 /* If the first character has to match a backreference, that means
3905 that the group was empty (since it already matched). Since this
3906 is the only case that interests us here, we can assume that the
3907 backreference must match the empty string. */
3908 p++;
3909 continue;
3910
3911
3912 /* Following are the cases which match a character. These end
3913 with `break'. */
3914
3915 case exactn:
3916 if (fastmap)
3917 {
3918 /* If multibyte is nonzero, the first byte of each
3919 character is an ASCII or a leading code. Otherwise,
3920 each byte is a character. Thus, this works in both
3921 cases. */
3922 fastmap[p[1]] = 1;
3923 if (! multibyte)
3924 {
3925 /* For the case of matching this unibyte regex
3926 against multibyte, we must set a leading code of
3927 the corresponding multibyte character. */
3928 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
3929
3930 fastmap[CHAR_LEADING_CODE (c)] = 1;
3931 }
3932 }
3933 break;
3934
3935
3936 case anychar:
3937 /* We could put all the chars except for \n (and maybe \0)
3938 but we don't bother since it is generally not worth it. */
3939 if (!fastmap) break;
3940 return -1;
3941
3942
3943 case charset_not:
3944 if (!fastmap) break;
3945 {
3946 /* Chars beyond end of bitmap are possible matches. */
3947 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3948 j < (1 << BYTEWIDTH); j++)
3949 fastmap[j] = 1;
3950 }
3951
3952 /* Fallthrough */
3953 case charset:
3954 if (!fastmap) break;
3955 not = (re_opcode_t) *(p - 1) == charset_not;
3956 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3957 j >= 0; j--)
3958 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3959 fastmap[j] = 1;
3960
3961 #ifdef emacs
3962 if (/* Any leading code can possibly start a character
3963 which doesn't match the specified set of characters. */
3964 not
3965 ||
3966 /* If we can match a character class, we can match any
3967 multibyte characters. */
3968 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3969 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
3970
3971 {
3972 if (match_any_multibyte_characters == false)
3973 {
3974 for (j = MIN_MULTIBYTE_LEADING_CODE;
3975 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
3976 fastmap[j] = 1;
3977 match_any_multibyte_characters = true;
3978 }
3979 }
3980
3981 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3982 && match_any_multibyte_characters == false)
3983 {
3984 /* Set fastmap[I] to 1 where I is a leading code of each
3985 multibyte character in the range table. */
3986 int c, count;
3987 unsigned char lc1, lc2;
3988
3989 /* Make P points the range table. `+ 2' is to skip flag
3990 bits for a character class. */
3991 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
3992
3993 /* Extract the number of ranges in range table into COUNT. */
3994 EXTRACT_NUMBER_AND_INCR (count, p);
3995 for (; count > 0; count--, p += 3)
3996 {
3997 /* Extract the start and end of each range. */
3998 EXTRACT_CHARACTER (c, p);
3999 lc1 = CHAR_LEADING_CODE (c);
4000 p += 3;
4001 EXTRACT_CHARACTER (c, p);
4002 lc2 = CHAR_LEADING_CODE (c);
4003 for (j = lc1; j <= lc2; j++)
4004 fastmap[j] = 1;
4005 }
4006 }
4007 #endif
4008 break;
4009
4010 case syntaxspec:
4011 case notsyntaxspec:
4012 if (!fastmap) break;
4013 #ifndef emacs
4014 not = (re_opcode_t)p[-1] == notsyntaxspec;
4015 k = *p++;
4016 for (j = 0; j < (1 << BYTEWIDTH); j++)
4017 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4018 fastmap[j] = 1;
4019 break;
4020 #else /* emacs */
4021 /* This match depends on text properties. These end with
4022 aborting optimizations. */
4023 return -1;
4024
4025 case categoryspec:
4026 case notcategoryspec:
4027 if (!fastmap) break;
4028 not = (re_opcode_t)p[-1] == notcategoryspec;
4029 k = *p++;
4030 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4031 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4032 fastmap[j] = 1;
4033
4034 /* Any leading code can possibly start a character which
4035 has or doesn't has the specified category. */
4036 if (match_any_multibyte_characters == false)
4037 {
4038 for (j = MIN_MULTIBYTE_LEADING_CODE;
4039 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4040 fastmap[j] = 1;
4041 match_any_multibyte_characters = true;
4042 }
4043 break;
4044
4045 /* All cases after this match the empty string. These end with
4046 `continue'. */
4047
4048 case before_dot:
4049 case at_dot:
4050 case after_dot:
4051 #endif /* !emacs */
4052 case no_op:
4053 case begline:
4054 case endline:
4055 case begbuf:
4056 case endbuf:
4057 case wordbound:
4058 case notwordbound:
4059 case wordbeg:
4060 case wordend:
4061 case symbeg:
4062 case symend:
4063 continue;
4064
4065
4066 case jump:
4067 EXTRACT_NUMBER_AND_INCR (j, p);
4068 if (j < 0)
4069 /* Backward jumps can only go back to code that we've already
4070 visited. `re_compile' should make sure this is true. */
4071 break;
4072 p += j;
4073 switch (*p)
4074 {
4075 case on_failure_jump:
4076 case on_failure_keep_string_jump:
4077 case on_failure_jump_loop:
4078 case on_failure_jump_nastyloop:
4079 case on_failure_jump_smart:
4080 p++;
4081 break;
4082 default:
4083 continue;
4084 };
4085 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4086 to jump back to "just after here". */
4087 /* Fallthrough */
4088
4089 case on_failure_jump:
4090 case on_failure_keep_string_jump:
4091 case on_failure_jump_nastyloop:
4092 case on_failure_jump_loop:
4093 case on_failure_jump_smart:
4094 EXTRACT_NUMBER_AND_INCR (j, p);
4095 if (p + j <= p1)
4096 ; /* Backward jump to be ignored. */
4097 else
4098 { /* We have to look down both arms.
4099 We first go down the "straight" path so as to minimize
4100 stack usage when going through alternatives. */
4101 int r = analyze_first (p, pend, fastmap, multibyte);
4102 if (r) return r;
4103 p += j;
4104 }
4105 continue;
4106
4107
4108 case jump_n:
4109 /* This code simply does not properly handle forward jump_n. */
4110 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4111 p += 4;
4112 /* jump_n can either jump or fall through. The (backward) jump
4113 case has already been handled, so we only need to look at the
4114 fallthrough case. */
4115 continue;
4116
4117 case succeed_n:
4118 /* If N == 0, it should be an on_failure_jump_loop instead. */
4119 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4120 p += 4;
4121 /* We only care about one iteration of the loop, so we don't
4122 need to consider the case where this behaves like an
4123 on_failure_jump. */
4124 continue;
4125
4126
4127 case set_number_at:
4128 p += 4;
4129 continue;
4130
4131
4132 case start_memory:
4133 case stop_memory:
4134 p += 1;
4135 continue;
4136
4137
4138 default:
4139 abort (); /* We have listed all the cases. */
4140 } /* switch *p++ */
4141
4142 /* Getting here means we have found the possible starting
4143 characters for one path of the pattern -- and that the empty
4144 string does not match. We need not follow this path further. */
4145 return 0;
4146 } /* while p */
4147
4148 /* We reached the end without matching anything. */
4149 return 1;
4150
4151 } /* analyze_first */
4152 \f
4153 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4154 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4155 characters can start a string that matches the pattern. This fastmap
4156 is used by re_search to skip quickly over impossible starting points.
4157
4158 Character codes above (1 << BYTEWIDTH) are not represented in the
4159 fastmap, but the leading codes are represented. Thus, the fastmap
4160 indicates which character sets could start a match.
4161
4162 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4163 area as BUFP->fastmap.
4164
4165 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4166 the pattern buffer.
4167
4168 Returns 0 if we succeed, -2 if an internal error. */
4169
4170 int
4171 re_compile_fastmap (struct re_pattern_buffer *bufp)
4172 {
4173 char *fastmap = bufp->fastmap;
4174 int analysis;
4175
4176 assert (fastmap && bufp->buffer);
4177
4178 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4179 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4180
4181 analysis = analyze_first (bufp->buffer, bufp->buffer + bufp->used,
4182 fastmap, RE_MULTIBYTE_P (bufp));
4183 bufp->can_be_null = (analysis != 0);
4184 return 0;
4185 } /* re_compile_fastmap */
4186 \f
4187 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4188 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4189 this memory for recording register information. STARTS and ENDS
4190 must be allocated using the malloc library routine, and must each
4191 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4192
4193 If NUM_REGS == 0, then subsequent matches should allocate their own
4194 register data.
4195
4196 Unless this function is called, the first search or match using
4197 PATTERN_BUFFER will allocate its own register data, without
4198 freeing the old data. */
4199
4200 void
4201 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4202 {
4203 if (num_regs)
4204 {
4205 bufp->regs_allocated = REGS_REALLOCATE;
4206 regs->num_regs = num_regs;
4207 regs->start = starts;
4208 regs->end = ends;
4209 }
4210 else
4211 {
4212 bufp->regs_allocated = REGS_UNALLOCATED;
4213 regs->num_regs = 0;
4214 regs->start = regs->end = 0;
4215 }
4216 }
4217 WEAK_ALIAS (__re_set_registers, re_set_registers)
4218 \f
4219 /* Searching routines. */
4220
4221 /* Like re_search_2, below, but only one string is specified, and
4222 doesn't let you say where to stop matching. */
4223
4224 regoff_t
4225 re_search (struct re_pattern_buffer *bufp, const char *string, size_t size,
4226 ssize_t startpos, ssize_t range, struct re_registers *regs)
4227 {
4228 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4229 regs, size);
4230 }
4231 WEAK_ALIAS (__re_search, re_search)
4232
4233 /* Head address of virtual concatenation of string. */
4234 #define HEAD_ADDR_VSTRING(P) \
4235 (((P) >= size1 ? string2 : string1))
4236
4237 /* Address of POS in the concatenation of virtual string. */
4238 #define POS_ADDR_VSTRING(POS) \
4239 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4240
4241 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4242 virtual concatenation of STRING1 and STRING2, starting first at index
4243 STARTPOS, then at STARTPOS + 1, and so on.
4244
4245 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4246
4247 RANGE is how far to scan while trying to match. RANGE = 0 means try
4248 only at STARTPOS; in general, the last start tried is STARTPOS +
4249 RANGE.
4250
4251 In REGS, return the indices of the virtual concatenation of STRING1
4252 and STRING2 that matched the entire BUFP->buffer and its contained
4253 subexpressions.
4254
4255 Do not consider matching one past the index STOP in the virtual
4256 concatenation of STRING1 and STRING2.
4257
4258 We return either the position in the strings at which the match was
4259 found, -1 if no match, or -2 if error (such as failure
4260 stack overflow). */
4261
4262 regoff_t
4263 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, size_t size1,
4264 const char *str2, size_t size2, ssize_t startpos, ssize_t range,
4265 struct re_registers *regs, ssize_t stop)
4266 {
4267 regoff_t val;
4268 re_char *string1 = (re_char*) str1;
4269 re_char *string2 = (re_char*) str2;
4270 register char *fastmap = bufp->fastmap;
4271 register RE_TRANSLATE_TYPE translate = bufp->translate;
4272 size_t total_size = size1 + size2;
4273 ssize_t endpos = startpos + range;
4274 boolean anchored_start;
4275 /* Nonzero if we are searching multibyte string. */
4276 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4277
4278 /* Check for out-of-range STARTPOS. */
4279 if (startpos < 0 || startpos > total_size)
4280 return -1;
4281
4282 /* Fix up RANGE if it might eventually take us outside
4283 the virtual concatenation of STRING1 and STRING2.
4284 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4285 if (endpos < 0)
4286 range = 0 - startpos;
4287 else if (endpos > total_size)
4288 range = total_size - startpos;
4289
4290 /* If the search isn't to be a backwards one, don't waste time in a
4291 search for a pattern anchored at beginning of buffer. */
4292 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4293 {
4294 if (startpos > 0)
4295 return -1;
4296 else
4297 range = 0;
4298 }
4299
4300 #ifdef emacs
4301 /* In a forward search for something that starts with \=.
4302 don't keep searching past point. */
4303 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4304 {
4305 range = PT_BYTE - BEGV_BYTE - startpos;
4306 if (range < 0)
4307 return -1;
4308 }
4309 #endif /* emacs */
4310
4311 /* Update the fastmap now if not correct already. */
4312 if (fastmap && !bufp->fastmap_accurate)
4313 re_compile_fastmap (bufp);
4314
4315 /* See whether the pattern is anchored. */
4316 anchored_start = (bufp->buffer[0] == begline);
4317
4318 #ifdef emacs
4319 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4320 {
4321 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4322
4323 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4324 }
4325 #endif
4326
4327 /* Loop through the string, looking for a place to start matching. */
4328 for (;;)
4329 {
4330 /* If the pattern is anchored,
4331 skip quickly past places we cannot match.
4332 We don't bother to treat startpos == 0 specially
4333 because that case doesn't repeat. */
4334 if (anchored_start && startpos > 0)
4335 {
4336 if (! ((startpos <= size1 ? string1[startpos - 1]
4337 : string2[startpos - size1 - 1])
4338 == '\n'))
4339 goto advance;
4340 }
4341
4342 /* If a fastmap is supplied, skip quickly over characters that
4343 cannot be the start of a match. If the pattern can match the
4344 null string, however, we don't need to skip characters; we want
4345 the first null string. */
4346 if (fastmap && startpos < total_size && !bufp->can_be_null)
4347 {
4348 register re_char *d;
4349 register re_wchar_t buf_ch;
4350
4351 d = POS_ADDR_VSTRING (startpos);
4352
4353 if (range > 0) /* Searching forwards. */
4354 {
4355 ssize_t irange = range, lim = 0;
4356
4357 if (startpos < size1 && startpos + range >= size1)
4358 lim = range - (size1 - startpos);
4359
4360 /* Written out as an if-else to avoid testing `translate'
4361 inside the loop. */
4362 if (RE_TRANSLATE_P (translate))
4363 {
4364 if (multibyte)
4365 while (range > lim)
4366 {
4367 int buf_charlen;
4368
4369 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4370 buf_ch = RE_TRANSLATE (translate, buf_ch);
4371 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4372 break;
4373
4374 range -= buf_charlen;
4375 d += buf_charlen;
4376 }
4377 else
4378 while (range > lim)
4379 {
4380 register re_wchar_t ch, translated;
4381
4382 buf_ch = *d;
4383 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4384 translated = RE_TRANSLATE (translate, ch);
4385 if (translated != ch
4386 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4387 buf_ch = ch;
4388 if (fastmap[buf_ch])
4389 break;
4390 d++;
4391 range--;
4392 }
4393 }
4394 else
4395 {
4396 if (multibyte)
4397 while (range > lim)
4398 {
4399 int buf_charlen;
4400
4401 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4402 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4403 break;
4404 range -= buf_charlen;
4405 d += buf_charlen;
4406 }
4407 else
4408 while (range > lim && !fastmap[*d])
4409 {
4410 d++;
4411 range--;
4412 }
4413 }
4414 startpos += irange - range;
4415 }
4416 else /* Searching backwards. */
4417 {
4418 if (multibyte)
4419 {
4420 buf_ch = STRING_CHAR (d);
4421 buf_ch = TRANSLATE (buf_ch);
4422 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4423 goto advance;
4424 }
4425 else
4426 {
4427 register re_wchar_t ch, translated;
4428
4429 buf_ch = *d;
4430 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4431 translated = TRANSLATE (ch);
4432 if (translated != ch
4433 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4434 buf_ch = ch;
4435 if (! fastmap[TRANSLATE (buf_ch)])
4436 goto advance;
4437 }
4438 }
4439 }
4440
4441 /* If can't match the null string, and that's all we have left, fail. */
4442 if (range >= 0 && startpos == total_size && fastmap
4443 && !bufp->can_be_null)
4444 return -1;
4445
4446 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4447 startpos, regs, stop);
4448
4449 if (val >= 0)
4450 return startpos;
4451
4452 if (val == -2)
4453 return -2;
4454
4455 advance:
4456 if (!range)
4457 break;
4458 else if (range > 0)
4459 {
4460 /* Update STARTPOS to the next character boundary. */
4461 if (multibyte)
4462 {
4463 re_char *p = POS_ADDR_VSTRING (startpos);
4464 int len = BYTES_BY_CHAR_HEAD (*p);
4465
4466 range -= len;
4467 if (range < 0)
4468 break;
4469 startpos += len;
4470 }
4471 else
4472 {
4473 range--;
4474 startpos++;
4475 }
4476 }
4477 else
4478 {
4479 range++;
4480 startpos--;
4481
4482 /* Update STARTPOS to the previous character boundary. */
4483 if (multibyte)
4484 {
4485 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4486 re_char *p0 = p;
4487 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4488
4489 /* Find the head of multibyte form. */
4490 PREV_CHAR_BOUNDARY (p, phead);
4491 range += p0 - 1 - p;
4492 if (range > 0)
4493 break;
4494
4495 startpos -= p0 - 1 - p;
4496 }
4497 }
4498 }
4499 return -1;
4500 } /* re_search_2 */
4501 WEAK_ALIAS (__re_search_2, re_search_2)
4502 \f
4503 /* Declarations and macros for re_match_2. */
4504
4505 static int bcmp_translate (re_char *s1, re_char *s2,
4506 register ssize_t len,
4507 RE_TRANSLATE_TYPE translate,
4508 const int multibyte);
4509
4510 /* This converts PTR, a pointer into one of the search strings `string1'
4511 and `string2' into an offset from the beginning of that string. */
4512 #define POINTER_TO_OFFSET(ptr) \
4513 (FIRST_STRING_P (ptr) \
4514 ? (ptr) - string1 \
4515 : (ptr) - string2 + (ptrdiff_t) size1)
4516
4517 /* Call before fetching a character with *d. This switches over to
4518 string2 if necessary.
4519 Check re_match_2_internal for a discussion of why end_match_2 might
4520 not be within string2 (but be equal to end_match_1 instead). */
4521 #define PREFETCH() \
4522 while (d == dend) \
4523 { \
4524 /* End of string2 => fail. */ \
4525 if (dend == end_match_2) \
4526 goto fail; \
4527 /* End of string1 => advance to string2. */ \
4528 d = string2; \
4529 dend = end_match_2; \
4530 }
4531
4532 /* Call before fetching a char with *d if you already checked other limits.
4533 This is meant for use in lookahead operations like wordend, etc..
4534 where we might need to look at parts of the string that might be
4535 outside of the LIMITs (i.e past `stop'). */
4536 #define PREFETCH_NOLIMIT() \
4537 if (d == end1) \
4538 { \
4539 d = string2; \
4540 dend = end_match_2; \
4541 } \
4542
4543 /* Test if at very beginning or at very end of the virtual concatenation
4544 of `string1' and `string2'. If only one string, it's `string2'. */
4545 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4546 #define AT_STRINGS_END(d) ((d) == end2)
4547
4548 /* Disabled due to a compiler bug -- see comment at case wordbound */
4549
4550 /* The comment at case wordbound is following one, but we don't use
4551 AT_WORD_BOUNDARY anymore to support multibyte form.
4552
4553 The DEC Alpha C compiler 3.x generates incorrect code for the
4554 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4555 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4556 macro and introducing temporary variables works around the bug. */
4557
4558 #if 0
4559 /* Test if D points to a character which is word-constituent. We have
4560 two special cases to check for: if past the end of string1, look at
4561 the first character in string2; and if before the beginning of
4562 string2, look at the last character in string1. */
4563 #define WORDCHAR_P(d) \
4564 (SYNTAX ((d) == end1 ? *string2 \
4565 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4566 == Sword)
4567
4568 /* Test if the character before D and the one at D differ with respect
4569 to being word-constituent. */
4570 #define AT_WORD_BOUNDARY(d) \
4571 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4572 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4573 #endif
4574
4575 /* Free everything we malloc. */
4576 #ifdef MATCH_MAY_ALLOCATE
4577 # define FREE_VAR(var) \
4578 do { \
4579 if (var) \
4580 { \
4581 REGEX_FREE (var); \
4582 var = NULL; \
4583 } \
4584 } while (0)
4585 # define FREE_VARIABLES() \
4586 do { \
4587 REGEX_FREE_STACK (fail_stack.stack); \
4588 FREE_VAR (regstart); \
4589 FREE_VAR (regend); \
4590 FREE_VAR (best_regstart); \
4591 FREE_VAR (best_regend); \
4592 REGEX_SAFE_FREE (); \
4593 } while (0)
4594 #else
4595 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4596 #endif /* not MATCH_MAY_ALLOCATE */
4597
4598 \f
4599 /* Optimization routines. */
4600
4601 /* If the operation is a match against one or more chars,
4602 return a pointer to the next operation, else return NULL. */
4603 static re_char *
4604 skip_one_char (const_re_char *p)
4605 {
4606 switch (*p++)
4607 {
4608 case anychar:
4609 break;
4610
4611 case exactn:
4612 p += *p + 1;
4613 break;
4614
4615 case charset_not:
4616 case charset:
4617 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4618 {
4619 int mcnt;
4620 p = CHARSET_RANGE_TABLE (p - 1);
4621 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4622 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4623 }
4624 else
4625 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4626 break;
4627
4628 case syntaxspec:
4629 case notsyntaxspec:
4630 #ifdef emacs
4631 case categoryspec:
4632 case notcategoryspec:
4633 #endif /* emacs */
4634 p++;
4635 break;
4636
4637 default:
4638 p = NULL;
4639 }
4640 return p;
4641 }
4642
4643
4644 /* Jump over non-matching operations. */
4645 static re_char *
4646 skip_noops (const_re_char *p, const_re_char *pend)
4647 {
4648 int mcnt;
4649 while (p < pend)
4650 {
4651 switch (*p)
4652 {
4653 case start_memory:
4654 case stop_memory:
4655 p += 2; break;
4656 case no_op:
4657 p += 1; break;
4658 case jump:
4659 p += 1;
4660 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4661 p += mcnt;
4662 break;
4663 default:
4664 return p;
4665 }
4666 }
4667 assert (p == pend);
4668 return p;
4669 }
4670
4671 /* Non-zero if "p1 matches something" implies "p2 fails". */
4672 static int
4673 mutually_exclusive_p (struct re_pattern_buffer *bufp, const_re_char *p1,
4674 const_re_char *p2)
4675 {
4676 re_opcode_t op2;
4677 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4678 unsigned char *pend = bufp->buffer + bufp->used;
4679
4680 assert (p1 >= bufp->buffer && p1 < pend
4681 && p2 >= bufp->buffer && p2 <= pend);
4682
4683 /* Skip over open/close-group commands.
4684 If what follows this loop is a ...+ construct,
4685 look at what begins its body, since we will have to
4686 match at least one of that. */
4687 p2 = skip_noops (p2, pend);
4688 /* The same skip can be done for p1, except that this function
4689 is only used in the case where p1 is a simple match operator. */
4690 /* p1 = skip_noops (p1, pend); */
4691
4692 assert (p1 >= bufp->buffer && p1 < pend
4693 && p2 >= bufp->buffer && p2 <= pend);
4694
4695 op2 = p2 == pend ? succeed : *p2;
4696
4697 switch (op2)
4698 {
4699 case succeed:
4700 case endbuf:
4701 /* If we're at the end of the pattern, we can change. */
4702 if (skip_one_char (p1))
4703 {
4704 DEBUG_PRINT (" End of pattern: fast loop.\n");
4705 return 1;
4706 }
4707 break;
4708
4709 case endline:
4710 case exactn:
4711 {
4712 register re_wchar_t c
4713 = (re_opcode_t) *p2 == endline ? '\n'
4714 : RE_STRING_CHAR (p2 + 2, multibyte);
4715
4716 if ((re_opcode_t) *p1 == exactn)
4717 {
4718 if (c != RE_STRING_CHAR (p1 + 2, multibyte))
4719 {
4720 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4721 return 1;
4722 }
4723 }
4724
4725 else if ((re_opcode_t) *p1 == charset
4726 || (re_opcode_t) *p1 == charset_not)
4727 {
4728 int not = (re_opcode_t) *p1 == charset_not;
4729
4730 /* Test if C is listed in charset (or charset_not)
4731 at `p1'. */
4732 if (! multibyte || IS_REAL_ASCII (c))
4733 {
4734 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4735 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4736 not = !not;
4737 }
4738 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4739 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4740
4741 /* `not' is equal to 1 if c would match, which means
4742 that we can't change to pop_failure_jump. */
4743 if (!not)
4744 {
4745 DEBUG_PRINT (" No match => fast loop.\n");
4746 return 1;
4747 }
4748 }
4749 else if ((re_opcode_t) *p1 == anychar
4750 && c == '\n')
4751 {
4752 DEBUG_PRINT (" . != \\n => fast loop.\n");
4753 return 1;
4754 }
4755 }
4756 break;
4757
4758 case charset:
4759 {
4760 if ((re_opcode_t) *p1 == exactn)
4761 /* Reuse the code above. */
4762 return mutually_exclusive_p (bufp, p2, p1);
4763
4764 /* It is hard to list up all the character in charset
4765 P2 if it includes multibyte character. Give up in
4766 such case. */
4767 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4768 {
4769 /* Now, we are sure that P2 has no range table.
4770 So, for the size of bitmap in P2, `p2[1]' is
4771 enough. But P1 may have range table, so the
4772 size of bitmap table of P1 is extracted by
4773 using macro `CHARSET_BITMAP_SIZE'.
4774
4775 In a multibyte case, we know that all the character
4776 listed in P2 is ASCII. In a unibyte case, P1 has only a
4777 bitmap table. So, in both cases, it is enough to test
4778 only the bitmap table of P1. */
4779
4780 if ((re_opcode_t) *p1 == charset)
4781 {
4782 int idx;
4783 /* We win if the charset inside the loop
4784 has no overlap with the one after the loop. */
4785 for (idx = 0;
4786 (idx < (int) p2[1]
4787 && idx < CHARSET_BITMAP_SIZE (p1));
4788 idx++)
4789 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4790 break;
4791
4792 if (idx == p2[1]
4793 || idx == CHARSET_BITMAP_SIZE (p1))
4794 {
4795 DEBUG_PRINT (" No match => fast loop.\n");
4796 return 1;
4797 }
4798 }
4799 else if ((re_opcode_t) *p1 == charset_not)
4800 {
4801 int idx;
4802 /* We win if the charset_not inside the loop lists
4803 every character listed in the charset after. */
4804 for (idx = 0; idx < (int) p2[1]; idx++)
4805 if (! (p2[2 + idx] == 0
4806 || (idx < CHARSET_BITMAP_SIZE (p1)
4807 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4808 break;
4809
4810 if (idx == p2[1])
4811 {
4812 DEBUG_PRINT (" No match => fast loop.\n");
4813 return 1;
4814 }
4815 }
4816 }
4817 }
4818 break;
4819
4820 case charset_not:
4821 switch (*p1)
4822 {
4823 case exactn:
4824 case charset:
4825 /* Reuse the code above. */
4826 return mutually_exclusive_p (bufp, p2, p1);
4827 case charset_not:
4828 /* When we have two charset_not, it's very unlikely that
4829 they don't overlap. The union of the two sets of excluded
4830 chars should cover all possible chars, which, as a matter of
4831 fact, is virtually impossible in multibyte buffers. */
4832 break;
4833 }
4834 break;
4835
4836 case wordend:
4837 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
4838 case symend:
4839 return ((re_opcode_t) *p1 == syntaxspec
4840 && (p1[1] == Ssymbol || p1[1] == Sword));
4841 case notsyntaxspec:
4842 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
4843
4844 case wordbeg:
4845 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
4846 case symbeg:
4847 return ((re_opcode_t) *p1 == notsyntaxspec
4848 && (p1[1] == Ssymbol || p1[1] == Sword));
4849 case syntaxspec:
4850 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
4851
4852 case wordbound:
4853 return (((re_opcode_t) *p1 == notsyntaxspec
4854 || (re_opcode_t) *p1 == syntaxspec)
4855 && p1[1] == Sword);
4856
4857 #ifdef emacs
4858 case categoryspec:
4859 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4860 case notcategoryspec:
4861 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4862 #endif /* emacs */
4863
4864 default:
4865 ;
4866 }
4867
4868 /* Safe default. */
4869 return 0;
4870 }
4871
4872 \f
4873 /* Matching routines. */
4874
4875 #ifndef emacs /* Emacs never uses this. */
4876 /* re_match is like re_match_2 except it takes only a single string. */
4877
4878 regoff_t
4879 re_match (struct re_pattern_buffer *bufp, const char *string,
4880 size_t size, ssize_t pos, struct re_registers *regs)
4881 {
4882 regoff_t result = re_match_2_internal (bufp, NULL, 0, (re_char*) string,
4883 size, pos, regs, size);
4884 return result;
4885 }
4886 WEAK_ALIAS (__re_match, re_match)
4887 #endif /* not emacs */
4888
4889 #ifdef emacs
4890 /* In Emacs, this is the string or buffer in which we
4891 are matching. It is used for looking up syntax properties. */
4892 Lisp_Object re_match_object;
4893 #endif
4894
4895 /* re_match_2 matches the compiled pattern in BUFP against the
4896 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4897 and SIZE2, respectively). We start matching at POS, and stop
4898 matching at STOP.
4899
4900 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4901 store offsets for the substring each group matched in REGS. See the
4902 documentation for exactly how many groups we fill.
4903
4904 We return -1 if no match, -2 if an internal error (such as the
4905 failure stack overflowing). Otherwise, we return the length of the
4906 matched substring. */
4907
4908 regoff_t
4909 re_match_2 (struct re_pattern_buffer *bufp, const char *string1,
4910 size_t size1, const char *string2, size_t size2, ssize_t pos,
4911 struct re_registers *regs, ssize_t stop)
4912 {
4913 regoff_t result;
4914
4915 #ifdef emacs
4916 ssize_t charpos;
4917 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4918 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4919 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4920 #endif
4921
4922 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4923 (re_char*) string2, size2,
4924 pos, regs, stop);
4925 return result;
4926 }
4927 WEAK_ALIAS (__re_match_2, re_match_2)
4928
4929
4930 /* This is a separate function so that we can force an alloca cleanup
4931 afterwards. */
4932 static regoff_t
4933 re_match_2_internal (struct re_pattern_buffer *bufp, const_re_char *string1,
4934 size_t size1, const_re_char *string2, size_t size2,
4935 ssize_t pos, struct re_registers *regs, ssize_t stop)
4936 {
4937 /* General temporaries. */
4938 int mcnt;
4939 size_t reg;
4940
4941 /* Just past the end of the corresponding string. */
4942 re_char *end1, *end2;
4943
4944 /* Pointers into string1 and string2, just past the last characters in
4945 each to consider matching. */
4946 re_char *end_match_1, *end_match_2;
4947
4948 /* Where we are in the data, and the end of the current string. */
4949 re_char *d, *dend;
4950
4951 /* Used sometimes to remember where we were before starting matching
4952 an operator so that we can go back in case of failure. This "atomic"
4953 behavior of matching opcodes is indispensable to the correctness
4954 of the on_failure_keep_string_jump optimization. */
4955 re_char *dfail;
4956
4957 /* Where we are in the pattern, and the end of the pattern. */
4958 re_char *p = bufp->buffer;
4959 re_char *pend = p + bufp->used;
4960
4961 /* We use this to map every character in the string. */
4962 RE_TRANSLATE_TYPE translate = bufp->translate;
4963
4964 /* Nonzero if BUFP is setup from a multibyte regex. */
4965 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4966
4967 /* Nonzero if STRING1/STRING2 are multibyte. */
4968 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4969
4970 /* Failure point stack. Each place that can handle a failure further
4971 down the line pushes a failure point on this stack. It consists of
4972 regstart, and regend for all registers corresponding to
4973 the subexpressions we're currently inside, plus the number of such
4974 registers, and, finally, two char *'s. The first char * is where
4975 to resume scanning the pattern; the second one is where to resume
4976 scanning the strings. */
4977 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4978 fail_stack_type fail_stack;
4979 #endif
4980 #ifdef DEBUG_COMPILES_ARGUMENTS
4981 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4982 #endif
4983
4984 #if defined REL_ALLOC && defined REGEX_MALLOC
4985 /* This holds the pointer to the failure stack, when
4986 it is allocated relocatably. */
4987 fail_stack_elt_t *failure_stack_ptr;
4988 #endif
4989
4990 /* We fill all the registers internally, independent of what we
4991 return, for use in backreferences. The number here includes
4992 an element for register zero. */
4993 size_t num_regs = bufp->re_nsub + 1;
4994
4995 /* Information on the contents of registers. These are pointers into
4996 the input strings; they record just what was matched (on this
4997 attempt) by a subexpression part of the pattern, that is, the
4998 regnum-th regstart pointer points to where in the pattern we began
4999 matching and the regnum-th regend points to right after where we
5000 stopped matching the regnum-th subexpression. (The zeroth register
5001 keeps track of what the whole pattern matches.) */
5002 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5003 re_char **regstart, **regend;
5004 #endif
5005
5006 /* The following record the register info as found in the above
5007 variables when we find a match better than any we've seen before.
5008 This happens as we backtrack through the failure points, which in
5009 turn happens only if we have not yet matched the entire string. */
5010 unsigned best_regs_set = false;
5011 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5012 re_char **best_regstart, **best_regend;
5013 #endif
5014
5015 /* Logically, this is `best_regend[0]'. But we don't want to have to
5016 allocate space for that if we're not allocating space for anything
5017 else (see below). Also, we never need info about register 0 for
5018 any of the other register vectors, and it seems rather a kludge to
5019 treat `best_regend' differently than the rest. So we keep track of
5020 the end of the best match so far in a separate variable. We
5021 initialize this to NULL so that when we backtrack the first time
5022 and need to test it, it's not garbage. */
5023 re_char *match_end = NULL;
5024
5025 #ifdef DEBUG_COMPILES_ARGUMENTS
5026 /* Counts the total number of registers pushed. */
5027 unsigned num_regs_pushed = 0;
5028 #endif
5029
5030 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5031
5032 REGEX_USE_SAFE_ALLOCA;
5033
5034 INIT_FAIL_STACK ();
5035
5036 #ifdef MATCH_MAY_ALLOCATE
5037 /* Do not bother to initialize all the register variables if there are
5038 no groups in the pattern, as it takes a fair amount of time. If
5039 there are groups, we include space for register 0 (the whole
5040 pattern), even though we never use it, since it simplifies the
5041 array indexing. We should fix this. */
5042 if (bufp->re_nsub)
5043 {
5044 regstart = REGEX_TALLOC (num_regs, re_char *);
5045 regend = REGEX_TALLOC (num_regs, re_char *);
5046 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5047 best_regend = REGEX_TALLOC (num_regs, re_char *);
5048
5049 if (!(regstart && regend && best_regstart && best_regend))
5050 {
5051 FREE_VARIABLES ();
5052 return -2;
5053 }
5054 }
5055 else
5056 {
5057 /* We must initialize all our variables to NULL, so that
5058 `FREE_VARIABLES' doesn't try to free them. */
5059 regstart = regend = best_regstart = best_regend = NULL;
5060 }
5061 #endif /* MATCH_MAY_ALLOCATE */
5062
5063 /* The starting position is bogus. */
5064 if (pos < 0 || pos > size1 + size2)
5065 {
5066 FREE_VARIABLES ();
5067 return -1;
5068 }
5069
5070 /* Initialize subexpression text positions to -1 to mark ones that no
5071 start_memory/stop_memory has been seen for. Also initialize the
5072 register information struct. */
5073 for (reg = 1; reg < num_regs; reg++)
5074 regstart[reg] = regend[reg] = NULL;
5075
5076 /* We move `string1' into `string2' if the latter's empty -- but not if
5077 `string1' is null. */
5078 if (size2 == 0 && string1 != NULL)
5079 {
5080 string2 = string1;
5081 size2 = size1;
5082 string1 = 0;
5083 size1 = 0;
5084 }
5085 end1 = string1 + size1;
5086 end2 = string2 + size2;
5087
5088 /* `p' scans through the pattern as `d' scans through the data.
5089 `dend' is the end of the input string that `d' points within. `d'
5090 is advanced into the following input string whenever necessary, but
5091 this happens before fetching; therefore, at the beginning of the
5092 loop, `d' can be pointing at the end of a string, but it cannot
5093 equal `string2'. */
5094 if (pos >= size1)
5095 {
5096 /* Only match within string2. */
5097 d = string2 + pos - size1;
5098 dend = end_match_2 = string2 + stop - size1;
5099 end_match_1 = end1; /* Just to give it a value. */
5100 }
5101 else
5102 {
5103 if (stop < size1)
5104 {
5105 /* Only match within string1. */
5106 end_match_1 = string1 + stop;
5107 /* BEWARE!
5108 When we reach end_match_1, PREFETCH normally switches to string2.
5109 But in the present case, this means that just doing a PREFETCH
5110 makes us jump from `stop' to `gap' within the string.
5111 What we really want here is for the search to stop as
5112 soon as we hit end_match_1. That's why we set end_match_2
5113 to end_match_1 (since PREFETCH fails as soon as we hit
5114 end_match_2). */
5115 end_match_2 = end_match_1;
5116 }
5117 else
5118 { /* It's important to use this code when stop == size so that
5119 moving `d' from end1 to string2 will not prevent the d == dend
5120 check from catching the end of string. */
5121 end_match_1 = end1;
5122 end_match_2 = string2 + stop - size1;
5123 }
5124 d = string1 + pos;
5125 dend = end_match_1;
5126 }
5127
5128 DEBUG_PRINT ("The compiled pattern is: ");
5129 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5130 DEBUG_PRINT ("The string to match is: \"");
5131 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5132 DEBUG_PRINT ("\"\n");
5133
5134 /* This loops over pattern commands. It exits by returning from the
5135 function if the match is complete, or it drops through if the match
5136 fails at this starting point in the input data. */
5137 for (;;)
5138 {
5139 DEBUG_PRINT ("\n%p: ", p);
5140
5141 if (p == pend)
5142 {
5143 ptrdiff_t dcnt;
5144
5145 /* End of pattern means we might have succeeded. */
5146 DEBUG_PRINT ("end of pattern ... ");
5147
5148 /* If we haven't matched the entire string, and we want the
5149 longest match, try backtracking. */
5150 if (d != end_match_2)
5151 {
5152 /* 1 if this match ends in the same string (string1 or string2)
5153 as the best previous match. */
5154 boolean same_str_p = (FIRST_STRING_P (match_end)
5155 == FIRST_STRING_P (d));
5156 /* 1 if this match is the best seen so far. */
5157 boolean best_match_p;
5158
5159 /* AIX compiler got confused when this was combined
5160 with the previous declaration. */
5161 if (same_str_p)
5162 best_match_p = d > match_end;
5163 else
5164 best_match_p = !FIRST_STRING_P (d);
5165
5166 DEBUG_PRINT ("backtracking.\n");
5167
5168 if (!FAIL_STACK_EMPTY ())
5169 { /* More failure points to try. */
5170
5171 /* If exceeds best match so far, save it. */
5172 if (!best_regs_set || best_match_p)
5173 {
5174 best_regs_set = true;
5175 match_end = d;
5176
5177 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5178
5179 for (reg = 1; reg < num_regs; reg++)
5180 {
5181 best_regstart[reg] = regstart[reg];
5182 best_regend[reg] = regend[reg];
5183 }
5184 }
5185 goto fail;
5186 }
5187
5188 /* If no failure points, don't restore garbage. And if
5189 last match is real best match, don't restore second
5190 best one. */
5191 else if (best_regs_set && !best_match_p)
5192 {
5193 restore_best_regs:
5194 /* Restore best match. It may happen that `dend ==
5195 end_match_1' while the restored d is in string2.
5196 For example, the pattern `x.*y.*z' against the
5197 strings `x-' and `y-z-', if the two strings are
5198 not consecutive in memory. */
5199 DEBUG_PRINT ("Restoring best registers.\n");
5200
5201 d = match_end;
5202 dend = ((d >= string1 && d <= end1)
5203 ? end_match_1 : end_match_2);
5204
5205 for (reg = 1; reg < num_regs; reg++)
5206 {
5207 regstart[reg] = best_regstart[reg];
5208 regend[reg] = best_regend[reg];
5209 }
5210 }
5211 } /* d != end_match_2 */
5212
5213 succeed_label:
5214 DEBUG_PRINT ("Accepting match.\n");
5215
5216 /* If caller wants register contents data back, do it. */
5217 if (regs && !bufp->no_sub)
5218 {
5219 /* Have the register data arrays been allocated? */
5220 if (bufp->regs_allocated == REGS_UNALLOCATED)
5221 { /* No. So allocate them with malloc. We need one
5222 extra element beyond `num_regs' for the `-1' marker
5223 GNU code uses. */
5224 regs->num_regs = max (RE_NREGS, num_regs + 1);
5225 regs->start = TALLOC (regs->num_regs, regoff_t);
5226 regs->end = TALLOC (regs->num_regs, regoff_t);
5227 if (regs->start == NULL || regs->end == NULL)
5228 {
5229 FREE_VARIABLES ();
5230 return -2;
5231 }
5232 bufp->regs_allocated = REGS_REALLOCATE;
5233 }
5234 else if (bufp->regs_allocated == REGS_REALLOCATE)
5235 { /* Yes. If we need more elements than were already
5236 allocated, reallocate them. If we need fewer, just
5237 leave it alone. */
5238 if (regs->num_regs < num_regs + 1)
5239 {
5240 regs->num_regs = num_regs + 1;
5241 RETALLOC (regs->start, regs->num_regs, regoff_t);
5242 RETALLOC (regs->end, regs->num_regs, regoff_t);
5243 if (regs->start == NULL || regs->end == NULL)
5244 {
5245 FREE_VARIABLES ();
5246 return -2;
5247 }
5248 }
5249 }
5250 else
5251 {
5252 /* These braces fend off a "empty body in an else-statement"
5253 warning under GCC when assert expands to nothing. */
5254 assert (bufp->regs_allocated == REGS_FIXED);
5255 }
5256
5257 /* Convert the pointer data in `regstart' and `regend' to
5258 indices. Register zero has to be set differently,
5259 since we haven't kept track of any info for it. */
5260 if (regs->num_regs > 0)
5261 {
5262 regs->start[0] = pos;
5263 regs->end[0] = POINTER_TO_OFFSET (d);
5264 }
5265
5266 /* Go through the first `min (num_regs, regs->num_regs)'
5267 registers, since that is all we initialized. */
5268 for (reg = 1; reg < min (num_regs, regs->num_regs); reg++)
5269 {
5270 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5271 regs->start[reg] = regs->end[reg] = -1;
5272 else
5273 {
5274 regs->start[reg] = POINTER_TO_OFFSET (regstart[reg]);
5275 regs->end[reg] = POINTER_TO_OFFSET (regend[reg]);
5276 }
5277 }
5278
5279 /* If the regs structure we return has more elements than
5280 were in the pattern, set the extra elements to -1. If
5281 we (re)allocated the registers, this is the case,
5282 because we always allocate enough to have at least one
5283 -1 at the end. */
5284 for (reg = num_regs; reg < regs->num_regs; reg++)
5285 regs->start[reg] = regs->end[reg] = -1;
5286 } /* regs && !bufp->no_sub */
5287
5288 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5289 nfailure_points_pushed, nfailure_points_popped,
5290 nfailure_points_pushed - nfailure_points_popped);
5291 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed);
5292
5293 dcnt = POINTER_TO_OFFSET (d) - pos;
5294
5295 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt);
5296
5297 FREE_VARIABLES ();
5298 return dcnt;
5299 }
5300
5301 /* Otherwise match next pattern command. */
5302 switch (*p++)
5303 {
5304 /* Ignore these. Used to ignore the n of succeed_n's which
5305 currently have n == 0. */
5306 case no_op:
5307 DEBUG_PRINT ("EXECUTING no_op.\n");
5308 break;
5309
5310 case succeed:
5311 DEBUG_PRINT ("EXECUTING succeed.\n");
5312 goto succeed_label;
5313
5314 /* Match the next n pattern characters exactly. The following
5315 byte in the pattern defines n, and the n bytes after that
5316 are the characters to match. */
5317 case exactn:
5318 mcnt = *p++;
5319 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt);
5320
5321 /* Remember the start point to rollback upon failure. */
5322 dfail = d;
5323
5324 #ifndef emacs
5325 /* This is written out as an if-else so we don't waste time
5326 testing `translate' inside the loop. */
5327 if (RE_TRANSLATE_P (translate))
5328 do
5329 {
5330 PREFETCH ();
5331 if (RE_TRANSLATE (translate, *d) != *p++)
5332 {
5333 d = dfail;
5334 goto fail;
5335 }
5336 d++;
5337 }
5338 while (--mcnt);
5339 else
5340 do
5341 {
5342 PREFETCH ();
5343 if (*d++ != *p++)
5344 {
5345 d = dfail;
5346 goto fail;
5347 }
5348 }
5349 while (--mcnt);
5350 #else /* emacs */
5351 /* The cost of testing `translate' is comparatively small. */
5352 if (target_multibyte)
5353 do
5354 {
5355 int pat_charlen, buf_charlen;
5356 int pat_ch, buf_ch;
5357
5358 PREFETCH ();
5359 if (multibyte)
5360 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5361 else
5362 {
5363 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5364 pat_charlen = 1;
5365 }
5366 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5367
5368 if (TRANSLATE (buf_ch) != pat_ch)
5369 {
5370 d = dfail;
5371 goto fail;
5372 }
5373
5374 p += pat_charlen;
5375 d += buf_charlen;
5376 mcnt -= pat_charlen;
5377 }
5378 while (mcnt > 0);
5379 else
5380 do
5381 {
5382 int pat_charlen;
5383 int pat_ch, buf_ch;
5384
5385 PREFETCH ();
5386 if (multibyte)
5387 {
5388 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5389 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5390 }
5391 else
5392 {
5393 pat_ch = *p;
5394 pat_charlen = 1;
5395 }
5396 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5397 if (! CHAR_BYTE8_P (buf_ch))
5398 {
5399 buf_ch = TRANSLATE (buf_ch);
5400 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5401 if (buf_ch < 0)
5402 buf_ch = *d;
5403 }
5404 else
5405 buf_ch = *d;
5406 if (buf_ch != pat_ch)
5407 {
5408 d = dfail;
5409 goto fail;
5410 }
5411 p += pat_charlen;
5412 d++;
5413 }
5414 while (--mcnt);
5415 #endif
5416 break;
5417
5418
5419 /* Match any character except possibly a newline or a null. */
5420 case anychar:
5421 {
5422 int buf_charlen;
5423 re_wchar_t buf_ch;
5424
5425 DEBUG_PRINT ("EXECUTING anychar.\n");
5426
5427 PREFETCH ();
5428 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5429 target_multibyte);
5430 buf_ch = TRANSLATE (buf_ch);
5431
5432 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5433 && buf_ch == '\n')
5434 || ((bufp->syntax & RE_DOT_NOT_NULL)
5435 && buf_ch == '\000'))
5436 goto fail;
5437
5438 DEBUG_PRINT (" Matched \"%d\".\n", *d);
5439 d += buf_charlen;
5440 }
5441 break;
5442
5443
5444 case charset:
5445 case charset_not:
5446 {
5447 register unsigned int c, corig;
5448 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5449 int len;
5450
5451 /* Start of actual range_table, or end of bitmap if there is no
5452 range table. */
5453 re_char *range_table IF_LINT (= NULL);
5454
5455 /* Nonzero if there is a range table. */
5456 int range_table_exists;
5457
5458 /* Number of ranges of range table. This is not included
5459 in the initial byte-length of the command. */
5460 int count = 0;
5461
5462 /* Whether matching against a unibyte character. */
5463 boolean unibyte_char = false;
5464
5465 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5466
5467 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5468
5469 if (range_table_exists)
5470 {
5471 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5472 EXTRACT_NUMBER_AND_INCR (count, range_table);
5473 }
5474
5475 PREFETCH ();
5476 corig = c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5477 if (target_multibyte)
5478 {
5479 int c1;
5480
5481 c = TRANSLATE (c);
5482 c1 = RE_CHAR_TO_UNIBYTE (c);
5483 if (c1 >= 0)
5484 {
5485 unibyte_char = true;
5486 c = c1;
5487 }
5488 }
5489 else
5490 {
5491 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5492
5493 if (! CHAR_BYTE8_P (c1))
5494 {
5495 c1 = TRANSLATE (c1);
5496 c1 = RE_CHAR_TO_UNIBYTE (c1);
5497 if (c1 >= 0)
5498 {
5499 unibyte_char = true;
5500 c = c1;
5501 }
5502 }
5503 else
5504 unibyte_char = true;
5505 }
5506
5507 if (unibyte_char && c < (1 << BYTEWIDTH))
5508 { /* Lookup bitmap. */
5509 /* Cast to `unsigned' instead of `unsigned char' in
5510 case the bit list is a full 32 bytes long. */
5511 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5512 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5513 not = !not;
5514 }
5515 #ifdef emacs
5516 else if (range_table_exists)
5517 {
5518 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5519
5520 if ( (class_bits & BIT_LOWER
5521 && (ISLOWER (c)
5522 || (corig != c
5523 && c == upcase (corig) && ISUPPER(c))))
5524 | (class_bits & BIT_MULTIBYTE)
5525 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5526 | (class_bits & BIT_SPACE && ISSPACE (c))
5527 | (class_bits & BIT_UPPER
5528 && (ISUPPER (c)
5529 || (corig != c
5530 && c == downcase (corig) && ISLOWER (c))))
5531 | (class_bits & BIT_WORD && ISWORD (c))
5532 | (class_bits & BIT_ALPHA && ISALPHA (c))
5533 | (class_bits & BIT_ALNUM && ISALNUM (c))
5534 | (class_bits & BIT_GRAPH && ISGRAPH (c))
5535 | (class_bits & BIT_PRINT && ISPRINT (c)))
5536 not = !not;
5537 else
5538 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5539 }
5540 #endif /* emacs */
5541
5542 if (range_table_exists)
5543 p = CHARSET_RANGE_TABLE_END (range_table, count);
5544 else
5545 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5546
5547 if (!not) goto fail;
5548
5549 d += len;
5550 }
5551 break;
5552
5553
5554 /* The beginning of a group is represented by start_memory.
5555 The argument is the register number. The text
5556 matched within the group is recorded (in the internal
5557 registers data structure) under the register number. */
5558 case start_memory:
5559 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p);
5560
5561 /* In case we need to undo this operation (via backtracking). */
5562 PUSH_FAILURE_REG (*p);
5563
5564 regstart[*p] = d;
5565 regend[*p] = NULL; /* probably unnecessary. -sm */
5566 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart[*p]));
5567
5568 /* Move past the register number and inner group count. */
5569 p += 1;
5570 break;
5571
5572
5573 /* The stop_memory opcode represents the end of a group. Its
5574 argument is the same as start_memory's: the register number. */
5575 case stop_memory:
5576 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p);
5577
5578 assert (!REG_UNSET (regstart[*p]));
5579 /* Strictly speaking, there should be code such as:
5580
5581 assert (REG_UNSET (regend[*p]));
5582 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5583
5584 But the only info to be pushed is regend[*p] and it is known to
5585 be UNSET, so there really isn't anything to push.
5586 Not pushing anything, on the other hand deprives us from the
5587 guarantee that regend[*p] is UNSET since undoing this operation
5588 will not reset its value properly. This is not important since
5589 the value will only be read on the next start_memory or at
5590 the very end and both events can only happen if this stop_memory
5591 is *not* undone. */
5592
5593 regend[*p] = d;
5594 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend[*p]));
5595
5596 /* Move past the register number and the inner group count. */
5597 p += 1;
5598 break;
5599
5600
5601 /* \<digit> has been turned into a `duplicate' command which is
5602 followed by the numeric value of <digit> as the register number. */
5603 case duplicate:
5604 {
5605 register re_char *d2, *dend2;
5606 int regno = *p++; /* Get which register to match against. */
5607 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno);
5608
5609 /* Can't back reference a group which we've never matched. */
5610 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5611 goto fail;
5612
5613 /* Where in input to try to start matching. */
5614 d2 = regstart[regno];
5615
5616 /* Remember the start point to rollback upon failure. */
5617 dfail = d;
5618
5619 /* Where to stop matching; if both the place to start and
5620 the place to stop matching are in the same string, then
5621 set to the place to stop, otherwise, for now have to use
5622 the end of the first string. */
5623
5624 dend2 = ((FIRST_STRING_P (regstart[regno])
5625 == FIRST_STRING_P (regend[regno]))
5626 ? regend[regno] : end_match_1);
5627 for (;;)
5628 {
5629 ptrdiff_t dcnt;
5630
5631 /* If necessary, advance to next segment in register
5632 contents. */
5633 while (d2 == dend2)
5634 {
5635 if (dend2 == end_match_2) break;
5636 if (dend2 == regend[regno]) break;
5637
5638 /* End of string1 => advance to string2. */
5639 d2 = string2;
5640 dend2 = regend[regno];
5641 }
5642 /* At end of register contents => success */
5643 if (d2 == dend2) break;
5644
5645 /* If necessary, advance to next segment in data. */
5646 PREFETCH ();
5647
5648 /* How many characters left in this segment to match. */
5649 dcnt = dend - d;
5650
5651 /* Want how many consecutive characters we can match in
5652 one shot, so, if necessary, adjust the count. */
5653 if (dcnt > dend2 - d2)
5654 dcnt = dend2 - d2;
5655
5656 /* Compare that many; failure if mismatch, else move
5657 past them. */
5658 if (RE_TRANSLATE_P (translate)
5659 ? bcmp_translate (d, d2, dcnt, translate, target_multibyte)
5660 : memcmp (d, d2, dcnt))
5661 {
5662 d = dfail;
5663 goto fail;
5664 }
5665 d += dcnt, d2 += dcnt;
5666 }
5667 }
5668 break;
5669
5670
5671 /* begline matches the empty string at the beginning of the string
5672 (unless `not_bol' is set in `bufp'), and after newlines. */
5673 case begline:
5674 DEBUG_PRINT ("EXECUTING begline.\n");
5675
5676 if (AT_STRINGS_BEG (d))
5677 {
5678 if (!bufp->not_bol) break;
5679 }
5680 else
5681 {
5682 unsigned c;
5683 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5684 if (c == '\n')
5685 break;
5686 }
5687 /* In all other cases, we fail. */
5688 goto fail;
5689
5690
5691 /* endline is the dual of begline. */
5692 case endline:
5693 DEBUG_PRINT ("EXECUTING endline.\n");
5694
5695 if (AT_STRINGS_END (d))
5696 {
5697 if (!bufp->not_eol) break;
5698 }
5699 else
5700 {
5701 PREFETCH_NOLIMIT ();
5702 if (*d == '\n')
5703 break;
5704 }
5705 goto fail;
5706
5707
5708 /* Match at the very beginning of the data. */
5709 case begbuf:
5710 DEBUG_PRINT ("EXECUTING begbuf.\n");
5711 if (AT_STRINGS_BEG (d))
5712 break;
5713 goto fail;
5714
5715
5716 /* Match at the very end of the data. */
5717 case endbuf:
5718 DEBUG_PRINT ("EXECUTING endbuf.\n");
5719 if (AT_STRINGS_END (d))
5720 break;
5721 goto fail;
5722
5723
5724 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5725 pushes NULL as the value for the string on the stack. Then
5726 `POP_FAILURE_POINT' will keep the current value for the
5727 string, instead of restoring it. To see why, consider
5728 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5729 then the . fails against the \n. But the next thing we want
5730 to do is match the \n against the \n; if we restored the
5731 string value, we would be back at the foo.
5732
5733 Because this is used only in specific cases, we don't need to
5734 check all the things that `on_failure_jump' does, to make
5735 sure the right things get saved on the stack. Hence we don't
5736 share its code. The only reason to push anything on the
5737 stack at all is that otherwise we would have to change
5738 `anychar's code to do something besides goto fail in this
5739 case; that seems worse than this. */
5740 case on_failure_keep_string_jump:
5741 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5742 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5743 mcnt, p + mcnt);
5744
5745 PUSH_FAILURE_POINT (p - 3, NULL);
5746 break;
5747
5748 /* A nasty loop is introduced by the non-greedy *? and +?.
5749 With such loops, the stack only ever contains one failure point
5750 at a time, so that a plain on_failure_jump_loop kind of
5751 cycle detection cannot work. Worse yet, such a detection
5752 can not only fail to detect a cycle, but it can also wrongly
5753 detect a cycle (between different instantiations of the same
5754 loop).
5755 So the method used for those nasty loops is a little different:
5756 We use a special cycle-detection-stack-frame which is pushed
5757 when the on_failure_jump_nastyloop failure-point is *popped*.
5758 This special frame thus marks the beginning of one iteration
5759 through the loop and we can hence easily check right here
5760 whether something matched between the beginning and the end of
5761 the loop. */
5762 case on_failure_jump_nastyloop:
5763 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5764 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5765 mcnt, p + mcnt);
5766
5767 assert ((re_opcode_t)p[-4] == no_op);
5768 {
5769 int cycle = 0;
5770 CHECK_INFINITE_LOOP (p - 4, d);
5771 if (!cycle)
5772 /* If there's a cycle, just continue without pushing
5773 this failure point. The failure point is the "try again"
5774 option, which shouldn't be tried.
5775 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5776 PUSH_FAILURE_POINT (p - 3, d);
5777 }
5778 break;
5779
5780 /* Simple loop detecting on_failure_jump: just check on the
5781 failure stack if the same spot was already hit earlier. */
5782 case on_failure_jump_loop:
5783 on_failure:
5784 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5785 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5786 mcnt, p + mcnt);
5787 {
5788 int cycle = 0;
5789 CHECK_INFINITE_LOOP (p - 3, d);
5790 if (cycle)
5791 /* If there's a cycle, get out of the loop, as if the matching
5792 had failed. We used to just `goto fail' here, but that was
5793 aborting the search a bit too early: we want to keep the
5794 empty-loop-match and keep matching after the loop.
5795 We want (x?)*y\1z to match both xxyz and xxyxz. */
5796 p += mcnt;
5797 else
5798 PUSH_FAILURE_POINT (p - 3, d);
5799 }
5800 break;
5801
5802
5803 /* Uses of on_failure_jump:
5804
5805 Each alternative starts with an on_failure_jump that points
5806 to the beginning of the next alternative. Each alternative
5807 except the last ends with a jump that in effect jumps past
5808 the rest of the alternatives. (They really jump to the
5809 ending jump of the following alternative, because tensioning
5810 these jumps is a hassle.)
5811
5812 Repeats start with an on_failure_jump that points past both
5813 the repetition text and either the following jump or
5814 pop_failure_jump back to this on_failure_jump. */
5815 case on_failure_jump:
5816 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5817 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5818 mcnt, p + mcnt);
5819
5820 PUSH_FAILURE_POINT (p -3, d);
5821 break;
5822
5823 /* This operation is used for greedy *.
5824 Compare the beginning of the repeat with what in the
5825 pattern follows its end. If we can establish that there
5826 is nothing that they would both match, i.e., that we
5827 would have to backtrack because of (as in, e.g., `a*a')
5828 then we can use a non-backtracking loop based on
5829 on_failure_keep_string_jump instead of on_failure_jump. */
5830 case on_failure_jump_smart:
5831 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5832 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5833 mcnt, p + mcnt);
5834 {
5835 re_char *p1 = p; /* Next operation. */
5836 /* Here, we discard `const', making re_match non-reentrant. */
5837 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5838 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5839
5840 p -= 3; /* Reset so that we will re-execute the
5841 instruction once it's been changed. */
5842
5843 EXTRACT_NUMBER (mcnt, p2 - 2);
5844
5845 /* Ensure this is a indeed the trivial kind of loop
5846 we are expecting. */
5847 assert (skip_one_char (p1) == p2 - 3);
5848 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5849 DEBUG_STATEMENT (debug += 2);
5850 if (mutually_exclusive_p (bufp, p1, p2))
5851 {
5852 /* Use a fast `on_failure_keep_string_jump' loop. */
5853 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5854 *p3 = (unsigned char) on_failure_keep_string_jump;
5855 STORE_NUMBER (p2 - 2, mcnt + 3);
5856 }
5857 else
5858 {
5859 /* Default to a safe `on_failure_jump' loop. */
5860 DEBUG_PRINT (" smart default => slow loop.\n");
5861 *p3 = (unsigned char) on_failure_jump;
5862 }
5863 DEBUG_STATEMENT (debug -= 2);
5864 }
5865 break;
5866
5867 /* Unconditionally jump (without popping any failure points). */
5868 case jump:
5869 unconditional_jump:
5870 IMMEDIATE_QUIT_CHECK;
5871 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5872 DEBUG_PRINT ("EXECUTING jump %d ", mcnt);
5873 p += mcnt; /* Do the jump. */
5874 DEBUG_PRINT ("(to %p).\n", p);
5875 break;
5876
5877
5878 /* Have to succeed matching what follows at least n times.
5879 After that, handle like `on_failure_jump'. */
5880 case succeed_n:
5881 /* Signedness doesn't matter since we only compare MCNT to 0. */
5882 EXTRACT_NUMBER (mcnt, p + 2);
5883 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt);
5884
5885 /* Originally, mcnt is how many times we HAVE to succeed. */
5886 if (mcnt != 0)
5887 {
5888 /* Here, we discard `const', making re_match non-reentrant. */
5889 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5890 mcnt--;
5891 p += 4;
5892 PUSH_NUMBER (p2, mcnt);
5893 }
5894 else
5895 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5896 goto on_failure;
5897 break;
5898
5899 case jump_n:
5900 /* Signedness doesn't matter since we only compare MCNT to 0. */
5901 EXTRACT_NUMBER (mcnt, p + 2);
5902 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt);
5903
5904 /* Originally, this is how many times we CAN jump. */
5905 if (mcnt != 0)
5906 {
5907 /* Here, we discard `const', making re_match non-reentrant. */
5908 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5909 mcnt--;
5910 PUSH_NUMBER (p2, mcnt);
5911 goto unconditional_jump;
5912 }
5913 /* If don't have to jump any more, skip over the rest of command. */
5914 else
5915 p += 4;
5916 break;
5917
5918 case set_number_at:
5919 {
5920 unsigned char *p2; /* Location of the counter. */
5921 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5922
5923 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5924 /* Here, we discard `const', making re_match non-reentrant. */
5925 p2 = (unsigned char*) p + mcnt;
5926 /* Signedness doesn't matter since we only copy MCNT's bits. */
5927 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5928 DEBUG_PRINT (" Setting %p to %d.\n", p2, mcnt);
5929 PUSH_NUMBER (p2, mcnt);
5930 break;
5931 }
5932
5933 case wordbound:
5934 case notwordbound:
5935 {
5936 boolean not = (re_opcode_t) *(p - 1) == notwordbound;
5937 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5938
5939 /* We SUCCEED (or FAIL) in one of the following cases: */
5940
5941 /* Case 1: D is at the beginning or the end of string. */
5942 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5943 not = !not;
5944 else
5945 {
5946 /* C1 is the character before D, S1 is the syntax of C1, C2
5947 is the character at D, and S2 is the syntax of C2. */
5948 re_wchar_t c1, c2;
5949 int s1, s2;
5950 int dummy;
5951 #ifdef emacs
5952 ssize_t offset = PTR_TO_OFFSET (d - 1);
5953 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5954 UPDATE_SYNTAX_TABLE_FAST (charpos);
5955 #endif
5956 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5957 s1 = SYNTAX (c1);
5958 #ifdef emacs
5959 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos + 1);
5960 #endif
5961 PREFETCH_NOLIMIT ();
5962 GET_CHAR_AFTER (c2, d, dummy);
5963 s2 = SYNTAX (c2);
5964
5965 if (/* Case 2: Only one of S1 and S2 is Sword. */
5966 ((s1 == Sword) != (s2 == Sword))
5967 /* Case 3: Both of S1 and S2 are Sword, and macro
5968 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5969 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5970 not = !not;
5971 }
5972 if (not)
5973 break;
5974 else
5975 goto fail;
5976 }
5977
5978 case wordbeg:
5979 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5980
5981 /* We FAIL in one of the following cases: */
5982
5983 /* Case 1: D is at the end of string. */
5984 if (AT_STRINGS_END (d))
5985 goto fail;
5986 else
5987 {
5988 /* C1 is the character before D, S1 is the syntax of C1, C2
5989 is the character at D, and S2 is the syntax of C2. */
5990 re_wchar_t c1, c2;
5991 int s1, s2;
5992 int dummy;
5993 #ifdef emacs
5994 ssize_t offset = PTR_TO_OFFSET (d);
5995 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5996 UPDATE_SYNTAX_TABLE_FAST (charpos);
5997 #endif
5998 PREFETCH ();
5999 GET_CHAR_AFTER (c2, d, dummy);
6000 s2 = SYNTAX (c2);
6001
6002 /* Case 2: S2 is not Sword. */
6003 if (s2 != Sword)
6004 goto fail;
6005
6006 /* Case 3: D is not at the beginning of string ... */
6007 if (!AT_STRINGS_BEG (d))
6008 {
6009 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6010 #ifdef emacs
6011 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6012 #endif
6013 s1 = SYNTAX (c1);
6014
6015 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6016 returns 0. */
6017 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6018 goto fail;
6019 }
6020 }
6021 break;
6022
6023 case wordend:
6024 DEBUG_PRINT ("EXECUTING wordend.\n");
6025
6026 /* We FAIL in one of the following cases: */
6027
6028 /* Case 1: D is at the beginning of string. */
6029 if (AT_STRINGS_BEG (d))
6030 goto fail;
6031 else
6032 {
6033 /* C1 is the character before D, S1 is the syntax of C1, C2
6034 is the character at D, and S2 is the syntax of C2. */
6035 re_wchar_t c1, c2;
6036 int s1, s2;
6037 int dummy;
6038 #ifdef emacs
6039 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6040 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6041 UPDATE_SYNTAX_TABLE_FAST (charpos);
6042 #endif
6043 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6044 s1 = SYNTAX (c1);
6045
6046 /* Case 2: S1 is not Sword. */
6047 if (s1 != Sword)
6048 goto fail;
6049
6050 /* Case 3: D is not at the end of string ... */
6051 if (!AT_STRINGS_END (d))
6052 {
6053 PREFETCH_NOLIMIT ();
6054 GET_CHAR_AFTER (c2, d, dummy);
6055 #ifdef emacs
6056 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos);
6057 #endif
6058 s2 = SYNTAX (c2);
6059
6060 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6061 returns 0. */
6062 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6063 goto fail;
6064 }
6065 }
6066 break;
6067
6068 case symbeg:
6069 DEBUG_PRINT ("EXECUTING symbeg.\n");
6070
6071 /* We FAIL in one of the following cases: */
6072
6073 /* Case 1: D is at the end of string. */
6074 if (AT_STRINGS_END (d))
6075 goto fail;
6076 else
6077 {
6078 /* C1 is the character before D, S1 is the syntax of C1, C2
6079 is the character at D, and S2 is the syntax of C2. */
6080 re_wchar_t c1, c2;
6081 int s1, s2;
6082 #ifdef emacs
6083 ssize_t offset = PTR_TO_OFFSET (d);
6084 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6085 UPDATE_SYNTAX_TABLE_FAST (charpos);
6086 #endif
6087 PREFETCH ();
6088 c2 = RE_STRING_CHAR (d, target_multibyte);
6089 s2 = SYNTAX (c2);
6090
6091 /* Case 2: S2 is neither Sword nor Ssymbol. */
6092 if (s2 != Sword && s2 != Ssymbol)
6093 goto fail;
6094
6095 /* Case 3: D is not at the beginning of string ... */
6096 if (!AT_STRINGS_BEG (d))
6097 {
6098 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6099 #ifdef emacs
6100 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6101 #endif
6102 s1 = SYNTAX (c1);
6103
6104 /* ... and S1 is Sword or Ssymbol. */
6105 if (s1 == Sword || s1 == Ssymbol)
6106 goto fail;
6107 }
6108 }
6109 break;
6110
6111 case symend:
6112 DEBUG_PRINT ("EXECUTING symend.\n");
6113
6114 /* We FAIL in one of the following cases: */
6115
6116 /* Case 1: D is at the beginning of string. */
6117 if (AT_STRINGS_BEG (d))
6118 goto fail;
6119 else
6120 {
6121 /* C1 is the character before D, S1 is the syntax of C1, C2
6122 is the character at D, and S2 is the syntax of C2. */
6123 re_wchar_t c1, c2;
6124 int s1, s2;
6125 #ifdef emacs
6126 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6127 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6128 UPDATE_SYNTAX_TABLE_FAST (charpos);
6129 #endif
6130 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6131 s1 = SYNTAX (c1);
6132
6133 /* Case 2: S1 is neither Ssymbol nor Sword. */
6134 if (s1 != Sword && s1 != Ssymbol)
6135 goto fail;
6136
6137 /* Case 3: D is not at the end of string ... */
6138 if (!AT_STRINGS_END (d))
6139 {
6140 PREFETCH_NOLIMIT ();
6141 c2 = RE_STRING_CHAR (d, target_multibyte);
6142 #ifdef emacs
6143 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos + 1);
6144 #endif
6145 s2 = SYNTAX (c2);
6146
6147 /* ... and S2 is Sword or Ssymbol. */
6148 if (s2 == Sword || s2 == Ssymbol)
6149 goto fail;
6150 }
6151 }
6152 break;
6153
6154 case syntaxspec:
6155 case notsyntaxspec:
6156 {
6157 boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6158 mcnt = *p++;
6159 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6160 mcnt);
6161 PREFETCH ();
6162 #ifdef emacs
6163 {
6164 ssize_t offset = PTR_TO_OFFSET (d);
6165 ssize_t pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6166 UPDATE_SYNTAX_TABLE_FAST (pos1);
6167 }
6168 #endif
6169 {
6170 int len;
6171 re_wchar_t c;
6172
6173 GET_CHAR_AFTER (c, d, len);
6174 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6175 goto fail;
6176 d += len;
6177 }
6178 }
6179 break;
6180
6181 #ifdef emacs
6182 case before_dot:
6183 DEBUG_PRINT ("EXECUTING before_dot.\n");
6184 if (PTR_BYTE_POS (d) >= PT_BYTE)
6185 goto fail;
6186 break;
6187
6188 case at_dot:
6189 DEBUG_PRINT ("EXECUTING at_dot.\n");
6190 if (PTR_BYTE_POS (d) != PT_BYTE)
6191 goto fail;
6192 break;
6193
6194 case after_dot:
6195 DEBUG_PRINT ("EXECUTING after_dot.\n");
6196 if (PTR_BYTE_POS (d) <= PT_BYTE)
6197 goto fail;
6198 break;
6199
6200 case categoryspec:
6201 case notcategoryspec:
6202 {
6203 boolean not = (re_opcode_t) *(p - 1) == notcategoryspec;
6204 mcnt = *p++;
6205 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6206 not ? "not" : "", mcnt);
6207 PREFETCH ();
6208
6209 {
6210 int len;
6211 re_wchar_t c;
6212 GET_CHAR_AFTER (c, d, len);
6213 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6214 goto fail;
6215 d += len;
6216 }
6217 }
6218 break;
6219
6220 #endif /* emacs */
6221
6222 default:
6223 abort ();
6224 }
6225 continue; /* Successfully executed one pattern command; keep going. */
6226
6227
6228 /* We goto here if a matching operation fails. */
6229 fail:
6230 IMMEDIATE_QUIT_CHECK;
6231 if (!FAIL_STACK_EMPTY ())
6232 {
6233 re_char *str, *pat;
6234 /* A restart point is known. Restore to that state. */
6235 DEBUG_PRINT ("\nFAIL:\n");
6236 POP_FAILURE_POINT (str, pat);
6237 switch (*pat++)
6238 {
6239 case on_failure_keep_string_jump:
6240 assert (str == NULL);
6241 goto continue_failure_jump;
6242
6243 case on_failure_jump_nastyloop:
6244 assert ((re_opcode_t)pat[-2] == no_op);
6245 PUSH_FAILURE_POINT (pat - 2, str);
6246 /* Fallthrough */
6247
6248 case on_failure_jump_loop:
6249 case on_failure_jump:
6250 case succeed_n:
6251 d = str;
6252 continue_failure_jump:
6253 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6254 p = pat + mcnt;
6255 break;
6256
6257 case no_op:
6258 /* A special frame used for nastyloops. */
6259 goto fail;
6260
6261 default:
6262 abort ();
6263 }
6264
6265 assert (p >= bufp->buffer && p <= pend);
6266
6267 if (d >= string1 && d <= end1)
6268 dend = end_match_1;
6269 }
6270 else
6271 break; /* Matching at this starting point really fails. */
6272 } /* for (;;) */
6273
6274 if (best_regs_set)
6275 goto restore_best_regs;
6276
6277 FREE_VARIABLES ();
6278
6279 return -1; /* Failure to match. */
6280 }
6281 \f
6282 /* Subroutine definitions for re_match_2. */
6283
6284 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6285 bytes; nonzero otherwise. */
6286
6287 static int
6288 bcmp_translate (const_re_char *s1, const_re_char *s2, register ssize_t len,
6289 RE_TRANSLATE_TYPE translate, const int target_multibyte)
6290 {
6291 register re_char *p1 = s1, *p2 = s2;
6292 re_char *p1_end = s1 + len;
6293 re_char *p2_end = s2 + len;
6294
6295 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6296 different lengths, but relying on a single `len' would break this. -sm */
6297 while (p1 < p1_end && p2 < p2_end)
6298 {
6299 int p1_charlen, p2_charlen;
6300 re_wchar_t p1_ch, p2_ch;
6301
6302 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6303 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6304
6305 if (RE_TRANSLATE (translate, p1_ch)
6306 != RE_TRANSLATE (translate, p2_ch))
6307 return 1;
6308
6309 p1 += p1_charlen, p2 += p2_charlen;
6310 }
6311
6312 if (p1 != p1_end || p2 != p2_end)
6313 return 1;
6314
6315 return 0;
6316 }
6317 \f
6318 /* Entry points for GNU code. */
6319
6320 /* re_compile_pattern is the GNU regular expression compiler: it
6321 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6322 Returns 0 if the pattern was valid, otherwise an error string.
6323
6324 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6325 are set in BUFP on entry.
6326
6327 We call regex_compile to do the actual compilation. */
6328
6329 const char *
6330 re_compile_pattern (const char *pattern, size_t length,
6331 struct re_pattern_buffer *bufp)
6332 {
6333 reg_errcode_t ret;
6334
6335 /* GNU code is written to assume at least RE_NREGS registers will be set
6336 (and at least one extra will be -1). */
6337 bufp->regs_allocated = REGS_UNALLOCATED;
6338
6339 /* And GNU code determines whether or not to get register information
6340 by passing null for the REGS argument to re_match, etc., not by
6341 setting no_sub. */
6342 bufp->no_sub = 0;
6343
6344 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6345
6346 if (!ret)
6347 return NULL;
6348 return gettext (re_error_msgid[(int) ret]);
6349 }
6350 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6351 \f
6352 /* Entry points compatible with 4.2 BSD regex library. We don't define
6353 them unless specifically requested. */
6354
6355 #if defined _REGEX_RE_COMP || defined _LIBC
6356
6357 /* BSD has one and only one pattern buffer. */
6358 static struct re_pattern_buffer re_comp_buf;
6359
6360 char *
6361 # ifdef _LIBC
6362 /* Make these definitions weak in libc, so POSIX programs can redefine
6363 these names if they don't use our functions, and still use
6364 regcomp/regexec below without link errors. */
6365 weak_function
6366 # endif
6367 re_comp (const char *s)
6368 {
6369 reg_errcode_t ret;
6370
6371 if (!s)
6372 {
6373 if (!re_comp_buf.buffer)
6374 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6375 return (char *) gettext ("No previous regular expression");
6376 return 0;
6377 }
6378
6379 if (!re_comp_buf.buffer)
6380 {
6381 re_comp_buf.buffer = malloc (200);
6382 if (re_comp_buf.buffer == NULL)
6383 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6384 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6385 re_comp_buf.allocated = 200;
6386
6387 re_comp_buf.fastmap = malloc (1 << BYTEWIDTH);
6388 if (re_comp_buf.fastmap == NULL)
6389 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6390 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6391 }
6392
6393 /* Since `re_exec' always passes NULL for the `regs' argument, we
6394 don't need to initialize the pattern buffer fields which affect it. */
6395
6396 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6397
6398 if (!ret)
6399 return NULL;
6400
6401 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6402 return (char *) gettext (re_error_msgid[(int) ret]);
6403 }
6404
6405
6406 int
6407 # ifdef _LIBC
6408 weak_function
6409 # endif
6410 re_exec (const char *s)
6411 {
6412 const size_t len = strlen (s);
6413 return re_search (&re_comp_buf, s, len, 0, len, 0) >= 0;
6414 }
6415 #endif /* _REGEX_RE_COMP */
6416 \f
6417 /* POSIX.2 functions. Don't define these for Emacs. */
6418
6419 #ifndef emacs
6420
6421 /* regcomp takes a regular expression as a string and compiles it.
6422
6423 PREG is a regex_t *. We do not expect any fields to be initialized,
6424 since POSIX says we shouldn't. Thus, we set
6425
6426 `buffer' to the compiled pattern;
6427 `used' to the length of the compiled pattern;
6428 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6429 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6430 RE_SYNTAX_POSIX_BASIC;
6431 `fastmap' to an allocated space for the fastmap;
6432 `fastmap_accurate' to zero;
6433 `re_nsub' to the number of subexpressions in PATTERN.
6434
6435 PATTERN is the address of the pattern string.
6436
6437 CFLAGS is a series of bits which affect compilation.
6438
6439 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6440 use POSIX basic syntax.
6441
6442 If REG_NEWLINE is set, then . and [^...] don't match newline.
6443 Also, regexec will try a match beginning after every newline.
6444
6445 If REG_ICASE is set, then we considers upper- and lowercase
6446 versions of letters to be equivalent when matching.
6447
6448 If REG_NOSUB is set, then when PREG is passed to regexec, that
6449 routine will report only success or failure, and nothing about the
6450 registers.
6451
6452 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6453 the return codes and their meanings.) */
6454
6455 reg_errcode_t
6456 regcomp (regex_t *_Restrict_ preg, const char *_Restrict_ pattern,
6457 int cflags)
6458 {
6459 reg_errcode_t ret;
6460 reg_syntax_t syntax
6461 = (cflags & REG_EXTENDED) ?
6462 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6463
6464 /* regex_compile will allocate the space for the compiled pattern. */
6465 preg->buffer = 0;
6466 preg->allocated = 0;
6467 preg->used = 0;
6468
6469 /* Try to allocate space for the fastmap. */
6470 preg->fastmap = malloc (1 << BYTEWIDTH);
6471
6472 if (cflags & REG_ICASE)
6473 {
6474 unsigned i;
6475
6476 preg->translate = malloc (CHAR_SET_SIZE * sizeof *preg->translate);
6477 if (preg->translate == NULL)
6478 return (int) REG_ESPACE;
6479
6480 /* Map uppercase characters to corresponding lowercase ones. */
6481 for (i = 0; i < CHAR_SET_SIZE; i++)
6482 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6483 }
6484 else
6485 preg->translate = NULL;
6486
6487 /* If REG_NEWLINE is set, newlines are treated differently. */
6488 if (cflags & REG_NEWLINE)
6489 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6490 syntax &= ~RE_DOT_NEWLINE;
6491 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6492 }
6493 else
6494 syntax |= RE_NO_NEWLINE_ANCHOR;
6495
6496 preg->no_sub = !!(cflags & REG_NOSUB);
6497
6498 /* POSIX says a null character in the pattern terminates it, so we
6499 can use strlen here in compiling the pattern. */
6500 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6501
6502 /* POSIX doesn't distinguish between an unmatched open-group and an
6503 unmatched close-group: both are REG_EPAREN. */
6504 if (ret == REG_ERPAREN)
6505 ret = REG_EPAREN;
6506
6507 if (ret == REG_NOERROR && preg->fastmap)
6508 { /* Compute the fastmap now, since regexec cannot modify the pattern
6509 buffer. */
6510 re_compile_fastmap (preg);
6511 if (preg->can_be_null)
6512 { /* The fastmap can't be used anyway. */
6513 free (preg->fastmap);
6514 preg->fastmap = NULL;
6515 }
6516 }
6517 return ret;
6518 }
6519 WEAK_ALIAS (__regcomp, regcomp)
6520
6521
6522 /* regexec searches for a given pattern, specified by PREG, in the
6523 string STRING.
6524
6525 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6526 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6527 least NMATCH elements, and we set them to the offsets of the
6528 corresponding matched substrings.
6529
6530 EFLAGS specifies `execution flags' which affect matching: if
6531 REG_NOTBOL is set, then ^ does not match at the beginning of the
6532 string; if REG_NOTEOL is set, then $ does not match at the end.
6533
6534 We return 0 if we find a match and REG_NOMATCH if not. */
6535
6536 reg_errcode_t
6537 regexec (const regex_t *_Restrict_ preg, const char *_Restrict_ string,
6538 size_t nmatch, regmatch_t pmatch[_Restrict_arr_], int eflags)
6539 {
6540 regoff_t ret;
6541 struct re_registers regs;
6542 regex_t private_preg;
6543 size_t len = strlen (string);
6544 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6545
6546 private_preg = *preg;
6547
6548 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6549 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6550
6551 /* The user has told us exactly how many registers to return
6552 information about, via `nmatch'. We have to pass that on to the
6553 matching routines. */
6554 private_preg.regs_allocated = REGS_FIXED;
6555
6556 if (want_reg_info)
6557 {
6558 regs.num_regs = nmatch;
6559 regs.start = TALLOC (nmatch * 2, regoff_t);
6560 if (regs.start == NULL)
6561 return REG_NOMATCH;
6562 regs.end = regs.start + nmatch;
6563 }
6564
6565 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6566 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6567 was a little bit longer but still only matching the real part.
6568 This works because the `endline' will check for a '\n' and will find a
6569 '\0', correctly deciding that this is not the end of a line.
6570 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6571 a convenient '\0' there. For all we know, the string could be preceded
6572 by '\n' which would throw things off. */
6573
6574 /* Perform the searching operation. */
6575 ret = re_search (&private_preg, string, len,
6576 /* start: */ 0, /* range: */ len,
6577 want_reg_info ? &regs : 0);
6578
6579 /* Copy the register information to the POSIX structure. */
6580 if (want_reg_info)
6581 {
6582 if (ret >= 0)
6583 {
6584 unsigned r;
6585
6586 for (r = 0; r < nmatch; r++)
6587 {
6588 pmatch[r].rm_so = regs.start[r];
6589 pmatch[r].rm_eo = regs.end[r];
6590 }
6591 }
6592
6593 /* If we needed the temporary register info, free the space now. */
6594 free (regs.start);
6595 }
6596
6597 /* We want zero return to mean success, unlike `re_search'. */
6598 return ret >= 0 ? REG_NOERROR : REG_NOMATCH;
6599 }
6600 WEAK_ALIAS (__regexec, regexec)
6601
6602
6603 /* Returns a message corresponding to an error code, ERR_CODE, returned
6604 from either regcomp or regexec. We don't use PREG here.
6605
6606 ERR_CODE was previously called ERRCODE, but that name causes an
6607 error with msvc8 compiler. */
6608
6609 size_t
6610 regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size)
6611 {
6612 const char *msg;
6613 size_t msg_size;
6614
6615 if (err_code < 0
6616 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6617 /* Only error codes returned by the rest of the code should be passed
6618 to this routine. If we are given anything else, or if other regex
6619 code generates an invalid error code, then the program has a bug.
6620 Dump core so we can fix it. */
6621 abort ();
6622
6623 msg = gettext (re_error_msgid[err_code]);
6624
6625 msg_size = strlen (msg) + 1; /* Includes the null. */
6626
6627 if (errbuf_size != 0)
6628 {
6629 if (msg_size > errbuf_size)
6630 {
6631 memcpy (errbuf, msg, errbuf_size - 1);
6632 errbuf[errbuf_size - 1] = 0;
6633 }
6634 else
6635 strcpy (errbuf, msg);
6636 }
6637
6638 return msg_size;
6639 }
6640 WEAK_ALIAS (__regerror, regerror)
6641
6642
6643 /* Free dynamically allocated space used by PREG. */
6644
6645 void
6646 regfree (regex_t *preg)
6647 {
6648 free (preg->buffer);
6649 preg->buffer = NULL;
6650
6651 preg->allocated = 0;
6652 preg->used = 0;
6653
6654 free (preg->fastmap);
6655 preg->fastmap = NULL;
6656 preg->fastmap_accurate = 0;
6657
6658 free (preg->translate);
6659 preg->translate = NULL;
6660 }
6661 WEAK_ALIAS (__regfree, regfree)
6662
6663 #endif /* not emacs */