| blob | 8033ab9edaa7a7cceebac778428346c9d891190c |
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. */
536 #define false 0
537 #define true 1
542 ssize_t pos,
544 ssize_t stop));
545 \f
546 /* These are the command codes that appear in compiled regular
547 expressions. Some opcodes are followed by argument bytes. A
548 command code can specify any interpretation whatsoever for its
549 arguments. Zero bytes may appear in the compiled regular expression. */
552 {
555 /* Succeed right away--no more backtracking. */
556 succeed,
558 /* Followed by one byte giving n, then by n literal bytes. */
559 exactn,
561 /* Matches any (more or less) character. */
562 anychar,
564 /* Matches any one char belonging to specified set. First
565 following byte is number of bitmap bytes. Then come bytes
566 for a bitmap saying which chars are in. Bits in each byte
567 are ordered low-bit-first. A character is in the set if its
568 bit is 1. A character too large to have a bit in the map is
569 automatically not in the set.
571 If the length byte has the 0x80 bit set, then that stuff
572 is followed by a range table:
573 2 bytes of flags for character sets (low 8 bits, high 8 bits)
574 See RANGE_TABLE_WORK_BITS below.
575 2 bytes, the number of pairs that follow (upto 32767)
576 pairs, each 2 multibyte characters,
577 each multibyte character represented as 3 bytes. */
578 charset,
580 /* Same parameters as charset, but match any character that is
581 not one of those specified. */
582 charset_not,
584 /* Start remembering the text that is matched, for storing in a
585 register. Followed by one byte with the register number, in
586 the range 0 to one less than the pattern buffer's re_nsub
587 field. */
588 start_memory,
590 /* Stop remembering the text that is matched and store it in a
591 memory register. Followed by one byte with the register
592 number, in the range 0 to one less than `re_nsub' in the
593 pattern buffer. */
594 stop_memory,
596 /* Match a duplicate of something remembered. Followed by one
597 byte containing the register number. */
598 duplicate,
600 /* Fail unless at beginning of line. */
601 begline,
603 /* Fail unless at end of line. */
604 endline,
606 /* Succeeds if at beginning of buffer (if emacs) or at beginning
607 of string to be matched (if not). */
608 begbuf,
610 /* Analogously, for end of buffer/string. */
611 endbuf,
613 /* Followed by two byte relative address to which to jump. */
614 jump,
616 /* Followed by two-byte relative address of place to resume at
617 in case of failure. */
618 on_failure_jump,
620 /* Like on_failure_jump, but pushes a placeholder instead of the
621 current string position when executed. */
622 on_failure_keep_string_jump,
624 /* Just like `on_failure_jump', except that it checks that we
625 don't get stuck in an infinite loop (matching an empty string
626 indefinitely). */
627 on_failure_jump_loop,
629 /* Just like `on_failure_jump_loop', except that it checks for
630 a different kind of loop (the kind that shows up with non-greedy
631 operators). This operation has to be immediately preceded
632 by a `no_op'. */
633 on_failure_jump_nastyloop,
635 /* A smart `on_failure_jump' used for greedy * and + operators.
636 It analyses the loop before which it is put and if the
637 loop does not require backtracking, it changes itself to
638 `on_failure_keep_string_jump' and short-circuits the loop,
639 else it just defaults to changing itself into `on_failure_jump'.
640 It assumes that it is pointing to just past a `jump'. */
641 on_failure_jump_smart,
643 /* Followed by two-byte relative address and two-byte number n.
644 After matching N times, jump to the address upon failure.
645 Does not work if N starts at 0: use on_failure_jump_loop
646 instead. */
647 succeed_n,
649 /* Followed by two-byte relative address, and two-byte number n.
650 Jump to the address N times, then fail. */
651 jump_n,
653 /* Set the following two-byte relative address to the
654 subsequent two-byte number. The address *includes* the two
655 bytes of number. */
656 set_number_at,
667 /* Matches any character whose syntax is specified. Followed by
668 a byte which contains a syntax code, e.g., Sword. */
669 syntaxspec,
671 /* Matches any character whose syntax is not that specified. */
672 notsyntaxspec
674 #ifdef emacs
679 /* Matches any character whose category-set contains the specified
680 category. The operator is followed by a byte which contains a
681 category code (mnemonic ASCII character). */
682 categoryspec,
684 /* Matches any character whose category-set does not contain the
685 specified category. The operator is followed by a byte which
686 contains the category code (mnemonic ASCII character). */
687 notcategoryspec
690 \f
691 /* Common operations on the compiled pattern. */
693 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
695 #define STORE_NUMBER(destination, number) \
696 do { \
697 (destination)[0] = (number) & 0377; \
698 (destination)[1] = (number) >> 8; \
699 } while (0)
701 /* Same as STORE_NUMBER, except increment DESTINATION to
702 the byte after where the number is stored. Therefore, DESTINATION
703 must be an lvalue. */
705 #define STORE_NUMBER_AND_INCR(destination, number) \
706 do { \
707 STORE_NUMBER (destination, number); \
708 (destination) += 2; \
709 } while (0)
711 /* Put into DESTINATION a number stored in two contiguous bytes starting
712 at SOURCE. */
714 #define EXTRACT_NUMBER(destination, source) \
715 do { \
716 (destination) = *(source) & 0377; \
717 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
718 } while (0)
720 #ifdef DEBUG
722 static void
726 {
730 }
733 # undef EXTRACT_NUMBER
734 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
739 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
740 SOURCE must be an lvalue. */
742 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
743 do { \
744 EXTRACT_NUMBER (destination, source); \
745 (source) += 2; \
746 } while (0)
748 #ifdef DEBUG
751 static void
755 {
758 }
760 # ifndef EXTRACT_MACROS
761 # undef EXTRACT_NUMBER_AND_INCR
762 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
763 extract_number_and_incr (&dest, &src)
767 \f
768 /* Store a multibyte character in three contiguous bytes starting
769 DESTINATION, and increment DESTINATION to the byte after where the
770 character is stored. Therefore, DESTINATION must be an lvalue. */
772 #define STORE_CHARACTER_AND_INCR(destination, character) \
773 do { \
774 (destination)[0] = (character) & 0377; \
775 (destination)[1] = ((character) >> 8) & 0377; \
776 (destination)[2] = (character) >> 16; \
777 (destination) += 3; \
778 } while (0)
780 /* Put into DESTINATION a character stored in three contiguous bytes
781 starting at SOURCE. */
783 #define EXTRACT_CHARACTER(destination, source) \
784 do { \
785 (destination) = ((source)[0] \
786 | ((source)[1] << 8) \
787 | ((source)[2] << 16)); \
788 } while (0)
791 /* Macros for charset. */
793 /* Size of bitmap of charset P in bytes. P is a start of charset,
794 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
795 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
797 /* Nonzero if charset P has range table. */
798 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
800 /* Return the address of range table of charset P. But not the start
801 of table itself, but the before where the number of ranges is
802 stored. `2 +' means to skip re_opcode_t and size of bitmap,
803 and the 2 bytes of flags at the start of the range table. */
804 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
806 /* Extract the bit flags that start a range table. */
807 #define CHARSET_RANGE_TABLE_BITS(p) \
808 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
809 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
811 /* Return the address of end of RANGE_TABLE. COUNT is number of
812 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
813 is start of range and end of range. `* 3' is size of each start
814 and end. */
815 #define CHARSET_RANGE_TABLE_END(range_table, count) \
816 ((range_table) + (count) * 2 * 3)
818 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
819 COUNT is number of ranges in RANGE_TABLE. */
820 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
821 do \
822 { \
823 re_wchar_t range_start, range_end; \
824 re_char *rtp; \
825 re_char *range_table_end \
826 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
827 \
828 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
829 { \
830 EXTRACT_CHARACTER (range_start, rtp); \
831 EXTRACT_CHARACTER (range_end, rtp + 3); \
832 \
833 if (range_start <= (c) && (c) <= range_end) \
834 { \
835 (not) = !(not); \
836 break; \
837 } \
838 } \
839 } \
840 while (0)
842 /* Test if C is in range table of CHARSET. The flag NOT is negated if
843 C is listed in it. */
844 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
845 do \
846 { \
848 int count; \
849 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
850 \
851 EXTRACT_NUMBER_AND_INCR (count, range_table); \
852 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
853 } \
854 while (0)
855 \f
856 /* If DEBUG is defined, Regex prints many voluminous messages about what
857 it is doing (if the variable `debug' is nonzero). If linked with the
858 main program in `iregex.c', you can enter patterns and strings
859 interactively. And if linked with the main program in `main.c' and
860 the other test files, you can run the already-written tests. */
862 #ifdef DEBUG
864 /* We use standard I/O for debugging. */
865 # include <stdio.h>
867 /* It is useful to test things that ``must'' be true when debugging. */
868 # include <assert.h>
872 # define DEBUG_STATEMENT(e) e
873 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
874 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
875 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
876 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
877 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
878 if (debug > 0) print_partial_compiled_pattern (s, e)
879 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
880 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
883 /* Print the fastmap in human-readable form. */
885 void
888 {
893 {
895 {
899 {
901 i++;
902 }
904 {
907 }
908 }
909 }
911 }
914 /* Print a compiled pattern string in human-readable form, starting at
915 the START pointer into it and ending just before the pointer END. */
917 void
921 {
927 {
930 }
932 /* Loop over pattern commands. */
934 {
938 {
950 do
951 {
953 }
975 {
990 {
991 /* Are we starting a range? */
993 {
996 }
997 /* Have we broken a range? */
999 {
1002 }
1008 }
1018 {
1022 /* ??? Should print the range table; for now, just skip it. */
1026 }
1027 }
1122 # ifdef emacs
1158 }
1161 }
1164 }
1167 void
1170 {
1178 {
1181 }
1191 /* Perhaps we should print the translate table? */
1192 }
1195 void
1200 ssize_t size1;
1201 ssize_t size2;
1202 {
1203 ssize_t this_char;
1207 else
1208 {
1210 {
1215 }
1219 }
1220 }
1224 # undef assert
1225 # define assert(e)
1227 # define DEBUG_STATEMENT(e)
1228 # define DEBUG_PRINT1(x)
1229 # define DEBUG_PRINT2(x1, x2)
1230 # define DEBUG_PRINT3(x1, x2, x3)
1231 # define DEBUG_PRINT4(x1, x2, x3, x4)
1232 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1233 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1236 \f
1237 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1238 #ifdef lint
1239 # define IF_LINT(Code) Code
1240 #else
1242 #endif
1243 \f
1244 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1245 also be assigned to arbitrarily: each pattern buffer stores its own
1246 syntax, so it can be changed between regex compilations. */
1247 /* This has no initializer because initialized variables in Emacs
1248 become read-only after dumping. */
1249 reg_syntax_t re_syntax_options;
1252 /* Specify the precise syntax of regexps for compilation. This provides
1253 for compatibility for various utilities which historically have
1254 different, incompatible syntaxes.
1256 The argument SYNTAX is a bit mask comprised of the various bits
1257 defined in regex.h. We return the old syntax. */
1259 reg_syntax_t
1261 {
1266 }
1269 /* Regexp to use to replace spaces, or NULL meaning don't. */
1272 void
1274 {
1276 }
1278 \f
1279 /* This table gives an error message for each of the error codes listed
1280 in regex.h. Obviously the order here has to be same as there.
1281 POSIX doesn't require that we do anything for REG_NOERROR,
1282 but why not be nice? */
1285 {
1304 };
1305 \f
1306 /* Avoiding alloca during matching, to placate r_alloc. */
1308 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1309 searching and matching functions should not call alloca. On some
1310 systems, alloca is implemented in terms of malloc, and if we're
1311 using the relocating allocator routines, then malloc could cause a
1312 relocation, which might (if the strings being searched are in the
1313 ralloc heap) shift the data out from underneath the regexp
1314 routines.
1316 Here's another reason to avoid allocation: Emacs
1317 processes input from X in a signal handler; processing X input may
1318 call malloc; if input arrives while a matching routine is calling
1319 malloc, then we're scrod. But Emacs can't just block input while
1320 calling matching routines; then we don't notice interrupts when
1321 they come in. So, Emacs blocks input around all regexp calls
1322 except the matching calls, which it leaves unprotected, in the
1323 faith that they will not malloc. */
1325 /* Normally, this is fine. */
1326 #define MATCH_MAY_ALLOCATE
1328 /* The match routines may not allocate if (1) they would do it with malloc
1329 and (2) it's not safe for them to use malloc.
1330 Note that if REL_ALLOC is defined, matching would not use malloc for the
1331 failure stack, but we would still use it for the register vectors;
1332 so REL_ALLOC should not affect this. */
1333 #if defined REGEX_MALLOC && defined emacs
1334 # undef MATCH_MAY_ALLOCATE
1335 #endif
1337 \f
1338 /* Failure stack declarations and macros; both re_compile_fastmap and
1339 re_match_2 use a failure stack. These have to be macros because of
1340 REGEX_ALLOCATE_STACK. */
1343 /* Approximate number of failure points for which to initially allocate space
1344 when matching. If this number is exceeded, we allocate more
1345 space, so it is not a hard limit. */
1346 #ifndef INIT_FAILURE_ALLOC
1347 # define INIT_FAILURE_ALLOC 20
1348 #endif
1350 /* Roughly the maximum number of failure points on the stack. Would be
1351 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1352 This is a variable only so users of regex can assign to it; we never
1353 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1354 before using it, so it should probably be a byte-count instead. */
1355 # if defined MATCH_MAY_ALLOCATE
1356 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1357 whose default stack limit is 2mb. In order for a larger
1358 value to work reliably, you have to try to make it accord
1359 with the process stack limit. */
1361 # else
1363 # endif
1365 union fail_stack_elt
1366 {
1368 /* This should be the biggest `int' that's no bigger than a pointer. */
1370 };
1375 {
1382 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1385 /* Define macros to initialize and free the failure stack.
1386 Do `return -2' if the alloc fails. */
1388 #ifdef MATCH_MAY_ALLOCATE
1389 # define INIT_FAIL_STACK() \
1390 do { \
1391 fail_stack.stack = (fail_stack_elt_t *) \
1392 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1393 * sizeof (fail_stack_elt_t)); \
1394 \
1395 if (fail_stack.stack == NULL) \
1396 return -2; \
1397 \
1398 fail_stack.size = INIT_FAILURE_ALLOC; \
1399 fail_stack.avail = 0; \
1400 fail_stack.frame = 0; \
1401 } while (0)
1402 #else
1403 # define INIT_FAIL_STACK() \
1404 do { \
1405 fail_stack.avail = 0; \
1406 fail_stack.frame = 0; \
1407 } while (0)
1409 # define RETALLOC_IF(addr, n, t) \
1410 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1411 #endif
1414 /* Double the size of FAIL_STACK, up to a limit
1415 which allows approximately `re_max_failures' items.
1417 Return 1 if succeeds, and 0 if either ran out of memory
1418 allocating space for it or it was already too large.
1420 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1422 /* Factor to increase the failure stack size by
1423 when we increase it.
1424 This used to be 2, but 2 was too wasteful
1425 because the old discarded stacks added up to as much space
1426 were as ultimate, maximum-size stack. */
1427 #define FAIL_STACK_GROWTH_FACTOR 4
1429 #define GROW_FAIL_STACK(fail_stack) \
1430 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1431 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1432 ? 0 \
1433 : ((fail_stack).stack \
1434 = (fail_stack_elt_t *) \
1435 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1436 (fail_stack).size * sizeof (fail_stack_elt_t), \
1437 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1438 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1439 * FAIL_STACK_GROWTH_FACTOR))), \
1440 \
1441 (fail_stack).stack == NULL \
1442 ? 0 \
1443 : ((fail_stack).size \
1444 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1445 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1446 * FAIL_STACK_GROWTH_FACTOR)) \
1447 / sizeof (fail_stack_elt_t)), \
1448 1)))
1451 /* Push a pointer value onto the failure stack.
1452 Assumes the variable `fail_stack'. Probably should only
1453 be called from within `PUSH_FAILURE_POINT'. */
1454 #define PUSH_FAILURE_POINTER(item) \
1455 fail_stack.stack[fail_stack.avail++].pointer = (item)
1457 /* This pushes an integer-valued item onto the failure stack.
1458 Assumes the variable `fail_stack'. Probably should only
1459 be called from within `PUSH_FAILURE_POINT'. */
1460 #define PUSH_FAILURE_INT(item) \
1461 fail_stack.stack[fail_stack.avail++].integer = (item)
1463 /* These POP... operations complement the PUSH... operations.
1464 All assume that `fail_stack' is nonempty. */
1465 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1466 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1468 /* Individual items aside from the registers. */
1469 #define NUM_NONREG_ITEMS 3
1471 /* Used to examine the stack (to detect infinite loops). */
1472 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1473 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1474 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1475 #define TOP_FAILURE_HANDLE() fail_stack.frame
1478 #define ENSURE_FAIL_STACK(space) \
1479 while (REMAINING_AVAIL_SLOTS <= space) { \
1480 if (!GROW_FAIL_STACK (fail_stack)) \
1481 return -2; \
1484 }
1486 /* Push register NUM onto the stack. */
1487 #define PUSH_FAILURE_REG(num) \
1488 do { \
1489 char *destination; \
1490 ENSURE_FAIL_STACK(3); \
1492 num, regstart[num], regend[num]); \
1493 PUSH_FAILURE_POINTER (regstart[num]); \
1494 PUSH_FAILURE_POINTER (regend[num]); \
1495 PUSH_FAILURE_INT (num); \
1496 } while (0)
1498 /* Change the counter's value to VAL, but make sure that it will
1499 be reset when backtracking. */
1500 #define PUSH_NUMBER(ptr,val) \
1501 do { \
1502 char *destination; \
1503 int c; \
1504 ENSURE_FAIL_STACK(3); \
1505 EXTRACT_NUMBER (c, ptr); \
1507 PUSH_FAILURE_INT (c); \
1508 PUSH_FAILURE_POINTER (ptr); \
1509 PUSH_FAILURE_INT (-1); \
1510 STORE_NUMBER (ptr, val); \
1511 } while (0)
1513 /* Pop a saved register off the stack. */
1514 #define POP_FAILURE_REG_OR_COUNT() \
1515 do { \
1516 long pfreg = POP_FAILURE_INT (); \
1517 if (pfreg == -1) \
1518 { \
1521 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1522 pfreg = POP_FAILURE_INT (); \
1523 STORE_NUMBER (ptr, pfreg); \
1525 } \
1526 else \
1527 { \
1528 regend[pfreg] = POP_FAILURE_POINTER (); \
1529 regstart[pfreg] = POP_FAILURE_POINTER (); \
1531 pfreg, regstart[pfreg], regend[pfreg]); \
1532 } \
1533 } while (0)
1535 /* Check that we are not stuck in an infinite loop. */
1536 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1537 do { \
1538 ssize_t failure = TOP_FAILURE_HANDLE (); \
1540 while (failure > 0 \
1541 && (FAILURE_STR (failure) == string_place \
1542 || FAILURE_STR (failure) == NULL)) \
1543 { \
1544 assert (FAILURE_PAT (failure) >= bufp->buffer \
1545 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1546 if (FAILURE_PAT (failure) == pat_cur) \
1547 { \
1548 cycle = 1; \
1549 break; \
1550 } \
1552 failure = NEXT_FAILURE_HANDLE(failure); \
1553 } \
1555 } while (0)
1557 /* Push the information about the state we will need
1558 if we ever fail back to it.
1560 Requires variables fail_stack, regstart, regend and
1561 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1562 declared.
1564 Does `return FAILURE_CODE' if runs out of memory. */
1566 #define PUSH_FAILURE_POINT(pattern, string_place) \
1567 do { \
1568 char *destination; \
1569 /* Must be int, so when we don't save any registers, the arithmetic \
1571 \
1572 DEBUG_STATEMENT (nfailure_points_pushed++); \
1576 \
1577 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1578 \
1580 \
1582 PUSH_FAILURE_INT (fail_stack.frame); \
1583 \
1585 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1587 PUSH_FAILURE_POINTER (string_place); \
1588 \
1590 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1591 PUSH_FAILURE_POINTER (pattern); \
1592 \
1594 fail_stack.frame = fail_stack.avail; \
1595 } while (0)
1597 /* Estimate the size of data pushed by a typical failure stack entry.
1598 An estimate is all we need, because all we use this for
1599 is to choose a limit for how big to make the failure stack. */
1600 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1601 #define TYPICAL_FAILURE_SIZE 20
1603 /* How many items can still be added to the stack without overflowing it. */
1604 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1607 /* Pops what PUSH_FAIL_STACK pushes.
1609 We restore into the parameters, all of which should be lvalues:
1610 STR -- the saved data position.
1611 PAT -- the saved pattern position.
1612 REGSTART, REGEND -- arrays of string positions.
1614 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1615 `pend', `string1', `size1', `string2', and `size2'. */
1617 #define POP_FAILURE_POINT(str, pat) \
1618 do { \
1619 assert (!FAIL_STACK_EMPTY ()); \
1620 \
1625 \
1627 while (fail_stack.frame < fail_stack.avail) \
1628 POP_FAILURE_REG_OR_COUNT (); \
1629 \
1630 pat = POP_FAILURE_POINTER (); \
1632 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1633 \
1634 /* If the saved string location is NULL, it came from an \
1635 on_failure_keep_string_jump opcode, and we want to throw away the \
1637 str = POP_FAILURE_POINTER (); \
1639 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1641 \
1642 fail_stack.frame = POP_FAILURE_INT (); \
1644 \
1645 assert (fail_stack.avail >= 0); \
1646 assert (fail_stack.frame <= fail_stack.avail); \
1647 \
1648 DEBUG_STATEMENT (nfailure_points_popped++); \
1652 \f
1653 /* Registers are set to a sentinel when they haven't yet matched. */
1654 #define REG_UNSET(e) ((e) == NULL)
1655 \f
1656 /* Subroutine declarations and macros for regex_compile. */
1659 reg_syntax_t syntax,
1670 reg_syntax_t syntax));
1673 reg_syntax_t syntax));
1678 /* Fetch the next character in the uncompiled pattern, with no
1679 translation. */
1680 #define PATFETCH(c) \
1681 do { \
1682 int len; \
1683 if (p == pend) return REG_EEND; \
1684 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1685 p += len; \
1686 } while (0)
1689 /* If `translate' is non-null, return translate[D], else just D. We
1690 cast the subscript to translate because some data is declared as
1691 `char *', to avoid warnings when a string constant is passed. But
1692 when we use a character as a subscript we must make it unsigned. */
1693 #ifndef TRANSLATE
1694 # define TRANSLATE(d) \
1695 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1696 #endif
1699 /* Macros for outputting the compiled pattern into `buffer'. */
1701 /* If the buffer isn't allocated when it comes in, use this. */
1702 #define INIT_BUF_SIZE 32
1704 /* Make sure we have at least N more bytes of space in buffer. */
1705 #define GET_BUFFER_SPACE(n) \
1706 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1707 EXTEND_BUFFER ()
1709 /* Make sure we have one more byte of buffer space and then add C to it. */
1710 #define BUF_PUSH(c) \
1711 do { \
1712 GET_BUFFER_SPACE (1); \
1713 *b++ = (unsigned char) (c); \
1714 } while (0)
1717 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1718 #define BUF_PUSH_2(c1, c2) \
1719 do { \
1720 GET_BUFFER_SPACE (2); \
1721 *b++ = (unsigned char) (c1); \
1722 *b++ = (unsigned char) (c2); \
1723 } while (0)
1726 /* Store a jump with opcode OP at LOC to location TO. We store a
1727 relative address offset by the three bytes the jump itself occupies. */
1728 #define STORE_JUMP(op, loc, to) \
1729 store_op1 (op, loc, (to) - (loc) - 3)
1731 /* Likewise, for a two-argument jump. */
1732 #define STORE_JUMP2(op, loc, to, arg) \
1733 store_op2 (op, loc, (to) - (loc) - 3, arg)
1735 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1736 #define INSERT_JUMP(op, loc, to) \
1737 insert_op1 (op, loc, (to) - (loc) - 3, b)
1739 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1740 #define INSERT_JUMP2(op, loc, to, arg) \
1741 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1744 /* This is not an arbitrary limit: the arguments which represent offsets
1745 into the pattern are two bytes long. So if 2^15 bytes turns out to
1746 be too small, many things would have to change. */
1747 # define MAX_BUF_SIZE (1L << 15)
1750 /* Any other compiler which, like MSC, has allocation limit below 2^16
1751 bytes will have to use approach similar to what was done below for
1752 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1753 reallocating to 0 bytes. Such thing is not going to work too well.
1754 You have been warned!! */
1755 #if defined _MSC_VER && !defined WIN32
1756 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1757 # define MAX_BUF_SIZE 65500L
1758 #else
1759 # define MAX_BUF_SIZE (1L << 16)
1760 #endif
1763 /* Extend the buffer by twice its current size via realloc and
1764 reset the pointers that pointed into the old block to point to the
1765 correct places in the new one. If extending the buffer results in it
1766 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1767 #if __BOUNDED_POINTERS__
1768 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1769 # define MOVE_BUFFER_POINTER(P) \
1770 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1771 SET_HIGH_BOUND (P), \
1772 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1773 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1774 else \
1775 { \
1776 SET_HIGH_BOUND (b); \
1777 SET_HIGH_BOUND (begalt); \
1778 if (fixup_alt_jump) \
1779 SET_HIGH_BOUND (fixup_alt_jump); \
1780 if (laststart) \
1781 SET_HIGH_BOUND (laststart); \
1782 if (pending_exact) \
1783 SET_HIGH_BOUND (pending_exact); \
1784 }
1785 #else
1786 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1787 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1788 #endif
1789 #define EXTEND_BUFFER() \
1790 do { \
1791 unsigned char *old_buffer = bufp->buffer; \
1792 if (bufp->allocated == MAX_BUF_SIZE) \
1793 return REG_ESIZE; \
1794 bufp->allocated <<= 1; \
1795 if (bufp->allocated > MAX_BUF_SIZE) \
1796 bufp->allocated = MAX_BUF_SIZE; \
1797 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1798 if (bufp->buffer == NULL) \
1799 return REG_ESPACE; \
1801 if (old_buffer != bufp->buffer) \
1802 { \
1803 unsigned char *new_buffer = bufp->buffer; \
1804 MOVE_BUFFER_POINTER (b); \
1805 MOVE_BUFFER_POINTER (begalt); \
1806 if (fixup_alt_jump) \
1807 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1808 if (laststart) \
1809 MOVE_BUFFER_POINTER (laststart); \
1810 if (pending_exact) \
1811 MOVE_BUFFER_POINTER (pending_exact); \
1812 } \
1813 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1814 } while (0)
1817 /* Since we have one byte reserved for the register number argument to
1818 {start,stop}_memory, the maximum number of groups we can report
1819 things about is what fits in that byte. */
1820 #define MAX_REGNUM 255
1822 /* But patterns can have more than `MAX_REGNUM' registers. We just
1823 ignore the excess. */
1827 /* Macros for the compile stack. */
1829 /* Since offsets can go either forwards or backwards, this type needs to
1830 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1831 /* int may be not enough when sizeof(int) == 2. */
1835 {
1836 pattern_offset_t begalt_offset;
1837 pattern_offset_t fixup_alt_jump;
1838 pattern_offset_t laststart_offset;
1839 regnum_t regnum;
1844 {
1851 #define INIT_COMPILE_STACK_SIZE 32
1853 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1854 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1856 /* The next available element. */
1857 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1859 /* Explicit quit checking is only used on NTemacs and whenever we
1860 use polling to process input events. */
1861 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1863 # define IMMEDIATE_QUIT_CHECK \
1864 do { \
1865 if (immediate_quit) QUIT; \
1866 } while (0)
1867 #else
1868 # define IMMEDIATE_QUIT_CHECK ((void)0)
1869 #endif
1870 \f
1871 /* Structure to manage work area for range table. */
1872 struct range_table_work_area
1873 {
1878 };
1880 /* Make sure that WORK_AREA can hold more N multibyte characters.
1881 This is used only in set_image_of_range and set_image_of_range_1.
1882 It expects WORK_AREA to be a pointer.
1883 If it can't get the space, it returns from the surrounding function. */
1885 #define EXTEND_RANGE_TABLE(work_area, n) \
1886 do { \
1887 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1888 { \
1889 extend_range_table_work_area (&work_area); \
1890 if ((work_area).table == 0) \
1891 return (REG_ESPACE); \
1892 } \
1893 } while (0)
1895 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1896 (work_area).bits |= (bit)
1898 /* Bits used to implement the multibyte-part of the various character classes
1899 such as [:alnum:] in a charset's range table. */
1900 #define BIT_WORD 0x1
1901 #define BIT_LOWER 0x2
1902 #define BIT_PUNCT 0x4
1903 #define BIT_SPACE 0x8
1904 #define BIT_UPPER 0x10
1905 #define BIT_MULTIBYTE 0x20
1907 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1908 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1909 do { \
1910 EXTEND_RANGE_TABLE ((work_area), 2); \
1911 (work_area).table[(work_area).used++] = (range_start); \
1912 (work_area).table[(work_area).used++] = (range_end); \
1913 } while (0)
1915 /* Free allocated memory for WORK_AREA. */
1916 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1917 do { \
1918 if ((work_area).table) \
1919 free ((work_area).table); \
1920 } while (0)
1922 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1923 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1924 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1925 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1926 \f
1928 /* Set the bit for character C in a list. */
1929 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1932 #ifdef emacs
1934 /* Store characters in the range FROM to TO in the bitmap at B (for
1935 ASCII and unibyte characters) and WORK_AREA (for multibyte
1936 characters) while translating them and paying attention to the
1937 continuity of translated characters.
1939 Implementation note: It is better to implement these fairly big
1940 macros by a function, but it's not that easy because macros called
1941 in this macro assume various local variables already declared. */
1943 /* Both FROM and TO are ASCII characters. */
1945 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1946 do { \
1947 int C0, C1; \
1948 \
1949 for (C0 = (FROM); C0 <= (TO); C0++) \
1950 { \
1951 C1 = TRANSLATE (C0); \
1952 if (! ASCII_CHAR_P (C1)) \
1953 { \
1954 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1955 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1956 C1 = C0; \
1957 } \
1958 SET_LIST_BIT (C1); \
1959 } \
1960 } while (0)
1963 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1965 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1966 do { \
1967 int C0, C1, C2, I; \
1968 int USED = RANGE_TABLE_WORK_USED (work_area); \
1969 \
1970 for (C0 = (FROM); C0 <= (TO); C0++) \
1971 { \
1972 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1973 if (CHAR_BYTE8_P (C1)) \
1974 SET_LIST_BIT (C0); \
1975 else \
1976 { \
1977 C2 = TRANSLATE (C1); \
1978 if (C2 == C1 \
1979 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1980 C1 = C0; \
1981 SET_LIST_BIT (C1); \
1982 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1983 { \
1984 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1985 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1986 \
1987 if (C2 >= from - 1 && C2 <= to + 1) \
1988 { \
1989 if (C2 == from - 1) \
1990 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1991 else if (C2 == to + 1) \
1992 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1993 break; \
1994 } \
1995 } \
1996 if (I < USED) \
1997 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1998 } \
1999 } \
2000 } while (0)
2003 /* Both FROM and TO are multibyte characters. */
2005 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2006 do { \
2007 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2008 \
2009 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2010 for (C0 = (FROM); C0 <= (TO); C0++) \
2011 { \
2012 C1 = TRANSLATE (C0); \
2013 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2014 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2015 SET_LIST_BIT (C2); \
2016 if (C1 >= (FROM) && C1 <= (TO)) \
2017 continue; \
2018 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2019 { \
2020 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2021 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2022 \
2023 if (C1 >= from - 1 && C1 <= to + 1) \
2024 { \
2025 if (C1 == from - 1) \
2026 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2027 else if (C1 == to + 1) \
2028 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2029 break; \
2030 } \
2031 } \
2032 if (I < USED) \
2033 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2034 } \
2035 } while (0)
2039 /* Get the next unsigned number in the uncompiled pattern. */
2040 #define GET_UNSIGNED_NUMBER(num) \
2041 do { \
2042 if (p == pend) \
2043 FREE_STACK_RETURN (REG_EBRACE); \
2044 else \
2045 { \
2046 PATFETCH (c); \
2048 { \
2049 int prev; \
2050 if (num < 0) \
2051 num = 0; \
2052 prev = num; \
2054 if (num / 10 != prev) \
2055 FREE_STACK_RETURN (REG_BADBR); \
2056 if (p == pend) \
2057 FREE_STACK_RETURN (REG_EBRACE); \
2058 PATFETCH (c); \
2059 } \
2060 } \
2061 } while (0)
2062 \f
2063 #if ! WIDE_CHAR_SUPPORT
2065 /* Map a string to the char class it names (if any). */
2066 re_wctype_t
2068 {
2088 }
2090 /* True if CH is in the char class CC. */
2091 boolean
2093 {
2095 {
2116 }
2117 }
2119 /* Return a bit-pattern to use in the range-table bits to match multibyte
2120 chars of class CC. */
2121 static int
2123 {
2125 {
2137 }
2138 }
2139 #endif
2140 \f
2141 /* Filling in the work area of a range. */
2143 /* Actually extend the space in WORK_AREA. */
2145 static void
2147 {
2150 work_area->table
2152 else
2153 work_area->table
2155 }
2157 #if 0
2158 #ifdef emacs
2160 /* Carefully find the ranges of codes that are equivalent
2161 under case conversion to the range start..end when passed through
2162 TRANSLATE. Handle the case where non-letters can come in between
2163 two upper-case letters (which happens in Latin-1).
2164 Also handle the case of groups of more than 2 case-equivalent chars.
2166 The basic method is to look at consecutive characters and see
2167 if they can form a run that can be handled as one.
2169 Returns -1 if successful, REG_ESPACE if ran out of space. */
2171 static int
2174 RE_TRANSLATE_TYPE translate)
2175 {
2176 /* `one_case' indicates a character, or a run of characters,
2177 each of which is an isolate (no case-equivalents).
2178 This includes all ASCII non-letters.
2180 `two_case' indicates a character, or a run of characters,
2181 each of which has two case-equivalent forms.
2182 This includes all ASCII letters.
2184 `strange' indicates a character that has more than one
2185 case-equivalent. */
2189 /* Describe the run that is in progress,
2190 which the next character can try to extend.
2191 If run_type is strange, that means there really is no run.
2192 If run_type is one_case, then run_start...run_end is the run.
2193 If run_type is two_case, then the run is run_start...run_end,
2194 and the case-equivalents end at run_eqv_end. */
2199 Lisp_Object eqv_table;
2202 {
2207 }
2212 {
2217 /* Classify this character */
2222 else
2227 else
2230 /* Can this character extend the run in progress? */
2235 {
2236 /* No, end the run.
2237 Record each of its equivalent ranges. */
2239 {
2243 }
2245 {
2253 }
2255 }
2258 {
2259 /* For a strange character, add each of its equivalents, one
2260 by one. Don't start a range. */
2261 do
2262 {
2267 }
2269 }
2271 /* Add this char to the run, or start a new run. */
2273 {
2274 /* Initialize a new range. */
2279 }
2280 else
2281 {
2282 /* Extend a running range. */
2285 }
2286 }
2288 /* If a run is still in progress at the end, finish it now
2289 by recording its equivalent ranges. */
2291 {
2295 }
2297 {
2305 }
2308 }
2312 /* Record the image of the range start..end when passed through
2313 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2314 and is not even necessarily contiguous.
2315 Normally we approximate it with the smallest contiguous range that contains
2316 all the chars we need. However, for Latin-1 we go to extra effort
2317 to do a better job.
2319 This function is not called for ASCII ranges.
2321 Returns -1 if successful, REG_ESPACE if ran out of space. */
2323 static int
2326 RE_TRANSLATE_TYPE translate)
2327 {
2330 #ifdef emacs
2331 /* For Latin-1 ranges, use set_image_of_range_1
2332 to get proper handling of ranges that include letters and nonletters.
2333 For a range that includes the whole of Latin-1, this is not necessary.
2334 For other character sets, we don't bother to get this right. */
2337 {
2350 }
2351 #endif
2360 {
2364 {
2367 {
2370 else
2371 {
2374 }
2375 }
2376 }
2379 {
2383 }
2384 }
2387 }
2389 \f
2390 #ifndef MATCH_MAY_ALLOCATE
2392 /* If we cannot allocate large objects within re_match_2_internal,
2393 we make the fail stack and register vectors global.
2394 The fail stack, we grow to the maximum size when a regexp
2395 is compiled.
2396 The register vectors, we adjust in size each time we
2397 compile a regexp, according to the number of registers it needs. */
2401 /* Size with which the following vectors are currently allocated.
2402 That is so we can make them bigger as needed,
2403 but never make them smaller. */
2409 /* Make the register vectors big enough for NUM_REGS registers,
2410 but don't make them smaller. */
2412 static
2414 {
2416 {
2423 }
2424 }
2427 \f
2429 compile_stack,
2430 regnum_t regnum));
2432 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2433 Returns one of error codes defined in `regex.h', or zero for success.
2435 Assumes the `allocated' (and perhaps `buffer') and `translate'
2436 fields are set in BUFP on entry.
2438 If it succeeds, results are put in BUFP (if it returns an error, the
2439 contents of BUFP are undefined):
2440 `buffer' is the compiled pattern;
2441 `syntax' is set to SYNTAX;
2442 `used' is set to the length of the compiled pattern;
2443 `fastmap_accurate' is zero;
2444 `re_nsub' is the number of subexpressions in PATTERN;
2445 `not_bol' and `not_eol' are zero;
2447 The `fastmap' field is neither examined nor set. */
2449 /* Insert the `jump' from the end of last alternative to "here".
2450 The space for the jump has already been allocated. */
2451 #define FIXUP_ALT_JUMP() \
2452 do { \
2453 if (fixup_alt_jump) \
2454 STORE_JUMP (jump, fixup_alt_jump, b); \
2455 } while (0)
2458 /* Return, freeing storage we allocated. */
2459 #define FREE_STACK_RETURN(value) \
2460 do { \
2461 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2462 free (compile_stack.stack); \
2463 return value; \
2464 } while (0)
2466 static reg_errcode_t
2467 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2468 {
2469 /* We fetch characters from PATTERN here. */
2472 /* Points to the end of the buffer, where we should append. */
2475 /* Keeps track of unclosed groups. */
2476 compile_stack_type compile_stack;
2478 /* Points to the current (ending) position in the pattern. */
2479 #ifdef AIX
2480 /* `const' makes AIX compiler fail. */
2482 #else
2484 #endif
2487 /* How to translate the characters in the pattern. */
2490 /* Address of the count-byte of the most recently inserted `exactn'
2491 command. This makes it possible to tell if a new exact-match
2492 character can be added to that command or if the character requires
2493 a new `exactn' command. */
2496 /* Address of start of the most recently finished expression.
2497 This tells, e.g., postfix * where to find the start of its
2498 operand. Reset at the beginning of groups and alternatives. */
2501 /* Address of beginning of regexp, or inside of last group. */
2504 /* Place in the uncompiled pattern (i.e., the {) to
2505 which to go back if the interval is invalid. */
2508 /* Address of the place where a forward jump should go to the end of
2509 the containing expression. Each alternative of an `or' -- except the
2510 last -- ends with a forward jump of this sort. */
2513 /* Work area for range table of charset. */
2516 /* If the object matched can contain multibyte characters. */
2519 /* Nonzero if we have pushed down into a subpattern. */
2522 /* These hold the values of p, pattern, and pend from the main
2523 pattern when we have pushed into a subpattern. */
2528 #ifdef DEBUG
2529 debug++;
2532 {
2538 }
2541 /* Initialize the compile stack. */
2552 /* Initialize the pattern buffer. */
2558 /* Set `used' to zero, so that if we return an error, the pattern
2559 printer (for debugging) will think there's no pattern. We reset it
2560 at the end. */
2563 /* Always count groups, whether or not bufp->no_sub is set. */
2566 #if !defined emacs && !defined SYNTAX_TABLE
2567 /* Initialize the syntax table. */
2569 #endif
2572 {
2575 enough space. This loses if buffer's address is bogus, but
2576 that is the user's responsibility. */
2578 }
2579 else
2582 }
2586 }
2590 /* Loop through the uncompiled pattern until we're at the end. */
2592 {
2594 {
2595 /* If this is the end of an included regexp,
2596 pop back to the main regexp and try again. */
2598 {
2604 }
2605 /* If this is the end of the main regexp, we are done. */
2607 }
2612 {
2614 {
2617 /* If there's no special whitespace regexp, treat
2618 spaces normally. And don't try to do this recursively. */
2622 /* Peek past following spaces. */
2624 {
2627 p1++;
2628 }
2629 /* If the spaces are followed by a repetition op,
2630 treat them normally. */
2636 /* Replace the spaces with the whitespace regexp. */
2644 }
2647 {
2650 /* If context independent, it's an operator. */
2652 /* Otherwise, depends on what's come before. */
2655 else
2657 }
2662 {
2664 p == pend
2665 /* If context independent, it's an operator. */
2667 /* Otherwise, depends on what's next. */
2670 else
2672 }
2681 handle_plus:
2683 /* If there is no previous pattern... */
2685 {
2690 }
2692 {
2693 /* 1 means zero (many) matches is allowed. */
2697 /* If there is a sequence of repetition chars, collapse it
2698 down to just one (the right one). We can't combine
2699 interval operators with these because of, e.g., `a{2}*',
2700 which should only match an even number of `a's. */
2703 {
2707 else
2708 {
2711 }
2718 ;
2720 {
2725 else
2727 }
2728 else
2730 /* If we get here, we found another repeat character. */
2732 }
2734 /* Star, etc. applied to an empty pattern is equivalent
2735 to an empty pattern. */
2739 /* Now we know whether or not zero matches is allowed
2740 and also whether or not two or more matches is allowed. */
2742 {
2744 {
2747 re_opcode_t ofj =
2748 /* Check if the loop can match the empty string. */
2755 on_failure_keep_string_jump, we turn simple P+
2756 into PP* if P is simple. */
2764 }
2768 /* A + loop. */
2770 else
2771 /* Simple * loops can use on_failure_keep_string_jump
2772 depending on what follows. But since we don't know
2773 that yet, we leave the decision up to
2774 on_failure_jump_smart. */
2780 }
2781 else
2782 {
2783 /* A simple ? pattern. */
2788 }
2789 }
2795 {
2798 /* The non-greedy multiple match looks like
2799 a repeat..until: we only need a conditional jump
2800 at the end of the loop. */
2806 {
2807 /* The repeat...until naturally matches one or more.
2808 To also match zero times, we need to first jump to
2809 the end of the loop (its conditional jump). */
2812 }
2813 }
2814 else
2815 {
2816 /* non-greedy a?? */
2821 }
2822 }
2823 }
2835 {
2842 /* Ensure that we have enough space to push a charset: the
2843 opcode, the length count, and the bitset; 34 bytes in all. */
2848 /* We test `*p == '^' twice, instead of using an if
2849 statement, so we only need one BUF_PUSH. */
2852 p++;
2854 /* Remember the first position in the bracket expression. */
2857 /* Push the number of bytes in the bitmap. */
2860 /* Clear the whole map. */
2863 /* charset_not matches newline according to a syntax bit. */
2868 /* Read in characters and ranges, setting map bits. */
2870 {
2873 re_wchar_t ch;
2877 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2878 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2879 So the translation is done later in a loop. Example:
2880 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2883 /* \ might escape characters inside [...] and [^...]. */
2885 {
2890 }
2891 else
2892 {
2893 /* Could be the end of the bracket expression. If it's
2894 not (i.e., when the bracket expression is `[]' so
2895 far), the ']' character bit gets set way below. */
2898 }
2900 /* See if we're at the beginning of a possible character
2901 class. */
2905 {
2906 /* Leave room for the null. */
2914 /* If pattern is `[[:'. */
2918 {
2924 else
2925 /* This is in any case an invalid class name. */
2927 }
2930 /* If isn't a word bracketed by `[:' and `:]':
2931 undo the ending character, the letters, and
2932 leave the leading `:' and `[' (but set bits for
2933 them). */
2935 {
2941 /* Throw away the ] at the end of the character
2942 class. */
2947 #ifndef emacs
2950 {
2954 }
2956 /* Most character classes in a multibyte match
2957 just set a flag. Exceptions are is_blank,
2958 is_digit, is_cntrl, and is_xdigit, since
2959 they can only match ASCII characters. We
2960 don't need to handle them for multibyte.
2961 They are distinguished by a negative wctype. */
2963 /* Setup the gl_state object to its buffer-defined
2964 value. This hardcodes the buffer-global
2965 syntax-table for ASCII chars, while the other chars
2966 will obey syntax-table properties. It's not ideal,
2967 but it's the way it's been done until now. */
2971 {
2975 {
2984 }
2985 }
2986 SET_RANGE_TABLE_WORK_AREA_BIT
2989 /* In most cases the matching rule for char classes
2990 only uses the syntax table for multibyte chars,
2991 so that the content of the syntax-table it is not
2992 hardcoded in the range_table. SPACE and WORD are
2993 the two exceptions. */
2997 /* Repeat the loop. */
2999 }
3000 else
3001 {
3002 /* Go back to right after the "[:". */
3006 /* Because the `:' may starts the range, we
3007 can't simply set bit and repeat the loop.
3008 Instead, just set it to C and handle below. */
3010 }
3011 }
3014 {
3016 /* Discard the `-'. */
3019 /* Fetch the character which ends the range. */
3021 #ifdef emacs
3024 /* Treat the range from a multibyte character to
3025 raw-byte character as empty. */
3028 }
3029 else
3030 /* Range from C to C. */
3034 {
3037 /* Else, repeat the loop. */
3038 }
3039 else
3040 {
3041 #ifndef emacs
3042 /* Set the range into bitmap */
3044 {
3048 }
3051 {
3057 }
3059 {
3061 {
3066 }
3068 {
3070 }
3071 else
3072 {
3074 }
3075 }
3077 }
3078 }
3080 /* Discard any (non)matching list bytes that are all 0 at the
3081 end of the map. Decrease the map-length byte too. */
3086 /* Build real range table from work area. */
3089 {
3093 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3094 bytes for flags, two for COUNT, and three bytes for
3095 each character. */
3098 /* Indicate the existence of range table. */
3101 /* Store the character class flag bits into the range table.
3102 If not in emacs, these flag bits are always 0. */
3108 STORE_CHARACTER_AND_INCR
3110 }
3111 }
3118 else
3125 else
3132 else
3139 else
3146 else
3153 /* Do not translate the character after the \, so that we can
3154 distinguish, e.g., \B from \b, even if we normally would
3155 translate, e.g., B to b. */
3159 {
3164 handle_open:
3165 {
3169 {
3170 /* Look for a special (?...) construct */
3172 {
3175 {
3178 {
3181 /* An explicitly specified regnum must start
3182 with non-0. */
3189 /* Only (?:...) is supported right now. */
3191 }
3192 }
3193 }
3194 }
3204 /* Ideally, we'd want to check that the specified
3205 group can't have matched (i.e. all subgroups
3206 using the same regnum are in other branches of
3207 OR patterns), but we don't currently keep track
3208 of enough info to do that easily. */
3211 }
3212 else
3213 /* It's really shy. */
3217 {
3219 compile_stack_elt_t);
3223 }
3225 /* These are the values to restore when we hit end of this
3226 group. They are all relative offsets, so that if the
3227 whole pattern moves because of realloc, they will still
3228 be valid. */
3230 COMPILE_STACK_TOP.fixup_alt_jump
3235 /* Do not push a start_memory for groups beyond the last one
3236 we can represent in the compiled pattern. */
3245 /* If we've reached MAX_REGNUM groups, then this open
3246 won't actually generate any code, so we'll have to
3247 clear pending_exact explicitly. */
3250 }
3256 {
3259 else
3261 }
3263 handle_close:
3266 /* See similar code for backslashed left paren above. */
3268 {
3271 else
3273 }
3275 /* Since we just checked for an empty stack above, this
3276 ``can't happen''. */
3278 {
3279 /* We don't just want to restore into `regnum', because
3280 later groups should continue to be numbered higher,
3281 as in `(ab)c(de)' -- the second group is #2. */
3282 regnum_t regnum;
3286 fixup_alt_jump
3292 /* If we've reached MAX_REGNUM groups, then this open
3293 won't actually generate any code, so we'll have to
3294 clear pending_exact explicitly. */
3297 /* We're at the end of the group, so now we know how many
3298 groups were inside this one. */
3301 }
3308 handle_alt:
3312 /* Insert before the previous alternative a jump which
3313 jumps to this alternative if the former fails. */
3319 /* The alternative before this one has a jump after it
3320 which gets executed if it gets matched. Adjust that
3321 jump so it will jump to this alternative's analogous
3322 jump (put in below, which in turn will jump to the next
3323 (if any) alternative's such jump, etc.). The last such
3324 jump jumps to the correct final destination. A picture:
3325 _____ _____
3326 | | | |
3327 | v | v
3328 a | b | c
3330 If we are at `b', then fixup_alt_jump right now points to a
3331 three-byte space after `a'. We'll put in the jump, set
3332 fixup_alt_jump to right after `b', and leave behind three
3333 bytes which we'll fill in when we get to after `c'. */
3337 /* Mark and leave space for a jump after this alternative,
3338 to be filled in later either by next alternative or
3339 when know we're at the end of a series of alternatives. */
3350 /* If \{ is a literal. */
3352 /* If we're at `\{' and it's not the open-interval
3353 operator. */
3357 handle_interval:
3358 {
3359 /* If got here, then the syntax allows intervals. */
3361 /* At least (most) this many matches must be made. */
3370 else
3371 /* Interval such as `{1}' => match exactly once. */
3379 {
3385 }
3390 /* We just parsed a valid interval. */
3392 /* If it's invalid to have no preceding re. */
3394 {
3399 else
3401 }
3404 /* If the upper bound is zero, just drop the sub pattern
3405 altogether. */
3408 /* Just match it once: nothing to do here. */
3409 ;
3411 /* Otherwise, we have a nontrivial interval. When
3412 we're all done, the pattern will look like:
3413 set_number_at <jump count> <upper bound>
3414 set_number_at <succeed_n count> <lower bound>
3415 succeed_n <after jump addr> <succeed_n count>
3416 <body of loop>
3417 jump_n <succeed_n addr> <jump count>
3418 (The upper bound and `jump_n' are omitted if
3419 `upper_bound' is 1, though.) */
3420 else
3422 more at the end of the loop. */
3430 {
3431 /* A succeed_n that starts with 0 is really a
3432 a simple on_failure_jump_loop. */
3436 }
3437 else
3438 {
3439 /* Initialize lower bound of the `succeed_n', even
3440 though it will be set during matching by its
3441 attendant `set_number_at' (inserted next),
3442 because `re_compile_fastmap' needs to know.
3443 Jump to the `jump_n' we might insert below. */
3446 lower_bound);
3449 /* Code to initialize the lower bound. Insert
3450 before the `succeed_n'. The `5' is the last two
3451 bytes of this `set_number_at', plus 3 bytes of
3452 the following `succeed_n'. */
3456 }
3459 {
3460 /* A negative upper bound stands for infinity,
3461 in which case it degenerates to a plain jump. */
3464 }
3467 append a backward jump to the `succeed_n'
3468 that starts this interval.
3470 When we've reached this during matching,
3471 we'll have matched the interval once, so
3472 jump back only `upper_bound - 1' times. */
3477 /* The location we want to set is the second
3478 parameter of the `jump_n'; that is `b-2' as
3479 an absolute address. `laststart' will be
3480 the `set_number_at' we're about to insert;
3481 `laststart+3' the number to set, the source
3482 for the relative address. But we are
3483 inserting into the middle of the pattern --
3484 so everything is getting moved up by 5.
3485 Conclusion: (b - 2) - (laststart + 3) + 5,
3486 i.e., b - laststart.
3488 We insert this at the beginning of the loop
3489 so that if we fail during matching, we'll
3490 reinitialize the bounds. */
3494 }
3495 }
3498 }
3501 unfetch_interval:
3502 /* If an invalid interval, match the characters as literals. */
3507 /* normal_char and normal_backslash need `c'. */
3511 {
3514 }
3515 else
3518 #ifdef emacs
3519 /* There is no way to specify the before_dot and after_dot
3520 operators. rms says this is ok. --karl */
3588 else
3618 {
3619 regnum_t reg;
3627 /* Can't back reference to a subexp before its end. */
3633 }
3641 else
3645 normal_backslash:
3646 /* You might think it would be useful for \ to mean
3647 not to translate; but if we don't translate it
3648 it will never match anything. */
3650 }
3655 /* Expects the character in `c'. */
3656 normal_char:
3657 /* If no exactn currently being built. */
3660 /* If last exactn not at current position. */
3663 /* We have only one byte following the exactn for the count. */
3666 /* If followed by a repetition operator. */
3675 {
3676 /* Start building a new exactn. */
3682 }
3685 {
3689 {
3693 }
3694 else
3695 {
3698 {
3703 }
3706 }
3708 }
3715 /* Through the pattern now. */
3722 /* If we don't want backtracking, force success
3723 the first time we reach the end of the compiled pattern. */
3727 /* We have succeeded; set the length of the buffer. */
3730 #ifdef DEBUG
3732 {
3736 }
3737 debug--;
3740 #ifndef MATCH_MAY_ALLOCATE
3741 /* Initialize the failure stack to the largest possible stack. This
3742 isn't necessary unless we're trying to avoid calling alloca in
3743 the search and match routines. */
3744 {
3748 {
3752 fail_stack.stack
3755 else
3756 fail_stack.stack
3760 }
3763 }
3768 \f
3769 /* Subroutines for `regex_compile'. */
3771 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3773 static void
3775 {
3778 }
3781 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3783 static void
3785 {
3789 }
3792 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3793 for OP followed by two-byte integer parameter ARG. */
3795 static void
3797 {
3805 }
3808 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3810 static void
3812 {
3820 }
3823 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3824 after an alternative or a begin-subexpression. We assume there is at
3825 least one character before the ^. */
3827 static boolean
3829 {
3833 return
3834 /* After a subexpression? */
3836 /* After an alternative? */
3838 /* After a shy subexpression? */
3843 }
3846 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3847 at least one character after the $, i.e., `P < PEND'. */
3849 static boolean
3851 {
3856 return
3857 /* Before a subexpression? */
3860 /* Before an alternative? */
3863 }
3866 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3867 false if it's not. */
3869 static boolean
3871 {
3872 ssize_t this_element;
3876 this_element--)
3881 }
3882 \f
3883 /* analyse_first.
3884 If fastmap is non-NULL, go through the pattern and fill fastmap
3885 with all the possible leading chars. If fastmap is NULL, don't
3886 bother filling it up (obviously) and only return whether the
3887 pattern could potentially match the empty string.
3889 Return 1 if p..pend might match the empty string.
3890 Return 0 if p..pend matches at least one char.
3891 Return -1 if fastmap was not updated accurately. */
3893 static int
3895 {
3899 /* If all elements for base leading-codes in fastmap is set, this
3900 flag is set true. */
3905 /* The loop below works as follows:
3906 - It has a working-list kept in the PATTERN_STACK and which basically
3907 starts by only containing a pointer to the first operation.
3908 - If the opcode we're looking at is a match against some set of
3909 chars, then we add those chars to the fastmap and go on to the
3910 next work element from the worklist (done via `break').
3911 - If the opcode is a control operator on the other hand, we either
3912 ignore it (if it's meaningless at this point, such as `start_memory')
3913 or execute it (if it's a jump). If the jump has several destinations
3914 (i.e. `on_failure_jump'), then we push the other destination onto the
3915 worklist.
3916 We guarantee termination by ignoring backward jumps (more or less),
3917 so that `p' is monotonically increasing. More to the point, we
3918 never set `p' (or push) anything `<= p1'. */
3921 {
3922 /* `p1' is used as a marker of how far back a `on_failure_jump'
3923 can go without being ignored. It is normally equal to `p'
3924 (which prevents any backward `on_failure_jump') except right
3925 after a plain `jump', to allow patterns such as:
3926 0: jump 10
3927 3..9: <body>
3928 10: on_failure_jump 3
3929 as used for the *? operator. */
3933 {
3938 /* If the first character has to match a backreference, that means
3939 that the group was empty (since it already matched). Since this
3940 is the only case that interests us here, we can assume that the
3941 backreference must match the empty string. */
3942 p++;
3946 /* Following are the cases which match a character. These end
3947 with `break'. */
3951 {
3952 /* If multibyte is nonzero, the first byte of each
3953 character is an ASCII or a leading code. Otherwise,
3954 each byte is a character. Thus, this works in both
3955 cases. */
3958 {
3959 /* For the case of matching this unibyte regex
3960 against multibyte, we must set a leading code of
3961 the corresponding multibyte character. */
3965 }
3966 }
3971 /* We could put all the chars except for \n (and maybe \0)
3972 but we don't bother since it is generally not worth it. */
3979 {
3980 /* Chars beyond end of bitmap are possible matches. */
3984 }
3986 /* Fallthrough */
3995 #ifdef emacs
3997 which doesn't match the specified set of characters. */
3998 not
3999 ||
4000 /* If we can match a character class, we can match any
4001 multibyte characters. */
4005 {
4007 {
4012 }
4013 }
4017 {
4018 /* Set fastmap[I] to 1 where I is a leading code of each
4019 multibyte character in the range table. */
4023 /* Make P points the range table. `+ 2' is to skip flag
4024 bits for a character class. */
4027 /* Extract the number of ranges in range table into COUNT. */
4030 {
4031 /* Extract the start and end of each range. */
4039 }
4040 }
4041 #endif
4047 #ifndef emacs
4055 /* This match depends on text properties. These end with
4056 aborting optimizations. */
4068 /* Any leading code can possibly start a character which
4069 has or doesn't has the specified category. */
4071 {
4076 }
4079 /* All cases after this match the empty string. These end with
4080 `continue'. */
4103 /* Backward jumps can only go back to code that we've already
4104 visited. `re_compile' should make sure this is true. */
4108 {
4114 p++;
4118 };
4119 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4120 to jump back to "just after here". */
4121 /* Fallthrough */
4131 else
4133 We first go down the "straight" path so as to minimize
4134 stack usage when going through alternatives. */
4138 }
4143 /* This code simply does not properly handle forward jump_n. */
4146 /* jump_n can either jump or fall through. The (backward) jump
4147 case has already been handled, so we only need to look at the
4148 fallthrough case. */
4152 /* If N == 0, it should be an on_failure_jump_loop instead. */
4155 /* We only care about one iteration of the loop, so we don't
4156 need to consider the case where this behaves like an
4157 on_failure_jump. */
4176 /* Getting here means we have found the possible starting
4177 characters for one path of the pattern -- and that the empty
4178 string does not match. We need not follow this path further. */
4182 /* We reached the end without matching anything. */
4186 \f
4187 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4188 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4189 characters can start a string that matches the pattern. This fastmap
4190 is used by re_search to skip quickly over impossible starting points.
4192 Character codes above (1 << BYTEWIDTH) are not represented in the
4193 fastmap, but the leading codes are represented. Thus, the fastmap
4194 indicates which character sets could start a match.
4196 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4197 area as BUFP->fastmap.
4199 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4200 the pattern buffer.
4202 Returns 0 if we succeed, -2 if an internal error. */
4204 int
4206 {
4220 \f
4221 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4222 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4223 this memory for recording register information. STARTS and ENDS
4224 must be allocated using the malloc library routine, and must each
4225 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4227 If NUM_REGS == 0, then subsequent matches should allocate their own
4228 register data.
4230 Unless this function is called, the first search or match using
4231 PATTERN_BUFFER will allocate its own register data, without
4232 freeing the old data. */
4234 void
4235 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4236 {
4238 {
4243 }
4244 else
4245 {
4249 }
4250 }
4252 \f
4253 /* Searching routines. */
4255 /* Like re_search_2, below, but only one string is specified, and
4256 doesn't let you say where to stop matching. */
4258 regoff_t
4261 {
4264 }
4267 /* Head address of virtual concatenation of string. */
4268 #define HEAD_ADDR_VSTRING(P) \
4269 (((P) >= size1 ? string2 : string1))
4271 /* Address of POS in the concatenation of virtual string. */
4272 #define POS_ADDR_VSTRING(POS) \
4273 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4275 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4276 virtual concatenation of STRING1 and STRING2, starting first at index
4277 STARTPOS, then at STARTPOS + 1, and so on.
4279 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4281 RANGE is how far to scan while trying to match. RANGE = 0 means try
4282 only at STARTPOS; in general, the last start tried is STARTPOS +
4283 RANGE.
4285 In REGS, return the indices of the virtual concatenation of STRING1
4286 and STRING2 that matched the entire BUFP->buffer and its contained
4287 subexpressions.
4289 Do not consider matching one past the index STOP in the virtual
4290 concatenation of STRING1 and STRING2.
4292 We return either the position in the strings at which the match was
4293 found, -1 if no match, or -2 if error (such as failure
4294 stack overflow). */
4296 regoff_t
4300 {
4301 regoff_t val;
4308 boolean anchored_start;
4309 /* Nonzero if we are searching multibyte string. */
4312 /* Check for out-of-range STARTPOS. */
4316 /* Fix up RANGE if it might eventually take us outside
4317 the virtual concatenation of STRING1 and STRING2.
4318 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4324 /* If the search isn't to be a backwards one, don't waste time in a
4325 search for a pattern anchored at beginning of buffer. */
4327 {
4330 else
4332 }
4334 #ifdef emacs
4335 /* In a forward search for something that starts with \=.
4336 don't keep searching past point. */
4338 {
4342 }
4345 /* Update the fastmap now if not correct already. */
4349 /* See whether the pattern is anchored. */
4352 #ifdef emacs
4354 {
4358 }
4359 #endif
4361 /* Loop through the string, looking for a place to start matching. */
4363 {
4364 /* If the pattern is anchored,
4365 skip quickly past places we cannot match.
4366 We don't bother to treat startpos == 0 specially
4367 because that case doesn't repeat. */
4369 {
4374 }
4376 /* If a fastmap is supplied, skip quickly over characters that
4377 cannot be the start of a match. If the pattern can match the
4378 null string, however, we don't need to skip characters; we want
4379 the first null string. */
4381 {
4388 {
4395 /* Written out as an if-else to avoid testing `translate'
4396 inside the loop. */
4398 {
4401 {
4411 }
4412 else
4414 {
4425 d++;
4426 range--;
4427 }
4428 }
4429 else
4430 {
4433 {
4441 }
4442 else
4444 {
4445 d++;
4446 range--;
4447 }
4448 }
4450 }
4452 {
4454 {
4459 }
4460 else
4461 {
4472 }
4473 }
4474 }
4476 /* If can't match the null string, and that's all we have left, fail. */
4490 advance:
4494 {
4495 /* Update STARTPOS to the next character boundary. */
4497 {
4505 }
4506 else
4507 {
4508 range--;
4509 startpos++;
4510 }
4511 }
4512 else
4513 {
4514 range++;
4515 startpos--;
4517 /* Update STARTPOS to the previous character boundary. */
4519 {
4524 /* Find the head of multibyte form. */
4531 }
4532 }
4533 }
4537 \f
4538 /* Declarations and macros for re_match_2. */
4542 RE_TRANSLATE_TYPE translate,
4545 /* This converts PTR, a pointer into one of the search strings `string1'
4546 and `string2' into an offset from the beginning of that string. */
4547 #define POINTER_TO_OFFSET(ptr) \
4548 (FIRST_STRING_P (ptr) \
4549 ? ((regoff_t) ((ptr) - string1)) \
4550 : ((regoff_t) ((ptr) - string2 + size1)))
4552 /* Call before fetching a character with *d. This switches over to
4553 string2 if necessary.
4554 Check re_match_2_internal for a discussion of why end_match_2 might
4555 not be within string2 (but be equal to end_match_1 instead). */
4556 #define PREFETCH() \
4557 while (d == dend) \
4558 { \
4560 if (dend == end_match_2) \
4561 goto fail; \
4563 d = string2; \
4564 dend = end_match_2; \
4565 }
4567 /* Call before fetching a char with *d if you already checked other limits.
4568 This is meant for use in lookahead operations like wordend, etc..
4569 where we might need to look at parts of the string that might be
4570 outside of the LIMITs (i.e past `stop'). */
4571 #define PREFETCH_NOLIMIT() \
4572 if (d == end1) \
4573 { \
4574 d = string2; \
4575 dend = end_match_2; \
4576 } \
4578 /* Test if at very beginning or at very end of the virtual concatenation
4579 of `string1' and `string2'. If only one string, it's `string2'. */
4580 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4581 #define AT_STRINGS_END(d) ((d) == end2)
4583 /* Disabled due to a compiler bug -- see comment at case wordbound */
4585 /* The comment at case wordbound is following one, but we don't use
4586 AT_WORD_BOUNDARY anymore to support multibyte form.
4588 The DEC Alpha C compiler 3.x generates incorrect code for the
4589 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4590 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4591 macro and introducing temporary variables works around the bug. */
4593 #if 0
4594 /* Test if D points to a character which is word-constituent. We have
4595 two special cases to check for: if past the end of string1, look at
4596 the first character in string2; and if before the beginning of
4597 string2, look at the last character in string1. */
4598 #define WORDCHAR_P(d) \
4599 (SYNTAX ((d) == end1 ? *string2 \
4600 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4601 == Sword)
4603 /* Test if the character before D and the one at D differ with respect
4604 to being word-constituent. */
4605 #define AT_WORD_BOUNDARY(d) \
4606 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4607 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4608 #endif
4610 /* Free everything we malloc. */
4611 #ifdef MATCH_MAY_ALLOCATE
4612 # define FREE_VAR(var) \
4613 do { \
4614 if (var) \
4615 { \
4616 REGEX_FREE (var); \
4617 var = NULL; \
4618 } \
4619 } while (0)
4620 # define FREE_VARIABLES() \
4621 do { \
4622 REGEX_FREE_STACK (fail_stack.stack); \
4623 FREE_VAR (regstart); \
4624 FREE_VAR (regend); \
4625 FREE_VAR (best_regstart); \
4626 FREE_VAR (best_regend); \
4627 } while (0)
4628 #else
4632 \f
4633 /* Optimization routines. */
4635 /* If the operation is a match against one or more chars,
4636 return a pointer to the next operation, else return NULL. */
4639 {
4641 {
4652 {
4657 }
4658 else
4664 #ifdef emacs
4668 p++;
4673 }
4675 }
4678 /* Jump over non-matching operations. */
4681 {
4684 {
4686 {
4699 }
4700 }
4703 }
4705 /* Non-zero if "p1 matches something" implies "p2 fails". */
4706 static int
4708 {
4709 re_opcode_t op2;
4716 /* Skip over open/close-group commands.
4717 If what follows this loop is a ...+ construct,
4718 look at what begins its body, since we will have to
4719 match at least one of that. */
4721 /* The same skip can be done for p1, except that this function
4722 is only used in the case where p1 is a simple match operator. */
4723 /* p1 = skip_noops (p1, pend); */
4731 {
4734 /* If we're at the end of the pattern, we can change. */
4736 {
4739 }
4744 {
4745 register re_wchar_t c
4750 {
4752 {
4755 }
4756 }
4760 {
4763 /* Test if C is listed in charset (or charset_not)
4764 at `p1'. */
4766 {
4770 }
4774 /* `not' is equal to 1 if c would match, which means
4775 that we can't change to pop_failure_jump. */
4777 {
4780 }
4781 }
4784 {
4787 }
4788 }
4792 {
4794 /* Reuse the code above. */
4797 /* It is hard to list up all the character in charset
4798 P2 if it includes multibyte character. Give up in
4799 such case. */
4801 {
4802 /* Now, we are sure that P2 has no range table.
4803 So, for the size of bitmap in P2, `p2[1]' is
4804 enough. But P1 may have range table, so the
4805 size of bitmap table of P1 is extracted by
4806 using macro `CHARSET_BITMAP_SIZE'.
4808 In a multibyte case, we know that all the character
4809 listed in P2 is ASCII. In a unibyte case, P1 has only a
4810 bitmap table. So, in both cases, it is enough to test
4811 only the bitmap table of P1. */
4814 {
4816 /* We win if the charset inside the loop
4817 has no overlap with the one after the loop. */
4821 idx++)
4827 {
4830 }
4831 }
4833 {
4835 /* We win if the charset_not inside the loop lists
4836 every character listed in the charset after. */
4844 {
4847 }
4848 }
4849 }
4850 }
4855 {
4858 /* Reuse the code above. */
4861 /* When we have two charset_not, it's very unlikely that
4862 they don't overlap. The union of the two sets of excluded
4863 chars should cover all possible chars, which, as a matter of
4864 fact, is virtually impossible in multibyte buffers. */
4866 }
4890 #ifdef emacs
4898 ;
4899 }
4901 /* Safe default. */
4903 }
4905 \f
4906 /* Matching routines. */
4909 /* re_match is like re_match_2 except it takes only a single string. */
4911 regoff_t
4914 {
4918 }
4922 #ifdef emacs
4923 /* In Emacs, this is the string or buffer in which we
4924 are matching. It is used for looking up syntax properties. */
4925 Lisp_Object re_match_object;
4926 #endif
4928 /* re_match_2 matches the compiled pattern in BUFP against the
4929 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4930 and SIZE2, respectively). We start matching at POS, and stop
4931 matching at STOP.
4933 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4934 store offsets for the substring each group matched in REGS. See the
4935 documentation for exactly how many groups we fill.
4937 We return -1 if no match, -2 if an internal error (such as the
4938 failure stack overflowing). Otherwise, we return the length of the
4939 matched substring. */
4941 regoff_t
4945 {
4946 regoff_t result;
4948 #ifdef emacs
4949 ssize_t charpos;
4953 #endif
4959 }
4963 /* This is a separate function so that we can force an alloca cleanup
4964 afterwards. */
4965 static regoff_t
4969 {
4970 /* General temporaries. */
4971 ssize_t mcnt;
4974 /* Just past the end of the corresponding string. */
4977 /* Pointers into string1 and string2, just past the last characters in
4978 each to consider matching. */
4981 /* Where we are in the data, and the end of the current string. */
4984 /* Used sometimes to remember where we were before starting matching
4985 an operator so that we can go back in case of failure. This "atomic"
4986 behavior of matching opcodes is indispensable to the correctness
4987 of the on_failure_keep_string_jump optimization. */
4990 /* Where we are in the pattern, and the end of the pattern. */
4994 /* We use this to map every character in the string. */
4997 /* Nonzero if BUFP is setup from a multibyte regex. */
5000 /* Nonzero if STRING1/STRING2 are multibyte. */
5003 /* Failure point stack. Each place that can handle a failure further
5004 down the line pushes a failure point on this stack. It consists of
5005 regstart, and regend for all registers corresponding to
5006 the subexpressions we're currently inside, plus the number of such
5007 registers, and, finally, two char *'s. The first char * is where
5008 to resume scanning the pattern; the second one is where to resume
5009 scanning the strings. */
5011 fail_stack_type fail_stack;
5012 #endif
5013 #ifdef DEBUG
5015 #endif
5017 #if defined REL_ALLOC && defined REGEX_MALLOC
5018 /* This holds the pointer to the failure stack, when
5019 it is allocated relocatably. */
5021 #endif
5023 /* We fill all the registers internally, independent of what we
5024 return, for use in backreferences. The number here includes
5025 an element for register zero. */
5028 /* Information on the contents of registers. These are pointers into
5029 the input strings; they record just what was matched (on this
5030 attempt) by a subexpression part of the pattern, that is, the
5031 regnum-th regstart pointer points to where in the pattern we began
5032 matching and the regnum-th regend points to right after where we
5033 stopped matching the regnum-th subexpression. (The zeroth register
5034 keeps track of what the whole pattern matches.) */
5037 #endif
5039 /* The following record the register info as found in the above
5040 variables when we find a match better than any we've seen before.
5041 This happens as we backtrack through the failure points, which in
5042 turn happens only if we have not yet matched the entire string. */
5046 #endif
5048 /* Logically, this is `best_regend[0]'. But we don't want to have to
5049 allocate space for that if we're not allocating space for anything
5050 else (see below). Also, we never need info about register 0 for
5051 any of the other register vectors, and it seems rather a kludge to
5052 treat `best_regend' differently than the rest. So we keep track of
5053 the end of the best match so far in a separate variable. We
5054 initialize this to NULL so that when we backtrack the first time
5055 and need to test it, it's not garbage. */
5058 #ifdef DEBUG
5059 /* Counts the total number of registers pushed. */
5061 #endif
5067 #ifdef MATCH_MAY_ALLOCATE
5068 /* Do not bother to initialize all the register variables if there are
5069 no groups in the pattern, as it takes a fair amount of time. If
5070 there are groups, we include space for register 0 (the whole
5071 pattern), even though we never use it, since it simplifies the
5072 array indexing. We should fix this. */
5074 {
5081 {
5084 }
5085 }
5086 else
5087 {
5088 /* We must initialize all our variables to NULL, so that
5089 `FREE_VARIABLES' doesn't try to free them. */
5091 }
5094 /* The starting position is bogus. */
5096 {
5099 }
5101 /* Initialize subexpression text positions to -1 to mark ones that no
5102 start_memory/stop_memory has been seen for. Also initialize the
5103 register information struct. */
5107 /* We move `string1' into `string2' if the latter's empty -- but not if
5108 `string1' is null. */
5110 {
5115 }
5119 /* `p' scans through the pattern as `d' scans through the data.
5120 `dend' is the end of the input string that `d' points within. `d'
5121 is advanced into the following input string whenever necessary, but
5122 this happens before fetching; therefore, at the beginning of the
5123 loop, `d' can be pointing at the end of a string, but it cannot
5124 equal `string2'. */
5126 {
5127 /* Only match within string2. */
5131 }
5132 else
5133 {
5135 {
5136 /* Only match within string1. */
5138 /* BEWARE!
5139 When we reach end_match_1, PREFETCH normally switches to string2.
5140 But in the present case, this means that just doing a PREFETCH
5141 makes us jump from `stop' to `gap' within the string.
5142 What we really want here is for the search to stop as
5143 soon as we hit end_match_1. That's why we set end_match_2
5144 to end_match_1 (since PREFETCH fails as soon as we hit
5145 end_match_2). */
5147 }
5148 else
5150 moving `d' from end1 to string2 will not prevent the d == dend
5151 check from catching the end of string. */
5154 }
5157 }
5165 /* This loops over pattern commands. It exits by returning from the
5166 function if the match is complete, or it drops through if the match
5167 fails at this starting point in the input data. */
5169 {
5176 /* If we haven't matched the entire string, and we want the
5177 longest match, try backtracking. */
5179 {
5180 /* 1 if this match ends in the same string (string1 or string2)
5181 as the best previous match. */
5184 /* 1 if this match is the best seen so far. */
5185 boolean best_match_p;
5187 /* AIX compiler got confused when this was combined
5188 with the previous declaration. */
5191 else
5199 /* If exceeds best match so far, save it. */
5201 {
5208 {
5211 }
5212 }
5214 }
5216 /* If no failure points, don't restore garbage. And if
5217 last match is real best match, don't restore second
5218 best one. */
5220 {
5221 restore_best_regs:
5222 /* Restore best match. It may happen that `dend ==
5223 end_match_1' while the restored d is in string2.
5224 For example, the pattern `x.*y.*z' against the
5225 strings `x-' and `y-z-', if the two strings are
5226 not consecutive in memory. */
5234 {
5237 }
5238 }
5241 succeed_label:
5244 /* If caller wants register contents data back, do it. */
5246 {
5247 /* Have the register data arrays been allocated? */
5250 extra element beyond `num_regs' for the `-1' marker
5251 GNU code uses. */
5256 {
5259 }
5261 }
5264 allocated, reallocate them. If we need fewer, just
5265 leave it alone. */
5267 {
5272 {
5275 }
5276 }
5277 }
5278 else
5279 {
5280 /* These braces fend off a "empty body in an else-statement"
5281 warning under GCC when assert expands to nothing. */
5283 }
5285 /* Convert the pointer data in `regstart' and `regend' to
5286 indices. Register zero has to be set differently,
5287 since we haven't kept track of any info for it. */
5289 {
5292 }
5294 /* Go through the first `min (num_regs, regs->num_regs)'
5295 registers, since that is all we initialized. */
5297 {
5300 else
5301 {
5306 }
5307 }
5309 /* If the regs structure we return has more elements than
5310 were in the pattern, set the extra elements to -1. If
5311 we (re)allocated the registers, this is the case,
5312 because we always allocate enough to have at least one
5313 -1 at the end. */
5329 }
5331 /* Otherwise match next pattern command. */
5333 {
5334 /* Ignore these. Used to ignore the n of succeed_n's which
5335 currently have n == 0. */
5344 /* Match the next n pattern characters exactly. The following
5345 byte in the pattern defines n, and the n bytes after that
5346 are the characters to match. */
5351 /* Remember the start point to rollback upon failure. */
5354 #ifndef emacs
5355 /* This is written out as an if-else so we don't waste time
5356 testing `translate' inside the loop. */
5358 do
5359 {
5362 {
5365 }
5366 d++;
5367 }
5369 else
5370 do
5371 {
5374 {
5377 }
5378 }
5381 /* The cost of testing `translate' is comparatively small. */
5383 do
5384 {
5391 else
5392 {
5395 }
5399 {
5402 }
5407 }
5409 else
5410 do
5411 {
5417 {
5420 }
5421 else
5422 {
5425 }
5428 {
5433 }
5434 else
5437 {
5440 }
5442 d++;
5443 }
5445 #endif
5449 /* Match any character except possibly a newline or a null. */
5451 {
5453 re_wchar_t buf_ch;
5459 target_multibyte);
5470 }
5476 {
5481 /* Start of actual range_table, or end of bitmap if there is no
5482 range table. */
5485 /* Nonzero if there is a range table. */
5488 /* Number of ranges of range table. This is not included
5489 in the initial byte-length of the command. */
5492 /* Whether matching against a unibyte character. */
5500 {
5503 }
5508 {
5514 {
5517 }
5518 }
5519 else
5520 {
5524 {
5528 {
5531 }
5532 }
5533 else
5535 }
5539 /* Cast to `unsigned' instead of `unsigned char' in
5540 case the bit list is a full 32 bytes long. */
5544 }
5545 #ifdef emacs
5547 {
5557 else
5559 }
5564 else
5570 }
5574 /* The beginning of a group is represented by start_memory.
5575 The argument is the register number. The text
5576 matched within the group is recorded (in the internal
5577 registers data structure) under the register number. */
5581 /* In case we need to undo this operation (via backtracking). */
5588 /* Move past the register number and inner group count. */
5593 /* The stop_memory opcode represents the end of a group. Its
5594 argument is the same as start_memory's: the register number. */
5599 /* Strictly speaking, there should be code such as:
5601 assert (REG_UNSET (regend[*p]));
5602 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5604 But the only info to be pushed is regend[*p] and it is known to
5605 be UNSET, so there really isn't anything to push.
5606 Not pushing anything, on the other hand deprives us from the
5607 guarantee that regend[*p] is UNSET since undoing this operation
5608 will not reset its value properly. This is not important since
5609 the value will only be read on the next start_memory or at
5610 the very end and both events can only happen if this stop_memory
5611 is *not* undone. */
5616 /* Move past the register number and the inner group count. */
5621 /* \<digit> has been turned into a `duplicate' command which is
5622 followed by the numeric value of <digit> as the register number. */
5624 {
5629 /* Can't back reference a group which we've never matched. */
5633 /* Where in input to try to start matching. */
5636 /* Remember the start point to rollback upon failure. */
5639 /* Where to stop matching; if both the place to start and
5640 the place to stop matching are in the same string, then
5641 set to the place to stop, otherwise, for now have to use
5642 the end of the first string. */
5648 {
5649 /* If necessary, advance to next segment in register
5650 contents. */
5652 {
5656 /* End of string1 => advance to string2. */
5659 }
5660 /* At end of register contents => success */
5663 /* If necessary, advance to next segment in data. */
5666 /* How many characters left in this segment to match. */
5669 /* Want how many consecutive characters we can match in
5670 one shot, so, if necessary, adjust the count. */
5674 /* Compare that many; failure if mismatch, else move
5675 past them. */
5679 {
5682 }
5684 }
5685 }
5689 /* begline matches the empty string at the beginning of the string
5690 (unless `not_bol' is set in `bufp'), and after newlines. */
5695 {
5697 }
5698 else
5699 {
5704 }
5705 /* In all other cases, we fail. */
5709 /* endline is the dual of begline. */
5714 {
5716 }
5717 else
5718 {
5722 }
5726 /* Match at the very beginning of the data. */
5734 /* Match at the very end of the data. */
5742 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5743 pushes NULL as the value for the string on the stack. Then
5744 `POP_FAILURE_POINT' will keep the current value for the
5745 string, instead of restoring it. To see why, consider
5746 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5747 then the . fails against the \n. But the next thing we want
5748 to do is match the \n against the \n; if we restored the
5749 string value, we would be back at the foo.
5751 Because this is used only in specific cases, we don't need to
5752 check all the things that `on_failure_jump' does, to make
5753 sure the right things get saved on the stack. Hence we don't
5754 share its code. The only reason to push anything on the
5755 stack at all is that otherwise we would have to change
5756 `anychar's code to do something besides goto fail in this
5757 case; that seems worse than this. */
5766 /* A nasty loop is introduced by the non-greedy *? and +?.
5767 With such loops, the stack only ever contains one failure point
5768 at a time, so that a plain on_failure_jump_loop kind of
5769 cycle detection cannot work. Worse yet, such a detection
5770 can not only fail to detect a cycle, but it can also wrongly
5771 detect a cycle (between different instantiations of the same
5772 loop).
5773 So the method used for those nasty loops is a little different:
5774 We use a special cycle-detection-stack-frame which is pushed
5775 when the on_failure_jump_nastyloop failure-point is *popped*.
5776 This special frame thus marks the beginning of one iteration
5777 through the loop and we can hence easily check right here
5778 whether something matched between the beginning and the end of
5779 the loop. */
5786 {
5790 /* If there's a cycle, just continue without pushing
5791 this failure point. The failure point is the "try again"
5792 option, which shouldn't be tried.
5793 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5795 }
5798 /* Simple loop detecting on_failure_jump: just check on the
5799 failure stack if the same spot was already hit earlier. */
5801 on_failure:
5805 {
5809 /* If there's a cycle, get out of the loop, as if the matching
5810 had failed. We used to just `goto fail' here, but that was
5811 aborting the search a bit too early: we want to keep the
5812 empty-loop-match and keep matching after the loop.
5813 We want (x?)*y\1z to match both xxyz and xxyxz. */
5815 else
5817 }
5821 /* Uses of on_failure_jump:
5823 Each alternative starts with an on_failure_jump that points
5824 to the beginning of the next alternative. Each alternative
5825 except the last ends with a jump that in effect jumps past
5826 the rest of the alternatives. (They really jump to the
5827 ending jump of the following alternative, because tensioning
5828 these jumps is a hassle.)
5830 Repeats start with an on_failure_jump that points past both
5831 the repetition text and either the following jump or
5832 pop_failure_jump back to this on_failure_jump. */
5841 /* This operation is used for greedy *.
5842 Compare the beginning of the repeat with what in the
5843 pattern follows its end. If we can establish that there
5844 is nothing that they would both match, i.e., that we
5845 would have to backtrack because of (as in, e.g., `a*a')
5846 then we can use a non-backtracking loop based on
5847 on_failure_keep_string_jump instead of on_failure_jump. */
5852 {
5854 /* Here, we discard `const', making re_match non-reentrant. */
5859 instruction once it's been changed. */
5863 /* Ensure this is a indeed the trivial kind of loop
5864 we are expecting. */
5869 {
5870 /* Use a fast `on_failure_keep_string_jump' loop. */
5874 }
5875 else
5876 {
5877 /* Default to a safe `on_failure_jump' loop. */
5880 }
5882 }
5885 /* Unconditionally jump (without popping any failure points). */
5887 unconditional_jump:
5888 IMMEDIATE_QUIT_CHECK;
5896 /* Have to succeed matching what follows at least n times.
5897 After that, handle like `on_failure_jump'. */
5899 /* Signedness doesn't matter since we only compare MCNT to 0. */
5903 /* Originally, mcnt is how many times we HAVE to succeed. */
5905 {
5906 /* Here, we discard `const', making re_match non-reentrant. */
5908 mcnt--;
5911 }
5912 else
5913 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5918 /* Signedness doesn't matter since we only compare MCNT to 0. */
5922 /* Originally, this is how many times we CAN jump. */
5924 {
5925 /* Here, we discard `const', making re_match non-reentrant. */
5927 mcnt--;
5930 }
5931 /* If don't have to jump any more, skip over the rest of command. */
5932 else
5937 {
5942 /* Here, we discard `const', making re_match non-reentrant. */
5944 /* Signedness doesn't matter since we only copy MCNT's bits . */
5949 }
5953 {
5957 /* We SUCCEED (or FAIL) in one of the following cases: */
5959 /* Case 1: D is at the beginning or the end of string. */
5962 else
5963 {
5964 /* C1 is the character before D, S1 is the syntax of C1, C2
5965 is the character at D, and S2 is the syntax of C2. */
5969 #ifdef emacs
5973 #endif
5976 #ifdef emacs
5978 #endif
5985 /* Case 3: Both of S1 and S2 are Sword, and macro
5986 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5989 }
5992 else
5994 }
5999 /* We FAIL in one of the following cases: */
6001 /* Case 1: D is at the end of string. */
6004 else
6005 {
6006 /* C1 is the character before D, S1 is the syntax of C1, C2
6007 is the character at D, and S2 is the syntax of C2. */
6011 #ifdef emacs
6015 #endif
6020 /* Case 2: S2 is not Sword. */
6024 /* Case 3: D is not at the beginning of string ... */
6026 {
6028 #ifdef emacs
6030 #endif
6033 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6034 returns 0. */
6037 }
6038 }
6044 /* We FAIL in one of the following cases: */
6046 /* Case 1: D is at the beginning of string. */
6049 else
6050 {
6051 /* C1 is the character before D, S1 is the syntax of C1, C2
6052 is the character at D, and S2 is the syntax of C2. */
6056 #ifdef emacs
6060 #endif
6064 /* Case 2: S1 is not Sword. */
6068 /* Case 3: D is not at the end of string ... */
6070 {
6073 #ifdef emacs
6075 #endif
6078 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6079 returns 0. */
6082 }
6083 }
6089 /* We FAIL in one of the following cases: */
6091 /* Case 1: D is at the end of string. */
6094 else
6095 {
6096 /* C1 is the character before D, S1 is the syntax of C1, C2
6097 is the character at D, and S2 is the syntax of C2. */
6100 #ifdef emacs
6104 #endif
6109 /* Case 2: S2 is neither Sword nor Ssymbol. */
6113 /* Case 3: D is not at the beginning of string ... */
6115 {
6117 #ifdef emacs
6119 #endif
6122 /* ... and S1 is Sword or Ssymbol. */
6125 }
6126 }
6132 /* We FAIL in one of the following cases: */
6134 /* Case 1: D is at the beginning of string. */
6137 else
6138 {
6139 /* C1 is the character before D, S1 is the syntax of C1, C2
6140 is the character at D, and S2 is the syntax of C2. */
6143 #ifdef emacs
6147 #endif
6151 /* Case 2: S1 is neither Ssymbol nor Sword. */
6155 /* Case 3: D is not at the end of string ... */
6157 {
6160 #ifdef emacs
6162 #endif
6165 /* ... and S2 is Sword or Ssymbol. */
6168 }
6169 }
6174 {
6179 #ifdef emacs
6180 {
6184 }
6185 #endif
6186 {
6188 re_wchar_t c;
6194 }
6195 }
6198 #ifdef emacs
6219 {
6226 {
6228 re_wchar_t c;
6233 }
6234 }
6241 }
6245 /* We goto here if a matching operation fails. */
6246 fail:
6247 IMMEDIATE_QUIT_CHECK;
6249 {
6251 /* A restart point is known. Restore to that state. */
6255 {
6263 /* Fallthrough */
6269 continue_failure_jump:
6275 /* A special frame used for nastyloops. */
6280 }
6286 }
6287 else
6298 \f
6299 /* Subroutine definitions for re_match_2. */
6301 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6302 bytes; nonzero otherwise. */
6304 static int
6307 {
6312 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6313 different lengths, but relying on a single `len' would break this. -sm */
6315 {
6327 }
6333 }
6334 \f
6335 /* Entry points for GNU code. */
6337 /* re_compile_pattern is the GNU regular expression compiler: it
6338 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6339 Returns 0 if the pattern was valid, otherwise an error string.
6341 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6342 are set in BUFP on entry.
6344 We call regex_compile to do the actual compilation. */
6349 {
6350 reg_errcode_t ret;
6352 /* GNU code is written to assume at least RE_NREGS registers will be set
6353 (and at least one extra will be -1). */
6356 /* And GNU code determines whether or not to get register information
6357 by passing null for the REGS argument to re_match, etc., not by
6358 setting no_sub. */
6366 }
6368 \f
6369 /* Entry points compatible with 4.2 BSD regex library. We don't define
6370 them unless specifically requested. */
6372 #if defined _REGEX_RE_COMP || defined _LIBC
6374 /* BSD has one and only one pattern buffer. */
6378 # ifdef _LIBC
6379 /* Make these definitions weak in libc, so POSIX programs can redefine
6380 these names if they don't use our functions, and still use
6381 regcomp/regexec below without link errors. */
6382 weak_function
6383 # endif
6385 {
6386 reg_errcode_t ret;
6389 {
6391 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6394 }
6397 {
6400 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6406 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6408 }
6410 /* Since `re_exec' always passes NULL for the `regs' argument, we
6411 don't need to initialize the pattern buffer fields which affect it. */
6418 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6420 }
6423 int
6424 # ifdef _LIBC
6425 weak_function
6426 # endif
6428 {
6430 return
6432 }
6434 \f
6435 /* POSIX.2 functions. Don't define these for Emacs. */
6437 #ifndef emacs
6439 /* regcomp takes a regular expression as a string and compiles it.
6441 PREG is a regex_t *. We do not expect any fields to be initialized,
6442 since POSIX says we shouldn't. Thus, we set
6444 `buffer' to the compiled pattern;
6445 `used' to the length of the compiled pattern;
6446 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6447 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6448 RE_SYNTAX_POSIX_BASIC;
6449 `fastmap' to an allocated space for the fastmap;
6450 `fastmap_accurate' to zero;
6451 `re_nsub' to the number of subexpressions in PATTERN.
6453 PATTERN is the address of the pattern string.
6455 CFLAGS is a series of bits which affect compilation.
6457 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6458 use POSIX basic syntax.
6460 If REG_NEWLINE is set, then . and [^...] don't match newline.
6461 Also, regexec will try a match beginning after every newline.
6463 If REG_ICASE is set, then we considers upper- and lowercase
6464 versions of letters to be equivalent when matching.
6466 If REG_NOSUB is set, then when PREG is passed to regexec, that
6467 routine will report only success or failure, and nothing about the
6468 registers.
6470 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6471 the return codes and their meanings.) */
6473 reg_errcode_t
6476 {
6477 reg_errcode_t ret;
6478 reg_syntax_t syntax
6482 /* regex_compile will allocate the space for the compiled pattern. */
6487 /* Try to allocate space for the fastmap. */
6491 {
6494 preg->translate
6500 /* Map uppercase characters to corresponding lowercase ones. */
6503 }
6504 else
6507 /* If REG_NEWLINE is set, newlines are treated differently. */
6512 }
6513 else
6518 /* POSIX says a null character in the pattern terminates it, so we
6519 can use strlen here in compiling the pattern. */
6522 /* POSIX doesn't distinguish between an unmatched open-group and an
6523 unmatched close-group: both are REG_EPAREN. */
6529 buffer. */
6535 }
6536 }
6538 }
6542 /* regexec searches for a given pattern, specified by PREG, in the
6543 string STRING.
6545 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6546 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6547 least NMATCH elements, and we set them to the offsets of the
6548 corresponding matched substrings.
6550 EFLAGS specifies `execution flags' which affect matching: if
6551 REG_NOTBOL is set, then ^ does not match at the beginning of the
6552 string; if REG_NOTEOL is set, then $ does not match at the end.
6554 We return 0 if we find a match and REG_NOMATCH if not. */
6556 reg_errcode_t
6559 {
6560 regoff_t ret;
6562 regex_t private_preg;
6571 /* The user has told us exactly how many registers to return
6572 information about, via `nmatch'. We have to pass that on to the
6573 matching routines. */
6577 {
6583 }
6585 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6586 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6587 was a little bit longer but still only matching the real part.
6588 This works because the `endline' will check for a '\n' and will find a
6589 '\0', correctly deciding that this is not the end of a line.
6590 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6591 a convenient '\0' there. For all we know, the string could be preceded
6592 by '\n' which would throw things off. */
6594 /* Perform the searching operation. */
6599 /* Copy the register information to the POSIX structure. */
6601 {
6603 {
6607 {
6610 }
6611 }
6613 /* If we needed the temporary register info, free the space now. */
6615 }
6617 /* We want zero return to mean success, unlike `re_search'. */
6619 }
6623 /* Returns a message corresponding to an error code, ERR_CODE, returned
6624 from either regcomp or regexec. We don't use PREG here.
6626 ERR_CODE was previously called ERRCODE, but that name causes an
6627 error with msvc8 compiler. */
6629 size_t
6631 {
6637 /* Only error codes returned by the rest of the code should be passed
6638 to this routine. If we are given anything else, or if other regex
6639 code generates an invalid error code, then the program has a bug.
6640 Dump core so we can fix it. */
6648 {
6650 {
6653 }
6654 else
6656 }
6659 }
6663 /* Free dynamically allocated space used by PREG. */
6665 void
6667 {
6680 }