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) 1993,94,95,96,97,98,99,2000,04 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 2, or (at your option)
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,
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.
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
40 #if defined STDC_HEADERS && !defined emacs
43 /* We need this for `regex.h', and perhaps for the Emacs include files. */
44 # include <sys/types.h>
47 /* Whether to use ISO C Amendment 1 wide char functions.
48 Those should not be used for Emacs since it uses its own. */
50 #define WIDE_CHAR_SUPPORT 1
52 #define WIDE_CHAR_SUPPORT \
53 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
56 /* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
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(errcode, preg, errbuf, errbuf_size) \
70 __regerror(errcode, 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>
98 # define WEAK_ALIAS(a,b)
101 /* This is for other GNU distributions with internationalized messages. */
102 #if HAVE_LIBINTL_H || defined _LIBC
103 # include <libintl.h>
105 # define gettext(msgid) (msgid)
109 /* This define is so xgettext can find the internationalizable
111 # define gettext_noop(String) String
114 /* The `emacs' switch turns on certain matching commands
115 that make sense only in Emacs. */
121 /* Make syntax table lookup grant data in gl_state. */
122 # define SYNTAX_ENTRY_VIA_PROPERTY
125 # include "charset.h"
126 # include "category.h"
131 # define malloc xmalloc
135 # define realloc xrealloc
141 /* Converts the pointer to the char to BEG-based offset from the start. */
142 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
143 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
145 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
146 # define RE_STRING_CHAR(p, s) \
147 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
148 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
149 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
151 /* Set C a (possibly multibyte) character before P. P points into a
152 string which is the virtual concatenation of STR1 (which ends at
153 END1) or STR2 (which ends at END2). */
154 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
158 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
159 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
160 re_char *d0 = dtemp; \
161 PREV_CHAR_BOUNDARY (d0, dlimit); \
162 c = STRING_CHAR (d0, dtemp - d0); \
165 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
169 #else /* not emacs */
171 /* If we are not linking with Emacs proper,
172 we can't use the relocating allocator
173 even if config.h says that we can. */
176 # if defined STDC_HEADERS || defined _LIBC
183 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
184 If nothing else has been done, use the method below. */
185 # ifdef INHIBIT_STRING_HEADER
186 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
187 # if !defined bzero && !defined bcopy
188 # undef INHIBIT_STRING_HEADER
193 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
194 This is used in most programs--a few other programs avoid this
195 by defining INHIBIT_STRING_HEADER. */
196 # ifndef INHIBIT_STRING_HEADER
197 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
201 # define bzero(s, n) (memset (s, '\0', n), (s))
203 # define bzero(s, n) __bzero (s, n)
207 # include <strings.h>
209 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
212 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
217 /* Define the syntax stuff for \<, \>, etc. */
219 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
220 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
222 # ifdef SWITCH_ENUM_BUG
223 # define SWITCH_ENUM_CAST(x) ((int)(x))
225 # define SWITCH_ENUM_CAST(x) (x)
228 /* Dummy macros for non-Emacs environments. */
229 # define BASE_LEADING_CODE_P(c) (0)
230 # define CHAR_CHARSET(c) 0
231 # define CHARSET_LEADING_CODE_BASE(c) 0
232 # define MAX_MULTIBYTE_LENGTH 1
233 # define RE_MULTIBYTE_P(x) 0
234 # define WORD_BOUNDARY_P(c1, c2) (0)
235 # define CHAR_HEAD_P(p) (1)
236 # define SINGLE_BYTE_CHAR_P(c) (1)
237 # define SAME_CHARSET_P(c1, c2) (1)
238 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
239 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
240 # define STRING_CHAR(p, s) (*(p))
241 # define RE_STRING_CHAR STRING_CHAR
242 # define CHAR_STRING(c, s) (*(s) = (c), 1)
243 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
244 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
245 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
246 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
247 # define MAKE_CHAR(charset, c1, c2) (c1)
248 #endif /* not emacs */
251 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
252 # define RE_TRANSLATE_P(TBL) (TBL)
255 /* Get the interface, including the syntax bits. */
258 /* isalpha etc. are used for the character classes. */
263 /* 1 if C is an ASCII character. */
264 # define IS_REAL_ASCII(c) ((c) < 0200)
266 /* 1 if C is a unibyte character. */
267 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
269 /* The Emacs definitions should not be directly affected by locales. */
271 /* In Emacs, these are only used for single-byte characters. */
272 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
273 # define ISCNTRL(c) ((c) < ' ')
274 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
275 || ((c) >= 'a' && (c) <= 'f') \
276 || ((c) >= 'A' && (c) <= 'F'))
278 /* This is only used for single-byte characters. */
279 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
281 /* The rest must handle multibyte characters. */
283 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
284 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
287 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
288 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
291 # define ISALNUM(c) (IS_REAL_ASCII (c) \
292 ? (((c) >= 'a' && (c) <= 'z') \
293 || ((c) >= 'A' && (c) <= 'Z') \
294 || ((c) >= '0' && (c) <= '9')) \
295 : SYNTAX (c) == Sword)
297 # define ISALPHA(c) (IS_REAL_ASCII (c) \
298 ? (((c) >= 'a' && (c) <= 'z') \
299 || ((c) >= 'A' && (c) <= 'Z')) \
300 : SYNTAX (c) == Sword)
302 # define ISLOWER(c) (LOWERCASEP (c))
304 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
305 ? ((c) > ' ' && (c) < 0177 \
306 && !(((c) >= 'a' && (c) <= 'z') \
307 || ((c) >= 'A' && (c) <= 'Z') \
308 || ((c) >= '0' && (c) <= '9'))) \
309 : SYNTAX (c) != Sword)
311 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
313 # define ISUPPER(c) (UPPERCASEP (c))
315 # define ISWORD(c) (SYNTAX (c) == Sword)
317 #else /* not emacs */
319 /* Jim Meyering writes:
321 "... Some ctype macros are valid only for character codes that
322 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
323 using /bin/cc or gcc but without giving an ansi option). So, all
324 ctype uses should be through macros like ISPRINT... If
325 STDC_HEADERS is defined, then autoconf has verified that the ctype
326 macros don't need to be guarded with references to isascii. ...
327 Defining isascii to 1 should let any compiler worth its salt
328 eliminate the && through constant folding."
329 Solaris defines some of these symbols so we must undefine them first. */
332 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
333 # define ISASCII(c) 1
335 # define ISASCII(c) isascii(c)
338 /* 1 if C is an ASCII character. */
339 # define IS_REAL_ASCII(c) ((c) < 0200)
341 /* This distinction is not meaningful, except in Emacs. */
342 # define ISUNIBYTE(c) 1
345 # define ISBLANK(c) (ISASCII (c) && isblank (c))
347 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
350 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
352 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
356 # define ISPRINT(c) (ISASCII (c) && isprint (c))
357 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
358 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
359 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
360 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
361 # define ISLOWER(c) (ISASCII (c) && islower (c))
362 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
363 # define ISSPACE(c) (ISASCII (c) && isspace (c))
364 # define ISUPPER(c) (ISASCII (c) && isupper (c))
365 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
367 # define ISWORD(c) ISALPHA(c)
370 # define TOLOWER(c) _tolower(c)
372 # define TOLOWER(c) tolower(c)
375 /* How many characters in the character set. */
376 # define CHAR_SET_SIZE 256
380 extern char *re_syntax_table
;
382 # else /* not SYNTAX_TABLE */
384 static char re_syntax_table
[CHAR_SET_SIZE
];
395 bzero (re_syntax_table
, sizeof re_syntax_table
);
397 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
399 re_syntax_table
[c
] = Sword
;
401 re_syntax_table
['_'] = Ssymbol
;
406 # endif /* not SYNTAX_TABLE */
408 # define SYNTAX(c) re_syntax_table[(c)]
410 #endif /* not emacs */
413 # define NULL (void *)0
416 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
417 since ours (we hope) works properly with all combinations of
418 machines, compilers, `char' and `unsigned char' argument types.
419 (Per Bothner suggested the basic approach.) */
420 #undef SIGN_EXTEND_CHAR
422 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
423 #else /* not __STDC__ */
424 /* As in Harbison and Steele. */
425 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
428 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
429 use `alloca' instead of `malloc'. This is because using malloc in
430 re_search* or re_match* could cause memory leaks when C-g is used in
431 Emacs; also, malloc is slower and causes storage fragmentation. On
432 the other hand, malloc is more portable, and easier to debug.
434 Because we sometimes use alloca, some routines have to be macros,
435 not functions -- `alloca'-allocated space disappears at the end of the
436 function it is called in. */
440 # define REGEX_ALLOCATE malloc
441 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
442 # define REGEX_FREE free
444 #else /* not REGEX_MALLOC */
446 /* Emacs already defines alloca, sometimes. */
449 /* Make alloca work the best possible way. */
451 # define alloca __builtin_alloca
452 # else /* not __GNUC__ */
455 # endif /* HAVE_ALLOCA_H */
456 # endif /* not __GNUC__ */
458 # endif /* not alloca */
460 # define REGEX_ALLOCATE alloca
462 /* Assumes a `char *destination' variable. */
463 # define REGEX_REALLOCATE(source, osize, nsize) \
464 (destination = (char *) alloca (nsize), \
465 memcpy (destination, source, osize))
467 /* No need to do anything to free, after alloca. */
468 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
470 #endif /* not REGEX_MALLOC */
472 /* Define how to allocate the failure stack. */
474 #if defined REL_ALLOC && defined REGEX_MALLOC
476 # define REGEX_ALLOCATE_STACK(size) \
477 r_alloc (&failure_stack_ptr, (size))
478 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
479 r_re_alloc (&failure_stack_ptr, (nsize))
480 # define REGEX_FREE_STACK(ptr) \
481 r_alloc_free (&failure_stack_ptr)
483 #else /* not using relocating allocator */
487 # define REGEX_ALLOCATE_STACK malloc
488 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
489 # define REGEX_FREE_STACK free
491 # else /* not REGEX_MALLOC */
493 # define REGEX_ALLOCATE_STACK alloca
495 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
496 REGEX_REALLOCATE (source, osize, nsize)
497 /* No need to explicitly free anything. */
498 # define REGEX_FREE_STACK(arg) ((void)0)
500 # endif /* not REGEX_MALLOC */
501 #endif /* not using relocating allocator */
504 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
505 `string1' or just past its end. This works if PTR is NULL, which is
507 #define FIRST_STRING_P(ptr) \
508 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
510 /* (Re)Allocate N items of type T using malloc, or fail. */
511 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
512 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
513 #define RETALLOC_IF(addr, n, t) \
514 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
515 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
517 #define BYTEWIDTH 8 /* In bits. */
519 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
523 #define MAX(a, b) ((a) > (b) ? (a) : (b))
524 #define MIN(a, b) ((a) < (b) ? (a) : (b))
526 /* Type of source-pattern and string chars. */
527 typedef const unsigned char re_char
;
529 typedef char boolean
;
533 static int re_match_2_internal
_RE_ARGS ((struct re_pattern_buffer
*bufp
,
534 re_char
*string1
, int size1
,
535 re_char
*string2
, int size2
,
537 struct re_registers
*regs
,
540 /* These are the command codes that appear in compiled regular
541 expressions. Some opcodes are followed by argument bytes. A
542 command code can specify any interpretation whatsoever for its
543 arguments. Zero bytes may appear in the compiled regular expression. */
549 /* Succeed right away--no more backtracking. */
552 /* Followed by one byte giving n, then by n literal bytes. */
555 /* Matches any (more or less) character. */
558 /* Matches any one char belonging to specified set. First
559 following byte is number of bitmap bytes. Then come bytes
560 for a bitmap saying which chars are in. Bits in each byte
561 are ordered low-bit-first. A character is in the set if its
562 bit is 1. A character too large to have a bit in the map is
563 automatically not in the set.
565 If the length byte has the 0x80 bit set, then that stuff
566 is followed by a range table:
567 2 bytes of flags for character sets (low 8 bits, high 8 bits)
568 See RANGE_TABLE_WORK_BITS below.
569 2 bytes, the number of pairs that follow (upto 32767)
570 pairs, each 2 multibyte characters,
571 each multibyte character represented as 3 bytes. */
574 /* Same parameters as charset, but match any character that is
575 not one of those specified. */
578 /* Start remembering the text that is matched, for storing in a
579 register. Followed by one byte with the register number, in
580 the range 0 to one less than the pattern buffer's re_nsub
584 /* Stop remembering the text that is matched and store it in a
585 memory register. Followed by one byte with the register
586 number, in the range 0 to one less than `re_nsub' in the
590 /* Match a duplicate of something remembered. Followed by one
591 byte containing the register number. */
594 /* Fail unless at beginning of line. */
597 /* Fail unless at end of line. */
600 /* Succeeds if at beginning of buffer (if emacs) or at beginning
601 of string to be matched (if not). */
604 /* Analogously, for end of buffer/string. */
607 /* Followed by two byte relative address to which to jump. */
610 /* Followed by two-byte relative address of place to resume at
611 in case of failure. */
614 /* Like on_failure_jump, but pushes a placeholder instead of the
615 current string position when executed. */
616 on_failure_keep_string_jump
,
618 /* Just like `on_failure_jump', except that it checks that we
619 don't get stuck in an infinite loop (matching an empty string
621 on_failure_jump_loop
,
623 /* Just like `on_failure_jump_loop', except that it checks for
624 a different kind of loop (the kind that shows up with non-greedy
625 operators). This operation has to be immediately preceded
627 on_failure_jump_nastyloop
,
629 /* A smart `on_failure_jump' used for greedy * and + operators.
630 It analyses the loop before which it is put and if the
631 loop does not require backtracking, it changes itself to
632 `on_failure_keep_string_jump' and short-circuits the loop,
633 else it just defaults to changing itself into `on_failure_jump'.
634 It assumes that it is pointing to just past a `jump'. */
635 on_failure_jump_smart
,
637 /* Followed by two-byte relative address and two-byte number n.
638 After matching N times, jump to the address upon failure.
639 Does not work if N starts at 0: use on_failure_jump_loop
643 /* Followed by two-byte relative address, and two-byte number n.
644 Jump to the address N times, then fail. */
647 /* Set the following two-byte relative address to the
648 subsequent two-byte number. The address *includes* the two
652 wordbeg
, /* Succeeds if at word beginning. */
653 wordend
, /* Succeeds if at word end. */
655 wordbound
, /* Succeeds if at a word boundary. */
656 notwordbound
, /* Succeeds if not at a word boundary. */
658 symbeg
, /* Succeeds if at symbol beginning. */
659 symend
, /* Succeeds if at symbol end. */
661 /* Matches any character whose syntax is specified. Followed by
662 a byte which contains a syntax code, e.g., Sword. */
665 /* Matches any character whose syntax is not that specified. */
669 ,before_dot
, /* Succeeds if before point. */
670 at_dot
, /* Succeeds if at point. */
671 after_dot
, /* Succeeds if after point. */
673 /* Matches any character whose category-set contains the specified
674 category. The operator is followed by a byte which contains a
675 category code (mnemonic ASCII character). */
678 /* Matches any character whose category-set does not contain the
679 specified category. The operator is followed by a byte which
680 contains the category code (mnemonic ASCII character). */
685 /* Common operations on the compiled pattern. */
687 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
689 #define STORE_NUMBER(destination, number) \
691 (destination)[0] = (number) & 0377; \
692 (destination)[1] = (number) >> 8; \
695 /* Same as STORE_NUMBER, except increment DESTINATION to
696 the byte after where the number is stored. Therefore, DESTINATION
697 must be an lvalue. */
699 #define STORE_NUMBER_AND_INCR(destination, number) \
701 STORE_NUMBER (destination, number); \
702 (destination) += 2; \
705 /* Put into DESTINATION a number stored in two contiguous bytes starting
708 #define EXTRACT_NUMBER(destination, source) \
710 (destination) = *(source) & 0377; \
711 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
715 static void extract_number
_RE_ARGS ((int *dest
, re_char
*source
));
717 extract_number (dest
, source
)
721 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
722 *dest
= *source
& 0377;
726 # ifndef EXTRACT_MACROS /* To debug the macros. */
727 # undef EXTRACT_NUMBER
728 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
729 # endif /* not EXTRACT_MACROS */
733 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
734 SOURCE must be an lvalue. */
736 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
738 EXTRACT_NUMBER (destination, source); \
743 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
746 extract_number_and_incr (destination
, source
)
750 extract_number (destination
, *source
);
754 # ifndef EXTRACT_MACROS
755 # undef EXTRACT_NUMBER_AND_INCR
756 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
757 extract_number_and_incr (&dest, &src)
758 # endif /* not EXTRACT_MACROS */
762 /* Store a multibyte character in three contiguous bytes starting
763 DESTINATION, and increment DESTINATION to the byte after where the
764 character is stored. Therefore, DESTINATION must be an lvalue. */
766 #define STORE_CHARACTER_AND_INCR(destination, character) \
768 (destination)[0] = (character) & 0377; \
769 (destination)[1] = ((character) >> 8) & 0377; \
770 (destination)[2] = (character) >> 16; \
771 (destination) += 3; \
774 /* Put into DESTINATION a character stored in three contiguous bytes
775 starting at SOURCE. */
777 #define EXTRACT_CHARACTER(destination, source) \
779 (destination) = ((source)[0] \
780 | ((source)[1] << 8) \
781 | ((source)[2] << 16)); \
785 /* Macros for charset. */
787 /* Size of bitmap of charset P in bytes. P is a start of charset,
788 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
789 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
791 /* Nonzero if charset P has range table. */
792 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
794 /* Return the address of range table of charset P. But not the start
795 of table itself, but the before where the number of ranges is
796 stored. `2 +' means to skip re_opcode_t and size of bitmap,
797 and the 2 bytes of flags at the start of the range table. */
798 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
800 /* Extract the bit flags that start a range table. */
801 #define CHARSET_RANGE_TABLE_BITS(p) \
802 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
803 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
805 /* Test if C is listed in the bitmap of charset P. */
806 #define CHARSET_LOOKUP_BITMAP(p, c) \
807 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
808 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
810 /* Return the address of end of RANGE_TABLE. COUNT is number of
811 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
812 is start of range and end of range. `* 3' is size of each start
814 #define CHARSET_RANGE_TABLE_END(range_table, count) \
815 ((range_table) + (count) * 2 * 3)
817 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
818 COUNT is number of ranges in RANGE_TABLE. */
819 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
822 re_wchar_t range_start, range_end; \
824 re_char *range_table_end \
825 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
827 for (p = (range_table); p < range_table_end; p += 2 * 3) \
829 EXTRACT_CHARACTER (range_start, p); \
830 EXTRACT_CHARACTER (range_end, p + 3); \
832 if (range_start <= (c) && (c) <= range_end) \
841 /* Test if C is in range table of CHARSET. The flag NOT is negated if
842 C is listed in it. */
843 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
846 /* Number of ranges in range table. */ \
848 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
850 EXTRACT_NUMBER_AND_INCR (count, range_table); \
851 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
855 /* If DEBUG is defined, Regex prints many voluminous messages about what
856 it is doing (if the variable `debug' is nonzero). If linked with the
857 main program in `iregex.c', you can enter patterns and strings
858 interactively. And if linked with the main program in `main.c' and
859 the other test files, you can run the already-written tests. */
863 /* We use standard I/O for debugging. */
866 /* It is useful to test things that ``must'' be true when debugging. */
869 static int debug
= -100000;
871 # define DEBUG_STATEMENT(e) e
872 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
873 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
874 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
875 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
876 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
877 if (debug > 0) print_partial_compiled_pattern (s, e)
878 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
879 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
882 /* Print the fastmap in human-readable form. */
885 print_fastmap (fastmap
)
888 unsigned was_a_range
= 0;
891 while (i
< (1 << BYTEWIDTH
))
897 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
913 /* Print a compiled pattern string in human-readable form, starting at
914 the START pointer into it and ending just before the pointer END. */
917 print_partial_compiled_pattern (start
, end
)
927 fprintf (stderr
, "(null)\n");
931 /* Loop over pattern commands. */
934 fprintf (stderr
, "%d:\t", p
- start
);
936 switch ((re_opcode_t
) *p
++)
939 fprintf (stderr
, "/no_op");
943 fprintf (stderr
, "/succeed");
948 fprintf (stderr
, "/exactn/%d", mcnt
);
951 fprintf (stderr
, "/%c", *p
++);
957 fprintf (stderr
, "/start_memory/%d", *p
++);
961 fprintf (stderr
, "/stop_memory/%d", *p
++);
965 fprintf (stderr
, "/duplicate/%d", *p
++);
969 fprintf (stderr
, "/anychar");
975 register int c
, last
= -100;
976 register int in_range
= 0;
977 int length
= CHARSET_BITMAP_SIZE (p
- 1);
978 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
980 fprintf (stderr
, "/charset [%s",
981 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
984 fprintf (stderr
, " !extends past end of pattern! ");
986 for (c
= 0; c
< 256; c
++)
988 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
990 /* Are we starting a range? */
991 if (last
+ 1 == c
&& ! in_range
)
993 fprintf (stderr
, "-");
996 /* Have we broken a range? */
997 else if (last
+ 1 != c
&& in_range
)
999 fprintf (stderr
, "%c", last
);
1004 fprintf (stderr
, "%c", c
);
1010 fprintf (stderr
, "%c", last
);
1012 fprintf (stderr
, "]");
1016 if (has_range_table
)
1019 fprintf (stderr
, "has-range-table");
1021 /* ??? Should print the range table; for now, just skip it. */
1022 p
+= 2; /* skip range table bits */
1023 EXTRACT_NUMBER_AND_INCR (count
, p
);
1024 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1030 fprintf (stderr
, "/begline");
1034 fprintf (stderr
, "/endline");
1037 case on_failure_jump
:
1038 extract_number_and_incr (&mcnt
, &p
);
1039 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1042 case on_failure_keep_string_jump
:
1043 extract_number_and_incr (&mcnt
, &p
);
1044 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1047 case on_failure_jump_nastyloop
:
1048 extract_number_and_incr (&mcnt
, &p
);
1049 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1052 case on_failure_jump_loop
:
1053 extract_number_and_incr (&mcnt
, &p
);
1054 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1057 case on_failure_jump_smart
:
1058 extract_number_and_incr (&mcnt
, &p
);
1059 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1063 extract_number_and_incr (&mcnt
, &p
);
1064 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1068 extract_number_and_incr (&mcnt
, &p
);
1069 extract_number_and_incr (&mcnt2
, &p
);
1070 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1074 extract_number_and_incr (&mcnt
, &p
);
1075 extract_number_and_incr (&mcnt2
, &p
);
1076 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1080 extract_number_and_incr (&mcnt
, &p
);
1081 extract_number_and_incr (&mcnt2
, &p
);
1082 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1086 fprintf (stderr
, "/wordbound");
1090 fprintf (stderr
, "/notwordbound");
1094 fprintf (stderr
, "/wordbeg");
1098 fprintf (stderr
, "/wordend");
1102 fprintf (stderr
, "/symbeg");
1106 fprintf (stderr
, "/symend");
1110 fprintf (stderr
, "/syntaxspec");
1112 fprintf (stderr
, "/%d", mcnt
);
1116 fprintf (stderr
, "/notsyntaxspec");
1118 fprintf (stderr
, "/%d", mcnt
);
1123 fprintf (stderr
, "/before_dot");
1127 fprintf (stderr
, "/at_dot");
1131 fprintf (stderr
, "/after_dot");
1135 fprintf (stderr
, "/categoryspec");
1137 fprintf (stderr
, "/%d", mcnt
);
1140 case notcategoryspec
:
1141 fprintf (stderr
, "/notcategoryspec");
1143 fprintf (stderr
, "/%d", mcnt
);
1148 fprintf (stderr
, "/begbuf");
1152 fprintf (stderr
, "/endbuf");
1156 fprintf (stderr
, "?%d", *(p
-1));
1159 fprintf (stderr
, "\n");
1162 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1167 print_compiled_pattern (bufp
)
1168 struct re_pattern_buffer
*bufp
;
1170 re_char
*buffer
= bufp
->buffer
;
1172 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1173 printf ("%ld bytes used/%ld bytes allocated.\n",
1174 bufp
->used
, bufp
->allocated
);
1176 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1178 printf ("fastmap: ");
1179 print_fastmap (bufp
->fastmap
);
1182 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1183 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1184 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1185 printf ("no_sub: %d\t", bufp
->no_sub
);
1186 printf ("not_bol: %d\t", bufp
->not_bol
);
1187 printf ("not_eol: %d\t", bufp
->not_eol
);
1188 printf ("syntax: %lx\n", bufp
->syntax
);
1190 /* Perhaps we should print the translate table? */
1195 print_double_string (where
, string1
, size1
, string2
, size2
)
1208 if (FIRST_STRING_P (where
))
1210 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1211 putchar (string1
[this_char
]);
1216 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1217 putchar (string2
[this_char
]);
1221 #else /* not DEBUG */
1226 # define DEBUG_STATEMENT(e)
1227 # define DEBUG_PRINT1(x)
1228 # define DEBUG_PRINT2(x1, x2)
1229 # define DEBUG_PRINT3(x1, x2, x3)
1230 # define DEBUG_PRINT4(x1, x2, x3, x4)
1231 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1232 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1234 #endif /* not DEBUG */
1236 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1237 also be assigned to arbitrarily: each pattern buffer stores its own
1238 syntax, so it can be changed between regex compilations. */
1239 /* This has no initializer because initialized variables in Emacs
1240 become read-only after dumping. */
1241 reg_syntax_t re_syntax_options
;
1244 /* Specify the precise syntax of regexps for compilation. This provides
1245 for compatibility for various utilities which historically have
1246 different, incompatible syntaxes.
1248 The argument SYNTAX is a bit mask comprised of the various bits
1249 defined in regex.h. We return the old syntax. */
1252 re_set_syntax (syntax
)
1253 reg_syntax_t syntax
;
1255 reg_syntax_t ret
= re_syntax_options
;
1257 re_syntax_options
= syntax
;
1260 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1262 /* Regexp to use to replace spaces, or NULL meaning don't. */
1263 static re_char
*whitespace_regexp
;
1266 re_set_whitespace_regexp (regexp
)
1269 whitespace_regexp
= regexp
;
1271 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1273 /* This table gives an error message for each of the error codes listed
1274 in regex.h. Obviously the order here has to be same as there.
1275 POSIX doesn't require that we do anything for REG_NOERROR,
1276 but why not be nice? */
1278 static const char *re_error_msgid
[] =
1280 gettext_noop ("Success"), /* REG_NOERROR */
1281 gettext_noop ("No match"), /* REG_NOMATCH */
1282 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1283 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1284 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1285 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1286 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1287 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1288 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1289 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1290 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1291 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1292 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1293 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1294 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1295 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1296 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1297 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1300 /* Avoiding alloca during matching, to placate r_alloc. */
1302 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1303 searching and matching functions should not call alloca. On some
1304 systems, alloca is implemented in terms of malloc, and if we're
1305 using the relocating allocator routines, then malloc could cause a
1306 relocation, which might (if the strings being searched are in the
1307 ralloc heap) shift the data out from underneath the regexp
1310 Here's another reason to avoid allocation: Emacs
1311 processes input from X in a signal handler; processing X input may
1312 call malloc; if input arrives while a matching routine is calling
1313 malloc, then we're scrod. But Emacs can't just block input while
1314 calling matching routines; then we don't notice interrupts when
1315 they come in. So, Emacs blocks input around all regexp calls
1316 except the matching calls, which it leaves unprotected, in the
1317 faith that they will not malloc. */
1319 /* Normally, this is fine. */
1320 #define MATCH_MAY_ALLOCATE
1322 /* When using GNU C, we are not REALLY using the C alloca, no matter
1323 what config.h may say. So don't take precautions for it. */
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 C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1334 # undef MATCH_MAY_ALLOCATE
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
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. */
1360 size_t re_max_failures
= 40000;
1362 size_t re_max_failures
= 4000;
1365 union fail_stack_elt
1368 /* This should be the biggest `int' that's no bigger than a pointer. */
1372 typedef union fail_stack_elt fail_stack_elt_t
;
1376 fail_stack_elt_t
*stack
;
1378 size_t avail
; /* Offset of next open position. */
1379 size_t frame
; /* Offset of the cur constructed frame. */
1382 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1383 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1386 /* Define macros to initialize and free the failure stack.
1387 Do `return -2' if the alloc fails. */
1389 #ifdef MATCH_MAY_ALLOCATE
1390 # define INIT_FAIL_STACK() \
1392 fail_stack.stack = (fail_stack_elt_t *) \
1393 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1394 * sizeof (fail_stack_elt_t)); \
1396 if (fail_stack.stack == NULL) \
1399 fail_stack.size = INIT_FAILURE_ALLOC; \
1400 fail_stack.avail = 0; \
1401 fail_stack.frame = 0; \
1404 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1406 # define INIT_FAIL_STACK() \
1408 fail_stack.avail = 0; \
1409 fail_stack.frame = 0; \
1412 # define RESET_FAIL_STACK() ((void)0)
1416 /* Double the size of FAIL_STACK, up to a limit
1417 which allows approximately `re_max_failures' items.
1419 Return 1 if succeeds, and 0 if either ran out of memory
1420 allocating space for it or it was already too large.
1422 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1424 /* Factor to increase the failure stack size by
1425 when we increase it.
1426 This used to be 2, but 2 was too wasteful
1427 because the old discarded stacks added up to as much space
1428 were as ultimate, maximum-size stack. */
1429 #define FAIL_STACK_GROWTH_FACTOR 4
1431 #define GROW_FAIL_STACK(fail_stack) \
1432 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1433 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1435 : ((fail_stack).stack \
1436 = (fail_stack_elt_t *) \
1437 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1438 (fail_stack).size * sizeof (fail_stack_elt_t), \
1439 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1440 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1441 * FAIL_STACK_GROWTH_FACTOR))), \
1443 (fail_stack).stack == NULL \
1445 : ((fail_stack).size \
1446 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1447 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1448 * FAIL_STACK_GROWTH_FACTOR)) \
1449 / sizeof (fail_stack_elt_t)), \
1453 /* Push a pointer value onto the failure stack.
1454 Assumes the variable `fail_stack'. Probably should only
1455 be called from within `PUSH_FAILURE_POINT'. */
1456 #define PUSH_FAILURE_POINTER(item) \
1457 fail_stack.stack[fail_stack.avail++].pointer = (item)
1459 /* This pushes an integer-valued item onto the failure stack.
1460 Assumes the variable `fail_stack'. Probably should only
1461 be called from within `PUSH_FAILURE_POINT'. */
1462 #define PUSH_FAILURE_INT(item) \
1463 fail_stack.stack[fail_stack.avail++].integer = (item)
1465 /* Push a fail_stack_elt_t value onto the failure stack.
1466 Assumes the variable `fail_stack'. Probably should only
1467 be called from within `PUSH_FAILURE_POINT'. */
1468 #define PUSH_FAILURE_ELT(item) \
1469 fail_stack.stack[fail_stack.avail++] = (item)
1471 /* These three POP... operations complement the three PUSH... operations.
1472 All assume that `fail_stack' is nonempty. */
1473 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1474 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1475 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1477 /* Individual items aside from the registers. */
1478 #define NUM_NONREG_ITEMS 3
1480 /* Used to examine the stack (to detect infinite loops). */
1481 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1482 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1483 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1484 #define TOP_FAILURE_HANDLE() fail_stack.frame
1487 #define ENSURE_FAIL_STACK(space) \
1488 while (REMAINING_AVAIL_SLOTS <= space) { \
1489 if (!GROW_FAIL_STACK (fail_stack)) \
1491 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1492 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1495 /* Push register NUM onto the stack. */
1496 #define PUSH_FAILURE_REG(num) \
1498 char *destination; \
1499 ENSURE_FAIL_STACK(3); \
1500 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1501 num, regstart[num], regend[num]); \
1502 PUSH_FAILURE_POINTER (regstart[num]); \
1503 PUSH_FAILURE_POINTER (regend[num]); \
1504 PUSH_FAILURE_INT (num); \
1507 /* Change the counter's value to VAL, but make sure that it will
1508 be reset when backtracking. */
1509 #define PUSH_NUMBER(ptr,val) \
1511 char *destination; \
1513 ENSURE_FAIL_STACK(3); \
1514 EXTRACT_NUMBER (c, ptr); \
1515 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1516 PUSH_FAILURE_INT (c); \
1517 PUSH_FAILURE_POINTER (ptr); \
1518 PUSH_FAILURE_INT (-1); \
1519 STORE_NUMBER (ptr, val); \
1522 /* Pop a saved register off the stack. */
1523 #define POP_FAILURE_REG_OR_COUNT() \
1525 int reg = POP_FAILURE_INT (); \
1528 /* It's a counter. */ \
1529 /* Here, we discard `const', making re_match non-reentrant. */ \
1530 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1531 reg = POP_FAILURE_INT (); \
1532 STORE_NUMBER (ptr, reg); \
1533 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1537 regend[reg] = POP_FAILURE_POINTER (); \
1538 regstart[reg] = POP_FAILURE_POINTER (); \
1539 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1540 reg, regstart[reg], regend[reg]); \
1544 /* Check that we are not stuck in an infinite loop. */
1545 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1547 int failure = TOP_FAILURE_HANDLE (); \
1548 /* Check for infinite matching loops */ \
1549 while (failure > 0 \
1550 && (FAILURE_STR (failure) == string_place \
1551 || FAILURE_STR (failure) == NULL)) \
1553 assert (FAILURE_PAT (failure) >= bufp->buffer \
1554 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1555 if (FAILURE_PAT (failure) == pat_cur) \
1560 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1561 failure = NEXT_FAILURE_HANDLE(failure); \
1563 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1566 /* Push the information about the state we will need
1567 if we ever fail back to it.
1569 Requires variables fail_stack, regstart, regend and
1570 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1573 Does `return FAILURE_CODE' if runs out of memory. */
1575 #define PUSH_FAILURE_POINT(pattern, string_place) \
1577 char *destination; \
1578 /* Must be int, so when we don't save any registers, the arithmetic \
1579 of 0 + -1 isn't done as unsigned. */ \
1581 DEBUG_STATEMENT (nfailure_points_pushed++); \
1582 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1583 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1584 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1586 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1588 DEBUG_PRINT1 ("\n"); \
1590 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1591 PUSH_FAILURE_INT (fail_stack.frame); \
1593 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1594 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1595 DEBUG_PRINT1 ("'\n"); \
1596 PUSH_FAILURE_POINTER (string_place); \
1598 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1599 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1600 PUSH_FAILURE_POINTER (pattern); \
1602 /* Close the frame by moving the frame pointer past it. */ \
1603 fail_stack.frame = fail_stack.avail; \
1606 /* Estimate the size of data pushed by a typical failure stack entry.
1607 An estimate is all we need, because all we use this for
1608 is to choose a limit for how big to make the failure stack. */
1609 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1610 #define TYPICAL_FAILURE_SIZE 20
1612 /* How many items can still be added to the stack without overflowing it. */
1613 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1616 /* Pops what PUSH_FAIL_STACK pushes.
1618 We restore into the parameters, all of which should be lvalues:
1619 STR -- the saved data position.
1620 PAT -- the saved pattern position.
1621 REGSTART, REGEND -- arrays of string positions.
1623 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1624 `pend', `string1', `size1', `string2', and `size2'. */
1626 #define POP_FAILURE_POINT(str, pat) \
1628 assert (!FAIL_STACK_EMPTY ()); \
1630 /* Remove failure points and point to how many regs pushed. */ \
1631 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1632 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1633 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1635 /* Pop the saved registers. */ \
1636 while (fail_stack.frame < fail_stack.avail) \
1637 POP_FAILURE_REG_OR_COUNT (); \
1639 pat = POP_FAILURE_POINTER (); \
1640 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1641 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1643 /* If the saved string location is NULL, it came from an \
1644 on_failure_keep_string_jump opcode, and we want to throw away the \
1645 saved NULL, thus retaining our current position in the string. */ \
1646 str = POP_FAILURE_POINTER (); \
1647 DEBUG_PRINT2 (" Popping string %p: `", str); \
1648 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1649 DEBUG_PRINT1 ("'\n"); \
1651 fail_stack.frame = POP_FAILURE_INT (); \
1652 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1654 assert (fail_stack.avail >= 0); \
1655 assert (fail_stack.frame <= fail_stack.avail); \
1657 DEBUG_STATEMENT (nfailure_points_popped++); \
1658 } while (0) /* POP_FAILURE_POINT */
1662 /* Registers are set to a sentinel when they haven't yet matched. */
1663 #define REG_UNSET(e) ((e) == NULL)
1665 /* Subroutine declarations and macros for regex_compile. */
1667 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1668 reg_syntax_t syntax
,
1669 struct re_pattern_buffer
*bufp
));
1670 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1671 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1672 int arg1
, int arg2
));
1673 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1674 int arg
, unsigned char *end
));
1675 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1676 int arg1
, int arg2
, unsigned char *end
));
1677 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1679 reg_syntax_t syntax
));
1680 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1682 reg_syntax_t syntax
));
1683 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1684 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1685 char *fastmap
, const int multibyte
));
1687 /* Fetch the next character in the uncompiled pattern, with no
1689 #define PATFETCH(c) \
1692 if (p == pend) return REG_EEND; \
1693 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1698 /* If `translate' is non-null, return translate[D], else just D. We
1699 cast the subscript to translate because some data is declared as
1700 `char *', to avoid warnings when a string constant is passed. But
1701 when we use a character as a subscript we must make it unsigned. */
1703 # define TRANSLATE(d) \
1704 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1708 /* Macros for outputting the compiled pattern into `buffer'. */
1710 /* If the buffer isn't allocated when it comes in, use this. */
1711 #define INIT_BUF_SIZE 32
1713 /* Make sure we have at least N more bytes of space in buffer. */
1714 #define GET_BUFFER_SPACE(n) \
1715 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1718 /* Make sure we have one more byte of buffer space and then add C to it. */
1719 #define BUF_PUSH(c) \
1721 GET_BUFFER_SPACE (1); \
1722 *b++ = (unsigned char) (c); \
1726 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1727 #define BUF_PUSH_2(c1, c2) \
1729 GET_BUFFER_SPACE (2); \
1730 *b++ = (unsigned char) (c1); \
1731 *b++ = (unsigned char) (c2); \
1735 /* As with BUF_PUSH_2, except for three bytes. */
1736 #define BUF_PUSH_3(c1, c2, c3) \
1738 GET_BUFFER_SPACE (3); \
1739 *b++ = (unsigned char) (c1); \
1740 *b++ = (unsigned char) (c2); \
1741 *b++ = (unsigned char) (c3); \
1745 /* Store a jump with opcode OP at LOC to location TO. We store a
1746 relative address offset by the three bytes the jump itself occupies. */
1747 #define STORE_JUMP(op, loc, to) \
1748 store_op1 (op, loc, (to) - (loc) - 3)
1750 /* Likewise, for a two-argument jump. */
1751 #define STORE_JUMP2(op, loc, to, arg) \
1752 store_op2 (op, loc, (to) - (loc) - 3, arg)
1754 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1755 #define INSERT_JUMP(op, loc, to) \
1756 insert_op1 (op, loc, (to) - (loc) - 3, b)
1758 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1759 #define INSERT_JUMP2(op, loc, to, arg) \
1760 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1763 /* This is not an arbitrary limit: the arguments which represent offsets
1764 into the pattern are two bytes long. So if 2^15 bytes turns out to
1765 be too small, many things would have to change. */
1766 # define MAX_BUF_SIZE (1L << 15)
1768 #if 0 /* This is when we thought it could be 2^16 bytes. */
1769 /* Any other compiler which, like MSC, has allocation limit below 2^16
1770 bytes will have to use approach similar to what was done below for
1771 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1772 reallocating to 0 bytes. Such thing is not going to work too well.
1773 You have been warned!! */
1774 #if defined _MSC_VER && !defined WIN32
1775 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1776 # define MAX_BUF_SIZE 65500L
1778 # define MAX_BUF_SIZE (1L << 16)
1782 /* Extend the buffer by twice its current size via realloc and
1783 reset the pointers that pointed into the old block to point to the
1784 correct places in the new one. If extending the buffer results in it
1785 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1786 #if __BOUNDED_POINTERS__
1787 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1788 # define MOVE_BUFFER_POINTER(P) \
1789 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1790 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1793 SET_HIGH_BOUND (b); \
1794 SET_HIGH_BOUND (begalt); \
1795 if (fixup_alt_jump) \
1796 SET_HIGH_BOUND (fixup_alt_jump); \
1798 SET_HIGH_BOUND (laststart); \
1799 if (pending_exact) \
1800 SET_HIGH_BOUND (pending_exact); \
1803 # define MOVE_BUFFER_POINTER(P) (P) += incr
1804 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1806 #define EXTEND_BUFFER() \
1808 re_char *old_buffer = bufp->buffer; \
1809 if (bufp->allocated == MAX_BUF_SIZE) \
1811 bufp->allocated <<= 1; \
1812 if (bufp->allocated > MAX_BUF_SIZE) \
1813 bufp->allocated = MAX_BUF_SIZE; \
1814 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1815 if (bufp->buffer == NULL) \
1816 return REG_ESPACE; \
1817 /* If the buffer moved, move all the pointers into it. */ \
1818 if (old_buffer != bufp->buffer) \
1820 int incr = bufp->buffer - old_buffer; \
1821 MOVE_BUFFER_POINTER (b); \
1822 MOVE_BUFFER_POINTER (begalt); \
1823 if (fixup_alt_jump) \
1824 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1826 MOVE_BUFFER_POINTER (laststart); \
1827 if (pending_exact) \
1828 MOVE_BUFFER_POINTER (pending_exact); \
1830 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1834 /* Since we have one byte reserved for the register number argument to
1835 {start,stop}_memory, the maximum number of groups we can report
1836 things about is what fits in that byte. */
1837 #define MAX_REGNUM 255
1839 /* But patterns can have more than `MAX_REGNUM' registers. We just
1840 ignore the excess. */
1841 typedef int regnum_t
;
1844 /* Macros for the compile stack. */
1846 /* Since offsets can go either forwards or backwards, this type needs to
1847 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1848 /* int may be not enough when sizeof(int) == 2. */
1849 typedef long pattern_offset_t
;
1853 pattern_offset_t begalt_offset
;
1854 pattern_offset_t fixup_alt_jump
;
1855 pattern_offset_t laststart_offset
;
1857 } compile_stack_elt_t
;
1862 compile_stack_elt_t
*stack
;
1864 unsigned avail
; /* Offset of next open position. */
1865 } compile_stack_type
;
1868 #define INIT_COMPILE_STACK_SIZE 32
1870 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1871 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1873 /* The next available element. */
1874 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1876 /* Explicit quit checking is only used on NTemacs. */
1877 #if defined WINDOWSNT && defined emacs && defined QUIT
1878 extern int immediate_quit
;
1879 # define IMMEDIATE_QUIT_CHECK \
1881 if (immediate_quit) QUIT; \
1884 # define IMMEDIATE_QUIT_CHECK ((void)0)
1887 /* Structure to manage work area for range table. */
1888 struct range_table_work_area
1890 int *table
; /* actual work area. */
1891 int allocated
; /* allocated size for work area in bytes. */
1892 int used
; /* actually used size in words. */
1893 int bits
; /* flag to record character classes */
1896 /* Make sure that WORK_AREA can hold more N multibyte characters.
1897 This is used only in set_image_of_range and set_image_of_range_1.
1898 It expects WORK_AREA to be a pointer.
1899 If it can't get the space, it returns from the surrounding function. */
1901 #define EXTEND_RANGE_TABLE(work_area, n) \
1903 if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \
1905 extend_range_table_work_area (work_area); \
1906 if ((work_area)->table == 0) \
1907 return (REG_ESPACE); \
1911 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1912 (work_area).bits |= (bit)
1914 /* Bits used to implement the multibyte-part of the various character classes
1915 such as [:alnum:] in a charset's range table. */
1916 #define BIT_WORD 0x1
1917 #define BIT_LOWER 0x2
1918 #define BIT_PUNCT 0x4
1919 #define BIT_SPACE 0x8
1920 #define BIT_UPPER 0x10
1921 #define BIT_MULTIBYTE 0x20
1923 /* Set a range START..END to WORK_AREA.
1924 The range is passed through TRANSLATE, so START and END
1925 should be untranslated. */
1926 #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \
1929 tem = set_image_of_range (&work_area, start, end, translate); \
1931 FREE_STACK_RETURN (tem); \
1934 /* Free allocated memory for WORK_AREA. */
1935 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1937 if ((work_area).table) \
1938 free ((work_area).table); \
1941 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1942 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1943 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1944 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1947 /* Set the bit for character C in a list. */
1948 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1951 /* Get the next unsigned number in the uncompiled pattern. */
1952 #define GET_UNSIGNED_NUMBER(num) \
1955 FREE_STACK_RETURN (REG_EBRACE); \
1959 while ('0' <= c && c <= '9') \
1965 num = num * 10 + c - '0'; \
1966 if (num / 10 != prev) \
1967 FREE_STACK_RETURN (REG_BADBR); \
1969 FREE_STACK_RETURN (REG_EBRACE); \
1975 #if ! WIDE_CHAR_SUPPORT
1977 /* Map a string to the char class it names (if any). */
1982 const char *string
= str
;
1983 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
1984 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
1985 else if (STREQ (string
, "word")) return RECC_WORD
;
1986 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
1987 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
1988 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
1989 else if (STREQ (string
, "lower")) return RECC_LOWER
;
1990 else if (STREQ (string
, "print")) return RECC_PRINT
;
1991 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
1992 else if (STREQ (string
, "space")) return RECC_SPACE
;
1993 else if (STREQ (string
, "upper")) return RECC_UPPER
;
1994 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
1995 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
1996 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
1997 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
1998 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
1999 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2003 /* True iff CH is in the char class CC. */
2005 re_iswctype (ch
, cc
)
2011 case RECC_ALNUM
: return ISALNUM (ch
);
2012 case RECC_ALPHA
: return ISALPHA (ch
);
2013 case RECC_BLANK
: return ISBLANK (ch
);
2014 case RECC_CNTRL
: return ISCNTRL (ch
);
2015 case RECC_DIGIT
: return ISDIGIT (ch
);
2016 case RECC_GRAPH
: return ISGRAPH (ch
);
2017 case RECC_LOWER
: return ISLOWER (ch
);
2018 case RECC_PRINT
: return ISPRINT (ch
);
2019 case RECC_PUNCT
: return ISPUNCT (ch
);
2020 case RECC_SPACE
: return ISSPACE (ch
);
2021 case RECC_UPPER
: return ISUPPER (ch
);
2022 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2023 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2024 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2025 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2026 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2027 case RECC_WORD
: return ISWORD (ch
);
2028 case RECC_ERROR
: return false;
2034 /* Return a bit-pattern to use in the range-table bits to match multibyte
2035 chars of class CC. */
2037 re_wctype_to_bit (cc
)
2042 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2043 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2044 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2045 case RECC_LOWER
: return BIT_LOWER
;
2046 case RECC_UPPER
: return BIT_UPPER
;
2047 case RECC_PUNCT
: return BIT_PUNCT
;
2048 case RECC_SPACE
: return BIT_SPACE
;
2049 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2050 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2057 /* Filling in the work area of a range. */
2059 /* Actually extend the space in WORK_AREA. */
2062 extend_range_table_work_area (work_area
)
2063 struct range_table_work_area
*work_area
;
2065 work_area
->allocated
+= 16 * sizeof (int);
2066 if (work_area
->table
)
2068 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2071 = (int *) malloc (work_area
->allocated
);
2076 /* Carefully find the ranges of codes that are equivalent
2077 under case conversion to the range start..end when passed through
2078 TRANSLATE. Handle the case where non-letters can come in between
2079 two upper-case letters (which happens in Latin-1).
2080 Also handle the case of groups of more than 2 case-equivalent chars.
2082 The basic method is to look at consecutive characters and see
2083 if they can form a run that can be handled as one.
2085 Returns -1 if successful, REG_ESPACE if ran out of space. */
2088 set_image_of_range_1 (work_area
, start
, end
, translate
)
2089 RE_TRANSLATE_TYPE translate
;
2090 struct range_table_work_area
*work_area
;
2091 re_wchar_t start
, end
;
2093 /* `one_case' indicates a character, or a run of characters,
2094 each of which is an isolate (no case-equivalents).
2095 This includes all ASCII non-letters.
2097 `two_case' indicates a character, or a run of characters,
2098 each of which has two case-equivalent forms.
2099 This includes all ASCII letters.
2101 `strange' indicates a character that has more than one
2104 enum case_type
{one_case
, two_case
, strange
};
2106 /* Describe the run that is in progress,
2107 which the next character can try to extend.
2108 If run_type is strange, that means there really is no run.
2109 If run_type is one_case, then run_start...run_end is the run.
2110 If run_type is two_case, then the run is run_start...run_end,
2111 and the case-equivalents end at run_eqv_end. */
2113 enum case_type run_type
= strange
;
2114 int run_start
, run_end
, run_eqv_end
;
2116 Lisp_Object eqv_table
;
2118 if (!RE_TRANSLATE_P (translate
))
2120 EXTEND_RANGE_TABLE (work_area
, 2);
2121 work_area
->table
[work_area
->used
++] = (start
);
2122 work_area
->table
[work_area
->used
++] = (end
);
2126 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2128 for (; start
<= end
; start
++)
2130 enum case_type this_type
;
2131 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2132 int minchar
, maxchar
;
2134 /* Classify this character */
2136 this_type
= one_case
;
2137 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2138 this_type
= two_case
;
2140 this_type
= strange
;
2143 minchar
= start
, maxchar
= eqv
;
2145 minchar
= eqv
, maxchar
= start
;
2147 /* Can this character extend the run in progress? */
2148 if (this_type
== strange
|| this_type
!= run_type
2149 || !(minchar
== run_end
+ 1
2150 && (run_type
== two_case
2151 ? maxchar
== run_eqv_end
+ 1 : 1)))
2154 Record each of its equivalent ranges. */
2155 if (run_type
== one_case
)
2157 EXTEND_RANGE_TABLE (work_area
, 2);
2158 work_area
->table
[work_area
->used
++] = run_start
;
2159 work_area
->table
[work_area
->used
++] = run_end
;
2161 else if (run_type
== two_case
)
2163 EXTEND_RANGE_TABLE (work_area
, 4);
2164 work_area
->table
[work_area
->used
++] = run_start
;
2165 work_area
->table
[work_area
->used
++] = run_end
;
2166 work_area
->table
[work_area
->used
++]
2167 = RE_TRANSLATE (eqv_table
, run_start
);
2168 work_area
->table
[work_area
->used
++]
2169 = RE_TRANSLATE (eqv_table
, run_end
);
2174 if (this_type
== strange
)
2176 /* For a strange character, add each of its equivalents, one
2177 by one. Don't start a range. */
2180 EXTEND_RANGE_TABLE (work_area
, 2);
2181 work_area
->table
[work_area
->used
++] = eqv
;
2182 work_area
->table
[work_area
->used
++] = eqv
;
2183 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2185 while (eqv
!= start
);
2188 /* Add this char to the run, or start a new run. */
2189 else if (run_type
== strange
)
2191 /* Initialize a new range. */
2192 run_type
= this_type
;
2195 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2199 /* Extend a running range. */
2201 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2205 /* If a run is still in progress at the end, finish it now
2206 by recording its equivalent ranges. */
2207 if (run_type
== one_case
)
2209 EXTEND_RANGE_TABLE (work_area
, 2);
2210 work_area
->table
[work_area
->used
++] = run_start
;
2211 work_area
->table
[work_area
->used
++] = run_end
;
2213 else if (run_type
== two_case
)
2215 EXTEND_RANGE_TABLE (work_area
, 4);
2216 work_area
->table
[work_area
->used
++] = run_start
;
2217 work_area
->table
[work_area
->used
++] = run_end
;
2218 work_area
->table
[work_area
->used
++]
2219 = RE_TRANSLATE (eqv_table
, run_start
);
2220 work_area
->table
[work_area
->used
++]
2221 = RE_TRANSLATE (eqv_table
, run_end
);
2229 /* Record the the image of the range start..end when passed through
2230 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2231 and is not even necessarily contiguous.
2232 Normally we approximate it with the smallest contiguous range that contains
2233 all the chars we need. However, for Latin-1 we go to extra effort
2236 This function is not called for ASCII ranges.
2238 Returns -1 if successful, REG_ESPACE if ran out of space. */
2241 set_image_of_range (work_area
, start
, end
, translate
)
2242 RE_TRANSLATE_TYPE translate
;
2243 struct range_table_work_area
*work_area
;
2244 re_wchar_t start
, end
;
2246 re_wchar_t cmin
, cmax
;
2249 /* For Latin-1 ranges, use set_image_of_range_1
2250 to get proper handling of ranges that include letters and nonletters.
2251 For a range that includes the whole of Latin-1, this is not necessary.
2252 For other character sets, we don't bother to get this right. */
2253 if (RE_TRANSLATE_P (translate
) && start
< 04400
2254 && !(start
< 04200 && end
>= 04377))
2261 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2271 EXTEND_RANGE_TABLE (work_area
, 2);
2272 work_area
->table
[work_area
->used
++] = (start
);
2273 work_area
->table
[work_area
->used
++] = (end
);
2275 cmin
= -1, cmax
= -1;
2277 if (RE_TRANSLATE_P (translate
))
2281 for (ch
= start
; ch
<= end
; ch
++)
2283 re_wchar_t c
= TRANSLATE (ch
);
2284 if (! (start
<= c
&& c
<= end
))
2290 cmin
= MIN (cmin
, c
);
2291 cmax
= MAX (cmax
, c
);
2298 EXTEND_RANGE_TABLE (work_area
, 2);
2299 work_area
->table
[work_area
->used
++] = (cmin
);
2300 work_area
->table
[work_area
->used
++] = (cmax
);
2307 #ifndef MATCH_MAY_ALLOCATE
2309 /* If we cannot allocate large objects within re_match_2_internal,
2310 we make the fail stack and register vectors global.
2311 The fail stack, we grow to the maximum size when a regexp
2313 The register vectors, we adjust in size each time we
2314 compile a regexp, according to the number of registers it needs. */
2316 static fail_stack_type fail_stack
;
2318 /* Size with which the following vectors are currently allocated.
2319 That is so we can make them bigger as needed,
2320 but never make them smaller. */
2321 static int regs_allocated_size
;
2323 static re_char
** regstart
, ** regend
;
2324 static re_char
**best_regstart
, **best_regend
;
2326 /* Make the register vectors big enough for NUM_REGS registers,
2327 but don't make them smaller. */
2330 regex_grow_registers (num_regs
)
2333 if (num_regs
> regs_allocated_size
)
2335 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2336 RETALLOC_IF (regend
, num_regs
, re_char
*);
2337 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2338 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2340 regs_allocated_size
= num_regs
;
2344 #endif /* not MATCH_MAY_ALLOCATE */
2346 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2350 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2351 Returns one of error codes defined in `regex.h', or zero for success.
2353 Assumes the `allocated' (and perhaps `buffer') and `translate'
2354 fields are set in BUFP on entry.
2356 If it succeeds, results are put in BUFP (if it returns an error, the
2357 contents of BUFP are undefined):
2358 `buffer' is the compiled pattern;
2359 `syntax' is set to SYNTAX;
2360 `used' is set to the length of the compiled pattern;
2361 `fastmap_accurate' is zero;
2362 `re_nsub' is the number of subexpressions in PATTERN;
2363 `not_bol' and `not_eol' are zero;
2365 The `fastmap' field is neither examined nor set. */
2367 /* Insert the `jump' from the end of last alternative to "here".
2368 The space for the jump has already been allocated. */
2369 #define FIXUP_ALT_JUMP() \
2371 if (fixup_alt_jump) \
2372 STORE_JUMP (jump, fixup_alt_jump, b); \
2376 /* Return, freeing storage we allocated. */
2377 #define FREE_STACK_RETURN(value) \
2379 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2380 free (compile_stack.stack); \
2384 static reg_errcode_t
2385 regex_compile (pattern
, size
, syntax
, bufp
)
2388 reg_syntax_t syntax
;
2389 struct re_pattern_buffer
*bufp
;
2391 /* We fetch characters from PATTERN here. */
2392 register re_wchar_t c
, c1
;
2394 /* A random temporary spot in PATTERN. */
2397 /* Points to the end of the buffer, where we should append. */
2398 register unsigned char *b
;
2400 /* Keeps track of unclosed groups. */
2401 compile_stack_type compile_stack
;
2403 /* Points to the current (ending) position in the pattern. */
2405 /* `const' makes AIX compiler fail. */
2406 unsigned char *p
= pattern
;
2408 re_char
*p
= pattern
;
2410 re_char
*pend
= pattern
+ size
;
2412 /* How to translate the characters in the pattern. */
2413 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2415 /* Address of the count-byte of the most recently inserted `exactn'
2416 command. This makes it possible to tell if a new exact-match
2417 character can be added to that command or if the character requires
2418 a new `exactn' command. */
2419 unsigned char *pending_exact
= 0;
2421 /* Address of start of the most recently finished expression.
2422 This tells, e.g., postfix * where to find the start of its
2423 operand. Reset at the beginning of groups and alternatives. */
2424 unsigned char *laststart
= 0;
2426 /* Address of beginning of regexp, or inside of last group. */
2427 unsigned char *begalt
;
2429 /* Place in the uncompiled pattern (i.e., the {) to
2430 which to go back if the interval is invalid. */
2431 re_char
*beg_interval
;
2433 /* Address of the place where a forward jump should go to the end of
2434 the containing expression. Each alternative of an `or' -- except the
2435 last -- ends with a forward jump of this sort. */
2436 unsigned char *fixup_alt_jump
= 0;
2438 /* Counts open-groups as they are encountered. Remembered for the
2439 matching close-group on the compile stack, so the same register
2440 number is put in the stop_memory as the start_memory. */
2441 regnum_t regnum
= 0;
2443 /* Work area for range table of charset. */
2444 struct range_table_work_area range_table_work
;
2446 /* If the object matched can contain multibyte characters. */
2447 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2449 /* Nonzero if we have pushed down into a subpattern. */
2450 int in_subpattern
= 0;
2452 /* These hold the values of p, pattern, and pend from the main
2453 pattern when we have pushed into a subpattern. */
2455 re_char
*main_pattern
;
2460 DEBUG_PRINT1 ("\nCompiling pattern: ");
2463 unsigned debug_count
;
2465 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2466 putchar (pattern
[debug_count
]);
2471 /* Initialize the compile stack. */
2472 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2473 if (compile_stack
.stack
== NULL
)
2476 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2477 compile_stack
.avail
= 0;
2479 range_table_work
.table
= 0;
2480 range_table_work
.allocated
= 0;
2482 /* Initialize the pattern buffer. */
2483 bufp
->syntax
= syntax
;
2484 bufp
->fastmap_accurate
= 0;
2485 bufp
->not_bol
= bufp
->not_eol
= 0;
2487 /* Set `used' to zero, so that if we return an error, the pattern
2488 printer (for debugging) will think there's no pattern. We reset it
2492 /* Always count groups, whether or not bufp->no_sub is set. */
2495 #if !defined emacs && !defined SYNTAX_TABLE
2496 /* Initialize the syntax table. */
2497 init_syntax_once ();
2500 if (bufp
->allocated
== 0)
2503 { /* If zero allocated, but buffer is non-null, try to realloc
2504 enough space. This loses if buffer's address is bogus, but
2505 that is the user's responsibility. */
2506 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2509 { /* Caller did not allocate a buffer. Do it for them. */
2510 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2512 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2514 bufp
->allocated
= INIT_BUF_SIZE
;
2517 begalt
= b
= bufp
->buffer
;
2519 /* Loop through the uncompiled pattern until we're at the end. */
2524 /* If this is the end of an included regexp,
2525 pop back to the main regexp and try again. */
2529 pattern
= main_pattern
;
2534 /* If this is the end of the main regexp, we are done. */
2546 /* If there's no special whitespace regexp, treat
2547 spaces normally. And don't try to do this recursively. */
2548 if (!whitespace_regexp
|| in_subpattern
)
2551 /* Peek past following spaces. */
2558 /* If the spaces are followed by a repetition op,
2559 treat them normally. */
2561 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2562 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2565 /* Replace the spaces with the whitespace regexp. */
2569 main_pattern
= pattern
;
2570 p
= pattern
= whitespace_regexp
;
2571 pend
= p
+ strlen (p
);
2577 if ( /* If at start of pattern, it's an operator. */
2579 /* If context independent, it's an operator. */
2580 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2581 /* Otherwise, depends on what's come before. */
2582 || at_begline_loc_p (pattern
, p
, syntax
))
2583 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2592 if ( /* If at end of pattern, it's an operator. */
2594 /* If context independent, it's an operator. */
2595 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2596 /* Otherwise, depends on what's next. */
2597 || at_endline_loc_p (p
, pend
, syntax
))
2598 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2607 if ((syntax
& RE_BK_PLUS_QM
)
2608 || (syntax
& RE_LIMITED_OPS
))
2612 /* If there is no previous pattern... */
2615 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2616 FREE_STACK_RETURN (REG_BADRPT
);
2617 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2622 /* 1 means zero (many) matches is allowed. */
2623 boolean zero_times_ok
= 0, many_times_ok
= 0;
2626 /* If there is a sequence of repetition chars, collapse it
2627 down to just one (the right one). We can't combine
2628 interval operators with these because of, e.g., `a{2}*',
2629 which should only match an even number of `a's. */
2633 if ((syntax
& RE_FRUGAL
)
2634 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2638 zero_times_ok
|= c
!= '+';
2639 many_times_ok
|= c
!= '?';
2645 || (!(syntax
& RE_BK_PLUS_QM
)
2646 && (*p
== '+' || *p
== '?')))
2648 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2651 FREE_STACK_RETURN (REG_EESCAPE
);
2652 if (p
[1] == '+' || p
[1] == '?')
2653 PATFETCH (c
); /* Gobble up the backslash. */
2659 /* If we get here, we found another repeat character. */
2663 /* Star, etc. applied to an empty pattern is equivalent
2664 to an empty pattern. */
2665 if (!laststart
|| laststart
== b
)
2668 /* Now we know whether or not zero matches is allowed
2669 and also whether or not two or more matches is allowed. */
2674 boolean simple
= skip_one_char (laststart
) == b
;
2675 unsigned int startoffset
= 0;
2677 /* Check if the loop can match the empty string. */
2678 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2679 ? on_failure_jump
: on_failure_jump_loop
;
2680 assert (skip_one_char (laststart
) <= b
);
2682 if (!zero_times_ok
&& simple
)
2683 { /* Since simple * loops can be made faster by using
2684 on_failure_keep_string_jump, we turn simple P+
2685 into PP* if P is simple. */
2686 unsigned char *p1
, *p2
;
2687 startoffset
= b
- laststart
;
2688 GET_BUFFER_SPACE (startoffset
);
2689 p1
= b
; p2
= laststart
;
2695 GET_BUFFER_SPACE (6);
2698 STORE_JUMP (ofj
, b
, b
+ 6);
2700 /* Simple * loops can use on_failure_keep_string_jump
2701 depending on what follows. But since we don't know
2702 that yet, we leave the decision up to
2703 on_failure_jump_smart. */
2704 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2705 laststart
+ startoffset
, b
+ 6);
2707 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2712 /* A simple ? pattern. */
2713 assert (zero_times_ok
);
2714 GET_BUFFER_SPACE (3);
2715 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2719 else /* not greedy */
2720 { /* I wish the greedy and non-greedy cases could be merged. */
2722 GET_BUFFER_SPACE (7); /* We might use less. */
2725 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2727 /* The non-greedy multiple match looks like
2728 a repeat..until: we only need a conditional jump
2729 at the end of the loop. */
2730 if (emptyp
) BUF_PUSH (no_op
);
2731 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2732 : on_failure_jump
, b
, laststart
);
2736 /* The repeat...until naturally matches one or more.
2737 To also match zero times, we need to first jump to
2738 the end of the loop (its conditional jump). */
2739 INSERT_JUMP (jump
, laststart
, b
);
2745 /* non-greedy a?? */
2746 INSERT_JUMP (jump
, laststart
, b
+ 3);
2748 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2765 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2767 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2769 /* Ensure that we have enough space to push a charset: the
2770 opcode, the length count, and the bitset; 34 bytes in all. */
2771 GET_BUFFER_SPACE (34);
2775 /* We test `*p == '^' twice, instead of using an if
2776 statement, so we only need one BUF_PUSH. */
2777 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2781 /* Remember the first position in the bracket expression. */
2784 /* Push the number of bytes in the bitmap. */
2785 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2787 /* Clear the whole map. */
2788 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2790 /* charset_not matches newline according to a syntax bit. */
2791 if ((re_opcode_t
) b
[-2] == charset_not
2792 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2793 SET_LIST_BIT ('\n');
2795 /* Read in characters and ranges, setting map bits. */
2798 boolean escaped_char
= false;
2799 const unsigned char *p2
= p
;
2801 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2803 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2804 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2805 So the translation is done later in a loop. Example:
2806 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2809 /* \ might escape characters inside [...] and [^...]. */
2810 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2812 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2815 escaped_char
= true;
2819 /* Could be the end of the bracket expression. If it's
2820 not (i.e., when the bracket expression is `[]' so
2821 far), the ']' character bit gets set way below. */
2822 if (c
== ']' && p2
!= p1
)
2826 /* What should we do for the character which is
2827 greater than 0x7F, but not BASE_LEADING_CODE_P?
2830 /* See if we're at the beginning of a possible character
2833 if (!escaped_char
&&
2834 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2836 /* Leave room for the null. */
2837 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2838 const unsigned char *class_beg
;
2844 /* If pattern is `[[:'. */
2845 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2850 if ((c
== ':' && *p
== ']') || p
== pend
)
2852 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2855 /* This is in any case an invalid class name. */
2860 /* If isn't a word bracketed by `[:' and `:]':
2861 undo the ending character, the letters, and
2862 leave the leading `:' and `[' (but set bits for
2864 if (c
== ':' && *p
== ']')
2869 cc
= re_wctype (str
);
2872 FREE_STACK_RETURN (REG_ECTYPE
);
2874 /* Throw away the ] at the end of the character
2878 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2880 /* Most character classes in a multibyte match
2881 just set a flag. Exceptions are is_blank,
2882 is_digit, is_cntrl, and is_xdigit, since
2883 they can only match ASCII characters. We
2884 don't need to handle them for multibyte.
2885 They are distinguished by a negative wctype. */
2888 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work
,
2889 re_wctype_to_bit (cc
));
2891 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2893 int translated
= TRANSLATE (ch
);
2894 if (re_iswctype (btowc (ch
), cc
))
2895 SET_LIST_BIT (translated
);
2898 /* Repeat the loop. */
2903 /* Go back to right after the "[:". */
2907 /* Because the `:' may starts the range, we
2908 can't simply set bit and repeat the loop.
2909 Instead, just set it to C and handle below. */
2914 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2917 /* Discard the `-'. */
2920 /* Fetch the character which ends the range. */
2923 if (SINGLE_BYTE_CHAR_P (c
))
2925 if (! SINGLE_BYTE_CHAR_P (c1
))
2927 /* Handle a range starting with a
2928 character of less than 256, and ending
2929 with a character of not less than 256.
2930 Split that into two ranges, the low one
2931 ending at 0377, and the high one
2932 starting at the smallest character in
2933 the charset of C1 and ending at C1. */
2934 int charset
= CHAR_CHARSET (c1
);
2935 re_wchar_t c2
= MAKE_CHAR (charset
, 0, 0);
2937 SET_RANGE_TABLE_WORK_AREA (range_table_work
,
2942 else if (!SAME_CHARSET_P (c
, c1
))
2943 FREE_STACK_RETURN (REG_ERANGEX
);
2946 /* Range from C to C. */
2949 /* Set the range ... */
2950 if (SINGLE_BYTE_CHAR_P (c
))
2951 /* ... into bitmap. */
2953 re_wchar_t this_char
;
2954 re_wchar_t range_start
= c
, range_end
= c1
;
2956 /* If the start is after the end, the range is empty. */
2957 if (range_start
> range_end
)
2959 if (syntax
& RE_NO_EMPTY_RANGES
)
2960 FREE_STACK_RETURN (REG_ERANGE
);
2961 /* Else, repeat the loop. */
2965 for (this_char
= range_start
; this_char
<= range_end
;
2967 SET_LIST_BIT (TRANSLATE (this_char
));
2971 /* ... into range table. */
2972 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
2975 /* Discard any (non)matching list bytes that are all 0 at the
2976 end of the map. Decrease the map-length byte too. */
2977 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2981 /* Build real range table from work area. */
2982 if (RANGE_TABLE_WORK_USED (range_table_work
)
2983 || RANGE_TABLE_WORK_BITS (range_table_work
))
2986 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2988 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2989 bytes for flags, two for COUNT, and three bytes for
2991 GET_BUFFER_SPACE (4 + used
* 3);
2993 /* Indicate the existence of range table. */
2994 laststart
[1] |= 0x80;
2996 /* Store the character class flag bits into the range table.
2997 If not in emacs, these flag bits are always 0. */
2998 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
2999 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3001 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3002 for (i
= 0; i
< used
; i
++)
3003 STORE_CHARACTER_AND_INCR
3004 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3011 if (syntax
& RE_NO_BK_PARENS
)
3018 if (syntax
& RE_NO_BK_PARENS
)
3025 if (syntax
& RE_NEWLINE_ALT
)
3032 if (syntax
& RE_NO_BK_VBAR
)
3039 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3040 goto handle_interval
;
3046 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3048 /* Do not translate the character after the \, so that we can
3049 distinguish, e.g., \B from \b, even if we normally would
3050 translate, e.g., B to b. */
3056 if (syntax
& RE_NO_BK_PARENS
)
3057 goto normal_backslash
;
3064 /* Look for a special (?...) construct */
3065 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3067 PATFETCH (c
); /* Gobble up the '?'. */
3071 case ':': shy
= 1; break;
3073 /* Only (?:...) is supported right now. */
3074 FREE_STACK_RETURN (REG_BADPAT
);
3085 if (COMPILE_STACK_FULL
)
3087 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3088 compile_stack_elt_t
);
3089 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3091 compile_stack
.size
<<= 1;
3094 /* These are the values to restore when we hit end of this
3095 group. They are all relative offsets, so that if the
3096 whole pattern moves because of realloc, they will still
3098 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3099 COMPILE_STACK_TOP
.fixup_alt_jump
3100 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3101 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3102 COMPILE_STACK_TOP
.regnum
= shy
? -regnum
: regnum
;
3105 start_memory for groups beyond the last one we can
3106 represent in the compiled pattern. */
3107 if (regnum
<= MAX_REGNUM
&& !shy
)
3108 BUF_PUSH_2 (start_memory
, regnum
);
3110 compile_stack
.avail
++;
3115 /* If we've reached MAX_REGNUM groups, then this open
3116 won't actually generate any code, so we'll have to
3117 clear pending_exact explicitly. */
3123 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3125 if (COMPILE_STACK_EMPTY
)
3127 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3128 goto normal_backslash
;
3130 FREE_STACK_RETURN (REG_ERPAREN
);
3136 /* See similar code for backslashed left paren above. */
3137 if (COMPILE_STACK_EMPTY
)
3139 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3142 FREE_STACK_RETURN (REG_ERPAREN
);
3145 /* Since we just checked for an empty stack above, this
3146 ``can't happen''. */
3147 assert (compile_stack
.avail
!= 0);
3149 /* We don't just want to restore into `regnum', because
3150 later groups should continue to be numbered higher,
3151 as in `(ab)c(de)' -- the second group is #2. */
3152 regnum_t this_group_regnum
;
3154 compile_stack
.avail
--;
3155 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3157 = COMPILE_STACK_TOP
.fixup_alt_jump
3158 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3160 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3161 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3162 /* If we've reached MAX_REGNUM groups, then this open
3163 won't actually generate any code, so we'll have to
3164 clear pending_exact explicitly. */
3167 /* We're at the end of the group, so now we know how many
3168 groups were inside this one. */
3169 if (this_group_regnum
<= MAX_REGNUM
&& this_group_regnum
> 0)
3170 BUF_PUSH_2 (stop_memory
, this_group_regnum
);
3175 case '|': /* `\|'. */
3176 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3177 goto normal_backslash
;
3179 if (syntax
& RE_LIMITED_OPS
)
3182 /* Insert before the previous alternative a jump which
3183 jumps to this alternative if the former fails. */
3184 GET_BUFFER_SPACE (3);
3185 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3189 /* The alternative before this one has a jump after it
3190 which gets executed if it gets matched. Adjust that
3191 jump so it will jump to this alternative's analogous
3192 jump (put in below, which in turn will jump to the next
3193 (if any) alternative's such jump, etc.). The last such
3194 jump jumps to the correct final destination. A picture:
3200 If we are at `b', then fixup_alt_jump right now points to a
3201 three-byte space after `a'. We'll put in the jump, set
3202 fixup_alt_jump to right after `b', and leave behind three
3203 bytes which we'll fill in when we get to after `c'. */
3207 /* Mark and leave space for a jump after this alternative,
3208 to be filled in later either by next alternative or
3209 when know we're at the end of a series of alternatives. */
3211 GET_BUFFER_SPACE (3);
3220 /* If \{ is a literal. */
3221 if (!(syntax
& RE_INTERVALS
)
3222 /* If we're at `\{' and it's not the open-interval
3224 || (syntax
& RE_NO_BK_BRACES
))
3225 goto normal_backslash
;
3229 /* If got here, then the syntax allows intervals. */
3231 /* At least (most) this many matches must be made. */
3232 int lower_bound
= 0, upper_bound
= -1;
3236 GET_UNSIGNED_NUMBER (lower_bound
);
3239 GET_UNSIGNED_NUMBER (upper_bound
);
3241 /* Interval such as `{1}' => match exactly once. */
3242 upper_bound
= lower_bound
;
3244 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3245 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3246 FREE_STACK_RETURN (REG_BADBR
);
3248 if (!(syntax
& RE_NO_BK_BRACES
))
3251 FREE_STACK_RETURN (REG_BADBR
);
3253 FREE_STACK_RETURN (REG_EESCAPE
);
3258 FREE_STACK_RETURN (REG_BADBR
);
3260 /* We just parsed a valid interval. */
3262 /* If it's invalid to have no preceding re. */
3265 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3266 FREE_STACK_RETURN (REG_BADRPT
);
3267 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3270 goto unfetch_interval
;
3273 if (upper_bound
== 0)
3274 /* If the upper bound is zero, just drop the sub pattern
3277 else if (lower_bound
== 1 && upper_bound
== 1)
3278 /* Just match it once: nothing to do here. */
3281 /* Otherwise, we have a nontrivial interval. When
3282 we're all done, the pattern will look like:
3283 set_number_at <jump count> <upper bound>
3284 set_number_at <succeed_n count> <lower bound>
3285 succeed_n <after jump addr> <succeed_n count>
3287 jump_n <succeed_n addr> <jump count>
3288 (The upper bound and `jump_n' are omitted if
3289 `upper_bound' is 1, though.) */
3291 { /* If the upper bound is > 1, we need to insert
3292 more at the end of the loop. */
3293 unsigned int nbytes
= (upper_bound
< 0 ? 3
3294 : upper_bound
> 1 ? 5 : 0);
3295 unsigned int startoffset
= 0;
3297 GET_BUFFER_SPACE (20); /* We might use less. */
3299 if (lower_bound
== 0)
3301 /* A succeed_n that starts with 0 is really a
3302 a simple on_failure_jump_loop. */
3303 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3309 /* Initialize lower bound of the `succeed_n', even
3310 though it will be set during matching by its
3311 attendant `set_number_at' (inserted next),
3312 because `re_compile_fastmap' needs to know.
3313 Jump to the `jump_n' we might insert below. */
3314 INSERT_JUMP2 (succeed_n
, laststart
,
3319 /* Code to initialize the lower bound. Insert
3320 before the `succeed_n'. The `5' is the last two
3321 bytes of this `set_number_at', plus 3 bytes of
3322 the following `succeed_n'. */
3323 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3328 if (upper_bound
< 0)
3330 /* A negative upper bound stands for infinity,
3331 in which case it degenerates to a plain jump. */
3332 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3335 else if (upper_bound
> 1)
3336 { /* More than one repetition is allowed, so
3337 append a backward jump to the `succeed_n'
3338 that starts this interval.
3340 When we've reached this during matching,
3341 we'll have matched the interval once, so
3342 jump back only `upper_bound - 1' times. */
3343 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3347 /* The location we want to set is the second
3348 parameter of the `jump_n'; that is `b-2' as
3349 an absolute address. `laststart' will be
3350 the `set_number_at' we're about to insert;
3351 `laststart+3' the number to set, the source
3352 for the relative address. But we are
3353 inserting into the middle of the pattern --
3354 so everything is getting moved up by 5.
3355 Conclusion: (b - 2) - (laststart + 3) + 5,
3356 i.e., b - laststart.
3358 We insert this at the beginning of the loop
3359 so that if we fail during matching, we'll
3360 reinitialize the bounds. */
3361 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3362 upper_bound
- 1, b
);
3367 beg_interval
= NULL
;
3372 /* If an invalid interval, match the characters as literals. */
3373 assert (beg_interval
);
3375 beg_interval
= NULL
;
3377 /* normal_char and normal_backslash need `c'. */
3380 if (!(syntax
& RE_NO_BK_BRACES
))
3382 assert (p
> pattern
&& p
[-1] == '\\');
3383 goto normal_backslash
;
3389 /* There is no way to specify the before_dot and after_dot
3390 operators. rms says this is ok. --karl */
3398 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3404 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3410 BUF_PUSH_2 (categoryspec
, c
);
3416 BUF_PUSH_2 (notcategoryspec
, c
);
3422 if (syntax
& RE_NO_GNU_OPS
)
3425 BUF_PUSH_2 (syntaxspec
, Sword
);
3430 if (syntax
& RE_NO_GNU_OPS
)
3433 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3438 if (syntax
& RE_NO_GNU_OPS
)
3444 if (syntax
& RE_NO_GNU_OPS
)
3450 if (syntax
& RE_NO_GNU_OPS
)
3459 FREE_STACK_RETURN (REG_BADPAT
);
3463 if (syntax
& RE_NO_GNU_OPS
)
3465 BUF_PUSH (wordbound
);
3469 if (syntax
& RE_NO_GNU_OPS
)
3471 BUF_PUSH (notwordbound
);
3475 if (syntax
& RE_NO_GNU_OPS
)
3481 if (syntax
& RE_NO_GNU_OPS
)
3486 case '1': case '2': case '3': case '4': case '5':
3487 case '6': case '7': case '8': case '9':
3491 if (syntax
& RE_NO_BK_REFS
)
3492 goto normal_backslash
;
3496 /* Can't back reference to a subexpression before its end. */
3497 if (reg
> regnum
|| group_in_compile_stack (compile_stack
, reg
))
3498 FREE_STACK_RETURN (REG_ESUBREG
);
3501 BUF_PUSH_2 (duplicate
, reg
);
3508 if (syntax
& RE_BK_PLUS_QM
)
3511 goto normal_backslash
;
3515 /* You might think it would be useful for \ to mean
3516 not to translate; but if we don't translate it
3517 it will never match anything. */
3524 /* Expects the character in `c'. */
3526 /* If no exactn currently being built. */
3529 /* If last exactn not at current position. */
3530 || pending_exact
+ *pending_exact
+ 1 != b
3532 /* We have only one byte following the exactn for the count. */
3533 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3535 /* If followed by a repetition operator. */
3536 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3537 || ((syntax
& RE_BK_PLUS_QM
)
3538 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3539 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3540 || ((syntax
& RE_INTERVALS
)
3541 && ((syntax
& RE_NO_BK_BRACES
)
3542 ? p
!= pend
&& *p
== '{'
3543 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3545 /* Start building a new exactn. */
3549 BUF_PUSH_2 (exactn
, 0);
3550 pending_exact
= b
- 1;
3553 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3559 len
= CHAR_STRING (c
, b
);
3563 (*pending_exact
) += len
;
3568 } /* while p != pend */
3571 /* Through the pattern now. */
3575 if (!COMPILE_STACK_EMPTY
)
3576 FREE_STACK_RETURN (REG_EPAREN
);
3578 /* If we don't want backtracking, force success
3579 the first time we reach the end of the compiled pattern. */
3580 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3583 /* We have succeeded; set the length of the buffer. */
3584 bufp
->used
= b
- bufp
->buffer
;
3589 re_compile_fastmap (bufp
);
3590 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3591 print_compiled_pattern (bufp
);
3596 #ifndef MATCH_MAY_ALLOCATE
3597 /* Initialize the failure stack to the largest possible stack. This
3598 isn't necessary unless we're trying to avoid calling alloca in
3599 the search and match routines. */
3601 int num_regs
= bufp
->re_nsub
+ 1;
3603 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3605 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3607 if (! fail_stack
.stack
)
3609 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3610 * sizeof (fail_stack_elt_t
));
3613 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3615 * sizeof (fail_stack_elt_t
)));
3618 regex_grow_registers (num_regs
);
3620 #endif /* not MATCH_MAY_ALLOCATE */
3622 FREE_STACK_RETURN (REG_NOERROR
);
3623 } /* regex_compile */
3625 /* Subroutines for `regex_compile'. */
3627 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3630 store_op1 (op
, loc
, arg
)
3635 *loc
= (unsigned char) op
;
3636 STORE_NUMBER (loc
+ 1, arg
);
3640 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3643 store_op2 (op
, loc
, arg1
, arg2
)
3648 *loc
= (unsigned char) op
;
3649 STORE_NUMBER (loc
+ 1, arg1
);
3650 STORE_NUMBER (loc
+ 3, arg2
);
3654 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3655 for OP followed by two-byte integer parameter ARG. */
3658 insert_op1 (op
, loc
, arg
, end
)
3664 register unsigned char *pfrom
= end
;
3665 register unsigned char *pto
= end
+ 3;
3667 while (pfrom
!= loc
)
3670 store_op1 (op
, loc
, arg
);
3674 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3677 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3683 register unsigned char *pfrom
= end
;
3684 register unsigned char *pto
= end
+ 5;
3686 while (pfrom
!= loc
)
3689 store_op2 (op
, loc
, arg1
, arg2
);
3693 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3694 after an alternative or a begin-subexpression. We assume there is at
3695 least one character before the ^. */
3698 at_begline_loc_p (pattern
, p
, syntax
)
3699 re_char
*pattern
, *p
;
3700 reg_syntax_t syntax
;
3702 re_char
*prev
= p
- 2;
3703 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3706 /* After a subexpression? */
3707 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3708 /* After an alternative? */
3709 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3710 /* After a shy subexpression? */
3711 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3712 && prev
[-1] == '?' && prev
[-2] == '('
3713 && (syntax
& RE_NO_BK_PARENS
3714 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3718 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3719 at least one character after the $, i.e., `P < PEND'. */
3722 at_endline_loc_p (p
, pend
, syntax
)
3724 reg_syntax_t syntax
;
3727 boolean next_backslash
= *next
== '\\';
3728 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3731 /* Before a subexpression? */
3732 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3733 : next_backslash
&& next_next
&& *next_next
== ')')
3734 /* Before an alternative? */
3735 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3736 : next_backslash
&& next_next
&& *next_next
== '|');
3740 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3741 false if it's not. */
3744 group_in_compile_stack (compile_stack
, regnum
)
3745 compile_stack_type compile_stack
;
3750 for (this_element
= compile_stack
.avail
- 1;
3753 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3760 If fastmap is non-NULL, go through the pattern and fill fastmap
3761 with all the possible leading chars. If fastmap is NULL, don't
3762 bother filling it up (obviously) and only return whether the
3763 pattern could potentially match the empty string.
3765 Return 1 if p..pend might match the empty string.
3766 Return 0 if p..pend matches at least one char.
3767 Return -1 if fastmap was not updated accurately. */
3770 analyse_first (p
, pend
, fastmap
, multibyte
)
3773 const int multibyte
;
3778 /* If all elements for base leading-codes in fastmap is set, this
3779 flag is set true. */
3780 boolean match_any_multibyte_characters
= false;
3784 /* The loop below works as follows:
3785 - It has a working-list kept in the PATTERN_STACK and which basically
3786 starts by only containing a pointer to the first operation.
3787 - If the opcode we're looking at is a match against some set of
3788 chars, then we add those chars to the fastmap and go on to the
3789 next work element from the worklist (done via `break').
3790 - If the opcode is a control operator on the other hand, we either
3791 ignore it (if it's meaningless at this point, such as `start_memory')
3792 or execute it (if it's a jump). If the jump has several destinations
3793 (i.e. `on_failure_jump'), then we push the other destination onto the
3795 We guarantee termination by ignoring backward jumps (more or less),
3796 so that `p' is monotonically increasing. More to the point, we
3797 never set `p' (or push) anything `<= p1'. */
3801 /* `p1' is used as a marker of how far back a `on_failure_jump'
3802 can go without being ignored. It is normally equal to `p'
3803 (which prevents any backward `on_failure_jump') except right
3804 after a plain `jump', to allow patterns such as:
3807 10: on_failure_jump 3
3808 as used for the *? operator. */
3811 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3818 /* If the first character has to match a backreference, that means
3819 that the group was empty (since it already matched). Since this
3820 is the only case that interests us here, we can assume that the
3821 backreference must match the empty string. */
3826 /* Following are the cases which match a character. These end
3832 int c
= RE_STRING_CHAR (p
+ 1, pend
- p
);
3834 if (SINGLE_BYTE_CHAR_P (c
))
3843 /* We could put all the chars except for \n (and maybe \0)
3844 but we don't bother since it is generally not worth it. */
3845 if (!fastmap
) break;
3850 /* Chars beyond end of bitmap are possible matches.
3851 All the single-byte codes can occur in multibyte buffers.
3852 So any that are not listed in the charset
3853 are possible matches, even in multibyte buffers. */
3854 if (!fastmap
) break;
3855 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3856 j
< (1 << BYTEWIDTH
); j
++)
3860 if (!fastmap
) break;
3861 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3862 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3864 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3867 if ((not && multibyte
)
3868 /* Any character set can possibly contain a character
3869 which doesn't match the specified set of characters. */
3870 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3871 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3872 /* If we can match a character class, we can match
3873 any character set. */
3875 set_fastmap_for_multibyte_characters
:
3876 if (match_any_multibyte_characters
== false)
3878 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3879 if (BASE_LEADING_CODE_P (j
))
3881 match_any_multibyte_characters
= true;
3885 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3886 && match_any_multibyte_characters
== false)
3888 /* Set fastmap[I] 1 where I is a base leading code of each
3889 multibyte character in the range table. */
3892 /* Make P points the range table. `+ 2' is to skip flag
3893 bits for a character class. */
3894 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3896 /* Extract the number of ranges in range table into COUNT. */
3897 EXTRACT_NUMBER_AND_INCR (count
, p
);
3898 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3900 /* Extract the start of each range. */
3901 EXTRACT_CHARACTER (c
, p
);
3902 j
= CHAR_CHARSET (c
);
3903 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
3910 if (!fastmap
) break;
3912 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3914 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3915 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3919 /* This match depends on text properties. These end with
3920 aborting optimizations. */
3924 case notcategoryspec
:
3925 if (!fastmap
) break;
3926 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3928 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3929 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3933 /* Any character set can possibly contain a character
3934 whose category is K (or not). */
3935 goto set_fastmap_for_multibyte_characters
;
3938 /* All cases after this match the empty string. These end with
3960 EXTRACT_NUMBER_AND_INCR (j
, p
);
3962 /* Backward jumps can only go back to code that we've already
3963 visited. `re_compile' should make sure this is true. */
3966 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
3968 case on_failure_jump
:
3969 case on_failure_keep_string_jump
:
3970 case on_failure_jump_loop
:
3971 case on_failure_jump_nastyloop
:
3972 case on_failure_jump_smart
:
3978 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
3979 to jump back to "just after here". */
3982 case on_failure_jump
:
3983 case on_failure_keep_string_jump
:
3984 case on_failure_jump_nastyloop
:
3985 case on_failure_jump_loop
:
3986 case on_failure_jump_smart
:
3987 EXTRACT_NUMBER_AND_INCR (j
, p
);
3989 ; /* Backward jump to be ignored. */
3991 { /* We have to look down both arms.
3992 We first go down the "straight" path so as to minimize
3993 stack usage when going through alternatives. */
3994 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4002 /* This code simply does not properly handle forward jump_n. */
4003 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4005 /* jump_n can either jump or fall through. The (backward) jump
4006 case has already been handled, so we only need to look at the
4007 fallthrough case. */
4011 /* If N == 0, it should be an on_failure_jump_loop instead. */
4012 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4014 /* We only care about one iteration of the loop, so we don't
4015 need to consider the case where this behaves like an
4032 abort (); /* We have listed all the cases. */
4035 /* Getting here means we have found the possible starting
4036 characters for one path of the pattern -- and that the empty
4037 string does not match. We need not follow this path further. */
4041 /* We reached the end without matching anything. */
4044 } /* analyse_first */
4046 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4047 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4048 characters can start a string that matches the pattern. This fastmap
4049 is used by re_search to skip quickly over impossible starting points.
4051 Character codes above (1 << BYTEWIDTH) are not represented in the
4052 fastmap, but the leading codes are represented. Thus, the fastmap
4053 indicates which character sets could start a match.
4055 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4056 area as BUFP->fastmap.
4058 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4061 Returns 0 if we succeed, -2 if an internal error. */
4064 re_compile_fastmap (bufp
)
4065 struct re_pattern_buffer
*bufp
;
4067 char *fastmap
= bufp
->fastmap
;
4070 assert (fastmap
&& bufp
->buffer
);
4072 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4073 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4075 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4076 fastmap
, RE_MULTIBYTE_P (bufp
));
4077 bufp
->can_be_null
= (analysis
!= 0);
4079 } /* re_compile_fastmap */
4081 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4082 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4083 this memory for recording register information. STARTS and ENDS
4084 must be allocated using the malloc library routine, and must each
4085 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4087 If NUM_REGS == 0, then subsequent matches should allocate their own
4090 Unless this function is called, the first search or match using
4091 PATTERN_BUFFER will allocate its own register data, without
4092 freeing the old data. */
4095 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4096 struct re_pattern_buffer
*bufp
;
4097 struct re_registers
*regs
;
4099 regoff_t
*starts
, *ends
;
4103 bufp
->regs_allocated
= REGS_REALLOCATE
;
4104 regs
->num_regs
= num_regs
;
4105 regs
->start
= starts
;
4110 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4112 regs
->start
= regs
->end
= (regoff_t
*) 0;
4115 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4117 /* Searching routines. */
4119 /* Like re_search_2, below, but only one string is specified, and
4120 doesn't let you say where to stop matching. */
4123 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4124 struct re_pattern_buffer
*bufp
;
4126 int size
, startpos
, range
;
4127 struct re_registers
*regs
;
4129 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4132 WEAK_ALIAS (__re_search
, re_search
)
4134 /* Head address of virtual concatenation of string. */
4135 #define HEAD_ADDR_VSTRING(P) \
4136 (((P) >= size1 ? string2 : string1))
4138 /* End address of virtual concatenation of string. */
4139 #define STOP_ADDR_VSTRING(P) \
4140 (((P) >= size1 ? string2 + size2 : string1 + size1))
4142 /* Address of POS in the concatenation of virtual string. */
4143 #define POS_ADDR_VSTRING(POS) \
4144 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4146 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4147 virtual concatenation of STRING1 and STRING2, starting first at index
4148 STARTPOS, then at STARTPOS + 1, and so on.
4150 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4152 RANGE is how far to scan while trying to match. RANGE = 0 means try
4153 only at STARTPOS; in general, the last start tried is STARTPOS +
4156 In REGS, return the indices of the virtual concatenation of STRING1
4157 and STRING2 that matched the entire BUFP->buffer and its contained
4160 Do not consider matching one past the index STOP in the virtual
4161 concatenation of STRING1 and STRING2.
4163 We return either the position in the strings at which the match was
4164 found, -1 if no match, or -2 if error (such as failure
4168 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4169 struct re_pattern_buffer
*bufp
;
4170 const char *str1
, *str2
;
4174 struct re_registers
*regs
;
4178 re_char
*string1
= (re_char
*) str1
;
4179 re_char
*string2
= (re_char
*) str2
;
4180 register char *fastmap
= bufp
->fastmap
;
4181 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4182 int total_size
= size1
+ size2
;
4183 int endpos
= startpos
+ range
;
4184 boolean anchored_start
;
4186 /* Nonzero if we have to concern multibyte character. */
4187 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4189 /* Check for out-of-range STARTPOS. */
4190 if (startpos
< 0 || startpos
> total_size
)
4193 /* Fix up RANGE if it might eventually take us outside
4194 the virtual concatenation of STRING1 and STRING2.
4195 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4197 range
= 0 - startpos
;
4198 else if (endpos
> total_size
)
4199 range
= total_size
- startpos
;
4201 /* If the search isn't to be a backwards one, don't waste time in a
4202 search for a pattern anchored at beginning of buffer. */
4203 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4212 /* In a forward search for something that starts with \=.
4213 don't keep searching past point. */
4214 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4216 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4222 /* Update the fastmap now if not correct already. */
4223 if (fastmap
&& !bufp
->fastmap_accurate
)
4224 re_compile_fastmap (bufp
);
4226 /* See whether the pattern is anchored. */
4227 anchored_start
= (bufp
->buffer
[0] == begline
);
4230 gl_state
.object
= re_match_object
;
4232 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4234 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4238 /* Loop through the string, looking for a place to start matching. */
4241 /* If the pattern is anchored,
4242 skip quickly past places we cannot match.
4243 We don't bother to treat startpos == 0 specially
4244 because that case doesn't repeat. */
4245 if (anchored_start
&& startpos
> 0)
4247 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4248 : string2
[startpos
- size1
- 1])
4253 /* If a fastmap is supplied, skip quickly over characters that
4254 cannot be the start of a match. If the pattern can match the
4255 null string, however, we don't need to skip characters; we want
4256 the first null string. */
4257 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4259 register re_char
*d
;
4260 register re_wchar_t buf_ch
;
4262 d
= POS_ADDR_VSTRING (startpos
);
4264 if (range
> 0) /* Searching forwards. */
4266 register int lim
= 0;
4269 if (startpos
< size1
&& startpos
+ range
>= size1
)
4270 lim
= range
- (size1
- startpos
);
4272 /* Written out as an if-else to avoid testing `translate'
4274 if (RE_TRANSLATE_P (translate
))
4281 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4284 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4289 range
-= buf_charlen
;
4294 /* Convert *d to integer to shut up GCC's
4295 whining about comparison that is always
4300 && !fastmap
[RE_TRANSLATE (translate
, di
)])
4308 while (range
> lim
&& !fastmap
[*d
])
4314 startpos
+= irange
- range
;
4316 else /* Searching backwards. */
4318 int room
= (startpos
>= size1
4319 ? size2
+ size1
- startpos
4320 : size1
- startpos
);
4321 buf_ch
= RE_STRING_CHAR (d
, room
);
4322 buf_ch
= TRANSLATE (buf_ch
);
4324 if (! (buf_ch
>= 0400
4325 || fastmap
[buf_ch
]))
4330 /* If can't match the null string, and that's all we have left, fail. */
4331 if (range
>= 0 && startpos
== total_size
&& fastmap
4332 && !bufp
->can_be_null
)
4335 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4336 startpos
, regs
, stop
);
4337 #ifndef REGEX_MALLOC
4354 /* Update STARTPOS to the next character boundary. */
4357 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4358 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4359 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4377 /* Update STARTPOS to the previous character boundary. */
4380 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4382 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4384 /* Find the head of multibyte form. */
4385 PREV_CHAR_BOUNDARY (p
, phead
);
4386 range
+= p0
- 1 - p
;
4390 startpos
-= p0
- 1 - p
;
4396 WEAK_ALIAS (__re_search_2
, re_search_2
)
4398 /* Declarations and macros for re_match_2. */
4400 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4402 RE_TRANSLATE_TYPE translate
,
4403 const int multibyte
));
4405 /* This converts PTR, a pointer into one of the search strings `string1'
4406 and `string2' into an offset from the beginning of that string. */
4407 #define POINTER_TO_OFFSET(ptr) \
4408 (FIRST_STRING_P (ptr) \
4409 ? ((regoff_t) ((ptr) - string1)) \
4410 : ((regoff_t) ((ptr) - string2 + size1)))
4412 /* Call before fetching a character with *d. This switches over to
4413 string2 if necessary.
4414 Check re_match_2_internal for a discussion of why end_match_2 might
4415 not be within string2 (but be equal to end_match_1 instead). */
4416 #define PREFETCH() \
4419 /* End of string2 => fail. */ \
4420 if (dend == end_match_2) \
4422 /* End of string1 => advance to string2. */ \
4424 dend = end_match_2; \
4427 /* Call before fetching a char with *d if you already checked other limits.
4428 This is meant for use in lookahead operations like wordend, etc..
4429 where we might need to look at parts of the string that might be
4430 outside of the LIMITs (i.e past `stop'). */
4431 #define PREFETCH_NOLIMIT() \
4435 dend = end_match_2; \
4438 /* Test if at very beginning or at very end of the virtual concatenation
4439 of `string1' and `string2'. If only one string, it's `string2'. */
4440 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4441 #define AT_STRINGS_END(d) ((d) == end2)
4444 /* Test if D points to a character which is word-constituent. We have
4445 two special cases to check for: if past the end of string1, look at
4446 the first character in string2; and if before the beginning of
4447 string2, look at the last character in string1. */
4448 #define WORDCHAR_P(d) \
4449 (SYNTAX ((d) == end1 ? *string2 \
4450 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4453 /* Disabled due to a compiler bug -- see comment at case wordbound */
4455 /* The comment at case wordbound is following one, but we don't use
4456 AT_WORD_BOUNDARY anymore to support multibyte form.
4458 The DEC Alpha C compiler 3.x generates incorrect code for the
4459 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4460 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4461 macro and introducing temporary variables works around the bug. */
4464 /* Test if the character before D and the one at D differ with respect
4465 to being word-constituent. */
4466 #define AT_WORD_BOUNDARY(d) \
4467 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4468 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4471 /* Free everything we malloc. */
4472 #ifdef MATCH_MAY_ALLOCATE
4473 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4474 # define FREE_VARIABLES() \
4476 REGEX_FREE_STACK (fail_stack.stack); \
4477 FREE_VAR (regstart); \
4478 FREE_VAR (regend); \
4479 FREE_VAR (best_regstart); \
4480 FREE_VAR (best_regend); \
4483 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4484 #endif /* not MATCH_MAY_ALLOCATE */
4487 /* Optimization routines. */
4489 /* If the operation is a match against one or more chars,
4490 return a pointer to the next operation, else return NULL. */
4495 switch (SWITCH_ENUM_CAST (*p
++))
4506 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4509 p
= CHARSET_RANGE_TABLE (p
- 1);
4510 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4511 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4514 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4521 case notcategoryspec
:
4533 /* Jump over non-matching operations. */
4535 skip_noops (p
, pend
)
4541 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4550 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4561 /* Non-zero if "p1 matches something" implies "p2 fails". */
4563 mutually_exclusive_p (bufp
, p1
, p2
)
4564 struct re_pattern_buffer
*bufp
;
4568 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4569 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4571 assert (p1
>= bufp
->buffer
&& p1
< pend
4572 && p2
>= bufp
->buffer
&& p2
<= pend
);
4574 /* Skip over open/close-group commands.
4575 If what follows this loop is a ...+ construct,
4576 look at what begins its body, since we will have to
4577 match at least one of that. */
4578 p2
= skip_noops (p2
, pend
);
4579 /* The same skip can be done for p1, except that this function
4580 is only used in the case where p1 is a simple match operator. */
4581 /* p1 = skip_noops (p1, pend); */
4583 assert (p1
>= bufp
->buffer
&& p1
< pend
4584 && p2
>= bufp
->buffer
&& p2
<= pend
);
4586 op2
= p2
== pend
? succeed
: *p2
;
4588 switch (SWITCH_ENUM_CAST (op2
))
4592 /* If we're at the end of the pattern, we can change. */
4593 if (skip_one_char (p1
))
4595 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4603 register re_wchar_t c
4604 = (re_opcode_t
) *p2
== endline
? '\n'
4605 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4607 if ((re_opcode_t
) *p1
== exactn
)
4609 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4611 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4616 else if ((re_opcode_t
) *p1
== charset
4617 || (re_opcode_t
) *p1
== charset_not
)
4619 int not = (re_opcode_t
) *p1
== charset_not
;
4621 /* Test if C is listed in charset (or charset_not)
4623 if (SINGLE_BYTE_CHAR_P (c
))
4625 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4626 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4629 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4630 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4632 /* `not' is equal to 1 if c would match, which means
4633 that we can't change to pop_failure_jump. */
4636 DEBUG_PRINT1 (" No match => fast loop.\n");
4640 else if ((re_opcode_t
) *p1
== anychar
4643 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4651 if ((re_opcode_t
) *p1
== exactn
)
4652 /* Reuse the code above. */
4653 return mutually_exclusive_p (bufp
, p2
, p1
);
4655 /* It is hard to list up all the character in charset
4656 P2 if it includes multibyte character. Give up in
4658 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4660 /* Now, we are sure that P2 has no range table.
4661 So, for the size of bitmap in P2, `p2[1]' is
4662 enough. But P1 may have range table, so the
4663 size of bitmap table of P1 is extracted by
4664 using macro `CHARSET_BITMAP_SIZE'.
4666 Since we know that all the character listed in
4667 P2 is ASCII, it is enough to test only bitmap
4670 if ((re_opcode_t
) *p1
== charset
)
4673 /* We win if the charset inside the loop
4674 has no overlap with the one after the loop. */
4677 && idx
< CHARSET_BITMAP_SIZE (p1
));
4679 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4683 || idx
== CHARSET_BITMAP_SIZE (p1
))
4685 DEBUG_PRINT1 (" No match => fast loop.\n");
4689 else if ((re_opcode_t
) *p1
== charset_not
)
4692 /* We win if the charset_not inside the loop lists
4693 every character listed in the charset after. */
4694 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4695 if (! (p2
[2 + idx
] == 0
4696 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4697 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4702 DEBUG_PRINT1 (" No match => fast loop.\n");
4711 switch (SWITCH_ENUM_CAST (*p1
))
4715 /* Reuse the code above. */
4716 return mutually_exclusive_p (bufp
, p2
, p1
);
4718 /* When we have two charset_not, it's very unlikely that
4719 they don't overlap. The union of the two sets of excluded
4720 chars should cover all possible chars, which, as a matter of
4721 fact, is virtually impossible in multibyte buffers. */
4727 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4729 return ((re_opcode_t
) *p1
== syntaxspec
4730 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4732 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4735 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4737 return ((re_opcode_t
) *p1
== notsyntaxspec
4738 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4740 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4743 return (((re_opcode_t
) *p1
== notsyntaxspec
4744 || (re_opcode_t
) *p1
== syntaxspec
)
4749 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4750 case notcategoryspec
:
4751 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4763 /* Matching routines. */
4765 #ifndef emacs /* Emacs never uses this. */
4766 /* re_match is like re_match_2 except it takes only a single string. */
4769 re_match (bufp
, string
, size
, pos
, regs
)
4770 struct re_pattern_buffer
*bufp
;
4773 struct re_registers
*regs
;
4775 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4777 # if defined C_ALLOCA && !defined REGEX_MALLOC
4782 WEAK_ALIAS (__re_match
, re_match
)
4783 #endif /* not emacs */
4786 /* In Emacs, this is the string or buffer in which we
4787 are matching. It is used for looking up syntax properties. */
4788 Lisp_Object re_match_object
;
4791 /* re_match_2 matches the compiled pattern in BUFP against the
4792 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4793 and SIZE2, respectively). We start matching at POS, and stop
4796 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4797 store offsets for the substring each group matched in REGS. See the
4798 documentation for exactly how many groups we fill.
4800 We return -1 if no match, -2 if an internal error (such as the
4801 failure stack overflowing). Otherwise, we return the length of the
4802 matched substring. */
4805 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4806 struct re_pattern_buffer
*bufp
;
4807 const char *string1
, *string2
;
4810 struct re_registers
*regs
;
4817 gl_state
.object
= re_match_object
;
4818 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4819 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4822 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4823 (re_char
*) string2
, size2
,
4825 #if defined C_ALLOCA && !defined REGEX_MALLOC
4830 WEAK_ALIAS (__re_match_2
, re_match_2
)
4832 /* This is a separate function so that we can force an alloca cleanup
4835 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4836 struct re_pattern_buffer
*bufp
;
4837 re_char
*string1
, *string2
;
4840 struct re_registers
*regs
;
4843 /* General temporaries. */
4848 /* Just past the end of the corresponding string. */
4849 re_char
*end1
, *end2
;
4851 /* Pointers into string1 and string2, just past the last characters in
4852 each to consider matching. */
4853 re_char
*end_match_1
, *end_match_2
;
4855 /* Where we are in the data, and the end of the current string. */
4858 /* Used sometimes to remember where we were before starting matching
4859 an operator so that we can go back in case of failure. This "atomic"
4860 behavior of matching opcodes is indispensable to the correctness
4861 of the on_failure_keep_string_jump optimization. */
4864 /* Where we are in the pattern, and the end of the pattern. */
4865 re_char
*p
= bufp
->buffer
;
4866 re_char
*pend
= p
+ bufp
->used
;
4868 /* We use this to map every character in the string. */
4869 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4871 /* Nonzero if we have to concern multibyte character. */
4872 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4874 /* Failure point stack. Each place that can handle a failure further
4875 down the line pushes a failure point on this stack. It consists of
4876 regstart, and regend for all registers corresponding to
4877 the subexpressions we're currently inside, plus the number of such
4878 registers, and, finally, two char *'s. The first char * is where
4879 to resume scanning the pattern; the second one is where to resume
4880 scanning the strings. */
4881 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4882 fail_stack_type fail_stack
;
4885 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4888 #if defined REL_ALLOC && defined REGEX_MALLOC
4889 /* This holds the pointer to the failure stack, when
4890 it is allocated relocatably. */
4891 fail_stack_elt_t
*failure_stack_ptr
;
4894 /* We fill all the registers internally, independent of what we
4895 return, for use in backreferences. The number here includes
4896 an element for register zero. */
4897 size_t num_regs
= bufp
->re_nsub
+ 1;
4899 /* Information on the contents of registers. These are pointers into
4900 the input strings; they record just what was matched (on this
4901 attempt) by a subexpression part of the pattern, that is, the
4902 regnum-th regstart pointer points to where in the pattern we began
4903 matching and the regnum-th regend points to right after where we
4904 stopped matching the regnum-th subexpression. (The zeroth register
4905 keeps track of what the whole pattern matches.) */
4906 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4907 re_char
**regstart
, **regend
;
4910 /* The following record the register info as found in the above
4911 variables when we find a match better than any we've seen before.
4912 This happens as we backtrack through the failure points, which in
4913 turn happens only if we have not yet matched the entire string. */
4914 unsigned best_regs_set
= false;
4915 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4916 re_char
**best_regstart
, **best_regend
;
4919 /* Logically, this is `best_regend[0]'. But we don't want to have to
4920 allocate space for that if we're not allocating space for anything
4921 else (see below). Also, we never need info about register 0 for
4922 any of the other register vectors, and it seems rather a kludge to
4923 treat `best_regend' differently than the rest. So we keep track of
4924 the end of the best match so far in a separate variable. We
4925 initialize this to NULL so that when we backtrack the first time
4926 and need to test it, it's not garbage. */
4927 re_char
*match_end
= NULL
;
4930 /* Counts the total number of registers pushed. */
4931 unsigned num_regs_pushed
= 0;
4934 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4938 #ifdef MATCH_MAY_ALLOCATE
4939 /* Do not bother to initialize all the register variables if there are
4940 no groups in the pattern, as it takes a fair amount of time. If
4941 there are groups, we include space for register 0 (the whole
4942 pattern), even though we never use it, since it simplifies the
4943 array indexing. We should fix this. */
4946 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4947 regend
= REGEX_TALLOC (num_regs
, re_char
*);
4948 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4949 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
4951 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
4959 /* We must initialize all our variables to NULL, so that
4960 `FREE_VARIABLES' doesn't try to free them. */
4961 regstart
= regend
= best_regstart
= best_regend
= NULL
;
4963 #endif /* MATCH_MAY_ALLOCATE */
4965 /* The starting position is bogus. */
4966 if (pos
< 0 || pos
> size1
+ size2
)
4972 /* Initialize subexpression text positions to -1 to mark ones that no
4973 start_memory/stop_memory has been seen for. Also initialize the
4974 register information struct. */
4975 for (reg
= 1; reg
< num_regs
; reg
++)
4976 regstart
[reg
] = regend
[reg
] = NULL
;
4978 /* We move `string1' into `string2' if the latter's empty -- but not if
4979 `string1' is null. */
4980 if (size2
== 0 && string1
!= NULL
)
4987 end1
= string1
+ size1
;
4988 end2
= string2
+ size2
;
4990 /* `p' scans through the pattern as `d' scans through the data.
4991 `dend' is the end of the input string that `d' points within. `d'
4992 is advanced into the following input string whenever necessary, but
4993 this happens before fetching; therefore, at the beginning of the
4994 loop, `d' can be pointing at the end of a string, but it cannot
4998 /* Only match within string2. */
4999 d
= string2
+ pos
- size1
;
5000 dend
= end_match_2
= string2
+ stop
- size1
;
5001 end_match_1
= end1
; /* Just to give it a value. */
5007 /* Only match within string1. */
5008 end_match_1
= string1
+ stop
;
5010 When we reach end_match_1, PREFETCH normally switches to string2.
5011 But in the present case, this means that just doing a PREFETCH
5012 makes us jump from `stop' to `gap' within the string.
5013 What we really want here is for the search to stop as
5014 soon as we hit end_match_1. That's why we set end_match_2
5015 to end_match_1 (since PREFETCH fails as soon as we hit
5017 end_match_2
= end_match_1
;
5020 { /* It's important to use this code when stop == size so that
5021 moving `d' from end1 to string2 will not prevent the d == dend
5022 check from catching the end of string. */
5024 end_match_2
= string2
+ stop
- size1
;
5030 DEBUG_PRINT1 ("The compiled pattern is: ");
5031 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5032 DEBUG_PRINT1 ("The string to match is: `");
5033 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5034 DEBUG_PRINT1 ("'\n");
5036 /* This loops over pattern commands. It exits by returning from the
5037 function if the match is complete, or it drops through if the match
5038 fails at this starting point in the input data. */
5041 DEBUG_PRINT2 ("\n%p: ", p
);
5044 { /* End of pattern means we might have succeeded. */
5045 DEBUG_PRINT1 ("end of pattern ... ");
5047 /* If we haven't matched the entire string, and we want the
5048 longest match, try backtracking. */
5049 if (d
!= end_match_2
)
5051 /* 1 if this match ends in the same string (string1 or string2)
5052 as the best previous match. */
5053 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5054 == FIRST_STRING_P (d
));
5055 /* 1 if this match is the best seen so far. */
5056 boolean best_match_p
;
5058 /* AIX compiler got confused when this was combined
5059 with the previous declaration. */
5061 best_match_p
= d
> match_end
;
5063 best_match_p
= !FIRST_STRING_P (d
);
5065 DEBUG_PRINT1 ("backtracking.\n");
5067 if (!FAIL_STACK_EMPTY ())
5068 { /* More failure points to try. */
5070 /* If exceeds best match so far, save it. */
5071 if (!best_regs_set
|| best_match_p
)
5073 best_regs_set
= true;
5076 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5078 for (reg
= 1; reg
< num_regs
; reg
++)
5080 best_regstart
[reg
] = regstart
[reg
];
5081 best_regend
[reg
] = regend
[reg
];
5087 /* If no failure points, don't restore garbage. And if
5088 last match is real best match, don't restore second
5090 else if (best_regs_set
&& !best_match_p
)
5093 /* Restore best match. It may happen that `dend ==
5094 end_match_1' while the restored d is in string2.
5095 For example, the pattern `x.*y.*z' against the
5096 strings `x-' and `y-z-', if the two strings are
5097 not consecutive in memory. */
5098 DEBUG_PRINT1 ("Restoring best registers.\n");
5101 dend
= ((d
>= string1
&& d
<= end1
)
5102 ? end_match_1
: end_match_2
);
5104 for (reg
= 1; reg
< num_regs
; reg
++)
5106 regstart
[reg
] = best_regstart
[reg
];
5107 regend
[reg
] = best_regend
[reg
];
5110 } /* d != end_match_2 */
5113 DEBUG_PRINT1 ("Accepting match.\n");
5115 /* If caller wants register contents data back, do it. */
5116 if (regs
&& !bufp
->no_sub
)
5118 /* Have the register data arrays been allocated? */
5119 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5120 { /* No. So allocate them with malloc. We need one
5121 extra element beyond `num_regs' for the `-1' marker
5123 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5124 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5125 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5126 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5131 bufp
->regs_allocated
= REGS_REALLOCATE
;
5133 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5134 { /* Yes. If we need more elements than were already
5135 allocated, reallocate them. If we need fewer, just
5137 if (regs
->num_regs
< num_regs
+ 1)
5139 regs
->num_regs
= num_regs
+ 1;
5140 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5141 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5142 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5151 /* These braces fend off a "empty body in an else-statement"
5152 warning under GCC when assert expands to nothing. */
5153 assert (bufp
->regs_allocated
== REGS_FIXED
);
5156 /* Convert the pointer data in `regstart' and `regend' to
5157 indices. Register zero has to be set differently,
5158 since we haven't kept track of any info for it. */
5159 if (regs
->num_regs
> 0)
5161 regs
->start
[0] = pos
;
5162 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5165 /* Go through the first `min (num_regs, regs->num_regs)'
5166 registers, since that is all we initialized. */
5167 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5169 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5170 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5174 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5176 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5180 /* If the regs structure we return has more elements than
5181 were in the pattern, set the extra elements to -1. If
5182 we (re)allocated the registers, this is the case,
5183 because we always allocate enough to have at least one
5185 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5186 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5187 } /* regs && !bufp->no_sub */
5189 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5190 nfailure_points_pushed
, nfailure_points_popped
,
5191 nfailure_points_pushed
- nfailure_points_popped
);
5192 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5194 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5196 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5202 /* Otherwise match next pattern command. */
5203 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5205 /* Ignore these. Used to ignore the n of succeed_n's which
5206 currently have n == 0. */
5208 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5212 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5215 /* Match the next n pattern characters exactly. The following
5216 byte in the pattern defines n, and the n bytes after that
5217 are the characters to match. */
5220 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5222 /* Remember the start point to rollback upon failure. */
5225 /* This is written out as an if-else so we don't waste time
5226 testing `translate' inside the loop. */
5227 if (RE_TRANSLATE_P (translate
))
5232 int pat_charlen
, buf_charlen
;
5233 unsigned int pat_ch
, buf_ch
;
5236 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5237 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5239 if (RE_TRANSLATE (translate
, buf_ch
)
5248 mcnt
-= pat_charlen
;
5254 /* Avoid compiler whining about comparison being
5260 if (RE_TRANSLATE (translate
, di
) != *p
++)
5285 /* Match any character except possibly a newline or a null. */
5291 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5294 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5295 buf_ch
= TRANSLATE (buf_ch
);
5297 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5299 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5300 && buf_ch
== '\000'))
5303 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5312 register unsigned int c
;
5313 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5316 /* Start of actual range_table, or end of bitmap if there is no
5318 re_char
*range_table
;
5320 /* Nonzero if there is a range table. */
5321 int range_table_exists
;
5323 /* Number of ranges of range table. This is not included
5324 in the initial byte-length of the command. */
5327 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5329 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5331 if (range_table_exists
)
5333 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5334 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5338 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5339 c
= TRANSLATE (c
); /* The character to match. */
5341 if (SINGLE_BYTE_CHAR_P (c
))
5342 { /* Lookup bitmap. */
5343 /* Cast to `unsigned' instead of `unsigned char' in
5344 case the bit list is a full 32 bytes long. */
5345 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5346 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5350 else if (range_table_exists
)
5352 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5354 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5355 | (class_bits
& BIT_MULTIBYTE
)
5356 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5357 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5358 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5359 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5362 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5366 if (range_table_exists
)
5367 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5369 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5371 if (!not) goto fail
;
5378 /* The beginning of a group is represented by start_memory.
5379 The argument is the register number. The text
5380 matched within the group is recorded (in the internal
5381 registers data structure) under the register number. */
5383 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5385 /* In case we need to undo this operation (via backtracking). */
5386 PUSH_FAILURE_REG ((unsigned int)*p
);
5389 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5390 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5392 /* Move past the register number and inner group count. */
5397 /* The stop_memory opcode represents the end of a group. Its
5398 argument is the same as start_memory's: the register number. */
5400 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5402 assert (!REG_UNSET (regstart
[*p
]));
5403 /* Strictly speaking, there should be code such as:
5405 assert (REG_UNSET (regend[*p]));
5406 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5408 But the only info to be pushed is regend[*p] and it is known to
5409 be UNSET, so there really isn't anything to push.
5410 Not pushing anything, on the other hand deprives us from the
5411 guarantee that regend[*p] is UNSET since undoing this operation
5412 will not reset its value properly. This is not important since
5413 the value will only be read on the next start_memory or at
5414 the very end and both events can only happen if this stop_memory
5418 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5420 /* Move past the register number and the inner group count. */
5425 /* \<digit> has been turned into a `duplicate' command which is
5426 followed by the numeric value of <digit> as the register number. */
5429 register re_char
*d2
, *dend2
;
5430 int regno
= *p
++; /* Get which register to match against. */
5431 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5433 /* Can't back reference a group which we've never matched. */
5434 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5437 /* Where in input to try to start matching. */
5438 d2
= regstart
[regno
];
5440 /* Remember the start point to rollback upon failure. */
5443 /* Where to stop matching; if both the place to start and
5444 the place to stop matching are in the same string, then
5445 set to the place to stop, otherwise, for now have to use
5446 the end of the first string. */
5448 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5449 == FIRST_STRING_P (regend
[regno
]))
5450 ? regend
[regno
] : end_match_1
);
5453 /* If necessary, advance to next segment in register
5457 if (dend2
== end_match_2
) break;
5458 if (dend2
== regend
[regno
]) break;
5460 /* End of string1 => advance to string2. */
5462 dend2
= regend
[regno
];
5464 /* At end of register contents => success */
5465 if (d2
== dend2
) break;
5467 /* If necessary, advance to next segment in data. */
5470 /* How many characters left in this segment to match. */
5473 /* Want how many consecutive characters we can match in
5474 one shot, so, if necessary, adjust the count. */
5475 if (mcnt
> dend2
- d2
)
5478 /* Compare that many; failure if mismatch, else move
5480 if (RE_TRANSLATE_P (translate
)
5481 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5482 : memcmp (d
, d2
, mcnt
))
5487 d
+= mcnt
, d2
+= mcnt
;
5493 /* begline matches the empty string at the beginning of the string
5494 (unless `not_bol' is set in `bufp'), and after newlines. */
5496 DEBUG_PRINT1 ("EXECUTING begline.\n");
5498 if (AT_STRINGS_BEG (d
))
5500 if (!bufp
->not_bol
) break;
5505 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5509 /* In all other cases, we fail. */
5513 /* endline is the dual of begline. */
5515 DEBUG_PRINT1 ("EXECUTING endline.\n");
5517 if (AT_STRINGS_END (d
))
5519 if (!bufp
->not_eol
) break;
5523 PREFETCH_NOLIMIT ();
5530 /* Match at the very beginning of the data. */
5532 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5533 if (AT_STRINGS_BEG (d
))
5538 /* Match at the very end of the data. */
5540 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5541 if (AT_STRINGS_END (d
))
5546 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5547 pushes NULL as the value for the string on the stack. Then
5548 `POP_FAILURE_POINT' will keep the current value for the
5549 string, instead of restoring it. To see why, consider
5550 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5551 then the . fails against the \n. But the next thing we want
5552 to do is match the \n against the \n; if we restored the
5553 string value, we would be back at the foo.
5555 Because this is used only in specific cases, we don't need to
5556 check all the things that `on_failure_jump' does, to make
5557 sure the right things get saved on the stack. Hence we don't
5558 share its code. The only reason to push anything on the
5559 stack at all is that otherwise we would have to change
5560 `anychar's code to do something besides goto fail in this
5561 case; that seems worse than this. */
5562 case on_failure_keep_string_jump
:
5563 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5564 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5567 PUSH_FAILURE_POINT (p
- 3, NULL
);
5570 /* A nasty loop is introduced by the non-greedy *? and +?.
5571 With such loops, the stack only ever contains one failure point
5572 at a time, so that a plain on_failure_jump_loop kind of
5573 cycle detection cannot work. Worse yet, such a detection
5574 can not only fail to detect a cycle, but it can also wrongly
5575 detect a cycle (between different instantiations of the same
5577 So the method used for those nasty loops is a little different:
5578 We use a special cycle-detection-stack-frame which is pushed
5579 when the on_failure_jump_nastyloop failure-point is *popped*.
5580 This special frame thus marks the beginning of one iteration
5581 through the loop and we can hence easily check right here
5582 whether something matched between the beginning and the end of
5584 case on_failure_jump_nastyloop
:
5585 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5586 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5589 assert ((re_opcode_t
)p
[-4] == no_op
);
5592 CHECK_INFINITE_LOOP (p
- 4, d
);
5594 /* If there's a cycle, just continue without pushing
5595 this failure point. The failure point is the "try again"
5596 option, which shouldn't be tried.
5597 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5598 PUSH_FAILURE_POINT (p
- 3, d
);
5602 /* Simple loop detecting on_failure_jump: just check on the
5603 failure stack if the same spot was already hit earlier. */
5604 case on_failure_jump_loop
:
5606 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5607 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5611 CHECK_INFINITE_LOOP (p
- 3, d
);
5613 /* If there's a cycle, get out of the loop, as if the matching
5614 had failed. We used to just `goto fail' here, but that was
5615 aborting the search a bit too early: we want to keep the
5616 empty-loop-match and keep matching after the loop.
5617 We want (x?)*y\1z to match both xxyz and xxyxz. */
5620 PUSH_FAILURE_POINT (p
- 3, d
);
5625 /* Uses of on_failure_jump:
5627 Each alternative starts with an on_failure_jump that points
5628 to the beginning of the next alternative. Each alternative
5629 except the last ends with a jump that in effect jumps past
5630 the rest of the alternatives. (They really jump to the
5631 ending jump of the following alternative, because tensioning
5632 these jumps is a hassle.)
5634 Repeats start with an on_failure_jump that points past both
5635 the repetition text and either the following jump or
5636 pop_failure_jump back to this on_failure_jump. */
5637 case on_failure_jump
:
5638 IMMEDIATE_QUIT_CHECK
;
5639 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5640 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5643 PUSH_FAILURE_POINT (p
-3, d
);
5646 /* This operation is used for greedy *.
5647 Compare the beginning of the repeat with what in the
5648 pattern follows its end. If we can establish that there
5649 is nothing that they would both match, i.e., that we
5650 would have to backtrack because of (as in, e.g., `a*a')
5651 then we can use a non-backtracking loop based on
5652 on_failure_keep_string_jump instead of on_failure_jump. */
5653 case on_failure_jump_smart
:
5654 IMMEDIATE_QUIT_CHECK
;
5655 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5656 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5659 re_char
*p1
= p
; /* Next operation. */
5660 /* Here, we discard `const', making re_match non-reentrant. */
5661 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5662 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5664 p
-= 3; /* Reset so that we will re-execute the
5665 instruction once it's been changed. */
5667 EXTRACT_NUMBER (mcnt
, p2
- 2);
5669 /* Ensure this is a indeed the trivial kind of loop
5670 we are expecting. */
5671 assert (skip_one_char (p1
) == p2
- 3);
5672 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5673 DEBUG_STATEMENT (debug
+= 2);
5674 if (mutually_exclusive_p (bufp
, p1
, p2
))
5676 /* Use a fast `on_failure_keep_string_jump' loop. */
5677 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5678 *p3
= (unsigned char) on_failure_keep_string_jump
;
5679 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5683 /* Default to a safe `on_failure_jump' loop. */
5684 DEBUG_PRINT1 (" smart default => slow loop.\n");
5685 *p3
= (unsigned char) on_failure_jump
;
5687 DEBUG_STATEMENT (debug
-= 2);
5691 /* Unconditionally jump (without popping any failure points). */
5694 IMMEDIATE_QUIT_CHECK
;
5695 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5696 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5697 p
+= mcnt
; /* Do the jump. */
5698 DEBUG_PRINT2 ("(to %p).\n", p
);
5702 /* Have to succeed matching what follows at least n times.
5703 After that, handle like `on_failure_jump'. */
5705 /* Signedness doesn't matter since we only compare MCNT to 0. */
5706 EXTRACT_NUMBER (mcnt
, p
+ 2);
5707 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5709 /* Originally, mcnt is how many times we HAVE to succeed. */
5712 /* Here, we discard `const', making re_match non-reentrant. */
5713 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5716 PUSH_NUMBER (p2
, mcnt
);
5719 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5724 /* Signedness doesn't matter since we only compare MCNT to 0. */
5725 EXTRACT_NUMBER (mcnt
, p
+ 2);
5726 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5728 /* Originally, this is how many times we CAN jump. */
5731 /* Here, we discard `const', making re_match non-reentrant. */
5732 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5734 PUSH_NUMBER (p2
, mcnt
);
5735 goto unconditional_jump
;
5737 /* If don't have to jump any more, skip over the rest of command. */
5744 unsigned char *p2
; /* Location of the counter. */
5745 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5747 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5748 /* Here, we discard `const', making re_match non-reentrant. */
5749 p2
= (unsigned char*) p
+ mcnt
;
5750 /* Signedness doesn't matter since we only copy MCNT's bits . */
5751 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5752 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5753 PUSH_NUMBER (p2
, mcnt
);
5759 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5760 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5762 /* We SUCCEED (or FAIL) in one of the following cases: */
5764 /* Case 1: D is at the beginning or the end of string. */
5765 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5769 /* C1 is the character before D, S1 is the syntax of C1, C2
5770 is the character at D, and S2 is the syntax of C2. */
5774 int offset
= PTR_TO_OFFSET (d
- 1);
5775 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5776 UPDATE_SYNTAX_TABLE (charpos
);
5778 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5781 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5783 PREFETCH_NOLIMIT ();
5784 c2
= RE_STRING_CHAR (d
, dend
- d
);
5787 if (/* Case 2: Only one of S1 and S2 is Sword. */
5788 ((s1
== Sword
) != (s2
== Sword
))
5789 /* Case 3: Both of S1 and S2 are Sword, and macro
5790 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5791 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5800 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5802 /* We FAIL in one of the following cases: */
5804 /* Case 1: D is at the end of string. */
5805 if (AT_STRINGS_END (d
))
5809 /* C1 is the character before D, S1 is the syntax of C1, C2
5810 is the character at D, and S2 is the syntax of C2. */
5814 int offset
= PTR_TO_OFFSET (d
);
5815 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5816 UPDATE_SYNTAX_TABLE (charpos
);
5819 c2
= RE_STRING_CHAR (d
, dend
- d
);
5822 /* Case 2: S2 is not Sword. */
5826 /* Case 3: D is not at the beginning of string ... */
5827 if (!AT_STRINGS_BEG (d
))
5829 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5831 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5835 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5837 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5844 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5846 /* We FAIL in one of the following cases: */
5848 /* Case 1: D is at the beginning of string. */
5849 if (AT_STRINGS_BEG (d
))
5853 /* C1 is the character before D, S1 is the syntax of C1, C2
5854 is the character at D, and S2 is the syntax of C2. */
5858 int offset
= PTR_TO_OFFSET (d
) - 1;
5859 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5860 UPDATE_SYNTAX_TABLE (charpos
);
5862 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5865 /* Case 2: S1 is not Sword. */
5869 /* Case 3: D is not at the end of string ... */
5870 if (!AT_STRINGS_END (d
))
5872 PREFETCH_NOLIMIT ();
5873 c2
= RE_STRING_CHAR (d
, dend
- d
);
5875 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5879 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5881 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5888 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
5890 /* We FAIL in one of the following cases: */
5892 /* Case 1: D is at the end of string. */
5893 if (AT_STRINGS_END (d
))
5897 /* C1 is the character before D, S1 is the syntax of C1, C2
5898 is the character at D, and S2 is the syntax of C2. */
5902 int offset
= PTR_TO_OFFSET (d
);
5903 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5904 UPDATE_SYNTAX_TABLE (charpos
);
5907 c2
= RE_STRING_CHAR (d
, dend
- d
);
5910 /* Case 2: S2 is neither Sword nor Ssymbol. */
5911 if (s2
!= Sword
&& s2
!= Ssymbol
)
5914 /* Case 3: D is not at the beginning of string ... */
5915 if (!AT_STRINGS_BEG (d
))
5917 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5919 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5923 /* ... and S1 is Sword or Ssymbol. */
5924 if (s1
== Sword
|| s1
== Ssymbol
)
5931 DEBUG_PRINT1 ("EXECUTING symend.\n");
5933 /* We FAIL in one of the following cases: */
5935 /* Case 1: D is at the beginning of string. */
5936 if (AT_STRINGS_BEG (d
))
5940 /* C1 is the character before D, S1 is the syntax of C1, C2
5941 is the character at D, and S2 is the syntax of C2. */
5945 int offset
= PTR_TO_OFFSET (d
) - 1;
5946 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5947 UPDATE_SYNTAX_TABLE (charpos
);
5949 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5952 /* Case 2: S1 is neither Ssymbol nor Sword. */
5953 if (s1
!= Sword
&& s1
!= Ssymbol
)
5956 /* Case 3: D is not at the end of string ... */
5957 if (!AT_STRINGS_END (d
))
5959 PREFETCH_NOLIMIT ();
5960 c2
= RE_STRING_CHAR (d
, dend
- d
);
5962 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5966 /* ... and S2 is Sword or Ssymbol. */
5967 if (s2
== Sword
|| s2
== Ssymbol
)
5975 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
5977 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
5981 int offset
= PTR_TO_OFFSET (d
);
5982 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5983 UPDATE_SYNTAX_TABLE (pos1
);
5990 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5992 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6000 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6001 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6006 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6007 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6012 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6013 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6018 case notcategoryspec
:
6019 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6021 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
6027 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
6029 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6040 continue; /* Successfully executed one pattern command; keep going. */
6043 /* We goto here if a matching operation fails. */
6045 IMMEDIATE_QUIT_CHECK
;
6046 if (!FAIL_STACK_EMPTY ())
6049 /* A restart point is known. Restore to that state. */
6050 DEBUG_PRINT1 ("\nFAIL:\n");
6051 POP_FAILURE_POINT (str
, pat
);
6052 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6054 case on_failure_keep_string_jump
:
6055 assert (str
== NULL
);
6056 goto continue_failure_jump
;
6058 case on_failure_jump_nastyloop
:
6059 assert ((re_opcode_t
)pat
[-2] == no_op
);
6060 PUSH_FAILURE_POINT (pat
- 2, str
);
6063 case on_failure_jump_loop
:
6064 case on_failure_jump
:
6067 continue_failure_jump
:
6068 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6073 /* A special frame used for nastyloops. */
6080 assert (p
>= bufp
->buffer
&& p
<= pend
);
6082 if (d
>= string1
&& d
<= end1
)
6086 break; /* Matching at this starting point really fails. */
6090 goto restore_best_regs
;
6094 return -1; /* Failure to match. */
6097 /* Subroutine definitions for re_match_2. */
6099 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6100 bytes; nonzero otherwise. */
6103 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
6106 RE_TRANSLATE_TYPE translate
;
6107 const int multibyte
;
6109 register re_char
*p1
= s1
, *p2
= s2
;
6110 re_char
*p1_end
= s1
+ len
;
6111 re_char
*p2_end
= s2
+ len
;
6113 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6114 different lengths, but relying on a single `len' would break this. -sm */
6115 while (p1
< p1_end
&& p2
< p2_end
)
6117 int p1_charlen
, p2_charlen
;
6118 re_wchar_t p1_ch
, p2_ch
;
6120 p1_ch
= RE_STRING_CHAR_AND_LENGTH (p1
, p1_end
- p1
, p1_charlen
);
6121 p2_ch
= RE_STRING_CHAR_AND_LENGTH (p2
, p2_end
- p2
, p2_charlen
);
6123 if (RE_TRANSLATE (translate
, p1_ch
)
6124 != RE_TRANSLATE (translate
, p2_ch
))
6127 p1
+= p1_charlen
, p2
+= p2_charlen
;
6130 if (p1
!= p1_end
|| p2
!= p2_end
)
6136 /* Entry points for GNU code. */
6138 /* re_compile_pattern is the GNU regular expression compiler: it
6139 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6140 Returns 0 if the pattern was valid, otherwise an error string.
6142 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6143 are set in BUFP on entry.
6145 We call regex_compile to do the actual compilation. */
6148 re_compile_pattern (pattern
, length
, bufp
)
6149 const char *pattern
;
6151 struct re_pattern_buffer
*bufp
;
6155 /* GNU code is written to assume at least RE_NREGS registers will be set
6156 (and at least one extra will be -1). */
6157 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6159 /* And GNU code determines whether or not to get register information
6160 by passing null for the REGS argument to re_match, etc., not by
6164 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6168 return gettext (re_error_msgid
[(int) ret
]);
6170 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6172 /* Entry points compatible with 4.2 BSD regex library. We don't define
6173 them unless specifically requested. */
6175 #if defined _REGEX_RE_COMP || defined _LIBC
6177 /* BSD has one and only one pattern buffer. */
6178 static struct re_pattern_buffer re_comp_buf
;
6182 /* Make these definitions weak in libc, so POSIX programs can redefine
6183 these names if they don't use our functions, and still use
6184 regcomp/regexec below without link errors. */
6194 if (!re_comp_buf
.buffer
)
6195 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6196 return (char *) gettext ("No previous regular expression");
6200 if (!re_comp_buf
.buffer
)
6202 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6203 if (re_comp_buf
.buffer
== NULL
)
6204 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6205 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6206 re_comp_buf
.allocated
= 200;
6208 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6209 if (re_comp_buf
.fastmap
== NULL
)
6210 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6211 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6214 /* Since `re_exec' always passes NULL for the `regs' argument, we
6215 don't need to initialize the pattern buffer fields which affect it. */
6217 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6222 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6223 return (char *) gettext (re_error_msgid
[(int) ret
]);
6234 const int len
= strlen (s
);
6236 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6238 #endif /* _REGEX_RE_COMP */
6240 /* POSIX.2 functions. Don't define these for Emacs. */
6244 /* regcomp takes a regular expression as a string and compiles it.
6246 PREG is a regex_t *. We do not expect any fields to be initialized,
6247 since POSIX says we shouldn't. Thus, we set
6249 `buffer' to the compiled pattern;
6250 `used' to the length of the compiled pattern;
6251 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6252 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6253 RE_SYNTAX_POSIX_BASIC;
6254 `fastmap' to an allocated space for the fastmap;
6255 `fastmap_accurate' to zero;
6256 `re_nsub' to the number of subexpressions in PATTERN.
6258 PATTERN is the address of the pattern string.
6260 CFLAGS is a series of bits which affect compilation.
6262 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6263 use POSIX basic syntax.
6265 If REG_NEWLINE is set, then . and [^...] don't match newline.
6266 Also, regexec will try a match beginning after every newline.
6268 If REG_ICASE is set, then we considers upper- and lowercase
6269 versions of letters to be equivalent when matching.
6271 If REG_NOSUB is set, then when PREG is passed to regexec, that
6272 routine will report only success or failure, and nothing about the
6275 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6276 the return codes and their meanings.) */
6279 regcomp (preg
, pattern
, cflags
)
6280 regex_t
*__restrict preg
;
6281 const char *__restrict pattern
;
6286 = (cflags
& REG_EXTENDED
) ?
6287 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6289 /* regex_compile will allocate the space for the compiled pattern. */
6291 preg
->allocated
= 0;
6294 /* Try to allocate space for the fastmap. */
6295 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6297 if (cflags
& REG_ICASE
)
6302 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6303 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6304 if (preg
->translate
== NULL
)
6305 return (int) REG_ESPACE
;
6307 /* Map uppercase characters to corresponding lowercase ones. */
6308 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6309 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6312 preg
->translate
= NULL
;
6314 /* If REG_NEWLINE is set, newlines are treated differently. */
6315 if (cflags
& REG_NEWLINE
)
6316 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6317 syntax
&= ~RE_DOT_NEWLINE
;
6318 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6321 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6323 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6325 /* POSIX says a null character in the pattern terminates it, so we
6326 can use strlen here in compiling the pattern. */
6327 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6329 /* POSIX doesn't distinguish between an unmatched open-group and an
6330 unmatched close-group: both are REG_EPAREN. */
6331 if (ret
== REG_ERPAREN
)
6334 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6335 { /* Compute the fastmap now, since regexec cannot modify the pattern
6337 re_compile_fastmap (preg
);
6338 if (preg
->can_be_null
)
6339 { /* The fastmap can't be used anyway. */
6340 free (preg
->fastmap
);
6341 preg
->fastmap
= NULL
;
6346 WEAK_ALIAS (__regcomp
, regcomp
)
6349 /* regexec searches for a given pattern, specified by PREG, in the
6352 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6353 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6354 least NMATCH elements, and we set them to the offsets of the
6355 corresponding matched substrings.
6357 EFLAGS specifies `execution flags' which affect matching: if
6358 REG_NOTBOL is set, then ^ does not match at the beginning of the
6359 string; if REG_NOTEOL is set, then $ does not match at the end.
6361 We return 0 if we find a match and REG_NOMATCH if not. */
6364 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6365 const regex_t
*__restrict preg
;
6366 const char *__restrict string
;
6368 regmatch_t pmatch
[__restrict_arr
];
6372 struct re_registers regs
;
6373 regex_t private_preg
;
6374 int len
= strlen (string
);
6375 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6377 private_preg
= *preg
;
6379 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6380 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6382 /* The user has told us exactly how many registers to return
6383 information about, via `nmatch'. We have to pass that on to the
6384 matching routines. */
6385 private_preg
.regs_allocated
= REGS_FIXED
;
6389 regs
.num_regs
= nmatch
;
6390 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6391 if (regs
.start
== NULL
)
6392 return (int) REG_NOMATCH
;
6393 regs
.end
= regs
.start
+ nmatch
;
6396 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6397 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6398 was a little bit longer but still only matching the real part.
6399 This works because the `endline' will check for a '\n' and will find a
6400 '\0', correctly deciding that this is not the end of a line.
6401 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6402 a convenient '\0' there. For all we know, the string could be preceded
6403 by '\n' which would throw things off. */
6405 /* Perform the searching operation. */
6406 ret
= re_search (&private_preg
, string
, len
,
6407 /* start: */ 0, /* range: */ len
,
6408 want_reg_info
? ®s
: (struct re_registers
*) 0);
6410 /* Copy the register information to the POSIX structure. */
6417 for (r
= 0; r
< nmatch
; r
++)
6419 pmatch
[r
].rm_so
= regs
.start
[r
];
6420 pmatch
[r
].rm_eo
= regs
.end
[r
];
6424 /* If we needed the temporary register info, free the space now. */
6428 /* We want zero return to mean success, unlike `re_search'. */
6429 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6431 WEAK_ALIAS (__regexec
, regexec
)
6434 /* Returns a message corresponding to an error code, ERRCODE, returned
6435 from either regcomp or regexec. We don't use PREG here. */
6438 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6440 const regex_t
*preg
;
6448 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6449 /* Only error codes returned by the rest of the code should be passed
6450 to this routine. If we are given anything else, or if other regex
6451 code generates an invalid error code, then the program has a bug.
6452 Dump core so we can fix it. */
6455 msg
= gettext (re_error_msgid
[errcode
]);
6457 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6459 if (errbuf_size
!= 0)
6461 if (msg_size
> errbuf_size
)
6463 strncpy (errbuf
, msg
, errbuf_size
- 1);
6464 errbuf
[errbuf_size
- 1] = 0;
6467 strcpy (errbuf
, msg
);
6472 WEAK_ALIAS (__regerror
, regerror
)
6475 /* Free dynamically allocated space used by PREG. */
6481 if (preg
->buffer
!= NULL
)
6482 free (preg
->buffer
);
6483 preg
->buffer
= NULL
;
6485 preg
->allocated
= 0;
6488 if (preg
->fastmap
!= NULL
)
6489 free (preg
->fastmap
);
6490 preg
->fastmap
= NULL
;
6491 preg
->fastmap_accurate
= 0;
6493 if (preg
->translate
!= NULL
)
6494 free (preg
->translate
);
6495 preg
->translate
= NULL
;
6497 WEAK_ALIAS (__regfree
, regfree
)
6499 #endif /* not emacs */
6501 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6502 (do not change this comment) */