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., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
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 /* This table gives an error message for each of the error codes listed
1263 in regex.h. Obviously the order here has to be same as there.
1264 POSIX doesn't require that we do anything for REG_NOERROR,
1265 but why not be nice? */
1267 static const char *re_error_msgid
[] =
1269 gettext_noop ("Success"), /* REG_NOERROR */
1270 gettext_noop ("No match"), /* REG_NOMATCH */
1271 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1272 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1273 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1274 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1275 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1276 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1277 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1278 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1279 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1280 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1281 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1282 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1283 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1284 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1285 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1286 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1289 /* Avoiding alloca during matching, to placate r_alloc. */
1291 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1292 searching and matching functions should not call alloca. On some
1293 systems, alloca is implemented in terms of malloc, and if we're
1294 using the relocating allocator routines, then malloc could cause a
1295 relocation, which might (if the strings being searched are in the
1296 ralloc heap) shift the data out from underneath the regexp
1299 Here's another reason to avoid allocation: Emacs
1300 processes input from X in a signal handler; processing X input may
1301 call malloc; if input arrives while a matching routine is calling
1302 malloc, then we're scrod. But Emacs can't just block input while
1303 calling matching routines; then we don't notice interrupts when
1304 they come in. So, Emacs blocks input around all regexp calls
1305 except the matching calls, which it leaves unprotected, in the
1306 faith that they will not malloc. */
1308 /* Normally, this is fine. */
1309 #define MATCH_MAY_ALLOCATE
1311 /* When using GNU C, we are not REALLY using the C alloca, no matter
1312 what config.h may say. So don't take precautions for it. */
1317 /* The match routines may not allocate if (1) they would do it with malloc
1318 and (2) it's not safe for them to use malloc.
1319 Note that if REL_ALLOC is defined, matching would not use malloc for the
1320 failure stack, but we would still use it for the register vectors;
1321 so REL_ALLOC should not affect this. */
1322 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1323 # undef MATCH_MAY_ALLOCATE
1327 /* Failure stack declarations and macros; both re_compile_fastmap and
1328 re_match_2 use a failure stack. These have to be macros because of
1329 REGEX_ALLOCATE_STACK. */
1332 /* Approximate number of failure points for which to initially allocate space
1333 when matching. If this number is exceeded, we allocate more
1334 space, so it is not a hard limit. */
1335 #ifndef INIT_FAILURE_ALLOC
1336 # define INIT_FAILURE_ALLOC 20
1339 /* Roughly the maximum number of failure points on the stack. Would be
1340 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1341 This is a variable only so users of regex can assign to it; we never
1342 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1343 before using it, so it should probably be a byte-count instead. */
1344 # if defined MATCH_MAY_ALLOCATE
1345 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1346 whose default stack limit is 2mb. In order for a larger
1347 value to work reliably, you have to try to make it accord
1348 with the process stack limit. */
1349 size_t re_max_failures
= 40000;
1351 size_t re_max_failures
= 4000;
1354 union fail_stack_elt
1357 /* This should be the biggest `int' that's no bigger than a pointer. */
1361 typedef union fail_stack_elt fail_stack_elt_t
;
1365 fail_stack_elt_t
*stack
;
1367 size_t avail
; /* Offset of next open position. */
1368 size_t frame
; /* Offset of the cur constructed frame. */
1371 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1372 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1375 /* Define macros to initialize and free the failure stack.
1376 Do `return -2' if the alloc fails. */
1378 #ifdef MATCH_MAY_ALLOCATE
1379 # define INIT_FAIL_STACK() \
1381 fail_stack.stack = (fail_stack_elt_t *) \
1382 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1383 * sizeof (fail_stack_elt_t)); \
1385 if (fail_stack.stack == NULL) \
1388 fail_stack.size = INIT_FAILURE_ALLOC; \
1389 fail_stack.avail = 0; \
1390 fail_stack.frame = 0; \
1393 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1395 # define INIT_FAIL_STACK() \
1397 fail_stack.avail = 0; \
1398 fail_stack.frame = 0; \
1401 # define RESET_FAIL_STACK() ((void)0)
1405 /* Double the size of FAIL_STACK, up to a limit
1406 which allows approximately `re_max_failures' items.
1408 Return 1 if succeeds, and 0 if either ran out of memory
1409 allocating space for it or it was already too large.
1411 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1413 /* Factor to increase the failure stack size by
1414 when we increase it.
1415 This used to be 2, but 2 was too wasteful
1416 because the old discarded stacks added up to as much space
1417 were as ultimate, maximum-size stack. */
1418 #define FAIL_STACK_GROWTH_FACTOR 4
1420 #define GROW_FAIL_STACK(fail_stack) \
1421 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1422 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1424 : ((fail_stack).stack \
1425 = (fail_stack_elt_t *) \
1426 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1427 (fail_stack).size * sizeof (fail_stack_elt_t), \
1428 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1429 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1430 * FAIL_STACK_GROWTH_FACTOR))), \
1432 (fail_stack).stack == NULL \
1434 : ((fail_stack).size \
1435 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1436 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1437 * FAIL_STACK_GROWTH_FACTOR)) \
1438 / sizeof (fail_stack_elt_t)), \
1442 /* Push a pointer value onto the failure stack.
1443 Assumes the variable `fail_stack'. Probably should only
1444 be called from within `PUSH_FAILURE_POINT'. */
1445 #define PUSH_FAILURE_POINTER(item) \
1446 fail_stack.stack[fail_stack.avail++].pointer = (item)
1448 /* This pushes an integer-valued item onto the failure stack.
1449 Assumes the variable `fail_stack'. Probably should only
1450 be called from within `PUSH_FAILURE_POINT'. */
1451 #define PUSH_FAILURE_INT(item) \
1452 fail_stack.stack[fail_stack.avail++].integer = (item)
1454 /* Push a fail_stack_elt_t value onto the failure stack.
1455 Assumes the variable `fail_stack'. Probably should only
1456 be called from within `PUSH_FAILURE_POINT'. */
1457 #define PUSH_FAILURE_ELT(item) \
1458 fail_stack.stack[fail_stack.avail++] = (item)
1460 /* These three POP... operations complement the three PUSH... operations.
1461 All assume that `fail_stack' is nonempty. */
1462 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1463 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1464 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1466 /* Individual items aside from the registers. */
1467 #define NUM_NONREG_ITEMS 3
1469 /* Used to examine the stack (to detect infinite loops). */
1470 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1471 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1472 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1473 #define TOP_FAILURE_HANDLE() fail_stack.frame
1476 #define ENSURE_FAIL_STACK(space) \
1477 while (REMAINING_AVAIL_SLOTS <= space) { \
1478 if (!GROW_FAIL_STACK (fail_stack)) \
1480 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1481 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1484 /* Push register NUM onto the stack. */
1485 #define PUSH_FAILURE_REG(num) \
1487 char *destination; \
1488 ENSURE_FAIL_STACK(3); \
1489 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1490 num, regstart[num], regend[num]); \
1491 PUSH_FAILURE_POINTER (regstart[num]); \
1492 PUSH_FAILURE_POINTER (regend[num]); \
1493 PUSH_FAILURE_INT (num); \
1496 /* Change the counter's value to VAL, but make sure that it will
1497 be reset when backtracking. */
1498 #define PUSH_NUMBER(ptr,val) \
1500 char *destination; \
1502 ENSURE_FAIL_STACK(3); \
1503 EXTRACT_NUMBER (c, ptr); \
1504 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1505 PUSH_FAILURE_INT (c); \
1506 PUSH_FAILURE_POINTER (ptr); \
1507 PUSH_FAILURE_INT (-1); \
1508 STORE_NUMBER (ptr, val); \
1511 /* Pop a saved register off the stack. */
1512 #define POP_FAILURE_REG_OR_COUNT() \
1514 int reg = POP_FAILURE_INT (); \
1517 /* It's a counter. */ \
1518 /* Here, we discard `const', making re_match non-reentrant. */ \
1519 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1520 reg = POP_FAILURE_INT (); \
1521 STORE_NUMBER (ptr, reg); \
1522 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1526 regend[reg] = POP_FAILURE_POINTER (); \
1527 regstart[reg] = POP_FAILURE_POINTER (); \
1528 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1529 reg, regstart[reg], regend[reg]); \
1533 /* Check that we are not stuck in an infinite loop. */
1534 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1536 int failure = TOP_FAILURE_HANDLE (); \
1537 /* Check for infinite matching loops */ \
1538 while (failure > 0 \
1539 && (FAILURE_STR (failure) == string_place \
1540 || FAILURE_STR (failure) == NULL)) \
1542 assert (FAILURE_PAT (failure) >= bufp->buffer \
1543 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1544 if (FAILURE_PAT (failure) == pat_cur) \
1549 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1550 failure = NEXT_FAILURE_HANDLE(failure); \
1552 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1555 /* Push the information about the state we will need
1556 if we ever fail back to it.
1558 Requires variables fail_stack, regstart, regend and
1559 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1562 Does `return FAILURE_CODE' if runs out of memory. */
1564 #define PUSH_FAILURE_POINT(pattern, string_place) \
1566 char *destination; \
1567 /* Must be int, so when we don't save any registers, the arithmetic \
1568 of 0 + -1 isn't done as unsigned. */ \
1570 DEBUG_STATEMENT (nfailure_points_pushed++); \
1571 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1572 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1573 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1575 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1577 DEBUG_PRINT1 ("\n"); \
1579 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1580 PUSH_FAILURE_INT (fail_stack.frame); \
1582 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1583 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1584 DEBUG_PRINT1 ("'\n"); \
1585 PUSH_FAILURE_POINTER (string_place); \
1587 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1588 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1589 PUSH_FAILURE_POINTER (pattern); \
1591 /* Close the frame by moving the frame pointer past it. */ \
1592 fail_stack.frame = fail_stack.avail; \
1595 /* Estimate the size of data pushed by a typical failure stack entry.
1596 An estimate is all we need, because all we use this for
1597 is to choose a limit for how big to make the failure stack. */
1598 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1599 #define TYPICAL_FAILURE_SIZE 20
1601 /* How many items can still be added to the stack without overflowing it. */
1602 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1605 /* Pops what PUSH_FAIL_STACK pushes.
1607 We restore into the parameters, all of which should be lvalues:
1608 STR -- the saved data position.
1609 PAT -- the saved pattern position.
1610 REGSTART, REGEND -- arrays of string positions.
1612 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1613 `pend', `string1', `size1', `string2', and `size2'. */
1615 #define POP_FAILURE_POINT(str, pat) \
1617 assert (!FAIL_STACK_EMPTY ()); \
1619 /* Remove failure points and point to how many regs pushed. */ \
1620 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1621 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1622 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1624 /* Pop the saved registers. */ \
1625 while (fail_stack.frame < fail_stack.avail) \
1626 POP_FAILURE_REG_OR_COUNT (); \
1628 pat = POP_FAILURE_POINTER (); \
1629 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1630 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1632 /* If the saved string location is NULL, it came from an \
1633 on_failure_keep_string_jump opcode, and we want to throw away the \
1634 saved NULL, thus retaining our current position in the string. */ \
1635 str = POP_FAILURE_POINTER (); \
1636 DEBUG_PRINT2 (" Popping string %p: `", str); \
1637 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1638 DEBUG_PRINT1 ("'\n"); \
1640 fail_stack.frame = POP_FAILURE_INT (); \
1641 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1643 assert (fail_stack.avail >= 0); \
1644 assert (fail_stack.frame <= fail_stack.avail); \
1646 DEBUG_STATEMENT (nfailure_points_popped++); \
1647 } while (0) /* POP_FAILURE_POINT */
1651 /* Registers are set to a sentinel when they haven't yet matched. */
1652 #define REG_UNSET(e) ((e) == NULL)
1654 /* Subroutine declarations and macros for regex_compile. */
1656 static reg_errcode_t regex_compile
_RE_ARGS ((re_char
*pattern
, size_t size
,
1657 reg_syntax_t syntax
,
1658 struct re_pattern_buffer
*bufp
));
1659 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1660 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1661 int arg1
, int arg2
));
1662 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1663 int arg
, unsigned char *end
));
1664 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1665 int arg1
, int arg2
, unsigned char *end
));
1666 static boolean at_begline_loc_p
_RE_ARGS ((re_char
*pattern
,
1668 reg_syntax_t syntax
));
1669 static boolean at_endline_loc_p
_RE_ARGS ((re_char
*p
,
1671 reg_syntax_t syntax
));
1672 static re_char
*skip_one_char
_RE_ARGS ((re_char
*p
));
1673 static int analyse_first
_RE_ARGS ((re_char
*p
, re_char
*pend
,
1674 char *fastmap
, const int multibyte
));
1676 /* Fetch the next character in the uncompiled pattern, with no
1678 #define PATFETCH(c) \
1681 if (p == pend) return REG_EEND; \
1682 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1687 /* If `translate' is non-null, return translate[D], else just D. We
1688 cast the subscript to translate because some data is declared as
1689 `char *', to avoid warnings when a string constant is passed. But
1690 when we use a character as a subscript we must make it unsigned. */
1692 # define TRANSLATE(d) \
1693 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1697 /* Macros for outputting the compiled pattern into `buffer'. */
1699 /* If the buffer isn't allocated when it comes in, use this. */
1700 #define INIT_BUF_SIZE 32
1702 /* Make sure we have at least N more bytes of space in buffer. */
1703 #define GET_BUFFER_SPACE(n) \
1704 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1707 /* Make sure we have one more byte of buffer space and then add C to it. */
1708 #define BUF_PUSH(c) \
1710 GET_BUFFER_SPACE (1); \
1711 *b++ = (unsigned char) (c); \
1715 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1716 #define BUF_PUSH_2(c1, c2) \
1718 GET_BUFFER_SPACE (2); \
1719 *b++ = (unsigned char) (c1); \
1720 *b++ = (unsigned char) (c2); \
1724 /* As with BUF_PUSH_2, except for three bytes. */
1725 #define BUF_PUSH_3(c1, c2, c3) \
1727 GET_BUFFER_SPACE (3); \
1728 *b++ = (unsigned char) (c1); \
1729 *b++ = (unsigned char) (c2); \
1730 *b++ = (unsigned char) (c3); \
1734 /* Store a jump with opcode OP at LOC to location TO. We store a
1735 relative address offset by the three bytes the jump itself occupies. */
1736 #define STORE_JUMP(op, loc, to) \
1737 store_op1 (op, loc, (to) - (loc) - 3)
1739 /* Likewise, for a two-argument jump. */
1740 #define STORE_JUMP2(op, loc, to, arg) \
1741 store_op2 (op, loc, (to) - (loc) - 3, arg)
1743 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1744 #define INSERT_JUMP(op, loc, to) \
1745 insert_op1 (op, loc, (to) - (loc) - 3, b)
1747 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1748 #define INSERT_JUMP2(op, loc, to, arg) \
1749 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1752 /* This is not an arbitrary limit: the arguments which represent offsets
1753 into the pattern are two bytes long. So if 2^15 bytes turns out to
1754 be too small, many things would have to change. */
1755 # define MAX_BUF_SIZE (1L << 15)
1757 #if 0 /* This is when we thought it could be 2^16 bytes. */
1758 /* Any other compiler which, like MSC, has allocation limit below 2^16
1759 bytes will have to use approach similar to what was done below for
1760 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1761 reallocating to 0 bytes. Such thing is not going to work too well.
1762 You have been warned!! */
1763 #if defined _MSC_VER && !defined WIN32
1764 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1765 # define MAX_BUF_SIZE 65500L
1767 # define MAX_BUF_SIZE (1L << 16)
1771 /* Extend the buffer by twice its current size via realloc and
1772 reset the pointers that pointed into the old block to point to the
1773 correct places in the new one. If extending the buffer results in it
1774 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1775 #if __BOUNDED_POINTERS__
1776 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1777 # define MOVE_BUFFER_POINTER(P) \
1778 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1779 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1782 SET_HIGH_BOUND (b); \
1783 SET_HIGH_BOUND (begalt); \
1784 if (fixup_alt_jump) \
1785 SET_HIGH_BOUND (fixup_alt_jump); \
1787 SET_HIGH_BOUND (laststart); \
1788 if (pending_exact) \
1789 SET_HIGH_BOUND (pending_exact); \
1792 # define MOVE_BUFFER_POINTER(P) (P) += incr
1793 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1795 #define EXTEND_BUFFER() \
1797 re_char *old_buffer = bufp->buffer; \
1798 if (bufp->allocated == MAX_BUF_SIZE) \
1800 bufp->allocated <<= 1; \
1801 if (bufp->allocated > MAX_BUF_SIZE) \
1802 bufp->allocated = MAX_BUF_SIZE; \
1803 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1804 if (bufp->buffer == NULL) \
1805 return REG_ESPACE; \
1806 /* If the buffer moved, move all the pointers into it. */ \
1807 if (old_buffer != bufp->buffer) \
1809 int incr = bufp->buffer - old_buffer; \
1810 MOVE_BUFFER_POINTER (b); \
1811 MOVE_BUFFER_POINTER (begalt); \
1812 if (fixup_alt_jump) \
1813 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1815 MOVE_BUFFER_POINTER (laststart); \
1816 if (pending_exact) \
1817 MOVE_BUFFER_POINTER (pending_exact); \
1819 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1823 /* Since we have one byte reserved for the register number argument to
1824 {start,stop}_memory, the maximum number of groups we can report
1825 things about is what fits in that byte. */
1826 #define MAX_REGNUM 255
1828 /* But patterns can have more than `MAX_REGNUM' registers. We just
1829 ignore the excess. */
1830 typedef int regnum_t
;
1833 /* Macros for the compile stack. */
1835 /* Since offsets can go either forwards or backwards, this type needs to
1836 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1837 /* int may be not enough when sizeof(int) == 2. */
1838 typedef long pattern_offset_t
;
1842 pattern_offset_t begalt_offset
;
1843 pattern_offset_t fixup_alt_jump
;
1844 pattern_offset_t laststart_offset
;
1846 } compile_stack_elt_t
;
1851 compile_stack_elt_t
*stack
;
1853 unsigned avail
; /* Offset of next open position. */
1854 } compile_stack_type
;
1857 #define INIT_COMPILE_STACK_SIZE 32
1859 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1860 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1862 /* The next available element. */
1863 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1865 /* Explicit quit checking is only used on NTemacs. */
1866 #if defined WINDOWSNT && defined emacs && defined QUIT
1867 extern int immediate_quit
;
1868 # define IMMEDIATE_QUIT_CHECK \
1870 if (immediate_quit) QUIT; \
1873 # define IMMEDIATE_QUIT_CHECK ((void)0)
1876 /* Structure to manage work area for range table. */
1877 struct range_table_work_area
1879 int *table
; /* actual work area. */
1880 int allocated
; /* allocated size for work area in bytes. */
1881 int used
; /* actually used size in words. */
1882 int bits
; /* flag to record character classes */
1885 /* Make sure that WORK_AREA can hold more N multibyte characters.
1886 This is used only in set_image_of_range and set_image_of_range_1.
1887 It expects WORK_AREA to be a pointer.
1888 If it can't get the space, it returns from the surrounding function. */
1890 #define EXTEND_RANGE_TABLE(work_area, n) \
1892 if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \
1894 extend_range_table_work_area (work_area); \
1895 if ((work_area)->table == 0) \
1896 return (REG_ESPACE); \
1900 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1901 (work_area).bits |= (bit)
1903 /* Bits used to implement the multibyte-part of the various character classes
1904 such as [:alnum:] in a charset's range table. */
1905 #define BIT_WORD 0x1
1906 #define BIT_LOWER 0x2
1907 #define BIT_PUNCT 0x4
1908 #define BIT_SPACE 0x8
1909 #define BIT_UPPER 0x10
1910 #define BIT_MULTIBYTE 0x20
1912 /* Set a range START..END to WORK_AREA.
1913 The range is passed through TRANSLATE, so START and END
1914 should be untranslated. */
1915 #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \
1918 tem = set_image_of_range (&work_area, start, end, translate); \
1920 FREE_STACK_RETURN (tem); \
1923 /* Free allocated memory for WORK_AREA. */
1924 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1926 if ((work_area).table) \
1927 free ((work_area).table); \
1930 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1931 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1932 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1933 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1936 /* Set the bit for character C in a list. */
1937 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1940 /* Get the next unsigned number in the uncompiled pattern. */
1941 #define GET_UNSIGNED_NUMBER(num) \
1942 do { if (p != pend) \
1946 FREE_STACK_RETURN (REG_BADBR); \
1947 while ('0' <= c && c <= '9') \
1953 num = num * 10 + c - '0'; \
1954 if (num / 10 != prev) \
1955 FREE_STACK_RETURN (REG_BADBR); \
1961 FREE_STACK_RETURN (REG_BADBR); \
1965 #if ! WIDE_CHAR_SUPPORT
1967 /* Map a string to the char class it names (if any). */
1972 const char *string
= str
;
1973 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
1974 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
1975 else if (STREQ (string
, "word")) return RECC_WORD
;
1976 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
1977 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
1978 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
1979 else if (STREQ (string
, "lower")) return RECC_LOWER
;
1980 else if (STREQ (string
, "print")) return RECC_PRINT
;
1981 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
1982 else if (STREQ (string
, "space")) return RECC_SPACE
;
1983 else if (STREQ (string
, "upper")) return RECC_UPPER
;
1984 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
1985 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
1986 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
1987 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
1988 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
1989 else if (STREQ (string
, "blank")) return RECC_BLANK
;
1993 /* True iff CH is in the char class CC. */
1995 re_iswctype (ch
, cc
)
2001 case RECC_ALNUM
: return ISALNUM (ch
);
2002 case RECC_ALPHA
: return ISALPHA (ch
);
2003 case RECC_BLANK
: return ISBLANK (ch
);
2004 case RECC_CNTRL
: return ISCNTRL (ch
);
2005 case RECC_DIGIT
: return ISDIGIT (ch
);
2006 case RECC_GRAPH
: return ISGRAPH (ch
);
2007 case RECC_LOWER
: return ISLOWER (ch
);
2008 case RECC_PRINT
: return ISPRINT (ch
);
2009 case RECC_PUNCT
: return ISPUNCT (ch
);
2010 case RECC_SPACE
: return ISSPACE (ch
);
2011 case RECC_UPPER
: return ISUPPER (ch
);
2012 case RECC_XDIGIT
: return ISXDIGIT (ch
);
2013 case RECC_ASCII
: return IS_REAL_ASCII (ch
);
2014 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2015 case RECC_UNIBYTE
: return ISUNIBYTE (ch
);
2016 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2017 case RECC_WORD
: return ISWORD (ch
);
2018 case RECC_ERROR
: return false;
2024 /* Return a bit-pattern to use in the range-table bits to match multibyte
2025 chars of class CC. */
2027 re_wctype_to_bit (cc
)
2032 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2033 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2034 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2035 case RECC_LOWER
: return BIT_LOWER
;
2036 case RECC_UPPER
: return BIT_UPPER
;
2037 case RECC_PUNCT
: return BIT_PUNCT
;
2038 case RECC_SPACE
: return BIT_SPACE
;
2039 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2040 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2047 /* Filling in the work area of a range. */
2049 /* Actually extend the space in WORK_AREA. */
2052 extend_range_table_work_area (work_area
)
2053 struct range_table_work_area
*work_area
;
2055 work_area
->allocated
+= 16 * sizeof (int);
2056 if (work_area
->table
)
2058 = (int *) realloc (work_area
->table
, work_area
->allocated
);
2061 = (int *) malloc (work_area
->allocated
);
2066 /* Carefully find the ranges of codes that are equivalent
2067 under case conversion to the range start..end when passed through
2068 TRANSLATE. Handle the case where non-letters can come in between
2069 two upper-case letters (which happens in Latin-1).
2070 Also handle the case of groups of more than 2 case-equivalent chars.
2072 The basic method is to look at consecutive characters and see
2073 if they can form a run that can be handled as one.
2075 Returns -1 if successful, REG_ESPACE if ran out of space. */
2078 set_image_of_range_1 (work_area
, start
, end
, translate
)
2079 RE_TRANSLATE_TYPE translate
;
2080 struct range_table_work_area
*work_area
;
2081 re_wchar_t start
, end
;
2083 /* `one_case' indicates a character, or a run of characters,
2084 each of which is an isolate (no case-equivalents).
2085 This includes all ASCII non-letters.
2087 `two_case' indicates a character, or a run of characters,
2088 each of which has two case-equivalent forms.
2089 This includes all ASCII letters.
2091 `strange' indicates a character that has more than one
2094 enum case_type
{one_case
, two_case
, strange
};
2096 /* Describe the run that is in progress,
2097 which the next character can try to extend.
2098 If run_type is strange, that means there really is no run.
2099 If run_type is one_case, then run_start...run_end is the run.
2100 If run_type is two_case, then the run is run_start...run_end,
2101 and the case-equivalents end at run_eqv_end. */
2103 enum case_type run_type
= strange
;
2104 int run_start
, run_end
, run_eqv_end
;
2106 Lisp_Object eqv_table
;
2108 if (!RE_TRANSLATE_P (translate
))
2110 EXTEND_RANGE_TABLE (work_area
, 2);
2111 work_area
->table
[work_area
->used
++] = (start
);
2112 work_area
->table
[work_area
->used
++] = (end
);
2116 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2118 for (; start
<= end
; start
++)
2120 enum case_type this_type
;
2121 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2122 int minchar
, maxchar
;
2124 /* Classify this character */
2126 this_type
= one_case
;
2127 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2128 this_type
= two_case
;
2130 this_type
= strange
;
2133 minchar
= start
, maxchar
= eqv
;
2135 minchar
= eqv
, maxchar
= start
;
2137 /* Can this character extend the run in progress? */
2138 if (this_type
== strange
|| this_type
!= run_type
2139 || !(minchar
== run_end
+ 1
2140 && (run_type
== two_case
2141 ? maxchar
== run_eqv_end
+ 1 : 1)))
2144 Record each of its equivalent ranges. */
2145 if (run_type
== one_case
)
2147 EXTEND_RANGE_TABLE (work_area
, 2);
2148 work_area
->table
[work_area
->used
++] = run_start
;
2149 work_area
->table
[work_area
->used
++] = run_end
;
2151 else if (run_type
== two_case
)
2153 EXTEND_RANGE_TABLE (work_area
, 4);
2154 work_area
->table
[work_area
->used
++] = run_start
;
2155 work_area
->table
[work_area
->used
++] = run_end
;
2156 work_area
->table
[work_area
->used
++]
2157 = RE_TRANSLATE (eqv_table
, run_start
);
2158 work_area
->table
[work_area
->used
++]
2159 = RE_TRANSLATE (eqv_table
, run_end
);
2164 if (this_type
== strange
)
2166 /* For a strange character, add each of its equivalents, one
2167 by one. Don't start a range. */
2170 EXTEND_RANGE_TABLE (work_area
, 2);
2171 work_area
->table
[work_area
->used
++] = eqv
;
2172 work_area
->table
[work_area
->used
++] = eqv
;
2173 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2175 while (eqv
!= start
);
2178 /* Add this char to the run, or start a new run. */
2179 else if (run_type
== strange
)
2181 /* Initialize a new range. */
2182 run_type
= this_type
;
2185 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2189 /* Extend a running range. */
2191 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2195 /* If a run is still in progress at the end, finish it now
2196 by recording its equivalent ranges. */
2197 if (run_type
== one_case
)
2199 EXTEND_RANGE_TABLE (work_area
, 2);
2200 work_area
->table
[work_area
->used
++] = run_start
;
2201 work_area
->table
[work_area
->used
++] = run_end
;
2203 else if (run_type
== two_case
)
2205 EXTEND_RANGE_TABLE (work_area
, 4);
2206 work_area
->table
[work_area
->used
++] = run_start
;
2207 work_area
->table
[work_area
->used
++] = run_end
;
2208 work_area
->table
[work_area
->used
++]
2209 = RE_TRANSLATE (eqv_table
, run_start
);
2210 work_area
->table
[work_area
->used
++]
2211 = RE_TRANSLATE (eqv_table
, run_end
);
2219 /* Record the the image of the range start..end when passed through
2220 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2221 and is not even necessarily contiguous.
2222 Normally we approximate it with the smallest contiguous range that contains
2223 all the chars we need. However, for Latin-1 we go to extra effort
2226 This function is not called for ASCII ranges.
2228 Returns -1 if successful, REG_ESPACE if ran out of space. */
2231 set_image_of_range (work_area
, start
, end
, translate
)
2232 RE_TRANSLATE_TYPE translate
;
2233 struct range_table_work_area
*work_area
;
2234 re_wchar_t start
, end
;
2236 re_wchar_t cmin
, cmax
;
2239 /* For Latin-1 ranges, use set_image_of_range_1
2240 to get proper handling of ranges that include letters and nonletters.
2241 For a range that includes the whole of Latin-1, this is not necessary.
2242 For other character sets, we don't bother to get this right. */
2243 if (RE_TRANSLATE_P (translate
) && start
< 04400
2244 && !(start
< 04200 && end
>= 04377))
2251 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2261 EXTEND_RANGE_TABLE (work_area
, 2);
2262 work_area
->table
[work_area
->used
++] = (start
);
2263 work_area
->table
[work_area
->used
++] = (end
);
2265 cmin
= -1, cmax
= -1;
2267 if (RE_TRANSLATE_P (translate
))
2271 for (ch
= start
; ch
<= end
; ch
++)
2273 re_wchar_t c
= TRANSLATE (ch
);
2274 if (! (start
<= c
&& c
<= end
))
2280 cmin
= MIN (cmin
, c
);
2281 cmax
= MAX (cmax
, c
);
2288 EXTEND_RANGE_TABLE (work_area
, 2);
2289 work_area
->table
[work_area
->used
++] = (cmin
);
2290 work_area
->table
[work_area
->used
++] = (cmax
);
2297 #ifndef MATCH_MAY_ALLOCATE
2299 /* If we cannot allocate large objects within re_match_2_internal,
2300 we make the fail stack and register vectors global.
2301 The fail stack, we grow to the maximum size when a regexp
2303 The register vectors, we adjust in size each time we
2304 compile a regexp, according to the number of registers it needs. */
2306 static fail_stack_type fail_stack
;
2308 /* Size with which the following vectors are currently allocated.
2309 That is so we can make them bigger as needed,
2310 but never make them smaller. */
2311 static int regs_allocated_size
;
2313 static re_char
** regstart
, ** regend
;
2314 static re_char
**best_regstart
, **best_regend
;
2316 /* Make the register vectors big enough for NUM_REGS registers,
2317 but don't make them smaller. */
2320 regex_grow_registers (num_regs
)
2323 if (num_regs
> regs_allocated_size
)
2325 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2326 RETALLOC_IF (regend
, num_regs
, re_char
*);
2327 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2328 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2330 regs_allocated_size
= num_regs
;
2334 #endif /* not MATCH_MAY_ALLOCATE */
2336 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2340 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2341 Returns one of error codes defined in `regex.h', or zero for success.
2343 Assumes the `allocated' (and perhaps `buffer') and `translate'
2344 fields are set in BUFP on entry.
2346 If it succeeds, results are put in BUFP (if it returns an error, the
2347 contents of BUFP are undefined):
2348 `buffer' is the compiled pattern;
2349 `syntax' is set to SYNTAX;
2350 `used' is set to the length of the compiled pattern;
2351 `fastmap_accurate' is zero;
2352 `re_nsub' is the number of subexpressions in PATTERN;
2353 `not_bol' and `not_eol' are zero;
2355 The `fastmap' field is neither examined nor set. */
2357 /* Insert the `jump' from the end of last alternative to "here".
2358 The space for the jump has already been allocated. */
2359 #define FIXUP_ALT_JUMP() \
2361 if (fixup_alt_jump) \
2362 STORE_JUMP (jump, fixup_alt_jump, b); \
2366 /* Return, freeing storage we allocated. */
2367 #define FREE_STACK_RETURN(value) \
2369 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2370 free (compile_stack.stack); \
2374 static reg_errcode_t
2375 regex_compile (pattern
, size
, syntax
, bufp
)
2378 reg_syntax_t syntax
;
2379 struct re_pattern_buffer
*bufp
;
2381 /* We fetch characters from PATTERN here. */
2382 register re_wchar_t c
, c1
;
2384 /* A random temporary spot in PATTERN. */
2387 /* Points to the end of the buffer, where we should append. */
2388 register unsigned char *b
;
2390 /* Keeps track of unclosed groups. */
2391 compile_stack_type compile_stack
;
2393 /* Points to the current (ending) position in the pattern. */
2395 /* `const' makes AIX compiler fail. */
2396 unsigned char *p
= pattern
;
2398 re_char
*p
= pattern
;
2400 re_char
*pend
= pattern
+ size
;
2402 /* How to translate the characters in the pattern. */
2403 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2405 /* Address of the count-byte of the most recently inserted `exactn'
2406 command. This makes it possible to tell if a new exact-match
2407 character can be added to that command or if the character requires
2408 a new `exactn' command. */
2409 unsigned char *pending_exact
= 0;
2411 /* Address of start of the most recently finished expression.
2412 This tells, e.g., postfix * where to find the start of its
2413 operand. Reset at the beginning of groups and alternatives. */
2414 unsigned char *laststart
= 0;
2416 /* Address of beginning of regexp, or inside of last group. */
2417 unsigned char *begalt
;
2419 /* Place in the uncompiled pattern (i.e., the {) to
2420 which to go back if the interval is invalid. */
2421 re_char
*beg_interval
;
2423 /* Address of the place where a forward jump should go to the end of
2424 the containing expression. Each alternative of an `or' -- except the
2425 last -- ends with a forward jump of this sort. */
2426 unsigned char *fixup_alt_jump
= 0;
2428 /* Counts open-groups as they are encountered. Remembered for the
2429 matching close-group on the compile stack, so the same register
2430 number is put in the stop_memory as the start_memory. */
2431 regnum_t regnum
= 0;
2433 /* Work area for range table of charset. */
2434 struct range_table_work_area range_table_work
;
2436 /* If the object matched can contain multibyte characters. */
2437 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2441 DEBUG_PRINT1 ("\nCompiling pattern: ");
2444 unsigned debug_count
;
2446 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2447 putchar (pattern
[debug_count
]);
2452 /* Initialize the compile stack. */
2453 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2454 if (compile_stack
.stack
== NULL
)
2457 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2458 compile_stack
.avail
= 0;
2460 range_table_work
.table
= 0;
2461 range_table_work
.allocated
= 0;
2463 /* Initialize the pattern buffer. */
2464 bufp
->syntax
= syntax
;
2465 bufp
->fastmap_accurate
= 0;
2466 bufp
->not_bol
= bufp
->not_eol
= 0;
2468 /* Set `used' to zero, so that if we return an error, the pattern
2469 printer (for debugging) will think there's no pattern. We reset it
2473 /* Always count groups, whether or not bufp->no_sub is set. */
2476 #if !defined emacs && !defined SYNTAX_TABLE
2477 /* Initialize the syntax table. */
2478 init_syntax_once ();
2481 if (bufp
->allocated
== 0)
2484 { /* If zero allocated, but buffer is non-null, try to realloc
2485 enough space. This loses if buffer's address is bogus, but
2486 that is the user's responsibility. */
2487 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2490 { /* Caller did not allocate a buffer. Do it for them. */
2491 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2493 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2495 bufp
->allocated
= INIT_BUF_SIZE
;
2498 begalt
= b
= bufp
->buffer
;
2500 /* Loop through the uncompiled pattern until we're at the end. */
2509 if ( /* If at start of pattern, it's an operator. */
2511 /* If context independent, it's an operator. */
2512 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2513 /* Otherwise, depends on what's come before. */
2514 || at_begline_loc_p (pattern
, p
, syntax
))
2515 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2524 if ( /* If at end of pattern, it's an operator. */
2526 /* If context independent, it's an operator. */
2527 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2528 /* Otherwise, depends on what's next. */
2529 || at_endline_loc_p (p
, pend
, syntax
))
2530 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2539 if ((syntax
& RE_BK_PLUS_QM
)
2540 || (syntax
& RE_LIMITED_OPS
))
2544 /* If there is no previous pattern... */
2547 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2548 FREE_STACK_RETURN (REG_BADRPT
);
2549 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2554 /* 1 means zero (many) matches is allowed. */
2555 boolean zero_times_ok
= 0, many_times_ok
= 0;
2558 /* If there is a sequence of repetition chars, collapse it
2559 down to just one (the right one). We can't combine
2560 interval operators with these because of, e.g., `a{2}*',
2561 which should only match an even number of `a's. */
2565 if ((syntax
& RE_FRUGAL
)
2566 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2570 zero_times_ok
|= c
!= '+';
2571 many_times_ok
|= c
!= '?';
2577 || (!(syntax
& RE_BK_PLUS_QM
)
2578 && (*p
== '+' || *p
== '?')))
2580 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2583 FREE_STACK_RETURN (REG_EESCAPE
);
2584 if (p
[1] == '+' || p
[1] == '?')
2585 PATFETCH (c
); /* Gobble up the backslash. */
2591 /* If we get here, we found another repeat character. */
2595 /* Star, etc. applied to an empty pattern is equivalent
2596 to an empty pattern. */
2597 if (!laststart
|| laststart
== b
)
2600 /* Now we know whether or not zero matches is allowed
2601 and also whether or not two or more matches is allowed. */
2606 boolean simple
= skip_one_char (laststart
) == b
;
2607 unsigned int startoffset
= 0;
2609 /* Check if the loop can match the empty string. */
2610 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2611 ? on_failure_jump
: on_failure_jump_loop
;
2612 assert (skip_one_char (laststart
) <= b
);
2614 if (!zero_times_ok
&& simple
)
2615 { /* Since simple * loops can be made faster by using
2616 on_failure_keep_string_jump, we turn simple P+
2617 into PP* if P is simple. */
2618 unsigned char *p1
, *p2
;
2619 startoffset
= b
- laststart
;
2620 GET_BUFFER_SPACE (startoffset
);
2621 p1
= b
; p2
= laststart
;
2627 GET_BUFFER_SPACE (6);
2630 STORE_JUMP (ofj
, b
, b
+ 6);
2632 /* Simple * loops can use on_failure_keep_string_jump
2633 depending on what follows. But since we don't know
2634 that yet, we leave the decision up to
2635 on_failure_jump_smart. */
2636 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2637 laststart
+ startoffset
, b
+ 6);
2639 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2644 /* A simple ? pattern. */
2645 assert (zero_times_ok
);
2646 GET_BUFFER_SPACE (3);
2647 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2651 else /* not greedy */
2652 { /* I wish the greedy and non-greedy cases could be merged. */
2654 GET_BUFFER_SPACE (7); /* We might use less. */
2657 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2659 /* The non-greedy multiple match looks like
2660 a repeat..until: we only need a conditional jump
2661 at the end of the loop. */
2662 if (emptyp
) BUF_PUSH (no_op
);
2663 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2664 : on_failure_jump
, b
, laststart
);
2668 /* The repeat...until naturally matches one or more.
2669 To also match zero times, we need to first jump to
2670 the end of the loop (its conditional jump). */
2671 INSERT_JUMP (jump
, laststart
, b
);
2677 /* non-greedy a?? */
2678 INSERT_JUMP (jump
, laststart
, b
+ 3);
2680 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2697 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2699 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2701 /* Ensure that we have enough space to push a charset: the
2702 opcode, the length count, and the bitset; 34 bytes in all. */
2703 GET_BUFFER_SPACE (34);
2707 /* We test `*p == '^' twice, instead of using an if
2708 statement, so we only need one BUF_PUSH. */
2709 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2713 /* Remember the first position in the bracket expression. */
2716 /* Push the number of bytes in the bitmap. */
2717 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2719 /* Clear the whole map. */
2720 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2722 /* charset_not matches newline according to a syntax bit. */
2723 if ((re_opcode_t
) b
[-2] == charset_not
2724 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2725 SET_LIST_BIT ('\n');
2727 /* Read in characters and ranges, setting map bits. */
2730 boolean escaped_char
= false;
2731 const unsigned char *p2
= p
;
2733 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2735 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2736 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2737 So the translation is done later in a loop. Example:
2738 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2741 /* \ might escape characters inside [...] and [^...]. */
2742 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2744 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2747 escaped_char
= true;
2751 /* Could be the end of the bracket expression. If it's
2752 not (i.e., when the bracket expression is `[]' so
2753 far), the ']' character bit gets set way below. */
2754 if (c
== ']' && p2
!= p1
)
2758 /* What should we do for the character which is
2759 greater than 0x7F, but not BASE_LEADING_CODE_P?
2762 /* See if we're at the beginning of a possible character
2765 if (!escaped_char
&&
2766 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2768 /* Leave room for the null. */
2769 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2770 const unsigned char *class_beg
;
2776 /* If pattern is `[[:'. */
2777 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2782 if ((c
== ':' && *p
== ']') || p
== pend
)
2784 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2787 /* This is in any case an invalid class name. */
2792 /* If isn't a word bracketed by `[:' and `:]':
2793 undo the ending character, the letters, and
2794 leave the leading `:' and `[' (but set bits for
2796 if (c
== ':' && *p
== ']')
2801 cc
= re_wctype (str
);
2804 FREE_STACK_RETURN (REG_ECTYPE
);
2806 /* Throw away the ] at the end of the character
2810 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2812 /* Most character classes in a multibyte match
2813 just set a flag. Exceptions are is_blank,
2814 is_digit, is_cntrl, and is_xdigit, since
2815 they can only match ASCII characters. We
2816 don't need to handle them for multibyte.
2817 They are distinguished by a negative wctype. */
2820 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work
,
2821 re_wctype_to_bit (cc
));
2823 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2825 int translated
= TRANSLATE (ch
);
2826 if (re_iswctype (btowc (ch
), cc
))
2827 SET_LIST_BIT (translated
);
2830 /* Repeat the loop. */
2835 /* Go back to right after the "[:". */
2839 /* Because the `:' may starts the range, we
2840 can't simply set bit and repeat the loop.
2841 Instead, just set it to C and handle below. */
2846 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2849 /* Discard the `-'. */
2852 /* Fetch the character which ends the range. */
2855 if (SINGLE_BYTE_CHAR_P (c
))
2857 if (! SINGLE_BYTE_CHAR_P (c1
))
2859 /* Handle a range starting with a
2860 character of less than 256, and ending
2861 with a character of not less than 256.
2862 Split that into two ranges, the low one
2863 ending at 0377, and the high one
2864 starting at the smallest character in
2865 the charset of C1 and ending at C1. */
2866 int charset
= CHAR_CHARSET (c1
);
2867 re_wchar_t c2
= MAKE_CHAR (charset
, 0, 0);
2869 SET_RANGE_TABLE_WORK_AREA (range_table_work
,
2874 else if (!SAME_CHARSET_P (c
, c1
))
2875 FREE_STACK_RETURN (REG_ERANGEX
);
2878 /* Range from C to C. */
2881 /* Set the range ... */
2882 if (SINGLE_BYTE_CHAR_P (c
))
2883 /* ... into bitmap. */
2885 re_wchar_t this_char
;
2886 re_wchar_t range_start
= c
, range_end
= c1
;
2888 /* If the start is after the end, the range is empty. */
2889 if (range_start
> range_end
)
2891 if (syntax
& RE_NO_EMPTY_RANGES
)
2892 FREE_STACK_RETURN (REG_ERANGE
);
2893 /* Else, repeat the loop. */
2897 for (this_char
= range_start
; this_char
<= range_end
;
2899 SET_LIST_BIT (TRANSLATE (this_char
));
2903 /* ... into range table. */
2904 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
2907 /* Discard any (non)matching list bytes that are all 0 at the
2908 end of the map. Decrease the map-length byte too. */
2909 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2913 /* Build real range table from work area. */
2914 if (RANGE_TABLE_WORK_USED (range_table_work
)
2915 || RANGE_TABLE_WORK_BITS (range_table_work
))
2918 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2920 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2921 bytes for flags, two for COUNT, and three bytes for
2923 GET_BUFFER_SPACE (4 + used
* 3);
2925 /* Indicate the existence of range table. */
2926 laststart
[1] |= 0x80;
2928 /* Store the character class flag bits into the range table.
2929 If not in emacs, these flag bits are always 0. */
2930 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
2931 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
2933 STORE_NUMBER_AND_INCR (b
, used
/ 2);
2934 for (i
= 0; i
< used
; i
++)
2935 STORE_CHARACTER_AND_INCR
2936 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
2943 if (syntax
& RE_NO_BK_PARENS
)
2950 if (syntax
& RE_NO_BK_PARENS
)
2957 if (syntax
& RE_NEWLINE_ALT
)
2964 if (syntax
& RE_NO_BK_VBAR
)
2971 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2972 goto handle_interval
;
2978 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2980 /* Do not translate the character after the \, so that we can
2981 distinguish, e.g., \B from \b, even if we normally would
2982 translate, e.g., B to b. */
2988 if (syntax
& RE_NO_BK_PARENS
)
2989 goto normal_backslash
;
2996 /* Look for a special (?...) construct */
2997 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
2999 PATFETCH (c
); /* Gobble up the '?'. */
3003 case ':': shy
= 1; break;
3005 /* Only (?:...) is supported right now. */
3006 FREE_STACK_RETURN (REG_BADPAT
);
3017 if (COMPILE_STACK_FULL
)
3019 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3020 compile_stack_elt_t
);
3021 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3023 compile_stack
.size
<<= 1;
3026 /* These are the values to restore when we hit end of this
3027 group. They are all relative offsets, so that if the
3028 whole pattern moves because of realloc, they will still
3030 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3031 COMPILE_STACK_TOP
.fixup_alt_jump
3032 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3033 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3034 COMPILE_STACK_TOP
.regnum
= shy
? -regnum
: regnum
;
3037 start_memory for groups beyond the last one we can
3038 represent in the compiled pattern. */
3039 if (regnum
<= MAX_REGNUM
&& !shy
)
3040 BUF_PUSH_2 (start_memory
, regnum
);
3042 compile_stack
.avail
++;
3047 /* If we've reached MAX_REGNUM groups, then this open
3048 won't actually generate any code, so we'll have to
3049 clear pending_exact explicitly. */
3055 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3057 if (COMPILE_STACK_EMPTY
)
3059 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3060 goto normal_backslash
;
3062 FREE_STACK_RETURN (REG_ERPAREN
);
3068 /* See similar code for backslashed left paren above. */
3069 if (COMPILE_STACK_EMPTY
)
3071 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3074 FREE_STACK_RETURN (REG_ERPAREN
);
3077 /* Since we just checked for an empty stack above, this
3078 ``can't happen''. */
3079 assert (compile_stack
.avail
!= 0);
3081 /* We don't just want to restore into `regnum', because
3082 later groups should continue to be numbered higher,
3083 as in `(ab)c(de)' -- the second group is #2. */
3084 regnum_t this_group_regnum
;
3086 compile_stack
.avail
--;
3087 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3089 = COMPILE_STACK_TOP
.fixup_alt_jump
3090 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3092 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3093 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3094 /* If we've reached MAX_REGNUM groups, then this open
3095 won't actually generate any code, so we'll have to
3096 clear pending_exact explicitly. */
3099 /* We're at the end of the group, so now we know how many
3100 groups were inside this one. */
3101 if (this_group_regnum
<= MAX_REGNUM
&& this_group_regnum
> 0)
3102 BUF_PUSH_2 (stop_memory
, this_group_regnum
);
3107 case '|': /* `\|'. */
3108 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3109 goto normal_backslash
;
3111 if (syntax
& RE_LIMITED_OPS
)
3114 /* Insert before the previous alternative a jump which
3115 jumps to this alternative if the former fails. */
3116 GET_BUFFER_SPACE (3);
3117 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3121 /* The alternative before this one has a jump after it
3122 which gets executed if it gets matched. Adjust that
3123 jump so it will jump to this alternative's analogous
3124 jump (put in below, which in turn will jump to the next
3125 (if any) alternative's such jump, etc.). The last such
3126 jump jumps to the correct final destination. A picture:
3132 If we are at `b', then fixup_alt_jump right now points to a
3133 three-byte space after `a'. We'll put in the jump, set
3134 fixup_alt_jump to right after `b', and leave behind three
3135 bytes which we'll fill in when we get to after `c'. */
3139 /* Mark and leave space for a jump after this alternative,
3140 to be filled in later either by next alternative or
3141 when know we're at the end of a series of alternatives. */
3143 GET_BUFFER_SPACE (3);
3152 /* If \{ is a literal. */
3153 if (!(syntax
& RE_INTERVALS
)
3154 /* If we're at `\{' and it's not the open-interval
3156 || (syntax
& RE_NO_BK_BRACES
))
3157 goto normal_backslash
;
3161 /* If got here, then the syntax allows intervals. */
3163 /* At least (most) this many matches must be made. */
3164 int lower_bound
= 0, upper_bound
= -1;
3169 FREE_STACK_RETURN (REG_EBRACE
);
3171 GET_UNSIGNED_NUMBER (lower_bound
);
3174 GET_UNSIGNED_NUMBER (upper_bound
);
3176 /* Interval such as `{1}' => match exactly once. */
3177 upper_bound
= lower_bound
;
3179 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3180 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3181 FREE_STACK_RETURN (REG_BADBR
);
3183 if (!(syntax
& RE_NO_BK_BRACES
))
3186 FREE_STACK_RETURN (REG_BADBR
);
3192 FREE_STACK_RETURN (REG_BADBR
);
3194 /* We just parsed a valid interval. */
3196 /* If it's invalid to have no preceding re. */
3199 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3200 FREE_STACK_RETURN (REG_BADRPT
);
3201 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3204 goto unfetch_interval
;
3207 if (upper_bound
== 0)
3208 /* If the upper bound is zero, just drop the sub pattern
3211 else if (lower_bound
== 1 && upper_bound
== 1)
3212 /* Just match it once: nothing to do here. */
3215 /* Otherwise, we have a nontrivial interval. When
3216 we're all done, the pattern will look like:
3217 set_number_at <jump count> <upper bound>
3218 set_number_at <succeed_n count> <lower bound>
3219 succeed_n <after jump addr> <succeed_n count>
3221 jump_n <succeed_n addr> <jump count>
3222 (The upper bound and `jump_n' are omitted if
3223 `upper_bound' is 1, though.) */
3225 { /* If the upper bound is > 1, we need to insert
3226 more at the end of the loop. */
3227 unsigned int nbytes
= (upper_bound
< 0 ? 3
3228 : upper_bound
> 1 ? 5 : 0);
3229 unsigned int startoffset
= 0;
3231 GET_BUFFER_SPACE (20); /* We might use less. */
3233 if (lower_bound
== 0)
3235 /* A succeed_n that starts with 0 is really a
3236 a simple on_failure_jump_loop. */
3237 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3243 /* Initialize lower bound of the `succeed_n', even
3244 though it will be set during matching by its
3245 attendant `set_number_at' (inserted next),
3246 because `re_compile_fastmap' needs to know.
3247 Jump to the `jump_n' we might insert below. */
3248 INSERT_JUMP2 (succeed_n
, laststart
,
3253 /* Code to initialize the lower bound. Insert
3254 before the `succeed_n'. The `5' is the last two
3255 bytes of this `set_number_at', plus 3 bytes of
3256 the following `succeed_n'. */
3257 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3262 if (upper_bound
< 0)
3264 /* A negative upper bound stands for infinity,
3265 in which case it degenerates to a plain jump. */
3266 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3269 else if (upper_bound
> 1)
3270 { /* More than one repetition is allowed, so
3271 append a backward jump to the `succeed_n'
3272 that starts this interval.
3274 When we've reached this during matching,
3275 we'll have matched the interval once, so
3276 jump back only `upper_bound - 1' times. */
3277 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3281 /* The location we want to set is the second
3282 parameter of the `jump_n'; that is `b-2' as
3283 an absolute address. `laststart' will be
3284 the `set_number_at' we're about to insert;
3285 `laststart+3' the number to set, the source
3286 for the relative address. But we are
3287 inserting into the middle of the pattern --
3288 so everything is getting moved up by 5.
3289 Conclusion: (b - 2) - (laststart + 3) + 5,
3290 i.e., b - laststart.
3292 We insert this at the beginning of the loop
3293 so that if we fail during matching, we'll
3294 reinitialize the bounds. */
3295 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3296 upper_bound
- 1, b
);
3301 beg_interval
= NULL
;
3306 /* If an invalid interval, match the characters as literals. */
3307 assert (beg_interval
);
3309 beg_interval
= NULL
;
3311 /* normal_char and normal_backslash need `c'. */
3314 if (!(syntax
& RE_NO_BK_BRACES
))
3316 assert (p
> pattern
&& p
[-1] == '\\');
3317 goto normal_backslash
;
3323 /* There is no way to specify the before_dot and after_dot
3324 operators. rms says this is ok. --karl */
3332 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3338 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3344 BUF_PUSH_2 (categoryspec
, c
);
3350 BUF_PUSH_2 (notcategoryspec
, c
);
3356 if (syntax
& RE_NO_GNU_OPS
)
3359 BUF_PUSH_2 (syntaxspec
, Sword
);
3364 if (syntax
& RE_NO_GNU_OPS
)
3367 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3372 if (syntax
& RE_NO_GNU_OPS
)
3378 if (syntax
& RE_NO_GNU_OPS
)
3384 if (syntax
& RE_NO_GNU_OPS
)
3393 FREE_STACK_RETURN (REG_BADPAT
);
3397 if (syntax
& RE_NO_GNU_OPS
)
3399 BUF_PUSH (wordbound
);
3403 if (syntax
& RE_NO_GNU_OPS
)
3405 BUF_PUSH (notwordbound
);
3409 if (syntax
& RE_NO_GNU_OPS
)
3415 if (syntax
& RE_NO_GNU_OPS
)
3420 case '1': case '2': case '3': case '4': case '5':
3421 case '6': case '7': case '8': case '9':
3425 if (syntax
& RE_NO_BK_REFS
)
3426 goto normal_backslash
;
3430 /* Can't back reference to a subexpression before its end. */
3431 if (reg
> regnum
|| group_in_compile_stack (compile_stack
, reg
))
3432 FREE_STACK_RETURN (REG_ESUBREG
);
3435 BUF_PUSH_2 (duplicate
, reg
);
3442 if (syntax
& RE_BK_PLUS_QM
)
3445 goto normal_backslash
;
3449 /* You might think it would be useful for \ to mean
3450 not to translate; but if we don't translate it
3451 it will never match anything. */
3458 /* Expects the character in `c'. */
3460 /* If no exactn currently being built. */
3463 /* If last exactn not at current position. */
3464 || pending_exact
+ *pending_exact
+ 1 != b
3466 /* We have only one byte following the exactn for the count. */
3467 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3469 /* If followed by a repetition operator. */
3470 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3471 || ((syntax
& RE_BK_PLUS_QM
)
3472 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3473 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3474 || ((syntax
& RE_INTERVALS
)
3475 && ((syntax
& RE_NO_BK_BRACES
)
3476 ? p
!= pend
&& *p
== '{'
3477 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3479 /* Start building a new exactn. */
3483 BUF_PUSH_2 (exactn
, 0);
3484 pending_exact
= b
- 1;
3487 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3493 len
= CHAR_STRING (c
, b
);
3497 (*pending_exact
) += len
;
3502 } /* while p != pend */
3505 /* Through the pattern now. */
3509 if (!COMPILE_STACK_EMPTY
)
3510 FREE_STACK_RETURN (REG_EPAREN
);
3512 /* If we don't want backtracking, force success
3513 the first time we reach the end of the compiled pattern. */
3514 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3517 /* We have succeeded; set the length of the buffer. */
3518 bufp
->used
= b
- bufp
->buffer
;
3523 re_compile_fastmap (bufp
);
3524 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3525 print_compiled_pattern (bufp
);
3530 #ifndef MATCH_MAY_ALLOCATE
3531 /* Initialize the failure stack to the largest possible stack. This
3532 isn't necessary unless we're trying to avoid calling alloca in
3533 the search and match routines. */
3535 int num_regs
= bufp
->re_nsub
+ 1;
3537 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3539 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3541 if (! fail_stack
.stack
)
3543 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3544 * sizeof (fail_stack_elt_t
));
3547 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3549 * sizeof (fail_stack_elt_t
)));
3552 regex_grow_registers (num_regs
);
3554 #endif /* not MATCH_MAY_ALLOCATE */
3556 FREE_STACK_RETURN (REG_NOERROR
);
3557 } /* regex_compile */
3559 /* Subroutines for `regex_compile'. */
3561 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3564 store_op1 (op
, loc
, arg
)
3569 *loc
= (unsigned char) op
;
3570 STORE_NUMBER (loc
+ 1, arg
);
3574 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3577 store_op2 (op
, loc
, arg1
, arg2
)
3582 *loc
= (unsigned char) op
;
3583 STORE_NUMBER (loc
+ 1, arg1
);
3584 STORE_NUMBER (loc
+ 3, arg2
);
3588 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3589 for OP followed by two-byte integer parameter ARG. */
3592 insert_op1 (op
, loc
, arg
, end
)
3598 register unsigned char *pfrom
= end
;
3599 register unsigned char *pto
= end
+ 3;
3601 while (pfrom
!= loc
)
3604 store_op1 (op
, loc
, arg
);
3608 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3611 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3617 register unsigned char *pfrom
= end
;
3618 register unsigned char *pto
= end
+ 5;
3620 while (pfrom
!= loc
)
3623 store_op2 (op
, loc
, arg1
, arg2
);
3627 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3628 after an alternative or a begin-subexpression. We assume there is at
3629 least one character before the ^. */
3632 at_begline_loc_p (pattern
, p
, syntax
)
3633 re_char
*pattern
, *p
;
3634 reg_syntax_t syntax
;
3636 re_char
*prev
= p
- 2;
3637 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3640 /* After a subexpression? */
3641 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3642 /* After an alternative? */
3643 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
))
3644 /* After a shy subexpression? */
3645 || ((syntax
& RE_SHY_GROUPS
) && prev
- 2 >= pattern
3646 && prev
[-1] == '?' && prev
[-2] == '('
3647 && (syntax
& RE_NO_BK_PARENS
3648 || (prev
- 3 >= pattern
&& prev
[-3] == '\\')));
3652 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3653 at least one character after the $, i.e., `P < PEND'. */
3656 at_endline_loc_p (p
, pend
, syntax
)
3658 reg_syntax_t syntax
;
3661 boolean next_backslash
= *next
== '\\';
3662 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3665 /* Before a subexpression? */
3666 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3667 : next_backslash
&& next_next
&& *next_next
== ')')
3668 /* Before an alternative? */
3669 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3670 : next_backslash
&& next_next
&& *next_next
== '|');
3674 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3675 false if it's not. */
3678 group_in_compile_stack (compile_stack
, regnum
)
3679 compile_stack_type compile_stack
;
3684 for (this_element
= compile_stack
.avail
- 1;
3687 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3694 If fastmap is non-NULL, go through the pattern and fill fastmap
3695 with all the possible leading chars. If fastmap is NULL, don't
3696 bother filling it up (obviously) and only return whether the
3697 pattern could potentially match the empty string.
3699 Return 1 if p..pend might match the empty string.
3700 Return 0 if p..pend matches at least one char.
3701 Return -1 if fastmap was not updated accurately. */
3704 analyse_first (p
, pend
, fastmap
, multibyte
)
3707 const int multibyte
;
3712 /* If all elements for base leading-codes in fastmap is set, this
3713 flag is set true. */
3714 boolean match_any_multibyte_characters
= false;
3718 /* The loop below works as follows:
3719 - It has a working-list kept in the PATTERN_STACK and which basically
3720 starts by only containing a pointer to the first operation.
3721 - If the opcode we're looking at is a match against some set of
3722 chars, then we add those chars to the fastmap and go on to the
3723 next work element from the worklist (done via `break').
3724 - If the opcode is a control operator on the other hand, we either
3725 ignore it (if it's meaningless at this point, such as `start_memory')
3726 or execute it (if it's a jump). If the jump has several destinations
3727 (i.e. `on_failure_jump'), then we push the other destination onto the
3729 We guarantee termination by ignoring backward jumps (more or less),
3730 so that `p' is monotonically increasing. More to the point, we
3731 never set `p' (or push) anything `<= p1'. */
3735 /* `p1' is used as a marker of how far back a `on_failure_jump'
3736 can go without being ignored. It is normally equal to `p'
3737 (which prevents any backward `on_failure_jump') except right
3738 after a plain `jump', to allow patterns such as:
3741 10: on_failure_jump 3
3742 as used for the *? operator. */
3745 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3752 /* If the first character has to match a backreference, that means
3753 that the group was empty (since it already matched). Since this
3754 is the only case that interests us here, we can assume that the
3755 backreference must match the empty string. */
3760 /* Following are the cases which match a character. These end
3766 int c
= RE_STRING_CHAR (p
+ 1, pend
- p
);
3768 if (SINGLE_BYTE_CHAR_P (c
))
3777 /* We could put all the chars except for \n (and maybe \0)
3778 but we don't bother since it is generally not worth it. */
3779 if (!fastmap
) break;
3784 /* Chars beyond end of bitmap are possible matches.
3785 All the single-byte codes can occur in multibyte buffers.
3786 So any that are not listed in the charset
3787 are possible matches, even in multibyte buffers. */
3788 if (!fastmap
) break;
3789 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3790 j
< (1 << BYTEWIDTH
); j
++)
3794 if (!fastmap
) break;
3795 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3796 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3798 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3801 if ((not && multibyte
)
3802 /* Any character set can possibly contain a character
3803 which doesn't match the specified set of characters. */
3804 || (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3805 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3806 /* If we can match a character class, we can match
3807 any character set. */
3809 set_fastmap_for_multibyte_characters
:
3810 if (match_any_multibyte_characters
== false)
3812 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3813 if (BASE_LEADING_CODE_P (j
))
3815 match_any_multibyte_characters
= true;
3819 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3820 && match_any_multibyte_characters
== false)
3822 /* Set fastmap[I] 1 where I is a base leading code of each
3823 multibyte character in the range table. */
3826 /* Make P points the range table. `+ 2' is to skip flag
3827 bits for a character class. */
3828 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3830 /* Extract the number of ranges in range table into COUNT. */
3831 EXTRACT_NUMBER_AND_INCR (count
, p
);
3832 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3834 /* Extract the start of each range. */
3835 EXTRACT_CHARACTER (c
, p
);
3836 j
= CHAR_CHARSET (c
);
3837 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
3844 if (!fastmap
) break;
3846 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3848 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3849 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3853 /* This match depends on text properties. These end with
3854 aborting optimizations. */
3858 case notcategoryspec
:
3859 if (!fastmap
) break;
3860 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3862 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3863 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3867 /* Any character set can possibly contain a character
3868 whose category is K (or not). */
3869 goto set_fastmap_for_multibyte_characters
;
3872 /* All cases after this match the empty string. These end with
3894 EXTRACT_NUMBER_AND_INCR (j
, p
);
3896 /* Backward jumps can only go back to code that we've already
3897 visited. `re_compile' should make sure this is true. */
3900 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
3902 case on_failure_jump
:
3903 case on_failure_keep_string_jump
:
3904 case on_failure_jump_loop
:
3905 case on_failure_jump_nastyloop
:
3906 case on_failure_jump_smart
:
3912 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
3913 to jump back to "just after here". */
3916 case on_failure_jump
:
3917 case on_failure_keep_string_jump
:
3918 case on_failure_jump_nastyloop
:
3919 case on_failure_jump_loop
:
3920 case on_failure_jump_smart
:
3921 EXTRACT_NUMBER_AND_INCR (j
, p
);
3923 ; /* Backward jump to be ignored. */
3925 { /* We have to look down both arms.
3926 We first go down the "straight" path so as to minimize
3927 stack usage when going through alternatives. */
3928 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
3936 /* This code simply does not properly handle forward jump_n. */
3937 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
3939 /* jump_n can either jump or fall through. The (backward) jump
3940 case has already been handled, so we only need to look at the
3941 fallthrough case. */
3945 /* If N == 0, it should be an on_failure_jump_loop instead. */
3946 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
3948 /* We only care about one iteration of the loop, so we don't
3949 need to consider the case where this behaves like an
3966 abort (); /* We have listed all the cases. */
3969 /* Getting here means we have found the possible starting
3970 characters for one path of the pattern -- and that the empty
3971 string does not match. We need not follow this path further. */
3975 /* We reached the end without matching anything. */
3978 } /* analyse_first */
3980 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3981 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3982 characters can start a string that matches the pattern. This fastmap
3983 is used by re_search to skip quickly over impossible starting points.
3985 Character codes above (1 << BYTEWIDTH) are not represented in the
3986 fastmap, but the leading codes are represented. Thus, the fastmap
3987 indicates which character sets could start a match.
3989 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3990 area as BUFP->fastmap.
3992 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3995 Returns 0 if we succeed, -2 if an internal error. */
3998 re_compile_fastmap (bufp
)
3999 struct re_pattern_buffer
*bufp
;
4001 char *fastmap
= bufp
->fastmap
;
4004 assert (fastmap
&& bufp
->buffer
);
4006 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4007 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4009 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4010 fastmap
, RE_MULTIBYTE_P (bufp
));
4011 bufp
->can_be_null
= (analysis
!= 0);
4013 } /* re_compile_fastmap */
4015 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4016 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4017 this memory for recording register information. STARTS and ENDS
4018 must be allocated using the malloc library routine, and must each
4019 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4021 If NUM_REGS == 0, then subsequent matches should allocate their own
4024 Unless this function is called, the first search or match using
4025 PATTERN_BUFFER will allocate its own register data, without
4026 freeing the old data. */
4029 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4030 struct re_pattern_buffer
*bufp
;
4031 struct re_registers
*regs
;
4033 regoff_t
*starts
, *ends
;
4037 bufp
->regs_allocated
= REGS_REALLOCATE
;
4038 regs
->num_regs
= num_regs
;
4039 regs
->start
= starts
;
4044 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4046 regs
->start
= regs
->end
= (regoff_t
*) 0;
4049 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4051 /* Searching routines. */
4053 /* Like re_search_2, below, but only one string is specified, and
4054 doesn't let you say where to stop matching. */
4057 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4058 struct re_pattern_buffer
*bufp
;
4060 int size
, startpos
, range
;
4061 struct re_registers
*regs
;
4063 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4066 WEAK_ALIAS (__re_search
, re_search
)
4068 /* Head address of virtual concatenation of string. */
4069 #define HEAD_ADDR_VSTRING(P) \
4070 (((P) >= size1 ? string2 : string1))
4072 /* End address of virtual concatenation of string. */
4073 #define STOP_ADDR_VSTRING(P) \
4074 (((P) >= size1 ? string2 + size2 : string1 + size1))
4076 /* Address of POS in the concatenation of virtual string. */
4077 #define POS_ADDR_VSTRING(POS) \
4078 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4080 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4081 virtual concatenation of STRING1 and STRING2, starting first at index
4082 STARTPOS, then at STARTPOS + 1, and so on.
4084 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4086 RANGE is how far to scan while trying to match. RANGE = 0 means try
4087 only at STARTPOS; in general, the last start tried is STARTPOS +
4090 In REGS, return the indices of the virtual concatenation of STRING1
4091 and STRING2 that matched the entire BUFP->buffer and its contained
4094 Do not consider matching one past the index STOP in the virtual
4095 concatenation of STRING1 and STRING2.
4097 We return either the position in the strings at which the match was
4098 found, -1 if no match, or -2 if error (such as failure
4102 re_search_2 (bufp
, str1
, size1
, str2
, size2
, startpos
, range
, regs
, stop
)
4103 struct re_pattern_buffer
*bufp
;
4104 const char *str1
, *str2
;
4108 struct re_registers
*regs
;
4112 re_char
*string1
= (re_char
*) str1
;
4113 re_char
*string2
= (re_char
*) str2
;
4114 register char *fastmap
= bufp
->fastmap
;
4115 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4116 int total_size
= size1
+ size2
;
4117 int endpos
= startpos
+ range
;
4118 boolean anchored_start
;
4120 /* Nonzero if we have to concern multibyte character. */
4121 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4123 /* Check for out-of-range STARTPOS. */
4124 if (startpos
< 0 || startpos
> total_size
)
4127 /* Fix up RANGE if it might eventually take us outside
4128 the virtual concatenation of STRING1 and STRING2.
4129 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4131 range
= 0 - startpos
;
4132 else if (endpos
> total_size
)
4133 range
= total_size
- startpos
;
4135 /* If the search isn't to be a backwards one, don't waste time in a
4136 search for a pattern anchored at beginning of buffer. */
4137 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4146 /* In a forward search for something that starts with \=.
4147 don't keep searching past point. */
4148 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4150 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4156 /* Update the fastmap now if not correct already. */
4157 if (fastmap
&& !bufp
->fastmap_accurate
)
4158 re_compile_fastmap (bufp
);
4160 /* See whether the pattern is anchored. */
4161 anchored_start
= (bufp
->buffer
[0] == begline
);
4164 gl_state
.object
= re_match_object
;
4166 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4168 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4172 /* Loop through the string, looking for a place to start matching. */
4175 /* If the pattern is anchored,
4176 skip quickly past places we cannot match.
4177 We don't bother to treat startpos == 0 specially
4178 because that case doesn't repeat. */
4179 if (anchored_start
&& startpos
> 0)
4181 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4182 : string2
[startpos
- size1
- 1])
4187 /* If a fastmap is supplied, skip quickly over characters that
4188 cannot be the start of a match. If the pattern can match the
4189 null string, however, we don't need to skip characters; we want
4190 the first null string. */
4191 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4193 register re_char
*d
;
4194 register re_wchar_t buf_ch
;
4196 d
= POS_ADDR_VSTRING (startpos
);
4198 if (range
> 0) /* Searching forwards. */
4200 register int lim
= 0;
4203 if (startpos
< size1
&& startpos
+ range
>= size1
)
4204 lim
= range
- (size1
- startpos
);
4206 /* Written out as an if-else to avoid testing `translate'
4208 if (RE_TRANSLATE_P (translate
))
4215 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
4218 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4223 range
-= buf_charlen
;
4228 && !fastmap
[RE_TRANSLATE (translate
, *d
)])
4235 while (range
> lim
&& !fastmap
[*d
])
4241 startpos
+= irange
- range
;
4243 else /* Searching backwards. */
4245 int room
= (startpos
>= size1
4246 ? size2
+ size1
- startpos
4247 : size1
- startpos
);
4248 buf_ch
= RE_STRING_CHAR (d
, room
);
4249 buf_ch
= TRANSLATE (buf_ch
);
4251 if (! (buf_ch
>= 0400
4252 || fastmap
[buf_ch
]))
4257 /* If can't match the null string, and that's all we have left, fail. */
4258 if (range
>= 0 && startpos
== total_size
&& fastmap
4259 && !bufp
->can_be_null
)
4262 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4263 startpos
, regs
, stop
);
4264 #ifndef REGEX_MALLOC
4281 /* Update STARTPOS to the next character boundary. */
4284 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4285 re_char
*pend
= STOP_ADDR_VSTRING (startpos
);
4286 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
4304 /* Update STARTPOS to the previous character boundary. */
4307 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4309 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4311 /* Find the head of multibyte form. */
4312 PREV_CHAR_BOUNDARY (p
, phead
);
4313 range
+= p0
- 1 - p
;
4317 startpos
-= p0
- 1 - p
;
4323 WEAK_ALIAS (__re_search_2
, re_search_2
)
4325 /* Declarations and macros for re_match_2. */
4327 static int bcmp_translate
_RE_ARGS((re_char
*s1
, re_char
*s2
,
4329 RE_TRANSLATE_TYPE translate
,
4330 const int multibyte
));
4332 /* This converts PTR, a pointer into one of the search strings `string1'
4333 and `string2' into an offset from the beginning of that string. */
4334 #define POINTER_TO_OFFSET(ptr) \
4335 (FIRST_STRING_P (ptr) \
4336 ? ((regoff_t) ((ptr) - string1)) \
4337 : ((regoff_t) ((ptr) - string2 + size1)))
4339 /* Call before fetching a character with *d. This switches over to
4340 string2 if necessary.
4341 Check re_match_2_internal for a discussion of why end_match_2 might
4342 not be within string2 (but be equal to end_match_1 instead). */
4343 #define PREFETCH() \
4346 /* End of string2 => fail. */ \
4347 if (dend == end_match_2) \
4349 /* End of string1 => advance to string2. */ \
4351 dend = end_match_2; \
4354 /* Call before fetching a char with *d if you already checked other limits.
4355 This is meant for use in lookahead operations like wordend, etc..
4356 where we might need to look at parts of the string that might be
4357 outside of the LIMITs (i.e past `stop'). */
4358 #define PREFETCH_NOLIMIT() \
4362 dend = end_match_2; \
4365 /* Test if at very beginning or at very end of the virtual concatenation
4366 of `string1' and `string2'. If only one string, it's `string2'. */
4367 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4368 #define AT_STRINGS_END(d) ((d) == end2)
4371 /* Test if D points to a character which is word-constituent. We have
4372 two special cases to check for: if past the end of string1, look at
4373 the first character in string2; and if before the beginning of
4374 string2, look at the last character in string1. */
4375 #define WORDCHAR_P(d) \
4376 (SYNTAX ((d) == end1 ? *string2 \
4377 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4380 /* Disabled due to a compiler bug -- see comment at case wordbound */
4382 /* The comment at case wordbound is following one, but we don't use
4383 AT_WORD_BOUNDARY anymore to support multibyte form.
4385 The DEC Alpha C compiler 3.x generates incorrect code for the
4386 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4387 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4388 macro and introducing temporary variables works around the bug. */
4391 /* Test if the character before D and the one at D differ with respect
4392 to being word-constituent. */
4393 #define AT_WORD_BOUNDARY(d) \
4394 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4395 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4398 /* Free everything we malloc. */
4399 #ifdef MATCH_MAY_ALLOCATE
4400 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4401 # define FREE_VARIABLES() \
4403 REGEX_FREE_STACK (fail_stack.stack); \
4404 FREE_VAR (regstart); \
4405 FREE_VAR (regend); \
4406 FREE_VAR (best_regstart); \
4407 FREE_VAR (best_regend); \
4410 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4411 #endif /* not MATCH_MAY_ALLOCATE */
4414 /* Optimization routines. */
4416 /* If the operation is a match against one or more chars,
4417 return a pointer to the next operation, else return NULL. */
4422 switch (SWITCH_ENUM_CAST (*p
++))
4433 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4436 p
= CHARSET_RANGE_TABLE (p
- 1);
4437 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4438 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4441 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4448 case notcategoryspec
:
4460 /* Jump over non-matching operations. */
4462 skip_noops (p
, pend
)
4468 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4477 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4488 /* Non-zero if "p1 matches something" implies "p2 fails". */
4490 mutually_exclusive_p (bufp
, p1
, p2
)
4491 struct re_pattern_buffer
*bufp
;
4495 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4496 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4498 assert (p1
>= bufp
->buffer
&& p1
< pend
4499 && p2
>= bufp
->buffer
&& p2
<= pend
);
4501 /* Skip over open/close-group commands.
4502 If what follows this loop is a ...+ construct,
4503 look at what begins its body, since we will have to
4504 match at least one of that. */
4505 p2
= skip_noops (p2
, pend
);
4506 /* The same skip can be done for p1, except that this function
4507 is only used in the case where p1 is a simple match operator. */
4508 /* p1 = skip_noops (p1, pend); */
4510 assert (p1
>= bufp
->buffer
&& p1
< pend
4511 && p2
>= bufp
->buffer
&& p2
<= pend
);
4513 op2
= p2
== pend
? succeed
: *p2
;
4515 switch (SWITCH_ENUM_CAST (op2
))
4519 /* If we're at the end of the pattern, we can change. */
4520 if (skip_one_char (p1
))
4522 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4530 register re_wchar_t c
4531 = (re_opcode_t
) *p2
== endline
? '\n'
4532 : RE_STRING_CHAR (p2
+ 2, pend
- p2
- 2);
4534 if ((re_opcode_t
) *p1
== exactn
)
4536 if (c
!= RE_STRING_CHAR (p1
+ 2, pend
- p1
- 2))
4538 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4543 else if ((re_opcode_t
) *p1
== charset
4544 || (re_opcode_t
) *p1
== charset_not
)
4546 int not = (re_opcode_t
) *p1
== charset_not
;
4548 /* Test if C is listed in charset (or charset_not)
4550 if (SINGLE_BYTE_CHAR_P (c
))
4552 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4553 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4556 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4557 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4559 /* `not' is equal to 1 if c would match, which means
4560 that we can't change to pop_failure_jump. */
4563 DEBUG_PRINT1 (" No match => fast loop.\n");
4567 else if ((re_opcode_t
) *p1
== anychar
4570 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4578 if ((re_opcode_t
) *p1
== exactn
)
4579 /* Reuse the code above. */
4580 return mutually_exclusive_p (bufp
, p2
, p1
);
4582 /* It is hard to list up all the character in charset
4583 P2 if it includes multibyte character. Give up in
4585 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4587 /* Now, we are sure that P2 has no range table.
4588 So, for the size of bitmap in P2, `p2[1]' is
4589 enough. But P1 may have range table, so the
4590 size of bitmap table of P1 is extracted by
4591 using macro `CHARSET_BITMAP_SIZE'.
4593 Since we know that all the character listed in
4594 P2 is ASCII, it is enough to test only bitmap
4597 if ((re_opcode_t
) *p1
== charset
)
4600 /* We win if the charset inside the loop
4601 has no overlap with the one after the loop. */
4604 && idx
< CHARSET_BITMAP_SIZE (p1
));
4606 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4610 || idx
== CHARSET_BITMAP_SIZE (p1
))
4612 DEBUG_PRINT1 (" No match => fast loop.\n");
4616 else if ((re_opcode_t
) *p1
== charset_not
)
4619 /* We win if the charset_not inside the loop lists
4620 every character listed in the charset after. */
4621 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4622 if (! (p2
[2 + idx
] == 0
4623 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4624 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4629 DEBUG_PRINT1 (" No match => fast loop.\n");
4638 switch (SWITCH_ENUM_CAST (*p1
))
4642 /* Reuse the code above. */
4643 return mutually_exclusive_p (bufp
, p2
, p1
);
4645 /* When we have two charset_not, it's very unlikely that
4646 they don't overlap. The union of the two sets of excluded
4647 chars should cover all possible chars, which, as a matter of
4648 fact, is virtually impossible in multibyte buffers. */
4654 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4656 return ((re_opcode_t
) *p1
== syntaxspec
4657 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4659 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4662 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4664 return ((re_opcode_t
) *p1
== notsyntaxspec
4665 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4667 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4670 return (((re_opcode_t
) *p1
== notsyntaxspec
4671 || (re_opcode_t
) *p1
== syntaxspec
)
4676 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4677 case notcategoryspec
:
4678 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4690 /* Matching routines. */
4692 #ifndef emacs /* Emacs never uses this. */
4693 /* re_match is like re_match_2 except it takes only a single string. */
4696 re_match (bufp
, string
, size
, pos
, regs
)
4697 struct re_pattern_buffer
*bufp
;
4700 struct re_registers
*regs
;
4702 int result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
, size
,
4704 # if defined C_ALLOCA && !defined REGEX_MALLOC
4709 WEAK_ALIAS (__re_match
, re_match
)
4710 #endif /* not emacs */
4713 /* In Emacs, this is the string or buffer in which we
4714 are matching. It is used for looking up syntax properties. */
4715 Lisp_Object re_match_object
;
4718 /* re_match_2 matches the compiled pattern in BUFP against the
4719 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4720 and SIZE2, respectively). We start matching at POS, and stop
4723 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4724 store offsets for the substring each group matched in REGS. See the
4725 documentation for exactly how many groups we fill.
4727 We return -1 if no match, -2 if an internal error (such as the
4728 failure stack overflowing). Otherwise, we return the length of the
4729 matched substring. */
4732 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4733 struct re_pattern_buffer
*bufp
;
4734 const char *string1
, *string2
;
4737 struct re_registers
*regs
;
4744 gl_state
.object
= re_match_object
;
4745 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4746 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4749 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4750 (re_char
*) string2
, size2
,
4752 #if defined C_ALLOCA && !defined REGEX_MALLOC
4757 WEAK_ALIAS (__re_match_2
, re_match_2
)
4759 /* This is a separate function so that we can force an alloca cleanup
4762 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4763 struct re_pattern_buffer
*bufp
;
4764 re_char
*string1
, *string2
;
4767 struct re_registers
*regs
;
4770 /* General temporaries. */
4775 /* Just past the end of the corresponding string. */
4776 re_char
*end1
, *end2
;
4778 /* Pointers into string1 and string2, just past the last characters in
4779 each to consider matching. */
4780 re_char
*end_match_1
, *end_match_2
;
4782 /* Where we are in the data, and the end of the current string. */
4785 /* Used sometimes to remember where we were before starting matching
4786 an operator so that we can go back in case of failure. This "atomic"
4787 behavior of matching opcodes is indispensable to the correctness
4788 of the on_failure_keep_string_jump optimization. */
4791 /* Where we are in the pattern, and the end of the pattern. */
4792 re_char
*p
= bufp
->buffer
;
4793 re_char
*pend
= p
+ bufp
->used
;
4795 /* We use this to map every character in the string. */
4796 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4798 /* Nonzero if we have to concern multibyte character. */
4799 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4801 /* Failure point stack. Each place that can handle a failure further
4802 down the line pushes a failure point on this stack. It consists of
4803 regstart, and regend for all registers corresponding to
4804 the subexpressions we're currently inside, plus the number of such
4805 registers, and, finally, two char *'s. The first char * is where
4806 to resume scanning the pattern; the second one is where to resume
4807 scanning the strings. */
4808 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4809 fail_stack_type fail_stack
;
4812 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4815 #if defined REL_ALLOC && defined REGEX_MALLOC
4816 /* This holds the pointer to the failure stack, when
4817 it is allocated relocatably. */
4818 fail_stack_elt_t
*failure_stack_ptr
;
4821 /* We fill all the registers internally, independent of what we
4822 return, for use in backreferences. The number here includes
4823 an element for register zero. */
4824 size_t num_regs
= bufp
->re_nsub
+ 1;
4826 /* Information on the contents of registers. These are pointers into
4827 the input strings; they record just what was matched (on this
4828 attempt) by a subexpression part of the pattern, that is, the
4829 regnum-th regstart pointer points to where in the pattern we began
4830 matching and the regnum-th regend points to right after where we
4831 stopped matching the regnum-th subexpression. (The zeroth register
4832 keeps track of what the whole pattern matches.) */
4833 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4834 re_char
**regstart
, **regend
;
4837 /* The following record the register info as found in the above
4838 variables when we find a match better than any we've seen before.
4839 This happens as we backtrack through the failure points, which in
4840 turn happens only if we have not yet matched the entire string. */
4841 unsigned best_regs_set
= false;
4842 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4843 re_char
**best_regstart
, **best_regend
;
4846 /* Logically, this is `best_regend[0]'. But we don't want to have to
4847 allocate space for that if we're not allocating space for anything
4848 else (see below). Also, we never need info about register 0 for
4849 any of the other register vectors, and it seems rather a kludge to
4850 treat `best_regend' differently than the rest. So we keep track of
4851 the end of the best match so far in a separate variable. We
4852 initialize this to NULL so that when we backtrack the first time
4853 and need to test it, it's not garbage. */
4854 re_char
*match_end
= NULL
;
4857 /* Counts the total number of registers pushed. */
4858 unsigned num_regs_pushed
= 0;
4861 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4865 #ifdef MATCH_MAY_ALLOCATE
4866 /* Do not bother to initialize all the register variables if there are
4867 no groups in the pattern, as it takes a fair amount of time. If
4868 there are groups, we include space for register 0 (the whole
4869 pattern), even though we never use it, since it simplifies the
4870 array indexing. We should fix this. */
4873 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4874 regend
= REGEX_TALLOC (num_regs
, re_char
*);
4875 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
4876 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
4878 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
4886 /* We must initialize all our variables to NULL, so that
4887 `FREE_VARIABLES' doesn't try to free them. */
4888 regstart
= regend
= best_regstart
= best_regend
= NULL
;
4890 #endif /* MATCH_MAY_ALLOCATE */
4892 /* The starting position is bogus. */
4893 if (pos
< 0 || pos
> size1
+ size2
)
4899 /* Initialize subexpression text positions to -1 to mark ones that no
4900 start_memory/stop_memory has been seen for. Also initialize the
4901 register information struct. */
4902 for (reg
= 1; reg
< num_regs
; reg
++)
4903 regstart
[reg
] = regend
[reg
] = NULL
;
4905 /* We move `string1' into `string2' if the latter's empty -- but not if
4906 `string1' is null. */
4907 if (size2
== 0 && string1
!= NULL
)
4914 end1
= string1
+ size1
;
4915 end2
= string2
+ size2
;
4917 /* `p' scans through the pattern as `d' scans through the data.
4918 `dend' is the end of the input string that `d' points within. `d'
4919 is advanced into the following input string whenever necessary, but
4920 this happens before fetching; therefore, at the beginning of the
4921 loop, `d' can be pointing at the end of a string, but it cannot
4925 /* Only match within string2. */
4926 d
= string2
+ pos
- size1
;
4927 dend
= end_match_2
= string2
+ stop
- size1
;
4928 end_match_1
= end1
; /* Just to give it a value. */
4934 /* Only match within string1. */
4935 end_match_1
= string1
+ stop
;
4937 When we reach end_match_1, PREFETCH normally switches to string2.
4938 But in the present case, this means that just doing a PREFETCH
4939 makes us jump from `stop' to `gap' within the string.
4940 What we really want here is for the search to stop as
4941 soon as we hit end_match_1. That's why we set end_match_2
4942 to end_match_1 (since PREFETCH fails as soon as we hit
4944 end_match_2
= end_match_1
;
4947 { /* It's important to use this code when stop == size so that
4948 moving `d' from end1 to string2 will not prevent the d == dend
4949 check from catching the end of string. */
4951 end_match_2
= string2
+ stop
- size1
;
4957 DEBUG_PRINT1 ("The compiled pattern is: ");
4958 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4959 DEBUG_PRINT1 ("The string to match is: `");
4960 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4961 DEBUG_PRINT1 ("'\n");
4963 /* This loops over pattern commands. It exits by returning from the
4964 function if the match is complete, or it drops through if the match
4965 fails at this starting point in the input data. */
4968 DEBUG_PRINT2 ("\n%p: ", p
);
4971 { /* End of pattern means we might have succeeded. */
4972 DEBUG_PRINT1 ("end of pattern ... ");
4974 /* If we haven't matched the entire string, and we want the
4975 longest match, try backtracking. */
4976 if (d
!= end_match_2
)
4978 /* 1 if this match ends in the same string (string1 or string2)
4979 as the best previous match. */
4980 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4981 == FIRST_STRING_P (d
));
4982 /* 1 if this match is the best seen so far. */
4983 boolean best_match_p
;
4985 /* AIX compiler got confused when this was combined
4986 with the previous declaration. */
4988 best_match_p
= d
> match_end
;
4990 best_match_p
= !FIRST_STRING_P (d
);
4992 DEBUG_PRINT1 ("backtracking.\n");
4994 if (!FAIL_STACK_EMPTY ())
4995 { /* More failure points to try. */
4997 /* If exceeds best match so far, save it. */
4998 if (!best_regs_set
|| best_match_p
)
5000 best_regs_set
= true;
5003 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5005 for (reg
= 1; reg
< num_regs
; reg
++)
5007 best_regstart
[reg
] = regstart
[reg
];
5008 best_regend
[reg
] = regend
[reg
];
5014 /* If no failure points, don't restore garbage. And if
5015 last match is real best match, don't restore second
5017 else if (best_regs_set
&& !best_match_p
)
5020 /* Restore best match. It may happen that `dend ==
5021 end_match_1' while the restored d is in string2.
5022 For example, the pattern `x.*y.*z' against the
5023 strings `x-' and `y-z-', if the two strings are
5024 not consecutive in memory. */
5025 DEBUG_PRINT1 ("Restoring best registers.\n");
5028 dend
= ((d
>= string1
&& d
<= end1
)
5029 ? end_match_1
: end_match_2
);
5031 for (reg
= 1; reg
< num_regs
; reg
++)
5033 regstart
[reg
] = best_regstart
[reg
];
5034 regend
[reg
] = best_regend
[reg
];
5037 } /* d != end_match_2 */
5040 DEBUG_PRINT1 ("Accepting match.\n");
5042 /* If caller wants register contents data back, do it. */
5043 if (regs
&& !bufp
->no_sub
)
5045 /* Have the register data arrays been allocated? */
5046 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5047 { /* No. So allocate them with malloc. We need one
5048 extra element beyond `num_regs' for the `-1' marker
5050 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5051 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5052 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5053 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5058 bufp
->regs_allocated
= REGS_REALLOCATE
;
5060 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5061 { /* Yes. If we need more elements than were already
5062 allocated, reallocate them. If we need fewer, just
5064 if (regs
->num_regs
< num_regs
+ 1)
5066 regs
->num_regs
= num_regs
+ 1;
5067 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5068 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5069 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5078 /* These braces fend off a "empty body in an else-statement"
5079 warning under GCC when assert expands to nothing. */
5080 assert (bufp
->regs_allocated
== REGS_FIXED
);
5083 /* Convert the pointer data in `regstart' and `regend' to
5084 indices. Register zero has to be set differently,
5085 since we haven't kept track of any info for it. */
5086 if (regs
->num_regs
> 0)
5088 regs
->start
[0] = pos
;
5089 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5092 /* Go through the first `min (num_regs, regs->num_regs)'
5093 registers, since that is all we initialized. */
5094 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5096 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5097 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5101 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5103 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5107 /* If the regs structure we return has more elements than
5108 were in the pattern, set the extra elements to -1. If
5109 we (re)allocated the registers, this is the case,
5110 because we always allocate enough to have at least one
5112 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5113 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5114 } /* regs && !bufp->no_sub */
5116 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5117 nfailure_points_pushed
, nfailure_points_popped
,
5118 nfailure_points_pushed
- nfailure_points_popped
);
5119 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5121 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5123 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5129 /* Otherwise match next pattern command. */
5130 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5132 /* Ignore these. Used to ignore the n of succeed_n's which
5133 currently have n == 0. */
5135 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5139 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5142 /* Match the next n pattern characters exactly. The following
5143 byte in the pattern defines n, and the n bytes after that
5144 are the characters to match. */
5147 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5149 /* Remember the start point to rollback upon failure. */
5152 /* This is written out as an if-else so we don't waste time
5153 testing `translate' inside the loop. */
5154 if (RE_TRANSLATE_P (translate
))
5159 int pat_charlen
, buf_charlen
;
5160 unsigned int pat_ch
, buf_ch
;
5163 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
5164 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5166 if (RE_TRANSLATE (translate
, buf_ch
)
5175 mcnt
-= pat_charlen
;
5182 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5207 /* Match any character except possibly a newline or a null. */
5213 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5216 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
5217 buf_ch
= TRANSLATE (buf_ch
);
5219 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5221 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5222 && buf_ch
== '\000'))
5225 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5234 register unsigned int c
;
5235 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5238 /* Start of actual range_table, or end of bitmap if there is no
5240 re_char
*range_table
;
5242 /* Nonzero if there is a range table. */
5243 int range_table_exists
;
5245 /* Number of ranges of range table. This is not included
5246 in the initial byte-length of the command. */
5249 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5251 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5253 if (range_table_exists
)
5255 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5256 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5260 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5261 c
= TRANSLATE (c
); /* The character to match. */
5263 if (SINGLE_BYTE_CHAR_P (c
))
5264 { /* Lookup bitmap. */
5265 /* Cast to `unsigned' instead of `unsigned char' in
5266 case the bit list is a full 32 bytes long. */
5267 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5268 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5272 else if (range_table_exists
)
5274 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5276 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5277 | (class_bits
& BIT_MULTIBYTE
)
5278 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5279 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5280 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5281 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5284 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5288 if (range_table_exists
)
5289 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5291 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5293 if (!not) goto fail
;
5300 /* The beginning of a group is represented by start_memory.
5301 The argument is the register number. The text
5302 matched within the group is recorded (in the internal
5303 registers data structure) under the register number. */
5305 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5307 /* In case we need to undo this operation (via backtracking). */
5308 PUSH_FAILURE_REG ((unsigned int)*p
);
5311 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5312 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5314 /* Move past the register number and inner group count. */
5319 /* The stop_memory opcode represents the end of a group. Its
5320 argument is the same as start_memory's: the register number. */
5322 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5324 assert (!REG_UNSET (regstart
[*p
]));
5325 /* Strictly speaking, there should be code such as:
5327 assert (REG_UNSET (regend[*p]));
5328 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5330 But the only info to be pushed is regend[*p] and it is known to
5331 be UNSET, so there really isn't anything to push.
5332 Not pushing anything, on the other hand deprives us from the
5333 guarantee that regend[*p] is UNSET since undoing this operation
5334 will not reset its value properly. This is not important since
5335 the value will only be read on the next start_memory or at
5336 the very end and both events can only happen if this stop_memory
5340 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5342 /* Move past the register number and the inner group count. */
5347 /* \<digit> has been turned into a `duplicate' command which is
5348 followed by the numeric value of <digit> as the register number. */
5351 register re_char
*d2
, *dend2
;
5352 int regno
= *p
++; /* Get which register to match against. */
5353 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5355 /* Can't back reference a group which we've never matched. */
5356 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5359 /* Where in input to try to start matching. */
5360 d2
= regstart
[regno
];
5362 /* Remember the start point to rollback upon failure. */
5365 /* Where to stop matching; if both the place to start and
5366 the place to stop matching are in the same string, then
5367 set to the place to stop, otherwise, for now have to use
5368 the end of the first string. */
5370 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5371 == FIRST_STRING_P (regend
[regno
]))
5372 ? regend
[regno
] : end_match_1
);
5375 /* If necessary, advance to next segment in register
5379 if (dend2
== end_match_2
) break;
5380 if (dend2
== regend
[regno
]) break;
5382 /* End of string1 => advance to string2. */
5384 dend2
= regend
[regno
];
5386 /* At end of register contents => success */
5387 if (d2
== dend2
) break;
5389 /* If necessary, advance to next segment in data. */
5392 /* How many characters left in this segment to match. */
5395 /* Want how many consecutive characters we can match in
5396 one shot, so, if necessary, adjust the count. */
5397 if (mcnt
> dend2
- d2
)
5400 /* Compare that many; failure if mismatch, else move
5402 if (RE_TRANSLATE_P (translate
)
5403 ? bcmp_translate (d
, d2
, mcnt
, translate
, multibyte
)
5404 : memcmp (d
, d2
, mcnt
))
5409 d
+= mcnt
, d2
+= mcnt
;
5415 /* begline matches the empty string at the beginning of the string
5416 (unless `not_bol' is set in `bufp'), and after newlines. */
5418 DEBUG_PRINT1 ("EXECUTING begline.\n");
5420 if (AT_STRINGS_BEG (d
))
5422 if (!bufp
->not_bol
) break;
5427 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5431 /* In all other cases, we fail. */
5435 /* endline is the dual of begline. */
5437 DEBUG_PRINT1 ("EXECUTING endline.\n");
5439 if (AT_STRINGS_END (d
))
5441 if (!bufp
->not_eol
) break;
5445 PREFETCH_NOLIMIT ();
5452 /* Match at the very beginning of the data. */
5454 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5455 if (AT_STRINGS_BEG (d
))
5460 /* Match at the very end of the data. */
5462 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5463 if (AT_STRINGS_END (d
))
5468 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5469 pushes NULL as the value for the string on the stack. Then
5470 `POP_FAILURE_POINT' will keep the current value for the
5471 string, instead of restoring it. To see why, consider
5472 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5473 then the . fails against the \n. But the next thing we want
5474 to do is match the \n against the \n; if we restored the
5475 string value, we would be back at the foo.
5477 Because this is used only in specific cases, we don't need to
5478 check all the things that `on_failure_jump' does, to make
5479 sure the right things get saved on the stack. Hence we don't
5480 share its code. The only reason to push anything on the
5481 stack at all is that otherwise we would have to change
5482 `anychar's code to do something besides goto fail in this
5483 case; that seems worse than this. */
5484 case on_failure_keep_string_jump
:
5485 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5486 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5489 PUSH_FAILURE_POINT (p
- 3, NULL
);
5492 /* A nasty loop is introduced by the non-greedy *? and +?.
5493 With such loops, the stack only ever contains one failure point
5494 at a time, so that a plain on_failure_jump_loop kind of
5495 cycle detection cannot work. Worse yet, such a detection
5496 can not only fail to detect a cycle, but it can also wrongly
5497 detect a cycle (between different instantiations of the same
5499 So the method used for those nasty loops is a little different:
5500 We use a special cycle-detection-stack-frame which is pushed
5501 when the on_failure_jump_nastyloop failure-point is *popped*.
5502 This special frame thus marks the beginning of one iteration
5503 through the loop and we can hence easily check right here
5504 whether something matched between the beginning and the end of
5506 case on_failure_jump_nastyloop
:
5507 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5508 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5511 assert ((re_opcode_t
)p
[-4] == no_op
);
5514 CHECK_INFINITE_LOOP (p
- 4, d
);
5516 /* If there's a cycle, just continue without pushing
5517 this failure point. The failure point is the "try again"
5518 option, which shouldn't be tried.
5519 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5520 PUSH_FAILURE_POINT (p
- 3, d
);
5524 /* Simple loop detecting on_failure_jump: just check on the
5525 failure stack if the same spot was already hit earlier. */
5526 case on_failure_jump_loop
:
5528 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5529 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5533 CHECK_INFINITE_LOOP (p
- 3, d
);
5535 /* If there's a cycle, get out of the loop, as if the matching
5536 had failed. We used to just `goto fail' here, but that was
5537 aborting the search a bit too early: we want to keep the
5538 empty-loop-match and keep matching after the loop.
5539 We want (x?)*y\1z to match both xxyz and xxyxz. */
5542 PUSH_FAILURE_POINT (p
- 3, d
);
5547 /* Uses of on_failure_jump:
5549 Each alternative starts with an on_failure_jump that points
5550 to the beginning of the next alternative. Each alternative
5551 except the last ends with a jump that in effect jumps past
5552 the rest of the alternatives. (They really jump to the
5553 ending jump of the following alternative, because tensioning
5554 these jumps is a hassle.)
5556 Repeats start with an on_failure_jump that points past both
5557 the repetition text and either the following jump or
5558 pop_failure_jump back to this on_failure_jump. */
5559 case on_failure_jump
:
5560 IMMEDIATE_QUIT_CHECK
;
5561 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5562 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5565 PUSH_FAILURE_POINT (p
-3, d
);
5568 /* This operation is used for greedy *.
5569 Compare the beginning of the repeat with what in the
5570 pattern follows its end. If we can establish that there
5571 is nothing that they would both match, i.e., that we
5572 would have to backtrack because of (as in, e.g., `a*a')
5573 then we can use a non-backtracking loop based on
5574 on_failure_keep_string_jump instead of on_failure_jump. */
5575 case on_failure_jump_smart
:
5576 IMMEDIATE_QUIT_CHECK
;
5577 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5578 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5581 re_char
*p1
= p
; /* Next operation. */
5582 /* Here, we discard `const', making re_match non-reentrant. */
5583 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5584 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5586 p
-= 3; /* Reset so that we will re-execute the
5587 instruction once it's been changed. */
5589 EXTRACT_NUMBER (mcnt
, p2
- 2);
5591 /* Ensure this is a indeed the trivial kind of loop
5592 we are expecting. */
5593 assert (skip_one_char (p1
) == p2
- 3);
5594 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5595 DEBUG_STATEMENT (debug
+= 2);
5596 if (mutually_exclusive_p (bufp
, p1
, p2
))
5598 /* Use a fast `on_failure_keep_string_jump' loop. */
5599 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5600 *p3
= (unsigned char) on_failure_keep_string_jump
;
5601 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5605 /* Default to a safe `on_failure_jump' loop. */
5606 DEBUG_PRINT1 (" smart default => slow loop.\n");
5607 *p3
= (unsigned char) on_failure_jump
;
5609 DEBUG_STATEMENT (debug
-= 2);
5613 /* Unconditionally jump (without popping any failure points). */
5616 IMMEDIATE_QUIT_CHECK
;
5617 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5618 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5619 p
+= mcnt
; /* Do the jump. */
5620 DEBUG_PRINT2 ("(to %p).\n", p
);
5624 /* Have to succeed matching what follows at least n times.
5625 After that, handle like `on_failure_jump'. */
5627 /* Signedness doesn't matter since we only compare MCNT to 0. */
5628 EXTRACT_NUMBER (mcnt
, p
+ 2);
5629 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5631 /* Originally, mcnt is how many times we HAVE to succeed. */
5634 /* Here, we discard `const', making re_match non-reentrant. */
5635 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5638 PUSH_NUMBER (p2
, mcnt
);
5641 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5646 /* Signedness doesn't matter since we only compare MCNT to 0. */
5647 EXTRACT_NUMBER (mcnt
, p
+ 2);
5648 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5650 /* Originally, this is how many times we CAN jump. */
5653 /* Here, we discard `const', making re_match non-reentrant. */
5654 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5656 PUSH_NUMBER (p2
, mcnt
);
5657 goto unconditional_jump
;
5659 /* If don't have to jump any more, skip over the rest of command. */
5666 unsigned char *p2
; /* Location of the counter. */
5667 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5669 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5670 /* Here, we discard `const', making re_match non-reentrant. */
5671 p2
= (unsigned char*) p
+ mcnt
;
5672 /* Signedness doesn't matter since we only copy MCNT's bits . */
5673 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5674 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5675 PUSH_NUMBER (p2
, mcnt
);
5681 not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5682 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5684 /* We SUCCEED (or FAIL) in one of the following cases: */
5686 /* Case 1: D is at the beginning or the end of string. */
5687 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5691 /* C1 is the character before D, S1 is the syntax of C1, C2
5692 is the character at D, and S2 is the syntax of C2. */
5696 int offset
= PTR_TO_OFFSET (d
- 1);
5697 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5698 UPDATE_SYNTAX_TABLE (charpos
);
5700 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5703 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5705 PREFETCH_NOLIMIT ();
5706 c2
= RE_STRING_CHAR (d
, dend
- d
);
5709 if (/* Case 2: Only one of S1 and S2 is Sword. */
5710 ((s1
== Sword
) != (s2
== Sword
))
5711 /* Case 3: Both of S1 and S2 are Sword, and macro
5712 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5713 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5722 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5724 /* We FAIL in one of the following cases: */
5726 /* Case 1: D is at the end of string. */
5727 if (AT_STRINGS_END (d
))
5731 /* C1 is the character before D, S1 is the syntax of C1, C2
5732 is the character at D, and S2 is the syntax of C2. */
5736 int offset
= PTR_TO_OFFSET (d
);
5737 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5738 UPDATE_SYNTAX_TABLE (charpos
);
5741 c2
= RE_STRING_CHAR (d
, dend
- d
);
5744 /* Case 2: S2 is not Sword. */
5748 /* Case 3: D is not at the beginning of string ... */
5749 if (!AT_STRINGS_BEG (d
))
5751 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5753 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5757 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5759 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5766 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5768 /* We FAIL in one of the following cases: */
5770 /* Case 1: D is at the beginning of string. */
5771 if (AT_STRINGS_BEG (d
))
5775 /* C1 is the character before D, S1 is the syntax of C1, C2
5776 is the character at D, and S2 is the syntax of C2. */
5780 int offset
= PTR_TO_OFFSET (d
) - 1;
5781 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5782 UPDATE_SYNTAX_TABLE (charpos
);
5784 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5787 /* Case 2: S1 is not Sword. */
5791 /* Case 3: D is not at the end of string ... */
5792 if (!AT_STRINGS_END (d
))
5794 PREFETCH_NOLIMIT ();
5795 c2
= RE_STRING_CHAR (d
, dend
- d
);
5797 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
5801 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5803 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5810 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
5812 /* We FAIL in one of the following cases: */
5814 /* Case 1: D is at the end of string. */
5815 if (AT_STRINGS_END (d
))
5819 /* C1 is the character before D, S1 is the syntax of C1, C2
5820 is the character at D, and S2 is the syntax of C2. */
5824 int offset
= PTR_TO_OFFSET (d
);
5825 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5826 UPDATE_SYNTAX_TABLE (charpos
);
5829 c2
= RE_STRING_CHAR (d
, dend
- d
);
5832 /* Case 2: S2 is neither Sword nor Ssymbol. */
5833 if (s2
!= Sword
&& s2
!= Ssymbol
)
5836 /* Case 3: D is not at the beginning of string ... */
5837 if (!AT_STRINGS_BEG (d
))
5839 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5841 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5845 /* ... and S1 is Sword or Ssymbol. */
5846 if (s1
== Sword
|| s1
== Ssymbol
)
5853 DEBUG_PRINT1 ("EXECUTING symend.\n");
5855 /* We FAIL in one of the following cases: */
5857 /* Case 1: D is at the beginning of string. */
5858 if (AT_STRINGS_BEG (d
))
5862 /* C1 is the character before D, S1 is the syntax of C1, C2
5863 is the character at D, and S2 is the syntax of C2. */
5867 int offset
= PTR_TO_OFFSET (d
) - 1;
5868 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5869 UPDATE_SYNTAX_TABLE (charpos
);
5871 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5874 /* Case 2: S1 is neither Ssymbol nor Sword. */
5875 if (s1
!= Sword
&& s1
!= Ssymbol
)
5878 /* Case 3: D is not at the end of string ... */
5879 if (!AT_STRINGS_END (d
))
5881 PREFETCH_NOLIMIT ();
5882 c2
= RE_STRING_CHAR (d
, dend
- d
);
5884 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
5888 /* ... and S2 is Sword or Ssymbol. */
5889 if (s2
== Sword
|| s2
== Ssymbol
)
5897 not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
5899 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
5903 int offset
= PTR_TO_OFFSET (d
);
5904 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5905 UPDATE_SYNTAX_TABLE (pos1
);
5912 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5914 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
5922 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5923 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
5928 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5929 if (PTR_BYTE_POS (d
) != PT_BYTE
)
5934 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5935 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
5940 case notcategoryspec
:
5941 not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
5943 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt
);
5949 c
= RE_STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5951 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
5962 continue; /* Successfully executed one pattern command; keep going. */
5965 /* We goto here if a matching operation fails. */
5967 IMMEDIATE_QUIT_CHECK
;
5968 if (!FAIL_STACK_EMPTY ())
5971 /* A restart point is known. Restore to that state. */
5972 DEBUG_PRINT1 ("\nFAIL:\n");
5973 POP_FAILURE_POINT (str
, pat
);
5974 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
5976 case on_failure_keep_string_jump
:
5977 assert (str
== NULL
);
5978 goto continue_failure_jump
;
5980 case on_failure_jump_nastyloop
:
5981 assert ((re_opcode_t
)pat
[-2] == no_op
);
5982 PUSH_FAILURE_POINT (pat
- 2, str
);
5985 case on_failure_jump_loop
:
5986 case on_failure_jump
:
5989 continue_failure_jump
:
5990 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
5995 /* A special frame used for nastyloops. */
6002 assert (p
>= bufp
->buffer
&& p
<= pend
);
6004 if (d
>= string1
&& d
<= end1
)
6008 break; /* Matching at this starting point really fails. */
6012 goto restore_best_regs
;
6016 return -1; /* Failure to match. */
6019 /* Subroutine definitions for re_match_2. */
6021 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6022 bytes; nonzero otherwise. */
6025 bcmp_translate (s1
, s2
, len
, translate
, multibyte
)
6028 RE_TRANSLATE_TYPE translate
;
6029 const int multibyte
;
6031 register re_char
*p1
= s1
, *p2
= s2
;
6032 re_char
*p1_end
= s1
+ len
;
6033 re_char
*p2_end
= s2
+ len
;
6035 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6036 different lengths, but relying on a single `len' would break this. -sm */
6037 while (p1
< p1_end
&& p2
< p2_end
)
6039 int p1_charlen
, p2_charlen
;
6040 re_wchar_t p1_ch
, p2_ch
;
6042 p1_ch
= RE_STRING_CHAR_AND_LENGTH (p1
, p1_end
- p1
, p1_charlen
);
6043 p2_ch
= RE_STRING_CHAR_AND_LENGTH (p2
, p2_end
- p2
, p2_charlen
);
6045 if (RE_TRANSLATE (translate
, p1_ch
)
6046 != RE_TRANSLATE (translate
, p2_ch
))
6049 p1
+= p1_charlen
, p2
+= p2_charlen
;
6052 if (p1
!= p1_end
|| p2
!= p2_end
)
6058 /* Entry points for GNU code. */
6060 /* re_compile_pattern is the GNU regular expression compiler: it
6061 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6062 Returns 0 if the pattern was valid, otherwise an error string.
6064 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6065 are set in BUFP on entry.
6067 We call regex_compile to do the actual compilation. */
6070 re_compile_pattern (pattern
, length
, bufp
)
6071 const char *pattern
;
6073 struct re_pattern_buffer
*bufp
;
6077 /* GNU code is written to assume at least RE_NREGS registers will be set
6078 (and at least one extra will be -1). */
6079 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6081 /* And GNU code determines whether or not to get register information
6082 by passing null for the REGS argument to re_match, etc., not by
6086 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6090 return gettext (re_error_msgid
[(int) ret
]);
6092 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6094 /* Entry points compatible with 4.2 BSD regex library. We don't define
6095 them unless specifically requested. */
6097 #if defined _REGEX_RE_COMP || defined _LIBC
6099 /* BSD has one and only one pattern buffer. */
6100 static struct re_pattern_buffer re_comp_buf
;
6104 /* Make these definitions weak in libc, so POSIX programs can redefine
6105 these names if they don't use our functions, and still use
6106 regcomp/regexec below without link errors. */
6116 if (!re_comp_buf
.buffer
)
6117 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6118 return (char *) gettext ("No previous regular expression");
6122 if (!re_comp_buf
.buffer
)
6124 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6125 if (re_comp_buf
.buffer
== NULL
)
6126 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6127 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6128 re_comp_buf
.allocated
= 200;
6130 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6131 if (re_comp_buf
.fastmap
== NULL
)
6132 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6133 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6136 /* Since `re_exec' always passes NULL for the `regs' argument, we
6137 don't need to initialize the pattern buffer fields which affect it. */
6139 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6144 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6145 return (char *) gettext (re_error_msgid
[(int) ret
]);
6156 const int len
= strlen (s
);
6158 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6160 #endif /* _REGEX_RE_COMP */
6162 /* POSIX.2 functions. Don't define these for Emacs. */
6166 /* regcomp takes a regular expression as a string and compiles it.
6168 PREG is a regex_t *. We do not expect any fields to be initialized,
6169 since POSIX says we shouldn't. Thus, we set
6171 `buffer' to the compiled pattern;
6172 `used' to the length of the compiled pattern;
6173 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6174 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6175 RE_SYNTAX_POSIX_BASIC;
6176 `fastmap' to an allocated space for the fastmap;
6177 `fastmap_accurate' to zero;
6178 `re_nsub' to the number of subexpressions in PATTERN.
6180 PATTERN is the address of the pattern string.
6182 CFLAGS is a series of bits which affect compilation.
6184 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6185 use POSIX basic syntax.
6187 If REG_NEWLINE is set, then . and [^...] don't match newline.
6188 Also, regexec will try a match beginning after every newline.
6190 If REG_ICASE is set, then we considers upper- and lowercase
6191 versions of letters to be equivalent when matching.
6193 If REG_NOSUB is set, then when PREG is passed to regexec, that
6194 routine will report only success or failure, and nothing about the
6197 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6198 the return codes and their meanings.) */
6201 regcomp (preg
, pattern
, cflags
)
6202 regex_t
*__restrict preg
;
6203 const char *__restrict pattern
;
6208 = (cflags
& REG_EXTENDED
) ?
6209 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6211 /* regex_compile will allocate the space for the compiled pattern. */
6213 preg
->allocated
= 0;
6216 /* Try to allocate space for the fastmap. */
6217 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6219 if (cflags
& REG_ICASE
)
6224 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6225 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6226 if (preg
->translate
== NULL
)
6227 return (int) REG_ESPACE
;
6229 /* Map uppercase characters to corresponding lowercase ones. */
6230 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6231 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6234 preg
->translate
= NULL
;
6236 /* If REG_NEWLINE is set, newlines are treated differently. */
6237 if (cflags
& REG_NEWLINE
)
6238 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6239 syntax
&= ~RE_DOT_NEWLINE
;
6240 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6243 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6245 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6247 /* POSIX says a null character in the pattern terminates it, so we
6248 can use strlen here in compiling the pattern. */
6249 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6251 /* POSIX doesn't distinguish between an unmatched open-group and an
6252 unmatched close-group: both are REG_EPAREN. */
6253 if (ret
== REG_ERPAREN
)
6256 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6257 { /* Compute the fastmap now, since regexec cannot modify the pattern
6259 re_compile_fastmap (preg
);
6260 if (preg
->can_be_null
)
6261 { /* The fastmap can't be used anyway. */
6262 free (preg
->fastmap
);
6263 preg
->fastmap
= NULL
;
6268 WEAK_ALIAS (__regcomp
, regcomp
)
6271 /* regexec searches for a given pattern, specified by PREG, in the
6274 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6275 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6276 least NMATCH elements, and we set them to the offsets of the
6277 corresponding matched substrings.
6279 EFLAGS specifies `execution flags' which affect matching: if
6280 REG_NOTBOL is set, then ^ does not match at the beginning of the
6281 string; if REG_NOTEOL is set, then $ does not match at the end.
6283 We return 0 if we find a match and REG_NOMATCH if not. */
6286 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6287 const regex_t
*__restrict preg
;
6288 const char *__restrict string
;
6290 regmatch_t pmatch
[__restrict_arr
];
6294 struct re_registers regs
;
6295 regex_t private_preg
;
6296 int len
= strlen (string
);
6297 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6299 private_preg
= *preg
;
6301 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6302 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6304 /* The user has told us exactly how many registers to return
6305 information about, via `nmatch'. We have to pass that on to the
6306 matching routines. */
6307 private_preg
.regs_allocated
= REGS_FIXED
;
6311 regs
.num_regs
= nmatch
;
6312 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6313 if (regs
.start
== NULL
)
6314 return (int) REG_NOMATCH
;
6315 regs
.end
= regs
.start
+ nmatch
;
6318 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6319 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6320 was a little bit longer but still only matching the real part.
6321 This works because the `endline' will check for a '\n' and will find a
6322 '\0', correctly deciding that this is not the end of a line.
6323 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6324 a convenient '\0' there. For all we know, the string could be preceded
6325 by '\n' which would throw things off. */
6327 /* Perform the searching operation. */
6328 ret
= re_search (&private_preg
, string
, len
,
6329 /* start: */ 0, /* range: */ len
,
6330 want_reg_info
? ®s
: (struct re_registers
*) 0);
6332 /* Copy the register information to the POSIX structure. */
6339 for (r
= 0; r
< nmatch
; r
++)
6341 pmatch
[r
].rm_so
= regs
.start
[r
];
6342 pmatch
[r
].rm_eo
= regs
.end
[r
];
6346 /* If we needed the temporary register info, free the space now. */
6350 /* We want zero return to mean success, unlike `re_search'. */
6351 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6353 WEAK_ALIAS (__regexec
, regexec
)
6356 /* Returns a message corresponding to an error code, ERRCODE, returned
6357 from either regcomp or regexec. We don't use PREG here. */
6360 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6362 const regex_t
*preg
;
6370 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6371 /* Only error codes returned by the rest of the code should be passed
6372 to this routine. If we are given anything else, or if other regex
6373 code generates an invalid error code, then the program has a bug.
6374 Dump core so we can fix it. */
6377 msg
= gettext (re_error_msgid
[errcode
]);
6379 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6381 if (errbuf_size
!= 0)
6383 if (msg_size
> errbuf_size
)
6385 strncpy (errbuf
, msg
, errbuf_size
- 1);
6386 errbuf
[errbuf_size
- 1] = 0;
6389 strcpy (errbuf
, msg
);
6394 WEAK_ALIAS (__regerror
, regerror
)
6397 /* Free dynamically allocated space used by PREG. */
6403 if (preg
->buffer
!= NULL
)
6404 free (preg
->buffer
);
6405 preg
->buffer
= NULL
;
6407 preg
->allocated
= 0;
6410 if (preg
->fastmap
!= NULL
)
6411 free (preg
->fastmap
);
6412 preg
->fastmap
= NULL
;
6413 preg
->fastmap_accurate
= 0;
6415 if (preg
->translate
!= NULL
)
6416 free (preg
->translate
);
6417 preg
->translate
= NULL
;
6419 WEAK_ALIAS (__regfree
, regfree
)
6421 #endif /* not emacs */
6423 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6424 (do not change this comment) */