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) 1993-2012 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, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
36 /* Ignore some GCC warnings for now. This section should go away
37 once the Emacs and Gnulib regex code is merged. */
38 #if (__GNUC__ == 4 && 5 <= __GNUC_MINOR__) || 4 < __GNUC__
39 # pragma GCC diagnostic ignored "-Wstrict-overflow"
41 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
42 # pragma GCC diagnostic ignored "-Wunused-function"
43 # pragma GCC diagnostic ignored "-Wunused-macros"
44 # pragma GCC diagnostic ignored "-Wunused-result"
45 # pragma GCC diagnostic ignored "-Wunused-variable"
54 /* We need this for `regex.h', and perhaps for the Emacs include files. */
55 # include <sys/types.h>
58 /* Whether to use ISO C Amendment 1 wide char functions.
59 Those should not be used for Emacs since it uses its own. */
61 #define WIDE_CHAR_SUPPORT 1
63 #define WIDE_CHAR_SUPPORT \
64 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
67 /* For platform which support the ISO C amendment 1 functionality we
68 support user defined character classes. */
70 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
76 /* We have to keep the namespace clean. */
77 # define regfree(preg) __regfree (preg)
78 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
79 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
80 # define regerror(err_code, preg, errbuf, errbuf_size) \
81 __regerror (err_code, preg, errbuf, errbuf_size)
82 # define re_set_registers(bu, re, nu, st, en) \
83 __re_set_registers (bu, re, nu, st, en)
84 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
85 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
86 # define re_match(bufp, string, size, pos, regs) \
87 __re_match (bufp, string, size, pos, regs)
88 # define re_search(bufp, string, size, startpos, range, regs) \
89 __re_search (bufp, string, size, startpos, range, regs)
90 # define re_compile_pattern(pattern, length, bufp) \
91 __re_compile_pattern (pattern, length, bufp)
92 # define re_set_syntax(syntax) __re_set_syntax (syntax)
93 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
94 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
95 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
97 /* Make sure we call libc's function even if the user overrides them. */
98 # define btowc __btowc
99 # define iswctype __iswctype
100 # define wctype __wctype
102 # define WEAK_ALIAS(a,b) weak_alias (a, b)
104 /* We are also using some library internals. */
105 # include <locale/localeinfo.h>
106 # include <locale/elem-hash.h>
107 # include <langinfo.h>
109 # define WEAK_ALIAS(a,b)
112 /* This is for other GNU distributions with internationalized messages. */
113 #if HAVE_LIBINTL_H || defined _LIBC
114 # include <libintl.h>
116 # define gettext(msgid) (msgid)
120 /* This define is so xgettext can find the internationalizable
122 # define gettext_noop(String) String
125 /* The `emacs' switch turns on certain matching commands
126 that make sense only in Emacs. */
131 # include "character.h"
134 /* Make syntax table lookup grant data in gl_state. */
135 # define SYNTAX_ENTRY_VIA_PROPERTY
138 # include "category.h"
143 # define malloc xmalloc
147 # define realloc xrealloc
153 /* Converts the pointer to the char to BEG-based offset from the start. */
154 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
155 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
157 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
158 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
159 # define RE_STRING_CHAR(p, multibyte) \
160 (multibyte ? (STRING_CHAR (p)) : (*(p)))
161 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
162 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
164 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
166 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
168 /* Set C a (possibly converted to multibyte) character before P. P
169 points into a string which is the virtual concatenation of STR1
170 (which ends at END1) or STR2 (which ends at END2). */
171 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
173 if (target_multibyte) \
175 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
176 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
177 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
178 c = STRING_CHAR (dtemp); \
182 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
183 (c) = RE_CHAR_TO_MULTIBYTE (c); \
187 /* Set C a (possibly converted to multibyte) character at P, and set
188 LEN to the byte length of that character. */
189 # define GET_CHAR_AFTER(c, p, len) \
191 if (target_multibyte) \
192 (c) = STRING_CHAR_AND_LENGTH (p, len); \
197 (c) = RE_CHAR_TO_MULTIBYTE (c); \
201 #else /* not emacs */
203 /* If we are not linking with Emacs proper,
204 we can't use the relocating allocator
205 even if config.h says that we can. */
210 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
213 xmalloc (size_t size
)
215 void *val
= malloc (size
);
218 write (2, "virtual memory exhausted\n", 25);
225 xrealloc (void *block
, size_t size
)
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
233 val
= realloc (block
, size
);
236 write (2, "virtual memory exhausted\n", 25);
245 # define malloc xmalloc
249 # define realloc xrealloc
253 /* Define the syntax stuff for \<, \>, etc. */
255 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
256 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
258 # define SWITCH_ENUM_CAST(x) (x)
260 /* Dummy macros for non-Emacs environments. */
261 # define CHAR_CHARSET(c) 0
262 # define CHARSET_LEADING_CODE_BASE(c) 0
263 # define MAX_MULTIBYTE_LENGTH 1
264 # define RE_MULTIBYTE_P(x) 0
265 # define RE_TARGET_MULTIBYTE_P(x) 0
266 # define WORD_BOUNDARY_P(c1, c2) (0)
267 # define CHAR_HEAD_P(p) (1)
268 # define SINGLE_BYTE_CHAR_P(c) (1)
269 # define SAME_CHARSET_P(c1, c2) (1)
270 # define BYTES_BY_CHAR_HEAD(p) (1)
271 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
272 # define STRING_CHAR(p) (*(p))
273 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
274 # define CHAR_STRING(c, s) (*(s) = (c), 1)
275 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
276 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
277 # define RE_CHAR_TO_MULTIBYTE(c) (c)
278 # define RE_CHAR_TO_UNIBYTE(c) (c)
279 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
280 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
281 # define GET_CHAR_AFTER(c, p, len) \
283 # define MAKE_CHAR(charset, c1, c2) (c1)
284 # define BYTE8_TO_CHAR(c) (c)
285 # define CHAR_BYTE8_P(c) (0)
286 # define CHAR_LEADING_CODE(c) (c)
288 #endif /* not emacs */
291 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
292 # define RE_TRANSLATE_P(TBL) (TBL)
295 /* Get the interface, including the syntax bits. */
298 /* isalpha etc. are used for the character classes. */
303 /* 1 if C is an ASCII character. */
304 # define IS_REAL_ASCII(c) ((c) < 0200)
306 /* 1 if C is a unibyte character. */
307 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
309 /* The Emacs definitions should not be directly affected by locales. */
311 /* In Emacs, these are only used for single-byte characters. */
312 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
313 # define ISCNTRL(c) ((c) < ' ')
314 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
315 || ((c) >= 'a' && (c) <= 'f') \
316 || ((c) >= 'A' && (c) <= 'F'))
318 /* This is only used for single-byte characters. */
319 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
321 /* The rest must handle multibyte characters. */
323 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
324 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
327 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
328 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
331 # define ISALNUM(c) (IS_REAL_ASCII (c) \
332 ? (((c) >= 'a' && (c) <= 'z') \
333 || ((c) >= 'A' && (c) <= 'Z') \
334 || ((c) >= '0' && (c) <= '9')) \
335 : SYNTAX (c) == Sword)
337 # define ISALPHA(c) (IS_REAL_ASCII (c) \
338 ? (((c) >= 'a' && (c) <= 'z') \
339 || ((c) >= 'A' && (c) <= 'Z')) \
340 : SYNTAX (c) == Sword)
342 # define ISLOWER(c) lowercasep (c)
344 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
345 ? ((c) > ' ' && (c) < 0177 \
346 && !(((c) >= 'a' && (c) <= 'z') \
347 || ((c) >= 'A' && (c) <= 'Z') \
348 || ((c) >= '0' && (c) <= '9'))) \
349 : SYNTAX (c) != Sword)
351 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
353 # define ISUPPER(c) uppercasep (c)
355 # define ISWORD(c) (SYNTAX (c) == Sword)
357 #else /* not emacs */
359 /* 1 if C is an ASCII character. */
360 # define IS_REAL_ASCII(c) ((c) < 0200)
362 /* This distinction is not meaningful, except in Emacs. */
363 # define ISUNIBYTE(c) 1
366 # define ISBLANK(c) isblank (c)
368 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
371 # define ISGRAPH(c) isgraph (c)
373 # define ISGRAPH(c) (isprint (c) && !isspace (c))
376 /* Solaris defines ISPRINT so we must undefine it first. */
378 # define ISPRINT(c) isprint (c)
379 # define ISDIGIT(c) isdigit (c)
380 # define ISALNUM(c) isalnum (c)
381 # define ISALPHA(c) isalpha (c)
382 # define ISCNTRL(c) iscntrl (c)
383 # define ISLOWER(c) islower (c)
384 # define ISPUNCT(c) ispunct (c)
385 # define ISSPACE(c) isspace (c)
386 # define ISUPPER(c) isupper (c)
387 # define ISXDIGIT(c) isxdigit (c)
389 # define ISWORD(c) ISALPHA (c)
392 # define TOLOWER(c) _tolower (c)
394 # define TOLOWER(c) tolower (c)
397 /* How many characters in the character set. */
398 # define CHAR_SET_SIZE 256
402 extern char *re_syntax_table
;
404 # else /* not SYNTAX_TABLE */
406 static char re_syntax_table
[CHAR_SET_SIZE
];
409 init_syntax_once (void)
417 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
419 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
421 re_syntax_table
[c
] = Sword
;
423 re_syntax_table
['_'] = Ssymbol
;
428 # endif /* not SYNTAX_TABLE */
430 # define SYNTAX(c) re_syntax_table[(c)]
432 #endif /* not emacs */
434 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
436 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
437 use `alloca' instead of `malloc'. This is because using malloc in
438 re_search* or re_match* could cause memory leaks when C-g is used in
439 Emacs; also, malloc is slower and causes storage fragmentation. On
440 the other hand, malloc is more portable, and easier to debug.
442 Because we sometimes use alloca, some routines have to be macros,
443 not functions -- `alloca'-allocated space disappears at the end of the
444 function it is called in. */
448 # define REGEX_ALLOCATE malloc
449 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
450 # define REGEX_FREE free
452 #else /* not REGEX_MALLOC */
454 /* Emacs already defines alloca, sometimes. */
457 /* Make alloca work the best possible way. */
459 # define alloca __builtin_alloca
460 # else /* not __GNUC__ */
461 # ifdef HAVE_ALLOCA_H
463 # endif /* HAVE_ALLOCA_H */
464 # endif /* not __GNUC__ */
466 # endif /* not alloca */
468 # define REGEX_ALLOCATE alloca
470 /* Assumes a `char *destination' variable. */
471 # define REGEX_REALLOCATE(source, osize, nsize) \
472 (destination = (char *) alloca (nsize), \
473 memcpy (destination, source, osize))
475 /* No need to do anything to free, after alloca. */
476 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
478 #endif /* not REGEX_MALLOC */
480 /* Define how to allocate the failure stack. */
482 #if defined REL_ALLOC && defined REGEX_MALLOC
484 # define REGEX_ALLOCATE_STACK(size) \
485 r_alloc (&failure_stack_ptr, (size))
486 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
487 r_re_alloc (&failure_stack_ptr, (nsize))
488 # define REGEX_FREE_STACK(ptr) \
489 r_alloc_free (&failure_stack_ptr)
491 #else /* not using relocating allocator */
495 # define REGEX_ALLOCATE_STACK malloc
496 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
497 # define REGEX_FREE_STACK free
499 # else /* not REGEX_MALLOC */
501 # define REGEX_ALLOCATE_STACK alloca
503 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
504 REGEX_REALLOCATE (source, osize, nsize)
505 /* No need to explicitly free anything. */
506 # define REGEX_FREE_STACK(arg) ((void)0)
508 # endif /* not REGEX_MALLOC */
509 #endif /* not using relocating allocator */
512 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
513 `string1' or just past its end. This works if PTR is NULL, which is
515 #define FIRST_STRING_P(ptr) \
516 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
518 /* (Re)Allocate N items of type T using malloc, or fail. */
519 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
520 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
521 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
523 #define BYTEWIDTH 8 /* In bits. */
525 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
529 #define MAX(a, b) ((a) > (b) ? (a) : (b))
530 #define MIN(a, b) ((a) < (b) ? (a) : (b))
532 /* Type of source-pattern and string chars. */
534 typedef unsigned char re_char
;
536 typedef const unsigned char re_char
;
539 typedef char boolean
;
543 static regoff_t
re_match_2_internal (struct re_pattern_buffer
*bufp
,
544 re_char
*string1
, size_t size1
,
545 re_char
*string2
, size_t size2
,
547 struct re_registers
*regs
,
550 /* These are the command codes that appear in compiled regular
551 expressions. Some opcodes are followed by argument bytes. A
552 command code can specify any interpretation whatsoever for its
553 arguments. Zero bytes may appear in the compiled regular expression. */
559 /* Succeed right away--no more backtracking. */
562 /* Followed by one byte giving n, then by n literal bytes. */
565 /* Matches any (more or less) character. */
568 /* Matches any one char belonging to specified set. First
569 following byte is number of bitmap bytes. Then come bytes
570 for a bitmap saying which chars are in. Bits in each byte
571 are ordered low-bit-first. A character is in the set if its
572 bit is 1. A character too large to have a bit in the map is
573 automatically not in the set.
575 If the length byte has the 0x80 bit set, then that stuff
576 is followed by a range table:
577 2 bytes of flags for character sets (low 8 bits, high 8 bits)
578 See RANGE_TABLE_WORK_BITS below.
579 2 bytes, the number of pairs that follow (upto 32767)
580 pairs, each 2 multibyte characters,
581 each multibyte character represented as 3 bytes. */
584 /* Same parameters as charset, but match any character that is
585 not one of those specified. */
588 /* Start remembering the text that is matched, for storing in a
589 register. Followed by one byte with the register number, in
590 the range 0 to one less than the pattern buffer's re_nsub
594 /* Stop remembering the text that is matched and store it in a
595 memory register. Followed by one byte with the register
596 number, in the range 0 to one less than `re_nsub' in the
600 /* Match a duplicate of something remembered. Followed by one
601 byte containing the register number. */
604 /* Fail unless at beginning of line. */
607 /* Fail unless at end of line. */
610 /* Succeeds if at beginning of buffer (if emacs) or at beginning
611 of string to be matched (if not). */
614 /* Analogously, for end of buffer/string. */
617 /* Followed by two byte relative address to which to jump. */
620 /* Followed by two-byte relative address of place to resume at
621 in case of failure. */
624 /* Like on_failure_jump, but pushes a placeholder instead of the
625 current string position when executed. */
626 on_failure_keep_string_jump
,
628 /* Just like `on_failure_jump', except that it checks that we
629 don't get stuck in an infinite loop (matching an empty string
631 on_failure_jump_loop
,
633 /* Just like `on_failure_jump_loop', except that it checks for
634 a different kind of loop (the kind that shows up with non-greedy
635 operators). This operation has to be immediately preceded
637 on_failure_jump_nastyloop
,
639 /* A smart `on_failure_jump' used for greedy * and + operators.
640 It analyzes the loop before which it is put and if the
641 loop does not require backtracking, it changes itself to
642 `on_failure_keep_string_jump' and short-circuits the loop,
643 else it just defaults to changing itself into `on_failure_jump'.
644 It assumes that it is pointing to just past a `jump'. */
645 on_failure_jump_smart
,
647 /* Followed by two-byte relative address and two-byte number n.
648 After matching N times, jump to the address upon failure.
649 Does not work if N starts at 0: use on_failure_jump_loop
653 /* Followed by two-byte relative address, and two-byte number n.
654 Jump to the address N times, then fail. */
657 /* Set the following two-byte relative address to the
658 subsequent two-byte number. The address *includes* the two
662 wordbeg
, /* Succeeds if at word beginning. */
663 wordend
, /* Succeeds if at word end. */
665 wordbound
, /* Succeeds if at a word boundary. */
666 notwordbound
, /* Succeeds if not at a word boundary. */
668 symbeg
, /* Succeeds if at symbol beginning. */
669 symend
, /* Succeeds if at symbol end. */
671 /* Matches any character whose syntax is specified. Followed by
672 a byte which contains a syntax code, e.g., Sword. */
675 /* Matches any character whose syntax is not that specified. */
679 ,before_dot
, /* Succeeds if before point. */
680 at_dot
, /* Succeeds if at point. */
681 after_dot
, /* Succeeds if after point. */
683 /* Matches any character whose category-set contains the specified
684 category. The operator is followed by a byte which contains a
685 category code (mnemonic ASCII character). */
688 /* Matches any character whose category-set does not contain the
689 specified category. The operator is followed by a byte which
690 contains the category code (mnemonic ASCII character). */
695 /* Common operations on the compiled pattern. */
697 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
699 #define STORE_NUMBER(destination, number) \
701 (destination)[0] = (number) & 0377; \
702 (destination)[1] = (number) >> 8; \
705 /* Same as STORE_NUMBER, except increment DESTINATION to
706 the byte after where the number is stored. Therefore, DESTINATION
707 must be an lvalue. */
709 #define STORE_NUMBER_AND_INCR(destination, number) \
711 STORE_NUMBER (destination, number); \
712 (destination) += 2; \
715 /* Put into DESTINATION a number stored in two contiguous bytes starting
718 #define EXTRACT_NUMBER(destination, source) \
720 (destination) = *(source) & 0377; \
721 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
726 extract_number (int *dest
, re_char
*source
)
728 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
729 *dest
= *source
& 0377;
733 # ifndef EXTRACT_MACROS /* To debug the macros. */
734 # undef EXTRACT_NUMBER
735 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
736 # endif /* not EXTRACT_MACROS */
740 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
741 SOURCE must be an lvalue. */
743 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
745 EXTRACT_NUMBER (destination, source); \
751 extract_number_and_incr (int *destination
, re_char
**source
)
753 extract_number (destination
, *source
);
757 # ifndef EXTRACT_MACROS
758 # undef EXTRACT_NUMBER_AND_INCR
759 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
760 extract_number_and_incr (&dest, &src)
761 # endif /* not EXTRACT_MACROS */
765 /* Store a multibyte character in three contiguous bytes starting
766 DESTINATION, and increment DESTINATION to the byte after where the
767 character is stored. Therefore, DESTINATION must be an lvalue. */
769 #define STORE_CHARACTER_AND_INCR(destination, character) \
771 (destination)[0] = (character) & 0377; \
772 (destination)[1] = ((character) >> 8) & 0377; \
773 (destination)[2] = (character) >> 16; \
774 (destination) += 3; \
777 /* Put into DESTINATION a character stored in three contiguous bytes
778 starting at SOURCE. */
780 #define EXTRACT_CHARACTER(destination, source) \
782 (destination) = ((source)[0] \
783 | ((source)[1] << 8) \
784 | ((source)[2] << 16)); \
788 /* Macros for charset. */
790 /* Size of bitmap of charset P in bytes. P is a start of charset,
791 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
792 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
794 /* Nonzero if charset P has range table. */
795 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
797 /* Return the address of range table of charset P. But not the start
798 of table itself, but the before where the number of ranges is
799 stored. `2 +' means to skip re_opcode_t and size of bitmap,
800 and the 2 bytes of flags at the start of the range table. */
801 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
803 /* Extract the bit flags that start a range table. */
804 #define CHARSET_RANGE_TABLE_BITS(p) \
805 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
806 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
808 /* Return the address of end of RANGE_TABLE. COUNT is number of
809 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
810 is start of range and end of range. `* 3' is size of each start
812 #define CHARSET_RANGE_TABLE_END(range_table, count) \
813 ((range_table) + (count) * 2 * 3)
815 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
816 COUNT is number of ranges in RANGE_TABLE. */
817 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
820 re_wchar_t range_start, range_end; \
822 re_char *range_table_end \
823 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
825 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
827 EXTRACT_CHARACTER (range_start, rtp); \
828 EXTRACT_CHARACTER (range_end, rtp + 3); \
830 if (range_start <= (c) && (c) <= range_end) \
839 /* Test if C is in range table of CHARSET. The flag NOT is negated if
840 C is listed in it. */
841 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
844 /* Number of ranges in range table. */ \
846 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
848 EXTRACT_NUMBER_AND_INCR (count, range_table); \
849 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
853 /* If DEBUG is defined, Regex prints many voluminous messages about what
854 it is doing (if the variable `debug' is nonzero). If linked with the
855 main program in `iregex.c', you can enter patterns and strings
856 interactively. And if linked with the main program in `main.c' and
857 the other test files, you can run the already-written tests. */
861 /* We use standard I/O for debugging. */
864 /* It is useful to test things that ``must'' be true when debugging. */
867 static int debug
= -100000;
869 # define DEBUG_STATEMENT(e) e
870 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
871 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
872 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
873 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
874 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
875 if (debug > 0) print_partial_compiled_pattern (s, e)
876 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
877 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
880 /* Print the fastmap in human-readable form. */
883 print_fastmap (fastmap
)
886 unsigned was_a_range
= 0;
889 while (i
< (1 << BYTEWIDTH
))
895 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
911 /* Print a compiled pattern string in human-readable form, starting at
912 the START pointer into it and ending just before the pointer END. */
915 print_partial_compiled_pattern (start
, end
)
925 fprintf (stderr
, "(null)\n");
929 /* Loop over pattern commands. */
932 fprintf (stderr
, "%d:\t", p
- start
);
934 switch ((re_opcode_t
) *p
++)
937 fprintf (stderr
, "/no_op");
941 fprintf (stderr
, "/succeed");
946 fprintf (stderr
, "/exactn/%d", mcnt
);
949 fprintf (stderr
, "/%c", *p
++);
955 fprintf (stderr
, "/start_memory/%d", *p
++);
959 fprintf (stderr
, "/stop_memory/%d", *p
++);
963 fprintf (stderr
, "/duplicate/%d", *p
++);
967 fprintf (stderr
, "/anychar");
973 register int c
, last
= -100;
974 register int in_range
= 0;
975 int length
= CHARSET_BITMAP_SIZE (p
- 1);
976 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
978 fprintf (stderr
, "/charset [%s",
979 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
982 fprintf (stderr
, " !extends past end of pattern! ");
984 for (c
= 0; c
< 256; c
++)
986 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
988 /* Are we starting a range? */
989 if (last
+ 1 == c
&& ! in_range
)
991 fprintf (stderr
, "-");
994 /* Have we broken a range? */
995 else if (last
+ 1 != c
&& in_range
)
997 fprintf (stderr
, "%c", last
);
1002 fprintf (stderr
, "%c", c
);
1008 fprintf (stderr
, "%c", last
);
1010 fprintf (stderr
, "]");
1014 if (has_range_table
)
1017 fprintf (stderr
, "has-range-table");
1019 /* ??? Should print the range table; for now, just skip it. */
1020 p
+= 2; /* skip range table bits */
1021 EXTRACT_NUMBER_AND_INCR (count
, p
);
1022 p
= CHARSET_RANGE_TABLE_END (p
, count
);
1028 fprintf (stderr
, "/begline");
1032 fprintf (stderr
, "/endline");
1035 case on_failure_jump
:
1036 extract_number_and_incr (&mcnt
, &p
);
1037 fprintf (stderr
, "/on_failure_jump to %d", p
+ mcnt
- start
);
1040 case on_failure_keep_string_jump
:
1041 extract_number_and_incr (&mcnt
, &p
);
1042 fprintf (stderr
, "/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
1045 case on_failure_jump_nastyloop
:
1046 extract_number_and_incr (&mcnt
, &p
);
1047 fprintf (stderr
, "/on_failure_jump_nastyloop to %d", p
+ mcnt
- start
);
1050 case on_failure_jump_loop
:
1051 extract_number_and_incr (&mcnt
, &p
);
1052 fprintf (stderr
, "/on_failure_jump_loop to %d", p
+ mcnt
- start
);
1055 case on_failure_jump_smart
:
1056 extract_number_and_incr (&mcnt
, &p
);
1057 fprintf (stderr
, "/on_failure_jump_smart to %d", p
+ mcnt
- start
);
1061 extract_number_and_incr (&mcnt
, &p
);
1062 fprintf (stderr
, "/jump to %d", p
+ mcnt
- start
);
1066 extract_number_and_incr (&mcnt
, &p
);
1067 extract_number_and_incr (&mcnt2
, &p
);
1068 fprintf (stderr
, "/succeed_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1072 extract_number_and_incr (&mcnt
, &p
);
1073 extract_number_and_incr (&mcnt2
, &p
);
1074 fprintf (stderr
, "/jump_n to %d, %d times", p
- 2 + mcnt
- start
, mcnt2
);
1078 extract_number_and_incr (&mcnt
, &p
);
1079 extract_number_and_incr (&mcnt2
, &p
);
1080 fprintf (stderr
, "/set_number_at location %d to %d", p
- 2 + mcnt
- start
, mcnt2
);
1084 fprintf (stderr
, "/wordbound");
1088 fprintf (stderr
, "/notwordbound");
1092 fprintf (stderr
, "/wordbeg");
1096 fprintf (stderr
, "/wordend");
1100 fprintf (stderr
, "/symbeg");
1104 fprintf (stderr
, "/symend");
1108 fprintf (stderr
, "/syntaxspec");
1110 fprintf (stderr
, "/%d", mcnt
);
1114 fprintf (stderr
, "/notsyntaxspec");
1116 fprintf (stderr
, "/%d", mcnt
);
1121 fprintf (stderr
, "/before_dot");
1125 fprintf (stderr
, "/at_dot");
1129 fprintf (stderr
, "/after_dot");
1133 fprintf (stderr
, "/categoryspec");
1135 fprintf (stderr
, "/%d", mcnt
);
1138 case notcategoryspec
:
1139 fprintf (stderr
, "/notcategoryspec");
1141 fprintf (stderr
, "/%d", mcnt
);
1146 fprintf (stderr
, "/begbuf");
1150 fprintf (stderr
, "/endbuf");
1154 fprintf (stderr
, "?%d", *(p
-1));
1157 fprintf (stderr
, "\n");
1160 fprintf (stderr
, "%d:\tend of pattern.\n", p
- start
);
1165 print_compiled_pattern (bufp
)
1166 struct re_pattern_buffer
*bufp
;
1168 re_char
*buffer
= bufp
->buffer
;
1170 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1171 printf ("%ld bytes used/%ld bytes allocated.\n",
1172 bufp
->used
, bufp
->allocated
);
1174 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1176 printf ("fastmap: ");
1177 print_fastmap (bufp
->fastmap
);
1180 printf ("re_nsub: %d\t", bufp
->re_nsub
);
1181 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1182 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1183 printf ("no_sub: %d\t", bufp
->no_sub
);
1184 printf ("not_bol: %d\t", bufp
->not_bol
);
1185 printf ("not_eol: %d\t", bufp
->not_eol
);
1186 printf ("syntax: %lx\n", bufp
->syntax
);
1188 /* Perhaps we should print the translate table? */
1193 print_double_string (where
, string1
, size1
, string2
, size2
)
1206 if (FIRST_STRING_P (where
))
1208 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1209 putchar (string1
[this_char
]);
1214 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1215 putchar (string2
[this_char
]);
1219 #else /* not DEBUG */
1224 # define DEBUG_STATEMENT(e)
1225 # define DEBUG_PRINT1(x)
1226 # define DEBUG_PRINT2(x1, x2)
1227 # define DEBUG_PRINT3(x1, x2, x3)
1228 # define DEBUG_PRINT4(x1, x2, x3, x4)
1229 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1230 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1232 #endif /* not DEBUG */
1234 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1236 # define IF_LINT(Code) Code
1238 # define IF_LINT(Code) /* empty */
1241 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1242 also be assigned to arbitrarily: each pattern buffer stores its own
1243 syntax, so it can be changed between regex compilations. */
1244 /* This has no initializer because initialized variables in Emacs
1245 become read-only after dumping. */
1246 reg_syntax_t re_syntax_options
;
1249 /* Specify the precise syntax of regexps for compilation. This provides
1250 for compatibility for various utilities which historically have
1251 different, incompatible syntaxes.
1253 The argument SYNTAX is a bit mask comprised of the various bits
1254 defined in regex.h. We return the old syntax. */
1257 re_set_syntax (reg_syntax_t syntax
)
1259 reg_syntax_t ret
= re_syntax_options
;
1261 re_syntax_options
= syntax
;
1264 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1266 /* Regexp to use to replace spaces, or NULL meaning don't. */
1267 static re_char
*whitespace_regexp
;
1270 re_set_whitespace_regexp (const char *regexp
)
1272 whitespace_regexp
= (re_char
*) regexp
;
1274 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1276 /* This table gives an error message for each of the error codes listed
1277 in regex.h. Obviously the order here has to be same as there.
1278 POSIX doesn't require that we do anything for REG_NOERROR,
1279 but why not be nice? */
1281 static const char *re_error_msgid
[] =
1283 gettext_noop ("Success"), /* REG_NOERROR */
1284 gettext_noop ("No match"), /* REG_NOMATCH */
1285 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1286 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1287 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1288 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1289 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1290 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1291 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1292 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1293 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1294 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1295 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1296 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1297 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1298 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1299 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1300 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1303 /* Avoiding alloca during matching, to placate r_alloc. */
1305 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1306 searching and matching functions should not call alloca. On some
1307 systems, alloca is implemented in terms of malloc, and if we're
1308 using the relocating allocator routines, then malloc could cause a
1309 relocation, which might (if the strings being searched are in the
1310 ralloc heap) shift the data out from underneath the regexp
1313 Here's another reason to avoid allocation: Emacs
1314 processes input from X in a signal handler; processing X input may
1315 call malloc; if input arrives while a matching routine is calling
1316 malloc, then we're scrod. But Emacs can't just block input while
1317 calling matching routines; then we don't notice interrupts when
1318 they come in. So, Emacs blocks input around all regexp calls
1319 except the matching calls, which it leaves unprotected, in the
1320 faith that they will not malloc. */
1322 /* Normally, this is fine. */
1323 #define MATCH_MAY_ALLOCATE
1325 /* The match routines may not allocate if (1) they would do it with malloc
1326 and (2) it's not safe for them to use malloc.
1327 Note that if REL_ALLOC is defined, matching would not use malloc for the
1328 failure stack, but we would still use it for the register vectors;
1329 so REL_ALLOC should not affect this. */
1330 #if defined REGEX_MALLOC && defined emacs
1331 # undef MATCH_MAY_ALLOCATE
1335 /* Failure stack declarations and macros; both re_compile_fastmap and
1336 re_match_2 use a failure stack. These have to be macros because of
1337 REGEX_ALLOCATE_STACK. */
1340 /* Approximate number of failure points for which to initially allocate space
1341 when matching. If this number is exceeded, we allocate more
1342 space, so it is not a hard limit. */
1343 #ifndef INIT_FAILURE_ALLOC
1344 # define INIT_FAILURE_ALLOC 20
1347 /* Roughly the maximum number of failure points on the stack. Would be
1348 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1349 This is a variable only so users of regex can assign to it; we never
1350 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1351 before using it, so it should probably be a byte-count instead. */
1352 # if defined MATCH_MAY_ALLOCATE
1353 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1354 whose default stack limit is 2mb. In order for a larger
1355 value to work reliably, you have to try to make it accord
1356 with the process stack limit. */
1357 size_t re_max_failures
= 40000;
1359 size_t re_max_failures
= 4000;
1362 union fail_stack_elt
1365 /* This should be the biggest `int' that's no bigger than a pointer. */
1369 typedef union fail_stack_elt fail_stack_elt_t
;
1373 fail_stack_elt_t
*stack
;
1375 size_t avail
; /* Offset of next open position. */
1376 size_t frame
; /* Offset of the cur constructed frame. */
1379 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1382 /* Define macros to initialize and free the failure stack.
1383 Do `return -2' if the alloc fails. */
1385 #ifdef MATCH_MAY_ALLOCATE
1386 # define INIT_FAIL_STACK() \
1388 fail_stack.stack = \
1389 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1390 * sizeof (fail_stack_elt_t)); \
1392 if (fail_stack.stack == NULL) \
1395 fail_stack.size = INIT_FAILURE_ALLOC; \
1396 fail_stack.avail = 0; \
1397 fail_stack.frame = 0; \
1400 # define INIT_FAIL_STACK() \
1402 fail_stack.avail = 0; \
1403 fail_stack.frame = 0; \
1406 # define RETALLOC_IF(addr, n, t) \
1407 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1411 /* Double the size of FAIL_STACK, up to a limit
1412 which allows approximately `re_max_failures' items.
1414 Return 1 if succeeds, and 0 if either ran out of memory
1415 allocating space for it or it was already too large.
1417 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1419 /* Factor to increase the failure stack size by
1420 when we increase it.
1421 This used to be 2, but 2 was too wasteful
1422 because the old discarded stacks added up to as much space
1423 were as ultimate, maximum-size stack. */
1424 #define FAIL_STACK_GROWTH_FACTOR 4
1426 #define GROW_FAIL_STACK(fail_stack) \
1427 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1428 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1430 : ((fail_stack).stack \
1431 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1432 (fail_stack).size * sizeof (fail_stack_elt_t), \
1433 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1434 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1435 * FAIL_STACK_GROWTH_FACTOR))), \
1437 (fail_stack).stack == NULL \
1439 : ((fail_stack).size \
1440 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1441 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1442 * FAIL_STACK_GROWTH_FACTOR)) \
1443 / sizeof (fail_stack_elt_t)), \
1447 /* Push a pointer value onto the failure stack.
1448 Assumes the variable `fail_stack'. Probably should only
1449 be called from within `PUSH_FAILURE_POINT'. */
1450 #define PUSH_FAILURE_POINTER(item) \
1451 fail_stack.stack[fail_stack.avail++].pointer = (item)
1453 /* This pushes an integer-valued item onto the failure stack.
1454 Assumes the variable `fail_stack'. Probably should only
1455 be called from within `PUSH_FAILURE_POINT'. */
1456 #define PUSH_FAILURE_INT(item) \
1457 fail_stack.stack[fail_stack.avail++].integer = (item)
1459 /* These POP... operations complement the PUSH... operations.
1460 All assume that `fail_stack' is nonempty. */
1461 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1462 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1464 /* Individual items aside from the registers. */
1465 #define NUM_NONREG_ITEMS 3
1467 /* Used to examine the stack (to detect infinite loops). */
1468 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1469 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1470 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1471 #define TOP_FAILURE_HANDLE() fail_stack.frame
1474 #define ENSURE_FAIL_STACK(space) \
1475 while (REMAINING_AVAIL_SLOTS <= space) { \
1476 if (!GROW_FAIL_STACK (fail_stack)) \
1478 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1479 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1482 /* Push register NUM onto the stack. */
1483 #define PUSH_FAILURE_REG(num) \
1485 char *destination; \
1486 ENSURE_FAIL_STACK(3); \
1487 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1488 num, regstart[num], regend[num]); \
1489 PUSH_FAILURE_POINTER (regstart[num]); \
1490 PUSH_FAILURE_POINTER (regend[num]); \
1491 PUSH_FAILURE_INT (num); \
1494 /* Change the counter's value to VAL, but make sure that it will
1495 be reset when backtracking. */
1496 #define PUSH_NUMBER(ptr,val) \
1498 char *destination; \
1500 ENSURE_FAIL_STACK(3); \
1501 EXTRACT_NUMBER (c, ptr); \
1502 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1503 PUSH_FAILURE_INT (c); \
1504 PUSH_FAILURE_POINTER (ptr); \
1505 PUSH_FAILURE_INT (-1); \
1506 STORE_NUMBER (ptr, val); \
1509 /* Pop a saved register off the stack. */
1510 #define POP_FAILURE_REG_OR_COUNT() \
1512 long pfreg = POP_FAILURE_INT (); \
1515 /* It's a counter. */ \
1516 /* Here, we discard `const', making re_match non-reentrant. */ \
1517 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1518 pfreg = POP_FAILURE_INT (); \
1519 STORE_NUMBER (ptr, pfreg); \
1520 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, pfreg); \
1524 regend[pfreg] = POP_FAILURE_POINTER (); \
1525 regstart[pfreg] = POP_FAILURE_POINTER (); \
1526 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1527 pfreg, regstart[pfreg], regend[pfreg]); \
1531 /* Check that we are not stuck in an infinite loop. */
1532 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1534 ssize_t failure = TOP_FAILURE_HANDLE (); \
1535 /* Check for infinite matching loops */ \
1536 while (failure > 0 \
1537 && (FAILURE_STR (failure) == string_place \
1538 || FAILURE_STR (failure) == NULL)) \
1540 assert (FAILURE_PAT (failure) >= bufp->buffer \
1541 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1542 if (FAILURE_PAT (failure) == pat_cur) \
1547 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1548 failure = NEXT_FAILURE_HANDLE(failure); \
1550 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1553 /* Push the information about the state we will need
1554 if we ever fail back to it.
1556 Requires variables fail_stack, regstart, regend and
1557 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1560 Does `return FAILURE_CODE' if runs out of memory. */
1562 #define PUSH_FAILURE_POINT(pattern, string_place) \
1564 char *destination; \
1565 /* Must be int, so when we don't save any registers, the arithmetic \
1566 of 0 + -1 isn't done as unsigned. */ \
1568 DEBUG_STATEMENT (nfailure_points_pushed++); \
1569 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1570 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1571 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1573 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1575 DEBUG_PRINT1 ("\n"); \
1577 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1578 PUSH_FAILURE_INT (fail_stack.frame); \
1580 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1581 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1582 DEBUG_PRINT1 ("'\n"); \
1583 PUSH_FAILURE_POINTER (string_place); \
1585 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1586 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1587 PUSH_FAILURE_POINTER (pattern); \
1589 /* Close the frame by moving the frame pointer past it. */ \
1590 fail_stack.frame = fail_stack.avail; \
1593 /* Estimate the size of data pushed by a typical failure stack entry.
1594 An estimate is all we need, because all we use this for
1595 is to choose a limit for how big to make the failure stack. */
1596 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1597 #define TYPICAL_FAILURE_SIZE 20
1599 /* How many items can still be added to the stack without overflowing it. */
1600 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1603 /* Pops what PUSH_FAIL_STACK pushes.
1605 We restore into the parameters, all of which should be lvalues:
1606 STR -- the saved data position.
1607 PAT -- the saved pattern position.
1608 REGSTART, REGEND -- arrays of string positions.
1610 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1611 `pend', `string1', `size1', `string2', and `size2'. */
1613 #define POP_FAILURE_POINT(str, pat) \
1615 assert (!FAIL_STACK_EMPTY ()); \
1617 /* Remove failure points and point to how many regs pushed. */ \
1618 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1619 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1620 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1622 /* Pop the saved registers. */ \
1623 while (fail_stack.frame < fail_stack.avail) \
1624 POP_FAILURE_REG_OR_COUNT (); \
1626 pat = POP_FAILURE_POINTER (); \
1627 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1628 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1630 /* If the saved string location is NULL, it came from an \
1631 on_failure_keep_string_jump opcode, and we want to throw away the \
1632 saved NULL, thus retaining our current position in the string. */ \
1633 str = POP_FAILURE_POINTER (); \
1634 DEBUG_PRINT2 (" Popping string %p: `", str); \
1635 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1636 DEBUG_PRINT1 ("'\n"); \
1638 fail_stack.frame = POP_FAILURE_INT (); \
1639 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1641 assert (fail_stack.avail >= 0); \
1642 assert (fail_stack.frame <= fail_stack.avail); \
1644 DEBUG_STATEMENT (nfailure_points_popped++); \
1645 } while (0) /* POP_FAILURE_POINT */
1649 /* Registers are set to a sentinel when they haven't yet matched. */
1650 #define REG_UNSET(e) ((e) == NULL)
1652 /* Subroutine declarations and macros for regex_compile. */
1654 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1655 reg_syntax_t syntax
,
1656 struct re_pattern_buffer
*bufp
);
1657 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1658 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1659 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1660 int arg
, unsigned char *end
);
1661 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1662 int arg1
, int arg2
, unsigned char *end
);
1663 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1664 reg_syntax_t syntax
);
1665 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1666 reg_syntax_t syntax
);
1667 static re_char
*skip_one_char (re_char
*p
);
1668 static int analyse_first (re_char
*p
, re_char
*pend
,
1669 char *fastmap
, const int multibyte
);
1671 /* Fetch the next character in the uncompiled pattern, with no
1673 #define PATFETCH(c) \
1676 if (p == pend) return REG_EEND; \
1677 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1682 /* If `translate' is non-null, return translate[D], else just D. We
1683 cast the subscript to translate because some data is declared as
1684 `char *', to avoid warnings when a string constant is passed. But
1685 when we use a character as a subscript we must make it unsigned. */
1687 # define TRANSLATE(d) \
1688 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1692 /* Macros for outputting the compiled pattern into `buffer'. */
1694 /* If the buffer isn't allocated when it comes in, use this. */
1695 #define INIT_BUF_SIZE 32
1697 /* Make sure we have at least N more bytes of space in buffer. */
1698 #define GET_BUFFER_SPACE(n) \
1699 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1702 /* Make sure we have one more byte of buffer space and then add C to it. */
1703 #define BUF_PUSH(c) \
1705 GET_BUFFER_SPACE (1); \
1706 *b++ = (unsigned char) (c); \
1710 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1711 #define BUF_PUSH_2(c1, c2) \
1713 GET_BUFFER_SPACE (2); \
1714 *b++ = (unsigned char) (c1); \
1715 *b++ = (unsigned char) (c2); \
1719 /* Store a jump with opcode OP at LOC to location TO. We store a
1720 relative address offset by the three bytes the jump itself occupies. */
1721 #define STORE_JUMP(op, loc, to) \
1722 store_op1 (op, loc, (to) - (loc) - 3)
1724 /* Likewise, for a two-argument jump. */
1725 #define STORE_JUMP2(op, loc, to, arg) \
1726 store_op2 (op, loc, (to) - (loc) - 3, arg)
1728 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1729 #define INSERT_JUMP(op, loc, to) \
1730 insert_op1 (op, loc, (to) - (loc) - 3, b)
1732 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1733 #define INSERT_JUMP2(op, loc, to, arg) \
1734 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1737 /* This is not an arbitrary limit: the arguments which represent offsets
1738 into the pattern are two bytes long. So if 2^15 bytes turns out to
1739 be too small, many things would have to change. */
1740 # define MAX_BUF_SIZE (1L << 15)
1742 /* Extend the buffer by twice its current size via realloc and
1743 reset the pointers that pointed into the old block to point to the
1744 correct places in the new one. If extending the buffer results in it
1745 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1746 #if __BOUNDED_POINTERS__
1747 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1748 # define MOVE_BUFFER_POINTER(P) \
1749 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1750 SET_HIGH_BOUND (P), \
1751 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1752 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1755 SET_HIGH_BOUND (b); \
1756 SET_HIGH_BOUND (begalt); \
1757 if (fixup_alt_jump) \
1758 SET_HIGH_BOUND (fixup_alt_jump); \
1760 SET_HIGH_BOUND (laststart); \
1761 if (pending_exact) \
1762 SET_HIGH_BOUND (pending_exact); \
1765 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1766 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1768 #define EXTEND_BUFFER() \
1770 unsigned char *old_buffer = bufp->buffer; \
1771 if (bufp->allocated == MAX_BUF_SIZE) \
1773 bufp->allocated <<= 1; \
1774 if (bufp->allocated > MAX_BUF_SIZE) \
1775 bufp->allocated = MAX_BUF_SIZE; \
1776 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1777 if (bufp->buffer == NULL) \
1778 return REG_ESPACE; \
1779 /* If the buffer moved, move all the pointers into it. */ \
1780 if (old_buffer != bufp->buffer) \
1782 unsigned char *new_buffer = bufp->buffer; \
1783 MOVE_BUFFER_POINTER (b); \
1784 MOVE_BUFFER_POINTER (begalt); \
1785 if (fixup_alt_jump) \
1786 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1788 MOVE_BUFFER_POINTER (laststart); \
1789 if (pending_exact) \
1790 MOVE_BUFFER_POINTER (pending_exact); \
1792 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1796 /* Since we have one byte reserved for the register number argument to
1797 {start,stop}_memory, the maximum number of groups we can report
1798 things about is what fits in that byte. */
1799 #define MAX_REGNUM 255
1801 /* But patterns can have more than `MAX_REGNUM' registers. We just
1802 ignore the excess. */
1803 typedef int regnum_t
;
1806 /* Macros for the compile stack. */
1808 /* Since offsets can go either forwards or backwards, this type needs to
1809 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1810 /* int may be not enough when sizeof(int) == 2. */
1811 typedef long pattern_offset_t
;
1815 pattern_offset_t begalt_offset
;
1816 pattern_offset_t fixup_alt_jump
;
1817 pattern_offset_t laststart_offset
;
1819 } compile_stack_elt_t
;
1824 compile_stack_elt_t
*stack
;
1826 size_t avail
; /* Offset of next open position. */
1827 } compile_stack_type
;
1830 #define INIT_COMPILE_STACK_SIZE 32
1832 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1833 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1835 /* The next available element. */
1836 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1838 /* Explicit quit checking is only used on NTemacs and whenever we
1839 use polling to process input events. */
1840 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1841 extern int immediate_quit
;
1842 # define IMMEDIATE_QUIT_CHECK \
1844 if (immediate_quit) QUIT; \
1847 # define IMMEDIATE_QUIT_CHECK ((void)0)
1850 /* Structure to manage work area for range table. */
1851 struct range_table_work_area
1853 int *table
; /* actual work area. */
1854 int allocated
; /* allocated size for work area in bytes. */
1855 int used
; /* actually used size in words. */
1856 int bits
; /* flag to record character classes */
1859 /* Make sure that WORK_AREA can hold more N multibyte characters.
1860 This is used only in set_image_of_range and set_image_of_range_1.
1861 It expects WORK_AREA to be a pointer.
1862 If it can't get the space, it returns from the surrounding function. */
1864 #define EXTEND_RANGE_TABLE(work_area, n) \
1866 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1868 extend_range_table_work_area (&work_area); \
1869 if ((work_area).table == 0) \
1870 return (REG_ESPACE); \
1874 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1875 (work_area).bits |= (bit)
1877 /* Bits used to implement the multibyte-part of the various character classes
1878 such as [:alnum:] in a charset's range table. */
1879 #define BIT_WORD 0x1
1880 #define BIT_LOWER 0x2
1881 #define BIT_PUNCT 0x4
1882 #define BIT_SPACE 0x8
1883 #define BIT_UPPER 0x10
1884 #define BIT_MULTIBYTE 0x20
1886 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1887 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1889 EXTEND_RANGE_TABLE ((work_area), 2); \
1890 (work_area).table[(work_area).used++] = (range_start); \
1891 (work_area).table[(work_area).used++] = (range_end); \
1894 /* Free allocated memory for WORK_AREA. */
1895 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1897 if ((work_area).table) \
1898 free ((work_area).table); \
1901 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1902 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1903 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1904 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1907 /* Set the bit for character C in a list. */
1908 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1913 /* Store characters in the range FROM to TO in the bitmap at B (for
1914 ASCII and unibyte characters) and WORK_AREA (for multibyte
1915 characters) while translating them and paying attention to the
1916 continuity of translated characters.
1918 Implementation note: It is better to implement these fairly big
1919 macros by a function, but it's not that easy because macros called
1920 in this macro assume various local variables already declared. */
1922 /* Both FROM and TO are ASCII characters. */
1924 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1928 for (C0 = (FROM); C0 <= (TO); C0++) \
1930 C1 = TRANSLATE (C0); \
1931 if (! ASCII_CHAR_P (C1)) \
1933 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1934 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1937 SET_LIST_BIT (C1); \
1942 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1944 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1946 int C0, C1, C2, I; \
1947 int USED = RANGE_TABLE_WORK_USED (work_area); \
1949 for (C0 = (FROM); C0 <= (TO); C0++) \
1951 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1952 if (CHAR_BYTE8_P (C1)) \
1953 SET_LIST_BIT (C0); \
1956 C2 = TRANSLATE (C1); \
1958 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1960 SET_LIST_BIT (C1); \
1961 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1963 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1964 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1966 if (C2 >= from - 1 && C2 <= to + 1) \
1968 if (C2 == from - 1) \
1969 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1970 else if (C2 == to + 1) \
1971 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1976 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1982 /* Both FROM and TO are multibyte characters. */
1984 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1986 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1988 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1989 for (C0 = (FROM); C0 <= (TO); C0++) \
1991 C1 = TRANSLATE (C0); \
1992 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1993 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1994 SET_LIST_BIT (C2); \
1995 if (C1 >= (FROM) && C1 <= (TO)) \
1997 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1999 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2000 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2002 if (C1 >= from - 1 && C1 <= to + 1) \
2004 if (C1 == from - 1) \
2005 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2006 else if (C1 == to + 1) \
2007 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2012 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2018 /* Get the next unsigned number in the uncompiled pattern. */
2019 #define GET_UNSIGNED_NUMBER(num) \
2022 FREE_STACK_RETURN (REG_EBRACE); \
2026 while ('0' <= c && c <= '9') \
2032 num = num * 10 + c - '0'; \
2033 if (num / 10 != prev) \
2034 FREE_STACK_RETURN (REG_BADBR); \
2036 FREE_STACK_RETURN (REG_EBRACE); \
2042 #if ! WIDE_CHAR_SUPPORT
2044 /* Map a string to the char class it names (if any). */
2046 re_wctype (const re_char
*str
)
2048 const char *string
= (const char *) str
;
2049 if (STREQ (string
, "alnum")) return RECC_ALNUM
;
2050 else if (STREQ (string
, "alpha")) return RECC_ALPHA
;
2051 else if (STREQ (string
, "word")) return RECC_WORD
;
2052 else if (STREQ (string
, "ascii")) return RECC_ASCII
;
2053 else if (STREQ (string
, "nonascii")) return RECC_NONASCII
;
2054 else if (STREQ (string
, "graph")) return RECC_GRAPH
;
2055 else if (STREQ (string
, "lower")) return RECC_LOWER
;
2056 else if (STREQ (string
, "print")) return RECC_PRINT
;
2057 else if (STREQ (string
, "punct")) return RECC_PUNCT
;
2058 else if (STREQ (string
, "space")) return RECC_SPACE
;
2059 else if (STREQ (string
, "upper")) return RECC_UPPER
;
2060 else if (STREQ (string
, "unibyte")) return RECC_UNIBYTE
;
2061 else if (STREQ (string
, "multibyte")) return RECC_MULTIBYTE
;
2062 else if (STREQ (string
, "digit")) return RECC_DIGIT
;
2063 else if (STREQ (string
, "xdigit")) return RECC_XDIGIT
;
2064 else if (STREQ (string
, "cntrl")) return RECC_CNTRL
;
2065 else if (STREQ (string
, "blank")) return RECC_BLANK
;
2069 /* True if CH is in the char class CC. */
2071 re_iswctype (int ch
, re_wctype_t cc
)
2075 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2076 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2077 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2078 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2079 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2080 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2081 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2082 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2083 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2084 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2085 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2086 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2087 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2088 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2089 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2090 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2091 case RECC_WORD
: return ISWORD (ch
) != 0;
2092 case RECC_ERROR
: return false;
2098 /* Return a bit-pattern to use in the range-table bits to match multibyte
2099 chars of class CC. */
2101 re_wctype_to_bit (re_wctype_t cc
)
2105 case RECC_NONASCII
: case RECC_PRINT
: case RECC_GRAPH
:
2106 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2107 case RECC_ALPHA
: case RECC_ALNUM
: case RECC_WORD
: return BIT_WORD
;
2108 case RECC_LOWER
: return BIT_LOWER
;
2109 case RECC_UPPER
: return BIT_UPPER
;
2110 case RECC_PUNCT
: return BIT_PUNCT
;
2111 case RECC_SPACE
: return BIT_SPACE
;
2112 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2113 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2120 /* Filling in the work area of a range. */
2122 /* Actually extend the space in WORK_AREA. */
2125 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2127 work_area
->allocated
+= 16 * sizeof (int);
2128 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2134 /* Carefully find the ranges of codes that are equivalent
2135 under case conversion to the range start..end when passed through
2136 TRANSLATE. Handle the case where non-letters can come in between
2137 two upper-case letters (which happens in Latin-1).
2138 Also handle the case of groups of more than 2 case-equivalent chars.
2140 The basic method is to look at consecutive characters and see
2141 if they can form a run that can be handled as one.
2143 Returns -1 if successful, REG_ESPACE if ran out of space. */
2146 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2147 re_wchar_t start
, re_wchar_t end
,
2148 RE_TRANSLATE_TYPE translate
)
2150 /* `one_case' indicates a character, or a run of characters,
2151 each of which is an isolate (no case-equivalents).
2152 This includes all ASCII non-letters.
2154 `two_case' indicates a character, or a run of characters,
2155 each of which has two case-equivalent forms.
2156 This includes all ASCII letters.
2158 `strange' indicates a character that has more than one
2161 enum case_type
{one_case
, two_case
, strange
};
2163 /* Describe the run that is in progress,
2164 which the next character can try to extend.
2165 If run_type is strange, that means there really is no run.
2166 If run_type is one_case, then run_start...run_end is the run.
2167 If run_type is two_case, then the run is run_start...run_end,
2168 and the case-equivalents end at run_eqv_end. */
2170 enum case_type run_type
= strange
;
2171 int run_start
, run_end
, run_eqv_end
;
2173 Lisp_Object eqv_table
;
2175 if (!RE_TRANSLATE_P (translate
))
2177 EXTEND_RANGE_TABLE (work_area
, 2);
2178 work_area
->table
[work_area
->used
++] = (start
);
2179 work_area
->table
[work_area
->used
++] = (end
);
2183 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2185 for (; start
<= end
; start
++)
2187 enum case_type this_type
;
2188 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2189 int minchar
, maxchar
;
2191 /* Classify this character */
2193 this_type
= one_case
;
2194 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2195 this_type
= two_case
;
2197 this_type
= strange
;
2200 minchar
= start
, maxchar
= eqv
;
2202 minchar
= eqv
, maxchar
= start
;
2204 /* Can this character extend the run in progress? */
2205 if (this_type
== strange
|| this_type
!= run_type
2206 || !(minchar
== run_end
+ 1
2207 && (run_type
== two_case
2208 ? maxchar
== run_eqv_end
+ 1 : 1)))
2211 Record each of its equivalent ranges. */
2212 if (run_type
== one_case
)
2214 EXTEND_RANGE_TABLE (work_area
, 2);
2215 work_area
->table
[work_area
->used
++] = run_start
;
2216 work_area
->table
[work_area
->used
++] = run_end
;
2218 else if (run_type
== two_case
)
2220 EXTEND_RANGE_TABLE (work_area
, 4);
2221 work_area
->table
[work_area
->used
++] = run_start
;
2222 work_area
->table
[work_area
->used
++] = run_end
;
2223 work_area
->table
[work_area
->used
++]
2224 = RE_TRANSLATE (eqv_table
, run_start
);
2225 work_area
->table
[work_area
->used
++]
2226 = RE_TRANSLATE (eqv_table
, run_end
);
2231 if (this_type
== strange
)
2233 /* For a strange character, add each of its equivalents, one
2234 by one. Don't start a range. */
2237 EXTEND_RANGE_TABLE (work_area
, 2);
2238 work_area
->table
[work_area
->used
++] = eqv
;
2239 work_area
->table
[work_area
->used
++] = eqv
;
2240 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2242 while (eqv
!= start
);
2245 /* Add this char to the run, or start a new run. */
2246 else if (run_type
== strange
)
2248 /* Initialize a new range. */
2249 run_type
= this_type
;
2252 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2256 /* Extend a running range. */
2258 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2262 /* If a run is still in progress at the end, finish it now
2263 by recording its equivalent ranges. */
2264 if (run_type
== one_case
)
2266 EXTEND_RANGE_TABLE (work_area
, 2);
2267 work_area
->table
[work_area
->used
++] = run_start
;
2268 work_area
->table
[work_area
->used
++] = run_end
;
2270 else if (run_type
== two_case
)
2272 EXTEND_RANGE_TABLE (work_area
, 4);
2273 work_area
->table
[work_area
->used
++] = run_start
;
2274 work_area
->table
[work_area
->used
++] = run_end
;
2275 work_area
->table
[work_area
->used
++]
2276 = RE_TRANSLATE (eqv_table
, run_start
);
2277 work_area
->table
[work_area
->used
++]
2278 = RE_TRANSLATE (eqv_table
, run_end
);
2286 /* Record the image of the range start..end when passed through
2287 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2288 and is not even necessarily contiguous.
2289 Normally we approximate it with the smallest contiguous range that contains
2290 all the chars we need. However, for Latin-1 we go to extra effort
2293 This function is not called for ASCII ranges.
2295 Returns -1 if successful, REG_ESPACE if ran out of space. */
2298 set_image_of_range (struct range_table_work_area
*work_area
,
2299 re_wchar_t start
, re_wchar_t end
,
2300 RE_TRANSLATE_TYPE translate
)
2302 re_wchar_t cmin
, cmax
;
2305 /* For Latin-1 ranges, use set_image_of_range_1
2306 to get proper handling of ranges that include letters and nonletters.
2307 For a range that includes the whole of Latin-1, this is not necessary.
2308 For other character sets, we don't bother to get this right. */
2309 if (RE_TRANSLATE_P (translate
) && start
< 04400
2310 && !(start
< 04200 && end
>= 04377))
2317 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2327 EXTEND_RANGE_TABLE (work_area
, 2);
2328 work_area
->table
[work_area
->used
++] = (start
);
2329 work_area
->table
[work_area
->used
++] = (end
);
2331 cmin
= -1, cmax
= -1;
2333 if (RE_TRANSLATE_P (translate
))
2337 for (ch
= start
; ch
<= end
; ch
++)
2339 re_wchar_t c
= TRANSLATE (ch
);
2340 if (! (start
<= c
&& c
<= end
))
2346 cmin
= MIN (cmin
, c
);
2347 cmax
= MAX (cmax
, c
);
2354 EXTEND_RANGE_TABLE (work_area
, 2);
2355 work_area
->table
[work_area
->used
++] = (cmin
);
2356 work_area
->table
[work_area
->used
++] = (cmax
);
2364 #ifndef MATCH_MAY_ALLOCATE
2366 /* If we cannot allocate large objects within re_match_2_internal,
2367 we make the fail stack and register vectors global.
2368 The fail stack, we grow to the maximum size when a regexp
2370 The register vectors, we adjust in size each time we
2371 compile a regexp, according to the number of registers it needs. */
2373 static fail_stack_type fail_stack
;
2375 /* Size with which the following vectors are currently allocated.
2376 That is so we can make them bigger as needed,
2377 but never make them smaller. */
2378 static int regs_allocated_size
;
2380 static re_char
** regstart
, ** regend
;
2381 static re_char
**best_regstart
, **best_regend
;
2383 /* Make the register vectors big enough for NUM_REGS registers,
2384 but don't make them smaller. */
2387 regex_grow_registers (int num_regs
)
2389 if (num_regs
> regs_allocated_size
)
2391 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2392 RETALLOC_IF (regend
, num_regs
, re_char
*);
2393 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2394 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2396 regs_allocated_size
= num_regs
;
2400 #endif /* not MATCH_MAY_ALLOCATE */
2402 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2405 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2406 Returns one of error codes defined in `regex.h', or zero for success.
2408 Assumes the `allocated' (and perhaps `buffer') and `translate'
2409 fields are set in BUFP on entry.
2411 If it succeeds, results are put in BUFP (if it returns an error, the
2412 contents of BUFP are undefined):
2413 `buffer' is the compiled pattern;
2414 `syntax' is set to SYNTAX;
2415 `used' is set to the length of the compiled pattern;
2416 `fastmap_accurate' is zero;
2417 `re_nsub' is the number of subexpressions in PATTERN;
2418 `not_bol' and `not_eol' are zero;
2420 The `fastmap' field is neither examined nor set. */
2422 /* Insert the `jump' from the end of last alternative to "here".
2423 The space for the jump has already been allocated. */
2424 #define FIXUP_ALT_JUMP() \
2426 if (fixup_alt_jump) \
2427 STORE_JUMP (jump, fixup_alt_jump, b); \
2431 /* Return, freeing storage we allocated. */
2432 #define FREE_STACK_RETURN(value) \
2434 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2435 free (compile_stack.stack); \
2439 static reg_errcode_t
2440 regex_compile (const re_char
*pattern
, size_t size
, reg_syntax_t syntax
, struct re_pattern_buffer
*bufp
)
2442 /* We fetch characters from PATTERN here. */
2443 register re_wchar_t c
, c1
;
2445 /* Points to the end of the buffer, where we should append. */
2446 register unsigned char *b
;
2448 /* Keeps track of unclosed groups. */
2449 compile_stack_type compile_stack
;
2451 /* Points to the current (ending) position in the pattern. */
2453 /* `const' makes AIX compiler fail. */
2454 unsigned char *p
= pattern
;
2456 re_char
*p
= pattern
;
2458 re_char
*pend
= pattern
+ size
;
2460 /* How to translate the characters in the pattern. */
2461 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2463 /* Address of the count-byte of the most recently inserted `exactn'
2464 command. This makes it possible to tell if a new exact-match
2465 character can be added to that command or if the character requires
2466 a new `exactn' command. */
2467 unsigned char *pending_exact
= 0;
2469 /* Address of start of the most recently finished expression.
2470 This tells, e.g., postfix * where to find the start of its
2471 operand. Reset at the beginning of groups and alternatives. */
2472 unsigned char *laststart
= 0;
2474 /* Address of beginning of regexp, or inside of last group. */
2475 unsigned char *begalt
;
2477 /* Place in the uncompiled pattern (i.e., the {) to
2478 which to go back if the interval is invalid. */
2479 re_char
*beg_interval
;
2481 /* Address of the place where a forward jump should go to the end of
2482 the containing expression. Each alternative of an `or' -- except the
2483 last -- ends with a forward jump of this sort. */
2484 unsigned char *fixup_alt_jump
= 0;
2486 /* Work area for range table of charset. */
2487 struct range_table_work_area range_table_work
;
2489 /* If the object matched can contain multibyte characters. */
2490 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2492 /* Nonzero if we have pushed down into a subpattern. */
2493 int in_subpattern
= 0;
2495 /* These hold the values of p, pattern, and pend from the main
2496 pattern when we have pushed into a subpattern. */
2497 re_char
*main_p
IF_LINT (= NULL
);
2498 re_char
*main_pattern
IF_LINT (= NULL
);
2499 re_char
*main_pend
IF_LINT (= NULL
);
2503 DEBUG_PRINT1 ("\nCompiling pattern: ");
2506 unsigned debug_count
;
2508 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2509 putchar (pattern
[debug_count
]);
2514 /* Initialize the compile stack. */
2515 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2516 if (compile_stack
.stack
== NULL
)
2519 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2520 compile_stack
.avail
= 0;
2522 range_table_work
.table
= 0;
2523 range_table_work
.allocated
= 0;
2525 /* Initialize the pattern buffer. */
2526 bufp
->syntax
= syntax
;
2527 bufp
->fastmap_accurate
= 0;
2528 bufp
->not_bol
= bufp
->not_eol
= 0;
2529 bufp
->used_syntax
= 0;
2531 /* Set `used' to zero, so that if we return an error, the pattern
2532 printer (for debugging) will think there's no pattern. We reset it
2536 /* Always count groups, whether or not bufp->no_sub is set. */
2539 #if !defined emacs && !defined SYNTAX_TABLE
2540 /* Initialize the syntax table. */
2541 init_syntax_once ();
2544 if (bufp
->allocated
== 0)
2547 { /* If zero allocated, but buffer is non-null, try to realloc
2548 enough space. This loses if buffer's address is bogus, but
2549 that is the user's responsibility. */
2550 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2553 { /* Caller did not allocate a buffer. Do it for them. */
2554 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2556 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2558 bufp
->allocated
= INIT_BUF_SIZE
;
2561 begalt
= b
= bufp
->buffer
;
2563 /* Loop through the uncompiled pattern until we're at the end. */
2568 /* If this is the end of an included regexp,
2569 pop back to the main regexp and try again. */
2573 pattern
= main_pattern
;
2578 /* If this is the end of the main regexp, we are done. */
2590 /* If there's no special whitespace regexp, treat
2591 spaces normally. And don't try to do this recursively. */
2592 if (!whitespace_regexp
|| in_subpattern
)
2595 /* Peek past following spaces. */
2602 /* If the spaces are followed by a repetition op,
2603 treat them normally. */
2605 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2606 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2609 /* Replace the spaces with the whitespace regexp. */
2613 main_pattern
= pattern
;
2614 p
= pattern
= whitespace_regexp
;
2615 pend
= p
+ strlen ((const char *) p
);
2621 if ( /* If at start of pattern, it's an operator. */
2623 /* If context independent, it's an operator. */
2624 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2625 /* Otherwise, depends on what's come before. */
2626 || at_begline_loc_p (pattern
, p
, syntax
))
2627 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2636 if ( /* If at end of pattern, it's an operator. */
2638 /* If context independent, it's an operator. */
2639 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2640 /* Otherwise, depends on what's next. */
2641 || at_endline_loc_p (p
, pend
, syntax
))
2642 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2651 if ((syntax
& RE_BK_PLUS_QM
)
2652 || (syntax
& RE_LIMITED_OPS
))
2656 /* If there is no previous pattern... */
2659 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2660 FREE_STACK_RETURN (REG_BADRPT
);
2661 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2666 /* 1 means zero (many) matches is allowed. */
2667 boolean zero_times_ok
= 0, many_times_ok
= 0;
2670 /* If there is a sequence of repetition chars, collapse it
2671 down to just one (the right one). We can't combine
2672 interval operators with these because of, e.g., `a{2}*',
2673 which should only match an even number of `a's. */
2677 if ((syntax
& RE_FRUGAL
)
2678 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2682 zero_times_ok
|= c
!= '+';
2683 many_times_ok
|= c
!= '?';
2689 || (!(syntax
& RE_BK_PLUS_QM
)
2690 && (*p
== '+' || *p
== '?')))
2692 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2695 FREE_STACK_RETURN (REG_EESCAPE
);
2696 if (p
[1] == '+' || p
[1] == '?')
2697 PATFETCH (c
); /* Gobble up the backslash. */
2703 /* If we get here, we found another repeat character. */
2707 /* Star, etc. applied to an empty pattern is equivalent
2708 to an empty pattern. */
2709 if (!laststart
|| laststart
== b
)
2712 /* Now we know whether or not zero matches is allowed
2713 and also whether or not two or more matches is allowed. */
2718 boolean simple
= skip_one_char (laststart
) == b
;
2719 size_t startoffset
= 0;
2721 /* Check if the loop can match the empty string. */
2722 (simple
|| !analyse_first (laststart
, b
, NULL
, 0))
2723 ? on_failure_jump
: on_failure_jump_loop
;
2724 assert (skip_one_char (laststart
) <= b
);
2726 if (!zero_times_ok
&& simple
)
2727 { /* Since simple * loops can be made faster by using
2728 on_failure_keep_string_jump, we turn simple P+
2729 into PP* if P is simple. */
2730 unsigned char *p1
, *p2
;
2731 startoffset
= b
- laststart
;
2732 GET_BUFFER_SPACE (startoffset
);
2733 p1
= b
; p2
= laststart
;
2739 GET_BUFFER_SPACE (6);
2742 STORE_JUMP (ofj
, b
, b
+ 6);
2744 /* Simple * loops can use on_failure_keep_string_jump
2745 depending on what follows. But since we don't know
2746 that yet, we leave the decision up to
2747 on_failure_jump_smart. */
2748 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2749 laststart
+ startoffset
, b
+ 6);
2751 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2756 /* A simple ? pattern. */
2757 assert (zero_times_ok
);
2758 GET_BUFFER_SPACE (3);
2759 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2763 else /* not greedy */
2764 { /* I wish the greedy and non-greedy cases could be merged. */
2766 GET_BUFFER_SPACE (7); /* We might use less. */
2769 boolean emptyp
= analyse_first (laststart
, b
, NULL
, 0);
2771 /* The non-greedy multiple match looks like
2772 a repeat..until: we only need a conditional jump
2773 at the end of the loop. */
2774 if (emptyp
) BUF_PUSH (no_op
);
2775 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2776 : on_failure_jump
, b
, laststart
);
2780 /* The repeat...until naturally matches one or more.
2781 To also match zero times, we need to first jump to
2782 the end of the loop (its conditional jump). */
2783 INSERT_JUMP (jump
, laststart
, b
);
2789 /* non-greedy a?? */
2790 INSERT_JUMP (jump
, laststart
, b
+ 3);
2792 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2811 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2813 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2815 /* Ensure that we have enough space to push a charset: the
2816 opcode, the length count, and the bitset; 34 bytes in all. */
2817 GET_BUFFER_SPACE (34);
2821 /* We test `*p == '^' twice, instead of using an if
2822 statement, so we only need one BUF_PUSH. */
2823 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2827 /* Remember the first position in the bracket expression. */
2830 /* Push the number of bytes in the bitmap. */
2831 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2833 /* Clear the whole map. */
2834 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2836 /* charset_not matches newline according to a syntax bit. */
2837 if ((re_opcode_t
) b
[-2] == charset_not
2838 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2839 SET_LIST_BIT ('\n');
2841 /* Read in characters and ranges, setting map bits. */
2844 boolean escaped_char
= false;
2845 const unsigned char *p2
= p
;
2848 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2850 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2851 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2852 So the translation is done later in a loop. Example:
2853 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2856 /* \ might escape characters inside [...] and [^...]. */
2857 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2859 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2862 escaped_char
= true;
2866 /* Could be the end of the bracket expression. If it's
2867 not (i.e., when the bracket expression is `[]' so
2868 far), the ']' character bit gets set way below. */
2869 if (c
== ']' && p2
!= p1
)
2873 /* See if we're at the beginning of a possible character
2876 if (!escaped_char
&&
2877 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2879 /* Leave room for the null. */
2880 unsigned char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2881 const unsigned char *class_beg
;
2887 /* If pattern is `[[:'. */
2888 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2893 if ((c
== ':' && *p
== ']') || p
== pend
)
2895 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2898 /* This is in any case an invalid class name. */
2903 /* If isn't a word bracketed by `[:' and `:]':
2904 undo the ending character, the letters, and
2905 leave the leading `:' and `[' (but set bits for
2907 if (c
== ':' && *p
== ']')
2909 re_wctype_t cc
= re_wctype (str
);
2912 FREE_STACK_RETURN (REG_ECTYPE
);
2914 /* Throw away the ] at the end of the character
2918 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2921 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2922 if (re_iswctype (btowc (ch
), cc
))
2925 if (c
< (1 << BYTEWIDTH
))
2929 /* Most character classes in a multibyte match
2930 just set a flag. Exceptions are is_blank,
2931 is_digit, is_cntrl, and is_xdigit, since
2932 they can only match ASCII characters. We
2933 don't need to handle them for multibyte.
2934 They are distinguished by a negative wctype. */
2936 /* Setup the gl_state object to its buffer-defined
2937 value. This hardcodes the buffer-global
2938 syntax-table for ASCII chars, while the other chars
2939 will obey syntax-table properties. It's not ideal,
2940 but it's the way it's been done until now. */
2941 SETUP_BUFFER_SYNTAX_TABLE ();
2943 for (ch
= 0; ch
< 256; ++ch
)
2945 c
= RE_CHAR_TO_MULTIBYTE (ch
);
2946 if (! CHAR_BYTE8_P (c
)
2947 && re_iswctype (c
, cc
))
2953 if (ASCII_CHAR_P (c1
))
2955 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2959 SET_RANGE_TABLE_WORK_AREA_BIT
2960 (range_table_work
, re_wctype_to_bit (cc
));
2962 /* In most cases the matching rule for char classes
2963 only uses the syntax table for multibyte chars,
2964 so that the content of the syntax-table it is not
2965 hardcoded in the range_table. SPACE and WORD are
2966 the two exceptions. */
2967 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2968 bufp
->used_syntax
= 1;
2970 /* Repeat the loop. */
2975 /* Go back to right after the "[:". */
2979 /* Because the `:' may starts the range, we
2980 can't simply set bit and repeat the loop.
2981 Instead, just set it to C and handle below. */
2986 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2989 /* Discard the `-'. */
2992 /* Fetch the character which ends the range. */
2995 if (CHAR_BYTE8_P (c1
)
2996 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2997 /* Treat the range from a multibyte character to
2998 raw-byte character as empty. */
3003 /* Range from C to C. */
3008 if (syntax
& RE_NO_EMPTY_RANGES
)
3009 FREE_STACK_RETURN (REG_ERANGEX
);
3010 /* Else, repeat the loop. */
3015 /* Set the range into bitmap */
3016 for (; c
<= c1
; c
++)
3019 if (ch
< (1 << BYTEWIDTH
))
3026 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
3028 if (CHAR_BYTE8_P (c1
))
3029 c
= BYTE8_TO_CHAR (128);
3033 if (CHAR_BYTE8_P (c
))
3035 c
= CHAR_TO_BYTE8 (c
);
3036 c1
= CHAR_TO_BYTE8 (c1
);
3037 for (; c
<= c1
; c
++)
3042 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
3046 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
3053 /* Discard any (non)matching list bytes that are all 0 at the
3054 end of the map. Decrease the map-length byte too. */
3055 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3059 /* Build real range table from work area. */
3060 if (RANGE_TABLE_WORK_USED (range_table_work
)
3061 || RANGE_TABLE_WORK_BITS (range_table_work
))
3064 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
3066 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3067 bytes for flags, two for COUNT, and three bytes for
3069 GET_BUFFER_SPACE (4 + used
* 3);
3071 /* Indicate the existence of range table. */
3072 laststart
[1] |= 0x80;
3074 /* Store the character class flag bits into the range table.
3075 If not in emacs, these flag bits are always 0. */
3076 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3077 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3079 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3080 for (i
= 0; i
< used
; i
++)
3081 STORE_CHARACTER_AND_INCR
3082 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3089 if (syntax
& RE_NO_BK_PARENS
)
3096 if (syntax
& RE_NO_BK_PARENS
)
3103 if (syntax
& RE_NEWLINE_ALT
)
3110 if (syntax
& RE_NO_BK_VBAR
)
3117 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3118 goto handle_interval
;
3124 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3126 /* Do not translate the character after the \, so that we can
3127 distinguish, e.g., \B from \b, even if we normally would
3128 translate, e.g., B to b. */
3134 if (syntax
& RE_NO_BK_PARENS
)
3135 goto normal_backslash
;
3140 regnum_t regnum
= 0;
3143 /* Look for a special (?...) construct */
3144 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3146 PATFETCH (c
); /* Gobble up the '?'. */
3152 case ':': shy
= 1; break;
3154 /* An explicitly specified regnum must start
3157 FREE_STACK_RETURN (REG_BADPAT
);
3158 case '1': case '2': case '3': case '4':
3159 case '5': case '6': case '7': case '8': case '9':
3160 regnum
= 10*regnum
+ (c
- '0'); break;
3162 /* Only (?:...) is supported right now. */
3163 FREE_STACK_RETURN (REG_BADPAT
);
3170 regnum
= ++bufp
->re_nsub
;
3172 { /* It's actually not shy, but explicitly numbered. */
3174 if (regnum
> bufp
->re_nsub
)
3175 bufp
->re_nsub
= regnum
;
3176 else if (regnum
> bufp
->re_nsub
3177 /* Ideally, we'd want to check that the specified
3178 group can't have matched (i.e. all subgroups
3179 using the same regnum are in other branches of
3180 OR patterns), but we don't currently keep track
3181 of enough info to do that easily. */
3182 || group_in_compile_stack (compile_stack
, regnum
))
3183 FREE_STACK_RETURN (REG_BADPAT
);
3186 /* It's really shy. */
3187 regnum
= - bufp
->re_nsub
;
3189 if (COMPILE_STACK_FULL
)
3191 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3192 compile_stack_elt_t
);
3193 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3195 compile_stack
.size
<<= 1;
3198 /* These are the values to restore when we hit end of this
3199 group. They are all relative offsets, so that if the
3200 whole pattern moves because of realloc, they will still
3202 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3203 COMPILE_STACK_TOP
.fixup_alt_jump
3204 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3205 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3206 COMPILE_STACK_TOP
.regnum
= regnum
;
3208 /* Do not push a start_memory for groups beyond the last one
3209 we can represent in the compiled pattern. */
3210 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3211 BUF_PUSH_2 (start_memory
, regnum
);
3213 compile_stack
.avail
++;
3218 /* If we've reached MAX_REGNUM groups, then this open
3219 won't actually generate any code, so we'll have to
3220 clear pending_exact explicitly. */
3226 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3228 if (COMPILE_STACK_EMPTY
)
3230 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3231 goto normal_backslash
;
3233 FREE_STACK_RETURN (REG_ERPAREN
);
3239 /* See similar code for backslashed left paren above. */
3240 if (COMPILE_STACK_EMPTY
)
3242 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3245 FREE_STACK_RETURN (REG_ERPAREN
);
3248 /* Since we just checked for an empty stack above, this
3249 ``can't happen''. */
3250 assert (compile_stack
.avail
!= 0);
3252 /* We don't just want to restore into `regnum', because
3253 later groups should continue to be numbered higher,
3254 as in `(ab)c(de)' -- the second group is #2. */
3257 compile_stack
.avail
--;
3258 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3260 = COMPILE_STACK_TOP
.fixup_alt_jump
3261 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3263 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3264 regnum
= COMPILE_STACK_TOP
.regnum
;
3265 /* If we've reached MAX_REGNUM groups, then this open
3266 won't actually generate any code, so we'll have to
3267 clear pending_exact explicitly. */
3270 /* We're at the end of the group, so now we know how many
3271 groups were inside this one. */
3272 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3273 BUF_PUSH_2 (stop_memory
, regnum
);
3278 case '|': /* `\|'. */
3279 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3280 goto normal_backslash
;
3282 if (syntax
& RE_LIMITED_OPS
)
3285 /* Insert before the previous alternative a jump which
3286 jumps to this alternative if the former fails. */
3287 GET_BUFFER_SPACE (3);
3288 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3292 /* The alternative before this one has a jump after it
3293 which gets executed if it gets matched. Adjust that
3294 jump so it will jump to this alternative's analogous
3295 jump (put in below, which in turn will jump to the next
3296 (if any) alternative's such jump, etc.). The last such
3297 jump jumps to the correct final destination. A picture:
3303 If we are at `b', then fixup_alt_jump right now points to a
3304 three-byte space after `a'. We'll put in the jump, set
3305 fixup_alt_jump to right after `b', and leave behind three
3306 bytes which we'll fill in when we get to after `c'. */
3310 /* Mark and leave space for a jump after this alternative,
3311 to be filled in later either by next alternative or
3312 when know we're at the end of a series of alternatives. */
3314 GET_BUFFER_SPACE (3);
3323 /* If \{ is a literal. */
3324 if (!(syntax
& RE_INTERVALS
)
3325 /* If we're at `\{' and it's not the open-interval
3327 || (syntax
& RE_NO_BK_BRACES
))
3328 goto normal_backslash
;
3332 /* If got here, then the syntax allows intervals. */
3334 /* At least (most) this many matches must be made. */
3335 int lower_bound
= 0, upper_bound
= -1;
3339 GET_UNSIGNED_NUMBER (lower_bound
);
3342 GET_UNSIGNED_NUMBER (upper_bound
);
3344 /* Interval such as `{1}' => match exactly once. */
3345 upper_bound
= lower_bound
;
3347 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
3348 || (upper_bound
>= 0 && lower_bound
> upper_bound
))
3349 FREE_STACK_RETURN (REG_BADBR
);
3351 if (!(syntax
& RE_NO_BK_BRACES
))
3354 FREE_STACK_RETURN (REG_BADBR
);
3356 FREE_STACK_RETURN (REG_EESCAPE
);
3361 FREE_STACK_RETURN (REG_BADBR
);
3363 /* We just parsed a valid interval. */
3365 /* If it's invalid to have no preceding re. */
3368 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3369 FREE_STACK_RETURN (REG_BADRPT
);
3370 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3373 goto unfetch_interval
;
3376 if (upper_bound
== 0)
3377 /* If the upper bound is zero, just drop the sub pattern
3380 else if (lower_bound
== 1 && upper_bound
== 1)
3381 /* Just match it once: nothing to do here. */
3384 /* Otherwise, we have a nontrivial interval. When
3385 we're all done, the pattern will look like:
3386 set_number_at <jump count> <upper bound>
3387 set_number_at <succeed_n count> <lower bound>
3388 succeed_n <after jump addr> <succeed_n count>
3390 jump_n <succeed_n addr> <jump count>
3391 (The upper bound and `jump_n' are omitted if
3392 `upper_bound' is 1, though.) */
3394 { /* If the upper bound is > 1, we need to insert
3395 more at the end of the loop. */
3396 unsigned int nbytes
= (upper_bound
< 0 ? 3
3397 : upper_bound
> 1 ? 5 : 0);
3398 unsigned int startoffset
= 0;
3400 GET_BUFFER_SPACE (20); /* We might use less. */
3402 if (lower_bound
== 0)
3404 /* A succeed_n that starts with 0 is really a
3405 a simple on_failure_jump_loop. */
3406 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3412 /* Initialize lower bound of the `succeed_n', even
3413 though it will be set during matching by its
3414 attendant `set_number_at' (inserted next),
3415 because `re_compile_fastmap' needs to know.
3416 Jump to the `jump_n' we might insert below. */
3417 INSERT_JUMP2 (succeed_n
, laststart
,
3422 /* Code to initialize the lower bound. Insert
3423 before the `succeed_n'. The `5' is the last two
3424 bytes of this `set_number_at', plus 3 bytes of
3425 the following `succeed_n'. */
3426 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3431 if (upper_bound
< 0)
3433 /* A negative upper bound stands for infinity,
3434 in which case it degenerates to a plain jump. */
3435 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3438 else if (upper_bound
> 1)
3439 { /* More than one repetition is allowed, so
3440 append a backward jump to the `succeed_n'
3441 that starts this interval.
3443 When we've reached this during matching,
3444 we'll have matched the interval once, so
3445 jump back only `upper_bound - 1' times. */
3446 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3450 /* The location we want to set is the second
3451 parameter of the `jump_n'; that is `b-2' as
3452 an absolute address. `laststart' will be
3453 the `set_number_at' we're about to insert;
3454 `laststart+3' the number to set, the source
3455 for the relative address. But we are
3456 inserting into the middle of the pattern --
3457 so everything is getting moved up by 5.
3458 Conclusion: (b - 2) - (laststart + 3) + 5,
3459 i.e., b - laststart.
3461 We insert this at the beginning of the loop
3462 so that if we fail during matching, we'll
3463 reinitialize the bounds. */
3464 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3465 upper_bound
- 1, b
);
3470 beg_interval
= NULL
;
3475 /* If an invalid interval, match the characters as literals. */
3476 assert (beg_interval
);
3478 beg_interval
= NULL
;
3480 /* normal_char and normal_backslash need `c'. */
3483 if (!(syntax
& RE_NO_BK_BRACES
))
3485 assert (p
> pattern
&& p
[-1] == '\\');
3486 goto normal_backslash
;
3492 /* There is no way to specify the before_dot and after_dot
3493 operators. rms says this is ok. --karl */
3501 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3507 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3513 BUF_PUSH_2 (categoryspec
, c
);
3519 BUF_PUSH_2 (notcategoryspec
, c
);
3525 if (syntax
& RE_NO_GNU_OPS
)
3528 BUF_PUSH_2 (syntaxspec
, Sword
);
3533 if (syntax
& RE_NO_GNU_OPS
)
3536 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3541 if (syntax
& RE_NO_GNU_OPS
)
3547 if (syntax
& RE_NO_GNU_OPS
)
3553 if (syntax
& RE_NO_GNU_OPS
)
3562 FREE_STACK_RETURN (REG_BADPAT
);
3566 if (syntax
& RE_NO_GNU_OPS
)
3568 BUF_PUSH (wordbound
);
3572 if (syntax
& RE_NO_GNU_OPS
)
3574 BUF_PUSH (notwordbound
);
3578 if (syntax
& RE_NO_GNU_OPS
)
3584 if (syntax
& RE_NO_GNU_OPS
)
3589 case '1': case '2': case '3': case '4': case '5':
3590 case '6': case '7': case '8': case '9':
3594 if (syntax
& RE_NO_BK_REFS
)
3595 goto normal_backslash
;
3599 if (reg
> bufp
->re_nsub
|| reg
< 1
3600 /* Can't back reference to a subexp before its end. */
3601 || group_in_compile_stack (compile_stack
, reg
))
3602 FREE_STACK_RETURN (REG_ESUBREG
);
3605 BUF_PUSH_2 (duplicate
, reg
);
3612 if (syntax
& RE_BK_PLUS_QM
)
3615 goto normal_backslash
;
3619 /* You might think it would be useful for \ to mean
3620 not to translate; but if we don't translate it
3621 it will never match anything. */
3628 /* Expects the character in `c'. */
3630 /* If no exactn currently being built. */
3633 /* If last exactn not at current position. */
3634 || pending_exact
+ *pending_exact
+ 1 != b
3636 /* We have only one byte following the exactn for the count. */
3637 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3639 /* If followed by a repetition operator. */
3640 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3641 || ((syntax
& RE_BK_PLUS_QM
)
3642 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3643 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3644 || ((syntax
& RE_INTERVALS
)
3645 && ((syntax
& RE_NO_BK_BRACES
)
3646 ? p
!= pend
&& *p
== '{'
3647 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3649 /* Start building a new exactn. */
3653 BUF_PUSH_2 (exactn
, 0);
3654 pending_exact
= b
- 1;
3657 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3664 len
= CHAR_STRING (c
, b
);
3669 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3670 if (! CHAR_BYTE8_P (c1
))
3672 re_wchar_t c2
= TRANSLATE (c1
);
3674 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3680 (*pending_exact
) += len
;
3685 } /* while p != pend */
3688 /* Through the pattern now. */
3692 if (!COMPILE_STACK_EMPTY
)
3693 FREE_STACK_RETURN (REG_EPAREN
);
3695 /* If we don't want backtracking, force success
3696 the first time we reach the end of the compiled pattern. */
3697 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3700 /* We have succeeded; set the length of the buffer. */
3701 bufp
->used
= b
- bufp
->buffer
;
3706 re_compile_fastmap (bufp
);
3707 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3708 print_compiled_pattern (bufp
);
3713 #ifndef MATCH_MAY_ALLOCATE
3714 /* Initialize the failure stack to the largest possible stack. This
3715 isn't necessary unless we're trying to avoid calling alloca in
3716 the search and match routines. */
3718 int num_regs
= bufp
->re_nsub
+ 1;
3720 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3722 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3723 falk_stack
.stack
= realloc (fail_stack
.stack
,
3724 fail_stack
.size
* sizeof *falk_stack
.stack
);
3727 regex_grow_registers (num_regs
);
3729 #endif /* not MATCH_MAY_ALLOCATE */
3731 FREE_STACK_RETURN (REG_NOERROR
);
3732 } /* regex_compile */
3734 /* Subroutines for `regex_compile'. */
3736 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3739 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3741 *loc
= (unsigned char) op
;
3742 STORE_NUMBER (loc
+ 1, arg
);
3746 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3749 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3751 *loc
= (unsigned char) op
;
3752 STORE_NUMBER (loc
+ 1, arg1
);
3753 STORE_NUMBER (loc
+ 3, arg2
);
3757 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3758 for OP followed by two-byte integer parameter ARG. */
3761 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3763 register unsigned char *pfrom
= end
;
3764 register unsigned char *pto
= end
+ 3;
3766 while (pfrom
!= loc
)
3769 store_op1 (op
, loc
, arg
);
3773 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3776 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3778 register unsigned char *pfrom
= end
;
3779 register unsigned char *pto
= end
+ 5;
3781 while (pfrom
!= loc
)
3784 store_op2 (op
, loc
, arg1
, arg2
);
3788 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3789 after an alternative or a begin-subexpression. We assume there is at
3790 least one character before the ^. */
3793 at_begline_loc_p (const re_char
*pattern
, const re_char
*p
, reg_syntax_t syntax
)
3795 re_char
*prev
= p
- 2;
3796 boolean odd_backslashes
;
3798 /* After a subexpression? */
3800 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3802 /* After an alternative? */
3803 else if (*prev
== '|')
3804 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3806 /* After a shy subexpression? */
3807 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3809 /* Skip over optional regnum. */
3810 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3813 if (!(prev
- 2 >= pattern
3814 && prev
[-1] == '?' && prev
[-2] == '('))
3817 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3822 /* Count the number of preceding backslashes. */
3824 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3826 return (p
- prev
) & odd_backslashes
;
3830 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3831 at least one character after the $, i.e., `P < PEND'. */
3834 at_endline_loc_p (const re_char
*p
, const re_char
*pend
, reg_syntax_t syntax
)
3837 boolean next_backslash
= *next
== '\\';
3838 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3841 /* Before a subexpression? */
3842 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3843 : next_backslash
&& next_next
&& *next_next
== ')')
3844 /* Before an alternative? */
3845 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3846 : next_backslash
&& next_next
&& *next_next
== '|');
3850 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3851 false if it's not. */
3854 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3856 ssize_t this_element
;
3858 for (this_element
= compile_stack
.avail
- 1;
3861 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3868 If fastmap is non-NULL, go through the pattern and fill fastmap
3869 with all the possible leading chars. If fastmap is NULL, don't
3870 bother filling it up (obviously) and only return whether the
3871 pattern could potentially match the empty string.
3873 Return 1 if p..pend might match the empty string.
3874 Return 0 if p..pend matches at least one char.
3875 Return -1 if fastmap was not updated accurately. */
3878 analyse_first (const re_char
*p
, const re_char
*pend
, char *fastmap
, const int multibyte
)
3883 /* If all elements for base leading-codes in fastmap is set, this
3884 flag is set true. */
3885 boolean match_any_multibyte_characters
= false;
3889 /* The loop below works as follows:
3890 - It has a working-list kept in the PATTERN_STACK and which basically
3891 starts by only containing a pointer to the first operation.
3892 - If the opcode we're looking at is a match against some set of
3893 chars, then we add those chars to the fastmap and go on to the
3894 next work element from the worklist (done via `break').
3895 - If the opcode is a control operator on the other hand, we either
3896 ignore it (if it's meaningless at this point, such as `start_memory')
3897 or execute it (if it's a jump). If the jump has several destinations
3898 (i.e. `on_failure_jump'), then we push the other destination onto the
3900 We guarantee termination by ignoring backward jumps (more or less),
3901 so that `p' is monotonically increasing. More to the point, we
3902 never set `p' (or push) anything `<= p1'. */
3906 /* `p1' is used as a marker of how far back a `on_failure_jump'
3907 can go without being ignored. It is normally equal to `p'
3908 (which prevents any backward `on_failure_jump') except right
3909 after a plain `jump', to allow patterns such as:
3912 10: on_failure_jump 3
3913 as used for the *? operator. */
3916 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3922 /* If the first character has to match a backreference, that means
3923 that the group was empty (since it already matched). Since this
3924 is the only case that interests us here, we can assume that the
3925 backreference must match the empty string. */
3930 /* Following are the cases which match a character. These end
3936 /* If multibyte is nonzero, the first byte of each
3937 character is an ASCII or a leading code. Otherwise,
3938 each byte is a character. Thus, this works in both
3943 /* For the case of matching this unibyte regex
3944 against multibyte, we must set a leading code of
3945 the corresponding multibyte character. */
3946 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3948 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3955 /* We could put all the chars except for \n (and maybe \0)
3956 but we don't bother since it is generally not worth it. */
3957 if (!fastmap
) break;
3962 if (!fastmap
) break;
3964 /* Chars beyond end of bitmap are possible matches. */
3965 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3966 j
< (1 << BYTEWIDTH
); j
++)
3972 if (!fastmap
) break;
3973 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3974 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3976 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3980 if (/* Any leading code can possibly start a character
3981 which doesn't match the specified set of characters. */
3984 /* If we can match a character class, we can match any
3985 multibyte characters. */
3986 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3987 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3990 if (match_any_multibyte_characters
== false)
3992 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3993 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3995 match_any_multibyte_characters
= true;
3999 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
4000 && match_any_multibyte_characters
== false)
4002 /* Set fastmap[I] to 1 where I is a leading code of each
4003 multibyte character in the range table. */
4005 unsigned char lc1
, lc2
;
4007 /* Make P points the range table. `+ 2' is to skip flag
4008 bits for a character class. */
4009 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
4011 /* Extract the number of ranges in range table into COUNT. */
4012 EXTRACT_NUMBER_AND_INCR (count
, p
);
4013 for (; count
> 0; count
--, p
+= 3)
4015 /* Extract the start and end of each range. */
4016 EXTRACT_CHARACTER (c
, p
);
4017 lc1
= CHAR_LEADING_CODE (c
);
4019 EXTRACT_CHARACTER (c
, p
);
4020 lc2
= CHAR_LEADING_CODE (c
);
4021 for (j
= lc1
; j
<= lc2
; j
++)
4030 if (!fastmap
) break;
4032 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
4034 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4035 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
4039 /* This match depends on text properties. These end with
4040 aborting optimizations. */
4044 case notcategoryspec
:
4045 if (!fastmap
) break;
4046 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
4048 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
4049 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
4052 /* Any leading code can possibly start a character which
4053 has or doesn't has the specified category. */
4054 if (match_any_multibyte_characters
== false)
4056 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
4057 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
4059 match_any_multibyte_characters
= true;
4063 /* All cases after this match the empty string. These end with
4085 EXTRACT_NUMBER_AND_INCR (j
, p
);
4087 /* Backward jumps can only go back to code that we've already
4088 visited. `re_compile' should make sure this is true. */
4091 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4093 case on_failure_jump
:
4094 case on_failure_keep_string_jump
:
4095 case on_failure_jump_loop
:
4096 case on_failure_jump_nastyloop
:
4097 case on_failure_jump_smart
:
4103 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4104 to jump back to "just after here". */
4107 case on_failure_jump
:
4108 case on_failure_keep_string_jump
:
4109 case on_failure_jump_nastyloop
:
4110 case on_failure_jump_loop
:
4111 case on_failure_jump_smart
:
4112 EXTRACT_NUMBER_AND_INCR (j
, p
);
4114 ; /* Backward jump to be ignored. */
4116 { /* We have to look down both arms.
4117 We first go down the "straight" path so as to minimize
4118 stack usage when going through alternatives. */
4119 int r
= analyse_first (p
, pend
, fastmap
, multibyte
);
4127 /* This code simply does not properly handle forward jump_n. */
4128 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4130 /* jump_n can either jump or fall through. The (backward) jump
4131 case has already been handled, so we only need to look at the
4132 fallthrough case. */
4136 /* If N == 0, it should be an on_failure_jump_loop instead. */
4137 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4139 /* We only care about one iteration of the loop, so we don't
4140 need to consider the case where this behaves like an
4157 abort (); /* We have listed all the cases. */
4160 /* Getting here means we have found the possible starting
4161 characters for one path of the pattern -- and that the empty
4162 string does not match. We need not follow this path further. */
4166 /* We reached the end without matching anything. */
4169 } /* analyse_first */
4171 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4172 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4173 characters can start a string that matches the pattern. This fastmap
4174 is used by re_search to skip quickly over impossible starting points.
4176 Character codes above (1 << BYTEWIDTH) are not represented in the
4177 fastmap, but the leading codes are represented. Thus, the fastmap
4178 indicates which character sets could start a match.
4180 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4181 area as BUFP->fastmap.
4183 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4186 Returns 0 if we succeed, -2 if an internal error. */
4189 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4191 char *fastmap
= bufp
->fastmap
;
4194 assert (fastmap
&& bufp
->buffer
);
4196 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4197 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4199 analysis
= analyse_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4200 fastmap
, RE_MULTIBYTE_P (bufp
));
4201 bufp
->can_be_null
= (analysis
!= 0);
4203 } /* re_compile_fastmap */
4205 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4206 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4207 this memory for recording register information. STARTS and ENDS
4208 must be allocated using the malloc library routine, and must each
4209 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4211 If NUM_REGS == 0, then subsequent matches should allocate their own
4214 Unless this function is called, the first search or match using
4215 PATTERN_BUFFER will allocate its own register data, without
4216 freeing the old data. */
4219 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4223 bufp
->regs_allocated
= REGS_REALLOCATE
;
4224 regs
->num_regs
= num_regs
;
4225 regs
->start
= starts
;
4230 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4232 regs
->start
= regs
->end
= (regoff_t
*) 0;
4235 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4237 /* Searching routines. */
4239 /* Like re_search_2, below, but only one string is specified, and
4240 doesn't let you say where to stop matching. */
4243 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4244 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4246 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4249 WEAK_ALIAS (__re_search
, re_search
)
4251 /* Head address of virtual concatenation of string. */
4252 #define HEAD_ADDR_VSTRING(P) \
4253 (((P) >= size1 ? string2 : string1))
4255 /* Address of POS in the concatenation of virtual string. */
4256 #define POS_ADDR_VSTRING(POS) \
4257 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4259 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4260 virtual concatenation of STRING1 and STRING2, starting first at index
4261 STARTPOS, then at STARTPOS + 1, and so on.
4263 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4265 RANGE is how far to scan while trying to match. RANGE = 0 means try
4266 only at STARTPOS; in general, the last start tried is STARTPOS +
4269 In REGS, return the indices of the virtual concatenation of STRING1
4270 and STRING2 that matched the entire BUFP->buffer and its contained
4273 Do not consider matching one past the index STOP in the virtual
4274 concatenation of STRING1 and STRING2.
4276 We return either the position in the strings at which the match was
4277 found, -1 if no match, or -2 if error (such as failure
4281 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4282 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4283 struct re_registers
*regs
, ssize_t stop
)
4286 re_char
*string1
= (re_char
*) str1
;
4287 re_char
*string2
= (re_char
*) str2
;
4288 register char *fastmap
= bufp
->fastmap
;
4289 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4290 size_t total_size
= size1
+ size2
;
4291 ssize_t endpos
= startpos
+ range
;
4292 boolean anchored_start
;
4293 /* Nonzero if we are searching multibyte string. */
4294 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4296 /* Check for out-of-range STARTPOS. */
4297 if (startpos
< 0 || startpos
> total_size
)
4300 /* Fix up RANGE if it might eventually take us outside
4301 the virtual concatenation of STRING1 and STRING2.
4302 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4304 range
= 0 - startpos
;
4305 else if (endpos
> total_size
)
4306 range
= total_size
- startpos
;
4308 /* If the search isn't to be a backwards one, don't waste time in a
4309 search for a pattern anchored at beginning of buffer. */
4310 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4319 /* In a forward search for something that starts with \=.
4320 don't keep searching past point. */
4321 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4323 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4329 /* Update the fastmap now if not correct already. */
4330 if (fastmap
&& !bufp
->fastmap_accurate
)
4331 re_compile_fastmap (bufp
);
4333 /* See whether the pattern is anchored. */
4334 anchored_start
= (bufp
->buffer
[0] == begline
);
4337 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4339 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4341 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4345 /* Loop through the string, looking for a place to start matching. */
4348 /* If the pattern is anchored,
4349 skip quickly past places we cannot match.
4350 We don't bother to treat startpos == 0 specially
4351 because that case doesn't repeat. */
4352 if (anchored_start
&& startpos
> 0)
4354 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4355 : string2
[startpos
- size1
- 1])
4360 /* If a fastmap is supplied, skip quickly over characters that
4361 cannot be the start of a match. If the pattern can match the
4362 null string, however, we don't need to skip characters; we want
4363 the first null string. */
4364 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4366 register re_char
*d
;
4367 register re_wchar_t buf_ch
;
4369 d
= POS_ADDR_VSTRING (startpos
);
4371 if (range
> 0) /* Searching forwards. */
4373 register int lim
= 0;
4374 ssize_t irange
= range
;
4376 if (startpos
< size1
&& startpos
+ range
>= size1
)
4377 lim
= range
- (size1
- startpos
);
4379 /* Written out as an if-else to avoid testing `translate'
4381 if (RE_TRANSLATE_P (translate
))
4388 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4389 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4390 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4393 range
-= buf_charlen
;
4399 register re_wchar_t ch
, translated
;
4402 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4403 translated
= RE_TRANSLATE (translate
, ch
);
4404 if (translated
!= ch
4405 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4407 if (fastmap
[buf_ch
])
4420 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4421 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4423 range
-= buf_charlen
;
4427 while (range
> lim
&& !fastmap
[*d
])
4433 startpos
+= irange
- range
;
4435 else /* Searching backwards. */
4439 buf_ch
= STRING_CHAR (d
);
4440 buf_ch
= TRANSLATE (buf_ch
);
4441 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4446 register re_wchar_t ch
, translated
;
4449 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4450 translated
= TRANSLATE (ch
);
4451 if (translated
!= ch
4452 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4454 if (! fastmap
[TRANSLATE (buf_ch
)])
4460 /* If can't match the null string, and that's all we have left, fail. */
4461 if (range
>= 0 && startpos
== total_size
&& fastmap
4462 && !bufp
->can_be_null
)
4465 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4466 startpos
, regs
, stop
);
4479 /* Update STARTPOS to the next character boundary. */
4482 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4483 int len
= BYTES_BY_CHAR_HEAD (*p
);
4501 /* Update STARTPOS to the previous character boundary. */
4504 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4506 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4508 /* Find the head of multibyte form. */
4509 PREV_CHAR_BOUNDARY (p
, phead
);
4510 range
+= p0
- 1 - p
;
4514 startpos
-= p0
- 1 - p
;
4520 WEAK_ALIAS (__re_search_2
, re_search_2
)
4522 /* Declarations and macros for re_match_2. */
4524 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4525 register ssize_t len
,
4526 RE_TRANSLATE_TYPE translate
,
4527 const int multibyte
);
4529 /* This converts PTR, a pointer into one of the search strings `string1'
4530 and `string2' into an offset from the beginning of that string. */
4531 #define POINTER_TO_OFFSET(ptr) \
4532 (FIRST_STRING_P (ptr) \
4533 ? ((regoff_t) ((ptr) - string1)) \
4534 : ((regoff_t) ((ptr) - string2 + size1)))
4536 /* Call before fetching a character with *d. This switches over to
4537 string2 if necessary.
4538 Check re_match_2_internal for a discussion of why end_match_2 might
4539 not be within string2 (but be equal to end_match_1 instead). */
4540 #define PREFETCH() \
4543 /* End of string2 => fail. */ \
4544 if (dend == end_match_2) \
4546 /* End of string1 => advance to string2. */ \
4548 dend = end_match_2; \
4551 /* Call before fetching a char with *d if you already checked other limits.
4552 This is meant for use in lookahead operations like wordend, etc..
4553 where we might need to look at parts of the string that might be
4554 outside of the LIMITs (i.e past `stop'). */
4555 #define PREFETCH_NOLIMIT() \
4559 dend = end_match_2; \
4562 /* Test if at very beginning or at very end of the virtual concatenation
4563 of `string1' and `string2'. If only one string, it's `string2'. */
4564 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4565 #define AT_STRINGS_END(d) ((d) == end2)
4567 /* Disabled due to a compiler bug -- see comment at case wordbound */
4569 /* The comment at case wordbound is following one, but we don't use
4570 AT_WORD_BOUNDARY anymore to support multibyte form.
4572 The DEC Alpha C compiler 3.x generates incorrect code for the
4573 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4574 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4575 macro and introducing temporary variables works around the bug. */
4578 /* Test if D points to a character which is word-constituent. We have
4579 two special cases to check for: if past the end of string1, look at
4580 the first character in string2; and if before the beginning of
4581 string2, look at the last character in string1. */
4582 #define WORDCHAR_P(d) \
4583 (SYNTAX ((d) == end1 ? *string2 \
4584 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4587 /* Test if the character before D and the one at D differ with respect
4588 to being word-constituent. */
4589 #define AT_WORD_BOUNDARY(d) \
4590 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4591 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4594 /* Free everything we malloc. */
4595 #ifdef MATCH_MAY_ALLOCATE
4596 # define FREE_VAR(var) \
4604 # define FREE_VARIABLES() \
4606 REGEX_FREE_STACK (fail_stack.stack); \
4607 FREE_VAR (regstart); \
4608 FREE_VAR (regend); \
4609 FREE_VAR (best_regstart); \
4610 FREE_VAR (best_regend); \
4613 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4614 #endif /* not MATCH_MAY_ALLOCATE */
4617 /* Optimization routines. */
4619 /* If the operation is a match against one or more chars,
4620 return a pointer to the next operation, else return NULL. */
4622 skip_one_char (const re_char
*p
)
4624 switch (SWITCH_ENUM_CAST (*p
++))
4635 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4638 p
= CHARSET_RANGE_TABLE (p
- 1);
4639 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4640 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4643 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4650 case notcategoryspec
:
4662 /* Jump over non-matching operations. */
4664 skip_noops (const re_char
*p
, const re_char
*pend
)
4669 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
))
4678 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4689 /* Non-zero if "p1 matches something" implies "p2 fails". */
4691 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const re_char
*p1
, const re_char
*p2
)
4694 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4695 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4697 assert (p1
>= bufp
->buffer
&& p1
< pend
4698 && p2
>= bufp
->buffer
&& p2
<= pend
);
4700 /* Skip over open/close-group commands.
4701 If what follows this loop is a ...+ construct,
4702 look at what begins its body, since we will have to
4703 match at least one of that. */
4704 p2
= skip_noops (p2
, pend
);
4705 /* The same skip can be done for p1, except that this function
4706 is only used in the case where p1 is a simple match operator. */
4707 /* p1 = skip_noops (p1, pend); */
4709 assert (p1
>= bufp
->buffer
&& p1
< pend
4710 && p2
>= bufp
->buffer
&& p2
<= pend
);
4712 op2
= p2
== pend
? succeed
: *p2
;
4714 switch (SWITCH_ENUM_CAST (op2
))
4718 /* If we're at the end of the pattern, we can change. */
4719 if (skip_one_char (p1
))
4721 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4729 register re_wchar_t c
4730 = (re_opcode_t
) *p2
== endline
? '\n'
4731 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4733 if ((re_opcode_t
) *p1
== exactn
)
4735 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4737 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4742 else if ((re_opcode_t
) *p1
== charset
4743 || (re_opcode_t
) *p1
== charset_not
)
4745 int not = (re_opcode_t
) *p1
== charset_not
;
4747 /* Test if C is listed in charset (or charset_not)
4749 if (! multibyte
|| IS_REAL_ASCII (c
))
4751 if (c
< CHARSET_BITMAP_SIZE (p1
) * BYTEWIDTH
4752 && p1
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4755 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1
))
4756 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p1
);
4758 /* `not' is equal to 1 if c would match, which means
4759 that we can't change to pop_failure_jump. */
4762 DEBUG_PRINT1 (" No match => fast loop.\n");
4766 else if ((re_opcode_t
) *p1
== anychar
4769 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4777 if ((re_opcode_t
) *p1
== exactn
)
4778 /* Reuse the code above. */
4779 return mutually_exclusive_p (bufp
, p2
, p1
);
4781 /* It is hard to list up all the character in charset
4782 P2 if it includes multibyte character. Give up in
4784 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4786 /* Now, we are sure that P2 has no range table.
4787 So, for the size of bitmap in P2, `p2[1]' is
4788 enough. But P1 may have range table, so the
4789 size of bitmap table of P1 is extracted by
4790 using macro `CHARSET_BITMAP_SIZE'.
4792 In a multibyte case, we know that all the character
4793 listed in P2 is ASCII. In a unibyte case, P1 has only a
4794 bitmap table. So, in both cases, it is enough to test
4795 only the bitmap table of P1. */
4797 if ((re_opcode_t
) *p1
== charset
)
4800 /* We win if the charset inside the loop
4801 has no overlap with the one after the loop. */
4804 && idx
< CHARSET_BITMAP_SIZE (p1
));
4806 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4810 || idx
== CHARSET_BITMAP_SIZE (p1
))
4812 DEBUG_PRINT1 (" No match => fast loop.\n");
4816 else if ((re_opcode_t
) *p1
== charset_not
)
4819 /* We win if the charset_not inside the loop lists
4820 every character listed in the charset after. */
4821 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4822 if (! (p2
[2 + idx
] == 0
4823 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4824 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4829 DEBUG_PRINT1 (" No match => fast loop.\n");
4838 switch (SWITCH_ENUM_CAST (*p1
))
4842 /* Reuse the code above. */
4843 return mutually_exclusive_p (bufp
, p2
, p1
);
4845 /* When we have two charset_not, it's very unlikely that
4846 they don't overlap. The union of the two sets of excluded
4847 chars should cover all possible chars, which, as a matter of
4848 fact, is virtually impossible in multibyte buffers. */
4854 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4856 return ((re_opcode_t
) *p1
== syntaxspec
4857 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4859 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4862 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4864 return ((re_opcode_t
) *p1
== notsyntaxspec
4865 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4867 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4870 return (((re_opcode_t
) *p1
== notsyntaxspec
4871 || (re_opcode_t
) *p1
== syntaxspec
)
4876 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4877 case notcategoryspec
:
4878 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4890 /* Matching routines. */
4892 #ifndef emacs /* Emacs never uses this. */
4893 /* re_match is like re_match_2 except it takes only a single string. */
4896 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4897 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4899 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4900 size
, pos
, regs
, size
);
4903 WEAK_ALIAS (__re_match
, re_match
)
4904 #endif /* not emacs */
4907 /* In Emacs, this is the string or buffer in which we
4908 are matching. It is used for looking up syntax properties. */
4909 Lisp_Object re_match_object
;
4912 /* re_match_2 matches the compiled pattern in BUFP against the
4913 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4914 and SIZE2, respectively). We start matching at POS, and stop
4917 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4918 store offsets for the substring each group matched in REGS. See the
4919 documentation for exactly how many groups we fill.
4921 We return -1 if no match, -2 if an internal error (such as the
4922 failure stack overflowing). Otherwise, we return the length of the
4923 matched substring. */
4926 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4927 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4928 struct re_registers
*regs
, ssize_t stop
)
4934 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4935 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4936 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4939 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4940 (re_char
*) string2
, size2
,
4944 WEAK_ALIAS (__re_match_2
, re_match_2
)
4947 /* This is a separate function so that we can force an alloca cleanup
4950 re_match_2_internal (struct re_pattern_buffer
*bufp
, const re_char
*string1
,
4951 size_t size1
, const re_char
*string2
, size_t size2
,
4952 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4954 /* General temporaries. */
4958 /* Just past the end of the corresponding string. */
4959 re_char
*end1
, *end2
;
4961 /* Pointers into string1 and string2, just past the last characters in
4962 each to consider matching. */
4963 re_char
*end_match_1
, *end_match_2
;
4965 /* Where we are in the data, and the end of the current string. */
4968 /* Used sometimes to remember where we were before starting matching
4969 an operator so that we can go back in case of failure. This "atomic"
4970 behavior of matching opcodes is indispensable to the correctness
4971 of the on_failure_keep_string_jump optimization. */
4974 /* Where we are in the pattern, and the end of the pattern. */
4975 re_char
*p
= bufp
->buffer
;
4976 re_char
*pend
= p
+ bufp
->used
;
4978 /* We use this to map every character in the string. */
4979 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4981 /* Nonzero if BUFP is setup from a multibyte regex. */
4982 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4984 /* Nonzero if STRING1/STRING2 are multibyte. */
4985 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4987 /* Failure point stack. Each place that can handle a failure further
4988 down the line pushes a failure point on this stack. It consists of
4989 regstart, and regend for all registers corresponding to
4990 the subexpressions we're currently inside, plus the number of such
4991 registers, and, finally, two char *'s. The first char * is where
4992 to resume scanning the pattern; the second one is where to resume
4993 scanning the strings. */
4994 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4995 fail_stack_type fail_stack
;
4998 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5001 #if defined REL_ALLOC && defined REGEX_MALLOC
5002 /* This holds the pointer to the failure stack, when
5003 it is allocated relocatably. */
5004 fail_stack_elt_t
*failure_stack_ptr
;
5007 /* We fill all the registers internally, independent of what we
5008 return, for use in backreferences. The number here includes
5009 an element for register zero. */
5010 size_t num_regs
= bufp
->re_nsub
+ 1;
5012 /* Information on the contents of registers. These are pointers into
5013 the input strings; they record just what was matched (on this
5014 attempt) by a subexpression part of the pattern, that is, the
5015 regnum-th regstart pointer points to where in the pattern we began
5016 matching and the regnum-th regend points to right after where we
5017 stopped matching the regnum-th subexpression. (The zeroth register
5018 keeps track of what the whole pattern matches.) */
5019 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5020 re_char
**regstart
, **regend
;
5023 /* The following record the register info as found in the above
5024 variables when we find a match better than any we've seen before.
5025 This happens as we backtrack through the failure points, which in
5026 turn happens only if we have not yet matched the entire string. */
5027 unsigned best_regs_set
= false;
5028 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5029 re_char
**best_regstart
, **best_regend
;
5032 /* Logically, this is `best_regend[0]'. But we don't want to have to
5033 allocate space for that if we're not allocating space for anything
5034 else (see below). Also, we never need info about register 0 for
5035 any of the other register vectors, and it seems rather a kludge to
5036 treat `best_regend' differently than the rest. So we keep track of
5037 the end of the best match so far in a separate variable. We
5038 initialize this to NULL so that when we backtrack the first time
5039 and need to test it, it's not garbage. */
5040 re_char
*match_end
= NULL
;
5043 /* Counts the total number of registers pushed. */
5044 unsigned num_regs_pushed
= 0;
5047 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5051 #ifdef MATCH_MAY_ALLOCATE
5052 /* Do not bother to initialize all the register variables if there are
5053 no groups in the pattern, as it takes a fair amount of time. If
5054 there are groups, we include space for register 0 (the whole
5055 pattern), even though we never use it, since it simplifies the
5056 array indexing. We should fix this. */
5059 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5060 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5061 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5062 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5064 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5072 /* We must initialize all our variables to NULL, so that
5073 `FREE_VARIABLES' doesn't try to free them. */
5074 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5076 #endif /* MATCH_MAY_ALLOCATE */
5078 /* The starting position is bogus. */
5079 if (pos
< 0 || pos
> size1
+ size2
)
5085 /* Initialize subexpression text positions to -1 to mark ones that no
5086 start_memory/stop_memory has been seen for. Also initialize the
5087 register information struct. */
5088 for (reg
= 1; reg
< num_regs
; reg
++)
5089 regstart
[reg
] = regend
[reg
] = NULL
;
5091 /* We move `string1' into `string2' if the latter's empty -- but not if
5092 `string1' is null. */
5093 if (size2
== 0 && string1
!= NULL
)
5100 end1
= string1
+ size1
;
5101 end2
= string2
+ size2
;
5103 /* `p' scans through the pattern as `d' scans through the data.
5104 `dend' is the end of the input string that `d' points within. `d'
5105 is advanced into the following input string whenever necessary, but
5106 this happens before fetching; therefore, at the beginning of the
5107 loop, `d' can be pointing at the end of a string, but it cannot
5111 /* Only match within string2. */
5112 d
= string2
+ pos
- size1
;
5113 dend
= end_match_2
= string2
+ stop
- size1
;
5114 end_match_1
= end1
; /* Just to give it a value. */
5120 /* Only match within string1. */
5121 end_match_1
= string1
+ stop
;
5123 When we reach end_match_1, PREFETCH normally switches to string2.
5124 But in the present case, this means that just doing a PREFETCH
5125 makes us jump from `stop' to `gap' within the string.
5126 What we really want here is for the search to stop as
5127 soon as we hit end_match_1. That's why we set end_match_2
5128 to end_match_1 (since PREFETCH fails as soon as we hit
5130 end_match_2
= end_match_1
;
5133 { /* It's important to use this code when stop == size so that
5134 moving `d' from end1 to string2 will not prevent the d == dend
5135 check from catching the end of string. */
5137 end_match_2
= string2
+ stop
- size1
;
5143 DEBUG_PRINT1 ("The compiled pattern is: ");
5144 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5145 DEBUG_PRINT1 ("The string to match is: `");
5146 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5147 DEBUG_PRINT1 ("'\n");
5149 /* This loops over pattern commands. It exits by returning from the
5150 function if the match is complete, or it drops through if the match
5151 fails at this starting point in the input data. */
5154 DEBUG_PRINT2 ("\n%p: ", p
);
5157 { /* End of pattern means we might have succeeded. */
5158 DEBUG_PRINT1 ("end of pattern ... ");
5160 /* If we haven't matched the entire string, and we want the
5161 longest match, try backtracking. */
5162 if (d
!= end_match_2
)
5164 /* 1 if this match ends in the same string (string1 or string2)
5165 as the best previous match. */
5166 boolean same_str_p
= (FIRST_STRING_P (match_end
)
5167 == FIRST_STRING_P (d
));
5168 /* 1 if this match is the best seen so far. */
5169 boolean best_match_p
;
5171 /* AIX compiler got confused when this was combined
5172 with the previous declaration. */
5174 best_match_p
= d
> match_end
;
5176 best_match_p
= !FIRST_STRING_P (d
);
5178 DEBUG_PRINT1 ("backtracking.\n");
5180 if (!FAIL_STACK_EMPTY ())
5181 { /* More failure points to try. */
5183 /* If exceeds best match so far, save it. */
5184 if (!best_regs_set
|| best_match_p
)
5186 best_regs_set
= true;
5189 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5191 for (reg
= 1; reg
< num_regs
; reg
++)
5193 best_regstart
[reg
] = regstart
[reg
];
5194 best_regend
[reg
] = regend
[reg
];
5200 /* If no failure points, don't restore garbage. And if
5201 last match is real best match, don't restore second
5203 else if (best_regs_set
&& !best_match_p
)
5206 /* Restore best match. It may happen that `dend ==
5207 end_match_1' while the restored d is in string2.
5208 For example, the pattern `x.*y.*z' against the
5209 strings `x-' and `y-z-', if the two strings are
5210 not consecutive in memory. */
5211 DEBUG_PRINT1 ("Restoring best registers.\n");
5214 dend
= ((d
>= string1
&& d
<= end1
)
5215 ? end_match_1
: end_match_2
);
5217 for (reg
= 1; reg
< num_regs
; reg
++)
5219 regstart
[reg
] = best_regstart
[reg
];
5220 regend
[reg
] = best_regend
[reg
];
5223 } /* d != end_match_2 */
5226 DEBUG_PRINT1 ("Accepting match.\n");
5228 /* If caller wants register contents data back, do it. */
5229 if (regs
&& !bufp
->no_sub
)
5231 /* Have the register data arrays been allocated? */
5232 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5233 { /* No. So allocate them with malloc. We need one
5234 extra element beyond `num_regs' for the `-1' marker
5236 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
5237 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5238 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5239 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5244 bufp
->regs_allocated
= REGS_REALLOCATE
;
5246 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5247 { /* Yes. If we need more elements than were already
5248 allocated, reallocate them. If we need fewer, just
5250 if (regs
->num_regs
< num_regs
+ 1)
5252 regs
->num_regs
= num_regs
+ 1;
5253 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5254 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5255 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5264 /* These braces fend off a "empty body in an else-statement"
5265 warning under GCC when assert expands to nothing. */
5266 assert (bufp
->regs_allocated
== REGS_FIXED
);
5269 /* Convert the pointer data in `regstart' and `regend' to
5270 indices. Register zero has to be set differently,
5271 since we haven't kept track of any info for it. */
5272 if (regs
->num_regs
> 0)
5274 regs
->start
[0] = pos
;
5275 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5278 /* Go through the first `min (num_regs, regs->num_regs)'
5279 registers, since that is all we initialized. */
5280 for (reg
= 1; reg
< MIN (num_regs
, regs
->num_regs
); reg
++)
5282 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5283 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5287 = (regoff_t
) POINTER_TO_OFFSET (regstart
[reg
]);
5289 = (regoff_t
) POINTER_TO_OFFSET (regend
[reg
]);
5293 /* If the regs structure we return has more elements than
5294 were in the pattern, set the extra elements to -1. If
5295 we (re)allocated the registers, this is the case,
5296 because we always allocate enough to have at least one
5298 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5299 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5300 } /* regs && !bufp->no_sub */
5302 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5303 nfailure_points_pushed
, nfailure_points_popped
,
5304 nfailure_points_pushed
- nfailure_points_popped
);
5305 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
5307 mcnt
= POINTER_TO_OFFSET (d
) - pos
;
5309 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
5315 /* Otherwise match next pattern command. */
5316 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
5318 /* Ignore these. Used to ignore the n of succeed_n's which
5319 currently have n == 0. */
5321 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5325 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5328 /* Match the next n pattern characters exactly. The following
5329 byte in the pattern defines n, and the n bytes after that
5330 are the characters to match. */
5333 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
5335 /* Remember the start point to rollback upon failure. */
5339 /* This is written out as an if-else so we don't waste time
5340 testing `translate' inside the loop. */
5341 if (RE_TRANSLATE_P (translate
))
5345 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5365 /* The cost of testing `translate' is comparatively small. */
5366 if (target_multibyte
)
5369 int pat_charlen
, buf_charlen
;
5374 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5377 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5380 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5382 if (TRANSLATE (buf_ch
) != pat_ch
)
5390 mcnt
-= pat_charlen
;
5402 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5403 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5410 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5411 if (! CHAR_BYTE8_P (buf_ch
))
5413 buf_ch
= TRANSLATE (buf_ch
);
5414 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5420 if (buf_ch
!= pat_ch
)
5433 /* Match any character except possibly a newline or a null. */
5439 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5442 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5444 buf_ch
= TRANSLATE (buf_ch
);
5446 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
5448 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
5449 && buf_ch
== '\000'))
5452 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
5461 register unsigned int c
;
5462 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
5465 /* Start of actual range_table, or end of bitmap if there is no
5467 re_char
*range_table
IF_LINT (= NULL
);
5469 /* Nonzero if there is a range table. */
5470 int range_table_exists
;
5472 /* Number of ranges of range table. This is not included
5473 in the initial byte-length of the command. */
5476 /* Whether matching against a unibyte character. */
5477 boolean unibyte_char
= false;
5479 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5481 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
5483 if (range_table_exists
)
5485 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
5486 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
5490 c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5491 if (target_multibyte
)
5496 c1
= RE_CHAR_TO_UNIBYTE (c
);
5499 unibyte_char
= true;
5505 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5507 if (! CHAR_BYTE8_P (c1
))
5509 c1
= TRANSLATE (c1
);
5510 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5513 unibyte_char
= true;
5518 unibyte_char
= true;
5521 if (unibyte_char
&& c
< (1 << BYTEWIDTH
))
5522 { /* Lookup bitmap. */
5523 /* Cast to `unsigned' instead of `unsigned char' in
5524 case the bit list is a full 32 bytes long. */
5525 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
5526 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5530 else if (range_table_exists
)
5532 int class_bits
= CHARSET_RANGE_TABLE_BITS (&p
[-1]);
5534 if ( (class_bits
& BIT_LOWER
&& ISLOWER (c
))
5535 | (class_bits
& BIT_MULTIBYTE
)
5536 | (class_bits
& BIT_PUNCT
&& ISPUNCT (c
))
5537 | (class_bits
& BIT_SPACE
&& ISSPACE (c
))
5538 | (class_bits
& BIT_UPPER
&& ISUPPER (c
))
5539 | (class_bits
& BIT_WORD
&& ISWORD (c
)))
5542 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
5546 if (range_table_exists
)
5547 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
5549 p
+= CHARSET_BITMAP_SIZE (&p
[-1]) + 1;
5551 if (!not) goto fail
;
5558 /* The beginning of a group is represented by start_memory.
5559 The argument is the register number. The text
5560 matched within the group is recorded (in the internal
5561 registers data structure) under the register number. */
5563 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p
);
5565 /* In case we need to undo this operation (via backtracking). */
5566 PUSH_FAILURE_REG ((unsigned int)*p
);
5569 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5570 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
5572 /* Move past the register number and inner group count. */
5577 /* The stop_memory opcode represents the end of a group. Its
5578 argument is the same as start_memory's: the register number. */
5580 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p
);
5582 assert (!REG_UNSET (regstart
[*p
]));
5583 /* Strictly speaking, there should be code such as:
5585 assert (REG_UNSET (regend[*p]));
5586 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5588 But the only info to be pushed is regend[*p] and it is known to
5589 be UNSET, so there really isn't anything to push.
5590 Not pushing anything, on the other hand deprives us from the
5591 guarantee that regend[*p] is UNSET since undoing this operation
5592 will not reset its value properly. This is not important since
5593 the value will only be read on the next start_memory or at
5594 the very end and both events can only happen if this stop_memory
5598 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
5600 /* Move past the register number and the inner group count. */
5605 /* \<digit> has been turned into a `duplicate' command which is
5606 followed by the numeric value of <digit> as the register number. */
5609 register re_char
*d2
, *dend2
;
5610 int regno
= *p
++; /* Get which register to match against. */
5611 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
5613 /* Can't back reference a group which we've never matched. */
5614 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5617 /* Where in input to try to start matching. */
5618 d2
= regstart
[regno
];
5620 /* Remember the start point to rollback upon failure. */
5623 /* Where to stop matching; if both the place to start and
5624 the place to stop matching are in the same string, then
5625 set to the place to stop, otherwise, for now have to use
5626 the end of the first string. */
5628 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5629 == FIRST_STRING_P (regend
[regno
]))
5630 ? regend
[regno
] : end_match_1
);
5633 /* If necessary, advance to next segment in register
5637 if (dend2
== end_match_2
) break;
5638 if (dend2
== regend
[regno
]) break;
5640 /* End of string1 => advance to string2. */
5642 dend2
= regend
[regno
];
5644 /* At end of register contents => success */
5645 if (d2
== dend2
) break;
5647 /* If necessary, advance to next segment in data. */
5650 /* How many characters left in this segment to match. */
5653 /* Want how many consecutive characters we can match in
5654 one shot, so, if necessary, adjust the count. */
5655 if (mcnt
> dend2
- d2
)
5658 /* Compare that many; failure if mismatch, else move
5660 if (RE_TRANSLATE_P (translate
)
5661 ? bcmp_translate (d
, d2
, mcnt
, translate
, target_multibyte
)
5662 : memcmp (d
, d2
, mcnt
))
5667 d
+= mcnt
, d2
+= mcnt
;
5673 /* begline matches the empty string at the beginning of the string
5674 (unless `not_bol' is set in `bufp'), and after newlines. */
5676 DEBUG_PRINT1 ("EXECUTING begline.\n");
5678 if (AT_STRINGS_BEG (d
))
5680 if (!bufp
->not_bol
) break;
5685 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5689 /* In all other cases, we fail. */
5693 /* endline is the dual of begline. */
5695 DEBUG_PRINT1 ("EXECUTING endline.\n");
5697 if (AT_STRINGS_END (d
))
5699 if (!bufp
->not_eol
) break;
5703 PREFETCH_NOLIMIT ();
5710 /* Match at the very beginning of the data. */
5712 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5713 if (AT_STRINGS_BEG (d
))
5718 /* Match at the very end of the data. */
5720 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5721 if (AT_STRINGS_END (d
))
5726 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5727 pushes NULL as the value for the string on the stack. Then
5728 `POP_FAILURE_POINT' will keep the current value for the
5729 string, instead of restoring it. To see why, consider
5730 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5731 then the . fails against the \n. But the next thing we want
5732 to do is match the \n against the \n; if we restored the
5733 string value, we would be back at the foo.
5735 Because this is used only in specific cases, we don't need to
5736 check all the things that `on_failure_jump' does, to make
5737 sure the right things get saved on the stack. Hence we don't
5738 share its code. The only reason to push anything on the
5739 stack at all is that otherwise we would have to change
5740 `anychar's code to do something besides goto fail in this
5741 case; that seems worse than this. */
5742 case on_failure_keep_string_jump
:
5743 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5744 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5747 PUSH_FAILURE_POINT (p
- 3, NULL
);
5750 /* A nasty loop is introduced by the non-greedy *? and +?.
5751 With such loops, the stack only ever contains one failure point
5752 at a time, so that a plain on_failure_jump_loop kind of
5753 cycle detection cannot work. Worse yet, such a detection
5754 can not only fail to detect a cycle, but it can also wrongly
5755 detect a cycle (between different instantiations of the same
5757 So the method used for those nasty loops is a little different:
5758 We use a special cycle-detection-stack-frame which is pushed
5759 when the on_failure_jump_nastyloop failure-point is *popped*.
5760 This special frame thus marks the beginning of one iteration
5761 through the loop and we can hence easily check right here
5762 whether something matched between the beginning and the end of
5764 case on_failure_jump_nastyloop
:
5765 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5766 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5769 assert ((re_opcode_t
)p
[-4] == no_op
);
5772 CHECK_INFINITE_LOOP (p
- 4, d
);
5774 /* If there's a cycle, just continue without pushing
5775 this failure point. The failure point is the "try again"
5776 option, which shouldn't be tried.
5777 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5778 PUSH_FAILURE_POINT (p
- 3, d
);
5782 /* Simple loop detecting on_failure_jump: just check on the
5783 failure stack if the same spot was already hit earlier. */
5784 case on_failure_jump_loop
:
5786 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5787 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5791 CHECK_INFINITE_LOOP (p
- 3, d
);
5793 /* If there's a cycle, get out of the loop, as if the matching
5794 had failed. We used to just `goto fail' here, but that was
5795 aborting the search a bit too early: we want to keep the
5796 empty-loop-match and keep matching after the loop.
5797 We want (x?)*y\1z to match both xxyz and xxyxz. */
5800 PUSH_FAILURE_POINT (p
- 3, d
);
5805 /* Uses of on_failure_jump:
5807 Each alternative starts with an on_failure_jump that points
5808 to the beginning of the next alternative. Each alternative
5809 except the last ends with a jump that in effect jumps past
5810 the rest of the alternatives. (They really jump to the
5811 ending jump of the following alternative, because tensioning
5812 these jumps is a hassle.)
5814 Repeats start with an on_failure_jump that points past both
5815 the repetition text and either the following jump or
5816 pop_failure_jump back to this on_failure_jump. */
5817 case on_failure_jump
:
5818 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5819 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5822 PUSH_FAILURE_POINT (p
-3, d
);
5825 /* This operation is used for greedy *.
5826 Compare the beginning of the repeat with what in the
5827 pattern follows its end. If we can establish that there
5828 is nothing that they would both match, i.e., that we
5829 would have to backtrack because of (as in, e.g., `a*a')
5830 then we can use a non-backtracking loop based on
5831 on_failure_keep_string_jump instead of on_failure_jump. */
5832 case on_failure_jump_smart
:
5833 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5834 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5837 re_char
*p1
= p
; /* Next operation. */
5838 /* Here, we discard `const', making re_match non-reentrant. */
5839 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5840 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5842 p
-= 3; /* Reset so that we will re-execute the
5843 instruction once it's been changed. */
5845 EXTRACT_NUMBER (mcnt
, p2
- 2);
5847 /* Ensure this is a indeed the trivial kind of loop
5848 we are expecting. */
5849 assert (skip_one_char (p1
) == p2
- 3);
5850 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5851 DEBUG_STATEMENT (debug
+= 2);
5852 if (mutually_exclusive_p (bufp
, p1
, p2
))
5854 /* Use a fast `on_failure_keep_string_jump' loop. */
5855 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5856 *p3
= (unsigned char) on_failure_keep_string_jump
;
5857 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5861 /* Default to a safe `on_failure_jump' loop. */
5862 DEBUG_PRINT1 (" smart default => slow loop.\n");
5863 *p3
= (unsigned char) on_failure_jump
;
5865 DEBUG_STATEMENT (debug
-= 2);
5869 /* Unconditionally jump (without popping any failure points). */
5872 IMMEDIATE_QUIT_CHECK
;
5873 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5874 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5875 p
+= mcnt
; /* Do the jump. */
5876 DEBUG_PRINT2 ("(to %p).\n", p
);
5880 /* Have to succeed matching what follows at least n times.
5881 After that, handle like `on_failure_jump'. */
5883 /* Signedness doesn't matter since we only compare MCNT to 0. */
5884 EXTRACT_NUMBER (mcnt
, p
+ 2);
5885 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5887 /* Originally, mcnt is how many times we HAVE to succeed. */
5890 /* Here, we discard `const', making re_match non-reentrant. */
5891 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5894 PUSH_NUMBER (p2
, mcnt
);
5897 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5902 /* Signedness doesn't matter since we only compare MCNT to 0. */
5903 EXTRACT_NUMBER (mcnt
, p
+ 2);
5904 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5906 /* Originally, this is how many times we CAN jump. */
5909 /* Here, we discard `const', making re_match non-reentrant. */
5910 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5912 PUSH_NUMBER (p2
, mcnt
);
5913 goto unconditional_jump
;
5915 /* If don't have to jump any more, skip over the rest of command. */
5922 unsigned char *p2
; /* Location of the counter. */
5923 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5925 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5926 /* Here, we discard `const', making re_match non-reentrant. */
5927 p2
= (unsigned char*) p
+ mcnt
;
5928 /* Signedness doesn't matter since we only copy MCNT's bits . */
5929 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5930 DEBUG_PRINT3 (" Setting %p to %d.\n", p2
, mcnt
);
5931 PUSH_NUMBER (p2
, mcnt
);
5938 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5939 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5941 /* We SUCCEED (or FAIL) in one of the following cases: */
5943 /* Case 1: D is at the beginning or the end of string. */
5944 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5948 /* C1 is the character before D, S1 is the syntax of C1, C2
5949 is the character at D, and S2 is the syntax of C2. */
5954 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5955 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5956 UPDATE_SYNTAX_TABLE (charpos
);
5958 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5961 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5963 PREFETCH_NOLIMIT ();
5964 GET_CHAR_AFTER (c2
, d
, dummy
);
5967 if (/* Case 2: Only one of S1 and S2 is Sword. */
5968 ((s1
== Sword
) != (s2
== Sword
))
5969 /* Case 3: Both of S1 and S2 are Sword, and macro
5970 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5971 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5981 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5983 /* We FAIL in one of the following cases: */
5985 /* Case 1: D is at the end of string. */
5986 if (AT_STRINGS_END (d
))
5990 /* C1 is the character before D, S1 is the syntax of C1, C2
5991 is the character at D, and S2 is the syntax of C2. */
5996 ssize_t offset
= PTR_TO_OFFSET (d
);
5997 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5998 UPDATE_SYNTAX_TABLE (charpos
);
6001 GET_CHAR_AFTER (c2
, d
, dummy
);
6004 /* Case 2: S2 is not Sword. */
6008 /* Case 3: D is not at the beginning of string ... */
6009 if (!AT_STRINGS_BEG (d
))
6011 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6013 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6017 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6019 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6026 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6028 /* We FAIL in one of the following cases: */
6030 /* Case 1: D is at the beginning of string. */
6031 if (AT_STRINGS_BEG (d
))
6035 /* C1 is the character before D, S1 is the syntax of C1, C2
6036 is the character at D, and S2 is the syntax of C2. */
6041 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6042 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6043 UPDATE_SYNTAX_TABLE (charpos
);
6045 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6048 /* Case 2: S1 is not Sword. */
6052 /* Case 3: D is not at the end of string ... */
6053 if (!AT_STRINGS_END (d
))
6055 PREFETCH_NOLIMIT ();
6056 GET_CHAR_AFTER (c2
, d
, dummy
);
6058 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
6062 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6064 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6071 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6073 /* We FAIL in one of the following cases: */
6075 /* Case 1: D is at the end of string. */
6076 if (AT_STRINGS_END (d
))
6080 /* C1 is the character before D, S1 is the syntax of C1, C2
6081 is the character at D, and S2 is the syntax of C2. */
6085 ssize_t offset
= PTR_TO_OFFSET (d
);
6086 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6087 UPDATE_SYNTAX_TABLE (charpos
);
6090 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6093 /* Case 2: S2 is neither Sword nor Ssymbol. */
6094 if (s2
!= Sword
&& s2
!= Ssymbol
)
6097 /* Case 3: D is not at the beginning of string ... */
6098 if (!AT_STRINGS_BEG (d
))
6100 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6102 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6106 /* ... and S1 is Sword or Ssymbol. */
6107 if (s1
== Sword
|| s1
== Ssymbol
)
6114 DEBUG_PRINT1 ("EXECUTING symend.\n");
6116 /* We FAIL in one of the following cases: */
6118 /* Case 1: D is at the beginning of string. */
6119 if (AT_STRINGS_BEG (d
))
6123 /* C1 is the character before D, S1 is the syntax of C1, C2
6124 is the character at D, and S2 is the syntax of C2. */
6128 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6129 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6130 UPDATE_SYNTAX_TABLE (charpos
);
6132 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6135 /* Case 2: S1 is neither Ssymbol nor Sword. */
6136 if (s1
!= Sword
&& s1
!= Ssymbol
)
6139 /* Case 3: D is not at the end of string ... */
6140 if (!AT_STRINGS_END (d
))
6142 PREFETCH_NOLIMIT ();
6143 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6145 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
6149 /* ... and S2 is Sword or Ssymbol. */
6150 if (s2
== Sword
|| s2
== Ssymbol
)
6159 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6161 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt
);
6165 ssize_t offset
= PTR_TO_OFFSET (d
);
6166 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6167 UPDATE_SYNTAX_TABLE (pos1
);
6174 GET_CHAR_AFTER (c
, d
, len
);
6175 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6184 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6185 if (PTR_BYTE_POS (d
) >= PT_BYTE
)
6190 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6191 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6196 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6197 if (PTR_BYTE_POS (d
) <= PT_BYTE
)
6202 case notcategoryspec
:
6204 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6206 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n",
6207 not?"not":"", mcnt
);
6213 GET_CHAR_AFTER (c
, d
, len
);
6214 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6226 continue; /* Successfully executed one pattern command; keep going. */
6229 /* We goto here if a matching operation fails. */
6231 IMMEDIATE_QUIT_CHECK
;
6232 if (!FAIL_STACK_EMPTY ())
6235 /* A restart point is known. Restore to that state. */
6236 DEBUG_PRINT1 ("\nFAIL:\n");
6237 POP_FAILURE_POINT (str
, pat
);
6238 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *pat
++))
6240 case on_failure_keep_string_jump
:
6241 assert (str
== NULL
);
6242 goto continue_failure_jump
;
6244 case on_failure_jump_nastyloop
:
6245 assert ((re_opcode_t
)pat
[-2] == no_op
);
6246 PUSH_FAILURE_POINT (pat
- 2, str
);
6249 case on_failure_jump_loop
:
6250 case on_failure_jump
:
6253 continue_failure_jump
:
6254 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6259 /* A special frame used for nastyloops. */
6266 assert (p
>= bufp
->buffer
&& p
<= pend
);
6268 if (d
>= string1
&& d
<= end1
)
6272 break; /* Matching at this starting point really fails. */
6276 goto restore_best_regs
;
6280 return -1; /* Failure to match. */
6283 /* Subroutine definitions for re_match_2. */
6285 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6286 bytes; nonzero otherwise. */
6289 bcmp_translate (const re_char
*s1
, const re_char
*s2
, register ssize_t len
,
6290 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6292 register re_char
*p1
= s1
, *p2
= s2
;
6293 re_char
*p1_end
= s1
+ len
;
6294 re_char
*p2_end
= s2
+ len
;
6296 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6297 different lengths, but relying on a single `len' would break this. -sm */
6298 while (p1
< p1_end
&& p2
< p2_end
)
6300 int p1_charlen
, p2_charlen
;
6301 re_wchar_t p1_ch
, p2_ch
;
6303 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6304 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6306 if (RE_TRANSLATE (translate
, p1_ch
)
6307 != RE_TRANSLATE (translate
, p2_ch
))
6310 p1
+= p1_charlen
, p2
+= p2_charlen
;
6313 if (p1
!= p1_end
|| p2
!= p2_end
)
6319 /* Entry points for GNU code. */
6321 /* re_compile_pattern is the GNU regular expression compiler: it
6322 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6323 Returns 0 if the pattern was valid, otherwise an error string.
6325 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6326 are set in BUFP on entry.
6328 We call regex_compile to do the actual compilation. */
6331 re_compile_pattern (const char *pattern
, size_t length
,
6332 struct re_pattern_buffer
*bufp
)
6336 /* GNU code is written to assume at least RE_NREGS registers will be set
6337 (and at least one extra will be -1). */
6338 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6340 /* And GNU code determines whether or not to get register information
6341 by passing null for the REGS argument to re_match, etc., not by
6345 ret
= regex_compile ((re_char
*) pattern
, length
, re_syntax_options
, bufp
);
6349 return gettext (re_error_msgid
[(int) ret
]);
6351 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6353 /* Entry points compatible with 4.2 BSD regex library. We don't define
6354 them unless specifically requested. */
6356 #if defined _REGEX_RE_COMP || defined _LIBC
6358 /* BSD has one and only one pattern buffer. */
6359 static struct re_pattern_buffer re_comp_buf
;
6363 /* Make these definitions weak in libc, so POSIX programs can redefine
6364 these names if they don't use our functions, and still use
6365 regcomp/regexec below without link errors. */
6368 re_comp (const char *s
)
6374 if (!re_comp_buf
.buffer
)
6375 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6376 return (char *) gettext ("No previous regular expression");
6380 if (!re_comp_buf
.buffer
)
6382 re_comp_buf
.buffer
= malloc (200);
6383 if (re_comp_buf
.buffer
== NULL
)
6384 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6385 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6386 re_comp_buf
.allocated
= 200;
6388 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6389 if (re_comp_buf
.fastmap
== NULL
)
6390 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6391 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6394 /* Since `re_exec' always passes NULL for the `regs' argument, we
6395 don't need to initialize the pattern buffer fields which affect it. */
6397 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6402 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6403 return (char *) gettext (re_error_msgid
[(int) ret
]);
6411 re_exec (const char *s
)
6413 const size_t len
= strlen (s
);
6415 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6417 #endif /* _REGEX_RE_COMP */
6419 /* POSIX.2 functions. Don't define these for Emacs. */
6423 /* regcomp takes a regular expression as a string and compiles it.
6425 PREG is a regex_t *. We do not expect any fields to be initialized,
6426 since POSIX says we shouldn't. Thus, we set
6428 `buffer' to the compiled pattern;
6429 `used' to the length of the compiled pattern;
6430 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6431 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6432 RE_SYNTAX_POSIX_BASIC;
6433 `fastmap' to an allocated space for the fastmap;
6434 `fastmap_accurate' to zero;
6435 `re_nsub' to the number of subexpressions in PATTERN.
6437 PATTERN is the address of the pattern string.
6439 CFLAGS is a series of bits which affect compilation.
6441 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6442 use POSIX basic syntax.
6444 If REG_NEWLINE is set, then . and [^...] don't match newline.
6445 Also, regexec will try a match beginning after every newline.
6447 If REG_ICASE is set, then we considers upper- and lowercase
6448 versions of letters to be equivalent when matching.
6450 If REG_NOSUB is set, then when PREG is passed to regexec, that
6451 routine will report only success or failure, and nothing about the
6454 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6455 the return codes and their meanings.) */
6458 regcomp (regex_t
*__restrict preg
, const char *__restrict pattern
,
6463 = (cflags
& REG_EXTENDED
) ?
6464 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6466 /* regex_compile will allocate the space for the compiled pattern. */
6468 preg
->allocated
= 0;
6471 /* Try to allocate space for the fastmap. */
6472 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6474 if (cflags
& REG_ICASE
)
6478 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6479 if (preg
->translate
== NULL
)
6480 return (int) REG_ESPACE
;
6482 /* Map uppercase characters to corresponding lowercase ones. */
6483 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6484 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6487 preg
->translate
= NULL
;
6489 /* If REG_NEWLINE is set, newlines are treated differently. */
6490 if (cflags
& REG_NEWLINE
)
6491 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6492 syntax
&= ~RE_DOT_NEWLINE
;
6493 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6496 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6498 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6500 /* POSIX says a null character in the pattern terminates it, so we
6501 can use strlen here in compiling the pattern. */
6502 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6504 /* POSIX doesn't distinguish between an unmatched open-group and an
6505 unmatched close-group: both are REG_EPAREN. */
6506 if (ret
== REG_ERPAREN
)
6509 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6510 { /* Compute the fastmap now, since regexec cannot modify the pattern
6512 re_compile_fastmap (preg
);
6513 if (preg
->can_be_null
)
6514 { /* The fastmap can't be used anyway. */
6515 free (preg
->fastmap
);
6516 preg
->fastmap
= NULL
;
6521 WEAK_ALIAS (__regcomp
, regcomp
)
6524 /* regexec searches for a given pattern, specified by PREG, in the
6527 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6528 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6529 least NMATCH elements, and we set them to the offsets of the
6530 corresponding matched substrings.
6532 EFLAGS specifies `execution flags' which affect matching: if
6533 REG_NOTBOL is set, then ^ does not match at the beginning of the
6534 string; if REG_NOTEOL is set, then $ does not match at the end.
6536 We return 0 if we find a match and REG_NOMATCH if not. */
6539 regexec (const regex_t
*__restrict preg
, const char *__restrict string
,
6540 size_t nmatch
, regmatch_t pmatch
[__restrict_arr
], int eflags
)
6543 struct re_registers regs
;
6544 regex_t private_preg
;
6545 size_t len
= strlen (string
);
6546 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6548 private_preg
= *preg
;
6550 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6551 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6553 /* The user has told us exactly how many registers to return
6554 information about, via `nmatch'. We have to pass that on to the
6555 matching routines. */
6556 private_preg
.regs_allocated
= REGS_FIXED
;
6560 regs
.num_regs
= nmatch
;
6561 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6562 if (regs
.start
== NULL
)
6564 regs
.end
= regs
.start
+ nmatch
;
6567 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6568 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6569 was a little bit longer but still only matching the real part.
6570 This works because the `endline' will check for a '\n' and will find a
6571 '\0', correctly deciding that this is not the end of a line.
6572 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6573 a convenient '\0' there. For all we know, the string could be preceded
6574 by '\n' which would throw things off. */
6576 /* Perform the searching operation. */
6577 ret
= re_search (&private_preg
, string
, len
,
6578 /* start: */ 0, /* range: */ len
,
6579 want_reg_info
? ®s
: (struct re_registers
*) 0);
6581 /* Copy the register information to the POSIX structure. */
6588 for (r
= 0; r
< nmatch
; r
++)
6590 pmatch
[r
].rm_so
= regs
.start
[r
];
6591 pmatch
[r
].rm_eo
= regs
.end
[r
];
6595 /* If we needed the temporary register info, free the space now. */
6599 /* We want zero return to mean success, unlike `re_search'. */
6600 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6602 WEAK_ALIAS (__regexec
, regexec
)
6605 /* Returns a message corresponding to an error code, ERR_CODE, returned
6606 from either regcomp or regexec. We don't use PREG here.
6608 ERR_CODE was previously called ERRCODE, but that name causes an
6609 error with msvc8 compiler. */
6612 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6618 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6619 /* Only error codes returned by the rest of the code should be passed
6620 to this routine. If we are given anything else, or if other regex
6621 code generates an invalid error code, then the program has a bug.
6622 Dump core so we can fix it. */
6625 msg
= gettext (re_error_msgid
[err_code
]);
6627 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6629 if (errbuf_size
!= 0)
6631 if (msg_size
> errbuf_size
)
6633 memcpy (errbuf
, msg
, errbuf_size
- 1);
6634 errbuf
[errbuf_size
- 1] = 0;
6637 strcpy (errbuf
, msg
);
6642 WEAK_ALIAS (__regerror
, regerror
)
6645 /* Free dynamically allocated space used by PREG. */
6648 regfree (regex_t
*preg
)
6650 free (preg
->buffer
);
6651 preg
->buffer
= NULL
;
6653 preg
->allocated
= 0;
6656 free (preg
->fastmap
);
6657 preg
->fastmap
= NULL
;
6658 preg
->fastmap_accurate
= 0;
6660 free (preg
->translate
);
6661 preg
->translate
= NULL
;
6663 WEAK_ALIAS (__regfree
, regfree
)
6665 #endif /* not emacs */