1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993,94,95,96,97,98,2000 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace succeed_n + jump_n with a combined operation so that the counter
26 can simply be decremented when popping the failure_point without having
27 to stack up failure_count entries.
30 /* AIX requires this to be the first thing in the file. */
31 #if defined _AIX && !defined REGEX_MALLOC
42 /* We need this for `regex.h', and perhaps for the Emacs include files. */
43 #include <sys/types.h>
45 /* This is for other GNU distributions with internationalized messages. */
46 #if HAVE_LIBINTL_H || defined _LIBC
49 # define gettext(msgid) (msgid)
53 /* This define is so xgettext can find the internationalizable
55 # define gettext_noop(String) String
58 /* The `emacs' switch turns on certain matching commands
59 that make sense only in Emacs. */
65 /* Make syntax table lookup grant data in gl_state. */
66 # define SYNTAX_ENTRY_VIA_PROPERTY
70 # include "category.h"
72 # define malloc xmalloc
73 # define realloc xrealloc
76 /* Converts the pointer to the char to BEG-based offset from the start. */
77 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
78 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
80 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
81 # define RE_STRING_CHAR(p, s) \
82 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
83 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
84 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
86 /* Set C a (possibly multibyte) character before P. P points into a
87 string which is the virtual concatenation of STR1 (which ends at
88 END1) or STR2 (which ends at END2). */
89 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
93 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
94 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
95 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
96 c = STRING_CHAR (dtemp, (p) - dtemp); \
99 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
103 #else /* not emacs */
105 /* If we are not linking with Emacs proper,
106 we can't use the relocating allocator
107 even if config.h says that we can. */
110 # if defined STDC_HEADERS || defined _LIBC
117 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
118 If nothing else has been done, use the method below. */
119 # ifdef INHIBIT_STRING_HEADER
120 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
121 # if !defined bzero && !defined bcopy
122 # undef INHIBIT_STRING_HEADER
127 /* This is the normal way of making sure we have a bcopy and a bzero.
128 This is used in most programs--a few other programs avoid this
129 by defining INHIBIT_STRING_HEADER. */
130 # ifndef INHIBIT_STRING_HEADER
131 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
134 # define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
137 # define bcopy(s, d, n) memcpy ((d), (s), (n))
140 # define bzero(s, n) memset ((s), 0, (n))
143 # include <strings.h>
147 /* Define the syntax stuff for \<, \>, etc. */
149 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
150 enum syntaxcode { Swhitespace = 0, Sword = 1 };
152 # ifdef SWITCH_ENUM_BUG
153 # define SWITCH_ENUM_CAST(x) ((int)(x))
155 # define SWITCH_ENUM_CAST(x) (x)
158 # define SYNTAX(c) re_syntax_table[c]
160 /* Dummy macros for non-Emacs environments. */
161 # define BASE_LEADING_CODE_P(c) (0)
162 # define CHAR_CHARSET(c) 0
163 # define CHARSET_LEADING_CODE_BASE(c) 0
164 # define MAX_MULTIBYTE_LENGTH 1
165 # define RE_MULTIBYTE_P(x) 0
166 # define WORD_BOUNDARY_P(c1, c2) (0)
167 # define CHAR_HEAD_P(p) (1)
168 # define SINGLE_BYTE_CHAR_P(c) (1)
169 # define SAME_CHARSET_P(c1, c2) (1)
170 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
171 # define STRING_CHAR(p, s) (*(p))
172 # define RE_STRING_CHAR STRING_CHAR
173 # define CHAR_STRING(c, s) (*(s) = (c), 1)
174 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
175 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
176 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
177 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
178 # define MAKE_CHAR(charset, c1, c2) (c1)
179 #endif /* not emacs */
182 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
183 # define RE_TRANSLATE_P(TBL) (TBL)
186 /* Get the interface, including the syntax bits. */
189 /* isalpha etc. are used for the character classes. */
194 /* 1 if C is an ASCII character. */
195 # define IS_REAL_ASCII(c) ((c) < 0200)
197 /* 1 if C is a unibyte character. */
198 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
200 /* The Emacs definitions should not be directly affected by locales. */
202 /* In Emacs, these are only used for single-byte characters. */
203 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
204 # define ISCNTRL(c) ((c) < ' ')
205 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
206 || ((c) >= 'a' && (c) <= 'f') \
207 || ((c) >= 'A' && (c) <= 'F'))
209 /* This is only used for single-byte characters. */
210 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
212 /* The rest must handle multibyte characters. */
214 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
215 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
218 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
219 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
222 # define ISALNUM(c) (IS_REAL_ASCII (c) \
223 ? (((c) >= 'a' && (c) <= 'z') \
224 || ((c) >= 'A' && (c) <= 'Z') \
225 || ((c) >= '0' && (c) <= '9')) \
226 : SYNTAX (c) == Sword)
228 # define ISALPHA(c) (IS_REAL_ASCII (c) \
229 ? (((c) >= 'a' && (c) <= 'z') \
230 || ((c) >= 'A' && (c) <= 'Z')) \
231 : SYNTAX (c) == Sword)
233 # define ISLOWER(c) (LOWERCASEP (c))
235 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
236 ? ((c) > ' ' && (c) < 0177 \
237 && !(((c) >= 'a' && (c) <= 'z') \
238 || ((c) >= 'A' && (c) <= 'Z') \
239 || ((c) >= '0' && (c) <= '9'))) \
240 : SYNTAX (c) != Sword)
242 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
244 # define ISUPPER(c) (UPPERCASEP (c))
246 # define ISWORD(c) (SYNTAX (c) == Sword)
248 #else /* not emacs */
250 /* Jim Meyering writes:
252 "... Some ctype macros are valid only for character codes that
253 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
254 using /bin/cc or gcc but without giving an ansi option). So, all
255 ctype uses should be through macros like ISPRINT... If
256 STDC_HEADERS is defined, then autoconf has verified that the ctype
257 macros don't need to be guarded with references to isascii. ...
258 Defining isascii to 1 should let any compiler worth its salt
259 eliminate the && through constant folding." */
261 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
262 # define ISASCII(c) 1
264 # define ISASCII(c) isascii(c)
267 /* 1 if C is an ASCII character. */
268 # define IS_REAL_ASCII(c) ((c) < 0200)
270 /* This distinction is not meaningful, except in Emacs. */
271 # define ISUNIBYTE(c) 1
274 # define ISBLANK(c) (ISASCII (c) && isblank (c))
276 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
279 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
281 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
284 # define ISPRINT(c) (ISASCII (c) && isprint (c))
285 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
286 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
287 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
288 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
289 # define ISLOWER(c) (ISASCII (c) && islower (c))
290 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
291 # define ISSPACE(c) (ISASCII (c) && isspace (c))
292 # define ISUPPER(c) (ISASCII (c) && isupper (c))
293 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
295 # define ISWORD(c) ISALPHA(c)
299 extern char *re_syntax_table;
301 # else /* not SYNTAX_TABLE */
303 /* How many characters in the character set. */
304 # define CHAR_SET_SIZE 256
306 static char re_syntax_table[CHAR_SET_SIZE];
317 bzero (re_syntax_table, sizeof re_syntax_table);
319 for (c = 'a'; c <= 'z'; c++)
320 re_syntax_table[c] = Sword;
322 for (c = 'A'; c <= 'Z'; c++)
323 re_syntax_table[c] = Sword;
325 for (c = '0'; c <= '9'; c++)
326 re_syntax_table[c] = Sword;
328 re_syntax_table['_'] = Sword;
333 # endif /* not SYNTAX_TABLE */
335 #endif /* not emacs */
338 # define NULL (void *)0
341 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
342 since ours (we hope) works properly with all combinations of
343 machines, compilers, `char' and `unsigned char' argument types.
344 (Per Bothner suggested the basic approach.) */
345 #undef SIGN_EXTEND_CHAR
347 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
348 #else /* not __STDC__ */
349 /* As in Harbison and Steele. */
350 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
353 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
354 use `alloca' instead of `malloc'. This is because using malloc in
355 re_search* or re_match* could cause memory leaks when C-g is used in
356 Emacs; also, malloc is slower and causes storage fragmentation. On
357 the other hand, malloc is more portable, and easier to debug.
359 Because we sometimes use alloca, some routines have to be macros,
360 not functions -- `alloca'-allocated space disappears at the end of the
361 function it is called in. */
365 # define REGEX_ALLOCATE malloc
366 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
367 # define REGEX_FREE free
369 #else /* not REGEX_MALLOC */
371 /* Emacs already defines alloca, sometimes. */
374 /* Make alloca work the best possible way. */
376 # define alloca __builtin_alloca
377 # else /* not __GNUC__ */
380 # endif /* HAVE_ALLOCA_H */
381 # endif /* not __GNUC__ */
383 # endif /* not alloca */
385 # define REGEX_ALLOCATE alloca
387 /* Assumes a `char *destination' variable. */
388 # define REGEX_REALLOCATE(source, osize, nsize) \
389 (destination = (char *) alloca (nsize), \
390 bcopy (source, destination, osize), \
393 /* No need to do anything to free, after alloca. */
394 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
396 #endif /* not REGEX_MALLOC */
398 /* Define how to allocate the failure stack. */
400 #if defined REL_ALLOC && defined REGEX_MALLOC
402 # define REGEX_ALLOCATE_STACK(size) \
403 r_alloc (&failure_stack_ptr, (size))
404 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
405 r_re_alloc (&failure_stack_ptr, (nsize))
406 # define REGEX_FREE_STACK(ptr) \
407 r_alloc_free (&failure_stack_ptr)
409 #else /* not using relocating allocator */
413 # define REGEX_ALLOCATE_STACK malloc
414 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
415 # define REGEX_FREE_STACK free
417 # else /* not REGEX_MALLOC */
419 # define REGEX_ALLOCATE_STACK alloca
421 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
422 REGEX_REALLOCATE (source, osize, nsize)
423 /* No need to explicitly free anything. */
424 # define REGEX_FREE_STACK(arg) ((void)0)
426 # endif /* not REGEX_MALLOC */
427 #endif /* not using relocating allocator */
430 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
431 `string1' or just past its end. This works if PTR is NULL, which is
433 #define FIRST_STRING_P(ptr) \
434 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
436 /* (Re)Allocate N items of type T using malloc, or fail. */
437 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
438 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
439 #define RETALLOC_IF(addr, n, t) \
440 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
441 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
443 #define BYTEWIDTH 8 /* In bits. */
445 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
449 #define MAX(a, b) ((a) > (b) ? (a) : (b))
450 #define MIN(a, b) ((a) < (b) ? (a) : (b))
452 /* Type of source-pattern and string chars. */
453 typedef const unsigned char re_char;
455 typedef char boolean;
459 static int re_match_2_internal ();
461 /* These are the command codes that appear in compiled regular
462 expressions. Some opcodes are followed by argument bytes. A
463 command code can specify any interpretation whatsoever for its
464 arguments. Zero bytes may appear in the compiled regular expression. */
470 /* Succeed right away--no more backtracking. */
473 /* Followed by one byte giving n, then by n literal bytes. */
476 /* Matches any (more or less) character. */
479 /* Matches any one char belonging to specified set. First
480 following byte is number of bitmap bytes. Then come bytes
481 for a bitmap saying which chars are in. Bits in each byte
482 are ordered low-bit-first. A character is in the set if its
483 bit is 1. A character too large to have a bit in the map is
484 automatically not in the set.
486 If the length byte has the 0x80 bit set, then that stuff
487 is followed by a range table:
488 2 bytes of flags for character sets (low 8 bits, high 8 bits)
489 See RANGE_TABLE_WORK_BITS below.
490 2 bytes, the number of pairs that follow
491 pairs, each 2 multibyte characters,
492 each multibyte character represented as 3 bytes. */
495 /* Same parameters as charset, but match any character that is
496 not one of those specified. */
499 /* Start remembering the text that is matched, for storing in a
500 register. Followed by one byte with the register number, in
501 the range 0 to one less than the pattern buffer's re_nsub
505 /* Stop remembering the text that is matched and store it in a
506 memory register. Followed by one byte with the register
507 number, in the range 0 to one less than `re_nsub' in the
511 /* Match a duplicate of something remembered. Followed by one
512 byte containing the register number. */
515 /* Fail unless at beginning of line. */
518 /* Fail unless at end of line. */
521 /* Succeeds if at beginning of buffer (if emacs) or at beginning
522 of string to be matched (if not). */
525 /* Analogously, for end of buffer/string. */
528 /* Followed by two byte relative address to which to jump. */
531 /* Followed by two-byte relative address of place to resume at
532 in case of failure. */
535 /* Like on_failure_jump, but pushes a placeholder instead of the
536 current string position when executed. */
537 on_failure_keep_string_jump,
539 /* Just like `on_failure_jump', except that it checks that we
540 don't get stuck in an infinite loop (matching an empty string
542 on_failure_jump_loop,
544 /* Just like `on_failure_jump_loop', except that it checks for
545 a different kind of loop (the kind that shows up with non-greedy
546 operators). This operation has to be immediately preceded
548 on_failure_jump_nastyloop,
550 /* A smart `on_failure_jump' used for greedy * and + operators.
551 It analyses the loop before which it is put and if the
552 loop does not require backtracking, it changes itself to
553 `on_failure_keep_string_jump' and short-circuits the loop,
554 else it just defaults to changing itself into `on_failure_jump'.
555 It assumes that it is pointing to just past a `jump'. */
556 on_failure_jump_smart,
558 /* Followed by two-byte relative address and two-byte number n.
559 After matching N times, jump to the address upon failure.
560 Does not work if N starts at 0: use on_failure_jump_loop
564 /* Followed by two-byte relative address, and two-byte number n.
565 Jump to the address N times, then fail. */
568 /* Set the following two-byte relative address to the
569 subsequent two-byte number. The address *includes* the two
573 wordbeg, /* Succeeds if at word beginning. */
574 wordend, /* Succeeds if at word end. */
576 wordbound, /* Succeeds if at a word boundary. */
577 notwordbound, /* Succeeds if not at a word boundary. */
579 /* Matches any character whose syntax is specified. Followed by
580 a byte which contains a syntax code, e.g., Sword. */
583 /* Matches any character whose syntax is not that specified. */
587 ,before_dot, /* Succeeds if before point. */
588 at_dot, /* Succeeds if at point. */
589 after_dot, /* Succeeds if after point. */
591 /* Matches any character whose category-set contains the specified
592 category. The operator is followed by a byte which contains a
593 category code (mnemonic ASCII character). */
596 /* Matches any character whose category-set does not contain the
597 specified category. The operator is followed by a byte which
598 contains the category code (mnemonic ASCII character). */
603 /* Common operations on the compiled pattern. */
605 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
607 #define STORE_NUMBER(destination, number) \
609 (destination)[0] = (number) & 0377; \
610 (destination)[1] = (number) >> 8; \
613 /* Same as STORE_NUMBER, except increment DESTINATION to
614 the byte after where the number is stored. Therefore, DESTINATION
615 must be an lvalue. */
617 #define STORE_NUMBER_AND_INCR(destination, number) \
619 STORE_NUMBER (destination, number); \
620 (destination) += 2; \
623 /* Put into DESTINATION a number stored in two contiguous bytes starting
626 #define EXTRACT_NUMBER(destination, source) \
628 (destination) = *(source) & 0377; \
629 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
634 extract_number (dest, source)
636 unsigned char *source;
638 int temp = SIGN_EXTEND_CHAR (*(source + 1));
639 *dest = *source & 0377;
643 # ifndef EXTRACT_MACROS /* To debug the macros. */
644 # undef EXTRACT_NUMBER
645 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
646 # endif /* not EXTRACT_MACROS */
650 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
651 SOURCE must be an lvalue. */
653 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
655 EXTRACT_NUMBER (destination, source); \
661 extract_number_and_incr (destination, source)
663 unsigned char **source;
665 extract_number (destination, *source);
669 # ifndef EXTRACT_MACROS
670 # undef EXTRACT_NUMBER_AND_INCR
671 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
672 extract_number_and_incr (&dest, &src)
673 # endif /* not EXTRACT_MACROS */
677 /* Store a multibyte character in three contiguous bytes starting
678 DESTINATION, and increment DESTINATION to the byte after where the
679 character is stored. Therefore, DESTINATION must be an lvalue. */
681 #define STORE_CHARACTER_AND_INCR(destination, character) \
683 (destination)[0] = (character) & 0377; \
684 (destination)[1] = ((character) >> 8) & 0377; \
685 (destination)[2] = (character) >> 16; \
686 (destination) += 3; \
689 /* Put into DESTINATION a character stored in three contiguous bytes
690 starting at SOURCE. */
692 #define EXTRACT_CHARACTER(destination, source) \
694 (destination) = ((source)[0] \
695 | ((source)[1] << 8) \
696 | ((source)[2] << 16)); \
700 /* Macros for charset. */
702 /* Size of bitmap of charset P in bytes. P is a start of charset,
703 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
704 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
706 /* Nonzero if charset P has range table. */
707 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
709 /* Return the address of range table of charset P. But not the start
710 of table itself, but the before where the number of ranges is
711 stored. `2 +' means to skip re_opcode_t and size of bitmap,
712 and the 2 bytes of flags at the start of the range table. */
713 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
715 /* Extract the bit flags that start a range table. */
716 #define CHARSET_RANGE_TABLE_BITS(p) \
717 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
718 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
720 /* Test if C is listed in the bitmap of charset P. */
721 #define CHARSET_LOOKUP_BITMAP(p, c) \
722 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
723 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
725 /* Return the address of end of RANGE_TABLE. COUNT is number of
726 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
727 is start of range and end of range. `* 3' is size of each start
729 #define CHARSET_RANGE_TABLE_END(range_table, count) \
730 ((range_table) + (count) * 2 * 3)
732 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
733 COUNT is number of ranges in RANGE_TABLE. */
734 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
737 int range_start, range_end; \
739 unsigned char *range_table_end \
740 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
742 for (p = (range_table); p < range_table_end; p += 2 * 3) \
744 EXTRACT_CHARACTER (range_start, p); \
745 EXTRACT_CHARACTER (range_end, p + 3); \
747 if (range_start <= (c) && (c) <= range_end) \
756 /* Test if C is in range table of CHARSET. The flag NOT is negated if
757 C is listed in it. */
758 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
761 /* Number of ranges in range table. */ \
763 unsigned char *range_table = CHARSET_RANGE_TABLE (charset); \
765 EXTRACT_NUMBER_AND_INCR (count, range_table); \
766 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
770 /* If DEBUG is defined, Regex prints many voluminous messages about what
771 it is doing (if the variable `debug' is nonzero). If linked with the
772 main program in `iregex.c', you can enter patterns and strings
773 interactively. And if linked with the main program in `main.c' and
774 the other test files, you can run the already-written tests. */
778 /* We use standard I/O for debugging. */
781 /* It is useful to test things that ``must'' be true when debugging. */
784 static int debug = -100000;
786 # define DEBUG_STATEMENT(e) e
787 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
788 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
789 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
790 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
791 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
792 if (debug > 0) print_partial_compiled_pattern (s, e)
793 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
794 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
797 /* Print the fastmap in human-readable form. */
800 print_fastmap (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 (start, end)
833 unsigned char *start;
837 unsigned char *p = start;
838 unsigned char *pend = end;
846 /* Loop over pattern commands. */
849 printf ("%d:\t", p - start);
851 switch ((re_opcode_t) *p++)
863 printf ("/exactn/%d", mcnt);
873 printf ("/start_memory/%d", *p++);
877 printf ("/stop_memory/%d", *p++);
881 printf ("/duplicate/%d", *p++);
891 register int c, last = -100;
892 register int in_range = 0;
893 int length = CHARSET_BITMAP_SIZE (p - 1);
894 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
896 printf ("/charset [%s",
897 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
899 assert (p + *p < pend);
901 for (c = 0; c < 256; c++)
903 && (p[1 + (c/8)] & (1 << (c % 8))))
905 /* Are we starting a range? */
906 if (last + 1 == c && ! in_range)
911 /* Have we broken a range? */
912 else if (last + 1 != c && in_range)
934 printf ("has-range-table");
936 /* ??? Should print the range table; for now, just skip it. */
937 p += 2; /* skip range table bits */
938 EXTRACT_NUMBER_AND_INCR (count, p);
939 p = CHARSET_RANGE_TABLE_END (p, count);
952 case on_failure_jump:
953 extract_number_and_incr (&mcnt, &p);
954 printf ("/on_failure_jump to %d", p + mcnt - start);
957 case on_failure_keep_string_jump:
958 extract_number_and_incr (&mcnt, &p);
959 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
962 case on_failure_jump_nastyloop:
963 extract_number_and_incr (&mcnt, &p);
964 printf ("/on_failure_jump_nastyloop to %d", p + mcnt - start);
967 case on_failure_jump_loop:
968 extract_number_and_incr (&mcnt, &p);
969 printf ("/on_failure_jump_loop to %d", p + mcnt - start);
972 case on_failure_jump_smart:
973 extract_number_and_incr (&mcnt, &p);
974 printf ("/on_failure_jump_smart to %d", p + mcnt - start);
978 extract_number_and_incr (&mcnt, &p);
979 printf ("/jump to %d", p + mcnt - start);
983 extract_number_and_incr (&mcnt, &p);
984 extract_number_and_incr (&mcnt2, &p);
985 printf ("/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
989 extract_number_and_incr (&mcnt, &p);
990 extract_number_and_incr (&mcnt2, &p);
991 printf ("/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
995 extract_number_and_incr (&mcnt, &p);
996 extract_number_and_incr (&mcnt2, &p);
997 printf ("/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1001 printf ("/wordbound");
1005 printf ("/notwordbound");
1009 printf ("/wordbeg");
1013 printf ("/wordend");
1016 printf ("/syntaxspec");
1018 printf ("/%d", mcnt);
1022 printf ("/notsyntaxspec");
1024 printf ("/%d", mcnt);
1029 printf ("/before_dot");
1037 printf ("/after_dot");
1041 printf ("/categoryspec");
1043 printf ("/%d", mcnt);
1046 case notcategoryspec:
1047 printf ("/notcategoryspec");
1049 printf ("/%d", mcnt);
1062 printf ("?%d", *(p-1));
1068 printf ("%d:\tend of pattern.\n", p - start);
1073 print_compiled_pattern (bufp)
1074 struct re_pattern_buffer *bufp;
1076 unsigned char *buffer = bufp->buffer;
1078 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1079 printf ("%ld bytes used/%ld bytes allocated.\n", bufp->used, bufp->allocated);
1081 if (bufp->fastmap_accurate && bufp->fastmap)
1083 printf ("fastmap: ");
1084 print_fastmap (bufp->fastmap);
1087 printf ("re_nsub: %d\t", bufp->re_nsub);
1088 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1089 printf ("can_be_null: %d\t", bufp->can_be_null);
1090 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1091 printf ("no_sub: %d\t", bufp->no_sub);
1092 printf ("not_bol: %d\t", bufp->not_bol);
1093 printf ("not_eol: %d\t", bufp->not_eol);
1094 printf ("syntax: %d\n", bufp->syntax);
1096 /* Perhaps we should print the translate table? */
1101 print_double_string (where, string1, size1, string2, size2)
1114 if (FIRST_STRING_P (where))
1116 for (this_char = where - string1; this_char < size1; this_char++)
1117 putchar (string1[this_char]);
1122 for (this_char = where - string2; this_char < size2; this_char++)
1123 putchar (string2[this_char]);
1127 #else /* not DEBUG */
1132 # define DEBUG_STATEMENT(e)
1133 # define DEBUG_PRINT1(x)
1134 # define DEBUG_PRINT2(x1, x2)
1135 # define DEBUG_PRINT3(x1, x2, x3)
1136 # define DEBUG_PRINT4(x1, x2, x3, x4)
1137 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1138 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1140 #endif /* not DEBUG */
1142 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1143 also be assigned to arbitrarily: each pattern buffer stores its own
1144 syntax, so it can be changed between regex compilations. */
1145 /* This has no initializer because initialized variables in Emacs
1146 become read-only after dumping. */
1147 reg_syntax_t re_syntax_options;
1150 /* Specify the precise syntax of regexps for compilation. This provides
1151 for compatibility for various utilities which historically have
1152 different, incompatible syntaxes.
1154 The argument SYNTAX is a bit mask comprised of the various bits
1155 defined in regex.h. We return the old syntax. */
1158 re_set_syntax (syntax)
1159 reg_syntax_t syntax;
1161 reg_syntax_t ret = re_syntax_options;
1163 re_syntax_options = syntax;
1167 /* This table gives an error message for each of the error codes listed
1168 in regex.h. Obviously the order here has to be same as there.
1169 POSIX doesn't require that we do anything for REG_NOERROR,
1170 but why not be nice? */
1172 static const char *re_error_msgid[] =
1174 gettext_noop ("Success"), /* REG_NOERROR */
1175 gettext_noop ("No match"), /* REG_NOMATCH */
1176 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1177 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1178 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1179 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1180 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1181 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1182 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1183 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1184 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1185 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1186 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1187 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1188 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1189 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1190 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1193 /* Avoiding alloca during matching, to placate r_alloc. */
1195 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1196 searching and matching functions should not call alloca. On some
1197 systems, alloca is implemented in terms of malloc, and if we're
1198 using the relocating allocator routines, then malloc could cause a
1199 relocation, which might (if the strings being searched are in the
1200 ralloc heap) shift the data out from underneath the regexp
1203 Here's another reason to avoid allocation: Emacs
1204 processes input from X in a signal handler; processing X input may
1205 call malloc; if input arrives while a matching routine is calling
1206 malloc, then we're scrod. But Emacs can't just block input while
1207 calling matching routines; then we don't notice interrupts when
1208 they come in. So, Emacs blocks input around all regexp calls
1209 except the matching calls, which it leaves unprotected, in the
1210 faith that they will not malloc. */
1212 /* Normally, this is fine. */
1213 #define MATCH_MAY_ALLOCATE
1215 /* When using GNU C, we are not REALLY using the C alloca, no matter
1216 what config.h may say. So don't take precautions for it. */
1221 /* The match routines may not allocate if (1) they would do it with malloc
1222 and (2) it's not safe for them to use malloc.
1223 Note that if REL_ALLOC is defined, matching would not use malloc for the
1224 failure stack, but we would still use it for the register vectors;
1225 so REL_ALLOC should not affect this. */
1226 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1227 # undef MATCH_MAY_ALLOCATE
1231 /* Failure stack declarations and macros; both re_compile_fastmap and
1232 re_match_2 use a failure stack. These have to be macros because of
1233 REGEX_ALLOCATE_STACK. */
1236 /* Approximate number of failure points for which to initially allocate space
1237 when matching. If this number is exceeded, we allocate more
1238 space, so it is not a hard limit. */
1239 #ifndef INIT_FAILURE_ALLOC
1240 # define INIT_FAILURE_ALLOC 20
1243 /* Roughly the maximum number of failure points on the stack. Would be
1244 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1245 This is a variable only so users of regex can assign to it; we never
1246 change it ourselves. */
1247 #if defined MATCH_MAY_ALLOCATE
1248 /* Note that 4400 is 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 int re_max_failures = 40000;
1254 int re_max_failures = 4000;
1257 union fail_stack_elt
1259 const unsigned char *pointer;
1260 unsigned int integer;
1263 typedef union fail_stack_elt fail_stack_elt_t;
1267 fail_stack_elt_t *stack;
1269 unsigned avail; /* Offset of next open position. */
1270 unsigned frame; /* Offset of the cur constructed frame. */
1273 #define PATTERN_STACK_EMPTY() (fail_stack.avail == 0)
1274 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1275 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1278 /* Define macros to initialize and free the failure stack.
1279 Do `return -2' if the alloc fails. */
1281 #ifdef MATCH_MAY_ALLOCATE
1282 # define INIT_FAIL_STACK() \
1284 fail_stack.stack = (fail_stack_elt_t *) \
1285 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1286 * sizeof (fail_stack_elt_t)); \
1288 if (fail_stack.stack == NULL) \
1291 fail_stack.size = INIT_FAILURE_ALLOC; \
1292 fail_stack.avail = 0; \
1293 fail_stack.frame = 0; \
1296 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1298 # define INIT_FAIL_STACK() \
1300 fail_stack.avail = 0; \
1301 fail_stack.frame = 0; \
1304 # define RESET_FAIL_STACK() ((void)0)
1308 /* Double the size of FAIL_STACK, up to a limit
1309 which allows approximately `re_max_failures' items.
1311 Return 1 if succeeds, and 0 if either ran out of memory
1312 allocating space for it or it was already too large.
1314 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1316 /* Factor to increase the failure stack size by
1317 when we increase it.
1318 This used to be 2, but 2 was too wasteful
1319 because the old discarded stacks added up to as much space
1320 were as ultimate, maximum-size stack. */
1321 #define FAIL_STACK_GROWTH_FACTOR 4
1323 #define GROW_FAIL_STACK(fail_stack) \
1324 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1325 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1327 : ((fail_stack).stack \
1328 = (fail_stack_elt_t *) \
1329 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1330 (fail_stack).size * sizeof (fail_stack_elt_t), \
1331 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1332 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1333 * FAIL_STACK_GROWTH_FACTOR))), \
1335 (fail_stack).stack == NULL \
1337 : ((fail_stack).size \
1338 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1339 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1340 * FAIL_STACK_GROWTH_FACTOR)) \
1341 / sizeof (fail_stack_elt_t)), \
1345 /* Push pointer POINTER on FAIL_STACK.
1346 Return 1 if was able to do so and 0 if ran out of memory allocating
1348 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1349 ((FAIL_STACK_FULL () \
1350 && !GROW_FAIL_STACK (FAIL_STACK)) \
1352 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1354 #define POP_PATTERN_OP() POP_FAILURE_POINTER ()
1356 /* Push a pointer value onto the failure stack.
1357 Assumes the variable `fail_stack'. Probably should only
1358 be called from within `PUSH_FAILURE_POINT'. */
1359 #define PUSH_FAILURE_POINTER(item) \
1360 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1362 /* This pushes an integer-valued item onto the failure stack.
1363 Assumes the variable `fail_stack'. Probably should only
1364 be called from within `PUSH_FAILURE_POINT'. */
1365 #define PUSH_FAILURE_INT(item) \
1366 fail_stack.stack[fail_stack.avail++].integer = (item)
1368 /* Push a fail_stack_elt_t value onto the failure stack.
1369 Assumes the variable `fail_stack'. Probably should only
1370 be called from within `PUSH_FAILURE_POINT'. */
1371 #define PUSH_FAILURE_ELT(item) \
1372 fail_stack.stack[fail_stack.avail++] = (item)
1374 /* These three POP... operations complement the three PUSH... operations.
1375 All assume that `fail_stack' is nonempty. */
1376 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1377 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1378 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1380 /* Individual items aside from the registers. */
1381 #define NUM_NONREG_ITEMS 3
1383 /* Used to examine the stack (to detect infinite loops). */
1384 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1385 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1386 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1387 #define TOP_FAILURE_HANDLE() fail_stack.frame
1390 #define ENSURE_FAIL_STACK(space) \
1391 while (REMAINING_AVAIL_SLOTS <= space) { \
1392 if (!GROW_FAIL_STACK (fail_stack)) \
1394 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1395 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1398 /* Push register NUM onto the stack. */
1399 #define PUSH_FAILURE_REG(num) \
1401 char *destination; \
1402 ENSURE_FAIL_STACK(3); \
1403 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1404 num, regstart[num], regend[num]); \
1405 PUSH_FAILURE_POINTER (regstart[num]); \
1406 PUSH_FAILURE_POINTER (regend[num]); \
1407 PUSH_FAILURE_INT (num); \
1410 #define PUSH_FAILURE_COUNT(ptr) \
1412 char *destination; \
1414 ENSURE_FAIL_STACK(3); \
1415 EXTRACT_NUMBER (c, ptr); \
1416 DEBUG_PRINT3 (" Push counter %p = %d\n", ptr, c); \
1417 PUSH_FAILURE_INT (c); \
1418 PUSH_FAILURE_POINTER (ptr); \
1419 PUSH_FAILURE_INT (-1); \
1422 /* Pop a saved register off the stack. */
1423 #define POP_FAILURE_REG_OR_COUNT() \
1425 int reg = POP_FAILURE_INT (); \
1428 /* It's a counter. */ \
1429 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1430 reg = POP_FAILURE_INT (); \
1431 STORE_NUMBER (ptr, reg); \
1432 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1436 regend[reg] = POP_FAILURE_POINTER (); \
1437 regstart[reg] = POP_FAILURE_POINTER (); \
1438 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1439 reg, regstart[reg], regend[reg]); \
1443 /* Check that we are not stuck in an infinite loop. */
1444 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1446 int failure = TOP_FAILURE_HANDLE(); \
1447 /* Check for infinite matching loops */ \
1448 while (failure > 0 && \
1449 (FAILURE_STR (failure) == string_place \
1450 || FAILURE_STR (failure) == NULL)) \
1452 assert (FAILURE_PAT (failure) >= bufp->buffer \
1453 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1454 if (FAILURE_PAT (failure) == pat_cur) \
1456 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1457 failure = NEXT_FAILURE_HANDLE(failure); \
1459 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1462 /* Push the information about the state we will need
1463 if we ever fail back to it.
1465 Requires variables fail_stack, regstart, regend and
1466 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1469 Does `return FAILURE_CODE' if runs out of memory. */
1471 #define PUSH_FAILURE_POINT(pattern, string_place) \
1473 char *destination; \
1474 /* Must be int, so when we don't save any registers, the arithmetic \
1475 of 0 + -1 isn't done as unsigned. */ \
1477 DEBUG_STATEMENT (failure_id++); \
1478 DEBUG_STATEMENT (nfailure_points_pushed++); \
1479 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1480 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1481 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1483 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1485 DEBUG_PRINT1 ("\n"); \
1487 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1488 PUSH_FAILURE_INT (fail_stack.frame); \
1490 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1491 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1492 DEBUG_PRINT1 ("'\n"); \
1493 PUSH_FAILURE_POINTER (string_place); \
1495 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1496 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1497 PUSH_FAILURE_POINTER (pattern); \
1499 /* Close the frame by moving the frame pointer past it. */ \
1500 fail_stack.frame = fail_stack.avail; \
1503 /* Estimate the size of data pushed by a typical failure stack entry.
1504 An estimate is all we need, because all we use this for
1505 is to choose a limit for how big to make the failure stack. */
1507 #define TYPICAL_FAILURE_SIZE 20
1509 /* How many items can still be added to the stack without overflowing it. */
1510 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1513 /* Pops what PUSH_FAIL_STACK pushes.
1515 We restore into the parameters, all of which should be lvalues:
1516 STR -- the saved data position.
1517 PAT -- the saved pattern position.
1518 REGSTART, REGEND -- arrays of string positions.
1520 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1521 `pend', `string1', `size1', `string2', and `size2'. */
1523 #define POP_FAILURE_POINT(str, pat) \
1525 assert (!FAIL_STACK_EMPTY ()); \
1527 /* Remove failure points and point to how many regs pushed. */ \
1528 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1529 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1530 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1532 /* Pop the saved registers. */ \
1533 while (fail_stack.frame < fail_stack.avail) \
1534 POP_FAILURE_REG_OR_COUNT (); \
1536 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1537 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1538 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1540 /* If the saved string location is NULL, it came from an \
1541 on_failure_keep_string_jump opcode, and we want to throw away the \
1542 saved NULL, thus retaining our current position in the string. */ \
1543 str = (re_char *) POP_FAILURE_POINTER (); \
1544 DEBUG_PRINT2 (" Popping string %p: `", str); \
1545 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1546 DEBUG_PRINT1 ("'\n"); \
1548 fail_stack.frame = POP_FAILURE_INT (); \
1549 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1551 assert (fail_stack.avail >= 0); \
1552 assert (fail_stack.frame <= fail_stack.avail); \
1554 DEBUG_STATEMENT (nfailure_points_popped++); \
1555 } while (0) /* POP_FAILURE_POINT */
1559 /* Registers are set to a sentinel when they haven't yet matched. */
1560 #define REG_UNSET_VALUE NULL
1561 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1563 /* Subroutine declarations and macros for regex_compile. */
1565 static void store_op1 _RE_ARGS((re_opcode_t op, unsigned char *loc, int arg));
1566 static void store_op2 _RE_ARGS((re_opcode_t op, unsigned char *loc,
1567 int arg1, int arg2));
1568 static void insert_op1 _RE_ARGS((re_opcode_t op, unsigned char *loc,
1569 int arg, unsigned char *end));
1570 static void insert_op2 _RE_ARGS((re_opcode_t op, unsigned char *loc,
1571 int arg1, int arg2, unsigned char *end));
1572 static boolean at_begline_loc_p _RE_ARGS((const unsigned char *pattern,
1573 const unsigned char *p,
1574 reg_syntax_t syntax));
1575 static boolean at_endline_loc_p _RE_ARGS((const unsigned char *p,
1576 const unsigned char *pend,
1577 reg_syntax_t syntax));
1578 static unsigned char *skip_one_char _RE_ARGS((unsigned char *p));
1579 static int analyse_first _RE_ARGS((unsigned char *p, unsigned char *pend,
1580 char *fastmap, const int multibyte));
1582 /* Fetch the next character in the uncompiled pattern---translating it
1583 if necessary. Also cast from a signed character in the constant
1584 string passed to us by the user to an unsigned char that we can use
1585 as an array index (in, e.g., `translate'). */
1586 #define PATFETCH(c) \
1589 c = TRANSLATE (c); \
1592 /* Fetch the next character in the uncompiled pattern, with no
1594 #define PATFETCH_RAW(c) \
1597 if (p == pend) return REG_EEND; \
1598 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1603 /* If `translate' is non-null, return translate[D], else just D. We
1604 cast the subscript to translate because some data is declared as
1605 `char *', to avoid warnings when a string constant is passed. But
1606 when we use a character as a subscript we must make it unsigned. */
1608 # define TRANSLATE(d) \
1609 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1613 /* Macros for outputting the compiled pattern into `buffer'. */
1615 /* If the buffer isn't allocated when it comes in, use this. */
1616 #define INIT_BUF_SIZE 32
1618 /* Make sure we have at least N more bytes of space in buffer. */
1619 #define GET_BUFFER_SPACE(n) \
1620 while (b - bufp->buffer + (n) > bufp->allocated) \
1623 /* Make sure we have one more byte of buffer space and then add C to it. */
1624 #define BUF_PUSH(c) \
1626 GET_BUFFER_SPACE (1); \
1627 *b++ = (unsigned char) (c); \
1631 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1632 #define BUF_PUSH_2(c1, c2) \
1634 GET_BUFFER_SPACE (2); \
1635 *b++ = (unsigned char) (c1); \
1636 *b++ = (unsigned char) (c2); \
1640 /* As with BUF_PUSH_2, except for three bytes. */
1641 #define BUF_PUSH_3(c1, c2, c3) \
1643 GET_BUFFER_SPACE (3); \
1644 *b++ = (unsigned char) (c1); \
1645 *b++ = (unsigned char) (c2); \
1646 *b++ = (unsigned char) (c3); \
1650 /* Store a jump with opcode OP at LOC to location TO. We store a
1651 relative address offset by the three bytes the jump itself occupies. */
1652 #define STORE_JUMP(op, loc, to) \
1653 store_op1 (op, loc, (to) - (loc) - 3)
1655 /* Likewise, for a two-argument jump. */
1656 #define STORE_JUMP2(op, loc, to, arg) \
1657 store_op2 (op, loc, (to) - (loc) - 3, arg)
1659 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1660 #define INSERT_JUMP(op, loc, to) \
1661 insert_op1 (op, loc, (to) - (loc) - 3, b)
1663 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1664 #define INSERT_JUMP2(op, loc, to, arg) \
1665 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1668 /* This is not an arbitrary limit: the arguments which represent offsets
1669 into the pattern are two bytes long. So if 2^16 bytes turns out to
1670 be too small, many things would have to change. */
1671 #define MAX_BUF_SIZE (1L << 16)
1674 /* Extend the buffer by twice its current size via realloc and
1675 reset the pointers that pointed into the old block to point to the
1676 correct places in the new one. If extending the buffer results in it
1677 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1678 #define EXTEND_BUFFER() \
1680 unsigned char *old_buffer = bufp->buffer; \
1681 if (bufp->allocated == MAX_BUF_SIZE) \
1683 bufp->allocated <<= 1; \
1684 if (bufp->allocated > MAX_BUF_SIZE) \
1685 bufp->allocated = MAX_BUF_SIZE; \
1686 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1687 if (bufp->buffer == NULL) \
1688 return REG_ESPACE; \
1689 /* If the buffer moved, move all the pointers into it. */ \
1690 if (old_buffer != bufp->buffer) \
1692 b = (b - old_buffer) + bufp->buffer; \
1693 begalt = (begalt - old_buffer) + bufp->buffer; \
1694 if (fixup_alt_jump) \
1695 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1697 laststart = (laststart - old_buffer) + bufp->buffer; \
1698 if (pending_exact) \
1699 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1704 /* Since we have one byte reserved for the register number argument to
1705 {start,stop}_memory, the maximum number of groups we can report
1706 things about is what fits in that byte. */
1707 #define MAX_REGNUM 255
1709 /* But patterns can have more than `MAX_REGNUM' registers. We just
1710 ignore the excess. */
1711 typedef unsigned regnum_t;
1714 /* Macros for the compile stack. */
1716 /* Since offsets can go either forwards or backwards, this type needs to
1717 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1718 typedef int pattern_offset_t;
1722 pattern_offset_t begalt_offset;
1723 pattern_offset_t fixup_alt_jump;
1724 pattern_offset_t laststart_offset;
1726 } compile_stack_elt_t;
1731 compile_stack_elt_t *stack;
1733 unsigned avail; /* Offset of next open position. */
1734 } compile_stack_type;
1737 #define INIT_COMPILE_STACK_SIZE 32
1739 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1740 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1742 /* The next available element. */
1743 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1746 /* Structure to manage work area for range table. */
1747 struct range_table_work_area
1749 int *table; /* actual work area. */
1750 int allocated; /* allocated size for work area in bytes. */
1751 int used; /* actually used size in words. */
1752 int bits; /* flag to record character classes */
1755 /* Make sure that WORK_AREA can hold more N multibyte characters. */
1756 #define EXTEND_RANGE_TABLE_WORK_AREA(work_area, n) \
1758 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1760 (work_area).allocated += 16 * sizeof (int); \
1761 if ((work_area).table) \
1763 = (int *) realloc ((work_area).table, (work_area).allocated); \
1766 = (int *) malloc ((work_area).allocated); \
1767 if ((work_area).table == 0) \
1768 FREE_STACK_RETURN (REG_ESPACE); \
1772 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1773 (work_area).bits |= (bit)
1775 /* These bits represent the various character classes such as [:alnum:]
1776 in a charset's range table. */
1777 #define BIT_ALNUM 0x1
1778 #define BIT_ALPHA 0x2
1779 #define BIT_WORD 0x4
1780 #define BIT_ASCII 0x8
1781 #define BIT_NONASCII 0x10
1782 #define BIT_GRAPH 0x20
1783 #define BIT_LOWER 0x40
1784 #define BIT_PRINT 0x80
1785 #define BIT_PUNCT 0x100
1786 #define BIT_SPACE 0x200
1787 #define BIT_UPPER 0x400
1788 #define BIT_UNIBYTE 0x800
1789 #define BIT_MULTIBYTE 0x1000
1791 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1792 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1794 EXTEND_RANGE_TABLE_WORK_AREA ((work_area), 2); \
1795 (work_area).table[(work_area).used++] = (range_start); \
1796 (work_area).table[(work_area).used++] = (range_end); \
1799 /* Free allocated memory for WORK_AREA. */
1800 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1802 if ((work_area).table) \
1803 free ((work_area).table); \
1806 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1807 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1808 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1809 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1812 /* Set the bit for character C in a list. */
1813 #define SET_LIST_BIT(c) \
1814 (b[((unsigned char) (c)) / BYTEWIDTH] \
1815 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1818 /* Get the next unsigned number in the uncompiled pattern. */
1819 #define GET_UNSIGNED_NUMBER(num) \
1820 do { if (p != pend) \
1823 while (ISDIGIT (c)) \
1827 num = num * 10 + c - '0'; \
1835 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1837 #define IS_CHAR_CLASS(string) \
1838 (STREQ (string, "alpha") || STREQ (string, "upper") \
1839 || STREQ (string, "lower") || STREQ (string, "digit") \
1840 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1841 || STREQ (string, "space") || STREQ (string, "print") \
1842 || STREQ (string, "punct") || STREQ (string, "graph") \
1843 || STREQ (string, "cntrl") || STREQ (string, "blank") \
1844 || STREQ (string, "word") \
1845 || STREQ (string, "ascii") || STREQ (string, "nonascii") \
1846 || STREQ (string, "unibyte") || STREQ (string, "multibyte"))
1848 /* QUIT is only used on NTemacs. */
1849 #if !defined WINDOWSNT || !defined emacs
1854 #ifndef MATCH_MAY_ALLOCATE
1856 /* If we cannot allocate large objects within re_match_2_internal,
1857 we make the fail stack and register vectors global.
1858 The fail stack, we grow to the maximum size when a regexp
1860 The register vectors, we adjust in size each time we
1861 compile a regexp, according to the number of registers it needs. */
1863 static fail_stack_type fail_stack;
1865 /* Size with which the following vectors are currently allocated.
1866 That is so we can make them bigger as needed,
1867 but never make them smaller. */
1868 static int regs_allocated_size;
1870 static re_char ** regstart, ** regend;
1871 static re_char **best_regstart, **best_regend;
1873 /* Make the register vectors big enough for NUM_REGS registers,
1874 but don't make them smaller. */
1877 regex_grow_registers (num_regs)
1880 if (num_regs > regs_allocated_size)
1882 RETALLOC_IF (regstart, num_regs, re_char *);
1883 RETALLOC_IF (regend, num_regs, re_char *);
1884 RETALLOC_IF (best_regstart, num_regs, re_char *);
1885 RETALLOC_IF (best_regend, num_regs, re_char *);
1887 regs_allocated_size = num_regs;
1891 #endif /* not MATCH_MAY_ALLOCATE */
1893 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
1897 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1898 Returns one of error codes defined in `regex.h', or zero for success.
1900 Assumes the `allocated' (and perhaps `buffer') and `translate'
1901 fields are set in BUFP on entry.
1903 If it succeeds, results are put in BUFP (if it returns an error, the
1904 contents of BUFP are undefined):
1905 `buffer' is the compiled pattern;
1906 `syntax' is set to SYNTAX;
1907 `used' is set to the length of the compiled pattern;
1908 `fastmap_accurate' is zero;
1909 `re_nsub' is the number of subexpressions in PATTERN;
1910 `not_bol' and `not_eol' are zero;
1912 The `fastmap' and `newline_anchor' fields are neither
1913 examined nor set. */
1915 /* Insert the `jump' from the end of last alternative to "here".
1916 The space for the jump has already been allocated. */
1917 #define FIXUP_ALT_JUMP() \
1919 if (fixup_alt_jump) \
1920 STORE_JUMP (jump, fixup_alt_jump, b); \
1924 /* Return, freeing storage we allocated. */
1925 #define FREE_STACK_RETURN(value) \
1927 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
1928 free (compile_stack.stack); \
1932 static reg_errcode_t
1933 regex_compile (pattern, size, syntax, bufp)
1936 reg_syntax_t syntax;
1937 struct re_pattern_buffer *bufp;
1939 /* We fetch characters from PATTERN here. Even though PATTERN is
1940 `char *' (i.e., signed), we declare these variables as unsigned, so
1941 they can be reliably used as array indices. */
1942 register unsigned int c, c1;
1944 /* A random temporary spot in PATTERN. */
1947 /* Points to the end of the buffer, where we should append. */
1948 register unsigned char *b;
1950 /* Keeps track of unclosed groups. */
1951 compile_stack_type compile_stack;
1953 /* Points to the current (ending) position in the pattern. */
1955 /* `const' makes AIX compiler fail. */
1956 unsigned char *p = pattern;
1958 re_char *p = pattern;
1960 re_char *pend = pattern + size;
1962 /* How to translate the characters in the pattern. */
1963 RE_TRANSLATE_TYPE translate = bufp->translate;
1965 /* Address of the count-byte of the most recently inserted `exactn'
1966 command. This makes it possible to tell if a new exact-match
1967 character can be added to that command or if the character requires
1968 a new `exactn' command. */
1969 unsigned char *pending_exact = 0;
1971 /* Address of start of the most recently finished expression.
1972 This tells, e.g., postfix * where to find the start of its
1973 operand. Reset at the beginning of groups and alternatives. */
1974 unsigned char *laststart = 0;
1976 /* Address of beginning of regexp, or inside of last group. */
1977 unsigned char *begalt;
1979 /* Place in the uncompiled pattern (i.e., the {) to
1980 which to go back if the interval is invalid. */
1981 re_char *beg_interval;
1983 /* Address of the place where a forward jump should go to the end of
1984 the containing expression. Each alternative of an `or' -- except the
1985 last -- ends with a forward jump of this sort. */
1986 unsigned char *fixup_alt_jump = 0;
1988 /* Counts open-groups as they are encountered. Remembered for the
1989 matching close-group on the compile stack, so the same register
1990 number is put in the stop_memory as the start_memory. */
1991 regnum_t regnum = 0;
1993 /* Work area for range table of charset. */
1994 struct range_table_work_area range_table_work;
1996 /* If the object matched can contain multibyte characters. */
1997 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2001 DEBUG_PRINT1 ("\nCompiling pattern: ");
2004 unsigned debug_count;
2006 for (debug_count = 0; debug_count < size; debug_count++)
2007 putchar (pattern[debug_count]);
2012 /* Initialize the compile stack. */
2013 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2014 if (compile_stack.stack == NULL)
2017 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2018 compile_stack.avail = 0;
2020 range_table_work.table = 0;
2021 range_table_work.allocated = 0;
2023 /* Initialize the pattern buffer. */
2024 bufp->syntax = syntax;
2025 bufp->fastmap_accurate = 0;
2026 bufp->not_bol = bufp->not_eol = 0;
2028 /* Set `used' to zero, so that if we return an error, the pattern
2029 printer (for debugging) will think there's no pattern. We reset it
2033 /* Always count groups, whether or not bufp->no_sub is set. */
2036 #if !defined emacs && !defined SYNTAX_TABLE
2037 /* Initialize the syntax table. */
2038 init_syntax_once ();
2041 if (bufp->allocated == 0)
2044 { /* If zero allocated, but buffer is non-null, try to realloc
2045 enough space. This loses if buffer's address is bogus, but
2046 that is the user's responsibility. */
2047 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2050 { /* Caller did not allocate a buffer. Do it for them. */
2051 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2053 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2055 bufp->allocated = INIT_BUF_SIZE;
2058 begalt = b = bufp->buffer;
2060 /* Loop through the uncompiled pattern until we're at the end. */
2069 if ( /* If at start of pattern, it's an operator. */
2071 /* If context independent, it's an operator. */
2072 || syntax & RE_CONTEXT_INDEP_ANCHORS
2073 /* Otherwise, depends on what's come before. */
2074 || at_begline_loc_p (pattern, p, syntax))
2084 if ( /* If at end of pattern, it's an operator. */
2086 /* If context independent, it's an operator. */
2087 || syntax & RE_CONTEXT_INDEP_ANCHORS
2088 /* Otherwise, depends on what's next. */
2089 || at_endline_loc_p (p, pend, syntax))
2099 if ((syntax & RE_BK_PLUS_QM)
2100 || (syntax & RE_LIMITED_OPS))
2104 /* If there is no previous pattern... */
2107 if (syntax & RE_CONTEXT_INVALID_OPS)
2108 FREE_STACK_RETURN (REG_BADRPT);
2109 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2114 /* 1 means zero (many) matches is allowed. */
2115 boolean zero_times_ok = 0, many_times_ok = 0;
2118 /* If there is a sequence of repetition chars, collapse it
2119 down to just one (the right one). We can't combine
2120 interval operators with these because of, e.g., `a{2}*',
2121 which should only match an even number of `a's. */
2125 if ((syntax & RE_FRUGAL)
2126 && c == '?' && (zero_times_ok || many_times_ok))
2130 zero_times_ok |= c != '+';
2131 many_times_ok |= c != '?';
2137 || (!(syntax & RE_BK_PLUS_QM)
2138 && (*p == '+' || *p == '?')))
2140 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2143 FREE_STACK_RETURN (REG_EESCAPE);
2144 if (p[1] == '+' || p[1] == '?')
2145 PATFETCH (c); /* Gobble up the backslash. */
2151 /* If we get here, we found another repeat character. */
2155 /* Star, etc. applied to an empty pattern is equivalent
2156 to an empty pattern. */
2157 if (!laststart || laststart == b)
2160 /* Now we know whether or not zero matches is allowed
2161 and also whether or not two or more matches is allowed. */
2166 boolean simple = skip_one_char (laststart) == b;
2167 unsigned int startoffset = 0;
2169 (simple || !analyse_first (laststart, b, NULL, 0)) ?
2170 on_failure_jump : on_failure_jump_loop;
2171 assert (skip_one_char (laststart) <= b);
2173 if (!zero_times_ok && simple)
2174 { /* Since simple * loops can be made faster by using
2175 on_failure_keep_string_jump, we turn simple P+
2176 into PP* if P is simple. */
2177 unsigned char *p1, *p2;
2178 startoffset = b - laststart;
2179 GET_BUFFER_SPACE (startoffset);
2180 p1 = b; p2 = laststart;
2186 GET_BUFFER_SPACE (6);
2189 STORE_JUMP (ofj, b, b + 6);
2191 /* Simple * loops can use on_failure_keep_string_jump
2192 depending on what follows. But since we don't know
2193 that yet, we leave the decision up to
2194 on_failure_jump_smart. */
2195 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2196 laststart + startoffset, b + 6);
2198 STORE_JUMP (jump, b, laststart + startoffset);
2203 /* A simple ? pattern. */
2204 assert (zero_times_ok);
2205 GET_BUFFER_SPACE (3);
2206 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2210 else /* not greedy */
2211 { /* I wish the greedy and non-greedy cases could be merged. */
2213 GET_BUFFER_SPACE (7); /* We might use less. */
2216 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2218 /* The non-greedy multiple match looks like a repeat..until:
2219 we only need a conditional jump at the end of the loop */
2220 if (emptyp) BUF_PUSH (no_op);
2221 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2222 : on_failure_jump, b, laststart);
2226 /* The repeat...until naturally matches one or more.
2227 To also match zero times, we need to first jump to
2228 the end of the loop (its conditional jump). */
2229 INSERT_JUMP (jump, laststart, b);
2235 /* non-greedy a?? */
2236 INSERT_JUMP (jump, laststart, b + 3);
2238 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2255 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2257 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2259 /* Ensure that we have enough space to push a charset: the
2260 opcode, the length count, and the bitset; 34 bytes in all. */
2261 GET_BUFFER_SPACE (34);
2265 /* We test `*p == '^' twice, instead of using an if
2266 statement, so we only need one BUF_PUSH. */
2267 BUF_PUSH (*p == '^' ? charset_not : charset);
2271 /* Remember the first position in the bracket expression. */
2274 /* Push the number of bytes in the bitmap. */
2275 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2277 /* Clear the whole map. */
2278 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2280 /* charset_not matches newline according to a syntax bit. */
2281 if ((re_opcode_t) b[-2] == charset_not
2282 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2283 SET_LIST_BIT ('\n');
2285 /* Read in characters and ranges, setting map bits. */
2288 boolean escaped_char = false;
2289 const unsigned char *p2 = p;
2291 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2295 /* \ might escape characters inside [...] and [^...]. */
2296 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2298 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2301 escaped_char = true;
2305 /* Could be the end of the bracket expression. If it's
2306 not (i.e., when the bracket expression is `[]' so
2307 far), the ']' character bit gets set way below. */
2308 if (c == ']' && p2 != p1)
2312 /* What should we do for the character which is
2313 greater than 0x7F, but not BASE_LEADING_CODE_P?
2316 /* See if we're at the beginning of a possible character
2319 if (!escaped_char &&
2320 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2322 /* Leave room for the null. */
2323 char str[CHAR_CLASS_MAX_LENGTH + 1];
2324 const unsigned char *class_beg;
2330 /* If pattern is `[[:'. */
2331 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2336 if (c == ':' || c == ']' || p == pend
2337 || c1 == CHAR_CLASS_MAX_LENGTH)
2343 /* If isn't a word bracketed by `[:' and `:]':
2344 undo the ending character, the letters, and
2345 leave the leading `:' and `[' (but set bits for
2347 if (c == ':' && *p == ']')
2350 boolean is_alnum = STREQ (str, "alnum");
2351 boolean is_alpha = STREQ (str, "alpha");
2352 boolean is_ascii = STREQ (str, "ascii");
2353 boolean is_blank = STREQ (str, "blank");
2354 boolean is_cntrl = STREQ (str, "cntrl");
2355 boolean is_digit = STREQ (str, "digit");
2356 boolean is_graph = STREQ (str, "graph");
2357 boolean is_lower = STREQ (str, "lower");
2358 boolean is_multibyte = STREQ (str, "multibyte");
2359 boolean is_nonascii = STREQ (str, "nonascii");
2360 boolean is_print = STREQ (str, "print");
2361 boolean is_punct = STREQ (str, "punct");
2362 boolean is_space = STREQ (str, "space");
2363 boolean is_unibyte = STREQ (str, "unibyte");
2364 boolean is_upper = STREQ (str, "upper");
2365 boolean is_word = STREQ (str, "word");
2366 boolean is_xdigit = STREQ (str, "xdigit");
2368 if (!IS_CHAR_CLASS (str))
2369 FREE_STACK_RETURN (REG_ECTYPE);
2371 /* Throw away the ] at the end of the character
2375 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2377 /* Most character classes in a multibyte match
2378 just set a flag. Exceptions are is_blank,
2379 is_digit, is_cntrl, and is_xdigit, since
2380 they can only match ASCII characters. We
2381 don't need to handle them for multibyte. */
2387 if (is_alnum) bit = BIT_ALNUM;
2388 if (is_alpha) bit = BIT_ALPHA;
2389 if (is_ascii) bit = BIT_ASCII;
2390 if (is_graph) bit = BIT_GRAPH;
2391 if (is_lower) bit = BIT_LOWER;
2392 if (is_multibyte) bit = BIT_MULTIBYTE;
2393 if (is_nonascii) bit = BIT_NONASCII;
2394 if (is_print) bit = BIT_PRINT;
2395 if (is_punct) bit = BIT_PUNCT;
2396 if (is_space) bit = BIT_SPACE;
2397 if (is_unibyte) bit = BIT_UNIBYTE;
2398 if (is_upper) bit = BIT_UPPER;
2399 if (is_word) bit = BIT_WORD;
2401 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work,
2405 /* Handle character classes for ASCII characters. */
2406 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2408 int translated = TRANSLATE (ch);
2409 /* This was split into 3 if's to
2410 avoid an arbitrary limit in some compiler. */
2411 if ( (is_alnum && ISALNUM (ch))
2412 || (is_alpha && ISALPHA (ch))
2413 || (is_blank && ISBLANK (ch))
2414 || (is_cntrl && ISCNTRL (ch)))
2415 SET_LIST_BIT (translated);
2416 if ( (is_digit && ISDIGIT (ch))
2417 || (is_graph && ISGRAPH (ch))
2418 || (is_lower && ISLOWER (ch))
2419 || (is_print && ISPRINT (ch)))
2420 SET_LIST_BIT (translated);
2421 if ( (is_punct && ISPUNCT (ch))
2422 || (is_space && ISSPACE (ch))
2423 || (is_upper && ISUPPER (ch))
2424 || (is_xdigit && ISXDIGIT (ch)))
2425 SET_LIST_BIT (translated);
2426 if ( (is_ascii && IS_REAL_ASCII (ch))
2427 || (is_nonascii && !IS_REAL_ASCII (ch))
2428 || (is_unibyte && ISUNIBYTE (ch))
2429 || (is_multibyte && !ISUNIBYTE (ch)))
2430 SET_LIST_BIT (translated);
2432 if ( (is_word && ISWORD (ch)))
2433 SET_LIST_BIT (translated);
2436 /* Repeat the loop. */
2441 /* Go back to right after the "[:". */
2445 /* Because the `:' may starts the range, we
2446 can't simply set bit and repeat the loop.
2447 Instead, just set it to C and handle below. */
2452 if (p < pend && p[0] == '-' && p[1] != ']')
2455 /* Discard the `-'. */
2458 /* Fetch the character which ends the range. */
2461 if (SINGLE_BYTE_CHAR_P (c))
2463 if (! SINGLE_BYTE_CHAR_P (c1))
2465 /* Handle a range such as \177-\377 in
2466 multibyte mode. Split that into two
2467 ranges, the low one ending at 0237, and
2468 the high one starting at the smallest
2469 character in the charset of C1 and
2471 int charset = CHAR_CHARSET (c1);
2472 int c2 = MAKE_CHAR (charset, 0, 0);
2474 SET_RANGE_TABLE_WORK_AREA (range_table_work,
2479 else if (!SAME_CHARSET_P (c, c1))
2480 FREE_STACK_RETURN (REG_ERANGE);
2483 /* Range from C to C. */
2486 /* Set the range ... */
2487 if (SINGLE_BYTE_CHAR_P (c))
2488 /* ... into bitmap. */
2491 int range_start = c, range_end = c1;
2493 /* If the start is after the end, the range is empty. */
2494 if (range_start > range_end)
2496 if (syntax & RE_NO_EMPTY_RANGES)
2497 FREE_STACK_RETURN (REG_ERANGE);
2498 /* Else, repeat the loop. */
2502 for (this_char = range_start; this_char <= range_end;
2504 SET_LIST_BIT (TRANSLATE (this_char));
2508 /* ... into range table. */
2509 SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1);
2512 /* Discard any (non)matching list bytes that are all 0 at the
2513 end of the map. Decrease the map-length byte too. */
2514 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2518 /* Build real range table from work area. */
2519 if (RANGE_TABLE_WORK_USED (range_table_work)
2520 || RANGE_TABLE_WORK_BITS (range_table_work))
2523 int used = RANGE_TABLE_WORK_USED (range_table_work);
2525 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2526 bytes for flags, two for COUNT, and three bytes for
2528 GET_BUFFER_SPACE (4 + used * 3);
2530 /* Indicate the existence of range table. */
2531 laststart[1] |= 0x80;
2533 /* Store the character class flag bits into the range table.
2534 If not in emacs, these flag bits are always 0. */
2535 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
2536 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
2538 STORE_NUMBER_AND_INCR (b, used / 2);
2539 for (i = 0; i < used; i++)
2540 STORE_CHARACTER_AND_INCR
2541 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
2548 if (syntax & RE_NO_BK_PARENS)
2555 if (syntax & RE_NO_BK_PARENS)
2562 if (syntax & RE_NEWLINE_ALT)
2569 if (syntax & RE_NO_BK_VBAR)
2576 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2577 goto handle_interval;
2583 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2585 /* Do not translate the character after the \, so that we can
2586 distinguish, e.g., \B from \b, even if we normally would
2587 translate, e.g., B to b. */
2593 if (syntax & RE_NO_BK_PARENS)
2594 goto normal_backslash;
2601 /* Look for a special (?...) construct */
2602 if ((syntax & RE_SHY_GROUPS) && *p == '?')
2604 PATFETCH (c); /* Gobble up the '?'. */
2608 case ':': shy = 1; break;
2610 /* Only (?:...) is supported right now. */
2611 FREE_STACK_RETURN (REG_BADPAT);
2622 if (COMPILE_STACK_FULL)
2624 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2625 compile_stack_elt_t);
2626 if (compile_stack.stack == NULL) return REG_ESPACE;
2628 compile_stack.size <<= 1;
2631 /* These are the values to restore when we hit end of this
2632 group. They are all relative offsets, so that if the
2633 whole pattern moves because of realloc, they will still
2635 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2636 COMPILE_STACK_TOP.fixup_alt_jump
2637 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2638 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2639 COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum;
2642 start_memory for groups beyond the last one we can
2643 represent in the compiled pattern. */
2644 if (regnum <= MAX_REGNUM && !shy)
2645 BUF_PUSH_2 (start_memory, regnum);
2647 compile_stack.avail++;
2652 /* If we've reached MAX_REGNUM groups, then this open
2653 won't actually generate any code, so we'll have to
2654 clear pending_exact explicitly. */
2660 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2662 if (COMPILE_STACK_EMPTY)
2664 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2665 goto normal_backslash;
2667 FREE_STACK_RETURN (REG_ERPAREN);
2673 /* See similar code for backslashed left paren above. */
2674 if (COMPILE_STACK_EMPTY)
2676 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2679 FREE_STACK_RETURN (REG_ERPAREN);
2682 /* Since we just checked for an empty stack above, this
2683 ``can't happen''. */
2684 assert (compile_stack.avail != 0);
2686 /* We don't just want to restore into `regnum', because
2687 later groups should continue to be numbered higher,
2688 as in `(ab)c(de)' -- the second group is #2. */
2689 regnum_t this_group_regnum;
2691 compile_stack.avail--;
2692 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2694 = COMPILE_STACK_TOP.fixup_alt_jump
2695 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2697 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2698 this_group_regnum = COMPILE_STACK_TOP.regnum;
2699 /* If we've reached MAX_REGNUM groups, then this open
2700 won't actually generate any code, so we'll have to
2701 clear pending_exact explicitly. */
2704 /* We're at the end of the group, so now we know how many
2705 groups were inside this one. */
2706 if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0)
2707 BUF_PUSH_2 (stop_memory, this_group_regnum);
2712 case '|': /* `\|'. */
2713 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2714 goto normal_backslash;
2716 if (syntax & RE_LIMITED_OPS)
2719 /* Insert before the previous alternative a jump which
2720 jumps to this alternative if the former fails. */
2721 GET_BUFFER_SPACE (3);
2722 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2726 /* The alternative before this one has a jump after it
2727 which gets executed if it gets matched. Adjust that
2728 jump so it will jump to this alternative's analogous
2729 jump (put in below, which in turn will jump to the next
2730 (if any) alternative's such jump, etc.). The last such
2731 jump jumps to the correct final destination. A picture:
2737 If we are at `b', then fixup_alt_jump right now points to a
2738 three-byte space after `a'. We'll put in the jump, set
2739 fixup_alt_jump to right after `b', and leave behind three
2740 bytes which we'll fill in when we get to after `c'. */
2744 /* Mark and leave space for a jump after this alternative,
2745 to be filled in later either by next alternative or
2746 when know we're at the end of a series of alternatives. */
2748 GET_BUFFER_SPACE (3);
2757 /* If \{ is a literal. */
2758 if (!(syntax & RE_INTERVALS)
2759 /* If we're at `\{' and it's not the open-interval
2761 || (syntax & RE_NO_BK_BRACES)
2762 /* What is that? -sm */
2763 /* || (p - 2 == pattern && p == pend) */)
2764 goto normal_backslash;
2768 /* If got here, then the syntax allows intervals. */
2770 /* At least (most) this many matches must be made. */
2771 int lower_bound = 0, upper_bound = -1;
2777 if (syntax & RE_NO_BK_BRACES)
2778 goto unfetch_interval;
2780 FREE_STACK_RETURN (REG_EBRACE);
2783 GET_UNSIGNED_NUMBER (lower_bound);
2786 GET_UNSIGNED_NUMBER (upper_bound);
2788 /* Interval such as `{1}' => match exactly once. */
2789 upper_bound = lower_bound;
2791 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2792 || (upper_bound >= 0 && lower_bound > upper_bound))
2794 if (syntax & RE_NO_BK_BRACES)
2795 goto unfetch_interval;
2797 FREE_STACK_RETURN (REG_BADBR);
2800 if (!(syntax & RE_NO_BK_BRACES))
2802 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2809 if (syntax & RE_NO_BK_BRACES)
2810 goto unfetch_interval;
2812 FREE_STACK_RETURN (REG_BADBR);
2815 /* We just parsed a valid interval. */
2817 /* If it's invalid to have no preceding re. */
2820 if (syntax & RE_CONTEXT_INVALID_OPS)
2821 FREE_STACK_RETURN (REG_BADRPT);
2822 else if (syntax & RE_CONTEXT_INDEP_OPS)
2825 goto unfetch_interval;
2828 if (upper_bound == 0)
2829 /* If the upper bound is zero, just drop the sub pattern
2832 else if (lower_bound == 1 && upper_bound == 1)
2833 /* Just match it once: nothing to do here. */
2836 /* Otherwise, we have a nontrivial interval. When
2837 we're all done, the pattern will look like:
2838 set_number_at <jump count> <upper bound>
2839 set_number_at <succeed_n count> <lower bound>
2840 succeed_n <after jump addr> <succeed_n count>
2842 jump_n <succeed_n addr> <jump count>
2843 (The upper bound and `jump_n' are omitted if
2844 `upper_bound' is 1, though.) */
2846 { /* If the upper bound is > 1, we need to insert
2847 more at the end of the loop. */
2848 unsigned int nbytes = (upper_bound < 0 ? 3
2849 : upper_bound > 1 ? 5 : 0);
2850 unsigned int startoffset = 0;
2852 GET_BUFFER_SPACE (20); /* We might use less. */
2854 if (lower_bound == 0)
2856 /* A succeed_n that starts with 0 is really a
2857 a simple on_failure_jump_loop. */
2858 INSERT_JUMP (on_failure_jump_loop, laststart,
2864 /* Initialize lower bound of the `succeed_n', even
2865 though it will be set during matching by its
2866 attendant `set_number_at' (inserted next),
2867 because `re_compile_fastmap' needs to know.
2868 Jump to the `jump_n' we might insert below. */
2869 INSERT_JUMP2 (succeed_n, laststart,
2874 /* Code to initialize the lower bound. Insert
2875 before the `succeed_n'. The `5' is the last two
2876 bytes of this `set_number_at', plus 3 bytes of
2877 the following `succeed_n'. */
2878 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2883 if (upper_bound < 0)
2885 /* A negative upper bound stands for infinity,
2886 in which case it degenerates to a plain jump. */
2887 STORE_JUMP (jump, b, laststart + startoffset);
2890 else if (upper_bound > 1)
2891 { /* More than one repetition is allowed, so
2892 append a backward jump to the `succeed_n'
2893 that starts this interval.
2895 When we've reached this during matching,
2896 we'll have matched the interval once, so
2897 jump back only `upper_bound - 1' times. */
2898 STORE_JUMP2 (jump_n, b, laststart + startoffset,
2902 /* The location we want to set is the second
2903 parameter of the `jump_n'; that is `b-2' as
2904 an absolute address. `laststart' will be
2905 the `set_number_at' we're about to insert;
2906 `laststart+3' the number to set, the source
2907 for the relative address. But we are
2908 inserting into the middle of the pattern --
2909 so everything is getting moved up by 5.
2910 Conclusion: (b - 2) - (laststart + 3) + 5,
2911 i.e., b - laststart.
2913 We insert this at the beginning of the loop
2914 so that if we fail during matching, we'll
2915 reinitialize the bounds. */
2916 insert_op2 (set_number_at, laststart, b - laststart,
2917 upper_bound - 1, b);
2922 beg_interval = NULL;
2927 /* If an invalid interval, match the characters as literals. */
2928 assert (beg_interval);
2930 beg_interval = NULL;
2932 /* normal_char and normal_backslash need `c'. */
2935 if (!(syntax & RE_NO_BK_BRACES))
2937 assert (p > pattern && p[-1] == '\\');
2938 goto normal_backslash;
2944 /* There is no way to specify the before_dot and after_dot
2945 operators. rms says this is ok. --karl */
2953 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2959 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2965 BUF_PUSH_2 (categoryspec, c);
2971 BUF_PUSH_2 (notcategoryspec, c);
2978 BUF_PUSH_2 (syntaxspec, Sword);
2984 BUF_PUSH_2 (notsyntaxspec, Sword);
2997 BUF_PUSH (wordbound);
3001 BUF_PUSH (notwordbound);
3012 case '1': case '2': case '3': case '4': case '5':
3013 case '6': case '7': case '8': case '9':
3014 if (syntax & RE_NO_BK_REFS)
3020 FREE_STACK_RETURN (REG_ESUBREG);
3022 /* Can't back reference to a subexpression if inside of it. */
3023 if (group_in_compile_stack (compile_stack, c1))
3027 BUF_PUSH_2 (duplicate, c1);
3033 if (syntax & RE_BK_PLUS_QM)
3036 goto normal_backslash;
3040 /* You might think it would be useful for \ to mean
3041 not to translate; but if we don't translate it
3042 it will never match anything. */
3050 /* Expects the character in `c'. */
3052 /* If no exactn currently being built. */
3055 /* If last exactn not at current position. */
3056 || pending_exact + *pending_exact + 1 != b
3058 /* We have only one byte following the exactn for the count. */
3059 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3061 /* If followed by a repetition operator. */
3062 || (p != pend && (*p == '*' || *p == '^'))
3063 || ((syntax & RE_BK_PLUS_QM)
3064 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3065 : p != pend && (*p == '+' || *p == '?'))
3066 || ((syntax & RE_INTERVALS)
3067 && ((syntax & RE_NO_BK_BRACES)
3068 ? p != pend && *p == '{'
3069 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3071 /* Start building a new exactn. */
3075 BUF_PUSH_2 (exactn, 0);
3076 pending_exact = b - 1;
3079 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3084 len = CHAR_STRING (c, b);
3088 (*pending_exact) += len;
3093 } /* while p != pend */
3096 /* Through the pattern now. */
3100 if (!COMPILE_STACK_EMPTY)
3101 FREE_STACK_RETURN (REG_EPAREN);
3103 /* If we don't want backtracking, force success
3104 the first time we reach the end of the compiled pattern. */
3105 if (syntax & RE_NO_POSIX_BACKTRACKING)
3108 free (compile_stack.stack);
3110 /* We have succeeded; set the length of the buffer. */
3111 bufp->used = b - bufp->buffer;
3116 re_compile_fastmap (bufp);
3117 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3118 print_compiled_pattern (bufp);
3123 #ifndef MATCH_MAY_ALLOCATE
3124 /* Initialize the failure stack to the largest possible stack. This
3125 isn't necessary unless we're trying to avoid calling alloca in
3126 the search and match routines. */
3128 int num_regs = bufp->re_nsub + 1;
3130 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3132 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3134 if (! fail_stack.stack)
3136 = (fail_stack_elt_t *) malloc (fail_stack.size
3137 * sizeof (fail_stack_elt_t));
3140 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3142 * sizeof (fail_stack_elt_t)));
3145 regex_grow_registers (num_regs);
3147 #endif /* not MATCH_MAY_ALLOCATE */
3150 } /* regex_compile */
3152 /* Subroutines for `regex_compile'. */
3154 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3157 store_op1 (op, loc, arg)
3162 *loc = (unsigned char) op;
3163 STORE_NUMBER (loc + 1, arg);
3167 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3170 store_op2 (op, loc, arg1, arg2)
3175 *loc = (unsigned char) op;
3176 STORE_NUMBER (loc + 1, arg1);
3177 STORE_NUMBER (loc + 3, arg2);
3181 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3182 for OP followed by two-byte integer parameter ARG. */
3185 insert_op1 (op, loc, arg, end)
3191 register unsigned char *pfrom = end;
3192 register unsigned char *pto = end + 3;
3194 while (pfrom != loc)
3197 store_op1 (op, loc, arg);
3201 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3204 insert_op2 (op, loc, arg1, arg2, end)
3210 register unsigned char *pfrom = end;
3211 register unsigned char *pto = end + 5;
3213 while (pfrom != loc)
3216 store_op2 (op, loc, arg1, arg2);
3220 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3221 after an alternative or a begin-subexpression. We assume there is at
3222 least one character before the ^. */
3225 at_begline_loc_p (pattern, p, syntax)
3226 const unsigned char *pattern, *p;
3227 reg_syntax_t syntax;
3229 const unsigned char *prev = p - 2;
3230 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3233 /* After a subexpression? */
3234 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3235 /* After an alternative? */
3236 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3237 /* After a shy subexpression? */
3238 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3239 && prev[-1] == '?' && prev[-2] == '('
3240 && (syntax & RE_NO_BK_PARENS
3241 || (prev - 3 >= pattern && prev[-3] == '\\')));
3245 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3246 at least one character after the $, i.e., `P < PEND'. */
3249 at_endline_loc_p (p, pend, syntax)
3250 const unsigned char *p, *pend;
3251 reg_syntax_t syntax;
3253 const unsigned char *next = p;
3254 boolean next_backslash = *next == '\\';
3255 const unsigned char *next_next = p + 1 < pend ? p + 1 : 0;
3258 /* Before a subexpression? */
3259 (syntax & RE_NO_BK_PARENS ? *next == ')'
3260 : next_backslash && next_next && *next_next == ')')
3261 /* Before an alternative? */
3262 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3263 : next_backslash && next_next && *next_next == '|');
3267 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3268 false if it's not. */
3271 group_in_compile_stack (compile_stack, regnum)
3272 compile_stack_type compile_stack;
3277 for (this_element = compile_stack.avail - 1;
3280 if (compile_stack.stack[this_element].regnum == regnum)
3287 If fastmap is non-NULL, go through the pattern and fill fastmap
3288 with all the possible leading chars. If fastmap is NULL, don't
3289 bother filling it up (obviously) and only return whether the
3290 pattern could potentially match the empty string.
3292 Return 1 if p..pend might match the empty string.
3293 Return 0 if p..pend matches at least one char.
3294 Return -1 if p..pend matches at least one char, but fastmap was not
3296 Return -2 if an error occurred. */
3299 analyse_first (p, pend, fastmap, multibyte)
3300 unsigned char *p, *pend;
3302 const int multibyte;
3306 #ifdef MATCH_MAY_ALLOCATE
3307 fail_stack_type fail_stack;
3309 #ifndef REGEX_MALLOC
3313 #if defined REL_ALLOC && defined REGEX_MALLOC
3314 /* This holds the pointer to the failure stack, when
3315 it is allocated relocatably. */
3316 fail_stack_elt_t *failure_stack_ptr;
3319 /* Assume that each path through the pattern can be null until
3320 proven otherwise. We set this false at the bottom of switch
3321 statement, to which we get only if a particular path doesn't
3322 match the empty string. */
3323 boolean path_can_be_null = true;
3325 /* If all elements for base leading-codes in fastmap is set, this
3326 flag is set true. */
3327 boolean match_any_multibyte_characters = false;
3333 /* The loop below works as follows:
3334 - It has a working-list kept in the PATTERN_STACK and which basically
3335 starts by only containing a pointer to the first operation.
3336 - If the opcode we're looking at is a match against some set of
3337 chars, then we add those chars to the fastmap and go on to the
3338 next work element from the worklist (done via `break').
3339 - If the opcode is a control operator on the other hand, we either
3340 ignore it (if it's meaningless at this point, such as `start_memory')
3341 or execute it (if it's a jump). If the jump has several destinations
3342 (i.e. `on_failure_jump'), then we push the other destination onto the
3344 We guarantee termination by ignoring backward jumps (more or less),
3345 so that `p' is monotonically increasing. More to the point, we
3346 never set `p' (or push) anything `<= p1'. */
3348 /* If can_be_null is set, then the fastmap will not be used anyway. */
3351 /* `p1' is used as a marker of how far back a `on_failure_jump'
3352 can go without being ignored. It is normally equal to `p'
3353 (which prevents any backward `on_failure_jump') except right
3354 after a plain `jump', to allow patterns such as:
3357 10: on_failure_jump 3
3358 as used for the *? operator. */
3359 unsigned char *p1 = p;
3363 if (path_can_be_null)
3364 return (RESET_FAIL_STACK (), 1);
3366 /* We have reached the (effective) end of pattern. */
3367 if (PATTERN_STACK_EMPTY ())
3368 return (RESET_FAIL_STACK (), 0);
3370 p = (unsigned char*) POP_PATTERN_OP ();
3371 path_can_be_null = true;
3375 /* We should never be about to go beyond the end of the pattern. */
3378 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3385 /* If the first character has to match a backreference, that means
3386 that the group was empty (since it already matched). Since this
3387 is the only case that interests us here, we can assume that the
3388 backreference must match the empty string. */
3393 /* Following are the cases which match a character. These end
3399 int c = RE_STRING_CHAR (p + 1, pend - p);
3401 if (SINGLE_BYTE_CHAR_P (c))
3410 /* We could put all the chars except for \n (and maybe \0)
3411 but we don't bother since it is generally not worth it. */
3412 if (!fastmap) break;
3413 return (RESET_FAIL_STACK (), -1);
3417 /* Chars beyond end of bitmap are possible matches.
3418 All the single-byte codes can occur in multibyte buffers.
3419 So any that are not listed in the charset
3420 are possible matches, even in multibyte buffers. */
3421 if (!fastmap) break;
3422 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3423 j < (1 << BYTEWIDTH); j++)
3427 if (!fastmap) break;
3428 not = (re_opcode_t) *(p - 1) == charset_not;
3429 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3431 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3434 if ((not && multibyte)
3435 /* Any character set can possibly contain a character
3436 which doesn't match the specified set of characters. */
3437 || (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3438 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
3439 /* If we can match a character class, we can match
3440 any character set. */
3442 set_fastmap_for_multibyte_characters:
3443 if (match_any_multibyte_characters == false)
3445 for (j = 0x80; j < 0xA0; j++) /* XXX */
3446 if (BASE_LEADING_CODE_P (j))
3448 match_any_multibyte_characters = true;
3452 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3453 && match_any_multibyte_characters == false)
3455 /* Set fastmap[I] 1 where I is a base leading code of each
3456 multibyte character in the range table. */
3459 /* Make P points the range table. `+ 2' is to skip flag
3460 bits for a character class. */
3461 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
3463 /* Extract the number of ranges in range table into COUNT. */
3464 EXTRACT_NUMBER_AND_INCR (count, p);
3465 for (; count > 0; count--, p += 2 * 3) /* XXX */
3467 /* Extract the start of each range. */
3468 EXTRACT_CHARACTER (c, p);
3469 j = CHAR_CHARSET (c);
3470 fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1;
3477 if (!fastmap) break;
3479 not = (re_opcode_t)p[-1] == notsyntaxspec;
3481 for (j = 0; j < (1 << BYTEWIDTH); j++)
3482 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
3486 /* This match depends on text properties. These end with
3487 aborting optimizations. */
3488 return (RESET_FAIL_STACK (), -1);
3491 case notcategoryspec:
3492 if (!fastmap) break;
3493 not = (re_opcode_t)p[-1] == notcategoryspec;
3495 for (j = 0; j < (1 << BYTEWIDTH); j++)
3496 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
3500 /* Any character set can possibly contain a character
3501 whose category is K (or not). */
3502 goto set_fastmap_for_multibyte_characters;
3505 /* All cases after this match the empty string. These end with
3525 EXTRACT_NUMBER_AND_INCR (j, p);
3527 /* Backward jumps can only go back to code that we've already
3528 visited. `re_compile' should make sure this is true. */
3531 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
3533 case on_failure_jump:
3534 case on_failure_keep_string_jump:
3535 case on_failure_jump_loop:
3536 case on_failure_jump_nastyloop:
3537 case on_failure_jump_smart:
3543 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
3544 to jump back to "just after here". */
3547 case on_failure_jump:
3548 case on_failure_keep_string_jump:
3549 case on_failure_jump_nastyloop:
3550 case on_failure_jump_loop:
3551 case on_failure_jump_smart:
3552 EXTRACT_NUMBER_AND_INCR (j, p);
3554 ; /* Backward jump to be ignored. */
3555 else if (!PUSH_PATTERN_OP (p + j, fail_stack))
3556 return (RESET_FAIL_STACK (), -2);
3561 /* This code simply does not properly handle forward jump_n. */
3562 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
3564 /* jump_n can either jump or fall through. The (backward) jump
3565 case has already been handled, so we only need to look at the
3566 fallthrough case. */
3570 /* If N == 0, it should be an on_failure_jump_loop instead. */
3571 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
3573 /* We only care about one iteration of the loop, so we don't
3574 need to consider the case where this behaves like an
3591 abort (); /* We have listed all the cases. */
3594 /* Getting here means we have found the possible starting
3595 characters for one path of the pattern -- and that the empty
3596 string does not match. We need not follow this path further.
3597 Instead, look at the next alternative (remembered on the
3598 stack), or quit if no more. The test at the top of the loop
3599 does these things. */
3600 path_can_be_null = false;
3604 return (RESET_FAIL_STACK (), 0);
3605 } /* analyse_first */
3607 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3608 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3609 characters can start a string that matches the pattern. This fastmap
3610 is used by re_search to skip quickly over impossible starting points.
3612 Character codes above (1 << BYTEWIDTH) are not represented in the
3613 fastmap, but the leading codes are represented. Thus, the fastmap
3614 indicates which character sets could start a match.
3616 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3617 area as BUFP->fastmap.
3619 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3622 Returns 0 if we succeed, -2 if an internal error. */
3625 re_compile_fastmap (bufp)
3626 struct re_pattern_buffer *bufp;
3628 char *fastmap = bufp->fastmap;
3631 assert (fastmap && bufp->buffer);
3633 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3634 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3636 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
3637 fastmap, RE_MULTIBYTE_P (bufp));
3640 bufp->can_be_null = (analysis != 0);
3642 } /* re_compile_fastmap */
3644 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3645 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3646 this memory for recording register information. STARTS and ENDS
3647 must be allocated using the malloc library routine, and must each
3648 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3650 If NUM_REGS == 0, then subsequent matches should allocate their own
3653 Unless this function is called, the first search or match using
3654 PATTERN_BUFFER will allocate its own register data, without
3655 freeing the old data. */
3658 re_set_registers (bufp, regs, num_regs, starts, ends)
3659 struct re_pattern_buffer *bufp;
3660 struct re_registers *regs;
3662 regoff_t *starts, *ends;
3666 bufp->regs_allocated = REGS_REALLOCATE;
3667 regs->num_regs = num_regs;
3668 regs->start = starts;
3673 bufp->regs_allocated = REGS_UNALLOCATED;
3675 regs->start = regs->end = (regoff_t *) 0;
3679 /* Searching routines. */
3681 /* Like re_search_2, below, but only one string is specified, and
3682 doesn't let you say where to stop matching. */
3685 re_search (bufp, string, size, startpos, range, regs)
3686 struct re_pattern_buffer *bufp;
3688 int size, startpos, range;
3689 struct re_registers *regs;
3691 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3695 /* End address of virtual concatenation of string. */
3696 #define STOP_ADDR_VSTRING(P) \
3697 (((P) >= size1 ? string2 + size2 : string1 + size1))
3699 /* Address of POS in the concatenation of virtual string. */
3700 #define POS_ADDR_VSTRING(POS) \
3701 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
3703 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3704 virtual concatenation of STRING1 and STRING2, starting first at index
3705 STARTPOS, then at STARTPOS + 1, and so on.
3707 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3709 RANGE is how far to scan while trying to match. RANGE = 0 means try
3710 only at STARTPOS; in general, the last start tried is STARTPOS +
3713 In REGS, return the indices of the virtual concatenation of STRING1
3714 and STRING2 that matched the entire BUFP->buffer and its contained
3717 Do not consider matching one past the index STOP in the virtual
3718 concatenation of STRING1 and STRING2.
3720 We return either the position in the strings at which the match was
3721 found, -1 if no match, or -2 if error (such as failure
3725 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
3726 struct re_pattern_buffer *bufp;
3727 const char *str1, *str2;
3731 struct re_registers *regs;
3735 re_char *string1 = (re_char*) str1;
3736 re_char *string2 = (re_char*) str2;
3737 register char *fastmap = bufp->fastmap;
3738 register RE_TRANSLATE_TYPE translate = bufp->translate;
3739 int total_size = size1 + size2;
3740 int endpos = startpos + range;
3741 int anchored_start = 0;
3743 /* Nonzero if we have to concern multibyte character. */
3744 const boolean multibyte = RE_MULTIBYTE_P (bufp);
3746 /* Check for out-of-range STARTPOS. */
3747 if (startpos < 0 || startpos > total_size)
3750 /* Fix up RANGE if it might eventually take us outside
3751 the virtual concatenation of STRING1 and STRING2.
3752 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3754 range = 0 - startpos;
3755 else if (endpos > total_size)
3756 range = total_size - startpos;
3758 /* If the search isn't to be a backwards one, don't waste time in a
3759 search for a pattern anchored at beginning of buffer. */
3760 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3769 /* In a forward search for something that starts with \=.
3770 don't keep searching past point. */
3771 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3773 range = PT_BYTE - BEGV_BYTE - startpos;
3779 /* Update the fastmap now if not correct already. */
3780 if (fastmap && !bufp->fastmap_accurate)
3781 if (re_compile_fastmap (bufp) == -2)
3784 /* See whether the pattern is anchored. */
3785 if (bufp->buffer[0] == begline)
3789 gl_state.object = re_match_object;
3791 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
3793 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
3797 /* Loop through the string, looking for a place to start matching. */
3800 /* If the pattern is anchored,
3801 skip quickly past places we cannot match.
3802 We don't bother to treat startpos == 0 specially
3803 because that case doesn't repeat. */
3804 if (anchored_start && startpos > 0)
3806 if (! (bufp->newline_anchor
3807 && ((startpos <= size1 ? string1[startpos - 1]
3808 : string2[startpos - size1 - 1])
3813 /* If a fastmap is supplied, skip quickly over characters that
3814 cannot be the start of a match. If the pattern can match the
3815 null string, however, we don't need to skip characters; we want
3816 the first null string. */
3817 if (fastmap && startpos < total_size && !bufp->can_be_null)
3819 register re_char *d;
3820 register unsigned int buf_ch;
3822 d = POS_ADDR_VSTRING (startpos);
3824 if (range > 0) /* Searching forwards. */
3826 register int lim = 0;
3829 if (startpos < size1 && startpos + range >= size1)
3830 lim = range - (size1 - startpos);
3832 /* Written out as an if-else to avoid testing `translate'
3834 if (RE_TRANSLATE_P (translate))
3841 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
3844 buf_ch = RE_TRANSLATE (translate, buf_ch);
3849 range -= buf_charlen;
3854 && !fastmap[RE_TRANSLATE (translate, *d)])
3861 while (range > lim && !fastmap[*d])
3867 startpos += irange - range;
3869 else /* Searching backwards. */
3871 int room = (startpos >= size1
3872 ? size2 + size1 - startpos
3873 : size1 - startpos);
3874 buf_ch = RE_STRING_CHAR (d, room);
3875 buf_ch = TRANSLATE (buf_ch);
3877 if (! (buf_ch >= 0400
3878 || fastmap[buf_ch]))
3883 /* If can't match the null string, and that's all we have left, fail. */
3884 if (range >= 0 && startpos == total_size && fastmap
3885 && !bufp->can_be_null)
3888 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3889 startpos, regs, stop);
3890 #ifndef REGEX_MALLOC
3907 /* Update STARTPOS to the next character boundary. */
3910 re_char *p = POS_ADDR_VSTRING (startpos);
3911 re_char *pend = STOP_ADDR_VSTRING (startpos);
3912 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
3930 /* Update STARTPOS to the previous character boundary. */
3933 re_char *p = POS_ADDR_VSTRING (startpos);
3936 /* Find the head of multibyte form. */
3937 while (!CHAR_HEAD_P (*p))
3942 if (MULTIBYTE_FORM_LENGTH (p, len + 1) != (len + 1))
3959 /* Declarations and macros for re_match_2. */
3961 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
3963 RE_TRANSLATE_TYPE translate,
3964 const int multibyte));
3966 /* This converts PTR, a pointer into one of the search strings `string1'
3967 and `string2' into an offset from the beginning of that string. */
3968 #define POINTER_TO_OFFSET(ptr) \
3969 (FIRST_STRING_P (ptr) \
3970 ? ((regoff_t) ((ptr) - string1)) \
3971 : ((regoff_t) ((ptr) - string2 + size1)))
3973 /* Call before fetching a character with *d. This switches over to
3974 string2 if necessary.
3975 Check re_match_2_internal for a discussion of why end_match_2 might
3976 not be within string2 (but be equal to end_match_1 instead). */
3977 #define PREFETCH() \
3980 /* End of string2 => fail. */ \
3981 if (dend == end_match_2) \
3983 /* End of string1 => advance to string2. */ \
3985 dend = end_match_2; \
3988 /* Call before fetching a char with *d if you already checked other limits.
3989 This is meant for use in lookahead operations like wordend, etc..
3990 where we might need to look at parts of the string that might be
3991 outside of the LIMITs (i.e past `stop'). */
3992 #define PREFETCH_NOLIMIT() \
3996 dend = end_match_2; \
3999 /* Test if at very beginning or at very end of the virtual concatenation
4000 of `string1' and `string2'. If only one string, it's `string2'. */
4001 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4002 #define AT_STRINGS_END(d) ((d) == end2)
4005 /* Test if D points to a character which is word-constituent. We have
4006 two special cases to check for: if past the end of string1, look at
4007 the first character in string2; and if before the beginning of
4008 string2, look at the last character in string1. */
4009 #define WORDCHAR_P(d) \
4010 (SYNTAX ((d) == end1 ? *string2 \
4011 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4014 /* Disabled due to a compiler bug -- see comment at case wordbound */
4016 /* The comment at case wordbound is following one, but we don't use
4017 AT_WORD_BOUNDARY anymore to support multibyte form.
4019 The DEC Alpha C compiler 3.x generates incorrect code for the
4020 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4021 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4022 macro and introducing temporary variables works around the bug. */
4025 /* Test if the character before D and the one at D differ with respect
4026 to being word-constituent. */
4027 #define AT_WORD_BOUNDARY(d) \
4028 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4029 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4032 /* Free everything we malloc. */
4033 #ifdef MATCH_MAY_ALLOCATE
4034 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4035 # define FREE_VARIABLES() \
4037 REGEX_FREE_STACK (fail_stack.stack); \
4038 FREE_VAR (regstart); \
4039 FREE_VAR (regend); \
4040 FREE_VAR (best_regstart); \
4041 FREE_VAR (best_regend); \
4044 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4045 #endif /* not MATCH_MAY_ALLOCATE */
4048 /* Optimization routines. */
4050 /* If the operation is a match against one or more chars,
4051 return a pointer to the next operation, else return NULL. */
4052 static unsigned char *
4056 switch (SWITCH_ENUM_CAST (*p++))
4067 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4070 p = CHARSET_RANGE_TABLE (p - 1);
4071 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4072 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4075 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4082 case notcategoryspec:
4094 /* Jump over non-matching operations. */
4095 static unsigned char *
4096 skip_noops (p, pend)
4097 unsigned char *p, *pend;
4102 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4111 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4122 /* Non-zero if "p1 matches something" implies "p2 fails". */
4124 mutually_exclusive_p (bufp, p1, p2)
4125 struct re_pattern_buffer *bufp;
4126 unsigned char *p1, *p2;
4129 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4130 unsigned char *pend = bufp->buffer + bufp->used;
4132 assert (p1 >= bufp->buffer && p1 < pend
4133 && p2 >= bufp->buffer && p2 <= pend);
4135 /* Skip over open/close-group commands.
4136 If what follows this loop is a ...+ construct,
4137 look at what begins its body, since we will have to
4138 match at least one of that. */
4139 p2 = skip_noops (p2, pend);
4140 /* The same skip can be done for p1, except that this function
4141 is only used in the case where p1 is a simple match operator. */
4142 /* p1 = skip_noops (p1, pend); */
4144 assert (p1 >= bufp->buffer && p1 < pend
4145 && p2 >= bufp->buffer && p2 <= pend);
4147 op2 = p2 == pend ? succeed : *p2;
4149 switch (SWITCH_ENUM_CAST (op2))
4153 /* If we're at the end of the pattern, we can change. */
4154 if (skip_one_char (p1))
4156 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4162 if (!bufp->newline_anchor)
4167 register unsigned int c
4168 = (re_opcode_t) *p2 == endline ? '\n'
4169 : RE_STRING_CHAR(p2 + 2, pend - p2 - 2);
4171 if ((re_opcode_t) *p1 == exactn)
4173 if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2))
4175 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4180 else if ((re_opcode_t) *p1 == charset
4181 || (re_opcode_t) *p1 == charset_not)
4183 int not = (re_opcode_t) *p1 == charset_not;
4185 /* Test if C is listed in charset (or charset_not)
4187 if (SINGLE_BYTE_CHAR_P (c))
4189 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4190 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4193 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4194 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4196 /* `not' is equal to 1 if c would match, which means
4197 that we can't change to pop_failure_jump. */
4200 DEBUG_PRINT1 (" No match => fast loop.\n");
4204 else if ((re_opcode_t) *p1 == anychar
4207 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4216 if ((re_opcode_t) *p1 == exactn)
4217 /* Reuse the code above. */
4218 return mutually_exclusive_p (bufp, p2, p1);
4221 /* It is hard to list up all the character in charset
4222 P2 if it includes multibyte character. Give up in
4224 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4226 /* Now, we are sure that P2 has no range table.
4227 So, for the size of bitmap in P2, `p2[1]' is
4228 enough. But P1 may have range table, so the
4229 size of bitmap table of P1 is extracted by
4230 using macro `CHARSET_BITMAP_SIZE'.
4232 Since we know that all the character listed in
4233 P2 is ASCII, it is enough to test only bitmap
4239 /* We win if the charset inside the loop
4240 has no overlap with the one after the loop. */
4243 && idx < CHARSET_BITMAP_SIZE (p1));
4245 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4249 || idx == CHARSET_BITMAP_SIZE (p1))
4251 DEBUG_PRINT1 (" No match => fast loop.\n");
4255 else if ((re_opcode_t) *p1 == charset
4256 || (re_opcode_t) *p1 == charset_not)
4259 /* We win if the charset_not inside the loop lists
4260 every character listed in the charset after. */
4261 for (idx = 0; idx < (int) p2[1]; idx++)
4262 if (! (p2[2 + idx] == 0
4263 || (idx < CHARSET_BITMAP_SIZE (p1)
4264 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4269 DEBUG_PRINT1 (" No match => fast loop.\n");
4278 return ((re_opcode_t) *p1 == syntaxspec
4279 && p1[1] == (op2 == wordend ? Sword : p2[1]));
4283 return ((re_opcode_t) *p1 == notsyntaxspec
4284 && p1[1] == (op2 == wordend ? Sword : p2[1]));
4287 return (((re_opcode_t) *p1 == notsyntaxspec
4288 || (re_opcode_t) *p1 == syntaxspec)
4293 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4294 case notcategoryspec:
4295 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4307 /* Matching routines. */
4309 #ifndef emacs /* Emacs never uses this. */
4310 /* re_match is like re_match_2 except it takes only a single string. */
4313 re_match (bufp, string, size, pos, regs)
4314 struct re_pattern_buffer *bufp;
4317 struct re_registers *regs;
4319 int result = re_match_2_internal (bufp, NULL, 0, string, size,
4321 # if defined C_ALLOCA && !defined REGEX_MALLOC
4326 #endif /* not emacs */
4329 /* In Emacs, this is the string or buffer in which we
4330 are matching. It is used for looking up syntax properties. */
4331 Lisp_Object re_match_object;
4334 /* re_match_2 matches the compiled pattern in BUFP against the
4335 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4336 and SIZE2, respectively). We start matching at POS, and stop
4339 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4340 store offsets for the substring each group matched in REGS. See the
4341 documentation for exactly how many groups we fill.
4343 We return -1 if no match, -2 if an internal error (such as the
4344 failure stack overflowing). Otherwise, we return the length of the
4345 matched substring. */
4348 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
4349 struct re_pattern_buffer *bufp;
4350 const char *string1, *string2;
4353 struct re_registers *regs;
4360 gl_state.object = re_match_object;
4361 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4362 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4365 result = re_match_2_internal (bufp, string1, size1, string2, size2,
4367 #if defined C_ALLOCA && !defined REGEX_MALLOC
4373 /* This is a separate function so that we can force an alloca cleanup
4376 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
4377 struct re_pattern_buffer *bufp;
4378 re_char *string1, *string2;
4381 struct re_registers *regs;
4384 /* General temporaries. */
4389 /* Just past the end of the corresponding string. */
4390 re_char *end1, *end2;
4392 /* Pointers into string1 and string2, just past the last characters in
4393 each to consider matching. */
4394 re_char *end_match_1, *end_match_2;
4396 /* Where we are in the data, and the end of the current string. */
4399 /* Used sometimes to remember where we were before starting matching
4400 an operator so that we can go back in case of failure. This "atomic"
4401 behavior of matching opcodes is indispensable to the correctness
4402 of the on_failure_keep_string_jump optimization. */
4405 /* Where we are in the pattern, and the end of the pattern. */
4406 unsigned char *p = bufp->buffer;
4407 register unsigned char *pend = p + bufp->used;
4409 /* We use this to map every character in the string. */
4410 RE_TRANSLATE_TYPE translate = bufp->translate;
4412 /* Nonzero if we have to concern multibyte character. */
4413 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4415 /* Failure point stack. Each place that can handle a failure further
4416 down the line pushes a failure point on this stack. It consists of
4417 regstart, and regend for all registers corresponding to
4418 the subexpressions we're currently inside, plus the number of such
4419 registers, and, finally, two char *'s. The first char * is where
4420 to resume scanning the pattern; the second one is where to resume
4421 scanning the strings. */
4422 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4423 fail_stack_type fail_stack;
4426 static unsigned failure_id = 0;
4427 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4430 #if defined REL_ALLOC && defined REGEX_MALLOC
4431 /* This holds the pointer to the failure stack, when
4432 it is allocated relocatably. */
4433 fail_stack_elt_t *failure_stack_ptr;
4436 /* We fill all the registers internally, independent of what we
4437 return, for use in backreferences. The number here includes
4438 an element for register zero. */
4439 unsigned num_regs = bufp->re_nsub + 1;
4441 /* Information on the contents of registers. These are pointers into
4442 the input strings; they record just what was matched (on this
4443 attempt) by a subexpression part of the pattern, that is, the
4444 regnum-th regstart pointer points to where in the pattern we began
4445 matching and the regnum-th regend points to right after where we
4446 stopped matching the regnum-th subexpression. (The zeroth register
4447 keeps track of what the whole pattern matches.) */
4448 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4449 re_char **regstart, **regend;
4452 /* The following record the register info as found in the above
4453 variables when we find a match better than any we've seen before.
4454 This happens as we backtrack through the failure points, which in
4455 turn happens only if we have not yet matched the entire string. */
4456 unsigned best_regs_set = false;
4457 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4458 re_char **best_regstart, **best_regend;
4461 /* Logically, this is `best_regend[0]'. But we don't want to have to
4462 allocate space for that if we're not allocating space for anything
4463 else (see below). Also, we never need info about register 0 for
4464 any of the other register vectors, and it seems rather a kludge to
4465 treat `best_regend' differently than the rest. So we keep track of
4466 the end of the best match so far in a separate variable. We
4467 initialize this to NULL so that when we backtrack the first time
4468 and need to test it, it's not garbage. */
4469 re_char *match_end = NULL;
4472 /* Counts the total number of registers pushed. */
4473 unsigned num_regs_pushed = 0;
4476 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4480 #ifdef MATCH_MAY_ALLOCATE
4481 /* Do not bother to initialize all the register variables if there are
4482 no groups in the pattern, as it takes a fair amount of time. If
4483 there are groups, we include space for register 0 (the whole
4484 pattern), even though we never use it, since it simplifies the
4485 array indexing. We should fix this. */
4488 regstart = REGEX_TALLOC (num_regs, re_char *);
4489 regend = REGEX_TALLOC (num_regs, re_char *);
4490 best_regstart = REGEX_TALLOC (num_regs, re_char *);
4491 best_regend = REGEX_TALLOC (num_regs, re_char *);
4493 if (!(regstart && regend && best_regstart && best_regend))
4501 /* We must initialize all our variables to NULL, so that
4502 `FREE_VARIABLES' doesn't try to free them. */
4503 regstart = regend = best_regstart = best_regend = NULL;
4505 #endif /* MATCH_MAY_ALLOCATE */
4507 /* The starting position is bogus. */
4508 if (pos < 0 || pos > size1 + size2)
4514 /* Initialize subexpression text positions to -1 to mark ones that no
4515 start_memory/stop_memory has been seen for. Also initialize the
4516 register information struct. */
4517 for (mcnt = 1; mcnt < num_regs; mcnt++)
4518 regstart[mcnt] = regend[mcnt] = REG_UNSET_VALUE;
4520 /* We move `string1' into `string2' if the latter's empty -- but not if
4521 `string1' is null. */
4522 if (size2 == 0 && string1 != NULL)
4529 end1 = string1 + size1;
4530 end2 = string2 + size2;
4532 /* `p' scans through the pattern as `d' scans through the data.
4533 `dend' is the end of the input string that `d' points within. `d'
4534 is advanced into the following input string whenever necessary, but
4535 this happens before fetching; therefore, at the beginning of the
4536 loop, `d' can be pointing at the end of a string, but it cannot
4540 /* Only match within string2. */
4541 d = string2 + pos - size1;
4542 dend = end_match_2 = string2 + stop - size1;
4543 end_match_1 = end1; /* Just to give it a value. */
4549 /* Only match within string1. */
4550 end_match_1 = string1 + stop;
4552 When we reach end_match_1, PREFETCH normally switches to string2.
4553 But in the present case, this means that just doing a PREFETCH
4554 makes us jump from `stop' to `gap' within the string.
4555 What we really want here is for the search to stop as
4556 soon as we hit end_match_1. That's why we set end_match_2
4557 to end_match_1 (since PREFETCH fails as soon as we hit
4559 end_match_2 = end_match_1;
4562 { /* It's important to use this code when stop == size so that
4563 moving `d' from end1 to string2 will not prevent the d == dend
4564 check from catching the end of string. */
4566 end_match_2 = string2 + stop - size1;
4572 DEBUG_PRINT1 ("The compiled pattern is: ");
4573 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
4574 DEBUG_PRINT1 ("The string to match is: `");
4575 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
4576 DEBUG_PRINT1 ("'\n");
4578 /* This loops over pattern commands. It exits by returning from the
4579 function if the match is complete, or it drops through if the match
4580 fails at this starting point in the input data. */
4583 DEBUG_PRINT2 ("\n%p: ", p);
4586 { /* End of pattern means we might have succeeded. */
4587 DEBUG_PRINT1 ("end of pattern ... ");
4589 /* If we haven't matched the entire string, and we want the
4590 longest match, try backtracking. */
4591 if (d != end_match_2)
4593 /* 1 if this match ends in the same string (string1 or string2)
4594 as the best previous match. */
4595 boolean same_str_p = (FIRST_STRING_P (match_end)
4596 == FIRST_STRING_P (d));
4597 /* 1 if this match is the best seen so far. */
4598 boolean best_match_p;
4600 /* AIX compiler got confused when this was combined
4601 with the previous declaration. */
4603 best_match_p = d > match_end;
4605 best_match_p = !FIRST_STRING_P (d);
4607 DEBUG_PRINT1 ("backtracking.\n");
4609 if (!FAIL_STACK_EMPTY ())
4610 { /* More failure points to try. */
4612 /* If exceeds best match so far, save it. */
4613 if (!best_regs_set || best_match_p)
4615 best_regs_set = true;
4618 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4620 for (mcnt = 1; mcnt < num_regs; mcnt++)
4622 best_regstart[mcnt] = regstart[mcnt];
4623 best_regend[mcnt] = regend[mcnt];
4629 /* If no failure points, don't restore garbage. And if
4630 last match is real best match, don't restore second
4632 else if (best_regs_set && !best_match_p)
4635 /* Restore best match. It may happen that `dend ==
4636 end_match_1' while the restored d is in string2.
4637 For example, the pattern `x.*y.*z' against the
4638 strings `x-' and `y-z-', if the two strings are
4639 not consecutive in memory. */
4640 DEBUG_PRINT1 ("Restoring best registers.\n");
4643 dend = ((d >= string1 && d <= end1)
4644 ? end_match_1 : end_match_2);
4646 for (mcnt = 1; mcnt < num_regs; mcnt++)
4648 regstart[mcnt] = best_regstart[mcnt];
4649 regend[mcnt] = best_regend[mcnt];
4652 } /* d != end_match_2 */
4655 DEBUG_PRINT1 ("Accepting match.\n");
4657 /* If caller wants register contents data back, do it. */
4658 if (regs && !bufp->no_sub)
4660 /* Have the register data arrays been allocated? */
4661 if (bufp->regs_allocated == REGS_UNALLOCATED)
4662 { /* No. So allocate them with malloc. We need one
4663 extra element beyond `num_regs' for the `-1' marker
4665 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4666 regs->start = TALLOC (regs->num_regs, regoff_t);
4667 regs->end = TALLOC (regs->num_regs, regoff_t);
4668 if (regs->start == NULL || regs->end == NULL)
4673 bufp->regs_allocated = REGS_REALLOCATE;
4675 else if (bufp->regs_allocated == REGS_REALLOCATE)
4676 { /* Yes. If we need more elements than were already
4677 allocated, reallocate them. If we need fewer, just
4679 if (regs->num_regs < num_regs + 1)
4681 regs->num_regs = num_regs + 1;
4682 RETALLOC (regs->start, regs->num_regs, regoff_t);
4683 RETALLOC (regs->end, regs->num_regs, regoff_t);
4684 if (regs->start == NULL || regs->end == NULL)
4693 /* These braces fend off a "empty body in an else-statement"
4694 warning under GCC when assert expands to nothing. */
4695 assert (bufp->regs_allocated == REGS_FIXED);
4698 /* Convert the pointer data in `regstart' and `regend' to
4699 indices. Register zero has to be set differently,
4700 since we haven't kept track of any info for it. */
4701 if (regs->num_regs > 0)
4703 regs->start[0] = pos;
4704 regs->end[0] = POINTER_TO_OFFSET (d);
4707 /* Go through the first `min (num_regs, regs->num_regs)'
4708 registers, since that is all we initialized. */
4709 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
4711 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4712 regs->start[mcnt] = regs->end[mcnt] = -1;
4716 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4718 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4722 /* If the regs structure we return has more elements than
4723 were in the pattern, set the extra elements to -1. If
4724 we (re)allocated the registers, this is the case,
4725 because we always allocate enough to have at least one
4727 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
4728 regs->start[mcnt] = regs->end[mcnt] = -1;
4729 } /* regs && !bufp->no_sub */
4731 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4732 nfailure_points_pushed, nfailure_points_popped,
4733 nfailure_points_pushed - nfailure_points_popped);
4734 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4736 mcnt = POINTER_TO_OFFSET (d) - pos;
4738 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4744 /* Otherwise match next pattern command. */
4745 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4747 /* Ignore these. Used to ignore the n of succeed_n's which
4748 currently have n == 0. */
4750 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4754 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4757 /* Match the next n pattern characters exactly. The following
4758 byte in the pattern defines n, and the n bytes after that
4759 are the characters to match. */
4762 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4764 /* Remember the start point to rollback upon failure. */
4767 /* This is written out as an if-else so we don't waste time
4768 testing `translate' inside the loop. */
4769 if (RE_TRANSLATE_P (translate))
4774 int pat_charlen, buf_charlen;
4775 unsigned int pat_ch, buf_ch;
4778 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
4779 buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
4781 if (RE_TRANSLATE (translate, buf_ch)
4790 mcnt -= pat_charlen;
4797 if (RE_TRANSLATE (translate, *d) != *p++)
4822 /* Match any character except possibly a newline or a null. */
4826 unsigned int buf_ch;
4828 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4831 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
4832 buf_ch = TRANSLATE (buf_ch);
4834 if ((!(bufp->syntax & RE_DOT_NEWLINE)
4836 || ((bufp->syntax & RE_DOT_NOT_NULL)
4837 && buf_ch == '\000'))
4840 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4849 register unsigned int c;
4850 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4853 /* Start of actual range_table, or end of bitmap if there is no
4855 unsigned char *range_table;
4857 /* Nonzero if there is a range table. */
4858 int range_table_exists;
4860 /* Number of ranges of range table. This is not included
4861 in the initial byte-length of the command. */
4864 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4866 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
4868 if (range_table_exists)
4870 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
4871 EXTRACT_NUMBER_AND_INCR (count, range_table);
4875 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len);
4876 c = TRANSLATE (c); /* The character to match. */
4878 if (SINGLE_BYTE_CHAR_P (c))
4879 { /* Lookup bitmap. */
4880 /* Cast to `unsigned' instead of `unsigned char' in
4881 case the bit list is a full 32 bytes long. */
4882 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
4883 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4887 else if (range_table_exists)
4889 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
4891 if ( (class_bits & BIT_ALNUM && ISALNUM (c))
4892 | (class_bits & BIT_ALPHA && ISALPHA (c))
4893 | (class_bits & BIT_ASCII && IS_REAL_ASCII (c))
4894 | (class_bits & BIT_GRAPH && ISGRAPH (c))
4895 | (class_bits & BIT_LOWER && ISLOWER (c))
4896 | (class_bits & BIT_MULTIBYTE && !ISUNIBYTE (c))
4897 | (class_bits & BIT_NONASCII && !IS_REAL_ASCII (c))
4898 | (class_bits & BIT_PRINT && ISPRINT (c))
4899 | (class_bits & BIT_PUNCT && ISPUNCT (c))
4900 | (class_bits & BIT_SPACE && ISSPACE (c))
4901 | (class_bits & BIT_UNIBYTE && ISUNIBYTE (c))
4902 | (class_bits & BIT_UPPER && ISUPPER (c))
4903 | (class_bits & BIT_WORD && ISWORD (c)))
4906 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
4910 if (range_table_exists)
4911 p = CHARSET_RANGE_TABLE_END (range_table, count);
4913 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
4915 if (!not) goto fail;
4922 /* The beginning of a group is represented by start_memory.
4923 The argument is the register number. The text
4924 matched within the group is recorded (in the internal
4925 registers data structure) under the register number. */
4927 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
4929 /* In case we need to undo this operation (via backtracking). */
4930 PUSH_FAILURE_REG ((unsigned int)*p);
4933 regend[*p] = REG_UNSET_VALUE; /* probably unnecessary. -sm */
4934 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4936 /* Move past the register number and inner group count. */
4941 /* The stop_memory opcode represents the end of a group. Its
4942 argument is the same as start_memory's: the register number. */
4944 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
4946 assert (!REG_UNSET (regstart[*p]));
4947 /* Strictly speaking, there should be code such as:
4949 assert (REG_UNSET (regend[*p]));
4950 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
4952 But the only info to be pushed is regend[*p] and it is known to
4953 be UNSET, so there really isn't anything to push.
4954 Not pushing anything, on the other hand deprives us from the
4955 guarantee that regend[*p] is UNSET since undoing this operation
4956 will not reset its value properly. This is not important since
4957 the value will only be read on the next start_memory or at
4958 the very end and both events can only happen if this stop_memory
4962 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4964 /* Move past the register number and the inner group count. */
4969 /* \<digit> has been turned into a `duplicate' command which is
4970 followed by the numeric value of <digit> as the register number. */
4973 register re_char *d2, *dend2;
4974 int regno = *p++; /* Get which register to match against. */
4975 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4977 /* Can't back reference a group which we've never matched. */
4978 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4981 /* Where in input to try to start matching. */
4982 d2 = regstart[regno];
4984 /* Remember the start point to rollback upon failure. */
4987 /* Where to stop matching; if both the place to start and
4988 the place to stop matching are in the same string, then
4989 set to the place to stop, otherwise, for now have to use
4990 the end of the first string. */
4992 dend2 = ((FIRST_STRING_P (regstart[regno])
4993 == FIRST_STRING_P (regend[regno]))
4994 ? regend[regno] : end_match_1);
4997 /* If necessary, advance to next segment in register
5001 if (dend2 == end_match_2) break;
5002 if (dend2 == regend[regno]) break;
5004 /* End of string1 => advance to string2. */
5006 dend2 = regend[regno];
5008 /* At end of register contents => success */
5009 if (d2 == dend2) break;
5011 /* If necessary, advance to next segment in data. */
5014 /* How many characters left in this segment to match. */
5017 /* Want how many consecutive characters we can match in
5018 one shot, so, if necessary, adjust the count. */
5019 if (mcnt > dend2 - d2)
5022 /* Compare that many; failure if mismatch, else move
5024 if (RE_TRANSLATE_P (translate)
5025 ? bcmp_translate (d, d2, mcnt, translate, multibyte)
5026 : bcmp (d, d2, mcnt))
5031 d += mcnt, d2 += mcnt;
5037 /* begline matches the empty string at the beginning of the string
5038 (unless `not_bol' is set in `bufp'), and, if
5039 `newline_anchor' is set, after newlines. */
5041 DEBUG_PRINT1 ("EXECUTING begline.\n");
5043 if (AT_STRINGS_BEG (d))
5045 if (!bufp->not_bol) break;
5050 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5051 if (c == '\n' && bufp->newline_anchor)
5054 /* In all other cases, we fail. */
5058 /* endline is the dual of begline. */
5060 DEBUG_PRINT1 ("EXECUTING endline.\n");
5062 if (AT_STRINGS_END (d))
5064 if (!bufp->not_eol) break;
5068 PREFETCH_NOLIMIT ();
5069 if (*d == '\n' && bufp->newline_anchor)
5075 /* Match at the very beginning of the data. */
5077 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5078 if (AT_STRINGS_BEG (d))
5083 /* Match at the very end of the data. */
5085 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5086 if (AT_STRINGS_END (d))
5091 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5092 pushes NULL as the value for the string on the stack. Then
5093 `POP_FAILURE_POINT' will keep the current value for the
5094 string, instead of restoring it. To see why, consider
5095 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5096 then the . fails against the \n. But the next thing we want
5097 to do is match the \n against the \n; if we restored the
5098 string value, we would be back at the foo.
5100 Because this is used only in specific cases, we don't need to
5101 check all the things that `on_failure_jump' does, to make
5102 sure the right things get saved on the stack. Hence we don't
5103 share its code. The only reason to push anything on the
5104 stack at all is that otherwise we would have to change
5105 `anychar's code to do something besides goto fail in this
5106 case; that seems worse than this. */
5107 case on_failure_keep_string_jump:
5108 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5109 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5112 PUSH_FAILURE_POINT (p - 3, NULL);
5115 /* A nasty loop is introduced by the non-greedy *? and +?.
5116 With such loops, the stack only ever contains one failure point
5117 at a time, so that a plain on_failure_jump_loop kind of
5118 cycle detection cannot work. Worse yet, such a detection
5119 can not only fail to detect a cycle, but it can also wrongly
5120 detect a cycle (between different instantiations of the same
5122 So the method used for those nasty loops is a little different:
5123 We use a special cycle-detection-stack-frame which is pushed
5124 when the on_failure_jump_nastyloop failure-point is *popped*.
5125 This special frame thus marks the beginning of one iteration
5126 through the loop and we can hence easily check right here
5127 whether something matched between the beginning and the end of
5129 case on_failure_jump_nastyloop:
5130 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5131 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5134 assert ((re_opcode_t)p[-4] == no_op);
5135 CHECK_INFINITE_LOOP (p - 4, d);
5136 PUSH_FAILURE_POINT (p - 3, d);
5140 /* Simple loop detecting on_failure_jump: just check on the
5141 failure stack if the same spot was already hit earlier. */
5142 case on_failure_jump_loop:
5144 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5145 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5148 CHECK_INFINITE_LOOP (p - 3, d);
5149 PUSH_FAILURE_POINT (p - 3, d);
5153 /* Uses of on_failure_jump:
5155 Each alternative starts with an on_failure_jump that points
5156 to the beginning of the next alternative. Each alternative
5157 except the last ends with a jump that in effect jumps past
5158 the rest of the alternatives. (They really jump to the
5159 ending jump of the following alternative, because tensioning
5160 these jumps is a hassle.)
5162 Repeats start with an on_failure_jump that points past both
5163 the repetition text and either the following jump or
5164 pop_failure_jump back to this on_failure_jump. */
5165 case on_failure_jump:
5167 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5168 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5171 PUSH_FAILURE_POINT (p -3, d);
5174 /* This operation is used for greedy *.
5175 Compare the beginning of the repeat with what in the
5176 pattern follows its end. If we can establish that there
5177 is nothing that they would both match, i.e., that we
5178 would have to backtrack because of (as in, e.g., `a*a')
5179 then we can use a non-backtracking loop based on
5180 on_failure_keep_string_jump instead of on_failure_jump. */
5181 case on_failure_jump_smart:
5183 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5184 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5187 unsigned char *p1 = p; /* Next operation. */
5188 unsigned char *p2 = p + mcnt; /* Destination of the jump. */
5190 p -= 3; /* Reset so that we will re-execute the
5191 instruction once it's been changed. */
5193 EXTRACT_NUMBER (mcnt, p2 - 2);
5195 /* Ensure this is a indeed the trivial kind of loop
5196 we are expecting. */
5197 assert (skip_one_char (p1) == p2 - 3);
5198 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5199 DEBUG_STATEMENT (debug += 2);
5200 if (mutually_exclusive_p (bufp, p1, p2))
5202 /* Use a fast `on_failure_keep_string_jump' loop. */
5203 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5204 *p = (unsigned char) on_failure_keep_string_jump;
5205 STORE_NUMBER (p2 - 2, mcnt + 3);
5209 /* Default to a safe `on_failure_jump' loop. */
5210 DEBUG_PRINT1 (" smart default => slow loop.\n");
5211 *p = (unsigned char) on_failure_jump;
5213 DEBUG_STATEMENT (debug -= 2);
5217 /* Unconditionally jump (without popping any failure points). */
5221 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5222 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5223 p += mcnt; /* Do the jump. */
5224 DEBUG_PRINT2 ("(to %p).\n", p);
5228 /* Have to succeed matching what follows at least n times.
5229 After that, handle like `on_failure_jump'. */
5231 EXTRACT_NUMBER (mcnt, p + 2);
5232 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5234 /* Originally, mcnt is how many times we HAVE to succeed. */
5239 PUSH_FAILURE_COUNT (p);
5240 STORE_NUMBER_AND_INCR (p, mcnt);
5241 DEBUG_PRINT3 (" Setting %p to %d.\n", p, mcnt);
5244 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5249 EXTRACT_NUMBER (mcnt, p + 2);
5250 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5252 /* Originally, this is how many times we CAN jump. */
5256 PUSH_FAILURE_COUNT (p + 2);
5257 STORE_NUMBER (p + 2, mcnt);
5258 goto unconditional_jump;
5260 /* If don't have to jump any more, skip over the rest of command. */
5267 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5269 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5271 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5272 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
5273 PUSH_FAILURE_COUNT (p1);
5274 STORE_NUMBER (p1, mcnt);
5280 not = (re_opcode_t) *(p - 1) == notwordbound;
5281 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5283 /* We SUCCEED (or FAIL) in one of the following cases: */
5285 /* Case 1: D is at the beginning or the end of string. */
5286 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5290 /* C1 is the character before D, S1 is the syntax of C1, C2
5291 is the character at D, and S2 is the syntax of C2. */
5294 int offset = PTR_TO_OFFSET (d - 1);
5295 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5296 UPDATE_SYNTAX_TABLE (charpos);
5298 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5301 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
5303 PREFETCH_NOLIMIT ();
5304 c2 = RE_STRING_CHAR (d, dend - d);
5307 if (/* Case 2: Only one of S1 and S2 is Sword. */
5308 ((s1 == Sword) != (s2 == Sword))
5309 /* Case 3: Both of S1 and S2 are Sword, and macro
5310 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5311 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5320 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5322 /* We FAIL in one of the following cases: */
5324 /* Case 1: D is at the end of string. */
5325 if (AT_STRINGS_END (d))
5329 /* C1 is the character before D, S1 is the syntax of C1, C2
5330 is the character at D, and S2 is the syntax of C2. */
5333 int offset = PTR_TO_OFFSET (d);
5334 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5335 UPDATE_SYNTAX_TABLE (charpos);
5338 c2 = RE_STRING_CHAR (d, dend - d);
5341 /* Case 2: S2 is not Sword. */
5345 /* Case 3: D is not at the beginning of string ... */
5346 if (!AT_STRINGS_BEG (d))
5348 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5350 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
5354 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5356 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
5363 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5365 /* We FAIL in one of the following cases: */
5367 /* Case 1: D is at the beginning of string. */
5368 if (AT_STRINGS_BEG (d))
5372 /* C1 is the character before D, S1 is the syntax of C1, C2
5373 is the character at D, and S2 is the syntax of C2. */
5376 int offset = PTR_TO_OFFSET (d) - 1;
5377 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5378 UPDATE_SYNTAX_TABLE (charpos);
5380 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5383 /* Case 2: S1 is not Sword. */
5387 /* Case 3: D is not at the end of string ... */
5388 if (!AT_STRINGS_END (d))
5390 PREFETCH_NOLIMIT ();
5391 c2 = RE_STRING_CHAR (d, dend - d);
5393 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
5397 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5399 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
5407 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
5409 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
5413 int offset = PTR_TO_OFFSET (d);
5414 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5415 UPDATE_SYNTAX_TABLE (pos1);
5421 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len);
5423 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
5431 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5432 if (PTR_BYTE_POS (d) >= PT_BYTE)
5437 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5438 if (PTR_BYTE_POS (d) != PT_BYTE)
5443 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5444 if (PTR_BYTE_POS (d) <= PT_BYTE)
5449 case notcategoryspec:
5450 not = (re_opcode_t) *(p - 1) == notcategoryspec;
5452 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
5456 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len);
5458 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
5469 continue; /* Successfully executed one pattern command; keep going. */
5472 /* We goto here if a matching operation fails. */
5475 if (!FAIL_STACK_EMPTY ())
5479 /* A restart point is known. Restore to that state. */
5480 DEBUG_PRINT1 ("\nFAIL:\n");
5481 POP_FAILURE_POINT (str, pat);
5482 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
5484 case on_failure_keep_string_jump:
5485 assert (str == NULL);
5486 goto continue_failure_jump;
5488 case on_failure_jump_nastyloop:
5489 assert ((re_opcode_t)pat[-2] == no_op);
5490 PUSH_FAILURE_POINT (pat - 2, str);
5493 case on_failure_jump_loop:
5494 case on_failure_jump:
5497 continue_failure_jump:
5498 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
5503 /* A special frame used for nastyloops. */
5510 assert (p >= bufp->buffer && p <= pend);
5512 if (d >= string1 && d <= end1)
5516 break; /* Matching at this starting point really fails. */
5520 goto restore_best_regs;
5524 return -1; /* Failure to match. */
5527 /* Subroutine definitions for re_match_2. */
5529 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5530 bytes; nonzero otherwise. */
5533 bcmp_translate (s1, s2, len, translate, multibyte)
5536 RE_TRANSLATE_TYPE translate;
5537 const int multibyte;
5539 register re_char *p1 = s1, *p2 = s2;
5540 re_char *p1_end = s1 + len;
5541 re_char *p2_end = s2 + len;
5543 while (p1 != p1_end && p2 != p2_end)
5545 int p1_charlen, p2_charlen;
5548 p1_ch = RE_STRING_CHAR_AND_LENGTH (p1, p1_end - p1, p1_charlen);
5549 p2_ch = RE_STRING_CHAR_AND_LENGTH (p2, p2_end - p2, p2_charlen);
5551 if (RE_TRANSLATE (translate, p1_ch)
5552 != RE_TRANSLATE (translate, p2_ch))
5555 p1 += p1_charlen, p2 += p2_charlen;
5558 if (p1 != p1_end || p2 != p2_end)
5564 /* Entry points for GNU code. */
5566 /* re_compile_pattern is the GNU regular expression compiler: it
5567 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5568 Returns 0 if the pattern was valid, otherwise an error string.
5570 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5571 are set in BUFP on entry.
5573 We call regex_compile to do the actual compilation. */
5576 re_compile_pattern (pattern, length, bufp)
5577 const char *pattern;
5579 struct re_pattern_buffer *bufp;
5583 /* GNU code is written to assume at least RE_NREGS registers will be set
5584 (and at least one extra will be -1). */
5585 bufp->regs_allocated = REGS_UNALLOCATED;
5587 /* And GNU code determines whether or not to get register information
5588 by passing null for the REGS argument to re_match, etc., not by
5592 /* Match anchors at newline. */
5593 bufp->newline_anchor = 1;
5595 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5599 return gettext (re_error_msgid[(int) ret]);
5602 /* Entry points compatible with 4.2 BSD regex library. We don't define
5603 them unless specifically requested. */
5605 #if defined _REGEX_RE_COMP || defined _LIBC
5607 /* BSD has one and only one pattern buffer. */
5608 static struct re_pattern_buffer re_comp_buf;
5612 /* Make these definitions weak in libc, so POSIX programs can redefine
5613 these names if they don't use our functions, and still use
5614 regcomp/regexec below without link errors. */
5624 if (!re_comp_buf.buffer)
5625 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5626 return (char *) gettext ("No previous regular expression");
5630 if (!re_comp_buf.buffer)
5632 re_comp_buf.buffer = (unsigned char *) malloc (200);
5633 if (re_comp_buf.buffer == NULL)
5634 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5635 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
5636 re_comp_buf.allocated = 200;
5638 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5639 if (re_comp_buf.fastmap == NULL)
5640 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5641 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
5644 /* Since `re_exec' always passes NULL for the `regs' argument, we
5645 don't need to initialize the pattern buffer fields which affect it. */
5647 /* Match anchors at newlines. */
5648 re_comp_buf.newline_anchor = 1;
5650 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5655 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5656 return (char *) gettext (re_error_msgid[(int) ret]);
5667 const int len = strlen (s);
5669 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5671 #endif /* _REGEX_RE_COMP */
5673 /* POSIX.2 functions. Don't define these for Emacs. */
5677 /* regcomp takes a regular expression as a string and compiles it.
5679 PREG is a regex_t *. We do not expect any fields to be initialized,
5680 since POSIX says we shouldn't. Thus, we set
5682 `buffer' to the compiled pattern;
5683 `used' to the length of the compiled pattern;
5684 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5685 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5686 RE_SYNTAX_POSIX_BASIC;
5687 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5688 `fastmap' and `fastmap_accurate' to zero;
5689 `re_nsub' to the number of subexpressions in PATTERN.
5691 PATTERN is the address of the pattern string.
5693 CFLAGS is a series of bits which affect compilation.
5695 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5696 use POSIX basic syntax.
5698 If REG_NEWLINE is set, then . and [^...] don't match newline.
5699 Also, regexec will try a match beginning after every newline.
5701 If REG_ICASE is set, then we considers upper- and lowercase
5702 versions of letters to be equivalent when matching.
5704 If REG_NOSUB is set, then when PREG is passed to regexec, that
5705 routine will report only success or failure, and nothing about the
5708 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5709 the return codes and their meanings.) */
5712 regcomp (preg, pattern, cflags)
5714 const char *pattern;
5719 = (cflags & REG_EXTENDED) ?
5720 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5722 /* regex_compile will allocate the space for the compiled pattern. */
5724 preg->allocated = 0;
5727 /* Don't bother to use a fastmap when searching. This simplifies the
5728 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5729 characters after newlines into the fastmap. This way, we just try
5733 if (cflags & REG_ICASE)
5738 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5739 * sizeof (*(RE_TRANSLATE_TYPE)0));
5740 if (preg->translate == NULL)
5741 return (int) REG_ESPACE;
5743 /* Map uppercase characters to corresponding lowercase ones. */
5744 for (i = 0; i < CHAR_SET_SIZE; i++)
5745 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5748 preg->translate = NULL;
5750 /* If REG_NEWLINE is set, newlines are treated differently. */
5751 if (cflags & REG_NEWLINE)
5752 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5753 syntax &= ~RE_DOT_NEWLINE;
5754 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5755 /* It also changes the matching behavior. */
5756 preg->newline_anchor = 1;
5759 preg->newline_anchor = 0;
5761 preg->no_sub = !!(cflags & REG_NOSUB);
5763 /* POSIX says a null character in the pattern terminates it, so we
5764 can use strlen here in compiling the pattern. */
5765 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5767 /* POSIX doesn't distinguish between an unmatched open-group and an
5768 unmatched close-group: both are REG_EPAREN. */
5769 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5775 /* regexec searches for a given pattern, specified by PREG, in the
5778 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5779 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5780 least NMATCH elements, and we set them to the offsets of the
5781 corresponding matched substrings.
5783 EFLAGS specifies `execution flags' which affect matching: if
5784 REG_NOTBOL is set, then ^ does not match at the beginning of the
5785 string; if REG_NOTEOL is set, then $ does not match at the end.
5787 We return 0 if we find a match and REG_NOMATCH if not. */
5790 regexec (preg, string, nmatch, pmatch, eflags)
5791 const regex_t *preg;
5794 regmatch_t pmatch[];
5798 struct re_registers regs;
5799 regex_t private_preg;
5800 int len = strlen (string);
5801 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5803 private_preg = *preg;
5805 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5806 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5808 /* The user has told us exactly how many registers to return
5809 information about, via `nmatch'. We have to pass that on to the
5810 matching routines. */
5811 private_preg.regs_allocated = REGS_FIXED;
5815 regs.num_regs = nmatch;
5816 regs.start = TALLOC (nmatch, regoff_t);
5817 regs.end = TALLOC (nmatch, regoff_t);
5818 if (regs.start == NULL || regs.end == NULL)
5819 return (int) REG_NOMATCH;
5822 /* Perform the searching operation. */
5823 ret = re_search (&private_preg, string, len,
5824 /* start: */ 0, /* range: */ len,
5825 want_reg_info ? ®s : (struct re_registers *) 0);
5827 /* Copy the register information to the POSIX structure. */
5834 for (r = 0; r < nmatch; r++)
5836 pmatch[r].rm_so = regs.start[r];
5837 pmatch[r].rm_eo = regs.end[r];
5841 /* If we needed the temporary register info, free the space now. */
5846 /* We want zero return to mean success, unlike `re_search'. */
5847 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5851 /* Returns a message corresponding to an error code, ERRCODE, returned
5852 from either regcomp or regexec. We don't use PREG here. */
5855 regerror (errcode, preg, errbuf, errbuf_size)
5857 const regex_t *preg;
5865 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5866 /* Only error codes returned by the rest of the code should be passed
5867 to this routine. If we are given anything else, or if other regex
5868 code generates an invalid error code, then the program has a bug.
5869 Dump core so we can fix it. */
5872 msg = gettext (re_error_msgid[errcode]);
5874 msg_size = strlen (msg) + 1; /* Includes the null. */
5876 if (errbuf_size != 0)
5878 if (msg_size > errbuf_size)
5880 strncpy (errbuf, msg, errbuf_size - 1);
5881 errbuf[errbuf_size - 1] = 0;
5884 strcpy (errbuf, msg);
5891 /* Free dynamically allocated space used by PREG. */
5897 if (preg->buffer != NULL)
5898 free (preg->buffer);
5899 preg->buffer = NULL;
5901 preg->allocated = 0;
5904 if (preg->fastmap != NULL)
5905 free (preg->fastmap);
5906 preg->fastmap = NULL;
5907 preg->fastmap_accurate = 0;
5909 if (preg->translate != NULL)
5910 free (preg->translate);
5911 preg->translate = NULL;
5914 #endif /* not emacs */