1 /*************************************************************************
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 * Copyright 2008 by Sun Microsystems, Inc.
7 * OpenOffice.org - a multi-platform office productivity suite
9 * $RCSfile: reclass.cxx,v $
12 * This file is part of OpenOffice.org.
14 * OpenOffice.org is free software: you can redistribute it and/or modify
15 * it under the terms of the GNU Lesser General Public License version 3
16 * only, as published by the Free Software Foundation.
18 * OpenOffice.org is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU Lesser General Public License version 3 for more details
22 * (a copy is included in the LICENSE file that accompanied this code).
24 * You should have received a copy of the GNU Lesser General Public License
25 * version 3 along with OpenOffice.org. If not, see
26 * <http://www.openoffice.org/license.html>
27 * for a copy of the LGPLv3 License.
29 ************************************************************************/
32 // MARKER(update_precomp.py): autogen include statement, do not remove
33 #include "precompiled_regexp.hxx"
34 /* Extended regular expression matching and search library,
36 (Implements POSIX draft P1003.2/D11.2, except for some of the
37 internationalization features.)
38 Copyright (C) 1993, 94, 95, 96, 97, 98, 99 Free Software Foundation, Inc.
40 The GNU C Library is free software; you can redistribute it and/or
41 modify it under the terms of the GNU Library General Public License as
42 published by the Free Software Foundation; either version 2 of the
43 License, or (at your option) any later version.
45 The GNU C Library is distributed in the hope that it will be useful,
46 but WITHOUT ANY WARRANTY; without even the implied warranty of
47 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
48 Library General Public License for more details.
50 You should have received a copy of the GNU Library General Public
51 License along with the GNU C Library; see the file COPYING.LIB. If not,
52 write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
53 Boston, MA 02111-1307, USA. */
56 Modified for OpenOffice.org to use sal_Unicode and Transliteration service.
61 /* If for any reason (porting, debug) we can't use alloca() use malloc()
62 instead. Use alloca() if possible for performance reasons, this _is_
63 significant, with malloc() the re_match2() method makes heavy use of regexps
64 through the TextSearch interface up to three times slower. This is _the_
65 bottleneck in some spreadsheet documents. */
69 /* AIX requires this to be the first thing in the file. */
70 #if defined _AIX && !defined REGEX_MALLOC
77 #include <rtl/ustring.hxx>
78 #include <com/sun/star/i18n/TransliterationModules.hpp>
80 #include "reclass.hxx"
83 /* Maximum number of duplicates an interval can allow. Some systems
84 (erroneously) define this in other header files, but we want our
85 value, so remove any previous define. */
89 /* If sizeof(int) == 2, then ((1 << 15) - 1) overflows. */
90 #define RE_DUP_MAX (0x7fff)
93 /* If `regs_allocated' is REGS_UNALLOCATED in the pattern buffer,
94 `re_match_2' returns information about at least this many registers
95 the first time a `regs' structure is passed. */
102 #define INIT_COMPILE_STACK_SIZE 32
103 #define INIT_BUF_SIZE ((1 << BYTEWIDTH)/BYTEWIDTH)
104 #define MAX_BUF_SIZE 65535L
105 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
106 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
108 /* Since we have one byte reserved for the register number argument to
109 {start,stop}_memory, the maximum number of groups we can report
110 things about is what fits in that byte. */
111 #define MAX_REGNUM 255
113 #define MIN(x, y) ( (x) < (y) ? (x) : (y) )
114 #define MAX(x, y) ( (x) > (y) ? (x) : (y) )
117 // Always. We're not in Emacs and don't use relocating allocators.
118 #define MATCH_MAY_ALLOCATE
120 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
121 use `alloca' instead of `malloc'. This is because malloc is slower and
122 causes storage fragmentation. On the other hand, malloc is more portable,
125 Because we sometimes use alloca, some routines have to be macros,
126 not functions -- `alloca'-allocated space disappears at the end of the
127 function it is called in. */
131 # define REGEX_ALLOCATE malloc
132 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
133 # define REGEX_FREE free
135 #else /* not REGEX_MALLOC */
137 /* Emacs already defines alloca, sometimes. So does MSDEV. */
140 /* Make alloca work the best possible way. */
142 # define alloca __builtin_alloca
143 # else /* not __GNUC__ */
144 # include <sal/alloca.h>
145 # endif /* not __GNUC__ */
147 # endif /* not alloca */
149 # define REGEX_ALLOCATE alloca
151 /* Assumes a `char *destination' variable. */
152 # define REGEX_REALLOCATE(source, osize, nsize) \
153 (destination = (char *) alloca (nsize), \
154 memcpy (destination, source, osize))
156 /* No need to do anything to free, after alloca. */
157 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
159 #endif /* not REGEX_MALLOC */
162 /* Define how to allocate the failure stack. */
166 # define REGEX_ALLOCATE_STACK malloc
167 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
168 # define REGEX_FREE_STACK free
170 #else /* not REGEX_MALLOC */
172 # define REGEX_ALLOCATE_STACK alloca
174 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
175 REGEX_REALLOCATE (source, osize, nsize)
176 /* No need to explicitly free anything. */
177 # define REGEX_FREE_STACK(arg)
179 #endif /* not REGEX_MALLOC */
182 /* (Re)Allocate N items of type T using malloc, or fail. */
183 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
184 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
185 #define RETALLOC_IF(addr, n, t) \
186 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
187 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
189 #define BYTEWIDTH 16 /* In bits (assuming sizeof(sal_Unicode)*8) */
192 #define CHAR_CLASS_MAX_LENGTH 256
194 /* Fetch the next character in the uncompiled pattern, with no
196 #define PATFETCH_RAW(c) \
198 if (p == pend) return REG_EEND; \
199 c = (sal_Unicode) *p++; \
202 /* Go backwards one character in the pattern. */
203 #define PATUNFETCH p--
205 #define FREE_STACK_RETURN(value) \
206 return(free(compile_stack.stack), value)
208 #define GET_BUFFER_SPACE(n) \
209 while ((sal_uInt32)(b - bufp->buffer + (n)) > bufp->allocated) \
212 /* Extend the buffer by twice its current size via realloc and
213 reset the pointers that pointed into the old block to point to the
214 correct places in the new one. If extending the buffer results in it
215 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
216 #define EXTEND_BUFFER() \
218 sal_Unicode *old_buffer = bufp->buffer; \
219 if (bufp->allocated == MAX_BUF_SIZE) \
221 bufp->allocated <<= 1; \
222 if (bufp->allocated > MAX_BUF_SIZE) \
223 bufp->allocated = MAX_BUF_SIZE; \
224 bufp->buffer = (sal_Unicode *) realloc(bufp->buffer, \
226 sizeof(sal_Unicode)); \
227 if (bufp->buffer == NULL) \
229 /* If the buffer moved, move all the pointers into it. */ \
230 if (old_buffer != bufp->buffer) { \
231 b = (b - old_buffer) + bufp->buffer; \
232 begalt = (begalt - old_buffer) + bufp->buffer; \
233 if (fixup_alt_jump) \
234 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
236 laststart = (laststart - old_buffer) + bufp->buffer; \
238 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
242 #define BUF_PUSH(c) \
244 GET_BUFFER_SPACE(1); \
245 *b++ = (sal_Unicode)(c); \
248 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
249 #define BUF_PUSH_2(c1, c2) \
251 GET_BUFFER_SPACE(2); \
252 *b++ = (sal_Unicode) (c1); \
253 *b++ = (sal_Unicode) (c2); \
256 /* As with BUF_PUSH_2, except for three bytes. */
257 #define BUF_PUSH_3(c1, c2, c3) \
259 GET_BUFFER_SPACE(3); \
260 *b++ = (sal_Unicode) (c1); \
261 *b++ = (sal_Unicode) (c2); \
262 *b++ = (sal_Unicode) (c3); \
265 /* Store a jump with opcode OP at LOC to location TO. We store a
266 relative address offset by the three bytes the jump itself occupies. */
267 #define STORE_JUMP(op, loc, to) \
268 store_op1(op, loc, (int) ((to) - (loc) - 3))
270 /* Likewise, for a two-argument jump. */
271 #define STORE_JUMP2(op, loc, to, arg) \
272 store_op2(op, loc, (int) ((to) - (loc) - 3), arg)
274 /* Store NUMBER in two contiguous sal_Unicode starting at DESTINATION. */
278 Regexpr::store_number( sal_Unicode
* destination
, sal_Int32 number
)
280 (destination
)[0] = sal_Unicode((number
) & 0xffff);
281 (destination
)[1] = sal_Unicode((number
) >> 16);
284 /* Same as STORE_NUMBER, except increment DESTINATION to
285 the byte after where the number is stored. Therefore, DESTINATION
286 must be an lvalue. */
290 Regexpr::store_number_and_incr( sal_Unicode
*& destination
, sal_Int32 number
)
292 store_number( destination
, number
);
296 /* Put into DESTINATION a number stored in two contiguous sal_Unicode starting
299 inline void Regexpr::extract_number( sal_Int32
& dest
, sal_Unicode
*source
)
301 dest
= (((sal_Int32
) source
[1]) << 16) | (source
[0] & 0xffff);
304 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
305 #define INSERT_JUMP(op, loc, to) \
306 insert_op1(op, loc, (sal_Int32) ((to) - (loc) - 3), b)
308 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
309 #define INSERT_JUMP2(op, loc, to, arg) \
310 insert_op2(op, loc, (sal_Int32) ((to) - (loc) - 3), arg, b)
312 #define STREQ(s1, s2) (rtl_ustr_compare((s1), (s2)) ? (0) : (1))
314 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
315 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
317 /* The next available element. */
318 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
320 /* Get the next unsigned number in the uncompiled pattern. */
321 #define GET_UNSIGNED_NUMBER(num) { \
324 while (c >= (sal_Unicode)'0' && c <= (sal_Unicode)'9') { \
327 num = num * 10 + c - (sal_Unicode)'0'; \
335 /* Get the next hex number in the uncompiled pattern. */
336 #define GET_HEX_NUMBER(num) { \
338 sal_Bool stop = false; \
339 sal_Int16 hexcnt = 1; \
341 while ( (c >= (sal_Unicode)'0' && c <= (sal_Unicode)'9') || (c >= (sal_Unicode)'a' && c <= (sal_Unicode)'f') || (c >= (sal_Unicode)'A' && c <= (sal_Unicode)'F') ) { \
344 if ( c >= (sal_Unicode)'0' && c <= (sal_Unicode)'9' ) \
345 num = num * 16 + c - (sal_Unicode)'0'; \
346 else if ( c >= (sal_Unicode)'a' && c <= (sal_Unicode)'f' ) \
347 num = num * 16 + (10 + c - (sal_Unicode)'a'); \
349 num = num * 16 + (10 + c - (sal_Unicode)'A'); \
350 if (p == pend || hexcnt == 4) { \
362 if ( hexcnt > 4 || (num < 0 || num > 0xffff) ) num = -1;\
367 /* Number of failure points for which to initially allocate space
368 when matching. If this number is exceeded, we allocate more
369 space, so it is not a hard limit. */
370 #ifndef INIT_FAILURE_ALLOC
371 # define INIT_FAILURE_ALLOC 5
374 #define INIT_FAIL_STACK() \
376 fail_stack.stack = (fail_stack_elt_t *) \
377 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
379 if (fail_stack.stack == NULL) \
382 fail_stack.size = INIT_FAILURE_ALLOC; \
383 fail_stack.avail = 0; \
386 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
388 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
390 Return 1 if succeeds, and 0 if either ran out of memory
391 allocating space for it or it was already too large.
393 REGEX_REALLOCATE_STACK requires `destination' be declared. */
395 #define DOUBLE_FAIL_STACK(fail_stack) \
396 ((fail_stack).size > (sal_uInt32) (re_max_failures * MAX_FAILURE_ITEMS) \
398 : ((fail_stack).stack = (fail_stack_elt_t *) \
399 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
400 (fail_stack).size * sizeof (fail_stack_elt_t), \
401 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
403 (fail_stack).stack == NULL \
405 : ((fail_stack).size <<= 1, \
409 #define REG_UNSET_VALUE (®_unset_dummy)
410 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
412 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
413 #define IS_ACTIVE(R) ((R).bits.is_active)
414 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
415 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
417 /* Call this when have matched a real character; it sets `matched' flags
418 for the subexpressions which we are currently inside. Also records
419 that those subexprs have matched. */
420 #define SET_REGS_MATCHED() \
422 if (!set_regs_matched_done) { \
424 set_regs_matched_done = 1; \
425 for (r = lowest_active_reg; r <= highest_active_reg; r++) { \
426 MATCHED_SOMETHING(reg_info[r]) \
427 = EVER_MATCHED_SOMETHING(reg_info[r]) \
434 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
436 /* This converts PTR, a pointer into the search string `string2' into an offset from the beginning of that string. */
437 #define POINTER_TO_OFFSET(ptr) ((sal_Int32) ((ptr) - string2))
439 /* This is the number of items that are pushed and popped on the stack
440 for each register. */
441 #define NUM_REG_ITEMS 3
443 /* Individual items aside from the registers. */
444 # define NUM_NONREG_ITEMS 4
446 /* We push at most this many items on the stack. */
447 /* We used to use (num_regs - 1), which is the number of registers
448 this regexp will save; but that was changed to 5
449 to avoid stack overflow for a regexp with lots of parens. */
450 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
452 /* We actually push this many items. */
453 #define NUM_FAILURE_ITEMS \
455 ? 0 : highest_active_reg - lowest_active_reg + 1) \
459 /* How many items can still be added to the stack without overflowing it. */
460 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
462 /* Push a pointer value onto the failure stack.
463 Assumes the variable `fail_stack'. Probably should only
464 be called from within `PUSH_FAILURE_POINT'. */
465 #define PUSH_FAILURE_POINTER(item) \
466 fail_stack.stack[fail_stack.avail++].pointer = (sal_Unicode *) (item)
468 /* This pushes an integer-valued item onto the failure stack.
469 Assumes the variable `fail_stack'. Probably should only
470 be called from within `PUSH_FAILURE_POINT'. */
471 #define PUSH_FAILURE_INT(item) \
472 fail_stack.stack[fail_stack.avail++].integer = (item)
474 /* Push a fail_stack_elt_t value onto the failure stack.
475 Assumes the variable `fail_stack'. Probably should only
476 be called from within `PUSH_FAILURE_POINT'. */
477 #define PUSH_FAILURE_ELT(item) \
478 fail_stack.stack[fail_stack.avail++] = (item)
480 /* These three POP... operations complement the three PUSH... operations.
481 All assume that `fail_stack' is nonempty. */
482 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
483 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
484 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
486 /* Test if at very beginning or at very end of `string2'. */
487 #define AT_STRINGS_BEG(d) ((d) == string2 || !size2)
488 #define AT_STRINGS_END(d) ((d) == end2)
490 /* Checking for end of string */
500 Regexpr::iswordbegin(const sal_Unicode
*d
, sal_Unicode
*string
, sal_Int32 ssize
)
502 if ( d
== string
|| ! ssize
) return true;
504 if ( !unicode::isAlphaDigit(d
[-1]) && unicode::isAlphaDigit(d
[0])) {
511 Regexpr::iswordend(const sal_Unicode
*d
, sal_Unicode
*string
, sal_Int32 ssize
)
513 if ( d
== (string
+ssize
) ) return true;
515 if ( !unicode::isAlphaDigit(d
[0]) && unicode::isAlphaDigit(d
[-1])) {
521 /* Push the information about the state we will need
522 if we ever fail back to it.
524 Requires variables fail_stack, regstart, regend, and reg_info
525 be declared. DOUBLE_FAIL_STACK requires `destination'
528 Does `return FAILURE_CODE' if runs out of memory. */
530 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
533 /* Must be int, so when we don't save any registers, the arithmetic \
534 of 0 + -1 isn't done as unsigned. */ \
535 /* Can't be int, since there is not a shred of a guarantee that int \
536 is wide enough to hold a value of something to which pointer can \
538 sal_uInt32 this_reg; \
540 /* Ensure we have enough space allocated for what we will push. */ \
541 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) { \
542 if (!DOUBLE_FAIL_STACK(fail_stack)) \
543 return failure_code; \
546 /* Push the info, starting with the registers. */ \
548 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
550 PUSH_FAILURE_POINTER(regstart[this_reg]); \
552 PUSH_FAILURE_POINTER (regend[this_reg]); \
554 PUSH_FAILURE_ELT(reg_info[this_reg].word); \
557 PUSH_FAILURE_INT(lowest_active_reg); \
559 PUSH_FAILURE_INT(highest_active_reg); \
561 PUSH_FAILURE_POINTER(pattern_place); \
563 PUSH_FAILURE_POINTER(string_place); \
567 /* Pops what PUSH_FAIL_STACK pushes.
569 We restore into the parameters, all of which should be lvalues:
570 STR -- the saved data position.
571 PAT -- the saved pattern position.
572 LOW_REG, HIGH_REG -- the highest and lowest active registers.
573 REGSTART, REGEND -- arrays of string positions.
574 REG_INFO -- array of information about each subexpression.
576 Also assumes the variables `fail_stack' and (if debugging), `bufp',
577 `pend', `string2', and `size2'. */
579 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info) {\
580 sal_uInt32 this_reg; \
581 sal_Unicode *string_temp; \
583 assert(!FAIL_STACK_EMPTY()); \
585 /* Remove failure points and point to how many regs pushed. */ \
586 assert(fail_stack.avail >= NUM_NONREG_ITEMS); \
588 /* If the saved string location is NULL, it came from an \
589 on_failure_keep_string_jump opcode, and we want to throw away the \
590 saved NULL, thus retaining our current position in the string. */ \
591 string_temp = POP_FAILURE_POINTER(); \
592 if (string_temp != NULL) \
593 str = (const sal_Unicode *) string_temp; \
595 pat = (sal_Unicode *) POP_FAILURE_POINTER(); \
597 /* Restore register info. */ \
598 high_reg = (sal_uInt32) POP_FAILURE_INT(); \
600 low_reg = (sal_uInt32) POP_FAILURE_INT(); \
603 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) { \
605 reg_info[this_reg].word = POP_FAILURE_ELT(); \
607 regend[this_reg] = (const sal_Unicode *) POP_FAILURE_POINTER(); \
609 regstart[this_reg] = (const sal_Unicode *) POP_FAILURE_POINTER(); \
611 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) {\
612 reg_info[this_reg].word.integer = 0; \
613 regend[this_reg] = 0; \
614 regstart[this_reg] = 0; \
616 highest_active_reg = high_reg; \
619 set_regs_matched_done = 0; \
620 } /* POP_FAILURE_POINT */
624 Regexpr::extract_number_and_incr( sal_Int32
& destination
, sal_Unicode
*& source
)
626 extract_number(destination
, source
);
633 Regexpr::store_op1(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg
)
635 *loc
= (sal_Unicode
) op
;
636 store_number(loc
+ 1, arg
);
639 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
643 Regexpr::store_op2(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg1
, sal_Int32 arg2
)
645 *loc
= (sal_Unicode
) op
;
646 store_number(loc
+ 1, arg1
);
647 store_number(loc
+ 3, arg2
);
651 Regexpr::insert_op1(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg
, sal_Unicode
*end
)
653 register sal_Unicode
*pfrom
= end
;
654 register sal_Unicode
*pto
= end
+ 3;
656 while (pfrom
!= loc
) {
660 store_op1(op
, loc
, arg
);
664 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
667 Regexpr::insert_op2(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg1
, sal_Int32 arg2
, sal_Unicode
*end
)
669 register sal_Unicode
*pfrom
= end
;
670 register sal_Unicode
*pto
= end
+ 5;
675 store_op2 (op
, loc
, arg1
, arg2
);
678 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
679 after an alternative or a begin-subexpression. We assume there is at
680 least one character before the ^. */
683 Regexpr::at_begline_loc_p(const sal_Unicode
*local_pattern
, const sal_Unicode
*p
)
685 const sal_Unicode
*prev
= p
- 2;
686 sal_Bool prev_prev_backslash
= prev
> local_pattern
&& prev
[-1] == '\\';
689 /* After a subexpression? */
690 (*prev
== (sal_Unicode
)'(' && prev_prev_backslash
)
691 /* After an alternative? */
692 || (*prev
== (sal_Unicode
)'|' && prev_prev_backslash
));
695 /* The dual of at_begline_loc_p. This one is for $. We assume there is
696 at least one character after the $, i.e., `P < PEND'. */
699 Regexpr::at_endline_loc_p(const sal_Unicode
*p
, const sal_Unicode
* /* pend */ )
701 const sal_Unicode
*next
= p
;
702 //sal_Bool next_backslash = *next == (sal_Unicode)'\\';
703 //const sal_Unicode *next_next = p + 1 < pend ? p + 1 : 0;
706 /* Before a subexpression? */
707 *next
== (sal_Unicode
)')'
708 // (next_backslash && next_next && *next_next == (sal_Unicode)')')
709 /* Before an alternative? */
710 || *next
== (sal_Unicode
)'|' );
711 // || (next_backslash && next_next && *next_next == (sal_Unicode)'|'));
715 Regexpr::compile_range(sal_Unicode range_start
, sal_Unicode range_end
, sal_Unicode
*b
)
717 sal_uInt32 this_char
;
719 /* If the start is after the end, the range is empty. */
720 if (range_start
> range_end
)
723 /* Here we see why `this_char' has to be larger than an `sal_Unicode'
724 -- the range is inclusive, so if `range_end' == 0xffff
725 (assuming 16-bit characters), we would otherwise go into an infinite
726 loop, since all characters <= 0xffff. */
727 for (this_char
= range_start
; this_char
<= range_end
; this_char
++) {
728 set_list_bit( sal_Unicode(this_char
), b
);
734 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
735 false if it's not. */
738 Regexpr::group_in_compile_stack(compile_stack_type compile_stack
, sal_uInt32 regnum
)
740 sal_Int32 this_element
;
742 for (this_element
= compile_stack
.avail
- 1;
745 if (compile_stack
.stack
[this_element
].regnum
== regnum
) {
754 Regexpr::Regexpr( const ::com::sun::star::util::SearchOptions
& rOptions
,
755 ::com::sun::star::uno::Reference
<
756 ::com::sun::star::i18n::XExtendedTransliteration
> XTrans
)
761 if ( rOptions
.algorithmType
!= ::com::sun::star::util::SearchAlgorithms_REGEXP
) {
765 if ( rOptions
.searchString
== NULL
||
766 rOptions
.searchString
.getLength() <= 0) {
770 pattern
= (sal_Unicode
*)rOptions
.searchString
.getStr();
771 patsize
= rOptions
.searchString
.getLength();
773 re_max_failures
= 2000;
776 translate
= translit
.is() ? 1 : 0;
780 isIgnoreCase
= ((rOptions
.transliterateFlags
&
781 ::com::sun::star::i18n::TransliterationModules_IGNORE_CASE
) != 0);
783 // Compile Regular expression pattern
784 if ( regcomp() != REG_NOERROR
)
801 // translit->remove();
815 // sets a new line to search in (restore start/end_ptr)
817 Regexpr::set_line(const sal_Unicode
*new_line
, sal_Int32 len
)
823 // main function for searching the pattern
824 // returns negative or startpos and sets regs
826 Regexpr::re_search(struct re_registers
*regs
, sal_Int32 pOffset
)
828 // Check if pattern buffer is NULL
829 if ( bufp
== NULL
) {
840 linelen
= -(linelen
);
841 stoppos
= pOffset
+ 1;
847 sal_Int32 val
= re_match2(regs
, startpos
, stoppos
);
855 // Return success if match found
864 // If match only beginning of string (startpos)
869 // If search match from startpos to startpos+range
870 else if (range
> 0) { // Forward string search
873 } else { // Reverse string search
879 if ( regs
->num_of_match
> 0 )
888 bufp
= (struct re_pattern_buffer
*)malloc(sizeof(struct re_pattern_buffer
));
889 if ( bufp
== NULL
) {
897 //bufp->fastmap = (sal_Unicode*) malloc((1 << BYTEWIDTH) * sizeof(sal_Unicode));
898 // No fastmap with Unicode
899 bufp
->fastmap
= NULL
;
901 return(regex_compile());
905 Regexpr::regex_compile()
907 register sal_Unicode c
, c1
;
908 const sal_Unicode
*p1
;
909 register sal_Unicode
*b
;
911 /* Keeps track of unclosed groups. */
912 compile_stack_type compile_stack
;
914 /* Points to the current (ending) position in the pattern. */
915 const sal_Unicode
*p
= pattern
;
916 const sal_Unicode
*pend
= pattern
+ patsize
;
918 /* Address of the count-byte of the most recently inserted `exactn'
919 command. This makes it possible to tell if a new exact-match
920 character can be added to that command or if the character requires
921 a new `exactn' command. */
922 sal_Unicode
*pending_exact
= 0;
924 /* Address of start of the most recently finished expression.
925 This tells, e.g., postfix * where to find the start of its
926 operand. Reset at the beginning of groups and alternatives. */
927 sal_Unicode
*laststart
= 0;
929 /* Address of beginning of regexp, or inside of last group. */
932 /* Place in the uncompiled pattern (i.e., the {) to
933 which to go back if the interval is invalid. */
934 const sal_Unicode
*beg_interval
;
936 /* Address of the place where a forward jump should go to the end of
937 the containing expression. Each alternative of an `or' -- except the
938 last -- ends with a forward jump of this sort. */
939 sal_Unicode
*fixup_alt_jump
= 0;
941 /* Counts open-groups as they are encountered. Remembered for the
942 matching close-group on the compile stack, so the same register
943 number is put in the stop_memory as the start_memory. */
944 sal_Int32 regnum
= 0;
946 /* Initialize the compile stack. */
947 compile_stack
.stack
= (compile_stack_elt_t
*)malloc(INIT_COMPILE_STACK_SIZE
* sizeof(compile_stack_elt_t
));
948 if (compile_stack
.stack
== NULL
)
951 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
952 compile_stack
.avail
= 0;
954 /* Initialize the pattern buffer. */
955 bufp
->fastmap_accurate
= 0;
958 bufp
->newline_anchor
= 1;
960 /* Set `used' to zero, so that if we return an error, the pattern
961 printer (for debugging) will think there's no pattern. We reset it
965 /* Always count groups. */
968 if (bufp
->allocated
== 0) {
970 /* If zero allocated, but buffer is non-null, try to realloc
971 enough space. This loses if buffer's address is bogus, but
972 that is the user's responsibility. */
973 bufp
->buffer
= (sal_Unicode
*)realloc(bufp
->buffer
, INIT_BUF_SIZE
* sizeof(sal_Unicode
));
974 } else { /* Caller did not allocate a buffer. Do it for them. */
975 bufp
->buffer
= (sal_Unicode
*)malloc(INIT_BUF_SIZE
* sizeof(sal_Unicode
));
977 if (!bufp
->buffer
) FREE_STACK_RETURN(REG_ESPACE
);
979 bufp
->allocated
= INIT_BUF_SIZE
;
982 begalt
= b
= bufp
->buffer
;
984 /* Loop through the uncompiled pattern until we're at the end. */
989 case (sal_Unicode
)'^': {
990 if ( /* If at start of pattern, it's an operator. */
992 /* Otherwise, depends on what's come before. */
993 || at_begline_loc_p(pattern
, p
))
1000 case (sal_Unicode
)'$': {
1001 if ( /* If at end of pattern, it's an operator. */
1003 /* Otherwise, depends on what's next. */
1004 || at_endline_loc_p(p
, pend
)) {
1012 case (sal_Unicode
)'+':
1013 case (sal_Unicode
)'?':
1014 case (sal_Unicode
)'*':
1015 /* If there is no previous pattern... */
1021 /* Are we optimizing this jump? */
1022 sal_Bool keep_string_p
= false;
1024 /* 1 means zero (many) matches is allowed. */
1025 sal_Unicode zero_times_ok
= 0, many_times_ok
= 0;
1027 /* If there is a sequence of repetition chars, collapse it
1028 down to just one (the right one). We can't combine
1029 interval operators with these because of, e.g., `a{2}*',
1030 which should only match an even number of `a's. */
1033 zero_times_ok
|= c
!= (sal_Unicode
)'+';
1034 many_times_ok
|= c
!= (sal_Unicode
)'?';
1041 if (c
== (sal_Unicode
)'*' || (c
== (sal_Unicode
)'+'
1042 || c
== (sal_Unicode
)'?')) {
1048 /* If we get here, we found another repeat character. */
1051 /* Star, etc. applied to an empty pattern is equivalent
1052 to an empty pattern. */
1057 /* Now we know whether or not zero matches is allowed
1058 and also whether or not two or more matches is allowed. */
1059 if (many_times_ok
) {
1060 /* More than one repetition is allowed, so put in at the
1061 end a backward relative jump from `b' to before the next
1062 jump we're going to put in below (which jumps from
1063 laststart to after this jump).
1065 But if we are at the `*' in the exact sequence `.*\n',
1066 insert an unconditional jump backwards to the .,
1067 instead of the beginning of the loop. This way we only
1068 push a failure point once, instead of every time
1069 through the loop. */
1070 assert(p
- 1 > pattern
);
1072 /* Allocate the space for the jump. */
1073 GET_BUFFER_SPACE(3);
1075 /* We know we are not at the first character of the pattern,
1076 because laststart was nonzero. And we've already
1077 incremented `p', by the way, to be the character after
1078 the `*'. Do we have to do something analogous here
1079 for null bytes, because of RE_DOT_NOT_NULL? */
1080 if (*(p
- 2) == (sal_Unicode
)'.'
1082 && p
< pend
&& *p
== (sal_Unicode
)'\n') {
1084 STORE_JUMP(jump
, b
, laststart
);
1085 keep_string_p
= true;
1087 /* Anything else. */
1088 STORE_JUMP(maybe_pop_jump
, b
, laststart
- 3);
1091 /* We've added more stuff to the buffer. */
1095 /* On failure, jump from laststart to b + 3, which will be the
1096 end of the buffer after this jump is inserted. */
1097 GET_BUFFER_SPACE(3);
1098 INSERT_JUMP(keep_string_p
? on_failure_keep_string_jump
1104 if (!zero_times_ok
) {
1105 /* At least one repetition is required, so insert a
1106 `dummy_failure_jump' before the initial
1107 `on_failure_jump' instruction of the loop. This
1108 effects a skip over that instruction the first time
1109 we hit that loop. */
1110 GET_BUFFER_SPACE(3);
1111 INSERT_JUMP(dummy_failure_jump
, laststart
, laststart
+ 6);
1117 case (sal_Unicode
)'.':
1123 case (sal_Unicode
)'[': {
1124 sal_Bool have_range
= false;
1125 sal_Unicode last_char
= 0xffff;
1126 sal_Unicode first_range
= 0xffff;
1127 sal_Unicode second_range
= 0xffff;
1130 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1132 /* Ensure that we have enough space to push a charset: the
1133 opcode, the length count, and the bitset;
1134 1 + 1 + (1 << BYTEWIDTH) / BYTEWIDTH "bytes" in all. */
1135 bsiz
= 2 + ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1136 GET_BUFFER_SPACE(bsiz
);
1140 /* We test `*p == '^' twice, instead of using an if
1141 statement, so we only need one BUF_PUSH. */
1142 BUF_PUSH (*p
== (sal_Unicode
)'^' ? charset_not
: charset
);
1143 if (*p
== (sal_Unicode
)'^')
1146 /* Remember the first position in the bracket expression. */
1149 /* Push the number of "bytes" in the bitmap. */
1150 BUF_PUSH((1 << BYTEWIDTH
) / BYTEWIDTH
);
1152 /* Clear the whole map. */
1153 memset(b
, 0, ((1 << BYTEWIDTH
) / BYTEWIDTH
) * sizeof(sal_Unicode
));
1155 /* Read in characters and ranges, setting map bits. */
1157 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1161 if ( c
== (sal_Unicode
)'\\' ) {
1165 if ( c
== (sal_Unicode
)'x' ) {
1166 sal_Int32 UniChar
= -1;
1168 GET_HEX_NUMBER(UniChar
);
1169 if (UniChar
< 0 || UniChar
> 0xffff) FREE_STACK_RETURN(REG_BADPAT
);
1170 c
= (sal_Unicode
) UniChar
;
1172 set_list_bit(last_char
, b
);
1175 set_list_bit(last_char
, b
);
1177 } else if (c
== (sal_Unicode
)']') {
1178 /* Could be the end of the bracket expression. If it's
1179 not (i.e., when the bracket expression is `[]' so
1180 far), the ']' character bit gets set way below. */
1182 } else if ( c
== (sal_Unicode
)'-' ) {
1183 if ( !have_range
) {
1184 if ( last_char
!= 0xffff ) {
1185 first_range
= last_char
;
1189 last_char
= (sal_Unicode
)'-';
1190 set_list_bit(last_char
, b
);
1195 /* See if we're at the beginning of a possible character
1197 else if (c
== (sal_Unicode
)':' && p
[-2] == (sal_Unicode
)'[') {
1198 /* Leave room for the null. */
1199 sal_Unicode str
[CHAR_CLASS_MAX_LENGTH
+ 1];
1204 /* If pattern is `[[:'. */
1205 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1210 if ((c
== (sal_Unicode
)':' && *p
== (sal_Unicode
)']') || p
== pend
)
1212 if (c1
< CHAR_CLASS_MAX_LENGTH
)
1215 /* This is in any case an invalid class name. */
1216 str
[0] = (sal_Unicode
)'\0';
1218 str
[c1
] = (sal_Unicode
)'\0';
1220 /* If isn't a word bracketed by `[:' and `:]':
1221 undo the ending character, the letters, and leave
1222 the leading `:' and `[' (but set bits for them). */
1223 if (c
== (sal_Unicode
)':' && *p
== (sal_Unicode
)']') {
1225 // no support for GRAPH, PUNCT, or XDIGIT yet
1226 sal_Bool is_alnum
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"alnum").getStr());
1227 sal_Bool is_alpha
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"alpha").getStr());
1228 sal_Bool is_cntrl
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"cntrl").getStr());
1229 sal_Bool is_digit
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"digit").getStr());
1230 sal_Bool is_lower
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"lower").getStr());
1231 sal_Bool is_print
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"print").getStr());
1232 sal_Bool is_space
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"space").getStr());
1233 sal_Bool is_upper
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"upper").getStr());
1235 if (!(is_alnum
|| is_alpha
|| is_cntrl
||
1236 is_digit
|| is_lower
|| is_print
|| is_space
|| is_upper
) )
1237 FREE_STACK_RETURN(REG_ECTYPE
);
1239 /* Throw away the ] at the end of the character
1243 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1245 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++) {
1246 /* This was split into 3 if's to
1247 avoid an arbitrary limit in some compiler. */
1248 if ( (is_alnum
&& unicode::isAlphaDigit(sal_Unicode(ch
))) ||
1249 (is_alpha
&& unicode::isAlpha(sal_Unicode(ch
))) ||
1250 (is_cntrl
&& unicode::isControl(sal_Unicode(ch
))))
1251 set_list_bit(sal_Unicode(ch
), b
);
1252 if ( (is_digit
&& unicode::isDigit(sal_Unicode(ch
))) ||
1253 (is_lower
&& unicode::isLower(sal_Unicode(ch
))) ||
1254 (is_print
&& unicode::isPrint(sal_Unicode(ch
))))
1255 set_list_bit(sal_Unicode(ch
), b
);
1256 if ( (is_space
&& unicode::isSpace(sal_Unicode(ch
))) ||
1257 (is_upper
&& unicode::isUpper(sal_Unicode(ch
))) )
1258 set_list_bit(sal_Unicode(ch
), b
);
1259 if ( isIgnoreCase
&& (is_upper
|| is_lower
) &&
1260 (unicode::isUpper(sal_Unicode(ch
)) || unicode::isLower(sal_Unicode(ch
))))
1261 set_list_bit(sal_Unicode(ch
), b
);
1266 last_char
= (sal_Unicode
)':';
1267 set_list_bit(last_char
, b
);
1271 set_list_bit(last_char
, b
);
1274 if ( last_char
!= 0xffff ) {
1275 second_range
= last_char
;
1277 compile_range(first_range
, second_range
, b
);
1278 } else FREE_STACK_RETURN(REG_EBRACK
);
1280 if ( last_char
!= 0xffff ) {
1281 set_list_bit(last_char
, b
);
1282 } else FREE_STACK_RETURN(REG_EBRACK
);
1286 /* Discard any (non)matching list bytes that are all 0 at the
1287 end of the map. Decrease the map-length byte too. */
1289 while ((sal_Int16
) bsiz
> 0 && b
[bsiz
- 1] == 0)
1291 b
[-1] = (sal_Unicode
)bsiz
;
1296 case (sal_Unicode
)'(':
1299 case (sal_Unicode
)')':
1302 case (sal_Unicode
)'\n':
1305 case (sal_Unicode
)'|':
1308 case (sal_Unicode
)'{':
1309 goto handle_interval
;
1311 case (sal_Unicode
)'\\':
1312 if (p
== pend
) FREE_STACK_RETURN(REG_EESCAPE
);
1314 /* Do not translate the character after the \, so that we can
1315 distinguish, e.g., \B from \b, even if we normally would
1316 translate, e.g., B to b. */
1320 case (sal_Unicode
)'(':
1321 goto normal_backslash
;
1327 if (COMPILE_STACK_FULL
) {
1328 compile_stack
.stack
= (compile_stack_elt_t
*)realloc(compile_stack
.stack
, (compile_stack
.size
<< 1) * sizeof(compile_stack_elt_t
));
1329 if (compile_stack
.stack
== NULL
) return(REG_ESPACE
);
1331 compile_stack
.size
<<= 1;
1334 /* These are the values to restore when we hit end of this
1335 group. They are all relative offsets, so that if the
1336 whole pattern moves because of realloc, they will still
1338 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
1339 COMPILE_STACK_TOP
.fixup_alt_jump
1340 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
1341 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
1342 COMPILE_STACK_TOP
.regnum
= regnum
;
1344 /* We will eventually replace the 0 with the number of
1345 groups inner to this one. But do not push a
1346 start_memory for groups beyond the last one we can
1347 represent in the compiled pattern. */
1348 if (regnum
<= MAX_REGNUM
) {
1349 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
1350 BUF_PUSH_3 (start_memory
, regnum
, 0);
1353 compile_stack
.avail
++;
1358 /* If we've reached MAX_REGNUM groups, then this open
1359 won't actually generate any code, so we'll have to
1360 clear pending_exact explicitly. */
1365 case (sal_Unicode
)')':
1366 goto normal_backslash
;
1368 // unreachable (after goto):
1370 if (COMPILE_STACK_EMPTY
) {
1371 FREE_STACK_RETURN(REG_ERPAREN
);
1376 if (fixup_alt_jump
) {
1377 /* Push a dummy failure point at the end of the
1378 alternative for a possible future
1379 `pop_failure_jump' to pop. See comments at
1380 `push_dummy_failure' in `re_match2'. */
1381 BUF_PUSH(push_dummy_failure
);
1383 /* We allocated space for this jump when we assigned
1384 to `fixup_alt_jump', in the `handle_alt' case below. */
1385 STORE_JUMP(jump_past_alt
, fixup_alt_jump
, b
- 1);
1388 /* See similar code for backslashed left paren above. */
1389 if (COMPILE_STACK_EMPTY
) {
1390 FREE_STACK_RETURN(REG_ERPAREN
);
1393 /* Since we just checked for an empty stack above, this
1394 ``can't happen''. */
1395 assert (compile_stack
.avail
!= 0);
1398 /* We don't just want to restore into `regnum', because
1399 later groups should continue to be numbered higher,
1400 as in `(ab)c(de)' -- the second group is #2. */
1401 sal_Int32 this_group_regnum
;
1403 compile_stack
.avail
--;
1404 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
1406 = COMPILE_STACK_TOP
.fixup_alt_jump
1407 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
1409 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
1410 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
1411 /* If we've reached MAX_REGNUM groups, then this open
1412 won't actually generate any code, so we'll have to
1413 clear pending_exact explicitly. */
1416 /* We're at the end of the group, so now we know how many
1417 groups were inside this one. */
1418 if (this_group_regnum
<= MAX_REGNUM
) {
1419 sal_Unicode
*inner_group_loc
1420 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
1422 *inner_group_loc
= sal::static_int_cast
<sal_Unicode
>( regnum
- this_group_regnum
);
1423 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
1424 regnum
- this_group_regnum
);
1430 case (sal_Unicode
)'|': /* `\|'.
1432 goto normal_backslash
;
1435 /* Insert before the previous alternative a jump which
1436 jumps to this alternative if the former fails. */
1437 GET_BUFFER_SPACE (3);
1438 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
1442 /* The alternative before this one has a jump after it
1443 which gets executed if it gets matched. Adjust that
1444 jump so it will jump to this alternative's analogous
1445 jump (put in below, which in turn will jump to the next
1446 (if any) alternative's such jump, etc.). The last such
1447 jump jumps to the correct final destination. A picture:
1453 If we are at `b', then fixup_alt_jump right now points to a
1454 three-byte space after `a'. We'll put in the jump, set
1455 fixup_alt_jump to right after `b', and leave behind three
1456 bytes which we'll fill in when we get to after `c'. */
1459 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
1461 /* Mark and leave space for a jump after this alternative,
1462 to be filled in later either by next alternative or
1463 when know we're at the end of a series of alternatives. */
1465 GET_BUFFER_SPACE (3);
1473 case (sal_Unicode
)'{':
1474 goto normal_backslash
;
1478 /* allows intervals. */
1479 /* At least (most) this many matches must be made. */
1480 sal_Int32 lower_bound
= -1, upper_bound
= -1;
1482 beg_interval
= p
- 1;
1485 goto unfetch_interval
;
1488 GET_UNSIGNED_NUMBER(lower_bound
);
1490 if (c
== (sal_Unicode
)',') {
1491 GET_UNSIGNED_NUMBER(upper_bound
);
1492 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
1494 /* Interval such as `{1}' => match exactly once. */
1495 upper_bound
= lower_bound
;
1497 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
1498 || lower_bound
> upper_bound
) {
1499 goto unfetch_interval
;
1502 if (c
!= (sal_Unicode
)'}') {
1503 goto unfetch_interval
;
1506 /* We just parsed a valid interval. */
1508 /* If it's invalid to have no preceding re. */
1510 goto unfetch_interval
;
1513 /* If the upper bound is zero, don't want to succeed at
1514 all; jump from `laststart' to `b + 3', which will be
1515 the end of the buffer after we insert the jump. */
1516 if (upper_bound
== 0) {
1517 GET_BUFFER_SPACE(3);
1518 INSERT_JUMP(jump
, laststart
, b
+ 3);
1522 /* Otherwise, we have a nontrivial interval. When
1523 we're all done, the pattern will look like:
1524 set_number_at <jump count> <upper bound>
1525 set_number_at <succeed_n count> <lower bound>
1526 succeed_n <after jump addr> <succeed_n count>
1528 jump_n <succeed_n addr> <jump count>
1529 (The upper bound and `jump_n' are omitted if
1530 `upper_bound' is 1, though.) */
1532 /* If the upper bound is > 1, we need to insert
1533 more at the end of the loop. */
1534 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
1536 GET_BUFFER_SPACE(nbytes
);
1538 /* Initialize lower bound of the `succeed_n', even
1539 though it will be set during matching by its
1540 attendant `set_number_at' (inserted next),
1541 because `re_compile_fastmap' needs to know.
1542 Jump to the `jump_n' we might insert below. */
1543 INSERT_JUMP2(succeed_n
, laststart
,
1544 b
+ 5 + (upper_bound
> 1) * 5,
1548 /* Code to initialize the lower bound. Insert
1549 before the `succeed_n'. The `5' is the last two
1550 bytes of this `set_number_at', plus 3 bytes of
1551 the following `succeed_n'. */
1552 insert_op2(set_number_at
, laststart
, 5, lower_bound
, b
);
1555 if (upper_bound
> 1) {
1556 /* More than one repetition is allowed, so
1557 append a backward jump to the `succeed_n'
1558 that starts this interval.
1560 When we've reached this during matching,
1561 we'll have matched the interval once, so
1562 jump back only `upper_bound - 1' times. */
1563 STORE_JUMP2(jump_n
, b
, laststart
+ 5,
1567 /* The location we want to set is the second
1568 parameter of the `jump_n'; that is `b-2' as
1569 an absolute address. `laststart' will be
1570 the `set_number_at' we're about to insert;
1571 `laststart+3' the number to set, the source
1572 for the relative address. But we are
1573 inserting into the middle of the pattern --
1574 so everything is getting moved up by 5.
1575 Conclusion: (b - 2) - (laststart + 3) + 5,
1576 i.e., b - laststart.
1578 We insert this at the beginning of the loop
1579 so that if we fail during matching, we'll
1580 reinitialize the bounds. */
1581 insert_op2(set_number_at
, laststart
, b
- laststart
,
1582 upper_bound
- 1, b
);
1587 beg_interval
= NULL
;
1592 /* If an invalid interval, match the characters as literals. */
1593 assert (beg_interval
);
1595 beg_interval
= NULL
;
1597 /* normal_char and normal_backslash need `c'. */
1602 case (sal_Unicode
)'`':
1606 case (sal_Unicode
)'\'':
1610 case (sal_Unicode
)'1': case (sal_Unicode
)'2':
1611 case (sal_Unicode
)'3': case (sal_Unicode
)'4':
1612 case (sal_Unicode
)'5': case (sal_Unicode
)'6':
1613 case (sal_Unicode
)'7': case (sal_Unicode
)'8':
1614 case (sal_Unicode
)'9':
1615 c1
= c
- (sal_Unicode
)'0';
1618 FREE_STACK_RETURN(REG_ESUBREG
);
1620 /* Can't back reference to a subexpression if inside of it. */
1621 if (group_in_compile_stack(compile_stack
, (sal_uInt32
) c1
)) {
1626 BUF_PUSH_2(duplicate
, c1
);
1630 case (sal_Unicode
)'+':
1631 case (sal_Unicode
)'?':
1632 goto normal_backslash
;
1634 case (sal_Unicode
)'x': // Unicode char
1636 sal_Int32 UniChar
= -1;
1638 GET_HEX_NUMBER(UniChar
);
1639 if (UniChar
< 0 || UniChar
> 0xffff) FREE_STACK_RETURN(REG_BADPAT
);
1640 c
= (sal_Unicode
) UniChar
;
1643 // break; // unreachable - see goto above
1645 case (sal_Unicode
)'<': // begin Word boundary
1649 case (sal_Unicode
)'>': // end Word boundary
1653 case (sal_Unicode
)'n':
1657 case (sal_Unicode
)'t':
1668 /* Expects the character in `c'. */
1670 /* If no exactn currently being built. */
1671 if ( pending_exact
== NULL
1673 /* If last exactn not at current position. */
1674 || pending_exact
+ *pending_exact
+ 1 != b
1676 /* We have only one sal_Unicode char following the
1677 exactn for the count. */
1678 || *pending_exact
== (1 << BYTEWIDTH
) - 1
1680 /* If followed by a repetition operator. */
1681 || *p
== (sal_Unicode
)'*' || *p
== (sal_Unicode
)'^'
1682 || *p
== (sal_Unicode
)'+' || *p
== (sal_Unicode
)'?'
1683 || *p
== (sal_Unicode
) '{' ) {
1684 /* Start building a new exactn. */
1686 BUF_PUSH_2(exactn
, 0);
1687 pending_exact
= b
- 1;
1692 sal_Unicode tmp
= translit
->transliterateChar2Char(c
);
1695 } catch (::com::sun::star::i18n::MultipleCharsOutputException e
) {
1696 ::rtl::OUString
o2( translit
->transliterateChar2String( c
));
1697 sal_Int32 len2
= o2
.getLength();
1698 const sal_Unicode
* k2
= o2
.getStr();
1699 for (sal_Int32 nmatch
= 0; nmatch
< len2
; nmatch
++) {
1700 BUF_PUSH(k2
[nmatch
]);
1710 } /* while p != pend */
1712 /* Through the pattern now. */
1715 STORE_JUMP(jump_past_alt
, fixup_alt_jump
, b
);
1717 if (!COMPILE_STACK_EMPTY
)
1718 FREE_STACK_RETURN(REG_EPAREN
);
1720 // Assumes no backtracking
1723 if ( compile_stack
.stack
)
1724 free(compile_stack
.stack
);
1725 compile_stack
.stack
= NULL
;
1727 /* We have succeeded; set the length of the buffer. */
1728 bufp
->used
= b
- bufp
->buffer
;
1731 } /* regex_compile */
1733 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
1734 bytes; nonzero otherwise. */
1737 Regexpr::bcmp_translate(const sal_Unicode
*s1
, const sal_Unicode
*s2
, sal_Int32 len
)
1739 for (sal_Int32 nmatch
= 0; nmatch
< len
; nmatch
++) {
1740 if (*s1
++ != *s2
++) {
1749 /* We are passed P pointing to a register number after a start_memory.
1751 Return true if the pattern up to the corresponding stop_memory can
1752 match the empty string, and false otherwise.
1754 If we find the matching stop_memory, sets P to point to one past its number.
1755 Otherwise, sets P to an undefined byte less than or equal to END.
1757 We don't handle duplicates properly (yet). */
1760 Regexpr::group_match_null_string_p(sal_Unicode
**p
, sal_Unicode
*end
, register_info_type
*reg_info
)
1763 /* Point to after the args to the start_memory. */
1764 sal_Unicode
*p1
= *p
+ 2;
1767 /* Skip over opcodes that can match nothing, and return true or
1768 false, as appropriate, when we get to one that can't, or to the
1769 matching stop_memory. */
1771 switch ((re_opcode_t
) *p1
) {
1772 /* Could be either a loop or a series of alternatives. */
1773 case on_failure_jump
:
1775 extract_number_and_incr(mcnt
, p1
);
1777 /* If the next operation is not a jump backwards in the
1781 /* Go through the on_failure_jumps of the alternatives,
1782 seeing if any of the alternatives cannot match nothing.
1783 The last alternative starts with only a jump,
1784 whereas the rest start with on_failure_jump and end
1785 with a jump, e.g., here is the pattern for `a|b|c':
1787 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
1788 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
1791 So, we have to first go through the first (n-1)
1792 alternatives and then deal with the last one separately. */
1795 /* Deal with the first (n-1) alternatives, which start
1796 with an on_failure_jump (see above) that jumps to right
1797 past a jump_past_alt. */
1799 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
) {
1800 /* `mcnt' holds how many bytes long the alternative
1801 is, including the ending `jump_past_alt' and
1804 if (!alt_match_null_string_p(p1
, p1
+ mcnt
- 3, reg_info
))
1807 /* Move to right after this alternative, including the
1811 /* Break if it's the beginning of an n-th alternative
1812 that doesn't begin with an on_failure_jump. */
1813 if ((re_opcode_t
) *p1
!= on_failure_jump
)
1816 /* Still have to check that it's not an n-th
1817 alternative that starts with an on_failure_jump. */
1819 extract_number_and_incr(mcnt
, p1
);
1820 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
) {
1821 /* Get to the beginning of the n-th alternative. */
1827 /* Deal with the last alternative: go back and get number
1828 of the `jump_past_alt' just before it. `mcnt' contains
1829 the length of the alternative. */
1830 extract_number(mcnt
, p1
- 2);
1832 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
1835 p1
+= mcnt
; /* Get past the n-th alternative. */
1841 assert (p1
[1] == **p
);
1847 if (!common_op_match_null_string_p(&p1
, end
, reg_info
))
1850 } /* while p1 < end */
1853 } /* group_match_null_string_p */
1855 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
1856 It expects P to be the first byte of a single alternative and END one
1857 byte past the last. The alternative can contain groups. */
1860 Regexpr::alt_match_null_string_p(sal_Unicode
*p
, sal_Unicode
*end
, register_info_type
*reg_info
)
1863 sal_Unicode
*p1
= p
;
1866 /* Skip over opcodes that can match nothing, and break when we get
1867 to one that can't. */
1869 switch ((re_opcode_t
) *p1
) {
1871 case on_failure_jump
:
1873 extract_number_and_incr(mcnt
, p1
);
1878 if (!common_op_match_null_string_p(&p1
, end
, reg_info
))
1881 } /* while p1 < end */
1884 } /* alt_match_null_string_p */
1887 /* Deals with the ops common to group_match_null_string_p and
1888 alt_match_null_string_p.
1890 Sets P to one after the op and its arguments, if any. */
1893 Regexpr::common_op_match_null_string_p(sal_Unicode
**p
, sal_Unicode
*end
, register_info_type
*reg_info
)
1898 sal_Unicode
*p1
= *p
;
1900 switch ((re_opcode_t
) *p1
++) {
1910 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
1911 ret
= group_match_null_string_p(&p1
, end
, reg_info
);
1912 /* Have to set this here in case we're checking a group which
1913 contains a group and a back reference to it. */
1915 if (REG_MATCH_NULL_STRING_P(reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
1916 REG_MATCH_NULL_STRING_P(reg_info
[reg_no
]) = ret
;
1922 /* If this is an optimized succeed_n for zero times, make the jump. */
1924 extract_number_and_incr(mcnt
, p1
);
1932 /* Get to the number of times to succeed. */
1934 extract_number_and_incr(mcnt
, p1
);
1939 extract_number_and_incr(mcnt
, p1
);
1947 if (!REG_MATCH_NULL_STRING_P(reg_info
[*p1
]))
1955 /* All other opcodes mean we cannot match the empty string. */
1961 } /* common_op_match_null_string_p */
1965 /* Free everything we malloc. */
1966 #ifdef MATCH_MAY_ALLOCATE
1967 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
1968 # define FREE_VARIABLES() \
1970 REGEX_FREE_STACK (fail_stack.stack); \
1971 FREE_VAR (regstart); \
1972 FREE_VAR (regend); \
1973 FREE_VAR (old_regstart); \
1974 FREE_VAR (old_regend); \
1975 FREE_VAR (best_regstart); \
1976 FREE_VAR (best_regend); \
1977 FREE_VAR (reg_info); \
1978 FREE_VAR (reg_dummy); \
1979 FREE_VAR (reg_info_dummy); \
1982 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
1983 #endif /* not MATCH_MAY_ALLOCATE */
1985 /* This is a separate function so that we can force an alloca cleanup
1988 Regexpr::re_match2(struct re_registers
*regs
, sal_Int32 pos
, sal_Int32 range
)
1990 /* General temporaries. */
1994 /* Just past the end of the corresponding string. */
1997 /* Pointers into string2, just past the last characters in
1998 each to consider matching. */
1999 sal_Unicode
*end_match_2
;
2001 /* Where we are in the data, and the end of the current string. */
2002 const sal_Unicode
*d
, *dend
;
2004 /* Where we are in the compiled pattern, and the end of the compiled
2006 sal_Unicode
*p
= bufp
->buffer
;
2007 register sal_Unicode
*pend
= p
+ bufp
->used
;
2009 /* Mark the opcode just after a start_memory, so we can test for an
2010 empty subpattern when we get to the stop_memory. */
2011 sal_Unicode
*just_past_start_mem
= 0;
2013 /* Failure point stack. Each place that can handle a failure further
2014 down the line pushes a failure point on this stack. It consists of
2015 restart, regend, and reg_info for all registers corresponding to
2016 the subexpressions we're currently inside, plus the number of such
2017 registers, and, finally, two sal_Unicode *'s. The first
2018 sal_Unicode * is where to resume scanning the pattern; the second
2019 one is where to resume scanning the strings. If the latter is
2020 zero, the failure point is a ``dummy''; if a failure happens and
2021 the failure point is a dummy, it gets discarded and the next next
2023 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
2024 fail_stack_type fail_stack
;
2027 /* We fill all the registers internally, independent of what we
2028 return, for use in backreferences. The number here includes
2029 an element for register zero. */
2030 size_t num_regs
= bufp
->re_nsub
+ 1;
2032 /* The currently active registers. */
2033 sal_uInt32 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
2034 sal_uInt32 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
2036 /* Information on the contents of registers. These are pointers into
2037 the input strings; they record just what was matched (on this
2038 attempt) by a subexpression part of the pattern, that is, the
2039 regnum-th regstart pointer points to where in the pattern we began
2040 matching and the regnum-th regend points to right after where we
2041 stopped matching the regnum-th subexpression. (The zeroth register
2042 keeps track of what the whole pattern matches.) */
2043 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2044 const sal_Unicode
**regstart
, **regend
;
2047 /* If a group that's operated upon by a repetition operator fails to
2048 match anything, then the register for its start will need to be
2049 restored because it will have been set to wherever in the string we
2050 are when we last see its open-group operator. Similarly for a
2052 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2053 const sal_Unicode
**old_regstart
, **old_regend
;
2056 /* The is_active field of reg_info helps us keep track of which (possibly
2057 nested) subexpressions we are currently in. The matched_something
2058 field of reg_info[reg_num] helps us tell whether or not we have
2059 matched any of the pattern so far this time through the reg_num-th
2060 subexpression. These two fields get reset each time through any
2061 loop their register is in. */
2062 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
2063 register_info_type
*reg_info
;
2066 /* The following record the register info as found in the above
2067 variables when we find a match better than any we've seen before.
2068 This happens as we backtrack through the failure points, which in
2069 turn happens only if we have not yet matched the entire string. */
2070 //unsigned best_regs_set = false;
2071 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2072 const sal_Unicode
**best_regstart
, **best_regend
;
2075 /* Logically, this is `best_regend[0]'. But we don't want to have to
2076 allocate space for that if we're not allocating space for anything
2077 else (see below). Also, we never need info about register 0 for
2078 any of the other register vectors, and it seems rather a kludge to
2079 treat `best_regend' differently than the rest. So we keep track of
2080 the end of the best match so far in a separate variable. We
2081 initialize this to NULL so that when we backtrack the first time
2082 and need to test it, it's not garbage. */
2083 //const sal_Unicode *match_end = NULL;
2085 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
2086 sal_Int32 set_regs_matched_done
= 0;
2088 /* Used when we pop values we don't care about. */
2089 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2090 const sal_Unicode
**reg_dummy
;
2091 register_info_type
*reg_info_dummy
;
2096 #ifdef MATCH_MAY_ALLOCATE
2097 /* Do not bother to initialize all the register variables if there are
2098 no groups in the pattern, as it takes a fair amount of time. If
2099 there are groups, we include space for register 0 (the whole
2100 pattern), even though we never use it, since it simplifies the
2101 array indexing. We should fix this. */
2104 regstart
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2105 regend
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2106 old_regstart
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2107 old_regend
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2108 best_regstart
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2109 best_regend
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2110 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
2111 reg_dummy
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2112 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
2114 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
2115 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
2123 /* We must initialize all our variables to NULL, so that
2124 `FREE_VARIABLES' doesn't try to free them. */
2125 regstart
= regend
= old_regstart
= old_regend
= best_regstart
2126 = best_regend
= reg_dummy
= NULL
;
2127 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
2129 #endif /* MATCH_MAY_ALLOCATE */
2131 sal_Unicode
*string2
= (sal_Unicode
*)line
;
2132 sal_Int32 size2
= linelen
;
2133 sal_Int32 stop
= range
;
2135 /* The starting position is bogus. */
2136 if (pos
< 0 || pos
>= size2
|| linelen
<= 0 ) {
2141 /* Initialize subexpression text positions to -1 to mark ones that no
2142 start_memory/stop_memory has been seen for. Also initialize the
2143 register information struct. */
2144 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++) {
2145 regstart
[mcnt
] = regend
[mcnt
]
2146 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
2148 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
2149 IS_ACTIVE (reg_info
[mcnt
]) = 0;
2150 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
2151 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
2154 end2
= (sal_Unicode
*)(string2
+ size2
);
2156 end_match_2
= (sal_Unicode
*)(string2
+ stop
);
2158 /* `p' scans through the pattern as `d' scans through the data.
2159 `dend' is the end of the input string that `d' points within. `d'
2160 is advanced into the following input string whenever necessary, but
2161 this happens before fetching; therefore, at the beginning of the
2162 loop, `d' can be pointing at the end of a string, but it cannot
2167 /* This loops over pattern commands. It exits by returning from the
2168 function if the match is complete, or it drops through if the match
2169 fails at this starting point in the input data. */
2172 /* End of pattern means we might have succeeded. */
2174 /* If we haven't matched the entire string, and we want the
2175 longest match, try backtracking. */
2176 if (d
!= end_match_2
) {
2177 if (!FAIL_STACK_EMPTY()) {
2180 } /* d != end_match_2 */
2184 /* If caller wants register contents data back, do it. */
2186 /* Have the register data arrays been allocated? */
2187 if (regs
->num_regs
== 0) {
2188 /* No. So allocate them with malloc. We need one
2189 extra element beyond `num_regs' for the `-1' marker
2191 regs
->num_of_match
= 0;
2192 regs
->num_regs
= MAX(RE_NREGS
, num_regs
+ 1);
2193 regs
->start
= (sal_Int32
*) malloc(regs
->num_regs
* sizeof(sal_Int32
));
2194 regs
->end
= (sal_Int32
*) malloc(regs
->num_regs
* sizeof(sal_Int32
));
2195 if (regs
->start
== NULL
|| regs
->end
== NULL
) {
2199 } else if ( regs
->num_regs
> 0 ) {
2200 /* Yes. If we need more elements than were already
2201 allocated, reallocate them. If we need fewer, just
2203 if (regs
->num_regs
< num_regs
+ 1) {
2204 regs
->num_regs
= num_regs
+ 1;
2205 regs
->start
= (sal_Int32
*) realloc(regs
->start
, regs
->num_regs
* sizeof(sal_Int32
));
2206 regs
->end
= (sal_Int32
*) realloc(regs
->end
, regs
->num_regs
* sizeof(sal_Int32
));
2207 if (regs
->start
== NULL
|| regs
->end
== NULL
) {
2212 } else { // num_regs is negative
2217 /* Convert the pointer data in `regstart' and `regend' to
2218 indices. Register zero has to be set differently,
2219 since we haven't kept track of any info for it. */
2220 if (regs
->num_regs
> 0) {
2221 // Make sure a valid location
2222 sal_Int32 dpos
= d
- string2
;
2223 if (pos
== dpos
|| (d
- 1) >= dend
) {
2227 regs
->start
[regs
->num_of_match
] = pos
;
2228 regs
->end
[regs
->num_of_match
] = ((sal_Int32
) (d
- string2
));
2229 regs
->num_of_match
++;
2232 /* Go through the first `min (num_regs, regs->num_regs)'
2233 registers, since that is all we initialized. */
2234 for (mcnt
= regs
->num_of_match
; (unsigned) mcnt
< MIN(num_regs
, regs
->num_regs
);
2236 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
2237 if( !(REG_UNSET(regstart
[mcnt
]) || REG_UNSET(regend
[mcnt
])) ) {
2238 regs
->start
[regs
->num_of_match
] = (sal_Int32
) POINTER_TO_OFFSET(regstart
[mcnt
]);
2239 regs
->end
[regs
->num_of_match
] = (sal_Int32
) POINTER_TO_OFFSET(regend
[mcnt
]);
2240 regs
->num_of_match
++;
2244 /* If the regs structure we return has more elements than
2245 were in the pattern, set the extra elements to -1. If
2246 we (re)allocated the registers, this is the case,
2247 because we always allocate enough to have at least one
2249 for (mcnt
= regs
->num_of_match
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
2250 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
2253 mcnt
= d
- pos
- string2
;
2258 /* Otherwise match next pattern command. */
2259 switch ((re_opcode_t
) *p
++) {
2260 /* Ignore these. Used to ignore the n of succeed_n's which
2261 currently have n == 0. */
2268 /* Match the next n pattern characters exactly. The following
2269 byte in the pattern defines n, and the n bytes after that
2270 are the characters to match. */
2276 if ((sal_Unicode
)*d
++ != (sal_Unicode
) *p
++) goto fail
;
2281 /* Match any character except possibly a newline or a null. */
2285 if ( *d
== (sal_Unicode
)'\n' ||
2286 *d
== (sal_Unicode
)'\000' )
2295 register sal_Unicode c
;
2296 sal_Bool knot
= (re_opcode_t
) *(p
- 1) == charset_not
;
2299 c
= *d
; /* The character to match. */
2300 /* Cast to `sal_uInt32' instead of `sal_Unicode' in case the
2301 bit list is a full 32 bytes long. */
2302 if ((c
< (sal_uInt32
) (*p
* BYTEWIDTH
)) && (p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
))))
2316 /* The beginning of a group is represented by start_memory.
2317 The arguments are the register number in the next byte, and the
2318 number of groups inner to this one in the next. The text
2319 matched within the group is recorded (in the internal
2320 registers data structure) under the register number. */
2323 /* Find out if this group can match the empty string. */
2324 p1
= p
; /* To send to group_match_null_string_p. */
2326 if (REG_MATCH_NULL_STRING_P(reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
2327 REG_MATCH_NULL_STRING_P(reg_info
[*p
]) = group_match_null_string_p(&p1
, pend
, reg_info
);
2329 /* Save the position in the string where we were the last time
2330 we were at this open-group operator in case the group is
2331 operated upon by a repetition operator, e.g., with `(a*)*b'
2332 against `ab'; then we want to ignore where we are now in
2333 the string in case this attempt to match fails. */
2334 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P(reg_info
[*p
])
2335 ? REG_UNSET(regstart
[*p
]) ? d
: regstart
[*p
]
2340 IS_ACTIVE (reg_info
[*p
]) = 1;
2341 MATCHED_SOMETHING(reg_info
[*p
]) = 0;
2343 /* Clear this whenever we change the register activity status. */
2344 set_regs_matched_done
= 0;
2346 /* This is the new highest active register. */
2347 highest_active_reg
= *p
;
2349 /* If nothing was active before, this is the new lowest active
2351 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
2352 lowest_active_reg
= *p
;
2354 /* Move past the register number and inner group count. */
2356 just_past_start_mem
= p
;
2360 /* The stop_memory opcode represents the end of a group. Its
2361 arguments are the same as start_memory's: the register
2362 number, and the number of inner groups. */
2365 /* We need to save the string position the last time we were at
2366 this close-group operator in case the group is operated
2367 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
2368 against `aba'; then we want to ignore where we are now in
2369 the string in case this attempt to match fails. */
2370 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
2371 ? REG_UNSET(regend
[*p
]) ? d
: regend
[*p
]
2376 /* This register isn't active anymore. */
2377 IS_ACTIVE(reg_info
[*p
]) = 0;
2379 /* Clear this whenever we change the register activity status. */
2380 set_regs_matched_done
= 0;
2382 /* If this was the only register active, nothing is active
2384 if (lowest_active_reg
== highest_active_reg
) {
2385 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
2386 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
2387 } else { /* We must scan for the new highest active register, since
2388 it isn't necessarily one less than now: consider
2389 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
2390 new highest active register is 1. */
2391 sal_Unicode r
= *p
- 1;
2392 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
2395 /* If we end up at register zero, that means that we saved
2396 the registers as the result of an `on_failure_jump', not
2397 a `start_memory', and we jumped to past the innermost
2398 `stop_memory'. For example, in ((.)*) we save
2399 registers 1 and 2 as a result of the *, but when we pop
2400 back to the second ), we are at the stop_memory 1.
2401 Thus, nothing is active. */
2403 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
2404 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
2406 highest_active_reg
= r
;
2409 /* If just failed to match something this time around with a
2410 group that's operated on by a repetition operator, try to
2411 force exit from the ``loop'', and restore the register
2412 information for this group that we had before trying this
2414 if ((!MATCHED_SOMETHING (reg_info
[*p
])
2415 || just_past_start_mem
== p
- 1)
2416 && (p
+ 2) < pend
) {
2417 sal_Bool is_a_jump_n
= false;
2421 switch ((re_opcode_t
) *p1
++) {
2424 case pop_failure_jump
:
2425 case maybe_pop_jump
:
2427 case dummy_failure_jump
:
2428 extract_number_and_incr(mcnt
, p1
);
2438 /* If the next operation is a jump backwards in the pattern
2439 to an on_failure_jump right before the start_memory
2440 corresponding to this stop_memory, exit from the loop
2441 by forcing a failure after pushing on the stack the
2442 on_failure_jump's jump in the pattern, and d. */
2443 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
2444 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
) {
2445 /* If this group ever matched anything, then restore
2446 what its registers were before trying this last
2447 failed match, e.g., with `(a*)*b' against `ab' for
2448 regstart[1], and, e.g., with `((a*)*(b*)*)*'
2449 against `aba' for regend[3].
2451 Also restore the registers for inner groups for,
2452 e.g., `((a*)(b*))*' against `aba' (register 3 would
2453 otherwise get trashed). */
2455 if (EVER_MATCHED_SOMETHING (reg_info
[*p
])) {
2458 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
2460 /* Restore this and inner groups' (if any) registers. */
2461 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
2463 regstart
[r
] = old_regstart
[r
];
2465 /* xx why this test? */
2466 if (old_regend
[r
] >= regstart
[r
])
2467 regend
[r
] = old_regend
[r
];
2471 extract_number_and_incr(mcnt
, p1
);
2472 PUSH_FAILURE_POINT(p1
+ mcnt
, d
, -2);
2478 /* Move past the register number and the inner group count. */
2483 /* \<digit> has been turned into a `duplicate' command which is
2484 followed by the numeric value of <digit> as the register number. */
2487 register const sal_Unicode
*d2
, *dend2
;
2488 sal_Unicode regno
= *p
++; /* Get which register to match against. */
2490 /* Can't back reference a group which we've never matched. */
2491 if (REG_UNSET(regstart
[regno
]) || REG_UNSET(regend
[regno
])) {
2495 /* Where in input to try to start matching. */
2496 d2
= regstart
[regno
];
2498 /* Where to stop matching; if both the place to start and
2499 the place to stop matching are in the same string, then
2500 set to the place to stop, otherwise, for now have to use
2501 the end of the first string. */
2503 dend2
= regend
[regno
];
2505 /* If necessary, advance to next segment in register
2507 while (d2
== dend2
) {
2508 if (dend2
== end_match_2
) break;
2509 if (dend2
== regend
[regno
]) break;
2511 /* At end of register contents => success */
2512 if (d2
== dend2
) break;
2516 /* How many characters left in this segment to match. */
2519 /* Want how many consecutive characters we can match in
2520 one shot, so, if necessary, adjust the count. */
2521 if (mcnt
> dend2
- d2
)
2524 /* Compare that many; failure if mismatch, else move
2527 ? bcmp_translate(d
, d2
, mcnt
)
2528 : memcmp(d
, d2
, mcnt
* sizeof(sal_Unicode
))) {
2531 d
+= mcnt
, d2
+= mcnt
;
2532 /* Do this because we've match some characters. */
2538 /* begline matches the empty string at the beginning of the string
2539 (unless `not_bol' is set in `bufp'), and, if
2540 `newline_anchor' is set, after newlines. */
2543 if (AT_STRINGS_BEG (d
)) {
2544 if (!bufp
->not_bol
) break;
2545 } else if (d
[-1] == '\n' && bufp
->newline_anchor
) {
2548 /* In all other cases, we fail. */
2551 /* endline is the dual of begline. */
2554 if (AT_STRINGS_END(d
)) {
2555 if (!bufp
->not_eol
) break;
2556 } else if (*d
== '\n' && bufp
->newline_anchor
) {
2561 /* Match at the very beginning of the data. */
2563 if (AT_STRINGS_BEG (d
))
2568 /* Match at the very end of the data. */
2570 if (AT_STRINGS_END (d
))
2575 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
2576 pushes NULL as the value for the string on the stack. Then
2577 `pop_failure_point' will keep the current value for the
2578 string, instead of restoring it. To see why, consider
2579 matching `foo\nbar' against `.*\n'. The .* matches the foo;
2580 then the . fails against the \n. But the next thing we want
2581 to do is match the \n against the \n; if we restored the
2582 string value, we would be back at the foo.
2584 Because this is used only in specific cases, we don't need to
2585 check all the things that `on_failure_jump' does, to make
2586 sure the right things get saved on the stack. Hence we don't
2587 share its code. The only reason to push anything on the
2588 stack at all is that otherwise we would have to change
2589 `anychar's code to do something besides goto fail in this
2590 case; that seems worse than this. */
2591 case on_failure_keep_string_jump
:
2593 extract_number_and_incr(mcnt
, p
);
2595 PUSH_FAILURE_POINT(p
+ mcnt
, NULL
, -2);
2599 /* Uses of on_failure_jump:
2601 Each alternative starts with an on_failure_jump that points
2602 to the beginning of the next alternative. Each alternative
2603 except the last ends with a jump that in effect jumps past
2604 the rest of the alternatives. (They really jump to the
2605 ending jump of the following alternative, because tensioning
2606 these jumps is a hassle.)
2608 Repeats start with an on_failure_jump that points past both
2609 the repetition text and either the following jump or
2610 pop_failure_jump back to this on_failure_jump. */
2611 case on_failure_jump
:
2614 extract_number_and_incr(mcnt
, p
);
2616 /* If this on_failure_jump comes right before a group (i.e.,
2617 the original * applied to a group), save the information
2618 for that group and all inner ones, so that if we fail back
2619 to this point, the group's information will be correct.
2620 For example, in \(a*\)*\1, we need the preceding group,
2621 and in \(zz\(a*\)b*\)\2, we need the inner group. */
2623 /* We can't use `p' to check ahead because we push
2624 a failure point to `p + mcnt' after we do this. */
2627 /* We need to skip no_op's before we look for the
2628 start_memory in case this on_failure_jump is happening as
2629 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
2631 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
2634 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
) {
2635 /* We have a new highest active register now. This will
2636 get reset at the start_memory we are about to get to,
2637 but we will have saved all the registers relevant to
2638 this repetition op, as described above. */
2639 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
2640 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
2641 lowest_active_reg
= *(p1
+ 1);
2644 PUSH_FAILURE_POINT(p
+ mcnt
, d
, -2);
2647 /* A smart repeat ends with `maybe_pop_jump'.
2648 We change it to either `pop_failure_jump' or `jump'. */
2649 case maybe_pop_jump
:
2650 extract_number_and_incr(mcnt
, p
);
2652 register sal_Unicode
*p2
= p
;
2654 /* Compare the beginning of the repeat with what in the
2655 pattern follows its end. If we can establish that there
2656 is nothing that they would both match, i.e., that we
2657 would have to backtrack because of (as in, e.g., `a*a')
2658 then we can change to pop_failure_jump, because we'll
2659 never have to backtrack.
2661 This is not true in the case of alternatives: in
2662 `(a|ab)*' we do need to backtrack to the `ab' alternative
2663 (e.g., if the string was `ab'). But instead of trying to
2664 detect that here, the alternative has put on a dummy
2665 failure point which is what we will end up popping. */
2667 /* Skip over open/close-group commands.
2668 If what follows this loop is a ...+ construct,
2669 look at what begins its body, since we will have to
2670 match at least one of that. */
2673 && ((re_opcode_t
) *p2
== stop_memory
2674 || (re_opcode_t
) *p2
== start_memory
))
2676 else if (p2
+ 6 < pend
2677 && (re_opcode_t
) *p2
== dummy_failure_jump
)
2684 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
2685 to the `maybe_finalize_jump' of this case. Examine what
2688 /* If we're at the end of the pattern, we can change. */
2690 /* Consider what happens when matching ":\(.*\)"
2691 against ":/". I don't really understand this code
2693 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2694 } else if ((re_opcode_t
) *p2
== exactn
2695 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
)) {
2696 register sal_Unicode c
= *p2
== (sal_Unicode
) endline
? (sal_Unicode
)'\n' : p2
[2];
2698 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
) {
2699 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2700 } else if ((re_opcode_t
) p1
[3] == charset
2701 || (re_opcode_t
) p1
[3] == charset_not
) {
2702 sal_Int32 knot
= (re_opcode_t
) p1
[3] == charset_not
;
2704 if (c
< (sal_Unicode
) (p1
[4] * BYTEWIDTH
)
2705 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
2708 /* `not' is equal to 1 if c would match, which means
2709 that we can't change to pop_failure_jump. */
2711 p
[-3] = (unsigned char) pop_failure_jump
;
2714 } else if ((re_opcode_t
) *p2
== charset
) {
2715 /* We win if the first character of the loop is not part
2717 if ((re_opcode_t
) p1
[3] == exactn
2718 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
2719 && (p2
[2 + p1
[5] / BYTEWIDTH
]
2720 & (1 << (p1
[5] % BYTEWIDTH
))))) {
2721 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2722 } else if ((re_opcode_t
) p1
[3] == charset_not
) {
2724 /* We win if the charset_not inside the loop
2725 lists every character listed in the charset after. */
2726 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
2727 if (! (p2
[2 + idx
] == 0
2728 || (idx
< (int) p1
[4]
2729 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
2733 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2735 } else if ((re_opcode_t
) p1
[3] == charset
) {
2737 /* We win if the charset inside the loop
2738 has no overlap with the one after the loop. */
2740 idx
< (sal_Int32
) p2
[1] && idx
< (sal_Int32
) p1
[4];
2742 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
2745 if (idx
== p2
[1] || idx
== p1
[4]) {
2746 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2751 p
-= 2; /* Point at relative address again. */
2752 if ((re_opcode_t
) p
[-1] != pop_failure_jump
) {
2753 p
[-1] = (sal_Unicode
) jump
;
2754 goto unconditional_jump
;
2756 /* Note fall through. */
2759 /* The end of a simple repeat has a pop_failure_jump back to
2760 its matching on_failure_jump, where the latter will push a
2761 failure point. The pop_failure_jump takes off failure
2762 points put on by this pop_failure_jump's matching
2763 on_failure_jump; we got through the pattern to here from the
2764 matching on_failure_jump, so didn't fail. */
2765 case pop_failure_jump
:
2767 /* We need to pass separate storage for the lowest and
2768 highest registers, even though we don't care about the
2769 actual values. Otherwise, we will restore only one
2770 register from the stack, since lowest will == highest in
2771 `pop_failure_point'. */
2772 sal_uInt32 dummy_low_reg
, dummy_high_reg
;
2773 sal_Unicode
*pdummy
= NULL
;
2774 const sal_Unicode
*sdummy
= NULL
;
2776 POP_FAILURE_POINT(sdummy
, pdummy
,
2777 dummy_low_reg
, dummy_high_reg
,
2778 reg_dummy
, reg_dummy
, reg_info_dummy
);
2780 /* Note fall through. */
2783 /* Note fall through. */
2785 /* Unconditionally jump (without popping any failure points). */
2787 extract_number_and_incr(mcnt
, p
); /* Get the amount to jump. */
2788 p
+= mcnt
; /* Do the jump. */
2791 /* We need this opcode so we can detect where alternatives end
2792 in `group_match_null_string_p' et al. */
2794 goto unconditional_jump
;
2797 /* Normally, the on_failure_jump pushes a failure point, which
2798 then gets popped at pop_failure_jump. We will end up at
2799 pop_failure_jump, also, and with a pattern of, say, `a+', we
2800 are skipping over the on_failure_jump, so we have to push
2801 something meaningless for pop_failure_jump to pop. */
2802 case dummy_failure_jump
:
2803 /* It doesn't matter what we push for the string here. What
2804 the code at `fail' tests is the value for the pattern. */
2805 PUSH_FAILURE_POINT(NULL
, NULL
, -2);
2806 goto unconditional_jump
;
2809 /* At the end of an alternative, we need to push a dummy failure
2810 point in case we are followed by a `pop_failure_jump', because
2811 we don't want the failure point for the alternative to be
2812 popped. For example, matching `(a|ab)*' against `aab'
2813 requires that we match the `ab' alternative. */
2814 case push_dummy_failure
:
2815 /* See comments just above at `dummy_failure_jump' about the
2817 PUSH_FAILURE_POINT(NULL
, NULL
, -2);
2820 /* Have to succeed matching what follows at least n times.
2821 After that, handle like `on_failure_jump'. */
2823 extract_number(mcnt
, p
+ 2);
2826 /* Originally, this is how many times we HAVE to succeed. */
2830 store_number_and_incr (p
, mcnt
);
2831 } else if (mcnt
== 0) {
2832 p
[2] = (sal_Unicode
) no_op
;
2833 p
[3] = (sal_Unicode
) no_op
;
2839 extract_number(mcnt
, p
+ 2);
2841 /* Originally, this is how many times we CAN jump. */
2844 store_number (p
+ 2, mcnt
);
2845 goto unconditional_jump
;
2847 /* If don't have to jump any more, skip over the rest of command. */
2855 extract_number_and_incr(mcnt
, p
);
2857 extract_number_and_incr(mcnt
, p
);
2858 store_number (p1
, mcnt
);
2863 if (iswordbegin(d
, string2
, size2
))
2868 if (iswordend(d
, string2
, size2
))
2876 continue; /* Successfully executed one pattern command; keep going. */
2878 /* We goto here if a matching operation fails. */
2880 if (!FAIL_STACK_EMPTY()) {
2881 /* A restart point is known. Restore to that state. */
2882 POP_FAILURE_POINT(d
, p
,
2883 lowest_active_reg
, highest_active_reg
,
2884 regstart
, regend
, reg_info
);
2886 /* If this failure point is a dummy, try the next one. */
2890 /* If we failed to the end of the pattern, don't examine *p. */
2893 sal_Bool is_a_jump_n
= false;
2895 /* If failed to a backwards jump that's part of a repetition
2896 loop, need to pop this failure point and use the next
2898 switch ((re_opcode_t
) *p
) {
2901 case maybe_pop_jump
:
2902 case pop_failure_jump
:
2905 extract_number_and_incr(mcnt
, p1
);
2908 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
2910 && (re_opcode_t
) *p1
== on_failure_jump
)) {
2920 break; /* Matching at this starting point really fails. */
2926 return(-1); /* Failure to match. */
2929 /* Set the bit for character C in a list. */
2931 Regexpr::set_list_bit(sal_Unicode c
, sal_Unicode
*b
)
2935 sal_Unicode tmp
= translit
->transliterateChar2Char(c
);
2936 b
[tmp
/ BYTEWIDTH
] |= 1 << (tmp
% BYTEWIDTH
);
2937 } catch (::com::sun::star::i18n::MultipleCharsOutputException e
) {
2938 ::rtl::OUString
o2( translit
->transliterateChar2String( c
));
2939 sal_Int32 len2
= o2
.getLength();
2940 const sal_Unicode
* k2
= o2
.getStr();
2941 for (sal_Int32 nmatch
= 0; nmatch
< len2
; nmatch
++) {
2942 b
[k2
[nmatch
] / BYTEWIDTH
] |= 1 << (k2
[nmatch
] % BYTEWIDTH
);
2946 b
[c
/ BYTEWIDTH
] |= 1 << (c
% BYTEWIDTH
);
2950 /* vim: set ts=8 sw=2 noexpandtab: */