* lisp/textmodes/table.el (table-insert): Don't use `symbol-name' on
[emacs.git] / src / regex.c
blob7443eff39770bb260c099e1c5b84eb98fae4a000
1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
5 Copyright (C) 1993-2012 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
10 any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
20 USA. */
22 /* TODO:
23 - structure the opcode space into opcode+flag.
24 - merge with glibc's regex.[ch].
25 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
26 need to modify the compiled regexp so that re_match can be reentrant.
27 - get rid of on_failure_jump_smart by doing the optimization in re_comp
28 rather than at run-time, so that re_match can be reentrant.
31 /* AIX requires this to be the first thing in the file. */
32 #if defined _AIX && !defined REGEX_MALLOC
33 #pragma alloca
34 #endif
36 /* Ignore some GCC warnings for now. This section should go away
37 once the Emacs and Gnulib regex code is merged. */
38 #if (__GNUC__ == 4 && 5 <= __GNUC_MINOR__) || 4 < __GNUC__
39 # pragma GCC diagnostic ignored "-Wstrict-overflow"
40 # ifndef emacs
41 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
42 # pragma GCC diagnostic ignored "-Wunused-function"
43 # pragma GCC diagnostic ignored "-Wunused-macros"
44 # pragma GCC diagnostic ignored "-Wunused-result"
45 # pragma GCC diagnostic ignored "-Wunused-variable"
46 # endif
47 #endif
49 #include <config.h>
51 #include <stddef.h>
53 #ifdef emacs
54 /* We need this for `regex.h', and perhaps for the Emacs include files. */
55 # include <sys/types.h>
56 #endif
58 /* Whether to use ISO C Amendment 1 wide char functions.
59 Those should not be used for Emacs since it uses its own. */
60 #if defined _LIBC
61 #define WIDE_CHAR_SUPPORT 1
62 #else
63 #define WIDE_CHAR_SUPPORT \
64 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
65 #endif
67 /* For platform which support the ISO C amendment 1 functionality we
68 support user defined character classes. */
69 #if WIDE_CHAR_SUPPORT
70 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
71 # include <wchar.h>
72 # include <wctype.h>
73 #endif
75 #ifdef _LIBC
76 /* We have to keep the namespace clean. */
77 # define regfree(preg) __regfree (preg)
78 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
79 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
80 # define regerror(err_code, preg, errbuf, errbuf_size) \
81 __regerror (err_code, preg, errbuf, errbuf_size)
82 # define re_set_registers(bu, re, nu, st, en) \
83 __re_set_registers (bu, re, nu, st, en)
84 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
85 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
86 # define re_match(bufp, string, size, pos, regs) \
87 __re_match (bufp, string, size, pos, regs)
88 # define re_search(bufp, string, size, startpos, range, regs) \
89 __re_search (bufp, string, size, startpos, range, regs)
90 # define re_compile_pattern(pattern, length, bufp) \
91 __re_compile_pattern (pattern, length, bufp)
92 # define re_set_syntax(syntax) __re_set_syntax (syntax)
93 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
94 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
95 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
97 /* Make sure we call libc's function even if the user overrides them. */
98 # define btowc __btowc
99 # define iswctype __iswctype
100 # define wctype __wctype
102 # define WEAK_ALIAS(a,b) weak_alias (a, b)
104 /* We are also using some library internals. */
105 # include <locale/localeinfo.h>
106 # include <locale/elem-hash.h>
107 # include <langinfo.h>
108 #else
109 # define WEAK_ALIAS(a,b)
110 #endif
112 /* This is for other GNU distributions with internationalized messages. */
113 #if HAVE_LIBINTL_H || defined _LIBC
114 # include <libintl.h>
115 #else
116 # define gettext(msgid) (msgid)
117 #endif
119 #ifndef gettext_noop
120 /* This define is so xgettext can find the internationalizable
121 strings. */
122 # define gettext_noop(String) String
123 #endif
125 /* The `emacs' switch turns on certain matching commands
126 that make sense only in Emacs. */
127 #ifdef emacs
129 # include "lisp.h"
130 # include "character.h"
131 # include "buffer.h"
133 /* Make syntax table lookup grant data in gl_state. */
134 # define SYNTAX_ENTRY_VIA_PROPERTY
136 # include "syntax.h"
137 # include "category.h"
139 # ifdef malloc
140 # undef malloc
141 # endif
142 # define malloc xmalloc
143 # ifdef realloc
144 # undef realloc
145 # endif
146 # define realloc xrealloc
147 # ifdef free
148 # undef free
149 # endif
150 # define free xfree
152 /* Converts the pointer to the char to BEG-based offset from the start. */
153 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
154 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
156 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
157 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
158 # define RE_STRING_CHAR(p, multibyte) \
159 (multibyte ? (STRING_CHAR (p)) : (*(p)))
160 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
161 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
163 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
165 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
167 /* Set C a (possibly converted to multibyte) character before P. P
168 points into a string which is the virtual concatenation of STR1
169 (which ends at END1) or STR2 (which ends at END2). */
170 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
171 do { \
172 if (target_multibyte) \
174 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
175 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
176 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
177 c = STRING_CHAR (dtemp); \
179 else \
181 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
182 (c) = RE_CHAR_TO_MULTIBYTE (c); \
184 } while (0)
186 /* Set C a (possibly converted to multibyte) character at P, and set
187 LEN to the byte length of that character. */
188 # define GET_CHAR_AFTER(c, p, len) \
189 do { \
190 if (target_multibyte) \
191 (c) = STRING_CHAR_AND_LENGTH (p, len); \
192 else \
194 (c) = *p; \
195 len = 1; \
196 (c) = RE_CHAR_TO_MULTIBYTE (c); \
198 } while (0)
200 #else /* not emacs */
202 /* If we are not linking with Emacs proper,
203 we can't use the relocating allocator
204 even if config.h says that we can. */
205 # undef REL_ALLOC
207 # include <unistd.h>
209 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
211 static void *
212 xmalloc (size_t size)
214 void *val = malloc (size);
215 if (!val && size)
217 write (2, "virtual memory exhausted\n", 25);
218 exit (1);
220 return val;
223 static void *
224 xrealloc (void *block, size_t size)
226 void *val;
227 /* We must call malloc explicitly when BLOCK is 0, since some
228 reallocs don't do this. */
229 if (! block)
230 val = malloc (size);
231 else
232 val = realloc (block, size);
233 if (!val && size)
235 write (2, "virtual memory exhausted\n", 25);
236 exit (1);
238 return val;
241 # ifdef malloc
242 # undef malloc
243 # endif
244 # define malloc xmalloc
245 # ifdef realloc
246 # undef realloc
247 # endif
248 # define realloc xrealloc
250 # include <stdbool.h>
251 # include <string.h>
253 /* Define the syntax stuff for \<, \>, etc. */
255 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
256 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
258 /* Dummy macros for non-Emacs environments. */
259 # define CHAR_CHARSET(c) 0
260 # define CHARSET_LEADING_CODE_BASE(c) 0
261 # define MAX_MULTIBYTE_LENGTH 1
262 # define RE_MULTIBYTE_P(x) 0
263 # define RE_TARGET_MULTIBYTE_P(x) 0
264 # define WORD_BOUNDARY_P(c1, c2) (0)
265 # define CHAR_HEAD_P(p) (1)
266 # define SINGLE_BYTE_CHAR_P(c) (1)
267 # define SAME_CHARSET_P(c1, c2) (1)
268 # define BYTES_BY_CHAR_HEAD(p) (1)
269 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
270 # define STRING_CHAR(p) (*(p))
271 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
272 # define CHAR_STRING(c, s) (*(s) = (c), 1)
273 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
274 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
275 # define RE_CHAR_TO_MULTIBYTE(c) (c)
276 # define RE_CHAR_TO_UNIBYTE(c) (c)
277 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
278 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
279 # define GET_CHAR_AFTER(c, p, len) \
280 (c = *p, len = 1)
281 # define MAKE_CHAR(charset, c1, c2) (c1)
282 # define BYTE8_TO_CHAR(c) (c)
283 # define CHAR_BYTE8_P(c) (0)
284 # define CHAR_LEADING_CODE(c) (c)
286 #endif /* not emacs */
288 #ifndef RE_TRANSLATE
289 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
290 # define RE_TRANSLATE_P(TBL) (TBL)
291 #endif
293 /* Get the interface, including the syntax bits. */
294 #include "regex.h"
296 /* isalpha etc. are used for the character classes. */
297 #include <ctype.h>
299 #ifdef emacs
301 /* 1 if C is an ASCII character. */
302 # define IS_REAL_ASCII(c) ((c) < 0200)
304 /* 1 if C is a unibyte character. */
305 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
307 /* The Emacs definitions should not be directly affected by locales. */
309 /* In Emacs, these are only used for single-byte characters. */
310 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
311 # define ISCNTRL(c) ((c) < ' ')
312 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
313 || ((c) >= 'a' && (c) <= 'f') \
314 || ((c) >= 'A' && (c) <= 'F'))
316 /* This is only used for single-byte characters. */
317 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
319 /* The rest must handle multibyte characters. */
321 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
322 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
323 : 1)
325 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
326 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
327 : 1)
329 # define ISALNUM(c) (IS_REAL_ASCII (c) \
330 ? (((c) >= 'a' && (c) <= 'z') \
331 || ((c) >= 'A' && (c) <= 'Z') \
332 || ((c) >= '0' && (c) <= '9')) \
333 : SYNTAX (c) == Sword)
335 # define ISALPHA(c) (IS_REAL_ASCII (c) \
336 ? (((c) >= 'a' && (c) <= 'z') \
337 || ((c) >= 'A' && (c) <= 'Z')) \
338 : SYNTAX (c) == Sword)
340 # define ISLOWER(c) lowercasep (c)
342 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
343 ? ((c) > ' ' && (c) < 0177 \
344 && !(((c) >= 'a' && (c) <= 'z') \
345 || ((c) >= 'A' && (c) <= 'Z') \
346 || ((c) >= '0' && (c) <= '9'))) \
347 : SYNTAX (c) != Sword)
349 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
351 # define ISUPPER(c) uppercasep (c)
353 # define ISWORD(c) (SYNTAX (c) == Sword)
355 #else /* not emacs */
357 /* 1 if C is an ASCII character. */
358 # define IS_REAL_ASCII(c) ((c) < 0200)
360 /* This distinction is not meaningful, except in Emacs. */
361 # define ISUNIBYTE(c) 1
363 # ifdef isblank
364 # define ISBLANK(c) isblank (c)
365 # else
366 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
367 # endif
368 # ifdef isgraph
369 # define ISGRAPH(c) isgraph (c)
370 # else
371 # define ISGRAPH(c) (isprint (c) && !isspace (c))
372 # endif
374 /* Solaris defines ISPRINT so we must undefine it first. */
375 # undef ISPRINT
376 # define ISPRINT(c) isprint (c)
377 # define ISDIGIT(c) isdigit (c)
378 # define ISALNUM(c) isalnum (c)
379 # define ISALPHA(c) isalpha (c)
380 # define ISCNTRL(c) iscntrl (c)
381 # define ISLOWER(c) islower (c)
382 # define ISPUNCT(c) ispunct (c)
383 # define ISSPACE(c) isspace (c)
384 # define ISUPPER(c) isupper (c)
385 # define ISXDIGIT(c) isxdigit (c)
387 # define ISWORD(c) ISALPHA (c)
389 # ifdef _tolower
390 # define TOLOWER(c) _tolower (c)
391 # else
392 # define TOLOWER(c) tolower (c)
393 # endif
395 /* How many characters in the character set. */
396 # define CHAR_SET_SIZE 256
398 # ifdef SYNTAX_TABLE
400 extern char *re_syntax_table;
402 # else /* not SYNTAX_TABLE */
404 static char re_syntax_table[CHAR_SET_SIZE];
406 static void
407 init_syntax_once (void)
409 register int c;
410 static int done = 0;
412 if (done)
413 return;
415 memset (re_syntax_table, 0, sizeof re_syntax_table);
417 for (c = 0; c < CHAR_SET_SIZE; ++c)
418 if (ISALNUM (c))
419 re_syntax_table[c] = Sword;
421 re_syntax_table['_'] = Ssymbol;
423 done = 1;
426 # endif /* not SYNTAX_TABLE */
428 # define SYNTAX(c) re_syntax_table[(c)]
430 #endif /* not emacs */
432 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
434 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
435 use `alloca' instead of `malloc'. This is because using malloc in
436 re_search* or re_match* could cause memory leaks when C-g is used in
437 Emacs; also, malloc is slower and causes storage fragmentation. On
438 the other hand, malloc is more portable, and easier to debug.
440 Because we sometimes use alloca, some routines have to be macros,
441 not functions -- `alloca'-allocated space disappears at the end of the
442 function it is called in. */
444 #ifdef REGEX_MALLOC
446 # define REGEX_ALLOCATE malloc
447 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
448 # define REGEX_FREE free
450 #else /* not REGEX_MALLOC */
452 /* Emacs already defines alloca, sometimes. */
453 # ifndef alloca
455 /* Make alloca work the best possible way. */
456 # ifdef __GNUC__
457 # define alloca __builtin_alloca
458 # else /* not __GNUC__ */
459 # ifdef HAVE_ALLOCA_H
460 # include <alloca.h>
461 # endif /* HAVE_ALLOCA_H */
462 # endif /* not __GNUC__ */
464 # endif /* not alloca */
466 # define REGEX_ALLOCATE alloca
468 /* Assumes a `char *destination' variable. */
469 # define REGEX_REALLOCATE(source, osize, nsize) \
470 (destination = (char *) alloca (nsize), \
471 memcpy (destination, source, osize))
473 /* No need to do anything to free, after alloca. */
474 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
476 #endif /* not REGEX_MALLOC */
478 /* Define how to allocate the failure stack. */
480 #if defined REL_ALLOC && defined REGEX_MALLOC
482 # define REGEX_ALLOCATE_STACK(size) \
483 r_alloc (&failure_stack_ptr, (size))
484 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
485 r_re_alloc (&failure_stack_ptr, (nsize))
486 # define REGEX_FREE_STACK(ptr) \
487 r_alloc_free (&failure_stack_ptr)
489 #else /* not using relocating allocator */
491 # ifdef REGEX_MALLOC
493 # define REGEX_ALLOCATE_STACK malloc
494 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
495 # define REGEX_FREE_STACK free
497 # else /* not REGEX_MALLOC */
499 # define REGEX_ALLOCATE_STACK alloca
501 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
502 REGEX_REALLOCATE (source, osize, nsize)
503 /* No need to explicitly free anything. */
504 # define REGEX_FREE_STACK(arg) ((void)0)
506 # endif /* not REGEX_MALLOC */
507 #endif /* not using relocating allocator */
510 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
511 `string1' or just past its end. This works if PTR is NULL, which is
512 a good thing. */
513 #define FIRST_STRING_P(ptr) \
514 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
516 /* (Re)Allocate N items of type T using malloc, or fail. */
517 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
518 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
519 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
521 #define BYTEWIDTH 8 /* In bits. */
523 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
525 #undef MAX
526 #undef MIN
527 #define MAX(a, b) ((a) > (b) ? (a) : (b))
528 #define MIN(a, b) ((a) < (b) ? (a) : (b))
530 /* Type of source-pattern and string chars. */
531 #ifdef _MSC_VER
532 typedef unsigned char re_char;
533 #else
534 typedef const unsigned char re_char;
535 #endif
537 typedef char boolean;
539 static regoff_t re_match_2_internal (struct re_pattern_buffer *bufp,
540 re_char *string1, size_t size1,
541 re_char *string2, size_t size2,
542 ssize_t pos,
543 struct re_registers *regs,
544 ssize_t stop);
546 /* These are the command codes that appear in compiled regular
547 expressions. Some opcodes are followed by argument bytes. A
548 command code can specify any interpretation whatsoever for its
549 arguments. Zero bytes may appear in the compiled regular expression. */
551 typedef enum
553 no_op = 0,
555 /* Succeed right away--no more backtracking. */
556 succeed,
558 /* Followed by one byte giving n, then by n literal bytes. */
559 exactn,
561 /* Matches any (more or less) character. */
562 anychar,
564 /* Matches any one char belonging to specified set. First
565 following byte is number of bitmap bytes. Then come bytes
566 for a bitmap saying which chars are in. Bits in each byte
567 are ordered low-bit-first. A character is in the set if its
568 bit is 1. A character too large to have a bit in the map is
569 automatically not in the set.
571 If the length byte has the 0x80 bit set, then that stuff
572 is followed by a range table:
573 2 bytes of flags for character sets (low 8 bits, high 8 bits)
574 See RANGE_TABLE_WORK_BITS below.
575 2 bytes, the number of pairs that follow (upto 32767)
576 pairs, each 2 multibyte characters,
577 each multibyte character represented as 3 bytes. */
578 charset,
580 /* Same parameters as charset, but match any character that is
581 not one of those specified. */
582 charset_not,
584 /* Start remembering the text that is matched, for storing in a
585 register. Followed by one byte with the register number, in
586 the range 0 to one less than the pattern buffer's re_nsub
587 field. */
588 start_memory,
590 /* Stop remembering the text that is matched and store it in a
591 memory register. Followed by one byte with the register
592 number, in the range 0 to one less than `re_nsub' in the
593 pattern buffer. */
594 stop_memory,
596 /* Match a duplicate of something remembered. Followed by one
597 byte containing the register number. */
598 duplicate,
600 /* Fail unless at beginning of line. */
601 begline,
603 /* Fail unless at end of line. */
604 endline,
606 /* Succeeds if at beginning of buffer (if emacs) or at beginning
607 of string to be matched (if not). */
608 begbuf,
610 /* Analogously, for end of buffer/string. */
611 endbuf,
613 /* Followed by two byte relative address to which to jump. */
614 jump,
616 /* Followed by two-byte relative address of place to resume at
617 in case of failure. */
618 on_failure_jump,
620 /* Like on_failure_jump, but pushes a placeholder instead of the
621 current string position when executed. */
622 on_failure_keep_string_jump,
624 /* Just like `on_failure_jump', except that it checks that we
625 don't get stuck in an infinite loop (matching an empty string
626 indefinitely). */
627 on_failure_jump_loop,
629 /* Just like `on_failure_jump_loop', except that it checks for
630 a different kind of loop (the kind that shows up with non-greedy
631 operators). This operation has to be immediately preceded
632 by a `no_op'. */
633 on_failure_jump_nastyloop,
635 /* A smart `on_failure_jump' used for greedy * and + operators.
636 It analyzes the loop before which it is put and if the
637 loop does not require backtracking, it changes itself to
638 `on_failure_keep_string_jump' and short-circuits the loop,
639 else it just defaults to changing itself into `on_failure_jump'.
640 It assumes that it is pointing to just past a `jump'. */
641 on_failure_jump_smart,
643 /* Followed by two-byte relative address and two-byte number n.
644 After matching N times, jump to the address upon failure.
645 Does not work if N starts at 0: use on_failure_jump_loop
646 instead. */
647 succeed_n,
649 /* Followed by two-byte relative address, and two-byte number n.
650 Jump to the address N times, then fail. */
651 jump_n,
653 /* Set the following two-byte relative address to the
654 subsequent two-byte number. The address *includes* the two
655 bytes of number. */
656 set_number_at,
658 wordbeg, /* Succeeds if at word beginning. */
659 wordend, /* Succeeds if at word end. */
661 wordbound, /* Succeeds if at a word boundary. */
662 notwordbound, /* Succeeds if not at a word boundary. */
664 symbeg, /* Succeeds if at symbol beginning. */
665 symend, /* Succeeds if at symbol end. */
667 /* Matches any character whose syntax is specified. Followed by
668 a byte which contains a syntax code, e.g., Sword. */
669 syntaxspec,
671 /* Matches any character whose syntax is not that specified. */
672 notsyntaxspec
674 #ifdef emacs
675 ,before_dot, /* Succeeds if before point. */
676 at_dot, /* Succeeds if at point. */
677 after_dot, /* Succeeds if after point. */
679 /* Matches any character whose category-set contains the specified
680 category. The operator is followed by a byte which contains a
681 category code (mnemonic ASCII character). */
682 categoryspec,
684 /* Matches any character whose category-set does not contain the
685 specified category. The operator is followed by a byte which
686 contains the category code (mnemonic ASCII character). */
687 notcategoryspec
688 #endif /* emacs */
689 } re_opcode_t;
691 /* Common operations on the compiled pattern. */
693 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
695 #define STORE_NUMBER(destination, number) \
696 do { \
697 (destination)[0] = (number) & 0377; \
698 (destination)[1] = (number) >> 8; \
699 } while (0)
701 /* Same as STORE_NUMBER, except increment DESTINATION to
702 the byte after where the number is stored. Therefore, DESTINATION
703 must be an lvalue. */
705 #define STORE_NUMBER_AND_INCR(destination, number) \
706 do { \
707 STORE_NUMBER (destination, number); \
708 (destination) += 2; \
709 } while (0)
711 /* Put into DESTINATION a number stored in two contiguous bytes starting
712 at SOURCE. */
714 #define EXTRACT_NUMBER(destination, source) \
715 do { \
716 (destination) = *(source) & 0377; \
717 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
718 } while (0)
720 #ifdef DEBUG
721 static void
722 extract_number (int *dest, re_char *source)
724 int temp = SIGN_EXTEND_CHAR (*(source + 1));
725 *dest = *source & 0377;
726 *dest += temp << 8;
729 # ifndef EXTRACT_MACROS /* To debug the macros. */
730 # undef EXTRACT_NUMBER
731 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
732 # endif /* not EXTRACT_MACROS */
734 #endif /* DEBUG */
736 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
737 SOURCE must be an lvalue. */
739 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
740 do { \
741 EXTRACT_NUMBER (destination, source); \
742 (source) += 2; \
743 } while (0)
745 #ifdef DEBUG
746 static void
747 extract_number_and_incr (int *destination, re_char **source)
749 extract_number (destination, *source);
750 *source += 2;
753 # ifndef EXTRACT_MACROS
754 # undef EXTRACT_NUMBER_AND_INCR
755 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
756 extract_number_and_incr (&dest, &src)
757 # endif /* not EXTRACT_MACROS */
759 #endif /* DEBUG */
761 /* Store a multibyte character in three contiguous bytes starting
762 DESTINATION, and increment DESTINATION to the byte after where the
763 character is stored. Therefore, DESTINATION must be an lvalue. */
765 #define STORE_CHARACTER_AND_INCR(destination, character) \
766 do { \
767 (destination)[0] = (character) & 0377; \
768 (destination)[1] = ((character) >> 8) & 0377; \
769 (destination)[2] = (character) >> 16; \
770 (destination) += 3; \
771 } while (0)
773 /* Put into DESTINATION a character stored in three contiguous bytes
774 starting at SOURCE. */
776 #define EXTRACT_CHARACTER(destination, source) \
777 do { \
778 (destination) = ((source)[0] \
779 | ((source)[1] << 8) \
780 | ((source)[2] << 16)); \
781 } while (0)
784 /* Macros for charset. */
786 /* Size of bitmap of charset P in bytes. P is a start of charset,
787 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
788 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
790 /* Nonzero if charset P has range table. */
791 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
793 /* Return the address of range table of charset P. But not the start
794 of table itself, but the before where the number of ranges is
795 stored. `2 +' means to skip re_opcode_t and size of bitmap,
796 and the 2 bytes of flags at the start of the range table. */
797 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
799 /* Extract the bit flags that start a range table. */
800 #define CHARSET_RANGE_TABLE_BITS(p) \
801 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
802 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
804 /* Return the address of end of RANGE_TABLE. COUNT is number of
805 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
806 is start of range and end of range. `* 3' is size of each start
807 and end. */
808 #define CHARSET_RANGE_TABLE_END(range_table, count) \
809 ((range_table) + (count) * 2 * 3)
811 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
812 COUNT is number of ranges in RANGE_TABLE. */
813 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
814 do \
816 re_wchar_t range_start, range_end; \
817 re_char *rtp; \
818 re_char *range_table_end \
819 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
821 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
823 EXTRACT_CHARACTER (range_start, rtp); \
824 EXTRACT_CHARACTER (range_end, rtp + 3); \
826 if (range_start <= (c) && (c) <= range_end) \
828 (not) = !(not); \
829 break; \
833 while (0)
835 /* Test if C is in range table of CHARSET. The flag NOT is negated if
836 C is listed in it. */
837 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
838 do \
840 /* Number of ranges in range table. */ \
841 int count; \
842 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
844 EXTRACT_NUMBER_AND_INCR (count, range_table); \
845 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
847 while (0)
849 /* If DEBUG is defined, Regex prints many voluminous messages about what
850 it is doing (if the variable `debug' is nonzero). If linked with the
851 main program in `iregex.c', you can enter patterns and strings
852 interactively. And if linked with the main program in `main.c' and
853 the other test files, you can run the already-written tests. */
855 #ifdef DEBUG
857 /* We use standard I/O for debugging. */
858 # include <stdio.h>
860 /* It is useful to test things that ``must'' be true when debugging. */
861 # include <assert.h>
863 static int debug = -100000;
865 # define DEBUG_STATEMENT(e) e
866 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
867 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
868 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
869 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
870 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
871 if (debug > 0) print_partial_compiled_pattern (s, e)
872 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
873 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
876 /* Print the fastmap in human-readable form. */
878 void
879 print_fastmap (fastmap)
880 char *fastmap;
882 unsigned was_a_range = 0;
883 unsigned i = 0;
885 while (i < (1 << BYTEWIDTH))
887 if (fastmap[i++])
889 was_a_range = 0;
890 putchar (i - 1);
891 while (i < (1 << BYTEWIDTH) && fastmap[i])
893 was_a_range = 1;
894 i++;
896 if (was_a_range)
898 printf ("-");
899 putchar (i - 1);
903 putchar ('\n');
907 /* Print a compiled pattern string in human-readable form, starting at
908 the START pointer into it and ending just before the pointer END. */
910 void
911 print_partial_compiled_pattern (start, end)
912 re_char *start;
913 re_char *end;
915 int mcnt, mcnt2;
916 re_char *p = start;
917 re_char *pend = end;
919 if (start == NULL)
921 fprintf (stderr, "(null)\n");
922 return;
925 /* Loop over pattern commands. */
926 while (p < pend)
928 fprintf (stderr, "%d:\t", p - start);
930 switch ((re_opcode_t) *p++)
932 case no_op:
933 fprintf (stderr, "/no_op");
934 break;
936 case succeed:
937 fprintf (stderr, "/succeed");
938 break;
940 case exactn:
941 mcnt = *p++;
942 fprintf (stderr, "/exactn/%d", mcnt);
945 fprintf (stderr, "/%c", *p++);
947 while (--mcnt);
948 break;
950 case start_memory:
951 fprintf (stderr, "/start_memory/%d", *p++);
952 break;
954 case stop_memory:
955 fprintf (stderr, "/stop_memory/%d", *p++);
956 break;
958 case duplicate:
959 fprintf (stderr, "/duplicate/%d", *p++);
960 break;
962 case anychar:
963 fprintf (stderr, "/anychar");
964 break;
966 case charset:
967 case charset_not:
969 register int c, last = -100;
970 register int in_range = 0;
971 int length = CHARSET_BITMAP_SIZE (p - 1);
972 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
974 fprintf (stderr, "/charset [%s",
975 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
977 if (p + *p >= pend)
978 fprintf (stderr, " !extends past end of pattern! ");
980 for (c = 0; c < 256; c++)
981 if (c / 8 < length
982 && (p[1 + (c/8)] & (1 << (c % 8))))
984 /* Are we starting a range? */
985 if (last + 1 == c && ! in_range)
987 fprintf (stderr, "-");
988 in_range = 1;
990 /* Have we broken a range? */
991 else if (last + 1 != c && in_range)
993 fprintf (stderr, "%c", last);
994 in_range = 0;
997 if (! in_range)
998 fprintf (stderr, "%c", c);
1000 last = c;
1003 if (in_range)
1004 fprintf (stderr, "%c", last);
1006 fprintf (stderr, "]");
1008 p += 1 + length;
1010 if (has_range_table)
1012 int count;
1013 fprintf (stderr, "has-range-table");
1015 /* ??? Should print the range table; for now, just skip it. */
1016 p += 2; /* skip range table bits */
1017 EXTRACT_NUMBER_AND_INCR (count, p);
1018 p = CHARSET_RANGE_TABLE_END (p, count);
1021 break;
1023 case begline:
1024 fprintf (stderr, "/begline");
1025 break;
1027 case endline:
1028 fprintf (stderr, "/endline");
1029 break;
1031 case on_failure_jump:
1032 extract_number_and_incr (&mcnt, &p);
1033 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1034 break;
1036 case on_failure_keep_string_jump:
1037 extract_number_and_incr (&mcnt, &p);
1038 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1039 break;
1041 case on_failure_jump_nastyloop:
1042 extract_number_and_incr (&mcnt, &p);
1043 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1044 break;
1046 case on_failure_jump_loop:
1047 extract_number_and_incr (&mcnt, &p);
1048 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1049 break;
1051 case on_failure_jump_smart:
1052 extract_number_and_incr (&mcnt, &p);
1053 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1054 break;
1056 case jump:
1057 extract_number_and_incr (&mcnt, &p);
1058 fprintf (stderr, "/jump to %d", p + mcnt - start);
1059 break;
1061 case succeed_n:
1062 extract_number_and_incr (&mcnt, &p);
1063 extract_number_and_incr (&mcnt2, &p);
1064 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1065 break;
1067 case jump_n:
1068 extract_number_and_incr (&mcnt, &p);
1069 extract_number_and_incr (&mcnt2, &p);
1070 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1071 break;
1073 case set_number_at:
1074 extract_number_and_incr (&mcnt, &p);
1075 extract_number_and_incr (&mcnt2, &p);
1076 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1077 break;
1079 case wordbound:
1080 fprintf (stderr, "/wordbound");
1081 break;
1083 case notwordbound:
1084 fprintf (stderr, "/notwordbound");
1085 break;
1087 case wordbeg:
1088 fprintf (stderr, "/wordbeg");
1089 break;
1091 case wordend:
1092 fprintf (stderr, "/wordend");
1093 break;
1095 case symbeg:
1096 fprintf (stderr, "/symbeg");
1097 break;
1099 case symend:
1100 fprintf (stderr, "/symend");
1101 break;
1103 case syntaxspec:
1104 fprintf (stderr, "/syntaxspec");
1105 mcnt = *p++;
1106 fprintf (stderr, "/%d", mcnt);
1107 break;
1109 case notsyntaxspec:
1110 fprintf (stderr, "/notsyntaxspec");
1111 mcnt = *p++;
1112 fprintf (stderr, "/%d", mcnt);
1113 break;
1115 # ifdef emacs
1116 case before_dot:
1117 fprintf (stderr, "/before_dot");
1118 break;
1120 case at_dot:
1121 fprintf (stderr, "/at_dot");
1122 break;
1124 case after_dot:
1125 fprintf (stderr, "/after_dot");
1126 break;
1128 case categoryspec:
1129 fprintf (stderr, "/categoryspec");
1130 mcnt = *p++;
1131 fprintf (stderr, "/%d", mcnt);
1132 break;
1134 case notcategoryspec:
1135 fprintf (stderr, "/notcategoryspec");
1136 mcnt = *p++;
1137 fprintf (stderr, "/%d", mcnt);
1138 break;
1139 # endif /* emacs */
1141 case begbuf:
1142 fprintf (stderr, "/begbuf");
1143 break;
1145 case endbuf:
1146 fprintf (stderr, "/endbuf");
1147 break;
1149 default:
1150 fprintf (stderr, "?%d", *(p-1));
1153 fprintf (stderr, "\n");
1156 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1160 void
1161 print_compiled_pattern (bufp)
1162 struct re_pattern_buffer *bufp;
1164 re_char *buffer = bufp->buffer;
1166 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1167 printf ("%ld bytes used/%ld bytes allocated.\n",
1168 bufp->used, bufp->allocated);
1170 if (bufp->fastmap_accurate && bufp->fastmap)
1172 printf ("fastmap: ");
1173 print_fastmap (bufp->fastmap);
1176 printf ("re_nsub: %d\t", bufp->re_nsub);
1177 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1178 printf ("can_be_null: %d\t", bufp->can_be_null);
1179 printf ("no_sub: %d\t", bufp->no_sub);
1180 printf ("not_bol: %d\t", bufp->not_bol);
1181 printf ("not_eol: %d\t", bufp->not_eol);
1182 printf ("syntax: %lx\n", bufp->syntax);
1183 fflush (stdout);
1184 /* Perhaps we should print the translate table? */
1188 void
1189 print_double_string (where, string1, size1, string2, size2)
1190 re_char *where;
1191 re_char *string1;
1192 re_char *string2;
1193 ssize_t size1;
1194 ssize_t size2;
1196 ssize_t this_char;
1198 if (where == NULL)
1199 printf ("(null)");
1200 else
1202 if (FIRST_STRING_P (where))
1204 for (this_char = where - string1; this_char < size1; this_char++)
1205 putchar (string1[this_char]);
1207 where = string2;
1210 for (this_char = where - string2; this_char < size2; this_char++)
1211 putchar (string2[this_char]);
1215 #else /* not DEBUG */
1217 # undef assert
1218 # define assert(e)
1220 # define DEBUG_STATEMENT(e)
1221 # define DEBUG_PRINT1(x)
1222 # define DEBUG_PRINT2(x1, x2)
1223 # define DEBUG_PRINT3(x1, x2, x3)
1224 # define DEBUG_PRINT4(x1, x2, x3, x4)
1225 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1226 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1228 #endif /* not DEBUG */
1230 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1231 #ifdef lint
1232 # define IF_LINT(Code) Code
1233 #else
1234 # define IF_LINT(Code) /* empty */
1235 #endif
1237 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1238 also be assigned to arbitrarily: each pattern buffer stores its own
1239 syntax, so it can be changed between regex compilations. */
1240 /* This has no initializer because initialized variables in Emacs
1241 become read-only after dumping. */
1242 reg_syntax_t re_syntax_options;
1245 /* Specify the precise syntax of regexps for compilation. This provides
1246 for compatibility for various utilities which historically have
1247 different, incompatible syntaxes.
1249 The argument SYNTAX is a bit mask comprised of the various bits
1250 defined in regex.h. We return the old syntax. */
1252 reg_syntax_t
1253 re_set_syntax (reg_syntax_t syntax)
1255 reg_syntax_t ret = re_syntax_options;
1257 re_syntax_options = syntax;
1258 return ret;
1260 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1262 /* Regexp to use to replace spaces, or NULL meaning don't. */
1263 static re_char *whitespace_regexp;
1265 void
1266 re_set_whitespace_regexp (const char *regexp)
1268 whitespace_regexp = (re_char *) regexp;
1270 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1272 /* This table gives an error message for each of the error codes listed
1273 in regex.h. Obviously the order here has to be same as there.
1274 POSIX doesn't require that we do anything for REG_NOERROR,
1275 but why not be nice? */
1277 static const char *re_error_msgid[] =
1279 gettext_noop ("Success"), /* REG_NOERROR */
1280 gettext_noop ("No match"), /* REG_NOMATCH */
1281 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1282 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1283 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1284 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1285 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1286 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1287 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1288 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1289 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1290 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1291 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1292 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1293 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1294 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1295 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1296 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1299 /* Avoiding alloca during matching, to placate r_alloc. */
1301 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1302 searching and matching functions should not call alloca. On some
1303 systems, alloca is implemented in terms of malloc, and if we're
1304 using the relocating allocator routines, then malloc could cause a
1305 relocation, which might (if the strings being searched are in the
1306 ralloc heap) shift the data out from underneath the regexp
1307 routines.
1309 Here's another reason to avoid allocation: Emacs
1310 processes input from X in a signal handler; processing X input may
1311 call malloc; if input arrives while a matching routine is calling
1312 malloc, then we're scrod. But Emacs can't just block input while
1313 calling matching routines; then we don't notice interrupts when
1314 they come in. So, Emacs blocks input around all regexp calls
1315 except the matching calls, which it leaves unprotected, in the
1316 faith that they will not malloc. */
1318 /* Normally, this is fine. */
1319 #define MATCH_MAY_ALLOCATE
1321 /* The match routines may not allocate if (1) they would do it with malloc
1322 and (2) it's not safe for them to use malloc.
1323 Note that if REL_ALLOC is defined, matching would not use malloc for the
1324 failure stack, but we would still use it for the register vectors;
1325 so REL_ALLOC should not affect this. */
1326 #if defined REGEX_MALLOC && defined emacs
1327 # undef MATCH_MAY_ALLOCATE
1328 #endif
1331 /* Failure stack declarations and macros; both re_compile_fastmap and
1332 re_match_2 use a failure stack. These have to be macros because of
1333 REGEX_ALLOCATE_STACK. */
1336 /* Approximate number of failure points for which to initially allocate space
1337 when matching. If this number is exceeded, we allocate more
1338 space, so it is not a hard limit. */
1339 #ifndef INIT_FAILURE_ALLOC
1340 # define INIT_FAILURE_ALLOC 20
1341 #endif
1343 /* Roughly the maximum number of failure points on the stack. Would be
1344 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1345 This is a variable only so users of regex can assign to it; we never
1346 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1347 before using it, so it should probably be a byte-count instead. */
1348 # if defined MATCH_MAY_ALLOCATE
1349 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1350 whose default stack limit is 2mb. In order for a larger
1351 value to work reliably, you have to try to make it accord
1352 with the process stack limit. */
1353 size_t re_max_failures = 40000;
1354 # else
1355 size_t re_max_failures = 4000;
1356 # endif
1358 union fail_stack_elt
1360 re_char *pointer;
1361 /* This should be the biggest `int' that's no bigger than a pointer. */
1362 long integer;
1365 typedef union fail_stack_elt fail_stack_elt_t;
1367 typedef struct
1369 fail_stack_elt_t *stack;
1370 size_t size;
1371 size_t avail; /* Offset of next open position. */
1372 size_t frame; /* Offset of the cur constructed frame. */
1373 } fail_stack_type;
1375 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1378 /* Define macros to initialize and free the failure stack.
1379 Do `return -2' if the alloc fails. */
1381 #ifdef MATCH_MAY_ALLOCATE
1382 # define INIT_FAIL_STACK() \
1383 do { \
1384 fail_stack.stack = \
1385 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1386 * sizeof (fail_stack_elt_t)); \
1388 if (fail_stack.stack == NULL) \
1389 return -2; \
1391 fail_stack.size = INIT_FAILURE_ALLOC; \
1392 fail_stack.avail = 0; \
1393 fail_stack.frame = 0; \
1394 } while (0)
1395 #else
1396 # define INIT_FAIL_STACK() \
1397 do { \
1398 fail_stack.avail = 0; \
1399 fail_stack.frame = 0; \
1400 } while (0)
1402 # define RETALLOC_IF(addr, n, t) \
1403 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1404 #endif
1407 /* Double the size of FAIL_STACK, up to a limit
1408 which allows approximately `re_max_failures' items.
1410 Return 1 if succeeds, and 0 if either ran out of memory
1411 allocating space for it or it was already too large.
1413 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1415 /* Factor to increase the failure stack size by
1416 when we increase it.
1417 This used to be 2, but 2 was too wasteful
1418 because the old discarded stacks added up to as much space
1419 were as ultimate, maximum-size stack. */
1420 #define FAIL_STACK_GROWTH_FACTOR 4
1422 #define GROW_FAIL_STACK(fail_stack) \
1423 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1424 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1425 ? 0 \
1426 : ((fail_stack).stack \
1427 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1428 (fail_stack).size * sizeof (fail_stack_elt_t), \
1429 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1430 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1431 * FAIL_STACK_GROWTH_FACTOR))), \
1433 (fail_stack).stack == NULL \
1434 ? 0 \
1435 : ((fail_stack).size \
1436 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1437 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1438 * FAIL_STACK_GROWTH_FACTOR)) \
1439 / sizeof (fail_stack_elt_t)), \
1440 1)))
1443 /* Push a pointer value onto the failure stack.
1444 Assumes the variable `fail_stack'. Probably should only
1445 be called from within `PUSH_FAILURE_POINT'. */
1446 #define PUSH_FAILURE_POINTER(item) \
1447 fail_stack.stack[fail_stack.avail++].pointer = (item)
1449 /* This pushes an integer-valued item onto the failure stack.
1450 Assumes the variable `fail_stack'. Probably should only
1451 be called from within `PUSH_FAILURE_POINT'. */
1452 #define PUSH_FAILURE_INT(item) \
1453 fail_stack.stack[fail_stack.avail++].integer = (item)
1455 /* These POP... operations complement the PUSH... operations.
1456 All assume that `fail_stack' is nonempty. */
1457 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1458 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1460 /* Individual items aside from the registers. */
1461 #define NUM_NONREG_ITEMS 3
1463 /* Used to examine the stack (to detect infinite loops). */
1464 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1465 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1466 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1467 #define TOP_FAILURE_HANDLE() fail_stack.frame
1470 #define ENSURE_FAIL_STACK(space) \
1471 while (REMAINING_AVAIL_SLOTS <= space) { \
1472 if (!GROW_FAIL_STACK (fail_stack)) \
1473 return -2; \
1474 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1475 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1478 /* Push register NUM onto the stack. */
1479 #define PUSH_FAILURE_REG(num) \
1480 do { \
1481 char *destination; \
1482 ENSURE_FAIL_STACK(3); \
1483 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1484 num, regstart[num], regend[num]); \
1485 PUSH_FAILURE_POINTER (regstart[num]); \
1486 PUSH_FAILURE_POINTER (regend[num]); \
1487 PUSH_FAILURE_INT (num); \
1488 } while (0)
1490 /* Change the counter's value to VAL, but make sure that it will
1491 be reset when backtracking. */
1492 #define PUSH_NUMBER(ptr,val) \
1493 do { \
1494 char *destination; \
1495 int c; \
1496 ENSURE_FAIL_STACK(3); \
1497 EXTRACT_NUMBER (c, ptr); \
1498 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1499 PUSH_FAILURE_INT (c); \
1500 PUSH_FAILURE_POINTER (ptr); \
1501 PUSH_FAILURE_INT (-1); \
1502 STORE_NUMBER (ptr, val); \
1503 } while (0)
1505 /* Pop a saved register off the stack. */
1506 #define POP_FAILURE_REG_OR_COUNT() \
1507 do { \
1508 long pfreg = POP_FAILURE_INT (); \
1509 if (pfreg == -1) \
1511 /* It's a counter. */ \
1512 /* Here, we discard `const', making re_match non-reentrant. */ \
1513 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1514 pfreg = POP_FAILURE_INT (); \
1515 STORE_NUMBER (ptr, pfreg); \
1516 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, pfreg); \
1518 else \
1520 regend[pfreg] = POP_FAILURE_POINTER (); \
1521 regstart[pfreg] = POP_FAILURE_POINTER (); \
1522 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1523 pfreg, regstart[pfreg], regend[pfreg]); \
1525 } while (0)
1527 /* Check that we are not stuck in an infinite loop. */
1528 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1529 do { \
1530 ssize_t failure = TOP_FAILURE_HANDLE (); \
1531 /* Check for infinite matching loops */ \
1532 while (failure > 0 \
1533 && (FAILURE_STR (failure) == string_place \
1534 || FAILURE_STR (failure) == NULL)) \
1536 assert (FAILURE_PAT (failure) >= bufp->buffer \
1537 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1538 if (FAILURE_PAT (failure) == pat_cur) \
1540 cycle = 1; \
1541 break; \
1543 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1544 failure = NEXT_FAILURE_HANDLE(failure); \
1546 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1547 } while (0)
1549 /* Push the information about the state we will need
1550 if we ever fail back to it.
1552 Requires variables fail_stack, regstart, regend and
1553 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1554 declared.
1556 Does `return FAILURE_CODE' if runs out of memory. */
1558 #define PUSH_FAILURE_POINT(pattern, string_place) \
1559 do { \
1560 char *destination; \
1561 /* Must be int, so when we don't save any registers, the arithmetic \
1562 of 0 + -1 isn't done as unsigned. */ \
1564 DEBUG_STATEMENT (nfailure_points_pushed++); \
1565 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1566 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1567 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1569 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1571 DEBUG_PRINT1 ("\n"); \
1573 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1574 PUSH_FAILURE_INT (fail_stack.frame); \
1576 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1577 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1578 DEBUG_PRINT1 ("'\n"); \
1579 PUSH_FAILURE_POINTER (string_place); \
1581 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1582 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1583 PUSH_FAILURE_POINTER (pattern); \
1585 /* Close the frame by moving the frame pointer past it. */ \
1586 fail_stack.frame = fail_stack.avail; \
1587 } while (0)
1589 /* Estimate the size of data pushed by a typical failure stack entry.
1590 An estimate is all we need, because all we use this for
1591 is to choose a limit for how big to make the failure stack. */
1592 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1593 #define TYPICAL_FAILURE_SIZE 20
1595 /* How many items can still be added to the stack without overflowing it. */
1596 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1599 /* Pops what PUSH_FAIL_STACK pushes.
1601 We restore into the parameters, all of which should be lvalues:
1602 STR -- the saved data position.
1603 PAT -- the saved pattern position.
1604 REGSTART, REGEND -- arrays of string positions.
1606 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1607 `pend', `string1', `size1', `string2', and `size2'. */
1609 #define POP_FAILURE_POINT(str, pat) \
1610 do { \
1611 assert (!FAIL_STACK_EMPTY ()); \
1613 /* Remove failure points and point to how many regs pushed. */ \
1614 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1615 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1616 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1618 /* Pop the saved registers. */ \
1619 while (fail_stack.frame < fail_stack.avail) \
1620 POP_FAILURE_REG_OR_COUNT (); \
1622 pat = POP_FAILURE_POINTER (); \
1623 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1624 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1626 /* If the saved string location is NULL, it came from an \
1627 on_failure_keep_string_jump opcode, and we want to throw away the \
1628 saved NULL, thus retaining our current position in the string. */ \
1629 str = POP_FAILURE_POINTER (); \
1630 DEBUG_PRINT2 (" Popping string %p: `", str); \
1631 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1632 DEBUG_PRINT1 ("'\n"); \
1634 fail_stack.frame = POP_FAILURE_INT (); \
1635 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1637 assert (fail_stack.avail >= 0); \
1638 assert (fail_stack.frame <= fail_stack.avail); \
1640 DEBUG_STATEMENT (nfailure_points_popped++); \
1641 } while (0) /* POP_FAILURE_POINT */
1645 /* Registers are set to a sentinel when they haven't yet matched. */
1646 #define REG_UNSET(e) ((e) == NULL)
1648 /* Subroutine declarations and macros for regex_compile. */
1650 static reg_errcode_t regex_compile (re_char *pattern, size_t size,
1651 reg_syntax_t syntax,
1652 struct re_pattern_buffer *bufp);
1653 static void store_op1 (re_opcode_t op, unsigned char *loc, int arg);
1654 static void store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2);
1655 static void insert_op1 (re_opcode_t op, unsigned char *loc,
1656 int arg, unsigned char *end);
1657 static void insert_op2 (re_opcode_t op, unsigned char *loc,
1658 int arg1, int arg2, unsigned char *end);
1659 static boolean at_begline_loc_p (re_char *pattern, re_char *p,
1660 reg_syntax_t syntax);
1661 static boolean at_endline_loc_p (re_char *p, re_char *pend,
1662 reg_syntax_t syntax);
1663 static re_char *skip_one_char (re_char *p);
1664 static int analyse_first (re_char *p, re_char *pend,
1665 char *fastmap, const int multibyte);
1667 /* Fetch the next character in the uncompiled pattern, with no
1668 translation. */
1669 #define PATFETCH(c) \
1670 do { \
1671 int len; \
1672 if (p == pend) return REG_EEND; \
1673 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1674 p += len; \
1675 } while (0)
1678 /* If `translate' is non-null, return translate[D], else just D. We
1679 cast the subscript to translate because some data is declared as
1680 `char *', to avoid warnings when a string constant is passed. But
1681 when we use a character as a subscript we must make it unsigned. */
1682 #ifndef TRANSLATE
1683 # define TRANSLATE(d) \
1684 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1685 #endif
1688 /* Macros for outputting the compiled pattern into `buffer'. */
1690 /* If the buffer isn't allocated when it comes in, use this. */
1691 #define INIT_BUF_SIZE 32
1693 /* Make sure we have at least N more bytes of space in buffer. */
1694 #define GET_BUFFER_SPACE(n) \
1695 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1696 EXTEND_BUFFER ()
1698 /* Make sure we have one more byte of buffer space and then add C to it. */
1699 #define BUF_PUSH(c) \
1700 do { \
1701 GET_BUFFER_SPACE (1); \
1702 *b++ = (unsigned char) (c); \
1703 } while (0)
1706 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1707 #define BUF_PUSH_2(c1, c2) \
1708 do { \
1709 GET_BUFFER_SPACE (2); \
1710 *b++ = (unsigned char) (c1); \
1711 *b++ = (unsigned char) (c2); \
1712 } while (0)
1715 /* Store a jump with opcode OP at LOC to location TO. We store a
1716 relative address offset by the three bytes the jump itself occupies. */
1717 #define STORE_JUMP(op, loc, to) \
1718 store_op1 (op, loc, (to) - (loc) - 3)
1720 /* Likewise, for a two-argument jump. */
1721 #define STORE_JUMP2(op, loc, to, arg) \
1722 store_op2 (op, loc, (to) - (loc) - 3, arg)
1724 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1725 #define INSERT_JUMP(op, loc, to) \
1726 insert_op1 (op, loc, (to) - (loc) - 3, b)
1728 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1729 #define INSERT_JUMP2(op, loc, to, arg) \
1730 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1733 /* This is not an arbitrary limit: the arguments which represent offsets
1734 into the pattern are two bytes long. So if 2^15 bytes turns out to
1735 be too small, many things would have to change. */
1736 # define MAX_BUF_SIZE (1L << 15)
1738 /* Extend the buffer by twice its current size via realloc and
1739 reset the pointers that pointed into the old block to point to the
1740 correct places in the new one. If extending the buffer results in it
1741 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1742 #if __BOUNDED_POINTERS__
1743 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1744 # define MOVE_BUFFER_POINTER(P) \
1745 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1746 SET_HIGH_BOUND (P), \
1747 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1748 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1749 else \
1751 SET_HIGH_BOUND (b); \
1752 SET_HIGH_BOUND (begalt); \
1753 if (fixup_alt_jump) \
1754 SET_HIGH_BOUND (fixup_alt_jump); \
1755 if (laststart) \
1756 SET_HIGH_BOUND (laststart); \
1757 if (pending_exact) \
1758 SET_HIGH_BOUND (pending_exact); \
1760 #else
1761 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1762 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1763 #endif
1764 #define EXTEND_BUFFER() \
1765 do { \
1766 unsigned char *old_buffer = bufp->buffer; \
1767 if (bufp->allocated == MAX_BUF_SIZE) \
1768 return REG_ESIZE; \
1769 bufp->allocated <<= 1; \
1770 if (bufp->allocated > MAX_BUF_SIZE) \
1771 bufp->allocated = MAX_BUF_SIZE; \
1772 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1773 if (bufp->buffer == NULL) \
1774 return REG_ESPACE; \
1775 /* If the buffer moved, move all the pointers into it. */ \
1776 if (old_buffer != bufp->buffer) \
1778 unsigned char *new_buffer = bufp->buffer; \
1779 MOVE_BUFFER_POINTER (b); \
1780 MOVE_BUFFER_POINTER (begalt); \
1781 if (fixup_alt_jump) \
1782 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1783 if (laststart) \
1784 MOVE_BUFFER_POINTER (laststart); \
1785 if (pending_exact) \
1786 MOVE_BUFFER_POINTER (pending_exact); \
1788 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1789 } while (0)
1792 /* Since we have one byte reserved for the register number argument to
1793 {start,stop}_memory, the maximum number of groups we can report
1794 things about is what fits in that byte. */
1795 #define MAX_REGNUM 255
1797 /* But patterns can have more than `MAX_REGNUM' registers. We just
1798 ignore the excess. */
1799 typedef int regnum_t;
1802 /* Macros for the compile stack. */
1804 /* Since offsets can go either forwards or backwards, this type needs to
1805 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1806 /* int may be not enough when sizeof(int) == 2. */
1807 typedef long pattern_offset_t;
1809 typedef struct
1811 pattern_offset_t begalt_offset;
1812 pattern_offset_t fixup_alt_jump;
1813 pattern_offset_t laststart_offset;
1814 regnum_t regnum;
1815 } compile_stack_elt_t;
1818 typedef struct
1820 compile_stack_elt_t *stack;
1821 size_t size;
1822 size_t avail; /* Offset of next open position. */
1823 } compile_stack_type;
1826 #define INIT_COMPILE_STACK_SIZE 32
1828 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1829 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1831 /* The next available element. */
1832 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1834 /* Explicit quit checking is needed for Emacs, which uses polling to
1835 process input events. */
1836 #ifdef emacs
1837 # define IMMEDIATE_QUIT_CHECK \
1838 do { \
1839 if (immediate_quit) QUIT; \
1840 } while (0)
1841 #else
1842 # define IMMEDIATE_QUIT_CHECK ((void)0)
1843 #endif
1845 /* Structure to manage work area for range table. */
1846 struct range_table_work_area
1848 int *table; /* actual work area. */
1849 int allocated; /* allocated size for work area in bytes. */
1850 int used; /* actually used size in words. */
1851 int bits; /* flag to record character classes */
1854 /* Make sure that WORK_AREA can hold more N multibyte characters.
1855 This is used only in set_image_of_range and set_image_of_range_1.
1856 It expects WORK_AREA to be a pointer.
1857 If it can't get the space, it returns from the surrounding function. */
1859 #define EXTEND_RANGE_TABLE(work_area, n) \
1860 do { \
1861 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1863 extend_range_table_work_area (&work_area); \
1864 if ((work_area).table == 0) \
1865 return (REG_ESPACE); \
1867 } while (0)
1869 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1870 (work_area).bits |= (bit)
1872 /* Bits used to implement the multibyte-part of the various character classes
1873 such as [:alnum:] in a charset's range table. */
1874 #define BIT_WORD 0x1
1875 #define BIT_LOWER 0x2
1876 #define BIT_PUNCT 0x4
1877 #define BIT_SPACE 0x8
1878 #define BIT_UPPER 0x10
1879 #define BIT_MULTIBYTE 0x20
1881 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1882 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1883 do { \
1884 EXTEND_RANGE_TABLE ((work_area), 2); \
1885 (work_area).table[(work_area).used++] = (range_start); \
1886 (work_area).table[(work_area).used++] = (range_end); \
1887 } while (0)
1889 /* Free allocated memory for WORK_AREA. */
1890 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1891 do { \
1892 if ((work_area).table) \
1893 free ((work_area).table); \
1894 } while (0)
1896 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1897 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1898 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1899 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1902 /* Set the bit for character C in a list. */
1903 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1906 #ifdef emacs
1908 /* Store characters in the range FROM to TO in the bitmap at B (for
1909 ASCII and unibyte characters) and WORK_AREA (for multibyte
1910 characters) while translating them and paying attention to the
1911 continuity of translated characters.
1913 Implementation note: It is better to implement these fairly big
1914 macros by a function, but it's not that easy because macros called
1915 in this macro assume various local variables already declared. */
1917 /* Both FROM and TO are ASCII characters. */
1919 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1920 do { \
1921 int C0, C1; \
1923 for (C0 = (FROM); C0 <= (TO); C0++) \
1925 C1 = TRANSLATE (C0); \
1926 if (! ASCII_CHAR_P (C1)) \
1928 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1929 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1930 C1 = C0; \
1932 SET_LIST_BIT (C1); \
1934 } while (0)
1937 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1939 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1940 do { \
1941 int C0, C1, C2, I; \
1942 int USED = RANGE_TABLE_WORK_USED (work_area); \
1944 for (C0 = (FROM); C0 <= (TO); C0++) \
1946 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1947 if (CHAR_BYTE8_P (C1)) \
1948 SET_LIST_BIT (C0); \
1949 else \
1951 C2 = TRANSLATE (C1); \
1952 if (C2 == C1 \
1953 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1954 C1 = C0; \
1955 SET_LIST_BIT (C1); \
1956 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1958 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1959 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1961 if (C2 >= from - 1 && C2 <= to + 1) \
1963 if (C2 == from - 1) \
1964 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1965 else if (C2 == to + 1) \
1966 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1967 break; \
1970 if (I < USED) \
1971 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1974 } while (0)
1977 /* Both FROM and TO are multibyte characters. */
1979 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1980 do { \
1981 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1983 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1984 for (C0 = (FROM); C0 <= (TO); C0++) \
1986 C1 = TRANSLATE (C0); \
1987 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1988 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1989 SET_LIST_BIT (C2); \
1990 if (C1 >= (FROM) && C1 <= (TO)) \
1991 continue; \
1992 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1994 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1995 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1997 if (C1 >= from - 1 && C1 <= to + 1) \
1999 if (C1 == from - 1) \
2000 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2001 else if (C1 == to + 1) \
2002 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2003 break; \
2006 if (I < USED) \
2007 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2009 } while (0)
2011 #endif /* emacs */
2013 /* Get the next unsigned number in the uncompiled pattern. */
2014 #define GET_UNSIGNED_NUMBER(num) \
2015 do { \
2016 if (p == pend) \
2017 FREE_STACK_RETURN (REG_EBRACE); \
2018 else \
2020 PATFETCH (c); \
2021 while ('0' <= c && c <= '9') \
2023 int prev; \
2024 if (num < 0) \
2025 num = 0; \
2026 prev = num; \
2027 num = num * 10 + c - '0'; \
2028 if (num / 10 != prev) \
2029 FREE_STACK_RETURN (REG_BADBR); \
2030 if (p == pend) \
2031 FREE_STACK_RETURN (REG_EBRACE); \
2032 PATFETCH (c); \
2035 } while (0)
2037 #if ! WIDE_CHAR_SUPPORT
2039 /* Map a string to the char class it names (if any). */
2040 re_wctype_t
2041 re_wctype (const re_char *str)
2043 const char *string = (const char *) str;
2044 if (STREQ (string, "alnum")) return RECC_ALNUM;
2045 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2046 else if (STREQ (string, "word")) return RECC_WORD;
2047 else if (STREQ (string, "ascii")) return RECC_ASCII;
2048 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2049 else if (STREQ (string, "graph")) return RECC_GRAPH;
2050 else if (STREQ (string, "lower")) return RECC_LOWER;
2051 else if (STREQ (string, "print")) return RECC_PRINT;
2052 else if (STREQ (string, "punct")) return RECC_PUNCT;
2053 else if (STREQ (string, "space")) return RECC_SPACE;
2054 else if (STREQ (string, "upper")) return RECC_UPPER;
2055 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2056 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2057 else if (STREQ (string, "digit")) return RECC_DIGIT;
2058 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2059 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2060 else if (STREQ (string, "blank")) return RECC_BLANK;
2061 else return 0;
2064 /* True if CH is in the char class CC. */
2065 boolean
2066 re_iswctype (int ch, re_wctype_t cc)
2068 switch (cc)
2070 case RECC_ALNUM: return ISALNUM (ch) != 0;
2071 case RECC_ALPHA: return ISALPHA (ch) != 0;
2072 case RECC_BLANK: return ISBLANK (ch) != 0;
2073 case RECC_CNTRL: return ISCNTRL (ch) != 0;
2074 case RECC_DIGIT: return ISDIGIT (ch) != 0;
2075 case RECC_GRAPH: return ISGRAPH (ch) != 0;
2076 case RECC_LOWER: return ISLOWER (ch) != 0;
2077 case RECC_PRINT: return ISPRINT (ch) != 0;
2078 case RECC_PUNCT: return ISPUNCT (ch) != 0;
2079 case RECC_SPACE: return ISSPACE (ch) != 0;
2080 case RECC_UPPER: return ISUPPER (ch) != 0;
2081 case RECC_XDIGIT: return ISXDIGIT (ch) != 0;
2082 case RECC_ASCII: return IS_REAL_ASCII (ch) != 0;
2083 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2084 case RECC_UNIBYTE: return ISUNIBYTE (ch) != 0;
2085 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2086 case RECC_WORD: return ISWORD (ch) != 0;
2087 case RECC_ERROR: return false;
2088 default:
2089 abort ();
2093 /* Return a bit-pattern to use in the range-table bits to match multibyte
2094 chars of class CC. */
2095 static int
2096 re_wctype_to_bit (re_wctype_t cc)
2098 switch (cc)
2100 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2101 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2102 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2103 case RECC_LOWER: return BIT_LOWER;
2104 case RECC_UPPER: return BIT_UPPER;
2105 case RECC_PUNCT: return BIT_PUNCT;
2106 case RECC_SPACE: return BIT_SPACE;
2107 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2108 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2109 default:
2110 abort ();
2113 #endif
2115 /* Filling in the work area of a range. */
2117 /* Actually extend the space in WORK_AREA. */
2119 static void
2120 extend_range_table_work_area (struct range_table_work_area *work_area)
2122 work_area->allocated += 16 * sizeof (int);
2123 work_area->table = realloc (work_area->table, work_area->allocated);
2126 #if 0
2127 #ifdef emacs
2129 /* Carefully find the ranges of codes that are equivalent
2130 under case conversion to the range start..end when passed through
2131 TRANSLATE. Handle the case where non-letters can come in between
2132 two upper-case letters (which happens in Latin-1).
2133 Also handle the case of groups of more than 2 case-equivalent chars.
2135 The basic method is to look at consecutive characters and see
2136 if they can form a run that can be handled as one.
2138 Returns -1 if successful, REG_ESPACE if ran out of space. */
2140 static int
2141 set_image_of_range_1 (struct range_table_work_area *work_area,
2142 re_wchar_t start, re_wchar_t end,
2143 RE_TRANSLATE_TYPE translate)
2145 /* `one_case' indicates a character, or a run of characters,
2146 each of which is an isolate (no case-equivalents).
2147 This includes all ASCII non-letters.
2149 `two_case' indicates a character, or a run of characters,
2150 each of which has two case-equivalent forms.
2151 This includes all ASCII letters.
2153 `strange' indicates a character that has more than one
2154 case-equivalent. */
2156 enum case_type {one_case, two_case, strange};
2158 /* Describe the run that is in progress,
2159 which the next character can try to extend.
2160 If run_type is strange, that means there really is no run.
2161 If run_type is one_case, then run_start...run_end is the run.
2162 If run_type is two_case, then the run is run_start...run_end,
2163 and the case-equivalents end at run_eqv_end. */
2165 enum case_type run_type = strange;
2166 int run_start, run_end, run_eqv_end;
2168 Lisp_Object eqv_table;
2170 if (!RE_TRANSLATE_P (translate))
2172 EXTEND_RANGE_TABLE (work_area, 2);
2173 work_area->table[work_area->used++] = (start);
2174 work_area->table[work_area->used++] = (end);
2175 return -1;
2178 eqv_table = XCHAR_TABLE (translate)->extras[2];
2180 for (; start <= end; start++)
2182 enum case_type this_type;
2183 int eqv = RE_TRANSLATE (eqv_table, start);
2184 int minchar, maxchar;
2186 /* Classify this character */
2187 if (eqv == start)
2188 this_type = one_case;
2189 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2190 this_type = two_case;
2191 else
2192 this_type = strange;
2194 if (start < eqv)
2195 minchar = start, maxchar = eqv;
2196 else
2197 minchar = eqv, maxchar = start;
2199 /* Can this character extend the run in progress? */
2200 if (this_type == strange || this_type != run_type
2201 || !(minchar == run_end + 1
2202 && (run_type == two_case
2203 ? maxchar == run_eqv_end + 1 : 1)))
2205 /* No, end the run.
2206 Record each of its equivalent ranges. */
2207 if (run_type == one_case)
2209 EXTEND_RANGE_TABLE (work_area, 2);
2210 work_area->table[work_area->used++] = run_start;
2211 work_area->table[work_area->used++] = run_end;
2213 else if (run_type == two_case)
2215 EXTEND_RANGE_TABLE (work_area, 4);
2216 work_area->table[work_area->used++] = run_start;
2217 work_area->table[work_area->used++] = run_end;
2218 work_area->table[work_area->used++]
2219 = RE_TRANSLATE (eqv_table, run_start);
2220 work_area->table[work_area->used++]
2221 = RE_TRANSLATE (eqv_table, run_end);
2223 run_type = strange;
2226 if (this_type == strange)
2228 /* For a strange character, add each of its equivalents, one
2229 by one. Don't start a range. */
2232 EXTEND_RANGE_TABLE (work_area, 2);
2233 work_area->table[work_area->used++] = eqv;
2234 work_area->table[work_area->used++] = eqv;
2235 eqv = RE_TRANSLATE (eqv_table, eqv);
2237 while (eqv != start);
2240 /* Add this char to the run, or start a new run. */
2241 else if (run_type == strange)
2243 /* Initialize a new range. */
2244 run_type = this_type;
2245 run_start = start;
2246 run_end = start;
2247 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2249 else
2251 /* Extend a running range. */
2252 run_end = minchar;
2253 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2257 /* If a run is still in progress at the end, finish it now
2258 by recording its equivalent ranges. */
2259 if (run_type == one_case)
2261 EXTEND_RANGE_TABLE (work_area, 2);
2262 work_area->table[work_area->used++] = run_start;
2263 work_area->table[work_area->used++] = run_end;
2265 else if (run_type == two_case)
2267 EXTEND_RANGE_TABLE (work_area, 4);
2268 work_area->table[work_area->used++] = run_start;
2269 work_area->table[work_area->used++] = run_end;
2270 work_area->table[work_area->used++]
2271 = RE_TRANSLATE (eqv_table, run_start);
2272 work_area->table[work_area->used++]
2273 = RE_TRANSLATE (eqv_table, run_end);
2276 return -1;
2279 #endif /* emacs */
2281 /* Record the image of the range start..end when passed through
2282 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2283 and is not even necessarily contiguous.
2284 Normally we approximate it with the smallest contiguous range that contains
2285 all the chars we need. However, for Latin-1 we go to extra effort
2286 to do a better job.
2288 This function is not called for ASCII ranges.
2290 Returns -1 if successful, REG_ESPACE if ran out of space. */
2292 static int
2293 set_image_of_range (struct range_table_work_area *work_area,
2294 re_wchar_t start, re_wchar_t end,
2295 RE_TRANSLATE_TYPE translate)
2297 re_wchar_t cmin, cmax;
2299 #ifdef emacs
2300 /* For Latin-1 ranges, use set_image_of_range_1
2301 to get proper handling of ranges that include letters and nonletters.
2302 For a range that includes the whole of Latin-1, this is not necessary.
2303 For other character sets, we don't bother to get this right. */
2304 if (RE_TRANSLATE_P (translate) && start < 04400
2305 && !(start < 04200 && end >= 04377))
2307 int newend;
2308 int tem;
2309 newend = end;
2310 if (newend > 04377)
2311 newend = 04377;
2312 tem = set_image_of_range_1 (work_area, start, newend, translate);
2313 if (tem > 0)
2314 return tem;
2316 start = 04400;
2317 if (end < 04400)
2318 return -1;
2320 #endif
2322 EXTEND_RANGE_TABLE (work_area, 2);
2323 work_area->table[work_area->used++] = (start);
2324 work_area->table[work_area->used++] = (end);
2326 cmin = -1, cmax = -1;
2328 if (RE_TRANSLATE_P (translate))
2330 int ch;
2332 for (ch = start; ch <= end; ch++)
2334 re_wchar_t c = TRANSLATE (ch);
2335 if (! (start <= c && c <= end))
2337 if (cmin == -1)
2338 cmin = c, cmax = c;
2339 else
2341 cmin = MIN (cmin, c);
2342 cmax = MAX (cmax, c);
2347 if (cmin != -1)
2349 EXTEND_RANGE_TABLE (work_area, 2);
2350 work_area->table[work_area->used++] = (cmin);
2351 work_area->table[work_area->used++] = (cmax);
2355 return -1;
2357 #endif /* 0 */
2359 #ifndef MATCH_MAY_ALLOCATE
2361 /* If we cannot allocate large objects within re_match_2_internal,
2362 we make the fail stack and register vectors global.
2363 The fail stack, we grow to the maximum size when a regexp
2364 is compiled.
2365 The register vectors, we adjust in size each time we
2366 compile a regexp, according to the number of registers it needs. */
2368 static fail_stack_type fail_stack;
2370 /* Size with which the following vectors are currently allocated.
2371 That is so we can make them bigger as needed,
2372 but never make them smaller. */
2373 static int regs_allocated_size;
2375 static re_char ** regstart, ** regend;
2376 static re_char **best_regstart, **best_regend;
2378 /* Make the register vectors big enough for NUM_REGS registers,
2379 but don't make them smaller. */
2381 static
2382 regex_grow_registers (int num_regs)
2384 if (num_regs > regs_allocated_size)
2386 RETALLOC_IF (regstart, num_regs, re_char *);
2387 RETALLOC_IF (regend, num_regs, re_char *);
2388 RETALLOC_IF (best_regstart, num_regs, re_char *);
2389 RETALLOC_IF (best_regend, num_regs, re_char *);
2391 regs_allocated_size = num_regs;
2395 #endif /* not MATCH_MAY_ALLOCATE */
2397 static boolean group_in_compile_stack (compile_stack_type compile_stack,
2398 regnum_t regnum);
2400 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2401 Returns one of error codes defined in `regex.h', or zero for success.
2403 Assumes the `allocated' (and perhaps `buffer') and `translate'
2404 fields are set in BUFP on entry.
2406 If it succeeds, results are put in BUFP (if it returns an error, the
2407 contents of BUFP are undefined):
2408 `buffer' is the compiled pattern;
2409 `syntax' is set to SYNTAX;
2410 `used' is set to the length of the compiled pattern;
2411 `fastmap_accurate' is zero;
2412 `re_nsub' is the number of subexpressions in PATTERN;
2413 `not_bol' and `not_eol' are zero;
2415 The `fastmap' field is neither examined nor set. */
2417 /* Insert the `jump' from the end of last alternative to "here".
2418 The space for the jump has already been allocated. */
2419 #define FIXUP_ALT_JUMP() \
2420 do { \
2421 if (fixup_alt_jump) \
2422 STORE_JUMP (jump, fixup_alt_jump, b); \
2423 } while (0)
2426 /* Return, freeing storage we allocated. */
2427 #define FREE_STACK_RETURN(value) \
2428 do { \
2429 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2430 free (compile_stack.stack); \
2431 return value; \
2432 } while (0)
2434 static reg_errcode_t
2435 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2437 /* We fetch characters from PATTERN here. */
2438 register re_wchar_t c, c1;
2440 /* Points to the end of the buffer, where we should append. */
2441 register unsigned char *b;
2443 /* Keeps track of unclosed groups. */
2444 compile_stack_type compile_stack;
2446 /* Points to the current (ending) position in the pattern. */
2447 #ifdef AIX
2448 /* `const' makes AIX compiler fail. */
2449 unsigned char *p = pattern;
2450 #else
2451 re_char *p = pattern;
2452 #endif
2453 re_char *pend = pattern + size;
2455 /* How to translate the characters in the pattern. */
2456 RE_TRANSLATE_TYPE translate = bufp->translate;
2458 /* Address of the count-byte of the most recently inserted `exactn'
2459 command. This makes it possible to tell if a new exact-match
2460 character can be added to that command or if the character requires
2461 a new `exactn' command. */
2462 unsigned char *pending_exact = 0;
2464 /* Address of start of the most recently finished expression.
2465 This tells, e.g., postfix * where to find the start of its
2466 operand. Reset at the beginning of groups and alternatives. */
2467 unsigned char *laststart = 0;
2469 /* Address of beginning of regexp, or inside of last group. */
2470 unsigned char *begalt;
2472 /* Place in the uncompiled pattern (i.e., the {) to
2473 which to go back if the interval is invalid. */
2474 re_char *beg_interval;
2476 /* Address of the place where a forward jump should go to the end of
2477 the containing expression. Each alternative of an `or' -- except the
2478 last -- ends with a forward jump of this sort. */
2479 unsigned char *fixup_alt_jump = 0;
2481 /* Work area for range table of charset. */
2482 struct range_table_work_area range_table_work;
2484 /* If the object matched can contain multibyte characters. */
2485 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2487 /* Nonzero if we have pushed down into a subpattern. */
2488 int in_subpattern = 0;
2490 /* These hold the values of p, pattern, and pend from the main
2491 pattern when we have pushed into a subpattern. */
2492 re_char *main_p IF_LINT (= NULL);
2493 re_char *main_pattern IF_LINT (= NULL);
2494 re_char *main_pend IF_LINT (= NULL);
2496 #ifdef DEBUG
2497 debug++;
2498 DEBUG_PRINT1 ("\nCompiling pattern: ");
2499 if (debug > 0)
2501 unsigned debug_count;
2503 for (debug_count = 0; debug_count < size; debug_count++)
2504 putchar (pattern[debug_count]);
2505 putchar ('\n');
2507 #endif /* DEBUG */
2509 /* Initialize the compile stack. */
2510 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2511 if (compile_stack.stack == NULL)
2512 return REG_ESPACE;
2514 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2515 compile_stack.avail = 0;
2517 range_table_work.table = 0;
2518 range_table_work.allocated = 0;
2520 /* Initialize the pattern buffer. */
2521 bufp->syntax = syntax;
2522 bufp->fastmap_accurate = 0;
2523 bufp->not_bol = bufp->not_eol = 0;
2524 bufp->used_syntax = 0;
2526 /* Set `used' to zero, so that if we return an error, the pattern
2527 printer (for debugging) will think there's no pattern. We reset it
2528 at the end. */
2529 bufp->used = 0;
2531 /* Always count groups, whether or not bufp->no_sub is set. */
2532 bufp->re_nsub = 0;
2534 #if !defined emacs && !defined SYNTAX_TABLE
2535 /* Initialize the syntax table. */
2536 init_syntax_once ();
2537 #endif
2539 if (bufp->allocated == 0)
2541 if (bufp->buffer)
2542 { /* If zero allocated, but buffer is non-null, try to realloc
2543 enough space. This loses if buffer's address is bogus, but
2544 that is the user's responsibility. */
2545 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2547 else
2548 { /* Caller did not allocate a buffer. Do it for them. */
2549 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2551 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2553 bufp->allocated = INIT_BUF_SIZE;
2556 begalt = b = bufp->buffer;
2558 /* Loop through the uncompiled pattern until we're at the end. */
2559 while (1)
2561 if (p == pend)
2563 /* If this is the end of an included regexp,
2564 pop back to the main regexp and try again. */
2565 if (in_subpattern)
2567 in_subpattern = 0;
2568 pattern = main_pattern;
2569 p = main_p;
2570 pend = main_pend;
2571 continue;
2573 /* If this is the end of the main regexp, we are done. */
2574 break;
2577 PATFETCH (c);
2579 switch (c)
2581 case ' ':
2583 re_char *p1 = p;
2585 /* If there's no special whitespace regexp, treat
2586 spaces normally. And don't try to do this recursively. */
2587 if (!whitespace_regexp || in_subpattern)
2588 goto normal_char;
2590 /* Peek past following spaces. */
2591 while (p1 != pend)
2593 if (*p1 != ' ')
2594 break;
2595 p1++;
2597 /* If the spaces are followed by a repetition op,
2598 treat them normally. */
2599 if (p1 != pend
2600 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2601 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2602 goto normal_char;
2604 /* Replace the spaces with the whitespace regexp. */
2605 in_subpattern = 1;
2606 main_p = p1;
2607 main_pend = pend;
2608 main_pattern = pattern;
2609 p = pattern = whitespace_regexp;
2610 pend = p + strlen ((const char *) p);
2611 break;
2614 case '^':
2616 if ( /* If at start of pattern, it's an operator. */
2617 p == pattern + 1
2618 /* If context independent, it's an operator. */
2619 || syntax & RE_CONTEXT_INDEP_ANCHORS
2620 /* Otherwise, depends on what's come before. */
2621 || at_begline_loc_p (pattern, p, syntax))
2622 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2623 else
2624 goto normal_char;
2626 break;
2629 case '$':
2631 if ( /* If at end of pattern, it's an operator. */
2632 p == pend
2633 /* If context independent, it's an operator. */
2634 || syntax & RE_CONTEXT_INDEP_ANCHORS
2635 /* Otherwise, depends on what's next. */
2636 || at_endline_loc_p (p, pend, syntax))
2637 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2638 else
2639 goto normal_char;
2641 break;
2644 case '+':
2645 case '?':
2646 if ((syntax & RE_BK_PLUS_QM)
2647 || (syntax & RE_LIMITED_OPS))
2648 goto normal_char;
2649 handle_plus:
2650 case '*':
2651 /* If there is no previous pattern... */
2652 if (!laststart)
2654 if (syntax & RE_CONTEXT_INVALID_OPS)
2655 FREE_STACK_RETURN (REG_BADRPT);
2656 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2657 goto normal_char;
2661 /* 1 means zero (many) matches is allowed. */
2662 boolean zero_times_ok = 0, many_times_ok = 0;
2663 boolean greedy = 1;
2665 /* If there is a sequence of repetition chars, collapse it
2666 down to just one (the right one). We can't combine
2667 interval operators with these because of, e.g., `a{2}*',
2668 which should only match an even number of `a's. */
2670 for (;;)
2672 if ((syntax & RE_FRUGAL)
2673 && c == '?' && (zero_times_ok || many_times_ok))
2674 greedy = 0;
2675 else
2677 zero_times_ok |= c != '+';
2678 many_times_ok |= c != '?';
2681 if (p == pend)
2682 break;
2683 else if (*p == '*'
2684 || (!(syntax & RE_BK_PLUS_QM)
2685 && (*p == '+' || *p == '?')))
2687 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2689 if (p+1 == pend)
2690 FREE_STACK_RETURN (REG_EESCAPE);
2691 if (p[1] == '+' || p[1] == '?')
2692 PATFETCH (c); /* Gobble up the backslash. */
2693 else
2694 break;
2696 else
2697 break;
2698 /* If we get here, we found another repeat character. */
2699 PATFETCH (c);
2702 /* Star, etc. applied to an empty pattern is equivalent
2703 to an empty pattern. */
2704 if (!laststart || laststart == b)
2705 break;
2707 /* Now we know whether or not zero matches is allowed
2708 and also whether or not two or more matches is allowed. */
2709 if (greedy)
2711 if (many_times_ok)
2713 boolean simple = skip_one_char (laststart) == b;
2714 size_t startoffset = 0;
2715 re_opcode_t ofj =
2716 /* Check if the loop can match the empty string. */
2717 (simple || !analyse_first (laststart, b, NULL, 0))
2718 ? on_failure_jump : on_failure_jump_loop;
2719 assert (skip_one_char (laststart) <= b);
2721 if (!zero_times_ok && simple)
2722 { /* Since simple * loops can be made faster by using
2723 on_failure_keep_string_jump, we turn simple P+
2724 into PP* if P is simple. */
2725 unsigned char *p1, *p2;
2726 startoffset = b - laststart;
2727 GET_BUFFER_SPACE (startoffset);
2728 p1 = b; p2 = laststart;
2729 while (p2 < p1)
2730 *b++ = *p2++;
2731 zero_times_ok = 1;
2734 GET_BUFFER_SPACE (6);
2735 if (!zero_times_ok)
2736 /* A + loop. */
2737 STORE_JUMP (ofj, b, b + 6);
2738 else
2739 /* Simple * loops can use on_failure_keep_string_jump
2740 depending on what follows. But since we don't know
2741 that yet, we leave the decision up to
2742 on_failure_jump_smart. */
2743 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2744 laststart + startoffset, b + 6);
2745 b += 3;
2746 STORE_JUMP (jump, b, laststart + startoffset);
2747 b += 3;
2749 else
2751 /* A simple ? pattern. */
2752 assert (zero_times_ok);
2753 GET_BUFFER_SPACE (3);
2754 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2755 b += 3;
2758 else /* not greedy */
2759 { /* I wish the greedy and non-greedy cases could be merged. */
2761 GET_BUFFER_SPACE (7); /* We might use less. */
2762 if (many_times_ok)
2764 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2766 /* The non-greedy multiple match looks like
2767 a repeat..until: we only need a conditional jump
2768 at the end of the loop. */
2769 if (emptyp) BUF_PUSH (no_op);
2770 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2771 : on_failure_jump, b, laststart);
2772 b += 3;
2773 if (zero_times_ok)
2775 /* The repeat...until naturally matches one or more.
2776 To also match zero times, we need to first jump to
2777 the end of the loop (its conditional jump). */
2778 INSERT_JUMP (jump, laststart, b);
2779 b += 3;
2782 else
2784 /* non-greedy a?? */
2785 INSERT_JUMP (jump, laststart, b + 3);
2786 b += 3;
2787 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2788 b += 3;
2792 pending_exact = 0;
2793 break;
2796 case '.':
2797 laststart = b;
2798 BUF_PUSH (anychar);
2799 break;
2802 case '[':
2804 re_char *p1;
2806 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2808 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2810 /* Ensure that we have enough space to push a charset: the
2811 opcode, the length count, and the bitset; 34 bytes in all. */
2812 GET_BUFFER_SPACE (34);
2814 laststart = b;
2816 /* We test `*p == '^' twice, instead of using an if
2817 statement, so we only need one BUF_PUSH. */
2818 BUF_PUSH (*p == '^' ? charset_not : charset);
2819 if (*p == '^')
2820 p++;
2822 /* Remember the first position in the bracket expression. */
2823 p1 = p;
2825 /* Push the number of bytes in the bitmap. */
2826 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2828 /* Clear the whole map. */
2829 memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH);
2831 /* charset_not matches newline according to a syntax bit. */
2832 if ((re_opcode_t) b[-2] == charset_not
2833 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2834 SET_LIST_BIT ('\n');
2836 /* Read in characters and ranges, setting map bits. */
2837 for (;;)
2839 boolean escaped_char = false;
2840 const unsigned char *p2 = p;
2841 re_wchar_t ch;
2843 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2845 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2846 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2847 So the translation is done later in a loop. Example:
2848 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2849 PATFETCH (c);
2851 /* \ might escape characters inside [...] and [^...]. */
2852 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2854 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2856 PATFETCH (c);
2857 escaped_char = true;
2859 else
2861 /* Could be the end of the bracket expression. If it's
2862 not (i.e., when the bracket expression is `[]' so
2863 far), the ']' character bit gets set way below. */
2864 if (c == ']' && p2 != p1)
2865 break;
2868 /* See if we're at the beginning of a possible character
2869 class. */
2871 if (!escaped_char &&
2872 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2874 /* Leave room for the null. */
2875 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2876 const unsigned char *class_beg;
2878 PATFETCH (c);
2879 c1 = 0;
2880 class_beg = p;
2882 /* If pattern is `[[:'. */
2883 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2885 for (;;)
2887 PATFETCH (c);
2888 if ((c == ':' && *p == ']') || p == pend)
2889 break;
2890 if (c1 < CHAR_CLASS_MAX_LENGTH)
2891 str[c1++] = c;
2892 else
2893 /* This is in any case an invalid class name. */
2894 str[0] = '\0';
2896 str[c1] = '\0';
2898 /* If isn't a word bracketed by `[:' and `:]':
2899 undo the ending character, the letters, and
2900 leave the leading `:' and `[' (but set bits for
2901 them). */
2902 if (c == ':' && *p == ']')
2904 re_wctype_t cc = re_wctype (str);
2906 if (cc == 0)
2907 FREE_STACK_RETURN (REG_ECTYPE);
2909 /* Throw away the ] at the end of the character
2910 class. */
2911 PATFETCH (c);
2913 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2915 #ifndef emacs
2916 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
2917 if (re_iswctype (btowc (ch), cc))
2919 c = TRANSLATE (ch);
2920 if (c < (1 << BYTEWIDTH))
2921 SET_LIST_BIT (c);
2923 #else /* emacs */
2924 /* Most character classes in a multibyte match
2925 just set a flag. Exceptions are is_blank,
2926 is_digit, is_cntrl, and is_xdigit, since
2927 they can only match ASCII characters. We
2928 don't need to handle them for multibyte.
2929 They are distinguished by a negative wctype. */
2931 /* Setup the gl_state object to its buffer-defined
2932 value. This hardcodes the buffer-global
2933 syntax-table for ASCII chars, while the other chars
2934 will obey syntax-table properties. It's not ideal,
2935 but it's the way it's been done until now. */
2936 SETUP_BUFFER_SYNTAX_TABLE ();
2938 for (ch = 0; ch < 256; ++ch)
2940 c = RE_CHAR_TO_MULTIBYTE (ch);
2941 if (! CHAR_BYTE8_P (c)
2942 && re_iswctype (c, cc))
2944 SET_LIST_BIT (ch);
2945 c1 = TRANSLATE (c);
2946 if (c1 == c)
2947 continue;
2948 if (ASCII_CHAR_P (c1))
2949 SET_LIST_BIT (c1);
2950 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
2951 SET_LIST_BIT (c1);
2954 SET_RANGE_TABLE_WORK_AREA_BIT
2955 (range_table_work, re_wctype_to_bit (cc));
2956 #endif /* emacs */
2957 /* In most cases the matching rule for char classes
2958 only uses the syntax table for multibyte chars,
2959 so that the content of the syntax-table it is not
2960 hardcoded in the range_table. SPACE and WORD are
2961 the two exceptions. */
2962 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
2963 bufp->used_syntax = 1;
2965 /* Repeat the loop. */
2966 continue;
2968 else
2970 /* Go back to right after the "[:". */
2971 p = class_beg;
2972 SET_LIST_BIT ('[');
2974 /* Because the `:' may starts the range, we
2975 can't simply set bit and repeat the loop.
2976 Instead, just set it to C and handle below. */
2977 c = ':';
2981 if (p < pend && p[0] == '-' && p[1] != ']')
2984 /* Discard the `-'. */
2985 PATFETCH (c1);
2987 /* Fetch the character which ends the range. */
2988 PATFETCH (c1);
2989 #ifdef emacs
2990 if (CHAR_BYTE8_P (c1)
2991 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
2992 /* Treat the range from a multibyte character to
2993 raw-byte character as empty. */
2994 c = c1 + 1;
2995 #endif /* emacs */
2997 else
2998 /* Range from C to C. */
2999 c1 = c;
3001 if (c > c1)
3003 if (syntax & RE_NO_EMPTY_RANGES)
3004 FREE_STACK_RETURN (REG_ERANGEX);
3005 /* Else, repeat the loop. */
3007 else
3009 #ifndef emacs
3010 /* Set the range into bitmap */
3011 for (; c <= c1; c++)
3013 ch = TRANSLATE (c);
3014 if (ch < (1 << BYTEWIDTH))
3015 SET_LIST_BIT (ch);
3017 #else /* emacs */
3018 if (c < 128)
3020 ch = MIN (127, c1);
3021 SETUP_ASCII_RANGE (range_table_work, c, ch);
3022 c = ch + 1;
3023 if (CHAR_BYTE8_P (c1))
3024 c = BYTE8_TO_CHAR (128);
3026 if (c <= c1)
3028 if (CHAR_BYTE8_P (c))
3030 c = CHAR_TO_BYTE8 (c);
3031 c1 = CHAR_TO_BYTE8 (c1);
3032 for (; c <= c1; c++)
3033 SET_LIST_BIT (c);
3035 else if (multibyte)
3037 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3039 else
3041 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3044 #endif /* emacs */
3048 /* Discard any (non)matching list bytes that are all 0 at the
3049 end of the map. Decrease the map-length byte too. */
3050 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3051 b[-1]--;
3052 b += b[-1];
3054 /* Build real range table from work area. */
3055 if (RANGE_TABLE_WORK_USED (range_table_work)
3056 || RANGE_TABLE_WORK_BITS (range_table_work))
3058 int i;
3059 int used = RANGE_TABLE_WORK_USED (range_table_work);
3061 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3062 bytes for flags, two for COUNT, and three bytes for
3063 each character. */
3064 GET_BUFFER_SPACE (4 + used * 3);
3066 /* Indicate the existence of range table. */
3067 laststart[1] |= 0x80;
3069 /* Store the character class flag bits into the range table.
3070 If not in emacs, these flag bits are always 0. */
3071 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3072 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3074 STORE_NUMBER_AND_INCR (b, used / 2);
3075 for (i = 0; i < used; i++)
3076 STORE_CHARACTER_AND_INCR
3077 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3080 break;
3083 case '(':
3084 if (syntax & RE_NO_BK_PARENS)
3085 goto handle_open;
3086 else
3087 goto normal_char;
3090 case ')':
3091 if (syntax & RE_NO_BK_PARENS)
3092 goto handle_close;
3093 else
3094 goto normal_char;
3097 case '\n':
3098 if (syntax & RE_NEWLINE_ALT)
3099 goto handle_alt;
3100 else
3101 goto normal_char;
3104 case '|':
3105 if (syntax & RE_NO_BK_VBAR)
3106 goto handle_alt;
3107 else
3108 goto normal_char;
3111 case '{':
3112 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3113 goto handle_interval;
3114 else
3115 goto normal_char;
3118 case '\\':
3119 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3121 /* Do not translate the character after the \, so that we can
3122 distinguish, e.g., \B from \b, even if we normally would
3123 translate, e.g., B to b. */
3124 PATFETCH (c);
3126 switch (c)
3128 case '(':
3129 if (syntax & RE_NO_BK_PARENS)
3130 goto normal_backslash;
3132 handle_open:
3134 int shy = 0;
3135 regnum_t regnum = 0;
3136 if (p+1 < pend)
3138 /* Look for a special (?...) construct */
3139 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3141 PATFETCH (c); /* Gobble up the '?'. */
3142 while (!shy)
3144 PATFETCH (c);
3145 switch (c)
3147 case ':': shy = 1; break;
3148 case '0':
3149 /* An explicitly specified regnum must start
3150 with non-0. */
3151 if (regnum == 0)
3152 FREE_STACK_RETURN (REG_BADPAT);
3153 case '1': case '2': case '3': case '4':
3154 case '5': case '6': case '7': case '8': case '9':
3155 regnum = 10*regnum + (c - '0'); break;
3156 default:
3157 /* Only (?:...) is supported right now. */
3158 FREE_STACK_RETURN (REG_BADPAT);
3164 if (!shy)
3165 regnum = ++bufp->re_nsub;
3166 else if (regnum)
3167 { /* It's actually not shy, but explicitly numbered. */
3168 shy = 0;
3169 if (regnum > bufp->re_nsub)
3170 bufp->re_nsub = regnum;
3171 else if (regnum > bufp->re_nsub
3172 /* Ideally, we'd want to check that the specified
3173 group can't have matched (i.e. all subgroups
3174 using the same regnum are in other branches of
3175 OR patterns), but we don't currently keep track
3176 of enough info to do that easily. */
3177 || group_in_compile_stack (compile_stack, regnum))
3178 FREE_STACK_RETURN (REG_BADPAT);
3180 else
3181 /* It's really shy. */
3182 regnum = - bufp->re_nsub;
3184 if (COMPILE_STACK_FULL)
3186 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3187 compile_stack_elt_t);
3188 if (compile_stack.stack == NULL) return REG_ESPACE;
3190 compile_stack.size <<= 1;
3193 /* These are the values to restore when we hit end of this
3194 group. They are all relative offsets, so that if the
3195 whole pattern moves because of realloc, they will still
3196 be valid. */
3197 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3198 COMPILE_STACK_TOP.fixup_alt_jump
3199 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3200 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3201 COMPILE_STACK_TOP.regnum = regnum;
3203 /* Do not push a start_memory for groups beyond the last one
3204 we can represent in the compiled pattern. */
3205 if (regnum <= MAX_REGNUM && regnum > 0)
3206 BUF_PUSH_2 (start_memory, regnum);
3208 compile_stack.avail++;
3210 fixup_alt_jump = 0;
3211 laststart = 0;
3212 begalt = b;
3213 /* If we've reached MAX_REGNUM groups, then this open
3214 won't actually generate any code, so we'll have to
3215 clear pending_exact explicitly. */
3216 pending_exact = 0;
3217 break;
3220 case ')':
3221 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3223 if (COMPILE_STACK_EMPTY)
3225 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3226 goto normal_backslash;
3227 else
3228 FREE_STACK_RETURN (REG_ERPAREN);
3231 handle_close:
3232 FIXUP_ALT_JUMP ();
3234 /* See similar code for backslashed left paren above. */
3235 if (COMPILE_STACK_EMPTY)
3237 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3238 goto normal_char;
3239 else
3240 FREE_STACK_RETURN (REG_ERPAREN);
3243 /* Since we just checked for an empty stack above, this
3244 ``can't happen''. */
3245 assert (compile_stack.avail != 0);
3247 /* We don't just want to restore into `regnum', because
3248 later groups should continue to be numbered higher,
3249 as in `(ab)c(de)' -- the second group is #2. */
3250 regnum_t regnum;
3252 compile_stack.avail--;
3253 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3254 fixup_alt_jump
3255 = COMPILE_STACK_TOP.fixup_alt_jump
3256 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3257 : 0;
3258 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3259 regnum = COMPILE_STACK_TOP.regnum;
3260 /* If we've reached MAX_REGNUM groups, then this open
3261 won't actually generate any code, so we'll have to
3262 clear pending_exact explicitly. */
3263 pending_exact = 0;
3265 /* We're at the end of the group, so now we know how many
3266 groups were inside this one. */
3267 if (regnum <= MAX_REGNUM && regnum > 0)
3268 BUF_PUSH_2 (stop_memory, regnum);
3270 break;
3273 case '|': /* `\|'. */
3274 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3275 goto normal_backslash;
3276 handle_alt:
3277 if (syntax & RE_LIMITED_OPS)
3278 goto normal_char;
3280 /* Insert before the previous alternative a jump which
3281 jumps to this alternative if the former fails. */
3282 GET_BUFFER_SPACE (3);
3283 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3284 pending_exact = 0;
3285 b += 3;
3287 /* The alternative before this one has a jump after it
3288 which gets executed if it gets matched. Adjust that
3289 jump so it will jump to this alternative's analogous
3290 jump (put in below, which in turn will jump to the next
3291 (if any) alternative's such jump, etc.). The last such
3292 jump jumps to the correct final destination. A picture:
3293 _____ _____
3294 | | | |
3295 | v | v
3296 a | b | c
3298 If we are at `b', then fixup_alt_jump right now points to a
3299 three-byte space after `a'. We'll put in the jump, set
3300 fixup_alt_jump to right after `b', and leave behind three
3301 bytes which we'll fill in when we get to after `c'. */
3303 FIXUP_ALT_JUMP ();
3305 /* Mark and leave space for a jump after this alternative,
3306 to be filled in later either by next alternative or
3307 when know we're at the end of a series of alternatives. */
3308 fixup_alt_jump = b;
3309 GET_BUFFER_SPACE (3);
3310 b += 3;
3312 laststart = 0;
3313 begalt = b;
3314 break;
3317 case '{':
3318 /* If \{ is a literal. */
3319 if (!(syntax & RE_INTERVALS)
3320 /* If we're at `\{' and it's not the open-interval
3321 operator. */
3322 || (syntax & RE_NO_BK_BRACES))
3323 goto normal_backslash;
3325 handle_interval:
3327 /* If got here, then the syntax allows intervals. */
3329 /* At least (most) this many matches must be made. */
3330 int lower_bound = 0, upper_bound = -1;
3332 beg_interval = p;
3334 GET_UNSIGNED_NUMBER (lower_bound);
3336 if (c == ',')
3337 GET_UNSIGNED_NUMBER (upper_bound);
3338 else
3339 /* Interval such as `{1}' => match exactly once. */
3340 upper_bound = lower_bound;
3342 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3343 || (upper_bound >= 0 && lower_bound > upper_bound))
3344 FREE_STACK_RETURN (REG_BADBR);
3346 if (!(syntax & RE_NO_BK_BRACES))
3348 if (c != '\\')
3349 FREE_STACK_RETURN (REG_BADBR);
3350 if (p == pend)
3351 FREE_STACK_RETURN (REG_EESCAPE);
3352 PATFETCH (c);
3355 if (c != '}')
3356 FREE_STACK_RETURN (REG_BADBR);
3358 /* We just parsed a valid interval. */
3360 /* If it's invalid to have no preceding re. */
3361 if (!laststart)
3363 if (syntax & RE_CONTEXT_INVALID_OPS)
3364 FREE_STACK_RETURN (REG_BADRPT);
3365 else if (syntax & RE_CONTEXT_INDEP_OPS)
3366 laststart = b;
3367 else
3368 goto unfetch_interval;
3371 if (upper_bound == 0)
3372 /* If the upper bound is zero, just drop the sub pattern
3373 altogether. */
3374 b = laststart;
3375 else if (lower_bound == 1 && upper_bound == 1)
3376 /* Just match it once: nothing to do here. */
3379 /* Otherwise, we have a nontrivial interval. When
3380 we're all done, the pattern will look like:
3381 set_number_at <jump count> <upper bound>
3382 set_number_at <succeed_n count> <lower bound>
3383 succeed_n <after jump addr> <succeed_n count>
3384 <body of loop>
3385 jump_n <succeed_n addr> <jump count>
3386 (The upper bound and `jump_n' are omitted if
3387 `upper_bound' is 1, though.) */
3388 else
3389 { /* If the upper bound is > 1, we need to insert
3390 more at the end of the loop. */
3391 unsigned int nbytes = (upper_bound < 0 ? 3
3392 : upper_bound > 1 ? 5 : 0);
3393 unsigned int startoffset = 0;
3395 GET_BUFFER_SPACE (20); /* We might use less. */
3397 if (lower_bound == 0)
3399 /* A succeed_n that starts with 0 is really a
3400 a simple on_failure_jump_loop. */
3401 INSERT_JUMP (on_failure_jump_loop, laststart,
3402 b + 3 + nbytes);
3403 b += 3;
3405 else
3407 /* Initialize lower bound of the `succeed_n', even
3408 though it will be set during matching by its
3409 attendant `set_number_at' (inserted next),
3410 because `re_compile_fastmap' needs to know.
3411 Jump to the `jump_n' we might insert below. */
3412 INSERT_JUMP2 (succeed_n, laststart,
3413 b + 5 + nbytes,
3414 lower_bound);
3415 b += 5;
3417 /* Code to initialize the lower bound. Insert
3418 before the `succeed_n'. The `5' is the last two
3419 bytes of this `set_number_at', plus 3 bytes of
3420 the following `succeed_n'. */
3421 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3422 b += 5;
3423 startoffset += 5;
3426 if (upper_bound < 0)
3428 /* A negative upper bound stands for infinity,
3429 in which case it degenerates to a plain jump. */
3430 STORE_JUMP (jump, b, laststart + startoffset);
3431 b += 3;
3433 else if (upper_bound > 1)
3434 { /* More than one repetition is allowed, so
3435 append a backward jump to the `succeed_n'
3436 that starts this interval.
3438 When we've reached this during matching,
3439 we'll have matched the interval once, so
3440 jump back only `upper_bound - 1' times. */
3441 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3442 upper_bound - 1);
3443 b += 5;
3445 /* The location we want to set is the second
3446 parameter of the `jump_n'; that is `b-2' as
3447 an absolute address. `laststart' will be
3448 the `set_number_at' we're about to insert;
3449 `laststart+3' the number to set, the source
3450 for the relative address. But we are
3451 inserting into the middle of the pattern --
3452 so everything is getting moved up by 5.
3453 Conclusion: (b - 2) - (laststart + 3) + 5,
3454 i.e., b - laststart.
3456 We insert this at the beginning of the loop
3457 so that if we fail during matching, we'll
3458 reinitialize the bounds. */
3459 insert_op2 (set_number_at, laststart, b - laststart,
3460 upper_bound - 1, b);
3461 b += 5;
3464 pending_exact = 0;
3465 beg_interval = NULL;
3467 break;
3469 unfetch_interval:
3470 /* If an invalid interval, match the characters as literals. */
3471 assert (beg_interval);
3472 p = beg_interval;
3473 beg_interval = NULL;
3475 /* normal_char and normal_backslash need `c'. */
3476 c = '{';
3478 if (!(syntax & RE_NO_BK_BRACES))
3480 assert (p > pattern && p[-1] == '\\');
3481 goto normal_backslash;
3483 else
3484 goto normal_char;
3486 #ifdef emacs
3487 /* There is no way to specify the before_dot and after_dot
3488 operators. rms says this is ok. --karl */
3489 case '=':
3490 BUF_PUSH (at_dot);
3491 break;
3493 case 's':
3494 laststart = b;
3495 PATFETCH (c);
3496 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3497 break;
3499 case 'S':
3500 laststart = b;
3501 PATFETCH (c);
3502 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3503 break;
3505 case 'c':
3506 laststart = b;
3507 PATFETCH (c);
3508 BUF_PUSH_2 (categoryspec, c);
3509 break;
3511 case 'C':
3512 laststart = b;
3513 PATFETCH (c);
3514 BUF_PUSH_2 (notcategoryspec, c);
3515 break;
3516 #endif /* emacs */
3519 case 'w':
3520 if (syntax & RE_NO_GNU_OPS)
3521 goto normal_char;
3522 laststart = b;
3523 BUF_PUSH_2 (syntaxspec, Sword);
3524 break;
3527 case 'W':
3528 if (syntax & RE_NO_GNU_OPS)
3529 goto normal_char;
3530 laststart = b;
3531 BUF_PUSH_2 (notsyntaxspec, Sword);
3532 break;
3535 case '<':
3536 if (syntax & RE_NO_GNU_OPS)
3537 goto normal_char;
3538 BUF_PUSH (wordbeg);
3539 break;
3541 case '>':
3542 if (syntax & RE_NO_GNU_OPS)
3543 goto normal_char;
3544 BUF_PUSH (wordend);
3545 break;
3547 case '_':
3548 if (syntax & RE_NO_GNU_OPS)
3549 goto normal_char;
3550 laststart = b;
3551 PATFETCH (c);
3552 if (c == '<')
3553 BUF_PUSH (symbeg);
3554 else if (c == '>')
3555 BUF_PUSH (symend);
3556 else
3557 FREE_STACK_RETURN (REG_BADPAT);
3558 break;
3560 case 'b':
3561 if (syntax & RE_NO_GNU_OPS)
3562 goto normal_char;
3563 BUF_PUSH (wordbound);
3564 break;
3566 case 'B':
3567 if (syntax & RE_NO_GNU_OPS)
3568 goto normal_char;
3569 BUF_PUSH (notwordbound);
3570 break;
3572 case '`':
3573 if (syntax & RE_NO_GNU_OPS)
3574 goto normal_char;
3575 BUF_PUSH (begbuf);
3576 break;
3578 case '\'':
3579 if (syntax & RE_NO_GNU_OPS)
3580 goto normal_char;
3581 BUF_PUSH (endbuf);
3582 break;
3584 case '1': case '2': case '3': case '4': case '5':
3585 case '6': case '7': case '8': case '9':
3587 regnum_t reg;
3589 if (syntax & RE_NO_BK_REFS)
3590 goto normal_backslash;
3592 reg = c - '0';
3594 if (reg > bufp->re_nsub || reg < 1
3595 /* Can't back reference to a subexp before its end. */
3596 || group_in_compile_stack (compile_stack, reg))
3597 FREE_STACK_RETURN (REG_ESUBREG);
3599 laststart = b;
3600 BUF_PUSH_2 (duplicate, reg);
3602 break;
3605 case '+':
3606 case '?':
3607 if (syntax & RE_BK_PLUS_QM)
3608 goto handle_plus;
3609 else
3610 goto normal_backslash;
3612 default:
3613 normal_backslash:
3614 /* You might think it would be useful for \ to mean
3615 not to translate; but if we don't translate it
3616 it will never match anything. */
3617 goto normal_char;
3619 break;
3622 default:
3623 /* Expects the character in `c'. */
3624 normal_char:
3625 /* If no exactn currently being built. */
3626 if (!pending_exact
3628 /* If last exactn not at current position. */
3629 || pending_exact + *pending_exact + 1 != b
3631 /* We have only one byte following the exactn for the count. */
3632 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3634 /* If followed by a repetition operator. */
3635 || (p != pend && (*p == '*' || *p == '^'))
3636 || ((syntax & RE_BK_PLUS_QM)
3637 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3638 : p != pend && (*p == '+' || *p == '?'))
3639 || ((syntax & RE_INTERVALS)
3640 && ((syntax & RE_NO_BK_BRACES)
3641 ? p != pend && *p == '{'
3642 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3644 /* Start building a new exactn. */
3646 laststart = b;
3648 BUF_PUSH_2 (exactn, 0);
3649 pending_exact = b - 1;
3652 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3654 int len;
3656 if (multibyte)
3658 c = TRANSLATE (c);
3659 len = CHAR_STRING (c, b);
3660 b += len;
3662 else
3664 c1 = RE_CHAR_TO_MULTIBYTE (c);
3665 if (! CHAR_BYTE8_P (c1))
3667 re_wchar_t c2 = TRANSLATE (c1);
3669 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3670 c = c1;
3672 *b++ = c;
3673 len = 1;
3675 (*pending_exact) += len;
3678 break;
3679 } /* switch (c) */
3680 } /* while p != pend */
3683 /* Through the pattern now. */
3685 FIXUP_ALT_JUMP ();
3687 if (!COMPILE_STACK_EMPTY)
3688 FREE_STACK_RETURN (REG_EPAREN);
3690 /* If we don't want backtracking, force success
3691 the first time we reach the end of the compiled pattern. */
3692 if (syntax & RE_NO_POSIX_BACKTRACKING)
3693 BUF_PUSH (succeed);
3695 /* We have succeeded; set the length of the buffer. */
3696 bufp->used = b - bufp->buffer;
3698 #ifdef DEBUG
3699 if (debug > 0)
3701 re_compile_fastmap (bufp);
3702 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3703 print_compiled_pattern (bufp);
3705 debug--;
3706 #endif /* DEBUG */
3708 #ifndef MATCH_MAY_ALLOCATE
3709 /* Initialize the failure stack to the largest possible stack. This
3710 isn't necessary unless we're trying to avoid calling alloca in
3711 the search and match routines. */
3713 int num_regs = bufp->re_nsub + 1;
3715 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3717 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3718 falk_stack.stack = realloc (fail_stack.stack,
3719 fail_stack.size * sizeof *falk_stack.stack);
3722 regex_grow_registers (num_regs);
3724 #endif /* not MATCH_MAY_ALLOCATE */
3726 FREE_STACK_RETURN (REG_NOERROR);
3727 } /* regex_compile */
3729 /* Subroutines for `regex_compile'. */
3731 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3733 static void
3734 store_op1 (re_opcode_t op, unsigned char *loc, int arg)
3736 *loc = (unsigned char) op;
3737 STORE_NUMBER (loc + 1, arg);
3741 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3743 static void
3744 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2)
3746 *loc = (unsigned char) op;
3747 STORE_NUMBER (loc + 1, arg1);
3748 STORE_NUMBER (loc + 3, arg2);
3752 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3753 for OP followed by two-byte integer parameter ARG. */
3755 static void
3756 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end)
3758 register unsigned char *pfrom = end;
3759 register unsigned char *pto = end + 3;
3761 while (pfrom != loc)
3762 *--pto = *--pfrom;
3764 store_op1 (op, loc, arg);
3768 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3770 static void
3771 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end)
3773 register unsigned char *pfrom = end;
3774 register unsigned char *pto = end + 5;
3776 while (pfrom != loc)
3777 *--pto = *--pfrom;
3779 store_op2 (op, loc, arg1, arg2);
3783 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3784 after an alternative or a begin-subexpression. We assume there is at
3785 least one character before the ^. */
3787 static boolean
3788 at_begline_loc_p (const re_char *pattern, const re_char *p, reg_syntax_t syntax)
3790 re_char *prev = p - 2;
3791 boolean odd_backslashes;
3793 /* After a subexpression? */
3794 if (*prev == '(')
3795 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3797 /* After an alternative? */
3798 else if (*prev == '|')
3799 odd_backslashes = (syntax & RE_NO_BK_VBAR) == 0;
3801 /* After a shy subexpression? */
3802 else if (*prev == ':' && (syntax & RE_SHY_GROUPS))
3804 /* Skip over optional regnum. */
3805 while (prev - 1 >= pattern && prev[-1] >= '0' && prev[-1] <= '9')
3806 --prev;
3808 if (!(prev - 2 >= pattern
3809 && prev[-1] == '?' && prev[-2] == '('))
3810 return false;
3811 prev -= 2;
3812 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3814 else
3815 return false;
3817 /* Count the number of preceding backslashes. */
3818 p = prev;
3819 while (prev - 1 >= pattern && prev[-1] == '\\')
3820 --prev;
3821 return (p - prev) & odd_backslashes;
3825 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3826 at least one character after the $, i.e., `P < PEND'. */
3828 static boolean
3829 at_endline_loc_p (const re_char *p, const re_char *pend, reg_syntax_t syntax)
3831 re_char *next = p;
3832 boolean next_backslash = *next == '\\';
3833 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3835 return
3836 /* Before a subexpression? */
3837 (syntax & RE_NO_BK_PARENS ? *next == ')'
3838 : next_backslash && next_next && *next_next == ')')
3839 /* Before an alternative? */
3840 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3841 : next_backslash && next_next && *next_next == '|');
3845 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3846 false if it's not. */
3848 static boolean
3849 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
3851 ssize_t this_element;
3853 for (this_element = compile_stack.avail - 1;
3854 this_element >= 0;
3855 this_element--)
3856 if (compile_stack.stack[this_element].regnum == regnum)
3857 return true;
3859 return false;
3862 /* analyse_first.
3863 If fastmap is non-NULL, go through the pattern and fill fastmap
3864 with all the possible leading chars. If fastmap is NULL, don't
3865 bother filling it up (obviously) and only return whether the
3866 pattern could potentially match the empty string.
3868 Return 1 if p..pend might match the empty string.
3869 Return 0 if p..pend matches at least one char.
3870 Return -1 if fastmap was not updated accurately. */
3872 static int
3873 analyse_first (const re_char *p, const re_char *pend, char *fastmap, const int multibyte)
3875 int j, k;
3876 boolean not;
3878 /* If all elements for base leading-codes in fastmap is set, this
3879 flag is set true. */
3880 boolean match_any_multibyte_characters = false;
3882 assert (p);
3884 /* The loop below works as follows:
3885 - It has a working-list kept in the PATTERN_STACK and which basically
3886 starts by only containing a pointer to the first operation.
3887 - If the opcode we're looking at is a match against some set of
3888 chars, then we add those chars to the fastmap and go on to the
3889 next work element from the worklist (done via `break').
3890 - If the opcode is a control operator on the other hand, we either
3891 ignore it (if it's meaningless at this point, such as `start_memory')
3892 or execute it (if it's a jump). If the jump has several destinations
3893 (i.e. `on_failure_jump'), then we push the other destination onto the
3894 worklist.
3895 We guarantee termination by ignoring backward jumps (more or less),
3896 so that `p' is monotonically increasing. More to the point, we
3897 never set `p' (or push) anything `<= p1'. */
3899 while (p < pend)
3901 /* `p1' is used as a marker of how far back a `on_failure_jump'
3902 can go without being ignored. It is normally equal to `p'
3903 (which prevents any backward `on_failure_jump') except right
3904 after a plain `jump', to allow patterns such as:
3905 0: jump 10
3906 3..9: <body>
3907 10: on_failure_jump 3
3908 as used for the *? operator. */
3909 re_char *p1 = p;
3911 switch (*p++)
3913 case succeed:
3914 return 1;
3916 case duplicate:
3917 /* If the first character has to match a backreference, that means
3918 that the group was empty (since it already matched). Since this
3919 is the only case that interests us here, we can assume that the
3920 backreference must match the empty string. */
3921 p++;
3922 continue;
3925 /* Following are the cases which match a character. These end
3926 with `break'. */
3928 case exactn:
3929 if (fastmap)
3931 /* If multibyte is nonzero, the first byte of each
3932 character is an ASCII or a leading code. Otherwise,
3933 each byte is a character. Thus, this works in both
3934 cases. */
3935 fastmap[p[1]] = 1;
3936 if (! multibyte)
3938 /* For the case of matching this unibyte regex
3939 against multibyte, we must set a leading code of
3940 the corresponding multibyte character. */
3941 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
3943 fastmap[CHAR_LEADING_CODE (c)] = 1;
3946 break;
3949 case anychar:
3950 /* We could put all the chars except for \n (and maybe \0)
3951 but we don't bother since it is generally not worth it. */
3952 if (!fastmap) break;
3953 return -1;
3956 case charset_not:
3957 if (!fastmap) break;
3959 /* Chars beyond end of bitmap are possible matches. */
3960 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3961 j < (1 << BYTEWIDTH); j++)
3962 fastmap[j] = 1;
3965 /* Fallthrough */
3966 case charset:
3967 if (!fastmap) break;
3968 not = (re_opcode_t) *(p - 1) == charset_not;
3969 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3970 j >= 0; j--)
3971 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3972 fastmap[j] = 1;
3974 #ifdef emacs
3975 if (/* Any leading code can possibly start a character
3976 which doesn't match the specified set of characters. */
3979 /* If we can match a character class, we can match any
3980 multibyte characters. */
3981 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3982 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
3985 if (match_any_multibyte_characters == false)
3987 for (j = MIN_MULTIBYTE_LEADING_CODE;
3988 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
3989 fastmap[j] = 1;
3990 match_any_multibyte_characters = true;
3994 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3995 && match_any_multibyte_characters == false)
3997 /* Set fastmap[I] to 1 where I is a leading code of each
3998 multibyte character in the range table. */
3999 int c, count;
4000 unsigned char lc1, lc2;
4002 /* Make P points the range table. `+ 2' is to skip flag
4003 bits for a character class. */
4004 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4006 /* Extract the number of ranges in range table into COUNT. */
4007 EXTRACT_NUMBER_AND_INCR (count, p);
4008 for (; count > 0; count--, p += 3)
4010 /* Extract the start and end of each range. */
4011 EXTRACT_CHARACTER (c, p);
4012 lc1 = CHAR_LEADING_CODE (c);
4013 p += 3;
4014 EXTRACT_CHARACTER (c, p);
4015 lc2 = CHAR_LEADING_CODE (c);
4016 for (j = lc1; j <= lc2; j++)
4017 fastmap[j] = 1;
4020 #endif
4021 break;
4023 case syntaxspec:
4024 case notsyntaxspec:
4025 if (!fastmap) break;
4026 #ifndef emacs
4027 not = (re_opcode_t)p[-1] == notsyntaxspec;
4028 k = *p++;
4029 for (j = 0; j < (1 << BYTEWIDTH); j++)
4030 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4031 fastmap[j] = 1;
4032 break;
4033 #else /* emacs */
4034 /* This match depends on text properties. These end with
4035 aborting optimizations. */
4036 return -1;
4038 case categoryspec:
4039 case notcategoryspec:
4040 if (!fastmap) break;
4041 not = (re_opcode_t)p[-1] == notcategoryspec;
4042 k = *p++;
4043 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4044 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4045 fastmap[j] = 1;
4047 /* Any leading code can possibly start a character which
4048 has or doesn't has the specified category. */
4049 if (match_any_multibyte_characters == false)
4051 for (j = MIN_MULTIBYTE_LEADING_CODE;
4052 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4053 fastmap[j] = 1;
4054 match_any_multibyte_characters = true;
4056 break;
4058 /* All cases after this match the empty string. These end with
4059 `continue'. */
4061 case before_dot:
4062 case at_dot:
4063 case after_dot:
4064 #endif /* !emacs */
4065 case no_op:
4066 case begline:
4067 case endline:
4068 case begbuf:
4069 case endbuf:
4070 case wordbound:
4071 case notwordbound:
4072 case wordbeg:
4073 case wordend:
4074 case symbeg:
4075 case symend:
4076 continue;
4079 case jump:
4080 EXTRACT_NUMBER_AND_INCR (j, p);
4081 if (j < 0)
4082 /* Backward jumps can only go back to code that we've already
4083 visited. `re_compile' should make sure this is true. */
4084 break;
4085 p += j;
4086 switch (*p)
4088 case on_failure_jump:
4089 case on_failure_keep_string_jump:
4090 case on_failure_jump_loop:
4091 case on_failure_jump_nastyloop:
4092 case on_failure_jump_smart:
4093 p++;
4094 break;
4095 default:
4096 continue;
4098 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4099 to jump back to "just after here". */
4100 /* Fallthrough */
4102 case on_failure_jump:
4103 case on_failure_keep_string_jump:
4104 case on_failure_jump_nastyloop:
4105 case on_failure_jump_loop:
4106 case on_failure_jump_smart:
4107 EXTRACT_NUMBER_AND_INCR (j, p);
4108 if (p + j <= p1)
4109 ; /* Backward jump to be ignored. */
4110 else
4111 { /* We have to look down both arms.
4112 We first go down the "straight" path so as to minimize
4113 stack usage when going through alternatives. */
4114 int r = analyse_first (p, pend, fastmap, multibyte);
4115 if (r) return r;
4116 p += j;
4118 continue;
4121 case jump_n:
4122 /* This code simply does not properly handle forward jump_n. */
4123 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4124 p += 4;
4125 /* jump_n can either jump or fall through. The (backward) jump
4126 case has already been handled, so we only need to look at the
4127 fallthrough case. */
4128 continue;
4130 case succeed_n:
4131 /* If N == 0, it should be an on_failure_jump_loop instead. */
4132 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4133 p += 4;
4134 /* We only care about one iteration of the loop, so we don't
4135 need to consider the case where this behaves like an
4136 on_failure_jump. */
4137 continue;
4140 case set_number_at:
4141 p += 4;
4142 continue;
4145 case start_memory:
4146 case stop_memory:
4147 p += 1;
4148 continue;
4151 default:
4152 abort (); /* We have listed all the cases. */
4153 } /* switch *p++ */
4155 /* Getting here means we have found the possible starting
4156 characters for one path of the pattern -- and that the empty
4157 string does not match. We need not follow this path further. */
4158 return 0;
4159 } /* while p */
4161 /* We reached the end without matching anything. */
4162 return 1;
4164 } /* analyse_first */
4166 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4167 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4168 characters can start a string that matches the pattern. This fastmap
4169 is used by re_search to skip quickly over impossible starting points.
4171 Character codes above (1 << BYTEWIDTH) are not represented in the
4172 fastmap, but the leading codes are represented. Thus, the fastmap
4173 indicates which character sets could start a match.
4175 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4176 area as BUFP->fastmap.
4178 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4179 the pattern buffer.
4181 Returns 0 if we succeed, -2 if an internal error. */
4184 re_compile_fastmap (struct re_pattern_buffer *bufp)
4186 char *fastmap = bufp->fastmap;
4187 int analysis;
4189 assert (fastmap && bufp->buffer);
4191 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4192 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4194 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4195 fastmap, RE_MULTIBYTE_P (bufp));
4196 bufp->can_be_null = (analysis != 0);
4197 return 0;
4198 } /* re_compile_fastmap */
4200 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4201 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4202 this memory for recording register information. STARTS and ENDS
4203 must be allocated using the malloc library routine, and must each
4204 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4206 If NUM_REGS == 0, then subsequent matches should allocate their own
4207 register data.
4209 Unless this function is called, the first search or match using
4210 PATTERN_BUFFER will allocate its own register data, without
4211 freeing the old data. */
4213 void
4214 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4216 if (num_regs)
4218 bufp->regs_allocated = REGS_REALLOCATE;
4219 regs->num_regs = num_regs;
4220 regs->start = starts;
4221 regs->end = ends;
4223 else
4225 bufp->regs_allocated = REGS_UNALLOCATED;
4226 regs->num_regs = 0;
4227 regs->start = regs->end = (regoff_t *) 0;
4230 WEAK_ALIAS (__re_set_registers, re_set_registers)
4232 /* Searching routines. */
4234 /* Like re_search_2, below, but only one string is specified, and
4235 doesn't let you say where to stop matching. */
4237 regoff_t
4238 re_search (struct re_pattern_buffer *bufp, const char *string, size_t size,
4239 ssize_t startpos, ssize_t range, struct re_registers *regs)
4241 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4242 regs, size);
4244 WEAK_ALIAS (__re_search, re_search)
4246 /* Head address of virtual concatenation of string. */
4247 #define HEAD_ADDR_VSTRING(P) \
4248 (((P) >= size1 ? string2 : string1))
4250 /* Address of POS in the concatenation of virtual string. */
4251 #define POS_ADDR_VSTRING(POS) \
4252 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4254 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4255 virtual concatenation of STRING1 and STRING2, starting first at index
4256 STARTPOS, then at STARTPOS + 1, and so on.
4258 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4260 RANGE is how far to scan while trying to match. RANGE = 0 means try
4261 only at STARTPOS; in general, the last start tried is STARTPOS +
4262 RANGE.
4264 In REGS, return the indices of the virtual concatenation of STRING1
4265 and STRING2 that matched the entire BUFP->buffer and its contained
4266 subexpressions.
4268 Do not consider matching one past the index STOP in the virtual
4269 concatenation of STRING1 and STRING2.
4271 We return either the position in the strings at which the match was
4272 found, -1 if no match, or -2 if error (such as failure
4273 stack overflow). */
4275 regoff_t
4276 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, size_t size1,
4277 const char *str2, size_t size2, ssize_t startpos, ssize_t range,
4278 struct re_registers *regs, ssize_t stop)
4280 regoff_t val;
4281 re_char *string1 = (re_char*) str1;
4282 re_char *string2 = (re_char*) str2;
4283 register char *fastmap = bufp->fastmap;
4284 register RE_TRANSLATE_TYPE translate = bufp->translate;
4285 size_t total_size = size1 + size2;
4286 ssize_t endpos = startpos + range;
4287 boolean anchored_start;
4288 /* Nonzero if we are searching multibyte string. */
4289 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4291 /* Check for out-of-range STARTPOS. */
4292 if (startpos < 0 || startpos > total_size)
4293 return -1;
4295 /* Fix up RANGE if it might eventually take us outside
4296 the virtual concatenation of STRING1 and STRING2.
4297 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4298 if (endpos < 0)
4299 range = 0 - startpos;
4300 else if (endpos > total_size)
4301 range = total_size - startpos;
4303 /* If the search isn't to be a backwards one, don't waste time in a
4304 search for a pattern anchored at beginning of buffer. */
4305 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4307 if (startpos > 0)
4308 return -1;
4309 else
4310 range = 0;
4313 #ifdef emacs
4314 /* In a forward search for something that starts with \=.
4315 don't keep searching past point. */
4316 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4318 range = PT_BYTE - BEGV_BYTE - startpos;
4319 if (range < 0)
4320 return -1;
4322 #endif /* emacs */
4324 /* Update the fastmap now if not correct already. */
4325 if (fastmap && !bufp->fastmap_accurate)
4326 re_compile_fastmap (bufp);
4328 /* See whether the pattern is anchored. */
4329 anchored_start = (bufp->buffer[0] == begline);
4331 #ifdef emacs
4332 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4334 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4336 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4338 #endif
4340 /* Loop through the string, looking for a place to start matching. */
4341 for (;;)
4343 /* If the pattern is anchored,
4344 skip quickly past places we cannot match.
4345 We don't bother to treat startpos == 0 specially
4346 because that case doesn't repeat. */
4347 if (anchored_start && startpos > 0)
4349 if (! ((startpos <= size1 ? string1[startpos - 1]
4350 : string2[startpos - size1 - 1])
4351 == '\n'))
4352 goto advance;
4355 /* If a fastmap is supplied, skip quickly over characters that
4356 cannot be the start of a match. If the pattern can match the
4357 null string, however, we don't need to skip characters; we want
4358 the first null string. */
4359 if (fastmap && startpos < total_size && !bufp->can_be_null)
4361 register re_char *d;
4362 register re_wchar_t buf_ch;
4364 d = POS_ADDR_VSTRING (startpos);
4366 if (range > 0) /* Searching forwards. */
4368 register int lim = 0;
4369 ssize_t irange = range;
4371 if (startpos < size1 && startpos + range >= size1)
4372 lim = range - (size1 - startpos);
4374 /* Written out as an if-else to avoid testing `translate'
4375 inside the loop. */
4376 if (RE_TRANSLATE_P (translate))
4378 if (multibyte)
4379 while (range > lim)
4381 int buf_charlen;
4383 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4384 buf_ch = RE_TRANSLATE (translate, buf_ch);
4385 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4386 break;
4388 range -= buf_charlen;
4389 d += buf_charlen;
4391 else
4392 while (range > lim)
4394 register re_wchar_t ch, translated;
4396 buf_ch = *d;
4397 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4398 translated = RE_TRANSLATE (translate, ch);
4399 if (translated != ch
4400 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4401 buf_ch = ch;
4402 if (fastmap[buf_ch])
4403 break;
4404 d++;
4405 range--;
4408 else
4410 if (multibyte)
4411 while (range > lim)
4413 int buf_charlen;
4415 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4416 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4417 break;
4418 range -= buf_charlen;
4419 d += buf_charlen;
4421 else
4422 while (range > lim && !fastmap[*d])
4424 d++;
4425 range--;
4428 startpos += irange - range;
4430 else /* Searching backwards. */
4432 if (multibyte)
4434 buf_ch = STRING_CHAR (d);
4435 buf_ch = TRANSLATE (buf_ch);
4436 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4437 goto advance;
4439 else
4441 register re_wchar_t ch, translated;
4443 buf_ch = *d;
4444 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4445 translated = TRANSLATE (ch);
4446 if (translated != ch
4447 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4448 buf_ch = ch;
4449 if (! fastmap[TRANSLATE (buf_ch)])
4450 goto advance;
4455 /* If can't match the null string, and that's all we have left, fail. */
4456 if (range >= 0 && startpos == total_size && fastmap
4457 && !bufp->can_be_null)
4458 return -1;
4460 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4461 startpos, regs, stop);
4463 if (val >= 0)
4464 return startpos;
4466 if (val == -2)
4467 return -2;
4469 advance:
4470 if (!range)
4471 break;
4472 else if (range > 0)
4474 /* Update STARTPOS to the next character boundary. */
4475 if (multibyte)
4477 re_char *p = POS_ADDR_VSTRING (startpos);
4478 int len = BYTES_BY_CHAR_HEAD (*p);
4480 range -= len;
4481 if (range < 0)
4482 break;
4483 startpos += len;
4485 else
4487 range--;
4488 startpos++;
4491 else
4493 range++;
4494 startpos--;
4496 /* Update STARTPOS to the previous character boundary. */
4497 if (multibyte)
4499 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4500 re_char *p0 = p;
4501 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4503 /* Find the head of multibyte form. */
4504 PREV_CHAR_BOUNDARY (p, phead);
4505 range += p0 - 1 - p;
4506 if (range > 0)
4507 break;
4509 startpos -= p0 - 1 - p;
4513 return -1;
4514 } /* re_search_2 */
4515 WEAK_ALIAS (__re_search_2, re_search_2)
4517 /* Declarations and macros for re_match_2. */
4519 static int bcmp_translate (re_char *s1, re_char *s2,
4520 register ssize_t len,
4521 RE_TRANSLATE_TYPE translate,
4522 const int multibyte);
4524 /* This converts PTR, a pointer into one of the search strings `string1'
4525 and `string2' into an offset from the beginning of that string. */
4526 #define POINTER_TO_OFFSET(ptr) \
4527 (FIRST_STRING_P (ptr) \
4528 ? ((regoff_t) ((ptr) - string1)) \
4529 : ((regoff_t) ((ptr) - string2 + size1)))
4531 /* Call before fetching a character with *d. This switches over to
4532 string2 if necessary.
4533 Check re_match_2_internal for a discussion of why end_match_2 might
4534 not be within string2 (but be equal to end_match_1 instead). */
4535 #define PREFETCH() \
4536 while (d == dend) \
4538 /* End of string2 => fail. */ \
4539 if (dend == end_match_2) \
4540 goto fail; \
4541 /* End of string1 => advance to string2. */ \
4542 d = string2; \
4543 dend = end_match_2; \
4546 /* Call before fetching a char with *d if you already checked other limits.
4547 This is meant for use in lookahead operations like wordend, etc..
4548 where we might need to look at parts of the string that might be
4549 outside of the LIMITs (i.e past `stop'). */
4550 #define PREFETCH_NOLIMIT() \
4551 if (d == end1) \
4553 d = string2; \
4554 dend = end_match_2; \
4557 /* Test if at very beginning or at very end of the virtual concatenation
4558 of `string1' and `string2'. If only one string, it's `string2'. */
4559 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4560 #define AT_STRINGS_END(d) ((d) == end2)
4562 /* Disabled due to a compiler bug -- see comment at case wordbound */
4564 /* The comment at case wordbound is following one, but we don't use
4565 AT_WORD_BOUNDARY anymore to support multibyte form.
4567 The DEC Alpha C compiler 3.x generates incorrect code for the
4568 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4569 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4570 macro and introducing temporary variables works around the bug. */
4572 #if 0
4573 /* Test if D points to a character which is word-constituent. We have
4574 two special cases to check for: if past the end of string1, look at
4575 the first character in string2; and if before the beginning of
4576 string2, look at the last character in string1. */
4577 #define WORDCHAR_P(d) \
4578 (SYNTAX ((d) == end1 ? *string2 \
4579 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4580 == Sword)
4582 /* Test if the character before D and the one at D differ with respect
4583 to being word-constituent. */
4584 #define AT_WORD_BOUNDARY(d) \
4585 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4586 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4587 #endif
4589 /* Free everything we malloc. */
4590 #ifdef MATCH_MAY_ALLOCATE
4591 # define FREE_VAR(var) \
4592 do { \
4593 if (var) \
4595 REGEX_FREE (var); \
4596 var = NULL; \
4598 } while (0)
4599 # define FREE_VARIABLES() \
4600 do { \
4601 REGEX_FREE_STACK (fail_stack.stack); \
4602 FREE_VAR (regstart); \
4603 FREE_VAR (regend); \
4604 FREE_VAR (best_regstart); \
4605 FREE_VAR (best_regend); \
4606 } while (0)
4607 #else
4608 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4609 #endif /* not MATCH_MAY_ALLOCATE */
4612 /* Optimization routines. */
4614 /* If the operation is a match against one or more chars,
4615 return a pointer to the next operation, else return NULL. */
4616 static re_char *
4617 skip_one_char (const re_char *p)
4619 switch (*p++)
4621 case anychar:
4622 break;
4624 case exactn:
4625 p += *p + 1;
4626 break;
4628 case charset_not:
4629 case charset:
4630 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4632 int mcnt;
4633 p = CHARSET_RANGE_TABLE (p - 1);
4634 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4635 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4637 else
4638 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4639 break;
4641 case syntaxspec:
4642 case notsyntaxspec:
4643 #ifdef emacs
4644 case categoryspec:
4645 case notcategoryspec:
4646 #endif /* emacs */
4647 p++;
4648 break;
4650 default:
4651 p = NULL;
4653 return p;
4657 /* Jump over non-matching operations. */
4658 static re_char *
4659 skip_noops (const re_char *p, const re_char *pend)
4661 int mcnt;
4662 while (p < pend)
4664 switch (*p)
4666 case start_memory:
4667 case stop_memory:
4668 p += 2; break;
4669 case no_op:
4670 p += 1; break;
4671 case jump:
4672 p += 1;
4673 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4674 p += mcnt;
4675 break;
4676 default:
4677 return p;
4680 assert (p == pend);
4681 return p;
4684 /* Non-zero if "p1 matches something" implies "p2 fails". */
4685 static int
4686 mutually_exclusive_p (struct re_pattern_buffer *bufp, const re_char *p1, const re_char *p2)
4688 re_opcode_t op2;
4689 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4690 unsigned char *pend = bufp->buffer + bufp->used;
4692 assert (p1 >= bufp->buffer && p1 < pend
4693 && p2 >= bufp->buffer && p2 <= pend);
4695 /* Skip over open/close-group commands.
4696 If what follows this loop is a ...+ construct,
4697 look at what begins its body, since we will have to
4698 match at least one of that. */
4699 p2 = skip_noops (p2, pend);
4700 /* The same skip can be done for p1, except that this function
4701 is only used in the case where p1 is a simple match operator. */
4702 /* p1 = skip_noops (p1, pend); */
4704 assert (p1 >= bufp->buffer && p1 < pend
4705 && p2 >= bufp->buffer && p2 <= pend);
4707 op2 = p2 == pend ? succeed : *p2;
4709 switch (op2)
4711 case succeed:
4712 case endbuf:
4713 /* If we're at the end of the pattern, we can change. */
4714 if (skip_one_char (p1))
4716 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4717 return 1;
4719 break;
4721 case endline:
4722 case exactn:
4724 register re_wchar_t c
4725 = (re_opcode_t) *p2 == endline ? '\n'
4726 : RE_STRING_CHAR (p2 + 2, multibyte);
4728 if ((re_opcode_t) *p1 == exactn)
4730 if (c != RE_STRING_CHAR (p1 + 2, multibyte))
4732 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4733 return 1;
4737 else if ((re_opcode_t) *p1 == charset
4738 || (re_opcode_t) *p1 == charset_not)
4740 int not = (re_opcode_t) *p1 == charset_not;
4742 /* Test if C is listed in charset (or charset_not)
4743 at `p1'. */
4744 if (! multibyte || IS_REAL_ASCII (c))
4746 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4747 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4748 not = !not;
4750 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4751 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4753 /* `not' is equal to 1 if c would match, which means
4754 that we can't change to pop_failure_jump. */
4755 if (!not)
4757 DEBUG_PRINT1 (" No match => fast loop.\n");
4758 return 1;
4761 else if ((re_opcode_t) *p1 == anychar
4762 && c == '\n')
4764 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4765 return 1;
4768 break;
4770 case charset:
4772 if ((re_opcode_t) *p1 == exactn)
4773 /* Reuse the code above. */
4774 return mutually_exclusive_p (bufp, p2, p1);
4776 /* It is hard to list up all the character in charset
4777 P2 if it includes multibyte character. Give up in
4778 such case. */
4779 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4781 /* Now, we are sure that P2 has no range table.
4782 So, for the size of bitmap in P2, `p2[1]' is
4783 enough. But P1 may have range table, so the
4784 size of bitmap table of P1 is extracted by
4785 using macro `CHARSET_BITMAP_SIZE'.
4787 In a multibyte case, we know that all the character
4788 listed in P2 is ASCII. In a unibyte case, P1 has only a
4789 bitmap table. So, in both cases, it is enough to test
4790 only the bitmap table of P1. */
4792 if ((re_opcode_t) *p1 == charset)
4794 int idx;
4795 /* We win if the charset inside the loop
4796 has no overlap with the one after the loop. */
4797 for (idx = 0;
4798 (idx < (int) p2[1]
4799 && idx < CHARSET_BITMAP_SIZE (p1));
4800 idx++)
4801 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4802 break;
4804 if (idx == p2[1]
4805 || idx == CHARSET_BITMAP_SIZE (p1))
4807 DEBUG_PRINT1 (" No match => fast loop.\n");
4808 return 1;
4811 else if ((re_opcode_t) *p1 == charset_not)
4813 int idx;
4814 /* We win if the charset_not inside the loop lists
4815 every character listed in the charset after. */
4816 for (idx = 0; idx < (int) p2[1]; idx++)
4817 if (! (p2[2 + idx] == 0
4818 || (idx < CHARSET_BITMAP_SIZE (p1)
4819 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4820 break;
4822 if (idx == p2[1])
4824 DEBUG_PRINT1 (" No match => fast loop.\n");
4825 return 1;
4830 break;
4832 case charset_not:
4833 switch (*p1)
4835 case exactn:
4836 case charset:
4837 /* Reuse the code above. */
4838 return mutually_exclusive_p (bufp, p2, p1);
4839 case charset_not:
4840 /* When we have two charset_not, it's very unlikely that
4841 they don't overlap. The union of the two sets of excluded
4842 chars should cover all possible chars, which, as a matter of
4843 fact, is virtually impossible in multibyte buffers. */
4844 break;
4846 break;
4848 case wordend:
4849 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
4850 case symend:
4851 return ((re_opcode_t) *p1 == syntaxspec
4852 && (p1[1] == Ssymbol || p1[1] == Sword));
4853 case notsyntaxspec:
4854 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
4856 case wordbeg:
4857 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
4858 case symbeg:
4859 return ((re_opcode_t) *p1 == notsyntaxspec
4860 && (p1[1] == Ssymbol || p1[1] == Sword));
4861 case syntaxspec:
4862 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
4864 case wordbound:
4865 return (((re_opcode_t) *p1 == notsyntaxspec
4866 || (re_opcode_t) *p1 == syntaxspec)
4867 && p1[1] == Sword);
4869 #ifdef emacs
4870 case categoryspec:
4871 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4872 case notcategoryspec:
4873 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4874 #endif /* emacs */
4876 default:
4880 /* Safe default. */
4881 return 0;
4885 /* Matching routines. */
4887 #ifndef emacs /* Emacs never uses this. */
4888 /* re_match is like re_match_2 except it takes only a single string. */
4890 regoff_t
4891 re_match (struct re_pattern_buffer *bufp, const char *string,
4892 size_t size, ssize_t pos, struct re_registers *regs)
4894 regoff_t result = re_match_2_internal (bufp, NULL, 0, (re_char*) string,
4895 size, pos, regs, size);
4896 return result;
4898 WEAK_ALIAS (__re_match, re_match)
4899 #endif /* not emacs */
4901 #ifdef emacs
4902 /* In Emacs, this is the string or buffer in which we
4903 are matching. It is used for looking up syntax properties. */
4904 Lisp_Object re_match_object;
4905 #endif
4907 /* re_match_2 matches the compiled pattern in BUFP against the
4908 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4909 and SIZE2, respectively). We start matching at POS, and stop
4910 matching at STOP.
4912 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4913 store offsets for the substring each group matched in REGS. See the
4914 documentation for exactly how many groups we fill.
4916 We return -1 if no match, -2 if an internal error (such as the
4917 failure stack overflowing). Otherwise, we return the length of the
4918 matched substring. */
4920 regoff_t
4921 re_match_2 (struct re_pattern_buffer *bufp, const char *string1,
4922 size_t size1, const char *string2, size_t size2, ssize_t pos,
4923 struct re_registers *regs, ssize_t stop)
4925 regoff_t result;
4927 #ifdef emacs
4928 ssize_t charpos;
4929 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4930 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4931 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4932 #endif
4934 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4935 (re_char*) string2, size2,
4936 pos, regs, stop);
4937 return result;
4939 WEAK_ALIAS (__re_match_2, re_match_2)
4942 /* This is a separate function so that we can force an alloca cleanup
4943 afterwards. */
4944 static regoff_t
4945 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1,
4946 size_t size1, const re_char *string2, size_t size2,
4947 ssize_t pos, struct re_registers *regs, ssize_t stop)
4949 /* General temporaries. */
4950 ssize_t mcnt;
4951 size_t reg;
4953 /* Just past the end of the corresponding string. */
4954 re_char *end1, *end2;
4956 /* Pointers into string1 and string2, just past the last characters in
4957 each to consider matching. */
4958 re_char *end_match_1, *end_match_2;
4960 /* Where we are in the data, and the end of the current string. */
4961 re_char *d, *dend;
4963 /* Used sometimes to remember where we were before starting matching
4964 an operator so that we can go back in case of failure. This "atomic"
4965 behavior of matching opcodes is indispensable to the correctness
4966 of the on_failure_keep_string_jump optimization. */
4967 re_char *dfail;
4969 /* Where we are in the pattern, and the end of the pattern. */
4970 re_char *p = bufp->buffer;
4971 re_char *pend = p + bufp->used;
4973 /* We use this to map every character in the string. */
4974 RE_TRANSLATE_TYPE translate = bufp->translate;
4976 /* Nonzero if BUFP is setup from a multibyte regex. */
4977 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4979 /* Nonzero if STRING1/STRING2 are multibyte. */
4980 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4982 /* Failure point stack. Each place that can handle a failure further
4983 down the line pushes a failure point on this stack. It consists of
4984 regstart, and regend for all registers corresponding to
4985 the subexpressions we're currently inside, plus the number of such
4986 registers, and, finally, two char *'s. The first char * is where
4987 to resume scanning the pattern; the second one is where to resume
4988 scanning the strings. */
4989 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4990 fail_stack_type fail_stack;
4991 #endif
4992 #ifdef DEBUG
4993 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4994 #endif
4996 #if defined REL_ALLOC && defined REGEX_MALLOC
4997 /* This holds the pointer to the failure stack, when
4998 it is allocated relocatably. */
4999 fail_stack_elt_t *failure_stack_ptr;
5000 #endif
5002 /* We fill all the registers internally, independent of what we
5003 return, for use in backreferences. The number here includes
5004 an element for register zero. */
5005 size_t num_regs = bufp->re_nsub + 1;
5007 /* Information on the contents of registers. These are pointers into
5008 the input strings; they record just what was matched (on this
5009 attempt) by a subexpression part of the pattern, that is, the
5010 regnum-th regstart pointer points to where in the pattern we began
5011 matching and the regnum-th regend points to right after where we
5012 stopped matching the regnum-th subexpression. (The zeroth register
5013 keeps track of what the whole pattern matches.) */
5014 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5015 re_char **regstart, **regend;
5016 #endif
5018 /* The following record the register info as found in the above
5019 variables when we find a match better than any we've seen before.
5020 This happens as we backtrack through the failure points, which in
5021 turn happens only if we have not yet matched the entire string. */
5022 unsigned best_regs_set = false;
5023 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5024 re_char **best_regstart, **best_regend;
5025 #endif
5027 /* Logically, this is `best_regend[0]'. But we don't want to have to
5028 allocate space for that if we're not allocating space for anything
5029 else (see below). Also, we never need info about register 0 for
5030 any of the other register vectors, and it seems rather a kludge to
5031 treat `best_regend' differently than the rest. So we keep track of
5032 the end of the best match so far in a separate variable. We
5033 initialize this to NULL so that when we backtrack the first time
5034 and need to test it, it's not garbage. */
5035 re_char *match_end = NULL;
5037 #ifdef DEBUG
5038 /* Counts the total number of registers pushed. */
5039 unsigned num_regs_pushed = 0;
5040 #endif
5042 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5044 INIT_FAIL_STACK ();
5046 #ifdef MATCH_MAY_ALLOCATE
5047 /* Do not bother to initialize all the register variables if there are
5048 no groups in the pattern, as it takes a fair amount of time. If
5049 there are groups, we include space for register 0 (the whole
5050 pattern), even though we never use it, since it simplifies the
5051 array indexing. We should fix this. */
5052 if (bufp->re_nsub)
5054 regstart = REGEX_TALLOC (num_regs, re_char *);
5055 regend = REGEX_TALLOC (num_regs, re_char *);
5056 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5057 best_regend = REGEX_TALLOC (num_regs, re_char *);
5059 if (!(regstart && regend && best_regstart && best_regend))
5061 FREE_VARIABLES ();
5062 return -2;
5065 else
5067 /* We must initialize all our variables to NULL, so that
5068 `FREE_VARIABLES' doesn't try to free them. */
5069 regstart = regend = best_regstart = best_regend = NULL;
5071 #endif /* MATCH_MAY_ALLOCATE */
5073 /* The starting position is bogus. */
5074 if (pos < 0 || pos > size1 + size2)
5076 FREE_VARIABLES ();
5077 return -1;
5080 /* Initialize subexpression text positions to -1 to mark ones that no
5081 start_memory/stop_memory has been seen for. Also initialize the
5082 register information struct. */
5083 for (reg = 1; reg < num_regs; reg++)
5084 regstart[reg] = regend[reg] = NULL;
5086 /* We move `string1' into `string2' if the latter's empty -- but not if
5087 `string1' is null. */
5088 if (size2 == 0 && string1 != NULL)
5090 string2 = string1;
5091 size2 = size1;
5092 string1 = 0;
5093 size1 = 0;
5095 end1 = string1 + size1;
5096 end2 = string2 + size2;
5098 /* `p' scans through the pattern as `d' scans through the data.
5099 `dend' is the end of the input string that `d' points within. `d'
5100 is advanced into the following input string whenever necessary, but
5101 this happens before fetching; therefore, at the beginning of the
5102 loop, `d' can be pointing at the end of a string, but it cannot
5103 equal `string2'. */
5104 if (pos >= size1)
5106 /* Only match within string2. */
5107 d = string2 + pos - size1;
5108 dend = end_match_2 = string2 + stop - size1;
5109 end_match_1 = end1; /* Just to give it a value. */
5111 else
5113 if (stop < size1)
5115 /* Only match within string1. */
5116 end_match_1 = string1 + stop;
5117 /* BEWARE!
5118 When we reach end_match_1, PREFETCH normally switches to string2.
5119 But in the present case, this means that just doing a PREFETCH
5120 makes us jump from `stop' to `gap' within the string.
5121 What we really want here is for the search to stop as
5122 soon as we hit end_match_1. That's why we set end_match_2
5123 to end_match_1 (since PREFETCH fails as soon as we hit
5124 end_match_2). */
5125 end_match_2 = end_match_1;
5127 else
5128 { /* It's important to use this code when stop == size so that
5129 moving `d' from end1 to string2 will not prevent the d == dend
5130 check from catching the end of string. */
5131 end_match_1 = end1;
5132 end_match_2 = string2 + stop - size1;
5134 d = string1 + pos;
5135 dend = end_match_1;
5138 DEBUG_PRINT1 ("The compiled pattern is: ");
5139 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5140 DEBUG_PRINT1 ("The string to match is: `");
5141 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5142 DEBUG_PRINT1 ("'\n");
5144 /* This loops over pattern commands. It exits by returning from the
5145 function if the match is complete, or it drops through if the match
5146 fails at this starting point in the input data. */
5147 for (;;)
5149 DEBUG_PRINT2 ("\n%p: ", p);
5151 if (p == pend)
5152 { /* End of pattern means we might have succeeded. */
5153 DEBUG_PRINT1 ("end of pattern ... ");
5155 /* If we haven't matched the entire string, and we want the
5156 longest match, try backtracking. */
5157 if (d != end_match_2)
5159 /* 1 if this match ends in the same string (string1 or string2)
5160 as the best previous match. */
5161 boolean same_str_p = (FIRST_STRING_P (match_end)
5162 == FIRST_STRING_P (d));
5163 /* 1 if this match is the best seen so far. */
5164 boolean best_match_p;
5166 /* AIX compiler got confused when this was combined
5167 with the previous declaration. */
5168 if (same_str_p)
5169 best_match_p = d > match_end;
5170 else
5171 best_match_p = !FIRST_STRING_P (d);
5173 DEBUG_PRINT1 ("backtracking.\n");
5175 if (!FAIL_STACK_EMPTY ())
5176 { /* More failure points to try. */
5178 /* If exceeds best match so far, save it. */
5179 if (!best_regs_set || best_match_p)
5181 best_regs_set = true;
5182 match_end = d;
5184 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5186 for (reg = 1; reg < num_regs; reg++)
5188 best_regstart[reg] = regstart[reg];
5189 best_regend[reg] = regend[reg];
5192 goto fail;
5195 /* If no failure points, don't restore garbage. And if
5196 last match is real best match, don't restore second
5197 best one. */
5198 else if (best_regs_set && !best_match_p)
5200 restore_best_regs:
5201 /* Restore best match. It may happen that `dend ==
5202 end_match_1' while the restored d is in string2.
5203 For example, the pattern `x.*y.*z' against the
5204 strings `x-' and `y-z-', if the two strings are
5205 not consecutive in memory. */
5206 DEBUG_PRINT1 ("Restoring best registers.\n");
5208 d = match_end;
5209 dend = ((d >= string1 && d <= end1)
5210 ? end_match_1 : end_match_2);
5212 for (reg = 1; reg < num_regs; reg++)
5214 regstart[reg] = best_regstart[reg];
5215 regend[reg] = best_regend[reg];
5218 } /* d != end_match_2 */
5220 succeed_label:
5221 DEBUG_PRINT1 ("Accepting match.\n");
5223 /* If caller wants register contents data back, do it. */
5224 if (regs && !bufp->no_sub)
5226 /* Have the register data arrays been allocated? */
5227 if (bufp->regs_allocated == REGS_UNALLOCATED)
5228 { /* No. So allocate them with malloc. We need one
5229 extra element beyond `num_regs' for the `-1' marker
5230 GNU code uses. */
5231 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5232 regs->start = TALLOC (regs->num_regs, regoff_t);
5233 regs->end = TALLOC (regs->num_regs, regoff_t);
5234 if (regs->start == NULL || regs->end == NULL)
5236 FREE_VARIABLES ();
5237 return -2;
5239 bufp->regs_allocated = REGS_REALLOCATE;
5241 else if (bufp->regs_allocated == REGS_REALLOCATE)
5242 { /* Yes. If we need more elements than were already
5243 allocated, reallocate them. If we need fewer, just
5244 leave it alone. */
5245 if (regs->num_regs < num_regs + 1)
5247 regs->num_regs = num_regs + 1;
5248 RETALLOC (regs->start, regs->num_regs, regoff_t);
5249 RETALLOC (regs->end, regs->num_regs, regoff_t);
5250 if (regs->start == NULL || regs->end == NULL)
5252 FREE_VARIABLES ();
5253 return -2;
5257 else
5259 /* These braces fend off a "empty body in an else-statement"
5260 warning under GCC when assert expands to nothing. */
5261 assert (bufp->regs_allocated == REGS_FIXED);
5264 /* Convert the pointer data in `regstart' and `regend' to
5265 indices. Register zero has to be set differently,
5266 since we haven't kept track of any info for it. */
5267 if (regs->num_regs > 0)
5269 regs->start[0] = pos;
5270 regs->end[0] = POINTER_TO_OFFSET (d);
5273 /* Go through the first `min (num_regs, regs->num_regs)'
5274 registers, since that is all we initialized. */
5275 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5277 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5278 regs->start[reg] = regs->end[reg] = -1;
5279 else
5281 regs->start[reg]
5282 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5283 regs->end[reg]
5284 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5288 /* If the regs structure we return has more elements than
5289 were in the pattern, set the extra elements to -1. If
5290 we (re)allocated the registers, this is the case,
5291 because we always allocate enough to have at least one
5292 -1 at the end. */
5293 for (reg = num_regs; reg < regs->num_regs; reg++)
5294 regs->start[reg] = regs->end[reg] = -1;
5295 } /* regs && !bufp->no_sub */
5297 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5298 nfailure_points_pushed, nfailure_points_popped,
5299 nfailure_points_pushed - nfailure_points_popped);
5300 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5302 mcnt = POINTER_TO_OFFSET (d) - pos;
5304 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5306 FREE_VARIABLES ();
5307 return mcnt;
5310 /* Otherwise match next pattern command. */
5311 switch (*p++)
5313 /* Ignore these. Used to ignore the n of succeed_n's which
5314 currently have n == 0. */
5315 case no_op:
5316 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5317 break;
5319 case succeed:
5320 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5321 goto succeed_label;
5323 /* Match the next n pattern characters exactly. The following
5324 byte in the pattern defines n, and the n bytes after that
5325 are the characters to match. */
5326 case exactn:
5327 mcnt = *p++;
5328 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5330 /* Remember the start point to rollback upon failure. */
5331 dfail = d;
5333 #ifndef emacs
5334 /* This is written out as an if-else so we don't waste time
5335 testing `translate' inside the loop. */
5336 if (RE_TRANSLATE_P (translate))
5339 PREFETCH ();
5340 if (RE_TRANSLATE (translate, *d) != *p++)
5342 d = dfail;
5343 goto fail;
5345 d++;
5347 while (--mcnt);
5348 else
5351 PREFETCH ();
5352 if (*d++ != *p++)
5354 d = dfail;
5355 goto fail;
5358 while (--mcnt);
5359 #else /* emacs */
5360 /* The cost of testing `translate' is comparatively small. */
5361 if (target_multibyte)
5364 int pat_charlen, buf_charlen;
5365 int pat_ch, buf_ch;
5367 PREFETCH ();
5368 if (multibyte)
5369 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5370 else
5372 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5373 pat_charlen = 1;
5375 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5377 if (TRANSLATE (buf_ch) != pat_ch)
5379 d = dfail;
5380 goto fail;
5383 p += pat_charlen;
5384 d += buf_charlen;
5385 mcnt -= pat_charlen;
5387 while (mcnt > 0);
5388 else
5391 int pat_charlen;
5392 int pat_ch, buf_ch;
5394 PREFETCH ();
5395 if (multibyte)
5397 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5398 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5400 else
5402 pat_ch = *p;
5403 pat_charlen = 1;
5405 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5406 if (! CHAR_BYTE8_P (buf_ch))
5408 buf_ch = TRANSLATE (buf_ch);
5409 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5410 if (buf_ch < 0)
5411 buf_ch = *d;
5413 else
5414 buf_ch = *d;
5415 if (buf_ch != pat_ch)
5417 d = dfail;
5418 goto fail;
5420 p += pat_charlen;
5421 d++;
5423 while (--mcnt);
5424 #endif
5425 break;
5428 /* Match any character except possibly a newline or a null. */
5429 case anychar:
5431 int buf_charlen;
5432 re_wchar_t buf_ch;
5434 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5436 PREFETCH ();
5437 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5438 target_multibyte);
5439 buf_ch = TRANSLATE (buf_ch);
5441 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5442 && buf_ch == '\n')
5443 || ((bufp->syntax & RE_DOT_NOT_NULL)
5444 && buf_ch == '\000'))
5445 goto fail;
5447 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5448 d += buf_charlen;
5450 break;
5453 case charset:
5454 case charset_not:
5456 register unsigned int c;
5457 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5458 int len;
5460 /* Start of actual range_table, or end of bitmap if there is no
5461 range table. */
5462 re_char *range_table IF_LINT (= NULL);
5464 /* Nonzero if there is a range table. */
5465 int range_table_exists;
5467 /* Number of ranges of range table. This is not included
5468 in the initial byte-length of the command. */
5469 int count = 0;
5471 /* Whether matching against a unibyte character. */
5472 boolean unibyte_char = false;
5474 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5476 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5478 if (range_table_exists)
5480 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5481 EXTRACT_NUMBER_AND_INCR (count, range_table);
5484 PREFETCH ();
5485 c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5486 if (target_multibyte)
5488 int c1;
5490 c = TRANSLATE (c);
5491 c1 = RE_CHAR_TO_UNIBYTE (c);
5492 if (c1 >= 0)
5494 unibyte_char = true;
5495 c = c1;
5498 else
5500 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5502 if (! CHAR_BYTE8_P (c1))
5504 c1 = TRANSLATE (c1);
5505 c1 = RE_CHAR_TO_UNIBYTE (c1);
5506 if (c1 >= 0)
5508 unibyte_char = true;
5509 c = c1;
5512 else
5513 unibyte_char = true;
5516 if (unibyte_char && c < (1 << BYTEWIDTH))
5517 { /* Lookup bitmap. */
5518 /* Cast to `unsigned' instead of `unsigned char' in
5519 case the bit list is a full 32 bytes long. */
5520 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5521 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5522 not = !not;
5524 #ifdef emacs
5525 else if (range_table_exists)
5527 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5529 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5530 | (class_bits & BIT_MULTIBYTE)
5531 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5532 | (class_bits & BIT_SPACE && ISSPACE (c))
5533 | (class_bits & BIT_UPPER && ISUPPER (c))
5534 | (class_bits & BIT_WORD && ISWORD (c)))
5535 not = !not;
5536 else
5537 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5539 #endif /* emacs */
5541 if (range_table_exists)
5542 p = CHARSET_RANGE_TABLE_END (range_table, count);
5543 else
5544 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5546 if (!not) goto fail;
5548 d += len;
5550 break;
5553 /* The beginning of a group is represented by start_memory.
5554 The argument is the register number. The text
5555 matched within the group is recorded (in the internal
5556 registers data structure) under the register number. */
5557 case start_memory:
5558 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5560 /* In case we need to undo this operation (via backtracking). */
5561 PUSH_FAILURE_REG ((unsigned int)*p);
5563 regstart[*p] = d;
5564 regend[*p] = NULL; /* probably unnecessary. -sm */
5565 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5567 /* Move past the register number and inner group count. */
5568 p += 1;
5569 break;
5572 /* The stop_memory opcode represents the end of a group. Its
5573 argument is the same as start_memory's: the register number. */
5574 case stop_memory:
5575 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5577 assert (!REG_UNSET (regstart[*p]));
5578 /* Strictly speaking, there should be code such as:
5580 assert (REG_UNSET (regend[*p]));
5581 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5583 But the only info to be pushed is regend[*p] and it is known to
5584 be UNSET, so there really isn't anything to push.
5585 Not pushing anything, on the other hand deprives us from the
5586 guarantee that regend[*p] is UNSET since undoing this operation
5587 will not reset its value properly. This is not important since
5588 the value will only be read on the next start_memory or at
5589 the very end and both events can only happen if this stop_memory
5590 is *not* undone. */
5592 regend[*p] = d;
5593 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5595 /* Move past the register number and the inner group count. */
5596 p += 1;
5597 break;
5600 /* \<digit> has been turned into a `duplicate' command which is
5601 followed by the numeric value of <digit> as the register number. */
5602 case duplicate:
5604 register re_char *d2, *dend2;
5605 int regno = *p++; /* Get which register to match against. */
5606 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5608 /* Can't back reference a group which we've never matched. */
5609 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5610 goto fail;
5612 /* Where in input to try to start matching. */
5613 d2 = regstart[regno];
5615 /* Remember the start point to rollback upon failure. */
5616 dfail = d;
5618 /* Where to stop matching; if both the place to start and
5619 the place to stop matching are in the same string, then
5620 set to the place to stop, otherwise, for now have to use
5621 the end of the first string. */
5623 dend2 = ((FIRST_STRING_P (regstart[regno])
5624 == FIRST_STRING_P (regend[regno]))
5625 ? regend[regno] : end_match_1);
5626 for (;;)
5628 /* If necessary, advance to next segment in register
5629 contents. */
5630 while (d2 == dend2)
5632 if (dend2 == end_match_2) break;
5633 if (dend2 == regend[regno]) break;
5635 /* End of string1 => advance to string2. */
5636 d2 = string2;
5637 dend2 = regend[regno];
5639 /* At end of register contents => success */
5640 if (d2 == dend2) break;
5642 /* If necessary, advance to next segment in data. */
5643 PREFETCH ();
5645 /* How many characters left in this segment to match. */
5646 mcnt = dend - d;
5648 /* Want how many consecutive characters we can match in
5649 one shot, so, if necessary, adjust the count. */
5650 if (mcnt > dend2 - d2)
5651 mcnt = dend2 - d2;
5653 /* Compare that many; failure if mismatch, else move
5654 past them. */
5655 if (RE_TRANSLATE_P (translate)
5656 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5657 : memcmp (d, d2, mcnt))
5659 d = dfail;
5660 goto fail;
5662 d += mcnt, d2 += mcnt;
5665 break;
5668 /* begline matches the empty string at the beginning of the string
5669 (unless `not_bol' is set in `bufp'), and after newlines. */
5670 case begline:
5671 DEBUG_PRINT1 ("EXECUTING begline.\n");
5673 if (AT_STRINGS_BEG (d))
5675 if (!bufp->not_bol) break;
5677 else
5679 unsigned c;
5680 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5681 if (c == '\n')
5682 break;
5684 /* In all other cases, we fail. */
5685 goto fail;
5688 /* endline is the dual of begline. */
5689 case endline:
5690 DEBUG_PRINT1 ("EXECUTING endline.\n");
5692 if (AT_STRINGS_END (d))
5694 if (!bufp->not_eol) break;
5696 else
5698 PREFETCH_NOLIMIT ();
5699 if (*d == '\n')
5700 break;
5702 goto fail;
5705 /* Match at the very beginning of the data. */
5706 case begbuf:
5707 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5708 if (AT_STRINGS_BEG (d))
5709 break;
5710 goto fail;
5713 /* Match at the very end of the data. */
5714 case endbuf:
5715 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5716 if (AT_STRINGS_END (d))
5717 break;
5718 goto fail;
5721 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5722 pushes NULL as the value for the string on the stack. Then
5723 `POP_FAILURE_POINT' will keep the current value for the
5724 string, instead of restoring it. To see why, consider
5725 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5726 then the . fails against the \n. But the next thing we want
5727 to do is match the \n against the \n; if we restored the
5728 string value, we would be back at the foo.
5730 Because this is used only in specific cases, we don't need to
5731 check all the things that `on_failure_jump' does, to make
5732 sure the right things get saved on the stack. Hence we don't
5733 share its code. The only reason to push anything on the
5734 stack at all is that otherwise we would have to change
5735 `anychar's code to do something besides goto fail in this
5736 case; that seems worse than this. */
5737 case on_failure_keep_string_jump:
5738 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5739 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5740 mcnt, p + mcnt);
5742 PUSH_FAILURE_POINT (p - 3, NULL);
5743 break;
5745 /* A nasty loop is introduced by the non-greedy *? and +?.
5746 With such loops, the stack only ever contains one failure point
5747 at a time, so that a plain on_failure_jump_loop kind of
5748 cycle detection cannot work. Worse yet, such a detection
5749 can not only fail to detect a cycle, but it can also wrongly
5750 detect a cycle (between different instantiations of the same
5751 loop).
5752 So the method used for those nasty loops is a little different:
5753 We use a special cycle-detection-stack-frame which is pushed
5754 when the on_failure_jump_nastyloop failure-point is *popped*.
5755 This special frame thus marks the beginning of one iteration
5756 through the loop and we can hence easily check right here
5757 whether something matched between the beginning and the end of
5758 the loop. */
5759 case on_failure_jump_nastyloop:
5760 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5761 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5762 mcnt, p + mcnt);
5764 assert ((re_opcode_t)p[-4] == no_op);
5766 int cycle = 0;
5767 CHECK_INFINITE_LOOP (p - 4, d);
5768 if (!cycle)
5769 /* If there's a cycle, just continue without pushing
5770 this failure point. The failure point is the "try again"
5771 option, which shouldn't be tried.
5772 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5773 PUSH_FAILURE_POINT (p - 3, d);
5775 break;
5777 /* Simple loop detecting on_failure_jump: just check on the
5778 failure stack if the same spot was already hit earlier. */
5779 case on_failure_jump_loop:
5780 on_failure:
5781 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5782 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5783 mcnt, p + mcnt);
5785 int cycle = 0;
5786 CHECK_INFINITE_LOOP (p - 3, d);
5787 if (cycle)
5788 /* If there's a cycle, get out of the loop, as if the matching
5789 had failed. We used to just `goto fail' here, but that was
5790 aborting the search a bit too early: we want to keep the
5791 empty-loop-match and keep matching after the loop.
5792 We want (x?)*y\1z to match both xxyz and xxyxz. */
5793 p += mcnt;
5794 else
5795 PUSH_FAILURE_POINT (p - 3, d);
5797 break;
5800 /* Uses of on_failure_jump:
5802 Each alternative starts with an on_failure_jump that points
5803 to the beginning of the next alternative. Each alternative
5804 except the last ends with a jump that in effect jumps past
5805 the rest of the alternatives. (They really jump to the
5806 ending jump of the following alternative, because tensioning
5807 these jumps is a hassle.)
5809 Repeats start with an on_failure_jump that points past both
5810 the repetition text and either the following jump or
5811 pop_failure_jump back to this on_failure_jump. */
5812 case on_failure_jump:
5813 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5814 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5815 mcnt, p + mcnt);
5817 PUSH_FAILURE_POINT (p -3, d);
5818 break;
5820 /* This operation is used for greedy *.
5821 Compare the beginning of the repeat with what in the
5822 pattern follows its end. If we can establish that there
5823 is nothing that they would both match, i.e., that we
5824 would have to backtrack because of (as in, e.g., `a*a')
5825 then we can use a non-backtracking loop based on
5826 on_failure_keep_string_jump instead of on_failure_jump. */
5827 case on_failure_jump_smart:
5828 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5829 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5830 mcnt, p + mcnt);
5832 re_char *p1 = p; /* Next operation. */
5833 /* Here, we discard `const', making re_match non-reentrant. */
5834 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5835 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5837 p -= 3; /* Reset so that we will re-execute the
5838 instruction once it's been changed. */
5840 EXTRACT_NUMBER (mcnt, p2 - 2);
5842 /* Ensure this is a indeed the trivial kind of loop
5843 we are expecting. */
5844 assert (skip_one_char (p1) == p2 - 3);
5845 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5846 DEBUG_STATEMENT (debug += 2);
5847 if (mutually_exclusive_p (bufp, p1, p2))
5849 /* Use a fast `on_failure_keep_string_jump' loop. */
5850 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5851 *p3 = (unsigned char) on_failure_keep_string_jump;
5852 STORE_NUMBER (p2 - 2, mcnt + 3);
5854 else
5856 /* Default to a safe `on_failure_jump' loop. */
5857 DEBUG_PRINT1 (" smart default => slow loop.\n");
5858 *p3 = (unsigned char) on_failure_jump;
5860 DEBUG_STATEMENT (debug -= 2);
5862 break;
5864 /* Unconditionally jump (without popping any failure points). */
5865 case jump:
5866 unconditional_jump:
5867 IMMEDIATE_QUIT_CHECK;
5868 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5869 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5870 p += mcnt; /* Do the jump. */
5871 DEBUG_PRINT2 ("(to %p).\n", p);
5872 break;
5875 /* Have to succeed matching what follows at least n times.
5876 After that, handle like `on_failure_jump'. */
5877 case succeed_n:
5878 /* Signedness doesn't matter since we only compare MCNT to 0. */
5879 EXTRACT_NUMBER (mcnt, p + 2);
5880 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5882 /* Originally, mcnt is how many times we HAVE to succeed. */
5883 if (mcnt != 0)
5885 /* Here, we discard `const', making re_match non-reentrant. */
5886 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5887 mcnt--;
5888 p += 4;
5889 PUSH_NUMBER (p2, mcnt);
5891 else
5892 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5893 goto on_failure;
5894 break;
5896 case jump_n:
5897 /* Signedness doesn't matter since we only compare MCNT to 0. */
5898 EXTRACT_NUMBER (mcnt, p + 2);
5899 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5901 /* Originally, this is how many times we CAN jump. */
5902 if (mcnt != 0)
5904 /* Here, we discard `const', making re_match non-reentrant. */
5905 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5906 mcnt--;
5907 PUSH_NUMBER (p2, mcnt);
5908 goto unconditional_jump;
5910 /* If don't have to jump any more, skip over the rest of command. */
5911 else
5912 p += 4;
5913 break;
5915 case set_number_at:
5917 unsigned char *p2; /* Location of the counter. */
5918 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5920 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5921 /* Here, we discard `const', making re_match non-reentrant. */
5922 p2 = (unsigned char*) p + mcnt;
5923 /* Signedness doesn't matter since we only copy MCNT's bits . */
5924 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5925 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
5926 PUSH_NUMBER (p2, mcnt);
5927 break;
5930 case wordbound:
5931 case notwordbound:
5933 boolean not = (re_opcode_t) *(p - 1) == notwordbound;
5934 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5936 /* We SUCCEED (or FAIL) in one of the following cases: */
5938 /* Case 1: D is at the beginning or the end of string. */
5939 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5940 not = !not;
5941 else
5943 /* C1 is the character before D, S1 is the syntax of C1, C2
5944 is the character at D, and S2 is the syntax of C2. */
5945 re_wchar_t c1, c2;
5946 int s1, s2;
5947 int dummy;
5948 #ifdef emacs
5949 ssize_t offset = PTR_TO_OFFSET (d - 1);
5950 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5951 UPDATE_SYNTAX_TABLE (charpos);
5952 #endif
5953 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5954 s1 = SYNTAX (c1);
5955 #ifdef emacs
5956 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
5957 #endif
5958 PREFETCH_NOLIMIT ();
5959 GET_CHAR_AFTER (c2, d, dummy);
5960 s2 = SYNTAX (c2);
5962 if (/* Case 2: Only one of S1 and S2 is Sword. */
5963 ((s1 == Sword) != (s2 == Sword))
5964 /* Case 3: Both of S1 and S2 are Sword, and macro
5965 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5966 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5967 not = !not;
5969 if (not)
5970 break;
5971 else
5972 goto fail;
5975 case wordbeg:
5976 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5978 /* We FAIL in one of the following cases: */
5980 /* Case 1: D is at the end of string. */
5981 if (AT_STRINGS_END (d))
5982 goto fail;
5983 else
5985 /* C1 is the character before D, S1 is the syntax of C1, C2
5986 is the character at D, and S2 is the syntax of C2. */
5987 re_wchar_t c1, c2;
5988 int s1, s2;
5989 int dummy;
5990 #ifdef emacs
5991 ssize_t offset = PTR_TO_OFFSET (d);
5992 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5993 UPDATE_SYNTAX_TABLE (charpos);
5994 #endif
5995 PREFETCH ();
5996 GET_CHAR_AFTER (c2, d, dummy);
5997 s2 = SYNTAX (c2);
5999 /* Case 2: S2 is not Sword. */
6000 if (s2 != Sword)
6001 goto fail;
6003 /* Case 3: D is not at the beginning of string ... */
6004 if (!AT_STRINGS_BEG (d))
6006 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6007 #ifdef emacs
6008 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6009 #endif
6010 s1 = SYNTAX (c1);
6012 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6013 returns 0. */
6014 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6015 goto fail;
6018 break;
6020 case wordend:
6021 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6023 /* We FAIL in one of the following cases: */
6025 /* Case 1: D is at the beginning of string. */
6026 if (AT_STRINGS_BEG (d))
6027 goto fail;
6028 else
6030 /* C1 is the character before D, S1 is the syntax of C1, C2
6031 is the character at D, and S2 is the syntax of C2. */
6032 re_wchar_t c1, c2;
6033 int s1, s2;
6034 int dummy;
6035 #ifdef emacs
6036 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6037 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6038 UPDATE_SYNTAX_TABLE (charpos);
6039 #endif
6040 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6041 s1 = SYNTAX (c1);
6043 /* Case 2: S1 is not Sword. */
6044 if (s1 != Sword)
6045 goto fail;
6047 /* Case 3: D is not at the end of string ... */
6048 if (!AT_STRINGS_END (d))
6050 PREFETCH_NOLIMIT ();
6051 GET_CHAR_AFTER (c2, d, dummy);
6052 #ifdef emacs
6053 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6054 #endif
6055 s2 = SYNTAX (c2);
6057 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6058 returns 0. */
6059 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6060 goto fail;
6063 break;
6065 case symbeg:
6066 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6068 /* We FAIL in one of the following cases: */
6070 /* Case 1: D is at the end of string. */
6071 if (AT_STRINGS_END (d))
6072 goto fail;
6073 else
6075 /* C1 is the character before D, S1 is the syntax of C1, C2
6076 is the character at D, and S2 is the syntax of C2. */
6077 re_wchar_t c1, c2;
6078 int s1, s2;
6079 #ifdef emacs
6080 ssize_t offset = PTR_TO_OFFSET (d);
6081 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6082 UPDATE_SYNTAX_TABLE (charpos);
6083 #endif
6084 PREFETCH ();
6085 c2 = RE_STRING_CHAR (d, target_multibyte);
6086 s2 = SYNTAX (c2);
6088 /* Case 2: S2 is neither Sword nor Ssymbol. */
6089 if (s2 != Sword && s2 != Ssymbol)
6090 goto fail;
6092 /* Case 3: D is not at the beginning of string ... */
6093 if (!AT_STRINGS_BEG (d))
6095 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6096 #ifdef emacs
6097 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6098 #endif
6099 s1 = SYNTAX (c1);
6101 /* ... and S1 is Sword or Ssymbol. */
6102 if (s1 == Sword || s1 == Ssymbol)
6103 goto fail;
6106 break;
6108 case symend:
6109 DEBUG_PRINT1 ("EXECUTING symend.\n");
6111 /* We FAIL in one of the following cases: */
6113 /* Case 1: D is at the beginning of string. */
6114 if (AT_STRINGS_BEG (d))
6115 goto fail;
6116 else
6118 /* C1 is the character before D, S1 is the syntax of C1, C2
6119 is the character at D, and S2 is the syntax of C2. */
6120 re_wchar_t c1, c2;
6121 int s1, s2;
6122 #ifdef emacs
6123 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6124 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6125 UPDATE_SYNTAX_TABLE (charpos);
6126 #endif
6127 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6128 s1 = SYNTAX (c1);
6130 /* Case 2: S1 is neither Ssymbol nor Sword. */
6131 if (s1 != Sword && s1 != Ssymbol)
6132 goto fail;
6134 /* Case 3: D is not at the end of string ... */
6135 if (!AT_STRINGS_END (d))
6137 PREFETCH_NOLIMIT ();
6138 c2 = RE_STRING_CHAR (d, target_multibyte);
6139 #ifdef emacs
6140 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6141 #endif
6142 s2 = SYNTAX (c2);
6144 /* ... and S2 is Sword or Ssymbol. */
6145 if (s2 == Sword || s2 == Ssymbol)
6146 goto fail;
6149 break;
6151 case syntaxspec:
6152 case notsyntaxspec:
6154 boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6155 mcnt = *p++;
6156 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6157 PREFETCH ();
6158 #ifdef emacs
6160 ssize_t offset = PTR_TO_OFFSET (d);
6161 ssize_t pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6162 UPDATE_SYNTAX_TABLE (pos1);
6164 #endif
6166 int len;
6167 re_wchar_t c;
6169 GET_CHAR_AFTER (c, d, len);
6170 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6171 goto fail;
6172 d += len;
6175 break;
6177 #ifdef emacs
6178 case before_dot:
6179 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6180 if (PTR_BYTE_POS (d) >= PT_BYTE)
6181 goto fail;
6182 break;
6184 case at_dot:
6185 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6186 if (PTR_BYTE_POS (d) != PT_BYTE)
6187 goto fail;
6188 break;
6190 case after_dot:
6191 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6192 if (PTR_BYTE_POS (d) <= PT_BYTE)
6193 goto fail;
6194 break;
6196 case categoryspec:
6197 case notcategoryspec:
6199 boolean not = (re_opcode_t) *(p - 1) == notcategoryspec;
6200 mcnt = *p++;
6201 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n",
6202 not?"not":"", mcnt);
6203 PREFETCH ();
6206 int len;
6207 re_wchar_t c;
6208 GET_CHAR_AFTER (c, d, len);
6209 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6210 goto fail;
6211 d += len;
6214 break;
6216 #endif /* emacs */
6218 default:
6219 abort ();
6221 continue; /* Successfully executed one pattern command; keep going. */
6224 /* We goto here if a matching operation fails. */
6225 fail:
6226 IMMEDIATE_QUIT_CHECK;
6227 if (!FAIL_STACK_EMPTY ())
6229 re_char *str, *pat;
6230 /* A restart point is known. Restore to that state. */
6231 DEBUG_PRINT1 ("\nFAIL:\n");
6232 POP_FAILURE_POINT (str, pat);
6233 switch (*pat++)
6235 case on_failure_keep_string_jump:
6236 assert (str == NULL);
6237 goto continue_failure_jump;
6239 case on_failure_jump_nastyloop:
6240 assert ((re_opcode_t)pat[-2] == no_op);
6241 PUSH_FAILURE_POINT (pat - 2, str);
6242 /* Fallthrough */
6244 case on_failure_jump_loop:
6245 case on_failure_jump:
6246 case succeed_n:
6247 d = str;
6248 continue_failure_jump:
6249 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6250 p = pat + mcnt;
6251 break;
6253 case no_op:
6254 /* A special frame used for nastyloops. */
6255 goto fail;
6257 default:
6258 abort ();
6261 assert (p >= bufp->buffer && p <= pend);
6263 if (d >= string1 && d <= end1)
6264 dend = end_match_1;
6266 else
6267 break; /* Matching at this starting point really fails. */
6268 } /* for (;;) */
6270 if (best_regs_set)
6271 goto restore_best_regs;
6273 FREE_VARIABLES ();
6275 return -1; /* Failure to match. */
6276 } /* re_match_2 */
6278 /* Subroutine definitions for re_match_2. */
6280 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6281 bytes; nonzero otherwise. */
6283 static int
6284 bcmp_translate (const re_char *s1, const re_char *s2, register ssize_t len,
6285 RE_TRANSLATE_TYPE translate, const int target_multibyte)
6287 register re_char *p1 = s1, *p2 = s2;
6288 re_char *p1_end = s1 + len;
6289 re_char *p2_end = s2 + len;
6291 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6292 different lengths, but relying on a single `len' would break this. -sm */
6293 while (p1 < p1_end && p2 < p2_end)
6295 int p1_charlen, p2_charlen;
6296 re_wchar_t p1_ch, p2_ch;
6298 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6299 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6301 if (RE_TRANSLATE (translate, p1_ch)
6302 != RE_TRANSLATE (translate, p2_ch))
6303 return 1;
6305 p1 += p1_charlen, p2 += p2_charlen;
6308 if (p1 != p1_end || p2 != p2_end)
6309 return 1;
6311 return 0;
6314 /* Entry points for GNU code. */
6316 /* re_compile_pattern is the GNU regular expression compiler: it
6317 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6318 Returns 0 if the pattern was valid, otherwise an error string.
6320 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6321 are set in BUFP on entry.
6323 We call regex_compile to do the actual compilation. */
6325 const char *
6326 re_compile_pattern (const char *pattern, size_t length,
6327 struct re_pattern_buffer *bufp)
6329 reg_errcode_t ret;
6331 /* GNU code is written to assume at least RE_NREGS registers will be set
6332 (and at least one extra will be -1). */
6333 bufp->regs_allocated = REGS_UNALLOCATED;
6335 /* And GNU code determines whether or not to get register information
6336 by passing null for the REGS argument to re_match, etc., not by
6337 setting no_sub. */
6338 bufp->no_sub = 0;
6340 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6342 if (!ret)
6343 return NULL;
6344 return gettext (re_error_msgid[(int) ret]);
6346 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6348 /* Entry points compatible with 4.2 BSD regex library. We don't define
6349 them unless specifically requested. */
6351 #if defined _REGEX_RE_COMP || defined _LIBC
6353 /* BSD has one and only one pattern buffer. */
6354 static struct re_pattern_buffer re_comp_buf;
6356 char *
6357 # ifdef _LIBC
6358 /* Make these definitions weak in libc, so POSIX programs can redefine
6359 these names if they don't use our functions, and still use
6360 regcomp/regexec below without link errors. */
6361 weak_function
6362 # endif
6363 re_comp (const char *s)
6365 reg_errcode_t ret;
6367 if (!s)
6369 if (!re_comp_buf.buffer)
6370 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6371 return (char *) gettext ("No previous regular expression");
6372 return 0;
6375 if (!re_comp_buf.buffer)
6377 re_comp_buf.buffer = malloc (200);
6378 if (re_comp_buf.buffer == NULL)
6379 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6380 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6381 re_comp_buf.allocated = 200;
6383 re_comp_buf.fastmap = malloc (1 << BYTEWIDTH);
6384 if (re_comp_buf.fastmap == NULL)
6385 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6386 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6389 /* Since `re_exec' always passes NULL for the `regs' argument, we
6390 don't need to initialize the pattern buffer fields which affect it. */
6392 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6394 if (!ret)
6395 return NULL;
6397 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6398 return (char *) gettext (re_error_msgid[(int) ret]);
6403 # ifdef _LIBC
6404 weak_function
6405 # endif
6406 re_exec (const char *s)
6408 const size_t len = strlen (s);
6409 return
6410 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6412 #endif /* _REGEX_RE_COMP */
6414 /* POSIX.2 functions. Don't define these for Emacs. */
6416 #ifndef emacs
6418 /* regcomp takes a regular expression as a string and compiles it.
6420 PREG is a regex_t *. We do not expect any fields to be initialized,
6421 since POSIX says we shouldn't. Thus, we set
6423 `buffer' to the compiled pattern;
6424 `used' to the length of the compiled pattern;
6425 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6426 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6427 RE_SYNTAX_POSIX_BASIC;
6428 `fastmap' to an allocated space for the fastmap;
6429 `fastmap_accurate' to zero;
6430 `re_nsub' to the number of subexpressions in PATTERN.
6432 PATTERN is the address of the pattern string.
6434 CFLAGS is a series of bits which affect compilation.
6436 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6437 use POSIX basic syntax.
6439 If REG_NEWLINE is set, then . and [^...] don't match newline.
6440 Also, regexec will try a match beginning after every newline.
6442 If REG_ICASE is set, then we considers upper- and lowercase
6443 versions of letters to be equivalent when matching.
6445 If REG_NOSUB is set, then when PREG is passed to regexec, that
6446 routine will report only success or failure, and nothing about the
6447 registers.
6449 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6450 the return codes and their meanings.) */
6452 reg_errcode_t
6453 regcomp (regex_t *__restrict preg, const char *__restrict pattern,
6454 int cflags)
6456 reg_errcode_t ret;
6457 reg_syntax_t syntax
6458 = (cflags & REG_EXTENDED) ?
6459 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6461 /* regex_compile will allocate the space for the compiled pattern. */
6462 preg->buffer = 0;
6463 preg->allocated = 0;
6464 preg->used = 0;
6466 /* Try to allocate space for the fastmap. */
6467 preg->fastmap = malloc (1 << BYTEWIDTH);
6469 if (cflags & REG_ICASE)
6471 unsigned i;
6473 preg->translate = malloc (CHAR_SET_SIZE * sizeof *preg->translate);
6474 if (preg->translate == NULL)
6475 return (int) REG_ESPACE;
6477 /* Map uppercase characters to corresponding lowercase ones. */
6478 for (i = 0; i < CHAR_SET_SIZE; i++)
6479 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6481 else
6482 preg->translate = NULL;
6484 /* If REG_NEWLINE is set, newlines are treated differently. */
6485 if (cflags & REG_NEWLINE)
6486 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6487 syntax &= ~RE_DOT_NEWLINE;
6488 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6490 else
6491 syntax |= RE_NO_NEWLINE_ANCHOR;
6493 preg->no_sub = !!(cflags & REG_NOSUB);
6495 /* POSIX says a null character in the pattern terminates it, so we
6496 can use strlen here in compiling the pattern. */
6497 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6499 /* POSIX doesn't distinguish between an unmatched open-group and an
6500 unmatched close-group: both are REG_EPAREN. */
6501 if (ret == REG_ERPAREN)
6502 ret = REG_EPAREN;
6504 if (ret == REG_NOERROR && preg->fastmap)
6505 { /* Compute the fastmap now, since regexec cannot modify the pattern
6506 buffer. */
6507 re_compile_fastmap (preg);
6508 if (preg->can_be_null)
6509 { /* The fastmap can't be used anyway. */
6510 free (preg->fastmap);
6511 preg->fastmap = NULL;
6514 return ret;
6516 WEAK_ALIAS (__regcomp, regcomp)
6519 /* regexec searches for a given pattern, specified by PREG, in the
6520 string STRING.
6522 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6523 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6524 least NMATCH elements, and we set them to the offsets of the
6525 corresponding matched substrings.
6527 EFLAGS specifies `execution flags' which affect matching: if
6528 REG_NOTBOL is set, then ^ does not match at the beginning of the
6529 string; if REG_NOTEOL is set, then $ does not match at the end.
6531 We return 0 if we find a match and REG_NOMATCH if not. */
6533 reg_errcode_t
6534 regexec (const regex_t *__restrict preg, const char *__restrict string,
6535 size_t nmatch, regmatch_t pmatch[__restrict_arr], int eflags)
6537 regoff_t ret;
6538 struct re_registers regs;
6539 regex_t private_preg;
6540 size_t len = strlen (string);
6541 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6543 private_preg = *preg;
6545 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6546 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6548 /* The user has told us exactly how many registers to return
6549 information about, via `nmatch'. We have to pass that on to the
6550 matching routines. */
6551 private_preg.regs_allocated = REGS_FIXED;
6553 if (want_reg_info)
6555 regs.num_regs = nmatch;
6556 regs.start = TALLOC (nmatch * 2, regoff_t);
6557 if (regs.start == NULL)
6558 return REG_NOMATCH;
6559 regs.end = regs.start + nmatch;
6562 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6563 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6564 was a little bit longer but still only matching the real part.
6565 This works because the `endline' will check for a '\n' and will find a
6566 '\0', correctly deciding that this is not the end of a line.
6567 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6568 a convenient '\0' there. For all we know, the string could be preceded
6569 by '\n' which would throw things off. */
6571 /* Perform the searching operation. */
6572 ret = re_search (&private_preg, string, len,
6573 /* start: */ 0, /* range: */ len,
6574 want_reg_info ? &regs : (struct re_registers *) 0);
6576 /* Copy the register information to the POSIX structure. */
6577 if (want_reg_info)
6579 if (ret >= 0)
6581 unsigned r;
6583 for (r = 0; r < nmatch; r++)
6585 pmatch[r].rm_so = regs.start[r];
6586 pmatch[r].rm_eo = regs.end[r];
6590 /* If we needed the temporary register info, free the space now. */
6591 free (regs.start);
6594 /* We want zero return to mean success, unlike `re_search'. */
6595 return ret >= 0 ? REG_NOERROR : REG_NOMATCH;
6597 WEAK_ALIAS (__regexec, regexec)
6600 /* Returns a message corresponding to an error code, ERR_CODE, returned
6601 from either regcomp or regexec. We don't use PREG here.
6603 ERR_CODE was previously called ERRCODE, but that name causes an
6604 error with msvc8 compiler. */
6606 size_t
6607 regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size)
6609 const char *msg;
6610 size_t msg_size;
6612 if (err_code < 0
6613 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6614 /* Only error codes returned by the rest of the code should be passed
6615 to this routine. If we are given anything else, or if other regex
6616 code generates an invalid error code, then the program has a bug.
6617 Dump core so we can fix it. */
6618 abort ();
6620 msg = gettext (re_error_msgid[err_code]);
6622 msg_size = strlen (msg) + 1; /* Includes the null. */
6624 if (errbuf_size != 0)
6626 if (msg_size > errbuf_size)
6628 memcpy (errbuf, msg, errbuf_size - 1);
6629 errbuf[errbuf_size - 1] = 0;
6631 else
6632 strcpy (errbuf, msg);
6635 return msg_size;
6637 WEAK_ALIAS (__regerror, regerror)
6640 /* Free dynamically allocated space used by PREG. */
6642 void
6643 regfree (regex_t *preg)
6645 free (preg->buffer);
6646 preg->buffer = NULL;
6648 preg->allocated = 0;
6649 preg->used = 0;
6651 free (preg->fastmap);
6652 preg->fastmap = NULL;
6653 preg->fastmap_accurate = 0;
6655 free (preg->translate);
6656 preg->translate = NULL;
6658 WEAK_ALIAS (__regfree, regfree)
6660 #endif /* not emacs */