1 /*************************************************************************
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 * Copyright 2000, 2010 Oracle and/or its affiliates.
7 * OpenOffice.org - a multi-platform office productivity suite
9 * This file is part of OpenOffice.org.
11 * OpenOffice.org is free software: you can redistribute it and/or modify
12 * it under the terms of the GNU Lesser General Public License version 3
13 * only, as published by the Free Software Foundation.
15 * OpenOffice.org is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU Lesser General Public License version 3 for more details
19 * (a copy is included in the LICENSE file that accompanied this code).
21 * You should have received a copy of the GNU Lesser General Public License
22 * version 3 along with OpenOffice.org. If not, see
23 * <http://www.openoffice.org/license.html>
24 * for a copy of the LGPLv3 License.
26 ************************************************************************/
29 // MARKER(update_precomp.py): autogen include statement, do not remove
30 #include "precompiled_regexp.hxx"
31 /* Extended regular expression matching and search library,
33 (Implements POSIX draft P1003.2/D11.2, except for some of the
34 internationalization features.)
35 Copyright (C) 1993, 94, 95, 96, 97, 98, 99 Free Software Foundation, Inc.
37 The GNU C Library is free software; you can redistribute it and/or
38 modify it under the terms of the GNU Library General Public License as
39 published by the Free Software Foundation; either version 2 of the
40 License, or (at your option) any later version.
42 The GNU C Library is distributed in the hope that it will be useful,
43 but WITHOUT ANY WARRANTY; without even the implied warranty of
44 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
45 Library General Public License for more details.
47 You should have received a copy of the GNU Library General Public
48 License along with the GNU C Library; see the file COPYING.LIB. If not,
49 write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
50 Boston, MA 02111-1307, USA. */
53 Modified for OpenOffice.org to use sal_Unicode and Transliteration service.
58 /* If for any reason (porting, debug) we can't use alloca() use malloc()
59 instead. Use alloca() if possible for performance reasons, this _is_
60 significant, with malloc() the re_match2() method makes heavy use of regexps
61 through the TextSearch interface up to three times slower. This is _the_
62 bottleneck in some spreadsheet documents. */
66 /* AIX requires this to be the first thing in the file. */
67 #if defined _AIX && !defined REGEX_MALLOC
74 #include <rtl/ustring.hxx>
75 #include <com/sun/star/i18n/TransliterationModules.hpp>
77 #include "reclass.hxx"
80 /* Maximum number of duplicates an interval can allow. Some systems
81 (erroneously) define this in other header files, but we want our
82 value, so remove any previous define. */
86 /* If sizeof(int) == 2, then ((1 << 15) - 1) overflows. */
87 #define RE_DUP_MAX (0x7fff)
90 /* If `regs_allocated' is REGS_UNALLOCATED in the pattern buffer,
91 `re_match_2' returns information about at least this many registers
92 the first time a `regs' structure is passed. */
99 #define INIT_COMPILE_STACK_SIZE 32
100 #define INIT_BUF_SIZE ((1 << BYTEWIDTH)/BYTEWIDTH)
101 #define MAX_BUF_SIZE 65535L
102 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
103 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
105 /* Since we have one byte reserved for the register number argument to
106 {start,stop}_memory, the maximum number of groups we can report
107 things about is what fits in that byte. */
108 #define MAX_REGNUM 255
110 #define MIN(x, y) ( (x) < (y) ? (x) : (y) )
111 #define MAX(x, y) ( (x) > (y) ? (x) : (y) )
114 // Always. We're not in Emacs and don't use relocating allocators.
115 #define MATCH_MAY_ALLOCATE
117 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
118 use `alloca' instead of `malloc'. This is because malloc is slower and
119 causes storage fragmentation. On the other hand, malloc is more portable,
122 Because we sometimes use alloca, some routines have to be macros,
123 not functions -- `alloca'-allocated space disappears at the end of the
124 function it is called in. */
128 # define REGEX_ALLOCATE malloc
129 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
130 # define REGEX_FREE free
132 #else /* not REGEX_MALLOC */
134 /* Emacs already defines alloca, sometimes. So does MSDEV. */
137 /* Make alloca work the best possible way. */
139 # define alloca __builtin_alloca
140 # else /* not __GNUC__ */
141 # include <sal/alloca.h>
142 # endif /* not __GNUC__ */
144 # endif /* not alloca */
146 # define REGEX_ALLOCATE alloca
148 /* Assumes a `char *destination' variable. */
149 # define REGEX_REALLOCATE(source, osize, nsize) \
150 (destination = (char *) alloca (nsize), \
151 memcpy (destination, source, osize))
153 /* No need to do anything to free, after alloca. */
154 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
156 #endif /* not REGEX_MALLOC */
159 /* Define how to allocate the failure stack. */
163 # define REGEX_ALLOCATE_STACK malloc
164 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
165 # define REGEX_FREE_STACK free
167 #else /* not REGEX_MALLOC */
169 # define REGEX_ALLOCATE_STACK alloca
171 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
172 REGEX_REALLOCATE (source, osize, nsize)
173 /* No need to explicitly free anything. */
174 # define REGEX_FREE_STACK(arg)
176 #endif /* not REGEX_MALLOC */
179 /* (Re)Allocate N items of type T using malloc, or fail. */
180 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
181 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
182 #define RETALLOC_IF(addr, n, t) \
183 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
184 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
186 #define BYTEWIDTH 16 /* In bits (assuming sizeof(sal_Unicode)*8) */
189 #define CHAR_CLASS_MAX_LENGTH 256
191 /* Fetch the next character in the uncompiled pattern, with no
193 #define PATFETCH_RAW(c) \
195 if (p == pend) return REG_EEND; \
196 c = (sal_Unicode) *p++; \
199 /* Go backwards one character in the pattern. */
200 #define PATUNFETCH p--
202 #define FREE_STACK_RETURN(value) \
203 return(free(compile_stack.stack), value)
205 #define GET_BUFFER_SPACE(n) \
206 while ((sal_uInt32)(b - bufp->buffer + (n)) > bufp->allocated) \
209 /* Extend the buffer by twice its current size via realloc and
210 reset the pointers that pointed into the old block to point to the
211 correct places in the new one. If extending the buffer results in it
212 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
213 #define EXTEND_BUFFER() \
215 sal_Unicode *old_buffer = bufp->buffer; \
216 if (bufp->allocated == MAX_BUF_SIZE) \
218 bufp->allocated <<= 1; \
219 if (bufp->allocated > MAX_BUF_SIZE) \
220 bufp->allocated = MAX_BUF_SIZE; \
221 bufp->buffer = (sal_Unicode *) realloc(bufp->buffer, \
223 sizeof(sal_Unicode)); \
224 if (bufp->buffer == NULL) \
226 /* If the buffer moved, move all the pointers into it. */ \
227 if (old_buffer != bufp->buffer) { \
228 b = (b - old_buffer) + bufp->buffer; \
229 begalt = (begalt - old_buffer) + bufp->buffer; \
230 if (fixup_alt_jump) \
231 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
233 laststart = (laststart - old_buffer) + bufp->buffer; \
235 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
239 #define BUF_PUSH(c) \
241 GET_BUFFER_SPACE(1); \
242 *b++ = (sal_Unicode)(c); \
245 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
246 #define BUF_PUSH_2(c1, c2) \
248 GET_BUFFER_SPACE(2); \
249 *b++ = (sal_Unicode) (c1); \
250 *b++ = (sal_Unicode) (c2); \
253 /* As with BUF_PUSH_2, except for three bytes. */
254 #define BUF_PUSH_3(c1, c2, c3) \
256 GET_BUFFER_SPACE(3); \
257 *b++ = (sal_Unicode) (c1); \
258 *b++ = (sal_Unicode) (c2); \
259 *b++ = (sal_Unicode) (c3); \
262 /* Store a jump with opcode OP at LOC to location TO. We store a
263 relative address offset by the three bytes the jump itself occupies. */
264 #define STORE_JUMP(op, loc, to) \
265 store_op1(op, loc, (int) ((to) - (loc) - 3))
267 /* Likewise, for a two-argument jump. */
268 #define STORE_JUMP2(op, loc, to, arg) \
269 store_op2(op, loc, (int) ((to) - (loc) - 3), arg)
271 /* Store NUMBER in two contiguous sal_Unicode starting at DESTINATION. */
275 Regexpr::store_number( sal_Unicode
* destination
, sal_Int32 number
)
277 (destination
)[0] = sal_Unicode((number
) & 0xffff);
278 (destination
)[1] = sal_Unicode((number
) >> 16);
281 /* Same as STORE_NUMBER, except increment DESTINATION to
282 the byte after where the number is stored. Therefore, DESTINATION
283 must be an lvalue. */
287 Regexpr::store_number_and_incr( sal_Unicode
*& destination
, sal_Int32 number
)
289 store_number( destination
, number
);
293 /* Put into DESTINATION a number stored in two contiguous sal_Unicode starting
296 inline void Regexpr::extract_number( sal_Int32
& dest
, sal_Unicode
*source
)
298 dest
= (((sal_Int32
) source
[1]) << 16) | (source
[0] & 0xffff);
301 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
302 #define INSERT_JUMP(op, loc, to) \
303 insert_op1(op, loc, (sal_Int32) ((to) - (loc) - 3), b)
305 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
306 #define INSERT_JUMP2(op, loc, to, arg) \
307 insert_op2(op, loc, (sal_Int32) ((to) - (loc) - 3), arg, b)
309 #define STREQ(s1, s2) (rtl_ustr_compare((s1), (s2)) ? (0) : (1))
311 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
312 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
314 /* The next available element. */
315 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
317 /* Get the next unsigned number in the uncompiled pattern. */
318 #define GET_UNSIGNED_NUMBER(num) { \
321 while (c >= (sal_Unicode)'0' && c <= (sal_Unicode)'9') { \
324 num = num * 10 + c - (sal_Unicode)'0'; \
332 /* Get the next hex number in the uncompiled pattern. */
333 #define GET_HEX_NUMBER(num) { \
335 sal_Bool stop = false; \
336 sal_Int16 hexcnt = 1; \
338 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') ) { \
341 if ( c >= (sal_Unicode)'0' && c <= (sal_Unicode)'9' ) \
342 num = num * 16 + c - (sal_Unicode)'0'; \
343 else if ( c >= (sal_Unicode)'a' && c <= (sal_Unicode)'f' ) \
344 num = num * 16 + (10 + c - (sal_Unicode)'a'); \
346 num = num * 16 + (10 + c - (sal_Unicode)'A'); \
347 if (p == pend || hexcnt == 4) { \
359 if ( hexcnt > 4 || (num < 0 || num > 0xffff) ) num = -1;\
364 /* Number of failure points for which to initially allocate space
365 when matching. If this number is exceeded, we allocate more
366 space, so it is not a hard limit. */
367 #ifndef INIT_FAILURE_ALLOC
368 # define INIT_FAILURE_ALLOC 5
371 #define INIT_FAIL_STACK() \
373 fail_stack.stack = (fail_stack_elt_t *) \
374 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
376 if (fail_stack.stack == NULL) \
379 fail_stack.size = INIT_FAILURE_ALLOC; \
380 fail_stack.avail = 0; \
383 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
385 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
387 Return 1 if succeeds, and 0 if either ran out of memory
388 allocating space for it or it was already too large.
390 REGEX_REALLOCATE_STACK requires `destination' be declared. */
392 #define DOUBLE_FAIL_STACK(fail_stack) \
393 ((fail_stack).size > (sal_uInt32) (re_max_failures * MAX_FAILURE_ITEMS) \
395 : ((fail_stack).stack = (fail_stack_elt_t *) \
396 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
397 (fail_stack).size * sizeof (fail_stack_elt_t), \
398 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
400 (fail_stack).stack == NULL \
402 : ((fail_stack).size <<= 1, \
406 #define REG_UNSET_VALUE (®_unset_dummy)
407 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
409 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
410 #define IS_ACTIVE(R) ((R).bits.is_active)
411 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
412 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
414 /* Call this when have matched a real character; it sets `matched' flags
415 for the subexpressions which we are currently inside. Also records
416 that those subexprs have matched. */
417 #define SET_REGS_MATCHED() \
419 if (!set_regs_matched_done) { \
421 set_regs_matched_done = 1; \
422 for (r = lowest_active_reg; r <= highest_active_reg; r++) { \
423 MATCHED_SOMETHING(reg_info[r]) \
424 = EVER_MATCHED_SOMETHING(reg_info[r]) \
431 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
433 /* This converts PTR, a pointer into the search string `string2' into an offset from the beginning of that string. */
434 #define POINTER_TO_OFFSET(ptr) ((sal_Int32) ((ptr) - string2))
436 /* This is the number of items that are pushed and popped on the stack
437 for each register. */
438 #define NUM_REG_ITEMS 3
440 /* Individual items aside from the registers. */
441 # define NUM_NONREG_ITEMS 4
443 /* We push at most this many items on the stack. */
444 /* We used to use (num_regs - 1), which is the number of registers
445 this regexp will save; but that was changed to 5
446 to avoid stack overflow for a regexp with lots of parens. */
447 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
449 /* We actually push this many items. */
450 #define NUM_FAILURE_ITEMS \
452 ? 0 : highest_active_reg - lowest_active_reg + 1) \
456 /* How many items can still be added to the stack without overflowing it. */
457 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
459 /* Push a pointer value onto the failure stack.
460 Assumes the variable `fail_stack'. Probably should only
461 be called from within `PUSH_FAILURE_POINT'. */
462 #define PUSH_FAILURE_POINTER(item) \
463 fail_stack.stack[fail_stack.avail++].pointer = (sal_Unicode *) (item)
465 /* This pushes an integer-valued item onto the failure stack.
466 Assumes the variable `fail_stack'. Probably should only
467 be called from within `PUSH_FAILURE_POINT'. */
468 #define PUSH_FAILURE_INT(item) \
469 fail_stack.stack[fail_stack.avail++].integer = (item)
471 /* Push a fail_stack_elt_t value onto the failure stack.
472 Assumes the variable `fail_stack'. Probably should only
473 be called from within `PUSH_FAILURE_POINT'. */
474 #define PUSH_FAILURE_ELT(item) \
475 fail_stack.stack[fail_stack.avail++] = (item)
477 /* These three POP... operations complement the three PUSH... operations.
478 All assume that `fail_stack' is nonempty. */
479 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
480 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
481 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
483 /* Test if at very beginning or at very end of `string2'. */
484 #define AT_STRINGS_BEG(d) ((d) == string2 || !size2)
485 #define AT_STRINGS_END(d) ((d) == end2)
487 /* Checking for end of string */
497 Regexpr::iswordbegin(const sal_Unicode
*d
, sal_Unicode
*string
, sal_Int32 ssize
)
499 if ( d
== string
|| ! ssize
) return true;
501 if ( !unicode::isAlphaDigit(d
[-1]) && unicode::isAlphaDigit(d
[0])) {
508 Regexpr::iswordend(const sal_Unicode
*d
, sal_Unicode
*string
, sal_Int32 ssize
)
510 if ( d
== (string
+ssize
) ) return true;
512 if ( !unicode::isAlphaDigit(d
[0]) && unicode::isAlphaDigit(d
[-1])) {
518 /* Push the information about the state we will need
519 if we ever fail back to it.
521 Requires variables fail_stack, regstart, regend, and reg_info
522 be declared. DOUBLE_FAIL_STACK requires `destination'
525 Does `return FAILURE_CODE' if runs out of memory. */
527 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
530 /* Must be int, so when we don't save any registers, the arithmetic \
531 of 0 + -1 isn't done as unsigned. */ \
532 /* Can't be int, since there is not a shred of a guarantee that int \
533 is wide enough to hold a value of something to which pointer can \
535 sal_uInt32 this_reg; \
537 /* Ensure we have enough space allocated for what we will push. */ \
538 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) { \
539 if (!DOUBLE_FAIL_STACK(fail_stack)) \
540 return failure_code; \
543 /* Push the info, starting with the registers. */ \
545 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
547 PUSH_FAILURE_POINTER(regstart[this_reg]); \
549 PUSH_FAILURE_POINTER (regend[this_reg]); \
551 PUSH_FAILURE_ELT(reg_info[this_reg].word); \
554 PUSH_FAILURE_INT(lowest_active_reg); \
556 PUSH_FAILURE_INT(highest_active_reg); \
558 PUSH_FAILURE_POINTER(pattern_place); \
560 PUSH_FAILURE_POINTER(string_place); \
564 /* Pops what PUSH_FAIL_STACK pushes.
566 We restore into the parameters, all of which should be lvalues:
567 STR -- the saved data position.
568 PAT -- the saved pattern position.
569 LOW_REG, HIGH_REG -- the highest and lowest active registers.
570 REGSTART, REGEND -- arrays of string positions.
571 REG_INFO -- array of information about each subexpression.
573 Also assumes the variables `fail_stack' and (if debugging), `bufp',
574 `pend', `string2', and `size2'. */
576 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info) {\
577 sal_uInt32 this_reg; \
578 sal_Unicode *string_temp; \
580 assert(!FAIL_STACK_EMPTY()); \
582 /* Remove failure points and point to how many regs pushed. */ \
583 assert(fail_stack.avail >= NUM_NONREG_ITEMS); \
585 /* If the saved string location is NULL, it came from an \
586 on_failure_keep_string_jump opcode, and we want to throw away the \
587 saved NULL, thus retaining our current position in the string. */ \
588 string_temp = POP_FAILURE_POINTER(); \
589 if (string_temp != NULL) \
590 str = (const sal_Unicode *) string_temp; \
592 pat = (sal_Unicode *) POP_FAILURE_POINTER(); \
594 /* Restore register info. */ \
595 high_reg = (sal_uInt32) POP_FAILURE_INT(); \
597 low_reg = (sal_uInt32) POP_FAILURE_INT(); \
600 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) { \
602 reg_info[this_reg].word = POP_FAILURE_ELT(); \
604 regend[this_reg] = (const sal_Unicode *) POP_FAILURE_POINTER(); \
606 regstart[this_reg] = (const sal_Unicode *) POP_FAILURE_POINTER(); \
608 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) {\
609 reg_info[this_reg].word.integer = 0; \
610 regend[this_reg] = 0; \
611 regstart[this_reg] = 0; \
613 highest_active_reg = high_reg; \
616 set_regs_matched_done = 0; \
617 } /* POP_FAILURE_POINT */
621 Regexpr::extract_number_and_incr( sal_Int32
& destination
, sal_Unicode
*& source
)
623 extract_number(destination
, source
);
630 Regexpr::store_op1(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg
)
632 *loc
= (sal_Unicode
) op
;
633 store_number(loc
+ 1, arg
);
636 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
640 Regexpr::store_op2(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg1
, sal_Int32 arg2
)
642 *loc
= (sal_Unicode
) op
;
643 store_number(loc
+ 1, arg1
);
644 store_number(loc
+ 3, arg2
);
648 Regexpr::insert_op1(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg
, sal_Unicode
*end
)
650 register sal_Unicode
*pfrom
= end
;
651 register sal_Unicode
*pto
= end
+ 3;
653 while (pfrom
!= loc
) {
657 store_op1(op
, loc
, arg
);
661 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
664 Regexpr::insert_op2(re_opcode_t op
, sal_Unicode
*loc
, sal_Int32 arg1
, sal_Int32 arg2
, sal_Unicode
*end
)
666 register sal_Unicode
*pfrom
= end
;
667 register sal_Unicode
*pto
= end
+ 5;
672 store_op2 (op
, loc
, arg1
, arg2
);
675 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
676 after an alternative or a begin-subexpression. We assume there is at
677 least one character before the ^. */
680 Regexpr::at_begline_loc_p(const sal_Unicode
*local_pattern
, const sal_Unicode
*p
)
682 const sal_Unicode
*prev
= p
- 2;
683 sal_Bool prev_prev_backslash
= prev
> local_pattern
&& prev
[-1] == '\\';
686 /* After a subexpression? */
687 (*prev
== (sal_Unicode
)'(' && prev_prev_backslash
)
688 /* After an alternative? */
689 || (*prev
== (sal_Unicode
)'|' && prev_prev_backslash
));
692 /* The dual of at_begline_loc_p. This one is for $. We assume there is
693 at least one character after the $, i.e., `P < PEND'. */
696 Regexpr::at_endline_loc_p(const sal_Unicode
*p
, const sal_Unicode
* /* pend */ )
698 const sal_Unicode
*next
= p
;
699 //sal_Bool next_backslash = *next == (sal_Unicode)'\\';
700 //const sal_Unicode *next_next = p + 1 < pend ? p + 1 : 0;
703 /* Before a subexpression? */
704 *next
== (sal_Unicode
)')'
705 // (next_backslash && next_next && *next_next == (sal_Unicode)')')
706 /* Before an alternative? */
707 || *next
== (sal_Unicode
)'|' );
708 // || (next_backslash && next_next && *next_next == (sal_Unicode)'|'));
712 Regexpr::compile_range(sal_Unicode range_start
, sal_Unicode range_end
, sal_Unicode
*b
)
714 sal_uInt32 this_char
;
716 /* If the start is after the end, the range is empty. */
717 if (range_start
> range_end
)
720 /* Here we see why `this_char' has to be larger than an `sal_Unicode'
721 -- the range is inclusive, so if `range_end' == 0xffff
722 (assuming 16-bit characters), we would otherwise go into an infinite
723 loop, since all characters <= 0xffff. */
724 for (this_char
= range_start
; this_char
<= range_end
; this_char
++) {
725 set_list_bit( sal_Unicode(this_char
), b
);
731 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
732 false if it's not. */
735 Regexpr::group_in_compile_stack(compile_stack_type compile_stack
, sal_uInt32 regnum
)
737 sal_Int32 this_element
;
739 for (this_element
= compile_stack
.avail
- 1;
742 if (compile_stack
.stack
[this_element
].regnum
== regnum
) {
751 Regexpr::Regexpr( const ::com::sun::star::util::SearchOptions
& rOptions
,
752 ::com::sun::star::uno::Reference
<
753 ::com::sun::star::i18n::XExtendedTransliteration
> XTrans
)
758 if ( rOptions
.algorithmType
!= ::com::sun::star::util::SearchAlgorithms_REGEXP
) {
762 if ( rOptions
.searchString
== NULL
||
763 rOptions
.searchString
.getLength() <= 0) {
767 pattern
= (sal_Unicode
*)rOptions
.searchString
.getStr();
768 patsize
= rOptions
.searchString
.getLength();
770 re_max_failures
= 2000;
773 translate
= translit
.is() ? 1 : 0;
777 isIgnoreCase
= ((rOptions
.transliterateFlags
&
778 ::com::sun::star::i18n::TransliterationModules_IGNORE_CASE
) != 0);
780 // Compile Regular expression pattern
781 if ( regcomp() != REG_NOERROR
)
798 // translit->remove();
812 // sets a new line to search in (restore start/end_ptr)
814 Regexpr::set_line(const sal_Unicode
*new_line
, sal_Int32 len
)
820 // main function for searching the pattern
821 // returns negative or startpos and sets regs
823 Regexpr::re_search(struct re_registers
*regs
, sal_Int32 pOffset
)
825 // Check if pattern buffer is NULL
826 if ( bufp
== NULL
) {
837 linelen
= -(linelen
);
838 stoppos
= pOffset
+ 1;
844 sal_Int32 val
= re_match2(regs
, startpos
, stoppos
);
852 // Return success if match found
861 // If match only beginning of string (startpos)
866 // If search match from startpos to startpos+range
867 else if (range
> 0) { // Forward string search
870 } else { // Reverse string search
876 if ( regs
->num_of_match
> 0 )
885 bufp
= (struct re_pattern_buffer
*)malloc(sizeof(struct re_pattern_buffer
));
886 if ( bufp
== NULL
) {
894 //bufp->fastmap = (sal_Unicode*) malloc((1 << BYTEWIDTH) * sizeof(sal_Unicode));
895 // No fastmap with Unicode
896 bufp
->fastmap
= NULL
;
898 return(regex_compile());
902 Regexpr::regex_compile()
904 register sal_Unicode c
, c1
;
905 const sal_Unicode
*p1
;
906 register sal_Unicode
*b
;
908 /* Keeps track of unclosed groups. */
909 compile_stack_type compile_stack
;
911 /* Points to the current (ending) position in the pattern. */
912 const sal_Unicode
*p
= pattern
;
913 const sal_Unicode
*pend
= pattern
+ patsize
;
915 /* Address of the count-byte of the most recently inserted `exactn'
916 command. This makes it possible to tell if a new exact-match
917 character can be added to that command or if the character requires
918 a new `exactn' command. */
919 sal_Unicode
*pending_exact
= 0;
921 /* Address of start of the most recently finished expression.
922 This tells, e.g., postfix * where to find the start of its
923 operand. Reset at the beginning of groups and alternatives. */
924 sal_Unicode
*laststart
= 0;
926 /* Address of beginning of regexp, or inside of last group. */
929 /* Place in the uncompiled pattern (i.e., the {) to
930 which to go back if the interval is invalid. */
931 const sal_Unicode
*beg_interval
;
933 /* Address of the place where a forward jump should go to the end of
934 the containing expression. Each alternative of an `or' -- except the
935 last -- ends with a forward jump of this sort. */
936 sal_Unicode
*fixup_alt_jump
= 0;
938 /* Counts open-groups as they are encountered. Remembered for the
939 matching close-group on the compile stack, so the same register
940 number is put in the stop_memory as the start_memory. */
941 sal_Int32 regnum
= 0;
943 /* Initialize the compile stack. */
944 compile_stack
.stack
= (compile_stack_elt_t
*)malloc(INIT_COMPILE_STACK_SIZE
* sizeof(compile_stack_elt_t
));
945 if (compile_stack
.stack
== NULL
)
948 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
949 compile_stack
.avail
= 0;
951 /* Initialize the pattern buffer. */
952 bufp
->fastmap_accurate
= 0;
955 bufp
->newline_anchor
= 1;
957 /* Set `used' to zero, so that if we return an error, the pattern
958 printer (for debugging) will think there's no pattern. We reset it
962 /* Always count groups. */
965 if (bufp
->allocated
== 0) {
967 /* If zero allocated, but buffer is non-null, try to realloc
968 enough space. This loses if buffer's address is bogus, but
969 that is the user's responsibility. */
970 bufp
->buffer
= (sal_Unicode
*)realloc(bufp
->buffer
, INIT_BUF_SIZE
* sizeof(sal_Unicode
));
971 } else { /* Caller did not allocate a buffer. Do it for them. */
972 bufp
->buffer
= (sal_Unicode
*)malloc(INIT_BUF_SIZE
* sizeof(sal_Unicode
));
974 if (!bufp
->buffer
) FREE_STACK_RETURN(REG_ESPACE
);
976 bufp
->allocated
= INIT_BUF_SIZE
;
979 begalt
= b
= bufp
->buffer
;
981 /* Loop through the uncompiled pattern until we're at the end. */
986 case (sal_Unicode
)'^': {
987 if ( /* If at start of pattern, it's an operator. */
989 /* Otherwise, depends on what's come before. */
990 || at_begline_loc_p(pattern
, p
))
997 case (sal_Unicode
)'$': {
998 if ( /* If at end of pattern, it's an operator. */
1000 /* Otherwise, depends on what's next. */
1001 || at_endline_loc_p(p
, pend
)) {
1009 case (sal_Unicode
)'+':
1010 case (sal_Unicode
)'?':
1011 case (sal_Unicode
)'*':
1012 /* If there is no previous pattern... */
1018 /* Are we optimizing this jump? */
1019 sal_Bool keep_string_p
= false;
1021 /* 1 means zero (many) matches is allowed. */
1022 sal_Unicode zero_times_ok
= 0, many_times_ok
= 0;
1024 /* If there is a sequence of repetition chars, collapse it
1025 down to just one (the right one). We can't combine
1026 interval operators with these because of, e.g., `a{2}*',
1027 which should only match an even number of `a's. */
1030 zero_times_ok
|= c
!= (sal_Unicode
)'+';
1031 many_times_ok
|= c
!= (sal_Unicode
)'?';
1038 if (c
== (sal_Unicode
)'*' || (c
== (sal_Unicode
)'+'
1039 || c
== (sal_Unicode
)'?')) {
1045 /* If we get here, we found another repeat character. */
1048 /* Star, etc. applied to an empty pattern is equivalent
1049 to an empty pattern. */
1054 /* Now we know whether or not zero matches is allowed
1055 and also whether or not two or more matches is allowed. */
1056 if (many_times_ok
) {
1057 /* More than one repetition is allowed, so put in at the
1058 end a backward relative jump from `b' to before the next
1059 jump we're going to put in below (which jumps from
1060 laststart to after this jump).
1062 But if we are at the `*' in the exact sequence `.*\n',
1063 insert an unconditional jump backwards to the .,
1064 instead of the beginning of the loop. This way we only
1065 push a failure point once, instead of every time
1066 through the loop. */
1067 assert(p
- 1 > pattern
);
1069 /* Allocate the space for the jump. */
1070 GET_BUFFER_SPACE(3);
1072 /* We know we are not at the first character of the pattern,
1073 because laststart was nonzero. And we've already
1074 incremented `p', by the way, to be the character after
1075 the `*'. Do we have to do something analogous here
1076 for null bytes, because of RE_DOT_NOT_NULL? */
1077 if (*(p
- 2) == (sal_Unicode
)'.'
1079 && p
< pend
&& *p
== (sal_Unicode
)'\n') {
1081 STORE_JUMP(jump
, b
, laststart
);
1082 keep_string_p
= true;
1084 /* Anything else. */
1085 STORE_JUMP(maybe_pop_jump
, b
, laststart
- 3);
1088 /* We've added more stuff to the buffer. */
1092 /* On failure, jump from laststart to b + 3, which will be the
1093 end of the buffer after this jump is inserted. */
1094 GET_BUFFER_SPACE(3);
1095 INSERT_JUMP(keep_string_p
? on_failure_keep_string_jump
1101 if (!zero_times_ok
) {
1102 /* At least one repetition is required, so insert a
1103 `dummy_failure_jump' before the initial
1104 `on_failure_jump' instruction of the loop. This
1105 effects a skip over that instruction the first time
1106 we hit that loop. */
1107 GET_BUFFER_SPACE(3);
1108 INSERT_JUMP(dummy_failure_jump
, laststart
, laststart
+ 6);
1114 case (sal_Unicode
)'.':
1120 case (sal_Unicode
)'[': {
1121 sal_Bool have_range
= false;
1122 sal_Unicode last_char
= 0xffff;
1123 sal_Unicode first_range
= 0xffff;
1124 sal_Unicode second_range
= 0xffff;
1127 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1129 /* Ensure that we have enough space to push a charset: the
1130 opcode, the length count, and the bitset;
1131 1 + 1 + (1 << BYTEWIDTH) / BYTEWIDTH "bytes" in all. */
1132 bsiz
= 2 + ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1133 GET_BUFFER_SPACE(bsiz
);
1137 /* We test `*p == '^' twice, instead of using an if
1138 statement, so we only need one BUF_PUSH. */
1139 BUF_PUSH (*p
== (sal_Unicode
)'^' ? charset_not
: charset
);
1140 if (*p
== (sal_Unicode
)'^')
1143 /* Remember the first position in the bracket expression. */
1146 /* Push the number of "bytes" in the bitmap. */
1147 BUF_PUSH((1 << BYTEWIDTH
) / BYTEWIDTH
);
1149 /* Clear the whole map. */
1150 memset(b
, 0, ((1 << BYTEWIDTH
) / BYTEWIDTH
) * sizeof(sal_Unicode
));
1152 /* Read in characters and ranges, setting map bits. */
1154 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1158 if ( c
== (sal_Unicode
)'\\' ) {
1162 if ( c
== (sal_Unicode
)'x' ) {
1163 sal_Int32 UniChar
= -1;
1165 GET_HEX_NUMBER(UniChar
);
1166 if (UniChar
< 0 || UniChar
> 0xffff) FREE_STACK_RETURN(REG_BADPAT
);
1167 c
= (sal_Unicode
) UniChar
;
1169 set_list_bit(last_char
, b
);
1172 set_list_bit(last_char
, b
);
1174 } else if (c
== (sal_Unicode
)']') {
1175 /* Could be the end of the bracket expression. If it's
1176 not (i.e., when the bracket expression is `[]' so
1177 far), the ']' character bit gets set way below. */
1179 } else if ( c
== (sal_Unicode
)'-' ) {
1180 if ( !have_range
) {
1181 if ( last_char
!= 0xffff ) {
1182 first_range
= last_char
;
1186 last_char
= (sal_Unicode
)'-';
1187 set_list_bit(last_char
, b
);
1192 /* See if we're at the beginning of a possible character
1194 else if (c
== (sal_Unicode
)':' && p
[-2] == (sal_Unicode
)'[') {
1195 /* Leave room for the null. */
1196 sal_Unicode str
[CHAR_CLASS_MAX_LENGTH
+ 1];
1201 /* If pattern is `[[:'. */
1202 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1207 if ((c
== (sal_Unicode
)':' && *p
== (sal_Unicode
)']') || p
== pend
)
1209 if (c1
< CHAR_CLASS_MAX_LENGTH
)
1212 /* This is in any case an invalid class name. */
1213 str
[0] = (sal_Unicode
)'\0';
1215 str
[c1
] = (sal_Unicode
)'\0';
1217 /* If isn't a word bracketed by `[:' and `:]':
1218 undo the ending character, the letters, and leave
1219 the leading `:' and `[' (but set bits for them). */
1220 if (c
== (sal_Unicode
)':' && *p
== (sal_Unicode
)']') {
1222 // no support for GRAPH, PUNCT, or XDIGIT yet
1223 sal_Bool is_alnum
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"alnum").getStr());
1224 sal_Bool is_alpha
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"alpha").getStr());
1225 sal_Bool is_cntrl
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"cntrl").getStr());
1226 sal_Bool is_digit
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"digit").getStr());
1227 sal_Bool is_lower
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"lower").getStr());
1228 sal_Bool is_print
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"print").getStr());
1229 sal_Bool is_space
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"space").getStr());
1230 sal_Bool is_upper
= STREQ(str
, ::rtl::OUString::createFromAscii((const sal_Char
*)"upper").getStr());
1232 if (!(is_alnum
|| is_alpha
|| is_cntrl
||
1233 is_digit
|| is_lower
|| is_print
|| is_space
|| is_upper
) )
1234 FREE_STACK_RETURN(REG_ECTYPE
);
1236 /* Throw away the ] at the end of the character
1240 if (p
== pend
) FREE_STACK_RETURN(REG_EBRACK
);
1242 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++) {
1243 /* This was split into 3 if's to
1244 avoid an arbitrary limit in some compiler. */
1245 if ( (is_alnum
&& unicode::isAlphaDigit(sal_Unicode(ch
))) ||
1246 (is_alpha
&& unicode::isAlpha(sal_Unicode(ch
))) ||
1247 (is_cntrl
&& unicode::isControl(sal_Unicode(ch
))))
1248 set_list_bit(sal_Unicode(ch
), b
);
1249 if ( (is_digit
&& unicode::isDigit(sal_Unicode(ch
))) ||
1250 (is_lower
&& unicode::isLower(sal_Unicode(ch
))) ||
1251 (is_print
&& unicode::isPrint(sal_Unicode(ch
))))
1252 set_list_bit(sal_Unicode(ch
), b
);
1253 if ( (is_space
&& unicode::isSpace(sal_Unicode(ch
))) ||
1254 (is_upper
&& unicode::isUpper(sal_Unicode(ch
))) )
1255 set_list_bit(sal_Unicode(ch
), b
);
1256 if ( isIgnoreCase
&& (is_upper
|| is_lower
) &&
1257 (unicode::isUpper(sal_Unicode(ch
)) || unicode::isLower(sal_Unicode(ch
))))
1258 set_list_bit(sal_Unicode(ch
), b
);
1263 last_char
= (sal_Unicode
)':';
1264 set_list_bit(last_char
, b
);
1268 set_list_bit(last_char
, b
);
1271 if ( last_char
!= 0xffff ) {
1272 second_range
= last_char
;
1274 compile_range(first_range
, second_range
, b
);
1275 } else FREE_STACK_RETURN(REG_EBRACK
);
1277 if ( last_char
!= 0xffff ) {
1278 set_list_bit(last_char
, b
);
1279 } else FREE_STACK_RETURN(REG_EBRACK
);
1283 /* Discard any (non)matching list bytes that are all 0 at the
1284 end of the map. Decrease the map-length byte too. */
1286 while ((sal_Int16
) bsiz
> 0 && b
[bsiz
- 1] == 0)
1288 b
[-1] = (sal_Unicode
)bsiz
;
1293 case (sal_Unicode
)'(':
1296 case (sal_Unicode
)')':
1299 case (sal_Unicode
)'\n':
1302 case (sal_Unicode
)'|':
1305 case (sal_Unicode
)'{':
1306 goto handle_interval
;
1308 case (sal_Unicode
)'\\':
1309 if (p
== pend
) FREE_STACK_RETURN(REG_EESCAPE
);
1311 /* Do not translate the character after the \, so that we can
1312 distinguish, e.g., \B from \b, even if we normally would
1313 translate, e.g., B to b. */
1317 case (sal_Unicode
)'(':
1318 goto normal_backslash
;
1324 if (COMPILE_STACK_FULL
) {
1325 compile_stack
.stack
= (compile_stack_elt_t
*)realloc(compile_stack
.stack
, (compile_stack
.size
<< 1) * sizeof(compile_stack_elt_t
));
1326 if (compile_stack
.stack
== NULL
) return(REG_ESPACE
);
1328 compile_stack
.size
<<= 1;
1331 /* These are the values to restore when we hit end of this
1332 group. They are all relative offsets, so that if the
1333 whole pattern moves because of realloc, they will still
1335 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
1336 COMPILE_STACK_TOP
.fixup_alt_jump
1337 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
1338 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
1339 COMPILE_STACK_TOP
.regnum
= regnum
;
1341 /* We will eventually replace the 0 with the number of
1342 groups inner to this one. But do not push a
1343 start_memory for groups beyond the last one we can
1344 represent in the compiled pattern. */
1345 if (regnum
<= MAX_REGNUM
) {
1346 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
1347 BUF_PUSH_3 (start_memory
, regnum
, 0);
1350 compile_stack
.avail
++;
1355 /* If we've reached MAX_REGNUM groups, then this open
1356 won't actually generate any code, so we'll have to
1357 clear pending_exact explicitly. */
1362 case (sal_Unicode
)')':
1363 goto normal_backslash
;
1365 // unreachable (after goto):
1367 if (COMPILE_STACK_EMPTY
) {
1368 FREE_STACK_RETURN(REG_ERPAREN
);
1373 if (fixup_alt_jump
) {
1374 /* Push a dummy failure point at the end of the
1375 alternative for a possible future
1376 `pop_failure_jump' to pop. See comments at
1377 `push_dummy_failure' in `re_match2'. */
1378 BUF_PUSH(push_dummy_failure
);
1380 /* We allocated space for this jump when we assigned
1381 to `fixup_alt_jump', in the `handle_alt' case below. */
1382 STORE_JUMP(jump_past_alt
, fixup_alt_jump
, b
- 1);
1385 /* See similar code for backslashed left paren above. */
1386 if (COMPILE_STACK_EMPTY
) {
1387 FREE_STACK_RETURN(REG_ERPAREN
);
1390 /* Since we just checked for an empty stack above, this
1391 ``can't happen''. */
1392 assert (compile_stack
.avail
!= 0);
1395 /* We don't just want to restore into `regnum', because
1396 later groups should continue to be numbered higher,
1397 as in `(ab)c(de)' -- the second group is #2. */
1398 sal_Int32 this_group_regnum
;
1400 compile_stack
.avail
--;
1401 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
1403 = COMPILE_STACK_TOP
.fixup_alt_jump
1404 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
1406 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
1407 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
1408 /* If we've reached MAX_REGNUM groups, then this open
1409 won't actually generate any code, so we'll have to
1410 clear pending_exact explicitly. */
1413 /* We're at the end of the group, so now we know how many
1414 groups were inside this one. */
1415 if (this_group_regnum
<= MAX_REGNUM
) {
1416 sal_Unicode
*inner_group_loc
1417 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
1419 *inner_group_loc
= sal::static_int_cast
<sal_Unicode
>( regnum
- this_group_regnum
);
1420 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
1421 regnum
- this_group_regnum
);
1427 case (sal_Unicode
)'|': /* `\|'.
1429 goto normal_backslash
;
1432 /* Insert before the previous alternative a jump which
1433 jumps to this alternative if the former fails. */
1434 GET_BUFFER_SPACE (3);
1435 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
1439 /* The alternative before this one has a jump after it
1440 which gets executed if it gets matched. Adjust that
1441 jump so it will jump to this alternative's analogous
1442 jump (put in below, which in turn will jump to the next
1443 (if any) alternative's such jump, etc.). The last such
1444 jump jumps to the correct final destination. A picture:
1450 If we are at `b', then fixup_alt_jump right now points to a
1451 three-byte space after `a'. We'll put in the jump, set
1452 fixup_alt_jump to right after `b', and leave behind three
1453 bytes which we'll fill in when we get to after `c'. */
1456 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
1458 /* Mark and leave space for a jump after this alternative,
1459 to be filled in later either by next alternative or
1460 when know we're at the end of a series of alternatives. */
1462 GET_BUFFER_SPACE (3);
1470 case (sal_Unicode
)'{':
1471 goto normal_backslash
;
1475 /* allows intervals. */
1476 /* At least (most) this many matches must be made. */
1477 sal_Int32 lower_bound
= -1, upper_bound
= -1;
1479 beg_interval
= p
- 1;
1482 goto unfetch_interval
;
1485 GET_UNSIGNED_NUMBER(lower_bound
);
1487 if (c
== (sal_Unicode
)',') {
1488 GET_UNSIGNED_NUMBER(upper_bound
);
1489 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
1491 /* Interval such as `{1}' => match exactly once. */
1492 upper_bound
= lower_bound
;
1494 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
1495 || lower_bound
> upper_bound
) {
1496 goto unfetch_interval
;
1499 if (c
!= (sal_Unicode
)'}') {
1500 goto unfetch_interval
;
1503 /* We just parsed a valid interval. */
1505 /* If it's invalid to have no preceding re. */
1507 goto unfetch_interval
;
1510 /* If the upper bound is zero, don't want to succeed at
1511 all; jump from `laststart' to `b + 3', which will be
1512 the end of the buffer after we insert the jump. */
1513 if (upper_bound
== 0) {
1514 GET_BUFFER_SPACE(3);
1515 INSERT_JUMP(jump
, laststart
, b
+ 3);
1519 /* Otherwise, we have a nontrivial interval. When
1520 we're all done, the pattern will look like:
1521 set_number_at <jump count> <upper bound>
1522 set_number_at <succeed_n count> <lower bound>
1523 succeed_n <after jump addr> <succeed_n count>
1525 jump_n <succeed_n addr> <jump count>
1526 (The upper bound and `jump_n' are omitted if
1527 `upper_bound' is 1, though.) */
1529 /* If the upper bound is > 1, we need to insert
1530 more at the end of the loop. */
1531 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
1533 GET_BUFFER_SPACE(nbytes
);
1535 /* Initialize lower bound of the `succeed_n', even
1536 though it will be set during matching by its
1537 attendant `set_number_at' (inserted next),
1538 because `re_compile_fastmap' needs to know.
1539 Jump to the `jump_n' we might insert below. */
1540 INSERT_JUMP2(succeed_n
, laststart
,
1541 b
+ 5 + (upper_bound
> 1) * 5,
1545 /* Code to initialize the lower bound. Insert
1546 before the `succeed_n'. The `5' is the last two
1547 bytes of this `set_number_at', plus 3 bytes of
1548 the following `succeed_n'. */
1549 insert_op2(set_number_at
, laststart
, 5, lower_bound
, b
);
1552 if (upper_bound
> 1) {
1553 /* More than one repetition is allowed, so
1554 append a backward jump to the `succeed_n'
1555 that starts this interval.
1557 When we've reached this during matching,
1558 we'll have matched the interval once, so
1559 jump back only `upper_bound - 1' times. */
1560 STORE_JUMP2(jump_n
, b
, laststart
+ 5,
1564 /* The location we want to set is the second
1565 parameter of the `jump_n'; that is `b-2' as
1566 an absolute address. `laststart' will be
1567 the `set_number_at' we're about to insert;
1568 `laststart+3' the number to set, the source
1569 for the relative address. But we are
1570 inserting into the middle of the pattern --
1571 so everything is getting moved up by 5.
1572 Conclusion: (b - 2) - (laststart + 3) + 5,
1573 i.e., b - laststart.
1575 We insert this at the beginning of the loop
1576 so that if we fail during matching, we'll
1577 reinitialize the bounds. */
1578 insert_op2(set_number_at
, laststart
, b
- laststart
,
1579 upper_bound
- 1, b
);
1584 beg_interval
= NULL
;
1589 /* If an invalid interval, match the characters as literals. */
1590 assert (beg_interval
);
1592 beg_interval
= NULL
;
1594 /* normal_char and normal_backslash need `c'. */
1599 case (sal_Unicode
)'`':
1603 case (sal_Unicode
)'\'':
1607 case (sal_Unicode
)'1': case (sal_Unicode
)'2':
1608 case (sal_Unicode
)'3': case (sal_Unicode
)'4':
1609 case (sal_Unicode
)'5': case (sal_Unicode
)'6':
1610 case (sal_Unicode
)'7': case (sal_Unicode
)'8':
1611 case (sal_Unicode
)'9':
1612 c1
= c
- (sal_Unicode
)'0';
1615 FREE_STACK_RETURN(REG_ESUBREG
);
1617 /* Can't back reference to a subexpression if inside of it. */
1618 if (group_in_compile_stack(compile_stack
, (sal_uInt32
) c1
)) {
1623 BUF_PUSH_2(duplicate
, c1
);
1627 case (sal_Unicode
)'+':
1628 case (sal_Unicode
)'?':
1629 goto normal_backslash
;
1631 case (sal_Unicode
)'x': // Unicode char
1633 sal_Int32 UniChar
= -1;
1635 GET_HEX_NUMBER(UniChar
);
1636 if (UniChar
< 0 || UniChar
> 0xffff) FREE_STACK_RETURN(REG_BADPAT
);
1637 c
= (sal_Unicode
) UniChar
;
1640 // break; // unreachable - see goto above
1642 case (sal_Unicode
)'<': // begin Word boundary
1646 case (sal_Unicode
)'>': // end Word boundary
1650 case (sal_Unicode
)'n':
1654 case (sal_Unicode
)'t':
1665 /* Expects the character in `c'. */
1667 /* If no exactn currently being built. */
1668 if ( pending_exact
== NULL
1670 /* If last exactn not at current position. */
1671 || pending_exact
+ *pending_exact
+ 1 != b
1673 /* We have only one sal_Unicode char following the
1674 exactn for the count. */
1675 || *pending_exact
== (1 << BYTEWIDTH
) - 1
1677 /* If followed by a repetition operator. */
1678 || *p
== (sal_Unicode
)'*' || *p
== (sal_Unicode
)'^'
1679 || *p
== (sal_Unicode
)'+' || *p
== (sal_Unicode
)'?'
1680 || *p
== (sal_Unicode
) '{' ) {
1681 /* Start building a new exactn. */
1683 BUF_PUSH_2(exactn
, 0);
1684 pending_exact
= b
- 1;
1689 sal_Unicode tmp
= translit
->transliterateChar2Char(c
);
1692 } catch (::com::sun::star::i18n::MultipleCharsOutputException e
) {
1693 ::rtl::OUString
o2( translit
->transliterateChar2String( c
));
1694 sal_Int32 len2
= o2
.getLength();
1695 const sal_Unicode
* k2
= o2
.getStr();
1696 for (sal_Int32 nmatch
= 0; nmatch
< len2
; nmatch
++) {
1697 BUF_PUSH(k2
[nmatch
]);
1707 } /* while p != pend */
1709 /* Through the pattern now. */
1712 STORE_JUMP(jump_past_alt
, fixup_alt_jump
, b
);
1714 if (!COMPILE_STACK_EMPTY
)
1715 FREE_STACK_RETURN(REG_EPAREN
);
1717 // Assumes no backtracking
1720 if ( compile_stack
.stack
)
1721 free(compile_stack
.stack
);
1722 compile_stack
.stack
= NULL
;
1724 /* We have succeeded; set the length of the buffer. */
1725 bufp
->used
= b
- bufp
->buffer
;
1728 } /* regex_compile */
1730 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
1731 bytes; nonzero otherwise. */
1734 Regexpr::bcmp_translate(const sal_Unicode
*s1
, const sal_Unicode
*s2
, sal_Int32 len
)
1736 for (sal_Int32 nmatch
= 0; nmatch
< len
; nmatch
++) {
1737 if (*s1
++ != *s2
++) {
1746 /* We are passed P pointing to a register number after a start_memory.
1748 Return true if the pattern up to the corresponding stop_memory can
1749 match the empty string, and false otherwise.
1751 If we find the matching stop_memory, sets P to point to one past its number.
1752 Otherwise, sets P to an undefined byte less than or equal to END.
1754 We don't handle duplicates properly (yet). */
1757 Regexpr::group_match_null_string_p(sal_Unicode
**p
, sal_Unicode
*end
, register_info_type
*reg_info
)
1760 /* Point to after the args to the start_memory. */
1761 sal_Unicode
*p1
= *p
+ 2;
1764 /* Skip over opcodes that can match nothing, and return true or
1765 false, as appropriate, when we get to one that can't, or to the
1766 matching stop_memory. */
1768 switch ((re_opcode_t
) *p1
) {
1769 /* Could be either a loop or a series of alternatives. */
1770 case on_failure_jump
:
1772 extract_number_and_incr(mcnt
, p1
);
1774 /* If the next operation is not a jump backwards in the
1778 /* Go through the on_failure_jumps of the alternatives,
1779 seeing if any of the alternatives cannot match nothing.
1780 The last alternative starts with only a jump,
1781 whereas the rest start with on_failure_jump and end
1782 with a jump, e.g., here is the pattern for `a|b|c':
1784 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
1785 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
1788 So, we have to first go through the first (n-1)
1789 alternatives and then deal with the last one separately. */
1792 /* Deal with the first (n-1) alternatives, which start
1793 with an on_failure_jump (see above) that jumps to right
1794 past a jump_past_alt. */
1796 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
) {
1797 /* `mcnt' holds how many bytes long the alternative
1798 is, including the ending `jump_past_alt' and
1801 if (!alt_match_null_string_p(p1
, p1
+ mcnt
- 3, reg_info
))
1804 /* Move to right after this alternative, including the
1808 /* Break if it's the beginning of an n-th alternative
1809 that doesn't begin with an on_failure_jump. */
1810 if ((re_opcode_t
) *p1
!= on_failure_jump
)
1813 /* Still have to check that it's not an n-th
1814 alternative that starts with an on_failure_jump. */
1816 extract_number_and_incr(mcnt
, p1
);
1817 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
) {
1818 /* Get to the beginning of the n-th alternative. */
1824 /* Deal with the last alternative: go back and get number
1825 of the `jump_past_alt' just before it. `mcnt' contains
1826 the length of the alternative. */
1827 extract_number(mcnt
, p1
- 2);
1829 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
1832 p1
+= mcnt
; /* Get past the n-th alternative. */
1838 assert (p1
[1] == **p
);
1844 if (!common_op_match_null_string_p(&p1
, end
, reg_info
))
1847 } /* while p1 < end */
1850 } /* group_match_null_string_p */
1852 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
1853 It expects P to be the first byte of a single alternative and END one
1854 byte past the last. The alternative can contain groups. */
1857 Regexpr::alt_match_null_string_p(sal_Unicode
*p
, sal_Unicode
*end
, register_info_type
*reg_info
)
1860 sal_Unicode
*p1
= p
;
1863 /* Skip over opcodes that can match nothing, and break when we get
1864 to one that can't. */
1866 switch ((re_opcode_t
) *p1
) {
1868 case on_failure_jump
:
1870 extract_number_and_incr(mcnt
, p1
);
1875 if (!common_op_match_null_string_p(&p1
, end
, reg_info
))
1878 } /* while p1 < end */
1881 } /* alt_match_null_string_p */
1884 /* Deals with the ops common to group_match_null_string_p and
1885 alt_match_null_string_p.
1887 Sets P to one after the op and its arguments, if any. */
1890 Regexpr::common_op_match_null_string_p(sal_Unicode
**p
, sal_Unicode
*end
, register_info_type
*reg_info
)
1895 sal_Unicode
*p1
= *p
;
1897 switch ((re_opcode_t
) *p1
++) {
1907 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
1908 ret
= group_match_null_string_p(&p1
, end
, reg_info
);
1909 /* Have to set this here in case we're checking a group which
1910 contains a group and a back reference to it. */
1912 if (REG_MATCH_NULL_STRING_P(reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
1913 REG_MATCH_NULL_STRING_P(reg_info
[reg_no
]) = ret
;
1919 /* If this is an optimized succeed_n for zero times, make the jump. */
1921 extract_number_and_incr(mcnt
, p1
);
1929 /* Get to the number of times to succeed. */
1931 extract_number_and_incr(mcnt
, p1
);
1936 extract_number_and_incr(mcnt
, p1
);
1944 if (!REG_MATCH_NULL_STRING_P(reg_info
[*p1
]))
1952 /* All other opcodes mean we cannot match the empty string. */
1958 } /* common_op_match_null_string_p */
1962 /* Free everything we malloc. */
1963 #ifdef MATCH_MAY_ALLOCATE
1964 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
1965 # define FREE_VARIABLES() \
1967 REGEX_FREE_STACK (fail_stack.stack); \
1968 FREE_VAR (regstart); \
1969 FREE_VAR (regend); \
1970 FREE_VAR (old_regstart); \
1971 FREE_VAR (old_regend); \
1972 FREE_VAR (best_regstart); \
1973 FREE_VAR (best_regend); \
1974 FREE_VAR (reg_info); \
1975 FREE_VAR (reg_dummy); \
1976 FREE_VAR (reg_info_dummy); \
1979 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
1980 #endif /* not MATCH_MAY_ALLOCATE */
1982 /* This is a separate function so that we can force an alloca cleanup
1985 Regexpr::re_match2(struct re_registers
*regs
, sal_Int32 pos
, sal_Int32 range
)
1987 /* General temporaries. */
1991 /* Just past the end of the corresponding string. */
1994 /* Pointers into string2, just past the last characters in
1995 each to consider matching. */
1996 sal_Unicode
*end_match_2
;
1998 /* Where we are in the data, and the end of the current string. */
1999 const sal_Unicode
*d
, *dend
;
2001 /* Where we are in the compiled pattern, and the end of the compiled
2003 sal_Unicode
*p
= bufp
->buffer
;
2004 register sal_Unicode
*pend
= p
+ bufp
->used
;
2006 /* Mark the opcode just after a start_memory, so we can test for an
2007 empty subpattern when we get to the stop_memory. */
2008 sal_Unicode
*just_past_start_mem
= 0;
2010 /* Failure point stack. Each place that can handle a failure further
2011 down the line pushes a failure point on this stack. It consists of
2012 restart, regend, and reg_info for all registers corresponding to
2013 the subexpressions we're currently inside, plus the number of such
2014 registers, and, finally, two sal_Unicode *'s. The first
2015 sal_Unicode * is where to resume scanning the pattern; the second
2016 one is where to resume scanning the strings. If the latter is
2017 zero, the failure point is a ``dummy''; if a failure happens and
2018 the failure point is a dummy, it gets discarded and the next next
2020 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
2021 fail_stack_type fail_stack
;
2024 /* We fill all the registers internally, independent of what we
2025 return, for use in backreferences. The number here includes
2026 an element for register zero. */
2027 size_t num_regs
= bufp
->re_nsub
+ 1;
2029 /* The currently active registers. */
2030 sal_uInt32 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
2031 sal_uInt32 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
2033 /* Information on the contents of registers. These are pointers into
2034 the input strings; they record just what was matched (on this
2035 attempt) by a subexpression part of the pattern, that is, the
2036 regnum-th regstart pointer points to where in the pattern we began
2037 matching and the regnum-th regend points to right after where we
2038 stopped matching the regnum-th subexpression. (The zeroth register
2039 keeps track of what the whole pattern matches.) */
2040 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2041 const sal_Unicode
**regstart
, **regend
;
2044 /* If a group that's operated upon by a repetition operator fails to
2045 match anything, then the register for its start will need to be
2046 restored because it will have been set to wherever in the string we
2047 are when we last see its open-group operator. Similarly for a
2049 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2050 const sal_Unicode
**old_regstart
, **old_regend
;
2053 /* The is_active field of reg_info helps us keep track of which (possibly
2054 nested) subexpressions we are currently in. The matched_something
2055 field of reg_info[reg_num] helps us tell whether or not we have
2056 matched any of the pattern so far this time through the reg_num-th
2057 subexpression. These two fields get reset each time through any
2058 loop their register is in. */
2059 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
2060 register_info_type
*reg_info
;
2063 /* The following record the register info as found in the above
2064 variables when we find a match better than any we've seen before.
2065 This happens as we backtrack through the failure points, which in
2066 turn happens only if we have not yet matched the entire string. */
2067 //unsigned best_regs_set = false;
2068 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2069 const sal_Unicode
**best_regstart
, **best_regend
;
2072 /* Logically, this is `best_regend[0]'. But we don't want to have to
2073 allocate space for that if we're not allocating space for anything
2074 else (see below). Also, we never need info about register 0 for
2075 any of the other register vectors, and it seems rather a kludge to
2076 treat `best_regend' differently than the rest. So we keep track of
2077 the end of the best match so far in a separate variable. We
2078 initialize this to NULL so that when we backtrack the first time
2079 and need to test it, it's not garbage. */
2080 //const sal_Unicode *match_end = NULL;
2082 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
2083 sal_Int32 set_regs_matched_done
= 0;
2085 /* Used when we pop values we don't care about. */
2086 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
2087 const sal_Unicode
**reg_dummy
;
2088 register_info_type
*reg_info_dummy
;
2093 #ifdef MATCH_MAY_ALLOCATE
2094 /* Do not bother to initialize all the register variables if there are
2095 no groups in the pattern, as it takes a fair amount of time. If
2096 there are groups, we include space for register 0 (the whole
2097 pattern), even though we never use it, since it simplifies the
2098 array indexing. We should fix this. */
2101 regstart
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2102 regend
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2103 old_regstart
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2104 old_regend
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2105 best_regstart
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2106 best_regend
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2107 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
2108 reg_dummy
= REGEX_TALLOC (num_regs
, const sal_Unicode
*);
2109 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
2111 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
2112 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
2120 /* We must initialize all our variables to NULL, so that
2121 `FREE_VARIABLES' doesn't try to free them. */
2122 regstart
= regend
= old_regstart
= old_regend
= best_regstart
2123 = best_regend
= reg_dummy
= NULL
;
2124 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
2126 #endif /* MATCH_MAY_ALLOCATE */
2128 sal_Unicode
*string2
= (sal_Unicode
*)line
;
2129 sal_Int32 size2
= linelen
;
2130 sal_Int32 stop
= range
;
2132 /* The starting position is bogus. */
2133 if (pos
< 0 || pos
>= size2
|| linelen
<= 0 ) {
2138 /* Initialize subexpression text positions to -1 to mark ones that no
2139 start_memory/stop_memory has been seen for. Also initialize the
2140 register information struct. */
2141 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++) {
2142 regstart
[mcnt
] = regend
[mcnt
]
2143 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
2145 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
2146 IS_ACTIVE (reg_info
[mcnt
]) = 0;
2147 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
2148 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
2151 end2
= (sal_Unicode
*)(string2
+ size2
);
2153 end_match_2
= (sal_Unicode
*)(string2
+ stop
);
2155 /* `p' scans through the pattern as `d' scans through the data.
2156 `dend' is the end of the input string that `d' points within. `d'
2157 is advanced into the following input string whenever necessary, but
2158 this happens before fetching; therefore, at the beginning of the
2159 loop, `d' can be pointing at the end of a string, but it cannot
2164 /* This loops over pattern commands. It exits by returning from the
2165 function if the match is complete, or it drops through if the match
2166 fails at this starting point in the input data. */
2169 /* End of pattern means we might have succeeded. */
2171 /* If we haven't matched the entire string, and we want the
2172 longest match, try backtracking. */
2173 if (d
!= end_match_2
) {
2174 if (!FAIL_STACK_EMPTY()) {
2177 } /* d != end_match_2 */
2181 /* If caller wants register contents data back, do it. */
2183 /* Have the register data arrays been allocated? */
2184 if (regs
->num_regs
== 0) {
2185 /* No. So allocate them with malloc. We need one
2186 extra element beyond `num_regs' for the `-1' marker
2188 regs
->num_of_match
= 0;
2189 regs
->num_regs
= MAX(RE_NREGS
, num_regs
+ 1);
2190 regs
->start
= (sal_Int32
*) malloc(regs
->num_regs
* sizeof(sal_Int32
));
2191 regs
->end
= (sal_Int32
*) malloc(regs
->num_regs
* sizeof(sal_Int32
));
2192 if (regs
->start
== NULL
|| regs
->end
== NULL
) {
2196 } else if ( regs
->num_regs
> 0 ) {
2197 /* Yes. If we need more elements than were already
2198 allocated, reallocate them. If we need fewer, just
2200 if (regs
->num_regs
< num_regs
+ 1) {
2201 regs
->num_regs
= num_regs
+ 1;
2202 regs
->start
= (sal_Int32
*) realloc(regs
->start
, regs
->num_regs
* sizeof(sal_Int32
));
2203 regs
->end
= (sal_Int32
*) realloc(regs
->end
, regs
->num_regs
* sizeof(sal_Int32
));
2204 if (regs
->start
== NULL
|| regs
->end
== NULL
) {
2209 } else { // num_regs is negative
2214 /* Convert the pointer data in `regstart' and `regend' to
2215 indices. Register zero has to be set differently,
2216 since we haven't kept track of any info for it. */
2217 if (regs
->num_regs
> 0) {
2218 // Make sure a valid location
2219 sal_Int32 dpos
= d
- string2
;
2220 if (pos
== dpos
|| (d
- 1) >= dend
) {
2224 regs
->start
[regs
->num_of_match
] = pos
;
2225 regs
->end
[regs
->num_of_match
] = ((sal_Int32
) (d
- string2
));
2226 regs
->num_of_match
++;
2229 /* Go through the first `min (num_regs, regs->num_regs)'
2230 registers, since that is all we initialized. */
2231 for (mcnt
= regs
->num_of_match
; (unsigned) mcnt
< MIN(num_regs
, regs
->num_regs
);
2233 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
2234 if( !(REG_UNSET(regstart
[mcnt
]) || REG_UNSET(regend
[mcnt
])) ) {
2235 regs
->start
[regs
->num_of_match
] = (sal_Int32
) POINTER_TO_OFFSET(regstart
[mcnt
]);
2236 regs
->end
[regs
->num_of_match
] = (sal_Int32
) POINTER_TO_OFFSET(regend
[mcnt
]);
2237 regs
->num_of_match
++;
2241 /* If the regs structure we return has more elements than
2242 were in the pattern, set the extra elements to -1. If
2243 we (re)allocated the registers, this is the case,
2244 because we always allocate enough to have at least one
2246 for (mcnt
= regs
->num_of_match
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
2247 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
2250 mcnt
= d
- pos
- string2
;
2255 /* Otherwise match next pattern command. */
2256 switch ((re_opcode_t
) *p
++) {
2257 /* Ignore these. Used to ignore the n of succeed_n's which
2258 currently have n == 0. */
2265 /* Match the next n pattern characters exactly. The following
2266 byte in the pattern defines n, and the n bytes after that
2267 are the characters to match. */
2273 if ((sal_Unicode
)*d
++ != (sal_Unicode
) *p
++) goto fail
;
2278 /* Match any character except possibly a newline or a null. */
2282 if ( *d
== (sal_Unicode
)'\n' ||
2283 *d
== (sal_Unicode
)'\000' )
2292 register sal_Unicode c
;
2293 sal_Bool knot
= (re_opcode_t
) *(p
- 1) == charset_not
;
2296 c
= *d
; /* The character to match. */
2297 /* Cast to `sal_uInt32' instead of `sal_Unicode' in case the
2298 bit list is a full 32 bytes long. */
2299 if ((c
< (sal_uInt32
) (*p
* BYTEWIDTH
)) && (p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
))))
2313 /* The beginning of a group is represented by start_memory.
2314 The arguments are the register number in the next byte, and the
2315 number of groups inner to this one in the next. The text
2316 matched within the group is recorded (in the internal
2317 registers data structure) under the register number. */
2320 /* Find out if this group can match the empty string. */
2321 p1
= p
; /* To send to group_match_null_string_p. */
2323 if (REG_MATCH_NULL_STRING_P(reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
2324 REG_MATCH_NULL_STRING_P(reg_info
[*p
]) = group_match_null_string_p(&p1
, pend
, reg_info
);
2326 /* Save the position in the string where we were the last time
2327 we were at this open-group operator in case the group is
2328 operated upon by a repetition operator, e.g., with `(a*)*b'
2329 against `ab'; then we want to ignore where we are now in
2330 the string in case this attempt to match fails. */
2331 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P(reg_info
[*p
])
2332 ? REG_UNSET(regstart
[*p
]) ? d
: regstart
[*p
]
2337 IS_ACTIVE (reg_info
[*p
]) = 1;
2338 MATCHED_SOMETHING(reg_info
[*p
]) = 0;
2340 /* Clear this whenever we change the register activity status. */
2341 set_regs_matched_done
= 0;
2343 /* This is the new highest active register. */
2344 highest_active_reg
= *p
;
2346 /* If nothing was active before, this is the new lowest active
2348 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
2349 lowest_active_reg
= *p
;
2351 /* Move past the register number and inner group count. */
2353 just_past_start_mem
= p
;
2357 /* The stop_memory opcode represents the end of a group. Its
2358 arguments are the same as start_memory's: the register
2359 number, and the number of inner groups. */
2362 /* We need to save the string position the last time we were at
2363 this close-group operator in case the group is operated
2364 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
2365 against `aba'; then we want to ignore where we are now in
2366 the string in case this attempt to match fails. */
2367 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
2368 ? REG_UNSET(regend
[*p
]) ? d
: regend
[*p
]
2373 /* This register isn't active anymore. */
2374 IS_ACTIVE(reg_info
[*p
]) = 0;
2376 /* Clear this whenever we change the register activity status. */
2377 set_regs_matched_done
= 0;
2379 /* If this was the only register active, nothing is active
2381 if (lowest_active_reg
== highest_active_reg
) {
2382 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
2383 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
2384 } else { /* We must scan for the new highest active register, since
2385 it isn't necessarily one less than now: consider
2386 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
2387 new highest active register is 1. */
2388 sal_Unicode r
= *p
- 1;
2389 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
2392 /* If we end up at register zero, that means that we saved
2393 the registers as the result of an `on_failure_jump', not
2394 a `start_memory', and we jumped to past the innermost
2395 `stop_memory'. For example, in ((.)*) we save
2396 registers 1 and 2 as a result of the *, but when we pop
2397 back to the second ), we are at the stop_memory 1.
2398 Thus, nothing is active. */
2400 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
2401 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
2403 highest_active_reg
= r
;
2406 /* If just failed to match something this time around with a
2407 group that's operated on by a repetition operator, try to
2408 force exit from the ``loop'', and restore the register
2409 information for this group that we had before trying this
2411 if ((!MATCHED_SOMETHING (reg_info
[*p
])
2412 || just_past_start_mem
== p
- 1)
2413 && (p
+ 2) < pend
) {
2414 sal_Bool is_a_jump_n
= false;
2418 switch ((re_opcode_t
) *p1
++) {
2421 case pop_failure_jump
:
2422 case maybe_pop_jump
:
2424 case dummy_failure_jump
:
2425 extract_number_and_incr(mcnt
, p1
);
2435 /* If the next operation is a jump backwards in the pattern
2436 to an on_failure_jump right before the start_memory
2437 corresponding to this stop_memory, exit from the loop
2438 by forcing a failure after pushing on the stack the
2439 on_failure_jump's jump in the pattern, and d. */
2440 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
2441 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
) {
2442 /* If this group ever matched anything, then restore
2443 what its registers were before trying this last
2444 failed match, e.g., with `(a*)*b' against `ab' for
2445 regstart[1], and, e.g., with `((a*)*(b*)*)*'
2446 against `aba' for regend[3].
2448 Also restore the registers for inner groups for,
2449 e.g., `((a*)(b*))*' against `aba' (register 3 would
2450 otherwise get trashed). */
2452 if (EVER_MATCHED_SOMETHING (reg_info
[*p
])) {
2455 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
2457 /* Restore this and inner groups' (if any) registers. */
2458 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
2460 regstart
[r
] = old_regstart
[r
];
2462 /* xx why this test? */
2463 if (old_regend
[r
] >= regstart
[r
])
2464 regend
[r
] = old_regend
[r
];
2468 extract_number_and_incr(mcnt
, p1
);
2469 PUSH_FAILURE_POINT(p1
+ mcnt
, d
, -2);
2475 /* Move past the register number and the inner group count. */
2480 /* \<digit> has been turned into a `duplicate' command which is
2481 followed by the numeric value of <digit> as the register number. */
2484 register const sal_Unicode
*d2
, *dend2
;
2485 sal_Unicode regno
= *p
++; /* Get which register to match against. */
2487 /* Can't back reference a group which we've never matched. */
2488 if (REG_UNSET(regstart
[regno
]) || REG_UNSET(regend
[regno
])) {
2492 /* Where in input to try to start matching. */
2493 d2
= regstart
[regno
];
2495 /* Where to stop matching; if both the place to start and
2496 the place to stop matching are in the same string, then
2497 set to the place to stop, otherwise, for now have to use
2498 the end of the first string. */
2500 dend2
= regend
[regno
];
2502 /* If necessary, advance to next segment in register
2504 while (d2
== dend2
) {
2505 if (dend2
== end_match_2
) break;
2506 if (dend2
== regend
[regno
]) break;
2508 /* At end of register contents => success */
2509 if (d2
== dend2
) break;
2513 /* How many characters left in this segment to match. */
2516 /* Want how many consecutive characters we can match in
2517 one shot, so, if necessary, adjust the count. */
2518 if (mcnt
> dend2
- d2
)
2521 /* Compare that many; failure if mismatch, else move
2524 ? bcmp_translate(d
, d2
, mcnt
)
2525 : memcmp(d
, d2
, mcnt
* sizeof(sal_Unicode
))) {
2528 d
+= mcnt
, d2
+= mcnt
;
2529 /* Do this because we've match some characters. */
2535 /* begline matches the empty string at the beginning of the string
2536 (unless `not_bol' is set in `bufp'), and, if
2537 `newline_anchor' is set, after newlines. */
2540 if (AT_STRINGS_BEG (d
)) {
2541 if (!bufp
->not_bol
) break;
2542 } else if (d
[-1] == '\n' && bufp
->newline_anchor
) {
2545 /* In all other cases, we fail. */
2548 /* endline is the dual of begline. */
2551 if (AT_STRINGS_END(d
)) {
2552 if (!bufp
->not_eol
) break;
2553 } else if (*d
== '\n' && bufp
->newline_anchor
) {
2558 /* Match at the very beginning of the data. */
2560 if (AT_STRINGS_BEG (d
))
2565 /* Match at the very end of the data. */
2567 if (AT_STRINGS_END (d
))
2572 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
2573 pushes NULL as the value for the string on the stack. Then
2574 `pop_failure_point' will keep the current value for the
2575 string, instead of restoring it. To see why, consider
2576 matching `foo\nbar' against `.*\n'. The .* matches the foo;
2577 then the . fails against the \n. But the next thing we want
2578 to do is match the \n against the \n; if we restored the
2579 string value, we would be back at the foo.
2581 Because this is used only in specific cases, we don't need to
2582 check all the things that `on_failure_jump' does, to make
2583 sure the right things get saved on the stack. Hence we don't
2584 share its code. The only reason to push anything on the
2585 stack at all is that otherwise we would have to change
2586 `anychar's code to do something besides goto fail in this
2587 case; that seems worse than this. */
2588 case on_failure_keep_string_jump
:
2590 extract_number_and_incr(mcnt
, p
);
2592 PUSH_FAILURE_POINT(p
+ mcnt
, NULL
, -2);
2596 /* Uses of on_failure_jump:
2598 Each alternative starts with an on_failure_jump that points
2599 to the beginning of the next alternative. Each alternative
2600 except the last ends with a jump that in effect jumps past
2601 the rest of the alternatives. (They really jump to the
2602 ending jump of the following alternative, because tensioning
2603 these jumps is a hassle.)
2605 Repeats start with an on_failure_jump that points past both
2606 the repetition text and either the following jump or
2607 pop_failure_jump back to this on_failure_jump. */
2608 case on_failure_jump
:
2611 extract_number_and_incr(mcnt
, p
);
2613 /* If this on_failure_jump comes right before a group (i.e.,
2614 the original * applied to a group), save the information
2615 for that group and all inner ones, so that if we fail back
2616 to this point, the group's information will be correct.
2617 For example, in \(a*\)*\1, we need the preceding group,
2618 and in \(zz\(a*\)b*\)\2, we need the inner group. */
2620 /* We can't use `p' to check ahead because we push
2621 a failure point to `p + mcnt' after we do this. */
2624 /* We need to skip no_op's before we look for the
2625 start_memory in case this on_failure_jump is happening as
2626 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
2628 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
2631 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
) {
2632 /* We have a new highest active register now. This will
2633 get reset at the start_memory we are about to get to,
2634 but we will have saved all the registers relevant to
2635 this repetition op, as described above. */
2636 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
2637 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
2638 lowest_active_reg
= *(p1
+ 1);
2641 PUSH_FAILURE_POINT(p
+ mcnt
, d
, -2);
2644 /* A smart repeat ends with `maybe_pop_jump'.
2645 We change it to either `pop_failure_jump' or `jump'. */
2646 case maybe_pop_jump
:
2647 extract_number_and_incr(mcnt
, p
);
2649 register sal_Unicode
*p2
= p
;
2651 /* Compare the beginning of the repeat with what in the
2652 pattern follows its end. If we can establish that there
2653 is nothing that they would both match, i.e., that we
2654 would have to backtrack because of (as in, e.g., `a*a')
2655 then we can change to pop_failure_jump, because we'll
2656 never have to backtrack.
2658 This is not true in the case of alternatives: in
2659 `(a|ab)*' we do need to backtrack to the `ab' alternative
2660 (e.g., if the string was `ab'). But instead of trying to
2661 detect that here, the alternative has put on a dummy
2662 failure point which is what we will end up popping. */
2664 /* Skip over open/close-group commands.
2665 If what follows this loop is a ...+ construct,
2666 look at what begins its body, since we will have to
2667 match at least one of that. */
2670 && ((re_opcode_t
) *p2
== stop_memory
2671 || (re_opcode_t
) *p2
== start_memory
))
2673 else if (p2
+ 6 < pend
2674 && (re_opcode_t
) *p2
== dummy_failure_jump
)
2681 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
2682 to the `maybe_finalize_jump' of this case. Examine what
2685 /* If we're at the end of the pattern, we can change. */
2687 /* Consider what happens when matching ":\(.*\)"
2688 against ":/". I don't really understand this code
2690 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2691 } else if ((re_opcode_t
) *p2
== exactn
2692 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
)) {
2693 register sal_Unicode c
= *p2
== (sal_Unicode
) endline
? (sal_Unicode
)'\n' : p2
[2];
2695 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
) {
2696 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2697 } else if ((re_opcode_t
) p1
[3] == charset
2698 || (re_opcode_t
) p1
[3] == charset_not
) {
2699 sal_Int32 knot
= (re_opcode_t
) p1
[3] == charset_not
;
2701 if (c
< (sal_Unicode
) (p1
[4] * BYTEWIDTH
)
2702 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
2705 /* `not' is equal to 1 if c would match, which means
2706 that we can't change to pop_failure_jump. */
2708 p
[-3] = (unsigned char) pop_failure_jump
;
2711 } else if ((re_opcode_t
) *p2
== charset
) {
2712 /* We win if the first character of the loop is not part
2714 if ((re_opcode_t
) p1
[3] == exactn
2715 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
2716 && (p2
[2 + p1
[5] / BYTEWIDTH
]
2717 & (1 << (p1
[5] % BYTEWIDTH
))))) {
2718 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2719 } else if ((re_opcode_t
) p1
[3] == charset_not
) {
2721 /* We win if the charset_not inside the loop
2722 lists every character listed in the charset after. */
2723 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
2724 if (! (p2
[2 + idx
] == 0
2725 || (idx
< (int) p1
[4]
2726 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
2730 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2732 } else if ((re_opcode_t
) p1
[3] == charset
) {
2734 /* We win if the charset inside the loop
2735 has no overlap with the one after the loop. */
2737 idx
< (sal_Int32
) p2
[1] && idx
< (sal_Int32
) p1
[4];
2739 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
2742 if (idx
== p2
[1] || idx
== p1
[4]) {
2743 p
[-3] = (sal_Unicode
) pop_failure_jump
;
2748 p
-= 2; /* Point at relative address again. */
2749 if ((re_opcode_t
) p
[-1] != pop_failure_jump
) {
2750 p
[-1] = (sal_Unicode
) jump
;
2751 goto unconditional_jump
;
2753 /* Note fall through. */
2756 /* The end of a simple repeat has a pop_failure_jump back to
2757 its matching on_failure_jump, where the latter will push a
2758 failure point. The pop_failure_jump takes off failure
2759 points put on by this pop_failure_jump's matching
2760 on_failure_jump; we got through the pattern to here from the
2761 matching on_failure_jump, so didn't fail. */
2762 case pop_failure_jump
:
2764 /* We need to pass separate storage for the lowest and
2765 highest registers, even though we don't care about the
2766 actual values. Otherwise, we will restore only one
2767 register from the stack, since lowest will == highest in
2768 `pop_failure_point'. */
2769 sal_uInt32 dummy_low_reg
, dummy_high_reg
;
2770 sal_Unicode
*pdummy
= NULL
;
2771 const sal_Unicode
*sdummy
= NULL
;
2773 POP_FAILURE_POINT(sdummy
, pdummy
,
2774 dummy_low_reg
, dummy_high_reg
,
2775 reg_dummy
, reg_dummy
, reg_info_dummy
);
2777 /* Note fall through. */
2780 /* Note fall through. */
2782 /* Unconditionally jump (without popping any failure points). */
2784 extract_number_and_incr(mcnt
, p
); /* Get the amount to jump. */
2785 p
+= mcnt
; /* Do the jump. */
2788 /* We need this opcode so we can detect where alternatives end
2789 in `group_match_null_string_p' et al. */
2791 goto unconditional_jump
;
2794 /* Normally, the on_failure_jump pushes a failure point, which
2795 then gets popped at pop_failure_jump. We will end up at
2796 pop_failure_jump, also, and with a pattern of, say, `a+', we
2797 are skipping over the on_failure_jump, so we have to push
2798 something meaningless for pop_failure_jump to pop. */
2799 case dummy_failure_jump
:
2800 /* It doesn't matter what we push for the string here. What
2801 the code at `fail' tests is the value for the pattern. */
2802 PUSH_FAILURE_POINT(NULL
, NULL
, -2);
2803 goto unconditional_jump
;
2806 /* At the end of an alternative, we need to push a dummy failure
2807 point in case we are followed by a `pop_failure_jump', because
2808 we don't want the failure point for the alternative to be
2809 popped. For example, matching `(a|ab)*' against `aab'
2810 requires that we match the `ab' alternative. */
2811 case push_dummy_failure
:
2812 /* See comments just above at `dummy_failure_jump' about the
2814 PUSH_FAILURE_POINT(NULL
, NULL
, -2);
2817 /* Have to succeed matching what follows at least n times.
2818 After that, handle like `on_failure_jump'. */
2820 extract_number(mcnt
, p
+ 2);
2823 /* Originally, this is how many times we HAVE to succeed. */
2827 store_number_and_incr (p
, mcnt
);
2828 } else if (mcnt
== 0) {
2829 p
[2] = (sal_Unicode
) no_op
;
2830 p
[3] = (sal_Unicode
) no_op
;
2836 extract_number(mcnt
, p
+ 2);
2838 /* Originally, this is how many times we CAN jump. */
2841 store_number (p
+ 2, mcnt
);
2842 goto unconditional_jump
;
2844 /* If don't have to jump any more, skip over the rest of command. */
2852 extract_number_and_incr(mcnt
, p
);
2854 extract_number_and_incr(mcnt
, p
);
2855 store_number (p1
, mcnt
);
2860 if (iswordbegin(d
, string2
, size2
))
2865 if (iswordend(d
, string2
, size2
))
2873 continue; /* Successfully executed one pattern command; keep going. */
2875 /* We goto here if a matching operation fails. */
2877 if (!FAIL_STACK_EMPTY()) {
2878 /* A restart point is known. Restore to that state. */
2879 POP_FAILURE_POINT(d
, p
,
2880 lowest_active_reg
, highest_active_reg
,
2881 regstart
, regend
, reg_info
);
2883 /* If this failure point is a dummy, try the next one. */
2887 /* If we failed to the end of the pattern, don't examine *p. */
2890 sal_Bool is_a_jump_n
= false;
2892 /* If failed to a backwards jump that's part of a repetition
2893 loop, need to pop this failure point and use the next
2895 switch ((re_opcode_t
) *p
) {
2898 case maybe_pop_jump
:
2899 case pop_failure_jump
:
2902 extract_number_and_incr(mcnt
, p1
);
2905 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
2907 && (re_opcode_t
) *p1
== on_failure_jump
)) {
2917 break; /* Matching at this starting point really fails. */
2923 return(-1); /* Failure to match. */
2926 /* Set the bit for character C in a list. */
2928 Regexpr::set_list_bit(sal_Unicode c
, sal_Unicode
*b
)
2932 sal_Unicode tmp
= translit
->transliterateChar2Char(c
);
2933 b
[tmp
/ BYTEWIDTH
] |= 1 << (tmp
% BYTEWIDTH
);
2934 } catch (::com::sun::star::i18n::MultipleCharsOutputException e
) {
2935 ::rtl::OUString
o2( translit
->transliterateChar2String( c
));
2936 sal_Int32 len2
= o2
.getLength();
2937 const sal_Unicode
* k2
= o2
.getStr();
2938 for (sal_Int32 nmatch
= 0; nmatch
< len2
; nmatch
++) {
2939 b
[k2
[nmatch
] / BYTEWIDTH
] |= 1 << (k2
[nmatch
] % BYTEWIDTH
);
2943 b
[c
/ BYTEWIDTH
] |= 1 << (c
% BYTEWIDTH
);
2947 /* vim: set ts=8 sw=2 noexpandtab: */