| blob | 46065dda9e6f2b1a2ea2d0430a5ff0c56e98691d |
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.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
6 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
7 Free Software Foundation, Inc.
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3, or (at your option)
12 any later version.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
22 USA. */
24 /* TODO:
25 - structure the opcode space into opcode+flag.
26 - merge with glibc's regex.[ch].
27 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
28 need to modify the compiled regexp so that re_match can be reentrant.
29 - get rid of on_failure_jump_smart by doing the optimization in re_comp
30 rather than at run-time, so that re_match can be reentrant.
31 */
33 /* AIX requires this to be the first thing in the file. */
34 #if defined _AIX && !defined REGEX_MALLOC
35 #pragma alloca
36 #endif
38 #ifdef HAVE_CONFIG_H
39 # include <config.h>
40 #endif
42 #if defined STDC_HEADERS && !defined emacs
43 # include <stddef.h>
44 #else
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
47 #endif
49 /* Whether to use ISO C Amendment 1 wide char functions.
50 Those should not be used for Emacs since it uses its own. */
51 #if defined _LIBC
52 #define WIDE_CHAR_SUPPORT 1
53 #else
54 #define WIDE_CHAR_SUPPORT \
55 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
56 #endif
58 /* For platform which support the ISO C amendement 1 functionality we
59 support user defined character classes. */
60 #if WIDE_CHAR_SUPPORT
61 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
62 # include <wchar.h>
63 # include <wctype.h>
64 #endif
66 #ifdef _LIBC
67 /* We have to keep the namespace clean. */
68 # define regfree(preg) __regfree (preg)
69 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
70 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
71 # define regerror(err_code, preg, errbuf, errbuf_size) \
72 __regerror(err_code, preg, errbuf, errbuf_size)
73 # define re_set_registers(bu, re, nu, st, en) \
74 __re_set_registers (bu, re, nu, st, en)
75 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
76 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
77 # define re_match(bufp, string, size, pos, regs) \
78 __re_match (bufp, string, size, pos, regs)
79 # define re_search(bufp, string, size, startpos, range, regs) \
80 __re_search (bufp, string, size, startpos, range, regs)
81 # define re_compile_pattern(pattern, length, bufp) \
82 __re_compile_pattern (pattern, length, bufp)
83 # define re_set_syntax(syntax) __re_set_syntax (syntax)
84 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
85 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
86 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
88 /* Make sure we call libc's function even if the user overrides them. */
89 # define btowc __btowc
90 # define iswctype __iswctype
91 # define wctype __wctype
93 # define WEAK_ALIAS(a,b) weak_alias (a, b)
95 /* We are also using some library internals. */
96 # include <locale/localeinfo.h>
97 # include <locale/elem-hash.h>
98 # include <langinfo.h>
99 #else
100 # define WEAK_ALIAS(a,b)
101 #endif
103 /* This is for other GNU distributions with internationalized messages. */
104 #if HAVE_LIBINTL_H || defined _LIBC
105 # include <libintl.h>
106 #else
107 # define gettext(msgid) (msgid)
108 #endif
110 #ifndef gettext_noop
111 /* This define is so xgettext can find the internationalizable
112 strings. */
113 # define gettext_noop(String) String
114 #endif
116 /* The `emacs' switch turns on certain matching commands
117 that make sense only in Emacs. */
118 #ifdef emacs
120 # include <setjmp.h>
124 /* Make syntax table lookup grant data in gl_state. */
125 # define SYNTAX_ENTRY_VIA_PROPERTY
131 # ifdef malloc
132 # undef malloc
133 # endif
134 # define malloc xmalloc
135 # ifdef realloc
136 # undef realloc
137 # endif
138 # define realloc xrealloc
139 # ifdef free
140 # undef free
141 # endif
142 # define free xfree
144 /* Converts the pointer to the char to BEG-based offset from the start. */
145 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
146 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
148 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
149 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
150 # define RE_STRING_CHAR(p, multibyte) \
151 (multibyte ? (STRING_CHAR (p)) : (*(p)))
152 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
153 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
155 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
157 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
159 /* Set C a (possibly converted to multibyte) character before P. P
160 points into a string which is the virtual concatenation of STR1
161 (which ends at END1) or STR2 (which ends at END2). */
162 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
163 do { \
164 if (target_multibyte) \
165 { \
166 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
167 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
168 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
169 c = STRING_CHAR (dtemp); \
170 } \
171 else \
172 { \
173 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
174 (c) = RE_CHAR_TO_MULTIBYTE (c); \
175 } \
176 } while (0)
178 /* Set C a (possibly converted to multibyte) character at P, and set
179 LEN to the byte length of that character. */
180 # define GET_CHAR_AFTER(c, p, len) \
181 do { \
182 if (target_multibyte) \
183 (c) = STRING_CHAR_AND_LENGTH (p, len); \
184 else \
185 { \
186 (c) = *p; \
187 len = 1; \
188 (c) = RE_CHAR_TO_MULTIBYTE (c); \
189 } \
190 } while (0)
194 /* If we are not linking with Emacs proper,
195 we can't use the relocating allocator
196 even if config.h says that we can. */
197 # undef REL_ALLOC
199 # if defined STDC_HEADERS || defined _LIBC
200 # include <stdlib.h>
201 # else
204 # endif
206 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
211 {
215 {
218 }
220 }
226 {
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
232 else
235 {
238 }
240 }
242 # ifdef malloc
243 # undef malloc
244 # endif
245 # define malloc xmalloc
246 # ifdef realloc
247 # undef realloc
248 # endif
249 # define realloc xrealloc
251 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
252 If nothing else has been done, use the method below. */
253 # ifdef INHIBIT_STRING_HEADER
254 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
255 # if !defined bzero && !defined bcopy
256 # undef INHIBIT_STRING_HEADER
257 # endif
258 # endif
259 # endif
261 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
262 This is used in most programs--a few other programs avoid this
263 by defining INHIBIT_STRING_HEADER. */
264 # ifndef INHIBIT_STRING_HEADER
265 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
266 # include <string.h>
267 # ifndef bzero
268 # ifndef _LIBC
270 # else
271 # define bzero(s, n) __bzero (s, n)
272 # endif
273 # endif
274 # else
275 # include <strings.h>
276 # ifndef memcmp
277 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
278 # endif
279 # ifndef memcpy
280 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
281 # endif
282 # endif
283 # endif
285 /* Define the syntax stuff for \<, \>, etc. */
287 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
290 # define SWITCH_ENUM_CAST(x) (x)
292 /* Dummy macros for non-Emacs environments. */
293 # define CHAR_CHARSET(c) 0
294 # define CHARSET_LEADING_CODE_BASE(c) 0
295 # define MAX_MULTIBYTE_LENGTH 1
296 # define RE_MULTIBYTE_P(x) 0
297 # define RE_TARGET_MULTIBYTE_P(x) 0
298 # define WORD_BOUNDARY_P(c1, c2) (0)
299 # define CHAR_HEAD_P(p) (1)
300 # define SINGLE_BYTE_CHAR_P(c) (1)
301 # define SAME_CHARSET_P(c1, c2) (1)
302 # define BYTES_BY_CHAR_HEAD(p) (1)
303 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
304 # define STRING_CHAR(p) (*(p))
305 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
306 # define CHAR_STRING(c, s) (*(s) = (c), 1)
307 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
308 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
309 # define RE_CHAR_TO_MULTIBYTE(c) (c)
310 # define RE_CHAR_TO_UNIBYTE(c) (c)
311 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
312 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
313 # define GET_CHAR_AFTER(c, p, len) \
314 (c = *p, len = 1)
315 # define MAKE_CHAR(charset, c1, c2) (c1)
316 # define BYTE8_TO_CHAR(c) (c)
317 # define CHAR_BYTE8_P(c) (0)
318 # define CHAR_LEADING_CODE(c) (c)
322 #ifndef RE_TRANSLATE
323 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
324 # define RE_TRANSLATE_P(TBL) (TBL)
325 #endif
326 \f
327 /* Get the interface, including the syntax bits. */
330 /* isalpha etc. are used for the character classes. */
331 #include <ctype.h>
333 #ifdef emacs
335 /* 1 if C is an ASCII character. */
336 # define IS_REAL_ASCII(c) ((c) < 0200)
338 /* 1 if C is a unibyte character. */
339 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
341 /* The Emacs definitions should not be directly affected by locales. */
343 /* In Emacs, these are only used for single-byte characters. */
350 /* This is only used for single-byte characters. */
353 /* The rest must handle multibyte characters. */
355 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
357 : 1)
359 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
361 : 1)
363 # define ISALNUM(c) (IS_REAL_ASCII (c) \
367 : SYNTAX (c) == Sword)
369 # define ISALPHA(c) (IS_REAL_ASCII (c) \
372 : SYNTAX (c) == Sword)
374 # define ISLOWER(c) (LOWERCASEP (c))
376 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
381 : SYNTAX (c) != Sword)
383 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
385 # define ISUPPER(c) (UPPERCASEP (c))
387 # define ISWORD(c) (SYNTAX (c) == Sword)
391 /* Jim Meyering writes:
393 "... Some ctype macros are valid only for character codes that
394 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
395 using /bin/cc or gcc but without giving an ansi option). So, all
396 ctype uses should be through macros like ISPRINT... If
397 STDC_HEADERS is defined, then autoconf has verified that the ctype
398 macros don't need to be guarded with references to isascii. ...
399 Defining isascii to 1 should let any compiler worth its salt
400 eliminate the && through constant folding."
401 Solaris defines some of these symbols so we must undefine them first. */
403 # undef ISASCII
404 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
405 # define ISASCII(c) 1
406 # else
407 # define ISASCII(c) isascii(c)
408 # endif
410 /* 1 if C is an ASCII character. */
411 # define IS_REAL_ASCII(c) ((c) < 0200)
413 /* This distinction is not meaningful, except in Emacs. */
414 # define ISUNIBYTE(c) 1
416 # ifdef isblank
417 # define ISBLANK(c) (ISASCII (c) && isblank (c))
418 # else
420 # endif
421 # ifdef isgraph
422 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
423 # else
424 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
425 # endif
427 # undef ISPRINT
428 # define ISPRINT(c) (ISASCII (c) && isprint (c))
429 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
430 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
431 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
432 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
433 # define ISLOWER(c) (ISASCII (c) && islower (c))
434 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
435 # define ISSPACE(c) (ISASCII (c) && isspace (c))
436 # define ISUPPER(c) (ISASCII (c) && isupper (c))
437 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
439 # define ISWORD(c) ISALPHA(c)
441 # ifdef _tolower
442 # define TOLOWER(c) _tolower(c)
443 # else
444 # define TOLOWER(c) tolower(c)
445 # endif
447 /* How many characters in the character set. */
448 # define CHAR_SET_SIZE 256
450 # ifdef SYNTAX_TABLE
458 static void
460 {
476 }
480 # define SYNTAX(c) re_syntax_table[(c)]
483 \f
484 #ifndef NULL
485 # define NULL (void *)0
486 #endif
488 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
489 since ours (we hope) works properly with all combinations of
490 machines, compilers, `char' and `unsigned char' argument types.
491 (Per Bothner suggested the basic approach.) */
492 #undef SIGN_EXTEND_CHAR
493 #if __STDC__
494 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
496 /* As in Harbison and Steele. */
497 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
498 #endif
499 \f
500 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
501 use `alloca' instead of `malloc'. This is because using malloc in
502 re_search* or re_match* could cause memory leaks when C-g is used in
503 Emacs; also, malloc is slower and causes storage fragmentation. On
504 the other hand, malloc is more portable, and easier to debug.
506 Because we sometimes use alloca, some routines have to be macros,
507 not functions -- `alloca'-allocated space disappears at the end of the
508 function it is called in. */
510 #ifdef REGEX_MALLOC
512 # define REGEX_ALLOCATE malloc
513 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
514 # define REGEX_FREE free
518 /* Emacs already defines alloca, sometimes. */
519 # ifndef alloca
521 /* Make alloca work the best possible way. */
522 # ifdef __GNUC__
523 # define alloca __builtin_alloca
525 # ifdef HAVE_ALLOCA_H
526 # include <alloca.h>
532 # define REGEX_ALLOCATE alloca
534 /* Assumes a `char *destination' variable. */
535 # define REGEX_REALLOCATE(source, osize, nsize) \
536 (destination = (char *) alloca (nsize), \
537 memcpy (destination, source, osize))
539 /* No need to do anything to free, after alloca. */
544 /* Define how to allocate the failure stack. */
546 #if defined REL_ALLOC && defined REGEX_MALLOC
548 # define REGEX_ALLOCATE_STACK(size) \
549 r_alloc (&failure_stack_ptr, (size))
550 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
551 r_re_alloc (&failure_stack_ptr, (nsize))
552 # define REGEX_FREE_STACK(ptr) \
553 r_alloc_free (&failure_stack_ptr)
557 # ifdef REGEX_MALLOC
559 # define REGEX_ALLOCATE_STACK malloc
560 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
561 # define REGEX_FREE_STACK free
565 # define REGEX_ALLOCATE_STACK alloca
567 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
568 REGEX_REALLOCATE (source, osize, nsize)
569 /* No need to explicitly free anything. */
570 # define REGEX_FREE_STACK(arg) ((void)0)
576 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
577 `string1' or just past its end. This works if PTR is NULL, which is
578 a good thing. */
579 #define FIRST_STRING_P(ptr) \
580 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
582 /* (Re)Allocate N items of type T using malloc, or fail. */
583 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
584 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
585 #define RETALLOC_IF(addr, n, t) \
586 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
587 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
591 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
593 #undef MAX
594 #undef MIN
595 #define MAX(a, b) ((a) > (b) ? (a) : (b))
596 #define MIN(a, b) ((a) < (b) ? (a) : (b))
598 /* Type of source-pattern and string chars. */
602 #define false 0
603 #define true 1
611 \f
612 /* These are the command codes that appear in compiled regular
613 expressions. Some opcodes are followed by argument bytes. A
614 command code can specify any interpretation whatsoever for its
615 arguments. Zero bytes may appear in the compiled regular expression. */
618 {
621 /* Succeed right away--no more backtracking. */
622 succeed,
624 /* Followed by one byte giving n, then by n literal bytes. */
625 exactn,
627 /* Matches any (more or less) character. */
628 anychar,
630 /* Matches any one char belonging to specified set. First
631 following byte is number of bitmap bytes. Then come bytes
632 for a bitmap saying which chars are in. Bits in each byte
633 are ordered low-bit-first. A character is in the set if its
634 bit is 1. A character too large to have a bit in the map is
635 automatically not in the set.
637 If the length byte has the 0x80 bit set, then that stuff
638 is followed by a range table:
639 2 bytes of flags for character sets (low 8 bits, high 8 bits)
640 See RANGE_TABLE_WORK_BITS below.
641 2 bytes, the number of pairs that follow (upto 32767)
642 pairs, each 2 multibyte characters,
643 each multibyte character represented as 3 bytes. */
644 charset,
646 /* Same parameters as charset, but match any character that is
647 not one of those specified. */
648 charset_not,
650 /* Start remembering the text that is matched, for storing in a
651 register. Followed by one byte with the register number, in
652 the range 0 to one less than the pattern buffer's re_nsub
653 field. */
654 start_memory,
656 /* Stop remembering the text that is matched and store it in a
657 memory register. Followed by one byte with the register
658 number, in the range 0 to one less than `re_nsub' in the
659 pattern buffer. */
660 stop_memory,
662 /* Match a duplicate of something remembered. Followed by one
663 byte containing the register number. */
664 duplicate,
666 /* Fail unless at beginning of line. */
667 begline,
669 /* Fail unless at end of line. */
670 endline,
672 /* Succeeds if at beginning of buffer (if emacs) or at beginning
673 of string to be matched (if not). */
674 begbuf,
676 /* Analogously, for end of buffer/string. */
677 endbuf,
679 /* Followed by two byte relative address to which to jump. */
680 jump,
682 /* Followed by two-byte relative address of place to resume at
683 in case of failure. */
684 on_failure_jump,
686 /* Like on_failure_jump, but pushes a placeholder instead of the
687 current string position when executed. */
688 on_failure_keep_string_jump,
690 /* Just like `on_failure_jump', except that it checks that we
691 don't get stuck in an infinite loop (matching an empty string
692 indefinitely). */
693 on_failure_jump_loop,
695 /* Just like `on_failure_jump_loop', except that it checks for
696 a different kind of loop (the kind that shows up with non-greedy
697 operators). This operation has to be immediately preceded
698 by a `no_op'. */
699 on_failure_jump_nastyloop,
701 /* A smart `on_failure_jump' used for greedy * and + operators.
702 It analyses the loop before which it is put and if the
703 loop does not require backtracking, it changes itself to
704 `on_failure_keep_string_jump' and short-circuits the loop,
705 else it just defaults to changing itself into `on_failure_jump'.
706 It assumes that it is pointing to just past a `jump'. */
707 on_failure_jump_smart,
709 /* Followed by two-byte relative address and two-byte number n.
710 After matching N times, jump to the address upon failure.
711 Does not work if N starts at 0: use on_failure_jump_loop
712 instead. */
713 succeed_n,
715 /* Followed by two-byte relative address, and two-byte number n.
716 Jump to the address N times, then fail. */
717 jump_n,
719 /* Set the following two-byte relative address to the
720 subsequent two-byte number. The address *includes* the two
721 bytes of number. */
722 set_number_at,
733 /* Matches any character whose syntax is specified. Followed by
734 a byte which contains a syntax code, e.g., Sword. */
735 syntaxspec,
737 /* Matches any character whose syntax is not that specified. */
738 notsyntaxspec
740 #ifdef emacs
745 /* Matches any character whose category-set contains the specified
746 category. The operator is followed by a byte which contains a
747 category code (mnemonic ASCII character). */
748 categoryspec,
750 /* Matches any character whose category-set does not contain the
751 specified category. The operator is followed by a byte which
752 contains the category code (mnemonic ASCII character). */
753 notcategoryspec
756 \f
757 /* Common operations on the compiled pattern. */
759 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
761 #define STORE_NUMBER(destination, number) \
762 do { \
763 (destination)[0] = (number) & 0377; \
764 (destination)[1] = (number) >> 8; \
765 } while (0)
767 /* Same as STORE_NUMBER, except increment DESTINATION to
768 the byte after where the number is stored. Therefore, DESTINATION
769 must be an lvalue. */
771 #define STORE_NUMBER_AND_INCR(destination, number) \
772 do { \
773 STORE_NUMBER (destination, number); \
774 (destination) += 2; \
775 } while (0)
777 /* Put into DESTINATION a number stored in two contiguous bytes starting
778 at SOURCE. */
780 #define EXTRACT_NUMBER(destination, source) \
781 do { \
782 (destination) = *(source) & 0377; \
783 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
784 } while (0)
786 #ifdef DEBUG
788 static void
792 {
796 }
799 # undef EXTRACT_NUMBER
800 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
805 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
806 SOURCE must be an lvalue. */
808 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
809 do { \
810 EXTRACT_NUMBER (destination, source); \
811 (source) += 2; \
812 } while (0)
814 #ifdef DEBUG
817 static void
821 {
824 }
826 # ifndef EXTRACT_MACROS
827 # undef EXTRACT_NUMBER_AND_INCR
828 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
829 extract_number_and_incr (&dest, &src)
833 \f
834 /* Store a multibyte character in three contiguous bytes starting
835 DESTINATION, and increment DESTINATION to the byte after where the
836 character is stored. Therefore, DESTINATION must be an lvalue. */
838 #define STORE_CHARACTER_AND_INCR(destination, character) \
839 do { \
840 (destination)[0] = (character) & 0377; \
841 (destination)[1] = ((character) >> 8) & 0377; \
842 (destination)[2] = (character) >> 16; \
843 (destination) += 3; \
844 } while (0)
846 /* Put into DESTINATION a character stored in three contiguous bytes
847 starting at SOURCE. */
849 #define EXTRACT_CHARACTER(destination, source) \
850 do { \
851 (destination) = ((source)[0] \
852 | ((source)[1] << 8) \
853 | ((source)[2] << 16)); \
854 } while (0)
857 /* Macros for charset. */
859 /* Size of bitmap of charset P in bytes. P is a start of charset,
860 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
861 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
863 /* Nonzero if charset P has range table. */
864 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
866 /* Return the address of range table of charset P. But not the start
867 of table itself, but the before where the number of ranges is
868 stored. `2 +' means to skip re_opcode_t and size of bitmap,
869 and the 2 bytes of flags at the start of the range table. */
870 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
872 /* Extract the bit flags that start a range table. */
873 #define CHARSET_RANGE_TABLE_BITS(p) \
874 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
875 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
877 /* Test if C is listed in the bitmap of charset P. */
878 #define CHARSET_LOOKUP_BITMAP(p, c) \
879 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
880 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
882 /* Return the address of end of RANGE_TABLE. COUNT is number of
883 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
884 is start of range and end of range. `* 3' is size of each start
885 and end. */
886 #define CHARSET_RANGE_TABLE_END(range_table, count) \
887 ((range_table) + (count) * 2 * 3)
889 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
890 COUNT is number of ranges in RANGE_TABLE. */
891 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
892 do \
893 { \
894 re_wchar_t range_start, range_end; \
895 re_char *p; \
896 re_char *range_table_end \
897 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
898 \
899 for (p = (range_table); p < range_table_end; p += 2 * 3) \
900 { \
901 EXTRACT_CHARACTER (range_start, p); \
902 EXTRACT_CHARACTER (range_end, p + 3); \
903 \
904 if (range_start <= (c) && (c) <= range_end) \
905 { \
906 (not) = !(not); \
907 break; \
908 } \
909 } \
910 } \
911 while (0)
913 /* Test if C is in range table of CHARSET. The flag NOT is negated if
914 C is listed in it. */
915 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
916 do \
917 { \
919 int count; \
920 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
921 \
922 EXTRACT_NUMBER_AND_INCR (count, range_table); \
923 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
924 } \
925 while (0)
926 \f
927 /* If DEBUG is defined, Regex prints many voluminous messages about what
928 it is doing (if the variable `debug' is nonzero). If linked with the
929 main program in `iregex.c', you can enter patterns and strings
930 interactively. And if linked with the main program in `main.c' and
931 the other test files, you can run the already-written tests. */
933 #ifdef DEBUG
935 /* We use standard I/O for debugging. */
936 # include <stdio.h>
938 /* It is useful to test things that ``must'' be true when debugging. */
939 # include <assert.h>
943 # define DEBUG_STATEMENT(e) e
944 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
945 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
946 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
947 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
948 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
949 if (debug > 0) print_partial_compiled_pattern (s, e)
950 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
951 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
954 /* Print the fastmap in human-readable form. */
956 void
959 {
964 {
966 {
970 {
972 i++;
973 }
975 {
978 }
979 }
980 }
982 }
985 /* Print a compiled pattern string in human-readable form, starting at
986 the START pointer into it and ending just before the pointer END. */
988 void
992 {
998 {
1001 }
1003 /* Loop over pattern commands. */
1005 {
1009 {
1021 do
1022 {
1024 }
1046 {
1061 {
1062 /* Are we starting a range? */
1064 {
1067 }
1068 /* Have we broken a range? */
1070 {
1073 }
1079 }
1089 {
1093 /* ??? Should print the range table; for now, just skip it. */
1097 }
1098 }
1193 # ifdef emacs
1229 }
1232 }
1235 }
1238 void
1241 {
1249 {
1252 }
1262 /* Perhaps we should print the translate table? */
1263 }
1266 void
1273 {
1278 else
1279 {
1281 {
1286 }
1290 }
1291 }
1295 # undef assert
1296 # define assert(e)
1298 # define DEBUG_STATEMENT(e)
1299 # define DEBUG_PRINT1(x)
1300 # define DEBUG_PRINT2(x1, x2)
1301 # define DEBUG_PRINT3(x1, x2, x3)
1302 # define DEBUG_PRINT4(x1, x2, x3, x4)
1303 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1304 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1307 \f
1308 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1309 also be assigned to arbitrarily: each pattern buffer stores its own
1310 syntax, so it can be changed between regex compilations. */
1311 /* This has no initializer because initialized variables in Emacs
1312 become read-only after dumping. */
1313 reg_syntax_t re_syntax_options;
1316 /* Specify the precise syntax of regexps for compilation. This provides
1317 for compatibility for various utilities which historically have
1318 different, incompatible syntaxes.
1320 The argument SYNTAX is a bit mask comprised of the various bits
1321 defined in regex.h. We return the old syntax. */
1323 reg_syntax_t
1325 reg_syntax_t syntax;
1326 {
1331 }
1334 /* Regexp to use to replace spaces, or NULL meaning don't. */
1337 void
1340 {
1342 }
1344 \f
1345 /* This table gives an error message for each of the error codes listed
1346 in regex.h. Obviously the order here has to be same as there.
1347 POSIX doesn't require that we do anything for REG_NOERROR,
1348 but why not be nice? */
1351 {
1370 };
1371 \f
1372 /* Avoiding alloca during matching, to placate r_alloc. */
1374 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1375 searching and matching functions should not call alloca. On some
1376 systems, alloca is implemented in terms of malloc, and if we're
1377 using the relocating allocator routines, then malloc could cause a
1378 relocation, which might (if the strings being searched are in the
1379 ralloc heap) shift the data out from underneath the regexp
1380 routines.
1382 Here's another reason to avoid allocation: Emacs
1383 processes input from X in a signal handler; processing X input may
1384 call malloc; if input arrives while a matching routine is calling
1385 malloc, then we're scrod. But Emacs can't just block input while
1386 calling matching routines; then we don't notice interrupts when
1387 they come in. So, Emacs blocks input around all regexp calls
1388 except the matching calls, which it leaves unprotected, in the
1389 faith that they will not malloc. */
1391 /* Normally, this is fine. */
1392 #define MATCH_MAY_ALLOCATE
1394 /* The match routines may not allocate if (1) they would do it with malloc
1395 and (2) it's not safe for them to use malloc.
1396 Note that if REL_ALLOC is defined, matching would not use malloc for the
1397 failure stack, but we would still use it for the register vectors;
1398 so REL_ALLOC should not affect this. */
1399 #if defined REGEX_MALLOC && defined emacs
1400 # undef MATCH_MAY_ALLOCATE
1401 #endif
1403 \f
1404 /* Failure stack declarations and macros; both re_compile_fastmap and
1405 re_match_2 use a failure stack. These have to be macros because of
1406 REGEX_ALLOCATE_STACK. */
1409 /* Approximate number of failure points for which to initially allocate space
1410 when matching. If this number is exceeded, we allocate more
1411 space, so it is not a hard limit. */
1412 #ifndef INIT_FAILURE_ALLOC
1413 # define INIT_FAILURE_ALLOC 20
1414 #endif
1416 /* Roughly the maximum number of failure points on the stack. Would be
1417 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1418 This is a variable only so users of regex can assign to it; we never
1419 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1420 before using it, so it should probably be a byte-count instead. */
1421 # if defined MATCH_MAY_ALLOCATE
1422 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1423 whose default stack limit is 2mb. In order for a larger
1424 value to work reliably, you have to try to make it accord
1425 with the process stack limit. */
1427 # else
1429 # endif
1431 union fail_stack_elt
1432 {
1434 /* This should be the biggest `int' that's no bigger than a pointer. */
1436 };
1441 {
1448 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1449 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1452 /* Define macros to initialize and free the failure stack.
1453 Do `return -2' if the alloc fails. */
1455 #ifdef MATCH_MAY_ALLOCATE
1456 # define INIT_FAIL_STACK() \
1457 do { \
1458 fail_stack.stack = (fail_stack_elt_t *) \
1459 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1460 * sizeof (fail_stack_elt_t)); \
1461 \
1462 if (fail_stack.stack == NULL) \
1463 return -2; \
1464 \
1465 fail_stack.size = INIT_FAILURE_ALLOC; \
1466 fail_stack.avail = 0; \
1467 fail_stack.frame = 0; \
1468 } while (0)
1470 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1471 #else
1472 # define INIT_FAIL_STACK() \
1473 do { \
1474 fail_stack.avail = 0; \
1475 fail_stack.frame = 0; \
1476 } while (0)
1478 # define RESET_FAIL_STACK() ((void)0)
1479 #endif
1482 /* Double the size of FAIL_STACK, up to a limit
1483 which allows approximately `re_max_failures' items.
1485 Return 1 if succeeds, and 0 if either ran out of memory
1486 allocating space for it or it was already too large.
1488 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1490 /* Factor to increase the failure stack size by
1491 when we increase it.
1492 This used to be 2, but 2 was too wasteful
1493 because the old discarded stacks added up to as much space
1494 were as ultimate, maximum-size stack. */
1495 #define FAIL_STACK_GROWTH_FACTOR 4
1497 #define GROW_FAIL_STACK(fail_stack) \
1498 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1499 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1500 ? 0 \
1501 : ((fail_stack).stack \
1502 = (fail_stack_elt_t *) \
1503 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1504 (fail_stack).size * sizeof (fail_stack_elt_t), \
1505 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1506 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1507 * FAIL_STACK_GROWTH_FACTOR))), \
1508 \
1509 (fail_stack).stack == NULL \
1510 ? 0 \
1511 : ((fail_stack).size \
1512 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1513 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1514 * FAIL_STACK_GROWTH_FACTOR)) \
1515 / sizeof (fail_stack_elt_t)), \
1516 1)))
1519 /* Push a pointer value onto the failure stack.
1520 Assumes the variable `fail_stack'. Probably should only
1521 be called from within `PUSH_FAILURE_POINT'. */
1522 #define PUSH_FAILURE_POINTER(item) \
1523 fail_stack.stack[fail_stack.avail++].pointer = (item)
1525 /* This pushes an integer-valued item onto the failure stack.
1526 Assumes the variable `fail_stack'. Probably should only
1527 be called from within `PUSH_FAILURE_POINT'. */
1528 #define PUSH_FAILURE_INT(item) \
1529 fail_stack.stack[fail_stack.avail++].integer = (item)
1531 /* Push a fail_stack_elt_t value onto the failure stack.
1532 Assumes the variable `fail_stack'. Probably should only
1533 be called from within `PUSH_FAILURE_POINT'. */
1534 #define PUSH_FAILURE_ELT(item) \
1535 fail_stack.stack[fail_stack.avail++] = (item)
1537 /* These three POP... operations complement the three PUSH... operations.
1538 All assume that `fail_stack' is nonempty. */
1539 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1540 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1541 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1543 /* Individual items aside from the registers. */
1544 #define NUM_NONREG_ITEMS 3
1546 /* Used to examine the stack (to detect infinite loops). */
1547 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1548 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1549 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1550 #define TOP_FAILURE_HANDLE() fail_stack.frame
1553 #define ENSURE_FAIL_STACK(space) \
1554 while (REMAINING_AVAIL_SLOTS <= space) { \
1555 if (!GROW_FAIL_STACK (fail_stack)) \
1556 return -2; \
1559 }
1561 /* Push register NUM onto the stack. */
1562 #define PUSH_FAILURE_REG(num) \
1563 do { \
1564 char *destination; \
1565 ENSURE_FAIL_STACK(3); \
1567 num, regstart[num], regend[num]); \
1568 PUSH_FAILURE_POINTER (regstart[num]); \
1569 PUSH_FAILURE_POINTER (regend[num]); \
1570 PUSH_FAILURE_INT (num); \
1571 } while (0)
1573 /* Change the counter's value to VAL, but make sure that it will
1574 be reset when backtracking. */
1575 #define PUSH_NUMBER(ptr,val) \
1576 do { \
1577 char *destination; \
1578 int c; \
1579 ENSURE_FAIL_STACK(3); \
1580 EXTRACT_NUMBER (c, ptr); \
1582 PUSH_FAILURE_INT (c); \
1583 PUSH_FAILURE_POINTER (ptr); \
1584 PUSH_FAILURE_INT (-1); \
1585 STORE_NUMBER (ptr, val); \
1586 } while (0)
1588 /* Pop a saved register off the stack. */
1589 #define POP_FAILURE_REG_OR_COUNT() \
1590 do { \
1591 int reg = POP_FAILURE_INT (); \
1592 if (reg == -1) \
1593 { \
1596 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1597 reg = POP_FAILURE_INT (); \
1598 STORE_NUMBER (ptr, reg); \
1600 } \
1601 else \
1602 { \
1603 regend[reg] = POP_FAILURE_POINTER (); \
1604 regstart[reg] = POP_FAILURE_POINTER (); \
1606 reg, regstart[reg], regend[reg]); \
1607 } \
1608 } while (0)
1610 /* Check that we are not stuck in an infinite loop. */
1611 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1612 do { \
1613 int failure = TOP_FAILURE_HANDLE (); \
1615 while (failure > 0 \
1616 && (FAILURE_STR (failure) == string_place \
1617 || FAILURE_STR (failure) == NULL)) \
1618 { \
1619 assert (FAILURE_PAT (failure) >= bufp->buffer \
1620 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1621 if (FAILURE_PAT (failure) == pat_cur) \
1622 { \
1623 cycle = 1; \
1624 break; \
1625 } \
1627 failure = NEXT_FAILURE_HANDLE(failure); \
1628 } \
1630 } while (0)
1632 /* Push the information about the state we will need
1633 if we ever fail back to it.
1635 Requires variables fail_stack, regstart, regend and
1636 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1637 declared.
1639 Does `return FAILURE_CODE' if runs out of memory. */
1641 #define PUSH_FAILURE_POINT(pattern, string_place) \
1642 do { \
1643 char *destination; \
1644 /* Must be int, so when we don't save any registers, the arithmetic \
1646 \
1647 DEBUG_STATEMENT (nfailure_points_pushed++); \
1651 \
1652 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1653 \
1655 \
1657 PUSH_FAILURE_INT (fail_stack.frame); \
1658 \
1660 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1662 PUSH_FAILURE_POINTER (string_place); \
1663 \
1665 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1666 PUSH_FAILURE_POINTER (pattern); \
1667 \
1669 fail_stack.frame = fail_stack.avail; \
1670 } while (0)
1672 /* Estimate the size of data pushed by a typical failure stack entry.
1673 An estimate is all we need, because all we use this for
1674 is to choose a limit for how big to make the failure stack. */
1675 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1676 #define TYPICAL_FAILURE_SIZE 20
1678 /* How many items can still be added to the stack without overflowing it. */
1679 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1682 /* Pops what PUSH_FAIL_STACK pushes.
1684 We restore into the parameters, all of which should be lvalues:
1685 STR -- the saved data position.
1686 PAT -- the saved pattern position.
1687 REGSTART, REGEND -- arrays of string positions.
1689 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1690 `pend', `string1', `size1', `string2', and `size2'. */
1692 #define POP_FAILURE_POINT(str, pat) \
1693 do { \
1694 assert (!FAIL_STACK_EMPTY ()); \
1695 \
1700 \
1702 while (fail_stack.frame < fail_stack.avail) \
1703 POP_FAILURE_REG_OR_COUNT (); \
1704 \
1705 pat = POP_FAILURE_POINTER (); \
1707 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1708 \
1709 /* If the saved string location is NULL, it came from an \
1710 on_failure_keep_string_jump opcode, and we want to throw away the \
1712 str = POP_FAILURE_POINTER (); \
1714 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1716 \
1717 fail_stack.frame = POP_FAILURE_INT (); \
1719 \
1720 assert (fail_stack.avail >= 0); \
1721 assert (fail_stack.frame <= fail_stack.avail); \
1722 \
1723 DEBUG_STATEMENT (nfailure_points_popped++); \
1727 \f
1728 /* Registers are set to a sentinel when they haven't yet matched. */
1729 #define REG_UNSET(e) ((e) == NULL)
1730 \f
1731 /* Subroutine declarations and macros for regex_compile. */
1734 reg_syntax_t syntax,
1745 reg_syntax_t syntax));
1748 reg_syntax_t syntax));
1753 /* Fetch the next character in the uncompiled pattern, with no
1754 translation. */
1755 #define PATFETCH(c) \
1756 do { \
1757 int len; \
1758 if (p == pend) return REG_EEND; \
1759 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1760 p += len; \
1761 } while (0)
1764 /* If `translate' is non-null, return translate[D], else just D. We
1765 cast the subscript to translate because some data is declared as
1766 `char *', to avoid warnings when a string constant is passed. But
1767 when we use a character as a subscript we must make it unsigned. */
1768 #ifndef TRANSLATE
1769 # define TRANSLATE(d) \
1770 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1771 #endif
1774 /* Macros for outputting the compiled pattern into `buffer'. */
1776 /* If the buffer isn't allocated when it comes in, use this. */
1777 #define INIT_BUF_SIZE 32
1779 /* Make sure we have at least N more bytes of space in buffer. */
1780 #define GET_BUFFER_SPACE(n) \
1781 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1782 EXTEND_BUFFER ()
1784 /* Make sure we have one more byte of buffer space and then add C to it. */
1785 #define BUF_PUSH(c) \
1786 do { \
1787 GET_BUFFER_SPACE (1); \
1788 *b++ = (unsigned char) (c); \
1789 } while (0)
1792 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1793 #define BUF_PUSH_2(c1, c2) \
1794 do { \
1795 GET_BUFFER_SPACE (2); \
1796 *b++ = (unsigned char) (c1); \
1797 *b++ = (unsigned char) (c2); \
1798 } while (0)
1801 /* As with BUF_PUSH_2, except for three bytes. */
1802 #define BUF_PUSH_3(c1, c2, c3) \
1803 do { \
1804 GET_BUFFER_SPACE (3); \
1805 *b++ = (unsigned char) (c1); \
1806 *b++ = (unsigned char) (c2); \
1807 *b++ = (unsigned char) (c3); \
1808 } while (0)
1811 /* Store a jump with opcode OP at LOC to location TO. We store a
1812 relative address offset by the three bytes the jump itself occupies. */
1813 #define STORE_JUMP(op, loc, to) \
1814 store_op1 (op, loc, (to) - (loc) - 3)
1816 /* Likewise, for a two-argument jump. */
1817 #define STORE_JUMP2(op, loc, to, arg) \
1818 store_op2 (op, loc, (to) - (loc) - 3, arg)
1820 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1821 #define INSERT_JUMP(op, loc, to) \
1822 insert_op1 (op, loc, (to) - (loc) - 3, b)
1824 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1825 #define INSERT_JUMP2(op, loc, to, arg) \
1826 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1829 /* This is not an arbitrary limit: the arguments which represent offsets
1830 into the pattern are two bytes long. So if 2^15 bytes turns out to
1831 be too small, many things would have to change. */
1832 # define MAX_BUF_SIZE (1L << 15)
1835 /* Any other compiler which, like MSC, has allocation limit below 2^16
1836 bytes will have to use approach similar to what was done below for
1837 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1838 reallocating to 0 bytes. Such thing is not going to work too well.
1839 You have been warned!! */
1840 #if defined _MSC_VER && !defined WIN32
1841 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1842 # define MAX_BUF_SIZE 65500L
1843 #else
1844 # define MAX_BUF_SIZE (1L << 16)
1845 #endif
1848 /* Extend the buffer by twice its current size via realloc and
1849 reset the pointers that pointed into the old block to point to the
1850 correct places in the new one. If extending the buffer results in it
1851 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1852 #if __BOUNDED_POINTERS__
1853 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1854 # define MOVE_BUFFER_POINTER(P) \
1855 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1856 SET_HIGH_BOUND (P), \
1857 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1858 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1859 else \
1860 { \
1861 SET_HIGH_BOUND (b); \
1862 SET_HIGH_BOUND (begalt); \
1863 if (fixup_alt_jump) \
1864 SET_HIGH_BOUND (fixup_alt_jump); \
1865 if (laststart) \
1866 SET_HIGH_BOUND (laststart); \
1867 if (pending_exact) \
1868 SET_HIGH_BOUND (pending_exact); \
1869 }
1870 #else
1871 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1872 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1873 #endif
1874 #define EXTEND_BUFFER() \
1875 do { \
1876 unsigned char *old_buffer = bufp->buffer; \
1877 if (bufp->allocated == MAX_BUF_SIZE) \
1878 return REG_ESIZE; \
1879 bufp->allocated <<= 1; \
1880 if (bufp->allocated > MAX_BUF_SIZE) \
1881 bufp->allocated = MAX_BUF_SIZE; \
1882 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1883 if (bufp->buffer == NULL) \
1884 return REG_ESPACE; \
1886 if (old_buffer != bufp->buffer) \
1887 { \
1888 unsigned char *new_buffer = bufp->buffer; \
1889 MOVE_BUFFER_POINTER (b); \
1890 MOVE_BUFFER_POINTER (begalt); \
1891 if (fixup_alt_jump) \
1892 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1893 if (laststart) \
1894 MOVE_BUFFER_POINTER (laststart); \
1895 if (pending_exact) \
1896 MOVE_BUFFER_POINTER (pending_exact); \
1897 } \
1898 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1899 } while (0)
1902 /* Since we have one byte reserved for the register number argument to
1903 {start,stop}_memory, the maximum number of groups we can report
1904 things about is what fits in that byte. */
1905 #define MAX_REGNUM 255
1907 /* But patterns can have more than `MAX_REGNUM' registers. We just
1908 ignore the excess. */
1912 /* Macros for the compile stack. */
1914 /* Since offsets can go either forwards or backwards, this type needs to
1915 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1916 /* int may be not enough when sizeof(int) == 2. */
1920 {
1921 pattern_offset_t begalt_offset;
1922 pattern_offset_t fixup_alt_jump;
1923 pattern_offset_t laststart_offset;
1924 regnum_t regnum;
1929 {
1936 #define INIT_COMPILE_STACK_SIZE 32
1938 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1939 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1941 /* The next available element. */
1942 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1944 /* Explicit quit checking is only used on NTemacs and whenever we
1945 use polling to process input events. */
1946 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1948 # define IMMEDIATE_QUIT_CHECK \
1949 do { \
1950 if (immediate_quit) QUIT; \
1951 } while (0)
1952 #else
1953 # define IMMEDIATE_QUIT_CHECK ((void)0)
1954 #endif
1955 \f
1956 /* Structure to manage work area for range table. */
1957 struct range_table_work_area
1958 {
1963 };
1965 /* Make sure that WORK_AREA can hold more N multibyte characters.
1966 This is used only in set_image_of_range and set_image_of_range_1.
1967 It expects WORK_AREA to be a pointer.
1968 If it can't get the space, it returns from the surrounding function. */
1970 #define EXTEND_RANGE_TABLE(work_area, n) \
1971 do { \
1972 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1973 { \
1974 extend_range_table_work_area (&work_area); \
1975 if ((work_area).table == 0) \
1976 return (REG_ESPACE); \
1977 } \
1978 } while (0)
1980 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1981 (work_area).bits |= (bit)
1983 /* Bits used to implement the multibyte-part of the various character classes
1984 such as [:alnum:] in a charset's range table. */
1985 #define BIT_WORD 0x1
1986 #define BIT_LOWER 0x2
1987 #define BIT_PUNCT 0x4
1988 #define BIT_SPACE 0x8
1989 #define BIT_UPPER 0x10
1990 #define BIT_MULTIBYTE 0x20
1992 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1993 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1994 do { \
1995 EXTEND_RANGE_TABLE ((work_area), 2); \
1996 (work_area).table[(work_area).used++] = (range_start); \
1997 (work_area).table[(work_area).used++] = (range_end); \
1998 } while (0)
2000 /* Free allocated memory for WORK_AREA. */
2001 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2002 do { \
2003 if ((work_area).table) \
2004 free ((work_area).table); \
2005 } while (0)
2007 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2008 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2009 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2010 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2011 \f
2013 /* Set the bit for character C in a list. */
2014 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2017 #ifdef emacs
2019 /* Store characters in the range FROM to TO in the bitmap at B (for
2020 ASCII and unibyte characters) and WORK_AREA (for multibyte
2021 characters) while translating them and paying attention to the
2022 continuity of translated characters.
2024 Implementation note: It is better to implement these fairly big
2025 macros by a function, but it's not that easy because macros called
2026 in this macro assume various local variables already declared. */
2028 /* Both FROM and TO are ASCII characters. */
2030 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2031 do { \
2032 int C0, C1; \
2033 \
2034 for (C0 = (FROM); C0 <= (TO); C0++) \
2035 { \
2036 C1 = TRANSLATE (C0); \
2037 if (! ASCII_CHAR_P (C1)) \
2038 { \
2039 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2040 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2041 C1 = C0; \
2042 } \
2043 SET_LIST_BIT (C1); \
2044 } \
2045 } while (0)
2048 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2050 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2051 do { \
2052 int C0, C1, C2, I; \
2053 int USED = RANGE_TABLE_WORK_USED (work_area); \
2054 \
2055 for (C0 = (FROM); C0 <= (TO); C0++) \
2056 { \
2057 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2058 if (CHAR_BYTE8_P (C1)) \
2059 SET_LIST_BIT (C0); \
2060 else \
2061 { \
2062 C2 = TRANSLATE (C1); \
2063 if (C2 == C1 \
2064 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2065 C1 = C0; \
2066 SET_LIST_BIT (C1); \
2067 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2068 { \
2069 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2070 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2071 \
2072 if (C2 >= from - 1 && C2 <= to + 1) \
2073 { \
2074 if (C2 == from - 1) \
2075 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2076 else if (C2 == to + 1) \
2077 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2078 break; \
2079 } \
2080 } \
2081 if (I < USED) \
2082 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2083 } \
2084 } \
2085 } while (0)
2088 /* Both FROM and TO are multibyte characters. */
2090 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2091 do { \
2092 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2093 \
2094 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2095 for (C0 = (FROM); C0 <= (TO); C0++) \
2096 { \
2097 C1 = TRANSLATE (C0); \
2098 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2099 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2100 SET_LIST_BIT (C2); \
2101 if (C1 >= (FROM) && C1 <= (TO)) \
2102 continue; \
2103 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2104 { \
2105 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2106 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2107 \
2108 if (C1 >= from - 1 && C1 <= to + 1) \
2109 { \
2110 if (C1 == from - 1) \
2111 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2112 else if (C1 == to + 1) \
2113 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2114 break; \
2115 } \
2116 } \
2117 if (I < USED) \
2118 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2119 } \
2120 } while (0)
2124 /* Get the next unsigned number in the uncompiled pattern. */
2125 #define GET_UNSIGNED_NUMBER(num) \
2126 do { \
2127 if (p == pend) \
2128 FREE_STACK_RETURN (REG_EBRACE); \
2129 else \
2130 { \
2131 PATFETCH (c); \
2133 { \
2134 int prev; \
2135 if (num < 0) \
2136 num = 0; \
2137 prev = num; \
2139 if (num / 10 != prev) \
2140 FREE_STACK_RETURN (REG_BADBR); \
2141 if (p == pend) \
2142 FREE_STACK_RETURN (REG_EBRACE); \
2143 PATFETCH (c); \
2144 } \
2145 } \
2146 } while (0)
2147 \f
2148 #if ! WIDE_CHAR_SUPPORT
2150 /* Map a string to the char class it names (if any). */
2151 re_wctype_t
2154 {
2174 }
2176 /* True if CH is in the char class CC. */
2177 boolean
2180 re_wctype_t cc;
2181 {
2183 {
2204 }
2205 }
2207 /* Return a bit-pattern to use in the range-table bits to match multibyte
2208 chars of class CC. */
2209 static int
2211 re_wctype_t cc;
2212 {
2214 {
2226 }
2227 }
2228 #endif
2229 \f
2230 /* Filling in the work area of a range. */
2232 /* Actually extend the space in WORK_AREA. */
2234 static void
2237 {
2240 work_area->table
2242 else
2243 work_area->table
2245 }
2247 #if 0
2248 #ifdef emacs
2250 /* Carefully find the ranges of codes that are equivalent
2251 under case conversion to the range start..end when passed through
2252 TRANSLATE. Handle the case where non-letters can come in between
2253 two upper-case letters (which happens in Latin-1).
2254 Also handle the case of groups of more than 2 case-equivalent chars.
2256 The basic method is to look at consecutive characters and see
2257 if they can form a run that can be handled as one.
2259 Returns -1 if successful, REG_ESPACE if ran out of space. */
2261 static int
2263 RE_TRANSLATE_TYPE translate;
2266 {
2267 /* `one_case' indicates a character, or a run of characters,
2268 each of which is an isolate (no case-equivalents).
2269 This includes all ASCII non-letters.
2271 `two_case' indicates a character, or a run of characters,
2272 each of which has two case-equivalent forms.
2273 This includes all ASCII letters.
2275 `strange' indicates a character that has more than one
2276 case-equivalent. */
2280 /* Describe the run that is in progress,
2281 which the next character can try to extend.
2282 If run_type is strange, that means there really is no run.
2283 If run_type is one_case, then run_start...run_end is the run.
2284 If run_type is two_case, then the run is run_start...run_end,
2285 and the case-equivalents end at run_eqv_end. */
2290 Lisp_Object eqv_table;
2293 {
2298 }
2303 {
2308 /* Classify this character */
2313 else
2318 else
2321 /* Can this character extend the run in progress? */
2326 {
2327 /* No, end the run.
2328 Record each of its equivalent ranges. */
2330 {
2334 }
2336 {
2344 }
2346 }
2349 {
2350 /* For a strange character, add each of its equivalents, one
2351 by one. Don't start a range. */
2352 do
2353 {
2358 }
2360 }
2362 /* Add this char to the run, or start a new run. */
2364 {
2365 /* Initialize a new range. */
2370 }
2371 else
2372 {
2373 /* Extend a running range. */
2376 }
2377 }
2379 /* If a run is still in progress at the end, finish it now
2380 by recording its equivalent ranges. */
2382 {
2386 }
2388 {
2396 }
2399 }
2403 /* Record the image of the range start..end when passed through
2404 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2405 and is not even necessarily contiguous.
2406 Normally we approximate it with the smallest contiguous range that contains
2407 all the chars we need. However, for Latin-1 we go to extra effort
2408 to do a better job.
2410 This function is not called for ASCII ranges.
2412 Returns -1 if successful, REG_ESPACE if ran out of space. */
2414 static int
2416 RE_TRANSLATE_TYPE translate;
2419 {
2422 #ifdef emacs
2423 /* For Latin-1 ranges, use set_image_of_range_1
2424 to get proper handling of ranges that include letters and nonletters.
2425 For a range that includes the whole of Latin-1, this is not necessary.
2426 For other character sets, we don't bother to get this right. */
2429 {
2442 }
2443 #endif
2452 {
2456 {
2459 {
2462 else
2463 {
2466 }
2467 }
2468 }
2471 {
2475 }
2476 }
2479 }
2481 \f
2482 #ifndef MATCH_MAY_ALLOCATE
2484 /* If we cannot allocate large objects within re_match_2_internal,
2485 we make the fail stack and register vectors global.
2486 The fail stack, we grow to the maximum size when a regexp
2487 is compiled.
2488 The register vectors, we adjust in size each time we
2489 compile a regexp, according to the number of registers it needs. */
2493 /* Size with which the following vectors are currently allocated.
2494 That is so we can make them bigger as needed,
2495 but never make them smaller. */
2501 /* Make the register vectors big enough for NUM_REGS registers,
2502 but don't make them smaller. */
2504 static
2507 {
2509 {
2516 }
2517 }
2520 \f
2522 compile_stack,
2523 regnum_t regnum));
2525 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2526 Returns one of error codes defined in `regex.h', or zero for success.
2528 Assumes the `allocated' (and perhaps `buffer') and `translate'
2529 fields are set in BUFP on entry.
2531 If it succeeds, results are put in BUFP (if it returns an error, the
2532 contents of BUFP are undefined):
2533 `buffer' is the compiled pattern;
2534 `syntax' is set to SYNTAX;
2535 `used' is set to the length of the compiled pattern;
2536 `fastmap_accurate' is zero;
2537 `re_nsub' is the number of subexpressions in PATTERN;
2538 `not_bol' and `not_eol' are zero;
2540 The `fastmap' field is neither examined nor set. */
2542 /* Insert the `jump' from the end of last alternative to "here".
2543 The space for the jump has already been allocated. */
2544 #define FIXUP_ALT_JUMP() \
2545 do { \
2546 if (fixup_alt_jump) \
2547 STORE_JUMP (jump, fixup_alt_jump, b); \
2548 } while (0)
2551 /* Return, freeing storage we allocated. */
2552 #define FREE_STACK_RETURN(value) \
2553 do { \
2554 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2555 free (compile_stack.stack); \
2556 return value; \
2557 } while (0)
2559 static reg_errcode_t
2563 reg_syntax_t syntax;
2565 {
2566 /* We fetch characters from PATTERN here. */
2569 /* A random temporary spot in PATTERN. */
2572 /* Points to the end of the buffer, where we should append. */
2575 /* Keeps track of unclosed groups. */
2576 compile_stack_type compile_stack;
2578 /* Points to the current (ending) position in the pattern. */
2579 #ifdef AIX
2580 /* `const' makes AIX compiler fail. */
2582 #else
2584 #endif
2587 /* How to translate the characters in the pattern. */
2590 /* Address of the count-byte of the most recently inserted `exactn'
2591 command. This makes it possible to tell if a new exact-match
2592 character can be added to that command or if the character requires
2593 a new `exactn' command. */
2596 /* Address of start of the most recently finished expression.
2597 This tells, e.g., postfix * where to find the start of its
2598 operand. Reset at the beginning of groups and alternatives. */
2601 /* Address of beginning of regexp, or inside of last group. */
2604 /* Place in the uncompiled pattern (i.e., the {) to
2605 which to go back if the interval is invalid. */
2608 /* Address of the place where a forward jump should go to the end of
2609 the containing expression. Each alternative of an `or' -- except the
2610 last -- ends with a forward jump of this sort. */
2613 /* Work area for range table of charset. */
2616 /* If the object matched can contain multibyte characters. */
2619 /* If a target of matching can contain multibyte characters. */
2622 /* Nonzero if we have pushed down into a subpattern. */
2625 /* These hold the values of p, pattern, and pend from the main
2626 pattern when we have pushed into a subpattern. */
2631 #ifdef DEBUG
2632 debug++;
2635 {
2641 }
2644 /* Initialize the compile stack. */
2655 /* Initialize the pattern buffer. */
2661 /* Set `used' to zero, so that if we return an error, the pattern
2662 printer (for debugging) will think there's no pattern. We reset it
2663 at the end. */
2666 /* Always count groups, whether or not bufp->no_sub is set. */
2669 #if !defined emacs && !defined SYNTAX_TABLE
2670 /* Initialize the syntax table. */
2672 #endif
2675 {
2678 enough space. This loses if buffer's address is bogus, but
2679 that is the user's responsibility. */
2681 }
2682 else
2685 }
2689 }
2693 /* Loop through the uncompiled pattern until we're at the end. */
2695 {
2697 {
2698 /* If this is the end of an included regexp,
2699 pop back to the main regexp and try again. */
2701 {
2707 }
2708 /* If this is the end of the main regexp, we are done. */
2710 }
2715 {
2717 {
2720 /* If there's no special whitespace regexp, treat
2721 spaces normally. And don't try to do this recursively. */
2725 /* Peek past following spaces. */
2727 {
2730 p1++;
2731 }
2732 /* If the spaces are followed by a repetition op,
2733 treat them normally. */
2739 /* Replace the spaces with the whitespace regexp. */
2747 }
2750 {
2753 /* If context independent, it's an operator. */
2755 /* Otherwise, depends on what's come before. */
2758 else
2760 }
2765 {
2767 p == pend
2768 /* If context independent, it's an operator. */
2770 /* Otherwise, depends on what's next. */
2773 else
2775 }
2784 handle_plus:
2786 /* If there is no previous pattern... */
2788 {
2793 }
2795 {
2796 /* 1 means zero (many) matches is allowed. */
2800 /* If there is a sequence of repetition chars, collapse it
2801 down to just one (the right one). We can't combine
2802 interval operators with these because of, e.g., `a{2}*',
2803 which should only match an even number of `a's. */
2806 {
2810 else
2811 {
2814 }
2821 ;
2823 {
2828 else
2830 }
2831 else
2833 /* If we get here, we found another repeat character. */
2835 }
2837 /* Star, etc. applied to an empty pattern is equivalent
2838 to an empty pattern. */
2842 /* Now we know whether or not zero matches is allowed
2843 and also whether or not two or more matches is allowed. */
2845 {
2847 {
2850 re_opcode_t ofj =
2851 /* Check if the loop can match the empty string. */
2858 on_failure_keep_string_jump, we turn simple P+
2859 into PP* if P is simple. */
2867 }
2871 /* A + loop. */
2873 else
2874 /* Simple * loops can use on_failure_keep_string_jump
2875 depending on what follows. But since we don't know
2876 that yet, we leave the decision up to
2877 on_failure_jump_smart. */
2883 }
2884 else
2885 {
2886 /* A simple ? pattern. */
2891 }
2892 }
2898 {
2901 /* The non-greedy multiple match looks like
2902 a repeat..until: we only need a conditional jump
2903 at the end of the loop. */
2909 {
2910 /* The repeat...until naturally matches one or more.
2911 To also match zero times, we need to first jump to
2912 the end of the loop (its conditional jump). */
2915 }
2916 }
2917 else
2918 {
2919 /* non-greedy a?? */
2924 }
2925 }
2926 }
2938 {
2943 /* Ensure that we have enough space to push a charset: the
2944 opcode, the length count, and the bitset; 34 bytes in all. */
2949 /* We test `*p == '^' twice, instead of using an if
2950 statement, so we only need one BUF_PUSH. */
2953 p++;
2955 /* Remember the first position in the bracket expression. */
2958 /* Push the number of bytes in the bitmap. */
2961 /* Clear the whole map. */
2964 /* charset_not matches newline according to a syntax bit. */
2969 /* Read in characters and ranges, setting map bits. */
2971 {
2978 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2979 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2980 So the translation is done later in a loop. Example:
2981 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2984 /* \ might escape characters inside [...] and [^...]. */
2986 {
2991 }
2992 else
2993 {
2994 /* Could be the end of the bracket expression. If it's
2995 not (i.e., when the bracket expression is `[]' so
2996 far), the ']' character bit gets set way below. */
2999 }
3001 /* See if we're at the beginning of a possible character
3002 class. */
3006 {
3007 /* Leave room for the null. */
3015 /* If pattern is `[[:'. */
3019 {
3025 else
3026 /* This is in any case an invalid class name. */
3028 }
3031 /* If isn't a word bracketed by `[:' and `:]':
3032 undo the ending character, the letters, and
3033 leave the leading `:' and `[' (but set bits for
3034 them). */
3036 {
3037 re_wctype_t cc;
3045 /* Throw away the ] at the end of the character
3046 class. */
3051 #ifndef emacs
3054 {
3058 }
3060 /* Most character classes in a multibyte match
3061 just set a flag. Exceptions are is_blank,
3062 is_digit, is_cntrl, and is_xdigit, since
3063 they can only match ASCII characters. We
3064 don't need to handle them for multibyte.
3065 They are distinguished by a negative wctype. */
3067 /* Setup the gl_state object to its buffer-defined
3068 value. This hardcodes the buffer-global
3069 syntax-table for ASCII chars, while the other chars
3070 will obey syntax-table properties. It's not ideal,
3071 but it's the way it's been done until now. */
3075 {
3079 {
3088 }
3089 }
3090 SET_RANGE_TABLE_WORK_AREA_BIT
3093 /* In most cases the matching rule for char classes
3094 only uses the syntax table for multibyte chars,
3095 so that the content of the syntax-table it is not
3096 hardcoded in the range_table. SPACE and WORD are
3097 the two exceptions. */
3101 /* Repeat the loop. */
3103 }
3104 else
3105 {
3106 /* Go back to right after the "[:". */
3110 /* Because the `:' may starts the range, we
3111 can't simply set bit and repeat the loop.
3112 Instead, just set it to C and handle below. */
3114 }
3115 }
3118 {
3120 /* Discard the `-'. */
3123 /* Fetch the character which ends the range. */
3125 #ifdef emacs
3128 /* Treat the range from a multibyte character to
3129 raw-byte character as empty. */
3132 }
3133 else
3134 /* Range from C to C. */
3138 {
3141 /* Else, repeat the loop. */
3142 }
3143 else
3144 {
3145 #ifndef emacs
3146 /* Set the range into bitmap */
3148 {
3152 }
3155 {
3161 }
3163 {
3165 {
3170 }
3172 {
3174 }
3175 else
3176 {
3178 }
3179 }
3181 }
3182 }
3184 /* Discard any (non)matching list bytes that are all 0 at the
3185 end of the map. Decrease the map-length byte too. */
3190 /* Build real range table from work area. */
3193 {
3197 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3198 bytes for flags, two for COUNT, and three bytes for
3199 each character. */
3202 /* Indicate the existence of range table. */
3205 /* Store the character class flag bits into the range table.
3206 If not in emacs, these flag bits are always 0. */
3212 STORE_CHARACTER_AND_INCR
3214 }
3215 }
3222 else
3229 else
3236 else
3243 else
3250 else
3257 /* Do not translate the character after the \, so that we can
3258 distinguish, e.g., \B from \b, even if we normally would
3259 translate, e.g., B to b. */
3263 {
3268 handle_open:
3269 {
3273 {
3274 /* Look for a special (?...) construct */
3276 {
3279 {
3282 {
3285 /* An explicitly specified regnum must start
3286 with non-0. */
3293 /* Only (?:...) is supported right now. */
3295 }
3296 }
3297 }
3298 }
3308 /* Ideally, we'd want to check that the specified
3309 group can't have matched (i.e. all subgroups
3310 using the same regnum are in other branches of
3311 OR patterns), but we don't currently keep track
3312 of enough info to do that easily. */
3315 }
3316 else
3317 /* It's really shy. */
3321 {
3323 compile_stack_elt_t);
3327 }
3329 /* These are the values to restore when we hit end of this
3330 group. They are all relative offsets, so that if the
3331 whole pattern moves because of realloc, they will still
3332 be valid. */
3334 COMPILE_STACK_TOP.fixup_alt_jump
3339 /* Do not push a start_memory for groups beyond the last one
3340 we can represent in the compiled pattern. */
3349 /* If we've reached MAX_REGNUM groups, then this open
3350 won't actually generate any code, so we'll have to
3351 clear pending_exact explicitly. */
3354 }
3360 {
3363 else
3365 }
3367 handle_close:
3370 /* See similar code for backslashed left paren above. */
3372 {
3375 else
3377 }
3379 /* Since we just checked for an empty stack above, this
3380 ``can't happen''. */
3382 {
3383 /* We don't just want to restore into `regnum', because
3384 later groups should continue to be numbered higher,
3385 as in `(ab)c(de)' -- the second group is #2. */
3386 regnum_t regnum;
3390 fixup_alt_jump
3396 /* If we've reached MAX_REGNUM groups, then this open
3397 won't actually generate any code, so we'll have to
3398 clear pending_exact explicitly. */
3401 /* We're at the end of the group, so now we know how many
3402 groups were inside this one. */
3405 }
3412 handle_alt:
3416 /* Insert before the previous alternative a jump which
3417 jumps to this alternative if the former fails. */
3423 /* The alternative before this one has a jump after it
3424 which gets executed if it gets matched. Adjust that
3425 jump so it will jump to this alternative's analogous
3426 jump (put in below, which in turn will jump to the next
3427 (if any) alternative's such jump, etc.). The last such
3428 jump jumps to the correct final destination. A picture:
3429 _____ _____
3430 | | | |
3431 | v | v
3432 a | b | c
3434 If we are at `b', then fixup_alt_jump right now points to a
3435 three-byte space after `a'. We'll put in the jump, set
3436 fixup_alt_jump to right after `b', and leave behind three
3437 bytes which we'll fill in when we get to after `c'. */
3441 /* Mark and leave space for a jump after this alternative,
3442 to be filled in later either by next alternative or
3443 when know we're at the end of a series of alternatives. */
3454 /* If \{ is a literal. */
3456 /* If we're at `\{' and it's not the open-interval
3457 operator. */
3461 handle_interval:
3462 {
3463 /* If got here, then the syntax allows intervals. */
3465 /* At least (most) this many matches must be made. */
3474 else
3475 /* Interval such as `{1}' => match exactly once. */
3483 {
3489 }
3494 /* We just parsed a valid interval. */
3496 /* If it's invalid to have no preceding re. */
3498 {
3503 else
3505 }
3508 /* If the upper bound is zero, just drop the sub pattern
3509 altogether. */
3512 /* Just match it once: nothing to do here. */
3513 ;
3515 /* Otherwise, we have a nontrivial interval. When
3516 we're all done, the pattern will look like:
3517 set_number_at <jump count> <upper bound>
3518 set_number_at <succeed_n count> <lower bound>
3519 succeed_n <after jump addr> <succeed_n count>
3520 <body of loop>
3521 jump_n <succeed_n addr> <jump count>
3522 (The upper bound and `jump_n' are omitted if
3523 `upper_bound' is 1, though.) */
3524 else
3526 more at the end of the loop. */
3534 {
3535 /* A succeed_n that starts with 0 is really a
3536 a simple on_failure_jump_loop. */
3540 }
3541 else
3542 {
3543 /* Initialize lower bound of the `succeed_n', even
3544 though it will be set during matching by its
3545 attendant `set_number_at' (inserted next),
3546 because `re_compile_fastmap' needs to know.
3547 Jump to the `jump_n' we might insert below. */
3550 lower_bound);
3553 /* Code to initialize the lower bound. Insert
3554 before the `succeed_n'. The `5' is the last two
3555 bytes of this `set_number_at', plus 3 bytes of
3556 the following `succeed_n'. */
3560 }
3563 {
3564 /* A negative upper bound stands for infinity,
3565 in which case it degenerates to a plain jump. */
3568 }
3571 append a backward jump to the `succeed_n'
3572 that starts this interval.
3574 When we've reached this during matching,
3575 we'll have matched the interval once, so
3576 jump back only `upper_bound - 1' times. */
3581 /* The location we want to set is the second
3582 parameter of the `jump_n'; that is `b-2' as
3583 an absolute address. `laststart' will be
3584 the `set_number_at' we're about to insert;
3585 `laststart+3' the number to set, the source
3586 for the relative address. But we are
3587 inserting into the middle of the pattern --
3588 so everything is getting moved up by 5.
3589 Conclusion: (b - 2) - (laststart + 3) + 5,
3590 i.e., b - laststart.
3592 We insert this at the beginning of the loop
3593 so that if we fail during matching, we'll
3594 reinitialize the bounds. */
3598 }
3599 }
3602 }
3605 unfetch_interval:
3606 /* If an invalid interval, match the characters as literals. */
3611 /* normal_char and normal_backslash need `c'. */
3615 {
3618 }
3619 else
3622 #ifdef emacs
3623 /* There is no way to specify the before_dot and after_dot
3624 operators. rms says this is ok. --karl */
3692 else
3722 {
3723 regnum_t reg;
3731 /* Can't back reference to a subexp before its end. */
3737 }
3745 else
3749 normal_backslash:
3750 /* You might think it would be useful for \ to mean
3751 not to translate; but if we don't translate it
3752 it will never match anything. */
3754 }
3759 /* Expects the character in `c'. */
3760 normal_char:
3761 /* If no exactn currently being built. */
3764 /* If last exactn not at current position. */
3767 /* We have only one byte following the exactn for the count. */
3770 /* If followed by a repetition operator. */
3779 {
3780 /* Start building a new exactn. */
3786 }
3789 {
3793 {
3797 }
3798 else
3799 {
3802 {
3807 }
3810 }
3812 }
3819 /* Through the pattern now. */
3826 /* If we don't want backtracking, force success
3827 the first time we reach the end of the compiled pattern. */
3831 /* We have succeeded; set the length of the buffer. */
3834 #ifdef DEBUG
3836 {
3840 }
3841 debug--;
3844 #ifndef MATCH_MAY_ALLOCATE
3845 /* Initialize the failure stack to the largest possible stack. This
3846 isn't necessary unless we're trying to avoid calling alloca in
3847 the search and match routines. */
3848 {
3852 {
3856 fail_stack.stack
3859 else
3860 fail_stack.stack
3864 }
3867 }
3872 \f
3873 /* Subroutines for `regex_compile'. */
3875 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3877 static void
3879 re_opcode_t op;
3882 {
3885 }
3888 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3890 static void
3892 re_opcode_t op;
3895 {
3899 }
3902 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3903 for OP followed by two-byte integer parameter ARG. */
3905 static void
3907 re_opcode_t op;
3911 {
3919 }
3922 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3924 static void
3926 re_opcode_t op;
3930 {
3938 }
3941 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3942 after an alternative or a begin-subexpression. We assume there is at
3943 least one character before the ^. */
3945 static boolean
3948 reg_syntax_t syntax;
3949 {
3953 return
3954 /* After a subexpression? */
3956 /* After an alternative? */
3958 /* After a shy subexpression? */
3963 }
3966 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3967 at least one character after the $, i.e., `P < PEND'. */
3969 static boolean
3972 reg_syntax_t syntax;
3973 {
3978 return
3979 /* Before a subexpression? */
3982 /* Before an alternative? */
3985 }
3988 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3989 false if it's not. */
3991 static boolean
3993 compile_stack_type compile_stack;
3994 regnum_t regnum;
3995 {
4000 this_element--)
4005 }
4006 \f
4007 /* analyse_first.
4008 If fastmap is non-NULL, go through the pattern and fill fastmap
4009 with all the possible leading chars. If fastmap is NULL, don't
4010 bother filling it up (obviously) and only return whether the
4011 pattern could potentially match the empty string.
4013 Return 1 if p..pend might match the empty string.
4014 Return 0 if p..pend matches at least one char.
4015 Return -1 if fastmap was not updated accurately. */
4017 static int
4022 {
4026 /* If all elements for base leading-codes in fastmap is set, this
4027 flag is set true. */
4032 /* The loop below works as follows:
4033 - It has a working-list kept in the PATTERN_STACK and which basically
4034 starts by only containing a pointer to the first operation.
4035 - If the opcode we're looking at is a match against some set of
4036 chars, then we add those chars to the fastmap and go on to the
4037 next work element from the worklist (done via `break').
4038 - If the opcode is a control operator on the other hand, we either
4039 ignore it (if it's meaningless at this point, such as `start_memory')
4040 or execute it (if it's a jump). If the jump has several destinations
4041 (i.e. `on_failure_jump'), then we push the other destination onto the
4042 worklist.
4043 We guarantee termination by ignoring backward jumps (more or less),
4044 so that `p' is monotonically increasing. More to the point, we
4045 never set `p' (or push) anything `<= p1'. */
4048 {
4049 /* `p1' is used as a marker of how far back a `on_failure_jump'
4050 can go without being ignored. It is normally equal to `p'
4051 (which prevents any backward `on_failure_jump') except right
4052 after a plain `jump', to allow patterns such as:
4053 0: jump 10
4054 3..9: <body>
4055 10: on_failure_jump 3
4056 as used for the *? operator. */
4060 {
4066 /* If the first character has to match a backreference, that means
4067 that the group was empty (since it already matched). Since this
4068 is the only case that interests us here, we can assume that the
4069 backreference must match the empty string. */
4070 p++;
4074 /* Following are the cases which match a character. These end
4075 with `break'. */
4079 {
4080 /* If multibyte is nonzero, the first byte of each
4081 character is an ASCII or a leading code. Otherwise,
4082 each byte is a character. Thus, this works in both
4083 cases. */
4086 {
4087 /* For the case of matching this unibyte regex
4088 against multibyte, we must set a leading code of
4089 the corresponding multibyte character. */
4093 }
4094 }
4099 /* We could put all the chars except for \n (and maybe \0)
4100 but we don't bother since it is generally not worth it. */
4107 {
4108 /* Chars beyond end of bitmap are possible matches. */
4112 }
4114 /* Fallthrough */
4123 #ifdef emacs
4125 which doesn't match the specified set of characters. */
4126 not
4127 ||
4128 /* If we can match a character class, we can match any
4129 multibyte characters. */
4133 {
4135 {
4140 }
4141 }
4145 {
4146 /* Set fastmap[I] to 1 where I is a leading code of each
4147 multibyte characer in the range table. */
4151 /* Make P points the range table. `+ 2' is to skip flag
4152 bits for a character class. */
4155 /* Extract the number of ranges in range table into COUNT. */
4158 {
4159 /* Extract the start and end of each range. */
4167 }
4168 }
4169 #endif
4175 #ifndef emacs
4183 /* This match depends on text properties. These end with
4184 aborting optimizations. */
4196 /* Any leading code can possibly start a character which
4197 has or doesn't has the specified category. */
4199 {
4204 }
4207 /* All cases after this match the empty string. These end with
4208 `continue'. */
4231 /* Backward jumps can only go back to code that we've already
4232 visited. `re_compile' should make sure this is true. */
4236 {
4242 p++;
4246 };
4247 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4248 to jump back to "just after here". */
4249 /* Fallthrough */
4259 else
4261 We first go down the "straight" path so as to minimize
4262 stack usage when going through alternatives. */
4266 }
4271 /* This code simply does not properly handle forward jump_n. */
4274 /* jump_n can either jump or fall through. The (backward) jump
4275 case has already been handled, so we only need to look at the
4276 fallthrough case. */
4280 /* If N == 0, it should be an on_failure_jump_loop instead. */
4283 /* We only care about one iteration of the loop, so we don't
4284 need to consider the case where this behaves like an
4285 on_failure_jump. */
4304 /* Getting here means we have found the possible starting
4305 characters for one path of the pattern -- and that the empty
4306 string does not match. We need not follow this path further. */
4310 /* We reached the end without matching anything. */
4314 \f
4315 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4316 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4317 characters can start a string that matches the pattern. This fastmap
4318 is used by re_search to skip quickly over impossible starting points.
4320 Character codes above (1 << BYTEWIDTH) are not represented in the
4321 fastmap, but the leading codes are represented. Thus, the fastmap
4322 indicates which character sets could start a match.
4324 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4325 area as BUFP->fastmap.
4327 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4328 the pattern buffer.
4330 Returns 0 if we succeed, -2 if an internal error. */
4332 int
4335 {
4349 \f
4350 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4351 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4352 this memory for recording register information. STARTS and ENDS
4353 must be allocated using the malloc library routine, and must each
4354 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4356 If NUM_REGS == 0, then subsequent matches should allocate their own
4357 register data.
4359 Unless this function is called, the first search or match using
4360 PATTERN_BUFFER will allocate its own register data, without
4361 freeing the old data. */
4363 void
4369 {
4371 {
4376 }
4377 else
4378 {
4382 }
4383 }
4385 \f
4386 /* Searching routines. */
4388 /* Like re_search_2, below, but only one string is specified, and
4389 doesn't let you say where to stop matching. */
4391 int
4397 {
4400 }
4403 /* Head address of virtual concatenation of string. */
4404 #define HEAD_ADDR_VSTRING(P) \
4405 (((P) >= size1 ? string2 : string1))
4407 /* End address of virtual concatenation of string. */
4408 #define STOP_ADDR_VSTRING(P) \
4409 (((P) >= size1 ? string2 + size2 : string1 + size1))
4411 /* Address of POS in the concatenation of virtual string. */
4412 #define POS_ADDR_VSTRING(POS) \
4413 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4415 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4416 virtual concatenation of STRING1 and STRING2, starting first at index
4417 STARTPOS, then at STARTPOS + 1, and so on.
4419 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4421 RANGE is how far to scan while trying to match. RANGE = 0 means try
4422 only at STARTPOS; in general, the last start tried is STARTPOS +
4423 RANGE.
4425 In REGS, return the indices of the virtual concatenation of STRING1
4426 and STRING2 that matched the entire BUFP->buffer and its contained
4427 subexpressions.
4429 Do not consider matching one past the index STOP in the virtual
4430 concatenation of STRING1 and STRING2.
4432 We return either the position in the strings at which the match was
4433 found, -1 if no match, or -2 if error (such as failure
4434 stack overflow). */
4436 int
4445 {
4453 boolean anchored_start;
4454 /* Nonzero if we are searching multibyte string. */
4457 /* Check for out-of-range STARTPOS. */
4461 /* Fix up RANGE if it might eventually take us outside
4462 the virtual concatenation of STRING1 and STRING2.
4463 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4469 /* If the search isn't to be a backwards one, don't waste time in a
4470 search for a pattern anchored at beginning of buffer. */
4472 {
4475 else
4477 }
4479 #ifdef emacs
4480 /* In a forward search for something that starts with \=.
4481 don't keep searching past point. */
4483 {
4487 }
4490 /* Update the fastmap now if not correct already. */
4494 /* See whether the pattern is anchored. */
4497 #ifdef emacs
4499 {
4503 }
4504 #endif
4506 /* Loop through the string, looking for a place to start matching. */
4508 {
4509 /* If the pattern is anchored,
4510 skip quickly past places we cannot match.
4511 We don't bother to treat startpos == 0 specially
4512 because that case doesn't repeat. */
4514 {
4519 }
4521 /* If a fastmap is supplied, skip quickly over characters that
4522 cannot be the start of a match. If the pattern can match the
4523 null string, however, we don't need to skip characters; we want
4524 the first null string. */
4526 {
4533 {
4540 /* Written out as an if-else to avoid testing `translate'
4541 inside the loop. */
4543 {
4546 {
4556 }
4557 else
4559 {
4570 d++;
4571 range--;
4572 }
4573 }
4574 else
4575 {
4578 {
4586 }
4587 else
4589 {
4590 d++;
4591 range--;
4592 }
4593 }
4595 }
4597 {
4599 {
4604 }
4605 else
4606 {
4617 }
4618 }
4619 }
4621 /* If can't match the null string, and that's all we have left, fail. */
4635 advance:
4639 {
4640 /* Update STARTPOS to the next character boundary. */
4642 {
4651 }
4652 else
4653 {
4654 range--;
4655 startpos++;
4656 }
4657 }
4658 else
4659 {
4660 range++;
4661 startpos--;
4663 /* Update STARTPOS to the previous character boundary. */
4665 {
4670 /* Find the head of multibyte form. */
4677 }
4678 }
4679 }
4683 \f
4684 /* Declarations and macros for re_match_2. */
4688 RE_TRANSLATE_TYPE translate,
4691 /* This converts PTR, a pointer into one of the search strings `string1'
4692 and `string2' into an offset from the beginning of that string. */
4693 #define POINTER_TO_OFFSET(ptr) \
4694 (FIRST_STRING_P (ptr) \
4695 ? ((regoff_t) ((ptr) - string1)) \
4696 : ((regoff_t) ((ptr) - string2 + size1)))
4698 /* Call before fetching a character with *d. This switches over to
4699 string2 if necessary.
4700 Check re_match_2_internal for a discussion of why end_match_2 might
4701 not be within string2 (but be equal to end_match_1 instead). */
4702 #define PREFETCH() \
4703 while (d == dend) \
4704 { \
4706 if (dend == end_match_2) \
4707 goto fail; \
4709 d = string2; \
4710 dend = end_match_2; \
4711 }
4713 /* Call before fetching a char with *d if you already checked other limits.
4714 This is meant for use in lookahead operations like wordend, etc..
4715 where we might need to look at parts of the string that might be
4716 outside of the LIMITs (i.e past `stop'). */
4717 #define PREFETCH_NOLIMIT() \
4718 if (d == end1) \
4719 { \
4720 d = string2; \
4721 dend = end_match_2; \
4722 } \
4724 /* Test if at very beginning or at very end of the virtual concatenation
4725 of `string1' and `string2'. If only one string, it's `string2'. */
4726 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4727 #define AT_STRINGS_END(d) ((d) == end2)
4730 /* Test if D points to a character which is word-constituent. We have
4731 two special cases to check for: if past the end of string1, look at
4732 the first character in string2; and if before the beginning of
4733 string2, look at the last character in string1. */
4734 #define WORDCHAR_P(d) \
4735 (SYNTAX ((d) == end1 ? *string2 \
4736 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4737 == Sword)
4739 /* Disabled due to a compiler bug -- see comment at case wordbound */
4741 /* The comment at case wordbound is following one, but we don't use
4742 AT_WORD_BOUNDARY anymore to support multibyte form.
4744 The DEC Alpha C compiler 3.x generates incorrect code for the
4745 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4746 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4747 macro and introducing temporary variables works around the bug. */
4749 #if 0
4750 /* Test if the character before D and the one at D differ with respect
4751 to being word-constituent. */
4752 #define AT_WORD_BOUNDARY(d) \
4753 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4754 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4755 #endif
4757 /* Free everything we malloc. */
4758 #ifdef MATCH_MAY_ALLOCATE
4759 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4760 # define FREE_VARIABLES() \
4761 do { \
4762 REGEX_FREE_STACK (fail_stack.stack); \
4763 FREE_VAR (regstart); \
4764 FREE_VAR (regend); \
4765 FREE_VAR (best_regstart); \
4766 FREE_VAR (best_regend); \
4767 } while (0)
4768 #else
4772 \f
4773 /* Optimization routines. */
4775 /* If the operation is a match against one or more chars,
4776 return a pointer to the next operation, else return NULL. */
4780 {
4782 {
4793 {
4798 }
4799 else
4805 #ifdef emacs
4809 p++;
4814 }
4816 }
4819 /* Jump over non-matching operations. */
4823 {
4826 {
4828 {
4841 }
4842 }
4845 }
4847 /* Non-zero if "p1 matches something" implies "p2 fails". */
4848 static int
4852 {
4853 re_opcode_t op2;
4860 /* Skip over open/close-group commands.
4861 If what follows this loop is a ...+ construct,
4862 look at what begins its body, since we will have to
4863 match at least one of that. */
4865 /* The same skip can be done for p1, except that this function
4866 is only used in the case where p1 is a simple match operator. */
4867 /* p1 = skip_noops (p1, pend); */
4875 {
4878 /* If we're at the end of the pattern, we can change. */
4880 {
4883 }
4888 {
4889 register re_wchar_t c
4894 {
4896 {
4899 }
4900 }
4904 {
4907 /* Test if C is listed in charset (or charset_not)
4908 at `p1'. */
4910 {
4914 }
4918 /* `not' is equal to 1 if c would match, which means
4919 that we can't change to pop_failure_jump. */
4921 {
4924 }
4925 }
4928 {
4931 }
4932 }
4936 {
4938 /* Reuse the code above. */
4941 /* It is hard to list up all the character in charset
4942 P2 if it includes multibyte character. Give up in
4943 such case. */
4945 {
4946 /* Now, we are sure that P2 has no range table.
4947 So, for the size of bitmap in P2, `p2[1]' is
4948 enough. But P1 may have range table, so the
4949 size of bitmap table of P1 is extracted by
4950 using macro `CHARSET_BITMAP_SIZE'.
4952 In a multibyte case, we know that all the character
4953 listed in P2 is ASCII. In a unibyte case, P1 has only a
4954 bitmap table. So, in both cases, it is enough to test
4955 only the bitmap table of P1. */
4958 {
4960 /* We win if the charset inside the loop
4961 has no overlap with the one after the loop. */
4965 idx++)
4971 {
4974 }
4975 }
4977 {
4979 /* We win if the charset_not inside the loop lists
4980 every character listed in the charset after. */
4988 {
4991 }
4992 }
4993 }
4994 }
4999 {
5002 /* Reuse the code above. */
5005 /* When we have two charset_not, it's very unlikely that
5006 they don't overlap. The union of the two sets of excluded
5007 chars should cover all possible chars, which, as a matter of
5008 fact, is virtually impossible in multibyte buffers. */
5010 }
5034 #ifdef emacs
5042 ;
5043 }
5045 /* Safe default. */
5047 }
5049 \f
5050 /* Matching routines. */
5053 /* re_match is like re_match_2 except it takes only a single string. */
5055 int
5061 {
5065 }
5069 #ifdef emacs
5070 /* In Emacs, this is the string or buffer in which we
5071 are matching. It is used for looking up syntax properties. */
5072 Lisp_Object re_match_object;
5073 #endif
5075 /* re_match_2 matches the compiled pattern in BUFP against the
5076 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5077 and SIZE2, respectively). We start matching at POS, and stop
5078 matching at STOP.
5080 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5081 store offsets for the substring each group matched in REGS. See the
5082 documentation for exactly how many groups we fill.
5084 We return -1 if no match, -2 if an internal error (such as the
5085 failure stack overflowing). Otherwise, we return the length of the
5086 matched substring. */
5088 int
5096 {
5099 #ifdef emacs
5104 #endif
5110 }
5114 /* This is a separate function so that we can force an alloca cleanup
5115 afterwards. */
5116 static int
5124 {
5125 /* General temporaries. */
5130 /* Just past the end of the corresponding string. */
5133 /* Pointers into string1 and string2, just past the last characters in
5134 each to consider matching. */
5137 /* Where we are in the data, and the end of the current string. */
5140 /* Used sometimes to remember where we were before starting matching
5141 an operator so that we can go back in case of failure. This "atomic"
5142 behavior of matching opcodes is indispensable to the correctness
5143 of the on_failure_keep_string_jump optimization. */
5146 /* Where we are in the pattern, and the end of the pattern. */
5150 /* We use this to map every character in the string. */
5153 /* Nonzero if BUFP is setup from a multibyte regex. */
5156 /* Nonzero if STRING1/STRING2 are multibyte. */
5159 /* Failure point stack. Each place that can handle a failure further
5160 down the line pushes a failure point on this stack. It consists of
5161 regstart, and regend for all registers corresponding to
5162 the subexpressions we're currently inside, plus the number of such
5163 registers, and, finally, two char *'s. The first char * is where
5164 to resume scanning the pattern; the second one is where to resume
5165 scanning the strings. */
5167 fail_stack_type fail_stack;
5168 #endif
5169 #ifdef DEBUG
5171 #endif
5173 #if defined REL_ALLOC && defined REGEX_MALLOC
5174 /* This holds the pointer to the failure stack, when
5175 it is allocated relocatably. */
5177 #endif
5179 /* We fill all the registers internally, independent of what we
5180 return, for use in backreferences. The number here includes
5181 an element for register zero. */
5184 /* Information on the contents of registers. These are pointers into
5185 the input strings; they record just what was matched (on this
5186 attempt) by a subexpression part of the pattern, that is, the
5187 regnum-th regstart pointer points to where in the pattern we began
5188 matching and the regnum-th regend points to right after where we
5189 stopped matching the regnum-th subexpression. (The zeroth register
5190 keeps track of what the whole pattern matches.) */
5193 #endif
5195 /* The following record the register info as found in the above
5196 variables when we find a match better than any we've seen before.
5197 This happens as we backtrack through the failure points, which in
5198 turn happens only if we have not yet matched the entire string. */
5202 #endif
5204 /* Logically, this is `best_regend[0]'. But we don't want to have to
5205 allocate space for that if we're not allocating space for anything
5206 else (see below). Also, we never need info about register 0 for
5207 any of the other register vectors, and it seems rather a kludge to
5208 treat `best_regend' differently than the rest. So we keep track of
5209 the end of the best match so far in a separate variable. We
5210 initialize this to NULL so that when we backtrack the first time
5211 and need to test it, it's not garbage. */
5214 #ifdef DEBUG
5215 /* Counts the total number of registers pushed. */
5217 #endif
5223 #ifdef MATCH_MAY_ALLOCATE
5224 /* Do not bother to initialize all the register variables if there are
5225 no groups in the pattern, as it takes a fair amount of time. If
5226 there are groups, we include space for register 0 (the whole
5227 pattern), even though we never use it, since it simplifies the
5228 array indexing. We should fix this. */
5230 {
5237 {
5240 }
5241 }
5242 else
5243 {
5244 /* We must initialize all our variables to NULL, so that
5245 `FREE_VARIABLES' doesn't try to free them. */
5247 }
5250 /* The starting position is bogus. */
5252 {
5255 }
5257 /* Initialize subexpression text positions to -1 to mark ones that no
5258 start_memory/stop_memory has been seen for. Also initialize the
5259 register information struct. */
5263 /* We move `string1' into `string2' if the latter's empty -- but not if
5264 `string1' is null. */
5266 {
5271 }
5275 /* `p' scans through the pattern as `d' scans through the data.
5276 `dend' is the end of the input string that `d' points within. `d'
5277 is advanced into the following input string whenever necessary, but
5278 this happens before fetching; therefore, at the beginning of the
5279 loop, `d' can be pointing at the end of a string, but it cannot
5280 equal `string2'. */
5282 {
5283 /* Only match within string2. */
5287 }
5288 else
5289 {
5291 {
5292 /* Only match within string1. */
5294 /* BEWARE!
5295 When we reach end_match_1, PREFETCH normally switches to string2.
5296 But in the present case, this means that just doing a PREFETCH
5297 makes us jump from `stop' to `gap' within the string.
5298 What we really want here is for the search to stop as
5299 soon as we hit end_match_1. That's why we set end_match_2
5300 to end_match_1 (since PREFETCH fails as soon as we hit
5301 end_match_2). */
5303 }
5304 else
5306 moving `d' from end1 to string2 will not prevent the d == dend
5307 check from catching the end of string. */
5310 }
5313 }
5321 /* This loops over pattern commands. It exits by returning from the
5322 function if the match is complete, or it drops through if the match
5323 fails at this starting point in the input data. */
5325 {
5332 /* If we haven't matched the entire string, and we want the
5333 longest match, try backtracking. */
5335 {
5336 /* 1 if this match ends in the same string (string1 or string2)
5337 as the best previous match. */
5340 /* 1 if this match is the best seen so far. */
5341 boolean best_match_p;
5343 /* AIX compiler got confused when this was combined
5344 with the previous declaration. */
5347 else
5355 /* If exceeds best match so far, save it. */
5357 {
5364 {
5367 }
5368 }
5370 }
5372 /* If no failure points, don't restore garbage. And if
5373 last match is real best match, don't restore second
5374 best one. */
5376 {
5377 restore_best_regs:
5378 /* Restore best match. It may happen that `dend ==
5379 end_match_1' while the restored d is in string2.
5380 For example, the pattern `x.*y.*z' against the
5381 strings `x-' and `y-z-', if the two strings are
5382 not consecutive in memory. */
5390 {
5393 }
5394 }
5397 succeed_label:
5400 /* If caller wants register contents data back, do it. */
5402 {
5403 /* Have the register data arrays been allocated? */
5406 extra element beyond `num_regs' for the `-1' marker
5407 GNU code uses. */
5412 {
5415 }
5417 }
5420 allocated, reallocate them. If we need fewer, just
5421 leave it alone. */
5423 {
5428 {
5431 }
5432 }
5433 }
5434 else
5435 {
5436 /* These braces fend off a "empty body in an else-statement"
5437 warning under GCC when assert expands to nothing. */
5439 }
5441 /* Convert the pointer data in `regstart' and `regend' to
5442 indices. Register zero has to be set differently,
5443 since we haven't kept track of any info for it. */
5445 {
5448 }
5450 /* Go through the first `min (num_regs, regs->num_regs)'
5451 registers, since that is all we initialized. */
5453 {
5456 else
5457 {
5462 }
5463 }
5465 /* If the regs structure we return has more elements than
5466 were in the pattern, set the extra elements to -1. If
5467 we (re)allocated the registers, this is the case,
5468 because we always allocate enough to have at least one
5469 -1 at the end. */
5485 }
5487 /* Otherwise match next pattern command. */
5489 {
5490 /* Ignore these. Used to ignore the n of succeed_n's which
5491 currently have n == 0. */
5500 /* Match the next n pattern characters exactly. The following
5501 byte in the pattern defines n, and the n bytes after that
5502 are the characters to match. */
5507 /* Remember the start point to rollback upon failure. */
5510 #ifndef emacs
5511 /* This is written out as an if-else so we don't waste time
5512 testing `translate' inside the loop. */
5514 do
5515 {
5518 {
5521 }
5522 d++;
5523 }
5525 else
5526 do
5527 {
5530 {
5533 }
5534 }
5537 /* The cost of testing `translate' is comparatively small. */
5539 do
5540 {
5547 else
5548 {
5551 }
5555 {
5558 }
5563 }
5565 else
5566 do
5567 {
5573 {
5576 }
5577 else
5578 {
5581 }
5584 {
5589 }
5590 else
5593 {
5596 }
5598 d++;
5599 }
5601 #endif
5605 /* Match any character except possibly a newline or a null. */
5607 {
5609 re_wchar_t buf_ch;
5615 target_multibyte);
5626 }
5632 {
5637 /* Start of actual range_table, or end of bitmap if there is no
5638 range table. */
5641 /* Nonzero if there is a range table. */
5644 /* Number of ranges of range table. This is not included
5645 in the initial byte-length of the command. */
5648 /* Whether matching against a unibyte character. */
5656 {
5659 }
5664 {
5670 {
5673 }
5674 }
5675 else
5676 {
5680 {
5684 {
5687 }
5688 }
5689 else
5691 }
5695 /* Cast to `unsigned' instead of `unsigned char' in
5696 case the bit list is a full 32 bytes long. */
5700 }
5701 #ifdef emacs
5703 {
5713 else
5715 }
5720 else
5727 }
5730 /* The beginning of a group is represented by start_memory.
5731 The argument is the register number. The text
5732 matched within the group is recorded (in the internal
5733 registers data structure) under the register number. */
5737 /* In case we need to undo this operation (via backtracking). */
5744 /* Move past the register number and inner group count. */
5749 /* The stop_memory opcode represents the end of a group. Its
5750 argument is the same as start_memory's: the register number. */
5755 /* Strictly speaking, there should be code such as:
5757 assert (REG_UNSET (regend[*p]));
5758 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5760 But the only info to be pushed is regend[*p] and it is known to
5761 be UNSET, so there really isn't anything to push.
5762 Not pushing anything, on the other hand deprives us from the
5763 guarantee that regend[*p] is UNSET since undoing this operation
5764 will not reset its value properly. This is not important since
5765 the value will only be read on the next start_memory or at
5766 the very end and both events can only happen if this stop_memory
5767 is *not* undone. */
5772 /* Move past the register number and the inner group count. */
5777 /* \<digit> has been turned into a `duplicate' command which is
5778 followed by the numeric value of <digit> as the register number. */
5780 {
5785 /* Can't back reference a group which we've never matched. */
5789 /* Where in input to try to start matching. */
5792 /* Remember the start point to rollback upon failure. */
5795 /* Where to stop matching; if both the place to start and
5796 the place to stop matching are in the same string, then
5797 set to the place to stop, otherwise, for now have to use
5798 the end of the first string. */
5804 {
5805 /* If necessary, advance to next segment in register
5806 contents. */
5808 {
5812 /* End of string1 => advance to string2. */
5815 }
5816 /* At end of register contents => success */
5819 /* If necessary, advance to next segment in data. */
5822 /* How many characters left in this segment to match. */
5825 /* Want how many consecutive characters we can match in
5826 one shot, so, if necessary, adjust the count. */
5830 /* Compare that many; failure if mismatch, else move
5831 past them. */
5835 {
5838 }
5840 }
5841 }
5845 /* begline matches the empty string at the beginning of the string
5846 (unless `not_bol' is set in `bufp'), and after newlines. */
5851 {
5853 }
5854 else
5855 {
5860 }
5861 /* In all other cases, we fail. */
5865 /* endline is the dual of begline. */
5870 {
5872 }
5873 else
5874 {
5878 }
5882 /* Match at the very beginning of the data. */
5890 /* Match at the very end of the data. */
5898 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5899 pushes NULL as the value for the string on the stack. Then
5900 `POP_FAILURE_POINT' will keep the current value for the
5901 string, instead of restoring it. To see why, consider
5902 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5903 then the . fails against the \n. But the next thing we want
5904 to do is match the \n against the \n; if we restored the
5905 string value, we would be back at the foo.
5907 Because this is used only in specific cases, we don't need to
5908 check all the things that `on_failure_jump' does, to make
5909 sure the right things get saved on the stack. Hence we don't
5910 share its code. The only reason to push anything on the
5911 stack at all is that otherwise we would have to change
5912 `anychar's code to do something besides goto fail in this
5913 case; that seems worse than this. */
5922 /* A nasty loop is introduced by the non-greedy *? and +?.
5923 With such loops, the stack only ever contains one failure point
5924 at a time, so that a plain on_failure_jump_loop kind of
5925 cycle detection cannot work. Worse yet, such a detection
5926 can not only fail to detect a cycle, but it can also wrongly
5927 detect a cycle (between different instantiations of the same
5928 loop).
5929 So the method used for those nasty loops is a little different:
5930 We use a special cycle-detection-stack-frame which is pushed
5931 when the on_failure_jump_nastyloop failure-point is *popped*.
5932 This special frame thus marks the beginning of one iteration
5933 through the loop and we can hence easily check right here
5934 whether something matched between the beginning and the end of
5935 the loop. */
5942 {
5946 /* If there's a cycle, just continue without pushing
5947 this failure point. The failure point is the "try again"
5948 option, which shouldn't be tried.
5949 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5951 }
5954 /* Simple loop detecting on_failure_jump: just check on the
5955 failure stack if the same spot was already hit earlier. */
5957 on_failure:
5961 {
5965 /* If there's a cycle, get out of the loop, as if the matching
5966 had failed. We used to just `goto fail' here, but that was
5967 aborting the search a bit too early: we want to keep the
5968 empty-loop-match and keep matching after the loop.
5969 We want (x?)*y\1z to match both xxyz and xxyxz. */
5971 else
5973 }
5977 /* Uses of on_failure_jump:
5979 Each alternative starts with an on_failure_jump that points
5980 to the beginning of the next alternative. Each alternative
5981 except the last ends with a jump that in effect jumps past
5982 the rest of the alternatives. (They really jump to the
5983 ending jump of the following alternative, because tensioning
5984 these jumps is a hassle.)
5986 Repeats start with an on_failure_jump that points past both
5987 the repetition text and either the following jump or
5988 pop_failure_jump back to this on_failure_jump. */
5997 /* This operation is used for greedy *.
5998 Compare the beginning of the repeat with what in the
5999 pattern follows its end. If we can establish that there
6000 is nothing that they would both match, i.e., that we
6001 would have to backtrack because of (as in, e.g., `a*a')
6002 then we can use a non-backtracking loop based on
6003 on_failure_keep_string_jump instead of on_failure_jump. */
6008 {
6010 /* Here, we discard `const', making re_match non-reentrant. */
6015 instruction once it's been changed. */
6019 /* Ensure this is a indeed the trivial kind of loop
6020 we are expecting. */
6025 {
6026 /* Use a fast `on_failure_keep_string_jump' loop. */
6030 }
6031 else
6032 {
6033 /* Default to a safe `on_failure_jump' loop. */
6036 }
6038 }
6041 /* Unconditionally jump (without popping any failure points). */
6043 unconditional_jump:
6044 IMMEDIATE_QUIT_CHECK;
6052 /* Have to succeed matching what follows at least n times.
6053 After that, handle like `on_failure_jump'. */
6055 /* Signedness doesn't matter since we only compare MCNT to 0. */
6059 /* Originally, mcnt is how many times we HAVE to succeed. */
6061 {
6062 /* Here, we discard `const', making re_match non-reentrant. */
6064 mcnt--;
6067 }
6068 else
6069 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6074 /* Signedness doesn't matter since we only compare MCNT to 0. */
6078 /* Originally, this is how many times we CAN jump. */
6080 {
6081 /* Here, we discard `const', making re_match non-reentrant. */
6083 mcnt--;
6086 }
6087 /* If don't have to jump any more, skip over the rest of command. */
6088 else
6093 {
6098 /* Here, we discard `const', making re_match non-reentrant. */
6100 /* Signedness doesn't matter since we only copy MCNT's bits . */
6105 }
6112 /* We SUCCEED (or FAIL) in one of the following cases: */
6114 /* Case 1: D is at the beginning or the end of string. */
6117 else
6118 {
6119 /* C1 is the character before D, S1 is the syntax of C1, C2
6120 is the character at D, and S2 is the syntax of C2. */
6124 #ifdef emacs
6128 #endif
6131 #ifdef emacs
6133 #endif
6140 /* Case 3: Both of S1 and S2 are Sword, and macro
6141 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6144 }
6147 else
6153 /* We FAIL in one of the following cases: */
6155 /* Case 1: D is at the end of string. */
6158 else
6159 {
6160 /* C1 is the character before D, S1 is the syntax of C1, C2
6161 is the character at D, and S2 is the syntax of C2. */
6165 #ifdef emacs
6169 #endif
6174 /* Case 2: S2 is not Sword. */
6178 /* Case 3: D is not at the beginning of string ... */
6180 {
6182 #ifdef emacs
6184 #endif
6187 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6188 returns 0. */
6191 }
6192 }
6198 /* We FAIL in one of the following cases: */
6200 /* Case 1: D is at the beginning of string. */
6203 else
6204 {
6205 /* C1 is the character before D, S1 is the syntax of C1, C2
6206 is the character at D, and S2 is the syntax of C2. */
6210 #ifdef emacs
6214 #endif
6218 /* Case 2: S1 is not Sword. */
6222 /* Case 3: D is not at the end of string ... */
6224 {
6227 #ifdef emacs
6229 #endif
6232 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6233 returns 0. */
6236 }
6237 }
6243 /* We FAIL in one of the following cases: */
6245 /* Case 1: D is at the end of string. */
6248 else
6249 {
6250 /* C1 is the character before D, S1 is the syntax of C1, C2
6251 is the character at D, and S2 is the syntax of C2. */
6254 #ifdef emacs
6258 #endif
6263 /* Case 2: S2 is neither Sword nor Ssymbol. */
6267 /* Case 3: D is not at the beginning of string ... */
6269 {
6271 #ifdef emacs
6273 #endif
6276 /* ... and S1 is Sword or Ssymbol. */
6279 }
6280 }
6286 /* We FAIL in one of the following cases: */
6288 /* Case 1: D is at the beginning of string. */
6291 else
6292 {
6293 /* C1 is the character before D, S1 is the syntax of C1, C2
6294 is the character at D, and S2 is the syntax of C2. */
6297 #ifdef emacs
6301 #endif
6305 /* Case 2: S1 is neither Ssymbol nor Sword. */
6309 /* Case 3: D is not at the end of string ... */
6311 {
6314 #ifdef emacs
6316 #endif
6319 /* ... and S2 is Sword or Ssymbol. */
6322 }
6323 }
6332 #ifdef emacs
6333 {
6337 }
6338 #endif
6339 {
6341 re_wchar_t c;
6347 }
6350 #ifdef emacs
6375 {
6377 re_wchar_t c;
6383 }
6390 }
6394 /* We goto here if a matching operation fails. */
6395 fail:
6396 IMMEDIATE_QUIT_CHECK;
6398 {
6400 /* A restart point is known. Restore to that state. */
6404 {
6412 /* Fallthrough */
6418 continue_failure_jump:
6424 /* A special frame used for nastyloops. */
6429 }
6435 }
6436 else
6447 \f
6448 /* Subroutine definitions for re_match_2. */
6450 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6451 bytes; nonzero otherwise. */
6453 static int
6457 RE_TRANSLATE_TYPE translate;
6459 {
6464 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6465 different lengths, but relying on a single `len' would break this. -sm */
6467 {
6479 }
6485 }
6486 \f
6487 /* Entry points for GNU code. */
6489 /* re_compile_pattern is the GNU regular expression compiler: it
6490 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6491 Returns 0 if the pattern was valid, otherwise an error string.
6493 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6494 are set in BUFP on entry.
6496 We call regex_compile to do the actual compilation. */
6503 {
6504 reg_errcode_t ret;
6506 /* GNU code is written to assume at least RE_NREGS registers will be set
6507 (and at least one extra will be -1). */
6510 /* And GNU code determines whether or not to get register information
6511 by passing null for the REGS argument to re_match, etc., not by
6512 setting no_sub. */
6520 }
6522 \f
6523 /* Entry points compatible with 4.2 BSD regex library. We don't define
6524 them unless specifically requested. */
6526 #if defined _REGEX_RE_COMP || defined _LIBC
6528 /* BSD has one and only one pattern buffer. */
6532 # ifdef _LIBC
6533 /* Make these definitions weak in libc, so POSIX programs can redefine
6534 these names if they don't use our functions, and still use
6535 regcomp/regexec below without link errors. */
6536 weak_function
6537 # endif
6540 {
6541 reg_errcode_t ret;
6544 {
6546 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6549 }
6552 {
6555 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6561 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6563 }
6565 /* Since `re_exec' always passes NULL for the `regs' argument, we
6566 don't need to initialize the pattern buffer fields which affect it. */
6573 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6575 }
6578 int
6579 # ifdef _LIBC
6580 weak_function
6581 # endif
6584 {
6586 return
6588 }
6590 \f
6591 /* POSIX.2 functions. Don't define these for Emacs. */
6593 #ifndef emacs
6595 /* regcomp takes a regular expression as a string and compiles it.
6597 PREG is a regex_t *. We do not expect any fields to be initialized,
6598 since POSIX says we shouldn't. Thus, we set
6600 `buffer' to the compiled pattern;
6601 `used' to the length of the compiled pattern;
6602 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6603 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6604 RE_SYNTAX_POSIX_BASIC;
6605 `fastmap' to an allocated space for the fastmap;
6606 `fastmap_accurate' to zero;
6607 `re_nsub' to the number of subexpressions in PATTERN.
6609 PATTERN is the address of the pattern string.
6611 CFLAGS is a series of bits which affect compilation.
6613 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6614 use POSIX basic syntax.
6616 If REG_NEWLINE is set, then . and [^...] don't match newline.
6617 Also, regexec will try a match beginning after every newline.
6619 If REG_ICASE is set, then we considers upper- and lowercase
6620 versions of letters to be equivalent when matching.
6622 If REG_NOSUB is set, then when PREG is passed to regexec, that
6623 routine will report only success or failure, and nothing about the
6624 registers.
6626 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6627 the return codes and their meanings.) */
6629 int
6634 {
6635 reg_errcode_t ret;
6636 reg_syntax_t syntax
6640 /* regex_compile will allocate the space for the compiled pattern. */
6645 /* Try to allocate space for the fastmap. */
6649 {
6652 preg->translate
6658 /* Map uppercase characters to corresponding lowercase ones. */
6661 }
6662 else
6665 /* If REG_NEWLINE is set, newlines are treated differently. */
6670 }
6671 else
6676 /* POSIX says a null character in the pattern terminates it, so we
6677 can use strlen here in compiling the pattern. */
6680 /* POSIX doesn't distinguish between an unmatched open-group and an
6681 unmatched close-group: both are REG_EPAREN. */
6687 buffer. */
6693 }
6694 }
6696 }
6700 /* regexec searches for a given pattern, specified by PREG, in the
6701 string STRING.
6703 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6704 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6705 least NMATCH elements, and we set them to the offsets of the
6706 corresponding matched substrings.
6708 EFLAGS specifies `execution flags' which affect matching: if
6709 REG_NOTBOL is set, then ^ does not match at the beginning of the
6710 string; if REG_NOTEOL is set, then $ does not match at the end.
6712 We return 0 if we find a match and REG_NOMATCH if not. */
6714 int
6721 {
6724 regex_t private_preg;
6733 /* The user has told us exactly how many registers to return
6734 information about, via `nmatch'. We have to pass that on to the
6735 matching routines. */
6739 {
6745 }
6747 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6748 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6749 was a little bit longer but still only matching the real part.
6750 This works because the `endline' will check for a '\n' and will find a
6751 '\0', correctly deciding that this is not the end of a line.
6752 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6753 a convenient '\0' there. For all we know, the string could be preceded
6754 by '\n' which would throw things off. */
6756 /* Perform the searching operation. */
6761 /* Copy the register information to the POSIX structure. */
6763 {
6765 {
6769 {
6772 }
6773 }
6775 /* If we needed the temporary register info, free the space now. */
6777 }
6779 /* We want zero return to mean success, unlike `re_search'. */
6781 }
6785 /* Returns a message corresponding to an error code, ERR_CODE, returned
6786 from either regcomp or regexec. We don't use PREG here.
6788 ERR_CODE was previously called ERRCODE, but that name causes an
6789 error with msvc8 compiler. */
6791 size_t
6797 {
6803 /* Only error codes returned by the rest of the code should be passed
6804 to this routine. If we are given anything else, or if other regex
6805 code generates an invalid error code, then the program has a bug.
6806 Dump core so we can fix it. */
6814 {
6816 {
6819 }
6820 else
6822 }
6825 }
6829 /* Free dynamically allocated space used by PREG. */
6831 void
6834 {
6847 }
6852 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6853 (do not change this comment) */