1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993,94,95,96,97,98,99,2000 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 #if defined STDC_HEADERS && !defined emacs
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
49 /* Whether to use ISO C Amendment 1 wide char functions.
50 Those should not be used for Emacs since it uses its own. */
51 #define WIDE_CHAR_SUPPORT \
52 (defined _LIBC || HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
54 /* For platform which support the ISO C amendement 1 functionality we
55 support user defined character classes. */
57 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
63 /* We have to keep the namespace clean. */
64 # define regfree(preg) __regfree (preg)
65 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
66 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
67 # define regerror(errcode, preg, errbuf, errbuf_size) \
68 __regerror(errcode, preg, errbuf, errbuf_size)
69 # define re_set_registers(bu, re, nu, st, en) \
70 __re_set_registers (bu, re, nu, st, en)
71 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
72 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
73 # define re_match(bufp, string, size, pos, regs) \
74 __re_match (bufp, string, size, pos, regs)
75 # define re_search(bufp, string, size, startpos, range, regs) \
76 __re_search (bufp, string, size, startpos, range, regs)
77 # define re_compile_pattern(pattern, length, bufp) \
78 __re_compile_pattern (pattern, length, bufp)
79 # define re_set_syntax(syntax) __re_set_syntax (syntax)
80 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
81 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
82 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
84 /* Make sure we call libc's function even if the user overrides them. */
85 # define btowc __btowc
86 # define iswctype __iswctype
87 # define wctype __wctype
89 # define WEAK_ALIAS(a,b) weak_alias (a, b)
91 /* We are also using some library internals. */
92 # include <locale/localeinfo.h>
93 # include <locale/elem-hash.h>
94 # include <langinfo.h>
96 # define WEAK_ALIAS(a,b)
99 /* This is for other GNU distributions with internationalized messages. */
100 #if HAVE_LIBINTL_H || defined _LIBC
101 # include <libintl.h>
103 # define gettext(msgid) (msgid)
107 /* This define is so xgettext can find the internationalizable
109 # define gettext_noop(String) String
112 /* The `emacs' switch turns on certain matching commands
113 that make sense only in Emacs. */
119 /* Make syntax table lookup grant data in gl_state. */
120 # define SYNTAX_ENTRY_VIA_PROPERTY
123 # include "charset.h"
124 # include "category.h"
126 # define malloc xmalloc
127 # define realloc xrealloc
130 /* Converts the pointer to the char to BEG-based offset from the start. */
131 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
132 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
134 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
135 # define RE_STRING_CHAR(p, s) \
136 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
137 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
138 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
140 /* Set C a (possibly multibyte) character before P. P points into a
141 string which is the virtual concatenation of STR1 (which ends at
142 END1) or STR2 (which ends at END2). */
143 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
147 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
148 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
149 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
150 c = STRING_CHAR (dtemp, (p) - dtemp); \
153 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
157 #else /* not emacs */
159 /* If we are not linking with Emacs proper,
160 we can't use the relocating allocator
161 even if config.h says that we can. */
164 # if defined STDC_HEADERS || defined _LIBC
171 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
172 If nothing else has been done, use the method below. */
173 # ifdef INHIBIT_STRING_HEADER
174 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
175 # if !defined bzero && !defined bcopy
176 # undef INHIBIT_STRING_HEADER
181 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
182 This is used in most programs--a few other programs avoid this
183 by defining INHIBIT_STRING_HEADER. */
184 # ifndef INHIBIT_STRING_HEADER
185 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
189 # define bzero(s, n) (memset (s, '\0', n), (s))
191 # define bzero(s, n) __bzero (s, n)
195 # include <strings.h>
197 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
200 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
205 /* Define the syntax stuff for \<, \>, etc. */
207 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
208 enum syntaxcode { Swhitespace = 0, Sword = 1 };
210 # ifdef SWITCH_ENUM_BUG
211 # define SWITCH_ENUM_CAST(x) ((int)(x))
213 # define SWITCH_ENUM_CAST(x) (x)
216 /* Dummy macros for non-Emacs environments. */
217 # define BASE_LEADING_CODE_P(c) (0)
218 # define CHAR_CHARSET(c) 0
219 # define CHARSET_LEADING_CODE_BASE(c) 0
220 # define MAX_MULTIBYTE_LENGTH 1
221 # define RE_MULTIBYTE_P(x) 0
222 # define WORD_BOUNDARY_P(c1, c2) (0)
223 # define CHAR_HEAD_P(p) (1)
224 # define SINGLE_BYTE_CHAR_P(c) (1)
225 # define SAME_CHARSET_P(c1, c2) (1)
226 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
227 # define STRING_CHAR(p, s) (*(p))
228 # define RE_STRING_CHAR STRING_CHAR
229 # define CHAR_STRING(c, s) (*(s) = (c), 1)
230 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
231 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
232 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
233 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
234 # define MAKE_CHAR(charset, c1, c2) (c1)
235 #endif /* not emacs */
238 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
239 # define RE_TRANSLATE_P(TBL) (TBL)
242 /* Get the interface, including the syntax bits. */
245 /* isalpha etc. are used for the character classes. */
250 /* 1 if C is an ASCII character. */
251 # define IS_REAL_ASCII(c) ((c) < 0200)
253 /* 1 if C is a unibyte character. */
254 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
256 /* The Emacs definitions should not be directly affected by locales. */
258 /* In Emacs, these are only used for single-byte characters. */
259 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
260 # define ISCNTRL(c) ((c) < ' ')
261 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
262 || ((c) >= 'a' && (c) <= 'f') \
263 || ((c) >= 'A' && (c) <= 'F'))
265 /* This is only used for single-byte characters. */
266 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
268 /* The rest must handle multibyte characters. */
270 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
271 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
274 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
275 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
278 # define ISALNUM(c) (IS_REAL_ASCII (c) \
279 ? (((c) >= 'a' && (c) <= 'z') \
280 || ((c) >= 'A' && (c) <= 'Z') \
281 || ((c) >= '0' && (c) <= '9')) \
282 : SYNTAX (c) == Sword)
284 # define ISALPHA(c) (IS_REAL_ASCII (c) \
285 ? (((c) >= 'a' && (c) <= 'z') \
286 || ((c) >= 'A' && (c) <= 'Z')) \
287 : SYNTAX (c) == Sword)
289 # define ISLOWER(c) (LOWERCASEP (c))
291 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
292 ? ((c) > ' ' && (c) < 0177 \
293 && !(((c) >= 'a' && (c) <= 'z') \
294 || ((c) >= 'A' && (c) <= 'Z') \
295 || ((c) >= '0' && (c) <= '9'))) \
296 : SYNTAX (c) != Sword)
298 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
300 # define ISUPPER(c) (UPPERCASEP (c))
302 # define ISWORD(c) (SYNTAX (c) == Sword)
304 #else /* not emacs */
306 /* Jim Meyering writes:
308 "... Some ctype macros are valid only for character codes that
309 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
310 using /bin/cc or gcc but without giving an ansi option). So, all
311 ctype uses should be through macros like ISPRINT... If
312 STDC_HEADERS is defined, then autoconf has verified that the ctype
313 macros don't need to be guarded with references to isascii. ...
314 Defining isascii to 1 should let any compiler worth its salt
315 eliminate the && through constant folding."
316 Solaris defines some of these symbols so we must undefine them first. */
319 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
320 # define ISASCII(c) 1
322 # define ISASCII(c) isascii(c)
325 /* 1 if C is an ASCII character. */
326 # define IS_REAL_ASCII(c) ((c) < 0200)
328 /* This distinction is not meaningful, except in Emacs. */
329 # define ISUNIBYTE(c) 1
332 # define ISBLANK(c) (ISASCII (c) && isblank (c))
334 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
337 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
339 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
343 # define ISPRINT(c) (ISASCII (c) && isprint (c))
344 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
345 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
346 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
347 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
348 # define ISLOWER(c) (ISASCII (c) && islower (c))
349 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
350 # define ISSPACE(c) (ISASCII (c) && isspace (c))
351 # define ISUPPER(c) (ISASCII (c) && isupper (c))
352 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
354 # define ISWORD(c) ISALPHA(c)
357 # define TOLOWER(c) _tolower(c)
359 # define TOLOWER(c) tolower(c)
362 /* How many characters in the character set. */
363 # define CHAR_SET_SIZE 256
367 extern char *re_syntax_table;
369 # else /* not SYNTAX_TABLE */
371 static char re_syntax_table[CHAR_SET_SIZE];
382 bzero (re_syntax_table, sizeof re_syntax_table);
384 for (c = 0; c < CHAR_SET_SIZE; ++c)
386 re_syntax_table[c] = Sword;
388 re_syntax_table['_'] = Sword;
393 # endif /* not SYNTAX_TABLE */
395 # define SYNTAX(c) re_syntax_table[(c)]
397 #endif /* not emacs */
400 # define NULL (void *)0
403 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
404 since ours (we hope) works properly with all combinations of
405 machines, compilers, `char' and `unsigned char' argument types.
406 (Per Bothner suggested the basic approach.) */
407 #undef SIGN_EXTEND_CHAR
409 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
410 #else /* not __STDC__ */
411 /* As in Harbison and Steele. */
412 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
415 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
416 use `alloca' instead of `malloc'. This is because using malloc in
417 re_search* or re_match* could cause memory leaks when C-g is used in
418 Emacs; also, malloc is slower and causes storage fragmentation. On
419 the other hand, malloc is more portable, and easier to debug.
421 Because we sometimes use alloca, some routines have to be macros,
422 not functions -- `alloca'-allocated space disappears at the end of the
423 function it is called in. */
427 # define REGEX_ALLOCATE malloc
428 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
429 # define REGEX_FREE free
431 #else /* not REGEX_MALLOC */
433 /* Emacs already defines alloca, sometimes. */
436 /* Make alloca work the best possible way. */
438 # define alloca __builtin_alloca
439 # else /* not __GNUC__ */
442 # endif /* HAVE_ALLOCA_H */
443 # endif /* not __GNUC__ */
445 # endif /* not alloca */
447 # define REGEX_ALLOCATE alloca
449 /* Assumes a `char *destination' variable. */
450 # define REGEX_REALLOCATE(source, osize, nsize) \
451 (destination = (char *) alloca (nsize), \
452 memcpy (destination, source, osize))
454 /* No need to do anything to free, after alloca. */
455 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
457 #endif /* not REGEX_MALLOC */
459 /* Define how to allocate the failure stack. */
461 #if defined REL_ALLOC && defined REGEX_MALLOC
463 # define REGEX_ALLOCATE_STACK(size) \
464 r_alloc (&failure_stack_ptr, (size))
465 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
466 r_re_alloc (&failure_stack_ptr, (nsize))
467 # define REGEX_FREE_STACK(ptr) \
468 r_alloc_free (&failure_stack_ptr)
470 #else /* not using relocating allocator */
474 # define REGEX_ALLOCATE_STACK malloc
475 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
476 # define REGEX_FREE_STACK free
478 # else /* not REGEX_MALLOC */
480 # define REGEX_ALLOCATE_STACK alloca
482 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
483 REGEX_REALLOCATE (source, osize, nsize)
484 /* No need to explicitly free anything. */
485 # define REGEX_FREE_STACK(arg) ((void)0)
487 # endif /* not REGEX_MALLOC */
488 #endif /* not using relocating allocator */
491 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
492 `string1' or just past its end. This works if PTR is NULL, which is
494 #define FIRST_STRING_P(ptr) \
495 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
497 /* (Re)Allocate N items of type T using malloc, or fail. */
498 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
499 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
500 #define RETALLOC_IF(addr, n, t) \
501 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
502 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
504 #define BYTEWIDTH 8 /* In bits. */
506 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
510 #define MAX(a, b) ((a) > (b) ? (a) : (b))
511 #define MIN(a, b) ((a) < (b) ? (a) : (b))
513 /* Type of source-pattern and string chars. */
514 typedef const unsigned char re_char;
516 typedef char boolean;
520 static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
521 re_char *string1, int size1,
522 re_char *string2, int size2,
524 struct re_registers *regs,
527 /* These are the command codes that appear in compiled regular
528 expressions. Some opcodes are followed by argument bytes. A
529 command code can specify any interpretation whatsoever for its
530 arguments. Zero bytes may appear in the compiled regular expression. */
536 /* Succeed right away--no more backtracking. */
539 /* Followed by one byte giving n, then by n literal bytes. */
542 /* Matches any (more or less) character. */
545 /* Matches any one char belonging to specified set. First
546 following byte is number of bitmap bytes. Then come bytes
547 for a bitmap saying which chars are in. Bits in each byte
548 are ordered low-bit-first. A character is in the set if its
549 bit is 1. A character too large to have a bit in the map is
550 automatically not in the set.
552 If the length byte has the 0x80 bit set, then that stuff
553 is followed by a range table:
554 2 bytes of flags for character sets (low 8 bits, high 8 bits)
555 See RANGE_TABLE_WORK_BITS below.
556 2 bytes, the number of pairs that follow
557 pairs, each 2 multibyte characters,
558 each multibyte character represented as 3 bytes. */
561 /* Same parameters as charset, but match any character that is
562 not one of those specified. */
565 /* Start remembering the text that is matched, for storing in a
566 register. Followed by one byte with the register number, in
567 the range 0 to one less than the pattern buffer's re_nsub
571 /* Stop remembering the text that is matched and store it in a
572 memory register. Followed by one byte with the register
573 number, in the range 0 to one less than `re_nsub' in the
577 /* Match a duplicate of something remembered. Followed by one
578 byte containing the register number. */
581 /* Fail unless at beginning of line. */
584 /* Fail unless at end of line. */
587 /* Succeeds if at beginning of buffer (if emacs) or at beginning
588 of string to be matched (if not). */
591 /* Analogously, for end of buffer/string. */
594 /* Followed by two byte relative address to which to jump. */
597 /* Followed by two-byte relative address of place to resume at
598 in case of failure. */
601 /* Like on_failure_jump, but pushes a placeholder instead of the
602 current string position when executed. */
603 on_failure_keep_string_jump,
605 /* Just like `on_failure_jump', except that it checks that we
606 don't get stuck in an infinite loop (matching an empty string
608 on_failure_jump_loop,
610 /* Just like `on_failure_jump_loop', except that it checks for
611 a different kind of loop (the kind that shows up with non-greedy
612 operators). This operation has to be immediately preceded
614 on_failure_jump_nastyloop,
616 /* A smart `on_failure_jump' used for greedy * and + operators.
617 It analyses the loop before which it is put and if the
618 loop does not require backtracking, it changes itself to
619 `on_failure_keep_string_jump' and short-circuits the loop,
620 else it just defaults to changing itself into `on_failure_jump'.
621 It assumes that it is pointing to just past a `jump'. */
622 on_failure_jump_smart,
624 /* Followed by two-byte relative address and two-byte number n.
625 After matching N times, jump to the address upon failure.
626 Does not work if N starts at 0: use on_failure_jump_loop
630 /* Followed by two-byte relative address, and two-byte number n.
631 Jump to the address N times, then fail. */
634 /* Set the following two-byte relative address to the
635 subsequent two-byte number. The address *includes* the two
639 wordbeg, /* Succeeds if at word beginning. */
640 wordend, /* Succeeds if at word end. */
642 wordbound, /* Succeeds if at a word boundary. */
643 notwordbound, /* Succeeds if not at a word boundary. */
645 /* Matches any character whose syntax is specified. Followed by
646 a byte which contains a syntax code, e.g., Sword. */
649 /* Matches any character whose syntax is not that specified. */
653 ,before_dot, /* Succeeds if before point. */
654 at_dot, /* Succeeds if at point. */
655 after_dot, /* Succeeds if after point. */
657 /* Matches any character whose category-set contains the specified
658 category. The operator is followed by a byte which contains a
659 category code (mnemonic ASCII character). */
662 /* Matches any character whose category-set does not contain the
663 specified category. The operator is followed by a byte which
664 contains the category code (mnemonic ASCII character). */
669 /* Common operations on the compiled pattern. */
671 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
673 #define STORE_NUMBER(destination, number) \
675 (destination)[0] = (number) & 0377; \
676 (destination)[1] = (number) >> 8; \
679 /* Same as STORE_NUMBER, except increment DESTINATION to
680 the byte after where the number is stored. Therefore, DESTINATION
681 must be an lvalue. */
683 #define STORE_NUMBER_AND_INCR(destination, number) \
685 STORE_NUMBER (destination, number); \
686 (destination) += 2; \
689 /* Put into DESTINATION a number stored in two contiguous bytes starting
692 #define EXTRACT_NUMBER(destination, source) \
694 (destination) = *(source) & 0377; \
695 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
699 static void extract_number _RE_ARGS ((int *dest, re_char *source));
701 extract_number (dest, source)
703 unsigned char *source;
705 int temp = SIGN_EXTEND_CHAR (*(source + 1));
706 *dest = *source & 0377;
710 # ifndef EXTRACT_MACROS /* To debug the macros. */
711 # undef EXTRACT_NUMBER
712 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
713 # endif /* not EXTRACT_MACROS */
717 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
718 SOURCE must be an lvalue. */
720 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
722 EXTRACT_NUMBER (destination, source); \
727 static void extract_number_and_incr _RE_ARGS ((int *destination,
730 extract_number_and_incr (destination, source)
732 unsigned char **source;
734 extract_number (destination, *source);
738 # ifndef EXTRACT_MACROS
739 # undef EXTRACT_NUMBER_AND_INCR
740 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
741 extract_number_and_incr (&dest, &src)
742 # endif /* not EXTRACT_MACROS */
746 /* Store a multibyte character in three contiguous bytes starting
747 DESTINATION, and increment DESTINATION to the byte after where the
748 character is stored. Therefore, DESTINATION must be an lvalue. */
750 #define STORE_CHARACTER_AND_INCR(destination, character) \
752 (destination)[0] = (character) & 0377; \
753 (destination)[1] = ((character) >> 8) & 0377; \
754 (destination)[2] = (character) >> 16; \
755 (destination) += 3; \
758 /* Put into DESTINATION a character stored in three contiguous bytes
759 starting at SOURCE. */
761 #define EXTRACT_CHARACTER(destination, source) \
763 (destination) = ((source)[0] \
764 | ((source)[1] << 8) \
765 | ((source)[2] << 16)); \
769 /* Macros for charset. */
771 /* Size of bitmap of charset P in bytes. P is a start of charset,
772 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
773 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
775 /* Nonzero if charset P has range table. */
776 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
778 /* Return the address of range table of charset P. But not the start
779 of table itself, but the before where the number of ranges is
780 stored. `2 +' means to skip re_opcode_t and size of bitmap,
781 and the 2 bytes of flags at the start of the range table. */
782 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
784 /* Extract the bit flags that start a range table. */
785 #define CHARSET_RANGE_TABLE_BITS(p) \
786 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
787 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
789 /* Test if C is listed in the bitmap of charset P. */
790 #define CHARSET_LOOKUP_BITMAP(p, c) \
791 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
792 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
794 /* Return the address of end of RANGE_TABLE. COUNT is number of
795 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
796 is start of range and end of range. `* 3' is size of each start
798 #define CHARSET_RANGE_TABLE_END(range_table, count) \
799 ((range_table) + (count) * 2 * 3)
801 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
802 COUNT is number of ranges in RANGE_TABLE. */
803 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
806 int range_start, range_end; \
808 unsigned char *range_table_end \
809 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
811 for (p = (range_table); p < range_table_end; p += 2 * 3) \
813 EXTRACT_CHARACTER (range_start, p); \
814 EXTRACT_CHARACTER (range_end, p + 3); \
816 if (range_start <= (c) && (c) <= range_end) \
825 /* Test if C is in range table of CHARSET. The flag NOT is negated if
826 C is listed in it. */
827 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
830 /* Number of ranges in range table. */ \
832 unsigned char *range_table = CHARSET_RANGE_TABLE (charset); \
834 EXTRACT_NUMBER_AND_INCR (count, range_table); \
835 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
839 /* If DEBUG is defined, Regex prints many voluminous messages about what
840 it is doing (if the variable `debug' is nonzero). If linked with the
841 main program in `iregex.c', you can enter patterns and strings
842 interactively. And if linked with the main program in `main.c' and
843 the other test files, you can run the already-written tests. */
847 /* We use standard I/O for debugging. */
850 /* It is useful to test things that ``must'' be true when debugging. */
853 static int debug = -100000;
855 # define DEBUG_STATEMENT(e) e
856 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
857 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
858 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
859 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
860 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
861 if (debug > 0) print_partial_compiled_pattern (s, e)
862 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
863 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
866 /* Print the fastmap in human-readable form. */
869 print_fastmap (fastmap)
872 unsigned was_a_range = 0;
875 while (i < (1 << BYTEWIDTH))
881 while (i < (1 << BYTEWIDTH) && fastmap[i])
897 /* Print a compiled pattern string in human-readable form, starting at
898 the START pointer into it and ending just before the pointer END. */
901 print_partial_compiled_pattern (start, end)
902 unsigned char *start;
906 unsigned char *p = start;
907 unsigned char *pend = end;
915 /* Loop over pattern commands. */
918 printf ("%d:\t", p - start);
920 switch ((re_opcode_t) *p++)
932 printf ("/exactn/%d", mcnt);
942 printf ("/start_memory/%d", *p++);
946 printf ("/stop_memory/%d", *p++);
950 printf ("/duplicate/%d", *p++);
960 register int c, last = -100;
961 register int in_range = 0;
962 int length = CHARSET_BITMAP_SIZE (p - 1);
963 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
965 printf ("/charset [%s",
966 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
968 assert (p + *p < pend);
970 for (c = 0; c < 256; c++)
972 && (p[1 + (c/8)] & (1 << (c % 8))))
974 /* Are we starting a range? */
975 if (last + 1 == c && ! in_range)
980 /* Have we broken a range? */
981 else if (last + 1 != c && in_range)
1000 if (has_range_table)
1003 printf ("has-range-table");
1005 /* ??? Should print the range table; for now, just skip it. */
1006 p += 2; /* skip range table bits */
1007 EXTRACT_NUMBER_AND_INCR (count, p);
1008 p = CHARSET_RANGE_TABLE_END (p, count);
1014 printf ("/begline");
1018 printf ("/endline");
1021 case on_failure_jump:
1022 extract_number_and_incr (&mcnt, &p);
1023 printf ("/on_failure_jump to %d", p + mcnt - start);
1026 case on_failure_keep_string_jump:
1027 extract_number_and_incr (&mcnt, &p);
1028 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
1031 case on_failure_jump_nastyloop:
1032 extract_number_and_incr (&mcnt, &p);
1033 printf ("/on_failure_jump_nastyloop to %d", p + mcnt - start);
1036 case on_failure_jump_loop:
1037 extract_number_and_incr (&mcnt, &p);
1038 printf ("/on_failure_jump_loop to %d", p + mcnt - start);
1041 case on_failure_jump_smart:
1042 extract_number_and_incr (&mcnt, &p);
1043 printf ("/on_failure_jump_smart to %d", p + mcnt - start);
1047 extract_number_and_incr (&mcnt, &p);
1048 printf ("/jump to %d", p + mcnt - start);
1052 extract_number_and_incr (&mcnt, &p);
1053 extract_number_and_incr (&mcnt2, &p);
1054 printf ("/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1058 extract_number_and_incr (&mcnt, &p);
1059 extract_number_and_incr (&mcnt2, &p);
1060 printf ("/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1064 extract_number_and_incr (&mcnt, &p);
1065 extract_number_and_incr (&mcnt2, &p);
1066 printf ("/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1070 printf ("/wordbound");
1074 printf ("/notwordbound");
1078 printf ("/wordbeg");
1082 printf ("/wordend");
1085 printf ("/syntaxspec");
1087 printf ("/%d", mcnt);
1091 printf ("/notsyntaxspec");
1093 printf ("/%d", mcnt);
1098 printf ("/before_dot");
1106 printf ("/after_dot");
1110 printf ("/categoryspec");
1112 printf ("/%d", mcnt);
1115 case notcategoryspec:
1116 printf ("/notcategoryspec");
1118 printf ("/%d", mcnt);
1131 printf ("?%d", *(p-1));
1137 printf ("%d:\tend of pattern.\n", p - start);
1142 print_compiled_pattern (bufp)
1143 struct re_pattern_buffer *bufp;
1145 unsigned char *buffer = bufp->buffer;
1147 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1148 printf ("%ld bytes used/%ld bytes allocated.\n",
1149 bufp->used, bufp->allocated);
1151 if (bufp->fastmap_accurate && bufp->fastmap)
1153 printf ("fastmap: ");
1154 print_fastmap (bufp->fastmap);
1157 printf ("re_nsub: %d\t", bufp->re_nsub);
1158 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1159 printf ("can_be_null: %d\t", bufp->can_be_null);
1160 printf ("no_sub: %d\t", bufp->no_sub);
1161 printf ("not_bol: %d\t", bufp->not_bol);
1162 printf ("not_eol: %d\t", bufp->not_eol);
1163 printf ("syntax: %lx\n", bufp->syntax);
1165 /* Perhaps we should print the translate table? */
1170 print_double_string (where, string1, size1, string2, size2)
1183 if (FIRST_STRING_P (where))
1185 for (this_char = where - string1; this_char < size1; this_char++)
1186 putchar (string1[this_char]);
1191 for (this_char = where - string2; this_char < size2; this_char++)
1192 putchar (string2[this_char]);
1196 #else /* not DEBUG */
1201 # define DEBUG_STATEMENT(e)
1202 # define DEBUG_PRINT1(x)
1203 # define DEBUG_PRINT2(x1, x2)
1204 # define DEBUG_PRINT3(x1, x2, x3)
1205 # define DEBUG_PRINT4(x1, x2, x3, x4)
1206 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1207 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1209 #endif /* not DEBUG */
1211 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1212 also be assigned to arbitrarily: each pattern buffer stores its own
1213 syntax, so it can be changed between regex compilations. */
1214 /* This has no initializer because initialized variables in Emacs
1215 become read-only after dumping. */
1216 reg_syntax_t re_syntax_options;
1219 /* Specify the precise syntax of regexps for compilation. This provides
1220 for compatibility for various utilities which historically have
1221 different, incompatible syntaxes.
1223 The argument SYNTAX is a bit mask comprised of the various bits
1224 defined in regex.h. We return the old syntax. */
1227 re_set_syntax (syntax)
1228 reg_syntax_t syntax;
1230 reg_syntax_t ret = re_syntax_options;
1232 re_syntax_options = syntax;
1235 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1237 /* This table gives an error message for each of the error codes listed
1238 in regex.h. Obviously the order here has to be same as there.
1239 POSIX doesn't require that we do anything for REG_NOERROR,
1240 but why not be nice? */
1242 static const char *re_error_msgid[] =
1244 gettext_noop ("Success"), /* REG_NOERROR */
1245 gettext_noop ("No match"), /* REG_NOMATCH */
1246 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1247 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1248 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1249 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1250 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1251 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1252 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1253 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1254 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1255 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1256 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1257 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1258 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1259 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1260 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1263 /* Avoiding alloca during matching, to placate r_alloc. */
1265 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1266 searching and matching functions should not call alloca. On some
1267 systems, alloca is implemented in terms of malloc, and if we're
1268 using the relocating allocator routines, then malloc could cause a
1269 relocation, which might (if the strings being searched are in the
1270 ralloc heap) shift the data out from underneath the regexp
1273 Here's another reason to avoid allocation: Emacs
1274 processes input from X in a signal handler; processing X input may
1275 call malloc; if input arrives while a matching routine is calling
1276 malloc, then we're scrod. But Emacs can't just block input while
1277 calling matching routines; then we don't notice interrupts when
1278 they come in. So, Emacs blocks input around all regexp calls
1279 except the matching calls, which it leaves unprotected, in the
1280 faith that they will not malloc. */
1282 /* Normally, this is fine. */
1283 #define MATCH_MAY_ALLOCATE
1285 /* When using GNU C, we are not REALLY using the C alloca, no matter
1286 what config.h may say. So don't take precautions for it. */
1291 /* The match routines may not allocate if (1) they would do it with malloc
1292 and (2) it's not safe for them to use malloc.
1293 Note that if REL_ALLOC is defined, matching would not use malloc for the
1294 failure stack, but we would still use it for the register vectors;
1295 so REL_ALLOC should not affect this. */
1296 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1297 # undef MATCH_MAY_ALLOCATE
1301 /* Failure stack declarations and macros; both re_compile_fastmap and
1302 re_match_2 use a failure stack. These have to be macros because of
1303 REGEX_ALLOCATE_STACK. */
1306 /* Approximate number of failure points for which to initially allocate space
1307 when matching. If this number is exceeded, we allocate more
1308 space, so it is not a hard limit. */
1309 #ifndef INIT_FAILURE_ALLOC
1310 # define INIT_FAILURE_ALLOC 20
1313 /* Roughly the maximum number of failure points on the stack. Would be
1314 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1315 This is a variable only so users of regex can assign to it; we never
1316 change it ourselves. */
1317 # if defined MATCH_MAY_ALLOCATE
1318 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1319 whose default stack limit is 2mb. In order for a larger
1320 value to work reliably, you have to try to make it accord
1321 with the process stack limit. */
1322 size_t re_max_failures = 40000;
1324 size_t re_max_failures = 4000;
1327 union fail_stack_elt
1329 const unsigned char *pointer;
1330 /* This should be the biggest `int' that's no bigger than a pointer. */
1334 typedef union fail_stack_elt fail_stack_elt_t;
1338 fail_stack_elt_t *stack;
1340 size_t avail; /* Offset of next open position. */
1341 size_t frame; /* Offset of the cur constructed frame. */
1344 #define PATTERN_STACK_EMPTY() (fail_stack.avail == 0)
1345 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1346 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1349 /* Define macros to initialize and free the failure stack.
1350 Do `return -2' if the alloc fails. */
1352 #ifdef MATCH_MAY_ALLOCATE
1353 # define INIT_FAIL_STACK() \
1355 fail_stack.stack = (fail_stack_elt_t *) \
1356 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1357 * sizeof (fail_stack_elt_t)); \
1359 if (fail_stack.stack == NULL) \
1362 fail_stack.size = INIT_FAILURE_ALLOC; \
1363 fail_stack.avail = 0; \
1364 fail_stack.frame = 0; \
1367 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1369 # define INIT_FAIL_STACK() \
1371 fail_stack.avail = 0; \
1372 fail_stack.frame = 0; \
1375 # define RESET_FAIL_STACK() ((void)0)
1379 /* Double the size of FAIL_STACK, up to a limit
1380 which allows approximately `re_max_failures' items.
1382 Return 1 if succeeds, and 0 if either ran out of memory
1383 allocating space for it or it was already too large.
1385 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1387 /* Factor to increase the failure stack size by
1388 when we increase it.
1389 This used to be 2, but 2 was too wasteful
1390 because the old discarded stacks added up to as much space
1391 were as ultimate, maximum-size stack. */
1392 #define FAIL_STACK_GROWTH_FACTOR 4
1394 #define GROW_FAIL_STACK(fail_stack) \
1395 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1396 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1398 : ((fail_stack).stack \
1399 = (fail_stack_elt_t *) \
1400 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1401 (fail_stack).size * sizeof (fail_stack_elt_t), \
1402 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1403 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1404 * FAIL_STACK_GROWTH_FACTOR))), \
1406 (fail_stack).stack == NULL \
1408 : ((fail_stack).size \
1409 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1410 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1411 * FAIL_STACK_GROWTH_FACTOR)) \
1412 / sizeof (fail_stack_elt_t)), \
1416 /* Push pointer POINTER on FAIL_STACK.
1417 Return 1 if was able to do so and 0 if ran out of memory allocating
1419 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1420 ((FAIL_STACK_FULL () \
1421 && !GROW_FAIL_STACK (FAIL_STACK)) \
1423 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1425 #define POP_PATTERN_OP() POP_FAILURE_POINTER ()
1427 /* Push a pointer value onto the failure stack.
1428 Assumes the variable `fail_stack'. Probably should only
1429 be called from within `PUSH_FAILURE_POINT'. */
1430 #define PUSH_FAILURE_POINTER(item) \
1431 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1433 /* This pushes an integer-valued item onto the failure stack.
1434 Assumes the variable `fail_stack'. Probably should only
1435 be called from within `PUSH_FAILURE_POINT'. */
1436 #define PUSH_FAILURE_INT(item) \
1437 fail_stack.stack[fail_stack.avail++].integer = (item)
1439 /* Push a fail_stack_elt_t value onto the failure stack.
1440 Assumes the variable `fail_stack'. Probably should only
1441 be called from within `PUSH_FAILURE_POINT'. */
1442 #define PUSH_FAILURE_ELT(item) \
1443 fail_stack.stack[fail_stack.avail++] = (item)
1445 /* These three POP... operations complement the three PUSH... operations.
1446 All assume that `fail_stack' is nonempty. */
1447 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1448 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1449 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1451 /* Individual items aside from the registers. */
1452 #define NUM_NONREG_ITEMS 3
1454 /* Used to examine the stack (to detect infinite loops). */
1455 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1456 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1457 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1458 #define TOP_FAILURE_HANDLE() fail_stack.frame
1461 #define ENSURE_FAIL_STACK(space) \
1462 while (REMAINING_AVAIL_SLOTS <= space) { \
1463 if (!GROW_FAIL_STACK (fail_stack)) \
1465 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1466 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1469 /* Push register NUM onto the stack. */
1470 #define PUSH_FAILURE_REG(num) \
1472 char *destination; \
1473 ENSURE_FAIL_STACK(3); \
1474 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1475 num, regstart[num], regend[num]); \
1476 PUSH_FAILURE_POINTER (regstart[num]); \
1477 PUSH_FAILURE_POINTER (regend[num]); \
1478 PUSH_FAILURE_INT (num); \
1481 #define PUSH_FAILURE_COUNT(ptr) \
1483 char *destination; \
1485 ENSURE_FAIL_STACK(3); \
1486 EXTRACT_NUMBER (c, ptr); \
1487 DEBUG_PRINT3 (" Push counter %p = %d\n", ptr, c); \
1488 PUSH_FAILURE_INT (c); \
1489 PUSH_FAILURE_POINTER (ptr); \
1490 PUSH_FAILURE_INT (-1); \
1493 /* Pop a saved register off the stack. */
1494 #define POP_FAILURE_REG_OR_COUNT() \
1496 int reg = POP_FAILURE_INT (); \
1499 /* It's a counter. */ \
1500 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1501 reg = POP_FAILURE_INT (); \
1502 STORE_NUMBER (ptr, reg); \
1503 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1507 regend[reg] = POP_FAILURE_POINTER (); \
1508 regstart[reg] = POP_FAILURE_POINTER (); \
1509 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1510 reg, regstart[reg], regend[reg]); \
1514 /* Check that we are not stuck in an infinite loop. */
1515 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1517 int failure = TOP_FAILURE_HANDLE(); \
1518 /* Check for infinite matching loops */ \
1519 while (failure > 0 && \
1520 (FAILURE_STR (failure) == string_place \
1521 || FAILURE_STR (failure) == NULL)) \
1523 assert (FAILURE_PAT (failure) >= bufp->buffer \
1524 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1525 if (FAILURE_PAT (failure) == pat_cur) \
1527 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1528 failure = NEXT_FAILURE_HANDLE(failure); \
1530 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1533 /* Push the information about the state we will need
1534 if we ever fail back to it.
1536 Requires variables fail_stack, regstart, regend and
1537 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1540 Does `return FAILURE_CODE' if runs out of memory. */
1542 #define PUSH_FAILURE_POINT(pattern, string_place) \
1544 char *destination; \
1545 /* Must be int, so when we don't save any registers, the arithmetic \
1546 of 0 + -1 isn't done as unsigned. */ \
1548 DEBUG_STATEMENT (nfailure_points_pushed++); \
1549 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1550 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1551 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1553 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1555 DEBUG_PRINT1 ("\n"); \
1557 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1558 PUSH_FAILURE_INT (fail_stack.frame); \
1560 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1561 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1562 DEBUG_PRINT1 ("'\n"); \
1563 PUSH_FAILURE_POINTER (string_place); \
1565 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1566 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1567 PUSH_FAILURE_POINTER (pattern); \
1569 /* Close the frame by moving the frame pointer past it. */ \
1570 fail_stack.frame = fail_stack.avail; \
1573 /* Estimate the size of data pushed by a typical failure stack entry.
1574 An estimate is all we need, because all we use this for
1575 is to choose a limit for how big to make the failure stack. */
1577 #define TYPICAL_FAILURE_SIZE 20
1579 /* How many items can still be added to the stack without overflowing it. */
1580 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1583 /* Pops what PUSH_FAIL_STACK pushes.
1585 We restore into the parameters, all of which should be lvalues:
1586 STR -- the saved data position.
1587 PAT -- the saved pattern position.
1588 REGSTART, REGEND -- arrays of string positions.
1590 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1591 `pend', `string1', `size1', `string2', and `size2'. */
1593 #define POP_FAILURE_POINT(str, pat) \
1595 assert (!FAIL_STACK_EMPTY ()); \
1597 /* Remove failure points and point to how many regs pushed. */ \
1598 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1599 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1600 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1602 /* Pop the saved registers. */ \
1603 while (fail_stack.frame < fail_stack.avail) \
1604 POP_FAILURE_REG_OR_COUNT (); \
1606 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1607 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1608 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1610 /* If the saved string location is NULL, it came from an \
1611 on_failure_keep_string_jump opcode, and we want to throw away the \
1612 saved NULL, thus retaining our current position in the string. */ \
1613 str = (re_char *) POP_FAILURE_POINTER (); \
1614 DEBUG_PRINT2 (" Popping string %p: `", str); \
1615 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1616 DEBUG_PRINT1 ("'\n"); \
1618 fail_stack.frame = POP_FAILURE_INT (); \
1619 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1621 assert (fail_stack.avail >= 0); \
1622 assert (fail_stack.frame <= fail_stack.avail); \
1624 DEBUG_STATEMENT (nfailure_points_popped++); \
1625 } while (0) /* POP_FAILURE_POINT */
1629 /* Registers are set to a sentinel when they haven't yet matched. */
1630 #define REG_UNSET(e) ((e) == NULL)
1632 /* Subroutine declarations and macros for regex_compile. */
1634 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1635 reg_syntax_t syntax,
1636 struct re_pattern_buffer *bufp));
1637 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1638 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1639 int arg1, int arg2));
1640 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1641 int arg, unsigned char *end));
1642 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1643 int arg1, int arg2, unsigned char *end));
1644 static boolean at_begline_loc_p _RE_ARGS ((const unsigned char *pattern,
1645 const unsigned char *p,
1646 reg_syntax_t syntax));
1647 static boolean at_endline_loc_p _RE_ARGS ((const unsigned char *p,
1648 const unsigned char *pend,
1649 reg_syntax_t syntax));
1650 static unsigned char *skip_one_char _RE_ARGS ((unsigned char *p));
1651 static int analyse_first _RE_ARGS ((unsigned char *p, unsigned char *pend,
1652 char *fastmap, const int multibyte));
1654 /* Fetch the next character in the uncompiled pattern---translating it
1655 if necessary. Also cast from a signed character in the constant
1656 string passed to us by the user to an unsigned char that we can use
1657 as an array index (in, e.g., `translate'). */
1658 #define PATFETCH(c) \
1661 c = TRANSLATE (c); \
1664 /* Fetch the next character in the uncompiled pattern, with no
1666 #define PATFETCH_RAW(c) \
1669 if (p == pend) return REG_EEND; \
1670 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1675 /* If `translate' is non-null, return translate[D], else just D. We
1676 cast the subscript to translate because some data is declared as
1677 `char *', to avoid warnings when a string constant is passed. But
1678 when we use a character as a subscript we must make it unsigned. */
1680 # define TRANSLATE(d) \
1681 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1685 /* Macros for outputting the compiled pattern into `buffer'. */
1687 /* If the buffer isn't allocated when it comes in, use this. */
1688 #define INIT_BUF_SIZE 32
1690 /* Make sure we have at least N more bytes of space in buffer. */
1691 #define GET_BUFFER_SPACE(n) \
1692 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1695 /* Make sure we have one more byte of buffer space and then add C to it. */
1696 #define BUF_PUSH(c) \
1698 GET_BUFFER_SPACE (1); \
1699 *b++ = (unsigned char) (c); \
1703 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1704 #define BUF_PUSH_2(c1, c2) \
1706 GET_BUFFER_SPACE (2); \
1707 *b++ = (unsigned char) (c1); \
1708 *b++ = (unsigned char) (c2); \
1712 /* As with BUF_PUSH_2, except for three bytes. */
1713 #define BUF_PUSH_3(c1, c2, c3) \
1715 GET_BUFFER_SPACE (3); \
1716 *b++ = (unsigned char) (c1); \
1717 *b++ = (unsigned char) (c2); \
1718 *b++ = (unsigned char) (c3); \
1722 /* Store a jump with opcode OP at LOC to location TO. We store a
1723 relative address offset by the three bytes the jump itself occupies. */
1724 #define STORE_JUMP(op, loc, to) \
1725 store_op1 (op, loc, (to) - (loc) - 3)
1727 /* Likewise, for a two-argument jump. */
1728 #define STORE_JUMP2(op, loc, to, arg) \
1729 store_op2 (op, loc, (to) - (loc) - 3, arg)
1731 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1732 #define INSERT_JUMP(op, loc, to) \
1733 insert_op1 (op, loc, (to) - (loc) - 3, b)
1735 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1736 #define INSERT_JUMP2(op, loc, to, arg) \
1737 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1740 /* This is not an arbitrary limit: the arguments which represent offsets
1741 into the pattern are two bytes long. So if 2^16 bytes turns out to
1742 be too small, many things would have to change. */
1743 /* Any other compiler which, like MSC, has allocation limit below 2^16
1744 bytes will have to use approach similar to what was done below for
1745 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1746 reallocating to 0 bytes. Such thing is not going to work too well.
1747 You have been warned!! */
1748 #if defined _MSC_VER && !defined WIN32
1749 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1750 # define MAX_BUF_SIZE 65500L
1752 # define MAX_BUF_SIZE (1L << 16)
1755 /* Extend the buffer by twice its current size via realloc and
1756 reset the pointers that pointed into the old block to point to the
1757 correct places in the new one. If extending the buffer results in it
1758 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1759 #if __BOUNDED_POINTERS__
1760 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1761 # define MOVE_BUFFER_POINTER(P) \
1762 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1763 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1766 SET_HIGH_BOUND (b); \
1767 SET_HIGH_BOUND (begalt); \
1768 if (fixup_alt_jump) \
1769 SET_HIGH_BOUND (fixup_alt_jump); \
1771 SET_HIGH_BOUND (laststart); \
1772 if (pending_exact) \
1773 SET_HIGH_BOUND (pending_exact); \
1776 # define MOVE_BUFFER_POINTER(P) (P) += incr
1777 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1779 #define EXTEND_BUFFER() \
1781 unsigned char *old_buffer = bufp->buffer; \
1782 if (bufp->allocated == MAX_BUF_SIZE) \
1784 bufp->allocated <<= 1; \
1785 if (bufp->allocated > MAX_BUF_SIZE) \
1786 bufp->allocated = MAX_BUF_SIZE; \
1787 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1788 if (bufp->buffer == NULL) \
1789 return REG_ESPACE; \
1790 /* If the buffer moved, move all the pointers into it. */ \
1791 if (old_buffer != bufp->buffer) \
1793 int incr = bufp->buffer - old_buffer; \
1794 MOVE_BUFFER_POINTER (b); \
1795 MOVE_BUFFER_POINTER (begalt); \
1796 if (fixup_alt_jump) \
1797 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1799 MOVE_BUFFER_POINTER (laststart); \
1800 if (pending_exact) \
1801 MOVE_BUFFER_POINTER (pending_exact); \
1803 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1807 /* Since we have one byte reserved for the register number argument to
1808 {start,stop}_memory, the maximum number of groups we can report
1809 things about is what fits in that byte. */
1810 #define MAX_REGNUM 255
1812 /* But patterns can have more than `MAX_REGNUM' registers. We just
1813 ignore the excess. */
1814 typedef unsigned regnum_t;
1817 /* Macros for the compile stack. */
1819 /* Since offsets can go either forwards or backwards, this type needs to
1820 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1821 /* int may be not enough when sizeof(int) == 2. */
1822 typedef long pattern_offset_t;
1826 pattern_offset_t begalt_offset;
1827 pattern_offset_t fixup_alt_jump;
1828 pattern_offset_t laststart_offset;
1830 } compile_stack_elt_t;
1835 compile_stack_elt_t *stack;
1837 unsigned avail; /* Offset of next open position. */
1838 } compile_stack_type;
1841 #define INIT_COMPILE_STACK_SIZE 32
1843 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1844 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1846 /* The next available element. */
1847 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1850 /* Structure to manage work area for range table. */
1851 struct range_table_work_area
1853 int *table; /* actual work area. */
1854 int allocated; /* allocated size for work area in bytes. */
1855 int used; /* actually used size in words. */
1856 int bits; /* flag to record character classes */
1859 /* Make sure that WORK_AREA can hold more N multibyte characters. */
1860 #define EXTEND_RANGE_TABLE_WORK_AREA(work_area, n) \
1862 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1864 (work_area).allocated += 16 * sizeof (int); \
1865 if ((work_area).table) \
1867 = (int *) realloc ((work_area).table, (work_area).allocated); \
1870 = (int *) malloc ((work_area).allocated); \
1871 if ((work_area).table == 0) \
1872 FREE_STACK_RETURN (REG_ESPACE); \
1876 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1877 (work_area).bits |= (bit)
1879 /* Bits used to implement the multibyte-part of the various character classes
1880 such as [:alnum:] in a charset's range table. */
1881 #define BIT_WORD 0x1
1882 #define BIT_LOWER 0x2
1883 #define BIT_PUNCT 0x4
1884 #define BIT_SPACE 0x8
1885 #define BIT_UPPER 0x10
1886 #define BIT_MULTIBYTE 0x20
1888 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1889 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1891 EXTEND_RANGE_TABLE_WORK_AREA ((work_area), 2); \
1892 (work_area).table[(work_area).used++] = (range_start); \
1893 (work_area).table[(work_area).used++] = (range_end); \
1896 /* Free allocated memory for WORK_AREA. */
1897 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1899 if ((work_area).table) \
1900 free ((work_area).table); \
1903 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1904 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1905 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1906 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1909 /* Set the bit for character C in a list. */
1910 #define SET_LIST_BIT(c) \
1911 (b[((unsigned char) (c)) / BYTEWIDTH] \
1912 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1915 /* Get the next unsigned number in the uncompiled pattern. */
1916 #define GET_UNSIGNED_NUMBER(num) \
1917 do { if (p != pend) \
1920 while ('0' <= c && c <= '9') \
1924 num = num * 10 + c - '0'; \
1932 #if WIDE_CHAR_SUPPORT
1933 /* The GNU C library provides support for user-defined character classes
1934 and the functions from ISO C amendement 1. */
1935 # ifdef CHARCLASS_NAME_MAX
1936 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1938 /* This shouldn't happen but some implementation might still have this
1939 problem. Use a reasonable default value. */
1940 # define CHAR_CLASS_MAX_LENGTH 256
1942 typedef wctype_t re_wctype_t;
1943 # define re_wctype wctype
1944 # define re_iswctype iswctype
1945 # define re_wctype_to_bit(cc) 0
1947 # define CHAR_CLASS_MAX_LENGTH 9 /* Namely, `multibyte'. */
1950 /* Character classes' indices. */
1951 typedef enum { RECC_ERROR = 0,
1952 RECC_ALNUM, RECC_ALPHA, RECC_WORD,
1953 RECC_GRAPH, RECC_PRINT,
1954 RECC_LOWER, RECC_UPPER,
1955 RECC_PUNCT, RECC_CNTRL,
1956 RECC_DIGIT, RECC_XDIGIT,
1957 RECC_BLANK, RECC_SPACE,
1958 RECC_MULTIBYTE, RECC_NONASCII,
1959 RECC_ASCII, RECC_UNIBYTE
1962 /* Map a string to the char class it names (if any). */
1965 unsigned char *string;
1967 if (STREQ (string, "alnum")) return RECC_ALNUM;
1968 else if (STREQ (string, "alpha")) return RECC_ALPHA;
1969 else if (STREQ (string, "word")) return RECC_WORD;
1970 else if (STREQ (string, "ascii")) return RECC_ASCII;
1971 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
1972 else if (STREQ (string, "graph")) return RECC_GRAPH;
1973 else if (STREQ (string, "lower")) return RECC_LOWER;
1974 else if (STREQ (string, "print")) return RECC_PRINT;
1975 else if (STREQ (string, "punct")) return RECC_PUNCT;
1976 else if (STREQ (string, "space")) return RECC_SPACE;
1977 else if (STREQ (string, "upper")) return RECC_UPPER;
1978 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
1979 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
1980 else if (STREQ (string, "digit")) return RECC_DIGIT;
1981 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
1982 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
1983 else if (STREQ (string, "blank")) return RECC_BLANK;
1987 /* True iff CH is in the char class CC. */
1989 re_iswctype (ch, cc)
1995 case RECC_ALNUM: return ISALNUM (ch);
1996 case RECC_ALPHA: return ISALPHA (ch);
1997 case RECC_BLANK: return ISBLANK (ch);
1998 case RECC_CNTRL: return ISCNTRL (ch);
1999 case RECC_DIGIT: return ISDIGIT (ch);
2000 case RECC_GRAPH: return ISGRAPH (ch);
2001 case RECC_LOWER: return ISLOWER (ch);
2002 case RECC_PRINT: return ISPRINT (ch);
2003 case RECC_PUNCT: return ISPUNCT (ch);
2004 case RECC_SPACE: return ISSPACE (ch);
2005 case RECC_UPPER: return ISUPPER (ch);
2006 case RECC_XDIGIT: return ISXDIGIT (ch);
2007 case RECC_ASCII: return IS_REAL_ASCII (ch);
2008 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2009 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2010 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2011 case RECC_WORD: return ISWORD (ch);
2012 case RECC_ERROR: return false;
2016 /* Return a bit-pattern to use in the range-table bits to match multibyte
2017 chars of class CC. */
2019 re_wctype_to_bit (cc)
2024 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2025 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2026 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2027 case RECC_LOWER: return BIT_LOWER;
2028 case RECC_UPPER: return BIT_UPPER;
2029 case RECC_PUNCT: return BIT_PUNCT;
2030 case RECC_SPACE: return BIT_SPACE;
2031 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2032 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2037 /* QUIT is only used on NTemacs. */
2038 #if !defined WINDOWSNT || !defined emacs || !defined QUIT
2043 #ifndef MATCH_MAY_ALLOCATE
2045 /* If we cannot allocate large objects within re_match_2_internal,
2046 we make the fail stack and register vectors global.
2047 The fail stack, we grow to the maximum size when a regexp
2049 The register vectors, we adjust in size each time we
2050 compile a regexp, according to the number of registers it needs. */
2052 static fail_stack_type fail_stack;
2054 /* Size with which the following vectors are currently allocated.
2055 That is so we can make them bigger as needed,
2056 but never make them smaller. */
2057 static int regs_allocated_size;
2059 static re_char ** regstart, ** regend;
2060 static re_char **best_regstart, **best_regend;
2062 /* Make the register vectors big enough for NUM_REGS registers,
2063 but don't make them smaller. */
2066 regex_grow_registers (num_regs)
2069 if (num_regs > regs_allocated_size)
2071 RETALLOC_IF (regstart, num_regs, re_char *);
2072 RETALLOC_IF (regend, num_regs, re_char *);
2073 RETALLOC_IF (best_regstart, num_regs, re_char *);
2074 RETALLOC_IF (best_regend, num_regs, re_char *);
2076 regs_allocated_size = num_regs;
2080 #endif /* not MATCH_MAY_ALLOCATE */
2082 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2086 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2087 Returns one of error codes defined in `regex.h', or zero for success.
2089 Assumes the `allocated' (and perhaps `buffer') and `translate'
2090 fields are set in BUFP on entry.
2092 If it succeeds, results are put in BUFP (if it returns an error, the
2093 contents of BUFP are undefined):
2094 `buffer' is the compiled pattern;
2095 `syntax' is set to SYNTAX;
2096 `used' is set to the length of the compiled pattern;
2097 `fastmap_accurate' is zero;
2098 `re_nsub' is the number of subexpressions in PATTERN;
2099 `not_bol' and `not_eol' are zero;
2101 The `fastmap' field is neither examined nor set. */
2103 /* Insert the `jump' from the end of last alternative to "here".
2104 The space for the jump has already been allocated. */
2105 #define FIXUP_ALT_JUMP() \
2107 if (fixup_alt_jump) \
2108 STORE_JUMP (jump, fixup_alt_jump, b); \
2112 /* Return, freeing storage we allocated. */
2113 #define FREE_STACK_RETURN(value) \
2115 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2116 free (compile_stack.stack); \
2120 static reg_errcode_t
2121 regex_compile (pattern, size, syntax, bufp)
2124 reg_syntax_t syntax;
2125 struct re_pattern_buffer *bufp;
2127 /* We fetch characters from PATTERN here. Even though PATTERN is
2128 `char *' (i.e., signed), we declare these variables as unsigned, so
2129 they can be reliably used as array indices. */
2130 register unsigned int c, c1;
2132 /* A random temporary spot in PATTERN. */
2135 /* Points to the end of the buffer, where we should append. */
2136 register unsigned char *b;
2138 /* Keeps track of unclosed groups. */
2139 compile_stack_type compile_stack;
2141 /* Points to the current (ending) position in the pattern. */
2143 /* `const' makes AIX compiler fail. */
2144 unsigned char *p = pattern;
2146 re_char *p = pattern;
2148 re_char *pend = pattern + size;
2150 /* How to translate the characters in the pattern. */
2151 RE_TRANSLATE_TYPE translate = bufp->translate;
2153 /* Address of the count-byte of the most recently inserted `exactn'
2154 command. This makes it possible to tell if a new exact-match
2155 character can be added to that command or if the character requires
2156 a new `exactn' command. */
2157 unsigned char *pending_exact = 0;
2159 /* Address of start of the most recently finished expression.
2160 This tells, e.g., postfix * where to find the start of its
2161 operand. Reset at the beginning of groups and alternatives. */
2162 unsigned char *laststart = 0;
2164 /* Address of beginning of regexp, or inside of last group. */
2165 unsigned char *begalt;
2167 /* Place in the uncompiled pattern (i.e., the {) to
2168 which to go back if the interval is invalid. */
2169 re_char *beg_interval;
2171 /* Address of the place where a forward jump should go to the end of
2172 the containing expression. Each alternative of an `or' -- except the
2173 last -- ends with a forward jump of this sort. */
2174 unsigned char *fixup_alt_jump = 0;
2176 /* Counts open-groups as they are encountered. Remembered for the
2177 matching close-group on the compile stack, so the same register
2178 number is put in the stop_memory as the start_memory. */
2179 regnum_t regnum = 0;
2181 /* Work area for range table of charset. */
2182 struct range_table_work_area range_table_work;
2184 /* If the object matched can contain multibyte characters. */
2185 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2189 DEBUG_PRINT1 ("\nCompiling pattern: ");
2192 unsigned debug_count;
2194 for (debug_count = 0; debug_count < size; debug_count++)
2195 putchar (pattern[debug_count]);
2200 /* Initialize the compile stack. */
2201 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2202 if (compile_stack.stack == NULL)
2205 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2206 compile_stack.avail = 0;
2208 range_table_work.table = 0;
2209 range_table_work.allocated = 0;
2211 /* Initialize the pattern buffer. */
2212 bufp->syntax = syntax;
2213 bufp->fastmap_accurate = 0;
2214 bufp->not_bol = bufp->not_eol = 0;
2216 /* Set `used' to zero, so that if we return an error, the pattern
2217 printer (for debugging) will think there's no pattern. We reset it
2221 /* Always count groups, whether or not bufp->no_sub is set. */
2224 #if !defined emacs && !defined SYNTAX_TABLE
2225 /* Initialize the syntax table. */
2226 init_syntax_once ();
2229 if (bufp->allocated == 0)
2232 { /* If zero allocated, but buffer is non-null, try to realloc
2233 enough space. This loses if buffer's address is bogus, but
2234 that is the user's responsibility. */
2235 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2238 { /* Caller did not allocate a buffer. Do it for them. */
2239 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2241 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2243 bufp->allocated = INIT_BUF_SIZE;
2246 begalt = b = bufp->buffer;
2248 /* Loop through the uncompiled pattern until we're at the end. */
2257 if ( /* If at start of pattern, it's an operator. */
2259 /* If context independent, it's an operator. */
2260 || syntax & RE_CONTEXT_INDEP_ANCHORS
2261 /* Otherwise, depends on what's come before. */
2262 || at_begline_loc_p (pattern, p, syntax))
2263 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2272 if ( /* If at end of pattern, it's an operator. */
2274 /* If context independent, it's an operator. */
2275 || syntax & RE_CONTEXT_INDEP_ANCHORS
2276 /* Otherwise, depends on what's next. */
2277 || at_endline_loc_p (p, pend, syntax))
2278 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2287 if ((syntax & RE_BK_PLUS_QM)
2288 || (syntax & RE_LIMITED_OPS))
2292 /* If there is no previous pattern... */
2295 if (syntax & RE_CONTEXT_INVALID_OPS)
2296 FREE_STACK_RETURN (REG_BADRPT);
2297 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2302 /* 1 means zero (many) matches is allowed. */
2303 boolean zero_times_ok = 0, many_times_ok = 0;
2306 /* If there is a sequence of repetition chars, collapse it
2307 down to just one (the right one). We can't combine
2308 interval operators with these because of, e.g., `a{2}*',
2309 which should only match an even number of `a's. */
2313 if ((syntax & RE_FRUGAL)
2314 && c == '?' && (zero_times_ok || many_times_ok))
2318 zero_times_ok |= c != '+';
2319 many_times_ok |= c != '?';
2325 || (!(syntax & RE_BK_PLUS_QM)
2326 && (*p == '+' || *p == '?')))
2328 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2331 FREE_STACK_RETURN (REG_EESCAPE);
2332 if (p[1] == '+' || p[1] == '?')
2333 PATFETCH (c); /* Gobble up the backslash. */
2339 /* If we get here, we found another repeat character. */
2343 /* Star, etc. applied to an empty pattern is equivalent
2344 to an empty pattern. */
2345 if (!laststart || laststart == b)
2348 /* Now we know whether or not zero matches is allowed
2349 and also whether or not two or more matches is allowed. */
2354 boolean simple = skip_one_char (laststart) == b;
2355 unsigned int startoffset = 0;
2357 (simple || !analyse_first (laststart, b, NULL, 0)) ?
2358 on_failure_jump : on_failure_jump_loop;
2359 assert (skip_one_char (laststart) <= b);
2361 if (!zero_times_ok && simple)
2362 { /* Since simple * loops can be made faster by using
2363 on_failure_keep_string_jump, we turn simple P+
2364 into PP* if P is simple. */
2365 unsigned char *p1, *p2;
2366 startoffset = b - laststart;
2367 GET_BUFFER_SPACE (startoffset);
2368 p1 = b; p2 = laststart;
2374 GET_BUFFER_SPACE (6);
2377 STORE_JUMP (ofj, b, b + 6);
2379 /* Simple * loops can use on_failure_keep_string_jump
2380 depending on what follows. But since we don't know
2381 that yet, we leave the decision up to
2382 on_failure_jump_smart. */
2383 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2384 laststart + startoffset, b + 6);
2386 STORE_JUMP (jump, b, laststart + startoffset);
2391 /* A simple ? pattern. */
2392 assert (zero_times_ok);
2393 GET_BUFFER_SPACE (3);
2394 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2398 else /* not greedy */
2399 { /* I wish the greedy and non-greedy cases could be merged. */
2401 GET_BUFFER_SPACE (7); /* We might use less. */
2404 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2406 /* The non-greedy multiple match looks like a repeat..until:
2407 we only need a conditional jump at the end of the loop */
2408 if (emptyp) BUF_PUSH (no_op);
2409 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2410 : on_failure_jump, b, laststart);
2414 /* The repeat...until naturally matches one or more.
2415 To also match zero times, we need to first jump to
2416 the end of the loop (its conditional jump). */
2417 INSERT_JUMP (jump, laststart, b);
2423 /* non-greedy a?? */
2424 INSERT_JUMP (jump, laststart, b + 3);
2426 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2443 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2445 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2447 /* Ensure that we have enough space to push a charset: the
2448 opcode, the length count, and the bitset; 34 bytes in all. */
2449 GET_BUFFER_SPACE (34);
2453 /* We test `*p == '^' twice, instead of using an if
2454 statement, so we only need one BUF_PUSH. */
2455 BUF_PUSH (*p == '^' ? charset_not : charset);
2459 /* Remember the first position in the bracket expression. */
2462 /* Push the number of bytes in the bitmap. */
2463 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2465 /* Clear the whole map. */
2466 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2468 /* charset_not matches newline according to a syntax bit. */
2469 if ((re_opcode_t) b[-2] == charset_not
2470 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2471 SET_LIST_BIT ('\n');
2473 /* Read in characters and ranges, setting map bits. */
2476 boolean escaped_char = false;
2477 const unsigned char *p2 = p;
2479 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2483 /* \ might escape characters inside [...] and [^...]. */
2484 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2486 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2489 escaped_char = true;
2493 /* Could be the end of the bracket expression. If it's
2494 not (i.e., when the bracket expression is `[]' so
2495 far), the ']' character bit gets set way below. */
2496 if (c == ']' && p2 != p1)
2500 /* What should we do for the character which is
2501 greater than 0x7F, but not BASE_LEADING_CODE_P?
2504 /* See if we're at the beginning of a possible character
2507 if (!escaped_char &&
2508 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2510 /* Leave room for the null. */
2511 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2512 const unsigned char *class_beg;
2518 /* If pattern is `[[:'. */
2519 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2524 if ((c == ':' && *p == ']') || p == pend)
2526 if (c1 < CHAR_CLASS_MAX_LENGTH)
2529 /* This is in any case an invalid class name. */
2534 /* If isn't a word bracketed by `[:' and `:]':
2535 undo the ending character, the letters, and
2536 leave the leading `:' and `[' (but set bits for
2538 if (c == ':' && *p == ']')
2543 cc = re_wctype (str);
2546 FREE_STACK_RETURN (REG_ECTYPE);
2548 /* Throw away the ] at the end of the character
2552 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2554 /* Most character classes in a multibyte match
2555 just set a flag. Exceptions are is_blank,
2556 is_digit, is_cntrl, and is_xdigit, since
2557 they can only match ASCII characters. We
2558 don't need to handle them for multibyte.
2559 They are distinguished by a negative wctype. */
2562 SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work,
2563 re_wctype_to_bit (cc));
2565 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
2567 int translated = TRANSLATE (ch);
2568 if (re_iswctype (btowc (ch), cc))
2569 SET_LIST_BIT (translated);
2572 /* Repeat the loop. */
2577 /* Go back to right after the "[:". */
2581 /* Because the `:' may starts the range, we
2582 can't simply set bit and repeat the loop.
2583 Instead, just set it to C and handle below. */
2588 if (p < pend && p[0] == '-' && p[1] != ']')
2591 /* Discard the `-'. */
2594 /* Fetch the character which ends the range. */
2597 if (SINGLE_BYTE_CHAR_P (c))
2599 if (! SINGLE_BYTE_CHAR_P (c1))
2601 /* Handle a range starting with a
2602 character of less than 256, and ending
2603 with a character of not less than 256.
2604 Split that into two ranges, the low one
2605 ending at 0377, and the high one
2606 starting at the smallest character in
2607 the charset of C1 and ending at C1. */
2608 int charset = CHAR_CHARSET (c1);
2609 int c2 = MAKE_CHAR (charset, 0, 0);
2611 SET_RANGE_TABLE_WORK_AREA (range_table_work,
2616 else if (!SAME_CHARSET_P (c, c1))
2617 FREE_STACK_RETURN (REG_ERANGE);
2620 /* Range from C to C. */
2623 /* Set the range ... */
2624 if (SINGLE_BYTE_CHAR_P (c))
2625 /* ... into bitmap. */
2628 int range_start = c, range_end = c1;
2630 /* If the start is after the end, the range is empty. */
2631 if (range_start > range_end)
2633 if (syntax & RE_NO_EMPTY_RANGES)
2634 FREE_STACK_RETURN (REG_ERANGE);
2635 /* Else, repeat the loop. */
2639 for (this_char = range_start; this_char <= range_end;
2641 SET_LIST_BIT (TRANSLATE (this_char));
2645 /* ... into range table. */
2646 SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1);
2649 /* Discard any (non)matching list bytes that are all 0 at the
2650 end of the map. Decrease the map-length byte too. */
2651 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2655 /* Build real range table from work area. */
2656 if (RANGE_TABLE_WORK_USED (range_table_work)
2657 || RANGE_TABLE_WORK_BITS (range_table_work))
2660 int used = RANGE_TABLE_WORK_USED (range_table_work);
2662 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2663 bytes for flags, two for COUNT, and three bytes for
2665 GET_BUFFER_SPACE (4 + used * 3);
2667 /* Indicate the existence of range table. */
2668 laststart[1] |= 0x80;
2670 /* Store the character class flag bits into the range table.
2671 If not in emacs, these flag bits are always 0. */
2672 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
2673 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
2675 STORE_NUMBER_AND_INCR (b, used / 2);
2676 for (i = 0; i < used; i++)
2677 STORE_CHARACTER_AND_INCR
2678 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
2685 if (syntax & RE_NO_BK_PARENS)
2692 if (syntax & RE_NO_BK_PARENS)
2699 if (syntax & RE_NEWLINE_ALT)
2706 if (syntax & RE_NO_BK_VBAR)
2713 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2714 goto handle_interval;
2720 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2722 /* Do not translate the character after the \, so that we can
2723 distinguish, e.g., \B from \b, even if we normally would
2724 translate, e.g., B to b. */
2730 if (syntax & RE_NO_BK_PARENS)
2731 goto normal_backslash;
2738 /* Look for a special (?...) construct */
2739 if ((syntax & RE_SHY_GROUPS) && *p == '?')
2741 PATFETCH (c); /* Gobble up the '?'. */
2745 case ':': shy = 1; break;
2747 /* Only (?:...) is supported right now. */
2748 FREE_STACK_RETURN (REG_BADPAT);
2759 if (COMPILE_STACK_FULL)
2761 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2762 compile_stack_elt_t);
2763 if (compile_stack.stack == NULL) return REG_ESPACE;
2765 compile_stack.size <<= 1;
2768 /* These are the values to restore when we hit end of this
2769 group. They are all relative offsets, so that if the
2770 whole pattern moves because of realloc, they will still
2772 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2773 COMPILE_STACK_TOP.fixup_alt_jump
2774 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2775 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2776 COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum;
2779 start_memory for groups beyond the last one we can
2780 represent in the compiled pattern. */
2781 if (regnum <= MAX_REGNUM && !shy)
2782 BUF_PUSH_2 (start_memory, regnum);
2784 compile_stack.avail++;
2789 /* If we've reached MAX_REGNUM groups, then this open
2790 won't actually generate any code, so we'll have to
2791 clear pending_exact explicitly. */
2797 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2799 if (COMPILE_STACK_EMPTY)
2801 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2802 goto normal_backslash;
2804 FREE_STACK_RETURN (REG_ERPAREN);
2810 /* See similar code for backslashed left paren above. */
2811 if (COMPILE_STACK_EMPTY)
2813 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2816 FREE_STACK_RETURN (REG_ERPAREN);
2819 /* Since we just checked for an empty stack above, this
2820 ``can't happen''. */
2821 assert (compile_stack.avail != 0);
2823 /* We don't just want to restore into `regnum', because
2824 later groups should continue to be numbered higher,
2825 as in `(ab)c(de)' -- the second group is #2. */
2826 regnum_t this_group_regnum;
2828 compile_stack.avail--;
2829 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2831 = COMPILE_STACK_TOP.fixup_alt_jump
2832 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2834 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2835 this_group_regnum = COMPILE_STACK_TOP.regnum;
2836 /* If we've reached MAX_REGNUM groups, then this open
2837 won't actually generate any code, so we'll have to
2838 clear pending_exact explicitly. */
2841 /* We're at the end of the group, so now we know how many
2842 groups were inside this one. */
2843 if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0)
2844 BUF_PUSH_2 (stop_memory, this_group_regnum);
2849 case '|': /* `\|'. */
2850 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2851 goto normal_backslash;
2853 if (syntax & RE_LIMITED_OPS)
2856 /* Insert before the previous alternative a jump which
2857 jumps to this alternative if the former fails. */
2858 GET_BUFFER_SPACE (3);
2859 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2863 /* The alternative before this one has a jump after it
2864 which gets executed if it gets matched. Adjust that
2865 jump so it will jump to this alternative's analogous
2866 jump (put in below, which in turn will jump to the next
2867 (if any) alternative's such jump, etc.). The last such
2868 jump jumps to the correct final destination. A picture:
2874 If we are at `b', then fixup_alt_jump right now points to a
2875 three-byte space after `a'. We'll put in the jump, set
2876 fixup_alt_jump to right after `b', and leave behind three
2877 bytes which we'll fill in when we get to after `c'. */
2881 /* Mark and leave space for a jump after this alternative,
2882 to be filled in later either by next alternative or
2883 when know we're at the end of a series of alternatives. */
2885 GET_BUFFER_SPACE (3);
2894 /* If \{ is a literal. */
2895 if (!(syntax & RE_INTERVALS)
2896 /* If we're at `\{' and it's not the open-interval
2898 || (syntax & RE_NO_BK_BRACES))
2899 goto normal_backslash;
2903 /* If got here, then the syntax allows intervals. */
2905 /* At least (most) this many matches must be made. */
2906 int lower_bound = 0, upper_bound = -1;
2911 FREE_STACK_RETURN (REG_EBRACE);
2913 GET_UNSIGNED_NUMBER (lower_bound);
2916 GET_UNSIGNED_NUMBER (upper_bound);
2918 /* Interval such as `{1}' => match exactly once. */
2919 upper_bound = lower_bound;
2921 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2922 || (upper_bound >= 0 && lower_bound > upper_bound))
2923 FREE_STACK_RETURN (REG_BADBR);
2925 if (!(syntax & RE_NO_BK_BRACES))
2928 FREE_STACK_RETURN (REG_BADBR);
2934 FREE_STACK_RETURN (REG_BADBR);
2936 /* We just parsed a valid interval. */
2938 /* If it's invalid to have no preceding re. */
2941 if (syntax & RE_CONTEXT_INVALID_OPS)
2942 FREE_STACK_RETURN (REG_BADRPT);
2943 else if (syntax & RE_CONTEXT_INDEP_OPS)
2946 goto unfetch_interval;
2949 if (upper_bound == 0)
2950 /* If the upper bound is zero, just drop the sub pattern
2953 else if (lower_bound == 1 && upper_bound == 1)
2954 /* Just match it once: nothing to do here. */
2957 /* Otherwise, we have a nontrivial interval. When
2958 we're all done, the pattern will look like:
2959 set_number_at <jump count> <upper bound>
2960 set_number_at <succeed_n count> <lower bound>
2961 succeed_n <after jump addr> <succeed_n count>
2963 jump_n <succeed_n addr> <jump count>
2964 (The upper bound and `jump_n' are omitted if
2965 `upper_bound' is 1, though.) */
2967 { /* If the upper bound is > 1, we need to insert
2968 more at the end of the loop. */
2969 unsigned int nbytes = (upper_bound < 0 ? 3
2970 : upper_bound > 1 ? 5 : 0);
2971 unsigned int startoffset = 0;
2973 GET_BUFFER_SPACE (20); /* We might use less. */
2975 if (lower_bound == 0)
2977 /* A succeed_n that starts with 0 is really a
2978 a simple on_failure_jump_loop. */
2979 INSERT_JUMP (on_failure_jump_loop, laststart,
2985 /* Initialize lower bound of the `succeed_n', even
2986 though it will be set during matching by its
2987 attendant `set_number_at' (inserted next),
2988 because `re_compile_fastmap' needs to know.
2989 Jump to the `jump_n' we might insert below. */
2990 INSERT_JUMP2 (succeed_n, laststart,
2995 /* Code to initialize the lower bound. Insert
2996 before the `succeed_n'. The `5' is the last two
2997 bytes of this `set_number_at', plus 3 bytes of
2998 the following `succeed_n'. */
2999 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3004 if (upper_bound < 0)
3006 /* A negative upper bound stands for infinity,
3007 in which case it degenerates to a plain jump. */
3008 STORE_JUMP (jump, b, laststart + startoffset);
3011 else if (upper_bound > 1)
3012 { /* More than one repetition is allowed, so
3013 append a backward jump to the `succeed_n'
3014 that starts this interval.
3016 When we've reached this during matching,
3017 we'll have matched the interval once, so
3018 jump back only `upper_bound - 1' times. */
3019 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3023 /* The location we want to set is the second
3024 parameter of the `jump_n'; that is `b-2' as
3025 an absolute address. `laststart' will be
3026 the `set_number_at' we're about to insert;
3027 `laststart+3' the number to set, the source
3028 for the relative address. But we are
3029 inserting into the middle of the pattern --
3030 so everything is getting moved up by 5.
3031 Conclusion: (b - 2) - (laststart + 3) + 5,
3032 i.e., b - laststart.
3034 We insert this at the beginning of the loop
3035 so that if we fail during matching, we'll
3036 reinitialize the bounds. */
3037 insert_op2 (set_number_at, laststart, b - laststart,
3038 upper_bound - 1, b);
3043 beg_interval = NULL;
3048 /* If an invalid interval, match the characters as literals. */
3049 assert (beg_interval);
3051 beg_interval = NULL;
3053 /* normal_char and normal_backslash need `c'. */
3056 if (!(syntax & RE_NO_BK_BRACES))
3058 assert (p > pattern && p[-1] == '\\');
3059 goto normal_backslash;
3065 /* There is no way to specify the before_dot and after_dot
3066 operators. rms says this is ok. --karl */
3074 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3080 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3086 BUF_PUSH_2 (categoryspec, c);
3092 BUF_PUSH_2 (notcategoryspec, c);
3098 if (syntax & RE_NO_GNU_OPS)
3101 BUF_PUSH_2 (syntaxspec, Sword);
3106 if (syntax & RE_NO_GNU_OPS)
3109 BUF_PUSH_2 (notsyntaxspec, Sword);
3114 if (syntax & RE_NO_GNU_OPS)
3120 if (syntax & RE_NO_GNU_OPS)
3126 if (syntax & RE_NO_GNU_OPS)
3128 BUF_PUSH (wordbound);
3132 if (syntax & RE_NO_GNU_OPS)
3134 BUF_PUSH (notwordbound);
3138 if (syntax & RE_NO_GNU_OPS)
3144 if (syntax & RE_NO_GNU_OPS)
3149 case '1': case '2': case '3': case '4': case '5':
3150 case '6': case '7': case '8': case '9':
3151 if (syntax & RE_NO_BK_REFS)
3157 FREE_STACK_RETURN (REG_ESUBREG);
3159 /* Can't back reference to a subexpression if inside of it. */
3160 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
3164 BUF_PUSH_2 (duplicate, c1);
3170 if (syntax & RE_BK_PLUS_QM)
3173 goto normal_backslash;
3177 /* You might think it would be useful for \ to mean
3178 not to translate; but if we don't translate it
3179 it will never match anything. */
3187 /* Expects the character in `c'. */
3189 /* If no exactn currently being built. */
3192 /* If last exactn not at current position. */
3193 || pending_exact + *pending_exact + 1 != b
3195 /* We have only one byte following the exactn for the count. */
3196 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3198 /* If followed by a repetition operator. */
3199 || (p != pend && (*p == '*' || *p == '^'))
3200 || ((syntax & RE_BK_PLUS_QM)
3201 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3202 : p != pend && (*p == '+' || *p == '?'))
3203 || ((syntax & RE_INTERVALS)
3204 && ((syntax & RE_NO_BK_BRACES)
3205 ? p != pend && *p == '{'
3206 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3208 /* Start building a new exactn. */
3212 BUF_PUSH_2 (exactn, 0);
3213 pending_exact = b - 1;
3216 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3221 len = CHAR_STRING (c, b);
3225 (*pending_exact) += len;
3230 } /* while p != pend */
3233 /* Through the pattern now. */
3237 if (!COMPILE_STACK_EMPTY)
3238 FREE_STACK_RETURN (REG_EPAREN);
3240 /* If we don't want backtracking, force success
3241 the first time we reach the end of the compiled pattern. */
3242 if (syntax & RE_NO_POSIX_BACKTRACKING)
3245 free (compile_stack.stack);
3247 /* We have succeeded; set the length of the buffer. */
3248 bufp->used = b - bufp->buffer;
3253 re_compile_fastmap (bufp);
3254 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3255 print_compiled_pattern (bufp);
3260 #ifndef MATCH_MAY_ALLOCATE
3261 /* Initialize the failure stack to the largest possible stack. This
3262 isn't necessary unless we're trying to avoid calling alloca in
3263 the search and match routines. */
3265 int num_regs = bufp->re_nsub + 1;
3267 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3269 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3271 if (! fail_stack.stack)
3273 = (fail_stack_elt_t *) malloc (fail_stack.size
3274 * sizeof (fail_stack_elt_t));
3277 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3279 * sizeof (fail_stack_elt_t)));
3282 regex_grow_registers (num_regs);
3284 #endif /* not MATCH_MAY_ALLOCATE */
3287 } /* regex_compile */
3289 /* Subroutines for `regex_compile'. */
3291 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3294 store_op1 (op, loc, arg)
3299 *loc = (unsigned char) op;
3300 STORE_NUMBER (loc + 1, arg);
3304 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3307 store_op2 (op, loc, arg1, arg2)
3312 *loc = (unsigned char) op;
3313 STORE_NUMBER (loc + 1, arg1);
3314 STORE_NUMBER (loc + 3, arg2);
3318 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3319 for OP followed by two-byte integer parameter ARG. */
3322 insert_op1 (op, loc, arg, end)
3328 register unsigned char *pfrom = end;
3329 register unsigned char *pto = end + 3;
3331 while (pfrom != loc)
3334 store_op1 (op, loc, arg);
3338 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3341 insert_op2 (op, loc, arg1, arg2, end)
3347 register unsigned char *pfrom = end;
3348 register unsigned char *pto = end + 5;
3350 while (pfrom != loc)
3353 store_op2 (op, loc, arg1, arg2);
3357 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3358 after an alternative or a begin-subexpression. We assume there is at
3359 least one character before the ^. */
3362 at_begline_loc_p (pattern, p, syntax)
3363 const unsigned char *pattern, *p;
3364 reg_syntax_t syntax;
3366 const unsigned char *prev = p - 2;
3367 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3370 /* After a subexpression? */
3371 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3372 /* After an alternative? */
3373 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3374 /* After a shy subexpression? */
3375 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3376 && prev[-1] == '?' && prev[-2] == '('
3377 && (syntax & RE_NO_BK_PARENS
3378 || (prev - 3 >= pattern && prev[-3] == '\\')));
3382 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3383 at least one character after the $, i.e., `P < PEND'. */
3386 at_endline_loc_p (p, pend, syntax)
3387 const unsigned char *p, *pend;
3388 reg_syntax_t syntax;
3390 const unsigned char *next = p;
3391 boolean next_backslash = *next == '\\';
3392 const unsigned char *next_next = p + 1 < pend ? p + 1 : 0;
3395 /* Before a subexpression? */
3396 (syntax & RE_NO_BK_PARENS ? *next == ')'
3397 : next_backslash && next_next && *next_next == ')')
3398 /* Before an alternative? */
3399 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3400 : next_backslash && next_next && *next_next == '|');
3404 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3405 false if it's not. */
3408 group_in_compile_stack (compile_stack, regnum)
3409 compile_stack_type compile_stack;
3414 for (this_element = compile_stack.avail - 1;
3417 if (compile_stack.stack[this_element].regnum == regnum)
3424 If fastmap is non-NULL, go through the pattern and fill fastmap
3425 with all the possible leading chars. If fastmap is NULL, don't
3426 bother filling it up (obviously) and only return whether the
3427 pattern could potentially match the empty string.
3429 Return 1 if p..pend might match the empty string.
3430 Return 0 if p..pend matches at least one char.
3431 Return -1 if p..pend matches at least one char, but fastmap was not
3433 Return -2 if an error occurred. */
3436 analyse_first (p, pend, fastmap, multibyte)
3437 unsigned char *p, *pend;
3439 const int multibyte;
3443 #ifdef MATCH_MAY_ALLOCATE
3444 fail_stack_type fail_stack;
3446 #ifndef REGEX_MALLOC
3450 #if defined REL_ALLOC && defined REGEX_MALLOC
3451 /* This holds the pointer to the failure stack, when
3452 it is allocated relocatably. */
3453 fail_stack_elt_t *failure_stack_ptr;
3456 /* Assume that each path through the pattern can be null until
3457 proven otherwise. We set this false at the bottom of switch
3458 statement, to which we get only if a particular path doesn't
3459 match the empty string. */
3460 boolean path_can_be_null = true;
3462 /* If all elements for base leading-codes in fastmap is set, this
3463 flag is set true. */
3464 boolean match_any_multibyte_characters = false;
3470 /* The loop below works as follows:
3471 - It has a working-list kept in the PATTERN_STACK and which basically
3472 starts by only containing a pointer to the first operation.
3473 - If the opcode we're looking at is a match against some set of
3474 chars, then we add those chars to the fastmap and go on to the
3475 next work element from the worklist (done via `break').
3476 - If the opcode is a control operator on the other hand, we either
3477 ignore it (if it's meaningless at this point, such as `start_memory')
3478 or execute it (if it's a jump). If the jump has several destinations
3479 (i.e. `on_failure_jump'), then we push the other destination onto the
3481 We guarantee termination by ignoring backward jumps (more or less),
3482 so that `p' is monotonically increasing. More to the point, we
3483 never set `p' (or push) anything `<= p1'. */
3487 /* `p1' is used as a marker of how far back a `on_failure_jump'
3488 can go without being ignored. It is normally equal to `p'
3489 (which prevents any backward `on_failure_jump') except right
3490 after a plain `jump', to allow patterns such as:
3493 10: on_failure_jump 3
3494 as used for the *? operator. */
3495 unsigned char *p1 = p;
3499 if (path_can_be_null)
3500 return (RESET_FAIL_STACK (), 1);
3502 /* We have reached the (effective) end of pattern. */
3503 if (PATTERN_STACK_EMPTY ())
3504 return (RESET_FAIL_STACK (), 0);
3506 p = (unsigned char*) POP_PATTERN_OP ();
3507 path_can_be_null = true;
3511 /* We should never be about to go beyond the end of the pattern. */
3514 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3521 /* If the first character has to match a backreference, that means
3522 that the group was empty (since it already matched). Since this
3523 is the only case that interests us here, we can assume that the
3524 backreference must match the empty string. */
3529 /* Following are the cases which match a character. These end
3535 int c = RE_STRING_CHAR (p + 1, pend - p);
3537 if (SINGLE_BYTE_CHAR_P (c))
3546 /* We could put all the chars except for \n (and maybe \0)
3547 but we don't bother since it is generally not worth it. */
3548 if (!fastmap) break;
3549 return (RESET_FAIL_STACK (), -1);
3553 /* Chars beyond end of bitmap are possible matches.
3554 All the single-byte codes can occur in multibyte buffers.
3555 So any that are not listed in the charset
3556 are possible matches, even in multibyte buffers. */
3557 if (!fastmap) break;
3558 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3559 j < (1 << BYTEWIDTH); j++)
3563 if (!fastmap) break;
3564 not = (re_opcode_t) *(p - 1) == charset_not;
3565 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3567 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3570 if ((not && multibyte)
3571 /* Any character set can possibly contain a character
3572 which doesn't match the specified set of characters. */
3573 || (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3574 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
3575 /* If we can match a character class, we can match
3576 any character set. */
3578 set_fastmap_for_multibyte_characters:
3579 if (match_any_multibyte_characters == false)
3581 for (j = 0x80; j < 0xA0; j++) /* XXX */
3582 if (BASE_LEADING_CODE_P (j))
3584 match_any_multibyte_characters = true;
3588 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3589 && match_any_multibyte_characters == false)
3591 /* Set fastmap[I] 1 where I is a base leading code of each
3592 multibyte character in the range table. */
3595 /* Make P points the range table. `+ 2' is to skip flag
3596 bits for a character class. */
3597 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
3599 /* Extract the number of ranges in range table into COUNT. */
3600 EXTRACT_NUMBER_AND_INCR (count, p);
3601 for (; count > 0; count--, p += 2 * 3) /* XXX */
3603 /* Extract the start of each range. */
3604 EXTRACT_CHARACTER (c, p);
3605 j = CHAR_CHARSET (c);
3606 fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1;
3613 if (!fastmap) break;
3615 not = (re_opcode_t)p[-1] == notsyntaxspec;
3617 for (j = 0; j < (1 << BYTEWIDTH); j++)
3618 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
3622 /* This match depends on text properties. These end with
3623 aborting optimizations. */
3624 return (RESET_FAIL_STACK (), -1);
3627 case notcategoryspec:
3628 if (!fastmap) break;
3629 not = (re_opcode_t)p[-1] == notcategoryspec;
3631 for (j = 0; j < (1 << BYTEWIDTH); j++)
3632 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
3636 /* Any character set can possibly contain a character
3637 whose category is K (or not). */
3638 goto set_fastmap_for_multibyte_characters;
3641 /* All cases after this match the empty string. These end with
3661 EXTRACT_NUMBER_AND_INCR (j, p);
3663 /* Backward jumps can only go back to code that we've already
3664 visited. `re_compile' should make sure this is true. */
3667 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
3669 case on_failure_jump:
3670 case on_failure_keep_string_jump:
3671 case on_failure_jump_loop:
3672 case on_failure_jump_nastyloop:
3673 case on_failure_jump_smart:
3679 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
3680 to jump back to "just after here". */
3683 case on_failure_jump:
3684 case on_failure_keep_string_jump:
3685 case on_failure_jump_nastyloop:
3686 case on_failure_jump_loop:
3687 case on_failure_jump_smart:
3688 EXTRACT_NUMBER_AND_INCR (j, p);
3690 ; /* Backward jump to be ignored. */
3691 else if (!PUSH_PATTERN_OP (p + j, fail_stack))
3692 return (RESET_FAIL_STACK (), -2);
3697 /* This code simply does not properly handle forward jump_n. */
3698 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
3700 /* jump_n can either jump or fall through. The (backward) jump
3701 case has already been handled, so we only need to look at the
3702 fallthrough case. */
3706 /* If N == 0, it should be an on_failure_jump_loop instead. */
3707 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
3709 /* We only care about one iteration of the loop, so we don't
3710 need to consider the case where this behaves like an
3727 abort (); /* We have listed all the cases. */
3730 /* Getting here means we have found the possible starting
3731 characters for one path of the pattern -- and that the empty
3732 string does not match. We need not follow this path further.
3733 Instead, look at the next alternative (remembered on the
3734 stack), or quit if no more. The test at the top of the loop
3735 does these things. */
3736 path_can_be_null = false;
3740 return (RESET_FAIL_STACK (), 0);
3741 } /* analyse_first */
3743 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3744 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3745 characters can start a string that matches the pattern. This fastmap
3746 is used by re_search to skip quickly over impossible starting points.
3748 Character codes above (1 << BYTEWIDTH) are not represented in the
3749 fastmap, but the leading codes are represented. Thus, the fastmap
3750 indicates which character sets could start a match.
3752 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3753 area as BUFP->fastmap.
3755 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3758 Returns 0 if we succeed, -2 if an internal error. */
3761 re_compile_fastmap (bufp)
3762 struct re_pattern_buffer *bufp;
3764 char *fastmap = bufp->fastmap;
3767 assert (fastmap && bufp->buffer);
3769 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3770 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3772 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
3773 fastmap, RE_MULTIBYTE_P (bufp));
3774 bufp->can_be_null = (analysis != 0);
3778 } /* re_compile_fastmap */
3780 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3781 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3782 this memory for recording register information. STARTS and ENDS
3783 must be allocated using the malloc library routine, and must each
3784 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3786 If NUM_REGS == 0, then subsequent matches should allocate their own
3789 Unless this function is called, the first search or match using
3790 PATTERN_BUFFER will allocate its own register data, without
3791 freeing the old data. */
3794 re_set_registers (bufp, regs, num_regs, starts, ends)
3795 struct re_pattern_buffer *bufp;
3796 struct re_registers *regs;
3798 regoff_t *starts, *ends;
3802 bufp->regs_allocated = REGS_REALLOCATE;
3803 regs->num_regs = num_regs;
3804 regs->start = starts;
3809 bufp->regs_allocated = REGS_UNALLOCATED;
3811 regs->start = regs->end = (regoff_t *) 0;
3814 WEAK_ALIAS (__re_set_registers, re_set_registers)
3816 /* Searching routines. */
3818 /* Like re_search_2, below, but only one string is specified, and
3819 doesn't let you say where to stop matching. */
3822 re_search (bufp, string, size, startpos, range, regs)
3823 struct re_pattern_buffer *bufp;
3825 int size, startpos, range;
3826 struct re_registers *regs;
3828 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3831 WEAK_ALIAS (__re_search, re_search)
3833 /* End address of virtual concatenation of string. */
3834 #define STOP_ADDR_VSTRING(P) \
3835 (((P) >= size1 ? string2 + size2 : string1 + size1))
3837 /* Address of POS in the concatenation of virtual string. */
3838 #define POS_ADDR_VSTRING(POS) \
3839 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
3841 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3842 virtual concatenation of STRING1 and STRING2, starting first at index
3843 STARTPOS, then at STARTPOS + 1, and so on.
3845 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3847 RANGE is how far to scan while trying to match. RANGE = 0 means try
3848 only at STARTPOS; in general, the last start tried is STARTPOS +
3851 In REGS, return the indices of the virtual concatenation of STRING1
3852 and STRING2 that matched the entire BUFP->buffer and its contained
3855 Do not consider matching one past the index STOP in the virtual
3856 concatenation of STRING1 and STRING2.
3858 We return either the position in the strings at which the match was
3859 found, -1 if no match, or -2 if error (such as failure
3863 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
3864 struct re_pattern_buffer *bufp;
3865 const char *str1, *str2;
3869 struct re_registers *regs;
3873 re_char *string1 = (re_char*) str1;
3874 re_char *string2 = (re_char*) str2;
3875 register char *fastmap = bufp->fastmap;
3876 register RE_TRANSLATE_TYPE translate = bufp->translate;
3877 int total_size = size1 + size2;
3878 int endpos = startpos + range;
3879 boolean anchored_start;
3881 /* Nonzero if we have to concern multibyte character. */
3882 const boolean multibyte = RE_MULTIBYTE_P (bufp);
3884 /* Check for out-of-range STARTPOS. */
3885 if (startpos < 0 || startpos > total_size)
3888 /* Fix up RANGE if it might eventually take us outside
3889 the virtual concatenation of STRING1 and STRING2.
3890 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3892 range = 0 - startpos;
3893 else if (endpos > total_size)
3894 range = total_size - startpos;
3896 /* If the search isn't to be a backwards one, don't waste time in a
3897 search for a pattern anchored at beginning of buffer. */
3898 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3907 /* In a forward search for something that starts with \=.
3908 don't keep searching past point. */
3909 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3911 range = PT_BYTE - BEGV_BYTE - startpos;
3917 /* Update the fastmap now if not correct already. */
3918 if (fastmap && !bufp->fastmap_accurate)
3919 if (re_compile_fastmap (bufp) == -2)
3922 /* See whether the pattern is anchored. */
3923 anchored_start = (bufp->buffer[0] == begline);
3926 gl_state.object = re_match_object;
3928 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
3930 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
3934 /* Loop through the string, looking for a place to start matching. */
3937 /* If the pattern is anchored,
3938 skip quickly past places we cannot match.
3939 We don't bother to treat startpos == 0 specially
3940 because that case doesn't repeat. */
3941 if (anchored_start && startpos > 0)
3943 if (! ((startpos <= size1 ? string1[startpos - 1]
3944 : string2[startpos - size1 - 1])
3949 /* If a fastmap is supplied, skip quickly over characters that
3950 cannot be the start of a match. If the pattern can match the
3951 null string, however, we don't need to skip characters; we want
3952 the first null string. */
3953 if (fastmap && startpos < total_size && !bufp->can_be_null)
3955 register re_char *d;
3956 register unsigned int buf_ch;
3958 d = POS_ADDR_VSTRING (startpos);
3960 if (range > 0) /* Searching forwards. */
3962 register int lim = 0;
3965 if (startpos < size1 && startpos + range >= size1)
3966 lim = range - (size1 - startpos);
3968 /* Written out as an if-else to avoid testing `translate'
3970 if (RE_TRANSLATE_P (translate))
3977 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
3980 buf_ch = RE_TRANSLATE (translate, buf_ch);
3985 range -= buf_charlen;
3990 && !fastmap[RE_TRANSLATE (translate, *d)])
3997 while (range > lim && !fastmap[*d])
4003 startpos += irange - range;
4005 else /* Searching backwards. */
4007 int room = (startpos >= size1
4008 ? size2 + size1 - startpos
4009 : size1 - startpos);
4010 buf_ch = RE_STRING_CHAR (d, room);
4011 buf_ch = TRANSLATE (buf_ch);
4013 if (! (buf_ch >= 0400
4014 || fastmap[buf_ch]))
4019 /* If can't match the null string, and that's all we have left, fail. */
4020 if (range >= 0 && startpos == total_size && fastmap
4021 && !bufp->can_be_null)
4024 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4025 startpos, regs, stop);
4026 #ifndef REGEX_MALLOC
4043 /* Update STARTPOS to the next character boundary. */
4046 re_char *p = POS_ADDR_VSTRING (startpos);
4047 re_char *pend = STOP_ADDR_VSTRING (startpos);
4048 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
4066 /* Update STARTPOS to the previous character boundary. */
4069 re_char *p = POS_ADDR_VSTRING (startpos);
4072 /* Find the head of multibyte form. */
4073 while (!CHAR_HEAD_P (*p))
4078 if (MULTIBYTE_FORM_LENGTH (p, len + 1) != (len + 1))
4094 WEAK_ALIAS (__re_search_2, re_search_2)
4096 /* Declarations and macros for re_match_2. */
4098 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4100 RE_TRANSLATE_TYPE translate,
4101 const int multibyte));
4103 /* This converts PTR, a pointer into one of the search strings `string1'
4104 and `string2' into an offset from the beginning of that string. */
4105 #define POINTER_TO_OFFSET(ptr) \
4106 (FIRST_STRING_P (ptr) \
4107 ? ((regoff_t) ((ptr) - string1)) \
4108 : ((regoff_t) ((ptr) - string2 + size1)))
4110 /* Call before fetching a character with *d. This switches over to
4111 string2 if necessary.
4112 Check re_match_2_internal for a discussion of why end_match_2 might
4113 not be within string2 (but be equal to end_match_1 instead). */
4114 #define PREFETCH() \
4117 /* End of string2 => fail. */ \
4118 if (dend == end_match_2) \
4120 /* End of string1 => advance to string2. */ \
4122 dend = end_match_2; \
4125 /* Call before fetching a char with *d if you already checked other limits.
4126 This is meant for use in lookahead operations like wordend, etc..
4127 where we might need to look at parts of the string that might be
4128 outside of the LIMITs (i.e past `stop'). */
4129 #define PREFETCH_NOLIMIT() \
4133 dend = end_match_2; \
4136 /* Test if at very beginning or at very end of the virtual concatenation
4137 of `string1' and `string2'. If only one string, it's `string2'. */
4138 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4139 #define AT_STRINGS_END(d) ((d) == end2)
4142 /* Test if D points to a character which is word-constituent. We have
4143 two special cases to check for: if past the end of string1, look at
4144 the first character in string2; and if before the beginning of
4145 string2, look at the last character in string1. */
4146 #define WORDCHAR_P(d) \
4147 (SYNTAX ((d) == end1 ? *string2 \
4148 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4151 /* Disabled due to a compiler bug -- see comment at case wordbound */
4153 /* The comment at case wordbound is following one, but we don't use
4154 AT_WORD_BOUNDARY anymore to support multibyte form.
4156 The DEC Alpha C compiler 3.x generates incorrect code for the
4157 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4158 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4159 macro and introducing temporary variables works around the bug. */
4162 /* Test if the character before D and the one at D differ with respect
4163 to being word-constituent. */
4164 #define AT_WORD_BOUNDARY(d) \
4165 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4166 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4169 /* Free everything we malloc. */
4170 #ifdef MATCH_MAY_ALLOCATE
4171 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4172 # define FREE_VARIABLES() \
4174 REGEX_FREE_STACK (fail_stack.stack); \
4175 FREE_VAR (regstart); \
4176 FREE_VAR (regend); \
4177 FREE_VAR (best_regstart); \
4178 FREE_VAR (best_regend); \
4181 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4182 #endif /* not MATCH_MAY_ALLOCATE */
4185 /* Optimization routines. */
4187 /* If the operation is a match against one or more chars,
4188 return a pointer to the next operation, else return NULL. */
4189 static unsigned char *
4193 switch (SWITCH_ENUM_CAST (*p++))
4204 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4207 p = CHARSET_RANGE_TABLE (p - 1);
4208 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4209 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4212 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4219 case notcategoryspec:
4231 /* Jump over non-matching operations. */
4232 static unsigned char *
4233 skip_noops (p, pend)
4234 unsigned char *p, *pend;
4239 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4248 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4259 /* Non-zero if "p1 matches something" implies "p2 fails". */
4261 mutually_exclusive_p (bufp, p1, p2)
4262 struct re_pattern_buffer *bufp;
4263 unsigned char *p1, *p2;
4266 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4267 unsigned char *pend = bufp->buffer + bufp->used;
4269 assert (p1 >= bufp->buffer && p1 < pend
4270 && p2 >= bufp->buffer && p2 <= pend);
4272 /* Skip over open/close-group commands.
4273 If what follows this loop is a ...+ construct,
4274 look at what begins its body, since we will have to
4275 match at least one of that. */
4276 p2 = skip_noops (p2, pend);
4277 /* The same skip can be done for p1, except that this function
4278 is only used in the case where p1 is a simple match operator. */
4279 /* p1 = skip_noops (p1, pend); */
4281 assert (p1 >= bufp->buffer && p1 < pend
4282 && p2 >= bufp->buffer && p2 <= pend);
4284 op2 = p2 == pend ? succeed : *p2;
4286 switch (SWITCH_ENUM_CAST (op2))
4290 /* If we're at the end of the pattern, we can change. */
4291 if (skip_one_char (p1))
4293 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4301 register unsigned int c
4302 = (re_opcode_t) *p2 == endline ? '\n'
4303 : RE_STRING_CHAR(p2 + 2, pend - p2 - 2);
4305 if ((re_opcode_t) *p1 == exactn)
4307 if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2))
4309 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4314 else if ((re_opcode_t) *p1 == charset
4315 || (re_opcode_t) *p1 == charset_not)
4317 int not = (re_opcode_t) *p1 == charset_not;
4319 /* Test if C is listed in charset (or charset_not)
4321 if (SINGLE_BYTE_CHAR_P (c))
4323 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4324 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4327 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4328 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4330 /* `not' is equal to 1 if c would match, which means
4331 that we can't change to pop_failure_jump. */
4334 DEBUG_PRINT1 (" No match => fast loop.\n");
4338 else if ((re_opcode_t) *p1 == anychar
4341 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4350 if ((re_opcode_t) *p1 == exactn)
4351 /* Reuse the code above. */
4352 return mutually_exclusive_p (bufp, p2, p1);
4355 /* It is hard to list up all the character in charset
4356 P2 if it includes multibyte character. Give up in
4358 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4360 /* Now, we are sure that P2 has no range table.
4361 So, for the size of bitmap in P2, `p2[1]' is
4362 enough. But P1 may have range table, so the
4363 size of bitmap table of P1 is extracted by
4364 using macro `CHARSET_BITMAP_SIZE'.
4366 Since we know that all the character listed in
4367 P2 is ASCII, it is enough to test only bitmap
4373 /* We win if the charset inside the loop
4374 has no overlap with the one after the loop. */
4377 && idx < CHARSET_BITMAP_SIZE (p1));
4379 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4383 || idx == CHARSET_BITMAP_SIZE (p1))
4385 DEBUG_PRINT1 (" No match => fast loop.\n");
4389 else if ((re_opcode_t) *p1 == charset
4390 || (re_opcode_t) *p1 == charset_not)
4393 /* We win if the charset_not inside the loop lists
4394 every character listed in the charset after. */
4395 for (idx = 0; idx < (int) p2[1]; idx++)
4396 if (! (p2[2 + idx] == 0
4397 || (idx < CHARSET_BITMAP_SIZE (p1)
4398 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4403 DEBUG_PRINT1 (" No match => fast loop.\n");
4412 return ((re_opcode_t) *p1 == syntaxspec
4413 && p1[1] == (op2 == wordend ? Sword : p2[1]));
4417 return ((re_opcode_t) *p1 == notsyntaxspec
4418 && p1[1] == (op2 == wordend ? Sword : p2[1]));
4421 return (((re_opcode_t) *p1 == notsyntaxspec
4422 || (re_opcode_t) *p1 == syntaxspec)
4427 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4428 case notcategoryspec:
4429 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4441 /* Matching routines. */
4443 #ifndef emacs /* Emacs never uses this. */
4444 /* re_match is like re_match_2 except it takes only a single string. */
4447 re_match (bufp, string, size, pos, regs)
4448 struct re_pattern_buffer *bufp;
4451 struct re_registers *regs;
4453 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
4455 # if defined C_ALLOCA && !defined REGEX_MALLOC
4460 WEAK_ALIAS (__re_match, re_match)
4461 #endif /* not emacs */
4464 /* In Emacs, this is the string or buffer in which we
4465 are matching. It is used for looking up syntax properties. */
4466 Lisp_Object re_match_object;
4469 /* re_match_2 matches the compiled pattern in BUFP against the
4470 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4471 and SIZE2, respectively). We start matching at POS, and stop
4474 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4475 store offsets for the substring each group matched in REGS. See the
4476 documentation for exactly how many groups we fill.
4478 We return -1 if no match, -2 if an internal error (such as the
4479 failure stack overflowing). Otherwise, we return the length of the
4480 matched substring. */
4483 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
4484 struct re_pattern_buffer *bufp;
4485 const char *string1, *string2;
4488 struct re_registers *regs;
4495 gl_state.object = re_match_object;
4496 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4497 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4500 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4501 (re_char*) string2, size2,
4503 #if defined C_ALLOCA && !defined REGEX_MALLOC
4508 WEAK_ALIAS (__re_match_2, re_match_2)
4510 /* This is a separate function so that we can force an alloca cleanup
4513 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
4514 struct re_pattern_buffer *bufp;
4515 re_char *string1, *string2;
4518 struct re_registers *regs;
4521 /* General temporaries. */
4526 /* Just past the end of the corresponding string. */
4527 re_char *end1, *end2;
4529 /* Pointers into string1 and string2, just past the last characters in
4530 each to consider matching. */
4531 re_char *end_match_1, *end_match_2;
4533 /* Where we are in the data, and the end of the current string. */
4536 /* Used sometimes to remember where we were before starting matching
4537 an operator so that we can go back in case of failure. This "atomic"
4538 behavior of matching opcodes is indispensable to the correctness
4539 of the on_failure_keep_string_jump optimization. */
4542 /* Where we are in the pattern, and the end of the pattern. */
4543 unsigned char *p = bufp->buffer;
4544 register unsigned char *pend = p + bufp->used;
4546 /* We use this to map every character in the string. */
4547 RE_TRANSLATE_TYPE translate = bufp->translate;
4549 /* Nonzero if we have to concern multibyte character. */
4550 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4552 /* Failure point stack. Each place that can handle a failure further
4553 down the line pushes a failure point on this stack. It consists of
4554 regstart, and regend for all registers corresponding to
4555 the subexpressions we're currently inside, plus the number of such
4556 registers, and, finally, two char *'s. The first char * is where
4557 to resume scanning the pattern; the second one is where to resume
4558 scanning the strings. */
4559 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4560 fail_stack_type fail_stack;
4563 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4566 #if defined REL_ALLOC && defined REGEX_MALLOC
4567 /* This holds the pointer to the failure stack, when
4568 it is allocated relocatably. */
4569 fail_stack_elt_t *failure_stack_ptr;
4572 /* We fill all the registers internally, independent of what we
4573 return, for use in backreferences. The number here includes
4574 an element for register zero. */
4575 size_t num_regs = bufp->re_nsub + 1;
4577 /* Information on the contents of registers. These are pointers into
4578 the input strings; they record just what was matched (on this
4579 attempt) by a subexpression part of the pattern, that is, the
4580 regnum-th regstart pointer points to where in the pattern we began
4581 matching and the regnum-th regend points to right after where we
4582 stopped matching the regnum-th subexpression. (The zeroth register
4583 keeps track of what the whole pattern matches.) */
4584 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4585 re_char **regstart, **regend;
4588 /* The following record the register info as found in the above
4589 variables when we find a match better than any we've seen before.
4590 This happens as we backtrack through the failure points, which in
4591 turn happens only if we have not yet matched the entire string. */
4592 unsigned best_regs_set = false;
4593 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4594 re_char **best_regstart, **best_regend;
4597 /* Logically, this is `best_regend[0]'. But we don't want to have to
4598 allocate space for that if we're not allocating space for anything
4599 else (see below). Also, we never need info about register 0 for
4600 any of the other register vectors, and it seems rather a kludge to
4601 treat `best_regend' differently than the rest. So we keep track of
4602 the end of the best match so far in a separate variable. We
4603 initialize this to NULL so that when we backtrack the first time
4604 and need to test it, it's not garbage. */
4605 re_char *match_end = NULL;
4608 /* Counts the total number of registers pushed. */
4609 unsigned num_regs_pushed = 0;
4612 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4616 #ifdef MATCH_MAY_ALLOCATE
4617 /* Do not bother to initialize all the register variables if there are
4618 no groups in the pattern, as it takes a fair amount of time. If
4619 there are groups, we include space for register 0 (the whole
4620 pattern), even though we never use it, since it simplifies the
4621 array indexing. We should fix this. */
4624 regstart = REGEX_TALLOC (num_regs, re_char *);
4625 regend = REGEX_TALLOC (num_regs, re_char *);
4626 best_regstart = REGEX_TALLOC (num_regs, re_char *);
4627 best_regend = REGEX_TALLOC (num_regs, re_char *);
4629 if (!(regstart && regend && best_regstart && best_regend))
4637 /* We must initialize all our variables to NULL, so that
4638 `FREE_VARIABLES' doesn't try to free them. */
4639 regstart = regend = best_regstart = best_regend = NULL;
4641 #endif /* MATCH_MAY_ALLOCATE */
4643 /* The starting position is bogus. */
4644 if (pos < 0 || pos > size1 + size2)
4650 /* Initialize subexpression text positions to -1 to mark ones that no
4651 start_memory/stop_memory has been seen for. Also initialize the
4652 register information struct. */
4653 for (mcnt = 1; mcnt < num_regs; mcnt++)
4654 regstart[mcnt] = regend[mcnt] = NULL;
4656 /* We move `string1' into `string2' if the latter's empty -- but not if
4657 `string1' is null. */
4658 if (size2 == 0 && string1 != NULL)
4665 end1 = string1 + size1;
4666 end2 = string2 + size2;
4668 /* `p' scans through the pattern as `d' scans through the data.
4669 `dend' is the end of the input string that `d' points within. `d'
4670 is advanced into the following input string whenever necessary, but
4671 this happens before fetching; therefore, at the beginning of the
4672 loop, `d' can be pointing at the end of a string, but it cannot
4676 /* Only match within string2. */
4677 d = string2 + pos - size1;
4678 dend = end_match_2 = string2 + stop - size1;
4679 end_match_1 = end1; /* Just to give it a value. */
4685 /* Only match within string1. */
4686 end_match_1 = string1 + stop;
4688 When we reach end_match_1, PREFETCH normally switches to string2.
4689 But in the present case, this means that just doing a PREFETCH
4690 makes us jump from `stop' to `gap' within the string.
4691 What we really want here is for the search to stop as
4692 soon as we hit end_match_1. That's why we set end_match_2
4693 to end_match_1 (since PREFETCH fails as soon as we hit
4695 end_match_2 = end_match_1;
4698 { /* It's important to use this code when stop == size so that
4699 moving `d' from end1 to string2 will not prevent the d == dend
4700 check from catching the end of string. */
4702 end_match_2 = string2 + stop - size1;
4708 DEBUG_PRINT1 ("The compiled pattern is: ");
4709 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
4710 DEBUG_PRINT1 ("The string to match is: `");
4711 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
4712 DEBUG_PRINT1 ("'\n");
4714 /* This loops over pattern commands. It exits by returning from the
4715 function if the match is complete, or it drops through if the match
4716 fails at this starting point in the input data. */
4719 DEBUG_PRINT2 ("\n%p: ", p);
4722 { /* End of pattern means we might have succeeded. */
4723 DEBUG_PRINT1 ("end of pattern ... ");
4725 /* If we haven't matched the entire string, and we want the
4726 longest match, try backtracking. */
4727 if (d != end_match_2)
4729 /* 1 if this match ends in the same string (string1 or string2)
4730 as the best previous match. */
4731 boolean same_str_p = (FIRST_STRING_P (match_end)
4732 == FIRST_STRING_P (d));
4733 /* 1 if this match is the best seen so far. */
4734 boolean best_match_p;
4736 /* AIX compiler got confused when this was combined
4737 with the previous declaration. */
4739 best_match_p = d > match_end;
4741 best_match_p = !FIRST_STRING_P (d);
4743 DEBUG_PRINT1 ("backtracking.\n");
4745 if (!FAIL_STACK_EMPTY ())
4746 { /* More failure points to try. */
4748 /* If exceeds best match so far, save it. */
4749 if (!best_regs_set || best_match_p)
4751 best_regs_set = true;
4754 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4756 for (mcnt = 1; mcnt < num_regs; mcnt++)
4758 best_regstart[mcnt] = regstart[mcnt];
4759 best_regend[mcnt] = regend[mcnt];
4765 /* If no failure points, don't restore garbage. And if
4766 last match is real best match, don't restore second
4768 else if (best_regs_set && !best_match_p)
4771 /* Restore best match. It may happen that `dend ==
4772 end_match_1' while the restored d is in string2.
4773 For example, the pattern `x.*y.*z' against the
4774 strings `x-' and `y-z-', if the two strings are
4775 not consecutive in memory. */
4776 DEBUG_PRINT1 ("Restoring best registers.\n");
4779 dend = ((d >= string1 && d <= end1)
4780 ? end_match_1 : end_match_2);
4782 for (mcnt = 1; mcnt < num_regs; mcnt++)
4784 regstart[mcnt] = best_regstart[mcnt];
4785 regend[mcnt] = best_regend[mcnt];
4788 } /* d != end_match_2 */
4791 DEBUG_PRINT1 ("Accepting match.\n");
4793 /* If caller wants register contents data back, do it. */
4794 if (regs && !bufp->no_sub)
4796 /* Have the register data arrays been allocated? */
4797 if (bufp->regs_allocated == REGS_UNALLOCATED)
4798 { /* No. So allocate them with malloc. We need one
4799 extra element beyond `num_regs' for the `-1' marker
4801 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4802 regs->start = TALLOC (regs->num_regs, regoff_t);
4803 regs->end = TALLOC (regs->num_regs, regoff_t);
4804 if (regs->start == NULL || regs->end == NULL)
4809 bufp->regs_allocated = REGS_REALLOCATE;
4811 else if (bufp->regs_allocated == REGS_REALLOCATE)
4812 { /* Yes. If we need more elements than were already
4813 allocated, reallocate them. If we need fewer, just
4815 if (regs->num_regs < num_regs + 1)
4817 regs->num_regs = num_regs + 1;
4818 RETALLOC (regs->start, regs->num_regs, regoff_t);
4819 RETALLOC (regs->end, regs->num_regs, regoff_t);
4820 if (regs->start == NULL || regs->end == NULL)
4829 /* These braces fend off a "empty body in an else-statement"
4830 warning under GCC when assert expands to nothing. */
4831 assert (bufp->regs_allocated == REGS_FIXED);
4834 /* Convert the pointer data in `regstart' and `regend' to
4835 indices. Register zero has to be set differently,
4836 since we haven't kept track of any info for it. */
4837 if (regs->num_regs > 0)
4839 regs->start[0] = pos;
4840 regs->end[0] = POINTER_TO_OFFSET (d);
4843 /* Go through the first `min (num_regs, regs->num_regs)'
4844 registers, since that is all we initialized. */
4845 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
4847 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4848 regs->start[mcnt] = regs->end[mcnt] = -1;
4852 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4854 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4858 /* If the regs structure we return has more elements than
4859 were in the pattern, set the extra elements to -1. If
4860 we (re)allocated the registers, this is the case,
4861 because we always allocate enough to have at least one
4863 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
4864 regs->start[mcnt] = regs->end[mcnt] = -1;
4865 } /* regs && !bufp->no_sub */
4867 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4868 nfailure_points_pushed, nfailure_points_popped,
4869 nfailure_points_pushed - nfailure_points_popped);
4870 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4872 mcnt = POINTER_TO_OFFSET (d) - pos;
4874 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4880 /* Otherwise match next pattern command. */
4881 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4883 /* Ignore these. Used to ignore the n of succeed_n's which
4884 currently have n == 0. */
4886 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4890 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4893 /* Match the next n pattern characters exactly. The following
4894 byte in the pattern defines n, and the n bytes after that
4895 are the characters to match. */
4898 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4900 /* Remember the start point to rollback upon failure. */
4903 /* This is written out as an if-else so we don't waste time
4904 testing `translate' inside the loop. */
4905 if (RE_TRANSLATE_P (translate))
4910 int pat_charlen, buf_charlen;
4911 unsigned int pat_ch, buf_ch;
4914 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
4915 buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
4917 if (RE_TRANSLATE (translate, buf_ch)
4926 mcnt -= pat_charlen;
4933 if (RE_TRANSLATE (translate, *d) != *p++)
4958 /* Match any character except possibly a newline or a null. */
4962 unsigned int buf_ch;
4964 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4967 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
4968 buf_ch = TRANSLATE (buf_ch);
4970 if ((!(bufp->syntax & RE_DOT_NEWLINE)
4972 || ((bufp->syntax & RE_DOT_NOT_NULL)
4973 && buf_ch == '\000'))
4976 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4985 register unsigned int c;
4986 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4989 /* Start of actual range_table, or end of bitmap if there is no
4991 unsigned char *range_table;
4993 /* Nonzero if there is a range table. */
4994 int range_table_exists;
4996 /* Number of ranges of range table. This is not included
4997 in the initial byte-length of the command. */
5000 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5002 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5004 if (range_table_exists)
5006 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5007 EXTRACT_NUMBER_AND_INCR (count, range_table);
5011 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len);
5012 c = TRANSLATE (c); /* The character to match. */
5014 if (SINGLE_BYTE_CHAR_P (c))
5015 { /* Lookup bitmap. */
5016 /* Cast to `unsigned' instead of `unsigned char' in
5017 case the bit list is a full 32 bytes long. */
5018 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5019 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5023 else if (range_table_exists)
5025 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5027 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5028 | (class_bits & BIT_MULTIBYTE)
5029 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5030 | (class_bits & BIT_SPACE && ISSPACE (c))
5031 | (class_bits & BIT_UPPER && ISUPPER (c))
5032 | (class_bits & BIT_WORD && ISWORD (c)))
5035 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5039 if (range_table_exists)
5040 p = CHARSET_RANGE_TABLE_END (range_table, count);
5042 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5044 if (!not) goto fail;
5051 /* The beginning of a group is represented by start_memory.
5052 The argument is the register number. The text
5053 matched within the group is recorded (in the internal
5054 registers data structure) under the register number. */
5056 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5058 /* In case we need to undo this operation (via backtracking). */
5059 PUSH_FAILURE_REG ((unsigned int)*p);
5062 regend[*p] = NULL; /* probably unnecessary. -sm */
5063 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5065 /* Move past the register number and inner group count. */
5070 /* The stop_memory opcode represents the end of a group. Its
5071 argument is the same as start_memory's: the register number. */
5073 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5075 assert (!REG_UNSET (regstart[*p]));
5076 /* Strictly speaking, there should be code such as:
5078 assert (REG_UNSET (regend[*p]));
5079 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5081 But the only info to be pushed is regend[*p] and it is known to
5082 be UNSET, so there really isn't anything to push.
5083 Not pushing anything, on the other hand deprives us from the
5084 guarantee that regend[*p] is UNSET since undoing this operation
5085 will not reset its value properly. This is not important since
5086 the value will only be read on the next start_memory or at
5087 the very end and both events can only happen if this stop_memory
5091 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5093 /* Move past the register number and the inner group count. */
5098 /* \<digit> has been turned into a `duplicate' command which is
5099 followed by the numeric value of <digit> as the register number. */
5102 register re_char *d2, *dend2;
5103 int regno = *p++; /* Get which register to match against. */
5104 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5106 /* Can't back reference a group which we've never matched. */
5107 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5110 /* Where in input to try to start matching. */
5111 d2 = regstart[regno];
5113 /* Remember the start point to rollback upon failure. */
5116 /* Where to stop matching; if both the place to start and
5117 the place to stop matching are in the same string, then
5118 set to the place to stop, otherwise, for now have to use
5119 the end of the first string. */
5121 dend2 = ((FIRST_STRING_P (regstart[regno])
5122 == FIRST_STRING_P (regend[regno]))
5123 ? regend[regno] : end_match_1);
5126 /* If necessary, advance to next segment in register
5130 if (dend2 == end_match_2) break;
5131 if (dend2 == regend[regno]) break;
5133 /* End of string1 => advance to string2. */
5135 dend2 = regend[regno];
5137 /* At end of register contents => success */
5138 if (d2 == dend2) break;
5140 /* If necessary, advance to next segment in data. */
5143 /* How many characters left in this segment to match. */
5146 /* Want how many consecutive characters we can match in
5147 one shot, so, if necessary, adjust the count. */
5148 if (mcnt > dend2 - d2)
5151 /* Compare that many; failure if mismatch, else move
5153 if (RE_TRANSLATE_P (translate)
5154 ? bcmp_translate (d, d2, mcnt, translate, multibyte)
5155 : memcmp (d, d2, mcnt))
5160 d += mcnt, d2 += mcnt;
5166 /* begline matches the empty string at the beginning of the string
5167 (unless `not_bol' is set in `bufp'), and after newlines. */
5169 DEBUG_PRINT1 ("EXECUTING begline.\n");
5171 if (AT_STRINGS_BEG (d))
5173 if (!bufp->not_bol) break;
5178 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5182 /* In all other cases, we fail. */
5186 /* endline is the dual of begline. */
5188 DEBUG_PRINT1 ("EXECUTING endline.\n");
5190 if (AT_STRINGS_END (d))
5192 if (!bufp->not_eol) break;
5196 PREFETCH_NOLIMIT ();
5203 /* Match at the very beginning of the data. */
5205 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5206 if (AT_STRINGS_BEG (d))
5211 /* Match at the very end of the data. */
5213 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5214 if (AT_STRINGS_END (d))
5219 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5220 pushes NULL as the value for the string on the stack. Then
5221 `POP_FAILURE_POINT' will keep the current value for the
5222 string, instead of restoring it. To see why, consider
5223 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5224 then the . fails against the \n. But the next thing we want
5225 to do is match the \n against the \n; if we restored the
5226 string value, we would be back at the foo.
5228 Because this is used only in specific cases, we don't need to
5229 check all the things that `on_failure_jump' does, to make
5230 sure the right things get saved on the stack. Hence we don't
5231 share its code. The only reason to push anything on the
5232 stack at all is that otherwise we would have to change
5233 `anychar's code to do something besides goto fail in this
5234 case; that seems worse than this. */
5235 case on_failure_keep_string_jump:
5236 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5237 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5240 PUSH_FAILURE_POINT (p - 3, NULL);
5243 /* A nasty loop is introduced by the non-greedy *? and +?.
5244 With such loops, the stack only ever contains one failure point
5245 at a time, so that a plain on_failure_jump_loop kind of
5246 cycle detection cannot work. Worse yet, such a detection
5247 can not only fail to detect a cycle, but it can also wrongly
5248 detect a cycle (between different instantiations of the same
5250 So the method used for those nasty loops is a little different:
5251 We use a special cycle-detection-stack-frame which is pushed
5252 when the on_failure_jump_nastyloop failure-point is *popped*.
5253 This special frame thus marks the beginning of one iteration
5254 through the loop and we can hence easily check right here
5255 whether something matched between the beginning and the end of
5257 case on_failure_jump_nastyloop:
5258 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5259 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5262 assert ((re_opcode_t)p[-4] == no_op);
5263 CHECK_INFINITE_LOOP (p - 4, d);
5264 PUSH_FAILURE_POINT (p - 3, d);
5268 /* Simple loop detecting on_failure_jump: just check on the
5269 failure stack if the same spot was already hit earlier. */
5270 case on_failure_jump_loop:
5272 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5273 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5276 CHECK_INFINITE_LOOP (p - 3, d);
5277 PUSH_FAILURE_POINT (p - 3, d);
5281 /* Uses of on_failure_jump:
5283 Each alternative starts with an on_failure_jump that points
5284 to the beginning of the next alternative. Each alternative
5285 except the last ends with a jump that in effect jumps past
5286 the rest of the alternatives. (They really jump to the
5287 ending jump of the following alternative, because tensioning
5288 these jumps is a hassle.)
5290 Repeats start with an on_failure_jump that points past both
5291 the repetition text and either the following jump or
5292 pop_failure_jump back to this on_failure_jump. */
5293 case on_failure_jump:
5295 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5296 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5299 PUSH_FAILURE_POINT (p -3, d);
5302 /* This operation is used for greedy *.
5303 Compare the beginning of the repeat with what in the
5304 pattern follows its end. If we can establish that there
5305 is nothing that they would both match, i.e., that we
5306 would have to backtrack because of (as in, e.g., `a*a')
5307 then we can use a non-backtracking loop based on
5308 on_failure_keep_string_jump instead of on_failure_jump. */
5309 case on_failure_jump_smart:
5311 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5312 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5315 unsigned char *p1 = p; /* Next operation. */
5316 unsigned char *p2 = p + mcnt; /* Destination of the jump. */
5318 p -= 3; /* Reset so that we will re-execute the
5319 instruction once it's been changed. */
5321 EXTRACT_NUMBER (mcnt, p2 - 2);
5323 /* Ensure this is a indeed the trivial kind of loop
5324 we are expecting. */
5325 assert (skip_one_char (p1) == p2 - 3);
5326 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5327 DEBUG_STATEMENT (debug += 2);
5328 if (mutually_exclusive_p (bufp, p1, p2))
5330 /* Use a fast `on_failure_keep_string_jump' loop. */
5331 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5332 *p = (unsigned char) on_failure_keep_string_jump;
5333 STORE_NUMBER (p2 - 2, mcnt + 3);
5337 /* Default to a safe `on_failure_jump' loop. */
5338 DEBUG_PRINT1 (" smart default => slow loop.\n");
5339 *p = (unsigned char) on_failure_jump;
5341 DEBUG_STATEMENT (debug -= 2);
5345 /* Unconditionally jump (without popping any failure points). */
5349 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5350 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5351 p += mcnt; /* Do the jump. */
5352 DEBUG_PRINT2 ("(to %p).\n", p);
5356 /* Have to succeed matching what follows at least n times.
5357 After that, handle like `on_failure_jump'. */
5359 EXTRACT_NUMBER (mcnt, p + 2);
5360 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5362 /* Originally, mcnt is how many times we HAVE to succeed. */
5367 PUSH_FAILURE_COUNT (p);
5368 DEBUG_PRINT3 (" Setting %p to %d.\n", p, mcnt);
5369 STORE_NUMBER_AND_INCR (p, mcnt);
5372 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5377 EXTRACT_NUMBER (mcnt, p + 2);
5378 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5380 /* Originally, this is how many times we CAN jump. */
5384 PUSH_FAILURE_COUNT (p + 2);
5385 STORE_NUMBER (p + 2, mcnt);
5386 goto unconditional_jump;
5388 /* If don't have to jump any more, skip over the rest of command. */
5395 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5397 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5399 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5400 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
5401 PUSH_FAILURE_COUNT (p1);
5402 STORE_NUMBER (p1, mcnt);
5408 not = (re_opcode_t) *(p - 1) == notwordbound;
5409 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5411 /* We SUCCEED (or FAIL) in one of the following cases: */
5413 /* Case 1: D is at the beginning or the end of string. */
5414 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5418 /* C1 is the character before D, S1 is the syntax of C1, C2
5419 is the character at D, and S2 is the syntax of C2. */
5422 int offset = PTR_TO_OFFSET (d - 1);
5423 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5424 UPDATE_SYNTAX_TABLE (charpos);
5426 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5429 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
5431 PREFETCH_NOLIMIT ();
5432 c2 = RE_STRING_CHAR (d, dend - d);
5435 if (/* Case 2: Only one of S1 and S2 is Sword. */
5436 ((s1 == Sword) != (s2 == Sword))
5437 /* Case 3: Both of S1 and S2 are Sword, and macro
5438 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5439 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5448 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5450 /* We FAIL in one of the following cases: */
5452 /* Case 1: D is at the end of string. */
5453 if (AT_STRINGS_END (d))
5457 /* C1 is the character before D, S1 is the syntax of C1, C2
5458 is the character at D, and S2 is the syntax of C2. */
5461 int offset = PTR_TO_OFFSET (d);
5462 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5463 UPDATE_SYNTAX_TABLE (charpos);
5466 c2 = RE_STRING_CHAR (d, dend - d);
5469 /* Case 2: S2 is not Sword. */
5473 /* Case 3: D is not at the beginning of string ... */
5474 if (!AT_STRINGS_BEG (d))
5476 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5478 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
5482 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5484 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
5491 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5493 /* We FAIL in one of the following cases: */
5495 /* Case 1: D is at the beginning of string. */
5496 if (AT_STRINGS_BEG (d))
5500 /* C1 is the character before D, S1 is the syntax of C1, C2
5501 is the character at D, and S2 is the syntax of C2. */
5504 int offset = PTR_TO_OFFSET (d) - 1;
5505 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5506 UPDATE_SYNTAX_TABLE (charpos);
5508 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5511 /* Case 2: S1 is not Sword. */
5515 /* Case 3: D is not at the end of string ... */
5516 if (!AT_STRINGS_END (d))
5518 PREFETCH_NOLIMIT ();
5519 c2 = RE_STRING_CHAR (d, dend - d);
5521 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
5525 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5527 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
5535 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
5537 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
5541 int offset = PTR_TO_OFFSET (d);
5542 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5543 UPDATE_SYNTAX_TABLE (pos1);
5549 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len);
5551 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
5559 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5560 if (PTR_BYTE_POS (d) >= PT_BYTE)
5565 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5566 if (PTR_BYTE_POS (d) != PT_BYTE)
5571 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5572 if (PTR_BYTE_POS (d) <= PT_BYTE)
5577 case notcategoryspec:
5578 not = (re_opcode_t) *(p - 1) == notcategoryspec;
5580 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
5584 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len);
5586 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
5597 continue; /* Successfully executed one pattern command; keep going. */
5600 /* We goto here if a matching operation fails. */
5603 if (!FAIL_STACK_EMPTY ())
5607 /* A restart point is known. Restore to that state. */
5608 DEBUG_PRINT1 ("\nFAIL:\n");
5609 POP_FAILURE_POINT (str, pat);
5610 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
5612 case on_failure_keep_string_jump:
5613 assert (str == NULL);
5614 goto continue_failure_jump;
5616 case on_failure_jump_nastyloop:
5617 assert ((re_opcode_t)pat[-2] == no_op);
5618 PUSH_FAILURE_POINT (pat - 2, str);
5621 case on_failure_jump_loop:
5622 case on_failure_jump:
5625 continue_failure_jump:
5626 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
5631 /* A special frame used for nastyloops. */
5638 assert (p >= bufp->buffer && p <= pend);
5640 if (d >= string1 && d <= end1)
5644 break; /* Matching at this starting point really fails. */
5648 goto restore_best_regs;
5652 return -1; /* Failure to match. */
5655 /* Subroutine definitions for re_match_2. */
5657 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5658 bytes; nonzero otherwise. */
5661 bcmp_translate (s1, s2, len, translate, multibyte)
5664 RE_TRANSLATE_TYPE translate;
5665 const int multibyte;
5667 register re_char *p1 = s1, *p2 = s2;
5668 re_char *p1_end = s1 + len;
5669 re_char *p2_end = s2 + len;
5671 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
5672 different lengths, but relying on a single `len' would break this. -sm */
5673 while (p1 < p1_end && p2 < p2_end)
5675 int p1_charlen, p2_charlen;
5678 p1_ch = RE_STRING_CHAR_AND_LENGTH (p1, p1_end - p1, p1_charlen);
5679 p2_ch = RE_STRING_CHAR_AND_LENGTH (p2, p2_end - p2, p2_charlen);
5681 if (RE_TRANSLATE (translate, p1_ch)
5682 != RE_TRANSLATE (translate, p2_ch))
5685 p1 += p1_charlen, p2 += p2_charlen;
5688 if (p1 != p1_end || p2 != p2_end)
5694 /* Entry points for GNU code. */
5696 /* re_compile_pattern is the GNU regular expression compiler: it
5697 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5698 Returns 0 if the pattern was valid, otherwise an error string.
5700 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5701 are set in BUFP on entry.
5703 We call regex_compile to do the actual compilation. */
5706 re_compile_pattern (pattern, length, bufp)
5707 const char *pattern;
5709 struct re_pattern_buffer *bufp;
5713 /* GNU code is written to assume at least RE_NREGS registers will be set
5714 (and at least one extra will be -1). */
5715 bufp->regs_allocated = REGS_UNALLOCATED;
5717 /* And GNU code determines whether or not to get register information
5718 by passing null for the REGS argument to re_match, etc., not by
5722 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
5726 return gettext (re_error_msgid[(int) ret]);
5728 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
5730 /* Entry points compatible with 4.2 BSD regex library. We don't define
5731 them unless specifically requested. */
5733 #if defined _REGEX_RE_COMP || defined _LIBC
5735 /* BSD has one and only one pattern buffer. */
5736 static struct re_pattern_buffer re_comp_buf;
5740 /* Make these definitions weak in libc, so POSIX programs can redefine
5741 these names if they don't use our functions, and still use
5742 regcomp/regexec below without link errors. */
5752 if (!re_comp_buf.buffer)
5753 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5754 return (char *) gettext ("No previous regular expression");
5758 if (!re_comp_buf.buffer)
5760 re_comp_buf.buffer = (unsigned char *) malloc (200);
5761 if (re_comp_buf.buffer == NULL)
5762 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5763 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
5764 re_comp_buf.allocated = 200;
5766 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5767 if (re_comp_buf.fastmap == NULL)
5768 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5769 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
5772 /* Since `re_exec' always passes NULL for the `regs' argument, we
5773 don't need to initialize the pattern buffer fields which affect it. */
5775 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5780 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5781 return (char *) gettext (re_error_msgid[(int) ret]);
5792 const int len = strlen (s);
5794 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5796 #endif /* _REGEX_RE_COMP */
5798 /* POSIX.2 functions. Don't define these for Emacs. */
5802 /* regcomp takes a regular expression as a string and compiles it.
5804 PREG is a regex_t *. We do not expect any fields to be initialized,
5805 since POSIX says we shouldn't. Thus, we set
5807 `buffer' to the compiled pattern;
5808 `used' to the length of the compiled pattern;
5809 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5810 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5811 RE_SYNTAX_POSIX_BASIC;
5812 `fastmap' to an allocated space for the fastmap;
5813 `fastmap_accurate' to zero;
5814 `re_nsub' to the number of subexpressions in PATTERN.
5816 PATTERN is the address of the pattern string.
5818 CFLAGS is a series of bits which affect compilation.
5820 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5821 use POSIX basic syntax.
5823 If REG_NEWLINE is set, then . and [^...] don't match newline.
5824 Also, regexec will try a match beginning after every newline.
5826 If REG_ICASE is set, then we considers upper- and lowercase
5827 versions of letters to be equivalent when matching.
5829 If REG_NOSUB is set, then when PREG is passed to regexec, that
5830 routine will report only success or failure, and nothing about the
5833 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5834 the return codes and their meanings.) */
5837 regcomp (preg, pattern, cflags)
5839 const char *pattern;
5844 = (cflags & REG_EXTENDED) ?
5845 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5847 /* regex_compile will allocate the space for the compiled pattern. */
5849 preg->allocated = 0;
5852 /* Try to allocate space for the fastmap. */
5853 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
5855 if (cflags & REG_ICASE)
5860 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5861 * sizeof (*(RE_TRANSLATE_TYPE)0));
5862 if (preg->translate == NULL)
5863 return (int) REG_ESPACE;
5865 /* Map uppercase characters to corresponding lowercase ones. */
5866 for (i = 0; i < CHAR_SET_SIZE; i++)
5867 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
5870 preg->translate = NULL;
5872 /* If REG_NEWLINE is set, newlines are treated differently. */
5873 if (cflags & REG_NEWLINE)
5874 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5875 syntax &= ~RE_DOT_NEWLINE;
5876 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5879 syntax |= RE_NO_NEWLINE_ANCHOR;
5881 preg->no_sub = !!(cflags & REG_NOSUB);
5883 /* POSIX says a null character in the pattern terminates it, so we
5884 can use strlen here in compiling the pattern. */
5885 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
5887 /* POSIX doesn't distinguish between an unmatched open-group and an
5888 unmatched close-group: both are REG_EPAREN. */
5889 if (ret == REG_ERPAREN)
5892 if (ret == REG_NOERROR && preg->fastmap)
5893 { /* Compute the fastmap now, since regexec cannot modify the pattern
5895 re_compile_fastmap (preg);
5896 if (preg->can_be_null)
5897 { /* The fastmap can't be used anyway. */
5898 free (preg->fastmap);
5899 preg->fastmap = NULL;
5904 WEAK_ALIAS (__regcomp, regcomp)
5907 /* regexec searches for a given pattern, specified by PREG, in the
5910 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5911 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5912 least NMATCH elements, and we set them to the offsets of the
5913 corresponding matched substrings.
5915 EFLAGS specifies `execution flags' which affect matching: if
5916 REG_NOTBOL is set, then ^ does not match at the beginning of the
5917 string; if REG_NOTEOL is set, then $ does not match at the end.
5919 We return 0 if we find a match and REG_NOMATCH if not. */
5922 regexec (preg, string, nmatch, pmatch, eflags)
5923 const regex_t *preg;
5926 regmatch_t pmatch[];
5930 struct re_registers regs;
5931 regex_t private_preg;
5932 int len = strlen (string);
5933 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
5935 private_preg = *preg;
5937 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5938 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5940 /* The user has told us exactly how many registers to return
5941 information about, via `nmatch'. We have to pass that on to the
5942 matching routines. */
5943 private_preg.regs_allocated = REGS_FIXED;
5947 regs.num_regs = nmatch;
5948 regs.start = TALLOC (nmatch * 2, regoff_t);
5949 if (regs.start == NULL)
5950 return (int) REG_NOMATCH;
5951 regs.end = regs.start + nmatch;
5954 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
5955 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
5956 was a little bit longer but still only matching the real part.
5957 This works because the `endline' will check for a '\n' and will find a
5958 '\0', correctly deciding that this is not the end of a line.
5959 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
5960 a convenient '\0' there. For all we know, the string could be preceded
5961 by '\n' which would throw things off. */
5963 /* Perform the searching operation. */
5964 ret = re_search (&private_preg, string, len,
5965 /* start: */ 0, /* range: */ len,
5966 want_reg_info ? ®s : (struct re_registers *) 0);
5968 /* Copy the register information to the POSIX structure. */
5975 for (r = 0; r < nmatch; r++)
5977 pmatch[r].rm_so = regs.start[r];
5978 pmatch[r].rm_eo = regs.end[r];
5982 /* If we needed the temporary register info, free the space now. */
5986 /* We want zero return to mean success, unlike `re_search'. */
5987 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5989 WEAK_ALIAS (__regexec, regexec)
5992 /* Returns a message corresponding to an error code, ERRCODE, returned
5993 from either regcomp or regexec. We don't use PREG here. */
5996 regerror (errcode, preg, errbuf, errbuf_size)
5998 const regex_t *preg;
6006 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6007 /* Only error codes returned by the rest of the code should be passed
6008 to this routine. If we are given anything else, or if other regex
6009 code generates an invalid error code, then the program has a bug.
6010 Dump core so we can fix it. */
6013 msg = gettext (re_error_msgid[errcode]);
6015 msg_size = strlen (msg) + 1; /* Includes the null. */
6017 if (errbuf_size != 0)
6019 if (msg_size > errbuf_size)
6021 strncpy (errbuf, msg, errbuf_size - 1);
6022 errbuf[errbuf_size - 1] = 0;
6025 strcpy (errbuf, msg);
6030 WEAK_ALIAS (__regerror, regerror)
6033 /* Free dynamically allocated space used by PREG. */
6039 if (preg->buffer != NULL)
6040 free (preg->buffer);
6041 preg->buffer = NULL;
6043 preg->allocated = 0;
6046 if (preg->fastmap != NULL)
6047 free (preg->fastmap);
6048 preg->fastmap = NULL;
6049 preg->fastmap_accurate = 0;
6051 if (preg->translate != NULL)
6052 free (preg->translate);
6053 preg->translate = NULL;
6055 WEAK_ALIAS (__regfree, regfree)
6057 #endif /* not emacs */