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) 1993-2013 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
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, see <http://www.gnu.org/licenses/>. */
21 - structure the opcode space into opcode+flag.
22 - merge with glibc's regex.[ch].
23 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
24 need to modify the compiled regexp so that re_match can be reentrant.
25 - get rid of on_failure_jump_smart by doing the optimization in re_comp
26 rather than at run-time, so that re_match can be reentrant.
29 /* AIX requires this to be the first thing in the file. */
30 #if defined _AIX && !defined REGEX_MALLOC
34 /* Ignore some GCC warnings for now. This section should go away
35 once the Emacs and Gnulib regex code is merged. */
36 #if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5)
37 # pragma GCC diagnostic ignored "-Wstrict-overflow"
39 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
40 # pragma GCC diagnostic ignored "-Wunused-function"
41 # pragma GCC diagnostic ignored "-Wunused-macros"
42 # pragma GCC diagnostic ignored "-Wunused-result"
43 # pragma GCC diagnostic ignored "-Wunused-variable"
52 /* We need this for `regex.h', and perhaps for the Emacs include files. */
53 # include <sys/types.h>
56 /* Whether to use ISO C Amendment 1 wide char functions.
57 Those should not be used for Emacs since it uses its own. */
59 #define WIDE_CHAR_SUPPORT 1
61 #define WIDE_CHAR_SUPPORT \
62 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
65 /* For platform which support the ISO C amendment 1 functionality we
66 support user defined character classes. */
68 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
74 /* We have to keep the namespace clean. */
75 # define regfree(preg) __regfree (preg)
76 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
77 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
78 # define regerror(err_code, preg, errbuf, errbuf_size) \
79 __regerror (err_code, preg, errbuf, errbuf_size)
80 # define re_set_registers(bu, re, nu, st, en) \
81 __re_set_registers (bu, re, nu, st, en)
82 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
83 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
84 # define re_match(bufp, string, size, pos, regs) \
85 __re_match (bufp, string, size, pos, regs)
86 # define re_search(bufp, string, size, startpos, range, regs) \
87 __re_search (bufp, string, size, startpos, range, regs)
88 # define re_compile_pattern(pattern, length, bufp) \
89 __re_compile_pattern (pattern, length, bufp)
90 # define re_set_syntax(syntax) __re_set_syntax (syntax)
91 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
92 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
93 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
95 /* Make sure we call libc's function even if the user overrides them. */
96 # define btowc __btowc
97 # define iswctype __iswctype
98 # define wctype __wctype
100 # define WEAK_ALIAS(a,b) weak_alias (a, b)
102 /* We are also using some library internals. */
103 # include <locale/localeinfo.h>
104 # include <locale/elem-hash.h>
105 # include <langinfo.h>
107 # define WEAK_ALIAS(a,b)
110 /* This is for other GNU distributions with internationalized messages. */
111 #if HAVE_LIBINTL_H || defined _LIBC
112 # include <libintl.h>
114 # define gettext(msgid) (msgid)
118 /* This define is so xgettext can find the internationalizable
120 # define gettext_noop(String) String
123 /* The `emacs' switch turns on certain matching commands
124 that make sense only in Emacs. */
128 # include "character.h"
131 /* Make syntax table lookup grant data in gl_state. */
132 # define SYNTAX_ENTRY_VIA_PROPERTY
135 # include "category.h"
140 # define malloc xmalloc
144 # define realloc xrealloc
150 /* Converts the pointer to the char to BEG-based offset from the start. */
151 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
152 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
154 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
155 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
156 # define RE_STRING_CHAR(p, multibyte) \
157 (multibyte ? (STRING_CHAR (p)) : (*(p)))
158 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
159 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
161 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
163 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
165 /* Set C a (possibly converted to multibyte) character before P. P
166 points into a string which is the virtual concatenation of STR1
167 (which ends at END1) or STR2 (which ends at END2). */
168 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
170 if (target_multibyte) \
172 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
173 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
174 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
175 c = STRING_CHAR (dtemp); \
179 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
180 (c) = RE_CHAR_TO_MULTIBYTE (c); \
184 /* Set C a (possibly converted to multibyte) character at P, and set
185 LEN to the byte length of that character. */
186 # define GET_CHAR_AFTER(c, p, len) \
188 if (target_multibyte) \
189 (c) = STRING_CHAR_AND_LENGTH (p, len); \
194 (c) = RE_CHAR_TO_MULTIBYTE (c); \
198 #else /* not emacs */
200 /* If we are not linking with Emacs proper,
201 we can't use the relocating allocator
202 even if config.h says that we can. */
207 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
210 xmalloc (size_t size
)
212 void *val
= malloc (size
);
215 write (2, "virtual memory exhausted\n", 25);
222 xrealloc (void *block
, size_t size
)
225 /* We must call malloc explicitly when BLOCK is 0, since some
226 reallocs don't do this. */
230 val
= realloc (block
, size
);
233 write (2, "virtual memory exhausted\n", 25);
242 # define malloc xmalloc
246 # define realloc xrealloc
248 # include <stdbool.h>
251 /* Define the syntax stuff for \<, \>, etc. */
253 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
254 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
256 /* Dummy macros for non-Emacs environments. */
257 # define CHAR_CHARSET(c) 0
258 # define CHARSET_LEADING_CODE_BASE(c) 0
259 # define MAX_MULTIBYTE_LENGTH 1
260 # define RE_MULTIBYTE_P(x) 0
261 # define RE_TARGET_MULTIBYTE_P(x) 0
262 # define WORD_BOUNDARY_P(c1, c2) (0)
263 # define CHAR_HEAD_P(p) (1)
264 # define SINGLE_BYTE_CHAR_P(c) (1)
265 # define SAME_CHARSET_P(c1, c2) (1)
266 # define BYTES_BY_CHAR_HEAD(p) (1)
267 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
268 # define STRING_CHAR(p) (*(p))
269 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
270 # define CHAR_STRING(c, s) (*(s) = (c), 1)
271 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
272 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
273 # define RE_CHAR_TO_MULTIBYTE(c) (c)
274 # define RE_CHAR_TO_UNIBYTE(c) (c)
275 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
276 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
277 # define GET_CHAR_AFTER(c, p, len) \
279 # define MAKE_CHAR(charset, c1, c2) (c1)
280 # define BYTE8_TO_CHAR(c) (c)
281 # define CHAR_BYTE8_P(c) (0)
282 # define CHAR_LEADING_CODE(c) (c)
284 #endif /* not emacs */
287 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
288 # define RE_TRANSLATE_P(TBL) (TBL)
291 /* Get the interface, including the syntax bits. */
294 /* isalpha etc. are used for the character classes. */
299 /* 1 if C is an ASCII character. */
300 # define IS_REAL_ASCII(c) ((c) < 0200)
302 /* 1 if C is a unibyte character. */
303 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
305 /* The Emacs definitions should not be directly affected by locales. */
307 /* In Emacs, these are only used for single-byte characters. */
308 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
309 # define ISCNTRL(c) ((c) < ' ')
310 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
311 || ((c) >= 'a' && (c) <= 'f') \
312 || ((c) >= 'A' && (c) <= 'F'))
314 /* This is only used for single-byte characters. */
315 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
317 /* The rest must handle multibyte characters. */
319 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
320 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
323 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
324 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
327 # define ISALNUM(c) (IS_REAL_ASCII (c) \
328 ? (((c) >= 'a' && (c) <= 'z') \
329 || ((c) >= 'A' && (c) <= 'Z') \
330 || ((c) >= '0' && (c) <= '9')) \
331 : SYNTAX (c) == Sword)
333 # define ISALPHA(c) (IS_REAL_ASCII (c) \
334 ? (((c) >= 'a' && (c) <= 'z') \
335 || ((c) >= 'A' && (c) <= 'Z')) \
336 : SYNTAX (c) == Sword)
338 # define ISLOWER(c) lowercasep (c)
340 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
341 ? ((c) > ' ' && (c) < 0177 \
342 && !(((c) >= 'a' && (c) <= 'z') \
343 || ((c) >= 'A' && (c) <= 'Z') \
344 || ((c) >= '0' && (c) <= '9'))) \
345 : SYNTAX (c) != Sword)
347 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
349 # define ISUPPER(c) uppercasep (c)
351 # define ISWORD(c) (SYNTAX (c) == Sword)
353 #else /* not emacs */
355 /* 1 if C is an ASCII character. */
356 # define IS_REAL_ASCII(c) ((c) < 0200)
358 /* This distinction is not meaningful, except in Emacs. */
359 # define ISUNIBYTE(c) 1
362 # define ISBLANK(c) isblank (c)
364 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
367 # define ISGRAPH(c) isgraph (c)
369 # define ISGRAPH(c) (isprint (c) && !isspace (c))
372 /* Solaris defines ISPRINT so we must undefine it first. */
374 # define ISPRINT(c) isprint (c)
375 # define ISDIGIT(c) isdigit (c)
376 # define ISALNUM(c) isalnum (c)
377 # define ISALPHA(c) isalpha (c)
378 # define ISCNTRL(c) iscntrl (c)
379 # define ISLOWER(c) islower (c)
380 # define ISPUNCT(c) ispunct (c)
381 # define ISSPACE(c) isspace (c)
382 # define ISUPPER(c) isupper (c)
383 # define ISXDIGIT(c) isxdigit (c)
385 # define ISWORD(c) ISALPHA (c)
388 # define TOLOWER(c) _tolower (c)
390 # define TOLOWER(c) tolower (c)
393 /* How many characters in the character set. */
394 # define CHAR_SET_SIZE 256
398 extern char *re_syntax_table
;
400 # else /* not SYNTAX_TABLE */
402 static char re_syntax_table
[CHAR_SET_SIZE
];
405 init_syntax_once (void)
413 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
415 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
417 re_syntax_table
[c
] = Sword
;
419 re_syntax_table
['_'] = Ssymbol
;
424 # endif /* not SYNTAX_TABLE */
426 # define SYNTAX(c) re_syntax_table[(c)]
428 #endif /* not emacs */
430 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
432 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
433 use `alloca' instead of `malloc'. This is because using malloc in
434 re_search* or re_match* could cause memory leaks when C-g is used in
435 Emacs; also, malloc is slower and causes storage fragmentation. On
436 the other hand, malloc is more portable, and easier to debug.
438 Because we sometimes use alloca, some routines have to be macros,
439 not functions -- `alloca'-allocated space disappears at the end of the
440 function it is called in. */
444 # define REGEX_ALLOCATE malloc
445 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
446 # define REGEX_FREE free
448 #else /* not REGEX_MALLOC */
450 /* Emacs already defines alloca, sometimes. */
453 /* Make alloca work the best possible way. */
455 # define alloca __builtin_alloca
456 # else /* not __GNUC__ */
457 # ifdef HAVE_ALLOCA_H
459 # endif /* HAVE_ALLOCA_H */
460 # endif /* not __GNUC__ */
462 # endif /* not alloca */
464 # define REGEX_ALLOCATE alloca
466 /* Assumes a `char *destination' variable. */
467 # define REGEX_REALLOCATE(source, osize, nsize) \
468 (destination = (char *) alloca (nsize), \
469 memcpy (destination, source, osize))
471 /* No need to do anything to free, after alloca. */
472 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
474 #endif /* not REGEX_MALLOC */
476 /* Define how to allocate the failure stack. */
478 #if defined REL_ALLOC && defined REGEX_MALLOC
480 # define REGEX_ALLOCATE_STACK(size) \
481 r_alloc (&failure_stack_ptr, (size))
482 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
483 r_re_alloc (&failure_stack_ptr, (nsize))
484 # define REGEX_FREE_STACK(ptr) \
485 r_alloc_free (&failure_stack_ptr)
487 #else /* not using relocating allocator */
491 # define REGEX_ALLOCATE_STACK malloc
492 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
493 # define REGEX_FREE_STACK free
495 # else /* not REGEX_MALLOC */
497 # define REGEX_ALLOCATE_STACK alloca
499 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
500 REGEX_REALLOCATE (source, osize, nsize)
501 /* No need to explicitly free anything. */
502 # define REGEX_FREE_STACK(arg) ((void)0)
504 # endif /* not REGEX_MALLOC */
505 #endif /* not using relocating allocator */
508 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
509 `string1' or just past its end. This works if PTR is NULL, which is
511 #define FIRST_STRING_P(ptr) \
512 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
514 /* (Re)Allocate N items of type T using malloc, or fail. */
515 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
516 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
517 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
519 #define BYTEWIDTH 8 /* In bits. */
521 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
525 #define MAX(a, b) ((a) > (b) ? (a) : (b))
526 #define MIN(a, b) ((a) < (b) ? (a) : (b))
528 /* Type of source-pattern and string chars. */
530 typedef unsigned char re_char
;
532 typedef const unsigned char re_char
;
535 typedef char boolean
;
537 static regoff_t
re_match_2_internal (struct re_pattern_buffer
*bufp
,
538 re_char
*string1
, size_t size1
,
539 re_char
*string2
, size_t size2
,
541 struct re_registers
*regs
,
544 /* These are the command codes that appear in compiled regular
545 expressions. Some opcodes are followed by argument bytes. A
546 command code can specify any interpretation whatsoever for its
547 arguments. Zero bytes may appear in the compiled regular expression. */
553 /* Succeed right away--no more backtracking. */
556 /* Followed by one byte giving n, then by n literal bytes. */
559 /* Matches any (more or less) character. */
562 /* Matches any one char belonging to specified set. First
563 following byte is number of bitmap bytes. Then come bytes
564 for a bitmap saying which chars are in. Bits in each byte
565 are ordered low-bit-first. A character is in the set if its
566 bit is 1. A character too large to have a bit in the map is
567 automatically not in the set.
569 If the length byte has the 0x80 bit set, then that stuff
570 is followed by a range table:
571 2 bytes of flags for character sets (low 8 bits, high 8 bits)
572 See RANGE_TABLE_WORK_BITS below.
573 2 bytes, the number of pairs that follow (upto 32767)
574 pairs, each 2 multibyte characters,
575 each multibyte character represented as 3 bytes. */
578 /* Same parameters as charset, but match any character that is
579 not one of those specified. */
582 /* Start remembering the text that is matched, for storing in a
583 register. Followed by one byte with the register number, in
584 the range 0 to one less than the pattern buffer's re_nsub
588 /* Stop remembering the text that is matched and store it in a
589 memory register. Followed by one byte with the register
590 number, in the range 0 to one less than `re_nsub' in the
594 /* Match a duplicate of something remembered. Followed by one
595 byte containing the register number. */
598 /* Fail unless at beginning of line. */
601 /* Fail unless at end of line. */
604 /* Succeeds if at beginning of buffer (if emacs) or at beginning
605 of string to be matched (if not). */
608 /* Analogously, for end of buffer/string. */
611 /* Followed by two byte relative address to which to jump. */
614 /* Followed by two-byte relative address of place to resume at
615 in case of failure. */
618 /* Like on_failure_jump, but pushes a placeholder instead of the
619 current string position when executed. */
620 on_failure_keep_string_jump
,
622 /* Just like `on_failure_jump', except that it checks that we
623 don't get stuck in an infinite loop (matching an empty string
625 on_failure_jump_loop
,
627 /* Just like `on_failure_jump_loop', except that it checks for
628 a different kind of loop (the kind that shows up with non-greedy
629 operators). This operation has to be immediately preceded
631 on_failure_jump_nastyloop
,
633 /* A smart `on_failure_jump' used for greedy * and + operators.
634 It analyzes the loop before which it is put and if the
635 loop does not require backtracking, it changes itself to
636 `on_failure_keep_string_jump' and short-circuits the loop,
637 else it just defaults to changing itself into `on_failure_jump'.
638 It assumes that it is pointing to just past a `jump'. */
639 on_failure_jump_smart
,
641 /* Followed by two-byte relative address and two-byte number n.
642 After matching N times, jump to the address upon failure.
643 Does not work if N starts at 0: use on_failure_jump_loop
647 /* Followed by two-byte relative address, and two-byte number n.
648 Jump to the address N times, then fail. */
651 /* Set the following two-byte relative address to the
652 subsequent two-byte number. The address *includes* the two
656 wordbeg
, /* Succeeds if at word beginning. */
657 wordend
, /* Succeeds if at word end. */
659 wordbound
, /* Succeeds if at a word boundary. */
660 notwordbound
, /* Succeeds if not at a word boundary. */
662 symbeg
, /* Succeeds if at symbol beginning. */
663 symend
, /* Succeeds if at symbol end. */
665 /* Matches any character whose syntax is specified. Followed by
666 a byte which contains a syntax code, e.g., Sword. */
669 /* Matches any character whose syntax is not that specified. */
673 ,before_dot
, /* Succeeds if before point. */
674 at_dot
, /* Succeeds if at point. */
675 after_dot
, /* Succeeds if after point. */
677 /* Matches any character whose category-set contains the specified
678 category. The operator is followed by a byte which contains a
679 category code (mnemonic ASCII character). */
682 /* Matches any character whose category-set does not contain the
683 specified category. The operator is followed by a byte which
684 contains the category code (mnemonic ASCII character). */
689 /* Common operations on the compiled pattern. */
691 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
693 #define STORE_NUMBER(destination, number) \
695 (destination)[0] = (number) & 0377; \
696 (destination)[1] = (number) >> 8; \
699 /* Same as STORE_NUMBER, except increment DESTINATION to
700 the byte after where the number is stored. Therefore, DESTINATION
701 must be an lvalue. */
703 #define STORE_NUMBER_AND_INCR(destination, number) \
705 STORE_NUMBER (destination, number); \
706 (destination) += 2; \
709 /* Put into DESTINATION a number stored in two contiguous bytes starting
712 #define EXTRACT_NUMBER(destination, source) \
713 ((destination) = extract_number (source))
716 extract_number (re_char
*source
)
718 return (SIGN_EXTEND_CHAR (source
[1]) << 8) + source
[0];
721 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
722 SOURCE must be an lvalue. */
724 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
725 ((destination) = extract_number_and_incr (&source))
728 extract_number_and_incr (re_char
**source
)
730 int num
= extract_number (*source
);
735 /* Store a multibyte character in three contiguous bytes starting
736 DESTINATION, and increment DESTINATION to the byte after where the
737 character is stored. Therefore, DESTINATION must be an lvalue. */
739 #define STORE_CHARACTER_AND_INCR(destination, character) \
741 (destination)[0] = (character) & 0377; \
742 (destination)[1] = ((character) >> 8) & 0377; \
743 (destination)[2] = (character) >> 16; \
744 (destination) += 3; \
747 /* Put into DESTINATION a character stored in three contiguous bytes
748 starting at SOURCE. */
750 #define EXTRACT_CHARACTER(destination, source) \
752 (destination) = ((source)[0] \
753 | ((source)[1] << 8) \
754 | ((source)[2] << 16)); \
758 /* Macros for charset. */
760 /* Size of bitmap of charset P in bytes. P is a start of charset,
761 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
762 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
764 /* Nonzero if charset P has range table. */
765 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
767 /* Return the address of range table of charset P. But not the start
768 of table itself, but the before where the number of ranges is
769 stored. `2 +' means to skip re_opcode_t and size of bitmap,
770 and the 2 bytes of flags at the start of the range table. */
771 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
773 /* Extract the bit flags that start a range table. */
774 #define CHARSET_RANGE_TABLE_BITS(p) \
775 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
776 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
778 /* Return the address of end of RANGE_TABLE. COUNT is number of
779 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
780 is start of range and end of range. `* 3' is size of each start
782 #define CHARSET_RANGE_TABLE_END(range_table, count) \
783 ((range_table) + (count) * 2 * 3)
785 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
786 COUNT is number of ranges in RANGE_TABLE. */
787 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
790 re_wchar_t range_start, range_end; \
792 re_char *range_table_end \
793 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
795 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
797 EXTRACT_CHARACTER (range_start, rtp); \
798 EXTRACT_CHARACTER (range_end, rtp + 3); \
800 if (range_start <= (c) && (c) <= range_end) \
809 /* Test if C is in range table of CHARSET. The flag NOT is negated if
810 C is listed in it. */
811 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
814 /* Number of ranges in range table. */ \
816 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
818 EXTRACT_NUMBER_AND_INCR (count, range_table); \
819 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
823 /* If DEBUG is defined, Regex prints many voluminous messages about what
824 it is doing (if the variable `debug' is nonzero). If linked with the
825 main program in `iregex.c', you can enter patterns and strings
826 interactively. And if linked with the main program in `main.c' and
827 the other test files, you can run the already-written tests. */
831 /* We use standard I/O for debugging. */
834 /* It is useful to test things that ``must'' be true when debugging. */
837 static int debug
= -100000;
839 # define DEBUG_STATEMENT(e) e
840 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
841 # define DEBUG_COMPILES_ARGUMENTS
842 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
843 if (debug > 0) print_partial_compiled_pattern (s, e)
844 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
845 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
848 /* Print the fastmap in human-readable form. */
851 print_fastmap (char *fastmap
)
853 unsigned was_a_range
= 0;
856 while (i
< (1 << BYTEWIDTH
))
862 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
878 /* Print a compiled pattern string in human-readable form, starting at
879 the START pointer into it and ending just before the pointer END. */
882 print_partial_compiled_pattern (re_char
*start
, re_char
*end
)
890 fprintf (stderr
, "(null)\n");
894 /* Loop over pattern commands. */
897 fprintf (stderr
, "%td:\t", p
- start
);
899 switch ((re_opcode_t
) *p
++)
902 fprintf (stderr
, "/no_op");
906 fprintf (stderr
, "/succeed");
911 fprintf (stderr
, "/exactn/%d", mcnt
);
914 fprintf (stderr
, "/%c", *p
++);
920 fprintf (stderr
, "/start_memory/%d", *p
++);
924 fprintf (stderr
, "/stop_memory/%d", *p
++);
928 fprintf (stderr
, "/duplicate/%d", *p
++);
932 fprintf (stderr
, "/anychar");
938 register int c
, last
= -100;
939 register int in_range
= 0;
940 int length
= CHARSET_BITMAP_SIZE (p
- 1);
941 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
943 fprintf (stderr
, "/charset [%s",
944 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
947 fprintf (stderr
, " !extends past end of pattern! ");
949 for (c
= 0; c
< 256; c
++)
951 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
953 /* Are we starting a range? */
954 if (last
+ 1 == c
&& ! in_range
)
956 fprintf (stderr
, "-");
959 /* Have we broken a range? */
960 else if (last
+ 1 != c
&& in_range
)
962 fprintf (stderr
, "%c", last
);
967 fprintf (stderr
, "%c", c
);
973 fprintf (stderr
, "%c", last
);
975 fprintf (stderr
, "]");
982 fprintf (stderr
, "has-range-table");
984 /* ??? Should print the range table; for now, just skip it. */
985 p
+= 2; /* skip range table bits */
986 EXTRACT_NUMBER_AND_INCR (count
, p
);
987 p
= CHARSET_RANGE_TABLE_END (p
, count
);
993 fprintf (stderr
, "/begline");
997 fprintf (stderr
, "/endline");
1000 case on_failure_jump
:
1001 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1002 fprintf (stderr
, "/on_failure_jump to %td", p
+ mcnt
- start
);
1005 case on_failure_keep_string_jump
:
1006 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1007 fprintf (stderr
, "/on_failure_keep_string_jump to %td",
1011 case on_failure_jump_nastyloop
:
1012 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1013 fprintf (stderr
, "/on_failure_jump_nastyloop to %td",
1017 case on_failure_jump_loop
:
1018 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1019 fprintf (stderr
, "/on_failure_jump_loop to %td",
1023 case on_failure_jump_smart
:
1024 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1025 fprintf (stderr
, "/on_failure_jump_smart to %td",
1030 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1031 fprintf (stderr
, "/jump to %td", p
+ mcnt
- start
);
1035 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1036 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1037 fprintf (stderr
, "/succeed_n to %td, %d times",
1038 p
- 2 + mcnt
- start
, mcnt2
);
1042 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1043 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1044 fprintf (stderr
, "/jump_n to %td, %d times",
1045 p
- 2 + mcnt
- start
, mcnt2
);
1049 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1050 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1051 fprintf (stderr
, "/set_number_at location %td to %d",
1052 p
- 2 + mcnt
- start
, mcnt2
);
1056 fprintf (stderr
, "/wordbound");
1060 fprintf (stderr
, "/notwordbound");
1064 fprintf (stderr
, "/wordbeg");
1068 fprintf (stderr
, "/wordend");
1072 fprintf (stderr
, "/symbeg");
1076 fprintf (stderr
, "/symend");
1080 fprintf (stderr
, "/syntaxspec");
1082 fprintf (stderr
, "/%d", mcnt
);
1086 fprintf (stderr
, "/notsyntaxspec");
1088 fprintf (stderr
, "/%d", mcnt
);
1093 fprintf (stderr
, "/before_dot");
1097 fprintf (stderr
, "/at_dot");
1101 fprintf (stderr
, "/after_dot");
1105 fprintf (stderr
, "/categoryspec");
1107 fprintf (stderr
, "/%d", mcnt
);
1110 case notcategoryspec
:
1111 fprintf (stderr
, "/notcategoryspec");
1113 fprintf (stderr
, "/%d", mcnt
);
1118 fprintf (stderr
, "/begbuf");
1122 fprintf (stderr
, "/endbuf");
1126 fprintf (stderr
, "?%d", *(p
-1));
1129 fprintf (stderr
, "\n");
1132 fprintf (stderr
, "%td:\tend of pattern.\n", p
- start
);
1137 print_compiled_pattern (struct re_pattern_buffer
*bufp
)
1139 re_char
*buffer
= bufp
->buffer
;
1141 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1142 printf ("%ld bytes used/%ld bytes allocated.\n",
1143 bufp
->used
, bufp
->allocated
);
1145 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1147 printf ("fastmap: ");
1148 print_fastmap (bufp
->fastmap
);
1151 printf ("re_nsub: %zu\t", bufp
->re_nsub
);
1152 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1153 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1154 printf ("no_sub: %d\t", bufp
->no_sub
);
1155 printf ("not_bol: %d\t", bufp
->not_bol
);
1156 printf ("not_eol: %d\t", bufp
->not_eol
);
1157 printf ("syntax: %lx\n", bufp
->syntax
);
1159 /* Perhaps we should print the translate table? */
1164 print_double_string (re_char
*where
, re_char
*string1
, ssize_t size1
,
1165 re_char
*string2
, ssize_t size2
)
1173 if (FIRST_STRING_P (where
))
1175 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1176 putchar (string1
[this_char
]);
1181 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1182 putchar (string2
[this_char
]);
1186 #else /* not DEBUG */
1191 # define DEBUG_STATEMENT(e)
1192 # if __STDC_VERSION__ < 199901L
1193 # define DEBUG_COMPILES_ARGUMENTS
1194 # define DEBUG_PRINT /* 'DEBUG_PRINT (x, y)' discards X and Y. */ (void)
1196 # define DEBUG_PRINT(...)
1198 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1199 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1201 #endif /* not DEBUG */
1203 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1205 # define IF_LINT(Code) Code
1207 # define IF_LINT(Code) /* empty */
1210 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1211 also be assigned to arbitrarily: each pattern buffer stores its own
1212 syntax, so it can be changed between regex compilations. */
1213 /* This has no initializer because initialized variables in Emacs
1214 become read-only after dumping. */
1215 reg_syntax_t re_syntax_options
;
1218 /* Specify the precise syntax of regexps for compilation. This provides
1219 for compatibility for various utilities which historically have
1220 different, incompatible syntaxes.
1222 The argument SYNTAX is a bit mask comprised of the various bits
1223 defined in regex.h. We return the old syntax. */
1226 re_set_syntax (reg_syntax_t syntax
)
1228 reg_syntax_t ret
= re_syntax_options
;
1230 re_syntax_options
= syntax
;
1233 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1235 /* Regexp to use to replace spaces, or NULL meaning don't. */
1236 static re_char
*whitespace_regexp
;
1239 re_set_whitespace_regexp (const char *regexp
)
1241 whitespace_regexp
= (re_char
*) regexp
;
1243 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1245 /* This table gives an error message for each of the error codes listed
1246 in regex.h. Obviously the order here has to be same as there.
1247 POSIX doesn't require that we do anything for REG_NOERROR,
1248 but why not be nice? */
1250 static const char *re_error_msgid
[] =
1252 gettext_noop ("Success"), /* REG_NOERROR */
1253 gettext_noop ("No match"), /* REG_NOMATCH */
1254 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1255 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1256 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1257 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1258 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1259 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1260 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1261 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1262 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1263 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1264 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1265 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1266 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1267 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1268 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1269 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1272 /* Avoiding alloca during matching, to placate r_alloc. */
1274 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1275 searching and matching functions should not call alloca. On some
1276 systems, alloca is implemented in terms of malloc, and if we're
1277 using the relocating allocator routines, then malloc could cause a
1278 relocation, which might (if the strings being searched are in the
1279 ralloc heap) shift the data out from underneath the regexp
1282 Here's another reason to avoid allocation: Emacs
1283 processes input from X in a signal handler; processing X input may
1284 call malloc; if input arrives while a matching routine is calling
1285 malloc, then we're scrod. But Emacs can't just block input while
1286 calling matching routines; then we don't notice interrupts when
1287 they come in. So, Emacs blocks input around all regexp calls
1288 except the matching calls, which it leaves unprotected, in the
1289 faith that they will not malloc. */
1291 /* Normally, this is fine. */
1292 #define MATCH_MAY_ALLOCATE
1294 /* The match routines may not allocate if (1) they would do it with malloc
1295 and (2) it's not safe for them to use malloc.
1296 Note that if REL_ALLOC is defined, matching would not use malloc for the
1297 failure stack, but we would still use it for the register vectors;
1298 so REL_ALLOC should not affect this. */
1299 #if defined REGEX_MALLOC && defined emacs
1300 # undef MATCH_MAY_ALLOCATE
1304 /* Failure stack declarations and macros; both re_compile_fastmap and
1305 re_match_2 use a failure stack. These have to be macros because of
1306 REGEX_ALLOCATE_STACK. */
1309 /* Approximate number of failure points for which to initially allocate space
1310 when matching. If this number is exceeded, we allocate more
1311 space, so it is not a hard limit. */
1312 #ifndef INIT_FAILURE_ALLOC
1313 # define INIT_FAILURE_ALLOC 20
1316 /* Roughly the maximum number of failure points on the stack. Would be
1317 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1318 This is a variable only so users of regex can assign to it; we never
1319 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1320 before using it, so it should probably be a byte-count instead. */
1321 # if defined MATCH_MAY_ALLOCATE
1322 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1323 whose default stack limit is 2mb. In order for a larger
1324 value to work reliably, you have to try to make it accord
1325 with the process stack limit. */
1326 size_t re_max_failures
= 40000;
1328 size_t re_max_failures
= 4000;
1331 union fail_stack_elt
1334 /* This should be the biggest `int' that's no bigger than a pointer. */
1338 typedef union fail_stack_elt fail_stack_elt_t
;
1342 fail_stack_elt_t
*stack
;
1344 size_t avail
; /* Offset of next open position. */
1345 size_t frame
; /* Offset of the cur constructed frame. */
1348 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1351 /* Define macros to initialize and free the failure stack.
1352 Do `return -2' if the alloc fails. */
1354 #ifdef MATCH_MAY_ALLOCATE
1355 # define INIT_FAIL_STACK() \
1357 fail_stack.stack = \
1358 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1359 * sizeof (fail_stack_elt_t)); \
1361 if (fail_stack.stack == NULL) \
1364 fail_stack.size = INIT_FAILURE_ALLOC; \
1365 fail_stack.avail = 0; \
1366 fail_stack.frame = 0; \
1369 # define INIT_FAIL_STACK() \
1371 fail_stack.avail = 0; \
1372 fail_stack.frame = 0; \
1375 # define RETALLOC_IF(addr, n, t) \
1376 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1380 /* Double the size of FAIL_STACK, up to a limit
1381 which allows approximately `re_max_failures' items.
1383 Return 1 if succeeds, and 0 if either ran out of memory
1384 allocating space for it or it was already too large.
1386 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1388 /* Factor to increase the failure stack size by
1389 when we increase it.
1390 This used to be 2, but 2 was too wasteful
1391 because the old discarded stacks added up to as much space
1392 were as ultimate, maximum-size stack. */
1393 #define FAIL_STACK_GROWTH_FACTOR 4
1395 #define GROW_FAIL_STACK(fail_stack) \
1396 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1397 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1399 : ((fail_stack).stack \
1400 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1401 (fail_stack).size * sizeof (fail_stack_elt_t), \
1402 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1403 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1404 * FAIL_STACK_GROWTH_FACTOR))), \
1406 (fail_stack).stack == NULL \
1408 : ((fail_stack).size \
1409 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1410 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1411 * FAIL_STACK_GROWTH_FACTOR)) \
1412 / sizeof (fail_stack_elt_t)), \
1416 /* Push a pointer value onto the failure stack.
1417 Assumes the variable `fail_stack'. Probably should only
1418 be called from within `PUSH_FAILURE_POINT'. */
1419 #define PUSH_FAILURE_POINTER(item) \
1420 fail_stack.stack[fail_stack.avail++].pointer = (item)
1422 /* This pushes an integer-valued item onto the failure stack.
1423 Assumes the variable `fail_stack'. Probably should only
1424 be called from within `PUSH_FAILURE_POINT'. */
1425 #define PUSH_FAILURE_INT(item) \
1426 fail_stack.stack[fail_stack.avail++].integer = (item)
1428 /* These POP... operations complement the PUSH... operations.
1429 All assume that `fail_stack' is nonempty. */
1430 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1431 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1433 /* Individual items aside from the registers. */
1434 #define NUM_NONREG_ITEMS 3
1436 /* Used to examine the stack (to detect infinite loops). */
1437 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1438 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1439 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1440 #define TOP_FAILURE_HANDLE() fail_stack.frame
1443 #define ENSURE_FAIL_STACK(space) \
1444 while (REMAINING_AVAIL_SLOTS <= space) { \
1445 if (!GROW_FAIL_STACK (fail_stack)) \
1447 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1448 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1451 /* Push register NUM onto the stack. */
1452 #define PUSH_FAILURE_REG(num) \
1454 char *destination; \
1456 ENSURE_FAIL_STACK(3); \
1457 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1458 n, regstart[n], regend[n]); \
1459 PUSH_FAILURE_POINTER (regstart[n]); \
1460 PUSH_FAILURE_POINTER (regend[n]); \
1461 PUSH_FAILURE_INT (n); \
1464 /* Change the counter's value to VAL, but make sure that it will
1465 be reset when backtracking. */
1466 #define PUSH_NUMBER(ptr,val) \
1468 char *destination; \
1470 ENSURE_FAIL_STACK(3); \
1471 EXTRACT_NUMBER (c, ptr); \
1472 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1473 PUSH_FAILURE_INT (c); \
1474 PUSH_FAILURE_POINTER (ptr); \
1475 PUSH_FAILURE_INT (-1); \
1476 STORE_NUMBER (ptr, val); \
1479 /* Pop a saved register off the stack. */
1480 #define POP_FAILURE_REG_OR_COUNT() \
1482 long pfreg = POP_FAILURE_INT (); \
1485 /* It's a counter. */ \
1486 /* Here, we discard `const', making re_match non-reentrant. */ \
1487 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1488 pfreg = POP_FAILURE_INT (); \
1489 STORE_NUMBER (ptr, pfreg); \
1490 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1494 regend[pfreg] = POP_FAILURE_POINTER (); \
1495 regstart[pfreg] = POP_FAILURE_POINTER (); \
1496 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1497 pfreg, regstart[pfreg], regend[pfreg]); \
1501 /* Check that we are not stuck in an infinite loop. */
1502 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1504 ssize_t failure = TOP_FAILURE_HANDLE (); \
1505 /* Check for infinite matching loops */ \
1506 while (failure > 0 \
1507 && (FAILURE_STR (failure) == string_place \
1508 || FAILURE_STR (failure) == NULL)) \
1510 assert (FAILURE_PAT (failure) >= bufp->buffer \
1511 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1512 if (FAILURE_PAT (failure) == pat_cur) \
1517 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1518 failure = NEXT_FAILURE_HANDLE(failure); \
1520 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1523 /* Push the information about the state we will need
1524 if we ever fail back to it.
1526 Requires variables fail_stack, regstart, regend and
1527 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1530 Does `return FAILURE_CODE' if runs out of memory. */
1532 #define PUSH_FAILURE_POINT(pattern, string_place) \
1534 char *destination; \
1535 /* Must be int, so when we don't save any registers, the arithmetic \
1536 of 0 + -1 isn't done as unsigned. */ \
1538 DEBUG_STATEMENT (nfailure_points_pushed++); \
1539 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1540 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1541 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1543 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1545 DEBUG_PRINT ("\n"); \
1547 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1548 PUSH_FAILURE_INT (fail_stack.frame); \
1550 DEBUG_PRINT (" Push string %p: `", string_place); \
1551 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1552 DEBUG_PRINT ("'\n"); \
1553 PUSH_FAILURE_POINTER (string_place); \
1555 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1556 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1557 PUSH_FAILURE_POINTER (pattern); \
1559 /* Close the frame by moving the frame pointer past it. */ \
1560 fail_stack.frame = fail_stack.avail; \
1563 /* Estimate the size of data pushed by a typical failure stack entry.
1564 An estimate is all we need, because all we use this for
1565 is to choose a limit for how big to make the failure stack. */
1566 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1567 #define TYPICAL_FAILURE_SIZE 20
1569 /* How many items can still be added to the stack without overflowing it. */
1570 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1573 /* Pops what PUSH_FAIL_STACK pushes.
1575 We restore into the parameters, all of which should be lvalues:
1576 STR -- the saved data position.
1577 PAT -- the saved pattern position.
1578 REGSTART, REGEND -- arrays of string positions.
1580 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1581 `pend', `string1', `size1', `string2', and `size2'. */
1583 #define POP_FAILURE_POINT(str, pat) \
1585 assert (!FAIL_STACK_EMPTY ()); \
1587 /* Remove failure points and point to how many regs pushed. */ \
1588 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1589 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1590 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1592 /* Pop the saved registers. */ \
1593 while (fail_stack.frame < fail_stack.avail) \
1594 POP_FAILURE_REG_OR_COUNT (); \
1596 pat = POP_FAILURE_POINTER (); \
1597 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1598 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1600 /* If the saved string location is NULL, it came from an \
1601 on_failure_keep_string_jump opcode, and we want to throw away the \
1602 saved NULL, thus retaining our current position in the string. */ \
1603 str = POP_FAILURE_POINTER (); \
1604 DEBUG_PRINT (" Popping string %p: `", str); \
1605 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1606 DEBUG_PRINT ("'\n"); \
1608 fail_stack.frame = POP_FAILURE_INT (); \
1609 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1611 assert (fail_stack.avail >= 0); \
1612 assert (fail_stack.frame <= fail_stack.avail); \
1614 DEBUG_STATEMENT (nfailure_points_popped++); \
1615 } while (0) /* POP_FAILURE_POINT */
1619 /* Registers are set to a sentinel when they haven't yet matched. */
1620 #define REG_UNSET(e) ((e) == NULL)
1622 /* Subroutine declarations and macros for regex_compile. */
1624 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1625 reg_syntax_t syntax
,
1626 struct re_pattern_buffer
*bufp
);
1627 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1628 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1629 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1630 int arg
, unsigned char *end
);
1631 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1632 int arg1
, int arg2
, unsigned char *end
);
1633 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1634 reg_syntax_t syntax
);
1635 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1636 reg_syntax_t syntax
);
1637 static re_char
*skip_one_char (re_char
*p
);
1638 static int analyse_first (re_char
*p
, re_char
*pend
,
1639 char *fastmap
, const int multibyte
);
1641 /* Fetch the next character in the uncompiled pattern, with no
1643 #define PATFETCH(c) \
1646 if (p == pend) return REG_EEND; \
1647 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1652 /* If `translate' is non-null, return translate[D], else just D. We
1653 cast the subscript to translate because some data is declared as
1654 `char *', to avoid warnings when a string constant is passed. But
1655 when we use a character as a subscript we must make it unsigned. */
1657 # define TRANSLATE(d) \
1658 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1662 /* Macros for outputting the compiled pattern into `buffer'. */
1664 /* If the buffer isn't allocated when it comes in, use this. */
1665 #define INIT_BUF_SIZE 32
1667 /* Make sure we have at least N more bytes of space in buffer. */
1668 #define GET_BUFFER_SPACE(n) \
1669 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1672 /* Make sure we have one more byte of buffer space and then add C to it. */
1673 #define BUF_PUSH(c) \
1675 GET_BUFFER_SPACE (1); \
1676 *b++ = (unsigned char) (c); \
1680 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1681 #define BUF_PUSH_2(c1, c2) \
1683 GET_BUFFER_SPACE (2); \
1684 *b++ = (unsigned char) (c1); \
1685 *b++ = (unsigned char) (c2); \
1689 /* Store a jump with opcode OP at LOC to location TO. We store a
1690 relative address offset by the three bytes the jump itself occupies. */
1691 #define STORE_JUMP(op, loc, to) \
1692 store_op1 (op, loc, (to) - (loc) - 3)
1694 /* Likewise, for a two-argument jump. */
1695 #define STORE_JUMP2(op, loc, to, arg) \
1696 store_op2 (op, loc, (to) - (loc) - 3, arg)
1698 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1699 #define INSERT_JUMP(op, loc, to) \
1700 insert_op1 (op, loc, (to) - (loc) - 3, b)
1702 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1703 #define INSERT_JUMP2(op, loc, to, arg) \
1704 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1707 /* This is not an arbitrary limit: the arguments which represent offsets
1708 into the pattern are two bytes long. So if 2^15 bytes turns out to
1709 be too small, many things would have to change. */
1710 # define MAX_BUF_SIZE (1L << 15)
1712 /* Extend the buffer by twice its current size via realloc and
1713 reset the pointers that pointed into the old block to point to the
1714 correct places in the new one. If extending the buffer results in it
1715 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1716 #if __BOUNDED_POINTERS__
1717 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1718 # define MOVE_BUFFER_POINTER(P) \
1719 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1720 SET_HIGH_BOUND (P), \
1721 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1722 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1725 SET_HIGH_BOUND (b); \
1726 SET_HIGH_BOUND (begalt); \
1727 if (fixup_alt_jump) \
1728 SET_HIGH_BOUND (fixup_alt_jump); \
1730 SET_HIGH_BOUND (laststart); \
1731 if (pending_exact) \
1732 SET_HIGH_BOUND (pending_exact); \
1735 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1736 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1738 #define EXTEND_BUFFER() \
1740 unsigned char *old_buffer = bufp->buffer; \
1741 if (bufp->allocated == MAX_BUF_SIZE) \
1743 bufp->allocated <<= 1; \
1744 if (bufp->allocated > MAX_BUF_SIZE) \
1745 bufp->allocated = MAX_BUF_SIZE; \
1746 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1747 if (bufp->buffer == NULL) \
1748 return REG_ESPACE; \
1749 /* If the buffer moved, move all the pointers into it. */ \
1750 if (old_buffer != bufp->buffer) \
1752 unsigned char *new_buffer = bufp->buffer; \
1753 MOVE_BUFFER_POINTER (b); \
1754 MOVE_BUFFER_POINTER (begalt); \
1755 if (fixup_alt_jump) \
1756 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1758 MOVE_BUFFER_POINTER (laststart); \
1759 if (pending_exact) \
1760 MOVE_BUFFER_POINTER (pending_exact); \
1762 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1766 /* Since we have one byte reserved for the register number argument to
1767 {start,stop}_memory, the maximum number of groups we can report
1768 things about is what fits in that byte. */
1769 #define MAX_REGNUM 255
1771 /* But patterns can have more than `MAX_REGNUM' registers. We just
1772 ignore the excess. */
1773 typedef int regnum_t
;
1776 /* Macros for the compile stack. */
1778 /* Since offsets can go either forwards or backwards, this type needs to
1779 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1780 /* int may be not enough when sizeof(int) == 2. */
1781 typedef long pattern_offset_t
;
1785 pattern_offset_t begalt_offset
;
1786 pattern_offset_t fixup_alt_jump
;
1787 pattern_offset_t laststart_offset
;
1789 } compile_stack_elt_t
;
1794 compile_stack_elt_t
*stack
;
1796 size_t avail
; /* Offset of next open position. */
1797 } compile_stack_type
;
1800 #define INIT_COMPILE_STACK_SIZE 32
1802 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1803 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1805 /* The next available element. */
1806 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1808 /* Explicit quit checking is needed for Emacs, which uses polling to
1809 process input events. */
1811 # define IMMEDIATE_QUIT_CHECK \
1813 if (immediate_quit) QUIT; \
1816 # define IMMEDIATE_QUIT_CHECK ((void)0)
1819 /* Structure to manage work area for range table. */
1820 struct range_table_work_area
1822 int *table
; /* actual work area. */
1823 int allocated
; /* allocated size for work area in bytes. */
1824 int used
; /* actually used size in words. */
1825 int bits
; /* flag to record character classes */
1828 /* Make sure that WORK_AREA can hold more N multibyte characters.
1829 This is used only in set_image_of_range and set_image_of_range_1.
1830 It expects WORK_AREA to be a pointer.
1831 If it can't get the space, it returns from the surrounding function. */
1833 #define EXTEND_RANGE_TABLE(work_area, n) \
1835 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1837 extend_range_table_work_area (&work_area); \
1838 if ((work_area).table == 0) \
1839 return (REG_ESPACE); \
1843 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1844 (work_area).bits |= (bit)
1846 /* Bits used to implement the multibyte-part of the various character classes
1847 such as [:alnum:] in a charset's range table. */
1848 #define BIT_WORD 0x1
1849 #define BIT_LOWER 0x2
1850 #define BIT_PUNCT 0x4
1851 #define BIT_SPACE 0x8
1852 #define BIT_UPPER 0x10
1853 #define BIT_MULTIBYTE 0x20
1855 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1856 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1858 EXTEND_RANGE_TABLE ((work_area), 2); \
1859 (work_area).table[(work_area).used++] = (range_start); \
1860 (work_area).table[(work_area).used++] = (range_end); \
1863 /* Free allocated memory for WORK_AREA. */
1864 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1866 if ((work_area).table) \
1867 free ((work_area).table); \
1870 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1871 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1872 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1873 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1876 /* Set the bit for character C in a list. */
1877 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1882 /* Store characters in the range FROM to TO in the bitmap at B (for
1883 ASCII and unibyte characters) and WORK_AREA (for multibyte
1884 characters) while translating them and paying attention to the
1885 continuity of translated characters.
1887 Implementation note: It is better to implement these fairly big
1888 macros by a function, but it's not that easy because macros called
1889 in this macro assume various local variables already declared. */
1891 /* Both FROM and TO are ASCII characters. */
1893 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1897 for (C0 = (FROM); C0 <= (TO); C0++) \
1899 C1 = TRANSLATE (C0); \
1900 if (! ASCII_CHAR_P (C1)) \
1902 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1903 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1906 SET_LIST_BIT (C1); \
1911 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1913 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1915 int C0, C1, C2, I; \
1916 int USED = RANGE_TABLE_WORK_USED (work_area); \
1918 for (C0 = (FROM); C0 <= (TO); C0++) \
1920 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1921 if (CHAR_BYTE8_P (C1)) \
1922 SET_LIST_BIT (C0); \
1925 C2 = TRANSLATE (C1); \
1927 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1929 SET_LIST_BIT (C1); \
1930 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1932 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1933 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1935 if (C2 >= from - 1 && C2 <= to + 1) \
1937 if (C2 == from - 1) \
1938 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1939 else if (C2 == to + 1) \
1940 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1945 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1951 /* Both FROM and TO are multibyte characters. */
1953 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1955 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1957 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1958 for (C0 = (FROM); C0 <= (TO); C0++) \
1960 C1 = TRANSLATE (C0); \
1961 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1962 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1963 SET_LIST_BIT (C2); \
1964 if (C1 >= (FROM) && C1 <= (TO)) \
1966 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1968 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1969 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1971 if (C1 >= from - 1 && C1 <= to + 1) \
1973 if (C1 == from - 1) \
1974 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1975 else if (C1 == to + 1) \
1976 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1981 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1987 /* Get the next unsigned number in the uncompiled pattern. */
1988 #define GET_UNSIGNED_NUMBER(num) \
1991 FREE_STACK_RETURN (REG_EBRACE); \
1995 while ('0' <= c && c <= '9') \
2001 num = num * 10 + c - '0'; \
2002 if (num / 10 != prev) \
2003 FREE_STACK_RETURN (REG_BADBR); \
2005 FREE_STACK_RETURN (REG_EBRACE); \
2011 #if ! WIDE_CHAR_SUPPORT
2013 /* Map a string to the char class it names (if any). */
2015 re_wctype (const re_char
*str
)
2017 const char *string
= (const char *) str
;
2018 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2019 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2020 else if (STREQ (string
, "word")) return RECC_WORD
;
2021 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2022 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2023 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2024 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2025 else if (STREQ (string
, "print")) return RECC_PRINT
;
2026 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2027 else if (STREQ (string
, "space")) return RECC_SPACE
;
2028 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2029 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2030 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2031 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2032 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2033 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2034 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2038 /* True if CH is in the char class CC. */
2040 re_iswctype (int ch
, re_wctype_t cc
)
2044 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2045 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2046 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2047 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2048 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2049 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2050 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2051 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2052 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2053 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2054 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2055 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2056 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2057 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2058 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2059 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2060 case RECC_WORD
: return ISWORD (ch
) != 0;
2061 case RECC_ERROR
: return false;
2067 /* Return a bit-pattern to use in the range-table bits to match multibyte
2068 chars of class CC. */
2070 re_wctype_to_bit (re_wctype_t cc
)
2074 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2075 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2076 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2077 case RECC_LOWER
: return BIT_LOWER
;
2078 case RECC_UPPER
: return BIT_UPPER
;
2079 case RECC_PUNCT
: return BIT_PUNCT
;
2080 case RECC_SPACE
: return BIT_SPACE
;
2081 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2082 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2089 /* Filling in the work area of a range. */
2091 /* Actually extend the space in WORK_AREA. */
2094 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2096 work_area
->allocated
+= 16 * sizeof (int);
2097 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2103 /* Carefully find the ranges of codes that are equivalent
2104 under case conversion to the range start..end when passed through
2105 TRANSLATE. Handle the case where non-letters can come in between
2106 two upper-case letters (which happens in Latin-1).
2107 Also handle the case of groups of more than 2 case-equivalent chars.
2109 The basic method is to look at consecutive characters and see
2110 if they can form a run that can be handled as one.
2112 Returns -1 if successful, REG_ESPACE if ran out of space. */
2115 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2116 re_wchar_t start
, re_wchar_t end
,
2117 RE_TRANSLATE_TYPE translate
)
2119 /* `one_case' indicates a character, or a run of characters,
2120 each of which is an isolate (no case-equivalents).
2121 This includes all ASCII non-letters.
2123 `two_case' indicates a character, or a run of characters,
2124 each of which has two case-equivalent forms.
2125 This includes all ASCII letters.
2127 `strange' indicates a character that has more than one
2130 enum case_type
{one_case
, two_case
, strange
};
2132 /* Describe the run that is in progress,
2133 which the next character can try to extend.
2134 If run_type is strange, that means there really is no run.
2135 If run_type is one_case, then run_start...run_end is the run.
2136 If run_type is two_case, then the run is run_start...run_end,
2137 and the case-equivalents end at run_eqv_end. */
2139 enum case_type run_type
= strange
;
2140 int run_start
, run_end
, run_eqv_end
;
2142 Lisp_Object eqv_table
;
2144 if (!RE_TRANSLATE_P (translate
))
2146 EXTEND_RANGE_TABLE (work_area
, 2);
2147 work_area
->table
[work_area
->used
++] = (start
);
2148 work_area
->table
[work_area
->used
++] = (end
);
2152 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2154 for (; start
<= end
; start
++)
2156 enum case_type this_type
;
2157 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2158 int minchar
, maxchar
;
2160 /* Classify this character */
2162 this_type
= one_case
;
2163 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2164 this_type
= two_case
;
2166 this_type
= strange
;
2169 minchar
= start
, maxchar
= eqv
;
2171 minchar
= eqv
, maxchar
= start
;
2173 /* Can this character extend the run in progress? */
2174 if (this_type
== strange
|| this_type
!= run_type
2175 || !(minchar
== run_end
+ 1
2176 && (run_type
== two_case
2177 ? maxchar
== run_eqv_end
+ 1 : 1)))
2180 Record each of its equivalent ranges. */
2181 if (run_type
== one_case
)
2183 EXTEND_RANGE_TABLE (work_area
, 2);
2184 work_area
->table
[work_area
->used
++] = run_start
;
2185 work_area
->table
[work_area
->used
++] = run_end
;
2187 else if (run_type
== two_case
)
2189 EXTEND_RANGE_TABLE (work_area
, 4);
2190 work_area
->table
[work_area
->used
++] = run_start
;
2191 work_area
->table
[work_area
->used
++] = run_end
;
2192 work_area
->table
[work_area
->used
++]
2193 = RE_TRANSLATE (eqv_table
, run_start
);
2194 work_area
->table
[work_area
->used
++]
2195 = RE_TRANSLATE (eqv_table
, run_end
);
2200 if (this_type
== strange
)
2202 /* For a strange character, add each of its equivalents, one
2203 by one. Don't start a range. */
2206 EXTEND_RANGE_TABLE (work_area
, 2);
2207 work_area
->table
[work_area
->used
++] = eqv
;
2208 work_area
->table
[work_area
->used
++] = eqv
;
2209 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2211 while (eqv
!= start
);
2214 /* Add this char to the run, or start a new run. */
2215 else if (run_type
== strange
)
2217 /* Initialize a new range. */
2218 run_type
= this_type
;
2221 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2225 /* Extend a running range. */
2227 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2231 /* If a run is still in progress at the end, finish it now
2232 by recording its equivalent ranges. */
2233 if (run_type
== one_case
)
2235 EXTEND_RANGE_TABLE (work_area
, 2);
2236 work_area
->table
[work_area
->used
++] = run_start
;
2237 work_area
->table
[work_area
->used
++] = run_end
;
2239 else if (run_type
== two_case
)
2241 EXTEND_RANGE_TABLE (work_area
, 4);
2242 work_area
->table
[work_area
->used
++] = run_start
;
2243 work_area
->table
[work_area
->used
++] = run_end
;
2244 work_area
->table
[work_area
->used
++]
2245 = RE_TRANSLATE (eqv_table
, run_start
);
2246 work_area
->table
[work_area
->used
++]
2247 = RE_TRANSLATE (eqv_table
, run_end
);
2255 /* Record the image of the range start..end when passed through
2256 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2257 and is not even necessarily contiguous.
2258 Normally we approximate it with the smallest contiguous range that contains
2259 all the chars we need. However, for Latin-1 we go to extra effort
2262 This function is not called for ASCII ranges.
2264 Returns -1 if successful, REG_ESPACE if ran out of space. */
2267 set_image_of_range (struct range_table_work_area
*work_area
,
2268 re_wchar_t start
, re_wchar_t end
,
2269 RE_TRANSLATE_TYPE translate
)
2271 re_wchar_t cmin
, cmax
;
2274 /* For Latin-1 ranges, use set_image_of_range_1
2275 to get proper handling of ranges that include letters and nonletters.
2276 For a range that includes the whole of Latin-1, this is not necessary.
2277 For other character sets, we don't bother to get this right. */
2278 if (RE_TRANSLATE_P (translate
) && start
< 04400
2279 && !(start
< 04200 && end
>= 04377))
2286 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2296 EXTEND_RANGE_TABLE (work_area
, 2);
2297 work_area
->table
[work_area
->used
++] = (start
);
2298 work_area
->table
[work_area
->used
++] = (end
);
2300 cmin
= -1, cmax
= -1;
2302 if (RE_TRANSLATE_P (translate
))
2306 for (ch
= start
; ch
<= end
; ch
++)
2308 re_wchar_t c
= TRANSLATE (ch
);
2309 if (! (start
<= c
&& c
<= end
))
2315 cmin
= MIN (cmin
, c
);
2316 cmax
= MAX (cmax
, c
);
2323 EXTEND_RANGE_TABLE (work_area
, 2);
2324 work_area
->table
[work_area
->used
++] = (cmin
);
2325 work_area
->table
[work_area
->used
++] = (cmax
);
2333 #ifndef MATCH_MAY_ALLOCATE
2335 /* If we cannot allocate large objects within re_match_2_internal,
2336 we make the fail stack and register vectors global.
2337 The fail stack, we grow to the maximum size when a regexp
2339 The register vectors, we adjust in size each time we
2340 compile a regexp, according to the number of registers it needs. */
2342 static fail_stack_type fail_stack
;
2344 /* Size with which the following vectors are currently allocated.
2345 That is so we can make them bigger as needed,
2346 but never make them smaller. */
2347 static int regs_allocated_size
;
2349 static re_char
** regstart
, ** regend
;
2350 static re_char
**best_regstart
, **best_regend
;
2352 /* Make the register vectors big enough for NUM_REGS registers,
2353 but don't make them smaller. */
2356 regex_grow_registers (int num_regs
)
2358 if (num_regs
> regs_allocated_size
)
2360 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2361 RETALLOC_IF (regend
, num_regs
, re_char
*);
2362 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2363 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2365 regs_allocated_size
= num_regs
;
2369 #endif /* not MATCH_MAY_ALLOCATE */
2371 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2374 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2375 Returns one of error codes defined in `regex.h', or zero for success.
2377 Assumes the `allocated' (and perhaps `buffer') and `translate'
2378 fields are set in BUFP on entry.
2380 If it succeeds, results are put in BUFP (if it returns an error, the
2381 contents of BUFP are undefined):
2382 `buffer' is the compiled pattern;
2383 `syntax' is set to SYNTAX;
2384 `used' is set to the length of the compiled pattern;
2385 `fastmap_accurate' is zero;
2386 `re_nsub' is the number of subexpressions in PATTERN;
2387 `not_bol' and `not_eol' are zero;
2389 The `fastmap' field is neither examined nor set. */
2391 /* Insert the `jump' from the end of last alternative to "here".
2392 The space for the jump has already been allocated. */
2393 #define FIXUP_ALT_JUMP() \
2395 if (fixup_alt_jump) \
2396 STORE_JUMP (jump, fixup_alt_jump, b); \
2400 /* Return, freeing storage we allocated. */
2401 #define FREE_STACK_RETURN(value) \
2403 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2404 free (compile_stack.stack); \
2408 static reg_errcode_t
2409 regex_compile (const re_char
*pattern
, size_t size
, reg_syntax_t syntax
, struct re_pattern_buffer
*bufp
)
2411 /* We fetch characters from PATTERN here. */
2412 register re_wchar_t c
, c1
;
2414 /* Points to the end of the buffer, where we should append. */
2415 register unsigned char *b
;
2417 /* Keeps track of unclosed groups. */
2418 compile_stack_type compile_stack
;
2420 /* Points to the current (ending) position in the pattern. */
2422 /* `const' makes AIX compiler fail. */
2423 unsigned char *p
= pattern
;
2425 re_char
*p
= pattern
;
2427 re_char
*pend
= pattern
+ size
;
2429 /* How to translate the characters in the pattern. */
2430 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2432 /* Address of the count-byte of the most recently inserted `exactn'
2433 command. This makes it possible to tell if a new exact-match
2434 character can be added to that command or if the character requires
2435 a new `exactn' command. */
2436 unsigned char *pending_exact
= 0;
2438 /* Address of start of the most recently finished expression.
2439 This tells, e.g., postfix * where to find the start of its
2440 operand. Reset at the beginning of groups and alternatives. */
2441 unsigned char *laststart
= 0;
2443 /* Address of beginning of regexp, or inside of last group. */
2444 unsigned char *begalt
;
2446 /* Place in the uncompiled pattern (i.e., the {) to
2447 which to go back if the interval is invalid. */
2448 re_char
*beg_interval
;
2450 /* Address of the place where a forward jump should go to the end of
2451 the containing expression. Each alternative of an `or' -- except the
2452 last -- ends with a forward jump of this sort. */
2453 unsigned char *fixup_alt_jump
= 0;
2455 /* Work area for range table of charset. */
2456 struct range_table_work_area range_table_work
;
2458 /* If the object matched can contain multibyte characters. */
2459 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2461 /* Nonzero if we have pushed down into a subpattern. */
2462 int in_subpattern
= 0;
2464 /* These hold the values of p, pattern, and pend from the main
2465 pattern when we have pushed into a subpattern. */
2466 re_char
*main_p
IF_LINT (= NULL
);
2467 re_char
*main_pattern
IF_LINT (= NULL
);
2468 re_char
*main_pend
IF_LINT (= NULL
);
2472 DEBUG_PRINT ("\nCompiling pattern: ");
2475 unsigned debug_count
;
2477 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2478 putchar (pattern
[debug_count
]);
2483 /* Initialize the compile stack. */
2484 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2485 if (compile_stack
.stack
== NULL
)
2488 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2489 compile_stack
.avail
= 0;
2491 range_table_work
.table
= 0;
2492 range_table_work
.allocated
= 0;
2494 /* Initialize the pattern buffer. */
2495 bufp
->syntax
= syntax
;
2496 bufp
->fastmap_accurate
= 0;
2497 bufp
->not_bol
= bufp
->not_eol
= 0;
2498 bufp
->used_syntax
= 0;
2500 /* Set `used' to zero, so that if we return an error, the pattern
2501 printer (for debugging) will think there's no pattern. We reset it
2505 /* Always count groups, whether or not bufp->no_sub is set. */
2508 #if !defined emacs && !defined SYNTAX_TABLE
2509 /* Initialize the syntax table. */
2510 init_syntax_once ();
2513 if (bufp
->allocated
== 0)
2516 { /* If zero allocated, but buffer is non-null, try to realloc
2517 enough space. This loses if buffer's address is bogus, but
2518 that is the user's responsibility. */
2519 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2522 { /* Caller did not allocate a buffer. Do it for them. */
2523 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2525 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2527 bufp
->allocated
= INIT_BUF_SIZE
;
2530 begalt
= b
= bufp
->buffer
;
2532 /* Loop through the uncompiled pattern until we're at the end. */
2537 /* If this is the end of an included regexp,
2538 pop back to the main regexp and try again. */
2542 pattern
= main_pattern
;
2547 /* If this is the end of the main regexp, we are done. */
2559 /* If there's no special whitespace regexp, treat
2560 spaces normally. And don't try to do this recursively. */
2561 if (!whitespace_regexp
|| in_subpattern
)
2564 /* Peek past following spaces. */
2571 /* If the spaces are followed by a repetition op,
2572 treat them normally. */
2574 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2575 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2578 /* Replace the spaces with the whitespace regexp. */
2582 main_pattern
= pattern
;
2583 p
= pattern
= whitespace_regexp
;
2584 pend
= p
+ strlen ((const char *) p
);
2590 if ( /* If at start of pattern, it's an operator. */
2592 /* If context independent, it's an operator. */
2593 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2594 /* Otherwise, depends on what's come before. */
2595 || at_begline_loc_p (pattern
, p
, syntax
))
2596 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2605 if ( /* If at end of pattern, it's an operator. */
2607 /* If context independent, it's an operator. */
2608 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2609 /* Otherwise, depends on what's next. */
2610 || at_endline_loc_p (p
, pend
, syntax
))
2611 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2620 if ((syntax
& RE_BK_PLUS_QM
)
2621 || (syntax
& RE_LIMITED_OPS
))
2625 /* If there is no previous pattern... */
2628 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2629 FREE_STACK_RETURN (REG_BADRPT
);
2630 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2635 /* 1 means zero (many) matches is allowed. */
2636 boolean zero_times_ok
= 0, many_times_ok
= 0;
2639 /* If there is a sequence of repetition chars, collapse it
2640 down to just one (the right one). We can't combine
2641 interval operators with these because of, e.g., `a{2}*',
2642 which should only match an even number of `a's. */
2646 if ((syntax
& RE_FRUGAL
)
2647 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2651 zero_times_ok
|= c
!= '+';
2652 many_times_ok
|= c
!= '?';
2658 || (!(syntax
& RE_BK_PLUS_QM
)
2659 && (*p
== '+' || *p
== '?')))
2661 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2664 FREE_STACK_RETURN (REG_EESCAPE
);
2665 if (p
[1] == '+' || p
[1] == '?')
2666 PATFETCH (c
); /* Gobble up the backslash. */
2672 /* If we get here, we found another repeat character. */
2676 /* Star, etc. applied to an empty pattern is equivalent
2677 to an empty pattern. */
2678 if (!laststart
|| laststart
== b
)
2681 /* Now we know whether or not zero matches is allowed
2682 and also whether or not two or more matches is allowed. */
2687 boolean simple
= skip_one_char (laststart
) == b
;
2688 size_t startoffset
= 0;
2690 /* Check if the loop can match the empty string. */
2691 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2692 ? on_failure_jump
: on_failure_jump_loop
;
2693 assert (skip_one_char (laststart
) <= b
);
2695 if (!zero_times_ok
&& simple
)
2696 { /* Since simple * loops can be made faster by using
2697 on_failure_keep_string_jump, we turn simple P+
2698 into PP* if P is simple. */
2699 unsigned char *p1
, *p2
;
2700 startoffset
= b
- laststart
;
2701 GET_BUFFER_SPACE (startoffset
);
2702 p1
= b
; p2
= laststart
;
2708 GET_BUFFER_SPACE (6);
2711 STORE_JUMP (ofj
, b
, b
+ 6);
2713 /* Simple * loops can use on_failure_keep_string_jump
2714 depending on what follows. But since we don't know
2715 that yet, we leave the decision up to
2716 on_failure_jump_smart. */
2717 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2718 laststart
+ startoffset
, b
+ 6);
2720 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2725 /* A simple ? pattern. */
2726 assert (zero_times_ok
);
2727 GET_BUFFER_SPACE (3);
2728 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2732 else /* not greedy */
2733 { /* I wish the greedy and non-greedy cases could be merged. */
2735 GET_BUFFER_SPACE (7); /* We might use less. */
2738 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2740 /* The non-greedy multiple match looks like
2741 a repeat..until: we only need a conditional jump
2742 at the end of the loop. */
2743 if (emptyp
) BUF_PUSH (no_op
);
2744 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2745 : on_failure_jump
, b
, laststart
);
2749 /* The repeat...until naturally matches one or more.
2750 To also match zero times, we need to first jump to
2751 the end of the loop (its conditional jump). */
2752 INSERT_JUMP (jump
, laststart
, b
);
2758 /* non-greedy a?? */
2759 INSERT_JUMP (jump
, laststart
, b
+ 3);
2761 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2780 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2782 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2784 /* Ensure that we have enough space to push a charset: the
2785 opcode, the length count, and the bitset; 34 bytes in all. */
2786 GET_BUFFER_SPACE (34);
2790 /* We test `*p == '^' twice, instead of using an if
2791 statement, so we only need one BUF_PUSH. */
2792 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2796 /* Remember the first position in the bracket expression. */
2799 /* Push the number of bytes in the bitmap. */
2800 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2802 /* Clear the whole map. */
2803 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2805 /* charset_not matches newline according to a syntax bit. */
2806 if ((re_opcode_t
) b
[-2] == charset_not
2807 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2808 SET_LIST_BIT ('\n');
2810 /* Read in characters and ranges, setting map bits. */
2813 boolean escaped_char
= false;
2814 const unsigned char *p2
= p
;
2817 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2819 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2820 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2821 So the translation is done later in a loop. Example:
2822 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2825 /* \ might escape characters inside [...] and [^...]. */
2826 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2828 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2831 escaped_char
= true;
2835 /* Could be the end of the bracket expression. If it's
2836 not (i.e., when the bracket expression is `[]' so
2837 far), the ']' character bit gets set way below. */
2838 if (c
== ']' && p2
!= p1
)
2842 /* See if we're at the beginning of a possible character
2845 if (!escaped_char
&&
2846 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2848 /* Leave room for the null. */
2849 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2850 const unsigned char *class_beg
;
2856 /* If pattern is `[[:'. */
2857 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2862 if ((c
== ':' && *p
== ']') || p
== pend
)
2864 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2867 /* This is in any case an invalid class name. */
2872 /* If isn't a word bracketed by `[:' and `:]':
2873 undo the ending character, the letters, and
2874 leave the leading `:' and `[' (but set bits for
2876 if (c
== ':' && *p
== ']')
2878 re_wctype_t cc
= re_wctype (str
);
2881 FREE_STACK_RETURN (REG_ECTYPE
);
2883 /* Throw away the ] at the end of the character
2887 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2890 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2891 if (re_iswctype (btowc (ch
), cc
))
2894 if (c
< (1 << BYTEWIDTH
))
2898 /* Most character classes in a multibyte match
2899 just set a flag. Exceptions are is_blank,
2900 is_digit, is_cntrl, and is_xdigit, since
2901 they can only match ASCII characters. We
2902 don't need to handle them for multibyte.
2903 They are distinguished by a negative wctype. */
2905 /* Setup the gl_state object to its buffer-defined
2906 value. This hardcodes the buffer-global
2907 syntax-table for ASCII chars, while the other chars
2908 will obey syntax-table properties. It's not ideal,
2909 but it's the way it's been done until now. */
2910 SETUP_BUFFER_SYNTAX_TABLE ();
2912 for (ch
= 0; ch
< 256; ++ch
)
2914 c
= RE_CHAR_TO_MULTIBYTE (ch
);
2915 if (! CHAR_BYTE8_P (c
)
2916 && re_iswctype (c
, cc
))
2922 if (ASCII_CHAR_P (c1
))
2924 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2928 SET_RANGE_TABLE_WORK_AREA_BIT
2929 (range_table_work
, re_wctype_to_bit (cc
));
2931 /* In most cases the matching rule for char classes
2932 only uses the syntax table for multibyte chars,
2933 so that the content of the syntax-table it is not
2934 hardcoded in the range_table. SPACE and WORD are
2935 the two exceptions. */
2936 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2937 bufp
->used_syntax
= 1;
2939 /* Repeat the loop. */
2944 /* Go back to right after the "[:". */
2948 /* Because the `:' may starts the range, we
2949 can't simply set bit and repeat the loop.
2950 Instead, just set it to C and handle below. */
2955 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2958 /* Discard the `-'. */
2961 /* Fetch the character which ends the range. */
2964 if (CHAR_BYTE8_P (c1
)
2965 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2966 /* Treat the range from a multibyte character to
2967 raw-byte character as empty. */
2972 /* Range from C to C. */
2977 if (syntax
& RE_NO_EMPTY_RANGES
)
2978 FREE_STACK_RETURN (REG_ERANGEX
);
2979 /* Else, repeat the loop. */
2984 /* Set the range into bitmap */
2985 for (; c
<= c1
; c
++)
2988 if (ch
< (1 << BYTEWIDTH
))
2995 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
2997 if (CHAR_BYTE8_P (c1
))
2998 c
= BYTE8_TO_CHAR (128);
3002 if (CHAR_BYTE8_P (c
))
3004 c
= CHAR_TO_BYTE8 (c
);
3005 c1
= CHAR_TO_BYTE8 (c1
);
3006 for (; c
<= c1
; c
++)
3011 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3015 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3022 /* Discard any (non)matching list bytes that are all 0 at the
3023 end of the map. Decrease the map-length byte too. */
3024 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3028 /* Build real range table from work area. */
3029 if (RANGE_TABLE_WORK_USED (range_table_work
)
3030 || RANGE_TABLE_WORK_BITS (range_table_work
))
3033 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3035 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3036 bytes for flags, two for COUNT, and three bytes for
3038 GET_BUFFER_SPACE (4 + used
* 3);
3040 /* Indicate the existence of range table. */
3041 laststart
[1] |= 0x80;
3043 /* Store the character class flag bits into the range table.
3044 If not in emacs, these flag bits are always 0. */
3045 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3046 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3048 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3049 for (i
= 0; i
< used
; i
++)
3050 STORE_CHARACTER_AND_INCR
3051 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3058 if (syntax
& RE_NO_BK_PARENS
)
3065 if (syntax
& RE_NO_BK_PARENS
)
3072 if (syntax
& RE_NEWLINE_ALT
)
3079 if (syntax
& RE_NO_BK_VBAR
)
3086 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3087 goto handle_interval
;
3093 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3095 /* Do not translate the character after the \, so that we can
3096 distinguish, e.g., \B from \b, even if we normally would
3097 translate, e.g., B to b. */
3103 if (syntax
& RE_NO_BK_PARENS
)
3104 goto normal_backslash
;
3109 regnum_t regnum
= 0;
3112 /* Look for a special (?...) construct */
3113 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3115 PATFETCH (c
); /* Gobble up the '?'. */
3121 case ':': shy
= 1; break;
3123 /* An explicitly specified regnum must start
3126 FREE_STACK_RETURN (REG_BADPAT
);
3127 case '1': case '2': case '3': case '4':
3128 case '5': case '6': case '7': case '8': case '9':
3129 regnum
= 10*regnum
+ (c
- '0'); break;
3131 /* Only (?:...) is supported right now. */
3132 FREE_STACK_RETURN (REG_BADPAT
);
3139 regnum
= ++bufp
->re_nsub
;
3141 { /* It's actually not shy, but explicitly numbered. */
3143 if (regnum
> bufp
->re_nsub
)
3144 bufp
->re_nsub
= regnum
;
3145 else if (regnum
> bufp
->re_nsub
3146 /* Ideally, we'd want to check that the specified
3147 group can't have matched (i.e. all subgroups
3148 using the same regnum are in other branches of
3149 OR patterns), but we don't currently keep track
3150 of enough info to do that easily. */
3151 || group_in_compile_stack (compile_stack
, regnum
))
3152 FREE_STACK_RETURN (REG_BADPAT
);
3155 /* It's really shy. */
3156 regnum
= - bufp
->re_nsub
;
3158 if (COMPILE_STACK_FULL
)
3160 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3161 compile_stack_elt_t
);
3162 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3164 compile_stack
.size
<<= 1;
3167 /* These are the values to restore when we hit end of this
3168 group. They are all relative offsets, so that if the
3169 whole pattern moves because of realloc, they will still
3171 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3172 COMPILE_STACK_TOP
.fixup_alt_jump
3173 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3174 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3175 COMPILE_STACK_TOP
.regnum
= regnum
;
3177 /* Do not push a start_memory for groups beyond the last one
3178 we can represent in the compiled pattern. */
3179 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3180 BUF_PUSH_2 (start_memory
, regnum
);
3182 compile_stack
.avail
++;
3187 /* If we've reached MAX_REGNUM groups, then this open
3188 won't actually generate any code, so we'll have to
3189 clear pending_exact explicitly. */
3195 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3197 if (COMPILE_STACK_EMPTY
)
3199 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3200 goto normal_backslash
;
3202 FREE_STACK_RETURN (REG_ERPAREN
);
3208 /* See similar code for backslashed left paren above. */
3209 if (COMPILE_STACK_EMPTY
)
3211 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3214 FREE_STACK_RETURN (REG_ERPAREN
);
3217 /* Since we just checked for an empty stack above, this
3218 ``can't happen''. */
3219 assert (compile_stack
.avail
!= 0);
3221 /* We don't just want to restore into `regnum', because
3222 later groups should continue to be numbered higher,
3223 as in `(ab)c(de)' -- the second group is #2. */
3226 compile_stack
.avail
--;
3227 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3229 = COMPILE_STACK_TOP
.fixup_alt_jump
3230 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3232 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3233 regnum
= COMPILE_STACK_TOP
.regnum
;
3234 /* If we've reached MAX_REGNUM groups, then this open
3235 won't actually generate any code, so we'll have to
3236 clear pending_exact explicitly. */
3239 /* We're at the end of the group, so now we know how many
3240 groups were inside this one. */
3241 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3242 BUF_PUSH_2 (stop_memory
, regnum
);
3247 case '|': /* `\|'. */
3248 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3249 goto normal_backslash
;
3251 if (syntax
& RE_LIMITED_OPS
)
3254 /* Insert before the previous alternative a jump which
3255 jumps to this alternative if the former fails. */
3256 GET_BUFFER_SPACE (3);
3257 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3261 /* The alternative before this one has a jump after it
3262 which gets executed if it gets matched. Adjust that
3263 jump so it will jump to this alternative's analogous
3264 jump (put in below, which in turn will jump to the next
3265 (if any) alternative's such jump, etc.). The last such
3266 jump jumps to the correct final destination. A picture:
3272 If we are at `b', then fixup_alt_jump right now points to a
3273 three-byte space after `a'. We'll put in the jump, set
3274 fixup_alt_jump to right after `b', and leave behind three
3275 bytes which we'll fill in when we get to after `c'. */
3279 /* Mark and leave space for a jump after this alternative,
3280 to be filled in later either by next alternative or
3281 when know we're at the end of a series of alternatives. */
3283 GET_BUFFER_SPACE (3);
3292 /* If \{ is a literal. */
3293 if (!(syntax
& RE_INTERVALS
)
3294 /* If we're at `\{' and it's not the open-interval
3296 || (syntax
& RE_NO_BK_BRACES
))
3297 goto normal_backslash
;
3301 /* If got here, then the syntax allows intervals. */
3303 /* At least (most) this many matches must be made. */
3304 int lower_bound
= 0, upper_bound
= -1;
3308 GET_UNSIGNED_NUMBER (lower_bound
);
3311 GET_UNSIGNED_NUMBER (upper_bound
);
3313 /* Interval such as `{1}' => match exactly once. */
3314 upper_bound
= lower_bound
;
3316 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3317 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3318 FREE_STACK_RETURN (REG_BADBR
);
3320 if (!(syntax
& RE_NO_BK_BRACES
))
3323 FREE_STACK_RETURN (REG_BADBR
);
3325 FREE_STACK_RETURN (REG_EESCAPE
);
3330 FREE_STACK_RETURN (REG_BADBR
);
3332 /* We just parsed a valid interval. */
3334 /* If it's invalid to have no preceding re. */
3337 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3338 FREE_STACK_RETURN (REG_BADRPT
);
3339 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3342 goto unfetch_interval
;
3345 if (upper_bound
== 0)
3346 /* If the upper bound is zero, just drop the sub pattern
3349 else if (lower_bound
== 1 && upper_bound
== 1)
3350 /* Just match it once: nothing to do here. */
3353 /* Otherwise, we have a nontrivial interval. When
3354 we're all done, the pattern will look like:
3355 set_number_at <jump count> <upper bound>
3356 set_number_at <succeed_n count> <lower bound>
3357 succeed_n <after jump addr> <succeed_n count>
3359 jump_n <succeed_n addr> <jump count>
3360 (The upper bound and `jump_n' are omitted if
3361 `upper_bound' is 1, though.) */
3363 { /* If the upper bound is > 1, we need to insert
3364 more at the end of the loop. */
3365 unsigned int nbytes
= (upper_bound
< 0 ? 3
3366 : upper_bound
> 1 ? 5 : 0);
3367 unsigned int startoffset
= 0;
3369 GET_BUFFER_SPACE (20); /* We might use less. */
3371 if (lower_bound
== 0)
3373 /* A succeed_n that starts with 0 is really a
3374 a simple on_failure_jump_loop. */
3375 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3381 /* Initialize lower bound of the `succeed_n', even
3382 though it will be set during matching by its
3383 attendant `set_number_at' (inserted next),
3384 because `re_compile_fastmap' needs to know.
3385 Jump to the `jump_n' we might insert below. */
3386 INSERT_JUMP2 (succeed_n
, laststart
,
3391 /* Code to initialize the lower bound. Insert
3392 before the `succeed_n'. The `5' is the last two
3393 bytes of this `set_number_at', plus 3 bytes of
3394 the following `succeed_n'. */
3395 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3400 if (upper_bound
< 0)
3402 /* A negative upper bound stands for infinity,
3403 in which case it degenerates to a plain jump. */
3404 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3407 else if (upper_bound
> 1)
3408 { /* More than one repetition is allowed, so
3409 append a backward jump to the `succeed_n'
3410 that starts this interval.
3412 When we've reached this during matching,
3413 we'll have matched the interval once, so
3414 jump back only `upper_bound - 1' times. */
3415 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3419 /* The location we want to set is the second
3420 parameter of the `jump_n'; that is `b-2' as
3421 an absolute address. `laststart' will be
3422 the `set_number_at' we're about to insert;
3423 `laststart+3' the number to set, the source
3424 for the relative address. But we are
3425 inserting into the middle of the pattern --
3426 so everything is getting moved up by 5.
3427 Conclusion: (b - 2) - (laststart + 3) + 5,
3428 i.e., b - laststart.
3430 We insert this at the beginning of the loop
3431 so that if we fail during matching, we'll
3432 reinitialize the bounds. */
3433 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3434 upper_bound
- 1, b
);
3439 beg_interval
= NULL
;
3444 /* If an invalid interval, match the characters as literals. */
3445 assert (beg_interval
);
3447 beg_interval
= NULL
;
3449 /* normal_char and normal_backslash need `c'. */
3452 if (!(syntax
& RE_NO_BK_BRACES
))
3454 assert (p
> pattern
&& p
[-1] == '\\');
3455 goto normal_backslash
;
3461 /* There is no way to specify the before_dot and after_dot
3462 operators. rms says this is ok. --karl */
3470 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3476 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3482 BUF_PUSH_2 (categoryspec
, c
);
3488 BUF_PUSH_2 (notcategoryspec
, c
);
3494 if (syntax
& RE_NO_GNU_OPS
)
3497 BUF_PUSH_2 (syntaxspec
, Sword
);
3502 if (syntax
& RE_NO_GNU_OPS
)
3505 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3510 if (syntax
& RE_NO_GNU_OPS
)
3516 if (syntax
& RE_NO_GNU_OPS
)
3522 if (syntax
& RE_NO_GNU_OPS
)
3531 FREE_STACK_RETURN (REG_BADPAT
);
3535 if (syntax
& RE_NO_GNU_OPS
)
3537 BUF_PUSH (wordbound
);
3541 if (syntax
& RE_NO_GNU_OPS
)
3543 BUF_PUSH (notwordbound
);
3547 if (syntax
& RE_NO_GNU_OPS
)
3553 if (syntax
& RE_NO_GNU_OPS
)
3558 case '1': case '2': case '3': case '4': case '5':
3559 case '6': case '7': case '8': case '9':
3563 if (syntax
& RE_NO_BK_REFS
)
3564 goto normal_backslash
;
3568 if (reg
> bufp
->re_nsub
|| reg
< 1
3569 /* Can't back reference to a subexp before its end. */
3570 || group_in_compile_stack (compile_stack
, reg
))
3571 FREE_STACK_RETURN (REG_ESUBREG
);
3574 BUF_PUSH_2 (duplicate
, reg
);
3581 if (syntax
& RE_BK_PLUS_QM
)
3584 goto normal_backslash
;
3588 /* You might think it would be useful for \ to mean
3589 not to translate; but if we don't translate it
3590 it will never match anything. */
3597 /* Expects the character in `c'. */
3599 /* If no exactn currently being built. */
3602 /* If last exactn not at current position. */
3603 || pending_exact
+ *pending_exact
+ 1 != b
3605 /* We have only one byte following the exactn for the count. */
3606 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3608 /* If followed by a repetition operator. */
3609 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3610 || ((syntax
& RE_BK_PLUS_QM
)
3611 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3612 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3613 || ((syntax
& RE_INTERVALS
)
3614 && ((syntax
& RE_NO_BK_BRACES
)
3615 ? p
!= pend
&& *p
== '{'
3616 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3618 /* Start building a new exactn. */
3622 BUF_PUSH_2 (exactn
, 0);
3623 pending_exact
= b
- 1;
3626 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3633 len
= CHAR_STRING (c
, b
);
3638 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3639 if (! CHAR_BYTE8_P (c1
))
3641 re_wchar_t c2
= TRANSLATE (c1
);
3643 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3649 (*pending_exact
) += len
;
3654 } /* while p != pend */
3657 /* Through the pattern now. */
3661 if (!COMPILE_STACK_EMPTY
)
3662 FREE_STACK_RETURN (REG_EPAREN
);
3664 /* If we don't want backtracking, force success
3665 the first time we reach the end of the compiled pattern. */
3666 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3669 /* We have succeeded; set the length of the buffer. */
3670 bufp
->used
= b
- bufp
->buffer
;
3675 re_compile_fastmap (bufp
);
3676 DEBUG_PRINT ("\nCompiled pattern: \n");
3677 print_compiled_pattern (bufp
);
3682 #ifndef MATCH_MAY_ALLOCATE
3683 /* Initialize the failure stack to the largest possible stack. This
3684 isn't necessary unless we're trying to avoid calling alloca in
3685 the search and match routines. */
3687 int num_regs
= bufp
->re_nsub
+ 1;
3689 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3691 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3692 falk_stack
.stack
= realloc (fail_stack
.stack
,
3693 fail_stack
.size
* sizeof *falk_stack
.stack
);
3696 regex_grow_registers (num_regs
);
3698 #endif /* not MATCH_MAY_ALLOCATE */
3700 FREE_STACK_RETURN (REG_NOERROR
);
3701 } /* regex_compile */
3703 /* Subroutines for `regex_compile'. */
3705 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3708 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3710 *loc
= (unsigned char) op
;
3711 STORE_NUMBER (loc
+ 1, arg
);
3715 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3718 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3720 *loc
= (unsigned char) op
;
3721 STORE_NUMBER (loc
+ 1, arg1
);
3722 STORE_NUMBER (loc
+ 3, arg2
);
3726 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3727 for OP followed by two-byte integer parameter ARG. */
3730 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3732 register unsigned char *pfrom
= end
;
3733 register unsigned char *pto
= end
+ 3;
3735 while (pfrom
!= loc
)
3738 store_op1 (op
, loc
, arg
);
3742 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3745 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3747 register unsigned char *pfrom
= end
;
3748 register unsigned char *pto
= end
+ 5;
3750 while (pfrom
!= loc
)
3753 store_op2 (op
, loc
, arg1
, arg2
);
3757 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3758 after an alternative or a begin-subexpression. We assume there is at
3759 least one character before the ^. */
3762 at_begline_loc_p (const re_char
*pattern
, const re_char
*p
, reg_syntax_t syntax
)
3764 re_char
*prev
= p
- 2;
3765 boolean odd_backslashes
;
3767 /* After a subexpression? */
3769 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3771 /* After an alternative? */
3772 else if (*prev
== '|')
3773 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3775 /* After a shy subexpression? */
3776 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3778 /* Skip over optional regnum. */
3779 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3782 if (!(prev
- 2 >= pattern
3783 && prev
[-1] == '?' && prev
[-2] == '('))
3786 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3791 /* Count the number of preceding backslashes. */
3793 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3795 return (p
- prev
) & odd_backslashes
;
3799 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3800 at least one character after the $, i.e., `P < PEND'. */
3803 at_endline_loc_p (const re_char
*p
, const re_char
*pend
, reg_syntax_t syntax
)
3806 boolean next_backslash
= *next
== '\\';
3807 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3810 /* Before a subexpression? */
3811 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3812 : next_backslash
&& next_next
&& *next_next
== ')')
3813 /* Before an alternative? */
3814 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3815 : next_backslash
&& next_next
&& *next_next
== '|');
3819 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3820 false if it's not. */
3823 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3825 ssize_t this_element
;
3827 for (this_element
= compile_stack
.avail
- 1;
3830 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3837 If fastmap is non-NULL, go through the pattern and fill fastmap
3838 with all the possible leading chars. If fastmap is NULL, don't
3839 bother filling it up (obviously) and only return whether the
3840 pattern could potentially match the empty string.
3842 Return 1 if p..pend might match the empty string.
3843 Return 0 if p..pend matches at least one char.
3844 Return -1 if fastmap was not updated accurately. */
3847 analyse_first (const re_char
*p
, const re_char
*pend
, char *fastmap
, const int multibyte
)
3852 /* If all elements for base leading-codes in fastmap is set, this
3853 flag is set true. */
3854 boolean match_any_multibyte_characters
= false;
3858 /* The loop below works as follows:
3859 - It has a working-list kept in the PATTERN_STACK and which basically
3860 starts by only containing a pointer to the first operation.
3861 - If the opcode we're looking at is a match against some set of
3862 chars, then we add those chars to the fastmap and go on to the
3863 next work element from the worklist (done via `break').
3864 - If the opcode is a control operator on the other hand, we either
3865 ignore it (if it's meaningless at this point, such as `start_memory')
3866 or execute it (if it's a jump). If the jump has several destinations
3867 (i.e. `on_failure_jump'), then we push the other destination onto the
3869 We guarantee termination by ignoring backward jumps (more or less),
3870 so that `p' is monotonically increasing. More to the point, we
3871 never set `p' (or push) anything `<= p1'. */
3875 /* `p1' is used as a marker of how far back a `on_failure_jump'
3876 can go without being ignored. It is normally equal to `p'
3877 (which prevents any backward `on_failure_jump') except right
3878 after a plain `jump', to allow patterns such as:
3881 10: on_failure_jump 3
3882 as used for the *? operator. */
3891 /* If the first character has to match a backreference, that means
3892 that the group was empty (since it already matched). Since this
3893 is the only case that interests us here, we can assume that the
3894 backreference must match the empty string. */
3899 /* Following are the cases which match a character. These end
3905 /* If multibyte is nonzero, the first byte of each
3906 character is an ASCII or a leading code. Otherwise,
3907 each byte is a character. Thus, this works in both
3912 /* For the case of matching this unibyte regex
3913 against multibyte, we must set a leading code of
3914 the corresponding multibyte character. */
3915 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3917 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3924 /* We could put all the chars except for \n (and maybe \0)
3925 but we don't bother since it is generally not worth it. */
3926 if (!fastmap
) break;
3931 if (!fastmap
) break;
3933 /* Chars beyond end of bitmap are possible matches. */
3934 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3935 j
< (1 << BYTEWIDTH
); j
++)
3941 if (!fastmap
) break;
3942 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3943 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3945 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3949 if (/* Any leading code can possibly start a character
3950 which doesn't match the specified set of characters. */
3953 /* If we can match a character class, we can match any
3954 multibyte characters. */
3955 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3956 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3959 if (match_any_multibyte_characters
== false)
3961 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3962 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3964 match_any_multibyte_characters
= true;
3968 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3969 && match_any_multibyte_characters
== false)
3971 /* Set fastmap[I] to 1 where I is a leading code of each
3972 multibyte character in the range table. */
3974 unsigned char lc1
, lc2
;
3976 /* Make P points the range table. `+ 2' is to skip flag
3977 bits for a character class. */
3978 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3980 /* Extract the number of ranges in range table into COUNT. */
3981 EXTRACT_NUMBER_AND_INCR (count
, p
);
3982 for (; count
> 0; count
--, p
+= 3)
3984 /* Extract the start and end of each range. */
3985 EXTRACT_CHARACTER (c
, p
);
3986 lc1
= CHAR_LEADING_CODE (c
);
3988 EXTRACT_CHARACTER (c
, p
);
3989 lc2
= CHAR_LEADING_CODE (c
);
3990 for (j
= lc1
; j
<= lc2
; j
++)
3999 if (!fastmap
) break;
4001 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4003 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4004 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4008 /* This match depends on text properties. These end with
4009 aborting optimizations. */
4013 case notcategoryspec
:
4014 if (!fastmap
) break;
4015 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4017 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4018 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4021 /* Any leading code can possibly start a character which
4022 has or doesn't has the specified category. */
4023 if (match_any_multibyte_characters
== false)
4025 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4026 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4028 match_any_multibyte_characters
= true;
4032 /* All cases after this match the empty string. These end with
4054 EXTRACT_NUMBER_AND_INCR (j
, p
);
4056 /* Backward jumps can only go back to code that we've already
4057 visited. `re_compile' should make sure this is true. */
4062 case on_failure_jump
:
4063 case on_failure_keep_string_jump
:
4064 case on_failure_jump_loop
:
4065 case on_failure_jump_nastyloop
:
4066 case on_failure_jump_smart
:
4072 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4073 to jump back to "just after here". */
4076 case on_failure_jump
:
4077 case on_failure_keep_string_jump
:
4078 case on_failure_jump_nastyloop
:
4079 case on_failure_jump_loop
:
4080 case on_failure_jump_smart
:
4081 EXTRACT_NUMBER_AND_INCR (j
, p
);
4083 ; /* Backward jump to be ignored. */
4085 { /* We have to look down both arms.
4086 We first go down the "straight" path so as to minimize
4087 stack usage when going through alternatives. */
4088 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4096 /* This code simply does not properly handle forward jump_n. */
4097 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4099 /* jump_n can either jump or fall through. The (backward) jump
4100 case has already been handled, so we only need to look at the
4101 fallthrough case. */
4105 /* If N == 0, it should be an on_failure_jump_loop instead. */
4106 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4108 /* We only care about one iteration of the loop, so we don't
4109 need to consider the case where this behaves like an
4126 abort (); /* We have listed all the cases. */
4129 /* Getting here means we have found the possible starting
4130 characters for one path of the pattern -- and that the empty
4131 string does not match. We need not follow this path further. */
4135 /* We reached the end without matching anything. */
4138 } /* analyse_first */
4140 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4141 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4142 characters can start a string that matches the pattern. This fastmap
4143 is used by re_search to skip quickly over impossible starting points.
4145 Character codes above (1 << BYTEWIDTH) are not represented in the
4146 fastmap, but the leading codes are represented. Thus, the fastmap
4147 indicates which character sets could start a match.
4149 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4150 area as BUFP->fastmap.
4152 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4155 Returns 0 if we succeed, -2 if an internal error. */
4158 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4160 char *fastmap
= bufp
->fastmap
;
4163 assert (fastmap
&& bufp
->buffer
);
4165 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4166 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4168 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4169 fastmap
, RE_MULTIBYTE_P (bufp
));
4170 bufp
->can_be_null
= (analysis
!= 0);
4172 } /* re_compile_fastmap */
4174 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4175 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4176 this memory for recording register information. STARTS and ENDS
4177 must be allocated using the malloc library routine, and must each
4178 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4180 If NUM_REGS == 0, then subsequent matches should allocate their own
4183 Unless this function is called, the first search or match using
4184 PATTERN_BUFFER will allocate its own register data, without
4185 freeing the old data. */
4188 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4192 bufp
->regs_allocated
= REGS_REALLOCATE
;
4193 regs
->num_regs
= num_regs
;
4194 regs
->start
= starts
;
4199 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4201 regs
->start
= regs
->end
= (regoff_t
*) 0;
4204 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4206 /* Searching routines. */
4208 /* Like re_search_2, below, but only one string is specified, and
4209 doesn't let you say where to stop matching. */
4212 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4213 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4215 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4218 WEAK_ALIAS (__re_search
, re_search
)
4220 /* Head address of virtual concatenation of string. */
4221 #define HEAD_ADDR_VSTRING(P) \
4222 (((P) >= size1 ? string2 : string1))
4224 /* Address of POS in the concatenation of virtual string. */
4225 #define POS_ADDR_VSTRING(POS) \
4226 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4228 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4229 virtual concatenation of STRING1 and STRING2, starting first at index
4230 STARTPOS, then at STARTPOS + 1, and so on.
4232 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4234 RANGE is how far to scan while trying to match. RANGE = 0 means try
4235 only at STARTPOS; in general, the last start tried is STARTPOS +
4238 In REGS, return the indices of the virtual concatenation of STRING1
4239 and STRING2 that matched the entire BUFP->buffer and its contained
4242 Do not consider matching one past the index STOP in the virtual
4243 concatenation of STRING1 and STRING2.
4245 We return either the position in the strings at which the match was
4246 found, -1 if no match, or -2 if error (such as failure
4250 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4251 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4252 struct re_registers
*regs
, ssize_t stop
)
4255 re_char
*string1
= (re_char
*) str1
;
4256 re_char
*string2
= (re_char
*) str2
;
4257 register char *fastmap
= bufp
->fastmap
;
4258 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4259 size_t total_size
= size1
+ size2
;
4260 ssize_t endpos
= startpos
+ range
;
4261 boolean anchored_start
;
4262 /* Nonzero if we are searching multibyte string. */
4263 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4265 /* Check for out-of-range STARTPOS. */
4266 if (startpos
< 0 || startpos
> total_size
)
4269 /* Fix up RANGE if it might eventually take us outside
4270 the virtual concatenation of STRING1 and STRING2.
4271 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4273 range
= 0 - startpos
;
4274 else if (endpos
> total_size
)
4275 range
= total_size
- startpos
;
4277 /* If the search isn't to be a backwards one, don't waste time in a
4278 search for a pattern anchored at beginning of buffer. */
4279 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4288 /* In a forward search for something that starts with \=.
4289 don't keep searching past point. */
4290 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4292 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4298 /* Update the fastmap now if not correct already. */
4299 if (fastmap
&& !bufp
->fastmap_accurate
)
4300 re_compile_fastmap (bufp
);
4302 /* See whether the pattern is anchored. */
4303 anchored_start
= (bufp
->buffer
[0] == begline
);
4306 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4308 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4310 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4314 /* Loop through the string, looking for a place to start matching. */
4317 /* If the pattern is anchored,
4318 skip quickly past places we cannot match.
4319 We don't bother to treat startpos == 0 specially
4320 because that case doesn't repeat. */
4321 if (anchored_start
&& startpos
> 0)
4323 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4324 : string2
[startpos
- size1
- 1])
4329 /* If a fastmap is supplied, skip quickly over characters that
4330 cannot be the start of a match. If the pattern can match the
4331 null string, however, we don't need to skip characters; we want
4332 the first null string. */
4333 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4335 register re_char
*d
;
4336 register re_wchar_t buf_ch
;
4338 d
= POS_ADDR_VSTRING (startpos
);
4340 if (range
> 0) /* Searching forwards. */
4342 register int lim
= 0;
4343 ssize_t irange
= range
;
4345 if (startpos
< size1
&& startpos
+ range
>= size1
)
4346 lim
= range
- (size1
- startpos
);
4348 /* Written out as an if-else to avoid testing `translate'
4350 if (RE_TRANSLATE_P (translate
))
4357 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4358 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4359 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4362 range
-= buf_charlen
;
4368 register re_wchar_t ch
, translated
;
4371 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4372 translated
= RE_TRANSLATE (translate
, ch
);
4373 if (translated
!= ch
4374 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4376 if (fastmap
[buf_ch
])
4389 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4390 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4392 range
-= buf_charlen
;
4396 while (range
> lim
&& !fastmap
[*d
])
4402 startpos
+= irange
- range
;
4404 else /* Searching backwards. */
4408 buf_ch
= STRING_CHAR (d
);
4409 buf_ch
= TRANSLATE (buf_ch
);
4410 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4415 register re_wchar_t ch
, translated
;
4418 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4419 translated
= TRANSLATE (ch
);
4420 if (translated
!= ch
4421 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4423 if (! fastmap
[TRANSLATE (buf_ch
)])
4429 /* If can't match the null string, and that's all we have left, fail. */
4430 if (range
>= 0 && startpos
== total_size
&& fastmap
4431 && !bufp
->can_be_null
)
4434 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4435 startpos
, regs
, stop
);
4448 /* Update STARTPOS to the next character boundary. */
4451 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4452 int len
= BYTES_BY_CHAR_HEAD (*p
);
4470 /* Update STARTPOS to the previous character boundary. */
4473 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4475 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4477 /* Find the head of multibyte form. */
4478 PREV_CHAR_BOUNDARY (p
, phead
);
4479 range
+= p0
- 1 - p
;
4483 startpos
-= p0
- 1 - p
;
4489 WEAK_ALIAS (__re_search_2
, re_search_2
)
4491 /* Declarations and macros for re_match_2. */
4493 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4494 register ssize_t len
,
4495 RE_TRANSLATE_TYPE translate
,
4496 const int multibyte
);
4498 /* This converts PTR, a pointer into one of the search strings `string1'
4499 and `string2' into an offset from the beginning of that string. */
4500 #define POINTER_TO_OFFSET(ptr) \
4501 (FIRST_STRING_P (ptr) \
4503 : (ptr) - string2 + (ptrdiff_t) size1)
4505 /* Call before fetching a character with *d. This switches over to
4506 string2 if necessary.
4507 Check re_match_2_internal for a discussion of why end_match_2 might
4508 not be within string2 (but be equal to end_match_1 instead). */
4509 #define PREFETCH() \
4512 /* End of string2 => fail. */ \
4513 if (dend == end_match_2) \
4515 /* End of string1 => advance to string2. */ \
4517 dend = end_match_2; \
4520 /* Call before fetching a char with *d if you already checked other limits.
4521 This is meant for use in lookahead operations like wordend, etc..
4522 where we might need to look at parts of the string that might be
4523 outside of the LIMITs (i.e past `stop'). */
4524 #define PREFETCH_NOLIMIT() \
4528 dend = end_match_2; \
4531 /* Test if at very beginning or at very end of the virtual concatenation
4532 of `string1' and `string2'. If only one string, it's `string2'. */
4533 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4534 #define AT_STRINGS_END(d) ((d) == end2)
4536 /* Disabled due to a compiler bug -- see comment at case wordbound */
4538 /* The comment at case wordbound is following one, but we don't use
4539 AT_WORD_BOUNDARY anymore to support multibyte form.
4541 The DEC Alpha C compiler 3.x generates incorrect code for the
4542 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4543 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4544 macro and introducing temporary variables works around the bug. */
4547 /* Test if D points to a character which is word-constituent. We have
4548 two special cases to check for: if past the end of string1, look at
4549 the first character in string2; and if before the beginning of
4550 string2, look at the last character in string1. */
4551 #define WORDCHAR_P(d) \
4552 (SYNTAX ((d) == end1 ? *string2 \
4553 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4556 /* Test if the character before D and the one at D differ with respect
4557 to being word-constituent. */
4558 #define AT_WORD_BOUNDARY(d) \
4559 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4560 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4563 /* Free everything we malloc. */
4564 #ifdef MATCH_MAY_ALLOCATE
4565 # define FREE_VAR(var) \
4573 # define FREE_VARIABLES() \
4575 REGEX_FREE_STACK (fail_stack.stack); \
4576 FREE_VAR (regstart); \
4577 FREE_VAR (regend); \
4578 FREE_VAR (best_regstart); \
4579 FREE_VAR (best_regend); \
4582 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4583 #endif /* not MATCH_MAY_ALLOCATE */
4586 /* Optimization routines. */
4588 /* If the operation is a match against one or more chars,
4589 return a pointer to the next operation, else return NULL. */
4591 skip_one_char (const re_char
*p
)
4604 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4607 p
= CHARSET_RANGE_TABLE (p
- 1);
4608 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4609 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4612 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4619 case notcategoryspec
:
4631 /* Jump over non-matching operations. */
4633 skip_noops (const re_char
*p
, const re_char
*pend
)
4647 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4658 /* Non-zero if "p1 matches something" implies "p2 fails". */
4660 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const re_char
*p1
, const re_char
*p2
)
4663 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4664 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4666 assert (p1
>= bufp
->buffer
&& p1
< pend
4667 && p2
>= bufp
->buffer
&& p2
<= pend
);
4669 /* Skip over open/close-group commands.
4670 If what follows this loop is a ...+ construct,
4671 look at what begins its body, since we will have to
4672 match at least one of that. */
4673 p2
= skip_noops (p2
, pend
);
4674 /* The same skip can be done for p1, except that this function
4675 is only used in the case where p1 is a simple match operator. */
4676 /* p1 = skip_noops (p1, pend); */
4678 assert (p1
>= bufp
->buffer
&& p1
< pend
4679 && p2
>= bufp
->buffer
&& p2
<= pend
);
4681 op2
= p2
== pend
? succeed
: *p2
;
4687 /* If we're at the end of the pattern, we can change. */
4688 if (skip_one_char (p1
))
4690 DEBUG_PRINT (" End of pattern: fast loop.\n");
4698 register re_wchar_t c
4699 = (re_opcode_t
) *p2
== endline
? '\n'
4700 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4702 if ((re_opcode_t
) *p1
== exactn
)
4704 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4706 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4711 else if ((re_opcode_t
) *p1
== charset
4712 || (re_opcode_t
) *p1
== charset_not
)
4714 int not = (re_opcode_t
) *p1
== charset_not
;
4716 /* Test if C is listed in charset (or charset_not)
4718 if (! multibyte
|| IS_REAL_ASCII (c
))
4720 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4721 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4724 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4725 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4727 /* `not' is equal to 1 if c would match, which means
4728 that we can't change to pop_failure_jump. */
4731 DEBUG_PRINT (" No match => fast loop.\n");
4735 else if ((re_opcode_t
) *p1
== anychar
4738 DEBUG_PRINT (" . != \\n => fast loop.\n");
4746 if ((re_opcode_t
) *p1
== exactn
)
4747 /* Reuse the code above. */
4748 return mutually_exclusive_p (bufp
, p2
, p1
);
4750 /* It is hard to list up all the character in charset
4751 P2 if it includes multibyte character. Give up in
4753 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4755 /* Now, we are sure that P2 has no range table.
4756 So, for the size of bitmap in P2, `p2[1]' is
4757 enough. But P1 may have range table, so the
4758 size of bitmap table of P1 is extracted by
4759 using macro `CHARSET_BITMAP_SIZE'.
4761 In a multibyte case, we know that all the character
4762 listed in P2 is ASCII. In a unibyte case, P1 has only a
4763 bitmap table. So, in both cases, it is enough to test
4764 only the bitmap table of P1. */
4766 if ((re_opcode_t
) *p1
== charset
)
4769 /* We win if the charset inside the loop
4770 has no overlap with the one after the loop. */
4773 && idx
< CHARSET_BITMAP_SIZE (p1
));
4775 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4779 || idx
== CHARSET_BITMAP_SIZE (p1
))
4781 DEBUG_PRINT (" No match => fast loop.\n");
4785 else if ((re_opcode_t
) *p1
== charset_not
)
4788 /* We win if the charset_not inside the loop lists
4789 every character listed in the charset after. */
4790 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4791 if (! (p2
[2 + idx
] == 0
4792 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4793 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4798 DEBUG_PRINT (" No match => fast loop.\n");
4811 /* Reuse the code above. */
4812 return mutually_exclusive_p (bufp
, p2
, p1
);
4814 /* When we have two charset_not, it's very unlikely that
4815 they don't overlap. The union of the two sets of excluded
4816 chars should cover all possible chars, which, as a matter of
4817 fact, is virtually impossible in multibyte buffers. */
4823 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4825 return ((re_opcode_t
) *p1
== syntaxspec
4826 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4828 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4831 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4833 return ((re_opcode_t
) *p1
== notsyntaxspec
4834 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4836 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4839 return (((re_opcode_t
) *p1
== notsyntaxspec
4840 || (re_opcode_t
) *p1
== syntaxspec
)
4845 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4846 case notcategoryspec
:
4847 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4859 /* Matching routines. */
4861 #ifndef emacs /* Emacs never uses this. */
4862 /* re_match is like re_match_2 except it takes only a single string. */
4865 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4866 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4868 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4869 size
, pos
, regs
, size
);
4872 WEAK_ALIAS (__re_match
, re_match
)
4873 #endif /* not emacs */
4876 /* In Emacs, this is the string or buffer in which we
4877 are matching. It is used for looking up syntax properties. */
4878 Lisp_Object re_match_object
;
4881 /* re_match_2 matches the compiled pattern in BUFP against the
4882 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4883 and SIZE2, respectively). We start matching at POS, and stop
4886 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4887 store offsets for the substring each group matched in REGS. See the
4888 documentation for exactly how many groups we fill.
4890 We return -1 if no match, -2 if an internal error (such as the
4891 failure stack overflowing). Otherwise, we return the length of the
4892 matched substring. */
4895 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4896 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4897 struct re_registers
*regs
, ssize_t stop
)
4903 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4904 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4905 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4908 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4909 (re_char
*) string2
, size2
,
4913 WEAK_ALIAS (__re_match_2
, re_match_2
)
4916 /* This is a separate function so that we can force an alloca cleanup
4919 re_match_2_internal (struct re_pattern_buffer
*bufp
, const re_char
*string1
,
4920 size_t size1
, const re_char
*string2
, size_t size2
,
4921 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4923 /* General temporaries. */
4927 /* Just past the end of the corresponding string. */
4928 re_char
*end1
, *end2
;
4930 /* Pointers into string1 and string2, just past the last characters in
4931 each to consider matching. */
4932 re_char
*end_match_1
, *end_match_2
;
4934 /* Where we are in the data, and the end of the current string. */
4937 /* Used sometimes to remember where we were before starting matching
4938 an operator so that we can go back in case of failure. This "atomic"
4939 behavior of matching opcodes is indispensable to the correctness
4940 of the on_failure_keep_string_jump optimization. */
4943 /* Where we are in the pattern, and the end of the pattern. */
4944 re_char
*p
= bufp
->buffer
;
4945 re_char
*pend
= p
+ bufp
->used
;
4947 /* We use this to map every character in the string. */
4948 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4950 /* Nonzero if BUFP is setup from a multibyte regex. */
4951 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4953 /* Nonzero if STRING1/STRING2 are multibyte. */
4954 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4956 /* Failure point stack. Each place that can handle a failure further
4957 down the line pushes a failure point on this stack. It consists of
4958 regstart, and regend for all registers corresponding to
4959 the subexpressions we're currently inside, plus the number of such
4960 registers, and, finally, two char *'s. The first char * is where
4961 to resume scanning the pattern; the second one is where to resume
4962 scanning the strings. */
4963 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4964 fail_stack_type fail_stack
;
4966 #ifdef DEBUG_COMPILES_ARGUMENTS
4967 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4970 #if defined REL_ALLOC && defined REGEX_MALLOC
4971 /* This holds the pointer to the failure stack, when
4972 it is allocated relocatably. */
4973 fail_stack_elt_t
*failure_stack_ptr
;
4976 /* We fill all the registers internally, independent of what we
4977 return, for use in backreferences. The number here includes
4978 an element for register zero. */
4979 size_t num_regs
= bufp
->re_nsub
+ 1;
4981 /* Information on the contents of registers. These are pointers into
4982 the input strings; they record just what was matched (on this
4983 attempt) by a subexpression part of the pattern, that is, the
4984 regnum-th regstart pointer points to where in the pattern we began
4985 matching and the regnum-th regend points to right after where we
4986 stopped matching the regnum-th subexpression. (The zeroth register
4987 keeps track of what the whole pattern matches.) */
4988 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4989 re_char
**regstart
, **regend
;
4992 /* The following record the register info as found in the above
4993 variables when we find a match better than any we've seen before.
4994 This happens as we backtrack through the failure points, which in
4995 turn happens only if we have not yet matched the entire string. */
4996 unsigned best_regs_set
= false;
4997 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4998 re_char
**best_regstart
, **best_regend
;
5001 /* Logically, this is `best_regend[0]'. But we don't want to have to
5002 allocate space for that if we're not allocating space for anything
5003 else (see below). Also, we never need info about register 0 for
5004 any of the other register vectors, and it seems rather a kludge to
5005 treat `best_regend' differently than the rest. So we keep track of
5006 the end of the best match so far in a separate variable. We
5007 initialize this to NULL so that when we backtrack the first time
5008 and need to test it, it's not garbage. */
5009 re_char
*match_end
= NULL
;
5011 #ifdef DEBUG_COMPILES_ARGUMENTS
5012 /* Counts the total number of registers pushed. */
5013 unsigned num_regs_pushed
= 0;
5016 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5020 #ifdef MATCH_MAY_ALLOCATE
5021 /* Do not bother to initialize all the register variables if there are
5022 no groups in the pattern, as it takes a fair amount of time. If
5023 there are groups, we include space for register 0 (the whole
5024 pattern), even though we never use it, since it simplifies the
5025 array indexing. We should fix this. */
5028 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5029 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5030 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5031 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5033 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5041 /* We must initialize all our variables to NULL, so that
5042 `FREE_VARIABLES' doesn't try to free them. */
5043 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5045 #endif /* MATCH_MAY_ALLOCATE */
5047 /* The starting position is bogus. */
5048 if (pos
< 0 || pos
> size1
+ size2
)
5054 /* Initialize subexpression text positions to -1 to mark ones that no
5055 start_memory/stop_memory has been seen for. Also initialize the
5056 register information struct. */
5057 for (reg
= 1; reg
< num_regs
; reg
++)
5058 regstart
[reg
] = regend
[reg
] = NULL
;
5060 /* We move `string1' into `string2' if the latter's empty -- but not if
5061 `string1' is null. */
5062 if (size2
== 0 && string1
!= NULL
)
5069 end1
= string1
+ size1
;
5070 end2
= string2
+ size2
;
5072 /* `p' scans through the pattern as `d' scans through the data.
5073 `dend' is the end of the input string that `d' points within. `d'
5074 is advanced into the following input string whenever necessary, but
5075 this happens before fetching; therefore, at the beginning of the
5076 loop, `d' can be pointing at the end of a string, but it cannot
5080 /* Only match within string2. */
5081 d
= string2
+ pos
- size1
;
5082 dend
= end_match_2
= string2
+ stop
- size1
;
5083 end_match_1
= end1
; /* Just to give it a value. */
5089 /* Only match within string1. */
5090 end_match_1
= string1
+ stop
;
5092 When we reach end_match_1, PREFETCH normally switches to string2.
5093 But in the present case, this means that just doing a PREFETCH
5094 makes us jump from `stop' to `gap' within the string.
5095 What we really want here is for the search to stop as
5096 soon as we hit end_match_1. That's why we set end_match_2
5097 to end_match_1 (since PREFETCH fails as soon as we hit
5099 end_match_2
= end_match_1
;
5102 { /* It's important to use this code when stop == size so that
5103 moving `d' from end1 to string2 will not prevent the d == dend
5104 check from catching the end of string. */
5106 end_match_2
= string2
+ stop
- size1
;
5112 DEBUG_PRINT ("The compiled pattern is: ");
5113 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5114 DEBUG_PRINT ("The string to match is: `");
5115 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5116 DEBUG_PRINT ("'\n");
5118 /* This loops over pattern commands. It exits by returning from the
5119 function if the match is complete, or it drops through if the match
5120 fails at this starting point in the input data. */
5123 DEBUG_PRINT ("\n%p: ", p
);
5129 /* End of pattern means we might have succeeded. */
5130 DEBUG_PRINT ("end of pattern ... ");
5132 /* If we haven't matched the entire string, and we want the
5133 longest match, try backtracking. */
5134 if (d
!= end_match_2
)
5136 /* 1 if this match ends in the same string (string1 or string2)
5137 as the best previous match. */
5138 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5139 == FIRST_STRING_P (d
));
5140 /* 1 if this match is the best seen so far. */
5141 boolean best_match_p
;
5143 /* AIX compiler got confused when this was combined
5144 with the previous declaration. */
5146 best_match_p
= d
> match_end
;
5148 best_match_p
= !FIRST_STRING_P (d
);
5150 DEBUG_PRINT ("backtracking.\n");
5152 if (!FAIL_STACK_EMPTY ())
5153 { /* More failure points to try. */
5155 /* If exceeds best match so far, save it. */
5156 if (!best_regs_set
|| best_match_p
)
5158 best_regs_set
= true;
5161 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5163 for (reg
= 1; reg
< num_regs
; reg
++)
5165 best_regstart
[reg
] = regstart
[reg
];
5166 best_regend
[reg
] = regend
[reg
];
5172 /* If no failure points, don't restore garbage. And if
5173 last match is real best match, don't restore second
5175 else if (best_regs_set
&& !best_match_p
)
5178 /* Restore best match. It may happen that `dend ==
5179 end_match_1' while the restored d is in string2.
5180 For example, the pattern `x.*y.*z' against the
5181 strings `x-' and `y-z-', if the two strings are
5182 not consecutive in memory. */
5183 DEBUG_PRINT ("Restoring best registers.\n");
5186 dend
= ((d
>= string1
&& d
<= end1
)
5187 ? end_match_1
: end_match_2
);
5189 for (reg
= 1; reg
< num_regs
; reg
++)
5191 regstart
[reg
] = best_regstart
[reg
];
5192 regend
[reg
] = best_regend
[reg
];
5195 } /* d != end_match_2 */
5198 DEBUG_PRINT ("Accepting match.\n");
5200 /* If caller wants register contents data back, do it. */
5201 if (regs
&& !bufp
->no_sub
)
5203 /* Have the register data arrays been allocated? */
5204 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5205 { /* No. So allocate them with malloc. We need one
5206 extra element beyond `num_regs' for the `-1' marker
5208 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5209 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5210 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5211 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5216 bufp
->regs_allocated
= REGS_REALLOCATE
;
5218 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5219 { /* Yes. If we need more elements than were already
5220 allocated, reallocate them. If we need fewer, just
5222 if (regs
->num_regs
< num_regs
+ 1)
5224 regs
->num_regs
= num_regs
+ 1;
5225 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5226 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5227 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5236 /* These braces fend off a "empty body in an else-statement"
5237 warning under GCC when assert expands to nothing. */
5238 assert (bufp
->regs_allocated
== REGS_FIXED
);
5241 /* Convert the pointer data in `regstart' and `regend' to
5242 indices. Register zero has to be set differently,
5243 since we haven't kept track of any info for it. */
5244 if (regs
->num_regs
> 0)
5246 regs
->start
[0] = pos
;
5247 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5250 /* Go through the first `min (num_regs, regs->num_regs)'
5251 registers, since that is all we initialized. */
5252 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5254 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5255 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5258 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5259 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5263 /* If the regs structure we return has more elements than
5264 were in the pattern, set the extra elements to -1. If
5265 we (re)allocated the registers, this is the case,
5266 because we always allocate enough to have at least one
5268 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5269 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5270 } /* regs && !bufp->no_sub */
5272 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5273 nfailure_points_pushed
, nfailure_points_popped
,
5274 nfailure_points_pushed
- nfailure_points_popped
);
5275 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5277 dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5279 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5285 /* Otherwise match next pattern command. */
5288 /* Ignore these. Used to ignore the n of succeed_n's which
5289 currently have n == 0. */
5291 DEBUG_PRINT ("EXECUTING no_op.\n");
5295 DEBUG_PRINT ("EXECUTING succeed.\n");
5298 /* Match the next n pattern characters exactly. The following
5299 byte in the pattern defines n, and the n bytes after that
5300 are the characters to match. */
5303 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5305 /* Remember the start point to rollback upon failure. */
5309 /* This is written out as an if-else so we don't waste time
5310 testing `translate' inside the loop. */
5311 if (RE_TRANSLATE_P (translate
))
5315 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5335 /* The cost of testing `translate' is comparatively small. */
5336 if (target_multibyte
)
5339 int pat_charlen
, buf_charlen
;
5344 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5347 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5350 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5352 if (TRANSLATE (buf_ch
) != pat_ch
)
5360 mcnt
-= pat_charlen
;
5372 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5373 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5380 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5381 if (! CHAR_BYTE8_P (buf_ch
))
5383 buf_ch
= TRANSLATE (buf_ch
);
5384 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5390 if (buf_ch
!= pat_ch
)
5403 /* Match any character except possibly a newline or a null. */
5409 DEBUG_PRINT ("EXECUTING anychar.\n");
5412 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5414 buf_ch
= TRANSLATE (buf_ch
);
5416 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5418 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5419 && buf_ch
== '\000'))
5422 DEBUG_PRINT (" Matched `%d'.\n", *d
);
5431 register unsigned int c
;
5432 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5435 /* Start of actual range_table, or end of bitmap if there is no
5437 re_char
*range_table
IF_LINT (= NULL
);
5439 /* Nonzero if there is a range table. */
5440 int range_table_exists
;
5442 /* Number of ranges of range table. This is not included
5443 in the initial byte-length of the command. */
5446 /* Whether matching against a unibyte character. */
5447 boolean unibyte_char
= false;
5449 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5451 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5453 if (range_table_exists
)
5455 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5456 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5460 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5461 if (target_multibyte
)
5466 c1
= RE_CHAR_TO_UNIBYTE (c
);
5469 unibyte_char
= true;
5475 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5477 if (! CHAR_BYTE8_P (c1
))
5479 c1
= TRANSLATE (c1
);
5480 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5483 unibyte_char
= true;
5488 unibyte_char
= true;
5491 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5492 { /* Lookup bitmap. */
5493 /* Cast to `unsigned' instead of `unsigned char' in
5494 case the bit list is a full 32 bytes long. */
5495 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5496 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5500 else if (range_table_exists
)
5502 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5504 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5505 | (class_bits
& BIT_MULTIBYTE
)
5506 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5507 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5508 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5509 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5512 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5516 if (range_table_exists
)
5517 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5519 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5521 if (!not) goto fail
;
5528 /* The beginning of a group is represented by start_memory.
5529 The argument is the register number. The text
5530 matched within the group is recorded (in the internal
5531 registers data structure) under the register number. */
5533 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5535 /* In case we need to undo this operation (via backtracking). */
5536 PUSH_FAILURE_REG (*p
);
5539 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5540 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5542 /* Move past the register number and inner group count. */
5547 /* The stop_memory opcode represents the end of a group. Its
5548 argument is the same as start_memory's: the register number. */
5550 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5552 assert (!REG_UNSET (regstart
[*p
]));
5553 /* Strictly speaking, there should be code such as:
5555 assert (REG_UNSET (regend[*p]));
5556 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5558 But the only info to be pushed is regend[*p] and it is known to
5559 be UNSET, so there really isn't anything to push.
5560 Not pushing anything, on the other hand deprives us from the
5561 guarantee that regend[*p] is UNSET since undoing this operation
5562 will not reset its value properly. This is not important since
5563 the value will only be read on the next start_memory or at
5564 the very end and both events can only happen if this stop_memory
5568 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5570 /* Move past the register number and the inner group count. */
5575 /* \<digit> has been turned into a `duplicate' command which is
5576 followed by the numeric value of <digit> as the register number. */
5579 register re_char
*d2
, *dend2
;
5580 int regno
= *p
++; /* Get which register to match against. */
5581 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5583 /* Can't back reference a group which we've never matched. */
5584 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5587 /* Where in input to try to start matching. */
5588 d2
= regstart
[regno
];
5590 /* Remember the start point to rollback upon failure. */
5593 /* Where to stop matching; if both the place to start and
5594 the place to stop matching are in the same string, then
5595 set to the place to stop, otherwise, for now have to use
5596 the end of the first string. */
5598 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5599 == FIRST_STRING_P (regend
[regno
]))
5600 ? regend
[regno
] : end_match_1
);
5605 /* If necessary, advance to next segment in register
5609 if (dend2
== end_match_2
) break;
5610 if (dend2
== regend
[regno
]) break;
5612 /* End of string1 => advance to string2. */
5614 dend2
= regend
[regno
];
5616 /* At end of register contents => success */
5617 if (d2
== dend2
) break;
5619 /* If necessary, advance to next segment in data. */
5622 /* How many characters left in this segment to match. */
5625 /* Want how many consecutive characters we can match in
5626 one shot, so, if necessary, adjust the count. */
5627 if (dcnt
> dend2
- d2
)
5630 /* Compare that many; failure if mismatch, else move
5632 if (RE_TRANSLATE_P (translate
)
5633 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5634 : memcmp (d
, d2
, dcnt
))
5639 d
+= dcnt
, d2
+= dcnt
;
5645 /* begline matches the empty string at the beginning of the string
5646 (unless `not_bol' is set in `bufp'), and after newlines. */
5648 DEBUG_PRINT ("EXECUTING begline.\n");
5650 if (AT_STRINGS_BEG (d
))
5652 if (!bufp
->not_bol
) break;
5657 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5661 /* In all other cases, we fail. */
5665 /* endline is the dual of begline. */
5667 DEBUG_PRINT ("EXECUTING endline.\n");
5669 if (AT_STRINGS_END (d
))
5671 if (!bufp
->not_eol
) break;
5675 PREFETCH_NOLIMIT ();
5682 /* Match at the very beginning of the data. */
5684 DEBUG_PRINT ("EXECUTING begbuf.\n");
5685 if (AT_STRINGS_BEG (d
))
5690 /* Match at the very end of the data. */
5692 DEBUG_PRINT ("EXECUTING endbuf.\n");
5693 if (AT_STRINGS_END (d
))
5698 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5699 pushes NULL as the value for the string on the stack. Then
5700 `POP_FAILURE_POINT' will keep the current value for the
5701 string, instead of restoring it. To see why, consider
5702 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5703 then the . fails against the \n. But the next thing we want
5704 to do is match the \n against the \n; if we restored the
5705 string value, we would be back at the foo.
5707 Because this is used only in specific cases, we don't need to
5708 check all the things that `on_failure_jump' does, to make
5709 sure the right things get saved on the stack. Hence we don't
5710 share its code. The only reason to push anything on the
5711 stack at all is that otherwise we would have to change
5712 `anychar's code to do something besides goto fail in this
5713 case; that seems worse than this. */
5714 case on_failure_keep_string_jump
:
5715 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5716 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5719 PUSH_FAILURE_POINT (p
- 3, NULL
);
5722 /* A nasty loop is introduced by the non-greedy *? and +?.
5723 With such loops, the stack only ever contains one failure point
5724 at a time, so that a plain on_failure_jump_loop kind of
5725 cycle detection cannot work. Worse yet, such a detection
5726 can not only fail to detect a cycle, but it can also wrongly
5727 detect a cycle (between different instantiations of the same
5729 So the method used for those nasty loops is a little different:
5730 We use a special cycle-detection-stack-frame which is pushed
5731 when the on_failure_jump_nastyloop failure-point is *popped*.
5732 This special frame thus marks the beginning of one iteration
5733 through the loop and we can hence easily check right here
5734 whether something matched between the beginning and the end of
5736 case on_failure_jump_nastyloop
:
5737 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5738 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5741 assert ((re_opcode_t
)p
[-4] == no_op
);
5744 CHECK_INFINITE_LOOP (p
- 4, d
);
5746 /* If there's a cycle, just continue without pushing
5747 this failure point. The failure point is the "try again"
5748 option, which shouldn't be tried.
5749 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5750 PUSH_FAILURE_POINT (p
- 3, d
);
5754 /* Simple loop detecting on_failure_jump: just check on the
5755 failure stack if the same spot was already hit earlier. */
5756 case on_failure_jump_loop
:
5758 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5759 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5763 CHECK_INFINITE_LOOP (p
- 3, d
);
5765 /* If there's a cycle, get out of the loop, as if the matching
5766 had failed. We used to just `goto fail' here, but that was
5767 aborting the search a bit too early: we want to keep the
5768 empty-loop-match and keep matching after the loop.
5769 We want (x?)*y\1z to match both xxyz and xxyxz. */
5772 PUSH_FAILURE_POINT (p
- 3, d
);
5777 /* Uses of on_failure_jump:
5779 Each alternative starts with an on_failure_jump that points
5780 to the beginning of the next alternative. Each alternative
5781 except the last ends with a jump that in effect jumps past
5782 the rest of the alternatives. (They really jump to the
5783 ending jump of the following alternative, because tensioning
5784 these jumps is a hassle.)
5786 Repeats start with an on_failure_jump that points past both
5787 the repetition text and either the following jump or
5788 pop_failure_jump back to this on_failure_jump. */
5789 case on_failure_jump
:
5790 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5791 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5794 PUSH_FAILURE_POINT (p
-3, d
);
5797 /* This operation is used for greedy *.
5798 Compare the beginning of the repeat with what in the
5799 pattern follows its end. If we can establish that there
5800 is nothing that they would both match, i.e., that we
5801 would have to backtrack because of (as in, e.g., `a*a')
5802 then we can use a non-backtracking loop based on
5803 on_failure_keep_string_jump instead of on_failure_jump. */
5804 case on_failure_jump_smart
:
5805 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5806 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5809 re_char
*p1
= p
; /* Next operation. */
5810 /* Here, we discard `const', making re_match non-reentrant. */
5811 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5812 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5814 p
-= 3; /* Reset so that we will re-execute the
5815 instruction once it's been changed. */
5817 EXTRACT_NUMBER (mcnt
, p2
- 2);
5819 /* Ensure this is a indeed the trivial kind of loop
5820 we are expecting. */
5821 assert (skip_one_char (p1
) == p2
- 3);
5822 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5823 DEBUG_STATEMENT (debug
+= 2);
5824 if (mutually_exclusive_p (bufp
, p1
, p2
))
5826 /* Use a fast `on_failure_keep_string_jump' loop. */
5827 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5828 *p3
= (unsigned char) on_failure_keep_string_jump
;
5829 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5833 /* Default to a safe `on_failure_jump' loop. */
5834 DEBUG_PRINT (" smart default => slow loop.\n");
5835 *p3
= (unsigned char) on_failure_jump
;
5837 DEBUG_STATEMENT (debug
-= 2);
5841 /* Unconditionally jump (without popping any failure points). */
5844 IMMEDIATE_QUIT_CHECK
;
5845 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5846 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5847 p
+= mcnt
; /* Do the jump. */
5848 DEBUG_PRINT ("(to %p).\n", p
);
5852 /* Have to succeed matching what follows at least n times.
5853 After that, handle like `on_failure_jump'. */
5855 /* Signedness doesn't matter since we only compare MCNT to 0. */
5856 EXTRACT_NUMBER (mcnt
, p
+ 2);
5857 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5859 /* Originally, mcnt is how many times we HAVE to succeed. */
5862 /* Here, we discard `const', making re_match non-reentrant. */
5863 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5866 PUSH_NUMBER (p2
, mcnt
);
5869 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5874 /* Signedness doesn't matter since we only compare MCNT to 0. */
5875 EXTRACT_NUMBER (mcnt
, p
+ 2);
5876 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5878 /* Originally, this is how many times we CAN jump. */
5881 /* Here, we discard `const', making re_match non-reentrant. */
5882 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5884 PUSH_NUMBER (p2
, mcnt
);
5885 goto unconditional_jump
;
5887 /* If don't have to jump any more, skip over the rest of command. */
5894 unsigned char *p2
; /* Location of the counter. */
5895 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5897 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5898 /* Here, we discard `const', making re_match non-reentrant. */
5899 p2
= (unsigned char*) p
+ mcnt
;
5900 /* Signedness doesn't matter since we only copy MCNT's bits . */
5901 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5902 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5903 PUSH_NUMBER (p2
, mcnt
);
5910 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5911 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5913 /* We SUCCEED (or FAIL) in one of the following cases: */
5915 /* Case 1: D is at the beginning or the end of string. */
5916 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5920 /* C1 is the character before D, S1 is the syntax of C1, C2
5921 is the character at D, and S2 is the syntax of C2. */
5926 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5927 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5928 UPDATE_SYNTAX_TABLE (charpos
);
5930 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5933 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5935 PREFETCH_NOLIMIT ();
5936 GET_CHAR_AFTER (c2
, d
, dummy
);
5939 if (/* Case 2: Only one of S1 and S2 is Sword. */
5940 ((s1
== Sword
) != (s2
== Sword
))
5941 /* Case 3: Both of S1 and S2 are Sword, and macro
5942 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5943 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5953 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5955 /* We FAIL in one of the following cases: */
5957 /* Case 1: D is at the end of string. */
5958 if (AT_STRINGS_END (d
))
5962 /* C1 is the character before D, S1 is the syntax of C1, C2
5963 is the character at D, and S2 is the syntax of C2. */
5968 ssize_t offset
= PTR_TO_OFFSET (d
);
5969 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5970 UPDATE_SYNTAX_TABLE (charpos
);
5973 GET_CHAR_AFTER (c2
, d
, dummy
);
5976 /* Case 2: S2 is not Sword. */
5980 /* Case 3: D is not at the beginning of string ... */
5981 if (!AT_STRINGS_BEG (d
))
5983 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5985 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5989 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5991 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5998 DEBUG_PRINT ("EXECUTING wordend.\n");
6000 /* We FAIL in one of the following cases: */
6002 /* Case 1: D is at the beginning of string. */
6003 if (AT_STRINGS_BEG (d
))
6007 /* C1 is the character before D, S1 is the syntax of C1, C2
6008 is the character at D, and S2 is the syntax of C2. */
6013 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6014 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6015 UPDATE_SYNTAX_TABLE (charpos
);
6017 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6020 /* Case 2: S1 is not Sword. */
6024 /* Case 3: D is not at the end of string ... */
6025 if (!AT_STRINGS_END (d
))
6027 PREFETCH_NOLIMIT ();
6028 GET_CHAR_AFTER (c2
, d
, dummy
);
6030 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6034 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6036 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6043 DEBUG_PRINT ("EXECUTING symbeg.\n");
6045 /* We FAIL in one of the following cases: */
6047 /* Case 1: D is at the end of string. */
6048 if (AT_STRINGS_END (d
))
6052 /* C1 is the character before D, S1 is the syntax of C1, C2
6053 is the character at D, and S2 is the syntax of C2. */
6057 ssize_t offset
= PTR_TO_OFFSET (d
);
6058 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6059 UPDATE_SYNTAX_TABLE (charpos
);
6062 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6065 /* Case 2: S2 is neither Sword nor Ssymbol. */
6066 if (s2
!= Sword
&& s2
!= Ssymbol
)
6069 /* Case 3: D is not at the beginning of string ... */
6070 if (!AT_STRINGS_BEG (d
))
6072 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6074 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6078 /* ... and S1 is Sword or Ssymbol. */
6079 if (s1
== Sword
|| s1
== Ssymbol
)
6086 DEBUG_PRINT ("EXECUTING symend.\n");
6088 /* We FAIL in one of the following cases: */
6090 /* Case 1: D is at the beginning of string. */
6091 if (AT_STRINGS_BEG (d
))
6095 /* C1 is the character before D, S1 is the syntax of C1, C2
6096 is the character at D, and S2 is the syntax of C2. */
6100 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6101 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6102 UPDATE_SYNTAX_TABLE (charpos
);
6104 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6107 /* Case 2: S1 is neither Ssymbol nor Sword. */
6108 if (s1
!= Sword
&& s1
!= Ssymbol
)
6111 /* Case 3: D is not at the end of string ... */
6112 if (!AT_STRINGS_END (d
))
6114 PREFETCH_NOLIMIT ();
6115 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6117 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6121 /* ... and S2 is Sword or Ssymbol. */
6122 if (s2
== Sword
|| s2
== Ssymbol
)
6131 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6133 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6138 ssize_t offset
= PTR_TO_OFFSET (d
);
6139 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6140 UPDATE_SYNTAX_TABLE (pos1
);
6147 GET_CHAR_AFTER (c
, d
, len
);
6148 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6157 DEBUG_PRINT ("EXECUTING before_dot.\n");
6158 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6163 DEBUG_PRINT ("EXECUTING at_dot.\n");
6164 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6169 DEBUG_PRINT ("EXECUTING after_dot.\n");
6170 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6175 case notcategoryspec
:
6177 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6179 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6180 not ? "not" : "", mcnt
);
6186 GET_CHAR_AFTER (c
, d
, len
);
6187 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6199 continue; /* Successfully executed one pattern command; keep going. */
6202 /* We goto here if a matching operation fails. */
6204 IMMEDIATE_QUIT_CHECK
;
6205 if (!FAIL_STACK_EMPTY ())
6208 /* A restart point is known. Restore to that state. */
6209 DEBUG_PRINT ("\nFAIL:\n");
6210 POP_FAILURE_POINT (str
, pat
);
6213 case on_failure_keep_string_jump
:
6214 assert (str
== NULL
);
6215 goto continue_failure_jump
;
6217 case on_failure_jump_nastyloop
:
6218 assert ((re_opcode_t
)pat
[-2] == no_op
);
6219 PUSH_FAILURE_POINT (pat
- 2, str
);
6222 case on_failure_jump_loop
:
6223 case on_failure_jump
:
6226 continue_failure_jump
:
6227 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6232 /* A special frame used for nastyloops. */
6239 assert (p
>= bufp
->buffer
&& p
<= pend
);
6241 if (d
>= string1
&& d
<= end1
)
6245 break; /* Matching at this starting point really fails. */
6249 goto restore_best_regs
;
6253 return -1; /* Failure to match. */
6256 /* Subroutine definitions for re_match_2. */
6258 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6259 bytes; nonzero otherwise. */
6262 bcmp_translate (const re_char
*s1
, const re_char
*s2
, register ssize_t len
,
6263 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6265 register re_char
*p1
= s1
, *p2
= s2
;
6266 re_char
*p1_end
= s1
+ len
;
6267 re_char
*p2_end
= s2
+ len
;
6269 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6270 different lengths, but relying on a single `len' would break this. -sm */
6271 while (p1
< p1_end
&& p2
< p2_end
)
6273 int p1_charlen
, p2_charlen
;
6274 re_wchar_t p1_ch
, p2_ch
;
6276 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6277 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6279 if (RE_TRANSLATE (translate
, p1_ch
)
6280 != RE_TRANSLATE (translate
, p2_ch
))
6283 p1
+= p1_charlen
, p2
+= p2_charlen
;
6286 if (p1
!= p1_end
|| p2
!= p2_end
)
6292 /* Entry points for GNU code. */
6294 /* re_compile_pattern is the GNU regular expression compiler: it
6295 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6296 Returns 0 if the pattern was valid, otherwise an error string.
6298 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6299 are set in BUFP on entry.
6301 We call regex_compile to do the actual compilation. */
6304 re_compile_pattern (const char *pattern
, size_t length
,
6305 struct re_pattern_buffer
*bufp
)
6309 /* GNU code is written to assume at least RE_NREGS registers will be set
6310 (and at least one extra will be -1). */
6311 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6313 /* And GNU code determines whether or not to get register information
6314 by passing null for the REGS argument to re_match, etc., not by
6318 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6322 return gettext (re_error_msgid
[(int) ret
]);
6324 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6326 /* Entry points compatible with 4.2 BSD regex library. We don't define
6327 them unless specifically requested. */
6329 #if defined _REGEX_RE_COMP || defined _LIBC
6331 /* BSD has one and only one pattern buffer. */
6332 static struct re_pattern_buffer re_comp_buf
;
6336 /* Make these definitions weak in libc, so POSIX programs can redefine
6337 these names if they don't use our functions, and still use
6338 regcomp/regexec below without link errors. */
6341 re_comp (const char *s
)
6347 if (!re_comp_buf
.buffer
)
6348 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6349 return (char *) gettext ("No previous regular expression");
6353 if (!re_comp_buf
.buffer
)
6355 re_comp_buf
.buffer
= malloc (200);
6356 if (re_comp_buf
.buffer
== NULL
)
6357 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6358 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6359 re_comp_buf
.allocated
= 200;
6361 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6362 if (re_comp_buf
.fastmap
== NULL
)
6363 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6364 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6367 /* Since `re_exec' always passes NULL for the `regs' argument, we
6368 don't need to initialize the pattern buffer fields which affect it. */
6370 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6375 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6376 return (char *) gettext (re_error_msgid
[(int) ret
]);
6384 re_exec (const char *s
)
6386 const size_t len
= strlen (s
);
6387 return (re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0)
6390 #endif /* _REGEX_RE_COMP */
6392 /* POSIX.2 functions. Don't define these for Emacs. */
6396 /* regcomp takes a regular expression as a string and compiles it.
6398 PREG is a regex_t *. We do not expect any fields to be initialized,
6399 since POSIX says we shouldn't. Thus, we set
6401 `buffer' to the compiled pattern;
6402 `used' to the length of the compiled pattern;
6403 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6404 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6405 RE_SYNTAX_POSIX_BASIC;
6406 `fastmap' to an allocated space for the fastmap;
6407 `fastmap_accurate' to zero;
6408 `re_nsub' to the number of subexpressions in PATTERN.
6410 PATTERN is the address of the pattern string.
6412 CFLAGS is a series of bits which affect compilation.
6414 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6415 use POSIX basic syntax.
6417 If REG_NEWLINE is set, then . and [^...] don't match newline.
6418 Also, regexec will try a match beginning after every newline.
6420 If REG_ICASE is set, then we considers upper- and lowercase
6421 versions of letters to be equivalent when matching.
6423 If REG_NOSUB is set, then when PREG is passed to regexec, that
6424 routine will report only success or failure, and nothing about the
6427 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6428 the return codes and their meanings.) */
6431 regcomp (regex_t
*__restrict preg
, const char *__restrict pattern
,
6436 = (cflags
& REG_EXTENDED
) ?
6437 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6439 /* regex_compile will allocate the space for the compiled pattern. */
6441 preg
->allocated
= 0;
6444 /* Try to allocate space for the fastmap. */
6445 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6447 if (cflags
& REG_ICASE
)
6451 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6452 if (preg
->translate
== NULL
)
6453 return (int) REG_ESPACE
;
6455 /* Map uppercase characters to corresponding lowercase ones. */
6456 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6457 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6460 preg
->translate
= NULL
;
6462 /* If REG_NEWLINE is set, newlines are treated differently. */
6463 if (cflags
& REG_NEWLINE
)
6464 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6465 syntax
&= ~RE_DOT_NEWLINE
;
6466 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6469 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6471 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6473 /* POSIX says a null character in the pattern terminates it, so we
6474 can use strlen here in compiling the pattern. */
6475 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6477 /* POSIX doesn't distinguish between an unmatched open-group and an
6478 unmatched close-group: both are REG_EPAREN. */
6479 if (ret
== REG_ERPAREN
)
6482 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6483 { /* Compute the fastmap now, since regexec cannot modify the pattern
6485 re_compile_fastmap (preg
);
6486 if (preg
->can_be_null
)
6487 { /* The fastmap can't be used anyway. */
6488 free (preg
->fastmap
);
6489 preg
->fastmap
= NULL
;
6494 WEAK_ALIAS (__regcomp
, regcomp
)
6497 /* regexec searches for a given pattern, specified by PREG, in the
6500 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6501 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6502 least NMATCH elements, and we set them to the offsets of the
6503 corresponding matched substrings.
6505 EFLAGS specifies `execution flags' which affect matching: if
6506 REG_NOTBOL is set, then ^ does not match at the beginning of the
6507 string; if REG_NOTEOL is set, then $ does not match at the end.
6509 We return 0 if we find a match and REG_NOMATCH if not. */
6512 regexec (const regex_t
*__restrict preg
, const char *__restrict string
,
6513 size_t nmatch
, regmatch_t pmatch
[__restrict_arr
], int eflags
)
6516 struct re_registers regs
;
6517 regex_t private_preg
;
6518 size_t len
= strlen (string
);
6519 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6521 private_preg
= *preg
;
6523 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6524 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6526 /* The user has told us exactly how many registers to return
6527 information about, via `nmatch'. We have to pass that on to the
6528 matching routines. */
6529 private_preg
.regs_allocated
= REGS_FIXED
;
6533 regs
.num_regs
= nmatch
;
6534 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6535 if (regs
.start
== NULL
)
6537 regs
.end
= regs
.start
+ nmatch
;
6540 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6541 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6542 was a little bit longer but still only matching the real part.
6543 This works because the `endline' will check for a '\n' and will find a
6544 '\0', correctly deciding that this is not the end of a line.
6545 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6546 a convenient '\0' there. For all we know, the string could be preceded
6547 by '\n' which would throw things off. */
6549 /* Perform the searching operation. */
6550 ret
= re_search (&private_preg
, string
, len
,
6551 /* start: */ 0, /* range: */ len
,
6552 want_reg_info
? ®s
: (struct re_registers
*) 0);
6554 /* Copy the register information to the POSIX structure. */
6561 for (r
= 0; r
< nmatch
; r
++)
6563 pmatch
[r
].rm_so
= regs
.start
[r
];
6564 pmatch
[r
].rm_eo
= regs
.end
[r
];
6568 /* If we needed the temporary register info, free the space now. */
6572 /* We want zero return to mean success, unlike `re_search'. */
6573 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6575 WEAK_ALIAS (__regexec
, regexec
)
6578 /* Returns a message corresponding to an error code, ERR_CODE, returned
6579 from either regcomp or regexec. We don't use PREG here.
6581 ERR_CODE was previously called ERRCODE, but that name causes an
6582 error with msvc8 compiler. */
6585 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6591 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6592 /* Only error codes returned by the rest of the code should be passed
6593 to this routine. If we are given anything else, or if other regex
6594 code generates an invalid error code, then the program has a bug.
6595 Dump core so we can fix it. */
6598 msg
= gettext (re_error_msgid
[err_code
]);
6600 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6602 if (errbuf_size
!= 0)
6604 if (msg_size
> errbuf_size
)
6606 memcpy (errbuf
, msg
, errbuf_size
- 1);
6607 errbuf
[errbuf_size
- 1] = 0;
6610 strcpy (errbuf
, msg
);
6615 WEAK_ALIAS (__regerror
, regerror
)
6618 /* Free dynamically allocated space used by PREG. */
6621 regfree (regex_t
*preg
)
6623 free (preg
->buffer
);
6624 preg
->buffer
= NULL
;
6626 preg
->allocated
= 0;
6629 free (preg
->fastmap
);
6630 preg
->fastmap
= NULL
;
6631 preg
->fastmap_accurate
= 0;
6633 free (preg
->translate
);
6634 preg
->translate
= NULL
;
6636 WEAK_ALIAS (__regfree
, regfree
)
6638 #endif /* not emacs */