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) 1993-2016 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>. */
21 - structure the opcode space into opcode+flag.
22 - merge with glibc's regex.[ch].
23 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
24 need to modify the compiled regexp so that re_match can be reentrant.
25 - get rid of on_failure_jump_smart by doing the optimization in re_comp
26 rather than at run-time, so that re_match can be reentrant.
29 /* AIX requires this to be the first thing in the file. */
30 #if defined _AIX && !defined REGEX_MALLOC
34 /* Ignore some GCC warnings for now. This section should go away
35 once the Emacs and Gnulib regex code is merged. */
36 #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) || defined __clang__
37 # pragma GCC diagnostic ignored "-Wstrict-overflow"
39 # pragma GCC diagnostic ignored "-Wunused-function"
40 # pragma GCC diagnostic ignored "-Wunused-macros"
41 # pragma GCC diagnostic ignored "-Wunused-result"
42 # pragma GCC diagnostic ignored "-Wunused-variable"
46 #if 4 < __GNUC__ + (6 <= __GNUC_MINOR__) && ! defined __clang__
47 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
56 /* We need this for `regex.h', and perhaps for the Emacs include files. */
57 # include <sys/types.h>
60 /* Whether to use ISO C Amendment 1 wide char functions.
61 Those should not be used for Emacs since it uses its own. */
63 #define WIDE_CHAR_SUPPORT 1
65 #define WIDE_CHAR_SUPPORT \
66 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
69 /* For platform which support the ISO C amendment 1 functionality we
70 support user defined character classes. */
72 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
78 /* We have to keep the namespace clean. */
79 # define regfree(preg) __regfree (preg)
80 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
81 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
82 # define regerror(err_code, preg, errbuf, errbuf_size) \
83 __regerror (err_code, preg, errbuf, errbuf_size)
84 # define re_set_registers(bu, re, nu, st, en) \
85 __re_set_registers (bu, re, nu, st, en)
86 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
87 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
88 # define re_match(bufp, string, size, pos, regs) \
89 __re_match (bufp, string, size, pos, regs)
90 # define re_search(bufp, string, size, startpos, range, regs) \
91 __re_search (bufp, string, size, startpos, range, regs)
92 # define re_compile_pattern(pattern, length, bufp) \
93 __re_compile_pattern (pattern, length, bufp)
94 # define re_set_syntax(syntax) __re_set_syntax (syntax)
95 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
96 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
97 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
99 /* Make sure we call libc's function even if the user overrides them. */
100 # define btowc __btowc
101 # define iswctype __iswctype
102 # define wctype __wctype
104 # define WEAK_ALIAS(a,b) weak_alias (a, b)
106 /* We are also using some library internals. */
107 # include <locale/localeinfo.h>
108 # include <locale/elem-hash.h>
109 # include <langinfo.h>
111 # define WEAK_ALIAS(a,b)
114 /* This is for other GNU distributions with internationalized messages. */
115 #if HAVE_LIBINTL_H || defined _LIBC
116 # include <libintl.h>
118 # define gettext(msgid) (msgid)
122 /* This define is so xgettext can find the internationalizable
124 # define gettext_noop(String) String
127 /* The `emacs' switch turns on certain matching commands
128 that make sense only in Emacs. */
132 # include "character.h"
136 # include "category.h"
138 /* Make syntax table lookup grant data in gl_state. */
139 # define SYNTAX(c) syntax_property (c, 1)
144 # define malloc xmalloc
148 # define realloc xrealloc
154 /* Converts the pointer to the char to BEG-based offset from the start. */
155 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
156 /* Strings are 0-indexed, buffers are 1-indexed; we pun on the boolean
157 result to get the right base index. */
158 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
160 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
161 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
162 # define RE_STRING_CHAR(p, multibyte) \
163 (multibyte ? (STRING_CHAR (p)) : (*(p)))
164 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
165 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
167 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
169 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
171 /* Set C a (possibly converted to multibyte) character before P. P
172 points into a string which is the virtual concatenation of STR1
173 (which ends at END1) or STR2 (which ends at END2). */
174 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
176 if (target_multibyte) \
178 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
179 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
180 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
181 c = STRING_CHAR (dtemp); \
185 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
186 (c) = RE_CHAR_TO_MULTIBYTE (c); \
190 /* Set C a (possibly converted to multibyte) character at P, and set
191 LEN to the byte length of that character. */
192 # define GET_CHAR_AFTER(c, p, len) \
194 if (target_multibyte) \
195 (c) = STRING_CHAR_AND_LENGTH (p, len); \
200 (c) = RE_CHAR_TO_MULTIBYTE (c); \
204 #else /* not emacs */
206 /* If we are not linking with Emacs proper,
207 we can't use the relocating allocator
208 even if config.h says that we can. */
213 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
216 xmalloc (size_t size
)
218 void *val
= malloc (size
);
221 write (STDERR_FILENO
, "virtual memory exhausted\n", 25);
228 xrealloc (void *block
, size_t size
)
231 /* We must call malloc explicitly when BLOCK is 0, since some
232 reallocs don't do this. */
236 val
= realloc (block
, size
);
239 write (STDERR_FILENO
, "virtual memory exhausted\n", 25);
248 # define malloc xmalloc
252 # define realloc xrealloc
254 # include <stdbool.h>
257 /* Define the syntax stuff for \<, \>, etc. */
259 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
260 enum syntaxcode
{ Swhitespace
= 0, Sword
= 1, Ssymbol
= 2 };
262 /* Dummy macros for non-Emacs environments. */
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 BYTES_BY_CHAR_HEAD(p) (1)
268 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
269 # define STRING_CHAR(p) (*(p))
270 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
271 # define CHAR_STRING(c, s) (*(s) = (c), 1)
272 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
273 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
274 # define RE_CHAR_TO_MULTIBYTE(c) (c)
275 # define RE_CHAR_TO_UNIBYTE(c) (c)
276 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
277 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
278 # define GET_CHAR_AFTER(c, p, len) \
280 # define CHAR_BYTE8_P(c) (0)
281 # define CHAR_LEADING_CODE(c) (c)
283 #endif /* not emacs */
286 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
287 # define RE_TRANSLATE_P(TBL) (TBL)
290 /* Get the interface, including the syntax bits. */
293 /* isalpha etc. are used for the character classes. */
298 /* 1 if C is an ASCII character. */
299 # define IS_REAL_ASCII(c) ((c) < 0200)
301 /* 1 if C is a unibyte character. */
302 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
304 /* The Emacs definitions should not be directly affected by locales. */
306 /* In Emacs, these are only used for single-byte characters. */
307 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
308 # define ISCNTRL(c) ((c) < ' ')
309 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
310 || ((c) >= 'a' && (c) <= 'f') \
311 || ((c) >= 'A' && (c) <= 'F'))
313 /* This is only used for single-byte characters. */
314 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
316 /* The rest must handle multibyte characters. */
318 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
319 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0240) \
322 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
323 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
326 # define ISALNUM(c) (IS_REAL_ASCII (c) \
327 ? (((c) >= 'a' && (c) <= 'z') \
328 || ((c) >= 'A' && (c) <= 'Z') \
329 || ((c) >= '0' && (c) <= '9')) \
332 # define ISALPHA(c) (IS_REAL_ASCII (c) \
333 ? (((c) >= 'a' && (c) <= 'z') \
334 || ((c) >= 'A' && (c) <= 'Z')) \
337 # define ISLOWER(c) lowercasep (c)
339 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
340 ? ((c) > ' ' && (c) < 0177 \
341 && !(((c) >= 'a' && (c) <= 'z') \
342 || ((c) >= 'A' && (c) <= 'Z') \
343 || ((c) >= '0' && (c) <= '9'))) \
344 : SYNTAX (c) != Sword)
346 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
348 # define ISUPPER(c) uppercasep (c)
350 # define ISWORD(c) (SYNTAX (c) == Sword)
352 #else /* not emacs */
354 /* 1 if C is an ASCII character. */
355 # define IS_REAL_ASCII(c) ((c) < 0200)
357 /* This distinction is not meaningful, except in Emacs. */
358 # define ISUNIBYTE(c) 1
361 # define ISBLANK(c) isblank (c)
363 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
366 # define ISGRAPH(c) isgraph (c)
368 # define ISGRAPH(c) (isprint (c) && !isspace (c))
371 /* Solaris defines ISPRINT so we must undefine it first. */
373 # define ISPRINT(c) isprint (c)
374 # define ISDIGIT(c) isdigit (c)
375 # define ISALNUM(c) isalnum (c)
376 # define ISALPHA(c) isalpha (c)
377 # define ISCNTRL(c) iscntrl (c)
378 # define ISLOWER(c) islower (c)
379 # define ISPUNCT(c) ispunct (c)
380 # define ISSPACE(c) isspace (c)
381 # define ISUPPER(c) isupper (c)
382 # define ISXDIGIT(c) isxdigit (c)
384 # define ISWORD(c) ISALPHA (c)
387 # define TOLOWER(c) _tolower (c)
389 # define TOLOWER(c) tolower (c)
392 /* How many characters in the character set. */
393 # define CHAR_SET_SIZE 256
397 extern char *re_syntax_table
;
399 # else /* not SYNTAX_TABLE */
401 static char re_syntax_table
[CHAR_SET_SIZE
];
404 init_syntax_once (void)
412 memset (re_syntax_table
, 0, sizeof re_syntax_table
);
414 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
416 re_syntax_table
[c
] = Sword
;
418 re_syntax_table
['_'] = Ssymbol
;
423 # endif /* not SYNTAX_TABLE */
425 # define SYNTAX(c) re_syntax_table[(c)]
427 #endif /* not emacs */
429 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
431 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
432 use `alloca' instead of `malloc'. This is because using malloc in
433 re_search* or re_match* could cause memory leaks when C-g is used in
434 Emacs; also, malloc is slower and causes storage fragmentation. On
435 the other hand, malloc is more portable, and easier to debug.
437 Because we sometimes use alloca, some routines have to be macros,
438 not functions -- `alloca'-allocated space disappears at the end of the
439 function it is called in. */
443 # define REGEX_ALLOCATE malloc
444 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
445 # define REGEX_FREE free
447 #else /* not REGEX_MALLOC */
450 # define REGEX_USE_SAFE_ALLOCA USE_SAFE_ALLOCA
451 # define REGEX_SAFE_FREE() SAFE_FREE ()
452 # define REGEX_ALLOCATE SAFE_ALLOCA
455 # define REGEX_ALLOCATE alloca
458 /* Assumes a `char *destination' variable. */
459 # define REGEX_REALLOCATE(source, osize, nsize) \
460 (destination = REGEX_ALLOCATE (nsize), \
461 memcpy (destination, source, osize))
463 /* No need to do anything to free, after alloca. */
464 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
466 #endif /* not REGEX_MALLOC */
468 #ifndef REGEX_USE_SAFE_ALLOCA
469 # define REGEX_USE_SAFE_ALLOCA ((void) 0)
470 # define REGEX_SAFE_FREE() ((void) 0)
473 /* Define how to allocate the failure stack. */
475 #if defined REL_ALLOC && defined REGEX_MALLOC
477 # define REGEX_ALLOCATE_STACK(size) \
478 r_alloc (&failure_stack_ptr, (size))
479 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
480 r_re_alloc (&failure_stack_ptr, (nsize))
481 # define REGEX_FREE_STACK(ptr) \
482 r_alloc_free (&failure_stack_ptr)
484 #else /* not using relocating allocator */
486 # define REGEX_ALLOCATE_STACK(size) REGEX_ALLOCATE (size)
487 # define REGEX_REALLOCATE_STACK(source, o, n) REGEX_REALLOCATE (source, o, n)
488 # define REGEX_FREE_STACK(ptr) REGEX_FREE (ptr)
490 #endif /* not using relocating allocator */
493 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
494 `string1' or just past its end. This works if PTR is NULL, which is
496 #define FIRST_STRING_P(ptr) \
497 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
499 /* (Re)Allocate N items of type T using malloc, or fail. */
500 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
501 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
502 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
504 #define BYTEWIDTH 8 /* In bits. */
509 # define max(a, b) ((a) > (b) ? (a) : (b))
510 # define min(a, b) ((a) < (b) ? (a) : (b))
513 /* Type of source-pattern and string chars. */
515 typedef unsigned char re_char
;
516 typedef const re_char const_re_char
;
518 typedef const unsigned char re_char
;
519 typedef re_char const_re_char
;
522 typedef char boolean
;
524 static regoff_t
re_match_2_internal (struct re_pattern_buffer
*bufp
,
525 re_char
*string1
, size_t size1
,
526 re_char
*string2
, size_t size2
,
528 struct re_registers
*regs
,
531 /* These are the command codes that appear in compiled regular
532 expressions. Some opcodes are followed by argument bytes. A
533 command code can specify any interpretation whatsoever for its
534 arguments. Zero bytes may appear in the compiled regular expression. */
540 /* Succeed right away--no more backtracking. */
543 /* Followed by one byte giving n, then by n literal bytes. */
546 /* Matches any (more or less) character. */
549 /* Matches any one char belonging to specified set. First
550 following byte is number of bitmap bytes. Then come bytes
551 for a bitmap saying which chars are in. Bits in each byte
552 are ordered low-bit-first. A character is in the set if its
553 bit is 1. A character too large to have a bit in the map is
554 automatically not in the set.
556 If the length byte has the 0x80 bit set, then that stuff
557 is followed by a range table:
558 2 bytes of flags for character sets (low 8 bits, high 8 bits)
559 See RANGE_TABLE_WORK_BITS below.
560 2 bytes, the number of pairs that follow (upto 32767)
561 pairs, each 2 multibyte characters,
562 each multibyte character represented as 3 bytes. */
565 /* Same parameters as charset, but match any character that is
566 not one of those specified. */
569 /* Start remembering the text that is matched, for storing in a
570 register. Followed by one byte with the register number, in
571 the range 0 to one less than the pattern buffer's re_nsub
575 /* Stop remembering the text that is matched and store it in a
576 memory register. Followed by one byte with the register
577 number, in the range 0 to one less than `re_nsub' in the
581 /* Match a duplicate of something remembered. Followed by one
582 byte containing the register number. */
585 /* Fail unless at beginning of line. */
588 /* Fail unless at end of line. */
591 /* Succeeds if at beginning of buffer (if emacs) or at beginning
592 of string to be matched (if not). */
595 /* Analogously, for end of buffer/string. */
598 /* Followed by two byte relative address to which to jump. */
601 /* Followed by two-byte relative address of place to resume at
602 in case of failure. */
605 /* Like on_failure_jump, but pushes a placeholder instead of the
606 current string position when executed. */
607 on_failure_keep_string_jump
,
609 /* Just like `on_failure_jump', except that it checks that we
610 don't get stuck in an infinite loop (matching an empty string
612 on_failure_jump_loop
,
614 /* Just like `on_failure_jump_loop', except that it checks for
615 a different kind of loop (the kind that shows up with non-greedy
616 operators). This operation has to be immediately preceded
618 on_failure_jump_nastyloop
,
620 /* A smart `on_failure_jump' used for greedy * and + operators.
621 It analyzes the loop before which it is put and if the
622 loop does not require backtracking, it changes itself to
623 `on_failure_keep_string_jump' and short-circuits the loop,
624 else it just defaults to changing itself into `on_failure_jump'.
625 It assumes that it is pointing to just past a `jump'. */
626 on_failure_jump_smart
,
628 /* Followed by two-byte relative address and two-byte number n.
629 After matching N times, jump to the address upon failure.
630 Does not work if N starts at 0: use on_failure_jump_loop
634 /* Followed by two-byte relative address, and two-byte number n.
635 Jump to the address N times, then fail. */
638 /* Set the following two-byte relative address to the
639 subsequent two-byte number. The address *includes* the two
643 wordbeg
, /* Succeeds if at word beginning. */
644 wordend
, /* Succeeds if at word end. */
646 wordbound
, /* Succeeds if at a word boundary. */
647 notwordbound
, /* Succeeds if not at a word boundary. */
649 symbeg
, /* Succeeds if at symbol beginning. */
650 symend
, /* Succeeds if at symbol end. */
652 /* Matches any character whose syntax is specified. Followed by
653 a byte which contains a syntax code, e.g., Sword. */
656 /* Matches any character whose syntax is not that specified. */
660 , at_dot
, /* Succeeds if at point. */
662 /* Matches any character whose category-set contains the specified
663 category. The operator is followed by a byte which contains a
664 category code (mnemonic ASCII character). */
667 /* Matches any character whose category-set does not contain the
668 specified category. The operator is followed by a byte which
669 contains the category code (mnemonic ASCII character). */
674 /* Common operations on the compiled pattern. */
676 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
678 #define STORE_NUMBER(destination, number) \
680 (destination)[0] = (number) & 0377; \
681 (destination)[1] = (number) >> 8; \
684 /* Same as STORE_NUMBER, except increment DESTINATION to
685 the byte after where the number is stored. Therefore, DESTINATION
686 must be an lvalue. */
688 #define STORE_NUMBER_AND_INCR(destination, number) \
690 STORE_NUMBER (destination, number); \
691 (destination) += 2; \
694 /* Put into DESTINATION a number stored in two contiguous bytes starting
697 #define EXTRACT_NUMBER(destination, source) \
698 ((destination) = extract_number (source))
701 extract_number (re_char
*source
)
703 unsigned leading_byte
= SIGN_EXTEND_CHAR (source
[1]);
704 return (leading_byte
<< 8) + source
[0];
707 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
708 SOURCE must be an lvalue. */
710 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
711 ((destination) = extract_number_and_incr (&source))
714 extract_number_and_incr (re_char
**source
)
716 int num
= extract_number (*source
);
721 /* Store a multibyte character in three contiguous bytes starting
722 DESTINATION, and increment DESTINATION to the byte after where the
723 character is stored. Therefore, DESTINATION must be an lvalue. */
725 #define STORE_CHARACTER_AND_INCR(destination, character) \
727 (destination)[0] = (character) & 0377; \
728 (destination)[1] = ((character) >> 8) & 0377; \
729 (destination)[2] = (character) >> 16; \
730 (destination) += 3; \
733 /* Put into DESTINATION a character stored in three contiguous bytes
734 starting at SOURCE. */
736 #define EXTRACT_CHARACTER(destination, source) \
738 (destination) = ((source)[0] \
739 | ((source)[1] << 8) \
740 | ((source)[2] << 16)); \
744 /* Macros for charset. */
746 /* Size of bitmap of charset P in bytes. P is a start of charset,
747 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
748 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
750 /* Nonzero if charset P has range table. */
751 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
753 /* Return the address of range table of charset P. But not the start
754 of table itself, but the before where the number of ranges is
755 stored. `2 +' means to skip re_opcode_t and size of bitmap,
756 and the 2 bytes of flags at the start of the range table. */
757 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
760 /* Extract the bit flags that start a range table. */
761 #define CHARSET_RANGE_TABLE_BITS(p) \
762 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
763 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
766 /* Return the address of end of RANGE_TABLE. COUNT is number of
767 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
768 is start of range and end of range. `* 3' is size of each start
770 #define CHARSET_RANGE_TABLE_END(range_table, count) \
771 ((range_table) + (count) * 2 * 3)
773 /* If DEBUG is defined, Regex prints many voluminous messages about what
774 it is doing (if the variable `debug' is nonzero). If linked with the
775 main program in `iregex.c', you can enter patterns and strings
776 interactively. And if linked with the main program in `main.c' and
777 the other test files, you can run the already-written tests. */
781 /* We use standard I/O for debugging. */
784 /* It is useful to test things that ``must'' be true when debugging. */
787 static int debug
= -100000;
789 # define DEBUG_STATEMENT(e) e
790 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
791 # define DEBUG_COMPILES_ARGUMENTS
792 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
793 if (debug > 0) print_partial_compiled_pattern (s, e)
794 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
795 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
798 /* Print the fastmap in human-readable form. */
801 print_fastmap (char *fastmap
)
803 unsigned was_a_range
= 0;
806 while (i
< (1 << BYTEWIDTH
))
812 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
828 /* Print a compiled pattern string in human-readable form, starting at
829 the START pointer into it and ending just before the pointer END. */
832 print_partial_compiled_pattern (re_char
*start
, re_char
*end
)
840 fprintf (stderr
, "(null)\n");
844 /* Loop over pattern commands. */
847 fprintf (stderr
, "%td:\t", p
- start
);
849 switch ((re_opcode_t
) *p
++)
852 fprintf (stderr
, "/no_op");
856 fprintf (stderr
, "/succeed");
861 fprintf (stderr
, "/exactn/%d", mcnt
);
864 fprintf (stderr
, "/%c", *p
++);
870 fprintf (stderr
, "/start_memory/%d", *p
++);
874 fprintf (stderr
, "/stop_memory/%d", *p
++);
878 fprintf (stderr
, "/duplicate/%d", *p
++);
882 fprintf (stderr
, "/anychar");
888 register int c
, last
= -100;
889 register int in_range
= 0;
890 int length
= CHARSET_BITMAP_SIZE (p
- 1);
891 int has_range_table
= CHARSET_RANGE_TABLE_EXISTS_P (p
- 1);
893 fprintf (stderr
, "/charset [%s",
894 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
897 fprintf (stderr
, " !extends past end of pattern! ");
899 for (c
= 0; c
< 256; c
++)
901 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
903 /* Are we starting a range? */
904 if (last
+ 1 == c
&& ! in_range
)
906 fprintf (stderr
, "-");
909 /* Have we broken a range? */
910 else if (last
+ 1 != c
&& in_range
)
912 fprintf (stderr
, "%c", last
);
917 fprintf (stderr
, "%c", c
);
923 fprintf (stderr
, "%c", last
);
925 fprintf (stderr
, "]");
932 fprintf (stderr
, "has-range-table");
934 /* ??? Should print the range table; for now, just skip it. */
935 p
+= 2; /* skip range table bits */
936 EXTRACT_NUMBER_AND_INCR (count
, p
);
937 p
= CHARSET_RANGE_TABLE_END (p
, count
);
943 fprintf (stderr
, "/begline");
947 fprintf (stderr
, "/endline");
950 case on_failure_jump
:
951 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
952 fprintf (stderr
, "/on_failure_jump to %td", p
+ mcnt
- start
);
955 case on_failure_keep_string_jump
:
956 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
957 fprintf (stderr
, "/on_failure_keep_string_jump to %td",
961 case on_failure_jump_nastyloop
:
962 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
963 fprintf (stderr
, "/on_failure_jump_nastyloop to %td",
967 case on_failure_jump_loop
:
968 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
969 fprintf (stderr
, "/on_failure_jump_loop to %td",
973 case on_failure_jump_smart
:
974 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
975 fprintf (stderr
, "/on_failure_jump_smart to %td",
980 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
981 fprintf (stderr
, "/jump to %td", p
+ mcnt
- start
);
985 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
986 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
987 fprintf (stderr
, "/succeed_n to %td, %d times",
988 p
- 2 + mcnt
- start
, mcnt2
);
992 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
993 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
994 fprintf (stderr
, "/jump_n to %td, %d times",
995 p
- 2 + mcnt
- start
, mcnt2
);
999 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
1000 EXTRACT_NUMBER_AND_INCR (mcnt2
, p
);
1001 fprintf (stderr
, "/set_number_at location %td to %d",
1002 p
- 2 + mcnt
- start
, mcnt2
);
1006 fprintf (stderr
, "/wordbound");
1010 fprintf (stderr
, "/notwordbound");
1014 fprintf (stderr
, "/wordbeg");
1018 fprintf (stderr
, "/wordend");
1022 fprintf (stderr
, "/symbeg");
1026 fprintf (stderr
, "/symend");
1030 fprintf (stderr
, "/syntaxspec");
1032 fprintf (stderr
, "/%d", mcnt
);
1036 fprintf (stderr
, "/notsyntaxspec");
1038 fprintf (stderr
, "/%d", mcnt
);
1043 fprintf (stderr
, "/at_dot");
1047 fprintf (stderr
, "/categoryspec");
1049 fprintf (stderr
, "/%d", mcnt
);
1052 case notcategoryspec
:
1053 fprintf (stderr
, "/notcategoryspec");
1055 fprintf (stderr
, "/%d", mcnt
);
1060 fprintf (stderr
, "/begbuf");
1064 fprintf (stderr
, "/endbuf");
1068 fprintf (stderr
, "?%d", *(p
-1));
1071 fprintf (stderr
, "\n");
1074 fprintf (stderr
, "%td:\tend of pattern.\n", p
- start
);
1079 print_compiled_pattern (struct re_pattern_buffer
*bufp
)
1081 re_char
*buffer
= bufp
->buffer
;
1083 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
1084 printf ("%ld bytes used/%ld bytes allocated.\n",
1085 bufp
->used
, bufp
->allocated
);
1087 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1089 printf ("fastmap: ");
1090 print_fastmap (bufp
->fastmap
);
1093 printf ("re_nsub: %zu\t", bufp
->re_nsub
);
1094 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1095 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1096 printf ("no_sub: %d\t", bufp
->no_sub
);
1097 printf ("not_bol: %d\t", bufp
->not_bol
);
1098 printf ("not_eol: %d\t", bufp
->not_eol
);
1100 printf ("syntax: %lx\n", bufp
->syntax
);
1103 /* Perhaps we should print the translate table? */
1108 print_double_string (re_char
*where
, re_char
*string1
, ssize_t size1
,
1109 re_char
*string2
, ssize_t size2
)
1117 if (FIRST_STRING_P (where
))
1119 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1120 putchar (string1
[this_char
]);
1125 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1126 putchar (string2
[this_char
]);
1130 #else /* not DEBUG */
1135 # define DEBUG_STATEMENT(e)
1136 # define DEBUG_PRINT(...)
1137 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1138 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1140 #endif /* not DEBUG */
1144 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1145 also be assigned to arbitrarily: each pattern buffer stores its own
1146 syntax, so it can be changed between regex compilations. */
1147 /* This has no initializer because initialized variables in Emacs
1148 become read-only after dumping. */
1149 reg_syntax_t re_syntax_options
;
1152 /* Specify the precise syntax of regexps for compilation. This provides
1153 for compatibility for various utilities which historically have
1154 different, incompatible syntaxes.
1156 The argument SYNTAX is a bit mask comprised of the various bits
1157 defined in regex.h. We return the old syntax. */
1160 re_set_syntax (reg_syntax_t syntax
)
1162 reg_syntax_t ret
= re_syntax_options
;
1164 re_syntax_options
= syntax
;
1167 WEAK_ALIAS (__re_set_syntax
, re_set_syntax
)
1171 /* This table gives an error message for each of the error codes listed
1172 in regex.h. Obviously the order here has to be same as there.
1173 POSIX doesn't require that we do anything for REG_NOERROR,
1174 but why not be nice? */
1176 static const char *re_error_msgid
[] =
1178 gettext_noop ("Success"), /* REG_NOERROR */
1179 gettext_noop ("No match"), /* REG_NOMATCH */
1180 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1181 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1182 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1183 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1184 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1185 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1186 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1187 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1188 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1189 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1190 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1191 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1192 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1193 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1194 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1195 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1198 /* Avoiding alloca during matching, to placate r_alloc. */
1200 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1201 searching and matching functions should not call alloca. On some
1202 systems, alloca is implemented in terms of malloc, and if we're
1203 using the relocating allocator routines, then malloc could cause a
1204 relocation, which might (if the strings being searched are in the
1205 ralloc heap) shift the data out from underneath the regexp
1208 Here's another reason to avoid allocation: Emacs
1209 processes input from X in a signal handler; processing X input may
1210 call malloc; if input arrives while a matching routine is calling
1211 malloc, then we're scrod. But Emacs can't just block input while
1212 calling matching routines; then we don't notice interrupts when
1213 they come in. So, Emacs blocks input around all regexp calls
1214 except the matching calls, which it leaves unprotected, in the
1215 faith that they will not malloc. */
1217 /* Normally, this is fine. */
1218 #define MATCH_MAY_ALLOCATE
1220 /* The match routines may not allocate if (1) they would do it with malloc
1221 and (2) it's not safe for them to use malloc.
1222 Note that if REL_ALLOC is defined, matching would not use malloc for the
1223 failure stack, but we would still use it for the register vectors;
1224 so REL_ALLOC should not affect this. */
1225 #if defined REGEX_MALLOC && defined emacs
1226 # undef MATCH_MAY_ALLOCATE
1230 /* Failure stack declarations and macros; both re_compile_fastmap and
1231 re_match_2 use a failure stack. These have to be macros because of
1232 REGEX_ALLOCATE_STACK. */
1235 /* Approximate number of failure points for which to initially allocate space
1236 when matching. If this number is exceeded, we allocate more
1237 space, so it is not a hard limit. */
1238 #ifndef INIT_FAILURE_ALLOC
1239 # define INIT_FAILURE_ALLOC 20
1242 /* Roughly the maximum number of failure points on the stack. Would be
1243 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1244 This is a variable only so users of regex can assign to it; we never
1245 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1246 before using it, so it should probably be a byte-count instead. */
1247 # if defined MATCH_MAY_ALLOCATE
1248 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1249 whose default stack limit is 2mb. In order for a larger
1250 value to work reliably, you have to try to make it accord
1251 with the process stack limit. */
1252 size_t re_max_failures
= 40000;
1254 size_t re_max_failures
= 4000;
1257 union fail_stack_elt
1260 /* This should be the biggest `int' that's no bigger than a pointer. */
1264 typedef union fail_stack_elt fail_stack_elt_t
;
1268 fail_stack_elt_t
*stack
;
1270 size_t avail
; /* Offset of next open position. */
1271 size_t frame
; /* Offset of the cur constructed frame. */
1274 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1277 /* Define macros to initialize and free the failure stack.
1278 Do `return -2' if the alloc fails. */
1280 #ifdef MATCH_MAY_ALLOCATE
1281 # define INIT_FAIL_STACK() \
1283 fail_stack.stack = \
1284 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1285 * sizeof (fail_stack_elt_t)); \
1287 if (fail_stack.stack == NULL) \
1290 fail_stack.size = INIT_FAILURE_ALLOC; \
1291 fail_stack.avail = 0; \
1292 fail_stack.frame = 0; \
1295 # define INIT_FAIL_STACK() \
1297 fail_stack.avail = 0; \
1298 fail_stack.frame = 0; \
1301 # define RETALLOC_IF(addr, n, t) \
1302 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1306 /* Double the size of FAIL_STACK, up to a limit
1307 which allows approximately `re_max_failures' items.
1309 Return 1 if succeeds, and 0 if either ran out of memory
1310 allocating space for it or it was already too large.
1312 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1314 /* Factor to increase the failure stack size by
1315 when we increase it.
1316 This used to be 2, but 2 was too wasteful
1317 because the old discarded stacks added up to as much space
1318 were as ultimate, maximum-size stack. */
1319 #define FAIL_STACK_GROWTH_FACTOR 4
1321 #define GROW_FAIL_STACK(fail_stack) \
1322 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1323 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1325 : ((fail_stack).stack \
1326 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1327 (fail_stack).size * sizeof (fail_stack_elt_t), \
1328 min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1329 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1330 * FAIL_STACK_GROWTH_FACTOR))), \
1332 (fail_stack).stack == NULL \
1334 : ((fail_stack).size \
1335 = (min (re_max_failures * TYPICAL_FAILURE_SIZE, \
1336 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1337 * FAIL_STACK_GROWTH_FACTOR)) \
1338 / sizeof (fail_stack_elt_t)), \
1342 /* Push a pointer value onto the failure stack.
1343 Assumes the variable `fail_stack'. Probably should only
1344 be called from within `PUSH_FAILURE_POINT'. */
1345 #define PUSH_FAILURE_POINTER(item) \
1346 fail_stack.stack[fail_stack.avail++].pointer = (item)
1348 /* This pushes an integer-valued item onto the failure stack.
1349 Assumes the variable `fail_stack'. Probably should only
1350 be called from within `PUSH_FAILURE_POINT'. */
1351 #define PUSH_FAILURE_INT(item) \
1352 fail_stack.stack[fail_stack.avail++].integer = (item)
1354 /* These POP... operations complement the PUSH... operations.
1355 All assume that `fail_stack' is nonempty. */
1356 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1357 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1359 /* Individual items aside from the registers. */
1360 #define NUM_NONREG_ITEMS 3
1362 /* Used to examine the stack (to detect infinite loops). */
1363 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1364 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1365 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1366 #define TOP_FAILURE_HANDLE() fail_stack.frame
1369 #define ENSURE_FAIL_STACK(space) \
1370 while (REMAINING_AVAIL_SLOTS <= space) { \
1371 if (!GROW_FAIL_STACK (fail_stack)) \
1373 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1374 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1377 /* Push register NUM onto the stack. */
1378 #define PUSH_FAILURE_REG(num) \
1380 char *destination; \
1382 ENSURE_FAIL_STACK(3); \
1383 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1384 n, regstart[n], regend[n]); \
1385 PUSH_FAILURE_POINTER (regstart[n]); \
1386 PUSH_FAILURE_POINTER (regend[n]); \
1387 PUSH_FAILURE_INT (n); \
1390 /* Change the counter's value to VAL, but make sure that it will
1391 be reset when backtracking. */
1392 #define PUSH_NUMBER(ptr,val) \
1394 char *destination; \
1396 ENSURE_FAIL_STACK(3); \
1397 EXTRACT_NUMBER (c, ptr); \
1398 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1399 PUSH_FAILURE_INT (c); \
1400 PUSH_FAILURE_POINTER (ptr); \
1401 PUSH_FAILURE_INT (-1); \
1402 STORE_NUMBER (ptr, val); \
1405 /* Pop a saved register off the stack. */
1406 #define POP_FAILURE_REG_OR_COUNT() \
1408 long pfreg = POP_FAILURE_INT (); \
1411 /* It's a counter. */ \
1412 /* Here, we discard `const', making re_match non-reentrant. */ \
1413 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1414 pfreg = POP_FAILURE_INT (); \
1415 STORE_NUMBER (ptr, pfreg); \
1416 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1420 regend[pfreg] = POP_FAILURE_POINTER (); \
1421 regstart[pfreg] = POP_FAILURE_POINTER (); \
1422 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1423 pfreg, regstart[pfreg], regend[pfreg]); \
1427 /* Check that we are not stuck in an infinite loop. */
1428 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1430 ssize_t failure = TOP_FAILURE_HANDLE (); \
1431 /* Check for infinite matching loops */ \
1432 while (failure > 0 \
1433 && (FAILURE_STR (failure) == string_place \
1434 || FAILURE_STR (failure) == NULL)) \
1436 assert (FAILURE_PAT (failure) >= bufp->buffer \
1437 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1438 if (FAILURE_PAT (failure) == pat_cur) \
1443 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1444 failure = NEXT_FAILURE_HANDLE(failure); \
1446 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1449 /* Push the information about the state we will need
1450 if we ever fail back to it.
1452 Requires variables fail_stack, regstart, regend and
1453 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1456 Does `return FAILURE_CODE' if runs out of memory. */
1458 #define PUSH_FAILURE_POINT(pattern, string_place) \
1460 char *destination; \
1461 /* Must be int, so when we don't save any registers, the arithmetic \
1462 of 0 + -1 isn't done as unsigned. */ \
1464 DEBUG_STATEMENT (nfailure_points_pushed++); \
1465 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1466 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1467 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1469 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1471 DEBUG_PRINT ("\n"); \
1473 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1474 PUSH_FAILURE_INT (fail_stack.frame); \
1476 DEBUG_PRINT (" Push string %p: \"", string_place); \
1477 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1478 DEBUG_PRINT ("\"\n"); \
1479 PUSH_FAILURE_POINTER (string_place); \
1481 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1482 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1483 PUSH_FAILURE_POINTER (pattern); \
1485 /* Close the frame by moving the frame pointer past it. */ \
1486 fail_stack.frame = fail_stack.avail; \
1489 /* Estimate the size of data pushed by a typical failure stack entry.
1490 An estimate is all we need, because all we use this for
1491 is to choose a limit for how big to make the failure stack. */
1492 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1493 #define TYPICAL_FAILURE_SIZE 20
1495 /* How many items can still be added to the stack without overflowing it. */
1496 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1499 /* Pops what PUSH_FAIL_STACK pushes.
1501 We restore into the parameters, all of which should be lvalues:
1502 STR -- the saved data position.
1503 PAT -- the saved pattern position.
1504 REGSTART, REGEND -- arrays of string positions.
1506 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1507 `pend', `string1', `size1', `string2', and `size2'. */
1509 #define POP_FAILURE_POINT(str, pat) \
1511 assert (!FAIL_STACK_EMPTY ()); \
1513 /* Remove failure points and point to how many regs pushed. */ \
1514 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1515 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1516 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1518 /* Pop the saved registers. */ \
1519 while (fail_stack.frame < fail_stack.avail) \
1520 POP_FAILURE_REG_OR_COUNT (); \
1522 pat = POP_FAILURE_POINTER (); \
1523 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1524 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1526 /* If the saved string location is NULL, it came from an \
1527 on_failure_keep_string_jump opcode, and we want to throw away the \
1528 saved NULL, thus retaining our current position in the string. */ \
1529 str = POP_FAILURE_POINTER (); \
1530 DEBUG_PRINT (" Popping string %p: \"", str); \
1531 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1532 DEBUG_PRINT ("\"\n"); \
1534 fail_stack.frame = POP_FAILURE_INT (); \
1535 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1537 assert (fail_stack.avail >= 0); \
1538 assert (fail_stack.frame <= fail_stack.avail); \
1540 DEBUG_STATEMENT (nfailure_points_popped++); \
1541 } while (0) /* POP_FAILURE_POINT */
1545 /* Registers are set to a sentinel when they haven't yet matched. */
1546 #define REG_UNSET(e) ((e) == NULL)
1548 /* Subroutine declarations and macros for regex_compile. */
1550 static reg_errcode_t
regex_compile (re_char
*pattern
, size_t size
,
1552 bool posix_backtracking
,
1553 const char *whitespace_regexp
,
1555 reg_syntax_t syntax
,
1557 struct re_pattern_buffer
*bufp
);
1558 static void store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
);
1559 static void store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
);
1560 static void insert_op1 (re_opcode_t op
, unsigned char *loc
,
1561 int arg
, unsigned char *end
);
1562 static void insert_op2 (re_opcode_t op
, unsigned char *loc
,
1563 int arg1
, int arg2
, unsigned char *end
);
1564 static boolean
at_begline_loc_p (re_char
*pattern
, re_char
*p
,
1565 reg_syntax_t syntax
);
1566 static boolean
at_endline_loc_p (re_char
*p
, re_char
*pend
,
1567 reg_syntax_t syntax
);
1568 static re_char
*skip_one_char (re_char
*p
);
1569 static int analyze_first (re_char
*p
, re_char
*pend
,
1570 char *fastmap
, const int multibyte
);
1572 /* Fetch the next character in the uncompiled pattern, with no
1574 #define PATFETCH(c) \
1577 if (p == pend) return REG_EEND; \
1578 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1583 /* If `translate' is non-null, return translate[D], else just D. We
1584 cast the subscript to translate because some data is declared as
1585 `char *', to avoid warnings when a string constant is passed. But
1586 when we use a character as a subscript we must make it unsigned. */
1588 # define TRANSLATE(d) \
1589 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1593 /* Macros for outputting the compiled pattern into `buffer'. */
1595 /* If the buffer isn't allocated when it comes in, use this. */
1596 #define INIT_BUF_SIZE 32
1598 /* Make sure we have at least N more bytes of space in buffer. */
1599 #define GET_BUFFER_SPACE(n) \
1600 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1603 /* Make sure we have one more byte of buffer space and then add C to it. */
1604 #define BUF_PUSH(c) \
1606 GET_BUFFER_SPACE (1); \
1607 *b++ = (unsigned char) (c); \
1611 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1612 #define BUF_PUSH_2(c1, c2) \
1614 GET_BUFFER_SPACE (2); \
1615 *b++ = (unsigned char) (c1); \
1616 *b++ = (unsigned char) (c2); \
1620 /* Store a jump with opcode OP at LOC to location TO. We store a
1621 relative address offset by the three bytes the jump itself occupies. */
1622 #define STORE_JUMP(op, loc, to) \
1623 store_op1 (op, loc, (to) - (loc) - 3)
1625 /* Likewise, for a two-argument jump. */
1626 #define STORE_JUMP2(op, loc, to, arg) \
1627 store_op2 (op, loc, (to) - (loc) - 3, arg)
1629 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1630 #define INSERT_JUMP(op, loc, to) \
1631 insert_op1 (op, loc, (to) - (loc) - 3, b)
1633 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1634 #define INSERT_JUMP2(op, loc, to, arg) \
1635 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1638 /* This is not an arbitrary limit: the arguments which represent offsets
1639 into the pattern are two bytes long. So if 2^15 bytes turns out to
1640 be too small, many things would have to change. */
1641 # define MAX_BUF_SIZE (1L << 15)
1643 /* Extend the buffer by twice its current size via realloc and
1644 reset the pointers that pointed into the old block to point to the
1645 correct places in the new one. If extending the buffer results in it
1646 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1647 #define EXTEND_BUFFER() \
1649 unsigned char *old_buffer = bufp->buffer; \
1650 if (bufp->allocated == MAX_BUF_SIZE) \
1652 bufp->allocated <<= 1; \
1653 if (bufp->allocated > MAX_BUF_SIZE) \
1654 bufp->allocated = MAX_BUF_SIZE; \
1655 ptrdiff_t b_off = b - old_buffer; \
1656 ptrdiff_t begalt_off = begalt - old_buffer; \
1657 bool fixup_alt_jump_set = !!fixup_alt_jump; \
1658 bool laststart_set = !!laststart; \
1659 bool pending_exact_set = !!pending_exact; \
1660 ptrdiff_t fixup_alt_jump_off, laststart_off, pending_exact_off; \
1661 if (fixup_alt_jump_set) fixup_alt_jump_off = fixup_alt_jump - old_buffer; \
1662 if (laststart_set) laststart_off = laststart - old_buffer; \
1663 if (pending_exact_set) pending_exact_off = pending_exact - old_buffer; \
1664 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1665 if (bufp->buffer == NULL) \
1666 return REG_ESPACE; \
1667 unsigned char *new_buffer = bufp->buffer; \
1668 b = new_buffer + b_off; \
1669 begalt = new_buffer + begalt_off; \
1670 if (fixup_alt_jump_set) fixup_alt_jump = new_buffer + fixup_alt_jump_off; \
1671 if (laststart_set) laststart = new_buffer + laststart_off; \
1672 if (pending_exact_set) pending_exact = new_buffer + pending_exact_off; \
1676 /* Since we have one byte reserved for the register number argument to
1677 {start,stop}_memory, the maximum number of groups we can report
1678 things about is what fits in that byte. */
1679 #define MAX_REGNUM 255
1681 /* But patterns can have more than `MAX_REGNUM' registers. We just
1682 ignore the excess. */
1683 typedef int regnum_t
;
1686 /* Macros for the compile stack. */
1688 /* Since offsets can go either forwards or backwards, this type needs to
1689 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1690 /* int may be not enough when sizeof(int) == 2. */
1691 typedef long pattern_offset_t
;
1695 pattern_offset_t begalt_offset
;
1696 pattern_offset_t fixup_alt_jump
;
1697 pattern_offset_t laststart_offset
;
1699 } compile_stack_elt_t
;
1704 compile_stack_elt_t
*stack
;
1706 size_t avail
; /* Offset of next open position. */
1707 } compile_stack_type
;
1710 #define INIT_COMPILE_STACK_SIZE 32
1712 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1713 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1715 /* The next available element. */
1716 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1718 /* Explicit quit checking is needed for Emacs, which uses polling to
1719 process input events. */
1721 # define IMMEDIATE_QUIT_CHECK \
1723 if (immediate_quit) QUIT; \
1726 # define IMMEDIATE_QUIT_CHECK ((void)0)
1729 /* Structure to manage work area for range table. */
1730 struct range_table_work_area
1732 int *table
; /* actual work area. */
1733 int allocated
; /* allocated size for work area in bytes. */
1734 int used
; /* actually used size in words. */
1735 int bits
; /* flag to record character classes */
1740 /* Make sure that WORK_AREA can hold more N multibyte characters.
1741 This is used only in set_image_of_range and set_image_of_range_1.
1742 It expects WORK_AREA to be a pointer.
1743 If it can't get the space, it returns from the surrounding function. */
1745 #define EXTEND_RANGE_TABLE(work_area, n) \
1747 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1749 extend_range_table_work_area (&work_area); \
1750 if ((work_area).table == 0) \
1751 return (REG_ESPACE); \
1755 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1756 (work_area).bits |= (bit)
1758 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1759 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1761 EXTEND_RANGE_TABLE ((work_area), 2); \
1762 (work_area).table[(work_area).used++] = (range_start); \
1763 (work_area).table[(work_area).used++] = (range_end); \
1768 /* Free allocated memory for WORK_AREA. */
1769 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1771 if ((work_area).table) \
1772 free ((work_area).table); \
1775 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1776 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1777 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1778 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1780 /* Bits used to implement the multibyte-part of the various character classes
1781 such as [:alnum:] in a charset's range table. The code currently assumes
1782 that only the low 16 bits are used. */
1783 #define BIT_WORD 0x1
1784 #define BIT_LOWER 0x2
1785 #define BIT_PUNCT 0x4
1786 #define BIT_SPACE 0x8
1787 #define BIT_UPPER 0x10
1788 #define BIT_MULTIBYTE 0x20
1789 #define BIT_ALPHA 0x40
1790 #define BIT_ALNUM 0x80
1791 #define BIT_GRAPH 0x100
1792 #define BIT_PRINT 0x200
1795 /* Set the bit for character C in a list. */
1796 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1801 /* Store characters in the range FROM to TO in the bitmap at B (for
1802 ASCII and unibyte characters) and WORK_AREA (for multibyte
1803 characters) while translating them and paying attention to the
1804 continuity of translated characters.
1806 Implementation note: It is better to implement these fairly big
1807 macros by a function, but it's not that easy because macros called
1808 in this macro assume various local variables already declared. */
1810 /* Both FROM and TO are ASCII characters. */
1812 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1816 for (C0 = (FROM); C0 <= (TO); C0++) \
1818 C1 = TRANSLATE (C0); \
1819 if (! ASCII_CHAR_P (C1)) \
1821 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1822 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1825 SET_LIST_BIT (C1); \
1830 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1832 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1834 int C0, C1, C2, I; \
1835 int USED = RANGE_TABLE_WORK_USED (work_area); \
1837 for (C0 = (FROM); C0 <= (TO); C0++) \
1839 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1840 if (CHAR_BYTE8_P (C1)) \
1841 SET_LIST_BIT (C0); \
1844 C2 = TRANSLATE (C1); \
1846 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1848 SET_LIST_BIT (C1); \
1849 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1851 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1852 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1854 if (C2 >= from - 1 && C2 <= to + 1) \
1856 if (C2 == from - 1) \
1857 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1858 else if (C2 == to + 1) \
1859 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1864 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1870 /* Both FROM and TO are multibyte characters. */
1872 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1874 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1876 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1877 for (C0 = (FROM); C0 <= (TO); C0++) \
1879 C1 = TRANSLATE (C0); \
1880 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1881 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1882 SET_LIST_BIT (C2); \
1883 if (C1 >= (FROM) && C1 <= (TO)) \
1885 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1887 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1888 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1890 if (C1 >= from - 1 && C1 <= to + 1) \
1892 if (C1 == from - 1) \
1893 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1894 else if (C1 == to + 1) \
1895 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1900 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1906 /* Get the next unsigned number in the uncompiled pattern. */
1907 #define GET_INTERVAL_COUNT(num) \
1910 FREE_STACK_RETURN (REG_EBRACE); \
1914 while ('0' <= c && c <= '9') \
1918 if (RE_DUP_MAX / 10 - (RE_DUP_MAX % 10 < c - '0') < num) \
1919 FREE_STACK_RETURN (REG_BADBR); \
1920 num = num * 10 + c - '0'; \
1922 FREE_STACK_RETURN (REG_EBRACE); \
1928 #if ! WIDE_CHAR_SUPPORT
1930 /* Parse a character class, i.e. string such as "[:name:]". *strp
1931 points to the string to be parsed and limit is length, in bytes, of
1934 If *strp point to a string that begins with "[:name:]", where name is
1935 a non-empty sequence of lower case letters, *strp will be advanced past the
1936 closing square bracket and RECC_* constant which maps to the name will be
1937 returned. If name is not a valid character class name zero, or RECC_ERROR,
1940 Otherwise, if *strp doesn’t begin with "[:name:]", -1 is returned.
1942 The function can be used on ASCII and multibyte (UTF-8-encoded) strings.
1945 re_wctype_parse (const unsigned char **strp
, unsigned limit
)
1947 const char *beg
= (const char *)*strp
, *it
;
1949 if (limit
< 4 || beg
[0] != '[' || beg
[1] != ':')
1952 beg
+= 2; /* skip opening ‘[:’ */
1953 limit
-= 3; /* opening ‘[:’ and half of closing ‘:]’; --limit handles rest */
1954 for (it
= beg
; it
[0] != ':' || it
[1] != ']'; ++it
)
1958 *strp
= (const unsigned char *)(it
+ 2);
1960 /* Sort tests in the length=five case by frequency the classes to minimize
1961 number of times we fail the comparison. The frequencies of character class
1962 names used in Emacs sources as of 2016-07-27:
1964 $ find \( -name \*.c -o -name \*.el \) -exec grep -h '\[:[a-z]*:]' {} + |
1965 sed 's/]/]\n/g' |grep -o '\[:[a-z]*:]' |sort |uniq -c |sort -nr
1983 If you update this list, consider also updating chain of or’ed conditions
1984 in execute_charset function.
1989 if (!memcmp (beg
, "word", 4)) return RECC_WORD
;
1992 if (!memcmp (beg
, "alnum", 5)) return RECC_ALNUM
;
1993 if (!memcmp (beg
, "alpha", 5)) return RECC_ALPHA
;
1994 if (!memcmp (beg
, "space", 5)) return RECC_SPACE
;
1995 if (!memcmp (beg
, "digit", 5)) return RECC_DIGIT
;
1996 if (!memcmp (beg
, "blank", 5)) return RECC_BLANK
;
1997 if (!memcmp (beg
, "upper", 5)) return RECC_UPPER
;
1998 if (!memcmp (beg
, "lower", 5)) return RECC_LOWER
;
1999 if (!memcmp (beg
, "punct", 5)) return RECC_PUNCT
;
2000 if (!memcmp (beg
, "ascii", 5)) return RECC_ASCII
;
2001 if (!memcmp (beg
, "graph", 5)) return RECC_GRAPH
;
2002 if (!memcmp (beg
, "print", 5)) return RECC_PRINT
;
2003 if (!memcmp (beg
, "cntrl", 5)) return RECC_CNTRL
;
2006 if (!memcmp (beg
, "xdigit", 6)) return RECC_XDIGIT
;
2009 if (!memcmp (beg
, "unibyte", 7)) return RECC_UNIBYTE
;
2012 if (!memcmp (beg
, "nonascii", 8)) return RECC_NONASCII
;
2015 if (!memcmp (beg
, "multibyte", 9)) return RECC_MULTIBYTE
;
2022 /* True if CH is in the char class CC. */
2024 re_iswctype (int ch
, re_wctype_t cc
)
2028 case RECC_ALNUM
: return ISALNUM (ch
) != 0;
2029 case RECC_ALPHA
: return ISALPHA (ch
) != 0;
2030 case RECC_BLANK
: return ISBLANK (ch
) != 0;
2031 case RECC_CNTRL
: return ISCNTRL (ch
) != 0;
2032 case RECC_DIGIT
: return ISDIGIT (ch
) != 0;
2033 case RECC_GRAPH
: return ISGRAPH (ch
) != 0;
2034 case RECC_LOWER
: return ISLOWER (ch
) != 0;
2035 case RECC_PRINT
: return ISPRINT (ch
) != 0;
2036 case RECC_PUNCT
: return ISPUNCT (ch
) != 0;
2037 case RECC_SPACE
: return ISSPACE (ch
) != 0;
2038 case RECC_UPPER
: return ISUPPER (ch
) != 0;
2039 case RECC_XDIGIT
: return ISXDIGIT (ch
) != 0;
2040 case RECC_ASCII
: return IS_REAL_ASCII (ch
) != 0;
2041 case RECC_NONASCII
: return !IS_REAL_ASCII (ch
);
2042 case RECC_UNIBYTE
: return ISUNIBYTE (ch
) != 0;
2043 case RECC_MULTIBYTE
: return !ISUNIBYTE (ch
);
2044 case RECC_WORD
: return ISWORD (ch
) != 0;
2045 case RECC_ERROR
: return false;
2051 /* Return a bit-pattern to use in the range-table bits to match multibyte
2052 chars of class CC. */
2054 re_wctype_to_bit (re_wctype_t cc
)
2059 case RECC_MULTIBYTE
: return BIT_MULTIBYTE
;
2060 case RECC_ALPHA
: return BIT_ALPHA
;
2061 case RECC_ALNUM
: return BIT_ALNUM
;
2062 case RECC_WORD
: return BIT_WORD
;
2063 case RECC_LOWER
: return BIT_LOWER
;
2064 case RECC_UPPER
: return BIT_UPPER
;
2065 case RECC_PUNCT
: return BIT_PUNCT
;
2066 case RECC_SPACE
: return BIT_SPACE
;
2067 case RECC_GRAPH
: return BIT_GRAPH
;
2068 case RECC_PRINT
: return BIT_PRINT
;
2069 case RECC_ASCII
: case RECC_DIGIT
: case RECC_XDIGIT
: case RECC_CNTRL
:
2070 case RECC_BLANK
: case RECC_UNIBYTE
: case RECC_ERROR
: return 0;
2077 /* Filling in the work area of a range. */
2079 /* Actually extend the space in WORK_AREA. */
2082 extend_range_table_work_area (struct range_table_work_area
*work_area
)
2084 work_area
->allocated
+= 16 * sizeof (int);
2085 work_area
->table
= realloc (work_area
->table
, work_area
->allocated
);
2091 /* Carefully find the ranges of codes that are equivalent
2092 under case conversion to the range start..end when passed through
2093 TRANSLATE. Handle the case where non-letters can come in between
2094 two upper-case letters (which happens in Latin-1).
2095 Also handle the case of groups of more than 2 case-equivalent chars.
2097 The basic method is to look at consecutive characters and see
2098 if they can form a run that can be handled as one.
2100 Returns -1 if successful, REG_ESPACE if ran out of space. */
2103 set_image_of_range_1 (struct range_table_work_area
*work_area
,
2104 re_wchar_t start
, re_wchar_t end
,
2105 RE_TRANSLATE_TYPE translate
)
2107 /* `one_case' indicates a character, or a run of characters,
2108 each of which is an isolate (no case-equivalents).
2109 This includes all ASCII non-letters.
2111 `two_case' indicates a character, or a run of characters,
2112 each of which has two case-equivalent forms.
2113 This includes all ASCII letters.
2115 `strange' indicates a character that has more than one
2118 enum case_type
{one_case
, two_case
, strange
};
2120 /* Describe the run that is in progress,
2121 which the next character can try to extend.
2122 If run_type is strange, that means there really is no run.
2123 If run_type is one_case, then run_start...run_end is the run.
2124 If run_type is two_case, then the run is run_start...run_end,
2125 and the case-equivalents end at run_eqv_end. */
2127 enum case_type run_type
= strange
;
2128 int run_start
, run_end
, run_eqv_end
;
2130 Lisp_Object eqv_table
;
2132 if (!RE_TRANSLATE_P (translate
))
2134 EXTEND_RANGE_TABLE (work_area
, 2);
2135 work_area
->table
[work_area
->used
++] = (start
);
2136 work_area
->table
[work_area
->used
++] = (end
);
2140 eqv_table
= XCHAR_TABLE (translate
)->extras
[2];
2142 for (; start
<= end
; start
++)
2144 enum case_type this_type
;
2145 int eqv
= RE_TRANSLATE (eqv_table
, start
);
2146 int minchar
, maxchar
;
2148 /* Classify this character */
2150 this_type
= one_case
;
2151 else if (RE_TRANSLATE (eqv_table
, eqv
) == start
)
2152 this_type
= two_case
;
2154 this_type
= strange
;
2157 minchar
= start
, maxchar
= eqv
;
2159 minchar
= eqv
, maxchar
= start
;
2161 /* Can this character extend the run in progress? */
2162 if (this_type
== strange
|| this_type
!= run_type
2163 || !(minchar
== run_end
+ 1
2164 && (run_type
== two_case
2165 ? maxchar
== run_eqv_end
+ 1 : 1)))
2168 Record each of its equivalent ranges. */
2169 if (run_type
== one_case
)
2171 EXTEND_RANGE_TABLE (work_area
, 2);
2172 work_area
->table
[work_area
->used
++] = run_start
;
2173 work_area
->table
[work_area
->used
++] = run_end
;
2175 else if (run_type
== two_case
)
2177 EXTEND_RANGE_TABLE (work_area
, 4);
2178 work_area
->table
[work_area
->used
++] = run_start
;
2179 work_area
->table
[work_area
->used
++] = run_end
;
2180 work_area
->table
[work_area
->used
++]
2181 = RE_TRANSLATE (eqv_table
, run_start
);
2182 work_area
->table
[work_area
->used
++]
2183 = RE_TRANSLATE (eqv_table
, run_end
);
2188 if (this_type
== strange
)
2190 /* For a strange character, add each of its equivalents, one
2191 by one. Don't start a range. */
2194 EXTEND_RANGE_TABLE (work_area
, 2);
2195 work_area
->table
[work_area
->used
++] = eqv
;
2196 work_area
->table
[work_area
->used
++] = eqv
;
2197 eqv
= RE_TRANSLATE (eqv_table
, eqv
);
2199 while (eqv
!= start
);
2202 /* Add this char to the run, or start a new run. */
2203 else if (run_type
== strange
)
2205 /* Initialize a new range. */
2206 run_type
= this_type
;
2209 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2213 /* Extend a running range. */
2215 run_eqv_end
= RE_TRANSLATE (eqv_table
, run_end
);
2219 /* If a run is still in progress at the end, finish it now
2220 by recording its equivalent ranges. */
2221 if (run_type
== one_case
)
2223 EXTEND_RANGE_TABLE (work_area
, 2);
2224 work_area
->table
[work_area
->used
++] = run_start
;
2225 work_area
->table
[work_area
->used
++] = run_end
;
2227 else if (run_type
== two_case
)
2229 EXTEND_RANGE_TABLE (work_area
, 4);
2230 work_area
->table
[work_area
->used
++] = run_start
;
2231 work_area
->table
[work_area
->used
++] = run_end
;
2232 work_area
->table
[work_area
->used
++]
2233 = RE_TRANSLATE (eqv_table
, run_start
);
2234 work_area
->table
[work_area
->used
++]
2235 = RE_TRANSLATE (eqv_table
, run_end
);
2243 /* Record the image of the range start..end when passed through
2244 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2245 and is not even necessarily contiguous.
2246 Normally we approximate it with the smallest contiguous range that contains
2247 all the chars we need. However, for Latin-1 we go to extra effort
2250 This function is not called for ASCII ranges.
2252 Returns -1 if successful, REG_ESPACE if ran out of space. */
2255 set_image_of_range (struct range_table_work_area
*work_area
,
2256 re_wchar_t start
, re_wchar_t end
,
2257 RE_TRANSLATE_TYPE translate
)
2259 re_wchar_t cmin
, cmax
;
2262 /* For Latin-1 ranges, use set_image_of_range_1
2263 to get proper handling of ranges that include letters and nonletters.
2264 For a range that includes the whole of Latin-1, this is not necessary.
2265 For other character sets, we don't bother to get this right. */
2266 if (RE_TRANSLATE_P (translate
) && start
< 04400
2267 && !(start
< 04200 && end
>= 04377))
2274 tem
= set_image_of_range_1 (work_area
, start
, newend
, translate
);
2284 EXTEND_RANGE_TABLE (work_area
, 2);
2285 work_area
->table
[work_area
->used
++] = (start
);
2286 work_area
->table
[work_area
->used
++] = (end
);
2288 cmin
= -1, cmax
= -1;
2290 if (RE_TRANSLATE_P (translate
))
2294 for (ch
= start
; ch
<= end
; ch
++)
2296 re_wchar_t c
= TRANSLATE (ch
);
2297 if (! (start
<= c
&& c
<= end
))
2303 cmin
= min (cmin
, c
);
2304 cmax
= max (cmax
, c
);
2311 EXTEND_RANGE_TABLE (work_area
, 2);
2312 work_area
->table
[work_area
->used
++] = (cmin
);
2313 work_area
->table
[work_area
->used
++] = (cmax
);
2321 #ifndef MATCH_MAY_ALLOCATE
2323 /* If we cannot allocate large objects within re_match_2_internal,
2324 we make the fail stack and register vectors global.
2325 The fail stack, we grow to the maximum size when a regexp
2327 The register vectors, we adjust in size each time we
2328 compile a regexp, according to the number of registers it needs. */
2330 static fail_stack_type fail_stack
;
2332 /* Size with which the following vectors are currently allocated.
2333 That is so we can make them bigger as needed,
2334 but never make them smaller. */
2335 static int regs_allocated_size
;
2337 static re_char
** regstart
, ** regend
;
2338 static re_char
**best_regstart
, **best_regend
;
2340 /* Make the register vectors big enough for NUM_REGS registers,
2341 but don't make them smaller. */
2344 regex_grow_registers (int num_regs
)
2346 if (num_regs
> regs_allocated_size
)
2348 RETALLOC_IF (regstart
, num_regs
, re_char
*);
2349 RETALLOC_IF (regend
, num_regs
, re_char
*);
2350 RETALLOC_IF (best_regstart
, num_regs
, re_char
*);
2351 RETALLOC_IF (best_regend
, num_regs
, re_char
*);
2353 regs_allocated_size
= num_regs
;
2357 #endif /* not MATCH_MAY_ALLOCATE */
2359 static boolean
group_in_compile_stack (compile_stack_type compile_stack
,
2362 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2363 Returns one of error codes defined in `regex.h', or zero for success.
2365 If WHITESPACE_REGEXP is given (only #ifdef emacs), it is used instead of
2366 a space character in PATTERN.
2368 Assumes the `allocated' (and perhaps `buffer') and `translate'
2369 fields are set in BUFP on entry.
2371 If it succeeds, results are put in BUFP (if it returns an error, the
2372 contents of BUFP are undefined):
2373 `buffer' is the compiled pattern;
2374 `syntax' is set to SYNTAX;
2375 `used' is set to the length of the compiled pattern;
2376 `fastmap_accurate' is zero;
2377 `re_nsub' is the number of subexpressions in PATTERN;
2378 `not_bol' and `not_eol' are zero;
2380 The `fastmap' field is neither examined nor set. */
2382 /* Insert the `jump' from the end of last alternative to "here".
2383 The space for the jump has already been allocated. */
2384 #define FIXUP_ALT_JUMP() \
2386 if (fixup_alt_jump) \
2387 STORE_JUMP (jump, fixup_alt_jump, b); \
2391 /* Return, freeing storage we allocated. */
2392 #define FREE_STACK_RETURN(value) \
2394 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2395 free (compile_stack.stack); \
2399 static reg_errcode_t
2400 regex_compile (const_re_char
*pattern
, size_t size
,
2402 # define syntax RE_SYNTAX_EMACS
2403 bool posix_backtracking
,
2404 const char *whitespace_regexp
,
2406 reg_syntax_t syntax
,
2407 # define posix_backtracking (!(syntax & RE_NO_POSIX_BACKTRACKING))
2409 struct re_pattern_buffer
*bufp
)
2411 /* We fetch characters from PATTERN here. */
2412 register re_wchar_t c
, c1
;
2414 /* Points to the end of the buffer, where we should append. */
2415 register unsigned char *b
;
2417 /* Keeps track of unclosed groups. */
2418 compile_stack_type compile_stack
;
2420 /* Points to the current (ending) position in the pattern. */
2422 /* `const' makes AIX compiler fail. */
2423 unsigned char *p
= pattern
;
2425 re_char
*p
= pattern
;
2427 re_char
*pend
= pattern
+ size
;
2429 /* How to translate the characters in the pattern. */
2430 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2432 /* Address of the count-byte of the most recently inserted `exactn'
2433 command. This makes it possible to tell if a new exact-match
2434 character can be added to that command or if the character requires
2435 a new `exactn' command. */
2436 unsigned char *pending_exact
= 0;
2438 /* Address of start of the most recently finished expression.
2439 This tells, e.g., postfix * where to find the start of its
2440 operand. Reset at the beginning of groups and alternatives. */
2441 unsigned char *laststart
= 0;
2443 /* Address of beginning of regexp, or inside of last group. */
2444 unsigned char *begalt
;
2446 /* Place in the uncompiled pattern (i.e., the {) to
2447 which to go back if the interval is invalid. */
2448 re_char
*beg_interval
;
2450 /* Address of the place where a forward jump should go to the end of
2451 the containing expression. Each alternative of an `or' -- except the
2452 last -- ends with a forward jump of this sort. */
2453 unsigned char *fixup_alt_jump
= 0;
2455 /* Work area for range table of charset. */
2456 struct range_table_work_area range_table_work
;
2458 /* If the object matched can contain multibyte characters. */
2459 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
2462 /* Nonzero if we have pushed down into a subpattern. */
2463 int in_subpattern
= 0;
2465 /* These hold the values of p, pattern, and pend from the main
2466 pattern when we have pushed into a subpattern. */
2468 re_char
*main_pattern
;
2474 DEBUG_PRINT ("\nCompiling pattern: ");
2477 unsigned debug_count
;
2479 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2480 putchar (pattern
[debug_count
]);
2485 /* Initialize the compile stack. */
2486 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2487 if (compile_stack
.stack
== NULL
)
2490 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2491 compile_stack
.avail
= 0;
2493 range_table_work
.table
= 0;
2494 range_table_work
.allocated
= 0;
2496 /* Initialize the pattern buffer. */
2498 bufp
->syntax
= syntax
;
2500 bufp
->fastmap_accurate
= 0;
2501 bufp
->not_bol
= bufp
->not_eol
= 0;
2502 bufp
->used_syntax
= 0;
2504 /* Set `used' to zero, so that if we return an error, the pattern
2505 printer (for debugging) will think there's no pattern. We reset it
2509 /* Always count groups, whether or not bufp->no_sub is set. */
2512 #if !defined emacs && !defined SYNTAX_TABLE
2513 /* Initialize the syntax table. */
2514 init_syntax_once ();
2517 if (bufp
->allocated
== 0)
2520 { /* If zero allocated, but buffer is non-null, try to realloc
2521 enough space. This loses if buffer's address is bogus, but
2522 that is the user's responsibility. */
2523 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2526 { /* Caller did not allocate a buffer. Do it for them. */
2527 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2529 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2531 bufp
->allocated
= INIT_BUF_SIZE
;
2534 begalt
= b
= bufp
->buffer
;
2536 /* Loop through the uncompiled pattern until we're at the end. */
2542 /* If this is the end of an included regexp,
2543 pop back to the main regexp and try again. */
2547 pattern
= main_pattern
;
2553 /* If this is the end of the main regexp, we are done. */
2566 /* If there's no special whitespace regexp, treat
2567 spaces normally. And don't try to do this recursively. */
2568 if (!whitespace_regexp
|| in_subpattern
)
2571 /* Peek past following spaces. */
2578 /* If the spaces are followed by a repetition op,
2579 treat them normally. */
2581 && (*p1
== '*' || *p1
== '+' || *p1
== '?'
2582 || (*p1
== '\\' && p1
+ 1 != pend
&& p1
[1] == '{')))
2585 /* Replace the spaces with the whitespace regexp. */
2589 main_pattern
= pattern
;
2590 p
= pattern
= (re_char
*) whitespace_regexp
;
2591 pend
= p
+ strlen (whitespace_regexp
);
2598 if ( /* If at start of pattern, it's an operator. */
2600 /* If context independent, it's an operator. */
2601 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2602 /* Otherwise, depends on what's come before. */
2603 || at_begline_loc_p (pattern
, p
, syntax
))
2604 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? begbuf
: begline
);
2613 if ( /* If at end of pattern, it's an operator. */
2615 /* If context independent, it's an operator. */
2616 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2617 /* Otherwise, depends on what's next. */
2618 || at_endline_loc_p (p
, pend
, syntax
))
2619 BUF_PUSH ((syntax
& RE_NO_NEWLINE_ANCHOR
) ? endbuf
: endline
);
2628 if ((syntax
& RE_BK_PLUS_QM
)
2629 || (syntax
& RE_LIMITED_OPS
))
2633 /* If there is no previous pattern... */
2636 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2637 FREE_STACK_RETURN (REG_BADRPT
);
2638 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2643 /* 1 means zero (many) matches is allowed. */
2644 boolean zero_times_ok
= 0, many_times_ok
= 0;
2647 /* If there is a sequence of repetition chars, collapse it
2648 down to just one (the right one). We can't combine
2649 interval operators with these because of, e.g., `a{2}*',
2650 which should only match an even number of `a's. */
2654 if ((syntax
& RE_FRUGAL
)
2655 && c
== '?' && (zero_times_ok
|| many_times_ok
))
2659 zero_times_ok
|= c
!= '+';
2660 many_times_ok
|= c
!= '?';
2666 || (!(syntax
& RE_BK_PLUS_QM
)
2667 && (*p
== '+' || *p
== '?')))
2669 else if (syntax
& RE_BK_PLUS_QM
&& *p
== '\\')
2672 FREE_STACK_RETURN (REG_EESCAPE
);
2673 if (p
[1] == '+' || p
[1] == '?')
2674 PATFETCH (c
); /* Gobble up the backslash. */
2680 /* If we get here, we found another repeat character. */
2684 /* Star, etc. applied to an empty pattern is equivalent
2685 to an empty pattern. */
2686 if (!laststart
|| laststart
== b
)
2689 /* Now we know whether or not zero matches is allowed
2690 and also whether or not two or more matches is allowed. */
2695 boolean simple
= skip_one_char (laststart
) == b
;
2696 size_t startoffset
= 0;
2698 /* Check if the loop can match the empty string. */
2699 (simple
|| !analyze_first (laststart
, b
, NULL
, 0))
2700 ? on_failure_jump
: on_failure_jump_loop
;
2701 assert (skip_one_char (laststart
) <= b
);
2703 if (!zero_times_ok
&& simple
)
2704 { /* Since simple * loops can be made faster by using
2705 on_failure_keep_string_jump, we turn simple P+
2706 into PP* if P is simple. */
2707 unsigned char *p1
, *p2
;
2708 startoffset
= b
- laststart
;
2709 GET_BUFFER_SPACE (startoffset
);
2710 p1
= b
; p2
= laststart
;
2716 GET_BUFFER_SPACE (6);
2719 STORE_JUMP (ofj
, b
, b
+ 6);
2721 /* Simple * loops can use on_failure_keep_string_jump
2722 depending on what follows. But since we don't know
2723 that yet, we leave the decision up to
2724 on_failure_jump_smart. */
2725 INSERT_JUMP (simple
? on_failure_jump_smart
: ofj
,
2726 laststart
+ startoffset
, b
+ 6);
2728 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
2733 /* A simple ? pattern. */
2734 assert (zero_times_ok
);
2735 GET_BUFFER_SPACE (3);
2736 INSERT_JUMP (on_failure_jump
, laststart
, b
+ 3);
2740 else /* not greedy */
2741 { /* I wish the greedy and non-greedy cases could be merged. */
2743 GET_BUFFER_SPACE (7); /* We might use less. */
2746 boolean emptyp
= analyze_first (laststart
, b
, NULL
, 0);
2748 /* The non-greedy multiple match looks like
2749 a repeat..until: we only need a conditional jump
2750 at the end of the loop. */
2751 if (emptyp
) BUF_PUSH (no_op
);
2752 STORE_JUMP (emptyp
? on_failure_jump_nastyloop
2753 : on_failure_jump
, b
, laststart
);
2757 /* The repeat...until naturally matches one or more.
2758 To also match zero times, we need to first jump to
2759 the end of the loop (its conditional jump). */
2760 INSERT_JUMP (jump
, laststart
, b
);
2766 /* non-greedy a?? */
2767 INSERT_JUMP (jump
, laststart
, b
+ 3);
2769 INSERT_JUMP (on_failure_jump
, laststart
, laststart
+ 6);
2788 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2790 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2792 /* Ensure that we have enough space to push a charset: the
2793 opcode, the length count, and the bitset; 34 bytes in all. */
2794 GET_BUFFER_SPACE (34);
2798 /* We test `*p == '^' twice, instead of using an if
2799 statement, so we only need one BUF_PUSH. */
2800 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2804 /* Remember the first position in the bracket expression. */
2807 /* Push the number of bytes in the bitmap. */
2808 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2810 /* Clear the whole map. */
2811 memset (b
, 0, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2813 /* charset_not matches newline according to a syntax bit. */
2814 if ((re_opcode_t
) b
[-2] == charset_not
2815 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2816 SET_LIST_BIT ('\n');
2818 /* Read in characters and ranges, setting map bits. */
2821 boolean escaped_char
= false;
2822 const unsigned char *p2
= p
;
2826 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2828 /* See if we're at the beginning of a possible character
2830 if (syntax
& RE_CHAR_CLASSES
&&
2831 (cc
= re_wctype_parse(&p
, pend
- p
)) != -1)
2834 FREE_STACK_RETURN (REG_ECTYPE
);
2837 FREE_STACK_RETURN (REG_EBRACK
);
2840 for (ch
= 0; ch
< (1 << BYTEWIDTH
); ++ch
)
2841 if (re_iswctype (btowc (ch
), cc
))
2844 if (c
< (1 << BYTEWIDTH
))
2848 /* Most character classes in a multibyte match just set
2849 a flag. Exceptions are is_blank, is_digit, is_cntrl, and
2850 is_xdigit, since they can only match ASCII characters.
2851 We don't need to handle them for multibyte. */
2853 /* Setup the gl_state object to its buffer-defined value.
2854 This hardcodes the buffer-global syntax-table for ASCII
2855 chars, while the other chars will obey syntax-table
2856 properties. It's not ideal, but it's the way it's been
2858 SETUP_BUFFER_SYNTAX_TABLE ();
2860 for (c
= 0; c
< 0x80; ++c
)
2861 if (re_iswctype (c
, cc
))
2867 if (ASCII_CHAR_P (c1
))
2869 else if ((c1
= RE_CHAR_TO_UNIBYTE (c1
)) >= 0)
2872 SET_RANGE_TABLE_WORK_AREA_BIT
2873 (range_table_work
, re_wctype_to_bit (cc
));
2875 /* In most cases the matching rule for char classes only
2876 uses the syntax table for multibyte chars, so that the
2877 content of the syntax-table is not hardcoded in the
2878 range_table. SPACE and WORD are the two exceptions. */
2879 if ((1 << cc
) & ((1 << RECC_SPACE
) | (1 << RECC_WORD
)))
2880 bufp
->used_syntax
= 1;
2882 /* Repeat the loop. */
2886 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2887 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2888 So the translation is done later in a loop. Example:
2889 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2892 /* \ might escape characters inside [...] and [^...]. */
2893 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2895 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2898 escaped_char
= true;
2902 /* Could be the end of the bracket expression. If it's
2903 not (i.e., when the bracket expression is `[]' so
2904 far), the ']' character bit gets set way below. */
2905 if (c
== ']' && p2
!= p1
)
2909 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2912 /* Discard the `-'. */
2915 /* Fetch the character which ends the range. */
2918 if (CHAR_BYTE8_P (c1
)
2919 && ! ASCII_CHAR_P (c
) && ! CHAR_BYTE8_P (c
))
2920 /* Treat the range from a multibyte character to
2921 raw-byte character as empty. */
2926 /* Range from C to C. */
2931 if (syntax
& RE_NO_EMPTY_RANGES
)
2932 FREE_STACK_RETURN (REG_ERANGEX
);
2933 /* Else, repeat the loop. */
2938 /* Set the range into bitmap */
2939 for (; c
<= c1
; c
++)
2942 if (ch
< (1 << BYTEWIDTH
))
2949 SETUP_ASCII_RANGE (range_table_work
, c
, ch
);
2951 if (CHAR_BYTE8_P (c1
))
2952 c
= BYTE8_TO_CHAR (128);
2956 if (CHAR_BYTE8_P (c
))
2958 c
= CHAR_TO_BYTE8 (c
);
2959 c1
= CHAR_TO_BYTE8 (c1
);
2960 for (; c
<= c1
; c
++)
2965 SETUP_MULTIBYTE_RANGE (range_table_work
, c
, c1
);
2969 SETUP_UNIBYTE_RANGE (range_table_work
, c
, c1
);
2976 /* Discard any (non)matching list bytes that are all 0 at the
2977 end of the map. Decrease the map-length byte too. */
2978 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2982 /* Build real range table from work area. */
2983 if (RANGE_TABLE_WORK_USED (range_table_work
)
2984 || RANGE_TABLE_WORK_BITS (range_table_work
))
2987 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2989 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2990 bytes for flags, two for COUNT, and three bytes for
2992 GET_BUFFER_SPACE (4 + used
* 3);
2994 /* Indicate the existence of range table. */
2995 laststart
[1] |= 0x80;
2997 /* Store the character class flag bits into the range table.
2998 If not in emacs, these flag bits are always 0. */
2999 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) & 0xff;
3000 *b
++ = RANGE_TABLE_WORK_BITS (range_table_work
) >> 8;
3002 STORE_NUMBER_AND_INCR (b
, used
/ 2);
3003 for (i
= 0; i
< used
; i
++)
3004 STORE_CHARACTER_AND_INCR
3005 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
3012 if (syntax
& RE_NO_BK_PARENS
)
3019 if (syntax
& RE_NO_BK_PARENS
)
3026 if (syntax
& RE_NEWLINE_ALT
)
3033 if (syntax
& RE_NO_BK_VBAR
)
3040 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3041 goto handle_interval
;
3047 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3049 /* Do not translate the character after the \, so that we can
3050 distinguish, e.g., \B from \b, even if we normally would
3051 translate, e.g., B to b. */
3057 if (syntax
& RE_NO_BK_PARENS
)
3058 goto normal_backslash
;
3063 regnum_t regnum
= 0;
3066 /* Look for a special (?...) construct */
3067 if ((syntax
& RE_SHY_GROUPS
) && *p
== '?')
3069 PATFETCH (c
); /* Gobble up the '?'. */
3075 case ':': shy
= 1; break;
3077 /* An explicitly specified regnum must start
3080 FREE_STACK_RETURN (REG_BADPAT
);
3081 case '1': case '2': case '3': case '4':
3082 case '5': case '6': case '7': case '8': case '9':
3083 regnum
= 10*regnum
+ (c
- '0'); break;
3085 /* Only (?:...) is supported right now. */
3086 FREE_STACK_RETURN (REG_BADPAT
);
3093 regnum
= ++bufp
->re_nsub
;
3095 { /* It's actually not shy, but explicitly numbered. */
3097 if (regnum
> bufp
->re_nsub
)
3098 bufp
->re_nsub
= regnum
;
3099 else if (regnum
> bufp
->re_nsub
3100 /* Ideally, we'd want to check that the specified
3101 group can't have matched (i.e. all subgroups
3102 using the same regnum are in other branches of
3103 OR patterns), but we don't currently keep track
3104 of enough info to do that easily. */
3105 || group_in_compile_stack (compile_stack
, regnum
))
3106 FREE_STACK_RETURN (REG_BADPAT
);
3109 /* It's really shy. */
3110 regnum
= - bufp
->re_nsub
;
3112 if (COMPILE_STACK_FULL
)
3114 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3115 compile_stack_elt_t
);
3116 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3118 compile_stack
.size
<<= 1;
3121 /* These are the values to restore when we hit end of this
3122 group. They are all relative offsets, so that if the
3123 whole pattern moves because of realloc, they will still
3125 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
3126 COMPILE_STACK_TOP
.fixup_alt_jump
3127 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
3128 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
3129 COMPILE_STACK_TOP
.regnum
= regnum
;
3131 /* Do not push a start_memory for groups beyond the last one
3132 we can represent in the compiled pattern. */
3133 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3134 BUF_PUSH_2 (start_memory
, regnum
);
3136 compile_stack
.avail
++;
3141 /* If we've reached MAX_REGNUM groups, then this open
3142 won't actually generate any code, so we'll have to
3143 clear pending_exact explicitly. */
3149 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3151 if (COMPILE_STACK_EMPTY
)
3153 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3154 goto normal_backslash
;
3156 FREE_STACK_RETURN (REG_ERPAREN
);
3162 /* See similar code for backslashed left paren above. */
3163 if (COMPILE_STACK_EMPTY
)
3165 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3168 FREE_STACK_RETURN (REG_ERPAREN
);
3171 /* Since we just checked for an empty stack above, this
3172 ``can't happen''. */
3173 assert (compile_stack
.avail
!= 0);
3175 /* We don't just want to restore into `regnum', because
3176 later groups should continue to be numbered higher,
3177 as in `(ab)c(de)' -- the second group is #2. */
3180 compile_stack
.avail
--;
3181 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
3183 = COMPILE_STACK_TOP
.fixup_alt_jump
3184 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3186 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
3187 regnum
= COMPILE_STACK_TOP
.regnum
;
3188 /* If we've reached MAX_REGNUM groups, then this open
3189 won't actually generate any code, so we'll have to
3190 clear pending_exact explicitly. */
3193 /* We're at the end of the group, so now we know how many
3194 groups were inside this one. */
3195 if (regnum
<= MAX_REGNUM
&& regnum
> 0)
3196 BUF_PUSH_2 (stop_memory
, regnum
);
3201 case '|': /* `\|'. */
3202 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3203 goto normal_backslash
;
3205 if (syntax
& RE_LIMITED_OPS
)
3208 /* Insert before the previous alternative a jump which
3209 jumps to this alternative if the former fails. */
3210 GET_BUFFER_SPACE (3);
3211 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
3215 /* The alternative before this one has a jump after it
3216 which gets executed if it gets matched. Adjust that
3217 jump so it will jump to this alternative's analogous
3218 jump (put in below, which in turn will jump to the next
3219 (if any) alternative's such jump, etc.). The last such
3220 jump jumps to the correct final destination. A picture:
3226 If we are at `b', then fixup_alt_jump right now points to a
3227 three-byte space after `a'. We'll put in the jump, set
3228 fixup_alt_jump to right after `b', and leave behind three
3229 bytes which we'll fill in when we get to after `c'. */
3233 /* Mark and leave space for a jump after this alternative,
3234 to be filled in later either by next alternative or
3235 when know we're at the end of a series of alternatives. */
3237 GET_BUFFER_SPACE (3);
3246 /* If \{ is a literal. */
3247 if (!(syntax
& RE_INTERVALS
)
3248 /* If we're at `\{' and it's not the open-interval
3250 || (syntax
& RE_NO_BK_BRACES
))
3251 goto normal_backslash
;
3255 /* If got here, then the syntax allows intervals. */
3257 /* At least (most) this many matches must be made. */
3258 int lower_bound
= 0, upper_bound
= -1;
3262 GET_INTERVAL_COUNT (lower_bound
);
3265 GET_INTERVAL_COUNT (upper_bound
);
3267 /* Interval such as `{1}' => match exactly once. */
3268 upper_bound
= lower_bound
;
3271 || (0 <= upper_bound
&& upper_bound
< lower_bound
))
3272 FREE_STACK_RETURN (REG_BADBR
);
3274 if (!(syntax
& RE_NO_BK_BRACES
))
3277 FREE_STACK_RETURN (REG_BADBR
);
3279 FREE_STACK_RETURN (REG_EESCAPE
);
3284 FREE_STACK_RETURN (REG_BADBR
);
3286 /* We just parsed a valid interval. */
3288 /* If it's invalid to have no preceding re. */
3291 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3292 FREE_STACK_RETURN (REG_BADRPT
);
3293 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3296 goto unfetch_interval
;
3299 if (upper_bound
== 0)
3300 /* If the upper bound is zero, just drop the sub pattern
3303 else if (lower_bound
== 1 && upper_bound
== 1)
3304 /* Just match it once: nothing to do here. */
3307 /* Otherwise, we have a nontrivial interval. When
3308 we're all done, the pattern will look like:
3309 set_number_at <jump count> <upper bound>
3310 set_number_at <succeed_n count> <lower bound>
3311 succeed_n <after jump addr> <succeed_n count>
3313 jump_n <succeed_n addr> <jump count>
3314 (The upper bound and `jump_n' are omitted if
3315 `upper_bound' is 1, though.) */
3317 { /* If the upper bound is > 1, we need to insert
3318 more at the end of the loop. */
3319 unsigned int nbytes
= (upper_bound
< 0 ? 3
3320 : upper_bound
> 1 ? 5 : 0);
3321 unsigned int startoffset
= 0;
3323 GET_BUFFER_SPACE (20); /* We might use less. */
3325 if (lower_bound
== 0)
3327 /* A succeed_n that starts with 0 is really a
3328 a simple on_failure_jump_loop. */
3329 INSERT_JUMP (on_failure_jump_loop
, laststart
,
3335 /* Initialize lower bound of the `succeed_n', even
3336 though it will be set during matching by its
3337 attendant `set_number_at' (inserted next),
3338 because `re_compile_fastmap' needs to know.
3339 Jump to the `jump_n' we might insert below. */
3340 INSERT_JUMP2 (succeed_n
, laststart
,
3345 /* Code to initialize the lower bound. Insert
3346 before the `succeed_n'. The `5' is the last two
3347 bytes of this `set_number_at', plus 3 bytes of
3348 the following `succeed_n'. */
3349 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3354 if (upper_bound
< 0)
3356 /* A negative upper bound stands for infinity,
3357 in which case it degenerates to a plain jump. */
3358 STORE_JUMP (jump
, b
, laststart
+ startoffset
);
3361 else if (upper_bound
> 1)
3362 { /* More than one repetition is allowed, so
3363 append a backward jump to the `succeed_n'
3364 that starts this interval.
3366 When we've reached this during matching,
3367 we'll have matched the interval once, so
3368 jump back only `upper_bound - 1' times. */
3369 STORE_JUMP2 (jump_n
, b
, laststart
+ startoffset
,
3373 /* The location we want to set is the second
3374 parameter of the `jump_n'; that is `b-2' as
3375 an absolute address. `laststart' will be
3376 the `set_number_at' we're about to insert;
3377 `laststart+3' the number to set, the source
3378 for the relative address. But we are
3379 inserting into the middle of the pattern --
3380 so everything is getting moved up by 5.
3381 Conclusion: (b - 2) - (laststart + 3) + 5,
3382 i.e., b - laststart.
3384 We insert this at the beginning of the loop
3385 so that if we fail during matching, we'll
3386 reinitialize the bounds. */
3387 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3388 upper_bound
- 1, b
);
3393 beg_interval
= NULL
;
3398 /* If an invalid interval, match the characters as literals. */
3399 assert (beg_interval
);
3401 beg_interval
= NULL
;
3403 /* normal_char and normal_backslash need `c'. */
3406 if (!(syntax
& RE_NO_BK_BRACES
))
3408 assert (p
> pattern
&& p
[-1] == '\\');
3409 goto normal_backslash
;
3423 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3429 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3435 BUF_PUSH_2 (categoryspec
, c
);
3441 BUF_PUSH_2 (notcategoryspec
, c
);
3447 if (syntax
& RE_NO_GNU_OPS
)
3450 BUF_PUSH_2 (syntaxspec
, Sword
);
3455 if (syntax
& RE_NO_GNU_OPS
)
3458 BUF_PUSH_2 (notsyntaxspec
, Sword
);
3463 if (syntax
& RE_NO_GNU_OPS
)
3470 if (syntax
& RE_NO_GNU_OPS
)
3477 if (syntax
& RE_NO_GNU_OPS
)
3486 FREE_STACK_RETURN (REG_BADPAT
);
3490 if (syntax
& RE_NO_GNU_OPS
)
3492 BUF_PUSH (wordbound
);
3496 if (syntax
& RE_NO_GNU_OPS
)
3498 BUF_PUSH (notwordbound
);
3502 if (syntax
& RE_NO_GNU_OPS
)
3508 if (syntax
& RE_NO_GNU_OPS
)
3513 case '1': case '2': case '3': case '4': case '5':
3514 case '6': case '7': case '8': case '9':
3518 if (syntax
& RE_NO_BK_REFS
)
3519 goto normal_backslash
;
3523 if (reg
> bufp
->re_nsub
|| reg
< 1
3524 /* Can't back reference to a subexp before its end. */
3525 || group_in_compile_stack (compile_stack
, reg
))
3526 FREE_STACK_RETURN (REG_ESUBREG
);
3529 BUF_PUSH_2 (duplicate
, reg
);
3536 if (syntax
& RE_BK_PLUS_QM
)
3539 goto normal_backslash
;
3543 /* You might think it would be useful for \ to mean
3544 not to translate; but if we don't translate it
3545 it will never match anything. */
3552 /* Expects the character in `c'. */
3554 /* If no exactn currently being built. */
3557 /* If last exactn not at current position. */
3558 || pending_exact
+ *pending_exact
+ 1 != b
3560 /* We have only one byte following the exactn for the count. */
3561 || *pending_exact
>= (1 << BYTEWIDTH
) - MAX_MULTIBYTE_LENGTH
3563 /* If followed by a repetition operator. */
3564 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
3565 || ((syntax
& RE_BK_PLUS_QM
)
3566 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3567 : p
!= pend
&& (*p
== '+' || *p
== '?'))
3568 || ((syntax
& RE_INTERVALS
)
3569 && ((syntax
& RE_NO_BK_BRACES
)
3570 ? p
!= pend
&& *p
== '{'
3571 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
3573 /* Start building a new exactn. */
3577 BUF_PUSH_2 (exactn
, 0);
3578 pending_exact
= b
- 1;
3581 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH
);
3588 len
= CHAR_STRING (c
, b
);
3593 c1
= RE_CHAR_TO_MULTIBYTE (c
);
3594 if (! CHAR_BYTE8_P (c1
))
3596 re_wchar_t c2
= TRANSLATE (c1
);
3598 if (c1
!= c2
&& (c1
= RE_CHAR_TO_UNIBYTE (c2
)) >= 0)
3604 (*pending_exact
) += len
;
3609 } /* while p != pend */
3612 /* Through the pattern now. */
3616 if (!COMPILE_STACK_EMPTY
)
3617 FREE_STACK_RETURN (REG_EPAREN
);
3619 /* If we don't want backtracking, force success
3620 the first time we reach the end of the compiled pattern. */
3621 if (!posix_backtracking
)
3624 /* We have succeeded; set the length of the buffer. */
3625 bufp
->used
= b
- bufp
->buffer
;
3630 re_compile_fastmap (bufp
);
3631 DEBUG_PRINT ("\nCompiled pattern: \n");
3632 print_compiled_pattern (bufp
);
3637 #ifndef MATCH_MAY_ALLOCATE
3638 /* Initialize the failure stack to the largest possible stack. This
3639 isn't necessary unless we're trying to avoid calling alloca in
3640 the search and match routines. */
3642 int num_regs
= bufp
->re_nsub
+ 1;
3644 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
3646 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3647 falk_stack
.stack
= realloc (fail_stack
.stack
,
3648 fail_stack
.size
* sizeof *falk_stack
.stack
);
3651 regex_grow_registers (num_regs
);
3653 #endif /* not MATCH_MAY_ALLOCATE */
3655 FREE_STACK_RETURN (REG_NOERROR
);
3660 # undef posix_backtracking
3662 } /* regex_compile */
3664 /* Subroutines for `regex_compile'. */
3666 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3669 store_op1 (re_opcode_t op
, unsigned char *loc
, int arg
)
3671 *loc
= (unsigned char) op
;
3672 STORE_NUMBER (loc
+ 1, arg
);
3676 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3679 store_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
)
3681 *loc
= (unsigned char) op
;
3682 STORE_NUMBER (loc
+ 1, arg1
);
3683 STORE_NUMBER (loc
+ 3, arg2
);
3687 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3688 for OP followed by two-byte integer parameter ARG. */
3691 insert_op1 (re_opcode_t op
, unsigned char *loc
, int arg
, unsigned char *end
)
3693 register unsigned char *pfrom
= end
;
3694 register unsigned char *pto
= end
+ 3;
3696 while (pfrom
!= loc
)
3699 store_op1 (op
, loc
, arg
);
3703 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3706 insert_op2 (re_opcode_t op
, unsigned char *loc
, int arg1
, int arg2
, unsigned char *end
)
3708 register unsigned char *pfrom
= end
;
3709 register unsigned char *pto
= end
+ 5;
3711 while (pfrom
!= loc
)
3714 store_op2 (op
, loc
, arg1
, arg2
);
3718 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3719 after an alternative or a begin-subexpression. We assume there is at
3720 least one character before the ^. */
3723 at_begline_loc_p (const_re_char
*pattern
, const_re_char
*p
, reg_syntax_t syntax
)
3725 re_char
*prev
= p
- 2;
3726 boolean odd_backslashes
;
3728 /* After a subexpression? */
3730 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3732 /* After an alternative? */
3733 else if (*prev
== '|')
3734 odd_backslashes
= (syntax
& RE_NO_BK_VBAR
) == 0;
3736 /* After a shy subexpression? */
3737 else if (*prev
== ':' && (syntax
& RE_SHY_GROUPS
))
3739 /* Skip over optional regnum. */
3740 while (prev
- 1 >= pattern
&& prev
[-1] >= '0' && prev
[-1] <= '9')
3743 if (!(prev
- 2 >= pattern
3744 && prev
[-1] == '?' && prev
[-2] == '('))
3747 odd_backslashes
= (syntax
& RE_NO_BK_PARENS
) == 0;
3752 /* Count the number of preceding backslashes. */
3754 while (prev
- 1 >= pattern
&& prev
[-1] == '\\')
3756 return (p
- prev
) & odd_backslashes
;
3760 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3761 at least one character after the $, i.e., `P < PEND'. */
3764 at_endline_loc_p (const_re_char
*p
, const_re_char
*pend
, reg_syntax_t syntax
)
3767 boolean next_backslash
= *next
== '\\';
3768 re_char
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3771 /* Before a subexpression? */
3772 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3773 : next_backslash
&& next_next
&& *next_next
== ')')
3774 /* Before an alternative? */
3775 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3776 : next_backslash
&& next_next
&& *next_next
== '|');
3780 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3781 false if it's not. */
3784 group_in_compile_stack (compile_stack_type compile_stack
, regnum_t regnum
)
3786 ssize_t this_element
;
3788 for (this_element
= compile_stack
.avail
- 1;
3791 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3798 If fastmap is non-NULL, go through the pattern and fill fastmap
3799 with all the possible leading chars. If fastmap is NULL, don't
3800 bother filling it up (obviously) and only return whether the
3801 pattern could potentially match the empty string.
3803 Return 1 if p..pend might match the empty string.
3804 Return 0 if p..pend matches at least one char.
3805 Return -1 if fastmap was not updated accurately. */
3808 analyze_first (const_re_char
*p
, const_re_char
*pend
, char *fastmap
,
3809 const int multibyte
)
3814 /* If all elements for base leading-codes in fastmap is set, this
3815 flag is set true. */
3816 boolean match_any_multibyte_characters
= false;
3820 /* The loop below works as follows:
3821 - It has a working-list kept in the PATTERN_STACK and which basically
3822 starts by only containing a pointer to the first operation.
3823 - If the opcode we're looking at is a match against some set of
3824 chars, then we add those chars to the fastmap and go on to the
3825 next work element from the worklist (done via `break').
3826 - If the opcode is a control operator on the other hand, we either
3827 ignore it (if it's meaningless at this point, such as `start_memory')
3828 or execute it (if it's a jump). If the jump has several destinations
3829 (i.e. `on_failure_jump'), then we push the other destination onto the
3831 We guarantee termination by ignoring backward jumps (more or less),
3832 so that `p' is monotonically increasing. More to the point, we
3833 never set `p' (or push) anything `<= p1'. */
3837 /* `p1' is used as a marker of how far back a `on_failure_jump'
3838 can go without being ignored. It is normally equal to `p'
3839 (which prevents any backward `on_failure_jump') except right
3840 after a plain `jump', to allow patterns such as:
3843 10: on_failure_jump 3
3844 as used for the *? operator. */
3853 /* If the first character has to match a backreference, that means
3854 that the group was empty (since it already matched). Since this
3855 is the only case that interests us here, we can assume that the
3856 backreference must match the empty string. */
3861 /* Following are the cases which match a character. These end
3867 /* If multibyte is nonzero, the first byte of each
3868 character is an ASCII or a leading code. Otherwise,
3869 each byte is a character. Thus, this works in both
3874 /* For the case of matching this unibyte regex
3875 against multibyte, we must set a leading code of
3876 the corresponding multibyte character. */
3877 int c
= RE_CHAR_TO_MULTIBYTE (p
[1]);
3879 fastmap
[CHAR_LEADING_CODE (c
)] = 1;
3886 /* We could put all the chars except for \n (and maybe \0)
3887 but we don't bother since it is generally not worth it. */
3888 if (!fastmap
) break;
3893 if (!fastmap
) break;
3895 /* Chars beyond end of bitmap are possible matches. */
3896 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3897 j
< (1 << BYTEWIDTH
); j
++)
3903 if (!fastmap
) break;
3904 not = (re_opcode_t
) *(p
- 1) == charset_not
;
3905 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3907 if (!!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))) ^ not)
3911 if (/* Any leading code can possibly start a character
3912 which doesn't match the specified set of characters. */
3915 /* If we can match a character class, we can match any
3916 multibyte characters. */
3917 (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3918 && CHARSET_RANGE_TABLE_BITS (&p
[-2]) != 0))
3921 if (match_any_multibyte_characters
== false)
3923 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3924 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3926 match_any_multibyte_characters
= true;
3930 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3931 && match_any_multibyte_characters
== false)
3933 /* Set fastmap[I] to 1 where I is a leading code of each
3934 multibyte character in the range table. */
3936 unsigned char lc1
, lc2
;
3938 /* Make P points the range table. `+ 2' is to skip flag
3939 bits for a character class. */
3940 p
+= CHARSET_BITMAP_SIZE (&p
[-2]) + 2;
3942 /* Extract the number of ranges in range table into COUNT. */
3943 EXTRACT_NUMBER_AND_INCR (count
, p
);
3944 for (; count
> 0; count
--, p
+= 3)
3946 /* Extract the start and end of each range. */
3947 EXTRACT_CHARACTER (c
, p
);
3948 lc1
= CHAR_LEADING_CODE (c
);
3950 EXTRACT_CHARACTER (c
, p
);
3951 lc2
= CHAR_LEADING_CODE (c
);
3952 for (j
= lc1
; j
<= lc2
; j
++)
3961 if (!fastmap
) break;
3963 not = (re_opcode_t
)p
[-1] == notsyntaxspec
;
3965 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3966 if ((SYNTAX (j
) == (enum syntaxcode
) k
) ^ not)
3970 /* This match depends on text properties. These end with
3971 aborting optimizations. */
3975 case notcategoryspec
:
3976 if (!fastmap
) break;
3977 not = (re_opcode_t
)p
[-1] == notcategoryspec
;
3979 for (j
= (1 << BYTEWIDTH
); j
>= 0; j
--)
3980 if ((CHAR_HAS_CATEGORY (j
, k
)) ^ not)
3983 /* Any leading code can possibly start a character which
3984 has or doesn't has the specified category. */
3985 if (match_any_multibyte_characters
== false)
3987 for (j
= MIN_MULTIBYTE_LEADING_CODE
;
3988 j
<= MAX_MULTIBYTE_LEADING_CODE
; j
++)
3990 match_any_multibyte_characters
= true;
3994 /* All cases after this match the empty string. These end with
4014 EXTRACT_NUMBER_AND_INCR (j
, p
);
4016 /* Backward jumps can only go back to code that we've already
4017 visited. `re_compile' should make sure this is true. */
4022 case on_failure_jump
:
4023 case on_failure_keep_string_jump
:
4024 case on_failure_jump_loop
:
4025 case on_failure_jump_nastyloop
:
4026 case on_failure_jump_smart
:
4032 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4033 to jump back to "just after here". */
4036 case on_failure_jump
:
4037 case on_failure_keep_string_jump
:
4038 case on_failure_jump_nastyloop
:
4039 case on_failure_jump_loop
:
4040 case on_failure_jump_smart
:
4041 EXTRACT_NUMBER_AND_INCR (j
, p
);
4043 ; /* Backward jump to be ignored. */
4045 { /* We have to look down both arms.
4046 We first go down the "straight" path so as to minimize
4047 stack usage when going through alternatives. */
4048 int r
= analyze_first (p
, pend
, fastmap
, multibyte
);
4056 /* This code simply does not properly handle forward jump_n. */
4057 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
); assert (j
< 0));
4059 /* jump_n can either jump or fall through. The (backward) jump
4060 case has already been handled, so we only need to look at the
4061 fallthrough case. */
4065 /* If N == 0, it should be an on_failure_jump_loop instead. */
4066 DEBUG_STATEMENT (EXTRACT_NUMBER (j
, p
+ 2); assert (j
> 0));
4068 /* We only care about one iteration of the loop, so we don't
4069 need to consider the case where this behaves like an
4086 abort (); /* We have listed all the cases. */
4089 /* Getting here means we have found the possible starting
4090 characters for one path of the pattern -- and that the empty
4091 string does not match. We need not follow this path further. */
4095 /* We reached the end without matching anything. */
4098 } /* analyze_first */
4100 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4101 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4102 characters can start a string that matches the pattern. This fastmap
4103 is used by re_search to skip quickly over impossible starting points.
4105 Character codes above (1 << BYTEWIDTH) are not represented in the
4106 fastmap, but the leading codes are represented. Thus, the fastmap
4107 indicates which character sets could start a match.
4109 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4110 area as BUFP->fastmap.
4112 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4115 Returns 0 if we succeed, -2 if an internal error. */
4118 re_compile_fastmap (struct re_pattern_buffer
*bufp
)
4120 char *fastmap
= bufp
->fastmap
;
4123 assert (fastmap
&& bufp
->buffer
);
4125 memset (fastmap
, 0, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4126 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4128 analysis
= analyze_first (bufp
->buffer
, bufp
->buffer
+ bufp
->used
,
4129 fastmap
, RE_MULTIBYTE_P (bufp
));
4130 bufp
->can_be_null
= (analysis
!= 0);
4132 } /* re_compile_fastmap */
4134 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4135 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4136 this memory for recording register information. STARTS and ENDS
4137 must be allocated using the malloc library routine, and must each
4138 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4140 If NUM_REGS == 0, then subsequent matches should allocate their own
4143 Unless this function is called, the first search or match using
4144 PATTERN_BUFFER will allocate its own register data, without
4145 freeing the old data. */
4148 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
, unsigned int num_regs
, regoff_t
*starts
, regoff_t
*ends
)
4152 bufp
->regs_allocated
= REGS_REALLOCATE
;
4153 regs
->num_regs
= num_regs
;
4154 regs
->start
= starts
;
4159 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4161 regs
->start
= regs
->end
= 0;
4164 WEAK_ALIAS (__re_set_registers
, re_set_registers
)
4166 /* Searching routines. */
4168 /* Like re_search_2, below, but only one string is specified, and
4169 doesn't let you say where to stop matching. */
4172 re_search (struct re_pattern_buffer
*bufp
, const char *string
, size_t size
,
4173 ssize_t startpos
, ssize_t range
, struct re_registers
*regs
)
4175 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4178 WEAK_ALIAS (__re_search
, re_search
)
4180 /* Head address of virtual concatenation of string. */
4181 #define HEAD_ADDR_VSTRING(P) \
4182 (((P) >= size1 ? string2 : string1))
4184 /* Address of POS in the concatenation of virtual string. */
4185 #define POS_ADDR_VSTRING(POS) \
4186 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4188 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4189 virtual concatenation of STRING1 and STRING2, starting first at index
4190 STARTPOS, then at STARTPOS + 1, and so on.
4192 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4194 RANGE is how far to scan while trying to match. RANGE = 0 means try
4195 only at STARTPOS; in general, the last start tried is STARTPOS +
4198 In REGS, return the indices of the virtual concatenation of STRING1
4199 and STRING2 that matched the entire BUFP->buffer and its contained
4202 Do not consider matching one past the index STOP in the virtual
4203 concatenation of STRING1 and STRING2.
4205 We return either the position in the strings at which the match was
4206 found, -1 if no match, or -2 if error (such as failure
4210 re_search_2 (struct re_pattern_buffer
*bufp
, const char *str1
, size_t size1
,
4211 const char *str2
, size_t size2
, ssize_t startpos
, ssize_t range
,
4212 struct re_registers
*regs
, ssize_t stop
)
4215 re_char
*string1
= (re_char
*) str1
;
4216 re_char
*string2
= (re_char
*) str2
;
4217 register char *fastmap
= bufp
->fastmap
;
4218 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4219 size_t total_size
= size1
+ size2
;
4220 ssize_t endpos
= startpos
+ range
;
4221 boolean anchored_start
;
4222 /* Nonzero if we are searching multibyte string. */
4223 const boolean multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4225 /* Check for out-of-range STARTPOS. */
4226 if (startpos
< 0 || startpos
> total_size
)
4229 /* Fix up RANGE if it might eventually take us outside
4230 the virtual concatenation of STRING1 and STRING2.
4231 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4233 range
= 0 - startpos
;
4234 else if (endpos
> total_size
)
4235 range
= total_size
- startpos
;
4237 /* If the search isn't to be a backwards one, don't waste time in a
4238 search for a pattern anchored at beginning of buffer. */
4239 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
4248 /* In a forward search for something that starts with \=.
4249 don't keep searching past point. */
4250 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
4252 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
4258 /* Update the fastmap now if not correct already. */
4259 if (fastmap
&& !bufp
->fastmap_accurate
)
4260 re_compile_fastmap (bufp
);
4262 /* See whether the pattern is anchored. */
4263 anchored_start
= (bufp
->buffer
[0] == begline
);
4266 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4268 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos
));
4270 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4274 /* Loop through the string, looking for a place to start matching. */
4277 /* If the pattern is anchored,
4278 skip quickly past places we cannot match.
4279 We don't bother to treat startpos == 0 specially
4280 because that case doesn't repeat. */
4281 if (anchored_start
&& startpos
> 0)
4283 if (! ((startpos
<= size1
? string1
[startpos
- 1]
4284 : string2
[startpos
- size1
- 1])
4289 /* If a fastmap is supplied, skip quickly over characters that
4290 cannot be the start of a match. If the pattern can match the
4291 null string, however, we don't need to skip characters; we want
4292 the first null string. */
4293 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
4295 register re_char
*d
;
4296 register re_wchar_t buf_ch
;
4298 d
= POS_ADDR_VSTRING (startpos
);
4300 if (range
> 0) /* Searching forwards. */
4302 ssize_t irange
= range
, lim
= 0;
4304 if (startpos
< size1
&& startpos
+ range
>= size1
)
4305 lim
= range
- (size1
- startpos
);
4307 /* Written out as an if-else to avoid testing `translate'
4309 if (RE_TRANSLATE_P (translate
))
4316 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4317 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
4318 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4321 range
-= buf_charlen
;
4327 register re_wchar_t ch
, translated
;
4330 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4331 translated
= RE_TRANSLATE (translate
, ch
);
4332 if (translated
!= ch
4333 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4335 if (fastmap
[buf_ch
])
4348 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
4349 if (fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4351 range
-= buf_charlen
;
4355 while (range
> lim
&& !fastmap
[*d
])
4361 startpos
+= irange
- range
;
4363 else /* Searching backwards. */
4367 buf_ch
= STRING_CHAR (d
);
4368 buf_ch
= TRANSLATE (buf_ch
);
4369 if (! fastmap
[CHAR_LEADING_CODE (buf_ch
)])
4374 register re_wchar_t ch
, translated
;
4377 ch
= RE_CHAR_TO_MULTIBYTE (buf_ch
);
4378 translated
= TRANSLATE (ch
);
4379 if (translated
!= ch
4380 && (ch
= RE_CHAR_TO_UNIBYTE (translated
)) >= 0)
4382 if (! fastmap
[TRANSLATE (buf_ch
)])
4388 /* If can't match the null string, and that's all we have left, fail. */
4389 if (range
>= 0 && startpos
== total_size
&& fastmap
4390 && !bufp
->can_be_null
)
4393 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4394 startpos
, regs
, stop
);
4407 /* Update STARTPOS to the next character boundary. */
4410 re_char
*p
= POS_ADDR_VSTRING (startpos
);
4411 int len
= BYTES_BY_CHAR_HEAD (*p
);
4429 /* Update STARTPOS to the previous character boundary. */
4432 re_char
*p
= POS_ADDR_VSTRING (startpos
) + 1;
4434 re_char
*phead
= HEAD_ADDR_VSTRING (startpos
);
4436 /* Find the head of multibyte form. */
4437 PREV_CHAR_BOUNDARY (p
, phead
);
4438 range
+= p0
- 1 - p
;
4442 startpos
-= p0
- 1 - p
;
4448 WEAK_ALIAS (__re_search_2
, re_search_2
)
4450 /* Declarations and macros for re_match_2. */
4452 static int bcmp_translate (re_char
*s1
, re_char
*s2
,
4453 register ssize_t len
,
4454 RE_TRANSLATE_TYPE translate
,
4455 const int multibyte
);
4457 /* This converts PTR, a pointer into one of the search strings `string1'
4458 and `string2' into an offset from the beginning of that string. */
4459 #define POINTER_TO_OFFSET(ptr) \
4460 (FIRST_STRING_P (ptr) \
4462 : (ptr) - string2 + (ptrdiff_t) size1)
4464 /* Call before fetching a character with *d. This switches over to
4465 string2 if necessary.
4466 Check re_match_2_internal for a discussion of why end_match_2 might
4467 not be within string2 (but be equal to end_match_1 instead). */
4468 #define PREFETCH() \
4471 /* End of string2 => fail. */ \
4472 if (dend == end_match_2) \
4474 /* End of string1 => advance to string2. */ \
4476 dend = end_match_2; \
4479 /* Call before fetching a char with *d if you already checked other limits.
4480 This is meant for use in lookahead operations like wordend, etc..
4481 where we might need to look at parts of the string that might be
4482 outside of the LIMITs (i.e past `stop'). */
4483 #define PREFETCH_NOLIMIT() \
4487 dend = end_match_2; \
4490 /* Test if at very beginning or at very end of the virtual concatenation
4491 of `string1' and `string2'. If only one string, it's `string2'. */
4492 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4493 #define AT_STRINGS_END(d) ((d) == end2)
4495 /* Disabled due to a compiler bug -- see comment at case wordbound */
4497 /* The comment at case wordbound is following one, but we don't use
4498 AT_WORD_BOUNDARY anymore to support multibyte form.
4500 The DEC Alpha C compiler 3.x generates incorrect code for the
4501 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4502 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4503 macro and introducing temporary variables works around the bug. */
4506 /* Test if D points to a character which is word-constituent. We have
4507 two special cases to check for: if past the end of string1, look at
4508 the first character in string2; and if before the beginning of
4509 string2, look at the last character in string1. */
4510 #define WORDCHAR_P(d) \
4511 (SYNTAX ((d) == end1 ? *string2 \
4512 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4515 /* Test if the character before D and the one at D differ with respect
4516 to being word-constituent. */
4517 #define AT_WORD_BOUNDARY(d) \
4518 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4519 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4522 /* Free everything we malloc. */
4523 #ifdef MATCH_MAY_ALLOCATE
4524 # define FREE_VAR(var) \
4532 # define FREE_VARIABLES() \
4534 REGEX_FREE_STACK (fail_stack.stack); \
4535 FREE_VAR (regstart); \
4536 FREE_VAR (regend); \
4537 FREE_VAR (best_regstart); \
4538 FREE_VAR (best_regend); \
4539 REGEX_SAFE_FREE (); \
4542 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4543 #endif /* not MATCH_MAY_ALLOCATE */
4546 /* Optimization routines. */
4548 /* If the operation is a match against one or more chars,
4549 return a pointer to the next operation, else return NULL. */
4551 skip_one_char (const_re_char
*p
)
4564 if (CHARSET_RANGE_TABLE_EXISTS_P (p
- 1))
4567 p
= CHARSET_RANGE_TABLE (p
- 1);
4568 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4569 p
= CHARSET_RANGE_TABLE_END (p
, mcnt
);
4572 p
+= 1 + CHARSET_BITMAP_SIZE (p
- 1);
4579 case notcategoryspec
:
4591 /* Jump over non-matching operations. */
4593 skip_noops (const_re_char
*p
, const_re_char
*pend
)
4607 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4618 /* Test if C matches charset op. *PP points to the charset or charset_not
4619 opcode. When the function finishes, *PP will be advanced past that opcode.
4620 C is character to test (possibly after translations) and CORIG is original
4621 character (i.e. without any translations). UNIBYTE denotes whether c is
4622 unibyte or multibyte character. */
4624 execute_charset (const_re_char
**pp
, unsigned c
, unsigned corig
, bool unibyte
)
4626 re_char
*p
= *pp
, *rtp
= NULL
;
4627 bool not = (re_opcode_t
) *p
== charset_not
;
4629 if (CHARSET_RANGE_TABLE_EXISTS_P (p
))
4632 rtp
= CHARSET_RANGE_TABLE (p
);
4633 EXTRACT_NUMBER_AND_INCR (count
, rtp
);
4634 *pp
= CHARSET_RANGE_TABLE_END ((rtp
), (count
));
4637 *pp
+= 2 + CHARSET_BITMAP_SIZE (p
);
4639 if (unibyte
&& c
< (1 << BYTEWIDTH
))
4640 { /* Lookup bitmap. */
4641 /* Cast to `unsigned' instead of `unsigned char' in
4642 case the bit list is a full 32 bytes long. */
4643 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (p
) * BYTEWIDTH
)
4644 && p
[2 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4650 int class_bits
= CHARSET_RANGE_TABLE_BITS (p
);
4651 re_wchar_t range_start
, range_end
;
4653 /* Sort tests by the most commonly used classes with some adjustment to which
4654 tests are easiest to perform. Take a look at comment in re_wctype_parse
4655 for table with frequencies of character class names. */
4657 if ((class_bits
& BIT_MULTIBYTE
) ||
4658 (class_bits
& BIT_ALNUM
&& ISALNUM (c
)) ||
4659 (class_bits
& BIT_ALPHA
&& ISALPHA (c
)) ||
4660 (class_bits
& BIT_SPACE
&& ISSPACE (c
)) ||
4661 (class_bits
& BIT_WORD
&& ISWORD (c
)) ||
4662 ((class_bits
& BIT_UPPER
) &&
4663 (ISUPPER (c
) || (corig
!= c
&&
4664 c
== downcase (corig
) && ISLOWER (c
)))) ||
4665 ((class_bits
& BIT_LOWER
) &&
4666 (ISLOWER (c
) || (corig
!= c
&&
4667 c
== upcase (corig
) && ISUPPER(c
)))) ||
4668 (class_bits
& BIT_PUNCT
&& ISPUNCT (c
)) ||
4669 (class_bits
& BIT_GRAPH
&& ISGRAPH (c
)) ||
4670 (class_bits
& BIT_PRINT
&& ISPRINT (c
)))
4673 for (p
= *pp
; rtp
< p
; rtp
+= 2 * 3)
4675 EXTRACT_CHARACTER (range_start
, rtp
);
4676 EXTRACT_CHARACTER (range_end
, rtp
+ 3);
4677 if (range_start
<= c
&& c
<= range_end
)
4685 /* Non-zero if "p1 matches something" implies "p2 fails". */
4687 mutually_exclusive_p (struct re_pattern_buffer
*bufp
, const_re_char
*p1
,
4691 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4692 unsigned char *pend
= bufp
->buffer
+ bufp
->used
;
4694 assert (p1
>= bufp
->buffer
&& p1
< pend
4695 && p2
>= bufp
->buffer
&& p2
<= pend
);
4697 /* Skip over open/close-group commands.
4698 If what follows this loop is a ...+ construct,
4699 look at what begins its body, since we will have to
4700 match at least one of that. */
4701 p2
= skip_noops (p2
, pend
);
4702 /* The same skip can be done for p1, except that this function
4703 is only used in the case where p1 is a simple match operator. */
4704 /* p1 = skip_noops (p1, pend); */
4706 assert (p1
>= bufp
->buffer
&& p1
< pend
4707 && p2
>= bufp
->buffer
&& p2
<= pend
);
4709 op2
= p2
== pend
? succeed
: *p2
;
4715 /* If we're at the end of the pattern, we can change. */
4716 if (skip_one_char (p1
))
4718 DEBUG_PRINT (" End of pattern: fast loop.\n");
4726 register re_wchar_t c
4727 = (re_opcode_t
) *p2
== endline
? '\n'
4728 : RE_STRING_CHAR (p2
+ 2, multibyte
);
4730 if ((re_opcode_t
) *p1
== exactn
)
4732 if (c
!= RE_STRING_CHAR (p1
+ 2, multibyte
))
4734 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c
, p1
[2]);
4739 else if ((re_opcode_t
) *p1
== charset
4740 || (re_opcode_t
) *p1
== charset_not
)
4742 if (!execute_charset (&p1
, c
, c
, !multibyte
|| IS_REAL_ASCII (c
)))
4744 DEBUG_PRINT (" No match => fast loop.\n");
4748 else if ((re_opcode_t
) *p1
== anychar
4751 DEBUG_PRINT (" . != \\n => fast loop.\n");
4759 if ((re_opcode_t
) *p1
== exactn
)
4760 /* Reuse the code above. */
4761 return mutually_exclusive_p (bufp
, p2
, p1
);
4763 /* It is hard to list up all the character in charset
4764 P2 if it includes multibyte character. Give up in
4766 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
4768 /* Now, we are sure that P2 has no range table.
4769 So, for the size of bitmap in P2, `p2[1]' is
4770 enough. But P1 may have range table, so the
4771 size of bitmap table of P1 is extracted by
4772 using macro `CHARSET_BITMAP_SIZE'.
4774 In a multibyte case, we know that all the character
4775 listed in P2 is ASCII. In a unibyte case, P1 has only a
4776 bitmap table. So, in both cases, it is enough to test
4777 only the bitmap table of P1. */
4779 if ((re_opcode_t
) *p1
== charset
)
4782 /* We win if the charset inside the loop
4783 has no overlap with the one after the loop. */
4786 && idx
< CHARSET_BITMAP_SIZE (p1
));
4788 if ((p2
[2 + idx
] & p1
[2 + idx
]) != 0)
4792 || idx
== CHARSET_BITMAP_SIZE (p1
))
4794 DEBUG_PRINT (" No match => fast loop.\n");
4798 else if ((re_opcode_t
) *p1
== charset_not
)
4801 /* We win if the charset_not inside the loop lists
4802 every character listed in the charset after. */
4803 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4804 if (! (p2
[2 + idx
] == 0
4805 || (idx
< CHARSET_BITMAP_SIZE (p1
)
4806 && ((p2
[2 + idx
] & ~ p1
[2 + idx
]) == 0))))
4811 DEBUG_PRINT (" No match => fast loop.\n");
4824 /* Reuse the code above. */
4825 return mutually_exclusive_p (bufp
, p2
, p1
);
4827 /* When we have two charset_not, it's very unlikely that
4828 they don't overlap. The union of the two sets of excluded
4829 chars should cover all possible chars, which, as a matter of
4830 fact, is virtually impossible in multibyte buffers. */
4836 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == Sword
);
4838 return ((re_opcode_t
) *p1
== syntaxspec
4839 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4841 return ((re_opcode_t
) *p1
== syntaxspec
&& p1
[1] == p2
[1]);
4844 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == Sword
);
4846 return ((re_opcode_t
) *p1
== notsyntaxspec
4847 && (p1
[1] == Ssymbol
|| p1
[1] == Sword
));
4849 return ((re_opcode_t
) *p1
== notsyntaxspec
&& p1
[1] == p2
[1]);
4852 return (((re_opcode_t
) *p1
== notsyntaxspec
4853 || (re_opcode_t
) *p1
== syntaxspec
)
4858 return ((re_opcode_t
) *p1
== notcategoryspec
&& p1
[1] == p2
[1]);
4859 case notcategoryspec
:
4860 return ((re_opcode_t
) *p1
== categoryspec
&& p1
[1] == p2
[1]);
4872 /* Matching routines. */
4874 #ifndef emacs /* Emacs never uses this. */
4875 /* re_match is like re_match_2 except it takes only a single string. */
4878 re_match (struct re_pattern_buffer
*bufp
, const char *string
,
4879 size_t size
, ssize_t pos
, struct re_registers
*regs
)
4881 regoff_t result
= re_match_2_internal (bufp
, NULL
, 0, (re_char
*) string
,
4882 size
, pos
, regs
, size
);
4885 WEAK_ALIAS (__re_match
, re_match
)
4886 #endif /* not emacs */
4889 /* In Emacs, this is the string or buffer in which we are matching.
4890 See the declaration in regex.h for details. */
4891 Lisp_Object re_match_object
;
4894 /* re_match_2 matches the compiled pattern in BUFP against the
4895 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4896 and SIZE2, respectively). We start matching at POS, and stop
4899 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4900 store offsets for the substring each group matched in REGS. See the
4901 documentation for exactly how many groups we fill.
4903 We return -1 if no match, -2 if an internal error (such as the
4904 failure stack overflowing). Otherwise, we return the length of the
4905 matched substring. */
4908 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
,
4909 size_t size1
, const char *string2
, size_t size2
, ssize_t pos
,
4910 struct re_registers
*regs
, ssize_t stop
)
4916 gl_state
.object
= re_match_object
; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4917 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos
));
4918 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4921 result
= re_match_2_internal (bufp
, (re_char
*) string1
, size1
,
4922 (re_char
*) string2
, size2
,
4926 WEAK_ALIAS (__re_match_2
, re_match_2
)
4929 /* This is a separate function so that we can force an alloca cleanup
4932 re_match_2_internal (struct re_pattern_buffer
*bufp
, const_re_char
*string1
,
4933 size_t size1
, const_re_char
*string2
, size_t size2
,
4934 ssize_t pos
, struct re_registers
*regs
, ssize_t stop
)
4936 /* General temporaries. */
4940 /* Just past the end of the corresponding string. */
4941 re_char
*end1
, *end2
;
4943 /* Pointers into string1 and string2, just past the last characters in
4944 each to consider matching. */
4945 re_char
*end_match_1
, *end_match_2
;
4947 /* Where we are in the data, and the end of the current string. */
4950 /* Used sometimes to remember where we were before starting matching
4951 an operator so that we can go back in case of failure. This "atomic"
4952 behavior of matching opcodes is indispensable to the correctness
4953 of the on_failure_keep_string_jump optimization. */
4956 /* Where we are in the pattern, and the end of the pattern. */
4957 re_char
*p
= bufp
->buffer
;
4958 re_char
*pend
= p
+ bufp
->used
;
4960 /* We use this to map every character in the string. */
4961 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4963 /* Nonzero if BUFP is setup from a multibyte regex. */
4964 const boolean multibyte
= RE_MULTIBYTE_P (bufp
);
4966 /* Nonzero if STRING1/STRING2 are multibyte. */
4967 const boolean target_multibyte
= RE_TARGET_MULTIBYTE_P (bufp
);
4969 /* Failure point stack. Each place that can handle a failure further
4970 down the line pushes a failure point on this stack. It consists of
4971 regstart, and regend for all registers corresponding to
4972 the subexpressions we're currently inside, plus the number of such
4973 registers, and, finally, two char *'s. The first char * is where
4974 to resume scanning the pattern; the second one is where to resume
4975 scanning the strings. */
4976 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4977 fail_stack_type fail_stack
;
4979 #ifdef DEBUG_COMPILES_ARGUMENTS
4980 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4983 #if defined REL_ALLOC && defined REGEX_MALLOC
4984 /* This holds the pointer to the failure stack, when
4985 it is allocated relocatably. */
4986 fail_stack_elt_t
*failure_stack_ptr
;
4989 /* We fill all the registers internally, independent of what we
4990 return, for use in backreferences. The number here includes
4991 an element for register zero. */
4992 size_t num_regs
= bufp
->re_nsub
+ 1;
4994 /* Information on the contents of registers. These are pointers into
4995 the input strings; they record just what was matched (on this
4996 attempt) by a subexpression part of the pattern, that is, the
4997 regnum-th regstart pointer points to where in the pattern we began
4998 matching and the regnum-th regend points to right after where we
4999 stopped matching the regnum-th subexpression. (The zeroth register
5000 keeps track of what the whole pattern matches.) */
5001 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5002 re_char
**regstart
, **regend
;
5005 /* The following record the register info as found in the above
5006 variables when we find a match better than any we've seen before.
5007 This happens as we backtrack through the failure points, which in
5008 turn happens only if we have not yet matched the entire string. */
5009 unsigned best_regs_set
= false;
5010 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5011 re_char
**best_regstart
, **best_regend
;
5014 /* Logically, this is `best_regend[0]'. But we don't want to have to
5015 allocate space for that if we're not allocating space for anything
5016 else (see below). Also, we never need info about register 0 for
5017 any of the other register vectors, and it seems rather a kludge to
5018 treat `best_regend' differently than the rest. So we keep track of
5019 the end of the best match so far in a separate variable. We
5020 initialize this to NULL so that when we backtrack the first time
5021 and need to test it, it's not garbage. */
5022 re_char
*match_end
= NULL
;
5024 #ifdef DEBUG_COMPILES_ARGUMENTS
5025 /* Counts the total number of registers pushed. */
5026 unsigned num_regs_pushed
= 0;
5029 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5031 REGEX_USE_SAFE_ALLOCA
;
5035 #ifdef MATCH_MAY_ALLOCATE
5036 /* Do not bother to initialize all the register variables if there are
5037 no groups in the pattern, as it takes a fair amount of time. If
5038 there are groups, we include space for register 0 (the whole
5039 pattern), even though we never use it, since it simplifies the
5040 array indexing. We should fix this. */
5043 regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5044 regend
= REGEX_TALLOC (num_regs
, re_char
*);
5045 best_regstart
= REGEX_TALLOC (num_regs
, re_char
*);
5046 best_regend
= REGEX_TALLOC (num_regs
, re_char
*);
5048 if (!(regstart
&& regend
&& best_regstart
&& best_regend
))
5056 /* We must initialize all our variables to NULL, so that
5057 `FREE_VARIABLES' doesn't try to free them. */
5058 regstart
= regend
= best_regstart
= best_regend
= NULL
;
5060 #endif /* MATCH_MAY_ALLOCATE */
5062 /* The starting position is bogus. */
5063 if (pos
< 0 || pos
> size1
+ size2
)
5069 /* Initialize subexpression text positions to -1 to mark ones that no
5070 start_memory/stop_memory has been seen for. Also initialize the
5071 register information struct. */
5072 for (reg
= 1; reg
< num_regs
; reg
++)
5073 regstart
[reg
] = regend
[reg
] = NULL
;
5075 /* We move `string1' into `string2' if the latter's empty -- but not if
5076 `string1' is null. */
5077 if (size2
== 0 && string1
!= NULL
)
5084 end1
= string1
+ size1
;
5085 end2
= string2
+ size2
;
5087 /* `p' scans through the pattern as `d' scans through the data.
5088 `dend' is the end of the input string that `d' points within. `d'
5089 is advanced into the following input string whenever necessary, but
5090 this happens before fetching; therefore, at the beginning of the
5091 loop, `d' can be pointing at the end of a string, but it cannot
5095 /* Only match within string2. */
5096 d
= string2
+ pos
- size1
;
5097 dend
= end_match_2
= string2
+ stop
- size1
;
5098 end_match_1
= end1
; /* Just to give it a value. */
5104 /* Only match within string1. */
5105 end_match_1
= string1
+ stop
;
5107 When we reach end_match_1, PREFETCH normally switches to string2.
5108 But in the present case, this means that just doing a PREFETCH
5109 makes us jump from `stop' to `gap' within the string.
5110 What we really want here is for the search to stop as
5111 soon as we hit end_match_1. That's why we set end_match_2
5112 to end_match_1 (since PREFETCH fails as soon as we hit
5114 end_match_2
= end_match_1
;
5117 { /* It's important to use this code when stop == size so that
5118 moving `d' from end1 to string2 will not prevent the d == dend
5119 check from catching the end of string. */
5121 end_match_2
= string2
+ stop
- size1
;
5127 DEBUG_PRINT ("The compiled pattern is: ");
5128 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5129 DEBUG_PRINT ("The string to match is: \"");
5130 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5131 DEBUG_PRINT ("\"\n");
5133 /* This loops over pattern commands. It exits by returning from the
5134 function if the match is complete, or it drops through if the match
5135 fails at this starting point in the input data. */
5138 DEBUG_PRINT ("\n%p: ", p
);
5142 /* End of pattern means we might have succeeded. */
5143 DEBUG_PRINT ("end of pattern ... ");
5145 /* If we haven't matched the entire string, and we want the
5146 longest match, try backtracking. */
5147 if (d
!= end_match_2
)
5149 /* True if this match is the best seen so far. */
5153 /* True if this match ends in the same string (string1
5154 or string2) as the best previous match. */
5155 bool same_str_p
= (FIRST_STRING_P (match_end
)
5156 == FIRST_STRING_P (d
));
5158 /* AIX compiler got confused when this was combined
5159 with the previous declaration. */
5161 best_match_p
= d
> match_end
;
5163 best_match_p
= !FIRST_STRING_P (d
);
5166 DEBUG_PRINT ("backtracking.\n");
5168 if (!FAIL_STACK_EMPTY ())
5169 { /* More failure points to try. */
5171 /* If exceeds best match so far, save it. */
5172 if (!best_regs_set
|| best_match_p
)
5174 best_regs_set
= true;
5177 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5179 for (reg
= 1; reg
< num_regs
; reg
++)
5181 best_regstart
[reg
] = regstart
[reg
];
5182 best_regend
[reg
] = regend
[reg
];
5188 /* If no failure points, don't restore garbage. And if
5189 last match is real best match, don't restore second
5191 else if (best_regs_set
&& !best_match_p
)
5194 /* Restore best match. It may happen that `dend ==
5195 end_match_1' while the restored d is in string2.
5196 For example, the pattern `x.*y.*z' against the
5197 strings `x-' and `y-z-', if the two strings are
5198 not consecutive in memory. */
5199 DEBUG_PRINT ("Restoring best registers.\n");
5202 dend
= ((d
>= string1
&& d
<= end1
)
5203 ? end_match_1
: end_match_2
);
5205 for (reg
= 1; reg
< num_regs
; reg
++)
5207 regstart
[reg
] = best_regstart
[reg
];
5208 regend
[reg
] = best_regend
[reg
];
5211 } /* d != end_match_2 */
5214 DEBUG_PRINT ("Accepting match.\n");
5216 /* If caller wants register contents data back, do it. */
5217 if (regs
&& !bufp
->no_sub
)
5219 /* Have the register data arrays been allocated? */
5220 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
5221 { /* No. So allocate them with malloc. We need one
5222 extra element beyond `num_regs' for the `-1' marker
5224 regs
->num_regs
= max (RE_NREGS
, num_regs
+ 1);
5225 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
5226 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
5227 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5232 bufp
->regs_allocated
= REGS_REALLOCATE
;
5234 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
5235 { /* Yes. If we need more elements than were already
5236 allocated, reallocate them. If we need fewer, just
5238 if (regs
->num_regs
< num_regs
+ 1)
5240 regs
->num_regs
= num_regs
+ 1;
5241 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
5242 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
5243 if (regs
->start
== NULL
|| regs
->end
== NULL
)
5252 /* These braces fend off a "empty body in an else-statement"
5253 warning under GCC when assert expands to nothing. */
5254 assert (bufp
->regs_allocated
== REGS_FIXED
);
5257 /* Convert the pointer data in `regstart' and `regend' to
5258 indices. Register zero has to be set differently,
5259 since we haven't kept track of any info for it. */
5260 if (regs
->num_regs
> 0)
5262 regs
->start
[0] = pos
;
5263 regs
->end
[0] = POINTER_TO_OFFSET (d
);
5266 /* Go through the first `min (num_regs, regs->num_regs)'
5267 registers, since that is all we initialized. */
5268 for (reg
= 1; reg
< min (num_regs
, regs
->num_regs
); reg
++)
5270 if (REG_UNSET (regstart
[reg
]) || REG_UNSET (regend
[reg
]))
5271 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5274 regs
->start
[reg
] = POINTER_TO_OFFSET (regstart
[reg
]);
5275 regs
->end
[reg
] = POINTER_TO_OFFSET (regend
[reg
]);
5279 /* If the regs structure we return has more elements than
5280 were in the pattern, set the extra elements to -1. If
5281 we (re)allocated the registers, this is the case,
5282 because we always allocate enough to have at least one
5284 for (reg
= num_regs
; reg
< regs
->num_regs
; reg
++)
5285 regs
->start
[reg
] = regs
->end
[reg
] = -1;
5286 } /* regs && !bufp->no_sub */
5288 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5289 nfailure_points_pushed
, nfailure_points_popped
,
5290 nfailure_points_pushed
- nfailure_points_popped
);
5291 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed
);
5293 ptrdiff_t dcnt
= POINTER_TO_OFFSET (d
) - pos
;
5295 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt
);
5301 /* Otherwise match next pattern command. */
5304 /* Ignore these. Used to ignore the n of succeed_n's which
5305 currently have n == 0. */
5307 DEBUG_PRINT ("EXECUTING no_op.\n");
5311 DEBUG_PRINT ("EXECUTING succeed.\n");
5314 /* Match the next n pattern characters exactly. The following
5315 byte in the pattern defines n, and the n bytes after that
5316 are the characters to match. */
5319 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt
);
5321 /* Remember the start point to rollback upon failure. */
5325 /* This is written out as an if-else so we don't waste time
5326 testing `translate' inside the loop. */
5327 if (RE_TRANSLATE_P (translate
))
5331 if (RE_TRANSLATE (translate
, *d
) != *p
++)
5351 /* The cost of testing `translate' is comparatively small. */
5352 if (target_multibyte
)
5355 int pat_charlen
, buf_charlen
;
5360 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5363 pat_ch
= RE_CHAR_TO_MULTIBYTE (*p
);
5366 buf_ch
= STRING_CHAR_AND_LENGTH (d
, buf_charlen
);
5368 if (TRANSLATE (buf_ch
) != pat_ch
)
5376 mcnt
-= pat_charlen
;
5388 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pat_charlen
);
5389 pat_ch
= RE_CHAR_TO_UNIBYTE (pat_ch
);
5396 buf_ch
= RE_CHAR_TO_MULTIBYTE (*d
);
5397 if (! CHAR_BYTE8_P (buf_ch
))
5399 buf_ch
= TRANSLATE (buf_ch
);
5400 buf_ch
= RE_CHAR_TO_UNIBYTE (buf_ch
);
5406 if (buf_ch
!= pat_ch
)
5419 /* Match any character except possibly a newline or a null. */
5424 reg_syntax_t syntax
;
5426 DEBUG_PRINT ("EXECUTING anychar.\n");
5429 buf_ch
= RE_STRING_CHAR_AND_LENGTH (d
, buf_charlen
,
5431 buf_ch
= TRANSLATE (buf_ch
);
5434 syntax
= RE_SYNTAX_EMACS
;
5436 syntax
= bufp
->syntax
;
5439 if ((!(syntax
& RE_DOT_NEWLINE
) && buf_ch
== '\n')
5440 || ((syntax
& RE_DOT_NOT_NULL
) && buf_ch
== '\000'))
5443 DEBUG_PRINT (" Matched \"%d\".\n", *d
);
5452 register unsigned int c
, corig
;
5455 /* Whether matching against a unibyte character. */
5456 boolean unibyte_char
= false;
5458 DEBUG_PRINT ("EXECUTING charset%s.\n",
5459 (re_opcode_t
) *(p
- 1) == charset_not
? "_not" : "");
5462 corig
= c
= RE_STRING_CHAR_AND_LENGTH (d
, len
, target_multibyte
);
5463 if (target_multibyte
)
5468 c1
= RE_CHAR_TO_UNIBYTE (c
);
5471 unibyte_char
= true;
5477 int c1
= RE_CHAR_TO_MULTIBYTE (c
);
5479 if (! CHAR_BYTE8_P (c1
))
5481 c1
= TRANSLATE (c1
);
5482 c1
= RE_CHAR_TO_UNIBYTE (c1
);
5485 unibyte_char
= true;
5490 unibyte_char
= true;
5494 if (!execute_charset (&p
, c
, corig
, unibyte_char
))
5502 /* The beginning of a group is represented by start_memory.
5503 The argument is the register number. The text
5504 matched within the group is recorded (in the internal
5505 registers data structure) under the register number. */
5507 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p
);
5509 /* In case we need to undo this operation (via backtracking). */
5510 PUSH_FAILURE_REG (*p
);
5513 regend
[*p
] = NULL
; /* probably unnecessary. -sm */
5514 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart
[*p
]));
5516 /* Move past the register number and inner group count. */
5521 /* The stop_memory opcode represents the end of a group. Its
5522 argument is the same as start_memory's: the register number. */
5524 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p
);
5526 assert (!REG_UNSET (regstart
[*p
]));
5527 /* Strictly speaking, there should be code such as:
5529 assert (REG_UNSET (regend[*p]));
5530 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5532 But the only info to be pushed is regend[*p] and it is known to
5533 be UNSET, so there really isn't anything to push.
5534 Not pushing anything, on the other hand deprives us from the
5535 guarantee that regend[*p] is UNSET since undoing this operation
5536 will not reset its value properly. This is not important since
5537 the value will only be read on the next start_memory or at
5538 the very end and both events can only happen if this stop_memory
5542 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend
[*p
]));
5544 /* Move past the register number and the inner group count. */
5549 /* \<digit> has been turned into a `duplicate' command which is
5550 followed by the numeric value of <digit> as the register number. */
5553 register re_char
*d2
, *dend2
;
5554 int regno
= *p
++; /* Get which register to match against. */
5555 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno
);
5557 /* Can't back reference a group which we've never matched. */
5558 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
5561 /* Where in input to try to start matching. */
5562 d2
= regstart
[regno
];
5564 /* Remember the start point to rollback upon failure. */
5567 /* Where to stop matching; if both the place to start and
5568 the place to stop matching are in the same string, then
5569 set to the place to stop, otherwise, for now have to use
5570 the end of the first string. */
5572 dend2
= ((FIRST_STRING_P (regstart
[regno
])
5573 == FIRST_STRING_P (regend
[regno
]))
5574 ? regend
[regno
] : end_match_1
);
5579 /* If necessary, advance to next segment in register
5583 if (dend2
== end_match_2
) break;
5584 if (dend2
== regend
[regno
]) break;
5586 /* End of string1 => advance to string2. */
5588 dend2
= regend
[regno
];
5590 /* At end of register contents => success */
5591 if (d2
== dend2
) break;
5593 /* If necessary, advance to next segment in data. */
5596 /* How many characters left in this segment to match. */
5599 /* Want how many consecutive characters we can match in
5600 one shot, so, if necessary, adjust the count. */
5601 if (dcnt
> dend2
- d2
)
5604 /* Compare that many; failure if mismatch, else move
5606 if (RE_TRANSLATE_P (translate
)
5607 ? bcmp_translate (d
, d2
, dcnt
, translate
, target_multibyte
)
5608 : memcmp (d
, d2
, dcnt
))
5613 d
+= dcnt
, d2
+= dcnt
;
5619 /* begline matches the empty string at the beginning of the string
5620 (unless `not_bol' is set in `bufp'), and after newlines. */
5622 DEBUG_PRINT ("EXECUTING begline.\n");
5624 if (AT_STRINGS_BEG (d
))
5626 if (!bufp
->not_bol
) break;
5631 GET_CHAR_BEFORE_2 (c
, d
, string1
, end1
, string2
, end2
);
5635 /* In all other cases, we fail. */
5639 /* endline is the dual of begline. */
5641 DEBUG_PRINT ("EXECUTING endline.\n");
5643 if (AT_STRINGS_END (d
))
5645 if (!bufp
->not_eol
) break;
5649 PREFETCH_NOLIMIT ();
5656 /* Match at the very beginning of the data. */
5658 DEBUG_PRINT ("EXECUTING begbuf.\n");
5659 if (AT_STRINGS_BEG (d
))
5664 /* Match at the very end of the data. */
5666 DEBUG_PRINT ("EXECUTING endbuf.\n");
5667 if (AT_STRINGS_END (d
))
5672 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5673 pushes NULL as the value for the string on the stack. Then
5674 `POP_FAILURE_POINT' will keep the current value for the
5675 string, instead of restoring it. To see why, consider
5676 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5677 then the . fails against the \n. But the next thing we want
5678 to do is match the \n against the \n; if we restored the
5679 string value, we would be back at the foo.
5681 Because this is used only in specific cases, we don't need to
5682 check all the things that `on_failure_jump' does, to make
5683 sure the right things get saved on the stack. Hence we don't
5684 share its code. The only reason to push anything on the
5685 stack at all is that otherwise we would have to change
5686 `anychar's code to do something besides goto fail in this
5687 case; that seems worse than this. */
5688 case on_failure_keep_string_jump
:
5689 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5690 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5693 PUSH_FAILURE_POINT (p
- 3, NULL
);
5696 /* A nasty loop is introduced by the non-greedy *? and +?.
5697 With such loops, the stack only ever contains one failure point
5698 at a time, so that a plain on_failure_jump_loop kind of
5699 cycle detection cannot work. Worse yet, such a detection
5700 can not only fail to detect a cycle, but it can also wrongly
5701 detect a cycle (between different instantiations of the same
5703 So the method used for those nasty loops is a little different:
5704 We use a special cycle-detection-stack-frame which is pushed
5705 when the on_failure_jump_nastyloop failure-point is *popped*.
5706 This special frame thus marks the beginning of one iteration
5707 through the loop and we can hence easily check right here
5708 whether something matched between the beginning and the end of
5710 case on_failure_jump_nastyloop
:
5711 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5712 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5715 assert ((re_opcode_t
)p
[-4] == no_op
);
5718 CHECK_INFINITE_LOOP (p
- 4, d
);
5720 /* If there's a cycle, just continue without pushing
5721 this failure point. The failure point is the "try again"
5722 option, which shouldn't be tried.
5723 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5724 PUSH_FAILURE_POINT (p
- 3, d
);
5728 /* Simple loop detecting on_failure_jump: just check on the
5729 failure stack if the same spot was already hit earlier. */
5730 case on_failure_jump_loop
:
5732 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5733 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5737 CHECK_INFINITE_LOOP (p
- 3, d
);
5739 /* If there's a cycle, get out of the loop, as if the matching
5740 had failed. We used to just `goto fail' here, but that was
5741 aborting the search a bit too early: we want to keep the
5742 empty-loop-match and keep matching after the loop.
5743 We want (x?)*y\1z to match both xxyz and xxyxz. */
5746 PUSH_FAILURE_POINT (p
- 3, d
);
5751 /* Uses of on_failure_jump:
5753 Each alternative starts with an on_failure_jump that points
5754 to the beginning of the next alternative. Each alternative
5755 except the last ends with a jump that in effect jumps past
5756 the rest of the alternatives. (They really jump to the
5757 ending jump of the following alternative, because tensioning
5758 these jumps is a hassle.)
5760 Repeats start with an on_failure_jump that points past both
5761 the repetition text and either the following jump or
5762 pop_failure_jump back to this on_failure_jump. */
5763 case on_failure_jump
:
5764 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5765 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5768 PUSH_FAILURE_POINT (p
-3, d
);
5771 /* This operation is used for greedy *.
5772 Compare the beginning of the repeat with what in the
5773 pattern follows its end. If we can establish that there
5774 is nothing that they would both match, i.e., that we
5775 would have to backtrack because of (as in, e.g., `a*a')
5776 then we can use a non-backtracking loop based on
5777 on_failure_keep_string_jump instead of on_failure_jump. */
5778 case on_failure_jump_smart
:
5779 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5780 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5783 re_char
*p1
= p
; /* Next operation. */
5784 /* Here, we discard `const', making re_match non-reentrant. */
5785 unsigned char *p2
= (unsigned char*) p
+ mcnt
; /* Jump dest. */
5786 unsigned char *p3
= (unsigned char*) p
- 3; /* opcode location. */
5788 p
-= 3; /* Reset so that we will re-execute the
5789 instruction once it's been changed. */
5791 EXTRACT_NUMBER (mcnt
, p2
- 2);
5793 /* Ensure this is a indeed the trivial kind of loop
5794 we are expecting. */
5795 assert (skip_one_char (p1
) == p2
- 3);
5796 assert ((re_opcode_t
) p2
[-3] == jump
&& p2
+ mcnt
== p
);
5797 DEBUG_STATEMENT (debug
+= 2);
5798 if (mutually_exclusive_p (bufp
, p1
, p2
))
5800 /* Use a fast `on_failure_keep_string_jump' loop. */
5801 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5802 *p3
= (unsigned char) on_failure_keep_string_jump
;
5803 STORE_NUMBER (p2
- 2, mcnt
+ 3);
5807 /* Default to a safe `on_failure_jump' loop. */
5808 DEBUG_PRINT (" smart default => slow loop.\n");
5809 *p3
= (unsigned char) on_failure_jump
;
5811 DEBUG_STATEMENT (debug
-= 2);
5815 /* Unconditionally jump (without popping any failure points). */
5818 IMMEDIATE_QUIT_CHECK
;
5819 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5820 DEBUG_PRINT ("EXECUTING jump %d ", mcnt
);
5821 p
+= mcnt
; /* Do the jump. */
5822 DEBUG_PRINT ("(to %p).\n", p
);
5826 /* Have to succeed matching what follows at least n times.
5827 After that, handle like `on_failure_jump'. */
5829 /* Signedness doesn't matter since we only compare MCNT to 0. */
5830 EXTRACT_NUMBER (mcnt
, p
+ 2);
5831 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt
);
5833 /* Originally, mcnt is how many times we HAVE to succeed. */
5836 /* Here, we discard `const', making re_match non-reentrant. */
5837 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5840 PUSH_NUMBER (p2
, mcnt
);
5843 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5848 /* Signedness doesn't matter since we only compare MCNT to 0. */
5849 EXTRACT_NUMBER (mcnt
, p
+ 2);
5850 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt
);
5852 /* Originally, this is how many times we CAN jump. */
5855 /* Here, we discard `const', making re_match non-reentrant. */
5856 unsigned char *p2
= (unsigned char*) p
+ 2; /* counter loc. */
5858 PUSH_NUMBER (p2
, mcnt
);
5859 goto unconditional_jump
;
5861 /* If don't have to jump any more, skip over the rest of command. */
5868 unsigned char *p2
; /* Location of the counter. */
5869 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5871 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5872 /* Here, we discard `const', making re_match non-reentrant. */
5873 p2
= (unsigned char*) p
+ mcnt
;
5874 /* Signedness doesn't matter since we only copy MCNT's bits. */
5875 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5876 DEBUG_PRINT (" Setting %p to %d.\n", p2
, mcnt
);
5877 PUSH_NUMBER (p2
, mcnt
);
5884 boolean
not = (re_opcode_t
) *(p
- 1) == notwordbound
;
5885 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5887 /* We SUCCEED (or FAIL) in one of the following cases: */
5889 /* Case 1: D is at the beginning or the end of string. */
5890 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5894 /* C1 is the character before D, S1 is the syntax of C1, C2
5895 is the character at D, and S2 is the syntax of C2. */
5900 ssize_t offset
= PTR_TO_OFFSET (d
- 1);
5901 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5902 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5904 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5907 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
+ 1);
5909 PREFETCH_NOLIMIT ();
5910 GET_CHAR_AFTER (c2
, d
, dummy
);
5913 if (/* Case 2: Only one of S1 and S2 is Sword. */
5914 ((s1
== Sword
) != (s2
== Sword
))
5915 /* Case 3: Both of S1 and S2 are Sword, and macro
5916 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5917 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5927 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5929 /* We FAIL in one of the following cases: */
5931 /* Case 1: D is at the end of string. */
5932 if (AT_STRINGS_END (d
))
5936 /* C1 is the character before D, S1 is the syntax of C1, C2
5937 is the character at D, and S2 is the syntax of C2. */
5942 ssize_t offset
= PTR_TO_OFFSET (d
);
5943 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5944 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5947 GET_CHAR_AFTER (c2
, d
, dummy
);
5950 /* Case 2: S2 is not Sword. */
5954 /* Case 3: D is not at the beginning of string ... */
5955 if (!AT_STRINGS_BEG (d
))
5957 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5959 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5963 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5965 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5972 DEBUG_PRINT ("EXECUTING wordend.\n");
5974 /* We FAIL in one of the following cases: */
5976 /* Case 1: D is at the beginning of string. */
5977 if (AT_STRINGS_BEG (d
))
5981 /* C1 is the character before D, S1 is the syntax of C1, C2
5982 is the character at D, and S2 is the syntax of C2. */
5987 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
5988 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
5989 UPDATE_SYNTAX_TABLE_FAST (charpos
);
5991 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5994 /* Case 2: S1 is not Sword. */
5998 /* Case 3: D is not at the end of string ... */
5999 if (!AT_STRINGS_END (d
))
6001 PREFETCH_NOLIMIT ();
6002 GET_CHAR_AFTER (c2
, d
, dummy
);
6004 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
);
6008 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6010 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
6017 DEBUG_PRINT ("EXECUTING symbeg.\n");
6019 /* We FAIL in one of the following cases: */
6021 /* Case 1: D is at the end of string. */
6022 if (AT_STRINGS_END (d
))
6026 /* C1 is the character before D, S1 is the syntax of C1, C2
6027 is the character at D, and S2 is the syntax of C2. */
6031 ssize_t offset
= PTR_TO_OFFSET (d
);
6032 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6033 UPDATE_SYNTAX_TABLE_FAST (charpos
);
6036 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6039 /* Case 2: S2 is neither Sword nor Ssymbol. */
6040 if (s2
!= Sword
&& s2
!= Ssymbol
)
6043 /* Case 3: D is not at the beginning of string ... */
6044 if (!AT_STRINGS_BEG (d
))
6046 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6048 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
6052 /* ... and S1 is Sword or Ssymbol. */
6053 if (s1
== Sword
|| s1
== Ssymbol
)
6060 DEBUG_PRINT ("EXECUTING symend.\n");
6062 /* We FAIL in one of the following cases: */
6064 /* Case 1: D is at the beginning of string. */
6065 if (AT_STRINGS_BEG (d
))
6069 /* C1 is the character before D, S1 is the syntax of C1, C2
6070 is the character at D, and S2 is the syntax of C2. */
6074 ssize_t offset
= PTR_TO_OFFSET (d
) - 1;
6075 ssize_t charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6076 UPDATE_SYNTAX_TABLE_FAST (charpos
);
6078 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
6081 /* Case 2: S1 is neither Ssymbol nor Sword. */
6082 if (s1
!= Sword
&& s1
!= Ssymbol
)
6085 /* Case 3: D is not at the end of string ... */
6086 if (!AT_STRINGS_END (d
))
6088 PREFETCH_NOLIMIT ();
6089 c2
= RE_STRING_CHAR (d
, target_multibyte
);
6091 UPDATE_SYNTAX_TABLE_FORWARD_FAST (charpos
+ 1);
6095 /* ... and S2 is Sword or Ssymbol. */
6096 if (s2
== Sword
|| s2
== Ssymbol
)
6105 boolean
not = (re_opcode_t
) *(p
- 1) == notsyntaxspec
;
6107 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6112 ssize_t offset
= PTR_TO_OFFSET (d
);
6113 ssize_t pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (offset
);
6114 UPDATE_SYNTAX_TABLE_FAST (pos1
);
6121 GET_CHAR_AFTER (c
, d
, len
);
6122 if ((SYNTAX (c
) != (enum syntaxcode
) mcnt
) ^ not)
6131 DEBUG_PRINT ("EXECUTING at_dot.\n");
6132 if (PTR_BYTE_POS (d
) != PT_BYTE
)
6137 case notcategoryspec
:
6139 boolean
not = (re_opcode_t
) *(p
- 1) == notcategoryspec
;
6141 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6142 not ? "not" : "", mcnt
);
6148 GET_CHAR_AFTER (c
, d
, len
);
6149 if ((!CHAR_HAS_CATEGORY (c
, mcnt
)) ^ not)
6161 continue; /* Successfully executed one pattern command; keep going. */
6164 /* We goto here if a matching operation fails. */
6166 IMMEDIATE_QUIT_CHECK
;
6167 if (!FAIL_STACK_EMPTY ())
6170 /* A restart point is known. Restore to that state. */
6171 DEBUG_PRINT ("\nFAIL:\n");
6172 POP_FAILURE_POINT (str
, pat
);
6175 case on_failure_keep_string_jump
:
6176 assert (str
== NULL
);
6177 goto continue_failure_jump
;
6179 case on_failure_jump_nastyloop
:
6180 assert ((re_opcode_t
)pat
[-2] == no_op
);
6181 PUSH_FAILURE_POINT (pat
- 2, str
);
6184 case on_failure_jump_loop
:
6185 case on_failure_jump
:
6188 continue_failure_jump
:
6189 EXTRACT_NUMBER_AND_INCR (mcnt
, pat
);
6194 /* A special frame used for nastyloops. */
6201 assert (p
>= bufp
->buffer
&& p
<= pend
);
6203 if (d
>= string1
&& d
<= end1
)
6207 break; /* Matching at this starting point really fails. */
6211 goto restore_best_regs
;
6215 return -1; /* Failure to match. */
6218 /* Subroutine definitions for re_match_2. */
6220 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6221 bytes; nonzero otherwise. */
6224 bcmp_translate (const_re_char
*s1
, const_re_char
*s2
, register ssize_t len
,
6225 RE_TRANSLATE_TYPE translate
, const int target_multibyte
)
6227 register re_char
*p1
= s1
, *p2
= s2
;
6228 re_char
*p1_end
= s1
+ len
;
6229 re_char
*p2_end
= s2
+ len
;
6231 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6232 different lengths, but relying on a single `len' would break this. -sm */
6233 while (p1
< p1_end
&& p2
< p2_end
)
6235 int p1_charlen
, p2_charlen
;
6236 re_wchar_t p1_ch
, p2_ch
;
6238 GET_CHAR_AFTER (p1_ch
, p1
, p1_charlen
);
6239 GET_CHAR_AFTER (p2_ch
, p2
, p2_charlen
);
6241 if (RE_TRANSLATE (translate
, p1_ch
)
6242 != RE_TRANSLATE (translate
, p2_ch
))
6245 p1
+= p1_charlen
, p2
+= p2_charlen
;
6248 if (p1
!= p1_end
|| p2
!= p2_end
)
6254 /* Entry points for GNU code. */
6256 /* re_compile_pattern is the GNU regular expression compiler: it
6257 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6258 Returns 0 if the pattern was valid, otherwise an error string.
6260 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6261 are set in BUFP on entry.
6263 We call regex_compile to do the actual compilation. */
6266 re_compile_pattern (const char *pattern
, size_t length
,
6268 bool posix_backtracking
, const char *whitespace_regexp
,
6270 struct re_pattern_buffer
*bufp
)
6274 /* GNU code is written to assume at least RE_NREGS registers will be set
6275 (and at least one extra will be -1). */
6276 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6278 /* And GNU code determines whether or not to get register information
6279 by passing null for the REGS argument to re_match, etc., not by
6283 ret
= regex_compile ((re_char
*) pattern
, length
,
6294 return gettext (re_error_msgid
[(int) ret
]);
6296 WEAK_ALIAS (__re_compile_pattern
, re_compile_pattern
)
6298 /* Entry points compatible with 4.2 BSD regex library. We don't define
6299 them unless specifically requested. */
6301 #if defined _REGEX_RE_COMP || defined _LIBC
6303 /* BSD has one and only one pattern buffer. */
6304 static struct re_pattern_buffer re_comp_buf
;
6308 /* Make these definitions weak in libc, so POSIX programs can redefine
6309 these names if they don't use our functions, and still use
6310 regcomp/regexec below without link errors. */
6313 re_comp (const char *s
)
6319 if (!re_comp_buf
.buffer
)
6320 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6321 return (char *) gettext ("No previous regular expression");
6325 if (!re_comp_buf
.buffer
)
6327 re_comp_buf
.buffer
= malloc (200);
6328 if (re_comp_buf
.buffer
== NULL
)
6329 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6330 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6331 re_comp_buf
.allocated
= 200;
6333 re_comp_buf
.fastmap
= malloc (1 << BYTEWIDTH
);
6334 if (re_comp_buf
.fastmap
== NULL
)
6335 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6336 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
6339 /* Since `re_exec' always passes NULL for the `regs' argument, we
6340 don't need to initialize the pattern buffer fields which affect it. */
6342 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6347 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6348 return (char *) gettext (re_error_msgid
[(int) ret
]);
6356 re_exec (const char *s
)
6358 const size_t len
= strlen (s
);
6359 return re_search (&re_comp_buf
, s
, len
, 0, len
, 0) >= 0;
6361 #endif /* _REGEX_RE_COMP */
6363 /* POSIX.2 functions. Don't define these for Emacs. */
6367 /* regcomp takes a regular expression as a string and compiles it.
6369 PREG is a regex_t *. We do not expect any fields to be initialized,
6370 since POSIX says we shouldn't. Thus, we set
6372 `buffer' to the compiled pattern;
6373 `used' to the length of the compiled pattern;
6374 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6375 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6376 RE_SYNTAX_POSIX_BASIC;
6377 `fastmap' to an allocated space for the fastmap;
6378 `fastmap_accurate' to zero;
6379 `re_nsub' to the number of subexpressions in PATTERN.
6381 PATTERN is the address of the pattern string.
6383 CFLAGS is a series of bits which affect compilation.
6385 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6386 use POSIX basic syntax.
6388 If REG_NEWLINE is set, then . and [^...] don't match newline.
6389 Also, regexec will try a match beginning after every newline.
6391 If REG_ICASE is set, then we considers upper- and lowercase
6392 versions of letters to be equivalent when matching.
6394 If REG_NOSUB is set, then when PREG is passed to regexec, that
6395 routine will report only success or failure, and nothing about the
6398 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6399 the return codes and their meanings.) */
6402 regcomp (regex_t
*_Restrict_ preg
, const char *_Restrict_ pattern
,
6407 = (cflags
& REG_EXTENDED
) ?
6408 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6410 /* regex_compile will allocate the space for the compiled pattern. */
6412 preg
->allocated
= 0;
6415 /* Try to allocate space for the fastmap. */
6416 preg
->fastmap
= malloc (1 << BYTEWIDTH
);
6418 if (cflags
& REG_ICASE
)
6422 preg
->translate
= malloc (CHAR_SET_SIZE
* sizeof *preg
->translate
);
6423 if (preg
->translate
== NULL
)
6424 return (int) REG_ESPACE
;
6426 /* Map uppercase characters to corresponding lowercase ones. */
6427 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6428 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6431 preg
->translate
= NULL
;
6433 /* If REG_NEWLINE is set, newlines are treated differently. */
6434 if (cflags
& REG_NEWLINE
)
6435 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6436 syntax
&= ~RE_DOT_NEWLINE
;
6437 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6440 syntax
|= RE_NO_NEWLINE_ANCHOR
;
6442 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6444 /* POSIX says a null character in the pattern terminates it, so we
6445 can use strlen here in compiling the pattern. */
6446 ret
= regex_compile ((re_char
*) pattern
, strlen (pattern
), syntax
, preg
);
6448 /* POSIX doesn't distinguish between an unmatched open-group and an
6449 unmatched close-group: both are REG_EPAREN. */
6450 if (ret
== REG_ERPAREN
)
6453 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6454 { /* Compute the fastmap now, since regexec cannot modify the pattern
6456 re_compile_fastmap (preg
);
6457 if (preg
->can_be_null
)
6458 { /* The fastmap can't be used anyway. */
6459 free (preg
->fastmap
);
6460 preg
->fastmap
= NULL
;
6465 WEAK_ALIAS (__regcomp
, regcomp
)
6468 /* regexec searches for a given pattern, specified by PREG, in the
6471 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6472 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6473 least NMATCH elements, and we set them to the offsets of the
6474 corresponding matched substrings.
6476 EFLAGS specifies `execution flags' which affect matching: if
6477 REG_NOTBOL is set, then ^ does not match at the beginning of the
6478 string; if REG_NOTEOL is set, then $ does not match at the end.
6480 We return 0 if we find a match and REG_NOMATCH if not. */
6483 regexec (const regex_t
*_Restrict_ preg
, const char *_Restrict_ string
,
6484 size_t nmatch
, regmatch_t pmatch
[_Restrict_arr_
], int eflags
)
6487 struct re_registers regs
;
6488 regex_t private_preg
;
6489 size_t len
= strlen (string
);
6490 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0 && pmatch
;
6492 private_preg
= *preg
;
6494 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6495 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6497 /* The user has told us exactly how many registers to return
6498 information about, via `nmatch'. We have to pass that on to the
6499 matching routines. */
6500 private_preg
.regs_allocated
= REGS_FIXED
;
6504 regs
.num_regs
= nmatch
;
6505 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6506 if (regs
.start
== NULL
)
6508 regs
.end
= regs
.start
+ nmatch
;
6511 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6512 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6513 was a little bit longer but still only matching the real part.
6514 This works because the `endline' will check for a '\n' and will find a
6515 '\0', correctly deciding that this is not the end of a line.
6516 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6517 a convenient '\0' there. For all we know, the string could be preceded
6518 by '\n' which would throw things off. */
6520 /* Perform the searching operation. */
6521 ret
= re_search (&private_preg
, string
, len
,
6522 /* start: */ 0, /* range: */ len
,
6523 want_reg_info
? ®s
: 0);
6525 /* Copy the register information to the POSIX structure. */
6532 for (r
= 0; r
< nmatch
; r
++)
6534 pmatch
[r
].rm_so
= regs
.start
[r
];
6535 pmatch
[r
].rm_eo
= regs
.end
[r
];
6539 /* If we needed the temporary register info, free the space now. */
6543 /* We want zero return to mean success, unlike `re_search'. */
6544 return ret
>= 0 ? REG_NOERROR
: REG_NOMATCH
;
6546 WEAK_ALIAS (__regexec
, regexec
)
6549 /* Returns a message corresponding to an error code, ERR_CODE, returned
6550 from either regcomp or regexec. We don't use PREG here.
6552 ERR_CODE was previously called ERRCODE, but that name causes an
6553 error with msvc8 compiler. */
6556 regerror (int err_code
, const regex_t
*preg
, char *errbuf
, size_t errbuf_size
)
6562 || err_code
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6563 /* Only error codes returned by the rest of the code should be passed
6564 to this routine. If we are given anything else, or if other regex
6565 code generates an invalid error code, then the program has a bug.
6566 Dump core so we can fix it. */
6569 msg
= gettext (re_error_msgid
[err_code
]);
6571 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6573 if (errbuf_size
!= 0)
6575 if (msg_size
> errbuf_size
)
6577 memcpy (errbuf
, msg
, errbuf_size
- 1);
6578 errbuf
[errbuf_size
- 1] = 0;
6581 strcpy (errbuf
, msg
);
6586 WEAK_ALIAS (__regerror
, regerror
)
6589 /* Free dynamically allocated space used by PREG. */
6592 regfree (regex_t
*preg
)
6594 free (preg
->buffer
);
6595 preg
->buffer
= NULL
;
6597 preg
->allocated
= 0;
6600 free (preg
->fastmap
);
6601 preg
->fastmap
= NULL
;
6602 preg
->fastmap_accurate
= 0;
6604 free (preg
->translate
);
6605 preg
->translate
= NULL
;
6607 WEAK_ALIAS (__regfree
, regfree
)
6609 #endif /* not emacs */