| blob | 31f188efa9983fe7b5672eb3c1223a8f8e75ae21 |
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 # ifdef HAVE_UNISTD_H
200 # include <unistd.h>
201 # endif
203 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
207 {
211 {
214 }
216 }
220 {
222 /* We must call malloc explicitly when BLOCK is 0, since some
223 reallocs don't do this. */
226 else
229 {
232 }
234 }
236 # ifdef malloc
237 # undef malloc
238 # endif
239 # define malloc xmalloc
240 # ifdef realloc
241 # undef realloc
242 # endif
243 # define realloc xrealloc
245 /* This is the normal way of making sure we have memcpy, memcmp and memset. */
246 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
247 # include <string.h>
248 # else
249 # include <strings.h>
250 # ifndef memcmp
251 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
252 # endif
253 # ifndef memcpy
254 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
255 # endif
256 # endif
258 /* Define the syntax stuff for \<, \>, etc. */
260 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
263 # define SWITCH_ENUM_CAST(x) (x)
265 /* Dummy macros for non-Emacs environments. */
266 # define CHAR_CHARSET(c) 0
267 # define CHARSET_LEADING_CODE_BASE(c) 0
268 # define MAX_MULTIBYTE_LENGTH 1
269 # define RE_MULTIBYTE_P(x) 0
270 # define RE_TARGET_MULTIBYTE_P(x) 0
271 # define WORD_BOUNDARY_P(c1, c2) (0)
272 # define CHAR_HEAD_P(p) (1)
273 # define SINGLE_BYTE_CHAR_P(c) (1)
274 # define SAME_CHARSET_P(c1, c2) (1)
275 # define BYTES_BY_CHAR_HEAD(p) (1)
276 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
277 # define STRING_CHAR(p) (*(p))
278 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
279 # define CHAR_STRING(c, s) (*(s) = (c), 1)
280 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
281 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
282 # define RE_CHAR_TO_MULTIBYTE(c) (c)
283 # define RE_CHAR_TO_UNIBYTE(c) (c)
284 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
285 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
286 # define GET_CHAR_AFTER(c, p, len) \
287 (c = *p, len = 1)
288 # define MAKE_CHAR(charset, c1, c2) (c1)
289 # define BYTE8_TO_CHAR(c) (c)
290 # define CHAR_BYTE8_P(c) (0)
291 # define CHAR_LEADING_CODE(c) (c)
295 #ifndef RE_TRANSLATE
296 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
297 # define RE_TRANSLATE_P(TBL) (TBL)
298 #endif
299 \f
300 /* Get the interface, including the syntax bits. */
303 /* isalpha etc. are used for the character classes. */
304 #include <ctype.h>
306 #ifdef emacs
308 /* 1 if C is an ASCII character. */
309 # define IS_REAL_ASCII(c) ((c) < 0200)
311 /* 1 if C is a unibyte character. */
312 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
314 /* The Emacs definitions should not be directly affected by locales. */
316 /* In Emacs, these are only used for single-byte characters. */
323 /* This is only used for single-byte characters. */
326 /* The rest must handle multibyte characters. */
328 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
330 : 1)
332 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
334 : 1)
336 # define ISALNUM(c) (IS_REAL_ASCII (c) \
340 : SYNTAX (c) == Sword)
342 # define ISALPHA(c) (IS_REAL_ASCII (c) \
345 : SYNTAX (c) == Sword)
347 # define ISLOWER(c) (LOWERCASEP (c))
349 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
354 : SYNTAX (c) != Sword)
356 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
358 # define ISUPPER(c) (UPPERCASEP (c))
360 # define ISWORD(c) (SYNTAX (c) == Sword)
364 /* Jim Meyering writes:
366 "... Some ctype macros are valid only for character codes that
367 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
368 using /bin/cc or gcc but without giving an ansi option). So, all
369 ctype uses should be through macros like ISPRINT... If
370 STDC_HEADERS is defined, then autoconf has verified that the ctype
371 macros don't need to be guarded with references to isascii. ...
372 Defining isascii to 1 should let any compiler worth its salt
373 eliminate the && through constant folding."
374 Solaris defines some of these symbols so we must undefine them first. */
376 # undef ISASCII
377 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
378 # define ISASCII(c) 1
379 # else
380 # define ISASCII(c) isascii(c)
381 # endif
383 /* 1 if C is an ASCII character. */
384 # define IS_REAL_ASCII(c) ((c) < 0200)
386 /* This distinction is not meaningful, except in Emacs. */
387 # define ISUNIBYTE(c) 1
389 # ifdef isblank
390 # define ISBLANK(c) (ISASCII (c) && isblank (c))
391 # else
393 # endif
394 # ifdef isgraph
395 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
396 # else
397 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
398 # endif
400 # undef ISPRINT
401 # define ISPRINT(c) (ISASCII (c) && isprint (c))
402 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
403 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
404 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
405 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
406 # define ISLOWER(c) (ISASCII (c) && islower (c))
407 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
408 # define ISSPACE(c) (ISASCII (c) && isspace (c))
409 # define ISUPPER(c) (ISASCII (c) && isupper (c))
410 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
412 # define ISWORD(c) ISALPHA(c)
414 # ifdef _tolower
415 # define TOLOWER(c) _tolower(c)
416 # else
417 # define TOLOWER(c) tolower(c)
418 # endif
420 /* How many characters in the character set. */
421 # define CHAR_SET_SIZE 256
423 # ifdef SYNTAX_TABLE
431 static void
433 {
449 }
453 # define SYNTAX(c) re_syntax_table[(c)]
456 \f
457 #ifndef NULL
458 # define NULL (void *)0
459 #endif
461 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
462 since ours (we hope) works properly with all combinations of
463 machines, compilers, `char' and `unsigned char' argument types.
464 (Per Bothner suggested the basic approach.) */
465 #undef SIGN_EXTEND_CHAR
466 #if __STDC__
467 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
469 /* As in Harbison and Steele. */
470 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
471 #endif
472 \f
473 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
474 use `alloca' instead of `malloc'. This is because using malloc in
475 re_search* or re_match* could cause memory leaks when C-g is used in
476 Emacs; also, malloc is slower and causes storage fragmentation. On
477 the other hand, malloc is more portable, and easier to debug.
479 Because we sometimes use alloca, some routines have to be macros,
480 not functions -- `alloca'-allocated space disappears at the end of the
481 function it is called in. */
483 #ifdef REGEX_MALLOC
485 # define REGEX_ALLOCATE malloc
486 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
487 # define REGEX_FREE free
491 /* Emacs already defines alloca, sometimes. */
492 # ifndef alloca
494 /* Make alloca work the best possible way. */
495 # ifdef __GNUC__
496 # define alloca __builtin_alloca
498 # ifdef HAVE_ALLOCA_H
499 # include <alloca.h>
505 # define REGEX_ALLOCATE alloca
507 /* Assumes a `char *destination' variable. */
508 # define REGEX_REALLOCATE(source, osize, nsize) \
509 (destination = (char *) alloca (nsize), \
510 memcpy (destination, source, osize))
512 /* No need to do anything to free, after alloca. */
517 /* Define how to allocate the failure stack. */
519 #if defined REL_ALLOC && defined REGEX_MALLOC
521 # define REGEX_ALLOCATE_STACK(size) \
522 r_alloc (&failure_stack_ptr, (size))
523 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
524 r_re_alloc (&failure_stack_ptr, (nsize))
525 # define REGEX_FREE_STACK(ptr) \
526 r_alloc_free (&failure_stack_ptr)
530 # ifdef REGEX_MALLOC
532 # define REGEX_ALLOCATE_STACK malloc
533 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
534 # define REGEX_FREE_STACK free
538 # define REGEX_ALLOCATE_STACK alloca
540 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
541 REGEX_REALLOCATE (source, osize, nsize)
542 /* No need to explicitly free anything. */
543 # define REGEX_FREE_STACK(arg) ((void)0)
549 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
550 `string1' or just past its end. This works if PTR is NULL, which is
551 a good thing. */
552 #define FIRST_STRING_P(ptr) \
553 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
555 /* (Re)Allocate N items of type T using malloc, or fail. */
556 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
557 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
558 #define RETALLOC_IF(addr, n, t) \
559 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
560 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
564 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
566 #undef MAX
567 #undef MIN
568 #define MAX(a, b) ((a) > (b) ? (a) : (b))
569 #define MIN(a, b) ((a) < (b) ? (a) : (b))
571 /* Type of source-pattern and string chars. */
575 #define false 0
576 #define true 1
584 \f
585 /* These are the command codes that appear in compiled regular
586 expressions. Some opcodes are followed by argument bytes. A
587 command code can specify any interpretation whatsoever for its
588 arguments. Zero bytes may appear in the compiled regular expression. */
591 {
594 /* Succeed right away--no more backtracking. */
595 succeed,
597 /* Followed by one byte giving n, then by n literal bytes. */
598 exactn,
600 /* Matches any (more or less) character. */
601 anychar,
603 /* Matches any one char belonging to specified set. First
604 following byte is number of bitmap bytes. Then come bytes
605 for a bitmap saying which chars are in. Bits in each byte
606 are ordered low-bit-first. A character is in the set if its
607 bit is 1. A character too large to have a bit in the map is
608 automatically not in the set.
610 If the length byte has the 0x80 bit set, then that stuff
611 is followed by a range table:
612 2 bytes of flags for character sets (low 8 bits, high 8 bits)
613 See RANGE_TABLE_WORK_BITS below.
614 2 bytes, the number of pairs that follow (upto 32767)
615 pairs, each 2 multibyte characters,
616 each multibyte character represented as 3 bytes. */
617 charset,
619 /* Same parameters as charset, but match any character that is
620 not one of those specified. */
621 charset_not,
623 /* Start remembering the text that is matched, for storing in a
624 register. Followed by one byte with the register number, in
625 the range 0 to one less than the pattern buffer's re_nsub
626 field. */
627 start_memory,
629 /* Stop remembering the text that is matched and store it in a
630 memory register. Followed by one byte with the register
631 number, in the range 0 to one less than `re_nsub' in the
632 pattern buffer. */
633 stop_memory,
635 /* Match a duplicate of something remembered. Followed by one
636 byte containing the register number. */
637 duplicate,
639 /* Fail unless at beginning of line. */
640 begline,
642 /* Fail unless at end of line. */
643 endline,
645 /* Succeeds if at beginning of buffer (if emacs) or at beginning
646 of string to be matched (if not). */
647 begbuf,
649 /* Analogously, for end of buffer/string. */
650 endbuf,
652 /* Followed by two byte relative address to which to jump. */
653 jump,
655 /* Followed by two-byte relative address of place to resume at
656 in case of failure. */
657 on_failure_jump,
659 /* Like on_failure_jump, but pushes a placeholder instead of the
660 current string position when executed. */
661 on_failure_keep_string_jump,
663 /* Just like `on_failure_jump', except that it checks that we
664 don't get stuck in an infinite loop (matching an empty string
665 indefinitely). */
666 on_failure_jump_loop,
668 /* Just like `on_failure_jump_loop', except that it checks for
669 a different kind of loop (the kind that shows up with non-greedy
670 operators). This operation has to be immediately preceded
671 by a `no_op'. */
672 on_failure_jump_nastyloop,
674 /* A smart `on_failure_jump' used for greedy * and + operators.
675 It analyses the loop before which it is put and if the
676 loop does not require backtracking, it changes itself to
677 `on_failure_keep_string_jump' and short-circuits the loop,
678 else it just defaults to changing itself into `on_failure_jump'.
679 It assumes that it is pointing to just past a `jump'. */
680 on_failure_jump_smart,
682 /* Followed by two-byte relative address and two-byte number n.
683 After matching N times, jump to the address upon failure.
684 Does not work if N starts at 0: use on_failure_jump_loop
685 instead. */
686 succeed_n,
688 /* Followed by two-byte relative address, and two-byte number n.
689 Jump to the address N times, then fail. */
690 jump_n,
692 /* Set the following two-byte relative address to the
693 subsequent two-byte number. The address *includes* the two
694 bytes of number. */
695 set_number_at,
706 /* Matches any character whose syntax is specified. Followed by
707 a byte which contains a syntax code, e.g., Sword. */
708 syntaxspec,
710 /* Matches any character whose syntax is not that specified. */
711 notsyntaxspec
713 #ifdef emacs
718 /* Matches any character whose category-set contains the specified
719 category. The operator is followed by a byte which contains a
720 category code (mnemonic ASCII character). */
721 categoryspec,
723 /* Matches any character whose category-set does not contain the
724 specified category. The operator is followed by a byte which
725 contains the category code (mnemonic ASCII character). */
726 notcategoryspec
729 \f
730 /* Common operations on the compiled pattern. */
732 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
734 #define STORE_NUMBER(destination, number) \
735 do { \
736 (destination)[0] = (number) & 0377; \
737 (destination)[1] = (number) >> 8; \
738 } while (0)
740 /* Same as STORE_NUMBER, except increment DESTINATION to
741 the byte after where the number is stored. Therefore, DESTINATION
742 must be an lvalue. */
744 #define STORE_NUMBER_AND_INCR(destination, number) \
745 do { \
746 STORE_NUMBER (destination, number); \
747 (destination) += 2; \
748 } while (0)
750 /* Put into DESTINATION a number stored in two contiguous bytes starting
751 at SOURCE. */
753 #define EXTRACT_NUMBER(destination, source) \
754 do { \
755 (destination) = *(source) & 0377; \
756 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
757 } while (0)
759 #ifdef DEBUG
761 static void
765 {
769 }
772 # undef EXTRACT_NUMBER
773 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
778 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
779 SOURCE must be an lvalue. */
781 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
782 do { \
783 EXTRACT_NUMBER (destination, source); \
784 (source) += 2; \
785 } while (0)
787 #ifdef DEBUG
790 static void
794 {
797 }
799 # ifndef EXTRACT_MACROS
800 # undef EXTRACT_NUMBER_AND_INCR
801 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
802 extract_number_and_incr (&dest, &src)
806 \f
807 /* Store a multibyte character in three contiguous bytes starting
808 DESTINATION, and increment DESTINATION to the byte after where the
809 character is stored. Therefore, DESTINATION must be an lvalue. */
811 #define STORE_CHARACTER_AND_INCR(destination, character) \
812 do { \
813 (destination)[0] = (character) & 0377; \
814 (destination)[1] = ((character) >> 8) & 0377; \
815 (destination)[2] = (character) >> 16; \
816 (destination) += 3; \
817 } while (0)
819 /* Put into DESTINATION a character stored in three contiguous bytes
820 starting at SOURCE. */
822 #define EXTRACT_CHARACTER(destination, source) \
823 do { \
824 (destination) = ((source)[0] \
825 | ((source)[1] << 8) \
826 | ((source)[2] << 16)); \
827 } while (0)
830 /* Macros for charset. */
832 /* Size of bitmap of charset P in bytes. P is a start of charset,
833 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
834 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
836 /* Nonzero if charset P has range table. */
837 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
839 /* Return the address of range table of charset P. But not the start
840 of table itself, but the before where the number of ranges is
841 stored. `2 +' means to skip re_opcode_t and size of bitmap,
842 and the 2 bytes of flags at the start of the range table. */
843 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
845 /* Extract the bit flags that start a range table. */
846 #define CHARSET_RANGE_TABLE_BITS(p) \
847 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
848 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
850 /* Test if C is listed in the bitmap of charset P. */
851 #define CHARSET_LOOKUP_BITMAP(p, c) \
852 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
853 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
855 /* Return the address of end of RANGE_TABLE. COUNT is number of
856 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
857 is start of range and end of range. `* 3' is size of each start
858 and end. */
859 #define CHARSET_RANGE_TABLE_END(range_table, count) \
860 ((range_table) + (count) * 2 * 3)
862 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
863 COUNT is number of ranges in RANGE_TABLE. */
864 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
865 do \
866 { \
867 re_wchar_t range_start, range_end; \
868 re_char *p; \
869 re_char *range_table_end \
870 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
871 \
872 for (p = (range_table); p < range_table_end; p += 2 * 3) \
873 { \
874 EXTRACT_CHARACTER (range_start, p); \
875 EXTRACT_CHARACTER (range_end, p + 3); \
876 \
877 if (range_start <= (c) && (c) <= range_end) \
878 { \
879 (not) = !(not); \
880 break; \
881 } \
882 } \
883 } \
884 while (0)
886 /* Test if C is in range table of CHARSET. The flag NOT is negated if
887 C is listed in it. */
888 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
889 do \
890 { \
892 int count; \
893 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
894 \
895 EXTRACT_NUMBER_AND_INCR (count, range_table); \
896 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
897 } \
898 while (0)
899 \f
900 /* If DEBUG is defined, Regex prints many voluminous messages about what
901 it is doing (if the variable `debug' is nonzero). If linked with the
902 main program in `iregex.c', you can enter patterns and strings
903 interactively. And if linked with the main program in `main.c' and
904 the other test files, you can run the already-written tests. */
906 #ifdef DEBUG
908 /* We use standard I/O for debugging. */
909 # include <stdio.h>
911 /* It is useful to test things that ``must'' be true when debugging. */
912 # include <assert.h>
916 # define DEBUG_STATEMENT(e) e
917 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
918 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
919 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
920 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
921 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
922 if (debug > 0) print_partial_compiled_pattern (s, e)
923 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
924 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
927 /* Print the fastmap in human-readable form. */
929 void
932 {
937 {
939 {
943 {
945 i++;
946 }
948 {
951 }
952 }
953 }
955 }
958 /* Print a compiled pattern string in human-readable form, starting at
959 the START pointer into it and ending just before the pointer END. */
961 void
965 {
971 {
974 }
976 /* Loop over pattern commands. */
978 {
982 {
994 do
995 {
997 }
1019 {
1034 {
1035 /* Are we starting a range? */
1037 {
1040 }
1041 /* Have we broken a range? */
1043 {
1046 }
1052 }
1062 {
1066 /* ??? Should print the range table; for now, just skip it. */
1070 }
1071 }
1166 # ifdef emacs
1202 }
1205 }
1208 }
1211 void
1214 {
1222 {
1225 }
1235 /* Perhaps we should print the translate table? */
1236 }
1239 void
1246 {
1251 else
1252 {
1254 {
1259 }
1263 }
1264 }
1268 # undef assert
1269 # define assert(e)
1271 # define DEBUG_STATEMENT(e)
1272 # define DEBUG_PRINT1(x)
1273 # define DEBUG_PRINT2(x1, x2)
1274 # define DEBUG_PRINT3(x1, x2, x3)
1275 # define DEBUG_PRINT4(x1, x2, x3, x4)
1276 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1277 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1280 \f
1281 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1282 also be assigned to arbitrarily: each pattern buffer stores its own
1283 syntax, so it can be changed between regex compilations. */
1284 /* This has no initializer because initialized variables in Emacs
1285 become read-only after dumping. */
1286 reg_syntax_t re_syntax_options;
1289 /* Specify the precise syntax of regexps for compilation. This provides
1290 for compatibility for various utilities which historically have
1291 different, incompatible syntaxes.
1293 The argument SYNTAX is a bit mask comprised of the various bits
1294 defined in regex.h. We return the old syntax. */
1296 reg_syntax_t
1298 {
1303 }
1306 /* Regexp to use to replace spaces, or NULL meaning don't. */
1309 void
1311 {
1313 }
1315 \f
1316 /* This table gives an error message for each of the error codes listed
1317 in regex.h. Obviously the order here has to be same as there.
1318 POSIX doesn't require that we do anything for REG_NOERROR,
1319 but why not be nice? */
1322 {
1341 };
1342 \f
1343 /* Avoiding alloca during matching, to placate r_alloc. */
1345 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1346 searching and matching functions should not call alloca. On some
1347 systems, alloca is implemented in terms of malloc, and if we're
1348 using the relocating allocator routines, then malloc could cause a
1349 relocation, which might (if the strings being searched are in the
1350 ralloc heap) shift the data out from underneath the regexp
1351 routines.
1353 Here's another reason to avoid allocation: Emacs
1354 processes input from X in a signal handler; processing X input may
1355 call malloc; if input arrives while a matching routine is calling
1356 malloc, then we're scrod. But Emacs can't just block input while
1357 calling matching routines; then we don't notice interrupts when
1358 they come in. So, Emacs blocks input around all regexp calls
1359 except the matching calls, which it leaves unprotected, in the
1360 faith that they will not malloc. */
1362 /* Normally, this is fine. */
1363 #define MATCH_MAY_ALLOCATE
1365 /* The match routines may not allocate if (1) they would do it with malloc
1366 and (2) it's not safe for them to use malloc.
1367 Note that if REL_ALLOC is defined, matching would not use malloc for the
1368 failure stack, but we would still use it for the register vectors;
1369 so REL_ALLOC should not affect this. */
1370 #if defined REGEX_MALLOC && defined emacs
1371 # undef MATCH_MAY_ALLOCATE
1372 #endif
1374 \f
1375 /* Failure stack declarations and macros; both re_compile_fastmap and
1376 re_match_2 use a failure stack. These have to be macros because of
1377 REGEX_ALLOCATE_STACK. */
1380 /* Approximate number of failure points for which to initially allocate space
1381 when matching. If this number is exceeded, we allocate more
1382 space, so it is not a hard limit. */
1383 #ifndef INIT_FAILURE_ALLOC
1384 # define INIT_FAILURE_ALLOC 20
1385 #endif
1387 /* Roughly the maximum number of failure points on the stack. Would be
1388 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1389 This is a variable only so users of regex can assign to it; we never
1390 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1391 before using it, so it should probably be a byte-count instead. */
1392 # if defined MATCH_MAY_ALLOCATE
1393 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1394 whose default stack limit is 2mb. In order for a larger
1395 value to work reliably, you have to try to make it accord
1396 with the process stack limit. */
1398 # else
1400 # endif
1402 union fail_stack_elt
1403 {
1405 /* This should be the biggest `int' that's no bigger than a pointer. */
1407 };
1412 {
1419 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1420 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1423 /* Define macros to initialize and free the failure stack.
1424 Do `return -2' if the alloc fails. */
1426 #ifdef MATCH_MAY_ALLOCATE
1427 # define INIT_FAIL_STACK() \
1428 do { \
1429 fail_stack.stack = (fail_stack_elt_t *) \
1430 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1431 * sizeof (fail_stack_elt_t)); \
1432 \
1433 if (fail_stack.stack == NULL) \
1434 return -2; \
1435 \
1436 fail_stack.size = INIT_FAILURE_ALLOC; \
1437 fail_stack.avail = 0; \
1438 fail_stack.frame = 0; \
1439 } while (0)
1441 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1442 #else
1443 # define INIT_FAIL_STACK() \
1444 do { \
1445 fail_stack.avail = 0; \
1446 fail_stack.frame = 0; \
1447 } while (0)
1449 # define RESET_FAIL_STACK() ((void)0)
1450 #endif
1453 /* Double the size of FAIL_STACK, up to a limit
1454 which allows approximately `re_max_failures' items.
1456 Return 1 if succeeds, and 0 if either ran out of memory
1457 allocating space for it or it was already too large.
1459 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1461 /* Factor to increase the failure stack size by
1462 when we increase it.
1463 This used to be 2, but 2 was too wasteful
1464 because the old discarded stacks added up to as much space
1465 were as ultimate, maximum-size stack. */
1466 #define FAIL_STACK_GROWTH_FACTOR 4
1468 #define GROW_FAIL_STACK(fail_stack) \
1469 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1470 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1471 ? 0 \
1472 : ((fail_stack).stack \
1473 = (fail_stack_elt_t *) \
1474 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1475 (fail_stack).size * sizeof (fail_stack_elt_t), \
1476 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1477 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1478 * FAIL_STACK_GROWTH_FACTOR))), \
1479 \
1480 (fail_stack).stack == NULL \
1481 ? 0 \
1482 : ((fail_stack).size \
1483 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1484 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1485 * FAIL_STACK_GROWTH_FACTOR)) \
1486 / sizeof (fail_stack_elt_t)), \
1487 1)))
1490 /* Push a pointer value onto the failure stack.
1491 Assumes the variable `fail_stack'. Probably should only
1492 be called from within `PUSH_FAILURE_POINT'. */
1493 #define PUSH_FAILURE_POINTER(item) \
1494 fail_stack.stack[fail_stack.avail++].pointer = (item)
1496 /* This pushes an integer-valued item onto the failure stack.
1497 Assumes the variable `fail_stack'. Probably should only
1498 be called from within `PUSH_FAILURE_POINT'. */
1499 #define PUSH_FAILURE_INT(item) \
1500 fail_stack.stack[fail_stack.avail++].integer = (item)
1502 /* Push a fail_stack_elt_t value onto the failure stack.
1503 Assumes the variable `fail_stack'. Probably should only
1504 be called from within `PUSH_FAILURE_POINT'. */
1505 #define PUSH_FAILURE_ELT(item) \
1506 fail_stack.stack[fail_stack.avail++] = (item)
1508 /* These three POP... operations complement the three PUSH... operations.
1509 All assume that `fail_stack' is nonempty. */
1510 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1511 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1512 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1514 /* Individual items aside from the registers. */
1515 #define NUM_NONREG_ITEMS 3
1517 /* Used to examine the stack (to detect infinite loops). */
1518 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1519 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1520 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1521 #define TOP_FAILURE_HANDLE() fail_stack.frame
1524 #define ENSURE_FAIL_STACK(space) \
1525 while (REMAINING_AVAIL_SLOTS <= space) { \
1526 if (!GROW_FAIL_STACK (fail_stack)) \
1527 return -2; \
1530 }
1532 /* Push register NUM onto the stack. */
1533 #define PUSH_FAILURE_REG(num) \
1534 do { \
1535 char *destination; \
1536 ENSURE_FAIL_STACK(3); \
1538 num, regstart[num], regend[num]); \
1539 PUSH_FAILURE_POINTER (regstart[num]); \
1540 PUSH_FAILURE_POINTER (regend[num]); \
1541 PUSH_FAILURE_INT (num); \
1542 } while (0)
1544 /* Change the counter's value to VAL, but make sure that it will
1545 be reset when backtracking. */
1546 #define PUSH_NUMBER(ptr,val) \
1547 do { \
1548 char *destination; \
1549 int c; \
1550 ENSURE_FAIL_STACK(3); \
1551 EXTRACT_NUMBER (c, ptr); \
1553 PUSH_FAILURE_INT (c); \
1554 PUSH_FAILURE_POINTER (ptr); \
1555 PUSH_FAILURE_INT (-1); \
1556 STORE_NUMBER (ptr, val); \
1557 } while (0)
1559 /* Pop a saved register off the stack. */
1560 #define POP_FAILURE_REG_OR_COUNT() \
1561 do { \
1562 int reg = POP_FAILURE_INT (); \
1563 if (reg == -1) \
1564 { \
1567 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1568 reg = POP_FAILURE_INT (); \
1569 STORE_NUMBER (ptr, reg); \
1571 } \
1572 else \
1573 { \
1574 regend[reg] = POP_FAILURE_POINTER (); \
1575 regstart[reg] = POP_FAILURE_POINTER (); \
1577 reg, regstart[reg], regend[reg]); \
1578 } \
1579 } while (0)
1581 /* Check that we are not stuck in an infinite loop. */
1582 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1583 do { \
1584 int failure = TOP_FAILURE_HANDLE (); \
1586 while (failure > 0 \
1587 && (FAILURE_STR (failure) == string_place \
1588 || FAILURE_STR (failure) == NULL)) \
1589 { \
1590 assert (FAILURE_PAT (failure) >= bufp->buffer \
1591 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1592 if (FAILURE_PAT (failure) == pat_cur) \
1593 { \
1594 cycle = 1; \
1595 break; \
1596 } \
1598 failure = NEXT_FAILURE_HANDLE(failure); \
1599 } \
1601 } while (0)
1603 /* Push the information about the state we will need
1604 if we ever fail back to it.
1606 Requires variables fail_stack, regstart, regend and
1607 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1608 declared.
1610 Does `return FAILURE_CODE' if runs out of memory. */
1612 #define PUSH_FAILURE_POINT(pattern, string_place) \
1613 do { \
1614 char *destination; \
1615 /* Must be int, so when we don't save any registers, the arithmetic \
1617 \
1618 DEBUG_STATEMENT (nfailure_points_pushed++); \
1622 \
1623 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1624 \
1626 \
1628 PUSH_FAILURE_INT (fail_stack.frame); \
1629 \
1631 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1633 PUSH_FAILURE_POINTER (string_place); \
1634 \
1636 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1637 PUSH_FAILURE_POINTER (pattern); \
1638 \
1640 fail_stack.frame = fail_stack.avail; \
1641 } while (0)
1643 /* Estimate the size of data pushed by a typical failure stack entry.
1644 An estimate is all we need, because all we use this for
1645 is to choose a limit for how big to make the failure stack. */
1646 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1647 #define TYPICAL_FAILURE_SIZE 20
1649 /* How many items can still be added to the stack without overflowing it. */
1650 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1653 /* Pops what PUSH_FAIL_STACK pushes.
1655 We restore into the parameters, all of which should be lvalues:
1656 STR -- the saved data position.
1657 PAT -- the saved pattern position.
1658 REGSTART, REGEND -- arrays of string positions.
1660 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1661 `pend', `string1', `size1', `string2', and `size2'. */
1663 #define POP_FAILURE_POINT(str, pat) \
1664 do { \
1665 assert (!FAIL_STACK_EMPTY ()); \
1666 \
1671 \
1673 while (fail_stack.frame < fail_stack.avail) \
1674 POP_FAILURE_REG_OR_COUNT (); \
1675 \
1676 pat = POP_FAILURE_POINTER (); \
1678 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1679 \
1680 /* If the saved string location is NULL, it came from an \
1681 on_failure_keep_string_jump opcode, and we want to throw away the \
1683 str = POP_FAILURE_POINTER (); \
1685 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1687 \
1688 fail_stack.frame = POP_FAILURE_INT (); \
1690 \
1691 assert (fail_stack.avail >= 0); \
1692 assert (fail_stack.frame <= fail_stack.avail); \
1693 \
1694 DEBUG_STATEMENT (nfailure_points_popped++); \
1698 \f
1699 /* Registers are set to a sentinel when they haven't yet matched. */
1700 #define REG_UNSET(e) ((e) == NULL)
1701 \f
1702 /* Subroutine declarations and macros for regex_compile. */
1705 reg_syntax_t syntax,
1716 reg_syntax_t syntax));
1719 reg_syntax_t syntax));
1724 /* Fetch the next character in the uncompiled pattern, with no
1725 translation. */
1726 #define PATFETCH(c) \
1727 do { \
1728 int len; \
1729 if (p == pend) return REG_EEND; \
1730 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1731 p += len; \
1732 } while (0)
1735 /* If `translate' is non-null, return translate[D], else just D. We
1736 cast the subscript to translate because some data is declared as
1737 `char *', to avoid warnings when a string constant is passed. But
1738 when we use a character as a subscript we must make it unsigned. */
1739 #ifndef TRANSLATE
1740 # define TRANSLATE(d) \
1741 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1742 #endif
1745 /* Macros for outputting the compiled pattern into `buffer'. */
1747 /* If the buffer isn't allocated when it comes in, use this. */
1748 #define INIT_BUF_SIZE 32
1750 /* Make sure we have at least N more bytes of space in buffer. */
1751 #define GET_BUFFER_SPACE(n) \
1752 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1753 EXTEND_BUFFER ()
1755 /* Make sure we have one more byte of buffer space and then add C to it. */
1756 #define BUF_PUSH(c) \
1757 do { \
1758 GET_BUFFER_SPACE (1); \
1759 *b++ = (unsigned char) (c); \
1760 } while (0)
1763 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1764 #define BUF_PUSH_2(c1, c2) \
1765 do { \
1766 GET_BUFFER_SPACE (2); \
1767 *b++ = (unsigned char) (c1); \
1768 *b++ = (unsigned char) (c2); \
1769 } while (0)
1772 /* As with BUF_PUSH_2, except for three bytes. */
1773 #define BUF_PUSH_3(c1, c2, c3) \
1774 do { \
1775 GET_BUFFER_SPACE (3); \
1776 *b++ = (unsigned char) (c1); \
1777 *b++ = (unsigned char) (c2); \
1778 *b++ = (unsigned char) (c3); \
1779 } while (0)
1782 /* Store a jump with opcode OP at LOC to location TO. We store a
1783 relative address offset by the three bytes the jump itself occupies. */
1784 #define STORE_JUMP(op, loc, to) \
1785 store_op1 (op, loc, (to) - (loc) - 3)
1787 /* Likewise, for a two-argument jump. */
1788 #define STORE_JUMP2(op, loc, to, arg) \
1789 store_op2 (op, loc, (to) - (loc) - 3, arg)
1791 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1792 #define INSERT_JUMP(op, loc, to) \
1793 insert_op1 (op, loc, (to) - (loc) - 3, b)
1795 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1796 #define INSERT_JUMP2(op, loc, to, arg) \
1797 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1800 /* This is not an arbitrary limit: the arguments which represent offsets
1801 into the pattern are two bytes long. So if 2^15 bytes turns out to
1802 be too small, many things would have to change. */
1803 # define MAX_BUF_SIZE (1L << 15)
1806 /* Any other compiler which, like MSC, has allocation limit below 2^16
1807 bytes will have to use approach similar to what was done below for
1808 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1809 reallocating to 0 bytes. Such thing is not going to work too well.
1810 You have been warned!! */
1811 #if defined _MSC_VER && !defined WIN32
1812 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1813 # define MAX_BUF_SIZE 65500L
1814 #else
1815 # define MAX_BUF_SIZE (1L << 16)
1816 #endif
1819 /* Extend the buffer by twice its current size via realloc and
1820 reset the pointers that pointed into the old block to point to the
1821 correct places in the new one. If extending the buffer results in it
1822 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1823 #if __BOUNDED_POINTERS__
1824 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1825 # define MOVE_BUFFER_POINTER(P) \
1826 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1827 SET_HIGH_BOUND (P), \
1828 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1829 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1830 else \
1831 { \
1832 SET_HIGH_BOUND (b); \
1833 SET_HIGH_BOUND (begalt); \
1834 if (fixup_alt_jump) \
1835 SET_HIGH_BOUND (fixup_alt_jump); \
1836 if (laststart) \
1837 SET_HIGH_BOUND (laststart); \
1838 if (pending_exact) \
1839 SET_HIGH_BOUND (pending_exact); \
1840 }
1841 #else
1842 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1843 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1844 #endif
1845 #define EXTEND_BUFFER() \
1846 do { \
1847 unsigned char *old_buffer = bufp->buffer; \
1848 if (bufp->allocated == MAX_BUF_SIZE) \
1849 return REG_ESIZE; \
1850 bufp->allocated <<= 1; \
1851 if (bufp->allocated > MAX_BUF_SIZE) \
1852 bufp->allocated = MAX_BUF_SIZE; \
1853 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1854 if (bufp->buffer == NULL) \
1855 return REG_ESPACE; \
1857 if (old_buffer != bufp->buffer) \
1858 { \
1859 unsigned char *new_buffer = bufp->buffer; \
1860 MOVE_BUFFER_POINTER (b); \
1861 MOVE_BUFFER_POINTER (begalt); \
1862 if (fixup_alt_jump) \
1863 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1864 if (laststart) \
1865 MOVE_BUFFER_POINTER (laststart); \
1866 if (pending_exact) \
1867 MOVE_BUFFER_POINTER (pending_exact); \
1868 } \
1869 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1870 } while (0)
1873 /* Since we have one byte reserved for the register number argument to
1874 {start,stop}_memory, the maximum number of groups we can report
1875 things about is what fits in that byte. */
1876 #define MAX_REGNUM 255
1878 /* But patterns can have more than `MAX_REGNUM' registers. We just
1879 ignore the excess. */
1883 /* Macros for the compile stack. */
1885 /* Since offsets can go either forwards or backwards, this type needs to
1886 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1887 /* int may be not enough when sizeof(int) == 2. */
1891 {
1892 pattern_offset_t begalt_offset;
1893 pattern_offset_t fixup_alt_jump;
1894 pattern_offset_t laststart_offset;
1895 regnum_t regnum;
1900 {
1907 #define INIT_COMPILE_STACK_SIZE 32
1909 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1910 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1912 /* The next available element. */
1913 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1915 /* Explicit quit checking is only used on NTemacs and whenever we
1916 use polling to process input events. */
1917 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1919 # define IMMEDIATE_QUIT_CHECK \
1920 do { \
1921 if (immediate_quit) QUIT; \
1922 } while (0)
1923 #else
1924 # define IMMEDIATE_QUIT_CHECK ((void)0)
1925 #endif
1926 \f
1927 /* Structure to manage work area for range table. */
1928 struct range_table_work_area
1929 {
1934 };
1936 /* Make sure that WORK_AREA can hold more N multibyte characters.
1937 This is used only in set_image_of_range and set_image_of_range_1.
1938 It expects WORK_AREA to be a pointer.
1939 If it can't get the space, it returns from the surrounding function. */
1941 #define EXTEND_RANGE_TABLE(work_area, n) \
1942 do { \
1943 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1944 { \
1945 extend_range_table_work_area (&work_area); \
1946 if ((work_area).table == 0) \
1947 return (REG_ESPACE); \
1948 } \
1949 } while (0)
1951 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1952 (work_area).bits |= (bit)
1954 /* Bits used to implement the multibyte-part of the various character classes
1955 such as [:alnum:] in a charset's range table. */
1956 #define BIT_WORD 0x1
1957 #define BIT_LOWER 0x2
1958 #define BIT_PUNCT 0x4
1959 #define BIT_SPACE 0x8
1960 #define BIT_UPPER 0x10
1961 #define BIT_MULTIBYTE 0x20
1963 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1964 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1965 do { \
1966 EXTEND_RANGE_TABLE ((work_area), 2); \
1967 (work_area).table[(work_area).used++] = (range_start); \
1968 (work_area).table[(work_area).used++] = (range_end); \
1969 } while (0)
1971 /* Free allocated memory for WORK_AREA. */
1972 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1973 do { \
1974 if ((work_area).table) \
1975 free ((work_area).table); \
1976 } while (0)
1978 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1979 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1980 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1981 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1982 \f
1984 /* Set the bit for character C in a list. */
1985 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1988 #ifdef emacs
1990 /* Store characters in the range FROM to TO in the bitmap at B (for
1991 ASCII and unibyte characters) and WORK_AREA (for multibyte
1992 characters) while translating them and paying attention to the
1993 continuity of translated characters.
1995 Implementation note: It is better to implement these fairly big
1996 macros by a function, but it's not that easy because macros called
1997 in this macro assume various local variables already declared. */
1999 /* Both FROM and TO are ASCII characters. */
2001 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2002 do { \
2003 int C0, C1; \
2004 \
2005 for (C0 = (FROM); C0 <= (TO); C0++) \
2006 { \
2007 C1 = TRANSLATE (C0); \
2008 if (! ASCII_CHAR_P (C1)) \
2009 { \
2010 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2011 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2012 C1 = C0; \
2013 } \
2014 SET_LIST_BIT (C1); \
2015 } \
2016 } while (0)
2019 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2021 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2022 do { \
2023 int C0, C1, C2, I; \
2024 int USED = RANGE_TABLE_WORK_USED (work_area); \
2025 \
2026 for (C0 = (FROM); C0 <= (TO); C0++) \
2027 { \
2028 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2029 if (CHAR_BYTE8_P (C1)) \
2030 SET_LIST_BIT (C0); \
2031 else \
2032 { \
2033 C2 = TRANSLATE (C1); \
2034 if (C2 == C1 \
2035 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2036 C1 = C0; \
2037 SET_LIST_BIT (C1); \
2038 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2039 { \
2040 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2041 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2042 \
2043 if (C2 >= from - 1 && C2 <= to + 1) \
2044 { \
2045 if (C2 == from - 1) \
2046 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2047 else if (C2 == to + 1) \
2048 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2049 break; \
2050 } \
2051 } \
2052 if (I < USED) \
2053 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2054 } \
2055 } \
2056 } while (0)
2059 /* Both FROM and TO are multibyte characters. */
2061 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2062 do { \
2063 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2064 \
2065 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2066 for (C0 = (FROM); C0 <= (TO); C0++) \
2067 { \
2068 C1 = TRANSLATE (C0); \
2069 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2070 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2071 SET_LIST_BIT (C2); \
2072 if (C1 >= (FROM) && C1 <= (TO)) \
2073 continue; \
2074 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2075 { \
2076 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2077 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2078 \
2079 if (C1 >= from - 1 && C1 <= to + 1) \
2080 { \
2081 if (C1 == from - 1) \
2082 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2083 else if (C1 == to + 1) \
2084 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2085 break; \
2086 } \
2087 } \
2088 if (I < USED) \
2089 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2090 } \
2091 } while (0)
2095 /* Get the next unsigned number in the uncompiled pattern. */
2096 #define GET_UNSIGNED_NUMBER(num) \
2097 do { \
2098 if (p == pend) \
2099 FREE_STACK_RETURN (REG_EBRACE); \
2100 else \
2101 { \
2102 PATFETCH (c); \
2104 { \
2105 int prev; \
2106 if (num < 0) \
2107 num = 0; \
2108 prev = num; \
2110 if (num / 10 != prev) \
2111 FREE_STACK_RETURN (REG_BADBR); \
2112 if (p == pend) \
2113 FREE_STACK_RETURN (REG_EBRACE); \
2114 PATFETCH (c); \
2115 } \
2116 } \
2117 } while (0)
2118 \f
2119 #if ! WIDE_CHAR_SUPPORT
2121 /* Map a string to the char class it names (if any). */
2122 re_wctype_t
2124 {
2144 }
2146 /* True if CH is in the char class CC. */
2147 boolean
2149 {
2151 {
2172 }
2173 }
2175 /* Return a bit-pattern to use in the range-table bits to match multibyte
2176 chars of class CC. */
2177 static int
2179 {
2181 {
2193 }
2194 }
2195 #endif
2196 \f
2197 /* Filling in the work area of a range. */
2199 /* Actually extend the space in WORK_AREA. */
2201 static void
2203 {
2206 work_area->table
2208 else
2209 work_area->table
2211 }
2213 #if 0
2214 #ifdef emacs
2216 /* Carefully find the ranges of codes that are equivalent
2217 under case conversion to the range start..end when passed through
2218 TRANSLATE. Handle the case where non-letters can come in between
2219 two upper-case letters (which happens in Latin-1).
2220 Also handle the case of groups of more than 2 case-equivalent chars.
2222 The basic method is to look at consecutive characters and see
2223 if they can form a run that can be handled as one.
2225 Returns -1 if successful, REG_ESPACE if ran out of space. */
2227 static int
2229 RE_TRANSLATE_TYPE translate;
2232 {
2233 /* `one_case' indicates a character, or a run of characters,
2234 each of which is an isolate (no case-equivalents).
2235 This includes all ASCII non-letters.
2237 `two_case' indicates a character, or a run of characters,
2238 each of which has two case-equivalent forms.
2239 This includes all ASCII letters.
2241 `strange' indicates a character that has more than one
2242 case-equivalent. */
2246 /* Describe the run that is in progress,
2247 which the next character can try to extend.
2248 If run_type is strange, that means there really is no run.
2249 If run_type is one_case, then run_start...run_end is the run.
2250 If run_type is two_case, then the run is run_start...run_end,
2251 and the case-equivalents end at run_eqv_end. */
2256 Lisp_Object eqv_table;
2259 {
2264 }
2269 {
2274 /* Classify this character */
2279 else
2284 else
2287 /* Can this character extend the run in progress? */
2292 {
2293 /* No, end the run.
2294 Record each of its equivalent ranges. */
2296 {
2300 }
2302 {
2310 }
2312 }
2315 {
2316 /* For a strange character, add each of its equivalents, one
2317 by one. Don't start a range. */
2318 do
2319 {
2324 }
2326 }
2328 /* Add this char to the run, or start a new run. */
2330 {
2331 /* Initialize a new range. */
2336 }
2337 else
2338 {
2339 /* Extend a running range. */
2342 }
2343 }
2345 /* If a run is still in progress at the end, finish it now
2346 by recording its equivalent ranges. */
2348 {
2352 }
2354 {
2362 }
2365 }
2369 /* Record the image of the range start..end when passed through
2370 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2371 and is not even necessarily contiguous.
2372 Normally we approximate it with the smallest contiguous range that contains
2373 all the chars we need. However, for Latin-1 we go to extra effort
2374 to do a better job.
2376 This function is not called for ASCII ranges.
2378 Returns -1 if successful, REG_ESPACE if ran out of space. */
2380 static int
2382 RE_TRANSLATE_TYPE translate;
2385 {
2388 #ifdef emacs
2389 /* For Latin-1 ranges, use set_image_of_range_1
2390 to get proper handling of ranges that include letters and nonletters.
2391 For a range that includes the whole of Latin-1, this is not necessary.
2392 For other character sets, we don't bother to get this right. */
2395 {
2408 }
2409 #endif
2418 {
2422 {
2425 {
2428 else
2429 {
2432 }
2433 }
2434 }
2437 {
2441 }
2442 }
2445 }
2447 \f
2448 #ifndef MATCH_MAY_ALLOCATE
2450 /* If we cannot allocate large objects within re_match_2_internal,
2451 we make the fail stack and register vectors global.
2452 The fail stack, we grow to the maximum size when a regexp
2453 is compiled.
2454 The register vectors, we adjust in size each time we
2455 compile a regexp, according to the number of registers it needs. */
2459 /* Size with which the following vectors are currently allocated.
2460 That is so we can make them bigger as needed,
2461 but never make them smaller. */
2467 /* Make the register vectors big enough for NUM_REGS registers,
2468 but don't make them smaller. */
2470 static
2473 {
2475 {
2482 }
2483 }
2486 \f
2488 compile_stack,
2489 regnum_t regnum));
2491 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2492 Returns one of error codes defined in `regex.h', or zero for success.
2494 Assumes the `allocated' (and perhaps `buffer') and `translate'
2495 fields are set in BUFP on entry.
2497 If it succeeds, results are put in BUFP (if it returns an error, the
2498 contents of BUFP are undefined):
2499 `buffer' is the compiled pattern;
2500 `syntax' is set to SYNTAX;
2501 `used' is set to the length of the compiled pattern;
2502 `fastmap_accurate' is zero;
2503 `re_nsub' is the number of subexpressions in PATTERN;
2504 `not_bol' and `not_eol' are zero;
2506 The `fastmap' field is neither examined nor set. */
2508 /* Insert the `jump' from the end of last alternative to "here".
2509 The space for the jump has already been allocated. */
2510 #define FIXUP_ALT_JUMP() \
2511 do { \
2512 if (fixup_alt_jump) \
2513 STORE_JUMP (jump, fixup_alt_jump, b); \
2514 } while (0)
2517 /* Return, freeing storage we allocated. */
2518 #define FREE_STACK_RETURN(value) \
2519 do { \
2520 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2521 free (compile_stack.stack); \
2522 return value; \
2523 } while (0)
2525 static reg_errcode_t
2526 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2527 {
2528 /* We fetch characters from PATTERN here. */
2531 /* A random temporary spot in PATTERN. */
2534 /* Points to the end of the buffer, where we should append. */
2537 /* Keeps track of unclosed groups. */
2538 compile_stack_type compile_stack;
2540 /* Points to the current (ending) position in the pattern. */
2541 #ifdef AIX
2542 /* `const' makes AIX compiler fail. */
2544 #else
2546 #endif
2549 /* How to translate the characters in the pattern. */
2552 /* Address of the count-byte of the most recently inserted `exactn'
2553 command. This makes it possible to tell if a new exact-match
2554 character can be added to that command or if the character requires
2555 a new `exactn' command. */
2558 /* Address of start of the most recently finished expression.
2559 This tells, e.g., postfix * where to find the start of its
2560 operand. Reset at the beginning of groups and alternatives. */
2563 /* Address of beginning of regexp, or inside of last group. */
2566 /* Place in the uncompiled pattern (i.e., the {) to
2567 which to go back if the interval is invalid. */
2570 /* Address of the place where a forward jump should go to the end of
2571 the containing expression. Each alternative of an `or' -- except the
2572 last -- ends with a forward jump of this sort. */
2575 /* Work area for range table of charset. */
2578 /* If the object matched can contain multibyte characters. */
2581 /* If a target of matching can contain multibyte characters. */
2584 /* Nonzero if we have pushed down into a subpattern. */
2587 /* These hold the values of p, pattern, and pend from the main
2588 pattern when we have pushed into a subpattern. */
2593 #ifdef DEBUG
2594 debug++;
2597 {
2603 }
2606 /* Initialize the compile stack. */
2617 /* Initialize the pattern buffer. */
2623 /* Set `used' to zero, so that if we return an error, the pattern
2624 printer (for debugging) will think there's no pattern. We reset it
2625 at the end. */
2628 /* Always count groups, whether or not bufp->no_sub is set. */
2631 #if !defined emacs && !defined SYNTAX_TABLE
2632 /* Initialize the syntax table. */
2634 #endif
2637 {
2640 enough space. This loses if buffer's address is bogus, but
2641 that is the user's responsibility. */
2643 }
2644 else
2647 }
2651 }
2655 /* Loop through the uncompiled pattern until we're at the end. */
2657 {
2659 {
2660 /* If this is the end of an included regexp,
2661 pop back to the main regexp and try again. */
2663 {
2669 }
2670 /* If this is the end of the main regexp, we are done. */
2672 }
2677 {
2679 {
2682 /* If there's no special whitespace regexp, treat
2683 spaces normally. And don't try to do this recursively. */
2687 /* Peek past following spaces. */
2689 {
2692 p1++;
2693 }
2694 /* If the spaces are followed by a repetition op,
2695 treat them normally. */
2701 /* Replace the spaces with the whitespace regexp. */
2709 }
2712 {
2715 /* If context independent, it's an operator. */
2717 /* Otherwise, depends on what's come before. */
2720 else
2722 }
2727 {
2729 p == pend
2730 /* If context independent, it's an operator. */
2732 /* Otherwise, depends on what's next. */
2735 else
2737 }
2746 handle_plus:
2748 /* If there is no previous pattern... */
2750 {
2755 }
2757 {
2758 /* 1 means zero (many) matches is allowed. */
2762 /* If there is a sequence of repetition chars, collapse it
2763 down to just one (the right one). We can't combine
2764 interval operators with these because of, e.g., `a{2}*',
2765 which should only match an even number of `a's. */
2768 {
2772 else
2773 {
2776 }
2783 ;
2785 {
2790 else
2792 }
2793 else
2795 /* If we get here, we found another repeat character. */
2797 }
2799 /* Star, etc. applied to an empty pattern is equivalent
2800 to an empty pattern. */
2804 /* Now we know whether or not zero matches is allowed
2805 and also whether or not two or more matches is allowed. */
2807 {
2809 {
2812 re_opcode_t ofj =
2813 /* Check if the loop can match the empty string. */
2820 on_failure_keep_string_jump, we turn simple P+
2821 into PP* if P is simple. */
2829 }
2833 /* A + loop. */
2835 else
2836 /* Simple * loops can use on_failure_keep_string_jump
2837 depending on what follows. But since we don't know
2838 that yet, we leave the decision up to
2839 on_failure_jump_smart. */
2845 }
2846 else
2847 {
2848 /* A simple ? pattern. */
2853 }
2854 }
2860 {
2863 /* The non-greedy multiple match looks like
2864 a repeat..until: we only need a conditional jump
2865 at the end of the loop. */
2871 {
2872 /* The repeat...until naturally matches one or more.
2873 To also match zero times, we need to first jump to
2874 the end of the loop (its conditional jump). */
2877 }
2878 }
2879 else
2880 {
2881 /* non-greedy a?? */
2886 }
2887 }
2888 }
2900 {
2905 /* Ensure that we have enough space to push a charset: the
2906 opcode, the length count, and the bitset; 34 bytes in all. */
2911 /* We test `*p == '^' twice, instead of using an if
2912 statement, so we only need one BUF_PUSH. */
2915 p++;
2917 /* Remember the first position in the bracket expression. */
2920 /* Push the number of bytes in the bitmap. */
2923 /* Clear the whole map. */
2926 /* charset_not matches newline according to a syntax bit. */
2931 /* Read in characters and ranges, setting map bits. */
2933 {
2940 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2941 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2942 So the translation is done later in a loop. Example:
2943 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2946 /* \ might escape characters inside [...] and [^...]. */
2948 {
2953 }
2954 else
2955 {
2956 /* Could be the end of the bracket expression. If it's
2957 not (i.e., when the bracket expression is `[]' so
2958 far), the ']' character bit gets set way below. */
2961 }
2963 /* See if we're at the beginning of a possible character
2964 class. */
2968 {
2969 /* Leave room for the null. */
2977 /* If pattern is `[[:'. */
2981 {
2987 else
2988 /* This is in any case an invalid class name. */
2990 }
2993 /* If isn't a word bracketed by `[:' and `:]':
2994 undo the ending character, the letters, and
2995 leave the leading `:' and `[' (but set bits for
2996 them). */
2998 {
2999 re_wctype_t cc;
3007 /* Throw away the ] at the end of the character
3008 class. */
3013 #ifndef emacs
3016 {
3020 }
3022 /* Most character classes in a multibyte match
3023 just set a flag. Exceptions are is_blank,
3024 is_digit, is_cntrl, and is_xdigit, since
3025 they can only match ASCII characters. We
3026 don't need to handle them for multibyte.
3027 They are distinguished by a negative wctype. */
3029 /* Setup the gl_state object to its buffer-defined
3030 value. This hardcodes the buffer-global
3031 syntax-table for ASCII chars, while the other chars
3032 will obey syntax-table properties. It's not ideal,
3033 but it's the way it's been done until now. */
3037 {
3041 {
3050 }
3051 }
3052 SET_RANGE_TABLE_WORK_AREA_BIT
3055 /* In most cases the matching rule for char classes
3056 only uses the syntax table for multibyte chars,
3057 so that the content of the syntax-table it is not
3058 hardcoded in the range_table. SPACE and WORD are
3059 the two exceptions. */
3063 /* Repeat the loop. */
3065 }
3066 else
3067 {
3068 /* Go back to right after the "[:". */
3072 /* Because the `:' may starts the range, we
3073 can't simply set bit and repeat the loop.
3074 Instead, just set it to C and handle below. */
3076 }
3077 }
3080 {
3082 /* Discard the `-'. */
3085 /* Fetch the character which ends the range. */
3087 #ifdef emacs
3090 /* Treat the range from a multibyte character to
3091 raw-byte character as empty. */
3094 }
3095 else
3096 /* Range from C to C. */
3100 {
3103 /* Else, repeat the loop. */
3104 }
3105 else
3106 {
3107 #ifndef emacs
3108 /* Set the range into bitmap */
3110 {
3114 }
3117 {
3123 }
3125 {
3127 {
3132 }
3134 {
3136 }
3137 else
3138 {
3140 }
3141 }
3143 }
3144 }
3146 /* Discard any (non)matching list bytes that are all 0 at the
3147 end of the map. Decrease the map-length byte too. */
3152 /* Build real range table from work area. */
3155 {
3159 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3160 bytes for flags, two for COUNT, and three bytes for
3161 each character. */
3164 /* Indicate the existence of range table. */
3167 /* Store the character class flag bits into the range table.
3168 If not in emacs, these flag bits are always 0. */
3174 STORE_CHARACTER_AND_INCR
3176 }
3177 }
3184 else
3191 else
3198 else
3205 else
3212 else
3219 /* Do not translate the character after the \, so that we can
3220 distinguish, e.g., \B from \b, even if we normally would
3221 translate, e.g., B to b. */
3225 {
3230 handle_open:
3231 {
3235 {
3236 /* Look for a special (?...) construct */
3238 {
3241 {
3244 {
3247 /* An explicitly specified regnum must start
3248 with non-0. */
3255 /* Only (?:...) is supported right now. */
3257 }
3258 }
3259 }
3260 }
3270 /* Ideally, we'd want to check that the specified
3271 group can't have matched (i.e. all subgroups
3272 using the same regnum are in other branches of
3273 OR patterns), but we don't currently keep track
3274 of enough info to do that easily. */
3277 }
3278 else
3279 /* It's really shy. */
3283 {
3285 compile_stack_elt_t);
3289 }
3291 /* These are the values to restore when we hit end of this
3292 group. They are all relative offsets, so that if the
3293 whole pattern moves because of realloc, they will still
3294 be valid. */
3296 COMPILE_STACK_TOP.fixup_alt_jump
3301 /* Do not push a start_memory for groups beyond the last one
3302 we can represent in the compiled pattern. */
3311 /* If we've reached MAX_REGNUM groups, then this open
3312 won't actually generate any code, so we'll have to
3313 clear pending_exact explicitly. */
3316 }
3322 {
3325 else
3327 }
3329 handle_close:
3332 /* See similar code for backslashed left paren above. */
3334 {
3337 else
3339 }
3341 /* Since we just checked for an empty stack above, this
3342 ``can't happen''. */
3344 {
3345 /* We don't just want to restore into `regnum', because
3346 later groups should continue to be numbered higher,
3347 as in `(ab)c(de)' -- the second group is #2. */
3348 regnum_t regnum;
3352 fixup_alt_jump
3358 /* If we've reached MAX_REGNUM groups, then this open
3359 won't actually generate any code, so we'll have to
3360 clear pending_exact explicitly. */
3363 /* We're at the end of the group, so now we know how many
3364 groups were inside this one. */
3367 }
3374 handle_alt:
3378 /* Insert before the previous alternative a jump which
3379 jumps to this alternative if the former fails. */
3385 /* The alternative before this one has a jump after it
3386 which gets executed if it gets matched. Adjust that
3387 jump so it will jump to this alternative's analogous
3388 jump (put in below, which in turn will jump to the next
3389 (if any) alternative's such jump, etc.). The last such
3390 jump jumps to the correct final destination. A picture:
3391 _____ _____
3392 | | | |
3393 | v | v
3394 a | b | c
3396 If we are at `b', then fixup_alt_jump right now points to a
3397 three-byte space after `a'. We'll put in the jump, set
3398 fixup_alt_jump to right after `b', and leave behind three
3399 bytes which we'll fill in when we get to after `c'. */
3403 /* Mark and leave space for a jump after this alternative,
3404 to be filled in later either by next alternative or
3405 when know we're at the end of a series of alternatives. */
3416 /* If \{ is a literal. */
3418 /* If we're at `\{' and it's not the open-interval
3419 operator. */
3423 handle_interval:
3424 {
3425 /* If got here, then the syntax allows intervals. */
3427 /* At least (most) this many matches must be made. */
3436 else
3437 /* Interval such as `{1}' => match exactly once. */
3445 {
3451 }
3456 /* We just parsed a valid interval. */
3458 /* If it's invalid to have no preceding re. */
3460 {
3465 else
3467 }
3470 /* If the upper bound is zero, just drop the sub pattern
3471 altogether. */
3474 /* Just match it once: nothing to do here. */
3475 ;
3477 /* Otherwise, we have a nontrivial interval. When
3478 we're all done, the pattern will look like:
3479 set_number_at <jump count> <upper bound>
3480 set_number_at <succeed_n count> <lower bound>
3481 succeed_n <after jump addr> <succeed_n count>
3482 <body of loop>
3483 jump_n <succeed_n addr> <jump count>
3484 (The upper bound and `jump_n' are omitted if
3485 `upper_bound' is 1, though.) */
3486 else
3488 more at the end of the loop. */
3496 {
3497 /* A succeed_n that starts with 0 is really a
3498 a simple on_failure_jump_loop. */
3502 }
3503 else
3504 {
3505 /* Initialize lower bound of the `succeed_n', even
3506 though it will be set during matching by its
3507 attendant `set_number_at' (inserted next),
3508 because `re_compile_fastmap' needs to know.
3509 Jump to the `jump_n' we might insert below. */
3512 lower_bound);
3515 /* Code to initialize the lower bound. Insert
3516 before the `succeed_n'. The `5' is the last two
3517 bytes of this `set_number_at', plus 3 bytes of
3518 the following `succeed_n'. */
3522 }
3525 {
3526 /* A negative upper bound stands for infinity,
3527 in which case it degenerates to a plain jump. */
3530 }
3533 append a backward jump to the `succeed_n'
3534 that starts this interval.
3536 When we've reached this during matching,
3537 we'll have matched the interval once, so
3538 jump back only `upper_bound - 1' times. */
3543 /* The location we want to set is the second
3544 parameter of the `jump_n'; that is `b-2' as
3545 an absolute address. `laststart' will be
3546 the `set_number_at' we're about to insert;
3547 `laststart+3' the number to set, the source
3548 for the relative address. But we are
3549 inserting into the middle of the pattern --
3550 so everything is getting moved up by 5.
3551 Conclusion: (b - 2) - (laststart + 3) + 5,
3552 i.e., b - laststart.
3554 We insert this at the beginning of the loop
3555 so that if we fail during matching, we'll
3556 reinitialize the bounds. */
3560 }
3561 }
3564 }
3567 unfetch_interval:
3568 /* If an invalid interval, match the characters as literals. */
3573 /* normal_char and normal_backslash need `c'. */
3577 {
3580 }
3581 else
3584 #ifdef emacs
3585 /* There is no way to specify the before_dot and after_dot
3586 operators. rms says this is ok. --karl */
3654 else
3684 {
3685 regnum_t reg;
3693 /* Can't back reference to a subexp before its end. */
3699 }
3707 else
3711 normal_backslash:
3712 /* You might think it would be useful for \ to mean
3713 not to translate; but if we don't translate it
3714 it will never match anything. */
3716 }
3721 /* Expects the character in `c'. */
3722 normal_char:
3723 /* If no exactn currently being built. */
3726 /* If last exactn not at current position. */
3729 /* We have only one byte following the exactn for the count. */
3732 /* If followed by a repetition operator. */
3741 {
3742 /* Start building a new exactn. */
3748 }
3751 {
3755 {
3759 }
3760 else
3761 {
3764 {
3769 }
3772 }
3774 }
3781 /* Through the pattern now. */
3788 /* If we don't want backtracking, force success
3789 the first time we reach the end of the compiled pattern. */
3793 /* We have succeeded; set the length of the buffer. */
3796 #ifdef DEBUG
3798 {
3802 }
3803 debug--;
3806 #ifndef MATCH_MAY_ALLOCATE
3807 /* Initialize the failure stack to the largest possible stack. This
3808 isn't necessary unless we're trying to avoid calling alloca in
3809 the search and match routines. */
3810 {
3814 {
3818 fail_stack.stack
3821 else
3822 fail_stack.stack
3826 }
3829 }
3834 \f
3835 /* Subroutines for `regex_compile'. */
3837 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3839 static void
3841 {
3844 }
3847 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3849 static void
3851 {
3855 }
3858 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3859 for OP followed by two-byte integer parameter ARG. */
3861 static void
3863 {
3871 }
3874 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3876 static void
3878 {
3886 }
3889 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3890 after an alternative or a begin-subexpression. We assume there is at
3891 least one character before the ^. */
3893 static boolean
3895 {
3899 return
3900 /* After a subexpression? */
3902 /* After an alternative? */
3904 /* After a shy subexpression? */
3909 }
3912 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3913 at least one character after the $, i.e., `P < PEND'. */
3915 static boolean
3917 {
3922 return
3923 /* Before a subexpression? */
3926 /* Before an alternative? */
3929 }
3932 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3933 false if it's not. */
3935 static boolean
3937 {
3942 this_element--)
3947 }
3948 \f
3949 /* analyse_first.
3950 If fastmap is non-NULL, go through the pattern and fill fastmap
3951 with all the possible leading chars. If fastmap is NULL, don't
3952 bother filling it up (obviously) and only return whether the
3953 pattern could potentially match the empty string.
3955 Return 1 if p..pend might match the empty string.
3956 Return 0 if p..pend matches at least one char.
3957 Return -1 if fastmap was not updated accurately. */
3959 static int
3961 {
3965 /* If all elements for base leading-codes in fastmap is set, this
3966 flag is set true. */
3971 /* The loop below works as follows:
3972 - It has a working-list kept in the PATTERN_STACK and which basically
3973 starts by only containing a pointer to the first operation.
3974 - If the opcode we're looking at is a match against some set of
3975 chars, then we add those chars to the fastmap and go on to the
3976 next work element from the worklist (done via `break').
3977 - If the opcode is a control operator on the other hand, we either
3978 ignore it (if it's meaningless at this point, such as `start_memory')
3979 or execute it (if it's a jump). If the jump has several destinations
3980 (i.e. `on_failure_jump'), then we push the other destination onto the
3981 worklist.
3982 We guarantee termination by ignoring backward jumps (more or less),
3983 so that `p' is monotonically increasing. More to the point, we
3984 never set `p' (or push) anything `<= p1'. */
3987 {
3988 /* `p1' is used as a marker of how far back a `on_failure_jump'
3989 can go without being ignored. It is normally equal to `p'
3990 (which prevents any backward `on_failure_jump') except right
3991 after a plain `jump', to allow patterns such as:
3992 0: jump 10
3993 3..9: <body>
3994 10: on_failure_jump 3
3995 as used for the *? operator. */
3999 {
4005 /* If the first character has to match a backreference, that means
4006 that the group was empty (since it already matched). Since this
4007 is the only case that interests us here, we can assume that the
4008 backreference must match the empty string. */
4009 p++;
4013 /* Following are the cases which match a character. These end
4014 with `break'. */
4018 {
4019 /* If multibyte is nonzero, the first byte of each
4020 character is an ASCII or a leading code. Otherwise,
4021 each byte is a character. Thus, this works in both
4022 cases. */
4025 {
4026 /* For the case of matching this unibyte regex
4027 against multibyte, we must set a leading code of
4028 the corresponding multibyte character. */
4032 }
4033 }
4038 /* We could put all the chars except for \n (and maybe \0)
4039 but we don't bother since it is generally not worth it. */
4046 {
4047 /* Chars beyond end of bitmap are possible matches. */
4051 }
4053 /* Fallthrough */
4062 #ifdef emacs
4064 which doesn't match the specified set of characters. */
4065 not
4066 ||
4067 /* If we can match a character class, we can match any
4068 multibyte characters. */
4072 {
4074 {
4079 }
4080 }
4084 {
4085 /* Set fastmap[I] to 1 where I is a leading code of each
4086 multibyte character in the range table. */
4090 /* Make P points the range table. `+ 2' is to skip flag
4091 bits for a character class. */
4094 /* Extract the number of ranges in range table into COUNT. */
4097 {
4098 /* Extract the start and end of each range. */
4106 }
4107 }
4108 #endif
4114 #ifndef emacs
4122 /* This match depends on text properties. These end with
4123 aborting optimizations. */
4135 /* Any leading code can possibly start a character which
4136 has or doesn't has the specified category. */
4138 {
4143 }
4146 /* All cases after this match the empty string. These end with
4147 `continue'. */
4170 /* Backward jumps can only go back to code that we've already
4171 visited. `re_compile' should make sure this is true. */
4175 {
4181 p++;
4185 };
4186 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4187 to jump back to "just after here". */
4188 /* Fallthrough */
4198 else
4200 We first go down the "straight" path so as to minimize
4201 stack usage when going through alternatives. */
4205 }
4210 /* This code simply does not properly handle forward jump_n. */
4213 /* jump_n can either jump or fall through. The (backward) jump
4214 case has already been handled, so we only need to look at the
4215 fallthrough case. */
4219 /* If N == 0, it should be an on_failure_jump_loop instead. */
4222 /* We only care about one iteration of the loop, so we don't
4223 need to consider the case where this behaves like an
4224 on_failure_jump. */
4243 /* Getting here means we have found the possible starting
4244 characters for one path of the pattern -- and that the empty
4245 string does not match. We need not follow this path further. */
4249 /* We reached the end without matching anything. */
4253 \f
4254 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4255 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4256 characters can start a string that matches the pattern. This fastmap
4257 is used by re_search to skip quickly over impossible starting points.
4259 Character codes above (1 << BYTEWIDTH) are not represented in the
4260 fastmap, but the leading codes are represented. Thus, the fastmap
4261 indicates which character sets could start a match.
4263 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4264 area as BUFP->fastmap.
4266 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4267 the pattern buffer.
4269 Returns 0 if we succeed, -2 if an internal error. */
4271 int
4273 {
4287 \f
4288 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4289 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4290 this memory for recording register information. STARTS and ENDS
4291 must be allocated using the malloc library routine, and must each
4292 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4294 If NUM_REGS == 0, then subsequent matches should allocate their own
4295 register data.
4297 Unless this function is called, the first search or match using
4298 PATTERN_BUFFER will allocate its own register data, without
4299 freeing the old data. */
4301 void
4302 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4303 {
4305 {
4310 }
4311 else
4312 {
4316 }
4317 }
4319 \f
4320 /* Searching routines. */
4322 /* Like re_search_2, below, but only one string is specified, and
4323 doesn't let you say where to stop matching. */
4325 int
4326 re_search (struct re_pattern_buffer *bufp, const char *string, int size, int startpos, int range, struct re_registers *regs)
4327 {
4330 }
4333 /* Head address of virtual concatenation of string. */
4334 #define HEAD_ADDR_VSTRING(P) \
4335 (((P) >= size1 ? string2 : string1))
4337 /* End address of virtual concatenation of string. */
4338 #define STOP_ADDR_VSTRING(P) \
4339 (((P) >= size1 ? string2 + size2 : string1 + size1))
4341 /* Address of POS in the concatenation of virtual string. */
4342 #define POS_ADDR_VSTRING(POS) \
4343 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4345 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4346 virtual concatenation of STRING1 and STRING2, starting first at index
4347 STARTPOS, then at STARTPOS + 1, and so on.
4349 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4351 RANGE is how far to scan while trying to match. RANGE = 0 means try
4352 only at STARTPOS; in general, the last start tried is STARTPOS +
4353 RANGE.
4355 In REGS, return the indices of the virtual concatenation of STRING1
4356 and STRING2 that matched the entire BUFP->buffer and its contained
4357 subexpressions.
4359 Do not consider matching one past the index STOP in the virtual
4360 concatenation of STRING1 and STRING2.
4362 We return either the position in the strings at which the match was
4363 found, -1 if no match, or -2 if error (such as failure
4364 stack overflow). */
4366 int
4367 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, int size1, const char *str2, int size2, int startpos, int range, struct re_registers *regs, int stop)
4368 {
4376 boolean anchored_start;
4377 /* Nonzero if we are searching multibyte string. */
4380 /* Check for out-of-range STARTPOS. */
4384 /* Fix up RANGE if it might eventually take us outside
4385 the virtual concatenation of STRING1 and STRING2.
4386 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4392 /* If the search isn't to be a backwards one, don't waste time in a
4393 search for a pattern anchored at beginning of buffer. */
4395 {
4398 else
4400 }
4402 #ifdef emacs
4403 /* In a forward search for something that starts with \=.
4404 don't keep searching past point. */
4406 {
4410 }
4413 /* Update the fastmap now if not correct already. */
4417 /* See whether the pattern is anchored. */
4420 #ifdef emacs
4422 {
4426 }
4427 #endif
4429 /* Loop through the string, looking for a place to start matching. */
4431 {
4432 /* If the pattern is anchored,
4433 skip quickly past places we cannot match.
4434 We don't bother to treat startpos == 0 specially
4435 because that case doesn't repeat. */
4437 {
4442 }
4444 /* If a fastmap is supplied, skip quickly over characters that
4445 cannot be the start of a match. If the pattern can match the
4446 null string, however, we don't need to skip characters; we want
4447 the first null string. */
4449 {
4456 {
4463 /* Written out as an if-else to avoid testing `translate'
4464 inside the loop. */
4466 {
4469 {
4479 }
4480 else
4482 {
4493 d++;
4494 range--;
4495 }
4496 }
4497 else
4498 {
4501 {
4509 }
4510 else
4512 {
4513 d++;
4514 range--;
4515 }
4516 }
4518 }
4520 {
4522 {
4527 }
4528 else
4529 {
4540 }
4541 }
4542 }
4544 /* If can't match the null string, and that's all we have left, fail. */
4558 advance:
4562 {
4563 /* Update STARTPOS to the next character boundary. */
4565 {
4574 }
4575 else
4576 {
4577 range--;
4578 startpos++;
4579 }
4580 }
4581 else
4582 {
4583 range++;
4584 startpos--;
4586 /* Update STARTPOS to the previous character boundary. */
4588 {
4593 /* Find the head of multibyte form. */
4600 }
4601 }
4602 }
4606 \f
4607 /* Declarations and macros for re_match_2. */
4611 RE_TRANSLATE_TYPE translate,
4614 /* This converts PTR, a pointer into one of the search strings `string1'
4615 and `string2' into an offset from the beginning of that string. */
4616 #define POINTER_TO_OFFSET(ptr) \
4617 (FIRST_STRING_P (ptr) \
4618 ? ((regoff_t) ((ptr) - string1)) \
4619 : ((regoff_t) ((ptr) - string2 + size1)))
4621 /* Call before fetching a character with *d. This switches over to
4622 string2 if necessary.
4623 Check re_match_2_internal for a discussion of why end_match_2 might
4624 not be within string2 (but be equal to end_match_1 instead). */
4625 #define PREFETCH() \
4626 while (d == dend) \
4627 { \
4629 if (dend == end_match_2) \
4630 goto fail; \
4632 d = string2; \
4633 dend = end_match_2; \
4634 }
4636 /* Call before fetching a char with *d if you already checked other limits.
4637 This is meant for use in lookahead operations like wordend, etc..
4638 where we might need to look at parts of the string that might be
4639 outside of the LIMITs (i.e past `stop'). */
4640 #define PREFETCH_NOLIMIT() \
4641 if (d == end1) \
4642 { \
4643 d = string2; \
4644 dend = end_match_2; \
4645 } \
4647 /* Test if at very beginning or at very end of the virtual concatenation
4648 of `string1' and `string2'. If only one string, it's `string2'. */
4649 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4650 #define AT_STRINGS_END(d) ((d) == end2)
4653 /* Test if D points to a character which is word-constituent. We have
4654 two special cases to check for: if past the end of string1, look at
4655 the first character in string2; and if before the beginning of
4656 string2, look at the last character in string1. */
4657 #define WORDCHAR_P(d) \
4658 (SYNTAX ((d) == end1 ? *string2 \
4659 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4660 == Sword)
4662 /* Disabled due to a compiler bug -- see comment at case wordbound */
4664 /* The comment at case wordbound is following one, but we don't use
4665 AT_WORD_BOUNDARY anymore to support multibyte form.
4667 The DEC Alpha C compiler 3.x generates incorrect code for the
4668 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4669 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4670 macro and introducing temporary variables works around the bug. */
4672 #if 0
4673 /* Test if the character before D and the one at D differ with respect
4674 to being word-constituent. */
4675 #define AT_WORD_BOUNDARY(d) \
4676 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4677 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4678 #endif
4680 /* Free everything we malloc. */
4681 #ifdef MATCH_MAY_ALLOCATE
4682 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4683 # define FREE_VARIABLES() \
4684 do { \
4685 REGEX_FREE_STACK (fail_stack.stack); \
4686 FREE_VAR (regstart); \
4687 FREE_VAR (regend); \
4688 FREE_VAR (best_regstart); \
4689 FREE_VAR (best_regend); \
4690 } while (0)
4691 #else
4695 \f
4696 /* Optimization routines. */
4698 /* If the operation is a match against one or more chars,
4699 return a pointer to the next operation, else return NULL. */
4702 {
4704 {
4715 {
4720 }
4721 else
4727 #ifdef emacs
4731 p++;
4736 }
4738 }
4741 /* Jump over non-matching operations. */
4744 {
4747 {
4749 {
4762 }
4763 }
4766 }
4768 /* Non-zero if "p1 matches something" implies "p2 fails". */
4769 static int
4771 {
4772 re_opcode_t op2;
4779 /* Skip over open/close-group commands.
4780 If what follows this loop is a ...+ construct,
4781 look at what begins its body, since we will have to
4782 match at least one of that. */
4784 /* The same skip can be done for p1, except that this function
4785 is only used in the case where p1 is a simple match operator. */
4786 /* p1 = skip_noops (p1, pend); */
4794 {
4797 /* If we're at the end of the pattern, we can change. */
4799 {
4802 }
4807 {
4808 register re_wchar_t c
4813 {
4815 {
4818 }
4819 }
4823 {
4826 /* Test if C is listed in charset (or charset_not)
4827 at `p1'. */
4829 {
4833 }
4837 /* `not' is equal to 1 if c would match, which means
4838 that we can't change to pop_failure_jump. */
4840 {
4843 }
4844 }
4847 {
4850 }
4851 }
4855 {
4857 /* Reuse the code above. */
4860 /* It is hard to list up all the character in charset
4861 P2 if it includes multibyte character. Give up in
4862 such case. */
4864 {
4865 /* Now, we are sure that P2 has no range table.
4866 So, for the size of bitmap in P2, `p2[1]' is
4867 enough. But P1 may have range table, so the
4868 size of bitmap table of P1 is extracted by
4869 using macro `CHARSET_BITMAP_SIZE'.
4871 In a multibyte case, we know that all the character
4872 listed in P2 is ASCII. In a unibyte case, P1 has only a
4873 bitmap table. So, in both cases, it is enough to test
4874 only the bitmap table of P1. */
4877 {
4879 /* We win if the charset inside the loop
4880 has no overlap with the one after the loop. */
4884 idx++)
4890 {
4893 }
4894 }
4896 {
4898 /* We win if the charset_not inside the loop lists
4899 every character listed in the charset after. */
4907 {
4910 }
4911 }
4912 }
4913 }
4918 {
4921 /* Reuse the code above. */
4924 /* When we have two charset_not, it's very unlikely that
4925 they don't overlap. The union of the two sets of excluded
4926 chars should cover all possible chars, which, as a matter of
4927 fact, is virtually impossible in multibyte buffers. */
4929 }
4953 #ifdef emacs
4961 ;
4962 }
4964 /* Safe default. */
4966 }
4968 \f
4969 /* Matching routines. */
4972 /* re_match is like re_match_2 except it takes only a single string. */
4974 int
4977 {
4981 }
4985 #ifdef emacs
4986 /* In Emacs, this is the string or buffer in which we
4987 are matching. It is used for looking up syntax properties. */
4988 Lisp_Object re_match_object;
4989 #endif
4991 /* re_match_2 matches the compiled pattern in BUFP against the
4992 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4993 and SIZE2, respectively). We start matching at POS, and stop
4994 matching at STOP.
4996 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4997 store offsets for the substring each group matched in REGS. See the
4998 documentation for exactly how many groups we fill.
5000 We return -1 if no match, -2 if an internal error (such as the
5001 failure stack overflowing). Otherwise, we return the length of the
5002 matched substring. */
5004 int
5005 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, const char *string2, int size2, int pos, struct re_registers *regs, int stop)
5006 {
5009 #ifdef emacs
5014 #endif
5020 }
5024 /* This is a separate function so that we can force an alloca cleanup
5025 afterwards. */
5026 static int
5027 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1, int size1, const re_char *string2, int size2, int pos, struct re_registers *regs, int stop)
5028 {
5029 /* General temporaries. */
5034 /* Just past the end of the corresponding string. */
5037 /* Pointers into string1 and string2, just past the last characters in
5038 each to consider matching. */
5041 /* Where we are in the data, and the end of the current string. */
5044 /* Used sometimes to remember where we were before starting matching
5045 an operator so that we can go back in case of failure. This "atomic"
5046 behavior of matching opcodes is indispensable to the correctness
5047 of the on_failure_keep_string_jump optimization. */
5050 /* Where we are in the pattern, and the end of the pattern. */
5054 /* We use this to map every character in the string. */
5057 /* Nonzero if BUFP is setup from a multibyte regex. */
5060 /* Nonzero if STRING1/STRING2 are multibyte. */
5063 /* Failure point stack. Each place that can handle a failure further
5064 down the line pushes a failure point on this stack. It consists of
5065 regstart, and regend for all registers corresponding to
5066 the subexpressions we're currently inside, plus the number of such
5067 registers, and, finally, two char *'s. The first char * is where
5068 to resume scanning the pattern; the second one is where to resume
5069 scanning the strings. */
5071 fail_stack_type fail_stack;
5072 #endif
5073 #ifdef DEBUG
5075 #endif
5077 #if defined REL_ALLOC && defined REGEX_MALLOC
5078 /* This holds the pointer to the failure stack, when
5079 it is allocated relocatably. */
5081 #endif
5083 /* We fill all the registers internally, independent of what we
5084 return, for use in backreferences. The number here includes
5085 an element for register zero. */
5088 /* Information on the contents of registers. These are pointers into
5089 the input strings; they record just what was matched (on this
5090 attempt) by a subexpression part of the pattern, that is, the
5091 regnum-th regstart pointer points to where in the pattern we began
5092 matching and the regnum-th regend points to right after where we
5093 stopped matching the regnum-th subexpression. (The zeroth register
5094 keeps track of what the whole pattern matches.) */
5097 #endif
5099 /* The following record the register info as found in the above
5100 variables when we find a match better than any we've seen before.
5101 This happens as we backtrack through the failure points, which in
5102 turn happens only if we have not yet matched the entire string. */
5106 #endif
5108 /* Logically, this is `best_regend[0]'. But we don't want to have to
5109 allocate space for that if we're not allocating space for anything
5110 else (see below). Also, we never need info about register 0 for
5111 any of the other register vectors, and it seems rather a kludge to
5112 treat `best_regend' differently than the rest. So we keep track of
5113 the end of the best match so far in a separate variable. We
5114 initialize this to NULL so that when we backtrack the first time
5115 and need to test it, it's not garbage. */
5118 #ifdef DEBUG
5119 /* Counts the total number of registers pushed. */
5121 #endif
5127 #ifdef MATCH_MAY_ALLOCATE
5128 /* Do not bother to initialize all the register variables if there are
5129 no groups in the pattern, as it takes a fair amount of time. If
5130 there are groups, we include space for register 0 (the whole
5131 pattern), even though we never use it, since it simplifies the
5132 array indexing. We should fix this. */
5134 {
5141 {
5144 }
5145 }
5146 else
5147 {
5148 /* We must initialize all our variables to NULL, so that
5149 `FREE_VARIABLES' doesn't try to free them. */
5151 }
5154 /* The starting position is bogus. */
5156 {
5159 }
5161 /* Initialize subexpression text positions to -1 to mark ones that no
5162 start_memory/stop_memory has been seen for. Also initialize the
5163 register information struct. */
5167 /* We move `string1' into `string2' if the latter's empty -- but not if
5168 `string1' is null. */
5170 {
5175 }
5179 /* `p' scans through the pattern as `d' scans through the data.
5180 `dend' is the end of the input string that `d' points within. `d'
5181 is advanced into the following input string whenever necessary, but
5182 this happens before fetching; therefore, at the beginning of the
5183 loop, `d' can be pointing at the end of a string, but it cannot
5184 equal `string2'. */
5186 {
5187 /* Only match within string2. */
5191 }
5192 else
5193 {
5195 {
5196 /* Only match within string1. */
5198 /* BEWARE!
5199 When we reach end_match_1, PREFETCH normally switches to string2.
5200 But in the present case, this means that just doing a PREFETCH
5201 makes us jump from `stop' to `gap' within the string.
5202 What we really want here is for the search to stop as
5203 soon as we hit end_match_1. That's why we set end_match_2
5204 to end_match_1 (since PREFETCH fails as soon as we hit
5205 end_match_2). */
5207 }
5208 else
5210 moving `d' from end1 to string2 will not prevent the d == dend
5211 check from catching the end of string. */
5214 }
5217 }
5225 /* This loops over pattern commands. It exits by returning from the
5226 function if the match is complete, or it drops through if the match
5227 fails at this starting point in the input data. */
5229 {
5236 /* If we haven't matched the entire string, and we want the
5237 longest match, try backtracking. */
5239 {
5240 /* 1 if this match ends in the same string (string1 or string2)
5241 as the best previous match. */
5244 /* 1 if this match is the best seen so far. */
5245 boolean best_match_p;
5247 /* AIX compiler got confused when this was combined
5248 with the previous declaration. */
5251 else
5259 /* If exceeds best match so far, save it. */
5261 {
5268 {
5271 }
5272 }
5274 }
5276 /* If no failure points, don't restore garbage. And if
5277 last match is real best match, don't restore second
5278 best one. */
5280 {
5281 restore_best_regs:
5282 /* Restore best match. It may happen that `dend ==
5283 end_match_1' while the restored d is in string2.
5284 For example, the pattern `x.*y.*z' against the
5285 strings `x-' and `y-z-', if the two strings are
5286 not consecutive in memory. */
5294 {
5297 }
5298 }
5301 succeed_label:
5304 /* If caller wants register contents data back, do it. */
5306 {
5307 /* Have the register data arrays been allocated? */
5310 extra element beyond `num_regs' for the `-1' marker
5311 GNU code uses. */
5316 {
5319 }
5321 }
5324 allocated, reallocate them. If we need fewer, just
5325 leave it alone. */
5327 {
5332 {
5335 }
5336 }
5337 }
5338 else
5339 {
5340 /* These braces fend off a "empty body in an else-statement"
5341 warning under GCC when assert expands to nothing. */
5343 }
5345 /* Convert the pointer data in `regstart' and `regend' to
5346 indices. Register zero has to be set differently,
5347 since we haven't kept track of any info for it. */
5349 {
5352 }
5354 /* Go through the first `min (num_regs, regs->num_regs)'
5355 registers, since that is all we initialized. */
5357 {
5360 else
5361 {
5366 }
5367 }
5369 /* If the regs structure we return has more elements than
5370 were in the pattern, set the extra elements to -1. If
5371 we (re)allocated the registers, this is the case,
5372 because we always allocate enough to have at least one
5373 -1 at the end. */
5389 }
5391 /* Otherwise match next pattern command. */
5393 {
5394 /* Ignore these. Used to ignore the n of succeed_n's which
5395 currently have n == 0. */
5404 /* Match the next n pattern characters exactly. The following
5405 byte in the pattern defines n, and the n bytes after that
5406 are the characters to match. */
5411 /* Remember the start point to rollback upon failure. */
5414 #ifndef emacs
5415 /* This is written out as an if-else so we don't waste time
5416 testing `translate' inside the loop. */
5418 do
5419 {
5422 {
5425 }
5426 d++;
5427 }
5429 else
5430 do
5431 {
5434 {
5437 }
5438 }
5441 /* The cost of testing `translate' is comparatively small. */
5443 do
5444 {
5451 else
5452 {
5455 }
5459 {
5462 }
5467 }
5469 else
5470 do
5471 {
5477 {
5480 }
5481 else
5482 {
5485 }
5488 {
5493 }
5494 else
5497 {
5500 }
5502 d++;
5503 }
5505 #endif
5509 /* Match any character except possibly a newline or a null. */
5511 {
5513 re_wchar_t buf_ch;
5519 target_multibyte);
5530 }
5536 {
5541 /* Start of actual range_table, or end of bitmap if there is no
5542 range table. */
5545 /* Nonzero if there is a range table. */
5548 /* Number of ranges of range table. This is not included
5549 in the initial byte-length of the command. */
5552 /* Whether matching against a unibyte character. */
5560 {
5563 }
5568 {
5574 {
5577 }
5578 }
5579 else
5580 {
5584 {
5588 {
5591 }
5592 }
5593 else
5595 }
5599 /* Cast to `unsigned' instead of `unsigned char' in
5600 case the bit list is a full 32 bytes long. */
5604 }
5605 #ifdef emacs
5607 {
5617 else
5619 }
5624 else
5631 }
5634 /* The beginning of a group is represented by start_memory.
5635 The argument is the register number. The text
5636 matched within the group is recorded (in the internal
5637 registers data structure) under the register number. */
5641 /* In case we need to undo this operation (via backtracking). */
5648 /* Move past the register number and inner group count. */
5653 /* The stop_memory opcode represents the end of a group. Its
5654 argument is the same as start_memory's: the register number. */
5659 /* Strictly speaking, there should be code such as:
5661 assert (REG_UNSET (regend[*p]));
5662 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5664 But the only info to be pushed is regend[*p] and it is known to
5665 be UNSET, so there really isn't anything to push.
5666 Not pushing anything, on the other hand deprives us from the
5667 guarantee that regend[*p] is UNSET since undoing this operation
5668 will not reset its value properly. This is not important since
5669 the value will only be read on the next start_memory or at
5670 the very end and both events can only happen if this stop_memory
5671 is *not* undone. */
5676 /* Move past the register number and the inner group count. */
5681 /* \<digit> has been turned into a `duplicate' command which is
5682 followed by the numeric value of <digit> as the register number. */
5684 {
5689 /* Can't back reference a group which we've never matched. */
5693 /* Where in input to try to start matching. */
5696 /* Remember the start point to rollback upon failure. */
5699 /* Where to stop matching; if both the place to start and
5700 the place to stop matching are in the same string, then
5701 set to the place to stop, otherwise, for now have to use
5702 the end of the first string. */
5708 {
5709 /* If necessary, advance to next segment in register
5710 contents. */
5712 {
5716 /* End of string1 => advance to string2. */
5719 }
5720 /* At end of register contents => success */
5723 /* If necessary, advance to next segment in data. */
5726 /* How many characters left in this segment to match. */
5729 /* Want how many consecutive characters we can match in
5730 one shot, so, if necessary, adjust the count. */
5734 /* Compare that many; failure if mismatch, else move
5735 past them. */
5739 {
5742 }
5744 }
5745 }
5749 /* begline matches the empty string at the beginning of the string
5750 (unless `not_bol' is set in `bufp'), and after newlines. */
5755 {
5757 }
5758 else
5759 {
5764 }
5765 /* In all other cases, we fail. */
5769 /* endline is the dual of begline. */
5774 {
5776 }
5777 else
5778 {
5782 }
5786 /* Match at the very beginning of the data. */
5794 /* Match at the very end of the data. */
5802 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5803 pushes NULL as the value for the string on the stack. Then
5804 `POP_FAILURE_POINT' will keep the current value for the
5805 string, instead of restoring it. To see why, consider
5806 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5807 then the . fails against the \n. But the next thing we want
5808 to do is match the \n against the \n; if we restored the
5809 string value, we would be back at the foo.
5811 Because this is used only in specific cases, we don't need to
5812 check all the things that `on_failure_jump' does, to make
5813 sure the right things get saved on the stack. Hence we don't
5814 share its code. The only reason to push anything on the
5815 stack at all is that otherwise we would have to change
5816 `anychar's code to do something besides goto fail in this
5817 case; that seems worse than this. */
5826 /* A nasty loop is introduced by the non-greedy *? and +?.
5827 With such loops, the stack only ever contains one failure point
5828 at a time, so that a plain on_failure_jump_loop kind of
5829 cycle detection cannot work. Worse yet, such a detection
5830 can not only fail to detect a cycle, but it can also wrongly
5831 detect a cycle (between different instantiations of the same
5832 loop).
5833 So the method used for those nasty loops is a little different:
5834 We use a special cycle-detection-stack-frame which is pushed
5835 when the on_failure_jump_nastyloop failure-point is *popped*.
5836 This special frame thus marks the beginning of one iteration
5837 through the loop and we can hence easily check right here
5838 whether something matched between the beginning and the end of
5839 the loop. */
5846 {
5850 /* If there's a cycle, just continue without pushing
5851 this failure point. The failure point is the "try again"
5852 option, which shouldn't be tried.
5853 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5855 }
5858 /* Simple loop detecting on_failure_jump: just check on the
5859 failure stack if the same spot was already hit earlier. */
5861 on_failure:
5865 {
5869 /* If there's a cycle, get out of the loop, as if the matching
5870 had failed. We used to just `goto fail' here, but that was
5871 aborting the search a bit too early: we want to keep the
5872 empty-loop-match and keep matching after the loop.
5873 We want (x?)*y\1z to match both xxyz and xxyxz. */
5875 else
5877 }
5881 /* Uses of on_failure_jump:
5883 Each alternative starts with an on_failure_jump that points
5884 to the beginning of the next alternative. Each alternative
5885 except the last ends with a jump that in effect jumps past
5886 the rest of the alternatives. (They really jump to the
5887 ending jump of the following alternative, because tensioning
5888 these jumps is a hassle.)
5890 Repeats start with an on_failure_jump that points past both
5891 the repetition text and either the following jump or
5892 pop_failure_jump back to this on_failure_jump. */
5901 /* This operation is used for greedy *.
5902 Compare the beginning of the repeat with what in the
5903 pattern follows its end. If we can establish that there
5904 is nothing that they would both match, i.e., that we
5905 would have to backtrack because of (as in, e.g., `a*a')
5906 then we can use a non-backtracking loop based on
5907 on_failure_keep_string_jump instead of on_failure_jump. */
5912 {
5914 /* Here, we discard `const', making re_match non-reentrant. */
5919 instruction once it's been changed. */
5923 /* Ensure this is a indeed the trivial kind of loop
5924 we are expecting. */
5929 {
5930 /* Use a fast `on_failure_keep_string_jump' loop. */
5934 }
5935 else
5936 {
5937 /* Default to a safe `on_failure_jump' loop. */
5940 }
5942 }
5945 /* Unconditionally jump (without popping any failure points). */
5947 unconditional_jump:
5948 IMMEDIATE_QUIT_CHECK;
5956 /* Have to succeed matching what follows at least n times.
5957 After that, handle like `on_failure_jump'. */
5959 /* Signedness doesn't matter since we only compare MCNT to 0. */
5963 /* Originally, mcnt is how many times we HAVE to succeed. */
5965 {
5966 /* Here, we discard `const', making re_match non-reentrant. */
5968 mcnt--;
5971 }
5972 else
5973 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5978 /* Signedness doesn't matter since we only compare MCNT to 0. */
5982 /* Originally, this is how many times we CAN jump. */
5984 {
5985 /* Here, we discard `const', making re_match non-reentrant. */
5987 mcnt--;
5990 }
5991 /* If don't have to jump any more, skip over the rest of command. */
5992 else
5997 {
6002 /* Here, we discard `const', making re_match non-reentrant. */
6004 /* Signedness doesn't matter since we only copy MCNT's bits . */
6009 }
6016 /* We SUCCEED (or FAIL) in one of the following cases: */
6018 /* Case 1: D is at the beginning or the end of string. */
6021 else
6022 {
6023 /* C1 is the character before D, S1 is the syntax of C1, C2
6024 is the character at D, and S2 is the syntax of C2. */
6028 #ifdef emacs
6032 #endif
6035 #ifdef emacs
6037 #endif
6044 /* Case 3: Both of S1 and S2 are Sword, and macro
6045 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6048 }
6051 else
6057 /* We FAIL in one of the following cases: */
6059 /* Case 1: D is at the end of string. */
6062 else
6063 {
6064 /* C1 is the character before D, S1 is the syntax of C1, C2
6065 is the character at D, and S2 is the syntax of C2. */
6069 #ifdef emacs
6073 #endif
6078 /* Case 2: S2 is not Sword. */
6082 /* Case 3: D is not at the beginning of string ... */
6084 {
6086 #ifdef emacs
6088 #endif
6091 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6092 returns 0. */
6095 }
6096 }
6102 /* We FAIL in one of the following cases: */
6104 /* Case 1: D is at the beginning of string. */
6107 else
6108 {
6109 /* C1 is the character before D, S1 is the syntax of C1, C2
6110 is the character at D, and S2 is the syntax of C2. */
6114 #ifdef emacs
6118 #endif
6122 /* Case 2: S1 is not Sword. */
6126 /* Case 3: D is not at the end of string ... */
6128 {
6131 #ifdef emacs
6133 #endif
6136 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6137 returns 0. */
6140 }
6141 }
6147 /* We FAIL in one of the following cases: */
6149 /* Case 1: D is at the end of string. */
6152 else
6153 {
6154 /* C1 is the character before D, S1 is the syntax of C1, C2
6155 is the character at D, and S2 is the syntax of C2. */
6158 #ifdef emacs
6162 #endif
6167 /* Case 2: S2 is neither Sword nor Ssymbol. */
6171 /* Case 3: D is not at the beginning of string ... */
6173 {
6175 #ifdef emacs
6177 #endif
6180 /* ... and S1 is Sword or Ssymbol. */
6183 }
6184 }
6190 /* We FAIL in one of the following cases: */
6192 /* Case 1: D is at the beginning of string. */
6195 else
6196 {
6197 /* C1 is the character before D, S1 is the syntax of C1, C2
6198 is the character at D, and S2 is the syntax of C2. */
6201 #ifdef emacs
6205 #endif
6209 /* Case 2: S1 is neither Ssymbol nor Sword. */
6213 /* Case 3: D is not at the end of string ... */
6215 {
6218 #ifdef emacs
6220 #endif
6223 /* ... and S2 is Sword or Ssymbol. */
6226 }
6227 }
6236 #ifdef emacs
6237 {
6241 }
6242 #endif
6243 {
6245 re_wchar_t c;
6251 }
6254 #ifdef emacs
6279 {
6281 re_wchar_t c;
6287 }
6294 }
6298 /* We goto here if a matching operation fails. */
6299 fail:
6300 IMMEDIATE_QUIT_CHECK;
6302 {
6304 /* A restart point is known. Restore to that state. */
6308 {
6316 /* Fallthrough */
6322 continue_failure_jump:
6328 /* A special frame used for nastyloops. */
6333 }
6339 }
6340 else
6351 \f
6352 /* Subroutine definitions for re_match_2. */
6354 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6355 bytes; nonzero otherwise. */
6357 static int
6360 {
6365 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6366 different lengths, but relying on a single `len' would break this. -sm */
6368 {
6380 }
6386 }
6387 \f
6388 /* Entry points for GNU code. */
6390 /* re_compile_pattern is the GNU regular expression compiler: it
6391 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6392 Returns 0 if the pattern was valid, otherwise an error string.
6394 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6395 are set in BUFP on entry.
6397 We call regex_compile to do the actual compilation. */
6401 {
6402 reg_errcode_t ret;
6404 /* GNU code is written to assume at least RE_NREGS registers will be set
6405 (and at least one extra will be -1). */
6408 /* And GNU code determines whether or not to get register information
6409 by passing null for the REGS argument to re_match, etc., not by
6410 setting no_sub. */
6418 }
6420 \f
6421 /* Entry points compatible with 4.2 BSD regex library. We don't define
6422 them unless specifically requested. */
6424 #if defined _REGEX_RE_COMP || defined _LIBC
6426 /* BSD has one and only one pattern buffer. */
6430 # ifdef _LIBC
6431 /* Make these definitions weak in libc, so POSIX programs can redefine
6432 these names if they don't use our functions, and still use
6433 regcomp/regexec below without link errors. */
6434 weak_function
6435 # endif
6438 {
6439 reg_errcode_t ret;
6442 {
6444 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6447 }
6450 {
6453 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6459 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6461 }
6463 /* Since `re_exec' always passes NULL for the `regs' argument, we
6464 don't need to initialize the pattern buffer fields which affect it. */
6471 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6473 }
6476 int
6477 # ifdef _LIBC
6478 weak_function
6479 # endif
6482 {
6484 return
6486 }
6488 \f
6489 /* POSIX.2 functions. Don't define these for Emacs. */
6491 #ifndef emacs
6493 /* regcomp takes a regular expression as a string and compiles it.
6495 PREG is a regex_t *. We do not expect any fields to be initialized,
6496 since POSIX says we shouldn't. Thus, we set
6498 `buffer' to the compiled pattern;
6499 `used' to the length of the compiled pattern;
6500 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6501 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6502 RE_SYNTAX_POSIX_BASIC;
6503 `fastmap' to an allocated space for the fastmap;
6504 `fastmap_accurate' to zero;
6505 `re_nsub' to the number of subexpressions in PATTERN.
6507 PATTERN is the address of the pattern string.
6509 CFLAGS is a series of bits which affect compilation.
6511 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6512 use POSIX basic syntax.
6514 If REG_NEWLINE is set, then . and [^...] don't match newline.
6515 Also, regexec will try a match beginning after every newline.
6517 If REG_ICASE is set, then we considers upper- and lowercase
6518 versions of letters to be equivalent when matching.
6520 If REG_NOSUB is set, then when PREG is passed to regexec, that
6521 routine will report only success or failure, and nothing about the
6522 registers.
6524 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6525 the return codes and their meanings.) */
6527 int
6530 {
6531 reg_errcode_t ret;
6532 reg_syntax_t syntax
6536 /* regex_compile will allocate the space for the compiled pattern. */
6541 /* Try to allocate space for the fastmap. */
6545 {
6548 preg->translate
6554 /* Map uppercase characters to corresponding lowercase ones. */
6557 }
6558 else
6561 /* If REG_NEWLINE is set, newlines are treated differently. */
6566 }
6567 else
6572 /* POSIX says a null character in the pattern terminates it, so we
6573 can use strlen here in compiling the pattern. */
6576 /* POSIX doesn't distinguish between an unmatched open-group and an
6577 unmatched close-group: both are REG_EPAREN. */
6583 buffer. */
6589 }
6590 }
6592 }
6596 /* regexec searches for a given pattern, specified by PREG, in the
6597 string STRING.
6599 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6600 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6601 least NMATCH elements, and we set them to the offsets of the
6602 corresponding matched substrings.
6604 EFLAGS specifies `execution flags' which affect matching: if
6605 REG_NOTBOL is set, then ^ does not match at the beginning of the
6606 string; if REG_NOTEOL is set, then $ does not match at the end.
6608 We return 0 if we find a match and REG_NOMATCH if not. */
6610 int
6613 {
6616 regex_t private_preg;
6625 /* The user has told us exactly how many registers to return
6626 information about, via `nmatch'. We have to pass that on to the
6627 matching routines. */
6631 {
6637 }
6639 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6640 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6641 was a little bit longer but still only matching the real part.
6642 This works because the `endline' will check for a '\n' and will find a
6643 '\0', correctly deciding that this is not the end of a line.
6644 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6645 a convenient '\0' there. For all we know, the string could be preceded
6646 by '\n' which would throw things off. */
6648 /* Perform the searching operation. */
6653 /* Copy the register information to the POSIX structure. */
6655 {
6657 {
6661 {
6664 }
6665 }
6667 /* If we needed the temporary register info, free the space now. */
6669 }
6671 /* We want zero return to mean success, unlike `re_search'. */
6673 }
6677 /* Returns a message corresponding to an error code, ERR_CODE, returned
6678 from either regcomp or regexec. We don't use PREG here.
6680 ERR_CODE was previously called ERRCODE, but that name causes an
6681 error with msvc8 compiler. */
6683 size_t
6685 {
6691 /* Only error codes returned by the rest of the code should be passed
6692 to this routine. If we are given anything else, or if other regex
6693 code generates an invalid error code, then the program has a bug.
6694 Dump core so we can fix it. */
6702 {
6704 {
6707 }
6708 else
6710 }
6713 }
6717 /* Free dynamically allocated space used by PREG. */
6719 void
6721 {
6734 }