This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
PATCH: [perl #133997] Assert fail with script runs
[perl5.git] / regexec.c
CommitLineData
a0d0e21e
LW
1/* regexec.c
2 */
3
4/*
f65819ce
CO
5 * One Ring to rule them all, One Ring to find them
6 *
4ac71550
TC
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
a0d0e21e
LW
10 */
11
61296642
DM
12/* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
166f8a29 14 * a regular expression.
e4a054ea
DM
15 *
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM
20 */
21
a687059c
LW
22/* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
24 */
25
26/* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
29 */
30
e50aee73
AD
31/* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
34*/
35
b9d5759e 36#ifdef PERL_EXT_RE_BUILD
54df2634 37#include "re_top.h"
9041c2e3 38#endif
56953603 39
a687059c 40/*
e50aee73 41 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
42 *
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
45 *
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
49 *
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
52 * from defects in it.
53 *
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
56 *
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
59 *
60 **** Alterations to Henry's code are...
61 ****
4bb101f2 62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
a687059c 65 ****
9ef589d8
LW
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
a687059c
LW
68 *
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGEXEC_C
a687059c 75#include "perl.h"
0f5d15d6 76
54df2634
NC
77#ifdef PERL_IN_XSUB_RE
78# include "re_comp.h"
79#else
80# include "regcomp.h"
81#endif
a687059c 82
b992490d 83#include "invlist_inline.h"
1b0f46bf 84#include "unicode_constants.h"
81e983c1 85
bbac6b20
KW
86#define B_ON_NON_UTF8_LOCALE_IS_WRONG \
87 "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale"
88
a0bd1a30
KW
89static const char utf8_locale_required[] =
90 "Use of (?[ ]) for non-UTF-8 locale is wrong. Assuming a UTF-8 locale";
91
e1cf74e3
CB
92#ifdef DEBUGGING
93/* At least one required character in the target string is expressible only in
94 * UTF-8. */
e7a474c0 95static const char non_utf8_target_but_utf8_required[]
e1cf74e3
CB
96 = "Can't match, because target string needs to be in UTF-8\n";
97#endif
98
7b031478 99#define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
6ad9a8ab 100 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%s", non_utf8_target_but_utf8_required));\
7b031478 101 goto target; \
e1cf74e3
CB
102} STMT_END
103
c74f6de9
KW
104#define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
105
a687059c
LW
106#ifndef STATIC
107#define STATIC static
108#endif
109
c277df42
IZ
110/*
111 * Forwards.
112 */
113
f2ed9b32 114#define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
a0ed51b3 115
3dab1dad 116#define HOPc(pos,off) \
ba44c216 117 (char *)(reginfo->is_utf8_target \
220db18a 118 ? reghop3((U8*)pos, off, \
9d9163fb 119 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
3dab1dad 120 : (U8*)(pos + off))
557f47af 121
bb152a4b
DM
122/* like HOPMAYBE3 but backwards. lim must be +ve. Returns NULL on overshoot */
123#define HOPBACK3(pos, off, lim) \
124 (reginfo->is_utf8_target \
125 ? reghopmaybe3((U8*)pos, (SSize_t)0-off, (U8*)(lim)) \
126 : (pos - off >= lim) \
127 ? (U8*)pos - off \
3dab1dad 128 : NULL)
efb30f32 129
bb152a4b
DM
130#define HOPBACKc(pos, off) ((char*)HOPBACK3(pos, off, reginfo->strbeg))
131
ba44c216 132#define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
1aa99e6b 133#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
1aa99e6b 134
557f47af
DM
135/* lim must be +ve. Returns NULL on overshoot */
136#define HOPMAYBE3(pos,off,lim) \
137 (reginfo->is_utf8_target \
138 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
139 : ((U8*)pos + off <= lim) \
140 ? (U8*)pos + off \
141 : NULL)
142
8e9f2289
DM
143/* like HOP3, but limits the result to <= lim even for the non-utf8 case.
144 * off must be >=0; args should be vars rather than expressions */
145#define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
146 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
147 : (U8*)((pos + off) > lim ? lim : (pos + off)))
67853908 148#define HOP3clim(pos,off,lim) ((char*)HOP3lim(pos,off,lim))
8e9f2289 149
2974eaec
DM
150#define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
151 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
152 : (U8*)(pos + off))
153#define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
7016d6eb 154
c7304fe2 155#define PLACEHOLDER /* Something for the preprocessor to grab onto */
3dab1dad
YO
156/* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
157
5f80c4cf 158/* for use after a quantifier and before an EXACT-like node -- japhy */
c35dcbe2
YO
159/* it would be nice to rework regcomp.sym to generate this stuff. sigh
160 *
161 * NOTE that *nothing* that affects backtracking should be in here, specifically
162 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
163 * node that is in between two EXACT like nodes when ascertaining what the required
164 * "follow" character is. This should probably be moved to regex compile time
165 * although it may be done at run time beause of the REF possibility - more
166 * investigation required. -- demerphq
167*/
baa60164
KW
168#define JUMPABLE(rn) ( \
169 OP(rn) == OPEN || \
24be3102
YO
170 (OP(rn) == CLOSE && \
171 !EVAL_CLOSE_PAREN_IS(cur_eval,ARG(rn)) ) || \
baa60164
KW
172 OP(rn) == EVAL || \
173 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
174 OP(rn) == PLUS || OP(rn) == MINMOD || \
175 OP(rn) == KEEPS || \
176 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
e2d8ce26 177)
ee9b8eae 178#define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
e2d8ce26 179
ee9b8eae
YO
180#define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
181
a84d97b6
HS
182/*
183 Search for mandatory following text node; for lookahead, the text must
184 follow but for lookbehind (rn->flags != 0) we skip to the next step.
185*/
baa60164 186#define FIND_NEXT_IMPT(rn) STMT_START { \
3dab1dad
YO
187 while (JUMPABLE(rn)) { \
188 const OPCODE type = OP(rn); \
189 if (type == SUSPEND || PL_regkind[type] == CURLY) \
e2d8ce26 190 rn = NEXTOPER(NEXTOPER(rn)); \
3dab1dad 191 else if (type == PLUS) \
cca55fe3 192 rn = NEXTOPER(rn); \
3dab1dad 193 else if (type == IFMATCH) \
a84d97b6 194 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
e2d8ce26 195 else rn += NEXT_OFF(rn); \
3dab1dad 196 } \
5f80c4cf 197} STMT_END
74750237 198
006f26b2
DM
199#define SLAB_FIRST(s) (&(s)->states[0])
200#define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
201
a75351a1 202static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
bf2039a9 203static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
bf2039a9 204static regmatch_state * S_push_slab(pTHX);
51371543 205
87c0511b 206#define REGCP_PAREN_ELEMS 3
f067efbf 207#define REGCP_OTHER_ELEMS 3
e0fa7e2b 208#define REGCP_FRAME_ELEMS 1
620d5b66
NC
209/* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
210 * are needed for the regexp context stack bookkeeping. */
211
76e3520e 212STATIC CHECKPOINT
21553840 213S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen _pDEPTH)
a0d0e21e 214{
a3b680e6 215 const int retval = PL_savestack_ix;
92da3157
DM
216 const int paren_elems_to_push =
217 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
e0fa7e2b
NC
218 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
219 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
87c0511b 220 I32 p;
40a82448 221 GET_RE_DEBUG_FLAGS_DECL;
a0d0e21e 222
b93070ed
DM
223 PERL_ARGS_ASSERT_REGCPPUSH;
224
e49a9654 225 if (paren_elems_to_push < 0)
e8a85d26
JH
226 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
227 (int)paren_elems_to_push, (int)maxopenparen,
228 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
e49a9654 229
e0fa7e2b 230 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
147e3846 231 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %" UVuf
5df417d0 232 " out of range (%lu-%ld)",
92da3157
DM
233 total_elems,
234 (unsigned long)maxopenparen,
235 (long)parenfloor);
e0fa7e2b 236
620d5b66 237 SSGROW(total_elems + REGCP_FRAME_ELEMS);
7f69552c 238
495f47a5 239 DEBUG_BUFFERS_r(
92da3157 240 if ((int)maxopenparen > (int)parenfloor)
2b1a3689 241 Perl_re_exec_indentf( aTHX_
147e3846 242 "rex=0x%" UVxf " offs=0x%" UVxf ": saving capture indices:\n",
2b1a3689
YO
243 depth,
244 PTR2UV(rex),
495f47a5
DM
245 PTR2UV(rex->offs)
246 );
247 );
92da3157 248 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
b1ce53c5 249/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
99a90e59
FC
250 SSPUSHIV(rex->offs[p].end);
251 SSPUSHIV(rex->offs[p].start);
1ca2007e 252 SSPUSHINT(rex->offs[p].start_tmp);
2b1a3689 253 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
147e3846 254 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "\n",
2b1a3689
YO
255 depth,
256 (UV)p,
495f47a5
DM
257 (IV)rex->offs[p].start,
258 (IV)rex->offs[p].start_tmp,
259 (IV)rex->offs[p].end
40a82448 260 ));
a0d0e21e 261 }
b1ce53c5 262/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
92da3157 263 SSPUSHINT(maxopenparen);
b93070ed
DM
264 SSPUSHINT(rex->lastparen);
265 SSPUSHINT(rex->lastcloseparen);
e0fa7e2b 266 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
41123dfd 267
a0d0e21e
LW
268 return retval;
269}
270
c277df42 271/* These are needed since we do not localize EVAL nodes: */
ab3bbdeb
YO
272#define REGCP_SET(cp) \
273 DEBUG_STATE_r( \
147e3846
KW
274 Perl_re_exec_indentf( aTHX_ \
275 "Setting an EVAL scope, savestack=%" IVdf ",\n", \
cb41e5d6
YO
276 depth, (IV)PL_savestack_ix \
277 ) \
278 ); \
ab3bbdeb 279 cp = PL_savestack_ix
c3464db5 280
ab3bbdeb 281#define REGCP_UNWIND(cp) \
e4f74956 282 DEBUG_STATE_r( \
cb41e5d6 283 if (cp != PL_savestack_ix) \
147e3846
KW
284 Perl_re_exec_indentf( aTHX_ \
285 "Clearing an EVAL scope, savestack=%" \
286 IVdf "..%" IVdf "\n", \
cb41e5d6
YO
287 depth, (IV)(cp), (IV)PL_savestack_ix \
288 ) \
289 ); \
ab3bbdeb 290 regcpblow(cp)
c277df42 291
8e9ed7cc
DM
292/* set the start and end positions of capture ix */
293#define CLOSE_CAPTURE(ix, s, e) \
294 rex->offs[ix].start = s; \
295 rex->offs[ix].end = e; \
3be2a9fd
DM
296 if (ix > rex->lastparen) \
297 rex->lastparen = ix; \
298 rex->lastcloseparen = ix; \
69cd2617 299 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
befca383 300 "CLOSE: rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf "..%" IVdf " max: %" UVuf "\n", \
69cd2617
DM
301 depth, \
302 PTR2UV(rex), \
303 PTR2UV(rex->offs), \
8e9ed7cc
DM
304 (UV)ix, \
305 (IV)rex->offs[ix].start, \
3be2a9fd
DM
306 (IV)rex->offs[ix].end, \
307 (UV)rex->lastparen \
69cd2617
DM
308 ))
309
a8d1f4b4 310#define UNWIND_PAREN(lp, lcp) \
befca383
DM
311 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
312 "UNWIND_PAREN: rex=0x%" UVxf " offs=0x%" UVxf ": invalidate (%" UVuf "..%" UVuf "] set lcp: %" UVuf "\n", \
313 depth, \
314 PTR2UV(rex), \
315 PTR2UV(rex->offs), \
316 (UV)(lp), \
317 (UV)(rex->lastparen), \
318 (UV)(lcp) \
319 )); \
a8d1f4b4
DM
320 for (n = rex->lastparen; n > lp; n--) \
321 rex->offs[n].end = -1; \
322 rex->lastparen = n; \
323 rex->lastcloseparen = lcp;
324
325
f067efbf 326STATIC void
21553840 327S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p _pDEPTH)
a0d0e21e 328{
e0fa7e2b 329 UV i;
87c0511b 330 U32 paren;
a3621e74
YO
331 GET_RE_DEBUG_FLAGS_DECL;
332
7918f24d
NC
333 PERL_ARGS_ASSERT_REGCPPOP;
334
b1ce53c5 335 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
c6bf6a65 336 i = SSPOPUV;
e0fa7e2b
NC
337 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
338 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
b93070ed
DM
339 rex->lastcloseparen = SSPOPINT;
340 rex->lastparen = SSPOPINT;
92da3157 341 *maxopenparen_p = SSPOPINT;
b1ce53c5 342
620d5b66 343 i -= REGCP_OTHER_ELEMS;
b1ce53c5 344 /* Now restore the parentheses context. */
495f47a5
DM
345 DEBUG_BUFFERS_r(
346 if (i || rex->lastparen + 1 <= rex->nparens)
2b1a3689 347 Perl_re_exec_indentf( aTHX_
147e3846 348 "rex=0x%" UVxf " offs=0x%" UVxf ": restoring capture indices to:\n",
2b1a3689
YO
349 depth,
350 PTR2UV(rex),
495f47a5
DM
351 PTR2UV(rex->offs)
352 );
353 );
92da3157 354 paren = *maxopenparen_p;
620d5b66 355 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
ea3daa5d 356 SSize_t tmps;
1ca2007e 357 rex->offs[paren].start_tmp = SSPOPINT;
99a90e59
FC
358 rex->offs[paren].start = SSPOPIV;
359 tmps = SSPOPIV;
b93070ed
DM
360 if (paren <= rex->lastparen)
361 rex->offs[paren].end = tmps;
2b1a3689 362 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
147e3846 363 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "%s\n",
2b1a3689
YO
364 depth,
365 (UV)paren,
495f47a5
DM
366 (IV)rex->offs[paren].start,
367 (IV)rex->offs[paren].start_tmp,
368 (IV)rex->offs[paren].end,
369 (paren > rex->lastparen ? "(skipped)" : ""));
c277df42 370 );
87c0511b 371 paren--;
a0d0e21e 372 }
daf18116 373#if 1
dafc8851
JH
374 /* It would seem that the similar code in regtry()
375 * already takes care of this, and in fact it is in
376 * a better location to since this code can #if 0-ed out
377 * but the code in regtry() is needed or otherwise tests
378 * requiring null fields (pat.t#187 and split.t#{13,14}
daf18116
JH
379 * (as of patchlevel 7877) will fail. Then again,
380 * this code seems to be necessary or otherwise
225593e1
DM
381 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
382 * --jhi updated by dapm */
b93070ed 383 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
92da3157 384 if (i > *maxopenparen_p)
b93070ed
DM
385 rex->offs[i].start = -1;
386 rex->offs[i].end = -1;
2b1a3689 387 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
147e3846 388 " \\%" UVuf ": %s ..-1 undeffing\n",
2b1a3689
YO
389 depth,
390 (UV)i,
92da3157 391 (i > *maxopenparen_p) ? "-1" : " "
495f47a5 392 ));
a0d0e21e 393 }
dafc8851 394#endif
a0d0e21e
LW
395}
396
74088413
DM
397/* restore the parens and associated vars at savestack position ix,
398 * but without popping the stack */
399
400STATIC void
21553840 401S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p _pDEPTH)
74088413
DM
402{
403 I32 tmpix = PL_savestack_ix;
85882954
YO
404 PERL_ARGS_ASSERT_REGCP_RESTORE;
405
74088413 406 PL_savestack_ix = ix;
21553840 407 regcppop(rex, maxopenparen_p);
74088413
DM
408 PL_savestack_ix = tmpix;
409}
410
02db2b7b 411#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
a0d0e21e 412
9637d2a5
CB
413#ifndef PERL_IN_XSUB_RE
414
24e16d7b
KW
415bool
416Perl_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
31c7f561
KW
417{
418 /* Returns a boolean as to whether or not 'character' is a member of the
419 * Posix character class given by 'classnum' that should be equivalent to a
420 * value in the typedef '_char_class_number'.
421 *
422 * Ideally this could be replaced by a just an array of function pointers
423 * to the C library functions that implement the macros this calls.
424 * However, to compile, the precise function signatures are required, and
425 * these may vary from platform to to platform. To avoid having to figure
426 * out what those all are on each platform, I (khw) am using this method,
7aee35ff
KW
427 * which adds an extra layer of function call overhead (unless the C
428 * optimizer strips it away). But we don't particularly care about
429 * performance with locales anyway. */
31c7f561
KW
430
431 switch ((_char_class_number) classnum) {
15861f94 432 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
31c7f561 433 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
e8d596e0
KW
434 case _CC_ENUM_ASCII: return isASCII_LC(character);
435 case _CC_ENUM_BLANK: return isBLANK_LC(character);
cee69f79 436 case _CC_ENUM_CASED: return isLOWER_LC(character)
b0d691b2 437 || isUPPER_LC(character);
e8d596e0 438 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
31c7f561
KW
439 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
440 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
441 case _CC_ENUM_LOWER: return isLOWER_LC(character);
442 case _CC_ENUM_PRINT: return isPRINT_LC(character);
443 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
e8d596e0 444 case _CC_ENUM_SPACE: return isSPACE_LC(character);
31c7f561
KW
445 case _CC_ENUM_UPPER: return isUPPER_LC(character);
446 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
31c7f561 447 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
31c7f561
KW
448 default: /* VERTSPACE should never occur in locales */
449 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
450 }
451
e5964223 452 NOT_REACHED; /* NOTREACHED */
31c7f561
KW
453 return FALSE;
454}
455
9637d2a5
CB
456#endif
457
53362e85
KW
458PERL_STATIC_INLINE I32
459S_foldEQ_latin1_s2_folded(const char *s1, const char *s2, I32 len)
460{
461 /* Compare non-UTF-8 using Unicode (Latin1) semantics. s2 must already be
462 * folded. Works on all folds representable without UTF-8, except for
463 * LATIN_SMALL_LETTER_SHARP_S, and does not check for this. Nor does it
464 * check that the strings each have at least 'len' characters.
465 *
466 * There is almost an identical API function where s2 need not be folded:
467 * Perl_foldEQ_latin1() */
468
469 const U8 *a = (const U8 *)s1;
470 const U8 *b = (const U8 *)s2;
471
472 PERL_ARGS_ASSERT_FOLDEQ_LATIN1_S2_FOLDED;
473
474 assert(len >= 0);
475
476 while (len--) {
477 assert(! isUPPER_L1(*b));
478 if (toLOWER_L1(*a) != *b) {
479 return 0;
480 }
481 a++, b++;
482 }
483 return 1;
484}
485
3018b823 486STATIC bool
a78c2fa6 487S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character, const U8* e)
3018b823
KW
488{
489 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
490 * 'character' is a member of the Posix character class given by 'classnum'
491 * that should be equivalent to a value in the typedef
492 * '_char_class_number'.
493 *
494 * This just calls isFOO_lc on the code point for the character if it is in
2f306ab9 495 * the range 0-255. Outside that range, all characters use Unicode
3018b823 496 * rules, ignoring any locale. So use the Unicode function if this class
4c404f26 497 * requires an inversion list, and use the Unicode macro otherwise. */
3018b823 498
1565c085
DM
499 dVAR;
500
3018b823
KW
501 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
502
503 if (UTF8_IS_INVARIANT(*character)) {
504 return isFOO_lc(classnum, *character);
505 }
506 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
507 return isFOO_lc(classnum,
a62b247b 508 EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1)));
3018b823
KW
509 }
510
a78c2fa6 511 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, e);
613abc6d 512
3018b823 513 switch ((_char_class_number) classnum) {
779cf272 514 case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character);
3018b823
KW
515 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
516 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
517 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
4d3d8522
KW
518 default:
519 return _invlist_contains_cp(PL_XPosix_ptrs[classnum],
a78c2fa6 520 utf8_to_uvchr_buf(character, e, NULL));
3018b823
KW
521 }
522
e1ee3960 523 return FALSE; /* Things like CNTRL are always below 256 */
3018b823
KW
524}
525
2b1f9c71
KW
526STATIC U8 *
527S_find_span_end(U8 * s, const U8 * send, const U8 span_byte)
ab1efbdc
KW
528{
529 /* Returns the position of the first byte in the sequence between 's' and
530 * 'send-1' inclusive that isn't 'span_byte'; returns 'send' if none found.
531 * */
532
533 PERL_ARGS_ASSERT_FIND_SPAN_END;
534
535 assert(send >= s);
536
537 if ((STRLEN) (send - s) >= PERL_WORDSIZE
538 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
539 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
540 {
541 PERL_UINTMAX_T span_word;
542
543 /* Process per-byte until reach word boundary. XXX This loop could be
544 * eliminated if we knew that this platform had fast unaligned reads */
545 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
546 if (*s != span_byte) {
547 return s;
548 }
549 s++;
550 }
551
552 /* Create a word filled with the bytes we are spanning */
553 span_word = PERL_COUNT_MULTIPLIER * span_byte;
554
555 /* Process per-word as long as we have at least a full word left */
556 do {
557
558 /* Keep going if the whole word is composed of 'span_byte's */
559 if ((* (PERL_UINTMAX_T *) s) == span_word) {
560 s += PERL_WORDSIZE;
561 continue;
562 }
563
23a7ee81
KW
564 /* Here, at least one byte in the word isn't 'span_byte'. */
565
566#ifdef EBCDIC
567
568 break;
569
570#else
571
572 /* This xor leaves 1 bits only in those non-matching bytes */
ab1efbdc
KW
573 span_word ^= * (PERL_UINTMAX_T *) s;
574
575 /* Make sure the upper bit of each non-matching byte is set. This
576 * makes each such byte look like an ASCII platform variant byte */
577 span_word |= span_word << 1;
578 span_word |= span_word << 2;
579 span_word |= span_word << 4;
580
581 /* That reduces the problem to what this function solves */
582 return s + _variant_byte_number(span_word);
583
23a7ee81
KW
584#endif
585
ab1efbdc
KW
586 } while (s + PERL_WORDSIZE <= send);
587 }
588
589 /* Process the straggler bytes beyond the final word boundary */
590 while (s < send) {
591 if (*s != span_byte) {
592 return s;
593 }
594 s++;
595 }
596
597 return s;
598}
599
2b1f9c71
KW
600STATIC U8 *
601S_find_next_masked(U8 * s, const U8 * send, const U8 byte, const U8 mask)
2813d4ad
KW
602{
603 /* Returns the position of the first byte in the sequence between 's'
604 * and 'send-1' inclusive that when ANDed with 'mask' yields 'byte';
605 * returns 'send' if none found. It uses word-level operations instead of
606 * byte to speed up the process */
607
608 PERL_ARGS_ASSERT_FIND_NEXT_MASKED;
609
610 assert(send >= s);
611 assert((byte & mask) == byte);
612
23a7ee81
KW
613#ifndef EBCDIC
614
2813d4ad
KW
615 if ((STRLEN) (send - s) >= PERL_WORDSIZE
616 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
617 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
618 {
1d8aafa0 619 PERL_UINTMAX_T word, mask_word;
2813d4ad
KW
620
621 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
2b1f9c71 622 if (((*s) & mask) == byte) {
2813d4ad
KW
623 return s;
624 }
625 s++;
626 }
627
1d8aafa0
KW
628 word = PERL_COUNT_MULTIPLIER * byte;
629 mask_word = PERL_COUNT_MULTIPLIER * mask;
2813d4ad
KW
630
631 do {
632 PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;
633
1d8aafa0
KW
634 /* If 'masked' contains bytes with the bit pattern of 'byte' within
635 * it, xoring with 'word' will leave each of the 8 bits in such
636 * bytes be 0, and no byte containing any other bit pattern will be
637 * 0. */
638 masked ^= word;
2813d4ad
KW
639
640 /* This causes the most significant bit to be set to 1 for any
641 * bytes in the word that aren't completely 0 */
642 masked |= masked << 1;
643 masked |= masked << 2;
644 masked |= masked << 4;
645
646 /* The msbits are the same as what marks a byte as variant, so we
647 * can use this mask. If all msbits are 1, the word doesn't
648 * contain 'byte' */
649 if ((masked & PERL_VARIANTS_WORD_MASK) == PERL_VARIANTS_WORD_MASK) {
650 s += PERL_WORDSIZE;
651 continue;
652 }
653
654 /* Here, the msbit of bytes in the word that aren't 'byte' are 1,
655 * and any that are, are 0. Complement and re-AND to swap that */
656 masked = ~ masked;
657 masked &= PERL_VARIANTS_WORD_MASK;
658
659 /* This reduces the problem to that solved by this function */
660 s += _variant_byte_number(masked);
661 return s;
662
663 } while (s + PERL_WORDSIZE <= send);
664 }
665
23a7ee81
KW
666#endif
667
2813d4ad 668 while (s < send) {
2b1f9c71 669 if (((*s) & mask) == byte) {
2813d4ad
KW
670 return s;
671 }
672 s++;
673 }
674
675 return s;
676}
677
070e8b2e
KW
678STATIC U8 *
679S_find_span_end_mask(U8 * s, const U8 * send, const U8 span_byte, const U8 mask)
680{
681 /* Returns the position of the first byte in the sequence between 's' and
682 * 'send-1' inclusive that when ANDed with 'mask' isn't 'span_byte'.
683 * 'span_byte' should have been ANDed with 'mask' in the call of this
684 * function. Returns 'send' if none found. Works like find_span_end(),
685 * except for the AND */
686
687 PERL_ARGS_ASSERT_FIND_SPAN_END_MASK;
688
689 assert(send >= s);
690 assert((span_byte & mask) == span_byte);
691
692 if ((STRLEN) (send - s) >= PERL_WORDSIZE
693 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
694 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
695 {
696 PERL_UINTMAX_T span_word, mask_word;
697
698 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
2b1f9c71 699 if (((*s) & mask) != span_byte) {
070e8b2e
KW
700 return s;
701 }
702 s++;
703 }
704
705 span_word = PERL_COUNT_MULTIPLIER * span_byte;
706 mask_word = PERL_COUNT_MULTIPLIER * mask;
707
708 do {
709 PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;
710
711 if (masked == span_word) {
712 s += PERL_WORDSIZE;
713 continue;
714 }
715
23a7ee81
KW
716#ifdef EBCDIC
717
718 break;
719
720#else
721
070e8b2e
KW
722 masked ^= span_word;
723 masked |= masked << 1;
724 masked |= masked << 2;
725 masked |= masked << 4;
726 return s + _variant_byte_number(masked);
727
23a7ee81
KW
728#endif
729
070e8b2e
KW
730 } while (s + PERL_WORDSIZE <= send);
731 }
732
733 while (s < send) {
2b1f9c71 734 if (((*s) & mask) != span_byte) {
070e8b2e
KW
735 return s;
736 }
737 s++;
738 }
739
740 return s;
741}
742
a687059c 743/*
e50aee73 744 * pregexec and friends
a687059c
LW
745 */
746
76234dfb 747#ifndef PERL_IN_XSUB_RE
a687059c 748/*
c277df42 749 - pregexec - match a regexp against a string
a687059c 750 */
c277df42 751I32
5aaab254 752Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
ea3daa5d 753 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
8fd1a950
DM
754/* stringarg: the point in the string at which to begin matching */
755/* strend: pointer to null at end of string */
756/* strbeg: real beginning of string */
757/* minend: end of match must be >= minend bytes after stringarg. */
758/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
759 * itself is accessed via the pointers above */
760/* nosave: For optimizations. */
c277df42 761{
7918f24d
NC
762 PERL_ARGS_ASSERT_PREGEXEC;
763
c277df42 764 return
9041c2e3 765 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
c277df42
IZ
766 nosave ? 0 : REXEC_COPY_STR);
767}
76234dfb 768#endif
22e551b9 769
cad2e5aa 770
6eb5f6b9 771
1a4edc3c
DM
772/* re_intuit_start():
773 *
774 * Based on some optimiser hints, try to find the earliest position in the
775 * string where the regex could match.
776 *
777 * rx: the regex to match against
778 * sv: the SV being matched: only used for utf8 flag; the string
779 * itself is accessed via the pointers below. Note that on
780 * something like an overloaded SV, SvPOK(sv) may be false
781 * and the string pointers may point to something unrelated to
782 * the SV itself.
783 * strbeg: real beginning of string
784 * strpos: the point in the string at which to begin matching
785 * strend: pointer to the byte following the last char of the string
786 * flags currently unused; set to 0
787 * data: currently unused; set to NULL
788 *
789 * The basic idea of re_intuit_start() is to use some known information
790 * about the pattern, namely:
791 *
792 * a) the longest known anchored substring (i.e. one that's at a
793 * constant offset from the beginning of the pattern; but not
794 * necessarily at a fixed offset from the beginning of the
795 * string);
796 * b) the longest floating substring (i.e. one that's not at a constant
797 * offset from the beginning of the pattern);
798 * c) Whether the pattern is anchored to the string; either
799 * an absolute anchor: /^../, or anchored to \n: /^.../m,
800 * or anchored to pos(): /\G/;
801 * d) A start class: a real or synthetic character class which
802 * represents which characters are legal at the start of the pattern;
803 *
804 * to either quickly reject the match, or to find the earliest position
805 * within the string at which the pattern might match, thus avoiding
806 * running the full NFA engine at those earlier locations, only to
807 * eventually fail and retry further along.
808 *
809 * Returns NULL if the pattern can't match, or returns the address within
810 * the string which is the earliest place the match could occur.
811 *
812 * The longest of the anchored and floating substrings is called 'check'
813 * and is checked first. The other is called 'other' and is checked
814 * second. The 'other' substring may not be present. For example,
815 *
816 * /(abc|xyz)ABC\d{0,3}DEFG/
817 *
818 * will have
819 *
820 * check substr (float) = "DEFG", offset 6..9 chars
821 * other substr (anchored) = "ABC", offset 3..3 chars
822 * stclass = [ax]
823 *
824 * Be aware that during the course of this function, sometimes 'anchored'
825 * refers to a substring being anchored relative to the start of the
826 * pattern, and sometimes to the pattern itself being anchored relative to
827 * the string. For example:
828 *
829 * /\dabc/: "abc" is anchored to the pattern;
830 * /^\dabc/: "abc" is anchored to the pattern and the string;
831 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
832 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
833 * but the pattern is anchored to the string.
52a21eb3
DM
834 */
835
cad2e5aa 836char *
52a21eb3
DM
837Perl_re_intuit_start(pTHX_
838 REGEXP * const rx,
839 SV *sv,
840 const char * const strbeg,
841 char *strpos,
842 char *strend,
843 const U32 flags,
844 re_scream_pos_data *data)
cad2e5aa 845{
8d919b0a 846 struct regexp *const prog = ReANY(rx);
b2ad2123
DM
847 SSize_t start_shift = prog->check_offset_min;
848 /* Should be nonnegative! */
849 SSize_t end_shift = 0;
0fc004dd
DM
850 /* current lowest pos in string where the regex can start matching */
851 char *rx_origin = strpos;
eb578fdb 852 SV *check;
f2ed9b32 853 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
6480a6c4 854 U8 other_ix = 1 - prog->substrs->check_ix;
6ad5ffb3 855 bool ml_anch = 0;
8f4bf5fc 856 char *other_last = strpos;/* latest pos 'other' substr already checked to */
bd61b366 857 char *check_at = NULL; /* check substr found at this pos */
bbe252da 858 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
f8fc2ecf 859 RXi_GET_DECL(prog,progi);
02d5137b
DM
860 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
861 regmatch_info *const reginfo = &reginfo_buf;
a3621e74
YO
862 GET_RE_DEBUG_FLAGS_DECL;
863
7918f24d 864 PERL_ARGS_ASSERT_RE_INTUIT_START;
c33e64f0
FC
865 PERL_UNUSED_ARG(flags);
866 PERL_UNUSED_ARG(data);
7918f24d 867
6ad9a8ab 868 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0
DM
869 "Intuit: trying to determine minimum start position...\n"));
870
fb9bbddb 871 /* for now, assume that all substr offsets are positive. If at some point
f67a5002 872 * in the future someone wants to do clever things with lookbehind and
fb9bbddb
DM
873 * -ve offsets, they'll need to fix up any code in this function
874 * which uses these offsets. See the thread beginning
875 * <20140113145929.GF27210@iabyn.com>
876 */
877 assert(prog->substrs->data[0].min_offset >= 0);
878 assert(prog->substrs->data[0].max_offset >= 0);
879 assert(prog->substrs->data[1].min_offset >= 0);
880 assert(prog->substrs->data[1].max_offset >= 0);
881 assert(prog->substrs->data[2].min_offset >= 0);
882 assert(prog->substrs->data[2].max_offset >= 0);
883
f7022b5a 884 /* for now, assume that if both present, that the floating substring
83f2232d 885 * doesn't start before the anchored substring.
f7022b5a
DM
886 * If you break this assumption (e.g. doing better optimisations
887 * with lookahead/behind), then you'll need to audit the code in this
888 * function carefully first
889 */
890 assert(
891 ! ( (prog->anchored_utf8 || prog->anchored_substr)
892 && (prog->float_utf8 || prog->float_substr))
893 || (prog->float_min_offset >= prog->anchored_offset));
894
1a4edc3c
DM
895 /* byte rather than char calculation for efficiency. It fails
896 * to quickly reject some cases that can't match, but will reject
897 * them later after doing full char arithmetic */
c344f387 898 if (prog->minlen > strend - strpos) {
6ad9a8ab 899 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 900 " String too short...\n"));
cad2e5aa 901 goto fail;
2c2d71f5 902 }
d8da0584 903
196a02af 904 RXp_MATCH_UTF8_set(prog, utf8_target);
6c3fea77 905 reginfo->is_utf8_target = cBOOL(utf8_target);
bf2039a9 906 reginfo->info_aux = NULL;
9d9163fb 907 reginfo->strbeg = strbeg;
220db18a 908 reginfo->strend = strend;
aed7b151 909 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
02d5137b 910 reginfo->intuit = 1;
1cb48e53
DM
911 /* not actually used within intuit, but zero for safety anyway */
912 reginfo->poscache_maxiter = 0;
02d5137b 913
f2ed9b32 914 if (utf8_target) {
2814f4b3
HS
915 if ((!prog->anchored_utf8 && prog->anchored_substr)
916 || (!prog->float_utf8 && prog->float_substr))
33b8afdf
JH
917 to_utf8_substr(prog);
918 check = prog->check_utf8;
919 } else {
7e0d5ad7
KW
920 if (!prog->check_substr && prog->check_utf8) {
921 if (! to_byte_substr(prog)) {
6b54ddc5 922 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
7e0d5ad7
KW
923 }
924 }
33b8afdf
JH
925 check = prog->check_substr;
926 }
274cd312 927
1dc475d0
DM
928 /* dump the various substring data */
929 DEBUG_OPTIMISE_MORE_r({
930 int i;
931 for (i=0; i<=2; i++) {
932 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
933 : prog->substrs->data[i].substr);
934 if (!sv)
935 continue;
936
6ad9a8ab 937 Perl_re_printf( aTHX_
147e3846
KW
938 " substrs[%d]: min=%" IVdf " max=%" IVdf " end shift=%" IVdf
939 " useful=%" IVdf " utf8=%d [%s]\n",
1dc475d0
DM
940 i,
941 (IV)prog->substrs->data[i].min_offset,
942 (IV)prog->substrs->data[i].max_offset,
943 (IV)prog->substrs->data[i].end_shift,
944 BmUSEFUL(sv),
945 utf8_target ? 1 : 0,
946 SvPEEK(sv));
947 }
948 });
949
8e1490ee 950 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
9fc7410e
DM
951
952 /* ml_anch: check after \n?
953 *
0fa70a06 954 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
9fc7410e
DM
955 * with /.*.../, these flags will have been added by the
956 * compiler:
957 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
958 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
959 */
7d2d37f5
DM
960 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
961 && !(prog->intflags & PREGf_IMPLICIT);
cad2e5aa 962
343c8a29 963 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
c889ccc8
DM
964 /* we are only allowed to match at BOS or \G */
965
57fcbfa7 966 /* trivially reject if there's a BOS anchor and we're not at BOS.
7bb3b9eb
DM
967 *
968 * Note that we don't try to do a similar quick reject for
969 * \G, since generally the caller will have calculated strpos
970 * based on pos() and gofs, so the string is already correctly
971 * anchored by definition; and handling the exceptions would
972 * be too fiddly (e.g. REXEC_IGNOREPOS).
57fcbfa7 973 */
7bb3b9eb 974 if ( strpos != strbeg
d3d47aac 975 && (prog->intflags & PREGf_ANCH_SBOL))
c889ccc8 976 {
6ad9a8ab 977 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 978 " Not at start...\n"));
c889ccc8
DM
979 goto fail;
980 }
981
a5d12a4b
DM
982 /* in the presence of an anchor, the anchored (relative to the
983 * start of the regex) substr must also be anchored relative
66b7ec5c
DM
984 * to strpos. So quickly reject if substr isn't found there.
985 * This works for \G too, because the caller will already have
986 * subtracted gofs from pos, and gofs is the offset from the
987 * \G to the start of the regex. For example, in /.abc\Gdef/,
988 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
989 * caller will have set strpos=pos()-4; we look for the substr
990 * at position pos()-4+1, which lines up with the "a" */
a5d12a4b 991
33c28ab2 992 if (prog->check_offset_min == prog->check_offset_max) {
c889ccc8 993 /* Substring at constant offset from beg-of-str... */
b2ad2123 994 SSize_t slen = SvCUR(check);
343c8a29 995 char *s = HOP3c(strpos, prog->check_offset_min, strend);
1de06328 996
6ad9a8ab 997 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 998 " Looking for check substr at fixed offset %" IVdf "...\n",
1dc475d0
DM
999 (IV)prog->check_offset_min));
1000
7742aa66
DM
1001 if (SvTAIL(check)) {
1002 /* In this case, the regex is anchored at the end too.
1003 * Unless it's a multiline match, the lengths must match
b2ad2123 1004 * exactly, give or take a \n. NB: slen >= 1 since
7742aa66
DM
1005 * the last char of check is \n */
1006 if (!multiline
b2ad2123
DM
1007 && ( strend - s > slen
1008 || strend - s < slen - 1
1009 || (strend - s == slen && strend[-1] != '\n')))
c889ccc8 1010 {
6ad9a8ab 1011 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 1012 " String too long...\n"));
c889ccc8
DM
1013 goto fail_finish;
1014 }
b2ad2123
DM
1015 /* Now should match s[0..slen-2] */
1016 slen--;
c889ccc8 1017 }
b2ad2123 1018 if (slen && (strend - s < slen
26fb2318 1019 || *SvPVX_const(check) != *s
b2ad2123 1020 || (slen > 1 && (memNE(SvPVX_const(check), s, slen)))))
d307bf57 1021 {
6ad9a8ab 1022 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 1023 " String not equal...\n"));
d307bf57
DM
1024 goto fail_finish;
1025 }
c889ccc8
DM
1026
1027 check_at = s;
1028 goto success_at_start;
cad2e5aa 1029 }
cad2e5aa 1030 }
cad2e5aa 1031 }
0fc004dd 1032
b2ad2123 1033 end_shift = prog->check_end_shift;
cad2e5aa 1034
19188028 1035#ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
b2ad2123 1036 if (end_shift < 0)
147e3846 1037 Perl_croak(aTHX_ "panic: end_shift: %" IVdf " pattern:\n%s\n ",
b2ad2123 1038 (IV)end_shift, RX_PRECOMP(rx));
2c2d71f5
JH
1039#endif
1040
2c2d71f5 1041 restart:
1de06328 1042
66b7ec5c
DM
1043 /* This is the (re)entry point of the main loop in this function.
1044 * The goal of this loop is to:
1045 * 1) find the "check" substring in the region rx_origin..strend
b2ad2123 1046 * (adjusted by start_shift / end_shift). If not found, reject
66b7ec5c
DM
1047 * immediately.
1048 * 2) If it exists, look for the "other" substr too if defined; for
1049 * example, if the check substr maps to the anchored substr, then
1050 * check the floating substr, and vice-versa. If not found, go
1051 * back to (1) with rx_origin suitably incremented.
1052 * 3) If we find an rx_origin position that doesn't contradict
1053 * either of the substrings, then check the possible additional
1054 * constraints on rx_origin of /^.../m or a known start class.
1055 * If these fail, then depending on which constraints fail, jump
1056 * back to here, or to various other re-entry points further along
1057 * that skip some of the first steps.
1058 * 4) If we pass all those tests, update the BmUSEFUL() count on the
1059 * substring. If the start position was determined to be at the
1060 * beginning of the string - so, not rejected, but not optimised,
1061 * since we have to run regmatch from position 0 - decrement the
1062 * BmUSEFUL() count. Otherwise increment it.
1063 */
1064
1a4edc3c
DM
1065
1066 /* first, look for the 'check' substring */
1067
1de06328 1068 {
c33e64f0
FC
1069 U8* start_point;
1070 U8* end_point;
c889ccc8 1071
c889ccc8 1072 DEBUG_OPTIMISE_MORE_r({
6ad9a8ab 1073 Perl_re_printf( aTHX_
147e3846
KW
1074 " At restart: rx_origin=%" IVdf " Check offset min: %" IVdf
1075 " Start shift: %" IVdf " End shift %" IVdf
1076 " Real end Shift: %" IVdf "\n",
675e93ee 1077 (IV)(rx_origin - strbeg),
c889ccc8 1078 (IV)prog->check_offset_min,
b2ad2123
DM
1079 (IV)start_shift,
1080 (IV)end_shift,
c889ccc8
DM
1081 (IV)prog->check_end_shift);
1082 });
1de06328 1083
b2ad2123 1084 end_point = HOPBACK3(strend, end_shift, rx_origin);
bb152a4b
DM
1085 if (!end_point)
1086 goto fail_finish;
b2ad2123 1087 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
33c28ab2
DM
1088 if (!start_point)
1089 goto fail_finish;
c889ccc8 1090
557f47af 1091
e0362b86 1092 /* If the regex is absolutely anchored to either the start of the
d3d47aac 1093 * string (SBOL) or to pos() (ANCH_GPOS), then
e0362b86
DM
1094 * check_offset_max represents an upper bound on the string where
1095 * the substr could start. For the ANCH_GPOS case, we assume that
1096 * the caller of intuit will have already set strpos to
1097 * pos()-gofs, so in this case strpos + offset_max will still be
1098 * an upper bound on the substr.
1099 */
c19c836a
DM
1100 if (!ml_anch
1101 && prog->intflags & PREGf_ANCH
e0362b86 1102 && prog->check_offset_max != SSize_t_MAX)
c19c836a 1103 {
b2ad2123 1104 SSize_t check_len = SvCUR(check) - !!SvTAIL(check);
e0362b86
DM
1105 const char * const anchor =
1106 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
b2ad2123 1107 SSize_t targ_len = (char*)end_point - anchor;
2ce94a86 1108
b2ad2123 1109 if (check_len > targ_len) {
2ce94a86 1110 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
12453e29 1111 "Target string too short to match required substring...\n"));
2ce94a86
DM
1112 goto fail_finish;
1113 }
e0362b86
DM
1114
1115 /* do a bytes rather than chars comparison. It's conservative;
1116 * so it skips doing the HOP if the result can't possibly end
1117 * up earlier than the old value of end_point.
1118 */
b2ad2123
DM
1119 assert(anchor + check_len <= (char *)end_point);
1120 if (prog->check_offset_max + check_len < targ_len) {
e0362b86
DM
1121 end_point = HOP3lim((U8*)anchor,
1122 prog->check_offset_max,
b2ad2123 1123 end_point - check_len
2ce94a86 1124 )
b2ad2123 1125 + check_len;
12453e29
YO
1126 if (end_point < start_point)
1127 goto fail_finish;
e0362b86 1128 }
d6ef1678
DM
1129 }
1130
ae5d4331 1131 check_at = fbm_instr( start_point, end_point,
7fba1cd6 1132 check, multiline ? FBMrf_MULTILINE : 0);
c889ccc8 1133
6ad9a8ab 1134 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1135 " doing 'check' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
675e93ee
DM
1136 (IV)((char*)start_point - strbeg),
1137 (IV)((char*)end_point - strbeg),
1138 (IV)(check_at ? check_at - strbeg : -1)
1139 ));
1140
8fd34720
DM
1141 /* Update the count-of-usability, remove useless subpatterns,
1142 unshift s. */
1143
1144 DEBUG_EXECUTE_r({
1145 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1146 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
6ad9a8ab 1147 Perl_re_printf( aTHX_ " %s %s substr %s%s%s",
8fd34720
DM
1148 (check_at ? "Found" : "Did not find"),
1149 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
1150 ? "anchored" : "floating"),
1151 quoted,
1152 RE_SV_TAIL(check),
1153 (check_at ? " at offset " : "...\n") );
1154 });
2c2d71f5 1155
8fd34720
DM
1156 if (!check_at)
1157 goto fail_finish;
8fd34720
DM
1158 /* set rx_origin to the minimum position where the regex could start
1159 * matching, given the constraint of the just-matched check substring.
1160 * But don't set it lower than previously.
1161 */
fdc003fd 1162
8fd34720
DM
1163 if (check_at - rx_origin > prog->check_offset_max)
1164 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
675e93ee 1165 /* Finish the diagnostic message */
6ad9a8ab 1166 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1167 "%ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
1168 (long)(check_at - strbeg),
1169 (IV)(rx_origin - strbeg)
1170 ));
8fd34720 1171 }
fdc003fd
DM
1172
1173
1a4edc3c 1174 /* now look for the 'other' substring if defined */
2c2d71f5 1175
fd8def15
HS
1176 if (prog->substrs->data[other_ix].utf8_substr
1177 || prog->substrs->data[other_ix].substr)
1de06328 1178 {
30944b6d 1179 /* Take into account the "other" substring. */
6c3343a6
DM
1180 char *last, *last1;
1181 char *s;
1182 SV* must;
1183 struct reg_substr_datum *other;
1184
1185 do_other_substr:
1186 other = &prog->substrs->data[other_ix];
fd8def15
HS
1187 if (!utf8_target && !other->substr) {
1188 if (!to_byte_substr(prog)) {
1189 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
1190 }
1191 }
6c3343a6
DM
1192
1193 /* if "other" is anchored:
1194 * we've previously found a floating substr starting at check_at.
1195 * This means that the regex origin must lie somewhere
1196 * between min (rx_origin): HOP3(check_at, -check_offset_max)
1197 * and max: HOP3(check_at, -check_offset_min)
1198 * (except that min will be >= strpos)
1199 * So the fixed substr must lie somewhere between
1200 * HOP3(min, anchored_offset)
1201 * HOP3(max, anchored_offset) + SvCUR(substr)
1202 */
1203
1204 /* if "other" is floating
1205 * Calculate last1, the absolute latest point where the
1206 * floating substr could start in the string, ignoring any
1207 * constraints from the earlier fixed match. It is calculated
1208 * as follows:
1209 *
1210 * strend - prog->minlen (in chars) is the absolute latest
1211 * position within the string where the origin of the regex
1212 * could appear. The latest start point for the floating
1213 * substr is float_min_offset(*) on from the start of the
1214 * regex. last1 simply combines thee two offsets.
1215 *
1216 * (*) You might think the latest start point should be
1217 * float_max_offset from the regex origin, and technically
1218 * you'd be correct. However, consider
1219 * /a\d{2,4}bcd\w/
1220 * Here, float min, max are 3,5 and minlen is 7.
1221 * This can match either
1222 * /a\d\dbcd\w/
1223 * /a\d\d\dbcd\w/
1224 * /a\d\d\d\dbcd\w/
1225 * In the first case, the regex matches minlen chars; in the
1226 * second, minlen+1, in the third, minlen+2.
1227 * In the first case, the floating offset is 3 (which equals
1228 * float_min), in the second, 4, and in the third, 5 (which
1229 * equals float_max). In all cases, the floating string bcd
1230 * can never start more than 4 chars from the end of the
1231 * string, which equals minlen - float_min. As the substring
1232 * starts to match more than float_min from the start of the
1233 * regex, it makes the regex match more than minlen chars,
1234 * and the two cancel each other out. So we can always use
1235 * float_min - minlen, rather than float_max - minlen for the
1236 * latest position in the string.
1237 *
1238 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1239 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1240 */
1241
e7a14a9c 1242 assert(prog->minlen >= other->min_offset);
6c3343a6
DM
1243 last1 = HOP3c(strend,
1244 other->min_offset - prog->minlen, strbeg);
1245
4d006249 1246 if (other_ix) {/* i.e. if (other-is-float) */
6c3343a6
DM
1247 /* last is the latest point where the floating substr could
1248 * start, *given* any constraints from the earlier fixed
1249 * match. This constraint is that the floating string starts
1250 * <= float_max_offset chars from the regex origin (rx_origin).
1251 * If this value is less than last1, use it instead.
eb3831ce 1252 */
6c3343a6
DM
1253 assert(rx_origin <= last1);
1254 last =
1255 /* this condition handles the offset==infinity case, and
1256 * is a short-cut otherwise. Although it's comparing a
1257 * byte offset to a char length, it does so in a safe way,
1258 * since 1 char always occupies 1 or more bytes,
1259 * so if a string range is (last1 - rx_origin) bytes,
1260 * it will be less than or equal to (last1 - rx_origin)
1261 * chars; meaning it errs towards doing the accurate HOP3
1262 * rather than just using last1 as a short-cut */
1263 (last1 - rx_origin) < other->max_offset
1264 ? last1
1265 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1266 }
1267 else {
b2ad2123
DM
1268 assert(strpos + start_shift <= check_at);
1269 last = HOP4c(check_at, other->min_offset - start_shift,
6c3343a6
DM
1270 strbeg, strend);
1271 }
ead917d0 1272
6c3343a6
DM
1273 s = HOP3c(rx_origin, other->min_offset, strend);
1274 if (s < other_last) /* These positions already checked */
1275 s = other_last;
1276
1277 must = utf8_target ? other->utf8_substr : other->substr;
1278 assert(SvPOK(must));
675e93ee
DM
1279 {
1280 char *from = s;
1281 char *to = last + SvCUR(must) - (SvTAIL(must)!=0);
1282
71a9d105
DM
1283 if (to > strend)
1284 to = strend;
88203927
DM
1285 if (from > to) {
1286 s = NULL;
6ad9a8ab 1287 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1288 " skipping 'other' fbm scan: %" IVdf " > %" IVdf "\n",
88203927
DM
1289 (IV)(from - strbeg),
1290 (IV)(to - strbeg)
1291 ));
1292 }
1293 else {
1294 s = fbm_instr(
1295 (unsigned char*)from,
1296 (unsigned char*)to,
1297 must,
1298 multiline ? FBMrf_MULTILINE : 0
1299 );
6ad9a8ab 1300 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1301 " doing 'other' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
88203927
DM
1302 (IV)(from - strbeg),
1303 (IV)(to - strbeg),
1304 (IV)(s ? s - strbeg : -1)
1305 ));
1306 }
675e93ee
DM
1307 }
1308
6c3343a6
DM
1309 DEBUG_EXECUTE_r({
1310 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1311 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
6ad9a8ab 1312 Perl_re_printf( aTHX_ " %s %s substr %s%s",
6c3343a6
DM
1313 s ? "Found" : "Contradicts",
1314 other_ix ? "floating" : "anchored",
1315 quoted, RE_SV_TAIL(must));
1316 });
ead917d0 1317
ead917d0 1318
6c3343a6
DM
1319 if (!s) {
1320 /* last1 is latest possible substr location. If we didn't
1321 * find it before there, we never will */
1322 if (last >= last1) {
6ad9a8ab 1323 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
675e93ee 1324 "; giving up...\n"));
6c3343a6 1325 goto fail_finish;
ead917d0
DM
1326 }
1327
6c3343a6
DM
1328 /* try to find the check substr again at a later
1329 * position. Maybe next time we'll find the "other" substr
1330 * in range too */
6c3343a6
DM
1331 other_last = HOP3c(last, 1, strend) /* highest failure */;
1332 rx_origin =
4d006249 1333 other_ix /* i.e. if other-is-float */
6c3343a6
DM
1334 ? HOP3c(rx_origin, 1, strend)
1335 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
6ad9a8ab 1336 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1337 "; about to retry %s at offset %ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
1338 (other_ix ? "floating" : "anchored"),
1339 (long)(HOP3c(check_at, 1, strend) - strbeg),
1340 (IV)(rx_origin - strbeg)
1341 ));
6c3343a6
DM
1342 goto restart;
1343 }
1344 else {
4d006249 1345 if (other_ix) { /* if (other-is-float) */
6c3343a6
DM
1346 /* other_last is set to s, not s+1, since its possible for
1347 * a floating substr to fail first time, then succeed
1348 * second time at the same floating position; e.g.:
1349 * "-AB--AABZ" =~ /\wAB\d*Z/
1350 * The first time round, anchored and float match at
1351 * "-(AB)--AAB(Z)" then fail on the initial \w character
1352 * class. Second time round, they match at "-AB--A(AB)(Z)".
1353 */
1354 other_last = s;
ead917d0
DM
1355 }
1356 else {
6c3343a6
DM
1357 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1358 other_last = HOP3c(s, 1, strend);
ead917d0 1359 }
6ad9a8ab 1360 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1361 " at offset %ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
1362 (long)(s - strbeg),
1363 (IV)(rx_origin - strbeg)
1364 ));
1365
6c3343a6 1366 }
cad2e5aa 1367 }
acba93e8
DM
1368 else {
1369 DEBUG_OPTIMISE_MORE_r(
6ad9a8ab 1370 Perl_re_printf( aTHX_
147e3846
KW
1371 " Check-only match: offset min:%" IVdf " max:%" IVdf
1372 " check_at:%" IVdf " rx_origin:%" IVdf " rx_origin-check_at:%" IVdf
1373 " strend:%" IVdf "\n",
acba93e8
DM
1374 (IV)prog->check_offset_min,
1375 (IV)prog->check_offset_max,
675e93ee
DM
1376 (IV)(check_at-strbeg),
1377 (IV)(rx_origin-strbeg),
1c1c599d 1378 (IV)(rx_origin-check_at),
675e93ee 1379 (IV)(strend-strbeg)
acba93e8
DM
1380 )
1381 );
1382 }
2c2d71f5 1383
acba93e8 1384 postprocess_substr_matches:
0991020e 1385
1a4edc3c 1386 /* handle the extra constraint of /^.../m if present */
e3c6feb0 1387
7d2d37f5 1388 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
4620cb61
DM
1389 char *s;
1390
6ad9a8ab 1391 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
a62659bd 1392 " looking for /^/m anchor"));
d0880ea7
DM
1393
1394 /* we have failed the constraint of a \n before rx_origin.
2e759faa
DM
1395 * Find the next \n, if any, even if it's beyond the current
1396 * anchored and/or floating substrings. Whether we should be
1397 * scanning ahead for the next \n or the next substr is debatable.
1398 * On the one hand you'd expect rare substrings to appear less
1399 * often than \n's. On the other hand, searching for \n means
675e93ee 1400 * we're effectively flipping between check_substr and "\n" on each
2e759faa
DM
1401 * iteration as the current "rarest" string candidate, which
1402 * means for example that we'll quickly reject the whole string if
1403 * hasn't got a \n, rather than trying every substr position
1404 * first
1405 */
d0880ea7 1406
4620cb61
DM
1407 s = HOP3c(strend, - prog->minlen, strpos);
1408 if (s <= rx_origin ||
1409 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1410 {
6ad9a8ab 1411 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
d0880ea7
DM
1412 " Did not find /%s^%s/m...\n",
1413 PL_colors[0], PL_colors[1]));
a62659bd
DM
1414 goto fail_finish;
1415 }
d0880ea7 1416
4ada1233
DM
1417 /* earliest possible origin is 1 char after the \n.
1418 * (since *rx_origin == '\n', it's safe to ++ here rather than
1419 * HOP(rx_origin, 1)) */
1420 rx_origin++;
d0880ea7 1421
f4f115de 1422 if (prog->substrs->check_ix == 0 /* check is anchored */
4ada1233 1423 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
f4f115de 1424 {
d0880ea7
DM
1425 /* Position contradicts check-string; either because
1426 * check was anchored (and thus has no wiggle room),
4ada1233 1427 * or check was float and rx_origin is above the float range */
6ad9a8ab 1428 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
675e93ee
DM
1429 " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
1430 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
d0880ea7
DM
1431 goto restart;
1432 }
1433
1434 /* if we get here, the check substr must have been float,
2e759faa 1435 * is in range, and we may or may not have had an anchored
d0880ea7
DM
1436 * "other" substr which still contradicts */
1437 assert(prog->substrs->check_ix); /* check is float */
1438
1439 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1440 /* whoops, the anchored "other" substr exists, so we still
1441 * contradict. On the other hand, the float "check" substr
1442 * didn't contradict, so just retry the anchored "other"
1443 * substr */
6ad9a8ab 1444 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1445 " Found /%s^%s/m, rescanning for anchored from offset %" IVdf " (rx_origin now %" IVdf ")...\n",
d0880ea7 1446 PL_colors[0], PL_colors[1],
73e8ff00
DM
1447 (IV)(rx_origin - strbeg + prog->anchored_offset),
1448 (IV)(rx_origin - strbeg)
675e93ee 1449 ));
d0880ea7
DM
1450 goto do_other_substr;
1451 }
1452
1453 /* success: we don't contradict the found floating substring
1454 * (and there's no anchored substr). */
6ad9a8ab 1455 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
675e93ee
DM
1456 " Found /%s^%s/m with rx_origin %ld...\n",
1457 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
e3c6feb0
DM
1458 }
1459 else {
6ad9a8ab 1460 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
fe4f3442 1461 " (multiline anchor test skipped)\n"));
e3c6feb0
DM
1462 }
1463
ffad1e6a 1464 success_at_start:
e3c6feb0 1465
cad2e5aa 1466
dd170ff5
DM
1467 /* if we have a starting character class, then test that extra constraint.
1468 * (trie stclasses are too expensive to use here, we are better off to
1469 * leave it to regmatch itself) */
1470
f8fc2ecf 1471 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
f8fc2ecf 1472 const U8* const str = (U8*)STRING(progi->regstclass);
0991020e 1473
b2ad2123
DM
1474 /* XXX this value could be pre-computed */
1475 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
2c75e362
DM
1476 ? (reginfo->is_utf8_pat
1477 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1478 : STR_LEN(progi->regstclass))
66e933ab 1479 : 1);
1de06328 1480 char * endpos;
fa3bb21d 1481 char *s;
000dfd2d
DM
1482 /* latest pos that a matching float substr constrains rx start to */
1483 char *rx_max_float = NULL;
1484
c75a3985
DM
1485 /* if the current rx_origin is anchored, either by satisfying an
1486 * anchored substring constraint, or a /^.../m constraint, then we
1487 * can reject the current origin if the start class isn't found
1488 * at the current position. If we have a float-only match, then
1489 * rx_origin is constrained to a range; so look for the start class
1490 * in that range. if neither, then look for the start class in the
1491 * whole rest of the string */
1492
dd170ff5
DM
1493 /* XXX DAPM it's not clear what the minlen test is for, and why
1494 * it's not used in the floating case. Nothing in the test suite
1495 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1496 * Here are some old comments, which may or may not be correct:
1497 *
1498 * minlen == 0 is possible if regstclass is \b or \B,
1499 * and the fixed substr is ''$.
1500 * Since minlen is already taken into account, rx_origin+1 is
1501 * before strend; accidentally, minlen >= 1 guaranties no false
1502 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1503 * 0) below assumes that regstclass does not come from lookahead...
1504 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1505 * This leaves EXACTF-ish only, which are dealt with in
1506 * find_byclass().
1507 */
1508
7d2d37f5 1509 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
b2ad2123 1510 endpos = HOP3clim(rx_origin, (prog->minlen ? cl_l : 0), strend);
000dfd2d 1511 else if (prog->float_substr || prog->float_utf8) {
b2ad2123
DM
1512 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1513 endpos = HOP3clim(rx_max_float, cl_l, strend);
000dfd2d 1514 }
1de06328
YO
1515 else
1516 endpos= strend;
1517
6ad9a8ab 1518 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
b2ad2123 1519 " looking for class: start_shift: %" IVdf " check_at: %" IVdf
147e3846 1520 " rx_origin: %" IVdf " endpos: %" IVdf "\n",
b2ad2123 1521 (IV)start_shift, (IV)(check_at - strbeg),
c43b5520 1522 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
d8080198 1523
c43b5520 1524 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
f9176b44 1525 reginfo);
be778b1a 1526 if (!s) {
6eb5f6b9 1527 if (endpos == strend) {
6ad9a8ab 1528 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1dc475d0 1529 " Could not match STCLASS...\n") );
6eb5f6b9
JH
1530 goto fail;
1531 }
6ad9a8ab 1532 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1dc475d0 1533 " This position contradicts STCLASS...\n") );
e0eb31e7
DM
1534 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1535 && !(prog->intflags & PREGf_IMPLICIT))
653099ff 1536 goto fail;
9fed8d02 1537
6eb5f6b9 1538 /* Contradict one of substrings */
97136c8a
DM
1539 if (prog->anchored_substr || prog->anchored_utf8) {
1540 if (prog->substrs->check_ix == 1) { /* check is float */
1541 /* Have both, check_string is floating */
b2ad2123
DM
1542 assert(rx_origin + start_shift <= check_at);
1543 if (rx_origin + start_shift != check_at) {
97136c8a 1544 /* not at latest position float substr could match:
c75a3985
DM
1545 * Recheck anchored substring, but not floating.
1546 * The condition above is in bytes rather than
1547 * chars for efficiency. It's conservative, in
1548 * that it errs on the side of doing 'goto
88203927
DM
1549 * do_other_substr'. In this case, at worst,
1550 * an extra anchored search may get done, but in
1551 * practice the extra fbm_instr() is likely to
1552 * get skipped anyway. */
6ad9a8ab 1553 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
147e3846 1554 " about to retry anchored at offset %ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
1555 (long)(other_last - strbeg),
1556 (IV)(rx_origin - strbeg)
1557 ));
97136c8a 1558 goto do_other_substr;
3369914b 1559 }
3369914b
DM
1560 }
1561 }
97136c8a 1562 else {
9fed8d02
DM
1563 /* float-only */
1564
7d2d37f5 1565 if (ml_anch) {
c75a3985
DM
1566 /* In the presence of ml_anch, we might be able to
1567 * find another \n without breaking the current float
1568 * constraint. */
1569
1570 /* strictly speaking this should be HOP3c(..., 1, ...),
1571 * but since we goto a block of code that's going to
1572 * search for the next \n if any, its safe here */
9fed8d02 1573 rx_origin++;
6ad9a8ab 1574 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
675e93ee 1575 " about to look for /%s^%s/m starting at rx_origin %ld...\n",
9fed8d02 1576 PL_colors[0], PL_colors[1],
675e93ee 1577 (long)(rx_origin - strbeg)) );
9fed8d02 1578 goto postprocess_substr_matches;
ab60c45a 1579 }
c75a3985
DM
1580
1581 /* strictly speaking this can never be true; but might
1582 * be if we ever allow intuit without substrings */
1583 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
9fed8d02 1584 goto fail;
c75a3985 1585
000dfd2d 1586 rx_origin = rx_max_float;
9fed8d02
DM
1587 }
1588
c75a3985
DM
1589 /* at this point, any matching substrings have been
1590 * contradicted. Start again... */
1591
9fed8d02 1592 rx_origin = HOP3c(rx_origin, 1, strend);
557f47af
DM
1593
1594 /* uses bytes rather than char calculations for efficiency.
1595 * It's conservative: it errs on the side of doing 'goto restart',
1596 * where there is code that does a proper char-based test */
b2ad2123 1597 if (rx_origin + start_shift + end_shift > strend) {
6ad9a8ab 1598 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
9fed8d02
DM
1599 " Could not match STCLASS...\n") );
1600 goto fail;
1601 }
6ad9a8ab 1602 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
147e3846 1603 " about to look for %s substr starting at offset %ld (rx_origin now %" IVdf ")...\n",
9fed8d02 1604 (prog->substrs->check_ix ? "floating" : "anchored"),
b2ad2123 1605 (long)(rx_origin + start_shift - strbeg),
675e93ee
DM
1606 (IV)(rx_origin - strbeg)
1607 ));
9fed8d02 1608 goto restart;
6eb5f6b9 1609 }
9fed8d02 1610
c75a3985
DM
1611 /* Success !!! */
1612
5f9c6575 1613 if (rx_origin != s) {
6ad9a8ab 1614 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 1615 " By STCLASS: moving %ld --> %ld\n",
675e93ee 1616 (long)(rx_origin - strbeg), (long)(s - strbeg))
b7953727
JH
1617 );
1618 }
1619 else {
6ad9a8ab 1620 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 1621 " Does not contradict STCLASS...\n");
b7953727
JH
1622 );
1623 }
6eb5f6b9 1624 }
ffad1e6a
DM
1625
1626 /* Decide whether using the substrings helped */
1627
1628 if (rx_origin != strpos) {
1629 /* Fixed substring is found far enough so that the match
1630 cannot start at strpos. */
1631
6ad9a8ab 1632 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " try at offset...\n"));
ffad1e6a
DM
1633 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1634 }
1635 else {
70563e16
DM
1636 /* The found rx_origin position does not prohibit matching at
1637 * strpos, so calling intuit didn't gain us anything. Decrement
1638 * the BmUSEFUL() count on the check substring, and if we reach
1639 * zero, free it. */
1640 if (!(prog->intflags & PREGf_NAUGHTY)
ffad1e6a
DM
1641 && (utf8_target ? (
1642 prog->check_utf8 /* Could be deleted already */
1643 && --BmUSEFUL(prog->check_utf8) < 0
1644 && (prog->check_utf8 == prog->float_utf8)
1645 ) : (
1646 prog->check_substr /* Could be deleted already */
1647 && --BmUSEFUL(prog->check_substr) < 0
1648 && (prog->check_substr == prog->float_substr)
1649 )))
1650 {
1651 /* If flags & SOMETHING - do not do it many times on the same match */
6ad9a8ab 1652 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " ... Disabling check substring...\n"));
ffad1e6a
DM
1653 /* XXX Does the destruction order has to change with utf8_target? */
1654 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1655 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1656 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1657 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1658 check = NULL; /* abort */
ffad1e6a
DM
1659 /* XXXX This is a remnant of the old implementation. It
1660 looks wasteful, since now INTUIT can use many
1661 other heuristics. */
1662 prog->extflags &= ~RXf_USE_INTUIT;
ffad1e6a
DM
1663 }
1664 }
1665
6ad9a8ab 1666 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
ffad1e6a 1667 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
675e93ee 1668 PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );
ffad1e6a 1669
c765d6e0 1670 return rx_origin;
2c2d71f5
JH
1671
1672 fail_finish: /* Substring not found */
33b8afdf 1673 if (prog->check_substr || prog->check_utf8) /* could be removed already */
f2ed9b32 1674 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
cad2e5aa 1675 fail:
6ad9a8ab 1676 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch rejected by optimizer%s\n",
e4584336 1677 PL_colors[4], PL_colors[5]));
bd61b366 1678 return NULL;
cad2e5aa 1679}
9661b544 1680
70563e16 1681
a0a388a1 1682#define DECL_TRIE_TYPE(scan) \
e7fd4aa1 1683 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
a4525e78 1684 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
3ed3004a 1685 trie_utf8l, trie_flu8, trie_flu8_latin } \
e7fd4aa1
KW
1686 trie_type = ((scan->flags == EXACT) \
1687 ? (utf8_target ? trie_utf8 : trie_plain) \
a4525e78
KW
1688 : (scan->flags == EXACTL) \
1689 ? (utf8_target ? trie_utf8l : trie_plain) \
89829bb5 1690 : (scan->flags == EXACTFAA) \
a4525e78
KW
1691 ? (utf8_target \
1692 ? trie_utf8_exactfa_fold \
1693 : trie_latin_utf8_exactfa_fold) \
1694 : (scan->flags == EXACTFLU8 \
3ed3004a
YO
1695 ? (utf8_target \
1696 ? trie_flu8 \
1697 : trie_flu8_latin) \
a4525e78
KW
1698 : (utf8_target \
1699 ? trie_utf8_fold \
3ed3004a 1700 : trie_latin_utf8_fold)))
fab2782b 1701
9ad8cac4
KW
1702/* 'uscan' is set to foldbuf, and incremented, so below the end of uscan is
1703 * 'foldbuf+sizeof(foldbuf)' */
1704#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uc_end, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
baa60164 1705STMT_START { \
fab2782b 1706 STRLEN skiplen; \
baa60164 1707 U8 flags = FOLD_FLAGS_FULL; \
fab2782b 1708 switch (trie_type) { \
a4525e78 1709 case trie_flu8: \
780fcc9f 1710 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3ed3004a 1711 if (UTF8_IS_ABOVE_LATIN1(*uc)) { \
e1a2878a 1712 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end); \
613abc6d 1713 } \
a4525e78 1714 goto do_trie_utf8_fold; \
31f05a37 1715 case trie_utf8_exactfa_fold: \
baa60164 1716 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1717 /* FALLTHROUGH */ \
fab2782b 1718 case trie_utf8_fold: \
a4525e78 1719 do_trie_utf8_fold: \
fab2782b 1720 if ( foldlen>0 ) { \
9ad8cac4 1721 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \
fab2782b
YO
1722 foldlen -= len; \
1723 uscan += len; \
1724 len=0; \
1725 } else { \
9ad8cac4 1726 uvc = _toFOLD_utf8_flags( (const U8*) uc, uc_end, foldbuf, &foldlen, \
a1a5ec35 1727 flags); \
695717ad 1728 len = UTF8_SAFE_SKIP(uc, uc_end); \
5f560d8a 1729 skiplen = UVCHR_SKIP( uvc ); \
fab2782b
YO
1730 foldlen -= skiplen; \
1731 uscan = foldbuf + skiplen; \
1732 } \
1733 break; \
3ed3004a
YO
1734 case trie_flu8_latin: \
1735 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1736 goto do_trie_latin_utf8_fold; \
baa60164
KW
1737 case trie_latin_utf8_exactfa_fold: \
1738 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1739 /* FALLTHROUGH */ \
fab2782b 1740 case trie_latin_utf8_fold: \
3ed3004a 1741 do_trie_latin_utf8_fold: \
fab2782b 1742 if ( foldlen>0 ) { \
9ad8cac4 1743 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \
fab2782b
YO
1744 foldlen -= len; \
1745 uscan += len; \
1746 len=0; \
1747 } else { \
1748 len = 1; \
31f05a37 1749 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
5f560d8a 1750 skiplen = UVCHR_SKIP( uvc ); \
fab2782b
YO
1751 foldlen -= skiplen; \
1752 uscan = foldbuf + skiplen; \
1753 } \
1754 break; \
a4525e78 1755 case trie_utf8l: \
780fcc9f 1756 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
613abc6d 1757 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
14f657d4 1758 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end); \
613abc6d 1759 } \
780fcc9f 1760 /* FALLTHROUGH */ \
fab2782b 1761 case trie_utf8: \
9ad8cac4 1762 uvc = utf8n_to_uvchr( (const U8*) uc, uc_end - uc, &len, uniflags ); \
fab2782b
YO
1763 break; \
1764 case trie_plain: \
1765 uvc = (UV)*uc; \
1766 len = 1; \
1767 } \
1768 if (uvc < 256) { \
1769 charid = trie->charmap[ uvc ]; \
1770 } \
1771 else { \
1772 charid = 0; \
1773 if (widecharmap) { \
1774 SV** const svpp = hv_fetch(widecharmap, \
1775 (char*)&uvc, sizeof(UV), 0); \
1776 if (svpp) \
1777 charid = (U16)SvIV(*svpp); \
1778 } \
1779 } \
4cadc6a9
YO
1780} STMT_END
1781
cb41e5d6 1782#define DUMP_EXEC_POS(li,s,doutf8,depth) \
ae7c5b9b 1783 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
cb41e5d6 1784 startpos, doutf8, depth)
ae7c5b9b 1785
da10aa09
KW
1786#define REXEC_FBC_SCAN(UTF8, CODE) \
1787 STMT_START { \
1788 while (s < strend) { \
1789 CODE \
1790 s += ((UTF8) ? UTF8SKIP(s) : 1); \
1791 } \
1792 } STMT_END
4cadc6a9 1793
d990bd30
KW
1794#define REXEC_FBC_CLASS_SCAN(UTF8, COND) \
1795 STMT_START { \
1796 while (s < strend) { \
1797 REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND) \
1798 } \
1799 } STMT_END
4cadc6a9 1800
d990bd30 1801#define REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND) \
05bd126c 1802 if (COND) { \
19719145 1803 FBC_CHECK_AND_TRY \
21d1ed54
KW
1804 s += ((UTF8) ? UTF8SKIP(s) : 1); \
1805 previous_occurrence_end = s; \
05bd126c 1806 } \
21d1ed54
KW
1807 else { \
1808 s += ((UTF8) ? UTF8SKIP(s) : 1); \
1809 }
4cadc6a9 1810
c84a03c5 1811#define REXEC_FBC_CSCAN(CONDUTF8,COND) \
baa60164 1812 if (utf8_target) { \
da10aa09 1813 REXEC_FBC_CLASS_SCAN(1, CONDUTF8); \
e1d1eefb
YO
1814 } \
1815 else { \
da10aa09 1816 REXEC_FBC_CLASS_SCAN(0, COND); \
d981ef24 1817 }
05bd126c 1818
a9448551
KW
1819/* We keep track of where the next character should start after an occurrence
1820 * of the one we're looking for. Knowing that, we can see right away if the
1821 * next occurrence is adjacent to the previous. When 'doevery' is FALSE, we
1822 * don't accept the 2nd and succeeding adjacent occurrences */
19719145
KW
1823#define FBC_CHECK_AND_TRY \
1824 if ( ( doevery \
1825 || s != previous_occurrence_end) \
1826 && (reginfo->intuit || regtry(reginfo, &s))) \
1827 { \
1828 goto got_it; \
1829 }
1830
a9448551
KW
1831
1832/* This differs from the above macros in that it calls a function which returns
1833 * the next occurrence of the thing being looked for in 's'; and 'strend' if
1834 * there is no such occurrence. */
1835#define REXEC_FBC_FIND_NEXT_SCAN(UTF8, f) \
1836 while (s < strend) { \
2b1f9c71 1837 s = (f); \
a9448551
KW
1838 if (s >= strend) { \
1839 break; \
1840 } \
1841 \
1842 FBC_CHECK_AND_TRY \
1843 s += (UTF8) ? UTF8SKIP(s) : 1; \
1844 previous_occurrence_end = s; \
1845 }
1846
ff8b8b42
KW
1847/* This differs from the above macros in that it is passed a single byte that
1848 * is known to begin the next occurrence of the thing being looked for in 's'.
1849 * It does a memchr to find the next occurrence of 'byte', before trying 'COND'
1850 * at that position. */
1851#define REXEC_FBC_FIND_NEXT_UTF8_BYTE_SCAN(byte, COND) \
1852 while (s < strend) { \
1853 s = (char *) memchr(s, byte, strend -s); \
1854 if (s == NULL) { \
1855 s = (char *) strend; \
1856 break; \
1857 } \
1858 \
04532561
KW
1859 if (COND) { \
1860 FBC_CHECK_AND_TRY \
1861 s += UTF8SKIP(s); \
1862 previous_occurrence_end = s; \
1863 } \
1864 else { \
1865 s += UTF8SKIP(s); \
1866 } \
ff8b8b42
KW
1867 }
1868
05bd126c
KW
1869/* The three macros below are slightly different versions of the same logic.
1870 *
1871 * The first is for /a and /aa when the target string is UTF-8. This can only
1872 * match ascii, but it must advance based on UTF-8. The other two handle the
1873 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1874 * for the boundary (or non-boundary) between a word and non-word character.
1875 * The utf8 and non-utf8 cases have the same logic, but the details must be
1876 * different. Find the "wordness" of the character just prior to this one, and
1877 * compare it with the wordness of this one. If they differ, we have a
1878 * boundary. At the beginning of the string, pretend that the previous
1879 * character was a new-line.
1880 *
1881 * All these macros uncleanly have side-effects with each other and outside
1882 * variables. So far it's been too much trouble to clean-up
1883 *
1884 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1885 * a word character or not.
1886 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1887 * word/non-word
1888 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1889 *
1890 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1891 * are looking for a boundary or for a non-boundary. If we are looking for a
1892 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1893 * see if this tentative match actually works, and if so, to quit the loop
1894 * here. And vice-versa if we are looking for a non-boundary.
1895 *
1896 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
da10aa09 1897 * REXEC_FBC_SCAN loops is a loop invariant, a bool giving the return of
05bd126c
KW
1898 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1899 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1900 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1901 * complement. But in that branch we complement tmp, meaning that at the
1902 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1903 * which means at the top of the loop in the next iteration, it is
1904 * TEST_NON_UTF8(s-1) */
b2f4e957 1905#define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW
1906 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1907 tmp = TEST_NON_UTF8(tmp); \
da10aa09 1908 REXEC_FBC_SCAN(1, /* 1=>is-utf8; advances s while s < strend */ \
05bd126c
KW
1909 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1910 tmp = !tmp; \
1911 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1912 } \
1913 else { \
1914 IF_FAIL; \
1915 } \
1916 ); \
1917
1918/* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1919 * TEST_UTF8 is a macro that for the same input code points returns identically
1920 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
236d82fd 1921#define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW
1922 if (s == reginfo->strbeg) { \
1923 tmp = '\n'; \
1924 } \
1925 else { /* Back-up to the start of the previous character */ \
1926 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1927 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
3db24e1e 1928 0, UTF8_ALLOW_DEFAULT); \
05bd126c
KW
1929 } \
1930 tmp = TEST_UV(tmp); \
da10aa09 1931 REXEC_FBC_SCAN(1, /* 1=>is-utf8; advances s while s < strend */ \
7a207065 1932 if (tmp == ! (TEST_UTF8((U8 *) s, (U8 *) reginfo->strend))) { \
05bd126c
KW
1933 tmp = !tmp; \
1934 IF_SUCCESS; \
1935 } \
1936 else { \
1937 IF_FAIL; \
1938 } \
1939 );
cfaf538b 1940
05bd126c
KW
1941/* Like the above two macros. UTF8_CODE is the complete code for handling
1942 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1943 * macros below */
baa60164 1944#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1945 if (utf8_target) { \
05bd126c 1946 UTF8_CODE \
63ac0dad
KW
1947 } \
1948 else { /* Not utf8 */ \
9d9163fb 1949 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad 1950 tmp = TEST_NON_UTF8(tmp); \
da10aa09 1951 REXEC_FBC_SCAN(0, /* 0=>not-utf8; advances s while s < strend */ \
63ac0dad 1952 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
63ac0dad 1953 IF_SUCCESS; \
760cfa8e 1954 tmp = !tmp; \
63ac0dad
KW
1955 } \
1956 else { \
1957 IF_FAIL; \
1958 } \
1959 ); \
1960 } \
c8519dc7
KW
1961 /* Here, things have been set up by the previous code so that tmp is the \
1962 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1963 * utf8ness of the target). We also have to check if this matches against \
1964 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1965 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1966 * string */ \
1967 if (tmp == ! TEST_NON_UTF8('\n')) { \
1968 IF_SUCCESS; \
1969 } \
1970 else { \
1971 IF_FAIL; \
1972 }
63ac0dad 1973
ae7c5b9b 1974/* This is the macro to use when we want to see if something that looks like it
b92b2705
KW
1975 * could match, actually does, and if so exits the loop. It needs to be used
1976 * only for bounds checking macros, as it allows for matching beyond the end of
1977 * string (which should be zero length without having to look at the string
1978 * contents) */
1979#define REXEC_FBC_TRYIT \
1980 if ((reginfo->intuit || (s <= reginfo->strend && regtry(reginfo, &s)))) \
ae7c5b9b
KW
1981 goto got_it
1982
1983/* The only difference between the BOUND and NBOUND cases is that
1984 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1985 * NBOUND. This is accomplished by passing it as either the if or else clause,
1986 * with the other one being empty (PLACEHOLDER is defined as empty).
1987 *
1988 * The TEST_FOO parameters are for operating on different forms of input, but
1989 * all should be ones that return identically for the same underlying code
1990 * points */
1991#define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1992 FBC_BOUND_COMMON( \
1993 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1994 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1995
44129e46 1996#define FBC_BOUND_A(TEST_NON_UTF8) \
ae7c5b9b
KW
1997 FBC_BOUND_COMMON( \
1998 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1999 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
2000
2001#define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
2002 FBC_BOUND_COMMON( \
2003 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
2004 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
2005
44129e46 2006#define FBC_NBOUND_A(TEST_NON_UTF8) \
ae7c5b9b
KW
2007 FBC_BOUND_COMMON( \
2008 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
2009 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
2010
8bde5eaf
JH
2011#ifdef DEBUGGING
2012static IV
2013S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) {
9ffbcbcf 2014 IV cp_out = _invlist_search(invlist, cp_in);
8bde5eaf
JH
2015 assert(cp_out >= 0);
2016 return cp_out;
2017}
2018# define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
2019 invmap[S_get_break_val_cp_checked(invlist, cp)]
2020#else
2021# define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
2022 invmap[_invlist_search(invlist, cp)]
2023#endif
2024
64935bc6
KW
2025/* Takes a pointer to an inversion list, a pointer to its corresponding
2026 * inversion map, and a code point, and returns the code point's value
2027 * according to the two arrays. It assumes that all code points have a value.
2028 * This is used as the base macro for macros for particular properties */
2029#define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \
8bde5eaf 2030 _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp)
64935bc6
KW
2031
2032/* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
2033 * of a code point, returning the value for the first code point in the string.
2034 * And it takes the particular macro name that finds the desired value given a
2035 * code point. Merely convert the UTF-8 to code point and call the cp macro */
2036#define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \
2037 (__ASSERT_(pos < strend) \
2038 /* Note assumes is valid UTF-8 */ \
2039 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))
2040
2041/* Returns the GCB value for the input code point */
2042#define getGCB_VAL_CP(cp) \
2043 _generic_GET_BREAK_VAL_CP( \
2044 PL_GCB_invlist, \
02f811dd 2045 _Perl_GCB_invmap, \
64935bc6
KW
2046 (cp))
2047
2048/* Returns the GCB value for the first code point in the UTF-8 encoded string
2049 * bounded by pos and strend */
2050#define getGCB_VAL_UTF8(pos, strend) \
2051 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)
05bd126c 2052
6b659339
KW
2053/* Returns the LB value for the input code point */
2054#define getLB_VAL_CP(cp) \
2055 _generic_GET_BREAK_VAL_CP( \
2056 PL_LB_invlist, \
2057 _Perl_LB_invmap, \
2058 (cp))
2059
2060/* Returns the LB value for the first code point in the UTF-8 encoded string
2061 * bounded by pos and strend */
2062#define getLB_VAL_UTF8(pos, strend) \
2063 _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend)
2064
06ae2722
KW
2065
2066/* Returns the SB value for the input code point */
2067#define getSB_VAL_CP(cp) \
2068 _generic_GET_BREAK_VAL_CP( \
2069 PL_SB_invlist, \
bf4268fa 2070 _Perl_SB_invmap, \
06ae2722
KW
2071 (cp))
2072
2073/* Returns the SB value for the first code point in the UTF-8 encoded string
2074 * bounded by pos and strend */
2075#define getSB_VAL_UTF8(pos, strend) \
2076 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)
2077
ae3bb8ea
KW
2078/* Returns the WB value for the input code point */
2079#define getWB_VAL_CP(cp) \
2080 _generic_GET_BREAK_VAL_CP( \
2081 PL_WB_invlist, \
bf4268fa 2082 _Perl_WB_invmap, \
ae3bb8ea
KW
2083 (cp))
2084
2085/* Returns the WB value for the first code point in the UTF-8 encoded string
2086 * bounded by pos and strend */
2087#define getWB_VAL_UTF8(pos, strend) \
2088 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)
2089
786e8c11 2090/* We know what class REx starts with. Try to find this position... */
02d5137b 2091/* if reginfo->intuit, its a dryrun */
786e8c11
YO
2092/* annoyingly all the vars in this routine have different names from their counterparts
2093 in regmatch. /grrr */
3c3eec57 2094STATIC char *
07be1b83 2095S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 2096 const char *strend, regmatch_info *reginfo)
a687059c 2097{
73104a1b 2098 dVAR;
a9448551
KW
2099
2100 /* TRUE if x+ need not match at just the 1st pos of run of x's */
73104a1b 2101 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
a9448551 2102
73104a1b
KW
2103 char *pat_string; /* The pattern's exactish string */
2104 char *pat_end; /* ptr to end char of pat_string */
2105 re_fold_t folder; /* Function for computing non-utf8 folds */
2106 const U8 *fold_array; /* array for folding ords < 256 */
2107 STRLEN ln;
2108 STRLEN lnc;
73104a1b
KW
2109 U8 c1;
2110 U8 c2;
743dd5b8 2111 char *e = NULL;
21d1ed54
KW
2112
2113 /* In some cases we accept only the first occurence of 'x' in a sequence of
2114 * them. This variable points to just beyond the end of the previous
2115 * occurrence of 'x', hence we can tell if we are in a sequence. (Having
2116 * it point to beyond the 'x' allows us to work for UTF-8 without having to
2117 * hop back.) */
2118 char * previous_occurrence_end = 0;
2119
3b6c52ce 2120 I32 tmp; /* Scratch variable */
ba44c216 2121 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 2122 UV utf8_fold_flags = 0;
f9176b44 2123 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW
2124 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
2125 with a result inverts that result, as 0^1 =
2126 1 and 1^1 = 0 */
2127 _char_class_number classnum;
2128
73104a1b 2129 RXi_GET_DECL(prog,progi);
2f7f8cb1 2130
73104a1b 2131 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 2132
73104a1b
KW
2133 /* We know what class it must start with. */
2134 switch (OP(c)) {
3edce4f5 2135 case ANYOFPOSIXL:
a4525e78 2136 case ANYOFL:
780fcc9f 2137 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
a0bd1a30 2138
d1c40ef5 2139 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) {
a0bd1a30
KW
2140 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
2141 }
2142
780fcc9f 2143 /* FALLTHROUGH */
ac44c12e 2144 case ANYOFD:
73104a1b
KW
2145 case ANYOF:
2146 if (utf8_target) {
da10aa09 2147 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
3db24e1e 2148 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
73104a1b 2149 }
daced5ae
KW
2150 else if (ANYOF_FLAGS(c) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
2151 /* We know that s is in the bitmap range since the target isn't
2152 * UTF-8, so what happens for out-of-range values is not relevant,
2153 * so exclude that from the flags */
da10aa09 2154 REXEC_FBC_CLASS_SCAN(0, reginclass(prog,c, (U8*)s, (U8*)s+1, 0));
1451f692 2155 }
73104a1b 2156 else {
da10aa09 2157 REXEC_FBC_CLASS_SCAN(0, ANYOF_BITMAP_TEST(c, *((U8*)s)));
73104a1b
KW
2158 }
2159 break;
73104a1b 2160
2813d4ad 2161 case ANYOFM: /* ARG() is the base byte; FLAGS() the mask byte */
0a5ed81e
KW
2162 /* UTF-8ness doesn't matter because only matches UTF-8 invariants, so
2163 * use 0 */
2813d4ad 2164 REXEC_FBC_FIND_NEXT_SCAN(0,
12475f8b
KW
2165 (char *) find_next_masked((U8 *) s, (U8 *) strend,
2166 (U8) ARG(c), FLAGS(c)));
2813d4ad
KW
2167 break;
2168
0a5ed81e
KW
2169 case NANYOFM: /* UTF-8ness does matter because can match UTF-8 variants.
2170 */
2171 REXEC_FBC_FIND_NEXT_SCAN(utf8_target,
3db0bccc
KW
2172 (char *) find_span_end_mask((U8 *) s, (U8 *) strend,
2173 (U8) ARG(c), FLAGS(c)));
2174 break;
2175
c316b824 2176 case ANYOFH:
ff8b8b42
KW
2177 if (utf8_target) { /* Can't possibly match a non-UTF-8 target */
2178 U8 first_byte = FLAGS(c);
2179
2180 if (first_byte) { /* We know what the first byte of any matched
2181 string should be */
2182 REXEC_FBC_FIND_NEXT_UTF8_BYTE_SCAN(first_byte,
2183 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
2184 }
2185 else {
2186 REXEC_FBC_CLASS_SCAN(TRUE,
c316b824 2187 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
ff8b8b42
KW
2188 }
2189 }
c316b824
KW
2190 break;
2191
89829bb5 2192 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
098b07d5 2193 assert(! is_utf8_pat);
924ba076 2194 /* FALLTHROUGH */
89829bb5 2195 case EXACTFAA:
aa419ff3
KW
2196 if (is_utf8_pat) {
2197 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII
2198 |FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
2199 goto do_exactf_utf8;
2200 }
2201 else if (utf8_target) {
2202
2203 /* Here, and elsewhere in this file, the reason we can't consider a
2204 * non-UTF-8 pattern already folded in the presence of a UTF-8
2205 * target is because any MICRO SIGN in the pattern won't be folded.
2206 * Since the fold of the MICRO SIGN requires UTF-8 to represent, we
2207 * can consider a non-UTF-8 pattern folded when matching a
2208 * non-UTF-8 target */
73104a1b
KW
2209 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
2210 goto do_exactf_utf8;
2211 }
aa419ff3
KW
2212
2213 /* Latin1 folds are not affected by /a, except it excludes the sharp s,
2214 * which these functions don't handle anyway */
2215 fold_array = PL_fold_latin1;
2216 folder = foldEQ_latin1_s2_folded;
2217 goto do_exactf_non_utf8;
77a6d856 2218
2fdb7295
KW
2219 case EXACTF: /* This node only generated for non-utf8 patterns */
2220 assert(! is_utf8_pat);
73104a1b 2221 if (utf8_target) {
73104a1b
KW
2222 goto do_exactf_utf8;
2223 }
2224 fold_array = PL_fold;
2225 folder = foldEQ;
2226 goto do_exactf_non_utf8;
2227
2228 case EXACTFL:
780fcc9f 2229 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
31f05a37 2230 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
cea315b6 2231 utf8_fold_flags = FOLDEQ_LOCALE;
73104a1b
KW
2232 goto do_exactf_utf8;
2233 }
2234 fold_array = PL_fold_locale;
2235 folder = foldEQ_locale;
2236 goto do_exactf_non_utf8;
3c760661 2237
627a7895
KW
2238 case EXACTFUP: /* Problematic even though pattern isn't UTF-8. Use
2239 full functionality normally not done except for
2240 UTF-8 */
a56633d1 2241 assert(! is_utf8_pat);
73104a1b 2242 goto do_exactf_utf8;
16d951b7 2243
a4525e78
KW
2244 case EXACTFLU8:
2245 if (! utf8_target) { /* All code points in this node require
2246 UTF-8 to express. */
2247 break;
2248 }
613abc6d
KW
2249 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
2250 | FOLDEQ_S2_FOLDS_SANE;
a4525e78
KW
2251 goto do_exactf_utf8;
2252
a9f8c7ac
KW
2253 case EXACTFU_ONLY8:
2254 if (! utf8_target) {
2255 break;
2256 }
2257 assert(is_utf8_pat);
2258 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
2259 goto do_exactf_utf8;
2260
73104a1b 2261 case EXACTFU:
984e6dd1 2262 if (is_utf8_pat || utf8_target) {
0fbec7cf 2263 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
73104a1b
KW
2264 goto do_exactf_utf8;
2265 }
fac1af77 2266
73104a1b
KW
2267 /* Any 'ss' in the pattern should have been replaced by regcomp,
2268 * so we don't have to worry here about this single special case
2269 * in the Latin1 range */
2270 fold_array = PL_fold_latin1;
0fbec7cf 2271 folder = foldEQ_latin1_s2_folded;
73104a1b 2272
924ba076 2273 /* FALLTHROUGH */
73104a1b 2274
c52b8b12 2275 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
73104a1b
KW
2276 are no glitches with fold-length differences
2277 between the target string and pattern */
2278
2279 /* The idea in the non-utf8 EXACTF* cases is to first find the
2280 * first character of the EXACTF* node and then, if necessary,
2281 * case-insensitively compare the full text of the node. c1 is the
2282 * first character. c2 is its fold. This logic will not work for
2283 * Unicode semantics and the german sharp ss, which hence should
2284 * not be compiled into a node that gets here. */
2285 pat_string = STRING(c);
2286 ln = STR_LEN(c); /* length to match in octets/bytes */
2287
2288 /* We know that we have to match at least 'ln' bytes (which is the
2289 * same as characters, since not utf8). If we have to match 3
2290 * characters, and there are only 2 availabe, we know without
2291 * trying that it will fail; so don't start a match past the
2292 * required minimum number from the far end */
ea3daa5d 2293 e = HOP3c(strend, -((SSize_t)ln), s);
dda01918
HS
2294 if (e < s)
2295 break;
fac1af77 2296
73104a1b
KW
2297 c1 = *pat_string;
2298 c2 = fold_array[c1];
2299 if (c1 == c2) { /* If char and fold are the same */
c05cc3b6
KW
2300 while (s <= e) {
2301 s = (char *) memchr(s, c1, e + 1 - s);
2302 if (s == NULL) {
2303 break;
2304 }
2305
2306 /* Check that the rest of the node matches */
2307 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2308 && (reginfo->intuit || regtry(reginfo, &s)) )
2309 {
2310 goto got_it;
2311 }
2312 s++;
2313 }
73104a1b
KW
2314 }
2315 else {
c05cc3b6
KW
2316 U8 bits_differing = c1 ^ c2;
2317
2318 /* If the folds differ in one bit position only, we can mask to
2319 * match either of them, and can use this faster find method. Both
2320 * ASCII and EBCDIC tend to have their case folds differ in only
2321 * one position, so this is very likely */
2322 if (LIKELY(PL_bitcount[bits_differing] == 1)) {
2323 bits_differing = ~ bits_differing;
2324 while (s <= e) {
2b1f9c71 2325 s = (char *) find_next_masked((U8 *) s, (U8 *) e + 1,
c05cc3b6
KW
2326 (c1 & bits_differing), bits_differing);
2327 if (s > e) {
2328 break;
2329 }
2330
2331 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2332 && (reginfo->intuit || regtry(reginfo, &s)) )
2333 {
2334 goto got_it;
2335 }
2336 s++;
2337 }
2338 }
2339 else { /* Otherwise, stuck with looking byte-at-a-time. This
2340 should actually happen only in EXACTFL nodes */
2341 while (s <= e) {
2342 if ( (*(U8*)s == c1 || *(U8*)s == c2)
2343 && (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2344 && (reginfo->intuit || regtry(reginfo, &s)) )
2345 {
2346 goto got_it;
2347 }
2348 s++;
2349 }
2350 }
73104a1b
KW
2351 }
2352 break;
fac1af77 2353
c52b8b12
KW
2354 do_exactf_utf8:
2355 {
73104a1b
KW
2356 unsigned expansion;
2357
2358 /* If one of the operands is in utf8, we can't use the simpler folding
2359 * above, due to the fact that many different characters can have the
2360 * same fold, or portion of a fold, or different- length fold */
2361 pat_string = STRING(c);
2362 ln = STR_LEN(c); /* length to match in octets/bytes */
2363 pat_end = pat_string + ln;
984e6dd1 2364 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW
2365 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
2366 : ln;
2367
2368 /* We have 'lnc' characters to match in the pattern, but because of
2369 * multi-character folding, each character in the target can match
2370 * up to 3 characters (Unicode guarantees it will never exceed
2371 * this) if it is utf8-encoded; and up to 2 if not (based on the
2372 * fact that the Latin 1 folds are already determined, and the
2373 * only multi-char fold in that range is the sharp-s folding to
2374 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
2375 * string character. Adjust lnc accordingly, rounding up, so that
2376 * if we need to match at least 4+1/3 chars, that really is 5. */
2377 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
2378 lnc = (lnc + expansion - 1) / expansion;
2379
2380 /* As in the non-UTF8 case, if we have to match 3 characters, and
2381 * only 2 are left, it's guaranteed to fail, so don't start a
2382 * match that would require us to go beyond the end of the string
2383 */
ea3daa5d 2384 e = HOP3c(strend, -((SSize_t)lnc), s);
73104a1b 2385
73104a1b
KW
2386 /* XXX Note that we could recalculate e to stop the loop earlier,
2387 * as the worst case expansion above will rarely be met, and as we
2388 * go along we would usually find that e moves further to the left.
2389 * This would happen only after we reached the point in the loop
2390 * where if there were no expansion we should fail. Unclear if
2391 * worth the expense */
2392
2393 while (s <= e) {
2394 char *my_strend= (char *)strend;
2395 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 2396 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 2397 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW
2398 {
2399 goto got_it;
2400 }
2401 s += (utf8_target) ? UTF8SKIP(s) : 1;
2402 }
2403 break;
2404 }
236d82fd 2405
73104a1b 2406 case BOUNDL:
780fcc9f 2407 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
64935bc6 2408 if (FLAGS(c) != TRADITIONAL_BOUND) {
89ad707a
KW
2409 if (! IN_UTF8_CTYPE_LOCALE) {
2410 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
64935bc6 2411 B_ON_NON_UTF8_LOCALE_IS_WRONG);
89ad707a 2412 }
64935bc6
KW
2413 goto do_boundu;
2414 }
2415
7a207065 2416 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
73104a1b 2417 break;
64935bc6 2418
73104a1b 2419 case NBOUNDL:
780fcc9f 2420 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
64935bc6 2421 if (FLAGS(c) != TRADITIONAL_BOUND) {
89ad707a
KW
2422 if (! IN_UTF8_CTYPE_LOCALE) {
2423 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
64935bc6 2424 B_ON_NON_UTF8_LOCALE_IS_WRONG);
89ad707a 2425 }
64935bc6
KW
2426 goto do_nboundu;
2427 }
2428
7a207065 2429 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
73104a1b 2430 break;
64935bc6
KW
2431
2432 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2433 meaning */
2434 assert(FLAGS(c) == TRADITIONAL_BOUND);
2435
7a207065 2436 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
73104a1b 2437 break;
64935bc6
KW
2438
2439 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2440 meaning */
2441 assert(FLAGS(c) == TRADITIONAL_BOUND);
2442
44129e46 2443 FBC_BOUND_A(isWORDCHAR_A);
73104a1b 2444 break;
64935bc6
KW
2445
2446 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2447 meaning */
2448 assert(FLAGS(c) == TRADITIONAL_BOUND);
2449
7a207065 2450 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
73104a1b 2451 break;
64935bc6
KW
2452
2453 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2454 meaning */
2455 assert(FLAGS(c) == TRADITIONAL_BOUND);
2456
44129e46 2457 FBC_NBOUND_A(isWORDCHAR_A);
73104a1b 2458 break;
64935bc6 2459
73104a1b 2460 case NBOUNDU:
64935bc6 2461 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
7a207065 2462 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
64935bc6
KW
2463 break;
2464 }
2465
2466 do_nboundu:
2467
2468 to_complement = 1;
2469 /* FALLTHROUGH */
2470
2471 case BOUNDU:
2472 do_boundu:
2473 switch((bound_type) FLAGS(c)) {
2474 case TRADITIONAL_BOUND:
7a207065 2475 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
64935bc6
KW
2476 break;
2477 case GCB_BOUND:
a7a8bd1e 2478 if (s == reginfo->strbeg) {
67481c39 2479 if (reginfo->intuit || regtry(reginfo, &s))
64935bc6
KW
2480 {
2481 goto got_it;
2482 }
a7a8bd1e
KW
2483
2484 /* Didn't match. Try at the next position (if there is one) */
64935bc6 2485 s += (utf8_target) ? UTF8SKIP(s) : 1;
a7a8bd1e
KW
2486 if (UNLIKELY(s >= reginfo->strend)) {
2487 break;
2488 }
64935bc6
KW
2489 }
2490
2491 if (utf8_target) {
85e5f08b 2492 GCB_enum before = getGCB_VAL_UTF8(
64935bc6
KW
2493 reghop3((U8*)s, -1,
2494 (U8*)(reginfo->strbeg)),
2495 (U8*) reginfo->strend);
2496 while (s < strend) {
85e5f08b 2497 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
64935bc6 2498 (U8*) reginfo->strend);
b0e24409
KW
2499 if ( (to_complement ^ isGCB(before,
2500 after,
2501 (U8*) reginfo->strbeg,
2502 (U8*) s,
2503 utf8_target))
00e3344b
KW
2504 && (reginfo->intuit || regtry(reginfo, &s)))
2505 {
2506 goto got_it;
64935bc6 2507 }
43a7bd62 2508 before = after;
64935bc6
KW
2509 s += UTF8SKIP(s);
2510 }
2511 }
2512 else { /* Not utf8. Everything is a GCB except between CR and
2513 LF */
2514 while (s < strend) {
00e3344b
KW
2515 if ((to_complement ^ ( UCHARAT(s - 1) != '\r'
2516 || UCHARAT(s) != '\n'))
2517 && (reginfo->intuit || regtry(reginfo, &s)))
64935bc6 2518 {
00e3344b 2519 goto got_it;
64935bc6 2520 }
43a7bd62 2521 s++;
64935bc6
KW
2522 }
2523 }
2524
6de80efc
KW
2525 /* And, since this is a bound, it can match after the final
2526 * character in the string */
67481c39 2527 if ((reginfo->intuit || regtry(reginfo, &s))) {
64935bc6
KW
2528 goto got_it;
2529 }
2530 break;
ae3bb8ea 2531
6b659339
KW
2532 case LB_BOUND:
2533 if (s == reginfo->strbeg) {
2534 if (reginfo->intuit || regtry(reginfo, &s)) {
2535 goto got_it;
2536 }
2537 s += (utf8_target) ? UTF8SKIP(s) : 1;
2538 if (UNLIKELY(s >= reginfo->strend)) {
2539 break;
2540 }
2541 }
2542
2543 if (utf8_target) {
2544 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
2545 -1,
2546 (U8*)(reginfo->strbeg)),
2547 (U8*) reginfo->strend);
2548 while (s < strend) {
2549 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
2550 if (to_complement ^ isLB(before,
2551 after,
2552 (U8*) reginfo->strbeg,
2553 (U8*) s,
2554 (U8*) reginfo->strend,
2555 utf8_target)
2556 && (reginfo->intuit || regtry(reginfo, &s)))
2557 {
2558 goto got_it;
2559 }
2560 before = after;
2561 s += UTF8SKIP(s);
2562 }
2563 }
2564 else { /* Not utf8. */
2565 LB_enum before = getLB_VAL_CP((U8) *(s -1));
2566 while (s < strend) {
2567 LB_enum after = getLB_VAL_CP((U8) *s);
2568 if (to_complement ^ isLB(before,
2569 after,
2570 (U8*) reginfo->strbeg,
2571 (U8*) s,
2572 (U8*) reginfo->strend,
2573 utf8_target)
2574 && (reginfo->intuit || regtry(reginfo, &s)))
2575 {
2576 goto got_it;
2577 }
2578 before = after;
2579 s++;
2580 }
2581 }
2582
2583 if (reginfo->intuit || regtry(reginfo, &s)) {
2584 goto got_it;
2585 }
2586
2587 break;
2588
06ae2722 2589 case SB_BOUND:
a7a8bd1e 2590 if (s == reginfo->strbeg) {
67481c39 2591 if (reginfo->intuit || regtry(reginfo, &s)) {
06ae2722
KW
2592 goto got_it;
2593 }
06ae2722 2594 s += (utf8_target) ? UTF8SKIP(s) : 1;
a7a8bd1e
KW
2595 if (UNLIKELY(s >= reginfo->strend)) {
2596 break;
2597 }
06ae2722
KW
2598 }
2599
2600 if (utf8_target) {
85e5f08b 2601 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
06ae2722
KW
2602 -1,
2603 (U8*)(reginfo->strbeg)),
2604 (U8*) reginfo->strend);
2605 while (s < strend) {
85e5f08b 2606 SB_enum after = getSB_VAL_UTF8((U8*) s,
06ae2722 2607 (U8*) reginfo->strend);
00e3344b
KW
2608 if ((to_complement ^ isSB(before,
2609 after,
2610 (U8*) reginfo->strbeg,
2611 (U8*) s,
2612 (U8*) reginfo->strend,
2613 utf8_target))
2614 && (reginfo->intuit || regtry(reginfo, &s)))
06ae2722 2615 {
00e3344b 2616 goto got_it;
06ae2722 2617 }
43a7bd62 2618 before = after;
06ae2722
KW
2619 s += UTF8SKIP(s);
2620 }
2621 }
2622 else { /* Not utf8. */
85e5f08b 2623 SB_enum before = getSB_VAL_CP((U8) *(s -1));
06ae2722 2624 while (s < strend) {
85e5f08b 2625 SB_enum after = getSB_VAL_CP((U8) *s);
00e3344b
KW
2626 if ((to_complement ^ isSB(before,
2627 after,
2628 (U8*) reginfo->strbeg,
2629 (U8*) s,
2630 (U8*) reginfo->strend,
2631 utf8_target))
2632 && (reginfo->intuit || regtry(reginfo, &s)))
06ae2722 2633 {
00e3344b 2634 goto got_it;
06ae2722 2635 }
43a7bd62 2636 before = after;
06ae2722
KW
2637 s++;
2638 }
2639 }
2640
2641 /* Here are at the final position in the target string. The SB
2642 * value is always true here, so matches, depending on other
2643 * constraints */
67481c39 2644 if (reginfo->intuit || regtry(reginfo, &s)) {
06ae2722
KW
2645 goto got_it;
2646 }
2647
2648 break;
2649
ae3bb8ea
KW
2650 case WB_BOUND:
2651 if (s == reginfo->strbeg) {
67481c39 2652 if (reginfo->intuit || regtry(reginfo, &s)) {
ae3bb8ea
KW
2653 goto got_it;
2654 }
2655 s += (utf8_target) ? UTF8SKIP(s) : 1;
a7a8bd1e
KW
2656 if (UNLIKELY(s >= reginfo->strend)) {
2657 break;
2658 }
ae3bb8ea
KW
2659 }
2660
2661 if (utf8_target) {
2662 /* We are at a boundary between char_sub_0 and char_sub_1.
2663 * We also keep track of the value for char_sub_-1 as we
2664 * loop through the line. Context may be needed to make a
2665 * determination, and if so, this can save having to
2666 * recalculate it */
85e5f08b
KW
2667 WB_enum previous = WB_UNKNOWN;
2668 WB_enum before = getWB_VAL_UTF8(
ae3bb8ea
KW
2669 reghop3((U8*)s,
2670 -1,
2671 (U8*)(reginfo->strbeg)),
2672 (U8*) reginfo->strend);
2673 while (s < strend) {
85e5f08b 2674 WB_enum after = getWB_VAL_UTF8((U8*) s,
ae3bb8ea 2675 (U8*) reginfo->strend);
00e3344b
KW
2676 if ((to_complement ^ isWB(previous,
2677 before,
2678 after,
2679 (U8*) reginfo->strbeg,
2680 (U8*) s,
2681 (U8*) reginfo->strend,
2682 utf8_target))
2683 && (reginfo->intuit || regtry(reginfo, &s)))
ae3bb8ea 2684 {
00e3344b 2685 goto got_it;
ae3bb8ea 2686 }
43a7bd62
KW
2687 previous = before;
2688 before = after;
ae3bb8ea
KW
2689 s += UTF8SKIP(s);
2690 }
2691 }
2692 else { /* Not utf8. */
85e5f08b
KW
2693 WB_enum previous = WB_UNKNOWN;
2694 WB_enum before = getWB_VAL_CP((U8) *(s -1));
ae3bb8ea 2695 while (s < strend) {
85e5f08b 2696 WB_enum after = getWB_VAL_CP((U8) *s);
00e3344b
KW
2697 if ((to_complement ^ isWB(previous,
2698 before,
2699 after,
2700 (U8*) reginfo->strbeg,
2701 (U8*) s,
2702 (U8*) reginfo->strend,
2703 utf8_target))
2704 && (reginfo->intuit || regtry(reginfo, &s)))
ae3bb8ea 2705 {
00e3344b 2706 goto got_it;
ae3bb8ea 2707 }
43a7bd62
KW
2708 previous = before;
2709 before = after;
ae3bb8ea
KW
2710 s++;
2711 }
2712 }
2713
67481c39 2714 if (reginfo->intuit || regtry(reginfo, &s)) {
ae3bb8ea
KW
2715 goto got_it;
2716 }
64935bc6 2717 }
73104a1b 2718 break;
64935bc6 2719
73104a1b
KW
2720 case LNBREAK:
2721 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2722 is_LNBREAK_latin1_safe(s, strend)
2723 );
2724 break;
3018b823
KW
2725
2726 /* The argument to all the POSIX node types is the class number to pass to
2727 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2728
2729 case NPOSIXL:
2730 to_complement = 1;
2731 /* FALLTHROUGH */
2732
2733 case POSIXL:
780fcc9f 2734 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
a78c2fa6 2735 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s, (U8 *) strend)),
3018b823 2736 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 2737 break;
3018b823
KW
2738
2739 case NPOSIXD:
2740 to_complement = 1;
2741 /* FALLTHROUGH */
2742
2743 case POSIXD:
2744 if (utf8_target) {
2745 goto posix_utf8;
2746 }
2747 goto posixa;
2748
2749 case NPOSIXA:
2750 if (utf8_target) {
2751 /* The complement of something that matches only ASCII matches all
837226c8 2752 * non-ASCII, plus everything in ASCII that isn't in the class. */
da10aa09
KW
2753 REXEC_FBC_CLASS_SCAN(1, ! isASCII_utf8_safe(s, strend)
2754 || ! _generic_isCC_A(*s, FLAGS(c)));
3018b823
KW
2755 break;
2756 }
2757
2758 to_complement = 1;
4a6c6db5 2759 goto posixa;
3018b823 2760
73104a1b
KW
2761 case POSIXA:
2762 /* Don't need to worry about utf8, as it can match only a single
4a6c6db5
KW
2763 * byte invariant character. But we do anyway for performance reasons,
2764 * as otherwise we would have to examine all the continuation
2765 * characters */
2766 if (utf8_target) {
da10aa09 2767 REXEC_FBC_CLASS_SCAN(1, _generic_isCC_A(*s, FLAGS(c)));
4a6c6db5
KW
2768 break;
2769 }
2770
2771 posixa:
da10aa09 2772 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
3018b823 2773 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 2774 break;
3018b823
KW
2775
2776 case NPOSIXU:
2777 to_complement = 1;
2778 /* FALLTHROUGH */
2779
2780 case POSIXU:
2781 if (! utf8_target) {
da10aa09
KW
2782 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
2783 to_complement ^ cBOOL(_generic_isCC(*s,
3018b823
KW
2784 FLAGS(c))));
2785 }
2786 else {
2787
c52b8b12 2788 posix_utf8:
3018b823 2789 classnum = (_char_class_number) FLAGS(c);
8d692afb
KW
2790 switch (classnum) {
2791 default:
2792 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2793 to_complement ^ cBOOL(_invlist_contains_cp(
2794 PL_XPosix_ptrs[classnum],
2795 utf8_to_uvchr_buf((U8 *) s,
2796 (U8 *) strend,
2797 NULL))));
2798 break;
779cf272 2799 case _CC_ENUM_SPACE:
da10aa09 2800 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
7a207065 2801 to_complement ^ cBOOL(isSPACE_utf8_safe(s, strend)));
3018b823
KW
2802 break;
2803
2804 case _CC_ENUM_BLANK:
da10aa09 2805 REXEC_FBC_CLASS_SCAN(1,
7a207065 2806 to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
3018b823
KW
2807 break;
2808
2809 case _CC_ENUM_XDIGIT:
da10aa09 2810 REXEC_FBC_CLASS_SCAN(1,
7a207065 2811 to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
3018b823
KW
2812 break;
2813
2814 case _CC_ENUM_VERTSPACE:
da10aa09 2815 REXEC_FBC_CLASS_SCAN(1,
7a207065 2816 to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
3018b823
KW
2817 break;
2818
2819 case _CC_ENUM_CNTRL:
da10aa09 2820 REXEC_FBC_CLASS_SCAN(1,
7a207065 2821 to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
3018b823 2822 break;
3018b823
KW
2823 }
2824 }
2825 break;
2826
73104a1b
KW
2827 case AHOCORASICKC:
2828 case AHOCORASICK:
2829 {
2830 DECL_TRIE_TYPE(c);
2831 /* what trie are we using right now */
2832 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2833 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2834 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2835
2836 const char *last_start = strend - trie->minlen;
6148ee25 2837#ifdef DEBUGGING
73104a1b 2838 const char *real_start = s;
6148ee25 2839#endif
73104a1b
KW
2840 STRLEN maxlen = trie->maxlen;
2841 SV *sv_points;
2842 U8 **points; /* map of where we were in the input string
2843 when reading a given char. For ASCII this
2844 is unnecessary overhead as the relationship
2845 is always 1:1, but for Unicode, especially
2846 case folded Unicode this is not true. */
2847 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2848 U8 *bitmap=NULL;
2849
2850
2851 GET_RE_DEBUG_FLAGS_DECL;
2852
2853 /* We can't just allocate points here. We need to wrap it in
2854 * an SV so it gets freed properly if there is a croak while
2855 * running the match */
2856 ENTER;
2857 SAVETMPS;
2858 sv_points=newSV(maxlen * sizeof(U8 *));
2859 SvCUR_set(sv_points,
2860 maxlen * sizeof(U8 *));
2861 SvPOK_on(sv_points);
2862 sv_2mortal(sv_points);
2863 points=(U8**)SvPV_nolen(sv_points );
2864 if ( trie_type != trie_utf8_fold
2865 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2866 {
2867 if (trie->bitmap)
2868 bitmap=(U8*)trie->bitmap;
2869 else
2870 bitmap=(U8*)ANYOF_BITMAP(c);
2871 }
2872 /* this is the Aho-Corasick algorithm modified a touch
2873 to include special handling for long "unknown char" sequences.
2874 The basic idea being that we use AC as long as we are dealing
2875 with a possible matching char, when we encounter an unknown char
2876 (and we have not encountered an accepting state) we scan forward
2877 until we find a legal starting char.
2878 AC matching is basically that of trie matching, except that when
2879 we encounter a failing transition, we fall back to the current
2880 states "fail state", and try the current char again, a process
2881 we repeat until we reach the root state, state 1, or a legal
2882 transition. If we fail on the root state then we can either
2883 terminate if we have reached an accepting state previously, or
2884 restart the entire process from the beginning if we have not.
2885
2886 */
2887 while (s <= last_start) {
2888 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2889 U8 *uc = (U8*)s;
2890 U16 charid = 0;
2891 U32 base = 1;
2892 U32 state = 1;
2893 UV uvc = 0;
2894 STRLEN len = 0;
2895 STRLEN foldlen = 0;
2896 U8 *uscan = (U8*)NULL;
2897 U8 *leftmost = NULL;
2898#ifdef DEBUGGING
2899 U32 accepted_word= 0;
786e8c11 2900#endif
73104a1b
KW
2901 U32 pointpos = 0;
2902
2903 while ( state && uc <= (U8*)strend ) {
2904 int failed=0;
2905 U32 word = aho->states[ state ].wordnum;
2906
2907 if( state==1 ) {
2908 if ( bitmap ) {
2909 DEBUG_TRIE_EXECUTE_r(
2910 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2911 dump_exec_pos( (char *)uc, c, strend, real_start,
cb41e5d6 2912 (char *)uc, utf8_target, 0 );
6ad9a8ab 2913 Perl_re_printf( aTHX_
73104a1b
KW
2914 " Scanning for legal start char...\n");
2915 }
2916 );
2917 if (utf8_target) {
2918 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2919 uc += UTF8SKIP(uc);
2920 }
2921 } else {
2922 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2923 uc++;
2924 }
786e8c11 2925 }
73104a1b 2926 s= (char *)uc;
07be1b83 2927 }
73104a1b
KW
2928 if (uc >(U8*)last_start) break;
2929 }
2930
2931 if ( word ) {
2932 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2933 if (!leftmost || lpos < leftmost) {
2934 DEBUG_r(accepted_word=word);
2935 leftmost= lpos;
7016d6eb 2936 }
73104a1b 2937 if (base==0) break;
7016d6eb 2938
73104a1b
KW
2939 }
2940 points[pointpos++ % maxlen]= uc;
2941 if (foldlen || uc < (U8*)strend) {
9ad8cac4
KW
2942 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
2943 (U8 *) strend, uscan, len, uvc,
2944 charid, foldlen, foldbuf,
2945 uniflags);
73104a1b
KW
2946 DEBUG_TRIE_EXECUTE_r({
2947 dump_exec_pos( (char *)uc, c, strend,
cb41e5d6 2948 real_start, s, utf8_target, 0);
6ad9a8ab 2949 Perl_re_printf( aTHX_
147e3846 2950 " Charid:%3u CP:%4" UVxf " ",
73104a1b
KW
2951 charid, uvc);
2952 });
2953 }
2954 else {
2955 len = 0;
2956 charid = 0;
2957 }
07be1b83 2958
73104a1b
KW
2959
2960 do {
6148ee25 2961#ifdef DEBUGGING
73104a1b 2962 word = aho->states[ state ].wordnum;
6148ee25 2963#endif
73104a1b
KW
2964 base = aho->states[ state ].trans.base;
2965
2966 DEBUG_TRIE_EXECUTE_r({
2967 if (failed)
2968 dump_exec_pos( (char *)uc, c, strend, real_start,
cb41e5d6 2969 s, utf8_target, 0 );
6ad9a8ab 2970 Perl_re_printf( aTHX_
147e3846 2971 "%sState: %4" UVxf ", word=%" UVxf,
73104a1b
KW
2972 failed ? " Fail transition to " : "",
2973 (UV)state, (UV)word);
2974 });
2975 if ( base ) {
2976 U32 tmp;
2977 I32 offset;
2978 if (charid &&
2979 ( ((offset = base + charid
2980 - 1 - trie->uniquecharcount)) >= 0)
2981 && ((U32)offset < trie->lasttrans)
2982 && trie->trans[offset].check == state
2983 && (tmp=trie->trans[offset].next))
2984 {
2985 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 2986 Perl_re_printf( aTHX_ " - legal\n"));
73104a1b
KW
2987 state = tmp;
2988 break;
07be1b83
YO
2989 }
2990 else {
786e8c11 2991 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 2992 Perl_re_printf( aTHX_ " - fail\n"));
786e8c11 2993 failed = 1;
73104a1b 2994 state = aho->fail[state];
07be1b83 2995 }
07be1b83 2996 }
73104a1b
KW
2997 else {
2998 /* we must be accepting here */
2999 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 3000 Perl_re_printf( aTHX_ " - accepting\n"));
73104a1b
KW
3001 failed = 1;
3002 break;
786e8c11 3003 }
73104a1b
KW
3004 } while(state);
3005 uc += len;
3006 if (failed) {
3007 if (leftmost)
3008 break;
3009 if (!state) state = 1;
07be1b83 3010 }
73104a1b
KW
3011 }
3012 if ( aho->states[ state ].wordnum ) {
3013 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
3014 if (!leftmost || lpos < leftmost) {
3015 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
3016 leftmost = lpos;
07be1b83
YO
3017 }
3018 }
73104a1b
KW
3019 if (leftmost) {
3020 s = (char*)leftmost;
3021 DEBUG_TRIE_EXECUTE_r({
147e3846 3022 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
73104a1b
KW
3023 (UV)accepted_word, (IV)(s - real_start)
3024 );
3025 });
02d5137b 3026 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW
3027 FREETMPS;
3028 LEAVE;
3029 goto got_it;
3030 }
3031 s = HOPc(s,1);
3032 DEBUG_TRIE_EXECUTE_r({
6ad9a8ab 3033 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
73104a1b
KW
3034 });
3035 } else {
3036 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 3037 Perl_re_printf( aTHX_ "No match.\n"));
73104a1b
KW
3038 break;
3039 }
3040 }
3041 FREETMPS;
3042 LEAVE;
3043 }
3044 break;
3045 default:
3046 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
73104a1b
KW
3047 }
3048 return 0;
3049 got_it:
3050 return s;
6eb5f6b9
JH
3051}
3052
60165aa4
DM
3053/* set RX_SAVED_COPY, RX_SUBBEG etc.
3054 * flags have same meanings as with regexec_flags() */
3055
749f4950
DM
3056static void
3057S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM
3058 char *strbeg,
3059 char *strend,
3060 SV *sv,
3061 U32 flags,
3062 bool utf8_target)
3063{
3064 struct regexp *const prog = ReANY(rx);
3065
60165aa4
DM
3066 if (flags & REXEC_COPY_STR) {
3067#ifdef PERL_ANY_COW
3068 if (SvCANCOW(sv)) {
eb8fc9fe 3069 DEBUG_C(Perl_re_printf( aTHX_
60165aa4 3070 "Copy on write: regexp capture, type %d\n",
eb8fc9fe 3071 (int) SvTYPE(sv)));
5411a0e5
DM
3072 /* Create a new COW SV to share the match string and store
3073 * in saved_copy, unless the current COW SV in saved_copy
3074 * is valid and suitable for our purpose */
3075 if (( prog->saved_copy
3076 && SvIsCOW(prog->saved_copy)
3077 && SvPOKp(prog->saved_copy)
3078 && SvIsCOW(sv)
3079 && SvPOKp(sv)
3080 && SvPVX(sv) == SvPVX(prog->saved_copy)))
a76b0e90 3081 {
5411a0e5
DM
3082 /* just reuse saved_copy SV */
3083 if (RXp_MATCH_COPIED(prog)) {
3084 Safefree(prog->subbeg);
3085 RXp_MATCH_COPIED_off(prog);
3086 }
3087 }
3088 else {
3089 /* create new COW SV to share string */
196a02af 3090 RXp_MATCH_COPY_FREE(prog);
a76b0e90 3091 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
a76b0e90 3092 }
5411a0e5
DM
3093 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
3094 assert (SvPOKp(prog->saved_copy));
60165aa4
DM
3095 prog->sublen = strend - strbeg;
3096 prog->suboffset = 0;
3097 prog->subcoffset = 0;
3098 } else
3099#endif
3100 {
99a90e59
FC
3101 SSize_t min = 0;
3102 SSize_t max = strend - strbeg;
ea3daa5d 3103 SSize_t sublen;
60165aa4
DM
3104
3105 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 3106 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
3107 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
3108 ) { /* don't copy $' part of string */
3109 U32 n = 0;
3110 max = -1;
3111 /* calculate the right-most part of the string covered
f67a5002 3112 * by a capture. Due to lookahead, this may be to
60165aa4
DM
3113 * the right of $&, so we have to scan all captures */
3114 while (n <= prog->lastparen) {
3115 if (prog->offs[n].end > max)
3116 max = prog->offs[n].end;
3117 n++;
3118 }
3119 if (max == -1)
3120 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
3121 ? prog->offs[0].start
3122 : 0;
3123 assert(max >= 0 && max <= strend - strbeg);
3124 }
3125
3126 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 3127 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
3128 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
3129 ) { /* don't copy $` part of string */
3130 U32 n = 0;
3131 min = max;
3132 /* calculate the left-most part of the string covered
f67a5002 3133 * by a capture. Due to lookbehind, this may be to
60165aa4
DM
3134 * the left of $&, so we have to scan all captures */
3135 while (min && n <= prog->lastparen) {
3136 if ( prog->offs[n].start != -1
3137 && prog->offs[n].start < min)
3138 {
3139 min = prog->offs[n].start;
3140 }
3141 n++;
3142 }
3143 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
3144 && min > prog->offs[0].end
3145 )
3146 min = prog->offs[0].end;
3147
3148 }
3149
3150 assert(min >= 0 && min <= max && min <= strend - strbeg);
3151 sublen = max - min;
3152
196a02af 3153 if (RXp_MATCH_COPIED(prog)) {
60165aa4
DM
3154 if (sublen > prog->sublen)
3155 prog->subbeg =
3156 (char*)saferealloc(prog->subbeg, sublen+1);
3157 }
3158 else
3159 prog->subbeg = (char*)safemalloc(sublen+1);
3160 Copy(strbeg + min, prog->subbeg, sublen, char);
3161 prog->subbeg[sublen] = '\0';
3162 prog->suboffset = min;
3163 prog->sublen = sublen;
196a02af 3164 RXp_MATCH_COPIED_on(prog);
60165aa4
DM
3165 }
3166 prog->subcoffset = prog->suboffset;
3167 if (prog->suboffset && utf8_target) {
3168 /* Convert byte offset to chars.
3169 * XXX ideally should only compute this if @-/@+
3170 * has been seen, a la PL_sawampersand ??? */
3171
3172 /* If there's a direct correspondence between the
3173 * string which we're matching and the original SV,
3174 * then we can use the utf8 len cache associated with
3175 * the SV. In particular, it means that under //g,
3176 * sv_pos_b2u() will use the previously cached
3177 * position to speed up working out the new length of
3178 * subcoffset, rather than counting from the start of
3179 * the string each time. This stops
3180 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
3181 * from going quadratic */
3182 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
ea3daa5d
FC
3183 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
3184 SV_GMAGIC|SV_CONST_RETURN);
60165aa4
DM
3185 else
3186 prog->subcoffset = utf8_length((U8*)strbeg,
3187 (U8*)(strbeg+prog->suboffset));
3188 }
3189 }
3190 else {
196a02af 3191 RXp_MATCH_COPY_FREE(prog);
60165aa4
DM
3192 prog->subbeg = strbeg;
3193 prog->suboffset = 0;
3194 prog->subcoffset = 0;
3195 prog->sublen = strend - strbeg;
3196 }
3197}
3198
3199
3200
fae667d5 3201
6eb5f6b9
JH
3202/*
3203 - regexec_flags - match a regexp against a string
3204 */
3205I32
5aaab254 3206Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
ea3daa5d 3207 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
8fd1a950
DM
3208/* stringarg: the point in the string at which to begin matching */
3209/* strend: pointer to null at end of string */
3210/* strbeg: real beginning of string */
3211/* minend: end of match must be >= minend bytes after stringarg. */
3212/* sv: SV being matched: only used for utf8 flag, pos() etc; string
3213 * itself is accessed via the pointers above */
3214/* data: May be used for some additional optimizations.
d058ec57 3215 Currently unused. */
a340edde 3216/* flags: For optimizations. See REXEC_* in regexp.h */
8fd1a950 3217
6eb5f6b9 3218{
8d919b0a 3219 struct regexp *const prog = ReANY(rx);
5aaab254 3220 char *s;
eb578fdb 3221 regnode *c;
03c83e26 3222 char *startpos;
ea3daa5d
FC
3223 SSize_t minlen; /* must match at least this many chars */
3224 SSize_t dontbother = 0; /* how many characters not to try at end */
f2ed9b32 3225 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 3226 I32 multiline;
f8fc2ecf 3227 RXi_GET_DECL(prog,progi);
02d5137b
DM
3228 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
3229 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 3230 regexp_paren_pair *swap = NULL;
006f26b2 3231 I32 oldsave;
a3621e74
YO
3232 GET_RE_DEBUG_FLAGS_DECL;
3233
7918f24d 3234 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 3235 PERL_UNUSED_ARG(data);
6eb5f6b9
JH
3236
3237 /* Be paranoid... */
3dc78631 3238 if (prog == NULL) {
6eb5f6b9 3239 Perl_croak(aTHX_ "NULL regexp parameter");
6eb5f6b9
JH
3240 }
3241
6c3fea77 3242 DEBUG_EXECUTE_r(
03c83e26 3243 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM
3244 "Matching");
3245 );
8adc0f72 3246
b342a604
DM
3247 startpos = stringarg;
3248
4cf1a867
DM
3249 /* set these early as they may be used by the HOP macros below */
3250 reginfo->strbeg = strbeg;
3251 reginfo->strend = strend;
3252 reginfo->is_utf8_target = cBOOL(utf8_target);
3253
58430ea8 3254 if (prog->intflags & PREGf_GPOS_SEEN) {
d307c076
DM
3255 MAGIC *mg;
3256
fef7148b
DM
3257 /* set reginfo->ganch, the position where \G can match */
3258
3259 reginfo->ganch =
3260 (flags & REXEC_IGNOREPOS)
3261 ? stringarg /* use start pos rather than pos() */
3dc78631 3262 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
25fdce4a
FC
3263 /* Defined pos(): */
3264 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
fef7148b
DM
3265 : strbeg; /* pos() not defined; use start of string */
3266
6ad9a8ab 3267 DEBUG_GPOS_r(Perl_re_printf( aTHX_
147e3846 3268 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
fef7148b 3269
03c83e26
DM
3270 /* in the presence of \G, we may need to start looking earlier in
3271 * the string than the suggested start point of stringarg:
0b2c2a84 3272 * if prog->gofs is set, then that's a known, fixed minimum
03c83e26
DM
3273 * offset, such as
3274 * /..\G/: gofs = 2
3275 * /ab|c\G/: gofs = 1
3276 * or if the minimum offset isn't known, then we have to go back
3277 * to the start of the string, e.g. /w+\G/
3278 */
2bfbe302 3279
8e1490ee 3280 if (prog->intflags & PREGf_ANCH_GPOS) {
4cf1a867
DM
3281 if (prog->gofs) {
3282 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
3283 if (!startpos ||
3284 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
3285 {
6ad9a8ab 3286 DEBUG_r(Perl_re_printf( aTHX_
4cf1a867
DM
3287 "fail: ganch-gofs before earliest possible start\n"));
3288 return 0;
3289 }
2bfbe302 3290 }
4cf1a867
DM
3291 else
3292 startpos = reginfo->ganch;
2bfbe302
DM
3293 }
3294 else if (prog->gofs) {
4cf1a867
DM
3295 startpos = HOPBACKc(startpos, prog->gofs);
3296 if (!startpos)
b342a604 3297 startpos = strbeg;
03c83e26 3298 }
58430ea8 3299 else if (prog->intflags & PREGf_GPOS_FLOAT)
b342a604 3300 startpos = strbeg;
03c83e26
DM
3301 }
3302
3303 minlen = prog->minlen;
b342a604 3304 if ((startpos + minlen) > strend || startpos < strbeg) {
6ad9a8ab 3305 DEBUG_r(Perl_re_printf( aTHX_
03c83e26
DM
3306 "Regex match can't succeed, so not even tried\n"));
3307 return 0;
3308 }
3309
63a3746a
DM
3310 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
3311 * which will call destuctors to reset PL_regmatch_state, free higher
3312 * PL_regmatch_slabs, and clean up regmatch_info_aux and
3313 * regmatch_info_aux_eval */
3314
3315 oldsave = PL_savestack_ix;
3316
dfa77d06
DM
3317 s = startpos;
3318
e322109a 3319 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM
3320 && !(flags & REXEC_CHECKED))
3321 {
dfa77d06 3322 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 3323 flags, NULL);
dfa77d06 3324 if (!s)
7fadf4a7
DM
3325 return 0;
3326
e322109a 3327 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM
3328 /* we can match based purely on the result of INTUIT.
3329 * Set up captures etc just for $& and $-[0]
3330 * (an intuit-only match wont have $1,$2,..) */
3331 assert(!prog->nparens);
d5e7783a
DM
3332
3333 /* s/// doesn't like it if $& is earlier than where we asked it to
3334 * start searching (which can happen on something like /.\G/) */
3335 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3336 && (s < stringarg))
3337 {
3338 /* this should only be possible under \G */
58430ea8 3339 assert(prog->intflags & PREGf_GPOS_SEEN);
6ad9a8ab 3340 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
d5e7783a
DM
3341 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3342 goto phooey;
3343 }
3344
7fadf4a7
DM
3345 /* match via INTUIT shouldn't have any captures.
3346 * Let @-, @+, $^N know */
3347 prog->lastparen = prog->lastcloseparen = 0;
196a02af 3348 RXp_MATCH_UTF8_set(prog, utf8_target);
3ff69bd6
DM
3349 prog->offs[0].start = s - strbeg;
3350 prog->offs[0].end = utf8_target
8875b6de 3351 ? (char*)utf8_hop_forward((U8*)s, prog->minlenret, (U8 *) strend) - strbeg
3ff69bd6 3352 : s - strbeg + prog->minlenret;
7fadf4a7 3353 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 3354 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM
3355 strbeg, strend,
3356 sv, flags, utf8_target);
3357
7fadf4a7
DM
3358 return 1;
3359 }
3360 }
3361
6c3fea77 3362 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 3363
dfa77d06 3364 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
6ad9a8ab 3365 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
a72c7584
JH
3366 "String too short [regexec_flags]...\n"));
3367 goto phooey;
1aa99e6b 3368 }
1de06328 3369
6eb5f6b9 3370 /* Check validity of program. */
f8fc2ecf 3371 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH
3372 Perl_croak(aTHX_ "corrupted regexp program");
3373 }
3374
196a02af
DM
3375 RXp_MATCH_TAINTED_off(prog);
3376 RXp_MATCH_UTF8_set(prog, utf8_target);
1738e041 3377
6c3fea77
DM
3378 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3379 reginfo->intuit = 0;
02d5137b
DM
3380 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3381 reginfo->warned = FALSE;
02d5137b 3382 reginfo->sv = sv;
1cb48e53 3383 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
6eb5f6b9 3384 /* see how far we have to get to not match where we matched before */
fe3974be 3385 reginfo->till = stringarg + minend;
6eb5f6b9 3386
60779a30 3387 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
82c23608
FC
3388 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3389 S_cleanup_regmatch_info_aux has executed (registered by
3390 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3391 magic belonging to this SV.
3392 Not newSVsv, either, as it does not COW.
3393 */
3394 reginfo->sv = newSV(0);
4cba5ac0 3395 SvSetSV_nosteal(reginfo->sv, sv);
82c23608
FC
3396 SAVEFREESV(reginfo->sv);
3397 }
3398
331b2dcc
DM
3399 /* reserve next 2 or 3 slots in PL_regmatch_state:
3400 * slot N+0: may currently be in use: skip it
3401 * slot N+1: use for regmatch_info_aux struct
3402 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3403 * slot N+3: ready for use by regmatch()
3404 */
bf2039a9 3405
331b2dcc
DM
3406 {
3407 regmatch_state *old_regmatch_state;
3408 regmatch_slab *old_regmatch_slab;
3409 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3410
3411 /* on first ever match, allocate first slab */
3412 if (!PL_regmatch_slab) {
3413 Newx(PL_regmatch_slab, 1, regmatch_slab);
3414 PL_regmatch_slab->prev = NULL;
3415 PL_regmatch_slab->next = NULL;
3416 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3417 }
bf2039a9 3418
331b2dcc
DM
3419 old_regmatch_state = PL_regmatch_state;
3420 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 3421
331b2dcc
DM
3422 for (i=0; i <= max; i++) {
3423 if (i == 1)
3424 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3425 else if (i ==2)
3426 reginfo->info_aux_eval =
3427 reginfo->info_aux->info_aux_eval =
3428 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 3429
331b2dcc
DM
3430 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3431 PL_regmatch_state = S_push_slab(aTHX);
3432 }
bf2039a9 3433
331b2dcc
DM
3434 /* note initial PL_regmatch_state position; at end of match we'll
3435 * pop back to there and free any higher slabs */
bf2039a9 3436
331b2dcc
DM
3437 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3438 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 3439 reginfo->info_aux->poscache = NULL;
bf2039a9 3440
331b2dcc 3441 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 3442
331b2dcc
DM
3443 if ((prog->extflags & RXf_EVAL_SEEN))
3444 S_setup_eval_state(aTHX_ reginfo);
3445 else
3446 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 3447 }
d3aa529c 3448
6eb5f6b9 3449 /* If there is a "must appear" string, look for it. */
6eb5f6b9 3450
288b8c02 3451 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG
3452 /* We have to be careful. If the previous successful match
3453 was from this regex we don't want a subsequent partially
3454 successful match to clobber the old results.
3455 So when we detect this possibility we add a swap buffer
d8da0584
KW
3456 to the re, and switch the buffer each match. If we fail,
3457 we switch it back; otherwise we leave it swapped.
e9105d30
GG
3458 */
3459 swap = prog->offs;
7d79ca09
DM
3460 /* avoid leak if we die, or clean up anyway if match completes */
3461 SAVEFREEPV(swap);
e9105d30 3462 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2b1a3689 3463 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
147e3846 3464 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
2b1a3689
YO
3465 0,
3466 PTR2UV(prog),
495f47a5
DM
3467 PTR2UV(swap),
3468 PTR2UV(prog->offs)
3469 ));
c74340f9 3470 }
6eb5f6b9 3471
ba6840fb
YO
3472 if (prog->recurse_locinput)
3473 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3474
0fa70a06
DM
3475 /* Simplest case: anchored match need be tried only once, or with
3476 * MBOL, only at the beginning of each line.
3477 *
3478 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3479 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3480 * match at the start of the string then it won't match anywhere else
3481 * either; while with /.*.../, if it doesn't match at the beginning,
3482 * the earliest it could match is at the