1
/*************************************************
2
* Perl-Compatible Regular Expressions *
3
*************************************************/
5
/* PCRE is a library of functions to support regular expressions whose syntax
6
and semantics are as close as possible to those of the Perl 5 language.
8
Written by Philip Hazel
9
Copyright (c) 1997-2012 University of Cambridge
11
-----------------------------------------------------------------------------
12
Redistribution and use in source and binary forms, with or without
13
modification, are permitted provided that the following conditions are met:
15
* Redistributions of source code must retain the above copyright notice,
16
this list of conditions and the following disclaimer.
18
* Redistributions in binary form must reproduce the above copyright
19
notice, this list of conditions and the following disclaimer in the
20
documentation and/or other materials provided with the distribution.
22
* Neither the name of the University of Cambridge nor the names of its
23
contributors may be used to endorse or promote products derived from
24
this software without specific prior written permission.
26
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36
POSSIBILITY OF SUCH DAMAGE.
37
-----------------------------------------------------------------------------
41
/* This module contains the external function pcre_study(), along with local
42
supporting functions. */
45
#ifdef PCRE_HAVE_CONFIG_H
49
#include "pcre_internal.h"
51
#define SET_BIT(c) start_bits[c/8] |= (1 << (c&7))
53
/* Returns from set_start_bits() */
55
enum { SSB_FAIL, SSB_DONE, SSB_CONTINUE, SSB_UNKNOWN };
59
/*************************************************
60
* Find the minimum subject length for a group *
61
*************************************************/
63
/* Scan a parenthesized group and compute the minimum length of subject that
64
is needed to match it. This is a lower bound; it does not mean there is a
65
string of that length that matches. In UTF8 mode, the result is in characters
69
code pointer to start of group (the bracket)
70
startcode pointer to start of the whole pattern
71
options the compiling options
74
Returns: the minimum length
75
-1 if \C in UTF-8 mode or (*ACCEPT) was encountered
76
-2 internal error (missing capturing bracket)
77
-3 internal error (opcode not listed)
81
find_minlength(const pcre_uchar *code, const pcre_uchar *startcode, int options,
85
/* PCRE_UTF16 has the same value as PCRE_UTF8. */
86
BOOL utf = (options & PCRE_UTF8) != 0;
87
BOOL had_recurse = FALSE;
88
register int branchlength = 0;
89
register pcre_uchar *cc = (pcre_uchar *)code + 1 + LINK_SIZE;
91
if (*code == OP_CBRA || *code == OP_SCBRA ||
92
*code == OP_CBRAPOS || *code == OP_SCBRAPOS) cc += IMM2_SIZE;
94
/* Scan along the opcodes for this branch. If we get to the end of the
95
branch, check the length against that of the other branches. */
101
register int op = *cc;
108
/* If there is only one branch in a condition, the implied branch has zero
109
length, so we don't add anything. This covers the DEFINE "condition"
112
cs = cc + GET(cc, 1);
115
cc = cs + 1 + LINK_SIZE;
119
/* Otherwise we can fall through and treat it the same as any other
132
d = find_minlength(cc, startcode, options, recurse_depth);
135
do cc += GET(cc, 1); while (*cc == OP_ALT);
139
/* ACCEPT makes things far too complicated; we have to give up. */
142
case OP_ASSERT_ACCEPT:
145
/* Reached end of a branch; if it's a ket it is the end of a nested
146
call. If it's ALT it is an alternation in a nested call. If it is END it's
147
the end of the outer call. All can be handled by the same code. If an
148
ACCEPT was previously encountered, use the length that was in force at that
149
time, and pass back the shortest ACCEPT length. */
157
if (length < 0 || (!had_recurse && branchlength < length))
158
length = branchlength;
159
if (op != OP_ALT) return length;
165
/* Skip over assertive subpatterns */
170
case OP_ASSERTBACK_NOT:
171
do cc += GET(cc, 1); while (*cc == OP_ALT);
174
/* Skip over things that don't match chars */
191
case OP_NOT_WORD_BOUNDARY:
192
case OP_WORD_BOUNDARY:
193
cc += PRIV(OP_lengths)[*cc];
196
/* Skip over a subpattern that has a {0} or {0,x} quantifier */
202
cc += PRIV(OP_lengths)[*cc];
203
do cc += GET(cc, 1); while (*cc == OP_ALT);
207
/* Handle literal characters and + repetitions */
228
if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
236
cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2;
239
/* Handle exact repetitions. The count is already in characters, but we
240
need to skip over a multibyte character in UTF8 mode. */
246
branchlength += GET2(cc,1);
249
if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
254
branchlength += GET2(cc,1);
255
cc += 2 + IMM2_SIZE + ((cc[1 + IMM2_SIZE] == OP_PROP
256
|| cc[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
259
/* Handle single-char non-literal matchers */
268
case OP_NOT_WHITESPACE:
270
case OP_NOT_WORDCHAR:
283
/* "Any newline" might match two characters, but it also might match just
291
/* The single-byte matcher means we can't proceed in UTF-8 mode. (In
292
non-UTF-8 mode \C will actually be turned into OP_ALLANY, so won't ever
293
appear, but leave the code, just in case.) */
303
/* For repeated character types, we have to test for \p and \P, which have
304
an extra two bytes of parameters. */
309
case OP_TYPEMINQUERY:
311
case OP_TYPEPOSQUERY:
312
if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
313
cc += PRIV(OP_lengths)[op];
319
if (cc[1 + IMM2_SIZE] == OP_PROP
320
|| cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
321
cc += PRIV(OP_lengths)[op];
324
/* Check a class for variable quantification */
326
#if defined SUPPORT_UTF || !defined COMPILE_PCRE8
328
cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
334
cc += PRIV(OP_lengths)[OP_CLASS];
352
branchlength += GET2(cc,1);
353
cc += 1 + 2 * IMM2_SIZE;
362
/* Backreferences and subroutine calls are treated in the same way: we find
363
the minimum length for the subpattern. A recursion, however, causes an
364
a flag to be set that causes the length of this branch to be ignored. The
365
logic is that a recursion can only make sense if there is another
366
alternation that stops the recursing. That will provide the minimum length
367
(when no recursion happens). A backreference within the group that it is
368
referencing behaves in the same way.
370
If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
371
matches an empty string (by default it causes a matching failure), so in
372
that case we must set the minimum length to zero. */
376
if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
378
ce = cs = (pcre_uchar *)PRIV(find_bracket)(startcode, utf, GET2(cc, 1));
379
if (cs == NULL) return -2;
380
do ce += GET(ce, 1); while (*ce == OP_ALT);
381
if (cc > cs && cc < ce)
388
d = find_minlength(cs, startcode, options, recurse_depth);
394
/* Handle repeated back references */
415
cc += 1 + 2 * IMM2_SIZE;
423
branchlength += min * d;
426
/* We can easily detect direct recursion, but not mutual recursion. This is
427
caught by a recursion depth count. */
430
cs = ce = (pcre_uchar *)startcode + GET(cc, 1);
431
do ce += GET(ce, 1); while (*ce == OP_ALT);
432
if ((cc > cs && cc < ce) || recurse_depth > 10)
436
branchlength += find_minlength(cs, startcode, options, recurse_depth + 1);
441
/* Anything else does not or need not match a character. We can get the
442
item's length from the table, but for those that can match zero occurrences
443
of a character, we must take special action for UTF-8 characters. As it
444
happens, the "NOT" versions of these opcodes are used at present only for
445
ASCII characters, so they could be omitted from this list. However, in
446
future that may change, so we include them here so as not to leave a
447
gotcha for a future maintainer. */
482
case OP_NOTMINQUERYI:
486
case OP_NOTPOSQUERYI:
488
cc += PRIV(OP_lengths)[op];
490
if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
494
/* Skip these, but we need to add in the name length. */
500
cc += PRIV(OP_lengths)[op] + cc[1];
503
/* The remaining opcodes are just skipped over. */
512
cc += PRIV(OP_lengths)[op];
515
/* This should not occur: we list all opcodes explicitly so that when
516
new ones get added they are properly considered. */
522
/* Control never gets here */
527
/*************************************************
528
* Set a bit and maybe its alternate case *
529
*************************************************/
531
/* Given a character, set its first byte's bit in the table, and also the
532
corresponding bit for the other version of a letter if we are caseless. In
533
UTF-8 mode, for characters greater than 127, we can only do the caseless thing
534
when Unicode property support is available.
537
start_bits points to the bit map
538
p points to the character
539
caseless the caseless flag
540
cd the block with char table pointers
541
utf TRUE for UTF-8 / UTF-16 mode
543
Returns: pointer after the character
546
static const pcre_uchar *
547
set_table_bit(pcre_uint8 *start_bits, const pcre_uchar *p, BOOL caseless,
548
compile_data *cd, BOOL utf)
563
c = UCD_OTHERCASE(c);
564
(void)PRIV(ord2utf)(c, buff);
572
/* Not UTF-8 mode, or character is less than 127. */
574
if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
578
#ifdef COMPILE_PCRE16
593
c = UCD_OTHERCASE(c);
603
if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
610
/*************************************************
611
* Set bits for a positive character type *
612
*************************************************/
614
/* This function sets starting bits for a character type. In UTF-8 mode, we can
615
only do a direct setting for bytes less than 128, as otherwise there can be
616
confusion with bytes in the middle of UTF-8 characters. In a "traditional"
617
environment, the tables will only recognize ASCII characters anyway, but in at
618
least one Windows environment, some higher bytes bits were set in the tables.
619
So we deal with that case by considering the UTF-8 encoding.
622
start_bits the starting bitmap
623
cbit type the type of character wanted
624
table_limit 32 for non-UTF-8; 16 for UTF-8
625
cd the block with char table pointers
631
set_type_bits(pcre_uint8 *start_bits, int cbit_type, int table_limit,
635
for (c = 0; c < table_limit; c++) start_bits[c] |= cd->cbits[c+cbit_type];
636
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
637
if (table_limit == 32) return;
638
for (c = 128; c < 256; c++)
640
if ((cd->cbits[c/8] & (1 << (c&7))) != 0)
643
(void)PRIV(ord2utf)(c, buff);
651
/*************************************************
652
* Set bits for a negative character type *
653
*************************************************/
655
/* This function sets starting bits for a negative character type such as \D.
656
In UTF-8 mode, we can only do a direct setting for bytes less than 128, as
657
otherwise there can be confusion with bytes in the middle of UTF-8 characters.
658
Unlike in the positive case, where we can set appropriate starting bits for
659
specific high-valued UTF-8 characters, in this case we have to set the bits for
660
all high-valued characters. The lowest is 0xc2, but we overkill by starting at
661
0xc0 (192) for simplicity.
664
start_bits the starting bitmap
665
cbit type the type of character wanted
666
table_limit 32 for non-UTF-8; 16 for UTF-8
667
cd the block with char table pointers
673
set_nottype_bits(pcre_uint8 *start_bits, int cbit_type, int table_limit,
677
for (c = 0; c < table_limit; c++) start_bits[c] |= ~cd->cbits[c+cbit_type];
678
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
679
if (table_limit != 32) for (c = 24; c < 32; c++) start_bits[c] = 0xff;
685
/*************************************************
686
* Create bitmap of starting bytes *
687
*************************************************/
689
/* This function scans a compiled unanchored expression recursively and
690
attempts to build a bitmap of the set of possible starting bytes. As time goes
691
by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
692
useful for parenthesized groups in patterns such as (a*)b where the group
693
provides some optional starting bytes but scanning must continue at the outer
694
level to find at least one mandatory byte. At the outermost level, this
695
function fails unless the result is SSB_DONE.
698
code points to an expression
699
start_bits points to a 32-byte table, initialized to 0
700
utf TRUE if in UTF-8 / UTF-16 mode
701
cd the block with char table pointers
703
Returns: SSB_FAIL => Failed to find any starting bytes
704
SSB_DONE => Found mandatory starting bytes
705
SSB_CONTINUE => Found optional starting bytes
706
SSB_UNKNOWN => Hit an unrecognized opcode
710
set_start_bits(const pcre_uchar *code, pcre_uint8 *start_bits, BOOL utf,
714
int yield = SSB_DONE;
715
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
716
int table_limit = utf? 16:32;
718
int table_limit = 32;
722
/* ========================================================================= */
723
/* The following comment and code was inserted in January 1999. In May 2006,
724
when it was observed to cause compiler warnings about unused values, I took it
725
out again. If anybody is still using OS/2, they will have to put it back
728
/* This next statement and the later reference to dummy are here in order to
729
trick the optimizer of the IBM C compiler for OS/2 into generating correct
730
code. Apparently IBM isn't going to fix the problem, and we would rather not
731
disable optimization (in this module it actually makes a big difference, and
732
the pcre module can use all the optimization it can get). */
735
/* ========================================================================= */
740
BOOL try_next = TRUE;
741
const pcre_uchar *tcode = code + 1 + LINK_SIZE;
743
if (*code == OP_CBRA || *code == OP_SCBRA ||
744
*code == OP_CBRAPOS || *code == OP_SCBRAPOS) tcode += IMM2_SIZE;
746
while (try_next) /* Loop for items in this branch */
752
/* If we reach something we don't understand, it means a new opcode has
753
been created that hasn't been added to this code. Hopefully this problem
754
will be discovered during testing. */
759
/* Fail for a valid opcode that implies no starting bits. */
762
case OP_ASSERT_ACCEPT:
789
case OP_NOTMINQUERYI:
799
case OP_NOTPOSQUERYI:
830
#if defined SUPPORT_UTF || !defined COMPILE_PCRE8
835
/* We can ignore word boundary tests. */
837
case OP_WORD_BOUNDARY:
838
case OP_NOT_WORD_BOUNDARY:
842
/* If we hit a bracket or a positive lookahead assertion, recurse to set
843
bits from within the subpattern. If it can't find anything, we have to
844
give up. If it finds some mandatory character(s), we are done for this
845
branch. Otherwise, carry on scanning after the subpattern. */
858
rc = set_start_bits(tcode, start_bits, utf, cd);
859
if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
860
if (rc == SSB_DONE) try_next = FALSE; else
862
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
863
tcode += 1 + LINK_SIZE;
867
/* If we hit ALT or KET, it means we haven't found anything mandatory in
868
this branch, though we might have found something optional. For ALT, we
869
continue with the next alternative, but we have to arrange that the final
870
result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
871
return SSB_CONTINUE: if this is the top level, that indicates failure,
872
but after a nested subpattern, it causes scanning to continue. */
875
yield = SSB_CONTINUE;
885
/* Skip over callout */
888
tcode += 2 + 2*LINK_SIZE;
891
/* Skip over lookbehind and negative lookahead assertions */
895
case OP_ASSERTBACK_NOT:
896
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
897
tcode += 1 + LINK_SIZE;
900
/* BRAZERO does the bracket, but carries on. */
905
rc = set_start_bits(++tcode, start_bits, utf, cd);
906
if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
907
/* =========================================================================
908
See the comment at the head of this function concerning the next line,
909
which was an old fudge for the benefit of OS/2.
911
========================================================================= */
912
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
913
tcode += 1 + LINK_SIZE;
916
/* SKIPZERO skips the bracket. */
920
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
921
tcode += 1 + LINK_SIZE;
924
/* Single-char * or ? sets the bit and tries the next item */
932
tcode = set_table_bit(start_bits, tcode + 1, FALSE, cd, utf);
941
tcode = set_table_bit(start_bits, tcode + 1, TRUE, cd, utf);
944
/* Single-char upto sets the bit and tries the next */
949
tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, FALSE, cd, utf);
955
tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, TRUE, cd, utf);
958
/* At least one single char sets the bit and stops */
967
(void)set_table_bit(start_bits, tcode + 1, FALSE, cd, utf);
978
(void)set_table_bit(start_bits, tcode + 1, TRUE, cd, utf);
982
/* Special spacing and line-terminating items. These recognize specific
983
lists of characters. The difference between VSPACE and ANYNL is that the
984
latter can match the two-character CRLF sequence, but that is not
985
relevant for finding the first character, so their code here is
995
SET_BIT(0xC2); /* For U+00A0 */
996
SET_BIT(0xE1); /* For U+1680, U+180E */
997
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
998
SET_BIT(0xE3); /* For U+3000 */
1000
#ifdef COMPILE_PCRE16
1002
SET_BIT(0xFF); /* For characters > 255 */
1006
#endif /* SUPPORT_UTF */
1009
#ifdef COMPILE_PCRE16
1010
SET_BIT(0xFF); /* For characters > 255 */
1025
#ifdef COMPILE_PCRE8
1026
SET_BIT(0xC2); /* For U+0085 */
1027
SET_BIT(0xE2); /* For U+2028, U+2029 */
1029
#ifdef COMPILE_PCRE16
1031
SET_BIT(0xFF); /* For characters > 255 */
1035
#endif /* SUPPORT_UTF */
1038
#ifdef COMPILE_PCRE16
1039
SET_BIT(0xFF); /* For characters > 255 */
1045
/* Single character types set the bits and stop. Note that if PCRE_UCP
1046
is set, we do not see these op codes because \d etc are converted to
1047
properties. Therefore, these apply in the case when only characters less
1048
than 256 are recognized to match the types. */
1051
set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
1056
set_type_bits(start_bits, cbit_digit, table_limit, cd);
1060
/* The cbit_space table has vertical tab as whitespace; we have to
1061
ensure it is set as not whitespace. */
1063
case OP_NOT_WHITESPACE:
1064
set_nottype_bits(start_bits, cbit_space, table_limit, cd);
1065
start_bits[1] |= 0x08;
1069
/* The cbit_space table has vertical tab as whitespace; we have to
1070
not set it from the table. */
1073
c = start_bits[1]; /* Save in case it was already set */
1074
set_type_bits(start_bits, cbit_space, table_limit, cd);
1075
start_bits[1] = (start_bits[1] & ~0x08) | c;
1079
case OP_NOT_WORDCHAR:
1080
set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1085
set_type_bits(start_bits, cbit_word, table_limit, cd);
1089
/* One or more character type fudges the pointer and restarts, knowing
1090
it will hit a single character type and stop there. */
1093
case OP_TYPEMINPLUS:
1094
case OP_TYPEPOSPLUS:
1099
tcode += 1 + IMM2_SIZE;
1102
/* Zero or more repeats of character types set the bits and then
1106
case OP_TYPEMINUPTO:
1107
case OP_TYPEPOSUPTO:
1108
tcode += IMM2_SIZE; /* Fall through */
1111
case OP_TYPEMINSTAR:
1112
case OP_TYPEPOSSTAR:
1114
case OP_TYPEMINQUERY:
1115
case OP_TYPEPOSQUERY:
1126
#ifdef COMPILE_PCRE8
1129
#ifdef COMPILE_PCRE8
1130
SET_BIT(0xC2); /* For U+00A0 */
1131
SET_BIT(0xE1); /* For U+1680, U+180E */
1132
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
1133
SET_BIT(0xE3); /* For U+3000 */
1135
#ifdef COMPILE_PCRE16
1137
SET_BIT(0xFF); /* For characters > 255 */
1141
#endif /* SUPPORT_UTF */
1151
#ifdef COMPILE_PCRE8
1154
#ifdef COMPILE_PCRE8
1155
SET_BIT(0xC2); /* For U+0085 */
1156
SET_BIT(0xE2); /* For U+2028, U+2029 */
1158
#ifdef COMPILE_PCRE16
1160
SET_BIT(0xFF); /* For characters > 255 */
1164
#endif /* SUPPORT_UTF */
1169
set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
1173
set_type_bits(start_bits, cbit_digit, table_limit, cd);
1176
/* The cbit_space table has vertical tab as whitespace; we have to
1177
ensure it gets set as not whitespace. */
1179
case OP_NOT_WHITESPACE:
1180
set_nottype_bits(start_bits, cbit_space, table_limit, cd);
1181
start_bits[1] |= 0x08;
1184
/* The cbit_space table has vertical tab as whitespace; we have to
1185
avoid setting it. */
1188
c = start_bits[1]; /* Save in case it was already set */
1189
set_type_bits(start_bits, cbit_space, table_limit, cd);
1190
start_bits[1] = (start_bits[1] & ~0x08) | c;
1193
case OP_NOT_WORDCHAR:
1194
set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1198
set_type_bits(start_bits, cbit_word, table_limit, cd);
1205
/* Character class where all the information is in a bit map: set the
1206
bits and either carry on or not, according to the repeat count. If it was
1207
a negative class, and we are operating with UTF-8 characters, any byte
1208
with a value >= 0xc4 is a potentially valid starter because it starts a
1209
character with a value > 255. */
1212
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
1215
start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
1216
memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
1219
#ifdef COMPILE_PCRE16
1220
SET_BIT(0xFF); /* For characters > 255 */
1228
map = (pcre_uint8 *)tcode;
1230
/* In UTF-8 mode, the bits in a bit map correspond to character
1231
values, not to byte values. However, the bit map we are constructing is
1232
for byte values. So we have to do a conversion for characters whose
1233
value is > 127. In fact, there are only two possible starting bytes for
1234
characters in the range 128 - 255. */
1236
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
1239
for (c = 0; c < 16; c++) start_bits[c] |= map[c];
1240
for (c = 128; c < 256; c++)
1242
if ((map[c/8] && (1 << (c&7))) != 0)
1244
int d = (c >> 6) | 0xc0; /* Set bit for this starter */
1245
start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
1246
c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
1253
/* In non-UTF-8 mode, the two bit maps are completely compatible. */
1254
for (c = 0; c < 32; c++) start_bits[c] |= map[c];
1257
/* Advance past the bit map, and act on what follows. For a zero
1258
minimum repeat, continue; otherwise stop processing. */
1260
tcode += 32 / sizeof(pcre_uchar);
1272
if (GET2(tcode, 1) == 0) tcode += 1 + 2 * IMM2_SIZE;
1273
else try_next = FALSE;
1281
break; /* End of bitmap class handling */
1283
} /* End of switch */
1284
} /* End of try_next loop */
1286
code += GET(code, 1); /* Advance to next branch */
1288
while (*code == OP_ALT);
1296
/*************************************************
1297
* Study a compiled expression *
1298
*************************************************/
1300
/* This function is handed a compiled expression that it must study to produce
1301
information that will speed up the matching. It returns a pcre[16]_extra block
1302
which then gets handed back to pcre_exec().
1305
re points to the compiled expression
1306
options contains option bits
1307
errorptr points to where to place error messages;
1308
set NULL unless error
1310
Returns: pointer to a pcre[16]_extra block, with study_data filled in and
1311
the appropriate flags set;
1312
NULL on error or if no optimization possible
1315
#ifdef COMPILE_PCRE8
1316
PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
1317
pcre_study(const pcre *external_re, int options, const char **errorptr)
1319
PCRE_EXP_DEFN pcre16_extra * PCRE_CALL_CONVENTION
1320
pcre16_study(const pcre16 *external_re, int options, const char **errorptr)
1324
BOOL bits_set = FALSE;
1325
pcre_uint8 start_bits[32];
1326
PUBL(extra) *extra = NULL;
1327
pcre_study_data *study;
1328
const pcre_uint8 *tables;
1330
compile_data compile_block;
1331
const REAL_PCRE *re = (const REAL_PCRE *)external_re;
1335
if (re == NULL || re->magic_number != MAGIC_NUMBER)
1337
*errorptr = "argument is not a compiled regular expression";
1341
if ((re->flags & PCRE_MODE) == 0)
1343
#ifdef COMPILE_PCRE8
1344
*errorptr = "argument is compiled in 16 bit mode";
1346
*errorptr = "argument is compiled in 8 bit mode";
1351
if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
1353
*errorptr = "unknown or incorrect option bit(s) set";
1357
code = (pcre_uchar *)re + re->name_table_offset +
1358
(re->name_count * re->name_entry_size);
1360
/* For an anchored pattern, or an unanchored pattern that has a first char, or
1361
a multiline pattern that matches only at "line starts", there is no point in
1362
seeking a list of starting bytes. */
1364
if ((re->options & PCRE_ANCHORED) == 0 &&
1365
(re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
1369
/* Set the character tables in the block that is passed around */
1371
tables = re->tables;
1373
#ifdef COMPILE_PCRE8
1375
(void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1379
(void)pcre16_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1383
compile_block.lcc = tables + lcc_offset;
1384
compile_block.fcc = tables + fcc_offset;
1385
compile_block.cbits = tables + cbits_offset;
1386
compile_block.ctypes = tables + ctypes_offset;
1388
/* See if we can find a fixed set of initial characters for the pattern. */
1390
memset(start_bits, 0, 32 * sizeof(pcre_uint8));
1391
rc = set_start_bits(code, start_bits, (re->options & PCRE_UTF8) != 0,
1393
bits_set = rc == SSB_DONE;
1394
if (rc == SSB_UNKNOWN)
1396
*errorptr = "internal error: opcode not recognized";
1401
/* Find the minimum length of subject string. */
1403
switch(min = find_minlength(code, code, re->options, 0))
1405
case -2: *errorptr = "internal error: missing capturing bracket"; return NULL;
1406
case -3: *errorptr = "internal error: opcode not recognized"; return NULL;
1410
/* If a set of starting bytes has been identified, or if the minimum length is
1411
greater than zero, or if JIT optimization has been requested, get a
1412
pcre[16]_extra block and a pcre_study_data block. The study data is put in the
1413
latter, which is pointed to by the former, which may also get additional data
1414
set later by the calling program. At the moment, the size of pcre_study_data
1415
is fixed. We nevertheless save it in a field for returning via the
1416
pcre_fullinfo() function so that if it becomes variable in the future,
1417
we don't have to change that code. */
1419
if (bits_set || min > 0
1421
|| (options & PCRE_STUDY_JIT_COMPILE) != 0
1425
extra = (PUBL(extra) *)(PUBL(malloc))
1426
(sizeof(PUBL(extra)) + sizeof(pcre_study_data));
1429
*errorptr = "failed to get memory";
1433
study = (pcre_study_data *)((char *)extra + sizeof(PUBL(extra)));
1434
extra->flags = PCRE_EXTRA_STUDY_DATA;
1435
extra->study_data = study;
1437
study->size = sizeof(pcre_study_data);
1440
/* Set the start bits always, to avoid unset memory errors if the
1441
study data is written to a file, but set the flag only if any of the bits
1442
are set, to save time looking when none are. */
1446
study->flags |= PCRE_STUDY_MAPPED;
1447
memcpy(study->start_bits, start_bits, sizeof(start_bits));
1449
else memset(study->start_bits, 0, 32 * sizeof(pcre_uint8));
1454
pcre_uint8 *ptr = start_bits;
1457
printf("Start bits:\n");
1458
for (i = 0; i < 32; i++)
1459
printf("%3d: %02x%s", i * 8, *ptr++, ((i + 1) & 0x7) != 0? " " : "\n");
1463
/* Always set the minlength value in the block, because the JIT compiler
1464
makes use of it. However, don't set the bit unless the length is greater than
1465
zero - the interpretive pcre_exec() and pcre_dfa_exec() needn't waste time
1466
checking the zero case. */
1470
study->flags |= PCRE_STUDY_MINLEN;
1471
study->minlength = min;
1473
else study->minlength = 0;
1475
/* If JIT support was compiled and requested, attempt the JIT compilation.
1476
If no starting bytes were found, and the minimum length is zero, and JIT
1477
compilation fails, abandon the extra block and return NULL. */
1480
extra->executable_jit = NULL;
1481
if ((options & PCRE_STUDY_JIT_COMPILE) != 0) PRIV(jit_compile)(re, extra);
1482
if (study->flags == 0 && (extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) == 0)
1484
#ifdef COMPILE_PCRE8
1485
pcre_free_study(extra);
1487
#ifdef COMPILE_PCRE16
1488
pcre16_free_study(extra);
1499
/*************************************************
1500
* Free the study data *
1501
*************************************************/
1503
/* This function frees the memory that was obtained by pcre_study().
1505
Argument: a pointer to the pcre[16]_extra block
1509
#ifdef COMPILE_PCRE8
1511
pcre_free_study(pcre_extra *extra)
1514
pcre16_free_study(pcre16_extra *extra)
1520
if ((extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) != 0 &&
1521
extra->executable_jit != NULL)
1522
PRIV(jit_free)(extra->executable_jit);
1527
/* End of pcre_study.c */