| blob | ccbad58abb34f17d923b5e756410f9369fa8945b |
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) 1993-2011 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
10 any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
20 USA. */
22 /* TODO:
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
29 */
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
33 #pragma alloca
34 #endif
36 #ifdef HAVE_CONFIG_H
37 # include <config.h>
38 #endif
40 #include <stddef.h>
42 #ifdef emacs
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
45 #endif
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
49 #if defined _LIBC
50 #define WIDE_CHAR_SUPPORT 1
51 #else
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
54 #endif
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
58 #if WIDE_CHAR_SUPPORT
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60 # include <wchar.h>
61 # include <wctype.h>
62 #endif
64 #ifdef _LIBC
65 /* We have to keep the namespace clean. */
66 # define regfree(preg) __regfree (preg)
67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69 # define regerror(err_code, preg, errbuf, errbuf_size) \
70 __regerror (err_code, preg, errbuf, errbuf_size)
71 # define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75 # define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77 # define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79 # define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81 # define re_set_syntax(syntax) __re_set_syntax (syntax)
82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
86 /* Make sure we call libc's function even if the user overrides them. */
87 # define btowc __btowc
88 # define iswctype __iswctype
89 # define wctype __wctype
91 # define WEAK_ALIAS(a,b) weak_alias (a, b)
93 /* We are also using some library internals. */
94 # include <locale/localeinfo.h>
95 # include <locale/elem-hash.h>
96 # include <langinfo.h>
97 #else
98 # define WEAK_ALIAS(a,b)
99 #endif
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
104 #else
105 # define gettext(msgid) (msgid)
106 #endif
108 #ifndef gettext_noop
109 /* This define is so xgettext can find the internationalizable
110 strings. */
111 # define gettext_noop(String) String
112 #endif
114 /* The `emacs' switch turns on certain matching commands
115 that make sense only in Emacs. */
116 #ifdef emacs
118 # include <setjmp.h>
122 /* Make syntax table lookup grant data in gl_state. */
123 # define SYNTAX_ENTRY_VIA_PROPERTY
129 # ifdef malloc
130 # undef malloc
131 # endif
132 # define malloc xmalloc
133 # ifdef realloc
134 # undef realloc
135 # endif
136 # define realloc xrealloc
137 # ifdef free
138 # undef free
139 # endif
140 # define free xfree
142 /* Converts the pointer to the char to BEG-based offset from the start. */
143 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
144 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
146 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
147 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
148 # define RE_STRING_CHAR(p, multibyte) \
149 (multibyte ? (STRING_CHAR (p)) : (*(p)))
150 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
151 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
153 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
155 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
157 /* Set C a (possibly converted to multibyte) character before P. P
158 points into a string which is the virtual concatenation of STR1
159 (which ends at END1) or STR2 (which ends at END2). */
160 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
161 do { \
162 if (target_multibyte) \
163 { \
164 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
165 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
166 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
167 c = STRING_CHAR (dtemp); \
168 } \
169 else \
170 { \
171 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
172 (c) = RE_CHAR_TO_MULTIBYTE (c); \
173 } \
174 } while (0)
176 /* Set C a (possibly converted to multibyte) character at P, and set
177 LEN to the byte length of that character. */
178 # define GET_CHAR_AFTER(c, p, len) \
179 do { \
180 if (target_multibyte) \
181 (c) = STRING_CHAR_AND_LENGTH (p, len); \
182 else \
183 { \
184 (c) = *p; \
185 len = 1; \
186 (c) = RE_CHAR_TO_MULTIBYTE (c); \
187 } \
188 } while (0)
192 /* If we are not linking with Emacs proper,
193 we can't use the relocating allocator
194 even if config.h says that we can. */
195 # undef REL_ALLOC
197 # include <unistd.h>
199 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
203 {
207 {
210 }
212 }
216 {
218 /* We must call malloc explicitly when BLOCK is 0, since some
219 reallocs don't do this. */
222 else
225 {
228 }
230 }
232 # ifdef malloc
233 # undef malloc
234 # endif
235 # define malloc xmalloc
236 # ifdef realloc
237 # undef realloc
238 # endif
239 # define realloc xrealloc
241 # include <string.h>
243 /* Define the syntax stuff for \<, \>, etc. */
245 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
248 # define SWITCH_ENUM_CAST(x) (x)
250 /* Dummy macros for non-Emacs environments. */
251 # define CHAR_CHARSET(c) 0
252 # define CHARSET_LEADING_CODE_BASE(c) 0
253 # define MAX_MULTIBYTE_LENGTH 1
254 # define RE_MULTIBYTE_P(x) 0
255 # define RE_TARGET_MULTIBYTE_P(x) 0
256 # define WORD_BOUNDARY_P(c1, c2) (0)
257 # define CHAR_HEAD_P(p) (1)
258 # define SINGLE_BYTE_CHAR_P(c) (1)
259 # define SAME_CHARSET_P(c1, c2) (1)
260 # define BYTES_BY_CHAR_HEAD(p) (1)
261 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
262 # define STRING_CHAR(p) (*(p))
263 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
264 # define CHAR_STRING(c, s) (*(s) = (c), 1)
265 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
266 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
267 # define RE_CHAR_TO_MULTIBYTE(c) (c)
268 # define RE_CHAR_TO_UNIBYTE(c) (c)
269 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
270 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
271 # define GET_CHAR_AFTER(c, p, len) \
272 (c = *p, len = 1)
273 # define MAKE_CHAR(charset, c1, c2) (c1)
274 # define BYTE8_TO_CHAR(c) (c)
275 # define CHAR_BYTE8_P(c) (0)
276 # define CHAR_LEADING_CODE(c) (c)
280 #ifndef RE_TRANSLATE
281 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
282 # define RE_TRANSLATE_P(TBL) (TBL)
283 #endif
284 \f
285 /* Get the interface, including the syntax bits. */
288 /* isalpha etc. are used for the character classes. */
289 #include <ctype.h>
291 #ifdef emacs
293 /* 1 if C is an ASCII character. */
294 # define IS_REAL_ASCII(c) ((c) < 0200)
296 /* 1 if C is a unibyte character. */
297 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
299 /* The Emacs definitions should not be directly affected by locales. */
301 /* In Emacs, these are only used for single-byte characters. */
308 /* This is only used for single-byte characters. */
311 /* The rest must handle multibyte characters. */
313 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
315 : 1)
317 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
319 : 1)
321 # define ISALNUM(c) (IS_REAL_ASCII (c) \
325 : SYNTAX (c) == Sword)
327 # define ISALPHA(c) (IS_REAL_ASCII (c) \
330 : SYNTAX (c) == Sword)
332 # define ISLOWER(c) lowercasep (c)
334 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
339 : SYNTAX (c) != Sword)
341 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
343 # define ISUPPER(c) uppercasep (c)
345 # define ISWORD(c) (SYNTAX (c) == Sword)
349 /* 1 if C is an ASCII character. */
350 # define IS_REAL_ASCII(c) ((c) < 0200)
352 /* This distinction is not meaningful, except in Emacs. */
353 # define ISUNIBYTE(c) 1
355 # ifdef isblank
356 # define ISBLANK(c) isblank (c)
357 # else
359 # endif
360 # ifdef isgraph
361 # define ISGRAPH(c) isgraph (c)
362 # else
363 # define ISGRAPH(c) (isprint (c) && !isspace (c))
364 # endif
366 /* Solaris defines ISPRINT so we must undefine it first. */
367 # undef ISPRINT
368 # define ISPRINT(c) isprint (c)
369 # define ISDIGIT(c) isdigit (c)
370 # define ISALNUM(c) isalnum (c)
371 # define ISALPHA(c) isalpha (c)
372 # define ISCNTRL(c) iscntrl (c)
373 # define ISLOWER(c) islower (c)
374 # define ISPUNCT(c) ispunct (c)
375 # define ISSPACE(c) isspace (c)
376 # define ISUPPER(c) isupper (c)
377 # define ISXDIGIT(c) isxdigit (c)
379 # define ISWORD(c) ISALPHA (c)
381 # ifdef _tolower
382 # define TOLOWER(c) _tolower (c)
383 # else
384 # define TOLOWER(c) tolower (c)
385 # endif
387 /* How many characters in the character set. */
388 # define CHAR_SET_SIZE 256
390 # ifdef SYNTAX_TABLE
398 static void
400 {
416 }
420 # define SYNTAX(c) re_syntax_table[(c)]
423 \f
424 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
425 since ours (we hope) works properly with all combinations of
426 machines, compilers, `char' and `unsigned char' argument types.
427 (Per Bothner suggested the basic approach.) */
428 #undef SIGN_EXTEND_CHAR
429 #if __STDC__
430 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
432 /* As in Harbison and Steele. */
433 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
434 #endif
435 \f
436 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
437 use `alloca' instead of `malloc'. This is because using malloc in
438 re_search* or re_match* could cause memory leaks when C-g is used in
439 Emacs; also, malloc is slower and causes storage fragmentation. On
440 the other hand, malloc is more portable, and easier to debug.
442 Because we sometimes use alloca, some routines have to be macros,
443 not functions -- `alloca'-allocated space disappears at the end of the
444 function it is called in. */
446 #ifdef REGEX_MALLOC
448 # define REGEX_ALLOCATE malloc
449 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
450 # define REGEX_FREE free
454 /* Emacs already defines alloca, sometimes. */
455 # ifndef alloca
457 /* Make alloca work the best possible way. */
458 # ifdef __GNUC__
459 # define alloca __builtin_alloca
461 # ifdef HAVE_ALLOCA_H
462 # include <alloca.h>
468 # define REGEX_ALLOCATE alloca
470 /* Assumes a `char *destination' variable. */
471 # define REGEX_REALLOCATE(source, osize, nsize) \
472 (destination = (char *) alloca (nsize), \
473 memcpy (destination, source, osize))
475 /* No need to do anything to free, after alloca. */
480 /* Define how to allocate the failure stack. */
482 #if defined REL_ALLOC && defined REGEX_MALLOC
484 # define REGEX_ALLOCATE_STACK(size) \
485 r_alloc (&failure_stack_ptr, (size))
486 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
487 r_re_alloc (&failure_stack_ptr, (nsize))
488 # define REGEX_FREE_STACK(ptr) \
489 r_alloc_free (&failure_stack_ptr)
493 # ifdef REGEX_MALLOC
495 # define REGEX_ALLOCATE_STACK malloc
496 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
497 # define REGEX_FREE_STACK free
501 # define REGEX_ALLOCATE_STACK alloca
503 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
504 REGEX_REALLOCATE (source, osize, nsize)
505 /* No need to explicitly free anything. */
506 # define REGEX_FREE_STACK(arg) ((void)0)
512 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
513 `string1' or just past its end. This works if PTR is NULL, which is
514 a good thing. */
515 #define FIRST_STRING_P(ptr) \
516 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
518 /* (Re)Allocate N items of type T using malloc, or fail. */
519 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
520 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
521 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
525 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
527 #undef MAX
528 #undef MIN
529 #define MAX(a, b) ((a) > (b) ? (a) : (b))
530 #define MIN(a, b) ((a) < (b) ? (a) : (b))
532 /* Type of source-pattern and string chars. */
533 #ifdef _MSC_VER
535 #else
537 #endif
540 #define false 0
541 #define true 1
546 ssize_t pos,
548 ssize_t stop));
549 \f
550 /* These are the command codes that appear in compiled regular
551 expressions. Some opcodes are followed by argument bytes. A
552 command code can specify any interpretation whatsoever for its
553 arguments. Zero bytes may appear in the compiled regular expression. */
556 {
559 /* Succeed right away--no more backtracking. */
560 succeed,
562 /* Followed by one byte giving n, then by n literal bytes. */
563 exactn,
565 /* Matches any (more or less) character. */
566 anychar,
568 /* Matches any one char belonging to specified set. First
569 following byte is number of bitmap bytes. Then come bytes
570 for a bitmap saying which chars are in. Bits in each byte
571 are ordered low-bit-first. A character is in the set if its
572 bit is 1. A character too large to have a bit in the map is
573 automatically not in the set.
575 If the length byte has the 0x80 bit set, then that stuff
576 is followed by a range table:
577 2 bytes of flags for character sets (low 8 bits, high 8 bits)
578 See RANGE_TABLE_WORK_BITS below.
579 2 bytes, the number of pairs that follow (upto 32767)
580 pairs, each 2 multibyte characters,
581 each multibyte character represented as 3 bytes. */
582 charset,
584 /* Same parameters as charset, but match any character that is
585 not one of those specified. */
586 charset_not,
588 /* Start remembering the text that is matched, for storing in a
589 register. Followed by one byte with the register number, in
590 the range 0 to one less than the pattern buffer's re_nsub
591 field. */
592 start_memory,
594 /* Stop remembering the text that is matched and store it in a
595 memory register. Followed by one byte with the register
596 number, in the range 0 to one less than `re_nsub' in the
597 pattern buffer. */
598 stop_memory,
600 /* Match a duplicate of something remembered. Followed by one
601 byte containing the register number. */
602 duplicate,
604 /* Fail unless at beginning of line. */
605 begline,
607 /* Fail unless at end of line. */
608 endline,
610 /* Succeeds if at beginning of buffer (if emacs) or at beginning
611 of string to be matched (if not). */
612 begbuf,
614 /* Analogously, for end of buffer/string. */
615 endbuf,
617 /* Followed by two byte relative address to which to jump. */
618 jump,
620 /* Followed by two-byte relative address of place to resume at
621 in case of failure. */
622 on_failure_jump,
624 /* Like on_failure_jump, but pushes a placeholder instead of the
625 current string position when executed. */
626 on_failure_keep_string_jump,
628 /* Just like `on_failure_jump', except that it checks that we
629 don't get stuck in an infinite loop (matching an empty string
630 indefinitely). */
631 on_failure_jump_loop,
633 /* Just like `on_failure_jump_loop', except that it checks for
634 a different kind of loop (the kind that shows up with non-greedy
635 operators). This operation has to be immediately preceded
636 by a `no_op'. */
637 on_failure_jump_nastyloop,
639 /* A smart `on_failure_jump' used for greedy * and + operators.
640 It analyses the loop before which it is put and if the
641 loop does not require backtracking, it changes itself to
642 `on_failure_keep_string_jump' and short-circuits the loop,
643 else it just defaults to changing itself into `on_failure_jump'.
644 It assumes that it is pointing to just past a `jump'. */
645 on_failure_jump_smart,
647 /* Followed by two-byte relative address and two-byte number n.
648 After matching N times, jump to the address upon failure.
649 Does not work if N starts at 0: use on_failure_jump_loop
650 instead. */
651 succeed_n,
653 /* Followed by two-byte relative address, and two-byte number n.
654 Jump to the address N times, then fail. */
655 jump_n,
657 /* Set the following two-byte relative address to the
658 subsequent two-byte number. The address *includes* the two
659 bytes of number. */
660 set_number_at,
671 /* Matches any character whose syntax is specified. Followed by
672 a byte which contains a syntax code, e.g., Sword. */
673 syntaxspec,
675 /* Matches any character whose syntax is not that specified. */
676 notsyntaxspec
678 #ifdef emacs
683 /* Matches any character whose category-set contains the specified
684 category. The operator is followed by a byte which contains a
685 category code (mnemonic ASCII character). */
686 categoryspec,
688 /* Matches any character whose category-set does not contain the
689 specified category. The operator is followed by a byte which
690 contains the category code (mnemonic ASCII character). */
691 notcategoryspec
694 \f
695 /* Common operations on the compiled pattern. */
697 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
699 #define STORE_NUMBER(destination, number) \
700 do { \
701 (destination)[0] = (number) & 0377; \
702 (destination)[1] = (number) >> 8; \
703 } while (0)
705 /* Same as STORE_NUMBER, except increment DESTINATION to
706 the byte after where the number is stored. Therefore, DESTINATION
707 must be an lvalue. */
709 #define STORE_NUMBER_AND_INCR(destination, number) \
710 do { \
711 STORE_NUMBER (destination, number); \
712 (destination) += 2; \
713 } while (0)
715 /* Put into DESTINATION a number stored in two contiguous bytes starting
716 at SOURCE. */
718 #define EXTRACT_NUMBER(destination, source) \
719 do { \
720 (destination) = *(source) & 0377; \
721 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
722 } while (0)
724 #ifdef DEBUG
726 static void
730 {
734 }
737 # undef EXTRACT_NUMBER
738 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
743 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
744 SOURCE must be an lvalue. */
746 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
747 do { \
748 EXTRACT_NUMBER (destination, source); \
749 (source) += 2; \
750 } while (0)
752 #ifdef DEBUG
755 static void
759 {
762 }
764 # ifndef EXTRACT_MACROS
765 # undef EXTRACT_NUMBER_AND_INCR
766 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
767 extract_number_and_incr (&dest, &src)
771 \f
772 /* Store a multibyte character in three contiguous bytes starting
773 DESTINATION, and increment DESTINATION to the byte after where the
774 character is stored. Therefore, DESTINATION must be an lvalue. */
776 #define STORE_CHARACTER_AND_INCR(destination, character) \
777 do { \
778 (destination)[0] = (character) & 0377; \
779 (destination)[1] = ((character) >> 8) & 0377; \
780 (destination)[2] = (character) >> 16; \
781 (destination) += 3; \
782 } while (0)
784 /* Put into DESTINATION a character stored in three contiguous bytes
785 starting at SOURCE. */
787 #define EXTRACT_CHARACTER(destination, source) \
788 do { \
789 (destination) = ((source)[0] \
790 | ((source)[1] << 8) \
791 | ((source)[2] << 16)); \
792 } while (0)
795 /* Macros for charset. */
797 /* Size of bitmap of charset P in bytes. P is a start of charset,
798 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
799 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
801 /* Nonzero if charset P has range table. */
802 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
804 /* Return the address of range table of charset P. But not the start
805 of table itself, but the before where the number of ranges is
806 stored. `2 +' means to skip re_opcode_t and size of bitmap,
807 and the 2 bytes of flags at the start of the range table. */
808 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
810 /* Extract the bit flags that start a range table. */
811 #define CHARSET_RANGE_TABLE_BITS(p) \
812 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
813 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
815 /* Return the address of end of RANGE_TABLE. COUNT is number of
816 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
817 is start of range and end of range. `* 3' is size of each start
818 and end. */
819 #define CHARSET_RANGE_TABLE_END(range_table, count) \
820 ((range_table) + (count) * 2 * 3)
822 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
823 COUNT is number of ranges in RANGE_TABLE. */
824 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
825 do \
826 { \
827 re_wchar_t range_start, range_end; \
828 re_char *rtp; \
829 re_char *range_table_end \
830 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
831 \
832 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
833 { \
834 EXTRACT_CHARACTER (range_start, rtp); \
835 EXTRACT_CHARACTER (range_end, rtp + 3); \
836 \
837 if (range_start <= (c) && (c) <= range_end) \
838 { \
839 (not) = !(not); \
840 break; \
841 } \
842 } \
843 } \
844 while (0)
846 /* Test if C is in range table of CHARSET. The flag NOT is negated if
847 C is listed in it. */
848 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
849 do \
850 { \
852 int count; \
853 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
854 \
855 EXTRACT_NUMBER_AND_INCR (count, range_table); \
856 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
857 } \
858 while (0)
859 \f
860 /* If DEBUG is defined, Regex prints many voluminous messages about what
861 it is doing (if the variable `debug' is nonzero). If linked with the
862 main program in `iregex.c', you can enter patterns and strings
863 interactively. And if linked with the main program in `main.c' and
864 the other test files, you can run the already-written tests. */
866 #ifdef DEBUG
868 /* We use standard I/O for debugging. */
869 # include <stdio.h>
871 /* It is useful to test things that ``must'' be true when debugging. */
872 # include <assert.h>
876 # define DEBUG_STATEMENT(e) e
877 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
878 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
879 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
880 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
881 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
882 if (debug > 0) print_partial_compiled_pattern (s, e)
883 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
884 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
887 /* Print the fastmap in human-readable form. */
889 void
892 {
897 {
899 {
903 {
905 i++;
906 }
908 {
911 }
912 }
913 }
915 }
918 /* Print a compiled pattern string in human-readable form, starting at
919 the START pointer into it and ending just before the pointer END. */
921 void
925 {
931 {
934 }
936 /* Loop over pattern commands. */
938 {
942 {
954 do
955 {
957 }
979 {
994 {
995 /* Are we starting a range? */
997 {
1000 }
1001 /* Have we broken a range? */
1003 {
1006 }
1012 }
1022 {
1026 /* ??? Should print the range table; for now, just skip it. */
1030 }
1031 }
1126 # ifdef emacs
1162 }
1165 }
1168 }
1171 void
1174 {
1182 {
1185 }
1195 /* Perhaps we should print the translate table? */
1196 }
1199 void
1204 ssize_t size1;
1205 ssize_t size2;
1206 {
1207 ssize_t this_char;
1211 else
1212 {
1214 {
1219 }
1223 }
1224 }
1228 # undef assert
1229 # define assert(e)
1231 # define DEBUG_STATEMENT(e)
1232 # define DEBUG_PRINT1(x)
1233 # define DEBUG_PRINT2(x1, x2)
1234 # define DEBUG_PRINT3(x1, x2, x3)
1235 # define DEBUG_PRINT4(x1, x2, x3, x4)
1236 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1237 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1240 \f
1241 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1242 #ifdef lint
1243 # define IF_LINT(Code) Code
1244 #else
1246 #endif
1247 \f
1248 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1249 also be assigned to arbitrarily: each pattern buffer stores its own
1250 syntax, so it can be changed between regex compilations. */
1251 /* This has no initializer because initialized variables in Emacs
1252 become read-only after dumping. */
1253 reg_syntax_t re_syntax_options;
1256 /* Specify the precise syntax of regexps for compilation. This provides
1257 for compatibility for various utilities which historically have
1258 different, incompatible syntaxes.
1260 The argument SYNTAX is a bit mask comprised of the various bits
1261 defined in regex.h. We return the old syntax. */
1263 reg_syntax_t
1265 {
1270 }
1273 /* Regexp to use to replace spaces, or NULL meaning don't. */
1276 void
1278 {
1280 }
1282 \f
1283 /* This table gives an error message for each of the error codes listed
1284 in regex.h. Obviously the order here has to be same as there.
1285 POSIX doesn't require that we do anything for REG_NOERROR,
1286 but why not be nice? */
1289 {
1308 };
1309 \f
1310 /* Avoiding alloca during matching, to placate r_alloc. */
1312 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1313 searching and matching functions should not call alloca. On some
1314 systems, alloca is implemented in terms of malloc, and if we're
1315 using the relocating allocator routines, then malloc could cause a
1316 relocation, which might (if the strings being searched are in the
1317 ralloc heap) shift the data out from underneath the regexp
1318 routines.
1320 Here's another reason to avoid allocation: Emacs
1321 processes input from X in a signal handler; processing X input may
1322 call malloc; if input arrives while a matching routine is calling
1323 malloc, then we're scrod. But Emacs can't just block input while
1324 calling matching routines; then we don't notice interrupts when
1325 they come in. So, Emacs blocks input around all regexp calls
1326 except the matching calls, which it leaves unprotected, in the
1327 faith that they will not malloc. */
1329 /* Normally, this is fine. */
1330 #define MATCH_MAY_ALLOCATE
1332 /* The match routines may not allocate if (1) they would do it with malloc
1333 and (2) it's not safe for them to use malloc.
1334 Note that if REL_ALLOC is defined, matching would not use malloc for the
1335 failure stack, but we would still use it for the register vectors;
1336 so REL_ALLOC should not affect this. */
1337 #if defined REGEX_MALLOC && defined emacs
1338 # undef MATCH_MAY_ALLOCATE
1339 #endif
1341 \f
1342 /* Failure stack declarations and macros; both re_compile_fastmap and
1343 re_match_2 use a failure stack. These have to be macros because of
1344 REGEX_ALLOCATE_STACK. */
1347 /* Approximate number of failure points for which to initially allocate space
1348 when matching. If this number is exceeded, we allocate more
1349 space, so it is not a hard limit. */
1350 #ifndef INIT_FAILURE_ALLOC
1351 # define INIT_FAILURE_ALLOC 20
1352 #endif
1354 /* Roughly the maximum number of failure points on the stack. Would be
1355 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1356 This is a variable only so users of regex can assign to it; we never
1357 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1358 before using it, so it should probably be a byte-count instead. */
1359 # if defined MATCH_MAY_ALLOCATE
1360 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1361 whose default stack limit is 2mb. In order for a larger
1362 value to work reliably, you have to try to make it accord
1363 with the process stack limit. */
1365 # else
1367 # endif
1369 union fail_stack_elt
1370 {
1372 /* This should be the biggest `int' that's no bigger than a pointer. */
1374 };
1379 {
1386 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1389 /* Define macros to initialize and free the failure stack.
1390 Do `return -2' if the alloc fails. */
1392 #ifdef MATCH_MAY_ALLOCATE
1393 # define INIT_FAIL_STACK() \
1394 do { \
1395 fail_stack.stack = (fail_stack_elt_t *) \
1396 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1397 * sizeof (fail_stack_elt_t)); \
1398 \
1399 if (fail_stack.stack == NULL) \
1400 return -2; \
1401 \
1402 fail_stack.size = INIT_FAILURE_ALLOC; \
1403 fail_stack.avail = 0; \
1404 fail_stack.frame = 0; \
1405 } while (0)
1406 #else
1407 # define INIT_FAIL_STACK() \
1408 do { \
1409 fail_stack.avail = 0; \
1410 fail_stack.frame = 0; \
1411 } while (0)
1413 # define RETALLOC_IF(addr, n, t) \
1414 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1415 #endif
1418 /* Double the size of FAIL_STACK, up to a limit
1419 which allows approximately `re_max_failures' items.
1421 Return 1 if succeeds, and 0 if either ran out of memory
1422 allocating space for it or it was already too large.
1424 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1426 /* Factor to increase the failure stack size by
1427 when we increase it.
1428 This used to be 2, but 2 was too wasteful
1429 because the old discarded stacks added up to as much space
1430 were as ultimate, maximum-size stack. */
1431 #define FAIL_STACK_GROWTH_FACTOR 4
1433 #define GROW_FAIL_STACK(fail_stack) \
1434 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1435 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1436 ? 0 \
1437 : ((fail_stack).stack \
1438 = (fail_stack_elt_t *) \
1439 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1440 (fail_stack).size * sizeof (fail_stack_elt_t), \
1441 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1442 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1443 * FAIL_STACK_GROWTH_FACTOR))), \
1444 \
1445 (fail_stack).stack == NULL \
1446 ? 0 \
1447 : ((fail_stack).size \
1448 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1449 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1450 * FAIL_STACK_GROWTH_FACTOR)) \
1451 / sizeof (fail_stack_elt_t)), \
1452 1)))
1455 /* Push a pointer value onto the failure stack.
1456 Assumes the variable `fail_stack'. Probably should only
1457 be called from within `PUSH_FAILURE_POINT'. */
1458 #define PUSH_FAILURE_POINTER(item) \
1459 fail_stack.stack[fail_stack.avail++].pointer = (item)
1461 /* This pushes an integer-valued item onto the failure stack.
1462 Assumes the variable `fail_stack'. Probably should only
1463 be called from within `PUSH_FAILURE_POINT'. */
1464 #define PUSH_FAILURE_INT(item) \
1465 fail_stack.stack[fail_stack.avail++].integer = (item)
1467 /* These POP... operations complement the PUSH... operations.
1468 All assume that `fail_stack' is nonempty. */
1469 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1470 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1472 /* Individual items aside from the registers. */
1473 #define NUM_NONREG_ITEMS 3
1475 /* Used to examine the stack (to detect infinite loops). */
1476 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1477 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1478 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1479 #define TOP_FAILURE_HANDLE() fail_stack.frame
1482 #define ENSURE_FAIL_STACK(space) \
1483 while (REMAINING_AVAIL_SLOTS <= space) { \
1484 if (!GROW_FAIL_STACK (fail_stack)) \
1485 return -2; \
1488 }
1490 /* Push register NUM onto the stack. */
1491 #define PUSH_FAILURE_REG(num) \
1492 do { \
1493 char *destination; \
1494 ENSURE_FAIL_STACK(3); \
1496 num, regstart[num], regend[num]); \
1497 PUSH_FAILURE_POINTER (regstart[num]); \
1498 PUSH_FAILURE_POINTER (regend[num]); \
1499 PUSH_FAILURE_INT (num); \
1500 } while (0)
1502 /* Change the counter's value to VAL, but make sure that it will
1503 be reset when backtracking. */
1504 #define PUSH_NUMBER(ptr,val) \
1505 do { \
1506 char *destination; \
1507 int c; \
1508 ENSURE_FAIL_STACK(3); \
1509 EXTRACT_NUMBER (c, ptr); \
1511 PUSH_FAILURE_INT (c); \
1512 PUSH_FAILURE_POINTER (ptr); \
1513 PUSH_FAILURE_INT (-1); \
1514 STORE_NUMBER (ptr, val); \
1515 } while (0)
1517 /* Pop a saved register off the stack. */
1518 #define POP_FAILURE_REG_OR_COUNT() \
1519 do { \
1520 long pfreg = POP_FAILURE_INT (); \
1521 if (pfreg == -1) \
1522 { \
1525 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1526 pfreg = POP_FAILURE_INT (); \
1527 STORE_NUMBER (ptr, pfreg); \
1529 } \
1530 else \
1531 { \
1532 regend[pfreg] = POP_FAILURE_POINTER (); \
1533 regstart[pfreg] = POP_FAILURE_POINTER (); \
1535 pfreg, regstart[pfreg], regend[pfreg]); \
1536 } \
1537 } while (0)
1539 /* Check that we are not stuck in an infinite loop. */
1540 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1541 do { \
1542 ssize_t failure = TOP_FAILURE_HANDLE (); \
1544 while (failure > 0 \
1545 && (FAILURE_STR (failure) == string_place \
1546 || FAILURE_STR (failure) == NULL)) \
1547 { \
1548 assert (FAILURE_PAT (failure) >= bufp->buffer \
1549 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1550 if (FAILURE_PAT (failure) == pat_cur) \
1551 { \
1552 cycle = 1; \
1553 break; \
1554 } \
1556 failure = NEXT_FAILURE_HANDLE(failure); \
1557 } \
1559 } while (0)
1561 /* Push the information about the state we will need
1562 if we ever fail back to it.
1564 Requires variables fail_stack, regstart, regend and
1565 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1566 declared.
1568 Does `return FAILURE_CODE' if runs out of memory. */
1570 #define PUSH_FAILURE_POINT(pattern, string_place) \
1571 do { \
1572 char *destination; \
1573 /* Must be int, so when we don't save any registers, the arithmetic \
1575 \
1576 DEBUG_STATEMENT (nfailure_points_pushed++); \
1580 \
1581 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1582 \
1584 \
1586 PUSH_FAILURE_INT (fail_stack.frame); \
1587 \
1589 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1591 PUSH_FAILURE_POINTER (string_place); \
1592 \
1594 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1595 PUSH_FAILURE_POINTER (pattern); \
1596 \
1598 fail_stack.frame = fail_stack.avail; \
1599 } while (0)
1601 /* Estimate the size of data pushed by a typical failure stack entry.
1602 An estimate is all we need, because all we use this for
1603 is to choose a limit for how big to make the failure stack. */
1604 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1605 #define TYPICAL_FAILURE_SIZE 20
1607 /* How many items can still be added to the stack without overflowing it. */
1608 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1611 /* Pops what PUSH_FAIL_STACK pushes.
1613 We restore into the parameters, all of which should be lvalues:
1614 STR -- the saved data position.
1615 PAT -- the saved pattern position.
1616 REGSTART, REGEND -- arrays of string positions.
1618 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1619 `pend', `string1', `size1', `string2', and `size2'. */
1621 #define POP_FAILURE_POINT(str, pat) \
1622 do { \
1623 assert (!FAIL_STACK_EMPTY ()); \
1624 \
1629 \
1631 while (fail_stack.frame < fail_stack.avail) \
1632 POP_FAILURE_REG_OR_COUNT (); \
1633 \
1634 pat = POP_FAILURE_POINTER (); \
1636 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1637 \
1638 /* If the saved string location is NULL, it came from an \
1639 on_failure_keep_string_jump opcode, and we want to throw away the \
1641 str = POP_FAILURE_POINTER (); \
1643 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1645 \
1646 fail_stack.frame = POP_FAILURE_INT (); \
1648 \
1649 assert (fail_stack.avail >= 0); \
1650 assert (fail_stack.frame <= fail_stack.avail); \
1651 \
1652 DEBUG_STATEMENT (nfailure_points_popped++); \
1656 \f
1657 /* Registers are set to a sentinel when they haven't yet matched. */
1658 #define REG_UNSET(e) ((e) == NULL)
1659 \f
1660 /* Subroutine declarations and macros for regex_compile. */
1663 reg_syntax_t syntax,
1674 reg_syntax_t syntax));
1677 reg_syntax_t syntax));
1682 /* Fetch the next character in the uncompiled pattern, with no
1683 translation. */
1684 #define PATFETCH(c) \
1685 do { \
1686 int len; \
1687 if (p == pend) return REG_EEND; \
1688 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1689 p += len; \
1690 } while (0)
1693 /* If `translate' is non-null, return translate[D], else just D. We
1694 cast the subscript to translate because some data is declared as
1695 `char *', to avoid warnings when a string constant is passed. But
1696 when we use a character as a subscript we must make it unsigned. */
1697 #ifndef TRANSLATE
1698 # define TRANSLATE(d) \
1699 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1700 #endif
1703 /* Macros for outputting the compiled pattern into `buffer'. */
1705 /* If the buffer isn't allocated when it comes in, use this. */
1706 #define INIT_BUF_SIZE 32
1708 /* Make sure we have at least N more bytes of space in buffer. */
1709 #define GET_BUFFER_SPACE(n) \
1710 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1711 EXTEND_BUFFER ()
1713 /* Make sure we have one more byte of buffer space and then add C to it. */
1714 #define BUF_PUSH(c) \
1715 do { \
1716 GET_BUFFER_SPACE (1); \
1717 *b++ = (unsigned char) (c); \
1718 } while (0)
1721 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1722 #define BUF_PUSH_2(c1, c2) \
1723 do { \
1724 GET_BUFFER_SPACE (2); \
1725 *b++ = (unsigned char) (c1); \
1726 *b++ = (unsigned char) (c2); \
1727 } while (0)
1730 /* Store a jump with opcode OP at LOC to location TO. We store a
1731 relative address offset by the three bytes the jump itself occupies. */
1732 #define STORE_JUMP(op, loc, to) \
1733 store_op1 (op, loc, (to) - (loc) - 3)
1735 /* Likewise, for a two-argument jump. */
1736 #define STORE_JUMP2(op, loc, to, arg) \
1737 store_op2 (op, loc, (to) - (loc) - 3, arg)
1739 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1740 #define INSERT_JUMP(op, loc, to) \
1741 insert_op1 (op, loc, (to) - (loc) - 3, b)
1743 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1744 #define INSERT_JUMP2(op, loc, to, arg) \
1745 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1748 /* This is not an arbitrary limit: the arguments which represent offsets
1749 into the pattern are two bytes long. So if 2^15 bytes turns out to
1750 be too small, many things would have to change. */
1751 # define MAX_BUF_SIZE (1L << 15)
1754 /* Any other compiler which, like MSC, has allocation limit below 2^16
1755 bytes will have to use approach similar to what was done below for
1756 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1757 reallocating to 0 bytes. Such thing is not going to work too well.
1758 You have been warned!! */
1759 #if defined _MSC_VER && !defined WIN32
1760 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1761 # define MAX_BUF_SIZE 65500L
1762 #else
1763 # define MAX_BUF_SIZE (1L << 16)
1764 #endif
1767 /* Extend the buffer by twice its current size via realloc and
1768 reset the pointers that pointed into the old block to point to the
1769 correct places in the new one. If extending the buffer results in it
1770 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1771 #if __BOUNDED_POINTERS__
1772 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1773 # define MOVE_BUFFER_POINTER(P) \
1774 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1775 SET_HIGH_BOUND (P), \
1776 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1777 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1778 else \
1779 { \
1780 SET_HIGH_BOUND (b); \
1781 SET_HIGH_BOUND (begalt); \
1782 if (fixup_alt_jump) \
1783 SET_HIGH_BOUND (fixup_alt_jump); \
1784 if (laststart) \
1785 SET_HIGH_BOUND (laststart); \
1786 if (pending_exact) \
1787 SET_HIGH_BOUND (pending_exact); \
1788 }
1789 #else
1790 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1791 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1792 #endif
1793 #define EXTEND_BUFFER() \
1794 do { \
1795 unsigned char *old_buffer = bufp->buffer; \
1796 if (bufp->allocated == MAX_BUF_SIZE) \
1797 return REG_ESIZE; \
1798 bufp->allocated <<= 1; \
1799 if (bufp->allocated > MAX_BUF_SIZE) \
1800 bufp->allocated = MAX_BUF_SIZE; \
1801 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1802 if (bufp->buffer == NULL) \
1803 return REG_ESPACE; \
1805 if (old_buffer != bufp->buffer) \
1806 { \
1807 unsigned char *new_buffer = bufp->buffer; \
1808 MOVE_BUFFER_POINTER (b); \
1809 MOVE_BUFFER_POINTER (begalt); \
1810 if (fixup_alt_jump) \
1811 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1812 if (laststart) \
1813 MOVE_BUFFER_POINTER (laststart); \
1814 if (pending_exact) \
1815 MOVE_BUFFER_POINTER (pending_exact); \
1816 } \
1817 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1818 } while (0)
1821 /* Since we have one byte reserved for the register number argument to
1822 {start,stop}_memory, the maximum number of groups we can report
1823 things about is what fits in that byte. */
1824 #define MAX_REGNUM 255
1826 /* But patterns can have more than `MAX_REGNUM' registers. We just
1827 ignore the excess. */
1831 /* Macros for the compile stack. */
1833 /* Since offsets can go either forwards or backwards, this type needs to
1834 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1835 /* int may be not enough when sizeof(int) == 2. */
1839 {
1840 pattern_offset_t begalt_offset;
1841 pattern_offset_t fixup_alt_jump;
1842 pattern_offset_t laststart_offset;
1843 regnum_t regnum;
1848 {
1855 #define INIT_COMPILE_STACK_SIZE 32
1857 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1858 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1860 /* The next available element. */
1861 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1863 /* Explicit quit checking is only used on NTemacs and whenever we
1864 use polling to process input events. */
1865 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1867 # define IMMEDIATE_QUIT_CHECK \
1868 do { \
1869 if (immediate_quit) QUIT; \
1870 } while (0)
1871 #else
1872 # define IMMEDIATE_QUIT_CHECK ((void)0)
1873 #endif
1874 \f
1875 /* Structure to manage work area for range table. */
1876 struct range_table_work_area
1877 {
1882 };
1884 /* Make sure that WORK_AREA can hold more N multibyte characters.
1885 This is used only in set_image_of_range and set_image_of_range_1.
1886 It expects WORK_AREA to be a pointer.
1887 If it can't get the space, it returns from the surrounding function. */
1889 #define EXTEND_RANGE_TABLE(work_area, n) \
1890 do { \
1891 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1892 { \
1893 extend_range_table_work_area (&work_area); \
1894 if ((work_area).table == 0) \
1895 return (REG_ESPACE); \
1896 } \
1897 } while (0)
1899 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1900 (work_area).bits |= (bit)
1902 /* Bits used to implement the multibyte-part of the various character classes
1903 such as [:alnum:] in a charset's range table. */
1904 #define BIT_WORD 0x1
1905 #define BIT_LOWER 0x2
1906 #define BIT_PUNCT 0x4
1907 #define BIT_SPACE 0x8
1908 #define BIT_UPPER 0x10
1909 #define BIT_MULTIBYTE 0x20
1911 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1912 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1913 do { \
1914 EXTEND_RANGE_TABLE ((work_area), 2); \
1915 (work_area).table[(work_area).used++] = (range_start); \
1916 (work_area).table[(work_area).used++] = (range_end); \
1917 } while (0)
1919 /* Free allocated memory for WORK_AREA. */
1920 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1921 do { \
1922 if ((work_area).table) \
1923 free ((work_area).table); \
1924 } while (0)
1926 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1927 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1928 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1929 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1930 \f
1932 /* Set the bit for character C in a list. */
1933 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1936 #ifdef emacs
1938 /* Store characters in the range FROM to TO in the bitmap at B (for
1939 ASCII and unibyte characters) and WORK_AREA (for multibyte
1940 characters) while translating them and paying attention to the
1941 continuity of translated characters.
1943 Implementation note: It is better to implement these fairly big
1944 macros by a function, but it's not that easy because macros called
1945 in this macro assume various local variables already declared. */
1947 /* Both FROM and TO are ASCII characters. */
1949 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1950 do { \
1951 int C0, C1; \
1952 \
1953 for (C0 = (FROM); C0 <= (TO); C0++) \
1954 { \
1955 C1 = TRANSLATE (C0); \
1956 if (! ASCII_CHAR_P (C1)) \
1957 { \
1958 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1959 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1960 C1 = C0; \
1961 } \
1962 SET_LIST_BIT (C1); \
1963 } \
1964 } while (0)
1967 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1969 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1970 do { \
1971 int C0, C1, C2, I; \
1972 int USED = RANGE_TABLE_WORK_USED (work_area); \
1973 \
1974 for (C0 = (FROM); C0 <= (TO); C0++) \
1975 { \
1976 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1977 if (CHAR_BYTE8_P (C1)) \
1978 SET_LIST_BIT (C0); \
1979 else \
1980 { \
1981 C2 = TRANSLATE (C1); \
1982 if (C2 == C1 \
1983 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1984 C1 = C0; \
1985 SET_LIST_BIT (C1); \
1986 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1987 { \
1988 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1989 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1990 \
1991 if (C2 >= from - 1 && C2 <= to + 1) \
1992 { \
1993 if (C2 == from - 1) \
1994 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1995 else if (C2 == to + 1) \
1996 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1997 break; \
1998 } \
1999 } \
2000 if (I < USED) \
2001 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2002 } \
2003 } \
2004 } while (0)
2007 /* Both FROM and TO are multibyte characters. */
2009 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2010 do { \
2011 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2012 \
2013 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2014 for (C0 = (FROM); C0 <= (TO); C0++) \
2015 { \
2016 C1 = TRANSLATE (C0); \
2017 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2018 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2019 SET_LIST_BIT (C2); \
2020 if (C1 >= (FROM) && C1 <= (TO)) \
2021 continue; \
2022 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2023 { \
2024 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2025 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2026 \
2027 if (C1 >= from - 1 && C1 <= to + 1) \
2028 { \
2029 if (C1 == from - 1) \
2030 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2031 else if (C1 == to + 1) \
2032 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2033 break; \
2034 } \
2035 } \
2036 if (I < USED) \
2037 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2038 } \
2039 } while (0)
2043 /* Get the next unsigned number in the uncompiled pattern. */
2044 #define GET_UNSIGNED_NUMBER(num) \
2045 do { \
2046 if (p == pend) \
2047 FREE_STACK_RETURN (REG_EBRACE); \
2048 else \
2049 { \
2050 PATFETCH (c); \
2052 { \
2053 int prev; \
2054 if (num < 0) \
2055 num = 0; \
2056 prev = num; \
2058 if (num / 10 != prev) \
2059 FREE_STACK_RETURN (REG_BADBR); \
2060 if (p == pend) \
2061 FREE_STACK_RETURN (REG_EBRACE); \
2062 PATFETCH (c); \
2063 } \
2064 } \
2065 } while (0)
2066 \f
2067 #if ! WIDE_CHAR_SUPPORT
2069 /* Map a string to the char class it names (if any). */
2070 re_wctype_t
2072 {
2092 }
2094 /* True if CH is in the char class CC. */
2095 boolean
2097 {
2099 {
2120 }
2121 }
2123 /* Return a bit-pattern to use in the range-table bits to match multibyte
2124 chars of class CC. */
2125 static int
2127 {
2129 {
2141 }
2142 }
2143 #endif
2144 \f
2145 /* Filling in the work area of a range. */
2147 /* Actually extend the space in WORK_AREA. */
2149 static void
2151 {
2154 work_area->table
2156 else
2157 work_area->table
2159 }
2161 #if 0
2162 #ifdef emacs
2164 /* Carefully find the ranges of codes that are equivalent
2165 under case conversion to the range start..end when passed through
2166 TRANSLATE. Handle the case where non-letters can come in between
2167 two upper-case letters (which happens in Latin-1).
2168 Also handle the case of groups of more than 2 case-equivalent chars.
2170 The basic method is to look at consecutive characters and see
2171 if they can form a run that can be handled as one.
2173 Returns -1 if successful, REG_ESPACE if ran out of space. */
2175 static int
2178 RE_TRANSLATE_TYPE translate)
2179 {
2180 /* `one_case' indicates a character, or a run of characters,
2181 each of which is an isolate (no case-equivalents).
2182 This includes all ASCII non-letters.
2184 `two_case' indicates a character, or a run of characters,
2185 each of which has two case-equivalent forms.
2186 This includes all ASCII letters.
2188 `strange' indicates a character that has more than one
2189 case-equivalent. */
2193 /* Describe the run that is in progress,
2194 which the next character can try to extend.
2195 If run_type is strange, that means there really is no run.
2196 If run_type is one_case, then run_start...run_end is the run.
2197 If run_type is two_case, then the run is run_start...run_end,
2198 and the case-equivalents end at run_eqv_end. */
2203 Lisp_Object eqv_table;
2206 {
2211 }
2216 {
2221 /* Classify this character */
2226 else
2231 else
2234 /* Can this character extend the run in progress? */
2239 {
2240 /* No, end the run.
2241 Record each of its equivalent ranges. */
2243 {
2247 }
2249 {
2257 }
2259 }
2262 {
2263 /* For a strange character, add each of its equivalents, one
2264 by one. Don't start a range. */
2265 do
2266 {
2271 }
2273 }
2275 /* Add this char to the run, or start a new run. */
2277 {
2278 /* Initialize a new range. */
2283 }
2284 else
2285 {
2286 /* Extend a running range. */
2289 }
2290 }
2292 /* If a run is still in progress at the end, finish it now
2293 by recording its equivalent ranges. */
2295 {
2299 }
2301 {
2309 }
2312 }
2316 /* Record the image of the range start..end when passed through
2317 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2318 and is not even necessarily contiguous.
2319 Normally we approximate it with the smallest contiguous range that contains
2320 all the chars we need. However, for Latin-1 we go to extra effort
2321 to do a better job.
2323 This function is not called for ASCII ranges.
2325 Returns -1 if successful, REG_ESPACE if ran out of space. */
2327 static int
2330 RE_TRANSLATE_TYPE translate)
2331 {
2334 #ifdef emacs
2335 /* For Latin-1 ranges, use set_image_of_range_1
2336 to get proper handling of ranges that include letters and nonletters.
2337 For a range that includes the whole of Latin-1, this is not necessary.
2338 For other character sets, we don't bother to get this right. */
2341 {
2354 }
2355 #endif
2364 {
2368 {
2371 {
2374 else
2375 {
2378 }
2379 }
2380 }
2383 {
2387 }
2388 }
2391 }
2393 \f
2394 #ifndef MATCH_MAY_ALLOCATE
2396 /* If we cannot allocate large objects within re_match_2_internal,
2397 we make the fail stack and register vectors global.
2398 The fail stack, we grow to the maximum size when a regexp
2399 is compiled.
2400 The register vectors, we adjust in size each time we
2401 compile a regexp, according to the number of registers it needs. */
2405 /* Size with which the following vectors are currently allocated.
2406 That is so we can make them bigger as needed,
2407 but never make them smaller. */
2413 /* Make the register vectors big enough for NUM_REGS registers,
2414 but don't make them smaller. */
2416 static
2418 {
2420 {
2427 }
2428 }
2431 \f
2433 compile_stack,
2434 regnum_t regnum));
2436 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2437 Returns one of error codes defined in `regex.h', or zero for success.
2439 Assumes the `allocated' (and perhaps `buffer') and `translate'
2440 fields are set in BUFP on entry.
2442 If it succeeds, results are put in BUFP (if it returns an error, the
2443 contents of BUFP are undefined):
2444 `buffer' is the compiled pattern;
2445 `syntax' is set to SYNTAX;
2446 `used' is set to the length of the compiled pattern;
2447 `fastmap_accurate' is zero;
2448 `re_nsub' is the number of subexpressions in PATTERN;
2449 `not_bol' and `not_eol' are zero;
2451 The `fastmap' field is neither examined nor set. */
2453 /* Insert the `jump' from the end of last alternative to "here".
2454 The space for the jump has already been allocated. */
2455 #define FIXUP_ALT_JUMP() \
2456 do { \
2457 if (fixup_alt_jump) \
2458 STORE_JUMP (jump, fixup_alt_jump, b); \
2459 } while (0)
2462 /* Return, freeing storage we allocated. */
2463 #define FREE_STACK_RETURN(value) \
2464 do { \
2465 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2466 free (compile_stack.stack); \
2467 return value; \
2468 } while (0)
2470 static reg_errcode_t
2471 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2472 {
2473 /* We fetch characters from PATTERN here. */
2476 /* Points to the end of the buffer, where we should append. */
2479 /* Keeps track of unclosed groups. */
2480 compile_stack_type compile_stack;
2482 /* Points to the current (ending) position in the pattern. */
2483 #ifdef AIX
2484 /* `const' makes AIX compiler fail. */
2486 #else
2488 #endif
2491 /* How to translate the characters in the pattern. */
2494 /* Address of the count-byte of the most recently inserted `exactn'
2495 command. This makes it possible to tell if a new exact-match
2496 character can be added to that command or if the character requires
2497 a new `exactn' command. */
2500 /* Address of start of the most recently finished expression.
2501 This tells, e.g., postfix * where to find the start of its
2502 operand. Reset at the beginning of groups and alternatives. */
2505 /* Address of beginning of regexp, or inside of last group. */
2508 /* Place in the uncompiled pattern (i.e., the {) to
2509 which to go back if the interval is invalid. */
2512 /* Address of the place where a forward jump should go to the end of
2513 the containing expression. Each alternative of an `or' -- except the
2514 last -- ends with a forward jump of this sort. */
2517 /* Work area for range table of charset. */
2520 /* If the object matched can contain multibyte characters. */
2523 /* Nonzero if we have pushed down into a subpattern. */
2526 /* These hold the values of p, pattern, and pend from the main
2527 pattern when we have pushed into a subpattern. */
2532 #ifdef DEBUG
2533 debug++;
2536 {
2542 }
2545 /* Initialize the compile stack. */
2556 /* Initialize the pattern buffer. */
2562 /* Set `used' to zero, so that if we return an error, the pattern
2563 printer (for debugging) will think there's no pattern. We reset it
2564 at the end. */
2567 /* Always count groups, whether or not bufp->no_sub is set. */
2570 #if !defined emacs && !defined SYNTAX_TABLE
2571 /* Initialize the syntax table. */
2573 #endif
2576 {
2579 enough space. This loses if buffer's address is bogus, but
2580 that is the user's responsibility. */
2582 }
2583 else
2586 }
2590 }
2594 /* Loop through the uncompiled pattern until we're at the end. */
2596 {
2598 {
2599 /* If this is the end of an included regexp,
2600 pop back to the main regexp and try again. */
2602 {
2608 }
2609 /* If this is the end of the main regexp, we are done. */
2611 }
2616 {
2618 {
2621 /* If there's no special whitespace regexp, treat
2622 spaces normally. And don't try to do this recursively. */
2626 /* Peek past following spaces. */
2628 {
2631 p1++;
2632 }
2633 /* If the spaces are followed by a repetition op,
2634 treat them normally. */
2640 /* Replace the spaces with the whitespace regexp. */
2648 }
2651 {
2654 /* If context independent, it's an operator. */
2656 /* Otherwise, depends on what's come before. */
2659 else
2661 }
2666 {
2668 p == pend
2669 /* If context independent, it's an operator. */
2671 /* Otherwise, depends on what's next. */
2674 else
2676 }
2685 handle_plus:
2687 /* If there is no previous pattern... */
2689 {
2694 }
2696 {
2697 /* 1 means zero (many) matches is allowed. */
2701 /* If there is a sequence of repetition chars, collapse it
2702 down to just one (the right one). We can't combine
2703 interval operators with these because of, e.g., `a{2}*',
2704 which should only match an even number of `a's. */
2707 {
2711 else
2712 {
2715 }
2722 ;
2724 {
2729 else
2731 }
2732 else
2734 /* If we get here, we found another repeat character. */
2736 }
2738 /* Star, etc. applied to an empty pattern is equivalent
2739 to an empty pattern. */
2743 /* Now we know whether or not zero matches is allowed
2744 and also whether or not two or more matches is allowed. */
2746 {
2748 {
2751 re_opcode_t ofj =
2752 /* Check if the loop can match the empty string. */
2759 on_failure_keep_string_jump, we turn simple P+
2760 into PP* if P is simple. */
2768 }
2772 /* A + loop. */
2774 else
2775 /* Simple * loops can use on_failure_keep_string_jump
2776 depending on what follows. But since we don't know
2777 that yet, we leave the decision up to
2778 on_failure_jump_smart. */
2784 }
2785 else
2786 {
2787 /* A simple ? pattern. */
2792 }
2793 }
2799 {
2802 /* The non-greedy multiple match looks like
2803 a repeat..until: we only need a conditional jump
2804 at the end of the loop. */
2810 {
2811 /* The repeat...until naturally matches one or more.
2812 To also match zero times, we need to first jump to
2813 the end of the loop (its conditional jump). */
2816 }
2817 }
2818 else
2819 {
2820 /* non-greedy a?? */
2825 }
2826 }
2827 }
2839 {
2846 /* Ensure that we have enough space to push a charset: the
2847 opcode, the length count, and the bitset; 34 bytes in all. */
2852 /* We test `*p == '^' twice, instead of using an if
2853 statement, so we only need one BUF_PUSH. */
2856 p++;
2858 /* Remember the first position in the bracket expression. */
2861 /* Push the number of bytes in the bitmap. */
2864 /* Clear the whole map. */
2867 /* charset_not matches newline according to a syntax bit. */
2872 /* Read in characters and ranges, setting map bits. */
2874 {
2877 re_wchar_t ch;
2881 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2882 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2883 So the translation is done later in a loop. Example:
2884 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2887 /* \ might escape characters inside [...] and [^...]. */
2889 {
2894 }
2895 else
2896 {
2897 /* Could be the end of the bracket expression. If it's
2898 not (i.e., when the bracket expression is `[]' so
2899 far), the ']' character bit gets set way below. */
2902 }
2904 /* See if we're at the beginning of a possible character
2905 class. */
2909 {
2910 /* Leave room for the null. */
2918 /* If pattern is `[[:'. */
2922 {
2928 else
2929 /* This is in any case an invalid class name. */
2931 }
2934 /* If isn't a word bracketed by `[:' and `:]':
2935 undo the ending character, the letters, and
2936 leave the leading `:' and `[' (but set bits for
2937 them). */
2939 {
2945 /* Throw away the ] at the end of the character
2946 class. */
2951 #ifndef emacs
2954 {
2958 }
2960 /* Most character classes in a multibyte match
2961 just set a flag. Exceptions are is_blank,
2962 is_digit, is_cntrl, and is_xdigit, since
2963 they can only match ASCII characters. We
2964 don't need to handle them for multibyte.
2965 They are distinguished by a negative wctype. */
2967 /* Setup the gl_state object to its buffer-defined
2968 value. This hardcodes the buffer-global
2969 syntax-table for ASCII chars, while the other chars
2970 will obey syntax-table properties. It's not ideal,
2971 but it's the way it's been done until now. */
2975 {
2979 {
2988 }
2989 }
2990 SET_RANGE_TABLE_WORK_AREA_BIT
2993 /* In most cases the matching rule for char classes
2994 only uses the syntax table for multibyte chars,
2995 so that the content of the syntax-table it is not
2996 hardcoded in the range_table. SPACE and WORD are
2997 the two exceptions. */
3001 /* Repeat the loop. */
3003 }
3004 else
3005 {
3006 /* Go back to right after the "[:". */
3010 /* Because the `:' may starts the range, we
3011 can't simply set bit and repeat the loop.
3012 Instead, just set it to C and handle below. */
3014 }
3015 }
3018 {
3020 /* Discard the `-'. */
3023 /* Fetch the character which ends the range. */
3025 #ifdef emacs
3028 /* Treat the range from a multibyte character to
3029 raw-byte character as empty. */
3032 }
3033 else
3034 /* Range from C to C. */
3038 {
3041 /* Else, repeat the loop. */
3042 }
3043 else
3044 {
3045 #ifndef emacs
3046 /* Set the range into bitmap */
3048 {
3052 }
3055 {
3061 }
3063 {
3065 {
3070 }
3072 {
3074 }
3075 else
3076 {
3078 }
3079 }
3081 }
3082 }
3084 /* Discard any (non)matching list bytes that are all 0 at the
3085 end of the map. Decrease the map-length byte too. */
3090 /* Build real range table from work area. */
3093 {
3097 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3098 bytes for flags, two for COUNT, and three bytes for
3099 each character. */
3102 /* Indicate the existence of range table. */
3105 /* Store the character class flag bits into the range table.
3106 If not in emacs, these flag bits are always 0. */
3112 STORE_CHARACTER_AND_INCR
3114 }
3115 }
3122 else
3129 else
3136 else
3143 else
3150 else
3157 /* Do not translate the character after the \, so that we can
3158 distinguish, e.g., \B from \b, even if we normally would
3159 translate, e.g., B to b. */
3163 {
3168 handle_open:
3169 {
3173 {
3174 /* Look for a special (?...) construct */
3176 {
3179 {
3182 {
3185 /* An explicitly specified regnum must start
3186 with non-0. */
3193 /* Only (?:...) is supported right now. */
3195 }
3196 }
3197 }
3198 }
3208 /* Ideally, we'd want to check that the specified
3209 group can't have matched (i.e. all subgroups
3210 using the same regnum are in other branches of
3211 OR patterns), but we don't currently keep track
3212 of enough info to do that easily. */
3215 }
3216 else
3217 /* It's really shy. */
3221 {
3223 compile_stack_elt_t);
3227 }
3229 /* These are the values to restore when we hit end of this
3230 group. They are all relative offsets, so that if the
3231 whole pattern moves because of realloc, they will still
3232 be valid. */
3234 COMPILE_STACK_TOP.fixup_alt_jump
3239 /* Do not push a start_memory for groups beyond the last one
3240 we can represent in the compiled pattern. */
3249 /* If we've reached MAX_REGNUM groups, then this open
3250 won't actually generate any code, so we'll have to
3251 clear pending_exact explicitly. */
3254 }
3260 {
3263 else
3265 }
3267 handle_close:
3270 /* See similar code for backslashed left paren above. */
3272 {
3275 else
3277 }
3279 /* Since we just checked for an empty stack above, this
3280 ``can't happen''. */
3282 {
3283 /* We don't just want to restore into `regnum', because
3284 later groups should continue to be numbered higher,
3285 as in `(ab)c(de)' -- the second group is #2. */
3286 regnum_t regnum;
3290 fixup_alt_jump
3296 /* If we've reached MAX_REGNUM groups, then this open
3297 won't actually generate any code, so we'll have to
3298 clear pending_exact explicitly. */
3301 /* We're at the end of the group, so now we know how many
3302 groups were inside this one. */
3305 }
3312 handle_alt:
3316 /* Insert before the previous alternative a jump which
3317 jumps to this alternative if the former fails. */
3323 /* The alternative before this one has a jump after it
3324 which gets executed if it gets matched. Adjust that
3325 jump so it will jump to this alternative's analogous
3326 jump (put in below, which in turn will jump to the next
3327 (if any) alternative's such jump, etc.). The last such
3328 jump jumps to the correct final destination. A picture:
3329 _____ _____
3330 | | | |
3331 | v | v
3332 a | b | c
3334 If we are at `b', then fixup_alt_jump right now points to a
3335 three-byte space after `a'. We'll put in the jump, set
3336 fixup_alt_jump to right after `b', and leave behind three
3337 bytes which we'll fill in when we get to after `c'. */
3341 /* Mark and leave space for a jump after this alternative,
3342 to be filled in later either by next alternative or
3343 when know we're at the end of a series of alternatives. */
3354 /* If \{ is a literal. */
3356 /* If we're at `\{' and it's not the open-interval
3357 operator. */
3361 handle_interval:
3362 {
3363 /* If got here, then the syntax allows intervals. */
3365 /* At least (most) this many matches must be made. */
3374 else
3375 /* Interval such as `{1}' => match exactly once. */
3383 {
3389 }
3394 /* We just parsed a valid interval. */
3396 /* If it's invalid to have no preceding re. */
3398 {
3403 else
3405 }
3408 /* If the upper bound is zero, just drop the sub pattern
3409 altogether. */
3412 /* Just match it once: nothing to do here. */
3413 ;
3415 /* Otherwise, we have a nontrivial interval. When
3416 we're all done, the pattern will look like:
3417 set_number_at <jump count> <upper bound>
3418 set_number_at <succeed_n count> <lower bound>
3419 succeed_n <after jump addr> <succeed_n count>
3420 <body of loop>
3421 jump_n <succeed_n addr> <jump count>
3422 (The upper bound and `jump_n' are omitted if
3423 `upper_bound' is 1, though.) */
3424 else
3426 more at the end of the loop. */
3434 {
3435 /* A succeed_n that starts with 0 is really a
3436 a simple on_failure_jump_loop. */
3440 }
3441 else
3442 {
3443 /* Initialize lower bound of the `succeed_n', even
3444 though it will be set during matching by its
3445 attendant `set_number_at' (inserted next),
3446 because `re_compile_fastmap' needs to know.
3447 Jump to the `jump_n' we might insert below. */
3450 lower_bound);
3453 /* Code to initialize the lower bound. Insert
3454 before the `succeed_n'. The `5' is the last two
3455 bytes of this `set_number_at', plus 3 bytes of
3456 the following `succeed_n'. */
3460 }
3463 {
3464 /* A negative upper bound stands for infinity,
3465 in which case it degenerates to a plain jump. */
3468 }
3471 append a backward jump to the `succeed_n'
3472 that starts this interval.
3474 When we've reached this during matching,
3475 we'll have matched the interval once, so
3476 jump back only `upper_bound - 1' times. */
3481 /* The location we want to set is the second
3482 parameter of the `jump_n'; that is `b-2' as
3483 an absolute address. `laststart' will be
3484 the `set_number_at' we're about to insert;
3485 `laststart+3' the number to set, the source
3486 for the relative address. But we are
3487 inserting into the middle of the pattern --
3488 so everything is getting moved up by 5.
3489 Conclusion: (b - 2) - (laststart + 3) + 5,
3490 i.e., b - laststart.
3492 We insert this at the beginning of the loop
3493 so that if we fail during matching, we'll
3494 reinitialize the bounds. */
3498 }
3499 }
3502 }
3505 unfetch_interval:
3506 /* If an invalid interval, match the characters as literals. */
3511 /* normal_char and normal_backslash need `c'. */
3515 {
3518 }
3519 else
3522 #ifdef emacs
3523 /* There is no way to specify the before_dot and after_dot
3524 operators. rms says this is ok. --karl */
3592 else
3622 {
3623 regnum_t reg;
3631 /* Can't back reference to a subexp before its end. */
3637 }
3645 else
3649 normal_backslash:
3650 /* You might think it would be useful for \ to mean
3651 not to translate; but if we don't translate it
3652 it will never match anything. */
3654 }
3659 /* Expects the character in `c'. */
3660 normal_char:
3661 /* If no exactn currently being built. */
3664 /* If last exactn not at current position. */
3667 /* We have only one byte following the exactn for the count. */
3670 /* If followed by a repetition operator. */
3679 {
3680 /* Start building a new exactn. */
3686 }
3689 {
3693 {
3697 }
3698 else
3699 {
3702 {
3707 }
3710 }
3712 }
3719 /* Through the pattern now. */
3726 /* If we don't want backtracking, force success
3727 the first time we reach the end of the compiled pattern. */
3731 /* We have succeeded; set the length of the buffer. */
3734 #ifdef DEBUG
3736 {
3740 }
3741 debug--;
3744 #ifndef MATCH_MAY_ALLOCATE
3745 /* Initialize the failure stack to the largest possible stack. This
3746 isn't necessary unless we're trying to avoid calling alloca in
3747 the search and match routines. */
3748 {
3752 {
3756 fail_stack.stack
3759 else
3760 fail_stack.stack
3764 }
3767 }
3772 \f
3773 /* Subroutines for `regex_compile'. */
3775 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3777 static void
3779 {
3782 }
3785 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3787 static void
3789 {
3793 }
3796 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3797 for OP followed by two-byte integer parameter ARG. */
3799 static void
3801 {
3809 }
3812 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3814 static void
3816 {
3824 }
3827 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3828 after an alternative or a begin-subexpression. We assume there is at
3829 least one character before the ^. */
3831 static boolean
3833 {
3837 return
3838 /* After a subexpression? */
3840 /* After an alternative? */
3842 /* After a shy subexpression? */
3847 }
3850 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3851 at least one character after the $, i.e., `P < PEND'. */
3853 static boolean
3855 {
3860 return
3861 /* Before a subexpression? */
3864 /* Before an alternative? */
3867 }
3870 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3871 false if it's not. */
3873 static boolean
3875 {
3876 ssize_t this_element;
3880 this_element--)
3885 }
3886 \f
3887 /* analyse_first.
3888 If fastmap is non-NULL, go through the pattern and fill fastmap
3889 with all the possible leading chars. If fastmap is NULL, don't
3890 bother filling it up (obviously) and only return whether the
3891 pattern could potentially match the empty string.
3893 Return 1 if p..pend might match the empty string.
3894 Return 0 if p..pend matches at least one char.
3895 Return -1 if fastmap was not updated accurately. */
3897 static int
3899 {
3903 /* If all elements for base leading-codes in fastmap is set, this
3904 flag is set true. */
3909 /* The loop below works as follows:
3910 - It has a working-list kept in the PATTERN_STACK and which basically
3911 starts by only containing a pointer to the first operation.
3912 - If the opcode we're looking at is a match against some set of
3913 chars, then we add those chars to the fastmap and go on to the
3914 next work element from the worklist (done via `break').
3915 - If the opcode is a control operator on the other hand, we either
3916 ignore it (if it's meaningless at this point, such as `start_memory')
3917 or execute it (if it's a jump). If the jump has several destinations
3918 (i.e. `on_failure_jump'), then we push the other destination onto the
3919 worklist.
3920 We guarantee termination by ignoring backward jumps (more or less),
3921 so that `p' is monotonically increasing. More to the point, we
3922 never set `p' (or push) anything `<= p1'. */
3925 {
3926 /* `p1' is used as a marker of how far back a `on_failure_jump'
3927 can go without being ignored. It is normally equal to `p'
3928 (which prevents any backward `on_failure_jump') except right
3929 after a plain `jump', to allow patterns such as:
3930 0: jump 10
3931 3..9: <body>
3932 10: on_failure_jump 3
3933 as used for the *? operator. */
3937 {
3942 /* If the first character has to match a backreference, that means
3943 that the group was empty (since it already matched). Since this
3944 is the only case that interests us here, we can assume that the
3945 backreference must match the empty string. */
3946 p++;
3950 /* Following are the cases which match a character. These end
3951 with `break'. */
3955 {
3956 /* If multibyte is nonzero, the first byte of each
3957 character is an ASCII or a leading code. Otherwise,
3958 each byte is a character. Thus, this works in both
3959 cases. */
3962 {
3963 /* For the case of matching this unibyte regex
3964 against multibyte, we must set a leading code of
3965 the corresponding multibyte character. */
3969 }
3970 }
3975 /* We could put all the chars except for \n (and maybe \0)
3976 but we don't bother since it is generally not worth it. */
3983 {
3984 /* Chars beyond end of bitmap are possible matches. */
3988 }
3990 /* Fallthrough */
3999 #ifdef emacs
4001 which doesn't match the specified set of characters. */
4002 not
4003 ||
4004 /* If we can match a character class, we can match any
4005 multibyte characters. */
4009 {
4011 {
4016 }
4017 }
4021 {
4022 /* Set fastmap[I] to 1 where I is a leading code of each
4023 multibyte character in the range table. */
4027 /* Make P points the range table. `+ 2' is to skip flag
4028 bits for a character class. */
4031 /* Extract the number of ranges in range table into COUNT. */
4034 {
4035 /* Extract the start and end of each range. */
4043 }
4044 }
4045 #endif
4051 #ifndef emacs
4059 /* This match depends on text properties. These end with
4060 aborting optimizations. */
4072 /* Any leading code can possibly start a character which
4073 has or doesn't has the specified category. */
4075 {
4080 }
4083 /* All cases after this match the empty string. These end with
4084 `continue'. */
4107 /* Backward jumps can only go back to code that we've already
4108 visited. `re_compile' should make sure this is true. */
4112 {
4118 p++;
4122 };
4123 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4124 to jump back to "just after here". */
4125 /* Fallthrough */
4135 else
4137 We first go down the "straight" path so as to minimize
4138 stack usage when going through alternatives. */
4142 }
4147 /* This code simply does not properly handle forward jump_n. */
4150 /* jump_n can either jump or fall through. The (backward) jump
4151 case has already been handled, so we only need to look at the
4152 fallthrough case. */
4156 /* If N == 0, it should be an on_failure_jump_loop instead. */
4159 /* We only care about one iteration of the loop, so we don't
4160 need to consider the case where this behaves like an
4161 on_failure_jump. */
4180 /* Getting here means we have found the possible starting
4181 characters for one path of the pattern -- and that the empty
4182 string does not match. We need not follow this path further. */
4186 /* We reached the end without matching anything. */
4190 \f
4191 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4192 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4193 characters can start a string that matches the pattern. This fastmap
4194 is used by re_search to skip quickly over impossible starting points.
4196 Character codes above (1 << BYTEWIDTH) are not represented in the
4197 fastmap, but the leading codes are represented. Thus, the fastmap
4198 indicates which character sets could start a match.
4200 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4201 area as BUFP->fastmap.
4203 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4204 the pattern buffer.
4206 Returns 0 if we succeed, -2 if an internal error. */
4208 int
4210 {
4224 \f
4225 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4226 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4227 this memory for recording register information. STARTS and ENDS
4228 must be allocated using the malloc library routine, and must each
4229 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4231 If NUM_REGS == 0, then subsequent matches should allocate their own
4232 register data.
4234 Unless this function is called, the first search or match using
4235 PATTERN_BUFFER will allocate its own register data, without
4236 freeing the old data. */
4238 void
4239 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4240 {
4242 {
4247 }
4248 else
4249 {
4253 }
4254 }
4256 \f
4257 /* Searching routines. */
4259 /* Like re_search_2, below, but only one string is specified, and
4260 doesn't let you say where to stop matching. */
4262 regoff_t
4265 {
4268 }
4271 /* Head address of virtual concatenation of string. */
4272 #define HEAD_ADDR_VSTRING(P) \
4273 (((P) >= size1 ? string2 : string1))
4275 /* Address of POS in the concatenation of virtual string. */
4276 #define POS_ADDR_VSTRING(POS) \
4277 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4279 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4280 virtual concatenation of STRING1 and STRING2, starting first at index
4281 STARTPOS, then at STARTPOS + 1, and so on.
4283 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4285 RANGE is how far to scan while trying to match. RANGE = 0 means try
4286 only at STARTPOS; in general, the last start tried is STARTPOS +
4287 RANGE.
4289 In REGS, return the indices of the virtual concatenation of STRING1
4290 and STRING2 that matched the entire BUFP->buffer and its contained
4291 subexpressions.
4293 Do not consider matching one past the index STOP in the virtual
4294 concatenation of STRING1 and STRING2.
4296 We return either the position in the strings at which the match was
4297 found, -1 if no match, or -2 if error (such as failure
4298 stack overflow). */
4300 regoff_t
4304 {
4305 regoff_t val;
4312 boolean anchored_start;
4313 /* Nonzero if we are searching multibyte string. */
4316 /* Check for out-of-range STARTPOS. */
4320 /* Fix up RANGE if it might eventually take us outside
4321 the virtual concatenation of STRING1 and STRING2.
4322 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4328 /* If the search isn't to be a backwards one, don't waste time in a
4329 search for a pattern anchored at beginning of buffer. */
4331 {
4334 else
4336 }
4338 #ifdef emacs
4339 /* In a forward search for something that starts with \=.
4340 don't keep searching past point. */
4342 {
4346 }
4349 /* Update the fastmap now if not correct already. */
4353 /* See whether the pattern is anchored. */
4356 #ifdef emacs
4358 {
4362 }
4363 #endif
4365 /* Loop through the string, looking for a place to start matching. */
4367 {
4368 /* If the pattern is anchored,
4369 skip quickly past places we cannot match.
4370 We don't bother to treat startpos == 0 specially
4371 because that case doesn't repeat. */
4373 {
4378 }
4380 /* If a fastmap is supplied, skip quickly over characters that
4381 cannot be the start of a match. If the pattern can match the
4382 null string, however, we don't need to skip characters; we want
4383 the first null string. */
4385 {
4392 {
4399 /* Written out as an if-else to avoid testing `translate'
4400 inside the loop. */
4402 {
4405 {
4415 }
4416 else
4418 {
4429 d++;
4430 range--;
4431 }
4432 }
4433 else
4434 {
4437 {
4445 }
4446 else
4448 {
4449 d++;
4450 range--;
4451 }
4452 }
4454 }
4456 {
4458 {
4463 }
4464 else
4465 {
4476 }
4477 }
4478 }
4480 /* If can't match the null string, and that's all we have left, fail. */
4494 advance:
4498 {
4499 /* Update STARTPOS to the next character boundary. */
4501 {
4509 }
4510 else
4511 {
4512 range--;
4513 startpos++;
4514 }
4515 }
4516 else
4517 {
4518 range++;
4519 startpos--;
4521 /* Update STARTPOS to the previous character boundary. */
4523 {
4528 /* Find the head of multibyte form. */
4535 }
4536 }
4537 }
4541 \f
4542 /* Declarations and macros for re_match_2. */
4546 RE_TRANSLATE_TYPE translate,
4549 /* This converts PTR, a pointer into one of the search strings `string1'
4550 and `string2' into an offset from the beginning of that string. */
4551 #define POINTER_TO_OFFSET(ptr) \
4552 (FIRST_STRING_P (ptr) \
4553 ? ((regoff_t) ((ptr) - string1)) \
4554 : ((regoff_t) ((ptr) - string2 + size1)))
4556 /* Call before fetching a character with *d. This switches over to
4557 string2 if necessary.
4558 Check re_match_2_internal for a discussion of why end_match_2 might
4559 not be within string2 (but be equal to end_match_1 instead). */
4560 #define PREFETCH() \
4561 while (d == dend) \
4562 { \
4564 if (dend == end_match_2) \
4565 goto fail; \
4567 d = string2; \
4568 dend = end_match_2; \
4569 }
4571 /* Call before fetching a char with *d if you already checked other limits.
4572 This is meant for use in lookahead operations like wordend, etc..
4573 where we might need to look at parts of the string that might be
4574 outside of the LIMITs (i.e past `stop'). */
4575 #define PREFETCH_NOLIMIT() \
4576 if (d == end1) \
4577 { \
4578 d = string2; \
4579 dend = end_match_2; \
4580 } \
4582 /* Test if at very beginning or at very end of the virtual concatenation
4583 of `string1' and `string2'. If only one string, it's `string2'. */
4584 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4585 #define AT_STRINGS_END(d) ((d) == end2)
4587 /* Disabled due to a compiler bug -- see comment at case wordbound */
4589 /* The comment at case wordbound is following one, but we don't use
4590 AT_WORD_BOUNDARY anymore to support multibyte form.
4592 The DEC Alpha C compiler 3.x generates incorrect code for the
4593 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4594 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4595 macro and introducing temporary variables works around the bug. */
4597 #if 0
4598 /* Test if D points to a character which is word-constituent. We have
4599 two special cases to check for: if past the end of string1, look at
4600 the first character in string2; and if before the beginning of
4601 string2, look at the last character in string1. */
4602 #define WORDCHAR_P(d) \
4603 (SYNTAX ((d) == end1 ? *string2 \
4604 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4605 == Sword)
4607 /* Test if the character before D and the one at D differ with respect
4608 to being word-constituent. */
4609 #define AT_WORD_BOUNDARY(d) \
4610 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4611 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4612 #endif
4614 /* Free everything we malloc. */
4615 #ifdef MATCH_MAY_ALLOCATE
4616 # define FREE_VAR(var) \
4617 do { \
4618 if (var) \
4619 { \
4620 REGEX_FREE (var); \
4621 var = NULL; \
4622 } \
4623 } while (0)
4624 # define FREE_VARIABLES() \
4625 do { \
4626 REGEX_FREE_STACK (fail_stack.stack); \
4627 FREE_VAR (regstart); \
4628 FREE_VAR (regend); \
4629 FREE_VAR (best_regstart); \
4630 FREE_VAR (best_regend); \
4631 } while (0)
4632 #else
4636 \f
4637 /* Optimization routines. */
4639 /* If the operation is a match against one or more chars,
4640 return a pointer to the next operation, else return NULL. */
4643 {
4645 {
4656 {
4661 }
4662 else
4668 #ifdef emacs
4672 p++;
4677 }
4679 }
4682 /* Jump over non-matching operations. */
4685 {
4688 {
4690 {
4703 }
4704 }
4707 }
4709 /* Non-zero if "p1 matches something" implies "p2 fails". */
4710 static int
4712 {
4713 re_opcode_t op2;
4720 /* Skip over open/close-group commands.
4721 If what follows this loop is a ...+ construct,
4722 look at what begins its body, since we will have to
4723 match at least one of that. */
4725 /* The same skip can be done for p1, except that this function
4726 is only used in the case where p1 is a simple match operator. */
4727 /* p1 = skip_noops (p1, pend); */
4735 {
4738 /* If we're at the end of the pattern, we can change. */
4740 {
4743 }
4748 {
4749 register re_wchar_t c
4754 {
4756 {
4759 }
4760 }
4764 {
4767 /* Test if C is listed in charset (or charset_not)
4768 at `p1'. */
4770 {
4774 }
4778 /* `not' is equal to 1 if c would match, which means
4779 that we can't change to pop_failure_jump. */
4781 {
4784 }
4785 }
4788 {
4791 }
4792 }
4796 {
4798 /* Reuse the code above. */
4801 /* It is hard to list up all the character in charset
4802 P2 if it includes multibyte character. Give up in
4803 such case. */
4805 {
4806 /* Now, we are sure that P2 has no range table.
4807 So, for the size of bitmap in P2, `p2[1]' is
4808 enough. But P1 may have range table, so the
4809 size of bitmap table of P1 is extracted by
4810 using macro `CHARSET_BITMAP_SIZE'.
4812 In a multibyte case, we know that all the character
4813 listed in P2 is ASCII. In a unibyte case, P1 has only a
4814 bitmap table. So, in both cases, it is enough to test
4815 only the bitmap table of P1. */
4818 {
4820 /* We win if the charset inside the loop
4821 has no overlap with the one after the loop. */
4825 idx++)
4831 {
4834 }
4835 }
4837 {
4839 /* We win if the charset_not inside the loop lists
4840 every character listed in the charset after. */
4848 {
4851 }
4852 }
4853 }
4854 }
4859 {
4862 /* Reuse the code above. */
4865 /* When we have two charset_not, it's very unlikely that
4866 they don't overlap. The union of the two sets of excluded
4867 chars should cover all possible chars, which, as a matter of
4868 fact, is virtually impossible in multibyte buffers. */
4870 }
4894 #ifdef emacs
4902 ;
4903 }
4905 /* Safe default. */
4907 }
4909 \f
4910 /* Matching routines. */
4913 /* re_match is like re_match_2 except it takes only a single string. */
4915 regoff_t
4918 {
4922 }
4926 #ifdef emacs
4927 /* In Emacs, this is the string or buffer in which we
4928 are matching. It is used for looking up syntax properties. */
4929 Lisp_Object re_match_object;
4930 #endif
4932 /* re_match_2 matches the compiled pattern in BUFP against the
4933 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4934 and SIZE2, respectively). We start matching at POS, and stop
4935 matching at STOP.
4937 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4938 store offsets for the substring each group matched in REGS. See the
4939 documentation for exactly how many groups we fill.
4941 We return -1 if no match, -2 if an internal error (such as the
4942 failure stack overflowing). Otherwise, we return the length of the
4943 matched substring. */
4945 regoff_t
4949 {
4950 regoff_t result;
4952 #ifdef emacs
4953 ssize_t charpos;
4957 #endif
4963 }
4967 /* This is a separate function so that we can force an alloca cleanup
4968 afterwards. */
4969 static regoff_t
4973 {
4974 /* General temporaries. */
4975 ssize_t mcnt;
4978 /* Just past the end of the corresponding string. */
4981 /* Pointers into string1 and string2, just past the last characters in
4982 each to consider matching. */
4985 /* Where we are in the data, and the end of the current string. */
4988 /* Used sometimes to remember where we were before starting matching
4989 an operator so that we can go back in case of failure. This "atomic"
4990 behavior of matching opcodes is indispensable to the correctness
4991 of the on_failure_keep_string_jump optimization. */
4994 /* Where we are in the pattern, and the end of the pattern. */
4998 /* We use this to map every character in the string. */
5001 /* Nonzero if BUFP is setup from a multibyte regex. */
5004 /* Nonzero if STRING1/STRING2 are multibyte. */
5007 /* Failure point stack. Each place that can handle a failure further
5008 down the line pushes a failure point on this stack. It consists of
5009 regstart, and regend for all registers corresponding to
5010 the subexpressions we're currently inside, plus the number of such
5011 registers, and, finally, two char *'s. The first char * is where
5012 to resume scanning the pattern; the second one is where to resume
5013 scanning the strings. */
5015 fail_stack_type fail_stack;
5016 #endif
5017 #ifdef DEBUG
5019 #endif
5021 #if defined REL_ALLOC && defined REGEX_MALLOC
5022 /* This holds the pointer to the failure stack, when
5023 it is allocated relocatably. */
5025 #endif
5027 /* We fill all the registers internally, independent of what we
5028 return, for use in backreferences. The number here includes
5029 an element for register zero. */
5032 /* Information on the contents of registers. These are pointers into
5033 the input strings; they record just what was matched (on this
5034 attempt) by a subexpression part of the pattern, that is, the
5035 regnum-th regstart pointer points to where in the pattern we began
5036 matching and the regnum-th regend points to right after where we
5037 stopped matching the regnum-th subexpression. (The zeroth register
5038 keeps track of what the whole pattern matches.) */
5041 #endif
5043 /* The following record the register info as found in the above
5044 variables when we find a match better than any we've seen before.
5045 This happens as we backtrack through the failure points, which in
5046 turn happens only if we have not yet matched the entire string. */
5050 #endif
5052 /* Logically, this is `best_regend[0]'. But we don't want to have to
5053 allocate space for that if we're not allocating space for anything
5054 else (see below). Also, we never need info about register 0 for
5055 any of the other register vectors, and it seems rather a kludge to
5056 treat `best_regend' differently than the rest. So we keep track of
5057 the end of the best match so far in a separate variable. We
5058 initialize this to NULL so that when we backtrack the first time
5059 and need to test it, it's not garbage. */
5062 #ifdef DEBUG
5063 /* Counts the total number of registers pushed. */
5065 #endif
5071 #ifdef MATCH_MAY_ALLOCATE
5072 /* Do not bother to initialize all the register variables if there are
5073 no groups in the pattern, as it takes a fair amount of time. If
5074 there are groups, we include space for register 0 (the whole
5075 pattern), even though we never use it, since it simplifies the
5076 array indexing. We should fix this. */
5078 {
5085 {
5088 }
5089 }
5090 else
5091 {
5092 /* We must initialize all our variables to NULL, so that
5093 `FREE_VARIABLES' doesn't try to free them. */
5095 }
5098 /* The starting position is bogus. */
5100 {
5103 }
5105 /* Initialize subexpression text positions to -1 to mark ones that no
5106 start_memory/stop_memory has been seen for. Also initialize the
5107 register information struct. */
5111 /* We move `string1' into `string2' if the latter's empty -- but not if
5112 `string1' is null. */
5114 {
5119 }
5123 /* `p' scans through the pattern as `d' scans through the data.
5124 `dend' is the end of the input string that `d' points within. `d'
5125 is advanced into the following input string whenever necessary, but
5126 this happens before fetching; therefore, at the beginning of the
5127 loop, `d' can be pointing at the end of a string, but it cannot
5128 equal `string2'. */
5130 {
5131 /* Only match within string2. */
5135 }
5136 else
5137 {
5139 {
5140 /* Only match within string1. */
5142 /* BEWARE!
5143 When we reach end_match_1, PREFETCH normally switches to string2.
5144 But in the present case, this means that just doing a PREFETCH
5145 makes us jump from `stop' to `gap' within the string.
5146 What we really want here is for the search to stop as
5147 soon as we hit end_match_1. That's why we set end_match_2
5148 to end_match_1 (since PREFETCH fails as soon as we hit
5149 end_match_2). */
5151 }
5152 else
5154 moving `d' from end1 to string2 will not prevent the d == dend
5155 check from catching the end of string. */
5158 }
5161 }
5169 /* This loops over pattern commands. It exits by returning from the
5170 function if the match is complete, or it drops through if the match
5171 fails at this starting point in the input data. */
5173 {
5180 /* If we haven't matched the entire string, and we want the
5181 longest match, try backtracking. */
5183 {
5184 /* 1 if this match ends in the same string (string1 or string2)
5185 as the best previous match. */
5188 /* 1 if this match is the best seen so far. */
5189 boolean best_match_p;
5191 /* AIX compiler got confused when this was combined
5192 with the previous declaration. */
5195 else
5203 /* If exceeds best match so far, save it. */
5205 {
5212 {
5215 }
5216 }
5218 }
5220 /* If no failure points, don't restore garbage. And if
5221 last match is real best match, don't restore second
5222 best one. */
5224 {
5225 restore_best_regs:
5226 /* Restore best match. It may happen that `dend ==
5227 end_match_1' while the restored d is in string2.
5228 For example, the pattern `x.*y.*z' against the
5229 strings `x-' and `y-z-', if the two strings are
5230 not consecutive in memory. */
5238 {
5241 }
5242 }
5245 succeed_label:
5248 /* If caller wants register contents data back, do it. */
5250 {
5251 /* Have the register data arrays been allocated? */
5254 extra element beyond `num_regs' for the `-1' marker
5255 GNU code uses. */
5260 {
5263 }
5265 }
5268 allocated, reallocate them. If we need fewer, just
5269 leave it alone. */
5271 {
5276 {
5279 }
5280 }
5281 }
5282 else
5283 {
5284 /* These braces fend off a "empty body in an else-statement"
5285 warning under GCC when assert expands to nothing. */
5287 }
5289 /* Convert the pointer data in `regstart' and `regend' to
5290 indices. Register zero has to be set differently,
5291 since we haven't kept track of any info for it. */
5293 {
5296 }
5298 /* Go through the first `min (num_regs, regs->num_regs)'
5299 registers, since that is all we initialized. */
5301 {
5304 else
5305 {
5310 }
5311 }
5313 /* If the regs structure we return has more elements than
5314 were in the pattern, set the extra elements to -1. If
5315 we (re)allocated the registers, this is the case,
5316 because we always allocate enough to have at least one
5317 -1 at the end. */
5333 }
5335 /* Otherwise match next pattern command. */
5337 {
5338 /* Ignore these. Used to ignore the n of succeed_n's which
5339 currently have n == 0. */
5348 /* Match the next n pattern characters exactly. The following
5349 byte in the pattern defines n, and the n bytes after that
5350 are the characters to match. */
5355 /* Remember the start point to rollback upon failure. */
5358 #ifndef emacs
5359 /* This is written out as an if-else so we don't waste time
5360 testing `translate' inside the loop. */
5362 do
5363 {
5366 {
5369 }
5370 d++;
5371 }
5373 else
5374 do
5375 {
5378 {
5381 }
5382 }
5385 /* The cost of testing `translate' is comparatively small. */
5387 do
5388 {
5395 else
5396 {
5399 }
5403 {
5406 }
5411 }
5413 else
5414 do
5415 {
5421 {
5424 }
5425 else
5426 {
5429 }
5432 {
5437 }
5438 else
5441 {
5444 }
5446 d++;
5447 }
5449 #endif
5453 /* Match any character except possibly a newline or a null. */
5455 {
5457 re_wchar_t buf_ch;
5463 target_multibyte);
5474 }
5480 {
5485 /* Start of actual range_table, or end of bitmap if there is no
5486 range table. */
5489 /* Nonzero if there is a range table. */
5492 /* Number of ranges of range table. This is not included
5493 in the initial byte-length of the command. */
5496 /* Whether matching against a unibyte character. */
5504 {
5507 }
5512 {
5518 {
5521 }
5522 }
5523 else
5524 {
5528 {
5532 {
5535 }
5536 }
5537 else
5539 }
5543 /* Cast to `unsigned' instead of `unsigned char' in
5544 case the bit list is a full 32 bytes long. */
5548 }
5549 #ifdef emacs
5551 {
5561 else
5563 }
5568 else
5574 }
5578 /* The beginning of a group is represented by start_memory.
5579 The argument is the register number. The text
5580 matched within the group is recorded (in the internal
5581 registers data structure) under the register number. */
5585 /* In case we need to undo this operation (via backtracking). */
5592 /* Move past the register number and inner group count. */
5597 /* The stop_memory opcode represents the end of a group. Its
5598 argument is the same as start_memory's: the register number. */
5603 /* Strictly speaking, there should be code such as:
5605 assert (REG_UNSET (regend[*p]));
5606 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5608 But the only info to be pushed is regend[*p] and it is known to
5609 be UNSET, so there really isn't anything to push.
5610 Not pushing anything, on the other hand deprives us from the
5611 guarantee that regend[*p] is UNSET since undoing this operation
5612 will not reset its value properly. This is not important since
5613 the value will only be read on the next start_memory or at
5614 the very end and both events can only happen if this stop_memory
5615 is *not* undone. */
5620 /* Move past the register number and the inner group count. */
5625 /* \<digit> has been turned into a `duplicate' command which is
5626 followed by the numeric value of <digit> as the register number. */
5628 {
5633 /* Can't back reference a group which we've never matched. */
5637 /* Where in input to try to start matching. */
5640 /* Remember the start point to rollback upon failure. */
5643 /* Where to stop matching; if both the place to start and
5644 the place to stop matching are in the same string, then
5645 set to the place to stop, otherwise, for now have to use
5646 the end of the first string. */
5652 {
5653 /* If necessary, advance to next segment in register
5654 contents. */
5656 {
5660 /* End of string1 => advance to string2. */
5663 }
5664 /* At end of register contents => success */
5667 /* If necessary, advance to next segment in data. */
5670 /* How many characters left in this segment to match. */
5673 /* Want how many consecutive characters we can match in
5674 one shot, so, if necessary, adjust the count. */
5678 /* Compare that many; failure if mismatch, else move
5679 past them. */
5683 {
5686 }
5688 }
5689 }
5693 /* begline matches the empty string at the beginning of the string
5694 (unless `not_bol' is set in `bufp'), and after newlines. */
5699 {
5701 }
5702 else
5703 {
5708 }
5709 /* In all other cases, we fail. */
5713 /* endline is the dual of begline. */
5718 {
5720 }
5721 else
5722 {
5726 }
5730 /* Match at the very beginning of the data. */
5738 /* Match at the very end of the data. */
5746 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5747 pushes NULL as the value for the string on the stack. Then
5748 `POP_FAILURE_POINT' will keep the current value for the
5749 string, instead of restoring it. To see why, consider
5750 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5751 then the . fails against the \n. But the next thing we want
5752 to do is match the \n against the \n; if we restored the
5753 string value, we would be back at the foo.
5755 Because this is used only in specific cases, we don't need to
5756 check all the things that `on_failure_jump' does, to make
5757 sure the right things get saved on the stack. Hence we don't
5758 share its code. The only reason to push anything on the
5759 stack at all is that otherwise we would have to change
5760 `anychar's code to do something besides goto fail in this
5761 case; that seems worse than this. */
5770 /* A nasty loop is introduced by the non-greedy *? and +?.
5771 With such loops, the stack only ever contains one failure point
5772 at a time, so that a plain on_failure_jump_loop kind of
5773 cycle detection cannot work. Worse yet, such a detection
5774 can not only fail to detect a cycle, but it can also wrongly
5775 detect a cycle (between different instantiations of the same
5776 loop).
5777 So the method used for those nasty loops is a little different:
5778 We use a special cycle-detection-stack-frame which is pushed
5779 when the on_failure_jump_nastyloop failure-point is *popped*.
5780 This special frame thus marks the beginning of one iteration
5781 through the loop and we can hence easily check right here
5782 whether something matched between the beginning and the end of
5783 the loop. */
5790 {
5794 /* If there's a cycle, just continue without pushing
5795 this failure point. The failure point is the "try again"
5796 option, which shouldn't be tried.
5797 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5799 }
5802 /* Simple loop detecting on_failure_jump: just check on the
5803 failure stack if the same spot was already hit earlier. */
5805 on_failure:
5809 {
5813 /* If there's a cycle, get out of the loop, as if the matching
5814 had failed. We used to just `goto fail' here, but that was
5815 aborting the search a bit too early: we want to keep the
5816 empty-loop-match and keep matching after the loop.
5817 We want (x?)*y\1z to match both xxyz and xxyxz. */
5819 else
5821 }
5825 /* Uses of on_failure_jump:
5827 Each alternative starts with an on_failure_jump that points
5828 to the beginning of the next alternative. Each alternative
5829 except the last ends with a jump that in effect jumps past
5830 the rest of the alternatives. (They really jump to the
5831 ending jump of the following alternative, because tensioning
5832 these jumps is a hassle.)
5834 Repeats start with an on_failure_jump that points past both
5835 the repetition text and either the following jump or
5836 pop_failure_jump back to this on_failure_jump. */
5845 /* This operation is used for greedy *.
5846 Compare the beginning of the repeat with what in the
5847 pattern follows its end. If we can establish that there
5848 is nothing that they would both match, i.e., that we
5849 would have to backtrack because of (as in, e.g., `a*a')
5850 then we can use a non-backtracking loop based on
5851 on_failure_keep_string_jump instead of on_failure_jump. */
5856 {
5858 /* Here, we discard `const', making re_match non-reentrant. */
5863 instruction once it's been changed. */
5867 /* Ensure this is a indeed the trivial kind of loop
5868 we are expecting. */
5873 {
5874 /* Use a fast `on_failure_keep_string_jump' loop. */
5878 }
5879 else
5880 {
5881 /* Default to a safe `on_failure_jump' loop. */
5884 }
5886 }
5889 /* Unconditionally jump (without popping any failure points). */
5891 unconditional_jump:
5892 IMMEDIATE_QUIT_CHECK;
5900 /* Have to succeed matching what follows at least n times.
5901 After that, handle like `on_failure_jump'. */
5903 /* Signedness doesn't matter since we only compare MCNT to 0. */
5907 /* Originally, mcnt is how many times we HAVE to succeed. */
5909 {
5910 /* Here, we discard `const', making re_match non-reentrant. */
5912 mcnt--;
5915 }
5916 else
5917 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5922 /* Signedness doesn't matter since we only compare MCNT to 0. */
5926 /* Originally, this is how many times we CAN jump. */
5928 {
5929 /* Here, we discard `const', making re_match non-reentrant. */
5931 mcnt--;
5934 }
5935 /* If don't have to jump any more, skip over the rest of command. */
5936 else
5941 {
5946 /* Here, we discard `const', making re_match non-reentrant. */
5948 /* Signedness doesn't matter since we only copy MCNT's bits . */
5953 }
5957 {
5961 /* We SUCCEED (or FAIL) in one of the following cases: */
5963 /* Case 1: D is at the beginning or the end of string. */
5966 else
5967 {
5968 /* C1 is the character before D, S1 is the syntax of C1, C2
5969 is the character at D, and S2 is the syntax of C2. */
5973 #ifdef emacs
5977 #endif
5980 #ifdef emacs
5982 #endif
5989 /* Case 3: Both of S1 and S2 are Sword, and macro
5990 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5993 }
5996 else
5998 }
6003 /* We FAIL in one of the following cases: */
6005 /* Case 1: D is at the end of string. */
6008 else
6009 {
6010 /* C1 is the character before D, S1 is the syntax of C1, C2
6011 is the character at D, and S2 is the syntax of C2. */
6015 #ifdef emacs
6019 #endif
6024 /* Case 2: S2 is not Sword. */
6028 /* Case 3: D is not at the beginning of string ... */
6030 {
6032 #ifdef emacs
6034 #endif
6037 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6038 returns 0. */
6041 }
6042 }
6048 /* We FAIL in one of the following cases: */
6050 /* Case 1: D is at the beginning of string. */
6053 else
6054 {
6055 /* C1 is the character before D, S1 is the syntax of C1, C2
6056 is the character at D, and S2 is the syntax of C2. */
6060 #ifdef emacs
6064 #endif
6068 /* Case 2: S1 is not Sword. */
6072 /* Case 3: D is not at the end of string ... */
6074 {
6077 #ifdef emacs
6079 #endif
6082 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6083 returns 0. */
6086 }
6087 }
6093 /* We FAIL in one of the following cases: */
6095 /* Case 1: D is at the end of string. */
6098 else
6099 {
6100 /* C1 is the character before D, S1 is the syntax of C1, C2
6101 is the character at D, and S2 is the syntax of C2. */
6104 #ifdef emacs
6108 #endif
6113 /* Case 2: S2 is neither Sword nor Ssymbol. */
6117 /* Case 3: D is not at the beginning of string ... */
6119 {
6121 #ifdef emacs
6123 #endif
6126 /* ... and S1 is Sword or Ssymbol. */
6129 }
6130 }
6136 /* We FAIL in one of the following cases: */
6138 /* Case 1: D is at the beginning of string. */
6141 else
6142 {
6143 /* C1 is the character before D, S1 is the syntax of C1, C2
6144 is the character at D, and S2 is the syntax of C2. */
6147 #ifdef emacs
6151 #endif
6155 /* Case 2: S1 is neither Ssymbol nor Sword. */
6159 /* Case 3: D is not at the end of string ... */
6161 {
6164 #ifdef emacs
6166 #endif
6169 /* ... and S2 is Sword or Ssymbol. */
6172 }
6173 }
6178 {
6183 #ifdef emacs
6184 {
6188 }
6189 #endif
6190 {
6192 re_wchar_t c;
6198 }
6199 }
6202 #ifdef emacs
6223 {
6230 {
6232 re_wchar_t c;
6237 }
6238 }
6245 }
6249 /* We goto here if a matching operation fails. */
6250 fail:
6251 IMMEDIATE_QUIT_CHECK;
6253 {
6255 /* A restart point is known. Restore to that state. */
6259 {
6267 /* Fallthrough */
6273 continue_failure_jump:
6279 /* A special frame used for nastyloops. */
6284 }
6290 }
6291 else
6302 \f
6303 /* Subroutine definitions for re_match_2. */
6305 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6306 bytes; nonzero otherwise. */
6308 static int
6311 {
6316 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6317 different lengths, but relying on a single `len' would break this. -sm */
6319 {
6331 }
6337 }
6338 \f
6339 /* Entry points for GNU code. */
6341 /* re_compile_pattern is the GNU regular expression compiler: it
6342 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6343 Returns 0 if the pattern was valid, otherwise an error string.
6345 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6346 are set in BUFP on entry.
6348 We call regex_compile to do the actual compilation. */
6353 {
6354 reg_errcode_t ret;
6356 /* GNU code is written to assume at least RE_NREGS registers will be set
6357 (and at least one extra will be -1). */
6360 /* And GNU code determines whether or not to get register information
6361 by passing null for the REGS argument to re_match, etc., not by
6362 setting no_sub. */
6370 }
6372 \f
6373 /* Entry points compatible with 4.2 BSD regex library. We don't define
6374 them unless specifically requested. */
6376 #if defined _REGEX_RE_COMP || defined _LIBC
6378 /* BSD has one and only one pattern buffer. */
6382 # ifdef _LIBC
6383 /* Make these definitions weak in libc, so POSIX programs can redefine
6384 these names if they don't use our functions, and still use
6385 regcomp/regexec below without link errors. */
6386 weak_function
6387 # endif
6389 {
6390 reg_errcode_t ret;
6393 {
6395 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6398 }
6401 {
6404 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6410 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6412 }
6414 /* Since `re_exec' always passes NULL for the `regs' argument, we
6415 don't need to initialize the pattern buffer fields which affect it. */
6422 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6424 }
6427 int
6428 # ifdef _LIBC
6429 weak_function
6430 # endif
6432 {
6434 return
6436 }
6438 \f
6439 /* POSIX.2 functions. Don't define these for Emacs. */
6441 #ifndef emacs
6443 /* regcomp takes a regular expression as a string and compiles it.
6445 PREG is a regex_t *. We do not expect any fields to be initialized,
6446 since POSIX says we shouldn't. Thus, we set
6448 `buffer' to the compiled pattern;
6449 `used' to the length of the compiled pattern;
6450 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6451 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6452 RE_SYNTAX_POSIX_BASIC;
6453 `fastmap' to an allocated space for the fastmap;
6454 `fastmap_accurate' to zero;
6455 `re_nsub' to the number of subexpressions in PATTERN.
6457 PATTERN is the address of the pattern string.
6459 CFLAGS is a series of bits which affect compilation.
6461 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6462 use POSIX basic syntax.
6464 If REG_NEWLINE is set, then . and [^...] don't match newline.
6465 Also, regexec will try a match beginning after every newline.
6467 If REG_ICASE is set, then we considers upper- and lowercase
6468 versions of letters to be equivalent when matching.
6470 If REG_NOSUB is set, then when PREG is passed to regexec, that
6471 routine will report only success or failure, and nothing about the
6472 registers.
6474 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6475 the return codes and their meanings.) */
6477 reg_errcode_t
6480 {
6481 reg_errcode_t ret;
6482 reg_syntax_t syntax
6486 /* regex_compile will allocate the space for the compiled pattern. */
6491 /* Try to allocate space for the fastmap. */
6495 {
6498 preg->translate
6504 /* Map uppercase characters to corresponding lowercase ones. */
6507 }
6508 else
6511 /* If REG_NEWLINE is set, newlines are treated differently. */
6516 }
6517 else
6522 /* POSIX says a null character in the pattern terminates it, so we
6523 can use strlen here in compiling the pattern. */
6526 /* POSIX doesn't distinguish between an unmatched open-group and an
6527 unmatched close-group: both are REG_EPAREN. */
6533 buffer. */
6539 }
6540 }
6542 }
6546 /* regexec searches for a given pattern, specified by PREG, in the
6547 string STRING.
6549 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6550 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6551 least NMATCH elements, and we set them to the offsets of the
6552 corresponding matched substrings.
6554 EFLAGS specifies `execution flags' which affect matching: if
6555 REG_NOTBOL is set, then ^ does not match at the beginning of the
6556 string; if REG_NOTEOL is set, then $ does not match at the end.
6558 We return 0 if we find a match and REG_NOMATCH if not. */
6560 reg_errcode_t
6563 {
6564 regoff_t ret;
6566 regex_t private_preg;
6575 /* The user has told us exactly how many registers to return
6576 information about, via `nmatch'. We have to pass that on to the
6577 matching routines. */
6581 {
6587 }
6589 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6590 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6591 was a little bit longer but still only matching the real part.
6592 This works because the `endline' will check for a '\n' and will find a
6593 '\0', correctly deciding that this is not the end of a line.
6594 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6595 a convenient '\0' there. For all we know, the string could be preceded
6596 by '\n' which would throw things off. */
6598 /* Perform the searching operation. */
6603 /* Copy the register information to the POSIX structure. */
6605 {
6607 {
6611 {
6614 }
6615 }
6617 /* If we needed the temporary register info, free the space now. */
6619 }
6621 /* We want zero return to mean success, unlike `re_search'. */
6623 }
6627 /* Returns a message corresponding to an error code, ERR_CODE, returned
6628 from either regcomp or regexec. We don't use PREG here.
6630 ERR_CODE was previously called ERRCODE, but that name causes an
6631 error with msvc8 compiler. */
6633 size_t
6635 {
6641 /* Only error codes returned by the rest of the code should be passed
6642 to this routine. If we are given anything else, or if other regex
6643 code generates an invalid error code, then the program has a bug.
6644 Dump core so we can fix it. */
6652 {
6654 {
6657 }
6658 else
6660 }
6663 }
6667 /* Free dynamically allocated space used by PREG. */
6669 void
6671 {
6684 }